QUANTUM SIGNATURES OF CHAOS

Fritz Haake

Fachbereich Physik
Universität-Gesamthochschule Essen
W-4300 Essen 1, F. R. Cermany

1 Quantum Chaos as a Game with Quadratic Equations

The distinction between level dustering and Ievel repulsion is one of the quantum ana-
logues of the classical distinction between globally regular and predorninantly chaotic
motion (see Figs. 1, 2, 3). In order to reveallevel repulsion under conditions of global
classical chaos special care may be necessary: (i) subspectra referring to different values
of the quantum numbers related to symmetries must be dealt with separately and (ii)
for systems with quantum localization only Ievels whose wavefunctions have overlapping
support must be admitted. A "Ievel" may either be an energy eigeuvalue E in t.he case
of autonomaus systems or, for periodically driven systems, a quasi-energy <p, i.e. an
eigenphase of the unitary Floquet operator transporting the wavevector from period to
period.
I shall here be concerned with three varieties of quantum systems displaying level
repnlsion. They differ in the degree ß of Ievel repulsion which is also referred to as the
universality dass index. The degrec in qucstion dcscribes the limiting behavior of the
distribution P(S) of nearest-neighbor spacings S for small S,

for ß-'> 0. (l)

The so-called orthogonal, unitary, and symplectic universality classes of dynamical sys-
tems have ß = 1,2, and 4, respectively (see Fig. 2). Which univcrsalit.y dass a given
dynarnical syst.ern rnay belong to is decided by the set of unitary and anti-unitary sym-
metries. Anti-unitary symmetries are related to (generalized) time reversal invariance
[1,2,3].
For a given spectrum consisting of ma.ny Ievels the level spacing distribution takes
the form of a histogram (see Fig. 2). As is well known, such histograms generically are
filithful to the smooth distributions predicted by Raudorn Matrix Theory for cnscmblcs
of N x N matrices in the Iimit N -> oo. Rather than aiming at thc construction of the
full functions Pß(S) for 0::; S < oo I shall confine myself to the discussion of the small-S
behavior eq. (1 ). This rather modest goal can be reached by using quite elementary
methods.

Quantum Coherence in ,\fesoscopic Systems

Edited by B. Kramer, Plenum Press, New York, 1991 583
584 F. HAAKE

~~illf~~XL
d

';·'.:,~;;·:;;; > '4.~~~~·~,r

z
Figure 1: Sequence of four classical phase space portraits
for a periodically kicked top. As a coupling constant is
increased from (a) to (b) and so forth, chaos is seen to
expand from small islands (a) to full coverage of the phase
space (d).
QUANTUM SIGNATURES OF CHAOS 585

a b
1.

1. d

P(S)

0.5

o. ,._..__~....__,_~~-'--..~-.) "-"-'-~..___L~~_:t".;:::....__.__J

Figure 2: Level spacing distributions for periodically kicked

tops. The histogram in (a) refers to a case with predomi-
nantly regular motion in the classicallimit, while (b), (c)
and (d) correspond to conditions of global classical chaos.
The latter three cases differ by thc dcgrce of Ievel rcpul-
sion which is linear (b), quadratic (c), and quartic (cl); the
smooth cmves describe the predictions of Random Matrix
Theory for the circular orthogonal, unitary, and symplectic
ensembles.

More specifically, I propose to show that (i) ß = 1 if there is a (generalized) time

reversal invariance T squaring to +1, (ii) ß = 2 if there is no such invariance, ancl (iii)
ß = 4 if there is an invariance with T = -1, the latter case being characteristic of
systems with Kramer's degeneracy[1,2,3].
Somc remarks about time reversal are in orcler now. I shall consicler autonomaus
systems: the generalization to perioclically driven systems may be looked up in the
Iiterature [3]. Schröclinger's equation

(2)
586 F. HAAKE

is said tobe time reversa.l invariant if for every solution 'lj!(t) there is another one,

(3)
with T an anti-unitary operatorsquaring to either +1 or -1,

(4)

By demanding that ;;;( t) obey the same Schrödinger equation as 1/;( l) one easily finds
that the Hamiltonian must c01nmute with T,

THT- 1 = H. (5)
The reader may recall that a unitary operator leaves the scalar product of two arbitrary
state vectors unchanged in the sense (1/>lx) = (Uif>IUx). An anti-unitary operator T, on
the othcr hand, docs change a scalar product into its complex conjugatc,

(T1/;ITx) = (1/;lx)* = (xllj]). (6)

The squared modulus of a scala,r product, i.e. probabilities, a,re left invariant by both
unitary and anti-unitary operations. The additional requirement eq. (4) distinguishes
the special anti-unitary operators deserving the nametime reversal operations [4]: any
wave function should be reproduced, up to a constant factor c, when operated upon
twice by T, T 2 '1jJ = c'I/J; it is easy to see that ±1 are the only possible values for c.
It may be weil to give a few examples. The most weil known one is conventional
time reversal, which leaves all coordinates x unchanged, reverses the sign of all momenta
and angular momenta, and replaccs wavefunctions (in the position representation) by
thcir complcx conjugatcs

X --+ ToxT0- 1 =x
p --+ TopT0- 1 = -p
L --+ ToLT0- 1 = -L
1/J(x) --+ To1/;(x) = '1/J*(x). (7)
This yields an invariance for a free particle with the Hamiltonian H = p 2 /2m but not
for a charged particle in a magnetic field since T(p- eA)2T- 1 = (p + eA )2 .
However, a charged particle in a homogeneaus magnetic field does have another
anti-unitary symmetry. Let B = (0, 0, B) point in the z-dircction and the vector potential
be A = (-1/2)x x iJ suchthat the Hamiltonian reads

(8)

Clearly, this H cornmutes with T1 = UT0 where 10 is conventional time reversal and U
a rotation by 1r around any axis perpcndicular to B. Thcrc is nothing "false" about the
non-conventional time reversal operatiou T1 : it obeys cqs. (2), (3), (4), (5).
To break the invariance undcr T1 it suffices to switch on a homogeneaus electric field
E tilted arbitrarily against the magnetic field iJ. Even then, another non-conventional
time reversal invariance remains which is easily checked to be the product of T 0 with a
QUANTUM SIG)JATURES OF CHAOS 587

reflection about the plane spanned by B and E. To verify that symmetry it must be
realized that B and E behave differently under spatial reftections.
Breaking any kind to time reversal invariance in an atom or nucleus obviously
requires inhomogeneaus external fields. \Vhile sizeable inhomogeneities across a nucleus
are certainly inaccessible to the human cxpcrirnenter, the most highly excited Rydberg
atoms may not make for hopeless tasks forever.
One rnore ingredient will be needed to prove eq. (1 ), nearly degenerate perturbation
theory. Imagine the Hamiltonian of some dynamical systems to have the form II =
H0 +>.V and consider the fate of two neighboring eigenvalues E 1 (>.) and E 2 (>.) when the
control paramctcr >. is varicd. Clearly, oncc the distance IE1 (>.)- E 2 (>.)1 is smaller than
the distance of either Ievel to any third one, the close encounter in consideration can be
studied by nearly degenerate perturbation theory. One may assume the two eigenvalues
a.nd the corresponding eigenvectors known for a particular value ,\ 0 of >. and diagonalize
H in that basis. If the two Ievels are non-degenerate the basis is two-dimensional and
the cigenvalues of thc two by two rnatrix have thc diffcrencc

(9)
At this point it becornes obvious that by varying a singlc parameter like >. the
distance of neighboring Ievels can bc steercd through a minimum but not, in general,
made to vanish. The discriminant in eq. (9) being a sum of non-negative squares the
number of parameters needed to enforce a crossing is two or three, depending on whether
the matrix Hij is real symmetric or complex Hermitian.
The density of Ievel spacings may be defined as

(10)
with the average to be performed over all neighboring Ievels of the Hamiltonian. Al-
ternatively, one may average over the set of two by two matrices H associated with
close encounters of Ievel pairs, provided a suitable distribution function for the matrix
elernents is available. The latter average takcs thc form [5]

P(S) = jdxdydzW(x,y,z)5[S-Vx2+y2+z2]

8 j dxdydzW(Sx,Sy,Sz)5 [1- jx
2 2 + y 2 + z 2] , (11)

where I have assumed that no symmetry restricts the Hamiltonian to be a real matrix.
If the density vV( x, y, z) exists as a reasonably smooth function and neither diverges nor
vanishes at x = y = z --+ 0, the spacing distribution approaches the power-law behavior
eq. (1) with ß = 2. This degree is typical of the so-called unitary ( Gaussian or circular)
ensembles of random matrices [1-3].
An anti-unitary symmetry of H with 1' 2 = 1 modifies the situation just described
in a small but consequcntial dctail: the Hamiltonian can then be represented as a real
symmetric matrix, Hij = H;i by employing T invariant orthonormal basis vectors, ~'i =
1''1/Ji, (l/>ill/>j) = Dij· An arbitrary basis {cp;} is easily turned into a Tinvariant one in the
following way. Let l/> 1 ~ cp 1 +Tcp 1 suchthat T ~· 1 = ~'r and normalize, (~; 1 11/; 1 ) = 1; then
take a combination ;{2 of the 'Pi orthogonal to l/>I(l/>11;{2) = 0, choose l/>2 ~ ;j;2 + T;j;2,
and again normalize etc. The reality of H then follows from thc anti-unitary of T and
588 F.HAAKE

[II,T] = 0: indeed, (1~1IHI1~z) = (T1/J1ITHI1~z)' = (7joliJJIVJz)*; due to its Ilermiticity the

Hamiltonian matrix is then also symmetric. It follows that the discriminant in eq. (9)
is a sum of only two non-negative reals. The number of integrations in eq. (11) is thns
reduced to two and the degree of Ievel repulsion to ß = 1 which also characterizes the
so-called orthogonal (Gaussian or circular) ensembles of random matrices.
Finally, when a (generalized) time reversal invariance holds with T 2 = -1 an
interesting complication arises due to Kramer's clegeneracy, a double degeneracy of each
level. To check on this statement, imagine 1j; to be an eigenvector of H; t.hen TVJ is also
an eigenvector pertaining to the same eigenvalue due to [H, T] = 0; but y6 and TI/J are
orthogonal: (1/>ITI/J) = (T1/JIT 2 7jo)* = -(Ty61,P)* = -(1/JIT7jo) = 0. Therefore, in treating a
near miss of two Ievels by ncarly eiegenerate perturbation thcory I must diagonalize a four
by four matrix. That matrix is restrictecl by Ilermiticity ancl the symmetry [JJ, T] = 0
in precisely such a way that its secnlar determinant takes the form of the square of a
two by two determinant. It is easy to verify that [6]

o+ß 0 I - icr -E- io )

H = ( 0. a- ß E - i8 1 + icr (12)
I+ w- E + i8 a-ß 0 '
-c + i8 ~~- icr 0 a-ß
where a, ß, 1, 8, c, er are six real paramctcrs; indcccl, aftcr arranging thc basis vectors in
the orcler 11), Tl1), 12),112) consicler, e.g. (1IIITI1) = ('1'1l'l'HTI1)* = -(T11HI1)*
- (1IHT]l) = 0. The two eigenvalues have the difference
(13)
When determining the sma.ll-S' behavior of P(S) in analogy t.o eq. (11) I now encounter a
fivefolcl integral ancl thus the quartic degree of Ievel repulsion, ß = 4, otherwise known to
be characteristic of the so-callecl symplectic (Gaussian or circular) enscmbles of random
matrices [1-3].

2 Random Matrix Theory as Statistical Mechanics

Random Matrix Theory is well known tobe quite succcssful in clcscribing JJuctuations in
the quantmn spcctra of dynamical system with global chaos in the classicallimit. That
theory moclels Hamiltonians as random I-Iermitian matrices and Floquet operators as
ranclom unitary matrices. For the sake of a change I shall cleal with Floquet operators,
F, F Fl = 1, in this section; the transcription to Hamiltonia.ns is mostly a matter of
nota.tion, though.
Dyson's circular ensembles [1-3] of unitary N X N ranclorn rnatrices irnply the
following joint probability clensities for thc N cigenphases 'Pi

(14)

The degree oflevel repulsion ß appears again here ancl clistinguishes the three universality
classes as
orthogonal
unitary (15)
symplectic
QUAI\TUM SIG~ATURES OF CHAOS 589

"
Figure 3: Quasi-energy spedrum of a perioclically kickecl
top versus some coupling constant >.;top ancl coupling con-
stant the same as in Fig. 1; Fig. la, b, c, ancl cl corresponcl
to A = 1, A = 2.5, A = 3, ancl A = 6, respectively. Note
that the spectrum changes its chctracter from levcl duster-
ing to lcvcl repulsion in thc same range of A, A r:::: 3 ± 0.5, in
which the classical transition from preclominance of regular
motion to preclominance of chaos takes place. The range
A ;<; 3 has A inclepenclent spectral fluctuations; this is the
range for equilibrium statistics treatecl in section 2. As ).
incrcases from 0 to 3 a relaxation into equilibrium takes
place such as the one treatecl in lecture 3.

Fl·orn the clistribution eqs. (14), (15) all measures of fluctuations in the spectrum can be
clerivecl by suitable integrations.
Rather than clescribing the amazing clegree to which the Floquet spectra of classi-
cally chaotic systems are faithful to the preclictions of Ranclom Matrix Theory, I shall
here focus on the question as to why tlmt shoulcl be the case. Tothat end I shall pointout
that Raudorn Matrix Theory car1 be rcinterpretecl as equilibrium statistical mechanics
of a certain fictitious classical gas.
\Vhen contemplating Fig. 3 the reader will reali7,e the analogy of the bundle of
N level curves <p(A) with a bunclle of trajectories of N particles in a one-climensional
configuration space with Aas a time; the particle-particle interaction appears as repulsive.
In fact, on the basis of a cliscovery of Pechukas [7] the analogy can be extcndecl into a
rigorous one-to-onc rclation. Consicler a Floq uet operator of the form
(16)
with two Hermitian operators H 0 ancl V, both inclepcnclent of thc control parameter >.;
such Floquct opcrators arisc for periodically kickecl systems for which the perturbation
AV is switchecl on perioclically ancl impulsively. The control parameter clepenclence of
the eigenvalue problem
m=1,2, ... N (17)
can be shown tobe equivalent to the classical motion generatecl by the Hamiltonian [8,3]
2
1 "" 1 "" J€mn (18)
Pm + g L_., · 2 ((
2 j
'H = 7)'""" L_.,
rn m::J.n SlTI Ym
_
r..pn
)/2) ·
590 F. HAAKE

The correspondence between eqs. (17) and (18) is as follows

.\ +-+ time

'Pm +-+ particle Coordinates

Pm = (miVIm) +-+ momenta

lmn = -2(m1VIn) (ei('Pm-'Pn)- 1) +-+ angular momenta. (19)

The matrix elements of V occurring here are meant in the Floquet basis eq. (17). The
only unusual feature of the dynamics in question is that the strengths of the pair inter-
action involve dynamical variables lmn rather than being constants; consequently, the
phase space encountered has the dimension

Nß = 2N + ßN(N -1)/2, (20)

the universality class index ,B enters herein counting the number of independent parame-
ters in the off-diagonal elements of the matrix V. To establish the Ha.miltonian equations
of motion the usual Poisson brackets for coorclinates and momenta must be employecl as
weil as indepenclent such brackcts for the angular rnomcnta; the latter, looking different
for thc thrce universality classes [3,9], constitute Lie algebras of generators of rotations
in appropriate (ßN) dimensional vector spaces - hence the name "angular momenta".
Previous to the cliscovery of its relation to the quantum mechanical eigenvalue
problern eqs. (16) n.nd (17) the classical N particle system with the Hamiltonian eq. (18)
had been known in the rnathematicalliterature as a variant of the Calogero-Moser dy-
namics [10,11,12]. As a byproduct of the equivalence in question the integrability of the
fictitious-particle dynamics is worth being noted. (The reader may recall that evcn the
harmonic chain, a prototypical integrable N particle system, requires diagonalization of
an N X N matrix.) In fact, by employing for a moment the eigenbasis of V it is easy
to see that the .\ clepenclence of the matrix F in eq. (16) can be built with N harmonic
oscillations; in other words, the classical trajectories implied by the Harniltonian 'H. are
confinecl to an N torus within the /vß dimensional phase space; moreover, the N fre-
qucncies involved have no reason to be commensurate and thus in general give rise to an
ergodie motion in the N torus [13].
An infinity of constants of the motion of the fictitious-particle dynamics,

(21)
becornes accessible once it is recognized that the Hamiltonian equations can be written
in thc so-called Lax from [14], V = [g, V] and f = [g, V] with a suitablc "generating
matrix" g. Due to the rcla.tion between v;nn and lmn given in eq. (19) the pha.se space
functions C 11 in genera.l diverge as two coordinates coincide, 'Pm -<.pn --+ 0 at lmn f 0; this
fact may either be interpreted in quantum parlance ( avoided crossings, Ievel repulsion)
or in the classical-particle picture (repulsive pair interaction). Unfortunately, it is not
known at present whether or not the set of constants of the motion eq. (21) contains a
subset of Nß- N inclepenclent ones allowing to nail down the N torus.
In the Iimit N--+ cxJ the fictitious gas invites a statistical description [15,16,3]. The
appropriate equilibrium ensemble would be the generalized microcanonical one which
QUANTl:M SIGKATURES OP CIIAOS 591

specifies the N torus by assigning sharp values to suitable ;\(ß - N constants of thc
motion. Technically easier to work with is the canonical ensemble

(22)

with Lagrange parameters aw The argument to follow assumes all C'" in eq. (22) to be
of the form eq. (21), real valued, and boundcd from below. Of special interest is the
reduced distribution of all N coordinates, obtained by integrating out all N momenta
and the Nß - N independent angular momenta

(23)

Clearly, this distribution vanishes at particle crossings, due to the divergence of the
Cl' mentioned above. To ascertain the behavior near crossings I abandon the angular
moment.a emn in favour of the Vmn as int.egra.t.ion variables,

P(r.p)

(24)

Note tha.t for fixed finite Vmn the function P(r.p) has no reason va.nish a.t a. pa.rticle
crossing. The degree to which the fict.itious particles a.void crossings is t.herefore solely
e
determined by the Ja.cobian of the tra.nsforma.tion -+ V, det( D€/ DV) ~ P( r.p )/ P( r.p ),
which a.ctua.lly is nothing but the joint eigenvalue density eq. (14) predicted by Random
.Matrix Theory.
Any noticea.ble r.p dependence of the function P would make for a. difference between
the eigenva.lue statistics implied by Random Matrix Theory and equilibrium statistica.l
mechanics of the fictitious gas. Interestingly, could I exclude from the ensemble eq. (22)
all constants of the motion eq. (21) with 11 1 = p 3 = ft 5 = ... = 0, i.e. admit only trV 1',
I would face a P strictly independent of the r.p a.nd thus rigorously recover Random
Matrix Theory, even more interestingly, the usual microcanonical ensemble exp( -,61i)
does have this nice property since 1{ ~ hV 2 , unfortunately, however, of the tr V 1' only N
a.rc indcpcndent a.nd thesc a.rc definitely insufficient in number to na.il down the N torus.
It follows that there can be no rigorous equivalence, with respect to the distribution P( r.p)
of all N coordinates, between ergodie motion of t.he N torus and Ra.ndom Matrix Theory
for fixed finite N. vVhat does hold instead is a weaker, asymptotic correspondence: the
Ievel spacing distribution [16,3] (as weil as certain other quantities which typically involve
corrcla.tions between only a few Ievels) come out the samein the two approaches, provided
the Iimit N -+ oo is taken. For the proof of tha.t a.symptotic equivalence I refer the reader
to the originalliterature. The proof is based on representing the r.p dependence of P( r.p)
by an N fold Fourier series; none of the finite-frequency terms can compctc, for N-+ oo,
with the zero-frcqucncy one in its effect on the level-spacing distribution.
Some non-trivial modifications of the foregoing consicleration, become necessary
for the Ievel dyna.mics of quantum Hamiltonians

H(,\) = H 0 +,\V (25)

592 F.HAAKE

of autonomaus systems. The equivalent fictitious gas was identified by Pechukas [7] as
governed by thc Hamiltonian function

(26)

Here, the particle coordinates Em correspond to the eigenvalues of H, in contrast to

the eigenphases 'Pm of Floquet operators the Em are not confined to a finite interval
but rather to 0 < Ern < =· The diagonal elements Vmn of the perturbation V in the
eigenrepresentation of H()..) again play thc rolc of rnornents whilc Rmn = -2V,nn(Ern -
En)·
Unfortunately, due totheinfinite range of the Em and the absence of any confining
potential in eq. (26), the repulsive pa.ir interaction causes indefinite expa.nsion of the
Pechukas gas and thus defines the application of equilibrium statistical mechanics. The
instability of the gas corresponds to the structure of the original quantum Hamiltonian
H =Ho+ ).V: for).,-+ = the eigenenergies must fiy apart as Em(A) ~ )\; their spectral
density thins out correspondingly. In as much as only fluctuations in the spectrum of
H(A) are of interest it is intuitive to rescale as H()..) = g()..)(H0 +)\ V) suchthat the mean
density of Ievels remains independent of A. For the fictitious gas that rescaling H -+ Ji
means a non-canonical transformation with new coordinates Ern = gEm, a new time
r,)., -+ )., (T), g = g ( T). By req uiring the transformed dynamics to be again Hamiltonian
in character with a Hamiltonian function H independent of T, two differential equations
for the two functions g( T) and ).,( T) are obtained. The resulting Hamiltonian function
[17,3]

(27)

differs from Pechuka's one by a hannonic binding potential which prcvents thc modificd
gas from exploding as T -+ =·
Integrability is not destroyed by the confining potential in eq. (27); the Hamil-
tonian equations of motion can again be written in the Lax form and the expression
eq. (21) reappear as constants of the motion. As an additional conserved quantity the
Hamiltonian H itself must be listed since now H is not of the form eq. (21 ).
Equilibrium statistical mechanics does make sense for the confined gas. A restricted
canonical ensemble involving H as well as at most N different constants of the motion
trV~' now rigorously implics the dcnsities P(E) of all N Ievels predicted by the Gaussian
ensembles of random Hermitian matrices [1,2,3],

(28)

here the product of eigenvalue differences arises as a J a.cobian like in eq. (24) while the
Gaussian factor is due to the confining potential 1/2 .Lm E?;, in thc additional constant
of the motion H; in writing down eq. (28) I have absorbed the Lagrange parameter
a.ssocia.ted with H in the energy scale.
No proof is available right now for the conjccture that the complete equilibrium
ensemble (which spccifics the submanifold of the iJ,~ dimensional phase space accessible
to a trajectory) also implies the eigenvalue density eq. (28) in the limit N-+ =·
QUANTUM SIGNATURES OF CHAOS 593

3 Relaxation of Spectra into Equilibrium

Fluctuations within quantum spectra often change their character when a perturbation
is switched on. Transitions from non-generic to generic behavior for either integrable or
non-integrable systems, crossover from level clustering to level rep11lsing upon clestroying
integrability, change of universality dass when breaking or restoring an anti-unitary
symmetry are examples frequently met with.
The level dynamics discussed in the previous section allows to interpret such tran-
sitions as relaxations into equilibrium. (Needless to say, the apparent contracliction
between classical Hamiltonian clynamics of the fictitious gas and effective relaxation of
certain "macroscopic" quantities is the same here as for real many-particle systems ancl
so is its resolution.)
For the sake of concreteness I shall now clescribe a particular relaxation, the change
of the level spacing distribution P(S, A) accornpanying thc brcaking of time revcrsal
invariancc in a Hamiltonian H( A) = H 0 +),V. To that end I assume H 0 invariant und er
some anti-unitary T but V not so restricted and H(,A) classically non-integrable for any
value of \ moreover, with respect to their separate spectra H 0 and V are taken to be
typical numbers of the Gaussian orthogonal ancl unitary ensembles (GOE and GUE),
respectively. Fluctuations in the spectrum shoulcl Ums reflcct a transition from the
orthogonal to the unitary universality dass.
As a joint density of all matrix elements of H(A) I may consicler [18,3]

P(H, ,A) = (& (H- (Ho+), V))), (29)

where the brackets denote an average over the GOE with respect to H 0 and over the
GUE with respect to V. After performing the N 2 fold integration ovcr thc elcments of
V I obtain the convolution

P(H, ,A) = JdHoPcoE(Ho)PcuE H-~) ;,- ~ .

( --),- (30)

At this point it is important to recall that PcuE(H) is a product of N 2 Gaussians with

aseparate factor for each matrix element. It follows that PcuE((H- H 0 )/,A),A-N' can
be interpreted as the Green's function of an N 2 dimensional free diffusion process with
), 2 as the time; indeed, for ), --> 0 that function shrinks in width to the N 2 dimensional

delta function &(H- H0 ) and then correctly reproduces P(H, 0) = PcoE(H) in eq. (30);
for ), --> oo, like the Green's function of any free diffusion it shrinks to z;ero height
and expands to infinite width, kccping constant global weight with respect to all of its
independent variables. Similarly "sick" is the behavior of the convolutecl distribution
at !arge ),,P(IJ,),)--> ;,-N'PcuE(H/A). Of course, P(H,A) must fail to approach a
stationary limit as ), --> oo since the Hamiltonian H = H 0 + ), V --> ), V itself grows
indefinitely with the coupling constant ,\; that problem was already considcred at the
end of the prcvious section.
With the aim of kecping thc mean density of Ievels constant as ), is varied I again
rcscalc the cnergy by a function g(,A) suchthat eqs. (29) ancl (30) are replaced by

P(II, ,A) = (8 [JI- g(Ho +),V)])

J dH0 PcoE(Ho)(,Ag)- N' Pc:;UF; (H-AggHo) · (31)

 F.HAAKE

l3y imposing the bounda.ry conditions

). --+ { 0 for >. --+ 0

g 1 for >. --+ oo (32)

I secure the desired limiting behavior for the interpolating distribution

P(H, >.) --+ { PaoE(H) for A --+ 0 (33)

PauE(H) for >. --+ oo

As already indicated above the rescaling function g should be fixed so as to enforce

vVigner's semi-circle law [1,2,3] for the mean dcnsity of Ievels, which holds at >. = 0 and
A --+ oo, in the whole interval 0 :::; A < oo. A slightly indirect way of implementing
that goal is to allow for a new "time" T, i.e. for two functions JI(T) and g(T), and to
require that (AgtN 2 PmTE((H- gHo)f>.g) =
G(H, TjHo) remain the Green's function
of some l'viarkovian random process, i.e. obey the Chapman-Kolmogorov equa.tion [19].
In contrast to the free diffusion encountered for g = 1 the new Markovian process
respecting the boundary conditions eq. (32) rnust necessarily involve genera.li;oed forces
preventing the Ievels of the family of rcsca.lcd Hamiltonian g(Ho + AV) frorn flying apart.
In fact, thc Cha.pman-Kolmogorov equation yields two functional equations for ,\( T) a.nd
G( T) which, tagether with the boundary conditions eq. (32) have the unique solution
>.(T) 2 = e 27 - 1,g(T) = e- 7 • The resulting random process is the N 2 dimensional
Ornstein- Uhlenbeck proccss [19]: cach matrix element Hij beha.vcs likc the amplitude of
an overdamped harmonic oscillator driven by Gaussia.n white noise. The difference to
the free diffusion encountered before the rescaling lies in the systema.tic linear restoring
force effective for each matrix element; that restoring force is analogaus both in origin
and in its e1Tects to the one encountered for the modified Pechukas gas in the previous
section.
Within Random Matrix Theory the matrix-valued Ornstein- Uhlenbeck process has
long been known a.s Dyson's-Brownia.n-motion model [20,2]. Originally introduced by
Dyson as an ad hoc dynami;oa.Lion of thc Gaussian ensemble, it now reappears with its
status slightly elevated: since the time T us uniquely related to the control para.mcter ,\
in the quantum Hamiltonian the model amounts to a rigorous description of the depen-
dence of fiuctuations in the spectrum of g(H0 + JIV) on A. From the point of view of the
fictitious ga.s dynamics discussed in the previous section the Ornstein- Uhlenbeck process
appears due to (i) a. pa.rticular non-linear reparametrization of the "time" and (ii), more
irnportantly, a "tcleological" average: the Green's function in eq. (31) ows its Gaussian
character to the average of the delta clistribution over the final Gaussian unitary ensem-
ble. One may think of o(H- g(H0 +>.V-)) as tracing out a single trajectory in an N 2
dimensional space while the Grccn's function rcfers to a bundlc of such trajectories, the
bundle bcing dcsigned faithful to the asymptotic ensembletobe approached as >. --+ oo.
Incidentally, the Gaussian nature of both PaoE and PauE in eq. (31) is responsible
for Wigner's semi-circle law to hold for 0 :::; A < oo or, equivalently, 0 :::; T < oo; more
specifically, the width of the convoluted Gaussia.n for ea.ch off-diagonal matrix element,
given by the mea.n va.lue of (ReHij) 2 + (ImH;j) 2 , is independentofT and therefore the
traclitional construction of the semi-circle law goes through unchanged [3].
H is intcrcsting to note that the Ornstein- Uhlenbeck process for the matrix H =
g(H0+>.V) contains a closecl subdynamics for the eigenva.lues E;( T) which is also Marko-
QUANTUM SIGNATURES OF CI-lAOS 595

vian in character [19]. That process may be clescribed by the Fokker- Planck equation

a N
ihW(E,T)=L { a
8E ( E;+2LE-E
1 1 ) 8
+8E
2
2
}
vV(E,T). (34)
i=l ' j ( #i) ' J '

The confining effect of the rescaling is manifest in the linear restoring force while the
pair intcraction force takes care of Ievel rcpulsion. In vicw of that lattcr intcracLion thc
model in consideration is often referred to as to Dysons's Coulomb gas.

References
[1] Statistical Thcory of Spectra, C.E. Porter, ed., Acaclemic Press, New York (1965)
[2] M.L. Mehta, Ranclom Matrices and the Statistica.l Theory of Spectra, Academic
Press, New York (1965)
[3] F. Haakc, 'Quantum Signatures of Chaos', Springer-Verlag, Berlin, to appear (1990)
[4] KP. \Vigner, Grm1p Theory and its Applications to the Quantum Mechanics of
Atomic Spectra, Acaclemic Press, New York (1959)
[5] M.V. Berry, in: Chaotic Behavior of Deterministic Systems, G. looss, lUI.G. Helle-
mann, and R. Stora, ecls., Les Hauches Session XXXVI, 1981, North Holland, Am-
sterdam (1983)
[6] R. Scharf, B. Dietz, l'vi. Küs, F. Haake, ancl M.V. Berry, Europhys. Lett. 5, 38:3
(1988)
[7] P. Pcchukas, Phys. Rev. Lett. 51, 943 (1983)
[8] F. Haake, M. Küs, ancl R. Scharf, Lecture Notes in Phys. 282, F. Ehlotzky, ecl.,
Springer, Berlin (1987)
[9] l\1. Kt1s, to be publishecl
[10] F. Calogero ancl C. Marchioro, J. Ma.th. Phys. 15, 1425 (1974)
[11] 13. Sutherlancl, Phys. Rev. A 5, 1372 (1972)
[12] J. Moser, Aclv. Math. 16, 1 (1975)
[13] M. Kus, Europhys. Lett. 5, 1 (1988)
[14] P.D. Lax, Comm. Pure Appl. Ma.th. 21, 467 (1968)
[15] T. Yukawa, Phys. Rev. Lett. 54, 1883 (1985); Phys. Lett. 116 A, 227 (1986)
[16] B. Dietz and F. Haake, Europhys. Lett. 9, 1 (1989); Z. Phys., to bc publishccl
[17] F. Haake ancl G. Lenz, tobe publishecl
[18] G. Lenz ancl F. Haa.ke, to be publishecl
[19] H. Risken, The Fokker Planck Equation, Springer, ßerlin (1984)
[20] F. Dyson, .J. Math. Phys. 3, 140 (1982)