16 THEORETICAL AND COMPUTATIONAL CHEMISTRY
Computational Photochemistry
THEORETICAL AND COMPUTATIONAL CHEMISTRY
S ERIES E DITORS Professor P. Politzer Department of Chemistry University of New Orleans New Orleans, LA 70148, U.S.A.
Professor Z.B. Maksic Rudjer Boškovic Institute P.O. Box 1016, 10001 Zagreb, Croatia
VOLUME 1
VOLUME 9
Quantitative Treatments of Solute/Solvent Interactions P. Politzer and J.S. Murray (Editors)
Theoretical Biochemistry: Processes and Properties of Biological Systems L.A. Eriksson (Editor)
VOLUME 2
VOLUME 10
Modern Density Functional Theory: A Tool for Chemistry J.M. Seminario and P. Politzer (Editors)
Valence Bond Theory D.L. Cooper (Editor)
VOLUME 3 Molecular Electrostatic Potentials: Concepts and Applications J.S. Murray and K. Sen (Editors)
VOLUME 4 Recent Developments and Applications of Modern Density Functional Theory J.M. Seminario (Editor)
VOLUME 5 Theoretical Organic Chemistry C. Párkányi (Editor)
VOLUME 6 Pauling’s Legacy: Modern Modelling of the Chemical Bond Z.B. Maksic and W.J. Orville-Thomas (Editors)
VOLUME 7 Molecular Dynamics: From Classical to Quantum Methods P.B. Balbuena and J.M. Seminario (Editors)
VOLUME 8 Computational Molecular Biology J. Leszczynski (Editor)
VOLUME 11 Relativistic Electronic Structure Theory, Part 1. Fundamentals P. Schwerdtfeger (Editor)
VOLUME 12 Energetic Materials, Part 1. Decomposition, Crystal and Molecular Properties P. Politzer and J.S. Murray (Editors)
VOLUME 13 Energetic Materials, Part 2. Detonation, Combustion P. Politzer and J.S. Murray (Editors)
VOLUME 14 Relativistic Electronic Structure Theory, Part 2. Applications P. Schwerdtfeger (Editor)
VOLUME 15 Computational Materials Science J. Leszczynski (Editor)
VOLUME 16 Computational Photochemistry M. Olivucci (Editor)
16 THEORETICAL AND COMPUTATIONAL CHEMISTRY
Computational Photochemistry
Edited by M. Olivucci Dipartimento di Chimica dell’Universita di Siena Siena, Italy
ELSEVIER 2005 Amsterdam – Boston – Heidelberg – London – New York – Oxford – Paris San Diego – San Francisco – Singapore – Sydney – Tokyo
ELSEVIER B.V. Radarweg 29 P.O. Box 211, 1000 AE Amsterdam, The Netherlands
ELSEVIER Inc. 525 B Street Suite 1900, San Diego CA 92101-4495, USA
ELSEVIER Ltd. The Boulevard Langford Lane, Kidlington, Oxford OX5 1GB, UK
ELSEVIER Ltd. 84 Theobalds Road London WC1X 8RR UK
© 2005 Elsevier B.V. All rights reserved. This work is protected under copyright by Elsevier B.V., and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier’s Rights Department in Oxford, UK: phone (+44) 1865 843830, fax (+44) 1865 853333, e-mail:
[email protected]. Requests may also be completed on-line via the Elsevier homepage (http://www. elsevier.com/locate/permissions). In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: (+1) (978) 7508400, fax: (+1) (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London W1P 0LP, UK; phone: (+44) 20 7631 5555; fax: (+44) 20 7631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of the Publisher is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier’s Rights Department, at the fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made.
First edition 2005 Library of Congress Cataloging in Publication Data A catalog record is available from the Library of Congress. British Library Cataloguing in Publication Data A catalogue record is available from the British Library.
ISBN-13: 978-0-444-52110-1 ISBN-10: 0-444-52110-0 ISSN: 1380-7323 (Series) The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.
This book is dedicated to my beloved parents Armando and Anna
This Page is Intentionally Left Blank
Vll
Contents Foreword by Josef Michl Preface I.
Computational Photochemistry Massimo Olivucci and Adalgisa Sinicropi
ix xiii 1
II. Ab initio Methods for Excited States Manuela Merchan and Luis Serrano-Andres
35
III. Density Functional Methods for Excited States: Equilibrium Structure and Electronic Spectra Filipp Furche and Dmitrij Rappoport
93
IV. Electronic and Vibronic Spectra of Molecular Systems: Models and Simulations based on Quantum Chemically Computed Molecular Parameters Fabrizia Negri and G. Orlandi
129
V. Semiclassical Nonadiabatic Trajectory Computations In Photochemistry: Is The Reaction Path Enough To Understand A Photochemical Reaction Mechanism? G. A. Worth, M. J. Bearpark and Michael A. Robb
171
VI. Computation of Photochemical Reaction Mechanisms in Organic Chemistry Marco Garavelli, Fernando Bernardi and A. Cembran
191
VII. Computation of Reaction Mechanisms and Dynamics in Photobiology Seth Olsen, Alessandro Toniolo, Chaehyuk Ko, Leslie Manohar, Kristina Lamothe, and Todd J. Martinez
225
VIII. Development of Theory with Computation Howard Zimmerman
255
IX. Calculations of Electronic Spectra of Transition Metal Complexes Kerstin Pierloot
279
Vlll
X. Perspectives in Calculations on Excited State in Molecular Systems Bjorn Roos
317
Index
349
IX
Foreword Josef Michl Department of Chemistry and Biochemistry University of Colorado Boulder, CO 80309-0215
Anyone who has dealt with ground state mechanistic chemistry of organic or inorganic reactions appreciates how complex and demanding it can be. Nevertheless, from theoretical and computational standpoints, its complexity pales compared to mechanistic photochemistry, with its maze of paths that can be followed after the absorption of a photon of U V-visible light by a molecule. Even for the simplest photoreactions, it is a major feat to figure out the details of the routes followed by the molecule in a qualitative way and to rationalize the nature of the final ground state product. A reliable calculation of a quantity as fundamental to the experimental photochemist as the quantum yield of product formation remains a distant goal. There is competition between nonradiative and radiative processes, between spin-allowed and spinforbidden processes, between adiabatic and diabatic processes, between vibrational relaxation into one and another minimum after return to the ground state. There is the issue of possible violations of Kasha's rule, since a variation of the initial excitation energy is not always without consequences, even in solution photochemistry. Energy transfer and electron transfer possibilities often lurk in the background. Solvent effects are complex and manifold. No wonder a distinguished ground-state computational chemist friend who attended a theoretical photochemistry meeting with me a few years ago shook his head in disbelief after the first day of lectures, and asked something like "Isn't there an easier way to earn a living?". Photochemists thrive on complexity, theoreticians more than most. Although we cannot predict the quantum yield of a simple reaction any more accurately than we could when my interest in photochemistry was first piqued nearly half a century ago, great conceptual progress has been made. Then, it was not even very clear just what to calculate. Today, there is little doubt that we need dynamics on lowest potential energy surfaces. The concept of a potential energy surface guiding an excited molecule, with only occasional hops from one surface to another, was new to most experimental photochemists then. It has proven its heuristic and computational value since, and pervades the present book. Only for molecules with a very high
density of lowest-energy electronic states, such as those of saturated compounds, is it likely to be inadequate. The basic notions that are so familiar today were established in the sixties and seventies, and perusal of a 1974 review article[l] reveals the whole slew of the necessary concepts: excited state barriers and funnels for ultrafast return to the ground state, reactions with vibrationally equilibrated intermediates and direct reactions proceeding through state crossings, vibrationally equilibrated and "hot" excited and ground state reactions, internal and external heavy atom effects on intersystem crossing, etc. Back then, we used symmetry considerations, correlation diagrams, or calculations on simple models to estimate at what geometries barriers and funnels are likely to lie. Our first numerical computation of a funnel (conical intersection) relevant for a photochemical isomerization in an organic molecule was published only twenty years ago,[2] and it was made possible by the presence of symmetry at the state touching point. Today, advances in computer technology and in quantum chemical methodology, especially in multireference methods, many due to the authors of the chapters that follow, permit quite reliable calculations of these essential features at general geometries, and a thick book on conical intersections has just appeared. [3] Yes, difficulties remain, especially in evaluating the relative energies of covalent ("dotdot") and zwitterionic ("hole-pair") states with sufficient accuracy. Another problem is the proper treatment of reaction dynamics in all but the smallest molecules. After all, only those conical intersections are relevant that can be reached by the excited molecule in the short time available to it. And yes, in my opinion, too much emphasis has been put in recent years on the geometries of the lowest energy point of a conical intersection. This is an issue on which I have had a gentle disagreement with many. I would argue that these points are usually nearly irrelevant, because a molecule that has reached the seam of a conical intersection will fall to the lower surface right away and will not have time to ride the seam, looking for its lowest energy point. Thus, the effective funnel locations are those in which the seam is first reached, and not the lowest energy point in the intersection subspace. Unfortunately, the former are harder to calculate. In fact, much of the wave packet most likely seeps to the lower surface at geometries at which the state touching is still weakly avoided, simply because of their higher dimensionality, and in that sense the regions of weakly avoided crossings need not be as immaterial as they are sometimes made out to be. In spite of these minor quibbles, we all agree that the improvement from the level of mechanistic interpretations standard a quarter of a century ago to that common today is striking. Another interesting comparison is with a book on theoretical photochemistry that was published fifteen years ago.[4] It was a monograph rather than an edited multiauthor volume, and was organized differently in that it attempted a systematic treatment of all important classes of organic photoreactions, organized by Salem's concept of topicity. However, the ab initio calculations presented were hardly more than glorified correlation diagrams, involved no geometry
XI
optimization, and were pathetic by the present book's standards. The qualitative concepts are all that survives. Clearly, computational photochemistry has made tremendous strides in recent decades, and continues to do so. The present collection of ten outstanding contributions provides a fine illustration of this statement.
REFERENCES [1] Michl, J. "Physical Basis of Qualitative MO Arguments in Organic Photochemistry", Topics in Current Chemistry 1974, 46,1. [2] Bonacic-Koutecky, V.; Michl, J. "Photochemical Syn-Anti Isomerization of a Schiff Base: A Two-Dimensional Description of a Conical Intersection in Formaldimine", Theor. Chim. Ada 1985, 68, 45. [3] Domcke, W.; Yarkony, D. R.; Koppel, Editors, Conical Intersections: Electronic Structure, Dynamics & Spectroscopy, World Scientific Publishing Co., Singapore 2004. [4] Michl, J.; Bonacic-Koutecky, V. Electronic Aspects of Organic Photochemistry, John Wiley and Sons, Inc.: New York, 1990.
This Page is Intentionally Left Blank
Xlll
Preface The chemical community have recently witnessed a growing interest in the application of computational methods to problems involving electronically excited molecules. This is mostly due to a change in the field of photochemical sciences. In fact, until two decades ago, these were dominated by the search for novel photochemical or photophysical properties. In other words, photochemists were reporting on the effects of light at the molecular-level. In contrast, contemporary photochemists look for ways to exploit light to drive various molecular-level actions such as pollutant scavenging and removal, mechanical motion, sensing and signalling, photocatalysis and others. While such raising technologies require the preparation of molecules capable of performing specific functions, the lack of knowledge on the molecular mechanisms of light energy exploitation and wastage constitutes a severe limitation to the design of such systems. In this book, a selected group of experts show how the development, implementation and application of quantum chemical methods in photochemistry and spectroscopy provide a way to tackle this problem. Until recently the computer-aided investigation of photochemical reactions (i.e. reactions that are initiated by light absorption rather then by heat) was unpractical if not impossible. Because of this the simulation of fundamental chemical and biological events such as bleaching, fluorescence, phosphorescence, photochromism, vision, photosynthesis, phototropism, and others could not be performed. Thus, despite the growing availability of computer power, there were neither computer tools nor a clear theoretical basis for the investigation of photoexcited molecules. One key point for the solution of this problem was to establish the nature of the spatial arrangement of the atoms that allows a photoexcited molecule to efficiently decay from the excited state to the ground state thus initiating product formation. Loosely, this critical molecular structure, often called "photochemical funnel", plays in photochemistry, the role played by the transition structure in a thermal process. Thus the description of a reaction pathway in photochemistry must involve the description of the path leading to and departing from the photochemical funnel. About fifteen years ago different research lines started to change this unfavorable situation. Few lines were merely related to the elucidation of the general mechanism of photochemical reactions, including the nature of the photochemical funnel, while others involved the development of software tools allowing for an accurate evaluation and mapping of the potential energy surface of photoexcited molecules. For instance, at the end of the 80's improved ab initio quantum chemical methodologies became available which were suitable for computing, in a balanced way, excited and ground state energy surfaces taking into account the complete set of the 3N-6 nuclear degrees of freedom of the reacting system (N is the number of atoms). Such progress made possible to provide clear evidence that the ideas of
XIV
Edward Teller, Lionel Salem, Howard Zimmerman and Josef Michl, stating that for singlet photochemical reactions the photochemical funnel corresponds to a conical intersection of the excited and ground state energy surfaces, were correct. (A book on conical intersections has recently appeared: "Conical Intersections: Electronic Structure, Dynamics and Spectroscopy"; Domcke, W., Yarkony, D. R., Koppel, H., Eds.; World Scientific: Singapore, 2004) The target of the present book is two-fold. The first, and most ambitious one, is to contribute to establish a branch of computational chemistry that deals with the properties and reactivity of photoexcited molecules (see Chapter 1). Accordingly the book should give not only an historical view of the "scientific adventure" (see the Foreword written by one of the original and major player) that led to the emergence of the field but also a survey of the work that characterizes it. In order to satisfy this requirement the book provides an overview of few general strategies currently employed to investigate photochemical processes. The second target of the book is to give an account of the status of knowledge in either the mechanistic (conceptual) and methodological research lines in computational photochemistry. In fact, during the last ten years the potential energy surfaces of several organic chromophores were mapped. The resulting "maps" reveal prototype photochemical reaction mechanisms and form a firm body of computational photochemistry results. Accordingly, three book chapters focus on instructive case-studies comprising: (i) organic chromophores (Chapter 1, 6 and 8), (ii) biologically related chromophores (Chapter 7), (ii) photochemical funnels and reactive intermediates (Chapter 8 and 9). Such (still ongoing) systematic investigation could not be carried out without the development of novel computational tools that, nowadays, constitute the computational photochemist toolbox. These tools belong to four classes that will be reviewed in the remaining book chapters: (i) tools for the accurate computation of the excited state potential energy (Chapters 2, 3 and 10), (ii) tools for the prediction of absorption, fluorescence and Resonance Raman spectra (Chapter 4) (iii) tools for the mapping of excited state potential energy surfaces (including locating photochemical funnels and excited state reaction paths, Chapters 6) and, finally, (iv) tools for the computation of "photochemical" semi-classical trajectories (i.e. trajectories that start on the excited state energy surface and continue along the ground state surface, Chapter 5 and 7). A final chapter (Chapter 10), written by one of the major experts of electronic structure theories, provides a review and a perspective on the technologies for the ab initio computation of excited state energy surfaces. All authors have made an effort to write the chapters in a plain and simple way. Thus "Computational Photochemistry" should be readable not only by computational and theoretical chemists but also by chemists (e.g. photochemists, photobiologists and material scientists) interested in using computer tools in their laboratories. I feel deeply indebted to all authors that, not only have readily accepted my invitation, but have felt that the book may have provided a first, probably still crude, picture of an expanding field of computational chemistry. However, it is important to stress that many other scientists
XV
have given important contributions to the field and, indirectly, to the material reported in this book. The Editor feels indebted to the many colleagues including spectroscopists, photochemists, organic chemists and theoreticians that through both discussions and criticism have stimulated the present editorial effort. There remains only the pleasant task of thanking those who have otherwise been of help in the preparation of the book. Prof. Professor Zvonimir B. Maksic for originally inviting me to plan and edit the book and Andrew Gent of Elsevier for the attention devoted to progress of our work. A very special thank goes to Dr. Adalgisa Sinicropi (one of the author of Chapter 1) for taking care of many of the practical problems related to the assembly and revision of the chapters, for the production of the index and for her willingness to share the many, sometime frustrating, decision that I had to take during the various stages of the manuscript handling. As this is my first editorial work I cannot fail to acknowledge the fundamental role played, in the development of my scientific personality and career, by the chemists that have not only cast my education but shared with me their enthusiasm for many wonderful years: Fernando Bernardi and Michael A. Robb. Finally I am deeply indebted to my wife Matilde and my children Paolo, Enrico and Lidia for never let me feel alone during the too many days that I spend away from them. Massimo Olivucci Professor of Organic Chemistry Universita di Siena
This Page is Intentionally Left Blank
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 Elsevier B.V. All rights reserved
I. Computational Photochemistry Massimo Olivucci and Adalgisa Sinicropi Dipartimento di Chimica, Universita di Siena, Italy 1. INTRODUCTION The study of photochemical problems by means of computer simulations using specialized software tools and strategies enable us to get an understanding at the microscopic level of what happens to a molecule after absorption of a photon. A detailed understanding of the properties of electronically excited state species and the knowledge of the molecular mechanisms which control the fate of the energy deposited on a molecule after absorption, increase our ability to design efficient photochemical reactions and artificial photosynthetic systems. Furthermore, this represents a fundamental requirement for the rational design of novel materials, molecular devices and molecular level machines. On a more general ground, the ability to simulate, using complementary computational strategies, photoinduced events often allows to explore areas of chemistry that experiment could touch only indirectly. Together with the mechanistic ideas discussed below these strategies define the field of "computational chemistry". The application of quantum mechanics to chemical problems goes back to the end of the 1950s when computers came into use and it was possible to handle very complicated mathematical equations describing such complex systems as molecules. Note that even if the "... fundamental laws necessary for the mathematical treatment of a large part of physics and the whole of chemistry were completely known ...", as Dirac affirmed in 1929, accurate calculations of molecular properties and chemical reaction pathways were not possible at that time. In 1970 Pople published the first release of the GAUSSIAN program [1], making thus computational methods available to scientists [2]. The following growth of the speed of computers along with the development of more and more accurate quantum chemistry tools and methodologies implemented in commercially available program have made computational investigation of molecular structures and reactivity a standard practise. The way in which a chemical reaction step is investigated involves the computation of the transition state structure (TS) that connects a reactant to a product along with the associated energy barriers (Fig. 1). In particular, the energy of the transition state provides information on the time scale of the reaction while the geometrical structure of the transition state (TS) provides information on the stereochemistry of the reaction and sensitivity to different substituents. The progression of the molecular structure of the reactant toward the TS and the product constitutes the so called "reaction path" which can be mapped computing the
R Fig. 1. Schematic representation of the structure of the potential energy surface for a thermal chemical reaction. The dashed curve indicates the minimum energy path. R and P are local energy minima corresponding to reactants and products. TS is a saddle point corresponding to the transition structure.
minimum energy path (MEP) connecting the reactant (R) to a product (P) along the 3N-6 dimensional potential energy surface (N is the number of nuclei in the reacting system). One of the first computations in photochemistry/photophysics has to be ascribed to R. G. Parr and R. Pariser. In the early 1950s, they developed a semi-empirical method based on LCAOMO jt-electron theory to predict the electronic spectra and electronic structure of complex unsatured molecules, initially of benzene and N-heterocyclic analogues [3; 4]. Later on, using one of the first computers, an IBM 701, they were able to assign the electronic structure and electronic spectra of azulene and of the polyacenes [5-8]. The theoretical approach of Parr and Pariser together with the contribution of Pople formed the basis of the Pariser-Parr-Pople (PPP) theory that is one of the first semi-empirical method based on the ZDO (zerodifferential overlap) approximation. During '60s, '70s and even '80s many researchers and experts in the field of organic photochemistry shared their knowledge and published several papers [9-40] trying to understand the behaviour of electronically excited molecules and make a wide-ranging classification of photochemical reactions. The formulation of photoreduction mechanisms was mainly based on the construction of correlation diagrams. Although the interest for a unique theory of photochemical reactions was well recognized, the computational investigation of photochemical reaction mechanisms could not be easily implemented at the same level seen for the thermal chemistry. This frustrating status is somehow described in a 1990 paper [41], where N. J. Turro stated: "... the use of computational methods to elucidate reaction mechanisms has not really made a major impact on the way organic photochemists think about such mechanisms. The Woodward-Hoffmann rules and Salem diagrams of the 1960s and 1970s still serve as the basis for the day-to-day analysis of photoreactions...". On the other hand, other research groups in the field of organic photochemistry were tackling this
computational problem and realized that a detailed knowledge of the excited state molecular structure could lie at the basis of the "resolution" of the reaction mechanism. A detailed account of the development of mechanistic ideas and early results in the laboratory of H. E. Zimmerman will be given on Chapter 8. The major difficulty encountered by chemists in doing an exhaustive investigation of photochemical reactivity resided in the absence of robust computer tools able to map the MEP for excited state species. In fact, while a thermal reaction is governed by the topography of a single potential energy surface (starts and ends on the ground state of the reacting system), a photochemical reaction path evolves at least on two potential energy surfaces. Thus, in order to compute such path one needs to connect a reactant that is located on an excited state energy surface to products that are located on the ground state energy surface. This could only be done establishing the nature of the spatial arrangement of the atoms that allows a photoexcited molecule to efficiently decay from the excited state to the ground state thus initiating product formation. Loosely, this critical molecular structure, often called "photochemical funnel" plays, in photochemistry, the role of the transition state of a thermal process. As we will detail below, the characterization of the molecular structure and relative stability of the "photochemical funnel" in terms of conical intersections and singlet/triplet crossings is of central importance in mechanistic photochemistry. Nowadays, computational strategies are available for locating conical intersection and singlet/triplet crossing points and for constructing inter-state "photochemical" reaction pathways. These tools comprise methodologies for the optimisation of low-lying crossings between pair of potential energy surfaces and the computation of relaxation paths from a photoexcited reactant (e.g. from the Franck-Condon (FC) structure) to a deactivation channel. More in general, it is possible to compute the entire pathway connecting an excited state molecule to its ground state product. The major computational tools as well as few case-studies in the field of organic photochemistry will be revised by Cembran et al. in Chapter 6. The field of computational photochemistry is a relatively young field, especially when applied to the study of ultrafast reactions, but it is now established as a branch of computational chemistry and as a powerful, sometimes unique, way to simulate the molecular mechanism underlying fundamental chemical and biological events such as vision, primitive photosynthesis, phototropism, photochromism, bleaching, fluorescence, phosphorescence. Accordingly, the 3rd edition of the "Glossary of Terms Used in Photochemistry" (to be published in 2005) will contain new terms related to the use of computational tools in photochemistry (like Conical intersection, Photochemical Reaction Path, Minimum Energy Reaction Path).
2. PHOTOCHEMISTRY, PHOTOPHYSICS AND PHOTOBIOLOGY MEDIATED BY CONICAL INTERSECTION FUNNELS As mentioned above, in the past, correlation diagrams were, in many cases, the only practical tools available to the chemists to formulate reaction mechanism for thermal and photochemical reactions. For instance for pericyclic reactions Woodward-Hoffmann orbital correlation diagrams [42] and Longuet-Higgins and Abrahamson state correlation diagrams [43-45] were used. In the field of photochemical reactions the Van der Lugt-Oosteroff diagrams [40] were based on the hypothesis that avoided crossings provide the point of return of an excited state species to the ground state. At such an avoided crossing, if the energy gap is larger than few kcal mol"1, the excited state species will rapidly thermalize and the decay probability will be determined by the Fermi Golden Rule. However, within this model, the probability of decay should be small (unless the energy gap is small) and the radiationless decay process should occur on the same time scale of fluorescence (in ns [46; 47]). On the other hand, it is well known that many photochemical processes are extremely fast (well below one picosecond, i.e. on the timescale of a single molecular vibration) and associated with a complete lack of fluorescence. Furthermore, they are often stereospecific, implying a concerted mechanism. Indeed, femtosecond excited state lifetimes have been observed, for instance, for simple dienes [48], cyclohexadienes [48-50], hexatrienes [51], and in both free [52] and opsin-bound [53] retinal protonated Schiff bases. These observations suggest that a real crossing is accessible to the system. At such surface crossing the probability of decay is very high and the corresponding molecular identify the photochemical funnel (such name for the excited state decay channel suggests that the excited reactant must be "funnelled" through this point to initiate product formation). Thus, a photochemical funnel corresponds to a molecular structure that "lives" for only few femtoseconds (10" seconds). The history of conical intersection goes back to more than 60 years ago when, in 1937, the physicist Edward Teller giving a lecture at the Symposium on Molecular Structure [54] suggested that it was the electronic factors that may play the dominant role in the efficiency of radiationless decay. Teller made two general observations: in a polyatomic molecule the non-crossing rule, which is rigorously valid for diatomics, fails and two electronic states, even if they have the same symmetry, are allowed to cross at a conical intersection. radiationless decay from the upper to the lower intersecting state occurs within a single vibrational period when the system "travels" in the vicinity of such intersection points. On the basis of these observations, in 1969, at the Twentieth Farkas Memorial Symposium, Teller proposed that conical intersections may provide a common and very fast decay channel from the lowest excited states of polyatomics, which would explain the lack of fluorescence of the funnel[55].
In 1966, the organic chemist H. E. Zimmerman presented an alternative approach to the well known Woodward-Hoffmann method to predict the factors controlling ground and excited state reactions [9-11]. Zimmerman proposed an "MO Following" procedure that was capable of dealing with reactions lacking the symmetry to construct correlation diagrams. As an example he used the butadiene to cyclobutene closure and he found that along the reaction route a crossing (i.e., degeneracy) occurs. Thus, he concluded that such crossing point are significant in organic photochemistry and may provide a route for conversion of excited state reactant to ground state product. Michl [32; 33] proposed, independently, the same idea and documented such features in ab initio calculations on the H4 system [29; 30]. In 1970, Evleth and co-workers, in their work on the photolysis of aryldiazonium salts, interpreted their quantum yield measurements in terms of a complex energy surface crossing patterns. In the same years, Salem [35] proposed his state correlation diagrams that illustrated the occurrence of conical intersections at symmetric geometries in the photochemistry of carbonyl compounds. A continuous exchange of ideas between Salem, Turro e Dauben (as documented by Turro in a recent paper [56]) lead to the first complete classification of photochemical reactions [21; 36] using Salem's development of energy surface theory. Subsequently, geometries of few conical intersections were computed for Schiff base syn-anti isomerization by Bonacic-Koutecky and Michl [57], using ab initio procedures and the "3x3" model of biradicaloid electronic structure [58] was elaborated to permit qualitative prediction of geometries at which Si/So conical intersections take place [19; 34]. More recently, Yarkony [59; 60] and Ruedenberg [61] identified conical intersections geometries in small molecules. Despite the fact that the idea of Teller, Zimmerman, Michl and Salem represented an important refinement of the avoided crossing model, conical intersections were thought to be extremely rare or inaccessible (i.e. located too high in energy) in organic compounds and thus were disregarded. The main difficult has to be ascribed to the fact that, in practice, excited state quantum chemical computations require non-conventional methodologies and strategies based upon the use of multi-reference wavefunctions. (i.e., the so called post-SCF methods) rather than the standard single-reference SCF wavefunction. For this reason excited state computations were not routinely used by chemists. At the end of the 80's improved ab initio quantum chemical methodologies became available which were suitable for computing, in a balanced way, excited and ground state potential energy surfaces. In particular the ab initio Multiconfigurational Self-Consistent Field (MCSCF) method, developed by M. A. Robb in London, had an analytical gradient which could be employed for efficient geometry optimisation (the search for the structure corresponding to energy minima and transition states) taking into account the complete set of the 3N-6 nuclear degrees of freedom of the reacting system (N is the number of atoms). With this new methodology it was possible to overcome the limitations of the model proposed previously by Van der Lugt and Devaquet [31]. These limitations mainly regarded the computation of the excited state reaction path that was assumed to correspond to an interpolation between the reactant and the product geometrical structure. With the new tools one could determine real excited state reaction path where the reaction coordinates is not assumed but computed in a substantially unbiased way.
Conical Intersection
Interpolated Coordinate
Fig. 2. The relationship between the Van der Lugt - Oosterhoff model and the Conical Intersection (CI) model. The inset indicates the position of the Van der Lugt - Oosterhoff avoided crossing in the conical intersection region.
In Fig. 2 we show the relation between an avoided crossing and the double cone topology of a real conical intersection. In two dimensions, the Van der Lugt and Oosteroff model is refined by replacing the "avoided crossing" with an "unavoided crossing", i.e., a conical intersection (CI). In other words, the Van der Lugt and Oosteroff avoided crossing path R -> P is replaced by a path involving a real surface crossing R ->CI ->P. Bernardi, Olivucci, Robb and co-workers [62; 63] in 1990, reported a first application of the ab initio MCSCF method. This was a study of the photoinduced cycloaddition of two ethylene molecules and showed that: -
A conical intersection exists right at the bottom of the excited state energy surface. The molecular structure of the conical intersection is related to the observed photoproducts and stereochemistry of the reaction.
As discussed below, the original idea of Teller, Zimmerman Michl and Salem are now fully supported by the results of further computational work [64] which definitely demonstrates, when taken in conjunction with modern experimental results, that frequently radiationless deactivation occur via a conical intersection between excited and ground states. Radiationless decay at a conical intersection implies that: a) The internal conversion process may be 100% efficient (i.e. the Landau-Zener [65] decay probability will be unity)
b) Any observed retardation in the internal conversion or reaction rate (i.e. the competition with fluorescence) may reflect the presence of some excited state energy barrier which separates M* from the intersection structure and c) In the case where the decay leads to a chemical reaction, the molecular structure at the intersection must be related to the structure of the photoproducts. Points a-c provide the theoretical basis for the computational modelling of photochemical reactions. Between 1992 and 2002, a long-term computational project involving one of the authors has been carried out to prove the general validity of the hypothesis supporting the existence of low-lying conical intersections in organic molecules. The systematic search was performed on different classes of organic molecules with the intensive use of the MCSCF quantum chemical method. The application of powerful tools led to a detailed mapping of the potential energy surfaces of ca. 25 different organic chromophores, thus allowing the characterization of the conical intersections involved in the reaction mechanisms. The first result of such an extensive computational effort is that conical intersections may mediate all types of chemical events such as bond making, bond breaking, group exchange, intermolecular and intramolecular hydrogen transfer, charge transfer. The second outcome of the research is that conical intersections do not necessarily take part in a successful chemical reaction (i.e. reaction where the light energy is exploited to produce chemical species different from the reactants) but can also mediate light energy wastage mechanisms such as in quenching and internal conversion processes. Recently, an Si/So conical intersection has been characterized even in protein and solution environments using an hybrid Quantum Mechanics/Molecular Mechanics (QM/MM) strategy. Olivucci and co-workers [66; 67] located a 90°-twisted low-lying Si/So conical intersection in Rhodopsin and Bacteriorhodopsin using a CASPT2//CASSCF/AMBER level of theory. Toniolo et al [68] characterized the solution-phase conical intersections of the Green Fluorescent Protein (GFP) and Photoactive Yellow Protein (PYP) chromophores at semiempirical CAS/CI level. In conclusion, conical intersections could, contrary to common belief, be frequent (if not ubiquitous) in organic and bio-organic systems and, for many reaction, they constitute the photochemically relevant decay channel. The majority of the conical intersection structures documented for organic and bio-organic chromophores corresponds to low-lying conical intersections located at the bottom of the excited state relaxation path. As discussed in Chapter 6 of this book, these points could be only located using methods that allows for the computation of the so-called photochemical reaction path. A conical intersection is a point of crossing between two electronic states of the same spin multiplicity (most commonly singlet or triplet). Moreover, if we plot the energies of the two intersecting states against two specific internal coordinates Xi and X2 we obtain a typical double cone shape (see Fig. 3a). The Xi and X2 molecular modes define the so-called "branching" [61] or "g, h" [69] plane and the (n-2)-dimensional subspace of the n nuclear coordinates is called the intersection space, or seam of intersection [69], an hyperline
Ground Slalc PES ]
(a)
(b)
Fig. 3. (a) Representation of the typical double-cone topology for a conical intersection, (b) Relation between the "branching space" and the "intersection space".
consisting of an infinite number of conical intersection points (see Fig. 3b) and it is, locally, orthogonal to the two-dimensional branching plane. A molecular structure deformation along the branching plane lifts the Si/So energy degeneracy. Furthermore the ground and excited state wavefunctions undergo a dramatic change when the molecular structure is changed along a closed loop lying on the plane defined by the Xj and X2 modes and comprising the conical intersection. In particular, these wavefunctions exchange their character along the loop. The rest of this chapter contains some case studies involving the analysis of the branching plane structure and of the behaviour of the wavefunctions of the Si and So states in the region of the conical intersection. This will give a clear idea of the importance of such an analysis in providing information on the nature of the "reactive" process mediated by the such mechanistic entities. The results reported for each example have been produced using a common strategy based on the CASPT2//CASSCF level of theory. In this "mixed" computational method the full reaction coordinate is computed at the CASSCF level while the energy profile is computed at the CASPT2 level. This means that one applies multireference second-order perturbation correction to the CASSCF potential energy surface to incorporate dynamic correlation effects. For further details on these methods, the reader shall refer to Chapter 2 by Merchan and Serrano-Andres and, in part, to Chapter 10 by Roos. A discussion of the alternative and currently emerging Time-Dependent Density Functional Theory (TDDFT) methodology for computing the potential energy of electronically excited states will be given by Furche et al. in Chapter 3.
2.1 An example of a photoisomerization mediated by a conical intersection: the cis/trans isomerization in a rhodopsin chromophore model. The protonated Schiff base of retinal (PSB) is the chromophore of rhodopsin proteins. A light-induced cis->trans isomerization of the chromophore triggers the biological activity of rhodopsin, which, in turn, induces a conformational change in the protein. The detailed structure of the excited and ground state potential energy surfaces of the rhodopsin retinal chromophore model ?Z^-penta-3,5-dieniminium cation (CW-C5H6NH2) (1) and in particular the structure of the excited and ground state reaction path branches has been fully elucidated. Furthermore the reduced dimension of the model has allowed for the computations of abinitio CASSCF semi-classical trajectories and evaluation of the excited state lifetime and time scale of the photochemical isomerization. The results demonstrated that 1 provides a reasonable model for more realistic structures. In particular, the two-state two-mode nature of the reaction coordinate computed and observed (both in solution and in the protein) is maintained in the minimal model and the computed ultrafast excited state dynamics is still characterized by two different timescales corresponding to a very initial stretching relaxation (i.e. an inversion of the single bond/double bond positions) and to the following torsional deformation (about the central C2-C3 bond) respectively.
1 In Fig. 4 we plot the branching plane vectors (Xi and X2) at the conical intersection of 1. The conical intersection structure features one highly twisted double bond (about 92°) and involves two electronic configurations, an ionic and a covalent state, that differ for the transfer of one electron between the C5-C4-C3- and -C2-C1-N fragments.
10
Fig. 4. Branching (or g,h) plane vectors for the CA structure of 1. The Xi and X2 vectors correspond to the derivative coupling (or non-adiabatic coupling) and gradient difference vectors between the Si and So states
From the structure of the branching plane it is apparent that in this molecule Xi and X2 describe two types of processes. As shown in Scheme 1, motion along the Xi corresponds to a coupled pyramidalization (wagging) modes at the Ci and C4 centers of the it-chain. This motion allows for a widening of the C4-C3-C2-C1 dihedral angle leading to a it-bond breaking process. The X2 mode is characterized by a stretching deformation (a double bond expansion and single bond contraction mode) of the N=Ci-C2=C3-C4=C5 chain segment. Thus, motion along the X2 direction would ultimately yield two structures which may be represented by (resonance) formulas with inverted single and double bonds and with the positive charge shifted from the N-terminal to the Cs-terminal. These two 92° twisted structures will be less stable than the generated by motion along the wagging mode since the deformation along Xi allows for reconstitution of the central double bond providing strong coupling with the Z/E double bond isomerization coordinate. Thus, structural analysis of the branching plane suggests that upon decay from CI the molecule will generate the Z and E stereoisomers.
11
NHz
-•KES)
NH2
+{GS)
CI
+ Nhfe Scheme 1 The analysis of the wavefunction, taken together with the analysis of the branching plane, provides the basis for the rationalization of the electronic structure of the ground state energy surface comprising the reactant and product valleys (and, eventually, the transition structures connecting them). The result of such an analysis for chromophore 1 is shown in Scheme 2 where the wavefunction is analyzed in terms of point charges of the C5-C4-C3- and -C2-C1-N fragments along a loop centered on the CI and lying along the plane defined by the Xj and X2 modes. The charge distribution of the system demonstrates the existence of two different regions. The first region 0°< co < 30°, 200°< w < 360° is characterized by a structure where the charge is mainly localized on the N-terminal part of the molecule. The second region 30°< co < 200° is characterized by a structure where the positive charge is mainly located on the Cterminal part of the molecule. The border between the two regions corresponds to the electron transfer events between the two fragments. Notice that the wavefunction changes are associated with the two minima in the energy gap diagram.
12
/ degrees
Scheme 2 As we have previously underlined, the low-lying conical intersections could be only provided through the computation of the photochemical minimum energy path (MEP). However, some cases have been documented where excited state reaction path does not necessarily hit the lowest energy point belonging to the intersection space (IS) and the decay may not occur in this region. One of these cases regards the excited state relaxation path of the PSB chromophore 1. Indeed the mapping of the low-lying segments of the IS for this chromophore (see Fig. 5), by means of constrained MEP computations, demonstrated that it ends at a conical intersection with a ca. 70° (Cl7o°) twisted structure. The intersection space remains then coincident with the reaction path up to the lowest energy intersection (Cl92°) that has a 92° twisted structure [70]. Notice that in this situation the main locus of excited state (Si) decay is predicted to be Cl7o°. Semi-classical dynamics calculations on this model together with the calculations of the intersection seam connecting the So and Si potential energy surface spanning the entire range of twisting of the central double bond from 0° to 180° reveals the relationship between the excited-state reaction path and the intersection seam. The results show that motion along the reaction path is a good description of the photochemistry of the rhodopsin model. For a complete discussion about dynamical considerations on this example see Section 4 of Chapter 5 of this book. More in general, Chapter 5 and Chapter 7 provide an introduction to the use of non-adiabatic molecular dynamics to the investigation of photochemical reactions. Such studies bring the description of the reaction mechanism well beyond the photochemical reaction path picture discussed in this Chapter and in Chapter 6.
13
hv MaJorS,->S, daeay channel
Reaction Coordinate
Fig. 5. The excited state reaction path of the cation 1 intercepts the conical intersection point CI70" located ca 5 kcal mol"1 above the minimum energy conical intersection C I ^ . FC->Cl7o° The values of the relevant structural parameters are given in A and degrees. Redrawn with permission from reference [71] © 2004 World Scientific Publishing Co. Pte. Ltd.
The same model has been chosen to test the applicability of TDDFT method to excited state reactivity problems. Vertical excitation energies computed using the two different CASPT2//CASSCF and TDDFT//CASSCF treatments are in agreement. On the other hand, quantitative discrepancies are found along the reaction coordinate demonstrating that the quality of TDDFT must be further investigated especially with respect to the calculation of excited state reaction coordinate. In spite of these differences, TDDFT//CASSCF energy profile is found to describe the conical intersection region [72]. For a complete and detailed description of TDDFT performances in excitation energy and excited state structure computations see Chapter 3.
14 2.2 An example of photoaddition mediated by a conical intersection: a peryciclic reaction (ethylene + ethylene) The conical intersection seen in Fig. 4 for a protonated polyene Schiff base is an example of conical intersection between two electronic states, which are related by a charge transfer from a region of the molecule to another. Here, we show an example of conical intersection between two states, which does not differ for a different charge distribution but for a different spin distribution among the active orbitals. The Si/So conical intersection for the [2jts+2its] cycloaddition of two ethylene molecules is shown in Fig. 6a [62; 63]. The intersection is formed by two interacting olefinic fragment bound in a rigid rhomboidal structure (2.19 A interfragment distance and 110° rhomboidal distortion) with C2h symmetry. The branching plane vectors are shown in Fig. 6b. Motion along Xi lifts the degeneracy by changing the intermolecular C1-C4 interfragment distance whereas motion along X2 lifts the degeneracy by changing the angle of attack. Both motions are schematically illustrated in Scheme 3a where it is clear that the evolution along Xi leads to the reactant pair (-Xi) and cyclobutane (Xi). Alternatively, the evolution along X2 yields the tetramethylene biradical (-X2) or a rectangular saddle point (X2). The same prediction can be made analyzing the electronic structure of the conical intersection. Indeed, Si is described by a doubly excited state configuration, '(jt|2-jt32), involving excitation of two electrons from the ethylene-dimer HOMO (112) to the ethylenedimer LUMO (113) and can be identified as the combination of two ethylene molecules in their one-electron excited n-n* state. The electronic structure is tetraradical: the four unpaired electrons can recouple in different ways leading to the ethylene dimer reactant (the coupling is C1-C2, C3-C4), cyclobutane product (the coupling is C1-C3, C2-C4), and tetramethylene biradical (the coupling is C1-C4).
15
2 28 A
(a)
(b)
Fig. 6. (a) S|/So conical intersection for the [2jts+2res] cycloaddition of two ethylene molecules, (b) Branching (or g,h) plane vectors, X! and X2, for the CI structure in (a).
A combined Natural Bond Orbital (NBO) and wavefunction analysis along a small loop around the conical intersection shows that, in contrast to the conical intersection of a retinal model, there is no substantial variation in charge distributions. In this case, the analysis based on a representation in terms of localized 2-center spin orbitals shows that there is a change in the electronic distribution between aa-c3(C2-C4) and Jtci-c2(C3-C4) bonds. Note that even in this case the two dramatic wavefunction (i.e. electronic structure) changes correspond to the two minima in the Si-So energy difference diagram of Scheme 3b.
16
(a) Scheme 3 2.3 An example of photofragmentation mediated by a conical intersection: the photodenitrogenation of a bicyclic azoalkane. The photochemical denitrogenation of the 'n-it* 2,3-diazabicyclo[2.2.1]hept-2-ene (DBH) has been the subject of an intense investigation since the first report of Salomon[73] and coworkers in 1968 especially because of the unusual stereoselectivity observed during the nitrogen extrusion and formation of the housane. Upon thermal or photochemical excitation, in fact, DBH and its derivatives lose molecular nitrogen.
hv DBH
I
N
linear-axial Cf
exo-ax/a/ C,-/*e 1 DZ
17
Fig. 7. X[ and X2 vectors for the CI structure of DBH (showed in the inset) corresponding to two orthogonal bendings of the NNC angle.
The intersection structure that mediates the C-N a-cleavage is characterized by a linear-axial arrangement of the NNC fragment in which one of the two CN bond is still intact (1.48 A) [74; 75]. In Fig. 7 we plot the branching plane vectors Xi and X2 which correspond to two orthogonal bendings of the NNC angle. As illustrated in Scheme 4, after structural considerations, it is clear that the two CNN bending prompt the formation of ground state diazenyl diradical (*DZ in the Scheme above), either in the exo or endo or endo-exo forms.
18
•N A—1.18 A
(N.
exo
CI
N i/
N
endo-exo Scheme 4
The electronic structure of the intersecting states of the a-CN bond cleavage of DBH is shown in Scheme 5. The azoalkane Si state is described by a tetraradical configuration with one electron residing in each a-CN a-orbital, one electron residing in the excited nitrogen lone pair and one electron inside the p-orbital of the other nitrogen. The azoalkane So state is described by a biradical configuration in which the two a-CN a-orbitals are singly occupied.
Si1n-re* Scheme 5
19
Reaction path calculations indicate that the observed inversion of stereoselectivity has to be ascribed to the impulsive population of the vibrational mode that triggers an axial-toequatorial ring inversion. This idea is supported by classical trajectories calculation. In fact, as shown in Fig. 8, after a first oscillation (within 40 femtoseconds) in the direction of an unstable (transient) bicyclic intermediate, the molecule reaches a highly strained structure form which it can only relax following the initial direction of motion. After 60 femtoseconds the structures reaches the exo-axial 'DZ configuration and after 80 femtoseconds the axial to equatorial transition structure. The inverted configuration is then reached in 100 femtoseconds. Such ground state trajectory computation has been started from a point close to the conical intersection. Chapter 5 and 7, in this book, deals with the computations of photochemical (non-adiabatic) trajectories that start on the excited state energy surface and end on the ground state photoproduct minima.
El kcal mol
-25 120 Time (fs) Fig. 8. Triplet DFT energy profile along the trajectory computations started at a point closed to the CI of DBH. The structures document the molecular changes along the simulation. Redrawn with permission from reference [74] © 2003 American Chemical Society.
20
2.4 An example of charge transfer and an hydrogen transfer (intermolecular) processes mediated by a conical intersection: the fluorescence quenching of bicyclic azoalkanes. In contrast to DBH, there exist a different class of azoalkanes that are essentially inert to photochemical denitrogenation, the so-called "reluctant" azoalkanes. A representative system is the 23-diazabicyclo[2.2.2]oct-2-ene (DBO, see Scheme 6) which display exceedingly long singlet n,jt*-excited lifetimes (up to 1 |is).
hv
CHClj
Z5 or INEt,
Scheme 6 Using the 1,2-diazacyclopent-l-ene (pyrazoline) as a reduced model of DBO, allowing for the use of these accurate but expensive methods, it has been possible to demonstrate, in combination with the experimental evidence, that there are two basic mechanisms for the quenching of 'n,jt* states [76-79]. Indeed, DBO is efficiently quenched by hydrogen donors (such as non protic solvents like chloroform, methanol, benzene) via either a concerted or a stepwise process and by electron donors (such as amines like triethylamine) via a concerted process only. The computations have been carried out using CH2CI2 and methanol to model hydrogen donors, and trimethylamine and dimethyl ether as prototypes of strong and weak electron donor solvents, respectively. As shown in Fig. 9 both quenching routes involve bimolecular photochemical reactions (that is an electron or an hydrogen atom transfer photoreductions) and a full deactivation through a Si/So CI channel, which is located roughly halfway along the reaction coordinate and prompts a reaction path branching. The first branch (full arrows) is associated with an "aborted" chemical reaction. The second branch (light arrows) is associated to production of a very unstable transient species that may not accumulate but reverts to the original material by passage through a low-lying transition state located on a ground state reaction coordinate (dashed curve). Thus, for both routes a chemical transformation is initiated but it is not achieved.
21
Pyrawiline
•
CHjCl.orNfCHj),
quencher chromophorc Reacliun Cmirtlinale
Fig. 9. Potential energy diagram showing the interplay between ground (So) and excited (Si) state surfaces in the fluorescence quenching of n,jt* state chromophores due to an hydrogen donor or electron donor species. The full and light arrows describe the concerted and stepwise energy wastage route.
The conical intersection (see the ball and stick structure below) for the hydrogen abstraction mechanism (in the present of CH2G2) is characterized, with respect to the reactant pair, by a shortening of the H—N distance (from 2.22 to 1.02 A) and by a simultaneous expansion of the C—H distance (from 1.07 to 2.01 A).
The branching plane vectors are shown in Fig. 10. Motion along Xi lifts the degeneracy by stretching the intermolecular N-H distance whereas motion along X2 lifts the degeneracy by an out of plane pyrazoline ring distorsion. The evolution along both motions is schematically illustrated in Scheme 7. Distortion along Xi leads to production of a radical pair (RP) while distortion along -Xi leads to production of an unstable ion pair (IP). Evolution towards the X2 and -X2 directions leads to two equivalent ground state transition structures which feature a distorted pyrazoline ring.
22
Fig. 10. Xi and X2 vectors for the CI structure of pyrazoline + CH2C12 involved in the hydrogen abstraction mechanism of Fig. 9.
ci
}
O^ oi < 200°
0°< w c 100° * N
H
290°< w < 0°
CI
CI
IP
i
RP
-x,
/
I* Scheme 7
200°* w < 290'
23
The result of the wavefunction analysis indicate that the n,jt*-excited state correlates with a radical pair structure (derived from a complete hydrogen atom abstraction), and the ground state correlates with an ion pair (derived from proton abstraction) (see Fig. 1 la). Similarly to retinal PSB models, the configurations that describe the intersecting states are interchanged by a charge transfer (intermolecular instead of intramolecular). Indeed, the charge distribution of the system (see Fig. 1 lb) obtained computing fragment charges along a small loop lying along the plane defined by the Xi and X2 modes and centred around CI demonstrated that exist two different region. The first region (0°< co< 100° and 290°< co< 360°) is characterized by an ion pair structure where the charge on the pyrazoline is positive while on CHCI2 is negative. The second region 100°< co< 290° is characterized by a covalent structure. The border between the two regions corresponds to two sudden electron transfer events, one from the CHCI2 anion to the pyrazoline cation yielding the RP configuration (co = 90°) and the second in opposite direction (00 = 270°).
(a)
(b)
0
60
120
180
240
300
360
CO / degrees
Fig. 11. (a) Modified state correlation diagram of the n,it*-excited state (ES) of pyrazoline + CH2C12 correlating with the radical pair (RP) derived from hydrogen atom abstraction and the ground state (GS) correlating with the ion pair (IP) derived from proton abstraction, (b) So fragment charges [a.u.] along a loop centered around the CI (pyrazoline fragment, open triangles; hydrogen atom, open squares; CHCI2 fragment, open circles). Redrawn with permission from reference [77] © 2001 WileyVCH Verlag GmbH.
24
The geometrical structure of the conical intersection and the vectors of the branching plane for the charge transfer process (i.e. the present of trimethylamine) are given in Fig. 12. Notice that Xi is dominated by the out of plane deformation of the pyrazoline ring and X2 is dominated by the interfragment distance. The computed photochemical reaction path demonstrates that the excited state branch of the path is dominated by the decrease in distance between the pyrazoline and N(CH3)3 fragment. After a small excited state barrier, the progression along the path leads to the formation of an exciplex located in the close vicinity of a conical intersection. The intersection is accessed when the distance between the pyrazoline and amine nitrogen atoms is ca. 2 A. At the exciplex the computed amount of charge transfer from the trimethylamine lone pair to the excited state half-vacant nonbonding orbital of one pyrazoline nitrogen atom is 0.3 electrons.
Fig. 12. X[ and X2 vectors for the CI structure of pyrazoline + trimethylamine involved in the charge transfer process of Fig. 9.
25
Q
O
N CH
... (
3)3
.
N CH
( 3)3
CI
Interfragment Distance
Fig. 13. Modified correlation diagram for the interaction of the n,jt*-excited state of pyrazoline with an electron donor such as the trimethylamine reflecting the occurrence of an exciplexes and a conical intersection along the reaction pathway. Redrawn with permission from reference [76] © 2000 WileyVCH Verlag GmbH.
The exciplex state N...N two-orbital/three-electron bond can be viewed as a mixture of a covalent (N=N*---:NMe3) and an ionic (N=N*~---+*NMe3) electronic configuration. The steep rise of the ground state energy surface toward the conical intersection is due to a destabilizing two-orbital/four-electron repulsive interaction (N=N5--- :NMe3) (see Fig. 13).
2.5 An example of stereoselective photochemical reaction mediated by conical intersection: a Norrish Yang photocyclization. The hydrogen abstraction process mediated by a conical intersection, and documented above for pyrazoline, can also be found for structure 1 (see Scheme 8). Here the hydrogen atom transfer is intramolecular rather than intermolecular. In this structure a carbonyl function replaces the azo function (-N=N-) of the n^it* chromophore. Structure 1 models an alanine derivative, which undergoes, upon photoexcitation, a Norrish-Yang photocyclization reaction. Minimum energy path calculations demonstrated that this is an example of 'n,jt* photochemical reaction displaying a chiral memory effect [80] in agreement with experimental results [81].
hv
26 Scheme 8
Intermediate
Z5
.,«OH
r
Reaction Coordinate
Intermediate
Fig. 14. Potential energy diagram showing the ground (So) and excited (Si) state potential energy surfaces of the Norrish-Yang photocyclization of 1.
In fact, as illustrated in Fig. 14, the computed reaction coordinate leads to a CI displaying an incomplete hydrogen atom transfer to the 'n-jt* carbonyl oxygen. Once again, there are two ground state relaxation paths that develop from the intersection. Accordingly, while the first path leads back to the starting material (light arrows), the other path (full arrows) leads to a diradical species (intermediate). This is also demonstrated by plotting the branching plane vectors, which are dominated by the transfer of the H atom. Most important, no torsional components which can lead to a loss of stereochemistry are present (Fig. 15). However, in contrast to the pyrazoline (i.e. DBO) quenching, the intermediate structure does not easily revert to the starting material but is the precursor of the stereospecific product 2.
27
Fig. 15. Branching (or g,h) plane vectors (Xi and X2) for the CI structure of Structure 1. This model of an alanine derivative undergoes, upon photoexcitation, a Norrish-Yang photocyclization reaction.
2.6 Towards Computational Photobiology: the Rhodopsin Proteins The recent implementation of a QM/MM computational method based on the use of an ab initio CASPT2//CASSCF/6-31G* strategy (i.e. geometry optimization at the CASSCF level and energy evaluation at the CASPT2 level) coupled with a protein force field such as AMBER (or CHARMM) paved the way for excited state computations and conical intersection search in proteins (e.g. in rhodopsins, fluorescent proteins and others). This new appealing method has been applied to study the spectroscopy of two photoreactive proteins: the visual pigment Rhodopsin (Rh) and the Green Fluorescent Protein (GFP). The results demonstrate that the method is capable to provide a qualitatively correct description of the geometrical and electronic structure of the protein chromophores and their Si/So energy gap (the absorption and emission (for GFP) maxima) within a <40 nm error. The model used for the Rh computations is shown in Fig. 16a and the QM/MM method was based on a carefully parameterized hydrogen link-atom scheme [82; 83] with the frontier placed at the C6-CE bond of the Lys296 side chain. Using this strategy, it has been possible to map the lowest-lying (one dimensional) IS segment in Rh [66] whose coordinate is dominated by the torsional mode describing the photoisomerization of the 11 -cis isomer of the retinal chromophore (PSB11) to its all-trans isomer (PSBT) (Scheme 9).
28
_,. H!N
PSB11
14V 12
PSBT
Scheme 9
The IS corresponds to an energy plateau characterized by a 80°-l 10° change in the Cio-CnC12-C13 torsion with the 90° twisted intersection point corresponding both to the lowestenergy IS point and Si absolute energy minimum. The branching plane vectors (Xi and X2) computed at the 90° point is shown in Fig. 16. It is characterized by the combination of a stretching deformation (a coupled double-bond expansion and single-bond compression mode) and a coupled wagging mode (simultaneous pyramidalization at the CIO and C13 centers of the it-chain) of the PSB11 skeleton that are, despite the effect of the protein environment exactly the same vectors seen for model 1 of section 2.1. As we have already discussed for the minimal model in Section 2.1, the evolution along the X2 and -X2 directions would lead to two structures characterized by inverted single and double bonds, whereas evolution along Xi and -Xi would lead to two structures with Z and E central double bond. Thus, even in this case the analysis of the branching plane suggests that upon decay from CI the molecule will generate the Z and E stereoisomers.
29
(a)
(b)
Fig. 16. (a) The 1 \-cis retinal chromophore (PSB11) of the visual pigment rhodopsin (Rh) bounded to a lysine residue (Lys296) via a protonated Schiff base linkage (see green substructure), (b) Branching space vectors computed at the lowest-energy CI of Rh (left). Vector Xi corresponds to the gradient difference while X2 corresponds to the derivative coupling. Redrawn with permission from reference [66] © 2004 The Royal Society of Chemistry
3. CONCLUSIONS Above we have tried to give a (partial!) view of the potential of computational photochemistry: a section of computational chemistry devoted to the investigation of light induced chemical reactions. We have centered the discussion on the concept of photochemical reaction path concept and on its use in mechanistic investigations. As shown in Scheme 10, the description of a photochemical event through the photochemical reaction path may be seen as an intermediated step towards the complete and rigorous simulation of the reaction. Nevertheless, it represents a presently established concept that, on one hand, allows for comparison with the available experimental data and, on the other hand, may be used to formulate general explanations and theories. Photochemical reaction paths also provide the bases for the future development of the field. Firstly one is seeking for more accurate and less expensive ab initio or TDDFT methodologies allowing more quantitative reaction path computations and, in turn, a more direct comparison with spectral and reactivity data (top-left of Scheme 10). Secondly, one needs to develop a robust and accurate description of the
30 reaction that goes beyond the static description offered by paths. In particular, one wants to carry out robust non-adiabatic (either semi-classical or quantum) molecular dynamics simulations to allow, for instance, for the evaluation of quantum yields and reaction timescales at a level that comparison with the experiments is possible (top-right). Finally, it is important to stress, that the calculation of photochemical reaction paths has been attempted for only few cases dealing with organometallics, solvated molecules and biological macromolecules. Both more efficient computational strategies and quantum mechanics/molecular mechanics hybrid methods must be developed to fulfil this important task (bottom-left). In the following chapters the authors attempt to provide an overview of the status of the knowledge in such diverse directions.
Better comparison with the experiment
Belter description of the mechanism
Simulation of Electronic Spectra: see for instance Chapters 4 and 9
T
Quantum Chemical Methods for Excited States: see for instance Chapter 2, 3 and It)
Exploratory Theoretical/Computational Work: see for instance Chapter 8
Non-Adiabatic Dynamics see for instance Chapters 5 and 7
Photochemical Reaction Path: see for instance Chapter 1 and 6
Photochemistry of Organometallic Compounds: see for instance Chapter 9 Photochemistry in Proteins. Solution and Sttpramolecular Compounds: sec for instance Chapter 1 and 7 Extension of the applicability
Scheme 10
31 ACKNOWLEDGMENT Funds have been provided by the Universita di Siena (Progetto di Ateneo 02/04) and HFSP (RG 0229/2000-M). We thank CINECA for granted calculation time. REFERENCES [I] [2] [3] [4] [5] [6] [7] [8] [9] [10] [II] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38]
W. J. Hehre, R. F. Steward, J. Pople, J. Chem. Phys., 51 (1969) 2657. J. A. Pople, Angew. Chem. Int. Ed., 38 (1999) 1894. R. Pariser, R. G. Parr, J. Chem. Phys., 21 (1953) 466. R. Pariser, R. G. Parr, J. Chem. Phys., 21 (1953) 767. R. G. Parr, R. Pariser, J. Chem. Phys., 23 (1955) 711. R. Pariser, J. Chem. Phys., 25 (1956) 1112. R. Pariser, J. Chem. Phys., 24 (1956) 250. L. Cerruti, E. Riuniti (eds.), (2003). H. E. Zimmerman, J. Am. Chem. Soc., 88 (1966) 1566. H. E. Zimmerman, Science, 153 (1966) 837. H. E. Zimmerman, J. Am. Chem. Soc, 88 (1966) 1564. H. E. Zimmerman, Ace. Chem. Res., 4 (1971) 272. H. E. Zimmerman, Ace. Chem. Res., 5 (1972) 393. H. E. Zimmerman, Ace. Chem. Res., 15 (1982) 312. H. E. Zimmerman, R. C. Hahn, H. Morrison, M. C. Wani, J. Am. Chem. Soc., 87 (1965) 1138. H. E. Zimmerman, D. I. Schuster, J. Am. Chem. Soc, 83 (1961) 4486. B. Bigot, A. Devaquet, N. J. Turro, J. Am. Chem. Soc, 103 (1981)6. V. Bonacic-Koutecky, J. Kohler, J. Michl, Chem. Phys. Lett., 104 (1984) 440. V. Bonacic-Koutecky, J. Koutecky, J. Michl, Angew. Chem. Int. Ed., 26 (1987) 170. V. Bonacic-Koutecky, K. Schoffel, J. Michl, J. Am. Chem. Soc, 111 (1989) 6140. W. G. Dauben, L. Salem, N. J. Turro, Ace Chem. Res., 8 (1975) 41. R. C. Dougherty, J. Am. Chem. Soc, 93 (1971) 7187. N. D. Epiotis, S. Shaik, J. Am. Chem. Soc, 100 (1978) 1. N. D. Epiotis, S. Shaik, J. Am. Chem. Soc, 100 (1978) 9. N. D. Epiotis, S. Shaik, J. Am. Chem. Soc, 100 (1978) 18. N. D. Epiotis, S. Shaik, J. Am. Chem. Soc, 100 (1978) 29. T. Forster, Pure Appl. Chem., 24 (1970) 443. T. Forster, Pure Appl. Chem., 34 (1973) 225. W. Gerhartz, R. D. Poshusta, J. Michl, J. Am. Chem. Soc, 98 (1976) 6427. W. Gerhartz, R. D. Poshusta, J. Michl, J. Am. Chem. Soc, 99 (1977) 4263. D. Grimbert, G. Segal, A. Devaquet, J. Am. Chem. Soc, 97 (1975) 6629. J. Michl, Mol. Photochem., 4 (1972) 243. J. Michl, Topics Curr. Chem., 46 (1974) 1. J. Michl, V. Bonacic-Koutecky, Electronic Aspects of Organic Photochemistry, J. Wiley and Sons, INc, New York, (1990). L. Salem, J. Am. Chem. Soc, 96 (1974) 3486. L. Salem, W. G. Dauben, N. J. Turro, J. Chim. Phys., 70 (1973) 694. L. Salem, C. Leforestier, G. Segal, R. Wetmore, J. Am. Chem. Soc, 97 (1975) 479. L. Salem, C. Rowland, Angew. Chem. Int. Ed., 11 (1972) 92.
32 [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71]
[72] [73] [74] [75] [76] [77] [78] [79]
N. J. Turro, J. McVey, V. Ramamurthy, P. Lechtken, Angew. Chem. Int. Ed. Engl., 18 (1979) 572. W. T. A. M. Van der Lugt, L. J. Oosteroff, J. Am. Chem. Soc, 91 (1969) 6042. N. J. Turro, J. Photochem. Photobiol., A: Chemistry, 51 (1990) 63. R. B. Woodward, R. Hoffmann, Angew. Chem. Int. Ed. Engl., 8 (1969) 781. G. Herzberg, H. C. Longuette-Higgins, Discuss. Faraday. Soc, 35 (1963) 77. H. C. Longuet-Higgins, E. W. Abrahamson, J. Am. Chem. Soc, 87 (1965) 2045. H. C. Longuette-Higgins, Proc R. Soc. Lond. A, 344 (1975) 147. J. R. Lakowicz, Principles of Fluorescence Spectroscopy, Kluwer Academic/Plenum Publishers, New York, Boston, Dordrecht, London, Moscow, (1999). A. Gilbert, J. Baggott, (eds.), Essential of Molecular Photochemistry, Blackwell Scientific Publication, Oxford, (1991). M. O. Trulson, R. A. Mathies, J. Phys. Chem., 94 (1990) 5741. P. J. Reid, S. J. Doig, S. D. Wickham, R. A. Mathies, J. Am. Chem. Soc, 115 (1993) 4754. S. Pullen, L. A. W. II, B. Donovan, R. J. Sension, Chem. Phys. Lett., 242 (1995) 415. D. R. Cyr, C. C. Hayden, J. Chem. Phys., 104 (1996) 771. H. Kandori, Y. Katsuta, M. Ito, H. Sasabe, J. Am. Chem. Soc, 117 (1995) 2669. Q. Wang, R. W. Schoenlein, L. A. Peteanu, R. A. Mathies, C. V. Shank, Science, 266 (1994) 422. E. Teller, J. Phys. Chem., 41 (1937) 109. E. Teller, Israel Journal of Chemistry, 7 (1969) 227. N. J. Turro, J. Mol. Struct. (Theochem), 424 (1998) 77. V. Bonacic-Koutecky, J. Michl, Teor. Chim. Acta, 68 (1985) 45. W. T. Borden, (eds.), Diradicals, Wiley, New York, (1982). D. R. Yarkony, Rev. Mod. Phys., 68 (1996) 985. D. R. Yarkony, J. Chem. Phys., 92 (1990) 2457. G. J. Atchity, S. S. Xantheas, K. J. Ruendeberg, J. Chem. Phys., 95 (1991) 1862. F. Bernardi, S. De, M. Olivucci, M. A. Robb, J. Am. Chem. Soc, 112 (1990) 1737. F. Bernardi, M. Olivucci, M. A. Robb, Ace Chem. Res., 23 (1990) 405. F. Bernardi, M. Olivucci, M. A. Robb, Chem. Soc. Rev., 25 (1996) 321. M. Desouter-Lecomte, J. C. Lorquet, J. Chem. Phys., 71 (1979) 4391. A. Migani, A. Sinicropi, N. Ferre, A. Cembran, M. Garavelli, M. Olivucci, Faraday Discussion, 127(2004) 179. T. Andruniow, N. Ferre, M. Olivucci, PNAS USA, 101 (2004) 17908. A. Toniolo, G. Granucci, J. T. Martinez, J. Phys. Chem. A, 107 (2003) 3822. D. R. Yarkony, J. Phys. Chem. A, 105 (2001) 6277. A. Migani, M. A. Robb, M. Olivucci, J. Am. Chem. Soc, 125 (2003) 2804. A. Migani, M. Olivucci, in "Conical Intersections: Electronic Structure, Dynamics and Spectroscopy" (Eds.: W. Domcke, D. R. Yarkony, H. Koppel), World Scientific, Singapore, 2004. S. Fantacci, A. Migani, M. Olivucci, J. Phys. Chem. A, 108 (2004) 1208. B. S. Solomon, T. F. Thomas, C. Steel, J. Am. Chem. Soc, 90 (1968) 2249. A. Sinicropi, C. S. Page, W. Adam, M. Olivucci, J. Am. Chem. Soc, 125 (2003) 10945. N. Yamamoto, M. Olivucci, P. Celani, F. Bernardi, M. A. Robb, J. Am. Chem. Soc, 120 (1998)2391. A. Sinicropi, U. Pischel, R. Basosi, W. M. Nau, M. Olivucci, Angew. Chem. Int. Ed., 39 (2000) 4582. A. Sinicropi, R. Pogni, R. Basosi, M. A. Robb, G. Gramlich, W. M. Nau, M. Olivucci, Angew. Chem. Int. Ed., 40 (2001) 4185. A. Sinicropi, W. M. Nau, M. Olivucci, Photochemical & Photobiological Sciences, 1 (2002) 537. A. Sinicropi, Ph. D. thesis, University of Siena, (2002).
33 [80] [81] [82] [83]
A. Sinicropi, F. Barbosa, R. Basosi, B. Giese, M. Olivucci, Angew. Chem. Int. Ed., 44 (2005) 2390-2393 B. Giese, P. Wettstein, C. Stahelin, F. Barbosa, M. Neuburger, M. Zehnder, P. Wessig, Angew. Chem. Int. Ed. Engl., 17 (1999) 2586. N. Ferre, M. Olivucci, Theochem, 632 (2003) 71. N. Ferre, M. Olivucci, J. Am. Chem. Soc, 125 (2003) i
This Page is Intentionally Left Blank
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 Elsevier B .V. All rights reserved
35
II. Ab Initio Methods for Excited States Manuela Merchan and Luis Serrano-Andres Departamento de Quimica Fisica Instituto de Ciencia Molecular Universitat de Valencia Dr. Moliner 50, Burjassot ES-46100 Valencia, Spain Tel. 34 963543155/34 963544333 Fax. 34 963543156 Email:
[email protected],
[email protected] 1. INTRODUCTION From the perspective of a chemist all sort of matter is essentially composed by a few types of elementary particles that can be combined in different ways. These particles do not follow the laws of classic mechanics but behave according to the laws of quantum mechanics. They present certain features, such as symmetry laws and exchange phenomena, without correspondence in a Newtonian world, which have to be taken into account theoretically. Constitution of matter is, therefore, a quantum-chemical problem involving many particles. Knowledge on ab initio grounds of the true solutions for the full non-relativistic timeindependent Schrodinger equation of molecules, within the Born-Oppenheimer approximation, has been considered as one of the Grand Challenge problems in science since the birth of quantum mechanics at the beginning of the twentieth century. The term "ab initio" is Latin and the English meaning is "from the start", that is, from the first principles, implying that no parametrization at all is employed. Unfortunately, the Schrodinger equation for a molecule, except for small systems, cannot be exactly solved at present and we are forced to look for appropriate algorithms to obtain approximate solutions. Within the framework of a particular technique (variation principle, perturbation theory, or other schemes), the procedure can be still performed at the ab initio level. With the need of methodological development, the discipline of Quantum Chemistry emerged and, in order to perform the applications of interest, a large number of approximate quantum-chemical methods is currently available. The main ideas for many of those methods come from the earliest methodological attempts but significant new algorithms have been developed and implemented into efficient software in the last decade. Chemists have been some of the most active and innovative participants in the rapid expansion of computational science. Computational chemistry can be regarded as the
36 application of chemical, mathematical, and computing skills to the solutions of chemical problems. Obtaining approximate solutions to the Schrodinger equation is the basis for most of the computational chemistry performed today. The quantum-chemical applications performed serve many times as source of inspiration for new methodological developments. In what follows we shall consider reliable ab initio methods, those that offer reasonable answers for well-defined chemical problems. In practice, it is usually difficult to find a given method that can be applied to the successful calculation of many distinct chemical properties. Consequently, it is of major importance the selection of the proper method to be employed in the computation of a given molecular property. In the present chapter, we shall focus our attention mainly on the performance of ab initio methods for the description of spectroscopic molecular properties of compounds. The material presented is probably biased to our own work, but a fair coverage or other viewpoints can also be found. 2. GENERAL OVERVIEW Most of the quantum-chemical methods developed up to date have been based on the concept of the one-electron wave function. The electronic states of a system with N electrons can be described by a double expansion. Molecular orbitals (MOs) are one-electron wave functions expressed as linear combinations of a known one-electron basis set {K} and the Nelectron wave function is formulated in a many-electron basis set formed by determinants (or linear combination of them to form spin-adapted wave functions), built as normalized antisymmetric products of MOs. In principle, if the one-electron basis set {K} is complete, and a complete many-electron basis set can be generated by considering all possible occupations for the corresponding MOs, the true solution of the Schrodinger equation can be achieved. Such a computation is not possible technically in most cases and in actual applications the one-electron basis set has somehow to be truncated. Nevertheless, when all the N-electron wave functions are taken into account, the calculation is named full configuration interaction (FCI) and the corresponding eigenvalues and eigenvectors computed are exact within the space spanned by the finite basis set. Despite the great advances in FCI technology in the last few years, the size of the eigenvalue problem becomes rapidly too large to be handled by modern computers. As a result, FCI solutions are only available for relatively small molecular systems. We have, unfortunately, to land in the field of truncations, in both the one- and many-electron basis sets. Truncations performed in the one-electron basis sets together with the limitations introduced in the many-electron basis sets, which are normally truncated at a given degree of excitation (considering up to singly, doubly, triply, ... excited determinants) are the most important source of inaccuracies in the quantum-chemical calculations. The type of truncations carried out in conjunction with the class of techniques employed (variation principle, perturbation theory, and others) characterizes most of the methods currently employed through the available commercial software. Since the ground state of a large number of molecules at the equilibrium geometry is well described qualitatively by a single electronic configuration, it is not surprising that great
37 efforts have been devoted in the development of treatments such as Moller-Plesset perturbation theory (MP2, MP3, MP4), singles and doubles configuration interaction (CISD), and related non-variational approaches like coupled-electron pair approximation (CEPA), as well as coupled-cluster (CC) methods, in which the starting point is the Hartree-Fock (HF) wave function [1-4]. In contrast, the situation is quite different for the description of electronically excited states, which normally have several configurations equally relevant. The same may occur in certain regions of the ground-state hypersurface, far away from the equilibrium structure, for instance, in a transition state (TS) or in the dissociation limit of a homolitic breaking process of a covalent bond. In order to gather satisfactory results, one has to supply a wave function bearing enough flexibility to treat the required number of configurations on an equal footing. The goal can be nicely accomplished by the multiconfigurational self-consistent field (MCSCF) approach. For just a single configuration, it is equivalent to the MO model most commonly used in quantum chemistry: the HF SCF procedure. The complete active space SCF (CASSCF) is a variant of the MCSCF method that has become particularly popular because of its technical and conceptual simplicity. In the CASSCF method the active electrons are distributed among the active orbitals in all possible ways consistent with a given spatial and spin symmetry of the electronic state. The number and nature of the active orbitals and electrons are decided by the user. Normally, it is a crucial step for the successful performance of the approach, which must be guided by a deep knowledge of the chemical process under consideration in order to offer the required flexibility. It is not a question of chemical intuition but of chemical knowledge. Nothing easier than getting meaningless CASSCF results if the active space is meaningless for a given application. It is worth mentioning at this point that the ultimate responsibility for the selection and use of a given method relies on the user. For this purpose, calibration calculations are often enlightening. At the CASSCF level one usually takes into account longrange effects related to the so-called non-dynamic (static) correlation effects, making it possible the proper treatment of several nearly degenerate configurations. The remaining electron correlation effects, associated with the instantaneous short-range electron-electron interaction, can be accomplished by using variational methods like muti-reference CI (MRCI) [5-9] or employing perturbation theory by means, for instance, of the CASPT2 method (complete active space perturbation theory to second order) [10-12] or other related multireference perturbation theory (MRPT) schemes [13, 14]. According to the number of configurations considered initially, the methods can be classified in the following two categories: •
Single-configuration methods. They are typically based in the HF reference, which determines the MOs. The electron correlation treatment is usually performed at the CI, CC or MP levels. The coupled-cluster methods with singly and doubly configurations including the effect of triple excitations by perturbation theory CCSD(T), as well as related approaches, may yield accurate results. In general, the applicability of the methods in this group is restricted to situations where a single reference wave function is adequate for the description of a chemical process.
38 •
Multiconfigurational methods. Part of the electronic correlation is already included in the reference wave function, normally by using a MCSCF wave function, which determines a set of MOs. The remaining electron correlation effects are accounted for by MRCI, MRCC or MRPT techniques. They have a more ample range of applicability (ground state, excited states, TS, ...).
The accurate CASSCF/MRC1 protocol is computationally demanding, and quite often is not tractable, because a satisfactory selection of the active space requires huge technical resources. On the other hand, since a decade ago the multiconfigurational second-order perturbation theory CASPT2 has shown to be an efficient alternative, yielding an advantageous rate between the quality and the computational cost for the description of excited states in systems of relatively large molecular size [15-18]. Apart from MRCI, CASPT2 and other different MRPT algorithms, a quick inspection of the recent literature on excited states reveals that the following methods are also used quite often: Cl-singles (CIS) [19], Random-Phase Approximation (RPA) and related approaches [20], as well as coupledcluster based methods, Symmetry-Adapted Cluster CI (SAC-CI) [21], Equation-of-Motion CC (EOM-CC) [22], and linear response CCn [23]. They shall be reviewed in Section 4 from a practical point of view, making special emphasis on the expected advantages, disadvantages, and applicability in the qualitative/quantitative understanding of the electronic states. In addition, the performance of the time-dependent density functional (TD-DFT) approach, which is becoming widely used for the treatment of excited states, shall also be discussed. Because of the primordial role that electronic correlation plays in the relative placement of electronic states, the essentials shall be considered separately in the next section. 3. ELECTRON CORRELATION IN MOLECULES The extended treatment of electron correlation has traditionally been the bottleneck to achieve accurate results for excited states. Therefore, let us consider in this section the meaning of electron correlation in molecules from different perspectives. The major goal in quantum-chemical methodology for a molecular system formed by N electrons (i, j , ..) and M nuclei (A, B, ..) is finding reliable approximate solutions of the stationary electronic states as solutions of the Schrodinger equation H0 = s®
(1)
where the electronic wave function
depends explicitly on the N electronic coordinates xi, X2, X3,...XN and parametrically on the nuclear coordinates, within the well-known BornOppenheimer approximation. The three spatial coordinates rj and the one spin coordinate co; are denoted collectively by Xj. In atomic units (au), the electronic Hamiltonian operator is
39 The first term in Eq. (2) describes the kinetic energy of the electrons; the second term represents the Coulomb attraction between electrons and nuclei; the third term corresponds to the repulsion between electrons. The total energy for fixed nuclei includes the nuclear repulsion, a constant at a given geometry, (3)
and provides a potential for the nuclear motion. Eq. (1) constitutes the electronic problem, which has been during decades of major concern in methodological developments. The physics of electron correlation is hidden in the Hamiltonian itself. The Coulomb repulsion given by the term rj 1 , the inverse distance between two electrons, increases enormously in the regions close to rn = 0, preventing that two electrons may occupy the same space. Therefore, the motion of any two electrons is not independent but it is correlated. The phenomenon is known as electron correlation. Moreover, the statement that two electrons are correlated is equivalent to express that the probability of finding two electrons at the same point in space is zero. The instantaneous position of electron ; forms the centre of a region that electron j will avoid. For this reason, it is stated that each electron, as described by the exact wave function , is surrounded by a Coulomb hole. However, electron correlation is not taken into account properly by many approximate methods. The effect of neglecting electron correlation partly in approximate quantum-chemical approaches has great impact in the molecular spectroscopic properties of interest (computed transition energy, nature of the electronically excited states, related oscillator strengths, etc). The simplest wave function to describe a many-electron system is a Slater determinant built by orthogonal one-electron wave functions. Electrons are fermions and accordingly they have to be described by an antisymmetric wave function. For an N-electron system the Slater determinant has the form (x,)
X 2 (x,)
X,(x2)
X2O2) (4)
A,]\AN/
A-2VAN/
The constant (N!)"1/2 is a normalization factor. The wave function for an electron that describes both the spatial distribution and its spin is called spin orbital, Xi(xj). Since the Hamiltonian employed does not depend on the electronic spin (see Eq. (2)), each spin orbital can be expressed by multiplying the spatial orbital, \|/j(r0> by the spin function, r|(ot)i) (5)
40
A complete set for describing the spin of an electron consists of two orthogonal functions a(
(6)
A single-determinant wave function has several interesting properties. Firstly, it is worth noting that spin orbitals must be linearly independent, otherwise the value of the determinant is zero. It is obvious that interchanging two rows of the Slater determinant, which is equivalent to interchanging the coordinates of two electrons, changes the sign of the determinant. The requirement of the antisymmetry principle is automatically fulfilled. Having two columns of the determinant identical, that is, two electrons occupying the same spin orbital, makes the determinant zero. Thus, no more than one electron can occupy a spin orbital (Pauli exclusion principle). When a linear transformation of the set {xi} is carried out, (7)
where Aj, is an element of the matrix A of dimension NxN, with a value for its determinant, det(A), different from zero, then r = det(A) T
(8)
The wave functions *F' and ^ differ just in a constant and, therefore, represent the same physical situation. Since the set of spin orbitals is linearly independent, we can always choose a transformation matrix A so that the resulting spin orbitals %\ become orthonormal. Therefore, no restriction at all is imposed when we choose from the beginning an orthonormal set of spin orbitals. It just makes the computation of the Hamiltonian matrix elements involving Slater determinants easier. A Slater determinant is completely specified by the spin orbitals used to build it and any unitary transformation of them is equally valid. Two sets of spin orbitals related by a unitary transformation (A1 = A~'), which keeps the orthonormality of the spin orbitals, yield the same Slater determinant (see Eq. (8)). Slater determinants formed from orthonormal spin orbitals are normalized and N-electron Slater determinants that have different spin orbitals are orthogonal. The Slater determinant fulfils the basic symmetry law derived from the identity principle, because it describes N electrons occupying N spin orbitals (/, %2 •••%N) without specifying which electron is in each orbital. From a physical viewpoint, the use of a Slater determinant
41 wave function to describe a many-electron system implies that we are immersed in a model of independent electrons, where the electrons are not correlated, and the Coulomb hole is discarded. Nevertheless, it can be easily demonstrated that in a Slater determinant the motion of two electrons with the same spin function is correlated, that is, the probability of finding two electrons with parallel spins at the same point in the space is zero, the so-called exchange correlation, which is incorporated by the antisymmetric condition of the wave function for fermions. The phenomenon is known as the Fermi hole. We are, therefore, facing a model of independent particles where the behavior of certain electrons is not fully independent, because the Fermi hole simulates somehow the Coulomb hole. Since the motion of electrons with different spin function remains uncorrelated (there is a finite probability of finding two electrons with opposite spins at the same point in space), a single determinant wave function is commonly referred as an uncorrelated wave function. The Hartree-Fock approximation usually constitutes the first step towards more accurate approximations and has played a crucial role in elucidating modern chemistry. Indeed, many of the quantum-chemical methods can be considered either as simplifications of the HF method or going beyond it. The HF method provides the mathematical tools to obtain the unknown spin orbitals to build the best Slater determinant by making use of the variation principle. Let us consider a single Slater determinant to describe the ground state of an Nelectron system
The variation principle states that the best wave function of this functional form (single determinant type) is the one giving the lowest energy E 0 =(«F 0 |H|«F 0 )
(10)
where H is the electronic Hamiltonian. By minimizing Eo with respect to the choice of spin orbitals one can arrive to the Hartree-Fock conditions, which can be expressed in many different manners, and in particular the canonical expression takes the form ft.=e.X.
a=l...N
(11)
where f is an effective one-electron operator, called the Fock operator, which actually depends on its eigenfunctions. Thus, the HF equation (11) is not linear and must be solved iteratively through the SCF method. The Fock operator is the sum of a core-Hamiltonian operator and an effective one-electron potential operator. The former is the kinetic energy and potential energy for attraction to the nuclei and the latter is the average potential experienced by the electron described by the occupied spin orbital Xa^ue to the presence of the remaining N-l electrons. The solution of the HF eigenvalue problem, Eq. (11), yields a set of orthonormal canonical spin orbitals {%m} with orbital energies {sm}. The N spin orbitals with
42
the lowest energies are called occupied orbitals and span the Fock space. The remaining MOs, the virtual orbitals, span the complementary Fock space. Any unitary transformation within the Fock subspace leaves the HF energy invariant, Eq. (10). Transformations among the virtual orbitals can also be performed as long as the subspace spanned by the virtual spin orbitals remains orthogonal to the Fock subspace. They are particularly useful to improve (localize) the virtual MOs prior a MRCI calculation, which makes the convergence of the CI expansion more efficient [24]. In practice, the HF equation is solved by introducing a finite basis set of spatial basis functions {(^ | u = 1,2,...K} resulting in different matrix equations: Roothaan equations for closed-shell restricted determinants, Pople-Nesbet equations for unrestricted determinants, and Roothaan-Hartree-Fock equations for open-shell restricted determinants [1, 25-27]. Increasing the flexibility of the one-electron basis set {^}, the HF energy Eo will progressively reach a limit, called the Hartree-Fock limit (the exact HF energy). This limit cannot be usually achieved and the computed HF energy with a finite basis set is somewhat above it. The correlation energy (Ecorr) is defined as the difference between the exact non-relativistic energy of the system (so) and the HF energy Eo in the limit that the basis set approaches completeness ECOrr=So-Eo
(12)
Since so is lower than Eo, the correlation energy is negative. Because of the use of a complete basis set is prohibitive, or simply impossible, the exact electron correlation of a system cannot be computed, except for small systems. Definition of Ecorr corresponds then to the difference between the energy computed at a given level of the electron correlation treatment and the corresponding HF energy, both computed with the same, flexible enough, one-electron basis set The HF wave function, as it is a Slater determinant, the best one indeed in the sense of the variation principle, enjoys the basic features discussed above for determinants. Therefore, the HF wave function is uncorrelated, which leads to certain limitations in actual applications. For instance, it is well known that the restricted HF method cannot describe the dissociation of molecules into open-shell fragments (e.g. H2 —> 2H). Let us address this aspect with a model: the hydrogen molecule described in a minimal basis set, which also serves to introduce in a natural way more complicated functions including electron correlation. In a minimal basis model of the H2 molecule there are only two MOs, which are linear combinations of the two functions cpA and cpB placed on the nuclei HA and HB, respectively. The occupied molecular orbital, the bonding orbital of a g symmetry, has the lowest energy, and the virtual orbital corresponds to the antibonding combination of o u symmetry
43
(O
-n-
(13)
(14)
where s = <(pA|cpB> is the overlap between the basis functions. The case is simple enough that the solutions to Roothaan's equations are determined by symmetry arguments. The symmetry of the breaking process is maintained along the dissociation path and, therefore, the MOs have the same form independently of the interatomic distance R. Close to the equilibrium geometry, the ground-state wave function is (15) which expanded in terms of the basis functions leads to (16)
2(1+ s)
In Eq. (16), the first and second terms correspond to ionic configurations in the valence bond (VB) theory HA~ HB + and HA + HB~, respectively, while the third and fourth terms represent covalent situations. The four terms share the same coefficient, therefore, their weight is the same at a given distance R. Consequently, as R—»oo, s—»(), and the dissociation limit obtained is
Eo=-
(17) -1 n
The limit, rather than being twice the energy of the hydrogen atom in the same basis set (2E(H)), includes the spurious term E(H~)/2 reflecting the contribution of the ionic structures even at infinity. However, the incorrect behavior of the restricted HF theoiy at long interatomic distances for systems that dissociate into open-shell products does not detract from the validity of the approach around the equilibrium geometry, where the HF method has been shown to be remarkably successful for closed-shell ground-state systems. The ground state of the H2 system in the dissociation limit corresponds to two ground-state hydrogen atoms ( S) and the correct spin adapted wave function is
(18)
44 that in terms of MOs (Eqs. (13) and (14) with s = 0) can be rewritten as
The previous expression suggests that in order to decrease the weight of ionic determinants in Eq. (16) at interatomic distances close to the equilibrium geometry, R=RoPt, one has to combine the determinants corresponding to the ground-state (o g ) 2 and doubly excited (a* )2 configurations. The procedure is known as the configuration interaction (CI) method 2)
(20)
being X< 0 the variational parameter. As R—»co, then X —> - 1 , and the functions (20) and (19), except for a normalization factor, become equivalent. The method offers a proper treatment of the spatial correlation of the electrons, called left-right correlation, making it possible that the two electrons belong to different nuclei. In summary, in order to get a correct dissociation the doubly excited configuration involving the antibonding orbital has to be invoked because it has at infinity the same weight as the closed-shell ground-state configuration. Alternatively, one could think that the unrestricted HF (UHF) approximation might be a solution. It can be shown that the UHF energy goes to the correct limit but the total wave function does not. The UHF solution for the H2 molecule is not a pure singlet but it is contaminated by a triplet, which is required to make the UHF wave function a single determinant. At the dissociation limit the triplet contamination represents 50% of the wave function. Therefore, an unrestricted solution does not provide the best starting point neither for configuration interaction nor perturbation calculations. A similar reasoning employing CI wave functions can be also performed to analyze the two-electron correlation relative to the nuclear positions. These short-range correlation effects are typically called radial and angular correlation. They are related to the larger preference, in relation to the HF description, that two electrons actually have to be far apart of each other: close/far from the nucleus of an atom (radial) and in opposite ways in a given direction, up/down, in the space surrounding the nucleus (angular). It is worth noting, however, that approaches based on many-electron basis sets (determinants or Configuration State Functions, CSFs) built as products of one-electron wave functions (orbitals) cannot represent exactly the shape of the Coulomb hole, although it becomes reasonably described with large CI expansions. Unfortunately, as stated above, the convergence of the CI expansion is slow. The reader is referred to the interesting and recent contribution reported by Knowles, Schutz, and Werner for a more detailed discussion on the topic [14]. In molecular systems, electron correlation is usually computed in two steps. Firstly, nondynamic electron correlation is accounted for by using a CASSCF wave function or a selected
45
number of configurations. In a second step, the remaining electron correlation effects (dynamic correlation) are estimated by considering the singly and doubly replacements from the MRCI or the CASSCF wave functions. The borderline between non-dynamic (static) and dynamic correlation is not clearly defined in most cases. Normally, correlation energy arising from long-range terms allowing the correct asymptotic behavior in a molecular dissociation is referred to as non-dynamic correlation. The remaining correlation energy dealing with shortrange effects relevant in describing the Coulomb hole as accurate as possible is associated with dynamic correlation. Inclusion of both types of correlation is crucial in order to gather accurate results. For instance, a good (uncorrelated) HF SCF calculation tends to underestimated bond lengths with respect to a (reliable, gas phase) datum determined experimentally because in a HF wave function the ionic and covalent terms (in the VB sense) are equally weighted, making single covalent bonds too tightly bound. As can be easily deduced from Eq. (20), non-dynamic electron correlation decreases the effect of ionic forms in the CI wave function, leading to too long bond distances. A subsequent introduction of dynamic correlation recovers such an overestimation, leading the computed geometry to a closer agreement with the experimental datum. One of the most efficient manners of dealing with non-dynamic electron correlation is by using the CASSCF wave function comprising as active the valence MOs and valence electrons. For instance, it leads in a diatomic molecular system to the correct dissociation limit, that is, to the sum of the energies for the isolated atoms. This property is termed sizeconsistency, which requires that the energy of two non-interacting systems be the sum of the individual system energies. Many of the contributions of the valence CASSCF are not actually required to get the adequate dissociation limit. Shorter MCSCF wave functions would equally make it correctly, as the GVB-PP approach, which becomes equivalent to a selected pair-wise MCSCF wave function, but the valence CASSCF is still preferred because its implementations have technical advantages that, in general, make the computation easier. However, the idea that the valence CASSCF wave function is the most general way to obtain a proper treatment of non-dynamic electron correlation leads sometimes to surprises. For instance, the valence CASSCF yields for the water molecule to an asymmetric structure, with two non-equivalent O-H bonds. The right C2V structure can be recovered by enlarging the active space with an extra orbital which makes then the calculation balanced with respect to the treatment of both bonds (see discussion in Ref. [28]). Thus, the idea that the best choice of active orbitals corresponds to the valence MOs is not always supported, although the fact that valence electron correlation is linked to non-dynamic correlation effects holds true. In the computation of excited states for small-medium molecular systems, the valence excited states are usually interleaved among a number of Rydberg states. Therefore, the oneelectron basis set has to be flexible enough to compute both valence and Rydberg states. Consequently, the active space has to include the necessary valence and Rydberg MOs. In those situations, the CASSCF wave function takes into account some dynamic and nondynamic electron correlation. In summary, selection of the active space has to be performed in accordance with the application at hand. It is not a black-box procedure and might not be
46 straightforward, although is far from being impossible because there is a large body of information accumulated. As in any scientific research, one has to become familiar with the available literature on the issue prior the actual computation gets started. In this sense, the choice of the active space for CASSCF calculations is not an exception. The effort is rewarding because, when a proper CASSCF wave function is employed as reference function for a subsequent treatment of dynamic correlation (MRCI or MRPT), the results are accurate within 0.1-0.2 eV, which is an error bar sufficient for many spectroscopic applications. Otherwise, if higher accuracy were required the whole selection procedure for the proper methodology has to be redone. For instance, in the description of weak interacting systems like in the NO dimer, N2O2 [29], the Averaged Coupled-Pair Functional (ACPF) [30] method was preferred because it is strictly size-extensive, that means it has the correct scaling with the number of particles, whereas any other forms of truncated CIs, as well as the CASPT2 method, are not [14, 16]. The natural orbitals (NOs) derived from a Restricted Active Space SCF (RASSCF) calculation, an extension of the CASSCF method [31], are a good choice for a single-root MRCI computation [32]. Natural orbitals are defined as density matrix eigenvectors. Since the trace of the density matrix is equal to the number of electrons, the associated eigenvalues are interpreted as the corresponding occupation numbers of the NO. It can be shown that for twoelectron systems, a CI wave function written in terms of doubly excited determinants built from NOs offer the most compact one-electron basis set [1, 14]. Implicit to a variational calculation is the fact that the largest contributions come from pairs of electrons occupying the same region of physical space (a, ,o* ;nl ,n\; etc). One of the best manners of localizing a pair of MOs in a given spatial region is through a pair-wise MCSCF computation; the decrease of the energy associated with that pair of electrons to fulfill the variation principle is directly related to the increase of the corresponding exchange integral between the two MOs. As larger is the exchange integral involving a pair, more localized the two MOs would become. The main effect is carried out on the virtual MO of the pair, especially when canonical MOs are used as starting point. Virtual canonical MOs are computed in the mean field of N electrons, and therefore they become too diffuse. In contrast, the occupied canonical MOs are properly obtained in the mean field of the remaining N-l electrons. In general, canonical MOs represent the most inefficient choice of one-electron basis set for the purpose of CI calculations. The convergence of the CI expansion is significantly improved with a set of localized MOs. Among many different choices of MOs available, the best corresponds to NOs derived from an MCSCF calculation [24]. The CASSCF/MRCI approach is capable of yielding accurate results on medium size molecules as shown in the eighties by many authors, in particular Bauschlicher and co-workers [33].
4. AB INITIO METHODS: ESSENTIALS FOR EXCITED STATES In order to write the present section, two criteria have been used. Firstly, the methods included respond to the fact they are frequently used for the study of excited states, as it can
47
be easily checked from a rapid search on papers published during the last few years. However, it does not mean that in the authors' opinion all of them are appropriate for the treatment of excited states. It just reflects in certain cases that, despite of their well-known limitations, they are commonly used nowadays. Secondly, in an attempt to keep this contribution as practical as possible for a researcher interested in using ab initio methods for computational chemistry, the methodological details are kept to the minimum. In order to get further insight into the methods, the reader is referred to the original papers or more advanced material. Advantages and disadvantages on the applicability of the corresponding approaches shall, however, be emphasized. According to the classification introduced in Section 2, under the headings single-configuration and multiconfigurational methods, as appropriate, the most popular approaches used today are next discussed. 4.1. Single-configuration methods Within the molecular-orbital model, the simplest manner to describe the excited states of a molecule that one can think of is by one- (or two)-electron promotion(s) from the occupied to the virtual canonical MOs obtained from the SCF calculation at the equilibrium geometry of the corresponding ground state. Within this scheme, the energy difference of two orbital energies can be related to the vertical excitation energy absorbed by a molecular system. The smallest energy difference occurs between the lowest unoccupied molecular orbital (LUMO) and the highest occupied molecular orbital (HOMO). That the excited state described mainly by the HOMO—>LUMO one-electron promotion does not always correspond to the lowestenergy transition is strongly supported by high-level ab initio results, as well as experimental evidence [15-18]. In summary, this simple approach usually yields predictions that might not be even correct qualitatively. Nevertheless, analysis of the leading configurations of a complicated multiconfigurational wave function on the basis of the corresponding NOs, which are topologically similar to the canonical MOs, is very helpful. In fact, this is the underlying meaning of an excited state labeled, for instance, as HOMO->LUMO coming from an extended CAS SCF computation. Giving a step forward, a vertical excitation energy can be estimated by two HF/UHF SCF calculations of the respective ground and excited states. As it occurs also in the computation of ionization potentials at the HF level, the accuracy of the result for the lowest excited state, independently of its multiplicity, strongly depends on the molecule under consideration. The source of the errors is related to the intrinsic limitations of the HF approach: lack of electron correlation and spin contamination for an unrestricted open-shell wave function. Because only exchange correlation is included, UHF triplet states are strongly favored with respect to openshell singlet states, which leads to the wrong ground state for some biradicals [34]. All these approaches are sometimes known as the ASCF method [2, 35]. The next step in complexity leads to the widespread method called CIS (Configuration Interaction-Singles) [19]. The essence of the method is to consider that an excited state can be described by a singly excited determinant formed by replacing, with respect to the HF wave function, an occupied spin orbital with a virtual spin orbital. The drawbacks of such a
48 description may be partially compensated if a linear combination of all possible single excited determinants (cp^) is used to build the excited state wave function as:
(21) where the coefficients csak are the components of the eigenvector for state k, M is the number of occupied orbitals (i) from which excitation is allowed, and N the number of virtual orbitals (a) into which excitation is considered. The singly excited states are orthogonal to the ground state because of the Brillouin theorem, but not necessarily to each other. The orthogonalization of the states is the essence of the CIS (Cl-Singles) technique. The CIsingles procedure involves diagonalization of the CI matrix formed from the HF reference and all single excited configurations. The final outcome is a set of energy eigenvalues associated with eigenvectors in which the coefficients of the singly excited determinants, variationally obtained, characterize the state. The technique is size-consistent, allows spin flips in the excited electrons, and therefore to describe singlet and triplet wave functions, and also to obtain analytical gradients [19]. The description of the ground state in the CIS method is kept at the HF level. The CIS method clearly differs from a ground-state CI calculation in the sense that the former simply requires from the ground state the optimal HF MOs and the CI is performed to orthogonalize the singly excited states, while the latter includes CI excitations, higher than singles, to improve the description of the ground state itself. The main flaw of the CIS method is the lack of correlation energy. Further attempts to solve the problem by using double excitations [36] or perturbation theory [19] were not successful. Improved results have been obtained with semiempirical parametrizations of the CIS matrix elements, for instance using the INDO/S approach [2]. In general the CIS excitation energies are largely overestimated due to the absence of electron correlation energy. For instance, a calculation on the singlet excited states of benzene reported a mean absolute error of 0.7 eV, with deviations as large as 1.4 eV, although larger errors are common [37]. Despite the claims that the results are qualitatively useful because the states are correctly ordered [2], the facts speak to the contrary. There are more cases in the literature [38-40] of failures in the prediction of the CIS energy ordering that successes, simply because the differential correlation energy affects the excited states unevenly and because the intrinsic character of the states is multiconfigurational. The well-known 2'Ag state of polyenes, one of the main protagonists of their photochemistry, is a good example. For trans-\,3-butadiene there is no CIS 'Ag7i7i* valence excited state below 9.0 eV [41], more than 3 eV higher than most of high-quality multiconfigurational methods [42, 43]. In a CASSCF description, the doubly excited configurations contribute by 42% to the state wave function. CIS, as most of the single-configuration methods, will have enormous problems to describe such states. In systems were purely doubly excited states exist at low energies [44], these methods cannot even represent such states. In summary, the CIS method is clearly unsafe and its use is discouraged because the obtained results are usually misleading [38-40].
49 A large number of methods to compute excited states is included under the denomination of propagator approaches [20]. As such, the underlying technique, also called Green's function approach, equation-of-motion or linear response theory in its different formulations, can be applied to various types of methodologies whether single- or multiconfigurational configuration interaction, coupled-cluster, or density functional. The basis of the technique considers that once a molecule is subjected to a linear time-dependent electric field fluctuating with frequency co, a second-order property as the frequency-dependent groundstate polarizability of the system is well approximated by
where the denominator of the expression involves the frequency of the field and the excitation energies (AEj) characterizing the excited states (i), while the numerator of each term is the square of the transition dipole moment between the ground and the corresponding excited state [2, 20]. Using complex function analysis it is possible to obtain the poles of the expression, that is, the values for which the frequency corresponds to the excitation energies and the denominator goes to zero, while the residues provide the numerators, in this case the one-photon absorption matrix elements. Higher-order quadratic response theory determines third-order molecular properties, as the first hyperpolarizabilities, and from them two-photon absorption matrix elements [45], The peculiarity of the propagator approaches is that the wave functions of the individual states are not necessarily computed to obtain excitation energies and transition probabilities, while its quality relies on the type of reference wave function. A hierarchy of approximate propagator methods can be defined as function of the selection of the order of the particle-hole replacement operators. The most popular among the polarization propagator methods is the Random-Phase Approximation (RPA) or TimeDependent Hartree-Fock (TDHF) approach, where the used reference is the HF ground state and a single replacement operator is employed [20], and that is equivalent to the Coupled Hartree-Fock (CHF) approximation for time-independent perturbing fields [46]. Further developments of the method have included second-order perturbation (MP2) based approaches such as the Second-Order Polarization Propagator Approach (SOPPA) method, in which the so-called density-shift terms, particle-particle and hole-hole, are included [47, 48]. Many other variants have been developed, but our goal is not to perform a systematic review [20]. Just to mention the second- and third-order Algebraic-Diagrammatic Construction (ADC) approach, a Green's function one-electron propagator approach which has been applied in recent years to several problems [49]. Methods based on Green's function belong to the same hierarchy of approaches. They are typically expressed in the energy-dependent formalism, which can be transformed to the time-dependent propagator formalism by using a Fourier transform. One-particle many-body Green's functions methods are basically employed to compute ionization potentials (IPs) and electron affinities (EAs) [1].
50
Usual errors of the RPA method fall into ±1-2 eV in the excitation energies, while oscillator strengths may differ in one order of magnitude. It is also frequent to find singlet and triplet instabilities [47]. The effect of double excitations has been included by perturbation theory in order to slightly improve the excitation energies in the RPA(D) approach [50]. The SOPPA approach may improve the results within ±0.6 eV. A similar behavior is displayed by the ADC(2) method, slightly improved in the third-order version [49]. In any case, all these methods do not include non-dynamic correlation effects and, because of their singleconfiguration character, are extremely deficient when computing multiconfigurational states. Moreover, they cannot treat double excited states or open-shell ground-state excited states [47]. As a general advice, the improved versions of the methods, SOPPA and ADC(3), can be used to obtain a good qualitative description of the spectrum, although they lack generality. Compared to other methods they have, supposedly, the desired black-box behavior, and with respect to the TD-DFT approaches, their failures are not erratic, but well justified. Multiconfigurational response methods have been developed in recent years proving to be accurate in the calculation of molecular properties, not so much on energies [51], because of the lack of dynamical correlation. The most recent family of methods for excited states based on a single reference that have known practical use, at least for small systems, are those based on the size-extensive coupledcluster (CC) approach. CC methods for the ground state have become, in practice, the most accurate quantum-chemical approaches for many systems. Exceptions are however well recognized, which are related to low levels of excitations employed in situations where the single-configuration reference is clearly poor, for instance dissociating or quasi-degenerated situations or systems like ozone, C2, N2, the NO dimer and others [2, 35, 52-54]. The key point of single-configuration CC approaches for excited states is that they use a HF zerothorder reference, which is in general rather poor to represent excited states, and, in order to recover a large amount of correlation, high orders in the excitation level have to be included [55]. Three groups of methods have been most employed: the Symmetry-Adapted Cluster Configuration Interaction (SAC-CI) approach [21], the Equation-of-motion Coupled-Cluster (EOM-CC) method [22], and the hierarchy of linear response CCn approaches [23]. Although they have different formulations, their performance for the common truncated and nonapproximated coupled-cluster models is similar. The SAC-CI method, which has been used to compute very large systems by approximate procedures [56], and has been also extended to open-shell references [57], is comparable to the EOM-CCSD approach, which includes up to double excited cluster operators [22]. In parallel to the usual CC equations, in EOM-CC theory, excited state wave functions are represented by a linear expansion in the space spanned by all states [22, 35]: jJ^o)
(23)
51 where TA and ^cc, refers to the CC excited and ground states, respectively, % to the HF ground state, and T is the linear coupled-cluster excitation operator:
f = f o + f 1 + f 2 + f 3 + . . = XcllTM
(24)
V
where c^ denotes the cluster amplitudes and TMthe excitation operators including none, single, double, triple, and higher-order replacements. These methods use commutation relations to solve the Schrodinger equation for the excited state, and may be regarded as a conventional CI theory in which the configuration expansions carry the information about the excitation structure while a similarity-transformed Hamiltonian carries information about electron correlation [35, 57]. EOM-CC methods in different versions have been also extended to deal with open-shell ground states and compute IPs and EAs [58]. The family of methods CCS, CC2, CCSD, CC3, and CCDST is based on response theory [59]. Poles and residues of the linear-response CC equations yield excitation energies and transition matrix elements. CCS, CCSD, and CCSDT give a complete coupled cluster treatment of single, single-double, and single-double-triple spaces, respectively, for excited states, but just in systems with closed-shell ground states. The CCS approach is equivalent to the single excited configuration interaction or Tamm-Dancoff approach [60]. The iterative hybrid CC2 and CC3 procedures introduce approximations of similar nature although differing in the level of excitation. In this way, in CC2 the doubles of the CCSD approach and in CC3 the triples of the CCSDT approach are approximated by using perturbation theory up to first- and second-order, respectively. In this way, for instance, the CC3 wavefuntion is obtained as [35]: iT2+Q3)|y0)
(25)
where To refers to the HF reference, and the T operators include the single and double excitation cluster operators, while the effect of the triple excitations are introduced by the Q operators iteratively. CC3 include the single and double excitations at third order and the triple excitations at second order in the fluctuation potential, all of them one order higher than CCSD. This approximate way to include higher excitation levels allows less demanding computational procedures, scaling N4"7 from CCS to CC3, respectively [23, 60]. In order to get accurate excitation energies and properties, the single-configuration coupled-cluster methods should include high excitation levels to compensate both the poor reference wave function and the multiconfigurational character of the excited states. In situations where the HF reference is good enough, CC-based methods are, up-to-date and in practice, the most accurate methods to compute excited states in small to medium size molecules with closed-shell ground states, but only for those states which are well described by singly excited configurations and in systems were the ground state has a clear singleconfiguration character. In those cases, and in order to get accuracy better than typical
52
propagator or TD-DFT methods, triple excitations have to be included in the cluster expansion. Approaches valid for practical cases are, in general, EOM-CCSD(T) and CC3, while those only including double excitations as SAC-CI, EOM-CCSD or CC2, can be considered of lower quality than, for instance, multireference perturbation methods such as CASPT2 or similar approaches [15-18, 52, 61-63]. The precision of the CC methods decreases in systems with open-shell ground states. Less accurate is the behavior of the methods when the character of the states is clearly multiconfigurational. For instance, the mentioned 2'Ag state of polyenes, a multiconfigurational state with a large contribution of doubly excited configurations in the CASSCF description, is a good example. The CASPT2 vertical result, 6.27 eV, can be considered here a good benchmark, because the CASPT2 method has proved its accuracy on matching the experimental two-photon value for the analogous state in hexatriene [42]. EOM-CCSD, CCSD(T), and CCSD(T), the latter a noniterative version for the inclusion of triple excitations, deviate 1.0, 0.7, and 0.5 eV from the CASPT2 value, respectively [62]. Other example shows up in the l'Ei g state of ferrocene, with an error of 1.5 eV from experiment at the CC2 level [61]. More dramatic is the situation in other systems, in which the excited state is clearly multiconfigurational, where not even the inclusion of triple excitations can lead to accurate results. For instance, the EOM-CCSD description of the 2*Ai state of ozone leads to huge errors, 5-6 eV, that approximate inclusion of triple excitations cannot solve [54]. A similar situation occurs for some excited states of C2, showing deviations with respect to FCI of 2.05, 0.86, and 0.41 eV at the EOM-CCSD, CC3, and EOM-CCSDT levels [54], the second and third CC methods differing in the perturbative or variational procedure to include triple excitations. For the description of the states of the NO dimer, N2O2, EOM-CCSD fails on describing even the ground state, and EOM-CCSDT shows large inaccuracies to describe the excited states [53]. The inclusion of quadruple excitations, unpractical so far, would improve some of those results, but the only solution in prospect to beat in accuracy the lower level and less expensive multireference perturbation approaches such as CASPT2, is to use multireference coupled-cluster (MRCC) methods [64], in which the required excitation level will be certainly lower. Additionally to the calculation of energy eigenvalues, molecular properties for the different electronic states, and transition properties must be computed to define the spectroscopy of a molecular system. Properties such as the electric dipole moment, the frequency-dependent polarizability tensor, the nuclear magnetic shielding tensor, among others, are intrinsic properties of the system responsible of many spectroscopic and even structural phenomena whose calculation require also accurate ab initio approaches. If we focus on the molecular electromagnetic properties, they can be derived either as derivatives of the electronic energy or as derivatives of molecular electromagnetic moments and fields [65]. In non-variationally optimized wave functions, the Hellmann-Feynman theorem is not satisfied and the properties obtained as derivatives of the energy do not agree with the expectation values of the properties. This is the case, in general, of CC or MP approaches, which should be preferred in order to get results including most of correlation energy, although, in many cases, MCSCF properties can be considered reasonable [15-18]. Redefinitions of the expectation values in high-level methods have to be performed. Most ab
53 initio methods used to compute properties can be divided in three types: (1) those which evaluate properties by approximations to exact perturbation theory, such as the sum-overstates (SOS) procedures and the polarization propagator methods, RPA, SOPPA or MCRPA; (2) those which use perturbation theory with approximate wave functions, such as the Coupled Hartree-Fock (CHF) method or the response methods using different wave functions, in particular MCSCF or coupled-cluster, and (3) the derivative-based methods, where numerical or analytical evaluation of the derivatives of the electronic energy or properties in the presence of the perturbing field is performed, for instance, by finite field approaches. A comprehensive review can be found elsewhere [65]. Also, transition properties, such as transition multipole moments needed to obtain intensities, transition probabilities and radiative lifetimes and kinetic constants, have been computed at different levels, although the response approaches are becoming common to get accurate results. Finally, we shall comment and summarize advantages and disadvantages of the mentioned methods. Of course, it reflects our own opinion on the subject. Here we have not included all methods available or developed, just those more widely used. Nowadays (2004) singleconfiguration ab initio methods are useful to describe excitation energies, excited state properties, and transition probabilities, always subordinated to certain restrictions. CISderived methods cannot be recommended in practically any situation. A low-level general description of the excited states structure can be better obtained by means of carefully calibrated TD-DFT methods (the mixed and empirically corrected DFT/MRCI approach can be considered the best) [66], provided that the limitations of the DFT approaches and their low accuracy are well known. Regarding the propagator methods, they have, as the other methods of this section, the advantage of being (only partially) black-box approaches, and, therefore, they can be used also to get a qualitative picture of the spectrum, although neither all states nor all systems. Finally, coupled-cluster based approaches, assume that the effect of triple excitations are included, yield an accurate account of many states and systems, but not all of them. Regarding molecular and transition properties, the CC-based methods are surely the most accurate procedures available, provided that the approach is appropriate, although they are computationally more demanding and further improvements in their performance are necessary. In order to have an overall accurate description of all types of excited states, it is necessary to point out that, at present, the multireference perturbative methods, with CASPT2 as the most widely used approach [15-18], represents the only generally applicable method for the calculation of excited states, in all type of molecular systems, closed- and open-shells, multiconfigurational and degenerated situations, dissociations, etc [35]. It can be expected that in the near future, the multiconfigurational coupled-cluster approaches reach the maturity to be of practical use in molecular systems of reasonable size, and then, higher accuracy will be available in all cases. The full development of the methods will require also the implementation of geometry optimizers, reaction paths algorithms, etc, and some years will pass until all the needed tools become available. Up-to-date, analytical gradients for excited states in single-configuration methods are available, at a high computational cost, at the SACCl [67], EOM-CCSD [68], and CC2 [69] levels.
54 4.2. Ylulticonfiguiational methods Let us considered the wave function of CI type (26) expanded in a many-electron basis set of determinants. As in the H2 molecule, one can select a number of determinants to describe the correct dissociation limit. When the energy is minimized with respect to the coefficients of the expansion we are using the configuration interaction (CI) method. It should be kept in mind that actual calculations are performed using either spin-adapted CSFs or determinants. In case that the expansion contains more than one configuration, the process is denoted as multireference CI (MRCI). The wave function of Eq. (26) is a multireference function and, at least, the singly and doubly excited determinants generated from each reference determinant m) are taken into account. When the reference wave function consists of a single configuration, such as a closed-shell HF wave function, we are in the framework of single-reference CI wave functions, e.g., SDCI (singly and doubly excited CI). Including up to N-tuply excited determinants Eq. (26) would represent the full CI wave function. A comprehensive discussion on the distinct types of the MRCI method, including technical aspects on its efficient implementation, can be found elsewhere [14]. Coming back to Eq. (26), the MCSCF energy is obtained by minimizing (^JH 1*) to determine both the optimum CI expansion coefficients (as in the CI method) and the optimum form of the orbitals used to build m). The orbital optimization is similar to that carried out in the HF SCF method, therefore the approach is known as the multiconfigurational selfconsistent field (MCSCF). The MCSCF energy is usually expressed within the secondquantization formalism [35, 46]. The electronic Hamiltonian given in Eq. (2), has the following expression (physicists' notation) [1] in second quantization
H = S(iNj)a;a i +^( i Jl kl )a;W k ij
(27)
z
<;,k]
where the sums run over the set of spin orbitals, and a- and a; are the creation and annihilation operators, respectively. In the notation often referred to as the chemists' notation, the Hamiltonian is expressed has the form H = X)[i|h[ j]afa, + l £ [ i j | kl] a ^ a . a , ij
( 28 )
^ijki
Summing over the spin leads to H=2(i|h|J)Eii+^X(ij|kl)(EiiEkl-5ikEil)
(29)
55
(30)
where the sums run over the molecular orbitals and the spin summed excitation operators, defined as (31) a=u. |3
have also been introduced. In Eq. (30) the elements hi} include the kinetic energy for electrons and nucleus-to-electron attraction; the two-electron integrals involving the molecular orbitals, in chemists' notation, are denoted by gijkl. Given the wave function (26) as linear combination of a finite set of determinants, the expectation value of the Hamitonian is
ijk I
Z
C
-
i jk I
being Dj™ = (m EV) n), one-electron coupling coefficients. Their possible values are -1,0, 1, 2 in a many-electron basis set of Slater determinants, P
ijk" = ~ ( m | E i j E k i ~Sjk E ii | n ) > two-electron coupling coefficients,
,,, element of the first-order reduced density matrix,
PLjkl = ^ c"mPpCn, element of the second-order reduced density matrix. The expression for the energy (32) gives the clue for derivation of optimization algorithms employed in the MCSCF methods. It worth noting that information about the MOs is entirely contained in the one- and two-electron integrals, whereas the Cl coefficients are involved in the matrices D and P. Thus, the parameters to be varied are the CI coefficients and the MOs, which is made by considering their variations as rotations within an orthonormalized vector space. Since the exponential of an anti-Hermitian matrix is the most general expression of a unitary matrix, the rotations in the MCSCF optimization procedure are done by using that
56 type of matrices. There are several techniques to carry out such a process. MCSCF optimization methods can be classified as first-order and second-order methods, depending on the convergence type. A first-order treatment is only based in the computation of the energy and the first derivatives of the energy with respect to the variational parameters. The secondorder MCSCF methods are based on an energy expansion up to second order and the second derivatives of the energy are also computed. They are characterized by a quadratic convergence in the final steps of the optimization procedure. The Newton-Raphson approximation up to second order, where the energy is expanded in a Taylor series as function of the variational parameters, is probably the most prominent algorithm. The new values for the parameters are obtained by solving a set of linear equations. The convergence process is rapid and efficient in the proximities of the final solution (quadratic convergence), although the trust region is usually small. It is the standard optimization method and most of the remaining approaches can be related to it either as modified or simplified algorithms. Details on the particular techniques can be found in the specialized literature [28, 35, 70]. From a practical point of view, the selection of a particular algorithm would depend on the necessities of the application at hand. An advantage of a second-order algorithm is that the stationary point is characterized, that is, the condition of minimum can be established. Comparatively, in a second-order method a full iteration is more time consuming as compared to simplified approaches, although the number of iterations required to converge are relatively smaller. One has to balance the advantages and disadvantages in the two type of approaches prior utilizing one of them in a particular study. Anyway, because of the computational resources available today, one can freely take the decision of enjoying the advantages of both. Independently of the chosen algorithm, it is highly recommended to supply the MCSCF optimization calculation with good starting orbitals, in accordance with the physics of the problem to elucidate. Otherwise, unwanted solutions might easily come out of the calculation, which are simply nonsense. We have assumed that the N-electron basis set is constituted by Slater determinants. Nevertheless, the Hamiltonian operator, as well as the orbital rotations, can be expressed as function of the orbital excitation operators, which commute with the spin operators. It is, therefore, possible to work entirely in a N-electron basis set formed by spin-adapted CSFs. There are many manners to build eigenvectors of the spin operators from Slater determinants. Among them the graphical unitary group approach (GUGA) has played an outstanding role. Indeed, the excitation operators fulfill the same commutation relationships as the generators of the unitary group of dimension n, and for that reason the E^ operators are often referred as to generators. A CSFs basis set leads to shorter CI expansions that a basis set of Slater determinants, but the CI algorithms employing determinants are more efficient. The issue was in the past a subject of great debate. A certain consensus has been reached today, which is reflected in available software [71], making practical use of the advantages of both types of N-electron basis sets.
57
As stated above, the CASSCF method [72] is probably the MCSCF method more widely used at present. In the CASSCF method, the orbitals are classified in three categories, depending on the role they play in building the many-electron wave function: inactive, active, and secondary orbitals. Inactive and active orbitals are occupied in the wave functions, whereas the remaining of the orbital space, given by the size of the one-electron basis set employed, is constituted by secondary orbitals, also called external or virtual. Inactive orbitals are doubly occupied in all the CASSCF configurations. The number of electrons occupying inactive orbitals is, therefore, twice the number of inactive orbitals. The rest of the electrons (called active electrons) occupy active orbitals. The CASSCF wave function is formed by a linear combination of all the possible configurations that can be built by distributing the active electrons among the active orbitals and are consistent with a given spatial and spin symmetry. That is, in the configuration space spanned by the active orbitals, the CASSCF function is complete (or full). Inactive orbitals are also optimized in the variational process but they are treated as in the restricted HF function. The CASSCF energy is invariant to rotations among the active orbitals. Several states that belong to a same symmetry are usually computed by means of a StateAverage (SA) CASSCF calculation, where a functional of energy is defined as average of a number of states (1=1, M)
Eaverage = 2 > , E,
(33)
i
being to, the factors of the relative weight for each state considered. From a SA-CASSCF calculation comes out a set of average orbitals and a number of orthogonal wave functions equal to the number of roots used in the average process. In this manner, it is sometimes possible to overcome the problem of "root flipping", the interchange of roots along the CASSCF optimization procedure. For a given spatial and spin symmetry, the treatment of excited states is preferably performed by using SA-CASSCF calculations. In principle, it is also possible to make a single CASSCF calculation for higher roots (I > 1), optimizing just one state. Nevertheless, experience shows that in most cases, it can only be achieved for 1=2, the second root of a given irreducible representation. The active space provided by the user of a CASSCF software represents a key point to obtain accurate theoretical predictions, once that dynamic correlation has subsequently been taken into account, for instance at the CASPT2 level. The properties of a CASSCF wave function depend on the active space. Thus, a valence CASSCF is size-extensive and the corresponding CASPT2 results become also nearly size-extensive (formally the CASPT2 method is not size-extensive). As in any quantum-chemical approach one has to make sure that the method has enough flexibility to describe the chemical process under consideration. The flexibility in a CASSCF wave function is determined by the active space.
58 The CASPT2 method can be seen as a conventional non-degenerate perturbation theory, that is, a single reference function is considered, with the particularity that such reference function (zeroth-order wave function) is a CASSCF wave function and uses internal contraction in its formulation. The solution to the equation (34)
is expanded in power series of X: (35) (36) (37) Correction of order k to the wave function, T110, and to the energy, E
where the energy correction to order k, using intermediate normalization, is
(o) ft,
7(k-D
(39)
Therefore, we can write (40)
(41) The first-order correction to the wave function is given by (42)
with R<0) being the reduced resolvent (43)
59 The set of functions required to compute the first-order correction of the wave function, ¥ '\ is formed by those that interact with the zeroth-order wave function through the Hamiltonian in the perturbation theory of Rayleigh-Schrodinger, and it is known as the firstorder interacting space [73]. Taking into account the one and two particle nature of the Hamiltonian, the first-order interaction space, called hereafter VSD, comprises the functions generated by singly and doubly excited configurations from the zeroth-order wave function. In the CASPT2 formulation, the VSD space is divided in eight subsets, according to the nature of the excited configurations (see details in Ref. [10]). The corresponding functions are built by applying products of excitation operators, Epq Ers, to the zeroth-order wave function o), Eq. (26), with the coefficients cm kept as determined in the CASSCF wave function. It is the reason why the CASPT2 method is internally contracted. In non-contracted methods [13, 14], the excitation operators act directly on the functions m) of the linear combination described by Eq. (26). Both singly and doubly replacements from the CASSCF wave function are considered. The singly excited configurations can be explicitly seen as linear combinations of certain products of excitation operators. Internal contraction makes the perturbation series considerably shorter, without affecting significantly the quality of the results [10]. Nevertheless, although the dimension of the first-order interacting space is considerably smaller because of the internal contraction, the complexity to obtain the firstorder wave function increases because the resulting functions belonging to VSD are not in general orthogonal and may also have linear dependencies. x {
The first-order correction of the wave functions is expanded in the basis of the functions j) belonging to the VSD space M
M > dim VSD
(44)
The coefficients (Cj, j = 1,..., M} are obtained from the system of linear equations >
i=l,...M
(45)
where E(o) =(o|H°|o) is the zeroth-order wave function and must be solved iteratively. In standard CASPT2, the zeroth-order Hamiltonian is expressed in terms of a generalized Fock operator, which can be written as a sum of a diagonal, FD, and non-diagonal, FN, contributions F T =F D +F N
(46)
The operator is defined in such way that for a closed-shell HF reference wave function is equivalent to the Moller-Plesset Hamiltonian. For multiconfigurational single-reference perturbation theory, the choice of the zeroth-order Hamiltonian is not unique and it has been
60 the subject of active research and discussions yielding a number of different successful variants [16]. Eq. (45) can then b e rewritten as '|o)
i = l,...M
(47)
In order to simplify the notation, the following matrices and vectors are introduced X = D, N J
V; = i H O
(48) (49) (50)
where i, j = 1, ...M. The column vector C contains the coefficients Cj of the expansion. The difficulties in the resolution of Eq. (47) depend on the choice of the one-particle operator. In the simplest case, using F D , it leads to [F D -E (O) S]C = -
(51)
and the second-order correction to the energy comes out as the product of V'C. In most cases M > dim VSD and the linear dependences have to be removed. It is done by diagonalizing the overlap matrix S and discarding the eigenvectors with eigenvalues equal (or close) to zero. The resulting vectors are then orthogonalized and a subsequent diagonalization of the Fock matrix written in the orthogonal basis takes place. The E(2) correction is easily evaluated as function of the transformed matrices. The process becomes much more elaborated when the full operator FT is used [10]. It is the recommended procedure. The normalized wave function corrected up to first order is given by = C0 o) + C,
(52)
with C^ +C 2 = 1. The weight of the reference function (C^) can be used as a simple and rapid criterion of quality for the perturbation treatment carried out. Ideally, in order to get a fast convergence in the perturbation series, the weight should be close to unity. Nevertheless, its value depends on the number of correlated electrons [28]. Thus, upon enlarging the molecular system the reference weight decreases. The electronic excited states considered should have a similar magnitude for the weight as compared to the ground state, employing the same active space. Sometimes intruder states appear in the second-order calculation, which are normally related to the occurrence of large coefficients in the first-order expansion, leading to a low
61 value for the reference weight. Analysis of the states with large coefficients (intruder sates) may give a hint about the type of reformulation in the perturbation partition necessary to overcome the problem. Thus, a new CASSCF calculation might be designed comprising in the active space the orbitals implied in the description of the previous intruder states. It is the proper action to be taken when intruder states are strongly interacting with the CASSCF reference wave function, with contribution to the second-order energy larger than 0.1 au, because it points out to obvious deficiencies in the choice of the active space. Intruder states are often present in the treatment of excited states of small organic compounds when the active space does not include the full n valence system. Thus, the low weight for the zerothorder wave function in such a case just tells us that the active space has to be enlarged in a way that previous intruder states would be treated variationally, that is, they should be moved to the CAS-CI space. It is also frequent to find calculations where the reference weight of the excited state is "somewhat low" compared to that of the ground state, but a particular state cannot be identified as intruder in the first-order wave function, which is instead characterized by a large number of low-energy minor contributions. It occurs often in the simultaneous computation of valence and Rydberg states, where the one-electron valence basis set has been augmented with Rydberg-type functions. We have to face then accidental near-degeneracy effects, implying weakly interacting intruder states, and the level-shift (LS) technique is especially useful in order to check the validity of the perturbation treatment performed. Many times one has to apply both strategies: enlargement of the active space to overcome the problem of severe intruder states, and, with the enlarged active space, the LS technique is applied in order to minimize the effect of weak interacting intruder states. The level-shift CASPT2 (LS-CASPT2) method removes efficiently weak intruder states by the addition of a shift parameter, s, to the zeroth-order Hamiltonian and a subsequent back correction of its effect to the second-order energy [16, 28, 74]. It can be shown that the corrected level-shift second-order energy, Ej2*!, is equal to the standard CASPT2 energy, E(2), in first order of s 1
(53)
CO
where E(2)and ffi (weight of the CASSCF wave function) were obtained by using the shifted Hamiltonian. The relationship (53) might not be valid when intruder states appear in the firstorder interacting space. It is highly recommended to make an analysis of the trends for the weights a, total, and excitation energies upon varying the values of s. For instance, results at s = 0.0 (standard CASPT2), 0.1, 0.2, 0.3, 0.4 au are sufficient to establish the proper behavior of the LS-CASPT2 results. It is extremely dangerous to rely on just one result, because the appearance of an accidental near degeneracy might lead to large errors in the excitation energies. In order to demonstrate the proper performance of the LS-CASPT2 technique, calibration calculations of that type always have to be carried out. The best choice for s is the lowest possible value capable of removing intruder states. In the absence of intruder states the ELs<2) energy varies only slightly with respect to the value of 8. As can be seen, in this type of
62 approaches there is a large interaction researcher-software. The responsibility for the decisions taken along the computational process belongs, of course, to the user (researcher) of the tool (program). Other formulations of the multireference perturbation theory have been developed although they have not widespread use [75-80]. The multi-state CASPT2 (MS-CASPT2) [81, 82] procedure represents an extension of the CASPT2 method for the perturbation treatment of chemical situations that require two or more reference states. For instance, situations such as avoided crossings and near-degeneracy of valence and Rydberg states, which cannot be fully accounted for by just using a singlereference perturbation treatment. In the MS-CASPT2 method an effective Hamiltonian matrix is constructed where the diagonal elements correspond to the CASPT2 energies and the off-diagonal elements introduce the coupling up to second order in the dynamic correlation energy. Let us assume that we have performed two CASPT2 calculations for the corresponding reference wave functions©; (i=l, 2), obtained by using average CASSCF for those two roots and a set of average molecular orbitals is, therefore, available. In order to build the matrix representation of the Hamiltonian using as basis set the two normalized wave functions corrected up to first order, 4'i = C>i +4'| 1) , the following matrices are defined: =8ij+Sij
(54)
(55)
H(D; =8 i ; E ;
(56) Notice that the two wave functions are not orthogonal, since (O, and l } = 0, but (y "} = s n . On the other hand, the CASSCF energy for state ith is represented by E, and the elements en are the CASPT2 correlation energies. For each state, the Hamiltonian can be expressed as the sum of a zeroth-order contribution and a Hamiltonian taking care of the remaining effects (
(57)
= H°+H:
Therefore, up to second order it holds true that H
H;
H°
(58)
H| VF(.I)) correspond to third order corrections and, consequently, they The elements ( are not considered. The matrix representation of the Hamiltonian is not symmetric Hl2 * H,
63 Assuming that the off-diagonal terms are very similar, as it is implicit from Eq. (58), the matrix is made symmetric by using the average value [/(I)
H
H°
«F")I
(59)
The matrix element including zeroth-, first-, and second-order corrections takes the general form E i + i ^ + e ^ +^ E r ' + E f ^
(60)
By solving the corresponding secular equation (H-ES)C=0, the eigenfunctions and eigenvalues can be obtained. They correspond to the MS-CASPT2 wave functions and energies, respectively. The MS-CASPT2 wave function can be finally written as
where i^ are the CASSCF reference functions and 4^" is the first-order wave function for state p. Accordingly, the function formed by a linear combination of the CAS states involved in the MS-CASPT2 calculation is the model state and can be considered as a new reference function for state p. This reference function is the so-called Perturbation Modified CAS (PMCAS) [82]. It is used for the computation of transition properties and expectation values at the MS-CASPT2 level. For the proper use of the MS-CASPT2 method, the condition (58) has to be fulfilled. In practice, it means that the asymmetric effective Hamiltonian matrix should have small and similar off-diagonal elements. Otherwise, the average process carried out, (H ] 2 +H 2 l )/2, may lead to unphysical results, in both the MS-CASPT2 energies and eigenfunctions. The condition that Hl2 = H2I can be achieved by enlarging the active space, which implies a redefinition of the zeroth-order Hamiltonian. Large active spaces, beyond the main valence MOs, are used naturally in the simultaneous treatment of valence and Rydberg states, where the MS-CASPT2 approach has proved to be extremely useful. Especial caution has to be exercised, however, for the computation of a crossing point between two surfaces, as in the case of conical intersections (and avoided crossings), crucial in photochemistry. The states involved in a conical intersection have usually different nature. Quite often one state has covalent character, whereas the other is zwitterionic [83]. They are described by hole-hole and hole-pair VB structure, respectively. The effect of dynamic correlation is usually much more pronounced for zwitterionic than for covalent states. As a result, with
64 moderate (valence) active spaces, the off-diagonal elements become very different, because the covalent state is comparatively described more accurate than the zwitterionic state. Active spaces comprising MOs beyond the valence shell would be required to make Hl2 = H 2I . In addition, the structure of the 2x2 effective Hamiltonian is
: "XT £) where E" 2 =E, +e n and E2T2 = E 2 +e 2 2 are the CASPT2 energies of the two states and A = (H12 + H 2 ] ) / 2 . If the states are degenerate at the CASPT2 level, E|>12 = E212 = E, and the multi-state energies and wave functions are E+ = E ± A
(63)
^=-^(¥,±¥3)
(64)
As A = 0 the MS-CASPT2 and the CASPT2 solutions are equivalent, what is expected to occur at the conical intersection. Therefore, by providing enough flexibility to the active space, one has to make sure that the condition H,, s H2I is satisfied and A becomes small (< 2 kcal/mol). As a conclusion, computation of surface crossings at the MS-CASPT2 level (so far numerically) is expected to require more extended active spaces than those done at the CASSCF level. What does it happen if A is larger than 2 kcal/mol? For systems of large molecular size one cannot be sure whether that result points out to the presence of an avoided crossing or it is just spurious because of the limited active space employed. As shall be illustrated in Section 6, dynamic correlation plays sometimes a crucial role in determining the nature of the lowest surface crossings. Nevertheless, except for small molecular systems, the MS-CASPT2 approach in its present formulation does not represent a practical solution for this purpose. Methodological efforts are certainly required to improve the present situation. In this respect, recent advances on analytic energy gradients for general MRPT methods seem very promising [84]. On the order hand, computation of conical intersections at the CASPT2 level uses two non-orthogonal wave functions and how it might affect to the structure of the singular point so obtained is not yet known. Unfortunately, localization of conical intersections including dynamic correlation by using variational strategies, at the MRCI level for instance, is currently limited to small-size molecular systems [85-87]. 5. EXCITED STATES AND SPECTROSCOPY 5.1. General Considerations Quantum-chemical methods provide information for excited states directly applicable to explain and predict the spectroscopy of molecular systems. A balanced description of the different electronic states is required in order to obtain the basic data, that is, energy
65 differences and transition probabilities, in an accurate way. This goal is a much more difficult task for excited states as compared to the ground state. First, one has to deal with many classes of excited states, each one requiring different amounts of electronic correlation and flexible one-electron basis functions able to describe all effects simultaneously. Then, it is necessary to compute extremely complicated potential energy hypersurfaces where the number of minima, transition states, and surface crossings such conical intersections, is multiplied. Because of the inherent complexity of the problems, the methods and algorithms to compute excited states are not so efficient as for ground states or are still under development [2,4]. The selection of the proper one-electron basis set is the first decision a quantum-chemist has to take in order to plan a calculation, and will determine the accuracy of the obtained results. In general, excited state quantum-chemical calculations require the use of large, diffuse, and flexible basis sets, able to describe at the same level states of compact nature, such as valence, and diffuse, such as Rydberg or anionic states. Atomic Natural Orbital (ANO) basis sets supplemented with diffuse functions or augmented correlation-consistent basis sets (aug-cc-pVXZ, with X=D,T,Q,...), are the best general choice in order to get all type of excited states in the different regions of the spectrum [88]. Because of their balanced construction, ANO basis sets usually get better results with less number of functions than other sets. In a typical calculation in electronic spectroscopy for a medium-size molecule, an ANO contraction of the triple-zeta plus polarization type has been shown to give accurate and reliable results for valence states [15-17]. Specific diffuse functions with small Gaussian exponents, whether distributed on the atomic centers or centred in the molecule, are required to compute Rydberg and anionic excited states. Basis sets of the type 6-31G* are not as accurate but considerably cheaper. If used, they should be carefully calibrated for the studied problem. They may work in cases where the Rydberg states are not competitive with the valence states in the studied energy region. Normally, a full study including both valence and Rydberg states is required in order to validate the quality of the valence results employing smaller basis sets. It particularly holds true in molecular systems with a strong valenceRydberg mixing [42, 89]. In order to properly compare to the recorded spectroscopic data [18], the excited states have to be computed at significant points in the potential energy hypersurface (PES), which should be previously located by using appropriate optimization algorithms. At the ground or excited state minima, vertical absorption (EVA) and emission (EVE) energy differences are obtained comparable, within the spirit of the Franck-Condon (FC) principle, to the absorption and emission band maxima, respectively. This is just a convenient approach. In order to identify on theoretical grounds the true maxima, a full determination of the vibrational profile of the electronic transition would have to be performed. The vertical transition is however a quite useful concept. The obtained vertical excitation energies and oscillator strengths, together with properties such as the charge distribution in the different states, multipole moments, etc, give an overall view of the structure of the excited states and electronic transitions, although further refinements are required to achieve higher accuracy. Typical
66 differences between vertical absorption and band maxima range 0.1-0.2 eV in systems where the excited state structure undergoes small changes as compared to that of the ground state [90].
u'=0
Ground state u"=0
Fig. 1. Computed energy differences describing the molecular photophysics.
In order to gain more spectroscopic insight it is necessary to compute adiabatic transitions, that is, energy differences between states at distinct regions of the PESs [18]. In particular, the energy difference between the excited and the ground state at their respective optimized equilibrium geometries can be related to the electronic band origin, both in absorption and emission. In the former case, it is the smallest possible energy difference allowed in absorption under the assumption that all excitations begin from the relaxed ground state. In emission it is related to the largest energy emitted from the relaxed excited state. As displayed in Fig. 1 such transition is preferably coined Te. In many cases determination of Te provides enough information to assign band origins. If more accurate results are needed, the Zero-point Vibrational Energy (ZVE) has to be included in both initial and final states to obtain the vibrational band origin To (also named 0-0 or 0^ transition), which is strictly comparable to the experimental datum {see Fig. 1). Usually, determination of the vibrational frequencies (COQ) at the state minima is performed within the harmonic approach to simplify the calculations [18]: To = Te
y =T e +E i . vib (0)-E j . vib (0) =
J-Q
(65)
Locating singular points in the hypersurface is a difficult and time-consuming task. It is frequent that geometry optimizations, frequency or property determinations are performed at levels of theory lower than the energy calculations. In most cases it is a question of balance in the results, in particular when highly correlated methods are too expensive for the system under study and low-level approaches have to be employed. For instance, in molecules with double bonds, an enlargement of the bond length is observed when increasing the amount of
67 correlation energy included. In particular, it has been observed, even for ground states, that in regions with large n delocalization, the difference between methods such as MP2 and CASSCF is quite large (up to 0.03 A). The former, and the same holds true for DFT, usually overestimates n conjugation [91]. In these cases, TT-CASSCF may be a preferable approach to get geometries close to the gas-phase results, because the effects of the ^-correlation compensate the lack of o-correlation. Even more dangerous can be to use crystallographic data. X-ray crystal determinations are known to underestimate double bond lengths (up to 0.02 A) [92, 93]. Apart from energy differences at specific geometries, spectroscopic determinations require the calculation of transition probabilities in order to get band intensities, emission lifetimes, and kinetic rate constants [94, 95]. Within the static picture and using Fermi's Golden rule, the calculation of transition multipole (dipole approach) moments, together with transition energies, leads to transition probabilities in the form of oscillator strengths: f = -E V A M(Q 0 ) 2 3
(66)
where EVA is the vertical absorption energy and M(Qo) is the modulus of the transition dipole moment, computed as the transition dipole components (Mx, My, Mz) between the initial and the final state at the ground state equilibrium geometry. The oscillator strength can be directly related to the experimental observation, based on band shapes and half-widths. More precise determinations of band profiles require the calculation of vibronic transition moments and frequencies. From the calculation of transition dipole moments, radiative lifetimes can also be obtained, both in fluorescence and phosphorescence, for the electronic or vibrational states by using the Einstein coefficients (A21) and the Strickler-Berg relationships [95]: A2I = — = 2.142005 • l O ^ M t Q j ' E ^ T
(67)
rad
where irad is the radiative lifetime measured in s (the other magnitudes in atomic units) and EVE is the emission maximum, which can be also replaced by To. In the case of phosphorescence, the spin-orbit coupling has to be considered to get M(Qo), which is considerably smaller. Vibronic contributions can be then crucial, in particular in phosphorescence. Intersystem crossing rates can be also obtained in a similar way. In systems including heavy atoms, the spin-orbit coupling can be large enough, the difference between fluorescence and phosphorescence vanishes, and emission is traditionally named luminescence [96]. 5.2. On the Valence-Rydberg Mixing: Anti Conformer of w-Tetrasilane. According to the nature of the MOs involved in the description of an electronic state, two basic types of excited states can be found in actual calculations in neutral molecules: valence and Rydberg. The latter have large radial extension of atom-like character, covering the whole
68 molecule. In a good approximation, a Rydberg state can be described as the result of a oneelectron promotion from an occupied orbital to an atom-like orbital of higher quantum number. Therefore, the electron in the Rydberg orbital "feels" the molecule like a cation acting as a point charge, and the presence of the state is justified by the electrostatic interaction between the electron and the cation. In a molecular system formed by atoms of the second period, with electrons of maximum principal quantum number n=2, the Rydberg orbitals begin with n-3, and the Rydberg states 3s, 3px, 3py, 3pz, 3dX2.y2, 3dZ2, 3dxz, 3dyz, 3dxy, are usually found among the lowest Rydberg states, where the promoted electron comes from the HOMO. Valence and Rydberg states can be characterized by their spatial extent, measured through the expectation values <x2>, , and . Valence excited states are described mainly by valence MOs (bonding, lone pairs, and antibonding) and, therefore, they are more compact as compared to diffuse Rydberg states. Comparison of the relative values computed for the second Cartesian moment (<x2>, , ) can be used as criterion to determine in a simple manner the nature of the excited state; results close to those of the ground state indicate the valence nature of the excited state under consideration. It can be also used to identify a particular type of Rydberg state according to its radial extension. Excited states of intermediate valence-Rydberg nature come out quite often from the computation. To elucidate whether those states actually correspond to spectroscopic states or are just an erroneous consequence of the truncated level of theory employed is not an obvious task and the actions to be taken depend on the particular case. Higher levels of theory are usually required to give a clue in the right direction. Experience shows, however, that valence-Rydberg mixing found in vertical transitions is in most cases spurious and it progressively vanishes upon the increasing level in the treatment of dynamic correlation [97, 98]. Well-known examples of a strong valence-Rydberg interaction at the CASSCF level are the excited states of ethene and butadiene [42, 82]. A similar situation was found in the anti conformer of n-tetrasilane (see Fig. 2), where the SA-CASSCF calculations lead to a strong mixing of the Rydberg and valence states [98]. The MS-CASPT2 method is able to rectify the problem yielding an effective separation of the computed states, which can be clearly identified as valence and Rydberg [82]. Tetrasilane can be regarded as the simplest oligosilane for which the contribution of the conformers, gauche and anti forms, on the electronic spectra can be analyzed, providing an step further to the understanding of siliconcontaining compounds of great impact in modern technology [98, 99]. Table 1 compiles the results computed at the CASSCF, CASPT2, and MS-CASPT2 levels at the equilibrium geometry of the ground state using an ANO-type basis set with the contraction scheme Si[6s5p2d]/H[2slp], which was augmented with a lslpld set of Rydberg functions placed in the centre of the system. The computations were carried out within the C2i, symmetry constraints, with the silicon atoms placed in the xy plane and the y axis parallel to the terminal SiSi bonds. The active space comprises the a and o* Si-Si bond orbitals,
69 extended for each irreducible representation to include Rydberg orbitals and 7i*-symmetry Si-H antibonding orbitals, as appropriate [98]. Six valence states occur below the lowest Rydberg transition at the MS-CASPT2 level. Let us focus our attention on the singlet excited states of Bu symmetry. Transition to the lowest excited state (1 'Bu) at 6.33 eV, computed with the larger oscillator strength (around 1.12), can be clearly attributed to the low-energy experimental band with a maximum at 6.14 eV in the matrix spectrum [100]. The l'B u state in terms of the occupation numbers associated with natural orbitals obtained from the PMCAS wave function corresponds to the expected oneelectron promotion HOMO^LUMO of aa* character. On the other hand, transition to the 2'B U (V5) state at 6.96 eV has a smaller oscillator strength (0.15) and it probably contributes to the overall shape of the high-energy band with maximum at 6.89 eV observed in the matrix spectrum [99, 100]. The 3'BU and 4'BU states are 4p Rydberg states, placed at 7.46 and 7.87 eV, respectively (MS-CASPT2 results). The CASSCF calculation was carried out as four-root average of the singlet states of Bu symmetry. From a comparison of the MS-CASPT2 to the CASSCF and CASPT2 results, one can easily conclude that the CASSCF procedure leads to a too pronounced valence-Rydberg mixing that a single-reference multiconfigurational perturbation theory such as CASPT2 cannot fully recover. As a consequence, the two Bu states of valence character are placed energetically too high at the CASPT2 level, whereas the two Rydberg states are stabilized too much. Accordingly, the CASPT2 oscillator strengths for the valence transitions are underestimated (because of the interference of the Rydberg states). The opposite is true for the Rydberg transitions, which are computed with larger oscillator strengths at the CASPT2 level, with respect to the MS-CASPT2 findings, because of the mixing with the valence excited states. A comparison between the CASSCF and MS-CASPT2 results demonstrates that dynamic correlation effects contribute the most (more than 2 eV) to the excitation energy of the lowest valence excited state Vj. This effect is typically found in zwitterionic states (i.e., described by hole-pair ionic VB structures) [83], which are particularly difficult to characterize theoretically. In order to get further insight into the valence-Rydberg mixing Table 2 lists the results obtained with the Si[6s5p2d]/H[2slp] ANO-type basis set valence basis set (omitting the Rydberg functions). The number of roots in the average CASSCF process was just the required to compute the valence excites states (2 for 'B u symmetry, 3 for 'Ag symmetry including the ground state, and 1 for the remaining). As can be readily seen from Tables 1 and 2, similar results for the valence excited states are obtained with both basis sets at the MSCASPT2 level. Furthermore, when the valence basis set is employed, the CASPT2 and MSCASPT2 results agree (cf. Table 2).
70
Fig. 2. The ground-state structure of the anti conformer of n-tetrasilane. Table 1 Excitation energies (AE) and oscillator strengths for anti «-tetrasilane employing the Si[6s5p2d]/H[2slp] + l s l p l d (Rydberg functions) ANO-type basis set [98], CASSCF State
CASPT2
MS-CASPT2
AE (eV)
Osc. Str.
AE (eV)
Osc. Str.
AE (eV)
Osc. Str.
l'B u (V,)
8.50
1.098
6.90
0.891
6.33
1.115
2'Ag(V2)
6.94
forb.
7.12
forb.
6.55
forb.
0.003
6.86
0.003
6.68
0.005
l'Au(V3)
7.58
3'Ag(V4)
6.46
forb.
6.76
forb.
6.87
forb.
2'BU(V5)
8.81
0.275
7.50
0.234
6.96
0.154
l'Bg(V6)
8.06
forb.
7.51
forb.
7.10
forb.
4'Ag(H->4s)
7.10
forb.
7.33
forb.
7.40
forb.
3'Bu(H->4p)
7.76
0.051
7.09
0.046
7.46
0.006
2'A u (H^4p)
7.60
0.003
7.28
0.003
7.46
0.000
5'A g (H^3d)
7.29
forb.
7.87
forb.
7.86
forb.
4 Bll(H->4p)
7.79
0.279
7.13
0.255
7.87
0.029
2'B g (H^3d)
7.78
forb.
7.74
forb.
7.93
forb.
6'Ag(H->3d)
7.61
forb.
7.70
forb.
7.99
forb.
3'B g (H^3d)
7.84
forb.
7.84
forb.
8.10
forb.
7'A g (H^3d)
7.76
forb.
8.09
forb.
8.22
forb.
l
Table 2. Excitation energies (AE) and oscillator strengths for anti n-tetrasilane employing the Si[6s5p2d]/H[2slp] ANO-type basis set [98], CASSCF State
CASPT2
MS-CASPT2
AE (eV)
Osc. Str.
AE (eV)
Osc. Str.
AE (eV)
Osc. Str.
l'B u (Vi) 2'A g (V 2 )
7.98
1.542
6.40
1.237
6.36
1.175
7.15
forb.
6.69
forb.
6.68
forb.
0.003
6.66
0.002
6.66
0.002
1 'AU(V3)
7.39
3'A g (V 4 )
7.39
forb.
6.96
forb.
6.96
forb.
2 l B ll (V 5 )
7.84
0.121
6.88
0.107
6.92
0.165
l'B g (V 6 )
7.82
forb.
7.12
forb.
7.12
forb.
71 However, by computing instead three roots of 'Bu symmetry with the valence basis set, the lowest excited state moves to higher energy destroying the nice agreement with the experimental datum and the assignment for the third root is really uncertain, somewhat between valence and Rydberg character. Here we have again an unbalanced situation of valence-Rydberg mixing! Despite the one-electron basis set does not have Rydberg functions, the computation tries to simulate the 3'BU(H—>4p) Rydberg state, as much as the diffuseness of the basis set allows. Thus, one can get good results of valence excited states using a valence basis set as long as only valence states are computed. In other words, the right number of roots for the average CASSCF step has to be considered, making sure that at the CASSCF level the valence states of a given symmetry are more stable than the Rydberg states. Unfortunately, that theoretical information comes out only from the full computation with the extended basis set. The conclusion is, therefore, that the number of valence excited states within an energy interval can only be determined from the complete consideration of both valence and Rydberg states, as it also occurs in organic compounds [15-18]. In order to achieve the goal, flexible enough basis sets have to be supplied, employing high-level methodology with inherent flexibility to overcome the possible erratic valence-Rydberg mixing. The MS-CASPT2 method in conjunction with ANO-type (valence and centred diffuse) basis set is certainly one of the low-cost possibilities. The reader can find additional discussions on the spectroscopic features of the system, as well as comparison to earlier CIS results, in the original publications [98, 99]. 5.3. Computational Strategies. An Illustration: Cyclooctatetraene. The thermal and photochemical reactivity of cyclooctatetraene (COT) has been very well studied from both experimental and theoretical standpoints (for a recent contribution see Ref. [101]). The electronic spectra of COT have comparatively received less attention, so we decided about a couple years ago to go deeply into the subject. The electron energy-loss spectrum (EELS) [102] of cycloocta-l,3,5,7-tetraene, at 50 eV impact energy and a scattering angle of 10°, can be described as a broad band of low intensity over the region 4-4.8 eV with a maximum at 4.43 eV and an intense band peaking at 6.42 eV, which has a shoulder around 6 eV. With these experimental conditions the observed features can be considered to be essentially singlet—>singlet transitions [102]. The study was mainly addressed for determining the nature of those observed bands. In principle, it was expected that they could be attributed to the electronic transitions calculated vertically. The research was subsequently extended by including the singlet—^triplet spectrum (relevant for the understanding of COT as triplet quencher), lowest ionization potentials, and electron affinity of COT. They shall also be briefly considered in turn. In order to design the calculation one has to decide about the equilibrium geometry of the ground state, active space, and basis set. Previous work on the COT system clearly revealed that the ground state of neutral COT has four equivalent D2C/ local minima connected by two independent reaction paths: ring inversion (with a D41, transition state) and bond shifting (through a Dgh transition state) (see the 1996 landmark paper of Wenthold et al. [103]). Because of the tub-shaped structure belonging to the Z^tf symmetry (see Fig. 3), both through-bond (0-71 interaction or
72
hyperconjugation) and through-space interactions have to be taken into account, which is a true challenge for any theoretical method [104]. On the other hand, it was also known at the time we started the project that the lowest triplet state has an octagonal structure, and that the ground state of COT radical anion is also planar belonging to the D^ symmetry. A consistent treatment for the ground-state geometry optimization of these systems (neutral, lowest triplet, and radical anion) was carried out at the CASSCF level employing the n valence MOs (and the respective n electrons) active (denoted hereafter as 71-CASSCF). As illustrated in many applications, analysis of the nature and spacing of the canonical MOs is usually helpful to rationalize the most important spectroscopic features obtained from more complex CASSCF wave functions. It also serves as a qualitative guide for predicting the type of valence excited states to be computed in order to choose an active space in accordance with the expectations. This is an important step because the active space has to be supplied by the user and has, therefore, to bear enough flexibility to describe all possible type of excited states. From the electronic structure shown schematically in Fig. 4, at least five candidates can be expected as low-lying singlet excited states: •
The ]A2 state described mainly by the HOMO^LUMO one-electron promotion.
•
States of 'E symmetry involving the HOMO^LUMO+1 and HOMO-l^LUMO singly excited configurations. Because of the similar orbital energy differences (12.4 eV versus 12.5 eV), these nearly degenerate one-electron promotions can further interact leading to a plus and minus linear combinations in the actual CASSCF wave function. As a result of the interaction, the minus and plus state are pushed down and up, respectively. The vertical transition from the ground to the minus excited state can be predicted with low intensity because of the subtraction of the corresponding transition moments. On the other hand, transition involving the plus state should carry most of the intensity. (Notice, however, that intensities are not computed but oscillator strengths).
•
The 'Bi state described primarily by the singly excited configuration from the deepest n orbital, 4b2, to the LUMO (3a2).
•
The ]B2 state involving the highest occupied a orbital, 3bi, and the LUMO.
Transitions from the ground to the !A2 and 'Bi states are dipole forbidden in D211 symmetry, as it is the expected transition to the r A| state of doubly excited character. In addition, experience shows that in systems of a similar molecular size, Rydberg states converging on the lowest ionization potential are interleaved among the valence excited states. Therefore, the lowest 3s, 3p, and 3d members of the Rydberg series have also to be taken into account. The Rydberg states are described primordially by one-electron promotion from the HOMO (5aj) to the 3s, 3p, 3d atomic-like MOs covering the whole molecule. The aim of the study is then clearly defined and it only remains to select the active spaces
73 accordingly. In this case, the process was somewhat laborious from a technical standpoint because the D2d symmetry is not implemented in the software employed (MOLCAS package) [71]. Therefore, actual calculations were performing in C2V symmetry. As usual, the 7i-valence active space was extended to include Rydberg orbitals of the different symmetries, as appropriate (see details in Ref. [105]).
Fig. 3. The ground-state structure of cycloocta-1,3,5,7-tetraene.
i
K (eV)
5blW
+6.75
+4 13
9e(7t)
9c(jt)
3a2(;t)
+2.41
LUMO HOMO
-8.^6
10 11 -11.84 -13.05
7CO0
7c(7t)
4b 2 W
3b,(a)
Fig. 4. Schematic electronic structure of cycloocta-1,3,5,7-tetraene including the highest five occupied and 7r-valence virutal canonical MOs, together with the orbital energies, computed with the ANO-type C[4s3pld]/H[2slp] basis set at the TI-CASSCF equilibrium geometry.
74
The CASSCF and CASPT2 results for the vertical singlet—^singlet electronic transitions obtained with the ANO-type C[4s3pld]/H[2slp] + 2s2p2d (Rydberg functions) basis set are listed in Table 3. The MS-CASPT2 and CASPT2 findings lead essentially to the same picture of the electronic spectrum. For the sake of simplicity only CASPT2 results shall next be considered. The computed vertical excitation energy, 3.79 eV, is somewhat too low compared to the maximum observed by electron energy-loss spectroscopy, 4.43 eV. The previous ab initio MRCI results reported by Palmer [106] predicted the lowest-energy band to peak at 4.37 eV. The source of the discrepancy between the MRCI results and the CASPT2 finding can be attributed to the limited basis set (split-valence quality) used for the MRCI calculations. Table 4 compiles the 7T-CASPT2 results computed for the lowest vertical transition l'Ai—»l'A2(5ai—»3a2) upon improving the contraction scheme. Employing the ANO-type basis set with the contraction C[3s2p]/H[2s] the CASPT2 result, 4.45 eV, is consistent with the earlier MRCI result. Thus, the message is clear. For the lowest singlet—^singlet vertical transition of COT, the MRCI and CASPT2 results yield about 4.4 eV when a split-valence basis set is employed. The agreement with experiment is perfect. If we define a right theoretical result as it matches with the experimental datum, this is the typical "right answer" for the "wrong reason". Upon improving the quality of the contraction in the ANO-type basis set, which always has the same number of primitive functions, the nice agreement with the experimental datum is destroyed. Adding d polarization functions on the carbon atoms, the computed vertical transition drops to 3.92 eV (a similar result was obtained at the CASPT2 level with the 6-31G* basis set at a slightly different geometry) [101]. Inclusion of/? polarization functions on the hydrogen atoms has a minor influence on the transition energy. A slight decrease of the excitation energy also occurs employing the C[4s3pld]/H[2slp] contraction. Adding 2s2p2d diffuse functions at the symmetry centre of the molecule, a further decrease takes place (0.11 eV with respect to the valence plus polarization basis set) and the excitation energy so computed, 3.79 eV, is similar with the result obtained with the largest valence basis set explored, 3.80 eV, which includes up to/and d functions on the carbon and hydrogen atoms, respectively. In this manner, the best theoretical estimation for the lowest vertical transition is computed to be within the energy range 3.80-4.00 eV, about half an eV below the peak observed experimentally. It seems to point out that the maximum of the low-intensity band does not correspond to the vertical excitation. Although the transition is dipole-forbidden, it is observed optically, probably through a vibronic coupling mechanism with a nearby dipole-allowed transition. If higher accuracy on theoretical grounds were required, vibronic resolution of the band would have to be performed. One of the main advantages of an ab initio approach relies on the fact that one is aware of the limitations introduced in the study knowing the direction to push further (if required) the methodology. It is a well-defined hierarchical framework. Pioneering INDO results [107] already predicted the l'A2(5ai—>3a2) state around 4 eV but the deviation with respect to the experimental datum was ascribed to the inherent limitations of the semiempirical methods. On the other hand, we were curious about the performance of the TDDFT method with the B3LYP functional to describe the electronic spectrum of COT. Employing the same geometry and the ANO-type C[4s3pld]/H[2slp] + 2s2p2d basis set, the lowest singlet excited state was found to be at 3.51 eV. Without the CASPT2 results at hand,
75
the large deviation with respect to experiment obtained by using semiempirical or TD-DFT methods could then be related to the poor performance of the methods. What are the open alternatives to solve that? Just one, to move to another parametrization and see it the result gets closer to the experimental datum. In summary, the predictive power is lost and we would have missed the main key point coming out from the ab initio research carried out at the CASSCF/CASPT2 level, namely the observed feature cannot be related to the vertical transition and most probably vibrational resolution of the band might be required to fully characterize it. Table 3 Computed CASSCF and CASPT2 excitation energies (eV) and related oscillator strengths for the vertical singlet—»singlet electronic transitions of cycloocta-1,3,5,7tetraene employing the ANO-type C[4s3pld]/H[2slp] + 2s2p2d (Rydberg functions) basis set [105], State
CASSCF
CASPT2
Osc. Str.
EELS data"
l'A 2 (5a,->3a 2 )
6.70
3.79
forbidden
4.43
l'E(7I7I*)
7.88
5.56
0.0075 forbidden
2'A,(5ai->3s)
5.81
5.58
21E(5a1->3px.y)
6.69
5.93
0.0004
3'A,(7nr*)
6.84
6.14
forbidden
l'B 2 (3b,->3a 2 )
10.09
6.14
0.011
2'B2(5a,->3pz)
6.21
6.17
0.0532
1 'B,(4b2->3a2)
8.02
6.36
forbidden
3'E(7C7t*)
10.28
6.40
1.1096
5a,-»3d
6.75-7.35
6.57-6.80
0.0014
"Observed by electron energy-loss spectroscopy (EELS) [102].
Table 4 Convergence pattern for the lowest vertical transition 1 'A|—»1 'A 2 (5ai^3a 2 ) upon improving the contraction scheme. Basis set*
7I-CASPT2 (eV)
Previous (eV)
C[3s2p]/H[2s]
4.45
4.37b
C[3s2pld]/H[2s]
3.92
4.00°
C[3s2pld]/H[2slp]
3.92
C[4s3pld]/H[2slp]
3.90
C[4s3pld]/H[2slp] + 2s2p2d
3.79d
C[5s4p2dlf]/H[3s2pld]
3.80
"Primitive sets: C(14s9p4d3f)/H(8p4p3d) ANO-type basis set b
MRCI result [106].
C d
TI-CASPT2/6-31G* result taken from Ref. [101].
From Table 3.
6.02 6.42
76 Coming back to Table 3, the following remarks are pertinent: •
The most intense feature is related to the 3'E(7ITI*) state, the plus state described above. It is placed at about 10 eV at the CASSCF level and the CASPT2 result, 6.40 eV, is in agreement with experiment. The effect of dynamic correlation is crucial for the accurate location of the state. The l'E(7r7i*) state corresponds to the minus state and is close to the lowest Rydberg state.
•
The 3'Ai(7i7i*) state has a prominent weight (33.2%) of the doubly excited configuration (HOMO->LUMO)2 in the CASSCF wave function. The 3'AI(TI71*) and l'B2(o7i*) states are degenerate. The most plausible assignment responsible of the observed shoulder at 6.02 eV is the Rydberg transition to the 3pz orbital, although the valence state an* might also contribute to this feature in the gas phase. The primary Rydberg character for the shoulder recorded in the gas phase is supported by the fact that it is not observed in the absorption spectrum of COT in hexane. It is well recognized that Rydberg states are usually perturbed in condensed phases and they collapse in solution. Many different earlier assignments for the observed shoulder can be found in the literature [105]. However, the issue is now clarified theoretically and it would be highly desirable that the assignment could be confirmed unambiguously by experimental research.
As we see both valence and Rydberg states coexist in the same energy region. It is more a rule than an exception for molecules of medium molecular size [15-18]. COT is employed for efficient laser operation of dye solutions because of the unique properties of its lowest triplet state. During the operation of laser dye solutions, the triplet excited levels of the dyes are populated along with the singlet states, which causes a detrimental effect in their operation. A fraction of the excited singlet state population responsible for the laser action becomes deactivated by the intersystem-crossing mechanism. These triplet dye molecules exhibit broad optical absorption triplets-triplet spectra with relatively high intensities, with the consequent loss of laser efficiency. Hence, it is essential to use a suitable triplet scavenger, able to remove the triplet dye molecules, without interfering in the laser efficiency. The acceptor COT fulfils the requirement and it is widely used for this purpose. As can be seen from Fig. 5, where the main CASPT2 results for the So—»Tj transition are depicted, the energy difference between the vertical and adiabatic excitation energies is large, about 2 eV. Simultaneous to the electronic excitation of COT, a progressive structural reorganization towards planarity takes place. Therefore, COT has a pronounced non-vertical behavior and covers a wide range of triplet donors, D*(Ti). The origin of the So^Ti transition, about 0.8 eV, can be considered as an estimate of the lower limit for the triplet energy of a donor that the acceptor COT could still react with [108]. On the other hand, the vertical phosphorescence is predicted in the infrared range.
77
vertical 1
i
V emission
adiabatic
' "}-0.22eV
~i U 0.78 eV
2.82 eV '
0
J
So"
J
C
Fig. 5. The lowest singlet—^triplet electronic transition. CASPT2 results for the vertical absorption, vertical emission (phosphorescence), and adiabatic excitation energies for COT. The triplet energy of a donor D is also represented.
Additional information about the lowest triplet state was obtained from the photoelectron (PE) spectrum of the radical anion, where photodetachment to two distinct electronic states of neutral COT was observed. Wenthold et al. [103] identified these electronic states as: •
The l'Aig (D4h) state at 1.1 eV, which corresponds to the transition state of COT ring inversion along the So hypersurface, and
•
The l3A2g (D8h) state; the lowest triplet state, at 1.62 eV.
Employing the ground state structure optimized for the COT radical anion, the PE spectrum was computed. At the CASPT2 level, the ground state of the neutral system is found to be at 1.11 eV and the lowest triplet state at 1.47 eV in reasonable accordance with the experimental data. As can be seen in Table 5, the results at the CASSCF level are poor. The minus sign in the CASSCF result (-0.86 eV) implies that the radical anion is not bound. Therefore, the CASPT2 approach is capable of recovering (qualitatively and quantitatively) the right relative position between the respective states of the anion and the neutral systems. It is really amazing, especially if one realizes that CASPT2 is just a second-order perturbation approach. In this type of difficult cases, involving large differential dynamic correlation contributions, the CASPT2 method certainly plays an outstanding role. Similar comments are valid for the computed electron affinity. The vertical EA is negative at both CASSCF and
78 CASPT2 levels of theory. Therefore, we can confidently conclude that the radical anion is not actually bound at the ground-state equilibrium geometry of COT (1 Ai (D2d))- In addition, the CASPT2 adiabatic EA, 0.56 eV, is in agreement with recent experimental determinations (0.57 eV) [109]. The agreement is entirely due to the inclusion of dynamic correlation because at the TI-CASSCF level the ground-state radical anion is above the ground state of the neutral system by nearly one and half eV (-1.43 eV in Table 5). Table 5 Computed photoelectron spectrum for the planar cyclooctatetraene radical anion and electron affinity of COT employing the ANO-type C[4s3pld]/H[2slp] + 2s2p2d (Rydberg functions) basis set. Energy differences in eV [105], State
TI-CASSCF
71-CASPT2
Experimental
Ground state of cyclooctatetraene radical anion: 1 B hl (D4i, symmetry) -0.86
1.11
1.10a
0.16
1.47
1.62a
vertical EA
-2.79
-0.49
adiabatic EA
-1.43
0.56
1'Aig 3
l A 2e
0.57b
"Taken from the recorded photoelectron spectrum [103]. b
See Ref. [109] and cited therein.
5.4. Up-to-date Theoretical Spectroscopy. We shall leave for the next section the purely photochemical problems, that is, those in which different photoproducts are generated or those in which non-adiabatic state transitions occur, and focus here on spectroscopy, understood as the assignment of absorption and emission band positions and intensities, radiative lifetimes, and environmental effects. Theoretical ab initio spectroscopy can provide nowadays extremely accurate data that help to interpret and rationalize the experimental recordings and to predict new findings. Not all systems can be equally computed at the same level of accuracy. Excitation energies and oscillator strengths for organic systems up to the size of, for instance, the free base porphin molecule (C20N4H14), have been studied using accurate methods and basis sets: CASPT2 [110], in which a novel interpretation of the spectrum was put forward, SAC-CI [111], and EOM-CCSD [112], although the single-configuration methods showed less accuracy. In order to compute the low-energy spectra of larger systems such as fused zinc porphyrin dimers (Zn2C4oNsH22) at the SAC-CI level [56], severe approximations, such as lack of polarization functions in the basis sets or partial removal of virtual orbitals, were performed, undoubtedly decreasing the accuracy of the results. As a rough estimation, error boundaries smaller than 0.3 eV are required in order to obtain reliable interpretations of many spectra. DFT approaches for excited states, TD-DFT theories basically, were expected to be able to deal, although at low level of accuracy, with large systems were ab initio methods become too expensive. Unfortunately, recent findings have proved that the TD-DFT methods fail
79 dramatically in too many situations: charge transfer states [113], multiconfigurational states [113], doubly or highly-excited states [61, 113], valence states of large it extended systems such as acenes, from naphthalene to octaacene [114, 115], polyacetylene fragments or oligoporphyrins [116], polyenes, from butadiene to decapentaene [117], and the list increases every day. In some cases the errors are larger than 5 eV [114]. Regarding inorganic electronic spectroscopy, only multireference perturbation theory, CASPT2 basically, has been able to obtain general and accurate results in systems so different as ionic transition metal (TM) molecules, covalent actinide complexes or organometallic metal-ligand compounds. Typical examples are chromium hexafluoride and hexachloride anions, iron porphyrins, tetra-, pentaand hexacarbonyl or cyano TM complexes, TM dihalides, cyclometalated compounds, blue copper protein chromophores, and lanthanide and actinide oxides [16, 118-121]. The virtual extension of the multiconfigurational approaches to systems with several transition metal atoms is complicated because the selection of the reference becomes challenging [16, 118]. In the other side of the scale, ab initio methods can yield extremely accurate information for small systems, provided that high-level approaches and large basis sets are used. In these cases, the required accuracy is not far from the usually recognized as chemical accuracy, 1-2 kcal/mol. Not only electronic data are needed, also detailed vibrational or rotational spectroscopic information, and, typically, also vibronic or spin couplings have to be included. In very small systems, the MRC1 method can be considered extremely accurate and general, provided that the problem of the size-extensivity is corrected or estimated [5, 7]. If the system fulfils certain requirements such as a closed-shell ground state well represented by a single reference and excited states of clear singly excited character, single-configuration coupledcluster approaches including triple excitations, EOM-CCSD(T) or CC3 for instance, may offer high accuracy. Those methods are size extensive and can in practice be extended further than the MRCI approaches. In any case, the only generally applicable methods are the multireference perturbation approaches, which means, CASPT2 and related. CASPT2 is a non size-extensive methodology but, in practice, it can be shown that, in the calculation of spectroscopic properties, the corresponding effect is negligible [16]. The expected accuracy, being simply a second-order perturbation theory, cannot be as large as more elaborated approaches, except for the fact that it does not present unexpected failures in difficult cases, assuming a computation free of intruder states. As an example of the required accuracy needed to solve spectroscopic problems, a CCSD(T) study of the ground state of the van der Waals Ar-CO complex required the use of a basis set composed by aug-cc-pVQZ plus midbond functions in order to get an accuracy close to 0.3-0.4 cm" and assign conflictive rovibrational bands [122]. In general, methods for electronic excited states cannot reach the same precision. Last decade has known tremendous breakthroughs in the field of quantum chemistry of the excited state. The number, size, and accuracy of the computed problems have grown up to the point of being comparable in certain cases with the experimental measurements, in particular for gas-phase spectroscopy. Solvent simulations in spectroscopy, basically by the Reaction Field (RF) or Quantum Mechanics/Molecular Mechanics (QM/MM) approaches, cannot be
80 considered quantitative so far, although they are helpful to elucidate spectroscopic assignments [2, 123, 124]. If we summarize a number of achievements made by modern quantum-chemical theories in the field of spectroscopy, it is worth remembering that, nowadays, all type of states can be computed accurately, whether valence, Rydberg or multipole-bound anionic states, optically one-photon allowed or forbidden (dark) states, and covalent, ionic, and zwitterionic states [18]. Band origins (Te or To transitions) [15-18, 42, 125] and vibrational profiles for electronic absorption and emission bands, involving ground and excited states geometry optimizations and knowledge of the states force fields are also computed with high accuracy for medium size systems such as benzene [90], pyrrole [126], and /i-benzosemiquinone radical anion [127] leading to straightforward comparisons with experiment. Even the effects of the anharmonicities in the vibrational bands positions and intensities can be computed at different levels for, at least, small systems. As an example, the low-lying absorption and emission spectra of the formyl radical obtained at the CASPT2 level, which required the calculations of quartic potentials built by computing hundreds of points in the hypersurfaces [128]. Methods to incorporate vibronic couplings at different levels and obtain refined effects on the intensity of the vibrational bands, become also available, although at high cost [128-130]. Examples of the inclusion of accurate calculation of vibronic couplings considering the interaction of several electronic states include the CASSCF/MRCI description of the S2(TI7I*) absorption band of pyrazine [130], the Green's function treatment of the photoelectron spectrum of benzene [130], and CASPT2 and EOMCCSD studies of pyridazine and pyrimidine [131, 132]. Other consequences of the breakdown of the Born-Oppenheimer approximation such as the Jahn-Teller and Renner-Teller couplings have been widely studied in small systems, where high accuracy is needed [4]. Finally, spinorbit couplings and relativistie effects computed at ab initio levels are becoming generally available for the excited states of systems including heavy atoms [4]. An example is the recent implementations of the combination of two-component relativistie formulations using a Douglas-Kroll Hamiltonian to incorporate the scalar effects and the use of multiconfigurational CASSCF/CASPT2 or shifted RASSCF methods with relativistie basis sets to solve the spin-orbit Hamiltonian. This approach proved to get errors in the relativistie effects negligible if compared with the accuracy of the methods to account for the correlation energy [133]. The whole previous discussion leads to one simple conclusion: within certain limitations related to the size of the systems, quantum-chemical methods applied to theoretical spectroscopy have reached the point where a real and constructive interplay can be established with experiment [134-137]. Both approaches, experimental and theoretical, will become more accurate in different cases. For instance, nowadays, none experimental determination can probably match the theoretical calculation of the ground-state structure for an isolated molecule, that is, modeling the system in the vapor phase. In other cases, such as hyperfine couplings at different levels or situations where the environment produce fine effects, the theoretical methods do not have enough accuracy as compared with recorded data. Apart from energies, excited states properties and transition probabilities are now routinely computed for many systems. In some cases, such as multipole moments in excited states, the
81 accuracy reached by the theoretical methods is also unmatched by the experimental measurements. Many representative examples can be given of this new era in the quantum chemistry of the excited state in which the ab initio methods, especially the multiconfigurational CASPT2 approach [15-18], have been the main protagonists. Considering the confirmed weaknesses of the TD-DFT theory to deal with excited states and the accuracy needed to solve spectroscopic problems, the ab initio methods will probably be the basic tool in the near future. Better implementations of the methods and development of efficient geometry optimizers will be required to proceed and they are, indeed, becoming available [84]. 6. EXCITED STATES AND PHOTOCHEMISTRY This section is devoted to the computation of excited states specifically involved in photochemical reactions, that is, reactions initiated by light. The borderline between spectroscopy and photochemistry is extremely dim and vague. We can jump from one area to the other without even notice it. For instance, if one is interested in the calculation of the vertical excitation energies of cytosine [138], the results produced are certainly in the area of theoretical spectroscopy. Now, let us assume one wants to give a step forward by computing the equilibrium structures of the main valence singlet excited states [139], namely '(TOT*), '(no?!*), and '(n^*), one immediately enters in the field of non-adiabatic photochemistry. The CASSCF geometry optimization of the '(riN7t*) state (where nN refers to the lone pair located on the nitrogen atom) leads directly to a conical intersection with the ground state. On the other hand, the CASSCF equilibrium structure for the '(JITT*) state is essentially coincident with a conical intersection involving the excited states '(TTTT*) and '(nojr*). There is no problem to reach the minimum for the '(noft*) state, which becomes the lowest excited state at the CASSCF level [139]. However, when dynamic electron correlation is taken into account the photochemical picture is somewhat different, becoming the '(7171*) state the most stable [140]. Incidentally, the computation of these spectroscopic properties of cytosine by employing the CIS method, with the purpose in mind of getting a rapid qualitative vision of the situation, becomes a nightmare, facing all sort of "convergence" problems (not surprisingly), leading to meaningless results where the '(no7t*) state is completely missed. Let us start from the very beginning. Considering the excited and ground state potential energy surfaces and the different reaction paths that a system might evolve through, the molecular processes can normally be identified as photophysics, adiabatic photochemistry, and non-adiabatic photochemistry [141]. Absorption and emission can be regarded as photophysical processes. From the theoretical viewpoint they involve calculations at similar molecular structures. In an adiabatic reaction path, once that the vertical absorption takes place, the system proceeds along the hypersurface of the excited state to reach a local (or absolute) minimum leading eventually to an emitting feature. For instance, the dual fluorescence observed for dimethylaminobenzonitriles [142] and 1-phenylpyrrole [143] in polar solvents can be explained in terms of a photoadiabatic reaction that takes place in the lowest excited state. In those cases, the polar environment decreases the reaction barriers and
82 favors the process. In a non-adiabatic photochemical reaction path, part of the reaction occurs on the excited state hypersurface and after a non-radiative jump at the surface crossing (or funnel) continues on the ground state hypersurface. When the two hypersurfaces have the same multiplicity (e.g. singlet/singlet) the radiationless jump is denoted as internal conversion (IC), and intersystem crossing (ISC) is reserved for cases of different multiplicity (e.g. singlet/triplet). Internal conversion may occur through an avoided crossing (AC) or a conical intersection (CI). Among several researchers, Robb, Olivucci, Bernardi, and co-workers have specifically shown during the last decade the important role that conical intersections play in organic photochemistry [141]. A large number of photochemical reactivity problems has been studied in the last years involving CIs, including photoisomerizations, photocycloadditions, photorearrangements, and photodecompositions. Depending on the nature of the CI [144], the corresponding radiationless transition can yield specific photoproducts or relax the energy towards the ground-state initial situation. Geometry determination of a conical intersection, as well as localization of minima and transition states, is usually performed at the CASSCF level. In a subsequent step, the energy differences are corrected by including dynamic correlation. If it is done at the CASPT2 level, the protocol is denoted as CASPT2//CASSCF, which stands for geometry optimization at the CASSCF level and CASPT2 calculation at the optimized CASSCF structure. Two main situations do actually occur. In cases where the PES computed at the CASSCF and CASPT2//CASSCF level behave approximately parallel (CASE A), the CASSCF optimized geometries will be in general correct, despite they have been computed at a lower level of theory. It means that dynamic correlation contributions are quite regular and similar in ample regions of the PES. The photochemistry of the protonated Schiff bases constitutes a nice CASE-A example, where the CASPT2//CASSCF computational strategy can be confidently applied. As can be seen in Fig. 3 of Ref. [145], the CASSCF minimum energy path runs parallel to that obtained at the CASPT2//CASSCF level. When dynamic correlation is markedly different for the states considered and varies significantly along the PES of interest, geometry optimization has to be carried out at the highest correlated level (CASE B). Otherwise, the uneven contributions of dynamic correlation may lead to unphysical crossings and interactions between the two electronic states. A clear representative study of CASE B corresponds to the characterization of the nature of the So/Si crossing responsible for the radiationless decay in singlet excited cytosine. The excited DNA bases have a lifetime so small that they relax to their ground state before a photochemical reaction may take place. In fact, the excited-state lifetimes of the nucleic acid molecules fall in the sub-picosecond time scale, suggesting the presence of an ultrafast internal conversion channel [146, 147]. It is an intrinsic molecular property because very short lifetimes have also been determined in the gas phase for the isolated purine and pyrimidine bases [148]. The CASPT2 results [140] suggest that the conical intersection between the ground state and the nn* state, denoted by (gs/7t7i*)ci, is responsible for the ultrafast decay of singlet excited cytosine, which is in contrast to the picture offered by the CASSCF method [139]. Moreover, the no7i* state is involved in a S2/Si crossing and it does not contribute directly to the ultrafast repopulation of the ground state [140]. As stated above, optimization of the singlet nNit* state leads directly to
83 a conical intersection with the ground state but it is not found to be the preferential path of the observed decay. Whether this is a general relaxation mechanism for all the excited nucleobases or not is the subject of current research. A situation like cytosine where the CASSCF and CASPT2 reaction paths do not run parallel (CASE B) manifests an urgent necessity of efficient algorithms for computing conical intersection with inclusion of dynamic correlation. The two main open routes available at present, through the MRCI [149] and the MS-CASPT2 methods [150], are limited to systems of small molecular size. However, in order to tackle general CASE-B problems, where the CASPT2//CASSCF (or MRCI//CASSCF) protocols are not valid, no methodology is available in practice. It is clear that most of the biomolecules of interest cannot be confidently treated today at the MRCI level because of the severe truncations that have to be performed. On the other hand, caution has to be exercised when applying the MS-CASPT2 method to locate conical intersections [151], which is next addressed by using as example the penta-2,4dieniminium cation. De Vico et al. [152] have recently reported the optimized structures for the Si /So conical intersection computed at the MS-CASPT2 and CASSCF levels, hereafter denoted as Geom. I and Geom. II, respectively. Table 6 shows the CASSCF, CASPT2, and MS-CASPT2 energy differences (AE) between Si and So that we have computed at those geometries. The 6-31G* basis set was used throughout. Incidentally, because photochemical studies are mainly related to the lowest valence states, basis sets smaller than those used in spectroscopic studies, are frequently employed, which should be alright as far as no competitive Rydberg states are placed around the studied region. When the energy difference is less than 2 kcal/mol, the minimum reached is considered technically as a conical intersection; otherwise (AE > 2 kcal/mol) we are facing an avoiding crossing. The MS-CASPT2(6MOs/6e) result, 3.89 kcal/mol, is similar to the CASPT2 finding employing Geom. I, and the off-diagonal matrix elements of the asymmetric effective Hamiltonian (Heff) are small (less than 2 kcal/mol). Everything seems to be quite consistent. Apparently an avoiding crossing has been found at the MS-CASPT2(6MOs/6e) level. Using Geom. II, the CASSCF(6MOs/6e) and CASPT2(6MOs/6e) results for AE are within 1 kcal/mol. It indicates that the optimal geometry determined for the conical intersection at both levels of theory is probably very similar. However, the states become separated by 7.57 kcal/mol when they are allowed to interact. As can be seen in Table 6, the off-diagonal elements of the Heff are very different, 6.12 and 1.38 kcal/mol. Because the states are nearly degenerate at the CASPT2 level, the result for the off-diagonal symmetric Heff just comes out from averaging: (6.12+1.38)/2. As a consequence, the CASPT2 states are pushed down and up by that amount, 3.75 kcal/mol. Such interaction is definitely unphysical! Enlarging the active space with two extra orbitals (8MOs/6e results), which allows for radial correlation of the electrons involved in the 90°twisted double bond, the Hi 2 and H21 asymmetric elements become small enough, which reflects that the corresponding zeroth-order Hamiltonians are capable of yielding a balanced description for both states. Accordingly, the CASPT2 and MS-CASPT2 splitting between the Si and So states becomes small. In summary, the computed geometry at the
84 CASSCF(6MOS/6e) level represents also a conical intersection at the MSCASPT2(8MOs/6e) level, which confirms that protonated Schiff bases behave as CASE A. Unfortunately, in larger molecular systems the active space cannot be extended to the extreme that the off-diagonal elements become less than 2 kcal/mol and the MS-CASPT2 method may be forced to yield avoiding crossings. It is certainly a circumstance to be prevented in future applications of the MS-CASPT2 method. Table 6 Energy difference between Si and So, AE, computed at the optimized structures of the penta-2,4-dieniminium cation Si/S0 conical intersection3 at the MSCASPT2 (Geom. I) and the CASSCF (Geom. II) levels. The 6-31G* basis set was used throughout. The off-diagonal elements of the MS-CASPT2 effective Hamiltonian (Heff) are also included. Method
Geom. I
Geom. II
(6MOs/6e)
(6MOs/6e)
Geom. II (8MOs/6e)
AE(S,-So) (kcal/mol) CASSCF
3.40
0.07
4.78
CASPT2
3.83
0.99
0.60
MS-CASPT2
3.89
7.57
0.64
eff
Off-diagonal elements of H
(kcal/mol)
H|2(asymmetric)
0.62
6.12
0.18
H2| (asymmetric)
0.09
1.38
0.04
Hi2=H2] (symmetric)
0.35
3.75
0.11
"Optimized geometrical parameters taken from De Vico et al. [152].
7. FINAL REMARKS It is clear that we are living in a new era where experimental and theoretical research can talk to each other on an equal footing. In this privileged situation we should be able to join efforts addressed to elucidate the big challenges our society faces at present in the realms of atmospheric chemistry, material science, photobiology, and nanotechnology. Experimental and theoretical research work shares at least one characteristic: the results produced have to be interpreted. The most cumbersome task is to compare experimental and theoretical derived data properly. In many cases recorded values do not directly yield the studied property, which has to be obtained by indirect procedures within a given scheme. On the other hand, theoretical results are usually obtained for simplified models. The resolution of the scientific problem certainly requires a constructive interplay between both viewpoints. We must be able to design a research strategy in computational chemistry (RESICC) leading to results with predictive character, independent of any experimental information. Fig. 6 shows a proposed RESICC algorithm. Basic steps include:
85 1. Define objectives. This is surely one of the most important parts of a research. The aim of the study has to be clearly defined, as precisely as possible: What is the purpose of the computation? 2. Literature reviewing: What is the scientific background on the topic? Analysis of previous information has to be critically reviewed with open mind, because it can be extremely helpful to design the computation. 3. Actual computation. According to the previous steps the actual computation takes place at a given level of theory. 4. Once the results have been carefully analyzed two key questions rise. Are the obtained conclusions stable with respect to further theoretical improvements? Do they fulfill the initial objectives? 5. A proper action has to be taken if the calculation does not guarantee the required levels of quality. Theory has to be pushed further until stable conclusions are achieved. It is worth noting that ab initio methods, because of their well-defined hierarchical structure that allows convergence of the results upon the increasing level or theory, are currently the only type of quantum-chemical tools able to fulfill the requirements implicit in the RESICC scheme. The decision is up to you!
f
START
Define Objectives Literature Reviewing/
Actual Computation Improved Level of Theory Stable \ ^ .Conclusions?,
No
Fig. 6. Research strategy in computational chemistry.
86 8. ACKNOWLEDGMENTS We thank our co-workers for their valuable contributions. MCYT of Spain, projects BQU2001-2926 and BQU2004-01739, and Generalitat Valenciana, project GV04B-228, have financed the research.
REFERENCES [I] A. Szabo and N.S. Ostlund, Modern Quantum Chemistry, McGraw-Hill, New York, 1989. [2] C.J. Cramer, Essentials of Computational Chemistry, Wiley, Chichester, 2002. [3] T. Veszpremi and M. Feher, Quantum Chemistry. Fundamentals to Applications, Kluwer, Dordretch, 1999. [4] P. Jensen and P.R. Bunker (eds.), Computational Molecular Spectroscopy, Wiley, Chichester, 2000. [5] B.O. Roos and P.E.M. Siegbahn, Int. J. Quantum Chem., 17 (1980) 485-500. [6] S. Peyerimhoff, Spectroscopy: Computational Methods, in Encyclopedia of Computational Chemistry, P.v.R. Schleyer et al. (eds.), Wiley, Chichester, 1998, pp. 2646-2664. [7] R.J. Buenker, G. Hirsch, Y. Li, J.-P. Gu, A.B. Alekseyev, H.-P. Liebermann, and M. Kimura, Ab Initio Calculations of Excited State Potential Functions, in Computational Molecular Spectroscopy, P. Jensen and P. R. Bunker (eds.), Wiley, Chichester, 2000, pp. 135-168. [8] J.L. Heully, J.P. Malrieu, I. Nebot-Gil, and J. Sanchez-Marin, Chem. Phys. Lett., 256 (1996) 589-594. [9] E.R. Davidson and L.E. McMurchie, Ab Initio Calculations of Excited-State Potential Surfaces of Polyatomic Molecules, in Excited States, E.C. Lim (ed.), Academic Press, London, 1982, pp. 1-35. [10] K. Andersson, P.-A. Malmqvist, and B.O. Roos, J. Chem. Phys., 96 (1992) 1218-1226. [II] K. Andersson and B.O. Roos, Chem. Phys. Lett, 191 (1992) 507-514. [12] B.O. Roos, K. Andersson, and M.P. Fulscher, Chem. Phys. Lett., 192 (1992) 5-13. [13] K. Hirao (ed.), Recent Advances in Multireference Methods, World Scientific Publishing, Singapore, 1999. [14] P. Knowles, M. Schiitz, and H.-J. Werner, in Modern Methods and Algorithms of Quantum Chemistry, J. Grotendorst (ed.), John von Neumann Institute for Computing, NIC Series, Vol 1, Jiilich, 2000, pp. 69-151. [15] B.O. Roos, M.P. Fulscher, P.-A. Malmqvist, M. Merchan, and L. Serrano-Andres, in Quantum Mechanical Electronic Structure Calculations with Chemical Accuracy, S. R. Langhoff (ed.), Kluwer Academic Publishers, Dordrecht, 1995, pp. 357-438. [16] B.O. Roos, M.P. Fulscher, P.-A. Malmqvist, L. Serrano-Andres, K. Pierloot, and M. Merchan, in Advances in Chemical Physics: New Methods in Computational Quantum Mechanics, I. Prigogine and S. A. Rice (eds.), John Wiley and Sons, New York, 1996, pp- 219-331. [17] M. Merchan, L. Serrano-Andres, M.P. Fulscher, and B.O. Roos, in Recent Advances in Multireference Methods, K. Hirao (ed.), World Scientific Publishing, Singapore, 1999.
87 [18] L. Serrano-Andres and M. Merchan, in Encyclopedia of Computational Chemistry, P.v.R. Schleyer et al. (eds.), Wiley, Chichester, 2004. [19] J.B. Foresman, M. Head-Gordon, J.A. Pople, and M.J. Frisch, J. Phys. Chem., 96 (1992) 135149. [20] J. Oddershede, Propagator Methods, in Ab Initio Methods in Quantum Chemistry II, Advances in Chemical Physics, Vol. 69, K. P. Lawley (ed.), Wiley, Chichester, 1987, pp. 201-239. [21] H. Nakatsuji, Chem. Phys. Lett., 67 (1979) 334-342. [22] J.F. Stanton and R.J. Bartlett, J. Chem. Phys., 98 (1993) 7029-7039. [23] O. Christiansen, H. Koch, and P. Jorgensen, Chem. Phys. Lett., 243 (1995) 409-418. [24] F. Mas, M. Merchan, M. Pelissier, and J.P. Malrieu, Chem. Phys., 107 (1986) 361-380. [25] P.O. Lowdin, Annu. Rev. Phys. Chem., 11 (1960) 107-132. [26] C.CJ. Roothaan, Rev. Mod. Phys., 23 (1951) 69-89. [27] A.C. Hurley, Introduction to the Electron Theory of Small Molecules, Academic Press, New York, 1976. [28] B.O. Roos, in Multiconfigurational (MC) Self-Consistent (SCF) Theory: European Summerschool in Quantum Chemistry, Vol. 2, B.O. Roos and P.-O. Widmark (eds.), Lund University, 2000, pp. 285-359. [29] R. Gonzalez-Luque, M. Merchan, and B.O. Roos, Theor. Chim. Acta, 88 (1994) 425-435. [30] R. J. Gdanitz and R. Ahlrichs, Chem. Phys. Lett, 143 (1988) 413-420. [31] P.-A. Malmqvist, A. Rendell, and B.O. Roos, J. Phys. Chem., 94 (1990) 5477-5482. [32] R. Gonzalez-Luque, M. Merchan, M.P. Fulscher, and B.O. Roos, Chem. Phys. Lett., 204 (1993) 323-332. [33] H. Partridge, S.R. Langhoff, C.W. Bauschlicher Jr, in Quantum Mechanical Electronic Structure Calculations with Chemical Accuracy, S.R. Langhoff (ed.), Kluwer Academic Publishers, Dordrecht, 1995, pp. 209-260. [34] M. Cantos, M. Merchan, F. Tomas-Vert, and B.O. Roos, Chem. Phys. Lett., 229 (1994) 181190. [35] T. Helgaker, P. J0rgensen, and J. Olsen, Molecular Electronic-Structure Theory, Wiley, Chichester, 2000. [36] M. Head-Gordon, R.J. Rico, M. Oumi, and T.J. Lee, Chem. Phys. Lett., 219 (1994) 21-29. [37] R.E. Stratmann, G.E. Scuseria, and M.J. Frisch, J. Chem. Phys., 109 (1998) 8218-8224. [38] V. Molina, B.R. Smith, and M. Merchan, Chem. Phys. Lett., 309 (1999) 486-494. [39] L. Serrano-Andres, M. Merchan, and M. Jablonski, J. Chem. Phys., 119 (2003) 4294-4304. [40] A.C. Borin, L. Serrano-Andres, V. Ludwig, and S. Canute, Phys. Chem. Chem. Phys., 5 (2003) 5001-5009. [41] K.B. Wiberg, CM. Hadad, G.B. Ellison, and J.B. Foresman, J. Phys. Chem., 97 (1993) 1358613597. [42] L. Serrano-Andres, M. Merchan, I. Nebot-Gil, R. Lindh, and B.O. Roos, J. Chem. Phys., 98 (1993)3151-3162. [43] J. Lappe and R.J. Cave, J. Phys. Chem., 104 (2000) 2294-2300. [44] M. Merchan, L. Serrano-Andres, L.S. Slater, B.O. Roos, R. McDiarmid, and X. Xing, J. Phys. Chem. A, 103 (1999) 5468-5476. [45] J. Olsen and P. Jorgensen, J. Chem. Phys., 82 (1985) 3235-3264.
[46] P. Jorgensen and J. Simons, Second Quantization-Based Methods in Quantum Chemistry, Academic Press, New York, 1981. [47] E.S. Nielsen, P. Jorgensen, and J. Oddershede, J. Chem. Phys., 73 (1980) 6238-6246. [48] M.J. Packer, E.K. Dalskov, T. Enevoldsen, HJ. Aa. Jensen, and J. Oddershede, J. Chem. Phys., 105(1996)5886-5900. [49] A.B. Trofimov, G. Stelter, and J. Schirmer, J. Chem. Phys., 117 (2002) 6402-6410. [50] O. Christiansen, K.L. Bak, H. Koch, and S.P.A. Bauer, Chem. Phys. Lett., 284 (1998) 47-55. [51] P. Cronstrand, O. Christiansen, P. Norman, and H. Agren, Phys. Chem. Chem. Phys., 2 (2000) 5357-5363. [52] J.W. Krogh and J. Olsen, Chem. Phys. Lett., 344 (2001) 578-586. [53] M. Tobita, S.S. Perera, M. Musial, R.J. Bartlett, M. Nooijen, and J.S. Lee, J. Chem. Phys., 119 (2003)10713-10723. [54] K. Kowalski and P. Piecuch, J. Chem. Phys., 120 (2004) 1715-1738. [55] T.J. Lee and G.E. Scuseria, in Quantum Mechanical Electronic Structure Calculations with Chemical Accuracy, S.R. Langhoff (ed.), Kluwer Academic Publishers, Dordrecht, 1995, pp. 47-108. [56] T. Miyahara, H. Nakatsuji, J. Hasegawa, A. Osuka, N. Aratani, and A. Tsuda, J. Chem. Phys., 117(2002)11196-11207. [57] K. Hirao, J. Chem. Phys., 95 (1991) 3589-3595. [58] M. Musial, S.A. Kucharski, and R.J. Bartlett, J. Chem. Phys., 118 (2003) 1128-1136. [59] H. Koch and P. Jorgensen, J. Chem. Phys., 93 (1990) 3333-3344. [60] H. Koch, O. Christiansen, P. Jorgensen, and J. Olsen, Chem. Phys. Lett., 244 (1995) 75-82. [61] M. Parac and S. Grimme, J. Phys. Chem. A, 106 (2002) 6844-6850. [62] J.D. Watts and R.J. Bartlett, Chem. Phys. Lett., 258 (1996) 581-588. [63] K.W. Sattelmeyer, J.F. Stanton, J. Olsen, and J. Gauss, Chem. Phys. Lett., 347 (2001) 499-504. [64] S. Chattopadhyay, U.S. Mahapatra, and D. Mukherjee, in Recent Advances in Multireference Methods, K. Hirao (ed.), World Scientific Publishing, Singapore, 1999. [65] S.P.A. Sauer and M.J. Packer, The Ab Initio Calculation of Molecular Properties Other than the Potential Energy Surface, in Computational Molecular Spectroscopy, P. Jensen and P.R. Bunker (eds.), Wiley, Chichester, 2000, pp. 221-252. [66] S. Grimme and M. Waletzke, J. Chem. Phys., 111 (1999) 5645-5655. [67] T. Nakajima and H. Nakatsuji, Chem. Phys. Lett., 280 (1997) 79-84. [68] J.F. Stanton and J. Gauss, J. Chem. Phys., 100 (1994), 4695-4698. [69] A. Kohn and C. Hattig, J. Chem. Phys., 119 (2003) 5021-5036. [70] K. P. Lawley (ed.), Ab Initio Methods in Quantum Chemistry, Part II, Wiley, Chichester, 1987. [71] K. Andersson, M. Baryz, A. Bernhardsson, M.R.A. Blomberg, P. Boussard, D.L. Cooper, T. Fleig, M.P. Fulscher, B. Hess, G. Karlstrom, R. Lindh, P.-A. Malmqvist, P. Neogrady, J. Olsen, B.O. Roos, A.J. Sadlej, B. Schimmelpfennig, M. Schiltz, L. Seijo, L. Serrano-Andres, P.E.M. Siegbahn, J. Staking, T. Thorsteinsson, V. Veryazov, U. Wahlgren, P.-O. Widmark, MOLCAS, version 5.0; Department of Theoretical Chemistry, Chemical Centre, University of Lund, P.O.B. 124, S-221 00 Lund: Sweden, 2000. [72] B.O. Roos and P.R. Taylor, Chem. Phys., 48 (1980) 157-173. [73] A.D. McLean and B. Liu, J. Chem. Phys., 58 (1973) 1066-1078.
89 [74] B.O. Roos, K. Andersson, M.P. Fulscher, L. Serrano-Andres, K. Pierloot, M. Merchan, and V. Molina, J. Mol. Struct. (Theochem), 388 (1996) 257-276. [75] K. Hirao, Chem. Phys. Lett., 190 (1992) 374-380. [76] P.M. Kozlowski and E.R. Davidson, J. Chem. Phys., 100 (1994) 3672-3682. [77] J.P. Finley and K.F. Freed, J. Chem. Phys., 102 (1995) 1306-1333. [78] A. Zaitsevskii and J.P. Malrieu, Chem. Phys. Lett., 233 (1995) 597-604. [79] H. J. Werner, Mol. Phys., 89 (1996) 645-661. [80] S. Grimme and M. Waletzke, Phys. Chem. Chem. Phys., 2 (2000) 2075-2081. [81] B.O. Roos, A Perturbation-Variation Treatment of the Multi-State Problem in CASPT2, Lund University, 1996. [82] J. Finley, P.-A. Malmqvist, B.O. Roos, and L. Serrano-Andres, Chem. Phys. Lett., 288 (1998) 299-306. [83] M. Klessinger and J. Michl, Excited States and Photochemistry of Organic Molecules, VCH Publishers, New York, 1995. [84] P. Celani and H.-J. Werner, J. Chem. Phys., 119 (2003) 5044-5047. [85] M.P. Dallos, T. Muller, H. Lischka, and R. Shepard, J. Chem. Phys., 114 (2001) 746-757. [86] S.S. Han and D.S. Yarkony, J. Chem. Phys., 119 (2003) 11561-11569. [87] A.J.C. Varandas, On the Geometric Phase Effect in Jahn-Teller Systems, in Fundamental World of Quantum Chemistry, Vol II, E.J. Brandas and E.S. Kryachko (eds.), Kluwer, Netherlands, 2003, pp. 33-92. [88] P.R. Taylor, in Accurate Calculations and Calibration: European Summerschool in Quantum Chemistry, Vol. 3, B.O. Roos and P.-O. Widmark (eds.), Lund University, 2000, pp. 667-676. [89] T. Muller, M. Dallos, and H. Lischka, J. Chem. Phys., 110 (1999) 7176-7184. [90] A. Bernhardsson, N. Forsberg, P.-A. Malmqvist, B.O. Roos, and L. Serrano-Andres, J. Chem. Phys., 112 (2000) 2798-2809. [91] C. Page, M. Merchan, L. Serrano-Andres, and M. Olivucci, J. Phys. Chem. A, 103 (1999) 98649871. [92] L. Serrano-Andres, R. Lindh, B.O. Roos, and M. Merchan, J. Phys. Chem., 97 (1993) 93609368. [93] V. Molina, M. Merchan, and B.O. Roos, J. Phys. Chem. A, 101 (1997) 3478-3487. [94] J.R. Lakowicz, Principles of Fluorescence Spectroscopy, Kluwer, New York, 1999. [95] J.L. McHale, Molecular Spectroscopy, Prentice Hall, Upper Saddle River, 1999. [96] N.J. Turro, Modern Molecular Photochemistry, University Science Books, Sausalito, 1991. [97] M. Merchan, R. Gonzalez-Luque, and B.O. Roos, Theor. Chim. Acta, 94 (1996) 143-154. [98] R. Crespo, M. Merchan, and J. Michl, J. Phys. Chem. A, 104 (2000) 8593-8599. [99] M.C. Piqueras, M. Merchan, R. Crespo, and J. Michl, J. Phys. Chem. A, 106 (2002) 9868-9873. [100] B. Albinsson, H. Teramae, H.S. Plitt, L.M. Goss, H. Schmidbaur, and J. Michl, J. Phys. Chem., 100(1996)8681-8691. [101] M. Garavelli, F. Bernardi, A. Cembran, O. Castano, L.M. Frutos, M. Merchan, and M. Olivucci, J. Am. Chem. Soc, 124 (2002) 13770-13789. [102] R.P. Frueholz and A. Kuppermann, J. Chem. Phys., 69 (1978) 3614-3621. [103] P.G. Wenthold, D.P. Hrovath, W.T. Borden, and W.C. Lineberger, Science, 272 (1996) 14561459.
90 [104] B.O. Roos, M. Merchan, R. McDiarmid, and X. Xing, J. Am. Chem. Soc, 116 (1994) 59275936. [105] L.-M. Frutos, O. Castafio, and M. Merchan, J. Phys. Chem. A, 107 (2003) 5472-5478. [106] M.H. Palmer, J. Mol. Struct, 178 (1988) 79-87. [107] F.A. Van-Catledge, J. Am. Chem. Soc, 93 (1971) 4365-4374. [108] L.M. Frutos, O. Castano, J.L. Andres, M. Merchan, and A.U. Acufia, J. Chem. Phys., 120 (2004) 1208-1216. [109] T.M. Miller, A.A. Viggiano, and A.E.S. Miller, J. Phys. Chem. A, 106 (2002) 10200-10204. [110] L. Serrano-Andres, M. Merchan, M. Rubio, and B.O. Roos, Chem. Phys. Lett., 295 (1998) 195203. [111] H. Nakatsuji, J.Y. Hasegawa, and M. Hada, J. Chem. Phys., 104 (1996) 2321-2329. [112] M. Nooijen and R.J. Bartlett, J. Chem. Phys., 106 (1997) 6449-6455. [113] D.J. Tozer, R.D. Amos, N.C. Handy, B.O. Roos, and L. Serrano-Andres, Mol. Phys., 97 (1999) 859-868. [114] M. Parac and S. Grimme, Chem. Phys. 292 (2003) 11-21. [115] S. Grimme and M. Parac, ChemPhysChem, 3 (2003) 292-295. [116] Z.-L. Cai, K. Sendt, and J.R. Reimers, J. Chem. Phys., 117 (2002) 5543-5549. [117] M. Wanko, M. Garavelli, F. Bernardi, T.A. Niehaus, T. Frauenheim, and M. Elstner, J. Chem. Phys., 120 (2004) 1674-1692. [118] K. Pierloot, Mol. Phys., 13 (2003) 2083-2094. [119] K. Pierloot, A. Ceulemans, M. Merchan, and L. Serrano-Andres, J. Phys. Chem. A, 104 (2000) 4374-4382. [120] U. Ryde, M.H.M. Olsson, B.O. Roos, K. Pierloot, J.O.A. De Kerpel, Protein, Blue Copper: Electronic Spectra, in Encyclopedia of Computational Chemistry, P. v. R. Schleyer et al. (eds.), Wiley, Chichester, 1998, pp. 2255-2270. [121] B.O. Roos, P.-O. Widmark, and L. Gagliardi, Faraday Discuss., 124 (2003) 57-62. [122] J. Lopez-Cacheiro, B. Fernandez, T.B. Pedersen, and H. Koch, J. Chem. Phys., 118 (2003) 9596-9607. [123] L. Serrano-Andres, M.P. Fulscher, and G. Karlstrom, Int. J. Quantum Chem., 65 (1997) 167181. [124] N. Ferre and M. Olivucci, J. Am. Chem. Soc, 125 (2003) 6868-6869. [125] L. Serrano-Andres and B.O. Roos, J. Am. Chem. Soc, 118 (1996) 185-195. [126] B. O. Roos, P.-A. Malmqvist, V. Molina, L. Serrano-Andres, and M. Merchan, J. Chem. Phys., 116(2002)7526-7537. [127] R. Pou-Amerigo, L. Serrano-Andres, M. Merchan, E. Orti, and N. Forsberg, J. Am. Chem. Soc, 122 (2000) 6067-6077. [128] L. Serrano-Andres, N. Forsberg, and P.-A. Malmqvist, J. Chem. Phys., 108 (1998) 7202-7216. [129] G. Fischer, Vibronic Coupling, Academic Press, London, 1984. [130] H. Koppel and W. Domcke, Vibronic Dynamics in Polyatomic Molecules, in Encyclopedia of Computational Chemistry, P.v.R. Schleyer et al. (eds.), Wiley, Chichester, 1998, pp. 3166-3182. [131] G. Fischer and P. Wormell, Chem. Phys., 257 (2000) 1-20. [132] G. Fischer, Z.L. Cai, J.R. Reimers, and P. Wormell, J. Phys. Chem. A, 107 (2003) 3093-3106. [133] P.-A. Malmqvist, B.O. Roos, and B. Schimmelpfennig, Chem. Phys. Lett., 357 (2002) 230-240.
91 [134] M. Merchan, B.O. Roos, R. McDiarmid, and X. Xing, J. Chem. Phys., 104 (1996) 1791-1804. [135] J.P. Dinnocenzo, M. Merchan, B.O. Roos, S. Shaik, and H. Zuilhof, J. Phys. Chem. A, 102 (1998)8979-8987. [136] L. Serrano-Andres and M.P. Fulscher, J. Phys. Chem. B, 105 (2001) 9323-9330. [137] V. Molina and M. Merchan, Proc. Nat. Acad. Sci. USA, 98 (2001) 4299-4304. [138] M.P. Fulscher and B.O. Roos, J. Am. Chem. Soc, 117 (1995) 2089-2095. [139] N. Ismail, L. Blancafort, M. Olivucci, B. Kohler, and M.A. Robb, J. Am. Chem. Soc, 124 (2002)6818-6819. [140] M. Merchan and L. Serrano-Andres, J. Am. Chem. Soc, 125 (2003) 8108-8109. [141] M. A. Robb, M. Olivucci, and F. Bernardi, Photochemistry, in Encyclopedia of Computational Chemistry, P.v.R. Schleyer et al. (eds.), Wiley, Chichester, 1998, pp. 2056-2064. [142] L. Serrano-Andres, M. Merchan, R. Lindh, and B.O. Roos, J. Am. Chem. Soc, 117 (1995) 3189-3202. [143] B. Proppe, M. Merchan, and L. Serrano-Andres, J. Phys. Chem. A, 104 (2000) 1608-1616. [144] G.J. Atchity, S.S. Xantheas, and K. Ruedenberg, J. Chem. Phys., 95 (1991) 1862-1876. [145] R. Gonzalez-Luque, M. Garavelli, F. Bernardi, M. Merchan, M.A. Robb, and M. Olivucci, Proc. Nat. Acad. Sci. USA, 97 (2000) 9379-9384. [146] J.-M. Pecourt, J. Peon, and B. Kohler, J. Am. Chem. Soc, 123 (2001) 10370-10378. [147] J. Peon and A.H. Zewail, Chem. Phys. Lett, 348 (2001) 255-262. [148] H. Kang, K.T. Lee, B. Jung, Y.J. Ko, and S.K. Kim, J. Am. Chem. Soc, 124 (2002) 1295812959. [149] D.R. Yarkony, J. Phys. Chem. A, 105 (2001) 6277-6293. [150] C.S. Page and M. Olivucci, J. Comput. Chem., 24 (2003) 298-309. [151] L. Serrano-Andres, M. Merchan, and R. Lindh, to be published. [152] L. De Vico, C.S. Page, M. Garavelli, F. Bernardi, R. Basosi, and M. Olivucci, J. Am. Chem. Soc, 124 (2002) 4124-4134.
This Page is Intentionally Left Blank
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 Elsevier B .V. All rights reserved
93
III. Density Functional Methods for Excited States: Equilibrium Structure and Electronic Spectra Filipp Furche* and Dmitrij Rappoport Institut fiir Physikalische Chemie, Universitat Karlsruhe, Kaiserstrafie 12. 76128 Karlsruhe, Germany 1. INTRODUCTION Density functional theory (DFT) is nowadays one of the most popular methods for ground state electronic structure calculations in quantum chemistry and solid state physics. Compared to traditional ab initio and semi-empirical approaches, contemporary density functional methods show a favorable balance between accuracy and computational efficiency. A number of commercial programs is available, and DFT calculations of ground state energies, structures, and many other properties are routinely performed by nonexperts in (bio-)chemistry, physics, and materials sciences. Hohenberg-Kohn density functional theory is strictly limited to ground states [1], which excludes applications to photochemistry. This is a serious drawback, because photoexcited molecules are experimentally much more difficult to characterize than molecules in their ground states. Reliable theoretical predictions for excited states are thus especially valuable. Several routes have been followed to extend conventional DFT to excited states (see, e.g., Refs. [2-5]). In the present review, we focus on time-dependent density functional theory (TDDFT), which is presently the most popular method to treat excited states in a DFT framework. Extensive reviews on TDDFT exist [6-10]; most of them emphasize formal aspects of the theory. The aim of the present work is to survey the use of TDDFT in photochemistry. It is primarily written for non-experts with little background in DFT. The literature in this field is growing rapidly, and we cannot claim to be exhaustive; instead, we give a selective introduction to important concepts and recent developments from a rather personal perspective. Sec. 2 contains a brief introduction to the theory. We do not give any derivations and merely state the most important results and explain their meaning. An overview of popular density functionals is given in Sec. 2.3. Algorithms to compute spectra and excited state properties are reviewed in Sec. 3. We mostly describe the steps necessary in a TDDFT excited state calculation and give details only where necessary. Some timings for typical applications are presented in Sec. 3.4. Sec. 4 summarizes the performance of TDDFT excitation energies, transition moments, and excited state properties * Electronic address: [email protected].
94 in benchmark studies. This section is recommended to the reader interested in the accuracy of TDDFT in general. Situations where present functionals fail are discussed as well. Specific applications are surveyed in Sec. 5. Classes of compounds include aromatic systems and fullerenes, porphyrins and related compounds, transition metal compounds, metal and semiconductor clusters, organic polymers, and biologically relevant systems. We close with an outlook in Sec. 6.
2. THEORETICAL FOUNDATIONS 2.1. Time-dependent response theory approach to excited states Excited states are solutions of the time-independent stationary Schrodinger equation; time-dependent response theory is used as a trick to reduce electronic excitations to ground state properties. Consider a molecule in its electronic ground state subject to a periodic perturbation by a uniform electric field E oscillating at frequency ui. The distribution of the electronic charge and current density of the molecule is described by the one-particle density matrix 7(i). j(t) will perform driven oscillations about its ground state value 7 ^ . The amplitude of these oscillations is given by the Fourier transform of 7(t), denoted j(u>) for simplicity. As a result of elementary perturbation theory. 7(0;) has the following expansion in powers of the field E: 7H
=7(0) - £ (-^»- - ^f)
E + O(E?)
(1)
If the frequency ui approaches an excitation energy O()n of the system, there is a resonance catastrophe and the amplitude of the oscillation diverges. Keeping the analogy to a system of harmonic oscillators [11,12], the excitation energies Jlo,,. are the eigenfrequencies of the electrons in the molecule, and the transition density matrices -fOn are the corresponding collective modes. After inversion of the relation between 7(0;) and E, the excitation energies are obtained as eigenvalues of an electronic Hessian which may be imagined as the matrix of second derivatives of the electron energy with respect to the electronic degrees of freedom. In this way, any ground-state theory can be extended to excited states, provided the time-dependent response is well-defined. The thus obtained excitation energies and transition moments are in turn consistent with ground-state properties because sum over states (SOS) expressions as in Eq. (1) hold. Both is generally not true for state-based methods. On the other hand, the reliability of response theory based methods crucially depends on the stability of the ground state (see below). The formal basis for an extension of common ground-state density functional methods to time-dependent perturbations is TDDFT. Within the time-dependent Kohn-Sham (TDKS) framework [13], one considers a system of N non-interacting fermions whose density is constrained to the physical density p(t,x). This leads to the time-dependent Kohn-Sham equations i^:j(t,x)=H[p](t,x)cbJ(t,x).
(2)
The effective TDKS one-particle Hamiltonian H[p\(t,x) = n2(t, x)/2 + vs[p\(t, x) consists of a kinetic energy part and a local time-dependent external potential vs [p]. The latter is
95 a unique functional of p(t, x) (up to a gauge transformation) for a given initial state, as stated by Runge and Gross [13]. vs is usually decomposed according to vs[p}(t,x)=vext(t,x)+vc[p]{t,x)+vxc[p}{t,x)
(3)
into the external one-particle potential vtxt. the time-dependent Coulomb (or Hartree) potential vc[p](t,x) = J dx'p(t,x')/\r — r'|. and the time-dependent exchange-correlation potential vxt.[p](t,x). Equivalently, one may consider the TDKS one-particle density matrix 7(£), which is related to the TDKS orbitals via the spectral representation
Its time evolution is governed by the von-Neumann equation (5)
subject to the idempotency constraint f
>y(t,x,x') =
dx1-y(t,x,x1)>y(t,x1,x').
(6)
This density matrix based approach [12,14-16] is particularly convenient for response theory, because the equations determining the first and higher order response of 7 can be derived by straightforward differentiation of Eqs. (5) and (6) with respect to an external perturbation [15]. Complicated intermediates such as perturbed orbitals or response functions are avoided. Equations (5) and (6) describe a non-interacting system and are therefore computationally manageable, while the solution of the full interacting TV-electron problem is exponentially more complex. This is, somewhat simplified, the main cause of the computational advantage of density functional methods over conventional wave-function methods. The price for this improved efficiency is that the potential vxc[p](t, x) has to be approximated. The construction of accurate and inexpensive approximations to vxc [p] is a central problem of TDDFT and will be discussed in Sec. 2.3. Formally, the TDKS construction implies [6] that 7(4) yields the interacting density p and the interacting current density j according to p(t,x) = j(t,x,x) (7) This means that the frequency-dependent TDKS density matrix response must have an SOS expansion of the type (1). Therefore, the physical excitation energies are accessible from the TDKS response, and the corresponding eigenmodes yield physical transition moments.
96 2.2. Excited state properties
2.2.1. The Lagrangian of the excitation energy The time-dependent response theory approach outlined in the last section provides a route to excitation energies and transition moments. Excited state total energies are accessible by adding the corresponding ground state energy to the excitation energy. But how to compute other excited state properties such as dipole moments without an excited state wavefunction? - First it is important to remember that the wavefunction is only an intermediate that relates properties of a system such as energies or densities to a Hamiltonian, i.e., external potentials. Properties of a stationary state may be defined without reference to the wavefunction by the dependence of the energy on an applied external perturbation. For example, the dipole moment may be denned as the first derivative of the energy with respect to a constant electric field at zero field strength. More generally, the excited state density can be defined as the functional derivative of the excited state energy with respect to an external perturbing potential at zero coupling. It is therefore sufficient to know the dependence of the excited state energy on the external potential to compute static excited state properties. The energy of a stationary state is stable with respect to the wavefunction: this leads to the Hellmann-Feynman theorem for first-order properties and to the more general Wigner 2n + 1 rule. The latter states that the wavefunction through n-th order determines all properties through order 2n + 1. The Lagrangian method establishes an analogous variational principle for excited states in TDDFT. Here we present a summary only; for a detailed derivation, the reader is referred to Ref. [17]. The Lagrangian of the excitation energy is defined by L[X,Y,Q,C,Z,W] = (X,Y\A\X,Y) ~n({X,Y\A\X,Y) - 1)
F is the ground state Fock matrix, and S denotes the overlap matrix. L depends on the ground state Kohn-Sham (KS) molecular orbital (MO) coefficients C; the latter are related to the ground state KS MOs <j)pa via the LCAO (linear combination of atomic orbitals) jKr(r) = YlC^Xu(r),
(9)
where \v are atom-centered basis functions. Indices i,j,... are used for occupied, a, 6,... for virtual, and p,q.... for general MOs. We assume that the MOs are real and eigenfunctions of the z component of the total spin. X and Y parameterize the transition density matrix 70,. of the n-th excited state, ^ a f f (r)^ f f (r') + ^ ^ ( r ^ r ' ) ) ;
(10)
we shall always refer to the n-th state and omit state labels where possible. X and Y are conveniently gathered in the two component "transition vector" y)=\X,Y).
(11)
97 Q, Z. and W are Lagrange multipliers enforcing additional constraints, as discussed below. If L becomes stationary, the additional "penalty" terms introduced by fi, Z. and W vanish by construction. One is thus left with the term (X. Y\A\X, Y) representing the excitation energy. It may be considered an expectation value of the orbital rotation Hessian A evaluated for the transition vector \X,Y). A and A are 2x2 "super-operators".
where A and B have the matrix representation (.4 + B)lar,:jbrj,
= (eaa - ela)5i:j5ab5arj> + 2(iaa\jba')
+ 2f™a]ba>
- cxSa(7'[(jaa\ibcj) + (aba\ija)} (A-B)itmjba>
= (eaa - el(J)5i:j5ab5arji + cxSaaf[(jaa\iba)
(13a) - (aba\ija)}.
(13b)
(pqa\rsa') is a two-electron repulsion integral in Mulliken notation, and f^aTsa> represents a matrix element of the exchange-correlation kernel in the adiabatic approximation (AA),
where i?xc is the static exchange-correlation energy functional. The hybrid mixing parameter cx [18,19] is used to interpolate between the limits of "pure" density functionals (cx = 0) and time-dependent Hartree-Fock (TDHF) theory (cx = l , £ x c = 0). 2.2.2. Stationarity conditions for L The following stationarity conditions determine the excited state energy and first order properties. 1. The ground-state KS equations (in unitary invariant form), - ^ - = Flaa = 0,
(15)
implying that the occupied-virtual block of the ground-state Fock operator F is zero. The Lagrange multiplier W enforces orthonormality of the KS MOs, J T T ^ - = Spqa - 5pq = 0.
(16)
2. The TDKS eigenvalue problem (EVP) - ^ -
= (A-nA)|X,y>=0,
(17)
together with the non-standard normalization condition for the transition vectors (X.Y\A\X.Y)-l
= 0.,
(18)
which is enforced by f2. The form of Eqs. (17) and (18) is familiar from Hartree-Fock (HF) theory [20]. This analogy was first recognized by Zangwil and Soven [21] and later
98 generalized by Casida [14]. Other schemes, including density based methods [22] and Dyson-type procedures [23] are special cases of the density matrix based formalism. The eigenvalues Q of A are electronic excitation energies, and the corresponding transition vectors \X, Y) are collective eigenmodes of the TDKS density matrix. Vt and \X, Y) are the solutions of the TDKS EVP (17). The normalization condition (18) can be used to assign a state in terms of excitations from occupied to virtual KS MOs. The weight of a one-particle excitation from the occupied orbital i to the virtual orbital a is
The configuration mixing reflects the change in the Coulomb and exchange-correlation potentials upon excitation. More elaborate methods to analyze transition vectors use transition natural orbitals [24] or attachment and detachment densities [25]. Denoting the electronic dipole moment operator by /it, the oscillator strength for the transition n <— 0 is given by
/ 0 n = ^n n |(/x|x n ,y n >| 2 .
(20)
Similarly, the rotatory strength is ROn = Im((/x|X n , Yn) • (Xn, y n | m » ,
(21)
where m denotes the magnetic dipole moment operator, /j, can be expressed in various forms, e.g., the dipole-length or the dipole-velocity form [26] which are related by a gauge transformation. Since the TDKS formalism is gauge invariant, the different forms of /J, lead to the same result in the basis set limit [15]. As expected for a response theory based approach, the oscillator strength and the rotatory strength satisfy sum rules. For example, the isotropic polarizability of the the ground state at frequency LO has the SOS expansion
This is true independent of the basis set and functional. 3. The aZ vector" equation and the determining equations for W. They follow from the stationarity condition ^ = 0 .
(23)
The Z vector equation is a static perturbed KS equation of the form (A + B)Z = -R.
(24)
The expressions for R and W involve third order functional derivatives and are explicity given in Ref. [17]. The difference between the excited and ground state density matrices is given by P = T + Z,
(25)
99 where the "unrelaxed" part T contains products of the excitation vectors only. Z accounts for relaxation of the ground state orbitals; it can be of the same order of magnitude as T. The information contained in P is complementary to the information contained in the transition vector. The latter is related to matrix elements between the ground and excited state, while P is related to the difference of expectation values for the excited and the ground state. For example, tr(Pfi) is the change of the dipole moment upon excitation from the ground state; by adding the ground state density matrix to P, excited state properties can be computed in this way. Population analysis or graphical representation of P can give insight in the re-distribution of the electronic charge due to the excitation process. The remaining Lagrange multiplier W accounts for first-order changes in the energy due to changes in the overlap matrix. W is therefore an "energy weighted difference density matrix", and is needed for gradient calculations only. The total gradient of L with respect to a perturbation £ has the form [17]
fiver
/iva
E
fifnXaa'
+ £ ^ ( O i V + E 01xAX + Y)lwa(X + Y)KXar,
(26)
h is the sum the kinetic and potential energy one-particle operators and Vxc is the static exchange-correlation potential. F is an effective two-particle density matrix that separates into two-index quantities. £ may represent, e.g., a component of an external electric field, in which case all terms except the first are zero: or it may represent a nuclear coordinate. Parentheses indicate that derivatives need to be taken only with respect to basis functions; MO coefficient derivatives do not occur as a consequence of the 2n+ 1 rule. L^ has nearly the same form as the ground state energy gradient [27], the definitions of P, F, and W being the main difference. Total excited state properties are obtained by simply adding the ground state contributions. 2.3. Approximate exchange-correlation functionals There are different approaches to the construction of approximate functionals. Empirical functionals contain a large number of parameters fitted to a "training set" of accurate experimental or calculated data. Non-empirical functionals contain few or no fitted parameters and are designed to satisfy known constraints. Empirical functionals should be accurate for systems and properties contained in the training set, but they can fail for other systems. In contrast, non-empirical functional usually exhibit a more uniform accuracy [28], The accuracy of approximate exchange-correlation functionals is limited by their form, i.e., there is a certain maximum accuracy that can be expected for local, semi-local, etc. functionals. The "perfect agreement" with experiment reported in some density functional studies should therefore rather give rise to concern, especially if highly parameterized or exotic functionals are used. The most common and universally used approximation in TDDFT is the above-mentioned A A [29]. It replaces the time-dependent exchange-correlation potential by its static counterpart, evaluated at the time-dependent density. The resulting potential is instantaneous, in contrast to the exact one, which has a "memory" of all times t' < t. In response
100 theory, the AA makes the exchange-correlation kernel and all higher derivatives of the exchange-correlation potential independent of the frequency. The AA has been considered uncritical for a long time. Only recently it has been clarified that the lack of higher excited states in TDDFT excitation spectra is a consequence of the AA [30]. This may be related to the failure of the AA in dissociating H2, where doubly excited states are important [31]. 2.3.1. Local and semi-local functionals Semi-local functionals have the form Exc = /d A r /(p a (r),p0(r), Vp Q (r), V M r ) , . . . ) .
(27)
In the local spin density approximation (LSDA), / depends on the spin densities at r only. The LSDA is derived from the exchange-correlation energy per particle of a uniform electron gas. which has been accurately parameterized [32,33]. For functionals of the generalized gradient approximation (GGA), / also depends on the gradient of the spin densities. Popular GGA functionals with few empirical parameters are Becke's 1988 exchange functional [34] together with the correlation functional of Lee, Yang, and Parr (BLYP) [35], or Perdew's 1986 correlation functional (BP86) [36]. The GGA of Pewdew, Burke, and Ernzerhof [37] (PBE) is parameter free, while Hamprecht, Cohen, Tozer, and Handy (HCTH) have proposed an empirical GGA functional [38]. In meta-GGA functionals, / depends on additional local information such as the kinetic energy density or the Laplacian of the density. Examples are the 21 parameter meta-GGA of Van Voorhis and Scuseria (VS98) [39], or the non-empirical meta-GGA of Tao, Perdew, Staroverov, and Scuseria (TPSS) [28]. 2.3.2. Hybrid functionals Hybrid functionals interpolate between HF theory and semi-local functionals [18,19]; the fraction of HF exchange is controlled by the exchange mixing parameter cx. The exchange is treated as in HF theory, using non-local potentials. This interpolation leads to an error compensation for many properties. Popular hybrid functionals are, e.g., B3LYP [40], B3PW91 [19], or PBEO [41]. 2.3.3. Optimized effective potential (OEP) based functionals Exact exchange (EXX) as a functional of the KS density matrix has the same form as HF exchange. Differences arise in the variation of the energy. In HF theory, the energy is minimized with respect to the density matrix. The resulting exchange potential is the well-known non-local exchange operator in HF theory, while it is a local multiplicative potential in KS theory. For a fixed density, this potential can be determined by an energy optimization procedure, as first shown for atoms by Talman and Shadwick [42]. Computation of the local exchange potential in molecules is a non-trivial problem [43], but there has been recent progress in developing more efficient methods [44,45] and approximations [46-48]. Full OEP calculations of the frequency-dependent exchange kernel have been reported for solids, but not for molecules so far [49]; see Refs. [50, 51] for a review. In most TDDFT applications, KS orbitals and orbital energies from an OEP calculation are combined with adiabatic LSDA or GGA exchange-correlation kernels.
101
2.3.4- Asymptotic corrections The exchange-correlation potentials of semi-local functionals decay too fast in the asymptotic region outside a molecule. In most cases, the decay is exponential, instead of the correct — 1/r. As a result, diffuse excited states are often predicted too low in energy, and higher Rydberg excitations may be absent from the bound spectrum [52], Various correction schemes have been suggest to remedy this problem [53-55]. These corrected potentials are not the derivative of any exchange-correlation energy functional, however. This does not affect vertical excitation energies, but makes a consistent definition of excited state total energies and properties difficult. 2.3.5. Current-dependent functionals Some deficiencies of semi-local functionals can be cured by using the current density j instead of the density. Vignale and Kohn have shown that the time-dependent exchangecorrelation vector potential of weakly inhomogeneous systems possesses a gradient expansion as a functional of j but not of p [56,57]. Current dependent functionals capture macroscopic polarization effects in solids which are ultra-non-local in the density [58]. First applications to molecular excitation energies [59] show a somewhat mixed picture, however. 3. COMPUTATIONAL STRATEGIES 3.1. Basis set methods As explained in the last section, performing a T D D F T excited state calculation amounts to finding the stationary points of the Lagrangian L. Introduction of a finite basis set (usually atom-centered) generates a finite number of MOs through the LCAO expansion (9). If the basis set is suitably chosen, the excited state energy may be well approximated by optimizing L on the corresponding subspace. We thus arrive at a finite-dimensional optimization problem which can be solved by matrix algebra. The basis set incompleteness can be checked by using hierarchical basis sets of different size, compare Sec. 3.3. The steps necessary to compute excited state energy and gradients parallel the stationarity conditions for L discussed in Sec. 2.2.2. A summary is given in Table 1, including the scaling of the computational cost with the system size measured by N.
Table 1 Steps in an excited state energy and gradient calculation, formal and asymptotic scaling of computational cost. Scaling Formal Asymptotic Ground state energy and wavefunction TV4 N2 4 Excitation energy iV A"2 4 Relaxed density a n d gradient A" A2
T h e first step, solution of t h e g r o u n d - s t a t e KS equations in a finite basis set, is a s t a n d a r d procedure in q u a n t u m chemistry a n d needs no further discussion here. In t h e
102 second step, (approximate) excitation energies and transition vectors are calculated by solving the finite-dimensional TDKS EVP. Complete diagonalization of the electronic Hessian A scales as TV6 and is prohibitive for systems with more than 10 heavy atoms. In most applications, however, especially in larger systems, only the lowest excited states are of interest. By iterative methods, the lowest part of the spectrum of A can be calculated much more efficiently than by complete diagonalization. Iterative methods minimize L by expanding the excitation vector on a subspace whose dimension is small compared to the full problem. One usually starts from unit vectors, i.e., the KS one-particle excitations. In each iteration, the best approximation to the excitation energy is calculated by a small diagonalization on the current subspace (Ritz step). The error is controlled by the norm of the residual which corresponds to the gradient of L. If the error is small enough, the process terminates; otherwise, the subspace is extended in the direction of the (preconditioned) gradient and a new iteration starts. Similar ideas can be found in the early work of Lanczos [60] and Hestenes and Stiefel [61] already, but it was only the preconditioning introduced by Davidson [62, 63] that made these iterative algorithms useful for quantum chemistry. The extension to the special EVPs occurring in response theory goes back to Olsen, Jensen, and J0rgensen [64]; in the meantime, several modifications have been suggested [65-68]. If "pure" functionals are used, it is favorable to transform the TDKS EVP to a symmetric problem of half the original dimension [69]; the latter is amenable to standard algorithms for symmetric-positive EVPs. The time-determining step in all iterative methods is the computation of matrix-vectorproducts \U, V) = A\X, Y). where \X, Y) is a subspace basis vector. This is most efficiently performed as (U + V) = {A + B)(X + Y),
(28a)
(U-V)
(28b)
= (A-B)(X -Y),
because the symmetry of (A ± B) (as a super-operator) and of (X ± Y) can be fully exploited. The diagonal contribution to (A ± B) resulting from the orbital energy differences, cf. Eqs. (13), is trivial to compute. The multiplication by the the remaining four-index integrals is best performed by transforming the vectors to the AO basis, in the spirit of direct CI methods [70] in an AO formulation [71, 72]. Denoting the transformed vectors by Greek indices, we have
(X ± Y)lwa = l- J^iX ± y\u,,{ClllaCvan
± CtumCmrT).
(29)
ia
With respect to the AO indices, (X + Y) is a symmetric and (X — Y) a skew-symmetric square matrix. After that, one computes
(U + V)llva = ] KXU'
(30a) (U - V % - = ^ c x < W [ ( H ^ ) - (II\\VK)](X
- YW-
(30b)
103
Back-transformation finally yields the product vectors in the MO basis,
(U ± V)iaa -+l- J2(U ± V)IW(J(ClaiJC,J(m ± CtumCvin).
(31)
The part resulting from the two-electron integrals is fully equivalent to a ground-state Fock matrix construction for a complex density matrix [65]. This means that highly efficient direct SCF techniques available for ground states can be carried over to excited state calculations with minimal modifications. Thus, in each iteration, only O(N2) non-zero two-electron integrals (/iz/|«;A) are calculated "on the fly", i.e., they are completely or partly discarded after use and not stored. In contrast, an integral transformation would lead to an O(7V5) scaling of CPU-time and O(iV4) I/O, because (A ± B) is generally not sparse in the MO basis. The analogy to ground-state calculations also holds for the contribution arising from the exchange-correlation kernel. The four-index quantities f*'vaKxa> are never actually calculated; instead, the contributions to (U + V) are formed directly on the quadrature grid and integrated, which is virtually equivalent to setting up the matrix of the ground-state exchange-correlation potential [17, 69]. For semi-local functionals, prescreening leads to a scaling of O(N) for the exchange-correlation contribution to the matrix-vector-products (U + V). The vector transformation steps (29) and (31) have a formal O(N'i) scaling: however, if efficient linear algebra subroutines are used, the cost is negligible for systems with up to ca. 10000 basis functions. For simulating electronic excitation spectra of larger systems, block iteration methods lead to dramatic further savings of computation time [73, 74]. In these methods, a number of states is treated simultaneously. This means that the two-electron integrals need to be calculated only once for all vectors of a block. In addition, block methods often show favorable convergence compared to single-vector methods. Molecular point group symmetry can be exploited in the MO basis by Clebsch-Gordan reduction of MO products and in the AO basis by skeleton operator techniques [65, 74, 75]. This leads to an overall reduction of computational cost by approximately the order of the point group. Advantage can be taken of spin symmetry as well. For closed-shell singlet ground states, the TDKS EVP decomposes into two separate EVPs for singlet and triplet excitations. A restricted open shell scheme for high spin ground states has been proposed recently [76]. If first-order excited state properties are to be calculated, the Z vector equation (24) needs to be solved in the third step. This is best done iteratively again, using the techniques outlined above. Once the relaxed density matrices P and W have been obtained, excited state properties can be evaluated in almost the same manner as ground state properties. It is important that the thus obtained relaxed density matrices do not depend on the perturbation. The cost for computing analytical gradients of the excited state energy is therefore independent of the number of nuclear degrees of freedom. In contrast, numerical differentiation leads to a cost that increases linearly with the number of nuclei. The cost for computing excited state energies and first-order properties differs from the cost for the corresponding ground-state calculations by a constant factor only. In conclusion, excited state geometry optimizations within the TDDFT framework are hence not significantly more expensive than conventional DFT ground state optimizations [17]. The prerequisite is, however, that the 2n + 1 rule is used and full advantage is taken of
104
the similarity to efficient ground state algorithms. 3.2. Approximations and extensions 3.2.1. Efficient treatment of the Coulomb energy As explained in Sec. 3.1, computation of the two-electron integrals (JJLI>\K\) is the bottleneck in larger TDDFT response calculations. For non-hybrid functionals (cx = 0), these integrals contribute to the Coulomb part of the excitation energy only.
Ec\pan] = 2 Yl (X + ^W(H«*)(* + y W 2}
|r-r'|
The last expression is identical to the ground state Coulomb energy functional, evaluated at the spin-averaged transition density pOn(r) = \ ^ 7(w(r. r) = T ^ ( ^ + Y) ma\ ,(r)xi,(r)
(33)
\1V<7
(7
RI-J techniques for a fast evaluation of the ground state Coulomb energy [77, 78] can thus be carried over to excited state calculations in a straightforward manner. The key idea of RI-J approximation is to introduce an auxiliary expansion of the density in a set of one-center functions Xp(r) (usually atom-centered Gaussians) [79-83], A,n(r) = ^ c p X p ( r ) . v
(34)
The expansion coefficients cp are determined by minimizing the error in the Coulomb norm \\pon - P0n\\c =2j
\r^V\
^
This leads to the Coulomb energy in the RI-J approximation, Ec[pon] = — 2^
/-^(X + Y)llvr7(p,v\p){p\q)~l(q\KX){X + Y)K\ai.
(36)
In this expression only three- and two-center electron repulsion integrals occur, with products of basis functions replaced by auxiliary functions (denoted by labels p, q). In the auxiliary basis set limit, Eq. (36) is formally obtained from Eq. (32) by inserting the identity. The choice of the Coulomb metric implies that the error Ec[pan\ ~ Ec[pon] is positive and quadratic in the error in the density. This variational stability ensures that good accuracy can be achieved with relatively small auxiliary basis sets. The RI-J approximation has several computational advantages. First, the calculation of four-center integrals ([J,I/\K\) that formally scales as TV4 is replaced with two TV3 steps. By means of integral pre-screening, the scaling can be further reduced to TV2, as in the conventional case; the pre-factor is much lower, however. Secondly, a large amount of
105
integrals (fj,i/\p) can be pre-computed and stored in memory. The inverse (p\q) 1 is never actually calculated; instead, linear equation systems are solved using the Cholesky decomposition of (p\q) [84], This very fast O(N:i) step is performed once before the iteration starts and has almost no effect on total computation times for systems that are currently feasible. Speedups of a factor of 10 and more are achieved for the Coulomb contribution to the excitation energy compared to conventional methods [85]. This leads to a significant reduction of total timings for large systems, where the exchange-correlation part becomes less important due to its favorable O(N) scaling. Auxiliary basis sets developed for ground state calculations [77, 86-90] are sufficient for most TDDFT applications, although in some cases additional diffuse basis functions must be included. For excitation energies RI-J errors of less than 0.005 eV are observed for valence excitations, whereas for Rydberg excitations somewhat greater deviations up to 0.05 eV are found. These deviations are usually much smaller than both errors due to the incompleteness of the one-particle basis set and due to the use of approximate functionals, compare Sees. 3.3 and 4). TDDFT implementations using fitting basis sets are available, e.g., in the deMon [91], TURBOMOLE [85], and PARAGAUSS [22,92] program packages which make use of Gaussian auxiliary basis functions. Basis sets of Slater-type (STO) are employed in the ADF [67,93,94] program. Some of these implementations use other norms for the auxiliary expansion, e.g., the overlap norm, or norms based on the exchange-correlation kernel / x c or the full TDDFT response kernel instead of the Coulomb interaction [22]. A similar resolution of the identity approach has been developed for non-local HF exchange [87]. This Hl-JK approach can be useful for TDDFT calculations with hybrid functionals [95], but is more demanding and requires larger auxiliary basis sets than the RI-J method. Typical speedups are in the range of 2 — 4 compared to the full calculation of four-center integrals. Analytical gradient calculations for excited states can take advantage of the RI-J approximation as well [96]. RI-J may be used in the determination of excitation energies and transition vectors {X ± Y) and in the iterative solution of the Z vector equation (24). The calculation of excited state gradients can be carried out along the same lines as for ground state gradients. The total computational effort for excited state optimizations is reduced by at least a factor of 4-6 by the RI-J approximation. This allows to perform excited state optimizations on medium-size and large molecules with more than 100 atoms. RI-J errors in optimized bond lengths and angles amount to less then 0.5 pm and 1 degree, respectively. For adiabatic excitation energies, RI-J errors of 0.01-0.02 eV are found. 3.2.2. The Tamm-Dancoff-approximation (TDA) The TDA amounts to constraining Y = 0 in the variation of L. As a result, the TDKS EVP reduces to the symmetric-positive EVP AXTDA = n'rDAXTDA.
(37)
For TDHF, the TDA is equivalent to the configuration interaction singles (CIS) method, where the excited states are determined by diagonahzing the singles part of the stationary Hamiltonian. The TDA was introduced to TDDFT by Grimme [97], who used additional
106
empirical parameters to correct some of the systematic errors: the above form (37) is due to Hirata and Head-Gordon [98]. A frequently used motivation for the TDA is its apparent computational advantage due to the reduction of dimensionality. This argument overlooks that, in an integral direct algorithm, the cost for computing a matrix-vector-product AX is approximately the same as the cost for computing the two matrix-vector-products (A + B) (X + Y) and (A — B)(X — Y). This is due to the lack of symmetry of the AO-transformed vector Xtll,a, which is neither symmetric nor skew-symmetric. In fact, in an integral-driven algorithm, where only non-redundant integrals {JJ,V\K\) are calculated in the innermost loop, AX has to be computed according to [65]
AX = ±[(A + B)X + (A-B)X];
(38)
this involves approximately the same operation count as the simultaneous formation of (A + B)(X + Y) and (A — B)(X — Y). The vector-vector operations performed in the MO basis are much less expensive and affect total CPU timings only marginally. A positive aspect of the TDA is its improved stability. It is well known that closedshell HF solutions may be unstable with respect to a spin-symmetry breaking [99]. The resulting instabilities lead to negative or imaginary excitation energies and a breakdown of the response formalism in its usual form. Triplet instabilities are a common limitation in TDHF theory, especially at geometries that differ significantly from the ground-state minimum. The KS reference is generally less susceptible to instabilities [100]: nevertheless, there is still a tendency to underestimate triplet excitation energies. The TDA alleviates this problem, because the variational constraint leads to systematically higher excitation energies. Transition moments are somewhat ill-defined in the TDA because of its lack of gauge invariance. For example, the length and velocity forms of the transition dipole moment may differ even in the basis set limit. Furthermore, the TDA does not satisfy the usual sum rules. These problems do not affect singlet-triplet excitations, where the transition moments vanish due to spin symmetry. 3.2.3. Other approximations Many approximations commonly made in ground state calculations are easily carried over to TDDFT. Examples are the frozen core approximation or the use of effective core potentials. We also mention semi-empirical approximations such as tight-binding DFT [101] here. The single-pole approximation [23,102] is mainly used in physics and corresponds to first-order perturbation theory for the excitation energies starting from the KS orbital energy differences as zeroth order. It is often appropriate in small systems but breaks down in situations where excited states are nearly degenerate and strong configuration mixing occurs. 3.2.4- Solvent effects Electronic absorption and CD spectra usually exhibit a marked solvent dependence. A common approach to include these effects in quantum chemical calculations is based on classical electrostatic solvent models, e.g., the polarizable continuum model (PCM) [103] or COSMO [104]. In these models, the solvent is approximated by a polarizable
107 continuum, while the solute molecule is placed in a cavity, whose dielectric constant is set to one. The presence of the solvent leads to an additional external potential which depends itself on the charge density of the electrons. An extension of the PCM to TDDFT vertical excitation energies has been reported by Cossi and Barone [105]. The computed solvent shifts were found to be fairly accurate in benchmark applications to small molecules [106]. Solvent effects on excited state geometries have been studied in an approximate TDDFT framework by Tomasi and coworkers [107], A hybrid Car-Parrinello quantum mechanical/molecular mechanical (QM/MM) approach which includes the solvent explicitly has recently been applied to the ground and first excited singlet state of acetone in water [108]. 3.3. Basis set effects Flexible Gaussian basis sets developed for ground states are usually well suited for lowlying valence excited states. Split valence basis sets with polarization functions on all nonhydrogen atoms such as 6-31G* [109] or SV(P) [110] are useful in exploratory calculations or larger applications. These basis sets systematically overestimate excitation energies by several tenths of an eV, and transition moments are qualitative only. Exceptions are larger planar systems, where transitions in the molecular plane can be accurate in small basis sets already. For states with Rydberg character and higher excitations, diffuse augmentation is necessary. Usually one adds atom-centered or molecule-centered primitive Gaussians whose exponents are determined by downward extrapolation or by optimization for atomic anions [111]. In general, diffuse functions should be used sparingly, to avoid imbalance and unnecessary computational cost. Continuum excitations show poor or no basis set convergence [52] and require special techniques. The KS ionization threshold should therefore always be checked in TDDFT excited state calculations. The basis set dependence of excited state structures, dipole moments, and force constant parallels that observed in ground states [74]; for example, C-C bonds lengths are usually overestimated by ca. 1 pm in split-valence basis sets. For most applications, triple zeta valence basis sets with two sets of polarization functions, i.e., 2dlf for first-row elements, yield basis set errors well below the systematic errors of current functionals. Examples of triple zeta valence basis sets are the segmented contracted TZVPP [112], or Dunning's cc-pVTZ [113], which uses generalized contractions. Larger basis sets are necessary for benchmark and basis set convergence studies. For DFT total energies, basis set convergence within "chemical accuracy" is reached at the quadruple zeta valence level [114]. 3.4. Examples A number of commercial quantum chemistry programs support the calculation of TDDFT vertical excitation energies, e.g., ADF [115], CADPAC [116], deMon [91], Gaussian [117], Q-Chem [118], PARAGAUSS [22], and TURBOMOLE [119]. The demonstrative CPU timings in Table 2 are from Ref. [74] and were obtained using TURBOMOLE V5-4. The RI-J method was employed for the cases denoted "RI-TDDFT": no other approximations were made. The examples in Table 2 show that TDDFT calculations are practicable for systems with several hundreds of atoms and several thousands of basis functions, even on low-end personal computers. TDDFT is thus becoming a challenge for semi-empirical methods, which have almost exclusively been used for applications of this size.
108 Table 2 CPU timings (hours) for the calculation of excitation and CD spectra, p is the number of (symmetry allowed) excitations including degeneracy, A ^ F is the number of Cartesian basis functions. The computer platforms (P) include a 1.2 GHz Athlon PC (A) and a 440 MHz HP J5000 (B) workstation (both single processor). System Sym. Method Basis/Grid NBF V CPU P Tris(alanine)-Co ln C3 B3LYP SVP/m3 386 100 12:04 B Cu-phthalocyanine" D4h B3LYP SVP/4 706 90 40:24 B Tetrathia-[7]helicene C 2 B3LYP SVP(s)/3 482 50 30:13 B SVP/m5 8100 3 19:17 B Fullerene C54<) Ih BP/RI SV(P)/m3 2804 300 128:04 A "Cd 10 Se 16 " ( ' T BP/RI S V ( P ) 7 m 3 1294 100 46:08 B Vancomycin d BP/RI SV(P) rf /m3 1600 100 62:18 A Methylcobalamine C\ BP/RI "Open shell (l CdloSe4(SePh)12(P"Pr3)4 "Optimized SZ basis sets on all weakly polarized alkyl und phonyl moieties ''Larger TZVDP+f basis set for cobalt
CD
E z> Q_
o
10
15
20
25
30
35
40
Fig. 1. CPU time for computing a single-point excited state energy plus gradient with and without the RI- J approximation as a function of the number of thiophene rings n. We used the BP86 functional and a TZVPP basis set. The calculations were performed on a 1.2 GHz Athlon PC.
109 The geometry optimization of the 2lA state of chlorophyll a may serve as an example for the efficiency of the RI-J approximation, as implemented in TuRBOMOLE. The BP86 functional and a SV(P) basis set were used, leading to a total of 1114 Cartesian basis functions. The overall calculation took 13 geometry cycles starting from the optimized ground state geometry and required 29:57 h of CPU time on a 2.4 GHz Pentium IV PC. Fig. 3.4 displays the scaling of computational cost for single-point excited state gradient calculations with and without the RI-J approximation. We consider a, a'-oligothiophenes with increasing chain length. Both methods show the expected N2 scaling, but with different pre-factors. For the larger members of the series, the RI-J approximation leads to a reduction in total computation times of a factor of 4-6. 4. VALIDATION 4.1. Vertical excitation and CD spectra Semi-local functionals predict low-lying valence excitation energies with errors in the range of 0.4 eV [69,91,98,120-127], There is a systematic underestimation [69] which may be due to the missing integer derivative discontinuity [128]. This underestimation is larger for singlet-triplet excitations [129,130]. Hybrid functionals yield smaller but less systematic errors, at somewhat higher cost. Contemporary TDDFT methods certainly cannot claim "chemical accuracy" (errors < 0.05 eV), but they are often accurate enough to make useful predictions. Calibration with accurate experimental or theoretical results for small systems is always recommendable. The domain of TDDFT are larger systems, where experimental inaccuracies may be comparable to the systematic errors of TDDFT, and correlated ah initio methods are (still) too expensive. With errors of 1-2 eV and more, traditional CIS and TDHF methods are considerably less accurate than TDDFT, despite similar or higher computational requirements. There are situations where semi-local functional tend to produce much larger errors, though. Lower-lying diffuse states are often too low in energy, and higher Rydberg states are spuriously unbound [52,131]. Similarly, the excitation energies of charge transfer (CT) and ionic states may be considerably underestimated [128,132.133]. In conjugated aromatic compounds [134] and polymers [135], the error in CT excitation energies increases with the chain length, and excitons may be erroneously unbound [136]. These failures may partly be traced to the self-interaction problem of semi-local functionals which has been known for a long time [137]. The classical Coulomb energy contains self-interaction which semi-local functionals do not cancel properly in strongly inhomogeneous systems. As a result, an electron "sees" the effective charge of N rather than N — I other electrons in the asymptotic tail of the density. The asymptotic correction schemes mentioned in Sec. 2.3.4 partly remedy this problem by imposing the correct — 1/r-behavior on the exchange-correlation potential. They do not improve the description of CT states, however. Correction schemes have been devised to estimate the missing derivative discontinuity in CT excitation energies from A SCF calculations [132,133]. At present, these approaches are mainly of diagnostic value because they depend on assumptions such as complete charge separation that may not be satisfied in many situations. The EXX methodology offers a more fundamental solution to the self-interaction problem. EXX potentials are self-interaction free and lead to a correct description of diffuse
110
states [138-140], and optical properties of conjugated polymers are improved [141]. Efficient methods to generate exact [44, 45] or approximate [46-48] EXX potentials for molecular systems are available. So far. they have been combined with adiabatic LSDA or GGA kernels; the EXX kernel is frequency-dependent and applications have been reported for solids only [49]. The dilemma of the EXX method is that, although it solves the Coulomb self-interaction problem, it does not improve consistently upon semi-local functionals for all systems and properties. For example, excitation energies of valence excited states are not better or even worse [138,140]. Unfortunately, the error cancellation between approximate exchange and correlation in semi-local functionals is lost when exact exchange is combined with semilocal correlation functionals. Hybrid functionals compromise between these extremes by using only a fraction of exact exchange. While this is not a general solution, it works often surprisingly well even for CT [142] and diffuse [143] states. In the long term, the development of correlation functionals compatible with exact exchange remains desirable. Oscillator strengths of well-separated states are usually predicted with errors in the 10% range [125]. They can be qualitatively wrong for strongly coupled states (as in most other methods). As the excitation energy approaches the KS ionization threshold, i.e., the negative HOMO energy, the density of states increases and a reliable assignment of individual transitions becomes impossible. This can be a major limitation in applications, especially to smaller systems and negative anions, because GGA potentials are too repulsive which results in too few bound states, as explained above. In other cases, one finds spurious intruder states which "steal" intensity from adjacent transitions of the same symmetry [144], Nevertheless, apart form the technical difficulties associated with continuum states, the overall shape of the computed spectra is often accurate [145]. This is also true if states with strong double excitation character are involved [30]. Pure double excitations are entirely missing in the TDDFT spectra [129], as a consequence of the AA. Trends observed for calculated rotatory strengths are generally similar to those observed for oscillator strengths [73,146]. Rotatory strengths of individual transitions may even have the wrong sign; but the overall CD spectra are often fairly accurate. The use of gauge origin invariant London orbitals does not seem to be necessary [147]. The simulation of CD spectra by TDDFT calculations is becoming increasingly popular as an inexpensive method to determine the absolute configuration; additional information is provided by optical rotations which can be calculated as well [148-151]. TDDFT works for inherently chiral chromophores [152] and transition metal complexes [153,154], but has problems with weakly disturbed, inherently achiral chromophores and systems with Rydberg-valence mixing [155]. 4.2. Excited state properties As analytical gradients of the excited state energy have become available only recently [17,156-158], the literature on excited state properties obtained with TDDFT is still limited. A comparison with accurate spectroscopic data for small systems shows that TDDFT excited state structures, dipole moments, and vibrational frequencies are of similar accuracy as the corresponding DFT ground state properties [17]. Case studies for other systems [159,160] and correlated ah initio results [161] corroborate this finding, which is somewhat unexpected in view of the relatively large errors in the excitation en-
Ill
ergies. Obviously, properties such as structures or dipole moments are less sensitive to deficiencies of current exchange-correlation functionals, e.g., self-interaction. The traditional CIS method, which has almost exclusively been used for geometry optimization of excited states in larger systems, is considerably less accurate at similar or even larger computational cost. Another significant advantage of TDDFT over HF-based methods for excited states is the enhanced stability of the KS reference compared to the HF reference, as discussed in Sec. 3.2.2. As a result, even excited state minima distant form the ground state minimum are mostly reasonable with TDDFT. Adiabatic excitation energies thus show basically the same error pattern as vertical excitation energies. Excited state vibrational frequencies can be used to identify the structure of excited states by comparison with, e.g., time-dependent infrared (TIR) or time-dependent resonance Raman (TRR) spectra from pump-probe experiments [162]. This is a promising combination, because TDDFT is applicable to fairly large systems and the information contained in the experimental spectra is difficult to interpret. In addition, the vibronic fine structure of UV spectra can be simulated within the Franck-Condon and HerzbergTeller approximations. Applications to aromatic hydrocarbons show a very encouraging agreement with experiments [163]. 4.3. Excited state dynamics Early work by Casida [164] and Domcke and coworkers [165] indicated that TDDFT can provide qualitatively correct excited state reaction paths. The validation is difficult and has to rely almost exclusively on accurate ab initio results. For the conical intersection in the retinal model Z-penta-2,4-dieniminium, TDDFT and CASPT2 (complete active space self-consistent field plus second order perturbation theory) single-point results are in agreement, while deviations have been reported for other systems [166]. A limitation most studies is that the calculated reaction paths do not correspond to minimum energy paths (MEPs), i.e., the internal degrees of freedom other than the reaction coordinate are not relaxed. The first full MEP calculations using TDDFT have been performed only recently [162]. For an adequate treatment of conical intersections and excited state dynamics, non-adiabatic coupling needs to be taken into account [167,168]. It seems unlikely that present functionals are accurate enough for predicting, e.g., barrier heights, but definite conclusions will have to await further studies. 5. APPLICATIONS 5.1. Aromatic compounds and fullerenes Aromatic compounds are among the most frequently investigated molecules in TDDFT studies. Several papers on singlet and triplet excitation energies of condensed polycyclic aromatic hydrocarbons (PAHs) [92,169-172] have appeared. In a recent study Grimme and Parac [134] have pointed out that the energy of the ionic La states [173] is significantly underestimated by common functionals. PAHs and their cations have also attracted interest due to their proposed occurrence the dark interstellar matter [127,174-177]. Chiroptical properties of a series of helicenes have been investigated in a joint experimental and theoretical study [73]. The simulated CD spectra are accurate enough to assign the absolute configuration and can even be used to distinguish derivatives with
112
substituents coupling to the aromatic TT system. CD spectra calculated with the DFT/SCI method have been used by Grimme and co-workers for structure elucidation of paracyclophanes [178]. Recently, the absolute configuration of enantiopure 9,9'-biathryls could be assigned by means of CD calculations [179]. For small aromatic heterocycles accurate excited state calculations with correlated ab initio methods are available. TDDFT studies focus on solvation effects [180,181], excited state dynamics [182-185], and larger systems [186-191]. Moreover, TDDFT calculations complement experimental investigations of newly synthesized ring systems like tetrathiafulvalene [192,193] and trithiapentalene [194]. Other recent TDDFT studies deal with indole derivatives related to tryptophane metabolism and melanin formation [195,196]. Laaksonen an co-workers [197] have investigated photochemical properties of urocanic acid, a human skin chromophore which plays a role in photo-immunosuppression and skin cancer. Mechanisms of photoisomerization of prototypical molecular switches azobenzene [198,199] and stilbene [200,201] have been the subject of other studies. Finally, the biological activity of the naturally occurring heterocycles luciferin [202] and flavins [203] has been investigated with TDDFT. Luciferin is responsible for the bioluminescence of fireflies, while flavins play a role in hydrogen transfer in cells. So far, the only practicable route to prepare pure fullerenes is based on soot extraction. Because of the extremely small yields, electronic absorption spectroscopy is, besides NMR measurements, the most important method for the characterization of fullerenes. Apart from a uniform red-shift, TDDFT using GGA functionals predicts the absorption spectra of large gap fullerenes with surprising accuracy [204], Small gap fullerenes are highly reactive and can presently only be studied theoretically. For example, of the seven isomers of Cgo obeying the isolated pentagon rule, only three have a large gap, and two of those have been observed [205]. Other studies focus on functionalized and substituted fullerenes [206, 207], carbon nanotubes [208] and sheets [209, 210]. Lower symmetric larger fullerenes frequently exhibit inherent chirality. In contrast to semi-empirical methods, TDDFT is well suited to determine the absolute configuration of chiral fullerenes, as has been shown for -D2-C84 [152] as well as C76 and C78 isomers [74]. TDDFT calculations on C ^ have been used to assign the photoelectron spectrum of stable Cs4 dianions [211]. 5.2. Porphyrins and related compounds Porphyrins, phthalocyanines, porphyrazines, and similar heterocyclic systems show a variety of optical and photochemical properties that are of interest from a biochemical as well as a technological point of view. The first rationale of the characteristic features observed in the absorption spectra of porphyrins was given by Gouterman [212, 213] in 1961. It is based on a simple perimeter model for [18]-annulene, the basic building unit of porphyrins. In Gouterman's scheme two energetically close pairs of orbitals, the two highest occupied molecular orbitals (HOMO and HOMO-1) and the two lowest virtual MOs (LUMO and LUMO+1), are involved in the lowest singlet transitions and are responsible for the so-called Q- and B-bands of porphyrins. For the free base porphin, the weaker pair of Q-bands (Q;,, and Qy) is found in the visible region whereas the substantially more intensive B-band (Soret band) is located in the near UV, see Fig. 5.2. The Q.,; and Qy bands were ascribed to the HOMO —> LUMO transition and the antisymmetric combination of HOMO —> LUMO+1 and HOMO-1
113 B
1 -
intens;ity (arb. unr
CO
A 1
0.8 0.6 -
I
0.4 0.2 -
M
L
J
ll
PL
0 200 250 300 350 400 450 500 550 600 650 700 wavelength (nm)
Fig. 2. The absorption spectrum of free base porphin. The experimental spectrum is from Ref. [214]. Calculated BP86/aug-SVP oscillator strengths [215] are indicated by sticks.
—> LUMO transitions, respectively. The symmetric combination of the latter two was considered as the origin of the Soret band. Porphyrin derivatives and analogues exhibit characteristic energy shifts and intensity patterns in the same energy range. The first TDDFT results on free base porphin were reported by Bauernschmitt and Ahlrichs [69] and later confirmed by Scuseria and co-workers [66]. Subsequent studies by van Gisbergen, Baerends and co-workers and by Sundholm addressed the validation of the four-orbital model of Gouterman for the free base porphin and the assignment of its UV/VIS spectrum [215-218]. Investigations by Parusel and co-workers employed the DFT/SCI [219] and DFT/MRCI methods [220] for the same purpose. While a correspondence to the Gouterman model can be established for the Q bands, the origin of the intense B band is still under discussion. It appears that lower occupied orbitals are significantly involved in these transitions [215, 217], and a non-negligible contribution from double excitations is suggested from DFT/MRCI results [220]; thus, the simple four-orbital model does not hold. A similar picture emerges for porphyrazine [217, 221], corrphycene [222] and corrin [223] molecules where Gouterman's model provides a rough description of low-lying electronic transitions. Positions of electronic excitations in porphyrins are further strongly affected by conformational flexibility of the macrocycle, deviations from planarity leading to red shifts of Q- and B-bands. The suggestion that nonplanarity of hemes in hemoproteins and photosynthetic proteins may influence their biological activity [224] stimulated much research on saddled and ruffled forms on porphyrins. Porphyrin diacids [225,226] and complexes bearing aromatic substituents
114 [227-230] have been investigated as well. Porphyrinoid systems have a, tendency to form chelate complexes with various metal cations. Two large groups of complexes can be distinguished by their spectral behavior, denoted regular and irregular porphyrins by Gouterman [213]. Main group and closedshell transition metal cations form regular complexes that largely resemble the parent macrocycles because the contribution of the metal to the frontier orbitals is small. This was shown by Nguyen, Baerends, and co-workers for Zn11 [231-235] and by Sundholm for Mg n complexes [236]. In irregular metal complexes incomplete d-shells of transition metal cations interact strongly with the 7r-system of the ligand; substantially different optical properties [237-243] result. The most important representatives of this class are iron and cobalt complexes which are closely related to heme [244,245] and vitamin B12 [246,247]. 5.3. Transition metal compounds For calculations of optical properties of transition metal complexes, TDDFT is often the method of choice. In most cases the accuracy of TDDFT is sufficient for an assignment of excitations in closed-shell oxide, carbonyl and cyclopentadienyl complexes [121,248250]: hybrid functionals do not always lead to an improvement for these systems. Ligand field d to d transitions appear at too high energies as a result of self-interaction error, as Autschbach and co-workers have shown for Co111 and Rh m complexes [154]. Difficulties are encountered for small open-shell molecules such as ScO or VO [251,252]. The diversity of photophysical and photochemical properties of transition metal complexes is reflected in TDDFT investigations on this class of compounds. Possible applications in photocatalysis and solar energy conversion have triggered research on complexes of copper [253,254], chromium [255], ruthenium [256,257], paladium [258], platinum [259] and rhenium [260] with aromatic heterocyclic ligands as 9,10-phenantroline (phen), a, a'-bipyridyl (bipy) or dipyrido[3,2-a:2',3'-c]phenazine (dppz). Dissociation and rearrangement dynamics upon photoexcitation has been discussed in connection with [Fe(CN)5(NO)]2~ [261, 262], [Cr(CO)5L] and [Fe(CO)4L] [263], as well as on [Ru(PH3)3(CO)(H2)] complexes [264]. The catalytical activity of titanium complexes for polymerization and oxidation reactions has motivated several studies on titanocenes [265, 266] and alkoxy complexes [267]. Optical properties and bond dissociation of alkylplatinum complexes are the subject of a recent study by van Slageren and co-workers [268]. TDDFT calculations for neutral dithiolene complexes of nickel, palladium, and platinum have explained the uncommon properties of these compounds, especially the presence of an exceedingly strong absorption in the near IR region [269]. Other studies investigate the photophysics and the luminescence behavior of cyclometalated complexes of rhodium [270] and indium [271]. 5.4. Metal and semiconductor clusters Metal clusters differ substantially in their properties from the bulk phase [272, 273] and have received much attention in connection with possible applications in nanotechnology and heterogenous catalysis. Experimental structure determination is a difficult task even for small clusters, and theoretical results are particularly helpful. Flexible structures, a large number of competing minima, and low-lying excited states are difficult challenges for all electronic structure methods. Most theoretical work therefore address the most simple class of metal cluster compounds, alkali metal clusters, for which reliable experimental data as well as accurate quantum chemical calculations [273, 274] exist. TDDFT
115 applications on alkali metal clusters range from simple jellium [275-279] to full TDDFT calculations employing GGA functionals [280-282], Comparison with available experimental data indicates a good accuracy of TDDFT results with typical errors of 0.1-0.2 eV or less in excitation energies [283-285]. For the dimers Li2, Na2 and K2, experimental vertical excitation energies are overestimated by TDDFT [281], in contrast to the usual behavior of the method. Photoabsorption spectra are reproduced satisfactorily as well [285-287]: finite temperature effects have been investigated by molecular dynamics simulations [288,289]. Similar studies have been performed for Al clusters [290,291], Coinage metal (Cu, Ag, Au) clusters are more complicated due to the presence of rather polarizable d-electrons. Very little direct structural information is available from experiment. Of particular interest is the transition from the planar structures that are the most stable isomers for small clusters to bulk-like three-dimensional aggregates [292]. While the simple jellium model does not perform very well in this case, the polarizable cluster core approximation [293-295] or full TDDFT calculations [296-298] provide better results for photoabsorption spectra. Nevertheless, transitions with s —> d character are notoriously in error, which is a consequence of self-interaction [137], In summary, TDDFT absorption spectra can give useful hints, but are presently not accurate enough for a unique determination of the geometric structure of most metal clusters. The band gap of semiconductor clusters can be altered over a wide range by varying the particle size; this makes them suitable materials for optoelectronic devices [299]. Recent TDDFT investigations have addressed optical properties of silicon [283, 287, 300305], gallium arsenide [287], as well as zinc sulfide, cadmium selenide, and related 12-16 clusters [306-311]. Most studies focus on the size dependence of the optical gap. With increasing cluster size the band gap is reduced as a result of quantum confinement, e.g. for hydrogenated Si clusters from 3.8 eV for Si47H107 to 2.5 eV for Si147H247 [300], Another important factor is the constitution of the cluster surface, with abstraction of hydrogen or oxidation leading to a substantial decrease of the absorption edge [299]. The definition of the optical gap is not straightforward, however, since the lowest electronic transitions are very weakly allowed in large clusters. Within these limitations both LSD A and gradient corrected functionals yield results in good agreement with experimental data. 5.5. Organic polymers Two different theoretical approaches have been used for polymers: solid state methods employing periodical boundary conditions, and oligomer methods considering discrete fragments of increasing size. For calculations of excitation energies of organic polymers, the latter seems to be more widespread, although a LCAO-crystalline orbital implementation of excitation energies of extended systems has been reported [312, 313]. For oligomer methods, the convergence of the calculated properties to the bulk limit and the quality of extrapolated properties are of primary interest. Several papers by Ratner, Zojer, and coworkers summarize computational results on different classes of polymers [314, 315], e.g., polyenes, polythiophenes, and polyphenylenes. From these results, the authors concluded [316] that extrapolation techniques are capable of providing correct band gaps for the polymers. However, empirical extrapolations with respect to 1/n, where n is the number of monomer units, may show significant systematic errors. Cai and co-workers [135] note a tendency to spurious metallic behavior and wrong ground state multiplicities in large
116 conjugated yr-systems. For polyenes the relative stability of l 1 ^ and 21A!j states (in C-ih symmetry), which is of importance for carotenoids of the light harvesting complex, has been extensively discussed [166,317-319]. Polythiophene [320-326] and polypyrrole [327, 328] polymers are important industrial materials for optoelectronic devices such as light emitting diodes (LEDs) have been the subject of numerous TDDFT studies. 5.6. Charge and proton transfer The geometric and electronic structure of a molecule can significantly change upon photoexcitation. Transfer of charge or protons are among the most simple photochemical reactions, and excitation energy transfer plays a fundamental role for the photosynthesis. In work of Parusel, Grimme. and others, intramolecular charge transfer (ICT) in donoracceptor substituted aromatic systems was investigated by TDDFT [329], DFT/SCI [330, 331], and DFT/MRCI [332-334] methods (see Ref. [329] for an overview). Most of the studies addressed 4-(N,N)-dimethylaminobenzonitrile (DMABN), a prototypical dual fluorescent compound showing a strong emission from the ICT state in polar solutions. In extensive studies by Jamorski and co-workers [142,335-337], the accuracy of TDDFT for exploration of intramolecular charge transfer phenomena has been assessed, and a classification for the emission properties of these compounds was presented [338,339]. A definite assignment of the structure of the two lowest singlet states has recently been given by means of TDDFT calculations [162] and confirmed by coupled cluster calculations [340]. Further investigations have dealt with solvent effects and photophysical properties of donor-substituted pyridine derivatives [341,342], Excited state proton transfer phenomena have been the subject of a number of TDDFT studies. So far, excited state proton transfer in salicylic acid and related aromatic compounds [165,343-346] as well as in 7-azonindole-water complexes [347] has been investigated. 5.7. Biologically relevant systems Most molecules of biological relevance are a challenge due to their size. Calculations of optical properties of chlorophylls and bacteriochlorophylls by Sundholm [144, 236, 348, 349] and Yamaguchi [350, 351] showed that good accuracy can be achieved with the BP86 and B3LYP functionals. Different aspects of the interaction between chlorophyll molecules and carotenoids and of the dynamics in the photosynthetic apparatus have been extensively studied by Dreuw, Fleming and co-workers [317, 352-354], Pullerits and co-workers have investigated the dependence of excitation energies of bacteriochlorphyll on the local environment represented by a uniform electric field [355]. The dissociation dynamics of CO-hemoglobin complexes has recently been studied by Head-Gordon and co-workers [244, 245] who showed that excitation into the & A" and the 3lA' states of the complex leads to repulsive interaction and dissociation of the CO molecule. Photochemistry of nucleic acid bases is relevant for an understanding of DNA damage by UV irradiation and cellular repair mechanisms. Absorption spectra, tautomeric equilibria, and excited state geometries of adenine [356] and cytosine [357] have been reported. A comprehensive study on absorption properties of DNA bases has appeared recently [358]. The thermochemistry of thymine dimer formation and photoinduced cycloreversion reactions occuring in DNA repair mechanisms have been investigated by Durbeej and Eriksson [359,360], TDDFT calculations on complexes of thymine with psoralene have
117 been performed to clarify the effect of psoralenes which are utilized in photochemotherapy [361]. TDDFT calculations allow to go beyond model compounds and investigate larger fragments of biological systems like the photoactive centers of green fluorescent protein [362] or photoactive yellow protein [363]. Future improvements of TDDFT such as a better description of solvation effects or QM/MM coupling may help to provide deeper insight into photochemical processes in living organisms. 6. OUTLOOK Many phenomena in photochemistry are still not well understood, even in small model systems. The enormous complexity of photochemical processes will require a combined effort of theory and experiment to extend the frontier of our knowledge to real systems of technical and biological interest. It is clear by now that TDDFT has the potential to play an important role in this development, besides more accurate methods and experimental techniques. Nevertheless, contemporary TDDFT is not a black box method, and every user of commercial TDDFT codes should be aware of its limitations. ACKNOWLEDGMENTS We would like to thank R. Ahlrichs for helpful comments. This work was supported by the Center for Functional Nanostructures (CFN) of the Deutsche Forschungsgemeinschaft (DFG) within project C2.1. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9]
[10] [11] [12] [13] [14]
[15] [16]
P. Hohenberg and W. Kohn, Phys. Rev., 136 (1964) 864. L. N. Oliveira, E. K. U. Gross, and W. Kohn. Phys. Rev. A, 37 (1988) 2821. I. Frank, J. Hutter, D. Marx, and M. Parrinello, J. Chem. Phys., 108 (1998) 4060. A. Gorling, Phys. Rev. A, 59 (1999) 3359. M. Levy and A. Nagy, Phys. Rev. Lett., 83 (1999) 4361. E. K. U. Gross, J. F. Dobson, and M. Petersilka, Top. Curr. Chem., 181 (1996) 81. R. Van Leeuwen, J. Mod. Phys. B, 15 (2001) 1969. G. Onida, L. Reining, and A. Rubio, Rev. Mod. Phys., 74 (2002) 601. N. T. Maitra, A. Wasserman, and K. Burke, in A. Gonis, N. Kioussis, and M. Ciftan (eds.), Electron correlations and materials properties 2, Kluwer Academic/Plenum Press, New York, 2003 pp. 285-298. F. Furche and K. Burke, Ann. Rev. Comp. Chem., 1 (2005) in press. D. J. Thouless, The Quantum Mechanics of Many-Body Systems, vol. 11 of Pure and Applied Physics, Academic Press, New York, 2. edn., 1972. S. Tretiak and V. Chernyak, J. Chem. Phys., 119 (2003) 8809. E. Runge and E. K. U. Gross, Phys. Rev. Lett., 52 (1984) 997. M. E. Casida, in D. E. Chong (ed.), Recent Advances in Density Functional Methods, World Scientific, Singapore, vol. 1 of Recent Advances in Computational Chemistry, 1995 pp. 155-192. F. Furche, J. Chem. Phys., 114 (2001) 5982. O. V. P Salek, T. Helgaker, and H. Agren, J. Chem. Phys., 117 (2002) 9630.
118 [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50]
[51]
[52]
F. Furche and R. Ahlrichs, J. Chem. Phys., 117 (2002) 7433, J. Chem. Phys., in press (E). A. D. Becke, J. Chem. Phys., 98 (1993) 1372. A. D. Becke, J. Chem. Phys., 98 (1993) 5648. A. D. McLachlan and M. A. Ball, Rev. Mod. Phys., 36 (1964) 844. A. Zangwil and P. Soven, Phys. Rev. A, 21 (1980) 1561. A. Gorling, H. H. Heinze, S. P. Ruzankin, M. Staufer, and N. Rosch, J. Chem. Phys., 110 (1999) 2785. M. Petersilka, U. J. Gossmann, and E. K. U. Gross, Phys. Rev. Lett., 76 (1996) 1212. R. L. Martin, J. Chem. Phys., 118 (2003) 4475. M. Head-Gordon, A. M. Grana, D. Maurice, and C. A. White, J. Phys. Chem., 99 (1995) 14261. A. E. Hansen and T. D. Bouman, Adv. Chem. Phys., 44 (1980) 545. B. G. Johnson, P. M. W. Gill, and J. A. Pople, J. Chem. Phys., 98 (5612) 5612. J. Tao, J. P. Perdew, V. N. Staroverov, and G. E. Scuseria, Phys. Rev. Lett., 91 (2003) 146401. E. K. U. Gross and W. Kohn, Adv. Quantum Chemistry, 21 (1990) 255. N. T. Maitra, F. Zhang, R. J. Cave, and K. Burke, J. Chem. Phys., 120 (2004) 5932. O. V. Gritsenko, S. J. A. van Gisbergen, A. Gorling, and E. J. Baerends, J. Chem. Phys., 113 (2000) 8478. S. H. Vosko, L. Wilk, and M. Nusair, Can. J. Phys., 58 (1980) 1200. J. P. Perdew and Y. Wang, Phys. Rev. B, 45 (1992) 13244. A. D. Becke, Phys. Rev. A, 38 (1988) 3098. C. Lee, W. Yang, and R. G. Parr, Phys. Rev. B, 37 (1988) 785. J. P. Perdew, Phys. Rev. B, 33 (1986) 8822. J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett., 77 (1996) 3865. F. A. Hamprecht, A. J. Cohen, D. J. Tozer, and N. C. Handy, J. Chem. Phys., 109 (1998) 6264. T. Van Voorhis and G. E. Scuseria, J. Chem. Phys., 109 (1998) 400. P. J. Stephens, F. J. Devlin, C. F. Chabalowski, and M. J. Frisch, J. Phys. Chem., 98 (1994) 11623. J. P. Perdew, M. Ernzerhof, and K. Burke, J. Chem. Phys., 105 (1996) 9982. J. D. Talman and W. F. Shadwick, Phys. Rev. A, 14 (1976) 36. S. Hirata, S. Ivanov, I. Grabowski, R. J. Bartlett, K. Burke, and J. D. Talman, J. Chem. Phys., 115 (2001) 1635. S. Kiimmel and J. P. Perdew, Phys. Rev. Lett., 90 (2003) 043004. S. Kiimmel and J. P. Perdew, Phys. Rev. B, 68 (2003) 035103. F. Delia Sala and A. Gorling, J. Chem. Phys., 115 (2001) 5718. F. Delia Sala and A. Gorling, J. Chem. Phys., 116 (2002) 5374. M. Gruening, O. V. Gritsenko, and E. J. Baerends, J. Chem. Phys., 116 (2002) 6435. Y. H. Kim and A. Gorling, Phys. Rev. Lett., 89 (2002) 096402. T. Grabo, T. Kreibich, S. Kurth, and E. K. U. Gross, in V. I. Anisimov (ed.), Strong Coulomb Correlations in Electronic Structure Calculations: Beyond the Local Density Approximation, Gordon and Breach, Tokyo, 2000 pp. 203-311. E. K. U. Gross, T. Kreibich, M. Lein, and M. Petersilka, in A. Gonis, N. Kioussis, and M. Ciftan (eds.), Electron Correlations and Materials Properties, Plenum Press, New York, 1999 pp. 393-427. M. E. Casida, C. Jamorski, K. C. Casida, and D. R. Salahub, J. Chem. Phys., 108 (1998) 4439.
119 [53] D. J. Tozer and N. C. Handy, J. Chem. Phys., 109 (1998) 10180. [54] M. E. Casida and D. R. Salahub, J. Chem. Phys., 113 (2000) 8918. [55] P. R. T. Schipper, O. V. Gritsenko, S. J. A. van Gisbergen, and E. J. Baerends, J. Chem. Phys., 112 (2000) 1344. [56] G. Vignale and W. Kohn, Phys. Rev. Lett., 77 (1996) 2037. [57] G. Vignale, C. A. Ullrich, and S. Conti, Phys. Rev. Lett., 79 (1997) 4878. [58] P. L. de Boeij, F. Kootstra, J. A. Berger, R. van Leeuwen, and J. G. Snijders, J. Chem. Phys., 115 (2001) 1995. [59] M. van Faassen and P. L. de Boeij, J. Chem. Phys., 120 (2004) 8353. [60] C. Lanczos, J. Res. Nat. Bur. Stand., 45 (1950) 255. [61] M. R. Hestenes and E. Stiefel, J. Res. Nat. Bur. Stand., 49 (1952) 409. [62] E. R. Davidson, J. Comp. Phys., 17 (1975) 87. [63] M. Crouzeix, B. Philippe, and M. Sadkane, SIAM J. Sci. Comput., 15 (1994) 62. [64] J. Olsen, H. J. A. Jensen, and P. J0rgensen, J. Comp. Phys., 74 (1988) 265. [65] H. Weiss, R. Ahlrichs, and M. Haser, J. Chem. Phys., 99 (1993) 1262. [66] R. E. Stratmann, G. E. Scuseria, and M. J. Frisch, J. Chem. Phys., 109 (1998) 8218. [67] S. J. A. van Gisbergen, J. G. Snijders, and E. J. Baerends, Comput. Phys. Comm., 118 (1999) 119. [68] V. Chernyak, M. F. Schulz, S. Mukamel, S. Tretiak, and E. V. Tsiper, J. Chem. Phys., 113 (2000) 36. [69] R. Bauernschmitt and R. Ahlrichs, Chem. Phys. Lett., 256 (1996) 454. [70] B. O. Roos, Chem. Phys. Lett., 15 (1972) 153. [71] R. Ahlrichs, Comput. Phys. Commun., 17 (1979) 31. [72] R. Ahlrichs, W. Meyer, and C. Dykstra, in C. E. Dykstra (ed.), Advanced Theories and Computational Approaches to the Electronic Structure of Molecules, Reidel, Dordrecht, vol. 133 of NATO ASI Series C: Mathematical and Physical Sciences, 1984 pp. 19-38. [73] F. Furche, R. Ahlrichs, C. Wachsmann, E. Weber, A. Sobanski, F. Vogtle, and S. Grimme, J. Am. Chem. Soc, 122 (2000) 1717. [74] F. Furche, Dichtefunktionalmethoden fur elektronisch angeregte Molekule. Theorie - Implementierung - Anwendung, Ph.D. thesis, Universitat Karlsruhe, 2002. [75] P. R. Taylor, Int. J. Quant. Chem., 27 (1985) 89. [76] Z. Rinkevicius, I. Tunell, P. Salek, O. Vahtras, and H. Agren, J. Chem. Phys., 119 (2003) 34. [77] K. Eichkorn, O. Treutler, H. Ohm, M. Haser, and R. Ahlrichs, Chem. Phys. Lett., 240 (1995) 283. [78] K. Eichkorn, O. Treutler, H. Ohm, M. Haser, and R. Ahlrichs, Chem. Phys. Lett., 242 (1995) 652. [79] J. L. Whitten, J. Chem. Phys., 58 (1973) 4496. [80] B. I. Dunlap, J. W. D. Connolly, and J. R. Sabin, J. Chem. Phys., 71 (1979) 3396. [81] J. W. Mintmire and B. I. Dunlap, Phys. Rev. A, 25 (1982) 88. [82] O. Vahtras, J. Almlof, and M. W. Feyereisen, Chem. Phys. Lett., 213 (1993) 514. [83] B. I. Dunlap, J. Mol. Struct. (THEOCHEM), 529 (2000) 37. [84] R. Ahlrichs and K. Tsereteli, J. Comput. Chem., 23 (2002) 306. [85] R. Bauernschmitt, M. Haser, O. Treutler, and R. Ahlrichs, Chem. Phys. Lett., 264 (1997) 573. [86] K. Eichkorn, F. Weigend, O. Treutler, and R. Ahlrichs, Theor. Chem. Ace, 97 (1997) 119. [87] F. Weigend, Phys. Chem. Chem. Phys., 4 (2002) 4285.
120 [88] N. Godbout, D. R. Salahub, J. Andzelm, and E. Wimmer, Can. J. Chem., 70 (1992) 560. [89] K. R. Glaesemann and M. S. Gordon, J. Chem. Phys., 112 (2000) 10738. [90] C. K. Skylaris, L. Gagliardi, N. C. Handy, A. G. Ioannou, S. Spencer, and A. Willetts, J. Mol. Struct. (THEOCHEM), 501 (2000) 229. [91] C. Jamorski, M. E. Casida, and D. R. Salahub, J. Chem. Phys., 104 (1996) 5134. [92] H. H. Heinze, A. Goling, and N. R6sch, J. Chem. Phys., 113 (2000) 2088. [93] S. J. A. van Gisbergen, J. G. Snijders, and E. J. Baerends, J. Chem. Phys., 103 (1995) 9347. [94] S. J. A. van Gisbergen, C. F. Guerra, and E. J. Baerends, J. Comput. Chem., 21 (2000) 1511. [95] F. Neese and G. Olbrich, Chem. Phys. Lett., 362 (2002) 170. [96] D. Rappoport and F. Furche, submitted. [97] S. Grimme, Chem. Phys. Lett., 259 (1996) 128. [98] S. Hirata and M. Head-Gordon, Chem. Phys. Lett., 314 (1999) 291. [99] J. Cizek and J. Paldus, J. Chem. Phys., 47 (1967) 3976. [100] R. Bauernschmitt and R. Ahlrichs, J. Chem. Phys., 104 (1996) 9047. [101] T. Frauenheim, G. Seifert, M. Elstner, T. Niehaus, C. Kohler, M. Amkreutz, M. Sternberg, Z. Hajnal, A. Di Carlo, and S. Suhai, J. Phys. Cond. Mat., 14 (2002) 3015. [102] H. Appel, E. K. U. Gross, and K. Burke, Phys. Rev. Lett., 90 (2003) 043005. [103] C. Amovilli, V. Barone, R. Cammi, M. Cances, M. Cossi, B. Mennucci, C. S. Pomelli, and J. Tomasi, Adv. Quantum Chem., 32 (1998) 264. [104] A. Klamt and G. Schiiurman, J. Chem. Soc, Perkin Trans., 2 (1993) 799. [105] M. Cossi and V. Barone, J. Chem. Phys., 115 (2001) 4708. [106] F. Aquilante, M. Cossi, O. Crescenzi, G. Scalmani, and V. Barone, Mol. Phys., 101 (2003) 1945. [107] R. Cammi, B. Mennucci, and J. Tomasi, J. Phys. Chem. A, 104 (2000) 5631. [108] U. F. Rohrig, I. Frank, J. Hutter, A. L. J. VandeVondele, and U. Rothlisberger, ChemPhysChem, 4 (2003) 1177. [109] P. C. Hariharan and J. A. Pople, Theoret. Chimica Acta, 28 (1973) 213. [110] A. Schafer, H. Horn, and R. Ahlrichs, J. Chem. Phys., 97 (1992) 2571. [Ill] R. A. Kendall, T. H. Dunning, Jr., and R. J. Harrison, J. Chem. Phys., 96 (1992) 6796. [112] A. Schafer, C. Huber, and R. Ahlrichs, J. Chem. Phys., 100 (1994) 5829. [113] T. H. Dunning, Jr., J. Chem. Phys., 90 (1989) 1007. [114] F. Weigend, F. Furche, and R. Ahlrichs, J. Chem. Phys., 119 (2003) 12753. [115] G. te Velde, F. Bickelhaupt, E. Baerends, C. Fonseca Guerra, S. J. A. van Gisbergen, J. Snijders, and T. Ziegler, J. Comput. Chem, 22 (2001) 931. [116] R. D. Amos, I. L. Alberts, J. S. Andrews, S. M. C. ell, N. C. Handy, D. Jayatilaka, P. J. Knowles, R. Kobayashi, G. J. Laming, A. M. Lee, P. E. Maslen, C. W. Murray, P. Palmieri, J. E. Rice, E. D. Simandiras, A. J. Stone, M.-D. Su, and D. J. Tozer, CADPAC6.5, The Cambridge Analytic Derivatives Package, Cambridge, UK, 1998. [117] M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R. Cheeseman, J. A. Montgomery Jr., T. Vreven, K. N. Kudin, J. C. Burant, J. M. Millam, S. S. Iyengar, J. Tomasi, V. Barone, B. Mennucci, M. Cossi, G. Scalmani, N. Rega, G. A. Petersson, H. Nakatsuji, M. Hada, M. Ehara, K. Toyota, R. Fukuda, J. Hasegawa, M. Ishida, T. Nakajima, Y. Honda, O. Kitao, H. Nakai, M. Klene, X. Li, J. E. Knox, H. P. Hratchian, J. B. Cross, C. Adamo, J. Jaramillo, R. Gomperts, R. E. Stratmann, O. Yazyev, A. J. Austin, R. Cammi, C. Pomelli, J. W. Ochterski, P. Y. Ayala, K. Morokuma, G. A. Voth, P. Salvador, J. J. Dannenberg, V. G. Zakrzewski, S. Dapprich, A. D. Daniels, M. C. Strain,
121
[118]
[119] [120] [121] [122] [123]
O. Farkas, D. K. Malick, A. D. Rabuck, K. Raghavachari, J. B. Foresman, J. V. Ortiz, Q. Cui, A. G. Baboul, S. Clifford, J. Cioslowski, B. B. Stefanov, G. Liu, A. Liashenko, P. Piskorz, I. Komaromi, R. L. Martin, D. J. Fox, T. Keith, M. A. Al-Laham, C. Y. Peng, A. Nanayakkara, M. Challacombe, P. M. W. Gill, B. Johnson, W. Chen, M. W. Wong, C. Gonzalez, and J. A. Pople, Gaussian Inc., Pittsburgh PA, 2003. J. Kong, C. A. White, A. I. Krylov, D. Sherrill, R. D. Adamson, T. R. Furlani, M. S. Lee, A. M. Lee, S. R. Gwaltney, T. R. Adams, C. Ochsenfeld, A. T. B. Gilbert, G. S. Kedziora, V. A. Rassolov, D. R. Maurice, N. Nair, Y. Shao, N. A. Besley, P. E. Maslen, J. P. Dombroski, H. Daschel, W. Zhang, P. P. Korambath, J. Baker, E. F. C. Byrd, T. V. Voorhis, M. Oumi, S. Hirata, C.-P. Hsu, N. Ishikawa, J. Florian, A. Warshel, B. G. Johnson, P. M. W. Gill, M. Head-Gordon, and J. A. Pople, J. Comput. Chem., 21 (2000) 16. R. Ahlrichs, M. Bar, M. Haser, H. Horn, and C. Kolmel, Chem. Phys. Lett., 162 (1989) 165, current version: see http://www.turbomole.com. S. Hirata and M. Head-Gordon, Chem. Phys. Lett., 302 (1999) 375 . S. J. A. van Gisbergen, J. A. Groeneveld, A. Rosa, J. G. Snijders, and E. J. Baerends, J. Phys. Chem. A, 103 (1999) 6835. C. Adamo and V. Barone, Theo. Chem. Ace, 105 (2000) 169. J. Guan, M. E. Casida, and D. R. Salahub, J. Mol. Struct. (THEOCHEM), 527 (2000) 229.
[124] D. Guillaumont and S. Nakamura, Dyes and Pigments, 46 (2000) 85. [125] N. N. Matsuzawa, A. Ishitani, D. A. Dixon, and T. Uda, J. Phys. Chem. A, 105 (2001) 4953. [126] M. Parac and S. Grimme, J. Phys. Chem. A, 101 (2002) 6844. [127] S. Hirata, M. Head-Gordon, J. Szczepanski, and M. Vala, J. Phys. Chem. A, 107 (2003) 4940. [128] D. J. Tozer, J. Chem. Phys., 119 (2003) 12697. [129] D. Tozer and N. C. Handy, Phys. Chem. Chem. Phys., 2 (2000) 2117. [130] M. N. Paddon-Row and M. J. Shephard, J. Phys. Chem. A, 106 (2002) 2935. [131] A. Wasserman, N. T. Maitra, and K. Burke, Phys. Rev. Lett., 91 (2003) 263001. [132] M. E. Casida, F. Gutierrez, J. G. Guan, F. X. Gadea, D. R. Salahub, and J. P. Daudey, J. Chem. Phys., 113 (2000) 7062. [133] A. Dreuw, J. L. Weisman, and M. Head-Gordon, J. Chem. Phys., 119 (2003) 2943. [134] S. Grimme and M. Parac, ChemPhysChem, 4 (2003) 292. [135] Z. L. Cai, K. Sendt, and J. R. Reimers, J. Chem. Phys., 117 (2002) 5543. [136] S. Tretiak, K. Igumenshchev, and V. Chernyak, preprint. [137] J. P. Perdew and A. Zunger, Phys. Rev. B, 23 (1981) 5048. [138] S. Hirata, S. Ivanov, I. Grabowski, and R. J. Bartlett, J. Chem. Phys., 116 (2002) 6468. [139] S. Hamel, M. E. Casida, and D. R. Salahub, J. Chem. Phys., 116 (2002) 8276. [140] A. G. Delia Sala, F, Int. J. Quant. Chem., 91 (2003) 131. [141] S. J. A. van Gisbergen, P. R. T. Schipper, O. V. Gritsenko, E. J. Baerends, J. G. Snijders, B. Champagne, and B. Kirtman, Phys. Rev. Lett., 83 (1999) 694. [142] C. Jamorski, J. B. Foresman, C. Thilgen, and H. P. Luthi, J. Chem. Phys., 116 (2002) 8761. [143] C. Adamo, G. E. Scuseria, and V. Barone, J. Chem. Phys., I l l (1999) 2889. [144] D. Sundholm, Phys. Chem. Chem. Phys., 5 (2003) 4265. [145] A. Wassermann, N. T. Maitra, and K. Burke, Phys. Rev. Lett., 91 (2003) 263001. [146] J. Autschbach, T. Ziegler, S. J. A. van Gisbergen, and E. J. Baerends, J. Chem. Phys.,
122 116 (2002) 6930. [147] M. Pecul, K. Ruud, and T. Helgaker, Chem. Phys. Lett., 388 (2004) 110. [148] P. J. Stephens, F. J. Devlin, J. R. Cheeseman, and M. J. Frisch, J. Phys. Chem. A, 105 (2001) 5356. [149] S. Grimme, F. Furche, and R. Ahlrichs, Chem. Phys. Lett., 361 (2002) 321. [150] J. Autschbach, S. Patchkovskii, T. Ziegler, S. J. A. van Gisbergen, and E. J. Baerends, J. Chem. Phys., 117 (2002) 581. [151] P. Norman, K. Ruud, and T. Helgaker, J. Chem. Phys., 120 (2004) 5027. [152] F. Furche and R. Ahlrichs, J. Am. Chem. Soc, 124 (2002) 3804. [153] J. Autschbach, F. R. Jorge, and T. Ziegler, J. Comput. Chem., 42 (2003) 2867. [154] J. Autschbach, F. E. Jorge, and T. Ziegler, Inorg. Chem., 42 (2003) 2867. [155] C. Diedrich and S. Grimme, J. Phys. Chem. A, 107 (2003) 2524. [156] C. van Caillie and R. D. Amos, Chem. Phys. Lett., 308 (1999) 249. [157] C. van Caillie and R. D. Amos, Chem. Phys. Lett., 317 (2000) 159. [158] J. Hutter, J. Chem. Phys., 118 (2003) 3928. [159] R. Burcl, R. D. Amos, and N. C. Handy, Chem. Phys. Lett., 355 (2002) 8. [160] M. Odelius, D. Laikov, and J. Hutter, J. Mol. Struct. (THEOCHEM), 630 (2003) 163. [161] A. Kohn and C. Hattig, J. Chem. Phys., 119 (2003) 5021. [162] D. Rappoport and F. Furche., J. Am. Chem. Soc, 126 (2004) 1277. [163] M. Dierksen and S. Grimme, J. Chem. Phys., 120 (2004) 3544. [164] M. E. Casida, K. C. Casida, and D. R. Salahub, Int. J. Quant. Chem., 70 (1998) 933. [165] A. L. Sobolewski and W. Domcke, Phys. Chem. Chem. Phys., 1 (1999) 3065. [166] M. Wanko, M. Garavelli, F. Bernardi, A. Niehaus, T. Frauenheim, and M. Elstner, J. Chem. Phys., 120 (2004) 1674. [167] V. Chernyak and S. Mukamel, J. Chem. Phys., 112 (2000) 3572. [168] N. L. Doltsinis and D. Marx, Phys. Rev. Lett., 88 (2002) 166402. [169] S. Hirata, T. J. Lee, and M. Head-Gordon, J. Chem. Phys., I l l (1999) 8904. [170] T. A. Niehaus, S. Suhai, F. Delia Sala, P. Lugli, M. Elstner, G. Seifert, and T. Frauenheim, Phys. Rev. B, 63 (2001) 085108. [171] T. M. Halasinski, J. L. Weisman, R. Ruiterkamp, T. J. Lee, F. Salama, and M. HeadGordon, J. Phys. Chem. A, 107 (2003) 3660. [172] M. Parac and S. Grimme, Chem. Phys., 292 (2003) 11. [173] J. R. Platt, J. Chem. Phys., 17 (1949) 484. [174] J. L. Weisman, T. J. Lee, and M. Head-Gordon, Spectrochim. Acta A, 57 (2001) 931. [175] J. Szczepanski, J. Banisauskas, M. Vala, S. Hirata, R. J. Bartlett, and M. Head-Gordon, J. Phys. Chem. A, 106 (2002) 63. [176] J. L. Weisman, T. J. Lee, F. Salama, and M. Head-Gordon, Astrophys. J., 587 (2003) 256. [177] J. Banisaukas, J. Szczepanski, J. Eyler, M. Vala, S. Hirata, M. Head-Gordon, J. Oomens, G. Meijer. and G. von Helden, J. Phys. Chem. A, 107 (2003) 782. [178] S. Grimme, I. Pischel, S. Laufenberg, and F. Vogtle, Chirality, 10 (1998) 147. [179] S. Toyota, T. Shimasaki, N. Tanifuji, and K. Wakamatsu, Tetr. Assym., 14 (2003) 1623. [180] Z. L. Cai and J. R. Reimers, J. Phys. Chem. A, 106 (2002) 8769. [181] C. Adamo and V. Barone, Chem. Phys. Lett., 330 (2000) 152. [182] J. Herbich, M. Kijak, A. Zielinska, and R. P. T. J. Waluk, J. Phys. Chem. A, 106 (2002) 2158. [183] M. Chachisvilis and A. H. Zewail, J. Phys. Chem. A, 103 (1999) 7408. [184] C. Dedonder-Lardeux, C. Jouvet, S. Perun, and A. L. Sobolewski, Phys. Chem. Chem. Phys., 5 (2003) 5118.
123 [185] C. W. Chang, Y. T. Kao, and E. W. G. Diau, Chem. Phys. Lett., 374 (2003) 110. [186] D. Delaere, M. T. Nguyen, and L. G. Vanquickenborne, Chem. Phys. Lett., 333 (2001) 103. [187] K. A. Nguyen, J. Kennel, and R. Pachter, J. Chem. Phys., 117 (2002) 7128. [188] D. Delaere, M. T. Nguyen, and L. G. Vanquickenborne, J. Phys. Chem. A, 107 (2003) 838. [189] W. P. O. Nina Sadlej-Sosnowska and A. Krowczynski, Chem. Phys., 294 (2003) 65. [190] J. Spanget-Larsen and E. W. Thulstrup, J. Mol. Struct. (THEOCHEM), 661 (2003) 603. [191] I. Deperasinska and J. Prochorow, Acta Phys. Pol. A, 104 (2003) 601. [192] R. Pou-Amerigo, P. M. Viruela, R. Viruela, M. Rubio, and E. Orti, Chem. Phys. Lett., 352 (2002) 491. [193] R. Andreu, J. Garin, and J. Orduna, Tetrahedr., 57 (2001) 7883. [194] J. Spanget-Larsen, Spectrochim. Acta A, 58 (2002) 2245. [195] A. Crespo, A. G. Turjanski, and D. A. Estrin, Chem. Phys. Lett., 365 (2002) 15. [196] Y. V. Il'iechev and J. D. Simon, J. Phys. Chem. B, 107 (2003) 7162. [197] J. Danielsson, J. Ulicny, and A. Laaksonen, J. Am. Chem. Soc, 123 (2001) 9817. [198] T. Schultz, J. Quenneville, B. Levine, A. Toniolo, T. J. Martinez, S. Lochbrunner, M. Schmitt, J. P. Shaffer, M. Z. Zgierski, and A. Stolow, J. Am. Chem. Soc, 125 (2003) 8098. [199] M. Kurihara, A. Hirooka, S. Kume, M. Sugimoto, and H. Nishihara, J. Am. Chem. Soc, 124 (2002) 8800. [200] S. P. Kwasniewski, M. S. Deleuze, and J. P. Francois, Int. J. Quant. Chem., 80 (2000) 672. [201] R. Improta, F. Santoro, C. Dietl, E. Papastathopoulos, and G. Gerber, Chem. Phys. Lett., 387 (2004) 509. [202] G. Orlova, J. D. Goddard, and L. Y. Brovko, J. Am. Chem. Soc, 125 (2003) 6962. [203] C. Neiss, P. Saalfrank, M. Parac, and S. Grimme, J. Phys. Chem. A, 107 (2003) 140. [204] R. Bauernschmitt, R. Ahlrichs, F. H. Hennrich, and M. M. Kappes, J. Am. Chem. Soc, 120 (1998) 5052. [205] F. Furche and R. Ahlrichs, J. Chem. Phys., 114 (2001) 10362. [206] X.-D. Li, W.-D. Cheng, D.-S. Wu, H. Zhang, Y.-J. Gong, and Y.-Z. Lan, Chem. Phys. Lett., 380 (2003) 480. [207] R.-H. Xie, G. W. Bryant, G. Sun, M. C. Nicklaus, D. Heringer, T. Frauenheim, M. R. Manaa, V. H. Smith, Jr., Y. Araki, and O. Ito, J. Chem. Phys., 120 (2004) 5133. [208] D.-S. Wu, W.-D. Cheng, H. Zhang, X.-D. Li, Y.-Z. Lan, D.-G. Chen, Y.-J. Gong, and Y.-C. Zhang, Phys. Rev. B, 68 (2003) 125402. [209] A. Rubio, J. A. Alonso, J. M. Lopez, and M. J. Stott, Physica B, 183 (1993) 247. [210] C. Yannouleas, E. N. Bogachek, and U. Landman, Phys. Rev. B, 53 (1996) 10225. [211] O. T. Ehrler, J. M. Weber, F. Furche, and M. M. Kappes, Phys. Rev. Lett., 91 (2003) 113006. [212] M. Gouterman, Mol. Spectr., 6 (1961) 138. [213] M. Gouterman, in D. H. Dolphin (ed.), Porphyrins, Academic Press, New York, vol. 3, 1978 pp. 1-165. [214] L. Edwards, D. H. Dolphin, M. Gouterman, and A. D. Adler, J. Mol. Spectr., 38 (1971) 16. [215] D. Sundholm, Phys. Chem. Chem. Phys., 2 (2000) 2275. [216] S. J. A. van Gisbergen, A. Rosa, G. Ricciardi, and E. J. Baerends, J. Chem. Phys., I l l (1999) 2499.
124 [217] E. J. Baerends, G. Ricciardi, A. Rosa, and S. J. A. van Gisbergen, Coord. Chem. Rev., 230 (2002) 5. [218] X. J. Liu, J. K. Feng, A. M. Ren, and X. Zhou, Chem. Phys. Lett., 373 (2003) 197. [219] A. B. J. Parusel and A. Ghosh, J. Phys. Chem. A, 104 (2000) 2504. [220] A. B. J. Parusel and S. Grimme, J. Porphyrins Phthalocyanines, 5 (2001) 225. [221] I. Infante and F. Lelj, Chem. Phys. Lett., 367 (2003) 308. [222] A. Gorski, E. Vogel, J. L. Sessler, and J. Waluk, J. Phys. Chem. A, 106 (2002) 8139. [223] M. Jaworska, G. Kazibut, and P. Lodowski, J. Phys. Chem. A, 107 (2003) 1339. [224] J. A. Shelnutt, X.-Z. Song, J.-G. Ma, S.-L. Jia, W. Jentzen, and C. J. Medforth, Chem. Soc. Rev., 27 (1998) 31. [225] D. M. Chen, X. Liu, T. J. He, and F. C. Liu, Chem. Phys., 289 (2003) 397. [226] A. Rosa, G. Ricciardi, E. J. Baerends, A. Romeo, and L. M. Scolar, J. Phys. Chem. A, 107 (2003) 11468. [227] A. B. J. Parusel, T. Wondimagegn, and A. Ghosh, J. Am. Chem. Soc, 122 (2000) 6371. [228] A. K. Wertsehing, A. S. Koch, and S. G. DiMagno, J. Am. Chem. Soc, 123 (2001) 3932. [229] H. Ryeng and A. Ghosh, J. Am. Chem. Soc, 124 (2002) 8099. [230] A. Kyrychenko, J. Andreasson, J. Maartensson, and B. Albinsson, J. Phys. Chem. B, 106 (2002) 12613. [231] K. A. Nguyen, P. N. Day, and R. Pachter, J. Phys. Chem. A, 103 (1999) 9378. [232] G. Ricciardi, A. Rosa, and E. J. Baerends, J. Phys. Chem. A, 105 (2001) 5242. [233] K. A. Nguyen and R. Pachter, J. Chem. Phys., 114 (2001) 10757. [234] K. A. Nguyen, P. N. Day, R. Pachter, S. Tretiak, V. Chernyak, and S. Mukamel, J. Phys. Chem. A, 106 (2002) 10285. [235] K. A. Nguyen and R. Pachter, J. Chem. Phys., 118 (2003) 5802. [236] D. Sundholm, Chem. Phys. Lett., 317 (2000) 392. [237] G. Ricciardi, A. Rosa, S. J. A. van Gisbergen, and E. J. Baerends, J. Phys. Chem. A, 104 (2000) 635. [238] A. Rosa, G. Ricciardi, E. J. Baerends, and S. J. A. van Gisbergen, J. Phys. Chem. A, 105 (2001) 3311. [239] I. H. Wasbotten, T. Wondimagegn, and A. Ghosh, J. Am. Chem. Soc, 124 (2002) 8104. [240] G. Ricciardi, A. Rosa, E. J. Baerends, and S. J. A. van Gisbergen, J. Am. Chem. Soc, 124 (2002) 12319. [241] J. E. Rogers, K. A. Nguyen, D. C. Hufnagle, D. G. McLean, W. Su, K. M. Gossett, A. R. Burke, S. A. Vinogradov, R. Pachter, and P. A. Fleitz, J. Phys. Chem. A, 107 (2003) 11331. [242] E. M. Bauer, M. P. Donzello, C. Ercolani, E. Masetti, S. Panero, G. Ricciardi, A. Rosa, A. Chiesi-Villa, and C. Rizzoli, Inorg. Chem., 42 (2003) 283. [243] M. Jaworska and P. Lodowski, J. Mol. Struct. (THEOCHEM), 631 (2003) 209. [244] A. Dreuw, B. D. Dunietz, and M. Head-Gordon, J. Am. Chem. Soc, 124 (2002) 12070. [245] B. D. Dunietz, A. Dreuw, and M. Head-Gordon, J. Phys. Chem. B, 107 (2003) 5623. [246] T. Andruniow, P. M. Kozlowski, and M. Z. Zgierski, J. Chem. Phys., 115 (2001) 7522. [247] T. A. Stich, A. J. Brooks, N. R. Buan, and T. C. Brunold, J. Am. Chem. Soc, 125 (2003) 5897. [248] A. Rosa, E. J. Baerends, S. J. A. van Gisbergen, E. van Lenthe, J. A. Groeneveld, and J. G. Snijders, J. Am. Chem. Soc, 121 (1999) 10356. [249] P. Boulet, H. Chermette, C. Daul, F. Gilardoni, F. Rogemond, J. Weber, and G. Zuber, J. Phys. Chem. A, 105 (2001) 885. [250] J. Full, L. Gonzalez, and C. Daniel, J. Phys. Chem. A, 105 (2001) 184.
125 [251] [252] [253] [254] [255] [256] [257] [258] [259] [260]
[261] [262] [263] [264] [265] [266] [267] [268] [269] [270] [271] [272] [273] [274] [275]
[276] [277] [278] [279] [280] [281]
E. Broclawik and T. Borowski, Chem. Phys. Lett., 339 (2001) 433. B. Dai, K. M. Deng, J. L. Yang, and Q. S. Zhu, J. Chem. Phys., 118 (2003) 9608. M. Z. Zgierski, J. Chem. Phys., 118 (2003) 4045. J. Cody, J. Dennisson, J. Gilmore, D. G. VanDerveer, M. M. Henary, A. Gabrielli, C. D. Sherrill, Y. Zhang, C.-P. Pan, C. Burda, and C. J. Fahrni, Inorg. Chem., 42 (2003) 4918. I. R. Farrell, F. Hartl, S. Zalis, T. Mahabiersing, and J. Vlcek, Antonin, Dalton Trans., (2000) 4323. J. E. Monat, J. H. Rodriguez, and J. K. McCusker, J. Phys. Chem. A, 106 (2002) 7399. J. F. Guillemoles, V. Barone, L. Joubert, and C. Adamo, J. Phys. Chem. A, 106 (2002) 11354. C. Makedonas, C. A. Mitsopoulou, F. J. Lahoz, and A. I. Balana, Inorg. Chem., 42 (2003) 8853. S. R. Stoyanov, J. M. Villegas, and D. P. Rillema, Inorg. Chem., 42 (2003) 7852. J. Dyer, W. J. Blau, C. G. Coates, C. M. C. dn J D Gavey, M. W. George, D. C. Grills, S. Hudson, J. M. Kelly, P. Matousek, J. J. McGarvey, J. McMaster, A. W. Parker, M. Towrie, and J. A. Weinstein, Photochem. Photobiol. Sci., 2 (2003) 542. S. I. Gorelsky and A. B. P. Lever, Int. J. Quant. Chem., 80 (2000) 636. P. Boulet, M. Buchs, H. Chermette, C. Daul, E. Furet, F. Gilardoni, F. Rogemond, C. W. Schlaepfer, and J. Weber, J. Phys. Chem. A, 105 (2001) 8999. T. P. M. Goumans, A. W. Ehlers, M. C. van Hemert, A. Rosa, E. J. Baerends, and K. Lammertsma, J. Am. Chem. Soc, 125 (2003) 3558. L. Torres, R. Gelabert, M. Moreno, and J. M. Lluch, Chem. Phys., 286 (2003) 149. V. Cavillot and B. Champagne, Chem. Phys. Lett., 354 (2002) 449. X. Wang, L. Chen, A. Endou, M. Kubo, and A. Miyamoto, J. Organomet. Chem., 678 (2003) 156. V. A. de la Pena O'Shea, M. Capel-Sanchez, G. Blanco-Brieva, J. M. Campos-Martin, and J. L. G. Fierro, Angew. Chem. Int. Ed., 42 (2003) 5851. J. van Slageren, A. Klein, and S. Zalis, Coord. Chem. Rev., 230 (2002) 193. P. Romaniello, M. C. Aragoni, M. Area, T. Cassano, C. Denotti, F. A. Devillanova, F. Esaia, F. Lelj, V. Lippolis, and R. Tommasi, J. Phys. Chem. A, 107 (2003) 9679. L. Ghizdavu, O. Lentzen, S. Schumm, A. Brodkorb, C. Moucheron, and A. Kirsch-De Mesmaeker, Inorg. Chem., 42 (2003) 1935. P. J. Hay, J. Phys. Chem. A, 106 (2002) 1634. W. Ekardt (ed.), Metal Clusters, Wiley Series in Theoretical Chemistry, Vol. X, Wiley, New York, 1999. V. Bonacic-Koutecky, P. Fantucci, and J. Koutecky, Chem. Rev., 91 (1991) 1035. G. Onida, L. Reining, R. W. Godby, R. D. Sole, and W. Andreani, Phys. Rev. Lett., 75 (1995) 818. W. A. de Heer, W. D. Knight, M. Y. Chou, and M. L. Cohen, in H. Ehrenreich, F. Seitz, and D. Turnbull (eds.), Solid State Physics, Academic Press, New York, vol. 40, 1987 p. 98. W. Ekardt, Phys. Rev. B, 31 (1985) 6360. W. Ekardt and Z. Penzar, Phys. Rev. B, 43 (1991) 1322. W. A. de Heer, Rev. Mod. Phys., 65 (1993) 611. M. Brack, Rev. Mod. Phys., 65 (1993) 677. N. Umezawa and S. Susumi, Materials Research Society Symposium Proceedings, 579 (2000) 75. S. J. A. van Gisbergen, J. M. Pacheco, and E. J. Baerends, Phys. Rev. A, 63 (2001)
126 063201. [282] M. B. Torres and L. C. Balbas, Int. J. Quant. Chem., 91 (2003) 263. [283] A. Rubio, J. A. Alonso, X. Blase, L. C. Balbas, and S. G. Louie, Phys. Rev. Lett., 77 (1996) 247. [284] F. Calvayrac, C. Kohl, P. G. Reinhard, E. Suraud, and C. Ullrich, Cond. Matt. Theor., 13 (1998) 83. [285] I. Vasiliev, S. Ogiit, and J. R. Chelikowsky, Phys. Rev. Lett., 82 (1999) 1919. [286] J. M. Pacheco and J. L. Martins, J. Chem. Phys., 106 (1997) 6039. [287] I. Vasiliev, S. Ogiit, and J. R. Chelikowski, Phys. Rev. B, 65 (2002) 115416. [288] J. M. Pacheco and W.-D. Schone, Phys. Rev. Lett., 79 (1997) 4986. [289] M. Moseler, H. Hakkinen, and U. Landman, Phys. Rev. Lett., 87 (2001) 053401. [290] A. G. Eguiluz and W. Ku, in A. Gonis, N. Kioussis, and M. Ciftan (eds.), Electron correlation and Materials Properties, Kluwer Academic/Plenum Press, New York, 1999 pp. 329-359. [291] M. D. Deshpande, D. G. Kanhere, I. Vasiliev, and R. M. Martin, Phys. Rev. B, 68 (2003) 035428. [292] F. Furche, R. Ahlrichs, P. Weis, C. Jacob, S. Gilb, T. Bierweiler, and M. M. Kappes, J. Chem. Phys., 117 (2002) 6982. [293] L. Serra and A. Rubio, Phys. Rev. Lett., 78 (1997) 1428. [294] J. Lerme, B. Palpant, B. Prevel, E. Cottancin, M. Pellarin, M. Treilleux, J. L. Vialle, A. Perez, and M. Broyer, Eur. Phys. J. D, 4 (1998) 95. [295] K. Yabana and G. F. Bertsch, Phys. Rev. A, 60 (1999) 3809. [296] D. Schoofi, S. Gilb, J. Kaller, M. M. Kappes, F. Furche, A. Kohn, K. May, and R. Ahlrichs, J. Chem. Phys., 113 (2000) 5361. [297] X. J. Wang, X. H. Wan, H. Zhou, S. Takami, M. Kubo, and A. Miyamoto, J. Mol. Struct. (THEOCHEM), 579 (2002) 221. [298] A. Schweizer, J. M. Weber, S. Gilb, H. Schneider, D. Schoofi, and M. M. Kappes, J. Chem. Phys., 119 (2003) 3699. [299] J. R. Chelikowski, L. Kronik, and I. Vasiliev, J. Phys. Cond. Matt., 15 (2003) R1517. [300] C. S. Garoufalis and A. D. Z. ans S Grimme, Phys. Rev. Lett., 87 (2001) 276402. [301] D. Sundholm, Nano Lett., 3 (2003) 847. [302] S. 6 . I Vasiliev and J. R. Chelikowski, Phys. Rev. Lett., 86 (2001) 1813. [303] L. X. Benedict, A. Puzder, A. J. Williamson, J. C. Grossman, G. Galli, J. E. Klepeis, J.-Y. Raty, and O. Pankratov, Phys. Rev. B, 68 (2003) 085310. [304] V. Kumar, T. M. Briere, and Y. Kawazoe, Phys. Rev. B, 68 (2003) 155412. [305] M. Yu, C. S. Jayanthi, D. A. Drabold, and S. Y. Wu, Phys. Rev. B, 68 (2003) 035404. [306] K. Eichkorn and R. Ahlrichs, Chem. Phys. Lett., 288 (1998) 235. [307] P. Deglmann, R. Ahlrichs, and K. Tsereteli, J. Chem. Phys., 116 (2003) 1585. [308] J. M. Matxain, A. Irigoras, J. E. Fowler, and J. M. Ugalde, Phys. Rev. A, 63 (2000) 013202. [309] J. M. Matxain, A. Iringoras, J. E. Fowler, and J. M. Ugalde, Phys. Rev. A, 64 (2001) 013201. [310] J. M. Matxain, J. M. Mercero, J. E. Fowler, and J. M. Ugalde, J. Am. Chem. Soc, 125 (2003) 9494.
[311] [312] [313] [314]
M. C. Troparevsky, L. Kronik, and J. R. Chelikowsky, J. Chem. Phys., 119 (2003) 2284. S. Hirata, M. Head-Gordon, and R. J. Bartlett, J. Chem. Phys., I l l (1999) 10774. M. Tobita, S. Hirata, and R. J. Bartlett, J. Chem. Phys., 114 (2001) 9130. G. R. Hutchison, M. A. Ratner, and T. J. Marks, J. Phys. Chem. A, 106 (2002) 10596.
127 [315] A. Pogantsch, G. Heimel, and E. Zojer, J. Chem. Phys., 117 (2002) 5921. [316] G. R. Hutchison, Y. J. Zhao, B. Delley, A. J. Freeman, M. A. Ratner, and I. J. Marks, Phys. Rev. B, 68 (2003) 035204. [317] C. P. Hsu, P. J. Walla, M. Head-Gordon, and G. R. Fleming, J. Phys. Chem. B, 105 (2001) 11016. [318] J. Catalan, J. Chem. Phys., 119 (2003) 1373. [319] J. Catalan and J. L. G. de Paz, J. Chem. Phys., 120 (2004) 1864. [320] O. Kwon and M. L. McKee, J. Phys. Chem. A, 104 (2000) 7106. [321] F. Delia Sala, H. H. Heinze, and A. Gorling, Chem. Phys. Lett., 339 (2001) 343. [322] N. Elmaci and E. Yurtsever, J. Phys. Chem. A, 106 (2002) 11981. [323] J. Casado, L. L. Miller, K. R. Mann, T. M. Pappenfus, Y. Kanemitsu, E. Orti, P. M. Viruela, R. Pou-Amerigo, V. Hernandez, and J. T. Lopez Navarette. J. Phys. Chem. B, 106 (2002) 3872. [324] J. Ma, S. Li, and Y. Jiang, Macromol., 35 (2002) 1109. [325] A. Tamulis, J. Tamuliene, and V. Tamulis, NATO Sci. Ser. II, 96 (2003) 1. [326] F. Delia Sala, M. F. Raganato, M. Anni, R. Cingolani, M. Weimer, A. Gorling, L. Favaretto, G. Barbarello, and GGigli, Synth. Met., 139 (2003) 897. [327] M. Yurtsever, G. Sonmez, and A. S. Sarac, Synth. Met., 135 (2003) 463. [328] Z. Zhu, Y. Wang, and Y. Lu, Macromol., 36 (2003) 9585. [329] A. B. J. Parusel, W. Rettig, and W. Sudholt, J. Phys. Chem. A, 106 (2002) 804. [330] A. B. J. Parusel, G. Kohler, and S. Grimme, DFT/SCI, 102 (1998) 6297. [331] C. Bulliard, M. Allan, G. Wirtz, E. Haselbach, K. A. Zachariasse, N. Detzer, and S. Grimme, J. Phys. Chem. A, 103 (1999) 7766. [332] A. B. J. Parusel, Phys. Chem. Chem. Phys., 2 (2000) 5545. [333] A. B. J. Parusel and G. Kohler, Int. J. Quant. Chem., 84 (2001) 149. [334] A. B. J. Parusel, Chem. Phys. Lett., 340 (2001) 531. [335] C. Jamorski Jodicke and H.-P. Liithi, J. Chem. Phys., 117 (2002) 4146. [336] C. Jamorski Jodicke and H.-P. Liithi, J. Chem. Phys., 117 (2002) 4157. [337] C. Jamorski J6dicke and H.-P. Liithi, J. Am. Chem. Soc, 125 (2003) 252. [338] C. Jamorski Jodicke and H.-P. Liithi, Chem. Phys. Lett., 368 (2003) 561. [339] C. Jamorski Jodicke and H.-P. Liithi, J. Chem. Phys., 119 (2003) 12852. [340] A. Kohn and C. Hattig, J. Am. Chem. Soc, 126 (2004) 7399. [341] I. Szydlowska, A. Kyrychenko, A. Gorski, J. Waluk, and J. Herbich, Photochem. Photobiol. Sci., 2 (2003) 187. [342] I. Szydlowska, A. Kyrychenko, J. Nowacki, and J. Herbich, Phys. Chem. Chem. Phys., 5 (2003) 1032. [343] E. Falkovskaia, V. G. Pivovarenko, and J. C. del Valle, Chem. Phys. Lett., 352 (2002) 415. [344] E. Falkovskaia, V. G. Pivovarenko, and J. C. Del Valle, J. Phys. Chem. A, 107 (2003) 3316. [345] C. J. Fahrni, M. M. Henary, and D. G. VanDerveer, J. Phys. Chem. A, 106 (2002) 7655. [346] V. Barone, A. Palma, and N. Sanna, Chem. Phys. Lett., 381 (2003) 451. [347] R. Casadesus, M. Moreno, and J. M. Lluch, Chem. Phys., 290 (2003) 319. [348] D. Sundholm, Chem. Phys. Lett., 302 (1999) 480. [349] D. Sundholm, Chem. Phys. Lett., 317 (2000) 545. [350] Y. Yamaguchi, J. Porph. Phthal., 6 (2002) 617. [351] Y. Yamaguchi, S. Yokoyama, and S. Mashiko, J. Chem. Phys., 116 (2002) 6541. [352] A. Dreuw, G. R. Fleming, and M. Head-Gordon, Phys. Chem. Chem. Phys., 5 (2003)
128 3247. [353] A. Dreuw, G. R. Fleming, and M. Head-Gordon, J. Phys. Chem. B, 107 (2003) 6500. [354] H. M. Vaswani, C. P. Hsu, M. Head-Gordon, and G. R. Fleming, J. Phys. Chem. B, 107 (2003) 7940.
[355] [356] [357] [358] [359] [360] [361] [362]
P. Kjellberg, Z. He, and T. Pullerits, J. Phys. Chem. B, 107 (2003) 13737. B. Mennucci, A. Toniolo, and J. Tomasi, J. Phys. Chem. A, 105 (2001) 4749. J. L. M K Shukla, J. Phys. Chem. A, 106 (2002) 11338. M. K. Shukla and J. Leszczynski, J. Comp. Chem., 25 (2004) 768. B. Durbeej and L. A. Eriksson, J. Am. Chem. Soc, 122 (2000) 10126. B. Durbeej and L. A. Eriksson, J. Photochem. Photobiol. A, 152 (2002) 95. A. Nakata, T. Baba, H. Takahashi, and H. Nakai, J. Comput. Chem., 25 (2003) 179. M. A. L. Marques, X. Lopez, D. Varsano, A. Castro, and A. Rubio, Phys. Rev. Lett., 90 (2003) 258101. [363] M. J. Thompson, D. Bashford, L. Noodleman, and E. D. Getzoff, J. Am. Chem. Soc, 125 (2003) 8186.
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 Elsevier B.V. All rights reserved
129
IV. Electronic and Vibronic Spectra of Molecular Systems: Models and Simulations based on Quantum Chemically Computed Molecular Parameters. F. Negri and G. Orlandi Dipartimento di Chimica 'G. Ciamician', Via F. Selmi, 2, 40126 Bologna, and INSTM, UdR Bologna, Italy 1. INTRODUCTION As the title of this chapter suggests, out of the huge field of molecular spectroscopy, we restrict here to a relatively narrow group of spectroscopic techniques, namely absorption, emission (fluorescence, phosphorescence) and resonance Raman scattering, and we will be concerned, specifically, with the vibronic structure associated to these spectra and the information that can be extracted from it. The interaction of radiation with molecules that occurs in all the spectroscopic measurements might be thought to pertain more closely to the field of photophysics rather than to the photochemistry of molecules. Thus, the subject of this chapter, namely the simulation of electronic spectra such as the absorption, emission or resonance Raman and, more specifically, of their associated vibronic structures, might appear only marginally related with the photochemistry of excited states. However, as we will try to make clear, electronic spectra may contain a wealth of information on the early dynamics of molecules in their excited states, hidden in the vibronic structure associated with the electronic transitions. In this sense, the analysis and interpretation of the vibronic structure associated with an electronic spectrum can reveal important details on the initial photodynamical driving forces in the selected excited state. This happens because, as it will be seen, very often vibronic intensities are proportional to the changes that occur on potential energy surfaces (PES) upon electronic excitation. Specifically, changes occur along selected nuclear motions and the corresponding vibrational normal modes are activated. These nuclear motions, often skeleton stretching modes, but also angular or torsional modes, are associated with the initial step in excited state dynamics. In this sense, the identification of active vibrations in electronic spectra may provide the bridge between spectroscopy and photochemistry, since the vibronic activity of one specific mode in an electronic spectrum may indicate that a similar nuclear motion characterizes the initial evolution of the excited molecule. Vibronic structures can be associated with electronic spectra (absorption, fluorescence, phosphorescence), but often the vibronic information is hidden under the broad band shape of these spectra. In these cases the most interesting information is lost under the diffuse bands. An alternative spectroscopic technique that has been shown to be extremely versatile and powerful to study the structure of molecules and their primary evolution in excited states is resonance Raman [1, 2]. In this case the wavelength employed to excite the Raman scattering falls within an electronic absorption band and it causes selective enhancement of the vibrations of the absorbing species. The selective enhancement is particularly useful since it allows to obtain the vibrational spectrum of a chromophore
130 embedded, for instance in a protein, without interference from the vibrations of the amino acid environment [1]. A second reason for discussing the vibronic activity associated with electronic transitions is that while most standard ab initio or semi-empirical codes calculate routinely excited states and electronic transition dipole moments or oscillator strengths, thus, they provide almost routinely parameters connected with pure electronic absorption or emission properties of molecules, the simulation of the associated vibronic structures is not a problem solvable with one standard equation or approach. The reason is that the vibronic structure can be originated by different mechanisms[3], and hence it is difficult to generalise the problem and provide a solution with a single universal approach. For this reason, one scope of this chapter is to provide an overview on the models generally employed for the simulation of the vibronic spectra of molecules and on the level of theory necessary to obtain quantum-chemically reliable molecular parameters required for the simulations. Notice that these two aspects, models and simulations, represent two separate steps that lead to the correct simulation of spectroscopic properties. The choice of the model is important since different degrees of approximation must be selected in accord with the dominant mechanism that originates the spectral intensities. For instance electronic transitions can be dipole allowed or dipole forbidden or spin forbidden. In Section 2 we will go through the different types of electronic transitions and we will see that they cannot be dealt with the same model. Conversely, various degrees of approximation can be retained or relaxed in view of the dominant mechanism that leads to the observed intensities. Once the model is chosen appropriately, molecular parameters have to be predicted with quantum-chemical methods, and introduced in the model to generate simulated intensities. The choice of the more appropriate level of theory for this purpose must take into account the fact that ground and excited states must be treated at comparable levels of theory and that accurate vibrational frequencies are required. The correct model implemented with unreliable molecular parameters can lead to discrepancies between observed and computed intensities. Similarly, discrepancies can occur if the model is too approximate, even if molecular parameters are computed with state of the art level of theory. In addition, it should be kept in mind that the definition of the 'more appropriate level of theory' in this case may be different from that required to study other molecular properties such as, for instance, photo-excited reaction paths. In the following we will try to make more clear all the concepts briefly outlined above, by presenting, first, a theoretical introduction on the origin of vibronic structures in the electronic spectra of molecules. We will introduce approaches of increasing complexity starting with the simplest situation of an allowed transition and moving to more complex situations involving symmetry forbidden transitions and spin forbidden transitions. In the second part of this contribution we will review some applications covering several classes of molecules ranging from polyenes to fullerenes and strictly related to the models presented in the first part of this chapter. 2. MODELS AND COMPUTATIONAL DETAILS In this section we review the fundamental concepts concerning the interaction of radiation with molecules which leads to absorption, emission spectra or resonance Raman scattering. The starting point is the concept of molecular vibronic state, namely a molecular level corresponding to a given electronic and vibrational excitation, indicated, in the following as for the initial state and f) for the final state.
131
Excited state Final vibronic state | j - \
Ground state nuclear coordinate Fig.l. Schematic representation of a vibronic transition. The Born Oppenheimer approximation enables the factorization of the wavefunctions representing the vibronic states, as simple products of electronic and vibrational wavefunctions: i) = | g, m) = g)| m)
and similarly
| f) = | e, n) = | e)| n)
where |g) and |e) indicate pure electronic wavefunctions and m) and n) indicate pure vibrational wavefunctions. If a photon of energy E = Ef — E( is absorbed, the molecule will be subject to a vibronic transition and it will be excited to the final vibronic state f) (see Fig. 1). The probability of such a transition is related to the matrix element of the electrical transition dipole moment between the two states, Mj f : electrons
Miif=e(i| £ rj |f)
(1)
i
where r; represents the coordinate of the \th electron e. The vectorial quantity M if can be expressed through its three Cartesian components (M*f ,M*f ,M*f ) and it can be interpreted as a measure of the charge migration during the transition that brings the molecule to the final state f) from the initial state i). Beside the transition dipole moment, a strictly related quantity is the oscillator strength / often employed to quantify the intensity of a transition./is connected to the transition dipole moment by the following relation
132
'
8n2mC
v, f M?, f
(2)
2.1. Absorption and emission: the Franck-Condon principle for allowed electronic transitions With the basic definitions given above we can now consider in more detail the problem of calculating the vibronic transition dipole moments corresponding to the vibronic transitions in an absorption or emission spectrum. According to Eq. (2) the intensity in absorption or emission is related to the square of the transition dipole moment. In Section 2.3 we will see that intensities in Raman scattering are also related to these fundamental integrals, the transition dipole moments. The vibronic transition dipole moment M _>c n is defined as
Mg.,m
(3)
= {m \(g | e 5 > , | e)\ n) = (m \Mg,e (Q) n)
where \x, c (Q) represents the electronic transition dipole moment
MB,c (Q) = (gI©" & Ie) = (g IA| e)
(4)
and it depends on the nuclear variable Q owing to the fact that electronic wavefunctions depend, parametrically, from nuclear coordinates |g) = |g(r,Q)>;
|e) = |e(r,Q)>
Eq. (3) shows that, owing to the Q-dependence of the electronic wavefunctions, the correct procedure to calculate the full transition dipole moment would be to determine the Qdependence of the electronic transition dipole moment Hge(Q) and then to integrate over nuclear coordinates. The problem can be simplified by adopting the Condon approximation, namely by assuming that the electronic transition dipole moment is constant and equal to the value computed at a reference geometry, usually the equilibrium geometry Qo of the ground electronic state of the molecule under investigation:
Under the above condition Eq. (3) simplifies to: ,c(Q^n) = ^ c ( m | n ) = ^ c S m , n
(6)
where Sm,n represents the overlap integral between vibrational wavefunctions of the initial and final states. Thus, the intensity of a vibronic transition is given by
133
g.m—>e,n
L^g,e J
(7)
m,n
^ n is known as the Franck-Condon (FC) factor of the transition.
Excited state
yy=3 Vertical electronic transition
v'=0
Ground state v"=l v"=0
n
O
Vg
-
nuclear coordinate
Energy
Fig.2. Schematic representation of the Franck-Condon principle (left) and of the resulting absorption spectrum (right).
b) Fig.3. Two in-plane vibrational normal coordinates of 1,3,5,-frvms-hexatriene. a) a totally symmetric mode and b) a non-totally symmetric mode. Eq. (7) clearly indicates that the intensity in an absorption or emission spectrum will be modulated by the FC factor and that the crucial point in the simulation of the vibronic
134 structure is the evaluation of the Sm,n overlap integrals. This concept, known also as the Franck-Condon principle, is schematically represented in Fig. 2. The difficulties in the calculation of overlap integrals are due to the fact that potential energy curves generally deviate from harmonicity, but this fact can often be neglected. The major difficulty arises since the integral is multidimensional. Indeed a molecule with N atoms will have 3N-6 vibrational coordinates, thus the vibrational wavefunction will be factorized in terms of the 3N-6 normal coordinates
)=n K 3N-6
m
where the product is extended to all the normal vibrational coordinates. The vibronic transition dipole moment under the validity of the Franck-Condon approximation is thus given by M g , m ^ c , n = H°,cSm,n = U L k
k
and the intensity becomes
J n<mk |nk)2 = k J f l k . n l 2
(10)
Eq. (10) shows that for a polyatomic molecule, the intensity of the vibronic transition g)| m) —> e)| n) is related to the product of the FC factors of each normal vibration belonging to the molecule. From a computational point of view, the exact evaluation of the full FC factor can be very difficult and time consuming, however several simplifications, briefly described in the following, can be introduced. First of all, the use of symmetry can reduce the dimensionality of the problem. Vibrational normal coordinates can be classified according to the irreducible representations of the symmetry point group to which the molecule belongs. For instance the vibrational normal coordinates of benzene, which belongs to the D^ symmetry point group, can be classified into the following irreducible representations: Aig, A2g, B2g, Eig, ..., E2U, etc... To simplify the calculation in Eq. (10) an easier classification is, however, sufficient, namely the separation between totally symmetric (TS) vibrations and non-totally symmetric (NTS) vibrations. TS vibrations can be easily distinguished from NTS vibrations owing to the fact that the first do not break the symmetry of the molecules while the second do break it. To further clarify this point, in Fig. 3 we present as an example, a typical TS and NTS vibration of 1,3,5-frans-hexatriene.
135
Excited state
NTS coordinate
Fig.4. (left): profile of the ground and excited state PES along a typical TS normal coordinate; (right) profile of the ground and excited state PES along a typical NTS normal coordinate. An additional way to distinguish between TS and NTS vibrations is to consider the potential energy surface (PES) profile of the ground and excited state along the selected normal coordinate. If the normal coordinate is TS, the PES profile will show, generally, the minimum of the ground state Qg shifted with respect to the minimum of the excited state Qe (see Fig. 4, left). In other words, the two PESs are displaced along the TS coordinate. Conversely, the two PESs are not displaced along a NTS coordinate. Indeed a displacement of the two minima along a NTS coordinate can occur only if the excited state belongs to a symmetry point group different (lower) from that of the ground state. If the symmetry in the two electronic states is the same, a displacement cannot occur (see Fig. 4, right). According to the above classification, the multidimensional FC factor in Eq. (10) can be factorized in two terms:
A further simplification arises if the harmonic approximation is adopted. The harmonic approximation is generally acceptable for medium-high frequency modes, while it can be unacceptable for very low frequency modes whose PES profiles are highly anharmonic even for small displacements away from the equilibrium position. Thus, if the vibronic activity is dominated by medium-high frequency vibrations, the harmonic approximation can be adopted with confidence, otherwise the simulated results must be taken with care or more appropriate procedures to evaluate overlap integrals must be considered [4]. Under the harmonic approximation, and assuming equal vibrational frequencies (cof = co^) in both the ground and excited states, simple analytical expressions can be derived for the overlap integrals involving wavefunctions of TS or NTS vibrations[2, 5]. Notice that in typical conditions molecules are in their ground electronic state and ground vibrational state, thus, the FC factors to be calculated correspond to m) = 10}:
136 6-
-
NTS .
-
TS -
-
kJ =nkJnkJ
Each term in the product above, for instance the sequence [s^0J ; [s^, J ;[s,^ 2 J ...;[s^n represents a progression in the given normal mode k, and is described by a simple Poisson distribution:
where
and Bk is the dimensionless displacement parameter defined as [6]
rkE
(15)
here Q^>g represents the projection of the geometry change occurring upon g) —> e) electronic excitation, expressed in Cartesian coordinates, on the kth vibrational coordinate Qk, namely:
k
(16)
In Eq.(16) xgje is the 3N-dimensional vector of the Cartesian coordinates corresponding to the equilibrium geometry of the e and g electronic states. M is the diagonal 3NX3N matrix of the atomic masses and Lk is the 3N-dimensional vector of the normal coordinate Qk in terms of mass weighted Cartesian coordinates. It is worth noting at this point, that the extension of the vibronic progression of a given vibrational mode is governed by the magnitude of the displacement parameter, as it can be appreciated by inspecting Fig. 5, where three progressions corresponding to different magnitudes of the Bk parameter are compared. From the above definition of the dimensionless displacement parameter Bk and considering the characteristic PES profile behaviour typical for TS and NTS modes, it is easy to conclude that, being yk =0 for the latter, Eq. (13) reduces to =0 and
n>l
(17)
137
1000
2000 3000 Energy / cm-1
4000
5000
Fig. 5. Extension of the vibronic progression of a single mode of frequency co=1000 cm"1, as a function of the magnitude of the displacement parameter Bk. From bottom to top, Bk=1.0, 1.5, 2.0.
(18) It should now be clear why it is convenient to separate TS from NTS modes, since the FC factor of Eq.(12) simplifies to (19)
namely, Eq.(19) shows that the vibronic structure in the electronic spectra of polyatomic molecules, under the validity of the FC approximation, is governed predominantly by TS modes, through their displacement parameters Bi and hence their j \ parameters.
138
1000
2000
3000
4000
5000
Energy / cm-1
Fig.6. Vibronic structure due to the superposition of two active modes with the following parameters: co,=1000 cm'1, B,= 1.0 and co2=100 cm"1, B2=2.0.
Emission
Absorption
r" a "S
Energy Fig. 7. Schematic representation of the mirror symmetry between absorption and emission spectra.
139 Although in general only TS are active according to Eq. (19), it should be kept in mind that the number of TS modes can be quite large for polyatomic molecules. This can be reflected in quite complicated vibronic structures, due to the superposition of the progressions of modes characterized by different frequencies and different activities (displacement parameters). In Fig. 6 we show a simple example, namely the distribution of vibronic activities for a twomode system with considerably different vibrational frequencies and displacement parameters. If vibrational frequencies are substantially different in the ground and excited states, the constraint of equal vibrational frequencies must be relaxed and the conclusions above must be slightly modified since features corresponding to NTS modes can now appear in the spectra. Indeed, owing to the fact that {DgTS * roeNTS, FC factors such as [s^ s ] 2 and [soNf]2 can be significantly different from zero and less approximated relations for the FC factors must be employed. For instance, the more exact expression of [S^ s J reads:
(20) (D g +(D C
Notice that, if the approximation of equal frequencies in the ground and excited states is adopted, the equations so far discussed also imply a vibronic structure in emission specular to the vibronic structure observed in absorption (see Fig. 7). The mirror symmetry between absorption and emission spectra is indeed observed when frequency variations upon excitations are minor and when rotation of vibrational coordinates (Duschinsky effect [7]) can be neglected. Indeed, the change in the orbital nature of the electronic state upon excitation is usually reflected in a remarkable change of the PES. This change is usually associated with some degree of normal coordinate rotation. Normal coordinates belonging to the same irreducible representation can mix according to the following relation
where the Rkj coefficients are the elements of the Duschinsky rotation matrix. The main consequence of Duschinsky rotation is the redistribution of vibronic intensities between the absorption and emission spectra, namely the loss of mirror symmetry. Nevertheless, the Duschinsky effect can be reasonably taken into account in a simple way, namely without the direct inclusion of Duschinsky rotation matrices, by calculating the Bk displacement parameters (Eq. 15) using vibrational normal coordinates and frequencies of the ground state in the case of emission spectra and vibrational normal coordinates and frequencies of the excited state for the simulation of absorption spectra [8]. 2.2. Beyond the Condon approximation: forbidden electronic transitions In the previous section we discussed the vibronic structure that can be predicted according to the Condon approximation and we concluded that it will be governed by progressions of TS modes and, only marginally, by features due to NTS modes, if remarkable frequency changes occur upon electronic excitation.
140 The Condon approximation relies on Eq. (5) namely on the assumption that the electronic transition dipole moment can be considered independent of the nuclear coordinates. This is not so in general, and thus the usual procedure is to expand the electronic transition dipole moment as a Taylor series in the nuclear coordinates about the equilibrium nuclear configuration Qo [3]. (22)
The above expansion of the electronic transition dipole moment parallels the Herzberg-Teller (HT) expansion of adiabatic electronic wavefunctions in terms of crude adiabatic (CA) wavefunctions, i.e. electronic wavefunctions defined at a specific nuclear configuration Qo, so that the Q dependence is removed [3]. The HT expansion reads |r(q,Q)} = |r(q,Q0)) + Xa s , r (Q)|s(q,Q 0 ))
(23)
where r(q,Q0)) and |s(q,Q0)) are CA wavefunctions. The Q dependence in the above expansion is also known as the breakdown of the Condon approximation. The CA electronic wavefunctions satisfy the so called static Schrodinger equation, namely H cl (q,Q o Mq.Qo))= Er(Qo)|r(q,Qo))
(24)
The static electronic Hamiltonian Hei(q,Qo) in Eq. (24) is related to the complete, dynamic electronic Hamiltonian Hd(q,Q) through Eq. (25) H c ] (q,Q)=T(q)+U(q,Q)=T(q)+u(q,Q 0 )+AU(q,Q)=H c l (q,Q 0 )+AU(q,Q)
(25)
where T is the kinetic energy term and U is the potential energy term. Thus, the coefficients in Eq. (23), which depend on the nuclear coordinates, are given by the perturbation relation
a
-(Q)=
(S(q,Q0)|AU(q,QHq,Q,,)) Er(Q.)-E.(Q.)
+
-
The potential AU(q,Q) in Eq.(26) can be expanded as a Taylor series about Qo to get
(27)
Substitution of Eq.(27) into Eq.(26) leads to the following expression for the coupling constants as,r:
141
d
*Q)l
-(,,QJ (28)
E,(Q.)-E,(Q.)
or in a more compact form
' AE
(29)
where K k r is the adiabatic vibronic interaction between the s) and |r) electronic states, perturbation mediated through the Mi vibrational coordinate. Being a perturbation expansion, the HT approach can only be used if the interacting states in Eq. (28) are well separated in energy, namely for weak coupling regimes. If the energy separation is small, other approaches must be adopted, for instance perturbation approaches taking into account the vibrational contribution to the energy gaps which separate vibronically coupled states [9, 10] or exact diagonalization of the Hamiltonian matrix in the vibronic basis [11, 12]. We will not go into the details of these additional approaches, since they go beyond the scope of this chapter. Notice that Eq.(28) or (29) show the same Qk dependence of the electronic transition dipole moment in Eq.(22). Coming back to Eq.(22) we notice that the Condon approximation discussed in the previous section requires the neglect of all but the first term. In this limit we have seen that the transition intensities are proportional to the FC factors. In the breakdown of the Condon approximation, more generally called the HT vibronic coupling, the higher order terms in Eq.(22) are introduced. The most dramatic effects due to HT vibronic coupling are observed, generally, for dipole forbidden transitions. In this case the first term in Eq.(22) vanishes and the Condon approximation would imply absence of intensities in correspondence of a similar electronic transition. This conclusion is, however, in marked contrast with the experimental evidence. The simpler example is represented by the absorption spectrum of benzene, shown in Fig. 8, where, disregarding the lowest energy spin and symmetry forbidden S0(Alg ) —> T^B^J transition, appreciable intensities are observed not only for the symmetry allowed S0(A]g j—> S 3 (E 1 U ) transition but also for the S0(A,,,)—> S,(B 2 U ) transition whose transition dipole moment is zero by symmetry. The discrepancy between the experimental evidence and the prediction provided by the Condon approximation implies that the model is too approximated in this case, and one has to move to the more accurate HT vibronic coupling approach. In the HT vibronic coupling approach, the first term in Eq.(22) is zero, but higher terms are retained and they account for the intensities observed for symmetry forbidden transitions. To clarify this point we can consider an even simpler example, namely a | g) —> e) symmetry forbidden electronic transition of a molecule with a single NTS mode such that —
# 0 . Following Eq. (22), the intensity I gm ^ cn of the g)|m) —> |e)|n) vibronic
transition in this NTS mode is given by
142
150
X (r*m)
Fig.8. Low resolution spectrum of benzene from ref.[13].
=
m NTS
'NTS
.0
r-g.e n "N TS MTS
)+ m (30)
where the vibrational overlap multiplied by (a°c has been neglected in the last line of Eq.(30) because the electronic transition is assumed to be dipole forbidden. It is interesting to employ Eq. (30) to obtain the expressions for two specific vibronic transitions, namely the | g)| 0} -» | e)| 0) and the | g)| 0} -> | e)| l): =0
(31) (32)
The first of the two (Eq.(31)) is zero owing to the properties of the vibrational wavefunctions. This implies that the g}| 0} —> e)| 0} transition, namely the origin band, of a forbidden transition will never appear in the absorption spectrum. Eq.(32), in contrast, shows that the g)| 0) —> e)| l) band can appear in the region of a forbidden transition, owing to the properties of the vibrational wavefunctions which ensure that the integral <0|QNTS 1) is different from zero.
143 (0NTS.0TS)"-(1NTS>4 T S)'
(O N T S ,O T S )"^(l N T S ,l T S )' (O NTS ,O TS )"^(1 NTS ,O TS )' false origin
(0 N T S ,0 T S )"
(0NTS,0TS)"^(0NTS,0TS)' true origin, not observed
Energy Fig.9. Schematic representation of the vbronic structure in a symmetry forbidden transition. The example shows the structure expected for a model system with one NTS mode and one TS mode. The vibrational frequency of the NTS mode is ca. double that of the TS mode. The true origin of the electronic spectrum is missing (dotted line). The first band observed in the spectrum is the false orgin (thick solid line) due to a NTS vibration (ONTS—>TNTS)- The remaining vibronic structure is due to the progression of the TS mode, built on the false origin.
HT induced transition (false origin)
vibronic coupling via mode e,n
dipole forbidden transition (true origin)
S0(Ag)
Fig. 10. Schematic representation of the typical HT mechanism of intensity borrowing
144 This simple example states two important and new characteristics (compared with the previous description of dipole-allowed transitions) typical for forbidden electronic transitions: 1) the orgin band is always missing and 2) the first band that can appear in the spectrum is the 0—>\ transition in a NTS mode. This band, being the first (lowest energy) band in the absorption spectrum is also called the false origin of the spectrum. A schematic representation of the typical vibronic structure of a forbidden transition is depicted in Fig. 9. Bands originated through the above mechanism are commonly said to be induced vibronically and the theory that accounts quantitatively for their appearance is the HT theory of vibronic coupling anticipated in Eq.(23-29). If we rewrite the electronic transition dipole moment using Eq.(23) to expand, for simplicity, only the excited electronic state and Eq.(29) to express the mixing coefficients, we ends up with the following expression K
(33)
which clarifies the origin of the induced intensity. Indeed, Eq.(33) says that if the electronic transition is dipole forbidden, that is ^g(q,Q0|(i|e(q,Q0)^ = 0, then intensity can be induced by borrowing or stealing a contribution from the g(q,Q0))—» s(q,Q(l)) dipole allowed transition. This mechanism, known as the HT intensity borrowing, is schematically shown in Fig. 10 where the relevant states involved in the case of benzene are considered. In this case, the HT intensity borrowing mechanism implies that the appearance of intensity in the S 0 ( A , E ) - > S,(B 2 U ) and S 0 (A lg )-> S 2 (B kl ) regions of the absorption spectrum of benzene can be accounted for by considering the coupling between dipole forbidden states (B2U, B[u) with the dipole allowed Eiu state, mediated by vibrational coordinates of appropriate symmetry[14, 15]. The case of the S0(AlgJ—> S,(B 2 U ) transition is represented in Fig. 10. 2.3. Resonance Raman spectra So far we have overviewed the two basic mechanisms that account for vibronic intensities in absorption (or emission) spectra associated with dipole allowed or dipole forbidden transitions. We now move to a slightly more complicated scheme, the equations governing Raman and resonance Raman intensities for which the same models described above, namely the FC approximation or the HT intensity borrowing, can be applied. For a Raman transition between two states |i) and f), the intensity is proportional to the square of the transition polarizability a£,° [2]. In the Born-Oppenheimer approximation, the vibronic states i), | f) and the intermediate state p) are formed by products of pure vibrational and pure electronic states. If we assume that the system is initially and finally in the ground electronic state g) we may write i) = | g}| n), | f) = g)| m) and | p) = e)| v), and the transition polarizability can be written as:
145
(Hk=]p v)(v kj" co cv -o) gn
-CD ascr
P IM 8 .CJ° v)(v k, g ] |n)
+ i r cv
c
+ 03 asCT +
ircv
(34)
where iFcv is a damping factor associated with the g) —> e) electronic transition and k e f ' " *s t n e P u r e electronic transition moment. The sum in Eq. (34) extends to all electronic |e) and vibrational |v) states. Notice that v) indicates a multidimensional vibrational state as discussed in the previous sections. h(ogn, ^cogm and 7z
(35)
v +...
where the sum runs over the set of k normal coordinates Qk and
In
the HT perturbation description of vibronic coupling discussed in the previous section, the derivative of the transition dipole-moment can be expressed in terms of the adiabatic vibronic interactions KJJS (see Eq. (28-29)) which can mix the state e) with other states s) and the analogous K.^ terms corresponding to couplings of the ground electronic state with other excited states s). Substitution of Eq.(35) in Eq. (34) leads to an expression than can be grouped in four terms, that are labeled the A, B, C and D [2]. The latter three terms contain expansions that include the Kkes and KJL vibronic coupling parameters, while the A term [16] is given simply by: a f p f(A-term) = (m v)(v n) G>™ — C O ™
-CO,.
m v•>(v|n) + ir\.
K) m ,
(36)
-CO™, +C0,.
According to the discussion of Section 2.1, it is clear that the A-term includes only the Franck-Condon effects and ignores vibronic coupling effects. The latter are taken into account in the remaining B, C and D terms. A thorough discussion of vibronic coupling effects in resonance Raman spectroscopy can be found in ref. [6]. Here we restrict the discussion to the A-term since, as it will be shown in the applications discussed in Section 3, this term accounts for the most prominent features in the resonance Raman spectra of several classes of molecules, whenever the excitation wavelength is in resonance with a dipole allowed transition.
146 Off-resonance Raman scattering
Resonance Raman scattering Virtual state
Virtual state
Fig. 11. Off-resonance and resonance regimes of Raman scattering. We have anticipated the concept of resonance Raman, but indeed Eq. (36), if we neglect vibronic coupling effects, is completely general, and can be used to calculate either resonance and off-resonance Raman intensities. The difference among various regimes of Raman excitations is summarized in Fig. 11. Similarly to the intensity in absorption or emission spectra, Eq. (36) indicates that, since for a given electronic transition the pure electronic transition dipole moments [u/!cJ are constant, the modulation of vibronic intensities occurs because of the vibrational overlaps. The multi-dimensional vibrational overlap integrals in Eq. (36) can be expanded in terms of the contributions from NTS and TS vibrations as seen in the previous sections. NTS modes can contribute to vibrational overlaps (other than the (0 0)) in case of large frequency variations upon excitations, but generally these effects are negligible. Thus, to a good a approximation, according to the Franck-Condon mechanism only TS vibrational modes can contribute to Raman bands. In other words, we can restrict the calculation of vibrational = 0. integrals to TS modes since for the fth NTS vibration | n,) = v, = The latter TS vibrational overlap integrals are readily evaluated by further adopting the harmonic approximation, and by assuming identical vibrational frequencies and normal coordinates (no Dushinsky rotation [7]) in the ground and excited electronic states. Analytical or recurrence formulas [2, 5, 17] can be easily employed to this end. Vibrational overlap integrals expressed in terms of the dimensionless displacement parameters Bk can be calculated for each Mi TS vibration according to Eq. (15). In summary, for a fundamental band of a TS mode k in the Raman spectrum, the product of multi-dimensional vibrational overlap integrals in Eq. (36) reads:
147
vk
V
n
t
2vkE
k-.)_B(
°p,
P
0
TS e
2
J
2v
(37)
P
n
Notice that Eq. (36) can be drastically simplified if only one excited electronic state e) contributes to the Raman scattering and if we assume that the sum running over the vibrational states is dominated by v=0. These conditions are satisfied, for instance, if we are in perfect resonance with a 0-0 dipole allowed electronic transition, well separated in energy from other dipole-allowed electronic transitions. In this case the contribution from the A term simplifies to:
(38)
where the second term in Eq (36) has been neglected because of the resonance condition. The above relation shows that to evaluate resonance Raman intensities the computation of the Bk parameters is essential since (39) and the resonance Raman intensity of the kth fundamental vibration is (40)
proportional to the square of the displacement parameter. The evaluation of the Bk parameters, in turn, requires the determination of equilibrium structures of the ground state and of the excited electronic state in resonance with the excitation wavelength. The simplifications discussed above will be useful to rationalize in a simple way the computed and observed resonance Raman intensities discussed in the application section. 2.4. Spin forbidden transitions: phosphorescence spectra In the previous Sections we have considered spectroscopic conditions of increasing complexity, by moving from dipole allowed transitions, generally dominated by the FC activity of TS modes, to dipole forbidden transitions for which the role of vibronic coupling mediated by NTS modes becomes dominant. The last topic we consider, in this first part covering the minimal models required to account for vibronic intensities in electronic spectra, is an electronic transition which is both spin and symmetry forbidden. The different spin of the initial and final states is an additional source of complexity.
148 Spin-orbit coupling H s o
T, spin forbidden transition spin allowed transition
Fig. 12. The spin-orbit perturbation induces weak So —> T, intensity. Our purpose is to review the relations required to predict vibronic intensities in this case, which typically corresponds to the simulation of phosphorescence spectral structures or So —> T, absorption spectra. The calculation of the So -O- T, transition dipole moment requires the inclusion of the spin-orbit (SO) perturbation, which is responsible for the mixing among singlet and triplet states (see Fig. 12). The spin-orbit induced transition dipole moment can be computed according to first order perturbation theory H Ey — Eg
HS
o
,T S
Es - E T
(41)
In the expression above r and s run over the full space of singlet and triplet states of the molecular system. Notice that Eq. (41) provides an estimate of the pure electronic transition dipole moment. If the electronic transition dipole moment is different from zero, that is, if the So <-» T, transition is only spin and not symmetry forbidden, then Eq. (41) provides the equivalent of the constant factor (i°e in Eq. (7). In other words, the vibronic structure of such a transition will be governed by the FC activity of TS modes. However, for several conjugated molecules, (simple and common examples are benzene and Cgo fullerene) the So <-> T, transition is both spin and symmetry forbidden. This implies that the vibronic structure of the phosphorescence spectra is dominated by HT induced false origins on top of which progressions of TS modes can be built. Thus, the estimate of the 0 —»1 vibronically induced intensity in the phosphorescence spectrum (0 and 1 being the initial and final vibrational quantum numbers of the 4th NTS
149 mode) requires additional evaluation of the numerical derivatives of the |j, s ^_T transition dipole moment with respect to each normal coordinate Qk that may be active on the basis of symmetry selection rules (42)
3Qk
The vibronic transition moment in Eq. (42) is readily obtained by integration over the i
vibrational integral, which, in the harmonic approximation is equal to
l2cok Conversely, the Franck-Condon (FC) activity of the fth TS mode in the spectra is governed by the displacement parameter B; defined in Eq. (15). 2.5. Simulations based on quantum-chemically computed molecular parameters In the previous Sections we have discussed the models required to account for the major features of the vibronic structure associated with electronic spectra. Here we wish to consider more practical aspects concerning the simulation of vibronic structures: more specifically, the calculation with quantum-chemical methods, of the molecular parameters required by the models overviewed above. In the simplest approach discussed, namely under the Condon approximation, the molecular parameters required to simulate the vibronic structure associated with an absorption or emission transition, or in RR spectra, are the displacement parameters of Eq. (15). These, in turn, require the evaluation of equilibrium structures of the two electronic states involved in the electronic excitation and vibrational frequencies of either the ground or the excited state. These parameters can nowadays be obtained at ab-initio level also for molecules of mediumlarge size. Several studies on conjugated organic molecules [18-20] have shown that the CASSCF level with standard 3-21G, 6-31G or 6-31G* basis set provides reliable estimates for both geometries and vibrational frequencies. When CASSCF calculations cannot be performed because of the size of the chromophore, the HF level for the ground state and the CIS level for the excited states provide very reasonable estimates of the required molecular parameters. Obviously enough, CIS level is acceptable only for excited states whose nature does not require substantial corrections from multiple excitations. It is clear that both CASSCF or CIS levels do not provide reliable estimates of excitation energies, but this aspect is irrelevant since vibronic intensities depend only on the quality of the description of geometry changes upon excitation and not on the accuracy of the description of the energy separation between excited and ground state potential energy surfaces. Beside geometry changes, the quality of vibrational normal coordinates is of utmost importance. The reason is that minor unbalanced descriptions in the internal coordinate composition of vibrational normal coordinates can lead to completely unsatisfactory distributions of intensity in a vibronic spectrum. This fact should be taken into account very seriously when choosing the level of theory. In summary, the generation of the best computed molecular parameters requires two separate decisions: 1) the choice of most appropriate structural model to mimic the molecular system investigated. Usually, large chromophores cannot be computed in their complete form, because of their large dimension, and an appropriate reduced system must be identified. For instance, in the case of a conjugated chromophore with side methyl groups, undergoing
150 isomerization upon excitation, the best reduced model could be represented just by the conjugated chain, if the major features of photoexcited reaction path and deactivation are studied. However, if the model must be used to simulate vibronic spectra and to compare them with experiment, the inclusion of at least some of the side groups (methyl) will be necessary, since the normal vibrations of these groups are generally coupled with skeleton modes, and their neglect can influence dramatically the vibronic intensity distributions. 2) the second decision concerns the choice of the most appropriate level of theory. We have just mentioned that CASSCF provides very reliable molecular parameters, and we will see few examples in the next section. However, CASSCF level of theory may lead to unsatisfactory vibronic intensities, if the size of the molecular system imposes restrictions over the CAS space that can imply an unbalanced description of selected molecular regions. These molecular regions may be irrelevant for the overall description of the excited state deactivation pathway, but they may influence dramatically the quality of vibrational coordinates and hence of the simulated vibronic intensities. For very large molecules, such as fullerenes or the large graphenic structures described in the next section, ab initio methods are often not practicable, especially to obtain excited state structures. In this case semiempirical methods such as the QCFF/PI method [21], with its more recent implementations [22, 23] has been shown to be able to catch the most relevant excited state changes occurring in rat* states of highly conjugated organic compounds [2427]. Moving to the more complex models requiring vibronic interaction calculations or spin-orbit interaction calculations, the major problem in these cases is that interactions are required among large numbers of excited states. Indeed, Eq. (23) required to correct the description of electronic states, involves the mixing with all the diabatic excited states through the mixing coefficients ars(Q) whose magnitude, in turn, depends on the strength of the vibronic interaction between the states s and r. Generally it is not known, a priori, how many s states are required to correct the diabatic r state. Similarly, for spin forbidden transitions, Eq. (41) implies that both the initial and final electronic states involved in the transition must be corrected by mixing them with states of different multiplicity. Singlet states will have to be mixed with triplet states and triplet states with singlet states. The magnitude of the mixing coefficient is determined by the magnitude of the spin-orbit interaction, and also in this case the larger the number of interaction included, the more accurate will be the resulting prediction. Thus, owing to the requirement of a large number of excited interacting states, to model vibronically forbidden transitions or spin forbidden transitions one has to resort to semiempirical calculations. Electronic energies and transition dipole moments have been obtained, in several studies on unsaturated hydrocarbons and heterocycles[28, 29], by employing the Complete Neglect of Differential Overlaps for Spectroscopy (CNDO/S) Hamiltonian [30] followed by CI calculations involving singles (CIS) or singles and doubles excitations (CISD). To calculate the intensity of false origins in dipole forbidden transitions the HT approach is adopted, and the adiabatic wavefunctions are expanded in terms of diabatic wavefunctions, following Eq. (23). The mixing coefficients are given by Eq. (28) and (29) from which we can write
&Q'T <XQ)=-
fau(g,Q)
ao
E r (Q 0 )-E s (Q 0 )
k
E r (Q 0 )-E s (Q 0 )
(43)
151 In practice, the adiabatic vibronic coupling integrals are easily evaluated in the diabatic wavefunction basis, which is chosen to maintain the MOs and CI coefficients fixed at the equilibrium structure values, while the basis of atomic orbitals is allowed to float with the nuclei [31]. Thus, they are obtained by computing the following numerical derivative Kk.=
'd(s(q,Q0)|H|r(q,Q
From Eq. (43) it follows that the a^s
(44)
coefficients are nonzero only if
r(s(q,Q0)))xr(Qk)xrjr(q,Q0)))=rTC
(45)
where F(x) is the irreducible representation associated with the x function. Eq. (45) allows to establish the symmetry of vibrational normal coordinates (among all the NTS vibrations) that can induce false origins in the spectra. The induced vibronic transition dipole moment for the false origin in the mode k of the g) —> e) forbidden transition will be
M.,
,K=
K
Vcok
(46)
and the expression of \x c can be obtained by equating Eq. (22) with Eq. (33), namely
(47)
For spin-forbidden transitions Eq. (41) and (42) can be employed. In this case the additional required parameters are the spin-orbit interactions among a large number of states with different spin multiplicity. When several SO couplings are required, again the semiempirical approach is favored. Thus, for a singlet-triplet transition the spin-orbit couplings H^°a between every singlet state Sr and the three components a = x,y, z of every triplet state Ts resulting from CIS calculations, can be evaluated in the one-electron approximation, according to the approach of ref.[32]: H
= S
T"
(48)
152 where ln and sn are orbital and spin angular momentum operators of the nth electron, A runs over the atoms of the molecule and C,A is the SO constant for atom A [32].
3. APPLICATIONS In this section we consider few examples, selected from recent literature, representative of the different models that must be employed to simulate vibronic structures. The include polyenes and their derivatives of biological relevance, fullerenes, whose electronic spectra are rich in features originated by vibronic interactions and, finally, an application in a field closer to material science, namely the simulation of the wavelength dependence of the Raman signal of carbonaceous materials. 3.1. Polyenes and biological chromophores Polyenes are conjugated carbon compounds of fundamental interest. Their skeleton structure is characterized by carbon-carbon bonds of alternate length, and they are subject to isomerization reactions upon photo-excitation. Owing to this property, polyene derivatives constitute the active chromophores in several biological systems, and their photoisomerization reactions represent the fundamental and ultrafast step in several biological cycles such as the visual photocycle.
Fig. 13. Equilibrium structure of the Si state of cyclobutene. Reprinted with permission from F. Negri, and G. Orlandi, F. Zerbetto, and M. Z. Zgierski, 'The Resonance Raman-Spectrum of Cyclobutene', J. Chem. Phys. 103, 5911-5918, Copyright (1995), American Institute of Physics (ref. [33]).
Resonance Raman spectroseopy has been used extensively to investigate the initial steps of photo-isomerization reactions for the reasons outlined in the introduction, and in section 2,
153 namely because RR intensities are sensitive to the geometry changes occurring upon excitation and owing to the fact that only chromophore bands are selectively enhanced, of particular relevance for chromophores embedded in protein environments. In this section we discuss first two relatively simple examples of RR spectra simulations, concerning a cyclic compound with a single CC double bond, namely cyclobutene and a short polyene, 1,3,5-hexatriene (135HT) and its triplet state RR spectrum. Then we move to a larger polyene derivative, the 11-cis protonated Schiff base of retinal and show how computed RR spectra have clarified the difference between initial photoreaction in solution and in the protein. 3.1.1. Cyclobutene Mathies and coworkers (A) measured the RR spectrum of cyclobutene (see Fig. 13) generated by an exciting beam in resonance with the excited singlet state dominated by the HOMO-LUMO excitation [34]. The scattering distribution among the different vibrational bands was used to obtain information about the initial stages of the dynamics of the resonant state. Clearly, the analysis of the RR spectrum and its implications on the dynamic aspects of the Si decay depend crucially both on the assignment of the observed vibrational frequencies and on the identification of the mechanisms of RR scattering. We calculated the ground state vibrational frequencies of cyclobutene at the MP2 and BLYP/6-31G* level and assigned the vibrational spectrum [33] improving upon previous analyses [35, 36]. In particular, we have shown that the coordinate associated with the conrotatory ring opening is the a2 (1145 cm"1) normal mode, while the disrotatory ring opening is associated with the bi (848 cm"1) normal mode. According the WoodwardHoffmann rules and potential energy surface calculations [37] the thermal ring opening occurs along the conrotatory (a2) path while the photochemical ring opening (in Si) follows the disrotatory (bi) path. Since So and Si belong to the Ai and B2 symmetry species, respectively, the bi modes, that are NTS, may appear in the RR spectrum only in an even number of quanta. Based on quantum-chemical calculations (So molecular parameters at HF/6-31G* and Si molecular parameters at CIS/6-31G* levels of theory) followed by RR spectra simulations (see Fig. 14) we analyzed the RR spectra of cyclobutene measured by Mathies et al. [34]. The analysis leads to the following conclusions: i) all the &\ modes, with the exception of the CH2CH2 stretch, are easily identified in the RR spectrum (at 983, 1110, 1180, 1440, 1570 cm"1). Among them, the band at 1570 cm"1 assigned to the C=C stretch is clearly prominent for its intensity; ii) the intensity of fundamentals of a2 modes is acquired via a vibronic coupling mechanism (terms B, C and D mentioned in Section 2.3). The activity of NTS modes corresponding to an even number of quanta is due to S0-S1 frequency shift via the FranckCondon mechanism (see Section 2.1). Only two a2 modes, at 328 and 902 cm"1, appear in the spectrum as fundamentals (902 cm"1) or in combination bands with one TS quantum (328+1563, 902+1110, 902+1563 cm"1) or as two quanta overtones (656, 1797, 2030 cm"1); iii) no bi fundamentals are observed. The 1075 cm"1 mode appears as two quanta overtone in the 2150 cm"1 band. The wide shoulder of the 1180 cm"1 band can be in part attributed to the overtone of the 635 cm"1 bi mode. The CH2 twist mode of 848 cm"1, corresponding to the disrotatory ring opening, could appear as an overtone of 1690 cm"1, which is on the shoulder of the 1650 cm"1 band that is attributed to a photoproduct. In summary, the analysis of the RR spectrum, based on quantum-chemically computed parameters, indicates (through the activity of the ai modes) that the molecule after excitation on the Sj(B2) state relaxes toward the new equilibrium CC bond lengths, in particular by lengthening the C=C bond.
154
c 0>
c o
E D
1000
1500 2000 2500 3000
Energy / cm" 1
Fig. 14. Experimental (bottom) from ref. [34] and simulated (top, only TS modes) resonance Raman spectrum of cyclobutene. Reprinted with permission from F. Negri, and G. Orlandi, F. Zerbetto, and M. Z. Zgierski, 'The Resonance Raman-Spectrum of Cyclobutene', J. Chem. Phys. 103, 5911-5918, Copyright (1995), American Institute of Physics (ref.[33]). The activity of the 328 and 902 cm"1 a2 modes indicates that the excited molecule relaxes toward the equilibrium geometry also by a non-planar motion along ring puckering and CH wag. In contrast, the lack of activity of modes related to the conrotatory and disrotatory ring openings indicates that during the first picoseconds after the excitation, the molecule relaxes toward the steepest descent direction on the S| PES, along the coordinates revealed by the RR spectrum, but does not begin the photoreactive motion of ring opening. Thus, the initial dynamics of cyclobutene inferred by this analysis agrees with the results of theoretical schemes such as the Woodward-Hoffmann rules, in that no vestiges of the conrotatory ring opening are observed. However, the spectrum does not show indications of disrotatory ring opening motions and in this sense this example serves to distinguish between initial relaxation dynamics and overall relaxation, that is the photochemical outcome. In fact, the photochemical process, in this case, develops on a time scale longer than the time interval explored by RR scattering while this spectroscopic technique provides a snapshot of the initial relaxation dynamics.
155 3.1.2. 1,3,5-hexatriene The recent advances in pulsed laser technology have led to the development of timeresolved spectroscopic methods that can follow the time-evolution of short-lived molecular species, typically molecules in excited states or radicals. Thus, time resolved absorption or Raman studies contribute not only to the knowledge of static molecular properties, but also to the dynamics of the short-lived species. Compared to singlet states, the lifetime of the lowest triplet states of polyenes is longer. This implies that, for sufficiently low energy barriers, the photo-excited species (conformers) that correspond to minima on the triplet PES, can equilibrate.
1,3,5-hexatriene-dO
600
1000
1400
1800
Wavenumber(cm-I) Fig. 15. Experimental (bottom) and simulated (top) resonance Raman spectrum of 135HT. Reprinted with permission from F. Negri, and G. Orlandi, 'The Ti Resonance Raman-Spectra of 1,3,5Hexatriene and Its Deuterated Isotopomers - an Ab-Initio Reinvestigation', J. Chem. Phys. 103, 24122419, Copyright (1995), American Institute of Physics (ref.[19]).
156 If equilibration is fast enough, the vibrational structure observed in the RR spectrum, whose measurement is time-delayed with respect to the triplet state pumping, is a source of indirect information on the shape (barriers and minima) of the triplet PES, via the contribution of equilibrated species to the RR spectrum. The triplet state RR spectra of both trans- and cisforms of 135HT were found to be identical [24]. This fact was initially interpreted as evidence for common, perpendicularly twisted intermediate, responsible for the RR spectra.
Fig. 16. Structure and atom numbering of the 11-cis protonated Schiff base of retinal.
Fig. 17. Structure of two models of PSB11 employed in computational studies: 1 was used in the study of Si PES and MEP [38], 2 was used in the simulation of RR intensities [20].
Later, on the basis of semiempirical QCFF/PI calculations, the contribution of the perpendicular form to the spectrum was ruled out because of the off-resonance conditions of the T, —> Tn transition [24], and the observed vibrational structure was assigned to the planar species. More recently [19] we have shown that ab initio calculations at CASSCF level predict the trans- form to be more stable than the cis- form by about 2 kcal/mol, in agreement with the indirect estimate deduced from experimental data [19]. Thus, a low barrier separates the two isomers, and equilibration in Ti can occur during the lifetime of the triplet state (ca 100 ns). As a consequence, the measured RR spectra (about 60 ns after triplet pumping) can only detect the equilibrated mixture. Accordingly the vibrational structure of the RR spectrum of both cis- and trans- forms is dominated by the trans- form since this is the lowest energy species in the equilibrated population. The comparison between the observed and simulated RR spectrum of 135HT is shown in Fig. 15. The agreement between simulated and observed RR spectra is excellent, which indicates that the CASSCF/6-31G* predicted molecular parameters are very reliable. Notice that the 6 electron - 6 n orbital space was considered in these calculations, so that the conjugated framework was described in an balanced way and similarly vibrational frequencies associated with skeleton motions.
157
Initial relaxation /
\
g
FC
Energy
Fig. 18. Structure of the Si PES along the relaxation path of the PSB11 model 1. Reprinted with permission from M. Garavelli, F. Negri, and M. Olivucci, J. Am. Chem. Soc. 121, 1023-1029, 'Initial excited-state relaxation of the isolated 11-cis protonated Schiff base of retinal: Evidence for in-plane motion from ab initio quantum chemical simulation of the resonance Raman spectrum', Copyright (1999), American Chemical Society (ref. [20]).
3.1.3. The 11-cis protonated Schiff base of retinal The 1 l-cis protonated Schiff base of retinal (PSB11) (see Fig. 16) is the chromophore of the rhodopsin protein, the human retina visual pigment. The photoisomerization of PSB11 to its all-trans isomer (PSBT) in rhodopsin is one of the fastest chemical reactions [39]: indeed, ground state PSBT is formed within 200 fs. In contrast, the photochemistry of free PSB 11 in solution is almost two orders of magnitude slower [40]. Before femtosecond experiments were available, resonance Raman spectroscopy was employed to study the initial dynamics of the primary light induced events in rhodopsin [41]. In these studies the effect of protein environment was identified in an enhanced torsional activity by comparing the RR response of PSB 11 in solution and in the protein [42]. The RR spectra of PSB 11 in solution and of rhodopsin are very similar, especially in the structurally sensitive fingerprint region [42, 43]. However, a major difference occurs in the activity of the band observed at ca. 970 cm"1, which is much more intense in rhodopsin. This mode was assigned to the out-of-phase 11,12 hydrogen out-of-plane (HOOP) wagging motion [44] and its activity was interpreted in terms of increased ground-state distortions about the Cio-Cn, Cn=Ci2 and C12-C13 bonds of the chromophore in the protein. The assumption of ground state distortion led to the appealing suggestion that the isomerization process would be primed in rhodopsin. From the experiments, however, it is not clear to what extent such twisting
158 should be reduced in PSB11 in solution, in order to account for both the decreased RR activity and the slower dynamics. In a recent computational study [20], RR intensities were simulated on the basis of entirely computed molecular parameters. The simulations were performed on a model carefully chosen to contain the most significant molecular characteristics of the retinal chromophore. To describe as accurately as possible the details of the vibronic structure, especially in the fingerprint region of the RR spectrum, model 1 in Fig. 17 was shown to be unsatisfactory, while the y,r|-dimethyl analogue (model 2 in Fig. 17) was proven to be the minimal model able to reproduce the observed vibronic features. Notice that this example demonstrates that the structural requirements necessary to model relaxation paths are generally different from those required to model spectroscopic properties and fine details. Indeed, model 1 was used to successfully model the PES of Si and the relaxation path upon excitation [38] (see Fig. 18) and inclusion of the additional methyl group in 2 does not change the conclusions drawn from the PES study carried out for 1. In contrast, inclusion of the methyl group is necessary for the simulation of RR intensities, since the methyl group internal coordinates couple quite strongly with the conjugated carbon skeleton coordinates.
B
1
600
800
1000 1200 1400 wavenumbers (cm1)
1600
1800
Fig. 19. Comparison between the experimental (bottom) and simulated (top) resonance Raman spectrum of PSB11. For the simulated spectrum, the model structure 2 ( see Fig. 17) was employed. Reprinted with permission from M. Garavelli, F. Negri, and M. Olivucci, J. Am. Chem. Soc. 121, 1023-1029, 'Initial excited-state relaxation of the isolated 11-cis protonated Schiff base of retinal: Evidence for in-plane motion from ab initio quantum chemical simulation of the resonance Raman spectrum', Copyright (1999), American Chemical Society (ref.[20]).
159 Such couplings are reflected in the internal coordinate composition of the vibrational normal coordinates of 2, which provide a nice representation of those of the full PSB11 chromophore, while those of 1 do not. The computation of equilibrium structures and vibrational normal coordinates was performed at CASSCF/6-31G* level of theory. The full active space of 10 JI electrons and 10 orbitals was employed. The resulting RR simulated spectrum is depicted in Fig. 19 where it is compared with the experimental spectrum of PSB11 in solution taken from ref. [43]. A Lorentzian line width of 15 cm"1 was used to plot the simulated spectrum. The similarity between simulated and observed spectra is remarkable and suggests that the PSB11 model chosen to mimic the visual chromophore provides a realistic representation of the So and Si potential energy changes. Although considerable intensity appears in the highfrequency region, significantly more important is the structurally sensitive fingerprint region (1100-1330 cm"1) of the RR spectrum. The activity in this frequency region identifies the isomeric form of the chromophore and the choice of a representative model, even though truncated with respect to the full PSB 11 chromophore, ensures realistic spectral simulations. Indeed four main bands are observed in this frequency region of the RR spectrum of PSB 11 in solution and in rhodopsin, at 1190, 1218, 1237 and 1276 cm" , and correspondingly four bands are computed for model 2, at 1163, 1201, 1225 and 1272 cm"', with intensity distributions very close to the observed spectra. Beside this region, the most interesting region in the RR spectrum, as regard the interpretation of the different photoreaction dynamics of PSB 11 in solution and in the protein, is between 900 and 1100 cm"1. The simulated spectra show, in this region, some activity at about 970 cm"1 in agreement with the observed spectra. However, this activity is not due to the HOOP mode described above, since the equilibrium structure of the chromophore is planar, and only in-plane TS modes appear in the simulated RR spectrum. Thus, the results of the simulations of RR spectra show that the initial in-plane relaxation predicted by the PES studies on model 1, are perfectly consistent with the RR spectra of the chromophore in solution, since an in-plane mode of about 970 cm"1 exists, which is responsible for the observed intensity. In other words, the experimental RR spectrum of PSB 11 in solution is perfectly compatible with an initial in-plane relaxation of the excited chromophore, and the HOOP activity does not have to be invoked to account for the spectral features. Conversely, this mode is certainly responsible for the increased activity in this region of the spectrum of rhodopsin, where both in-plane and the HOOP modes contribute to the intensity in this frequency region. Thus, this is an example that shows how the simulation of spectral intensities on one side allow for the correct attribution of the observed bands and, on the other side, support and reinforce the conclusions drawn on the basis of PES and MEP studies [38] which indicated a photo-isomerization model characterized by initial skeletal relaxation followed by slow evolution along a plateau on the torsional subspace of the Si PES. 3.2. Fullerenes Among the properties of fullerenes, the recently discovered carbon clusters [ref], related to their electronic structure and symmetry, the optical properties are of particular relevance for potential applications that might be based on them, such as the production of fullerene-based optical limiters [45]. The optical spectroscopy of the two most abundant fullerenes, C6o and C70 (see Fig 20), along with theoretical analyses and simulations of the electronic spectra, have recently been reviewed [46].
160
Fig. 20. The two most abundant fullerenes C(,o (right) and C70 (left).
HT induced transition (false origin)
vibronic coupling via mode e2o
\ i
J. Sn(Tlu) /
1/
"--
r
V
NTS=1
s
i(Tig'T28.
G
g)
dipole allowed transition dipole forbidden transition (true origin)
Fig. 21. Schematic representation of the HT intensity borrowing mechanism active in C6o
Here we discuss briefly the electronic spectroscopy of Cgo, not only because it is the most important fullerene, but even because it represents a recently discovered molecular system whose electronic spectroscopy interpretation can be accounted for only by considering the HT mechanism of intensity borrowing discussed in Section 2.2. Thus, in the context of this chapter, it represents an up to date example where the simpler FC approach fails to explain spectroscopic features. On the other hand we restrict to C60 and we do not discuss other fullerenes such as C70, since the lower symmetry of the latter implies energy gaps between vibronically coupled electronic states that cannot be dealt with the HT perturbation approach described in Section 2.2, and thus, their discussion goes beyond the scope of this chapter. The density of electronic states is impressive also for Ceo, but owing to its higher symmetry, the
161 dipole forbidden states responsible for the emission and for the onset of the absorption spectrum, are well separated from dipole allowed states. The lowest singlet excited states of C6o belong to T[g, T2g and Gg irreducible representations of the Ih symmetry point group, and their transition from So is dipole forbidden. These three states are computed to be quasi degenerate [28], and because excitation energies cannot be predicted (with quantum-chemical methods) with accuracy of tens of cm"1, the identification of the lowest singlet excited state cannot rely simply on the order predicted from calculations, but must be accompanied by simulation of the spectra and comparison with experimental data. The three states discussed above, gain intensity by borrowing it from T[U dipole allowed states, the lowest of which is ca. 1 eV above the dipole forbidden states. Thanks to this considerable energy difference, the perturbative approach of Section 2.2 can be employed to simulate the intensity of false origins. A schematic representation of the HT mechanism for Ceo is shown in Fig. 21. The intensities of false origins were computed along the lines described in Section 2.5, for the three lowest electronic states. These intensities were combined with the FC activities of TS modes and of Jahn-Teller (JT) active modes which, in the limit of strong JT coupling leading to static deformations, can be described in terms of displacement parameters (Eq. (15) similarly to TS modes [47, 48]. The emission spectra were then simulated by assuming a mixed nature (in terms of Tig, T2g, and Gg character) of the emitting state and were compared with the experimental spectra recorded at 4 K in Neon and Argon matrix. The simulated and experimental spectra are collected in Fig. 22. The agreement between observed and simulated spectra is surprising, especially if one considers the size of the molecule and the complexity of the mechanism which activates the absorption bands. The experimental emission spectra are different in the two matrices, and this fact was interpreted with a different degree of matrix induced mixing of the three Tig, T2g, and Gg states. To account for the Ne matrix spectrum, the mixing coefficients were 37, 56 and 7%, respectively, while for the argon matrix, simulations had to be carried out by imposing the following mixing coefficients: 50, 25 and 25%, respectively. The Argon matrix thus seems to mix more extensively the three lowest energy states, an indication that argon matrix perturbations are stronger than those in Neon matrix. The potential applications of C6o as an optical limiting material are related, among others, to the long lifetime of its triplet state Ti. The lifetime of Ti is remarkable (about 400 u.s in hydrocarbon matrices below 10 K [49, 50]) owing to the inefficiency of both radiative and radiationless T, —> So deactivation routes. According to quantum-chemical calculations, Ti belongs to T2g symmetry, and thus its transition to So (phosphorescence) is both symmetry and spin forbidden and is induced by spin-orbit and vibronic interactions as described in Section 2.4. Since the phosphorescence transition is very weak, a common method to enhance, experimentally, its intensity, is to make use of the heavy atom affect, by dispersing Ceo molecules in matrices containing heavy atoms which enhance the spin-orbit perturbations. In addition, if the symmetry of the molecule remains Ih, the transition is still symmetry forbidden, and one expects to see only false origins, induced vibronically, along with progressions of TS and JT active modes, built on the false origins. Although phosphorescence spectra have been measured also without the heavy-atom effect [50] here we restrict the attention on the phosphorescence spectrum measured making use of the heavy-atom effect [51], which is shown in Fig. 23(a). This spectrum shows a strong 0-0 band (the first band in the spectrum) which indicates a symmetry lowering of C60 in the Xe matrix environment.
162
a)
c)
. , . < •»
41)11
Mill
1200
1600
Frequency Shift from Origin (cm1)
Fig. 22. Experimental (a),(c), from ref. [47] and correspondingly simulated (b),(d) emission spectra of C60. (a) emission of C60 in Ne matrix, (c) emission observed in Ar matrix. Reprinted with permission from A. Sassara, G. Zerza, M. Chergui, F. Negri, and G. Orlandi, 'The visible emission and absorption spectrum of C M \ J. Chem. Phys. 107, 8731-8741, Copyright (1997), American Institute of Physics (ref. [47]).
163
I
500
1000
1500
2000
250C
wavenumber / cm~
Fig. 23. Simulated and observed phosphorescence spectra of C6o. (a) the experimental spectrum obtained in Xe matrix at 30 K [51] (b) the simulated spectrum. Reprinted with permission from M. G. Giuffreda, F. Negri, and G. Orlandi, J. Phys. Chem. A 105, 9123-9129, 'Quantum-chemical modeling and analysis of the vibrational structure in the phosphorescence spectrum of CM', Copyright (2001), American Chemical Society (ref.[52]). In Fig. 23 (b) the simulated spectrum is plotted for comparison [52]. This latter spectrum was obtained by computing the intensities of false origins along the lines described in Section 2.4 and 2.5. In addition, the FC activity of TS and JT modes was evaluated by computing the displacement parameters for the T, —> S o transition, according to Eq. (15), an the intensity of a member of the progression in the kth FC or JT active mode is related to the intensity of the false origin on which the progression is built, by the following expression
I t ( v t ) = l(false-origin)) x
To compare the computed and the observed spectrum in Fig. 23, we included in the former a strong 0-0 band and calculated the intensities of the associated progressions of TS and JT
164 active modes. The comparison between computed and observed intensities leads to the identification of several false origins in the spectrum. If one considers that band intensities result from the combination of two perturbations, namely spin-orbit and vibronic coupling, and that the size of the molecule is considerable, the agreement between observed and computed spectra is really remarkable.
3.3. Graphitic materials The main characteristic in the Raman spectra of amorphous or graphitic carbon materials is the presence of the so called D band (D for defect) which is originated by the presence of sp2 nanostructures in the amorphous material or by defects in regular graphitic sheets. The defects can be, for instance, portions of graphenes of molecular size, and in this sense they can be nicely represented by polycyclic aromatic hydrocarbons (PAHs). This implies that localized n electron regions of the size of PAHs can be responsible for the spectral features observed in the carbonaceous materials. Thus, we can model the changes occurring in the Raman spectra of carbon based materials by studying, quantum-chemically, the spectral features of PAH molecules with different shapes and sizes. The experimental studies carried out on several kinds of carbonaceous materials have shown evidences of a resonance Raman mechanism for the activation of the D band, whose dispersion with respect to the excitation wavelength [53] is presented in Fig. 24. The simulations of resonance Raman intensities were carried out on the PAHs presented in Fig. 25, characterized by different shape and size in order to mimic a distribution of defects in the carbon material.
0 1000
1200
1400 1600 1800 Raman Shift [cm 1 ]
2000
Fig. 24. Position of the D peak in the Raman spectra of polycrystalline graphite, as a function of excitation energy. Reprinted from J. Non-Cryst. Solids, 230, I. Pocsik, M. Hundhausen, M. Koos, and L. Ley, 'Origin of the D peak in the Raman spectrum of microcrystalline graphite', 1083-1086, Copyright (1998), with permission from Elsevier (ref. [53]).
165 D&, Bynanetty
C222
C294
C366
C102
C132
C384
D21, symmetry
C9G
Fig. 25. Molecular structure of the polycyclic aromatic hydrocarbons considered to simulate the localized defects in graphitic materials. Reprinted with permission from F. Negri, E. di Donato, M. Tommasini, C. Castiglioni, G. Zerbi, and K. Muellen, 'Resonance Raman contribution to the D band of carbon materials: Modeling defects with quantum chemistry', J. Chem. Phys. 120, 11889-11900, Copyright (2004), American Institute of Physics (ref. [54]).
In previous studies it was shown that the Raman activity in the D band region is dominated by TS vibrations, thus, simulations of resonance Raman intensities can be carried out by restricting to the A term of Eq. (36). As shown in Fig. 25, the sample of PAHs consists of very large molecules. In addition, the molecular parameters required to simulate RR intensities include equilibrium structures of excited states and vibrational frequencies.
166
1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
Excitation Energy [eV] 1600
(600
1400
I
W
1
1200
1000
800
.
, , . 1 . . . .
0.5
.
15
.
.
•
.
2.5
,
,
.
.... 3.5
4.5
Fig. 26. Frequency dispersion with respect to the excitation energy. (Top) observed G band and D band frequency dispersion. Reprinted from J. Non-Cryst. Solids, 230, I. Pocsik, M. Hundhausen, M. Koos, and L. Ley, 'Origin of the D peak in the Raman spectrum of microcrystalline graphite', 10831086, Copyright (1998), with permission from Elsevier (ref. [53]). (Bottom) simulated D band frequency dispersion. Reprinted with permission from F. Negri, E. di Donato, M. Tommasini, C. Castiglioni, G. Zerbi, and K. Muellen, 'Resonance Raman contribution to the D band of carbon materials: Modeling defects with quantum chemistry', J. Chem. Phys. 120, 11889-11900, Copyright (2004), American Institute of Physics (ref. [54]).
167 The size of some of the molecules in Fig. 25 prevents the use of ab initio methods to calculate excited state parameters. Thus we restricted to the QCFF/PI semiempirical approach that, as discussed in the previous section, was employed to describe successfully structural and vibrational properties of conjugated and aromatic compounds. The computed displacement parameters (Eq. (15)) indicate that the largest Raman activity is expected in the D-band region, namely in the 1200-1350 cm"1 frequency region [54]. In order to simulate the Raman frequency dispersion in graphitic materials, we assume that such dispersion is due to the selective enhancement of the Raman bands due to specific defects (sp2 nanostructures) resonantly enhanced by the chosen excitation light. In other words, for a given excitation wavelength, one or more defects (in our model PAHs) will have dipole allowed excited states in resonance with the light and hence their resonance Raman activity will dominate the Raman spectrum. PAHs are good models for the description of sp nanostructures, but they differ due to the presence of hydrogen atoms along their edges. For this reason we described the embedding of the sp nanostructures in the surrounding carbon matrix by replacing the usual masses of hydrogen with a larger mass, according to a united atom approach. With this procedure we decouple the Raman active vibrations of the sp2 nanostructure from spurious vibrations of the PAH, such as CH bending motions. The Raman signal is then simulated by considering the whole distribution of PAHs. When the laser energy is in resonance with a given excited state of a given PAH molecule, strong resonance occurs and it can be shown [54] that the signal is proportional to the square of the largest displacement parameter Bk of the PAH selected by resonance excitation. By repeating the simulation for several excitation wavelengths we can plot the frequency positions of the intensity maxima versus excitation wavelength. This is done in Fig. 26 (bottom) where a comparison with the observed [53] frequency dependence (dispersion) is also presented (top). It is remarkable that the dispersion slope of the D band obtained from the simulations (54 cm'VeV), matches closely the slope extracted from experimental studies (~ 50 cm"7eV). 4. CONCLUSIONS AND PERSPECTIVES The models of increasing complexity, discussed in this chapter to account for the vibronic structure in electronic spectra, were proven to be extremely useful for the interpretation of the experimental data of several classes of molecules. These models are not new, and were initially applied to small molecular systems of limited interest and formed by few atoms. However, in recent years, the tremendous growth in computational resources and the development of efficient algorithms for reliable quantum-chemical predictions of molecular parameters has favoured the application of the same models to molecular systems of increasing dimension and broader interest. These studies have contributed to elucidate the role that nuclear motions can have in coupling electronic states or in determining the early evolution of excited states. Since computational resources keep increasing, we believe that, in future years, a careful choice of the appropriate models will help to disentangle problems connected with even larger and complex molecular systems than those described here. Acknowledgements The financial support from MIUR (Project: "Modellistica delle Proprieta Spettroscopiche di Sistemi Molecolari Complessi", funds ex 60%; project: "Dinamiche
168 Molecolari in Sistemi di Interesse Chimico", funds ex 40%, and FIRB project: "Carbon based micro and nano structures", RBNE019NKS) as well as from the University of Bologna (Funds for Selected Research Topics) is gratefully acknowledged. REFERENCES [I] A. B. Myers and R. A. Mathies, in Biological Apllications of Raman Spectroscopy, Vol. 2 (T. G. Spiro, ed.), John Wiley & Sons, 1987, p. 1. [2] R. J. H. Clark and T. J. Dines, Angew. Chem. Int. Ed. Engl. 25 (1986) 131. [3] G. Fischer, Vibronic Coupling, Academic Press, London, 1984. [4] R. Neumann and C. Engler, Chem. Phys. 161 (1992) 229. [5] C. Manneback, Physica (Utrecht) 17 (1951) 1001. [6] W. Siebrand and M. Z. Zgierski, in Excited States, Vol. 4 (E. C. Lim, ed.), Academic Press Inc., 1979, p. 1. [7] F. Duschinsky, Acta Phisicochim. USSR 7 (1937) 551. [8] M. Z. Zgierski, Chem. Phys. 108 (1986) 61. [9] F. Negri and M. Z. Zgierski, J. Chem. Phys. 104 (1996) 3486. [10] M. Z. Zgierski, J. Chem. Phys. 85 (1986) 109. [II] A. R. Gregory, W. H. Henneker, W. Siebrand, and M. Z. Zgierski, J. Chem. Phys. 67 (1977)3175. [12] F. Negri, G. Orlandi, F. Zerbetto, and M. Z. Zgierski, J. Chem. Phys. 93 (1990) 600. [13] K. S. Pitzer, Quantum Chemistry, Prentice-Hall, Englewood Cliffs, New Jersey, 1953. [14] D. C. Harris and M. D. Bertolucci, Symmetry and Spectroscopy: an Introduction to Vibrational and Electronic Spectroscopy, Oxford University Press, 1978. [15] G. Orlandi and F. Zerbetto, Chem. Phys. 108 (1986) 187. [16] A. C. Albrecht, J. Chem. Phys. 34 (1961) 1476. [17] W. Siebrand and M. Z. Zgierski, J. Chem. Phys. 71 (1979) 3561. [18] F. Negri and M. Z. Zgierski, J. Chem. Phys. 99 (1993) 4318. [19] F. Negri and G. Orlandi, J. Chem. Phys. 103 (1995) 2412. [20] M. Garavelli, F. Negri, and M. Olivucci, J. Am. Chem. Soc. 121 (1999) 1023. [21] A. Warshel and M. Karplus, J. Am. Chem. Soc. 19 (1972) 5612. [22] F. Negri and G. Orlandi, J. Phys. Chem. 93 (1989) 4470. [23] F. Zerbetto, M. Z. Zgierski, F. Negri, and G. Orlandi, J. Chem. Phys. 89 (1988) 3681. [24] F. Negri, G. Orlandi, A. M. Brouwer, F. W. Langkilde, and R. Wilbrandt, J. Chem. Phys. 90(1989)5944. [25] F. Negri and M. Z. Zgierski, J. Chem. Phys. 97 (1992) 7124. [26] F. Negri and M. Z. Zgierski, J. Chem. Phys. 100 (1994) 2571. [27] F. Negri and G. Orlandi, J. Photochem. Photobiol. A-Chem. 105 (1997) 209. [28] F. Negri, G. Orlandi, and F. Zerbetto, J. Chem. Phys. 97 (1992) 6496. [29] F. Negri and G. Orlandi, J. Phys. B-At. Mol. Opt. Phys. 29 (1996) 5077. [30] J. Del Bene and H. H. Jaffe1, J. Chem. Phys. 48 (1968) 1807. [31] G. Orlandi, Chem. Phys. Lett. 44 (1976) 277. [32] C. A. Masmanidis, H. H. Jaffe1, and R. L. Ellis, J. Phys. Chem. 79 (1975) 19. [33] F. Negri, G. Orlandi, F. Zerbetto, and M. Z. Zgierski, J. Chem. Phys. 103 (1995) 5911. [34] M. K. Lawless, S. W. Wickam, and R. A. Mathies, J. Am. Chem. Soc. 116 (1994) 1593. [35] N. C. Craig, S. S. Borick, T. R. Tucker, and Y.-Z. Xiao, J. Phys. Chem. 95 (1991) 3549. [36] K. B. Wiberg and R. E. Rosenberg, J. Phys. Chem. 96 (1992) 8282. [37] W. T. A. M. van der Lugt and L. J. Oosterhoff, J. Am. Chem. Soc. 91 (1969) 6042.
169 [38] M. Garavelli, T. Vreven, P. Celani, F. Bernardi, M. A. Robb, and M. Olivucci, J. Am. Chem. Soc. 120(1998) 1285. [39] T. Yoshizawa and O. Kuwata, in CRC Handbook of Organic Photochemistry and Photobiology (W. M. Horspool and P.-S. Song, eds.), CRC Press, Boca raton, FL, 1995, p. 1493. [40] H. Kandori, Y. Katsuta, M. Ito, and H. Sasabe, J. Am. Chem. Soc. 117 (1995) 2669. [41] G. R. Loppnow and R. A. Mathies, Biophys. J. 54 (1988) 35. [42] I. Palings, J. A. Pardoen, E. van den Berg, C. Winkel, J. Lugtenburg, and R. A. Mathies, Biochemistry 26 (1987) 2544. [43] R. A. Mathies, T. B. Freedman, and L. Stryer, J. Mol. Biol. 109 (1977) 367. [44] G. Eyring, B. Curry, A. Broek, J. Lugtenburg, and R. A. Mathies, Biochemistry 21 (1982)384. [45] L. W. Tutt and A. Kost, Nature 356 (1992) 225. [46] G. Orlandi and F. Negri, Photochem. Photobiol. Sci. 1 (2002) 289. [47] A. Sassara, G. Zerza, M. Chergui, F. Negri, and G. Orlandi, J. Chem. Phys. 107 (1997) 8731. [48] C. Cepek, A. Goldoni, S. Modesti, F. Negri, G. Orlandi, and F. Zerbetto, Chem. Phys. Lett. 250(1996)537. [49] M. R. Wasielewski, M. P. Oneil, K. R. Lykke, M. J. Pellin, and D. M. Gruen, J. Am. Chem. Soc. 113(1991)2774. [50] D. J. Vandenheuvel, I. Y. Chan, E. J. J. Groenen, J. Schmidt, and G. Meijer, Chem. Phys. Lett. 231 (1994) 111. [51] A. Sassara, G. Zerza, and M. Chergui, Chem. Phys. Lett. 261 (1996) 213. [52] M. G. Giuffreda, F. Negri, and G. Orlandi, J. Phys. Chem. A 105 (2001) 9123. [53] I. Pocsik, M. Hundhausen, M. Koos, and L. Ley, J. Non-Cryst. Solids 230 (1998) 1083. [54] F. Negri, E. di Donato, M. Tommasini, C. Castiglioni, G. Zerbi, and K. Muellen, J. Chem. Phys. 120 (2004) 11889.
This Page is Intentionally Left Blank
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 Elsevier B.V. All rights reserved
171
V. Semiclassical Nonadiabatic Trajectory Computations In Photochemistry: Is The Reaction Path Enough To Understand A Photochemical Reaction Mechanism? Graham A. Worth, Michael J. Bearpark and Michael A. Robb Chemistry Department, Imperial College London, South Kensington Campus, London SW7 2AZUK
1. Conceptual Introduction: There Are Three Important Coordinates In Photochemistry Fig. 1 shows parts of the ground and excited state potential energy surfaces for the photochemical 2S+2S cycloaddition of two ethylenes [1,2]. Since this is a well-known textbook example of mechanistic photochemistry, we will use it to introduce some of the concepts that are necessary to read this chapter.
Fig. 1. The photochemical 2S+2S cycloaddition of two ethylenes.
The photocycloaddition of two ethylenes is also an appropriate starting point because, historically, it was the first organic photochemistry problem for which the mechanistic involvement of a conical intersection was demonstrated [1].
172 The obvious choice of reaction coordinate is R, the interfragment separation (Fig. 1, bottom right). This coordinate brings the two ethylenes together face-to-face in a Woodward Hoffman allowed process. Woodward Hoffmann theory predicts that along this coordinate there will be a minimum A on the excited state, and a corresponding maximum A' on the ground state. This concept of an avoided crossing funnel is where mechanistic photochemistry begins [4-8]. In the past ten years, however, both theory and experiment have shown that this onedimensional picture is misleading, because fast radiationless decay does not take place at the resulting avoided crossing; the energy gap is too large. To resolve this problem, one needs to add a second dimension to the picture. This 'extra' coordinate is shown in Fig. 1 (bottom left) as a rhomboidal distortion x, orthogonal to R. Now A (which was a minimum along R alone) becomes a transition structure, and the negative direction of curvature leads through a surface crossing E - a conical intersection (CI) [3-10] into the region of an open tetramethylene biradicaloid C on the ground state. Thus ground state photochemical products are produced by decay of the excited state through a conical intersection. The minimum number of coordinates that you need to describe a radiationless decay event via a conical intersection is two: for the ethylene photocycloaddition these are R and x (Fig. 1), both of which can lift the degeneracy. Finally, we should point out that Fig. 1 is a 'cartoon', intended to illustrate mechanistic ideas. In the computations themselves, we normally use all geometric degrees of freedom without constraint, but afterwards we try to rationalize the results in terms of a small number of coordinates that can be visualized. When describing the CI itself another aspect that becomes important is the cone topology [10]. In two dimensions there are two types of intersection: peaked (Fig. 2a) and sloped (Fig. 2b). The important point about a peaked intersection is that you can move away from the cone on the lower surface in seemingly any direction, with the possibility of generating multiple products. This passage through the intersection can be described as 'sand flowing through a funnel'. In contrast, if you have a sloped intersection you have a more directed crossing, with the greater likelihood of multiple crossings between ground and excited states and a purely photophysical event leading to reactant regeneration. Marked on Fig. 2 are arrows indicating the trajectory followed by a (classical) ball rolling through the intersection. This is the reaction path: the route of minimum energy joining critical points on the surfaces.
Fig. 2: Peaked (a) and sloped (b) conical intersections.
173 Understanding Fig.s 1 and 2 gives an essential insight into conical intersections and mechanistic photochemistry. But more recently, we have begun to appreciate that many nonadiabatic processes can only be fully understood by introducing a further coordinate, bringing the total to three. We now introduce this idea with some simple mathematical preliminaries. Fig. 3 shows the shape of a conical intersection for a model problem when the energy is plotted in the space of the two degeneracy-lifting (branching space) coordinates q\ and q2. (The model used is three orbitals and three electrons, but the conclusions are general). Here, a 'sand through the funnel' reaction path from excited state reactants (R"a) to ground state products (Pa) passes through the lowest energy point of conical intersection (CI R*a// Pa).
Cl (R*a/Pa)
Fig. 3: A three orbital, three electron conical intersection showing energy as a function of the two branching space (degeneracy-lifting) coordinates.
However, if you keep one of the branching space coordinates (say gi) fixed at the intersection, and plot the energy in the other branching space coordinate (^2) along with one of the coordinates orthogonal to the branching space (labeled ^3), then the conical intersection occurs as a 'seam' as shown in Fig. 4. This happens when q2 is at its value for the intersection, as moving along 93 cannot lift the degeneracy. CI(R b /Pb) \
Fig. 4: A three orbital, three electron conical intersection showing energy as a function of one of the two branching space (degeneracy-lifting) coordinates q2, and one of the 37V-8 orthogonal intersection space (degeneracy-preserving) coordinates g3.
174 Now consider a reaction path R b to R b in a valley parallel to the qs axis (Fig. 4). This reaction path does not lead to the intersection: to access it at all, one needs energy in the orthogonal branching space coordinate qi, which could result in crossing away from the lowest energy region of intersection. As drawn in Fig. 4, this type of surface is of no direct interest for photochemistry, as both reactant Rb and product Pb lie on the ground state surface (although this is relevant for electron transfer problems [11]). Our purpose here is to show that, if the reaction path lies outside the branching space of the crossing (Fig. 4), there are three independent coordinates that need to be considered. Contrast this with Fig. 3, where the reaction path is a combination of the branching space coordinates: there are three coordinates, but only two are independent. The difference between the paths illustrated in Fig.s 3 and 4 has important mechanistic consequences, as we shall show with examples later. We have just conjectured that to understand photochemistry (and nonadiabatic processes such as electron transfer), one needs three geometric variables: the two branching space variables q\ or 92 that lift the degeneracy and the reaction path itself. We can see that the reaction path may lie in the space of the two branching space variables q\ or 92 (corresponding to 'sand flowing through the funnel', Fig. 3), or in the intersection space (i.e. 'parallel' to the 'seam', Fig. 4). It should be obvious that in the second case, reactivity is determined by the extent to which energy flows into the degeneracy lifting coordinates. In general, there are many coordinates orthogonal to the two branching space coordinates, but the one that is really interesting is the reaction path, if it is independent (as in Fig. 4). When the reaction path does not lie in the space of the degeneracy-lifting branching coordinates, a study of the reaction path and knowledge of the lowest energy point of a conical intersection [12] is not enough to understand reactivity. Rather, the central idea in photochemistry should be the relationship between the reaction coordinate, the intersection (seam), and the path actually followed by the system. The only general way to investigate this is with molecular dynamics, and the focus of this article is to describe how dynamics studies can provide the mechanistic information required. While it is possible to map out segments of the seam [13], dynamics studies (since they are energy or temperature dependent) will explore the 'chemically relevant' parts of the seam. Let us restate what the central theme of this chapter will be. Consider Fig. 5. Here we have plotted the reaction path for a reaction that involves a ground state barrier in one dimension. The other dimension is one of the two (degeneracy-lifting) coordinates from the branching space. The conical intersection thus appears as a seam between two surfaces (as in Fig. 4, but in a different orientation).
175
Reaction Coordinate
Fig. 5: Model potential energy surface, showing a reactivity problem in the space of a reaction coordinate and a branching space coordinate.
Fig. 5 suggests some interesting mechanistic possibilities. Suppose that there are reactant and product basins on So, (left and right, foreground of Fig. 5). If you promote the system onto the excited state and there is no energy in motions orthogonal to the reaction coordinate, then the model of sand flowing down the funnel would work: the system will follow the downhill reaction path (Rx) to the excited state minimum. In this case, there is a conical intersection minimum near the Si minimum and the reaction path happens to intersect with the seam. Alternatively, one can also imagine that if there is vibrational motion orthogonal to the reaction path direction along a branching space coordinate, then the system can evolve along this coordinate as well and decay anywhere along the seam: either in the reactant region (left) before descent on the reaction path on Si has begun, or in the product region (right) after the system has relaxed along the Si reaction path. This suggests important experimental consequences. In a coherent control experiment, one aims to optimize a particular product by using laser light to modify the momentum components of a molecular wavepacket [14]. Where the mechanistic picture sketched in Fig. 5 is valid, it shows a principle that could be used for designing such experiments: if the laser used to excite the molecule puts energy into the branching space modes, decay at specific regions of the potential surface can be prompted, and hence the desired product obtained. The plan for the remainder of this chapter is first, to give a simple guide to molecular dynamics studies on photochemical systems, and then to try to illustrate some of the ideas we have just introduced with some specific case studies. The reader is advised to see recent reviews for further details of applying dynamics to photochemical system [15] and to the general theory of quantum dynamics and nonadiabatic transitions [16,17].
176 2. Molecular Dynamics And Mechanistic Photochemistry In the introduction we showed that to understand photochemical reactivity it may be necessary to go beyond the idea of following a reaction path through a minimum energy conical intersection, and to look not only at how the reaction coordinate relates to the intersection seam, but also at the actual dynamical path of the system. In short, we need to know where the system meets the intersection in order to determine where it crosses to the ground state, and this depends not only on the topology of the potential energy surfaces but also on the dynamics of the system, i.e. the momenta of the particles. The dynamics of a molecular system are of course governed by the time-dependent Schrodinger equation,
dt
where V is the nuclear wavefunction and H the Hamiltonian operator. The simplest model of photoexcitation is a vertical transition, moving the ground-state molecular wavefunction onto the excited-state potential energy surface without changing the nuclear geometry (Fig. 5, left). The wavepacket is then no longer a stationary state, and motion results. A vertical transition is sketched in Fig. 6. The Hamiltonian in this case consists of the nuclear kinetic energy operator and potential energy surfaces provided by the electrons. In more complicated models, a time-
dependent excitation operator (such as a laser pulse) could also be included. Fig. 6: Molecular dynamics following vertical excitation by light with energy Iru. Motion 1 takes the wavepacket away from the Franck-Condon geometry. Motion 2 leads towards the conical intersection. Motion 3 leads away from the conical intersection on the lower surface: 3a is an adiabatic transition and 3b is diabatic.
Using numerical methods the time-dependent Schrodinger equation can be solved, but only for small systems containing 3-4 atoms. Larger systems can be treated either by reducing the number of degrees of freedom, or else by using simple model Hamiltonians [17]. In this way, much information has been gained about what happens during inter-state crossing.
177 For treating the polyatomic molecules of interest in photochemistry, rather than solving the full Schrodinger equation to obtain the quantum dynamics of the system, it is usual to use an approximation that turns the wavepacket and its motion into a 'swarm of trajectories'. This is a set of trajectories starting with all of the possible initial positions and momenta contained in the wavepacket, that move according to the classical equations of motion, mx = - W where x are the particle coordinates and - W i s the force on the nuclei provided by the derivative of the potential energy surface. In our calculations, the potential energy surface experienced by the molecules is provided by 'on-the-fly' calculations, i.e. quantum chemistry is used to calculate the electronic energy and gradients for a molecular configuration on a trajectory when it is required. Such direct dynamics schemes are the only feasible methods for calculations on polyatomic molecules, as determination of an accurate many-dimensional potential function is unfeasible. Ignoring for the moment quantum mechanical effects such as interference, the major dynamics of a wavepacket are determined by the motion of its centre and the change in its width: the spread. According to the Ehrenfest theorem, the centre of a wavepacket follows the classical trajectory. A vertical transition does not put any energy into the nuclear degrees of freedom, and so if the system was initially at equilibrium on the lower state surface, the centre of the packet has no initial momentum. The initial direction of motion of the system will thus be along the force vector away from the Franck-Condon point, and the system will follow the reaction path, picking up momentum as it moves down the slope. This is labeled as motion 1 in Fig. 6. The width of the wavepacket grows with time according to the uncertainty principle. This change in shape is modeled by the many different trajectories in the swarm with non-zero momenta and positions away from the centre of the wavepacket. The distribution of initial conditions is determined by the quantum wavepacket. In principle these should be chosen from an appropriate Wigner distribution, but a simple classical distribution sampling constant energy conditions may be good enough. At some point the system will approach the conical intersection seam and, due to the nonadiabatic coupling, will start to feel the effect of the lower potential surface as the surfaces get closer. This is the motion in region 2 of Fig. 6. The passage through a conical intersection is a quantum mechanical process. The simple classical picture used for the initial motion is no longer applicable, and semiclassical methods must be used such as trajectory surface hopping or Ehrenfest dynamics, to introduce the quantum mechanical information.
178 In the surface-hopping picture, on entering the intersection region a trajectory has a probability of changing (hopping) to the other state. The essential factor can be taken as the time evolution of the electronic state amplitude, c,
where |c,.| is the probability of the system being in state i, Vi is the potential energy surface, R is the particle velocity vector and
is the nonadiabatic derivative coupling vector. When the probability of being in the lower state reaches a certain threshold, a hop takes place. Thus efficient transfer is due to a coupling between the velocity vector of the system and the nonadiabatic coupling vector. For a mechanistic description, the points where hops take place define the range of geometries at which decay takes place. Two basic types of transition can occur. The first is when a hop takes place before the intersection is reached. This is called an adiabatic transition, and is denoted in Fig. 6 by motion 3a. The second is when the trajectory moves through the intersection in what is called a diabatic transition (motion 3b) and ends up on the lower surface on the other side of the intersection. As the name suggests, no change in electronic configuration occurs during a diabatic transition, whereas an adiabatic transition is accompanied by a change in electronic configuration. In Ehrenfest dynamics, each trajectory experiences a potential energy due to both surfaces, weighted by the state populations - a mixed-state trajectory. This means that the nonadiabatic transition is not done in the ad hoc way of surface hopping, but it introduces the problem that unphysical trajectories can result which move away from the intersection but remain in a mixed state. After passing through the intersection, the trajectories then follow the ground-state surface to one of the possible photoproducts. As for the initial relaxation on the excited state, this part of the process can be reasonably described by classical trajectories. It would be useful to know how good the approximations used in the semiclassical methods are. In model studies, it is found that the semiclassical methods are able to capture the major features of a nonadiabatic transition. In particular the rate of population transfer is usually well described. The major difference found between the quantum and semiclassical descriptions is that in the quantum picture the wavepacket bifurcates at the intersection and interference effects take place between the different parts. Significantly, these parts move
179 coherently, but this coherence is lost in the semiclassical description, and so the motion on the lower surface is different. Unfortunately, limitations on the quantum dynamics mean that it has not really been possible to compare the two methods in the long-time limit for realistic, full-dimension calculations so far. We are presently developing a method to do this: the direct-dynamics variational multi-configuration Gaussian (DD-vMCG) method [18]. In this, the nuclear wavefunction is expanded as a superposition of Gaussian functions
where As are expansion coefficients and g, a Gaussian wavepacket basis set
8i =
Using a variational principle, equations of motion are set up for the expansion coefficients and the Gaussian parameters, A, § and r\. The Gaussian functions effectively follow trajectories, but they are coupled and so all the quantum mechanical information is correctly included. The method promises to be not only more accurate than semiclassical calculations, but also more efficient.
3. Photochemical Reactivity Involving Intramolecular Single Bond Breaking: Diarylethylene And Dihydroazulene/Vinylheptafulvene Photochromism Diarylethylenes are remarkable photochromic systems [19-23] where the chemical transformation is single bond breaking (Fig. 7).
CHD Fig. 7: Diarylethylenes with heterocyclic aryl and bisthienylethylene-based compounds (X=S) exhibit remarkable switching sensitivity (i.e. high quantum yield) and rapid response.
Because the distribution of Jt-bonds is different in both isomers, they have distinct absorption spectra. The photophysics and the efficiency of the system are completely controlled by the relationship between the reaction path and the degeneracy-lifting coordinates.
180 Fig. 8 shows the energy profile along a bond-breaking coordinate, and Fig. 9 the coordinates that lift the degeneracy. Like many of the examples we will discuss, the coordinates that lift the degeneracy are just skeletal deformations.
closed-ring isomtr
•
! - : I - I I I I -
•
HT q ring-opening reaction coordinate
Fig.s 8: Diarylethylenes - energy profile along the single bond breaking reaction path [24].
Fig. 9: Diarylethylenes - degeneracy-lifting coordinates.
There is a ground state thermal reaction path involving a transition state TSo in Fig- 8. In addition, there is an adiabatic reaction path on the excited state involving a minimum, a transition state TS, and another minimum. One can also find three critical points (local minima) on the conical intersection line indicated by crosses in Fig. 8: one near the products, one near the reactants, and also one near the transition state on the adiabatic excited-state reaction path. Thus, apparently disconnected conical intersections lie displaced along the intersection space coordinates (one of which is the transition vector shown in Fig. 10). \ \ Fig. 10: Diarylethylenes - reaction coordinate (single bond made or broken).
181 Fig. 11 shows a cartoon of the diarylethylene potential surfaces that can be distilled from the dynamics computations reported in [24].
reaction coordinate Fig. 11: Diarylethylenes - ground and excited state potential energy surfaces. The conical intersection appears as a line in this picture: the reaction path is in the foreground (left to right), and the conical intersection line (CI) lies in the background. Thus the reaction coordinate is parallel to the seam and so decay to So is controlled by motion orthogonal to the reaction path. Furthermore, the transition state (TS) is a narrow bottleneck. Just because the system has enough energy to get over the transition state, it doesn't mean that it will find it. In fact, passing through the transition state will be a very rare event: not only do you need the energy to be right, but also you need it to be distributed in exactly the right coordinates. A trajectory started at the Franck-Condon structure on the HT side of the reaction will remain in the HT* minimum for a long time, and will not necessarily find the transition state. By contrast, the conical intersection is a many dimensional hypersurface, and consequently more accessible. It is not a dynamical bottleneck in the same way as a transition state. Once the crossing seam is reached, decay to the ground state is immediate. The minimum energy point on the conical intersection line (middle cross, CI 3 in Fig. 8) is the one that is located using gradient driven optimization [12]. While it appears quite close to the transition state on the excited state, it is actually on the HT side of the barrier. In the dynamics computations [24], a typical trajectory started from the Franck-Condon region on the HT side of the reaction path lives in the HT* minimum for almost two picoseconds, until a chance vibration that has enough energy orthogonal to the reaction coordinate drives it towards the CI seam and decay takes place to the ground state. The central point is that knowledge of the excited-state reaction path does not yield an understanding of the photochromism of this system. Simply finding the conical intersection points does not yield a complete picture, because they do not lie on the reaction path. Indeed, a large segment
182 of the intersection seam in this region is energetically accessible. However, to demonstrate and understand this, you need to run dynamics calculations. Experimentally, one observes a fluorescence that is red-shifted, confirming that the position of the minimum is different on the excited state. The cyclization quantum yield (HT* to CHD) is high, which arises from the fact that a trajectory from HT* can sample the whole intersection seam, at right angles to the reaction path. On the other hand, for a trajectory starting from CHD*, the probability of decay to So is low, because the main locus of the conical intersection seam appears to be on the HT* side of the transition state. Thus to reach HT from CHD* you have to pass through the transition state on the excited state reaction path. Thus there is a competition between passing through the transition state to reach the reactive conical intersection on the HT* side of the TS, and decay at a nearby crossing on the CHD* side of the transition state, which does not lead to any reaction. Thus the quantum yield is quite low in this direction. There remains the question of whether one could design a pulsed laser sequence to 'control' this reaction. For the HT isomer, one would want a wavepacket with excess momentum in the branching space direction so that decay would take place quickly. In contrast, for the CHD isomer, one would need a wavepacket with excess momentum in the direction of the reaction coordinate, which would drive the reaction towards the transition state on the excited-state and avoid competition with any nearby conical intersection points. We now turn to another similar reaction, which seems to have the optimum design in terms of the orientation of the reaction path to the seam direction: Dihydroazulene (DHA) / Vinylheptafulvene (VHF), Fig. 12 (see [25] and [26] and references cited therein).
DHA
VHF
Fig. 12: Dihydroazulene (DHA) / vinylheptafulvene (VHF) photochromism.
When you photolyse DHA, it ring-opens to VHF. This photochromic DHA/VHF couple is very efficient: the forward reaction is photochemical, but the backward reaction only takes place thermally.
183
(a)
Fig. 13: The reaction path (a) and degeneracy-lifting coordinates (b) for a model [26] for the DHA/VHF photochemical reaction.
Fig. 13a shows the reaction coordinate (mainly a single bond breaking/making) and Fig. 13b the degeneracy-lifting coordinates. Our dynamics study [26] of this system shows that the reaction path from DHA intersects the conical intersection branching space well after the excited state TS and near a minimum Cl structure that is VHF-like. The potential surface is similar to the model surface in Fig. 5. However in the case of DHA/VHF, the excited state minimum and the CI minimum are at the same geometry (peaked intersection) and very much displaced towards the vicinity of the ground state VHF minimum, rather than occurring near the TS as in Fig. 5. In this system the relationship between the reaction path and the branching space directions is different on the DHA side of the reaction path (where the seam and the reaction path are parallel like diarylethylenes) and the VHF side (where 'sand through the funnel' is a good approximation because the reaction path intersects the branching space funnel). Because of the latter, the excitation of VHF prompts skeletal deformation in the branching space, rather than motion along the reaction coordinate. So the VHF to DHA transformation occurs only along a ground state thermal reaction coordinate. 4. Photochemical Reactivity Involving Cis-Trans Double Bond Isomerization We now turn to two examples of double bond cis-trans isomerization reactions, where the cartoon shown in Fig. 5 is a good model. We shall briefly describe two limiting situations: one where the seam is accessible only near the 'half-isomerized geometry' and so the 'sand in the funnel model' works well, and another where the majority of the trajectories decay near the reactant geometry because the seam is accessible all along the reaction path. We start with a minimal model of the retinal chromophore of rhodopsin, the human visual pigment.
184
1
2
For the Z-penta-3,5-dieniminium cation 1 the relationship between the seam and the reaction path has been documented in [13] and is similar to that shown in Fig. 5. In this case, the reaction path is the cis-trans isomerization coordinate. The branching space is spanned by a skeletal deformation coordinate, which results in a single-double bondlength inversion and an NH2 pyramidal distortion. The reaction path runs directly into the surface-crossing seam at a twist angle of ca. 76 degrees while the minimum point of the CI seam occurs at ca. 90 degrees. Fig. 14 summarizes the results of running trajectories on this surface [27]. Clearly, although the seam exists at small angles, the majority of the trajectories decay in the 60-80 degree region. Thus the photodynamics of the cis-trans isomerization is controlled by a small segment of the potential surface where the reaction path enters the low energy (near the minimum) part of the conical intersection seam.
Zi.
—
—
»-
1
1 as-ire
is.
ID
0-
so
^ ^ 1 ^m m 100 torsion angte [dag]
i
1
Fig. 14: Distribution of the twist angle at the surface hop geometries for 1. We now turn to the cyanine dye l,l'-diethyl-4,4'-cyanine (1144-C) shown in Fig. 15. We have [28] mapped the potential surface for the model shown in Fig. 16.
Fig.s 15: Cyanine dye 1,1 '-diethyl-4,4'-cyanine (1144-C).
185
Fig. 16: Three-carbon model of 1144-C (Fig. 15).
Femtosecond dynamics measurements by Sundstrom [29] show a bi-modal distribution at Franck-Condon energies, as well as an intermediate with a first order decay. In the model surface shown in Fig. 5, the reaction coordinate is cis-trans isomerization. The branching space is skeletal deformation coupled with pyramidal NH2 distortion. Our dynamics results are summarized in Fig.s 17 and 18.
Time (Femtoseconds)
Fig.s 17: Excited state lifetime for three-carbon model of 1144-C. Minimi urn Ct Structure 1 T|=104'
I..,,,,,,,,, |AT| (decrees)
Fig. 18: Surface hop angle along cis-trans isomerization coordinate for three-carbon model of 1144-C.
186 Fig. 18 shows the geometries where the system hops; the minimum on the seam occurs at a dihedral angle of around 104 degrees. However, most of the trajectories hop before the molecule has begun to rotate. Thus the seam is accessible along the skeletal deformation coordinate before cis-trans isomerization can occur. In Fig. 17 we show the Si lifetime. There is a large distribution that hops very quickly before rotation takes place, and a second broad distribution that corresponds to population that progresses along the cis-trans coordinate and decays at the half-twisted geometry. Thus the Z-penta-3,5-dieniminium cation (PSB) and the three carbon model 1144-C have similar potential energy surfaces, characterized by the fact that the cis/trans isomerization coordinate intersects the CI seam at the half rotation geometry. However, dynamics computations suggest that for the cyanine dye example, the sand flowing through the funnel model does not work and the system samples the seam at all torsion angles.
5. Cis-Trans Isomerization In A Protein As a last example, we discuss [30] the cis-trans isomerization of a double bond in the covalently bound />-coumaric acid chromophore (Fig. 19) in Photoactive Yellow Protein (PYP), an archetypal reversible protein photoreceptor. A combination of ab initio (CASSCF) dynamics with surface hopping and classical molecular dynamics (MD) simulation techniques has been used to directly simulate the process of photoisomerization within the protein.
Fig. 19: The p-coumaric acid chromophore in PYP. The chromophore is covalently linked to the sidechain of Cys69 through a thioester bond. The p-hydroxypheny 1 moiety is deprotonated, but stabilized by hydrogen bonding interactions with the side chains of Tyr42 and Glu46.
We have used CASSCF for the chromophore itself and molecular mechanics for the remainder of the system. A cartoon of the potential energy surfaces in vacuo and in the protein is shown in Fig. 20.
187
vacuo
protein
* ^ ^
#
Fig. 20: Potential energy surfaces of the excited and ground states in the trans-to-cis isomerization coordinate (torsion b, Fig. 19) and a skeletal deformation of the bonds: in vacuo (a) and in the protein (b).
In vacuo, there is an excited-state transition state barrier and minimum where the chromophore is partly twisted, and a minimum with a half twist. The main point is that the conical intersections in the protein and in the gas phase are significantly different. Crossing in the gas phase involves substantial additional motion to reach the seam at the half twist geometry. Again, the coordinates that lift the degeneracy are the skeletal deformations. In the gas phase dynamics, using the same initial conditions as in the protein simulations, the system never makes it over the first partial twist torsion barrier. In contrast, in the protein, the excited state is specifically stabilized by the charge distribution of the protein: one observes a decrease of the Si-So energy gap in the region of the twisted intermediate (from 80 kJ mol"' in vacuo to less than 1 kJ mol"' in the protein), accompanied by a displacement of the crossing seam closer to the global minimum. One also sees a decrease of the energy barrier separating the early planar Si minimum and the twisted Si minimum. In total, 14 dynamics simulations are discussed in [30]. In the protein, the lifetime of the excited state ranged from 129 to 2293 fs. The ratio of the number of successful isomerizations to the number of excitedstate trajectories is ~0.3, close to the experimental quantum yield of 0.35. Statistically, the number of trajectories is small, but they nevertheless yield a consistent mechanistic picture. PYP is probably the most dramatic example of a situation where the reaction path is simple (just torsion), and orthogonal to the degeneracy-lifting coordinates (mainly skeletal deformations). In this case, the reactivity is changed when you add the electric field of the protein. Nature has been very careful to position one charged residue in exactly the right place.
188 6. Conclusions In this chapter we have shown that it is often essential to appreciate that a conical intersection between potential energy surfaces forms a 'seam', even though it is the minimum energy point along this seam that is usually referred to as the 'conical intersection'. Several examples show that inter-state crossing takes place after accessing this intersection seam away from its minimum energy point, along coordinates orthogonal to the reaction path. As a result, if theoretical investigations are limited to the calculation of the minimum energy path, then one may miss many chemically interesting effects. These effects can be explored using 'on the fly' dynamics. Finally, we believe that some of the ideas we have been discussing may be useful in designing laser pulses for coherent control experiments. The title question was: is the reaction path enough to understand a photochemical reaction mechanism? The answer is: it depends on the relationship between the excited-state reaction path and the branching space that defines any conical intersection involved. If the reaction path lies in the branching space of the conical intersection, then one can probably understand the reaction mechanism without worrying much about dynamics explicitly. If the reaction path is orthogonal to the branching space, and three independent coordinates are important, then one needs dynamics to know where radiationless decay will occur, as it will probably be away from the lowest energy point on the intersection. This contrasts with a reaction on a single potential energy surface, in which a single coordinate—the reaction path at a transition structure—is most important. As for the choice of dynamics method: semiclassical dynamics (classical dynamics + surface hopping) is sufficient to obtain mechanistic information, as we have shown with a number of examples. Full quantum calculations for photochemical reactions are being developed at the time of writing.
189 References [I] Mechanism Of Ground-State-Forbidden Photochemical Pericyclic Reactions - Evidence For Real Conical Intersections, F. Bernardi, S. De, M. Olivucci and M.A. Robb, J. Am. Chem. Soc, 112 (1990) 1737-1744. [2] Geometry Optimization On A Hypersphere - Application To Finding Reaction Paths From A Conical Intersection, P. Celani, M.A. Robb, M. Garavelli, F. Bernardi and M. Olivucci, Chem. Phys. Lett., 243(1995)1-8. [3] Internal Conversion In Polyatomic Molecules, E. Teller, Isr. J. Chem., 7 (1969) 227. [4] Molecular Orbital Correlation Diagrams Mobius Systems And Factors Controlling Ground- And Excited-State Reactions 2, H.E. Zimmerman, J. Am. Chem. Soc, 88 (1966) 1566. [5] J. Michl, Mol. Photochem., 4 (1972) 243. [6] A. Gilbert and J. Baggott, Essentials of Molecular Photochemistry, Blackwell Scientific Publications, Oxford, 1991. [7] J. Michl and V. Bonacic-Koutecky, Electronic Aspects of Organic Photochemistry, Wiley, New York, 1990. [8] M. Klessinger and J. Michl, Excited States and Photochemistry of Organic Molecules, VCH Publishers, New York, 1994. [9] Intersection Of Potential Energy Surfaces In Polyatomic Molecules, G. Herzberg and H.C. Longuet-Higgins, Discuss. Faraday Soc, 35 (1963) 77. [10] Potential Energy Surfaces Near Intersections, G. J. Atchity, S.S. Xantheas and K. Ruedenberg, J. Chem. Phys., 95 (1991) 1862-1876. [II] Intramolecular Electron Transfer: Independent (Ground State) Adiabatic (Chemical) And Nonadiabatic Reaction Pathways In Bis(Hydrazine) Radical Cations, E. Fernandez, L. Blancafort, M. Olivucci and M.A. Robb, J. Am. Chem. Soc, 122 (2000) 7528-7533. [12] A Direct Method For The Location Of The Lowest Energy Point On A Potential Surface Crossing, M.J. Bearpark, M.A. Robb and H. B. Schlegel, Chem. Phys. Lett., 223 (1994) 269-274. [13] Relationship Between Photoisomerization Path And Intersection Space In A Retinal Chromophore Model, A. Migani, M.A. Robb and M. Olivucci, J. Am. Chem. Soc, 125 (2003) 2804-2808. [14] Applications Of Wavepacket Methodology, A. Stolow, Phil. Trans. Roy. Soc. Lond. A, 356 (1998) 345-362; Coherent Control For Ultrafast Photochemical Reactions, R. de Vivie-Riedle, L. Kurtz and A. Hofmann, Pure Appl. Chem., 73 (2001) 525-528. [15] Applying Direct Molecular Dynamics To Non-Adiabatic Systems, G.A. Worth and M.A. Robb., Adv. Chem. Phys. 124 (2002) 355-432. [16] Beyond Born-Oppenheimer: Molecular Dynamics Through A Conical Intersection, G.A. Worth and L.S. Cederbaum., Annu. Rev. Phys. Chem. 55 (2004) 127. [17] Multidimensional Dynamics Involving A Conical Intersection: Wave-Packet Calculations Using The MCTDH Method, G.A. Worth, H.-D. Meyer and L.S. Cederbaum, in Conical Intersections: Electronic Structure, Dynamics And Spectroscopy, eds. W. Domcke, D.R. Yarkony and H. Koppel., World Scientific, Singapore, 2004. [18] A Novel Algorithm For Non-Adiabatic Direct Dynamics Using Variational Gaussian Wavepackets. G.A. Worth, M.A. Robb and I. Burghardt. Farad. Disc. 127 (2004) xxx [19] Organic Photochromism, H. Bouas-Laurent and Diirr, H., Pure Appl. Chem., 73 (2001) 639-665. [20] Spiropyrans And Spirooxazines For Memories And Switches, G. Berkovic, V. Krongauz and V. Weiss,. Chem. Rev., 100 (2000) 1741-1753. [21] Fulgides For Memories And Switches, Y. Yokoyama,. Chem. Rev., 100 (2000) 1717-1739. [22] Diarylethenes For Memories And Switches, M. Irie,. Chem. Rev., 100 (2000) 1685-1716. [23] Ultrafast Dynamics Of Photochromic Systems, N. Tamai and H. Miyasaka, Chem. Rev., 100 (2000)1875-1890.
190 [24] Can Diarylethene Photochromism Be Explained By A Reaction Path Alone? A CASSCF Study With Model MMVB Dynamics, M. Boggio-Pasqua, M. Ravaglia, M.J. Bearpark, M. Garavelli and M.A. Robb, J. Phys. Chem. A, 107 (2003) 11139-11152. [25] Dihydroazulene/Vinylheptafulvene Photochromism: Dynamics Of The Photochemical RingOpening Reaction, J. Ern, M. Petermann, T. Mrozek, J. Daub, K. Kuldova and C. Kryschi, Chem. Phys., 259(2000)331-337. [26] Dihydroazulene / Vinylheptafulvene Photochromism: A CASSCF Model For One-Way Photochemistry Via A Conical Intersection, M. Boggio-Pasqua, M.J. Bearpark, P.A. Hunt and M.A. Robb, J. Am. Chem. Soc, 124 (2002) 1456-1470. [27] Probing The Photochemical Funnel Of A Retinal Chromophore Model Via Zero Point Energy Sampling Semiclassical Dynamics, O. Weingart, A. Migani, M. Olivucci, M.A. Robb, V. BuB and P.A. Hunt, J. Phys. Chem. A, (2004) in press. [28] Ultrafast Radiationless Deactivation Of Organic Dyes: Evidence For A Two-State Two-Mode Pathway In Polymethine Cyanines, A. Sanchez-Galvez, P. Hunt, M.A. Robb, M. Olivucci, T. Vreven and H. B. Schlegel, J. Am. Chem. Soc, 122 (2000) 2911-2924. [29] Overdamped Wavepacket Motion Along A Barrierless Potential-Energy Surface In Excited-State Isomerization, A. Yartsev, J. L. Alvarez, U. Aberg and V. Sundstrom, Chem. Phys. Lett., 243 (1995)281-289. [30] Photoactivation Of The Photoactive Yellow Protein: Why Photon Absorption Triggers A TransTo-Cis Isomerization Of The Chromophore In The Protein, G. Groenhof, M. Bouxin-Cademartory, B. Hess, S.P. de Visser, H.J.C. Berendsen, M. Olivucci, A.E. Mark and M.A. Robb, J. Am. Chem. Soc, 126(2004), 4228-4233.
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 ElsevierB.V. All rights reserved
191
VI. Computation of Photochemical Reaction Mechanisms in Organic Chemistry M. Garavelli, F. Bernardi, and A. Cembran Dipartimento di Chimica "G. Ciamician" dell'Universita di Bologna, Via Selmi 2, 40126 Bologna, Italy 1. INTRODUCTION Modern technology exploits and controls organic materials with a precision that was inconceivable only few decades ago.[l] This is a consequence of the fact that the performance and property of a material can be related to the structure and behaviour of the constituting molecules. For instance, considering the topic of the present book, the photostability of polymers, sunscreens, drugs, paints, etc. depends on the ability to waste, at the molecular level, the absorbed photon energy either via non-radiative (e.g. internal conversion) or radiative (e.g. fosforescence) channels. In general, the more control we have on molecular properties, the more a material will fit our needs and, for this reason, in the past chemists have learned to synthesize, for instance, photostable and luminescent molecules. A novel research target in the area of the control of molecular properties is represented by the design of molecular machines: molecules that react to a certain external signal by displacing, usually reversibly, one or more of their parts. Modern chemistry and technology are rapidly moving along this way: molecular devices and machines are, nowadays, under investigation, giving rise to the so-called molecular technology, or else termed nanotechnology.[\] Between other molecular devices, those based on reversible photochemical reactions have a great interest. These devices are operated irradiating the molecule at the wavelength required to trigger the photochemical process. In principle, the design of the right reagent, allows for a direct control of the reaction rate, efficiency and photostability even when the interconversion between the two (or more) "states" of the device need to be repeated over a large number of cycles.[2-5] It is apparent that the elucidation of the factors controlling photochemical reactions is imperative for the rational design of such a material. In particular, it is apparent that, in order to achieve this goal, it is mandatory to elucidate the details of the photochemical reaction mechanism. Among others, this requirement provides a timely and solid motivation for the topic developed in the present chapter. In recent years computational chemistry has gained increasing consideration as a valid tool for the detailed investigation of photochemical reaction mechanisms. Below we will outline the strategy and operational approach to the practical computational investigation of
192 reaction mechanisms in organic photochemistry. The aim is to show how this task can be achieved through high-level ab initio quantum chemical computations and ad hoc optimization tools, using either real or model (i.e. simplified) systems. Another purpose of the present Chapter is to show that, nowadays, a computational chemist can adapt his/her "instruments" (the method, the approach and the level of accuracy) to the problem under investigation, as every other scientist does when there is a problem to study and a methodology to be chosen. In particular, different and often complementary computational tools may be used as "virtual spectrometers" to characterize the molecular reactivity of a given chromophore. The general approach used to follow the course of the photochemical reaction involves the construction and characterization of the so called "photochemical reaction path". This is a minimum energy paths MEP[6] starting at the reactant structure and developing along the potential energy surfaces (PES) of the photochemically relevant states. Such interstate path usually originates at the Franck Condon (FC) point on the spectroscopic state and ends at the ground state photoproduct valley. Such an approach has been named the photochemical reaction path (see also Chapter 1) or, more briefly, pathway approach[7, 8]. Within this approach one pays attention to local properties of the potential energy surfaces such as slopes, minima, saddle points, barriers and crossings between states. The information accessible with this method is structural: i.e. the calculated path describes, strictly, the motion of a vibrationally cold molecule moving with infinitesimal momentum. While the path does not represent any "real" trajectory, it allows for a rationalization of different experimental data such as the excite-state lifetimes, the nature of the photoproducts and, more qualitatively, the quantum yields and transient absorption and emission spectra. As we will see in Section 2 this approach can be related to the common way of describing photochemical processes with the motion of the centre of a wave packet along the potential energy surfaces. [9] Notice also that the analysis of the photochemical reaction path is currently receiving new attention as a consequence of the recent advances in femtosecond spectroscopy and ultrafast techniques. [8] In most past work, the energy surface structural features and, ultimately, the entire reaction path have been computed by determining the molecular wavefunction with state-ofthe-art ab initio methods. In particular, a combined ab initio CASPT2[10, 11]//CASSCF[1215] methodology has been extensively used since it has been proven to reproduce data with nearly experimental accuracy.[8, 16] This approach will be described in detail in Sections 3 together with a number of commonly used potential energy surface mapping tools. The operational procedure for approaching a photochemical problem will then be described and discussed in Section 4. The applications of such a procedure to the intriguing problems of determining the mechanism of the photoinduced cis-trans isomerization of a retinal protonated Schiff base (RPSBs) model and of azobenzene (Ab) will be discussed in Sections 5 and 6 respectively. Both these chromophores have an extended conjugated it-system and are characterized by ultrafast and efficient cis-trans isomerizations taking place upon photoexcitation. Thus, these systems can potentially be employed in nanotechnology for the design and construction of molecular devices such as random access memories, photon
193 counters, picosecond photo detectors, neural-type logic gates, optical computing, lightswitchable receptors and sensors, light addressable memories and molecular motors just to mention a few.[2, 17, 18] Finally, the complex network of reaction paths underlying the photochemical reactivity of cyclooctatetraene (taken as a representative of cyclic conjugated hydrocarbons) will be discussed in Section 7 to illustrate both general and subtle aspects of photochemical organic reaction mechanisms. 2. MODELLING PHOTOCHEMICAL REACTIVITY Following their impressive development, in the past few decades ab initio quantum chemical methods have been widely applied to the investigation of the reactivity of molecules in their ground electronic state. [19, 20] Consequently, it is now possible to calculate structures and relative energies for fairly large molecules with high precision. For instance, it is possible to locate the transition structure (i.e. the structure connecting the reactant to the product) and the associated energy barrier with a ca. one kcal mol"1 error. Similarly the entire thermal reaction path (i.e. the progression of the reactant molecular structure towards the transition state and the product) can be determined in an unbiased way by computing the corresponding MEP[6] along the 3N-6 dimensional potential energy surface of the system. The corresponding development for photochemical reactions has been slower. One reason is the larger complexity of excited state wavefunctions. In fact, while, for most systems close to their equilibrium geometry, the electronic ground state is well described by a single electronic configuration, this is not the case for excited states. There, the contributions of different electronic configurations have to be accounted for since these may be close in energy and mix heavily. It is often difficult to decide a priori which configurations will be important in a given photochemical reaction. As also described in Chapter 2 the multiconfigurational approach[14] avoids this decision by dividing the orbitals into three sets: inactive, active, and secondary orbitals. The secondary orbitals are never occupied, in any of the considered electronic configurations. The inactive orbitals are doubly occupied for all configurations (they contain electrons which are perfectly described as closed shell electrons): they do not participate in excitation processes and are not involved in any chemical rearrangement which may take place on the potential energy surface. The remaining (na) orbitals are active and are chosen according to the specific chemical problem we are interested in. For instance if one is interested in a chemical reaction involving the photoexcitation and rearrangement of the jr-system of a molecule, the full set of jt electrons and it orbitals is included (if possible) into the active space. There are typically less than 2«a electrons housed in the active orbitals, therefore they can give rise to a number of excited configurations. A linear combination of them originates the excited state wavefunction of interest. Indeed the method is called complete active space self consistent field (CASSCF) since it computes all the configurations within the given active space, optimizing the linear combination coefficients, together with the orbitals included in the active and inactive spaces.[12-15] The extensive testing carried out over the years has shown that the CASSCF method well describes the topography of the reactive potential energy surfaces. On the other
194 hand, energies may not be accurate enough since it does not fully account for the electron correlation energy. Therefore, in order to refine the energetics by including electronic correlation, CASPT2[10, 11] single point calculations have to be performed for selected points along the MEP determined at the CASSCF level. Below we will focus on the results of this CASPT2//CASSCF approach, although, as shown in other chapters of this book, it is not the only one available for excited state reactivity studies. As already mentioned above photochemical reaction path (where the reactant typically resides on an excited state PES and the products accumulate on the ground state) is expected to have at least two branches: one located on the excited state and the other located on the ground state potential energy surface. The main difficulty associated with such a computation lies in the correct definition and practical computation of the "funnel" where the excited state reactant or intermediate is delivered to the ground state. While the progression on the excited state energy surface {i.e. the excited state "branch" of the reaction path) may be investigated with the same methods used for thermal reactions, a general way of locating the "locus" where the excited state branch of the reaction path is connected to the ground state branch or branches requires non-standard methods. Thus during the last decade, computational chemists have been able to develop novel concepts, tools and strategies to solve this fundamental problem.[14, 16, 21-27] In particular a systematic computational investigation of a wide range of photochemical organic reactions has lead to novel concepts that allow the definition of the photochemical reaction mechanism in a rigorous way and with a language which is familiar to chemists. Here, we will focus on the new tools (partly developed by our group), which permit this thorough description to be achieved.
Radiationless Decay Mechanisms (a)
(b)
Energy
Energy
Reactant Reaction Coordinate Avoided Crossing Scheme 1 (From ref. [28])
Reactant Reaction Coordinate Crossing (Conical Intersection)
195 2.1. Conical Intersections in Photoinduced Processes The classic text-book view of photochemical reactions is mainly due to the 1969 computational work of Van der Lugt and Oosteroff.[29] These authors proposed the decay of an excited species taking place at an excited state energy minimum corresponding to an avoided crossing of the excited and ground state potential energy surfaces (Scheme la). Zimmerman[30], Teller[31] and Michl[32] were the first to suggest, independently, that certain photoproducts may originate by decay of the excited state species through a conical intersection (CI) of the excited and ground state potential energy surfaces (Scheme lb). Zimmerman and Michl used the term "funnel" for this feature that corresponds, in contrast to Van der Lugt and Oosteroff results, to a real crossing of two potential energy surfaces possessing the same spin and spatial symmetry. Nearly two decades of systematic computational studies showed that conical intersections represent a general mechanistic entity in photochemistry, in analogy with transition states for thermal reactions. Furthermore, low-lying real crossings between photochemically relevant electronic states where found with a previously unsuspected frequency.^,, 16, 27] Consequently, an excited state has a high probability of entering a region where the excited state crosses the ground state. Such crossings provide a very efficient "funnel" for radiationless deactivation (i.e. internal conversion), which may occur in a single molecular vibration (i. e. in a subpicosecond timescale) and, in turn, prompt photoproducts formation. These results imply that when two or more surfaces are considered in the exploration of a photo-process, some reaction path features are expected. In summary one expects: the existence of an accessible funnel corresponding to a CI connecting the excited to the ground state branches of the reaction path; the overall excited state motion is guided by the MEP (i.e. a steepest descent path) and therefore by local structural features of the potential energy surfaces; the branching of the reaction path at the position corresponding to the CI leading to two or more ground state relaxation paths leading to the final photoproducts or back to the original reactant. Radiationless decay at a conical intersection implies: (a) a 100% efficient internal conversion (i.e. the Landau-Zener[23, 25] decay probability will be unity) that makes the intersection a structural bottleneck for the reaction, (b) a slow decay rate (e.g. the competition with fluorescence) may reflect the presence of some excited state energy barrier which separates the excited state intermediate M from the intersection structure CI (Fig. 1) and (c) in the case where the decay leads to a chemical reaction, the molecular structure at the intersection must be related to the structure of the observed photoproducts (P), as in thermal reactions the structure of the transition state is related to that of the products.
196
FC Energy Conical Intersection (CI)
*
Absorbtion Emission
\ \ Reactant Reaction Coordinate Fig. 1. Schematic reaction path for a barrier controlled "opening" of a fast radiationless decay channel. (From ref. [28])
2.2. Computational Photochemistry Points a-c presented above provide the theoretical basis for the modelling of photochemical reactions or, in other words, for computational photochemistry. The molecular motion is assumed to be controlled by the structure of the relevant excited and ground state potential energy surfaces. Thus, information on the excited state lifetime and on the type of photoproducts generated is obtained by computing and analyzing the reaction coordinates and energies (i.e. the reaction path) connecting the Franck-Condon point, to the excited state intermediate M* (if existing), to the ground state. In conclusion, as illustrated in Scheme 2, the strategy used in computational photochemistry is based on the mapping of the photochemical reaction path computed by following the MEP from the starting (e.g. FranckCondon structure FC) to the final (e.g. ground state photoproducts P' and P") points through, for example, a conical intersection CI. As mentioned above this approach, which is already employed in textbooks [25], has an intimate connection to the method using the motion of wave packets or semi-classical trajectories on potential surfaces to describe ultrafast photochemical processes. 2.3. MEP Mapping versus Standard Geometry Optimization The strategy outlined above provides information on the structure and accessibility of the photochemical reaction paths (i.e. the MEPs in computational terms) from a chosen starting point (e.g. the FC point). This technique has the advantage of self-limiting the investigation only to the region of the PES which is relevant for the description of the photochemical reaction. In other words, by following the MEP, we immediately focus onto the driving forces responsible for the photoinduced nuclear motion. Therefore, only those intermediates, transition states and funnels which are directly accessible by the system, will
197 be located as "travelling-points" along the MEP. Many other stationary points and crossing regions (which may be located, in principle, via systematic geometry optimizations) may be very far from the followed reaction channels (MEPs), both in energy and geometry, and are not important for the description of the process. Thus, for example, the decay point (i.e. the photochemical funnel) intercepted by the MEP may differ from the optimized lowest-energy one. This explains why the information given by the MEP may be different (or complementary) from that provided by locating stationary points and low-lying crossings thus yielding a more general and extended description of the shape of the potential energy surface. We will give examples of this in Section 5. \u:i>
Irjji-ctory """J i- ^
Reaction Path Modelling STATIC, LOCAL (iviiiiU'trt Optimizations
MEP (IRD, IRC1 < immli \ Optimizations of lhib Cunieal Intersection O
RclaxiiIkiiil'iilhslMEl')
fromCKIKIMKC) DYNAMIC. GLOBAL Q Semi-classical Trajectories Wave packet D.vnumics
Ground State
'^T I"
p"
Scheme 2
2.4. Topography versus Dynamics. Since reacting molecules have usually a finite amount of kinetic energy, a trajectory will not follow the MEP and may, in principle, deviate quite dramatically from it (e.g. in the case of "hot" systems where there is a large excess vibrational energy). In this case, regions of the potential energy surface far from the computed photochemical reaction path may become important and a dynamical treatment of the reaction is unavoidable. In other words in these cases the topography of the potential energy surfaces may not provide a correct representation of the molecular dynamics. On the other hand, at room temperature the reaction path usually becomes a useful mechanistic entity since a molecule will move (on the average) along the valley defined by the MEP. In this situation, one may compute the Reaction Path Hamiltonians[33] and apply the Transition State Theory (in its traditional[34] or variational[35] formulation) to estimate many properties of the system. Of course the photochemical reaction path becomes even more informative for vibrationally cold systems
198 {e.g. photochemical reactions where the excited state reactant has a small/controlled amount of vibrational excess energy), providing insight into the mechanism of processes such as those encountered in many experiments where slow motion or/and thermal equilibration is possible {e.g. in cool jets, cold matrices and solution). Under these conditions semi-classical dynamics yield, substantially, the same mechanistic information as from a topological investigation of the PES,[36, 37] because its topography is expected to play the dominant role in determining molecular motion (see the MEP line versus the real trajectory in Scheme 2). 3. COMPUTATIONAL TOOLS Modelling techniques used in studying photochemical reactions are limited to those that can properly describe excited states. In principle, every method that allows for a correct evaluation of the excited states energy, gradient and Hessian with respect to the geometrical coordinate of the reactant may be used for this purpose. The ab initio CASSCF[14] method is one of the main multi-reference wavefunction methods used for locating excited state minima, transition states and crossing points and for computing MEPs. In fact, it permits the analytical evaluation of gradients and second derivatives ultimately yielding an efficient mapping of the topography of excited state potential energy surfaces. Furthermore it is available in widely distributed software packages such as Gaussian [38] and MOLCAS.[39] In the applications described below (i.e. in Sections 5 to 7) it has never been necessary to go beyond the 6-31G* basis set as, for the investigated neutral or cationic systems, it produces reliable results. Nevertheless, as mentioned above, when accurate energies are required, a treatment accounting for the effect of dynamic electron correlation via a multireference-MP2 method such as CASPT2 [10, 11] (that takes the CASSCF wavefunction as the zeroth-order wavefunction) has been employed {i.e. errors within 3 kcal moH). In conclusion, in the application presented below we have always employed the ab initio CASPT2//CASSCF/631G* methodology. Standard methods for molecular structure optimization of stationary points as well as for MEP (e.g. the intrinsic reaction coordinate (IRC) method)[40] are employed for the mapping of both the ground and the excited states. However, description of the crossing region requires special methods as two potential energy surfaces become degenerate and the gradient and Hessian cannot be unambiguously computed. These tools are currently available in standard software packages such as Gaussian 03[38] and will not be presented here since they have recently been discussed in previous reviews.[41, 42] In contrast, below we will focus on the branching of the photochemical reaction path occurring upon decay from a higher to a lower laying state. The inter-state nature of such paths requires special methodologies to locate the energy valleys describing the relaxation process (e.g. the ground state relaxation occurring after the decay at the crossing). Methods for computing relaxation paths starting from a crossing point (or, more generally, from non-stationary points) are still matter of research and to our knowledge are not yet distributed.
199
Fig. 2. (a) Model PES showing a transition state (TS) connecting a reactant (R) to a product (P) in standard reactivity, (b) PES for a "model" elliptic conical intersection, and the corresponding energy profile (as a function of the angle a) along a circular cross section centred on the conical intersection point and with radius d. (From ref. [28])
3.1. Locating Relaxation Paths from a Conical Intersection. As mentioned above, an accessible conical intersection forms a structural bottleneck that separates the excited state branch of a photochemical reaction path from one or more ground state branches connecting the excited state reactant to one or more ground state products. The number and nature of the products generated following decay at a surface crossing will depend on the population of such branches each one corresponding to a different relaxation path. We have recently implemented a gradient-driven algorithm[37, 43] to locate and characterize all the accessible branches via the calculation of the initial relaxation directions (IRD) departing from a single conical intersection point. The MEP starting along these relaxation directions correspond to the possible relaxation paths which, in turn, locate the ground state valleys developing from the intersection region and comprising the energy minima of the corresponding photoproducts. Thus, by connecting the excited state and ground state paths a full description of the photochemical process from energy absorption to photoproductformation may be accomplished. The MEP connecting the reactant (R) to the product (P) of a thermal reaction is uniquely defined by the associated transition structure (TS). The direction of the transition vector {i.e. the normal co-ordinate corresponding to the imaginary frequency of the TS) is used to start a MEP computation. One takes a small step along this vector xi (shown in Fig.
200
2a) towards R or P and then follows the MEP connecting this point to the product or reactant well. The small initial step vector defines the IRD towards the product or reactant. This procedure cannot be used to find the IRD for a photochemical reaction since, as discussed above, a conical intersection is a "singularity" and there is no such unique direction for this first step (i.e. a frequency computation cannot be performed at a conical intersection point). The general situation is illustrated in Fig. 2b for an elliptic cone (i.e. linear approximation). In this case, two steep sides exist in the immediate vicinity of the apex of the cone. It is thus obvious that there are two preferred directions for downhill motion along these steep sides of the ground state cone surface. As one moves away from the apex along these steep directions, real reaction valleys eventually develop (leading to the final photoproducts minima). A simple procedure for defining these directions involves the computation of the energy profile along a circular cross-section centred on the vertex of the cone. This energy profile is given in Fig. 2b as a function of the angle a and for a suitable choice of the radius d. It can be seen that the profile contains two different energy minima. These minima (Mi and M2 in Fig. 2b) uniquely define the two IRDs (IRDi and IRD2) from the vertex of the cone. The two steepest descent lines (in mass-weighted co-ordinates) starting at M, and M2 define two MEPs which describe the relaxation processes in the same way the transition vector Xj (see Fig. 2a) defines the MEP connecting reactants to products. Thus, while there is no analogue for the transition vector in conical intersections, the simple case of an elliptic cone shows that the IRD are still uniquely defined in terms of Mi and M2. At this stage, one should notice that while the IRD from a TS connects the reactant to the product, there are two distinct IRDs from an elliptic conical intersection leading to two different photoproduct valleys (where one of these photoproducts may actually correspond to the original reactant). Although this model of the potential energy sheets at a conical intersection point is not general enough to give a correct description of all the relaxation paths encountered in a real system, the ideas introduced above can be easily extended to a full «-dimensional space search surrounding a conical intersection point, by replacing the circular cross-section with a (hyper)spherical cross-section centred at the vertex of the cone. Hence, locating stationary points on the n-\ dimensional hypersphere involves constrained geometry optimization, in mass-weighted co-ordinates, with a "frozen" variable d (i.e. the radius). The IRD is then defined as the vector joining the starting point to the optimized hyperminimum. The full mathematical details have been presented elsewhere.[43] We must emphasize that the procedure outlined above is designed to locate the points where the relaxation paths begin (i.e., they define the IRD). Once these points have been found for some small value of d, then one must compute the associated MEP (i.e. the relaxation path leading to a ground state energy minimum) as the steepest descent line in mass-weighted Cartesian coordinates, with the IRD vector defining the initial direction to follow. The standard IRC method [40] can be used for that purpose. As a consequence, the approach outlined above provides a systematic way to find the MEP connecting the vertex of the cone to the various ground state photoproduct wells.
201 Although this methodology has been presented for the case of ground state reaction paths departing from a conical intersection, it may also be used to locate paths beginning from a different non-stationary point, for example to follow the initial excited state relaxation channel starting from the FC point on the spectroscopic excited state. In conclusion, above we have discussed a two-step strategy for the computation of the full photochemical reaction path. Within the first step, keeping to the MEP from the FC region, key structures like minima and CIs can be located. In the second step, the MEPs describing the So relaxation process from the CIs (determined in the first step) are calculated with the strategy outlined above and the paths leading to the various photoproducts determined. 4. THE PROCEDURE Below we describe the protocol most commonly adopted by our group to approach a photochemical problem. The most common technical problems (and related solutions) are also discussed. 4.1. Choice of the Model. A chemist willing to perform accurate (e.g. CASPT2//CASSCF) computations frequently deals with the overwhelming issue of molecular size: it is often necessary to reduce the size of the target molecule to make the computation feasible. In order to ensure a qualitatively correct description of the reaction, this operation has to be done with great care. On the other hand, if the model has been carefully chosen, it can still catch the essential features of the reaction mechanism even if the observed excitation energies and barriers may not be accurately reproduced. A first rule of thumb is to maintain the atoms that are directly involved in the reactive process in the model. Furthermore, atoms that may affect the reactivity indirectly (i.e. through steric or electronic effects) have also to be considered. In photochemistry it is usually a good approximation to begin the study by considering the isolated chromophore moiety. This is the molecular sub-unit responsible for light absorption and, in general, the reaction is mainly driven by its reactivity. For instance, bulky alkyl groups may often be replaced with smaller methyl groups without loosing the molecule electronic properties and reaction mechanism. Of course, if bulky groups stabilize the "native" conformation of the reactant such an approximation becomes not applicable. A possible solution is to constrain the molecule in the desired conformation by freezing some geometrical parameters. A better solution would be describing the steric effect with minimal basis set computations or by adding to the total energy the contribution of suitably parametriezed "classic" MM potentials which mimic the encumbering effects[44]. Electronic effects are much more difficult to simulate and, in general, groups showing important contributions of this type cannot be eliminated from the model. Notice that, in all cases the correct/best choice of the molecular model is strongly influenced by the chemical knowledge of the chemist.
202
The computational investigation of the photochemical cis-trans isomerization of RPSBs[28, 45-50], also treated in Section 5, provides a good case study for the choice of the model. The retinal chromophore is both a conjugated poly ene and a protonated Schiff'base. In order to keep its qualitative features in a reduced model, we have to preserve both organic functions. In Scheme 3 we show how retinal models 1, 2, and 3 satisfy these requirement to an extent that depends on the nature and accuracy of the computational results one wish to obtain. While model 1 is very short and lacks of any alkyl substituent, it still preserves the cis configuration for the central double bond and can be seen as a minimal model for 11-cis RPSB. This model allows for very expensive calculations including ab initio CASSCF dynamics computations.[51] The larger model 2, may be used in studies dealing with the problem of competitive cis-trans isomerization of adjacent double bonds. Finally, model 3 represents the most realistic reduction of the target molecule. In fact, experimental data show that in solution (and also for the chromophore in Rhodopsin)[52-54] the |3-ionone ring is twisted by about 60°, making the conjugation of the final double bond almost negligible. Thus, we expect model 3 to deliver energies and reaction paths closer to that of the native molecule. EXCITED STATE DYNAMICS
^ H H
,
Target Molecule
,
STEREOSELECTWITY
2
H
ISOMERIZATION MECHANISM and ENERGETICS
ii
Scheme 3 (Adapted from ref. [28])
4.2. Choice of the Active Space. Selecting the CASSCF active space is not a trivial task. As for the choice of the model the correct/best choice of the active space is often driven by the chemical knowledge of the chemist. First of all, in order to describe the entire reaction path, the active space must not change along the reaction coordinate and thus the selection requires some valuable hypothesis for the complete reaction mechanism. Once the mechanism is defined the chemist must, in principle, include in the active space the orbitals and electrons that may possibly contribute to
203
the description of the different electronic states and reaction steps involved in the reaction. Some general guidelines for selecting the active space can be the following: if covalent bonds are broken or formed, the orbitals (bonding and antibonding) describing the bond must be included; if we are interested in the reactivity of a specific excited state, or if we think that other states may become important as the reaction proceeds, we have to be sure that the orbitals required for their description are included; The most important orbitals can be found in the highest occupied (HOMOs) and in the lowest unoccupied (LUMOs) orbitals resulting from a low-level (HF or MP2) calculation. Since the number of configurations generated in a CASSCF wavefunction grows factorially with the size of the active space, there is a practical limit to the dimension of the latter. Our experience shows that, on modern workstations, the limit for a non-symmetric system carrying about 300 basis functions is reached with an active space of 12 electrons and 12 orbitals, while slightly larger spaces can be managed if the number of determinants can be cut when certain elements of symmetry are conserved along the reaction coordinate. As a result, very often happens that not all the desired orbitals can be included in the active space. If selecting the desired orbitals is a difficult task, it is even more difficult to decide which of them may be excluded from the active space with a minor loss of accuracy. A general rule of thumb is that CASSCF (or RASSCF,[55]; see Section 6) active-space orbitals likely to have occupation numbers close to 2 or 0 can be transferred in the occupied or secondary spaces respectively with minor errors. Of course, in doing that, one must be convinced that the discarded orbitals will conserve such occupancies along the full reaction coordinate. The case of azobenzene, discussed in Section 6, nicely illustrates the rationale employed for activespace selection/reduction. 4.3. Vertical Excitation. Following a CASSCF optimization for the ground state reactant and known photoproducts and corrected the excited and ground state energies with single point CASPT2 computations, one should validate the method used comparing the results with experimental data (when the chosen model has not been investigated experimentally, one has to rely on experiments performed on similar molecules or on the comparison of relative quantities such as the reactant/product energy difference or the reactant/product shift in absorption maxima or previous computational results obtained with different methods). If the relative ground state stability of the reactant and products are not correctly reproduced, errors have been possibly made in the active space selection or in the choice of the model. Alternatively there are other important details that may have not been accounted for in the model such as environment (e.g. solvent) effects, stereo-electronic substituent effects, etc. The next step is to map the vertical excited states manifold above the reactant and (possibly) product minima. If no experimental evidence exists for the involvement of states of different spin multiplicity, then one can focus on states of the same multiplicity (e.g. usually singlet states for closed-shell organic
204
molecules). In order to locate the spectroscopic state (and characterize the other states that may be involved in the photochemistry of the system as well), it is important to consider several excited states in the computation and analyze them thoroughly. As discussed above single point CASPT2 calculations are applied to allow for quantitative comparison of the computed vertical and adiabatic excitation energies with the available experimental data. In order to simulate the transition probability between the electronic states we have to calculate the dipole transition moments (RASSI calculations[56, 57] implemented in MOLCAS 5[39] are used for this purpose) and evaluate the oscillator strength (/) associated to each transition. The analysis of the CASSCF wavefunction discloses the nature of the electronic state. In particular, the values of the configuration-interaction coefficients reveal the electronic configurations that, together with the analysis of the orbital shape, allow to "label" the state. For charged systems (as the RPSBs discussed above and in Section 5) useful information also comes from the analysis of the charge distribution of the different electronic states and from the change in dipole moments. These data are used to establish the nature {i.e. ionic vs. covaleni) of the different states.[45] Further insight may be achieved by Valence Bond (VB) analysis[58] of the CASSCF wavefunction. In this procedure each active space molecular orbital is "localized" on a single atomic centre. After the localization each configurationinteraction coefficients provides the weight of a specific valence bond description of the molecular system. In conclusion, the elements necessary to classify and characterize the states (including the non-spectroscopic ones) can be predicted and the methodology used validated by comparison with the observed absorption spectrum of the molecule. The observed band maxima should match the calculated energy gaps for the states with high values of/and the intensities (the area of the bands, to be precise) should be proportional to the corresponding/ values. Further, more complete spectral analysis is also occasionally done when vibrationally resolved electronic spectra are available (absorption, fluorescence, Raman, Resonance Raman etc.) and when the molecule is small enough to allow for the Hessian (i.e. vibrational frequency) computation. 4.4. Computing the photochemical reaction path 4.4.1. The excited state branch Once the nature and order of the states have been established, the MEP starting at FC on the spectroscopic state is constructed. Similarly the excited state mechanistic entities {i.e. minima, transition states and conical intersections) are located and interconnected between them and to FC. The resulting picture of the molecular rearrangement, including bond breaking and forming events, is described by looking at the progress from FC to one minimum and from this to another or to the conical intersections. This procedure must ultimately define the excited state branch of the photochemical reaction path. Since the FC point is a non-equilibrium structure on the excited state energy surface (see Fig. 3), the reactant structure usually undergoes a fast relaxation, which leads in general to a minimum or, for barrierless processes, to a conical intersection funnel giving access to
205 the ground state. The excited state relaxation path starting at the FC point is located by computing the IRD as described in Section 3.1. The full downhill path is then determined performing an IRC calculation.
Ionic State fspcctrcscopic stale)
Fig. 3. General evolution of a photochemical reaction. A, reactant. B, L photoproducts. C, D excited state intermediates. E, I excited state transition states. F pecked conical intersection. G, D avoided crossing regions. H sloped conical intersection.
When an excited state minimum (i.e. an excited state intermediate, see points C or D in Fig. 3) is reached by the IRC a more complex situation arises. In this case, the initial excited state relaxation indicate population of an excited state intermediate and one has to make reasonable hypotheses concerning the elementary reactions that may lead to other excited minima and/or deactivation funnels (e.g. a CI). These hypotheses must be computationally verified, searching all the possible transition state structures (Fig. 3, points E and I) that have to be overcome. In fact, there could be more than a single exit channel (in this case the minimum is a bifurcation point of the reaction path) and, in order to understand the reactivity, each possibility has to be explored. Indeed, the most probable reaction path can be established locating the possible competing transition states and comparing their corresponding barriers. In this way, some routes may be ruled out due to their unfavourable energetics (i.e. if a TS is much higher in energy than the others, the associated reaction is very unlikely to occur). If a detailed description of the reaction coordinate is required, IRC computations starting from the
206 energetically accessible TSs must be performed towards the original minimum and towards the unexplored part of the PES which may turn out to conduct to another minimum (Fig. 3, D) or to a picked[8, 59] conical intersection with the ground state energy surface(Fig. 3, F). In some cases, a sloped[&, 59] conical intersection (Fig. 3, H) may be reached from a minimum without overcoming a TS. 4.4.2. CASPT2 corrections and related mechanistic effects One major problem that may arise when the CASPT2 correction is applied to a set of MEP points whose structure is determined at the CASSCF level (i.e. when applying the CASPT2//CASSCF strategy), is that of state swapping. In other words after the correction an higher or lower excited state that, at the original CASSCF level, is well separated from the electronic state of the MEP, crosses such state. In principle, if a crossing with an upper state occurs, this should lead, in asymmetric systems, to an avoided crossing along the MEP (G in Fig. 3). On the other hand, a conical intersection is expected along the MEP if a crossing with a lower state occurs (Fig. 3, F and H). The difference is that in the first case the "true" (i.e. CASPT2 corrected) reaction coordinate may deviate with respect to the computed one to avoid the crossing region {i.e. in the case of a CI, the lower tip of the double cone) thus allowing the mixing of the two states and passing through an avoided crossing. Such an avoided crossing is thus characterized by a smooth change of the wavefunction along the MEP. However, in the second case the "true" (i.e., again, CASPT2 corrected) reaction coordinate is likely to cross the intersection seam (i.e. in the case of a CI, going from the upper to the lower part of the double cone). In this case the wavefunction does not change along the corrected MEP (see also Chapter 1). To correctly trace this behaviour, CASPT2 single point refinements have to be done on selected points along the computed MEP both on the reactive state, and the lower/higher lying states, if they are close in energy. In fact, as pointed out above, within the CASPT2//CASSCF strategy the CASSCF method is assumed to describe correctly the reaction coordinate (i.e. the shape of the potential energy surfaces) but it may fail evaluating the energy gaps (and therefore the correct energy order of the relevant states involved). For example, if the state right above the reactive one is, say, less than 0.5 eV at the CASSCF level, it might happen that the CASPT2 correction leads to an inversion in the energy order of the original states. If this is the case, then a crossing emerges (missed by CASSCF), and it is the state above (at the CASSCF level) that has to be followed (i.e. optimized by MEP computations) after this crossing point. Fig. 4 exemplifies this situation: the dotted line represents the upper state as it results from CASSCF, while the solid grey line shows the CASPT2 corrected state. After the scaling, a crossing arises (this may be actually an avoided crossing as discussed above in this same section. However, for practical reasons, one usually still describes this situations in terms of a real crossing) and the dark arrows represent the erroneous path (i.e. the erroneous reaction coordinate) that would be followed if the path were entirely on the state depicted in dark. With regard to the subject of Section 4.4.3, in such cases, the starting point for the relaxation channel computation on the lower lying state would be different than the one
207 predicted by the CASSCF level. The point of CASPT2 crossing should be chosen as the starting point for the relaxation path search. An analogous situation may also arise with a lower lying state (e.g. the ground state) where an early CI (i.e. a deactivation funnel) could emerge upon CASPT2 correction. This would lead to an earlier decay with respect to the one predicted at the CASSCF level.
CASSCF profile CASPT2 profile —•
Right path Wrong path
Fig. 4. Exemplification of the switching of two states originating by the CASPT2 correction. The dark arrows represent the wrong reaction path followed if the MEP would be calculated at the CASSCF level for the bottom state.
4.4.3. The ground state branch By means of the procedures described above, the excited state PES is mapped until a decay point with a lower state (usually with the ground state) is reached. This point could be a conical intersection. From the decay point, the approach followed for mapping the relaxation path on the ground state does not substantially differ from that followed for the excited state. All the possible channels departing from the CI on the ground state have to be systematically located and mapped (with the methods described in Section 3.1) until the minima corresponding to all photoproducts (or reactant) are reached. When no conical intersections are located, the decay point can be identified as an excited state minimum (Fig. 3, C and D), where competing radiative (i.e. fluorescence) and non-radiative processes may occur, depending on the Si-So energy gap.[23] Even complicated photochemical processes can be fully investigated by following this procedure. An example of that is provided by the complex network of reaction paths elucidating the photochemistry of cyclooctatetrane (COT) and presented in Section 7.
208 4.5. IRC Calculations. The step size of an IRC calculation[40] defines the distance between two consecutive points along the MEP (i.e. computed in terms of the steepest descent path in mass-weighted coordinates). Its value has to be carefully chosen: if it is too large the calculated path may deviate considerably from the real path, while if it is too small the calculation time required may be exceedingly long. In excited state calculations (as well as in diradical chemistry) the IRC calculation reaches very flat potential energy surface regions. In these cases, the algorithm may fail to converge or converge in a misleading direction. Two alternative solutions may be used. The first is to run a series of hyperminimum optimizations (using the IRD methodology discussed in Section 3.1), centring each new optimization in the hyperminimum located by the previous one. This procedure, that requires a tentative direction in input, seems more robust than the standard IRC computation that requires a starting Hessian matrix (since the Hessian matrix gets updated with the forces, accurate forces are critical for a successful run). In case this approach fails, a more drastic solution is to perform a relaxed scan along the molecular mode that is supposed (i.e. according to the chemical sense) to dominate the reaction coordinate, until the flat region is left behind. In any case, one should consider that when the MEP gets in into a flat region, usually a flat energy minimum is in the vicinity. A standard geometry optimization, with strict convergence parameters, easily shows if this is the case. Minima are important structures to be located along the MEP, as they represent basins collecting molecular population. 5. RPSB: FROM QUALITATIVE TOWARDS QUANTITATIVE INFORMATION The results of an application of the computational strategy outlined above are summarized in Fig. 5. At the bottom left of the figure we report the photochemical reaction path of the "minimal" retinal chromophore model 1 (cis-CsH()NH2+) along the MEPs connecting the FC structure to the So trans and cis product.[45, 47, 49] The solid circle on Si indicates the IRD point (see above) defining the starting point for the excited state MEP computation. The MEP describes the photoisomerization of the central double bond occurring along a barrierless reaction path developing entirely along the Si spectroscopic state (i.e., an ionic 1BU-Iike state) potential energy surface. Along this path, the energy difference between Si and S2 (i.e. a covalent 2Ag-like state) states is large (>25 kcal mob1). As a consequence, S2 is never involved in the reaction. Moreover, evolution of the charges along the reaction coordinate (see bar diagrams in Fig. 5) shows an increasing ionic character for the Si state consistently with the increasing S1-S2 energy separation along the MEP (that leads to a reduction in state mixing). The Si MEP ends at a Si/So conical intersection. This intersection has a-80° twisted central double bond which provides a route for efficient radiationless decay and non-adiabatic cis -» trans isomerization. Interestingly, as illustrated on the right side of Figure 5, the CI structure corresponds to a twisted intramolecular charge transfer (TICT) state: a single itelectron has smoothly migrated, along the Si MEP, from the left CH2-CH-CH- allyl fragment to the right -CH-CH-NH2 fragment and, consequently, the positive charge is almost
209 completely delocalized on the left allyl fragment. At the conical intersection, the electron shifts back to the left side and the original So charge distribution is restored (see top inset in Fig. 5). Starting from the crossing point we have located two IRD on So (see the two solid circles close to the CI), where two corresponding relaxation paths (i.e. MEPs) begin. The first path is a continuation of the excited state path and terminates at the all-trans CsHgNI-^ photoproduct well. The second path describes the back-formation of the reactant. In summary, such a computational approach has lead to a detailed description of what happens in model 1 along the reaction coordinate from absorption to photoproduct formation. The photoisomerization process is thus fully characterized.
right-ha IF
0.0
5.0
10.0
15.0
MEP co-ordinate
(a- u. )
Fig. 5. Energy profiles along the three MEPs describing the relaxation from the Franck-Condon (FC) and conical intersection (CI) points of model 1. Open and full squares define the excited (1BU-Iike) and ground state branches of the cis —» trans photoisomerization path, respectively. Full triangles define the ground state cis back-formation path. Open circles show the dark (2Ag-like) state energy along the excited state branch of the photoisomerization path. The structures (geometrical parameters in A and degrees) document the geometrical progression along the photoisomerization path, while the bar diagrams describe the change in the So. Si and S2 Mulliken charges (as % unit of the net positive charge) at the C=C-C- (left) and -C-C=N (right) fragments. The top box displays the Ji-electrons cloud at the CI for the excited and the ground state. (Adapted from ref. [28])
210 The same computational strategy has been applied to the study of the photoisomerization of longer RPSB models[46, 48, 50, 60] such as 2 and 3, leading the same qualitative results (See Fig. 6-7). Again a barrierless MEP with an ionic (charge-transfer) character leads to a Si/So CI point corresponding to a TICT state. However, in contrast to 1, model 2 allows for the investigation of competitive photoisomerization paths (whose presence is of interest for the investigation of the solution photoisomerization of RPSBs). However, model 3 of Scheme 4 and Figure 7 is the best candidate when accurate energetics and comparison with experimental data is the final goal of the study, as discussed in Section 4.1.
4
6
8 12
MEP co-ordinate (a. u.)
14 I eft-ha If
right-half
Fig. 6. Energy profiles along the MEPs describing the two competing (open and full squares) excited state (1BU-Iike) isomerization paths from the Franck-Condon point (FC) to the decay points (conical intersections) CI, and CI2 of model 2. The relaxed planar stationary point is labelled SP. Open circles show the dark (2Ag-like) state energy along the Si branch of the favoured photoisomerization path. The structures (geometrical parameters in A and degrees) document the geometrical progression along the two paths, while the bar diagrams describe the change in the So. Si and S2 Mulliken charges (as % unit of the net positive charge) of the C=C-C=C-C- (left) and -C-C=N (right) fragments. (From ref. [28])
Another interesting feature emerges, which is common to all the models: the Si reaction co-ordinate along the computed minimum energy path is curved, being consecutively dominated by two different perpendicular modes. The first mode is totally symmetric (preserving the planarity of the system) and drives the initial dynamics (< 50 fs)[61] out of the Franck-Condon point through a concerted double-bond stretch and single-bond compression
211 process (involving C-C bond order inversion). The second mode is asymmetric and dominated by a cis-trans isomerization motion: only after that the initial planar stretching has occurred, the reaction coordinate changes sharply its direction and this second asymmetric mode gets populated. Analytical frequency computations confirm the structure of the Si potential energy surface illustrated in Fig. 8a-c: around the FC point a steep valley exists with a slope (i.e., a gradient) driving the system along a planar backbone rearrangement. This reaction channel changes the shape on its way to the relaxed planar stationary point (SP), and the valley evolves (via an inflection point) into a ridge: this provides the driving force for the reaction coordinate sharp turn along the asymmetric twisting mode. Thus, for example, the two competing isomerization paths found in model 2 (see Fig. 6) have a common initial part, while the MEP bifurcates (see Fig. 8b) later on. Initial relaxation along a totally symmetric mode followed by motion along non totally symmetric (twisting) modes seems to be a general feature of the excited state behaviour of these and closely related compounds. [46]
4
6
8
MEP co-ordinate (a. u.) Fig. 7. Energy profiles along the S, reaction coordinate of the PSB11 model 3 (open squares) from the Franck-Condon point (FC) to the decay point CI (conical intersection). Open circles and diamonds show the S2 and So energy cross-sections along the same coordinate. The structures (geometrical parameters in A and degrees) document the progression of the molecular structure along the coordinate, while the bar diagrams describe the change in the So. Si and S2 Mulliken charges (as % unit of the net positive charge) of the C=C-C=C-C- (left) and -C-C=C-C=N (right) fragments. The top-left inset displays the observed and simulated RR spectra for PSB11 and model 3, respectively. (Adapted from ref. [28])
212 In spite of the similarities described above, Fig. 8 illustrates a striking difference between the short (1) and the longer (2 and 3) RPSB models: an increasing energy plateau along the photoisomerization path is observed as the length of the conjugated chain is extended (see the insets in Fig. 5-7). Accordingly, while the qualitative features of the Si energy surface remain the same for all the models, a difference in the dynamics is expected as the chain length increases: due to the reduced steepness along the twisting mode, a retardation in the photoisomerization (i.e., a longer Si lifetime) is expected for the longer models.[46] In other words, the shape of the Si energy surface and the two-mode photoisomerization coordinate suggest the dynamic behaviour illustrated in Fig. 8, where a metastable species performs many skeletal oscillations along an energy plateau before the reactive torsional mode gets effectively populated (i.e., intramolecular vibrational energy redistribution has to occur). This metastable species may be assigned to the "fluorescent-state" observed for RPSB in solution.
<=>
Energy
Trajeaory
Fig. 8. Shape of the F C ^ S P region of the Si energy surface for 1 (a), 2 (b), and 3 (c). The structures of the relaxed planar stationary points (SP) are shown (geometrical parameters in A and degrees). (From ref. [28])
Validation of 3 as a realistic model for the 11 -cis retinal chromophore (PSB11) of rhodopsin comes from the energy and charge data computed along the photoisomerization path. The computed absorption maximum (482 nm at planar FC) and fluorescence maximum (594-614 nm, depending on the selected point on the MEP along the energy plateau) compare reasonably well with the observed values in hexane solution (458 and 620 nm respectively)
213 and suggest that the general structure of the potential energy surface for the isolated cation 3 may be similar to that of PSB11 in solution.[50] Moreover, the computed So-Si dipole moment change |A,u| (14.0 Debyes) matches very well with the observed value (12.0 Debyes),[50] and support the strong ionic character of the Si state versus the covalent nature of So and S2. A further evidence for the high quality of model 3 comes from a recent simulation of the resonance Raman (RR) spectra. [60] As it has been shown, the computed frequencies and RR intensities are in strong agreement with those observed for PSB11 and its isotopomers in solution, this supporting the high quality of the simulated Si force field (see the top-left inset in Fig. 7). Fig. 8 provides also a clear example of what has been stated in Section 2.4. The MEP does not travel through the optimized planar stationary point SP. In contrast, due to the negative curvature at this point (i.e., it is a transition state), deviates from the planarsymmetric path and is funnelled toward the crossing point along the active twisting mode, before reaching SP. This situation is particularly evident for big negative curvatures such as in 1 (Fig. 8a). In 2 and (even more) 3 (see Fig. 8b-c), the flatness of the energy surface at SP accounts for an initial symmetric MEP that evolves (via a sharp-90° curve) along the asymmetric twisting mode only in the close vicinity of SP. Thus, while SP allows to characterize the topology of the Si energy surface, the system does not travel through this point and the information we get from MEP mapping and geometry optimizations are complementary. The same is for the funnel point: the optimized lower-energy crossing point is different from the conical intersection structure intercepted by the MEP. While, in general, the difference (both in energy and geometry) may be not too large, the point where decay actually occurs is effectively the latter. The computational results illustrated above provide information on the radiative to kinetic energy interconversion (i.e., photoisomerization) for RPSBs. These results, in concert with very recent experimental evidences, call for a revision of the models previously proposed for the primary event in rhodopsin proteins, and support a two-state/two-mode model for the photoisomerization of retinal chromophores.[50] 6. AZOBENZENE PHOTOCHEMISTRY 6.1. Active Space Selection Azobenzene (Ab) offers an example of how unaffordable large active space can be reduced keeping an acceptable description of the photochemical reaction path. Here we are interested in studying the photoisomerization process involving the central N=N double bond, which takes place on the Si (n-»jr*) state.[62-65] Selection of all n orbitals would result in a 14 electrons in 14 orbitals active space that is far too large to be handled by the available implementations of the CASSCF method. In addition, since the Si excited state has an m:* nature, the nitrogen non-bonding orbitals and electrons (i.e. the two lone-pairs) responsible of this excitation must be included yielding an even larger 18 electrons in 16 orbitals active space (see Scheme 4). Furthermore, along the torsion mechanism, the nitrogen lone pairs and
214 the 71/71* N=N double bond orbitals belong to the same symmetry and may overlap and mix. In order to make the computation feasible, one has to discard at least four orbitals and four electrons yielding a 14 electrons in 12 orbitals active space. As discussed in Section 4.2, a first rule for the reduction of the active space requires the removal of the orbitals with occupation closer to 2 and 0 (for the electronic states of interest) as this leads to minor errors. Unfortunately, for Ab even a single CASSCF calculation with the "chemical" active space (i.e. with 18 electrons and 16 orbitals) is impossible. In this case one may proceed by estimating the occupancy at the less expensive (but also less accurate) restricted active space (RASSCF) [55] level. In contrast with the CASSCF level, the RASSCF is defined by a wavefunction containing only a subset of all possible electronic configurations generated by the same 18 electron and 16 orbital active space (i.e the active space is not "complete" anymore). In particular, for Ab we have constructed a wavefunction containing the configurations corresponding exclusively to single, double, triple and quadruples excitations for both So and S|. In Scheme 4, the 16 Ab active orbitals used in RASSCF calculations are listed with the dashed line comprising the reduced (14 orbitals) active space for CASSCF. In Fig. 9a we report the occupancies and shapes of the most and least occupied RASSCF orbitals. Although the nonbonding orbitals {i.e., the lone pairs) are more occupied than the 711 and 7C2 benzene "sandwiches" in So, their occupations decrease in the Si state. Ag
Au
Bu
E-Azobcnzenc Target Active Space Reduced Active Space
Scheme 4
Since the RASSCF computation indicates that the benzene "sandwiches" (711 and 712) and "anti-sandwiches" (71 13 and n 14) are well-localized and show similar occupations (and the higher and lower occupations, respectively!) in both chemically relevant states, we can
215 move them outside the original select 18 electrons and 16 orbitals active space. The generated 14 electrons and 12 orbitals active space makes a CASSCF possible. Albeit the shapes of the excluded orbitals have changed (see Fig. 9b for the n\ and 712 orbitals), they are still localized on the benzene moieties. Since these orbitals are not uniformly distributed over the entire benzene moieties (they are mainly localized on eight carbon atoms, see Fig. 9b), a little unbalanced description of the bond distances of the aromatic rings may arise. Nevertheless, since the excluded orbitals have no component on the N=N fragment, one can hope that the "chemically-important" part of the molecule, as well as the N=N bond isomerization, can be correctly described.
So 1.998019 S, 1.056887
50 1.963582 51 1.971944
So 0.034000 S, 0.027887
So 1.996863 S, 1.942373
So 1.963229 S, 1.972556
So 0.033468 S, 0.027711
71 13
(b) Fig. 9. (a) RASSCF most and less occupied orbitals with So and Si occupations. The other active orbitals occupation is between 1.93 and 0.07 for So, and between 1.95 and 0.07 for Si; (b) The two (doubly occupied) excluded orbitals as they result after the CASSCF calculation with a restricted 14 electrons and 12 orbitals active space.
216
E (kcal/mol)
0 -1
85 º 84 92 º 94 º 108 º 120º
E-Ab (S0
º
i º
º
126.9 109.9 174.3º 133.4º
180º - 180º
! º
117.3 136.0º
º
125.2 128.9º
º
^
C N N C
º
128.8 115.1 128.8º 115.1º
NNC
trans
Fig. 10. Singlet (So and Si) and triplet (Ti) reaction paths (i.e. MEPs) for the E —» Z isomerization in azobenzene. Energy profiles are schematized for sake of clarity and have been scaled to match CASPT2 values. Open circles represent Si/So CIs: CItOrs-i and CIjnv structures (geometrical parameters in A and degrees) are reported. The horizontal axis represents the CNNC torsion coordinate. Values for NNC bending angles are also reported. The paths of Ti and Si radiationless decays are shown. (Adapted from ref. [66])
6.2. Photoisomerization With the selected 14 electrons and 12 orbitals active space, a detailed study of the So, Si and Ti potential energy surfaces in the isolated Ab molecule has been performed to identify the most efficient decay and isomerization routes.[66] Using quantum chemical methods, we have computed MEPs, transition states and Si/So conical intersections leading to the Si deactivation and to cis-trans photoisomerization. These results are collected in Fig. 10. On Si we have found only one transition state TStOrs(Si), where a 60° torsion about the CNNC dihedral and an activation energy of only 2 kcal moH, from the trans (E) isomer, are required. The lowest energy CI (CItOrs-i), lying also 2 kcal mol"1 above the Si minimum, is found on the torsion pathway for a CNNC angle of 94°. From the structure of the lowest energy CIs (CItOrs-i) one can see that torsion is the most important coordinate for the Ab
217 photoisomerization (as opposed to the notion that the photoisomerization of isolated Ab in the Si(nji*) state may also occur along the inversion coordinate)[65, 67] and explain the short lifetime of S|. At a qualitative level, the barrierless versus the barrier-controlled twisting path accounts for the slower decay of the E compared to the cis (Z) isomer,[68] as well as for the observed higher yield of the Z -» E with respect to the E -> Z photoisomerization process. [69] Cis connected with the inversion pathway have also been detected (CIjnv), but these are much higher in energy (ca. 25 kcal mol"1 higher than the Si minimum E-Ab(Si)) and cannot possibly represent competing deactivation channels for Si. The quantum yield of the Z -» E photosensitized photoisomerization is close to one, while the yield of the opposite process is about 0.01.[69, 70] These observations are accounted for by the fact that the lower energy So/Ti crossing is on the E side and is almost degenerate with the Ti minimum (see Fig. 10). Furthermore, calculations of So-Ti spin-orbit interactions reveal that they are large enough to justify the short lifetime of Ti (estimated lifetime -10 ps, thus explaining the lack of T| observation so far) and suggest that thermal isomerization can proceed via the non-adiabatic torsion route involving the So - Ti - So crossings with pre-exponential factor and activation energy in agreement with the values obtained from kinetic measures.[65, 67, 71] In conclusion, the computed reaction scheme unveils the preferred isomerization pathways for Ab in the So, Ti and Si states, and provides a consistent interpretation of the experimental results. 7. COT PHOTOCHEMISTRY An example of a rather intricate network of photochemical reactions (including bifurcations, branching, generation of more than one photoproduct, adiabatic and non-adiabatic paths, etc) that have been disclosed using the CASPT2//CASSCF strategy is provided by the photochemistry of the antiaromatic molecule cyclooctatetrane (COT), an eight-membered ring conjugated hydrocarbon.[72] A summary of the full reaction network is shown in Scheme 5, while a cartoon-like drawing of the computed MEPs is shown in Scheme 6. This also reports the relative CASPT2-corrected energies (with respect to the excited state intermediate COT*, grey numbers) for all the key-structures found. The computations show that evolution of photoexcited COT out of the FC region prompts an efficient radiationless decay of the spectroscopic S2 (*E) state into the dark and lower Si ('A|) state (i.e. the photochemically relevant state), leading to population of the planar D<% excited state intermidiate COT*. Non-adiabatic transitions to So appear to be controlled by two different tetraradical-type conical intersections (CIst and CIt>), directly connected to COT* by independent excited state reaction paths. CIst belongs to the family of the -(CH)3- kinked crossings found in smaller cyclic hydrocarbons and linear polyenes (This features three adjacent weakly interacting electrons plus one delocalized on a it-fragment).[8, 16, 36, 37, 46, 73] The decay channel corresponding to CIst prompts ground-state relaxation to both three (or four) membered rings (BIC1, BIC2) and cis^-trans isomerization (E-COT) (i.e., the typical photoproducts observed from -(CH)3- kinked intersections), [73] see Scheme
218 6. However, this channel is too high in energy to provide an efficient way for radiationless decay. The other conical intersection funnel (Clb) has only two unpaired electrons centred on single carbon atoms, plus two allyl radical moieties. The decay channel corresponding to Clb is lower in energy (due to the higher stability of the allyl radicals and to a decreased ring strain) and thus the favoured ground state relaxation paths originate from this point. Si/S0Clst Photochemical process (unfavored)
Scheme 5 (Adapted from ref. [72])
Both electronic and strain factors play a central role in determining the stabilization and selection of low-energy conical intersection funnels. In ref. [72] a crude model has been presented to predict the electronic structure of low-energy crossings in unsaturated cyclic hydrocarbon systems. Indeed, it has been shown[72] how the photochemical production of the primary photoproduct semibullvalene (SBV) as well as the double bond shifting are boosted by COT* deactivation through deactivation at Clb that, therefore, constitutes the locus for the ground state branching of the photochemical process. This result provides not only a mechanistic explanation for the experimentally observed[74] COT^SBV photoisomerization in the gas phase (thus correcting previously proposed mechanistic hypothesis),[75] but also rationalizes photochemical double bond shifting. In fact, both processes are controlled from the same conical intersection that interconnects excited state reaction paths with ground state relaxation paths. Thus, again, the main "chemical" result is that production of semibullvalene and double bond shifting are intrinsically bounded processes. This feature should be properly considered when designing double bond shift-based molecular switches or devices.
219 Thermal processes (i.e. thermal conversions between COT photoproducts, and reactions such as valence isomerization (VI), Cope rearrangement (CR), double bond shifting (DBS), ring inversion), have also been investigated in order to draw a comprehensive reactivity scheme (see Scheme 5), which rationalizes observations and embraces the complex network of photochemical (Si) versus thermal (So) reaction channels, and their interconnections (see Scheme 6).[72] *
UIHHLJM TRANSITION STATE Si'S 0 CI
COT
Scheme 6 (Adapted from ref. [72])
8. CONCLUSIONS Above we have provided the basis for the computation, analysis and characterization of the photochemical reaction path of an organic chromophore. We believe that, through the presented examples, we have also provided evidence that a description of the reaction, from energy absorption to photoproduct formation, is nowadays technically possible. The basic conceptual problems that need to be solved to apply the available quantum chemical tools to photochemical problems mainly concern the computation of MEPs and the characterization of the conical intersection funnels. As clearly shown in our works and pointed out in recent reviews,[7-9, 16, 27] conical intersections are well far from being an abstract feature of quantum chemistry, since they do represent key mechanistic elements in photochemical reactions. As a consequence conical intersections often provide a straightforward
220
interpretation of basic experimental features such as the branching of the excited state reaction paths and the reaction stereochemistry (determined by their structures). Moreover, it has been shown that the photoisomerization of RPSB involves a conical intersection. This is intimately connected to the process of vision and, therefore, to an everyday life photobiological event. Obviously, any mechanistic study undertaken using quantum chemistry methods requires considerable physical and chemical insight. As for a thermal reaction there is no robust method that will automatically generate all the possible mechanistic pathways that might be relevant, thus in excited state chemistry one still needs to apply the chemical intuition and make sensible hypotheses that can then be explored computationally. This may be difficult when more electronic states are involved. Anyway, as illustrated by the examples above, a rational approach is now available for these studies, that allow new interesting information to be collected, leading to an unprecedented detailed view of photochemical processes. ACKNOWLEDGEMENTS This chapter is based (in part) on a lecture course given by M. Garavelli at the University of Paderborn, Theoretical Physics Department, Germany, and at the University of Bologna, Department of Chemistry 'G. Ciamician', Italy. We are grateful to the students and participants for their questions. We are also pleased to acknowledge Dr. Marcella Ravaglia and Piero Altoe who have commented on various parts of the manuscript.
REFERENCES [1 ] K. E. Drexler, Engines of Creation: the Coming Era of Nanotechnology, Anchor Books Editions, USA 1986. [2] C. Dugave, and L. Demange, Chem. Rev. 103 (2003) 2475. [31 N. Tamai, and H. Miyasaka, Chem. Rev. 100 (2000) 1875. [4] Y. Yokoyama, Chem. Rev. 1000 (2000) 1717. [51 M. Irie, Chem. Rev. 100 (2000) 1685. [6J D. G. Truhlar, and M. S. Gordon, Science 249 (1990) 491. [7J W. Fuss, S. Lochbrunner, A. M. Muller, T. Schikarski, W. E. Schmid, and S. A. Trushin, Chem. Phys. 232(1998) 161. [8J M. A. Robb, M. Garavelli, M. Olivucci, and F. Bernardi, Rev. Comp. Chem. 15 (2000) 87. [91 W. Fuss, K. L. Kompa, S. Lochbrunner, and A. M. Muller, Chem. Phys. Phys. Chem. 101 (1997) 500. [ 10| K. Andersson, P.-A. Malmqvist, and B. O. Roos, J. Chem. Phys. 96 (1992) 1218. [11J J. J. W. McDouall, K. Peasley, and M. A. Robb, Chem. Phys. Lett. 148 (1988) 183. [121 B. O. Roos, Int. J. Quantum Chem. 14 (1980) 175. [13J B. O. Roos, P. R. Taylor, and P. E. M. Siegbahn, Chem. Phys. 48 (1980) 157.
221 [14J B. O. Roos, The Complete Active Space Self Consistent Field Method and its Applications in Electronic Structure Calculations, in Ab initio Methods in Quantum Chemistry - II, Lawley, K. P., (ed.), Wiley, New York, 1987, pp. 399-446. L15J P. E. M. Siegbahn, J. Almlof, A. Heiberg, and B. O. Roos, J. Chem. Phys. 74 (1981) 2384. [ 16] F. Bernardi, M. Olivucci, and M. A. Robb, Chem. Soc. Rev. 25 (1996) 321. L17J A. Natansohn, and P. Rochon, Chem. Rev. 102 (2002) 4139. [181 N. Hampp, Chem. Rev. 100 (2000) 1755. [ 19] K. P. Lawley, Ab Initio Methods in Quantum Chemistry - I, in Adv. Chem. Phys. John Wiley & Sons Ltd., Chichester, England, 1987. [20] K. P. Lawley, Ab Initio Methods in Quantum Chemistry - II, in Adv. Chem. Phys. John Wiley & Sons Ltd., Chichester, England, 1987. [211 D. R- Yarkony, J. Phys. Chem. 100 (1996) 18612. 122] D. R. Yarkony, Ace. Chem. Res. 31 (1998) 511. [231 M. Desouter-Lecomte, and J. C. Lorquet, J. Chem. Phys. 71 (1979) 4391. [24] A. Gilbert, and J. Baggott, Essentials of Molecular Photochemistry, Blackwell Scientific Publications, Oxford 1991. [25J M. Klessinger, and J. Michl, Excited States and Photochemistry of Organic Molecules, VCH Publishers, New York 1994. [26J J- Michl, and V. Bonacic-Koutecky, Electronic Aspects of Organic Photochemistry, Wiley & Sons, Inc. 1990. L27J F. Bernardi, M. Olivucci, J. Michl, and M. A. Robb, The Spectrum 9 (1997) 1. L28J M. Garavelli, F. Bernardi, M. A. Robb, and M. Olivucci, Int. J. Photoenergy 4 (2002) 57. [291 W. T. A. M. Van der Lugt, and L. J. Oosterhoff, J. Am. Chem. Soc. 91 (1969) 6042. [30J H. E. Zimmerman, J. Am. Chem. Soc. 88 (1966) 1566. [311 E. Teller, Isr. J. Chem. 7 (1969) 227. L32J J. Michl, Molecular Photochemistry 4 (1972) 243. [33] W. H. Miller, H. N. C, and A. J. E., J. Chem. Phys. 72 (1980) 99. [341 H. Eyring, J. Chem. Phys. 3 (1935) 107. [35J D. G. Truhlar, and B. C. Garret, Annual Rev. Phys. Chem. 35 (1984) 159. [361 M. Garavelli, B. R. Smith, M. J. Bearpark, F. Bernardi, M. Olivucci, and M. A. Robb, J. Am. Chem. Soc. 122 (2000) 5568. [37] M. Garavelli, P. Celani, M. Fato, M. J. Bearpark, B. R. Smith, M. Olivucci, and M. A. Robb, J. Phys. Chem. A 101 (1997) 2023. [38] Gaussian 98, M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R. Cheeseman, V. G. Zakrzewski, J. A. Montgomery, R. E. S. Jr., J. C. Burant, S. Dapprich, J. M. Millam, A. D. Daniels, K. N. Kudin, M. C. Strain, O. Farkas, J. Tomasi, V. Barone, M. Cossi, R. Cammi, B. Menucci, C. Pomelli, C. Adamo, S. Clifford, J. Ochterski, G. A. Petersson, P. Y. Ayala, Q. Cui, K. Morokuma, D. K. Malick, A. D. Rabuck, K. Raghavachari, J. B. Foresman, J. Cioslowski, J. V. Ortiz, B. B. Stefanov, G. Liu, A. Liashenko, P. Piskorz, I. Komaromi, R. Gomperts, R. L. Martin, D. J. Fox, T. Keith, M. A. Al-Laham, C. Y. Peng, A. Nanayakkara, C. Gonzalez, M. Challacombe, P.
222 M. W. Gill, B. G. Johnson, W. Chen, M. W. Wong, J. L. Andres, C. Gonzalez, M. Head-Gordon, E. S. Replogle, and J. A. Pople, Gaussian Inc., Pittsburgh PA, 1998. [39] MOLCAS, Version 5.0, K. Andersson, M. R. A. Blomberg, M. P. Fiilscher, G. Karlstrom, R. Lindh, P.-A. Malmqvist, P. Neogrady, J. Olsen, B. O. Roos, A. J. Sadlej, M. Schutz, L. Seijo, L. Serrano-Andres, P. E. M. Siegbahn, and P.-O. Widmark, Lund University, Lund, 1999. [40J C. Gonzalez, and H. B. Schlegel, J. Phys. Chem. 94 (1990) 5523. [411 J- B. Foresman, and F. Aeleen, Exploring Chemistry with Electronic Structure Methods, Gaussian, Inc., Pittsburgh PA 1996. [421 H. B. Schlegel, in Ab initio Methods in Quantum Chemistry, Lawley, K. P. (ed.), Wiley, New York, 1987, pp. 249-286. L43J P. Celani, M. A. Robb, M. Garavelli, F. Bernardi, and M. Olivucci, Chem. Phys. Lett. 243 (1995) 1.
[44] M. Garavelli, B. Frabboni, M. Fato, P. Celani, F. Bernardi, M. A. Robb, and M. Olivucci, J. Am. Chem. Soc. 121 (1999) 1537. [45] M. Garavelli, P. Celani, F. Bernardi, M. A. Robb, and M. Olivucci, J. Am. Chem. Soc. 119 (1997)6891. [46J M. Garavelli, F. Bernardi, M. Olivucci, T. Vreven, S. Klein, P. Celani, and M. A. Robb, Faraday Discuss. 110(1998)51. L47J M. Garavelli, F. Bernardi, P. Celani, M. A. Robb, and M. Olivucci, J. Photochem. Photobiol. A: Chem. 114(1998) 109. L48J M. Garavelli, T. Vreven, P. Celani, F. Bernardi, M. A. Robb, and M. Olivucci, J. Am. Chem. Soc. 120(1998) 1285. [491 M. Garavelli, F. Bernardi, M. A. Robb, and M. Olivucci, J. Mol. Struc. 463 (1999) 59. [50J R- Gonzalez-Luque, M. Garavelli, F. Bernardi, M. Merchan, M. A. Robb, and M. Olivucci, Proc. Natl. Acad. Sci. U.S.A. 97 (2000) 9379. L51] T. Vreven, F. Bernardi, M. Garavelli, M. Olivucci, M. A. Robb, and H. B. Schlegel, J. Am. Chem. Soc. 119(1997) 12687. [521 D. C. Teller, T. Okada, C. A. Behnke, K. Palczewski, and R. E. Stenkamp, Biochemistry 40 (2001)7761. [531 R. Rajamani, and J. Gao, J. Comp. Chem. 23 (2001) 96. L54J A. Albeck, N. Livnah, H. Gottlieb, and M. Sheves, J. Am. Chem. Soc. 114 (1992) 2400. [55] J. Olsen, B. O. Roos, P. J0rgensen, and H. J. A. Jensen, J. Chem. Phys. 89 (1998) 2185. L56J P. A. Malmqvist, and B. Roos, Chem. Phys. Lett. 155 (1989) 189. [57J P. A. Malmqvist, Int. J. Quantum Chem. 30 (1986) 479. [581 S. F. Boys, Rev. Mod. Phys. 32 (1960) 296. L59J G. J. Atchity, S. S. Xantheas, and K. Ruedenberg, J. Chem. Phys. 95 (1991) 1862. [60] M. Garavelli, F. Negri, and M. Olivucci, J. Am. Chem. Soc. 121 (1999) 1023. [611 S. Ruhman, B. Hou, N. Friedman, M. Ottolenghi, and M. Sheves, J. Am. Chem. Soc. 124 (2002) 8854. [621 G. Zimmermann, L. Y. Chow, and U. I. Park, J. Am. Chem. Soc. 80 (1958) 3528. [63] E. Fischer, J. Am. Chem. Soc. 82 (1960) 3249.
223 L64J E. Fischer, and S. Malkin, J. Phys. Chem. 66 (1962) 2482. [651 H. Rau, and E. Luddeke, J. Am. Chem. Soc. 104 (1982) 1616. [66] A. Cembran, F. Bernardi, M. Garavelli, L. Gagliardi, and G. Orlandi, J. Am. Chem. Soc. 126 (2004) 3234. [67] H. Rau, in Photocromism, Molecules and Systems, Dtirr, H., and Bounas-Laurent, H. (eds.), Elsevier, Amsterdam, 1990, pp. 165-192. [681 H. Satzger, S. Sporlein, C. Root, J. Wachtveitl, W. Zinth, and P. Gilch, Chem. Phys. Lett. 372 (2003) 216. [69] P. Bortolus, and S. Monti, J. Phys. Chem. 83 (1979) 648. [70] L. B. Jones, and G. S. Hammond, J. Am. Chem. Soc. 87 (1965) 4219. [71] E. V. Brown, and G. R. Grunneman, J. Am. Chem. Soc. 97 (1975) 621. [72] M. Garavelli, F. Bernardi, A. Cembran, O. Castano, L. M. Frutos, M. Merchan, and M. Olivucci, J. Am. Chem. Soc. 124 (2002) 13770. L73] P. Celani, M. Garavelli, S. Ottani, F. Bernardi, M. A. Robb, and M. Olivucci, J. Am. Chem. Soc. 117(1995) 11584. [741 N. J. Turro, J.-M. Liu, H. E. Zimmerman, and R. E. Factor, J. Org. Chem. 45 (1980) 3511. [75] H. E. Zimmerman, and H. Iwamura, J. Am. Chem. Soc. 92 (1970) 2015.
This Page is Intentionally Left Blank
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 Elsevier B .V. All rights reserved
VII. Computation of Reaction Dynamics in Photobiology
225
Mechanisms
and
Seth Olsen, Alessandro Toniolo, Chaehyuk Ko, Leslie Manohar, Kristina Lamothe, and Todd J. Martinez Department of Chemistry, Beckman Institute, and Center for Biophysics and Computational Biology, University of Illinois at Urbana-Champaign, Urbana, IL 61801 1. INTRODUCTION All three domains of life utilize light energy for various purposes.1 Photon-driven ion pumps such as halorhodopsin and bacteriorhodopsin are found in Archae. In the Eukarya, phytochoromes monitor the light environment and regulate plant growth and rhodopsins are the photoreceptors which initiate visual signal transduction. Photoactive proteins are also found in the Eubacteria, for example as receptors in phototaxis - movement in response to light. Many of the photoprocesses used by biology involve conversion of the initial photon absorption to a localized mechanical response on the femtosecond timescale. The protein environment often acts as an amplifier, transforming an Angstrom-scale mechanical motion into larger scale directed motion. A particularly striking example in this context is the photoactive yellow protein (PYP) of Ectothiorhodospira halophila, which is a photoreceptor used by the organism to avoid blue light. In this case, the initial absorption of a blue photon leads immediately to trans-cis isomerization of the PYP chromophore (~lA movement), which in turn leads to large-scale conformational change of the PYP protein («10A movement). Through a signal transduction pathway this leads to a change in flagellar motion, resulting in a reorientation of the organism and ultimately motion on the micron and larger length scales. Thus, the initial photon absorption leads to a cascade of amplified motions from the Angstrom to micron length scales. Modeling the entire process will require techniques spanning many orders of magnitude in both space and time. However, it all starts with the smallest time (fs) and space (A) scales which are now amenable to direct modeling, thanks to advances in both computational resources and theoretical methods. A complete understanding of photobiological processes would be desirable for purely scientific reasons. However, such understanding could lead to a much greater impact by enabling the design of optically-switched and/or optically-powered molecular machines. The exquisite control of light afforded by modern lasers on both ultrafast temporal and ultrashort spatial scales could in principle be transferred to molecules through such devices. Photobiology does not actually use this fine control - the light which is used for signaling or powering photoactive proteins is typically neither time-gated nor spatially focused. However, the first steps towards
226 rationally designed light-driven molecular devices will almost certainly benefit from understanding structure-function relationships in photobiological systems. As detailed elsewhere in this book, the modern picture of photochemistry centers on conical intersection points - molecular geometries where two or more electronic states are exactly degenerate.2"5 In general, these intersections are not points at all, but rather seams of dimension as high as N-2, where N is the number of internal degrees of freedom in the molecule. It is also worth noting that geometries where multiple electronic states are degenerate need not be strictly cone-shaped.6 For example, the electronic state degeneracy is lifted only in second order as the molecule is displaced from Renner-Teller intersections. Nevertheless, the tendency in the literature is to use the words "conical intersection" to describe these degenerate points unless it is known that the intersection is not conical. This is not unreasonable since almost all intersection points will be conical unless the molecular geometry has special symmetry. We will follow this practice in this article. The importance of conical intersection points comes from the ease of changing electronic state character at these points. This was recognized early on, but it was often thought that these intersection points would rarely be found at chemically-relevant energies unless their presence was dictated by symmetry.4 However, as should be abundantly clear from other chapters in this book, it is now recognized that the existence of low-lying conical intersections is much more the rule than the exception. Indeed, in many cases7"10 conical intersections are absolute minima on an excited state potential energy surface." Energetic considerations naturally shift the focus to "minimal energy conical intersections" or MECIs, which are defined as the lowest energy point on a seam of conical intersections. It is tempting to elevate MECIs to the status of "transition states" for photochemical reactions. After absorbing a photon, the molecule finds itself in an unfavorable geometry. It relaxes, but remains trapped on the excited electronic state until it reaches a conical intersection. At such point it rapidly "quenches" to the lower electronic state and descends to form ground-state products. A schematic reaction path illustrating these ideas is shown in Fig. 1. While the picture of MECIs as "transition states" for photochemical reactions is often qualitatively correct, one must be careful to note the possible differences between MECIs and the transition state of a ground state reaction. As noted above, the MECI can often be an absolute minimum on the excited electronic state. This is quite different from the transition state in a normal ground state reaction, which can only be reached by molecules that are much more energetic than the average molecule in the reactive degree of freedom. The typical assumption in ground state reactions that energy is statistically distributed prior to reaching the transition state is clearly suspect in the photochemical case. Dynamical effects can thus play a significant role in the outcome of a photochemical reaction - much more so than one usually finds in ground state reactions. Indeed, it can happen in photochemical reactions that the MECI is never reached because the molecules encounter a higher energy point on the conical intersection seam before
227
Z e S0/S1MECJ
a
o a.
Reaction Coordinate Fig. 1. Schematic potential energy curves along the reaction coordinate of a photochemical reaction. Important points along the reaction path are the Franck-Condon point (SoMinG) which defines the initial geometry of the molecule after photon absorption, possible local minima on the excited state (S]MinL) from which the molecule could fluoresce, the global minimum on the excited state (SiMinG) and the minimal energy So/Si conical intersection (S0/S1MECI). Often the SiMinG and So/SiMECI geometries are found to be identical, i.e. the global minimum on the excited state is a conical intersection. there is sufficient time to completely relax on the excited state PES. There is therefore clearly an important, even critical, role for explicit modeling of dynamics in photochemical problems. Turning back to photobiology, one of the key questions is how the protein environment modifies the fate of the excited chromophore. The first step in answering this question is determining how the chromophore responds to light in isolation and in condensed phases. These environments provide the reference with which the behavior in the protein environment may be compared. Does photobiological diversity stem from a diversity of chromophores or from a diversity of protein environments? What is the extent of and mechanism for influence of excited
228 state dynamics by protein environments? The development of techniques which can provide detailed answers to these questions including a realistic treatment of the chromophore and its surrounding environment are only now emerging. In this article we provide a rather selective overview of some of the methods involved and highlight some interesting results. 2. EXCITED STATES OF BIOLOGICAL CHROMOPHORES The first question which must be asked in photobiological problems is the intrinsic behavior expected from the chromophore. This question can be split into two parts according to a chronological view of the mechanism. At the earliest time, the question centers on the nature of the excited electronic state which is initially populated. At later times, one wants to know the low-lying MECI geometries which will dominate quenching behavior and provide the initial conditions for subsequent evolution on the ground electronic state. While still far from a trivial problem, there are a number of methods which can be used for the computation of vertical excitation energies and characterization of the excited state manifold in the Franck-Condon region. For molecules with less than ten atoms, there are many options for reliable ab initio methods. The basic issue which stymies their application to the large chromophores that are typically found in photobiology is a poor scaling with system size. For example, the equation-of-motion coupled cluster12 (EOM-CC) and multireference perturbation theory, e.g. CASPT2,' 3 methods can achieve 0.5eV or better accuracy in vertical excitation energies but both scale as N6 or worse, where N indicates molecular size. It is now possible to push these approaches to model chromophores of biological significance,14"1 but generally only at a limited number of geometries and with modest basis sets. Time-dependent density functional theory (TDDFT),19 which as usually applied does not involve time explicitly at all and should not be confused with dynamics methods, is an alternative method for excited electronic states which shows considerable promise. The key approximations in most implementations (beyond the usual ignorance of the exact exchange-correlation functional in DFT) are the adiabatic and linear response approximations. Within the Kohn-Sham framework, linear response restricts the model wavefunction used to generate the density and transition densities to a single determinant and all single electron excitations from it. The adiabatic approximation ignores the possible, in fact strictly speaking necessary, energy dependence of the exchange-correlation functional. In spite of these approximations, TDDFT often predicts vertical excitation energies with 0.5eV accuracy,20 as long as there is no charge transfer or Rydberg character in the excited state. Corrected functionals have been proposed which can improve the performance for Rydberg states, but charge transfer remains problematic22'23 with the currently available functionals. States with significant doubly-excited character are also not accessible to TDDFT in the usual adiabatic linear response approximations, although a recent proposal24'25 to include frequency dependence in the exchange-correlation functional (going beyond the adiabatic approximation) seems to be a promising avenue for improving this.
229 Although vertical excitation energies have an important role given their direct relationship to absorption maxima in electronic spectroscopy, they are only the beginning of the story. In determining reaction mechanisms, it is critical that the global shape of the excited state potential energy surface be well-reproduced. Especially the geometries of excited state minima and lowlying conical intersections need to be correct. Often, this requires a multi-reference treatment. This is especially easy to see around conical intersections, where the degeneracy of multiple electronic states makes any single-reference representation of the wavefunction arbitrary and unreliable. It is also important to note that doubly-excited states, which are usually dark and therefore of little interest in modeling absorption spectra, may become important after relaxation on the excited state. This makes techniques such as TDDFT, which are both single-reference and restricted to singly-excited states (relative to the ground electronic state) quite suspect. Nevertheless, in some cases TDDFT works quite well for the lowest singlet excited state.26 In Fig. 2, we show a comparison of CASPT2 and TDDFT (using the B3LYP functional) for an Si coordinate driving path obtained using SA-5-CASSCF(6/5).27 The molecule in this case is an analog of the chromophore in photoactive yellow protein (PYP). The agreement for So and Si is almost quantitative, although the S2 state is not well-predicted by TDDFT. Such agreement does not hold up at conical intersections,28 although one is encouraged that an accurate TDDFT applicable to photochemistry may be possible in the future.
230
2 r 1 0 >
-1
-2 -3 -4 -5 0
2
4
6
8
10
Mass-Weighted Distance / Angstrom * amu
1/2
Fig. 2. Comparison of TDDFT-B3LYP (dotted lines) and CASPT2 (solid lines) for the three lowest singlet electronic states of a PYP chromophore analog along a coordinate-driving path on S|. Geometries are obtained by minimizing the energy on Si using a SA-5-CASSCF(6/5) wavefunction subject to fixed torsional angle
231 consistent field (CASSCF) method is used, including all electronic configurations consistent with a number of active orbitals and electrons. This is not required, but again avoids any unintended bias towards one of a pair of degenerate states. Unlike CASPT2, which adds dynamic electron correlation to CASSCF through perturbation theory, the CASSCF method is sufficiently inexpensive to be applicable for geometry optimizations and even dynamics calculations.16'29 However, such calculations generally omit solvent molecules or surrounding protein residues. This is a clear deficiency in the context of photobiology, but is being remedied in the most recent work of several groups, which we discuss below. It is critical to recognize that the choice of active space in CASSCF calculations is somewhat arbitrary. Although in principle it is true that a larger active space is better, this is not always borne out by calculations. The CASSCF method does include a varying degree of dynamic electron correlation as the active space is increased. This is clear because CASSCF becomes full configuration interaction (CI) (and hence exact within the chosen basis set) in the limit of an active space including all electrons and orbitals in the molecule. However, one is in practice always far from this limit, and the CASSCF method is well-known to be a very inefficient way of including dynamic electron correlation effects. Thus, we take a pragmatic view that the active space should be chosen on energetic criteria. One useful way to determine an initial guess for the number of electrons and orbitals in the active space is to compute the vertical excitation energies using EOM-CCSD. From analysis of the EOM-CCSD coefficients for the low-lying electronic states, one can deduce a good guess at the required number of electrons and orbitals in the active space. Then, several calculations are carried out varying the number of electrons and orbitals around the EOM-CCSD predicted values. From each of these, a CASPT2 calculation is carried out. If the active spaces are all reasonable, the vertical excitation energies computed by CASPT2 will be nearly independent of the underlying CASSCF. From this procedure, one can collect a set of different active spaces which have the electronic states ordered correctly and also are apparently equally valid as judged by the resulting CASPT2 vertical excitation energy. Ideally, one continues the verification procedure by tracing out reaction paths on the excited state using both CASSCF and CASPT2 with the active spaces which remain under consideration. The CASSCF active space which gives the best agreement with CASPT2 is the one which is then used. The number of electronic states included in the state-averaging procedure can also be varied in this process, using the same general criteria that a reasonable CASSCF calculation will reproduce the features predicted by CASPT2. This type of procedure is much more effective at determining a robust active space than one based on chemical intuition, which often ignores the influence of dynamic electron correlation. When used in the context of dynamics, it is necessary to repeat this process several times. Essentially, one first determines which active spaces are reasonable at the Franck-Condon point as discussed above and then carries out a few dynamics simulations. From the observed dynamics, one can extract coordinates which are important and construct paths along which the CASSCF and CASPT2 potential energy surfaces can be compared. Ultimately, one may supplant this procedure with dynamics and/or optimizations using CASPT2 (or perhaps an improved form of TDDFT)
232 directly. However, until this is computationally practical, the above-outlined approach provides a means for obtaining semi-quantitative agreement between CASSCF and CASPT2, which then allows for extensive dynamics calculations addressing the photochemical mechanisms. In Fig. 3. we demonstrate the kind of agreement which can be achieved using this technique for a biological chromophore - the same analog of the PYP chromophore presented in Fig. 2. The agreement is semi-quantitative, providing justification for the use of the chosen active space in geometry optimizations, reaction path calculations and dynamics studies.
00
ft ^ ^ S
SA-5-CASSCF(6/5) 10
10
120
Mass-Weighted Distance / am u
12
Angstrom
Fig. 3. Comparison of CASSCF and CASPT2 for analog of the PYP chromophore. The geometries in the chosen path are chosen as in Fig. 2. The active space was chosen according to the procedure detailed in the text. Unfortunately, even CASSCF methods are quite expensive in the context of photodynamics of biological chromophores. Thus, there is incentive to seek alternative electronic structure methods with similar, or perhaps even better, accuracy but much less computational cost. Semiempirical methods are quite promising in this regard. However, most previous parameterizations have emphasized ground state properties. In the few cases where excited electronic states were considered,30'31 vertical excitation energies were emphasized and not global features of excited state PESs. Furthermore, most semiempirical methods are based on a single-reference electronic wavefunction. The MNDOC method32"34 has been employed using a multi-reference single and double excitation configuration interaction (MRSDCI) wavefunction, " but with molecular orbitals derived from a single determinant SCF calculation,
233 i.e. Hartree-Fock. An interesting multi-reference semiempirical model was proposed by Cullen, but only applied to ground states. ' We have recently explored the utility of a multi-reference semiempirical method which approximates the ab initio state-averaged CASSCF method. The floating occupation molecular orbital (FOMO) method40"42 is used to obtain molecular orbitals which avoid bias to a particular electronic state. These molecular orbitals are "best-compromise" orbitals for the low-lying electronic states and thus the procedure must be followed by CI. We usually perform a CASCI for this purpose, although other forms of CI may be used if necessary.40"42 Electron repulsion integrals are approximated using a "neglect of differential diatomic overlap" (NDDO) scheme.43 Our first investigations of the FOMO-CASCI semiempirical method focused on the predicted geometries and energetics of conical intersections. ' We found that MECI geometries were predicted quite well using this approach with standard semiempirical parameter sets such as MNDO,44'45 AMI, 46 and PM3. 47 In Fig. 4, we show an example of the agreement which is achieved for an So/Si MECI geometry in the case of the Green Fluorescent Protein (GFP) chromophore. The ab initio and semiempirical predicted MECI geometries are almost indistinguishable. However, the energetics predicted by the semiempirical FOMO-CI method is often qualitatively incorrect. For example, the MECI which is an absolute excited state minimum for ethylene using ab initio methods '4 ' was predicted to lie above the Franck-Condon point using the FOMO-CASCI method with the MNDO parameterization. This would lead to qualitatively incorrect dynamics, since the MECI is energetically inaccessible after photoexcitation. Nevertheless, the good agreement for MECI geometries suggests that the semiempirical FOMO-CI method may be useful if the semiempirical parameters are optimized for the specific molecule of interest.
Fig. 4. Comparison of ab initio (black) and semiempirical (grey) So/S i MECI geometries for the neutral form of the GFP chromophore. The ab initio results are obtained using a SA-2-CAS(2/2) electronic wavefunction with the 6-31G basis set and the semiempirical results are obtained with the FOMO-CI method using the AMI semimepirical Hamiltonian.
234 We apply high-level ab initio methods at the most important geometries of the molecule under study (ground and excited state minima and MECIs) for all the electronic states of interest. The semiempirical parameters are then optimized in order to reproduce the ab initio results. Such system-specific reparameterization was almost routine in early implementations of semiempirical methods, and was most recently revived under the acronym SRP ("specific reaction parameters") for ground state reaction rate calculations.50 A similar approach has been used in the context of photochemical reactions by Olivucci and coworkers under the rubric of the molecularmechanics/valence-bond (MMVB) method.51"54 However, the valence bond form of the wavefunction used in the MMVB parameterization is not expected to treat ionic states accurately, restricting its application to cases where only covalent states are involved. The best choice of what data to include in reparameterization of the semiempirical FOMO-CI method is a current topic of our research. In our work on the GFP chromophore,55 we have included stationary points (both local minima and transition states) on the ground and excited electronic states and also selected MECIs. At each of these molecular geometries, the ground and excited state energies were included in the data set to be reproduced. In Fig. 5, we show a representative example of the results obtained in this case. The reparameterized semiempirical FOMO-CI method is seen to be intermediate between the CASSCF and CASPT2 methods, at a computational cost which is significantly less than CASSCF. A simpler and more automatic approach to reparameterization may be possible. For benzene, we tried using only a few molecular geometries corresponding to local minima on the ground electronic state. For each of these geometries, both energies and gradients of the low-lying electronic states were included. This strategy was sufficient to produce reasonable agreement of both geometries and energetics of MECIs and also of potential energy surfaces along pathways connecting photochemically-important points such as the Franck-Condon point and MECIs.56
235
LLJ
20
40
60
I-Twist Angle / Degrees Fig. 5. Comparison between reparameterized FOMO-CI (dotted lines) So and Si PESs along an Sicoordinate-driving path (optimizing all coordinates except for one of the torsion angles on the S, PES) and the corresponding PESs from SA-2-CAS(2,2) (solid lines) and SA-2-CAS(2/2)-PT2 (dashed lines). The molecule is the neutral form of the GFP chromophore and is depicted in the inset. The FOMO-CI and CASSCF paths are optimized separately. The CASPT2 results are obtained along the CASSCF-optimized paths. The zero of energy is in all cases chosen to be the So minimum at the respective level of theory. Although the reparameterized semiempirical FOMO-CI approach is much less computationally challenging than CASSCF methods, it is still too demanding for direct application in condensed phase and protein environments. However, the most significant effects of electronic excitation are often confined to a well-defined chromophore. Thus, a quantum mechanical/molecular mechanical (QM/MM) description may be appropriate. First implemented by Warshel and Levitt57, the basic idea in QM/MM methods is to describe a relatively small region of the system quantum mechanically (this could use either ab initio or semiempirical methods or even some combination), while the remainder is described with a classical force field. For example, in a photoactive protein, the QM region could be limited to the chromophore. The environment does exert an influence on the chromophore, through electrostatic and steric effects.
236 However, the electrons in the environment are not treated explicitly and are effectively treated adiabatically. The basic QM/MM method partitions the molecular Hamiltonian operator as follows: HTOT — tt(jM + HgM/MM
+ ti
MM
(1)
where HQM is the usual molecular electrostatic Hamiltonian dependent on the atoms and electrons in the quantum mechanical region, HMM contains the classical force-field terms for atoms in the molecular mechanics region, and HQMIMM couples the two subsystems. The QM/MM interaction usually contains both electrostatic and van der Waals terms:
^
(2)
where /and a are QM indices for electrons and nuclei respectively, m represents MM atoms, Z are nuclear (QM) or effective atomic (MM) charges and e and a are the van der Waals parameters. Only the first double summation in Eq. (2) contains the electronic coordinates, and this can be denoted as H^M/ml. For a given set of nuclear coordinates for both QM and MM atoms, one then carries out the usual QM optimization of the electronic wavefunction using the augmented Hamiltonian HQM +H^)MIMM in place of H0M . The procedure becomes considerably more complicated if the classical force field includes explicit electronic polarization effects. It can also be complicated when the QM and MM regions are connected by one or more covalent bonds, as is often found in protein environments. A number of solutions58"62 have been proposed for this "covalent embedding" problem, beginning with the "link atom" concept of Singh and Kollman.63 Recent work compared several of these approaches and found little advantage to the more complicated approaches. However, these comparisons were limited to semiempirical methods on the ground electronic state. It is not clear whether similar conclusions would be reached for excited states and/or ab initio methods. The QM/MM methods are intended to reproduce the results of a fully QM calculation on the same system. There will however be errors stemming from inadequacy of the classical force fields and neglected effects such as electronic polarization of the MM region. Thus, it is important to quantify the accuracy of the approximation. We show one example here for an So/Si MECI geometry obtained for the neutral form of the GFP chromophore surrounded by 27 water molecules (roughly one solvation shell.) This is sufficiently small that it is possible to carry out an MECI optimization with all atoms treated quantum mechanically using the semiempirical FOMO-CI method. We have also carried out the same calculation treating all but one of the water molecules using the SPC force field.65 The chromophore and the water molecule closest to the OH of phenol are treated with semiempirical FOMO-CI. Starting from the geometry determined quantum mechanically, we searched for an MECI using the QM/MM semiempirical
237
Fig. 6. Comparison of So/Si MECI geometries obtained for the neutral form of the GFP chromophore in a cluster of 27 water molecules using QM (black) and QM/MM (grey) methods. In both cases, the QM method is semiempirical FOMO-CASCI(12/8) using AMI parameters. The MM water molecules in the QM/MM method are modeled using the SPC force field. FOMO-CI method. The resulting QM and QM/MM geometries are compared in Fig. 6. The geometries are seen to be quite similar and a quantitative comparison can be given in the root mean square deviation of the structures, which is computed to be 0.28 Angstroms. Further comparisons including energetics and topography of PESs around MECIs can be found in our previous work. Several implementations of QM/MM using ab initio and/or reparameterized semiempirical methods for the QM region in the context of photobiology have been described recently. The ONIOM method of Morokuma and coworkers has been applied to photoisomerization of the retinal protonated Schiff base (RPSB) chromophore in the gas phase.66 Olivucci and coworkers have included dynamic electron correlation in the quantum mechanical region using the CASPT2 method in a QM/MM study of RPSB in the rhodopsin protein environment.67 This study was limited to geometries near the Franck-Condon point, but a more recent study of PYP using QM/MM with CASSCF for the QM region has also included excited state dynamics.68 The Schulten group69 has investigated the excited state dynamics of RPSB in bacteriorhodopsin (bR) using a QM/MM method with a CASSCF description of the QM region.
238 Warshel has used a limited reparameterization of the QCFF/PI semiempirical method in a followup7 to earlier QM/MM work on bR using QCFF/PI without reparameterization.71 We have used QM/MM methods with a semiempirical FOMO-CI description for the QM region in both condensed phase and protein environments.42'55 We have both optimized MECI locations and followed excited state dynamics including electronic quenching. 3. EXCITED STATE REACTION PATHS AND DYNAMICS The concept of a minimal energy path (MEP) connecting reactants and products is very useful for reactions occurring on the ground state, in spite of the paradox that it is only strictly valid in the infinite friction limit where activated reactions cannot occur. Likewise, the concept remains useful for excited state reactions even though it is only a model. In particular, the ultrafast nature of many photochemical reactions leads one to expect that vibrational energy relaxation rates could be bottlenecks to reaction. Olivucci and coworkers have used the MEP to connect points expected to be important in photochemistry,15'72'73 such as the Franck-Condon point, excited state minima, and MECIs (see Fig. 1). From these MEPs, reaction mechanisms can be inferred. However, ultimately dynamics becomes important in order to determine lifetimes for connection with experiment - how long does it take to reach a conical intersection and how efficiently is population quenched at the intersection? If the dynamics is to be modeled specifically, then one immediately faces the need for a quantum mechanical, or at least semiclassical, description of the nuclear degrees of freedom. Changes in electronic state signal the failure of the Born-Oppenheimer approximation and are most easily treated within a formalism which recognizes that the electrons are quantized. There are three basic approaches to model nonadiabatic effects which can be currently applied to large molecules in condensed phases: Ehrenfest dynamics, surface-hopping dynamics, ' and multiple spawning.9'77"80 We only sketch the methods here, as they have been adequately described and compared in the literature. In the Ehrenfest method, each classical trajectory moves on a potential energy surface which is a weighted average of all the electronic states. The weights in this average correspond to the probability of being on a particular state, which is determined by solving the Schrodinger equation along the prescribed trajectory. The drawback of this approach is that the potential energy surface can be quite unphysical when several electronic states which are very different from each other are populated. This has been discussed at length by Tully and others.76 Even when the electronic states in the problem are quite different, Ehrenfest dynamics can still be reasonable as long as population is transferred between electronic states quickly. The real problem comes when a trajectory propagates on an average of electronic states for an extended period of time and these states are very different from each other. Surface-hopping ameliorates the problems of Ehrenfest dynamics by demanding that the forces governing a trajectory are always derived from a single adiabatic electronic state. However, the electronic state with which a trajectory is associated changes discontinuously.
239 Typically this "hopping" occurs according to a probabilistic rule. In the "fewest switches" version, this rule is itself derived from an Ehrenfest-like dynamics, where a Schrodinger equation is solved for the electronic amplitudes along the trajectory. The multiple spawning method starts as a basis set description of wavepacket dynamics, and thus always has access to a nuclear wavefunction representing the system. Each nuclear basis function is like a classical trajectory in that its phase space evolution follows Hamilton's equations for a specified electronic state. New nuclear basis functions are created adaptively when the nonadiabatic matrix elements that signal failure of the Born-Oppenheimer approximation become large. Instead of trajectories hopping between electronic states, new trajectory basis functions are created and these are populated according to the solution of the nuclear Schrodinger equation in the expanded basis set. Since the new trajectory basis functions always begin with no population, i.e. they are initially "virtual" basis functions, the nuclear wavefunction evolves continuously with time. In principle, the advantage of this formalism is that it is guaranteed to converge to exact solution of the nuclear Schrodinger equation as the basis set is enlarged. Although exact numerical solution of the nuclear wavefunction may not be feasible for molecules with hundreds of degrees of freedom, it is nevertheless possible to test the convergence of specific observables as the basis set is expanded. This is not possible in either of Ehrenfest dynamics or surface-hopping, which are not easily cast as members of a hierarchy of methods culminating in exact solution of the nuclear Schrodinger equation. Furthermore, the availability of a nuclear wavefunction in multiple spawning means that correlation functions and hence spectra can be computed directly from the simulation results. ' When the multiple spawning method is used in conjunction with simultaneous solution of the electronic structure for the potential energy surfaces and their nonadiabatic couplings, we refer to it as "ab initio multiple spawning" (AIMS). 4. APPLICATION TO GREEN FLUORESCENT PROTEIN Now we turn to a representative application of the methods discussed so far. The Green Fluorescent Protein83 (GFP) is found in a variety of coelenterates, including the jellyfish A. victoria. The protein absorbs blue light emitted (approx. 470nm) in a chemiluminescent reaction by the protein aequorin and emits green light (508nm). It is still not entirely clear why the jellyfish should want or need to do this wavelength conversion, but the autofiuorescent nature of GFP has made it a useful tool in molecular biology. The GFP chromophore is formed autocatalytically from a Ser-Tyr-Gly tripeptide after expression and folding under aerobic conditions. Since there is no need for an external cofactor, chimeric proteins can be engineered such that GFP is attached to a desired protein and becomes a fluorescent probe reporting on its expression. Curiously, the chromophore of GFP (p-hydroxybenzylidene-imidazolinone, see Fig. 5) has been established to be virtually non-fluorescent when it is either synthesized or when the protein environment is disrupted. This raises the question of the mechanism for nonradiative
240 decay in the solvated chromophore and the means by which the protein prevents this decay. It is further known that the photocycle in the protein involves proton transfer,84 so that at least two protonation states of the chromophore are important. Early semiempirical studies were not very successful at determining which protonation states were involved,85'86 but the experimental evidence favors the neutral and anionic (phenolate) forms.87"90 Fluorescence is observed primarily from the anionic form and mutant GFPs have been engineered which stabilize the anionic form in order to increase fluorescence quantum yield.91 Previous theoretical work of Weber, et al. using semiempirical methods led to the suggestion that nonradiative decay occurred through conical intersections associated with torsion about one or both of the bridge bonds connecting the phenol (P-bond) and imidazolinone (I-bond) rings to the central carbon atom.92 These calculations also suggested the possibility of "hula twist" motion,93"95 i.e. concerted torsion about the I- and P-bonds, as an important coordinate leading to So/Si degeneracy. A number of subsequent theoretical studies concentrated on possible tuning of the electronic absorption of the GFP chromophore by the protein environment17 and also on the vibrational spectroscopy,96 partially in response to the early semiempirical work which shed doubt on the assignment of the chromophore protonation states involved in the photocycle. The most recent work using CASSCF and CASPT2 has focused on the anionic protonation state and suggested that while hula twist motion is disfavored on Si in the gas phase, it does lead to an So/Si MEC1 and was the only explanation which could be found for the nonradiative decay.97 As will be seen below, our calculations suggest that there are So/Si MECIs very close to two distinct but nearly-degenerate global minima on Si, which would explain the lack of fluorescence in the chromophore. What is more, solvent accelerates this decay significantly. As suggested in Fig. 5, photoexcitation of the neutral GFP chromophore is expected to lead to torsion about the "1-bond." Using the SA-2-CAS(2/2) method with a 6-31G basis set and searching for a Si/So MECI from the Si global minimum (SiMinG, twisted by 90° about the Ibond) leads to a conical intersection which is 7 kcal/mol above SiMino- The molecular geometry of the So/Si MECI is shown in Fig. 7, along with geometrical parameters for both Si Mine and the MECI. Torsion about the "P-bond," the bridge bond adjacent to the I-bond, is not favored in the neutral chromophore as might have been expected due to the propensity for bond alternation in the bright excited state of conjugated molecules. This situation changes in the anion form of the chromophore, where two twisted local minima exist one for each of the I- and P-bonds. These are essentially isoenergetic, and again the So/Si MECI geometries are energetically above the corresponding twisted Si minima. A schematic of the important points on the PES is shown in Fig. 8. Energetics obtained with both SA-2-CAS(2/2) and SA-2-CAS(2/2)-PT2 are shown, using the 6-31G* basis set. These results confirm the suggestion of Weber, et al. that torsion represents a viable pathway for nonradiative decay, although we should point out that the picture is quite different in details. Weber, et al. did not find a torsional deactivation pathway for the anion form of the chromophore which was energetically accessible from the Franck-Condon region, and even the inaccessible pathway which they found required simultaneous torsion of the I- and P-
241 bonds. However, these disagreements are not surprising in view of their use of an uncalibrated semiempirical method.
Fig. 7. The imidazolinone-twisted minimal energy conical intersection for the neutral form of the GFP chromophore. Included are heavy-atom bond lengths (A), bridge dihedrals and interior heavy-atom angles for the So/S| MECI and for the imidazolinone-twisted SiMinG (in parentheses). The relevant structures were optimized on S, using an SA2-CAS(2,2)/6-31G wavefunction.
242
Fig. 8. Fluorescence lifetime of neutral and anionic forms of GFP chromophore as predicted by CAS(2/2)/6-31G dynamics. There are four trajectory basis functions corresponding to the neutral form and three trajectory basis functions corresponding to the anion. The lifetime of the neutral is slightly shorter and the amplitude of the decay is larger. Larger sets of initial conditions are needed to make firm conclusions, but the results are in agreement with solution phase experiments. The immediate question which follows the preceding observations is whether there are any barriers to torsion in either of the neutral or anionic forms of the chromophore. Recent experiments have measured the fluorescence lifetime of both the neutral and anionic forms in different solution environments. The authors find a multiexponential decay of fluorescence with the shortest lifetime ranging from 70fs to 460fs. This suggests that any barrier which does exist must be quite small, at least in the solution environment. Furthermore, the lifetime of the neutral is found to be consistently shorter than the anion. We have carried out AIMS simulations to investigate this, using the SA-2-CAS(2/2) electronic wavefunction in the 6-31G basis set. Initial conditions are sampled from the Wigner distribution corresponding to the So minimum in the harmonic approximation. The resulting fluorescence is shown in Fig. 9, which is averaged over the trajectory basis functions used to represent the initial state. The neutral and anion are seen to be quite similar, with fluorescence from the neutral form decaying slightly faster. This is in broad agreement with the experimental results, but one should note that these simulations are in the gas phase and furthermore that the number of initial conditions used is not yet sufficient to
243 draw firm conclusions regarding the lifetime. An important point to note is that the observed fluorescence decay comes strictly from torsion of the chromophore and not from electronic quenching. As expected, the So/Si transition dipole moment governing the fluorescence intensity becomes very small when the molecule is twisted about either of the I- or P-bonds. Although Fig. 9 does include the proper weighting of the electronic state population, i.e. each trajectory basis function is weighted according to its population on Si, there is very little (< 10%) quenching to the ground electronic state in the time shown.
50
100
150
200
250
300
350
Fig. 9. Fluorescence lifetime of neutral and anionic forms of GFP chromophore as predicted by AIMS dynamics using CAS(2/2)/6-31G for the electronic structure. The square of the S]/So transition dipole moment at the center of each nuclear basis function, averaged over all trajectory basis functions, is shown. There are four trajectory basis functions corresponding to the neutral form and three trajectory basis functions corresponding to the anion. The lifetime of the neutral is slightly shorter and the amplitude of the decay is larger. Larger sets of initial conditions are needed to make firm conclusions, but the results are in agreement with solution phase experiments.
244 Thus, the decay of fluorescence should not match the rise time which would be measured by probing ground state recovery. However, we cannot say yet how much different these lifetimes should be, except to place a lower bound of lOOfs on the difference. Is there any significant effect from the solvent environment? In order to investigate this question, we turn to reparameterized semiempirical methods within which we are able to include an environment explicitly. Only the neutral form of the chromophore will be addressed here. We reparameterized the FOMO-CI semiempirical method for the GFP chromophore, using ab initio results for the calibration data.55 We refer the reader back to Fig. 5 for an example of the quality of the resulting PESs. Similar results are obtained for the neutral chromophore. The solvent environment is modeled using the SPC force field representation of water,65 with 51 water molecules surrounding the chromophore. Carrying out AIMS simulations with the reparameterized semiempirical FOMO-CASCI(12/8) method leads to the lifetimes shown in Fig. 10. These are now reflective of the population on Si, and cannot be directly compared to the fluorescence lifetimes shown in Fig. 9. What is initially striking about this data is the large decrease in excited state lifetime which is observed on solvation. Further investigation shows that this is a consequence of a significant change in the PES of the chromophore in solution as compared to the gas phase. In particular, the So/Si MECI in the gas phase is located approximately 7 kcal/mol above the I-bond twisted global minimum. Upon aqueous solvation, the So/Si MECI becomes the absolute minimum on Si, i.e. the Si Mine and So/Si MECI points become identical. Thus, access to the MECI becomes nonactivated, even for molecules which are able to dissipate energy rapidly. This is a consequence of the charge-transfer character involved in the states which are connected at the MECI, and is expected to be a general phenomenon. The lifetime we observe in the aqueous environment is in good agreement with the 70fs short time constant measured for the neutral chromophore in water, providing considerable support to this picture.
245 I
I
i
I
i
I
^Vacuum 0.8
-
^
-
^
^
^
O
0.6
-
Water
Q.
O CL
0.4
-
-
0.2
-
-
n
50
I
I
I
I
I
100
150
200
250
300
350
Time / fs Fig. 10. Population on Si as a function of time for neutral GFP chromophore in vacuum and water environments. The chromophore in both cases is modeled with the reparameterized semiempirical FOMO-CASCI(12/8) method. Water molecules are treated using a QM/MM method with the SPC force field. Finally, one can ask how the protein environment affects the excited state dynamics. Several X-ray structures are available99"101 and we use the wild-type A. victoria structure in this work. A truncated model was constructed which includes all residues within a 10 Angstrom radius of the phenol moiety of the GFP chromophore. This is depicted in Fig. 11, along with a cartoon representation of the GFP protein which shows the novel "y?-can" fold. We have used the
246
Fig. 11. Left: cartoon representation of the GFP protein, showing the chromophore buried inside the "(5can" structure. Right: Computational model used as a representation of the protein environment. The model has a total of almost 500 atoms, of which 39 are treated with the FOMO-CI method and the remainder with the AMBER force field. AMBER force field1 to represent the surrounding protein residues and the reparameterized semiempirical FOMO-CI method used for the chromophore. The chromophore is connected to the protein residues using "connection atoms" which have been reparameterized in order to reproduce the purely QM results for excitation energies in a smaller model including only the nearest protein residue explicitly. Our treatment of these connection atoms follows the strategy set forth by Antes and Thiel,103 and these connection atoms must be reparameterized (or at least the parameterization must be verified) for each new system. We first investigate the MEP for the chromophore in order to compare the vacuum, solvated, and protein environments. A simple way to generate an approximate MEP is by carrying out dynamics in a high-friction environment. We do this by using a Langevin dynamics:
dE(r) Pa =
~ d
YaPa
(3)
wherep a =mara is the conjugated momentum of the ra coordinate and ma is the associated mass. The usual Langevin equation also contains a term representing the random energy exchange between solute and solvent, which we ignore since our purpose is just to generate an
247 approximate MEP. In principle, the friction coefficients ya are related to the molecular diffusion coefficient D: D
-
^
-
(4)
2 However, for our purposes they do not have to represent any real solvent and we chose ya = 1014 sec"'. The numerical implementation has been coded in a development version of the MOPAC package, along with the other methods we describe in this paper.104 The resulting MEP is depicted in Fig. 12 for the neutral GFP chromophore in isolated, water, and protein environments. The distinction between the vacuum and aqueous environments we mentioned earlier is immediately evident. In the vacuum case, the energy gap between So and Si remains large when the minimum on Si is reached. In contrast, the So/Si gap vanishes even before the minimum on Si is reached for the aqueous environment. Thus, the chromophore is directed to a conical intersection seam and follows the seam toward the absolute minimum, which is itself a conical intersection. The rapid excited state decay in the aqueous case is thus easily explained. In the protein environment, the chromophore geometry does not change very much before a minimum is reached. This minimum is planar and is expected to fluoresce. Hence, the protein environment has induced a barrier to torsion which traps the chromophore in a fluorescent state. We have not yet determined the height of the barrier and dynamics simulations are underway to determine what fraction of the excited chromophores in the protein environment are trapped in this fluorescent local minimum. A further interesting question is the origin of the barrier, which could be dominated by steric or electrostatic effects.
248
Vacuum
& CD C
LLJ
10
15
20
25
30
35
40
Reaction Coordinate / amu1'2 Angstrom Fig. 12. Minimal energy path for the neutral GFP chromophore on Si in different environments. The water environment contains 51 water molecules modeled with the SPC force field. The protein environment corresponds to the truncated model depicted in Fig.ll, where the protein surroundings are modeled with the AMBER force field. In order to better compare the results, the reaction coordinate includes only the motion of the chromophore atoms, excluding the MM environment.
249 5. CONCLUSIONS We have provided an overview of methods to determine reaction mechanisms for photobiological systems. Early efforts using empirical potentials or restricted semiempirical methods have not been emphasized. Neither have we discussed in any detail the determination of reaction paths in isolated molecules of biological significance. Instead, we have focused on emerging methods which can include the environment in some detail while at the same time being able to provide a flexible and accurate description of the chromophore. Multireference electronic structure theory methods combined with classical force fields in QM/MM fashion seem to be the preferred way of simulating these systems. The methodology is still developing and the number of applications to date is rather few. However, there is every reason to believe that this will be a rapidly-growing field over the next few years and that we are on the verge of a new level of accuracy in photobiological simulation. Indeed, we have presented some applications where not only reaction paths but also dynamic evolution has been modeled, including quantum mechanical effects of both electrons and nuclei. Using GFP as a model application, we have shown how it is possible to characterize reaction paths and follow dynamics in a chromophore from the gas phase through to solution and protein environments. Electrostatic effects of the environment can change the photochemical outcome dramatically, providing an order of magnitude decrease in the excited state lifetime for the case of GFP chromophore in an aqueous environment. We have shown that our approach does predict a barrier to torsion in the protein environment, and therefore an expected increase in fluorescence in agreement with experiment. Future studies will characterize the nature of this barrier and quantify its effect on the excited state lifetime. The next step, which can be expected in the near future, is to begin designing complex environments for photochemical reactions. As the accuracy of the methods becomes quantified through widespread application and comparison to experiment and as computational power continues to increase, this will be a realistic and useful goal. ACKNOWLEDGEMENTS This work has been supported by the National Science Foundation and the Department of Energy. TJM is grateful to the Packard and Dreyfus Foundations for support through a Packard fellowship and Teacher-Scholar award, respectively.
250 REFERENCES 1. M. A. van der Horst and K. J. Hellingwerf, Ace. Chem. Res. 37, 13-20 (2004). 2. M. Klessinger and J. Michl, Excited States and Photochemistry of Organic Molecules (VCH Publishers, New York, 1995). 3. M. A. Robb, F. Bernardi, and M. Olivucci, Pure Appl. Chem. 67, 783-789 (1995). 4. E. Teller, J. Phys. Chem. 41, 109-116 (1937). 5. D. R. Yarkony, J. Phys. Chem. 105A, 6277-6293 (2001). 6. T. Carrington, Ace. Chem. Res. 7, 20-25 (1974). 7. M. Ben-Nun and T. J. Martinez, Chem. Phys. Lett. 298, 57-65 (1998). 8. M. Ben-Nun and T. J. Martinez, Chem. Phys. 259, 237-248 (2000). 9. M. Ben-Nun, J. Quenneville, and T. J. Martinez, J. Phys. Chem. 104A, 5161-5175 (2000). 10. F. Molnar, M. Ben-Nun, T. J. Martinez, and K. Schulten, J. Mol. Struct. (THEOCHEM) 506, 169-178(2000). 11. We use the term "absolute minima" because these are clearly not local minima in the usual sense of the word-the gradient vector of the adiabatic electronic state is ill defined at a conical intersection geometry. 12. J. F. Stanton and R. J. Bartlett, J. Chem. Phys. 98, 7029-7039 (1993). 13. B. O. Roos, Ace. Chem. Res. 32, 137-144 (1999). 14. V. Helms, Curr. Op. Struct. Biol. 12, 169-175 (2002). 15. M. Garavelli, F. Bernardi, P. Celani, M. A. Robb, and M. Olivucci, J. Photochem. Photobio. 114A, 109-116(1998). 16. C. Ko, B. Levine, A. Toniolo, L. Manohar, S. Olsen, H.-J. Werner, and T. J. Martinez, J. Amer. Chem. Soc. 125, 12710-12711 (2003). 17. V. Helms, C. Winstead, and P. W. Langhoff, J. Mol. Struct. THEOCHEM 506, 179-189 (2000). 18. V. Molina and M. Merchan, Proc. Natl. Acad. Sci. 98, 4299-4304 (2001). 19. E. K. U. Gross and W. Kohn, Adv. Quantum Chem. 21, 255-291 (1990). 20. D. J. Tozer, R. D. Amos, N. C. Handy, B. O. Roos, and L. Serrano-Andres, Mol. Phys. 97, 859-868(1999). 21. D. J. Tozer andN. C. Handy, J. Chem. Phys. 109, 10180-10189 (1998). 22. A. Dreuw, J. L. Weisman, and M. Head-Gordon, J. Chem. Phys. 119, 2943-2946 (2003). 23. S. Grimme and M. Parac, ChemPhysChem 4, 292-295 (2003). 24. N. T. Maitra, F. Zhang, R. J. Cave, and K. Burke, J. Chem. Phys. 120, 5932-5937 (2004). 25. R. J. Cave, F. Zhang, N. T. Maitra, and K. Burke, Chem. Phys. Lett. 389, 39-42 (2004). 26. M. Wanko, M. Garavelli, F. Bernardi, T. A. Niehaus, T. Frauenheim, and M. Elstner, J. Chem. Phys. 120, 1674-1692 (2004). 27. The notation we use for state-averaged CASSCF is that SA-n-CAS(x/y) is a calculation with an active space of x electrons in y orbitals where the orbitals are determined to minimize
251
28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 54. 55. 56.
the energy of an equally weighted average of the n lowest electronic states of the desired spin symmetry (typically singlet). B. Levine, J. Quenneville, C. Ko, and T. J. Martinez, manuscript in preparation T. Vreven, F. Bernardi, M. Graravelli, M. Olivucci, M. A. Robb, and H. B. Schlegel, J. Amer. Chem. Soc. 119, 12687-12688 (1997). J. Del Bene and H. H. Jaffe, J. Chem. Phys. 48, 1807 (1968). J. Ridley and M. Zerner, Theo. Chim. Acta 32, 111 (1973). W. Thiel, J. Amer. Chem. Soc. 103, 1413-1420(1981). W. Thiel, J. Amer. Chem. Soc. 103, 1420-1425 (1981). A. Schweig and W. Thiel, J. Amer. Chem. Soc. 103, 1425-1431 (1981). M. Reinsch, U. Howeler, and M. Klessinger, Angew. Chem. 99, 250-252 (1987). M. Reinsch, U. Howeler, and M. Klessinger, J. Mol. Struct. (THEOCHEM) 167, 301-306 (1988). M. Klessinger, T. Potter, and C. v. Wullen, Theo. Chim. Acta 80, 1-17 (1991). J. Dreyer and M. Klessinger, J. Chem. Phys. 101, 10655-10665 (1994). J. M. Cullen, Int. J. Quantum Chem. 56, 97-113 (1995). G. Granucci and A. Toniolo, Chem. Phys. Lett. 325, 79 (2000). A. Toniolo, M. Ben-Nun, and T. J. Martinez, J. Phys. Chem. 106A, 4679-4689 (2002). A. Toniolo, G. Granucci, and T. J. Martinez, J. Phys. Chem. 107A, 3822-3830 (2003). G. A. Segal, Semiempirical methods of electronic structure calculation (Plenum Press, New York, 1977). M. J. S. Dewar and W. Thiel, J. Amer. Chem. Soc. 99, 4899-4907 (1977). M. J. S. Dewar and W. Thiel, J. Amer. Chem. Soc. 99, 4907-4917 (1977). M. J. S. Dewar, E. G. Zoebisch, E. F. Healy, and J. J. P. Stewart, J. Amer. Chem. Soc. 107, 3902-3909(1985). J. J. P. Stewart, J. Comp. Chem. 10, 209-220 (1989). J. Quenneville, M. Ben-Nun, and T. J. Martinez, J. Photochem. Photobiol. A 144, 229-235 (2001). J. Quenneville and T. J. Martinez, J. Phys. Chem. 107A, 829-837 (2003). I. Rossi and D. G. Truhlar, Chem. Phys. Lett. 233, 231-236 (1995). M. J. Bearpark, M. A. Robb, F. Bernardi, and M. Olivucci, Chem. Phys. Lett. 217, 513-519 (1994). M. J. Bearpark, F. Bernardi, S. Clifford, M. Olivucci, M. A. Robb, B. R. Smith, and T. Vreven, J. Amer. Chem. Soc. 118, 169-175 (1996). M. Deumal, M. J. Bearpark, B. R. Smith, M. Olivucci, F. Bernardi, and M. A. Robb, J. Org. Chem. 63, 4594-4600 (1998). B. R. Smith, M. J. Bearpark, M. A. Robb, F. Bernardi, and M. Olivucci, Chem. Phys. Lett. 242,27-32(1995). A. Toniolo, S. Olsen, L. Manohar, and T. J. Martinez, Faraday Disc. 127, 149-163 (2004). A. Toniolo, A. L. Thompson, and T. J. Martinez, Chem. Phys. 304, 133-145 (2004).
252 57. 58. 59. 60. 61. 62. 63. 64. 65. 66. 67. 68. 69. 70. 71. 72. 73. 74. 75. 76. 77. 78. 79. 80. 81. 82. 83. 84. 85. 86. 87. 88. 89.
A. Warshel and M. Levitt, J. Mol. Biol. 103, 227-249 (1976). J. Gao, P. Amara, C. Alhambra, and M. J. Field, J. Phys. Chem. 102A, 4714-4721 (1998). D. M. Philipp and R. A. Friesner, J. Comp. Chem. 20, 1468-1494 (1999). V. Kairys and J. H. Jensen, J. Phys. Chem. 104A, 6656-6665 (2000). Y. Zhang, T. Lee, and W. Yang, J. Chem. Phys. 110, 46-54 (1999). A. Toniolo, C. Ciminelli, G. Granucci, T. Laino, and M. Persico, Theo. Chem. Ace. I l l , 270279 (2004). U. C. Singh and P. A. Kollman, J. Comp. Chem. 7, 718 (1986). N. Reuter, A. Dejaegere, B. Maigret, and M. Karplus, J. Phys. Chem. 104A, 1720-1735 (2000). L. X. Dang and B. M. Pettitt, J. Phys. Chem. 91, 3349-3354 (1987). T. Vreven and K. Morokuma, J. Chem. Phys. 113, 2969-2975 (2000). N. Ferre and M. Olivucci, J. Amer. Chem. Soc. 125, 6868-6869 (2003). G. Groenhof, M. Bouxin-Cademartory, B. Hess, S. P. de Visser, H. J. C. Berendsen, M. Olivucci, and M. A. Robb, J. Amer. Chem. Soc. 126, 4228-4233 (2004). S. Hayashi, E. Tajkhorshid, and K. Schulten, Biophys. J. 85, 1440-1449 (2003). A. Warshel and Z. T. Chu, J. Phys. Chem. 105B, 9857-9871 (2001). A. Warshel, Z. T. Chu, and J.-K. Hwang, Chem. Phys. 158, 303-314 (1991). M. Garavelli, F. Bernardi, M. Olivucci, T. Vreven, S. Klein, P. Celani, and M. A. Robb, Faraday Disc, 51-70 (1998). M. Garavelli, P. Celani, M. Fato, M. J. Bearpark, B. R. Smith, M. Olivucci, and M. A. Robb, J. Phys. Chem. 101A, 2023-2032 (1997). G. D. Billing, Int. Rev. Phys. Chem. 13, 309 (1994). J. C. Tully and R. K. Preston, J. Chem. Phys. 55, 562 (1971). J. C. Tully, J. Chem. Phys. 93, 1061 (1990). M. Ben-Nun and T. J. Martinez, Adv. Chem. Phys. 121, 439-512 (2002). T. J. Martinez, M. Ben-Nun, and G. Ashkenazi, J. Chem. Phys. 104, 2847-2856 (1996). T. J. Martinez, M. Ben-Nun, and R. D. Levine, J. Phys. Chem. 100, 7884-7895 (1996). M. Ben-Nun and T. J. Martinez, J. Chem. Phys. 108, 7244-7257 (1998). K. K. Baeck and T. J. Martinez, Chem. Phys. Lett. 375, 299-308 (2003). M. Ben-Nun and T. J. Martinez, J. Phys. Chem. 103A, 10517-10527 (1999). R. Y. Tsien, Annu. Rev. Biochem. 67, 509-544 (1998). M. Chattoraj, B. A. King, G. U. Bublitz, and S. G. Boxer, Proc. Natl. Acad. Sci. 93, 83628367 (1996). A. A. Voituyk, M. Michel-Beyerle, and N. Rosch, Chem. Phys. Lett. 272, 162-167 (1997). A. A. Voituyk, M. Michel-Beyerle, and N. Rosch, Chem. Phys. Lett. 296, 269-276 (1998). X. A. He, A. F. Bell, and P. J. Tonge, J. Phys. Chem. 106B, 6056 -6066 (2002). A. F. Bell, X. He, R. M. Wachter, and P. J. Tonge, Biochem. 39, 4423-4431 (2000). A. P. Esposito, P. Schellengberg, W. W. Parson, and P. J. Reid, J. Mol. Struct. 569, 25-41 (2001).
253 90. P. Schellenberg, E. Johnson, A. Esposito, P. J. Reid, and W. W. Parson, J. Phys. Chem. 105B, 5316-5322(2001). 91. G. H. Patterson, S. M. Knobel, W. D. Sharif, S. R. Kain, and D. W. Piston, Biophys. J. 73, 2782-2790(1997). 92. W. Weber, V. Helms, J. A. McCammon, and P. W. Langhoff, Proc. Natl. Acad. Sci. 96, 6177-6182(1999). 93. R. S. H. Liu, D. Mead, and A. E. Asato, J. Amer. Chem. Soc. 107, 6609-6614 (1985). 94. R. S. H. Liu and A. E. Asato, Proc. Natl. Acad. Sci. 82, 259-263 (1985). 95. R. S. H. Liu, Ace. Chem. Res. 34, 555-562 (2001). 96. H. Yoo, J. A. Boatz, V. Helms, J. A. McCammon, and P. W. Langhoff, J. Phys. Chem. 105B, 2850-2857 (2001). 97. M. E. Martin, F. Negri, and M. Olivucci, J. Amer. Chem. Soc. 126, 5452-5464 (2004). 98. D. Mandal, T. Tahara, and S. Meech, J. Phys. Chem. 108B, 1102-1108 (2004). 99. F. Yang, L. G. Moss, and J. Phillips, G. N., Nat. Biotech. 14, 1246-1251 (1996). 100. J. Phillips, G. N., Curr. Op. Struct. Biol. 7, 821-827 (1997). 101. M. Ormo, A. Cubitt, K. Kallio, L. Gross, R. Y. Tsien, and S. Remington, Science 273, 1392-1395(1996). 102. S. J. Weiner, P. A. Kollmann, D. A. Case, U. C. Singh, C. Ghio, G. Alagona, S. Profeta, and P. Weiner, J. Amer. Chem. Soc. 106, 765-784 (1984). 103. I. Antes and W. Thiel, J. Phys. Chem. 103A, 9290-9295 (1999). 104. J. J. P. Stewart, (Fujitsu Limited, Tokyo, Japan, 1999).
This Page is Intentionally Left Blank
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 Elsevier B .V. All rights reserved
255
VIII. Development of Theory with Computation Howard E. Zimmerman Chemistry Department University of Wisconsin 1. INTRODUCTION - EARLY EFFORTS; THE BACKGROUND The present chapter describes the development of theory in our laboratory from the primitive beginnings at the outset to our present efforts. This includes both computational and theoretical aspects. The present author's interest in quantum mechanical and photochemical organic research began with his reading publications by Egbert Havinga in 1956 [1] and by Derek Barton in 1958 [2]. Thus, Egbert Havinga reported the curious behavior of p- and m-nitrophenyl phosphates in water. The para isotner hydrolyzed reasonably rapidly independent both in the dark and in light while the meta isomer was highly reactive in light but unreactive in the dark.
m NO2
hv H2O
(2) — No Acceleration Relative to the Dark Reaction
This solvolysis behavior on irradiation, of course, was in contrast to what we teach our undergraduates about ortho-para electron transmission. Egbert Havinga noted that there was no rationale in mechanisms of the time using qualitative resonance theory. It was clear to the present author that the limitation in understanding the phenomenon was not knowing the structure of the excited state of the phosphate esters.
256 Similarly, two years later Derek Barton had elucidated the structure of Lumisantonin, the product resulting from irradiation of Santonin. At the time, the best mechanism for this bizarre rearrangement (Note Eq. 3) consisted of "bond switching" which indicated which bonds were lost and which were gained. Again, it was clear that what was needed was a description of the excited state undergoing the reaction. But at the time there had been no real attempt to relate photochemical transformations to excited state structures.
These two examples - the m-nitrophosphate solvolysis and the Santonin to Lumisantonin rearrangement - which were typical of the time prompted the present author to investigate whether determination of the excited state structures involved would lead to an understanding of the experimental photochemical reactivity. 2. OUR EARLY EFFORTS ON THESE TWO PROBLEMS However, at the time, the best computational facility available was the IBM 1604 and the programming available was the Hiickel treatment employing a Jacobi diagonalization. Our first computation was for the benzylic species. With eight electrons this is isoconjugate with anisole, and the aim was to understand what transmission of unshared electrons occurs from an electron donor group, such as methoxy. Thus, Scheme 1 shows the local densities in the seven n MO's of the benzyl species. The ground state is populated as depicted with solid dots. The excitation process, in simplistic homo to lumo form, is represented by arrow B. It is seen that there is a large enhancement of the meta electron density by promotion of an electron from an MO which has no meta density to one which does.
257
+2.10
+ 1.26
T
,57
4^>.1 *14
o
-2.10
Scheme 1. Densities of Benzyl Species MO's. All solid arrows represent electrons in the case of an electron donor. Hollow arrows represent electrons in the case of a withdrawing group. In contrast, there is diminution of the para density. The simple Hiickel computation for anisole paralleled the computation for the isoconjugate benzylic anion as seen in Scheme 2. Thus we proposed the phenomenon of "meta electron transmission" in the first excited state which contrasts with the well-established ortho-para transmission so characteristic of the ground state chemistry.
Ground State
Scheme 2. Densities for So and Si Anisole
0.76 Excited State
258 With two meta methoxyl groups, the effect is enhanced and this led to study of the solvolysis of meta-methoxylated benzyl acetates. We discuss this chemistry below in the context of subsequent and more sophisticated computation of benzylic systems. With the Havinga report in mind, we again considered Scheme 1 but now with just six n electrons, since this corresponds to an aromatic ring with an electron withdrawing group. In this case, excitation corresponds to arrow A with the consequence of loss of meta electron density. The para site actually has an increase in density on excitation. Hence, again, we see meta electron transmission in the excited state. Having considered the basis of the Havinga phenomenon we turned to the esoteric rearrangement of Santonin described by Derek Barton (Note Eq. 3 above). A beginning point was the nature of the excited states of carbonyl compounds. While at the time the nature of the n-7t* and 71-71* excited states of formaldehyde, acetone and similar carbonyl compounds was in the physical chemical literature, organic chemists either were not aware of this or had not made the connection to organic photochemistry. The physical chemists, while certainly aware of these excited states, were not in a position to deal with the organic chemistry. Thus we noted that the three-dimensional n-Tt* excitation process (Scheme 3) provides two orthogonal systems with the ground state in electron distribution. First, the py (or n) orbital has lost an electron and is reminiscent of the orbital in oxy-free radicals. Second, there is a TC system that has an added electron and is isoconjugate with a metal ketyl (i.e. a radical-anion). It was suggested that, while thinking in three dimensions, the short-hand notation shown in Scheme 3 permits the organic chemist to draw excited state structures quickly. It was also suggested that excited state reactivity is guided by the reacting molecule selecting "energy valleys" just as observed for ground state reactions. Thus, the organic chemist's electron pushing may be applied to species on an excited state hypersurface. Thus, the Norrish II was described as in Scheme 3 in those early papers. It is seen that the n-Tt* excited state has an "electron hole" in the py orbital and this orbital is coplanar with the sigma bonds to the carbonyl carbon. With overlap between the py orbital and one of the sigma bonds, we see a system of three coplanar orbitals - the p y orbital and the two comprising the sigma bond. This linear system of three orbitals and three electrons is isoconjugate with the allyl radical species. The bond order between basis orbitals 1 and 2 in allyl is lower than unity; similarly, in the n-Tc* excited state here, delocalization of the electron-hole weakens the bond holding the alkyl group R with resulting scission and release of the alkyl free-radical (Scheme 4). This is one example of the role of the pv orbital in sigma system n-Tt* reactivity.
259
R R
hν O yy° o
R R
•• Oo y
SP HYBRID ELECTRONS
•
7i ELECTRONS
y
p y ELECTRONS
Scheme 3. n-jt* Excitation
Scheme 4. Acyl fission of an Alkyl group in the n-n* Excited State. Turning now to the Santonin to LumiSantonin rearrangement (Eq. 3), we found it easier to begin with 4,4-diphenylcyclohexadienone in Scheme 5, since this would provide new photochemistry and yet had the same 2,5-cyclohexadienone electronics. The proposed mechanism [3c] is outlined in Scheme 5 using the short-hand notation described above. The basic transformation involves four steps after excitation to S,.
260
p,(3-Bonding
hv, ISC
ISC
,© Pri
1,3-Shift
Ph
Scheme 5. The Type-A Rearrangement of 4,4-Dipheny lcyclohexadienone. The first is intersystem crossing to Ti. The second is bonding between the beta carbons of the dienone system. The third is radiationless decay with intersystem crossing to So in the form of a zwitterion. And the last is a cyclopropyl-carbinyl rearrangement involving a 1,3-shift. The beta-beta bonding found justification in simple Hiickel computations [4]. Note Scheme 6. While the beta-beta bond order in the ground state is negative, that for the singly promoted dienone is positive and bonding. This indicates that a perturbation with increased beta-beta bonding will lower the energy.
O -0.295 +0.173 +0.007 +0.053 P,p-bond order-0.070 i.e. antibonding
GROUND STATE
+0.419 -0.109 | -0.023 -0.131
EXCITED STATE
P,p-bond order+0.114 i.e. bonding
Scheme 6. Hiickel Computation of 4,4-Disubstituted 2,5-Cyclohexadienone; Electron Densities and Bond Orders
261 Application of the same mechanism to the Santonin to LumiSantonin rearrangement, not only leads to the correct product structure but also accounts for the stereochemistry. In this early work [3] we showed how the above type of reasoning accounted for the major fraction of photochemical reactions known at the time. This included the Norrish Types I and II, the Yang Reaction, a-expulsion of groups adjacent to carbonyl moieties, epoxyketone rearrangements, the Paterni-Biichi reaction, and the "Type-B Enone Rearrangement" which we had just discovered [5]. 3. THE MOBIUS-HUCKEL CORRELATION DIAGRAMS
CONCEPT
FOR
TRANSITION
STATES
AND
Hiickel computations proved useful again in 1966 when we proposed that pericyclic systems consisted both of the usual "Hiickel" variety and also of "Mobius" topology. See Scheme 7. At the time we derived [6] a circle mnemonic for Mobius systems, quite parallel to the FrostMusulin [7] for the Hiickel systems. We showed that for reacting molecules of the pericyclic variety, at half-reaction one had either a Hiickel (zero or an even number of basis set +/- overlaps around the orbital array) or a Mobius orbital array (with an odd number of +/- overlaps). The circle mnemonic afforded the MO eigenvalues and indicated the degeneracies. For each degeneracy one has a crossing of MO's. This then permits one to draw the correlation diagram easily. Hiickel computations confirmed the validity of the derivation.
: +1.00
+2.00 ' 0.0
o.o-2.00
-1.00
Hueckel Cyclopropenyl
Moebius Cyclopropenyl
MO 1 MO's1&2 Crossing ALLOWED WITH 2 ELECTRONS
ALLOWED WITH 4 ELECTRONS
Scheme 7. Example of Mobius-Huckel Determination of Degeneracies and MO Crossings Along the Reaction Coordinate.
262 This work noted the role of such degeneracies in giving conical intersections leading S, to So. In fact, it was in 1961 that the first organic reaction correlation diagram was put forth [8]. This was for the 1,2-shift of groups in the Grovenstein-Zimmerman Rearrangement of carbanions. There proved to be a real contrast between the 1,2-shift of an alkyl compared with a phenyl with one occupied bonding MO becoming badly anti-bonding at half-migration. With phenyl migrating, for several cases, that MO became only slightly antibonding. The MO's were drawn and the correlation was in words rather than being explicitly depicted. Also, in this study hybrid orbitals were required. These were obtained analytically and then used as a basis in the Huckel computation. This early philosophy used a truncated basis set consisting of p-orbitals and hybrid orbitals. The set selected was that which included basis orbitals involved in excitation and reaction for photochemistry and those sigma hybrids which were perturbed in the reaction studied. Thus far we had relied on Huckel (one-electron) computations. A comment is required regarding this. First, at the time, available computer facilities permitted nothing more elaborate. However, a further point is relevant. This is that Huckel theory is mathematically exact; it is identical with graph theory. The eigenvalues and eigenvectors for a set of points in space are fixed and determined by the topology of the species in space. This is independent of whether the figure subject to computation is a real molecule or a mathematician's graph. What is approximate are the assumptions that a one-electron operator is sufficient and ignoring electron-electron repulsion and exchange effects. Additionally, in our use of Huckel methodology the basis orbitals, such as p-orbitals and, possibly hybrids, are parametrized only roughly. Nevertheless, molecular topology seems most often to provide a semi-quantitative, or at least a qualitative guide, to molecular structure. Thus, while more sophisticated quantum mechanics can lead to precise numerical results, the simplicity of Huckel (graph) theory most often provides a guide to expectation of what greater sophistication will afford. Further, often a pad and pen will permit a quick assessment of molecular properties. 4. SUBSEQUENT DEVELOPMENT; EARLY SCF-CI COMPUTATIONS However, in about 1967 the author had purchased a PDP-8/I minicomputer. This was followed by a more powerful PDP-11/55 model. At the time we wrote our own programming which was based on a single Slater determinantal wavefunction for So. For Ti and S, the wavefunction was as in Eq. 4 (i.e. minus for S, and plus for Tj, all normalized). Here K refers to a bonding MO and L to an antibonding one. M O , ( 1 ) " M O 2 ( 2 ) ( i -• M 0 s ( N - l ) a M O
I l
( N ) ( ' | ± |MO, ( 1 ) " M O 2 ( 2 )
p
' MOK
(N-l)
P
MO,, (N) "| ( 4 )
263 Similarly, matrix elements between singly excited configurations were used. Thus configuration interaction was possible. We used an active space of thirteen or less and zero differential overlap. However, with limited memory, we used our hard-drive disk for virtual memory [9]. The computations were of K systems and included the py orbital with integrals taken from literature values. The main application was in obtaining wavefunctions, electron densities and energies of excited states. Interestingly, in the case of the 4,4-disubstituted cyclohexadienones, the beta-beta bonding of the n-n* triplet was positive and thus bonding, while that of the K-K* triplet was negative and antibonding. The earlier Huckel computations, of course, were spin-independent but similarly had predicted the positive beta-beta bond order. One rearrangement of particular interest was the Di-Tr-Methane Rearrangement of barrelene to semibullvalene [10]. The organic mechanism is given in Eq. 5. In 1967 the best treatment of the reaction we had made use [10] of extended Huckel computations [11]. The coordinates used were obtained by mechanical measurement. Remarkably, the basic surface obtained (note Scheme 8) is qualitatively similar to that obtained from our more sophisticated, subsequent ab initio computations subsequently (vide infra).
(5)
Acetone Barrelene, So
Dirad 1,T,
Dirad 2, T, Relaxed
Semibullvalene, So
Scheme 8. Extended Huckel Hypersurface. Circular dot for So and solid dots for the excited state.
264 5. SINGLE PHOTON COUNTING COMPUTATIONAL CHALLENGE
COMPUTATIONS
AS
ANOTHER
One other use of computation at that time was in our single photon counting measurements of very rapid excited state fluorescent decay [13]. In this we obtain the profile of the excitation lamp flash as shown in the I matrix in Eq. 6 with the subscripts referring to consecutive decay times. Also we get the profile of the fluorescence emission as in the E vector [Eo Ei E2]. The example given here has only three data points for simplicity; commonly there would be 512 or more. The D vector consists of the decay function values at the increasing time intervals. Io 0
0
I, Io 0
h h
Ir
Do D, D7
(6)
The decay function might consist of a simple negative exponential, ae"kl, or a sum of exponentials. Since the rate constant k is desired, it is the vector D which is the main unknown and wanted. However the lamp decay vector I matrix is singular due to the zero value of Io, and we cannot directly solve Eq. 6 for the D vector. However, we can make an initial assumption of the value of k, and a, and determine the error in the E vector obtained relative to experiment. By solving for the derivatives of the error with respect to k and a, we can determine what changes will minimize the error. With new values of k and a in the vector D, we can again determine the errors in the E vector. Our programming continues this iteratively until there is convergence. Thus, rather than a deconvolution, the method performs a convolution. With more data points than unknowns, the method is both accurate and useful. It also works with other decay functions. 6. A CURIOUS REACTION; A DIRADICAL REACTING IN TWO WAYS USES TWO STATES One unusual set of reactions is shown in Scheme 9 [14]. The reaction of the benzodiene A afforded photoproduct F while irradiation of F led to photoproducts C and A. Clearly a diradical of structure D must be involved in both irradiations. A superficial paradox is that a common species D led in two directions depending on the reactant. SCF-CI computations, summarized in Fig.s la and lb and in the notations in Scheme 9, clarify that there are two states of D which are involved. So leads selectively to A and C while S, affords product F. What is also revealed by the computations is that a conical intersection, or avoided crossing occurs in the D to F conversion. This is seen both at the MO as well as the state level; note Fig.s la and lb.
265 Another interesting aspect of the photochemistry in the F to C conversion is the dimethylcarbon of the three-membered ring moving along the surface of the % system. If one considers the hybridization of the three-membered ring as having two-sp5 orbitals aimed at the n-surface, this is then seen reminiscent of bicycling along the surface wherein the sp5 orbitals are the "bicycle wheels". In fact, this is a type reaction seen earlier in our research. Thus, a number of examples of the bicycle rearrangement are known [15]. One example is given in Eq.s 7a and 7b. The reaction is largely stereospecific with the Exo reactant giving the Anti-spiro product and the Endo reactant giving the Syn-spiro product. From the Endo-Bicyclic reactant there is a minor loss of stereochemistry. The topology of the reactions is depicted in Fig. 2. It is seen that the minor course (B) accounts for loss of complete stereospecificity.
Scheme 9. Rearrangements via Diradicals of Common Structure but Differing in States. Empty arrows for pathways originating from D as Si and dashed arrows for pathways originating from D as So.
266
Di-π A
-562
Spiro C Bicyclic F
-562
Fig.s la and lb. Correlation of MO's (in la) and States (in lb).
Ph
Ph
hu
H
hu (7a.b)
Ph'
Ph'
Exo-Bicyclic
Anti-Spiro (Major Product)
Endo-Bicyclic
Syn-Spiro
(Minor Product) from the Exo)
Fig. 2. Bicycle Topology; major (A) and minor (B).
267 Interestingly, minor amounts of benzenoid by-products derive uniquely from intermediate biradical intermediates along the topology shown. Thus the 2,5-diphenyltoluene, 3,4diphenyltoluene and 2,4-diphenyltoluene formed in minor amounts, give evidence for the reality of the three diradicals shown in Fig. 3.
From Main Path A
From Minor Path B
Fig. 3. Bicycle Derived Diradicals and Their Products. These 1,4-diradicals tend to undergo a Grob fragmentation, and we have shown the Grob fragmentation to occur only from So and not from S| diradicals [16]. Conical So-S, intersections were found for the diradicals by use of the SCF-CI computations discussed above. As a consequence we can conclude that the bicycle rearrangement occurs from Si diradicals D. Some of those subsequent diradicals which decay, rather than proceeding onward to the spiro products, are lost to the benzenoid by-products. Interestingly, the intermediate Diradicals, such as these in Fig. 3 and also D in Scheme 9, are the same species encountered in the Di-jt-Methane Rearrangement. In Scheme 9, the Si state of D actually comes from the Di-7t-Methane reactant A. The Di-71-Methane Rearrangement originally was discovered in the barrelene to semibullvalene rearrangement [10] as noted above. However, this case uses the triplet. Also, as noted, it utilizes diradical species, the cyclopropyldicarbinyl diradical, which are common to the bicycle rearrangement. Thus our SCF-CI computations have been applied to this rearrangement independently [14,15a, 15b, 16,17,]. The triptych mode of display of correlation diagrams has
268 been employed in a number of these studies. In these, conical crossings and/or avoided crossings have been uncovered of the type discussed for one example above. One example of the Di-71-Methane Rearrangement is shown in Eq. 8 [15c]. The regioselectivity favoring ring opening process "a" over "b" of the cyclopropyldicabinyl diradical is of intrinsic interest. The computations done on this system were carried out with just one carbomethoxyl group and, in place of the geminal diphenyls, we used single phenyl groups. The lower energy energy path corresponded to ring opening process "a" and also led to a conical intersection. Interestingly, the triplet process opening bond "b" rather than "a", and this was ascribed to the larger exchange integral for that process.
Ph
Ph Ph C O , Me CO,Me
7. QUANTITATIVE METHODOLOGY
Ph
SOLID
CO 2 Me
STATE
CO 2 Me CO 2 Me
STUDIES
AND
Ph
Ph
CO2Me CO 2 Me
COMPUTATIONAL
Solid state photochemistry had been known for a long time often to afford photoproducts which most often differ from those formed in solution [18]. Our aim was to put this photochemistry on a quantitative basis. With the coordinates of all atoms known from the X-ray data on each crystal, it was possible to define what we termed a "mini-crystal lattice", a portion of the crystal lattice large enough to be of use but small enough to permit computation. Our initial efforts utilized molecular mechanics (MM3) to assess overlap of alternative reaction intermediates with their neighbors and to determine the energy of the mini-lattice overall. A schematic mini-lattice is depicted in Fig. 4.[19,20] However, it was clear that molecular mechanics would include only steric but not electronic effects. One might argue that it is heresy to utilize MM3 with open-shell species such as triplet diradical intermediates. Thus an alternative was devised. A simple program, "Pairs", was written [21]. This obtained all distances between pairs of atoms, one atom in the reacting species in the mini-lattice and one atom in the surrounding molecules. These distances were then sorted with distance increasing. A cut-off was then selected and all the more distant atoms of the neighbors were computationally annihilated. This left neighboring atoms with "dangling" free valencies. The hydrogens were therefore computationally changed to helium atoms and the carbons, oxygens and nitrogens were changed to neons. Then the reacting intermediate imbedded in the inert gas shell was subjected to a Gaussian98 [22] ab initio computation with the shell kept fixed
269
/ • •
.-• /
/
/
/
,-".-'/
Fig. 4. A Mini-Crystal-Lattice. Each "R" represents a reactant molecule and "I" is an intermediate leading to one of the possible photoproducts. Other computational methods were least motion and molecular volume increase in formation of either product or the intermediate. These alternatives did not not correlate well with experiment. and the intermediate permitted to geometry optimize. These results, while including electronic effects, nevertheless led qualitatively to the same predictions of preferred reaction pathways. Nevertheless, this "inert gas shell" approach was of some interest [21]. One thing this does demonstrate is that steric influences are dominant over electronic factors and that in predicting the reaction course in the crystalline state, molecular mechanics approximation was not unreasonable. Nevertheless, one would like to explore this question further. At the time, Morkuma's Oniom QM/MM programming [23] had become available in Gaussian 98. We used this [24] with three concentric shells. The inner-most shell had the reacting molecule and was subjected to ab initio, Gaussian98 geometry optimization. The next layer was treated with molecular mechanics and also permitted to geometry optimize to account for relaxation effects. The outer-most layer was also treated with molecular mechanics but kept rigid. Additionally, the size of the middle layer was varied to ascertain how many molecules were needed in this shell for relaxation to lower the overall energy. It turned out that about one layer of molecules is really important. It was found that this ab initio/molecular mechanics treatment was in qualitative agreement with the original molecular mechanics and the subsequent inert gas shell methods. With these methods available we proceeded to study host-guest complexes [24] as well as further phenomena. One particularly exciting result was the finding that solid-state photochemistry proceeds in stages. [25] At a given point, commonly when each reactant molecule has one product molecule adjacent, there is a phase change and new products arise [26]. Often these are products not obtainable in other ways.
270 8. MORE SOPHISTICATED EFFORTS ON THE META-EFFECT. Our early work on the meta-effect had been limited by available computer methodology of the time. With Gaussian98 available, we carried out ab initio computations for S, of p-methylbenzyl acetate, m-methoxybenzyl acetate and 3,5-dimethoxybenzyl acetate systems at the CASSCF(8,8)/6-31G* level. Computations were carried on the ion-pairs and on the radical pairs. The separate cations and acetate energies were obtained and a method was devised for obtaining the ion-pairing energy. This total was compared with the sum of the benzylic and acetoxy radicals. The benzyl species were taken in the first excited state (i.e. S, and Di) while acetate anion and acetoxyl radical were taken in their ground states. The energetic preference was for heterolysis rather than homolysis. It was found that excited state heterolysis to form anion pairs is favored over homolysis to form a radical pairs in the case of the meta isomers. Also for heterolysis meta substitution is favored over para. Additionally, for homolysis, p-substitution leads to a slight energy lowering, in contrast with experiment. In the case of the 3,5dimethoxybenzyl cation a conical intersection of S, with So was encountered. [27] 9. THE DELTA DENSITY PREDICTION OF GROUND STATE AND EXCITED STATE REACTIVITY. The Delta-Density method was developed [28] to predict reactivity. Originally the method was applied to photochemistry but proved more general. The basic idea is to utilize the density matrix derived for a molecule, A, and then to consider that molecule perturbed in some fashion - by introduction of a photon, an electron to afford a radical anion, by loss of an electron to give a radical-cation, or by some small molecular deformation. The density matrix of the perturbed molecule, B, is also available from computation. Then the delta-density matrix is defined as in Eq. 9.
AB
or
DM D l 2 D r , D21 D 22 D 21
Dn D,2 D n D21 D, 2 D 2 3 D 3 , D32 D33
AD, | AD,2 AD,3 AD21 AD 2 2 AD 23 AD 3i AD 32 AD 33
AD r t = D b l t S b r t -D a r t S a r t
D3, D32 D33
B
(9a)
A
(9b)
271 For this we use the Weinhold natural hybrid orbitals [29] as our basis and to account for basis orbital orientation, the overlap between the hybrids r and t are included. The basic philosophy of the Delta-Density method is that the off-diagonal density matrix elements are effectively bond orders. On perturbation of molecule A, without geometry change, the resulting species B is formed with a number of bonds with less than ideal bond lengths or valence angles. These correspond to negative elements in the delta density matrix and to bonds which can relax by stretching. Positive elements indicate molecular relaxation by bond formation. Thus, one can predict which bonds are likely to be broken; these are the most negative ones. Similarly, new bonding is likely to occur with strongly positive delta density elements. In Scheme 10 there are listed fifteen reactants for the more common photochemical reactions. The most negative delta density matrix elements are labeled "a" and the next most negative elements are labeled "b". Near zero elements are labeled "c", while positive off-diagonal elements are labeled "f". One center positive diagonal elements are labeled "h". In each case, the bonds labeled "a" are, indeed those which are severed in the photochemistry. Thus, cyclopropyl ketones, as 1 and 6 undergo ring opening of with scission of the a-fS bond to the more substituted a-carbon. Ketones (e.g. 2, 7 and camphor 12) undergo the Norrish Type I cleavage. The cyclobutanones 3 and 8 undergo the Yates ring expansion to an oxacarbene.
t-Bu
/b
11
Scheme 10. Excited State Delta Density Predictions positive element.
272 The reaction has part of its mechanism involving a Norrish-like scission of the bond to the carbonyl carbon and is also predicted. Ketones bearing an a-substituent expel that substituent as an anion in polar media or as a radical in non-polar media. Again, this is the bond which has the most negative Delta-Density value (Note 5). The Di-71-Methane Rearrangement of barrelene 11 to semibullvalene has dissipation of two 71-bonds and vinyl-vinyl bridging as the initial step of its rearrangement, again predicted. Photochemical raeemization of optically active biphenyls results from planarization of the phenyl-phenyl bond in Si and here we see an increase in the aryl-aryl bond order in biphenyl species 4. It has been noted by a referee that in our research we had suggested that the course of reactions was determined by avoidance of excited state energy barriers and by the presence of conical intersections. The referee wondered then how this criterion based on the vertical Franck-Condon excited state could afford predictions. This is an interesting point. The answer is an important one, namely that the Delta-Density prediction merely gives the "primary photochemical step". This provides a necessary but not sufficient criterion for photochemical reactivity; without the primary step no photochemistry can occur. The Delta-Density method proved applicable to chemistry other than photochemical. Thus, introduction of an electron to afford a radical-anion or extraction of an electron to give a radicalcation leads to species with non-zero Delta-Density elements for the new species with the original geometry. These cases are quite parallel to the photochemistry examples, except that here the species in introduced or withdrawn is an electron rather than a photon. The application to the odd-electron species is illustrated in Eq. 10,
o + e"
(10)
Product
where on electron introduction it is the more substituted bond 2-3 which is severed. The carbonyl carbon of ketyl B is both electron-rich and odd-electron in character. But it is the odd-electron character which is dominant and leads to the more stable of two alternative opened species, this having a disubstituted carbon. Had the anionic character been controlling, ring opening to give a primary carbanion would have resulted.
273 10. SOME ADDITIONAL CHEMISTRY WITH INTERESTING INCLUDING SPIN-ORBIT COUPLING
COMPUTATIONS
The present author in early efforts had already suggested that the course of photochemical reactions was determined by excited state energy barriers and by conical intersections and/or points of intersystem crossing (vide supra). More recently he had added the requirement of a primary photochemical process of the Franck-Condon excited state (again note above). Thus, it was of interest to consider some of these factors further.
NB
Scheme 11 . Mechanisms of Benzobarrelene Rearrangements. The Di-ir-Methane Rearrangement of the barrelenes was selected for study [30]. The dimethylbenzobarrelene and the corresponding naphthobarrelene in Scheme 11 were selected for study. For reasons of space, structures are drawn just for the benzobarrelene (B).The basic mechanism is that given in our much earlier work. Two independent routes to the triplet "Diradical I" species (DR1). The first began with the barrelene B. The other began with the azo precursor A. We see that DR1 has two competing modes of three-ring opening, a and b, and this partition characterizes the species. Independent of precursor, B or A, in the naphthobarrelene case, the same ratio (8.3:1) of semibullvalene photophotoducts (SB1 and SB2) was formed. For the naphthobarrelene the triplet sensitizer, xanthone (ET = 74 kcal/mol) was used, while and for the azo precursor A benzil (ET = 53 kcal/mole) was employed. Both for the naphtho- and the benzo-barrelenes, the product ratio was independent of the reactant, B or A.
274 If DR1 were a transition state or a mere point on the triplet hypersurface then the partition between process "a" and "b" would differ depending on the direction of approach. Since the regioselectivity of the reaction of DR1 was independent of its source, we can conclude that it is a thermally equilibrated intermediate, Ti. Strikingly, with a higher energy xanthone (74 kcal/mole) sensitizer, the azo-naphtho reactant A afforded a quite different product ratio of 2.6:1. It is clear that the azo route with higher energy sensitizers proceeds via T2. Computations were carried out on the barrelene, benzobarrelene and and napthobarrelene triplets (Ti) using Gaussian98 and CASSCF((6,6)/6-31G*. A striking feature was one stretched double bond (1.51 A) with the remaining double bonds being more normal (1.34 A). The Jahn-Teller distortion results from a singly occupied degenerate pair in these cases. Computations for "Diradical 2" DR2 also led to an energy minimum. The triplets of DR1 and DR2 were found to cross the ground state surface and yet the quantum yields were of the order of 0.5, which signifies that Diradical I cannot be the site of major intersystem crossing to ground state. So of B is known to revert to reactant barrelenes. However, spin-orbit coupling computations utilizing GAMESS [33] revealed that spin-orbit coupling for DR1 is only l/4th of that for DR2. 11. SPIN-ORBIT COUPLING DISSECTION One question which arose was what structural features in a molecule or a reacting species gives rise to spin-orbit coupling. Our approach [31,32] to this problem involved doing some programming using the module in GAMESS [33] where the total SOC was computed. The approach was to take the loop as it obtained local contributions and collect these as basis orbital pairs. For this to be helpful, the Weinhold NHO (Natural Hybrid Orbital) basis [29] was used. What was found was that the main contributions come from geminal orbital pairs which tend to be geometrically orthogonal. In diradicals, the p-orbital with the odd electron and orthogonal NHO's of sigma bonds at the same center contribute the most. One interesting application of this approach was in the Type B cyclohexenone rearrangement. It was found that in this triplet process SOC become most heavy as the product geometry is reached with intersystem crossing then occurring as the last bond is being formed. Note Scheme 12.
275
Ph Closure with ISC
Phi
Ph
Scheme 12. Mechanism of the Type B Cyclohexenone Rearrangement.
12. CONCLUSION This presentation makes clear the role of computation in organic photochemistry. The computation ranges from simple Huckel to sophisticated CASSCF treatments. The upper level of computation has been a function of computer facilities available in a given period. However, it also is clear that Huckel treatments have not lost their importance in related sophisticated computations to molecular structure.
REFERENCES [1] Havinga, E.; De Jong, R. O.; Dorst, W., Rec. Trav. Chim., 1956,75, 378. [2] Barton, D. H. R.; DeMayo, P.; Shafiq, M., Proc. Chem. Soc, London, 1958, 205 which first clarified the reaction course. [3] (a) "A Mechanistic Approach to Organic Photochemistry", H. E. Zimmerman, Seventeenth National Organic Symposium of the Amer. Chem. Soc, Bloomington, Indiana, 1961, pgs. 31-41; (b). "Mechanistic Organic Photochemistry. IV. Photochemical Rearrangements of 4,4-Diphenylcyclohexadienone", H. E. Zimmerman and D. I. Schuster, J. Amer. Chem. Soc, 1962, 84 , 4527-4540; (c) "Mechanistic Organic Photochemistry. ", H. E. Zimmerman, Tetrahedron ,1963, Suppl. 2, 19, 393-401; (d) "A New Approach to Mechanistic Organic Photochemistry", Zimmerman, H. E., "Advances in Photochemistry", Editors: A. Noyes, Jr., G. S. Hammond and J. N. Pitts, Jr., Interscience, Vol. 1, 183-208, 1963; (e) "Interpretation of Some Organic Photochemistry", H. E. Zimmerman, Science, 1966, 153, 837-844.
276 [4] "Mechanistic Organic Photochemistry. VIII. Identification of the n-71* Triplet in Rearrangement of 4,4-Diphenylcyclohexadienone", Zimmerman, H. E.;Swenton, J. S. J. Amer. Chem. Soc. , 1964 , 86 , 1436-1437. [5] "A General Theory of Photochemical Reactions. VII. Mechanisms of Epoxy Ketone Reactions", Zimmerman, H. E.; Wilson, J. W., J. Amer. Chem. Soc., 1964, 86, 4036-4042. [6] "On Molecular Orbital Correlation Diagrams, the Occurrence of Moebius Systems in Cyclization Reactions, and Factors Controlling Ground and Excited State Reactions. I", H. E. Zimmerman, J. Amer. Chem. Soc., 1966 , 88 , 1564-1565. [7] Frost, A.; Musulin, B., J. Chem. Phys., 21, 572 (1953). [8] "Carbanion Rearrangements. II", Zimmerman, H. E.; Zweig, A., J. Amer. Chem. Soc, 1961 , 83 , 1196-1213. [9] Electronically Excited State Structures", Zimmerman, H.E.; Binkley, R. W.; McCullough, J. J; Zimmerman, G. A., J. Amer. Chem. Soc., 1967 , 89 , 6589-6595. [10] "Mechanistic Organic Photochemistry. XXIV. The Mechanism of the Conversion of Barrelene to Semibullvalene. A General Photochemical Process", H. E. Zimmerman, R. W. Binkley, R. S. Givens, and M. A. Sherwin, J. Amer. Chem. Soc., 1967, 89 , 3932-3933. [11] Zimmerman, H. E.; Zhu, Z. General theoretical treatments of solid-state photochemical rearrangements and a variety of contrasting crystal versus solution photochemistry. J. Am. Chem. Soc. 1995, 117,5245-5262. [12] Hoffmann, R.; Lipscome, W. N., J. Chem. Phys., 1962, 37, 2872. [13] "Mechanisms of Electron Demotion. Direct Measurement of Internal Conversion and Intersystem Crossing Rates. Mechanistic Organic Photochemistry, H. E. Zimmerman, K. S. Kamm and D. P. Werthemann, J. Amer. Chem. Soc, 1975, 97, 3718-3725. [14] "The Bicycle Rearrangement: Relationship to the Di-n-Methane Rearrangement and Control by Bifunnel Distortion. Mechanistic and Exploratory Organic Photochemistry," Zimmerman, H. E.;Factor, R. E. J. Amer. Chem. Soc, 1980, 102, 3538-3548. [15] (a) "Topology of the Photochemical Bicycle Reaction. Mechanistic and Exploratory Organic Photochemistry", Zimmerman, H. E.; Cutler, T. P. J.C.S. Chemical Communications, 1978, 232-234;(b) "Generality of the Photochemical Bicycle Rearrangement. Exploratory and Mechanistic Organic Photochemistry", Zimmerman, H. E.; Cutler, T. P. J. Org. Chem., 1978, 43, 3283-3303; (c) "Di-7iMethane Hypersurfaces and Reactivity; Multiplicity and Regioselectivity; Relationship Between the Di-7rMethane and Bicycle Rearrangements", Zimmerman, H. E.; Factor, R. E. Tetrahedron, 1981, 37, Supplement 1, 125-141; (d) "The Bicycle Rearrangement. A Review", Zimmerman, H. E. Chimia, 1982, 36, 423-428. [16] "Unusual Organic Photochemistry Effected by Cyano and Methoxy Substitution. Exploratory and Mechanistic Organic Photochemistry", Zimmerman, H. E.; Armesto, D.; Amezua, M. G.; Gannett, T. P.; Johnson, R. P., J. Amer. Chem. Soc., 1979, 101, 6367-6383. [17] "The Bicycle Rearrangement. A Review", H. E. Zimmerman, Chimia , 1982, 36 , 423-428. [18] Much elegant solid-state research preceded ours and is not surveyed here with out emphasis on our computational methodology. These references are found in our solid-state publications.
277 [19] (a) "Confinement Control in Solid State Photochemistry; Photochemistry in a Box", Zimmerman, H. E.; Zuraw, M. J., J. Am. Chem. Soc, 1989, 111 ,2358-2361. (b) "Photochemistry in a Box; Photochemical Reactions of Molecules Entrapped in Crystal Lattices; Mechanistic and Exploratory Organic Photochemistry", Zimmerman, H. E.; Zuraw, M. J., J. Am. Chem. Soc, 1989, 111 7974-7989. [20] (a)"A General Predictor for Photoreactivity in Crystal Lattices: Molecular Mechanics in Crystalline Media and Lock and Key Control; Reaction Examples" Zimmerman, H. E.; Zhu, Z., J. Am. Chem. Soc, 1994, 116 9757-9758; (b) "General Theoretical Treatments of Solid-State Photochemical Rearrangements; And, A Variety of Contrasting Crystal Versus Solution Photochemistry", Zimmerman, H. E.; Zhu, Z., J. Am. Chem. Soc. 1995, 117, 5245-5262. [21] (a) Zimmerman, H. E.; Sebek, P.; Zhu, Z. Ab initio computations of reacting species in crystal lattices; mechanistic and exploratory organic photochemistry. J. Am. Chem. Soc. 1998, 120, 8549-8550; (b) Zimmerman, H. E.; Sebek, P. Photochemistry in crystalline cage. Control of the type-B bicyclic reaction course: mechanistic and exploratory organic photochemistry. J. Am. Chem. Soc. 1997, 119, 3677-3690. [22] Gaussian 98, Revision A.6. Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Zakrzewski, V. G.; Montgomery, Jr., J. A.; Stratmann, R. E.; Burant, J. C; Millam, J. M.; Daniels, A. D.; Kudin, K. N.; Strain, M. C; Farkas, O.; Tomasi, J; Barone, V.; Cossi, M.; Cammi, R; Mennucci, B.; Pomelli, C; Adamo, C; Clifford, C; Ochterski, J.; Petersson, G. A.; Ayala, P. Y.; Cui, Q.; Morokuma, K.; Malick, D. K.; Rabuck, A. D.; Raghavachari, K.; Foresman, J. B.; Cioslowski, J.; Ortiz, J. V.; Stefanov, B. B.; Liu, G.; Liashenko, A.; Piskorz, P.; Komaromi, I.; Gomperts, R.; Martin, R. L.; Fox, D. J.; Keith, T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.; Gonzalez, C; Challacombe, M.; Gill, P. M. W.; Johnson, B.; Chen, W.; Wong, M. W.; Andres, J. L.; Gonzalez, C; Head-Gordon, M.; Replogle, E. S.; Pople, J. A. Gaussian, Inc., Pittsburgh PA, 1998. [23] (a) Maseras, F.; Morokuma, K. IMOMM: a new integrated ab initio + molecular mechanics geometry optimization scheme of equilibrium structures and transition states. J. Comput. Chem. 1995, 16, 11701179; (b) Matsubara, T.; Sieber. S.; Morokuma, K. A test of the new "integrated MO + MM " (IMOMM) method for the conformational energy of ethane and n-butane. Int. J. Quantum Chem. 1996, 60, 11011109. [24] Zimmerman, H. E.; Alabugin, I. V.; Smolenskaya, V. N. Experimental and theoretical host-guest photochemistry; control of reactivity with host variation and theoretical treatment with a stress shaped reaction cavity; mechanistic and exploratory organic photochemistry. Tetrahedron 2000, 56, 6821-5831. [25] "Crystal Lattice Photochemistry Often Proceeds in Discrete Stages; Mechanistic and Exploratory Organic Photochemistry", Zimmerman, H. E.; Nesterov, E. E. Organic Letters, 2000, 2, 1169-1171. [26] Zimmerman, H. E.; Alabugin, I. V.; Chen, W.; Zhu, Z. Dramatic effect of crystal morphology on solid state reaction course: control by crystal disorder; mechanistic and exploratory organic photochemistry. J. Am. Chem. Soc. 1999, 121, 11930-11931. [27] "The Meta Effect in Organic Photochemistry; Mechanistic and Exploratory Organic Photochemistry", Zimmerman, H. E., J. Am. Chem. Soc. 1995, 117, 8988-8991. "The Ortho-Meta Effect in Organic Photochemistry; Mechanistic and Exploratory Organic Photochemistry", Zimmerman, H. E., J. Phys. Chem., 1998, 102 5616-5621.
278 [28] (a)"Excited State Energy Distribution and Redistribution and Chemical Reactivity; Mechanistic and Exploratory Organic Photochemistry", Zimmerman, H. E.; Alabugin, I. V. J. Am. Chem. Soc, 2000, 122, 952-953; (b) "Energy Distribution and Redistribution and Chemical Reactivity. The Generalized Delta Overlap-Density Method for Ground State and Electron Transfer Reactions; A new Quantitative Counterpart of Electron Pushing", Zimmerman, H. E.; Alabugin, I. V. J. Am. Chem. Soc. 2001, 121, 2265-2270 [29] (a) Foster, J. P.; Weinhold, F., J. Am. Chem. Soc, 1980, 102, 7211-7218; (b) Reed, A.; Curtiss, L. A.; Weinhold, F. Chem. Rev., 1988, 88, 899-926. [30] "Excited State Reactivity as a Function of Diradical Structure; Evidence for Two Triplet Cyclopropyldicarbinyl Diradical Intermediates With Differing Reactivity", Zimmerman, H. E.; Kutateladze, A. G.; Maekawa, Y.; Mangette J. E. J. Am. Chem. Soc, 1994, 116, 9795-9796. [31] "Novel Dissection Analysis of Spin-Orbit Coupling in the Type B Cyclohexenone Photorearrangement. What Controls Photoreactivity. Mechanistic and Exploratory Organic Photochemistry", Zimmerman, H. E.; Kutateladze, A. G. J. Org Chem. 1995, 60, 6008-6009. [32] "Novel Dissection Analysis of Spin-Orbit Coupling in the Type B Cyclohexenone Photorearrangement. What Controls Photoreactivity. Mechanistic and Exploratory Organic Photochemistry", Zimmerman, H. E.; Kutateladze, A. G. J. Org Chem. 1995, 60, 6008-6009. [33] Schmidt, M. W., et. al. J. Comput. Chem., 1993 14, 1347-1363. ACKNOWLEDGMENT It is a pleasure to acknowledge the support of the National Science Foundation without which these studies would not have been possible.
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 Elsevier B .V. All rights reserved
279
IX. Calculations of Electronic Spectra of Transition Metal Complexes K. Pierloot Department of Chemistry, University of Leuven, Celestijnenlaan 200F, B-3001 Leuven, Belgium 1. INTRODUCTION If this chapter would have been written ten-twenty years ago this introduction might traditionally have started with a sentence like: "Transition metals and their compounds still present a major challenge to ab initio quantum chemistry ...". Today, the magic surrounding these difficult elements has to a large extent been lifted. This of course has everything to do with the development of density functional theory (DFT), a method which turned out to work marvelously well also for molecules containing transition metals (TM), even in cases that were traditionally considered to be quite difficult to treat computationally (because of the presence of important nondynamic correlation effects), e.g. organometallic systems, with Cr(CO)6 and ferrocene as prototypes [1-3] . The possibility of treating TM systems on a routine basis with methods that could generally be considered trustworthy, at least until proven differently, has led to a drastic increase of computations in the field, especially focusing on organometallic and biological molecules. Molecular properties that are today treated on an almost routine basis include ground state structures, vibrational frequencies and reaction energies [4]. Furthermore, the availability of very fast computers has enabled such computations on systems of real chemical interest. Things are still quite different when considering electronically excited states. Theoretical calculations of electronic spectroscopy have in general today not reached a "black box" status. The specific electronic structure of TM complexes, originating from a partially filled d shell surrounded by a number of potential electron donating or accepting ligands, adds a number of additional intricacies to these computations, often making them accessible only to researchers with experience both in the field of coordination and computational chemistry. This is true even when applying DFT, be it under its time-independent formalism, usually denoted as ASCF [5], or its time-dependent formalism TDDFT [6-8]. Test results obtained with these methods in transition metal spectroscopy are certainly promising in some cases, but have also pointed to quite a number of remaining problems (e.g. the spectra of MnC% (see section 4) and the blue copper proteins (see section 5)).
280 An introductory overview of the various electronically excited states to be met in TM complexes most conveniently starts from a simple ligand field picture. According to this picture the TM d orbitals are split in energy by the presence of the coordinating groups, the so-called ligand field environment. Depending on the extent of the splitting the available electrons are spread over the different d-orbitals giving rise to a high spin ground state, or are placed pairwise in the lowest lying orbitals, thus producing a low-spin (singlet or doublet) ground state. By absorption of light the electrons are excited from the filled into the singly occupied or unoccupied d orbitals. Since the splittings of the d shell are critically dependent on the number, character and position of the ligands, these ligand field (LF) transitions provide sensitive probes of the electronic structure and geometric surrounding of TM ions. LF transitions are parity forbidden, and as such generally have a low intensity in an absorption spectrum (but can be intense in CD spectrum or low-temperature MCD spectra [9]). Overlapping the LF transitions and to higher energy in the visible/UV spectral regions are the charge transfer (CT) transitions, often electric dipole allowed and therefore more intense in absorption spectra than LF transitions. CT transitions either involve the excitation of an electron from one of the filled d orbitals into a ligand centered unoccupied orbital (MLCT), or the excitation of an electron from filled valence orbitals on the ligands into one of the singly occupied or empty d orbitals (LMCT). MLCT transitions are typically observed in organometallic compounds, containing organic ligands with low-lying (relative to the d shell) TT* orbitals. On the other hand, electronic spectra of TM containing biological molecules often contain one or several low-lying LMCT transitions. The position and intensities of CT transitions are sensitive probes of the character of specific metal-ligand interactions. Low energy CT transition are the reflection of close-lying ligand and metal valence orbitals. Moreover, the intensity of a CT transition increases with the overlap extent of the donor and acceptor orbitals. Low energy, intense CT transitions therefore usually reflect highly covalent metal-ligand bonds. The list of possible electronic excitations in TM spectra is not complete without mentioning also intra-ligand excitations. With some exceptions [10, 11] such transitions do not occur in the low-energy part of the electronic spectra. They are not further considered in this chapter. The intention of this chapter is not to provide a detailed overview of possible computational methods for the calculation of excited states in transition metal complexes. Such an overview has recently been given in a review by C. Daniel [12]. Instead we would like to present a few selected examples out of our recent work, which we believe clearly illustrate the distinct power of accurate computations for the interpretation of experimental spectroscopic features and the discussion of the specific electronic structure and metal-ligand bonding interactions giving rise to these features. The examples are taken from various fields of chemistry and include different types of excitations. In section 3 we will focus on the ligand field spectra of Co(II) and Cu(II) as probes of the coordination environment of these ions when bound to a zeolite surface. Section 4 contains a discussion of the long-standing and heavily debated spectrum of the permaganate ion, i.e. a pure charge transfer spectrum. In section 5 we wil discuss the electronic structure of two important classes of redox proteins, namely the blue copper proteins and a series of mononuclear oxomolybdenum enzymes, modeled by the (Tp)MoO(bdt) compound (with Tp =
281 hydrotris(l-pyrazolyl)borate and bdt = benzenedithiolate). In both cases, the considered electronic spectra are built from a mixtures of LF states and a number of low-lying LMCT states. All excited state calculations reported in this chapter were performed by means of the CASPT2//CASSCF method [13] as implemented in the Molcas software [14]. Ground state structures were either taken from experimental X-ray diffraction (XRD) data or calculated by means of the B3LYP-DFT method, using either the Mulliken [15] or Turbomole [16] codes. Whenever available, comparisons are made to the results obtained by other computational methods. In the following section, we will give a short introduction to the CASPT2/CASSCF method, focusing on those aspects that are of particular importance when applying the method for computations of electronic spectra of transition metal complexes. 2. CALCULATION OF ELECTRONIC SPECTRA OF TM COMPLEXES WITH MULTICONFIGURATIONAL PERTURBATION THEORY The CASPT2/CASSCF method is a mixed variational/perturbational procedure. In the first, C ASSCF step, a zero-order wavefunction is constructed by distributing a limited number of valence electrons, i.e. the active electrons, over a limited range of valence orbitals, i.e. the active space, in all possible ways consistent with the spin and symmetry of the considered state. By variationally optimizing both the orbitals and the CI coefficients of this (limited) full CI wavefunction the best possible treatment of strong correlation effects can be obtained. In the second, C ASPT2 step, remaining correlation effects can then be treated by a second-order perturbational approach. This second step is quite straightforward. However, since perturbational theory is not able to cope with strong, static correlation effects, it is of crucial importance that all such effects are treated at the CASSCF level by including the appropriate orbitals in the active space. In a recent paper [17] we have illustrated by a few examples (Fe(CO)5, Fe(II)-porphin) how the omission of just one important orbital from the active space may lead to errors of several thousands of wavenumbers on excitation energies calculated by the CASPT2 method. The validity of the CASPT2 treatment entirely depends on the quality of the CASSCF wavefunction. A thorough discussion of important static correlation effects in TM complexes has been presented in a recent book chapter [18]. These effects can be subdivided into two categories: • The so-called double-shell effect is an atomic, radial correlation effect, manifested in first-row TM atoms or ions containing a large number of electrons in a compact 3d shell (e.g. the Ni atom [19]). This effect should be treated at the CASSCF level by including in the active space a second d shell, usually denoted as 3d' or Ad. As was shown in ref. [18] a proper treatment of the double-shell effect is of greater importance when considering electronic transitions involving a change of the 3d occupation number (e.g. 3d —¥ As or charge-transfer transitions) than for transitions within the 3d shell (i.e. LF transitions). Furthermore, the effect is drastically reduced when going to the second and third TM series.
282 • Covalent metal-ligand bonds, i.e. bonds involving a significant admixture of ligand character into the metal d orbitals and vice versa, invariably give rise to strong correlation effects. These effects should be accounted for in the zero-order wavefunction by including in the active space both the bonding and antibonding combination of valence orbitals involved in the covalent bond(s). Qualitatively speaking, covalent metal-ligand bonds arise when the energy separation between the metal d and ligand valence orbitals is small and/or when both orbitals are strongly overlapping. For the same ligand L, M-L covalency and concomitant correlation effects therefore increase (a) with an increasing formal charge on the metal M, and (b) when moving from left to right in the TM series. Test calculations in ref. [18] also indicated a significant reduction of static correlation effects connected to M-L covalency when going to the second and third TM series. This fact, together with the strongly reduced double-shell effect in the higher TM series, explains the increasing succes of single-reference methods such as MP2 or MCPF for complexes containing these heavier metals [20-22]. As concerns the ligand L, M-L covalency generally increases as L becomes less electronegative and more polarizable or "soft". When considering electronic spectra, the selection of valence orbitals to be included in the active space should be based on two criteria, i.e. (a) all strong correlation effects should be incorporated, and (b) any orbital that gets a population number different from either zero or two in any of the considered states should be included. At first sight it might seem as if both criteria should in fact lead to the same selection of valence orbitals, i.e. the highest occupied and lowest unoccupied orbitals. This is however not always the case in TM chemistry, since the orbitals involved in static correlation effects either consist of a double d shell or of pairs of bonding and antibonding combinations of the metal d and ligand orbitals, whereas lowest-lying charge-transfer states may instead involve excitation of an electron from or into non-bonding ligand orbital(s). As an illustrative example, let us look at the octahedral Cr(CO)g complex. As is well known, the metal-ligand bonding in this organometallic compound is built from a-donation from CO into the empty Cr 3d orbitals, counteracted by vr-backdonation from the filled Cr 3d orbitals into CO TT* . Both bonding types are covalent and give rise to important correlation effects [23]. These effects may be treated at the CASSCF level by considering an active space consisting of the bonding an antibonding combination of CO a and Cr (d Z2 ,dx2_y2) within eg symmetry and of Cr (dxy,d.xz,dyz) and CO TT* within t2g symmetry. However, the lowest-lying excited states in the absorption spectrum of Cr(CO)e arise from MLCT transitions from the occupied t2g (predominantly 3d) shell into the lowest unoccupied i l u , t2u (CO TT*) shells [24]. The latter are non-bonding and therefore lower in energy than the antibonding t 2g (predominantly CO TT*) orbitals. A multiconfigurational treatment of the absorption spectrum of Cr(CO)e therefore requires an active space consisting of at least sixteen orbitals, which is at the limit of today's computing power. As a viable and cheaper alternative, the calculation of different excited states may be performed using different active spaces, by selectively including only those orbitals that are either populated or depopulated in each state considered [24]. Complexes containing transition metals with exceptionally high formal oxidation states (i.e. +VI or higher) present a special problem for CASPT2//CASSCF, as for other ab initio meth-
283
ods. Well-known examples are CrFG, CrO^ and MnO4 . This may again easily be understood by considering a qualitative orbital picture, in which now the metal 3d and ligand valence orbitals literally become near-degenerate [17, 18]. As a consequence, static correlation effects on the M-L bonds become extremely important in these complexes. Furthermore, these effects are no longer limited to the bonding-antibonding pairs of covalently interacting orbitals, but rather involve all ligand valence orbitals. In CrF6 for example important contributions to the ground state wavefunction are observed from excitations into Cr 3d out of all eighteen F 2p orbitals [17]. In section 4 we will discuss the results obtained for the electronic spectrum of MnO^", using CASPT2 based on a zero-order wavefunction comprising 2.2 million configuration state functions (9.5 million determinants), built from an active space of seventeen (five Mn 3d + twelve O 2p) orbitals. Such calculations are really at the limit of what can be handled with presently available computing hardware/software. Still it is important to note that even for exceptionally hard cases such as MnO J the combined variational/perturbational treatment offered by the CASPT2 method is quite successful in providing an accurate treatment of all low-lying excited states in the spectrum (see further in section 4). Once an appropriate zero-order wavefunction is obtained, the perturbational part of the calculation does not require much extra attention, but for the choice of appropriate basis sets to accurately describe dynamic correlation effects. In order to limit basis set superposition errors, core electrons are preferebly not included in the correlation treatment. In this respect, the metal semi-core orbitals (e.g. 3s,3p in first-row, 4s,4p in second-row TM, ...) may deserve some special care. It has been shown [25] that correlation of these electrons may give very important contributions to the relative energy of different states. As such, the semi-core orbitals should preferably be included in the CASPT2 correlation treatment. However, standard transition metal basis sets are not always well-suited for this purpose. The use of such basis sets without extra precautions may therefore give rise to huge basis superposition errors, when considering for example M-L bond dissocation energies [26]. The problem is less stringent in calculations of vertical excitation energies, but then of course semi-core correlation effects can not be described to their full extent unless the basis sets used contain the necessary functions to do so. This may be accomplished by uncontracting or adding some extra functions in the appropriate region of space. Even when the CASSCF active space was chosen large enough to include all strong correlation, it may happen that the perturbational treatment fails due to the occurrence of so-called intruder states. Such states may arise when the energy gap between the active and inactive or between the virtual and active orbital space becomes small or even negative (the orbital energies of the active orbitals being the eigenvalues of the zeroth-order hamiltonian Ho, which may in practice have different forms [27]). As long as the intruder states are weakly interacting with the zero-order wavefunction, they may be effectively shifted away by applying a special levelshift technique [28, 29]. (if they are strongly interacting they should instead be included in the reference space). The actual size of the level shift can be obtained by performing a series of test calculations [30]. For calculations of electronic TM spectra, the systematic application of a level shift 0.3-0.35 a.u. has been found appropriate. Practice learns that with a level shift of
284 this size all weakly interacting intruder states are systematicaly removed, whereas the relative energies obtained with CASPT2 are not significantly affected. Furthermore, applying a level shift also greatly speeds up the convergence of the perturbation treatment. 3. LIGAND FIELD SPECTRA OF CO(II) AND CU(II) COORDINATED TO OXYGEN SIX-RINGS IN ZEOLITES Zeolites are inorganic crystalline materials characterized by a regular structure of channels and cages of molecular dimensions. Their structure is built from tetrahedral units (TO4), consisting of a central atom (T) - mostly Si(IV) - coordinated to four oxygen atoms. Other atoms, most often Al(III), can replace the central Si(IV). The tetrahedral units are linked by sharing all the oxygen atoms, leading to a wide variety of materials, differing in framework structure and chemical composition. [31] As each Al(III) incorporated in the silicate framework leads to one excess negative charge, an equivalent amount of extra-framework cations must be introduced to neutralize the structure. These cations, mostly Na(I), K(I), Ca(II), are present inside the cages and channels of the zeolite together with intrazeolitic water. They are not covalently bound to the zeolite framework and can therefore easily be replaced by transition metal ions via conventional aqueous ion exchange. After dehydration, these transition metal ions become localized. They are dispersed over the large internal surface of the zeolite, coordinating to the framework oxygen atoms. However, in the absence of adsorbed molecules, the resulting TM coordination environment is often not saturated. This makes the TM sites interesting sites for adsorption and opens the possibility for catalysis. The microporous character of the zeolite material puts limits to the size and shape of the interacting molecules, thus enabling shape selective reactions. A crucial step in the investigation of the catalytic potential of these TM centers is the study of the different coordination possibilities of the metal in the zeolite and its accompanying electronic structure. A powerful tool in this respect is provided by electronic spectroscopy, in particular diffuse reflectance spectroscopy (DRS) and electron spin resonance (ESR) spectroscopy. With a partially occupied 3d shell in an ionic coordination environment, the lowest excitations in the electronic spectra can be expected to originate from transitions within the 3d shell [32]. Both the ligand field excitation energies and the ESR spectra, as manifested by the g tensors and hyperfine splittings, are sensitive to the surroundings of the transition metal ion. As such, these spectra provide "fingerprints" of the specific TM environment(s) in the considered zeolite. However, the interpretation of these spectroscopic data is not always straightforward, and in many cases still a subject of debate. Until some years ago, the spectroscopic fingerprints were mainly interpreted from X-ray diffraction (XRD) results or by means of semi-empirical calculations (ligand field theory) [33-41]. However, a major limitation of XRD in the study of TM exchanged zeolites is that it can only provide average information, making no distinction between Si and Al in the lattice nor between occupied and empty sites. As a consequence, the apparent symmetry of the coordination environment of TM ions obtained with XRD is often too high. All semi-empirical calculations inevitably started from these average XRD
285
structures and were therefore performed in fictitious high symmetry coordination environments. The possibility of symmetry reduction, for example by an asymmetric aluminum surrounding, was not considered. As will be shown in what follows, this factor does indeed play a crucial role in the metal-zeolite interaction. During the last ten years many computational studies, predominantly using DFT, have been reported [42-64] in which the siting and catalytic potential of TM ions (most often Cu(I)) bound to zeolite surfaces was studied by means of cluster models of different sizes. Especially worth mentioning in this respect is the work of J. Sauer and coworkers, employing a combined quantum mechanics/interatomic potential function technique [53, 65, 66], in which the cation exchange site is treated by quantum mechanical methods, while the periodic lattice is treated by an ion pair shell model potential. Using this method, the location, structure and coordination of isolated Cu(I) and Cu(II) ions in ZSM-5 were studied. [57, 59, 62]. In a series of recent studies [67-72] we have investigated the structure and spectroscopic properties of the Co(II) and Cu(II) ions in different zeolite environment (e.g. zeolites A, Y, ZK4, ZSM-5 and mordenite). In these studies, DFT structure optimizations on cluster models were combined with CASPT2//CASSCF calculations of absorption spectra and ESR (/-factors. An important plus point of such combined studies [62] is that the plausibility of the calculated ground state structures can be directly tested by a confrontation with experimental (spectroscopic) data. In the next sections we will present the basic methodology and discuss the results obtained for the ligand field spectra of Co(II) and Cu(II) coordinated at trigonal six-membered ring sites in zeolites A and Y. For a further discussion of Cu(II) coordination in pentasil zeolites (mordenite, ZSM-5) and the results obtained for ESR p-factors we refer to the literature [67-72]. 3.1. Calculated electronic spectra for trigonal cluster models The framework structures of zeolites A and Y are schematically drawn in Fig. 1. These zeolites are built from a regular stacking of so-called sodalite cages, each consisting of 24 T (Si or Al) atoms that make up a truncated octahedron and 36 bridging oxygen atoms. In Fig. 1 the vertices give the position of the T atoms, while the lines connecting the vertices represent the oxygen atoms. The sodalite cages can be connected to each other in two different ways, either through the oxygen four-membered rings or through the six-membered rings. This gives rise to two different zeolite topologies, respectively denoted as LTA (Fig. l.A) and FAU (Fig. l.B). Zeolite A belongs to the LTA topology, whereas zeolite Y is a member of the FAU = faujasite topology. Both zeolites also differ in their Si/Al ratio: strictly one in zeolite A, but about two or higher in zeolite Y. The latter difference was shown to be at the basis of the different ESR signals observed for Cu(II) bound to both zeolites [68, 72]. Structures obtained from single-crystal XRD are available for Co(II) and Cu(II) exchanged zeolite A [73, 74]. Both ions are found on the threefold axes in the hexagonal six-membered oxygen rings, denoted as site II in Fig. l.A. The coordination environment of both ions in this site is shown in Fig. 2. Both TM ion are coordinated to three close-by oxygen atoms (O^), with a second shell of three oxygen atoms (OB) at a considerably larger distance. The site symmetry is Cy,v, but with the OA-M-OA angle very close to 120° in both cases (119° for
286
A
B
Fig. 1. Schematic framework structure of zeolite A (A) and zeolite Y (B)
M = Cu and 117.4° for M = Co), indicating an almost trigonal planar (D 3 t ) coordination. For Cu(II) and Co(II) exchanged X-ray [75-77] and neutron diffraction [78] studies indicate that also in this case the sing-membered rings (indicated as I and II in Fig. 1) are the favored sites for both ions, although other possibilities (e.g. the center of the hexagonal prism, I, or, in case of Cu(II), the four-membered ring sites III and III) have been proposed. Both for M = Co and M = Cu the experimental DRS spectra of the M(II)-A and M(II)-Y combinations show ligand field transitions at similar energies. The spectra of Co(II)-A and Co(II)-Y consist of three main features [67, 79]: a broad band with a maximum at 7 000 cm" 1 (band I), a second broad band at 15 000-19 000 cm" 1 (band II) and a weaker feature at 24 000-25 000 cm" 1 (band III). All three features are split further by 1 000-2 000 cm" 1 . For Cu(II), both spectra consist of a main broad band with a maximum at 10 400 cm" 1 and a shoulder at around 12 500 cm" 1 , and a second, weak, band appearing at 14 700-15 400 cm" 1 [41]. A first series of CASPT2//CASSCF calculations was performed on a MOeSi3Al3Hi2~ cluster, representing the oxygen six-ring site in zeolite A, and using the average structure obtained from XRD (Fig. 2). With three Al and Si put at alternating positions (according to Loewenstein's rule [80] two Al should always be separated by at least one Si) the site symmetry is reduced from C3,, to C3. For M = Co, CASSCF calculations were performed with an active space containing thirteen electrons in eight orbitals, namely Co 3p, 3d. The Co-0 bonds are very ionic [67], such that contributions to the wavefunctions from O 2p -4 Co 3d excitations are not important in this case [18]. Instead, when including O 2p orbitals into the active space these orbitals spontaneaously rotate into Co 3p. The latter reflects the importance of these semi-core orbitals in the correlation treatment. Even if this is a dynamic rather than a static correlation effect, slightly superior results for ligand field transitions may be obtained with 3p in the active space [25]. On the other hand, in order to take care of the double-shell effect and the increased covalent character of the M-0 bonds for M = Cu, the active space was in this case constructed from two Cu d shells (3d,3) as well as the bonding O 2y>-Cu 3d counterpart of the orbital
287
Al
Al
Fig. 2. M-OA and M-OB distances in the trigonal structures obtained from XRD on M(II)-zeolite A (M = Co, Cu)
which is singly occupied in the ground state (antibonding but predominantly Cu 3d [68, 72]). The effect of spin-orbit coupling on the ligand field transitions was taken into account by means of an effective one-electron operator [67, 68]. For Co(II), the effects of spin-orbit coupling are more limited. They are not included in the results reported here (but can be found in ref. [67]). For Cu(II) in a trigonal oxygen environment, the ligand field spectrum arises from a splitting of the sole d° term 2D into two 2E and one 2 A states, both 2E states being further split by spinorbit coupling. For Co(II), the states composing the ligand field spectrum correspond to two parent d7 states 4P and 4F, the latter being the ground state. The experimental 4 P- 4 F splitting in the free Co(II) ion is 14 561 cm" 1 [81]. In a C3 environment, these terms are split as follows: 4 F ->• 3 4A + 2 4E, 4P -)•4A + 4 E. In table 3.1 the results obtained from the CASPT2 calculations are compared to the experimental ligand field splittings in the zeolite environment. For the MOoSi3Al3Hi2~ cluster model at the crystal structure, the splitting of the free-ion states of both TM ions is obviously underestimated. For Cu(II) the predicted excitation energies do not reach higher than 8 500 cm" 1 , whereas the experimental spectra reveal the presence of ligand field states up to 15 000 cm" 1 . For Co(II), the calculated splitting of 3 500 cm" 1 for the 4 F state corresponds to only half of the experimental band maximum of band I. The calculated splitting of the 4P state, 7 000 cm" 1 , is closer to the experimental splitting of around 8 000 cm" 1 between bands II—III. The calculated excitation energy of the c 4E state, 15 719 cm" 1 , does correspond to the experimental band maximum of band II. However, the calculated result for the second state, 22 700 cm" 1 is too low when compared to band III. Such large errors are unlikely to be caused by deficiencies in the CASPT2 method. This strongly suggests that the origin must instead be found in structural deficiencies in the average XRD structures used in this first set of calculations. Obviously, the ligand field strength created by this fictitiously high symmetric coordination environment is much too low.
288 Table 1 CASPT2 spectral data for selected MO6Si(6-a.)Ala'Hi2~ clusters (M = Cu, Co) at crystal and DFT optimized structures exp. CuO6Si 3A13H12 cluster CuO6:3i3Al2Hi2 cluster B3LYP-DFT structure' spectrum'1 XRD structure" Energy (cm"1) State Energy (cm 1 )' ; State Energy (cm x )d 2 2 E A 0 0 2 1413 A 8411 2 2 E A 10 644 10 400 4 997 2 12 500 5 259 A 11303 2 2 A A 15 000 8 523 15 108 CoO6Si 3A13H12~ cluster CoO6:3i3Al2Hi2 cluster exp. B3LYP-DFT structure* spectrum^ XRD structure'' State Energy (cm : ) State Energy (cm 1) Energy (cm"1) 4 4 A 0 A 0 4 4 E A 562 63 4 4 E A 63 773 4 4 A 2 260 A 2 386 4 4 E 3 235 A 3 916 4 4 E A 3 235 4 576 4 4 A A 6 264 7 000 3 523 4 4 E A 16 000 15 519 15 886 4 4 E 15 519 A 16 410 4 4 25 000 A 22 793 A 25 163 "ref 74 ; see Figure 3.2; r;ref 41; ^including spin-orbit coupling; 'ref 73; 'ref 67
289 3.2. Structural deformations and concomitant ligand field strengthening In a second series of calculations, it was therefore decided to investigate possible local distortions of the zeolite surface produced by the coordination of the TM ions. This was done by DFT structure optimizations on clusters with the structural formula MO6Si6-a. Ala.(OH)i2(2~:i:)+ (x=0-3), containing the metal ion coordinated in the same sing six-ring site as above, but terminated by OH-groups instead of H. The starting geometry of the cluster models was based on the XRD data of M(II)-zeolite A [73, 74]. In order to allow full freedom of movement for the TM ions and the coordinating oxygens, these optimizations were performed without symmetry restrictions. However, in order to mimic the rigidity of the zeolite framework, restrictions were imposed on the orientation of the dangling OH bonds. Different aluminum contents and distributions were considered. Both the BP86 and B3LYP functionals were employed and produced similar results in case of Co(II) but not for of Cu(II), for which B3LYP proved to be superior [82]. All results included in this chapter were obtained using the B3LYP functional. In case of Cu(II), the accuracy of the obtained B3LYP-DFT structures was checked by performing additional restricted CASPT2 geometry optimizations for a few clusters. [68] Significant corrections were only obtained for the negatively charged CuO6Si3Al3(OH)12~ cluster model. Finally, the ligand field spectra corresponding to all optimized structures were calculated by means of the CASPT2 method. In order to reduce the computational effort, the terminal OH bonds were first replaced by hydrogens. Fig. 3 shows the results obtained for the structure of a representative cluster model containing two alumina. As compared to the crystal structures in Fig. 2, the largest changes are observed for the distances M-O^ between the central metal and the three distant oxygens. For M = Co two of these distances are shortened from 2.95 A to 2.67-2.42 A, while the third OB atom is driven away to 3.39 A. As such, the central Co(II) ion obtains a total coordination number of five instead of three, however with two bonds still signficantly weaker than the three original CO-OA bonds. On the other hand, for M = Cu two of the distant O B oxygens are pushed outwards, while the third O^ is moved towards the central Cu(II) to form a fourth strong bond with a Cu-O distance of 2.15 A (CASPT2 optimized; the corresponding B3LYP result is 2.21 A). This striving of Cu(II) to obtain a fourfold coordination environment when bound to a zeolite surface is a general given, observed in all our studies on Cu(II) bound to different zeolites, with and without external ligands [68-72]. It can be explained by considering the ground state d° electronic structure, with one singly occupied 3d orbital. Since this orbital has four, squarely oriented lobes, a maximum antibonding interaction with the ligand environment is obtained by a square-planar rather than by a trigonal oxygen environment. As the corresponding bonding orbital is doubly occupied, maximizing the antibonding interaction with the ligands leads to a stabilization. This explanation is in line with simple ligand field predictions. Co(II), with three singly occupied 3d orbitals, does not display this strong directional preference for one additional strong bond, but rather shows a more general tendency to increase its coordination number to four or five [67, 82]. However, it is important to note that in both cases (and all other cases included in our studies) the additional M-0 bonds are formed with O# oxygens that are bound to Al rather than Si. This indicates how the aluminum distribution in zeolites may
290 S Si iA
•• gf
29
an
AS>
^J9SI
Ji2-15
3.39
SSii
05 A 2 9M 9 2 67 - flT
A
0SlSi 12.42 02.4
3.39
Fig. 3. M-O^ and M-O# distances in the asymmetric structures obtained from B3LYP-DFT optimization of MO6Si4Al2(OH)12 (M = Co, Cu)
thoroughly affect the bonding properties and corresponding spectroscopic (electronic and ESR) properties of both ions in different zeolites. In particular, our calculations have indicated how different distributions of aluminums may actually lead to two distinctly different ESR signals in the zeolites Y, ZK4 and mordenite [68, 69, 72]. As for the ligand field spectra of Cu(II) and Co(II) coordinated to zeolites A and Y, similar excitation energies were obtained from CASPT2 calculations on optimized models with different numbers and distributions of aluminums. The results obtained for the model with two aluminums distributed as in Fig. 3 are included in Table 3.1. These data clearly reveal a much stronger ligand field surrounding of both ions as compared to the trigonal oxygen environment in the models with structures frozen at the XRD values. The splitting of the 2D ground state of Cu(II) is doubled, and the calculated transition energies now nicely correspond to the position of the two band maxima at 10 400 cm" 1 and 15 000 cm" 1 in the experimental spectra. For Co(II), the splitting of the 4F ground state is again almost doubled, but still remains slightly below the maximum of band I in the experimental spectrum. It is shown in ref [67] that the introduction of spin-orbit coupling leads to a further broadening by 800 cm" 1 of the transition energies belonging to 4F. Band II at 16 000 cm" 1 is now found to correspond to two distinct transitions, split by about 600 cm" 1 . This is in accord with the experimental splitting observed for this band, although the calculated splitting is still too small. Finally, the appearance of band III at 25 000 cm" 1 is well reproduced by the CASPT2 results. The experimental splitting of this band may at least be partly explained by spin-orbit coupling [67]. On the whole, the results presented here have indicated how the coordination of TM ions in zeolites may be succesfully explained and even predicted by means of cluster models and using a combined CASPT2//D FT approach. The CASPT2 excitation energies are especially useful since, in confrontation with experimental ligand field spectra, they provide a sensitive probe of the geometric distortions predicted by the DFT optimizations. Such distortions are also indicated by other spectroscopic experiments, e.g. IR, ESEEM and 27A1 NMR [83-85], but their actual extent has so far been impossible to obtain directly from experiment.
291
4. CHARGE-TRANSFER STATES: ELECTRONIC SPECTRUM OF THE PERMANGANATE ION The electronic spectrum of permanganate presents one of the theoretically most intensively studied spectra of transition metal systems. Based on a semi-empirical MO treatment, a basic interpretation of the electronic transitions was provided already in 1952 by Wolfsberg and Helmholtz [86]. During the seventies, numerous ab initio studies were reported, using either Hartree-Fock and limited configuration-interaction [87-91] or one of the earlier Hartree-FockSlater versions of DFT: HFS-SW [92], HFS-DVM [93]. More recently, MnOJ has been given the status of a prototype transition metal system, used as a benchmark for testing the performance of several newly developed methods for the calculation of excited states: DFT in its ASCF [94, 95] and time-dependent [96, 97] form and cluster expansion methods such as SACCI [98] and EOM- and extended EOM-CC [99]. This prototype status is however somewhat undeserved, given that when it comes to obtaining accurate electronic excitation energies for transition metal complexes, permanganate is an exceptionally hard rather than a representative case. The problem is related to the high (+7) formal charge on manganese in this molecule, giving rise to substantial static correlation effects involving excitations from all twelve doubly occupied orbitals originating from 0(2-) 2p, either bonding or nonbonding, into the five antibonding orbitals corresponding to Mn(7+) 3d, which are empty in the ground state [18] (see also section 2) . Moreover, these correlation effects change appreciably in the excited states [100]. Consequently, any single reference molecular orbital description of the permanganate ground state is quite dubious, and its shortcomings doomed to be reflected in the calculation of excitation energies using methods working with these ground state orbitals. The presence of important static correlation effects in permanganate also affects the DFT description of the excitation energies, with values that substantially differ between the ASCF and TD-DFT approaches, that fluctuate by up to 3 000 cm" 1 between different functionals, and that are on the whole considerably less satisfactory than the results obtained with the same methods for other transition metal systems [96, 97]. Another consequence if this general malaise is that the assignments of the permanganate spectrum have undergone a lot of changes throughout the years, and in fact a general consensus still has not been reached. Here, we will only discuss the most recent theoretical results, obtained with either TD-DFT or cluster expansion methods (an overview of the older results can be found in ref. [98]). Furthermore, we also present the results obtained from a CASPT2 treatment [101], based on CASSCF reference wavefunctions built from an active space comprising all seventeen valence orbitals originating from manganese 3d and oxygen 2p: 6a i, (l-2)e, lti, (5-7)^. CASSCF calculations with such a large active space have only recently become within the limits of computer power, but are still very time consuming. We report here the results obtained for the excitation energies of the lowest spin- and dipole-allowed 1Ai—>XT2 transitions in this tetrahedral molecule. A more detailed description of the spectrum will be presented in a separate paper [101]. In Table 4 calculated excitation energies of the four lowest XT2 states with different methods
Is) Is)
Table 2 Calculated electronic allowed transition energies (in cm" 1 ) of MnOj state Excitation energy (cm x)
Composition(CASSCF)(%) l*i
lr
a T2 bxT2
a
ASCF" LDA 21880 34 070 32 400 45 950
6
TDDFT BP/ALDA 22 825 31536 38 310 47 182
SAC-Cr E 20 728 30 003 28 874 46 940
ref 95; b ref 96-values differing by up
I CASPT2 18 066 29 035 29 600 46 779
18 389 27 906 31455 43 433
C
I 2e 48 3
6*2
l*i
6ai
6*2
2e 3 30 9
7*2
7*2
7*2
1 8 23
3 4
5 5
Assignment
l*i->7* 2 6*2^7*2
Experimental band position^ (cm"1) 18 300 28 000 32 200 43 956
293
are compared to the experimental band positions, shown in the rightmost column. Based on symmetry considerations it can easily be shown that the O 2p23—Mn 3d1 manifold can give rise to five lT2 states corresponding to one of the following excitations, or to a mixture of them: lti—>2e, 6£2—>2e, lii—>7t2, 6ai—>7t2, 6t2—>7£2- The calculated results in Table 4 have been ordered such that their principal character corresponds to the assignment in the second column from the right. One should however keep in mind that, with exception of the ASCF method, all methods describe the different excited states as a mixture of different orbital replacements. For example, with TDDFT the state b : T 2 which is assigned as 6£2—>2e in table 4 in fact consists of 63 % 6t2—>2e and 36 % l*i —>7t2 character, whereas the c *T2 state, assigned as lii—>7i2, is composed of 50 % lii—>7t2, 17 % 6t2—>2e and 20 % 6t2^-7t2 character [96]. Table 4 also shows the contribution of the five singly excited configurations to the CASSCF reference wavefunctions correponding to the calculated CASPT2 excitation energies. For one thing, these numbers reflect the extreme importance of nondynamic correlation and corresponding multiconfigurational character of the different state functions in permanganate, with a total contribution of singly excited character amounting to 51 % for the first a x T 2 state and decreasing further to only 9 % for the d XT2 state. The contribution of the HF configuration in the corresponding CASSCF ground state wavefunction is 58 %. All methods collected in Table 4 agree upon the assignment of the first band in the experimental spectrum to an excitation with predominantly lti—>2e character. Furthermore, as far as concerns single orbital replacements, the CASPT2/CASSCF results also agree with the other methods [96, 98, 99] that the fourth band in fact corresponds to an (almost equal) mix of the 6t2—>7t2 and (M\—>7t2 orbital transitions. However, the assignments of the second and third bands have so far remained controversial. The SAC-CI and ASCF methods assign the second band to \t\-^t7t2 and the third band to 6t2^2e. This also corresponds to the original assignment by Ballhausen and Gray [103], based on indirect evidence from various sources. On the other hand, according to the TDDFT, EOM-CC and also to the present CASPT2 results these assignments should be reversed. In view of the fact that (a) CA SPT2//CA SSC F is the only method that properly accounts for the strong multiconfigurational character of this molecule, and (b) the close correspondence (to within 1000 cm" 1 ) between the CASPT2 excitation energies and the experimental band maxima of all four bands, we hold it most likely that the assignment based on the CASPT2//CASSCF method is correct, and that the discussion on the assignment of the second and third bands in the permanganate spectrum can now finally be closed.
5. SPECTROSCOPY OF REDOX PROTEINS TM atoms or ions coordinated in proteins are usually found in ligand field environments consisting of very specific, sometimes remarkably complex ligands. The role of this environment is of course to provide the metal with the appropriate electronic structure to play its part in the catalytic cycle. In the case of redox active metalloenzymes an important task for the ligand environment is to modulate the metal reduction potential, the latter determining the driving force of the electron transfer reaction. The reduction potential is dependent on the ionization energy
294
of the reduced site as well as on the so-called reorganization energy, i.e the energy required to reorganize the ligand environment (inner-sphere reorganization energy) and the solvent and the rest of the protein (outer-sphere reorganization energy). Obviously the number, position and specific character of the ligands directly bound to the metal plays a crucial part in all this. Thus, strongly covalent interactions between the metal and one or more ligands will facilitate oxidation by lowering the effective charge on the metal. Moreover, the more strongly antibonding the redox active orbital the higher its energy and the easier ionization from this orbital. Since the same factors also determine the spectroscopic characteristics of the metal-ligand combination, electronic spectroscopy provides an excellent means of obtaining information concerning the catalytic potential of the metal active site in redox proteins. In the remaining of this section we will discuss the electronic structure of the active site in two important groups of redox active metalloenzymes, based on the computation and interpretation of electronic absorption spectra of model complexes. 5.1. Relation between the structure and spectroscopy of blue copper proteins Blue copper proteins are electron transfer proteins. The active site of all these proteins contains a single copper ion (with oxidation state varying between I and II), four- or five-coordinated with a typical short bond to the thiolate S of a cysteine residue. Two of the other ligands are N s atoms of histidine residues, and the fourth is normally a methionine thioether group. (Azurins have a fifth, weakly bound carbonyl oxygen ligand). A typical example is plastocyanin, a small protein involved in photosynthesis. In the oxidized form, the Cu(II) ligand surrounding in this protein is trigonal, the two histidine-N and cysteine-S forming a (distorted) trigonal plane, with the methionine S bound in an axial position at a large distance. Such a trigonal surrounding is quite peculiar for Cu(II): four-coordinated inorganic complexes of this ion are almost invariably (more or less distorted) square planar [32] (e.g. the square planar Cu(II) coordination in zeolites, discussed in section 3.2) . Instead, the trigonal Cu(II) surrounding in plastocyanin is closer to the (distorted) tetrahedral surrounding found for the reduced Cu(I) form of the same protein [104]. The close resemblance between the two geometries accounts for a small reorganization energy accompanying reduction, and hence to a high rate of electron transfer in this class of proteins [105]. This fact originally led to the suggestion that the unusual trigonal surrounding of the oxidized site is not the one preferred by Cu(II) itself, but is instead imposed upon it by the tertiary protein structure, in order to facilitate electron-transfer, the induced-rack [106, 107] andentatic state [108, 109] theories. However, DFT structure calculations on different realistic models of the Cu(II) surrounding but lacking the rest of the protein (e.g. Cu(imidazole)2(SCH3)(S(CH3)2)+) [9, 110, 111] have indicated instead that a trigonal structure is indeed what is preferred by the specific choice of Cu(II) ligands in plastocyanin. The ligand which is crucial in this respect is the Scv« thiolate group [112]. By forming a 7r-bond between the two lobes of its 3pw orbital and two of the lobes of the Cu 3d orbital in the trigonal plane, the S cy« ligand in fact plays the role of two ligands in an "apparent" square-planar coordination [113]. Electronic structure calculations further indicated that the Cu-Scy., yr-bond is highly covalent. The antibonding Cu 3d-S 3pw combination
295 involved is the orbital which is singly occupied in the Cu(II) ground state. Therefore, this is also the redox active orbital and its high covalency activates the blue copper site for its biological function as electron transfer site [9]. The highly covalent Cu-S^j,., 7r-bond in plastocyanin is also responsible for the appearance in the electronic spectrum of the "blue" band, i.e. an intense (e,,,.aa:=3000-6000 M~1cm~1) absorption band around 600 nm. This has been shown in several theoretical studies of the electronic spectrum of plastocyanin, where the blue band was found to originate from a charge-transfer excitation from the bonding to the antibonding Cu-thiolate IT orbital, gaining its high intensity from the large overlap between the two orbitals involved [114117]. 5.7.7. The spectrum of plastocyanin The electronic spectra of several blue copper proteins have been recorded by visual and nearinfrared absorption spectroscopy, circular dichroism and magnetic circular dichroism [9,115, 118-120]. For plastocyanin in total nine different absorption bands have been reported [115]. They all correspond to excitations of an electron into the Cu 3d-S 3pn antibonding combination. Four of these are LF transitions, i.e. where the electron originates from one of the other molecular orbitals with (predominantly) Cu 3d character. The other five are LMCT excitations, with the excited electron coming from a molecular orbital which is mainly centered on one of the ligands. The spectrum of plastocyanin was calculated by performing CASPT2/CASSCF calculations on different model systems, ranging from [Cu(NH3)2(SH)(SH2)]+ to [Cu(imidazole)2 (SC2H5)(S(CH3)2)]+ [117]. In order to include the important 3d radial correlation effects in the Cu2+ ion the active space of these calculation necessarily had to include a second d-shell (i.e. the double-shell effect; see also section 2). Since the model complexes which are closest to the real protein necessarily are unsymmetric, the size of the active space that could be used was limited to twelve orbitals, i.e two Cu d-shells, and the Scys 'ipa and 3pn lone pair orbitals. This active space allowed for the description of the lowest energy part of the spectrum, up to 20 000 cm"1, containing the ligand field states and the two most intense charge-transfer states, both originating from Scy!i- Two additional charge-transfer transitions at higher energy could be calculated for model systems that were forced into Cs symmetry, and were shown to originate from either imidazole or SMC*- For a more detailed discussion of these transitions and the corresponding experimental band positions we refer to ref. [117, 121]. The most accurate CASPT2 results for the spectrum of plastocyanin were obtained for the [Cu(imidazole)2(SH)(SH2)]+ model by making use of the plastocyanin crystal structure [122] but with CASPT2 optimized Cu-S^s a n c ' CU-SMC* distances, and after including the effect of the protein environment by means of a point-charge model [123]. These results are shown in Table 3. The ground state and the different excited states may most easily be characterized by analyzing the singly occupied natural orbital in each state. These orbitals are shown in Fig. 4. By choosing a coordinate system with the z axis along the Cu-S Met bond and the Cu-Sc,ys bond located in the xz plane, the Cu 3d orbitals involved can (approximately) be labeled as shown in the left column of Table 3.
Is)
Table 3 Comparison between the calculated and experimental spectra of plastocyanin and nitrite reductase " State Character Plastocyanin Nitrite reductase exp. exp. calc. calc. bA LF (Cu-S)o-*^(Cu-S)7r* 4 363(.0001) 4 408(.0000) 5 000(.0000) 5 600(.0000) 11 645(.0000) 10 800(.0031) 11 900(.0026) 12 329(.0003) cA LF Cu3dz2 —>(CU-S)TT* dA LF Cu3dyz -^(CU-S)TT* 12 981(.0025) 12 800(.0114) 13 500(.0086) 12 872(.0004) 12 671 (.0002) 13 95O(.OO43) 14 900(.0101) 13 873(.0028) eA LF Cu3dxz -KCU-S)TT* fA CT ( C U - S ) T T ^ ( C U - S K 15 654(.1162) 16 700(.0496) 17 550(.0198) 15 789(.0325) CT (CU-S)CT^(CU-S)TT* 21 974(.0012) 21 39O(.OO35) 21 900(.0299) 22 461(.1192) gA
Character LF (CU-S)TT*->(CU-S)CT *(+7T*)
LF LF LF CT(CU-S)TT^(CU-S)CT* (+7T*) CT(CU-S)CT->(CU-S)CT* (+7T*)
"Calculations were performed on a [Cu(imidazole)2(SCH3)(S(CH3)2)]+ model at the crystal geometry but with the Cu-Scya optimized at the CASPT2 level
and Cu-Sj^ e t distances
297
^
XA
bA
eA H Fig. 4. Plot of the singly occupied (natural) orbital in the ground state and the lowest excited states in [Cu(imidazole)2(SH)(SH2)]+ at the plastocyanin structure
As one can see, the ground state singly occupied orbital is indeed strongly delocalized between Cu 3dxv and Scy., 3p^, thus confirming the highly covalent character of this interaction. The four lowest excited states in the spectrum may formally be characterized as ligand field states. However, from Fig. 4 it is clear that all four in fact also contain a significant amount of charge-transfer character. Even if the departing orbitals in the second to fourth transition are almost pure Cu 3d, the accepting orbital is partly Scys 3pn such that these transitions involve a significant movement of charge from the copper into Scy.,. More importantly, the lowest excited state has a single electron in an orbital which is again strongly delocalized over the CuS<7y., bond, only now involving Cu 3dx2_y2 and Scya 3pCT. According to these CASPT2 results, the first band at 5 000 cm" 1 in the spectrum of plastocyanin therefore involves the excitation of an electron from the Cu-Sc,,., c* to IT* orbital. It should be noted that the CASPT2 assignment is different from the one reported by Solomon and coworkers [9,115], who instead assign the band at 5 000 cm" 1 as the excitation out of the Cu 3dz2 orbital and the second band at 10 800 cm" 1 as the excitation out of Cu 3d,l;2_y2. Fig. 4 also shows that the singly occupied orbitals in the highest two excited states included in the CASPT2 calculations are the bonding counterparts of the corresponding orbitals of the ground state and the first excited state respectively. As such, the first charge-transfer state at 16 600 cm" 1 in the spectrum of plastocyanin corresponds to a vr-to-Tr* Cu 3d-Scy« 3p^ transition. The large overlap between the two combinations involved explains the extremely high intensity of this transition. At higher energy a second charge-transfer band, corresponding
298
to a a-to-TT* transition, is predicted to appear with a considerably lower intensity, as is confirmed by the experimental spectrum. The CASPT2 results presented in Table 3 undoubtedly represent the most accurate set of computational results and analysis of the plastocyanin spectrum available today. Older theoretical studies, performed using the HFS-SW (Xa scattered-wave) DFT method [114, 115] and the semi-empirical CNDO/S method [116], could provide a qualitative but no quantitative explanation of the experimental spectrum. A DFT study of the spectrum was recently reported [9], containing results obtained with different approaches (ASCF, TD-DFT), functionals (GGA and hybrid) and basis sets (STO and GTO). Calculations were performed on the [Cu(imidazole)2(SCH3)(S(CH3)2)]+ model, fixed at the crystal structure of oxidized plastocyanin [122]. As it turns out, TDDFT completely fails in describing the electronically excited states of this model: all ligand field transitions are calculated much too high in energy, whereas the charge-transfer states are calculated much too low and in the wrong order. E.g. the lowestenergy band in the plastocyanin spectrum is predicted by TDDFT to be a SM<-J.—>Cu3d excitation ! According to the authors [9], the failure of DFT to describe the plastocyanin spectrum should be brought back to an overly covalent bonding description of the ground state, with a singly occupied HOMO containing too much S cys and too little Cu 3d character. A reasonable ASCF (BP86) description is obtained after introducing an adjusted effective nuclear charge on the central copper ion. 5.1.2. From blue to green Cu proteins The appearance of two S>cya—>Cu3d charge-transfer bands is characteristic of the electronic spectrum of all blue copper proteins, providing them with the label type 1 copper proteins (as opposed to type 2 copper proteins, the spectrum of which only contains a number of weak ligand field transitions in the same region). The band positions of the two transitions, around 600 nm (16 600 cm"1) and 450 nm (22 000 cm" 1 ), remain approximately the same for all proteins. However, the relative intensity of the two bands varies between different proteins [124, 125]. Thus axial type 1 proteins, like plastocyanin and azurin, show only little absorption in the 450 nm region, while the 450 nm band becomes much more prominent in rhombic type 1 proteins like pseudoazurin and cucumber basic protein (the classification of blue Cu proteins as axial or rhombic is based on their ESR characteristics). The increasing intensity of the 450 nm band in the latter proteins goes together with a decrease in intensity of the 600 nm band, so the sum of e^omn and eeoonm remains approximately constant [124]. A limiting case is nitrite reductase from Achromobacter cycloclastes. The intensity of the 600 nm absorption peak is in this enzyme reduced by a factor 3 compared to the classic proteins, and the 460 nm absorption has actually become the more intense, resulting in a green color of the enzyme. The gradual change in the spectral characteristics of different blue copper proteins goes together with a structure which is gradually more distorted with respect to the trigonal Cu(II) surrounding in plastocyanin. The distortion comes down to a flattening, and can most conveniently be quantified by considering the angle cj> between the planes formed by the N ffiil-CuNiH,, bonds and the S ^ - C U - S M C * bonds respectively [123]. In a strictly trigonal structure,
299 both planes are perpendicular ((/> = 90°). In practice, axial type 1 proteins have angles ranging between 77° and 89° (e.g. for the poplar plastocyanin described in the previous section >=82°). On the other hand, in the rhombic type 1 proteins significantly smaller angles are found, ranging between 70° and 75°. Again, nitrite reductase is a limiting case: different crystal structures of this enzyme have cf> angles ranging between 56° and 65°. Together with the flattening of the structure subtle variations of two important bond distances are observed: the Cu-S Met bond is considerably shortened in nitrite reductase as compared to plastocyanin, from 2.88 A to 2.56 A, whereas the Cu-Scj,., bond distance is increased, from 2.11 A to 2.17 A. In order to (qualitatively) investigate the relation between the structural variations between different blue copper proteins on the one hand and the variations in the relative intensity of the 460 nm and 600 nm bands on the other hand, test calculations were first performed on a small model [Cu(NH3)2(SH)(SH2)]+. A series of (B3LYP-DFT + restricted CASPT2) geometry optimizations was performed on this model at fixed <j> angles, ranging between = 90° (i.e. a trigonal structure) and <j> = 0° (i.e. a square-planar structure). A CASPT2//CASSCF calculation of the electronic spectrum was performed at different points along the twisting path. From these small model calculations, several important points were noted, all of them highly relevant for the structure-spectroscopy relationship in the actual proteins. First of all, these calculations indicated that the energy curve along the <j> twisting angle is extremely flat. At the CASPT2 level two minima were found, one for =90 ° (i.e. close to the plastocyanin ) and one for — 50° (i.e. close to the nitrite reductase cj>). Yet, the barrier between these two minima is low: only 8 kJ/mol. This explains how different protein environments are in fact able to stabilize a continuum of Cu(II) surroundings [9], ranging from the trigonal structure in plastocyanin to a distorted tetragonal structure in nitrite reductase. A second point emerging from these calculations (see also Fig. 5) is that the optimized Cu-SH and CU-SH2 distances indeed show the trends also observed for the actual proteins: the Cu-SH2 distance is gradually decreased by more than 0.3 A as 0 is reduced from 90° to 50°, while the Cu-SH bond is simultaneously elongated by about 0.1 A. As for the electronic spectra obtained for these small models, the relative intensity of the two HS—>Cu transitions, also indicated in Fig. 5, indeed follows the trend observed for the proteins. This trend can now also be related to the electronic structure of the ground state at different 4> angles, as reflected by the nature of the singly occupied orbital. In the trigonal ( = 0°) structure this orbital displays a TT antibonding interaction between copper 3d and the SH ~ S 3p orbital, similar to plastocyanin (see also the previous section and Fig. 4). As such, an intense TT-to-Tr* transition is calculated for this structure, whereas at higher energy the second, a-to-vr* transition, is calculated with little intensity. However, as angles between 90° and 0° an intermediate electronic structure is obtained, with the ground-
300 25
I
210
60 Twisting angle ( )
80
Fig. 5. Cu-SH (squares) and CU-SH2 (diamonds) distances, and calculated quotient of the oscillator strengths of the transitions around 460 nm and 600 nm (bullets), as a function of the cj> angle for the [Cu(NH3)2(SH)(SH2)]+ model.
state singly occupied orbital gaining more and more Cu-SH a character (at the expense of TT) as the structure becomes more flattened. As a consequence of this changing ground state electronic structure, the second charge-transfer transition in the electronic spectrum, originating from the Cu-SH a bonding orbital, gains intensity at the expense of the first transition, originating from Cu-SH IT. The "crossover", i.e. the point where the second transition becomes more intense than the first, is for [Cu(NH3)2(SH)(SH2)]+ predicted at cf> around 74°. As such, even these small model calculations predict nitrite reductase to be green. Armed with the above qualitative considerations, the analysis and assignment of the lowenergy part of the experimental spectrum of nitrite reductase becomes straightforward. CASPT2/CASSCF calculations of this spectrum were performed in a similar way as for plastocyanin, i.e. using the [Cu(imidazole)2(SH)(SH2)]+ at the nitrite reductase crystal structure of Achromobacter cycloclastes [126] but with CASPT2 optimized Cu-S>cys and Cu-S^d distances, and a charge model for the protein environment [123]. The results are compared to the experimental spectrum [127, 9] and the corresponding data for plastocyanin in Table 3. The singly occupied natural orbitals characterizing the ground and each of the excited states in nitrite reductase are shown in Fig. 6. As one can see, the ground state singly occupied orbital in the nitrite reductase model has indeed become more of a* than of TT* type, and vice versa for the corresponding orbital of the first excited state. Interestingly, the corresponding bonding orbitals, depopulated in the two charge-transfer states, do retain their pure a or TT character. As such, the intense bands
301
Fig. 6. Plot of the singly occupied (natural) orbital in the ground state and the lowest excited states in [Cu(imidazole)2(SH)(SH2)]+ at the nitrite reductase structure
in the nitrite reductase spectrum should be assigned as Scys—>Cu 3d 7r-to-(<7* + TT*) (17 550 cm"1) and a-to-(a* + TT*) (21 900 cm"1), respectively. As for plastocyanin, four ligand field states appear at lower energies. The experimental ligand field transition energies are significantly increased in nitrite reductase as compared to plastocyanin, indicating a strengthening of the ligand field as the Cu(II) surrounding becomes more tetragonally distorted [9]. This trend is also confirmed by the theoretical calculations [123]. Finally, the observed interchange of a- and vr-character between the ground and the first excited state with a decreasing twisting angle indicates that the flat ground state energy profile along this angle is in fact the result of a (strongly) forbidden transition between these two states. This also provides further credit to the fact that the lowest-energy band in both spectra should indeed correspond to the exchange of an electron between a* and TT* within the Cu-S^., bond, as opposed to Solomon's assignment of this band as a ligand field transition originating from Cu 3dzz. In summary, the model calculations presented in this section have clearly demonstrated how subtle changes in the structures of these blue copper proteins have a profound effect on the electronic structure of the Cu(II) environment and as such also on their spectroscopic characteristics. We believe that this case presents a nice example of the strength of accurate theoretical methods for both ground and excited states in explaining and predicting the electronic structure properties of biologically important molecules.
302 O
H
2
N
,,
N
SH
N
O
4
3
Fig. 7. Structure of the pyranopterin derived from protein crystallographic studies (shown in protonated form).
5.2. Electronic structure of oxomolybdenum enzymes modelled by the (Tp)MoO(bdt) complex Apart from the nitrogenases, all molybdoenzymes are oxotransferase enzymes, i.e. they catalyze the net transfer of an oxygen atom to or from a substrate: X + H2O « X O + 2H+ + 2e" e.g. sulfite to sulfate, an aldehyde to the corresponding carboxylic acid, or nitrite to nitrate. During enzymatic turnover, the molybdenym center is proposed to shuttle through Mo(VI)/ Mo(V)/ Mo(IV) oxidation states. All these enzymes, further subdivided into three families based on structure and reactivity (xanthine oxidase, DMSO reductase, and sulfite oxidase), contain a mononuclear molybdenum center ligated by one or two molecules of a special pyranopterin, originally termed "molybdopterin" (MPT: see Fig. 7). MPT consists of a pterin core that is coordinated to the metal atom via the dithiolene group on the four-carbon side chain. The mere fact that MPT invariably is a part of the active site strongly suggest that this ligand plays an active role in the catalytic process, by acting as an electron transfer conduit from the metal to other prosthetic groups [128], and by modulating the Mo center reduction potential [128-131]. At least one terminal oxo ligand is associated with the Mo active center during catalysis. The application of electronic absorption spectroscopy to study the electronic structure of the molybdenum center in these enzymes is hampered by the presence of additional prosthetic groups (e.g. hemes, iron sulfur centers and flavins) whose intense electronic absorptions obscure any absorbance that might arise from the oxomolybdenum center of the pterin-containing molybdenum cofactor. On the other hand, systematic spectroscopic investigations and analysis of well-characterized model complexes containing the molybdenyl ([MoO]3+) fragment have since long [132, 133] been used to probe the electronic structue and molybdenum-ligand bonding in the considered enzymes. An important class of such model compounds are of the type (Tp*)MoO(S-S), where Tp* = hydrotris(3,5-dimethyl-l-pyrazolyl)borate and (S-S) is a dithiolate ligand which forms a five-membered ring with Mo(V). Experimental absorption and CD spectra of (Tp*)MoO2+ centers with various S-S ligands (e.g. bdt = 1,2-benzenedithiolate, tdt = 2,4-toluenedithiolate, edt = 1,2-ethanedithiolate, qdt = quinoxaline-2,3-dithiolate, bdtCl 2 = 3,6-dichloro-l,2-benzenedithiolate) have been reported and assigned [128, 134, 135]. In a forthcoming publication we will present the results of a comparative computational study of these spectra [136]. Here we report a representative case, namely (Tp)MoO(bdt) with bdt =
303
Fig. 8. B3LYP-DFT optimized structure of (Tp)MoO(bdt)
1,2-benzenedithiolate, whereas in Tp the 3,5 dimethyl groups of Tp* are replaced by hydrogen atoms to simplify the computations. The structure of this molecule was first optimized with B3LYP-DFT. These calculations used 6-31G(d) basis sets for all atoms except molybdenum, the core region (up to 3d) of which was described by a relativistic ECP [137], whereas for the valence region a triple-zeta basis set with one f-type polarization function was used. The optimized structure is shown in Fig. 8. For the calculation and description of the electronic spectrum the complex was oriented such that the axial M=O bond lies along the z-axis, whereas the mirror plane perpendicular to the ene-dithiolate plane corresponds to the xz plane. The ligand field splitting of the d-orbital manifold in (Tp*)MoO(S-S) is dominated by the presence of a short, strongly covalent axial M=O bond. In particular, a interaction between the Mo dz2 and O pz orbitals and IT interaction between Mo dXZ4)z and O px/!J orbitals give rise to bonding-antibonding pairs of molecular orbitals, thus resulting in a strong destabilization of the antibonding, primarily metal based combinations. In the equatorial plane the Mo dxy orbital is involved in and destabilized by a interactions with two N and two S-donor ligands. This then leaves the dx2-V2 orbital as the Mo d orbital which is least destabilized by the ligand field, and which therefore becomes singly occupied in the Mo(V) ground state. This is also the redox active orbital, changing its occupation number between zero and two during the catalytic Mo(IV)-Mo(VI) cycle. Next to the ligand field transitions, the electronic absorption spectrum of Tp*MoO(S-S) complexes is characterized by a number of low-lying LMCT transitions originating from the highest four filled dithiolate orbitals, primarily sulfur in character, and schematically presented in Fig. 9. Two of them (denoted as a'(S,;?,) and a"(S,j,)) are oriented within the dithiolate plane, while the other two (denoted as a (SOJ)) and a (Sop)) are orthogonal to this plane. The a "(S,;2,) orbital is involved in covalent a bonding with the Mo 4dxy orbital. In order to describe the entire low-energy part of the spectrum by means of a series of CASPT2/CASSCF calculations, the active space was chosen to include the five orbitals with
304
3 \&op) Fig. 9. Schematic plot of the highest occupied molecular orbitals of S-S = ene-l,2-dithiolate. The small amplitudes of the wavefunction on the carbon atoms are not shown
predominant Mo Ad character and the four dithiolate orbitals a (Sip), a (S,;,,), a (Sap), a (SO!,). Furthermore, important correlation effects connected to the strongly covalent Mo=O bond were treated at the CASSCF level by including also the relevant Mo 4d—O 2p bonding molecular orbitals. This sums up to a total of twelve active orbitals containing 15 electrons. The CASPT2//CASSCF calculations were performed using ANO type basis sets [136]. Oscillator strengths were obtained using the RASSI method [138]. The results obtained for the spectrum of (Tp)MoO(bdt) are shown in Table 4. A plot of all orbitals involved in low-lying excitations is provided in Fig. 10. Three different groups of orbital excitations may be distinguished in the table. First, there are the LF excitations of the single electron from Mo dx2_y2 into the orbitals with predominant Mo dxz, dyz and dxy character. It is noticeable that dz2 is missing from this list: this orbital is destabilized so strongly by a interaction with the axial oxygen that any excitation into it falls outside the range of the calculated spectrum, i.e. up to about 35 000 cm" 1 . A second group are the LMCT excitations out of the Sop and S,p orbitals into Mo dT2_y2. Such excitations give rise to four excited states, which can be clearly recognized in Table 4 as the states b 2 A', a 2A", e 2A" and f 2A . The third group consists of LMCT excitations out of the two SO2, orbitals into Mo dxz, dyz and dxy. Excitations out of Srp into the same orbitals occur at higher energies, outside the range of the calculated spectrum. Each of the excitations in the third group can occur with two different spin couplings (with the unpaired electrons on the metal either singlet or triplet coupled), such that the total number of configuration state functions arising from excitations within this group amounts to 12. Together with the first two groups this brings the total number of excited states included in Table 4 to 19. As one can see, most of the calculated excited states in fact correspond to a mixture of different orbital excitations. A detailed experimental study of the electronic absorption and MCD spectra of (Tp*)MoO(bdt) was presented in 1999 by Inscore et al. [128]. The low-energy region of the spectrum consists of three distinct features at 9 100 cm" 1 , 13 100 cm" 1 and 19 400 cm" 1 . This band pattern seems to be characteristic for all (Tp*)MoO(S-S) complexes reported to date, which possess five membered chelate ring formed between the ene 1,2 dithiolate and molybdenum atom [128, 130, 134, 135]. The bands at 9 100 cm" 1 and 13 100 cm" 1 were assigned as LMCT transitions from the Sop orbitals into the singly occupied Mo 4dx2_y2 orbital. This assignment is confirmed by our calculations, which now also further establish the energy ordering of the
305 Table 4 Comparison between the calculated spectrum of (Tp)MoO(bdt) and the experimental spectrum of (Tp*)MoO(bdt)"
b 2 A' a 2 A" b 2A" c 2 A' c 2A" d 2 A' d 2 A" e 2 A" f 2A" e 2 A' g 2A"
calculated results exc.energy* main character" (osc.str.) 10 912(.O168) a (S OJ) )^a ( d ^ . , / ) 85 % 14 326(.0147) a"(Sop)->a'(clr2_!/2) 88 % 16 457(.0018) a'(d^- y O^a"(d, y ,) 45% ; a'(S OJ) )^a"(d y J 38 % 18 670(.0106) a'(d3;2_!;2)->a'(d:l!j 48% ; a'(SOj,)^a'(da.z) 36 % 21 657(.0068) a'(SOI,)->a"(d,,J 58 % ; a(dx2_y2)^a"(dyz) 26 % 23 029(.0002) a"(S op )^a"(d,, z ) 57 % ; a"(SOJ,)^a"(d,,y) 20 % 23 131(.O257) a"(S O! ,)^a'(d ci ) 74 % ; a'(d3;2_y2)->a"(d.1/,) 5 % 23 276(.0023) a"(S,;j,)^a'(d:c2_,y2) 90 % 23 467(.0044) a'(S o} ,)^a"(d y J 71 % ; a'(dal2_!/2)^a"(da;!/) 6 % 23 985(.0431) a'(S OJ) )^a'(d,, 0 ) 64 % ; a'(d:i;2_,y2)^a'(d,,z) 21 % 25 655(.0038) a (Sop)^a" (dyz) 17 % ; a(Sop)^a"(dxy) 39 %
f 2 A' h 2 A"
26 274(.0006) 26 308(.0144)
State
a'(S,:j,)->a'(d3:2_v2) 89 % a"(S o ,,)^a"(d 3 , z ) 50 % ; a'(d c 2_,, 2 )^a"(d,, z ) 21 % ;
g 2 A' 27 030(.0111) a'(S OJ ,)^a'(d l;0 ) 71 % ; a'(d :l; 2_, y 2)^a'(d l;z ) 14 % h 2 A' 27 395(.0131) a"(S o ,,)^a"(d,,J 62 % ; a"(SOj,)-J-a"(d:,:iy) 18 % i 2 A" 33 270(.0146) a (So,,)^'(d^y) 72 % ; a'(da!2_v2)^a"(d;!1!y) 13% i 2 A' 34 121(.0043) a"(S o p )^a"(d : r ; / ) 66 % ; a " ( S q , ) - » a " ( g 13 % j 2 A" 36 715(.0011) a.'(Sop)^'(dvy) 45 % ; a'(da!2_!/2)^a"(d;I!y) 35% j 2 A' 37 284(.0016) a"(S o p )^a"(d : r ; / ) 63 % ; a"(S^^a (dyZ) 17 % "Calculations were performed on the B3LYP-DFT structure Excitation energies are given in cm ~ : r Only the main contributions (> 5%) are shown
experiment exc.energy h (osc.str.) 9 100(.0056) 13 100(.0033) 15 800 (-) 19 400(.0160)
22 100(.0170)
25 100(.0920)
306
a(d,.2_y2)
a {dxz)
a (SOJ))
a'(So,,)
a"(S, :p ) Fig. 10. Plot of the (natural) orbitals involved in the lowest electronic excitations in (Tp)MoO(bdt)
307
Fig. 11. Schematic plot of the three-center pseudo-a bonding and antibonding combinations formed by the Mo Ad.xi-yi and a'(S,;p) orbitals
excitations out of a (Sop) and a (Sop), the former being the lowest. The calculated transition energies, 10 912 cm" 1 and 14 326 cm" 1 are 1 200-1 800 cm" 1 too high, whereas the oscillator strength of both transitions is grossly overestimated by the calculations. Nevertheless, there can be little or no doubt concerning the character of the lowest two excited states in these spectra. The same is not true for the feature at 19 400 cm" 1 . Although this band was originally (for (Tp*)MoO(tdt)) assigned as the Mo 4dx2_y2->Adxy LF transition [134], all later studies [128, 130, 135] report the assignment of the 19 400 cm" 1 as the in-plane a'(Sj,,->Mo 4rfa.2_y2 LMCT transition. The latter is based on the relatively high intensity of this band, which supposedly originates from a large overlap between the donor and acceptor orbital involved. In order to explain such overlap it was suggested that both orbitals are involved in a highly covalent threecentered pseudo a bond, as shown schematically in Fig. 11. However, our calculations do not give any indication for the existence of such a three-centered bond. As can be seen from Fig. 10 the Mo 4rfiI2_y2 and Srp orbitals do not show any significant mixing. Correspondingly, the calculated oscillator strength of the a'(S,p)->Mo 4dx2_y2 excitation is very low, 6 x 10~4. This excitation is predicted to occur at 26 274 cm" 1 (state f 2A' in Table 4). The corresponding a"(S,;}))—>Mo 4dx2_y2 excitation is predicted at 23 276 cm" 1 with an oscillator strength of 2.3 x 10"3 (state e 2A" in Table 4). Both transitions should in the experimental spectrum be obscured by the occurrence in the same region of much more intense transitions originating from a (SO2,) and a"(Sop. The present results therefore throw a new light on the possible role of the pyranopterin dithiolate Sn, orbital in the catalytic activity of oxomolybdenum enzymes, either acting as an electron transfer conduit or modulating the molybdenum reduction potential. Facile electron transfer in general requires that there be a high degree of electronic communication between the donor and acceptor sites. In ref. [128] it was suggested that the highly covalent three-center a interaction between the a'(S,{)) orbital and the redox active Mo 4dx2_y2 orbital should be involved in and responsible for efficient pyranopterin mediated electron transfer. The in-plane nature of this interaction is convenient, since it may be expected to give a minimal contribution to the reorganization energy involving distortions along the apical M=O bond during electron transfer reactions. Furthermore, this effective in-plane covalency was also anticipated [128]
308 to play an important role in modulating the electrochemical potential of the active site, by destabilizing the energy of the redox active (predominantly) Mo Ad,r2_y2 orbital. The latter role was however put into perspective by a number of comparative studies of (Tp*)MoO(S-S) complexes with different (S-S) ligands, such as tdt, qdt or bdtCl2. The electronic absorption spectra of these complexes all contain the same band around 19 400 cm" 1 with a very similar transition energy and intensity. This would suggest (still assuming that this band indeed corresponds to a transition between the bonding and antibonding combinations of Mo d X2_y2 and a'(S,j,)) that the covalency of the three-center pseudo a bond and its destabilizing effect on the redox active orbital is indifferent to the character of the (S-S), in particular on the presence of electron withdrawing or donating groups. On the other hand, electrochemical and He(I) PES studies [130, 131, 135, 139] have indicated that both the ease of oxidation and reduction of the Mo(V) center is indeed strongly affected by the presence of such groups. In particular, these studies indicate that the redox active orbital is stabilized as the ene 1,2 dithiolate ligand becomes more electron withdrawing, thus making ionization more difficult and reduction easier. Based on these observations it was proposed that instead anisotropic covalency contributions involving only the Sop orbitals of the coordinated thiolate control the Mo reduction potential by modulating the effective nuclear charge of the metal. In refs 135, 140 it is suggested that a regulating factor of this interaction is the S-S fold angle of the ene dithiolate chelate ring, i.e. the angle between the S-C=C-S and S-Mo-S planes. Folding the S-C=C-S plane upwards in the direction of the M=O bond leads to an increased overlap between the the a (Sop) and the Mo idx2_y2 orbital, thus giving rise to a reduced effective charge on the metal. Such overlap is indeed manifested by the plot of the bonding a'(SOJ)) orbital in Fig. 10, although it is virtually absent in the corresponding antibonding orbital, i.e. the ground state singly occupied orbital a(d.j.2_y2). The role of the SOJ) orbitals and the effect of the S-S fold angle on the electronic structure of these (Tp*)MoO(S-S) complexes and pyranopterin Mo enzymes they stand model for certainly warrants further investigation. Returning to the spectroscopic data for (Tp)MoO(bdt) in Table 4 and to the band at 19 400 cm"1 in particular, our calculations indicate that this band should instead be assigned as the Mo dx2-y2—>dxz LF transition, which is however strongly mixed with a'(Sop)—>dxz LMCT character. This mixing may explain the intensity of the 19 400 cm" 1 band, although the calculated oscillator strength is significantly lower than the experimental value. A second ligand field transition, Mo dx2_y2—tdyz is predicted to occur at lower energy, 16 457 cm" 1 , again strongly mixed with CT character out of a'(Sop) into the same id orbital. The calculated oscillator strength of this transition is very low. This is in agreement with the appearance of a very weak feature (only discernable in low-temperature MCD) at 15 800 cm" 1 in the experimental spectrum, which is therefore assigned as such. The third LF exitation, Mo dx2_y2^dxy is situated at a considerably higher energy. The lowest excited state containing a significant (6 %) dx2_y2^dxy contribution is the f 2A" state at 23 467 cm" 1 . However, none of the states included in Table 4 can even grosso modo be identified as the transition to dxy, the largest contribution, 35 %, being found for the j 2 A" state at 36 715 cm" 1 . The high energy of the metal dxy orbital is of course related to the strong antibonding character (primarily with S,p but also with the equatorial N u-donor
309 ligands) of this orbital, as is obvious from the orbital plot in Fig. 10. Ranging from 21 000 to 27 400 cm" 1 a whole series of transitions is calculated, corresponding to excitations from the a'(SOJ,), a"(SOJ>) orbitals into Mo dxz, dyz. In the experimental spectrum, two intense bands are distinguished in this region: a shoulder at 22 100 cm" 1 and a very strong band at 25 100 cm" 1 . They are assigned to the two excited states for which the calculations predict the highest oscillator strength: d 2A", calculated at 23 131 cm" 1 and e 2A', calculated at 23 985 cm" 1 . Both states primarily correspond to an excitation into Mo dxz. Corresponding excitations to Mo dyz are calculated at slightly lower energies but with a much lower oscillator strength. Finally, excitations out of the a'(SOj,), a"(SO2,) orbitals into Mo dxy are predicted to occur between 33 000 and 37 200 cm" 1 , i.e. outside the range of the experimental spectrum. 6. CONCLUSION In this chapter we have described a few theoretical studies of electronic spectra of transition metal ions in different coordination environments. We believe that these studies have convincingly demonstrated the strength of accurate computational approaches: (i) in explaining and predicting the relative position and intensities of different types of excited states (LF or CT) in these spectra; (ii) in relating this information to the ground state electronic structure and bonding characteristics of different metal-ligand combinations; (iii) in relating to the ground state electronic structure and spectroscopic characteristics other important properties such as the geometrical structure of the coordination environment or the possibility for redox activity of the central metal ion. An illustrative example is presented by the calculations on the Cu(II) ion in two thoroughly different coordination environments: the ionic oxygen ligand surrounding of a zeolite surface on the one hand, leading to an almost strictly square planar Cu(II) surrounding and a typical ligand field spectrum, and the trigonal Cu(II) surrounding in blue copper proteins on the other hand, caused by the presence of a highly covalent Cu-S cVa "" bond, the latter also giving rise to a low-lying intense LMCT band, i.e. the "blue" band. The calculations on (Tp)MoO(bdt) have indicated that the Mo(V)-dithiolate bond in these model complexes is in fact highly ionic, as opposed to the earlier suggested presence of a covalent three-center S-Mo-S pseudo a bond. The question of course remains whether this complex is indeed a representative model for pterin-containing molybdenum enzymes and how to explain the redox activity of such enzymes. Finally, the calculations on MnO J have demonstrated how the CASPT2/CASSCF method, used throughout this work, is able to provide very accurate results also for the notoriously difficult-to-calculate spectrum of this molecule, provided that the active space is chosen large enough to include all (very important) nondynamic correlation effects.
310 References [1] L. Fan, T. Ziegler, J. Chem. Phys. 95 (1991) 7401. [2] L. Fan, T. Ziegler, J. Phys. Chem. 96 (1993) 6940. [3] B. Delley, M. Wrinn, H. P. Liithi, J. Chem. Phys. 100 (1994) 5785. [4] M. Diedenhofen, T. Wagener, G. Frenking, in: R. Cundari (Ed.), Computational organometallic chemistry, Marcel Dekker Inc., New York-Basel, 2001, pp. 69-121. [5] C. Daul, Int. J. Quantum Chem. Symp. 52 (1994) 867. [6] M. E. Casida, in: D. P. Chong (Ed.), Recent Advances in Density Functional Theory, Vol. 1, World Scientific, Singapore, 1995, pp. 155-192. [7] C. Jamorski, M. E. Casida, D. R. Salahub, J. Chem. Phys. 104 (1996) 5134. [8] E. U. K. Gross, J. F. Dobson, M. Petersilka, Topics in Current Chemistry 181 (1996) 82-172. [9] E. I. Solomon, R. K. Szilagyi, S. D. George, L. Basumallick, Chem. Rev. 104 (2004) 419^158. [10] H. Yersin, W. Humbs, J. Strasser, in: H. Yersin (Ed.), Electronic and Vibronic Spectra of Transition Metal Complexes II, Vol. 191 of Topics in Current Chemistry, Springer, Berlin, 1997, pp. 153-249. [11] K. Pierloot, A. Ceulemans, M. Merchan, L. Serrano-Andres, J. Phys. Chem. A 4374-4382 (2000) 16545-16550. [12] C. Daniel, Coord. Chem. Rev. 238-239 (2003) 143-166. [13] K. Andersson, P.-A. Malmqvist, B. O. Roos, J. Chem. Phys. 96 (1992) 1218-1226. [14] G. Karlstrom, R. Lindh, P.-A. Malmqvist, B. O. Roos, U. Ryde, V. Veryazov, P.-O. Widmark, M. Cossi, B. Schimmelpfennig, P. Neogrady, L. Seijo, Computational Material Science 28 (2003) 222. [15] J. E. Rice, H. Horn, B. H. Lengsfield, A. D. McLean, J. T. Carter, E. S. Replogle, L. A. Barnes, S. A. Maluendes, G. C. Lie, M. Gutwski, W. E. Rude, S. P. A. Sauer, R. Lindh, K. Andersson, T. S. Chevalier, P.-O. Widmark, D. Bouzida, G. Pacansky, K. Singh, C. J. Gillan, P. Carnevali, W. C. Swope, B. Liu, Mulliken™ Version 2.31h, internal release, IBM Corporation, Almaden, USA (1995). [16] R. Ahlrichs, M. Bar, M. Haser, H. Horn, C. Kolmel, Chem. Phys. Lett. 162 (1989) 165. [17] K. Pierloot, Mol. Phys. 101 (2003) 2083-2094. [18] K. Pierloot, in: R. Cundari (Ed.), Computational organometallic chemistry, Marcel Dekker Inc., New York-Basel, 2001, pp. 123-158. [19] K. Andersson, B. O. Roos, Chem. Phys. Letters 191 (1992) 507-514. [20] P. E. M. Siegbahn, in: I. Prigogine, S. A. Rice (Eds.), Advances in Chemical Physics: New Methods in Computational Quantum Mechanics, Vol. XCIII, John Wiley & Sons, New York, 1996, pp. 333-387.
311 [21] G. Frenking, I. Antes, M. Bohme, S. Dapprich, A. W. Ehlers, V. Jonas, A. Neuhaus, M. Otto, R. Stegmann, A. Veldkamp, S. F. Vyboishchikov, in: K. B. Lipkowitz, D. B. Boyd (Eds.), Reviews in Computational Chemistry, Vol. 8, VCH Publishers, Inc., New York, 1996, pp. 63-144. [22] A. W. Ehlers, S. Dapprich, S. F. Vyboishchikov, G. Frenking, Organometallics 15 (1996) 105117. [23] B. J. Persson, B. O. Roos, K. Pierloot, J. Chem. Phys. 101 (1994) 6810-6821. [24] K. Pierloot, E. Tsokos, L. G. Vanquickenborne, J. Phys. Chem. 100 (1996) 16545-16550. [25] K. Pierloot, E. Tsokos, B. O. Roos, Chem. Phys. Letters 214 (1993) 583. [26] K. Pierloot, B. J. Persson, B. O. Roos, J. Phys. Chem. 99 (1995) 3465-3472. [27] K. Andersson, Theor. Chim. Acta 91 (1995) 31. [28] B. O. Roos, K. Andersson, Chem. Phys. Letters 245 (1995) 215-223. [29] B. O. Roos, K. Andersson, M. P. Fulscher, P.-A. Malmqvist, L. Serrano-Andres, K. Pierloot, M. Merchan, in: I. Prigogine, S. A. Rice (Eds.), Advances in Chemical Physics: New Methods in Computational Quantum Mechanics, Vol. XCIII, John Wiley & Sons, New York, 1996, pp. 219-331. [30] B. O. Roos, K. Andersson, M. P. Fulscher, L. Serrano-Andres, K. Pierloot, M. Merchan, V. Molina, J. Mol. Struct. (Theochem) 388 (1996) 257-276. [31] Atlas of Zeolite Structure Types, Structure Commission of the International Zeolite Association (IZA-SC), http://www.iza-structure.org/databases. [32] A. B. P. Lever, Inorganic Electronc Spectroscopy, Elsevier, Amsterdam, 1984. [33] K. Klier, Adv. Chem. Ser. 101 (1971)480. [34] K. Klier, P. J. Hutta, R. Kellerman, ACS Symp. Ser. 40 (1977) 108. [35] K. Klier, Langmuir 4 (1988) 13. [36] J. Texter, D. H. Storme, R. G. Herman, K. Klier, J. Phys. Chem. 81 (1977) 333. [37] D. H. Storme, K. Klier, ACS Symp. Ser. 135 (1980) 155. [38] R. A. Schoonheydt, Academiae Analecta 49 (1987) 3. [39] D. Packet, R. A. Schoonheydt, ACS Symp. Ser. 368 (1988) 203. [40] R. A. Schoonheydt, J. Phys. Chem. Solids 50 (1989) 523. [41] R. A. Schoonheydt, Catal. Rev. - Sci. Eng. 35 (1993) 129. [42] Y. Yokomichi, H. Ohtsuka, T. Tabata, O. Okada, Y. Yokoi, H. Ishikawa, R. Yamaguchi, H. Matsui, A. Tachibana, T. Yamabe, Catalysis Today 23 (1995) 431. [43] Y. Yokomichi, T. Yamabe, H. Ohtsuka, T. Kakumoto, J. Phys. Chem. 100 (1996) 14424.
312 [44] N. U. Zhanpeisov, H. Nakatsuji, H. Hada, M. Anpo, Catalysis Letters 42 (1996) 173. [45] B. L. Trout, A. K. Chakraborty, A. T. Bell, J. Phys. Chem. 100 (1996) 4173^1179. [46] B. L. Trout, A. K. Chakraborty, A. T. Bell, J. Phys. Chem. 100 (1996) 17582. [47] R. J. Blint, J. Phys. Chem. 100 (1996) 19518. [48] H. Oka, T. Okada, K. Hori, J. Mol. Catal. A: Chemical 109 (1996) 51. [49] W. F. Schneider, K. C. Hass, R. Ramprasad, J. B. Adams, J. Phys. Chem. 100 (1996) 6032. [50] W. F. Schneider, K. C. Hass, R. Ramprasad, J. B. Adams, J. Phys. Chem. B 101 (1997) 4353. [51] H. V. Brand, A. Redondo, P. J. Hay, J. Phys. Chem. B 101 (1997) 7691. [52] W. F. Schneider, K. C. Hass, R. Ramprasad, J. B. Adams, J. Phys. Chem. B 102 (1998) 36923705. [53] L. Rodriguez-Santiago, M. Sierka, V. Branchadell, M. Sodupe, J. Sauer, J. Am. Chem. Soc. 120 (1998) 1545. [54] B. R. Goodman, W. F. Schneider, K. C. Hass, J. B. Adams, Catal. Lett. 56 (1998) 183-188. [55] B. R. Goodman, K. C. Hass, W. F. Schneider, J. B. Adams, J. Phys. Chem. B 103 (1999) 1045210460. [56] N. Tajima, M. Hashimoto, F. Toyama, A. M. El-Nahas, K. Hirao, Phys. Chem. Chem. Phys. 1 (1999)3823. [57] D. Nachtigallova, P. Nachtigall, M. Sierka, J. Sauer, Phys. Chem. Chem. Phys. 1 (1999) 2019. [58] X. Solans-Monfort, V. Branchadell, M. Sodupe, J. Phys. Chem. A 104 (2000) 3225. [59] P. Nachtigall, D. Nachtigallova, J. Sauer, J. Phys. Chem. B 104 (2000) 1738. [60] M. J. Rice, A. K. Chakrabotry, A. T. Bell, J. of Catal. 194 (2000) 278. [61] M. J. Rice, A. K. Chakrabotry, A. T. Bell, J. Phys. Chem. B 104 (2000) 9987. [62] D. Nachtigallova, P. Nachtigall, J. Sauer, Phys. Chem. Chem. Phys. 3 (2001) 1552-1559. [63] P. J. Carl, S. L. Isley, S. C. Larsen, J. Phys. Chem. A 105 (2001) 4563. [64] D. Sengupta, J. B. Adams, W. F. Schneider, K. C. Hass, Catal. Lett. 74 (2001) 193-199. [65] M. Sierka, J. Sauer, Faraday Discuss. 106 (1997) 41. [66] J. Sauer, Chem. Rev. 89 (1989) 199. [67] K. Pierloot, A. Delabie, C. Ribbing, A. A. Verberckmoes, R. A. Schoonheydt, J. Phys. Chem. B 102(1998)10789. [68] K. Pierloot, A. Delabie, M. H. Groothaert, R. A. Schoonheydt, Phys. Chem. Chem. Phys. 3 (2001) 2174.
313 [69] A. Delabie, K. Pierloot, M. H. Groothaert, B. M. Weckhuysen, R. A. Schoonheydt, Phys. Chem. Chem. Phys. 4 (2001) 134-145. [70] M. H. Groothaert, K. Pierloot, A. Delabie, R. A. Schoonheydt, Phys. Chem. Chem. Phys. 5 (2003) 2135-2144. [71] A. Delabie, K. Pierloot, M. H. Groothaert, B. M. Weckhuysen, R. A. Schoonheydt, Micropor. Mesopor. Mat. 37 (2000) 209. [72] A. Delabie, K. Pierloot, M. H. Groothaert, R. H. Schoonheydt, L. G. Vanquickenborne, Eur. J. Inorg. Chem. (2002) 515-530. [73] P. E. Riley, K. Seff, Inorg. Chem. 13 (1974) 1355. [74] H. S. Lee, K. Seff, J. Phys. Chem. 85 (1981) 397. [75] P. Gallezot, B. Imelik, J. Chim. Phys. 71 (1974) 155. [76] P. Gallezot, Y. B. Taarit, B. Imelik, J. Catal. 26 (1972) 295. [77] I. E. Maxwell, J. J. de Boer, J. Phys. Chem. 79 (1975) 1874. [78] R. M. Haniffa, K. Seff, Micropor. Mesopor. Mat. 25 (1998) 137. [79] A. A. Verberckmoes, B. M. Weckhuysen, R. A. Schoonheydt, K. Ooms, I. Langhans, Anal. Chim. Acta 348 (1997) 267. [80] W. Loewenstein, Amer. Miner. 39 (1942) 92. [81] J. Sugar, C. Corliss, J. Phys. Chem. Ref. Data 6 (1977) 317. [82] K. Pierloot, A. Delabie, A. A. Verberckmoes, R. A. Schoonheydt, in: P. Geerlings, F. D. Proft, W. Langenaeker (Eds.), Density Functional Theory, a Bridge between Chemistry and Physics, VUB University Press, Brussels, 1999, pp. 169-188. [83] Z. Sobalfk, J. Dedecek, I. Ikonnikov, B. Wichterlova, Microporous and Mesoporous Materials 21 (1998)525. [84] Z. Sobalfk, Z. Tvariizkova, B. Wichterlova, J. Phys. Chem. B 102 (1998) 1077. [85] D. Goldfarb, L. Kevan, J. Magnet. Res. 82 (1989) 270. [86] M. Wolfsberg, L. Helmholz, J. Chem. Phys. 20 (1952) 837-843. [87] I. H. Hillier, V. R. Saunders, Proc. Roy. Soc. Lond. A. 320 (1970) 161-173. [88] I. H. Hillier, V. R. Saunders, Chem. Phys. Lett. 9 (1971) 219-221. [89] H. Hsu, C. Peterson, R. M. Pitzer, J. Chem. Phys. 64 (1976) 791-795. [90] H. Johanson, Mol. Phys. 49 (1983) 1209-1216. [91] H. Johanson, S. Rettrup, Chem. Phys. 74 (1983) 77. [92] K. H. Johnson, J. F. C. Smith, Chem. Phys. Lett. 10 (1971) 219-223.
314 [93] T. Ziegler, A. Rauk, E. J. Baerends, Chem. Phys. 16 (1976) 209-217. [94] R. M. Dickson, T. Ziegler, Int. J. Quantum Chem. 58 (1996) 681-687. [95] A. C. Stiickl, C. A. Daul, H. U. Giidel, J. Chem. Phys. 107 (1997) 4606^1617. [96] S. J. A. van Gisbergen, J. A. Groeneveld, A. Rosa, J. G. Snijders, E. J. Baerends, J. Phys. Chem. A 103(1999)6835-6844. [97] P. Boulet, H. Chermette, C. Daul, F. Gilardoni, F. Rogemond, J. Weber, G. Zuber, J. Phys. Chem. A 105(2001)885-894. [98] H. Nakai, Y. Ohmori, H. Nakatsuji, J. Phys. Chem. A 103 (1999) 6835-6844. [99] M. Nooijen, V. Lotrich, J. Chem. Phys. 113 (2000) 494-507. [100] M. A. Buijse, E. J. Baerends, J. Chem. Phys. 93 (1990) 4129^141. [101] K. Pierloot, The electronic spectrum of permanganate by a multiconfigurational perturbational approach, manuscript in preparation (2004). [102] S. L. Holt, C. J. Ballhausen, Theor. Chim. Acta 7 (1967) 313. [103] C. J. Ballhausen, H. B. Gray, in: A. E. Martell (Ed.), Coordination Chemistry, Vol. 1, Van Nostrand Reinhold, New York, 1971, Ch. 1, pp. 3-83. [104] J. M. Guss, P. R. Harrowell, M. Murata, V. A. Norris, H. C. Freeman, J. Mol. Biol. 192 (1986) 361-387. [ 105] R. A. Marcus, N. Sutin, Biochim. Biophys. Acta 811(1985) 265-322. [106] H. B. Gray, B. G. Malmstrom, Comments Inorg. Chem. 2 (1983) 203-209. [107] B. G. Malmstrom, Eur. J. Biochem. 223 (1994) 207-216. [108] B. L. Vallee, R. J. P. Williams, Proc. Nat. Acad. Sci. 59 (1968) 498-505. [109] R. J. P. Williams, Eur. J. Biochem. 234 (1995) 363-381. [110] U. Ryde, M. H. M. Olsson, K. Pierloot, B. O. Roos, J. Mol. Biol. 261 (1996) 586-596. [Ill] U. Ryde, M. H. M. Olsson, B. O. Roos, J. O. A. D. Kerpel, K. Pierloot, J. Biol. Inorg. Chem. 5 (2000) 565-574. [112] M. H. M. Olsson, U. Ryde, B. O. Roos, K. Pierloot, J. Biol. Inorg. Chem. 3 (1998) 109-125. [113] B. G. Malmstrom, J. Leckner, Current Opinion in Chemical Biology 2 (1998) 286-292. [114] K. W. Penfield, A. A. Gewirth, E. I. Solomon, J. Am. Chem. Soc. 107 (1985) 4519^519. [115] A. A. Gewirth, E. I. Solomon, J. Am. Chem. Soc. 110 (1988) 3811-3819. [116] S. Larsson, A. Broo, L. Sjolin, J. Phys. Chem. 99 (1995) 4860^1865. [117] K. Pierloot, J. O. A. De Kerpel, U. Ryde, B. O. Roos, J. Am. Chem. Soc. 119 (1997) 218-226.
315 [118] A. G. Sykes, Adv. Inorg. Chem. 36 (1990) 377-408. [119] E. I. Solomon, J. W. Hare, D. M. Dooley, J. H. Dawson, P. J. Stephens, H. B. Gray, J. Am. Chem. Soc. 102 (1980) 168. [120] K. W. Penfield, R. R. Gay, R. S. Himmelwright, N. C. Eickman, V. A. Norris, H. C. Freeman, E. I. Solomon, J. Am. Chem. Soc. 103 (1981) 4382. [121] U. Ryde, M. H. M. Olsson, K. Pierloot, in: L. A. Eriksson (Ed.), Theoretical and Computational Chemistry, Vol. 9, Elsevier, Amsterdam, 2001, pp. 1-56. [122] J. M. Guss, H. D. Bartunik, H. C. Freeman, Acta Crystallogr. B48 (1992) 790-807. [123] K. Pierloot, J. O. A. De Kerpel, U. Ryde, M. H. M. Olsson, B. O. Roos, J. Am. Chem. Soc. 120 (1998) 13156-13166. [124] J. Han, T. M. Loehr, Y. Lu, J. S. Valentine, B. A. Averill, J. Sanders-Loehr, J. Am. Chem. Soc. 115(1993)4256^1263. [125] Y. Lu, J. A. Roe, E. B. Gralla, J. S. Valentine, in: K. D. Karlin, Z. Tyeklar (Eds.), Bioinorganic Chemistry of Copper, Chapman & Hall, New York, 1993, p. 64. [126] E. T. Adman, J. W. Godden, S. Turley, J. Biol. Chem. 270 (1995) 27458-27474. [127] Y. Lu, L. B. Lacroix, M. D. Lowery, E. I. Solomon, C. J. Bender, J. Peisach, J. A. Roe, E. B. Gralla, J. S. Valentine, J. Am. Chem. Soc. 115 (1993) 5907. [128] F. E. Inscore, R. McNaughton, B. L. Westcott, M. E. Helton, R. Jones, I. K. Dhawan, J. E. Enemark, M. L. Kirk, Inorg. Chem. 38 (1999) 1401-1410. [129] R. Hille, Chem. Rev. 96 (1996) 2757-2816. [130] M. E. Helton, M. L. Kirk, Inorg. Chem. 38 (1999) 4384^1385. [131] M. E. Helton, N. E. Gruhn, R. I. Mcnaughton, M. L. Kirk, Inorg. Chem. 39 (2000) 2273-2278. [132] C. K. J0rgensen, Acta Chem. Scand. 11 (1957) 73-85. [133] D. J. Collison, J. Chem. Soc. Dalton Trans. (1990) 2999-3006. [134] M. D. Carducci, C. Brown, E. Solomon, J. H. Enemark, J. Am. Chem. Soc. 116 (1994) 1185611868. [135] F. E. Inscore, H. K. Joshi, A. E. McElhaney, J. H. Enemark, Inorg. Chim. Acta 331 (2002) 246256. [ 136] K. Pierloot, K. Gaethofs, manuscript in preparation (2004). [137] D. Andrae, U. Haeussermann, M. Dolg, H. Stoll, Theor. Chim. Acta 77 (1990) 123. [138] P. A. Malmqvist, B. O. Roos, Chem. Phys. Letters 155 (1989) 189-194. [139] A. E. McElhaney, F. E. Inscore, J. T. Schirlin, J. H. Enemark, Inorg. Chim. Acta 341 (2002) 85-90. [140] H. K. Joshi, J. J. A. Cooney, F. E. Inscore, N. E. Gruhn, D. L. Lichtenberger, J. H. Enemark, Proc. Nat. Acad. Sci. 100 (2003) 3719-3724.
This Page is Intentionally Left Blank
M. Olivucci (Editor) Computational Photochemistry Theoretical and Computational Chemistry, Vol. 16 © 2005 Elsevier B.V. All rights reserved
317
X. Perspectives in Calculations on Excited State in Molecular Systems Bjorn O. Roos Department of Theoretical Chemistry, Chemical Center P.O.B. 124, S-221 00 Lund, Sweden 1. INTRODUCTION The electronic Schrodinger equation has for most molecular systems an infinite number of bound solutions. Each of them corresponds to an electronic state with a specific electronic structure. Such a state gets populated when the molecule absorbs one or more quanta of light (photons). When a molecule is brought into one of the excited states, one of several things can happen. The energy surface for the excited state is different from that of the ground state, because the electronic structure has changed. The excitation process is fast, so the atoms will not have time to move. As a result, the molecule will find itself in a non-equilibrium position and will start vibrating. Ultimately, it may end up in the equilibrium geometry of the excited state (which in many cases means dissociation of one or several chemical bonds). That is, if there is time enough because the lifetime of the excited state is usually short and a competing process is emission of a light quantum and return to the ground state energy surface directly or via lower lying electronic states. But the dynamics may be even more complex. We show an example in Fig.l (for simplicity, the energy is shown for one geometrical coordinate only). A photon excites the system from the lowest vibrational level on the ground state energy surface, X, vertically to the first excited state surface, B. This state is bound but with another geometry than the ground state. Another state, A, crosses state B at a larger value for the geometry parameter. The crossing point lies below the energy that the molecule has adsorbed from the photon. So, the molecule has several possibilities. It can stay on surface B and loose the extra energy by thermal dissipation (possibly involving other molecules). Ultimately, it will then loose its energy by emitting a photon of lower energy (fluorescence). Or it can cross over the surface A and dissociate. It can also come back to surface B via the crossing point. Today it is possible to follow such dynamical processes experimentally using time-resolved laser spectroscopy. How can we deal with such problems theoretically and computationally? It is clearly not straightforward and simple. We need to be able to compute with good accuracy the energy surfaces for a number of excited states and we then need to be able to solve the dynamical problem for the nuclear motion on these surfaces. In this chapter we shall deal with the first of these problems. Chapters 5 and 7 will deal with different aspects of the nuclear dynamics.
318
hv
— X
Geometry
Fig.l. Energy surfaces in a hypothetical molecular system. Four electronic states are shown. The energies are given as functions of one arbitrary geometrical variable.
2. A BRIEF HISTORICAL BACKGROUND The history of electronic spectroscopy of atoms and molecules is older than quantum mechanics. Actually much of the experimental background that inspired the development of quantum mechanics in the beginning of the previous century was found in spectroscopy. It was a grand and immediate success of the Schrodinger equation [1] that it could explain with high accuracy the electronic spectrum of the hydrogen atom. The exact solution of this problem gave us the concept of the atomic orbital. It was almost immediately extended to many-electron systems by the introduction of the self-consistent field model, were the electron-electron interaction is modeled using a mean-field approximation. The model was introduced for atoms by R. D. Hartree in 1928 [2]. A similar model was actually proposed for molecules already in 1927 by F. Hund [3]. The approach was immediately used by R. S. Mulliken and others for the interpretation of electronic spectra of small (diatomic) molecules [4]. Today we know this approach as the Hartree-Fock (HF) method. The basic concept is the molecular orbital (MO) and its energy, leading to a shell model for molecules similar (but more complex in structure) to that of the atoms. This theoretical model explains the excited states of a molecule as follows: in the ground state the electrons reside in the MOs of lowest energy (the aufbau principle). If one electron absorbs a photon of the right energy it will move to an MO of higher energy, which is empty or only singly occupied in the ground state. This is a very powerful approach and forms still today the basis for our analysis of excited states in molecular systems. We now know that one has to go beyond the self-consistent field model to obtain quantitative results for excited states but we can always (almost) label them as excitations by one or more electrons from one MO to another. The development discussed above concerned mainly atoms and small molecules. A simple approach that was to become very influential for the development of the theory of electronic spectra of planar conjugated organic molecules was suggested by E. Huckel [5]. Huckel
319
Fig.2. The orbital energy levels for the 7t-electrons in the benzene molecule. simplified the problem of the many electrons by only considering the loosely bound electrons in the 7t-orbitals. The method that emerged was a simple semi-empirical scheme that could be used to compute the structure and energy of these 7i-orbitals. The use of the approach for the interpretation of electronic spectra was, however, limited because the important electronelectron repulsion terms were not accounted for. This problem was realized by M. GoeppertMayer and A. L. Sklar in their landmark study of the electronic spectrum of the benzene molecule in 1938 [6]. In order to understand the importance of the electron-electron interaction, let us take a closer look at the energy levels in benzene. Experimentally, four excited singlet states are known which originates from excitations within the valence 7i-orbitals with energies around 4.9, 6.2, 7.0, and 7.8 eV. The energy levels for the orbitals are shown in Fig. 2. Actually, we can form four excited configurations by moving one electron from an occupied to a virtual orbital and one might think that this corresponds to the four singly excited valence states seen in the electronic spectrum of benzene. This is, however, a too simple picture. Consider the singly excited configuration a ^ e ^ e ^ . We can form 16 Slater determinants for this configuration because of the double degeneracy of the e2U and eig orbitals in the D^ point group of the molecule. 12 of them correspond to triplet states and the four others to singlet states. A little bit of group theory shows that they are: 'B2U, 'BIU, and 'EH, . In a Hartree-Fock picture they have the same one-electron energy, but the electron-electron repulsion energies are different. Goeppert-Mayer and Sklar succeeded to calculate these energies, which was quite an achievement in 1938, even if they were forced to use crude approximations. We refer to their paper for details [6]. What about the fourth state at 7.8 eV? Today we know that this state has gerade symmetry. It can thus be formed from the single excitations aiu—>e2U or eig—»big . We know now that the wave function is a combination of these two configurations, but that also the double excitation efg —^e^ gives a sizable contribution. It is thus an oversimplification to assume that only single excitations are important for the excited states. We shall later see that double excitations are sometimes of crucial importance for the description of the electronic structure. It was to take another 15 years until further progress was made. A rather straightforward extension of the ideas of Goeppert-Mayer and Sklar would be the following: represent each of
320
the Ti-orbitals in a planar organic molecule with one basis function. Determine the MOs by solving a Hartree-Fock like problem. Then, expand the wave function for the excited states, V, as a linear combination of singly (and possibly multiply) excited configurations, O^ starting from the closed shell HF wave function:
(1)
^=1 ^
Use the variational principle to determine the expansion coefficients Q,. 50 years ago it was an impossible task to carry out such a calculation from first principles. Drastic approximations had to be made. One key problem was the calculation of all electron-electron repulsion integrals, which occur in the matrix element of the Hamiltonian:
(pg\rs)= \\^p{rl)^p — ^r{r2)Ur2)dVldV2 r
(2)
12
where <j>p etc. are the molecular 7t-orbitals. Even if each atomic basis function was a single Slater orbital, it was an impossible task to compute these integrals in 1950. Robert Parr then introduced a drastic approximation: the zero differential overlap, ZDO approximation [7]. All products of basis function, which where located on different centra were assumed to be zero everywhere. The only remaining integrals were then of the type (pp|rr) and the calculation of the matrix elements were drastically simplified. The integrals could be computed using a single Slater type orbital to represent the AOs. The one-electron matrix elements were taken from Hiickel theory. However, very accurate excitation energies were still not obtained. Rudolph Pariser [8] suggested that the one-center integrals for carbon should also be determined empirically from the valence state disproponation reaction: 2C->C~ +C+
(3)
The corresponding energy is (pp|pp) = IP - EA, where IP is the ionization energy and EA the electron affinity. Using experimental values for these quantities gave the value 11 eV for the one-center integral (pp|pp), a reduction of 6 eV from the theoretical value. These are the key ingredients in the, so called, Pariser-Parr-Pople (PPP) method to compute electronic spectra of planar conjugated hydrocarbons. It was later extended also to hetero-atoms and actually also to transition metal complexes with some success. For a nice historical review of the development we refer to articles by Pariser, Parr, and Pople in Ref. [9]. The PPP method was an immediate success and it became possible to assign a large number electronic states in conjugated organic molecules. Why was that? What was the underlying reason for the large reduction of the one-center two-electron integral? Actually, this was well understood immediately and Pariser wrote: the effect of the a-electrons can be approximately taken into account by changing the value of primarily one Coulomb repulsion integral [8]. Today we call this effect dynamic polarization. It is maybe easiest demonstrated on the Vstate of the ethene molecule C2H4. While the ground state of this molecule has two electron in the bonding 71-orbital, the V state has the electronic configurations (7ur*)s , where S stands for singlet coupling. It is easy to see that the state can be described (using localized orbitals) as a resonance between the two ionic states C+C~ and C~C+. The a-electrons will react dynamically to this polarization. One can actually show explicitly that if the o-system is
321 replaced by a polarizable medium, the effective Hamiltonian will have reduced values of the electron repulsion integrals. The one-center Coulomb integral is reduced from 16.2 to 11.5 eV, which is exactly what Pariser predicted [10]. It would take a long time before modern ab initio quantum chemistry was able to challenge the results obtained with the PPP method. However, chemistry is not only organic. What about the excited states of inorganic compounds. Actually, there was much less interest in this area with two exceptions, small molecules (mainly diatomics) and transition metal complexes. The chemical bond between transition metal ions and their ligands could not be explained using the classical valence picture. A different theory was suggested by Becquerel [11] and formulated more exactly by Bethe [12]. It was acronymed Crystal Field Theory. The applications in chemistry were initiated by the work of Van Vleck [13]. The metal ion was in this method assumed to be perturbed by an electric field from the surrounding ligands. The spectroscopy therefore was atomic in nature. Transition metal complexes often have high symmetry and much of the success of the theory was based on the possibility to use group theory for the analysis of the excited states. For octahedral complexes it was, for example, enough to introduce one single empirical parameter (lODq) to explain the perturbation on the d-type atomic orbitals by the ligands. The crystal field theory was, however, only moderately successful. It was clear that many experimental facts could only be explained if one assumed a delocalization of the open shell electrons onto the ligands. From the crystal field theory emerged Ligand Field Theory. The ligand orbitals directly interacting with the metal ion d-orbitals were introduced and a set of MOs were constructed as linear combinations of the d-orbitals and ligand orbitals. The open shell electrons are then delocalized onto the ligands and a number of spectroscopic properties could thereby be explained. Ligand field theory shares many of the features of the PPP method of organic chemistry both in its formulation and the semi-empirical parameterization. The two approaches have even been combined into one semi-empirical theory, which was used with some success to explain the spectra features of metal porphyrins and similar metalorganic systems [14]. For a detailed treatment of crystal and ligand field theory we refer to the classical book by Carl Ballhausen [15]. This was the general situation at the middle of the 60ies, which is the time when modern ab initio quantum chemistry started its development. We shall therefore leave the history here and continue with a description of the different methods to deal with excited states in molecules and complexes that have emerged during the last 30 years. 3. WAVE FUNCTIONS FOR EXCITED STATES The semi-empirical methods developed for excited state calculations were moderately successful in describing certain features of the electronic spectra. But they had severe deficiencies. The PPP method was, for example, only able to describe %—>%* excitations within the valence shell. It was not clear how to treat the Rydberg transition or n—>p* transitions that occur from a lone-pair orbitals («) in systems with hetero atoms. Today we know that such excitations form an important part of the electronic spectrum of conjugated molecules. Sometimes we also see strong interaction between valence and Rydberg excited states. A typical example is the V state in ethene, which was mentioned above. In transition metal chemistry we know that important excitations are of the charge-transfer type, which can hardly be treated using ligand-field theory. There was clearly a need for an unbiased nonempirical theory, which does not a priori make any assumption concerning the nature of the excited states. This was the aim, when the development of the ab initio methods started in the
322 middle of the 60ies. The first efforts were naturally concentrated to the ground states and the development for excited states came a little bit later. But let us start with the general formulation of modern quantum chemistry. 3.1. Orbitals and wave functions The quantum chemistry for the excited state is always formulated using molecular orbital theory. The basic entity is a set of one-electron functions, the molecular orbitals (MOs). They are multiplied with a spin function to yield a set of spin-orbitals (SOs). Usually, the MOs are expanded in a basis set consisting of functions centered on the different atoms in the molecule, the atomic orbitals (AOs). If we select« such functions, /p(p= 1, «), we can form n orthonormal MOs as linear combinations of the AOs (the LCAO method):
6- = y C- v
(4)
From the n MOs we can generate In SOs. This is the one-electron basis we use to build the wave function. Our results will depend strongly on the quality of the MOs and features not present here cannot be compensated anywhere else. A proper choice of the AO basis set is therefore crucial for the accuracy of the results. This is particularly important for excited states. Once the MO basis has been chosen we need to build a basis for the /V-electron wave function. We can generate In N Slater determinants, d^, by occupying the 2« SOs with N electrons in all possible ways. The total wave function can be expanded in these Slater determinants:
^ = EC /; O /(
(5)
The expansion coefficients can be determined using the variational principle, which leads to the well known secular problem:
J] Hfn,-ES/lr)Cr=0
(6)
When the AO basis set becomes complete, this equation becomes equivalent to the Schrodinger equation. In practice this is of course not possible, but we can (in principle) approach closer and closer to the exact solution by increasing the size of the AO basis set. This is the key feature of ab initio quantum chemistry in contrast to other approaches, like semi-empirical schemes, density functional theory, etc. With a finite basis set the, method is called Full Cl (FCI). It is the best (in the context of the variational principle) approximate solution to the Schrodinger equation that we can obtain with the chosen basis set. The problem is only that we cannot solve equation 6. The dimension is simply too large for anything but very limited basis sets and few electrons.
323
Therefore, a second approximation is needed. There are many of them. Actually, all of modern quantum chemistry (with the exception of density functional theory) attempts to approximate the FCI equation in one way or another. Many of the approximations concentrate on the ground state and treat excited states as a secondary problem. The simplest of these methods is obviously the Hartree-Fock method, which only keeps one determinant in the expansion 5. A more advanced approach is coupled cluster theory. The most important of the methods that treat ground and excited states in a balanced way, is the complete active space SCF method and the multi-reference CI methods. All these methods try to select the most important terms in the FCI expansion, using knowledge about the detailed structure of the electron-electron interaction in molecular systems. We shall take a look at some of the methods. There are essentially two different approaches used in excited state calculations. The first, attempts to determine wave functions and total energies for each state of interest. To this category belongs the configuration interaction based methods, which directly tries to approximate equation 6. The second approach starts with the ground state and considers the excitation process as a perturbation of the ground state wave function due to the interaction with the electromagnetic field. These are the linear response methods. Here we compute directly the excitation energies and the transition moments, but do not have access to a wave function for the excited states. To this category belongs time dependent density functional theory and coupled cluster linear response methods. Let us first take a look at the wave function based methods. 3.2. Hartree-Fock and configuration interaction The solution of the FCI equation 6 is independent of the form of the MOs because the FCI wave function is invariant to unitary transformations among them. It is no longer the case when we approximate the equation. We are then faced with two problems, to determine the most optimal form of both the MOs and the CI expansion coefficients. So, what MOs should we use? The most immediate answer to the question is to use the HF orbitals for the ground state (we shall assume that the reader is familiar with the HF method). The eigenfunctions of the Fock operator can be divided into one occupied and one empty (virtual) subspace of MOs. The occupied orbitals will, for a closed shell molecule, contain two electrons each (cf. Fig. 3.)
Ground State
Singly Excited States
Doubly Excited State
Fig.3. Occupied and empty orbitals in the HF model for excited states.
324
The HF method is quite successful in describing the electronic structure and energetics of many molecules. The reason is that the mean field approximation of the electron-electron interaction is able to recover almost all of the repulsion energy. The error in the total energy is in well behaved cases less than 0.5%. This is not always the case though. We shall later discuss cases, where it is not possible to start from a HF ground state, but here we shall assume that the HF wave function gives a good representation of the ground state electronic structure. The error in the HF energy compared to the energy of the exact electronic wave function comes from the mean field approximation for the electron-electron repulsion. Each electron will in HF theory experience the average electrostatic field from all the other electrons. The error is called correlation energy because it arises from the inability of the HF method to describe the correlated motion of the electrons, in particular when they come close together. The Pauli exclusion principle tells us that two electrons of the same spin will not occupy the same position in space. This Fermi correlation is included in HF theory through the antisymmetry requirement on the wave function. Thus, the major part of the correlation error comes from electrons of opposite spin. Only two electrons can simultaneously be close together. Thus, correlation can with a good approximation be described using pair-theories. With such a ground state, how do we describe excited states? It is rather obvious: by moving electrons from the occupied to the virtual orbitals. This will generate a set of electronic configurations, which we can use to describe the excited states (cf. Fig. 3). Will the configurations in themselves give a good representation of the electronic structure? Usually not. The reason is that the density of states is often large in the excited manifold and one can expect strong interaction between different configurations (the + and -combinations that occur in alternant hydrocarbons is a good example). So, we have to write the wave function as a linear configuration of the configurations (or more properly configuration state functions (CSFs) because we have to take linear combinations of Slater determinants in order to obtain functions with the correct spin symmetry, singlets or triplets). Our excited state wave function thus has the form:
^ = C 0 O 0 + X Ciaia + Z Cia,jbOiaJb
+....
(7)
where ®o is the HF wave functions, ia are singly excited CSFs and ®iajb are doubly excited CSFs, etc. Where do we stop the expansion? If we do not stop it at all, we shall recover the FCI wave function. We can expect that the singly excited CSFs will be most important, but we cannot exclude that in some cases also doubly excited CSFs will give important contributions. Actually, it is more common that one might expect that excited states dominated by double excited CSFs appear at low energies in the electronic spectrum. But if we believe that this is not the case, can we terminate the expansion after the singly excited CSFs? Such an approximation is today labeled CIS (CI singles). Let us consider an example, the V state of the ethene molecule. The dominating CSFs are presented in Table 1. The numbers are from an extensive multi-reference CI calculation [16]. All orbitals in the table are of cr-type, except those labeled n. The ground state configuration is (ITT)2. The excitations that involve a-orbitals describe dynamic polarization, which is the most important correlation effect for the V state in ethene (see the discussion in the previous section). We see in the table that the CSFs are both singly and triply excited with respect to the HF ground state, but they are, as expected, double excitations with respect to the main configuration of the V state. We conclude that CIS does not give a balanced treatment of the
325
dynamic correlation effects in the excited state [17]. Ideally, we ought to include up to triple excitations in the expansion, but such an expansion becomes quite long and difficult to handle computationally. There is another solution to the problem: instead of starting from a single HF configuration we add a second configuration (ITI*) 2 : ct>0 = a1(l^") + a2i}7r)
(8)
and determine the two expansion coefficients aj and 02 variationally. We then include in the CI expansion for the excited states singly and doubly excited states with respect to both the starting configurations. These are the multi-reference methods. Table 1 The most important CI coefficients for the V state in the ethene moleculea. CSF
Coefficient
lTT^lTI*
0.934
3a g —> 3biu
-0.101
S(D)
( l 7 i ) 2 3 a g ^ (171* ) 3b,u
0.071
T(D)
Ib 3 a ^2b 3 U
-0.064
S(D)
(17I) 2 ^(27C)(2TI*)
0.064
D(D)
Ib3g^3b3u
-0.063
S(D)
2b|U^4ag
-0.052
S(D)
(l7i) 2 3a g ^(27i*) 2 3b lu
-0.043
T(D)
2
Type
a
FromRef. [16].
b
S = single, D = double, T = triple. Within (): with respect to the V state.
But there is yet another way to view the table. Suppose we use the V state configuration (7i7i*)s , instead of the ground state, as the reference in the expansion 7. The most important CSFs will then be the double excitations. It is well known that the most important contributions to the correlation energy comes from the configurations, which are doubly excited with respect to a given reference state. So, it seems that the most natural way to approximate the FCI equation for a given excited state is to first find a good reference function, which describes well the qualitative features of the wave function and then keep in the CI expansion the reference function plus all doubly excited CSFs. For open shell systems, one also needs to include singly excited CSFs. Exactly the same result would be obtained if we tried to estimate the importance of the excited CSFs using perturbation theory. Only the singly and doubly excited CSFs will interact directly with the reference function and therefore appear in the low order corrections. The analysis points to a theory that first computes
326 reference wave functions for each excited state and then adds correlation using singly and doubly excited CSFs. This is the multiconfigurational approach. 3.3. Active orbitals and the CASSCF method The two configuration wave function given in Eq. (8) describes the ground state of ethene better than HF theory. Actually, the coefficients of the second configuration is as large as -0.17. The reason is the rather weak 71-bond, which leads to a large occupations of the antibonding orbital. It is clear that when such occupations become too large, the HF model will break down and we cannot describe the electronic structure with a single configuration. Such situations are obtained when bonds are weak or even broken (dissociation, transition states) and quite frequently in excited states. We need then to go beyond the single configurational model to describe the excited state. This type of electron correlation is called static. It is long range in contrast to the dynamic correlation, which describes what happens when two electrons come close together. The idea behind the Complete Active Space (CAS) model [18] is to find a wave function that is based only on the orbital concept (as is HF), but which can handle also situations where a single configuration is not sufficient to describe the electronic structure. This is achieved by partitioning the MOs into three subspaces: inactive, active, and virtual. The inactive orbitals are assumed to be HF like and thus doubly occupied. To them we add a number of active orbitals but we do not assume anything about their occupation. Instead we construct a full CI wave function in this orbital space. The number of active electrons is the total number minus the number occupying the inactive orbitals. The wave function is thus completely determined by the choice of active orbitals and the condition that it should be an eigenfunction of spin and have a given space symmetry (in the molecular point group): *V
=^C O
(9)
where the sum goes over all CSFs O^ that can be generated by occupying the active orbitals in all possible ways consistent with an overall spin and space symmetry. The coefficients of the MOs and the CI expansion coefficients are obtained using the variational principle. In the ethene case we would choose as active the n and TI* orbitals with two active electrons, which gives the wave function 8 for the ground state. The same active space can also be used to describe the V state and the corresponding triplet. If we start to rotate the molecule around the double bond we would find that also the, so-called, Z state, the singlet state orthogonal to 8 becomes interesting (see below). However, the choice of active orbital becomes less straightforward when we study excited states. We need to assume something about the nature of the excited states and the orbitals which are important for a qualitatively correct description of the wave functions. Take ethene as an example again. The problem to compute an accurate wave function and energy for the V state is related to a near degeneracy between the valence excited state and an excitation to a 3d Rydberg orbital. There is a qualitative difference between Rydberg and valence excited states. A Rydberg state has one electron in a very diffuse orbital which interacts only weakly with the other electrons. Thus, the correlation energy decreases and resembles more that of the positive ion. For the ionic V state, the correlation energy is, on the other hand, increased because of the strong dynamic polarization effects, which are more prominent here than they are in the ground state where the electronic structure has a more covalent character. In order
327
to balance the correlation effects, we therefore need very accurate wave functions where most of the correlation energy is included. In order to be able to describe both valence and Rydberg excited states we need to extend the active space with Rydberg orbitals. The first step is to extend the AO basis set with enough diffuse functions. How it is done will not be described in detail here. Instead we refer to a review that [19], where the choice of active orbitals for excited states of organic molecules are discussed in detail. Ideally, one would like to be able to solve Eq. (9) separately for each electronic state. It is in general not possible. Different electronic states can be close in energy and the corresponding energy surfaces may even cross. Separate optimization of two states, which are close in energy and of the same symmetry, could also yield wave functions, which are non-orthogonal. This is undesirable. The solution is state average optimization. Several roots are extracted from the secular equation. The average density matrices are constructed and a common set of orbitals is optimized. Such wave functions are per construction orthogonal to each other. The use of a common set of orbitals is normally not a problem. The wave functions will still be good reference functions, which describe the qualitative features of the electronic structure well. To summarize: the CASSCF method will determine a qualitatively correct wave function for each electronic state. This function can now be used as a reference function in the CI expansion 7. So, the next step would then be to solve this equation to obtain the final accurate description of the electronic states. 3.4. Dynamic correlation In the previous section we sketched a two step procedure for the calculation of wave functions for excited states. The first step consists of determining an optimal set of MOs together with a wave function, which describes the electronic structure of the excited state qualitatively correct. We saw that it could be achieved by extending the HF scheme to a full CI in a limited set of active orbitals, the CASSCF method. It should be noted that the method can, with a properly chosen active space, be used not only in the Frank-Condon region around the equilibrium geometry of the molecule, but also to map out full energy surfaces, describing chemical transformations, conical intersections and other curve crossings, etc. The HF method cannot be used in such situations. The second step consists of solving a large CI problem where the expansion comprises all CSFs that can be generated by single and double excitations in each of the reference configurations. We call this approach multi-reference CI (MRCI). It became feasible for large scale applications with the development of the direct CI methodology in the early 70ies [20,21]. It is probably the most accurate method available for general studies of excited states. Today it is possible to use more than 108 CSFs in the CI expansion. The most efficient implementation is probably the COLUMBUS code developed by H. Lischka and co-workers [22]. Even if modern MRCI programs can handle long CI expansions, it is easy to see that with a CASSCF reference function, the number of CSFs will easily reach 108, when the number of correlated electrons and the basis set increase. The calculations will then become quite timeconsuming and rather soon impossible. One way to reduce them is to use only a subset of the CAS configurations in the reference function, thereby introducing one more ambiguity. Another possibility, which has been used with some success, is to use the entire CASSCF wave function as the reference instead of the individual CSFs. This is the internally contracted form of the MRCI method [23,24].
328 The MRCI method has a deficiency: it is not a size-extensive method. To explain what this means, let us consider two non-interacting systems A and B with wave functions *PA and ^PB • The wave function for the total system is ^FAB = ^ A ^ B • Suppose now that the wave functions for the two sub-systems are of the MRCI type with singly and doubly excited CSFs from a given reference function (MR-SDCI). The total wave function VAB will then include up to quadruple excitations with respect to the product of the two reference functions. An MRSDCI calculation will not describe the total system with the same accuracy as the subsystems and the energy will not be the sum of the energies for A and B. It is not easy to correct for this deficiency of the MR-SDCI method. Several methods have been developed, which account approximately for the missing higher excitations, which are products of the double excitations, but none of them are completely satisfactory. The only completely efficient way out is to include the product terms in the formal expression for the wave function. This leads to the coupled cluster expansion, which however has so far only been developed into an effective computational method for a single determinant reference function. 3.5. Second order perturbation theory Another way to treat the dynamic correlation effects is to use perturbation theory. Such an approach has the virtue of being size-extensive and ought to be computationally more efficient than the MRCI approach. Moller-Plesset second order perturbation theory (MP2) has been used for a long time to treat electron correlation for ground states, where the reference function is a single determinant. It is known to give accurate results for structural and other properties of closed shell molecules. The approach was in the late 80ies extended to reference functions of the CAS type, the CASPT2 method [25,26]. The approach turned out to be very productive especially in molecular spectroscopy. A large number of applications has been made for organic molecules, transition metal complexes, and more recently also in heavy element chemistry [19,27,28]. A number of examples will be given in other chapters of this book. A CASPT2 calculation starts with a CASSCF calculation, which has been designed to include the electronic states of interest. Each of the CAS wave functions are then used as reference functions for a CASPT2 calculation. The first order wave function is generated as a sum over all single and double excitations with respect to the entire reference function. In this respect CASPT2 is similar to internally contracted CI, but the expansion coefficients, CM are only determined to first order and the energy to second order. They are determined by the equation: =-VOfl
(10)
Here, H{^ are matrix elements of a zeroth order Hamiltonian, which is chosen as a oneelectron operator in the spirit of MP2. Sf,v is an overlap matrix: the excited functions are not in general orthogonal to each other. Finally, VOu represents the interaction between the excited function and the CAS reference function. The difference between Eq. (10) and ordinary MP2 is the more complicated structure of the matrix elements of the zeroth order Hamiltonian. In ordinary MP2 it is a simple sum of orbital energies. Here it is a complex expression involving matrix elements of a generalized Fock operator combined with up to fourth order density matrices of the CAS wave function. We do not give further details here but refer to the original papers If one assumes the generalized Fock matrix to be diagonal, one can formally write the second order energy in the same form as for MP2:
329
(11) where K runs over a set of orthogonalized excited states and s are sums of generalized orbital energies. In ordinary MP2, the denominators are simply: £a+£b-s~£/, where a, b are virtual and i, j occupied orbitals. The CASPT2 denominators are more complex, but describes in principle the same energy difference. The CASPT2 approach has several advantages. It is computationally effective, as the many large scale applications have shown. It can use arbitrarily complex reference functions with maybe up to a million CSFs in the reference function. It is size extensive (a small deviation from size extensivity can occur with some choices of the active space, but the deviations are negligibly small). The generality of the reference functions makes it possible to treat, not only the Frank-Condon region but to follow excited states energy surfaces over energy barriers, through conical intersections, to dissociation of a chemical bond, etc. Other articles in the book will describe such situations in more detail. The accuracy of the excitation energies is high with errors generally smaller than about 0.2 eV. There are, however, some drawbacks with the method, which can sometimes cause problems. One is the limitation of the active space. It is today difficult to perform calculations with more than about one million CSFs in the reference function. This can in some cases make accurate calculations impossible because of requirements on the size of the active space that yields larger sizes of the CAS CI space. Another problem is caused by the, so called, intruder states. The denominator in ordinary MP2 will rarely become small because there is usually a large HOMO-LUMO gap in the orbital energies. It is not always the case in CASPT2 because the active orbitals have energies that are intermediate to the energies of the doubly occupied and virtual orbitals. A low order perturbation approach will become invalid when the denominators become small in comparison to the numerators in Eq. (11). The cause of such behavior is usually a too small active space, but it is not always possible to extend the space further. A level shifting technique has been developed that will remove the intruder state [29,30], but it should only be used for weak intruder states that will have only a minor effect on the computed correlation energy. The CASPT2 method also has a small systematic error, which is due to the definition of the zeroth order Hamiltonian. States with a closed shell character are favored in relation to open shell structures. This leads to too small excitation and dissociation energies in some cases. A modification of the zeroth order Hamiltonian that reduces the error considerably has recently been suggested [31]. The CASPT2 method relies on a well defined CAS reference function, which will not be strongly affected by the addition of dynamic correlation. Normally, this model is satisfactory, but it may happen in more complex cases that several states of the same symmetry appear close in energy. They may then interact strongly with large changes occurring in the reference functions. Typical situations are areas of avoided crossings, conical intersections, etc. A multi-state version of CASPT2 has been developed to handles such situations. This is multistate (MS) CASPT2 [32]. Here an effective Hamiltonian is constructed for a set of CAS reference functions that has the normal CASPT2 energies in the diagonal and includes interaction between the first order wave functions in the off-diagonal elements. A detailed
330
0.0
10.0
20.0
30.0
40.0
50.0
60.0
70.0
80.0
90.0
HCH Rotation angle Fig.4. Weights of the CSFs (n) (upper line) and (71*) for the N-state of ethene as a function of the rotation angle.
discussion of the possibilities and limitations of the MS-CASPT2 method is given by Merchan and Serrano in another chapter of this book [33]. 3.6. The ethene molecule As an example of the importance of a multiconfigurational treatment in photochemistry we consider a very simple but illustrative and important example: the rotation of one CH2 group around the double bond in the ethene molecule, C2H4. The calculations on which the results are based have been made with a small DZP basis set. Also, the only parameter that has been varied is the rotation angle. All other geometry parameters have been kept at the values of the equilibrium geometry, which is a serious approximation, in particular for the CC bond. We should therefore not pay much attention to the actual numbers, but merely to the qualitative features. We construct CASSCF wave functions for all states, which can be formed by occupying the two 71-orbitals, n and n* in all possible ways. This gives rise to four electronic states: The N-state, dominated by the CSF (nf The V-state, a singlet state with the configuration: (TC)(TT*) The T-state, a triplet state with the configuration: (TT)(7T*) The Z-state, dominated by the CSF (n*f
331
N-state Z-state V-state T-state
-1.00
S
1 -1.20
-1.40
0.0
10.0
20.0
30.0
40.0
50.0
60.0
70.0
80.0
90.0
Rotation angle
Fig. 5. The total energy of the N-, T-, V-, and Z-states as a function of the rotation angle
The two CSFs (TI)2 and (TT*)2 belong to the same symmetry and can thus mix, so both states will have some multiconfigurational character. The question is how it varies with the rotational angle. We show in Fig. 4 how the weights of the two CSFs vary with the rotation angle for the Nstate. The corresponding picture for the V-state is the mirror image in this simple model. For planar ethene there is not much mixing. The 7i*-orbital is high in energy, so the N-state will be quite HF like with one dominating CSF. But, when we start rotating one of the HCH groups around the double bond the energy drops and finally, at 90°, the two orbitals it and TI* become degenerate. The n bond is now completely dissociated. Thus, the two CSFs have the same energy and there will be complete mixing, which is typical of a dissociation process. The variation of the (CASPT2) energies with the angle is shown in Fig. 5. The large spread in energies at zero angle is decreased substantially at 90° and the four states become pairwise degenerate. The lower pair correspond to a singlet (S) and triplet (T) coupled biradical system. The upper pair of states are ionic with the two electrons resonating between the two centers. The description is oversimplified here. In reality, Rydberg like states in the same energy region interact strongly with the upper states and change the nature of the wave functions in the FC region. The two degenerate ionic states are split by a Jahn-Teller distortion. A pyramidalization of one HCH group will stabilize one of the states further. At the same time, the ground state is further destabilized and eventually a conical intersection is reached, which makes it possible to move from this surface back to the ground state, (a more detailed discussion of the subject can be found in Ref. [34]).
332
One might think that the above example is rather special and of little interest in photochemistry. On the contrary. This simple model is the key for the understanding of a large number of important photochemical processes in chemistry and biochemistry. Rotation around a double bond is a much used process in nature. The most striking example is probably the cis-trans isomerization of the retinal molecule in rhodopsin, the key molecular process in vision. A number of other examples could be given, for which we refer to other chapters in this book. We emphasize that these typical problems in photochemistry can only be dealt with using a multiconfigurational quantum chemical methods like CASSCF/CASPT2 or CASSCF/MRC1. With this example we finish the overview of the state specific methods. We shall next briefly discuss the linear response methods, where the focus is on the ground state wave function and the excitations are obtained through the response of the system to an electromagnetic field. 4. LINEAR RESPONSE THEORY An alternative way to treat excitation processes is to study the response of the ground state electronic structure perturbed by an electromagnetic field. The quality of such a method depends crucially on the quality of the ground state wave function. The first attempts along this line was based on Hartree-Fock theory and was named the Random Phase Approximation (RPA) [35]. The method is not much used today. Recent developments use more accurate wave functions and the two most popular approaches are based on Density Functional Theory (DFT) or the Coupled Cluster (CC) method. Because both these approaches assume the ground state to be single configurational HF like, the main area of application will be studies of electronic spectra in the FC region. It is not advisable to use LR methods far away from equilibrium where the ground state may be strongly multiconfigurational and doubly excited states are low in energy. The second limitation inherent in the methodology is that only excited states, which are dominated by singly excited configurations are included. LR theory cannot be use to study rotation around a double bond as in the ethene example discussed above. LR theory starts with the ground state wave function. A perturbation in the form of a time dependent electric field is applied: E(/) = E^cosiat)
(12)
The first order response function is then computed and the excitation energies are found as poles of this function while the transition moments are obtained from the residues. We shall not attempt to give a full account of the theory here but refer to the original papers. The method was first developed in CC theory by Monkhorst [36] and was extended into an effective computational tool by the Aarhus school (see Ref. [37] and further references therein). The CC-LR method is capable of giving a very accurate account of the excited states that are dominated by single excitations. It is available today in an integral direct implementation, which makes calculations with many basis functions possible. For HF like ground states and excitations dominated by singly excited CSFs it is probably the most accurate method available today. Due do its limitations it is, however, not useful for studies of photochemical processes. Linear response is in DFT a simple extension of the RPA approach. For a full account of LR in DFT we refer to the review by M. Casida where more references may be found [38]. The approach is available in most programs and is very simple to use. It gives a quick overview of
333
all singly excited states, provided that an adequate basis set is used. One needs, however, to remember the limitations. We shall illustrate the performance of both LR-DFT and LRCC in the next section. A detailed and critical evaluation of the approach is given in the chapter on Ab Initio methods by Merchan and Serrano-Andres [33]. 5. THE BENZENE MOLECULE Let us now for a while leave the theory and relax with an example. We choose the organic molecule, which has through the years been the prototype for testing methods for excited states, benzene. At the same time it will be a recapitulation of the history. In our historical review, we discussed the calculations of Goeppert-Mayer and Sklar in 1938 and the use of benzene for the testing of the PPP method in the 50ies. It is not surprising that these early successes would be a challenge for ab initio quantum chemistry when it became available for studies of excited states in the late 60ies. The first attempts were made using small CI expansions, involving only the Tt-orbitals. The first calculation was performed by R. Buenker et al. [39]. It was a (by today's standard) a small CI calculation involving only the six 7i-orbitals. The results are shown in the first column of Fig. 4 (BWP). The results were not very impressive for obvious reasons. Actually, the calculations resemble the PPP calculation by Pariser from 1952, but now without any empirical correction of the integrals. Thus there is no account for the dynamic polarization of the o-electrons, which is reflected in the errors, which are especially large for the more ionic states 'Biu and ' E i u . Some improvement was obtained in the calculation by Peyerimhoff and Buenker (PB in Fig. 7) by extending the basis set and increasing the number of 7i-orbitals from six to nine [40], but the errors are still large. Hay and Shavitt (HS) performed a calculation including 23 7t-orbitals in 1974 [41], which we might consider as limit of what can be obtained without explicit inclusion of the a-electrons. The errors are still larger than one eV for some of the states. One should remember that the largest error in the PPP calculation of Pariser was 0.5 eV for the singlet states and 0.9 eV for the triplets and later developments improved the results further.
Fig. 6. The benzene molecule.
334
eV 11 -,
10
8
•
BWP
PB
HS
R
MRM
Expt.
Fig. 7. Early CI results for the electronic spectrum of benzene. For the references, see the text. So, it was a hard task to improve on the semi-empirical methods. The first CI calculation that included the a-electrons was performed in 1976 by Rancurel et al. [42] in a minimal AO basis set (R in Fig. 7). The results did not improve, which is due to the too small basis set used, which did not allow for any orbital relaxation. A much larger CAS-CI calculation was performed by Matos et al. in 1987 [43]. It also included corrections for the size-consistency error. It was the largest Cl calculation performed on the benzene spectrum before 1990. As can be seen in Fig. 7, the errors in computed excitation energies are still sizable (more than 0,5 eV in some cases), but the quality is now at least comparable to that obtained with the PPP method 30 years earlier. The more recent developments of linear-response theory (both in DFT and CC theory) and the CASSCF/CASPT2 approaches have of course been tested on benzene. The first
335 CASSCF/CASPT2 calculation was performed in 1992 [44] and a more complete treatment of the vertical spectrum (also including the Rydberg states) was made in 1995 [45]. The results of these calculations are shown in column 2 of Fig. 5 (labeled CASPT2). There is now good agreement with experiments with errors of only 0.1-0.2 eV. One notices that most CASPT2 energies are slightly on the low side, reflecting the systematic error of the method. The CASPT2 results are in Fig. 8 surrounded by two results obtained using linear response theory. The first column shows the results of a TD-DFT calculation [46]. Again, we see improvement compared to the older CI results, but errors are as large as 0.5 eV for some states. Nevertheless, this very cheap method is able to recover most of the dynamic
eV
9
8
; 2g
1
B1u
6-
•3B.
•1B. 3
E<
'lu
DFT
CASPT2
CC3
Expt.
Fig. 8. More resent results for the benzene molecule. The references are given in the text.
336 correlation effects on the excitation energies. It is to an even larger extent true for the accurate coupled-cluster linear response studies (CC-LR) of Christiansen et al. [37] for the singlet states and Hald et al. [47] for the triplet states. The method gives energies that are slightly larger than the experimental results. Both TD-DFT and LR-CC rely on a Hartree-Fock like reference function and excited states dominated by single excitations. The first condition is reasonably well fulfilled for benzene. Among the excited states, one finds the largest contributions from doubly excited configurations for the 'E2g state. It is also the state that give the largest errors using the LR method, both in DFT and CC. So, today we have methods that allow an accurate description of the excited states of a molecule like benzene. It took ab initio quantum chemistry almost 30 years to achieve this goal and make quantitatively accurate calculations of excited states in larger molecules possible. The calculations presented above were all made at the ground state equilibrium geometry. The energies thus correspond to "vertical" excitations. This is not a measurable quantity. Instead, an electronic spectrum is resolved into a vibrational structure corresponding to excitations from a vibrational level in the ground state to a corresponding level in the excited state (very detailed spectra may also resolve the rotational fine-structure). The intensity distribution for allowed transitions is normally such that the most intense bands occur close to the vertical energy difference (the Frank-Condon region). But not exactly. And what happens for symmetry forbidden transitions? We can illustrate the problem for the lowest excited state in the benzene molecule, for which a calculation of the vibrationally resolved spectrum has been made using the CASSCF/CASPT2 method [48]. Such a calculation is more complex than a calculation at a single geometry. Now one has to compute not only energies, but also the equilibrium geometry of the ground and excited state, the force field (to obtain the vibrational frequencies) and the transition moment as a function of geometry (for the intensities). It was done for the two lowest bands in the benzene spectrum using the MULA program of the MOLCAS quantum chemistry software. This program solves the vibrational problem in the harmonic approximation and computes transition intensities for each vibrational mode [49]. The result for the !B2u band is shown in Fig. 9.
337 I.So-OS THEORETICAL
ee-oe
EXPERIMENTAL
I X
HiVrLCWOTH
Fig. 9. The theoretical and experimental spectrum for the 'B?u band in the benzene molecule.
The agreement between experiment and theory is spectacular. The intensity distribution among the vibronic bands are strongly dependent on the geometry difference between the ground and excited state. It is thus necessary to be able to predict the change in geometry with high precision. Here, the CASPT2 method was used, which for organic molecules normally predicts bond distances with an accuracy of a few thousands of an Angstrom. All intensity of the band comes from vibrations that break symmetry. The transition moment itself is zero in D6h symmetry. It was found that the vertical transition energy is about 0.1 eV larger than the energy of the most intense band, showing that relating vertical energies to band maxima is not a precise measure of accuracy. The example has shown that today we cannot only compute a set of excitation energies for a given geometry, but we can also compute the energy surfaces for the excited state. This
338 makes it possible to use theory to study photo chemical processes. A number of examples in other chapters of this book will show how this is done in practice. 6. HEAVY ELEMENT COMPOUNDS With heavy elements we shall here mean all atoms with an atomic number larger than 18 (Ar). Included among these atoms are the transition metals, the heavier main group elements, the lanthanides, and the actinides. Calculations of excited states for compounds containing such atoms are for several reasons more complex than for lighter atoms. The number of electronic states with low energy increases with increasing atomic number and the so do the orbitals involved. There is thus both a problem to choose an adequate active space and to know how many states one needs to include in the calculation. An additional problem is that one needs to account for relativistic effects. Relativistic quantum chemistry is an area where intense development is going on at the present time and many ideas have been suggested for treating relativistic effects ranging from four component theories based on the Dirac equation to simple addition of relativistic effects to a nonrelativistic method using perturbation theory. Most of these approaches has addressed mainly the ground electronic state. We shall not discuss all these methods in detail here but refer instead to a recent book where many approaches are discussed [50]. Instead we shall give some examples using an extension of the CASSCF/CASPT2 method to the relativistic regime. The approach is a compromise between the need to include relativistic effects and the requirement that one should be able to treat general electronic structures both in ground and excited states, the key feature of the multiconfigurational approach in quantum chemistry. In addition, the method should be applicable to large molecules. All these requirements rule out a four-component approach based on the Dirac equation. Instead a two step procedure has been introduced based on the Douglas-Kroll (DK) Hamiltonian [51] together with a meanfield approximation of the spin-orbit(SO) part of this Hamiltonian [52]. For a detailed discussion of this approach we refer to a recent paper where further references can be found [53]. The method can be characterized as a two step procedure: In a first step the scalar part of the DK Hamiltonian is introduced into a normal non-relativistic CASSCF/CASPT2 calculation. Specially designed AO basis sets are used, which account for the relativistic contraction of the inner shell orbitals. The method treats all electrons without the introduction of effective core-potentials, but deep core orbitals are described in a minimal basis set. Calculations are performed for a number of electronic states. The selection of the states depends on the specific problem and the importance of spin-orbit interaction. In a final step, the CASSCF wave functions for these electronic states are used as basis function in the construction of a matrix of the spin-orbit part of the DK Hamiltonian. The RASSCF state interaction method (RASSI) is used, which allows for the calculation of matrix elements of one-and two-electron operators in a basis of CASSCF wave functions, possibly using different (non-orthogonal) sets of orbitals [54]. The diagonal elements of the matrix are shifted to the CASPT2 energies in order to account for electron correlation. The eigenvalues of this matrix gives the final energies and wave functions including spin-orbit coupling. Can a method that treats relativistic effects in such an approximate way, really work? Actually it does. In connection with the development of a new set of relativistic ANO type basis sets, ANO-RCC, which also include correlation of the outermost core-electrons (which is necessary for heavy elements) we have studied the approach for all alkaline, alkaline earth, main group and noble gas atoms [55,56]. It turns out that errors in the spin-orbit splitting are small, with the possible exception of the heavier fifth row main group atoms (Bi-At). Ground
339 state potential curves for the dimer of Tl and Pb give spectroscopic constants in agreement with experiment [53]. The electronic spectrum of PbO could be successfully assigned with the present method [53]. Here we give an additional example from the third row transition metals. The energy levels of the platinum atom ANO-RCC basis sets for the transition metals are presently under development. In connection with the work we computed the electronic spectrum of the Pt atom. The ground state of Pt is (5df{6s), 3D, J=3. There is, however a large spread of the different J-level (J=l is located 10131 cm'1 above J=3), while the term (5df{6sf, 3F, has its lowest level, J=3, only 824 cm"' above the ground level. The term (5d)[0, 'S is found at 6140 cm"'. Is it at all possible to describe the spectrum theoretically? Let us see. We set up the following calculation: CASSCF/CASPT2 calculations are performed for all terms of the configurations (5
340
Table 2 CASSCF/CASPT2/RASSI results for the energy levels in the Pt atom (energies in cm"1). Level Theory Expt.a {5d)\6s\ 3D J=3 0 0 (5df(6s), 'D J=2
366
775
502
823
(5d)\6s), DJ=2
6105
6567
{5df, 'SJ=O
7278
6140
9932
10116
10204
10131
3
{5df(6sf, F J=4 3
3
(5df(6sf, F J=3 9
3
(5J) (fc), D J=l 3
13811
23
15494
15501
3
16853
16983
2 3
18522 22631 26647 48514
18567 21976 26638 -
(5df(6sf, P J=2 8
(5
fo) , P J=l (5df(6sf, 'G J=4 (5)8(fe)2, 'D J=2 (5aQ8(fo)2, 'SJ-0 Experimental data from Ref. [59].
The results are shown in Table 2. The agreement with experiment is impressive considering the many approximations involved. The largest error, 1138 cm"' (0.14 eV) is found for the (5d)10, 'S state. It is not surprising. It is well known that the different shape of the d-orbital and the strong correlation effects in the (5
341
3.5
3.0
2.5
1D 1G.
2.0 .3P
1.5
1.0 1D 0.5 3D
o.o L Theoretical
Experimental
Fig. 10. The electronic spectrum of the Pt atom computed without and with spin-orbit coupling.
6.1. Lanthanides and actinides The possibilities for lanthanides is not yet fully explored. Some promising results exist for small lanthanide molecules like SmO [60]. One problem is the size of the active space. At least the Af-, 6s-,and 6/?-orbitals needs to be included for the atom, which is already 11 orbitals. Adding ligand orbitals may easily extend the limits of the present implementations of the methodology. However, the chemistry of the lanthanides mainly involves metal in a high oxidation state. Then only the ^orbitals are needed. They are not involved in the bonding process and it would be interesting to develop a variant of the method, which allows inactive
342
open shell orbitals. It can easily be done at the CASSCF level, but is more complicated for CASPT2. But with the present programs one needs to include the whole 4f-shell in the active space. An additional problem is the number states that need to be treated. The spectrum is very dense and some experimentation is needed in order to find out how many states one needs to include for a given problem. A good example is the chemiionisation process studied for SmO as mentioned above. We shall certainly see some development of the multiconfigurational approach for lanthanide chemistry in the near future. More has been done in actinide chemistry. Experimental information is scarce and theory has a great opportunity to add to our knowledge about the chemistry of actinide compounds. Much of the early work in actinide chemistry dealt mostly with U(VI) compounds because of their closed shell structure. However, configuration interaction methods have also been used to study electronic spectra and other properties. We shall not discuss the earlier work here but refer instead to some recent reviews on the subject [50,61,62]. Instead, we shall illustrate the possibilities and problems that quantum chemistry faces in actinide chemistry by a recent example, the electronic spectrum of the UO2 molecule. For "normal" molecules one can usually determine the ground state electronic configuration using rather simple quantum chemical arguments based on standard theories of the chemical bond. This is not the case in actinide chemistry and UO2 is a striking example. The uranyl ion UO 2 + is maybe the most well studied species in uranium chemistry. It has a closed shell electronic structure with uranium, in the formal oxidation state VI, forming a triple bond to each of the oxygens. If we now add two electrons to form the neutral molecule, where do they go? Following the aufbau principle we would choose to place them in 5/orbitals on uranium with the resulting electronic state: (5f0)(5fo), JHg . However, is the aufbau principle valid here? The electronic configuration of the uranium atom is {5ff(6d)(7s)2, so we might suspect that there might be competition between different choices. As it turns out, the lowest configuration is (5f)(7s), 3u with the 3Hg state about 0.5 eV above [63]. But the story does not end here. Earlier studies of the vibrational spectrum in an Ne matrix had confirmed the electronic ground state to be 3u. But when these experiment were repeated in matrices of heavier rare gas atoms, a different spectrum was recorded [64] and guided by DFT and CCSD(T) calculations it was concluded that the ground state had changed to JHg. Measurements of the electronic spectrum in gas phase and in an Ar matrix gave, however, opposite evidence and suggested that the ground state is not changed but remains as 3 u also in the matrix [65,66]. The issue is yet unresolved, but it shows that in actinide chemistry we cannot take anything for granted. The density of states is so high that even weak interactions can change the electronic structure. It becomes quite complicated to study chemical transformations in such systems. We can illustrate the problem with a calculation of the electronic spectrum of the UO2 molecule that has recently been performed [67]. The calculations are of the same type as described for the Pt atom. In all 150 electronic levels were computed. The results are shown in Fig. 11. The two lowest levels with labels 2u and 3u come from the 3U state and the third level (at 0.20 eV) from the JHg state. As can be seen the spectrum is very dense. Actually, there are probably more lines in the spectrum at energies above about 2 eV. The calculations included in the active space (apart from orbitals from the UO bond) only the 5f, 7s and 7p orbitals because they give rise to allowed transitions from an ungerade ground state. However excitations to 6d orbitals on uranium will appear in the energy region above 2 eV. Electronic spectroscopy for actinide compounds is clearly not an easy task. If the electronic spectrum is so complex, what about dynamics and photochemistry? Here is a "simple" example to illustrate the problems one has to overcome if one wants to study the
343 TTieoretical{all)
eV
Theoretical(f>0.001)
Experimental Range
5.0
4.0
3.0
2.0
1.0
o.o-
a
Fig. 11. The electronic spectrum of the UO2 molecule. The first column shows all lines, the second those with oscillator strengths larger than 0.0001, and the third the measured spectrum. Grey lines are accessible from the 2u state and black lines from 3u.
dynamics of reactions involving heavy atoms. NUN is a stable molecule with a closed shell singlet ground state. It is produced by allowing uranium atoms to collide with nitrogen molecules in a molecular beam. The uranium atom has a Ls ground state (17-fold degenerate). A recent calculation shows that the transition state is one out of many close lying triplet states with a barrier that is completely determined by spin-orbit coupling. The reactions thus starts on a 17-fold degenerate surface, crosses over to a dense set of triplet states at the transition state region and finally settles on a singlet surface at equilibrium (for details, see Ref. [68]). Even with access to very accurate energy surfaces, how do we solve the dynamical problem?
344
7. THE FUTURE It is of course impossible to make any real predictions about the future developments of a research fields. Exploring the unknown cannot be predicted. The most we can do is to make some rather straightforward extrapolations from the present situation. As we have seen in this and other chapters of the book, the field is in quite a good shape. Many problems can be treated with reasonable accuracy and we can make sensible predictions in spectroscopy and in photochemistry. Quantum chemistry is today a valid and accurate tool for assigning electronic spectra of all sorts of molecules, ranging from organic systems and transition metal complexes to heavy element compounds involving main group elements, lanthanides and actinides. Photochemical reactions and excited state dynamics can be studied for organic systems even if we still have many things to learn here. The development is a little bit slower for transition metal complexes but progress is made also in this field. What are the major bottlenecks for extending the applications to larger and more complex systems? One is the standard problem of quantum chemistry: the integrals. Most methods used today for excited states still scale at least as n4 with the size of the basis set, n. Some progress is made in developing linear scaling models but the progress seems to be slow. Localization of orbitals offer another possibility for treating correlation only in an "active" part of a molecule. Other methods that will reduce the dependence on the size of the AO basis set are being developed. I have a feeling that a few years from now, this will no longer constitute a real problem. All the quantum chemical methods we are using today to treat excited states have inherent problems of one kind or another. The very cost-effective LR-DFT approach is not accurate enough and cannot treat all types of excited states in a balanced way, and it is restricted to a HF like ground state. It is possible that new functionals will become available in the future that will increase the applicability. However, another possibility seems to be more promising. One can formulate a density functional theory, which is not based on a single configurational model, but instead on a CAS wave function. Such a model would treat exchange exactly (A CAS model does not have the inherent problem of a delocalized exchange as HF has) and only use DFT to describe dynamic correlation. It would certainly open up new possibilities for excited state calculations. There will be no limitation on the electronic structure of the ground state, the electronic spin will be well defined for open shell systems, and all types of excited states can be treated. Many research groups have started research in this field and it is possible that we may have a working procedure soon (for a discussion of some aspects of the problem see for example Ref. [69]). In coupled cluster theory one would need a multiconfigurational analogue to the CC-LR methods available today. There are such developments but they are slow for molecules. The Fock space based methods can be very accurate for small systems like atoms, but are not general enough for photochemistry (See for example Ref. [70]). What about the CASSCF/CASPT2 approach. It is most likely going to continue to be useful tool for studies in photochemistry because of the general structure. In combination with methods that reduce the integral problem it should be able to treat very large systems. The method has, however, limitations that are not easy to overcome. The active space is limited to around 15 orbitals. This problem is solved if DFT is used instead of CASPT2 to treat dynamic correlation, because it is then possible to work with reduced CI expansions (RASSCF). Another problem with the CASPT2 method is its limited accuracy. In principle it may be increased by increasing the active space, but this is rarely an alternative in practical applications. There is hope that the error will be reduced in the near future by developments
345
of improved zeroth order Hamiltonians, which reduces the systematic error for open shell systems [31]. The development in heavy element chemistry is intense at present and we expect to see an increasing number of applications in this area in the future. The problem of the large active space needed for systems containing several heavy atoms (e.g. clusters of transition metals or actinides and other nanosystems) is still not solved and an accurate theoretical spectroscopy of such systems is not yet feasible. The same is true for many transition metal complexes where the metal is in a very high oxidation state. Generally, this area is full of challenging problems for theory, in particular, because it is needed to include relativistic effects, both scalar and spin-orbit coupling. But we are faced not only with the problem of computing accurate wave functions and energies. Most chemical processes (even in photochemistry) take place in an environment, a solvent, an active site in a protein, a crystal, etc. The methods we have today to treat environmental effects are rudimentary and need to be developed further. The problem of computing free energies instead of just enthalpies has not been solved in the general case. Will we in the future be able to do direct dynamics using accurate energies and forces, or will we instead see further development of global fitting procedures. We do not know today, but there are certainly many difficult problems to solve and exciting challenges to meet in order to develop the field into a true predictive tool in chemistry. Here is the ultimate challenge: develop theoretical tools that allow an accurate quantitative prediction of the entire process of transformation of solar energy into chemical energy in the photosystem of bacteria and plants.
346 ACKNOWLEDGMENTS This work has been supported by a grant from the Swedish Science Research Council, VR and the Swedish Foundation for Strategic Research (SSF). REFERENCES [I] E. Schrodinger. Ann.Phys., 79, 361, (1926). [2] D. R. Hartree. Proc. Cambridge Phil. Soc, 24, 89,111,426, (1928). [3] F. Hund. Z. Physik, 40, 742, (1927). [4] R. S. Mulliken. Phys. Rev., 32, 186, (1928). [5] E. Hiickel. Z. Physik, 60, 423, (1930). [6] M. Goeppert-Mayer and A. L. Sklar. J. Chem. Phys., 6, 645, (1938). [7] R. Parr. J. Chem. Phys., 20, 1499, (1952). [8] R. Pariser. J. Chem. Phys., 21, 568, (1953). [9] R. Pariser, R. Parr, and J. A. Pople. Int. J. Quantum Chem., 37, 319, (1990). [10] P.-A. Malmqvist and B. O. Roos. Theor. Chim. Acta, 83, 191-199, (1992). II1] J. Becquerel. Z. Physik, 58, 205, (1929). [12] H. Bethe. Ann.Phys., 3, 135, (1929). [13] J. H. Van Vleck. Theory of Magnetic and Electric Susceptibilities. Oxford University Press, Oxford, (1932). [14] B. O. Roos. Acta Chem. Scand., 20, 15, (1966). [15] C. J. Ballhausen. Ligand Field Theory. McGraw-Hill Book Company, Inc, New York, (1962). [16] R. Lindh and B. O. Roos. Int. J. Quantum Chem., 35, 813-825, (1989). [17] Comment: The CIS method has been introduced in some commercially available quantum chemistry softwares. I strongly recommend against the use of this approach in studies of excited states due to its unbalanced treatment of the effects of electron correlation. [18] B. O. Roos. The complete active space self-consistent field method and its applications in electronic structure calculations. In K. P. Lawley, editor, Advances in Chemical Physics; Ab Initio Methods in Quantum Chemistry II, chapter 69, page 399. John Wiley & Sons Ltd., Chichester, England, (1987). [19] B. O. Roos, M. P. Fulscher, P.-A Malmqvist, M. Merchan, and L. Serrano-Andres. Theoretical studies of electronic spectra of organic molecules. In S. R. Langhoff, editor, Quantum Mechanical Electronic Structure Calculations with Chemical Accuracy, pages 357— 438. Kluwer Academic Publishers, Understanding Chem. React., Dordrecht, The Netherlands, (1995).
[20] B. O. Roos. Chem. Phys. Letters, 15, 153, (1972). [21] P. E. M. Siegbahn. J. Chem. Phys., 72, 1647, (1980). [22] H. Lischka, R. Shepard, I. Shavitt, R. M. Pitzer, M. Dallos, T. Mller, P. G. Szalay, F. B. Brown, R. Ahlrichs, H. J. Bhm, A. Chang, D. C. Comeau, R. Gdanitz, H. Dachsel, C. Ehrhardt, M. Ernzerhof, P. Hchtl, S. Irle, G. Kedziora, T. Kovar, V. Parasuk, M. J. M. Pepper, P. Scharf, H. Schiffer, M. Schindler, M. Schler, M. Seth, E. A. Stahlberg, J.-G. Zhao, S. Yabushita, , and Z. Zhang. COLUMBUS, an ab initio electronic structure program, release 5.9, (2004). [23] H.-J. Werner and P. J. Knowles. J. Chem. Phys., 89, 5803, (1988). [24] P. J. Knowles and H.-J. Werner. Theor. Chim. Acta, 84, 95, (1992).
347
[25] K. Andersson, P.-A. Malmqvist, B. O. Roos, A. J. Sadlej, and K. Wolinski. i. Phys. Chem., 94, 5483-5488, (1990). [26] K. Andersson, P.- A. Malmqvist, and B. O. Roos. J. Chem. Phys., 96, 1218- 1226, (1992). [27]B.O. Roos, K. Andersson, M. P. Fulscher, P.-A. Malmqvist, L. Serrano-Andres, K. Pierloot, and M. Merchan. Multiconfigurational perturbation theory: Applications in electronic spectroscopy. In I. Prigogine and S. A. Rice, editors, Advances in Chemical Physics: New Methods in Computational Quantum Mechanics, Vol. XCIII: 219-331, pages 219-332. John Wiley & Sons, New York, (1996). [28] B. O. Roos. Ace. Chem. Res., 32, 137-144, (1999). [29] B. O. Roos and K. Andersson. Chem. Phys. Letters, 245, 215-223, (1995). [30] B. O. Roos, K. Andersson, M. P. Fulscher, L. Serrano-Andres, K. Pierloot, M. Merchan, and V. Molina. J. Mol. Struct. (Theochem), 388, 257-276, (1996). [31] G. Ghigo and B. O. Roos. Chem. Phys. Letters, submitted, (2004). [32] J. Finley, P.-A. Malmqvist, B. O. Roos, and L. Serrano-Andres. Chem. Phys. Letters, 288,299-306,(1998). [33] See Chapter 2 of this book. [34] V. Molina, M. Merchan, B. O. Roos, and P.-A. Malmqvist. Phys. Chem. Chem. Phys., 2, 2211-2217,(2000). [35] T. H. Dunning and v. McKoy. J. Chem. Phys., 47, 1735, (1967). [36] H. J. Monkhorst. Int. J. Quantum Chem., SI 1, 421, (1977). [37] O. Christiansen, H. Koch, A. Halkier, and P. Jorgensen. J. Chem. Phys., 105, 6921, (1996). [38] M. Casida. In D. P. Chong, editor, Recent Advances in Density Functional Theory. Part I. World Scientific, Singapore, (1995). [39] R. J. Buenker, J. L. Whitten, and J. D. Petke. J. Chem. Phys., 49, 2261, (1968). [40] S. D. Peyerimhoff and R. J. Buenker. Theor. Chim. Acta, 19, 1, (1970). [41] P. J. Hay and I. Shavitt. J. Chem. Phys., 60, 2865, (1974). [42] P. Rancurel, B. Huron, L. Praud, J. P. Malrieu, and G. Berthier. J. Mol. Spectrosc, 60, 259, (1976). [43] J. M. O. Matos, B. O. Roos, and P.-A Malmqvist. J. Chem. Phys., 86, 1458, (1987). [44] B. O. Roos, K. Andersson, and M. P. Fulscher. Chem. Phys. Letters, 192, 5-13, (1992). [45] J. Lorentzon, P.-A Malmqvist, M. P. Fulscher, and B. O. Roos. Theor. Chim. Acta, 91, 91-108,(1995). [46] D. J. Tozer, R. D. Amos, N. C. Handy, B. O. Roos, and L. Serrano-Andres. Mol. Phys., 97,859-868,(1999). [47] K. Hald, C. Hattig, and P. Jorgensen. J. Chem. Phys., 113, 7765, (2000). [48] A. Bernhardsson, N. Forsberg, P.-A. Malmqvist, B. O. Roos, and L. Serrano-Andres. J. Chem. Phys, 112, 2798-2809, (2000). [49] P.-A. Malmqvist andN. Forsberg. Chem. Phys., 228, 227-240, (1998). [50] In P. Schwerdtfeger, editor, Relativistic Electronic Structure Theory. Part I. Fundamentals. Elsevier, Amsterdam, (2002). [51] N. Douglas and N. M. Kroll. Ann.Phys, 82, 89, (1974). [52] B. A. Hess, CM. Marian, U. Wahlgren, and O. Gropen, Chem. Phys. Letters, 251, 365, (1996). [53] B. O. Roos and P.-A. Malmqvist. Adv. Quantum Chem., in press, (2004). [54] P.-A. Malmqvist, B. O. Roos, and B. Schimmelpfennig. Chem. Phys. Letters, 357, 230, (2002).
348 [55] B. O. Roos, V. Veryazov, and P.-O. Widmark. Theor. Chim. Acta, 111, 345, (2004). [56] B. O. Roos, R. Lindh, P.-A. Malmqvist, V. Veryazov, and P.-O. Widmark. J. Phys. Chem. A, 108,2851,(2004). [57] K. Andersson and B. O. Roos. Chem. Phys. Letters, 191, 507, (1992). [58] See Chapter 9 of this book. [59] J.E. Sansonetti, W.C. Martin, and S.L. Young. Handbook of Basic Atomic Spectroscopic Data (version 1.00). [Online] Available: http://physics.nist.gov/Handbook. National Institute of Standards and Technology, Gaithersburg, MD, (2003). [60] J. Paulovic, L. Gagliardi, J.M. Dyke, and K. Hirao. J. Chem. Phys., 120, 9998, (2004). [61] B. O. Roos, P.-A. Malmqvist, and L. Gagliardi. Heavy element quantum chemistry -the multiconfigurational approach. In (E. Brandas and E. Kryachko), editors, Fundamental World of Quantum Chemistry, page 425. Kluwer Acad. Publ., Dordrecht, Netherlands, (2003). [62] In B. Hess, editor, Relativisitc Effects in Heavy-Element Chemistry. Elsevier, Amsterdam, (2003). [63] L. Gagliardi, B. O. Roos, P.-A. Malmqvist, and J. M. Dyke. J. Phys. Chem. A, 105,
10602-10606, (2001). [64] J. Li, B. E. Bursten, and L. Andrews C. J. Marsden. J. Am. Chem. Soc, 126, 3424, (2004). [65] J. Han, V. Goncharov, L. A. Kaledin, A. V. Komissarov, and M. C. Heaven. J. Chem. Phys., 120,5155,(2004). [66] C. J. Lue, J. Jin, M. J. Ortiz, J. C. Rienstra-Kiracofe, and M. C. Heaven. J. Am. Chem. Soc, 126, 1812,(2004). [67] L. Gagliardi, M. C. Heaven, J. Wisborg Krogh, and B. O. Roos. J. Am. Chem. Soc, in press, (2004). [68] L. Gagliardi, G. La Manna, and B. O. Roos. Faraday Discuss., 124, 63, (2003). [69] S. Gusarov, P.-A. Malmqvist, R. Lindh, and B. O. Roos. Theor. Chim. Acta, 112, (2004). [70] A. Landau, E. Eliav, Y. Ishikawa, and U. Kaldor., J. Chem. Phys., 114, 2977, (2001)
349
Index 2,3-diazabicyclo[2.2.1]hept-2-ene(DBH) 16, 18,20 2,3-diazabicyclo[2.2.2]oct-2-ene (DBO) 20, 26 ASCF method 47, 279, 291, 293, 298 Absorption spectrum 76, 141-142, 144, 161,204,280,282,303 Actinides 338, 341, 344-345 Active electrons 37, 57, 281, 326 Active orbitals 14, 37, 45, 57, 193, 214, 215,231, 304,326-327,329 Adenine 116 Adiabatic approximation 97, 228 Adiabatic electronic - state 238 - wavefunction 140 Adiabatic transition(s) 66, 175, 178 Adiabatic excitation (energies) 76, 105, 111,204 Algebraic-Diagrammatic Construction (ADC) approach 49-50 Anisole 256-257 Antisymmetry (principle) 40, 324 Approximate propagator methods 49 Averaged Coupled-Pair Functional (ACPF) method 46 Avoided crossing(s) 4-6, 62-64, 82, 172, 195,206,264,268,329 Azoalkane 16, 18, 20 Azobenzene 112, 192, 203, 213 Azulene 2 Bacteriorhodopsin (bR) 7, 225, 237 Barrelene 263, 267, 272-274 Barrierless processes 204 Barrierless minimum energy path 210 Barrierless reaction path 208 Barrierless vs. barrier-controlled twisting path 217 Benzene 2, 20, 48, 80, 134, 141, 144, 148, 214-215,234,319,333-337 Benzenedithiolate (bdt) 281, 302-303 Benzobarrelene 273-274 Benzodiene 264 Bimolecular photochemical reactions 20 Biradicaloid electronic structure 5 Bi-(di-)radical 14, 17, 18, 26, 47, 208, 264, 267-268, 273-274, 331 Block iteration methods 103
Blue copper protein(s) 79, 279-280, 294295, 298-299, 309 Bottleneck 38, 104, 181, 195, 199, 238, 344 Butadiene 5, 68, 79 Cr(CO)6 279, 282 Carotenoids 116 Complete Active Space Self Consistent Field (CASSCF) method 37, 46, 57,82, 193, 198,206,213,231235, 326-327, 336, 338, 340 Complete Neglect of Differential Overlaps for Spectroscopy (CNDO/S) - Hamiltonian 150 - methods 298 Complete Active Space - Multireference Second Order Perturbation Theory (CASPT2) method 37, 46, 52, 5759,62,77,234,237,281,283, 287, 289, 328-329, 337, 339, 344 Charge transfer 7, 14, 20, 23-24, 116, 210, 228, 244, 280, 281, 295, 298, 300, 321 Charge transfer state(s) 79, 109, 208, 282, 291,295,297-298,300,306 Chlorophyll(s)109, 116 Cluster expansion method 291 Configuration Interaction (Cl) methods 44, 54, 323, 342 - full configuration interaction 36, 231,322 - multi-reference Cl (MRCI) 37-38, 42, 46, 54, 64, 74, 79, 83, 232, 327-328 - MRCI type with singly and doubly excited CSFs (MR-SDCI) 328 - singles and doubles configuration interaction (CISD) 37, 150 - singles configuration interaction (CIS) 37, 47-48, 51,81, 105, 109, 150 Configuration State Functions (CSFs) 44, 54, 56, 324-332 Conical intersection(s) 3-9, 12-16, 19-21, 24-25, 27, 63-65, 81-84, 111, 171177, 180-184, 187-188. 195-197, 199-201,204-211,213,216-218, 220, 226-227, 229, 230, 233, 238, 240-241, 247, 262, 264, 268, 270, 272-273,327,329,331. Coupled-Cluster (CC) methods 37-38, 4951, 53,79, 332, 336
350 - equation-of-Motion CC (EOM-CC) 50-53,78-79,228,291,293 - multireference coupled-cluster (MRCC) 52 Coupled-Electron Pair Approximation (CEPA) 37 Crossing point(s) 3, 5, 63, 198, 206, 209,
Ethene 68, 320-321, 324-326, 330-332 Ethylene(s) 6, 14, 171-172,233 Exact Exchange (EXX) functional 100, 109-110 Exchange correlation 41, 47, 95, 97-101, 103, 105, 109, 111,228, Exchange integral 46, 268
213,317 Crystal lattice 268 Cyanines 184, 186 Cyclobutane 14 Cyclobutanones 271 Cyclobutene 5, 14, 153-154 Cyclohexadienes 4 Cyclohexenone 274 Cyclooctatetraene (COT) 71-72, 74, 7678, 193,217-219 Cytosine 81-83, 116 Delta Density method 270-272 Density Functional Theory (DFT) 53, 67, 78, 93-128, 228, 279, 285, 289291,294,298,332,334,344 - Hartree-Fock-Slater versions of DFTHFS-SW291.298 - HFS-DVM 291 Derivative coupling 10, 29, 178 Di-7t-Methane Diarylethylene 179-181, 183 Dihydroazulene (DHA) 179, 182 Dipole allowed state(s) 161, 167 Dipole allowed transition(s) 74, 130, 144, 145, 147,280,291 Dipole forbidden state(s) 144, 161 Dipole forbidden transition(s) 72, 74, 130, 141-142, 144, 147, 150, 161 Direct-dynamics variational Multiconfiguration Gaussian (DDvMCG) method 179 Dynamic correlation 8, 45-46, 57, 62-64, 68-69, 76-78, 82-83, 283, 325-329, 344 Effective Hamiltonian 62-64, 83, 321, 329 Ehrenfest dynamics 177-178, 238-239 Electric field 49, 187, 321, 324, 332 Electron transfer 11, 23, 174, 239-295, 297, 307 Electrostatic effect 235, 247, 249 Electrostatic Hamiltonian 236 Electrostatic interaction 68 Electron Spin Resonance (ESR) spectra, g-factors 284-285, 290, 298
Ferrocene 52, 279 Floating Occupation Molecular Orbital (FOMO) method233 - FOMO-CASCI233, 244, 245-246 - FOMO-CI233-238 Fullerenes 94, 111-112, 130, 148, 150, 152, 159-160 Gaussian wavepacket 179 Gradient difference 10, 29 Graphitic materials 164, 167 Green's functions methods 49, 80 Green Fluorescent Protein (GFP) 7, 27, 233-234, 236, 239-240, 244-249 H2 42-44, 54, 100 Halorhodopsin 225 Hartree-Fock (HF) method 41-43, 318320, 323-324, 332 - coupled Hartree-Fock (CHF) 49, 53 - time-dependent Hartree-Fock (TDHF) (see also Random-Phase Approximation) 49, 50, 53, 97, 105, 106, 109,332 Hartree-Fock (HF) wavefunction 37 Hellmann-Feynman theorem 52, 96 Hexatriene(s) 4, 52 -1,3,5-trans-hexatriene 134-135,
155 Hybrid functionals 100, 105, 109-110, 114 Hybrid methods Hydrogen transfer 7, 20, 112, 297 Huckel methodology 262 Inactive orbitals 57, 193, 326 Indole 112 Initial Relaxation Directions (IRD) 199-200, 205, 208-209 Internal coordinates 7, 158 Intersection space 7, 12, 174, 180 Intrinsic Reaction Coordinate (IRC) method 198,200,205,208 Intruder states 60-61, 79, 110, 283-284,
329
351
Jahn-Teller coupling 80, 161 Jahn-Teller distorsion 274, 331 Lagrangian method 96 Lanthanides 79, 338, 341-342, 344 Level-shift CASPT2 (LS-CASPT2) method 61, Linear combination of atomic orbitals (LCAO)96, 101, 115, 322 Linear response method(s) 332, 336, 344 Luciferin 112 MnCV 279,283,291,309 Minimum energy path (MEP) 2, 12, 25, 82, 111, 188, 192,210 Molecular dynamics (MD) 12, 29, 115, 174-176, 186, 197 Molecular Mechanics/Valence Bond (MMVB) method 234 Mobius-Huckel theory 261 Mononuclear oxomolybdenum enzymes 280 Moller-Plesset (second-order) perturbation theory (MP2) 37, 49, 67, 153, 197, 203, 282, 328-329 Multiconfigurational self-consistent field (MCSCF) method 5-7, 37-38, 4546, 52-57 Multireference perturbation theory (MRPT) 37-38, 46, 62, 64, 79, 228 Multiple spawning method 238-239 - ab initio multiple spawning (AIMS) 239 Multi-state Complete Active Space Multireference Second Order Perturbation Theory (MS-CASPT2) 62-64, 6871,74,83-84,329,330 n-tetrasilane 67-70 N2O2 46, 52 Natural Orbitals (NOs) 46, 69, 98, 300 Non-dynamic correlation 45, 50 Normal coordinates 134, 139, 145-146, 149, 151, 159 Normal mode(s) 129, 136, 153 Ozone 50, 52 Pariser-Parr-Pople (PPP) method 2, 320-
321, 333-334 Penta-3,5-dieniminium cation 9, 184, 186 Pericyclic reaction 4, 261
Perturbation theory 35-37, 48, 50-51, 53, 58-59,79,94, 106, 111, 148,231, 325, 328, 338 Photoactive Yellow Protein (PYP) 7,117, 186-187,225,229,232,237 Photo- isomerization 9, 27, 82, 112, 150, 152, 157, 159, 208-218, 220, 237 -cis-trans 9, 10, 183-187, 192, 202, 208, 211, 216-217, 225, 332 Plastocyanin 294-300 Platinum atom 339-340 Polycyclic Aromatic Hydrocarbons (PAHs) 111, 164-165, 167 Polymers 94, 109-110, 115-116, 191 Polythiophene(s) 115-116 Porphyrin(s) 78-79, 94, 112-114, 321 Potential energy surface(s) 2-3, 5-9, 12, 26,81, 129, 135, 149, 153, 171, 176-178, 186, 188-189, 192-198, 206, 208, 211, 213, 216, 226, 229231,234,238-239 Protonated Schiff Base(s) (PSB) 4, 9, 14, 29, 82, 84, 153, 156-157, 192, 202, 237 Pyrazoline 20-26 Quantum Mechanics/Molecular Mechanics (QM/MM) approach 7, 27, 79, 107, 117, 235-238, 245-246, 249, 269 Radical anion 72, 77, 78, 80, 258, 270, 272 Radical pair 21, 23 Random-Phase Approximation (RPA) 49, 50,53,97, 105, 106, 109,332 Rrestricted Active Space State Interaction (RASSI) method 204, 304, 338-339 Reaction field (RF) approaches 79 Reaction path(way)(s) 1, 3, 5, 7, 9, 12, 19-
20,24,29-30,53,71,81-83, 111, 130, 150, 171-177, 179-184, 187188, 192-202, 204-205, 207-208, 213, 216-220, 226-227, 230-232, 238, 249, 269 Relativistic effect 80, 303, 338, 345 Renner-Teller coupling 80 Renner-Teller intersection(s) 226 Restricted active space self consistent field (RASSCF) 46, 80, 203, 214, 338, 344 Rhodopsin (Rh) 7, 9, 12, 27, 157, 159, 183, 202, 212-213, 225, 237, 332 Rydberg character 76, 107, 228
352 Rydberg excitation 101, 105 Rydberg orbitals 45, 68-69, 73, 326, 327 Rydberg (like) states 45, 61-63, 65, 68-69, 71-72, 76, 83, 109, 228, 321, 326327,331,335 Rydberg (-type) functions 61, 68-69, 71, 74-75 Rydberg transition(s) 69, 76, 321
Symmetry-allowed transition(s) 141 Symmetry-forbidden transition(s) 130, 141, 147-148, 161, 336 Surface hopping dynamics 177-178, 186,
188, 238-239e Tetraradical 14, 18, 217 Tamm-Dancoff approximation (TDA) 51,
105-106 Saddle point 14, 192 Salem correlation diagrams 2, 5-6 Second-Order Polarization Propagator Approach (SOPPA) method 49-50, 53 Semibullvalene 218, 263, 267, 272-273 Semi-classical dynamics methods 12, 29, 171, 177, 178, 179, 188, 196, 198, 238 Semiconductor clusters 94, 114-115 Semi-empirical methods 2, 74, 107, 232238, 244 - CAS/CI 7 - MNDOC 232 -QCFF/PI 150, 156, 167,238 Single-configuration methods 32, 47-48, 78 Slater-type orbital (s) (STO) 105 Solvation 112, 117,236,244 Spatial symmetry 37, 57, 103, 106, 195, 324 Spin-allowed transition(s) 141 Spin-forbidden transition(s) 130, 147, 150151, 161 Spin-orbit coupling 67, 80, 151, 273-274, 287, 290, 338, 340, 343, 345 Spin symmetry 37, 57, 103, 106, 195, 324 State-average CASSCF (method) 57, 68, 233, 327 Stationary point(s) 56, 101, 197-198, 200, 201,211,213,234 Stereochemistry 1, 6, 26, 220, 261, 265 Steric effect 235 Stilbene 112 Symmetry 4-5, 14, 35, 37-38, 40, 42-43, 50, 57,68-69,71-74, 102-103, 106, 110, 116, 130, 134-135, 139141, 143-144, 147-149, 151, 153, 159-161, 195, 203, 214, 226, 281282, 284-286, 293, 295, 319, 324, 326-327, 329, 331, 336-337, 339 Symmetry-Adapted Cluster Configuration Interaction (SAC-CI) 38, 50, 52-53, 78,291,293
Time-Dependent Density Functional Theory (TDDFT) 8, 13, 29, 38, 50, 53-53, 74-75, 78, 81, 93-128, 228-229, 231, 279, 291, 293, 298, 335-336, 342 Time-Dependent Kohn Sham framework (TDKS ) 94-95 - time-Dependent Kohn Sham eigenvalue problem (TDKS EVP) 97-98, 102, 105 Trans-1,3-butadiene 48 Transition metal compounds 94, 114, 340 Transition state 1, 3, 5, 20, 37, 65, 71, 77, 82, 180-182, 187, 193, 195-198, 204-205, 213, 216, 226, 234, 261, 274, 343 Transition state theory 197 Triplet state(s) 77, 148, 150-153, 155-156, 161,319,330,336,339,343 UO2 342 Urocanicacid 112 Variational principle 96, 179, 320, 322, 326 Vertical excitation energies 13, 47, 65, 74,
81, 101, 107, 109, 111, 115,203, 228,229,231-232,283 Vinylheptafulvene 179, 182 Woodward-Hoffmann rules 2, 5, 153-154, 172 Zeolites 284-285, 289-290, 294