The Journal of Antimicrobial Chemotherapy (Volume 62, Number 6, 2008)

701

A Versatile Analytical Expression for the Inverse Abel Transform Applied to Experimental Data with Noise Shuiliang Ma, Hongming Gao, Guangjun Zhang, and Lin Wu

Spectroscopic Techniques 708

A Strategy to Prevent Signal Losses, Analyte Decomposition, and Fluctuating Carbon Contamination Bands in Surface-Enhanced Raman Spectroscopy Boon-Siang Yeo, Thomas Schmid, Weihua Zhang, and Renato Zenobi

Notes 714

716

Mid-Infrared Laser-Induced Breakdown Spectroscopy Emissions from Alkali Metal Halides Clayton S.-C. Yang, E. Brown, Uwe Hommerich, Sudhir B. Trivedi, Alan C. Samuels, and A. Peter Snyder Evaluation and Comparison of Two Combinations of Pneumatic Nebulizers and Spray Chambers for Direct Slurry Aspiration and Multielement Analysis of Infant Milk Powders by Axial-Viewing Inductively Coupled Plasma Atomic Emission Spectrometry G. A. Zachariadis and L. I. Valianou

Columns and Features 136A 140A 144A 151A 153A

Cover Feature Spectroscopists’ Calendar What’s New 2008 Buyer’s Guide Applied Spectroscopy News Book Reviews

Cover Feature A confocal Raman microscope focuses a laser beam to a small diffraction-limited focal volume and uses an aperture to block scattered light originating from outside this region, thereby improving the spatial resolution. However, this attenuation is not 100% effective; there is an extended illumination volume that can excite detectable Raman scatter on either side of the optimum focus. In a transparent sample, every point in the extended volume can generate a few rays that pass through the focal point and hence through the confocal aperture, and the summed out-of-focus contribution can be significant compared to the signal from the optimum focus. The net result is that with transparent samples the surface specificity of confocal Raman microscopy is significantly worse than one would imagine and the spectrum of a surface is overlaid by significant features from subsurface domains. This can have confusing, counterintuitive ramifications, such as an increase in the signal from the substrate, relative to the surface, on moving the laser focus above the surface of a sample. These effects can be modeled using a simple analytical treatment. For more information, please see the paper ‘‘The Influence of Out-ofFocus Sample Regions on the Surface Specificity of Confocal Raman Microscopy’’, by Neil Everall.

136A

Volume 62, Number 6, 2008

Mary Carrabba, Editor

Spectroscopists’ Calendar is a regular feature in Applied Spectroscopy. Your cooperation in submitting timely information for the column is appreciated. For publication in one issue of Applied Spectroscopy, send information no less than four months in advance of the actual event. Please send announcements of meetings and symposia of interest to spectroscopists to Mary Carrabba, Department of Chemistry, Southern Oregon University, 1250 Siskiyou Blvd., Ashland, Oregon 97520, e-mail: carrabbam@sou. edu, Ph: (541)261-9800. Announcements of short courses, schools, workshops, or other educational activities should be sent to the SAS webmaster, Stephen Bialkowski ([email protected]).

CMA4CH Mediterranean Meeting: Multivariate Analysis and Chemometrics applied to Cultural Heritage and Environment, 1–4 June 2008, Ventotene Island, Latium, Italy; e-mail: [email protected], world wide web: http://w3.uniroma1.it/cma4ch/. ASMS 2008—56th ASMS Conference on Mass Spectrometry, 1–5 June 2008, Denver, Colorado; American Society for Mass Spectrometry, 2019 Galisteo Street, Building I, Santa Fe, New Mexico 87505 (USA), e-mail: [email protected], world wide web: http://www.asms.org, Ph: (505)989-4517, Fax: (505)989-1073. GeoRaman 08—8th International Conference on Raman Spectroscopy Applied to the Earth Sciences, 2–6 June 2008, Ghent, Belgium; Peter Vandenabeele, Department of Analytical Chemistry, Proeftuinstraat 86, B-9000 Ghent, Belgium, e-mail: GeoRaman@UGent. be, world wide web: http://www.georaman. ugent.be, Ph: 32 9 264 66 23, Fax: 32 9 264 66 99. 10th International Workshop on Physical Characterization of Pharmaceutical Solids, 8–13 June 2008, Bamberg, Germany; Sarah Passell, Communications Manager, Assa International, 3B East Lake Road, Danbury, Connecticut 06811, e-mail: communications@ assainternational.com, world wide web: www. assainternational.com, Ph: (203)312-0682, Fax: (203)312-0722. 63rd Ohio State University International Symposium on Molecular Spectroscopy, 16–20 June 2008, Columbus, Ohio; Terry A. Miller, Chair, International Symposium on Molecular Spectroscopy, Department of 140A

Volume 62, Number 6, 2008

Chemistry, The Ohio State University, 100 West 18th Avenue, Columbus, OH 43210, e-mail: [email protected]. edu, world wide web: http://molspect. chemistry.ohio-state.edu/symposium/, Ph: (614)292-2569, Fax: (614)292-194. EXRS 2008—13th European Conference on X-ray Spectrometry, 16–20 June 2008, Cavtat, Dubrovnik, Croatia; EXRS-2008 Secretariat, Rudjer Boskovic Institute, P.O. Box 180, 10002 Zagreb, Croatia, e-mail: [email protected], world wide web: http:// exrs2008.irb.hr/. ISEAC 35—25th International Symposium on Environmental Analytical Chemistry, 22–26 June 2008, Gdansk, Poland; International Association of Environmental Analytical Chemistry (IAEAC), Attn: Mrs. Marianne Frei, Postfach 46, CH-4123 Allschwil 2, Switzerland, e-mail: [email protected], world wide web: http://www.pg.gda.pl/chem/iaeac/index. htm. 21st International Conference on X-Ray and Inner-Shell Processes, 22–27 June 2008, Paris, France; world wide web: http://x08. spectro.jussieu.fr. 11th Conference on Chemometrics in Analytical Chemistry (CAC 2008), 30 June–4 July 2008, Montpellier, France; CAC-2006, Jean-Michel Roger, Cemagref, 361 rue JF Breton—BP 5095, 34196 Montpellier Cedex, France, e-mail: [email protected], world wide web: http://www.cac2008.org, Fax: 33 467046306. 21st International Activated Carbon Conference (IACC-21), 3–4 July 2008, Madrid, Spain; Dr. Henry Nowicki, Conference Chairman, PACS, Inc., 409 Meade Drive, Coraopolis, Pennsylvania 15108, e-mail: Henry@pacslabs. com, world wide web: http://www.pacslabs.com, Ph: (724)457-6576 or (800)367-2587, Fax: (724)457-1214. CHIRALITY—2008: 20th International Symposium on Chirality (ISCD-20), 6–9 July 2008, Geneva, Switzerland; Organizers Switzerland Ltd., 噦 Chirality 2008 Symposium, Obere Egg 2, 4312 Magden, Switzerland, e-mail: [email protected], world wide web: https://asp.artegis.com/lp/CHI08/ CHI08?1⫽1, Ph: 41 61 836 98 76, Fax: 41 61 836 98 77. EUROMAR 2008, 6–11 July 2008, St. Petersburg, Russia; e-mail: hq@euromar2008. com, world wide web: http://www. euromar2008.com.

14th Biennial National Atomic Spectroscopy Symposium (14th BNASS), 7–9 July 2008, Brighton, United Kingdom; 14th BNASS Secretariat, 噦 Dr Joanna Wragg, British Geological Society, Keyworth, Nottingham, NG12 5GG, United Kingdom; e-mail: [email protected], world wide web: http://www.bnass.org. The Sixteenth Annual International Conference on Composites/NANO Engineering (ICCE-16), 20–26 July 2008, Kunming, China; Professor David Hui, Conference Chair, Department of Mechanical Engineering, University of New Orleans, New Orleans, Louisiana 70148, e-mail: [email protected], world wide web: http://www.acad.polyu.edu.hk/ ⬃mmktlau/ICCE/ICCE㛮Main.htm, Ph: (504) 280-6652, Fax: (504)280-6192. 50th Rocky Mountain Conference on Analytical Chemistry, 27–31 July 2008, Breckenridge, Colorado; Milestone Presentations, LLC, 4255 S. Buckley Road, Suite 118, Aurora, Colorado 80013, e-mail: info@ rockychem.com, world wide web: http:// www.rockychem.com, Ph: (800)996-3233 or (303)690-3233, Fax: (888)996-3296 or (303)690-3278. 14th International Diffuse Reflectance Conference (IDRC 2008), 1–8 August 2008, Chambersburg, Pennsylvania; Steve Delwiche, General Chair, e-mail: stephen. [email protected], world wide web: http://www.idrc-chambersburg.org. Gordon Research Conference on Vibrational Spectroscopy, 3–8 August 2008, South Hadley, Massachusetts; Chair: Philip J. Reid, Chair, University of Washington, Department of Chemistry, Box 351700, Seattle, Washington 98195-1700, e-mail: [email protected], world wide web: http://www.grc.org/programs.aspx? year⫽2008&program⫽vibrspec. ASTM International Conference on Surface and Dermal Sampling, 4–8 August 2008, Boulder, Colorado; Dr. Kevin Ashley, Conference Chairman, CDC/NIOSH, 4676 Columbia Parkway, Mail Stop R-7, Cincinnati, OH, 45226-1998; e-mail: [email protected], world wide web: http://www.astm.org/MEETINGS/ COMMIT/d22boulder.html, Ph: (513)8414402, Fax: (513)458-7189. DXC 2008–57th Annual Denver X-ray Conference, 4–8 August 2008, Denver, Colorado; Denise Flaherty, DXC Conference Coordinator, 12 Campus Boulevard, Newtown Square, Pennsylvania 19073-3273, e-mail: flahertyj@

(Continued)

icdd.com, world dxcicdd.com.

wide

web:

http://www.

20444267 or 420 2 66053635, Fax: 420 2 86582307.

American Chemical Society 236th National Meeting & Exposition, 17–21 August 2008, Philadelphia, Pennsylvania; ACS Meetings, 1155 16th St., N.W., Washington, D.C. 200364899, e-mail: [email protected], world wide web: http://www.chemistry.org, Ph: (800)2275558, Fax: (202)872-6128.

SMASH 2008—Small Molecule NMR Conference, 7–10 September 2008, Santa Fe, New Mexico; Daneen Hadden, Conference Secretary, world wide web: http://www.smashnmr. org/main.asp, Ph: (317)655-2032.

21st International Conference on Raman Spectroscopy (ICORS 2008), 17–22 August 2008, London, United Kingdom; The Secretariat, Hampton Medical Conferences Ltd., 113-119 High Street, Hampton Hill, Middlesex TW12 1NJ, UK, e-mail: ICORS@ hamptonmedical.com, world wide web: www.icors2008.ukevents.org/contact. ICMRBS 2008—23rd International Conference on Magnetic Resonance in Biological Systems, 24–29 August 2008, San Diego, California; world wide web: http://www. icmrbs2008.org/. 13th International Conference on Capture Gamma-Ray Spectroscopy and Related Topics (CGS-13), 25–29 August 2008, Cologne, Germany; Prof. Jan Jolie, Institute for Nuclear Physics, University of Cologne, Zu¨lpicherstrasse 77, D-50937 Ko¨ln, Germany, e-mail: [email protected], world wide web: http://www.ikp.uni-koeln.de/cgs13/, Fax: 49 221 470 5168. 5th International Conference on Broadband Dielectric Spectroscopy and Its Applications, Joint meeting of the 5th Conference of the ‘‘International Dielectric Society’’ and the 10th Conference on ‘‘Dielectric and Related Phenomena’’, 26–29 August 2008, Lyon, France; BDS 2008 Secretariat, IMP/LMPB, UMR CNRS 5223, Universite´ Lyon 1, Baˆt. ISTIL, 43 Bd. du 11 Novembre 1918, 69622 Villeurbanne Cedex, France, e-mail: [email protected], world wide web: http://bds2008.univ-lyon1.fr, Fax: 33 4 78 89 25 83. 8th Atmospheric Spectroscopy Applications Meeting, 27–30 August 2008, Reims, France; Ludovic Daumont, e-mail: ludovic.daumont@ univ-reims.fr, world wide web: http://asa. univ-reims.fr. The 20th International Conference on High Resolution Molecular Spectroscopy (PRAHA2008), 2–6 September 2008, Prague, Czech Republic; Professor Sˇte˘pa´n Urban, Local Organizing Committee Co-Chair, PRAHA2008, Academy of Sciences of the Czech Republic, J. Heyrovsky´ Institute of Physical Chemistry, Dolejsˇkova 3, CZ-18223 Praha 8, Czech Republic, e-mail: praha06@jh-inst. cas.cz, world wide web: http://www.chem. uni-wuppertal.de/conference/, Ph: 420 2

30th Annual Meeting of the British Mass Spectrometry Society (BMSS 2008), 7–10 September 2008, York, United Kingdom; Dave Collison, 68 Bathurst Road, Winnersh, Wokingham, Berkshire RG41 5JF, United Kingdom, e-mail: [email protected], world wide web: http://www.massspectrum.co.uk/ york2008/, Ph: 44 0 118 961 9155, Fax: 44 0118 901 8359. XIII International Symposium on Luminescence Spectrometry. Analytical luminescence: new diagnostic tools in life sciences, food safety and cultural heritage (ISLS 2008), 7–11 September 2008, Bologna, Italy; Professor Aldo Roda, Symposium President. Department of Pharmaceutical Sciences, University of Bologna, Via Belmeloro 6, I-40126 Bologna, Italy, e-mail: isls2008.bologna@ unibo.it, world wide web: http://www. isls2008.unibo.it, Ph: 39 051 343398, Fax: 39 051 343398. 10th Rio Symposium on Atomic Spectrometry, 7–12 September 2008, Salvador, Bahia, Brazil; Dr. Bernhard Welz, Departamento de Quı´mica, Universidad Federal de Santa Catarina, 88040-900 Floriano´polis—SC, Brazil, e-mail: [email protected], world wide web: http://www.10riosymposium.ufba.br, Ph: 55 48 3733 8876, Fax: 55 48 3721 6850. European Symposium on Atomic Spectrometry, 28 September–1 October 2008, Weimar, Germany; world wide web: http:// www.esas-symposium.de. Expoquimia, International Chemical Industry Exhibition, 20–24 October 2008, Barcelona, Spain; Fira De Barcelona, Av. Reina M. Cristina, E-08004 Barcelona, Spain, e-mail: [email protected], world wide web: www.expoquimia.com, Ph: 34 3 2332200, Fax: 34 3 2332001. 25th LC/MS Montreux Symposium on LC/ MS, 12–14 November 2008, Montreux, Switzerland; Secretary, Marianne Frei, Postfach 46, CH-4123 Allschwil 2, Switzerland, e-mail: [email protected], world wide web: http://www.iaeac.ch/lcms-montreux.html, Ph: 41 61 481 27 89, Fax: 41 61 482 08 05. Third Asia-Pacific Winter Conference on Plasma Spectrochemistry, 16–21 November 2008, Tsukuba, Japan; Naoki Furuta, Chuo University, Faculty of Science and Engineering, Department of Applied Chemistry, Environmental Chemistry Laboratory, 1-13-27 Ka-

suga, Bunkyo-ku, Tokyo 112-8551, Japan, e-mail: [email protected], world wide web: http://envsun.chem.chuo-u.ac.jp/ plasma/2008apwccont.htm, Ph: 81 3 3817 1906, Fax: 81 3 3817 1699. Eastern Analytical Symposium & Exposition, 17–20 November 2008, Somerset, New Jersey; EAS Executive Secretary and Retort Editor, Eastern Analytical Symposium, Inc., P.O. Box 370, Walkersville, MD 21793, e-mail: [email protected], world wide web: http://www.eas.org, Ph: (301)682 3701, Fax: (301)668 4312. 7th Biennial Conference of the Australian and New Zealand Society for Magnetic Resonance (ANZMAG 2008), 6–11 December 2008, Queensland, Australia; world wide web: http://www.anzmag.com.au/anzmag2008.html. European Winter Conference on Plasma Spectrochemistry, 15–20 February 2009, Graz, Austria; Ms. Astrid Tuider, Conference Secretary, Karl-Franzens University Graz, Institute of Chemistry/Analytical Chemistry, Universita¨tsplatz 1, A-8010 Graz, Austria, e-mail: [email protected], world wide web: www.winterplasmagraz.at, Ph: 43 316 380 5300, Fax: 43 316 380 9845. The Pittsburgh Conference on Analytical Chemistry and Applied Spectroscopy (PITTCON 2009), 8–13 March 2009, Chicago, Illinois; PITTCON, 300 Penn Center Boulevard, Suite 332, Pittsburgh, Pennsylvania 15235, e-mail: [email protected], world wide web: http://www.pittcon.org, Ph: (412) 825-3220 or (800)825-3221, Fax: (412)8253224. American Chemical Society 235th National Meeting and Exposition, 22–26 March 2009, Salt Lake City, Utah; ACS Meetings, 1155 16th St., N.W., Washington, D.C. 20036-4899, e-mail: [email protected], world wide web: http://www.chemistry.org, Ph: (800)227-5558, Fax: (202)872-6128. Third International Congress on Operando Spectroscopy (Operando—III): Recent Developments and Future Perspectives in Spectroscopy of Working Catalysts, 19–23 April 2009, Rostock-Warnemu¨nde, Germany; Secretariat: Operando III, Leibniz-Institut fu¨r Katalyse, an der Universitat Rostock, AlbertEinstein Straße 29 a, 18059 Rostock, Germany, e-mail: [email protected], world wide web: www.catalysis.de/operando, Ph: 49 381 1281 169; Fax: 49 381 1281 5000. American Chemical Society 236th National Meeting and Exposition, 16–20 August 2009, Washington, D.C.; ACS Meetings, 1155 16th St., N.W., Washington, D.C. 20036-4899, e-mail: [email protected], world wide web: APPLIED SPECTROSCOPY

141A

(Continued)

http://www.chemistry.org, Ph: (800)227-5558, Fax: (202)872-6128.

chem.chuo-u.ac.jp/plasma/2008apwccont.htm, Ph: 81 3 3817 1906, Fax: 81 3 3817 1699.

14th International Conference on Near Infrared Spectroscopy (ICNIRS 2009), 7–16 November 2009, Bangkok, Thailand; Sirinnapa Saranwong, General Secretariat (Asian NIR Consortium), National Food Research Institute, 2-1-12 Kannondai Tsukuba JAPAN 305-8642, e-mail: [email protected], world wide web: http://www.nir2009.com, Ph: 81 29 838 8088, Fax: 81 29 838 7996.

45th Eastern Analytical Symposium and Exposition, 17–20 November 2008, Somerset, New Jersey; Eastern Analytical Symposium and Exposition, Inc., P.O. Box 633, Montchanin, Delaware 19710, e-mail: easinfo@ aol.com, world wide web: http://www.eas. org, Ph: (610)485-4633, Fax: (610)485-9467.

Third Asia-Pacific Winter Conference on Plasma Spectrochemistry (2008 APWC), 16–21 November 2008, Tsukuba, Ibaraki, Japan; Naoki Furuta, Chuo University, Faculty of Science and Engineering, Department of Applied Chemistry, Environmental Chemistry Laboratory, 1-13-27 Kasuga, Bunkyo-ku, Tokyo 112-8551, Japan, e-mail: nfuruta@chem. chuo-u.ac.jp, world wide web: http://envibm.

142A

Volume 62, Number 6, 2008

2008 Materials Research Society Fall Meeting, 1–5 December 2008, Boston, Massachusetts; Materials Research Society, 506 Keystone Drive, Warrendale, Pennsylvania 150867573, e-mail: [email protected], world wide web: http://www.mrs.org, Ph: (724)779-3003, Fax: (724)779-8313. The Pittsburgh Conference on Analytical Chemistry and Applied Spectroscopy (PITTCON 2009), 8–13 March 2009, Chicago, Illinois; PITTCON, 300 Penn Center

Boulevard, Suite 332, Pittsburgh, Pennsylvania 15235, e-mail: [email protected], world wide web: http://www.pittcon.org, Ph: (412)825-3220 or (800)825-3221, Fax: (412)825-3224. ASMS 2009–57th ASMS Conference on Mass Spectrometry, 31 May–4 June 2009, Philadelphia, Pennsylvania; American Society for Mass Spectrometry, 2019 Galisteo Street, Building I, Santa Fe, New Mexico 87505 (USA), e-mail: [email protected], world wide web: http://www.asms.org, Ph: (505)9894517, Fax: (505)989-1073. 36th Annual Conference of the Federation of Analytical Chemistry and Spectroscopy Societies (FACSS), 18–20 October 2009, Louisville, Kentucky; FACSS, P.O. Box 24379, Santa Fe, New Mexico 87502, e-mail: [email protected], world wide web: www.facss.org, Ph: (505)820-1648, Fax: (505)989-1073.

Lee Craven, Editor

WHAT’S NEW is provided as a service for our readers. It contains the latest news on the products, catalogs, tips, and supplies that manufacturers elect to highlight. Publication in WHAT’S NEW does not imply recommendation or endorsement by the Society for Applied Spectroscopy or the column editor. Contributions to WHAT’S NEW should be sent to Applied Spectroscopy, What’s New Editor, 201B Broadway Street, Frederick, MD 21701-6501 (Fax 301-6946860; e-mail [email protected]). Please note that only black-andwhite photos will be considered for publication. Although there is no charge to submit materials, preference is given to Applied Spectroscopy advertisers. Call 800-627-0932 for questions regarding advertising. ␭

Probes of Molecular Chirality and Structure New accessories from Olis. Fluorescence and circular dichroism (CD) spectroscopy are well represented in the literature of macromolecular studies. Polarization of fluorescence (POF) and circularly polarized luminescence (CPL) are less utilized highly sensitive probes of molecular interactions (e.g., protein folding, secondary structure formation). Whereas CD involves the ground state of a chiral molecule, CPL is emitted by a fluorophore if the excited state of the emitting species is chiral. Fluorescence offers insights into the environment surrounding the emitting molecule; anisotropy (POF) is sensitive to rotational motion of the fluorescent probe. These four probes provide complementary informa144A

Volume 62, Number 6, 2008

tion, with POF and CPL providing insights into the structure, interactions, and environment of the emitting molecule. A recently introduced Olis upgrade accessory adds both CPL and POF to any Olis CD or fluorescence spectrometer for kinetic or steady-state readings. Single purpose models are also available. OnLine Instrument Systems, 130 Conway Dr., Suites A and B, Bogart, GA 30622-1724, Ph: 706-353-6547, www.olisweb.com.

␭

vestigative, such as forensic trace evidence analysis, to routine manufacturing troubleshooting, as in the polymer industry. The DXR Raman microscope brings the power of Raman microscopy to academic studies, for characterization of geological specimens, for example; to routine product quality control; and to sample authentication in industries such as fine gemstones. This Raman microscope is designed from the ground up for routine analysis. In addition to the DXR Raman microscope, Thermo Fisher Scientific is also launching its new DXR SmartRaman spectrometer. It is a macro Raman analytical tool designed specifically to bring the power of Raman spectroscopy to busy multipurpose laboratories. Thermo Fisher Scientific, 81 Wyman Street, Waltham, MA 02454, Ph: 800532-4752, www.thermo.com.

␭

New DXR Raman Microscope Thermo Fisher Scientific Inc. launched its new Thermo Scientific DXR Raman microscope. The instrument is designed specifically to help non-specialist users achieve rapid sampling and analysis of particles, down to one-micron spatial resolution. The novel microscope offers excellent spatial resolution, superior performance, and unmatched reproducibility in a package that anyone can use. The DXR Raman microscope is equipped with fully integrated, pre-aligned components for fast and easy field installation and configuration. Interchangeable SMART 姞 components require no operator adjustment and ensure automated system configuration. Patented auto alignment and auto calibration ensure reliable results. A fiber probe option is available for remote sampling. Furthermore, the microscope utilizes the Thermo Scientific ValPro 姞 complete system validation package, allowing for compliance with cGMP and FDA regulatory requirements. Thermo Fisher Scientific also offers a large collection of Raman spectral libraries to aid in sample identification. The DXR Raman microscope is designed to make the Raman technique accessible to a much wider audience by replacing manual adjustments with system intelligence and automation. The newly designed instrument is ideal for a broad range of applications, from the in-

New Ultra-Violet/Visible (UV/VIS) Spectrophotometers PerkinElmer Life and Analytical Sciences announced the global introduction of the LAMBDA姟 XLS, a UV/Vis spectrophotometer for Quality Assurance/Quality Control (QA/ QC) and teaching laboratories, and the LAMBDA姟 Bio, a UV/Vis spectrophotometer designed specifically for biological science laboratories. Both the LAMBDA XLS and LAMBDA Bio are designed as low-cost, routine platforms with a number of pre-configured standard methods and the capability to add customized methods, addressing a wide range of applications. The LAMBDA XLS was designed with a focus on productivity and ease of use. The LAMBDA XLS and Bio models include a large, clear on-board display and

2008 BUYER’S GUIDE

hanced return on investment over previous technology. The Optima 7000 Series replaces the Optima 2100, 5100, 5200, and 5300 models. Perkin Elmer, Inc, 940 Winter Street, Waltham, MA 02451, Ph: 781-663-6900, www.perkinelmer.com.

␭ robust spill-resistant keypad. With its intuitive graphical interface and wide range of local language options, users in manufacturing QA/ QC, environmental, teaching, and food analysis laboratories are able to perform wavelength, scanning, concentration studies, and biological assays with ease. The LAMBDA Bio is pre-configured with standard methods for ease of access, including DNA, RNA, and oligonucleotide concentration and purity, protein assays, and cell density measurements. Both the LAMBDA XLS and LAMBDA Bio have no moving parts and feature an ultra-long-lifetime Xenon lamp, helping to ensure robustness, maximum uptime, and low cost of ownership. The high quality split-beam optical design provides high stability and run-to-run reproducibility for added assurance of results. Perkin Elmer, Inc, 940 Winter Street, Waltham, MA 02451, Ph: 781-663-6900, www.perkinelmer. com.

user input. Software packages include proteomics, polymer and copolymer analysis, LCMALDI, biomarker discovery, and tissue imaging. Shimadzu designed the AXIMA Confidence姟 and Assurance姟 instruments with general analytical and life-science laboratories in mind. Both systems are affordable, robust options for routine sample analysis performed manually or automatically. Positive- and negative-ion modes are standard, allowing users full flexibility. Shimadzu also incorporated a patented beam blanker to optionally remove unwanted low-mass ions and prevent detector saturation. The Confidence MALDI MS offers reliable mass information and MS/MS-derived structural detail. A linear mode enables high levels of analysis of high molecular weight samples. The reflectron mode, incorporating the patented curved-field reflectron, provides high resolution and mass accuracy. Confidence software packages are available specifically for proteomics experiments, LC-MALDI, polymer and copolymer analysis, biomarker discovery, tissue imaging, and oligonucleotide/primer analysis. Shimadzu Scientific Instruments, 7102 Riverwood Drive, Columbia, MD 21046, Ph: 800-477-1227, www.ssi.shimadzu.com.

␭

␭

Optima姟 7000 Series ICP-OES PerkinElmer Life and Analytical Sciences has announced the Optima姟 7000 Series of Inductively Coupled Plasma-Optical Emissions Spectrometers (ICP-OES). The Optima 7000 family is designed for best-in-class inorganic analysis and is used in a variety of markets, including environmental, geochemical, product testing, and biofuels. This newest generation in the Optima series includes a Universal Data Acquisition mode that records all of the spectral data for each sample. This enables customers to retrieve data that was not initially reported without needing to run the sample again, saving them time and increasing productivity. The Optima 7000 has several time-saving enhancements for laboratory personnel. For rapid review of results, the Optima 7000 generates an error flag if a sample result falls above or below a user-specified value. To ensure a consistently high level of data and without requiring additional resources, the instrumentation can automatically generate Quality Control (QC) charts that show results for analyses run over a specified period of time. The addition of modern electronics extends the instrument’s lifetime, providing en-

MALDI Mass Spectrometer Offering Enhanced Performance and Flexibility Shimadzu Scientific Instruments has updated its line of AXIMA MALDI TOF mass spectrometers (MS) with three new systems to meet the needs of researchers with varying applications and budgets. The AXIMA Performance姟 mass spectrometer provides the pinnacle of performance for high-energy MS/MS interrogation of proteomics and biological and organic samples. These include peptides, polymers, oligonucleotides, SNPs, metabolites, and carbohydrates. It also demonstrates performance in the analysis of high-mass species, such as intact proteins. In reflectron mode, high-resolution MS data is generated for peptide mass fingerprinting and complex mixture analysis. Full automation, advanced calibration algorithms, data-dependent workflows, and application-centric software help users achieve maximum results with minimal APPLIED SPECTROSCOPY

145A

2008 BUYER’S GUIDE

NanoDrop Family of Micro-volume Spectroscopy Instruments Thermo Fisher Scientific Inc. announces the NanoDrop line of micro-volume spectroscopy systems. These unique instruments utilize a patented technology that enables easy measurement of UV-Vis and fluorescence without the use of cuvettes. The Thermo Scientific NanoDrop family of products extends the company’s molecular spectroscopy port-

146A

Volume 62, Number 6, 2008

folio for increasing applications involving small sample volumes. The Thermo Scientific NanoDrop instruments eliminate cuvettes and associated dilutions, resulting in more reliable measurements. An undiluted one-microliter sample is pipetted directly onto the measurement surface. After a quick spectral reading, the sample is simply wiped away in preparation for the next sample. The instruments are used to measure the quantity and purity of nucleic acids and proteins as well as associated fluorescent labels. Such measurements are routinely needed for quality control and sample preparation at multiple process points in many applications, including microarray probe preparations, PCR template normalization, small molecule crystallization, sequencing, antibodies, microgenomics, proteomics, genotyping, and FRET (fluorescence resonance energy transfer). These applications are used in such diverse fields as cancer research, microbiology, drug discovery, forensics, histocompatibility, and diagnostics. The Thermo Scientific NanoDrop line of products includes the Thermo Scientific NanoDrop 1000 Spectrophotometer, which takes UV-Vis absorbance spectra of one-microliter samples; the Thermo Scientific NanoDrop 8000 Spectrophotometer, which takes UV-Vis absorbance spectra of eight samples simultaneously; and the Thermo Scientific NanoDrop 3300 Fluorospectrometer, which performs broad spec-

trum fluorescent analysis with patent-pending technology that delivers a wide excitation range without requiring filter changes or a monochromator. Thermo Fisher Scientific, 81 Wyman Street, Waltham, MA 02454, Ph: 800-532-4752, www.thermo.com.

␭

Advanced Series of Monochromators and Spectrographs Princeton Instruments introduces the Acton Advanced Series of monochromators and spectrographs. Unlike the standard Acton spectrometers, the new Acton Advanced Series combines high precision computer-controlled motorized slits, interchangeable triple grating turrets, and dual entrance ports. These new features offer researchers complete soft-

2008 BUYER’S GUIDE ware control of incoming light to the spectrometer as well as the flexibility to configure separate experiments for each entrance port. The Acton Advanced Series is available in four different focal lengths, 150 mm, 300 mm, 500 mm, and 750 mm, and is highly configurable for different types of detectors, grating combinations, and light sources. The new Acton Advanced Series, like the previous SpectraPro series of spectrometers, continues to provide high quality spectroscopy components for researchers who demand the best equipment for their experimental setup. For decades Acton products have produced data for hundreds of posters, presentations, and publications, making Acton the proven choice for spectroscopy requirements. Princeton Instruments-Acton, 15 Discovery Way, Acton, MA 01720, Ph: 978-263-3584, www. piacton.com.

␭

ion battery or from AC power. Featuring interchangeable internal reflection (ATR) and external reflection sampling interfaces, this system is ideal for sampling a wide variety of materials including solids, pastes, gels, and liquids. Coupled with the unique A2 Technologies diamond internal reflection sampling system, this system is also capable of large surface analysis. Controlled and operated via a PDA device, the A2 Technologies Exoscan spectrometer combines all the unique capabilities of a laboratory-based FT-IR with the advantages of a portable instrument that can be used in the field, ensuring real-time results. Capable of nondestructive analysis, this new analyzer is ideal for applications for which the sample is too large to bring to the laboratory or too valuable to require a small portion to be removed for analysis. The optical system of the Exoscan is a monolithic, highly rugged infrared modulator that enables routine operation of the system in non-routine applications and environments without sacrificing performance. This system operates in the 4000 to 650 wavenumber mid-infrared region and is capable of 4 cm⫺1 resolution. Additionally, the analyzer is equipped with three-level software architecture, providing methods development, supervisory, and operator functionality. A2

Technologies, 14 Commerce Drive, Danbury, CT 06810, Ph: 203-312-1100, www. a2technologies.com.

␭

New Fourier Transform Infrared Spectrophotometer Offers High Sensitivity Shimadzu Scientific Instruments designed the IRAffinity-1 Fourier Transform Infrared (FT-IR) spectrophotometer with highthroughput optics and a dynamic alignment mechanism to increase sensitivity, stability, and usability. The IRAffinity-1 is ideal for highprecision infrared analysis to confirm, identify, and detect foreign matter in raw materials,

Portable Exoscan FT-IR Spectrometer for Surface Analysis Applications A2 Technologies announces the launch of one of the most compact FT-IR spectrometers available, Exoscan姟. Designed to move spectroscopy out of the laboratory and into the field, the analyzer is designed for nondestructive on-site surface and bulk analysis applications. Weighing less than seven pounds, the A2 Technologies Exoscan features ease of use, coupled with analytical performance that rivals far larger and more expensive traditional analytical FT-IR spectrometers. Exoscan is also capable of handling attenuated total reflection (ATR) applications. Exoscan is powered by an on-board rechargeable lithium APPLIED SPECTROSCOPY

147A

2008 BUYER’S GUIDE medical products, packages, and coatings. The IRAffinity-1 offers high signal-to-noise ratio (sensitivity) at 30,000:1, with a maximum resolution of 0.5 cm⫺1. The unit achieves this level of sensitivity by using a high-energy ceramic light source, temperature-controlled, high-sensitivity DLATGS detector, and highthroughput optical elements. Also, the IRAffinity-1 includes optimized electrical and optical systems to minimize noise levels. The IRAffinity-1 includes patented technology to increase stability and precision. Its moving mirror is run smoothly and precisely by a flexible joint system patented by Shimadzu. The interferometer is optimized and stabilized by a dynamic alignment mechanism on which Shimadzu has a patent pending. The interferometer’s optical elements are protected from humidity and stabilized by a sealed interferometer, as well as by continuously removing moisture by a reactive polymeric desiccator and coating the beam splitter with moisture-resistant protective film. To increase reliability, the IRAffinity1 executes self-diagnosis at initialization and monitors the state of the instrument during operation. Users can check basic performance using a standard-feature validation program. IRsolution software, standard in the IRAffinity1, emphasizes operability and analysis support programs to perform data processing and analysis. With IRsolution specialized windows, users can easily perform standard operations, such as measurement, display, data processing, quantitative analysis, search, and printing. A variety of optional programs and accessories, such as those facilitating PLS and multilinear regression, deconvolution, and mapping measurement, are also available. Shimadzu Scientific Instruments, 7102 Riverwood Drive, Columbia, MD 21046, Ph: 800-477-1227, www.ssi.shimadzu.com.

␭

Miniature Lasers RPMC introduces the Mini series of lasers. The Mini lasers get their name from their compact size. All optical and electronic compo148A

Volume 62, Number 6, 2008

nents are housed in one conductively cooled enclosure. These robust, sealed, QCW diode pumped lasers are available at 1064 nm, 532 ␮m, 355 ␮m, and 1.57 ␮m. The Mini is available with energy levels up to 7 mJ at 1.57 ␮m and 20 mJ at 1064 nm with 5 to 8 ns pulse widths at a repetition rate up to 50 Hz. If you prefer, the system components can be housed in two enclosures. The Mini can be customized, for example, 4th and 5th harmonics can be offered, as well as the OPO idler ⬃3.1 ␮m output. RPMC Lasers Inc., 203 Joseph Street, O’Fallon, MO 63366, Ph: 636-2727227 ext. 22, www.rpmclasers.com.

␭

New ITQ Ion Trap GC/MS Series Thermo Fisher Scientific Inc. announced the launch of its Thermo Scientific ITQ姟 Series of gas chromatography/mass spectrometry (GC/MS) ion trap instruments, featuring external ionization. The ITQ 700姟, ITQ 900姟, and ITQ 1100姟 ion trap systems feature fully upgradeable systems designed to provide high performance and high specificity. Developed for a wide range of applications, from routine GC/MS to research-grade ion trap MS, these new systems address the analytical needs of the environmental, food safety, pharmaceutical QA/QC, forensics, and toxicology industries, as well as academic laboratories. Designed for routine full scan quantitation and general teaching applications, the cost-effective ITQ 700 GC/MS system combines full scan ion trap sensitivity with a small footprint for laboratories with limited space. The ITQ 700 is capable of achieving a mass range of 700 amu, suitable for most general GC/MS applications, including environmental, QA/QC, and forensics. The ITQ 900 incorporates full scan sensitivity of ion trap mass spectrometry with the increased flexibility of the Thermo Scientific TRACE GC Ultra姟 gas chromatograph. Combining a complete range of injector options including true Cold-on-Column, Programmable Temperature Vaporization (PTV) and a full complement of additional detectors, the ITQ 900 significantly expands the working mass range of a typical GC/MS system to 900 amu. The large volume PTV injection with

back-flush option of the TRACE GC Ultra enables accurate detection of trace level components in complex matrices. This system enables users to perform routine full scan GC/ MS with greater analytical flexibility. Offering a broad selection of innovative features available for ion trap GC/MS, the Thermo Scientific ITQ 1100 system features new advanced MS/ MS (MS n) functions. These high-performance ion-trap-based GC/MS systems are ideal for laboratories seeking the most powerful GC/ MS platform and flexibility. For the first time, Thermo Scientific patented Pulsed Q Dissociation Mode (PQD) is available on a GC ion trap system. PQD increases the number of product ions formed during collision-induced disassociation (CID), yielding richer information for qualitative MS n experiments. The ITQ 1100 also features a 250 L/s turbo-molecular pump, the Thermo Scientific Vacuum Probe Interlock, and Data Dependent姟 Scanning. This integrated system provides an extended mass range up to 1100 amu, considerably increasing the number of compounds that can be detected and identified. In addition, the ITQ Series of GC/MS systems can be fully upgraded to maximize investment and offer the flexibility to cater to each analytical requirement. Detection capability can be improved using the MS/MS upgrade, while adding chemical ionization or direct sample probes allows increased flexibility. Injection options can also be expanded beyond split/splitless injectors to take advantage of advanced sample introduction techniques. Thermo Fisher Scientific, 81 Wyman Street, Waltham, MA 02454, Ph: 800-532-4752, www.thermo. com.

␭

Vacuum Ultraviolet Light Source, 30 nmⴙ The McPherson, Inc. Model 629 is a windowless hollow cathode source in which the negative discharge glow is viewed directly. Ionized gas emission lines are produced and can be conveniently viewed with little or no absorption by neutral gas. The source’s watercooled anode and cathode are electrically isolated. A differential pumping stage is available as an accessory to allow direct connection to experimental systems operating at lower vacuum pressures. The McPherson Model 629 may be operated with a variety of inert gases. Helium (He) is among the most popular. Op-

2008 BUYER’S GUIDE erating pressures of ⬃3 ⫻ 10⫺1 torr and discharge current of ⬃500 mA are optimum for emission of He II ion lines around 30.4 nm. With unchanged discharge current, higher pressure (⬃1 torr) provides optimum output at He I, 58.4 nm. Argon and Neon are other popular gases. This source may be used directly for ultraviolet spectroscopy experiments where unfiltered multi-line excitation is acceptable. Combine the Model 629 source with a high throughput vacuum monochromator like the 200 mm focal length McPherson Model 234/302 to benefit applications requiring greater specificity. This McPherson monochromator provides sub nanometer wavelength resolution, easily discriminating between multiple emission lines and rejecting out of band energy. Line of sight optical path is optionally available for the Model 629 light source, simplifying alignment with optical or sampling systems. The Model 629 windowless hollow cathode source is useful for wavelength calibration and as a general emission source for wavelengths 30 nm and longer. McPherson, Inc., 7A Stuart Road, Chelmsford, MA 01824-4107, Ph: 978-256-4512, www.McPhersonInc.com.

Road, Suite 335, Waltham, MA 02452, Ph: 781-478-0170, www.LambdaSolutions.com.

␭

Raman WorkStation姟 Macro Raman Analyzer Kaiser Optical Systems Inc. (Kaiser) is pleased to announce the release of the most recent member of the RamanRxn Systems姟 suite of analyzers; the Raman WorkStation姟.

The Raman WorkStation姟 incorporates Kaiser’s P hAT technology into a next-generation macro Raman analyzer. P hAT technology revolutionizes Raman sampling of solids and crystals by eliminating sampling irreproducibility and focus sensitivity. P hAT technology measures a representative sample volume, while reducing local laser power density, eliminating sample damage concerns. The analyzer comes equipped for automated highthroughput screening with varying spot sizes available from 7 to 1 mm. One advantage is for analyzing well-plates in screening experiments. Application needs and sampling versatility have been designed into the Raman WorkStation姟 by incorporating Kaiser’s universal fiber-optic sampling station interface. This allows a basic Raman WorkStation姟 to be expanded for in situ reaction analysis, remote solids sampling, or micrometer level imaging that extend the analyzer’s capability from the millimeter sampling domain to the micrometer sampling domain. The Raman WorkStation姟 has capabilities for the tablet and gelcap area by using a transmission accessory option. Kaiser Optical Systems, Inc., 371 Parkland Plaza, Ann Arbor, MI 48103, Ph: 734 665 8083, www.kosi.com.

␭

␭

High Performance Flow-Through Raman System The Lambda Solutions Dimension FT-ABS represents a true high sensitivity on-line Raman System for critical fluid-phase monitoring. The XL-FT features provide exceptional sensitivity and stability. The quartz-teflon cell design offers solvent/acid/base resistance and LSI RealTime monitoring software and optional integrated UV-Vis absorption extend analytical capability. Wide ranging applications make the XL-FT ideal for semi-conductor wet chemistry, petrochemical production and formulation, fermentation, and chemical reaction monitoring. This flow-through system offers high performance and robustness, with proven 24/7 operation, coverage from 50 cm⫺1 to 3000 cm⫺1, and resolution to 1.5 pixels/cm⫺1. Lambda Solutions, Inc., 411 Waverley Oaks APPLIED SPECTROSCOPY

149A

Deborah K. Bradshaw, Editor Applied Spectroscopy News is a monthly feature in the journal. It includes information from the Society for Applied Spectroscopy, news from other societies and institutions, announcements of meetings, schools, or other activities, and reports of symposia from recent conferences. If you have news items, a meeting announcement, or a report from a symposium that would be of interest to readers of Applied Spectroscopy, contact Deborah K. Bradshaw, Applied Spectroscopy News Editor, P.O. Box 536307, Orlando, Florida 32853-6307, e-mail: [email protected], Ph: (407) 898-3823.

TRISP 2010 Due to the short period between the European Plasma Winter Conference 2009 in Graz (February 15–20, 2009; www.winterplasmagraz. at) and the proposed Trends in Sample Preparation (TRISP) 2009 in Seggau Castle, it has been decided to postpone the TRISP Conference for one year. TRISP 2010 will take place in Seggau Castle from June 27 to July 1, 2010. SAS-CHICAGO WORKSHOP The SAS-Chicago section presented a workshop entitled ‘‘Sample Preparation and Introduction for Atomic Spectroscopy: AA, ICP and ICP-MS’’ on Thursday, May 8, 2008, at The McCrone Group, 850 Pasquinelli Drive, Westmont, IL. A workbook of the presentation slides was provided to all attendees. The agenda for the workshop was: ● Atomic Spectrometry: Overview and New Developments, Dr. Jon Carnahan, Northern Illinois University ● Metal Digestions Made Easy (Hot Block), Les Orr, Environmental Express, Inc. ● A ‘‘How-To’’ Practical Approach to Microwave Digestion, Elaine Hasty, CEM Corp. ● Better, Faster, Stable-er Standards Preparation, Dr. Tom Rettburg, VHG Labs, Inc. ● The Role of the Nebulizer in ICP Sample Introduction, Dr. Geoff Coleman, Meinhard Glass Products ● Specialized Introduction Techniques for Liquid and Solid Samples Using ICP-AES and ICP-MS Detection, Dr. Fred Smith, CETAC Technologies ● The Role of Spray Chamber Temperature in ICP Spectrometry, Jerry Dulude, Glass Expansion, Inc. ● Laser Ablation Fundamentals and Applications for ICP-AES and ICP-MS, Mike Colucci, New Wave Research The Chicago Section would like to thank all those who participated! IN CELEBRATION . . . The following is an excerpt from a poster presented at PittCon 2008 by Marvin Margoshes and Leopold May entitled ‘‘In Celebration of the 50th Anniversary of the Society for Applied Spectroscopy: A Brief History of the Early Years.’’ During World War II there was a rapid increase in the use of spectroscopy, to meet the need for the rapid manufacture of war materials. New kinds of spectroscopic instruments were developed. They became commercial products soon after the War, including the Beckman IR1, the Baird IR and Emission Spectrometers, and the ARL

Quantometer. Many scientists became involved in spectroscopy with little or no prior experience. Many were not eligible for the American Chemical Society as they were trained as physicists. Those who were trained as chemists were not eligible to join the American Physical Society. Local groups formed so that the members could share their knowledge. For example, the group that later became the New York Section of SAS had its first meeting in the summer of 1945, as WW II neared its end. Shortly afterwards, the Baltimore-Washington Spectroscopy Society was started, including some who had moved from New York to the Baltimore Washington Area. In 1954 a number of spectroscopists, recognizing the existence of several regional societies of spectroscopists, organized a committee to promote a stronger exchange of information among these societies. The efforts of this committee resulted in the formation of the Federation of Spectroscopic Societies on March 1, 1956 in Pittsburgh, Pennsylvania. The Federation was made up of nine local groups. On March 7, 1957, the Federation held a meeting of the representatives of the member societies in Pittsburgh. The new officers elected were Mr. William J. Poehlman, President; Mrs. Sarah Degenkolb, Vice President; and Rev. James J. Devlin, S.J., Secretary-Treasurer. The major business of that meeting was the appointment of a fact finding committee to consider the advisability of founding a National Society of Applied Spectroscopy. After due inquiry, this committee reported a widespread interest in forming a national society and proceeded to draft a constitution. The enthusiastic response to the proposal of a national society and the expressed willingness of the New York area group to relinquish its name and journal culminated in the foundation of a National Society for Applied Spectroscopy at a meeting of the Federation of Spectroscopic Societies at the first Eastern Analytical Symposium in New York, on November 4, 1958. One of the first matters to be debated was choosing an appropriate name. The proposal to replace the word ‘‘Federation’’ with ‘‘Association’’ was rejected with not much debate after it was pointed out that the new initials for the society would be ASS. The New York SAS graciously gave its name and journal, Applied Spectroscopy, to the new national society. There were seventeen Founding Sections: Baltimore-Washington, Chicago, Cincinnati, Cleveland, Delaware Valley, Detroit, Indiana, Milwaukee, New England, New York, Niagara Frontier, Northern California, Ohio Valley, Pittsburgh, San Diego, Southeastern, and Southern California. The first issue of Applied Spectroscopy was in 1945, as a newsletter of the just-formed New York area group. It was in a newsletter format, but it included technical articles, mostly of a tutorial nature. After Vol. 6 it took on a journal format. The national society assumed the responsibility for publishing the journal with Vol. 12 in 1960. Dr. Frederick Strong III, who was editor at the time of transfer, continued in that capacity until 1961. The logo was designed in 1960 by Rockwell Kent, III. 2009 CRAVER AWARD NOMINATIONS The Coblentz Society announces its first solicitation of nominations for The Craver Award In Recognition of Young Investigators in Applied Analytical Vibrational Spectroscopy. The Coblentz Society created this award to recognize young individuals who have made significant contributions in applied analytical vibrational spectroscopy. The Craver Award is presented annually to an outstanding young molecular spectroscopist whose efforts are in the area of applied anaAPPLIED SPECTROSCOPY

151A

lytical vibrational spectroscopy. The candidate must be under the age of 45 on January 1st of the year of the award. The work may include any aspect of infrared (NIR, MIR, or Far), and/or THz, and/or Raman spectroscopy in applied analytical vibrational spectroscopy. The nominees may come from an academic or government lab, or industrial backgrounds. The award carries with it a $2000 honorarium and a plaque, plus a $500 travel allowance. Files of candidates will be kept active until the age of eligibility is exceeded. Annual updates of candidate files are encouraged and will be solicited from the nomination source by the Award’s committee chair. Nominations for 2009 must include a detailed description of the nominee’s accomplishments, a curriculum vitae or resume, and a minimum

152A

Volume 62, Number 6, 2008

of three supporting letters. Nominations for 2008 close on July 31, 2008. This award will be presented at the annual FACSS meeting. The awardee will be offered a 25-minute plenary lecture to the assembled FACSS Conference. Further, a separate half-day award symposium honoring the award recipient and highlighting the interests of the awardee will also occur at the same conference. Please send nomination packages to: Dr. Mark A. Druy Physical Sciences Inc. 20 New England Business Center Andover, MA 01810 Or by E-mail (preferred): [email protected]

Alexander Scheeline, Editor Please forward book reviews to the Book Review Editor, Alexander Scheeline, Department of Chemistry, University of Illinois at Urbana-Champaign, 600 S. Mathews Ave., Urbana, IL 61801.

Note from the Editor. It is unusual to review a biography in Applied Spectroscopy, particularly when a majority of the content is not focused on technical matters. However, since Howard V. Malmstadt is a scientific ancestor of at least 5% of the Society’s membership (including this editor and the book’s reviewer), it seemed appropriate here.—A. Scheeline Into the Light: The Academic and Spiritual Legacy of Dr. Howard Malmstadt. J. Feaver. Youth with a Mission Publishing, Seattle, WA, 2007. Pp. 128. Price: US$11.99. ISBN 1-57-658411-9. I found this book an enjoyable read. As the title implies, the book breaks down into two parts of Dr. Howard Malmstadt’s life, although it is obvious that these parts make up the whole of the late Dr. Howard Malmstadt. While the author of this biography (John Feaver) is more personally familiar with Dr. Malmstadt’s ‘‘spiritual legacy,’’ he does a nice job of recounting Malmstadt’s years as a cornerstone in modern analytical chemistry while he was a faculty member at the University of Illinois (1951–1977). In addition to Malmstadt’s scientific accomplishments and his pioneering work in teaching electronics and ‘‘modular instrumentation,’’ he is also recognized as the academic father, grandfather, etc., of several generations of leading spectroscopists and instrumentalists in the US analytical chemistry community. In the book, insights into his teaching and mentoring philosophies are smoothly revealed through conversations with former students as well as through anecdotal incidents relayed by these individuals. Judging by the success of his research program, his contributions to modern analytical chemistry and the legacy of scientists that he has sponsored, the book serves as a subliminal tutorial on people and research management. I found it interesting that his encouragement of group problem solving and collaborative research preceded by more than two decades what is now considered to be the modern research model. I particularly like the recollection of Camille Bishop, who quotes Howard as having said, ‘‘I function by identifying what needs to be done, explain why and let the people pick up on it. I make suggestions along the way and see if they pick up on them. That is one of the ways of seeing if you have people who are capable of moving into leadership; they really pick up on things.’’ After leaving U of I in 1977, he became one of the founders of a Christian-based university in Hawaii that was ultimately named University of the Nations. Here he served as a faculty member and enabling force for both academic and research activities. As an example, his recognition of the need for clean water in many underdeveloped parts of the world is an interesting case in point of engaging others in a humanitarian project with realistic ‘‘ideals’’ such as the need for cost-effective purification. His enthusiasm for the project served as a catalyst to involve other talented individuals in the project, a genius he displayed throughout his scientific career. JAMES A. HOLCOMBE Department of Chemistry and Biochemistry University of Texas at Austin Austin, TX 78712

Vibrational Spectroscopy of Polymers: Principles and Practice. N. J. Everall, J. M. Chalmers, and P. R. Griffiths, Eds. John Wiley and Sons, New York, 2007. Pp. 586. Price: US$260. ISBN 0-470016-62-0. This book of 17 chapters is a collection covering the application of vibrational spectroscopy to solving fundamental and applied problems in solid-state polymer science. Many of the chapters were originally part of the five-volume set Handbook of Vibrational Spectroscopy, edited by J. M. Chalmers and P. R. Griffiths, and have been updated and revised, while several chapters are entirely new. The book is written to an audience that must be versed in polymer science, as it does not cover the basics of polymer science. It is also highly technical and often quite theoretical in its presentation of vibrational spectroscopy, and it may be best used by someone who has already attained a level of expertise in the application of vibrational spectroscopic tools. It serves as a good reference for someone in the field who would like to increase his understanding of specialized spectroscopic tools or expand his knowledge about the application of tools that he may already be familiar with. Since many techniques and polymer systems are covered, this book may contain too much material to allow it to be used for a graduate course textbook either in spectroscopy or polymer science. The topics covered include specific techniques such as dichroic and trichroic orientation measurements by IR and Raman spectroscopy, infrared linear dichroism, dynamic infrared linear dichroism, and rheo-optical Fourier transform infrared spectroscopy. The chapter on optical and dielectric properties of polymers by D. M. Smith and A. R. Chugtai includes a good summary of ellipsometry. There are also chapters devoted to specific classes of polymers and how vibrational spectroscopy has been used in studies of such systems. These include conducting polymers, rubbers, and composites. Missing is work specifically addressing the study of polymer blends, copolymers (though they are considered under composites), and any solution properties of polymers. The book could have been titled ‘‘Solid-State Spectroscopy of Polymers’’. The book addresses mostly mid-infrared and Raman spectroscopy, with some passing mention of near-infrared spectroscopy. This is not surprising, as in research it is mostly the two former that have been used to elucidate structure–property relationships in polymeric materials. J. M. Chalmers’ chapter on spectral interpretation is a good overview, providing useful guidance on how to approach spectral interpretation. He also addresses some of the complexities of polymer structure that make polymer spectral interpretation difficult (i.e., though it may be counter intuitive, polymer tacticity is not very well studied by vibrational spectroscopy). Coupled with the first chapter in the book by J. M. Chalmers and N. J. Everall, which covers qualitative and quantitative analysis of polymers, one quickly understands the utility of vibrational spectroscopy in solving industrial problems with polymers. These two chapters can serve as a useful guide to practitioners who are new to the application of spectroscopic tools to studying polymers. A chapter by M. Papini-Arconcada and F. Papini on the measurement of thermal and solar properties of polymers is extremely theoretical, written in technical language that can be difficult to follow. The chapter covers solar radiation and provides information on thermal properties, but does not relate these to polymers. It is only in the last section of the chapter that polymers are addressed; but only in passing, as cover systems. Zerbi’s chapter on conducting polymers is a thorough and engagAPPLIED SPECTROSCOPY

153A

(Continued)

ing treatment of both the polymer science and the spectroscopy of this class of materials. This is followed by a chapter by Furukawa on fundamentals and applications of spectroscopy to conducting polymers, in which some specific examples of Raman and IR spectroscopy of doped conducting polymers are given. The chapter on the rheo-optical spectroscopy of polymers, from H. W. Seisler’s group in Essen, Germany, with M. Zahedi in Iran is one of the few chapters that discuss the impact of spectroscopic measurements on the understanding of the structure–property relationship of polymers. It is an excellent treatment of the subject, covering experimentation, examples, and the significance of the studies to polymer science. The layout of the book does not have an obviously logical flow. Perhaps the theory of vibrational spectroscopy of polymers would have been better placed at the start, followed by an ordering of chapters by technique, then by specific polymer systems studied. The chapters are not numbered, and when reference is made to other chapters in the book, sometimes the title is given in bold type, but not always, making it confusing to understand the significance of this formatting. For the reader, a detailed table of contents would make it easier to navigate the book and find topics of particular

154A

Volume 62, Number 6, 2008

interest. It would be helpful to have better cross-referencing amongst the chapters (which would be easier if chapters were numbered). Each chapter also has its own set of abbreviations at the end, which may have been easier on the editors, but takes away from the feeling that the volume is a cohesive piece of work. A general list of acronyms of at least the common abbreviations for polymers could have been provided. An additional confusion comes about from the fact that not every author defines the acronyms in the same way, though they come to mean the same thing (i.e., IRAV—IR active vibrational and infrared activated bands; LAM longitudinal acoustic mode and longitudinal accordion mode). The book does contain valuable information on vibrational spectroscopy and on polymers. Some of the chapters will serve as a useful guide to industrial scientists who wish to apply spectroscopic tools to studying polymers. The majority of the chapters cover very specific techniques, which at this time may be more suited to academic studies of either spectroscopic techniques or polymeric systems. KATHERINE A. BAKEEV GlaxoSmithKline 709 Swedeland King of Prussia, PA 19406

accelerated paper

The Influence of Out-of-Focus Sample Regions on the Surface Specificity of Confocal Raman Microscopy NEIL EVERALL Intertek MSG, Wilton, Redcar, Cleveland, TS10 4RF, UK

This paper considers the quantitative implications of out-of-focus regions on the lateral and depth resolution of Raman microscopy, with special regard for the surface specificity of the technique. It builds on work that has recently appeared in the literature which shows that with transparent samples, signals can originate throughout a large extended illumination volume, even though most of this region is out of focus with regard to the confocal aperture. This gives rise to weak but readily detectable spectral contributions from regions that are tens of micrometers from the point of tightest focus, an effect that is easily demonstrated if the laser is focused far above the sample surface. When we integrate the signals arising throughout this extended volume, the resulting total signal can be significant with respect to the Raman signals originating from the point of focus; this has obvious implications for surface specificity and depth resolution. Furthermore, as one moves the focal point through and above a sample surface, signals from thick transparent samples decay relatively slowly compared with thin or opaque ones, where the extended focal volume is irrelevant. This means that on moving above the surface of a thinly coated thick substrate during a confocal axial scan, the substrateto-coating signal ratio increases dramatically, contrary to intuition. Consequently, confusing spectral artifacts arise if one focuses above the sample surface, either inadvertently when mapping an uneven sample, or deliberately in an attempt to improve surface specificity. In this work we show how a simple analytical model can predict the surface/substrate signal ratio as a function of distance above the surface. The model is validated using experimental data from monofilms and coated films. Furthermore, we show how this effect is not limited to the confocal axial profiling geometry. Similar effects are obtained when one scans laterally beyond the edge of mechanically prepared cross-sections due to an extended, out-of-focus laser field that can sample lateral regions far to the side of the optimum focus. This effect can lead to very confusing results, such as spectra from the substrate increasing in absolute intensity as one moves beyond the edge of the coating into the air. These observations, which as far as we are aware have not previously been reported, are rationalized using a simple ray-tracing description, which shows the potential for coupling light into the cross-section, which acts as a waveguide. These effects have a completely different origin than the wellknown anomalies that are introduced by refraction and spherical aberration; even with a perfect, aberration-free system, the extended focal volume may cause significant degradation in depth resolution. Although the effects have been demonstrated with simple film systems, they have the potential to impact the results from Raman mapping and imaging of any samples that contain significant refractive index discontinuities, which can potentially cause refraction and waveguiding, or that have compositional depth gradients and an uneven sample surface. Received 27 February 2007; accepted 3 April 2008. E-mail: [email protected].

Volume 62, Number 6, 2008

Index Headings: Confocal Raman microscopy; Extended focal volume; Out-of-focus contributions; Depth resolution; Spatial resolution.

INTRODUCTION Factors Affecting Spatial Resolution in Confocal Raman Microscopy. Confocal Raman microscopy is routinely used in two configurations to obtain depth-resolved structural information (Fig. 1). The first, commonly known as axial, optical, or confocal depth profiling, uses a laser beam incident normal to the sample surface, and a series of spectra are obtained by incrementally displacing the laser focal point along the surface normal. Provided it is transparent, the sample can be analyzed without modification. The second arrangement, termed lateral profiling, is inherently destructive, because one has to mechanically cut or polish a cross-section, which is analyzed by acquiring a line map across its surface. Interpreting results from a lateral scan is simple because it is obvious where the laser beam is positioned on the cross-section. In contrast, considerable effort has been expended in recent years in understanding how to convert the depth scale of a confocal Raman profile into an absolute, accurate depth scale. Most of this work has concentrated on the artifacts that are introduced when using high numerical aperture metallurgical microscope objectives to focus below the surface of transparent samples. These artifacts, which arise from refraction and spherical aberration, are manifested by degradation in depth resolution combined with significant compression in the apparent depth scale. A recent paper described the effect of spherical aberration in detail and reviewed the key work that appeared in this field up to the end of 2006.1 This paper also gave a quantitative comparison of the performance of a number of microscope objectives that were designed to minimize spherical aberration and showed how axial profiling with a 1.4 NA oil immersion objective gave much better depth resolution and signal intensities than did a 0.9 NA metallurgical objective. This suggests that even though recent work has shown that it is possible to correct data from metallurgical objectives to provide accurate confocal depth scales,2,3 the raw data will have lower signal intensity, poorer signal-noise ratio, and poorer depth resolution compared with data from an immersion objective. To summarize, the effects of spherical aberration can be avoided by selecting an appropriate objective or the distorted

0003-7028/08/6206-0591$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

591

FIG. 1. Schematic showing (a) axial (confocal) and (b) lateral scanning approaches to depth profiling.

data from a metallurgical objective can be processed to give the correct depth scale. Furthermore, when focusing directly onto the surface of a sample with a metallurgical objective, spherical aberration should not be an issue. With this in mind, one might suppose that confocal Raman microscopy can, under ideal circumstances, approach the often quoted ‘‘text-book’’ performance of ;1 lm3 volume resolution. Unfortunately, this is not the case in practice. For example, when focusing an oil immersion objective onto the surface of a 20 lm poly(ethylene) (PE) coating on a poly(ethyleneterephthalate) (PET) film, strong Raman signals from the PET substrate were clearly observed superimposed on the PE spectrum, even though the PET surface was far from the optimum focal plane of the microscope.1 In another experiment, focusing a 0.9 NA metallurgical objective onto the surface of a 2 lm thick poly(vinylidenechloride) (PVDC) coated PET substrate gave a Raman spectrum that was dominated by PET signals, showing that the depth resolution was considerably worse than 1 lm.4 Finally, and most surprisingly, when analyzing an axial confocal depth profile of a thin (;5 lm) coating of poly(ethylenenaphthoate) (PEN) on thick PET, it was found that the PET/PEN band intensity ratio actually increased on focusing the laser above the surface, i.e., into air.5 At first sight the latter result seems counterintuitive, as one might expect the coating intensity to increase relative to the substrate upon moving upwards. The observation was rationalized at the time by suggesting that the effect arises from laser and Raman rays that propagate along the paraxial axis (Fig. 2a). When focusing above the coating, only the paraxial rays can pass through the confocal aperture, and they preferentially sample the substrate because it is thicker than the coating. No attempt was made to model the expected magnitude of this effect on a theoretical basis. Macdonald and Vaughan6 recently rectified this matter. They reported a model that shows that the paraxial rays alone cannot account for the signals originating from the material that lies far from the focus of the laser beam. The basis for their model is shown in Fig. 2b, which shows the cone of illumination that is generated within a sample by the focused laser beam (in this case having a point of tightest focus, P, located on the sample surface). The key aspect of the model is that the laser produces an extended illumination volume, and we have to consider the fate of rays that are generated at all points within this volume, not just at point P. For example, any point on the paraxial axis

592

Volume 62, Number 6, 2008

FIG. 2. Origin of out-of-focus contributions to Raman signals. (a) Paraxial rays sample deep within a sample but account for little intensity. (b) Within an extended illumination volume, certain rays from any point within the volume can propagate back through the focal point P, and these rays will pass through the confocal aperture.

(e.g., A) can scatter a Raman ray directly upwards, which will pass through point P and hence through the confocal aperture (corresponding to the simplified situation depicted in Fig. 2a). However, off-axis points such as B and C can also emit rays that will pass through point P and then be collimated by lens L and transmitted through the confocal aperture. In fact, every point within the entire illuminated volume will scatter some Raman rays that will pass through the focal point P. While the number of rays that pass through P will be very small compared to the total number of rays that are scattered from a given point, and will decrease with distance from P, the net contribution from all points integrated over the whole focal volume is significant. An analogous argument can be made for points that lie above the focal plane (a situation that is relevant when focusing below the sample surface). In this case, any ray path that passes from point P and through the point of interest (e.g., point E in Fig. 2a) will pass through the confocal aperture, so rays generated at point E that are collinear with this path will be accepted by the aperture. Macdonald and Vaughan produced a quantitative model for this effect and showed (1) that it can account for the quantitative variation in signal intensity observed when axially depth profiling below the surface of thick samples of differing transparency, and (2) that it predicts the existence of detectable signals from material located up to 40 micrometers below the focal point P. This latter point explains the observations that we have described above, which are consistent with other reports of high quality confocal Raman data being obtained when focusing tens of micrometers above the sample surface.7 This is important work that should have widespread influence on the interpretation of confocal Raman data. While the simulation discussed by Macdonald and Vaughan clearly highlights the physical origin and importance of signals from the out-of-focus domains, and correctly predicts their magnitude, it is important to note that other workers have also quantified the effect of object planes that are quite distant from the laser focus. For example, Bridges et al.8 used confocal Raman microscopy to depth profile polystyrene (PS) beads immersed in a solution of perchlorate ion and showed how a simple Lorentzian model for the microscope response could account for the observed signal variations, particularly the ratio

FIG. 3. Decay of signals when focusing above a sample surface, recorded for (a) silicon and (b) thick PET. The signals from a thick transparent sample decay more slowly because they are integrated over a greater proportion of the extended illumination volume.

of PS to perchlorate signal. This model is very easy to code and will be used to predict the spectral variations that should occur on moving the laser focus above a sample. Objectives of the Study. The objective of this work is to illustrate and quantify the importance of the contribution of signals from the ‘‘out-of-focus’’ regions using well-defined coated polymers; we will show how even with quite thick coatings, the substrate signal can still dominate the spectrum of the coating in the axial scanning configuration (Fig. 1a). Consequently, we will study how the surface/bulk sensitivity changes as a function of distance above the surface, using samples having much thicker coatings than those described in our previous work. However, these studies will then be extended to the lateral scanning configuration (Fig. 1b), where we will show that the lateral resolution is also adversely affected by the out-of-focus regions, and some very surprising results are obtained near and beyond the edge of the crosssection. As far as we are aware, this observation has not been reported previously in the open literature and could have ramifications for the interpretation of Raman maps.

EXPERIMENTAL Instrumentation. Raman spectra were acquired using a Labram confocal Raman spectrometer (Horiba-JY) fitted with a HeNe laser (633 nm) and 1200 grooves/mm gratings. The slit width was fixed at 100 lm, giving a nominal spatial resolution of 6 cm1. We used an OlympusTM metallurgical objective (1003, 0.9 NA (MPlan 1003)) and a confocal aperture of 200 lm throughout this work. Samples. Three polymer film samples were used to characterize microscope performance in both the axial and lateral scanning modes. The first was a bilayer polymer film, consisting of a thick (;100 lm) poly(ethyleneterephthalate) (PET) core, which was coated on one side with a thin (;15 lm) layer of polyethylene (PE). The second was a thick (100 lm) PET film that was coated with a mixed-acrylate resin that was UV cured to form a hard coating of ;20 lm thickness. These were both research samples previously provided by the former Films business of ICI PLC. The third polymer sample was a commercial PET monofilm of 100 lm thickness, supplied by DuPont-Teijin Films (Wilton, UK). A silicon wafer

FIG. 4. Plot of PET/silicon intensity ratio as a function of distance above the surface. This shows how the bulk/surface sensitivity increases approximately linearly with distance above the sample.

was also used to generate a confocal response from a strong absorber, from which almost all the Raman scatter should be generated near the surface. Confocal axial depth profiles were obtained by securing the film samples (coated side uppermost) to glass microscope slides using double-sided adhesive tape and recording spectra at incremental displacements along a line normal to the sample surface. In each case, position zero was defined as the coating/ air interface and negative displacements denote moving the laser focus upwards into the air. Samples for lateral line scans were obtained by cutting cross-sections from each film using a glass-bladed microtome and laying the section onto a glass microscope slide. Again, position zero was defined as the coating/air interface, with positive displacements defining motion into the coating.

RESULTS Simple Demonstration of Influence of Out-of-Focus Contributions. In a previous paper we demonstrated that the sample that is most commonly used to measure the depth resolution of a confocal Raman microscope, a bare silicon wafer, gives somewhat misleading results.1 This is because it does not accurately mimic a typical sample for confocal microscopy, where the regions of interest could be buried beneath a reasonably thick transparent overlayer. The fact that silicon is almost opaque to visible radiation means that it does not respond to out-of-focus radiation in the same manner as a thick transparent sample, which leads one to overestimate the depth resolution of the microscope. This is shown by Fig. 3, which shows axial depth profiles for a silicon wafer and a 100 lm thick (clear) PET film. The profiles, which were generated using the intensities of the 520 cm1 (Si) and 1612 cm1 (PET) bands, start from the sample surface and move upwards (into the air). The Si response indicates that the out-of-focus contribution fell by 203 and 1703 at 5 and 10 micrometers above the sample, but for PET the out-of-focus contribution only diminished by ;113 and 323 at the same heights—a much slower decline—because one samples more of the extended illumination volume with a thick transparent sample. Figure 4 plots the ratio of the PET (1600 cm1) and silicon (520 cm1) signals as a function of distance above the surface, normalizing the ratio to unity at the surface. Assuming that the

APPLIED SPECTROSCOPY

593

FIG. 5. Spectra from confocal depth scan through 15 lm coating of PE on 100 lm PET substrate. The substrate signal dominates above the sample surface.

silicon wafer acts purely as a surface source, this plot measures the bulk/surface sensitivity ratio as a function of distance above the sample. As expected, this ratio increases with distance because the signals from the bulk fall more slowly than those from the surface. This observation alone clearly demonstrates the need to account for an extended illumination source; if the Raman signal originated solely from a 1 lm focal plane, irrespective of sample thickness or clarity, the surface and bulk signals would both decline at the same rate. To summarize, on moving above the surface, the signal from a thick transparent sample decays significantly more slowly than that from a surface source. Polyethylene/Poly(ethyleneterephthalate) Laminate. Figure 5 shows spectra recorded at various displacements from the PE surface. The spectra were extracted from a confocal axial depth profile at the positions shown in the figure; the focal positions were corrected for refraction and should therefore approximate the true position of the laser focus within the sample. Each spectrum was normalized to give approximately equal heights for the PET ring stretching band near 1612 cm1. When focusing at þ20 lm, i.e., into the PET, a strong PET

FIG. 6. Confocal profiles for PE-coated PET (see Fig. 5). The profiles are offset for clarity. The PET signal decays more slowly than the PE response, and hence the PET/PE ratio increases with distance above the coating.

594

Volume 62, Number 6, 2008

FIG. 7. Schematic showing potential for out-of-focus contributions during lateral scanning.

spectrum was obtained. At þ10 lm (i.e., inside the PE) there were several strong bands from the PE, but there was a significant signal from the PET, which was maintained on moving the focus upwards through the PE onto the coating surface. However, on focusing 5 lm and 10 lm above the coating, the PE bands were diminished relative to the PET spectrum, confirming our earlier observations using more thinly coated PET films.4,5 This shows how substrate features can strongly influence the spectra of quite thick coatings; this is particularly apparent when focusing above the coating surface, but as Fig. 5 shows, the effect is also present when focusing into the coating. Figure 6 plots the PE, PET, and PET/PE signals as a function of depth; the profiles have been offset for clarity. It shows that (1) the PE/PET contrast was maximized ;1–2 lm below the surface of the coating, (2) the PET signal fell more slowly than the PE signal, and (3) the PET/PE ratio increased with distance above the coating surface. These observations are consistent with the data shown in Figs. 4 and 5. Note that in Fig. 6 the depth scale has not been corrected to account for refraction, so the PE layer appeared to be only ;8 lm thick rather than 15 lm; this is due to the depth scale compression that occurs when using metallurgical objectives.1 Lateral Scanning. We now ask whether similar issues arise with lateral scanning of mechanically prepared cross-sections (i.e., mode depicted in Fig. 1b). Figure 7a shows a schematic of the situation when the focus of the laser beam is displaced laterally beyond the edge of the sample. In this case, with a thin section, it is clear that none of the rays from the objective intersect the sample and so one should expect zero Raman signal from coating or substrate. In contrast, Fig. 7b shows an edge-on view of a case where the cross-section is sufficiently thick to allow illumination of the substrate even though the nominal laser focus is positioned beyond the air/coating interface. In this case there is at least the potential to observe signals from both the coating and the substrate. Figure 8 shows

FIG. 8. Spectra extracted from a lateral scan across the cross-section cut from PE-coated PET.

several Raman spectra extracted from a lateral line scan over a ;25 lm thick cross-section of the PE/PET laminate; the spectra have not been normalized, so the relative intensities of the different spectra are displayed correctly in this figure. When focusing ;13 lm into the PE (i.e., ;2 lm from the PET/PE interface), strong PE and PET bands were observed (PET bands at ;1615 and 1726 cm1 are denoted by the asterisks). At 8 lm into the PE (7 lm from the PET) the PE bands are strong, but the PET bands are clearly visible. At ;3 lm into the PE, the spectrum was dominated by PE, although weak PET bands were still detected. This is perhaps not surprising because PET is a much stronger Raman scatterer than PE. However, on focusing just 1 lm beyond the edge of the PE (i.e., into the air), we see that the PE signal dropped, as expected, but the PET bands actually increased in intensity! Furthermore, moving ;7 lm beyond the edge of the PE layer, the PET bands were still clearly visible, while the PE bands had vanished. Figure 9 shows the intensity variation of the PE and PET bands as a function of position in the line scan. The data represent the baseline-corrected integrated intensities of the 1296 cm1 (PE) and 1612 cm1 (PET) bands. The profiles were not scaled or normalized, but they are offset for clarity. Figure 9 highlights the fact that while the coating signal fell dramatically on moving the laser focus into the air, the substrate signal increased significantly in the same region. At first sight this is very surprising indeed; why should the substrate signal rise on moving the laser focus beyond the coating? It is certainly different from the situation with confocal axial scans, where the signal from the substrate decreases on moving above the coating, albeit more slowly than the coating signal itself. In order to check that this was not an anomalous result with one particular cross-section or sample, the experiment was repeated using a cross-section cut from the UV-cured acrylate-coated PET sample. Figure 10 shows that the same result was obtained; the substrate band increased on moving the laser focus beyond the surface of the coating. The reasons for this behavior are discussed below.

FIG. 9. PE and PET intensity profiles from a lateral scan across the PE-coated PET cross-section. Note that the substrate signal increases on moving beyond the edge of the coating into the air.

analytical model based on the work of Travis et al.8 These authors showed that in a situation where the projected size of the confocal aperture in the object plane (s0) is greater than the width of the laser focus (w), the Raman detection efficiency E(z) is given by Eq. 1: EðzÞ ’

1 ½1 þ ðz=lÞ2 

ð1Þ

where z is the distance from the optimum focal plane, l ¼ s0 cot(a), s0 ¼ s/M, M is the magnification of the objective, and s is the physical size of the confocal aperture. Since w is given, in the diffraction limit, by 0.61k/NA (which equates to ;0.43 lm for our system), and because s0 ¼ 200/100 ¼ 1 lm, we are justified in using Eq. 1. We make the additional assumption that the attenuation due to scattering and absorption by the sample is the same for both the laser and the Raman scattered wavelengths but can differ from sample to sample. The attenuation coefficient is denoted k. Figure 11a describes the situation for a sample of thickness T1 that is displaced from the

DISCUSSION Modeling the Axial Depth Profile. The confocal profiling data discussed above can be readily explained using a simple

FIG. 10. Intensity profiles from a lateral scan across UV-cured acrylate-coated PET. Note that the substrate signal increases on moving beyond the edge of the coating into the air.

APPLIED SPECTROSCOPY

595

FIG. 13. Comparison of predicted and observed PET/Si ratio versus distance above the surface. Excellent agreement was obtained.

FIG. 11. Schematic showing the calculation of variation in signal as a function of sample displacement (D) above the optimum focus (z ¼ 0).

Figure 11b shows the situation for a two-layer system, where the coating has thickness T1 and the substrate has thickness T2. The total signal from the second layer (i.e., the substrate) is given by Eq. 4: Z DþT1 þT2 Isub ðDÞ¼ EðzÞexpð2k1 T1 Þexp½2k2 ðz  D  T1 Þ dz DþT1

optimum focus by a distance D. The signal that is detected from depth z is given by Eq. 2: Rðz; DÞ ¼ EðzÞexp½2k1 ðz  DÞ

ð2Þ

The total signal from the layer as a function of displacement D is given by Eq. 3: IðDÞ ¼

Z

DþT1

EðzÞexp½2k1 ðz  DÞ dz

ð3Þ

D

FIG. 12. Comparison of predicted and observed Si and PET signal variation with distance above the surface. Reasonable agreement was observed between the predicted and observed values. However, difficulty in determining the exact position of the surface in the experiment affects the quality of the fit at low displacements.

596

Volume 62, Number 6, 2008

ð4Þ Using Eq. 3 we can calculate the variation in signal as a function of height above any sample of known thickness and attenuation index. Equations 3 and 4 allow us to predict the change in signal from both layers of a two-layer film as a function of height above the surface of the uppermost layer. When used for this purpose, the signal from each layer must be scaled to account for the relative Raman scattering crosssection of each material. Comparison of Predicted and Measured Responses. Figure 12 compares the predicted and experimental results for the silicon wafer and 100 lm (uncoated) PET films, calculated on the assumption that kSi ¼ 0.3 lm1 and kpet ¼ 0.008 lm1. The absorption index for Si at 633 nm was taken from the Energy Citations Database,9 and the PET value was scaled in order to find the best fit to the experimental data, particularly the ratio of the PET and Si responses. The fit between the observed and predicted responses is quite good, particularly when one considers that there is an inevitable error in defining the position z ¼ 0 in the experimental data. Figure 13 shows the predicted ratio of the PET and silicon responses as a function of displacement above the surface and overlays the experimental data from Fig. 4. Agreement is reasonable over the whole range, although the experimental data do not necessarily support the slight nonlinear variation that is expected from the model (a straight line fit gave R2 ¼ 0.94). The model was also used to predict the bulk/surface response shown in Fig. 6 for the 15 lm coating of PE on 100 lm PET (Fig. 14). The model gave a good fit to the experimental data, provided we assume that the PET is a stronger Raman scatterer by an order of magnitude than PE, which is certainly plausible. We cannot check the relative scattering cross-sections because this would require access to PET and PE films of the same

FIG. 14. Comparison of predicted and observed PET/PE ratio as a function of distance above the surface. Good agreement was obtained on the assumption that PET is a 353 stronger Raman scatterer than PE. Values of 0.001 and 0.008 lm1 were taken for the attenuation coefficients of PE and PET.

orientation and crystallinity as the layers present in the laminate; the absolute scattering intensity will be extremely sensitive to both of these factors. Figures 12 through 14 demonstrate that the Raman response as a function of distance above the sample surface can be predicted quite well using the simple Lorentzian model coupled with reasonable assumptions regarding attenuation coefficients and relative scattering efficiencies of the materials involved. Lateral Mapping. It is now necessary to consider the strange results shown in Figs. 8, 9, and 10, i.e., the fact that when performing a lateral line map across a cross-section, moving the laser beam beyond the edge of the coating into the air causes the signal from the substrate to suddenly increase. Figure 15 presents a plausible explanation for this effect. When the laser beam is focused into the coating layer (Fig. 15a) there is a finite displacement beyond which none of the laser rays can penetrate the substrate. Thus, the substrate signal falls rapidly as the laser focus moves towards the coating/air interface. However, when the laser focus is displaced into the air (Fig. 15b), refraction occurs when the rays strike the edge of the coating, and the rays refract towards the substrate layer. Provided that the cross-section is sufficiently thick, some of these rays will enter the substrate and generate Raman photons. Furthermore, the cross-section could act as a waveguide for those laser rays that strike its lower surface at greater than the critical angle; this will further enhance Raman photon generation within the substrate in preference to the coating, since the entire length of the cross-section could support a guided mode. As we have seen above, there is a finite probability that the Raman photons will emerge from the sample and pass through the laser focal point, thereby being collected and detected. We note that waveguiding can only be excited when the laser is focused beyond the edge of the crosssection (Fig. 15b); phase matching conditions prevent waveguiding when focusing onto the surface of the cross-section (Fig. 15a). This hypothesis explains why the substrate signal initially increases in intensity on first moving the focus into the air, whereas the coating signal falls abruptly. This effect is

FIG. 15. Schematic suggesting a possible explanation for the increase in substrate intensity on moving beyond the edge of the cross-section. At position (a) no rays penetrate the substrate, but at (b) refraction causes rays to pass into the substrate, generating signal. Subsequent waveguiding would further enhance the substrate/coating signal ratio.

particularly noticeable when the substrate is a strong Raman scatterer and the coating scatters weakly. This is the case for both of the examples discussed above, i.e., PE and cured acrylate on PET. If the hypothesis outlined in Fig. 15 is correct, then one should find that the cross-section thickness will be an important factor. If the section is too thin, refracted rays will not enter the substrate. If it is too thick, waveguiding will not occur. The influence of the thickness of the cross-section will be investigated in the future, but it seems likely that in order to obtain the best lateral resolution one should work with the thinnest possible cross-section. Relevance to Practical Raman Microscopy. The author is aware that readers may consider these experiments to address rather extreme situations that would not be encountered in practice. However, this is not the case. Let us return to the example shown in Fig. 10, namely line-scanning over a crosssection of the UV-cured acrylate-coated PET film. This sample arose from a research project that investigated degree of cure as a function of dose and position within the coating. Measurement of subtle variations in cure, particularly near the coating surface, where O2 inhibition of cure is potentially a problem, was a key requirement of the project. Figure 16 shows three Raman spectra sampled at 1 lm intervals in the region of the air/coating interface. At position 0 (just on the interface) the spectrum is dominated by partially cured acrylate (C¼C stretch at 1635cm1) but also has a weak PET band (*). At position 1 (;1 lm towards the air) we observe both acrylate C¼C and a relatively strong PET ring vibration. At position 2 lm (1 lm further into the air) we only observe PET bands. While the spectrum at 2 lm is obviously an artifact, due to refraction, the spectrum at 1 lm could easily be misinterpreted as containing an acrylate C¼C band at 1615 cm1 rather than correctly assigning this to the erroneous signal being picked up

APPLIED SPECTROSCOPY

597

FIG. 16. Spectra extracted from a lateral scan across the cross-section of 20 lm UV-cured acrylate coating on PET. At the coating/air interface (0 lm) only a weak PET band (*) is seen; at 2 lm (focused into the air beyond the coating) we see a strong PET spectrum, which is obviously an artifact. However, at 1 lm the PET contribution is quite strong and could easily be misinterpreted as another C¼C species. Thus, one needs to be very careful not to misinterpret substrate bands when focused near a coating surface.

from the (distant) PET. In other words, unless one is aware of the very large influence that signals from distant substrates can play near a surface, it would be quite easy to misinterpret Raman line maps. The same is, of course, true for confocal axial scans, where data near a surface will also have heavily weighted contributions from a substrate. Although these results have been produced from well-defined polymer films, the effects are quite general and will potentially influence results obtained from any Raman mapping or imaging experiment where one encounters sample/air interfaces. Mapping situations in which the sample height is not well controlled are particularly susceptible to errors arising from inconsistent contribution of signals from material below the surface. Finally, we note that confusing data should always be expected to arise near sample surfaces, irrespective of whether confocal or lateral profiling is applied. However, if care is taken, one can always obtain better spatial resolution near a surface by choosing the lateral scanning approach, provided one does not accidentally focus just outside the sample. Figure 17 illustrates this point by comparing spectra extracted from confocal and lateral profiles through the UV-cured acrylate coating discussed above. The spectra were extracted from the points on the profiles that gave the best surface/substrate contrast ratio. It is obvious that the lateral scan gave much better suppression of the PET band. This is expected because with the lateral scanning configuration most of the extended focal volume does not overlap with the PET substrate, in stark contrast to the confocal geometry.

CONCLUSION These results show that extreme care must be exercised when obtaining Raman confocal depth profiles, lateral line scans, or maps. Significant anomalies arise because the focused

598

Volume 62, Number 6, 2008

FIG. 17. Spectra demonstrating that lateral scanning provides much better surface specificity than does axial scanning. With the axial scan, the spectrum with the highest surface/substrate contrast had a much more intense substrate band than did the equivalent laterally scanned spectrum.

laser beam generates an extended illumination volume within a sample, and signals can be detected from regions within this volume that are far from the point of tightest focus. If one focuses above the surface of a thick, clear sample, the spectra become heavily biased towards subsurface regions. This is because the substrate signals decay more slowly than surface signals upon moving upwards. If one scans laterally across a mechanically prepared cross-section, similar effects arise, but the substrate features can even increase in absolute intensity upon focusing outside the sample. This effect is only observed for systems in which the substrate is offset from the surface, for example, by a coating, and arises from refraction at the coating/ air interface. With either illumination geometry, it is apparent that it is easy to generate spectral artifacts when analyzing sample surfaces. ACKNOWLEDGMENTS The author acknowledges Intertek PLC for granting permission to publish this work. He would particularly like to thank Drs. Don Clark and Ian Clegg (Pfizer Ltd) for interesting discussions regarding this work.

1. N. Everall, J. Lapham, F. Adar, A. Whitley, E. Lee, and S. Mamedov, Appl. Spectrosc. 61, 251 (2007). 2. J. Pablo Tomba, L. M. Arzondo, and J. M. Pastor, Appl. Spectrosc. 61, 177 (2007). 3. H. Reinecke, S. J. Spells, J. Sacristan, J. Yarwood, and C. Mijangos, Appl. Spectrosc. 55, 1660 (2001). 4. N. Everall, JCT CoatingsTech 2, 38 (2005). 5. N. J. Everall, Spectroscopy 19, 16 (2004). 6. A. M. Macdonald and A. S. Vaughan, J. Raman. Spectrosc. 38, 584 (2007). 7. A. M. Macdonald, A. S. Vaughan, and P. J. Wyeth, J. Raman. Spectrosc. 36, 185 (2005). 8. T. E. Bridges, M. P. Houlne, and J. M. Harris, Anal. Chem. 76, 576 (2004). 9. J. Geist, A. R. Schaefer, J. F. Song, Y. H. Wang, and E. F. Zalewski, Energy Citations Database report PB-91-144501/XAB, ‘‘Accurate value for the absorption coefficient of silicon at 633 nm’’ (http://www.osti.gov/ energycitations/).

submitted papers

Fourier Transform Infrared Spectroscopic Imaging of Anisotropic Poly(vinylidene fluoride) Films with Polarized Radiation CHRISTIAN VOGEL, ELKE WESSEL, and HEINZ W. SIESLER* Department of Physical Chemistry, University of Duisburg-Essen, Schuetzenbahn 70, D-45117 Essen, Germany (C.V., H.W.S.); and Beiersdorf AG, Research and Development, Unnastraße 48, D-20253 Hamburg, Germany (E.W.)

The technique of Fourier transform infrared (FT-IR) spectroscopic imaging with focal plane array detectors has proved to be a powerful technique for rapid chemical visualization of samples with a lateral resolution up to about 10 lm. However, the potential of FT-IR imaging for the characterization of anisotropic materials can be significantly enhanced by using polarized radiation. This issue will be addressed in the present communication, which reports for the first time imaging investigations based on the FT-IR polarization spectra of poly(vinylidene fluoride) films that have been uniaxially elongated below and above the threshold temperature of the II(a) ) I(b) phase transition. Index Headings: Fourier transform infrared spectroscopy; FT-IR spectroscopic imaging; Poly(vinylidene fluoride); Phase transition; Orientation.

INTRODUCTION Poly(vinylidene fluoride) (PVDF) is a polymer with exceptional technological and scientific properties. Depending upon the thermal, mechanical, and electrical pretreatment the polymer can exist in four different modifications: I(b), II(a), III(c), and IVp(d) or IVp(ap). The crystal structures of the crumpled a-form and the all-trans b-modification are orthorhombic and their molecular conformations are tgtg¯ and tt, respectively. The existence of these crystalline forms was first demonstrated by Lando et al.1 The different modifications, the conditions of their formation, and mutual transformation have been the subject of numerous publications.2–8 Thus, it has been shown that the II(a) modification can be converted into the I(b) form by tensile stress below 140 8C (Fig. 1). In the present communication PVDF films that have been uniaxially elongated to 400% strain at 100 8C and to 100% strain at 150 8C were analyzed by FT-IR imaging9–11 in the transmission mode with radiation polarized parallel and perpendicular, respectively, to the drawing direction of the investigated film samples. FT-IR spectroscopic imaging with a focal plane array (FPA) detector offers the possibility to combine spectral and spatial information, thereby enabling a detailed chemical visualization of samples. Thus, sample areas of 3.9 3 3.9 mm2 or 260 3 260 lm2 can be analyzed by FT-IR Received 31 October 2007; accepted 24 March 2008. * Author to whom correspondence should be sent. E-mail: hw.siesler@ uni-due.de.

Volume 62, Number 6, 2008

transmission spectroscopy without or in combination with a microscope, respectively, with a lateral resolution up to about 10 lm. Apart from the localization and characterization of the sample areas where the II(a) ) I(b) transformation has taken place, the use of polarized radiation provides additional information on the lateral distribution of anisotropy in the mechanically treated polymer films. Infrared spectroscopic studies of the different modifications8,12–14 showed that the II(a) form of PVDF has a characteristic absorption band at 975 cm1 that has a transition moment perpendicular to the chain axis and can be assigned to a ct(CH2) þ ms(CF2) vibration.8,13 If a polymer film in the II(a) modification is elongated below 140 8C, then this band completely disappears upon transformation into the I(b) form (Fig. 2, top). Elongation above 140 8C does not induce this conformational transition (Fig. 2, bottom) due to the significant reduction of the applied stress at elevated temperature (Fig. 3). By investigating PVDF II(a) film samples that had been elongated at 100 8C and 150 8C, the modification and orientational changes due to elongation below and above the threshold temperature of 140 8C can be monitored with the 975 cm1 absorption band. In the case where the 975 cm1 absorption had disappeared because of the conformational transition to the I(ß) phase, the m(CH) absorption, which also has a transition moment perpendicular to the chain axis, was used for the characterization of anisotropy. Only two previous publications15,16 have demonstrated that orientation function images can be calculated from polarization spectra. Thus, we have evaluated the orientation function f? of the 975 cm1 or the m(CH) absorption bands by f? ¼ 2

R1 Rþ2

ð1Þ

where R ¼ A||/A? is the dichroic ratio and A|| and A? are the integral intensities in the polarization spectra measured with radiation polarized parallel and perpendicular to the drawing direction of the polymer film, respectively. To monitor the transformation of the II(a) into the I(b) modification, the structural absorbance of the 975 cm1 band A0,975, which eliminates the effect of orientation on the band intensity, has

0003-7028/08/6206-0599$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

599

FIG. 1. The conformational changes occurring in the II (a) ) I (b) transformation of PVDF.

been used: A0 ¼

A þ 2  A? 3

ð2Þ

Inhomogeneities in the thickness of the investigated film samples were compensated by the ratio of the structural absorbance value of the 975 cm1 band against the integral structural absorbance of the m(CH) absorption (A0,m(CH)). For more detailed information on IR dichroism studies of anisotropic polymers, the reader is referred to the relevant literature.17–20

EXPERIMENTAL Materials. The film samples were prepared from PVDF granulate (supplied by Kureha Chem. Ind., Japan) by hot-

FIG. 2. FT-IR imaging spectra measured in (top) the clamp (—) and stretched (---) regions and (bottom) in the shoulder (—) and neck (---) regions of PVDF samples elongated to 400% strain at 100 8C and to 100% strain at 150 8C, respectively.

600

Volume 62, Number 6, 2008

FIG. 3. Stress-strain diagrams of PVDF films elongated to 400% strain at 100 8C and 150 8C.

pressing at 210 8C and melt crystallization at room temperature. The film samples had a density of 1.782 gcm3 measured in a carbon tetrachloride/ethylene bromide mixture. From these film specimens strips with a thickness between 0.030 and 0.035 mm and gauge dimensions of 8 3 4 mm2 were prepared for the drawing procedure. In the mechanical treatment, isotropic PVDF film samples in the II(a) modification were subjected to uniaxial elongation up to 400% strain (at 100 8C) and 100% strain (at 150 8C) with an elongation rate of 95% per minute. For the imaging measurements the samples were demounted from the clamps and selected areas were investigated. Instrumentation. The FT-IR spectra were recorded on a Bruker imaging system (Bruker Optik GmbH, Ettlingen, Germany), which consists of an IFS66/S FT-IR spectrometer coupled to an infrared microscope (Hyperion 3000), a macro imaging chamber (IMAC), and a 64 3 64 mercury cadmium telluride (MCT) focal plane array (FPA) detector (Santa Barbara Focalplane, Goleta, CA). The FPA detector is a 3.9 3 3.9 mm2 array with a pixel size of 61 3 61 lm2. Each image was measured in transmission in the rapid-scan mode for sample areas of 260 3 260 lm2 (with the 153 objective) and 3.9 3 3.9 mm2 (in the macro mode). To produce images with radiation polarized parallel and perpendicular to the drawing direction of the film samples, a KRS-5 wire grid polarizer was mounted behind the sample in the micro/macro-mode measurements. The spectral resolution was 8 cm1 and 10 scans were coadded per spectrum. The resulting spectra and imaging data were evaluated using the Bruker software OPUS 5.0. Because the absorption band at 975 cm1 has a strongly sloping baseline different integration limits were tested but yielded very similar results. The data shown here are based on the peak area with a baseline between 996 and 961 cm1. For the m(CH) reference band the area between 3060 and 2940 cm1 was integrated. The mechanical measurements were performed in a miniaturized stretching machine that can be operated between 50 8C and 175 8C.20

RESULTS AND DISCUSSION Figures 1 and 2 show the structural and spectroscopic changes occurring during the II(a) ) I(b) transformation of

FIG. 4. (a) Visual image, (b) A0,975/A0,m(CH) image, and (c) f? (m(CH)) image recorded of a 3.9 3 3.9 mm2 area in the unstretched/stretched regions of a PVDF sample elongated to 400% strain at 100 8C.

FIG. 5. (a) Visual image, (b) A0,975/A0,m(CH) image, and (c) f? (975 cm1) image recorded of a 3.9 3 3.9 mm2 area in the shoulder/neck region of a PVDF sample elongated to 100% strain at 150 8C.

FIG. 6. (a) Visual, (b) A0,975/A0,m(CH), and (c) f? (m(CH)) images recorded of a 260 3 260 lm2 area in the unstretched/stretched region of a PVDF sample elongated to 400% strain at 100 8C and (d) visual, (e) A0,975/A0,m(CH), and (f) f? (975 cm1) images recorded of a 260 3 260 lm2 area in the shoulder/neck region of a PVDF sample elongated to 100% strain at 150 8C.

APPLIED SPECTROSCOPY

601

PVDF, respectively. While the 975 cm1 absorption completely disappears in the film that has been elongated to 400% strain at 100 8C (Fig. 2, top), this II(a) phase-specific band is retained during elongation of the film at 150 8C (Fig. 2, bottom). The decrease of intensity is a consequence of the significant film thickness reduction in the neck of this elongated film only. The drastic reduction of stress level with the increase of temperature to 150 8C is demonstrated in the stress-strain diagrams of Fig. 3. Thus, from the retention of the 975 cm1 absorption band (Fig. 2, bottom) during this elongation procedure, it must be concluded that the applied stress is too low to induce the conformational transformation. Figure 4a shows the optical image of the film sample that had been elongated to 400% strain at 100 8C. The imaging area of 3.9 3 3.9 mm2 was selected at the border line of the clamp area (undeformed material) and the neck region of the elongated part of the PVDF sample film (at elongations of 400% strain the neck has extended over the whole sample between the clamps). The FT-IR image evaluated with the A0,975/A0,m(CH) band ratio is displayed in Fig. 4b. In the FT-IR contour plot the red and blue colors represent a high and low value of this band ratio, respectively, corresponding to the extent of the stress-induced II(a) ) I(b) transformation. The image shown in Fig. 4c is based on the f? orientation function of the m(CH) intensity. While the undeformed clamp area appears homogeneous in both content of II(a) form (Fig. 4b, top) and isotropy (orientation function f? ; 0) (Fig. 4c, top), the neck region below the clamp area is quite inhomogeneous with reference to the extent of II(a) ) I(b) transformation and anisotropy (variations of f? between 0.2 and about 0.5). Thus, the imaging technique now offers the possibility to visualize chemical as well as orientational inhomogeneities in this transition region. Figure 5a shows the optical image of a PVDF film stretched to 100% strain at 150 8C. Contrary to the 400% elongated sample, here the shoulder-neck region has not extended over the whole sample region between the clamps and three different sample regions are observable in the visual image (Fig. 5a): the undeformed clamp area (top), the (only very slightly oriented) shoulder region (center) and the deformed neck region (bottom). Because of the experimental temperature of 150 8C and the consequently lower stress applied during drawing (Fig. 3), no significant II(a) ) I(b) transformation is expected in the neck region. This is confirmed by the A0,975/ A0,m(CH) image (Fig. 5b) taken in the macro mode (3.9 3 3.9 mm2) at the shoulder-neck border line. Apart from some very small regions of modification changes (red spots) the slight differences in the contour colors are mainly a consequence of the strong intensity decrease of the m(CH) reference band and its influence on the calculation of the A0,975/A0,m(CH) ratio due to the thickness reduction in the neck region. Contrary to the chemical similarity, a very strong difference could be detected in the anisotropy of the shoulder and the neck region as represented by the orientation function (f?) image obtained with the 975 cm1 absorption band (Fig. 5c). Values between f? ¼ 0 and 0.2 were obtained for the shoulder region, whereas the neck region reflected f? values in the range 0.4–0.6. To demonstrate the results obtained upon 153 magnification, the same PVDF films were analyzed in the center of the 3.9 3 3.9 mm2 areas shown in Figs. 4 and 5 by focusing on a 260 3 260 lm2 area only along the border line between undeformed/ deformed area and shoulder/neck region, respectively. Figures

602

Volume 62, Number 6, 2008

6a–6c show the optical, the A0,975/A0,m(CH), and the f? (m(CH)) images, respectively, for the PVDF film elongated to 400% strain at 100 8C. The visual image (the orange color is due to the polarizer) shows a sharp transition region (;20 lm) between the clamp area and the neck region of the stretched part of the PVDF film. In the A0,975/A0,m(CH) image the width of this transition region is clearly extended more than three-fold (to about 65 lm) (yellow-green area with red dots in the center of Fig. 6b), reflecting an intermediate degree of II(a) ) I(b) transformation. Due to the focus on a much smaller area (260 3 260 lm2) of the transition region, the image of Fig. 6b is in general much more heterogeneous than the corresponding macro-mode image (3.9 3 3.9 mm2) (Fig. 4b). In analogy to the macro-mode measurements, the f? (m(CH)) image in Fig. 6c clearly reflects that the clamp area in the vicinity of the transition region is almost isotropic (f? close to 0), whereas the elongated neck region of the sample is heterogeneously oriented with f? values in the range 0.2 to 0.5. Finally, the 260 3 260 lm2 images of the PVDF film elongated to 100% strain at 150 8C (Figs. 6d and 6f) confirm the sharp transition between the slightly oriented (f? between 0 and 0.2) shoulder and the highly oriented (f? ; 0.5) neck region (both primarily in the II(a) form) of this sample. The A0,975/A0,m(CH) image of Fig. 6e also proves that only a very small proportion of PVDF segments have transformed to the I(ß) phase under the experimental conditions.

CONCLUSION Fourier transform infrared (FT-IR) imaging spectroscopy with polarized radiation provides a powerful tool to study orientation phenomena in polymeric materials. As an example, the technique was applied to characterize the temperaturedependent II(a) ) I(b) transformation and the stress-induced orientation in PVDF films stretched at different temperatures. By imaging the border line of the unstretched/stretched and shoulder/neck regions of samples elongated to 400% strain at 100 8C and to 100% strain at 150 8C, respectively, in the macro-mode (3.9 3 3.9 mm2) and with 153 magnification (260 3 260 lm2), a detailed picture of the structural and orientational changes across these transition regions could be obtained. ACKNOWLEDGMENTS The authors thank the Dr. Jost-Henkel-Stiftung (Du¨sseldorf, Germany) for the financial support of Christian Vogel.

1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

11. 12. 13.

J. B. Lando, H. G. Olf, and A. Peterlin, J. Polym. Sci. A1 4, 941 (1966). J. B. Lando and W. W. Doll, J. Macromol. Sci. Phys. B2 2, 205 (1968). J. B. Lando and W. W. Doll, J. Macromol. Sci. Phys. B2 2, 219 (1968). W. W. Doll and J. B. Lando, J. Macromol. Sci. Phys. B4 2, 309 (1970). K. Matsushige, K. Nagata, S. Imada, and T. Takemura, Polymer 21, 1391 (1980). Y. Takahashi, H. Tadokoro, and A. Odajima, Macromolecules 13, 1318 (1980). A. J. Lovinger, Macromolecules 15, 40 (1982). H. W. Siesler, J. Polym. Sci. Polym. Phys. Ed. 23, 2413 (1985). R. Bhargava, S. Q. Wang, and J. L. Koenig, Adv. Polym. Sci. 163, 137 (2003). J. L. Koenig, ‘‘FTIR imaging of multicomponent polymers’’, in Spectrochemical Analysis Using Infrared Multichannel Detectors, R. Bhargava and I. W. Levin, Eds. (Blackwell Publ. Ltd., Oxford, UK, 2005). A. Grupper, P. Wilhelm, M. Schmied, S. G. Kazarian, K. L. A. Chan, and J. Reussner, Appl. Spectrosc. 56, 1515 (2002). S. Enomoto, Y. Kawai, and M. Sugita, J. Polym. Sci. A2 6, 861 (1968). M. A. Bachmann and J. L. Koenig, J. Chem. Phys. 74, 5896 (1981).

14. M. Kobayashi, K. Tashiro, and H. Tadokoro, Macromolecules 8, 158 (1975). 15. B. Chernev and P. Wilhelm, Monatshefte fu¨r Chemie 137, 963 (2006). 16. C. M. Snively and J. L. Koenig, J. Polym. Sci., Part B: Polym. Phys. 37, 2353 (1999). 17. H. W. Siesler and K. Holland-Moritz, Infrared and Raman Spectroscopy of Polymers (Marcel Decker, New York, 1980). 18. H. W. Siesler, Adv. Polym. Sci. 65, 1 (1984).

19. J. Dechant, Ultrarotspektroskopische Untersuchungen an Polymeren (Akademie-Verlag, Leipzig, 1972). 20. H. W. Siesler, G. G. Hoffmann, O. Kolomiets, F. Pfeifer, and M. Zahedi, ‘‘Variable-Temperature Rheo-Optical Fourier-Transform Infrared Spectroscopy of Polymers’’, in Vibrational Spectroscopy of Polymers: Principles and Practice, N. J. Everall, J. M. Chalmers, and P. R. Griffiths, Eds. (John Wiley and Sons, Chichester, UK, 2007), pp. 313–347.

APPLIED SPECTROSCOPY

603

Fluorescence Resonance Energy Transfer Based MCM-EDTA-Tb3þ-MES Sensor CLINT B. SMITH,* JOHN E. ANDERSON, RICHARD D. MASSARO, BALAJI TATINENI, KINSON C. KAM, and GARY C. TEPPER U.S. Army Engineer Research and Development Center, Topographic Engineering Center, Alexandria, Virginia 22315 (C.B.S., J.E.A., R.D.M.); Virginia Commonwealth University Dept. of Chemistry, Richmond, Virginia 23284 (B.T.); University of California, Materials Research Laboratory, Santa Barbara, California 93106 (K.C.K.); and Sentor Technologies Inc., 11551 Nuckols Road, Suite Q, Glen Allen, Virginia 23060 (G.C.T.)

An in situ mesopourous surface imprinted polymeric (SIP) sensor was synthesized for a highly sensitive, selective, and kinetically faster detection of the high-vapor-pressure nerve gas surrogate methyl salicylate (MES). Visual detection occurred on the filtrate thin films at 25 pM. Other nerve gas surrogates, TP, DMP, DMMP, PMP, and 1,4-thioxane, were tested and showed a decrease in sensitivity compared to MES. In addition, 2,6dipicolinic acid (DPA), a biological indicator, was also investigated and showed a decrease in sensitivity compared to MES. Finally, the detection plateau was reached at 40 s and at 1.5 3 104 M from pH 6–11. Index Headings: Fluorescence resonance energy transfer; FRET; Surface imprinted polymeric sensors; SIP; Methyl salicylate; MCM silica; Tb3þ; Nerve gas surrogate detection.

INTRODUCTION The Department of Homeland Security and the Department of Defense are interested in the broad range of collectors and detectors with high selectivity and sensitivity for hazardous nerve gases such as Sarin, Soman, and toxic industrial chemicals due to the potential for their use in unorthodox activities. Among these, Sarin and Soman are nerve gas reagents that have been produced and stockpiled globally as a result of past conflicts and are now aging. The leaking of these types of chemicals from stockpiles poses a risk due to the high vapor pressure of these compounds, which leads to concern for the possible contamination of water supplies. This has promoted a desire to develop solid-state selective array detectors for on-site detection. Methyl salicylate (MES) has been recognized by the Department of Homeland Security and the Environmental Protection Agency as a nerve gas surrogate. MES is also known as oil of wintergreen, betula oil, methyl-2hydroxybenzoate, having high vapor pressure and is a natural product of many species of plants. Technology currently used for hazardous-material-type responses are gas chromatography–mass spectroscopy (GCMS) and high-performance liquid chromatography (HPLC). These intstruments are large and expensive and often require extensive procedures.1–3 These techniques also do not necessarily have the direct ability to warn of immediate danger. In view of these short falls, research investigations have been starting to incorporate near-real-time analysis via the use of small field-portable sampling instrumentation. The technology currently being developed uses on-site analyses that are based on enzyme chemistry, selective immunoassays, and blocking of photosynthetic activity.4–7 In spite of many Received 19 November 2007; accepted 28 March 2008. *Author to whom correspondence should be sent. E-mail: clint.b.smith@ erdc.usace.army.mil.

604

Volume 62, Number 6, 2008

advantages for field-portability, these sensors have many draw backs such as the fact that they are less stable, non-reusable, and can take up to 30 minutes to respond. Surface imprinted polymers (SIPs) and molecularly imprinted polymers (MIPs) have become more useful as inorganicbased materials for the construction of lock and key mechanisms for highly sensitive and specific binding of chemical species.8–11 The process involves the synthesis of highly cross-linked networks to entrap the template moiety by polymerizing the monomer along with the cross-linker surrounding the template moiety. Removal of the template moiety leaves a fixed array of ligand and customized binding pockets. Synthesis of MIP materials has shown that organic polymer matrices1,12,13 are inhomogeneous and there is little control over the pore size, surface area, and swelling properties, which indicates that slower kinetics occur.12,13 Graham et al.14 synthesized MIP sol gel materials that used regular organic polymer imprint procedures that have kinetic and selectivity drawbacks. However, many research groups15–18 have used the surface imprint phenomenon to create cavities on the surface of ordered and networked inorganic materials. Since the invention of hierarchically imprinted polymers by Dai et al.19 to control the pore structures and adsorption sites using double templates, the research on hierarchically imprinted polymers has increased. These types of imprinted materials have been used to imitate solid-phase extraction and antibody functions to resolve racemic mixtures. In addition, there have not been many studies that focus primarily on SIPs that are directly related to optical and electrical sensing of biological and chemical species. This research investigation involved the use of SIPs that optically sense MES, a Sarin gas surrogate. Due to the lack of highly sensitive, selective, and kinetically fast on-site reporting methods for the detection of nerve gas surrogates and based on our previous studies20,21 for the detection of heavy metals, we report the in situ synthesis of MCM-41-EDTA SIP, using Tb3þ as a reporter for the detection of MES (Scheme 1). Here we have observed the fluorescence resonance energy transfer22,23 (FRET) phenomena, in which MES acts as a donor and Tb3þ acts as a receptor. To our knowledge this is the first report of FRET based detection of the nerve gas surrogate MES. Methyl salicylate is theorized to exist in two different isomers (ketoA and ketoB) in equilibrium at room temperature (Fig. 1). Some researchers have proposed that the dual fluorescence emission profile of MES is due to the different emissions of the two isomers.24 However, ketoA is only thought to be about 1/70 of the amount of ketoB in equilibrium.25 It is now widely thought that the ketoB form undergoes an excited-state intermolecular proton transfer to the

0003-7028/08/6206-0604$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

SCHEME 1. Schematic representation of template-directed in situ molecular imprinting of mesoporous silica particles (SIP).

so called ‘‘enol’’ triplet form (Fig. 2). This excited enol triplet is a lower energy state than the ketoB excited state and likely creates the long-wavelength emission.26 Density functional theory (DFT) calculations were performed on these isomers to determine their optimized structures. The ketoA, ketoB, and enol geometries are shown in Figs. 1 and 2.

EXPERIMENTAL Materials. Analytical reagent grade chemicals and deionized ultra-pure distilled water were used throughout the experiment, unless otherwise indicated. Tetra ethyl orthosilicate (TEOS) and cetyl trimethyl ammonium bromide (CTAB) were purchased from Sigma-Aldrich. N-(trimethoxysilylpro-

FIG. 1. Isomers of MES in singlet states. (Left) ketoA, (right) ketoB.

APPLIED SPECTROSCOPY

605

FIG. 4. IR spectral data of MCM-41 and SIP (in situ MIP).

FIG. 2. The enol triplet isomer of MES.

pyl)ethylene diamine, triaceticacid, and Na salt (EDTA-Si) were purchased from Dojindo (Gaithersburg, MD). Buffer solutions (0.2 M) of KCl-HCl, CH3COOH-CH3COONa, 2(cyclohexylamino) ethane sulfonic acid (CHES)-NaOH, 3morpholinopropane sulfonic acid (MOPS)-NaOH, and Ncyclohexyl-3-aminopropane sulfonic acid (CAPS)-NaOH were employed to adjust the pH of the solutions containing TbCl36H2O. Nerve gas analogues such as dimethyl methyl phosphonate (DMMP), diethyl ethyl phosphonate (DEEP), triethyl phosphate (TEP), pinacolyl methyl phosphonate (PMP), 1,4-thioxane, and structural analogue DPA were

FIG. 3. XRD data of MCM-41 and SIP (in situ MIP).

606

Volume 62, Number 6, 2008

purchased from Sigma-Aldrich and solutions were prepared using 0.01 M HCl aqueous solution. Instrumentation. The BET surface area, pore-size, and pore volume were measured using BELSORB Mini in N2 isotherm. Powder X-ray diffraction patterns were recorded using a MAC Science powder diffractometer under CuKa radiation (50 kV, 200 mA), with a step size of 0.028 and 2 s step time over the range 0.78 , 2h , 68. Field emission scanning electron microscopy (SEM, Hitachi S-800) and transmission electron microscopy (TEM, JEOL-2000EX II, 200 KV) were used to confirm the morphology and phase purity of the samples, respectively (see Supplemental Material Figs. S1–S4 and Table S1). The absorbance of the MES was recorded using the Spectronic Genesis model 6 ultraviolet (UV)-Visible spectrometer with matched quartz cells. The fluorescence spectra of the solid samples were measured using a Horiba Jobin Yvon fluorescence spectrometer (FluoroLog) equipped with fiber optics. The pH measurements were carried out using a Horiba digital pH meter with a glass electrode. Synthesis of MCM-41. The mesoporous material was prepared according to procedure 1 with a slight modification in the reaction conditions, using the following molar ratio composition: 0.137 CTAB:1 TEOS:5.9 NH4OH:139 H2O. The solution was stirred for 2 h at room temperature. The resulting white precipitate was heated at 80 8C for 12 h. The precipitate was recovered by suction filtration and washed using distilled water until a pH of 7 was obtained for the filtered liquid. The mesostructure was then dried in a vacuum oven at 90 8C for 12 h. The dried mesostructure was calcined at 550 8C to remove the template and characterized. Synthesis of the Non-Surface Imprinting Polymer. The synthesized MCM-41 material was used to prepare the nonimprinting polymer with Tb. EDTA-Si in a molar ratio of 1:1 was kept at 50 8C for 2 h and filtered. The product was washed with acetone, ethanol, and dichloromethane to remove any physisorbed precursors inside the material. The resulting product was finally dried at 70 8C for 6 h, ground into a fine powder, and then sieved. Optimized Protocol. A working solution was prepared by taking 0.2 M CHES buffer (pH ¼ 9) and MES solution. Two milligrams (2 mg) of MCM-SIP material was added to the solution and its volume was adjusted to 20 mL. A blank solution was also prepared following the same procedure for comparison. The product was filtered using a 25 mm cellulose

FIG. 5. (a) Excitation–emission matrix scan of SIP with MES; (b) excitation–emission matrix scan of SIP without MES.

acetate filter paper (Sibata filter holder) and a UV lamp was used for visual fluorescence assessment. Concentrations of MES in the solid samples were measured quantitatively by comparing the fluorescence intensity of the samples with the standard known concentration of MES. Similarly, the same procedures were used to measure the non-SIP. Synthesis of the Methyl Salicylate–Surface Imprinted Polymeric Sensor. Synthesis of the MES-SIP sensor was carried out by dissolving cetyl trimethyl ammonium bromide (CTAB) in distilled water. A reaction mixture was prepared by reacting Tb3þ, MES, and EDTA-Si, and then the pH of the solution was adjusted to 8 using CHES buffer. TEOS was added to this mixture and stirred for 1 h. The resulting solution

was added to CTAB solution and the pH of the mixture was adjusted to 11 using NaOH and stirred for 1 h. The generated ethanol and excess water were removed by rotaevaporation and the resulting gel was kept at room temperature for 12 h. The molar ratio composition used in this synthesis procedure was 0.161 CTAB:1 TEOS:0.02 Tb:0.06 MES:0.02 EDTA-Si:0.55 NaOH:162 H2O. The synthesized SIP was filtered, washed with acetone and dichloromethane, and dried at room temperature. MES was removed from the in situ SIP material using methanol until no further absorbance appeared for MES at 310 nm in the filtrate. The resulting product was finally dried at 50 8C for 6 h, ground to a fine powder, and sieved. MCM-41 material was prepared according to the procedure7,8 with a

FIG. 6. (Left) Reaction of SIP (in situ MIP) with varying MES concentrations at pH 9, slits 3 nm, 1 cm off the surface (fiber optic), 2 mg material, excitation at 310 nm. (Right) Low concentration of MES reaction with SIP (in situ MIP).

APPLIED SPECTROSCOPY

607

FIG. 7. Effects of pH on the reaction of MES with the SIP (in situ MIP), MES concentration 5 3 105 M, material weight 2 mg, slits 3 nm, 1 cm off (fiber optic), and excitation at 310 nm.

slight modification in the reaction conditions, using the following molar ratio composition: 0.137 CTAB:1 TEOS:5.9 NH4OH:139 H2O. In addition, a non-SIP was also prepared, as discussed previously.

FIG. 9. SIP (in situ MIP) sensor reaction with MES; pH 9, material weight 2 mg, and excitation at 310 nm (a calibration graph).

The MES SIP sensor along with MCM-41 was characterized using X-ray diffraction (XRD), infrared (IR) spectroscopy, SEM, and TEM (the SEM and TEM figures are provided in the Supplemental Material). The presence of Tb3þ inside the material was indicated by broadening and shifting of the XRD peak compared to the MCM-41 material (Fig. 3). The decrease in peak intensity significantly indicates the pore size increase and also the reduction in the long-range mesoscopic order from a channel to a worm-like structure.8 TEM data indicates the presence of Tb3þ nanocrystals inside the channels of the SIP along with a worm-like pore array and without any long-range order, as indicated by XRD. The presence of Tb-MES-EDTA moiety in the network minimizes the growth of the SIP mesoporous silica. The high pore volume and pore size, along with moderate surface area of the SIP, could be due to the presence of the cross-linker, reporter, and template inside the network. Infrared spectra of the SIP indicate the presence of the

organic moiety corresponding to aliphatic C–H and the stronger C–N bonds at 2960 and 1400 cm1 respectively (Fig. 4). The fluorescence excitation–emission matrix scan of the post-SIP indicates that the terbium, due to its low quantum yield and low absorption cross-section, exhibits a very weak fluorescence, whereas the pre-SIP exhibits a very strong fluorescence. The reason for this phenomenon could be due to the FRET mechanism from donor MES to acceptor Tb3þ. In FRET, MES shows an excitation at 310 nm and an emission at 440 nm, which is taken by the acceptor Tb3þ, and shows a very bright green fluorescence emission (Figs. 5a and 5b), which is several orders of magnitude larger. In this case the excitation of MES is followed by intramolecular energy transfer from MES, possibly in the enol triplet state, to the lower lying emissive terbium excited state. Here the MES is also acting as an antenna or sensitizer to compensate the low absorption coefficient of terbium. The figures (Figs. 5 and 6) indicate that the reaction of MES with Tb3þ provides four characteristic emission bands 5D4!7F5 (J ¼ 6, 5, 4, 3) in the blue and green region, of which the 5D4!7F5 transition (545 nm) is the highest. The reason for high emission intensities of the SIP material could be the high concentration of metal ions, non-leaching of the photoactive molecules, and non-clustering of emitting centers.27,28 The conditions for the detection of MES were optimized chemically, as discussed previously. The studies on the effect of pH indicate that from pH 5 the reaction with MES initiates and the intensity gradually increases to pH 7.5 and then reaches

FIG. 8. Kinetics of MES reaction with SIP (in situ MIP) sensor; slits 3 nm, pH 9, excitation at 310 nm, and MES concentration 1.5 3 104 M.

FIG. 10. FRET effect with varying MES concentration on SIP at pH 9, material weight 2 mg.

RESULTS AND DISCUSSION

608

Volume 62, Number 6, 2008

FIG. 11. Reaction of various MES analogues with SIP (concentration of MES 5 3 106 M, analogues 2.5 3 104 M).

a maximum from pH 8 to 10 (Fig. 7). The data suggests that the fluorescence behavior of the complex is at a maximum when the pH is above the pKa29,30 value of MES, suggesting that the anionic chelate is the most emissive species. A kinetic study indicates (Fig. 8) the reaction of MES occurring spontaneously (;40 s) with the SIP material, and the fast kinetic reaction could be due to the uniform distribution of lanthanides, particle size, and pore volume. The sensitivity reaction of the MES-SIP sensor was carried from a 2.5 3 1010 to 2 3 103 molar MES concentration using 2 mg of material at pH 9, and it was found that the material was able to detect picomolar MES concentrations (Fig. 6). The range of detection was very broad and the fluorescence intensity of MES-Tb3þ at 550 nm gradually increases with the increase in MES concentration (Fig. 9). Figure 10 explains the application of our solid-state array detector material for on-site detection of MES in various filtered thin-filmed tickets. The detection limit of the various SIP materials was calculated using the equation DL ¼ kSb/m, where k ¼ 3, Sb is the standard deviation for the blank, and m is the slope of the calibration graph in the linear range. The MESSIP sensor was found to have a limit of detection and quantification of 8.6 3 1011 and 2.86 3 1010 (mol1), respectively. The experiments on the effect of the weight of the material show that 2 mg of material was good for detection. The selectivity of the in situ MIP (MES-SIP) was tested with other nerve gas analogues and structural analogues using a concentration of 2.5 3 104 M. The data shows that a very weak signal intensity was observed for the other analogues compared to the MES signal intensity for in situ MIP (MESSIP) (Fig.11), whereas the non-MIP shows good reaction for most of the analogues, even though the intensity was not comparable to the in situ MIP (MES-SIP). The high selectivity of the MES-SIP sensor was due to the creation of homogeneous imprinted cavities inside the material, which was in contrast to the presence of no cavities in the nonMIP material. The test for repeated use of the material indicates the stability and usefulness of the material when applied as a sensor that could be reclaimed and reused. The MES-SIP material was tested using methanol as an eluant and we found that MES was detectable for eight repeated cycles at 4 3 106

FIG. 12. Removal of MES from SIP and its reusability.

cps or higher (Fig. 12). The data shows that even after eight repeated uses, the performance of the detector was only decreased by ;15% and this could be due to the leaching of the Tb3þ from MES-SIP sensor. This indicates that the material can be reused multiple times without losing its FRET sensing capability. The photostability of the MES-SIP material was tested and it was found that it is not bleached for a period of 2 h and after that, due to MES high vapor pressure, the fluorescence values gradually decreased.

CONCLUSION This research demonstrates that a double template surface imprint phenomenon could be used to synthesize polymers that exhibit selective binding properties of predetermined input species with high sensitivity and ultra-fast detection capabilities. Unlike the mesoporous solid-state detectors synthesized by surface functionalization,20,21,31–33 precise control of the arrangement of the ligands can be achieved with this double template approach. The moderate surface area, high stability of the material, low cost, flexibility of designing select pore sizes and volumes, and its non-photobleaching properties achieved with the current synthetic protocol improve upon the organic molecularly imprinted polymer methods. This research has resulted in opening up new formulas for selective detection strategies for the identification of various chemical high vapor pressure toxins and other biologically hazardous constituents of interest. ACKNOWLEDGMENT We would like to thank the U.S. Army Engineer Research and Development Center basic research program for funding this research opportunity.

SUPPLEMENTAL MATERIAL All supplemental material mentioned in the text, which includes Figures S1, S2, S3, and S4 and Table S1, showing SEM and TEM images of MCM-41 and SIP and the structural properties of MCM-41 and SIP, respectively, is available online on the SAS homepage at http://www.s-a-s.org and in the electronic on-line version of the journal. 1. A. L. Jenkins and S. Y. Bae, Anal. Chim. Acta 542, 32 (2005). 2. A. L. Jenkins, O. M. Uy, and G. M. Murray, Anal. Chem. 71, 373 (1999).

APPLIED SPECTROSCOPY

609

3. G. M. Murray and G. E. Southard, IEEE Instrum. Measure. Mag. 5(4), 12 (2002). 4. Assessment of Chemical and Biological Sensor Technologies, Report of the National Research Council (NRC, Washington, D.C., 1984). 5. W. Trettnak, A. Reininger, E. Zinterl, and O. S. Wolfbeis, Sens. Actuators, B 11, 87 (1993). 6. J. L. Marty, D. Garcia, and R. Rouillion, Trends Anal. Chem. 14, 329 (1995). 7. A. L. Jenkin, R. Yin, and J. J. Jensen, Analyst (Cambridge, U.K.) 126, 798 (2001). 8. T. A. Sergeyeva, S. A. Piletsky, A. A. Brovko, E. A. Slinchenko, L. M. Sergeeva, and A. V. El’skaya, Anal. Chim. Acta 392, 105 (1999). 9. G. Wulff, Angew. Chem. Int. Ed. Engl. 34, 1812 (1995). 10. K. Mosbach, Trends Polym. Sci. 2, 166 (1994). 11. R. A. Bartsch and M. Maeda, Eds., Molecular and Ionic Recognition with Imprinted Polymers (American Chemical Society, Washington, D.C., 1998). 12. S. Santra, P. Zhang, K. Wang, R. Tapec, and W. Tan, Anal. Chem. 73, 4988 (2001). 13. Z. Ye, M. Tan, G. Wang, and J. Yuan, Anal. Chem. 76, 513 (2004). 14. A. L. Graham, C. A. Carison, and P. L. Edmiston, Anal. Chem. 74, 458 (2002). 15. K. R. Kloetstra, H. Van Bekkum, and J. C. Jansen, Chem. Commun. 23, 2281 (1997). 16. S. L. Burkett, S. D. Sims, and S. Mann, Chem. Commun. 11, 1367 (1996). 17. S. Dai, M. C. Burleigh, Y. Shin, C. C. Morrow, C. E. Barnes, and Z. Xue, Angew. Chem. Int. Ed. Engl. 38, 1235 (1999).

610

Volume 62, Number 6, 2008

18. R. Makote and M. M. Collinson, Chem. Mater. 10, 2440 (1998). 19. S. Dai, M. C. Burleigh, Y. H. Ju, H. J. Gao, J. S. Lin, S. J. Pennycook, C. E. Barnes, and Z. L. Xue, J. Am. Chem. Soc. 122, 992 (2000). 20. T. Balaji, S. A. El-Safty, T. Hanaoka, H. Matsunaga, and F. Mizukami, Angew. Chem. Int. Ed. 45, 7202 (2006). 21. T. Balaji, M. Sasidharan, and H. Matsunaga, Anal. Bioanal. Chem. 384, 488 (2006). 22. J. M. Mauro, B. Fisher, E. R. Goldman, H. Mattoussi, A. R. Clapp, and I. L. Medintz, Nature Mater. 2, 630 (2003). 23. M. Smoluch, H. Joshi, A. Gerssen, C. Gooijer, and G. V. Zwan, J. Phys. Chem., A 109, 535 (2005). 24. A. U. Acuna, J. Catalan, and F. Toriblo, J. Phys. Chem. 85, 241 (1981). 25. L. Helmbrook, J. E. Kenny, B. E. Kohler, and G. W. Scott, J. Phys. Chem. 87, 280 (1983). 26. J. Catalan and C. Diaz, J. Phys. Chem. A 102, 323 (1998). 27. E. Brunet, O. Juanes, and C. R. Ubis, Curr. Chem. Biol. 1, 11 (2007). 28. Q. M. Wang and B. Yan, J. Photochem. Photobiol. A: Chem. 178, 70 (2006). 29. Z. Ernst and F. Herring, Trans. Faraday Soc. 61, 454 (1965). 30. G. Crisponi, M. Casu, V. M. Nurchi, F. C. Marincola, T. Pivetta, and R. Silvagni, Talanta 56, 441 (2002). 31. P. P. Balog, T. B. Stanford, R. J. Nordstrom, and R. C. Burgener, Feasibility assessment of Piezoelectric crystals as chemical agent sensors (HQ Aerospace Medical Division, 1986). 32. M. S. Nieuwenhuizen and J. L. N. Harteveld, Talanta 141, 461 (1994). 33. M. S. Nieuwenhuizen and J. L. N. Harteveld, Sens. Actuators, B 40, 167 (1997).

Silica Colloidal Crystals as Porous Substrates for Total Internal Reflection Fluorescence Microscopy of Live Cells TOMIKA R. C. VELARDE and MARY J. WIRTH* Department of Chemistry, University of Arizona, 1306 E. University Blvd., Tucson, Arizona 85721

Total internal reflection fluorescence (TIRF) microscopy is a powerful means of probing biological cells because it reduces autofluorescence, but the need for direct contact between the cell surface and the microscope slide hinders chemical access to the cell surface. In this work, a submicrometer crystalline layer of colloidal silica on the microscope coverslip is shown to allow TIRF microscopy while also allowing chemical access to the cell surface. A 750 nm layer of 165 nm silica colloidal crystals was sintered onto a fused silica coverslip, and Chinese hamster ovary cells were successfully grown on this surface. This cell line over-expresses the human delta-opioid receptor, which enabled probing of the binding of a labeled ligand to the receptors on the cell surface. Total internal reflection and chemical access to the cell surface are demonstrated. The range of angles for total internal reflection is reduced only by 1/3 due to the lower index of refraction of the colloidal multilayer relative to fused silica. Index Headings: Total internal reflection fluorescence microscopy; TIRF; Colloid; Nanoparticles: Colloidal crystal; Cells; Membrane protein; Delta-opioid receptor.

INTRODUCTION Membrane proteins are among the most important cellular proteins to understand because they play active roles in membrane transport, cell-to-cell communication, and intracellular signaling pathways. From a pharmacological perspective, membrane proteins are the most common targets for drugs. In particular, the G-protein coupled receptors (GPCRs) have been estimated to constitute over half of the targets for drugs currently in development.1 When a ligand from the extracellular medium binds into a pocket created by the seven alpha helices of a GPCR, a conformational change is believed to occur, and signal transduction is then triggered by binding of a G-protein from the intracellular side. Understanding the details of how ligand recognition by the receptor fosters signal transduction would advance drug discovery, and new technology to probe this phenomenon is continually being evaluated.2–5 Total internal reflection fluorescence (TIRF) spectroscopy and TIRF microscopy are powerful ways to probe binding of ligands to receptors.6 In artificial bilayers, TIRF affords surface selectivity to distinguish bound from unbound ligands,7,8 and TIRF enables the control of polarization to provide orientational information.9 Very high index substrates allow depth profiling on the scale of tens of nanometers by changing the penetration depth of the evanescent wave.10 The use of cell membrane fragments allows for a native environment while still avoiding the complications of live cells, such as receptor internalization and autofluorescence.11–14 TIRF is most valuable for live biological cells, which have strong autofluorescence: the small penetration depth of total internal reflection Received 4 October 2007; accepted 26 February 2008. * Author to whom correspondence should be sent. E-mail: mwirth@ email.arizona.edu.

Volume 62, Number 6, 2008

microscopy enables sensitive fluorescence probing of cell surfaces.6,15 Single-molecule fluorescence microscopy allows direct observation of interactions between ligands and receptors, and it is valuable for assessing heterogeneity by exploring individual members of an ensemble.16 When combined with TIRF to decrease background, the single-molecule capability enables the acquisition of kinetic and orientational parameters for individual binding events.17 TIRF microscopy of single molecules of G-protein coupled receptors tagged with green fluorescent protein has been achieved in live cells to allow tracking of the life cycle of the receptor.18 Most pertinent, the binding of single ligands to epithelial receptors in live cells has been accomplished by TIRF microscopy.19 This last example is illustrative of the need addressed in this paper: the ligand cannot access the side of the cell in contact with the substrate; instead, TIRF was performed by directing the beam through the cell and reflecting it off of the outer cell wall. The beam thus illuminates the cell interior to give a significant amount of autofluorescence, which reduces the signal-to-background ratio in the single-molecule experiment. A combination of TIRF with the ability to probe ligand binding kinetics would be valuable. The purpose of this work is to investigate silica colloidal crystals as a type of porous substrate that allows TIRF microscopy with chemical access to the surface of the cell facing the high-index medium. The idea is illustrated in Fig. 1. The ability to achieve TIRF through the colloidal array depends on whether the refractive index of the porous medium is sufficiently high. The net refractive index of the colloidal crystal, nx, is determined by the volume fraction of the two media (silica and water) and the refractive indices of the two media, ns and nw, respectively: n2x ¼ 0:74  n2s þ 0:26  n2w

ð1Þ

For the fcc crystals, the volume fraction of silica is 0.74. The refractive indices of fused silica and water are 1.46 and 1.33, respectively, and the refractive index of silica nanoparticles has recently been shown to be 1.46;20 therefore, the net refractive index of the medium is 1.43. In principle, this is sufficiently high for TIRF. The TIRF objective has glass optics with an index of 1.518, and index-matching oil of the same refractive index is used to couple light into the coverslip. Despite the lower index for the colloidal crystal, total internal reflectance ought to occur through the layer of colloidal crystal once the critical angle for the glass–water interface has been achieved. This is included in the illustration of Fig. 1, which shows that incident rays at the critical angle for glass–water are refracted to enter silica at the critical angle for silica–water, and these in turn are refracted to enter the colloidal crystal at the critical angle for the interface of the colloidal crystal with water. Three essential questions are addressed in this work: do cells grow

0003-7028/08/6206-0611$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

611

FIG. 1. Illustration of concept for using TIRF with a colloidal crystal as a porous support for cells. Refraction through layers from the objective/oil (or glass), the silica coverslip, the colloidal crystal, and finally the cell. The range of angles that is allowed by the objective, but is inaccessible due to the presence of the colloidal crystal, is shown as the black triangle in the lower left.

and adhere to these crystals, does TIRF occur, and does the ligand access the cell surface?

EXPERIMENTAL Colloidal Crystals. Silica colloids of nominally 200 nm were synthesized using the Sto¨ber-Fink-Bohn method.21,22 The colloids were then calcinated at 600 8C three times, for 5 hours each, in a box furnace (Lindberg/Blue, BF1766A). Between each calcination step, the colloids were suspended in ethanol by sonication (VWR Ultrasonic Cleaner Model 75T) to minimize aggregates. Colloidal crystals were formed on fused silica coverslips by vertical deposition. Coverslips were cleaned with boiling methanol and lens tissue, followed by UV-ozone treatment for 15 minutes at 20 W/cm2 (Novascan Technologies PSD-UV). The clean coverslips were placed in a 15 mL suspension of 0.06% (wt %) colloids in absolute ethanol in a small flask and heated (Thermo Electron Corporation Precision Economy Incubator, Model 2EG) at 55 8C until all ethanol evaporated. After deposition, crystals were sintered at 1050 8C for 12 hours for durability, and a utility patent has been filed for this application. The colloidal arrays were then rehydroxylated in a pH 9.5 tertbutylammonium hydroxide solution for 48 hours at 60 8C to restore surface silanol groups removed in the sintering process. Crystal uniformity and thickness was studied using field emission scanning electron microscopy (SEM) and atomic force microscopy (AFM). SEM of platinum coated crystals was performed using a Hitachi S4500 field emission scanning electron microscope with Thermo-Noran digital imaging. Transmission spectra of the colloidal crystals were obtained with a diode array absorbance spectrometer (Agilent 8453). An uncoated silica coverslip was used as the blank. Cells and Ligands. Colloidal crystals were initially cleaned using boiling methanol, followed by UV-ozone treatment. The crystals were then autoclaved and a sub-confluent layer of the Chinese hamster ovary cells was cultured directly on the crystals in a university facility for mammalian cell growth. The cells were grown in HAM’s F-12 medium containing 10% fetal calf serum and antibiotics at 37 8C in air enriched with 5% CO2. Two types of Chinese hamster ovary cells were investigated.23,24 One type over-expressed the human deltaopioid receptor and the other type over-expressed the same receptor with an enhanced green fluorescent protein (EGFP) label on its C-terminus.

612

Volume 62, Number 6, 2008

An antagonist labeled with fluorophore was used to test ligand binding. The antagonist was TIPP, a tetra-peptide with the sequence Tyr-Tic-Phe-Phe. It was synthesized by solid phase peptide synthesis using fluorenylmethoxycarbonyl (Fmoc) chemistry. The peptide binds with the Tyr residue directed into the binding pocket;25 hence, labeling at the Cterminus is not expected to interfere with its ability to bind to the receptor. The sequence Tyr-Tic-Phe-Phe-Gly-Cys-Gly was used to allow labeling of the cysteine residue by maleimide coupling of cy5. Total Internal Reflection Fluorescence Microscopy. For imaging of whole cells and cell fragments, coverslips were assembled in a liquid flow cell designed for use with an inverted microscope. Growth media was rinsed away and replaced with BES-Tyrodes buffer (pH 7.5). For excitation of EGFP, an argon ion laser (35-IMA-0840–015) provided the necessary 488 nm line for EGFP excitation, and a krypton ion laser (Ion Laser Technology, model 5400) provided the 647 nm line for cy5 excitation. TIRF microscopy was obtained using an inverted optical microscope (TE2000-U, Nikon). The commercial accessory (TIRF, Nikon) couples the polarized laser beam into an optical fiber whose output is focused onto the backplane of the microscope objective (Plan ApoTIRF, 1003 NA 1.45, Nikon) to achieve a collimated beam through the coverslip. By translating the optical fiber toward the outer edge of the objective, angles above the critical angle can be achieved to produce total internal reflection. Fluorescence and white light images were taken using an intensified chargecoupled device (ICCD) camera, Cascade II (512 3 512, Roper Scientific). An EGFP filter cube was used for each image of the cells expressing EGFP fused to the receptor, and a cy5 cube was used for the images of TIPP-cy5 bound to the cell surfaces.

RESULTS The deposition of stable layers of thin colloidal crystals on coverslips is required to achieve a porous TIRF support that is sufficiently rugged for cell culturing. Figure 2a shows the SEM image of a microscope coverslip bearing a thin layer of colloidal crystal. The coverslip had been fractured to allow a side view of the crystal. The image shows that the colloids are 165 nm in diameter, and the thickness of the crystal is 750 nm. The image also shows the divots and protrusions resulting from attachment of the colloids during the sintering process. Figure 2b shows an AFM image, giving a top view of the slide over a

FIG. 2. (a) SEM cross-section and (b) AFM images of a sintered colloidal crystal. The scale bar in (a) is 500 nm. The full scale for the planar axes in part (b) is 0 to 20 lm.

20 lm scale. This image shows the crystalline order over the coverslip, and the hexagonal packing confirms that the nanoparticles crystallized in the expected fcc space group. It also shows how flat the colloidal crystal is on the scale of the size of a cell (;10 lm). The entire height scale spans the range of only one nanoparticle diameter. Examination of the crystal

over other regions showed that this region is representative. There are occasional voids of one or several nanoparticles. These defects would slightly change the fluorescence intensity because the distance from the particle in the lower layer (130 nm) approaches the predicted penetration depth of the evanescent wave (180 nm). The AFM image shows that there

APPLIED SPECTROSCOPY

613

FIG. 3. Fluorescence micrographs of Chinese hamster ovary cells having EGFP-labeled human delta-opioid receptor on (a) silica coverslip using epi-fluorescence, (b) silica coverslip using TIRF, (c) colloidal crystal using epi-fluorescence, and (d) colloidal crystal using TIRF. The images are each 46 lm across.

are no larger steps; therefore, the voids would likely not affect the ability to count active receptors accurately in a singlemolecule experiment. Chinese hamster ovary cells were found to grow as well on the silica colloidal crystals as they do on silica coverslips, and fluorescence images of such cells are shown in Fig. 3. The particular cell type was one that over-expresses the delta-opioid receptor, which itself was engineered to bear enhanced green fluorescent protein (EGFP). This same figure compares epifluorescence images and TIRF images for the silica coverslip and the colloidal crystal, each with these cells on the surface. The images also serve to demonstrate that cells grow on the colloidal crystal, as evidenced by the elongated shape of the cell in each case. The initial cells are spherical, and they elongate when they thrive. An examination of wide regions under the fluorescence microscope show that the cells grow equally well on the colloidal crystal as they do on flat surfaces. The TIRF microscopy of the cells is compared for the two substrates. Figures 3a and 3b show epi-fluorescence and TIRF, respectively, for the cells on the bare silica coverslip. The latter is much weaker because only EGFP within the ;200 nm penetration depth of the evanescent wave is excited. By contrast, in epi-fluorescence, all of the EGFP, including that in the membrane and the organelles, is excited. For the cells grown on the silica coverslip bearing the thin colloidal crystal, Figs. 3c and 3d show that the epi-fluorescence and TIRF images of a cell that happens to be about the same orientation as that for Figs. 3a and 3b. The epi-fluorescence is again much brighter than the TIRF image, as expected. The behavior of epi versus TIRF with the colloidal crystals is indistinguishable from that for the bare silica coverslip, providing a demonstration of the ability of the colloidal crystal to allow TIRF microscopy.

614

Volume 62, Number 6, 2008

The accessibility of the cell surface to ligands in solution was investigated by observing the binding of the antagonist TIPP-cy5 to the human delta opioid receptor, in this case without an EGFP tag, in the cell membranes of Chinese hamster ovary cells. Figure 4a is a white light image of cells cultured on the colloidal crystal. These same cells show no appreciable autofluorescence throughout the entire cell at the cy5 emission wavelengths, as demonstrated in Fig. 4b, which was acquired using epi-illumination. Upon adding 1 nM TIPPcy5, the fluorescence now reveals the outlines of the cells, shown in Fig. 4c, as the antagonist binds to the delta-opioid receptors. The epi-illumination of Fig. 4c results in a fluorescence image upon which it is difficult to focus, and one cannot readily distinguish the upper membrane from the lower one. The TIRF illumination of Fig. 4d is tightly in focus, and it was found that there is only one focal plane, as expected for TIRF illumination. This demonstrates that ligand is able to diffuse through the pores of the crystal and bind to available receptors on the side of the cell facing the microscope coverslip. Figure 5 shows a representative comparison between TIRF microscopy images for Chinese hamster ovary cells grown on a silica colloidal crystal versus a fused silica coverslip. Each was exposed to a 0.5 nM solution of fluorescence-labeled antagonist (TIPP-cy5) for 15 minutes. The images for the colloidal crystal were typically two-fold brighter both on the edges of the cell and in the middle compared to the cells on the bare coverslip. The brighter edges of the cells are due to the vertical extension of the membrane, which gives more receptors per unit area at the cell edges. For the bare coverslip, there is expected to be a layer of water between the cell and the substrate, which allows for some access of the solution to the surface. The colloidal crystal is also expected to have this layer

FIG. 4. Fluorescence micrographs demonstrating access of the cell surface to the ligand through the colloidal crystal. All four micrographs are for the same region of the sample: (a) white-light image of Chinese hamster ovary cells on a colloidal crystal, (b) blank (no ligand) for epi-fluorescence image at the cy-5 emission wavelength, (c) epi-fluorescence image after exposure to ligand (1 nM TIPP-cy5 antagonist), and (d) same as (c), but using TIRF. Each image is 31 lm across.

of water between the silica and the cell, but note that the diffusion distance is much shorter, which allows for faster access. It is possible that the cell surface follows the contour of the colloids over much of their exposed surface area, which would limit access. Experiments are underway to pattern the surface of the colloids by stamping an adhesive agent, such as c-aminopropylsilane, on the top of the colloidal crystal to promote adhesive focal points on the tops of the nanoparticles. This would increase the area exposed to the solution. The results presented in Fig. 5 constitute proof-of-principle, and improvements in access are likely. In this work, choices were made for the system, including the colloidal crystal, the cells, and the numerical aperture of the objective, and we discuss the impact of each of these on the generality of the technique. First, we chose the colloidal crystalline layer to be less than 1 lm in thickness to minimize Bragg diffraction, which is strong for thick colloidal crystals.26 We found TIRF to be fraught with diffraction for crystals of a few micrometers in thickness, and these problems are eliminated by sub-micrometer thicknesses. This is a consequence of changing from a volume grating to a surface grating, and the weaker diffraction of the latter for colloidal crystals is described by theory.27 Second, we chose colloids on the order of 200 nm in diameter, for which the limiting interstitial pore size is calculated by geometry to be 30 nm. This is comparable to the pore size of wide-pore chromatographic silica gel, which was chosen to allow fast access of the ligand, or even proteins, to the cell surface through the colloidal crystal. The colloidal crystalline layer is much like high quality chromatographic silica gel: it is comparable in thickness to give transport in and out on the millisecond scale, and its surface can be modified to minimize adsorption of proteins and small ligands. One advantage of colloidal crystals over microfabrication of sub-lm pores28 is that the silica nanoparticles self-assemble into crystals, which is less expensive. They can be formed by spin-coating in minutes.29 An additional advantage of colloidal crystals over microfabricated holes is that the colloidal crystalline layer is much thinner than microfabricated holes, which is essential for the fast mass transport needed for reliable kinetic studies.

Third, we chose pure silica colloids rather than higher index colloids clad with silica. The latter would be more generally applicable because the index could potentially be tuned to match that of glass. The pure silica colloids are more readily available, and the results show that these are sufficient for TIRF. We observed that it was easy to achieve TIRF with these cells on the colloid surface, yet there is evidence that cells have refractive indices on the order of 1.38,30,31 which is considerably higher than the 1.33 value for water. An index of 1.38 for cells would give critical angles for glass–water and glass–cell of 62.18 and 65.48, respectively, which would reduce the range of usable angles for TIRF to only 58. There was no noticeable difficulty in adjusting the angle to achieve TIRF, although our instrument does not allow a quantitative measure of the range of angles. Either this narrow range of TIRF angles is not noticeable to the operator, or more likely, the refractive index of the cells within the evanescent wave is closer to that of water. Fourth, the colloidal layer is calculated to reduce the numerical aperture of the system from 1.45 to 1.43, and this lost range of angles is depicted in Fig. 1 as the thin gray triangle. Loss of the most extreme angles occurs because the light from the immersion oil does not couple into the lower index silica crystal. By principle of reversibility of path, the

FIG. 5. TIRFM images for CHO cells with 0.5 nM of TIPP-cy5 on (a) a silica colloidal crystal and (b) a fused silica coverslip. A higher sensitivity scale was used than that used for the images of Fig. 4. Each image is 29 lm across.

APPLIED SPECTROSCOPY

615

most extreme angles are not collected, which reduces sensitivity. Quantitatively, the angle of incidence from the glass into the silica for this unwanted total internal reflection between the silica coverslip and the silica colloids is 70.48. This reduces the range of angles collected by only 28. This loss would usually be a small price to pay for the ability to probe the cell surface with colloidal crystals of pure silica. Having a higher refractive index for the colloids would recover these lost angles and allow higher sensitivity, which might be important in applications using single-molecule spectroscopy. Further, one could use refractive indices even higher than that of glass. For TIRF microscopy, a microscope objective with numerical apertures of 1.65, along with coverslips and oil having an index of refraction of 1.78, are now commercially available.32 To enable use of colloids with such higher index objectives and substrates, higher index colloids could be used, in principle. Zirconia colloids (n ¼ 2.0),33 and core-shell composite colloids of silica–zirconia34 can be made to be monodisperse. One might tailor the volume fractions of the composites to match the refractive index of 1.78, which would allow the wide angular range for TIRF with biological cells (48.38 to 68.08). The same principles presented in this paper can be extended to the use of higher index colloids. The refractive index also affects the penetration depth. A glass (n ¼ 1.52) coverslip in contact with water or cells as the low index medium gives a depth of penetration of less than 150 nm for green excitation light. The present colloids (n ¼ 1.43) increase this distance to 180 nm. The new objectives with n ¼ 1.78 decrease this distance to 80 nm. For single-molecule spectroscopy, every step to shorten the penetration depth reduces the background, which is important for studies of biological cells. Higher index colloids might thus prove valuable. Finally, total internal reflection microscopy allows the use of polarization, which enables study of the orientation of bound ligands.9 It is presently not known to what extent the cell membrane bends to follow the contour of the colloidal crystal, and bending of the membrane would give an artificially broadened orientational distribution. Smaller colloids, or colloids with patterned surface coverages, could resolve these issues. A related question is what fraction of the surface area of the membrane is accessible to the solution versus in contact with the colloidal crystal. These questions can potentially be addressed by study of colloids of varying size. ACKNOWLEDGMENT This work was supported by the National Science Foundation (CHE0649508). 1. T. A. Vortherms and B. L. Roth, Idrugs 8, 491 (2005).

616

Volume 62, Number 6, 2008

2. W. R. Leifert, A. L. Aloia, O. Bucco, R. V. Glatz, and E. J. McMurchie, J. Biomol. Screening 10, 765 (2005). 3. I. D. Alves, C. K. Park, and V. J. Hruby, Curr. Pro. Pep. Sci. 6, 293 (2005). 4. W. Thomsen, J. Frazer, and D. Unett, Curr. Opin. Biotechnol. 16, 655 (2005). 5. A. Waller, P. C. Simons, S. M. Biggs, B. S. Edwards, E. R. Prossnitz, and L. A. Sklar, Trends Pharmacol. Sci. 25, 663 (2004). 6. N. L. Thompson and B. C. Lagerholm, Curr. Opin. Biotechnol. 8, 58 (1997). 7. B. Cannon, N. Weaver, Q. S. Pu, V. Thiagarajan, S. R. Liu, J. Y. Huang, M. W. Vaughn, and K. H. Cheng, Langmuir 21, 9666 (2005). 8. J. C. Conboy, K. D. McReynolds, J. Gervay-Hague, and S. S. Saavedra, J. Am. Chem. Soc. 124, 968 (2002). 9. N. L. Thompson and T. P. Burghardt, Biophys. Chem. 25, 91 (1986). 10. C. M. Ajo-Franklin, L. Kam, and S. G. Boxer, Proc. Natl. Acad. Sci. U.S.A. 98, 13643 (2001). 11. J. Avery, D. J. Ellis, T. Lang, P. Holroyd, D. Riedel, R. M. Henderson, J. M. Edwardson, and R. Jahn, J. Cell Biol. 148, 317 (2000). 12. K. L. Martinez, B. H. Meyer, R. Hovius, K. Lundstrom, and H. Vogel, Langmuir 19, 10925 (2003). 13. J. B. Perez, K. L. Martinez, J. M. Segura, and H. Vogel, Adv. Funct. Mater. 16, 306 (2006). 14. M. Tanaka, A. P. Wong, F. Rehfeldt, M. Tutus, and S. Kaufmann, J. Am. Chem. Soc. 126, 3257 (2004). 15. D. W. Lee, X. F. Wu, E. Eisenberg, and L. E. Greene, J. Cell Sci. 119, 3502 (2006). 16. E. A. Smith, J. W. Coym, S. M. Cowell, T. Tokimoto, V. J. Hruby, H. I. Yamamura, and M. J. Wirth, Langmuir 21, 9644 (2005). 17. T. Tokimoto, T. R. C. Bethea, M. Zhou, I. Ghosh, and M. J. Wirth, Appl. Spectrosc. 61, 130 (2007). 18. V. Jacquier, M. Prummer, J. M. Segura, H. Pick, and H. Vogel, Proc. Natl. Acad. Sci. U.S.A. 103, 14325 (2006). 19. Y. Sako, S. Minoghchi, and T. Yanagida, Nature Cell Biol. 2, 168 (2000). 20. T. Van Le, E. E. Ross, T. R. C. Velarde, M. A. Legg, and M. J. Wirth, Langmuir 23, 8554 (2007). 21. G. H. Bogush, M. A. Tracy, and C. F. Zukoski, J. Non-Cryst. Solids 104, 95 (1988). 22. W. Stober, A. Fink, and E. Bohn, J. Colloid Interface Sci. 26, 62 (1968). 23. E. Malatynska, Y. Wang, R. J. Knapp, S. Waite, S. Calderon, K. Rice, V. J. Hruby, H. I. Yamamura, and W. R. Roeske, J. Pharmacol. Exp. Ther. 278, 1083 (1996). 24. T. Okura, E. V. Varga, Y. Hosohata, E. Navratilova, S. M. Cowell, K. Rice, H. Nagase, V. J. Hruby, W. R. Roeske, and H. I. Yamamura, Eur. J. Pharmacol. 459, 9 (2003). 25. K. Befort, L. Tabbara, D. Kling, B. Maigret, and B. L. Kieffer, J. Biol. Chem. 271, 10161 (1996). 26. P. Jiang, J. F. Bertone, K. S. Hwang, and V. L. Colvin, Chem. Mater. 11, 2132 (1999). 27. I. I. Tarhan and G. H. Watson, Phys. Rev. B 54, 7593 (1996). 28. C. Danelon, J. B. Perez, C. Santschi, J. Brugger, and H. Vogel, Langmuir 22, 22 (2006). 29. A. Mihi, M. Ocana, and H. Miguez, Adv. Mater. 18, 2244 (2006). 30. N. Lue, G. Popescu, T. Ikeda, R. R. Dasari, K. Badizadegan, and M. S. Feld, Opt. Lett. 31, 2759 (2006). 31. B. Rappaz, P. Marquet, E. Cuche, Y. Emery, C. Depeursinge, and P. J. Magistretti, Opt. Exp. 13, 9361 (2005). 32. A. L. Mattheyses and D. Axelrod, J. Biomed. Opt. 10, 054007 (2005). 33. K. Lee, A. Sathyagal, P. W. Carr, and A. V. McCormick, J. Am. Ceram. Soc. 82, 338 (1999). 34. H. R. Chen, J. H. Gao, M. L. Ruan, J. L. Shi, and D. S. Yan, Micropor. Mesopor. Mater. 76, 209 (2004).

Comparison of Discrete and Continuous Motion in Scanning Probe Microscopy Monitored via Confocal Raman Microspectroscopy DANIEL P. CHERNEY* and DONALD A. WINESETT ExxonMobil Chemical Company, 5200 Bayway Drive, Baytown, Texas 77520

Confocal Raman imaging is a relatively new analytical technique that combines the strengths of Raman microspectroscopy and confocal optics. The images collected by the microscope are obtained by monitoring specific bands in the Raman spectra that are collected at many points in a sample, with the number of spectra usually numbering in the hundreds or thousands. Some commercially available systems acquire data while the sample is continuously moving with respect to the microscope objective. The distance that the stage moves during a single acquisition is a parameter that can be set prior to data acquisition. Data in this report was acquired with both a static and continuously moving sample for comparison, utilizing the 520 cm1 Si phonon of a silicon wafer to monitor an edge. Scattering collected from each discrete step, i.e., no motion during spectral acquisition, showed excellent precision of location, but a loss in resolution was observed as the pixel size was increased beyond the maximum theoretical resolution of the instrument. A continuously moving stage contributed to erroneous position data as the pixel size was increased beyond the maximum theoretical resolution of the instrument. Index Headings: Confocal microscopy; Resolution; Raman imaging.

INTRODUCTION Spectroscopic imaging has grown in popularity over the last two decades as microelectronics have been developed to control movement of a sample in conjunction with timing of data acquisition.1 The application of imaging to confocal Raman spectroscopy exploits the excellent spatial resolution gained through the use of confocal optics by providing the capability to chemically describe a sample in multiple dimensions. Raman images can be generated in at least one of three ways: point scanned,2 sequential-line scanned,3,4 or wide-field scanned.5,6 Since point-scanned imaging will be used for these experiments, sequential line and wide-field imaging will not be discussed here. Point-scanned imaging is performed by acquiring data, i.e., a spectrum, for each individual point at some user-defined spacing and acquisition time. Usually, the spacing between points is defined to optimize resolution, signal-to-noise ratio, and total acquisition time. Point scanning can be performed by an instrument in at least two ways: with discrete steps or with continuous motion of the sample with respect to the probe. In an image acquired with discrete steps, the stage and probe remain motionless during spectral acquisition; the spectrum is acquired and assigned to a pixel in an image and then the stage moves to the next location, stops during spectral acquisition, and so on. In an image acquired with a continuously scanning stage, the probe is moving with respect to the sample during spectral acquisition. The length of a pixel is traversed during Received 27 September 2007; accepted 3 March 2008. * Author to whom correspondence should be sent. E-mail: daniel.p. [email protected].

Volume 62, Number 6, 2008

the integration time for each pixel. However, movement of a probe during acquisition will alter the shape of the incident beam profile at each pixel in an image such that it is broadened in one dimension in comparison to the original Gaussian function7 that approximates the profile of a static acquisition. A comparison between a static profile and continuous beam profiles is displayed in Fig. 1. The line that describes the motionless probe is described as a Gaussian of 512 points with a r of 20 points, and a maximum point-intensity of unity. To demonstrate the effect of a continuously moving probe, the intensity profiles for four different step sizes were modeled using the identical probe shape as with the discrete steps (Gaussian of r ¼ 20). For example, the profile for a pixel size of ½ r is the shape calculated by summing from ¼ r (5 steps from center) to ¼ r (þ5 steps) for a total of 11 Gaussians each scaled by 1/11. The scaling of each individual Gaussian in the sum maintains the total integrated intensity between the functions that describe the static and continuous analyses. This example represents the over-sampled case and there is little change in the beam profile. The larger step sizes, such as 2r and 4r, demonstrate the degree to which the step size affects the effective probe shape. The factors affecting the shape of an intensity profile of a static beam and the resolution of a confocal Raman probe have been well described8–11 and are dependent on the wavelength of incident radiation and the optics in the system. Based on the profiles displayed in Fig. 1, it is clear that the step size in relation to the beam shape can have a pronounced effect on the resolution in a continuously scanned probe. Microscopic knife edge tests are routinely used to evaluate resolution in scanning probe microscopies and have been previously reported to describe the response of an SEM probe12,13 and to measure the mechanical response of a stage used for confocal microscopy.14 However, there is no existing study in the literature of which the authors are aware that reports any effects of acquiring data with discrete steps in comparison to continuous motion of a sample with respect to a probe. The point-spread profile will be obtained by monitoring the 520 cm1 band of a silicon wafer as a probe is scanned across the edge.15 This report will analyze the effects on lateral spatial resolution and intensity variations between continuous and discrete stepping in one dimension with a confocal Raman microscope.

THEORY A calculation of the intensity profile across an interface using two different methods of data acquisition, either with a static system that takes discrete steps or with a system with continuous motion of the probe with respect to the sample during acquisition, is desired. In each case, for computational

0003-7028/08/6206-0617$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

617

same dwell time: Z Z ‘ gðxÞf ðxÞ dx RHS ¼ dt Z ‘ ‘ 2 2 ð2pr2 Þ1=2 ef½xðs=2Þ g=ð2r Þ dx ¼t 0    pffiffiffiffiffiffiffiffiffiffiffi 1 s 2pr2 erfc  pffiffiffi ¼ tð2pr2 Þ1=2 2 r 8   t s ¼ erfc  pffiffiffi 2 r 8

FIG. 1. Spatial plot of the relative illumination intensity of the laser with the probe motionless (solid line) and scanned continuously over distances of 60.25r (gray line), 60.5r (dash-dot line), 61r (dotted line), 62r (dashed line), and 64r (dash-dot-dot line). The profile becomes wider as the distance scanned increases.

simplicity, we use a Gaussian profile to describe the shape of the incident beam, defined by:16 gðxÞ ¼ ð2pr2 Þ1=2 e½ðxaÞ

2

=ð2r2 Þ

ð1Þ

where r is the standard deviation of the Gaussian and a is the location of the maximum of the distribution. We then define a step size, s, stage velocity, v, and an interface at x ¼ 0 as a perfectly sharp step function:  0 x,0 f ðxÞ ¼ ð2Þ 1 x0 A profile across this interface consists of individual pixels distributed on either side of this interface, the two closest to the interface being x ¼s to 0 on the left-hand side (LHS) and x ¼ 0 to s on the right-hand side (RHS). The intensity profile across this interface will be calculated for both the discrete and continuous systems for various step sizes. For the discrete case, the total number of counts for the LHS pixel is calculated by integrating the total intensity detected from a probe centered over the pixel at a ¼ s/2 for a specific dwell time, t : Z Z ‘ LHS ¼ dt gðxÞf ðxÞ dx Z ‘ ‘ 2 2 ð2pr2 Þ1=2 ef½xþðs=2Þ g=ð2r Þ dx ð3Þ ¼t 0    pffiffiffiffiffiffiffiffiffiffiffi 1 s 2pr2 erfc pffiffiffi ¼ tð2pr2 Þ1=2 ð4Þ 2 r 8   t s ð5Þ ¼ erfc pffiffiffi 2 r 8 where erfc is the complementary error function. The integration in Eq. 3 utilizes formula 15.74 from Ref. 17. The RHS intensity is a comparable calculation except the probe is now centered over the pixel on the opposite of the interface at a ¼þs/2 for the

618

Volume 62, Number 6, 2008

ð6Þ ð7Þ ð8Þ

To determine the complete functional form for the discrete case (fd), we generalize to calculate the intensity measured n steps away from the interface in incremental displacements of s. This is accomplished by replacing the probe location at x ¼s/2 and x ¼ s/2 in Eqs. 3 and 6 above with x ¼ (n þ ½)s, where n is integer steps from ‘ to ‘, and solve similarly to obtain the general form for the discrete profile: 

   n þ 12 s t pffiffiffi fd ¼ erfc  ð9Þ 2 r 2 To better compare to the continuous case, we can substitute for the dwell time, t ¼ s/v: 

   n þ 12 s s pffiffiffi erfc  fd ¼ ð10Þ 2v r 2 Equations 9 and 10 are the intensity profile across a sharp interface for a discretely scanned probe for all points in space, n, in terms of probe size, r, velocity, v, and step size, s. For the continuous case, we must first establish the functional form of a profile at each point in space. Again, we approximate the probe as a Gaussian (Eq. 1) and calculate the response to the step function (Eq. 2): Z ‘ Z ‘ 2 2 hðxÞ ¼ gðxÞf ðxÞ dt ¼ ð2pr2 Þ1=2 e½ðxaÞ =ð2r Þ dt ‘

0

ð11Þ The position of the probe at any given point in time, a(t), is dependent on the velocity, or a(t) ¼ vt. Substituting this into Eq. 11 above and integrating over all time yields the expected profile in space: Z ‘ 2 2 hðxÞ ¼ ð2pr2 Þ1=2 e½ðxvtÞ =ð2r Þ dt ð12Þ 0 pffiffiffiffiffiffiffiffiffiffiffi   1 x 2pr2 erfc  pffiffiffi ¼ ð2pr2 Þ1=2 ð13Þ 2v r 2   1 x erfc  pffiffiffi ð14Þ ¼ 2v r 2   1 1 x þ erf pffiffiffi ð15Þ ¼ 2v 2v r 2 This is the expected result for a Gaussian convoluted across a step function. With this functional form, we are now able to calculate the integrated intensity for each side of the interface. Proceeding as before, the intensity is calculated on each side of the interface by integrating h(x) (Eq. 14) spatially over an entire

pixel on each side. Therefore, the intensity on the LHS is:   Z Z 0 1 0 x hðxÞ dx ¼ 1 þ erf pffiffiffi dx ð16Þ LHS ¼ 2v s r 2 s " #x¼0   rffiffiffi 1 x 2 ðx2 =2r2 Þ x þ x erf pffiffiffi þ ¼ ð17Þ re 2v p r 2 x¼s "rffiffiffi #   rffiffiffi 1 2 s 2 ðs2 =2r2 Þ ¼ ð18Þ r þ s þ s erf pffiffiffi  re 2v p p r 2 The intensity of the RHS is computed similarly:   Z Z s 1 s x RHS ¼ hðxÞ dx ¼ 1 þ erf pffiffiffi dx ð19Þ 2v 0 r 2 0 " # x¼s   rffiffiffi 1 x 2 ðx2 =2r2 Þ x þ x erf pffiffiffi þ ð20Þ ¼ re 2v p r 2 x¼0 " rffiffiffi #   rffiffiffi 1 s 2 ðs2 =2r2 Þ 2 s þ s erf pffiffiffi þ ¼  re r ð21Þ 2v p p r 2 Again, the derivation is expanded to the complete functional form for the continuous case (fc), by generalizing in the above equations n steps away from the interface in incremental displacements of s.   Z Z ðnþ1Þs 1 ðnþ1Þs x fc ¼ hðxÞ dx ¼ 1 þ erf pffiffiffi dx ð22Þ 2v ns r 2 ns " #x¼ðnþ1Þs   rffiffiffi 1 x 2 ðx2 =2r2 Þ x þ x erf pffiffiffi þ ð23Þ re ¼ 2v p r 2 x¼ns    1 ns þ s pffiffiffi ¼ s þ ðn þ 1Þs erf 2v r 2 rffiffiffi   2 f½ðnsþsÞ2 =ð2r2 Þg ns  ns erf pffiffiffi þ re p r 2 rffiffiffi 2 2 f½ðnsÞ =ð2r2 Þg þ re ð24Þ p Equation 24 is the intensity profile across a sharp interface for a continuously moving probe for all points in space (n ¼ ‘ to ‘) in terms of probe size, r, velocity, v, and step size, s. Equations 10 and 24 are the theoretical basis for which we compare our experiments in the remainder of this paper.

EXPERIMENTAL Confocal Raman Microscope. A CRM 200 (Witec Instruments) was used for all confocal Raman spectroscopy measurements. A diode laser producing light at 785 nm was coupled to a single-mode fiber optic and connected to an upright microscope. The light was passed through a beam expander and reflected off a dichroic mirror into an objective. Two objectives were used for analysis: an M Plan IR 1003, 0.95 NA objective (Olympus) and a Plan Fluor 403, 0.6 NA objective (Nikon). The 1808 backscattered light was collected through the same objective and passed through a filter to remove Rayleigh scattering. Raman scattered light was then brought to a secondary focus at either a 50 lm diameter fiber optic, serving as the confocal aperture, or a digital video device

for viewing. The digital video device could be used to collect optical images of a sample by illumination with a white light source. Raman scattered photons were transmitted through the fiber optic to a grating (600 lines/mm) and counted by a backilluminated, deep depletion charge-coupled device (CCD, Andor Technology) at 79 8C with a Peltier cooler and additional water cooling. Samples were placed on a piezodriven stage with lateral spatial accuracy of ,4 nm; the depth position accuracy is ,1 nm. The theoretical lateral spatial resolution of the 1003 objective was 0.33 lm and that of the 403 was 0.52 lm. These values are determined by equations describing the resolution of microscope objectives and confocal optics.18 Spectral Acquisition. A cleaved silicon wafer (single-side polished, ,100., Virginia Semiconductor, Inc.) was used as a sample for all spectral acquisitions. The wafer was mounted in a home-made vice to orient a flat cleaved edge of the wafer beneath the objective so that the step was aligned with the y-axis. For discrete step measurements, the line scan function in the Witec ScanCtrlSpectroscopy Plus software was utilized since the probe stops for collection of each spectrum in this mode. For continuous step measurements, the image function in the software was utilized since the stage moves during image acquisition. The laser power at the sample was approximately 5 mW and a 0.1 s or 1 s acquisition per pixel/point was utilized. Data was collected with a 1 s acquisition with the continuous scans to ensure that the read time of the detector (18 ms) was small compared to the acquisition time. For both the continuous and discrete tests, the laser was focused on the surface of the silicon wafer using the digital video device. The sample was rotated to ensure that the interface was oriented along the y-axis. The stage was then moved off the wafer for the beginning of data acquisition. For discrete step measurements, five separate line acquisitions at each step size were averaged to report a single profile. For continuous step measurements, square images were taken and averaged in the direction parallel to the edge to yield a single profile. Step sizes tested were 0.1 lm, 0.25 lm, 0.33 lm, 0.5 lm, 0.66 lm, 0.75 lm, 1 lm, 1.5 lm, 2 lm, 3 lm, 4 lm, and 5 lm. In both the line profile and imaging modes, the sample stage is moved while the objective remains static. Data Analysis. The area underneath the Si phonon centered near 520 cm1 was used for all analysis. No background subtraction or intensity normalization was applied to the spectra. Averaging of discrete profiles was completed in the software that controlled the CRM 200 and exported as an ASCII file. Continuous step files were analyzed by collecting spectra in imaging mode and using the area beneath the 520 cm1 band. The images were exported as .dat files and then imported into Interactive Data Language (IDL, v. 5.6, Research Systems, Inc.). Within IDL the image was averaged in the y direction to form an average line profile, with care taken to ensure the Si interface was parallel to the y direction in each image. Derivatives of each line scan (both continuous and discrete) were also calculated using IDL’s internal analytical functions. Modeling was performed using equations developed in the Theory section with Matlab 7.x (Mathworks).

RESULTS AND DISCUSSION The response of a focused laser beam passing onto a step function can be modeled by making two approximations: the shape of the point-spread profile, i.e., the energy density of the

APPLIED SPECTROSCOPY

619

FIG. 2. (Left) Diagram of a Gaussian function moving to the right, approaching a step function, and the integration of the two functions (dotted line). (Right) Raman spectrum of a silicon wafer. The area under the band at 520 cm1 was used for analysis.

laser at the focus, has a Gaussian shape7 and the interface of the silicon wafer can be modeled as a step function. Although the shape of the Airy pattern is described by a Bessel function,19 the addition of mathematical complexity is unnecessary for the approximation of resolution. A drawing of the experiment and expected response function as a Gaussian moves across a step function is displayed in Fig. 2 along with an example spectrum

FIG. 3. Overplot of the area beneath the 520 cm1 band with a static probe with (A) the 403, 0.6 NA objective and (B) the 1003, 0.95 NA objective. Data points were collected with distances ranging from 0.1 lm to 5 lm between acquisitions. All of the data points fall very close to the observed trend, regardless of step size. Points shown were acquired with step sizes of 0.1 (^), 0.5 (3), 1 (*), 2 (–), and 5 (m) lm.

620

Volume 62, Number 6, 2008

of silicon. Data recorded during the experiment will be reported similarly, where each point represents the intensity of the Si phonon in a silicon wafer centered at 520 cm1 at a reported location. The beginning position of the probe in each experiment was to the left of the silicon wafer. During the experiment with a static probe, spectra were acquired at discrete locations and the probe was moved in a specific increment towards the wafer. Successive spectra were acquired over a sufficient distance so that the final location monitored was several micrometers beyond the interface. The precision of the stage was also monitored by acquiring data across the same line (i.e., not changing the position in the y- or z-axes). The results (data not shown) demonstrate that the stage location is very precise such that any effects of moving and stopping the stage do not significantly affect the apparent location of the interface during any of the experiments. Several curves were also collected, using discrete steps, with various y-axis positions to monitor the vertical skew of the wafer over the same spatial domain as the acquisitions with continuous steps. The spot size of the laser with the 403 objective was approximately 0.79 lm in diameter, and the skew of the wafer was observed to be less than a micrometer in the x direction over several tens of micrometers of wafer in the y direction (data not shown). Since the total distance traveled in the y direction was less than 5 lm, it is likely that there was no significant effect on the results because of wafer skew. The area beneath the 520 cm1 band in experiments with discrete steps using both the 403 and 1003 objectives is shown in Fig. 3, showing data points collected at several step sizes. At least five curves were acquired for each to ensure that the response of the instrument was consistent and to ensure that the wafer was not moved significantly during acquisition due to stage motion. This figure demonstrates that the intensity of each point can be plotted with very good location precision, regardless of step size. The shape of the profile is significantly different from an expected error function. It is believed that the maximum observed is necessary because the derivative of the probe response will be indicative of the shape of the incident beam profile, an Airy function, as will be demonstrated shortly. The area under the 520 cm1 Raman band from a silicon wafer acquired with a probe moving continuously across the step function is shown in Fig. 4 for spectra acquired with the 403 and 1003 objectives. The step size, i.e., the width of each pixel, depends on the acquisition parameters that are set by the user. Regardless of pixel size, however, Raman scattering is

FIG. 4. Overplot of the area beneath the 520 cm1 band with a continuously moving probe with (A) the 403, 0.6 NA objective and (B) the 1003, 0.95 NA objective. Data points were taken with pixel widths ranging from 0.1 lm to 5 lm. Lines in (B) are shown to help the eye link data points and are not measured data. Points shown were acquired with step sizes of 0.1 (^), 0.5 (3), 1 (þ), 2 (&), 3 (filled ^), 4(m), and 5 (&) lm.

collected while the sample is being moved with respect to the objective. Unlike spectra that were acquired in a static system, seen in Fig. 3, there are some points that were acquired with larger pixel sizes that do not lie on the response function acquired at the smallest pixel sizes. Deviation from the response function is apparent near the maximum of the spectra collected with the 403 objective with a one micrometer pixel size; the observed response is approximately 10% lower than the overall maximum that is observed with the 1 lm pixel size. The measured intensity of the probe over space can be used to describe the point-spread function at each pixel size, as will be discussed shortly. The deviation from the expected response function is much clearer in images acquired with the 1003 objective. Some pixels acquired with larger step sizes are assigned counts several micrometers before the interface is actually reached and where no silicon is observed in the discretely stepping system. This shifts the apparent location of the edge to the left of where it is actually located. The effects of step size on resolution for data acquired with both discrete and continuous steps can be related to the apparent shape of the incident probe by taking the derivative of the response functions. The resultant point-spread functions for selected pixel widths are displayed in Fig. 5. The maxima of the derivatives were aligned so that the width of each function

FIG. 5. (A, B) Derivative of the response functions for each sampling increment shown in Fig. 4, relating to the effective intensity profile of the incident laser beam for each pixel size. Lines are included to connect the data points and show a smooth function for each pixel size. (C) Enlargement of the region near the peak intensity value shown in (B) to better illustrate the increase in the effective probe width as the pixel size is increased above the theoretical limit of 0.33 lm. Curves in (B) and (C) were shifted to approximately center their maximum intensities. Pixel sizes are 0.1 (^, solid line), 0.33 (–, dashed line), 0.5 (3, dotted line), 1 (þ, dash-dotted line), 2 (&, solid line), 3 (filled ^, dashed line), 4 (n, dash-dotted line), and 5 (&, solid line) lm.

could be compared. To obtain these functions, it was necessary to interpolate between points for some curves obtained with larger step and pixel sizes. It is clear that the effective shape becomes broader with increasing step size for data acquired with both discrete and continuous steps. This translates to an increase in the probe width and, therefore, a decrease in the

APPLIED SPECTROSCOPY

621

these two extremes, depending on where a feature is located with respect to the center of the probe during acquisition and the step/pixel size used for acquisition. The modeled range of 2r values is not significantly different between the moving and static probes until the step/pixel size increases beyond the theoretical limit of the probe. Once this value is exceeded, however, the smallest effective r of the probe is always larger for data acquired with a continuously moving probe than for data acquired with discrete steps. Since the Rayleigh resolution limit is dependent on the full width at half-maximum of the point-spread function, an increase in this width will lead to a loss of resolution, i.e., a greater spacing is required for two points to be resolved. Regardless of whether or not the probe is moving, collection of data with a step/pixel size set to a value larger than the theoretical resolution of the system may lead to inadvertent interpretation errors. This is an important reminder that the real data in a Raman image is the Raman spectrum assigned to a spatial location. Raman images are graphic representations of the collected data. Knowledge of the probe resolution and pixel size of an image are critical because of the tendency of the eye to associate adjacent pixels as being immediately adjacent in space as well. When stepping at or below the resolution of the instrument, this is approximately true, though spatial information is incorrectly assigned with larger pixel sizes acquired with a continuously moving probe. This incorrect spatial information worsens as the step/pixel size is increased beyond the maximum resolution of the system. FIG. 6. Experimental and modeled effective 2r values versus step/pixel size collected with (A) the 403, 0.6 NA objective and (B) the 1003, 0.95 NA objective. Experimentally observed 2r values ((&) 0.1 s discrete steps; (*) 0.1 s continuous motion; (n) 1 s continuous motion) were obtained by determining the distance between 84% and 16% of the maximum observed area under the 520 cm1 band at the interface of the silicon wafer. The dotted and dashed lines show the modeled effective 2r values for the discrete and continuously moving probe, respectively.

spatial resolution with increasing step size. The data indicates that the increase in probe width is observable immediately after the step/pixel size is increased beyond the maximum theoretical lateral resolution of the probe. The effective probe width can be determined by calculating the effective sigma from the measured response functions shown in Fig. 4. The value of 2r can be measured directly from the plots by determining the difference of positions where the measured response was 84% and 16% of the maximum observed signal. These distances are plotted in Fig. 6 along with the theoretical 2r values determined by using equations detailed in the Theory section and described below. There are two pairs of lines shown for the modeled 2r values for each objective displayed in Fig. 6. These are based on the extremes of possible locations of the probe center with respect to the interface during collection. The top lines display the value of 2r for the probe where the center of the probe is exactly at the interface during acquisition with discrete steps or is centered in the middle of the acquisition for the moving probe. The lower lines display the value where the center of the probe is exactly at 6s/2 during static acquisition and centered at 6s/2 during acquisition with the probe in motion. The apparent shape of the probe should be somewhere between

622

Volume 62, Number 6, 2008

CONCLUSION Confocal Raman images were collected by stepping the probe across a silicon wafer acting as an artificial edge. Spectra were acquired both with a static probe and a probe that was moving during acquisition. The effective probe width was found to increase after increasing the step size to a distance larger than the lateral resolution of the probe. Data acquired with discrete steps, i.e., no motion of the sample with respect to the probe, yielded lateral spatial information that was accurate regardless of the distance between acquisitions, though the effective probe size increased as the step size increased beyond the maximum resolution of the instrument. Data collected with continuous steps, i.e., motion of the sample with respect to the probe during acquisition, yielded incorrectly assigned lateral spatial information in addition to an increase of effective probe size when the pixel size was larger than the theoretical resolution of the instrument. For any scanning instrument, lateral resolution is lost (compared with the maximum theoretical resolution) when the probe is moved by a distance greater than the theoretical spatial resolution of the instrument; the best obtainable resolution is lessened as the pixel/step distance increases above the maximum resolution of the instrument. As most spectral imaging techniques use a probe with a Gaussian-like profile, this result is applicable not only to confocal Raman imaging, but to any method of point-scanned microscopy where spatial and compositional information is being assigned, such as scanning electron microscopy (SEM) or scanning tunneling electron microscopy (STEM) microprobe analysis, near-field scanning optical microscopy (NSOM), scanning transmission X-ray microscopy (STXM), and many other techniques.

ACKNOWLEDGMENTS The authors wish to thank A.-J. Bons, G. M. Brown, and J. H. Butler for their helpful comments on the manuscript. 1. P. J. Treado and M. P. Nelson, ‘‘Raman Imaging’’, in Handbook of Raman Spectroscopy, I. R. Lewis and H. G. M. Edwards, Eds. (Marcel Dekker, Inc., New York, 2001), Chap. 5. 2. J. J. Andrew and T. M. Hancewicz, Appl. Spectrosc. 52, 797 (1998). 3. M. Bowden, D. J. Gardiner, G. Rice, and D. L. Gerrand, J. Raman Spectrosc. 21, 37 (1990). 4. N. L. Jestel, J. M. Shaver, and M. D. Morris, Appl. Spectrosc. 52, 64 (1998). 5. P. J. Treado and M. D. Morris, Spectrochim. Acta Rev. 13, 355 (1990). 6. K.-L. K. Liu, L.-H. Chen, R.-S. Sheng, and M. D. Morris, Appl. Spectrosc. 45, 1717 (1991). 7. K. J. Baldwin, D. N. Batchelder, and S. Webster, ‘‘Raman Microscopy: Confocal and Scanning Near-Field’’, in Handbook of Raman Spectroscopy, I. R. Lewis and H. G. M. Edwards, Eds. (Marcel Dekker, Inc., New York, 2001), Chap. 4, p. 153. 8. N. Everall, Appl. Spectrosc. 54, 773 (2000).

9. N. Everall, Appl. Spectrosc. 54, 1515 (2000). 10. K. J. Baldwin and D. N. Batchelder, Appl. Spectrosc. 55, 517 (2001). 11. J. P. Tomba, L. M. Arzondo, and J. M. Pastor, Appl. Spectrosc. 61, 177 (2007). 12. K. W. Lee and J. T. L. Thong, Meas. Sci. Technol. 10, 1070 (1999). 13. D. C. Joy, ‘‘SEM Parameters and Their Measurement’’, Scanning Electron Microscope, 1974 (Part 1) Proceedings of the 7th Annual Scanning Electron Microscope Symposium, ITT Research Institute. 14. P. J. Cronin, P. W. Fekete, M. R. Arnison, and C. J. Cogswell, Rev. Sci. Instrum. 71, 118 (2000). 15. J. A. Fairfield, IEEE Proc. Nucl. Sci. Symp. Conf. Record, 2005 IEEE 2, 23 (2005). 16. J. N. Miller and J. C. Miller, Statistics and Chemometrics for Analytical Chemistry (Pearson Education Ltd., Harlow, England, 2005), 5th ed., p. 21. 17. M. R. Spiegel and J. M. Lin, Schaum’s Mathematical Handbook of Formulas & Tables (McGraw-Hill, Columbus, OH, 1999), 2nd ed. 18. http://www.olympusmicro.com/primer/opticalmicroscopy.html (accessed 4/23/2007). 19. G. R. Fowles, Introduction to Modern Optics (Dover Publications, Inc., New York, 1989), 2nd ed., Chap. 5.

APPLIED SPECTROSCOPY

623

Micro Attenuated Total Reflection Spectra of Bulk Silica Glass: Effects of Experimental Conditions and Glass Thermal History on Appearance of a Surface Polariton in the Si–O Stretching Region ANDREI A. STOLOV* and DEBRA A. SIMOFF OFS, Specialty Photonics Division, 55 Darling Drive, Avon, Connecticut 06001

Micro attenuated total reflection (micro-ATR) spectra of bulk silica glass were investigated for a variety of samples, including fused quartz slides, an optical fiber preform, and a series of optical fiber claddings. The experiments were performed at varied distances between the internal reflection element (IRE) and the sample. At certain conditions, a surface polariton peak is observed in the region 1100–1160 cm1. The position of this peak is affected by the type of IRE (Ge, Si, ZnSe, or diamond), IRE– sample distance, and the material used as an interlayer between the IRE and the sample (air or Nujol). From the experimental data, the dielectric constant of silica is determined in the region between 1100 and 1160 cm1. The polariton peak is also observed when glass is coated with a thin (40 nm) layer of carbon. It has also been found that the polariton peak position is affected by the thermal history of the glass, and an attempt is made to correlate the observed changes with the glass fictive temperature. Index Headings: Optical fiber; Silica; Surface polariton; Surface plasmon; Carbon coating; Fictive temperature; Attenuated total reflection; ATR; Optical contact.

INTRODUCTION Silica glass is the basic material of most optical fibers. Typical optical fibers contain a doped silica core, pure silica cladding, and a polymer coating. Some specialty fibers have a pure silica core and a polymer cladding, where the latter also plays the role of a coating. There are several applications of optical fibers that require partial coating removal. Important examples include fiber Bragg gratings and metal-plated pigtails, which are utilized in a variety of telecommunication devices.1–3 The coating can be removed either by mechanical stripping or chemically, by immersing in an acid. It is assumed that, after the coating is stripped off, there is negligible residual organic material and/or any other contamination at the glass surface. Cleanliness of the glass surface must be controlled and, if a residue is found, the root cause must be understood. Among other characterization techniques, infrared micro attenuated total reflection (micro-ATR) spectroscopy can be successfully used for detecting contaminations at the glass surface.4 An advantage of infrared (IR) spectroscopy is that it provides direct information about the material’s chemical structure. Micro-ATR attachments make the technique extremely helpful, since, on one hand, it is surface sensitive and, on the other hand, relatively small samples (a few micrometers in diameter) can be analyzed.5–9 An inherent property of ATR spectroscopy is the depth of penetration of the IR beam into the sample, typically around 1– 5 lm. If the thickness of a contaminant on the glass is less than

5 lm, it is highly probable that the obtained spectrum will contain bands of the silica. In order to interpret the spectra, bands belonging to the silica must be subtracted, or at least identified. Knowledge of ATR spectra of silica is also important when ATR spectroscopy is used in studying coatings or particles absorbed on silica glass,10–15 reaction kinetics on silica surfaces,16,17 or when characterizing oxide layers on silicon.18–23 Most Fourier transform infrared (FT-IR) applications use a spectrum subtraction procedure in order to eliminate the bands irrelevant to the study (e.g., bands from solvents or the measurement cell). One might assume that an ATR spectrum of pure silica could be obtained separately and then subtracted from each spectrum of the contaminated glass sample. However, based on depth profiling theory and experimental data, the shapes and intensities of the substrate bands depend on several factors, including contaminant thickness, its refractive index, and the contact quality between the internal reflection element (IRE) and the sample.24,25 In addition, surface polaritons (also called ‘‘surface plasmons’’) can be observed at certain experimental conditions.26–29 Thus, collecting a single ATR spectrum of pure silica may not be sufficient for subtraction purposes. Vibrational spectra of silica, including bulk amorphous SiO2 and its thin coatings, have been extensively studied using absorbance,30–42 reflection,43–53 ATR,54,55 and Raman techniques.56–58 Separate attention has been given to studying surface polaritons, especially for crystalline SiO2, using a prism coupling method.26,56–63 Among the approaches, the ATR technique is the most sensitive one to the experimental conditions, especially when polaritons are under investigation. A unique opportunity for studying polariton bands is provided by micro-ATR spectroscopy since this method comprises advantages of exploiting different IRE materials, enabling IRE–sample gap variation and using different IRE–sample interlayers. In this work, we performed a study of micro-ATR spectra of pure silica glasses under different experimental conditions (IRE type, contact quality) and using different interlayer materials (air and Nujol) between the samples and the IRE. Major attention is given to the Si–O stretching region, in particular to the surface polariton that is observable in that region. The results of the study are reported herein.

EXPERIMENTAL Received 4 September 2007; accepted 26 February 2008. * Author to whom correspondence should be sent. E-mail: stolov@ ofsoptics.com.

624

Volume 62, Number 6, 2008

Several pure silica samples were investigated: quartz slides (product of Chemglass, Inc, part number CGQ-0640-01), a

0003-7028/08/6206-0624$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

FIG. 1. The geometry for micro-ATR experiments.

fiber optic preform (Heraeus, Inc.), and glass claddings of uncoated optical fibers. All the studied fibers were products of OFS. The preform diameter was ;30 mm; the fiber diameters were in the range 125–1000 lm. A carbon-coated optical fiber was prepared by in-line carbon deposition onto a 125 lm diameter optical fiber. The carbon coating thickness, estimated from electrical resistance, was ;40 nm. Before use, the surfaces of the quartz slides and the preform were cleaned by washing with warm soapy water, followed by continuous flushing with acetone combined with mechanical wiping. A few studied optical fibers were intentionally drawn without a coating, so their surface did not require cleaning. For those fibers that contained a polymer coating, their coating was removed before collecting the spectra. Two means of removing the coating were utilized: (1) mechanical stripping followed by wiping with isopropanol, and (2) immersion in sulfuric acid. Both methods provided spectra similar to those obtained from the quartz slides, the preform, and the fibers drawn with no coating, indicating that the glass surfaces were clean within the detection limits of FT-IR. In some experiments, we used Nujol as an interlayer between the samples and the internal reflection elements. Nujol (paraffin oil, chemical formula CH3–(CH2)n–CH3, nD ¼ 1.478) is a product of Swan (NDC 0869-0831-43, purity 99.9%). For the same purpose we used 1-bromonaphthalene (C10H7Br, nD ¼ 1.657, product of ThermoSpectronic). Calibration of the internal reflection angles was performed using tripropylene glycol diacrylate (TPGDA, H2C¼CHCO(OC3H6)3O(CO)CH¼CH2, nD ¼ 1.451, product of Sartomer Company) and polystyrene (PS, (–CH2CH(C6H5)–)n, nD ¼ 1.60). Most of the spectra, including transmission, reflection, and micro-ATR spectra, were collected in the range 4000–600 cm1 using a Thermo-Fisher Nexus 670 Fourier transform infrared spectrometer equipped with an LN2 cooled mercury cadmium telluride (MCT) detector. The interferograms were averaged over 64 or 128 scans, Happ–Genzel apodized, and Fourier transformed with no zero filling factor to yield spectra at a resolution of 2 cm1. Reflection and micro-ATR spectra were collected using a Centaurus microscope attached to the FT-IR main bench. This microscope has a TritonTM objective with a numerical aperture of 0.71. For ATR, three single-bounce slide-on accessories were used: Ge (n0 ¼ 4.00), Si (n0 ¼ 3.42), and ZnSe (n0 ¼ 2.40). The IRE crystals have hemispherical probe surfaces with curvature radii of ;3.5 mm (Ge, Si) and ;7 mm (ZnSe). The geometry for micro-ATR experiments is shown in Fig. 1. Generally speaking, the sample is separated from the

IRE by a distance of zav. An interlayer (air or Nujol) fills the IRE–sample gap. In a particular case, the distance zav can be decreased to nearly zero, which corresponds to ‘‘full optical contact.’’ The microscope is designed so that the average internal reflection angle is ;458 for all three accessories. The internal reflection angle values were refined by collecting ATR spectra (at full optical contact) and transmission spectra of TPGDA, Nujol, and PS and comparing their intensities in those spectra. Ideally, ‘‘full optical contact’’ is a situation in which there is no gap between the IRE and the sample at the spot where the IR beam falls at the IRE–sample interface. Such contact is easily achievable for liquids and more difficult for solid samples, such as a polystyrene film. We use IR peak intensities as indicators of the contact quality. To approach ‘‘full optical contact,’’ we utilize flat samples and take measurements at the highest allowed force between the IRE and the sample. To ensure that ‘‘full optical contact’’ is achieved (within the detection limits of FT-IR) we utilized the Ekgasit method.24 In this method, a liquid interlayer (in our case, Nujol) is introduced between the IRE and the sample. For ‘‘full optical contact,’’ this should not cause any intensity change, while if there is a gap between the IRE and the sample, the peak in question would become more intense (because of an increase of the refractive index of medium 1). The extinction coefficients ek were determined from the transmission spectra: ek ¼

Dk l

ð1Þ

where Dk is absorbance of the peak at a wavelength k, and l is the sample thickness. The only peaks selected for the study were those whose absorption index kk was below 0.105 (kk ¼ ek ln(10)/(4pm), where m is the band wavenumber) since in those cases the extinction coefficient concept is applicable.64 Those selected peaks are 2978, 1637, and 1453 cm1 (TPGDA), 1459 and 1377 cm1 (Nujol), and 3082, 3059, 2849, 1601, 1155, 1068, 1029, and 907 cm1 (PS). The transmission spectra of TPGDA and Nujol were obtained using demountable cells with CaF2 windows. The cell path lengths were 30, 40, 108, and 260 lm; the accuracy of the measurements was 64 lm. The polystyrene film used in the study had a thickness of 31.2 6 0.3 lm. In micro-ATR spectroscopy, the extinction coefficient is determined from the following expression:65,66 ek ¼

logð1=Rk0 Þ dek

ð2Þ

where Rk0 is the reflectance at full optical contact and dek is the effective thickness of the sample. The effective thickness is determined via the expression dek ¼

k  n0k n2k cosh  fk pðn20k

 n22k Þðn20k sin2 h  n22k Þ1=2

ð3Þ

where k is the wavelength in vacuum, h is the angle of incidence, n0k and n2k are refractive indices of the IRE and the sample, respectively, and fk is a factor responsible for polarization. Since we used non-polarized spectra, we assume that65

APPLIED SPECTROSCOPY

625

FIG. 2. Micro-ATR spectra of a quartz slide obtained using a Ge IRE. The applied force increases from bottom (;0.0 N) to top (2.0 N). An offset of 0.02 absorbance units is applied between the spectra. The arrow corresponds to the developed peak at ;1150 cm1.

" # 2n20k sin2 h  n22k 1 1 1þ 2 fk ¼ ð fk? þ fkjj Þ ¼ 2 2 ðn0k þ n22k Þsin2 h  n22k

ð4Þ

where fk|| and fk? are the factors for p- and s-polarized light. Combining Eqs. 1–4 and assuming that the ek magnitudes determined from ATR spectra match those determined from transmission spectra, we were able to compute the angle of incidence, h. The obtained magnitudes are 46.3, 42.4, and 47.08 for Ge, Si, and ZnSe IREs, respectively. The contact quality between the IREs and samples was regulated manually by raising the sample stage against the IRE. To avoid excessive forces, the Centaurus microscope is equipped with a Contact AlertTM system. The highest force applied in the study was 2.0 N. The diameter of the sample area (i.e., the beam width at the IRE–sample interface) was determined using a 140 lm diameter optical fiber with hard polymer coating, as described in our previous paper.25 With a fully open microscope aperture, we found that at the IRE–sample interface the beam is 35 lm wide. Besides the micro-ATR spectra, we obtained reflection and transmission spectra for several samples. Reflection spectra from optical fibers were collected using the Centaurus microscope of the Nexus 670. The reflection angle was 908. Transmission FT-IR spectra were obtained using the main optical bench of the Nexus 670. In addition, a few micro-ATR spectra were obtained using a Bio-Rad FTS 135 spectrometer equipped with a UMA 500 microscope, which has a single-bounce slide-on ATR accessory with a diamond IRE (n0 ¼ 2.42). The operational surface of the diamond IRE is flat, and the average internal reflection angle is 44.58. The sample-area diameter for this micro-ATR unit is 70 lm, as determined previously. The interferograms were averaged over 128 scans, medium Norton–Beer apodized, and Fourier transformed with a zero filling factor of 2 to yield spectra at a resolution of 4 cm1.

RESULTS AND DISCUSSION Assignment of Bands in the Attenuated Total Reflection Spectra. We investigated micro-ATR spectra of several silica

626

Volume 62, Number 6, 2008

FIG. 3. Micro-ATR spectra of a quartz slide obtained using a ZnSe IRE. The applied force increases from bottom (;0.0 N) to top (2.0 N). An offset of 0.02 absorbance units is applied between the spectra. The arrow corresponds to the developed peak at ;1150 cm1.

samples, including a quartz slide, optical fiber preforms, and glass claddings of several optical fibers. The main features of the spectra obtained from those samples were quite similar. For convenience, we will focus our attention below on describing ATR spectra from a quartz slide, refering to peculiarities of the other samples where appropriate. According to the literature,38–40,51,52 only two absorption bands for silica fall in the range 4000–600 cm1 of IR transmission spectra. The strongest observable band, mAS ; 1070–1100 cm1, represents a transverse optical (TO) asymmetric stretching mode, while a weaker absorption, mS ; 800 cm1, belongs to a TO symmetric stretching mode. Depending on the measurement technique, the band positions, shapes, and intensities may vary significantly. Figure 2 shows ATR spectra of a quartz slide obtained using the Ge IRE at different applied forces. The higher applied force corresponds to better optical contact, i.e., smaller average distance between the IRE and the sample. At the highest applied force (top trace in Fig. 2), one can see the peaks at 1038 and 802 cm1. The 1038 cm1 peak has a broad highfrequency shoulder at about 1200 cm1. Using a different IRE material and/or changing the internal reflectance angle leads to significant changes in the ATR spectra. Thus, Fig. 3 shows spectra of the same slide obtained using the ZnSe IRE. At the best optical contact (top trace in Fig. 3), the mAS and mS peaks shift to 980 and 779 cm1, respectively. The peaks also become much broader, so that they strongly overlap. The peak maxima positions and the observed band widths (full widths at half-maximum) are given in Table I. At lower applied forces, when the IRE separates from the sample, one can see that a relatively narrow peak develops at ;1150 cm1 (Figs. 2 and 3). Appearance of this peak was observed with all IREs used and for all silica samples studied. Upon increasing the gap between the IRE and the sample, the 1150 cm1 peak intensity passes through a maximum and then drops to zero together with the main absorption peak. Examples of such spectra are shown in Figs. 4b and 4d. It can be seen that at certain conditions the 1150 cm1 peak may be as strong as mAS. On the contrary, when the IRE approaches the sample, the peak intensity decreases and, below a certain distance, the 1150 cm1 peak vanishes.

TABLE I.

Spectroscopic parameters of bulk silica observed with different internal reflection elements at the best optical contact.

IRE

n0

Internal reflection angle (8)

Diamond ZnSe Si Ge

2.42 2.403 3.418 4.003

44.5 47.0 42.4 46.3

Peak position (cm1)

Band width (cm1)

mAS

mS

mAS

mS

969 980 1026 1038

778 779 798 802

194 156 103 74

71 63 61 59

To understand the nature of the 1150 cm1 peak, we compared the IR spectra of silica obtained by different techniques. Figure 5 compares the reflection, transmission, and some of the micro-ATR spectra of silica. The reflection spectrum was collected using the Centaurus microscope, at an incidence angle of 908. For the transmission spectrum, a Nujol mull of the material was prepared and placed between two KBr disks. Although the spectrum of the Nujol mull is not a ‘‘quantitative’’ one, it provides representative positions of the absorbance peaks. Transmission spectra of silica obtained at much better quality can be found elsewhere.31,51,67–69 Absorbance in the 1300–1000 cm1 range of silica is the subject of extensive investigation. In transmission spectra of thin SiO2 films obtained at normal incidence of the IR beam, a clear shoulder is observed at ;1150 cm1.30,51,68,69 The same shoulder was reported for a silica powder in a KBr pellet47 and in a Nujol mull (Fig. 5). It was suggested38,51 that the asymmetric Si–O–Si motion gives rise to two vibrational modes, manifesting at 1072 cm1 (strong peak) and ;1147 cm1 (the shoulder). In transmission spectra obtained at a 308 angle of incidence, the longitudinal optic (LO) mode at 1254 cm1 and a TO–LO pair at 1200 and 1170 cm1, respectively, were observed.39 In reflection spectra, the mAS peak shows up at ;1120 cm1.45,51,67–69 The behavior of the 1150 cm1 peak observed in ATR spectra clearly indicates that this mode is located at the sample surface; therefore, it cannot be assigned to the bulk TO or LO modes. It is known that the absorptivity of the mAS fundamental

(observed in transmission spectra at approximately 1100 cm1) is very high: ek ; 1.5 lm1.31,51,67–69 It follows from the Kramers–Kronig relation that the refractive index of silica and its dielectric constant e2 decrease at frequencies above 1100 cm1. At negative dielectric constants, the material behaves similarly to a metal, so that specific surface modes, called surface polaritons or surface plasmons, can develop.26–28 Based on the literature data,26,59–63 we attribute the ;1150 cm1 peak in ATR spectra to one such mode. Surface polaritons are non-radiative waves localized at the boundary between media having one positive and the other negative dielectric constant. The electric vector of a surface polariton mode is perpendicular to the surface, so polaritons can interact only with p-polarized radiation. Polariton peaks are observable when using non-polarized light (as in our experiments) since it contains a p-polarized component. Polariton modes are localized in a relatively thin layer, and they decay exponentially into both media (in our case, silica and air). The surface polariton peaks are known to be observed in the range between the LO and TO modes.62,63 Indeed, the 1150 cm1 feature is located in the middle of the range between the TO and LO modes of silica (1072 and 1257 cm1 , respectively).51 Generally speaking, a similar effect is expected for the symmetric stretching bands at ;800 cm1. However, we have not noticed any peculiar features in that spectral range. The absence of a polariton band in this range can be explained as follows. First, the TO and LO stretching symmetric modes

FIG. 4. Micro-ATR spectra of a quartz slide obtained at poor optical contact with Nujol and air interlayers. (a) ZnSe IRE, Nujol interlayer. The spectrum is shifted upwards by 0.13 absorbance units. (b) ZnSe IRE, air gap. The spectrum is shifted upwards by 0.10 absorbance units. (c) Ge IRE, Nujol interlayer. The spectrum is shifted upwards by 0.02 absorbance units. (d) Ge IRE, air gap. Asterisks show the intrinsic peaks of Nujol.

FIG. 5. Comparison of FT-IR spectra of silica obtained by different techniques. (a) Reflection spectrum of a quartz slide. (b) Transmission (absorbance) spectrum of ground quartz mull in Nujol between two KBr plates. The Nujol bands are subtracted. Micro-ATR spectra of a quartz slide obtained with a ZnSe IRE at (c) good optical contact and (d) poor optical contact. The yscales of the plots are normalized arbitrarily. No corrections are applied to any of these spectra.

APPLIED SPECTROSCOPY

627

are located at 800 and 818 cm1, respectively.51 Thus, the separation between the TO and LO modes is much less than the TO band width (Table I). Second, the band absorptivity is much lower than that for the asymmetric mode, which makes the dielectric function changes in the LO–TO range less prominent. Nevertheless, it is worth noting that for crystalline quartz (whose structure and spectra are different from those of the amorphous samples studied here), a surface polariton peak in the 800 cm1 region is observable.60 It should be noted that the assignment of the 1150 cm1 peak to a surface polariton is entirely based on studying the spectral transformations as a function of the IRE–sample distance. If the studied glass surface is contaminated by an organic compound, the 1150 cm1 feature can be easily confused with absorption of C–F or C–O covalent bonds. If one obtains only a single ATR spectrum at a non-zero gap between the IRE and the sample, it would be hard to discern whether the 1150 cm1 peak is due to the substrate silica or to an unknown contaminant. Moreover, disappearance of the 1150 cm1 peak with improved IRE–sample contact could be misinterpreted as thinning the contamination layer, followed by its removal from the tested area to the sides (see Fig. 1). Another indirect consequence of the very strong absorptivity of mAS is a significant difference in the appearance of this band in the ATR spectra obtained with different IREs (see Figs. 2 and 3 and Table I). Thus, at the best achievable optical contacts, the peak maximum shifts from 969 to 1038 cm1 when changing from the diamond to the Ge IRE. Pronounced changes in the band shape and its position are due to considerable variation of the refractive index of silica, n2k, in the range 1200–900 cm1. According to theoretical estimates,51 the refractive index reaches a minimum value of ;0.33 at ;1130 cm1 and a maximum value of ;2.9 at ;1040 cm1. Since n2k affects the IR beam penetration depth and the effective thickness of the sample (Eq. 3), it transforms the ATR band envelope, shifting its maximum to lower frequencies (with respect to the absorption spectrum). Stronger shifts are expected for IREs with lower refractive indices (i.e., diamond and ZnSe), which we indeed observe. Moderate shifts are observed with the Si IRE and the lowest shift with the Ge IRE. Estimation of the Internal Reflection Element–Sample Distance. Surface polaritons are excited and probed by the ATR evanescent waves. A necessary condition for observing the polariton modes is a gap between the IRE and the sample.26,59–61 The size of the gap, zav (see Fig. 1), can be correlated with the polariton peak intensity. We assume the coordinate system bound to the IRE, as shown in Fig. 1. The evanescent wave electric field Ek decays in the interlayer (medium 1 of Fig. 1), which in our case is air. The decay is described by the following equation:65,66 Ek ¼ E0  expðzav =dp1k Þ

ð5Þ

where E0 is the electric field at the IRE–medium 1 interface and dp1k is the penetration depth in medium 1: dp1k ¼

k 2pðn20k sin2 h

 n21k Þ1=2

ð6Þ

In the region 1300–700 cm1, medium 1 is IR transparent while the sample (medium 2) absorbs IR energy. The ATR spectra can be collected at various zav. A particular case is full optical contact, when zav ¼ 0. We denote the reflectivity

628

Volume 62, Number 6, 2008

FIG. 6. Intensity of the polariton peak as a function of the IRE–sample distance. The results are obtained for a quartz slide with an air interlayer for different IRE types.

observable at full optical contact by Rk0. The fraction of energy absorbed by medium 2 at that condition will be (1  Rk0). If zav 6¼ 0, then the corresponding fraction of absorbed energy, (1  Rk), can be obtained as follows: ð1  Rk Þ ¼ ð1  Rk0 Þ  expð2zav =dp1k Þ

ð7Þ

The value that is measured by the spectrometer, which is similar to the absorbance for transmission spectra, is log(1/Rk). For the case zav 6¼ 0, it will be logð1=Rk Þ ¼ log½1  ð1  Rk0 Þexpð2zav =dp1k Þ From this, the magnitude of zav can be evaluated:   dp1k 1  Rk0 ln zav ¼  2 1  Rk

ð8Þ

ð9Þ

Two things are worth noting with regard to Eq. 9. First, this equation is valid for both weak and strong bands since it is not based on using the ‘‘extinction coefficient concept.’’64 Thus, very strong TO bands of silica can be utilized for evaluating zav. Second, Eq. 9 provides accurate prediction of zav only for flat and parallel IRE and sample interfaces (assuming that the refractive indices of the media are known). It has been shown previously that the shape of the sample can affect the ATR intensities as well.25 The flatter the sample, the more accurate the estimate of zav calculated via Eq. 9 should be. While for strongly curved or non-parallel surfaces the calculated zav may not be accurate, it still represents a measure of the IRE–sample distance and can be utilized when comparing similar systems (such as the same sample when analyzed with different IREs or using different interlayers). The use of zav in correlating spectroscopic parameters of the polariton band with various experimental conditions will be illustrated in the following sections. We calculated the magnitudes zav from the intensity of the mS band (;800 cm1). The polariton peak intensity was determined in the log(1/Rk) scale with respect to a baseline drawn between ;1280 cm1 and the minimum point at ;1110 cm1 (see Figs. 4b and 4d). Figure 6 shows dependences of the polariton peak intensity on zav obtained for the quartz slide. For all of the IRE types, the curves have clear maxima. At larger

FIG. 7. Micro-ATR spectra of a quartz slide obtained using a Ge IRE and a Nujol interlayer. The applied force increases from bottom (;0.0 N) to top (2.0 N). An offset of 0.02–0.05 absorbance units is applied between the spectra. The arrow corresponds to the developed peak at ;1120 cm1; the asterisks indicate the intrinsic peaks of Nujol.

IRE–sample gaps, the evanescent wave from the IRE is weak at the glass surface, causing a decrease in the observed band intensity. On the contrary, for zav!0, the range where the polariton wave decays disappears completely, and the surface waves become unobservable. For each type of IRE, the experimental points can be well fit by the following function: logð1=Rk Þ ¼ a  z3av  expðzav =bÞ

ð10Þ

where a and b are adjustable coefficients. The dependences exhibit maxima at ;0.30, 0.33, and 0.58 lm for Ge, Si, and ZnSe IREs, respectively. These magnitudes are comparable with the penetration depths in air for the same IREs: 0.51, 0.66, and 0.96 lm, respectively. Effects of the Interlayer Dielectric Constant. For surface polaritons, the following dispersion relationship is known:26–28 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x 0 e10  e2x kSk ¼  ð11Þ 0 0 e1 þ e2x c where kSk is the wave vector, x ¼ (2pc/k) is the angular 0 frequency, and e10 and e2x are the real parts of the dielectric constants of media 1 (air) and 2 (silica glass), respectively. The 0 subscript x in e2x indicates that this value is frequency dependent. It follows from Eq. 11 that the wave propagation can be affected by the dielectric constant of medium 1. In addition to air, we utilized Nujol as an interlayer between the IRE and the silica. The advantage of Nujol is that it is TABLE II.

transparent in most of the infrared region, including the 1300– 750 cm1 range. The ATR spectra obtained with a Ge IRE are shown in Fig. 7. The positions of the intrinsic bands of Nujol (;1460, ;1380, and ;720 cm1) are marked by asterisks. Their intensities are related to the gap thickness (zav); thus the bands almost vanish at the best achievable optical contact between the glass and the IRE. With the Nujol interlayer, the surface polariton peak is shifted to lower frequency (;1130 cm1) and the band profile changes. The most significant differences between the spectra can be seen at larger IRE– sample distances, as shown in Fig. 4. The highest relative intensity of the polariton mode is observed with the Ge IRE and a Nujol interlayer (Fig. 4c). Polariton peak positions observed with air and Nujol interlayers are given in Table II. All of the frequency values in Table II correspond to the largest magnitudes of zav (weakest optical contact) at which the polariton peak is still observable with a reasonable signal-tonoise ratio. Dependences of the band frequencies on zav will be considered below. Using the above approach, we transformed the mS peak intensities into IRE–sample distances, zav. Figure 8 shows the polariton peak intensity plotted against zav, obtained with a Nujol interlayer. The shapes of the curves are similar to the ones obtained with air as the interlayer. The main difference is that the maxima are observed at larger values of zav: 0.32, 0.53, and 0.76 lm for Ge, Si, and ZnSe, respectively. These magnitudes correlate with the penetration depth values in Nujol: 0.57, 0.80, and 1.51 lm. Dielectric Constant of Silica in the LO–TO Frequency Region. Spectral information obtained using different IREs and interlayer materials can be utilized for measuring the dielectric constant of the glass sample in the LO–TO frequency region. The evanescent wave from the IRE propagates with a momentum kPk, which is described as:29,59–61 x kPk ¼ n0k   sinh ð12Þ c From the constraints of energy and momentum conservation, the resonance attenuation occurs when kSk ¼ kPk. From this, and taking into account Eqs. 11 and 12, the real part of the dielectric constant of the glass can be calculated: 0 ¼ e2x

e10 n20k  sin2 h e10  n20k  sin2 h

ð13Þ

0 The value e2x determined via Eq. 13 corresponds to the frequency at which the polariton is observed. The results of the 0 calculations are shown in Table II. The dependence of e2x upon 0 magnitude was frequency is shown in Fig. 9. The lowest e2x observed when using the diamond IRE in combination with Nujol. Additional data points can be collected if an interlayer

0 ) Observed polariton peak frequencies (mP) and calculated dielectric constant of silica (e2x

IRE

n0

Internal reflection angle (8)

Diamond ZnSe Si Ge

2.42 2.403 3.418 4.003

44.5 47.0 42.4 46.3

Interlayer–air (n1 ¼ 1)

Interlayer–Nujol (n1 ¼ 1.478)

mP (cm1)

0 e2x

mP (cm1)

0 e2x

1141.4 1145.4 1156.5 1152.5

1.53 1.48 1.23 1.14

1101.6 1107.0 1123.1 1126.2

9.10 7.48 3.71 2.96

APPLIED SPECTROSCOPY

629

FIG. 8. Intensity of the polariton peak as a function of IRE–sample distance. The results are obtained for a quartz slide with a Nujol interlayer and different IRE types.

with a different dielectric constant is used. Of special interest is the range from 1100 cm1 to the TO absorption peak frequency of 1072 cm1. To collect the data at ,1100 cm1, we attempted to use 1-bromonaphthalene (nD ¼ 1.657) as an interlayer. However, at that high magnitude of n1, the polariton band was not distinguished in the spectra. This should be attributed to a significant overlap of the red-shifted and broadened polariton peak with the mS band of silica. Effects of Carbon Coating. The phenomenon of surface polaritons is widely used in optical sensors for chemical and biochemical analysis.70 One of the popular sensor configurations uses a prism coupler method. In this configuration, a light wave is totally reflected at the interface between a prism coupler and a thin (;50 nm) metal layer and excites a polariton wave at the outer boundary of the metal by evanescently tunneling through the thin metal layer. We attempted to observe a similar effect for amorphous silica coated with a thin conducting layer. For this purpose we utilized a carbon-coated optical fiber. Carbon-coated optical fibers represent an important class of optical waveguides. The role of the carbon coating is to seal the silica fiber, preventing it from stress corrosion due to possible water and/or hydrogen ingression at the glass surface. With carbon coatings, the fibers are known to have superior resistance to harsh chemical environments.71,72 Carbon coating is applied to the glass cladding as an optional step during fiber draw and the carbon has a structure similar to graphite. The optimal thickness for such coating is ;40 nm (as used in our study). We have obtained micro-ATR spectra of two similar optical fibers: one with and another without carbon coating. The fiber diameters were 125 lm. Figures 10 and 11 compare the obtained spectra of the carbon-coated and uncoated fibers. It is clearly seen that the polariton band is observable even with the carbon coating. The effects of the carbon are (1) band broadening, (2) decrease of the peak intensity, and (3) a shift to higher frequencies. Assessment of Glass Fictive Temperature. Band positions in vibrational spectra of silica are sensitive to its molecular structure. One of the structure-related parameters is the glass fictive temperature. The fictive temperature is defined as the equilibrium temperature at the moment of solidification. It is

630

Volume 62, Number 6, 2008

FIG 9. Dielectric constant of bulk silica, calculated from Eq. 13.

known that the fictive temperature affects the average Si–O–Si bond angle, and therefore the vibrational frequencies. Linear correlations have been established between the fictive temperature and the position of the mAS peak in reflection spectra and its first overtone in absorption spectra of silica.53,68,69,73–77 Since the position of the entire band envelope is affected by the fictive temperature, we expected that this would also be the case for the polariton peak. It is of interest to determine whether the polariton peak position can be used for evaluating the glass fictive temperature. A definite advantage of utilizing the polariton band is its smaller width as compared with the mAS peak in all types of infrared spectra (see Fig. 5). Before studying possible effects of the glass fictive temperature, we needed to analyze other factors that might influence the polariton peak position in the spectra. It was found previously that the polariton peak position is affected by the size of the IRE–sample gap.26 We have obtained the polariton band frequencies for various experimental conditions. The peak positions were determined only from the maxima of log(1/R) in the spectra. Although at some conditions the polariton band strongly overlaps mAS, no attempts to separate

FIG. 10. Micro-ATR spectra of optical fibers with and without carbon coating. All spectra were obtained with a Ge IRE using an approximately 1 lm gap between the sample and the IRE. (a) Carbon-coated fiber, Nujol interlayer. (b) No carbon coating, Nujol. (c) Carbon coating, air interlayer. (d) No carbon coating, air interlayer. The asterisks indicate the intrinsic peaks of Nujol.

FIG. 11. Micro-ATR spectra of optical fibers with and without carbon coating. All spectra were obtained with a ZnSe IRE using an approximately 1 lm gap between the sample and the IRE. (a) Carbon-coated fiber, Nujol interlayer. (b) No carbon coating, Nujol. (c) Carbon coating, air interlayer. (d) No carbon coating, air interlayer. The asterisks indicate the intrinsic peaks of Nujol.

FIG. 13. Micro-ATR spectrum of an ‘‘as drawn’’ silica fiber. The spectrum is obtained at a poor optical contact using a Nujol interlayer between the fiber and a ZnSe IRE. (a) The original spectrum. (b) The same spectrum after smoothing. (c) The first derivative of the smoothed spectrum. The y-scale and the offsets are arbitrary.

the bands were made since the shapes of the individual components are unknown. The obtained results are shown in Fig. 12. As can be seen, the polariton frequencies strongly depend on zav. Thus, for a Si IRE combined with an air interlayer, the polariton band shifts by almost 25 cm1 when changing the gap from 0.2 to 1.2 lm. For comparison, typical frequency shifts due to the fictive temperature difference do not exceed 5 cm1. It has also been found that the band position is sensitive to the alignment of the sample relative to the IRE. The geometry shown in Fig. 1 tacitly assumes that the sample (fiber) is located directly below the IRE. However, if the fiber is misaligned, this creates a difference in the average orientation of the IRE and sample surfaces, which influences the projection of kPk parallel to the glass surface (Eq. 12). This, in turn, should affect the peak frequency and intensity. Since there are at least two additional factors affecting the polariton peak frequency, the fictive temperature shifts can be investigated only if some ‘‘standard’’ experimental condition is found at which all the effects other than the fictive temperature

cancel out. Among different IRE–interlayer pairs, we have chosen the ZnSe–Nujol pair. For this condition, the polariton band is observed at the largest zav so that possible effects of the sample misalignment are the smallest. In addition, as can be seen in Fig. 12, the frequency dependence upon zav is the weakest when the ZnSe IRE is combined with Nujol. For the experiments we used two optical fiber samples with pure silica cladding. The cladding diameters were each 125 lm. The first sample was investigated ‘‘as drawn,’’ i.e., without further annealing. It was believed that its fictive temperature was close to its draw temperature, i.e., ;1600 8C. Before collecting micro-ATR spectra, the fiber coating was removed. To investigate possible effects of the coating removal conditions for this particular fiber, we applied two approaches as described in the Experimental section: (1) mechanical stripping followed by isopropanol wiping, and (2) immersion in sulfuric acid. The second sample was obtained from the same fiber by annealing it at 1150 8C for 3 hours. Before the annealing, the coating was removed by immersing the fiber in sulfuric acid. Using the microscope imaging system, both fiber samples were carefully aligned at the microscope stage so that the fiber axis was directly below the center of the IRE. The misalignment uncertainty is believed to be within 610 lm. Since we do not have a direct method of monitoring the gap size, we obtained a series of spectra at varied zav for both of the samples. Since the polariton peak for the ZnSe–Nujol conditions is broad, we decided to utilize the steep part of its envelope (1120–1150 cm1) for the frequency analysis (see Fig. 13). The spectral data were processed in the following way. First the original spectrum was 15-point smoothed, then the first derivative of the smoothed spectrum was taken. The position of the observed minimum was determined. These operations were performed using Omnic software provided with the spectrometer. In parallel, from the intensity of the mS band, we calculated the magnitudes of zav. Several measurements taken at different zav provided frequency-distance dependence. For each fiber sample, the frequency-distance data were collected three times. The obtained results are given in Fig. 14. For the

FIG. 12. Position of the polariton peak as a function of IRE–sample distance obtained for a quartz slide and three IREs. The dependences are obtained with Nujol and air interlayers.

APPLIED SPECTROSCOPY

631

ACKNOWLEDGMENTS The authors wish to thank Dennis Trevor and Jim Fleming (OFS Labs, Somerset, NJ) for providing us with the annealed fiber samples. We are also thankful to Jie Li (OFS, Avon, CT) for reviewing the manuscript.

FIG. 14. The peak frequency of the derivative curve as a function of IRE– sample distance. All spectra were obtained from fibers with acid-stripped coatings using a ZnSe IRE in combination with a Nujol interlayer. For each fiber sample (‘‘annealed’’ and ‘‘as drawn’’), three series of spectral measurements were taken.

‘‘as drawn’’ fiber, only the data obtained from the fiber whose coating was acid stripped are shown. The shift between the annealed and ‘‘as drawn’’ samples is clearly seen. As expected, the lower fictive temperature yields a higher band frequency in the spectrum. The obtained frequency-distance dependences have a ‘‘plateau’’ in the region 1.5–2.5 lm. For the ‘‘as drawn’’ fiber, averaging the values that fall within this range of zav, we obtained 1116.2 6 0.4 and 1116.0 6 0.3 cm1 for the cases of mechanical removal and acid removal of the fiber coating, respectively. It follows that the manner in which the coating was removed did not affect the polariton peak frequency. At the same time, for the annealed sample we obtained a value of 1120.2 6 0.4 cm1. The above numbers should be compared with the frequencies obtained from reflection spectra. Thus, we collected reflection spectra at 908 using the Nexus 670 microscope. For each sample, the spectra were collected four times. An example of such spectra is shown in Fig. 5. The obtained peak frequencies are 1119.89 6 0.13 and 1123.29 6 0.07 cm1 for the ‘‘as drawn’’ and annealed samples, respectively. The mAS frequency shift is thus 3.4 cm1. This shift, being caused by a ;450 8C change in fictive temperature, agrees well with the literature data.73–77 It follows that the polariton band shifts in the same direction by a similar magnitude as its counterpart in the reflection spectra. Unfortunately, the accuracy of our micro-ATR technique is much lower than that of the reflection one. Hence, using the reflection spectra still seems more appropriate for the fictive temperature measurements.

CONCLUSION A surface polariton mode is observed at ;1150 cm1 in micro-ATR spectra of amorphous bulk silica. Changes of the internal reflection element (IRE) material, variation of the gap size between the IRE and the sample, and also introducing different interlayers in the gap provide significant effects on the spectra. The polariton band is also observable when the bulk silica is coated by a 40 nm carbon layer. From the experimental data, the dielectric constant of silica is determined in the 1160– 1100 cm1 frequency range. The polariton band position was found to be sensitive to the fictive temperature of silica glass.

632

Volume 62, Number 6, 2008

1. A. Othonos and K. Kalli, Fiber Bragg Gratings: Fundamentals and Applications in Telecommunications and Sensing (Artech House, Norwood, MA, 1999). 2. R. P. Espindola, R. M. Atkins, D. A. Simoff, K. T. Nelson, and M. A. Paczkowski, in OFC’97 Technical Digest (Dallas, TX, 1997), paper PD-4. 3. R. W. Filas, in Materials Research Society Symposium Proceedings (MRS Spring Meeting, San Francisco, CA, 1998), vol. 531, p. 263. 4. A. Shinozaki, K. Arima, M. Morita, I. Kojima, and Y. Azuma, Anal. Sci. 19, 1557 (2003). 5. A. Rahimi, S. Gharazi, A. Ershad-Langroudi, and D. Ghasemi, J. Appl. Polym. Sci. 102, 5322 (2006). 6. B. M. Patterson, N. D. Danielson, and A. J. Sommer, Anal. Chem. 75, 1418 (2003). 7. K. L. A. Chan and S. G. Kazarian, Appl. Spectrosc. 57, 381 (2003). 8. T. T. Do, M. Celina, and P. M. Fredericks, Polym. Degrad. Stab. 77, 417 (2002). 9. A. J. Sommer, L. G. Tisinger, G. Marcott, and G. L. Story, Appl. Spectrosc. 55, 252 (2001). 10. N. Yilmaz, M. Mizukami, and K. Kurihara, Langmuir 23, 6070 (2007). 11. M. Mizukami, Y. Nakagawa, and K. Kurihara, Langmuir 21, 9402 (2005). 12. E. Puzenat and P. Pichat, J. Photochem. Photobiol. 160, 127 (2003). 13. W. M. Cross, S. Ma, R. M. Winter, and J. J. Kellar, Colloids Surf. A: Physicochem. Eng. Aspects 154, 115 (1999). 14. C. E. Giacomelli, M. G. E. G. Bremer, and W. Norde, J. Colloid Interface Sci. 220, 13 (1999). 15. T. Reihs, M. Muller, and K. Lunkwitz, Colloids Surf. A: Physicochem. Eng. Aspects 212, 79 (2003). 16. H. Gerung, C. J. Brinker, S. R. J. Brueck, and S. M. Han, J. Vac. Sci. Technol. A 23, 347 (2005). 17. D. B. Parry and J. M. Harris, Appl. Spectrosc. 42, 997 (1988). 18. N. Rochat, A. Chabli, F. Bertin, C. Vergnaud, P. Mur, S. Petitdidier, and P. Besson, Mater. Sci. Eng. B 102, 16 (2003). 19. D. Rouchon, N. Rochat, F. Gustavo, A. Chabli, O. Renault, and P. Besson, Surf. Interface Anal. 34, 445 (2002). 20. K. Ishikawa and M. Sekine, J. Appl. Phys. 91, 1661 (2002). 21. J. E. Olsen and F. Shimura, Appl. Phys. Lett. 53, 1934 (1988). 22. K. Namba, T. Komeda, and Y. Nishioka, Appl. Surf. Sci. 117/118, 198 (1997). 23. H. Tsuchida, I. Kamata, and K. Izumi, Appl. Surf. Sci. 117/118, 225 (1997). 24. S. Ekgasit and A. Padermshoke, Appl. Spectrosc. 55, 1352 (2001). 25. A. A. Stolov and D. A. Simoff, Appl. Spectrosc. 60, 29 (2006). 26. D. N. Mirlin, ‘‘Surface Phonon Polaritons in Dielectrics and Semiconductors,’’ in Surface Polaritons, V. A. Agranovich and D. L. Mills, Eds. (North-Holland, Amsterdam, 1982), Chap. 1, p. 3. 27. J. M. Pitarke, V. M. Silkin, E. V. Chulkov, and P. M. Echenique, Rep. Prog. Phys. 70, 1 (2007). 28. S. A. Ramakrishna, Rep. Prog. Phys. 68, 449 (2007). 29. N. Kuroda, Y. Iida, T. Shigeta, Hasanudin, and J. Watanabe, Jpn. J. Appl. Phys. 42, L1241 (2003). 30. W. A. Pliskin, J. Vac. Sci. Technol. 14, 1064 (1977), and references therein. 31. J. Wong, J. Appl. Phys. 44, 5629 (1973). 32. M. L. Naiman, C. T. Kirk, R. J. Aucoin, F. L. Terry, P. W. Wyatt, and S. D. Senturia, J. Electrochem. Soc.: Solid State Sci. Technol. 131, 637 (1984). 33. M. Nakamura, Y. Mochizuki, K. Usami, Y. Itah, and T. Nazaki, J. Electrochem. Soc.: Solid State Sci. Technol. 132, 482 (1985). 34. I. W. Boyd and J. I. B. Wilson, J. Appl. Phys. 53, 4166 (1982). 35. I. W. Boyd and J. I. B. Wilson, Appl. Phys. Lett. 50, 320 (1987). 36. I. W. Boyd, Appl. Phys. Lett. 51, 418 (1987). 37. G. Lucovsky, M. Manitini, J. K. Srivastava, and E. A. Irene, J. Vac. Sci. Technol. B 5, 530 (1987). 38. C. T. Kirk, Phys. Rev. B 38, 1255 (1988). 39. P. Lange, J. Appl. Phys. 66, 201 (1989). 40. W. Bensch and W. Bergholz, Semicond. Sci. Technol. 5, 421 (1990). 41. C. Martinet and R. A. B. Devine, J. Appl. Phys. 77, 4343 (1995). 42. A. M. Efimov and V. G. Pogareva, Chem. Geol. 229, 198 (2006). 43. K. B. Koller, W. A. Schmidt, and J. E. Butler, J. Appl. Phys. 64, 4704 (1988).

44. J. E. Olsen and F. Shimura, J. Appl. Phys. 66, 1353 (1989). 45. R. Almeida, Phys. Rev. B 45, 161 (1992). 46. K. Ishikawa, Y. Uchiyama, H. Ogawa, and S. Fujimura, Appl. Surf. Sci. 117/118, 212 (1997). 47. C. Z. Tan, J. Arndt, and H. S. Xie, Physica B 252, 28 (1998). 48. C. Z. Tan and J. Arndt, J. Chem. Phys. 112, 5970 (2000). 49. C. Z. Tan, J. Phys. Chem. Solids 63, 179 (2002). 50. T. B. Wang, Z. G. Liu, and C. Z. Tan, J. Chem. Phys. 119, 505 (2003). 51. B. C. Trasferetti and C. U. Davanzo, Appl. Spectrosc. 54, 502 (2000). 52. P. Innocenzi, J. Non-Cryst. Solids 316, 309 (2003). 53. T. Hirose, K. Saito, and A. Ikushima, J. Non-Cryst. Solids 352, 2198 (2006). 54. N. Nagai and H. Hashimoto, Appl. Surf. Sci. 172, 307 (2001). 55. N. Rochat, A. Chabli, F. Bertin, M. Olivier, C. Vergnaud, and P. Mur, J. Appl. Phys. 91, 5029 (2002). 56. F. L. Galeener and G. Lucovsky, Phys. Rev. Lett. 37, 1474 (1976). 57. F. L. Galeener, A. J. Leadbetter, and M. W. Stringfellow, Phys. Rev. B 27, 1052 (1983). 58. A. G. Kalampounias, S. N. Yannopoulos, and G. N. Papatheodoru, J. Chem. Phys. 124, 014504-1 (2006). 59. H. J. Falge and A. Otto, Phys. Stat. Sol. B 56, 523 (1973). 60. H. J. Falge, A. Otto, and W. Sohler, Phys. Stat. Sol. B 63, 259 (1974). 61. H. J. Falge, G. Borstel, and A. Otto, Phys. Stat. Sol. B 65, 123 (1974). 62. P. Grosse and V. Offermann, Vib. Spectrosc. 8, 121 (1995). 63. V. Offermann, P. Grosse, M. Feuerbacher, and G. Dittmar, Vib. Spectrosc. 8, 135 (1995).

64. S. Ekgasit, Appl. Spectrosc. 54, 756 (2000). 65. M. W. Urban, Attenuated Total Reflectance Spectroscopy of Polymers (American Chemical Society, Washington, D.C., 1996). 66. N. J. Harrick, Internal Reflection Spectroscopy (Harrick Scientific Corporation, Ossining, New York, 1987). 67. J. E. Dial, R. E. Gong, and J. N. Fordemwalt, J. Electrochem. Soc. 115, 326 (1968). 68. M. Tomozawa, J.-W. Hong, and S.-R. Ryu, J. Non-Cryst. Solids 351, 1054 (2005). 69. S.-R. Ryu and M. Tomozawa, J. Non-Cryst. Solids 352, 3929 (2006). 70. J. Homola, S. S. Yee, and G. Gauglitz, Sens. Actuators 54, 3 (1999). 71. E. A. Lindholm, J. Li, A. Hokansson, B. Slyman, and D. Burgess, in Proceedings of the National Fiber Optics Engineers Conference (Orlando, FL, 2003), p. 390. 72. P. G. Lemaire and E. A. Lindholm, ‘‘Hermetic Optical fibers: CarbonCoated Fibers,’’ in Specialty Optical Fibers Handbook, A. Mendez and T. F. Morse, Eds. (Elsevier, Amsterdam, 2007, p. 453. 73. A. Koike, S.-R. Ryu, and M. Tomozawa, J. Non-Cryst. Solids 351, 3797 (2005). 74. M. Tomozawa and R. Hepburn, J. Non-Cryst. Solids 345–346, 449 (2004). 75. D. L. Kim and M. Tomozawa, J. Non-Cryst. Solids 286, 132 (2001). 76. D. L. Kim, M. Tomozawa, S. Dubois, and G. Orcel, J. Lightwave Technol. 19, 1155 (2001). 77. A. Agarwal, K. M. Davis, and M. Tomozawa, J. Non-Cryst. Solids 185, 191 (1995).

APPLIED SPECTROSCOPY

633

Responses of Hydrophobic and Hydrophilic Groups in Nafion Differentiated By Least Squares Modeling of Infrared Spectra Recorded During Thin Film Hydration CAROL KORZENIEWSKI,* EVAN ADAMS, and DI LIU Department of Chemistry and Biochemistry, Texas Tech University, Lubbock, Texas 79409-1061

Least squares modeling was applied to gain insights into changes that occur in the structure of Nafion polymer membrane during hydration. Transmission infrared spectra followed changes in the strong polymer bands in the range of 1400–950 cm1 during water uptake by initially dry membrane upon exposure to 100% relative humidity atmosphere. Spectra recorded during hydration were fit to a rate equation that modeled the loss of a dry state accompanied by the development of a hydrated state. The evolution of the two states was described by an equation for diffusion in a cylindrical pore in the long time limit. Comparison of the experimental spectra in a data set to spectra calculated from the pure components derived by least squares modeling gave an excellent match for bands of the –CF2 and C–O–C group modes, but agreement was not as close for bands arising from modes of the hydrophilic –SO3 group and (modeled separately) water. The differences are discussed in terms of the likelihood that the –SO3 groups have stronger interactions with bulk-like water condensed in the membrane and therefore undergo more complex changes than do more hydrophobic polymer regions during hydration. A different model is necessary to describe the evolution of spectral features for water and –SO3 end groups during water uptake into Nafion thin films. Index Headings: Infrared spectroscopy; Fourier transform infrared spectroscopy; FT-IR spectroscopy; Polymers; Nafion; Least squares modeling.

INTRODUCTION Nafion membrane has a history of application as an ionically conductive separator for preparative scale electrochemical cells and fuel cell power sources.1,2 Nafion is a perfluorosulfonate ionomer that contains a poly(tetrafluoroethylene) backbone and perfluoroether side chains that terminate in a sulfonate group.1 The structure of local environments within Nafion membrane continues to be investigated to improve the understanding of factors that affect water transport and ionic conductivity.1,3–6 Nuclear magnetic resonance (NMR) techniques are powerful probes of polymer chain motions4–6 and solvent transport properties.7,8 Vibrational spectroscopy techniques provide insights into the environment surrounding chemical functional groups.3,9–19 Simple infrared spectroscopy measurements have revealed detailed information about side chain,9,11,13,16 backbone,16,17,19 and water3,10–12,14,17,18,20,21 structure in hydrated Nafion membrane. Recently, ultra-fast infrared vibrational lifetime measurements were performed on water in Nafion.3 The experiments showed evidence for multiple types of water clusters in low water content Nafion and a complex evolution of the clusters with changes in membrane hydration state.3 In this report, the results of applying least squares modeling to infrared spectral data sets recorded during hydration of a Received 4 January 2008; accepted 18 March 2008. * Author to whom correspondence should be sent. E-mail: carol. [email protected].

634

Volume 62, Number 6, 2008

thin, cast Nafion film are presented. Transmission sampling was used to follow changes during the transition between an initially dry state and a hydrated state in equilibrium with 100% relative humidity air. The region encompassing the strong polymer modes (1400–950 cm1) shows the well-documented perturbations in vibrations of the –CF2,16,19 –SO3,9,11,13,22 and C–O–C13 groups. Least squares modeling brings to light different responses for –SO3 groups relative to more hydrophobic polymer regions. The results are discussed in terms of models for water in Nafion membrane.

EXPERIMENTAL Infrared spectra were recorded on a Mattson Instruments RS/1 Fourier transform infrared (FT-IR) spectrometer system operating with a liquid nitrogen cooled, narrow band mercury cadmium telluride (MCT) detector. Spectra were computed from the average of 128 interferograms measured at 2 cm1 resolution and apodized with a triangular function. The sample cell was constructed by modification of a standard infrared transmission cell (Demountable Liquid Cell, Aldrich)19 and is based on a previously reported design.22 ZnSe and CaF2 windows (32 mm diameter by 3 mm thick) were employed. Windows were held parallel in the cell and separated a distance of 6.5 mm by a Kel-F spacer. The spacer included a port for entry of a 23 gauge syringe needle and a shallow well beneath the port to hold a water droplet. The windows were polished with 0.05 lm alumina and then rinsed in deionized water (Barnstead Nanopure, 4-cartridge Infinity System) followed by acetone and then brief sonication in deionized water. In a final step, the windows were further rinsed in deionized water and then set in an oven at 110 8C to dry. Thin Nafion films were prepared by casting ;80 lL of Nafion solution (10 wt. % in water, Aldrich) across the surface of a ZnSe optical window. The window was placed inside an oven at 110 8C for approximately 30 min to remove solvent from the film. The film was then ion exchanged in 0.5 M NaCl for one hour,16 rinsed in deionized water, and returned to the oven to evaporate surface water. The window containing the film was then dried overnight inside a vacuum oven at 110 8C and a few torr of pressure. The thickness of the Nafion films produced was approximately 1 lm.19 Just prior to infrared experiments, the ZnSe window containing the Nafion film was removed from the vacuum oven, placed in a desiccator, and allowed to cool to room temperature before assembling it into the infrared cell. The cell contained an 8 mm diameter circular aperture that ensured the area of polymer sampled completely filled the infrared beam. A CaF2 window was used on the side of the cell opposite the ZnSe window to reduce the appearance of interference fringes in spectra. Background spectra were

0003-7028/08/6206-0634$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

recorded with two freshly cleaned and polished windows (one CaF2 and one ZnSe) in the cell. Least squares modeling was carried out according to the algorithm described by Rivera and Harris.23,24 The program was developed using Visual Fortran (Digital Equipment Corporation) within the Microsoft Developer Studio (Redmond, WA) programming environment on a PC computer running under the Windows XP operating system.

DATA MODELING Transmission infrared spectra recorded at fixed times during membrane hydration formed a data set. The absorbance values for each spectrum were written as a column vector, and the vectors from the data set were assembled into a data matrix (D). The data matrix was decomposed into the product of a matrix of pure component spectra (A) and a matrix of coefficients (C) that define the relative contribution of each component. The relationship is given by the following equation: D ¼ AC. The columns of A contain the component spectra and each row of C represents a state of the membrane (‘‘dry’’ or ‘‘hydrated’’ in the model considered). The j columns of D index to the membrane hydration time, t( j), and the elements of C (c(i, j)) give the fraction of membrane in state i at time t( j). The elements in C were determined by adapting the equation for diffusion in a cylindrical pore in the long time limit:24–27 (  ) ‘ X N 8 D 2 2 ¼ 1 exp  2 ð2n þ 1Þ p t ð1Þ 2 2 Nmax L n¼0 ð2n þ 1Þ p In Eq. 1, D is the diffusion coefficient for analyte being transported into the membrane, L is the pore length (usually taken as the membrane thickness),25–27 N is the number of species in the film undergoing transport at time t, and Nmax is the maximum N attained after long measurement periods. Equation 1 has been applied to characterize mass transport of water and ions in Nafion membrane25–27 and has been shown to predict changes in spectral absorbance for water in the O–H stretching region near 3525 cm1 during Nafion membrane hydration.28 In the present study, adapting Eq. 1 assumed N is directly (linearly) related to changes in the spectral absorbance of polymer functional groups affected by water uptake. The assumption is considered further in the Discussion section below. To simplify Eq. 1, a parameter s was defined as equal to the factor L2/D. Applying the model, the c(i, j) values that describe the growth of the hydrated state were given by N/Nmax in Eq. 1 calculated at time t( j) with s ¼ sh, where sh is a constant. Similarly, the c(i, j) values that describe the loss of the dry polymer state were calculated as (1  N/Nmax) for time t( j) and s ¼ sd, where sd is a constant. For accuracy at short times, the sum in Eq. 1 was carried out to n ¼ 50. To begin the least squares analysis, an initial C matrix was calculated from an estimate of sh and sd. D was then decomposed to give a matrix of estimated component spectra ˆ as follows: (A) ˆ DCT ½CCT 1 ¼ A

ð2Þ

The parameters sh and sd in the model were optimized by using a simplex algorithm29 to minimize the square of the residuals, ˆ To verify the quality and reasonableness R, where R ¼ D  AC. ˆ was calculated by least of the fit to the kinetic model, a matrix C squares projection of the pure component spectra onto the data

FIG. 1. Simulated spectral bands used to test least squares modeling with Eq. 1. (A) The solid lines show the Gaussian peaks used to calculate the spectra given by the solid lines in (B). The Gaussian function parameters employed for each peak are listed in Table I. The simulated spectra in (B) were determined by taking the sum of the two Gaussian functions in (A) and varying the intensities according to Eq. 1 (see text for details). The plot in (C) shows the variation in (1  N/Nmax) (dashed line, labeled 2045 cm1) and N/Nmax (solid line, labeled 2030 cm1) as a function of time (in arbitrary units). The values for the functions at times 0, 2, 4, 6, and 8 multiplied the Gaussian peak I0 values in the spectra labeled, respectively, a, b, c, d, and e in (B). The traces formed by the ˆ that symbols in (A) are the optimized pure component spectra comprising A result from least squares modeling of the simulated spectra in (B). The symbols ˆ product of the optimized in (B) give the spectra determined from the AC matrices.

ˆ were ˆ 1A ˆ TD, and the elements in C ˆ ¼ [A ˆ TA] according to C compared graphically with the elements in the optimized C matrix. It is useful to mention that the modeling approach being applied is a numerical method that identifies spectral regions within a data set that change in accord with a mathematical function (Eq. 1, in this case) over the course of an experiment. The approach provides insights into the environment surrounding chemical functional groups within a molecular system under study but does not make assumptions about, or require knowledge of, the molecular chemical properties or the molecular geometry of the system.

RESULTS Test Calculations. As a test of the two-state model, least squares analysis was applied to a set of simulated spectra (Fig. 1). Spectra were calculated from Gaussian band functions, and the intensities were varied in accord with the two-state model. Since the goal of the test calculations was to assess the stability of the algorithm in performing an optimization to the two-state model, noise was not included in the simulated data. In Fig. 1A, the Gaussian functions plotted as solid lines were summed to give the resultant spectra in Fig. 1B (solid lines). The intensity of the band at 2045 cm1 in Fig. 1B decreases as it progresses from spectrum a to spectrum e. The response is intended to represent the behavior of a band associated with a

APPLIED SPECTROSCOPY

635

TABLE I. Gaussian band parameters.a m¯ 0

I0

r

2030 2045

0.50 0.25

5 5

a

Parameters used to plot the calculated Gaussian bands in Fig. 1 according to the function I(¯m) ¼ I0 exp[(¯m  m¯ 0)2/2r2] where the center frequency (¯m0) has cm1 units, the band width parameter (r) has cm1 units, and the intensity (I0) has absorbance units (a.u.).

‘‘dry’’ membrane state during water uptake. The Gaussian peak intensity parameter (I0, Table I) for the 2045 cm1 band was attenuated by (1  N/Nmax), where N/Nmax is given by Eq. 1. The factor L2/D ¼ s ¼ sd in Eq. 1 was set equal to 40. Similarly, changes in the intensity of the 2030 cm1 band in Fig. 1B were modeled using Eq. 1. In progressing from spectrum a to spectrum e in Fig. 1B, growth at 2030 cm1 tracks the response expected for a band associated with a ‘‘hydrated’’ polymer state during water uptake. To simulate the behavior, I0 in Table I was multiplied by N/Nmax with L2/D ¼ s ¼ sh ¼ 40. Figure 1C displays the time behavior of the factors N/Nmax and (1  N/ Nmax) that multiplied the I0 parameters for the two Gaussian peaks. The spectra labeled a, b, c, d, and e in Fig. 1B correspond, respectively, to times 0, 2, 4, 6, and 8 in Fig. 1C. The intensity values for the simulated spectra in Fig. 1B were assembled into a D matrix. Least squares modeling of the data and optimization of sh and sd from a wide range of initial values returned sh ¼ sd ¼ 40 and the pure component spectra shown by the symbols in Fig. 1A. The spectra that result by ˆ the matrix containing the pure taking the product of A, components, and C, calculated using the optimized sh and sd values, are shown as symbols in Fig. 1B. The close agreement between the simulated data and the least squares modeling result demonstrates that the algorithm being implemented is stable, and in performing operations on the simulated data, there is good precision in the computational steps. Application to Experimental Data. Figure 2 displays a set of transmission infrared spectra that were recorded during hydration of a thin Nafion film following exchange by Naþ. The initial spectrum was recorded just prior to the addition of a water droplet to the cell. Arrows next to the major bands indicate the direction of intensity change with water uptake. The strongest features that appear (1235 cm1 and 1159 cm1) are primarily attributed to C–F stretching modes of –CF2 groups.11,15–17,30 In the 1350–1200 cm1 region, features from the asymmetric S–O stretching modes of the –SO3 groups are also believed to be present.11,15,16,19,30 With water uptake, the 1159 cm1 band grows, but the position remains fixed, while the 1235 cm1 band undergoes more complex changes.16,19 The sensitivity to water uptake of bands associated with hydrophobic fluorocarbon groups in Nafion has been observed and discussed in terms of possible contributions from –CF moieties in the side chains extending into water-rich membrane regions.16,19 Hydration of the hydrophilic –SO3 groups also is likely important, particularly in affecting spectral changes near 1235 cm1 and 1300 cm1.11,15,16,19,30 The band near 1064 cm1 arises from the symmetric S–O stretching mode of –SO3 groups.9,11,13,22 For dry, Naþ exchanged Nafion membrane, the band appears in the vicinity of 1064 cm1 and with increasing hydration shifts down in energy and becomes narrower.9,13 The features near 983 cm1 in Fig. 2 have been assigned to stretching modes of the side-chain C–O–C groups.11,13,15,16

636

Volume 62, Number 6, 2008

FIG. 2. Experimental infrared spectra of a thin Nafion film cast onto a ZnSe optical window recorded during hydration in 100% relative humidity air starting from a dry state. Spectra were recorded at times of 0 min, 0.75 min, 1.5 min, 2.25 min, 6.75 min, and 28 min after exposure to humid air. The arrows indicate the direction of band intensity change with increasing hydration. The inset shows the growth in the relative intensity in the band at 1159 cm1. The points in the plot are from experiment and the solid line is the fit to Eq. 1, assuming a linear (Beer’s law) relationship between N and the absorbance changes at 1159 cm1.

The 983 cm1 band is insensitive to water uptake,13 but the lower energy feature at ;972 cm1 is strikingly sensitive.13 The ;972 cm1 band has been shown to be associated with vibrations of the ether group that is closest to the –SO3 end group at the side-chain terminus.13 The electrostatic and hydrophilic interactions that affect the –SO3 groups during membrane hydration are thought to contribute to the spectral changes at ;972 cm1.13 The inset in Fig. 2 shows the changes that occur in the intensity of the 1159 cm1 band with hydration time. The solid line in the plot demonstrates that the response is consistent with Eq. 1. Since N in Eq. 1 is expected to correspond to the number of water molecules entering the film during hydration, adherence of the 1159 cm1 polymer band intensity to the equation suggests that small, possibly micrometer to nanometer scale, regions of polymer in the film transition from a dry to a hydrated state in proportion to the water absorbed. The behavior is discussed further below (see the Discussion section). The consistency of band growth at 1159 cm1 with Eq. 1 led to the use of Eq. 1 in analyzing the full data range in Fig. 2 through the two-state model. The absorbance values for the spectra in Fig. 2 were assembled into a D matrix. Initial values for sh and sd were estimated from the fit in the Fig. 2 inset. The optimized pure component spectra that resulted from the analysis are graphed in Fig. 3. The pure component spectrum for the dry state is a close match to the initial experimental spectrum, which was recorded immediately prior to exposure of the dry film to water vapor saturated air. In a similar manner, the pure component spectrum associated with the hydrated state maps onto the final experimental spectrum recorded after a long period of film hydration. Figure 4 displays the spectra ˆ and C calculated by taking the product of the optimized A matrices (solid lines). Data points from the experimental spectra are included for comparison. The linear combinations of the two pure components reproduce the experimental results

FIG. 3. (Symbols) Data from the initial (0 min) and final (28 min) experimental spectra in Fig. 2. (Solid lines) Optimized pure component spectra ˆ The simulation returned optimized values for sh ¼ sd ¼ 39.0 min. in A.

well. At the level of detail shown, the two-state model appears to give an excellent prediction of the experimental spectra. In Figs. 5 and 6, narrow spectral regions from Fig. 4 are shown expanded and experimental spectra are displayed with higher resolution. The regions dominated by features of hydrophobic –CF2 groups (Fig. 5) and the C–O–C groups (Fig. 6) are in good agreement with spectra calculated from the ˆ and C matrices derived using the two-state model. optimized A However, deviations are evident in the region dominated by the –SO3 group mode near 1064 cm1, where the match between experiment and the modeling result is not as close as in the other modes examined.

DISCUSSION Modeling Function. In studies of Nafion 112 (;50 lm thickness) membrane, infrared spectral absorbance changes

FIG. 4. (Symbols) Data from the experimental spectra in Fig. 2. A point is ˆ product plotted every 4 cm1. (Solid lines) Spectra calculated from the AC following optimization.

FIG. 5. (Symbols) Data from the experimental spectra in Fig. 2 with focus on the spectral region encompassing the –CF2 stretching modes. A point is plotted ˆ product following every 1 cm1. (Solid lines) Spectra calculated from the AC optimization.

across the water O–H stretching region were observed to follow Eq. 1 during water uptake in a low humidity atmosphere (;10% relative humidity).28 The two-state model employed in the current work was developed by building on these findings. Agreement between experimental time-dependent infrared absorbance data and an equation describing diffusion in a cylindrical pore is consistent with proposed structures of Nafion as consisting of a network of hydrophilic pores and channels.1 In extending Eq. 1 to model the infrared absorbance data in Fig. 2, it was assumed that a linear relationship exists between the polymer band spectral absorbance changes and membrane water content. The good correspondence between the model and the experimental results suggests that the assumption is valid under the measurement conditions employed. The spectral absorbance changes observed (Fig. 2) are small relative to the peak intensities, and it is possible that the weak perturbations enable a linear response. It is notable that the peak intensities in Fig. 2 are in a range typically optimal for

FIG. 6. (Symbols) Data from the experimental spectra in Fig. 2 with focus on the spectral region encompassing modes of the side-chain C–O–C and –SO3 groups. A point is plotted every 1 cm1. (Solid lines) Spectra calculated from ˆ product following optimization. the AC

APPLIED SPECTROSCOPY

637

Beer’s law measurements (;1 a.u. or less). Non-Beer’s law type responses have been reported for infrared bands of water in Nafion, but under conditions involving larger absorbance changes on more strongly absorbing samples.21 In principle, a diffusion coefficient for water in Nafion can be estimated from the optimized s values by assuming L in Eq. 1 is the membrane thickness. However, as discussed earlier,28 restrictions in the channel network throughout partially hydrated Nafion membrane limit the mean free path for water. Hence, the transport of water across initially dry Nafion membrane in humid air is much slower than expected based on the self-diffusion coefficients for water in fully hydrated Nafion (0.5 3 106–5 3 106 cm2/s).7,26,31,32 An analysis of water transport based on s determined from the experimental data in Fig. 2 will not be carried out until more is known about the effective value of L, which is likely on the nanoscale for Nafion in low hydration states. Hydration Effects on Functional Groups in Nafion. The results in Figs. 5 and 6 indicate that regions of the Nafion polymer associated with the main stretching vibrational modes of the fluorocarbon and ether groups undergo change during hydration consistent with the mass transport function given by Eq. 1, but the symmetric S–O stretching mode of the –SO3 group near 1064 cm1 has a more complex behavior. The differences likely can be traced, at least in part, to interactions of the polymer functional groups with interfacial versus bulk-like water in Nafion. For the –C–F and C–O–C groups, there is evidence that these moieties contact water molecules at the surface of condensed droplets inside membrane pores and channels.10–12 The water–polymer interfacial interactions are detected in infrared spectra as sharp bands toward the high energy side of the broad feature encompassing the O–H stretching modes for hydrogen bonded water.10–12,18,21,28 A sharp feature is often observed near 3670 cm1 due to water molecules that extend into fluorocarbon rich regions in Nafion.10–12,18,21,28 In contrast to the more hydrophobic –C–F and C–O–C groups, the charge-bearing –SO3 groups are stabilized by cations and ion-dipole forces exerted on assemblies of water molecules in bulk-like water clusters and solvation shells. During hydration of Nafion from a dry state, the environment surrounding –SO3 groups can progress through a series of anion-cation-water structures9 that have been shown to lead to complicated spectral responses for –SO3 group vibrational modes.9,11,13,22 The electrostatic and polarization forces involved in solvation of the –SO3 group likely produce changes in the 1064 cm1 band (Fig. 6) that are not easily predicted by Eq. 1. The weaker forces exerted by interfacial water on the fluorocarbon and ether groups produce smaller perturbations and responses that track water transport into the polymer. It is important to point out that the asymmetric S–O stretching modes of the –SO3 group are believed to contribute to responses in the 1350–1200 cm1 range, in addition to the –C–F vibrations. However, the region, including the features near 1305 cm1 (results not shown in detail), is adequately fit by Eq. 1 and the two-state model (Fig. 5). It may be that bands from modes of the fluorocarbon groups dominate the region (c.f., Ref. 17) and make the small deviations from the model that occur for features of the sulfonate groups difficult to detect. Better knowledge of Nafion vibrational band assignments is needed to understand the pronounced changes that occur near 1350–1200 cm1 during membrane hydration.16,19

638

Volume 62, Number 6, 2008

It was mentioned in the Results section that adherence of band growth at 1159 cm1 to Eq. 1 (Fig. 2, inset) suggests that water uptake causes small polymer domains throughout the film to transform from a dry to a hydrated state in proportion to the amount of water incorporated. The picture is consistent with general models for water transport and membrane swelling during Nafion hydration.1 There is a tendency for water to accumulate in –SO3 group rich, nanometer-scale pores and channels that occur randomly throughout the membrane.1 In the present experiments, water molecules enter membrane regions closest to the outer surface at random and reach pores in the film (;1 lm thick) interior by diffusion. The concentration gradient that develops across the film during hydration is not expected to appreciably affect linearity in the absorbance, as the infrared wavelengths are considerably greater than the film thickness. Blanchard and Nuzzo16 pointed out that sensitivity of the band near 1159 cm1 toward water uptake is unexpected, since the band is typically associated with the symmetric –CF2 stretching mode in Nafion.11,15,16,30 However, these workers noted that spectral changes in bands associated with –C–F vibrational modes may reflect the substantial rearrangements that take place in the side chains.16 It is possible that the changes at 1159 cm1 arise mainly from perturbations to side-chain –C–F moieties in response to water accumulation near –SO3 groups. Application of the Model to O–H Stretching Modes of Water. In a separate investigation, the two-state model was applied to the region between 3100 cm1 and 3800 cm1 where the O–H stretching modes of water condensed inside Nafion pores and channels appear.10,12,16–18,21,28 The region can contain multiple features that reflect the presence of both bulk-like and interfacial water in the membrane.10,12,18,21,28 In an earlier study of Nafion 112 hydration, the peak and integrated intensities of the band encompassing modes of hydrogen-bonded, bulk-like water (3525 cm1) and the feature for interfacial water in contact with fluorocarbon-rich regions (;3670 cm1) were shown to increase in accord with Eq. 1 during water uptake.28 Since the two bands overlap and the region includes an additional interfacial water band (;3712 cm1), it was of interest to use least squares modeling to attempt to resolve the peaks and probe the time-dependent behavior of the different environments. Applying the two-state model to the data set for Nafion 112 returned one pure component dominated by interfacial water bands and a second pure component that contained both bulk-like and interfacial water contributions. However, when the experimental spectra ˆ and C were compared to the product of the optimized A  matrices, agreement was similar to that for the –SO3 group in Fig. 6. Subsequently, the water O–H stretching region in the set of spectra in Fig. 2 was analyzed using the two-state model and the same results were obtained. It appears that a different model is required to explain the spectral response for water bands monitored during Nafion membrane hydration.

CONCLUSION In summary, least squares modeling of infrared spectral data recorded during Nafion thin film hydration is useful for probing the effects of water on functional groups in the ionomer. Except for the prominent band of the symmetric S–O stretching mode of –SO3 groups near 1064 cm1, all other bands in the 1400–950 cm1 region fit a model for the transformation of polymer from a dry to a hydrated state with a time dependence consistent with the equation for diffusion in a cylindrical pore

in the long time limit. The results suggest that polymer regions that experience mainly interfacial interactions with water (–CF and C–O–C rich regions) undergo weak spectral intensity changes that scale with film water content. In contrast, the solvent clusters that form in the vicinity of the polar, hydrophilic –SO3 groups cause stronger spectral perturbations that require more detailed models to predict. Analyzing the water O–H stretching region gave results similar to those obtained for the 1064 cm1 band of the –SO3 group. In future work, new models will be tested, and steps will be taken to analyze polymer and water mid-infrared spectral regions simultaneously. Experiments will be carried out on thin films prepared by casting methods that will improve durability33–35 and have potential to enable thin films to be treated by methods analogous to those used to clean and ion exchange practical fuel cell membrane.18,32 The least squares modeling approach is general and can be applied in the study of these various types of Nafion membrane materials. Finally, adherence to Beer’s law in infrared measurements of water uptake by Nafion needs to be investigated further.11,21 We plan to extend demonstrated approaches based on gravimetric methods21 to assess linearity in infrared spectroscopic studies of water uptake by Nafion 112 and thin, cast Nafion films. Experiments of this type are needed to enable quantitative determinations of membrane water content from infrared spectral data. ACKNOWLEDGMENTS We thank Prof. Dion Rivera and Prof. Joel M. Harris for helpful discussions concerning least squares modeling of vibrational spectra. We are grateful to the U.S. Department of Energy for support under Award No. DE-FG0205ER46234. 1. K. A. Mauritz and R. B. Moore, Chem. Rev. 104, 4535 (2004). 2. S. Gottesfeld and T. A. Zawodzinski, ‘‘Polymer Electrolyte Fuel Cells’’, in Advances in Electrochemical Science and Engineering, R. C. Alkire, H. Gerischer, D. M. Kolb, and C. W. Tobias, Eds. (Wiley-VCH, New York, 1997), vol. 5, p. 195. 3. D. E. Moilanen, I. R. Piletic, and M. D. Fayer, J. Phys. Chem. A 110, 9084 (2006). 4. K. A. Page, K. M. Cable, and R. B. Moore, Macromolecules 38, 6472 (2005). 5. Q. Chen and K. Schmidt-Rohr, Macromol. Chem. Phys. 208, 2189 (2007). 6. K. A. Page, W. Jarrett, and R. B. Moore, J. Polym. Sci.: Part B: Polym. Phys. 45, 2177 (2007). 7. T. A. Zawodzinski, Jr., M. Neeman, L. O. Sillerud, and S. Gottesfeld, J. Phys. Chem. 95, 6040 (1991).

8. H. A. Every, M. A. Hickner, J. E. McGrath, and T. A. Zawodzinski, Jr., J. Membrane Sci. 250, 183 (2005). 9. S. R. Lowry and K. A. Mauritz, J. Am. Chem. Soc. 102, 4665 (1980). 10. M. Falk, Can. J. Chem. 58, 1495 (1980). 11. M. Falk, ‘‘Infrared Spectra of Perfluorosulfonated Polymer and of Water in Perfluorosulfonated Polymer’’, in Perfluorinated Ionomer Membranes, A. Eisenberg and H. L. Yeager, Eds. (American Chemical Society, Washington, D.C., 1982), vol. 180, p. 139. 12. S. Quezado, J. C. T. Kwak, and M. Falk, Can. J. Chem. 62, 958 (1984). 13. K. M. Cable, K. A. Mauritz, and R. B. Moore, J. Polym. Sci.: Part B: Polym. Phys. 33, 1065 (1995). 14. R. Buzzoni, S. Bordiga, G. Ricchiardi, G. Spoto, and A. Zecchina, J. Phys. Chem. 99, 11937 (1995). 15. M. Laporta, M. Pegoraro, and L. Zanderighi, Phys. Chem. Chem. Phys. 1, 4619 (1999). 16. R. M. Blanchard and R. G. Nuzzo, J. Polym. Sci.: Part B: Polym. Phys. 38, 1512 (2000). 17. A. Gruger, A. Regis, T. Schmatko, and P. Colomban, Vib. Spectrosc. 26, 215 (2001). 18. R. Basnayake, G. R. Peterson, D. J. Casadonte, Jr., and C. Korzeniewski, J. Phys. Chem. B 110, 23938 (2006). 19. C. Korzeniewski, D. Snow, and R. Basnayake, Appl. Spectrosc. 60, 599 (2006). 20. R. Iwamoto, K. Oguro, M. Sato, and Y. Iseki, J. Phys. Chem. B 106, 6973 (2002). 21. Y. Wang, Y. Kawano, S. R. Aubuchon, and R. A. Palmer, Macromolecules 36, 1138 (2003). 22. M. Ludvigsson, J. Lindgren, and J. Tegenfeldt, Electrochim. Acta 45, 2267 (2000). 23. D. Rivera, P. E. Peterson, R. H. Uibel, and J. M. Harris, Anal. Chem. 72, 1543 (2000). 24. D. Rivera and J. M. Harris, Anal. Chem. 73, 411 (2001). 25. A. Goswami, A. Acharya, and A. K. Pandey, J. Phys. Chem. B 105, 9196 (2001). 26. G. Suresh, Y. M. Scindia, A. K. Pandey, and A. Goswami, J. Membrane Sci. 250, 39 (2005). 27. G. Suresh, Y. M. Scindia, A. K. Pandey, and A. Goswami, J. Phys. Chem. B 108, 4104 (2004). 28. R. Basnayake, W. Wever, and C. Korzeniewski, Electrochim. Acta 53, 1259 (2007). 29. W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling, Numerical Recipes (Cambridge University Press, Cambridge, U.K., 1989). 30. D. Malevich, V. Zamlynny, S.-G. Sun, and J. Lipkowski, Z. Phys. Chem. 217, 513 (2003). 31. A. Roy, M. A. Hickner, X. Yu, Y. Li, T. E. Glass, and J. E. McGrath, J. Polym. Sci.: Part B: Polym. Phys. 44, 2226 (2006). 32. T. A. Zawodzinski, Jr., C. Derouin, S. Radzinski, R. J. Sherman, V. T. Smith, T. E. Springer, and S. Gottesfeld, J. Electrochem. Soc. 140, 1041 (1993). 33. R. B. Moore III and C. A. Martin, Anal. Chem. 58, 2569 (1986). 34. R. B. Moore III and C. A. Martin, Macromolecules 21, 1334 (1988). 35. R. B. Moore III and C. A. Martin, Macromolecules 22, 3594 (1989).

APPLIED SPECTROSCOPY

639

Rapid Nondestructive On-Site Screening of Methylamphetamine Seizures by Attenuated Total Reflection Fourier Transform Infrared Spectroscopy CHING YONG GOH, WILHELM

VAN

BRONSWIJK,* and COLIN PRIDDIS

Department of Applied Chemistry, Curtin University of Technology, GPO Box U1987, Perth, WA 6845, Australia (C.Y.G., W.v.B.); and Forensic Science Laboratory, Chemistry Centre (Western Australia), 125 Hay Street, East Perth, WA 6004, Australia (C.P.)

The identification and quantification of illicit substances in the field is often desirable. Fourier transform infrared spectroscopy (FT-IR) has both qualitative and quantitative capabilities and field portable instruments are commercially available. Transmission infrared spectra of mixtures containing ephedrine hydrochloride, glucose, and caffeine and attenuated total reflection (ATR) infrared spectra of mixtures composed of methylamphetamine hydrochloride, glucose, and caffeine were used to develop principal component regression (PCR) calibration models. The root mean sum of errors of predictions (RMSEP) of all individual components in a mixture from a single measurement was ,6% w/w, which reduced to ;3% w/w when triplicates were averaged. Sample mixing and grinding are essential to minimize the effect of heterogeneity, as deviations of up to 20% w/w were observed for single measurements of unground samples. Poor predictions of the components in a mixture occurred when samples were ‘‘contaminated’’ with substances not present in the calibration set, as would be expected. When only a single analyte (drug) was targeted, using a calibration set that contained both contaminated and uncontaminated samples, an RMSEP of ;4% w/w was achieved. The results demonstrate that ATR-FT-IR has the potential to quantify methylamphetamine samples, and possibly other licit or illicit substances, in at-seizure and on-site scenarios. Index Headings: Methylamphetamine; Infrared spectroscopy; Fourier transform infrared spectroscopy; FT-IR spectroscopy; Attenuated total reflection; ATR; Raman spectroscopy; Quantitative analysis.

INTRODUCTION Gas chromatography–mass spectrometry (GC-MS) is the primary method used by forensic laboratories for the definitive identification of illicit drugs. Identification and quantification of illicit drugs in the laboratory can also be performed by highperformance liquid chromatography (HPLC), gas chromatography (GC), capillary electrophoresis (CE), and nuclear magnetic resonance spectroscopy (NMR).1–3 These methods all suffer from one or more of the following shortcomings for nondestructive and rapid on-site quantification of controlled substances: There is no confirmatory information about the structure of the compound; it does not provide quantitative information; the sample is destroyed; extensive sample preparation is required; and the instrument is not readily deployable in the field. Vibrational spectroscopy, both infrared (IR) and Raman, provides structural information by way of characteristic vibrational frequencies that are a molecular fingerprint of the sample being analyzed. While the commonly used pressed KBr disk technique for IR spectroscopy is destructive, nondestructive methods such as attenuated total reflection (ATR) can be Received 7 June 2007; accepted 19 March 2008. * Author to whom correspondence should be sent. E-mail: [email protected].

640

Volume 62, Number 6, 2008

implemented on both laboratory and field-portable instruments (e.g., HazMatIDTM). Raman spectroscopy is by its nature nondestructive, provided the sample is not destroyed by the excitation laser used, and field-portable instruments are also available (e.g., FirstDefenderTM). One major issue that needs to be addressed with both techniques is that of sample heterogeneity. As illicit drug samples are inevitably mixtures of individual solid materials, it is crucial that the sample area/ volume analyzed is representative of the bulk. This is relatively simple with ATR Fourier transform infrared (FT-IR) spectroscopy as its sampling area can be large (.10 mm2, Fig. 1). In a typical ATR configuration (Fig. 1), the infrared beam enters a high refractive index ATR crystal and is internally reflected.4 This results in an evanescent wave, which penetrates the sample in contact with the ATR crystal and produces a reflection/absorption spectrum. Examples of ATR crystal materials include zinc selenide, KRS-5 (thallium iodide/ thallium bromide), germanium, and diamond. Diamond is usually preferred when unknown samples are to be assessed, e.g., in forensic work, because of its hardness (scratch resistance) and chemical inertness. Infrared spectroscopy of solid mixtures has been predominantly aimed at qualitative analysis. It has been applied to the identification of major diluents in illicit tablets5 and diluents and adulterants in seized cocaine6 and heroin7,8 samples. ATR based field-portable instruments (HazMatIDTM) have been used by the Australian Federal Police and Chemistry Centre (Western Australia) at clandestine laboratories to identify (but not quantify) various illicit drugs, precursor materials, solvents, and diluents.  Micro-FT-IR spectroscopy has also been used to confirm the results of microcrystal tests by acquiring spectra of the crystals formed, especially in cases where crystal forms show similar appearance.9 Raman scattering is inherently a weak effect that is easily overwhelmed by fluorescence, but it has the advantage of requiring no sample preparation and water giving only weak bands because of its low scattering cross-section. The fluorescence problem can sometimes be overcome by using longer wavelength lasers (typically 785, 830, or 1064 nm) albeit with a reduction in Raman intensity. However, the spectrometers have a comparatively small sampling area (often ,1 mm2) and hence sample heterogeneity can be a greater problem than with IR. Raman spectral intensities and band widths are also more sensitive to crystallinity than those of IR spectra, which can lead to greater variance in their spectra. The advantage of the small Raman scattering cross-section of water, as opposed to its high IR absorptivity, is unlikely to be relevant   M. Collins, 2006, personal communication; C. Priddis, 2006, personal communication.

0003-7028/08/6206-0640$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

quantification of seized methylamphetamine samples using model compounds and methylamphetamine, with laboratory and field-portable instrumentation.

EXPERIMENTAL

FIG. 1. General setup for ATR-FT-IR spectroscopy.

for methylamphetamine samples as they are almost inevitably dry powders or tablets. Raman spectroscopy of solid mixtures has similarly been aimed at identifying individual components and its use in identifying controlled substances such as heroin, cocaine, and MDMA (3,4-methylenedioxymethylamphetamine) is well established.10–12 The problem of sample heterogeneity for quantitative work is sometimes overcome by dissolving the sample in a suitable solvent and obtaining the solution’s spectrum.13 This overcomes the heterogeneity problem but the method requires sample preparation and is destructive, which is undesirable for field work. However, Ryder et al.10 have demonstrated that quantitative solid-state Raman analyses are feasible with their quantification of cocaine diluted with glucose via partial least squares regression (PLSR). While IR and Raman spectra are rich in information, the spectra of mixtures can be difficult to interpret simply because of the wealth of information and the overlap of spectral features. Qualitative analysis is sometimes possible directly from the spectra, but interpretation of such spectra is greatly aided by multivariate techniques such as principal components analysis (PCA). Extending PCA to principal components regression (PCR) allows the spectra to be used for quantification of components in mixtures, e.g., PCA has been used by Ryder14 to distinguish between diluted samples of cocaine, heroin, and MDMA, and by Strachan et al.15 as a data reduction method in conjunction with multiple linear regression (MLR) to quantify the polymorphic forms (I and III) of carbamazepine in a mixture. More recently, Katainen et al.13 demonstrated the quantification of seized amphetamine samples using solution Raman spectra relative peak heights and partial least squares regression (PLSR). To address the issues associated with on-site analyses, sample preservation, and sample heterogeneity, this study has investigated the use of IR and Raman spectroscopy for the TABLE I.

Materials. The source and purpose of materials used in this project are summarized in Table I. Methylamphetamine hydrochloride samples (Table II) were provided by the Illicit Drugs Section of the Chemistry Centre (Western Australia). They had been synthesized by the three most commonly used reduction methods at clandestine laboratories in Western Australia: red phosphorus/iodine, hypophosphorus acid/iodine, and metal/liquid ammonia. Infrared and Raman Spectroscopy. Transmission infrared spectra of the calibration and validation mixtures were acquired using a Bruker IFS66 FT-IR spectrophotometer fitted with a deuterated triglycine sulfate (DTGS) detector. Measurements were recorded at a resolution of 4 cm1 from 4000–400 cm1 with 16 scans accumulated. Spectral measurements, manipulations, and evaluations were performed by OPUS v2.2 software (Bruker). The samples were prepared as potassium bromide disks by mixing and grinding the sample (;1% w/w) with potassium bromide. The mixture was compressed in a metal die under vacuum at 550 MPa. Attenuated total reflection (ATR) spectra of the methylamphetamine samples were acquired using a field portable HazMatIDTM spectrometer (SensIR Technologies, L.L.C.) fitted with a zinc selenide ATR crystal and a thermoelectrically cooled DTGS detector. Measurements were recorded at a resolution of 4 cm1 from 4000–650 cm1 with 32 scans accumulated. Spectral measurements, manipulations, and evaluations were performed by HazMatIDTM v1.2.1 software (SensIR Technologies, L.L.C.). Powdered samples were clamped down at a set pressure, which did not cause further sample crushing, to ensure consistent contact between the sample and the ATR crystal. The sampling area was ;7 mm2 (;3 mm diameter). Raman spectra were obtained with a FirstDefenderTM (Ahura Scientific) field portable spectrometer fitted with a 785 nm excitation laser (maximum output of 300 mW) and a 2048 pixel silicon charge-coupled device (CCD) detector. The methylamphetamine mixtures were placed in non-fluorescing 4 mL vials (Ahura Scientific, Part No. 548-00471) and subsequently placed in the spectrometer’s Integrated 4 mL Vial Holder. Spectra were acquired in ‘‘Auto’’ mode, which accumulates signal to a preset level to obtain maximum signal to noise ratios and prevent detector overload. Accumulation times were less than five minutes.

Materials used in this study. Material

Glucose Caffeine Paracetamol Ephedrine hydrochloride Pseudoephedrine hydrochloride Methylamphetamine hydrochloride

Source

Purpose

APS Finechem BDH Chemicals, Victoria (AR grade) Fluka AG, Chem. Fabrik (.98%, anhydrous powder)

Calibration and test mixtures Diluent for MA mixtures Calibration and test mixtures, adulterant for MA mixtures Adulterant for MA mixtures Calibration and test mixtures Test mixtures Reference material

Chemistry Centre (WA) Aldrich Chemical Company School of Pharmacy, Curtin University of Technology Chemistry Centre (WA), (96% purity as HCl salt, 77.2% as free base)

APPLIED SPECTROSCOPY

641

TABLE II. Composition of methylamphetamine HCl samples provided by the Illicit Drugs Section (CCWA), as determined by HPLC.

Sample ID

Synthesis method

Methylamphetamine (%)

Ephedrine (%)

CL2 CL4 RP2

Hypophosphorus acid/iodine Sodium/liquid ammonia Red phosphorous/iodine

79.5 79.2 81.4

0.0 3.2 0.0

Regression Software. Principal component analysis and regression were performed on the full FT-IR spectra with The Unscramblert v9.7 (Camo Software AS, Norway).

RESULTS AND DISCUSSION As a preliminary to quantifying mixed powder methylamphetamine samples, ephedrine hydrochloride was used as a model substance. These ephedrine hydrochloride mixtures were dispersed in KBr to minimize the effect of possible sample heterogeneity and spectra obtained from pressed disks. Methylamphetamine mixtures were analyzed as undiluted powders using field-portable ATR-FT-IR and Raman spectrometers. Quantitative Transmission Fourier Transform Infrared Spectroscopy. A constrained lattice (Fig. 2) design of calibration mixtures was used for method development on a laboratory-based instrument. The design consisted of three factors (ephedrine HCl, glucose, and caffeine) at five levels (0, 25, 50, 75, and 100% w/w), with the sum of the components always being 100% (i.e., only two independent variables). The calibration design removes co-linearity between component mixtures and leads to more robust calibrations. A constrained lattice design (Fig. 3) was also used for validation mixtures comprising ephedrine HCl, caffeine, and glucose. The design covers the full range of the lattice with a minimal number of mixtures and enables the predictive ability of calibration models to be evaluated independent of the calibration data. All mixtures were homogenized by hand in an agate mortar and pestle before being pressed into disks. Spectra were obtained in duplicate. Spectra Acquisition and Data Pretreatment. Transmission FT-IR spectra of the calibration mixtures showed varying baseline offsets and slopes (Fig. 4). In addition, the varying amounts of mixture used in the preparation of the KBr disks

FIG. 2. Constrained lattice mixture design at five levels (0, 25, 50, 75, and 100% w/w) with three factors (ephedrine HCl, caffeine, and glucose).

642

Volume 62, Number 6, 2008

FIG. 3. Constrained lattice design for validation mixtures containing ephedrine HCl diluted with glucose and caffeine.

results in variable intensities. To avoid the inclusion of such variability in the calibration, the data was pretreated by baseline zeroing (Eq. 1) followed by area normalization (Eq. 2), where Ij is a data point in the spectrum and I1934 is the intensity at 1934 cm1. The intensity at 1934 cm1 was selected for baseline zeroing as that region of the spectrum was featureless and was common to all the spectra. Baseline zeroing removes problems associated with baseline offsets (resulting from opacity/scattering effects associated with KBr disks), while area normalization adjusts for the variation of the amount of analyte in the beam. The normalization scaling factor (100) is arbitrary and was selected for ease of comparison and computational efficiency. The results of scaling are shown in Fig. 5. IjðZeroedÞ ¼ Ij  I1934

ð1Þ

IjðZeroedÞ 3 100 IðArea NormalizedÞÞ ¼ X n ½IðZeroedÞ 

ð2Þ

n¼1

An alternative treatment to baseline zeroing and intensity scaling to address disk quality and analyte quantity issues is to take derivatives. First derivatives remove the effect of baseline offsets and second derivatives will remove the effect of sloping baselines. However, derivatives are inherently noisier than the

FIG. 4. Unscaled transmission (KBr) FT-IR spectra acquired from the ephedrine HCl, glucose, and caffeine calibration mixtures.

FIG. 5. Baseline zeroed and area normalized transmission (KBr) FT-IR spectra of the ephedrine HCl, glucose, and caffeine calibration mixtures.

absorbance spectra and a 5-point Savitzky–Golay filter was applied to both derivatives before they were assessed, along with spectra scaling, for predicting analyte concentrations. Development and Evaluation of Principal Component Regression Models. Principal component analysis (PCA) and regression (PCR) were performed on scaled FT-IR spectra and derivatives of the calibration and evaluation sets. The covariance-based PCA was able to resolve the three pure components and intermediate mixtures and showed good agreement between duplicates using only two principal components (PCs) derived from the scaled spectra (Fig. 6). This is expected for a closed three-component system and illustrates the efficacy of pretreating the data, as no principal components are required to describe baseline offsets and/or overall intensity and the two PCs account for 98% of the variance in the absorbance data. Covariance rather than correlation was used for the PCA to prevent minor variations in the spectra exerting undue influence. The validity of this approach was confirmed as correlation-derived PCs did not

FIG. 7. Predictions of the (a) ephedrine HCl, (b) glucose, and (c) caffeine content in the validation mixtures using two PCs from the PCA model of baseline zeroed and normalized transmission FT-IR spectra.

FIG. 6. Scores from the PCA model of scaled infrared spectra of the calibration and validation sets; (u) ephedrine HCl, (n) glucose, (^) caffeine, and (*) mixture.

achieve lower prediction errors than covariance PCs and in some instances were worse. The subsequent prediction of ephedrine HCl (Fig. 7a), glucose (Fig. 7b), and caffeine (Fig. 7c) content in the validation set showed that it was possible to acceptably quantify solid mixtures using FT-IR, the root mean sum of squares of residuals (RMSSR) and error of predictions (RMSEP) being ;4% for all components. Regression models developed from the first and second derivatives of the spectra performed worse than the scaled

APPLIED SPECTROSCOPY

643

TABLE III. Composition of methylamphetamine HCl (MA) mixtures diluted with glucose, caffeine, and paracetamol. Mixture ID 33 34 35 36 37 38 39 40 41

MA (%) a

25.0 39.7b 40.7c 25.7b 26.9c 16.5a 77.2d 0.0 0.0

Glucose (%)

Caffeine (%)

68.6 49.8 50.0 32.3 33.1 45.3 0.0 100.0 0.0

0.0 0.0 0.0 35.3 33.8 33.9 0.0 0.0 100.0

a

Purity of methylamphetamine sample used: 79.5%. Purity of methylamphetamine sample used: 79.2%. Purity of methylamphetamine sample used: 81.4%. d Purity of methylamphetamine sample used: 77.2%. b c

transmission spectra derived models. Selection of the number of PCs required to account for the variance was more subjective in these cases as the first three PCs accounted for only 86% of the variance, with each successive PC beyond this accounting for ,1% of variance. Hence, only three PCs were used in the models. These models had RMSSRs and RMSEPs in the range 3–13% for the three analytes. The poorer performance is most probably due to the noise inherent in spectra derivatives and a degree of subjectivity in the selection of the level of filtering applied. Filtering reduces noise but it also decreases peak heights and increases peak widths, and there is thus an optimal size for a given data set. Our PCA/PCR analysis of 3-, 5-, 7-, and 9-point smoothed data showed that the 5-point filter gave the lowest error of prediction overall for the calibration sets. Even though 7-point smoothing gave lower errors in some instances, for consistency, 5-point smoothing was used throughout. Quantitative Field-Portable Attenuated Total Reflection Fourier Transform Infrared Spectroscopy. The three methylamphetamine HCl samples available could only be serially diluted with glucose and caffeine due to limited availability and hence there is some co-linearity between components (Table III). Each sample had been prepared by a different synthetic route. The spectra of the calibration mixtures (Table III) acquired on the HazMatIDTM had sharp peaks, without the baseline off-set problems due to opacity/scattering effects observed with the KBr disk preparations. Therefore,

FIG. 8. Normalized ATR-FT-IR spectra of methylamphetamine HCl mixtures.

644

Volume 62, Number 6, 2008

FIG. 9. Scores from the PCA model of scaled ATR-FT-IR spectra of methylamphetamine HCl calibration and validation samples; (u) methylamphetamine HCl, (n) glucose, (^) caffeine, and (*) mixture.

only area normalization (Fig. 8) was required to account for the variations in the intensities due to variations in the amount of sample in contact with the ATR crystal. The three components are clearly resolved in the PCA (Fig. 9), along with the intermediate mixtures, again requiring only two principal components to account for 98% of the variance. A PCR model was developed similar to that used for the transmission FT-IR data. Because of the limited amount of material available, the spectra of the methylamphetamine HCl calibration set were reacquired to evaluate the model. Predictions of methylamphetamine, glucose, and caffeine (Fig. 10) were acceptable, with most being predicted to within 6% w/w of their value. However, some predictions were in error by up to 20% w/w, which indicates that there are repeatability problems that probably arise from both heterogeneity and having variable amounts of sample in direct contact with the ATR crystal for each measurement. Averaging triplicates reduced the RMSRR to ;3%. As was found with the transmission spectra, using correlation-derived PCs did not improve the accuracy of prediction and the use of first or second derivatives leads to poorer predictions. Robustness of the Calibration Models. A constrained lattice (Fig. 11) was used to design additional mixtures that contained pseudoephedrine HCl in place of ephedrine HCl as the surrogate illicit drug to evaluate the robustness of calibrations. Similarly paracetamol, a common illicit drug diluent, was used to dilute the methylamphetamine HCl samples (Table IV). The ability to predict known components in mixtures in the presence of ‘‘unknown’’ components, which is the likely scenario for seized samples, was evaluated by assessing the performance of models that specified only the three ‘‘known’’ components, all four components, or solely the target component (ephedrine HCl or methylamphetamine HCl). The PCs selected for the models were again those that individually accounted for .1% of variance. It is evident from Table V that a model based only on ephedrine HCl, glucose, and caffeine performs poorly when

FIG. 11. Constrained lattice design for mixtures containing pseudoephedrine HCl diluted with glucose and caffeine used to assess PCR model robustness.

FIG. 10. Prediction of (a) methylamphetamine, (b) glucose, and (c) caffeine components in the methylamphetamine HCl mixtures using two PCs from the PCA model of normalized ATR-FT-IR spectra.

pseudoephedrine HCl is present. This is expected as the model cannot distinguish between pseudoephedrine HCl and ephedrine HCl (Fig. 12), because their spectra have only minor differences, and this also highlights the well-known problem that occurs with quantifying materials when an unknown component is not accounted for in the calibration population and has a significant influence on the model. Including pseudoephedrine HCl in the model gave similar quality fits to the data and predictions to those observed when only ephedrine HCl, glucose, and caffeine are present. RMSSRs and

RMSEPs using four PCs derived from the scaled spectra (accounting for 99% of variance) were in the 3–6% range for all four components. Calibration and predicting for only ephedrine HCl in ten randomly selected samples gave an RMSSR of 2.5% and an RMSEP of 4.3%. The methylamphetamine HCl, glucose, and caffeine model was similarly affected by dilution with paracetamol. Caffeine is typically predicted 30% w/w higher and glucose 20% w/w higher than their known value, and pure paracetamol was predicted to have ;10% w/w methylamphetamine, ;40% w/w glucose, and ;50% w/w caffeine (Table VI). Including paracetamol improves performance significantly, with discrepancies between predicted and known values being ,7% w/w for 45 of the 52 calibration and evaluation mixtures, with the worst case being 20% w/w (Table VII). Using three PCs (accounting for 98% of variance), RMSSRs were in the range 3–5% and RMSEPs were 3.5%, 5.4%, 2.4%, and 5.3%, respectively, for methylamphetamine, glucose, caffeine, and paracetamol. The poor prediction of pure paracetamol (sample 45) is most likely due to the significant mutual overlap of the spectra of the four components, while the poor prediction of samples 33, 34, and 37 may be due to the heterogeneity of the bulk sample, as more than one component (methylamphetamine HCl, glucose, and/or caffeine) is poorly predicted. Using the entire data set but leaving out six randomly selected samples as an evaluation set to predict only methylamphetamine content gave a similar quality of fit and prediction (RMSSR ¼ 3.9%, RMSEP ¼ 2.6%), without major discrepancies. The choice of regression strategy significantly affects outcomes; hence, a decision must be made as to whether the aim of the analysis is to target a specific analyte or to fully characterize the sample. If the former, then the components likely to be encountered in unknown samples must be present in at least some of the calibration set, but their concentrations TABLE IV. Percentage composition of samples of methylamphetamine HCl (MA) mixtures diluted with 50% paracetamol. Mixture ID 42 43 44 45 a b c

Paracetamol (%) 50.0 50.0 49.2 100.0

MA (%) a

12.9 13.5b 8.4c 0.0

Glucose (%)

Caffeine (%)

16.1 16.5 23.0 0.0

17.7 16.9 17.2 0.0

Purity of methylamphetamine sample used: 79.2%. Purity of methylamphetamine sample used: 81.4%. Purity of methylamphetamine sample used: 79.5%.

APPLIED SPECTROSCOPY

645

TABLE V. Prediction of ephedrine HCl, glucose, and caffeine components in pseudoephedrine HCl test mixtures using two PCs from PCA model of scaled infrared spectra. Ephedrine HCl (%)

Glucose (%)

Caffeine (%)

Mixture ID (Fig. 11)

Actual

Predicted

Actual

Predicted

Actual

Predicted

23 24 24 25 25 26 26 27 27 28 28 29 30 30 31 31 32

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

45.8 45.8 38.1 40.1 35.5 23.2 21.7 16.8 21.7 19.3 25.9 1.7 15.7 2.8 0.7 1.3 9.6

0.0 33.2 33.2 0.0 0.0 66.6 66.6 33.0 33.0 0.0 0.0 100.0 67.1 67.1 33.9 33.9 0.0

50.7 56.5 65.2 12.4 18.7 83.1 84.8 38.2 33.3 0.7 17.6 104.8 40.9 56.6 25.5 23.8 4.2

0.0 0.0 0.0 33.3 33.3 0.0 0.0 33.5 33.5 67.3 67.3 0.0 32.9 32.9 66.1 66.1 100.0

3.5 2.3 3.3 47.6 45.8 6.3 6.5 45.0 45.1 80.0 91.7 3.0 43.4 40.6 73.8 77.4 94.6

are not required. If the latter, all potential components and their concentrations must be encompassed by the calibration set. Field-Portable Raman Spectroscopy. The Raman spectra of selected methylamphetamine HCl mixtures obtained with the field-portable FirstDefenderTM spectrometer are shown in Fig. 13. The spectra are well resolved, of acceptable quality, and qualitatively clearly show the presence of methylamphetamine HCl from its peaks at 622, 836, and 1002 cm1. However, they show that even with long wavelength excitation (785 nm) some samples still fluoresce or have complex baselines. This along with a small sample area (;1.3 mm2) renders quantification more difficult than is the case with ATRFT-IR. Furthermore, Raman spectroscopy could be easily defeated by illicit drug producers with the addition of a trace amount of fluorescent material. Hence, we did not pursue it for quantitative analyses.

CONCLUSION Fourier transform infrared spectroscopy has been shown to have sufficient quantitative accuracy for the screening of seized methylamphetamine samples in the field. Both transmission

FIG. 12. Transmission (KBr) FT-IR spectra of ephedrine HCl and pseudoephedrine HCl.

646

Volume 62, Number 6, 2008

FT-IR and ATR-FT-IR are much less time consuming than traditional chromatographic methods and hence are suitable for rapid quantitative assessment of illicit drug mixtures. However, the transmission technique is less suitable for on-site analyses because of its lack of transportability. It seems unlikely that Raman spectroscopy would meet with the same degree of success as FT-IR spectroscopy as sample heterogeneity may be a greater problem and the method can be defeated by the addition of a small amount of fluorescent material to illegally manufactured substances. Attenuated total reflection Fourier transform infrared spectra are affected by the amount of an individual component in a sample that is in contact with the ATR crystal, which is dependant on particle size and sample heterogeneity. Hence, care must be taken to ensure that the calibration set and unknowns have the same particle size and distribution and are adequately mixed before spectra are obtained. For transmission FT-IR the particle size and distribution are less of an issue, but representative sampling of heterogeneous materials is obviously still crucial. For quantitative FT-IR work, baseline zeroing and area normalization of the spectra are necessary data pretreatment methods for transmission FT-IR as they reduce the influence of sample preparation variance, but baseline zeroing is not necessary for ATR-FT-IR spectra. Calibration models based on these spectra can be developed using principal component analysis (PCA) and principal component regression (PCR). With ATR-FT-IR field-portable instrumentation, the predictions of methylamphetamine, glucose, and caffeine, using a PCR model developed from two PCs from a covariance PCA, were generally within 6% w/w of the known values, with RMSEPs of ;4%, as was found for the transmission method. Predictions for methylamphetamine, glucose, caffeine, and paracetamol using a PCR model developed with three PCs from a covariance PCA gave similar results, while predicting only methylamphetamine has an RMSEP of 3%. Using PCs derived from correlation PCA for regression (to increase the weighting given to minor peaks in the spectra) does not significantly alter the accuracy of prediction, while using first or second derivatives generally leads to poorer predictions. To

TABLE VI. Prediction of methylamphetamine (MA), glucose, and caffeine components in the validation mixtures where paracetamol was an unknown major diluent using two PCs from a PCR model. MA (%)

Glucose (%)

Caffeine (%)

Mixture ID

Actual

Predicted

Actual

Predicted

Actual

Predicted

42 42 43 43 44 44 45 45

12.9 12.9 13.5 13.5 8.4 8.4 0.0 0.0

13.6 14.2 13.3 14.4 10.2 9.8 8.3 11.1

16.1 16.1 16.5 16.5 23.0 23.0 0.0 0.0

35.9 35.5 37.1 37.4 40.1 41.3 39.6 39.3

17.7 17.7 16.9 16.9 17.2 17.2 0.0 0.0

47.0 46.6 46.2 44.5 47.1 46.4 50.0 46.8

TABLE VII. Prediction of methylamphetamine (MA), glucose, caffeine, and paracetamol components in the validation mixtures where paracetamol was a known major diluent using three PCs from the PCR model of normalized ATR-FT-IR spectra. MA (%)

Glucose (%)

Caffeine (%)

Paracetamol (%)

Mixture ID

Actual

Predicted

Actual

Predicted

Actual

Predicted

Actual

Predicted

33 34 35 36 37 38 39 39 39 39 40 41 42 43 44 45

25.0 39.7 40.7 25.7 26.9 16.5 79.5 79.2 81.4 77.2 0.0 0.0 12.9 13.5 8.4 0.0

31.3 24.3 44.2 23.9 30.2 16.7 76.6 81.2 81.9 77.7 0.1 2.2 11.3 11.7 7.1 6.9

68.6 49.8 50.0 32.3 33.1 45.3 0.0 0.0 0.0 0.0 100.0 0.0 16.1 16.5 23.0 0.0

58.2 68.6 43.0 34.4 21.7 47.1 2.3 0.7 0.5 1.3 100.9 3.0 13.1 16.1 20.0 7.7

0.0 0.0 0.0 35.3 33.8 33.9 0.0 0.0 0.0 0.0 0.0 100.0 17.7 16.9 17.2 0.0

0.2 0.1 0.2 38.6 42.4 31.7 1.5 1.2 1.1 0.4 0.3 96.8 15.0 14.5 16.4 2.3

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 50.0 50.0 49.2 100.0

2.4 0.9 1.8 3.0 2.0 0.3 0.0 1.7 1.4 0.7 0.9 2.5 57.7 54.7 54.8 81.3

reduce noise in the derivatives a degree of smoothing is required, which can lead to a loss of resolution and consequently poorer quantification. As smoothing is always subjective, totally objective measurements such as the absorbance spectra are likely to be less prone to error.

FIG. 13. Selected dispersive Raman spectra of methylamphetamine samples acquired by the FirstDefenderTM, (a) methylamphetamine HCl, (b) methylamphetamine HCl/glucose, (c) methylamphetamine HCl/glucose/caffeine, and (d) methylamphetamine HCl/glucose/caffeine/paracetamol.

Assessment of the robustness of transmission FT-IR and ATR-FT-IR calibration models to the presence of ‘‘unknown’’ substances in the samples clearly points out that the validity of quantification is dependent on the comprehensiveness of the calibration set used. Samples that contain components that are not encompassed by the calibration set will almost inevitably produce erroneous results, be it for the determination of a single analyte or a more comprehensive analysis. Providing that such samples are recognized and quantified by other means, the calibration set would be able to be updated to include those components. It is clear that attenuated total reflection Fourier transform infrared spectroscopy, in combination with multivariate calibration methods such as PCR, has the potential to allow illicit substances to be quantitatively analyzed at the point of seizure or on-site at clandestine laboratories. To do this successfully will require larger calibration data sets than those used here and outlier detection to assure the quality of results. 1. M. D. Cole, The Analysis of Controlled Substances (John Wiley and Sons, Chichester, 2003), Chap. 2, p. 13. 2. I. S. Lurie, C. G. Bailey, D. S. Anex, M. J. Bethea, T. D. McKibben, and J. F. Casale, J. Chromatogr., A 870, 53 (2000). 3. P. A. Hays, J. Forensic Sci. 50(6), 1 (2005). 4. B. C. Smith, Fundamentals of Fourier Transform Infrared Spectroscopy (CRC Press, London, 1996), pp. 117–125.

APPLIED SPECTROSCOPY

647

5. P. J. Gomm and I. J. Humphreys, J. Forensic Sci. Soc. 15, 293 (1975). 6. M. Lo´pez-Arı´guez, A. Camea´n, and M. Repetto, J. Forensic Sci. 40, 602 (1995). 7. R. Levy, M. Ravneby, L. Meirovich, and O. Shapira-Heiman, J. Forensic Sci. 41, 6 (1996). 8. M. Ravreby and A. Gorski, J. Forensic Sci. 34, 918 (1989). 9. D. Wielbo and I. R. Tebbett, J. Forensic Sci. 37, 1134 (1992). 10. A. G. Ryder, G. M. O’Connor, and T. J. Glynn, J. Forensic Sci. 44, 1013 (1999).

648

Volume 62, Number 6, 2008

11. S. E. J. Bell, D. T. Burns, A. C. Dennis, L. J. Matchett, and S. J. Speers, Analyst (Cambridge, U.K.) 125, 1811 (2000). 12. S. E. J. Bell, D. T. Burns, A. C. Dennis, and S. J. Speers, Analyst (Cambridge, U.K.) 125, 541 (2000). 13. E. Katainen, M. Elomaa, U.-M. Laakkonen, E. Sippola, P. Niemela, J. Suhonen, and K. Jarvinen, J. Forensic Sci. 52, 88 (2007). 14. A. G. Ryder, J. Forensic Sci. 47, 275 (2002). 15. C. J. Strachan, D. Paratiwi, K. C. Gordon, and T. Rades, J. Raman Spectrosc. 35, 347 (2004).

Evanescent-Wave Cavity Ring-Down Spectroscopy for Enhanced Detection of Surface Binding Under Flow Injection Analysis Conditions L.

VAN DER

SNEPPEN, J. B. BUIJS, C. GOOIJER, W. UBACHS, and F. ARIESE*

Laser Centre, Vrije Universiteit, De Boelelaan 1081–1083, 1081 HV Amsterdam, The Netherlands

The feasibility of liquid-phase evanescent-wave cavity ring-down spectroscopy (EW-CRDS) for surface-binding studies under flow-injection analysis (FIA) conditions is demonstrated. The EW-CRDS setup consists of an anti-reflection coated Dove prism inside a linear cavity (with standard or super-polishing of the total internal reflective (TIR) surface). A teflon spacer with an elliptical hole clamped on this surface acts as a 20 lL sized flow cell. The baseline noise of this system is of the order of 104 absorbance units; the baseline remains stable over a prolonged time and the prism surface does not become contaminated during repeated injections of the reversibly adsorbing test dyes Crystal Violet (CV) and Direct Red 10 (DR10). At typical FIA or liquid chromatography (LC) flow rates, the system has sufficient specificity to discriminate between species with different surface affinities. For CV a much stronger decrease in ringdown time is observed than calculated based on its bulk concentration and the effective depth probed by the evanescent wave, indicating binding of this positively charged dye to the negatively charged prism surface. The amount of adsorption can be influenced by adjusting the flow rate or the eluent composition. At a flow rate of 0.5 mL/min, an enrichment factor of 60 was calculated for CV; for the poorly adsorbing dye DR10 it is 5. Super-polishing of the already polished TIR surface works counterproductively. The adsorbing dye Crystal Violet has a detection limit of 3 lM for the standard polished surface; less binding occurs on the superpolished surface and the detection limit is 5 lM. Possible applications of EW-CRDS for studying surface binding or the development of bio-assays are discussed. Index Headings: Evanescent-wave cavity ring-down spectroscopy; EWCRDS; Flow injection analysis; FIA; Surface interactions; Silica surface; Crystal violet.

INTRODUCTION Cavity ring-down spectroscopy (CRDS) is an absorbance spectroscopic technique that is based on the measurement of the decay rate of light after abrupt termination of the excitation of a high-finesse optical resonator. It has certain advantages over conventional absorbance spectroscopy: since it is a multipass technique, extremely low concentration detection limits are feasible. Alternatively, analytes with low extinction coefficients or in very thin layers can still be detected. Furthermore, since it is based on the measurement of the rate of decay of light rather than the measurement of DI/I0, sources with high intensity fluctuations such as pulsed lasers can be used. CRDS is well established and widespread in the gas phase, and applications of CRDS to liquid-phase studies are currently being developed.1,2 A special mode of CRDS that is gaining considerable interest is evanescent-wave cavity ring-down spectroscopy (EW-CRDS), a technique that combines the high sensitivity of the CRDS technique with surface-specificity. In EW-CRDS, Received 25 July 2007; accepted 7 March 2008. * Author to whom correspondence should be sent. E-mail: [email protected]. nl.

Volume 62, Number 6, 2008

one or more of the reflections inside the optical resonator are a total internal reflection (TIR) event. Only light of the evanescent wave associated with this TIR event is used for probing the sample. The evanescent wave decays exponentially with distance from the surface, and hence mostly species that are near, physisorbed, or chemisorbed to the surface will interact with the radiation field. The penetration depth of the evanescent wave is typically on the order of one wavelength or less. This explains the current interest in evanescent-wavebased techniques for detecting surface binding phenomena including bio-assays on surfaces.3–5 Currently, three different implementations of EW-CRDS have been explored. The use of a monolithic ring cavity was already suggested and demonstrated a decade ago6,7 by Pipino et al. In follow-up studies, monolithic or folded resonators were used;8–10 this is an elegant solution to minimize intrinsic cavity losses, but the requirements on the surface quality, the morphology of the TIR surfaces, and the purity of the resonator material are very stringent. The principle of fiber-loop CRDS has been explored by O’Keefe’s group, who constructed a high-finesse fiber cavity by utilizing fiber-Bragg gratings (FBGs) as reflectors.11 Independently, this technique was developed in the groups of Loock and Lehmann by looping the fiber onto itself and using input and output couplers for the injection and detection of the pulse train.12,13 The use of intra-cavity elements is the most straight-forward method to create an EW-CRDS setup:14–21 off-the-shelf or easy-to-produce custom-made optics can be used. Obviously, the surfaces placed inside the cavity cause reflection and refraction losses that affect the performance of EW-CRDS and the achievable detection limits. Possible intra-cavity elements include Dove prisms placed inside a linear cavity, standard right-angle prisms, and Pellin–Broca prisms. However, to ensure TIR at a silica–liquid interface, the angle of incidence on the TIR surface should be larger than about 668; hence, a linear cavity with a right-angle or Pellin–Broca prism is not appropriate for liquid-phase EW-CRDS studies on glass or quartz TIR surfaces. Usually, an anti-reflection coating is applied to minimize reflection losses at the entrance and exit surfaces. Normal-incidence geometries have several advantages over non-normal incidence geometries. Firstly, when (relatively) small cavities are used, additional reflection losses of the surfaces can be maintained within the cavity.22 In addition, both s- and p-polarized light can be used, permitting orientational studies using the dichroic ratio. The use of a triangular prism in a ring cavity was explored by the group of Zare, to determine the molecular orientation of a film of methylene blue at the air–silica interface.17 Shaw et al. studied adsorption isotherms of Crystal Violet to silica as a function of pH and in the presence of other

0003-7028/08/6206-0649$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

649

cations14,15 using an anti-reflection coated (R  0.25%) Dove prism inside a cavity. Furthermore, the adsorption isotherm of hemoglobin from urine samples was determined in order to assess the feasibility of using EW-CRDS diagnosis in hemoglobinuria.16 Adsorption of a charged and an uncharged dye at the silica/CH3CN surface has been measured by Fan et al.19 using a similar setup. Mazurenka et al. built an electrochemical cell on top of an anti-reflection coated, standard right-angle prism inside a ring cavity.20 Simultaneous EW-CRDS and cyclovoltammetry measurements allowed the determination of concentration changes of chromophoric [Fe(CN)6]3 upon oxidation of [Fe(CN)6]4. Everest et al. used a custom-made Dove prism with entrance and exit faces at normal incidence.21 The adsorption isotherm of hemoglobin to the silica interface as well as the average orientation of the hemoglobin transition dipole moment was measured. Until now, liquid-phase EW-CRDS studies have included steady-state measurements16,21 or measurements with an extremely low continuous flow (on the order of mL/h) of sample in a 190 lL sized flow cell.14,15 The experimental conditions were such that adsorption equilibrium could be reached and thermodynamic properties could be measured; reversibility was not a critical factor. In contrast, our ultimate goal is the immobilization of biomolecules such as monoclonal antibodies and to detect compounds of interest showing interaction with these biomolecules using flow injection analysis (FIA) or liquid chromatography (LC). The use of biologically modified surfaces will be the subject of forthcoming studies. In this paper the test compounds Crystal Violet (CV) and Direct Red 10 (DR10) on unmodified surfaces are used to examine the performance of EW-CRDS detection under flow injection analysis (FIA) or liquid chromatography (LC) conditions. A generally encountered difficulty when using such test compounds is their irreversible adsorption to the TIR surface, necessitating rigorous cleaning (applying acids or alkaline solutions) or even etching of the surface after each measurement. An important next step in liquid-phase EW-CRDS is making such measurements compatible with FIA or LC by minimizing the flow cell dimensions, using flow rates on the order of mL/min, and having control over the adsorption process so that the TIR surface can be used over a prolonged time for repeated measurements. In this study, flow injection measurements of the positively charged dye CV and the negatively charged dye DR10 are compared. The addition of an organic modifier to the eluent— as is quite common in reversed-phase LC—ensures a reversible chromatography-like adsorption process. The observed ringdown times are converted to absorbance units and compared with the predicted values in the absence of adsorption based on the layer thickness probed by the evanescent wave; the difference indicates specific surface binding. The hydrophilic, negatively charged silica surface shows a strong preference for CV over DR10. The adsorption of CV could be dramatically increased by lowering the flow rate; for DR10 such effects are far less pronounced. It should be stressed that the scope of the present study is to explore conditions for the application of EW-CRDS in flow systems or LC rather than the characterization of the test molecules used. Since the adsorption properties of CV have been investigated previously in several studies,14,15 CV was chosen for testing the reversibility of adsorption to the surface.

650

Volume 62, Number 6, 2008

EXPERIMENTAL Crystal Violet (CV, color index 42555, 98%, e532 ¼ 5.3 3 104 M1cm1) was obtained from Aldrich Chem. (Milwaukee, WI) and Direct Red 10 (DR10, e532 ¼ 1.0 3 104 M1cm1) was obtained from Sigma-Aldrich (Seelze, Germany). Solutions of CV, diluted from a 0.2 mM stock solution in Milli-Q and DR10 were prepared in the eluent that was used. The eluent was 50% (v/v) 10 mM potassium phosphate buffer (pH ¼ 4.3, 7.4, or 8.8) in HPLC-grade methanol; a flow of 0.5, 0.3, or 0.1 mL/min was delivered using an LC-pump; and 100 lL of sample was injected using a six-way injection port. The eluents were degassed by sonication before use. In order to test the influence of the buffer and the ionic strength on adsorption of CV to the surface, 10 mM HEPES (4-(2-hydroxyethyl)piperazine-1ethanesulfonic acid) buffer at pH 7.4, as well as 100 mM phosphate and HEPES buffers, were used in the eluent with 50% of methanol as well. A cell with a volume of 20 lL was constructed by clamping a 1 mm thick teflon spacer with an elliptical hole of 9.4 3 3.5 mm on top of a Dove prism (see Fig. 1). Two anti-reflection coated (R  0.25% at 532 nm, n ¼ 1.52, 458 facets) BK7 Dove prisms (Casix, Fuzhou, China) were used, one with a surface roughness rms  1.5 nm, k/2 at 632.8 nm as received from the manufacturer (‘‘standard polish’’) and one with a custom polishing (Layertec, Germany, surface roughness rms  0.2 nm, k/10 at 632.8 nm). Cleaning was performed by simply wiping the TIR surface of the prism with water or methanol. A cavity was constructed using mirrors (R  99.996% at 532 nm, 50 mm radius of curvature) from REO Inc. (Boulder, CO). Alignment of the Dove prism inside the linear cavity was relatively easy: the laser was centered on the entrance and exit mirror holders, using two pinholes, and was then optimized with the mirrors in place. In the absence of the Dove prism, this yielded a ring-down time of several microseconds. Next, the mirrors were replaced with the pinholes again, and the Dove prism (which was already in place on the prism mount) was inserted by vertical movement of the prism mount using a micrometer screw. A combination of horizontal and angle movement of the prism was used to make sure that the beam path is not altered by the Dove prism. The prism mount was subsequently moved out of the beam path and the mirrors were inserted and aligned again by optimizing the ring-down transient behind the linear cavity. The Dove prism was inserted again, and the photomultiplier tube (PMT) was moved under an angle of 90 degrees of the cavity axis so that the scatter losses of the exit face of the Dove prism (R  0.25% on each pass) are detected.20 This ensures that more light (0.25% on each pass, rather than 0.004%) will reach the detector. Ring-down times for both the super-polished and standard polished TIR surfaces were between 20 and 25 ns, indicating that the surface roughness of the TIR surface of the prisms does not contribute significantly to the total losses of the system. This short ringdown time is on the order of the maximum achievable ringdown time with the current cavity: with the anti-reflection coatings specified at R  0.25% per surface and with two surfaces per pass, 200 passes are expected. Considering the relatively short cavity (85 mm), this corresponds to a ringdown time of 57 ns. Of course, additional reflection and scatter losses will take place, resulting in a ring-down time on the order of 20 to 25 ns. The fact that about 10% of the original ring-down time of the empty cavity is maintained after

FIG. 1. Schematic of the EW-CRDS setup (not drawn to scale). An anti-reflection-coated Dove prism is mounted on a mirror holder in a 70 mm long linear cavity. The effective path length with the prism in place, taking into account the refractive index, is 85 mm. Reflection loss at the 458 exit surface is used for detection of the transient. A teflon spacer with an elliptical hole is clamped leak-tight on the TIR face of the prism, and flow is delivered by PEEK tubing connected with fingertights. The mirror holder, mounted on an x,z-stage permitting horizontal and vertical movement, enables two degrees of freedom for tilting and in-plane rotation.

insertion of an intra-cavity element is quite common (for example, Ref. 14). Laser pulses at a wavelength of 532 nm, 2.4 ns duration, and a nominal bandwidth of ’10 cm1 were used to excite the cavity. As previously discussed, excitation with narrow-band radiation from injection-seeded Nd:YAG lasers would result in mode-beating on the transients,23 and therefore the 532 nm radiation at sufficiently large bandwidth was produced through the use of an optical parametric oscillator (OPO) pumped with the third harmonic of a Coherent Infinity single-mode Nd:YAG laser (Santa Clara, CA) at 355 nm;2 this laser has the advantage that the repetition frequency can be varied between 10 and 100 Hz without modifications to the beam profile. For the CRDS measurements no mode-matching was performed, ensuring multi-mode excitation of the cavity. The optical path length was L ¼ 85 mm (including a correction for the Dove prism with n ¼ 1.52). A scheme of the setup is shown in Fig. 1; the laser beam was p-polarized with respect to the 458 facets of the prism. Transients were recorded using a photomultiplier tube (Hamamatsu, Shimokanzo, Japan) and a fast sampling oscilloscope of 1 GHz analog bandwidth (Tektronix 5104 5 GS/s). An auxiliary photodiode was used to trigger the detection system. Typically, the first 15 ns of the detected trace, which contains the instrumental response function, was rejected; 65 ns (corresponding to about three ring-down times s0) of the remaining trace was fitted. A moving average of 1 second or 100 data points was applied afterwards. Absorbance units (eCl) were derived from the ring-down times via the following equation:   aanal l L 1 1 ¼  eCl ¼ ð1Þ 2:303 2:303c s s0 where e is the molar extinction coefficient in M1cm1 at 532 nm, C is the concentration in M, and l is the effective path length of the evanescent wave through the sample in cm. L is the cavity length (85 3 103 m), c is the speed of light, aanal is the absorption coefficient in cm1, s and s0 are the ring-down times in the presence or absence of analyte. In the case of

adsorption to the surface, the local concentration C will increase, leading to a shorter ring-down time s.

RESULTS AND DISCUSSION In view of our future applications, the performance of EWCRDS is examined at flow rates that are compatible with FIA and LC (0.1–0.5 mL/min). Under such flow rate conditions no real Langmuir adsorption isotherm was observed for CV, indicating that the adsorption equilibrium could not be reached. Nevertheless, even at pH 4.3, where by far the largest fraction of the silanol groups is not dissociated (a silica surface typically consists of 19% geminal or Q2 (pKa ¼ 4.5) and 81% vicinal or Q3 (pKa ¼ 8.5) silanol sites),15 adsorption of CV can still be observed, indicating that in addition to ionic interactions (induced) dipole interactions also play a role. For a detailed understanding about the adsorption mechanism, the possible existence of shearing effects and formation of a boundary layer should also be considered, but this is not within the scope of the current study. For DR10, adsorption effects appear to be far less important. The baseline of the absorption curve remained stable over a prolonged time and after repetitive injections of adsorbing dye; the prism surface did not degrade and could be used for days without cleaning. The baseline drift is acceptable (stable baselines for 45 minute measurements have been observed) and the tailing of the CV and DR10 peaks is comparable with the tailing observed in LC studies at a flow rate of 0.5 mL/min. This demonstrates that the surface of the prisms does not become contaminated and irreversibility of adsorption does not play a major role; the surface can be used for measurements over a prolonged timescale. Typical flow-injection profiles of low CV concentrations (5–20 lM) at 0.5 mL/min can be seen in Fig. 2. Spikes on the baseline are sometimes observed due to the presence of air bubbles in the flow cell, despite the fact that all eluents were sonicated before use. At this flow rate the response of the system to CV does not change significantly with pH over the range of 4.3 to 8.8 (data not shown). The absolute absorbances observed in our flow system at 0.5 mL/min (linear flow 5.3 mm/s) are more than an order of magnitude lower than those

APPLIED SPECTROSCOPY

651

FIG. 2. Influence of surface roughness. FIA profiles showing triplicate injections of, respectively, 5, 10, 15, and 20 lM CV in 10 mM potassium phosphate buffer, pH ¼ 7.4, on the super-polished and the standard polished TIR surface. The flow rate was 0.5 mL/min. The values on the y-axis are in absorbance units.

FIG. 3. Response curves of CV at a flow rate of 0.5 mL/min and a pH of 7.4 as observed on the standard polished and the super-polished TIR surface. Each concentration was measured in triplicate; the error bars represent one standard deviation.

observed by Fisk et al.15 under semi-stationary conditions in a purely aqueous buffer. While Fisk et al. have successfully explained their observations with a multi-layer adsorption model, our system is far removed from equilibrium, thus complicating a detailed understanding of the data. On the standard polished surface, the response of CV is much stronger than on the super-polished surface, which is also obvious from the detection limits being 3 lM for the standard polished TIR surface and 5 lM for the super-polished TIR surface. These results indicate that for analytical purposes standard polished surfaces seem to be more appropriate. For the repeated injections the peak height variation is acceptable: typically 15%. Memory effects were not observed. In Fig. 3, flow injection peak areas for CV are plotted as a function of concentration at 0.5 mL/min. Unexpectedly, the super-polished TIR surface gives a nonlinear response to CV, whereas for the standard polished surface a linear regression line is obtained. Overall, at lower concentrations the performance of the standard polished TIR surface is better, while at higher concentrations the responses on the two different surfaces are more similar. Upon reducing the flow rates to 0.3 or 0.1 mL/min, a similar nonlinear response to the CV concentration is observed for the standard polished surface (data not shown). We assume that the higher surface area of the standard polished surface is advantageous in the detection of low concentrations of CV. Additionally, the surface roughness of the standard polished prism might cause a turbulent flow at high flow rates, ensuring a more efficient exchange between the boundary layer and the bulk solution and thus stimulating further binding of CV. The flow regime inside the flow cell (laminar or turbulent) may be estimated from the Reynolds number:15

surface upon lowering the flow rate as well. It seems therefore possible that the flow is in a transition from turbulent to laminar at the utilized flow rates and is more turbulent on the standard polished surface. Contrary to CV, the DR10 response is linear for both the super-polished and standard polished TIR surface, also at low flow rates (see further below). At low flow rates (0.1 mL/min) the tailing increased considerably (see Fig. 4), but within a time-span of minutes the ring-down time increased to the original value again. The width of the CV peak corresponds roughly to 1/flowrate, whereas the area of the peak corresponds to (1/flowrate)2. At a flow rate of 0.5 mL/min and a pH of 4.3 (where most silanol groups will be neutral) the tailing was far worse than at pH 7.4 and 8.8 (data not shown), indicating that the adsorption of CV to silica is of a more irreversible nature. For bulk absorbance, one can calculate the optical path length from the effective depth de for p-polarized (in the plane of the beam and the surface normal) light:24

Re ¼

qud l

Volume 62, Number 6, 2008

n21 kð2sin2 h  n221 Þcosh n1 pð1  n221 Þ½ð1 þ n221 Þsin2 h  n221 ðsin2 h  n221 Þ1=2

ð3Þ

ð2Þ

where q is the density of the fluid (the density of water, 998 kg m3 is assumed), l is the dynamic fluid viscosity (0.8909 3 103 N s m2), and d is the thickness of the flow cell (103 m). u is the flow velocity; with the flow cell volume of 14 lL and length of 9 mm, this is 5.3, 3.2, or 1.1 mm s1 for 0.5, 0.3, and 0.1 mL/min, respectively. Calculated Reynolds numbers are 5.9, 3.6, and 1.2, and the flow is likely to be turbulent (the critical Reynolds number is around 1; flows above this number are turbulent). Especially at low flows the flow regime will become more laminar, and as mentioned before, a nonlinear concentration dependence can be seen on the standard polished

652

de ¼

FIG. 4. Influence of flow rate on CV adsorption. Flow profiles showing an injection of 20 lM CV in 10 mM potassium phosphate buffer, pH ¼ 7.4 on the super-polished TIR surface. Flow rates are 0.1, 0.3, and 0.5 mL/min; sample injection volume is 100 lL. Profiles were horizontally shifted for ease of comparison. The values on the y-axis are in absorbance units. The inset shows the peak area as a function of flow rate as measured for 20 lM CV on the standard polished TIR surface.

FIG. 5. Response curves of CV and DR10 at different flow rates and pH ¼ 7.4 as observed on the super-polished TIR surface. Each concentration was measured in triplicate; the error bars represent one standard deviation. Injected concentrations of CV were varied between 5 and 40 lM (lower scale); those of DR10 ranged from 100 to 1000 lM (upper scale), plotted in the same graph for ease of comparison. Lines between the points are guides for the eye. For a flow of 0.1 mL/min only 2 data points are shown for CV since data points corresponding to higher concentrations are far off the shown vertical scale; for example, 40 lM of CV gave an area of 5.1.

where h is the angle of incidence (72.88 in our case), k is the wavelength of the light (532 nm), and n1 and n2 are the refractive indices of BK7 (1.52) and the liquid medium (1.33), respectively, and n21 ¼ n2/n1. In the literature on infrared spectroscopy this effective path length is commonly used to compare absorbances in transmission mode with those obtained with an attenuated total reflection (ATR) geometry. For our setup we calculated a de of 403 6 10 nm. The contribution of the bulk absorbance at a CV concentration of 20 lM using this de value is calculated as 4.2 3 105 A.U., assuming that the extinction coefficient of adsorbed CV equals that in bulk solution. This is much lower than the measured values: at a flow rate of 0.5 mL/min, the measured absorbance for the standard polished TIR surface is about 2.5 3 103 A.U. while it is 8 3 104 A.U. for the super-polished TIR surface (see Fig. 2). In fact, the observed absorption in EW-CRDS on the standard polished surface is about 60 times larger than expected based on the CV bulk concentration, indicating a significant enrichment at the surface. For DR10 the bulk absorbance at the detection limit of 100 lM is calculated to be 4 3 105 A.U. assuming that the extinction coefficients of adsorbed and bulk species are the same. The measured absorbance is on the order of 2 3 104 A.U. (data not shown), indicating that the hydrophilic dye DR10 also exhibits adsorption, but to a far lesser extent than CV. To estimate the surface coverage in the case of CV, the relatively large footprint of the laser at the TIR surface (about 7 mm2) should be considered. Assuming a probed layer thickness of roughly 400 nm, the detected volume is about 3 nL and the absolute detection limit is 9 fmol at the concentration limit of detection of 3 lM. Taking into account the enrichment factor of 60, the probed layer actually contains about 0.5 pmol of CV

on the probed surface area. Estimates for the number of available silanol sites range from 0.8 to 8 per square nm,15 which would correspond to 10 to 100 pmol on the probed surface (for the super-polished TIR surface). At this CV concentration, the actual surface coverage will therefore be on the order of 102 relative to the number of silanol groups. Both CV and DR10 are hydrophilic dyes; the former is positively charged above pH 3, whereas the latter is negatively charged above pH 4. At a flow rate of 0.5 mL/min, CV has a response that is about 10 to 15 times higher than that of DR10, indicating that the EW-CRDS detection system has sufficient specificity to discriminate between species with different surface affinities. Whereas the response of the system towards DR10 at a pH of 7.4 is only increased in width upon lowering the flow rate from 0.5 to 0.3 mL/min, the peak area of 20 lM CV is three times higher at 0.3 mL/min. The CV response (peak area) increases roughly quadratically with decreasing flow rates (see Fig. 4, inset). A higher surface specificity is obtained if CV is given enough time to become adsorbed to the surface (see Fig. 5). Whereas for CV the performance of the standard polished TIR surface is better than that of the superpolished surface, the responses of both surfaces to DR10 are comparable. Crystal Violet adsorption isotherms to silica with EW-CRDS have been measured by Shaw et al.;14,15 they used a syringe pump that delivered a very low flow of 4 mL/hour to a flow cell of 190 lL. Rather than using flow injection, they used a constant flow of CV solution at varying concentrations. They used purely aqueous buffer solutions without any modifier, causing the CV molecules to bind even more strongly to the surface, thus necessitating heavy cleaning after each measurement. Using a similar setup, their concentration detection limit

APPLIED SPECTROSCOPY

653

TABLE I. Peak area of 20 lM CV in eluents containing different buffers. Samples were dissolved in the same eluent as used.

Buffer

Area (arb. units)

Standard deviation (n ¼ 6)

10 mM phosphate 10 mM phosphate 10 mM HEPES 100 mM HEPES

0.061 0.025 0.080 0.037

0.0035 0.0054 0.0071 0.0081

was about 0.3 lM of CV for a steady-state measurement, one order of magnitude lower than in our case. This difference should be attributed to a higher degree of surface binding achieved after complete equilibration, which will additionally be higher in water than in 50/50 water/methanol. In order to study the influence of buffer ions (type and concentration), additional experiments have been carried out using 50% of 10 mM HEPES buffer at pH 7.4 in methanol. As expected, the flow-injection profiles for 20 lM CV were similar in shape, and somewhat higher absorbances were obtained than in phosphate buffer. HEPES and phosphate buffers were also tested at a higher concentration: 100 mM in methanol (50%, v/v). The results are summarized in Table I. The data were handled in the same way as for Figs. 3 and 5. It can be seen from the table that an increase in ionic strength leads to a decrease in CV adsorption. Furthermore, the signals as obtained for HEPES buffer are overall higher: this can be explained in a similar way. The ionic strength of a phosphate buffer containing a 10 mM mixture of HPO42 and H2PO4 will be higher than that of a 10 mM mixture of C8H17N2SO4H and C8H17N2SO4.

CONCLUSION In this study it has been demonstrated that EW-CRDS detection utilizing an anti-reflection coated Dove prism inside a linear cavity has distinct potential as a surface-selective detector in flow-injection and LC analysis. Only the analyte molecules present in a roughly 400 nm thick layer at the surface are detected, while the detection limits are, nevertheless, very low: 3 lM for Crystal Violet at a flow rate of 0.5 mL/ min. Contamination or degradation of the surface is minimal, so that repeated measurements can be done over a prolonged time (days) without cleaning or re-alignment. At the applied flow rates the system is far from thermodynamic equilibrium. Nevertheless, the discrimination between the strongly adsorbing CV and the weakly adsorbing DR10 is quite efficient: already at 0.5 mL/min the response of the system towards CV is ten times higher. This discriminating power between nonadsorbing and adsorbing molecules increases significantly upon lowering the flow rates. The EW-CRDS response for CV was 60 times higher than expected if only bulk solution absorbance were to play a role in the absence of any binding to the surface. For DR10 the enrichment factor as a result of binding was much smaller, i.e., five times.

654

Volume 62, Number 6, 2008

It should be noted that the ring-down times in this study were on the order of 5% of that of the empty cavity, mostly due to reflection losses. Therefore, substantial sensitivity improvement should be feasible. Using Brewster’s angles for the entrance and exit faces of a Dove prism would lead to a decrease of only 40% of the ring-down time.19 However, such prisms are not easy to produce. The current setup is capable of reproducible surface-specific adsorption and desorption of the compound of interest at flow rates that are compatible with FIA and LC studies. Future applications of the EW-CRDS setup include specific LC detection or biosensing. Since coating of silanol with, for instance, single-stranded DNA or antibodies is well studied and widespread in bioanalytical chemistry and biological techniques and coating runs are even commercially available, modification of the silica TIR face of the prism surface and surface-specific detection should be feasible. 1. K. L. Bechtel, R. N. Zare, A. A. Kachanov, S. S. Sanders, and B. A. Paldus, Anal. Chem. 77, 1177 (2005). 2. L. van der Sneppen, A. E. Wiskerke, F. Ariese, C. Gooijer, and W. Ubachs, Appl. Spectrosc. 60, 935 (2006). 3. P. Claudon, M. Donner, and J. F. Stoltz, J. Mater. Sci. Mater. Med. 2, 197 (1991). 4. T. Vo-Dinh, M. J. Sepianak, G. D. Griffin, and J. P. Alarie, Immunomethods 3, 85 (1993). 5. S. Rodriguez-Mozaz, M.-P. Marco, M. J. Lopez de Alda, and D. Barcelo, Anal. Bioanal. Chem. 378, 588 (2004). 6. A. C. R. Pipino, J. W. Hudgens, and R. E. Huie, Rev. Sci. Instrum. 68, 2978 (1997). 7. A. C. R. Pipino, J. W. Hudgens, and R. E. Huie, Chem. Phys. Lett. 280, 104 (1997). 8. A. C. R. Pipino, Phys. Rev. Lett. 83, 3093 (1999). 9. A. C. R. Pipino, J. P. M. Hoefnagels, and N. Watanabe, J. Chem. Phys. 120, 2879 (2004). 10. I. M. P. Aarts, A. C. R. Pipino, J. P. M. Hoefnagels, W. M. M. Kessels, and M. C. M. Van de Sanden, Phys. Rev. Lett. 95, 166104 (2005). 11. M. Gupta, H. Jiao, and A. O’Keefe, Opt. Lett. 27, 1878 (2002). 12. R. Brown, I. Kozin, Z. Tong, R. Oleschuk, and H.-P. Loock, J. Chem. Phys. 117, 10444 (2002). 13. P. B. Tarsa, P. Rabinowitz, and K. K. Lehmann, Chem. Phys. Lett. 383, 297 (2004). 14. A. M. Shaw, T. E. Hannon, F. Li, and R. N. Zare, J. Phys. Chem. B 107, 7070 (2003). 15. J. D. Fisk, R. Batten, G. Jones, J. P. O’Reilly, and A. M. Shaw, J. Phys. Chem. B 109, 14475 (2005). 16. W. B. Martin, S. Mirov, D. Martyshkin, and R. Venugopalan, J. Biomed. Opt. 10, 024025 (2005). 17. F. Li and R. N. Zare, J. Phys. Chem. B 109, 3330 (2005). 18. T. E. Hannon, S. Chah, and R. N. Zare, J. Phys. Chem. B 109, 7435 (2005). 19. H.-F. Fan, C.-Y. Hung, and K.-C. Lin, Anal. Chem. 78, 3583 (2006). 20. M. Mazurenka, L. Wilkins, J. V. Macpherson, P. R. Unwin, and S. R. Mackenzie, Anal. Chem. 78, 6833 (2006). 21. M. A. Everest, V. M. Black, A. S. Haehlen, G. A. Haveman, C. J. Kliewer, and H. A. Neill, J. Phys. Chem. B 110, 19461 (2006). 22. S. E. Fiedler, A. Hese, and A. A. Ruth, Chem. Phys. Lett. 371, 284 (2003). 23. H. Naus, I. H. M. van Stokkum, W. Hogervorst, and W. Ubachs, Appl. Opt. 40, 4416 (2001). 24. N. J. Harrick, Infrared Reflection Spectroscopy (Harrick Scientific Corporation, New York, 1987).

Fourier Transform Infrared Measurement of Solid-, Liquid-, and Gas-Phase Samples with a Single Photoacoustic Cell JUHO UOTILA* and JYRKI KAUPPINEN Laboratory of Optics and Spectroscopy, Department of Physics, FI-20014 University of Turku, Finland (J.U., J.K.); and Gasera Ltd., Tykisto¨katu 4, FI-20520 Turku, Finland (J.U.)

A photoacoustic detector based on the optical cantilever microphone has been built. The detector is capable of measuring solid-, liquid-, and gasphase samples. Photoacoustic Fourier transform infrared (FT-IR) measurement with three samples in different phases was demonstrated. Example samples were polyethene, sunflower oil, and methane. The sensitivity of the cell was compared to a commercial photoacoustic FT-IR detector. With the standard carbon black sample the cantilever detector gave approximately five times higher signal-to-noise ratio than the reference detector. The sensitivity with methane was also compared to the DTGS detector of the FT-IR instrument corresponding to an absorption path of 6.3 cm. Simulation of the photoacoustic signal showed that a compromise has to be made in the cell design between sensitivity for solid- and gas-phase samples but it is possible to highly enhance the sensitivity for all types of samples by reducing cantilever dimensions. Index Headings: Photoacoustic spectroscopy; PAS; Fourier transform infrared spectroscopy; FT-IR spectroscopy; Cantilever microphone.

INTRODUCTION The basic idea of photoacoustic spectroscopy (PAS) is to irradiate a sample with modulated electromagnetic radiation that is absorbed by the sample at characteristic wavelengths. The heating of the sample generates a pressure wave whose amplitude is detected with a microphone. Samples can be solids, semisolids, gels, liquids, powders, fibers, gas, or basically anything that converts the radiation into heat at some characteristic wavelengths. The microphone can be coupled to the sample directly or via a coupling medium. A gas-phase coupling medium can be used regardless of the sample. Photoacoustic spectroscopy has been applied to gas analysis for over 60 years and also for several decades to solid and liquid samples.1–4 Numerous publications5,6 regarding Fourier transform infrared (FT-IR) PAS of solid and liquid samples can be found, but only few publications7–10 cover FT-IR PAS of gas samples. In this paper we introduce a photoacoustic cell (Fig. 1) based on a cantilever microphone that is capable of measuring samples in all three phases with relatively good sensitivity. The advantages of photoacoustic spectroscopy compared to other FT-IR techniques are different in gas-phase detection than in analyses of solids or liquids. The most wellknown and important advantages with solids are the minimal sample preparation that is required, suitability for opaque materials, possibility for depth profiling, and nondestructive measurement, which means that the sample is not consumed. With gas samples the advantages are low sample volume, short absorption path length, and direct absorption measurement, which allow sensitive and selective measurements in the Received 30 January 2008; accepted 24 March 2008. * Author to whom correspondence should be sent. E-mail: juho.uotila@ utu.fi.

Volume 62, Number 6, 2008

presence of high amounts of interfering substances such as water vapor. A cantilever microphone11–14 has been used in gas-phase PAS in order to achieve a good signal-to-noise ratio (SNR) even with low absorbing samples. This is often the case, especially with low concentration gas samples. The cantilever is microfabricated of silicon and is coated with aluminum or gold.15 The cantilever is bent by the pressure difference between its two sides, and the displacement of the cantilever end is measured optically with a Michelson type laser interferometer. The drawing of the cantilever microphone and dimensions of the cantilever used in this study are shown in Fig. 2. A detailed description of the interferometric measurement of the displacement of the cantilever has been presented elsewhere.12 Other advantages of the cantilever microphone over capacitive or electret microphones besides sensitivity are the possibility for introducing a gas flow through the gap between the cantilever and its frame without breaking the microphone, the possibility for evacuating and heating of the cell, good temperature stability of the cantilever, and extremely high dynamical range.

SIGNAL GENERATION Liquid- and Solid-Phase Samples. In solids and liquids the principle of signal generation is similar.6 Modulated radiation is absorbed, causing periodic heating in the sample. This causes pressure variations to propagate in all directions, and a superposition of these acoustic waves at the sample surface generates a surface motion that is coupled to the surrounding gas (acoustic coupling). The periodic heat flow to the gas from the surface also generates expansion and contraction in a thin layer of gas close to the surface (thermal coupling).16 The signal detected in the gas by the microphone is the result of a combination of these two mechanisms (the composite piston model).17 In a typical solid-state PAS experiment thermal coupling is the dominant mechanism and acoustic coupling can be neglected. However, with some liquid samples acoustic coupling might even dominate the photoacoustic signal. For thermally thick samples, assuming one-dimensional heat flow, the complex pressure signal amplitude according to the composite piston model17 can be given as j cP0 I0 ðmÞ x lg 2qs Cps   aðmÞ þ bT f1  exp½aðmÞls g 3 rg T0 ðg þ 1Þðr þ 1Þ

pðxÞ ¼ 

ð1Þ where j is an imaginary unit, x is the angular frequency of the modulation, lg and c are the thickness and ratio of specific

0003-7028/08/6206-0655$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

655

FIG. 1. The photoacoustic cell with a sample holder for the gas sample and for solid and liquid samples.

heats of the gas, P0 and T0 are the gas pressure and temperature, I0 is the incident radiation intensity at wavenumber m, qs and Cps are the density and heat capacity of the sample, a is the absorption coefficient of the sample, ls is the thickness of the sample, bT is the (volume) coefficient for thermal expansion, g ¼ (jgrg)/(jsrs), rg ¼ (jx/Dg)1/2, rs ¼ (jx/ Ds)½, r ¼ a/rs, Dg and Ds are the thermal diffusivities of the gas and sample, and jg and js are the thermal conductivities of the gas and sample. The first term in the parentheses on the right-hand side of Eq. 1 describes thermal coupling and the second term, acoustic coupling. The composite piston model assumes the signal generation in the coupling gas to be an adiabatic process. The cell and cantilever frequency responses should be added to Eq. 1 in order to model the microphone output signal. In the frequency range (100 Hz to 1000 Hz) that is typical in infrared photoacoustic spectroscopy, the amplitude of the cantilever end movement dx is proportional to the

pressure variation dp as14 dx ¼

Ac Ac dp ¼ dp k k0 þ cP0 A2c =2:5V

where Ac is the area of the cantilever and V is the volume of the sample cell. The effective spring constant k is the sum of the spring constants of the cantilever and oscillating gas. The cantilever spring constant k0 depends on the cantilever length l, width w, thickness t, and Young’s modulus E of silicon as 2  t 3 ð3Þ k0 ¼ wE 3 l Thermal waves that are coupled with gas have short decay length L ¼ (Ds /x)1/2, which describes the depth from the sample surface that affects the photoacoustic signal. The majority of the thermal wave is reflected back to the sample from the surface without generating signal. With high

FIG. 2. Drawing of the cantilever microphone and cantilever dimensions used in this study.

656

Volume 62, Number 6, 2008

ð2Þ

TABLE I.

Measurement parameters. Parameter

Value

Number of scans Resolution due to drive length [cm1] Resolution due to source aperture [cm1] Laser frequency [kHz] OPD mirror velocity [cm/s] Measurement time [s] Pressure in PA cell [mbar]

100 8 4 at 3095 1.6 0.09 168 1005

modulation frequencies, the finite propagation time of the thermal wave can also decrease the photoacoustic signal due to the phase lag between modulated radiation and the signal. Normalization and linearization can be performed on the photoacoustic spectrum. Normalization compensates the effects of the interferometer and photoacoustic cell responses. It is performed by simply dividing the measured spectrum by a reference background spectrum, which is measured, e.g., using a carbon black sample. The linearization corrects the line shape distortions due to the exponential behavior of absorption in the sample. One way to do this is to use the equation18 Sl ¼

S2R þ S2I 1:414ðSI RR  SR RI Þ

ð4Þ

where Sl is the linearized spectrum, SR and SI are the real and imaginary components of the sample spectrum, and RR and RI are the real and imaginary components of the reference spectrum. Linearization is important when general spectral libraries are used or quantitative analyses are made. In order to optimize the signal for solid samples the cell should be low in volume and all radiation focused on to the sample. Also, the infrared radiation absorption path in the gas should be short. If the atmosphere in the cell is helium the signal is from two to three times higher than with nitrogen or air. Due to the damping of the gas spring, the volume of the photoacoustic cell has to be optimized for the different cantilever dimensions. Lower volumes require smaller cantilevers. Gas-Phase Samples. The absorption in a gas-phase sample is directly transformed into the heat and pressure wave that couples with the cantilever. In contrast to the piston model with solid samples, the process with gas-phase samples is assumed to be isochoric. The photoacoustic signal in the cantileverenhanced photoacoustic cell can be given as14 Ac ðc  1Þ ax ðmÞpx lI0 ðmÞ s1 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi SðmÞ ¼ V m ðx20  x2 Þ2 þ ðxcD =mÞ2 1 þ ðxs1 Þ2 xs23 1 3 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 1 þ ðxs4 Þ2 1 þ ðxs23 Þ

ð5Þ

where Ac is the area of the cantilever, V is the volume of the cell, px is the partial pressure of the gas of interest, l is the absorption path length, m is the effective mass of oscillating elements (cantilever and gas), x0 is the resonance angular frequency of the cantilever inside the cell, cD is the damping factor, and s1, s23, and s4 are the time constants due to the heating of the sample, gas flow through the cantilever frame gap, and relaxation from absorption to heat, respectively. Since

FIG. 3. The photoacoustic spectrum of polyethene.

the radiation is generated in the FT-IR interferometer, the wavenumber m is proportional to the angular velocity x as m ¼ x/OPV, where OPV is the velocity of the interferometer scanning mirror. If the absorption is strong due to the high absorption coefficient or long absorption path, the approximation (apxl) ’ 1  exp(apxl) is no longer valid and Beer’s law has to be taken into account. The absorbance spectrum A can be calculated as   SðmÞ AðmÞ ¼ ln 1  ð6Þ Sr ðmÞ where Sr is the reference spectrum. Ideally, the measurement of the reference spectrum should be done by using a gas that absorbs all wavelengths equally. Since there is not such a gas, a carbon black spectrum must be used as an approximation for a reference. Another way to normalize the spectrum is to measure simultaneously the transmission spectrum ST(m) with an infrared detector in the beam exit of the photoacoustic cell. The ratio of the PA signal and reference signal would give the following result: SðmÞ RPAS ðmÞ I0 ðmÞ½1  eaðmÞpx l  ¼ xl ST ðmÞ RIR ðmÞ I0 ðmÞeð1=2ÞaðmÞp   RPAS ðmÞ 1  2  sinh aðmÞpx l ¼ RIR ðmÞ 2

ð7Þ

Since the frequency responses of the photoacoustic cell RPAS(m) and infrared detector RIR(m) are well known, the absorbance A ¼ a(m)pxl can be solved. The cell design for gas samples requires long absorption path in the gas compared to the cell volume. Optimal design could be a circular multipass cell.7

EXPERIMENTS Measurement Setup. The photoacoustic cell (Fig. 1) consists of the sample holder, sample cell, cantilever, and balance cell. Depending on the sample phase, the holder can be selected. The holder is a pipe for gas samples, 7 mm long and 4 mm in diameter, with a mirror at the pipe end. Solid and liquid

APPLIED SPECTROSCOPY

657

FIG. 4. The photoacoustic spectrum of sunflower oil.

FIG. 6. Normalized spectrum of polyethene.

samples are held on a small plate at a distance of 2.5 mm to 1 mm from the window. The center of mass of the sample and balance cell is set at the same side of the cantilever to compensate the noise due to the movements of the cell.14 The gas valves are connected to the balance cell so that during the sample change the cell can be purged through the cantilever frame gap. Helium was used as a purging gas in the experiments presented in this paper. The gas samples were changed by first evacuating the cell and then letting the sample mixture flow into the cell. All gas samples were mixed with nitrogen. The photoacoustic detector was connected to a Mattson Galaxy 6020 series FT-IR interferometer. The collimated output beam from the interferometer was focused to the beam guide pipe with an off-axis parabolic mirror. The spectra were acquired as an average of 100 single-beam measurements. Measurement parameters are shown in Table I. Reference PAS measurements were made with a commercial photoacoustic detector MTEC model 300.19 Reference measurements were also made with the DTGS detector of the FT-IR interferometer.

Liquid- and Solid-Phase Samples. The measurements were demonstrated with a polyethene chip (Fig. 3) and a drop of sunflower oil (Fig. 4). Strong absorptions from –C–H (CH2) stretching, –C–H (CH2, CH3) bending, –(CH2)n– bending, and also very weak absorptions, e.g., from –HC¼CH– bending out of plane, –C¼C– (cis-) stretching, and –C¼O (ester) Fermi resonance vibration modes, are clearly visible in the spectra.20 A standard carbon black film was used for a background signal measurement by setting it instead of a mirror in the 7 mm long pipe end. The background single beam spectra of the cantilever PAS, the reference PAS, and the DTGS detectors are shown in Fig. 5. In the background spectra the water and carbon dioxide absorptions from the air in front of the detector are visible. The signal decrease at low wavenumbers with the cantilever PAS is due to the BaF2 window. The reference PAS detector had a KBr window and the DTGS had a CSi window. The polyethene chip was black in color and contained material that absorbed on a broad spectrum, generating a strong background signal (Fig. 3). In the normalized spectrum it can be seen as an offset level (Fig. 6).

FIG. 5. Single-beam background spectra of cantilever PAS, reference PAS, and DTGS detectors.

FIG. 7. Measured and difference spectra of 100 ppm methane and water vapor.

658

Volume 62, Number 6, 2008

TABLE II. SNR comparison of the background signal and 100 ppm methane signal with different detectors. Signal-to-noise ratio Detector

Normalized background

CH4, 100 ppm

Cantilever PAS Reference PAS DTGS

3480 679 4683

36.2 6.10 57.8

PAS and DTGS detectors. The measured methane spectra are shown in Fig. 9. A copper mirror was set in the bottom of the reference PAS sample cell and a 10 cm long absorption cell was used with the DTGS detector.

RESULTS AND DISCUSSIONS

FIG. 8. Normalized photoacoustic spectrum of 100 ppm methane and modeled absorbance spectrum using HITRAN database.

Gas-Phase Samples. The gas-phase measurement with methane was also demonstrated. The certified sample mixture was 100 ppm methane in nitrogen. In reality, the sample also contained water vapor. The measured spectrum is shown in Fig. 7. The difference spectrum of water and methane mixtures was calculated, and the spectrum was normalized using Eq. 6 and the carbon black reference spectrum. The normalized difference spectrum and the simulation using the HITRAN21 database are shown in Fig. 8. The simulation gave four times lower absorbances than the normalized spectrum. This suggests that the solid reference sample gave too low a signal compared to the gas-phase measurement. The peak height at 3012 cm1 and the 2r root mean square (rms) noise in the range from 2700 cm1 to 2800 cm1 gives a SNR of 36 for 100 ppm methane. This would correspond to a detection limit of 3 ppm. The same sample mixture was also measured with the reference

FIG. 9. Measured spectra of 100 ppm methane with the cantilever PAS, reference PAS, and DTGS (10 cm path length) detectors. An offset level has been added to the cantilever PAS and DTGS spectrum.

The photoacoustic detector with the cantilever microphone was compared to the reference PAS and DTGS detectors. The signal-to-noise ratio (SNR) with different detectors for the normalized background signal and for 100 ppm methane is presented in Table II. The normalized background spectrum was calculated as a ratio of two 10-scan background spectra for each detector (Fig. 10). The SNR was calculated from the 2r rms noise at 2000 cm1. The cantilever PAS detector has 5.1 times higher SNR with a standard carbon black sample than the reference PAS detector. This corresponds to the sensitivity ratio with solidand liquid-phase samples. For the methane sample the SNR was calculated from the peak height at 3012 cm1 and 2r rms noise at 2700 cm1 to 2800 cm1. With gas-phase samples the ratio between the sensitivities of the cantilever and reference PAS detector is 5.9. The ratio between sensitivities of the cantilever PAS and DTGS with 10 cm absorption path is 0.63. When the sensitivity of the cantilever PAS detector is transformed to the absorption cell length, the corresponding absorption path would be 6.3 cm with the DTGS detector of the FT-IR interferometer. The enhanced sensitivity compared to the reference PAS detector means that without a helium purge the cantilever PAS

FIG. 10. Normalized background spectra of the cantilever PAS, reference PAS, and DTGS detectors.

APPLIED SPECTROSCOPY

659

based on the cantilever microphone has been demonstrated. In comparison to a commercial device the cantilever cell proved to be approximately five times more sensitive. The gas-phase measurement sensitivity corresponded to a 6.3 cm absorption path with the DTGS detector of the FT-IR device. In the cell design a compromise between sensitivity for solid- and gasphase samples has to be made. Smaller cantilevers should be used in order to make significant enhancements to the sensitivity with all types of samples.

FIG. 11. Relative values of signal and optimization parameters for solid- and gas-phase samples as a function of sample cell length.

detector would still give better sensitivity than the reference detector with helium, or the measurement time could be 26 times shorter for the same sensitivity. The cantilever also allows reasonably good sensitivity for gas-phase measurements even if the cell is designed for a different type of measurement. The optimization of the photoacoustic detector can be done by changing the cantilever dimensions or the photoacoustic cell dimensions. The cell dimensions were close to optimal for solid-phase samples with the cantilever used. As an example of the cantilever dimension variation, the signal with cantilever dimensions of l ¼ 1 mm, w ¼ 0.5 mm, and t ¼ 1 lm would be approximately 10 times higher for both solid (Eqs. 1–3) and gas samples (Eq. 4). The optimization of the cell dimensions is different for gas- and solid-phase samples (Fig. 11). The cantilever sensitivity (Eq. 2) is best with high volumes due to the smaller gas spring constant, but the photoacoustic pressure variation is inversely proportional to the cell volume (Eqs. 1 and 5). With solid samples the PA signal is dominated by the cell volume even if the gas spring constant is high. With gasphase samples the longer absorption path compensates the effect on the photoacoustic signal due to the higher volume (Eq. 5) and the signal strength follows the cantilever sensitivity. In Fig. 11 the photoacoustic signal for solid- and gas-phase samples is modeled as a function of sample cell length when the cantilever dimensions and cell diameter are fixed.

CONCLUSION Fourier transform infrared measurement of solid-, liquid-, and gas-phase samples with the single photoacoustic detector

660

Volume 62, Number 6, 2008

1. F. J. M. Harren, G. Cotti, J. Oomens, and S. Lintel Hekkert, ‘‘Photoacoustic Spectroscopy in Trace Gas Monitoring’’, in Encyclopedia of Analytical Chemistry, R. A. Meyers, Ed. (John Wiley and Sons, Chichester, 2000), p. 2203. 2. K. F. Luft, Z. Technol. Phys. 5, 97 (1943). 3. L. B. Kreuzer, J. Appl. Phys. 42, 2934 (1971). 4. A. Rosencwaig, ‘‘Photoacoustics and Photoacoustic Spectroscopy’’, in Chemical Analysis, A Series of Monographs on analytical Chemistry and Its Applications, P. J. Elving and J. D. Winefordner, Eds. (John Wiley and Sons, New York, 1980), vol. 57. 5. K. H. Michaelian, ‘‘Photoacoustic Infrared Spectroscopy’’, in Chemical Analysis, A Series of Monographs on analytical Chemistry and Its Applications, J. D. Winefordner, Ed. (John Wiley and Sons, Hoboken, NJ, 2003), vol. 159. 6. J. F. McClelland, R. W. Jones, and S. J. Bajic, ‘‘Photoacoustic Spectroscopy’’, in Handbook of Vibrational Spectroscopy, J. M. Chalmers and P. R. Griffiths, Eds. (John Wiley and Sons, London, 2002), vol. 2, p. 1231. 7. A. Olafsson, G. I. Hansen, A. S. Loftsdottir, and S. Jakobsson, ‘‘FT-IR photoacoustic trace gas detection’’, in Photoacoustic and Photothermal Phenomena: 10th Interntional Conference, F. Scudieri and M. Bertolotti, Eds. (1999), p. 208. 8. R. S. Wright, G. B. Howe, and R. K. M. Jayanty, J. Air Waste Manage. Assoc. 48, 1077 (1998). 9. G. Busse and B. Bullemer, Infrared Phys. 18, 631 (1978). 10. D. P. Baldwin, R. W. Jones, and J. F. McClelland, ‘‘Exploration of FTIRBased PAS for On-Site Analysis of Volatile Contaminants in Air’’, in Photoacoustic and Photothermal Phenomena III, D. Bic´anic´, Ed. (Springer Verlag, Heidelberg, 1992), p. 3. 11. K. Wilcken and J. Kauppinen, Appl. Spectrosc. 57, 1087 (2003). 12. J. Kauppinen, K. Wilcken, I. Kauppinen, and V. Koskinen, Microchem. J. 76, 151 (2004). 13. V. Koskinen, J. Fonsen, J. Kauppinen, and I. Kauppinen, Vib. Spectrosc. 42, 239 (2006). 14. T. Kuusela and J. Kauppinen, Appl. Spectrosc. Rev. 42, 443 (2007). 15. P. Sievila¨, V.-P. Rytko¨nen, O. Hahtela, N. Chekurov, J. Kauppinen, and I. Tittonen, J. Microcmech. Microeng. 17, 852 (2007). 16. A. Rosencwaig and A. Gersho, J. Appl. Phys. 47, 64 (1975). 17. F. A. McDonald and G. C. Wetsel, J. Appl. Phys. 49, 2313 (1978). 18. R. O. Carter, Appl. Spectrosc. 46, 219 (1992). 19. MTEC Photoacoustics Inc., MTEC model 300 photoacoustic cell Instrument manual, www.mtecpas.com (2005). 20. M. D. Guillen and N. Cabo, J. Am. Oil Chem. Soc. 74, 1281 (1997). 21. L. S. Rothman, D. Jacquemart, A. Barbe, D. Chris Benner, M. Birk, L. R. Brown, M. R. Carleer, C. Chackerian, Jr., K. Chance, L. H. Coudert, V. Dana, V. M. Devi, J.-M. Flaud, R.R. Gamache, A. Goldman, J.-M. Hartmann, K. W. Jucks, A. G. Maki, J.-Y. Mandin, S. T. Massie, J. Orphal, A. Perrin, C. P. Rinsland, M. A. H. Smith, J. Tennyson, R. N. Tolchenov, R. A. Toth, J. Vander Auwera, P. Varanasi, and G. Wagner, J. Quant. Spectrosc. Radiat. Trans. 96, 139 (2005).

Novel Search Algorithms for a Mid-Infrared Spectral Library of Cotton Contaminants J. BRIAN LOUDERMILK, DAVID S. HIMMELSBACH, FRANKLIN E. BARTON II, and JAMES A. DE HASETH* Department of Chemistry, The University of Georgia, Athens, Georgia 30602-2556; and USDA, ARS, Richard B. Russell Research Center, Athens, Georgia 30604-5677

During harvest, a variety of plant based contaminants are collected along with cotton lint. The USDA previously created a mid-infrared, attenuated total reflection (ATR), Fourier transform infrared (FT-IR) spectral library of cotton contaminants for contaminant identification as the contaminants have negative impacts on yarn quality. This library has shown impressive identification rates for extremely similar cellulose based contaminants in cases where the library was representative of the samples searched. When spectra of contaminant samples from crops grown in different geographic locations, seasons, and conditions and measured with a different spectrometer and accessories were searched, identification rates for standard search algorithms decreased significantly. Six standard algorithms were examined: dot product, correlation, sum of absolute values of differences, sum of the square root of the absolute values of differences, sum of absolute values of differences of derivatives, and sum of squared differences of derivatives. Four categories of contaminants derived from cotton plants were considered: leaf, stem, seed coat, and hull. Experiments revealed that the performance of the standard search algorithms depended upon the category of sample being searched and that different algorithms provided complementary information about sample identity. These results indicated that choosing a single standard algorithm to search the library was not possible. Three voting scheme algorithms based on result frequency, result rank, category frequency, or a combination of these factors for the results returned by the standard algorithms were developed and tested for their capability to overcome the unpredictability of the standard algorithms’ performances. The group voting scheme search was based on the number of spectra from each category of samples represented in the library returned in the top ten results of the standard algorithms. This group algorithm was able to identify correctly as many test spectra as the best standard algorithm without relying on human choice to select a standard algorithm to perform the searches. Index Headings: Spectral search system; Search algorithm; Voting scheme algorithm; Spectral library; Infrared; Fourier transform infrared spectroscopy; FT-IR spectroscopy; Attenuated total reflection; ATR; Chemometrics; Cotton; Contaminants.

INTRODUCTION When cotton is mechanically harvested from the field, it can be contaminated by a variety of foreign matter. In addition to the desired cotton fiber, parts of the plants such as leaves, stems, seeds, and hulls are also collected. Parts of other plants growing in the field, inorganic matter such as clay or sand, plastic waste, and greases and oils from machinery can also contaminate the cotton, but the majority of contamination comes from the cotton plants themselves.1 These contaminants have a major impact on the quality of cotton yarn produced and, ultimately, on the profitability of processing. Contaminants are responsible for an increased number of yarn Received 14 August 2007; accepted 26 February 2008. * Author to whom correspondence should be sent. E-mail: dehaseth@uga. edu.

Volume 62, Number 6, 2008

breakages during spinning and lower the quality and price of the final yarn product.2,3 Because of the negative impact of these contaminants on quality and profit, their detection and removal is important. Knowledge of which contaminants are most abundant in the cotton and which contaminants create the most difficulties during processing allows the entire process from harvesting through production of the final product to be streamlined to limit those contaminants, but the identification of contaminants is not straightforward. From the time organic contaminants are first harvested along with the cotton, they begin to break down in physical size, making human visual identification difficult or impossible.3 Some attempts to identify contaminant particles by color4 or geometric features5 have been made in the past. These methods did not make use of the chemical information available through spectroscopic analysis of trash components. Over several years, the United States Department of Agriculture—Agricultural Research Service (USDA-ARS) has developed a mid-infrared, Fourier transform infrared (FT-IR), attenuated total reflection (ATR) spectral library of cotton contaminants.1,3 This library allows one to employ the power of molecular fingerprinting, inherent to the mid-infrared region of the spectrum, to be applied to the problem of cotton contaminant identification. Work by Himmelsbach et al.1,3 has shown the utility of the library and the high percentage of correct matches returned by searching when the library spectra are representative of the unknown samples. It should be noted that identification of cotton plant parts by library searching is significantly more difficult than standard searches for a diverse library, such as most commercial libraries. In the current study, all the samples have both spectra and chemical compositions that are extremely similar. Standard algorithms have been designed to distinguish between relatively different library entries. The current work demonstrates that when spectra of plant parts grown in different seasons and geographic locations and measured with a different spectrometer and ATR accessories from those represented in the library are searched, the identification rate drops significantly compared to searching spectra that have representatives from the same growing regions and seasons and are measured under the same conditions. The performance of a variety of standard searching algorithms was investigated, and it was found that identification depends on the type of plant part spectrum being searched. Additionally, different algorithms were found to give complementary information about sample identity instead of completely repetitive information. In other words, each algorithm may return correct answers that are not returned by every other algorithm. These results meant that it was impossible to pick a single standard algorithm that would yield the best matches from the library for every category of contaminant considered.

0003-7028/08/6206-0661$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

661

This work focuses on the creation of voting scheme algorithms that overcome the disadvantages of the standard algorithms and improve identification.

BACKGROUND A thorough history of the development of infrared spectral library search systems through the mid 1980s has been reported by Lowry et al.6 The first automated spectral library search systems based on punch cards and electric sorters were reported in the 1950s.7,8 These systems relied on encoded cards to record the location of spectral absorption bands. Electric sorters were then used to locate cards that represented spectra with the same bands as the unknown spectrum searched. These early systems even made use of the absence as well as the presence of bands to match an unknown spectrum to the most similar spectra in a library. In the 1960s, the computerized ASTM binary encoded spectral library came into use.9–11 Because this system allowed for computerized searching of electronically stored information, the time required to search through a given number of spectra decreased, but only band location, and not band intensity, information was still used to match spectra. In the 1970s, Azaraga12 introduced the EPA vapor-phase library, which was the first major digitized spectral library. With the introduction of this library, scientists began to use Euclidean distance as a metric to determine how closely related spectra in a library were to an unknown spectrum. Into the 1980s, work continued with correlation algorithms13,14 and variations on the Euclidean distance metric, such as metrics based on the differences between the derivatives of spectra,6 but in practice there has been little change in infrared spectral library searching since that time.

MATERIALS AND METHODS USDA Attenuated Total Reflection Fourier Transform Infrared Spectral Library. The USDA library contains 929 FT-IR spectra measured with a Nicolet Magna 850 FT-IR spectrometer (Thermo Fisher Scientific, Waltham, MA) and a DuraScope ATR sampling accessory (Smiths Detection, Danbury, CT). The spectrometer contained a ceramic source, a KBr beamsplitter, and a deuterated triglycine sulfate (DTGS) detector. The ATR accessory had a diamond-coated ZnSe internal reflection element (IRE). Spectra were measured over the range of 4000 to 650 cm1 at 8 cm1 resolution, with 128 interferograms co-added, and interferograms were processed with Happ–Genzel apodization prior to Fourier transformation. Spectra were collected with the use of Omnic E.S.P. 5.2 software (Thermo Fisher Scientific, Waltham, MA). All spectra were first converted to GRAMS format (Thermo Fisher Scientific, Waltham, MA) and then imported into MATLAB (The Math Works, Natick, MA). The library comprises the spectra of foreign materials found in harvested cotton and includes the following types of samples: cotton plant parts including bloom, bract, hull, leaf, seed coat, shale, and stem; other organic matter such as poultry feathers, cow leather, and weed parts; inorganic materials such as sand and clay; greases and oils; and plastics. Organic samples were representative of several geographic locations and growing conditions. Test Set Spectra. A set of 75 test spectra of samples from several geographic locations and seasons different from the samples in the USDA library were measured with the use of a

662

Volume 62, Number 6, 2008

Varian Excalibur Series FTS-4000 FT-IR spectrometer (Varian, Palo Alto, CA) and three different ATR sampling accessories. The samples were obtained from the USDAARS Cotton Quality Research Laboratory (Clemson, SC). The spectrometer contained a ceramic source, a KBr beamsplitter, and a DTGS detector. The ATR accessories were the Specac Golden Gate (Specac, Woodstock, GA) with a diamond-coated ZnSe IRE and the Harrick SplitPea and Seagull (Harrick Scientific, Pleasantville, NY) with Si and ZnSe IREs, respectively. Spectra were measured over the range of 4000 to 400 cm1 at 4 cm1 resolution, with 256 interferograms coadded. Interferograms were processed with Happ–Genzel apodization to be consistent with the USDA library. Each spectrum in the test set was the average of three replicate spectra. Spectra were collected with the use of Varian Resolutions Pro 4.0.5.009 and WinIR Pro 3.2 software (Varian, Palo Alto, CA). All spectra were first converted to GRAMS format and then imported into MATLAB. The test set samples contained hull, leaf, seed coat, and stem samples, both intact and powdered, from nine different growing locations. Standard Search Algorithms. Six standard search algorithms were used in the experiments: dot product, correlation, sum of the absolute values of differences, square root of the sum of the absolute values of differences, sum of the absolute values of differences of the derivatives, and sum of the squared differences of the derivatives. Each equation below can be thought of as a metric for the comparison of two spectra represented as vectors. The vector ~ Ij represents the jth spectrum in the library, and the vector ~ u represents the spectrum searched against the library. The subscript i represents the ith element of a vector, and n equals the number of elements in each vector or the number of resolution elements in each spectrum. The score describing how closely related the spectrum being searched is to the jth spectrum in the library is represented by Sj. The dot product metric is given by Eq. 1: Ij  ~ uj Sj ¼ ~

ð1Þ

It should be noted that the dot product metric gives the same comparative information as an algorithm based on the sum of the squares of the differences between spectral resolution elements (often called a least squares algorithm in the literature), but the dot product metric requires a much smaller number of computations. For spectra vectors normalized to unit magnitude, the dot product metric is equivalent to the cosine of the angle between the vectors. A perfect match has a score of 1 and orthogonal spectra have a score of 0. Equation 2 is the sum of the absolute values of the differences algorithm, and Eq. 3 is the sum of the square root of the absolute values of the differences: Sj ¼

n  X  ~ Iji  ~ uji 

ð2Þ

i¼1

Sj ¼

n  X 1=2 ~ Iji  ~ uji 

ð3Þ

i¼1

These first three metrics differ in the emphasis placed on small versus large differences between the spectra compared: from Eq. 1 to Eq. 2 to Eq. 3 more emphasis is placed on small differences between the spectra versus large differences. Equations 4 and 5 represent the sum of the absolute values

of the differences of the derivatives and the sum of the squared differences of the derivatives metrics: Sj ¼

n h   X i ~ uj;iþ1  ~ Ij;iþ1  ~ Iji  ~ uji

ð4Þ

i¼1

Sj ¼

n h   X i 2 ~ Ij;iþ1  ~ Iji  ~ uji uj;iþ1  ~

ð5Þ

i¼1

These derivative metrics are especially useful for the comparison of spectra with varying baseline shifts. For the metrics represented by Eqs. 2–5, the smaller the score between the unknown and library spectra, the more spectrally related the unknown spectrum is to the library spectrum, and a perfect match has a score of 0. Equation 6 is for the correlation metric: ! n n X X ~ ~ Iji uji n X i¼1 i¼1 ~ Iji~ uji  n i¼1 Sj n ¼ 8 2 !2 32 !2 391=2 n n X X > > > > ~ > ~ Iji 76 n uji 7> > > = <6 n X X 7 7 6 6 2 i¼1 i¼1 2 7 7 6 ~ 6 ~ I   u ji ji n n 76 7> 6 > > 54 i¼1 5> 4 i¼1 > > > > ; : ð6Þ For the correlation metric, a perfect match has a value of 1, and values of the score decrease as the similarity of the spectra decreases. In order to find the spectrum in a library that most closely matches the spectrum that is searched for a given algorithm, the score, Sj, for each spectrum in the library is calculated. The scores for a given standard algorithm are then ordered so that the spectra in the library are ranked by their similarity to the spectrum being searched. The scores returned by different standard search algorithms are not directly comparable, but the ranks of the results returned by each algorithm are. The voting scheme algorithms described in the next section are a method of incorporating information from all of the standard search algorithms described for a given spectrum searched against the library. Voting Scheme Algorithms. Three voting scheme algorithms were developed. Each algorithm combined the top 10 ranked results from the library spectra returned by each standard algorithm and then ranked these 60 matches according to three different criteria. The weighted frequency algorithm assigned a weight to each of the 60 matches. Any match that was ranked first by a standard algorithm received a weight of 10. Any match ranked second by a standard algorithm received a weight of 9. This pattern continued until the matches given a rank of 10 by each of the standard algorithms were given a weight of 1. All of the weights associated with a given spectrum from the library were summed, and all of the spectra were then ranked by their associated sum of weights from highest to lowest. The highest sums were considered to be the best matches. This algorithm considered both the ranks of a given spectrum in the results of the standard algorithms and the number of standard algorithms that matched the spectrum. The frequency algorithm ranked spectra solely on the number of standard algorithms that returned the given

spectrum. This search would have been equivalent to a weighted frequency search where all 60 spectra were assigned a weight of 1. The group algorithm counted the number of matches out of 60 that represented sample categories found in the library. The spectra in the library were assigned to belong to 1 of 21 possible groups. Group divisions included leaf, stem, seed coat, and hull. The number of matches that belonged to a particular group counted as the number of votes for a particular group. The group with the highest number of votes was considered to be the best match for the test spectrum that was searched. Library Searching. Before searching, all spectra were truncated so that they ranged from 3700 to 2700 cm1 and 1800 to 650 cm1. The region between 2700 to 1800 cm1 was removed because none of the library or test spectra contained significant absorption bands in this region. The spectra in the test set were also de-resolved to 8 cm1 resolution to match the library spectra. After these preprocessing steps, all spectra contained 558 resolution elements. All spectra were also given a common minimum by subtraction of the lowest absorbance value in each spectrum from the absorbance values for every resolution element in that spectrum. All spectra were vector normalized to unit magnitude. A 20 member subset of the test spectra set was chosen to test the performance of standard spectral library search algorithms. This subset was composed of five spectra each of hull, leaf, seed coat, and stem spectra. Spectra of powdered samples (80 mesh) accounted for two or three spectra out of each group of five spectra. These spectra were searched against the library with the use of the six standard search algorithms, and the top 10 results returned by each algorithm for each spectrum in the subset were recorded. The combined top 10 results returned for each test spectrum by the standard algorithms were then searched with the use of the three voting scheme algorithms. A second subset of 12 spectra from the 75 test spectra set and 12 spectra from the library were chosen to test the performance of the standard algorithms when test spectra were searched against a library that had been augmented by combining the 75 spectra test set into the USDA library. This 24 spectra test set contained six spectra each of hull, leaf, seed coat, and stem. The spectra for each category comprised three spectra from the original 929 spectra library and three spectra from the 75 member test set. When each spectrum was searched against the augmented library, all spectra in the library that were replicates of the spectrum to be searched were removed from the library before the search was executed. All algorithms were programmed in the MATLAB programming language and executed with MATLAB 7 software. The performances of the standard algorithms were judged by the number of rank one results returned that correctly identified the test samples and by the total number of correct matches in the top ten results returned for the test samples (e.g., if a leaf is being searched against the library, any leaf spectrum that is returned will be considered a correct result). The performances of the voting scheme algorithms were compared to the standard algorithms by the number of correct rank one results returned.

RESULTS AND DISCUSSION Distinguishing among the infrared spectra of plant based contaminants found in cotton is not a trivial task. Figure 1 shows the spectra of four cotton contaminant samples: a leaf, stem, seed coat, and hull. Each of these materials has cellulose

APPLIED SPECTROSCOPY

663

FIG. 1. Spectra of leaf, stem, seed coat, and hull from the cotton plant. The spectra have been offset for clarity.

as its main component, and thus, they all have very similar spectra. The absorbance band locations in these spectra are nearly identical from one spectrum to the next; however, the band intensities do vary. If all stems from all cotton plants had the same spectrum and if all hulls from all cotton plants had the same spectrum but the hull spectrum was different from the stem spectrum, then one would be able to distinguish the spectrum of a stem contaminant from the spectrum of a hull contaminant by the differences in band intensities between the two spectra. Unfortunately, this is not the case. Because of the differing concentrations of components that make up parts from different plants and multiple parts from the same plant, relatively large variations exist among the spectra of a given type of plant part when compared to the similarity of the spectra of different types of plant parts seen in Fig. 1. In other words, the within group variation is significant compared to the between group variation. Figure 2 demonstrates this point for a group of leaf spectra. This figure shows the spectra of four different cotton leaves. Despite these challenges, previous research has shown that more than 90% of organic based contaminant samples can be identified with the use of the USDA cotton contaminant library and standard spectral search algorithms when the spectra in the library are representative of the spectra being searched and the unknown spectra are acquired with the same instrument used to measure the library spectra.1 Representative means that spectra of cultivars from the same geographic region and time period and grown under similar conditions as the sample represented by the unknown spectrum being searched must be contained in the library. In this work, spectra of samples grown in different geographic regions and during different time periods than the library spectra were used to test the performance of standard algorithms. Additionally, the test spectra were measured with a

664

Volume 62, Number 6, 2008

different instrument and different ATR accessories than the library spectra. It is well known that transferring spectral calibrations from one spectrometer to another can be difficult because of instrumental differences, and these instrumental differences also affect the performance of library searches. For true versatility, a spectral library needs to provide consistent results despite these sources of variation, and the experiments conducted in this study were designed to check the performance of the library under these conditions. When a test set of 20 spectra that represented samples from growing regions, years, and environmental conditions not represented in the library and that were measured with different spectrometers and accessories was searched with the use of standard search algorithms, the standard search algorithm that produced the highest number of rank one identifications was able to correctly identify only 12 out of 20 spectra. More serious was the differing performance of the standard search algorithms. Figure 3 shows the number of correct rank one results returned by each of the six standard algorithms, broken down by the four categories of plant parts. One can see that a different algorithm performed best for each of the four categories of plant parts tested against the library. Furthermore, in some cases, the best performing algorithms for a given sample category were the worst performing algorithms for a different category of samples. For example, the correlation and square root algorithms correctly identified the most seed coat samples of any of the algorithms, but the correlation and square root algorithms identified fewer stem samples than any of the other algorithms. These data revealed that one can have significantly different rates of identification that depend upon the algorithm chosen to search this library. Since there is no single standard algorithm capable of identifying all sample types, the chances of successfully identifying an unknown

FIG. 2. Spectra of four different cotton leaves.

spectrum depend on the choice of algorithm, but for a true unknown sample, there is no way to know if the best algorithm has been chosen. These facts pointed to the need for a voting scheme algorithm.

Figure 4 shows the total number of correct results returned in the top 10 results by each of the six standard algorithms. From looking at these results, one might predict that it would be best to use the top ten matches returned by one of the derivative

FIG. 3. Number of samples in the 20 member test set correctly identified by the first result returned by each standard algorithm. The results are broken down by sample type as indicated in the legend.

APPLIED SPECTROSCOPY

665

FIG. 4. Total number of results in the top 10 results returned by each of the standard algorithms for each of the 20 test set samples that correctly identified a test spectrum.

algorithms as a basis for a voting algorithm, but Fig. 5 shows why this approach is not ideal. As with the number of correct rank one results (see Fig. 3), the algorithm that produces the greatest number of correct matches in the top ten results

returned varies by sample category. These results are indicative of the fact that for a given unknown sample the algorithm that produces the greatest number of correct answers cannot be predicted. For instance, Table I shows the actual results

FIG. 5. Total number of results in the top 10 results returned by each of the standard algorithms for each of the 20 test set samples that correctly identified a test spectrum. The results are broken down by sample type as indicated in the legend.

666

Volume 62, Number 6, 2008

TABLE I.

Search results for a leaf powder sample. Absolute value

Rank

Score

Category

1 2 3 4 5 6 7 8 9 10

1.318 1.636 1.651 1.702 1.785 1.788 1.815 1.843 1.874 1.877

Hull Hull Bract Stem Leaf Leaf Stem Leaf Leaf Leaf

a

Absolute value derivative Index numbera

Score

Category

703 697 65 518 363 364 805 525 523 526

0.2118 0.2129 0.2130 0.2157 0.2178 0.2214 0.2219 0.2223 0.2256 0.2261

Leaf Hull Hull Hull Hull Stem Hull Bloom Hull Hull

Index numbera 363 703 478 481 697 21 696 35 699 695

Unique index number assigned to the spectra from the library.

returned for a leaf sample from the test set when searching with the absolute value and the absolute value derivatives algorithms. From the results shown in Fig. 4, one would have predicted that the derivative algorithm would have given the greatest number of correct answers, but in this case (see Table I) the absolute value algorithm yielded five correct matches to the derivative algorithm’s one correct match. This example demonstrates why any successful voting scheme algorithm must incorporate information from all six standard algorithms. The three voting scheme algorithms developed were designed to take advantage of both complementary and repetitive information among the top ten results returned by the six algorithms. The results returned by the standard algorithms varied because each algorithm measures the similarity of spectra by a different metric, and since these spectra are so similar to begin with, different orderings of the TABLE II.

most likely candidate spectra occur. Despite this fact, one still expects that some of the algorithms will return some of the exact same spectra. This is the principle behind the frequency algorithm. The more times a particular result shows up in the hit lists of the six algorithms, the more likely that result correctly identifies the unknown contaminant. The weighted frequency algorithm modified this approach by considering both the number of algorithms that returned a particular result and the algorithm specific ranks of the results returned. This algorithm considers the facts that both frequency of a particular result among the 60 hits and the rank of those results among the individual ten-member hit lists are indicators of the probability of a particular result correctly matching the unknown. The group algorithm uses the frequency of particular results returned in a different manner. This algorithm is only

Top 10 ranked results for all standard algorithms for a cotton seed coat powder spectrum. Square root

Squared derivative

Absolute value

Rank

Score

Category

Index number

Score

Category

Index number

Score

1 2 3 4 5 6 7 8 9 10

25.15722 26.28441 27.73352 27.75385 29.44865 29.76650 29.76650 29.81726 29.83374 29.84427

Seed coat Seed coat Seed coat Hull Hull Seed coat Seed coat Hull Cacyx Bloom

165 166 624 703 481 164 168 696 122 35

0.0001371 0.0001542 0.0001582 0.0001615 0.0001620 0.0001622 0.0001630 0.0001632 0.0001641 0.0001726

Stem Hull Seed coat Seed coat Seed coat Hull Seed coat Seed coat Hull Hull

20 481 624 618 167 697 625 628 696 703

1.49286 1.50666 1.61960 1.75337 1.83423 1.83423 1.89044 1.91286 1.93826 1.95001

Dot product

Absolute value derivative

Rank

Score

Category

Index number

1 2 3 4 5 6 7 8 9 10

0.99224 0.99150 0.99150 0.99100 0.99080 0.99050 0.99048 0.99044 0.99032 0.99017

Seed coat Seed coat Seed coat Seed coat Leaf Seed coat Bract Hull Seed coat Stem

166 164 168 165 244 623 76 699 167 26

Category Seed Seed Seed Hull Seed Seed Hull Seed Hull Seed

coat coat coat coat coat coat coat

Index number 166 165 624 703 164 168 481 167 696 623

Correlation

Score

Category

Index number

Score

0.18702 0.19129 0.19322 0.19403 0.19486 0.19535 0.19782 0.19816 0.19844 0.19854

Stem Seed coat Seed coat Hull Hull Seed coat Hull Seed coat Hull Seed coat

20 624 628 481 697 167 489 625 703 618

0.99080 0.98959 0.98957 0.98754 0.98754 0.98581 0.98573 0.98537 0.98441 0.98384

Category Seed Seed Seed Seed Seed Seed Seed Hull Seed Hull

coat coat coat coat coat coat coat coat

APPLIED SPECTROSCOPY

Index number 166 624 165 164 168 167 627 703 623 481

667

TABLE III. Results from voting scheme algorithms for the same seed coat powder sample for which standard algorithm results are shown in Table II. Weighted frequency Rank 1 2 3 4 5 6 7 8 9 10 a b c

Frequency

Score

Substance name

Index/group numbera

Score

42 39 34 27 27 23 21 20 20 11

Seed coat Seed coat Seed coat Hull Seed coat Seed coat Seed coat Hull Stem Seed coat

624 166 165 481 164 168 167 703 20 628

5 5 5 5 4 4 4 4 3 3

Substance name Seed Hull Seed Hull Seed Seed Seed Seed Seed Hull

coat coat coat coat coat coat coat

Group Index/group numbera 167 481 624 703 164 165 166 168 623 696

Scoreb

Substance name

Index/group numberc

36 17 3 1 1 1 1

Seed coat Hull Stem Bloom Bract Leaf Cacyx

11 18 5 6 7 8 10

Unique index number assigned to the spectra from the library. All 60 possible spectra to be ranked are included in seven results. Unique numbers assigned to each category of spectra represented in the library.

considering the number of results returned in a particular category. Table II shows an example of how these algorithms work. The results returned by searching a particular seed coat spectrum from the test set with the six standard algorithms are shown. The group search algorithm would reveal that the group returned most often was seed coat, with 36 out of 60 results returned. The frequency search would focus on the fact that seed coat samples 167 and 624, appearing in the results five times each, tied for the first ranked result with hull samples 481 and 703, which also appeared five times each. Finally, the weighted voting search revealed that when frequency and rank are considered seed coat sample 624 is ranked number 1 with a score of 42. Table III summarizes the top matches returned by each of the voting scheme algorithms for the same seed coat test spectrum for which standard algorithm results are shown in Table II. The

results of the weighted frequency voting and group voting searches definitely identify the test spectrum as seed coat, while only four of the six standard algorithms had a correct rank one match for this test spectrum (see Table II). One should also notice that the derivative algorithms are the algorithms that did not have a correct rank one match for the test spectrum. Recall that in Fig. 4 the derivative algorithms were shown to provide the largest total of correct results in the top ten results returned by the six standard algorithms, but in this case the derivative algorithms did not give a correct rank 1 result. This outcome reiterates the fact that the six standard algorithms are not simply giving repetitive information: all of the algorithms seem to be giving some distinct information useful to correctly identify unknown spectra. The ability of these voting scheme algorithms to take advantage of both complementary and repetitive information provided by the standard algorithms’ results makes these voting scheme algorithms valuable.

FIG. 6. Number of samples in the 20 member test set correctly identified by the first result returned by each of the voting scheme algorithms.

668

Volume 62, Number 6, 2008

FIG. 7. Number of samples in the 24 member test set searched against the augmented library that was correctly identified by the first result returned.

The results for the voting scheme algorithms for the entire test set are shown in Fig. 6. The data for the group algorithm show impressive performance: The group search was able to yield as many correct rank one matches as the best standard

algorithm. By use of the group search algorithm, one does not need to make a decision as to which standard algorithm should be used and risk choosing a poorly performing algorithm. The group search uses the discriminating information given by all

FIG. 8. Number of samples in the 24 member test set searched against the augmented library that was correctly identified by the first result returned. The results are broken down by sample type as indicated in the legend.

APPLIED SPECTROSCOPY

669

of the standard algorithms to overcome the problem of different standard algorithms performing differently for different sample types (see Figs. 3 and 5). Figure 6 also reveals that while the frequency and weighted frequency algorithms performed better than some of the standard search algorithms, they do not have the same discriminating power that the group search algorithm has. One can gain some insight into this outcome by looking at the results shown in Table III. Regarding the frequency algorithm, the results show that many ties occur among the hits. This behavior is typical of the results obtained by this algorithm for all of the test samples, and this lack of discrimination means that this algorithm is not as successful at distinguishing among spectra as the other algorithms. The rank information considered by the weighted frequency algorithm, in addition to the frequency information, generally gives the weighted frequency algorithm more discrimination power than the frequency algorithm. This trend is also demonstrated in the results contained in Table III. The example in Table III cannot show us why the group search generally performs better than the weighted voting search, but one can infer the reasons from the search data for the test set on the whole. In the top ten results returned by the six standard algorithms, correct answers show up often, but are not always highly ranked. Also, these six standard algorithms all calculate spectral similarity so that they often return the correct category of sample in their results, but these algorithms use metrics to calculate spectral similarity that differ enough that particular spectra from the library are not consistently returned by all of the standard algorithms. These factors hurt both the frequency search and the weighted frequency search since these algorithms depend upon a particular spectrum from the library being returned consistently, or consistently and highly ranked, respectively, in the results from the standard algorithms. These same factors are beneficial to the group search algorithm because it does not rely on rank or the consistency of particular results being returned. The group search only considers the category of spectrum being returned. In order to demonstrate that the dependence of algorithm performance on sample category was related to the test set samples not being represented in the library, the 75 test spectra were added to the library. Twelve test spectra along with 12 spectra from the original library were searched against this augmented library, as described earlier, to show that when spectra representative of the test set were included in the library a high identification rate could be obtained by the standard algorithms. This experiment was a legitimate use of the 75 member test set because replicate spectra of the same plant parts represented by the test set were removed from the augmented library before each search was conducted. The results of this experiment are summarized in Fig. 7. The absolute value derivative and squared derivative algorithms yielded correct rank one matches for 22 out of 24 test spectra, and the number of correct rank one results improved for all of the standard algorithms. Figure 8 shows how each of the standard algorithms performed by sample category. The absolute value derivative and squared derivative algorithms consistently performed the best for all sample categories tested. The data show that when spectra representative of the test set were found in the library the performances of the standard search algorithms were predictable. These results demonstrated the importance of making the library representative of the spectra that are going to

670

Volume 62, Number 6, 2008

be searched against it, but in cases where the library cannot be made completely representative of the unknown spectra to be searched, the voting scheme algorithms provide a way to overcome the problem of not being able to choose a single standard algorithm for the searches.

CONCLUSION A spectral library of cotton contaminants had previously been developed to aid in the identification of foreign matter of extremely similar chemical composition and with closely related spectra found in cotton lint. Our experiments demonstrated that when this library was representative of the types of samples being searched against it, standard library searching algorithms accurately identified test samples, but when spectra of samples grown in different geographic locations, seasons, and environmental conditions and measured with a different spectrometer and ATR accessories were searched against the library, the identification rates for standard spectral search algorithms decreased significantly. Compounding this problem was the fact that under these conditions one could not reliably choose a standard search algorithm to search unknown spectra against the library because the performances of the standard algorithms varied by sample type; consequently, the best performing algorithm could not be predicted. Our experiments showed that by using the group voting scheme algorithm based on the numbers of samples returned from each category of samples represented in the library, a number of rank one identifications equal to the best standard algorithm could be obtained. The success of this voting scheme is due to the fact that the information gained from different standard search algorithms is complementary and repetitive. By using the information gained from multiple standard search algorithms, a more reliable and robust search algorithm was created. In addition to the identification of cotton contaminants in an underrepresented library, these voting scheme techniques could have applications to other spectral libraries with significant within group variation compared to between group variation, such as biological or forensic spectral libraries. ACKNOWLEDGMENTS The authors sincerely thank Drs. John Foulk and Angela Allen from the USDA–ARS Cotton Quality Research Laboratory in Clemson, South Carolina, for providing the contaminant samples that were used to measure the spectra test set. 1. D. S. Himmelsbach, J. W. Hellgeth, and D. D. McAlister, J. Agric. Food Chem. 54, 7405 (2006). 2. A. D. Brashears, R. V. Baker, and C. K. Bragg, ‘‘Effect of Bark on Spinning Efficiency of Cotton’’, in Proceedings of the Beltwide Cotton Conference (National Cotton Council, Memphis, TN, 1992), p. 1218. 3. J. Foulk, D. McAlister, D. Himmelsbach, and E. Hughs, J. Cotton Sci. 8, 243 (2004). 4. B. Xu, C. Fang, and R. Huang, Text. Res. J. 67, 881 (1997). 5. B. Xu and C. Fang, Text. Res. J. 69, 656 (1999). 6. S. R. Lowry, D. A. Huppler, and C. R. Anderson, J. Chem. Inf. Comp. Sci. 25, 235 (1985). 7. A. W. Baker, N. Wright, and A. Opler, Anal. Chem. 25, 1457 (1953). 8. L. E. Kuentzel, Anal. Chem. 23, 1413 (1952). 9. D. H. Anderson and G. L. Covert, Anal. Chem. 39, 11 (1967). 10. D. S. Erley, Anal. Chem. 40, 894 (1968). 11. R. A. Sparks, Storage and Retrieval of Wyandotte-ASTM Infrared Spectral Data Using an IBM 1401 Computer (ASTM, Philadelphia, PA, 1964). 12. A. Hanna, J. C. Marshall, and T. L. Isenhour, J. Chromatogr. Sci. 17, 434 (1979). 13. K. Tanabe and S. Sae¨ki, Anal. Chem. 47, 118 (1975). 14. L. A. Powell and G. M. Hieftje, Anal. Chim. Acta 100, 313 (1978).

Assessment of Near-Infrared Path Length in Fibrous Phantom and Muscle Tissue EUGENE GUSSAKOVSKY* and VALERY KUPRIYANOV Institute for Biodiagnostics, National Research Council Canada, Winnipeg, Canada

The first derivative of the pseudo-absorption spectrum of a water-loaded cotton wool (water–CW) phantom, which mimics muscle tissues, was used to determine the light path length in the near-infrared (NIR) region. The light path length increased as the density of the turbid medium decreased. It is independent of both water content in the range of 75–85% (by weight) and the diffuse reflecting reference used to determine the pseudoabsorbance. The path length determination procedure was verified by measurements of diffuse reflectance in chicken breast tissue for which the path length of 1.8 mm (differential path length factor, DPF ¼ 2.1) was found to be similar to the path length of NIR light of 1.5–2.2 mm (DPF ¼ 1.8–2.6) in a water–CW phantom of density similar to chicken breast. We conclude that the NIR light path length can serve as a characteristic of muscle tissue density. Index Headings: Near-infrared; NIR light path length; First derivative; Phantom; Muscle tissue.

INTRODUCTION Near-infrared (NIR) light absorption by water was used to estimate light path length in a turbid medium (tissue) using diffuse reflectance measurements.1 Diffuse reflectance measurements allow determination of the light absorption, which is represented by the pseudo-optical density (POD) spectrum:2–4 PODðkÞ ¼ log10

IðkÞ ¼ ala ðkÞL þ bðls0 Þ I0 ðkÞ

ð1Þ

where I and I0 are the diffuse reflectance intensities of the sample and the reference turbid media, respectively, la and ls0 are the absorption and reduced scattering coefficients, and L is a virtual light path length related to the differential path length and represents an average distance that light passes between emitting and light-collecting optodes in tissue.4 Previously, the second derivative of the light absorption or pseudo-absorption spectrum was used.1,2,5–7 Assuming a linear dependence of light scattering on wavelength, the second derivative was expected to eliminate the scattering contribution.2,3,5 A reduced light scattering coefficient l 0 s(k) (see Ref. 3 for example) for turbid media was proposed to be described by a power function ls0 ; kn that relates to Rayleigh scattering. Mie scattering can also contribute to total diffuse reflection (see Ref. 8 for example). However, experimentally, the scattering coefficient was reported to depend almost linearly on wavelength in the NIR range for phantoms and biological tissues.9–11 A significant disadvantage of the second-derivative approach is a high noise level. To eliminate its effect, a second differentiation procedure was applied after the absorption spectrum was smoothed3 or the second-derivative spectrum Received 11 October 2007; accepted 19 March 2008. * Author to whom correspondence should be sent. E-mail: Eugene. [email protected].

Volume 62, Number 6, 2008

had been smoothed.12 However, the smoothing depends on the numerical method used and may lead to undesired changes in the band shape and intensity. Moreover, the noise may still not be reduced to an acceptable level in the resulting curve (see smoothing filters in Ref. 13). Furthermore, the smoothing requires many spectral points (seven or more), which are not always available. It is obvious from Eq. 1 that it is impossible to determine the POD(k) without measuring the intensity of light reflected from the reference. For diffuse reflectance spectra, various materials have been used as references: a 20% by volume BaSO4 powder suspension having a flat spectral response across the visible range,14,15 a Kodak Gray Card with uniform 18% reflection,16 Spectralont,17,18 Liposyn19,20 and Hanks’s Balanced Salt Solution containing 0.9% lipids by volume with a scattering coefficient of 1 mm1 at 630 nm,19 polystyrene microspheres,21,22 2% Intralipid in gel23 or 10% Intralipid (a fat suspension of roughly spherical micelles of about 0.1–0.2 lm diameter) in combination with silica particles of 1–10 lm diameter,24–26 TiO2 powder in epoxy resin,27 homogenized milk,28 non-dairy creamer,29 and others. Phantoms were constructed to maximally mimic epithelial tissue30 or human oesophageal wall.24 However, no phantoms have been developed for muscle or myocardial tissues. Such tissues are characterized by long muscle and collagen fibers, a capillary network containing an aqueous solution of proteins, and suspended blood cells with 80–85% water content in free and bound form. Most of the suggested phantoms are spherical scatterers (liposomes, polystyrene beads, water-insoluble inorganic compounds) and contain not less than 90% water. No water-containing phantoms of fibrillar structures have been proposed. Until recently phantoms have been constructed to model tissue and contained water and absorbing chromophores as compulsory elements. However, they cannot be used as a reference for water absorption. Dry references such as Kodak Gray Card or MgO covered plates do not allow light penetration and hence are not good references for diffuse reflectance measurements. A dry thermoplastic resin, Spectralont, provides diffuse reflectance. However, it is much more dense (1.25–1.5 g/cm3) and hydrophobic and has a porous regular structure that differs from the fibrous structure of muscle. Diffuse reflectance computational models allow excluding the reference from consideration and can provide the light scattering contribution through spectral fitting procedures (see Refs. 31 and 32 for example). However, the fitting approaches require predetermined parameters such as the refractive index of tissue and others, which can only be approximated. In the present work, we (1) evaluated how the nature of the reference affects POD(k) and (2) applied first-derivative spectroscopy of NIR light absorption by water in the 650–

0003-7028/08/6206-0671$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

671

provided by 36 fiber-optic filaments with diameters of 15 lm randomly surrounding the illuminating filament. The mean effective distance between the illuminating and collecting filaments (optode distance) was determined to be 0.85 mm. The numerical aperture for both light-emitting and collecting optodes was measured to be 0.246. POD(k) spectra were calculated according to Eq. 1. Before differentiation, they were smoothed using a 5-point moving linear function algorithm. The smoothed value was attributed to the third point in the middle of the range. No smoothing of the first derivative was applied. A water–cotton wool (CW) phantom was constructed in a cylindrical well with a depth of 12.6 mm, diameter of 19.1 mm, and volume of 3.6 cm3. Illumination and light collection were performed from the top of the phantom-packed well. At 12.6 mm depth, such turbid medium may be considered to be semiinfinite. The cotton wool density was determined as its weight per the well volume (g/cm3). The water content was expressed as water volume per well volume. The total volume of waterþCW was adjusted to the well volume by packing the wet cotton wool, resulting in varying CW densities. Fresh chicken breast was cut into cubes of 0.5 to 1 g and kept in phosphate-buffered saline solution at a pH of 7.4. The external water was removed by blotting with filter paper and the POD(k) spectra were measured from four sides of each piece. The procedure to determine the light path length was applied to the average of these four spectra. Each piece was weighed (wet weight, Ww) and then dehydrated at 50 8C for 90 hours until constant weight was achieved (dry weight, Wd). The water content of the chicken breast tissue was calculated as 1  Wd /Ww. FIG. 1. (A) Pseudo-optical density (POD) spectra and (B) their first derivatives (d(POD)/dk) obtained from a CW–phantom at a cotton wool density of 20.7% soaked by 80.9% water. The spectra are presented relative to various references: (1) Kodak Gray Card; (2) white sponge with a density of 0.132 g/cm3; (3) white sponge with a density of 0.032 g/cm3; and (4) cotton wool with a density of 0.207 g/cm3. In (B), spectra 2–4 overlap in the range of 800–1050 nm. Vertical arrows show minor bands of the differential light absorption of water at 730 and 830 nm. No correction of intensity value or intensity shift was done.

1050 nm wavelength range to estimate the light path length in a water-loaded phantom that mimics muscle tissue and in fresh chicken breast muscle.

MATERIALS AND METHODS Spectra were acquired in the dark background-corrected counts mode from 480 to 1100 nm with an increment of 1 nm and an integration time of 10 ms at 25 8C using a Control Developments Inc. model PDA-512 VIS/NIR spectrometer (South Bend, IN). Fifty consecutive spectra in a set were averaged to increase the signal-to-noise ratio. Broadband visible/NIR light from a fiber-optic illuminator Oriel model 77501 (Stratford, CT) was transmitted to the sample through one arm of a bifurcated fiber-optic bundle. The common illumination/collection probe tip was placed in contact with the sample perpendicular to its plane, which allowed the collection of predominantly diffuse reflected light through the other arm of the fiber bundle to the spectrometer. The light from the light source was delivered by a single fiber-optic filament with a diameter of 650 lm. The reflected light collection was

672

Volume 62, Number 6, 2008

RESULTS AND DISCUSSION Choice of Phantom and Reference. We used water-loaded cotton wool (water–CW) as a model of skeletal muscle and myocardium to determine NIR light path length. Cotton wool is a good model for a number of reasons: (1) it has a fibrillar internal structure, (2) it may be prepared in various densities, (3) it is a good water sorbent (15–25%) that can also retain free water outside the fibrils, and (4) it can be prepared to 80–85% water content and a density of 1.06 g/cm3 (mammalian skeletal muscle)31,32 throughout the phantom. A phantom of such density that contains 80% water and 20% cotton wool consists of 0.85 mL of water and 0.212 g of cotton wool per 1 cm3. Dry cotton wool may serve as a good reference because its lightscattering structure is very similar to the water–CW phantom. To estimate the effects of different reference materials on POD, a Kodak Gray Card and two polyurethane white sponges of various densities (0.032 and 0.132 g/cm3) were also tested. Figure 1 shows POD(k) spectra and their first derivatives in the 650–1050 nm wavelength range for a water–CW phantom obtained using the references listed above. POD(k) spectra for the sponges and dry CW of the 0.207 g/cm3 density, which provide true diffuse reflectance, differed only by the offset with no significant slope in the wavelength range of about 700–900 nm, where the water absorption is minimal (1–10% of the 975 nm peak intensity).5 With the Kodak Gray Card as a reference, the spectrum had significant positive slope. The positive slope is not likely to be related to light scattering since light scattering is known to be described by a negative power function of wavelength8 and hence should have a negative slope. This establishes the effect of the reference on the

POD(k) spectra and demonstrates the necessity of taking it into account together with the actual light scattering. First derivatives of the POD(k) spectra of the water–CW phantom with the dry CW or sponge references did not differ within error of measurements and were close to zero outside of the differential water absorption bands at 730, 830, and .930 nm (Fig. 1B). For the Kodak Gray Card the first-derivative spectrum was significantly different. Therefore, both sponges and cotton wool were considered to be suitable references for the water–CW phantom because the first derivative mostly eliminated non-absorbing components and rendered the spectral appearance reference invariant. The Kodak Gray Card is less suitable, most likely because it does not provide true diffuse reflectance. First Derivative and Light Path Length. Figure 2 shows the coincidence between the shapes of derivative spectra of water light absorption in a cuvette measured as Aw(k) ¼ 4pkw(k)/k (the imaginary part of refractive index kw(k) was taken from Ref. 33) and in the water–CW phantom used in this study. This means that under our POD-measurement the steric conditions (fiber-optic probe attached to the phantom perpendicular to the plane with zero distance between them), the cotton wool in the phantom does not measurably affect the molecular vibrations of water (symmetric stretch m1, asymmetric stretch m3, and bend m2, the combination of which determines absorption bands of 739, 836, and 970 nm)39 because the first derivative reveals even small changes in the absorption spectrum induced by changes of the spectral constituents. Therefore, the NIR absorption properties of free water and water in the phantom may be considered to be the same. As pure water absorption Aw depends on the light path length L, as Aw ¼ lwL, and the absorption coefficient lw is known, the light path length L in the phantom, deduced from the pseudo-optical density POD(k) [ P(k) defined by Eq. 1, may be determined as L¼

P 0 ðkÞ  mP lw0 ðkÞ

ð2Þ

where lw0 (k) ¼ dlw/dk, P 0 (k) ¼ dP/dk in the water absorption band, and mP ¼ mean(P 0 ) over the ranges of 650–680 and/or 750–800 nm. At these wavelengths, the first derivative of pure water absorption Aw0 is close to zero.5,34–36 Therefore, the mP value relates only to the non-absorbing components (light scattering and reference contribution). Both POD(k) and lw(k) should be determined in accordance with the optical density definition (Eq. 1). We assume that light scattering in the NIR range is a linear function of wavelength according to Refs. 9–11. In general, either Mie or Rayleigh scattering is nonlinear. However, in a short wavelength range and in the presence of noise, this approximation is valid.9 This was confirmed experimentally when, for the scattering presented as ls0 ¼ akn, the power n has been found to almost equal 1 (n ¼ 0.82 by Bevilacqua et al.32 at 650–1000 nm; n ¼ 1.11 by Doornbox et al.31 at 600–900 nm). Moreover, the difference DP ¼ POD  Aw, which should relate to light scattering only, actually also depends on the reference, which has been used for the POD(k) determination according to Eq. 1. According to Fig. 1B, the Kodak Gray Card or the white sponge with a density of 0.132 g/cm3 used in this study as references do not provide P 0 (k) ¼ constant at the desired wavelength while P 0 (k) was constant for both the dry cotton

FIG. 2. Pseudo-optical density first-derivative d(POD)/dk of a water–CW phantom at a cotton wool density of 20.7% loaded by 80.9% water (thin line) and the first derivative of the water absorption spectrum published in Ref. 35 (bold line). The spectra were equalized by the water derivative intensity at 950 nm. Vertical arrows show minor bands of the differential light absorption of water at 730 and 830 nm.

wool and white sponge with a density of 0.032 g/cm3. This clearly shows that the estimate of the expected scattering depends only on the reference. Therefore, DP cannot be correctly described by either Rayleigh or Mie theory unless the reference has a structure similar to that of the sample. From a practical point of view, if the first derivative is constant and equal to mP ¼ mean(P 0 ) at 650–680 and/or 750–800 nm within the error of measurement, one can assume that the description of light absorption by the phantom as [P 0 (k)  mP] is valid at other wavelengths in the range of 650–1050 nm. In other words, mP serves as a simple offset. To reduce the effect of noise, after integration over the main absorption band Eq. 2 becomes R 0 ½P ðkÞ  mP  dk R 0 ð3Þ L¼ lw dk The integral presentation in Eq. 3 does not affect the L value because of the invariance of the first-derivative spectrum R shape. The value of lw0 dk was calculated to be 0.1545 cm1 for integration over the 920–960 nm wavelength range corresponding to the water absorption spectrum.35 Therefore, determination of the light path length in cm requires (1) differentiating the POD spectrum, (2) taking an integral over 920–960 nm, and (3) dividing it by 0.1545 cm1. No spectral or intensity adjustments are needed. Similarly, Eq. 2 may also be used to determine the light path length when the signal-to-noise ratio of the P 0 (k) spectrum provides better than 10% accuracy in the determination of L. A lw0 (k) function is available from the first derivative of the water absorption spectrum. In this case, the opportunity to determine the dependence of L on wavelength may be an advantage. What Does the Light Path Length Depend Upon? Let us consider the parameters that affect the results of calculations using Eq. 3. Reference. The reference used to determine the POD(k) spectrum is the first one. Diffuse reflectance depends on the turbid medium structure, and the optical geometry of the light source and collector of reflected light for both the sample and reference according to Eq. 1. Variations in POD(k) spectra for

APPLIED SPECTROSCOPY

673

TABLE I. Absolute light path length and DPF in a water–CW phantom at various water content.a Relative volume of water loaded, % 75.7 77.0 80.9 82.4 85.1 Mean absolute path length, L Mean DPF ¼ 2.54 6 0.14 a

FIG. 3. Dependence of the path length of 920–960 nm light on cotton wool density (first batch) at a water content of 80.9% determined from d(POD)/dk using different references: dry cotton wool of corresponding densities (filled circles and bold line), white sponge with a density of 0.132 g/cm3 (open circles and thin line), white sponge with a density of 0.023 g/cm3 (open triangles and dashed line), and Kodak Gray Card (open squares and dotted line). Diamonds and dashed-dotted line show the path length in water–CW determined versus dry cotton wool of the second batch. The straight lines indicate a trend for the cotton wool in the 0.05–0.25 g/cm3 density range and do not imply linear correlation.

FIG. 4. (A) Pseudo-optical density (POD) spectra and (B) their first derivatives (d(POD)/dk) for a CW phantom at a cotton wool density of 0.15 g/cm3 and various volumes of water added: (1) 75.7%; (2) 77.0%; (3) 80.9%; (4) 82.4%; and (5) 85.1%. In (B), all spectra overlap. No correction or intensity shift was done. The reference spectrum was acquired from dry cotton wool of the same density. The vertical arrows show minor bands of the differential light absorption of water at 730 and 830 nm.

674

Volume 62, Number 6, 2008

Light path length, mm 1.99 2.19 2.07 2.25 2.28 2.16 6 0.12

The cotton wool density in the phantom was 0.15 g/cm3. The differential path length factor (DPF) was determined as the absolute path length per optode distance (see Materials and Methods section).

different references clearly show the effect of medium structure (Fig. 1A). However when the first derivative was applied, no effect of the reference structure on the path length determination was found unless the reference does not provide a diffuse reflectance (Kodak Gray Card). Cotton Wool Density. The cotton wool density, the variation of which mimics different muscle tissue density (content of dry matter in a unit of volume) at a constant water content, is the second parameter. We varied both the cotton wool batch and cotton wool density from 0.05 to 0.23 g/cm3 in the water–CW phantom at the same water content of 80.9%. Figure 3 shows that lower cotton density leads to longer NIR light path length regardless of the reference or cotton wool batch. Straight lines in Fig. 3 do not necessarily imply linear correlation but rather indicate the general tendency, that a higher density of tissue reduces light path length, which can be associated with lower depth of the light penetration. Water Content. The water content in the water–CW phantom, which mimics variations in hydration of the tissue at a constant dry matter density, is the third parameter. Figure 4 shows that variation in the water content in the range of 75– 85% (similar to the water content in muscles)33 does not affect the first derivative, while the POD(k) spectrum appears to be different. The figure confirms the advantage of the first derivative in relation to the POD(k) spectrum. In agreement with the lack of effect of water content, Table I shows that the absolute NIR light path length (L ¼ 2.2 mm) and the differential path length factor (DPF ¼ L/d ¼ 2.5, where d ¼ 0.85 is an optode distance; see Materials and Methods section) are not functions of the water content within an accuracy of 5–6%. At 75–85% water, there is probably an excess of free water, which does not affect cotton-adsorbed water (it is known that the moisture saturation of cotton fibers is 15–25%; therefore, 50– 70% of water is not bound). As a consequence, the diffuse reflection remains unchanged. Path Length in Chicken Breast Tissue. To verify the firstderivative approach for actual muscle tissue, we applied it to diffuse reflectance of NIR light in chicken breast. This tissue is expected to contain little or no hemoglobin and myoglobin. Therefore, the NIR light absorption should be determined by water only. POD(k) spectra of chicken breast (Fig. 5A) varied when calculated relative to different references, as was the case with the water–CW phantom. There were no significant absorption bands in the range of 500–600 nm, indicating the absence of hemoglobin and myoglobin in the tissue (measured with diffuse reflectance technique; data are not shown). Therefore,

CONCLUSION Use of the first derivative of POD(k) spectra of turbid media is a useful approach to determine the NIR light path length. The first derivative is not affected by light scattering, does not depend on any diffuse reflecting reference, and is convenient for light path length determination. It is shown that lowering the density of scattering matter associates with longer NIR light path length. Light path length is independent of both water content in the range of 75–85% and the diffuse reflecting reference used to determine the pseudo-absorbance. The path length value determination procedure was verified by measurements of diffuse reflectance in chicken breast tissue and can be used as a characteristic of muscle tissue density. ACKNOWLEDGMENTS This work was supported in part by the Genomics and Health Initiative Programme (GHI-3) of National Research Council Canada (NRCC). We are thankful to Drs. R. Deslauriers, R. A. Shaw, and J. C. T. Rendell, Institute for Biodiagnostics, NRCC, for fruitful discussion.

FIG. 5. (A) Pseudo-optical density (POD) spectra and (B) their first derivatives (d(POD)/dk) of chicken breast. The spectra are presented relative to various references: (1) Kodak Gray Card; (2) white sponge with a density of 0.132 g/cm3; (3) white sponge with a density of 0.032 g/cm3; and (4) cotton wool with a density of 0.207 g/cm3. In (B), spectra 2–4 overlap in the range of 800–1050 nm. The bold line shows the differential spectrum of pure water equalized by intensity at 950 nm and offset at 745–770 nm to the values of the chicken breast differential spectrum. The vertical arrows show minor differential bands of the light absorption of water at 730 and 830 nm. No corrections of intensity value or intensity shift for chicken breast spectra were performed.

only water should determine the POD(k) spectrum in the NIR range. The first derivative of chicken breast POD(k) spectra did not depend on the reference with the exception of the Kodak Gray Card (Fig. 5B), which does not provide diffuse reflectance. When the first derivatives of the chicken breast (Fig. 5B; lines 2–4) and pure water were superimposed, they differed in intensity only and their shapes appeared to be the same within the measurement error. This finding indicates that water has the same light absorption parameters (vibrational structure) in the phantom and in the chicken breast. Therefore, the difference is a function of the light path length only. The path length of light at 920–960 nm in chicken breast obtained from the POD(k) first derivative under our measurement conditions was determined to be 1.80 6 0.04 mm (variation of diffuse-reflecting references) at breast water content of 79.8 6 0.9% (n ¼ 4). This L value is in the range of 1.5–2.2 mm found for the water–CW phantom at a water content of 79.3% and a dry cotton wool density of 0.2 g/cm3 (see Fig.3).

1. S. P. Nighswander-Rempel, V. V. Kupriyanov, and R. A. Shaw, J. Biomed. Opt. 10, 0240023 (2005). 2. S. J. Matcher and C. E. Cooper, Phys. Med. Biol. 39, 1295 (1994). 3. L. S. L. Arakaki, M. J. Kushmerick, and D. H. Burns, Appl. Spectrosc. 50, 697 (1996). 4. D. T. Delpy, M. Cope, P. van Zee, S. Arridge, S. Wray, and J. Wyatt, Phys. Med. Biol. 33, 1433 (1988). 5. S. J. Matcher, M. Cope, and D. T. Delpy, Phys. Med. Biol. 39, 177 (1994). 6. L. M. Klassen, B. J. MacIntosh, and R. S. Menon, Phys. Med. Biol. 47, 1573 (2002). 7. F. Holler, D. H. Burns, and J. B. Callis, Appl. Spectrosc. 43, 877 (1989). 8. I. S. Saidi, S. L. Jacques, and F. K. Tittel, Appl. Opt. 34, 7410 (1995). 9. S. J. Matcher, M. Cope, and D. T. Delpy, Appl. Opt. 36, 386 (1997). 10. S. J. Matcher, P. Kirkpatrick, K. Nahid, M. Cope, and D. T. Delpy, Proc. SPIE-Int. Soc. Opt. Eng. 2389, 486 (1995). 11. A. Cerussi, N. Shah, D. Hsiang, A. Dunkin, J. Butler, and B. J. Tromberg, J. Biomed. Opt. 11, 044005 (2006). 12. L. S. L. Arakaki and D. H. Burns, Appl. Spectrosc. 46, 1919 (1992). 13. W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes in C: The Art of Scientific Computing (Cambridge University Press, New York, 2002), 2nd ed. 14. G. Zonios, L. T. Perelman, V. Backman, R. Manohatran, M. Fitzmaurice, J. Van Dam, and M. S. Feld, Appl. Opt. 38, 6628 (2002). 15. M. Mirenda, M. G. lagorio, and E. San Roman, Langmuir 20, 3690 (2004). 16. V. V. Kupriyanov, S. Nighswander-Rempel, and B. Xiang, J. Mol. Cell. Cardiol. 37, 947 (2004). 17. J. R. Mourant, I. J. Bigio, J. Boyer, T. M. Johnson, J. A. Lacey, A. G. Bohorfoush, and M. Mellow, J. Biomed. Opt. 1, 192 (1996). 18. D. C. G. de Veld, M. Skurichina, M. J. H. Witjes, R. P. W. Buin, H. J. C. M. Sterenborg, and J. L. N. Roodenburg, Lasers Surg. Med. 36, 356 (2005). 19. J. C. Finlay and T. H. Foster, Opt. Lett. 29, 965 (2004). 20. F. Fabbri, M. A. Franceschini, and S. Fantini, Appl. Opt. 42, 3063 (2003). 21. N. M. Marin, A. Milbourne, H. Rhodes, T. Ehlen, D. Miller, L. Benedet, R. Richards-Kortum, and M. Follen, Gynecol. Oncol. 99, S116 (2005). 22. S. C. Gebhart, A. Mahadevan-Jansen, and W.-C. Lin, Appl. Opt. 44, 4884 (2005). 23. P. R. Bargo, S. A. Prahl, T. T. Goodell, R. A. Sleven, G. Koval, G. Blair, and S. L. Jacques, J. Biomed. Opt. 10, 034018 (2005). 24. G. Wagieres, S. Cheng, M. Zellweger, N. Utke, D. Barichotte, J.-P. Ballini, and H. van den Bergh, Phys. Med. Biol. 42, 1415 (1997). 25. H. J. van Staveren, C. J. M. Moes, J. van Marle, S. A. Prahl, and M. J. C. van Gemert, Appl. Opt. 30, 4507 (1991). 26. S. T. Flock, S. L. Jacques, B. C. Wilson, W. M. Star, and M. J. C. van Gemert, Lasers Surg. Med. 12, 510 (1992). 27. J. Swartling, J. S. Dam, and S. Anderson-Engels, Appl. Opt. 42, 4612 (2003). 28. B. Wilson, Y. Park, Y. Hefetz, M. Patterson, A. Madsen, and S. Jacques, Proc. Soc. Photo-Opt. Instrum. Eng. 1064, 97 (1989). 29. J. Linfird, S. Shalev, and J. Bews, Med. Phys. 13, 869 (1986).

APPLIED SPECTROSCOPY

675

30. K. Sokolov, J. Galvan, A. Myakov, A. Lacy, R. Lotan, and R. RichardsKortum, J. Biomed. Opt. 7, 148 (2002). 31. R. M. P. Doornbos, R. Lang, M. C. Aalders, F. W. Cross, and H. J. C. M. Sterenborg, Phys. Med. Biol. 44, 967 (1999). 32. F. Bevilacqua, A. J. Berger, A. E. Cerussi, D. Jakubowski, and B. J. Tromberg, Appl. Opt. 39, 6498 (2000). 33. J. Mendez and A. Key, Metabolism 9, 184 (1960). 34. S. S. Segal, T. P. White, and J. A. Faulkner, Am. J. Physiol. (Cell Physiol.) 250, C474 (1986).

676

Volume 62, Number 6, 2008

35. L. Kou, D. Labrie, and P. Chylek, Appl. Opt. 32, 3531 (1993). 36. R. C. Smith and K. S. Baker, Appl. Opt. 20, 177 (1981). 37. H. Buiteveld, J. H. M. Hakvoort, and M. Donze, Proc. SPIE-Int. Soc. Opt. Eng. 2258, 174 (1994). 38. R. A. J. Litjens, T. I. Quickenden, and C. G. Freeman, Appl. Opt. 38, 1216 (1999). 39. M. Chaplin, http://www.lsbu.ac.uk/water/vibrat.html (accessed July 20, 2007).

A Noninvasive Method for Assessing Interior Skin Damage Caused by Chronological Aging and Photoaging Based on Near-Infrared Diffuse Reflection Spectroscopy YUTA MIYAMAE,* YUMIKA YAMAKAWA, MARIE KAWABATA, and YUKIHIRO OZAKI POLA Chemical Industries, Inc., Corporate Planning Department, 27-1 Takashimadai, Kanagawa-ku, Yokohama 221-0833 Japan (Y.M., Y.Y., M.K.); and Department of Chemistry and Research Center for Near-Infrared Spectroscopy, School of Science and Technology, Kwansei-Gakuin University, Sanda 669-1337, Japan (Y.O.)

This paper reports a noninvasive method for evaluating skin aging based on near-infrared diffuse reflectance (NIR-DR) spectroscopy. Skin aging can be attributed to photoaging and chronological aging. Both types of aging are heavily involved in the skin changes that occur as we get older, for example, wrinkles or sagging skin. Our goal is to develop a noninvasive way to assess changes taking place inside the skin for each type of aging by using NIR-DR spectroscopy. Interior skin damages caused by photoaging and chronological aging were studied for an ultraviolet-B (UVB)irradiated hairless mouse group (24 mice) and a non-irradiated group (29 mice) by using NIR-DR spectroscopy and principal component analysis (PCA). The results suggested the possibility of monitoring the contribution and the quantitative assessment of both types of aging taking place inside the skin by using the 5990–5490 cm1 and 5000–4480 cm1 regions of NIR-DR spectra. For the photoaging, structural changes in proteins are most clearly reflected by a shift of the band near 4880 cm1 due to a combination of amide A and amide II modes. On the other hand, the chronological aging is associated with a change in collagen quantity as is seen in the intensity changes in NIR bands assigned to collagen. NIR-DR spectroscopy and PCA may allow us to noninvasively assess the degree of photoaging and chronological aging as the degeneration of elasticity in collagen protein and the degradation of protein quantity, respectively. Index Headings: Near-infrared spectroscopy; Diffuse reflection spectroscopy; NIR-DR spectroscopy; Aging; Principal component analysis; PCA; Collagen; Skin; Noninvasive analysis.

INTRODUCTION Skin aging involves both photoaging and chronological aging processes.1–6 The effects of these processes are often overlapping and include changes in both the stratified epithelium and the fibroblast-rich dermis. Photoaging takes place in human skin that is subjected to repeated exposure to ultraviolet irradiation, and chronological aging is induced with the passage of time. It is well known from previous studies on human skin and animal skin that skin surface properties and the state of changes taking place inside the skin differ vastly between the two types of aging.1–6 Each stage of the two types of aging progression is subject to considerable variations, depending on environmental climate, life style, activity, and heredity. In order to develop more effective beauty counseling methods for skin, it is very important to determine the contribution and the quantitative assessment of these two types of aging on the individual consumer’s skin. We have, therefore, attempted to develop a noninvasive near-infrared (NIR) diffuse reflection (DR) method for the analysis of both types of skin aging. With a precise and simple Received 14 November 2007; accepted 11 March 2008. * Author to whom correspondence should be sent. E-mail: y-miyamae@ pola.co.jp.

Volume 62, Number 6, 2008

way to assess aging taking place inside the skin that may not yet be visible on the surface, consumers will be alerted to treatment options at an earlier stage, when treatment can make more of a difference. Invasive methods of skin tissue analysis, however, will not be tolerated by many consumers regardless of the benefits offered. Studies on the physical properties of the skin have been conducted using cutemeters, analysis of the skin surface, or image analysis of skin echogenicity as noninvasive methods of assessing total skin aging.3,7,8 However, these methods have limitations in their ability to distinguish between the two types of aging, and it is very difficult to monitor some chemical changes in the interior of skin. Fluorescence measurement is another noninvasive method that has been employed to assess aging, but few researchers have satisfactorily applied this technique to humans.9,10 In cosmetic research, NIR-DR spectroscopy has previously been used for measuring water content in skin, hair, and nails and hemoglobin or quantity of reddening (erythemal reaction) of skin.9–11 However, to our best knowledge, no published research has utilized this technology to assess skin aging. The objective of the present study is to monitor some chemical changes in the interior of the skin and to develop a noninvasive method for assessment that distinguishes the photoaging and chronological aging of the skin. We analyzed various NIR-DR spectra of the skin of two hairless mouse groups, an ultravioletB (UVB)-irradiated group and a non-irradiated group, and we investigated relationships between the results of the NIR-DR spectra and principal component analysis (PCA) and states of aging taking place inside the skin.

MATERIALS AND METHODS Animal Models of the Irradiated Group and the Nonirradiated Group. Five-week-old female albino hairless mice (Hr-1) were purchased from Hoshino (Japan). The animals were acclimatized for one week prior to the study. Fifty-three female albino hairless mice were divided into two groups, one irradiated group (24 mice) and one non-irradiated group (29 mice). The irradiated animals were exposed to ultraviolet radiation (UVB dose of 50 mJ per cm2 measured by the UVB probe of a Topcon radiometer) from a bank of four Toshiba SE lamps without any filtering, placed 40 cm above the back of each animal. As previously reported,6 the peak spectral output of these lamps is approximately 310 nm, with no energy detectable below 260 nm and approximately 0.6% between 260 and 280 nm (UVC), 72.7% between 280 and 320 nm (UVB), and 26.7% between 320 and 400 nm (UVA). The energy output of the lamps was measured with a Topcon UV radiometer 305/365D. Under these conditions, the minimal

0003-7028/08/6206-0677$2.00/0  2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

677

TABLE I. Distribution of animal models by age in weeks. Age in weeks

6

8

10

14

16

27

UVB-irradiated group Non-irradiated group

– 6

6 –

6 6

6 6

6 6

– 5

erythema dose (MED) to these animals was 60 mJ per cm2 UVB. The animals were irradiated three times a week for 10 weeks. Under these conditions, mice did not show any erythema, edema, or scaling, so the irradiations were below their MED thresholds. The total dose for a 10-week period of predominantly UVB exposure amounted to 2.7 J per cm2. The animals were fed a standard diet, allowed water ad libidum, and housed in rooms where the lighting was free from both UVA and UVB emissions. A 12 h automated light and dark cycle was used. Non-irradiated animals were prepared to measure as 6-, 10-, 14-, 16-, and 27-week-old animals (Table I). Near-Infrared Diffuse Reflection Measurements and Data Analysis. Near-infrared diffuse reflection spectra of the purchased skin tissues and animal models were obtained with a Fourier transform near-infrared (FT-NIR) spectrometer, IFS28/ N (Bruker Optics, Ettlingen, German) using a fiber probe placed on the backs of the animals. Uniform measurement conditions were maintained. The indoor temperature was kept at 20 8C, and a distance of 1.2 mm from the skin surface to the NIR-DR fiber probe was maintained by using a stainless steel ring with a thickness of 1.2 mm. NIR-DR spectra from 8000 to 4000 cm1 were collected at 8 cm1 spectral resolution and 32 scans. Pirouette, Version 3.11 (InfoMetrix, Bothell, WA) was used for data analysis. To extract information about the aging of skins, we selected the 8000–4000 cm1 region of the NIRDR spectra and subjected that region to principal component analysis (PCA). Monitoring of the Inside of the Skin. The skin sections of each model were denuded. The sections were fixed in 10% buffered formalin, processed in paraffin wax, and stained by the standard haematoxylin and eosin method. The epidermal thickness and dermal thickness were measured by light microscopy. Scanning electron microscopic (SEM) measurements were performed for monitoring the ultrastructure of dermal collagen fiber bundles on mouse skin specimens on an S-570 (Hitachi, Tokyo, Japan).

RESULTS Verification of the Animal Models. For the group of UVirradiated mice, photodamage increases gradually with UV dosage, as manifested by increased formation of wrinkles, epidermal and dermal thickening, and changes in the dermal collagen fiber bundles. Therefore, it is verified to be appropriate for the photoaging model. On the other hand, the group of non-irradiated mice is not appropriate for chronological aging, because no change in skin tissue was observed in this group. In this study, we used animal models for the photoaging and increased aging group. Near-Infrared Diffuse Reflection Spectra of the Animal Models. Figures 1a and 1b show an original NIR-DR spectrum in the 8000–4000 cm1 region of a mouse after treatment with standard normal variate (SNV) and taking the second derivative (SD), respectively. Two of the most prominent peaks in the original spectrum are due to water absorption. A

678

Volume 62, Number 6, 2008

FIG. 1. NIR-DR spectra in the 8000–4000 cm1 region. (a) Original spectrum, (b) second-derivative spectrum.

band due to the combination of the O–H antisymmetric and symmetric stretching modes of water is found at 6900 cm1, whereas that arising from the combination of the O–H stretching and bending modes accounts for an absorption at 5180 cm1 . Bands due to the first overtones of the antisymmetric and symmetric C–H stretching modes are located at 5830 and 5680 cm1, respectively (see Fig. 1b). Combination bands including the N–H stretching and different amide modes are identified at about 4880 and 4590 cm1.12 The 5990–5490 cm1 and 5000–4480 cm1 regions were selected for the present purpose because the intensities of the peaks in these regions are stronger than those in other regions, except for the regions where the two strong water features are observed. It is also noted that the water band regions are affected by measurement environmental conditions such as humidity. Figure 2 depicts NIR second-derivative spectra in the 5000– 4480 cm1 region of the photoaging model (16 weeks old), chronological aging model (6 weeks old), and chronological aging model (27 weeks old). It is found from the comparison of the NIR-DR spectra in the 5000–4480 cm1 region between the photoaging model and the chronological aging model that a band near 4880 cm1 shows a large shift by about 10 cm1 for the photoaging and a peak near 4590 cm1 increases significantly. Amide A indicates a NH stretching mode of the amide groups, and amide II is a key indicator for the secondary structure of the peptide backbone.13 The characteristic peak shift of the band near 4880 cm1 indicates a structural variation



FIG. 2. NIR second-derivative spectra in the 5000–4480 cm1 region. ( ) Photoaging model, 16 weeks old, (–––) chronological aging model, 6 weeks old, (u) chronological aging model, 27 weeks old.

FIG. 4. NIR second-derivative spectra of the chronological aging model in the 5990–5490 cm1 region: (–––) 6 weeks old, (n) 10 weeks old, (m) 14 weeks old, (u) 16 weeks old, and (*) 27 weeks old.

in protein, e.g., denaturation of the protein.13,14 It was also reported that NIR-DR spectra between 4700 and 4500 cm1 involve information about protein secondary structures.13,14 Figure 3 depicts the wavenumber of the peak near 4880 cm1 versus age for both the photoaging and chronological aging models. Of note is that the photoaging model shows a much larger shift than the chronological aging model (closed testing procedure, p , 0.001) (Fig. 3). Figure 4 displays NIR-DR spectra in the 5990–5490 cm1 region of the chronological aging model. The 5990–5490 cm1 region is mainly concerned with the first overtone and combinations of CH, CH2, and CH3 groups.12 It can be seen from Fig. 4 that the chronological aging model shows remarkable changes near 5930, 5790, and 5670 cm1. In contrast, there is little corresponding change in this region for the photoaging model. It was reported that the weak feature near 5930 cm1 is assigned to proteins, and the peaks near 5790 and 5670 cm1 arise from lipids.12,15 With aging, the peak intensity near 5930 cm1 decreases while the intensity of the peaks near 5790 and 5670 cm1 increases. Figure 5 illustrates that a significant correlation is noticed between the intensity of the peak near 5790 cm1 and the age in weeks

(Pearson’s correlation coefficient r ¼ 0.86, p , 0.001***, Fig. 5). These changes in the 5990–5490 cm1 region are concerned with the increased age. Principal Component Analysis and Loading Plots of the Near-Infrared Diffuse Reflection Spectra in the 5990–5490 and 5000–4480 cm1 Regions. The NIR-DR spectra in the 5990–5490 and 5000–4480 cm1 regions of the two models were subjected to PCA. A PC2 and PC3 score plot of PCA is shown in Fig. 6. The two groups formed from the photoaging model and the chronological aging model are divided along the principal component PC2 and PC3 in the PCA score plot. When comparing the score plots between the photoaging model and the chronological aging model for the 16-week-old mice, there is a remarkable difference in the circled places. To explore what divides the two models clearly in the PCA score plot, we calculated the loading plots. Figure 7 shows loading plots of PC2 and PC3. A peak at 4480 cm1 appears to play a key role in differentiating between the photoaging model and the chronological aging model. However, the peaks near 4880 and 4590 cm1 are also useful in classifying the samples into the two groups along PC3. These peaks in the loading plot for PC3 indicate that the differences in

FIG. 3. The wavenumber of the peak top near 4880 cm1 in the secondderivative NIR-DR spectra for each model as a function of age. (u) Chronological aging model, (m) photoaging model.

FIG. 5. The intensity of the peak near 5790 cm1 in the second-derivative NIR-DR spectra versus age for the chronological aging model and photoaging model. (u) Chronological aging model (Pearson’s correlation coefficient r ¼ 0.86, p , 0.001***), (m) photoaging model (r ¼ 0.33).

APPLIED SPECTROSCOPY

679

FIG. 6. A Factor 2–Factor 3 score plot generated by PCA developed by using the 5990–5490 cm1 and 5000–4480 cm1 regions of the NIR-DR spectra of the phtoaging model and chronological aging model. (n) Photoaging model, 2 weeks irradiation, 8 weeks old, (m) photoaging model, 4 weeks irradiation, 10 weeks old, (u) photoaging model, 8 weeks irradiation, 14 weeks old, (n) photoaging model, 10 weeks irradiation, 16 weeks old. Chronological aging model ( ) 6 weeks old, (3) 10 weeks old, (¤) 14 weeks old, (§) 16 weeks old, and (þ) 27 weeks old. &&&

the amount of peak shift near 4880 cm1 and the increase in the peak intensity near 4590 cm1 are found between two models in the second-derivative (SD) NIR-DR spectra. According to the loading plot for PC2 (Fig. 7), peaks near 5930, 5790, and 5670 cm1 have great influence on classifying the NIR-DR data along PC2. These peaks reflect the remarkable changes in the spectra of the chronological aging model in the 2D NIR-DR spectra.

DISCUSSION The changes in the dermis with photoaging are tied to the collapse of dermal collagen fiber bundles and the degeneration of elastic fiber.5,6 The change in the dermis with chronological aging is the decrease of total collagen quantity.16 The intriguing score plot produced by PCA results from the specific processing; the 5990–5490 cm1 and 5000–4480 cm1 regions of the NIR-DR spectra are selected and subjected to PCA after the treatments of SNV and SD (Fig. 6). Along PC3 and PC2 in the score plot, we obtained the very interesting result that PC3 indicates the photodamage while PC2 reflects the chronological aging (Fig. 6). The peak shift near 4880 cm1, which is one of the key bands to divide the spectra along PC3 in the score plot, monitors a structural transition in proteins.13,14 Thus, significant differences in the degree of this peak shift are found for the spectra of the 10-week-old, 14week-old, and 16-week-old mice between the photoaging model and the chronological aging model (closed testing procedure, p , 0.001). When comparing the spectra of the 16week-old mice between the photoaging model and the chronological aging model, there is a remarkable difference in the circled places (Fig. 6). Therefore, it is suggested that PC3 reflects degeneration of the skin, which is a structural transition in dermal proteins with photoaging, for example, collapse of dermal collagen fiber bundles and degeneration of elastic fiber. On the other hand, the NIR-DR data from the chronological aging model may be classified in chronological order along PC2 (Fig. 6). The peaks, which contribute to the classification along PC2 in our work, are assigned to proteins or lipids.15 It

680

Volume 62, Number 6, 2008

FIG. 7. PC loading plots of PC2 and PC3 for the score plot in Fig. 6. (u) PC2, (m) PC3.

has been reported that the quantity of proteins decreases while that of lipids increases with aging. Therefore, it is suggested that PC2 reflects degradation of the skin, which is the decrease in the protein quantity in the dermis with chronological aging, for example, the decrease of collagen protein quantity with a decline in cell function/productivity.16 As for the increase in the quantity of lipids in the NIR-DR spectra, the details are unclear. It is, however, probable that the collagen quantity and density in the dermis is decreasing with age,17 and then, the NIR-DR spectra of the elder mice (16 and 27 weeks old) contain more information about subcutaneous fat under the dermis.

CONCLUSION We have developed a novel method for assessing skin aging based on NIR-DR spectroscopy and PCA. In this method, the photoaging and chronological aging can be differentiated by a score plot generated by PCA, which reflects the degeneration caused by the photoaging and the degradation caused by the chronological aging. The NIR-DR method detects changes not only in the quantity of skin components such as proteins and lipids, but also in their structures and is different from general measurement methods for physical properties of the skin. It is capable of capturing an aging state more precisely in situ by detecting changes in protein structures and of quantity inside the skin. This proposed method is a valuable tool for the investigations of skin aging. This method, which is simple, quick, and noninvasive, enables us to evaluate the consumer’s skin condition in greater detail so that we can make better recommendations for appropriate skincare. This approach will be very useful to the cosmetic industry and will better satisfy the needs of consumers. 1. Y. Nishimori, A. D. Pearse, C. Edwards, and R. Marks, Skin Res. Technol. 4, 79 (1998). 2. R. M. Lavker, Ed., Photodamage (Blackwell Science, Oxford, 1995), p. 122. 3. Y. Takema, Y. Yorimoto, Y. Kawai, and G. Imokawa, Brit. J. Dermatol. 131, 641 (1994). 4. P. Zheng, J. Invest. Dermatol. 100, 194 (1993). 5. Y. Nishimori, A. D. Pearse, C. Edwards, and R. Marks, J. Invest. Dermatol. 117, 1458 (2001).

6. G. Imokawa, Y. Takema, Y. Yorimoto, K. Tsukahara, M. Kawai, and S. Imayama, J. Invest. Dermatol. 105, 254 (1995). 7. S. Akazaki, H. Nakagawa, H. Kazama, O. Osanai, M. Kawai, Y. Takema, and G. Imokawa, Brit. J. Dermatol. 147, 689 (2002). 8. M. Gniadecka and G. B. Jemec, Brit. J. Dermatol. 139, 815 (1998). 9. N. Kollias, R. Gillies, M. Moran, I. E. Kochevar, and R. R. Anderson, J. Invest. Dermatol. 111, 776 (1998). 10. G. N. Stamatas, R. B. Estanislao, M. Suero, Z. S. Rivera, J. L. Khaiat, and N. Kollias, Brit. J. Dermatol. 154, 125 (2006). 11. K. Han, H. Choi, C. Won, J. Chung, K. Cho, H. Eun, and K. Kim, Mech. Ageing Dev. 126, 560 (2005). 12. B. Osborne, T. Fearn, and P. H. Hindle, Practical NIR Spectroscopy with

13. 14. 15. 16. 17.

Applications in Food and Beverage Analysis (Longman Scientific and Technical, Harlow, UK, 1993). Y. Liu, R. Cho, K. Sakurai, T. Miura, and Y. Ozaki, Appl. Spectrosc. 48, 1249 (1994). M. Miyazawa and M. Sonoyama, J. Near Infrared Spectrosc. 6, A253 (1998). N. Hirosawa, Y. Sakamoto, H. Katayama, and K. Yano, Anal. Biochem. 305, 156 (2002). J. A. Palka, W. Miltyk, and S. Wolczynski, Tokai J. Exp. Clin. Med. 21, 207 (1996). S. Shuster, M. Black, and E. Mcvitie, Brit. J. Dermatol. 93, 639 (1975).

APPLIED SPECTROSCOPY

681

Study of the Solid-Matrix Phosphorescence Properties of a Heterocyclic Aromatic Amine and the Heat Capacities of Glucose Glasses as the Temperature Decreases SARA E. HUBBARD and ROBERT J. HURTUBISE* Department of Chemistry, University of Wyoming, Laramie, Wyoming 82071

The heat capacities were obtained from 294 to 133 K for four glucose-glass systems. Two of the glasses were prepared from crystalline glucose. One of the glasses contained the heavy-atom salt NaI and the other glass did not contain NaI. The other two glasses were similar, but they were prepared from glucose melts. Correlations were developed between the solid-matrix phosphorescence (SMP) lifetimes and intensities of 2-amino-1-methyl-6phenylimidazo[4,5-b]pyridine (PhIP) in the glucose glasses and the heat capacities of the glucose sugar glasses as the temperature was lowered. Several plots of reciprocal SMP lifetime versus reciprocal temperature and reciprocal SMP lifetime versus reciprocal heat capacity were compared. Also, the reciprocal SMP intensity versus reciprocal temperature plots were compared with the corresponding reciprocal SMP intensity versus reciprocal heat capacity plots. In addition, basic photophysical equations were used to develop relationships among the lifetime data, the intensity data, and the heat capacity data. The heat capacity data and SMP lifetime data, obtained as the temperature was lowered, were discussed in relationship to low-frequency vibrational modes and brelaxation phenomena in the glucose glasses. The discussion of these phenomena offered explanations for some of the loss of the excited tripletstate energy of PhIP in the glucose sugar glasses. Index Headings: Solid-matrix phosphorescence; SMP; Heat capacity; Glucose glasses; Heavy-atom salt.

INTRODUCTION Glucose glasses have been shown to be effective solid matrices for obtaining the solid-matrix fluorescence (SMF) and phosphorescence (SMP) of heterocyclic aromatic amines (HAAs) at room temperature.1–4 Glucose forms clear, hard glasses that hold the lumiphors rigidly at room temperature, allowing for high SMF and SMP intensities and low limits of detection.1–5 Also, along with glucose, several new sugar glasses were recently investigated for their potential in solidmatrix luminescence.6 Sugar glasses also have other uses. Recently, Amorij et al.7 used sugar-glass technology to develop a stable influenza subunit vaccine in the dry state. Navati and Friedman8 used trehalose-derived glasses to assist in electron transfer processes over macroscopic distances. A novel use for the sugar amylose has been developed recently as an aid in dissolving nanotubes in aqueous solution.9,10 Little research has been done in studying the interactions of fluorophors and phosphors in sugar glasses. Mendonsa and Hurtubise11 obtained the SMP lifetimes for 2-amino-1-methyl6-phenylimidazo[4,5-b]pyridine (PhIP) as a function of temperature from 294 to 93 K in four different glucose-glass systems. Based upon the lifetime data, they were able to calculate two activation energies for each glucose-glass system. The smaller activation energy was related to low-frequency Received 17 November 2007; accepted 26 February 2008. * Author to whom correspondence should be sent. E-mail: hurtubis@ uwyo.edu.

682

Volume 62, Number 6, 2008

vibrations in the glucose glasses, and the larger activation energy was related to b-relaxation processes. Earlier, Hurtubise et al.12 developed a model for the nonradiative decay of organic phosphor triplet states in solid matrices using lifetime, temperature, and heat capacity data. They postulated that the low-frequency vibrational modes in the solid matrix could enhance the nonradiative transition from the excited triplet state to the ground state of an organic phosphor. Ludescher and coworkers13,14 have also studied the effects of temperature on the phosphorescence intensities and lifetimes of phosphors in several sugar glasses and polymer matrices to understand the influence of the molecular motions in solid matrices on phosphorescence. It is well known that the heat capacities of glasses decrease as the temperature is lowered.15–18 The decrease of temperature reduces the molecular motions within the glasses and they become increasingly rigid.17–20 Therefore, the heat capacity can be related to vibrational modes in the glasses. The heat capacities at constant pressure (Cp) of powdered solid matrices and the SMP lifetimes of adsorbed phosphors were examined by Hurtubise et al.12 over a large temperature range. Correlations between plots of ln(1/sp  1/s0p ) versus 1/T and versus 1/Cp were discussed, where sp is the SMP lifetime and s0p is the SMP lifetime that is independent of temperature. In this work, the heat capacities of the four glucose-glass systems prepared from crystalline glucose with and without the heavy-atom salt NaI and from the glucose melt with and without NaI were obtained from 294 to 133 K. The NaI enhanced the SMP intensity and shortened the SMP lifetime. Thus, in this work, a detailed comparison was made between glucose-glass systems that gave relatively strong SMP intensities and short SMP lifetimes with glucose glass systems that gave weaker SMP intensities and longer SMP lifetimes. The heat capacity data were then compared to the SMP lifetime data obtained earlier by Mendonsa and Hurtubise11 and to the SMP intensities of PhIP in these glasses.4 There have been no previous reports of the relationship between SMP lifetimes and intensities and the heat capacities of sugar glasses. However, it has been shown that the SMP lifetimes of adsorbed phosphors are related to the heat capacities of powdered solid matrices.12 In this work, graphs of ln(1/¯sp  1/¯s0p ) versus both 1/T and 1/Cp were acquired, where s¯ p is the average SMP lifetime, and s¯ 0p is the average lifetime that is independent of temperature. Also, graphs of ln(1/Ip  1/Ip0 ) versus both 1/T and 1/Cp were obtained, where Ip is the SMP intensity and Ip0 is the SMP intensity that is independent of temperature. PhIP was used as a model compound, as it was representative of heterocyclic aromatic amines (HAAs). HAAs are a class of carcinogenic/mutagenic compounds found in various foods.21,22 They have been linked to breast, liver, and intestinal cancers.23,24 PhIP is the most common food-based

0003-7028/08/6206-0682$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

HAA and has been previously used as a model compound in SMP sugar-glass studies.3,6,11,25

EXPERIMENTAL Instrumentation. Solid-matrix phosphorescence intensity and lifetime data were obtained previously by Mendonsa and Hurtubise4,11 with a Spex Fluorolog 2 spectrometer using DataMax version 4.09 software (Jobin Yvon, Inc., Edison, NJ). To acquire the steady-state SMP intensities for PhIP, a 150 W pulsed xenon lamp and a RS928 detector were used. The excitation slits were set at 5.0 mm and the emission slits were set at 3.0 mm. The delay time (td) was 1 ms, and the gate time (tg) was 10 ms. Because the excitation and emission wavelengths varied somewhat as a function of temperature, the excitation and emission wavelengths were the ones that corresponded to the maximum wavelengths at a given temperature. At room temperature, the excitation and emission wavelengths were 342 nm and 481 nm, respectively. The SMP excitation spectrum gave one broad band, as did the SMP emission spectrum. Typical SMP excitation and emission spectra for PhIP in glucose glasses at room temperature can be found in Ref. 2. The microwave oven used in this work for drying glasses was a Panasonic model NN-6475A domestic microwave oven with an output power of 925 W (Matsushita Electric Corporation of America, Secaucus, NJ). The differential scanning calorimeter (DSC) employed for acquiring the heat capacity data in this work was a TA 2920MDSC with TA Thermal Advantage (version 1.0 F) software, TA Universal Analysis (version 2.6 D) software, and a liquid nitrogen cooling accessory (TA Instruments, New Castle, DE). Reagents and Solutions. PhIP was purchased from Toronto Research Chemicals, Inc. (Ontario, Canada). D-(þ)-glucose (99.5%) was purchased from Sigma (St. Louis, MO). Sodium iodide (99.999%) was purchased from Alfa Aesar (Ward Hill, MA) and from Aldrich (Milwaukee, WI). HPLC-grade water and methanol were purchased from EMD Chemicals, Inc. (Gibbstown, NJ). Stock solutions of 200 lg/mL of PhIP in methanol:water (50:50) were used for the preparation of glasses formed from crystalline glucose, and 200 lg/mL of PhIP in methanol was used for glasses formed from the glucose melt. Preparation of Glasses. The glasses were prepared as discussed in previous work.6,11 For heat capacity measurements, mg amounts of the glasses were placed in aluminum sample pans (#02190062, Perkin Elmer) and sealed using a Volatile Sample Sealer (#02190061, Perkin Elmer). Heat Capacity. Several standards were used to calibrate the DSC. First, a baseline scan was performed using the experimental parameters described below. This was followed by several temperature standards using the same flow rate as the baseline scan: indium metal (m.p. ¼ 156.6 8C), water (0.0 8C), methylcyclohexane (126.6 8C), and toluene (95.0 8C). Indium was also used as a cell constant calibrant, to ensure that the change in heat capacity during melting was as expected. Sapphire (Al2O3) was used as a standard of known heat capacities to determine the values of the cell calibration constant at the temperatures of interest. These values were then used to determine the heat capacities of the glucose glasses as the temperature changed. All standards were run using the same experimental parameters as those used for the samples. For all experiments run on the DSC, both a pan containing the sample or a standard and an empty pan to serve as a reference were required. Before scans for heat capacity could

be performed on samples, blank pans were run so that the heat flow caused by the pans could be subtracted from the overall heat flow. This then gave the heat flow for the sample alone. Four blanks were run. For both blanks and samples, the instrument was equilibrated at 150 8C. This temperature was maintained for 5 minutes, and was then increased at a rate of 20 8C/min to 40 8C. The temperature was sustained for 2 minutes. The cell was purged with He gas at 25 mL/min.

RESULTS AND DISCUSSION Heat Capacities and Glass-Transition Temperatures of Glucose-Glass Systems. Very little work has been done in relating the heat capacity of the solid matrix to the SMP properties of the adsorbed phosphor. However, Hurtubise et al.12 acquired the heat capacities of several powdered solid matrices and correlated the heat capacity data to the SMP properties of various phosphors. They developed a model for the nonradiative decay of the organic phosphor triplet state in solid matrices. In the present work, the constant-pressure molar heat capacities (Cp) from 133 to 294 K were obtained for the four glucose-glass systems. These systems were PhIP in glucose glasses prepared from crystalline glucose with and without 10% NaI, and glucose glasses prepared from glucose melts with and without 10% NaI. The glucose glass systems containing phosphors have been discussed previously.4,11 The glass-transition temperature (Tg) values were obtained in this work for the four glucose-glass systems. The values were 284, 289, 292, and 291 K, respectively, for crystalline glucose, crystalline glucose–NaI, glucose melt, and glucose melt–NaI. The previous values are somewhat lower than the Tg value reported in the literature for glucose.26,27 This was most likely the result of residual water in the glasses. Methanol:water (50:50) was used in the preparation of the glasses from crystalline glucose.6 For the glasses prepared from the glucose melts, methanol was used as a solvent, but anhydrous conditions were not used in the preparation of the glasses. Also, NaI may have affected the Tg value of glasses containing this salt. The Cp values increased linearly with temperature until the glass-transition temperature (Tg) region of these glasses and then the Cp values changed more dramatically with temperature (Fig. 1). The increase in Cp from about 133 to 273 K in Fig. 1 is due to the glass becoming less rigid and more vibrational and rotational modes becoming active.18–20 The drastic increase in Cp in the Tg region, beginning at approximately 273 K, is because the glass became much more liquid-like and the vibrational and rotational modes became highly active. The rate of increase in Cp with temperature before the glass-transition temperature region was not the same for all of the glucose-glass systems. Also, the change in Cp in the glass-transition region was different from system to system. The values for the slopes of the linear portions of the plots of Cp versus temperature before the Tg region and the DCp values in the Tg region are listed in Table I. Though they were not the same, the slopes of the Cp versus T plots were of the same order of magnitude for all four of the glucose-glass systems. However, the DCp values in the Tg region showed much greater differences. Systems prepared from crystalline glucose exhibited DCp values that were more than three times greater than those for systems prepared from the glucose melt (Table I). For example, DCp in the Tg region of crystalline glucose without NaI was 221 J/molK, whereas the DCp value of the glucose melt without

APPLIED SPECTROSCOPY

683

TABLE I. Slopes of Cp versus temperature plots and DCp at Tg for glucose-glass systems with and without the heavy-atom salt NaI. Glucose-glass system Crystalline glucose no heavy atom Crystalline glucose þ 10% NaI Glucose melt, no heavy atom Glucose melt þ 10% NaI a b

Slope (J/molK2)a DCp at Tg (J/molK)b 1.14 1.17 1.37 1.02

6 6 6 6

0.05 0.14 0.04 0.05

221 6 8.5 238 69.8 65.9 6 14

The reproducibility is represented by the standard deviation of three trials. All DCp data were obtained in triplicate except for the glucose melt without NaI and crystalline glucose with 10% NaI. These samples were run in duplicate.

NaI was 69.8 J/molK. The large DCp values for crystalline glucose samples are most likely related to the crystalline nature of the samples. Samples with more crystalline qualities would be more ordered and would require more energy to undergo a glass transition. Solid-Matrix Phosphorescence Intensity and Heat Capacity. Figure 2 gives a plot of SMP intensity versus heat capacity for PhIP in crystalline glucose over a temperature range of 133 to 294 K. As shown in the figure, the heat capacity decreased from 536 J/molK at 294 K to 314 J/molK at 273 K. Thus, the heat capacity decreased by 222 units. Nevertheless, over the same temperature range the relative intensity only increased from 1.00 to 1.87. As the temperature was decreased further, from 273 to 253 K, the SMP intensity increased from 1.87 to 4.50, and the corresponding heat capacity changed from 314 to 273 (a decrease of 41 J/molK). The large change in heat capacity from 294 K to 273 K is the result of the glucose system going from the glass-transition region to a more rigid state. However, the glucose matrix at room temperature (294 K) is still rigid enough to permit relatively strong SMP intensity,2,3,11 but the large change in heat capacity from 294 to 273 K does not translate to a large change in SMP intensity. The SMP intensity increased from 1.87 to 11.5 when the temperature was dropped from 273 to

FIG. 1. The heat capacity of glasses formed from glucose-glass systems as the temperature was increased from 133 to 294 K. (A) Crystalline glucose without NaI, (B) crystalline glucose þ 10% NaI, (C) glucose melt without NaI, and (D) glucose melt þ 10% NaI. FIG. 2. Plot of SMP intensity versus Cp as the temperature was lowered for PhIP in crystalline glucose.

684

Volume 62, Number 6, 2008

133 K (Fig. 2). This corresponded to Cp values changing from 314 to 150 (a decrease of 164 J/molK). The decrease in heat capacity over this temperature range was considerably smaller than the decrease of Cp in the glass-transition region (222 J/molK). Figure 2 shows that there is approximately a linear increase in SMP intensity as Cp decreases in the temperature range of 273 to 133 K. This is related to the solid matrix becoming more rigid and several vibrational and rotational modes becoming inactive or less active. Even though the heat capacity decreased in the temperature range of 273 to 133 K, the magnitude of the decrease was less than from 294 to 273 K. Thus, the large increase in SMP intensity from 273 to 133 K cannot be completely explained by the decrease in heat capacity. Most likely, the rigidity of glasses is the main factor in increasing the SMP as the temperature is lowered. Similar plots of SMP intensity and Cp were obtained for the other three glucose-glass systems. However, the change in Cp from 294 to 273 was not as great with the glucose-melt systems (see Table I). Also, the same general patterns were obtained for the corresponding average SMP lifetime values versus Cp plots, which indicated that SMP lifetimes were affected in a similar fashion as SMP intensity. However, some small differences did appear between the relationships for SMP intensity and heat capacity and for SMP average lifetimes and heat capacity. Solid-Matrix Phosphorescence Lifetime and Heat Capacity. The average SMP lifetimes of PhIP in glucose-glass systems as a function of temperature have been previously discussed by Mendonsa and Hurtubise.4,11 The average SMP lifetime range for PhIP in the glucose glasses (crystalline and melt) without NaI ranged from 0.31 s at room temperature to 1.61 s at low temperature. For PhIP in glucose glasses (crystalline and melt) with 10% NaI, the average SMP lifetime range was from 13 to 48 ms.11 In this work, graphs of ln(1/¯sp  1/¯s0p ) versus 1/Cp for the four glucose-glass systems were obtained. Example plots of ln(1/¯sp  1/¯s0p ) versus 1/Cp for PhIP in glucose glasses are shown in Figs. 3 and 4 along with the corresponding ln(1/¯sp  1/¯s0p ) versus 1/T plots. It is clear by comparing Fig. 3A with Fig. 3B for the glucose crystalline system that the shapes of the graphs are similar, except for the glass-transition region (dashed line) in Fig. 3A. Not considering the glass-transition region, both Figs. 3A and 3B show two linear regions in each figure. In the respective regions, the changes in the slopes in both the 1/Cp and 1/T plots are related in that two lines can be obtained, one with a smaller slope and one with a larger slope. Also, the same conclusions can be made for Fig. 4A and Fig. 4B for PhIP in the glucose melt. However, because the DCp is much smaller in the glasstransition region for glucose melt, Fig. 4A shows that the heat capacity value obtained in this region could be used with the linear fit for the line having the larger slope. Comparison of graphs A and B in Figs. 3 and 4 indicates that there are most likely fundamental processes that underlie the heat capacities of solid matrices and the decay of organic phosphors from the excited triplet state as the temperature is lowered. Earlier, Hurtubise et al.12 developed a model correlating the bulk heat capacities of solids, which are a measure of thermally excited vibrational states in the bulk solid, to the nonradiative deactivation of the excited triplet state of a phosphor adsorbed in the solids. The nonradiative deactivation occurred by the coupling of thermally activated solid matrix vibrations with the excited triplet state of the

FIG. 3. (A) Plot of ln(1/¯sp  1/¯s0p ) versus 1/Cp for PhIP in crystalline glucose without NaI. (B) Plot of ln(1/¯sp  1/¯s0p ) versus 1/T for PhIP in crystalline glucose without NaI. Data taken from Ref. 11.

phosphor. They related the phosphorescence lifetimes of organic phosphors adsorbed on a variety of powdered solid matrices to the change in the heat capacities of the solid matrices as the temperature decreased. The phosphorescence lifetimes they obtained were monoexponential. In their study, linear relationships were obtained for all the ln(1/sp  1/s0p ) versus 1/Cp and ln(1/sp  1/s0p ) versus 1/T plots.12 This is in contrast to the graphs in Figs. 3 and 4 that show the data fit to two lines. Thus, the situation with glucose glasses is more complex. Mendonsa and Hurtubise11 postulated previously that the line with the smaller slope (lower temperature region) in Fig. 3B was related to the nonradiative deactivation of PhIP in the excited triplet state by the coupling of the low-frequency

APPLIED SPECTROSCOPY

685

vibrational modes in the glucose to the excited triplet state of the phosphor. Figure 3A in this work offers support for this because of how ln(1/¯sp  1/¯s0p ) changes with 1/Cp in the 1/Cp region of 0.0045 to 0.007 molK/J and how ln(1/¯sp  1/¯s0p ) changes with 1/T in the 1/T region of 0.005 to 0.008 K1 in Fig. 3B. For the line with the larger slope (higher temperature region) in Fig. 3B, Mendonsa and Hurtubise11 related this region to b-relaxation phenomena in the glucose glasses. This conclusion was based on the activation energies obtained from SMP lifetimes reported in Ref. 11, in the temperature range of 233 to 294 K, and the activation energies reported by others for b-relaxation in glucose.28–33 The same general conclusions can be made for PhIP with data acquired for the glucose melts (see Figs. 4A and 4B) as were made for the data in Figs. 3A and 3B. b-Relaxation is not fully understood. It occurs below the glass transition temperature, and some researchers believe it is related to the motion of functional groups in sugar glasses.30 In the earlier work by Hurtubise et al.,12 the activation energies they calculated from the SMP lifetime data were in the range of 332 to 488 cm1 for phosphors adsorbed on powders. Only one activation energy was obtained in the earlier work for each solid-matrix system because of the single lines obtained from the correlation of the SMP lifetime and heat capacity data as the temperature was lowered.12 However, Mendonsa and Hurtubise11 calculated two activation energies for the glucoseglass systems based on SMP lifetime data using ln(1/¯sp  1/¯s0p ) versus 1/T plots (Fig. 3B and Fig. 4B). Also, it was determined previously that the nonradiative rate constant (km) for intersystem crossing from the excited triplet state (T1) to the ground state (S0) for PhIP in glucose glasses is given by Eq. 1:25 I km ¼ km þ k1 eEa1 =RT þ k2 eEa2 =RT

FIG. 4. (A) Plot of ln(1/¯sp  1/¯s0p ) versus 1/Cp for PhIP in glucose melt without NaI. (B) Plot of ln(1/¯sp  1/¯s0p ) versus 1/T for PhIP in glucose melt without NaI. Data taken from Ref. 11.

I where km is the nonradiative rate constant from T1 to S0 when the exponential terms are negligible, k1 and k2 are preexponential factors, Ea1 and Ea2 are activation energies, R is the gas constant, and T is temperature in Kelvin. The details for the calculation of Ea1 and Ea2 values from SMP lifetime data have been discussed in the literature.11,25 As shown in Table II, the activation energies for the lower temperature region (see Figs. 3B and 4B) (Ea1)s, were in the range of 364 to 454 cm1. Thus, the similarity of the activation energies for phosphors on powders (range of 332 to 488 cm1) and in glucose glasses indicates a common mechanism. Most likely, low-frequency vibrational modes in the glucose matrices couple with the excited triplet state, which results in the increased probability of the nonradiative deactivation of the excited triplet state. The (Ea2)s values in Table II are related to b-relaxation

TABLE II. The activation energies (cm1) determined from SMP lifetime and intensity data as a function of temperature.a Glasses from crystalline glucose No NaI

Glasses from the glucose melt

10% NaI b,c

454 6 51 3658 6 432d

e

364 6 22 (230 6 35) 2473 6 61 (1981 6 329)

a

No NaI

10% NaI

409 6 16 3036 6 172

434 6 12 (495 6 28) 3025 6 391 (2730 6 463)

The reproducibility of the data are shown by the standard deviation. Activation energies from lifetime data were obtained from Ref. 11. c The numbers in the top row are the activation energies from the long-component lifetime data, (Ea1)s. d The numbers in the bottom row are the activation energies from the short-component lifetime data, (Ea2)s. e The numbers in parentheses represent (Ea1)I and (Ea2)I that were obtained from SMP intensity data for systems containing NaI. b

686

Volume 62, Number 6, 2008

ð1Þ

phenomena. The (Ea2)s values were obtained from the higher temperature regions (see Figs. 3B and 4B). Some researchers believe that b-relaxation in glucose is caused by the motion of the pendant hydroxyl methyl group in glucose.30 However, brelaxation in glucose most likely is more complex.31 Because of the similarities of the respective reciprocal lifetime versus heat capacity and temperature plots for the regions with greater slopes in Figs. 3A and 3B and the plots in Figs. 4A and 4B, added evidence is provided that b-relaxation can be involved in contributing to the deactivation of the excited triplet state of a phosphor in the glucose glasses. In the higher temperature region, group vibrations would be very active.19,20 They most likely play a role in contributing to the nonradiative deactivation of the excited triplet state. It would be difficult to sort out specifically which vibrational modes in the glucose solid matrices could interact with the excited triplet state of a phosphor. However, researchers have related bulk heat capacities to vibrational modes in solids. Wunderlich and co-workers18,34,35 have developed relationships for correlating heat capacities of bulk solid organic compounds and solid organic macromolecules to various vibrational modes in the solids as the temperature is altered. Also, Pyda19 and Pyda and Wunderlich20 used the Cp values of a-D-glucose, which were obtained from 8 to 490 K, to calculate the contribution from the group vibrations and the contribution from skeletal vibrations to the heat capacity over this temperature range. At room temperature, there were major contributions from both group and skeletal vibrations to the heat capacity. As the temperature was lowered, there was a significant decrease in the group vibrations and only a gradual decrease in the skeletal vibrations. At about 100 K, there was essentially no contribution of the group vibrations to heat capacity, and skeletal vibrations dominated the heat capacity. At room temperature, the group vibrations contributed about 30% to the heat capacity and skeletal vibrations contributed about 70% to the heat capacity of glucose.19,20 More work is needed to discover what vibrational activity is involved in glucose glasses that causes some of the loss in SMP radiation via nonradiative deactivation of the excited triplet state. However, because group vibrations are very active at room temperature, they could be contributing to the nonradiative loss of energy via b-relaxation. Solid-Matrix Phosphorescence Intensity and Temperature. It is of interest to consider in more detail the change in SMP intensity with temperature. The SMP intensities (Ip) of PhIP in the four glucose-glass systems were previously obtained.4,11 In this work, Ip0 values for PhIP in each of the glucose-glass systems were calculated in the same fashion as discussed previously by Mendonsa and Hurtubise for s¯ 0p values.11 The value of Ip0 is the low-temperature limit of the SMP intensity, or the temperature at which the intensity becomes independent of temperature. In most of the glucoseglass systems, a constant SMP intensity was not obtained experimentally, although the intensities approached constant values. If the SMP intensity did not reach a constant value, then the Ip0 value was obtained by arbitrarily selecting values of Ip0 so that a plot of 1/Ip  1/Ip0 versus 1/T gave the best straight line.11,25 Over the past several years, relationships between the SMP lifetimes and temperature for a variety of phosphors in different solid matrices have been reported via plots of ln(1/¯sp  1/¯s0p ) versus 1/T11 and of ln(1/sp  1/s0p ) versus 1/T.12,36,37

FIG. 5. Plot of (1/Ip  1/Ip0 ) versus 1/T for PhIP in crystalline glucose with 10% NaI.

However, there have been no previous reports of ln(1/Ip  1/Ip0 ) as a function of 1/T for phosphors in the glucose-glass systems. The experimental values of Ip and Ip0 obtained earlier were used to prepare plots of ln(1/Ip  1/Ip0 ) versus 1/T. Figure 5 shows an example plot of ln(1/Ip  1/Ip0 ) versus 1/T for PhIP in crystalline glucose with 10% NaI. As shown in Fig. 5, the plot of ln(1/Ip  1/Ip0 ) versus 1/T was fit to two straight lines. Also, the reciprocal intensity data for PhIP in glucose melts with 10% NaI gave two lines. Interestingly, the reciprocal intensity plots gave single lines for the glucose systems without NaI. It is important to consider the relationship between Ip and s¯ p and other fundamental luminescence parameters. Equation 2 relates Ip to a number of parameters: Ip ¼ DIUt kp s¯ p

ð2Þ

Ip is the SMP intensity, DI is the rate of absorption (photons/s), Ut is the intersystem crossing yield, kp is the rate constant for phosphorescence, which is constant with temperature, and s¯ p is the average phosphorescence lifetime. Mendonsa and Hurtubise11 showed that the Ip/¯sp ratios increased as temperature was lowered for glasses prepared from crystalline glucose and glucose melts without NaI present. (However, at about 150 K the Ip/¯sp ratio decreased somewhat for both the crystalline glucose and glucose melt.) For the two types of glucose glasses containing NaI, the Ip/¯sp ratios were essentially constant with temperature. Thus, for the systems without NaI, Ip is a function of both Ut and s¯ p because these parameters would change with temperature. On the other hand, for the systems containing NaI, Ip is only a function of s¯ p, because Ut and kp are independent of temperature (see Eq. 2). Since Ut is constant for glucose-glass systems containing NaI, Ut is equal to U0t , where U0t is independent of temperature. Therefore, Eq. 2 can be used to obtain Eq. 3, where K is a constant and s¯ 0p is independent of temperature.

APPLIED SPECTROSCOPY

687



1 1  Ip Ip0

 ¼

  1 1 1  0 K s¯ p s¯ p

ð3Þ

Equation 3 shows that for the glucose-glass systems containing NaI, (1/Ip  1/Ip0 ) is directly proportional to (1/¯sp  1/¯s0p ) as the temperature changes. Equation 3 is important because it shows that the same basic information can be obtained from the reciprocal intensity and reciprocal lifetime plots for systems containing NaI. In fact, essentially the same Ea1 and Ea2 values were obtained from the reciprocal intensity plots as for reciprocal lifetime plots for glasses with 10% NaI (Table II). However, the activation energies obtained with the SMP intensity data were somewhat lower except for the Ea1 value for PhIP in the glucose melt with 10% NaI. Things are more complicated when the temperature is lowered and Ut is not equal to U0t . This occurred when NaI was not used as a heavy-atom salt. Equation 2 shows that Ip is proportional to Uts¯ p. This indicates that there is a more complicated relationship for glucose glasses without NaI compared to Ip from glucose glasses containing NaI. Thus, the interpretation of the plots of ln(1/Ip  1/Ip0 ) versus 1/T for glucose glasses without NaI are complex and beyond the scope of this work. As mentioned earlier, the reciprocal intensity plots gave single lines for PhIP in the glucose systems without NaI. In general, plots of ln(1/¯sp  1/¯s0p ) versus 1/T are easier to interpret because they are directly related to the excited triplet state and one is directly dealing only with the processes from the excited triplet state. Also, SMP lifetimes are less prone to changes in experimental conditions compared to SMP intensities. Thus, the lifetime plots were emphasized in this work.

CONCLUSION Solid-matrix phosphorescence lifetime data were correlated with the glass-transition region and heat capacities for glucose sugar glasses in the temperature range of 294 to 133 K. Glucose glasses containing PhIP with and without NaI were investigated. In general, the SMP lifetime data plotted as reciprocal lifetime versus the reciprocal of temperature or versus the reciprocal of heat capacity showed two linear regions. The linear region at low temperatures was related to low-frequency vibrations of sugar glasses and the higher temperature region was correlated to b-relaxation in the sugar glasses. This is only the second study in which heat capacity data of the solid matrix were linked to SMP data. Earlier research showed that phosphors adsorbed on powders gave only one linear region for similar SMP lifetime plots versus the reciprocal of temperature or versus the reciprocal of heat capacity. In this work, basic relationships were developed between SMP lifetime data and the heat capacity data that explained some of the loss of the energy in the excited triplet state by the nonradiative transition from the excited triplet state. SMP intensity data were also related to heat capacity and temperature changes. However, the SMP lifetime data gave more useful correlations. The general approach discussed in

688

Volume 62, Number 6, 2008

this paper of employing bulk heat capacity of the solid matrix to investigate SMP phenomena could be combined with infrared spectrometry to investigate in more detail specifically what vibrational modes couple with the excited triplet state of the phosphor. ACKNOWLEDGMENTS This work was supported by Chemical Sciences, Geosciences and Biosciences Division, Offices of Basic Energy Sciences, U.S. Department of Energy (DE-FG02-04ER15545). We thank Jeff Yarger for the use of the TA 2920MDSC and sample sealer. We also thank Shaun Mendonsa for acquiring the SMP intensity and lifetime data as a function of temperature. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 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.

J. Wang and R. J. Hurtubise, Appl. Spectrosc. 50, 53 (1996). J. Wang and R. J. Hurtubise, Anal. Chim. Acta 332, 299 (1996). S. Mendonsa and R. J. Hurtubise, Appl. Spectrosc. 55, 1385 (2001). S. Mendonsa, Ph.D. Thesis. University of Wyoming, Laramie, Wyoming (2001). S. Mendonsa and R. J. Hurtubise, Appl. Spectrosc. 54, 456 (2000). S. E. Hubbard and R. J. Hurtubise, Talanta 72, 132 (2007). J.-P. Amorij, J. Meulenaar, W. L. J. Hinrichs, T. Stegmann, A. Huckriede, F. Coenen, and H. W. Frijlink, Vaccine 25, 6447 (2007). M. S. Navati and J. M. Friedman, J. Biol. Chem. 281, 36021 (2006). R. Dagani, Chem. Eng. News, 38 (2002). A. Star, D. W. Steuerman, J. R. Heath, and J. F. Stoddart, Angew. Chem. Int. Ed. 41, 2508 (2002). S. Mendonsa and R. J. Hurtubise, J. Lumin. 97, 19 (2002). R. J. Hurtubise, S. M. Ramasamy, J. Boerio-Goates, and R. Putnam, J. Lumin. 68, 55 (1996). S. Shirke and R. D. Ludescher, Carbohydr. Res. 340, 2654 (2005). Y. You and R. D. Ludescher, Appl. Spectrosc. 60, 813 (2006). G. S. Parks and S. B. Thomas, J. Am. Chem. Soc. 56, 1423 (1934). J. Boerio-Goates, J. Chem. Therm. 23, 403 (1991). B. Wunderlich, J. Phys. Chem. 64, 1052 (1960). B. Wunderlich, Thermal Analysis (Academic Press, San Diego, CA, 1990). M. Pyda, in The Nature of Biological Systems as Revealed by Thermal Methods (Springer, New York, 2004), vol. 5, p. 307. M. Pyda and B. Wunderlich, Proceedings of the 29th Annual Conference of the North American Thermal Analysis Society; Sept. 24–26, 2001, St. Louis, MO 29, 76 (2001). S. Manabe, H. Suzuki, O. Wada, and A. Ueki, Carcinogenesis 14, 899 (1993). P. Pais, C. P. Salmon, M. G. Knize, and J. S. Felton, J. Agric. Food Chem. 47, 1098 (1999). M. Murkovic, Eur. J. Lipid Sci. Technol. 106, 777 (2004). K. M. Gorlewska-Roberts, C. H. Teitel, J. O. J. Lay, D. W. Roberts, and F. F. Kadlubar, Chem. Res. Toxicol. 17, 1659 (2004). J. Wang and R. J. Hurtubise, Anal. Chem. 69, 1946 (1997). P. D. Orford, R. Parker, and S. G. Ring, Carbohydr. Res. 196, 11 (1990). Y. Roos, Carbohydr. Res. 238, 39 (1993). R. K. Chan, K. Pathmanathan, and G. P. Johari, J. Phys. Chem. 90, 6358 (1986). T. R. Noel, S. G. Ring, and M. A. Whittam, J. Phys. Chem. 96, 5662 (1992). T. R. Noel, R. Parker, and S. G. Ring, Carbohydr. Res. 282, 193 (1996). D. van Dusschoten, U. Tracht, A. Heuer, and H. W. Spiess, J. Phys. Chem. A 103, 8359 (1999). G. R. Moran, K. R. Jeffrey, J. M. Thomas, and J. R. Stevens, Carbohydr. Res. 328, 573 (2000). T. R. Noel, R. Parker, and S. G. Ring, Carbohydr. Res. 329, 839 (2000). H. S. Bu, S. Z. D. Cheng, and B. Wunderlich, J. Phys. Chem. 91, 4179 (1987). Y. Jin and B. Wunderlich, J. Phys. Chem. 95, 9000 (1991). S. M. Ramasamy and R. J. Hurtubise, Anal. Chem. 59, 432 (1987). S. M. Ramasamy and R. J. Hurtubise, Talanta 36, 315 (1989).

Method for the Estimation of the Mean Lorentzian Bandwidth in Spectra Composed of an Unknown Number of Highly Overlapped Bands ´ RENZ-FONFRI´A* and ESTEVE PADRO ´S VI´CTOR A. LO Unitat de Biofı´sica. Departament de Bioquı´mica i de Biologia Molecular. Facultat de Medicina, and Centre d’Estudis en Biofı´sica, Universitat Auto`noma de Barcelona, 08193 Bellaterra, Barcelona, Spain

We introduce a method for the estimation of the mean Lorentzian bandwidth of the component bands in a spectrum. The method is computationally simple, using only the module of the Fourier transform of the spectrum, and its first derivative. Moreover, the presented method does not require knowledge of the number of bands in the spectrum, their band positions, or their band areas. Furthermore, it works on spectra containing Lorentzian bands, as well as Gaussian and Voigtian bands. Therefore, the introduced method seems especially well suited for obtaining a representative Lorentzian width for highly overlapped bands, independent of their number and Lorentzian/Gaussian character. We describe how different experimental limitations (spectral truncation, offset error, presence of noise, etc.) may affect the performance of the method, and when required we propose effective alternatives to minimize their effects. Finally, we show the application of the method to an experimental spectrum: the amide I band of a dry film of the solubilized ADP/ATP carrier. The estimation of the mean Lorentzian width can allow, for instance, for a more objective selection of the deconvolution width in Fourier self-deconvolution, allowing for a more objective and reliable analysis of the amide I band of proteins. The mean Lorentzian width can also be useful to obtain an estimation of the homogenous broadening and vibrational relaxation of the amide I vibration of proteins, without requiring complex pump-probe experiments. Index Headings: Bandwidth estimation; Spectral quantification; Infrared spectroscopy; Amide I; Proteins; ADP/ATP carrier; Fourier transform; Fourier deconvolution; Vibrational relaxation; Homogenous broadening.

INTRODUCTION The analysis of overlapped bands is a common problem in spectroscopy.1,2 One relevant example is the amide I vibration of proteins, with corresponding bands in the infrared distributed between 1700 and 1615 cm1.3–6 In spite of its structure sensitivity, the amide I of proteins appears as a featureless envelope showing only one maximum, due to the high overlap of component bands. The analysis of overlapped bands in vibrational spectroscopy is often assisted by band-narrowing methods,7–15 curve-fitting procedures,16–19 or a combination of both.19–24 The former provides more resolved spectra by narrowing the component bands, while the latter potentially allows the estimation of their band parameters (positions, widths, and areas). Band deconvolution in the Fourier domain is probably the most commonly used band-narrowing method in research areas where band overlap is severe, whereas the use of the second (and higher) derivatives is the most widespread method in less specialized applications. Derivation does not preserve the band Received 29 November 2007; accepted 29 February 2008. * Author to whom correspondence should be sent. E-mail: victor.lorenz@ gmail.com. Present address: Department of Materials Science and Engineering, Nagoya Institute of Technology, Showa-ku, Nagoya 4668555, Japan.

Volume 62, Number 6, 2008

areas, the target band-narrowing is less easily controlled, and it provides distorted band shapes (with intense side lobes).25 As a benefit, derivation becomes very effective to minimize the contribution of baseline errors.24 In contrast to derivation, Fourier deconvolution preserves the band areas, the target band-narrowing can be specified (by the so-called narrowing factor), and band-narrowing can be obtained while preserving a band appearance.25 As a drawback, deconvolution requires as an input the deconvolution band shape and its bandwidth. For optimal performance the deconvolution bandwidth should be close to the actual widths of the bands in the spectra; otherwise the obtained solution can show lower band narrowing and higher noise enhancement than possible (mostly when the width is infra-estimated), or a distorted band shape (when the width is over-estimated).26 Because band overlap precludes direct observation of the component bands, the selection of the deconvolution bandwidth becomes arbitrary and user-dependent, raising many questions and doubts about the objectivity of any analysis based on deconvolution.6 Indeed, the scientific literature shows how different authors have deconvolved the amide I band of proteins using very different deconvolution Lorentzian widths, namely: 4–5 cm1,27,28 7 cm1,29 13–14 cm1,3,30 17–18 cm1,31,32 20 cm1,33 25 cm1,34 30 cm1,5 or 40 cm1.35 Therefore, a method providing a model-free estimation of the bandwidth in overlapped spectra would improve the objectivity in the selection of the deconvolution width in Fourier deconvolution, solving one of its weaker and more criticized points. Curve fitting requires the initial user specification of the number and shape of the bands, as well as initial values for the fitting parameters: band positions, areas, and widths (and any parameter controlling the band shape, if present). The band shape is usually selected on theoretical or practical grounds, and basically restricted to Lorentzian, Gaussian, or Voigtian band shapes. The number and initial position of the bands is usually obtained after a band-narrowing method is applied to the spectrum. The problem of providing good initial values for the bandwidths is rarely discussed in the literature,17 but being a nonlinear minimization with several potential local minima, using reasonable initial values for the bandwidths would help convergence to the physical minimum. Moreover, recent results show how probabilistic constrains in both the band positions and the bandwidths improve substantially the curvefitting robustness to systematic errors in the model and/or in the data.19 Therefore, a method able to provide a reasonable a priori value for the bandwidths would also be valuable in curve fitting, increasing both the chances of convergence to a physical minimum and the robustness of the solution. Here we present a method able to provide the mean Lorentzian width of the component bands in a spectrum. First, we describe the theoretical basis of the method and then

0003-7028/08/6206-0689$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

689

illustrate how it works under ideal conditions. Then, we introduce one by one some non-idealities present in realistic conditions; we describe how they can affect the reliability of the method, and we consider approaches minimizing their effect when necessary. Once the method is ready to work under non-ideal conditions, we present the analysis of experimental data: the amide I band of a membrane protein, providing for the first time an objective model-free estimation of the Lorentzian width of the underlying sub-component in the amide I band of a protein. In the last sections, we briefly comment on different areas where the introduced method could be of valuable help. Then, we summarize the characteristics and limitations of the introduced method, also in relation with previous methods to estimate the bandwidth in overlapped spectra.

We now define a new function, proportional to the ratio of the interferogram module first derivative and the interferogram module: c¯ L ðxÞ ¼ 

N X

Symbolic manipulations were performed in Derive v6. Numerical analysis was implemented in home-made routines working in Matlab v7.

THEORY A spectrum composed of N Lorentzian bands is given by EðvÞ ¼

N X i¼1

2Ai cL;i h i p c2L;i þ 4ðv  v0;i Þ2

ð1Þ

where A, cL, and v0 represent the band area, the Lorentzian width, and the band position, respectively. The Fourier transform of such a spectrum is given by EðxÞ ¼

N X

Ai expðpcL;i jxjÞexpði2pxv0;i Þ

ð2Þ

i¼1

ð5Þ

We named this function the mean Lorentzian width function, or Lorentzian width function for short, since when x tends to þ0 it gives the mean Lorentzian width of the bands in the spectrum (weighted by their area):

c¯ L ðxÞx!þ0

METHODS

1 ]jEðxÞj 3 p 3 jEðxÞj ]x

Ai cL;i   1 ]jEðxÞj i¼1  ¼ 3 ¼  N  p 3 jEðxÞjx¼0 ]x x!þ0 X  Ai    i¼1  ð6Þ

Note how Eq. 6 gives the mean Lorentzian width without requiring any prior knowledge or estimation of the number of component bands in a spectrum, their position, or their individual areas. Therefore, the method seems applicable to spectra with highly overlapped bands, where the true number of bands, their positions, and their areas may be impossible to estimate. The proposed method remains valid not only for Lorentzian bands, but also for Gaussian and Voigtian bands. For a spectrum made of N Gaussian bands: 9 "  8  N 2 # > X pc x >  > G;i > > pffiffiffiffiffiffiffi expði2pv ] A  i exp  0;i xÞ> >   = <  > 2 ln2 ]jEðxÞj i¼1 ¼ > ]x ]x x!þ0 > > > > > > > ; :

x¼0

A detailed and practical explanation of Fourier transform properties can be found elsewhere.36–38 From this point on, we will use the names spectrum and interferogram when referring to E(v) and E(x), respectively, using the nomenclature common to Fourier transform infrared (FT-IR) spectroscopy. Independent of the number of bands, their position, or width (or actual band shape), the value of the interferogram at x ¼ 0 gives the total band area. This holds also for the interferogram absolute value (interferogram module):    X N   ð3Þ Ai  jEðxÞjx¼0 ¼   i¼1  Interestingly, we found that for a spectrum made of Lorentzian bands, the first derivative of the interferogram module depends only on the product of the band areas and the bandwidths, as x tends to zero: 3 2  N   X Ai expðpcL;i jxjÞexpði2pv0;i xÞ7 6]   7 6 i¼1 ]jEðxÞj 7 ¼6 6 7 ]x ]x 4 5 x!þ0 ¼

N X

x!þ0

ð4Þ

Volume 62, Number 6, 2008

ð7Þ

and applying Eq. 6 we correctly arrive at the prediction that the mean Lorentzian width is zero. For a spectrum composed of N Voigtian bands:   ]jEðxÞj ]x x!þ0 9 "  8  N  # > X pcG;i x 2 > >  > > p ffiffiffiffiffiffi ffi ] A expðpc jxjÞexp  xÞ expði2pv >  i 0;i > L;i = <  > 2 ln2 i¼1 ¼ > > ]x > > > > > > ; :

x!þ0

¼

N X

Ai pcL;i

ð8Þ

i¼1

and the mean Lorentzian width of the Voigtian bands can be correctly obtained from Eq. 6. Therefore, the proposed method is able to provide the mean Lorentzian width independent of the Gaussian character present in the bands (i.e., it is valid for Lorentzian, Gaussian, and Voigtian bands).

PERFORMANCE UNDER IDEAL CONDITIONS

Ai pcL;i

i¼1

690

¼0

To visually illustrate the performance of the method, we apply it to several test spectra. First, we consider spectra made

FIG. 1. Illustration of the performance of the method for a single band. (A) Lorentzian band of 20 cm1 width, (B) Gaussian band of 20 cm1 width, and (C) Voigtian band with Lorentzian and Gaussian component widths of 20 cm1 and 10 cm1. (D–F) Interferogram module for the ideal (gray lines) and truncated (black lines) spectra. (G–I) First derivative of the interferogram module for the ideal (gray lines) and truncated (black lines) spectra. (J–L) Lorentzian width function for the ideal (gray lines) and truncated (black lines) spectra. The open circles correspond to the errorless points for the truncated data (see text for more details).

of a single band with a Lorentzian, Gaussian, or Voigtian band shape (Figs. 1A–1C), conceptually extending from minus to plus infinity. The second row in Fig. 1 shows in gray the corresponding ideal interferogram modules (Figs. 1D–1F), and the third row shows in gray the first derivative of the ideal interferogram module (Figs. 1G–1I). The ratio of the data in the third and second rows gives, after appropriate scaling (see Eq. 5), the ideal Lorentzian width function (see Figs. 1J–1L, gray lines), with the property that its value at the origin (x ¼ 0) gives the mean Lorentzian width of the bands in the spectrum. The method correctly predicts the Lorentzian width of the band in the input spectra, which was 20, 0, and 20 cm1 for the Lorentzian, Gaussian, and Voigtian bands, respectively. As a second test problem, we consider highly overlapped spectra made of nine Lorentzian, Gaussian, or Voigtian subcomponent bands with variable bandwidths (Figs. 2A–2C, respectively), conceptually extending from minus to plus

infinity. The band parameters used to generate these spectra are given in the corresponding figure legend. The corresponding ideal Lorentzian width functions are displayed in the bottom row (Figs. 2D–2F, gray lines), correctly predicting the mean Lorentzian width at zero distance, which was 21.91, 0, and 21.91 cm1 for the spectra composed of Lorentzian, Gaussian, and Voigtian bands, respectively.

DISTORTIONS CAUSED BY SPECTRAL TRUNCATION The expressions for Lorentzian, Gaussian, and Voigtian bands in the Fourier domain given in the Theory section correspond to an idealized spectrum going from minus infinity to plus infinity. In practice, however, spectra are truncated, containing data only in a limited wavenumber interval. Moreover, the fast Fourier transform adds zeros until the data

APPLIED SPECTROSCOPY

691

FIG. 2. Illustration of the performance of the method for highly overlapped bands. (A–C) Spectra made of nine highly overlapped bands, with positions at 1695, 1685, 1675, 1670, 1662, 1654, 1650, 1641, 1635, and 1625 cm1; relative areas (in %) of 2, 5, 10, 8, 15, 20, 10, 15, 10, and 5, respectively; and (A) Lorentzian, (B) Gaussian, or (C) Lorentzian component widths of 16, 25, 22, 18, 18, 18, 30, 30, 20, and 18 cm1, respectively; and (C) a Gaussian component width of 5, 10, 8, 10, 5, 0, 5, 15, 5, and 10 cm1, respectively. (D–F). Lorentzian width function for the ideal (gray lines) and truncated (black lines) spectra. The open circles correspond to the error-free points for the truncated data (see text for more details).

has the appropriate length. Under these conditions, a truncated spectrum can be regarded as an ideal one multiplied with a Box function, a function equal to one inside the data interval and zero otherwise. From the Fourier transform convolution theorem and shift theorem, we can arrive at an expression that relates the interferogram obtained from the truncated data, E(x)trun, and the ideal interferogram, E(x)ideal: Z vm þ0:5Dv EðxÞtrun ¼ EðvÞexpði2pxvÞdv vm 0:5Dv

¼ EðxÞideal  ½Dv 3 sincðpDvxÞ 3 expði2pvm xÞ ð9Þ where vm is the mean wavenumber of the spectral interval, and Dv is the length of the spectral interval. The distortions will depend on how the truncation is performed (Dv and vm) as well as on the form of the ideal interferogram itself. The distortions are illustrated below using several test cases. Figure 1A shows spectra containing a single Lorentzian band, which we will consider now to be truncated between 1450 and 1850 cm1 (vm ¼ 1650 cm1 and Dv ¼ 400 cm1). The interferogram module for the truncated spectra (Fig. 1D, black line) shows only minor distortions with respect to the ideal one (gray line), except at a very short distance, where some tiny oscillatory deviations are apparent. These oscillatory deviations are enhanced in the first derivative of the module (see Fig. 1G, black line) and transmitted to the Lorentzian width function (see Fig. 1J, black line). Note, however, how the Lorentzian width function obtained from the truncated spectra oscillates around the ideal solution, making it possible to estimate the mean Lorentzian width in the spectra by

692

Volume 62, Number 6, 2008

visually extrapolating the data to zero retardation. This possibility is further discussed in the following section. Very different behavior was observed for the truncated Gaussian band (Fig. 1B). There were no significant differences in the Fourier domain (Figs. 1E, 1H, and 1K) between the truncated (black line) and the ideal data (gray line, hidden behind the black line). The reason is easy to rationalize. The Gaussian band shape has very small wings, with values tending fast to zero as we separate from the band maximum. Therefore, data truncation introduces no significant changes unless performed close to the band maximum. This also explains why the distortions on truncation are so high for a Lorentzian band, which shows intense wings. It also suggests that any spectral preprocessing reducing the intensity of the band wings before data truncation will reduce the distortions (e.g., deconvolution). This possibility will be explored with success later in the paper. For the truncated Voigtian band (Fig. 1 right column of panels), there are also substantial differences between the ideal results and the results obtained with the truncated data. Again, the spectral truncation introduces oscillations in the Fourier domain, and the truncated data (black lines) oscillates around the ideal ones (gray lines). Incidentally, note how for both the truncated Lorentzian and Voigtian bands (Figs. 1A and 1C) the first derivative of the module goes to zero (Figs. 1G and 1I, black line). Previously, these two band shapes have been criticized as non-realistic band models for IR spectroscopy from the observation that interferograms calculated from experimental spectra showed a zero derivative at zero retardation, in contrast to the theoretical expectation for an ideal Lorentzian or Voigtian band (see gray

lines in Figs. 1I and 1G). Our data clearly shows how, for truncated Lorentzian and Voigtian bands, the derivative of the module goes to zero at zero retardation (see black lines in Figs. 1I and 1G). Therefore, the reported discrepancies between Lorentzian and Voigtian bands with experimental bands can be explained by mere truncation effects. We also tested the distortions introduced by spectral truncation in spectra made of many overlapped bands. Again, spectra were now considered to be truncated between 1450 and 1850 cm1 (Figs. 2A–2C). For Lorentzian and Voigtian bands, the Lorentzian width function oscillates around the ideal data (compare black and gray lines in Figs. 2D and 2F). For the truncated spectrum made of overlapped Gaussian bands no distortions were observed in the Lorentzian width function (compare black and gray lines in Fig. 2E).

center of the data, as illustrated in Figs. 3F–3J. The spectra considered are made of a Lorentzian band at 1650 cm1, with a spectral interval of 1850–1450 cm1 (Fig. 3F), 1900–1500 cm1 (Fig. 3G), 1950–1550 cm1 (Fig. 3H), and 2000–1600 cm1 (Fig. 3I). In these conditions the band position and the mean interval wavenumber (v0 and vm) are no longer equal, and as a consequence the distortions in the interferogram come not only from the sinc function but also from the imaginary exponential in Eq. 9. The Lorentzian width function from the truncated spectra still oscillated around the ideal one, but the oscillation does not display a unique characteristic frequency (see Fig. 3J). The presence of multiple oscillations becomes more evident as the band becomes closer to the spectral edge. As this happens, the errorless points also become less free from error, eventually becoming difficult to use them to obtain a reliable fit/extrapolation.

MEAN LORENTZIAN WIDTH ESTIMATION IN THE PRESENCE OF SPECTRAL TRUNCATION DISTORTIONS

MINIMIZING THE SPECTRAL TRUNCATION DISTORTIONS

Empirically, we determined that the Lorentzian width function obtained from the truncated and ideal spectra cross approximately at 0.62/Dv, 1.54/Dv, 2.54/Dv, 3.54/Dv, etc. We named these distances the errorless points, plotted in Fig. 1 and Fig. 2 as open black circles. From the errorless points it seems possible to obtain the mean Lorentzian width of the spectra in spite of the oscillations. The approach we propose is the following: fitting some of the errorless points to a polynomial, and then extrapolating the fit to zero distance to obtain an estimation of the mean Lorentzian width. We first performed this analysis on the single Lorentzian band (Fig. 1, left column). Fitting and extrapolating the first ten errorless points (Fig. 1J, open circles) with a straight line, we obtained a mean Lorentzian width of 20.1 cm1 (not shown), very close to the actual value of 20 cm1. For the Voigtian band (Fig. 1, right column), the fitting-extrapolation gave a 20.3 cm1 Lorentzian width, also close to the actual value of 20 cm1. For the truncated spectrum made of overlapped Lorentzian bands (Fig. 2A), the fit and extrapolation of the first five errorless points of the Lorentzian width function (Fig. 2D) to a second-order polynomial provided a Lorentzian width estimate of 22.0 cm1 (not shown). For the spectrum made of overlapped Gaussian bands the fit-extrapolation gave an estimate of 0.0 cm1 (Fig. 2, middle column), and 22.1 cm1 for the spectrum made of overlapped Voigtian bands (Fig. 2, right column). These values are very close to the actual ones: 21.91, 0, and 21.91 cm1. In Fig. 3, we explored to what degree the errorless points are free from error. Figures 3A–3D show a spectrum made of a Lorentzian band, with increasing truncation. The Lorentzian band shows a maximum at 1650 cm1, coincident with the center of the spectrum (vm ¼ 1650 cm1). As Dv decreases from 400 cm1 (Fig. 3A) to 300 cm1 (Fig. 3B), 200 cm1 (Fig. 3C), and finally to 100 cm1 (Fig. 3D), the oscillations in the Lorentzian width function decrease in frequency and increase in amplitude (see Fig. 3E), although still oscillating around the ideal signal (thick continuous gray line). The errorless points remain quite close to the ideal curve, but as the truncation increases some discrepancies become more evident. Nevertheless, the fit extrapolation of the error-free points still allowed an estimation of the Lorentzian width between 19.9 and 20.1 cm1, even at the lowest Dv value tested (not shown). A different situation occurs when the band is not in the

Besides the use of the errorless points, it seems recommendable to find an alternative and more general way to minimize the oscillations in the interferogram caused by spectral truncation. The idea we propose is to use deconvolution, trying to change the original band shape from a Lorentzian to another band shape with wings of lower intensity. As we already discussed, the artifactual oscillations in the interferogram originate from the spectral truncation, which in turn depends on the intensity of the wings of the component bands. Therefore, by changing the band shape we should be able to minimize the truncation artifacts. Once the band shape is changed by deconvolution, the deconvoluted spectrum is truncated, transformed to an interferogram, and the interferogram is multiplied by an appropriate function to invert the deconvolution process. This generates back the interferogram of the original spectrum, but with removed/attenuated artifactual oscillations caused by spectral truncation. The practical success of this approach is shown in Fig. 4. Firstly, a spectrum from 1850–1450 cm1 made of a Lorentzian band of width 20 cm1 was deconvoluted, and then truncated from 1850–1600 cm1. Deconvolution was performed with a 2.5 narrowing factor, using a Bessel filter, and deconvolution widths of 15 cm1 (Fig. 4A), 20 cm1 (Fig. 4B), and 30 cm1 (Fig. 4C). The Lorentzian width function showed strongly reduced oscillations (Fig. 4D), keeping very close to the ideal one (thick gray line). After this preprocessing, it becomes straightforward to extrapolate the Lorentzian width function to zero distance (even without using the errorless points), obtaining the mean Lorentzian width. As a curiosity, for the deconvoluted-truncated spectra the Lorentzian width function at zero distance does not tend to zero as previously, but tends to the width used in the deconvolution process (see Fig. 4D). It is also important to note that the improvement obtained by this data preprocessing is quite insensitive to the width used in deconvolution (compare the different traces in Fig. 4D), although the best results are obtained when deconvolution is performed with a Lorentzian width similar to the mean Lorentzian width in the spectrum. The narrowing factor used in deconvolution does not play a critical role either (not shown). Nevertheless, because using higher narrowing factors provides narrower deconvoluted bands, the use of a higher narrowing factor allows higher data truncation levels free from artifactual oscillations (not

APPLIED SPECTROSCOPY

693

FIG. 3. (A–E) Effect of increasing the spectral truncation. (F–J) Effect of asymmetrical spectral truncation. (A–D and F–I) Spectra containing a single Lorentzian band at 1650 cm1 of 20 cm1 width. The spectra are truncated between (A) 1850–1450 cm1, (B) 1800–1500 cm1, (C) 1750–1550 cm1, (D) 1700–1500 cm1, (F) 1850–1450 cm1, (G) 1900–1500 cm1, (H) 1950–1550 cm1, and (I) 2000–1500 cm1. (E) Lorentzian width function for the ideal spectrum (thick gray line), for the spectrum in (A) (continuous black line), in (B) (continuous gray line), in (C) (dashed black line), and in (D) (dashed gray line). The open black, open gray, filled black, and filled gray circles correspond to the error-free points, respectively. (J) Lorentzian width function for the ideal spectrum (thick gray line), for the spectrum in (F) (continuous black line), in (G) (continuous gray line), in (H) (dashed black line), and in (I) (dashed gray line). The open black, open gray, filled black, and filled gray circles correspond to the error-free points.

shown). The filter used also played a minor role (not shown). However, some filters can provide a band shape with less intense wings/oscillations. Therefore, some filters could be more efficient at reducing the artifactual oscillations in the interferogram. Nevertheless, such possibility was not explored in the present paper.

PRESENCE OF SEVERAL PEAKS IN THE SPECTRUM In a more general situation, an experimental spectrum is not composed of a unique peak made of overlapped bands, but by several peaks, each one potentially made of overlapped bands. An example of such a situation is shown in Fig. 5A, mimicking the spectra observed in the 1700–1500 cm1 region of proteins, with two resolved peaks at 1656 cm1 and 1548 cm1 corresponding to the amide I and amide II vibrations. First, we show what happens when one tries to analyze simultaneously the spectrum with two peaks. The result of the Lorentzian width analysis is shown in Fig. 5B, for both the ideal (gray line) and the truncated (black line) spectrum. The ideal line tends to 27.01 cm1 at zero distance (Fig. 5B, gray line), exactly the mean Lorentzian width of the component bands of the spectrum in Fig. 5A. The first problem of this

694

Volume 62, Number 6, 2008

approach is that we may be interested only in the mean width of the component bands of only one peak (e.g., the amide I), rather than the whole spectrum. A major practical problem is that the interferogram module shows intense oscillation, with a wavelength close to the inverse of the peaks’ separation: (1656–1548)1 ’ 0.092 cm. These intense non-artifactual oscillations are transmitted to the Lorentzian width function (Fig. 5B, black). As a consequence, it becomes impossible to obtain a reliable fit-extrapolation of the truncated data to zero distance (Fig. 5B, gray line), and the mean Lorentzian width cannot be reliably estimated. We tested two approaches to make the method work in spectra made of several peaks. The first one was to directly truncate the spectrum, removing the data below 1600 cm1 (Fig. 5C, gray line), working only with the peak in which we are interested (Fig. 5C, black line). Figure 5D shows the Lorentzian width function for the truncated spectrum. Fitting the first three errorless points to a straight line, we obtain a 14.6 cm1 width (Fig. 5D, dashed black line), and 20.9 cm1 when the data was fitted to a second-order polynomial (Fig. 5D, dashed gray line). The estimated width becomes sensitive to minor details, such as the polynomial order used to fitextrapolate the data, making the estimation user-dependent.

FIG. 4. Use of deconvolution to reduce the spectral truncation artifacts. (A–C) A spectrum containing a single Lorentzian band (at 1650 cm1 of 20 cm1 width) from 1850–1450 cm1 was deconvoluted and truncated between 1850– 1600 cm1. Deconvolution was performed with a narrowing factor of 2.5, a Bessel filter, and a Lorentzian width of (A) 15 cm1, (B) 20 cm1, and (C) 30 cm1. (D) Lorentzian width function for the ideal spectrum (thick gray line) and for the deconvoluted spectra after inverting the deconvolution process in the Fourier domain for the spectrum in (A) (continuous black line), in (B) (dashed black line), and in (C) (dashed gray line).

The second approach we propose is similar to the first one but involves the use of deconvolution before data truncation (Fig. 5E). As seen, working with the deconvoluted-truncated spectrum provides much better results (compare Fig. 5D and Fig. 5F). There is a clear straight region with four errorless points. From a linear fit/extrapolation to zero we obtained a 22.3 cm1 mean Lorentzian width (not shown). A quadratic extrapolation gave a similar mean Lorentzian width of 21.6 cm1 (not shown). The gray line in Fig. 5F reproduces the ideal Lorentzian width function for the bands in the amide I region (the same data shown in Fig. 2D). The similitude of both black and gray lines in Fig. 5F shows how using deconvolution for preprocessing not only attenuates the oscillations, but also allows obtaining a mean Lorentzian width free from contributions from regions in which we are not interested. Finally, Fig. 5G shows the use of deconvolution followed by a more extensive spectral truncation. In practical conditions, such extensive spectral truncation may be required when neighboring peaks are present at both sides of the peak in which we are interested. Even with this extensive spectral truncation, the Lorentzian width function shows good behavior, keeping certain linearity up to 0.016 cm (see Fig. 5H). However, due to the extensive spectral truncation, the initial straight section only contains two errorless data points. With only two data points, fitting is no longer possible, but linear extrapolation to zero is still possible. When the first two errorless points were extrapolated to zero, we obtained a 22.9 cm1 width (Fig. 5H, dashed black line), close to the actual value. In a less favorable case, it could be possible that the second error-free data point could be outside the initial lineal range. In such a case, it is still possible to also fit/extrapolate

FIG. 5. The problem of spectra with several resolved peaks. (A) The same spectrum shown in Fig. 2A, but with extra five Lorentzian bands at 1590, 1570, 1550, 1535, and 1514 cm1; 40, 35, 35, 35, and 10 cm1 width; and 2, 5, 40, 20, and 2 areas, respectively. The spectrum is truncated between 1850–1400 cm1. (B) Corresponding Lorentzian width function for the ideal (gray line) and truncated (black line) spectrum in (A). (C) Same spectrum as in (A), but truncated between 1850–1600 cm1 (black line). (D) Corresponding Lorentzian width function for the truncated spectrum in (B) (black line). Lineal fit/extrapolation (dashed black line) and quadratic extrapolation (dashed gray line) of the first three errorless points to zero distance. (E) Same spectrum as in (A), but deconvoluted with a narrowing factor of 2.5, a Bessel filter, and a 30 cm1 Lorentzian width and truncated between 1850–1600 cm1 (black line). (F) Corresponding Lorentzian width function for the truncated spectrum in (E) (black line). The figure also shows the ideal Lorentzian width function for the amide I bands in gray (same as Fig. 2D). (G) Same spectrum as in (E), but truncated between 1710–1600 cm1. (H) Corresponding Lorentzian width function for the truncated spectrum in (G) (black line). Lineal extrapolation of the first two error-free points to zero distance (dashed black). Lineal fit/extrapolation of the truncated Lorentzian width function to zero distance (dashed gray line).

APPLIED SPECTROSCOPY

695

bands, due to their large wings. The negative oscillations from the negative offset can in favorable conditions effectively compensate the oscillations caused by spectral truncation, shown in Fig. 6D by the dashed back line. Therefore, we recommend offsetting the spectrum before applying the present method. The offset can be more safely performed in a deconvoluted spectrum because it contains bands with less intense wings.

PRESENCE OF NOISE IN THE SPECTRUM

FIG. 6. Effect of an error in the spectral offset. (A–C) Spectra containing a single Lorentzian band at 1650 cm1 of 20 cm1 width. (A) No offset error, (B) an offset error equal to 1% of the Lorentzian band maximal intensity, and (C) spectral offset at the extremes. (D) Lorentzian width function for the ideal spectrum (thick gray line) and for the spectrum in (A) (continuous black line), in (B) (dashed gray line), and in (C) (dashed black line).

directly some lineal interval of the experimental Lorentzian width function. With such an approach we obtained a 19.7 cm1 Lorentzian width estimation, as shown in Fig. 5H by the dashed gray line.

ERROR IN THE SPECTRAL OFFSET An error in the spectral offset means that the spectrum contains a constant component. The offset error will introduce an artifactual component in the experimental interferogram, E(x)exp, given by the second term of the following equation: Z vm þ0:5Dv ½EðvÞ þ aexpði2pxvÞdv EðxÞexp ¼ vm 0:5Dv

¼ EðxÞtrun þ a 3 Dv 3 sincðpDvxÞ 3 expði2pvm xÞ ð10Þ where a is the offset error intensity in the spectrum. Note that in contrast to Eq. 9, we now have a sinc function with a direct contribution to the interferogram. As a consequence, the distortions caused by an offset error can be much higher than the errors caused by truncation, but also easier to solve, as we will show. Figure 6A shows a spectrum made of a Lorentzian band. To this band, we added a constant signal of 1% of the Lorentzian maximal intensity (Fig. 6B). Figure 6D shows the Lorentzian width function. When the spectrum contains an offset error, the Lorentzian width function shows high amplitude oscillations (see Fig. 6D, dashed gray line), much larger than the oscillations observed without offset errors (see Fig. 6D, black line). Offset errors are usually reduced by performing a spectral offset. In a similar way, Fig. 6C shows a Lorentzian band offset to be zero at the extremes. This correction will introduce a small negative offset error for a spectrum made of Lorentzian

696

Volume 62, Number 6, 2008

Experimental spectra always contain a certain amount of noise. Therefore, it is relevant to consider the robustness of the method to the presence of noise in the input spectra. A method potentially providing valuable information but unable to work reliably under reasonable signal-to-noise ratios would be completely useless in practice. The method we propose in this paper uses the earlier data points of the interferogram, the region with higher signal-to-noise ratio. Therefore, we can expect a good robustness to the presence of noise in the spectra. To illustrate this fact, we added computer-generated normal noise to the spectrum shown in Fig. 5A, to a final signal-tonoise ratio of 5000, 1000, and 200 (Fig. 7A). Figures 7B through 7D show the corresponding noisy spectra after deconvolution (gray lines) and truncation (black lines). Finally, the corresponding Lorentzian width functions are displayed in Fig. 7E, which shows a minimal effect of noise, even when the initial signal-to-noise ratio was 200 (compare the black and the dashed gray line in Fig. 7E). As expected, the method is very robust to the presence of noise in the analyzed spectrum and is applicable under the habitual high signal-to-noise ratios found in IR spectroscopy.

PRESENCE OF WATER VAPOR BANDS IN THE SPECTRUM Atmospheric water vapor shows a strong absorption in the mid-infrared region. Purging the sample and the optics chamber with dry air greatly reduces its contribution. Still, fluctuations in water vapor concentration between the reference and the sample acquisition can generate positive (or negative) water vapor bands in absorbance spectra. Since water vapor bands are very narrow and generally have a low intensity if proper care is taken, it is reasonable to predict that their effect in the estimated mean Lorentzian width will be minimal. However, we should note that vapor bands are strongly enhanced in intensity when band-narrowing methods are applied. Since our general approach uses Fourier deconvolution to minimize spectral truncation effects, we should carefully check to what extent residual vapor bands can limit the method applicability. We added an experimental spectrum of water vapor at 2 cm1 to the spectrum shown in Fig. 5A (Fig. 8A). The ratio of the amide I band intensity to the more intense vapor band was set to 120, 40, and 12, representing a moderate, a high, and a very high water vapor contribution. Figures 8B through 8D show the corresponding spectra after deconvolution (gray lines) and truncation (black lines). Finally, the corresponding Lorentzian width functions are displayed in Fig. 8E. Although trace amounts of water vapor bands have a severe negative effect in deconvolution (see Figs. 8B–8D), they barely affected the Lorentzian width function (Fig. 8E), except at very high water vapor levels (Fig. 8E, gray dashed line). We can

FIG. 7. Effect of the presence of noise on the performance of the method. (A) Spectrum shown in Fig. 5A with the following ratios of added synthetic normal noise: SNR of 5000 (black line), 1000 (gray line), and 200 (light gray line), where SNR is the ratio of the spectrum maximum to the noise standard deviation before deconvolution. (Inset) Expanded region for a better appreciation of the noise intensity. (B–D) Deconvoluted spectra (narrowing factor of 2.5, a Bessel filter, and a 30 cm1 Lorentzian width) of the noisy spectra: (B) SNR of 5000, (C) SNR of 1000, and (D) SNR of 200. Deconvoluted spectra (light gray lines) were truncated (black lines) before computation of the Lorentzian width function. (E) Ideal Lorentzian width function for the amide I bands (thick continuous gray line, same as Fig. 2D); experimental Lorentzian width function for the truncated noise-free deconvoluted spectrum (continuous black line); and experimental Lorentzian width function for the truncated noisy deconvoluted spectrum with SNR of 5000 (continuous gray line), 1000 (dashed black line), and 200 (dashed gray line).

FIG. 8. Effect of the presence of residual water vapor bands on the performance of the method. (A) Spectrum shown in Fig. 5A with different levels of added experimental water vapor: SVR of 120 (black line), 40 (gray line), and 12 (light gray line), where SVR stands for the ratio of the spectrum maximum to the water vapor bands maximum. (Inset) Expanded region for a better appreciation of the water vapor bands. (B–D) Deconvoluted spectra (narrowing factor of 2.5, a Bessel filter, and a 30 cm1 Lorentzian width) of the spectra with water vapor contribution: (B) SVR of 120, (C) SVR of 40, and (D) SVR of 12. Deconvoluted spectra (light gray lines) were truncated (black lines) before computation of the Lorentzian width function. (E) Ideal Lorentzian width function for the amide I bands (thick continuous gray line, same as Fig. 2D); experimental Lorentzian width function for the truncated vapor-free deconvoluted spectrum (continuous black line); and experimental Lorentzian width function for the truncated deconvoluted spectrum with SVR of 120 (continuous gray line), 40 (dashed black line), and 12 (dashed gray line).

conclude that the presented method is robust to the presence of water vapor bands at the levels usually present in experimental IR spectra.

amide II vibration, and the small peak at 1739.5 cm1 corresponds to lipids that remain with the protein even when solubilized.40 For details about sample purification, preparation, and data acquisition see the original paper.40 First, we deconvoluted the spectrum, using a narrowing factor of 3, a Bessel filter, and three different Lorentzian widths: 25 cm1 (Fig. 10A), 20 cm1 (Fig. 10B), and 30 cm1 (Fig. 10C). We checked that the data offset was correct, and then we truncated the deconvoluted spectra (Figs. 10A–10C, black lines), removing the data not related to amide I bands (Figs. 10A–10C, gray lines). From the deconvoluted-truncated spectra we obtained the interferogram. The interferogram was corrected to invert the deconvolution process. At this point we also corrected the interferogram to remove the apodization function used when the spectrum was recorded (a triangle apodization with a cut-point of 0.5 cm). The interferogram module was calculated, and then its first derivative. Their appropriate ratio gave the Lorentzian width function, which is

EXPERIMENTAL APPLICATION In this section, we show how the method can be applied to experimental data. We consider the very interesting problem of estimating the mean Lorentzian width of the amide I band of a protein, the ADP/ATP mitochondrial carrier.39 We used the infrared absorbance spectrum of a dry film of the ADP/ATP carrier, solubilized in the detergent dodecyl maltoside and inhibited with carboxyatractyloside. In contrast to the spectrum presented in Fig. 3A of Lo´renz et al.,40 in the spectrum used here the contribution of the detergent was subtracted (see Fig. 9). Three peaks are observed in the 1800–1500 cm1 region. The most intense peak at 1657.5 cm1 corresponds to the amide I vibration, the peak at 1546 cm1 corresponds to the

APPLIED SPECTROSCOPY

697

FIG. 9. Absorbance spectrum of a dry film of the solubilized ADP/ATP carrier inhibited with carboxyatractyloside. The contribution from the dodecyl maltoside detergent was subtracted. The peak at 1657.5 cm1 corresponds to the amide I band, the peak at 1546.0 cm1 to the amide II band, and the peak at 1739.5 cm1 to lipids that remain bonded to the protein even in the solubilized state.

shown in Figs. 10D and 10F. In ideal conditions, the Lorentzian width function at zero distance gives the mean Lorentzian width. However, under the present conditions the Lorentzian width function departs from the ideal one as we approach zero, tending to the Lorentzian width used in deconvolution and not to the mean Lorentzian width.

An estimation of the mean Lorentzian width can be made from the fit/extrapolation of the errorless points (see Fig. 10, open circles). However, in this case the separation of errorless points is too large, and the second errorless point already lies too far for a reliable extrapolation. Nevertheless, we extrapolated to zero the first two error-free points, to obtain a mean Lorentzian width of ’ 21 cm1 (Figs. 10D–10F, dashed back line). However, a visual inspection clearly indicates that the obtained value will overestimate the actual mean Lorentzian width. We also performed a linear fit/extrapolation directly from the experimental Lorentzian width function, and obtained a mean Lorentzian width estimation of 18 6 1 cm1 (Figs. 10D–10F, dashed gray line). In view of Figs. 10D–10F, this estimation seems quite reliable. Note, however, the negative lobule around 1700 cm1 in Figs. 10A–10C, also present when the spectrum was deconvoluted with an 18 cm1 Lorentzian width (not shown). This clearly suggests that the band resolved at 1695 cm1 is actually narrower than 18 cm1, being overdeconvoluted under all conditions. Since 18 cm1 is the estimated mean width of the amide I bands, the presence of an over-deconvoluted band indicates that both narrower and broader bands than 18 cm1 are present in the amide I region.

POTENTIAL APPLICATIONS As already mentioned in the Introduction, an objective estimation of the mean Lorentzian width can allow for a more objective and efficient use of Fourier deconvolution (or any other deconvolution method) and can also improve the

FIG. 10. Estimation of the mean Lorentzian width of the amide I band of the ADP/ATP carrier. (A–C) Deconvoluted spectra with a narrowing factor of 3.0, a Bessel filter, and a deconvolution Lorentzian width of (A) 25 cm1, (B), 20 cm1, and (C) 30 cm1 (gray lines). The deconvoluted spectra were truncated (black lines). (D–F) Lorentzian width functions obtained from the truncated spectra in (A–C). The Lorentzian width functions were extrapolated to zero distance by two methods. The first two errorless points (open circles) were extrapolated (dashed black line). A straight section of the Lorentzian width function was fitted/extrapolated to zero distance (dashed gray line). The obtained mean Lorentzian widths are indicated.

698

Volume 62, Number 6, 2008

convergence properties and robustness of curve fitting. These two improvements can help, for instance, in the quantitative analysis of the amide I band of proteins, and in general in any area where quantitative analysis of overlapped bands is required. The estimation of the mean Lorentzian width could have its interest in its own. Currently, time-resolved pump-probe IR experiments with sub-picosecond time-resolution are used to monitor the amide I relaxation of peptides and proteins.41–43 On the other hand, static pump-probe two-dimensional (2D) IR experiments provide the homogeneous broadening of the amide I vibrations along the anti-diagonal.42 However, both experiments require sophisticated and expensive equipment. The introduced method could allow, in a simple way, this information to be obtained. The mean Lorentzian width of the amide I band should correspond roughly to the amide I homogeneous broadening, directly related to the T1 relaxation of the amide I vibrations.44,45 A Lorentzian band of width c corresponds to an exponential relaxation constant in the time domain of s ¼ (p 3 c 3 c)1 (compare Eqs. 1 and 2), where c is the speed of light in cm/s. For instance, a mean Lorentzian width of 18 cm1 for the amide I band of a protein suggests a mean relaxation time constant for the amide I vibration of ;600 ps. Interestingly, a 2D-IR experiment of a transmembrane a-helical peptide determined a homogenous broadening of 16 6 1 cm1 for the amide I vibrations.42 The same authors also measured the vibrational relaxation of the amide I vibrations, showing a time constant of ;700 ps.42 These values are very close to our estimated mean Lorentzian width and deduced amide I relaxation for the ADP/ATP carrier, 18 cm1 and 600 ps, respectively. Therefore, it seems that the introduced method could allow the extraction, from the static one-dimensional IR spectrum of the amide I, of its homogenous broadening/vibrational relaxation. This could allow one to test in a simple way how the amide I relaxation depends on, for instance, the protein size; the main protein structure (helix/beta/ unordered); the protein type (membrane/soluble); the protein environment (i.e., solubilized/reconstituted); the hydration conditions (solution/wet film/dry film); the temperature, etc., giving clues to the factors affecting the amide I relaxation process.

DISCUSSION AND CONCLUSION We have introduced a method that allows the estimation of the mean Lorentzian width of the bands in a spectrum, weighted by their areas. The method does not require the specification of the number of bands, their positions, or their areas. Indeed, the method would work even if the number of bands is unlimited. Moreover, the method provides the correct mean Lorentzian width even if the spectrum contains Gaussian or Voigtian bands. The method is also very robust to the presence of noise and water vapor bands in the analyzed spectrum. The method introduced also has some intrinsic limitations. First, it only provides the mean Lorentzian width, weighted by the band areas. In the case of heterogeneous bandwidths, using the mean Lorentzian width in deconvolution can lead to the over-deconvolution of some bands (negative side-lobules) and the infra-deconvolution of others (reduced effective bandnarrowing a noise-enhancement). Nevertheless, the use of the mean Lorentzian width in deconvolution may lead in most cases to a good compromise, balancing the pernicious effects

of both excessive over- and infra-deconvolution. Second, the method does not provide any information about the Gaussian width of the bands. This limitation is basically irrelevant for its application to IR spectra, since IR bands are mostly Lorentzian in shape, but can become critical for the application of the method to UV-vis spectra. Third, for a spectrum containing several peaks (each peak with an unspecified number of overlapped bands), the spectrum has to be truncated and each peak analyzed separately. Finally, the method only provides the mean Lorentzian width in spectra where all the bands have positive areas, and so it is applicable to absorbance spectra, but not to linear dichroism or difference spectra. The method has also some practical limitations, mainly related to some distortions generated by the spectral truncation process. We have shown how data preprocessing, involving Fourier deconvolution, can effectively minimize these distortions, making the method applicable to experimental data. We are only aware of three methods able to provide an estimation of the bandwidth in overlapped spectra without specification of the number, positions, and areas of the component bands. The first method was introduced as a direct application of Fourier self-deconvolution.14 First, the overlapped bands must be completely resolved by Fourier selfdeconvolution. The original individual bands can be restored by extracting the resolved bands and inverting the deconvolution process, from which the individual original bandwidths and band shapes can be estimated.14,46 The main limitation of this method is the requirement that the individual bands become completely resolved after deconvolution, precluding its applicability to the highly overlapped amide I band of proteins. The second method was introduced by Saarinen et al.,47 and it is based in deconvolution and backward linear prediction, following by an optimization step. The method is not only able to determine the bandwidth but also the band shape of the component bands. It works equally well for spectra containing both positive and negative bands. In principle, the method is only applicable to spectra containing bands with very similar widths and band shapes. The spectrum analyzed can contain many bands, but the number should not be higher than a certain value related to the interferogram length. Moreover, its performance is very sensitive to the presence of noise in the spectrum, requiring high signal-to-noise ratios. We have implemented Saarinen et al.’s method with some modifications (unpublished) and used it in both synthetic and experimental data.19,23,48 The large number of parameters involved (especially in the linear prediction step) makes it time consuming and also, to a certain degree, user dependent. We found it difficult to apply the Saarinen et al. method to the analysis of spectra with highly overlapped bands (as the amide I band of proteins), but it becomes quite useful for the bandwidth analysis of spectra containing many moderately overlapped positive/negative bands, as in infrared difference spectroscopy.48 The last method is from Buslov et al.49 It provides an estimation of the bandwidth from the distance at which the interferogram module reaches the noise level x0. As for our method, Buslov et al.’s method is easy to implement. However, it shows some shortcomings. The main one is that the way the x0 is transformed into a bandwidth depends on the band shape assumed. The estimated bandwidth can be quite different for a Lorentzian and for a Gaussian band shape. Moreover, for a Voigtian band this method can only give a relation between the Lorentzian and the Gaussian bands, but not a bandwidth. In

APPLIED SPECTROSCOPY

699

addition, when a spectrum contains bands of different bandwidths, the estimated bandwidth does not correspond to the mean bandwidth (as in our method), but it is biased in an unpredictable way toward the narrower band in the spectrum. This makes the interpretation of the estimated bandwidth less precise. Finally, this method is also sensitive to spectral truncation and therefore cannot be applied to the analysis of the amide I band without some modifications. ACKNOWLEDGMENTS The mitochondrial ADT/ATP carrier inhibited with carboxyatractyloside and solubilized in dodecyl maltoside was kindly provided by Ge´rard Brandolin. This work was supported by grant BFU2006-04656/BMC from the Direccio´n General de Investigacio´n (Ministerio de Educacio´n y Ciencia). 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

D. S. Stephenson, Prog. Nucl. Mag. Reson. Spectrosc. 20, 515 (1988). L. Antonov and D. Nedeltcheva, Chem. Soc. Rev. 29, 217 (2000). D. M. Byler and H. Susi, Biopolymers 25, 469 (1986). J. L. Arrondo, A. Muga, J. Castresana, and F. M. Go˜ni, Prog. Biophys. Mol. Biol. 59, 23 (1993). E. Goormaghtigh, V. Cabiaux, and J. M. Ruysschaert, Eur. J. Biochem. 193, 409 (1990). M. Jackson and H. H. Mantsch, Crit. Rev. Biochem. Mol. Biol. 30, 95 (1995). J. K. Kauppinen, D. J. Moffatt, M. R. Hollberg, and H. H. Mantsch, Appl. Spectrosc. 45, 411 (1991). V. A. Lo´renz-Fonfrı´a and E. Padro´s, Appl. Spectrosc. 59, 474 (2005). A. Barth, Spectrochim. Acta, Part A 56, 1223 (2000). J. Zheng, H. Zhang, and H. Gao, Sci. China (series B) 43, 1 (2000). M. I. Rogojerov and M. G. Arnaudov, Vib. Spectrosc. 3, 239 (1992). I. W. Friesen and K. H. Michaelian, Appl. Spectrosc. 45, 50 (1991). P. E. Saarinen, Appl. Spectrosc. 51, 188 (1997). J. K. Kauppinen, D. J. Moffatt, H. H. Mantsch, and D. G. Cameron, Appl. Spectrosc. 35, 271 (1981). D. G. Cameron and D. J. Moffatt, Appl. Spectrosc. 41, 539 (1987). B. G. M. Vandeginste and L. De Galan, Anal. Chem. 47, 2124 (1975). W. F. Maddams, Appl. Spectrosc. 34, 245 (1980). A. P. De Weijer, C. B. Lucaslus, L. Buydens, G. Kateman, H. M. Heuvel, and H. Mannee, Anal. Chem. 66, 23 (1994). V. A. Lo´renz-Fonfrı´a and E. Padro´s, Analyst (Cambridge, U.K.) 129, 1243 (2004). J. A. Pierce, R. S. Jackson, K. W. Van Every, P. R. Griffiths, and G. Hongjin, Anal. Chem. 62, 477 (1990). R. S. Jackson and P. R. Griffiths, Anal. Chem. 63, 2557 (1991). X. Q. Zhang, J. B. Zheng, and H. Gao, Analyst (Cambridge, U.K.) 125, 915 (2000).

700

Volume 62, Number 6, 2008

23. V. A. Lo´renz-Fonfrı´a and E. Padro´s, Spectrochim. Acta, Part A 60, 2703 (2004). 24. F. Holler, D. H. Burns, and J. B. Callis, Appl. Spectrosc. 43, 877 (1989). 25. D. G. Cameron and D. J. Moffatt, J. Test. Eval. 12, 78 (1984). 26. V. A. Lo´renz-Fonfrı´a, J. Villaverde, and E. Padro´s, Appl. Spectrosc. 56, 232 (2002). 27. K. Rahmelow and W. Hu¨bner, Appl. Spectrosc. 50, 795 (1996). 28. S. D’Auria, F. Alfieri, M. Staiano, F. Pelella, M. Rossi, A. Scire`, F. Tanfani, E. Bertoli, Z. Grycznyski, and J. R. Lakowicz, Biotechnol. Prog. 20, 330 (2004). 29. A. Marabotti, A. Ausili, M. Staiano, A. Scire`, F. Tanfani, A. Parracino, A. Varriale, M. Rossi, and S. D’Auria, Biochemistry 45, 11885 (2006). 30. J. Cladera, M. Sabe´s, and E. Padro´s, Biochemistry 31, 12363 (1992). 31. T. Heimburg, M. Esmann, and D. Marsh, J. Biol. Chem. 272, 25685 (1997). 32. S. G. Taneva, J. M. Caaveiro, A. Muga, and F. M. Go˜ni, FEBS Lett. 367, 297 (1995). 33. D. Naumann, C. Schultz, U. Gro¨ ne-Tschelnokow, and F. Hucho, Biochemistry 32, 3162 (1993). 34. P. W. Holloway and H. H. Mantsch, Biochemistry 28, 931 (1989). 35. H. H. de Jongh, E. Goormaghtigh, and J. M. Ruysschaert, Biochemistry 36, 13593 (1997). 36. P. R. Griffiths and J. A. de Haseth, Fourier Transform Infrared Spectrometry (John Wiley and Sons, New York, 1986). 37. A. G. Marshall and F. R. Verdun, Fourier Transforms in NMR, Optical, and Mass Spectrometry (Elsevier, Amsterdam, Oxford, New York, Tokyo, 1990). 38. J. K. Kauppinen and J. Partanen, Fourier Transforms in Spectroscopy (Wiley-VCH, Berlin, 2001). 39. C. Dahout-Gonzalez, H. Nury, V. Tre´ze´guet, G. J. Lauquin, E. PebayPeyroula, and G. Brandolin, Physiology (Bethesda) 21, 242 (2006). 40. V. A. Lo´renz, J. Villaverde, V. Tre´ze´guet, G. J. Lauquin, G. Brandolin, and E. Padro´s, Biochemistry 40, 8821 (2001). 41. S. Woutersen, Y. Mu, G. Stock, and P. Hamm, Proc. Natl. Acad. Sci. U.S.A. 98, 11254 (2001). 42. P. Mukherjee, I. Kass, I. T. Arkin, and M. T. Zanni, Proc. Natl. Acad. Sci. U.S.A. 103, 3528 (2006). 43. A. Xie, L. van Der Meer, W. Hoff, and R. H. Austin, Phys. Rev. Lett. 84, 5435 (2000). 44. D. W. Oxtoby, Annu. Rev. Phys. Chem. 32, 77 (1981). 45. J. C. Owrutsky, D. Raftery, and R. M. Hochstrasser, Annu. Rev. Phys. Chem. 45, 519 (1994). 46. V. A. Lo´renz-Fonfrı´a, J. Villaverde, V. Tre´ze´guet, G. J. Lauquin, G. Brandolin, and E. Padro´s, Biophys. J. 85, 255 (2003). 47. P. E. Saarinen, J. K. Kauppinen, and J. O. Partanen, Appl. Spectrosc. 49, 1438 (1995). 48. X. Leo´n, V. A. Lo´renz-Fonfrı´a, R. Lemonnier, G. Leblanc, and E. Padro´s, Biochemistry 44, 3506 (2005). 49. D. K. Buslov and N. A. Nikonenko, Appl. Spectrosc. 52, 613 (1998).

A Versatile Analytical Expression for the Inverse Abel Transform Applied to Experimental Data with Noise SHUILIANG MA,* HONGMING GAO, GUANGJUN ZHANG, and LIN WU State Key Laboratory of Advanced Welding Production Technology, Harbin Institute of Technology, Harbin 150001, China

A method for reconstruction of radially distributed plasma emission coefficients from projections with noise is proposed. The method represents the projections based on overlapping piecewise polynomial least squares fitting to take the inversion. Parameters that affect the inversion accuracy are analyzed and discussed in detail. Results for profiles with various shapes are presented and compared with those obtained with other methods. It is shown that for data with different numbers of points and different levels of noise, our method is more accurate and yields markedly better results for very sparse data. In addition, excellent results have been obtained from experimental intensities of an arc plasma without filtering of noise. Index Headings: Abel inversion; Arc plasma; Emission spectroscopy.

INTRODUCTION Thermal plasmas have been extensively used in many industrial fields, such as cutting, welding, spraying, surface modification, synthesis of fine powders, and waste treatment.1,2 In order to improve and optimize the characteristics of these plasmas for their applications, one needs to have a good understanding of the behavior under different experimental conditions. The spatial distributions of plasma parameters, such as the electron density and temperature, could provide important insight into the basic processes that occur in the plasma. Therefore, spatially resolved measurements of these parameters are quite necessary and indispensable. Since such parameters in most circumstances are not uniformly distributed within the discharge, the collected intensity in spectroscopic measurements that is the projection of emission coefficients integrated along a chord in the plasma column cross-section should be converted to the original radial distribution (see Fig. 1). For cylindrically symmetric and optically thin plasmas, this can be implemented by using a method known as Abel inversion.3 Over the past years, a large number of methods have been developed for the inversion. Some methods are based on the polynomial interpolation technique,4,5 which is a commonly used approach for function approximation. Because the influence of noise is not considered in these inversions, uncertainties in experimental data are greatly magnified due to the differential operator in the inverse Abel integral equation. Though it is possible to apply a smoothing process6 before the inversion, in most cases the smoothing degree is very difficult to control. Algorithms using Fourier transform techniques7,8 are also sensitive to noise, and the Fourier–Hankel method9,10 cannot yield reasonable results for small sets of data, as demonstrated in Ref. 11. Methods employing spline techniques12–14 were reported to be superior to other methods, with a high stability for resisting of noise, but when the noise is to be a magnitude larger, the methods are no longer suitable for Received 3 December 2007; accepted 27 March 2008. * Author to whom correspondence should be sent. E-mail: shlgma@126. com.

Volume 62, Number 6, 2008

use, and the poor inversion accuracy for off-axis peak profiles with very sparse data will lead to unacceptable results.15 The only appropriate methods seem to be those using polynomial least squares techniques to fit the experimental data either in the whole segment16,17 or in several divided ones.18,19 Among them the method that breaks the whole interval into five segments19 is more effective for yielding the best result in most situations. However, the larger errors near both ends of each segment fitted greatly deteriorate the inverted result. For these methods a common problem is that the inversion accuracy is very poor for small sets of data. In our experiments, the data collected with an optical fiber of a spectrograph are very limited; usually the number of points is from 20 to 50. The noise in the measured data evidently affects the inverted result when using methods such as that of Nestor and Olsen;4 thus, it would be highly desirable to provide an appropriate method that is suitable for this case. In such a way, the noise in the experimental data does not need to be filtered or can be just partly filtered, and the result obtained is relatively good even for data with a very limited number of points. This paper proposes a versatile Abel inversion method based on overlapping piecewise polynomial least squares fitting, which has a rather simple analytical expression but is very accurate even for very sparse data. In different cases, due to the smoothing property of the polynomial least squares fitting and the use of the segments overlapping technique, highly accurate results can be obtained by only adjusting one parameter according to the level of noise in the data. In the following sections we first describe the inversion method, then analyze the influences of different parameters and optimize their combinations, and at last show results obtained for simulated profiles and experimental data.

INVERSION METHOD The forward Abel transform may be written as Z 1 eðrÞr pffiffiffiffiffiffiffiffiffiffiffiffiffiffi dr Ið yÞ ¼ 2 r 2  y2 y

ð1Þ

where I(y) represents the measured intensity and e(r) is the emission coefficient of the radiation source with a radius of unity. The inverse Abel transform for reconstruction of the emission coefficient is given by Z 1 1 I 0 ðyÞ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi dy ð2Þ eðrÞ ¼ p r y2  r 2 where I 0 ( y) ¼ dI(y)/dy. In practice, however, there are two difficulties in applying Eq. 2 directly to experimental intensities. Values of the intensity measured are always in sets of discrete data, while a smooth function of I(y) is required for calculating the integral.

0003-7028/08/6206-0701$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

701

FIG. 1. Schematic diagram showing the coordinate system and geometry of a cylindrically symmetric radiation source with respect to the z-axis, which is normal to the paper.

Moreover, the random errors inherent in the measured data will be greatly magnified by the inverse transform, which involves numerical differentiation. In this paper, both of the problems are resolved by representing the intensity profile with overlapping piecewise polynomials. Assume I( y) is symmetric around y ¼ 0, smooth and continuous as one period of a periodic function in the interval [1, 1]; it can be approximated as XK XM XK x x an;m y m ð3Þ IðyÞ ’ n ðyÞPn;M ðyÞ ¼ n ðyÞ n¼1 n¼1 m¼0 where xn( y) ¼ 1 for yn1  y  yn and xn(y) ¼ 0 otherwise, and Pn,M( y) is a polynomial of degree M defined over [yan , ybn ], as shown in Fig. 2. Substituting Eq. 3 into Eq. 2 and with some simplifications, we have ek ðrÞ ¼ 

1 XK XM man;m n¼k m¼1 p ½dn;k Im1 ðr; yn Þ þ ð1  dn;k ÞIm1 ðyn1 ; yn Þ;

yk1  r  yk

ð4Þ

where 1  k  K, ek(r) is the expression of the emission coefficient for the kth segment, d is the Kronecker d-function, and Z b ym pffiffiffiffiffiffiffiffiffiffiffiffiffiffi dy ð5Þ Im ða; bÞ ¼ y2  r 2 a Orthogonal polynomials have good properties for function least squares fitting.16 With fn,l ( y) as orthogonal polynomial basis functions,20 Pn,M( y) can be expanded as XM XM Xl Pn;M ðyÞ ¼ c cn;l dn;l;m y m ð6Þ n;l fn;l ðyÞ ¼ l¼0 l¼0 m¼0 where cn;l ¼

Z

ybn

yan

fn;l ðyÞIðyÞ dy

ð7Þ

FIG. 2. Schematic representation of the overlapping piecewise polynomial least squares approximation for the intensity curve I(y). Pn,M denotes the polynomial of degree M fitted over segment [yan , ybn ], and xn is the function for determining the segment [yn1, yn] for inversion.

where Pl(x) is the Legendre polynomial21 of degree l defined over [1, 1]. Taking an exchange in the order of the summation in Eq. 6 and comparing with Eq. 3, one obtains XM c d ð9Þ an;m ¼ l¼m n;l n;l;m Note that cn,l cannot be directly calculated using Eq. 7. Since I(y) is assumed to be a periodic function, intensity data can be substituted by cubic interpolation functions. Under this condition, there is no difficulty to calculate the values of cn,l. Therefore, using Eqs. 4, 5, and 9, the emission coefficient can be inverted readily. Also note that almost in all cases we force I(y) to have a zero slope at the center, i.e., I 0 (0) ¼ 0, which is desired for a cylindrically symmetric distribution. For Eq. 3 the condition can be satisfied only by setting ya1 ¼ yb1 . According to the arrangement of segments, the performance of the method can be very different. It is easy to see that the technique of Bockasten5 and those fitting the intensity data to one curve16,17 are two extremities of the method presented here. On the other hand, the algorithm of Cremers and Birkebak 19 is a compromise of the two kinds of methods. It is clear that if the combination of the parameters in Eq. 3 is optimized, the best performance can be achieved. This, as the main aim of this paper, is discussed in detail and presented in the following section.

RESULTS AND DISCUSSION Test Profiles and Methods. It is always desirable to test a numerical method using simulated data, as the exact inversion is known and making a comparison between results obtained with different methods is possible. For this purpose, a total of four test profiles of two types are chosen from numerical examples in the literature.14,19,22,23 Figure 3 shows the radial emission coefficient distributions of these profiles. Two of them are for the usual bell-shaped distributions. They are given by e1 ðrÞ ¼ ð1  rÞ2 ð1 þ 2rÞ;

0r1

ð10Þ

m

and dn,l,m is the coefficient of y in fn,l (y), which is defined as sffiffiffiffiffiffiffiffiffiffiffiffiffiffi   2l þ 1 2y  ybn  yan fn;l ðyÞ ¼ P ð8Þ l ybn  yan ybn  yan

702

Volume 62, Number 6, 2008

and  e2 ðrÞ ¼

1  2r 2 ; 2ð1  r 2 Þ;

0  r  0:5 0:5 , r  1

ð11Þ

FIG. 3. Radial distributions of the emission coefficient of four test profiles.

The other two are for the off-axis peak type of distributions, which are expressed as 8 3 2 3 > > 0  r  0:25 > 4 þ 12r  32r ; < ð12Þ e3 ðrÞ ¼ 16 > 2 > ð1  rÞ ð1 þ 8rÞ; 0:25 , r  1 > : 27 and e4 ðrÞ ¼ ð1  r 2 Þ2 ð1 þ 12rÞ;

0r1

ð13Þ

Both of the two types are commonly encountered in plasma diagnostics. In our experiment for a dc argon arc, the bellshaped distributions typically appear in the anode arc column, and the off-axis peak profiles are characteristic distributions of the cathode arc column with high temperatures. Intensity data are needed as the input of inversion; thus, the corresponding values of I1(y) to I4(y) are obtained by direct integration of Eq. 1. As demonstrated in many studies,14,19 the off-axis peak distribution is much more difficult to reconstruct faithfully than other types with simple shapes such as e1(r) and e2(r). Therefore, we use I4 to investigate the effect of different parameters on the accuracy of the new method, then use I1 through I3 to validate the possibility with the optimized parameters for inversion of profiles with different shapes and make a comparison with other methods. The experimentally measured data in practice are unavoidably distorted by noise. To test the properties of the method for input values with errors, the data of I1 through I4 were rounded off to two decimal places or included different levels of normally distributed random noise with an absolute scale independent of the test profiles to simulate experimental uncertainties. To estimate the performance of the method with different parameters and compare the results with those obtained using other methods, we calculated the absolute inversion error at each point: Deðri Þ ¼ et ðri Þ  ec ðri Þ

ð14Þ

where et(ri) is the theoretical emission coefficient at point ri, and ec(ri) is the calculated value at the corresponding point, and we also calculated the standard deviation r, defined by  X 1=2 1 N 2 r¼ ½e ðr Þ  ec ðri Þ ð15Þ i¼1 t i N

Influence of Different Parameters. As can be seen from Eq. 3, in order to achieve the best performance with the proposed method several parameters should be optimized, such as the arrangement of the segments for both approximation and inversion and the degree of polynomial used for least squares fitting. For simplicity, all the segments were set to an equal length, i.e., yn  yn1 are the same for all values of n, except n ¼ K, for which the length of the last segment may be equal to or less than those of other ones. Though the degree of the polynomial used for each segment can be different, here only the case in which all segments employ polynomials with the same degree will be explored. Considering the inability of using polynomials with too low degrees for approximation and the difficulty of suppressing noise using polynomials with too high degrees, the values of M were chosen as 4, 6, and 8. The first parameter that should be optimized is the length of the segment for approximation, which determines the overlapping amount of two near segments (see Fig. 2). It is defined as the overlapping factor q ¼ ðybn  yan Þ=R

ð16Þ

where R is the radius of the source, which is normalized to unity in this paper. There are two advantages to the near segments being overlapped. The larger fitting errors invariably encountered near the ends of the data interval (this is the so-called termination error)24 are avoided under this condition as both end parts of each fitted profile are dropped with the overlapping technique, so the inverted curve will be more smooth and accurate. Also, since the segment for approximation has a longer length than that for inversion, more data points have been used for approximation, and thus fitted accuracy can be improved, especially for small sets of data with noise. As can be found from the example presented below that the inversion error indeed decreased with the increase of the overlapping factor. To optimize the overlapping factor q, random noise with a standard deviation S ¼ 0.01 was added to the values of I4 before taking the inversion. Results obtained with various values of q are shown in Fig. 4. As expected, the inversion accuracy keeps improving with the increase of q. The standard deviation r bottomed out at certain positions for different values of M. A detailed study indicates that for large values of q, the approximated function for each segment failed to accurately represent the intensity profile, and thus r increases as q tends to 1. The increasing trend is most significant for M ¼ 4, since a polynomial with a lower degree is more difficult to fit to the real distribution of data. Therefore, the value of q should be chosen between 1/K and 0.7, 0.8, and 0.9 for M ¼ 4, 6, and 8, respectively, depending on the level of noise in the data to be inverted. For experimental data with weak noise, the value of q can be lowered appropriately. Then the effect of the number of segments on the inversion accuracy is analyzed using two sets of data with N ¼ 100. One was used as the input value of the presented method with a low level of noise of S ¼ 0.005, and the other included a high level of noise with S ¼ 0.05. The overlapping factors for polynomials with degrees M ¼ 4, 6, and 8 were set as 0.4, 0.5, and 0.6 for S ¼ 0.005, and 0.5, 0.7, and 0.8 for S ¼ 0.05. The calculated standard deviations are shown in Figs. 5a and 5b, respectively. It is clear that the influence of the number of segments is not significant. With the increase of K, r increases very slowly. For K  5 the inversion error is relatively small. However,

APPLIED SPECTROSCOPY

703

FIG. 4. Standard deviations calculated with K ¼ 4 and various values of q for I4 with N ¼ 24 and S ¼ 0.01.

there is one exception; for K ¼ 2, big errors appeared for the inversion with M ¼ 4, which once again indicates the limited ability for fitting profiles faithfully using polynomials with M  4. Therefore, for practical utilization we recommend that the value of K should be no less than 3, although for K ¼ 2 values of r obtained with M ¼ 6 and 8 are much smaller. Considering that the use of a large number of segments cannot improve inversion accuracy but increases computational time, the value of K should be chosen between 3 and 5. The inversion accuracy will also depend on the number of FIG. 6. Standard deviations calculated with K ¼ 5 and various values of N for I4 with (a) S ¼ 0.005, and (b) S ¼ 0.05. Other conditions are the same as Fig. 5.

FIG. 5. Standard deviations calculated with various values of K for I4 with N ¼ 100, (a) S ¼ 0.005, q ¼ 0.4, 0.5, and 0.6; and (b) S ¼ 0.05, q ¼ 0.5, 0.7, and 0.8 for M ¼ 4, 6, and 8, respectively.

704

Volume 62, Number 6, 2008

data points. To find this relation, two series of data sets the same as those used in Figs. 5a and 5b but with different values of N were generated. The results inverted with K ¼ 5 are shown in Figs. 6a and 6b. Since the value of the number of segments was chosen appropriately, there is no significant difference between the results inverted with M ¼ 4, 6, and 8. With the increase of N, r indeed decreased. But for N  50 the improvement becomes minimal for the data with either a low or high level of noise. Therefore, when N , 50, it is better to use techniques such as interpolation to improve the inversion accuracy. On the other hand, when N  50, there is no need to increase the number of data points before inversion. Comparison with Other Algorithms. According to the results of the previous section, we used the method to take an inversion with K ¼ 5, M ¼ 4. We chose M ¼ 4 because a good comparison with the Cremers and Birkebak method19 is possible. In general, other parameters may be possible for achieving better results. First, the method was applied to the data of I3 with and without uncertainties. The absolute inversion errors obtained are listed in Table I. This illustrates the differences in the results between our method and that of Cremers and Birkebak. For accurate data, the result of Cremers and Birkebak is a little superior. For data rounded off to two decimal places, which is approximately equivalent to those including normally distributed random noise with a standard deviation, S ’ 0.00289; however, our method gives more accurate results. Since both the methods use the same number of segments and the same polynomial degree, the improvement is considered to be due to

TABLE I.

Comparison of absolute inversion errors obtained for accurate data of I3 and those rounded off to two decimal places with N ¼ 30. S¼0

i 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 r

Cremers and Birkebak, 4th deg 1.8 3.0 1.2 1.2 0.0 3.6 3.5 7.3 2.9 1.3 2.9 1.0 3.0 4.0 3.0 2.0 0.0 0.0 3.0 4.0 3.0 2.0 0.0 1.0 4.0 4.0 1.0 3.0 3.0 4.0 0.0 1.8

3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3

3

10 104 103 103 104 103 103 103 103 103 103 104 104 104 104 104 104 104 104 104 104 104 104 104 104 104 104 104 104 104 100 103

S ’ 0.00289 This work, M ¼ 4, q ¼ 0.5 2.1 1.9 3.8 4.2 3.3 2.9 2.8 9.0 6.3 2.3 5.2 1.7 4.3 7.6 8.6 7.5 5.3 3.2 2.6 6.7 8.5 7.7 5.1 2.5 4.0 4.6 2.0 1.6 2.2 4.5 0.0 2.5

3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3

4

10 103 103 103 103 103 103 103 103 103 104 103 104 104 104 104 104 104 104 104 104 104 104 104 104 104 104 104 104 104 100 103

the use of the overlapping technique. It is interesting to notice that the superior method for accurate data now turns out to be the worst method when uncertainties are presented in the intensities. Then, with our method for the first three test profiles I1 to I3 with and without error, using a varying number of data points, standard deviations of the inversion are obtained, as presented in Table II, where the results of the fourth degree method of Cremers and Birkebak and the spline interpolation method14 are also listed for comparison. For data with S ¼ 0, the results of our method are presented only to show the highest accuracy that can be achieved. By adjusting parameters such as the degree of the polynomial and the number of segments, more accurate result will be obtained, as shown in Ref. 18. Since experimental data that unavoidably have uncertainties cannot be measured with a high accuracy, results inverted for data with error will be more valuable. For data rounded off to two decimal places our method is superior to the other two methods, except for the case of I3, for which the spline interpolation method yields more accurate results. In fact, rounding the profile of I3 to two decimal places is not equivalent to adding normally distributed random noise as at the region near the center of the source the magnitude of I3 varies very slow. Therefore, comparison for data with random noise will be more reliable. Finally, we studied the performance of the method for data with different levels of random noise. The results obtained are listed in Table III. When S , 0.1, for I2 the result of our method is still more accurate than that obtained with the

Cremers and Birkebak, 4th deg 3.9 3.1 1.3 4.3 1.2 1.4 5.0 8.4 4.0 0.0 1.5 8.0 1.3 2.2 2.3 1.5 2.0 4.5 4.6 3.0 6.0 1.9 3.8 5.7 1.8 2.7 4.6 4.6 2.5 5.0 0.0 1.0

3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3

2

10 102 103 103 102 103 103 103 103 104 103 104 103 103 103 103 104 103 103 103 104 103 103 103 103 103 103 103 103 104 100 102

This work, M ¼ 4, q ¼ 0.5 2.1 1.6 6.6 2.9 9.2 1.1 4.8 1.1 8.8 4.7 1.1 1.4 2.3 2.5 1.9 9.2 2.4 1.6 4.9 2.9 2.5 2.2 3.7 3.7 1.9 2.3 4.6 4.4 1.8 1.5 0.0 6.6

3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3

102 102 103 103 103 102 103 102 103 103 103 103 103 103 103 104 104 103 103 103 104 103 103 103 103 103 103 103 103 103 100 103

Cremers and Birkebak and spline interpolation methods, and for I3 the value of r obtained with the spline interpolation method is about 3 to 10 times higher than ours for N  30, and that of the Cremers and Birkebak method is also higher. Note that in contrast to the results shown in Table II, the spline interpolation method now becomes the worst for profile I3 with random noise. When S ¼ 0.1, results obtained with all three methods are unacceptable due to the high level of noise in the data, which is not very realistic in experiments. If such a case is encountered, a smoothing technique should be implemented before the inversion. From these examples, it is evident that our method yields much better results than other methods for data either with different numbers of points or with different levels of noise; even for N  15, our method is still very accurate, almost ten times as accurate as the spline interpolation method for the offaxis peak profile I3. Experimental Results. It would be of great interest to test the new method further with experimental data as all algorithms should ultimately be used in practice. The plasma intensities for the inversion were collected at several layers of a dc argon arc burning free at atmospheric pressure. The arc current was 200 A, and the distance from the cathode tip to the anode plate was 5 mm. Light from the arc was imaged at a magnification of 2:1 onto a plane parallel to the arc axis, where an optical fiber was moved by a motor across the axis in the horizontal direction. The light in the fiber was then collimated onto the entrance slit of a spectrograph and recorded with a charge-coupled device (CCD). In such a way, chordal

APPLIED SPECTROSCOPY

705

TABLE II. Standard deviations obtained for I1 to I3. Data for the spline interpolation method were taken from gc(r) in Ref. 14. S¼0 Cremers and Birkebak, 4th deg

N I1

S ’ 0.00289

Spline interpolation

This work, M ¼ 4, q ¼ 0.6

Cremers and Birkebak, 4th deg

Spline interpolation

This work, M ¼ 4, q ¼ 0.6

10 20 30 40 50 100

1.8 4.0 1.3 7.8 6.5

3 3 3 3 3

101 104 104 105 105

1.3 6.8 1.9 4.2 4.3 2.5

3 3 3 3 3 3

103 104 104 105 105 106

9.3 1.0 9.9 9.7 9.6 9.2

3 3 3 3 3 3

104 103 104 104 104 104

1.8 6.7 3.7 6.0 5.7

3 3 3 3 3

101 103 103 103 103

1.5 4.5 2.8 4.1 3.4 3.2

3 3 3 3 3 3

102 103 103 103 103 103

3.7 9.1 2.3 3.0 1.9 1.3

3 3 3 3 3 3

103 103 103 103 103 103

10 20 30 40 50 100

1.8 1.5 9.2 7.6 7.0

3 3 3 3 3

101 103 104 104 104

2.5 9.5 2.9 1.6 1.5 2.7

3 3 3 3 3 3

102 104 104 104 104 105

1.9 2.4 2.3 2.3 2.3 2.2

3 3 3 3 3 3

103 103 103 103 103 103

1.8 4.2 7.1 7.0 5.3

3 3 3 3 3

101 103 103 103 103

2.4 5.0 7.0 5.3 3.2 5.7

3 3 3 3 3 3

102 103 103 103 103 103

5.6 5.7 6.0 4.9 2.5 2.9

3 3 3 3 3 3

103 103 103 103 103 103

10 20 30 40 50 100

2.1 3.5 1.8 1.3 1.1

3 3 3 3 3

101 103 103 103 103

1.2 3.5 2.4 4.1 3.8 5.5

3 3 3 3 3 3

101 103 103 104 104 105

6.9 7.1 6.9 6.7 6.6 6.4

3 3 3 3 3 3

103 103 103 103 103 103

2.1 1.0 1.0 9.9 1.1

3 3 3 3 3

101 102 102 103 102

1.2 4.2 5.2 5.0 5.6

3 3 3 3 3

101 103 103 103 103

1.4 8.8 8.0 8.4 7.9 7.5

3 3 3 3 3 3

102 103 103 103 103 103

I2

I3

profiles in Fig. 7c are similar to those shown in Fig. 7b. This indicates that the original data were not distorted by the polynomial fitting technique, and thus the results obtained are reliable. The profiles inverted with our method are smooth and continuous; the noise in the original data is completely suppressed due to the excellent smoothing property of the polynomial least squares fitting technique. It must be noticed that the first three profiles are off-axis peak distributions and the last one belongs to the bell-shaped type. Good results obtained for these profiles again validate the ability of our method for successfully applying to experimental data with various shapes that are commonly encountered in plasma diagnostics.

projections from different layers of the arc were measured. Figure 7a shows the intensity distributions of the Ar I 696.5 nm line measured from layers of the arc with distances of 0.5, 1.5, 2.5, and 3.5 mm from the cathode tip; the number of data points for these profiles is 40, 50, 60, and 60, respectively. No interpolation was used on the experimental data as the number of data points is large enough. For the inversion, we use K ¼ 5, M ¼ 6. The noise in the measured data was estimated to have a standard deviation of S ¼ 0.007, and considering the long tails in the distributions of the intensity profiles, the overlapping factor was therefore determined as 0.35. The results inverted with the Nestor–Olsen method4 and the algorithm presented in this paper are shown in Figs. 7b and 7c, respectively. It is evident that the distributions of the

TABLE III. Standard deviations obtained for I2 and I3 with different levels of noise. Data for the spline interpolation method were taken from Ref. 15. Data from this work were inverted with q ¼ 0.6, 0.7, and 0.8 for S ¼ 0.005, 0.01, and 0.1, respectively. I3

I2 N S ¼ 0.005 5 10 15 30 50 S ¼ 0.01 5 10 15 30 50 S ¼ 0.1 5 10 15 30 50

706

Cremers and Birkebak, 4th deg

Spline interpolation

This work, M¼4

Cremers and Birkebak, 4th deg

Spline interpolation

This work, M¼4

2.3 1.5 1.3 7.8 6.6

3 3 3 3 3

102 102 102 103 103

4.0 2.2 6.7 5.6 3.9

3 3 3 3 3

102 102 103 103 103

9.8 1.5 5.0 6.2 9.3

3 3 3 3 3

103 102 103 103 103

2.1 1.7 1.5 7.9 6.7

3 3 3 3 3

102 102 102 103 103

1.2 1.1 3.5 1.8 9.2

3 3 3 3 3

101 101 102 102 103

1.4 1.8 6.0 7.1 9.5

3 3 3 3 3

102 102 103 103 103

3.6 2.6 2.5 1.5 1.3

3 3 3 3 3

102 102 102 102 102

4.5 2.3 2.3 1.1 7.4

3 3 3 3 3

102 102 102 102 103

1.7 3.0 5.4 8.2 1.2

3 3 3 3 3

102 102 103 103 102

2.9 2.8 2.7 1.5 1.3

3 3 3 3 3

102 102 102 102 102

1.2 1.1 1.2 2.9 1.3

3 3 3 3 3

101 101 101 102 102

2.1 3.2 8.4 9.7 1.2

3 3 3 3 3

102 102 103 103 102

3.5 2.5 2.4 1.5 1.3

3 3 3 3 3

101 101 101 101 101

5.2 1.0 9.3 5.2 6.2

3 3 3 3 3

102 101 102 103 102

1.6 2.2 5.6 5.5 9.8

3 3 3 3 3

101 101 102 102 102

3.4 2.5 2.4 1.5 1.3

3 3 3 3 3

101 101 101 101 101

2.8 1.2 7.8 8.6 1.1

3 3 3 3 3

101 101 102 102 101

1.6 2.2 5.4 5.2 1.0

3 3 3 3 3

101 101 102 102 101

Volume 62, Number 6, 2008

parameters, and it is superior to other methods, such as those using the techniques of polynomial least squares fitting and spline interpolation, especially for data with a very limited number of points. Excellent results were obtained with the proposed method being applied to experimental data without removal of noise. The good performance of the method is achieved due to the following aspects. First, polynomial least squares fitting greatly reduces the effect of noise, and thus a good inversion can be obtained even for data with experimental uncertainties. Second, dividing the interval of data measurement into several segments ensures that the measured profile is not distorted by the fitting process and at the same time improves the approximation accuracy; therefore, the method is possible for applying to profiles with various shapes with a high inversion accuracy. Lastly, the use of the overlapping segment technique not only avoids termination errors but also increases the number of data points for fitting, which leads to more accurate inversions even for small sets of data. Consequently, the method is more suitable than other ones for applying to experimental data, for which the noise filtering process is not required. ACKNOWLEDGMENTS This work is supported by the Chinese Natural Science Foundation under Contract No. 50775053. The comments of the anonymous reviewers are gratefully acknowledged.

FIG. 7. Distributions of (a) experimental intensities measured from different layers of an arc plasma, and the corresponding radial emission coefficients inverted with (b) the Nestor–Olsen method and (c) the algorithm proposed in this paper with K ¼ 5, M ¼ 6, and q ¼ 0.35.

CONCLUSION A versatile analytical expression as well as its Abel inverse was presented for reconstruction of plasma emission coefficients from projected intensities with noise. The investigation with simulated data indicated that the method can suppress noise very effectively by using the optimized combination of

1. P. Fauchais and A. Vardelle, IEEE Trans. Plasma Sci. 25, 1258 (1997). 2. U. Kogelschatz, Plasma Phys. Control. Fusion 46, B63 (2004). 3. H. R. Griem, Principles of Plasma Spectroscopy (Cambridge University Press, Cambridge, 1997). 4. O. H. Nestor and H. N. Olsen, SIAM Rev. 2, 200 (1960). 5. K. Bockasten, J. Opt. Soc. Am. 51, 943 (1961). ´ lvarez, A. Rodero, and M. C. Quintero, Spectrochim. Acta, Part B 57, 6. R. A 1665 (2002). 7. K. Tatekura, Appl. Opt. 22, 460 (1983). 8. M. Kalal and K. A. Nugent, Appl. Opt. 27, 1956 (1988). 9. L. M. Smith, D. R. Keefer, and S. I. Sudharsanan, J. Quant. Spectrosc. Radiat. Trans. 39, 367 (1988). 10. J. Dong and R. J. Kearney, J. Quant. Spectrosc. Radiat. Trans. 46, 141 (1991). 11. S. Ma, H. Gao, and L. Wu, Appl. Opt. 47, 1350 (2008). 12. J. Glasser, J. Chapelle, and J. C. Boettner, Appl. Opt. 17, 3750 (1978). 13. Y. E. Voskoboinikov and N. G. Preobrazhenskii, Opt. Spectrosc. 60, 111 (1986). 14. M. Deutsch and I. Beniaminy, J. Appl. Phys. 54, 137 (1983). 15. A. Sa´inz, A. Dı´az, D. Casas, M. Pineda, F. Cubillo, and M. D. Calzada, Appl. Spectrosc. 60, 229 (2006). 16. G. N. Minerbo and M. E. Levy, SIAM J. Numer. Anal. 6, 598 (1969). 17. X.-F. Li, L. Huang, and Y. Huang, J. Phys. A: Math. Theor. 40, 347 (2006). 18. S. Ma, H. Gao, G. Zhang, and L. Wu, J. Quant. Spectrosc. Radiat. Trans. 107, 61 (2007). 19. C. J. Cremers and R. C. Birkebak, Appl. Opt. 5, 1057 (1966). 20. S. A. Yousefi, Appl. Math. Comput. 175, 574 (2006). 21. G. E. Andrews, R. Askey, and R. Roy, Special Functions (Cambridge University Press, Cambridge, 1999). 22. M. J. Buie, J. T. P. Pender, J. P. Holloway, T. Vincent, P. L. G. Ventzek, and M. L. Brake, J. Quant. Spectrosc. Radiat. Trans. 55, 231 (1996). 23. G. C.-Y. Chan and G. M. Hieftje, Spectrochim. Acta, Part B 61, 31 (2006). 24. R. S. Anderssen, J. Inst. Math. Appl. 17, 329 (1976).

APPLIED SPECTROSCOPY

707

spectroscopic techniques

A Strategy to Prevent Signal Losses, Analyte Decomposition, and Fluctuating Carbon Contamination Bands in Surface-Enhanced Raman Spectroscopy BOON-SIANG YEO, THOMAS SCHMID, WEIHUA ZHANG, and RENATO ZENOBI* Department of Chemistry and Applied Biosciences, ETH Zurich, 8093 Zurich, Switzerland

Signal losses and fluctuating carbon contamination bands are ‘‘bottlenecks’’ in the application of surface-enhanced Raman spectroscopy (SERS) for reliable chemical analysis. They originate mainly from prolonged laser irradiation of the sample during data collection, which causes analyte decomposition and/or loss of the enhancing capabilities of the adsorption site. In this work, a laser illumination/signal collection technique, the ‘‘multiple points collection’’ (MPC) method is introduced to circumvent these problems. The MPC method is based on the use of a pair of galvanic mirrors to scan the laser beam rapidly and steadily across the sample surface. Each position is irradiated for ,10 ls, at a rate of ;0.5 Hz. The SER spectrum is obtained by summing the signals collected from a large array of non-overlapping sample points. The MPC is compared with the conventional ‘‘single point collection’’ method, in which the laser beam is statically focused onto a particular spot and the scattered signals acquired. The MPC has the following advantages: (1) illumination and collection efficiencies are not compromised, (2) signal losses originating from analyte decomposition and/or alteration of the enhancing capabilities of the adsorption site are avoided, (3) high-quality SER spectra for analytes such as biomolecules and dipicolinic acid (a common marker for bacteria spores) can be easily obtained, and (4) the occurrence of broad amorphous carbon bands and the commonly observed temporal fluctuations in SERS are prevented. The success of the MPC is attributed to the reduction of local sample heating, as the time interval between the laser irradiations of a spot is much longer than the actual irradiation time itself. Index Headings: Chemical analysis; Surface-enhanced Raman spectroscopy; SERS; Laser scanning confocal microscope; Amorphous carbon.

INTRODUCTION Surface-enhanced Raman spectroscopy (SERS) has gained prominence as a powerful tool for the chemical analysis of molecules adsorbed on noble metal nanostructures.1–6 As a result of its sensitivity heading towards reproducible single molecule detection, great efforts have been made to develop SERS platforms for chemical and biological sensing. These advances are crucial for rapid monitoring of harmful biological agents and explosives and will also lead to progress in proteomics and drug discovery. A common way of performing chemical analysis using SERS is to deposit the analyte on a roughened Ag substrate and Received 26 December 2007; accepted 6 March 2008. * Author to whom correspondence should be sent. E-mail: zenobi@org. chem.ethz.ch.

708

Volume 62, Number 6, 2008

illuminate the sample with a visible wavelength laser. When the wavelengths of the laser and metal surface plasmons are in resonance, strong local electromagnetic fields are created that will amplify the Raman signatures of the molecules. The scattered signal can then be analyzed spectroscopically. The most popular setup is the Raman microscope because of its superior detection efficiency.4,5,7–9 High numerical aperture (N.A.) objectives are generally used not only to improve spatial resolution, but more importantly to optimize the SERS enhancements (through the angle of the incident beam) and light collection.10–13 This reduces the signal collection time, which is crucial for the optimum performance of fast sensors. The usefulness of SERS for chemical analysis lies in its ability to give a high-quality vibrational spectrum of the analyte. However, this aim is often frustrated by SERS signal losses during the experiment. The reasons underlying the losses are still points of contention in the SERS community. A commonly invoked factor is the decomposition of the analyte triggered either by laser heating during the experiment, by catalytic activity of the substrate, or from contaminants adsorbed on the metal surface during substrate preparation.7,9,10,14,15 As a result, amorphous carbon, manifested as two broad bands at ;1360 cm1 (D-band) and ;1580 cm1 (G-band), or as rapidly fluctuating peaks, is frequently observed (spectral fluctuations). The latter is more problematic as the peaks occur randomly in duration, Raman shift, and intensity. They cannot be easily identified and subtracted away from the signal of the analyte.16 Morphological and chemical changes of the enhancing sites due to laser irradiation may also deteriorate their SERS activity and cause signal losses.9,17 These interferences create problems for routine chemical analysis by SERS. To circumvent these problems, low N.A. objectives, low incident laser powers, or non-resonance Raman spectroscopy can be used.7,10 These plausible solutions reduce the incident photon flux or mitigate the effects of laser-induced damage on the sample. However, their application will generally result in a longer spectra collection time. Carbon can be removed by chemical displacement using self-assembled thiol monolayers, rinsing samples in solvents, and/or O2 and Ar plasma cleaning, etc.18,19 However, these cleaning methods are neither suitable for all types of metal surfaces nor do they assure cleanliness of the metal substrates for an extended period of time.

0003-7028/08/6206-0708$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

Contamination can still appear during the analysis. The lack of suitable remedies for SERS carbon contamination has led to its classification as an ‘‘omnipresent’’ problem.10 In this work, a strategy to circumvent the problems highlighted above is presented, namely to prevent SERS signal losses and to suppress the fluctuating carbon contamination signals. A micro-Raman system based on a laser scanning confocal microscope (LSCM) is used for collecting the spectra. The heart of the instrument consists of a pair of galvanic mirrors that scan the laser beam rapidly and steadily across the sample plane.20,21 Vapor-deposited Ag thin films are used as SERS substrates due to their popularity and ease of preparation, but this methodology is applicable to any type of SERS substrate. To acquire a SER spectrum using a LSCM, we first obtain an image of the Ag surface (with detection of the backscattered light by a photomultiplier tube (PMT)). The laser is then focused onto a particular spot, from which a SER spectrum is collected. This method is equivalent to focusing light onto the sample using a commercial Raman microscope, widely used by the SERS community.2,4,5,7–9 It will be termed here the single point collection (SPC) method. The method presented here for the acquisition of a SER spectrum involves confocal imaging of the surface during the active collection time, but transmitting the backscattered light into the Raman spectrometer rather than into the PMT of the LSCM. As opposed to the SPC mode, the laser beam is scanned rapidly over a larger area, with the signal being continuously collected to give a single spectrum. This method will be termed as the multiple points collection (MPC) method. This approach leads to a reduction of local sample heating because the time intervals between the laser irradiations of a spot are much longer than its actual irradiation time. It will be demonstrated that the illumination/collection efficiency of the MPC and SPC methods is quantitatively identical. SERS signal losses, analyte decomposition, and fluctuating carbon contamination signals are shown to be easily prevented using the MPC method. High-quality SER spectra of several molecules, including, for the first time, the antibody immunoglobulin M, are presented here.

EXPERIMENTAL Chemicals and Sample Preparation. For the SERS experiments, glass slides (Paul Marienfeld GmbH & Co. KG, Germany) were cleaned in piranha solution (concentrated H2SO4:H2O2, 3:1) for 10 minutes. After rinsing with methanol and blowing them dry using N2 gas, these were mounted into a vapor-coating chamber (MED 020, Bal-Tec, Liechtenstein). Then, 6 nm Ag (Bal-Tec, 99.99%) was deposited on to the slides at a rate of 0.05 nm/s. The pressure of the system is ,1 3 105 mbar during the coating process. The analytes used were brilliant cresyl blue dye (BCB, Fluka), pinacyanol chloride (PC, Aldrich), dipicolinic acid (DPA, 99%, Aldrich), cytochrome c (Cc, from bovine heart, 95%, Fluka), and immunoglobulin M (IgM, from human serum, ;95%, Sigma). The two dyes were dissolved in methanol, while DPA and Cc were dissolved in H2O. IgM was purified of its buffer salts using desalting columns (Micro BioSpin with Bio-Gel P-30, Bio-Rad, Hercules, CA) and resolvated in H2O. The analyte was spin coated onto the Ag coated glass slides immediately after removing the slides from the coating chamber.

FIG. 1. Raman spectra of BCB on a glass slide collected using MPC (solid trace) and SPC (dotted trace). The incident laser power is 7 lW and the spectrum collection time is 10 s.

Instrumentation. Our home-built setup has been described elsewhere.22 It consists of an inverted LSCM (IX 70, Olympus, Japan) and a Raman spectrograph/charge-coupled device (CCD) detector (Kaiser HoloSpec, Ann Arbor, MI). A 1.4 N.A. oil-immersion objective was used to focus light from a 532 nm diode-pumped solid-state laser (Ventus, Laser Quantum, UK) onto the samples. The laser spot size is ;400 nm. Back-scattered light was collected through the same objective. By means of a flipping mirror, the photons can be fed either into the Raman spectrograph through fiber optics or to the PMT (through a pinhole) to give an image of the surface. The aperture of the fiber and the pinhole maintain confocality whether light is channeled to the spectrograph or to the PMT. All the experiments were done under ambient conditions. For the MPC method, the LSCM is made to scan an area of ;210 3 210 lm2 with a resolution of 512 3 512 pixels. Hence, each pixel corresponds to ;410 3 410 nm2, roughly equivalent to the size of the laser focus. A scan time of ;2 s was used for each frame, i.e., each pixel is scanned at ;0.5 Hz, with an exposure of ,10 ls. No additional modification of the instrument is required. Alternatively for the SPC method, the laser is focused onto randomly chosen points. The MPC and SPC methods were always performed consecutively on the same sample. No background subtraction or smoothing of the spectra was carried out. The data points were processed using Igor Pro Version 4.09A Carbon (WaveMetrics, Inc., Portland, OR).

RESULTS Validation of the Multiple Points Collection Method. For the MPC method to be useful, the illumination/collection efficiency of this method should be the same as that for the conventional SPC method. This is an important consideration that can be inadvertently neglected in the quest for scanning methods that better maintain sample integrity. For example, spinning a sample during an experiment may prevent it from heating up and decomposing, but mechanical instability from

APPLIED SPECTROSCOPY

709

FIG. 2. (a) Time-sequence plots of SER spectra of BCB on a Ag surface collected using MPC (solid traces) and SPC (dotted traces). Forty consecutively acquired spectra are shown. (b) SER spectra of PC on a Ag surface collected using MPC (solid traces) and SPC (dotted traces). The upper traces are the first set of spectra collected when the laser irradiation is first initiated, while the lower ones (the third set) are obtained ;6 s later.

710

Volume 62, Number 6, 2008

its movement can also lead to laser defocusing and hence signal losses. Figure 1 shows the Raman spectra of a homogeneous BCB thin film spin coated on a glass slide collected by both the MPC and SPC methods. BCB was chosen as the sample as it is completely stable at low laser irradiation; 7 lW of incident laser was utilized. Normal Raman spectroscopy instead of SERS was performed in order to eliminate unknown experimental parameters such as surface distribution of enhancing hot spots, which may invalidate the quantitative aspects of this test. It could be observed that the band intensities and signal-to-noise ratios (S/N: defined as the ratio of intensity of the desired signal over the background noise) are the same for both spectra, regardless of whether the MPC or SPC is used. Preventing Surface-Enhanced Raman Spectroscopy Signal Losses: Essential for Quantitative Analysis. Quantitative analysis using SERS surfaces is generally difficult because the prolonged focusing of the laser beam on a single spot leads to photo-induced and thermal decomposition of the molecules.7 Physical/chemical changes of the adsorption sites may also cause them to lose their enhancing capabilities.9 In addition, randomly situated SERS hot spots will give locally very highly enhanced Raman signals that do not reflect the actual physical concentration of the analyte. An example of the MPC and SPC methods applied to BCB deposited on Ag films is presented in Fig. 2a. BCB was again chosen as the analyte because we found that its decomposition products do not give strong amorphous carbon bands that will interfere with the quantitative aspects of this analysis.23 This is in contrast with analytes such as malachite green isothiocyanate.15 One of BCB’s vibrational peaks at 578 cm1 was used to monitor the intensity changes. The sample was illuminated with a high laser power of 150 lW and SER spectra were continuously acquired with a collection time of 1 ms (with 2–3 s intervals). The BCB signals collected with the MPC method can be observed to be stable, while those using the SPC have dropped to ,1/2 of their original value in less than 2 minutes. A more dramatic example is shown in Fig. 2b. Here, SERS of PC, a photosensitive dye, is investigated with the same laser power and collection time as the example presented above for BCB. In the very first set of SER spectra obtained, the signals collected by the SPC method have less than half of the intensity compared to those collected by the MPC method. This can be attributed to any of the three factors mentioned above. More importantly, by the third set of spectra acquired (in ,8 s of laser irradiation), the signal intensities in the SPC spectrum are almost at the background level, while those acquired with the MPC method have not changed. Rapid signal losses of PC caused by analyte decomposition and/or loss of SERS activity of the substrate when the SPC method is used is clearly exhibited here. These two experiments demonstrate that the MPC method better ensures the integrity of the analyte and the substrate, hence facilitating quantitative analysis using SERS. The MPC method of collecting spectra also gives ensemble information over the entire scanned surface. Unlike the SPC method, it avoids strong dependence on randomly located hot spots present on the metal surface or inhomogeneous analyte distribution. This ensures a more accurate measurement of the quantity of analyte present. Preventing Analyte Decomposition: Essential for Qualitative Analysis. The SER spectra of DPA (a marker for

FIG. 3. SER spectra of (a) DPA (incident laser power 150 lW, collection time 1 s), (b) Cc (incident laser power 150 lW, average of three spectra, collection time 10 s each), and (c) IgM (incident laser power 65 lW, average of six spectra, collection time 10 s each). The solid and dotted traces denote the spectra collected using the MPC and SPC methods, respectively. Note that for DPA, its degradation occurs even within a very short laser irradiation time.

bacteria spores, including those of Bacillus anthracis) and two biologically important molecules, Cc and IgM, are presented in Fig. 3. Cc exhibits an optical resonance at 532 nm, while DPA and IgM do not. Using the MPC method, high-quality SER spectra were obtained for all three analytes. The spectra of DPA and Cc are in excellent agreement with previous studies,24,25 while to the best of our knowledge, this is the first reported SER spectrum of IgM. In contrast, when the SPC method is applied, analyte signal losses, most probably originating from decomposition of the analytes, can be observed. This is corroborated by the presence of vibrational features attributable to amorphous carbon bands (e.g., broad peaks at ;1360 cm1 and ;1580 cm1). However, the carbon signals may not originate solely from

FIG. 4. SER spectral changes of a Ag surface as a function of irradiation time. Spectra are collected using (a) SPC (dotted traces) and (b) MPC (solid traces). Every 5th spectrum acquired is presented. The averaged SER spectra (bold traces) calculated from 28 recorded spectra using either the SPC or MPC method are presented at the bottom of each set of spectra. The incident laser power is 150 lW and the spectrum collection time is 1 s.

the decomposed analyte. They could also be from the decomposition of residual contaminants present on the Ag substrate even before analyte deposition.9 Amorphous carbon has a Raman cross-section four orders of magnitude larger than

APPLIED SPECTROSCOPY

711

that of benzene and is also known to have broad molecular resonance from the UV to the near-infrared.1,26 Thus, signals from amorphous carbon will overwhelm those from the analyte even if the latter had not decomposed. This renders any chemical analysis difficult; a solution is presented in the next section. Preventing the Occurrence of Amorphous Carbon Bands and Spectral Fluctuations. The most difficult issue facing SERS analysis is the ubiquitous carbon contamination.10 Figure 4a shows a series of spectra collected with the laser focused (SPC) onto a Ag film without any analyte. The spectra are collected with 1 s accumulation time and show temporally fluctuating signals. When they are averaged, two broad bands centered at 1360 cm1 and 1590 cm1 are obtained. These are assigned to amorphous carbon. This observation is not unexpected because carbon contamination cannot be completely avoided in the vapor-deposition process.9 In contrast, when the MPC method is applied on the same Ag film, SER spectra with well-defined peaks with fairly constant intensities are found (Fig. 4b). Most importantly, they do not exhibit random and temporal shifts in Raman frequencies. The vibrational features of the averaged spectrum do not resemble the broad amorphous carbon bands shown in Fig. 4a. The two intense peaks at 1146 and 1572 cm1 can be respectively assigned to the C–C and C¼C stretching modes of carbon polyenes containing six conjugated double bonds, while the weaker ones are ascribed to graphitic sheets.27,28 An insight into the physical origin of the commonly observed amorphous carbon SER bands can be obtained from this experiment. The two characteristic broad bands at 1360 and 1590 cm1 were observed only in the averaged spectrum for the SPC method. This reveals that prolonged focusing of the laser beam at a particular spot (SPC) plays a role in generating these bands, probably through decomposition of the distinct carbonaceous molecular precursors such as polyenes (polyenes have been spectroscopically recorded as intermediates during the laser-induced conversion of polydiacetylene to amorphous carbon on Ag surfaces).27 This is corroborated by the absence of these bands when the MPC method was used. If amorphous carbon had been present prior to sample irradiation, then it should be more easily observed with the MPC method because a larger area is sampled. Spectral fluctuations ceased in the present work when the MPC method was utilized. For contaminants that are firmly adsorbed on the surface, e.g., through strong Ag–S bonds, the implication here is that their vibrational signatures can be first identified and subsequently subtracted from the analyte spectrum. This was hitherto not possible because of the frequent presence of rapidly fluctuating signals when the SPC was used. The MPC will be particularly helpful in the study of analytes at low concentrations or compounds that have very weak Raman scattering cross-sections.

DISCUSSION Multiple Points Collection for Chemical Analysis Based on Surface-Enhanced Raman Spectroscopy. Surface-enhanced Raman spectroscopy signal losses, analyte decomposition, and amorphous carbon bands are great obstacles in SERS-based chemical analysis. However, rather little effort has been made to overcome these problems. At present, the common practice for collecting a good SER spectrum in a reasonable time is to tediously balance the excitation laser

712

Volume 62, Number 6, 2008

power and collection time or to use low N.A. objective/lenses. These measures usually sacrifice excitation/collection efficiency.7,10,15 The effectiveness of the MPC method is attributed to the reduction of local sample heating, as the time interval between the laser irradiations of one pixel is much longer than the actual irradiation time itself. This scanning configuration distributes sequentially the total photon dose over space, which facilitates heat dissipation. The amount of power a sample can tolerate also increases when light is distributed on it through many nonoverlapping spots (MPC), rather than through one big spot (both with a fixed total area).29 As a further illustration of the negative effects of prolonged laser heating, the quality of some of the SER spectra in this work deteriorated (e.g., signal intensity losses and/or band broadening were observed) when the ls irradiation time of each pixel was lengthened by just one order of magnitude. In the light of the present results, the effect of prolonged vs. intermittent laser irradiation of a sample was investigated. The fluorescence and Raman intensities of Rhodamine 6G and PC thin films were respectively measured. A sample spot was irradiated either continuously (A s) or intermittently (A/B s 3 B number of times), while maintaining the total photon dose. If a sample is degraded purely through photo effects, no intensity differences would be expected regardless of the way the photons are temporally distributed. This is because each absorption–emission cycle has the same probability of causing photobleaching. From our preliminary studies, increasing the intermittent irradiation of a sample enhances its stability by 30%. This confirms the destructive effects of continuous laser irradiation, i.e., heating, on the integrity of a sample. Investigations to elucidate the thermal and photo contributions towards the degradation of a molecule are now being pursued in our laboratory. An alternate setup for the MPC method could be realized with a rapid, stable translation of the sample (on a scan stage) with respect to a stationary laser beam.30 However, stages that meet the requirements for speed and stability presented in this work are not commercially available. Mechanical stability is crucial for the reproducible focusing of the laser beam onto the sample surface, especially when tight focusing of the laser spot and confocality of the microscope are used. In general, movements between the sample and focusing objective should be avoided, as even small perturbations may cause laser beam defocusing and hence loss of signals. There are other scanning configurations designed for normal Raman spectroscopy that could be useful in solving the problems of SERS signal losses and spectral fluctuations, e.g., line and global illumination scan modes.31–33 These alternatives can be home-built or commercially purchased. However, the MPC method enjoys advantages over some of these apparatuses, such as a faster scan speed, better signal collection efficiency, no loss of confocality (which occurs for some instruments using line illumination scans), etc. Full Raman imaging of samples could in principle be done in the MPC mode by the use of a CCD detector that can collect and read out at ls speed. A summation of signals collected from a particular pixel (after a certain number of scans over the entire area) could then be obtained to give a better S/N spectrum. To the best of our knowledge, such CCD detectors are not yet commercially available. It is, however, emphasized here that chemical identification by SERS for biological and

chemical applications usually do not require high spatial resolution detection. This is because the analyte is randomly spread over the SERS substrate during sample preparation or when flowing through the sensing device.4,6,34 A final remark on highly robust SERS sensors: metallic nanostructures are often used to build such sensors and prolonged laser illumination can lead to alteration of their optical properties. We predict that the MPC strategy, demonstrated through this work to be effective in preventing signal losses, will play an important role in the wider usage of SERS sensors in research, in industry, and in the consumer market.

CONCLUSION In this work, common problems in SERS such as signal losses and fluctuating carbon contamination bands have been overcome using the multiple points collection method. Illumination and detection efficiencies are not compromised by the MPC method. High-quality and clean SER spectra of DPA, Cc, and IgM have been obtained. The MPC method also prevented the occurrence of fluctuating carbon bands. This allows the identification of pre-adsorbed contamination, whose signals can be subtracted from the SER spectrum after adding the analyte. The MPC method is more than an effective collection method: the physical origin of amorphous carbon SER bands has been better understood by using it. The success of the MPC method is attributed to the reduction of local sample heating, as the time interval between the laser irradiations of one pixel (;2 s) is much longer than the actual irradiation time (,10 ls) itself. Higher laser power can also be more easily used with the MPC method, which will lead to a shorter spectrum collection time. The use of galvanic mirrors to scan the laser beam assures stable focusing of the laser spot on the sample. Finally, from the series of experiments performed for this work, the signal intensity of the analytes collected by the MPC was observed either to be the same (when low incident laser power or a very robust sample is used) or often higher compared to that using the conventional SPC method. Carbon contamination from analyte decomposition was hardly observed when the MPC method was employed. This demonstrates that signal losses, sample decomposition, and contamination, which are bottlenecks hindering the development of SERS, can be more easily overcome than commonly held. Since the MPC method can be easily implemented, we hope that it will soon add to the arsenal of experimental techniques that will render SERS a highly reliable and quantitative chemical analysis method.

1. R. Aroca, Surface-Enhanced Vibrational Spectroscopy (John Wiley and Sons, Ltd, West Sussex, 2006), 1st ed. 2. J. A. Dieringer, A. D. McFarland, N. C. Shah, D. A. Stuart, A. V. Whitney, C. R. Yonzon, M. A. Young, X. Y. Zhang, and R. P. Van Duyne, Faraday Discuss. 132, 9 (2006). 3. T. H. Reilly, S. H. Chang, J. D. Corbman, G. C. Schatz, and K. L. Rowlen, J. Phys. Chem. C 111, 1689 (2007). 4. G. Braun, S. J. Lee, M. Dante, T. Q. Nguyen, M. Moskovits, and N. Reich, J. Am. Chem. Soc. 129, 6378 (2007). 5. D. R. Ward, N. K. Grady, C. S. Levin, N. J. Halas, Y. P. Wu, P. Nordlander, and D. Natelson, Nano Lett. 7, 1396 (2007). 6. X. Y. Zhang, J. Zhao, A. V. Whitney, J. W. Elam, and R. P. Van Duyne, J. Am. Chem. Soc. 128, 10304 (2006). 7. I. Khan, E. Polwart, D. W. McComb, and W. E. Smith, Analyst (Cambridge, U.K.) 129, 950 (2004). 8. R. M. Sto¨ckle, V. Deckert, C. Fokas, and R. Zenobi, Appl. Spectrosc. 54, 1577 (2000). 9. M. L. Jacobson and K. L. Rowlen, J. Phys. Chem. B 110, 19491 (2006). 10. N. P. W. Pieczonka and R. F. Aroca, Chem. Phys. Chem. 6, 2473 (2005). 11. A. Wei, B. Kim, B. Sadtler, and S. L. Tripp, Chem. Phys. Chem. 2, 743 (2001). 12. K. A. Bosnick, J. Jiang, and L. E. Brus, J. Phys. Chem. B 106, 8096 (2002). 13. I. Baltog, M. Baibarac, and S. Lefrant, Phys. Rev. B 72, 2452402 (2005). 14. A. Kudelski and B. Pettinger, Chem. Phys. Lett. 321, 356 (2000). 15. K. F. Domke, D. Zhang, and B. Pettinger, J. Phys. Chem. C 111, 8611 (2007). 16. Y. C. Liu and R. L. McCreery, J. Am. Chem. Soc. 117, 11254 (1995). 17. W. Zhang, T. Schmid, B.-S. Yeo, and R. Zenobi, J. Phys. Chem. C 112, 2104 (2008). 18. K. L. Norrod and K. L. Rowlen, Anal. Chem. 70, 4218 (1998). 19. Y. Saito, J. J. Wang, D. N. Batchelder, and D. A. Smith, Langmuir 19, 6857 (2003). 20. R. A. Farrer, M. J. R. Previte, C. E. Olson, L. A. Peyser, J. T. Fourkas, and P. T. C. So, Opt. Lett. 24, 1832 (1999). 21. S. W. Paddock, Mol. Biotechnol. 16, 127 (2000). 22. C. Vannier, B. S. Yeo, J. E. Melanson, and R. Zenobi, Rev. Sci. Instrum. 77, 023104 (2006). 23. W. H. Zhang, B. S. Yeo, T. Schmid, and R. Zenobi, J. Phys. Chem. C 111, 1733 (2007). 24. P. Hildebrandt and M. Stockburger, J. Phys. Chem. 90, 6017 (1986). 25. S. E. J. Bell, J. N. Mackle, and N. M. S. Sirimuthu, Analyst (Cambridge, U.K.) 130, 545 (2005). 26. J. C. Tsang, J. E. Demuth, P. N. Sanda, and J. R. Kirthey, Chem. Phys. Lett. 76, 54 (1980). 27. K. Itoh, I. Kudryashov, J. Yamagata, T. Nishizawa, M. Fujii, and N. Osaka, J. Phys. Chem. B 109, 271 (2005). 28. H. E. Schaffer, R. R. Chance, R. J. Silbey, K. Knoll, and R. R. Schrock, J. Chem. Phys. 94, 4161 (1991). 29. D. M. Zhang, J. D. Hanna, Y. N. Jiang, and D. Ben-Amotz, Appl. Spectrosc. 55, 61 (2001). 30. M. A. De Jesus, K. S. Giesfeldt, and M. J. Sepaniak, Appl. Spectrosc. 57, 428 (2003). 31. N. Zimmerer and W. Kiefer, Appl. Spectrosc. 28, 279 (1974). 32. S. Mamedov, A. Kisliuk, S. Loheider, and D. Quitmann, Appl. Spectrosc. 49, 1199 (1995). 33. J. J. Andrew, ‘‘Raman Microscopy and Imaging’’, in Encyclopedia of Analytical Chemistry, R. A. Meyers, Ed. (John Wiley and Sons, Ltd., Chichester, 2000), vol. 15, p. 13078. 34. D. J. Anderson and M. Moskovits, J. Phys. Chem. B 110, 13722 (2006).

APPLIED SPECTROSCOPY

713

notes

Mid-Infrared Laser-Induced Breakdown Spectroscopy Emissions from Alkali Metal Halides CLAYTON S.-C. YANG,* E. BROWN, UWE HOMMERICH, SUDHIR B. TRIVEDI, ALAN C. SAMUELS, and A. PETER SNYDER Battelle Eastern Science and Technology Center, Aberdeen, Maryland 21001 (C.S.-C.Y.); Department of Physics, Hampton University, Hampton, Virginia 23668 (E.B., U.H.); Brimrose Corporation of America, Baltimore, Maryland 21152 (S.B.T.); and Edgewood Chemical Biological Center, Aberdeen Proving Ground, Maryland 21010-5424 (A.C.S., A.P.S.)

Index Headings: Mid-infrared spectroscopy; MIR spectroscopy; Atomic spectroscopy; Laser-induced breakdown spectroscopy; LIBS.

INTRODUCTION Laser-induced breakdown spectroscopy (LIBS) has recently emerged as a powerful analytical tool to determine the chemical compositions of gases, liquids, and solids. Recent advances in the LIBS technique and instrumentation in the ultraviolet (UV) to near-infrared (NIR) region have led to many applications in the fields of geology,1 remote sensing,2 metallurgy,3 and Homeland Security.4,5 It is a relatively simple technique in which an intense pulsed laser beam is focused onto a target substance to produce a short-lived plasma from the target material. Following the creation of the plasma, the hot target fragments emit intense UV to NIR emissions characteristic of the electronic transitions of the atoms, ions, and simple molecules. Important information regarding the identification and concentration of the target elements can be derived by analyzing the LIBS emission spectra. LIBS is a relatively attractive analytical spectroscopy tool for probing the elemental constituents of a sample. Some advantages of the LIBS technique include minimal sample preparation, the detection of all elemental species, and relatively high sensitivity for most of the elements. The LIBS technique can capture spectra in near real time, which makes the technique attractive for outdoor environmental investigations for elemental analytes in complex matrices.6 An important objective of this work is the recording of the observations of atomic IR emission peaks from analytes when induced by a laser-induced plasma event. The current work is a Received 4 September 2007; accepted 27 March 2008. * Author to whom correspondence should be sent. E-mail: clayton.yang@us. army.mil.

714

Volume 62, Number 6, 2008

continuation of our previous IR investigations7 using LIBS as the excitation source.

THEORY Nearly all previous LIBS analytical investigations in the literature reported atomic and molecular measurements only in the UV-NIR (200–980 nm) region. However, it is well known that molecules exhibit spectroscopic signatures in the longer wavelength mid-infrared (MIR) region due to molecular vibrational and rotational transitions. In conventional breakdown emission spectroscopy, the literature describes the use of flames to generate molecules that are excited, atomized, and ionized by the combustive thermal energy transfer. Emission features from atomic and molecular breakdown fragments were observed and monitored in the atomic (UV to near-infrared) region,8 molecular vibrational (mid-infrared) region,9 and molecular rotational (far-infrared) regions.10 Therefore, an extension of LIBS to the MIR region may provide additional information concerning the identification and classification of target substances that can augment UV-NIR LIBS measurements. There are few MIR studies that explore the atomic and molecular emissions from laser-induced plasma. In recent studies by Yang et al.,7 the MIR molecular emissions of the laser-induced breakdown products CO and CO2 molecules from various solid organic substrates were observed for the first time. These CO2 and CO molecules that emit in the MIR region result from the oxidation of laser-sputtered carbon fragments from the organic solid samples. Moreover, studies of the energy relaxation of a plasma induced by a laser from oxygen gas revealed atomic line emissions in the MIR region.11 However, no observation of MIR atomic transitions between high-lying Rydberg states from laser-induced breakdown products from condensed-phase samples have been reported in the literature. The initial phenomena involved in the formation of the breakdown plasma are different depending on the nature of the target sample.12 The laser-induced seed electrons generated from condensed-phase samples, in comparison to those created from gas samples, have more complex interactions with the environment,13 mechanisms of initiation,12 and mechanisms of energy dissipation.14

EXPERIMENTAL In this work, we report the first observation of LIBS MIR atomic line emissions from alkali metal halide solid tablets. We observed intense emission lines between 2 to 5.6 lm from various neutral alkali metal laser breakdown products. The MIR LIBS experimental setup has been reported elsewhere.7 The 1064 nm output from a Q-switched Nd:YAG laser (pulse width of 5 ns and 10 Hz repetition rate) was used as the excitation source. The pulsed laser beam was focused onto the solid sample surface with a convex lens (f ¼ 20 cm), resulting in a beam size diameter less than 0.2 mm. A beamsplitter (;4%) was employed to monitor the laser beam energy during the emission measurements. The laser energy throughout the

0003-7028/08/6206-0714$2.00/0 Ó 2008 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

TABLE I. LiCl, NaCl, KCl, RbCl, KF, and KBr assignments of dominant LIBS MIR emissions from the NIST atomic spectral database. k k (observed, lm) (literature, lm) Li Na

K

FIG. 1. Non-normalized MIR LIBS emissions between 2 and 5.6 lm from a NaCl tablet and (inset) the normalized emission transient of the 2.21 lm line.

experiment was found to be very stable, with less than 1% power fluctuation. The laser energy reaching the sample was 16 mJ/pulse and the peak intensity was on the order of 1016 W/m2. This laser parameter was sufficient to induce plasma from selected alkali metal halide tablets. The sample tablets were made from high pressure application of the vacuum-dried alkali halide powders. The vacuum-dried alkali halide powders were ground into relatively smaller-sized particles and pressed into 1-inch-diameter tablets using a hydraulic pellet press. A solid tablet was then vertically mounted onto a linear translation stage attached to a stepping motor. The speed of the spectral scan rate was 100 nm/min from 2000 to 5600 nm. The sample stage was moved at a speed of ;1.5 mm/min so the

Rb

2.45 2.69 4.05 2.21 2.34 3.42 4.05 2.71 2.73 3.15 3.18 3.65 3.68 3.73 3.76 4.03 2.26 2.30 2.74 2.80 3.99

2.45 2.69 4.05 2.21 2.34 3.41 4.04 2.70 2.72 3.11 3.16 3.63 3.66 3.70 3.73 4.02 2.25 2.29 2.73 2.79 3.99

Assigned transition 42S1/2 ! 32P1/2,3/2 32P1/2,3/2 ! 32S1/2 52G7/2,9/2 ! 42F5/2 42P3/2 ! 42S1/2 42D5/2 ! 42P3/2 52S1/2 ! 42P3/2 52G7/2,9/2 ! 42F5/2 52P3/2 ! 52S1/2 52P1/2 ! 52S1/2 52F5/2 ! 42D3/2 52P1/2 ! 32D1/2 62S1/2 ! 52P1/2 62S1/2 ! 52P3/2 42D3/2 ! 52P1/2 42D5/2 ! 52P3/2 52G7/2,9/2 ! 42F5/2 62P3/2 ! 42D5/2 62P1/2 ! 42D3/2 62P3/2 ! 62S1/2 62P1/2 ! 62S1/2 52G7/2,9/2 ! 42F5/2

Ek Ei (cm1) (cm1) 30925 27206 36630 25739 30272 30272 34586 21026 21026 27398 21536 24700 24702 24701 24720 28127 19355 19355 20132 20132 26792

35012 30925 39097 30272 34548 33200 37059 24720 24701 30606 24701 24750 24750 27398 27397 30617 23792 23715 23792 23715 29296

laser pulse would impinge upon fresh analyte materials during the 12 minute emission scan. The MIR emission from the sample was collimated and focused onto the entrance slit of a 0.15 m grating spectrometer (kblaze ¼ 4 lm, 150 gratings/mm, slits width ¼ 0.25 mm) using two CaF2 focusing lenses (f ¼ 15 cm, f ¼ 20 cm). The entrance and exit slits of the spectrometer were adjusted to 0.25 mm during the scans. A 2 lm long pass filter was placed in front of the entrance slit to block laser light scatter from entering the spectrometer. The signal was recorded by a liquid nitrogen cooled InSb detector and averaged (delay time 10 ls, 15 ls gate width) using a boxcar averager.

RESULTS AND DISCUSSION

FIG. 2. Non-normalized MIR LIBS emissions between 2 and 5.6 lm from LiCl, NaCl, KCl, RbCl, KF, and KBr tablets.

Figure 1 illustrates the 2–5.6 lm LIBS emissions from a NaCl tablet and the transient of the 2.21 lm line (inset). The 2.21 lm transient was characterized by a short-lived initial spike (0–5 ls blackbody continuum, spectrum not shown) followed by a slower decaying component (0–20 ls LIBS emission). Similar to the 2.21 lm emission line, the MIR LIBS emissions from NaCl, as well as all the metal halide samples tested in this work, generally have a decay time on the order of 20 ls after initiation of the breakdown process. Thus, the boxcar setting (delay time 10 ls, 15 ls gate width) is sufficiently delayed to gate out the fast blackbody continuum emission and produce MIR LIBS emissions for all samples tested. The dominant distinct emission features between 2–5.6 lm from NaCl originated from the neutral Na atom 42P3/2, 42D3/2, 42D5/2, and 52G7/2,9/2 energy levels. The line assignments were based on energy level measurements of sodium atoms compiled by Martin et al.15 No line emissions from chlorine atoms and ions were observed. Figure 2 illustrates the LIBS MIR emission from LiCl, NaCl, KCl, RbCl, KF, and KBr tablets. The emission features from these alkali metal halide compounds between 2–5.6 lm can be assigned to atomic transitions of neutral alkali metal (Li, Na, K, Rb) atoms. The experimental and NIST atomic spectral database16 emission wavelengths for the alkali metals are listed in Table I. The highest Rydberg state observed is the

APPLIED SPECTROSCOPY

715

laser-breakdown solid compounds demonstrates the promise of MIR LIBS to be a complement to UV-VIS LIBS as a potential chemical analysis probe.

FIG. 3. Normalized 2705 nm emission intensities from KCl, KBr, and KF as a function of laser energy. Three independent measurements for each sample were performed to determine the error bars.

lithium 52G7/2,9/2 state at 39079 cm1. Emission from the halide atoms and ions was not observed. The MIR characteristic emission of the alkali metal atoms was not influenced by the nature of the halide anion in the ionic crystal structure. To further explore the LIBS MIR emission mechanism, we studied the emission intensities from KCl, KBr, and KF as a function of laser energy, as shown in Fig. 3. The most prominent emission line from K, 52P3/2 at 2705 nm (3697 cm1), was chosen for the purpose of relatively high accuracy. When the laser energy dropped below 5 mJ (Fig. 3), the alkali line intensities from these three samples were not observed even though an intense white plasma spark was still visible. The emission threshold of the K 42P3/2 line from KF appears to be the highest among the three potassium halides, while that from the KBr sample appears to be the lowest in intensity. In the analysis of solid samples with LIBS, the laser plasma heats up the target materials to initiate the vaporization and atomization. The amount of the vaporized target material may depend on the thermal properties of the target compounds.9 The lattice energy, the amount of energy required to separate a mole of solid compound into a gas of its ions, of KF (821 KJ/mol) is higher than that of KCl (715 KJ/mol) and KBr (682 KJ/mol). One plausible reason for the observed MIR emission threshold differences between KF and KCl/KBr (Fig. 3) is that, due to the higher lattice energy, there is less sample material (in this case potassium atoms) ablated into the laserinduced plasma plume.

1. R. Brennetot, J. L. Lacour, E. Vors, A. Rivoallan, D. Vailhen, and S. Maurice, Appl. Spectrosc. 57, 744 (2003). 2. K. Stelmaszczyk, P. Rohwetter, G. Me´jean, J. Yu, E. Salmon, J. Kasparian, R. Ackermann, J.-P. Wolf, and L. Wo¨ste, Appl. Phys. Lett. 85, 3977 (2004). 3. D. E. Kim, K. J. Yoo, H. K. Park, K. J. Oh, and D. W. Kim, Appl. Spectrosc. 51, 22 (1997). 4. F. C. DeLucia, Jr., R. S. Harmon, K. L. McNesby, R. J. Winkel, Jr., and A. W. Miziolek, Appl. Opt. 42, 6148 (2003). 5. A. C. Samuels, F. C. DeLucia, Jr., K. L. McNesby, and A. W. Miziolek, Appl. Opt. 42, 6205 (2003). 6. R. S. Harmon, F. C. DeLucia, C. E. McManus, N. J. McMillan, R. Coveney, T. F. Jenkins, M. E. Walsh, and A. Miziolek, Appl. Geochem. 21, 730 (2006). 7. C. S.-C. Yang, E. E. Brown, U. H. Hommerich, S. B. Trivedi, A. C. Samuels, and A. P. Snyder, Appl. Spectrosc. 61, 321 (2007). 8. R. Obertacke, H. Wintrich, F. Wintrich, and A. Leipertz, Combust. Sci. Technol. 121, 133 (1996). 9. C. K. Y. Lam, D. C. Tilotta, K. W. Busch, and M. A. Busch, Appl. Spectrosc. 44, 318 (1990). 10. R. A. Cheville and D. Grischkowsky, Opt. Lett. 20, 1646 (1995). 11. J. B. Lurie, S. M. Miller, A. M. Blumberg, and R. A. Armstrong, Chem. Phys. Lett. 120, 481 (1985). 12. F.-Y. Yueh, J. P. Singh, and H. Zhang, in Encyclopedia of Analytical Chemistry: Applications, Theory, and Instrumentation, R. A. Meyers, Ed. (John Wiley and Sons, Chichester, 2000), p. 2066. 13. C. A. Sacchi, J. Opt. Soc. Am. B 8, 337 (1991). 14. R. F. Haglund, in Laser Ablation and Desorption-Volume 30-Experimental Methods in Physical Sciences, J. C. Miller and R. F. Haglund, Eds. (Academic Press, San Diego, 1998). 15. W. C. Martin and R. Zalubas, J. Phys. Ref. Data 10, 153 (1981). 16. Y. Ralchenko, F.-C. Jou, D. E. Kelleher, A. E. Kramida, A. Musgrove, J. Reader, W. L. Wiese, and K. Olson, in NIST Atomic Spectra Database (version 3.1.1) (National Institute of Standards and Technology, Gaithersburg, MD, 2007) http://physics.nist.gov/asd3.

Evaluation and Comparison of Two Combinations of Pneumatic Nebulizers and Spray Chambers for Direct Slurry Aspiration and Multielement Analysis of Infant Milk Powders by Axial-Viewing Inductively Coupled Plasma Atomic Emission Spectrometry

CONCLUSION

G. A. ZACHARIADIS* and L. I. VALIANOU

In summary, the first observation of LIBS MIR emission from solid ionic compounds has been reported. Intense infrared emissions between 2 to 5.6 lm due to atomic transitions between high lying Rydberg states of laser-induced breakdown products from various alkali metal halide tablets were readily detected. All the MIR emitting species in the LIBS experimental setup have been determined to be the neutral alkali metal atoms. The emission thresholds of these LIBS MIR lines were found to be correlated with the lattice energies of the alkali metal halides and the ionization energies of alkali metal atoms. The observation of these MIR atomic lines together with the vibrational and rotational emission features7 from

Laboratory of Analytical Chemistry, Department of Chemistry, Aristotle University, 54124 Thessaloniki, Greece

716

Volume 62, Number 6, 2008

Index Headings: Infant milk powder; Metals; Inductively coupled plasma; ICP; Atomic emission spectroscopy; AES.

Received 11 December 2007; accepted 29 February 2008. * Author to whom correspondence should be sent. E-mail: zacharia@chem. auth.gr.

laser-breakdown solid compounds demonstrates the promise of MIR LIBS to be a complement to UV-VIS LIBS as a potential chemical analysis probe.

FIG. 3. Normalized 2705 nm emission intensities from KCl, KBr, and KF as a function of laser energy. Three independent measurements for each sample were performed to determine the error bars.

lithium 52G7/2,9/2 state at 39079 cm1. Emission from the halide atoms and ions was not observed. The MIR characteristic emission of the alkali metal atoms was not influenced by the nature of the halide anion in the ionic crystal structure. To further explore the LIBS MIR emission mechanism, we studied the emission intensities from KCl, KBr, and KF as a function of laser energy, as shown in Fig. 3. The most prominent emission line from K, 52P3/2 at 2705 nm (3697 cm1), was chosen for the purpose of relatively high accuracy. When the laser energy dropped below 5 mJ (Fig. 3), the alkali line intensities from these three samples were not observed even though an intense white plasma spark was still visible. The emission threshold of the K 42P3/2 line from KF appears to be the highest among the three potassium halides, while that from the KBr sample appears to be the lowest in intensity. In the analysis of solid samples with LIBS, the laser plasma heats up the target materials to initiate the vaporization and atomization. The amount of the vaporized target material may depend on the thermal properties of the target compounds.9 The lattice energy, the amount of energy required to separate a mole of solid compound into a gas of its ions, of KF (821 KJ/mol) is higher than that of KCl (715 KJ/mol) and KBr (682 KJ/mol). One plausible reason for the observed MIR emission threshold differences between KF and KCl/KBr (Fig. 3) is that, due to the higher lattice energy, there is less sample material (in this case potassium atoms) ablated into the laserinduced plasma plume.

1. R. Brennetot, J. L. Lacour, E. Vors, A. Rivoallan, D. Vailhen, and S. Maurice, Appl. Spectrosc. 57, 744 (2003). 2. K. Stelmaszczyk, P. Rohwetter, G. Me´jean, J. Yu, E. Salmon, J. Kasparian, R. Ackermann, J.-P. Wolf, and L. Wo¨ste, Appl. Phys. Lett. 85, 3977 (2004). 3. D. E. Kim, K. J. Yoo, H. K. Park, K. J. Oh, and D. W. Kim, Appl. Spectrosc. 51, 22 (1997). 4. F. C. DeLucia, Jr., R. S. Harmon, K. L. McNesby, R. J. Winkel, Jr., and A. W. Miziolek, Appl. Opt. 42, 6148 (2003). 5. A. C. Samuels, F. C. DeLucia, Jr., K. L. McNesby, and A. W. Miziolek, Appl. Opt. 42, 6205 (2003). 6. R. S. Harmon, F. C. DeLucia, C. E. McManus, N. J. McMillan, R. Coveney, T. F. Jenkins, M. E. Walsh, and A. Miziolek, Appl. Geochem. 21, 730 (2006). 7. C. S.-C. Yang, E. E. Brown, U. H. Hommerich, S. B. Trivedi, A. C. Samuels, and A. P. Snyder, Appl. Spectrosc. 61, 321 (2007). 8. R. Obertacke, H. Wintrich, F. Wintrich, and A. Leipertz, Combust. Sci. Technol. 121, 133 (1996). 9. C. K. Y. Lam, D. C. Tilotta, K. W. Busch, and M. A. Busch, Appl. Spectrosc. 44, 318 (1990). 10. R. A. Cheville and D. Grischkowsky, Opt. Lett. 20, 1646 (1995). 11. J. B. Lurie, S. M. Miller, A. M. Blumberg, and R. A. Armstrong, Chem. Phys. Lett. 120, 481 (1985). 12. F.-Y. Yueh, J. P. Singh, and H. Zhang, in Encyclopedia of Analytical Chemistry: Applications, Theory, and Instrumentation, R. A. Meyers, Ed. (John Wiley and Sons, Chichester, 2000), p. 2066. 13. C. A. Sacchi, J. Opt. Soc. Am. B 8, 337 (1991). 14. R. F. Haglund, in Laser Ablation and Desorption-Volume 30-Experimental Methods in Physical Sciences, J. C. Miller and R. F. Haglund, Eds. (Academic Press, San Diego, 1998). 15. W. C. Martin and R. Zalubas, J. Phys. Ref. Data 10, 153 (1981). 16. Y. Ralchenko, F.-C. Jou, D. E. Kelleher, A. E. Kramida, A. Musgrove, J. Reader, W. L. Wiese, and K. Olson, in NIST Atomic Spectra Database (version 3.1.1) (National Institute of Standards and Technology, Gaithersburg, MD, 2007) http://physics.nist.gov/asd3.

Evaluation and Comparison of Two Combinations of Pneumatic Nebulizers and Spray Chambers for Direct Slurry Aspiration and Multielement Analysis of Infant Milk Powders by Axial-Viewing Inductively Coupled Plasma Atomic Emission Spectrometry

CONCLUSION

G. A. ZACHARIADIS* and L. I. VALIANOU

In summary, the first observation of LIBS MIR emission from solid ionic compounds has been reported. Intense infrared emissions between 2 to 5.6 lm due to atomic transitions between high lying Rydberg states of laser-induced breakdown products from various alkali metal halide tablets were readily detected. All the MIR emitting species in the LIBS experimental setup have been determined to be the neutral alkali metal atoms. The emission thresholds of these LIBS MIR lines were found to be correlated with the lattice energies of the alkali metal halides and the ionization energies of alkali metal atoms. The observation of these MIR atomic lines together with the vibrational and rotational emission features7 from

Laboratory of Analytical Chemistry, Department of Chemistry, Aristotle University, 54124 Thessaloniki, Greece

716

Volume 62, Number 6, 2008

Index Headings: Infant milk powder; Metals; Inductively coupled plasma; ICP; Atomic emission spectroscopy; AES.

Received 11 December 2007; accepted 29 February 2008. * Author to whom correspondence should be sent. E-mail: zacharia@chem. auth.gr.

INTRODUCTION Milk and its products, used in the human dairy diet, contain, among other important ingredients, trace elements absolutely essential for nutrition.1 The monitoring of milk in the European Union is regulated by Directive 96/23/EC; Directive 96/4/EC is also implemented as separate legislation referring only to infant formulas. According to the latter, the tolerance of toxic elements in infant formulas should be the minimum.2 Inductively coupled plasma coupled to atomic emission (ICP-AES)3,4 or mass spectrometric detectors (ICP-MS)5–8 is a powerful technique for rapid multielemental determinations of macro- and trace elements, and it is widely employed in the analysis of various types of food samples, including milk and milk products.7–9 Most proposed methods involve a preliminary time-consuming pretreatment of the sample9 such as the official HNO3– HClO4 wet-acid digestion method and/or dry ashing10 to eliminate the organic matrix or, alternatively, the mild solubilization method by using tetramethylammonium hydroxide.4,11 Although sample digestion is widely used for milk analysis and offers important advantages for the determination of the total concentration of several elements in infant formulas, the current trend is towards simpler and faster methods for routine analysis of samples.12–15 An alternative method is based on direct introduction of dilute milk solutions or slurry suspensions of milk powders into the plasma, avoiding the digestion step, and such a method has already been reported for the determination of nutritional elements.3 However, there is a lack of extensive information in the literature related to the direct introduction of milk powder samples in to an ICP-AES system. It is well known that a more careful selection of the nebulization configuration may improve the analytical performance of the direct introduction method.16 Various problems such as pulsated sample delivery and tube clogging may arise during the operation of the sample introduction system that can potentially affect the condition of the plasma atomizers and deteriorate the overall precision.3,8,11 The aim of this work was to develop and evaluate a fast and simple method for direct introduction of powdered milk samples for the simultaneous analysis of almost 20 elements without a digestion step or addition of surfactant reagents. In such a case, the sample introduction system is a critical parameter; thus, we report here on the comparison of two different nebulization systems for aspiration and transfer of sample suspension into the plasma. After thorough optimization, the maximum tolerant concentration of milk powder in the suspensions while keeping high sensitivity and avoiding the extinguishing of the plasma was estimated. For comparative purposes a conventional wet-acid digestion method was applied by using concentrated nitric acid and closed-vessel decomposition.

EXPERIMENTAL Instrumentation. An axially-viewed Optima 3100 XL (Perkin-Elmer, Waltham, MA) inductively coupled plasma atomic emission spectrometer equipped with a solid-state segmented array charge-coupled device (SCD) as detector was used. The operating conditions are described in detail in Table I. The analytical wavelengths were initially set at the two most important spectral atomic (I) or ionic (II) lines17 of each

TABLE I. Operating conditions and instrumentation of the ICP-AES. Parameter

Value/condition

RF generator RF incident power Torch, injector, id Nebulizer argon flow rate Auxiliary argon flow rate Plasma gas flow rate Sample uptake flow rate Spray chamber Pneumatic nebulizer Polychromator/resolution Rinse time/read delay Signal processing/integration Detector

40 MHz, free-running 1500 W Fassel type, Alumina, 2.0 mm 0.8 L min1 1.5 L min1 18 L min1 3 mL min1 1. Scott double-pass (Ryton); 2. Cyclonic (Glass) 1. Gem-tip cross-flow; 2. Babington type Echelle/0.006 nm at 200 nm 30 s/50 s Area/8–12 s Segmented-array charge-coupled device (SCD)

analyte, and eventually the most sensitive line was employed for further research. Argon gas of 99.995% purity was employed as both the plasma and carrier gas. We have examined high RF power levels (up to 1500 W) and optimized this parameter, although in the literature lower levels, e.g., 1200 W8,10 or 1100 W,9,12 are reported without further optimization. Nevertheless, high power combined with a low argon gas flow rate results in socalled robust conditions in the plasma.18 Two different sample introduction systems were used and compared for the aspiration of the undigested samples into the plasma. The first one was a combination of a cross-flow nebulizer with a double-pass spay chamber (Scott-type, Ryton). The second one was a combination of a V-groove type nebulizer with a cyclonic spray chamber. The V-groove nebulizer can be considered to be a modified Babington type nebulizer with improved resistance to blockage and is very useful for solutions containing high levels of solids. On the other hand, cross-flow nebulizers appear to have better aerosol particle size distribution, and they are more frequently found in regular ICP equipment. Additionally, it is well known that cyclonic spray chambers exhibit higher transport efficiency than double-pass designs, allowing the passage of larger droplets.19 Reagents and Solutions. All solutions were prepared using analytical grade reagents acquired from Merck (Darmstadt, Germany) and ultra pure water from Milli-Q (18.2 MX, Millipore, Bedford, MA). Nitric acid (HNO3) was used for both performing digestion and avoiding hydrolysis of the analytes. Multielement aqueous standard solutions were prepared by appropriate stepwise dilutions of stock ICP multielement standard IV CertiPURt containing 1000 mg L1 for each element in 1 mol L1. Working calibration standards were prepared daily from stock solutions containing 1% v/v HNO3. The slope of the calibration curves was used to estimate the sensitivity of the method. Reference Materials and Samples. A non-fat milk powder reference material, SRM-1549 (National Institute of Standards and Technology, Gaithersburg, MD), was analyzed in order to evaluate the accuracy of the ICP-AES method. Five commercially available brands of dry powder infant formulas were purchased from the local market. According to information from the local suppliers, these products are very commonly used and cover more than 80% of the total consumption.

APPLIED SPECTROSCOPY

717

TABLE II. Sensitivity data from the standard addition curves in the digested milk matrix and milk suspensions using both introduction systems. Digesteda

Slurry suspension

A

A

B

Analyte

Spectral line (nm)

Slope

St. error

Slope

St. error

Slope

St. error

Ag Al As B Ba Bi Ca Cd Co Cr Cu Fe Ga In Mg Mn Ni Pb Se Zn

328.068 308.215 193.696 249.772 233.527 223.061 317.933 214.440 228.616 283.563 324.752 238.204 294.364 325.609 280.271 257.610 232.003 217.000 196.026 213.857

45.6 16.3 0.60 83.8 16.9 2.10 70.5 29.7 12.4 61.5 173 48.2 10.8 7.39 1279 237 3.48 0.43 0.65 37.1

5.9 0.1 0.01 1.6 0.1 0.02 1.0 0.3 0.1 0.7 1 0.5 0.1 0.09 7 1 0.03 0.00 0.01 0.1

47.0 14.3 0.54 57.7 18.3 1.84 177 24.7 10.6 57.8 124 20.6 9.27 6.03 456 206 3.33 0.61 0.63 21.7

0.9 0.1 0.005 0.1 0.1 0.01 31 0.1 0.0 0.2 1 0.4 0.04 0.07 10 0 0.02 0.00 0.005 0.4

56.7 21.4 0.76 78.1 30.8 2.88 88.6 33.1 15.1 85.1 221 29.9 15.1 8.77 1200 315 5.63 1.16 0.28 45.1

6.4 0.4 0.05 1.9 0.1 0.08 53.5 0.4 0.1 1.0 5 2 0.3 0.36 43 3 0.04 0.04 0.01 2.0

a

(A) Gem-tip cross-flow nebulizer and Scott double-pass spray chamber; (B) Babington type nebulizer and cyclonic spray chamber.

Direct Sample Introduction. According to this procedure, the samples were introduced directly into the ICP-AES after dilution and suspension in HNO3 with a final concentration of 1% v/v. The presence of low concentrations of nitric acid may lead to partial decomposition or leaching from the suspended solid phase. Thus, the determination refers to the total fraction of the analytes in both phases. Various concentrations of milk powder, i.e., from 1% up to 40% m/v, were homogenized by magnetic stirring at 600 rpm to form stable aqueous slurry suspensions and then aspirated immediately into the plasma. Sample Digestion. Wet-acid digestion still remains the most popular and reliable method for the determination of total concentrations of each analyte and was employed for comparative purposes. Amounts of 0.2000 6 0.0004 g were accurately weighed into dry and clean poly-tetrafluoro-ethylene (PTFE) vessels and digested with HNO3 (65% m/m) at 130 8C for 2 h. The final solutions contained about 1% m/v milk powder. Using this procedure, a study for possible matrix effects was carried out by spiking a series of the non-digested samples with appropriate standard solutions.

RESULTS AND DISCUSSION Optimization of Operating Conditions. Three operational parameters were studied first, in order to obtain maximum signal intensity and subsequently maximum sensitivity. These were the incident power of the RF generator, the flow rate of sample uptake for both methods, and the suspension’s concentration of milk powder for the direct introduction method. The optimization of the above-mentioned parameters and the selection of the optimum analytical wavelength for each analyte were made in terms of background-corrected sensitivity and no internal standard was employed. All other

718

Volume 62, Number 6, 2008

parameters were maintained stable and were selected upon previous experience.17 The optimum values for all studied parameters are included in Table I while the selected analytical wavelengths appear in Table II. Effect of Plasma Incident Power. In order to improve the atomization performance of the plasma, incident power was investigated at three levels, i.e., 1300, 1400, and 1500 W. Other parameters, such as sample uptake flow rate (3 mL min1) and concentration of milk powder (1% m/v and 20% m/v for digested and not-digested samples, respectively) were kept constant. The results proved that using either of the two sample introduction configurations the highest sensitivity was obtained at 1500 W, which was applied to the rest of the work. This observation holds for almost all analytes, although in previous works the use of lower incident power is reported.8–10 In the presence of increased amounts of organic constituents, higher incident power entails higher plasma excitation efficiency, and subsequently a better atomization of the analytes is expected. Effect of the Sample Uptake Flow Rate. A critical parameter causing increased emission intensities and strongly affecting the overall sensitivity is the sample uptake flow rate. It should always be under consideration that a greater than adequate flow rate may provoke extinction of the plasma and reduction of the overall precision. This is especially true when the available instrumental setup, i.e., rf generator and torch, has a low tolerance for sample load. Consequently, the effect of increasing sample flow rates, e.g., 1, 2, and 3 mL min1, respectively, was examined for both methods. It was proved that at a flow rate of 3 mL min1, higher signals were observed in both the primary and secondary spectral lines for the majority of the analytes. For some of them, higher signals were obtained only at the primary spectral line. Thus, 3 mL min1 was eventually chosen as the optimum sample uptake flow rate. Effect of Milk Powder Concentration in Slurry. The last parameter to be optimized in direct sample introduction was the matrix concentration. The concentrations of milk powder in milk studied were 1%, 5%, 10%, 15%, 20%, and 40% m/v, respectively. The optimum concentration was found to be 20% m/v without extinguishing the plasma and also maintaining a very good precision. In addition, clogging of the injector by carbon deposits can be avoided by frequent injection of dilute nitric acid solution (1% v/v). Obviously, lower powder concentrations, e.g., 1–10% m/v, could also be used for analysis of major analytes with higher expected concentrations. Comparison of Nebulization Systems. An important factor for obtaining accurate and precise results in ICP-AES is the sample introduction system.9,16,17 As the V-groove nebulizer in combination with a cyclonic spray chamber allows the passage of larger droplets, there was a difference concerning the sensitivity and precision. There is a small but observable increase in sensitivity because of the larger amount of analytes reaching the plasma and being excited by using the second nebulization configuration as compared to the first one (Table II). However, taking into account the standard error of the slopes, it is observed that the precision is worse when using the second configuration. For example, the standard error is up to ten times higher than with the first nebulization configuration for Ag, As, B, Bi, Ga, etc. This is probably due to the destabilization of the plasma, resulting from the droplet size distribution and the higher amount of organic matter coming into it.

TABLE III. Limits of detection (mg kg1) and precision (RSD, %) for the acid digestion method and the direct introduction of slurry suspension method. Digesteda

Slurry suspensiona

A

A

TABLE IV. Analysis of milk standard reference material NIST SRM 1549 (mg kg1).

Analyte

Spectral line (nm)

Ala Ca Cu Fe Mg Mn Zn

308.215 317.933 324.752 238.204 280.271 257.610 213.857

B

LOD RSDb LOD RSDb RSDc LOD RSDb RSDc % lg kg1 % % lg kg1 % % Analyte lg kg1 Ag Al As B Ba Bi Ca Cd Co Cr Cu Fe Ga In Mg Mn Ni Pb Se Zn

8.1 21 43 0.7 3.1 23 0.4 2.8 3.5 3.0 0.6 4.0 6.5 12 1.4 0.2 25 72 56 1.7

0.8 0.4 2.0 0.4 1.1 5.3 0.7 1.4 1.0 0.3 1.0 0.4 2.3 1.1 0.7 0.7 1.3 2.5 12 0.8

2.6 19 38 8.5 2.6 36 0.1 3.1 3.7 5.9 3.4 4.8 5.6 30 0.7 0.6 9.5 73 76 2.6

2.4 0.8 1.0 1.8 1.6 2.3 0.8 1.9 1.0 1.2 2.0 1.5 1.5 1.6 0.7 1.7 1.1 2.8 2.0 1.6

3.8 3.5 5.6 2.4 1.8 4.0 3.9 3.0 1.1 4.7 2.6 4.3 3.3 5.5 2.4 4.6 3.8 2.8 4.6 1.6

17 43 n.c.d 9.6 23 n.c. 1.1 19 45 10 4.4 2.4 59 111 0.7 2.5 n.c. n.c. n.c. 14.2

4.4 3.9 n.c.d 3.8 2.2 n.c. 4.1 1.9 1.9 5.1 10 7.0 2.7 6.4 2.9 8.3 n.c. n.c. n.c. 6.9

7.1 4.3 n.c.d 3.9 4.6 n.c.d 4.0 5.8 2.2 6.7 9.3 8.9 2.7 7.8 3.4 6.9 n.c.d n.c.d n.c.d 6.1

a

(A) Gem-tip cross-flow nebulizer and Scott double-pass spray chamber; (B) Babington type nebulizer and cyclonic spray chamber. b Intra-day precision at 250 lg kg1 (n ¼ 10). c Between-day precision (n ¼ 5). d (n.c.) Not calculated.

Analytical Performance of the Method. Under the optimum conditions and for each particular analyte, (1) calibration curves were prepared using aqueous multielement standards solutions and (2) standard addition curves were prepared using spiked powdered milk suspensions. The slopes with the standard errors for the standard addition curves are given in Table II for the most sensitive spectral line of each analyte. For the majority of the analytes, a decrease in the slope of the standard addition curve as compared to that of the aqueous multielement standards was observed. This sensitivity loss is mainly caused by the presence of substances containing high levels of carbon in the plasma region, which may absorb part of the induced plasma energy. Thus, the use of the standard addition curve is recommended in order to compensate for this sensitivity variation. Comparing the digested sample introduction with the direct slurry suspension introduction method (Table II), there are few unimportant fluctuations in sensitivity for almost all analytes. Consequently, the direct introduction of powdered milk sample can be considered as a good and fast alternative to the conventional acid-digestion method. The limit of detection (LOD) of each analyte (lg kg1) was assessed as the concentration equivalent to three times the standard deviation of ten blank measurements divided by the slope of the calibration line. Accordingly, the limit of quantification (LOQ) could be estimated using ten times the standard deviation. From the results of the calculated LODs listed in Table III, it is proved that the detection limits are lower with the nebulization system (A) for most of the analytes. The intra-day precision was calculated at a concentration level of

a

Certified value 13000 0.7 1.78 1200 0.26 46.1

2 6 6 6 6 6 6

500 0.1 0.10 30 0.06 2.2

Determined value

Recovery (%)

6 6 6 6 6 6 6

114 93 101 95 98 98 103

2.377 11650 0.71 1.588 1147 0 .243 47.9

0.092 550 0.02 0.046 42 0.014 8

Values of the SRM are not certified and are given only as information.

250 lg kg1 (n ¼ 10) for digested and non-digested samples, respectively, and is expressed by means of the relative standard deviation (RSD) in Table III together with the between-day (n ¼ 5) precision of the slurry technique. The RSD for intra-day precision ranged from 0.7–2.8% for almost all analytes with the first nebulization system (A) and between 1.9–10% with the second system. Comparing the two sample introduction configurations in terms of intra-day and between-days precision, it is observed that the performance is better with the first configuration. At very low concentrations, however, the precision of several analytes is worse when using the second nebulization combination, and in these cases the detection limits were not calculated. The correlation coefficient for the calibration curves ranged between 0.9988–0.9999 and 0.9750–0.9999 for the (A) and (B) nebulization systems, respectively. Taking into consideration the above observations, the use of the first nebulization configuration was preferred. The accuracy of the method was evaluated by analyzing NIST SRM 1549, which is a standard reference material similar to commercially available infant formulas. Nevertheless, it has to be mentioned that it is skimmed powdered milk containing low lipid concentration, thus causing lower matrix effects than normal-fat samples. The results of the SRM analysis are presented in Table IV, and in general there is a good agreement with the NIST certified values. The calculated recoveries for the measured analytes were between 95–114%. Finally, in order to verify the accuracy of the slurry suspension introduction method of regular powdered infant milk products, recovery tests were performed in spiked samples. The recovery in this case varied between 90–109% for almost all analytes. Application to Commercial Milk Powders. The concentrations of toxic and nutrient elements in infant formulas are regulated by Directive 96/4/EC of the European Union.2 In this context, samples of five commercial infant formulas were analyzed by using the proposed slurry suspension method TABLE V. Analysis of commercial infant formulas by the slurry suspension method (mg kg1).

Analyte

Spectral line (nm)

IF1a

IF2

IF3

IF4

IF5

Ca Cu Fe Mg Mn Zn

317.933 324.752 238.204 280.271 257.610 213.857

3120 2.66 48.7 421 0.64 36.4

4285 3.45 40.6 471 0.53 38.9

3805 2.36 47.0 471 0.63 47.0

4643 3.00 54.5 334 0.35 41.7

3388 3.12 62.5 347 0.43 38.2

a

(IF) Infant formula, commercial products of various brands.

APPLIED SPECTROSCOPY

719

using the first nebulization system. In parallel, the acid digestion was applied for comparison. The results obtained from the analysis of non-digested samples are presented in Table V. The results show that no detectable concentrations of toxic metals, e.g., Cd, Pb, etc., were present in the examined samples. Moreover the concentration of all nutrient elements falls into the minimum required levels for infant diet. It is worth mentioning that generally there is sufficient agreement between the results and the list of contents given on the label by the producers.

CONCLUSION The multielement analysis of milk samples and infant milk formulas can be performed by ICP-AES after digestion of the sample, which is a time-consuming procedure. The direct introduction of slurry suspension of the sample in the plasma atomizer of the ICP-AES proved to be an efficient alternative providing a fast and robust methodology. Determination of several major and minor analytes can be achieved with limited sample handling, but for trace analytes other more sensitive techniques, e.g., ICP-MS, should be used. The proposed slurry suspension method proved to be comparable with the digestion method in terms of precision and sensitivity. Moreover, two nebulization systems were tested considering both the sensitivity and precision, and between them, the combination of a cross-flow nebulizer with a double-pass spray chamber performed better in terms of overall precision. The developed method could be applied as a fast screening method for milk powders and especially for infant formulas.

720

Volume 62, Number 6, 2008

1. M. Neville, P. Zhang, and J. Allen, ‘‘Minerals, Ions and trace elements in milk’’, in Handbook of Milk Composition, R. G. Jensen, Ed. (Academic Press Inc., London, 1995), Chap. 7, p. 623. 2. European Directives 96/4/EC and 96/23/EC, http://ec.europa.eu/index_en. htm. 3. P. J. McKinstry, H. E. Indyk, and N. D. Kim, Food Chem. 65, 245 (1999). 4. J. A. Nobrega, Y. Gelinas, A. Krushevska, and R. M. Barnes, J. Anal. At. Spectrom. 12, 1243 (1997). 5. R. W. Fonseca and N. J. Miller-Ihli, Appl. Spectrosc. 49, 1403 (1995). 6. R. R. de la Flor St. Remy, M. L. Ferna´ndez Sa´nchez, J. B. Lo´pez Sastre, and A. Sanz-Medel, J. Anal. At. Spectrom. 19, 616 (2004). 7. T. Prohaska, G. Ko¨llensperger, M. Krachler, K. De Winne, G. Stingeder, and L. Moens, J. Anal. At. Spectrom. 15, 335 (2000). 8. P. Cava-Montesinos, M. L. Cervera, A. Pastor, and M. Guardia, Anal. Chim. Acta 531, 111 (2005). 9. C. Sola-Larranaga and I. Navarro-Blasco, Anal. Chim. Acta 555, 354 (2006). 10. AOAC Official Methods of Analysis, 16th ed., Supplement Chapter 50, Official Method 984.27 (1996). 11. H. Matusiewicz and B. Golik, Microchem. J. 76, 23 (2004). 12. M. A. Murcia, A. Vera, M. Martinez-Tome, A. Munoz, M. HernandezCordoba, and R. Ortiz-Gonzalez, Lebensm-Wiss.u-Technol. 32, 175 (1999). 13. P. C. Aleixo and J. A. Nobrega, Food Chem. 83, 457 (2003). 14. D. M. Santos, M. M. Pedroso, L. M. Costa, A. R A. Nogueira, and J. A. Nobrega, Talanta 65, 505 (2005). 15. A. Ikem, A. Nwankwoala, S. Odueyungbo, K. Nyavor, and N. Egiebor, Food Chem. 77, 439 (2002). 16. R. Lobinski, W. van Borm, J. A. C. Broekaert, P. Tschopel, and G. Tolg, Fresenius’ J. Anal. Chem. 342, 563 (1992). 17. G. A. Zachariadis and D. C. Kapsimali, J. Pharm. Biomed. Anal. 41, 1212 (2006). 18. I. Novotny, J. C. Farinas, W. Jia-liang, E. Poussel, and J.-M. Mermet, Spectrochim. Acta, Part B 51, 1517 (1996). 19. B. L. Sharp, J. Anal At. Spectrom. 3, 939 (1988).