Giovanni Gallavotti ( E d.)
Statistical Mechanics Lectures given at a Summer School of the Centro Internazionale Matematico Estivo (C.I.M.E.), held in Bressanone (Bolzano), Italy, June 21-27, 1976
C.I.M.E. Foundation c/o Dipartimento di Matematica “U. Dini” Viale Morgagni n. 67/a 50134 Firenze Italy
[email protected]
ISBN 978-3-642-11107-5 e-ISBN: 978-3-642-11108-2 DOI:10.1007/978-3-642-11108-2 Springer Heidelberg Dordrecht London New York
©Springer-Verlag Berlin Heidelberg 2010 Reprint of the 1st ed. C.I.M.E., Ed. Liguori, Napoli 1976 With kind permission of C.I.M.E.
Printed on acid-free paper
Springer.com
C m T R O INTERNAZIONALE MATEMATICO ESTIVO
(c.I.M.E.)
I Ciclo
- Bressanone d a l
21 giugno a1 2 4 giugno
1976
STATISTICAL MECHANCIS
Coordinatore: Prof. Giovanni Gallavotti
P. Cartier:
Theorie de la mesure. Introduction B la mecanique statistique classique (Testo non pervenuto)
C. Cercignani:
A sketch of the theory of the Boltzmann equation.-
O.E. Lanford:
Qualitative and statistical theory of dissipative systems.-
E.H. Lieb:
many particle Coulomb systems.-
B. Tirozzi:
Report on renormalizationgroup.-
A. Wehrl:
Basic properties of entropy in quantum mechanics.
2 ENTRO INTERN AZIONALE MATEMATICO ESTIVO
(c.I.M.E.)
A SKETCH OF THE THEORY OF THE BOLTZMANN EQUATION
C . CERCIGNANI
Istituto d i M a t e m a t i c a ,
Politecnico d i Milano
C o r s o tenuto a B r e s s a n o n e d a l 21 giugno a1 24 giugno 1976
A Sketch of t h e Theory of t h e
Bolt zmann equation C a r l o Cercignani I s t i t u t o d i Matemeticz P o l i t e c n i c o d i Tfiilano Milano, I t a l y
I n t h i s seminar, I s h a l l b r i e f l y review t h e t h e o p of t h e Boltzmann equation. How t h e l a t t e r a r i s e s from t h e 1 , i o u v i l l e equation has been d i s c u s s e d i n 0. Lanf ord 's l e c t u r e s . We s h a l l w r i t e t h e Boltzmann e q u z t i o n i n t h i n form
where t , 2 ,
5
r i a b l e s , while
4
denote t h e time, space and v e l o c j t y vni s t h e d i s t r i b u t i o n f u n c t i o n , normalized
i n such a way t h a t
where
M
i s t h e mass contained i n t h e r e ~ i o nover which t h e
i n t e a r a t i o n with r e s p e c t t o
Q({,{) i s t h e
2 extends.
so c a l l e d c o l l i s i o n term, e x p l i c i t l y obtai-
n a b l e from t h e f o l l o w i n g d e f i n i t i o n
where
$
i s an a u s i l i a r y v e l o c i t y v e c t o r , V is t h e re-
l a t i v e speed, i.e..
t h e mamitude of t h e v e c t o r
#'=Q&~),$=
-f
to
and
etc.,
where
'
and
-5;
y = f -I*, are releted
f , through t h e r e l a t i o n s e x p r e s s i n g conservation
of momentum and energy i n a c o l l i s i o n
where
2 in
i s a u n i t v e c t o r , whose p o l a r a n g l e s a r e
2 R
p o l a r c o o r d i n a t e system with
Y as p o l a r a x i s .
T n t ~ r r ~ t i oextends n t o ~ l vla l u e s of and
rr/2 with r e r o e c t t o 6 , from
t o 6 . Finally
B ( ~ v )i s
4 ma
and between
6
to
0
21 w i t h respect
related t o the d i f f e r e n t i a l cross
section
q(qvby
and
i s t h e mass of a g a s molecule. For f u r t h e r d e t a i l s
m
the relation
one should c o n s u l t one of my books [I ,2]. Eq. (1 ) i s v a l i d f o r monatomic molecules
and i s more Ke-
n e r n l t h a n t h e Boltzrnann e q u a t i o n considered by Lanford i n
-
h i s l e c t u r e s , because i t i s n o t but a l l o w s molecules
r e s t r i c t e d t o r i g i d spheres, with any d i f f e r e n t i a l c r o s s s e c t i o n .
The c a s e of r i g i d spheres i s obtained by s p e c i a l i z i n g as f o l l o w s
B(~v)
where
d
i s t h e sphere diameter. Another importa?t cF.se i s
o f f e r e d by t h e so c a l l e d 1~;axwellmolecules. The l a t t e r a r e c l z s s i c a l point masses i n t e r a c t i n r w i t h s c e n t r a l f o r c e i n v e r s e l y p r o p o r t i o n a l t o t h e f i f t h Dower of t h e i r mutual d i s t e n c e ; a s a consequence, it t u r n s out t h a t
B(6,V)
i s independent of
V.
It i s c l e a r t h a t i n i t i a l and boundary c o n d i t i o n s a r e required i n o r d e r t o s o l v e t h e Boltzmann e q u a t i o n , s i n c e t h e l a t t r r contain3 t h e time and space d e r i v a t i v e s of
f.
The bound:-
4 C O~? I -
d i t i o n s a r e p a r t i c u l a r l y important s i n c e t h e y d e s c r i b e t h e int e r a c t i o n of t h e g a s molecules with s o l i d w a l l s , but p a r t ) cul a r d i f f i c u l t t o e s t a b l i s h ; t h e d i f f i c u l t i e s a r e due, mainly, t o o u r l a c k of b o w l e d g e of t h e s t r u c t u r e of t h e s u r f a c e l c y ~ r - . of s o l i d bodies and hence of t h e i n t e r a c t i o n p o t e n t i a l of tlir g a s molecules with molecules of t h e s o l i d . \'!hen a molecule i::p i n c e s upon s s u r f a c e , it i s adsorbed and may form chemi cn.1 bonds, d i s s o c i a t e , become i o n i z e d o r d i s p l a c e s u r f a c e atoms. The simplest p o s s i b l e model of t h e pas-surface
interaction
i s t o assume t h a t t h e molecules a r e s p e c u l a r l y r e f l e c t e d a t
t h e s o l i d boundary. T h i s assumption i s extremely u n r e n l i s t l c i n e e n e r a l and can be used only i n p a r t i c u l a r c a w s . Tn peneral, a molecule s t r i k i n g a s u r f a c e a t a v e l o c i t y from it a t a v e l o c i t y
-
9 /
reflects
which i s s t r i c t l y determined only
i f t h e path of t h e molecule within a w b l l can be computed exac-
t l y . T h i s computation i s impossible because i t depends upon a g r e a t number of d e t a i l s , such as t h e l o c a t i o n s and v e l o c i t i e s of a l l t h e molecules of t h e wall. Hence vze m&y only hope t o compute t h e p r o b a b i l i t y d e n s i t y
R
&' - 4 5 )
thrt
2
no1 cc:r? e
e' f. +df .
s t r i k i n g t h e surface with v e l o c i t y between emerces with v e l o c i t y between
-t
and
and If
-fi+dY -
re-
R i s hown,
@
i t i s easy t o w r i t e t h e boundary condition f o r
where g i s t h e u n i t v e c t o r nonnal t o t h e wall and we assumed must be replaced t h e wall t o be at r e s t (otherwise
2,
by
f-%,tL% , denot i n s t h e w a l l 's v e l o c i t y . ) I n ~ e n e r a l ,R w i l l be d i f f e r e n t a t d i f f e r e n t p o i n t s o f t h e
w ? 1 1 and d i f f e r e n t times; t h e dependence on 5 and t i s not shown
e x n l i c i t l y t o make t h e equations shorter. If t h e wall r e s t i t u t e s all the e a s molecules ( i . e .
i t i s non-
porous m d nonadsorbing 1, t h e t o t a 1 p r o b a b i l i t y f o r an impinginp aolecule t o be re-emitted,
with no matter what v e l o c i t y
-I
is
mity:
A n obvious property of t h e k e r n e l
Kt!?!) i s
t h a t i t cannot
assume negative values
Another b a s i c property of t h e kernel
R
,
which can be cal-
l e d t h e " r e c i p r o c i t y laww o r t h e " d e t a i l e d balance", a s follows
where
$&)
[I,
21
is
written
:
is pmportlond t o
u*p[-.!!y(2KQJ,where
To i s
{(S)
t h e temperature of t h e w e l l ( i n o t h e r wonls,
j5 E ;:F.Y-~?-
l i a n d i s t r i b u t i o n f o r a pas a t r e s t a t t h e t e n n e r a t u r e of t h o w a l l 1. We n o t e a simple consequence of r e c i p r o c i t y ; i f t l - e
4
d i s t r i b u t i o n i s t h e w a l l ilaxwellian at t h e w a l l according t o E q . ( l l ) ,
md aess
IS
Y
--rln u j
- .-
ror.t.rne6
t h e n t h e d i s t r i b u t i o n fimctior,
of t h e emerging molecules i s a g a i n
4
or, i n o t h e r words, t h e
w a l l >Iaxwellian s a t i s f i e s t h e boundary c o n d i t i o n s . I n f a c t ,
we i n t e p r a t e Eq. ( 1 3 ) with r e s p e c t t o obtain
-8'
1f
and u s e Eq. (1 1 ) we
and t h i s equation proves o u r statement, according t o Eq. (10). It i s t o be remarked t h a t Eq. (14 1, although a consenuence o f
Eq. ( 1 3 ) (when Eq. ( 1 1 ) h o l d s ) i s l e s s r e s t r i c t i v e t h m Xq.
(1')
and could be s a t i s f i e d even i f Eq. ( 1 3 ) f a i l e d . A s a consequence of t h e above p r o p e r t i e s , one can p m v e [2]
t h e f o l l o w i n g remarkable theorem: Let
C ( I ) be
arpment
2.
a s t r i c t l y convex continuous f u n c t i o n of i t s
Then f o r any s c a t t e r i n p k e r n e l
R(k'd3)
sati-
s f y i n e Eqs. (11 ), (121, ( 1 4 ) , t h e follow in^ i n e q u a l i t y h o l d s
where
g
i s t h e w a l l Idaxwellian,
3 = $/fo
and i n t e ~ r e t i o n
extends t o t h e f u l l ranges of v a l u e s of t h e components of' t h e v a l u e s of
e=
4
through Eq. (1.6).
for
Zq9Jbeing
r e l a t e d t o those f o r ) . 9 C ~
E q u a l i t y i n Eq. ( 1 5 ) h o l d s i f and only if
almost everywhere, u n l e s s
R (EL5 )
i s proportionzl
t o a d e l t a function. d s a corollary, t h e following inequality holds
- h e r e [q*%Jd
denotes t h e normal h e a t f l u x fed.
:
[2)
i n t o the g a s
by t h e s o l i d c o n s t i t u t i n g t h e w a l l and R i s t h e g a s c o n s t m t . We want t o g e n e r a l i z e t h e H-theorem,
considered
in
0. L a -
f o r d ' s l e c t u r e s , t o t h e c a s e of a g a s bounded by s o l i d w a l l s which may o r may n o t be a t r e s t . To t h i s end we d e f i n e
and observe t h a t
{A
RUZ
w i t h r e s p e c t t o i from 1 t o 3 i s understood).
Mow, t h e f o l l o w i n g i d e n t i t y h o l d s f o r any t h e i n t e g r a l s make sense:
~ , fa,
provided
This i d e n t i t y f o l l o w by s t r a i g h t f o r w a r d manipulations; f o r details, see
rl
,q
.
Applying Eq. ( 2 0 ) t o t h e c a s e obtain
P = (D8f ,$I{
(430)) we
where t h e i n e q u a l i t y f o l l o w s from t h e f a c t t h a t
(1-A) &8 ;\
is
always n e g a t i v e , except f o r )( = 1 , where i t i s zero. Hence eq u a l i t y i n Eq. ( 2 1 ) i s v a l i d i f and only i f
or letting
denote
(P+E =P*?'!* I
1
Q = 1 and, a s R y = r ; ( i = 1,2,2) and
T h i s equation i s s a t i s f i e d t r i v i a l l y by consequence of Eqs. ( 4 ) and (5),by
p ;it can be shown
Qr
[2
]
t h e r e a no o t h e r li-
that
n e a r l y independent c o l l i s i o n i n v a r i a n t s (such i s t h e neme f o r the solutions
of Eq. (23)). A s a consequence, t h e most qene-
r a l d i s t r i b u t i o n f u n c t i o n s a t i s f y i n g Eq. ( 2 2 ) i s given by
where a,
2 ,c
a r e constant. Eq. (24)
b e rewritten i n the
f o l l o w i n g form
where
9 , y,T
- -
a r e new c o n s t a n t s r e l a t e d t o t h e previous
onesand have t h e meaning of d e n s i t y , mass v e l o c i t y a d temper a t u r e a s s o c i a t e d with t h e d i s t r i b u t i o n f u n c t i o n t o well-known formulas
El ,2]
distribution. Eqs. (19) and ( 2 1 ) imply t h a t
.
Eq.
f
according
(25 ) gives a I~axwellian
where t h e e q u a l i t y s i g n a p p l i e s i f ar.d only i f l i e n , i.e.
f
i s Xaxwel
-
i s given by Eq. (25 ).
I f we i n t e g r a t e , both s i d e s of Ea.
( 2 6 ) with resnect t o 2
nver a region R b o u d e d by s o l i d walls, we have, i f t h e boun3 9 of R moves with v e l o c i t y .k,:
dary
(27)
where
dS
i s a surface element of t h e boundary
and
9R
;
t h e inward. normal. The second term i n t h e i n t e g r a l comes from the f a c t t h a t , i f t h e boundary i s moving, when forming t h e time d e r i v a t i v e of H we have t o t a k e i n t o account t h a t t h e region of in.tegration changes with time. I f we u s e Eq. (16),
when we replaced
Eq. (27) becomes:
f
by J-%. i n Eq. ( 1 6 ) as required. Eq. ( 2 6 ) ~ e n e r a l i z e st h e H-theorem, showing t h a t H decreases with
time if t h e r e i s no heat exchange walls.
~ l s o ,e q u a l i t y i n Eq. ( 2 8 ) a p p l i e s i f and only i f
lr-ellian. where
between t h e gas and t h e
1
4
is
-
Eq. ( 2 8 ) suggests t h a t
H,. be i n t e r p r e t e d a s ?/? i s t h e entropy of t h e gas, s i n c e it s a t i s f i e s t h e
i n e q u a l i t y (Clausius-Duhem inequality). This i d e n t i f i c a t i o n i s v a l i d a t e d by evaluatinq H a t equilibrium, when h ~ v et h e form indicated i n Eq. ( 2 5 ) ; i n such a case
turns out t o have t h e same dependence on t ropy i n ordinary thermodynamics.
P
and
f m~st
)LC- RH as the en-
Let u s now b r i e f l y examine t h e problem of solving t h e Boltzrnann equation; because of t h e nonlinear n a t u r e of t h e c o l l i s i o n
term
,
t h i s i s a d i f f i c u l t problem. A very p a r t i c u l a r
c l a s s of s o l u t i o n s i s o f f e r e d by ? f a x w e l l i e ~dist ri.bt:tj.ons, So.
(25 1, which d e s c r i b e s t a t e s c h a r a c t e r i z e d by t h e f a c t that r . ~ i t b e r hea.t f l u x n o r s t r e s s e s o t h e r tha? i s o t r o p i c pressurn are n r e s e n t . i f we want t o d e a c r i b e more r e a l i s t i c nonsquili'oriur:! s i t u a t i o n s , vie have t o rely upon approxi.cata methods, t y n i c e l 1 . v -perturbation t e c h n i q u e s , The sj.molest approach i s
where
4
which
mpy
ti0
virj.te
i s a Maxwellian end
g i s a "small pnraneter", o r nay n o t appear i n t h e Eoltzmzrm equ?ti.on. I n t h c w i l l appear i n t h e i n i t i a l and hour,dnxy concli
second case,
t i o n s and t h e e q u a t i o n f o r
kL
-
w i l l be
where
i s c a l l e d t h e l i n e a r i z e d Boltzmann operator. 3q. ( 3 0 ) , j.n t u r n ,
i s c a l l e d l i n e a r i z e d Boltzmann equation.
H i l b e r t space
then
L
'p
where t h e s c a l a r product i s given bv
i s a symmetric o p e r a t o r io
I n addition,
L
If one i n t r o d u c e s a
i s non-nepative
2:
pad t h e equzllt,!r
in.vsrimt. : n
i.e.
s i g h o l d s '1
4.n
a collisio~
zu?h a c s s e
t h e c o l l l s r 3 n l n v a r i m t s are eigenfunctlons assoclafed
with t h e f i v e f o l d tor
m d onl;.. i f h
t .,411 t h e s e
degenerate e i r e n v a l u e
= 0 of t h e open=-
propertjeJfollow inmediately f mrn Eq. ( ? ?
(20), i f t h e circumstance t h a t
fe
)
and
s a t i s f i e s Eq. ( 2 2 ) i s pro-
n e r l y taken i n t o account. 39
(35) su(~(re8-t. i n v e s t i g a t i n g t h e spectrum of
p r ~ b l e ma r i s e s when we look f o r t h e s o l u t i o n of Eq. t h e space homogeneo~sc a s e
(aR/aa = 0) .
Eq.
L ;t h i s (30) i n
(34 ) shows t h a t
t h e speotrun i s contained i n t h e n e r a t i v e r e a l s e n i a x i s of t h e
x-?lane;
it t u r n s out t h a t t h e spectrum i s extremely d e v n d e n t
u m n t h e form o f t h e choice of t h e f u n c t i o n
5 (0,V)
appea-
r i n ~i n Eq. ( 3 ) . It i s completely d i s c r e t e f o r t h e c a s e of Hax::~11 molecules, while i t i s p a r t l y d i s c r e t e an3 p a r t l y continuous i:~ t h e c a s e of r i g i d spheres. F o r f u r t h e r d e t a i l s , one should
[7,23
c o n s u l t Ref s.
.
An i n t e r e s t i n g ~ r o h l e nc r i s e s when one i n v e s t i q a t e e t h e so-
l u t i o n s which do n o t d e ~ e n don t i n e t and two space c o o r d i n e t s s , ; i n this c z s e one has t o s o l v e t h e equation ~ ? yx2 and x,
i n t h e unknown
k = &. (x,, b,
between t h i s e ~ a t i o nand Eq.
,k,,h)= (,(x~, - ). The similzr!ty (30) with
t h a t we look f o r s o l u t i o n s of t h e form
a & , / a= ~0
sup~ests
{*ere
satisfies
g
which i s t h e a n a l o g j e of
U=;\k.The f i r s t
question i s i-.rtho-
t h e sol-utions of Eq. (381 a r e s u f f i c i e n t t o c o n s t r u c t t h e p n e r z l s o l u t i o n of Eq.
( 3 6 ) by euperposition. Next c o m e s a study of
t h e s e t of vfilues of (different fron
h
f o r ~vhick!Eq.
( 3 8 ) has a s o l u t i o n
9 = 0).
The problem h e r e i s more d i f f i c u l t beczuse t h e r e i s w play between
L
and t h e m u l t i p l i c a t i v e o n e r a t o r
4 . .1
-
:~.-.r-r.
1-5-
t i o n t h e e x i s t e n c e of t h e c o l l i s i o n i n v a r i m t s s ~ t i s f y i n y :T r .
( 3 5 ) prevents
L
from beinp a s t r i c t l y n e ~ a t i v eoperator. In
s p i t e of t h i s , it i s p o s s i b l e t o show [2]
t h ~ tth e p e n e r a l
s o l u t i o n of Eq. (36) can be w r i t t e n a s follows:
a=O are the five a r e t h e e i g e n s o l u t i o n s of Eq. 8 c o l l i s i o n i n v a r i a n t s yo=f q<~i-3; ( ;= 1,2,3 1, \O! , ff? $R$
,
(dl
(To being t h e temperature i n t h e b a s i c I~laxwellian I
we have assumed t h a t t h e 1 5
&
1. Here
form a continuous s e t , othenvi-
s e t h e corresponding i n t e g r a l h a s t o be repleced by t h e sum It i s c l e a r t h a t t h e g e n e r a l s o l u t i o n given i n Eq. i s mede up of two p a r t s , hA and
, giver,
by
(?.a)
where t h e "eigenvalues" t h e mean f r e e path [21
a r e of t h e o r d e r of t h e inverse of
. It
is clear that
$
describes space
t r a n s i e n t s which a r e of importance i n t h e neighbourhood of boundaries and become n e g l i g i b l e a few mean f r e e paths f a r from them. The circumatanccthat Eq. (41) contains exponentials with both
A>
0
0 i s e x a c t l y what i s required t o describe
and
a decay e i t h e r f o r
x > y,
or
X,
4 yl
, where
i s the
C
x,
location of a boundary, The general s o l u t i o n given by Eq. (39) then shows t h a t , i f the region where t h e gas i s contafned ( e i t h e r a h a l f space o r a s l a b of thickness
d , because of t h e assumption t h a t
h is
independent of two space coordinates) i s much t h i c k e r than t h e mean f r e e path
l , then %
w i l l be n e g l i g i b l e except i n boun-
dary l a y e r s a few mean f r e e paths thick. These l a y e r s receive the name of "Knudsen l a y e r s " o r "Kinetic boundary layers". Outs i d e them t h e s o l u t i o n i s accurately described by t h e asymptot i c p a r t hA
, defined
by Eq. ( 4 0 ) ; i t can be shown [2]
that i f
we conpute t h e s t r e s s t e n s o r and heat f l u x v e c t o r s a r i s i n g from hA
, they
t u r n out t o be r e l a t e d t o t h e v e l o c i t y and temperature
r r a d i e n t s by t h e NavierStokes-Fourier
r e l a t i o n s ,with t h e f ol-
lowing expressions f o r t h e v i s c o s i t y c o e f f i c i e n t heat conduction coefficient
k
:
and t h e
These r e s u l t s can be extended t o more ~ e n e r ? lr r o t l e - .-
r.2
Very i n t e r e s t i n q problems z r l s e when he j n ~ o : > r l , ' ; - ~r:>>< 1s riot s a t i s f i e d , i . e .
the s l a b thiclmess
t h e mean f r e e math 1 s c o r m - r - P I P rlth
n, 2 1
; t h e i r t r e e t n e n t I.:;, l-onevrr,
1.3-
yond t h e l i m i t s of t h e present eemina,r.
REFERENCES 1
- C.
Cercignani
- **Blathematical! k t h o d s Plenum P r e s s , N.Y.
2
- C.
Cercignani
in K i n e t i c Theor;",
(1969)
- "Theory and Applicztion
of t h e fioltzmsml
Equation1*, S c o t t i s h Academic Press, E d i n burgh (1975).
CENTRO INTERNAZIONALE MATWATICO ESTIVO (c.I.M.E.)
QUALITATIVE AND S T A T I S T I C A L THEORY OF D I S S I P A T I V E SYSTEMS
O s c a r E.
LANFORD I11
D e p a r t m e n t of M a t h e m a t i c s , U n i v e r s i t y of C a l i f o r n i a B e r k e l e y , C a l i f o r n i a 94720
C o r s o tenuto a B r e s s a n o n e d a l 21 a1 2 4 g i u g n o 1 9 7 6
Qualitative
Statistical Theory
of Dissi~ativeSystems -
Oscar E. Lanford I11 Department of Mathematics University of California Berkeley, California 94720
Preparation of these notes was supported in part by NSF Grant MCS 75-05576. A01.
26 Chapter I.
Elementary Qualitative
Theory of Mfferential Eauations.
This series of l e d w e s w i l l be concerned with t h e s t a t i s t i c a l theory of dissipative system and, a t l e a s t metaphorically, with i t s applications t o hydrodynamics.
The priacipal objective will be t o t r y t o
clariFy the question of how t o construct the appropriate ensemble f o r the s t a t i s t i c a l theory of turbulence. t h i s point f o r some time.
We w i l l not, however, come t o
It should be noted a t the outset t h a t the
relevance of our discussion t o t h e theory of turbulence i s dependent on the guess t h a t , despite the fact t h a t f l u i d flow problems have i n f i n i t e dimensional s t a t e spaces, t h e important phenomena are essentially f i n i t e dimensional.*
This point of view i s ncrt universally accepted [ h ] . On
the other hand, the theory i s not restricted t o fluid flow problems; it also applies t o a large number of model systems arising, f o r example, i n mathematical biology 171. The methods we w i l l discuss a r e limited i n that they appear not t o have anything t o say about such traditionally central issues as the characteristic s p a t i a l properties of turbulent flow, t h e d y n d c s of vorticity, etc.
Instead, they attempt t o clarify t h e apparently stoch-
a s t i c character o f t h e f l o w and i t s peculiar dependence-independence on i n i t i a l conditions.
To explain what t h i s means, l e t us look b r i e f l y a t
two important but not completely precise distinctions
- between consenra-
t i v e and dissipative systems end between stable and unstable ones. Intuitively, when we say t h a t a system i s conservative, we mean t h a t , once it has been started i n motion, it w i l l keep going forever without
*It
may be t h a t t h i s ceases t o be t r u e for "fully developed turbulence"
wid that what we say here applies t o turbulence a t r e l a t i v e l y low Rey-
nolds numbers and not a t high Reynolds numbers.
f u r t h e r external driving.
Mathematically t h i s is usually reflected i n
t h e f a c t t h a t t h e equations of motion may b e written i n Hamiltonian f o m , with t h e consequent conservation of energy and phase space volume.
Amon@:
numerous examples, l e t u s note a.
t h e Newtonian two-body problem
b.
t h e motion of a f i n i t e number of f r i c t i o n l e s s and p e r f e c t l y e l a s t i c b i l l i a r d b a l l s on a rectangular table.
These examples i l l u s t r a t e t h e d i s t i n c t i o n between s t a b l e and unstable systems.
The N&onian
two-body system i s s t a b l e i n t h e sense t h a t t h e
e f f e c t s of small perturbations of t h e i n i t i a l conditions grow slowly i f a t a l l and hence t h a t long-term predictions about t h e s t a t e of t h e system are possible on t h e b a s i s of approximate information about t h e i n i t i a l state.
In t h e b i l l i a r d system, on t h e other hand, even very small changes
i n t h e initial s t a t e a r e soon amplified so t h a t they have l a r g e e f f e c t s . I f t h e system i s s t a r t e d out repeatedly, in almost but not exactly t h e same way, t h e long-term h i s t o r i e s w i l l almost c e r t a i n l y be t o t a l l y different.
I n t h i s sense, although t h e motion i s s t r i c t l y speaking
deterministic, it i s from a p r a c t i c a l point of view effectively random; t h e coarse features of t h e s t a t e of t h e system a t large times depend on unobservably f i n e d e t a i l s of t h e s t a t e a t time zero. Consider next d i s s i p a t i v e systems.
I n t u i t i v e l y , these have some
s o r t of f r i c t i o n a l mechanism which tends t o damp out motion and must therefore be driven by external forces i f they a r e not simply t o stop. A mathematical t r a n s c r i p t i o n of t h i s notion which i s as general as t h e
corresgondence "conservative
r ~amiltonian" does not seem t o e x i s t , but
it is ,-ruerally not d i f f i c u l t t o agree on whether a given dynamical system i s dissipative o r not.
We w i l l consider systems driven
~ time-
independent forces, such as a viscous f l u i d flowing through a pipe or e l e c t r i c c i r c u i t s driven by batteries.
I n many cases these systems
display behavior which i s simpler than that of conservative systems
--
they may tend, independent of how they are started out, t o a aynamical equilibrium i n which driving forces a r e exactly balanced by dissipation. A system which tends t o t h e same equilibrium, no matter where i n its
s t a t e space it starts, appears t o forget its i n i t i a l conditions and hence t o be "even more stable" than t h e conservative Newtonian two-body system considered above.
Long-term predictions can be made which don't depend
on the i n i t i a l s t a t e but only on t h e parameters appearing i n t h e equations of motion. The next simplest possible behavior i s t h e existence of a globally
attracting periodic solution or l i m i t cycle.
In t h i s case the equations
) x0(0) of motion admit a s o l ~ i o n x o ( t ) with x o ( ~ =
f o r some r > 0,
and every solution of t h e equations of motion converges t o the s e t
{xo(t): 0
< t < .r)
as t
+ m.
What usually happens i n t h i s situation
is in fact something more special: tl(xl)
nth
0
tl < T
For each i n i t i a l
xl there exists
such t h a t
Although t h e long-time behavior is no longer completely independent of the initial point, the r o l e of t h e i n i t i a l point i s simply t o determine the phase t
1'
Again t h e motion s a t i s f i e s our i n t u i t i v e criterion f o r
s t a b i l i t y ; t h e long-term effect of a s d l change in t h e i n i t i a l point is simply a small change i n the phase.
It is natural t o ask what comes next a f t e r periodic o r b i t s in t h e
hierarchy of complexity f o r dissipative systems.
One plausible guess,
advocated by Landau among others, i s t h a t instead of having a single period, t h e system may have two o r more independent periods
-- i .e .,
t h a t the s t a t e space may contain a t o r u s of dimension two o r greater which i s invariant under t h e solution flow, which a t t r a c t s a t l e a s t nearby solution curves, and on which t h e solution flow reduces i n approp r i a t e co-ordinates t o uniform velocity flow.
Although t h i s c e r t a i n l y
can happen, it i s not l i k e l y t o be common since it i s destroyed by most small perturbations when it does occur.
What t u r n s out t o be much more
l i k e l y i s t h e presence of what have come t o be called "strange a t t r a c t o r s " s e t s invariant under t h e solution flow and a t t r a c t i n g nearby o r b i t s but which, instead of being smooth manifolds l i k e t o r i , have a complicated Cantor-set -like s t r u c t u r e
.
We w i l l present shortly a simple example of
a system with such a strange a t t r a c t o r , but before doing so we need t o introduce some notions from t h e q u a l i t a t i v e theory of d i f f e r e n t i a l equations. The s t a t e s of t h e physical system we a r e considering w i l l be assumed t o form a menifold M which we w i l l t a k e t o b e finite-dimensional (although much of t h e formal theory extends e a s i l y t o infinite-dimensional manifolds).
The equations of motion w i l l be taken t o be first-order
ordinary d i f f e r e n t i a l equations on M which we w i l l write i n t h e c l a s s i -
c a l co-ordinate form
where n
is t h e dimension of
M.
To avoid uninteresting complications
we w i l l assume t h a t the right-hand side is an i n f i n i t e l y differentiable f b c t i o n of
xl,.
..
,X
t
n*
We w i l l denote t h e solution mappings by Tt ,
so T x i s the solution curve passing through x
will assume t h a t , for any x, {T~x:t > 0 )
t
T x
a t time zero.
We
exists for all t > 0 and There a r e many interesting cases in
is relatively campact.
which t h i s condition i s s a t i s f i e d but in which solution curves do not in general exist f o r a l l t < 0; i f t h e s t a t e space M
t h e condition i s automatically satisfied
i s compact.
The mathematical transcription of t h e existence of a dynamic equilibrium t o which the system tends no matter how it i s s t a r t e d out i s as follows:
There e x i s t s a stationary solution xo of t h e equations of
m t i o n such t h a t
t l i m T x = xo
tfor all x E M. tionary solution.
Such an xo is said t o be a Gobally attractinn staMore generally, a stationary solution xo i s locally
attracting i f
for a l l x
in some neighborhood of
x,,.
While it i s generally d i f f i c u l t
t o determine whether a stationary solution is globally attracting, there
i s a simple sufficient condition for a stationary solution t o be a local attractor:
It suffices t h a t the matrix of p a r t i a l derivatives
giving t h e linearized equation of motion st xo have ell i t s eigenvalues in t h e open l e f t half -plane. '
We have already defined what we mean by saying t h a t a periodic solution t o t h e equations of motion is globally attracting; we w i l l similarly say t h a t a periodic solution locally a t t r a c t i n g i f f o r all x set
{z(t): 0
< t < T)
(g(t)
with period
T
is
in some neighborhood of t h e points
we have
To give a linear c r i t e r i o n f o r periodic solution t o be .locally a t t r a c t ing
which i s analogous t o t h e one given above for stationary solutions,
we introduce t h e notion of t h e Poincard map associated with a periodic solution.
Take a
gmall
piece
t o t h e periodic solution.
E of n-1 dimensional surface transverse
For each y
t h e periodic solution define @(y) solution curve
on
C and sufficiently near t o
t o be t h e f i r s t point on t h e forward
{TtY: t > 0) which i s again in
point where t h e periodic solution crosses
E,
t. I f
then
yo
denotes t h e
@(yo)= yo.
In
order t h a t the periodic solution be l o c a l l y attracting it is sufficient t h a t the derivative matrix
have all i t s eigenvalues i n t h e open unit disk. some s e t
yl,.
..,ynml
ere, we have
of l o c a l co-ordinates f o r C
chosen
and expressed
@
i n terms of these co-ordinates.) The two simple situations described above solutions and attracting periodic solutions related v i a t h e
biiurcation.
- attracting stationary
- t u r n out t o be closely
suppose t h a t our d i f f e r e n t i d equation
r which may indicate, f o r example, how hard
depends on a parameter
t h e system i s being driven.
Suppose also t h a t f o r some value
a stationary solution x = xr
rc of
r
changes from stable t o unstable by having
a ccanplex conjugate pair of eigenvalues for t h e linearization
at
xr
cross from t h e l e f t t o t h e right half-plane at non-zero speed.
It turns out t h a t , under these circumstances, i f a certain complicated
combination of the f i r s t , second, end t h i r d p a r t i a l derivatives of with respect t o x
at x = xr,
s l i g h t l y larger than
F
r = rc, is positive, then for r
rc there is an attracting periodic solution which
can be regarded as making a small c i r c l e around t h e now-unstable station-
ary solution xr.
As
r decreases t o rc t h e periodic orbit shrinks
down t o t h e single point
xr
solution undergoes a normal
.
In t h i s case we say t h a t t h e stationary bifurcation t o a periodic solution.
is also possible f o r t h e above-mentioned complicated combination of
p a r t i a l derivatives of
F t o be negative.
In t h i s case no attracting
It
periodic solution i s formed.
Instead, f o r r s l i g h t l y smaller
r
as
increases t o
rc,
xr which shrinks d m t o
t h e r e e x i s t s an unstable periodic solution near x r
than
rc. In t h i s case we s a y t h a t t h e stationary
C
solution undergoes an inverted bifurcation.
(Other, more complicated,
things can happen i f t h e combination of partial derivatives i s zero, o r i f nore than two eigenvalues cross t h e imaginary axis simultaneously. ) For a detailed discussion of t h e Hopf b i f u r c a t i o n end r e l a t e d phenomena, see [ 6 1. We next need some more general notions which apply even i n t h e absence of stationary and periodic solutions. s t a t e space we define t h e
tn
+
-.
implies t h a t
s e t of
The assumption t h a t w(x)
i s not empty;
t {T x: t w(x)
t invariant under t h e s o l u t i o n flow T verges as
t
x,
denoted by
x
dx),
in the t o be
t ~ ( x ) i s t h e s e t of a l l c l u s t e r points of sequences T "x
Alternatively, with
*limit
For any point
+
-
.
> O}
i s r e l a t i v e l y compact
i s a l s o evidently closed and I f t h e solution curve
t o a s t a t i o n a r y solution
xo
conversely, if ~ ( x )contains only one point
then xo
t Tx
con-
w(x) = {x0} ;
then
xo
is a s t a t i o n -
ary solution and
l i m T'X = xo. Similar statements hold f o r solution tcurves converging t o a periodic solution. Let
t Then every forward solution c w e T x
converges as t
-+
-
to
5,
and
5
i s t h e smallest closed s e t with t h i s ~ropert'ty. I n order t o under-
stand t h e behavior of solution curves Tor large positive times, it is enouqh t o study t h e solution flow on and near
6,
i-e.,
is t h e
essential part of t h e s t a t e space from t h e long-tern point of view.
One
important difference between dissipative and non-dissipative systems i s "
that
fl tends t o be sma3.1 f o r dissipative systems and t o be t h e whole To see the l a t t e r fact we f i r s t
s t a t e space f o r conservative systems. note:
Proposition. T~
Any finite measure
ha8 support in
Proof.
)1
invariant under the s o t u t i a flow
6.
We want t o show t h a t
ydp = 0 for any continuous f b e t i o n
whose support i s compact and d i s j o i n t from of
p
under
fi.
By t h e assumed invariance
t T ,
Now, since t h e support of
(p
is disjoint from
lim ~ ( T ~ =x o)
t*
for
6, x
so by t h e dominated convergence theorem
I f , f o r example, we consider a Hroniltonian system with t h e property
that
{x: ~ ( x ) 9 E)
i s compact f o r each
E, then Licuville's Theorem
implies t h a t every point of t h e s t a t e space i s i n t h e support of some invariant measure and hence, by t h e proposition, t h a t
-
fi i s t h e whole
s t a t e space. We w i l l say t h a t a point for
T~
x
of t h e s t a t e space i s a wanderinq point
i f t h e r e i s a neighborhood
Tt U n U =
The non-wandering
U
of
x
such t h a t
f o r a l l suffliciently large t
R is t h e s e t of
.
all points which a r e not wander-
It follows a t once from t h e definition t h a t t h e s e t of wandering
ing.
points i s open and hence t h a t
Q i s closed; it i s also easy t o see t h a t
f o r all x.
.
Hence,
5
6 $ Q,
equality does hold i n most i n t e r e s t i n g cases (including flows
CQ
While it i s not d i f f i c u l t t o construct flows f o r which
which s a t i s f y h a l e ' s Axiom A, t o be described below).
If t h e solution
flaw has a globally a t t r a c t i n g stationary solution xo, then
a similar statement holds when t h e r e i s a globally a t t r a c t i n g periodic solution. The simplest asymptotic behavior a d i f f e r e n t i a l equation can have
i s f o r all solution curves t o converge t o t h e same stationary or periodic solution.
One trivial way in which t h e s i t u a t i o n can become more compli-
cated i s to have several locally attracting stationary and periodic solutions.
If
xo
i s a locally attracting stationary solution define
the basin of attraction
of
t o be
t B ( I X ~ I=') {x: lim T x = x0} t-
.
One can define in a similar way t h e basin of attraction of a periodic solution.
A basin of attraction i s open and invariant under t h e solution
('The fact that it i s open may not be quite obvious.
flow.
case of a l o c a l l y attracting stationary solution xo the basin of attraction of
xo.
Consider t h e
and l e t x be i n
Because x 0 is l o c a l l y attracting there
i s a neighborhood V of
xo such t h a t every solution curve beginning t in V converges t o %, and because T x converges t o xo there t exists t such that T x E V. Then (T '1-4 i s an open neighborhood 0 of
x
contained i n the basin of attraction of
x,,. ) Note t h a t , since
basins of attraction are open and disjoint, the s t a t e space M,
if
connected, cannot be written as t h e union of two o r more basins.
Hence,
i f there a r e a t l e a s t two locally attracting stationary o r periodic solution, there must be some solution curves (lying on the boundaries of t h e basins of attraction) which do not converge t o any of them. We are going t o investigate a t t r a c t o r s which are more complicated than single points and periodic solutions, and we should therefore define precisely and generally what we mean by an a t t r a c t o r .
Unfortunate-
l y , no such definition seems t o be agreed upon, so we w i l l improvise by
l i s t i n g a number of properties which an a t t r a c t o r ought t o have, being careful t h a t t h e conditions are indeed s a t i s f i e d in t h e special case of Axiom A a t t r a c t o r s , where t h e r e does exist an accepted definition ( t o be
discussed below).
To begin with, an a t t r a c t o r should be a closed
(compact?) subset
of t h e s t a t e space, invariant under t h e solution
X
f l a w , which a t t r a c t s nearby o r b i t s i n t h e sense t h a t there exists an
open s e t U
containing X
such t h a t , for any x
in U
t l i m d ( x~, ~ )= 0 tta,
(or equivalently,
.
w(x) C X )
then
t T x remains near X
B(x)
of
X
Second, we require t h a t i f
for all t > 0.
x i s near X
The basin of attraction
i s now defined t o be
The argument given above f o r attracting stationary solutions i s e a s i l y adapted t o show t h a t
B(X) i s open.
We also want t o put i n t o the defini-
tion some condition which prevents an a t t r a d o r from being decomposable i n t o a f i n i t e number of other a t t r a c t o r s ; a good way t o do t h i s i s t o require t h a t some solution curve contained i n t h e attractor i s dense i n t h e a t t r a c t o r , i.e.,
that t h e solution flox T~
t o p o l o ~ i c a l l yt r a n s i t i v e .
restricted t o
X
is
Nothing i n t h e above l i s t of conditions pre-
vents t h e whole s t a t e space from being an a t t r a c t o r ; it w i l l be one i f t h e r e i s a single solution curve dense in t h e whole s t a t e space, and t h i s frequently happens f o r conservative systems i s taken t o be a single energy surface).
( i f t h e s t a t e space
For dissipative systems, on the
other hand, a t t r a c t o r s w i l l generally be s m a l l a t l e a s t in t h e sense of having empty interiors. The preceding discussion has considered only continuous flows
T~.
For many purposes it is useful t o have a p a r a l l e l s e t of definitions f o r the corresponding discrete situation, the s e t of powers single smooth transformation
T of a manifold.
{ T ~ )of a
The task of adapting our
discussion t o t h i s s l i g h t l y different context i s straightforward; we leave it t o t h e reader. Up t o t h i s point, we have been dealing with elementarg general considerations.
Although necessary i n order t o get s t a r t e d , t h e ideas
developed so f a r do not seem t o be sufficiently specific t o lead t o any very interesting analysis.
In order t o go further we must impose addi-
t i o n a l r e s t r i c t i o n s on the systems we study.
In recent years, it has
turned out t o be particularly f r u i t f u l t o impose some s o r t of hyperboli-
city condition.
The fundamental reference i n this area is [ 1d; we w i l l
sketch here a few of t h e basic ideas. W e expleined above t h a t a stationary solution xo
i s a t least a
local a t t r a c t o r i f all eigenvalues of t h e matrix
are i n t h e open l e f t half-plane.
More generally, we say that
hyperbolic stationary solution i f no eigenvalues f o r DF on the imaginary axis.
xo i s a
are precisely
In this case. Efn s p l i t s into two complementary
subspaces
ES and EU, each invariant under DF,
values of
have s t r i c t l y negative real p&s while t h e eigenvalues ES have s t r i c t l y positive r e a l parts ; and E~ are called
of
such that the eigen-
DF~
DF~ En respectively t h e stable and unstable eigenspaces f o r DF. ?enote the linearization of
F a t xo, i .e.
I f we l e t
,. F
then a solution x"(t)
as t
converges t o xo to
xo
at
t
+
of t h e linearized equation
-
-
+
-
i f and only i f
i f and only i f
i ( 0 ) € xo + E'
and converges
;(o) € xO + E ~ . Going back t o t h e
f u l l (non-linear) equation we define t h e s t a b l e and unstable manifolds a t xO ( f ' ( x o )
wU(x0) respectively)
by
t wU(rO) = {x EM: T x
-+
xo
as t
+
- -1
From t h e definition it i s not apparent t h a t these s e t s are submanifolds, o r even t h a t they contain any points other than
xo,
Stable Manifold Theorem f o r Hyperbolic Fixed Points.
submmifolde* of
M,
but we have:
f' (xO) wd
w-ith dimensions equal ~espectiveZyt o
wU(x0) are
dim ~'(x,,)
and dim EU(rO). These ~ u b n m i f o Z Lcontain xo and are tangent a t xo t o E ' ( X ~ ) and
respectiveZy.
*
There i s a technical d i s t i n c t i o n which needs t o be noted here. The s t a b l e and unstable manifolds are immersed but not i n general imbedded submanifolds of M. This means t h a t , although made up of countably many smooth pieces, t h e y can fold back a r b i t r a r i l y near themselves. A simple example of an immersed one-dimensional submanifold of ~3 i s t h e "Lissajous figure"
whce
u1,u2,
end ul/u2
a r e all i r r a t i o n a l .
With these elementary examples for motivation, we will now give a general definition of hyperbolic set.
There are i n f a c t two definitions,
t
one for transformations 4 and one f o r flows T the definition f o r transformations. and invertible mapping and l e t by A.
Thus, l e t
.
We w i l l give only
9 be a differentiable
A be a compact s e t mapped onto i t s e l f
For t h e sake of concreteness we w i l l assume t h a t a single s e t of
co-ordinates can be chosen f o r an open s e t containing A ( i .e., we w i l l a d as i f
A
is contained in Eln); there i s , however, no d i f f i c u l t y in
eliminating t h i s assumption by giving a coordinate f r e e version of t h e definition.
We define t h e derivative of
4
at x
t o be t h e n
x
n
matrix
end similarly define
D O ~ ( X )f o r any integer m
( ~ o s i t i v eor negative).
EJy t h e chain r u l e
We are going t o define A t o be a hyperbolic set f o r
cP
i f , roughly
speaking, any infinitesimal displacement from a point
x
belonging t o A
can be decomposed as t h e sum of two infinitesimal displacements, o n e p f
which contracts exponentially under positive powers of
O
of which contracts exponentially under negative powers of
2recisely: x E h
h
i s a hyperbolic
set
for O
and the other O.
More
i f t h e r e e x i s t s f o r each
a s p l i t t i n g of B~ i n t o a direct sum of complementary subspaces
z 3 ( x ) , E~(X) such that:
Fur some c > 0, 1 < 1, which do not depend on
x
or
m,
I n addition t h e s p l i t t i n g i a required t o be invariant under 8 :
and t o vary continuously with x. ing:
For every xo € A,
such t h a t
ns = aim ~ ' ( x ) and n
,. ..,&(XI
ESb)
5
t h e r e e x i s t s an open neighborhood U
such t h a t t h e r e e x i s t El(x)
This l a s t condition means t h e follow-
n
of
U
xo
= dim EU(x) a r e constant on U
and
cont inuoua IRn-valued functions
defined on
is spanned b y c1(x)
u nA
such t h a t , f o r each x E A
,...,cn,(x)
n U,
and EU(x) is spanned by
b),...,$,(XI. Alternatively, a s i t t m r r o u t , i t i s e n o u g h t o
ns+l require t h a t
i ( x , ~ ) :E E E'(x))
a r e both closed subsets of
A x R";
and
'
{ ( x , ~ ) :5 E ~ ~ ( x ) )
continuity as formulated above then
follows automatically. It i s p a r t i c u l a r l y i n t e r e s t i n g t o apply t h i s definition with
mn-wandering s e t of
8.
We say t h a t
8 s a t i s f i e s Axiom A
if
h
the
1.
2.
i s a hyperbolic s e t
The periodic points f o r 8
are dense in
n.
This condition has proved very f r u i t f u l for mathematical analysis.
It i s ,
on the other hand, hard t o verify in practical applications and non-trivi a l examples axe r e l a t i v e l y scarce.
It i s a t t h i s time s t i l l too early
t o decide whether Axiom A as it stands is too r e s t r i c t i v e t o apply t o cases of i n t e r e s t , but e i t h e r it o r some weakened version of it seems l i k e l y t o p l w en important r o l e i n future developments. It may be helpful t o note here one difference between hyperbolic
fixed points and more general hyperbolic sets. x
cannot be an a t t r a c t o r unless
A hyperbolic fixed point
E ~ ( S )is t r i v i a l .
r e s t r i c t i o n f o r general hyperbolic s e t s .
There i s no such
It frequently happens t h a t a
hyperbolic a t t r a d o r is made up l o c a l l y of i n f i n i t e l y many smooth "leaves"
- lower-dimensional
surfaces which are everywhere tangent t o
~ ~ ( x )Two . nearby points on t h e same leaf mve apart under t h e action of the transformation, but t h e whole assembly of leaves is attracting.
Chapter 11. The Lorenz System
We t u r n now from generalities t o a discussion o f a p a r t i c u l a r system of equations.
This system could hardly be simpler
-- t h e
state
space i s three-dimensional and t h e equations a r e
with b ,o,
r
positive constants
- but
it displays a bewildering
assortment of non-trivial mathematical phenomena.
SO far as I know, t h i s
system of equations was f i r s t seriously investigated by E. N. Lorenz [ 5
1
some f i f t e e e n years ago; i n recent times it has been studied intensively by Yorke , Guckenheimer [ 3 ] , Martin and McLaughlin McCracken [ 6
1, and
Williams [ 123, emong crthers
[ 81, Marsden and
.
One of t h e appealing aspects of t h e Lorenz system is t h e fact t h a t
it was not constructed f o r t h e purpose of proving t h e p o s s i b i l i t y of complicated behavior; r a t h e r , it turned up i n t h e course of a p r a c t i c a l investigation.
In h i s o r i g i n a l paper, Lorenz was l e d t o t h i s system
by t h e following considerations:
Consider t h e equations of motion f o r two-
dimensional convection i n a container of height
H
and length
L.
These
equations can be viewed, h e u r i s t i c a l l y a t l e a s t , as a f i r s t order differe n t i a l equation on an i n f i n i t e dimensional s t a t e space; t h e points of t h e
s t a t e space are p a i r s consisting of a stream fbnction temperature f i e l d T (x,z)
+(x,z)
and a
subject t o appropriate boundary conditions.
Look f o r solutions of t h e form
where T-, T+
denote t h e temperatures at t h e bottom and t h e t o p of t h e
container respectively.
(Such solutions correspond t o free o r no-stress
boundary conditions on the velocity f i e l d and t o t h e absence of heat flow through t h e ends of t h e container.)
Express t h e equations of
motion d i r e c t l y i n terms of t h e Fourier coefficients
€lm,n(t), I J J ~ , ~ ( ~ )
and d r a s t i c a l l y truncate t h e resulting i n f i n i t e s e t of coupled differe n t i a l equations by putting a l l except
IJJm,n,8m,n identically equal t o zero
1111,l'81,lS 8 Now put 0,2-
and choose
cl, c2, c3 so as t o simplify the d i f f e r e n t i a l equation; t h e
result i s t h e Lorenz system. With t h i s derivation canes a physical interpretation f o r t h e parameters
b,U,r.
Specificslly,
b
is a simple geometric constant
(4/(1+ ( H / L ) ~ ) ) ,U is t h e Prandtl number (i.e., t h e r a t i o of viscosity t o thermal conductivity) and r
i s a numerical constant times t h e Rayleigh
number, i.e., difference
i s a dimensionless rider proportional t o t h e tenperature (T-
- T+).
and 10 respectively.
The values of
b
and o will be fixed a t 8 / 3
We w i l l f i r s t discuss schematically how t h e
behavior of t y p i c a l solutions changes with what happens f o r a p a r t i c u l a r value of
r ; then describe i n d e t a i l
r.
We begin with a number of elementary observations about t h e Lorenz system: i)
he
The equations are invariant under t h e transformation
physical origin of t h i s symmetry i s invariance of t h e equations of
m t i o n unde'r r e f l e c t i o n through a v e r t i c a l l i n e a t t h e center of t h e
.
container ) ii)
The solution flow
T~
volumes i n t h e s t a t e space lR3
generated by t h e Lorenz system shrinks a t a uniform r a t e .
This follous from
t h e equation
This r a t e i s in f a d quite l a r g e ; a s e t of s t a t e s occuping unit volume a t time zero occupies only t h e v o l ~e iii) values of
A l l solutions are bomded f o r
X,Y,Z
= t > 0,
a r e damped by t h e motion.
at time one. and very l a r g e i n i t i d
To show t h i s we introduce
An elementary computation gives
with constants on
r ,U,b.
cl, c2 which do not depend on
X,Y ,Z,
but may depend
he essential point is t h a t , despite t h e quadratic terms
i n t h e equations of motion, there are no cubic terms i n
du at .)
It
follows e a s i l y t h a t every solution curve eventually gets and stays i n t h e i n t e r i o r of t h e b a l l
B where u ( 2 c2/c1. This b a l l i s mapped t into i t s e l f by T , and by ii) t h e volume of its image under T~
goes t o zero as t
goes t o i n f i n i t y .
TbX converges t o large t
Hence, every solution curve
t o the set
which i s closed and has Lebesgue measure zero. f o r T~
The non-wandering s e t
is contained in t h i s intersection and therefore also has
measure zero. We now describe what happens as r is varied s t a r t i n g from zero. Recall t h a t , in t h e derivation f%m t h e convection equations,
r was
yroportional t o t h e imposed v e r t i c a l temperature difference and is therefore a measure of how hard t h e system i s being driven.
For
r
between zero and one, inclusive, it i s not hard t o show t h a t
2
is
a globally attracting stationary solution.
As
r
is increased past
one, t h i s solution becomes unstable and bifurcates i n t o a pair of locally a t t r a c t i n g stationary solutions
C=
(m)
, r-1) ,
.
$2
-C '
= (-~bm -A=) ), ,r-1).
These are easily checkecl t o be t h e
only stationary solutions aside from
2
; they remain present, but
not necessarily stable, f o r all r > 1. Physically, they reuresent steady convectim.
Also for a l l r > 1 t h e stationary solution
is
hyperbolic, with a two-dimensional stable manifold and a one-dimensional unstable manifold.
For
r s l i g h t l y greater than one, nearly a l l solution
curves converge e i t h e r t o C or t o
C) f o r large time; t h e only exceptions
are those on t h e two-dimensional stable manifold of The two steady convection solutions
r = 470/19
3
before t h m .
A t a special value of
f o r s l i g h t l y l a r g e r values of orbits. to
< remain stable u n t i l
but various interesting things happen
24.74 ( f o r u = 101,
sional unstable manifold for
and
E.
r
around r = 13.9 the one dimen-
returns t o
2
(homoclinic o r b i t ) , end
r there are two unstable hyperbolic periodic
It is not known whether t h e non-wandering s e t
{c,c',~a l l the way up t o t h i s value
of
r o r whether periodic
solutions or other kinds of recurrence appear e a r l i e r . t h e appearances are t h a t , u n t i l
R remains equal
r i s nearly equal t o
Nevertheless, 470/19, a l l
solution curves except f o r a s e t of measure zero converge t o one of This does not remain true a l l t h e way up t o
or
what happens f o r
r's
C
r = 470/19, but
s l i g h t l y below t h a t c r i t i c a l value i s most easily
understood i n terms of what happens above it. As
r passes t h e c r i t i c a l value of
470/19, both
C and
become unstable through having a complex-conjugate p a i r of non-real eigenvalues cross i n t o t h e right half plane.
This does n o t lead, v i a
a normal Hopi bifurcation, t o stable periodic orbits near
C
and
5'
for
r
s l i g h t l y above
470/19.
Instead (see [ 8
1, [ 6 1),
what
happens i s an inverted Hopf bifurcation i n which an unstable periodic solution contracts t o each of to
470/19.
c , ~ 'and
disappears as
The behavior above t h e c r i t i c a l v a u e of
r
increases
r seems not
t o be accessible t o analysis by "infinitesimal" b i f u r c a t i o n theory but requires a global investigation of t h e behavior of solutions which has so f a r been possible t o carry out only by following solutions nlnnerically on a computer.
The next stage of our discussion w i l l be
a description of t h e r e s u l t s of such a numerical investigation, carried
r = 28.
out f o r t h e a r b i t r a r i l y chosen value
In t h e v i c i n i t y of any one of t h e t h r e e s t a t i o n a r y solution
2,C,Cf
,
t h e m t i o n i s similar t o t h a t given by t h e l i n e a r i z a t i o n a t t h e s t a t i o n a r y solution.
For each of t h e
of complex eigenvalues
2
.094
+
and
C' ,
10.2 i
t h e l i n e a r i z a t i o n hes a p a i r
and a negative eigenvalue -13.85.
Hence, i n t h e l i n e a r i z e d motion, t h r e e things a r e going on a t quite d i f f e r e n t speeds: i ) t h e compnent in t h e negative eigendirection damps out
rapidly ii )
t h e component i n t h e two-dimensional r e a l eigenspace associated
with t h e complex eigenvalue p a i r r o t a t e s a t a moderate speed and d s o iii)
expands slowly.
More s p e c i f i c a l l y t h e "rotation period'' i s component i s multiplied by about 2 x
lo4
2nh0.2 = .62; t h e ccmtracting f o r each r o t a t i o n ; and the'
r o t a t i n g component expands by about 6% with each r o t ation.
The same
q u a l i t a t i v e picture holds f o r t h e correct (not linearized) motion near
-C
and
2'
.
Passing through each of these points i s a two dimensional
surface, i t s unstable manifold, which strongly a t t r a c t s nearby solution curves and along which solutions s p i r a l slowly outward.
The approximate
appearance of a t y p i c a l solution curve i s shown in figure 1.
Figure 1
The normal t o t h i s surface a t
C
has polar angle 70' with respect t o t h e
Z axis and azimuthal angle 153O with respect t o t h e X
axis.
A s it turns out, t h e r e i s quite a l a r g e domain around each of
where t h i s picture is q u a l i t a t i v e l y correct.
2, C'
This does not yet t e l l us
much about t h e asymptotic behavior o f t y p i c a l solutions since t h e steady growth of t h e r o t a t i n g component eventually drives t h e solution curve
out of t h i s domain, and we have t o look a t where it goes.
The key t o
understanding t h e recurrent behavior of t h e Lorenz system i s t h e fact t h a t it usually goes i n t o t h e corresponding d d n around t h e other steady convection solution, where it i s again a t t r a c t e d t o t h e unstable manifold, eventually pushed out again, returns t o t h e o r i g i n a l domain,
and proceeds t o repeat t h e whole process.
The r e p e t i t i o n i s , however,
typically only approximate and may d i f f e r quite a l o t from t h e f i r s t cycle i n d e t a i l .
Although most o r b i t s continue forever s h u t t l i n g back
and forth between
and
C' ,
they are only exceptionally periodic
o r even asymptotic t o periodic solutions. To form a more precise picture of t h i s process we take a section with t h e horizontal plane
Z = 27 containing
C
and
c'.
Solution curves
then become simply discrete s e t s of points, and we w i l l i n fact keep track only of these crossing points where We thus define a "~oincar; map"
4
dz
< 0 (downcrossings).
of the plane t o i t s e l f which takes
each point t o t h e next downcrossing on i t s solution curve. see,
4
i s not defined everywhere, but it does t u r n out t o be defined
almost every-where.) when
8
(AS we s h d
Figure 2 shows what happens t o a domain around
C
i s applied t o it a few times; t h e transverse scale is grossly
exaggerated and t h e s t r i p s are r e a l l y much thinner than indicgted:
Figure 2
The figure r a i s e s t h e question of what i r o n t h e domain around
2
i s attached t o
c',
of
C
t o t h e domain around
2'.
Since one end
and t h e other i s a t t r a c t e d t o t h e unstable manifold
continuity considerations would seem t o suggest t h a t t h e s t r i p
w i l l get pulled diagonally across from
not t h e case: map
O does t o t h e s t r i p running
to
Such i s , however,
2'.
Although t h e solution flow i s continuous, t h e Poincar6
@ need not be, and it i n f a d undergoes a ,jump discontinuity
between
2
and
2'
.
The source of t h i s discontinuity i s t h e t h i r d stationary salution at
2.
Recall t h a t
2
is a hyperbolic stationary solution with a two-
dimensional stable manifold. motion t h a t t h e
It i s easy t o see from t h e equations of
Z axis i s contained i n t h e s t a b l e manifold of
we should expect t h e s t a b l e manifold t o i n t e r s e c t t h e plane a curve passing through
X = Y
3
w i l l c a l l t h e curve i n question
so
Z = 27 i n
0.
This t u r n s out t o be correct; we
Z;
it is shown i n Figure 5
running from upper l e f t t o luwer right.
.
2,
as
( ~ c t u a l l y t, h i s is only one
of i n f i n i t e l y many pieces of t h e intersection of t h e s t a b l e manifold of
-0
with t h e plane
Z = 27.)
Solution curves s t a r t i n g on C
proceed monutonically t o
2
i s not defined along Z.
Let us investigate what
approaches
Z
and never return t o t h e plane
from t h e upper r i g h t by t r a c i n g t h e orbit
very s l i a t l y above C.
For a long time
2
Z = 27; as
@
z
Tt& f o r
&
t T & tracks i t s neighbors on
t h e s t a b l e manifold and hence gets very close t o to
@(x)does
simply
2
.
While it i s close
i t s motion is well approximated by t h e l i n e a r i z a t i o n of t h e
equations of m t i o n a t zero.
The l i n e a r i z a t i o n has two negative eigen-
values and one positive one; t h e eigenvector corresponding t o t h e posit i v e eigenvalue i s horizontal ( i .e. has
Z-component zero).
In the
linearized motion, t h e components in t h e negative eigendirections decay steadily t o zero while t h e component in the positive eigendirection grows.
I n i t i d l y t h e negative eigencomponents s r e much l a r g e r than
t h e positive one (since t h e t r a j e c t o r y comes in near t h e s t a b l e manifold), but, uuless t h e positive eigencomponent i s exactly zero, it w i l l eventually dominate t h e others and t h e t r a j e c t o r y w i l l move awq from zero along t h e positive eigendirectiorl.
The modifications introduced i n t o t h i s picture
by t h e non-linear terms i n t h e interaction are simple:
A trajectory
entering t h e v i c i n i t y of zero near t h e stable manifold leaves along t h e unstable manifold; t h e closer it is t o t h e stable manifold i n i t i a l l y , t h e closer it will get t o zero and t h e closer it w i l l be t o t h e unstable manifold when it leaves.
The unstable manifold consists of two solution
curves, growing out of t h e origin i n opposite directions; which of these branches w i l l be i d l o v e d is determined by which side of t h e stable manifold t h e t r a j e c t o r y l i e s on.
See figure 3.
Figure 3.
Since t h e solution we are following s t a r t s s l i g h t l y above
C in
figure 5, it w i l l eventually be picked up by t h e branch of t h e unstable manifold of
2 along which
X
and Y
i n i t i a l l y increase.
,
of t h e unstable manifold makes a large loop around
This piece
as shown
schematically i n figure 4, end makes i t s f i r s t downcrossing of t h e plane
A,
2 = 27 a t t h e paint
with co-ordinates
comparison, t h e co-ordinates of
2'
Figure
x i s any point @(x)i s near
Thus, if above
z,
below
2
,
A
@(x) i s near
.
A' ,
-C -A
X
+
-X, Y
+
-Y,
Z
+
o or
-8.3).
(-8.5, -9.5))
4
on t h e plane
2 = 27
lying near but s l i g h t l y
On t h e other hand,
if
is slightly
t h e f i r s t downcrossing of t h e other
branch of t h e unstable manifold of under
are
(-5.2,
2
(i.e., t h e symmetric image of A_
z).
The picture, as developed so f a r , i s shown in figure 5.
The arc
i s part of the intersection of t h e unstable manifold of
2 with
t h e plane object for
Z = 27;
C'
.
i t s synnnetric image
A'
i s t h e corresponding
We have indicated by arrows t h e images of a few
important points under
a,
and we have
put
B)
= @(A),
Br
@(A').
This f i g u r e , unlike t h e others i n t h i s c h a ~ t e r ,is drawn c a r e f u l l y t o scale.
Note t h a t , although
B'
appears t o l i e on
f a c t be s l i g h t l y t o t h e right of it.
Figure 5
c'A',
it must in
We are now able t o form a f a i r l y comprehensive image of t h e behavior of a t y p i c a l solution which s t a r t s , say, near
2.
quickly a t t r a c t e d t o t h e unstable manifold of
C; then proceeds t o s p i r a l
C exactly, it will eventually
out along it. Unless it happens t o h i t land somewhere on t h e part of ed as s p i r a l l i n g around
2'
w i l l be somewhere very near
neighborhood of
2,
It should then be regard-
and i t s next downcrossing
Since points near
B' ,
C go t o t h e
t h e general point can land
and A_', and what happens next depends on whether
B'
it lands above o r below
u n t i l it is above C lands above C
r a t h e r than
CIA_'.
.
below C
while A_ goes t o
A_'
anywhere between
C A_
Such a solution is
C.
I f it lands below
C,
it s p i r a l s around
B A_ ; i f it t o B A_ . The
and then makes a t r a n s i t i o n back t o
it immediately makes t h e t r a n s i t i o n back
motion continues i n t h i s way forever, a l t e r n a t e l y moving around
-
C';
t h e only way it can stop is f o r t h e solution curve t o h i t
and
Z
exactly, and t h i s is extremely =likely. A few other features of t h e motion should be noted here.
ell, although t h e motion may s t a r t a r b i t r a r i l y near t o
has gotten away it can never return t o
c5
(or t o
C9 once it
B'.)
thus gaps of non-zero s i z e between t h e stationary solutions
and t h e region where t h e o r b i t i s recurrent. t h e stationary solution a t very close t o
2
The a r c irom A_ t o i t s preimage under
e, a t y p i c a l o r b i t A
to
Then are
C
and
This i s not t h e case with
(but only very infrequently).
s t r e t c h e s out t h e a r c from
F i r s t of
can be expected t o approach Second, r e c a l l t h a t
1 along t h e full length of
g' A'
C i s t h e image under @ of t h e arc from C t o 8,
i .e
., of a s i n g l e cycle under
8 of
C' A
.
8
.
Tracing back through a few applications of
4
from
to
C
is t h e image mder
of a rather smell piece of t h e arc
x
with
just s l i g h t l y outside
know t h a t t h e distance from than the distance from
= to
t h e position of a pint along
@ we find t h a t the arc
8 (n 'about 24, C 4 running t r o m
as it happens)
(O(5) ,
to
g. From t h e l i n e a r approximation we
@(x) t o
i s only &out 6%greater
2, so an C near
uncertainty of about 6% i n leads t o complete uncertainty
about where i t s orbit will land on t h e arc from complete uncertainty about where on
B' A'
A
to
C and hence t o
it w i l l go next.
Thus,
although t h e moticm is completely deterministic, it is unstable in t h e sense t h a t s m a l l changes in t h e i n i t i a l position are amplified rapidly. This means t h a t t h e behavior is effectively random;
t o determine where
an orbit w i l l be a f t e r making a number of transitions from
-B'
A'
BA
to
and back requires unreasonably precise knowledge about i t s i n i t i a l
position.
F i n a l l y , we need t o c l a r i f y an apparently contradictory aspect
of the above description.
Points on
C are on the s t a b l e manifold of O_
and t h e corresponding solution curves approach zero f a i r l y directly. i s natural t o visualize t h e s t a b l e manifold of
as remaining more or
C
leas f l a t all t h e way t o i n f i n i t y and hence a s separating Since a r d u t i o n curve cannot croes t h e stable manifold of
C'
from
2
2'.
.
it would
seem t o be impossible for solutione t o cycle back and forth between and
It
C
The fallacy in t h i s argument l i e s in an incorrect guess about
the global structure of the stable manifold of passing through t h e upper left-hand part of s p i r a l arotlnd
C(
2.
Solution curves
C, when followed backwards,
and those n e i r t h e intersection of
A'
with
C
s p i r a l around it a r b i t r a r i l y often. manifold of
2
Thus, one p a r t of t h e s t a b l e
wraps i n f i n i t e l y often around
i n f i n i t e l y often around
c'.
Another p a r t *aw
C , but i n t h e opposite direction.
The global
i s quite complicated, and it
s t r u c t u r e of t h e s t a b l e manifold of
manages t o stay out of t h e way of t h e m t i o n as described
above.
Raving seen what a t y p i c a l solution c w e looks l i k e , we w i l l next t r y t o construct a more comprehensive view of beginning in a neighborhood of as sketched i n Figure
A'
A
to
O(B),
T: i s mapped t o t h e narrow sh&d
but s l i g h t l y below
two similar s t r i p s .
Consider a neighborhood S
and 6':
cD t h e part of S lying above C Is mapped
t o t h e narrow shaded s t r i p f r o m
to
l. ((B' JA').
6, consisting of two enlongated ovals
Under one application of
lying below
(BA)
& solution curves
C' A'.
while t h e part of s t r i p m i n g from
The image of
S'
S
B'
i s t h e union of
We can a t t h i s point simplify t h e picture substanti-
a l l y by exploiting t h e symmetry of t h e Lorenz system.
We will i d e n t i f y
points i n pairs
(-X, -Y,
27)
and
( ~ , ~ , 2 7 and ) represent each such
pair by i t s member i n the half-plane'
X
>Y
(with some appropriate
convention on t h e l i n e X = Y ) ; correspondingly, we replace CI
the quotient mapping
@
it l i e s in t h e half-plane in figure 6:
sh-
obtained by reflecting t h e image of Y > X.
8
,
to
and
A
maps
A
o r near
A' ,
a l l points near
S. Points near
t o a continuous mapping of a l l of
C
S t o the point
f~
A..
S
The extended
t o A_' i s
- it permits
I: are mapped by
cb
A)
so
but t h e reflection sends
C t o t h e v i c i n i t y of
B'
S into itself.
is a n o t h v advantage t o making the identification be defined on all of
cb whenever
This leaves only half of t h e picture
The shaded reglm running f m m
reflected t o run froln
cb by
.
to
A,
There
6
to
either near
i
sends
may therefore be extended
into i t s e l f , sending the arc may be visualized as obtained
by t h e following three steps: 1. Stretch S
out t o roughly twice i t s original length, while
shrinking it l a t e r a l l y . 2.
Pinch the resulting set along the image of
3.
Fold back i n t o S , w i t h the pinch going t o
Figure 7
C.
A.
Straightening out and broadening t h e picture a b i t , we obtain t h e shaded region below
Here,
,.
@
C goes t o
8s
A
t h e image 'of
S under
with pinching and
&
goes smoothly t o B_. Applying
again gives a s e t consisting of four long t h i n pieces, two inside t h e
upper shaded region of Figure 8 and two inside t h e lower. s t r i p s are pinched together at pinched together at
B
A
All four
and t h e two upper ones are a l s o
Similmly
A,
6 3 ( ~ ) consists o f e i g h t s t r i p s , all pinched together at
four pinched a t
B,
transverse l i n e
P
and two pinched a t
If we i n t e r s e c t with a
we f i n d successively
Figure 10
A
Continuing t h e process,
=
"
P(s)
consists of uncountably n=l many longitudinal arcs end i n t e r s e c t s any transverse a r c l i k e 3 in
Q = n
P
a Cantor s e t .
fi
The longitudinal arcs making up
are Joined together
in t h e following complicated ww:
Each arc i s pinched together a t each
end with uncountably many others.
h s t (but not quite a l l ) arcs pinch
together at
^3
$'(A) ,@ ( A )
A_
,
with t h e other pinches occurring a t
B = $(A),
,... . Note t h a t t h e pinching points a r e exactly t h e successive
downcrossings of t h e unstable manifold of
g.
There w i l l normally be
i n f i n i t e l y many of these pinching points scattered densely through
$,
although i s it also possible f o r t h e s e t of pinching points t o be f i n i t e . This l a t t e r s i t u a t i o n happens i f and only i f
6
A_ i s a periodic point
o r equivalently i f t h e unstable manifold f o r
2
i s contained i n t h e
s t a b l e menifold; we should expect t h i s t o be t h e case f o r a countable dense s e t of values of
for
r in any small neighborhood of 28.
Whether t h e s e t of pinching points i s f i n i t e o r not, it i s easy t o see t h a t
some orbit
$ has
which i s dense i n
i s contained in t h e non-wandering set f o r A
Q
a t t r a c t s all o r b i t s s t a r t i n g in. S,
A
8.
6
and hence t h a t OD
Since
6=
A in(s), n=l so it meets a l l the requirements
of our e a r l i e r provisional definition of a t t r a c t o r , (and it certainly deserves t h e epithet "strange.") solution flow T~
The corresponding a t t r a c t o r for t h e
i s now not hard t o visualize.
I n t h e vicinity of
t h e plane Z = 27 it consists l o c a l l y of stacks of uncountably many two dimensional pieces which intersect transverse arcs i n Cantor sets. Globally these two-dimensional leaves all pinch together along t h e unstable manifold of
g
consisting of two solution curves which we
should expect t o be dense in the attractor.
Although t h e question has
not been carefiilly investigated, it appears t h a t t h e basin of attraction
g, g'
f i l l s all of I R ~ except f o r
and t h e i r respective one-dinensionel
s t able manifolds. We now return, b r i e f l y , t o t h e behavior of t h e Lorenz system f o r
r
s l i g h t l y l e s s than t h e c r i t i c a l value of 470/19.
Because of t h e f i n i t e
gap between t h e stationary solutions
and t h e a t t r a c t o r , it
i s not r e a l l y necessary f o r
2
and
C and
2'
t o be unstable in order f o r t h e
a t t r a c t o r t o e x i s t ; all that i s necessary is t h a t the unstable manifold of
2
which forms t h e outside edge of t h e a t t r a c t o r not f a l l i n t o t h e
C
basin of attraction of periodic orbits near
-C
and
C'
C
or
and
2'. The existence of small unstable g' shows t h a t t h e basins of attraction
of
are not very large for r s l i g h t l y below 470/19, and it
turns out i n fact t h a t t h e unstable m i f o l d of
2
i s not attracted t o
-C
C' unless r i s l e s s than about 24.1.
and
Thus, f o r
24.1 < r < 24.74,
the system has ( a t l e a a t ) three d i s t i n c t a t t r a c t o r s , t h e point attractors and a strange a t t r a c t o r between them.
C, C'
Which a t t r a c t o r traps a
given orbit depends on where t h e o r b i t s t a r t s , but o r b i t s s t a r t i n g near
-0
go t o t h e strange attractor.
F'hysicaJly, t h e system d i s p l w s
hysteresis; it has several possible behaviors depending on i t s past I f we imagine increasing t h e temperature gradient slowly from
history.
zero t h e solution w i l l simply track one of the two stationary solutions
r = 470/19.
up t o
r
I f , on the other hand, a temperature gradient making
s l i g h t l y l e s s than 470/19 is turned on suddenly with t h e system i n i t i a l l y
at r e s t
,a
s t a t e of permanent chaotic mot ion results.
In t h e above discussion, nothing has been s a i d about t h e behavior of t h e Lorenz system f o r
r larger than 28.
Preliminary numerical experi-
ments indicate t h a t several further changes occur i n t h e qualitative behavior of t y p i c a l o r b i t s , but, t o
knowledge, a detailed analysis has
not yet been made. It may be interesting t o note t h a t the general structure of t h e Lorenz a t t r a c t o r
- t h e fact that
it is made frum two-dimensional unstable
manifolds of a p a i r of stationary solutions folded back on themselves i n f i n i t e l y often space.
A
-
does not depend on t h e dimensionality of t h e s t a t e
similar a t t r a c t o r can e a s i l y be constructed in a space of
an a r b i t r a r y number of dimensions, and s t i l l consists l o c a l l y of an uncountable family of two-dimensional sheets, stacked up i n a Cantor-setl i k e way.
It is thue a t l e a s t possible t h a t analogues of t h e Lorenz
a t t r a c t o r e x i s t f o r r e a l i s t i c approximations t o the equations of hydroQnamics (or even f o r these equations themselves).
Chapter 111. Ergodic Theory of Dissipative Systems
Let us now t r y t o see what physical conclusions could be drawn i f we knew t h a t t h e f u l l convection equations
- o r some f i n i t e
dimensional approximation t o them which i s s u f f i c i e n t l y detailed t o give an accurate description of t h e physical phenomena behavior similar t o t h a t of t h e Lorenz system.
-- had
Thus, consider a
system of equations with an a t t r a c t o r on which t h e motion depends i n a s e n s i t i v e way on i n i t i a l conditions and whose basin of a t t r a c t ion contains some physically relevant i n i t i a l s t a t e s .
If t h e system
i s s t a r t e d out i n t h e basin of a t t r a c t i o n , its s t a t e a t l a r g e positive times i s not arbitrary:
one can a t l e a s t predict with confidence
t h a t it w i l l be very near t o t h e a t t r a c t o r , which w i l l normally occupy a small f r a c t i o n of t h e whole basin of a t t r a c t i o n .
On t h e
other hand, because of t h e i n s t a b i l i t y of t h e motion on t h e a t t r a c t o r i t s e l f , we cannot reasonably hope t o be able t o make accurate pred i c t i o n s about where near t h e a t t r a c t o r t h e system w i l l be found. In other words, t h e s t a t e a t l a r g e positive times i s somewhat r e s t r i c t e d a s it must be near t h e a t t r a c t o r but otherwise appears t o be "random," i. e.
, not
t o depend i n a predictable way on t h e i n i t i a l s t a t e .
A s a p r a c t i c a l matter, t h e main objective of t h e theory of
convection i s t h e computation of such quantities a s t h e thermal conductivity of t h e convective layer, and these quantities are supposed t o depend on t h e physical parameters of t h e system (viscos-
i t y , temperature gradient, etc. ) but not on the i n i t i a l state.
At
f i r s t glance it appears t h a t these computations are impossible in principle i f t h e asyntptotic behavior i s determined by something l i k e t h e Lorenz attractor;
The instantaneoue r a t e of heat t r a n s f e r
can be expected t o depend both on the time and on t h e initial s t a t e and i s not l i k e l y t o approach a limiting value a t t
goes t o
On closer examination, however, t h e s i t u a t i o n i s not as
inflnity.
bad aa it seems.
What i s usually required f o r applications i s not,
f o r example, t h e instantaneous r a t e of heat t r a n s f e r , but rather t h e average of t h i s quantity over a long period of time, and it is only t h e limiting value of t h i s time average which needs t o be independent of initial conditions.
This suggests t h a t it would be
useful t o have some s o r t of ergodic theorem f o r dissipative systems. We w i l l now o u t l i n e one possible version of such a theorem, motivated on t h e one hand by its intended applications and on t h e other hand by what has been proved in special cases. Let
be a flow, A
Tt
a t t r a c t ion
B.
an a t t r a c t o r f o r T~ with basin of
By an ergodic theorem f o r
(Tt ,A)
we mean a theorem
asserting t h e e d s t e n c e of t h e following objects:
a) A probability measure U1\ on A, t solution flow T b)
A subset
X
of
B,
invariant under t h e
of Lebesgue measure zero such t h a t :
For any continuous Anction
f
on B
and any x i n B but not in
This formulation has a number of related aspects; we o f f e r t h e following remarks t o c l a r i e what it i s intended t o mean.
The main
thing being asserted i s t h a t forward time averages of "general" functions on t h e basin of attraction exist and are independent of the i n i t i a l state.
Independence of i n i t i a l conditions cannot be
expected t o be t m e e n t i r e l y without qualification.
For example,
most non-trivial a t t r a c t o r s contain i n f i n i t e l y many unstable periodic o r b i t s ; the time average s t a r t i n g at a point exactly on one of these orbits w i l l simply be t h e average over t h e o r b i t , which w i l l not be a t all l i k e t h e time average f o r a typical i n i t i a l point.
We must
therefore be prepared t o throw out an exceptional s e t of i n i t i a l conditions
- i n our formulation. the s e t - which ought t o be X
negligible from t h e physical point of view.
We have taken as our
c r i t e r i o n of physical n e g l i g i b i l i t y t h a t t h e s e t of exceptianal points have Lebesgue measure zero.
Note t h a t Lebesgue measure i t s e l f t has r e l a t i v e l y l i t t l e connection w i t h t h e flow T , and i n p a r t i c u l a r
is
supposed t o be invariant under Tt ; it has rather been pulled
in a r t i f i c i a l l y t o provide an elementary way of s t a t i n g t h a t a certain
set is negligible.
This criterion f o r n e g l i g i b i l i t y has a number of
drawbacks
-- notably,
it applies only t o flows on finite-dimensional
manifolds cmd not t o t h e convection equations themselves
- and
t h e r e are indications t h a t it could be replaced by a sharper condition f o m d a t e d i n terms of Hausdorff dimension. Next:
We are considering time averages only f o r continuous
functions and not, say, f o r general bomded Borel fur~ctions. Some such r e s t r i c t i o n is necessary t o avoid t r i v i a l counterexamples arising from t h e fact t h a t t h e flow is non-recurrent on B \ A. if
A
For example,
consists simply of an attracting stationary solution, it is
easy t o construct a bounded Borel function f such t h a t
does not e x i s t f o r any x
i n t h e basin of a t t r a c t i o n other than t h e
stationary solution i t s e l f . Third:
In our formulation, the exceptional s e t i s taken t o be
independent of t h e flmction able t o a l l o w it t o vary with
whereas it might seem more reason-
f,
f.
It turns out, however, t o be no
more r e s t r i c t i v e t o require t h e existence of a single exceptional set.
To see t h i s , assume t h a t time averages exist and are indepen-
dent of i n i t i a l condition f o r each continuous m c t i o n , but allowthe exceptional s e t t o depend on the function.
Choose a countable
set of continuous functions whose restrictions t o A t h e space of a l l continuous functions on A; provided t h a t
i s compact.
Let
are dense in
t h i s w i l l be possible,
X be t h e union of the exceptional
s e t s f o r these couutably many functions;
X w i l l again be a set
of Lebesgue measure zero and it is easy t o see t h a t time averages exist and are independent of initial condition i n B \ X for all continuous functions Finally:
f.
Our formulation of a general ergodic theorem requires
t h a t time averages be obtained as mean values with respect t o a probability measure yA on A.
If
A i s compact, t h i s is automatic
once time averages are known t o exiat and t o be essentially indepen-
1c
dent of i n i t i a l condition. value of
lim T-
;
To see t h i s , l e t
dt ~ ( T ~ xfor ) almost
is defined f o r a l l functions
f
? denote the common x.
continuous on B,
he quantity
?
but i s i s easily
seen t h a t two functions which are equal on A have t h e same average, so f
I+
? can be regarded as a f b c t i o n a l defined on the space of
continuolzs functions on A.
This functional is lineas, positive,
and takes t h e constant function 1t o 1, and hence, by the Riesz Representation Theorem, has t h e form
for a uniquely determined probability measure yA on A.
I n s p i t e of
t h e fact that i t s existence is automatic, t h e measure y,,
i s interest-
ing and important since it ought t o be possible t o describe it i n t r i n s i c a l l y and hence t o give a procedure f o r computing time averages other than by applying t h e definition.
We have here a close analogy
t o t h e usual view of t h e r o l e of t h e microcanonical ensemble i n c l a s s i c a l s t a t i s t i c a l mechanics, and t h e measure
uA
may t h e r e
fore be viewed as an equilibrium ensemble f o r t h e dissipative system.
One important practical difference from c l a s s i c a l s t a t i s t i -
c a l mechanics should be noted:
The microcanonical ensemble for a
Hamiltonisn system can be written down directly i n terms of t h e Hamiltonian.
To construct pA, on t h e other hand, it i s necessary
f i r s t t o locate t h e a t t r a c t o r
A
and then t o analyze exhaustively
t h e behavior of t h e solution f l o w on and n e w
A.
So f a r , t h i s
process appears t o require detailed informat ion about t h e solutions t o the equations of motion, as opposed t o simply knowing the different ial equations themselves. To get a complete picture of the behavior of t y p i c a l solutions of a s e t of d i f f e r e n t i a l equations, we would want t o do something l i k e t h e following: a.
show t h a t , except f o r a s e t of Lebesgue measure zero, t h e
s t a t e space s p l i t s into t h e basins of a f i n i t e number of attractors.
b.
prove an ergodic theorem f o r each of these attractors.
The asymptotic properties of a solution curve will then depend on which basin of a t t r a c t i o n it l i e s in, but essentially all solution curves i n a given basin w i l l have the same s t a t i s t i c a l properties over long periods of time.
This program has been completely carried
out by Ruelle and Bowen [ l o ] , [l]f o r flows on compact manifolds which
setis* Smale's Axiom A.
Rather than describe t h e proof of t h e Ruelle-Bowen theorem, we will t r y t o i l l u s t r a t e t h e idea of t h e proof by showing how
it miefit be adapted t o prove an ergodic theorem for the Lorenz system.
This procedure has the advantage of concreteness and
r e l a t i v e simplicity; it has t h e disadvantage t h a t it i s not r e a l l y a proof of anything as : a.
t h e argument s t a r t s from some qualitative features of
t h e Lorenz a t t r a c t o r which are strongly suggested by numerical experiments but which are certainly not proved b.
even assming these qualitative properties t o hold, t h e
proof of an ergodic theorem for t h e Lorenz system involves some algebraic and analytic complexities not present i n t h e Axiom A case snd not yet completely overcome. What we w i l l therefore actually do is t o reduce t h e proof of t h e
ergodic theorem of t h e Lorenz a t t r a c t o r t o a question about a onedimensional transformation and then suggest how t o t r e a t the onedimensional problem by considering a model problem with a number of technical simplif icatf OM. The f i r s t s t e p i n our proposed proof of an ergodic theorem
t f o r t h e Lorenz a t t r a c t o r is a reduction from t h e solution f l o w T t o t h e Poincard map
4 discussed in the preceding chapter,
is, we assme we have an ergodic theorem f o r
t get one f o r T
.
For
x
That
4 and show how t o
in t h e plane Z = 27, l e t Z(E) denote
t h e time required for t h e solution curve through 1~ t o return t o
i t s f i r s t damcrossing of t h e plane; i f t h e solution curve never
,qyturns, we put
For any
~ ( 5= )w.
_q
in the basin of attraction
whose solution curve eventually makes a damcrossing at a point time averages s t a r t i n g a t
c,
have t h e same l i m i t as time averages
s t a r t i n g a t & s o we may a s well consider only time averages s t a r t i n g at points 5 = ( x , Y , ~ T ) Mere X*Y < 72.
of
9
< 0. i . e . , where
We w i l l exclude immediately 5's
on t h e stable manifold
; as the stable manifold is a s e t of measure zero, t h i s w i l l
not affect t h e proof of an ergodic theorem. function
Then, f o r any continuous
fl
lim T-
f
[
dt fl(T
t
lim
M-l
1 fCBd
,=o 3 = wm. iTN-1 lim
**
1 rCPx)
n=o
providhd bath t h e l i m i t in t h e numerator and l i m i t in t h e denominator exist.
If
f and T
were continuous, an ergodic theorem for
ip
would say that both numerrrtor and denominator exist f o r almost a l l 5 and a r e essentially independent of f o r t h e l i m i t on t h e l e f t . it approaches i n f i n i t y as
x;
t h e same would then follow
Unfortunately, r
x
approaches
(x) is not
continuous;
Z. To complete t h e reduction
properly therefore requires an approximation argument using some
s p e c i d properties of t h e equilibrium ensemble f o r 8.
This argu-
ment i s i n e s s e n t i a l t o t h e main o u t l i n e of t h e proof; we w i l l not
give it. Next, and purely t o simplify t h e exposition, we w i l l r e s t r i c t cons ideration t o those continuous functions t h e symmetry (X,Y,z)
+
(-x,-Y ,z)
.
8
.
invariant under
This permits us t o consider only
one of t h e two p a r t s of t h e a t t r a c t o r f o r 4, A
f
end t o replace
8 by
General continuous functions can be handled by a straightforward
extension of t h e argument.
For t h e remainder of t h i s chapter we w i l l
A
always consider
4 r a t h e r than
8,
and we w i l l drop t h e
We must next examine in d e t a i l t h e action of
A
.
8 on and near t h e
a t t r a c t o r Q. The picture we want t o develop i s t h a t some neighborhood of
Q aecomposes i n t o a one-parameter family of non-intersecting
a r c s running transverse t o t h e a t t r a c t o r .
The arcs a r e characterized
by t h e property t h a t each of them contracts t o a point under repeated
application of distance from +m
.
8, i.e.,
en%+
to
if
n 4+ x
+, +x
are i n t h e same a r c then t h e
goes t o zero rapidly as n
goes t o
Accordingly, we w i l l r e f e r t o them as contracting arcs. The existence of contracting arcs i s suggested by t h e fact t h a t
8 compresses strongly i n a direction transverse t o t h e a t t r a c t o r . To see i n more d e t a i l what is happening, l e t us look a t a point 3~ on o r near t h e a t t r a c t o r , and a l i n e segment
passing e i t h e r
above o r below 3f and rouefily p a r a l l e l t o t h e a t t r a c t o r :
Figure 1 Applying
4 mves both & and
and a l s o s l i d e s attractor.
much closer t o t h e a t t r a c t o r
a and 8 s l i g h t l y
Since a and
away from
& along t h e
move away from 2 in opposite
d i r e c t i o n s , t h e r e must be points l i k e y on separation between the attractor. of
@,
4& and
such that t h e
remains a t a s u b s t a n t i a l angle t o
cby
Because of t h e strength of t h e transverse compression
t h i s condition locates
.
f a i r l y precisely along
y
apply
4 again and require t h a t t h e separation between
@2y
remain transverse t o t h e a t t r a c t o r ; t h i s w i l l locate
m r e precisely.
Now
425 and y
even
Continuing in t h i s w a y we construct a sequence of
successive approximations which ultimately y i e l d s a single point on
any
5
anx
with t h e property t h a t t h e separation between
i s transverse t o t h e a t t r a c t o r f o r all positive
of t h i s transversal separation end t h e fact t h a t t h e transverse direction, t h e distance from f o r each n
any
to
o t h e r than
anz y,
compresses in
@
5 to
is
For a fixed &,
any ,
decreases exponentially with
n.
on t h e other hand, w i l l eventually
be drawn away from 3~ by t h e stretching action of
attrador.
and
Because
a small f r a c t i o n of t h e distance from @nz t o
so t h e distance from Any point of
n.
y
t h e point8
y
4
along t h e
on t h e various possible
nearby
longitudinal segments
a6 s t r i n g together t o form a one-
dimensiand s e t which, by construction, is contracted under t h e action of
O.
Thus: Each point
t o l i e on a contracting arc.
=
s u f f i c i e n t l y near t o Si ought
Contracting arcs are uniquely deter-
mined l o c a l l y , and two contracting arcs which i n t e r s e c t must be continuations of each other.
There i s no apparent reason why cm-
t r a d i n g arcs must be unreasonably short; it ought t o be possible t o continue each of them a t l e a s t across t h e AiLl thickness of t h e attractor.
W e thus a r r i v e at a picture l i k e t h e following, where
t h e predominantly v e r t i c a l segments represent contracting arcs.
Figure 2 This figure i l l u s t r a t e s an important feature of t h e decomposition i n t o contracting arcs.
If
W
t h e ends as i l l u s t r a t e d , then another contracting arc.
i s a contracting arc, cut o f f a t @W w i l l be part but not rill of
Frequently, t h e r e w i l l be a second contract-
ing a r c W'
,
that
shares a contracting a r c with
@W'
running acrms t h e opposite end of t h e a t t r a c t o r , such @W.
I n t h i s case, i f
&E W
and &@E W'
then t h e distance from
exponentially as n
x,
x1
anz
and
anz'
goes t o zero
goes t o i n f l n i t y in s p i t e of t h e fact t h a t
are not in t h e same contracting arc.
For
t o be in t h e
same contracting a r c a s 5 it i s necessary but
sufficient t o
haw
an& and
anz'
approach each other as n
goes t o i n f i n i t y .
x
It i s i n f a c t t o be expected t h a t , f o r a t y p i c a l point
a t t r a c t o r , t h e s e t of points
&'
near t h e
near t h e a t t r a c t o r such t h a t
will consist of an i n f i n i t e (but countable) union o f contracting
arcs end w i l l b e dense i n a neighborhood of t h e a t t r a c t o r . We can now formulate a precise s e t of assumptions about t h e existence and properties of contracting arcs :
Asstmrptim:
Existence and Absolute Continuity of the Contmcting
Foliation.
There i s a continuous decomposition o f a neighborhood
of the a t t m c t o r R into a me-parameter family o f smooth arm ( a m t ~ ~ ~ u tarcs) i n g with the foZZ&g
a.
pmperties:
( @ n t r a c t i v i t y ) There e x i s t constants C,
0 < A < 1
, such that i f 5 , x2
A
,
with
are i n the same contracting arc
It would nar be natural t o construct a co-ordinate system f o r
a neighborhood of
such thet the contracting arcs are l i n e s
Cl
where one of t h e co-oranates i r constant.
It turns out ultimately
t o be more convenient t o do only part of the reparametrization, i .e. t o construct only the co-ordinate which i s constant on contracting arcs or, equivalently, t o parametrize t h e s e t of contracting arcs. To do t h i s we draw in some convenient way a smooth arc Y
running
t h e f u l l length of the a t t r a c t o r (but not necessarily in t h e attractor) which crosses each contracting arc exactly once and a t a non-zero angle. We l e t onto Y
(we w i l l refer t o such an arc Y
a s a lonaitudinal
arc).
denote t h e projection of a neighborhood of the a t t r a c t o r along contracting arcs, i .e
t h e unique point of
y
., f o r each
3 n
on t h e same contracting arc as
Figure 3
(x) denotes r.
We add t o our l i s t of assumptions a regularity property for r : c.
(Absolute continuity) If yl
arc, then n mrrtricted t o yl f2.om
yl
i s amj other longitudinal
is a differentiabZe mapping
t o y wiUl a Eb'lder continuous d e r i m t i v e .
This does nut complete t h e statement of t h e assumption, but a t t h i s point we digress t o r e l a t e these conditions t o l m o ~ f a d s about Axis A attractors.
( I t perhaps needs t o be mentioned t h a t
@
does
not s a t i s f y Adom A; it i s neither one-one nor differentiable on
2.
Its apparently mild f a i l u r e t o f u l f i l l the conditions turns out t o
have far-reaching consequences; t h e Lorenz a t t r a c t o r has a much w r e i n t r i c a t e and delicate structure than i s possible for an Axiom A attractor. )
For a general Axiom A a t t r a c t o r , each point of t h e
a t t r a c t o r has a neighborhood ( i n t h e manifold of s t a t e s ) which s p l i t s s continwusly i n t o an (n-n )-parameter family of smooth submaDifolds of dimension ns,
called contracting leaves and analogous t o the
contracting arcs of t h e above discussicn.
Each contracting leaf
shrinks exponentially under repeated applications of t h e transformat i o n ; moreover, wherever it passes through t h e a t t r a c t o r it is appearing in t h e s are two n-n dimensional sur-
tangent t o t h e infinitesimal stable space 'E statement of Axiom A.
If
Y and yl
faces each running transverse t o the contracting leaves, then proj e c t i m d o n g contracting leaves defines a continuous mapping from y1
t o y.
One of t h e unpleasant technical features of t h i s subject
is t h e fact t h a t , even i f t h e transformation i t s e l f is i n f i n i t e l y
differentiable, t h i s projection does not need t o b e continuously differentiable.
It i s , however, s u f f i c i e n t l y well behaved t o
send (n-ns)-dimensional Lebesgue measure on yl
t o t h e product of
Lebesgue measure on y with a ~ b ' l d e rcontinuous density.
This
property i s called absolute c o n t i n u i t ~of t h e contractina f o l i a t i o n ; in t h e special case where n s = n-1
( s o y , yl
are one-dimensional)
absolute continuity implies continuous d i f f e r e n t i a b i l i t y . We return now t o t h e problem of proving en ergodic theorem f o r @
.
The next s t e p in t h e argument i s t o show t h a t it suffices t o
prove t h e existence end e s s e n t i a l independence of i n i t i a l point f o r f'unct ions
f
which are continuous g&
const ant
contracting arcs.
To see t h i s , we assume t h e ergodic theorem f o r such functions and prove it for general continuous functions. continuous on a neighborhood
V
Thus, l e t
Then choose m
be uniformly
of t h e a t t r a c t o r which s p l i t s , as
i n t h e above assumption, i n t o contracting axcs. find 6 > 0 suchthat
f
d(x1,x2)<6
Choose E > 0,
and
implies If(xl) - f ( x 2 ) 1 <€/2.
large enough so t h a t t h e image under
contracting a r c has diameter s n a l l e r than
6.
am
of aay
The variation of
foam over any contracting a r c is thus no l a r g e r than 4 2 , and from t h i s it follows e a s i l y t h a t t h e r e e x i s t s a continuous function
g,
constant on contracting arcs, which d i f f e r s from foam by no more than
€
everywhere on V.
By assumptim
,
e x i s t s and is equal t o a constant almost everywhere; c a l l t h e canstant
C
8'
Hence
-I nIjl= focnn ~
limsup 1
w
m a lim inf IV-~
-I nI-1 1 = ~
both d i f f e r from C by l e s s than E almost evergwhere. Changing 8 sunnnation indices end making some simple estimates shows t h a t
also d i f f e r from C by less than 8 E approach zero, we see that
lim
I-
I-1
I1
I n=O
E
almost. eveqvhere.
Letting
foen
exists and is constant almost everyvhere on V We need now consider only f'uncticas
f
as desired.
on V which are
co~~tinuous end constant on contracting arcs, a d the next s t e p is t o exploit t h e projection r along contracting srcs t o reduce t o a one-dimensional problem.
Choose a smooth longitudinal arc as in
part c. of our assuiupticm above, s a d l e t
of
f
t o y.
Constancy of
f
,., f
denote t h e r e s t r i c t i o n
along contracting arcs ~ a n tsh a t
Define a mapping
(i.e.
, first
arcs )
.
of y t o i t s e l f by
apply
a, then project back t o
By the invariance of
, the
of
$ on ?. We now claim t h a t , in order f o r
lim
n-
@
on
'
along contracting
t h e decomposition i n t o contracting arcs
i .e.
action of
y
f i s obtained by "lifting" t h e action
1
'
fo@n
n=O
t o exist and be constant almost everywhere on V,
it suffices t h a t
e x i s t and be constant almost everywhere on y with respect t o onedimensional Lebesgue measure.
Indeed, i f
everywhere except on a s e t
everywhere on V
-
X
of l i n e a r measure zero, then
except on a - 5 .
By t h e absolute continuity of
t h e decomposition i n t o contracting arcs,
a'%
t u d i n a l a r c i n a s e t of l i n e a r measure zero. Theorem it follows r e a d i l y t h a t
7 1 9
i n t e r s e c t s any longiFrom t h i s and Fubini's
is a s e t of two-dimensional
measure zero. We have t h w reduced t h e proof of an ergodic theorem f o r t h e proof of a corresponding statement f o r t h e mapping itself.
-Q
of
Q to
y to
By introducing a smooth parametrization f o r y we may trans-
port everything t o t h e unit interval.
For definiteness we t a k e t h e
psrametrization so t h a t zero corresponds t o t h e end of stationary solution
near t h e
C, and we -dl1denote by v t h e mapping of
t h e u n i t i n t e r v a l t o i t s e l f obtained from dinates i n t h i s way.
y
by introducing coor-
Since q, i s a mapping of t h e u n i t i n t e r v a l
t o i t s e l f , it can be represented by a graph, and i t s graph looks schematically l i k e
Figure 4
where
a
corresponds t o t h e point on y where it crosses
There i s a cusp a t
a;
otherwise,
is continuously differen-
cp
t i a b l e with a ~glder-continuous f i r s t derivative. t o be i n f i n i t e l y differentiable away from obtained by introducing coordinates f o r
@ =
@
n
since it i s and, although
@
cannot be expected t o
s t r e t c h e s i n t h e longitudinal direction.
fLongitu&nat expansiveness). The pmvrmetrization for
my be chosen in such a way that 9' [O,a)
TO@
We cannot expect
We now add a f i n a l assumption which makes precise
t h e i n t u i t i v e notion t h a t
d.
a,
.,
i s i n f i n i t e l y d i f f e r e n t i a b l e aww from C,
be very smooth.
C.
y
i s s t r i c t l y greater than one on
andstrictty less than minus one a (a,l].
Given t h i s assumption, and t h e q u a l i t a t i v e properties of described above, we would l i k e t o prove t h a t , f o r any continuous function
f
on
[0,1],
e x i s t s and is constant almost everywhere with respect t o Lebesgue measure on
[0,1].
The next s t e p in our chain of reductions i s t o
reduce t h i s statement t o one about invariant
* measures
for p
.
To
Recall t h a t a measure p i s s a i d t o be invariant under a measurable cp (which need not be invertible]--l~) = p ( ~ ) f o r all measurable sets E. mapping
see har t h e reduction goes, let us imagine f o r a moment t h a t Lebesgue measure was invariant under cp.
Then t h e existence of
almost everywhere would follow imm%liately from Birkhoff' s Pointwise Ergodic Theorem; i f Lebesgue measure were
furthermore
ergodic,
then constancy of t h i s l i m i t almost everyvhere would e l s o be automatic. Although Lebesgue measure is not invariant under cp,
t h e same argu-
m n t could be applied i f we knev t h a t there existed some invariant measure equivalent t o ( i . e . , with the same s e t s of measure zero as) Lebesgue measure.
Thus:
Granting the assumptions made t o justify
t h e above reductions, proving an ergodic theorem f o r t h e Lorenz a t t r a c t o r reduces t o proving:
If 9 is a mapping o f
[0,1]
to i t s e l f with the quatitative
featurea descTibed abom, then there i s a probabitity measure 11
squimtent t o Lebesgw measure on [ O ,I] which i s invariant
and sxyodic w i t h respect to 9.
Proof of t h i s statement i s complicated by some unavoidable but secondazy features of t h e mapping (9.
We w i l l therefore present
t h e proof f o r a simpler class of mappings which display i n a m r e transparent way t h e techniques used f o r investigating existence and
83 uniqueness of absolutely continuous invariant measure eqe t h e relation between such measures end Gibbs s t ates for me-dimnsional l a t t i c e systems.
The mappings we w i l l consider are again continuous
mappings 9 of t h e unit interval t o i$gelf witb t h e folJaring properties : a ) For some a E (0,1), decreasing on ( a , l )
9 is monotonic
(0,a),
increasing
.
b ) v ( a ) = 1, ~ ( 1 =) 0 c)
cp is differentiable on
[0,a)
and on ( a , l ] ; i t s derivative
i s ~ E l & rcontinuo= on each of these intervals and approaches a f i n i t e l i m i t when a is approached e i t h e r from the l e f t or f r o m t h e right. d)
For some a < 1
for a l l x
Thus, t h e graph of cp looks like:
Figure 5
(except
a)
Our model 9
differs
model in two respects. cusp at
a;
from the 9
derived from the Lorenz
The more important i s the absence of a
the second i s t h e requirement e ) t h a t q ( 0 ) = a.
At
t h e expense of only s l i g h t complications, the l a t t e r condition could be replaced by the more r e a l i s t i c requirement t h a t , f o r some iO, we have
For t h e Lorenz system, on t h e other hand, there is no reason t o i
believe t h a t any q (0)
i s exactly equal t o
a,
and t h i s f a d
introduces an e x t r a layer of complexity.
9 satisfying
From nov on, then, we consider a mapping
a)
- e).
A s an a i d t o analyzing such a mapping, we introduce a coding of
[0,1] into a set of (one-sided i n f i n i t e ) sequences of zeros and ones. I n t u i t i v e l y , w e want t o associate with each (i0,il,i2,
x
a sequence
...Iwhere
i, =
0
if
q
k
(*I > a
and
\
k The prescription i s ambiguous i f Q (x) = a.
= 1 i f qk(x)
< a.
It would be easy t o
lift t h i s ambiguity by making one of t h e inequalities s t r i c t , but
f o r our purposes it i s b e t t e r t o allow such x ' s t o have more than one coding.
The inverse correspondence
- from sequences t o
x's
-
w i l l turn out t o be unambiguously defined and almost, but not
quite, one-one.
so
We s e t
,...) i s a c o m g of
(iO
x
it qk(x) E ~ ( i ~ f o r) k = 0,1,2,.
It follows readily from t h e assumed properties of
only element
.. .
0 i s the 3 x of A(1) with ~ ( x E) A(1) and, since Q (0) = 1,
t h a t there is no x with
x E ~ ( 1 ) .Q (x) E
CP
that
~ ( 1 1 q, ~ * ( x ) €
~(1).
Thus, no coding sequence has three successive ones and only x's such t h a t cpk ( x ) = 0 ones.
f o r some k
admit codings with two successive
I n t h e l a t t e r case, replacing t h e block ( 1 , ~ )wherever it
appears i n t h e coding sequence of t h e block (1,0) gives another coding of t h e same point, so:
Every point of
[O,I] &its
a coding without successive ones.
We w i l l say t h a t a f i n i t e or i n f i n i t e sequence of 0's and 1 ' s is admissible i f it contains no p a i r of successive l l s , and we w i l l let
denote t h e s e t of all admissible sequences.
In what foZla06
we c m i d e r only codings into acbnissible sequences. The next s t e p i n t h e analysis i s t o show t h a t every admissible sequence i s a coding of exactly one x. any f i n i t e admissible sequence (iO,il,.
To do t h i s , we define, f o r
..,ik),
.
A
(i.e.,
A(iO,il,..
begin with
i 1= : k
.,ih)
1
A
is t h e s e t of
for
x's
j = O,~,.+.,JS~
admitting codings which
( iO , . . . , ik).)
..,\I,
w m i t i a . For any finite acbnissibte sequence
...,ik)
A ( iO,
thun ak
Proof. for
is a non-enrpty closed intervat of length no larger
(where a-l
is the expansion o m t a n t of conhition dl
We argue by induction on
k = 0.
( t o ,il,.
k.
The statement i s c l e a r l y t r u e
For k > 0 we h a w by definition:
From t h e assumed properties of rp it follows t h a t continuously d i f f e r e n t i a b l e , snd expanding on cp(A(1)) = A(0); cp(A(0)) = [0,11.
...,
Thrs, since
q(Ai ) 3 A ( i l ) 3 A ( i l , i k ) s o q maps 0 A(il,. ,ik). Bp the induction hypothesis
..
A(iO), and t h a t (io,il)
A(iO,
...,\)
A(il,.
..,\)
non-empty closed i n t e r v a l , s o t h e same i s t r u e of
since
q ' > a
on A(io),
cp i s one-one,
# (l,l),
A(iO,.
is a
..,ik 1. .
where
X denotes length of an interval (or Lebesgue measure of
a more general s e t ) .
Again using t h e inductim hypothesis
as desired.
..
I f , now,
= ( i0,il,i2,. )
i s an admissible sequence,
is a decreasing sequence of closed intervals with length going
t o zero and hence t h e r e i s exactly one point x i n every A ( i O ,il,.
admits
..,ik).
i
We w i l l note t h i s
by n ( i ) ; it evidently
as a coding and is the only point which does.
evem admissible sequence point
x
Thus,
& is a coding of a uniquely defined
The mapping n is continuolrs from
n(&) i n [0,1].
A,
equigped with t h e to polo^ it i n h e r i t s as a subset of t h e compact product space
(0,lF
,
t o [0,1].
Although r i s not one-one, it
i s easy t o see t h a t it is a t most two-to-one and t h a t there a r e only
88 countably many x ' s with more than one admissible coding. a
denote t h e s h i f t mapping on
A,
is an admissible coding of
then i f
We l e t
02
x,
is an d i s s i b l e
coding of cp(x), i . e .
If a were exactly one-one
, it
between cp and t h e s h i f t mapping case,
a
0.
would s e t up an isomorphism Although t h i s i s not t h e
i s close enough t o a t r u e isomorphism f o r many purposes,
and t h i s i s t r u e i n p a r t i c u l a r f o r t h e analysis o f continuous* measures on [0,1]. Any probability measure under
rr t o a probability measure
The inverse operation
-
Tp
-- l i f t i n g measures
; on d
projects
on [0,1] defined by
from 10 ,1] t o
A-- i s
not q u i t e s o simple, but t o any continuous probability measure
u
on [0,1] t h e r e corresponds a unique continuous probability measurq Y
A measure i s continuous i f it assigns measure zero t o any s e t consisting of a s i n g l e point.
; on A
with
construct
TO
G = p .
;,
.. ,..., ik
we l e t
b(i0
denoke the ey&.der s e t
c
we define
'
on cylinder s e t s by
--
u(A(io,.
..,ikN
.
= u(A(io,*. , i k h
;t o
and we use standard measure theoly t o extend probability measure on
he
A,.,
a Bore1
construction does not work f o r s
completely general measure p because
i; as defined above will
,...,$)
not be f i n i t e l y additive i f t h e boundaries of t h e A(iO have non-zero
p-measure.
For continuous measures there i s no
problem since t h e boundary of points.)
Projection by
T,
A(iO,.
..,\)
contains only two
then, s e t s up an isomorphism between
t h e s e t s of continuous probability measures on
and on [0,1].
This isomorphism i s e a s i l y seen t o preserve most i n t e r e s t i n g r e l a t i o n s , e.g. and
1G2
only i f
ni;
c1
axe, and
and
"
u2
a r e equivalent i f and only i f
i s invariant (ergodic) under
61
a i f and
i s invariant (ergodic) under (9.
We can therefore adopt the following strategy f o r constructing a 9-invariant meesure equivalent t o Lebesgue meesure
A:
i ) L i f t Lebesgue measure t o a measure ii) Construct a s h i f t invariant measure
ergodic and equivalent t o iii)
Project on
under
-X
on j6
-. p on
.A
which i s
n t o obtain t h e desired measure p
[O,lI.
This strategy involves a trade-off.
It replaces t h e possibly compli-
cated mapping cp by t h e simple and standard s h i f t mapping a ,
but
it also replaces the simple and standard Lebesgue measure by t h e l e s s
.
simple measure 1 on
A.
The u t i l i t y of the trade-off depends on
whether o r not we can find methods t o control the behavior of key t o obtaining such control is t h e observation t h a t
x.
The
-,
X i s the
Gibbs s t a t e f o r a one-dimensional semi-infinite classical l a t t i c e system with a rapidly-decreasing s h i f t invariant many-body potential. b y semi-infinite we mean t h a t t h e l a t t i c e s i t e s m e labelled by t h e
non-negative integers rather than by all t h e integers.) configuration space i s
A
with no two consecutive 1's
- t h e s e t of sequences of 0's
- the l a t t i c e system
Since t h e and 1 ' s
w i l l have a nearest
n e i a b o r exclusion, but we w i l l argue shortly t h a t t h e potential i s othervise f i n i t e
.
Before Justiiying t h e claim t h a t f i r s t describe why t h i s fact i s useful.
is a Gibbs s t a t e , l e t us
Standard theorems about
one-dimensi onal classical st a t i s t i c a l mechanics can be applied t o show that t h e same interaction f o r t h e two-sided i n f i n i t e l a t t i c e
'
system has a unique Gibbs s t a t e and t h a t t h i s Gibbs s t a t e is trans-
lation (i.e.,
s h i f t ) invariant with very good ergodic properties.
(See Ruelle [9] for uniqueness, Gallavotti [ 2 ] f o r ergodic pro-
; be the measure on
perties. ) Let
d obtained from the invariant
Gibbs s t a t e by ignoring the part of t h e l a t t i c e system t o t h e l e f t of the origin (i.e., by projection).
Using the fact that the inter-
action between t h e part of the l a t t i c e system t o the l e f t of the origin and the part t o t h e right is bounded except for the e f f e c t s of the nearest neighbor exclusion, it i s easy t o show that absolutely continuous with respect t o
-1
$ is
with a Radon-Nikodp
derivative which i s both bomded above and bounded away from zero.
-1
Thus, once
has been identified as a Gibbs s t a t e , t h e standard
theory of Gibbs s t a t e s yields almost inmediately the existence and ergodicity of a shift-invariant measure
equivalent t o
and
hence of a cp-invariant measure equivalent t o Lebesgue measure.
X i s a Gibbs s t a t e , let us
To see why
..
sequence il ,i2,.
.
respect t o
A
.. .
il,i2,.
fix
an admissible
and compute t h e conditional probabilities with
of the two possible values
- 0 and 1 - of
i0 given
(TO complete t h e identification, we vill need t o
compute, more generally, conditional probabilities of the various values of
is trivial;
..
iO,. ,ik i0
given
has t o be
otherwise not be allwed.
%+l,.. . .)
If
i1 = 1 the computation
..
0 since t h e sequence iO,il,. would
We assume therefore t h a t
t h i s case the conditional probability is equal t o
il = 0.
In
ore
precisely: General theory assures us t h a t t h i s l i m i t e x i s t s
f o r almost all
(il,i2,.
..)
and i s equal almost everywhere t o the
desired conditional probability.
l i m i t e x i s t s f o r all
am
.
.
(ii,i2,..)
end give a formula for i t . )
i s obtained by transporting Lebesgue measure,
Because
Recall t h a t
We w i l l i n fact shm that t h e
A(iO,.
..,im )
is sn interval of length no greater than
Moreover, for each of t h e two possible values of
,...,im )
A(iO
write
xi
onto A(il
f o r n ( i O,il..
0 on all of
,...,i m ) .
h(iO,il,.
..,in)
..) then
If
m
9'
maps
i s large, and i f we
i s nearly equal t o
cf' (xi )
0 and we have
The approximation becomes exact as m +
Thus i f we define
iO,cp
w
so
we find t h a t t h e conditional probability of
i0 given
,...
il ,i2
is equal t o
Entirely similar arguments shcw t h a t t h e conditicaal probability of
..,\ given
iO,.
..
ik+l..
is equal t o
We can now construct t h e interaction f o r which state.
The i n t u i t i v e iaea is t h a t
h ( i O,il,.
-
X
i s a Gibbs
..) should represent
the contribution of t h e l a t t i c e s i t e zero t o t h e total energy.
This
i s not a well-defined concept, however, s o t h e r e will be some choices
t o be made in t h e construction of t h e i n t e r a c t i m .
We w i l l think of
our l a t t i c e system as a spin system r a t h e r than a l a t t i c e gas.
The
interaction i s then specified by giving, f o r each f i n i t e subset
X
of
Z,
a function
ax
defined on
{0,1)
X
,
forward translation-invarience requirements;
sub3ect t o a straight-
$
i s interpreted a s
t h e p o t e n t i a l energy due t o direct interaction among a l l t h e l a t t i c e s i t e s in X.
The t o t a l energy f o r a configuratian
I
defined on a f i n i t e s u b l a t t i c e A i s
We construct an interaction by defining: unless
X i s an interval (i.e., a s e t of t h e form
.
(il 7il+17. ,il+k)
90,1,, ,,7k)(io7"'7ik)
= $(io3. *.7ik)
- 5-l(i0,"'7ik-l)
where ~
(o,...,\ i
inf
f
7
t
\+29."
h(iO,-.. ,ik7$+17$+2s*..),k
= 0,1,.-.
=-1 = O
The function
ex
f o r an i n t e r v a l of length
zero is determined by t r a n s l a t i o n invariance:
k+l not s t a r t i n g a t
With these definitions it i s easy t o see t h a t
Taking the l i m i t of t h e second of these equations as m
+
with
k
held fixed gives
Comparing t h i s equation with the previously obtained formula f o r conditional probabilities relative t o
1 shows that
i s indeed
a Gibbs s t a t e f o r a semi-infinite l a t t i c e system with the interaction we have ccastructed. To apply standard results from s t a t i s t i c a l mechanics t o show t h a t t h i s interaction has a unique Gibbs s t a t e , we need t o know t h a t t h e interaction drops off rapidly at large separatians. that
It turns out
goes t o zero exponentially a s
k goes t o i n f i n i t y .
To see t h i s ,
observe:
I
.
..,\I1
, ,k) ( iO,.
minimum of loglq'(x)l
i s no l a r g e r than t h e maximum minus
..,\I A(iO,. ..,%)
on &(i0,.
The length of
k i s no l a r g e r t h a n a
is Hslder continuous on A(0) and on
C?'
A(1).
To conclude, l e t us s w e y b r i e f l y how t h e above development uould have t o be modified t o apply t o t h e Lorenz system.
The Lorenz
then decreasing from a
to
Q
which comes from t h e
cp i s again increasing t o some point
a;
1, and we have
Formally, we can approach t h i s mapping i n t h e same way as our model q : We code each point
x
of
[ O,1]
i n t o a sequence of ones and zeros
determined by whether t h e successive
ri&t of
a,
9k (XI'S are
t o the l e f t or the
and thus t r a n s l a t e t h e problen i n t o a s t a t i s t i c a l -
mechanical me.
The t e c h n i c a l complications are two-fold.
F i r s t of
all, t h e image of t h e coding is no longer as simple as before.
It
i s possible t o have more than two successive ones, but because
cp
moves points of
[O ,a] non-trivially t o t h e ri&t it is not possible
(The maximnu number is actually 25 for k A straightforward analysis shms t h a t , unless" q, ( 0 ) = a
t o have a r b i t r a r i l y many.
r = 28).
f o r some k ,
t h e image of t h e coding cannot be described by specify-
ing a f i n i t e number of excluded f i n i t e sequences.
In s t a t i s t i c a l
mechanical terms, t h i s means t h a t the corresponding classical l a t t i c e system has i n f i n i t e l y many "exclusions," of a r b i t r a r i l y long range, generalizing the nearest-nei&bor exclusion of t h e model
Q.
A
second difficulty is caused by the i n f i n i t e tangent t o the graph of 'p a t
a which means t h a t t h e contribution h
of a single l a t t i c e
s i t e t o the t o t a l energy i s not bounded above.
These features make
t h e s t a t i s t i c a l mechanical problem considerably mare d i f f i c u l t then t h e me we have considered.
References
[l] R. Bowen and D. Ruelle, The ergodic theory of Axiom A flows, Inventiones Math. 181-202 0 9 7 5 1.
a
[2] G. Gallavotti, Ising model and Bernoulli schemes in one dimension, Commun. Math. Phys. 2 (19731, 183-190.
[3]
J. Guckenheimer, A strange strange attractor, in [6], pp. 368-381.
[4] J. Leray, Sur le mowement dlun liquide visqueax emplissant llespace, Acta Math. 63 (1934)~193-248. [5] E. N. Lorenz, Deterministic nonperiodic flow, J. Atmos. Sci. (19631, 130-141. [6] J. E. Marsden and M. McCracken, The Hopf Bifurcation and its Applied Mathematical Sciences 19, Springer-
[TI R. M.
May, Simple mathematical models with very complicated dynamics, Nature 261 (19761, 459-467.
[8] J. B. McLaughlin and P. C. Martin, Transition to turbulence of a statically stressed fluid system, Phys. Rev. A 12 (19751, 186-203. [9] D. Ruelle, Statistical mechanics of a one-dimensional lattice gas, Commun. ~ath.Phys. 2 (1968), 267-278. [lo] D. Ruelle, A measure associated with Axiom A attractors, Amer. Sour. Math., to appear. [ll] S. Smale, Differentiable dynamical systems, Bull. Amer. Math. Soc 12 (19671, 747-817.
.
[12] R. F. Williams, The structure of Lorenz attractors, Preprint, Northwestern University (1976).
CEN TRO INTEF@lAZIONALE MATPlATICO ESTIVO (c. I.M.E.)
MANY PARTICLE COULOMB SYSTEMS
E l l i o t t H. LIEB
D e p a r t m e n t s of M a t h e m a t i c s and P h y s i c s Princeton University
- Princeton,
N.J.
08540
C o r s o tenuto a B r e s s a n o n e dal 21 giugno a1 24 giu@o 1976
MANY PARTICLE COULOMB SYSTEMS
E l l i o t t H. Lieb
*
Departments of Mathematics and Physics Princeton University Princeton, N .J. 08540
Lectures presented a t the 1976 session on s t a t i s t i c a l mechanics of the International Mathematical Summer Center (C.I.M.E.)
Bressanone, Italy,
June 21-27.
* Work
partially supported by U.S. National Science Foundation grant
MCS 75-21684.
With t h e introduction of t h e Schroedinger equation i n 1926 i t became possible t o resolve one of t h e fundamental paradoxes of t h e atomic theory of matter (which i t s e l f had only become universally accepted a few decades earlier):
Why do the electrons not f a l l i n t o the nucleus?(Jeans, 1915).
Following t h i s success, more complicated questions posed themselves.
Why
is the lowest energy of bulk matter extensive, i.e. why i s i t proportional
t o N, the number of p a r t i c l e s , instead of t o some higher power of N? Next, why do the ordinary laws of thermodynamics hold?
Why, i n s p i t e of
the long range Coulomb force, can a block of matter be broken i n t o two pieces which, a f t e r a microscopic separation, a r e independent of each other? The aim of these l e c t u r e s is t o answer the above questions i n a simple and coherent way.
It is a summary of research I have been engaged
i n f o r the p a s t few years, and i t has been my good fortune t o have had the b e n e f i t of collaboration with J.L. Thirring.
Lebowitz, B. Simon and W.E.
Without t h e i r i n s i g h t s and stimulation probably none of t h i s
could have been carried t o f ~ i t i o n . The accompanying flow c h a r t might be helpful.
I n s e c t i o n I atoms
a r e shown t o be s t a b l e because of the Sobolev inequality, not the Heisenberg uncertainty principle.
A new inequality r e l a t e d t o Sobolev's
2 3 N f o r functions i n the antisymmetric tensor product L (1 ) i s presented i n s e c t i o n 11. Thomas-Fed theory (which was introduced i n 1927 j u s t a f t e r the Schroedinger equation) i s analyzed i n s e c t i o n 111. This subject i s i n t e r e s t i n g f o r three reasons:
( i ) As an application of
nonlinear functional analysis; ( i i ) It turns out t h a t i t agrees asymptotically with the Schroedinger equation i n a l i m i t i n which the
nuclear charges go t o i n f i n i t y ; ( i i i ) The no-binding theorem of ThomasF e d theory, when combined with t h e i n e q u a l i t y of s e c t i o n 11, y i e l d s a simple proof of the s t a b i l i t y of matter. IV.
The l a t t e r i s given i n s e c t i o n
The f i r s t proof of s t a b i l i t y is due t o Dyson and Lenard i n 1967, but
the proof i n s e c t i o n I V i s much simpler. dynamic problem.
Section V deals with t h e thermo-
The d i f f i c u l t y here i s not the one of collapse, which
was s e t t l e d i n s e c t i o n I V , b u t t h e p o s s i b i l i t y of explosion caused by the long range p a r t of the Coulomb p o t e n t i a l .
Newton's theorem t h a t a charged
sphere behaves from the outside as though a l l i t s charge were concentrated a t the center, together with some geometric f a c t s about the packing of b a l l s , i s used t o tame the l / r p o t e n t i a l .
Section VI on Hartree-Fock
theory i s r e a l l y outside t h e c e n t r a l theme, b u t i t has been added a s a f u r t h e r exercise i n functional analysis and because i t is, a f t e r a l l , the most common approximation scheme t o solve t h e Schroedinger equation. Chapters I1 and I V come from (Lieb-Thirring, 19751, Chapter I11 from (Lieb-Simon, 1976), Chapter V from (Lieb-Lebatitz,
1972) and Chapter V I
from (Lieb-Simon, 1973). A n attempt w a s made t o present the main ideas i n as simple and
readable a form as possible, and therefore t o amit many technical d e t a i l s . There were two reasons f o r t h i s .
The f i r s t was t o t r y t o make t h e
l e c t u r e s accessible t o p h y s i c i s t s as w e l l as t o mathematicians.
This a l s o
c r e a t e s n o t a t i o n a l and semantic problems which, i t is hoped, have been a t l e a s t p a r t i a l l y resolved.
With t h i s aim i n mind, I hope t h e i n c l u s i o n of
such things as an explanation of young's i n e q u a l i t y w i l l be excused.
The
second reason stems from the b e l i e f t h a t i f enough h i n t s of a proof a r e given then a competent analyst would as soon supply the d e t a i l s f o r
himself as read about them. The bibliography i s not scholarly, but I believe no theorem has been quoted without proper credit.
I am most grateful t o S.B.
Treiman who generously devoted much time
to reading the manuscript and who made many valuable suggestions t o improve i t s c l a r i t y .
. V.
Newton's Theorem, Screening, and the Geometry of Ball Packing
I
d
1 I. Atomic Stability and the Uncertainty
' 11.
.--.-
Principle
Uncertainty Principle f o r
IV. H-Stability
+'
of Matter
V.
4.
N Fermions *
Limit and Stability
L
C
Fl VI.
The numbers indicate sections.
Thermodynamic
I I
Hartree-Pock
I.
The S t a b i l i t y of Atoms
By t h e phrase " s t a b i l i t y of an atom" is meant t h a t t h e ground s t a t e energy of an atom is f i n i t e .
This i s a wealrer notion than the concept of
8 - a t a b i l i t y of matter, t o be discussed i n Section I V , which means t h a t the ground state energy of a many-body system i s not merely bounded below but is a l s o bounded by a constant times t h e number of p a r t i c l e s .
This,
i n turn, is d i f f e r e n t from thermodynamic s t a b i l i t y discussed i n Section
v. Coneider t h e Hamiltonian f o r t h e hydrogenic atom:
(using u n i t s i n which
62 12
2 3 = 1, m = 1 and [ e l = 1) E a c t s on L (1 ), the
square i n t e g r a b l e functions on 3-space.
Why i s t h e ground s t a t e energy
f i n i t e , i.e. why is
f o r some E
0
>
-m?
The obvious elementary quantum mechanics textbook
answer is t h e Heisenberg uncertainty p r i n c i p l e (Heisenberg, 1927) : I f the k i n e t i c energy is defined by
T,,, 5 $ l ~ $ ( x ) l ~ d x and i f
:'-en when
T <x2> > 914
4
4-
.
(4)
The i n t u i t i a n behind applying t h e Heisenberg uncertainty p r i n c i p l e
(4) t o t h e ground s t a t e problem (2) is t h a t i f the e l e c t r o n t r i e s t o g e t
within a distance R of the nucleus, the k i n e t i c energy TJ,i s a t l e a s t a s large as R-~.
Consequently <+,El$> 2 R - 2 - ~ / ~ , and t h i s has a minimum
-zL/4 f o r R = 212. The above argument is false!
The Heisenberg uncertainty p r i n c i p l e
says no such thing, despite the endless invocation of the argument. Consider a J, consisting of two p a r t s , J, =
J,
1
is a narrow wave
packet of radius R centered a t the o r i g i n with ~ 1 1 ~ 1 ~ - 1 / 2J,2 . is spherically symmetric and has support i n a narrow s h e l l of mean radius L
11$2 12-1/2. I f L i s 11x1-' lJ,(x)1 2dx - 1/2R.
and
l a r g e then, roughly, j x
IJ,(x) I 2dx
Thus, from ( 4 ) we can conclude
2 2 T$ > 9 / 2 ~ and hence t h a t <J,,HJ,>'9/2L -Z/2R. and using
2
- L212 whereas
only t h a t
With t h i s wave function,
only the Heisenberg uncertainty p r i n c i p l e , we can make Eo
a r b i t r a r i l y negative by l e t t i n g R
+
0.
A more c o l o r f u l way t o put t h e s i t u a t i o n i s t h i s :
an e l e c t r o n
cannot have both a sharply defined p o s i t i o n and momentum.
I f one is
w i l l i n g t o place the e l e c t r o n i n two widely separated packets, however, say here and on the moon, then the Heisenberg uncertainty p r i n c i p l e alone does not preclude each packet -
from having a sharp p o s i t i o n and
momentum. Thus, while (4) is correct i t is a p a l e r e f l e c t i o n of t h e power of the operator -A t o prevent collapse. (i.e.
A b e t t e r uncertainty p r i n c i p l e
a lower bound f o r t h e k i n e t i c energy i n terms of some i n t e g r a l of
$ which does not involve derivatives) is needed, one which r e f l e c t s more
accurately the f a c t t h a t i f one t r i e s t o compress a wave function qwhere then the k i n e t i c energy w i l l increase. -
This p r i n c i p l e was provided
by Sobolev (Sobolev, 1938) and f o r some unknown reason h i s inequality, which is simple and goes d i r e c t l y t o the h e a r t of t h e matter, has not
made its way i n t o the quantum mechanics textbooks where i t belongs. Sobolev's inequality i n three dimensions (unlike (4) i t s form is dimension dependent) i s TS
a
-~ ~ 1 ~ 1 ,
j ~ v S ( d l ~ d2xK ~ ~ $ P ( X ) ~ & I " ~
where
i a the density and
i s known t o be the best possible constant.
( 5 ) is non-linear i n p , but
t h a t i s unimportant. A rigorous derivation of (5) would take too long t o present but i t
can be made plausible as follows
(Rosen, 1971) : Ks i s the minimum of
Let us accept t h a t a minimizing JI e x i s t s ( t h i s is t h e hard p a r t ) and t h a t i t s a t i s f i e s t h e obvious v a r i a t i o n a l equation
with a > 0 .
Assrnning a l s o t h a t t h e r e is a minimizing $ which i s now
negative and spherically symnetric ( t h i s can be proved by a rearrangement inequality), one finds by inspection t h a t
-
2 -1/2 ~(x) (~/~I~/.~(I+I~I When t h i s is inserted i n t o the expression f o r K' K
-
t h e r e s u l t is
( 1 2 ) ~ ~Th . e minimizing $ is not square integrable, bur t h a t is-
of no concern.
Naw let us make a simple calculation t o show how good (5) r e a l l y is. For anp
J,
and hence ~ ( x LO, ) Ip
<$,a$> ,minih(p):
.
1)
The l a t t e r calculation is t r i v i a l (for any potential) since gradients a r e One finds t h a t the solution t o the v a r i a t i o n a l equation is
not involved.
~(x= ) a[1x1'1-~-1]112 R - K s n -413~-1.
f o r 1x1 (R and p(x) = 0 f o r 1x1
hen
-
h ( ~ ) ~ ~ ( n / 2 =) - ~ ( 4 1~3 )~Z bq (Recall t h a t one Rydberg
Ry = 114 i n these units.)
2
R, with
. Thus, (5) leads
e a s i l y t o the conclusion
and t h i s is an excellent lower bound t o the correct especially since no d i f f e r e n t i a l equation had t o be solved. I n anticipation of l a t e r developments a weaker, but a l s o useful, By ~ & l d e r ' sinequality*
form of (5) can be derived.
and, since w e always take
1 T,
*
-
2 Ks
1, fp (x)S13dx
.
Hiilder 's inequality s t a t e s t h a t
when p-l+q-l q
2
-
312.
= 1 and p
2 1. To obtain (10) take f
p, g
-
p2I3,
p = 3,
Note t h a t there is now an exponent 1 outside the i n t e g r a l .
Although Ks
is the best constant i n (5) i t is not the b e s t constant i n (11).
the l a t t e r Kl. /p(x)dx
Call
K1 is the minimum of / ~ v ) ( x )12dx/1p(3S13dx subject t o
= 1. This leads t o a non-linear Schroedinger equation whose
numerical solution y i e l d s (J. Barnes, private communication) K1
=
9.578
.
In any. event and hence
KC is much bigger than Ks;
i t is t h e c l a s s i c a l value, and w i l l be
encountered again i n section I1 and i n section 111 where its significance w i l l be c l a r i f i e d .
We can repeat the minimization calculation analogous t o (8) using the bound (12) and the functional hC(p) = xC/p ( x ~ ~ / ~ d 1x1-' x -P ~ / (xldx (We could, of course, use the b e t t e r constant K1.)
for
I xl
5 R.
.
This time
R is determined by /p = 1 and one f i n d s t h a t
R = (Kc/Z) (4/n2) 'I3 and
3 l I 3 i s only 8.2% greater than 413. The Sobolev inequality (5) o r its v a r i a n t (12) is, f o r our purposes, a much b e t t e r uncertainty p r i n c i p l e than Heisenberg's a l s o f a i r l y accurate.
- indeed i t is
W= nov want t o extend (12) t o the N-particle case
i n order to establish the s t a b i l i t y of bulk matter.
The important new
fact that w i l l be invoked is that the N particles are fermions; that is to say the N-particle wave function is an antisymmetric function of the N-space, spin variables.
11.
Extension of t h e Uncertainty Principle t o Many Fermions A w e l l known elementary calculation is t h a t of t h e lowest k i n e t i c
v , of
energy, T
N fermions i n a cubic box of volume V. For l a r g e N one
finds that
where p
-
N/V and q is t h e number of s p i n s t a t e s a v a i l a b l e t o each
p a r t i c l e (q-2 f o r electrons).
(15) is obtained by merely adding up the
N/q lowest eigenvaluea of -A with Dirichlet ($60) boundary conditions on the walls of the box. proportional t o N~~~ were not fermions.
The important feature of (15) is t h a t i t is instead of N, as would be the case i f the p a r t i c l e s
The e x t r a f a c t o r N~~~ is e s s e n t i a l f o r the s t a b i l i t y
of matter; i f electrons were bosons matter would not be s t a b l e . -213 , ought t o (15) suggests t h a t (12), with a f a c t o r q extend t o t h e N-particle case i f p(x) is interpreted properly.
The idea
i s old, going back t o Lenz (Lenz, 1932) who got i t from Thomas-Fermi theory.
The proof t h a t something l i k e (12) is not only an appro-tion
but i s a l s o a lower bound is new. To say t h a t the N p a r t i c l e s a r e fermione with q s p i n s t a t e s means t h a t the N-particle wave function JI (xl, xi€ 13 and oiE { 1,2,.
...,x , ;
U1,
...,uN) defined f o r
..,¶I is antisymmetric i n the p a i r s (xi,ui).
norm is given by
Define
t o be the usual k i n e t i c energy of $ and define
The
t o be the s i n g l e p a r t i c l e density, i . e . t h e probability of finding a p a r t i c l e a t x. Theorem 1.
The analogue of (12) is the following.
If <$,p= 1 then T$ 2 (4nq)-2/3 K' jpy (x)5/3dx
Apart from the annoying f a c t o r ( 4 7 ~ ) - ~z/ ~0.185,
..
(18) says t h a t t h e
i n t u i t i o n behind considering (15) as a lower bound is correct.
We b e l i e v e
t h a t ( 4 7 ~ ) - ~does / ~ not belong i n (18) and hope t o eliminate i t someday. Recent work (Lieb, 1976) has improved t h e constant by a f a c t o r ( 1 . 8 3 1 ~ ' ~= 1.496, so we a r e now off from t h e conjectured constant q'2/3~C only by t h e f a c t o r 0.277.
It
The proof of Theorem 1 i s not long but i t is s l i g h t l y tricky.
is necessary f i r s t t o i n v e s t i g a t e the negative eigenvalues of a onep a r t i c l e Schroedinger equation when t h e p o t e n t i a l i s non-positive. Theorem 2.
=V(x)
50
be a p o t e n t i a l f o r t h e one-particle, three
2 3 dimensional Schroedinger operator H = -A+V(x) ,n L (1 ) NE(V) be the number of eigenstates of H with energies N
2
-1/2
E
where If(x)l- = If(x)l
Lf
Corollary.
-< ... -< 0 a r e
any) then
If el 2 e2
f(x) ( 0
jlv(x)-e/rlf
& If(x)l-
.
For E 2 0 let
2 E. Then dx
= 0 otherwise.
the negative eigenvalues of H
(Lf
-I
B 1e j 1
.
(D
I n s e r t (19) and do t h e a i n t e g r a t i o n N-a (V)da 0 f i r s t and then the x i n t e g r a t i o n . The r e s u l t i s (20).
Proof.
fi
We believe the f a c t o r (41) does not belong i n (20). From t h e Schroedinger equation HJ, = eJ, i t i s easy
Proof of Theorem 2.
t o deduce t h a t NE(V) is equal t o the number of eigenvalues which a r e
21
of t h e p o s i t i v e d e f i n i t e Birman-Schwinger operator (Birman, 1961, Schwinger, 1961)
E s s e n t i a l l y (21) comes from t h e f a c t t h a t i f H$ = e$ then (-A-e)$ I f one defines
1 ~ 1 ~ ' 5~ ) (, then
when e is an eigenvalue. operator on L ing i n E. el ( e 2 (
2
(a
(.
Thus Be has an eigenvalue 1
However, BE i s a compact p o s i t i v e semi-definite
) f o r E < 0 and, as an operator, BE i s monotone decreas-
Thus, i f BE has k eigenvalues
... (ek
( E such t h a t Be
2 NU2(- I V-E/2
V(x)-~/2).
=
3
Consequently NE(V) NE (V)
Be(
= IVl $.
1 -)
2 1,
t h e r e e x i s t k numbers
has eigenvalue 1.
1
5 T r BE(V) 2
.
On t h e o t h e r hand
by t h e v a r i a t i o n a l p r i n c i p l e (draw a graph of
Thus, s i n c e BE(V) has a k e r n e l BE(x,y) = Iv(x)] 1/2
exp{-1~1~/~lx-y~1[4~~x-~~]-~~~(~)~~/~
(19) r e s u l t s from applying Young's inequality
*
t o (22).
Alternatively,
one can do the convolution i n t e g r a l by Fourier transforms and note t h a t the Fourier transform of the l a s t f a c t o r has a maximum a t p = 0 where i t
is 4 * ( 2 1 ~ 1 ) - ~ / ~ .
@
Using (20), which is a statement about t h e energy l e v e l s of a s i n g l e p a r t i c l e Hamiltonian we can, surprisingly, prove Theorem'l which r e f e r s t o t h e k i n e t i c energy of N fermions. Consider the non-
Proof of Theorem 1. J, and hence p$(x)are given. p o s i t i v e s i n g l e p a r t i c l e p o t e n t i a l V(x) r -UP
J,
(xl2I3 where a is given by
(213x1 q a3l2 = 1. Next consider the following N-particle Hamiltonian
2 3 qN on L (X ;O ) energy of
.
I f Eo is the fermion ground s t a t e
we have t h a t Eo
,
q Ze
j
, where
the e
eigenvalues of the s i n g l e p a r t i c l e Hamiltonian h.
j
a r e the negative (We merely f i l l the
lowest negative energy l e v e l s q times u n t i l t h e N p a r t i c l e s a r e accounted I f N > kq,
f o r ; i f t h e r e a r e k such l e v e l s and i f N < kq then Eo > q Ee
1'
the surplus p a r t i c l e s can be placed i n wave packets f a r away from the o r i g i n with a r b i t r a r i l y small k i n e t i c energy.) Eo
2
<J,,H#
On the other hand,
= T - a j ~(xl5I3dx by the v a r i a t i o n a l principle.
'4
J,
I f these
two i n e q u a l i t i e s a r e combined together with (20), which says t h a t
I e. 1
*
2 - ( 4 / 1 5 ~ ) a ~l/p~$ ( ~ ) ~ / ~ dthen x , (18) i s t h e r e s u l t .
I
Young's inequality s t a t e s t h a t
when p-l+q-l+r'l
= 2 and p,q,r
2 1. For (22) take p = ~ and 2 q=l.
It might not be too much o u t of place t o explain a t t h i s point why K'
is c a l l e d the c l a s s i c a l constant.
The name does not stem from its
a n t i q u i t y , a s i n the i d e a l gas k i n e t i c energy ( U ) , but r a t h e r from c l a s s i c a l mechanics
-- more p r e c i s e l y
quantum mechanics.
This i n t u i t i v e idea is valuable.
the semiclassical approximation t o
A s the proof of Theorem 1 shows, the constant i n (18) f o r T
J,
is
simply r e l a t e d t o t h e constant i n (20) f o r the sum of t h e eigenvalues. The point is t h a t the semiclassical approximation t o this sum i s
and t h i s , i n turn, would y i e l d (18) without t h e
factor.
The
semiclassical approximation i s obtained by saying t h a t a region of volume (2n13 i n the 6-dimensional phase space (p,x) can accommodate one eigenstate.
Hence, i n t e g r a t i n g over the s e t 0 (H)
, in
which ~ ( ,x) p = P~+v(x)
i s negative,
I f a coupling constant g i s introduced, and i f V is replaced by gV, then i t is a theorem t h a t the semiclassical approximation i s asymptotic a l l y exact as g
+
OD
f o r any
v
in L
~ (R~). / ~
Theorem 1 gives a lower bound t o t h e - k i n e t i c energy of fermions which i s c r u c i a l f o r the H-stability of matter a s developed i n Section
IV
.
To appreciate the significance of Theorem 1 it should be
compared with t h e one-particle Sobolev bound (12). outside some fixed domain, Q, of volume V.
Suppose t h a t p(x)
Then since
by Halder's inequality, one sees t h a t T grows a t l e a s t a s f a s t a s N513
'4
Using (12) alone,one would only be able t o conclude t h a t T
J,
This d i s t i n c t i o n stems from the Pauli principle, i.e. nature of the N-particle wave function.
0
.
grows as N.
the antisymmetric
As we s h a l l see, t h i s N~~~
growth
i s e s s e n t i a l f o r the s t a b i l i t y of matter because without i t the ground s t a t e energy of N p a r t i c l e s with Coulomb forces would grow a t l e a s t as f a s t a s -N7l5
instead of -N.
The Fermi pressure is needed t o prevent a collapse, but t o l e a r n how t o e x p l o i t i t we must f i r s t turn t o another chapter i n the theory of Coulomb systems, namely Thomas-Fed theory.
111.
Thomas-Fenni Theory The s t a t i s t i c a l theory of atoms and molecules was invented independ-
e n t l y by Thomas and F e d (Thomas, 1927, Fermi,1927).
For many years t h e
TF theory was regarded a s an uncertain approximation t o t h e N-particle Schroedinger equation and much e f f o r t was devoted t o t r y i n g t o determine i t s v a l i d i t y (e.g.
~ o m b i s , 1949)
.
It was eventually noticed numerically
(Sheldon, 1955) t h a t molecules d i d not appear t o bind i n t h i s theory, and then T e l l e r ( T e l l e r , 1962) proved t h i s t o be a g e n e r a l theorem. I t is now understood t h a t TF theory i s r e a l l y a l a r g e Z theory
(Lieb-Simon,
1976); t o b e p r e c i s e i t is exact i n the limit Z
+ m.
For
f i n i t e 2, TF theory i s q u a l i t a t i v e l y c o r r e c t i n t h a t i t adequately describes t h e bulk of an atom o r molecule. give binding.
It is n o t p r e c i s e enough t o
Indeed, i t should n o t do s o because binding i n TF theory
would imply t h a t the cores of atoms bind, and t h i s does not happen. Atomic binding is a f i n e quantum e f f e c t .
Nevertheless, TF theory
deserves t o be w e l l understood because i t i s e x a c t i n a l i m i t ; the TF theory i s t o the many e l e c t r o n system a s t h e hydrogen atom i s t o the few e l e c t r o n system.
For t h i s reason the main f e a t u r e s of t h e theory a r e
presented h e r e , mostly without proof. A second reason f o r our i n t e r e s t i n TF theory i s t h i s :
i n the
next s e c t i o n t h e problem of t h e H-stability of matter w i l l b e reduced t o a TF problem.
The knowledge t h a t TF theory is H-stable
(this is a
c o r o l l a r y of t h e no binding theorem) w i l l enable us t o conclude t h a t t h e t r u e quantum system is H-stable. The Hamiltonian f o r N e l e c t r o n s with k s t a t i c n u c l e i of charges zi > 0 and l o c a t i o n s Ri i s
where
and
The nuclear-nuclear repulsion U i s , of course, a e o n s f e ~ tterm i n
$ but
it is included f o r two reasons: (i)
W e wish t o consider the dependence on the Ri of EQ ({z ,R }C1) :the ground s t a t e energy of I$ N j j j
.
( i i ) Without U the energy w i l l not be bounded by N. The nuclear k i n e t i c energy i s not included i n
%.
For t h e H-
Q s t a b i l i t y problem we a r e only i n t e r e s t e d i n finding a lower bound t o EN, and t h e nuclear k i n e t i c energy adds a p o s i t i v e term. inf
C R 1~
Q z j ,Rj %({
I n other words,
i s smaller than the ground s t a t e energy of t h e t r u e
Hamiltonian (defined i n (58)) i n which t h e nuclear k i n e t i c energy is included.
Later on when we do the proper thermodynamics of t h e whole
system we s h a l l have t o include the nuclear k i n e t i c energy. The problem of estimating
%Q is
a s old a s t h e Schroedinger equation.
The TF theory, a s i n t e r p r e t e d by Lenz
(Lenz, 1932), reads a s follows:
For fermions having q s p i n s t a t e s (q= 2 f o r elecfrons) define t h e e n e r a functional:
i s t h e TF energy f o r A e l e c t r o n s (1 need not be an i n t e g e r , of course). When A = N the minimizing
p
is supposed t o approximate the p
JI
given by
is
(17), wherein 9 i s the t r u e ground s t a t e wave function, and supposed t o approximate
Q %.
The second and fourth tenne on t h e r i g h t s i d e of (26) a r e exact but the f i r s t and t h i r d a r e not.
The f i r s t is t o some e x t e n t j u s t i f i e d by
t h e k i n e t i c energy i n e q u a l i t y , Theorem 1; the t h i r d term w i l l be discussed later.
I n any event, (26) and (27) define TF theory.
It would b e too much t o t r y t o reproduce h e r e the d e t a i l s of our analysis of TF theory.
A s h o r t summary of some of t h e main theorems w i l l
have t o s u f f i c e . The f i r s t question is whether o r not
E~P
(which, by simple estimates
using Young's and Hglder's i n e q u a l i t i e s can b e shown t o b e f i n i t e f o r a l l A ) is a minimum as d i s t i n c t from merely an infimum.
The d i s t i n c t i o n is
c r u c i a l because the TF equatcon (the Euler-Lagrange equation f o r (26) and
with
has a s o l u t i o n with
i f and only i f there is a minimizing
Ip-x
The b a s i c theorem is Theorem 3. (i)
k A
IZ I
I zj
j-1
than
(0) haa a minimum on the s e t I P ( X ) ~ I=CA .
YI
TF
f o r EA
.
(ii) and ( 2 9 ) .
TF
This minimizing P ( c a l l i t p X ) i s unique and s a t i s f i e s (281 p is non-negative,
a n d - u i s t h e chemical p o t e n t i a l , i.e.
-u (iii)
-
aEy/ax
.
There i s no other s o l u t i o n t o (28) and (29) (for any
IP=X
TF other than pX (iv)
u
) with
.
X = 2, p = 0.
Otherwise Y > 0,
&E? is s t r i c t l y
decreasing i n A. (v)
&X
v a r i e s from 0
to 2,
p v a r i e s continuously from
+-
0.
(vi) p i s a convex, decreasing function of A. TF
(di) $A (x) > 0 f o r a l l x (5/3)ICC q-2'3 TF
I f X > Z then E ( I ) is not with Jp = X. Negative ions -
EP
exists and Ef
=
E?
X.
Hence when X = Z
P ? ( x ) ~ / ~=
(F(3 .
a minimum and (28) and (29) have no s o l u t i o n do not e x i s t i n TF theory.
Nevertheless,
@ X 2 2.
The proof of Theorem 3 is an exercise i n functional analysis. Basically, one f i r s t shows t h a t &(P) i s bounded below s o t h a t Ef
exists.
The Banach-Alaoglu theorem is used t o find an L " ~ weakly convergent TF sequence of 0's such t h a t E ( P ) converges t o EX
.
Then one notes t h a t
&(PI is weakly lower semicontinuous s o t h a t a minimizing p e x i s t s . uniqueness comes from an important property of $(p), convex. -
The
namely t h a t i t is
This a l s o implies t h a t the minimizing p s a t i s f i e s /p
-
A.
A
major point t o notice is t h a t a s o l u t i o n of t h e TF equation is obtained as a byproduct of minimizing. g ( p ) ; a d i r e c t proof t h a t the TF equation has a solution would be very complicated.
Apart from t h e d e t a i l s presented i n Theorem 3, the main point i s t h a t TF theory i s w e l l defined.
I n p a r t i c u l a r the density p i s unique
-
a s t a t e of a f f a i r s i n marked contrast t o t h a t of Hartree-Fock theory. The TF density p p has t h e following properties:
Lf
Theorem 4. (i)
12Z
then
-
(5/3)KC q-213 ~ 7 ( x ) ~Z ~ ~I X~-
near each Ri. (ii)
I n the n e u t r a l caae, 1 = Z =
R ~ I - ~
k
1
j-1
as
1x1
+
(iii)
OD,
i r r e s ~ e c t i v eof t h e d i s t r i b u t i o n of t h e nuclei.
TF
TF
p X (x) a r e r e a l a n a l y t i c i n x away from a l l the Ri,
(x)
on a l l of 3-space i n t h e n e u t r a l case and on ix:
TF (x) > 11) i n the
p o s i t i v e i o n i c case. (32) is especially remarkable:
a t l a r g e distances one loses a l l
knowledge of the nuclear charges and configuration.
Property (i)r e c a l l s
the s i n g u l a r i t y found i n the minimization of hC(p) (see (13)). (31) can be seen from (28) and (29) by inspection.
(32) is more
s u b t l e b u t i t is consistent with t h e observation t h a t (28) and (29) can be rewritten (when u=O) a s c 312 -(4n)-'~ @fP(x) = -i(3/5)q213 ( y ( x ) h C 1 away from the R
i*
I f i t is assumed t h a t 4
of 1x1 then (32) follows. (Sommerfeld, 1932).
TF
z (x) goes t o zero as a p w e r
This observation was f i r s t made by Sonnnerfeld
The proof t h a t a paver law f a l l o f f actually occurs
is somewhat s u b t l e and involves p o t e n t i a l t h e o r e t i c ideas such a6 t h a t used i n the proof of Leauna 8.
As pointed o u t e a r l i e r , t h e connection between TF theory and t h e Schroedinger equation i s b e s t seen i n t h e l i m i t Z +
k, of n u c l e i be h e l d f i x e d , b u t l e t N
+ m
and zi +
-.
-
Let t h e n u d e r ,
i n such a way t h a t
k
1
t h e degree of i o n i z a t i o n N/Z i s constant,where Z = we make the following d e f i n i t i o n :
To this end
j=l zj'
Fix {z R lk and X. I t is n o t j' j j = l For each N = 1,2, define aN by
...
necessary t o assume t h a t X ( 2 .
his means (23) replace z j by zj% and R by R -'I3. j j a ~ t h a t t h e nuclei come together a s N + I f they s t a y a t f i x e d p o s i t i o n s
1%
In
= N.
l$
-.
then t h a t is equivalent, i n t h e limit, t o i s o l a t e d atoms, i . e .
i t is
equivalent t o s t a r t i n g with a l l t h e n u c l e i i n f i n i t e l y f a r from.each other. F i n a l l y f o r the nuclear configuration {%z j, ground s t a t e wave function,
41 be
the
Q t h e ground s t a t e energy, and pN(x) be the
s i n g l e p a r t i c l e density a s defined by (17). J,
sN-113~jl:=l l e t
[Note:
If
Q
i s degenerate,
can be any ground s t a t e wave function a s f a r a s Theorem 5 is concerned.
If
Q
i s not an eigenvalue, but merely i n £ spec
l$,
then i t i s p o s s i b l e
Q s t i l l given by (17), i n t o d e f i n e an approximating sequence qN, with pN such a way t h a t Theorem 5 holds.
We omit t h e d e t a i l s of t h i s construc-
t i o n here.] It i s important t o note t h a t t h e r e i s a simple and obvious s c a l i n g
r e l a t i o n f o r TF theory, namely
and
f o r any a
2 0.
Hence, f o r t h e above sequence of systems parametrized by
f o r a l l N. I f , on the other hand, the n u c l e i a r e held f i x e d then one can prove that
where E?(Z) i s t h e energy of an i s o l a t e d atom of nuclear charge z. The j k 1 a r e determined by t h e condition t h a t X = X i f X ( Z (otherwise 1 j=1j
1
Xj = Z) and t h a t the chemical p o t e n t i a l s of the
1=1
lc at-
same.
Another way t o say t h i s i s t h a t t h e X
(37).
With t h e nuclei fixed, the analogue of (36) i s
l i m <2 P?
N-'
(a-1'3(x-~j))
1
a r e a l l the
minimize the r i g h t s i d e of
TF
= p X (x) j
.
The r i g h t s i d e of (38) is t h e P f o r a s i n g l e atom of nuclear charge z and e l e c t r o n charge A
(37) and (38) a r e a p r e c i s e statement of t h e j' f a c t t h a t i s o l a t e d atoms r e s u l t from f i x i n g t h e R
5'
The TF energy f o r an i s o l a t e d , n e u t r a l atom of nuclear charge Z i s found numerically t o be E?
For f u t u r e w e note t h a t
= 4 2 . 2 1 ) q 2/3(Kc)-1 z7/3
~fis proportional
t o 1/KC.
Thus, i f one
considers a TF theory with KC replaced by some o t h e r constant a > 0 , a s
w i l l be necessary i n Section I V , then (39) is c o r r e c t i f KC i s replaced
Theore.5.
maNEN/P&{z
,%'1/3~ 1)
<7/3#"atlzj
'i'
R lk fixed j' j j=lhas a l i m i t a s N
TF
-t
-.
k
This l i m i t i s El (Ezj ,Rj Ijsl).
(ii)
Q %-7/3 %({%zj,Rj))
(iii)
has a l i m i t a s N -+
m.
This limit is the
r i g h t s i d e of (37). (iv) If P 5 Z -
G~ p:(<'l3x;
-t
{%zj ,
-.
z j , t h i s l i m i t i s p y ( x ) and the convergence i s i n weak
J -1
TF 1 3 ~ ' ( 1 ~ ) .If I > 2, the linit i s p Z ( ~ i)n weak LlOc(1 ). (v)
For fixed nuclei, <2
p:(<113(x-~
) .{ z R 1) has a l i m i t j ' a ~ j * j
( i n the same sense a s ( i v ) ) which is the r i g h t s i d e of (38). The proof of Theorem 5 does not use anything introduced so f a r .
is complicated, but elementary. s i d e s of order
z'lI3.
It
One p a r t i t i o n s 3-space i n t o boxes with
I n each box the p o t e n t i a l i s replaced by i t s
maximum (resp. minimum) and one obtains an upper (resp. lower) bound t o
# by imposing Diriehlet on the boxes. The -r"
($=O) (resp. Neumann (V$=O)) boundary conditions
The upper bound is e s s e n t i a l l y a Hartree-Fock calculation.
s i n g u l a r i t y near the nuclei poses a problem f o r the lower bound,
and i t is tamed by exploiting the concept of angular momentum b a r r i e r . What Theorem 5 says, f i r s t of a l l , i s t h a t the t r u e quantum energy has a l i m i t on the order of charge is held fixed.
z'lI3
the r a t i o of electron t o nuclear
Second, t h i s l i m i t is given correctly by TF
theory a s is shown i n (35). together as
z7I3 when
The requirement t h a t the nuclei move
should be regarded a s a refinement r a t h e r than as a
drawback, f o r i f the nuclei a r e fixed a l i m i t a l s o e x i s t s but i t i s an uninteresting one of i s o l a t e d atoms.
4 is proportional t o Z2 and Theorem 5 a l s o says t h a t t h e density, pH, I f h > Z, Theorem 5 s t a t e s
has a s c a l e length proportional t o Z-ll3.
t h a t the surplus charge moves off t o i n f i n i t y and t h e r e s u l t is an i s o l a t e d molecule.
This means t h a t l a r g e atoms o r molecules cannot have
a negative i o n i z a t i o n proportional t o t h e t o t a l nuclear charge: a t b e s t they can have a negative i o n i z a t i o n which is a vanishingly small f r a c t i o n of t h e t o t a l charge.
This r e s u l t i s physically obvious f o r e l e c t r o s t a t i c
reasons, b u t i t is nice t o have a proof of it. Theorem 5 a l s o resolves c e r t a i n "anomalies" of TF theory which are: (a)
I n r e a l atoms or molecules t h e e l e c t r o n density f a l l s off
exponentially, while i n TF theory (Theorem 4) the density f a l l s off a s
1.p.
(b)
The TF atom shrinks i n s i z e as
z ' ~ ' ~(cf.
(36)) while r e a l
l a r g e atoms have roughly constant s i z e . (c)
I n TF theory t h e r e is no molecular binding, a s we s h a l l show
next, b u t t h e r e is binding f o r r e a l molecules. (d)
I n r e a l molecules t h e e l e c t r o n density is f i n i t e a t the nuclei,
but i n TF theory i t goes t o i n f i n i t y a s z lx-R
3
As Theorem 5 shows,
molecules.
1-312
(Theorem 4).
TF theory i s r e a l l y a theory of heavy atoms o r
A l a r g e atom looks l i k e a s t e l l a r galaxy, p o e t i c a l l y speaking.
It has a core which shrinks as
electrons.
3
Z-'I3
and which contains most of the
2 The density (on a s c a l e of Z ) is not f i n i t e a t the nucleus
because, a s t h e simplest Bohr theory shows, the S-wave e l e c t r o n s have a3 2 density proportional t o Z which is i n f i n i t e on a s c a l e of Z
.
Outside
the core is a mantle i n which t h e density is proportional t o (cf. Theorem 4) ( 3 1 ~3[ ) ( 5 / 3 ) ~ ~ 2 - ~z /~~ /] (~ z x[16 ~ / ~which ~ is independent of Z!
This density i s correct t o i n f i n i t e distances on a length s c a l e The core and t h e mantle contain 100% of t h e electrons as Z*.
z''l3.
The t h i r d
region i s a t r a n s i t i o n region t o t h e outer s h e l l and while i t may contain many electrons, i t contains only a vanishingly small f r a c t i o n of them. The fourth region i s t h e outer s h e l l i n which chemistry and binding takes place.
TF theory has nothing t o say about t h i s r e g i ~ n .The ~ f i f t h region
is t h e one i n which t h e density drops off exponentially. Thus, TF theory deals only with t h e core and the mantle i n which the bulk of the energy and t h e electrons reside.
There ought not t o be
binding i n TF theory, and indeed there i s none, because TF energies a r e proportional t o
z7I3
and binding energies a r e of order one.
The binding
occurs i n the fourth l a y e r . An important question i s what is t h e next term i n the energy beyond the
z7I3 term
of TF theory.
Several corrections have been proposed:
(e.g. Dirac, 1930, Von ~ e i z s ~ c k e 1935, r, Kirzhnits, 1957, Kompaneets and Pavlovskii, 1956, S c o t t , 1952). these corrections a r e of order should be a
z6I3 correction
With the exception of the l a s t , a l l
z5I3.
S c o t t (as l a t e a s 1952!) s a i d there
because TF theory is not a b l e t o t r e a t Recall t h a t i n Bohr theory each
c o r r e c t l y the innermost core electrons.
inner e l e c t r o n alone has an energy proportional t o Z
2
.
As these inner
e l e c t r o n s a r e unscreened, t h e i r energies should be independent of t h e presence o r absence of t h e electron-electron repulsion. the Z
2
I n o t h e r words,
correction f o r a molecule should be precisely a sum of corrections,
one f o r each atom.
The atomic correction should be t h e difference
between t h e Bohr energy and the
z7I3 TF
energy f o r an atom i n which t h e
electron-electron repulsion i s neglected.
We already calculated t h e TF
energy f o r such an "atom" i n (14) (put 251 there and then use scaling; Thus, f o r a n e u t r a l atom without electron-
a l s o replace ICC by q'213~C). e l e c t r o n repulsion ZTF Z
,,(31/314)q2/3 z7/3
2 2 2 For the Bohr atom, each s h e l l of energy -2 /4n has n s t a t e s , so L
1
z/q = Nlq = n*l n ' + ( ~ + l ) ~ += ~ ~ / 3 + ~ ~ / 2 + ~ / 6 + ( ~ + 1 ) ~ + with 0
2
+ L', 1 being
One f i n d s L
I
the f r a c t i o n of t h e (L+l)th s h e l l t h a t is f i l l e d .
(3Z/q) 'I3-
112
- + + o(1)
and
Thus, t o the next order, the energy should be
4({53Rj lkp l )
TF
'%
since q = 2 f o r electrons. correction
k ({zj ,Rjlj-l) Note that
1
k
-4 j'l1 z2j + lower
- q2j3 while
order,
the Scott
- q.
I t i s remarkable t h a t (41) gives a p r e c i s e conjecture about the
next correction.
It i s simple t o understand physically,
yet we do not have the means t o prove it. The t h i r d main f a c t about TF theory is t h a t there i s no binding. This was proved by Teller i n 1962.
Considering the e f f o r t t h a t went i n t o
the study of TF theory s i n c e i t s inception i n 1927, it i s remarkable t h a t the no-binding phenomenon was not seriously noticed u n t i l the computer study of Sheldon i n 1955.
T e l l e r ' s o r i g i n a l proof involved some
questionable manipulation with 6-functions and f o r t h a t reason h i s r e s u l t
Ris ideas were b a s i c a l l y r i g h t , however, and we have
was questioned.
made them rigorous. I f t h e r e a r e a t l e a s t two nuclei, write the
Theorem 6 (no binding).
-I k
nuclear a t t r a c t i o n V ( 3
zj lx-Rj j=1
1
V = V 1+V 2
zjlx-Rjl-' jP1 TF energy f o r the n u c l e i 1,. ,m V (x) =
1-1 a s
t h e sum of two pieces,
,nd 1
cm < k.
.. (e U = 1 lli<j<m
z z j
course) and l e t E Y s 2 be the same f o r the n u c l e i ni+l, l e t A1 -
)0
& A2
= A-A1
m
men
z j=l
IR
-R
...,k.
j
I-', of Given A ,
) 0 be chosen t o minimize the sum of t h e energies
of the separate m l e c u l e s , i.e. Theorem 3, ll =
be t h e
By''
E?"
1
+ ETFv2. (c A = A2
Z =
k
1g
then by
j=l
Since the r i g h t s i d e of (42) is t h e energy of two widely separated molecules, with the r e l a t i v e nuclear positions unchanged within each molecule, Theorem 6 says t h a t the TF energy is unstable under every decomposition of t h e b i g molecule i n t o smaller molecules.
In
p a r t i c u l a r , a molecule is unstable under &composition i n t o i s o l a t e d atoms, and Theorem 9 is a simple consequence of t h i s f a c t .
One would
suppose t h a t i f A and t h e z a r e fixed, b u t the R a r e replaced by aR j J j then
TF z
E
1'
a
1
) is'monotone decreasing i n a.
jj=l
I n other words, the "pressure" i s always p o s i t i v e .
This i s an unproved
conjecture, but i t has been proved ( ~ a l ' a z s , 1967) i n the case k = 2 and
z1 = z2-
An i n t e r e s t i n g s i d e remark i s Theorem 7.
I f t h e TF energy (26), (27) i s redefined by excluding t h e
repulsion term U i n (26), then the inequality i n (42) i s reversed. Thus, the nuclear repulsion is e s s e n t i a l f o r t h e no-binding Theorem 6. Another useful f a c t f o r some f u r t h e r developments of the theory, especially t h e TF theory of s o l i d s and the TF theory of screening (LiebSimon, 1976) is t h e following lemma (also due t o Teller) which i s used t o prove t h e main no binding Theorem 6. {RjI&l
Lemma 8.
k {zj }j31.
O
~
and f i x p
2 0 i n the TF equation (28) but not
(This means t h a t a s the z 's a r e varied X w i l l v a q , b u t always
.-
X =Czj ~ Z~f p =
a r e two s e t s of
and i f hl
and X2
2's
J
o
A =
z always.)
~f {z:)iml
{z
2k I 1 j-1
such t h a t
a r e the corresponding X's f o r t h e two s e t s , then f o r
all x and hence
There i s s t r i c t inequality when p = 0.
I n s h o r t , increasing some z
increases the density everywhere, not j u s t on t h e averaEe. The proof of Lemma 8 involves a beautifully simple p o t e n t i a l t h e o r e t i c argument which we cannot resist giving.
J
TF
We want t o prove gl (x)
Proof of Lemma 8.
2 +2TF (x) f o r a l l x,
and w i l l content ourselves here with proving only I w h e n p-0. B = {x:
TF
Let
TF
(x) > +2 (x)). B i s an open s e t and B does not contain any R i 1 2 TF TI? f o r which zi < zi by the TF equation (29). Let $(x) = g1 (x) g2 (x). TF TF If x € B then $(x) > 0 and, by (28), pl (x) > p 2 (x). For xEB,
-
-(4s)-'~$(x)
= p
TF TF (x) < 0, s o $ i s subharmonic on B ( i . e . $(x) 5 2 (x)-p 1
the average of 9 on any sphere contained i n B and centered a t x). Hence $ has i t s maximum on the boundary of B o r a t
-,
a t a l l of which points $10.
a
Therefore B i s the empty s e t .
I n the p=O case it is easy t o show how Theorem 6 follows from Lemma
Proof of Theorem 6 when A =
k
I zj.
J=l
same ideas but is more complicated.
k A2 =
-E
TF
j*l
I 4 and 0.p
(zdl,.
The proof when X < Cz uses the m 1 Since A = I z j then A1 zj,
-I
f o r a11 three systems.
For a > 0 l e t
..,%;Rdl, ...,%) , where the three E~
systems (i.e. p=O f o r a l l a ) .
a r e defined f o r n e u t r a l
The goal is t o show t h a t f ( 1 )
f (0) = 0, i t is enough t o show t h a t df(a)/da
2 0.
jsl
2
0.
Since
From (26) and (27) i t
is t r u e , and almost obvious, t h a t
This is the TF version of the Feynman-Hellman theorem; notice how the nuclear-nuclear repulsion comes i n here.
where na(x)
-
TP
TF (x) and
Thus,
(y is the p o t e n t i a l f o r
.., a ~ ~ , z , ~ ,...,%; Rl,. ..,\I and )? is t h e p o t e n t i a l f o r TF TF {az, ,...,a zm ; R 1,...,Rm 1. (x) ) +2 (x) f o r a l l x by Lenrma 8, and
carl,.
'
ria (x)
hence
2
0.
Theorem 6 has a n a t u r a l application t o the s t a b i l i t y of matter problem.
A s w i l l be shown i n the next section, the TF energy (27) is,
with s u i t a b l y modified constants, a lower bound t o the true quantum energy for Theorem 9.
2.
By Theorem 3 ( i v ) and Theorem 6 we have t h a t
Q { z
R
k
Ik
and let Z =
j' j j=l-
1 zj.
Jal
Then f o r ~ 1 1 1 1 ~ 0
The l a t t e r constant, 2.21, i s obtained by numerically solving t h e TF equation f o r a s i n g l e , n e u t r a l atom (J. Barnes, p r i v a t e cormnunication). By scaling, (43) holds f o r any choice of K'
i n the d e f i n i t i o n (26) of
$5(PI. Theorem 9 is what w i l l be needed f o r the H-stability because i t says t h a t the TF system is H-stable,
i.e.
of matter,
the energy is
bounded below by a constant times the nuclear p a r t i c l e number (assuming t h a t the z
1
a r e bounded, of course).
Another application of Theorem 6 t h a t w i l l be needed i s the following strange inversion of the r o l e of electrons and n u c l e i i n TF theory.
It
w i l l enable us t o give a l a s e r bound t o the t r u e quantum-mechanical electron-electron repulsion.
This theorem has nothing t o do with quantum
mechanics per se; i t i s r e a l l y a theorem purely about e l e c t r o s t a t i c s even though i t i s derivqd from the TF no binding theorem.
Theorem 10. Suppose t h a t xl,
...,%
and define
a r e any N d i s t i n c t points i n 3-space
-1 N
V,(y)
lY-xj 1-l j =l
Let y -
> 0 and l e t p(x) be any nownegative function such t h a t Ip(x)dx <
and P! -
(x) 'I3dx
Proof. -
R and
x
<
hen
0.
Consider g ( p ) (26) with q=1, k d , E f replaced by y, zj E 1 and
,1
EY2
..
Let A = j p (x)dx.
N.
-(2.21)N/y
(45) is j u s t g ( p )
+
by Theorem 9. (2.21)NIy.
Then g(p)
2
E?
(by d e f i n i t i o n )
The difference of the two s i d e s i n
1
IV.
The S t a b i l i t y of Bulk Matter The various r e s u l t s of t h e l a s t two s e c t i o n s can now be assembled
t o prove t h a t the ground s t a t e energy (or infimum of t h e spectrum, i f t h i s is n o t an eigenvalue) of
% is
bounded below by an extensive
quantity, namely the t o t a l number of p a r t i c l e s , independent of the nuclear l o c a t i o n s {R 1. This is c a l l e d the H - s t a b i l i t y of m a t t e r t o j d i s t i n g u i s h i t from thermodynamic s t a b i l i t y introduced i n the next section.
A s explained b e f o r e , t h e i n c l u s i o n of t h e nuclear k i n e t i c
energy, a s w i l l be done i n t h e next s e c t i o n , can only r a i s e t h e energy. The f i r s t proof of t h e N-boundedness of t h e energy was given by Dyson and Lenard (Dyson-Lenard,
1967, Lenard-Dyson,
1968).
Their proof
is a remarkable a n a l y t i c t o u r de f o r c e , but a chain of s u f f i c i e n t l y many i n e q u a l i t i e s was used t h a t they ended up with an e s t i m a t e of something l i k e -1014 ~ y d b e r g s / p a r t i c l e . Using the r e s u l t s of t h e previous s e c t i o n s we w i l l end up with -23 Rydbergs/particle (see (55)). W e have i n mind, of course, t h a t the nuclear charges z i f they a r e js not a l l the same, a r e bounded above by some f i x e d charge z .
Take any fermion ~ ( x ~ , . . . , x ~ ; ~ ~ , . ). .which , is normalized and N antisymmetric i n t h e (xi,ui). particle density P
J,
Define the k i n e t i c energy T
a s i n (16) and (17).
J,
and the s i n g l e
We wish t o compute a lower
bound t o (46) with
% being
t h e N-particle Hamiltonian given i n (23) and
= I.
'
For t h e t h i r d term on t h e r i g h t s i d e of (23) Theorem 10 can be used with p taken t o be p*.
Then, f o r any y > 0
Notice how the f i r s t and second terms on the r i g h t s i d e of (45) combine t o give
+ 1/2
since
To control the k i n e t i c energy i n (23) Theorem 1 is used; the t o t a l r e s u l t is then
Q 2 a ~ ~ , ( x ) ~ / ~ d x - ~ ~ ( x )dx s , (+x$ )
E,
1 (x) ~ ~
111Y -0 '
J, (yldxdy
with
R e s t r i c t y, which w a s a r b i t r a r y , s o t h a t a > 0. Then, a p a r t from the constant term -(2.21)~y-l, (49) i s j u s t g a ( p ), the Thomas-Fenni J,
energy functional ga(p,)
-> E"a,N
6 applied
to p
$'
but with q'2'3~C
replaced by a.
Since
:inf { F a ( p ) : I p = ~ }(by d e f i n i t i o n ) , and since the n e u t r a l
case always has the lowest TF energy, a s shown i n Theorem 9, we have t h a t
Thus we have proved the following: Theorem 11. f
+
i s a normalized, antisymmetric function of space and
spin of N variables, and i f there a r e q s p i n s t a t e s associated with each
p a r t i c l e then, f o r any y > 0 such t h a t a defined by (50) is p o s i t i v e ,
The optimm choice f o r y
i n which case
# , -(2.21)
k
1/2 2
This is the desired r e s u l t , but some a d d i t i o n a l remarks a r e i n order. (1)
-
Since [l+a 'l2l2 < 2+2a,
a r e bounded above by some f i x e d z, Thus, provided t h e nuclear charges z j is indeed bounded below by a constant tfoes t h e t o t a l p a r t i c l e number
#
N+k
. (2)
Theorem 11 does not presuppose n e u t r a l i t y .
(3)
For electrons, q=2 and t h e prefactor i n (53) is -(2.08)N.
As
remarked a f t e r Theorem 1, t h e unwanted constant ( 4 ~ ) ~ has ' ~been improved t o [4n/ (1.83)
1'I3.
Using t h i s , the prefactor becomes
- (1.39)N.
If
zj = 1 (hydrogen atoms) and N = k ( n e u t r a l i t y ) then
# 2 -(5.56)N (4)
= -(22.24)N
By
.
The power law z7I3 cannot be improved upon f o r l a r g e z because
Theorem 5 a s s e r t s t h a t the energy of an atom is indeed proportional to z7I3 f o r l a r g e z.
(5)
It is a l s o possible t o show t h a t matter i s indeed bulky.
This
w i l l be proved f o r any J, and any nuclear configuration (not j u s t the minimum energy configuration) f o r which E~ < 0.
d
-
The minimizing nuclear
configuration is, of course, included i n t h i s hypothesis.
where
N . 8;1 i s (23) but with a f a c t o r 112 multiplying 1-1 I Ai.
,jy> 2 2%, where EN is
Then
By Theorem (11).
the ri&t s i d e of (53) (replace K'
by KC/2
Therefore, the f i r s t important f a c t i s t h a t
there).
and t h i s is bounded above by the t o t a l p a r t i c l e number.
2 0,
i t is easy t o check t h a t there is a C > 0 such P t h a t f o r any nonnegative p(x),
Next, f o r any p
{ ~ (x)5/3dx~p12 p j
1 XI
P (x) dx
2
cp{Jp (x)dx) 1+5p/6
It is easy t o f i n d a minimizing p f o r t h i s , and t o calculate C : p(xl2I3 = 1-1x1'
P
Since T
4'
-
f o r 1x1 < 1; ~ ( x )= 0, otherwise.
s a t i s f i e s (18) we have t h a t
-
with C' C (KC/4)*I2 (4ns)-~/3. P P I f it is assumed t h a t ~ Z ! / ~ / N is bounded, and hence t h a t J ( N ~ ' ~ / )'I2 > d l 3 f o r some A, we reach the conclusion t h a t the radius of the system i s a t l e a s t of t h e order N " ~ ,
a s i t should be.
The above analysis did not use any s p e c i f i c property of t h e Coulomb p o t e n t i a l , such as the v i r i a l theorem. general Hamiltontan H i n (58). n,k
It i s a l s o applicable t o t h e more
(6)
The q dependence was purposely retained i n (53) i n order t o say
something about bosom.
I f q=N, then i t is easy t o s e e t h a t the' require-
ment of antisymmetry i n $ is no r e s t r i c t i o n a t a l l . one has simply
2 3N over a l l of L (1 )
I n t h i s case then,
# = inf spec %
.
Therefore
It was shown by Dyson and Lenard (Dyson-Lenard, 1967) t h a t EQ (bosons) N
2 -(constant)N 5/3 ,
and by Dyson (Dyson, 1967) t h a t
Q %(bosone) Proving (57) was not easy.
5 -(constant) N 715
.
Dyson had t o construct a r a t h e r complicated
v a r i a t i o n a l function r e l a t e d t o the type used i n the BCS theory of superconductivity.
Therefore bosom a r e not s t a b l e under the a c t i o n of
Coulomb forces, but the exact power law is not yet known.
Dyson has
conjectured t h a t i t is 715. I n any event, the e s s e n t i a l point has been made t h a t F e d s t a t i s t i c s is
e s s e n t i a l f o r the s t a b i l i t y of matter.
The uncertainty principle
f o r one p a r t i c l e , even i n the strong form (5), together with i n t u i t i v e notions t h a t the e l e c t r o s t a t i c energy ought not t o b e very great, a r e insufficient for stability.
The additional physical f a c t t h a t is needed
is t h a t the k i n e t i c energy increases a s the 513 power of the fermion density.
-
V.
The Thermodynamic Limit
Q i s bounded below by t h e t o t a l p a r t i c l e Having e s t a b l i s h e d t h a t EN number, the next question t o consider i s whether, under appropriate conditions, $/N
has a l i m i t a s N +
-, a s
expected.
More generally, the
same question can be asked about the f r e e energy per p a r t i c l e when t h e t e m p e r a t u r e is not zero and t h e p a r t i c l e s a r e confined t o a box. It should be appreciated t h a t the d i f f i c u l t y i n obtaining t h e lower
bound t o
4
came almost e n t i r e l y from t h e r
t h e Coulomb p o t e n t i a l .
-1
s h o r t range s i n g u l a r i t y of
Other p o t e n t i a l s , such a s t h e Yukawa p o t e n t i a l ,
with t h e same s i n g u l a r i t y would present t h e same d i f f i c u l t y which would be resolved i n the same way.
The s i n g u l a r i t y was tamed by the
p
5/3
behavior of t h e fermion k i n e t i c energy. The d i f f i c u l t y f o r t h e thermodynamic l i m i t i s d i f f e r e n t .
-1 caused by t h e long range r behavior of t h e Coulomb p o t e n t i a l .
It i s
I n other
words, we a r e faced with t h e problem of explosion r a t h e r than implosion. Normally, a p o t e n t i a l t h a t f a l l s off with d i s t a n c e more slowly t h a n r'3'E f o r some
E
> 0 does = h a v e
a thermodynamic l i m i t .
Because the charges
have d i f f e r e n t signs, however, t h e r e i s hope t h a t a c a n c e l l a t i o n a t l a r g e d i s t a n c e s may occur. An a d d i t i o n a l physical hypothesis w i l l be needed, namely n e u t r a l i t y . To appreciate t h e importance of n e u t r a l i t y consider t h e case t h a t the e l e c t r o n s have p o s i t i v e , instead of negative charge. every term i n (23) would be p o s i t i v e .
Q > 0 because Then EN
While the H-stability question i s
t r i v i a l i n t h i s case, the thermodynamic l i m i t is not.
I f the p a r t i c l e s
a r e constrained t o be i n a domain R whose volume IRl is proportional t o N , the p a r t i c l e s w i l l r e p e l each other so s t r o n g l y t h a t they w i l l a l l go
t o t h e boundary of R i n order t o minimize t h e e l e c t r o s t a t i c energy.
The
minimum e l e c t r o s t a t i c energy w i l l be of the order +N
2
-1/3
_
,$i/3
Hence no thermodynamic l i m i t w i l l e x i s t . When the system is n e u t r a l , however, the energy can be expected t o be extensive, i.e. O(N).
For t h i s t o be so, d i f f e r e n t p a r t s of the
system f a r from each other must be approximately independent, despite the long range nature of the Coulomb force.
The fundamental physical, o r
r a t h e r e l e c t r o s t a t i c , f a c t t h a t underlies t h i s is screening;
the distri-
bution of the p a r t i c l e s must be s u f f i c i e n t l y n e u t r a l and i s o t r o p i c l o c a l l y s o t h a t according t o Newton's theorem (13 below) the e l e c t r i c p o t e n t i a l f a r away w i l l be zero.
The problem i s t o express t h i s idea i n precise
mathematical form. We begin by defining the Hamiltonian f o r the e n t i r e system consisting 2 of k nuclei, each of charge z and mass M, and n electrons (6 /2 = 1, m = 1,
The f i r s t and second terms i n (58) a r e , respectively, the k i n e t i c energies of the e l e c t r o n s and the nuclei.
The l a s t three terms are, respectively,
the electron-nuclear, electron-electron and nuclear-nuclear Coulomb interactions. a r e yi.
The electron coordinates a r e xi and the nuclear coordinates
The electrons a r e fermions with s p i n 1/2; the nuclei may be
e i t h e r bosons o r fermians. The b a s i c n e u t r a l i t y hypotheses is t h a t n and k a r e r e l a t e d by n-kz. It is assumed t h a t z i s r a t i o n a l .
The thermodynamic 1Mt t o be discussed here can be proved under more general assumptions, i.e. we can have s e v e r a l kinds of negative p a r t i c l e s (but they must a l l b e fermions i n order t h a t the b a s i c s t a b i l i t y estimate of Section I V holds) and s e v e r a l kinds of n u c l e i with Neutrality must always hold,
d i f f e r e n t s t a t i s t i c s , charges and masses. however.
Short range forces and hard cores, i n addition t o t h e Coulomb
forces, can a l s o be included with a considerable s a c r i f i c e i n s i m p l i c i t y of the proof.
H
n,k a s well).
a c t s on square integrable functions of n+k v a r i a b l e s (and s p i n
To complete t h e d e f i n i t i o n of H we must specify boundary n,k conditions: choose a domain Q (an open s e t , which need not be connected) and require t h a t JI = 0 i f xi o r yi a r e on t h e boundary of Q. For each non-negative integer j , choose an n
1
k
j
determined by ( 5 9 ) , and choose a domain Q
j'
and a corresponding
The symbol N
j
will
henceforth stand f o r the pair (n k ) and j' j
We require t h a t the d e n s i t i e s
be such t h a t
.
l i m pj = p jp
is then the density i n the thermodynamic l i m i t .
the Q
j
t o be a sequence of b a l l s of r a d i i R
It can be shown t h a t the
j
and s h a l l denote them by B
same thermodynamic
f r e e energy holds f o r any sequence N
1'
Q
j
Here we s h a l l choose j'
l i m i t f o r t h e energy and
and depends
9 on
the limiting
p and 6, and not on the "shape" of the
Q
j
, provided
the Sl
go t o i n f i n i t y
J
i n some reasonable way. The b a s i c quantity of i n t e r e s t is t h e canonical p a r t i t i o n function
2 where the t r a c e is on L (a)
lNI
and 0 * 1/T, T being the temperature i n
u n i t s i n which Boltzmann's constant is unity. The f r e e energy per u n i t volume is
and the problem i s t o show t h a t with
then
l i m F E F(p,B) J j*
exists.
A s i m i l a r problem is t o show t h a t
the ground s t a t e energy per u n i t volume, has a l i m i t e(p) = l i m E j j* where
The proof we w i l l give f o r t h e l i m i t F(p,B) w i l l hold equally well f o r can be s u b s t i t u t e d f o r F i n a l l statements. j j The b a s i c s t r a t e g y c o n s i s t s of two p a r t s . The e a s i e s t p a r t is t o
e ( p ) because E
show t h a t F
3
is bounded below.
W e already know t h i s f o r E
J
by the
r e s u l t s of s e c t i o n IY. sequence F
i s decreasing.
j
Theorem 12. p =
The second s t e p is t o show t h a t i n some sense t h e
Given N,Q
IN~/IQ~
This w i l l then imply t h e e x i s t e n c e of a l i m i t .
&B
t h e r e e x i s t s a constant C depending only on
0 such t h a t 'F(N,a,B)
Proof. -
Write H
2c
HA+%, where
i s h a l f the k i n e t i c energy.
Then HB
2 b IN I , with b depending only on
z,
by t h e r e s u l t s of Section I V (increasing t h e mass by a f a c t o r of 2 i n HB only changes t h e constant b)
.
Hence Z (N,R, 0)
5 emBblNITr exp (-BHA).
However, Tr exp(-@HA) i s t h e p a r t i t i o n function of an i d e a l gas and i t
is known by e x p l i c i t computation t h a t i t is bounded above by eedlNlwith
1
d depending only on p = I N / 1 Q
1
and 0.
Thus
For t h e second s t e p , two elementary b u t b a s i c i n e q u a l i t i e s used i n t h e general theory of t h e thermodynamic l i m i t a r e needed and they w i l l be described next. A. Domain p a r t i t i o n inequality:
Given t h e domain R and t h e p a r t i c l e
numbers N = (n,k), l e t n b e a p a r t i t i o n of R i n t o
n
a Q
,..., .
1
d i s j o i n t domains
Likewise N i s p a r t i t i o n e d i n t o a i n t e g r a l p a r t s (some of
which may b e zero):
1 N=N+...+N
e
Then f o r any such p a r t i t i o n , r, of 0 and N
.
Here ~r~ means t r a c e over
and
$ is
for the N
defined as i n (58) but with D i r i c h l e t ($I= 0) boundary conditions i
p a r t i c l e s on the boundary of Q
Simply s t a t e d , t h e f i r s t N N
2
2
t o fi
, etc.
1
i
( f o r i=1,...,&).
1 p a r t i c l e s a r e confined t o S1 , the second
The i n t e r a c t i o n among t h e p a r t i c l e s i n d i f f e r e n t domains
is s t i l l present i n
$.
(69) can be proved by t h e Peierls-Bogoliubov
v a r i a t i o n a l p r i n c i p l e f o r Tr ex.
Alternatively, (69) can b e viewed simply
a s t h e statement t h a t t h e i n s e r t i o n of a hard wall, i n f i n i t e p o t e n t i a l on i the boundaries of the fi only decreases 2; the f u r t h e r r e s t r i c t i o n of a i
d e f i n i t e p a r t i c l e number t o each fi
t h a t the t r a c e is then over only the
f u r t h e r reduces 2 because i t means
<
-invariant s u b s p a c e , k n , of t h e
f u l l H i l b e r t space. B.
Inequality f o r the interdomain i n t e r a c t i o n :
The second
i n e q u a l i t y is another consequence of t h e convexity of A + Tr eA (PeierlsBogoliubov inequality): Tr
2 Tr eA exp
where
Some t e c h n i c a l conditions a r e needed here, b u t (70) and (71) w i l l hold i n our application. To e x p l o i t (70), f i r s t make the same p a r t i t i o n n as i n i n e q u a l i t y A and then w r i t e
i
with H being t h a t part of the t o t a l Hamiltonian (58) involving only the N
i
particles i n Q
i
, and
Hi i s defined with the s t a t e d Dirichlet boundary
conditions on the boundary of R
i
.
W(X) , with X standing f o r a l l the
coordinates, i s the inter-domain Coulomb interaction.
I n other words,
W(X) i s t h a t p a r t of t h e l a s t three terms on the r i g h t s i d e of (58) which involves coordinates i n d i f f e r e n t blocks of t h e p a r t i t i o n n.
Technically,
W is a small perturbation of Ho.
With
-BBo and B = -BW
*
A
i n (70), we must calculate <W>.
Since eA = e"o'
i s a simple tensor
lNil,
product of operators on each L ~ ( Q ~ ) W i s merely the average i n t e r domain Coulomb energy i n a canonical ensemble i n which the Coulomb interaction i s present i n each subdomain but the II domains a r e independent of each other.
This b a s i c idea i s due t o Grif f i t h s (Grif f i t h s , 1969). i
other words, l e t q (x), xER
i
, denote
the average charge density i n R
In
i
f o r t h i s ensemble of independent domains, namely
with the notation: A
i n fli, dx
1
Xi stands f o r the coordinates of the
I N iI
particles
i means integration over a l l these coordinates ( i n D ) with the
exception of x
J'
and x
is s e t equal t o x; qj i s the charge (-1 o r + z )
of the j t h p a r t i c l e ; e q ( - f 3 H i
i i ) ( X ,Y ) is a kernel (x-space representation)
i
f o r exp ( - B H ~ ) . q (x) vanishes i f x
4 ni.
With the definitions (75) one has t h a t
(70), together w i t h (76) and (74) is t h e desired inequality f o r t h e interIt is q u i t e general i n t h a t an analogous inequality
domain interaction.
holds f o r a r b i t r a r y two-body potentials.
Neither s p e c i f i c properties of
the Coulomb p o t e n t i a l nor n e u t r a l i t y was used. Now we come t o t h e c r u c i a l point a t which screening is brought in. The following venerable r e s u l t from the Principia Mathematics is e s s e n t i a l . Theorem 13 (Newton)
.
L t e p (x) b e an integrable function on 3-space such
t h a t p(x) = p(y) If_ 1x1 = 1y
1
-
-
(isotropy) a 2 p(x) = 0 i f 1x1 > R f o r
be the Coulomb p o t e n t i a l generated by p .
Then i f 1x1 > R
-
The important point is t h a t an isotropic, n e u t r a l charge d i s t r i b u t i o n generates zero p o t e n t i a l outside its support, i r r e s p e c t i v e of how the charge i s d i s t r i b u t e d r a d i a l l y . Suppose t h a t N
i
i s neutral, i.e.
the electron number = z times the
nucleon number f o r each subdomain i n Sl.
Ci
i
is a
ball of i
radius R
i
centered a t a
i
invariant, q (x) = q (y) i f Ix-a i
and q (x) = 0 i f /=-ail involving q
i
> R
i
.
i
I
Suppose a l s o t h a t the subdomain i
.
= Iy-a
i
Then since H~ i s r o t a t i o n
1,
jqi (x)dx
= 0 (by n e u t r a l i t y )
Then, by Theorem 13, every term i n (76)
vanishes, because when
jZi,
q j(y) = 0 i f
1 y-ail
< R
i
since
Q1 i s d i s j o i n t from 0i
.
Consequently t h e average interdomain i n t e r a c t i o n ,
<W>, vanishes. I n the decomposition, n, of
.
a i n t g nl, ..,n II and
.,N'
N i n t o N1,..
we
w i l l arrange matters such t h a t (i) (ii) (iii)
..,nL-' N', ...,N'-' N~ = 0 . nl,.
are balls are neutral
Then <W> = 0 and, using (69) and (70)
I n addition t o ( i ) , ( i i ) , ( i i i ) i t w i l l a l s o be necessary t o arrange matters such t h a t when Q i s a b a l l BK i n the chosen sequence of domains, then t h e sub-domains
2,...,~'-li n
the p a r t i t i o n of BK a r e
smaller
b a l l s i n the same sequence., With these requirements i n mind the standard sequence, which depends on the l i m i t i n g density, p , is defined a s follows: (1) Choose p > 0. (2)
Choose any No s a t i s f y i n g t h e n e u t r a l i t y condition (59).
(3)
Choose Ro such t h a t 3 28(4n/3)pRo
(4)
lNOl
.
For j > 1 l e t
-
be the r a d i u s of the b a l l B and the p a r t i c l e number i n t h a t b a l l . j
It w i l l be noted t h a t t h e density i n a l l the b a l l s except the f i r s t
pjPps
j ) 1 ,
(82)
while t h e density i n the smallest b a l l i s much bigger:
This has been done s o t h a t when a b a l l BK, K 2 1 i s packed with smaller b a l l s i n t h e manner t o be described below, the density i n each b a l l w i l l come out r i g h t ; the higher density i n Bo compensates f o r t h e portion of BK not covered by smaller b a l l s .
The r a d i i increase geometrically,
namely by a f a c t o r of 28. The number 28 may be s u r p r i s i n g u n t i l i t is r e a l i z e d t h a t t h e objective is t o be a b l e t o pack BK with b a l l s of type BK-l,
BK-2,
i n such a way t h a t as much as p o s s i b l e of BK i s covered and
also t h a t
very l i t t l e of BK i s covered by very small b a l l s .
etc.
I f t h e r a t i o of r a d i i
were too c l o s e t o unity then the packing of BK would be i n e f f i c i e n t from t h i s point of view.
I n s h o r t , i f t h e number 28 i s replaced by a much
smaller number the analogue of t h e following b a s i c geometric theorem
w i l l not be true. Theorem 14 (Cheese theorem).
Xj a p o s i t i v e i n t e g e r define the i n t e g e r
Then f o r each p o s i t i v e i n t e g e r K 2 1 i t i s possible K-1 t o pack t h e b a l l BK of radius % (given by 81) with .U (t-j b a l l s of 1=0 radius R ) "Pack" means t h a t a l l t h e b a l l s i n the union a r e d i s j o i n t .
"1
' (27)1-1(28)2j. 1
.
W e w i l l not give a proof of Theorem 14 here, but note t h a t it
e n t a i l s showing t h a t ml b a l l s of radius
can be packed i n BK i n a
cubic array, then t h a t m2 b a l l s of radius %-2 a r r a y i n t h e i n t e r s t i t i a l region, e t c .
can b e packed i n a cubic
Theorem 14 s t a t e s t h a t BK can be packed with (28) 2 b a l l s of type
3
b a l l s of type BK-2,
is the f r a c t i o n of the K- j volume of BK occupied by a l l the b a l l s of radius R i n the packing, then (27) (28) B~-l'
etc.
If f
1
with
The packing i s asymptotically complete i n the sense t h a t lim K*
K-1
1
OD
j=o
f
K-j
= (1127)
1
yj
5
j=1
1
.
It is a l s o "geometrically rapid" because the f r a c t i o n of
IB K I
that is
uncovered i s
The necessary ingredients having been assembled, we can now prove Theorem 15.
Given p and
B > 0, the thermodynamic l i m i t s F(p,B)
e(p)
(65,67) e x i s t f o r t h e sequence of b a l l s and p a r t i c l e numbers s p e c i f i e d
b~ (80) Proof. -
(81). Let FK given by (64) b e the f r e e energy per u n i t volume f o r t h e
b a l l B with N p a r t i c l e s i n it. For K 2 1 , p a r t i t i o n BK i n t o d i s j o i n t K K 1 i domains D , ,nk , where t h e D f o r i=1,. ,I1-1 designate t h e smaller
..
...
b a l l s referred t o i n Theorem 14, and
n 9.
(which i s the "cheese" a f t e r the
holes have been removed) is t h e remainder of BK. copies of B
0
t o (81). 'N
= 0.
J'
2
j (K-1;
The smaller b a l l s a r e
i n each of these place N
The t o t a l p a r t i c l e number i n
%
p a r t i c l e s according j i s then
a s i t should be.
U s e t h e b a s i c i n e q u a l i t y (79) ; <W> = 0 s i n c e a l l the smaller b a l l s a r e n e u t r a l and dividing by
with f
with
j
I. contains no p a r t i c l e s .
IB ~ ,I we
have f o r K
= y j/27 and y = 27/28.
% -> 0.
,
Thus, taking logarithms and
1 that
This i n e q u a l i t y can be r e w r i t t e n as
(89) i s a renewal equation which can be solved e x p l i c i t l y
by inspection:
We now use the f i r s t s t e p , Theorem 13, on t h e boundedness of FK. 0
Since FK F -+ K
-.
1
d must be f i n i t e , f o r otherwise (90) would say t h a t j j=1 The convergence of the sum implies t h a t -+ 0 a s K + m. Hence
2
C,
%
the l i m i t e x i s t s ; s p e c i f i c a l l y
Theorem 15 is t h e desired goal, namely the existence of the thermodynamic l i m i t f o r the f r e e energy (or ground s t a t e energy) per u n i t volume.
There a r e , however, some a d d i t i o n a l p o i n t s t h a t deserve comment.
,A
For each given l i m i t i n g density P , a p a r t i c u l a r sequence of
domains, namely b a l l s , and p a r t i c l e numbers was used.
It can be shown
t h a t the same l i m i t is reached f o r general domains, with some mild conditions on t h e i r shape including, of course, b a l l s of d i f f e r e n t r a d i i than t h a t used here.
The argument involves packing the given domains
with b a l l s of the standard sequence and v i c e versa. but standard, and can b e found i n (Lieb-Lebowitz,
B. -
The proof i s tedious,
1972);
Here we have considered the thermodynamic l i m i t f o r r e a l matter,
i n which a l l the p a r t i c l e s a r e mobile. of some physical i n t e r e s t .
There a r e , however, other models
One i s Jellium i n which the p o s i t i v e n u c l e i
a r e replaced by a fixed, uniform background of p o s i t i v e charge.
With the
a i d of an additional t r i c k t h e thermodynamic limit can a l s o be proved f o r t h i s model (Lieb-Narnhofer,
1975).
Another, more important model is one
i n which the nuclei a r e fixed point charges arranged periodically i n a This i s the model of s o l i d s t a t e physics.
lattice.
Unfortunately, l o c a l
r o t a t i o n invariance is l o s t and Newton's Theorem 13 cannot be used.
This
problem i s s t i l l open and i t s s o l u t i o r i w i l l require a deeper i n s i g h t i n t o screening.
8
-
C. -
A n absolute physical requirement f o r BF(p,B), a s a function of
1/T, is t h a t i t b e concave.
This is equivalent t o t h e f a c t t h a t the
s p e c i f i c heat is non-negative since ( s p e c i f i c heat) Fortunately i t i s true.
-8
2 2 a 8F(p ,B)/a8 2
.
From the d e f i n i t i o n s (57), (58) we s e e t h a t
I n Z(N,Q,$) is convex i n 8 f o r every f i n i t e system and hence BF(N,O,B)i s concave.
Since t h e l i m i t of a sequence of concave functions is always
concave, the l i m i t BF(p,B) is concave i n 8.
Another a b s o l u t e requirement i s t h a t F(p, B) be convex a s a function of p.
This i s c a l l e d thermodynamic s t a b i l i t y a s d i s t i n c t from
t h e lower bound H - s t a b i l i t y of t h e previous s e c t i o n s . t o t h e f a c t t h a t t h e compressibility i s non-negative, (compressibility)-' t h e o r i e s (e.g.
=
PI ap
a2~
= p
( pfi)/ap2. ,
I t i s equivalent
since
Frequently, i n approximate
Van d e r Waals' theory of the vapor-liquid t r a n s i t i o n , some
f i e l d t h e o r i e s , o r some t h e o r i e s of magnetic systems i n which the magnetization per u n i t volume plays t h e r o l e of p ) , one introduces an F with a double bmp.
Such a n F is non-physical and never should a r i s e i n
an e x a c t theory. For a f i n i t e system, F i s defined only f o r i n t e g r a l N , and hence not for a l l r e a l p.
It can be defined f o r a l l p by l i n e a r i n t e r p o l a t i o n , f o r
example, but even s o i t can n e i t h e r b e expected, nor i s i t generally, convex, except i n the limit.
The i d e a behind t h e following proof i s
standard. Theorem 16.
E ( p ) i s a l s o a convex function of p .
f i x e d 8. Proof:
The l i m i t f u n c t i o n F(p,fi) i s a convex function of p f o r each
mis
means t h a t f o r p
and s i m i l a r l y f o r E(p).
a
Ap 1+(I-h)p2,
0
5
5
1,
A s F is bounded above on bounded p i n t e r v a l s
( t h i s can b e proved by a simple v a r i a t i o n a l c a l c u l a t i o n ) , i t i s s u f f i c i e n t t o prove (92) when A = 112.
To avoid t e c h n i c a l i t i e s (which can be
supplied) and concentrate on t h e main i d e a , w e s h a l l here prove (92) when p 2 and p1
are r a t i o n a l l y r e l a t e d :
ap
1
= bp
2
, a,
b, positive integers.
Choose any n e u t r a l p a r t i c l e number M and d e f i n e a sequence of b a l l s B
1
3 with r a d i i a s given i n (81) and with 28(4a/3)pR0 = (atb) system take N
0
s y s r m take N:
= (a+b)M, N j = Zb M, N'
j
= (28)
3j-1
No, j
= ( 2 8 1 ~ j - 4 : (resp. N:
-
j
even.
In
other h a l f place N
2 j
half
i n an obvious notation.
Since l i m
of t h e s e b a l l s p l a c e N'
particles, 0
5 j 5 K-1.
For t h e
2aM, ~:=(28)~'-11:).
b a l l s (Theorem 14) note t h a t the number of b a l l s B number is
.
2 1. For t h e p1 (resp. p2)
I n t h e canonical p a r t i t i o n ,
Consider t h e p system.
/MI
5
T,
is
of BK i n t o smaller j
and t h i s
p a r t i c l e s and i n t h e
Then i n place of (88) we g e t
I n s e r t i n g (89) on t h e r i g h t s i d e of (931,
-
Gp2
= 0, we can take t h e l i m i t K + i n (94) and o b t a i n (92). KE. The convexity i n p1 and concavity i n 8 of F(p ,B) has another -
Since F i s bounded below (Theorem 13) and
important consequence.
bounded above (by a simple v a r i a t i o n a l argument) on bounded sets i n t h e (p ,B) plane, the convexity /concavity implies t h a t i t i s j o i n t l y continuous i n (p,f3). FK + y
4(
This, together with t h e monotonicity i n K of
(see (go)), implies by a standard argument using Dini's theorem
t h a t the thermodynamic l i m i t i s uniform on bounded (p , B ) sets.
This
uniformity i s sometimes overlooked a s a b a s i c desideratum of t h e thermoWithout it one would have t o f i x p and 6 p r e c i s e l y i n
dynamic l i m i t . taking t h e l i m i t
- an
impossible t a s k experimentally.
With it, it is
s u f f i c i e n t t o have merely an increasing sequence of systems such t h a t
J
-t
p and B
+
.I
B.
The same r e s u l t holds f o r e (p)
.
F. -
An application of t h e uniformity of the l i m i t f o r e (p) is the Instead of confining the p a r t i c l e s t o a box (Dirichlet
following.
) one could consider H boundary condition f o r H defined on a l l of n,k n,k 'L (x3) , i.e. no confinement a t a l l . I n this case
IN(
is j u s t the ground s t a t e energy of a n e u t r a l molecule and i t is expected that
41I N ]has a l i m i t .
Indeed, t h i s l i m i t exists and i t is simply
l i m #/IN] = 1 i m p - l e(p)
P*
IW-
.
There i s no analogue of t h i s f o r F(p,B) because removing the box would cause the p a r t i t i o n function t o be i n f i n i t e even f o r a f i n i t e system. G. -
The ensemble used here i s the canonical ensemble.
It is
possible t o define and prove t h e existence of the thermodynamic l i m i t f o r the microcanonical and grand canonical ensembles and t o show t h a t a l l three ensembles a r e equivalent (i.e.
t h a t they y i e l d the same values f o r
a l l thermodynamic q u a n t i t i e s , such as the pressure).
H. -
(See Lieb-Lebowitz,
Charge n e u t r a l i t y was e s s e n t i a l f o r taming the long range
Coulomb force. this l e t N
What happens if the system i s not neutral?
To answer
il be a sequence of p a i r s of p a r t i c l e numbers and domains, 3 but without (59) being s a t i s f i e d . Let Q = zk -n be the n e t charge, 3 j j p j = INjl/liljl a s before, and p j p. One expects t h a t i f (i)
Qj = 0.
3'
-
~ ~ l ~ ~ 1 0- then ~ ' ~the same l i m i t +
On the other hand, i f
F(P ,$)
is achieved as i f
qj I 5 I
(it)
- ~ ~ then~ +
t h e r e is no l i m i t f o r F(N
j
,flj ,B) .
More
fl ,B) + because t h e minimum e l e c t r o s t a t i c energy is too j' j Both of these expectations can be proved t o b e c o r r e c t .
p r e c i s e l y F(N great.
The i n t e r e s t i n g case i s i f
l i m Qj
(iii)
j-
1 n j 1-213
=
0
exists
.
.
Then one expects a shape
A s s u m e t h a t the fl are geometrically j = Ano with Iflo[ = 1 and = p j with p j + p . Let C
dependent lidt t o e x i s t a s follows. s i m i l a r , i.e.
fl
j
b e t h e e l e c t r o s t a t i c capacity of capacity of fl is then C = C A. j .I t h e expectation is t h a t
no;
I N 1li3
it depends upon t h e shape of flO.
The
From elementary e l e c t r o s t a t i c s theory
.
l i m F(Nj,flj,i3) = F(p,6)+crL/2~ j-
I-'02/2c.
Note t h a t ( ~ ~ 11Ifl 2 ~ j
j
j
+
(95) can be proved f o r e l l i p s o i d s and b a l l s . complicated a s t h e r e s u l t is simple.
The proof i s a s
With work, t h e proof could probably
b e pushed through f o r o t h e r domains Ro with smooth boundaries. The r e s u l t (95) is amazing and shows how s p e c i a l t h e Coulomb f o r c e
is.
I t says t h a t t h e s u r p l u s charge Q. goes t o a t h i n l a y e r near t h e J
surface.
There, only i t s e l e c t r b s t a t i c energy, which overwhelms i t s
k i n e t i c energy, is s i g n i f i c a n t .
The bulk of Q
1
i s n e u t r a l and uninflu-
enced by the s u r f a c e l a y e r because t h e l a t t e r generates a constant p o t e n t i a l i n s i d e t h e bulk.
I t is seldom t h a t one has two s t r o n g l y i n t e r -
a c t i n g subsystems and t h a t t h e f i n a l r e s u l t has no cross terms, as i n (95).
I. -
There might be a temptation, which should be avoided, t o suppose
t h a t the thermodynamic l i m i t describes a s i n g l e phase system of uniform
density.
The temptation a r i s e s from the construction i n the proof of
Theorem 15 i n which a large domain BK i s partitioned i n t o smaller domains having e s s e n t i a l l y constant density. a large domain.
Several phases can be present i n s i d e
Indeed, i f B i s very large a s o l i d i s expected t o form,
and i f the average density, p , is smaller than the equilibrium density,
Ps* of t h e s o l i d a d i l u t e gas phase w i l l a l s o be present.
The location of
the s o l i d inside the l a r g e r domain w i l l be indeterminate.
From t h i s point of view, there is an amusing, although expected, aspect t o the theorem given i n (95). that p
C ps.
Suppose t h a t B is very large and
Suppose, also, t h a t a surplus charge Q = 0y2'3
where V is the volume of the container.
is present,
I n equilibrium, the surplus
charge w i l l never be bound t o the surface of the s o l i d , f o r t h a t would give r i s e t o a larger f r e e energy than i n (95). A s a f i n a l remark, the existence of the thermodynamic l i m i t (and hence the existence of intensive thermodynamic variables such as the pressure) does not e s t a b l i s h the existence of a thermodynamic s t a t e .
In
other words, i t has not been shown t h a t correlation functions, which always e x i s t f o r f i n i t e systems, have l i m i t s as the volume goes t o infinity. present.
Indeed, unique l i m i t s might not e x i s t i f s e v e r a l phases a r e For well behaved p o t e n t i a l s there a r e techniques available f o r
proving t h a t a s t a t e exists when the density is small, but these techniques do not work f o r t h e long-range Coulomb p o t e n t i a l .
Probably the
next chapter to be written i n t h i s subject w i l l consist of a proof that' correlation functions a r e well defined i n the thermodynamic l i m i t .
VI
.
Har tree-Fock Theory
Q (see (23) and (25)), As a p r a c t i c a l matter, a good estimate f o r EN even with fixed nuclei, i s d i f f i c u l t t o obtain.
An old method (Hartree,
1927, Fock, 1930, S l a t e r , 1930) is s t i l l much employed.
Indeed, chemists
r e f e r t o i t a s an ab i n i t i o calculation. Without taking any position on the usefulness of a HF calculation, i t might be worthwhile t o present the r e s u l t s of recent work (Lieb-Simon
1973) t o the e f f e c t t h a t HF theory is a t l e a s t well defined, i.e. equations have solutions.
the HF
Unlike the s i t u a t i o n f o r Thomas-Fed theory,
the solutions a r e not unique i n general. To define HF theory l e t 4
{O1,*.',ON)
denote a s e t of N s i n g l e p a r t i c l e functions of space and spin, +i, i n
L2 (P3 ;O 2). Two s p i n s t a t e s a r e assumed here.
Form the S l a t e r determinant
N
D,,,(xl,. ..,%;al,.
(N!)-~''
-.,flN)
detl Oi(xj,aj)
2 3 2N The L (R ;O ) norm of
which i s an antisymrnetric function of space-spin.
N D > = det IM' 4' 4 i j i,jP1
1
where M4 i s the overlap (Gram) matrix: ,
The HF energy is
f
= inf( :
and by the v a r i a t i o n a l o r i n c i ~ l e
J,'
D > 4
liSjal
-
1)
When the Oi a r e orthonormal one e a s i l y f i n d s t h a t E
EN($)+u(1zj ,Rj
where
V is given i n (24a).
W
i s a m u l t i p l i c a t i o n operator depending on x E l3
'4
and not on a:
K
'4
is a more complicated, s p i n dependent i n t e g r a l operator:
where
i s a p a r t i a l inner product.
(103) do n o t depend on i.
I t is important t o note t h a t W
'4
and K
'4
in
I n the e a r l i e r Hartree theory, which w i l l not
be discussed h e r e , t h e analogue of (103) does contain i dependent operators.
Nevertheless, the theorems t o be presented h e r e do carry over,
mutatis mutandis, t o Hartree theory. If
$ for
HF EN
is f i n i t e .
.
one expect?
The b a s i c question is whether t h e r e is a minimizing
This i s , of course, the same a s minimizing
& ('4) . What
should
I f N < Z+1 ( t h e "less than" is important) t h e r e should be a
minimum because otherwise t h e r e would be an uncompensated p o s i t i v e Coulomb p o t e n t i a l a t l a r g e d i s t a n c e s which, however weak, can always bind an a d d i t i o n a l e l e c t r o n . an integer.)
If N
(Recall t h a t Z = Ez
j
and z
j
2 Z+l the e x i s t e n c e of a minimum
> 0; Z need not be
i s less obvious.
It
may o r may not occur, depending on the d e t a i l s of t h e nuclear configurWe s h a l l have nothing t o say about t h i s l a t t e r case.
ation.
Theorem 17.
I f N < Z+1 = 1+Cz then, f o r any nuclear configuration, there j is s minimizing ( tor Furthermore, the (i & ( can be chosen t o be
g.
orthonormal, i.e. 'M
ij
-
6
id'
The proof of Theorem 1 7 involves a t r i c k which i n retrospect is obvious, but which took some time t o notice. (1)
Eere i s an o u t l i n e
Consider F N ( $ ) as defined by (103), (104) and (105).
This i s
a q u a r t i c expression i n the (i. Both GN(J,) and a r e i n v a r i a n t
JI 4'
under any unitary transformation of t h e form
with R being an N
x
N unitary matrix.
I f R is chosen t o diagonalize M$' ,
we can r e s t r i c t our a t t e n t i o n t o J, such t h a t the gi a r e orthogonal. The minimizing J, w i l l be constructed by taking a weak l i m i t of a sequence
such t h a t
The major d i f f i c u l t y is t h a t a weak l i m i t of orthonormal functions need not be orthogonal.
It could even happen t h a t l i m) ' :4
= ( (independent
IT-
of i ) .
The t r i c k t o overcome the d i f f i c u l t y is t h i s :
minimizing
with
gN(()
+
Instead of
subject t o = 1 consider instead $'
is the N x N i d e n t i t y matrix and the inequality i n (110) is t h a t IN-M JI i s positive semidefinite.
The obvious, but c r u c i a l , f a c t i s t h a t
a weak l i m i t of functions i n SN remains i n SN. I f there is a minimizing JI f o r
%,
the +i can be chosen t o be ortho-
gonal, possibly a f t e r a unitary transformation (107).
2 1. Assume
<$ i ,$ i> = 6i
6
i
> 0, a l l i.
Then, since *ESN,
Then t o see t h a t t h e 6i can be
chosen t o be unity, note t h a t g N ( $ ) i s quadratic i n each
+i, with
i f +i ,is replaced by (yi/6i)1'2 yi.
Clearly
aE N/ayi
5 0 (otherwise
yi > 0,
gNi s
Therefore
l i n e a r i n each
can be decreased i n taking yi=O,
which contradicts the assumption t h a t <+i,+i> thus
gNis
Oi.
> 0 a t the minimum), and
not increased i f yi i s taken t o be 1.
The problem, then, i s t o show two things: (2a)
there is a minimizing JI f o r eN;
(2b)
M$ does not have a zero eigenvalue. (2a).
This is an application of functional analysis.
one can find a sequence i n t h e Sobolev spaceR('W
Given (108)
converges weakly t o
such t h a t each
3
) , i.e.
.
Vi N
1
i s weakly laser semicontinuous, essenti$ J I i is1 a l l y because W -K is a p o s i t i v e operator and is bounded on ~ ' ( 1 ~ ) . The J, JI
The functional
p o s i t i v i t y of the function
I X - ~ I-'
6
on 1 i s used.
Finally,
because V is a r e l a t i v e l y compact perturbation of -A quadratic f o m s .
Thus J, minimizes EN($) on SN.
i n the sense of
(2b).
I f MQ has. a zero eigenvalue then
%
= eN-l,
$i vanishes ( a f t e r a unitary transformation (107)).
i . e . one of t h e
This is impossible
i f N < Z+l because one can always f i n d a $ orthogonal t o such t h a t 6N(41~""$N-199)
< ~N-l(+l,.-.,~N-l).
4N-1
The property of t h e
hydrogenic Ramiltonian (1) t h a t i t has i n f i n i t e l y many negative eigenvalues i s used i n an e s s e n t i a l way. By a standard argument i n the calculus of v a r i a t i o n s , t h e minimizing
Q s a t i s f i e s t h e Euler-Lagrange equation f o r P N ( Q ) a s follows. Theorem 18.
Lot
such t h a t M'
a
...,(N) be any minimizing Q tor $ arranged
I( = ( I $ ~ ,
5.
It is not necessary t o assume t h a t N < Z+1.
2 3 2 be t h e operator on L (1 ,P ):
a s defined i n (104) and (105).
f o r some hi < 0 .
(Al,.
(ii)
..,%I
..
Then f o r i = 1,. ,N
This is the HF equation. a r e t h e lowest N eigenvalues of H
%
N
$.
1
Xi because t h e r e a r e no f a c t o r s of 1/2 i n i-1 The only s l i g h t l y unusual point is ( i i ) which follows from t h e
I t is not t r u e t h a t
(112).
Let HQ
=
f a c t that g N ( Q ) is quadratic i n each Oi. t h e lowest N i s missing,
%
I f some eigenvalue of H
4'
among
can be lowered by using t h e missing eigen-
function i n s t e a d of t h e (N+j) t h eigenfunction. I n summary, j u s t a s i n the analogous case of TF theory, i t has been shown t h a t t h e nonlinear HF equation (113) not only has a s o l u t i o n , b u t
HF
t h a t among these s o l u t i o n s there i s one t h a t .minimizes t h e EF energy EN It i s not easy t o prove d i r e c t l y t h a t (113) has solutions.
.
I n general, i t is d i f f i c u l t t o say much about a minimizing $. One can-
Despite the deceptive notation, (113) is not a l i n e a r equation.
not say, as one could f o r the l i n e a r Schroedinger equation, t h a t the (Pi can be assumed t o be r e a l o r t h a t Oi(x,o) is a product fi(x)gi(a). assumptions a r e often made i n practice.
These
What can be done i s t o r e s t r i c t
the (Pi from the beginning t o be r e a l and/or product functions such t h a t f o r any i $ j gi = g o r g j
i
i s othogonal t o g
1'
Then the whole analysis
can be done afresh and Theorems 17 and 18 w i l l hold. i n t h i s r e s t r i c t e d class might be greater than
ENg",
manner other r e s t r i c t i o n s can be placed on the bi invariance, f o r example) with the same conclusion. requirement is that f o r any 4 and J, =
The minimum,
-a, EN
In the same
however.
(such as rotation The only e s s e n t i a l
i n the r e s t r i c t e d class,
H$"
is i n the same class.
The overriding question is, of course, h a close i s
%HF
to
I#? It
is d i f f i c u l t t o give a precise answer, but i n two limiting cases HF theory is exact.
One i s the hydrogen atom; the other is the 2
+
-
limit.
It was
i n f a c t a determinantal wave function (971, not the best one t o be sure, t h a t was used i n the variational upper bound leading t o Theorem 5.
i n the sense of Theorem 5.
Thus
References ~ a l i z s ,N.,
1967, Formation of s t a b l e molecules within t h e s t a t i s t i -
c a l theory of atoms, Phys. Rev. 5 6 , 42-47. Birman, M.S.,
1961, Mat. Sb.
2
(97), 125-174; The spectrum of
singular boundary value problems, Amer. Math. Soc. Transl. Ser. 2 (1966) , 52, 23-80. Dirac, P.A.M.,
1930, Note on exchange phenomena i n the Thomas atom,
Proc. Camb. P h i l . Soc. 26, 376-385. Dyson, F.J.,
1967, Ground-state energy of a f i n i t e system of charged
p a r t i c l e s , J. Math. Phys. j, 1538-1545. and A. Lenard, 1967, S t a b i l i t y of matter. I, J. Math.
Dyson, F.J.
Phys. 8, 423-434. C
F e d , E.,
1927, Un metodo s t a t i s t i c 0 per l a determinazione d i
alcune p r i o r e t h d e l l ' atome, Rend. Acad. Naz. Lincei Fock, V.,
5,
602-607.
1930, NHherungsmethode zur Liisung des quantenmechanischen
Mehrktrperproblems, Z e i t . Phys. N 61, 126-148; s e e a l s o V. Fock, "Self consistent f i e l d " m i t aus tausch fiir Natrium, Zeit. Phys.
2
(1930), 795-805. Gombgs, P.,
1949, "Die s t a t i s t i s c h e n Theorie des Atomes und i h r e
Anwendungen", Springer Verlag, Berlin. G r i f f i t h s , R.B.,
1969, Free energy of i n t e r a c t i n g magnetic dipoles,
Phys. Rev. 172, 655-659. /Y
Hartree, D.R.,
1927-28, The wave mechanics of an atom with a non-
Coulomb c e n t r a l f i e l d . SOC. 24, 89-110. r
P a r t I. Theory and methods, Proc. Camb. Phil.
Heisenberg, W.,
1927, Uber den anschaulichen I n h a l t der quanten-
theoretischen Kinematik und Mechanik, Z e i t s . Phys., 43, 172-198. Jeans, J.H.,
1915, The mathematical theory of e l e c t r i c i t y and
magnetism, Cambridge University Press, t h i r d e d i t i o n , page 168. Kirzhnits, D.A., 123.
1957, J. Exptl. Theoret. Phys. (U.S.S.R.)
2,115-
Engl. t r a n s l . Quantum corrections t o t h e Thomas-Fermi equation,
Sov. Phys
. JETP, 2 (1957),
Kompaneets, A.S. (U.S.S.R.)
and E.S.
3l, 427-438.
64-71. Pavlovskii, 1956, J. Exptl. Theoret. Phys.
Engl. transl. The self-consistent f i e l d
equations i n a n atom, Sov. Phys. JEW,
5 (1957),
328-336.
Lenard, A. and F.J. Dyson, 1968, S t a b i l i t y of matter. 11, J. Math. Phys
.2 , 698-711.
Lenz, W.,
1932, ijber d i e Anwendbarkeit der s t a t i s t i s c h e n Methode auf
Ionengitter, Zeit. Phys. 77, 713-721. Lieb, E.H.,
1976, Bounds on t h e eigenvalues of t h e Laplace and
Schroedinger operators, Bull. Amer. Math. Soc., i n press. Lieb, E.H. and J.L. Lebowitz, 1972, The c o n s t i t u t i o n of matter: existence of thermodynamics f o r systems composed of e l e c t r o n s and n u c l e i , Adv. i n Math.2,
316-398.
See a l s o J.L.
Lebowitz, and E.H.
Lieb , Existence of thermodynamics f o r r e a l matter with Coulomb forces, Phys. Rev. L e t t . Lieb, E.H.
2 (19691,
631-634.
and H. Narnhofer, 1975, The thermodynamic l i m i t f o r
jellium, J. S t a t . Phys. h 2 , 291-310.
Erratum:
J. S t a t . Phys.
3
(19761, No. 5. Lieb, E. H. and B. Simon, 1973, On s o l u t i o n s t o t h e Hartree-Fock problem f o r atoms and molecules, J. Chem. Phys. a longer paper i n preparation.
5,735-736.
Also
21.
.
Lieb ,E .H. and B Simon, 1975, The Thomas-Permi theory of atonp, molecules and s o l i d s , Adv. i n Math., i n press.
See a l s o E.H. Lieb
and B. Simon, Thomas-Fed theory r e v i s i t e d , Phys. Rev. Lett.33
w
(1973), 681-683. 22.
Lieb, E.H.
and W.E. Thirring, 1975, A bound f o r the k i n e t i c energy
of fermions which proves t h e s t a b i l i t y of matter, Phys. Rev. Lett.35
N'
687-689, Errata:
Phys. Rev. L e t t . (1975),
2,1116.
~or'more
d e t a i l s on k i n e t i c energy i n e q u a l i t i e s and t h e i r application, see a l s o E.B.
Lieb and W.E. Thirring, I n e q u a l i t i e s f o r the moments of
the Eigenvalues of the Schrijdinger Hamiltonian and t h e i r r e l a t i o n t o Sobolev i n e q u a l i t i e s , i n Studies i n Mathematical Physics:
Essays i n
Honor of Valentine Bargmann, E.H. Lieb, B. Simon and A.S. Wightman e d i t o r s , Princeton University Press, 1976. 23.
Rosen, G.,
1971, Minimum value f o r c i n t h e Sobolev i n e q u a l i t y
lid3 5 C I I V ( ( ~ ~SIAM , J. 24.
Appl. Math. N 21, 30-32.
Schwinger, J., 1961, On the bound s t a t e s of a given p o t e n t i a l , Proc. Nat. Acad. Sci. (U.S.) 57, 122-129.
25.
26.
Scott, J.M.C.,
1952, The binding energy of the Thomas Fermi atom,
Phil. Mag. ,3.4
859-867.
Sheldon, J.W.,
1955, Use of t h e s t a t i s t i c a l f i e l d approximation i n
molecular physics, Phys. Rev. 27.
S l a t e r , J.C.,
2,1291-1301.
1930, The theory of complex spectra, Phys. Rev.
3,
1293-1322. 28.
Sobolev, S.L.,
1938, Mat. Sb. 56, 471 (1938).
See a l s o S.L. Sobolev,
Applications of functional analysis i n mathematical physics, Leningrad (1950), Amer. Math. Soc. Transl. of Monographs,
(1963).
29.
Sonrmerfeld, A.,
1932, Aspmptotische Integration der Differential-
gleichung des Thomas-Femischen Atoms, Z e i t . Phys, 78, 283-308. 30.
Teller, E . , 1962, On the s t a b i l i t y of molecules i n the Thomas-Fermi theory, Rev. Mod. Phys. 34, 627-631.
31.
Thomas, L.H., Phil. Soc.
32.
1927, The calculation of atomic f i e l d s , Proc. Camb.
5,542-548.
Von Weizsacker, C.F.,
1935, Zur Theorie der Kernmassen, 96, 431-458. /u
CENTRO INTERNAZIONALE M A T W T I C O ESTIVO (c.I.M.E.
)
REPORT ON RENORMALIZATION GROUP
Be TIROZZI
Istituto d i Matematica, U n i v e r s i t i d i Camerino
C o r s o tenuto a B r e s s a n o n e d a l 21 giugno a1 24 $iugno 1976
REPORT ON RENORMALIZATION GROUP
P r o f . Benedetto T i r o z z i I s t i t u t o d i Matematica U n i v e r s i t s d i Camerino
Introduction
1. I n t e g r a l and l o c a l c e n t r a l l i m i t theorems o f p r o b a b i l i t y t h e o r y and t h e
r e n o r m a l i z a t i o n group method. In t h i s l e c t u r e we want t o p r e s e n t t h e problem o f t h e r e s e a r c h o f automodel p r o b a b i l i t y d i s t r i b u t i o n s i n comparison w i t h u s u a l i n t e g r a l and l o c a l c e n t r a l l i m i t theorems. We t h i n k t h a t t h i s approach is i n s t r u c t i v e f o r u n d e r s t a n d i n g t h e main mathematical i d e a underlyng t h i s kind o f problems. Consider a s t a t i o n a r y d i s c r e t e random f i e l d
-=i' k
S=I
E Zi
and suppose t h a t '0 J-I. f i e s t h e i n t e g r a l c e n t r a l l i m i t theorem i f
jj
, zL
. Then
J6
and t h e sequence
t h e random f i e l d w s a t i s -
G-
1.(1)
T h i s theorem was proven i n t h e p i o n e e r works o f Gnedenko (1),(2),(3) mogorov i n t h e c a s e i n which
5
a r e independent and e q u a l l y d i s t r i b u t e d . For
u s it i s more i n t e r e s t i n g t h e c a s e i n which t h e random v a r i a b l e s dependent and more p r e c i s e l y when
and Kol-
lfL,i e2
5,
a r e not in-
form a Gibbs random f i e l d , ( 4 ) ,
(51, corresponding t o a c e r t a i n p o t e n t i a l
+B 5 (
~ , t b ~IUI>A, )
-A36
(Kk&
K ~ E I
1. 2 )
where we suppose t h a t
$6
X,Xbeing
t h e s p a c e o f r e a l i z a t i o n o f t h e Gibbs
random f i e l d . It is w e l l known, from v e r y g e n e r a l arguments, t h a t a n e c e s s a r y and s u f f i i c i e n t c o n d i t i o h f a r 1.1 t o b e t r u e is t h a t t h e f i e l d $ . , ~ b t must b e s t r o n g
mixing ( 2 ) w i t h mixing c o e f f i c i e n t
-1
and t h a t t h e d i s p e r s i o n
O & * C ~ where
C i s some p o s i t i v e c o n s t a n t , f u r t h e r it is r e q u i r e d a c o n d i t i o n of t h e t y p e of Lindeberg ( 1 ) , ( 2 ) , analogous t o t h e one used f o r t h e c a s e o f independent variables:
where
@(@= ~ & , + - - + $ L is ~ ~some
p o s i t i v e c o n s t a n t and
The c o e f f i c i e n t of s t r o n g mixing i s d e f i n e d i n t h e f o l l o w i n g way: l e t q q p ) be a p r o b a b i l i t y s p a c e where t h e random v a r i a b l e s
4
a r e d e f i n e d and l e t u s
c a l l e t h e minimal r - a l g e b r a g e n e r a t e d by t h e e v e n t s
where
A-CA3and
Then we can d e f i n e t h e c o e f f i c i e n t o f mixing by t h e f o l l o w i n g , q u a n t i t y ' : 1. 4)
ac~)w = A€
1 pC~nB)-Pcn) PCB)\
mO_=
86Mz and we w i l l s a y t h a t t h e f i e l d (
*
se,t6b Z)
is s t r o n g mixing if we have t h a t
A'CK)-8
1. 5 )
K 4 0
Now we can show hcw i r i s p o s s i b l e from t h e knowledge o f 1. 5
t h e behaviour of
O.&
.
t o deduce
I n f a c t from a v e r y g e n e r a l theorem ( 2 ) f o r s t a t i o n a r y p r o c e s s e s we have that 1. 6 )
/ ESk54J -c c ~ ( ( i d h )
c,o
And s o we can g i v e t h e f o l l o w i n g e s t i m a t e f o r t h e d i s p e r s i o n of
E~&=o
g
Suppose now t h a t
Z o(Oj4)43
f o r a l l 9 t h e n we o b t a i n suddenly
,B =
6~23
DS--8-4,
1. 8 )
t h a t is t h e r e q u i r e d c o n d i t i o n on
0.c,
The proof o f t h e c e n t r a l l i m i t theorem under t h e h y p o t h e s i s s a i d above proc e e d s i n t h e f o l l o w i n g way. Divide t h e i n t e r v a l
Cd(*]
i n segments
Gl wl
such t h a t
/=P, I
wd/=q= O
f o r a l l i and s o we have t h a t
I t can be shown t h a t t h e random v a r i a b l e s t i c a l l v independent i f
S'=*% L
t6vi
gs
are a a p p t o .
I n t h i s c a s e we o b t a i n t h e f o l l o w i n g a s y m p t o t i c j o i n t p r o b a b i l i t y d i s t r i bution f o r t h e variables
xva - .. , SVj ,. .. 1
while using
1.8
it can be seen t h a t t h e v a r i a b l e s
go i n p r o b a b i l i t y t o z e r o . The i d e a o f t h i s method (method o f d e l e t i o n s ) i s due t o Bernstecn ( 2 ) . I t is p o s s i b l e t o f i n d i n t h e p a p e r s on S t a t i s t i c a l P h y s i c s o r on Probabil i t y t h e o r y many c a s e s i n which t h e c e n t r a l i n t e g r a l l i m i t theorem h a s been d e r i v e d f o r Gibbs f i e l d s w i t h g e n e r a l p o t e n t i a l s g i v e n by
1.2
u s i n g t h e ve-
r y fundamental i d e a s h e r e exposed o r p r o c e d u r e s which a r e founded on them. For example Nakhapitan ( 6 ) h a s g e n e r a l i z e d t h e method o f B e r n s t e i n ' s d e l e t i o n s used i n ( 2 ) f o r a Gibbs f i e l d i n any dimension and t h e r e he used t h e cond i t i o n o f t h e uniform s t r o n g mixing f o r t h e Gibbs f i e l d which he showed t o b e t r u e f o r a c e r t a i n c l a s s o f p o t e n t i a l s . Malishev (7) h a s been a b l e t o show t h a t , if t h e c o r r e l a t i o n s d e f i n e d i n
1.6
o f t h e Gibbs f i e l d , when t h e s p a c e
of r e a l i z a t i o n is f i n i t e , a r e e x p o n e n t i a l l y d e c r e a s i n g , t h e a s s e r t i o n g i v e n i n 1.11 is t r u e : h e u s e s t h e t e c h n i q u e o f e v a l u a t i n g s e m i i n v a r i a n t e s . He i s a l -
s o a b l e t o show t h a t i n t e g r a l c e n t r a l l i m i t theorem h o l d s f o r e v e r y t r a n s l a t i o n a l l y i n v a r i a n t s t a t e s i n t h e c a s e o f I s i n g models f o r
8,
and
6cb.
Then from a theorem shown by R . L. Dobrushin and B. T i r o z z i ( 8 ) it f o l l o w s t h a t a l s o t h e l o c a l l i m i t theorem i s t r u e f o r t h e s e Gibbs s t a t e s . These met h o d s t o g e t h e r w i t h o t h e r founded on p r o p e r t i e s o f a n a l y t i c i t y (9), t h a t t h e i n t e g r a l and l o c a l c e n t r a l l i m i t theorem a r e t r u e f o r
13ubf o r
( 1 0 ) show
p>>
and
a l a r g e c l a s s o f system o f i n t e r e s t o f S t a t i s t i c a l P h y s i c s .
Now we want t o begin t o examine t h e s i t u a t i o n when
P=rpgC. The
first
o b s e r v a t i c r . z o n s i s t s o f t h e f a c t t h a t it i s no more t r u e t h a t
because of t h e s m a l l e r r a t e o f d e c r e a s i n g o f c o r r e l a t i o n s and s o we e x p e c t a n o r m a l i z a t i o n f a c t o r g i v e n by
a s , i r 4 d . Furthermore,
g i v e n a c e r t a i n po-
t e n t i a l we a r e i n r s r e s t e d n o t o n l y i n t h e e x i s t e n c e o f t h e a s y m p t o t i c j o i n t
p r o b a b i l i t y d i s t r i b u t i o n s f o r t h e normed sums o f s p i n s b u t a l s o i n t h e d e t e r -
&. f o r
mination o f
p a r t i c u l a r h a m i l t o n i a n and a l s o it i s i m p o r t a n t t o
a
know how s t a b l e i s
w i t h r e s p e c t t o '!small1' changes of t h e p o t e n t i a l .
A s an example o f t h e s i t u a t i o n o f t h e known r e s u l t s a b o u t t h e s e problems we
f i n i s h t h i s i n t r o d u c t i o n showing an open problem f o r a onedimensional s p i n system w i t h a p a i r p o t e n t i a l g i v e n by
when
1344
Pa.
The q u e s t i o n is t h e f o l l o w i n g : i s , t h e l o c a l c e n t r a l l i m i t theorem v e r i f i e d const i n t h i s c a s e ? we know b u t we have no i n f o r m a t i o m ltL-u" about t h e c o e f f i c i e n t o f mixing.
E[T.J'J-
-
2 , Kadanoff r e n o r m a l i z a t i o-n group and g a u s s i a n automodel d i s t r i b u t i o n s .
We a r e going t o g i v e a more p r e c i s e f o r m u l a t i o n o f t h e above problems and t o d e s c r i b e some examples of l i m i t p r o b a b i l i t y d i s t r i b u t i o n s . Le?
R
be t h e s p a c e of a l l r e a l numbers. We s h a l l c o n s i d e r a random f i e l d
sb,&6zL.' which t a k e s v a l u e s i n R. -
Def i-a ii i -o n 2 . 1 i
d
,
fc
+e
Let u s c o n s i d e r a r e a l i z a t i o n
nzd
. We
3 /fiL,,-
w i l l d e f i n e an endomorphism on
R
H
of t h e random i n such a way:
K ~ Z ' '2.1.
kLJt5)-i
+
= (Aa(d1313= Y
The i r z n s f o r n a t i o n
b i l i t y measures d e f i n e d on
where
C 6 & R~ 4
i.e.
c
i d j o i n t of
&'
L & i4*cL F~~~~ 4
Ad<) a c t i n g
i n t h e s p a c e o f proba-
i s d e f i n e d by
b e l o n g s t o t h e @- a l g e b r a g e n e r a t e d by t h e random
4
hlt6Z.
field
Z~
R
A p r o b a b i l i t y d i s t r i b u t i o n on
Definition 2.2.
is a automodel i f
,L@COCJ p = p=P
2.3.
o r , i n ' o t h e r words,
A:c~)
mat i o n
Remark 1.
AK(,
P i s a automodel i f
it is a f i x e d p o i n t o f t h e t r a n s f o r -
zf
2" is an endomorphism o f
8
in
fi
and was f i r s t i n t r o d u c e d
by Kadanoff ( 1 1 ) . We d e s c r i b e e x p l i c i t y , f o r s a k e o f c l e a r n e s s , t h e t r a n s f o r m a t i o n 2.1. Consider t h e one dimensional l a t t i c e
2
i and suppose. t h a t t o each p o i n t . i of
+he l a t t i c e i s a s s o c i a t e d t h e v a l u e t a k e n by t h e random v a r i a b l e
jd
fig. 2.
Divide t h e l a t t i c e i n b l o c k s o f l e n g t h
-*
variables tor
((
K as in fig.
t h e n sum a l l t h e
2.,
which a r e i n t h e same b l o c k , m u l t i p l y f o r t h e n o r m a l i z a t i o n f a c -
2. g i v e a numeration t o t h e b l o c k s i n t h e f o l l o w i n g way: t h e b l a c k ha-
v i n g a n extreme l e f t s p i n
& has
index
% .
Then t h e t r a n s f o r m a t e d sequence w i l l be Y
where
-.
-
is e q u a l t o t h e weighted sum o f t h e v a r i a b l e s i n t h e block i d e f i n e d cr/
above.
We can f i n d t h e p r o b a b i l i t y measure
P
2.2, Let
2.4.
P
-
PI
be g i v e n by
Gc,
=
zi
.fwfJJ
$6c
d e f i n e d on
~/l&eef 6 R , Fa I
i s t h e p r o b a b i l i t y measure d e f i n e d on -
following measure on c y l i n d r i c a l s e t s
Dq
:
R
a*
using d e f i n i t i o n
+a ifif-, /?
zd g e n e r a t e d by t h e
g,
where
@CA
= { % M = ~ - l - - -
-1
-
$&=X~J 4
Remark 2. T t i s e a s y t o s e s t h a t t h e f a m i l y o f o p e r a t o r s m u l t i p l i c a t i o n s e m i g r o u p o f endomorphisms, t h a t i s
1 &Of
forms a
b ( ~ , 6(Pjt&(d)@):&f
From t h i s simple o b s e r v a t i o n it f o l l o w s e a s i l y t h e e q u i v a l e n c e between d e f i n i tion
2.2
and t h e f o r m u l a t i o n g i v e n i n t h e p r e c e e d i n g paragraph.
I n f a c t t h e skarch of a fixed point f o r
is r e l a t e d t o t h e conside-
r a t i o n of t h e l i m i t
and t h i s . c o r r e s p o n d s t o t h e i n v e s t i g a t i o n o f t h e a s y m p t o t i c j o i n t p r o b a b i l i t y d i s t r i b u t i o n o f t h e . random v a r i a b l e s
KMCju)-d
.
We a r e a b l e t o d e t e r m i n e t h e s p e c t r a l d e n s i t y o f a n automodel s t a t i o n a r y f i e l d
-
( fdz . L e t u s
i n t r o d u c e t h e g e n e r a l i z e d random f u n c t i o n
'f%aid,q&l
is t h e F o u r i e r t r a n s f o r m o f t h e f i e l d & r a n d
(WL if); has t h e fol-
lowing p r o p e r t i e s
We l o o k f o r t h e t r a n s f o r m a t i o n induced on it by Kadanoff's renorm group:
Cd+O K-A
+a *t Calling
y'le) sa * .:
f
3,
Fourier transforming the new field induce a transformation on
?&):
the generalized random function obfained
15:
+@
we have that
a(.( I(c&)-?$[&]. )
Ak(d
ui
We shall find it with the help of very simple calculations. In fact we have
we obtain the following equation of transformation:
S i n c e t h e f ie'ld
w fi f-@*)iss t a t i o n a r y
of t h e generalized p r o c e s s q [ t ) s i t y of t h e f i e l d
t$;$",-
we have t h a t t h e c o r ~ e l a t i o nf u n c t i o n
can be e x p r e s s e d i n t e r m s o f t h e s p e c t r a l den-
T h i s f o l l o w s from elementary c a l c u l a t i o n s ,
where
We have a l s o t h a t 2.12.
lee
From 2.10 function of
we o b t a i n t h e law of t r a n s f o r m a t i o n f o r
yet) t h e
correlation
&Ct)
Let u s look f o r i t s f i x e d p o i n t s . W e choose a s i n p u t f u n c t i o n
We w i l l i n d i c a t e t h e t r a n s f o r m a t i o n 2.13
with t h e symbol
Rh (4) , t h e n
we have t o c a l c u l a t e t h e f o l l o w i n g l i m i t keeping i n mind remark 2 ) :
We have
'rle have a i i m i t o n l y i n t h e c a s e
r = d and ~
s o we o b t a i n t h e e x p l i c i t
expression of
+*
The s p e c t r a l f u n c t i o n o f t h e p r o c e s s 1J.Jgean
be o b t a i n e d by u s i n g
1.11
and 2.12:
We have t h e f o l l o w i n g c o n d i t i o n on
d:i < o ( d t h e
f i r s t l i m i t i s due t o
t h e c o n s i d e r a t i o n made i n t h e i n t r o d u c t i o n , t h e second i s due a l s o t o t h e f a c t t h a t we want t h e d i v e r g e n c e i n t h e o r i g i n t o be i n t e g r a b l e . So we o b t a i n t h a t t h e f u n c t i o n
f'&)
h a s i n t h e o r i g i n a s i n g o l a r i t y of t h e
and t h e c o r r e l a t i o n f u n c t i o n o f t h e p r o c e s s w i l l have t h e f o l l o w i n g b e h a v i o u r a t infinity
-ci-4)-j
&%)
N
161
= 1% pa
3. S t a b i l i t y o f t h e automodell p r o b a b i l i t y d i s t r i b u t i o n s , non g a u s s i a n automox) del distribution
.
In t h e l a s t s e c t i o n we have shown how i t is p o s s i b l e t o f i n d g a u s s i a n a u t o m o d e l p r o b a b i l i t y d i s t r i b u t i o n s : i n f a c t we have c a l c u l a t e d t h e c o r r e l a t i o n
-
f u n c t i o n o f t h e automodel f i e l d s o i f it is g a u s s i a n , it is completely d e t e r mined once we have f i x e d t h e c o n d i t i o n
t $,=o
Now we want t o show t h a t a p p a r e n t l y t h e r e e x i s t a l s o o t h e r s automodel p r .
"In t h i s s e c t i o n a r e exposed t h e p r i n c i p a l c o n c e p t s c o n t a i n e d i n t h e paper (12).
d i s t r . , i . e . non g a u s s i a n a.p.d.
and we a r e a l s o going t o g i v e a method f o r
f i n d i n g t h e s e new t y p e s o f a . p r . d . We w i l l t a k e a p o i n t o f view i n analogy o f t h e c l a s s i c a l b i f u r c a t i o n t h e o r y o f dynamical systems (131, (141, ( 1 5 ) . Let u s c o n s i d e r a g a u s s i a n a.p.d. group
AKw i n
6L
o f t h e Kadanof f ' s r e n o r m a l i z a t ion
t h e one dimensional c a s e . We w i l l d e f i n e t h e " d i f f e r e n t i a l 1 *
o f t h e r e n o r m a l i z a t i o n group aKPfd) a c t i n g i n t h e " t a n g e n t space" t o t h e p r o b a b i l i t y d i s t r i b u t i o n s : t h e n we s t u d y t h e spectrum o f look f o r t h o s e v a l u e s o f of
&A*(<)
O(I
aKA*(d~and we
1 4 6 4 -2 f o r which t h e r e a p p e a r s a n e i g e n v a l u e
e q u a l t o one.
I n t h i s c a s e we e x p e c t t h a t i n t h e neighbourhood of t h e c o r r e s p o n d i n g 0( t h e l o o s e s its s t a b i l i t y and t h a t n e a r t h i s
gaussian a.p.distr.
b4
there
a p p e a r s a new branch o f automodel non g a u s s i a n d i s t r i b u t i o n s . Before going i n t o t h e d e t a i l s o f t h i s c o n s t r u c t i o n l e t u s d e f i n e more e x a c t l y what we mean under s t a b i l i t y i n our c a s e . Let u s t a k e a s an example t h e c l a s s i c a l i n t e g r a l c e n t r a l l i m i t theorem o f P r o b a b i l i t y t h e o r y . Let A2(41 be t h e Kadanoff's r e n o r m a l i z a t i o n group o b t a i ned f o r K=2,
4=&
t h e n we c o n s i d e r t h e t r a n s f o r m a t i o n
&*ti)
a c t i n g on proba-
b i l i t y d e n s i t i e s PC)()
M ' I t i s e v i d e n t t h a t t h e l i m i t ' d i s t r i b u t i o n o b t a i n e d by g p l y i n g t o the field
f
too 5; Jem
+oo
and, i f
13c].,
buted with
ii
must c o i n c i d e w i t h t h e l i m i t d i s t r i b u t i o n of
a f i e l d o f independent random v a r i a b l e s e q u a l l y d i s t r i -
ES~=O then
o f t h e l i m i t law:
(A~[*))~~*M
t h e dispersion,
D s i v 2 coincide with
t h e dispersion
t h e '!fixed p o i n t " of
% Afl
w i l l s a t i s f y t h e following equation:
p0
1
= ~ p c a x - ?(Y) ~ l dy
59 we c a n s t a t e t h a t t h e "manifold" o f t h e i n i t i a l random f i e l d s , t h e p o i n t of which w i l l converge t o t h e p r o b a b i l i : ~ d i s t r i b u t i o n
is determined by t h e d i s p e r s i o n .
applied n times,#+@, In t h e case o f d > P
is
when
3.2
e
t h i s argument d o e s n o t hold and wq w i l l t a k e t h e i d e a
of s t a b i l i t y from t h e c l a s s i c a l b i f u r c a t i o n t h e o r y .
~ e T: t MAN b e a d i f f e o m o r f i s m o f a n-dimensional manifold ( 1 6 ) i n i r s e l f
and l e t
x&M
be a f i x e d p o i n t o f T . ( f i g . 3 ) . Let f ( @ ) m e t r i z e d by
be a c u r v e i n
and l e t
* H
para-
be a n-1 submarr
n i f o l d of
such t h a t if
&6fd -
T'''&+x
then
ry
if~f=~'@f"l
and
d+e
a(@)belongs t o M ' i n t e r s e c t i o n between fig. 3
then t h e r e
g(P) and
is o n l y one M which w i l l
be o b t a i n e d f o r a c e r t a i n v a l u e o f
P.
T h i s s i t u a t i o n i s achieved when, and o n l y when, t h e d i f f e r e n t i a l o f T i n )(has o n l y one e i g e n v a l u e b i g g e r t h a n one, t h e n M ' is a d i r e c t i o n such t h a t T r e s t r i /r/ c t e d on M ' is expanding and T r e s t r i c t e d on M is c o n t r a c t i n g . In t h e c a s e when M i s t h e "manifold" o f p r o b a b i l i t y d i s t r i b u t i o n s , T is t h e r e n o r m a l i z a t i o n group
* [ d ) ,is~ & AK
t h a t is a a . p . d .
"manifold" o f p r o b a b i l i t y d i s t r i b u t i o n s such t h a t
4
We have t h a t M is a
3t AK Id)
a c t i n g on it is
c o n t r a c t i n g and M' i s t h e s e t o f i n i t i a l p r . d i s t r . and c l e a r l y t h e v a l u e o f
P
shown b e f o r e w i l l b e t h e c r i t i c a l t e m p e r a t u r e . I f such a s i t u a t i o n is v e r i f i e d we s h a l l s a y t h a t
6
is s t a b l e , t h u s
&
w i l l l o o s e i t s s t a b i l i t y when t h e r e
w i l l be two e i g e n v a l u e s b i g g e r o r e q u a l t h a n one and t h e n w e e x p e c t t h e appear i n g o f a new branch o f automodel p r o b a b i l i t y d i s t r i b u t i o n s . Now we w i l l e n t e r more i n t o t h e d e t a i l s and g i v e an e x p l i c i t c o n s t r u c t i o n of
We w i l l write f o r m a l expres-
t h e tangent space t o a gaussian aut'.prob.distr.
s i o n s f o r sake of s e m p l i c i t y b u t it i s p o s s i b l e t o g i v e t o them an e x a c t and r i g o r o u s meaning using t h e same procedure as in (17). Let
&
be a g a u s s i a n s t a t i o n a r y a u t . p r , d i s t r .
=
E
~
~
5
!2 ~ i &
5e +
=0
)
on
zs
k
and l e t ~ C C ) ;
J ~ A d~ I
, d
2
be t h e c o r r e l a t i o n f u n c t i o n , where
Thm
~(6)% g i v e n
by 2,18, Define t h e m a t r i x
3' can be w r i t t e n f o r m a l l y a s = e
- 9.6ZS Z a~,ks;'
For d e f i n i n g t h e t a n g e n t space i n
Gd
l e t us consider t h e s e t of s t n t i o i n t h e s e n s e t h a t t h e y a r e ab-
nary p r o b a b i l i t y d i s t r i b u t i o n s "near" t o s o l u t e l y c o n t i n u o u s w i t h r e s p e c t t o it
5:
3.6.
where (Li,jlk,
;
-2 e
it162
aiJj,4e aidj$$-6.L G% ~62'
C a r e r e a l numbers such t h a t : &hljhrr,k+y
The d e n s i t y of
6:
with respect t o
b4 w i l l
34 - J3 3K )e
G).m=a<jtk,
be c o n s i d e r e d a s a v e c t o r of
t h e t a n g e n t s p a c e , s o we w i l l g i v e t h e f o l l o w i n g d e f i n i t i o n s . D e finition -
3.1.
A form o f d e g r e e *
of t h e process
[~;f-Till
be t h e random
v a r i a b l e given by t h e f o l l o w i n g formal sum
where
aQC1., em a r e r e a l
numbers and symmetric, i n
eA,.- ., em
'
The forms o f d e g r e e can i n t r o d u c e i n Definition 3.2. that
form a v e c t o r space which we w i l l i n d i c a t e
3' . We
3* t h e t o p o l o g y g i v e n by t h e convergence o f t h e c o o r d i n a t e s . A s t a t i o n a r y form o f degree*
faC5+*.-,&+
3~
6%) w i l l be a form Q
d o e s n o t depend on ~6
e*m
Z'
The s p a c e o f s t a t i o n a r y forms o f d e g r e e 4 w i l l b e i n d i c a t e d a s c4tl I t is e v i d e n t t h a t i s a c l o s e d subspace o f
such
i&J
44
Y4
A l a r g e c l a s s of s t a t i o n a r y forms o f d e g r e e 4
way. L e t
d(&,,.. dA) b e
can be o b t a i n e d i n t h e f o l l o w i n g
a c o n t i n u o u s symmetric f u n c t i o n d e f i n e d on t h e 4 d i -
mensional t o r u s T o r n s u c h t h a t i t s F o u r i e r s e r i e s is a b s o l u t e l y convergent f o r
.
4
6
%here
Let u s c o n s i d e r t h e form
- 5 ~Z"ii'A'ocMJ
atA,..., e,,., -
f~rm
. Then
it i s e v i d e n t t h a t t h e
T~
belongs t o
t4tJ
3,
. From
t h e c o e f f i c i e n t s o f t h e form
3.9
w:
obzaifi t h s f o l l o w i n g r e p r e s e n t a t i o n f o r
gcw
T h i s means t h a t , i f t h e f u n c t i o n s
&''
c o i n c i d e on t h e diaq,iial of&+:
(4)
M '
l A . cZb l c ' 0 ( d l ) $ Let
414'
ment Q b S of d e g r e e
c&
t h e n t h e y g e n e r a t e t h s same form
be t h e a l g e b r a i c sum o f t h e l i n e a r s p a c e s
& c ~ L ~ d .
i& gm . Every
may be r e p r e s e n t e d a s a sum o f s i m i l a r s t a t i o n a r y forms
M,a:raCw' where Nl
f e r e n t from z e r o .
(461 -
eie-
c d &
i n t h i s sum o n l y a f i n i t e number of terms i s d i f -
0.3.3.
The s p a c e
3(461
w i l l be t h e t a n g e n t space of t h e g a u s s i a n s t a t i o n a r y
distribution. D i f f e r e n t i a l o f t h e r e n o r m a l i z a t i o n group.
*(4,.
Let u s now g i v e some f o r m a l arguments t o i n t r o d u c e t h e d i f f e r e n t i a l o f t h e renormgroup which we w i l l i n d i c a t e a s >K r e a d y given f o r m a l l y t h e r e l a t i o n between b a b i l i t y measure on
@"
A
?4
AKM&
In s e c t i o n 2 we have a l and
. For d e f i n i n g t h e d i f f e r e n t i a l
5
where G is a pro-
o f t h e a d j o i n t opera-
t o r o f t h e r e n o r m a l i z a t i o n group a c t i n g on t h e space o f p r o b a b i l i t y measures l e t u s c o n s i d e r two p r o b a b i l i t y d i s t r i b u t i o n s which a r e "near", introduced i n
3.5.
,
, then
3.6.
t h a t is
&
we have t h a t t h e y t r a n s f o r m i n t h e
f o l l o w i n g way:
where B i s some s e t b e l o n g i s t o Then t h e t a n g e n t v e c t o r i n
where
$=fAd4)'~
aR&rR1l,
can be f o r m a l l y d e f i n e d a s
Thus
A
d e f i n e d a s t h e t r a n s f o r m a t i o n on't'$ 0.3.4.
3.10.
3 K ~ * ~ d ) 4 1 *E =
w i l l be
g i v e n by
C Q ~ ~ ' ~ ~ )
Now we w i l l g i v e more d e t a i l s about t h i s d e f i n i t i o n . For t h i s aim we s h a l l u s e t h e n - s t o c h a s t i c i n t e g r a l d e f i n e d by I t o and a l s o some c o n c e p t s from g a u s s i a n dynamical systems (18), ( 1 9 ) . Let u s c o n s i d e r a g a u s s i a n dynamical system
,
where
zd 8 is t h e s p a c e
o f r e a l i z a t i o n s o f t h e g a u s s i r n random f i e l d
is t h e g a u s s i a n s t a t i o n a r y a u t . p r o b . d i s t r i b u t i o n
d e f i n e d i n 3.4;
SOD
lfi3 4 T is t h e
s h i f t operator:
Then, from t h e t h e o r y o f s t a t i o n a r y s t o c h a s t i c p r o c e s s e s we have t h a t it i s p o s s i b l e t o r e p r e s e n t t h e automodel g a u s s i a n random f i e l d w i t h t h e h e l p o f i t s s p e c t r a l random measure:
2cdLt) ( 1 8 ) ~ = i t ( d ) , & &f
where t h e random measure
-
measure
forms a c o n t i n u o u s complex normal random wherea
is t h e a l g e b r a o f b o r e 1 s e t s b e l o n -
C-il+g-
ging t o
The c o n t i n u o u s complex normal random measure
th), A&
is a system o f
At,..- ,dm6fi v a r i a b l e s Z['i),...r£(Al*)
random v a r i a b l e s such t h a t f o r e v e r y c o l l e c t i o n o f s e t s t h e j o i n t p r o b a b i l i t y d i s t r i b u t i o n o f t h e random is .qaussian w i t h t h e f o l l o w i n g moments:
Let
T*
3.13.
b e t h e a d j o i n t o f t h e o p e r a t o r T, t h a t is I. a c t s on
f (&)
: fro.
it is e a s y t o s e e t h a t
J* ZcdW = 4 "zc&)
3.15. Let(43,p)
b e t h e b a s i c p r o b a b i l i t y s p a c e where a l l t h e complex random vh-
.
riables
2 Cb)
D.3.5.
We w i l l s a y t h a t a complex random v a r i a b l e i s a 2 a i r e f u n c t i o n ( 1 8 )
of
a r e defined
it<&,noK
i f it b e l o n g s t o t h e minimal c l a s s 13 t h a t s a t i s f i e s one
o f t h e two c o n d i t i o n s : ( B l ) When
$
is a complex v a l u e d B a i r e f u n c t i o n o f 1) complex v a r i a b l e s i n t h e
usual sense (B2) I f
a;.
.-,
# [ a ~ d l ) ~ -~Ca*o)
belongs t o B
.
i s a sequence i n B which i s convergent f o r evoery NCRt h e n t h e
l i m i t belongs t o B. D.3.6.
ging
Lt( 2 ) w i l l i n d i c a t e t o LZCJL,3,PI.
t h e t o t a l i t y of t h e Baire functions of
-2
belon-
Let u s c o n s i d e r t h e n-dimensional I t o i n t e g r a l c o n s t r u c t e d w i t h t h e h e l p o f t h e c o n t i n u o u s complex normal random measure
where
%(&,-..rAm)
4 z[A), A 6 2 f
i s such t h a t
I*)jf
3.17.
S ( & ) - . . 9 ~ 2 n $ d & - - a em u
and t h e r e g i o n o f i n t e g r a t i o n i n 3.16 d o e s n o t c o n t a i n s p o i n t s (&.,&,\such that
&&=o
(19).
From t h e t h e o r y o f I t o i n t e g r a l s ( 1 8 ) we have t h e f o l l o w i n g p r o p o s i t i o n s usef u l f o r o u r aims: Let u s d e f i n e t h e s c a l a r product between two random v a r i a b l e s ) , ~ basic p r o b a b i l i t y space
L'C-% 4,PJ
P r o p o s i t i o n 3.2. The system
C IN[&),
?c=0,dr
where f s a t i s f i e s 3.17. P r c ~ o s i t i o n3.3.
where
is$ complete i n
[t..,t* -)?
-
&
using t h e
-. 5
L CZ)
'*8
~ : ~ # & ) ~ &and ) II*(w is
an h e r m i t e polynomial o f t h e r e a l v a r i a -
b l e x. Now we w i l l choose as base i n t h e s p a c e
3
L'[')
m i a l s and we w i l l f i n d a n e x p l i c i t e x p r e s s i o n f o r
hRzi
t h e h e r m i t e polyno-
3.10.
ry
Lemma 3.1.
*
Let
be t h e
defined a s i n
matical expectation
p-
3.9.
a l g e b r a g e n e r a t e d by t h e random v a r i a b l e s f o r some f i x e d k . Then t h e c o n d i t i o n a l mathe-
d&($) I&'~') rr
w i l l be g i v e n by
rV
where
z(d,.?,) is
t h e s p e c t r a l random measure o f
-
1fi]-kx,
and
[g*')~~W~,---~ , A?~~ J@~.)?;' J
-dj [ d d ) , j = i , and
zlri-4.
f+I~).cf~&)/le
-4
-,&
b e i n g g i v e n by ( 2.18. )
Proof. We w i l l show t h e c a s e n = l . We have
- -
k setting
4kiL'
we can compare t h i s e x p r e s s i o n w i t h
and t h u s
and we o b t a i n t h e f o l l o w i n g r e l a t i o n between
32 now r e s o l v s i 3 . - : . - :( 5 . 2 2 . )
I-;-;
,-1111
XK
PldA)
i
5%~
and
i
2ri-k
-d
£C&)
t h e t r a n s f o r m a r i o n cn i n t e r v a l s used
We c a n f i n d a f u n c t i o n
Cf/U)
such t h a t
-
I
where&d)must
b e i n d e p e n d e n t from
t h e s c a l a r p r o d u c t of
where we have choosen
;v
Thus, s e t t i n g
ZbR i ~ZC&J)C~ -~J,b6@ from 6'
3.24.
with
3.25.
c a n b e comvuted u s i n g
d i t i o n we h a v e :
t h e l e f t hand s i d e o f
ZC~) f o r e Y w yA C & + ~ ] . L e t u s make
A
i n s u c h a way t h a t
A
3.22.
t h e a b o v e con-
and o b t a i n
i n t e r s e c t s o n l y one o f t h e
k
d5h&'~
, and
e v a l u a t i n g t h e s c a l a r product on t h e r i p h t
hand s i d e :
from which we o b t a i n
I n t r o d u c i n g t h e new random measL- : ( 2 0 )
i t is p o s s i b l e t o w r i t e
-
yen) = K-
3.30.
4% is
where
i n a v e r y simple form;
3.24.
4
-* W K n )+ .y'(a)
independent w i t h r e s p e c t t o a l l t h e
?@), ~ c C 1-
6q'b)=0
and
Let u s now proof t h e lemma f o r
i
;
Let
us divide t h e interval
i n d i s j o i n t i n t i r v a l s of e q u a l l e n g t h &r[-
I(*" rc* l e t u s c o n s t r u c t the I t o i n r e g p a l f o r M. 3 w i t h a f u n c t i o n if
. where
-
t d D ~
, Then
r , ( n = & ~ £ c a ~Z ) =, & ~ ( L ) * l 4 3
/Gee b
. Then
3.32.
where
Now we can g i v e t h e fundamental r e s u l t found by J a . G , S i n a i ( 1 2 ) . Set
41
F~,+$J and
defined by
where
2
* means sum o v e r a l l t h e n-uples such t h a t
1:s
ti
Then we can e s t a b l i s h Theorem 3.1. Let nal
, X (-4)
t0,US
X(A]
be a symmetric
Let u s c o n s i d e r t h e I t o i n t e g r a l
1)
t h e form
4 9K
3.1.
A
.ffq)=
where f is a symmetrical
. In
. Then
& (q)
L ,(T~)LI'~') 6% (4)
9,A%)
p(=2 L.
TG*4 t
~ ( " s /m62
an eigenfunct ion f o r
From theorem
f u n c t i o n d e f i n e d on t h e dimensio-
, f o r a6
=tQ,1X)
(%+wj]=~*
c1
#
K~
f hi*) r
and we have t h a t
=
-M+d
f u n c t i o n d e f i n e d on Torn
cd)
then
i t is e v i d e n t t h a t f o r n=4 we expect a b i f u r c a t i o n f o r
t h e f o l l o w i n g l e c t u r e s we s h a l l g i v e some methods f o r f i n d i n g
non g a u s s i a n automodel p r o b a b i l i t y d i s t r i b u t i o n s .
Q. Eigenfunction and e i g e n v a l u e of t h e d i f f e r e n t i a l d e f i n e d f o r a Gaussian
&
automodel p r o b a b i l i t y d i s t r i b u t i o n and b.1.
-expansion
*.
A s we have seen i n t h e preceeding l e c t u r e s we have a new branch o f non
g a u s s i a n automodel d i s t r i b u t i o n s f o r v a l u e s o f
4
equalto
2 2
corresponding t o
t h e c a s e n=4. I n t h i s l e c t u r e we w i l l g i v e a n e x p l i c i t c o n s t r u c t i o n o f eigen-
*)'The c o n t e n t of t h i s l e c t u r e is exposed i n a n o t y e t published paper o f P.K. Bleher ( 2 1 ) .
f u n c t i o n s and e i g e n v a l u e s o f t h e d i f f e r e n t i a l o f t h e r e n o r m a l i z a t i o n g r o u p i n such a way t h a t it w i l l be p o s s i b l e t o d e t e r m i n e t h e h a m i l t o n i a n Heff o f t h i s new non g a u s s i a n automodel d i s t r i b u t i o n s u s i n g a n expansion method. The system of r e c u r r e n t e q u a t i o n s t h u s o b t a i n e d a r e r e s o l v e d up t o t h e second c o r d e r . From t h e r e s u l t s o b t a i n e d it is e v i d e n t t h a t t h e nongaussian terms c o r r e s pond t o a d e c r e a s e o f t h e p o t e n t i a l which i s slower t h a n t h a t o f t h e g a u s s i a n term: t h i s q u e s t i o n can be f u r t h e r i n v e s t i g a t e d r e s o l v i n g t h e t h i r d o r d e r o r by a p p l y i n g t h e methods h e r e developed. .We w i l l make u s e o f t h e random p r o c e s s ( g e n e r a l i z e d ) i n t r o d u c e d i n t h e f i r s t lecture
where
Y(t)
is a p e r i o d c c g e n e r a l i z e d s t a t i o n a r y g a u s s i a n p r o c e s s d e f i n e d a s
t h e F o u r i e r t r a n s f o r m o f t h e automodel Gaussian f i e l d ready determined t h e c o r r e l a t i o n f u n c t i o n o f
@(e) i n
2 3i f-tm LI
. Ye
have a l
t h e preeceding l e c t u r e s
where we have found t h a t it is e q u a l t o
where
PC*)
field
i3-3+? -a
is t h e s p e c t r a l f u n c t i o n o f t h e g a u s s i a n automodel s t a t i o n a r y
W? w i l l w r i t e a s t a t i o n a r y
where
,
.f(t&, .-. ta ) Z&
M-form
as
- - +G*)
%lt+,*..t
'*)
I t is p o s s i b l e , b u t d i f f i c o l t , t o show t h e . e q u i v a l e n c e between t h i s kind of approach and t h e one o b t a i n e d by u s i n g t h e I t o i n t e g r a l and we assume t h a t x1 t h i s f a c t is t r u e Sc we assume t h a t it i s p o s s i b l e t o make t h e f o r m a l change between
elf) and
t m ) and t h e n we f i n d t h e r e l a t i o n between t h e one between
&*)
for every
t and Q&)
i s g i v e n by:
E x p l i c i t e x p r e s s i o n f o r QK
For f i n d i n g how t h e (12-form ferential
&~*ld)
the variables
analogous t o
random p r o c e s s independent w i t h r e s p e c t t o a l l t h e
where G f i s t h e c o r r e l a t i o n f u n c t i o n o f t h e p r o c e s s 4.2.
B )(fJ
2~d-40)~ zkk~)
3 ~ tis) a g a u s s i a n
where
&) and
e'lf)
~*(d).
( 4 . 3 . ) t r a n s f o r m s under t h e a c t i o n o f t h e d i f -
we have t o c a l c u l a t e t h e c o n d i t i o n a l e x p e c t a t i o n o f
kf'[@~tl)
under t h e c o n d i t i o n s
6'
f i x e d . Thus
I
re^[@)= @and s i n c e i[tj)i s independent from t h e 6' we g&(tj) l@)=&J(q) and t h e c o n t r i b u t e t o ( 4 . 6 . ) d i f f e -
we have t h a t have t h a t
r e n t from z e r o w i l l come o n l y from t h e terms which c o n t a i n an even number of
Icej) and
s o we have o n l y t o c a l c u l a t e t h e mean o f p r o d u c t s o f
J(q)
and t h i s can be made u s i n g t h e Wick expansion and t h e c o r r e l a t i o n f u n c t i o n
4J&)Jej))
which can be o b t a i n e d by ( 4 . 3 . ) .
XjF.Dinaburg and J a . G . S i n a i have shown t h a t i s p o s s i b l e t o r e p r e s e n t t h e n-form u s i n g t h e same f u n c t i o n s a s h e r e b u t through s t o c h a s t i c i n t e g r a l s . (Unpublished r e s u l t ) .
So we have
:
BY s e t t i n g
&,,(6)aR(0) we
can f i n d immediatalv how it a c t s , i n
f a c t we have:
making t h e change o f v a r i a b l e s
Ilsing t h e e x p r e s s i o n ( 4 . 5 . )
K t i -.$ t)' we
obtain :
f o r a ( t ) and w r i t i n g
we have
where
Setttng
i~ t h e mean on t h e tcrur; Torn:
Stt~l.. .,tw)=?(h..It m1$(t3 -.$[t.) we csn
fi r d t h e r i g e n r t : c t o r s
a & ~ * ( d ) i f we f i n d t h e e i r e n f u n c t i o n o f t h e e a u a t i o n
of
Eigenfunctions of
ri. 3 .
adA46( d )
L e t u s look f o r t h e e i g e n f u n c t i o n s o f
-&
. We
choose t h e i n i t i a l fun-
c t i o n s w i t h i n t h e s e t of f u n c t i o n s which have t h e f o l l o w i n g form
for
(aST
, where
genfunction o f
&
Q
i s a homogeneus f u n c t i o n o f d e g r e e
can b e found u s i n g t h e l i m i t
We have :
R I - ' ~&
rfin)"~~*L
pQ(Q+Zit..yt*+ur~4t)
J,,.,l,=i t=.-@ +a
R+N
we can s e t
=
I f we s e t
-40 n o t i n t e g r a b l e on t h e
s o we o b t a i n l i m I(-'ICY ~kA(~~-,h
&+a
I .kc
ir~ta+aujJ+.+t,,+auiy)
c?h+~jd,..,t+~rj.lm * dcJa~-J$* p". a(t,+4~~,..t.4,+ar ~~ jb ~ + 4 & 7 ~ ~ 4 . .Jr,.rJa=-Q
r h m w e can s a y t h a t
-00
uM cc
i s an e i g e n f u n c t i o n w i t h e i g e n v a l u e For
+a0
'')t('
( * * ~ I - /t*tq') . is
&+mjf-fLf&=~ and
*
J
L k4 F($ C=-cO
we have t h a t Q
If 'HI>%-A
4.16.
a
.,twfl =
ic[&+..+**)~
* f=;&(%+-*+&)
4.15.
surface
o K z
ei-
iW [ Q O C ~ S , ..,tmfi (2
For
fQ
em) . The
M <W-A
we have t h a t
Tends t o i n t e g r a l o v e r t h e t o r u s Torn
iclt*+'.'+t4Q&,..,trnjJ "'I=
Am
K
cQ0 C~L~."/~*)I
thus t h e function
is e i g e n f u n c t i o n o f T
k'
Now we a r e a b l e t o f i n d t h e e i g e n f u n c t i o n s o f eq. (4.11.) the
using t h e f a c t t h a t
a r e i n v a r i a n t w i t h r e s p e c t t h e a c t i o n o f t h e r e n o r m a l i z a t i o n group:
P r o p o s i t i o n 4.1.
The e i g e n f u n c t i o n s of t h e t r a n s f o r m a t i o n (4.11.)
a r e given
by
1) for
/VH>(CI-i
with e i g e n v a l u e
-u+*-a++ K
a
2) t h e r e i s a l s o a n o t h e r e i g e n f u n c t i o n
fm[tA,..,t4)=T&
Remark. Note t h a t t h e g l o b a l t r a n s f o r m a t i o n
*
&A*(@)
is t ~ i a n g u l cs o
t h a t i t s g l o b a l p r o p e r t i e s a r e determined by t h e e i g e n v a l u e s and t h e e i g e n functions of t h e transformation
3!A COO.
we can a l s o n o t e t h a t it is pos-
s i b l e t o f i n d a l a r g e r c l a s s o f e i g e n f u n c t i o n s i f we a l l o w t h e & ( t ~ , - - , % ) t o be g e n e r a l i z e d f u n c t i o n s . Now we make a s e l e c t i o n w i t h i n t h e s e t o f t h e e i g e n f u n c t i o n o f p r o p o s i p i o n ( 4 . 1 . 1 , because we want t h a t t h e e f f e c t i v e harniltonian o f d e g r e e
t o s a t i s f y some p h y s i c a l p r o p e r t i e s , where it is e a s y t h a t
with
&
o f prop. ( 4 . 1 . )
by t h e f o l l o w i n g e q u a t i o n :
em
/M
i s connected
.
The r e q u i r e d p r o p e r t i e s a r e 1 ) The e f f e c t i v e h a m i l t o n i a n must be t r a n s l a t i o n i n v a r i a n t
Hm must
2)
be p o s i t i v e d e f i n i t e i n o r d e r t o g e n e r a t e a Gibbs measure
3 ) t h e p o t e n t i a l c o r r e s p o n d i n g t o t h e h a m i l t o n i a n s w i t h more t h a n two p a r t i c l e s
must d e c r e a s e f a s t e r t h a n t h e p o t e n t i a l coming from t h e g a u s s i a n term. Thus
+CP
From (4.21. ) it i s e v i d e n t t h a t t h e s i n g u l a r i t y o f must c a n c e l w i t h t h e b e h a v i o u r o f Let u s e v a l u a t e
Thus
((6)
$(t) CC
in the origin
]*-it
i n the origin
~5(~~'i(c4(T -f)-' iti (r
1 t l -A -q
and
t h u s we f i n d t h e e x p r e s s i o n f o r t h e e i g e n f u n c t i o n o f
ad~ ' ( 4 )
which have a l r e a d y been determined i n ( 1 2 ) . 4.4.
Compact formulae f o r t h e f u l l d i f f e r e n t i a l .
Now we a r e a b l e t o g i v e a r e p r e s e n t a t i o n o f t h e a c t i o n of t h e f u l l d i f f e r e n -
rial
& A*(H~.
We s h a l l Ejee how it a c t s on t h e s p a c e o f forms of t h e t y p e
where
e,,&~#.-~%)i s g i v e n by
(4.24. )
. It
i s u s e f u l t o u s e now o t h e r random
v a r i a b l e s d e f i n e d by
w (t)=
s f
-I e-it&
The t r a n s f o r m a t i o n induced on
0,
w[ej
by t h e p r e c e e d i n g f o r m u l a s ( 4 . 4 , ) ,
where
XCt)
by t h e a c t i o n o f
is a complex g a u s s i a n random p r o c e s s whose c o r r e l a t i o n f u n c t i o n is
~ ( twe)have
and is independent from
H(i)
a n o t h e r form f o r
...,ta)
where
isgivenby
r
4.29.
%,Ct.,...l
t4 1,
r,
Tf3t,+tnjK, K r i
j~,--.,j~
Thus
4.30.
is determined
(4.5.)
determined by t h a t o f W C * ) ~ U ) ' C k t ) Using
AK(Q()
+@ tff +lr f-l ( w )= (t rrl" (dr [dt, - [&d&.I--tuC&) em
-7f
-?i
wt(*f).
Let u s d e f i n e t h e f u n c t i o n
*(O()
So t h a t t h e e i g e n v e c t o r o f
We can
where
determine now how
>Ic gK)KA(4ja c t s
ex)is some polynomial.
We c a l c u l a t e ( 4 . 3 4 . )
w i l l have t h e form
on t h e more g e n e r a l form:
We have
i n t h i s way. S e t
, from
-7T -.
well known theorems ( 1 8 ) X w i l l be a complex g a u s s i a n random p r o c e s s whose var i a t i o n i s determined by t h a t of
Thus we can s e t :
~ ( t ,)and
wl[f) :
Let u s make t h e s u b s t i t u t i o n kt-7
I(
P, ('1%
since
+
Fj)=e&(~.,~t) ,
t/J(+?& and s e t t i n g ~
formuh f o r
4.39.
F
we h a v e :
-w
-Q)
changing
t
:
( % ] = ~ d [ ~ we ) o b t a i n a t r a s f o r m a t i 2 n
--xz
FJ~~,zI=
Choosing
F(x)
pending on
f
t o be a polynomial i n t h e v a r i a b l e
3
, with
c o e f f i c i e n t s de-
we c a n r e w r i t e (4.39. ) i n t h e f i n a l form:
.
w h e r e ~ ~ h ~In ' t~h i s way we have reduced t h e problem o f t h e a c f i o n o f t h e d i f f e r e n t i a l of t h e renormalization group
&A*&)
t o t h e problem o f s t u -
dying t h e p r o p e r t i e s o f t h e i n t e g r a l operator (4.40.). 5. E i g e n f u n c t i o n o f t h e d i f f e r e n t i a l of t h e r e n o r m a l i z a t i o n arouD.
It is ~ o s s i b l et o s t u d v t h e ~ r o ~ e r t i eo fs t h e o D e r a t o r ( 4 . 4 0 . ) . evaluate
x[c)
:
F i r s t we s h a l l
-
From t h e d e f i n i t i o n (4.27. ) of
P u t t i n g (4.42.)
Setting
X[t)
we have t h a t
-
ir.to ( 4 . 4 1 . ) we o b t a i n :
where
r way i f is p o s s i b l e t o show t h a t 4.46.
4 = P-'~(IZ)
and s o we o b t a i n
( Y [ ~ ) = ~ [ K Z ) ~ K ~ - ' ~ ( ~ )
Now it is p o s s i b l e t o f i n d t h e e x p l i c i t e x p r e s s i o n s o f t h e e i g e n f u n c t i o n s o f
pK4+(d) . Let
u s d e f i n e b e f o r e t h e h e r m i t e polynomials:
Then : Proposition 4.2.
The
functions
~(42)'-&&;qf~)) a r e
t h e transformat ion (4.40.) with eigenvalue equal t o
eigenfunctions of
&
Proof. The g a u s s o p e r a t o r
4
.
I -(D
t r a n s f o r m s a n h e r m i t e polynomial
c'? & (~;c(u-B))if T ~ U St r a n s f o r m a t i o n
ction
&, ( 2 , ~ )
i n a h e r m i t e polynomial
a7 6
( 4 . 40. ) t r a n s f o r m s t h e f u n c t i o n
R [ z i 7(a))
i n t h e fun-
KC-O&C~;C(~~K~)-X~ILI~= - K K- mf24?. * fZR/-Kd-tpq~'t))=~
-
The p r o p o s i t i o n is proven.
Remark. P r o p o s i t i o n ( 4 . 2 . ) i m p l i e s t h a t t h e form
is e i g e n f u n c t i o n of
& A (k' )
with e i g e n v a l u e
%-w+i 6%
. F u r t h e r ff$,
form an o r t h o g o n a l system o f random v a r i a b l e s w i t h r e s p e c t t o t h e g a u s s i a n p r o b a b i l i t y measure corresponding t o t h e automodel g a u s s i a n s t a t i o n a r y random field.
5. The
&
-expansion.
I t i s p o s s i b l e now t o g i v e a n a l g o r i t h m which a l l o w s t o f i n d t h e e f f e c t i v e h a m i l t o n i a n c o r r e s p o n d i n g t o a non g a u s s i a n automodel p r o b a b i l i t y d i s t r i b u t i o n . Let u s w r i t e t h e e f f e c t i v e h a m i l t o n i a n i n t h i s way / # ~ ) = ~ ~ i ~ ( U ) a nexpand d t h e corresponding measure i n power o f
where
where
N O is a
&
:
g a u s s i a n a u t . p r . d i s . . Let u s a p p l y
&A'@)
we have t a k e n o n l y t h e t e r m s up t o t h e second o r d e r i n
to
6
because we
a r e going t o f i n d an e x p l i c i t form o f t h e e f f e c t i v e h a m i l t o n i a n o n l y t o t h i s degree of a c c u r a c y . Thus we have t o f i n d
$(g
i n such a way
, BdW)
that:
We a r e s t u d y i n g t h e e f f e c t i v e h a m i l t o n i a n i n a neighbourhood of expected t o be a b i f u r c a t i o n p o i n t and s o If
d= 3
w i l l be d e f i n e d by
which i s
E=d- 2 t
.
t h e s o l u t i o n o f ( 5 . 3 a . ) should be w i t h e i g e n v a l u e 1 but s i n c e
N C Es t €=a-23
we can s t a t e
and s o we r e s o l v e t h e e q u a t i o n ( 5 . 3 a .
up t o t h e accurac.y
and p u t t i n g i n t o t h e e q u a t i o n (5.3b.1 t h e term
OCE)
OCg)
, setting
it w i l l a p p e a r a s :
where L is d e f i n e d by
The
aim o f t h i s s e c t i o n is t o f i n d a s o l u t i o n o f ( 5 . 6 . ) .
The space o f forms where we s h a l l l o o k f o r a s o l u t i o n of ( 5 . 6 . ) i s determined by t h e s t r u c t u r e o f t h e o p e r a t o r
LC&&])
. Let
u s s t u d y it i n some d e t a i l :
we u s e t h e r e p r e s e n t a t i o n o f t h e p r e c e e d i n g s e c t i o n :
tr
( t i t C ( 4 q ~and /%$a) -It c o e f f i c i e n t s depend on 2 . We have: where T[%)
3K~%40() HC ' WJ = E
+
w
whose
X
d
( $ dhldca , , F ( -rf[ ~ * s~,ftt,~~)K-'w't~k)+ -09
+-nI+" d f c t ~ d l * d , h - ) ~ ( t l ) ) .~
5.8-
is some polynomial i n
.(JT
-9 ( r ~ d t ~ & l t ~ , t ~ ~ -n
We can proceed i n an analogous way a s have done f o r f i n d i n g t h e compact form of t h e e i g e n f u n c t i o n s of t h e d i f f e r e n t i a l , t h e o n l y d i f f e r e n c e is t h a t now we w i l l make t h e c o n d i t i o n a l e x p e c t a t i o n w i t h r e s p e c t t o t h e by
and
-lr
we o b t a i n
where
and
-algebra generated
&=I d?h/&(t&k)X[t%) %=~t~(,&a&(*',W x&) -IT +JT
1%
,+a is a m a t r i x g i v e n by
and
making a n a l o g o u s c a l c u l a t i o n s a s b e f o r e w e o b t a i n
where
ltfq-fi)po(e) p,(zrrq-t) 8
yl%,q)
6w
making t h e i n t e g r a l s w i t h r e s p e c t t o
Xa 1 &
dt
and u s i n g t h e theorem
o f moments of g a u s s i a n s y s t e m o f v a r i a b l e s ( 2 0 ) ( w i c k t h e o r e m ) we o b t a i n a sltm
o n l y on t h e c o n n e c t e d g r a p h s b e c a u s e t h e d i a g o n a l s t e r m s o f
elpal t o
YCK~Z,)
.
are
From t h i s r e p r e s e n t a t i o n we car, a l s o d e d u c e t h a t we need
t o c o n s i d e r a s p a c e of forms o f t h e type:
R e p e a t i n g t h e c a i c u l a t i o n s made b e f o r e we o b t a i n t h a t t h e e i g e n f u n c t i o n s o f t h e d i f f e r e n t i a l i n t h e s p c e o f forms o f t h e t y p e ( 5 . 1 2 . ) a r e t h e multidimens i o n a l h e r m i t e polynomial;.
We g i v e t h i s s t a t e m e n t i n more p r e c i s e terms: P r o p o s i t i o n 5 .l. L e t
q(~, Y ) be d e f i n e d
a s i n (5.11. ) and l e t
c o r r e s p o n d i n g m a t r i x . Consider t h e g a u s s i a n random f i e l d sion matrix
ft : i (LCJ t~(ti) + -.+fin) nted
)=e
~ ( e Qr : :net w . . VC&)
5.13.
Let
6 ij
~(3 with
-2 uc& e,m=1
be t h e Wick polynomial o b t a i n e d
by
be t h e disper-
m,L) %(kJ ~'h,,..
making t h e o r t o g o n a l i z a t i o n w i t h r e s p e c t t o t h e g a u s s i a n d i s t r i b u t i o n (5.13.). Then
where
191 zG+-**+4*
The o r t o g o n a l i z a t i o n is j u s t t h e same a s used i n e u c l i d e a n quantum f i e l d theory:
t h e g r a f i c s designed above must be i n t e r p r e t e d w i t h t h e h e l p o f t h e f o l l o w i n g explanations:
-
t o every vertex corresponds a p o i r i t a
- t o every l i n e going -
i n t o the vertex
Jti
t h e l i n e j o i n i n g two v e r t e x e s c o r r e s p o n d s t o
Then t h e d e f i n i f ion of:
nq'&)
-
c o r r e s p o n d s a random v a r i a b l e r(ti)
-..flm (q*):
~ ~ I z ~ ) T ( ~ > = ~ i s g i v e n by
where t h e sums go o v e r a l l t h e p o s s i b l e g r a p h e s o b t a i n e d j o i n i n g p a i r o f l i n e s of t h e p i c t u r e above. We can make u s e o f P r o p o s i t i o n ( 5 . 1 . ) tion (5.6.).
In f a c t we can w r i t e :
where we have s u p p r e s s e d t h e i n d e x
f o r s o l v i n g equa-
.
b u t t h e Wick polynomials must be
understood i n t h e s e n s e of P r o p o s i t i o n ( 5 . 1 . ) .
Thus we c a n w r i t e
We can a l s o expand t h e q u a d r a t i c o p e r a t o r i n ( 5 . 6 . )
with t h e help of hermite
polynomials. I n f a c t we have
we c a n expand
9:
i n t e r m s o f ;he
eigenfunctions
of
gK~IC@)
It is p o s s i b l e t o s e e t h a t t h i s sum i s e q u i v a l e n t t o make t h e g r a f i c expansion
o f b e f o r e and keeping o n l y t h e connected termd. Thus it f o l l o w s from p r o p o s i t i o n ( 5 . 1 . ) t h a t :
and
The integralJY4(~h,kca)dbidta
tv
Of
will
‘r
d i v e r g e s because o f t h e p e r i o d i c proper-
=
: '#cu~+~,&+L) -l'(Cs%&)
, but
it g i v e s a c o n s t a n t which
g i v e no c o n t r i b u t i o n t o t h e p r o b a b i l i t y d i s t r i b u t i o n . T h i s d i v e r g e n c e
a r i s e s because we c a l c u l a t e d i r e c t l y t h e e f f e c t i v e h a m i l t o n i a n which is a sum o f some p o t e n t i a l o v e r a l l t h e l a t t i c e p o i n t s : t h e same c a l c u l a t i o n f o r t h e pot e n t i a l would have g i v e n a f i n i t e term. T u t t i n g (5.18.1,
(5.19.),
(5.15.)
i n t o eq. ( 5 . 6 . )
and n e g l e c t i n g t h e term
we o b t a i n t h e f o l l o w i n g e q u a t i o n s :
6
CY,~)
h t , w K
z-5[44)
(53)
-5
=
&,fit)
w I)K
i6af (YCKCA,K~E)-LYC& z*ca-Nq(,t =isa:[~
putting
OC&) i n t h e h i g h e r F,'3,3'
-
444,
-q(%~2))+ OCbI
o r d e r e q u a t i o n s we o b t a i n t h a t
[2a,24)
(TA,O*)=
=
16 (Y(%,h.t)
G~
7 2 ~ ' C ~ * , T Za)
'
P r o p o s i t i o n 5.2. The f u n c t i o n
(kte&)
belongs t o
flfl')
is r e a l , p e r i o d i c and
symmetric and h a s t h e f o l l o w i n g a s y m p t o t i c
4-t Proof.
The p r o p o s i t i o n f o l l o w s immediately from t h e d e f i n i t i o n (5.11. ) o t
Ute. P, \
. In
f a c t t h e Fourier transform o f
I t is s u f f i c i e n t t o look a t t h e b e h a v i o u r o f
From which it f o l l o w s t h a t :
y.
c"%t
C.4'l&,~~ ) is
bC.4)
g i v e n by
in the origin
From t h e d e f i n i t i o n of t o be i n t e g r a b l e on
R'
fa
( 5 . l Q b . ), it i s c l e a r t h a t we need
484
L%et)
and t o have s i n g u l a r i t i e s which a r e i n t e g r a b l e on t h e
plane.
i t s behaviour
F i r s t w k n o t e t h a t we caw s u b t r a c t from a t large
1%;
)
and t h e e q u a t i o n s o f t h e t y p e ( 5.20. ) w i l l b e s a t i s f i e d
j u s t t h e same. I n f a c t t h e r i g h t hand s i d e of (5.20.) w i l l n o t change.
K
~16 (ktod ~ -(#C,KQ~, ~ K O ~ -QC%i )
1%)
)=
5.23.
we need t o s u b t r a c t t h e behaviour a t l a r g e because
where Reg
only f o r
-
(33) , -
(4~)
q -
h a s a s u f f i c i e n t l y good d e c r e a s e p r o p e r t y a t i n f i n i t e .
But we have t h a t
5 ,z1
/%A-&)
~C"l'
now has, a non-integrable s i n g u l a r i t y
for
a s a consequence of t h e s u b t r a c t i o n . So we s h a l l s e t
is a g e n e r a l i z e d f u n c t i o n d e f i n e d by
-00 Now it i s p o s s i b l e t o show t h e following i d e n t i t y between g e n e r a l i z e d f u n c t i o n s
Which f o l l o w s from t h e d e f i n i t i o n :
Thus
cx.)
i
'
:k4"(fi8{
/ ~ 1 4 ' 2 dftv(*))+Z~"[qC0I k-5
5.27.
=('-@K
/*/MA
b8/L--
/~+-u
+-?
p - 4 -
( i
2d-3
- k3-a ) (7~x4,LPC*))
From (5.24.) and t h e f a c t t h a t neglected i n equation (5.20.)
-w2&w
i-K
3-44
Zol-3
6
we o b t a i n t h a t t h e t e r m
i s e q u a l t o t h e c o n t r i b u t e g i v e n by t h e
c t i o n t o t h e l e f t hand s i d e o f (5.6.)
and s o t h e
-
if
-expansion is c o m p l e t e l y r e s o l v e d up t o t h e second o r d e r .
Now we a r e g o i n g t o i n t e r p r e t t h e s e r e s u l t s i n term o f t h e e f f e c t i v e hamiltonian f o r t h e f i e l d
fa0 1 tlf-* .
We have found t h a t t h e non g a u s s i a n term o f t h e h a m i l t o n i a n h a s t h e f o l l o w i n g
form
Let us.-write
(5.28.)
We o b s e r v e t h a t
i n terms of
y'b) -
5.29.
where
S ( ~ I #. S~i n c e we have a l r e a d y s u b t r a c t e d
o f (5.28.)
from t h e f i r s t two terms
t h e g a u s s i a n b e h a v i o u r we compare o n l y t h e t h i r d term of ( 5 . 2 8 . )
w i t h the g a u s s i a n term. W e w i l l w r i t e t h e l a s t term o f (5.28.)
i n t h e f o l l o w i n a u s e f u l form:
we want t o compare t h e c o n t r i b u t i o n t o t h e e f f e c t i v e i n t e r a c t i o n between two spins
jplfPW
due t o , t h e ' n o n . gaussian:term
w i t h - t h e one-due t o - t h e
g a u s s i a n term. We w r i t e t h e l a s t one i n t h e f o l l o w i n g way:
For t h i s aim we have t o s t u d y t h e a n a l i t i c i t y p r o p e r t i e s o f f o r a l l t h e values of
. Thus we
/qS-
+(&,&)
have t o s t u d y t h e a n a l i t i c i -
t y p r o p e r t i e s o f t h e s e r i e s (5.22.). Let u s s t u d y b e f o r e t h e term i n (5.22.) w i t h
5.32.
q[ua,u*] = -3L- (luL1 -d-s +&, 1c(,/2Hz d+d
where
$o
~UI)
loped t h e f a c t o r
f15f Ut=0 M 4 ) ) -%
go(ar)dhIUt2
is an a n a l y t i c a l f u n c t i o n o f
(L+
IUlldtfg (llr))-id) *
UA
= $yug) , and
-- 1 *'
we have deve-
i n Taylor s e r i e s in t h e nei-
ghbourhood o f t h e o r i g i n , t h u s we can w r i t e
The second term g i v e s n o s i n g u l a r i t y i n t h e o r i g i n because it is a n a l y t i c a l while t h e f i r s t and t h e t h i r d w i l l g i v e
The s i n g u l a r i t y . a r i s i n g by t e r m s where
ffafu+=2~ k
gives a contribu-
t i o n no l a r g e r t h a n
a s it i s p o s s i b l e t o s e s from e l e m e n t a r y c o n s i d e r a t i o n s . Thus t h e main c o n t r i b u t i o n t o t h e e f f e c t i v e i n t e r a c t i o n p o t e n t i a l comes from
. ' e f i r s t term i n (5.33.
) 'and we have t h a t
which must be compared w i t h t h e behaviour o f t h e g a u s s i a n term
-' .
( 5 . 3 6 . ) g i v e n a p o t e n t i a l which d e c a y s a s which is slower t h a n
4-4
, while
(5.35.)
M-6
g i v e s +l
F u r t h e r i n v e s t i g a t i o n can be made i n two pos-
s i b l e directions: A ) To s t u d y t h e h i g h e r o r d e r e q u a t i o n s o f t h e
B) r e g u l a r i z e a l s o t h e t h i r d term o f ( 5 . 2 8 . ) .
E
-expansion
I n t h i s c n s e we would o b t a i n t h e
c o r r e c t behaviour o f t h e i n t e r a c t i o n p o t e n t i a l and a l s o t h e u n i c i t y of t h e s o l u t i o n o f eq. ( 5 . 6 . ) and s o B seems t o be t h e most f r u i t f u l way t o i n v e s t i gate.
6.
Some new r e s u l t s f o r s p i n systems
*1
Let u s c o n s i d e r a d-dimensional l a t t i c e system, l e t
g&)
be t h e c o r r e l a t i o n
f u n c t i o n of a Gaussian automodel random f i e l d
where Then t h e Gaussian a u t . p r . d i s t r .
can be w r i t t e n a s a Gibbs d i s t r i b u t i o n i n t h e
" ~ e c e n t l ~E. 1.Dinaburg and Ja.G. S i n a i proved t h a t for i 4dcg ' t h e gauss i a n automodel d i s t r i b u t i o n a p p e a r s a s t h e l i m i t p r o b a b i l i t y d i s t r i b u t i o n a t
p=w
t
f o r some one-dimensional t r a n s l a t i o n a l l y i n v a r i a n t system e with t h e long range p o t e n t i a l & I % )
-,,
form :
d>t
I t is ~ o s s i b l et o find,
for
l y t i c a l f u n c t i d n and s o
C(C%)
a c a s e when
l i m i t distribution
c(<2
t f o r [3=pur.
is a r e a l ana-
is a f u n c t i o n e x p o n e n t i a l l y d e c r e a s i n g
u LEI b C ~ s p tA . &-. I t is probable t h a t , f o r
f'l~)
A'C' such a p r o b a b i l i t y d i s t r i b u t i o n a p p e a r s a s
f o r a l a r g e c l a s s of s p i n systems with s h o r t
range i n t e r a c t i o n p o t e n t i a l .
7. Automodel p r o b a b i l i t y d i s t r i b u t i o n s i n t h e c o n t i n u o u s c a s e
z)
7.1. Gaussian c a s e . D.7.1.
Automodel p r o b a b i l i t y d i s t r i b u t i o n s .
L e t P b e a p r o b a b i l i t y measure d e f i n e d on t h e s p a c e o f g e n e r a l i z e d f u n c t i o r s
fi) 7.1.
If
on
R
. Consider t h e
following transformation:
{leLe)-u~-Kfl**) * ~ C O J ~ ) and
1oK@e)
have t h e same p r o b a b i l i t y d i s t r i b u t i o n s , t h e n P
"In t h i s s e c t i o n a r e g i v e n some r e s u l t s o b t a i n e d by R.L.Dobrurhin. s u l t s a r e c o l l e c t e d i n a p a p e r which must b e p u b l i s h e d .
These r e -
i s a n automodel p r o b a b i l i t y d i s t r i b u t i o n w i t h parameter K. It is p o s s i b l e t o
-
e s t a b l i s h t h e followine connection with t h e d i s c r e t e case. Suppose t h a t
f
is t h e s p a c e of. t e s t f u n c t i o n s which must s a t i s f y smoothness
p r o p e r t i e s . Consider t h e s p a c e o f f u n c t i o n s
M
which c o n t a i n s f
and t h e
c h a r a c t e r i s t i c f u n c t i o n s o f t h e u n i t s q u a r e s . I f it is p o s s i b l e t o extend t h e integral
J - ~ L ~ ~ c Qdt c.~I
t o t h e space
fowing d i s c r e t e random f i e l d : f o r each
where
flU
t€Rv
If
w i t h parameter K t h e n s
u&zV
is t h e u n i t s q u a r e w i t h c e n t e r i n t h e p o i n t
P r o p o s i t i o n 7.1.
danoff I
t h e n we can d e f i n e t h e f o l -
is automodel i n t h e s e n s e o f D.7.1.
4 & , ~ l t*f
r e n o m a l i r a t i o n group w i t h
&
is automodel w i t h r e s p e c t t o t h e Ka-
d = 2-
2
Gaussian automodel gen.
random f i e l d . A Gaussian g e n e r a l i z e d r.f. is d e f i n e d by means o f its s p e c t r a l measure
where
Q3
$&*)
i s a measure on
is t h e u n i t s p h e r e ,
$
R~ . Let
u s introduce t h e polar coordinates:
t h e u n i t vector,
c(GZ~,W>
then(&-,g*(d;g
Then : P r o p o s i t i o n 7.2.
A g e n e r a l i z e d g a u s s i a n random f i e l d
i s automodel when i t s
s p e c t r a l measure is g i v e n by
where
$
is any measure on t h e s u r f a c e o f t h e u n i t s p h e r e &
I t i s p o s s i b l e t o c o n s t r u c t a u t . d i s c r e t e g a u s s . random f i e l d n e r a l i z e d g a u s s . f i e l d . T h i s c a n be done when t h e measure
$(e),e~
Qt,
such t h a t t h e r e e x i s t
constants
C,
C >O
from t h i s gehas density f o r which
then it is p o s s i b l e t o c o n s t r u c t a d i s c r e t Gauss. r a n . f i e l d i f
OCkCt).
A very important example is t h e i s o t r o p i c gauss. r a n . f . f o r which
A f t e r t h e d i s c r e t i z a t i o n it is o b t a i n e d a c l a s s o f d i s c r e t Gaussian f i e l d s which c o n t a i n s t h a t o f S i n a i . The two c l a s s c o i n c i d e o n l y when
$
h a s a den-
sity. It i s important t o emphasize t h a t i n t h i s kind o f approach it is considered
a s p e c i a l c l a s s o f gen. random f i e l d . T h i s is determined choosing t h e t e s t func t i o n t o belong t o that i f
with
is t h e s p a c e o f smooth f u n c t i o n s such
where then:
Cfe&
-- 4f e
JLt- t
.Such a p r o c e s s i s s a i d t o be d e f i n e d "over
%
".
I t can be shown t h a t t h e c o n d i t i o n introduced on t h e t e s t f u n c t i o n s i m p l i e s
t h a t we c o n s i d e r random f i e l d s w i t h independent increments ( 2 2 ) . The u s e of such f i e l d s is u s e f u l because it i s p o s s i b l e t o a l l o w more c r i t i c a l behaviour in t h e origin f o r
6@ecr 6 E )
i n o r d e r it t o be g a u s s i a n .
In t h i s c a s e i n f a c t t h e f o l l o w i n g c o n d i t i o n must b e s a t i s f i e d :
Using t h e same concept t h e f o l l o w i n g r e s u l t s can b e o b t a i n e d : l e t
.'q3\.a
be
t h e Wick polynomial made w i t h r e s p e c t a gauss.gen. f i e l d t h e n i n t h e l i m i t
23 4) 7.2. Let
i n (7.1. )
:Y 'w :2
-
converges t o t h e m a s s l e s s g a u s s i a n f i e l d .
Non g a u s s i a n c a s e
f&)
b e a gen. rand. f . and P t h e corresponding s t a t e , t h a t is t h e
p r o b a b i l i t y measure d e f i n e d on t h e s p a c e o f g e n e r a l i z e d f u n c t i o n s Let u s c o n s i d e r a c l a s s o f f u n c t i o n a l s d e f i n e d on suppose t h a t t h e y belong t o
L.?(P)
. tie
9' .
f' /by,~ e f ' j
can d e f i n e i n
P(P)
and
and a l s o i n
-
t h i s c l a s s o f f u n c t i o n a l ' s a s h i f t t r a n s f o r m a t i o n induced by t h e s h i f t t r a n s f o r mation of
J(f)
Then we s a y t h a t t h e n e n e r a l i r e d r . p r o c . orinducedbygqif
f
kx)
is subordinated to)&')
isgivenby
I t can be shown t h a t t h e c l a s s of g e n . r . p r .
can be r e p r e s e n t e d u s i n g t h e I t o i n t e g r a l . Let
induced by a gen. Gauss. r . f .
Z C ~ b)e
t h e s p e c t r a l normal
random measure corresponding t o
Then it is p o s s i b l e t o show t h a t a l l t h e g e n e r a l i z e d random p r o c e s s e s of t h e t y p e (7.9.)
can be w r i t t e n a s :
4
if
4
-,
is 6uch t h a t
Let u s suppose now t h a t t h e gauss.gen.ran. with parameter Then
is automodel i n t h e s e n s of(D.7.1.)
&
P r o p o s i t i o n 7.3.
$
with parameter
if
i s automodel ( n m g a u s s i a n ) g e n e r a l i z e d ran. f i e l d
$
I t is a l s o p o s s i b l e t o r e p r e s e n t
f o r 04x4
$
when t h e Gaussian a u t o -
model prob. d i s t r . h a s parameter
where t h e
a r e bounded f u n c t i o n s .
fm
It is p o s s i b l e t o f o r m u l a t e P r o p o s i t i o n (7.3.)
:
Let u s e v a l u a t e t h e wick polynomial s i a n automodel
*
@
F'roposition 7.9.
then:
&x)
=:
El
$
i n a n o t h e r i n t e r e s t i n g way.
tf .
w i t h r e s p e c t t o t h e gaus-
.t
.
is a non g a u s s i a n automodel g e n . r a n . f .
We n o t e f o r example t h a t d i s c r e t i r i n g
:fix)%,
R o s e n b l a t t p r o c e s s (see G . G a l l a v o t t i , G.Iona-Lasinio A l l t h e f i e l d s r e p r e s e n t e d by ( 7 . 1 3 . )
n e s non g a u s s i a n automodel p r . d i s t r . .
v'
one o b t a i n s a
(23)).
can be d i s c r e t i z e d and s o one o b t a i -
There is a l s o a new b r a n c h i n g a u t - p r .
d i s t r , : t h e white noise. This r e s u l t
is new w i t h r e s p e c t t o t h o s e o b t a i n e d by B l e h e r e S i n a i and
it i s n o t c l e a r if t h i s new b i f u r c a t i n g p o i n t can b e found a s Gibbs d i s t r i b u tion for
p=h
f o r some models.
Thus we can draw t h e f o l l o w i n g p i c t u r e
white noise
*. :
p a r a m e t e r o f automodelity.
The c u r v e s i n d i c a t e a f a m i l y o f d i f f e r e n t non g a u s s automodel p r o b . d i s t r .
a s function of
For e v e r y a u t . g a u s s . p r . d i s t r . it i s p o s s i b l e t o f i n d
K
t h e branching o f a u t . p r . d i s t r . a s b e f o r e .
I am v e r y g r a t e f u l t o p r o f e s s o r s P.M.Bleher,
Acknowledgment.
Ja.G.Sinai,
R.L.
Dobrushin f o r t h e i r u s e f u l e x p l a n a t i o n s and f o r v e r y i n t e r e s t i n g d i s c u s s i o n s w i t h them. The a u t h o r acknonledeges v e r y Kind h o s p i t a l i t y o f t h e C h a i r of P r o b a b i l i t y t h e o r y o f t h e Moscow S t a t e U n i v e r s i t y , Lomonosov.
Bibliography S e c t i o n 1. 1 ) Gnedenko, B.V.,
Kolmogorov, A.N.,
Limit d i s t r i b u t i o n s f o r sums o f indepen-
d e n t random v a r i a b l e s , G o s t e c h i s d a t , Moscow (1949). 2 ) Ibragimov, I . A . ,
L i n n i k , J u , V:
Independent and s t a t i o n a r y sequences o f random v a r i a b l e s , Nauka, Moscow, (19651, E n g l i s h T r a n s l . Noordhoff, Groningen, (1971). 3 ) Gnedenko, B.V.,
Course on P r o b a b i l i t y Theory, Nauka, Moscow ( 1 9 6 5 ) .
4 ) Dobrushin, R.L.,
5 ) Lanford O.E.,
Funz. An. 2, 4 , 31, (1968).
R u e l l e D. Comm. Math. Phys.
6 ) Nakhapitan, B.C., 7 ) Malishev, V.A.,: 8) Dobrushin R.L., 9) Gallavotti, G.,:
Dokl. Ak. Nauk. U.R.S.S. T i r o z z i , B.:
224,(1975).
t o a p p e a r on Comm. Math. Phys.
Martin LBf, A. Nuovo Cimento, E B , 1, 425, (1975). Mat. Sbornik.
94,
136, (1974).
1 1 ) Kadanoff, L. e t a l . Rev. Mod. Phys.
12) J a . G. S i n a i , Teor. Ver. i e e Prim.,
27, --
(1969).
Dokl. A m i a n s k a i a Ak. Nauk. V61, (1975).
iO) Dobrushin, R.L.,
1 3 ) Arnold, V . I . , :
2, 194,
39,
395, (1967).
z, 1, (1976).
L e c t u r e s on b i f u r c a t i o n and v e r s a 1 f a m i l i e s , Usp. Mat. Nauk
5, 119, (1972).
1 u ) K o s t e r l i t z , J.M.,:
C r i t i c a l p r o p e r t i e s o f t h e one dimensional I s i n g model
wirh l o n g r a n g e i n t e r a c t i o n s , U n i v e r s i t y o f Birmingham p r e p r i n t . 1 5 ) P.M.
Bleher, Ya.G.Sinai,
Comm. Math. Phys.
2, 247,
(1975).
16) Arnold, V.I., Matematiceskie metodi classiceski mekaniki, Nauka, Moscow, (1974).
15,3, 469 (1970). Japan Journal of Mathematics, 22, 63, (1949).
17) Dobrushin, R.L., Teor. Ver. i ee Prim. 18) Ito, K.,
19) Ja. G. Sinai, Introduction to ergodic theory, Edited by Erevan and Moscow University, (1973). 20)
Ghikhman, I. I. ,~kho~?okhtod ,A .v., Theory of random processes, Ed. Nauka,
Moscow 1973, Transl. Springer-Verlag, Berlin (1975). 21) Bleher, P.M. 22) Iaglom, A.M.,
E-efiansion in Kadanoff renormalization group, Preprint Teor. Ver. i ee Prim., Vol. 2. N.3, p. 292, (1957).
23) G., Gallavotti, G.Iona-Lasinio, Comm. Math. Phys.
s,N.3, 301, (1975).
.
CENTRO INTERNAZIONALE MATEMATICO ESTIVO
(c.I.M.E.)
BASIC PROPERTIES OF ENTROPY IN,QUANTWM MECHANICS
A. WEHRL
I n s t i t u t f'b- T h e o r e t i s c h e P h y s i k
~ n i v e r s i t z tW i e n , A u s t r i a
BASIC PFOPERTIES OF ENTROPY I N QWLNTM MECHANICS
ALFRED W E H m I n s t i t u t ftir Theoretische Physik Universi-t
Wien
Austria
Introduction
Entropy i s one of t h e most important q u a n t i t i e s in physics, f o r it governs t h e behaviour of macroscopic systems and r e l a t e s macroscopic and microscopic q u a n t i t i e s . I n this l e c t u r e , 1 do not want t o go i n t o the thermodynamic foundations of entropy but s h a l l r a t h e r concentrate on some mathematical aspects of it. From t h e mathematical p o i n t of view, entropy can be considered as a measure of t h e i n t r i n s i c dispersion (degree of mixing, impurity, uncertainty, l a & of information, amount of chaos) of a quantum s t a t e .
(However, one should not
f o r g e t about t h e f a c t t h a t , a t l e a s t i n equilibrium, entropy i s a measurable q u a n t i t y ) . A s t a t e i n ordinary quantum mechanics being described by a density matrix p,i.e.
a l i n e a r operator,
2
0 , with t r a c e = 1, i t s entropy is given by
v. Neumann' s formula 1 )
I n what follows we s h a l l always p u t Boltzmann's constant kg = 1. V. Neumann's formula resembles very much Shannon's formula f o r .the information
of a f i n i t e , c l a s s i c a l p r o b a b i l i t y d i s t r i b u t i o n 2 ) : i f p = (pl,.
,pn) (pi
0,
1 pi
= 1) i s such a distribution, then 1Ip) =
- 1 pi
In pi.
In f a c t , many of
the theorems on quantum-mechanical entropy are more or l e s s obvious generalizations of r e s u l t s known from c l a s s i c a l information theory. However, i n some instances, the proofs are rather d i f f i c u l t (for instance, For the strong subadditivity), and, furthermore, there are classical r e s u l t s t h a t are f a l s e i n the quantum case.
Simple Properties
(a) Domain and range of entropy. Since a density matrix, being a compact operator, can be diagonalized,
(the Pi being one-dimensional projections), with P ( i ) 2 0,
( s ( x ) : = -X
In x i f x > 0,
-
1p (i)
0 i f x = 0 ) ;~ ( p ( ~ i) s) always
i s always defined, 20, but possibly =
m,
2
= 1,
0 , so t h a t S(p)
i f the Hilbert space is i n f i n i t e -
dimensional. One checks e a s i l y t h a t the range of S i s the whole extended r e a l half-line
[0,m].
Because s (x) = 0 + x = 0 or 1, S(p) = 0 implies p = pure, i .e. a one-dimensional projection. Pure s t a t e s are determined by a vector $ ("wave function"): since one-dimensional projections are of the form p =
I $) ($1,
in t h i s case
Since the wave-function, according to the principles of quantum mechanics, contains the maximal information t h a t can be obtained by measurements, it is plausible t h a t exactly i n t h a t case the entropy must be zero.
(b) P a r t i a l isometric invariance3). S(p) depends on the positive eigenvalues of P only, therefore, &y two density matrices with the same positive eigenvalues ( w i t h the same m u l t i p l i c i t i e s ) have the same entropy. Thus, i f U is an
arbitrary unitary operator, then S(p) = S (U p
u::).
O r , more generally, l e t
H 1'
H2 be two Hilbert spaces, p l , p 2 be two density matrices i n H1, o r H2, resp., V: H + H2 be a p a r t i a l isometry with i n i t i a l domain Dl and f i n a l domain D 1 2 such t h a t Ran p i c Di ( i = 1,2) and p2 = V p i v:, then S(pl) = S ( p 2 ) .
(c)
tbnotonicity with respect t o mixinq. C a l l two density matrices p l , p2 equi-
valent ( p l (hi ) 0, X1
%
p2) , . i f they have the same positive spectrum. I f p = A p
+
h2 = 11, then S(p)
1 1 + h2p2
2 S ( p l ) = S(p2). Heuristically speaking:
mixing of equivalent s t a t e s increases entropy. Proof: see section "Inequalities", concavity. Note, however, t h a t concavity does not follow from monotonicity.
Uhlmann ~ h e o ~ * ' ~ )
I f one does not take physics into account, there are of course many other
measures of i n t r i n s i c dispersion besides entropy. The most general concept i s due t o UhhaM and s t a t e s t h a t whatever quantity is introduced a s such a measure, it should f u l f i l properties (b) and ( c ) from l a s t section.. Uhlmam defines a density matrix p t o be "more mixed", o r "more chaotic", than another one, say p '
(and w r i t e s p
p',
or p '
4
p ) , i f p is i n the (weakly) closed
convex hull of the s e t of density matrices t h a t are equivalent ( i n the sense of ( b ) ) to p ' . Let us shortly s t a t e some results.
Main theorem. Let p ( l ) , p ( 2 ) 1 . . .
(or p ' ( * ) , p ' ( 2 ) , . . . l
resp.) be the eigen-
values of p (or p ' , resp. 1 , arranged i n decreasing order and repeated accord' ( l ) , p ( l ) p(2) < ing t o multiplicity. p p ' i f , and only i f , p (1)
>
<
-
(1)
+
(2)
,..., p ( l ) +
p (2)
+
... +
.-. + -(4,..
+
5p
p (n) ( p ' ( 1 )
+
(2) +
',I
Order relations. From t h i s theorem it follows immediately t h a t the relation is t r a n s i t i v e , and t h a t p
order relation.
>p
I ,
p
'>
p implies p
%
p ' , so t h a t
> is a pre-
Lattice structure of density matrices. One can show t h a t , with respect t o 2 , the equivalence classes of density matrices form a l a t t i c e . There exists always a "smallestn element, namely, the pure s t a t e s , but only i f the Hilbert space
is finite-dimensional, a "biggest one", namely ( d i m
Convex and concave functions. p
> p',
1.
i f , and only i f , for every convex (or
concave) function f , 2 0 , with f (0) = 0, T r f ( p ) 2 T r f ( p ' ) (or 2 0, resp. In particular, p
Coarse-grainins.
1 Pi
p
'
+ S(p )
)
.
2 S ( p ' ) , but the converse is not true.
Let Pi be a family of pair-wise orthogonal projections with
= 1. Then p (
1 Pi
p Pi.
(In theories about the measurement process, t h i s
is sometimes called "reduction of a s t a t e " ) . I f , i n addition, there e x i s t s a "coarse-grained" density matrix pc = then
2
pi p pi
1 A.
Pi such t h a t T r
< pc.
A i Pi =
T r p Pi,
Among the measures t h a t are compatible with Uhlmann's order relation, the quantum analogues of Renyi's entropies2) play a distinguished role. Let
for a > 0, # 1; So ( p ) = In d i m Ran p (the quantum analogue of the Hartley entropy), Sl( p ) = S ( p ) , Sm(p) =
- In
1 l p 11.
Then S ( p ) i s decreasing i n a6 ) , a
f i n i t e for a > 1, continuous w i t h respect t o the trace norm for 6 > 1 since
I(
~P ar 'Ia
-
(Tr p ' a ) l / a l
2 (Tr
Ip
- p ' l ) i /a
( t h i s is a consequence of the t r i a n g l e inequality7) f o r the v.Neumann-Schatten classes)
. Also
a-entropies look very much l i k e t h e r i g h t entropy, i n p a r t i c u l a r , they are a d d i t i v e (see s e c t i o n " I n e q u a l i t i e s (Two Spaces)") and have been used on s e v e r a l occasions, e.g.
i n non-equilibrium s t a t i s t i c a l mechanics.
Continuity P r o p e r t i e s o f Entropy
I f t h e H i l b e r t space is finite-dimensional,
entropy is c l e a r l y continuous. I n
t h e infinite-dimensional case, entropy is discontinuous with r e s p e c t t o the t r a c e norm, because every " b a l l " {p ' : Tr
Ip - p '1
<
E
(E
> 0)) contains density
matrices with i n f i n i t e entropy. This can be shown e x p l i c i t l y a s follows: l e t p(') < p ( 2 ) be thd eigenvalues of p. Choose N such t h a t
...
Let p
'
for i
2 N,
have t h e same eigenvectors as p , b u t eigenvalues p (i)
= p (i) f o r i < N,
provided t h a t
otherwise one can assume t h a t p (N-l) p ' ( N - l ) = P IN-''
-
C*
(c' <
L),
> 0 , then l e t p-*(i)
p ~ ( f~o r) i , N
= p(i) for i < N
- 1,
a s above. f n both cases, c is
to be chosen such t h a t
hen, T r
Ip - p'l
< E , b u t ~ ( p *=)
=.
Lower semi-continuity f o r entropy. Since S o ( @ )is continuous f o r a > l y a n d S(p)
= l i m Sa (P ) = sup Sa (p) , S (p ) is lower semi-continuous. a+l
Therefore, t h e
a >1
s e t s {p: S ( p ) ( n) a r e closed, t h e i r complements a r e dense, hence they a r e nowhere dense and
i s of f i r s t category. Besides lower semi-continuity,
some other r e s t r i c t e d continuity properties a r e
valid. The most t r i v i a l one is
Convergence of canonical approximations3). ~f p =
1 p ( i )pi,
arranged i n decreasing order, t h e Pi being one-dimensional,
( "canonical
approximation" )
. Then,
S ( pN ) + S (p )
the p(i) being let
.
Much l e s s t r i v i a l i s t h e
Dominated convergence theorem f o r entropyg). I f pn i s a sequence of densify matrices converging weakly t o p , and i f there e x i s t s a compact operator A 2 0 (not necessarily a density matrix) such t h a t pn <
m,
2A
f o r a l l n and -Tr A I n A <
then S(pn) + S ( p ) .
Entropy I n e q u a l i t i e s (One Space)
.
For a l l entropy i n e q u a l i t i e s , the reader i s r e f e r r e d t o t h e review a r t i c l e by Lieb 10)
.
Concavity. S(Xpl
+
(I-X)p2)
2 X
S(pl)
+
(1-1) S(p2) ( 0
2X 5
I ) . (This a l s o
proves monotonicity with respect t o mixing). Proof: This is true indeed f o r every concave function f
2 0. Let
orthonormal b a s i s of the H i l b e r t space t h a t diagonalizes p: = Xpl Then,
be an
+
(1-X)p2.
1 ~ ( ( O ~ I P ~ , ~ ->~ ~T )r ) f
Now, f ( ( d i l ~ 1 , 2 0 i ) )
( ~ ~ l f ( ~ ~ hence , ~ ) d ~ ) t
( ~ ~ , ~ ) .
For another proof, s e e ~ i e b " ) . Usually, concavity i s considered to be one of the most important p r o p e r t i e s of entropy. Concavity extends t o t h e following inequality: l e t p =
1 Xipi
(Xi
) 0,
Xi =
= 1 ) . Then,
The term
- 1 hi
I n Xi may be r e f e r r e d to as "mixing entropyn.
I t s u f f i c e s t o prove t h e r.h.s.
dimensional p r o j e c t i o n s Pi.
where the Q
i
only. Let us f i r s t assume t h a t t h e pi a r e one-
Then,
a r e a l s o one-dimensional projections, but, i n addition, a r e
mutually orthogonal4). This i s t r u e since
where the sup is taken over a l l p r o j e c t i o n s P of dimension ( n. (Ky Fan's in-
... v
equality 7 ) ) . Now this is 2 Tr p (Ql v
Qn)
, XI
+
... + An.
For t h e general
case, w r i t e
pi =
(j) p
1p i j
ij'
t h e Pij being t h e eigenprojections of pi. Then,
P =
hence Sip)
5-
1
Xi p
1 ij 1
(j)
Xi Pi
p
1 Xi
i
:,(pi)
- 1 Xi i
-1
0:)'
Pi0'
- 1 A1. I n
hi =
i
Xi. 4)
There is an e q u a l i t y on the r.h.s.
Coarse-graining.
) =
ij
ij
=
Pij.
i f Ran pi i s orthogonal to Ran p
j
f o r i f j.
The coarse-graining r e l a t i o n s known from Uhlmann Theory give
of course r i s e to the corresponding entropy i n e q u a l i t i e s .
Entropy I n e q u a l i t i e s (Two Spaces)
H
Additivity. Let all a
E
=
Hl 8 H2. ~f p = pl O p 2 , then Sa(p) = Sa(pl) + Salp2) f o r
s i n c e the eigenvalues o f p a r e p ( i ) 1""'
[o,-1,
( k ) , hence
A d d i t i v i t y e x p r e s s e s t h e f a c t t h a t , i f a system w n s i s t s o f two independent p a r t s (which is mathematically expressed by t h e d e n s i t y m a t r i x pl 8 p 2 ) , t h e n the information about the whole system i s j u s t the sum o f t h e informations a b u t its p a r t s .
S u b a d d i t i v i t y . Now l e t p be a d e n s i t y matrix i n
H
and l e t pl:
= Tr,, p, p2: = 2 = TrH p be t h e corresponding p a r t i a l t r a c e s . By p a r t i a l t r a c e t h e following 1 is meant: l e t (gi} be an orthonormal b a s i s f o r H l , ($i} be an orthononual
basis f o r
H2.
Then t h e m a t r i x elements of p l , which is an o p e r a t o r i n
HI, are
given by
This d e f i n i t i o n does n o t depend on the p a r t i c u l a r c h o i c e o f {$ checked t h a t , f o r A
E
8(H1),
3.
i
I t is e a s i l y
t h i s p r o p e r t y may a l s o be used as a d e f i n i t i o n o f t h e p a r t i a l t r a c e . Hence one can s a y , t h a t pl c o n t a i n s j u s t a l l those informations o f p t h a t r e f e r to t h e f i r s t subsystem only. The s t a t e m e n t , t h a t S ( p ) "subadditivity"
.
Proof. L e t {$I.) be an orthonormal b a s i s i n
5 S(pl) +
S (p2) .is c a l l e d
H2 t h a t d i a g o n a l i z e s p2. Then H may
w i t h Hi = /fl 8 $i. With r e s p e c t t o t h i s decomposition, p i has a m a t r i x r e p r e s e n t a t i o n be w r i t t e n a s @ H
p
i s the density matrix
w i t h A~ = Tr pii.
hence s ( p ) = S(pI)
+
1 pii,
whereas p 2 i s t h e numerical matrix
Now,
1
- 1 hi
~n ii
2
~ (oii)1 - 1 X~
In
(concavity) =
S ( p 2 ) . For o t h e r proofs, perhaps more e l e g a n t , s e e 10, 11. Also
So(p) is subadditive, but no o t h e r a-entropy.
Subadditivity c e r t a i n l y i s one
of the most important p r o p e r t i e s of entropy. I t may be i n t e r p r e t e d from t h e information-theoretical p o i n t of view i n such a way t h a t , i f one takes the p a r t i a l t r a c e s p l and p 2 and fits them together, a l l information about correl a t i o n s is l o s t and, therefore, t h e entropy of p
6) p 2 (= S ( p l )
+ S ( p 2 ) ) must
be bigger than the entropy of the o r i g i n a l p .
Monotonicity with r e s p e c t t o enlargening of t h e space. This would be t h e statement t h a t S ( p )
'
s ( ~ ~ Although ) .
t h i s i s t r u e i n the c l a s s i c a l case, it i s
f a l s e in the quantum case since p may be pure, b u t p l may not. I n f a c t , to every density matrix p l one can f i n d a H i l b e r t space H2 and a pure density matrix p i n Hl 8
H2 with
p l = Tr
"2
p.
By the way, i n t h a t case, t h e p o s i t i v e
12) s p e c t r a of p l and p2 coincide, hence S ( p ) = S(p2) 1
.
Triangle i n e q u a l i a 2 ) . l s ( p l ) The r.h.s.
- S(p2) I
5 S(P) 2 S(pl) +
s(p2)
being subadditivity, one has to prove the 1.h.s.
only. Let ff3 be a
Bilbert space and p ' be a pure density matrix in ffl PO ff p
P
= Tr
Let p 3 = Tr,,
p'.
H3
~ =~ TrH : p'.
1
p'.
Qff such t h a t 2 3 Then, S(p) = S ( p 3 ) . S ( p l ) = S(p23 ) , where
1 2 Subadditivity y i e l d s
and interchanging of 1 and 2 proves the triangle inequality:
Entropy I n e q u a l i t i e s (Three Spaces).
This is a group of i n e q u a l i t i e s centered around strong subadditivity. Strong subadditivity means the following: l e t
ti = ffl
@
ff2 QD ff3, p be a density matrix
i s one-dimensional, t h i s reduces t o subadditivity. 2 This i s a highly non-trivial r e s u l t and a proof of it requires i n e q u a l i t i e s I f ff
t h a t are very hard to derive, f o r instance t h a t ( f o r finite-dimensional matrices) t h e mapping
is concave14). Since t h i s is a f i e l d f u l l of t e c h n i c a l i t i e s , we have to r e f e r the reader to the l i t e r a t u r e l o , 13, however, it should be pointed out that strong a d d i t i v i t y has various important implications i n physics.
Aximatic Characterizations3)
Let Q be a mapping of the s e t of density matrices i n t o [O,@]. One may ask which conditions have to be imposed on 8 i n order t h a t , up to a constant
factor, Q is the entropy.
preliminary axioms. (PI) @ ( p ) is f i n i t e , i f p is of f i n i t e rank. ( ~ 2 ~f ) p is not of f i n i t e rank, then O(pN) + O ( p ) , where the pN are the canonical approximations of p. (P3) Q f u l f i l s p a r t i a l isometric invariance.
Characterization "B l a Renyi". Let Q f u l f i l (Pl) (I?) I f H = H1 @ hi
0,
1 hi
...
(B
Hn, and
p = Alpl 8
= I ) , then Q ( p ) =
i n the Hilbert space
cn with
1 hi
#(pi)
-
... 8 hnp n +
(P3) and (pi being density matrices,
@ ( A ) , where A is a density matrix
.., A n . (The Hi
eigenvalues A l p .
same dimension).
need not have the
Then, Q ( p ) = const-S(p). For the proof, it suffices t o consider the case t h a t a l l p i are of f i n i t e rank. Choosing suitable orthonormal bases i n the H
i'
and assuming t h a t a l l density
matrices under consideration commute, one is l e f t with the classical situation2) (since (P3) implies symmetry and expansibility), hence Q ( p ) = const-S (p)
,a
f o r t i o r i t h i s i s true for a l l p
.
Characterization "d l a ~ c z e l ,Forte, and Nq". Let Q f u l f i l (Pi)
-
(P3) and
additivity a s well a s subadditivity. Then Q i s a linear combination o f , s ( ~ ) and So(p). Proof. Again it is f a i r l y simple t o reduce the situation to the classical case. Then one can apply the very remarkable theorem of Aczel, Forte, and bIg1')
:
let
J, be a function, defined for a l l f i n i t e probability distributions (i.e. for
a l l n-tuples (pl,p2,.
(i)
..,pn) such t h a t pi
2 0,
1pi
= 1) with the properties
$ 2 0
(ii) J , ( P ~ ( ~ ) , . . . , P ~ ( ~ ) ) J,(pl,..-,p
n
),where P is any permutation of (l,..,n)
("synrmetry" (iii)J,(plr.-.,pn,O)
= $(pl,-..,p n
..,pn) + $(ql,. ..,%I ..., Pn) + $(ql, ...,%I,
( i v ) J,(plq1,-. ,pn%) = $(pl.. (v)
J,(rllr...,rm) 2 g ( p 1 .
("expansibility") ("additivity") where
Then, $(pi,
..,pn) = - A 1 pi
with A and B independent of n.
In p
i
+ B* (logarithm of the number of p ' s f
0) ,
The proof of t h i s theorem i n principle i s elementary except for one numbertheoretical argument, but very tricky. I t remains t o eliminate the quantum Hartley entropy S ( p ) 0
. This can be
done by
several very mild continuity conditions, for instance of t h a t kind: l e t p n be a sequence of density matrices of rank 2, P be a one-dimensional projection, +
- PI
1 /pn I -+ 0, then S ( pn ) + 0 (or one could even demand only t h a t S(P), which is not clear a p r i o r i ) 3 )
[ p , , ~ ] = 0, S(pn)
.
Of course, the charactkization "d l a Aczel, Forte, and Ng" i s much more related to physics than the one "d l a Renyi" since additivity and subadditivity have a rather appealing physical interpretation.
Acknowledgments
The author wishes to thank Prof. E l l i o t t E. Lieb f o r a c r i t i c a l reading of the manuscript and f o r making numerous suggestions, as well as Prof. O.E. I11 for useful remarks.
Lanford
References
1.
J.v.Neumann,
2.
A. Renyi,
2. Phys.
57, 30
(1929)
Wahrscheinlichkeitsrechnung, Deutscher Verlag der Wissenschaften,
Berlin 1966 Ochs, Rep. Math. Phys.
3.
W.
4.
A. Uhlmann, W i s s .
(1971);
21,
8,
109 (1975)
2. Karl-Marx-Univ.
Leipzig, Math.-Naturwiss.
2, 139 (1973); Phys. 6, 15 (1974)
421 (1972);
R.
p r i v a t e communications
5.
A. Wehrl, Rep. Math.
6.
G. Hardy, J. Littlewood, G. Polya, I n e q u a l i t i e s , Cambridge, 1967
7.
E. Beckenbach, R. Bellmann, Inequalities, Springer 1971
8.
A. Wehrl, Rep. Math. Phys.,
9.
B. Simon, appendix t o ref.
10. E. Lieb, Bull. AMS
81,
to be published 12
1 (1975)
11. D. Ruelle, S t a t i s t i c a l Mechanics, Benjamin, 1969 12. H. Araki, E. Lieb, Corn. Math. Phys.
13. E. Lieb, M. Ruskai, J. Math. Phys. 14. E. Lieb, Adv. Math.
11,267
IS,
160 (1970)
14, 1938
(1973)
(1973)
15. J. Aczel, B. Forte, C. Ng, Adv. Appl. Prob.
6,
131 (1974)
20,
633
