DYNAMICS AND STATISTICS OF UNSTABLE ... - Science Direct

13 downloads 0 Views 2MB Size Report
Abstract: The statistical theory of spectra formulated in terms of random ... Statistical properties of a quantum system were found to be profoundly interrelated.
Nuclear Physics A504 (1989) 562-588 Nosh-Holland, Amsterdam

DYNAMICS

AND STATISTICS

V.V. SOKOLOV Insfifufe

OF UNSTABLE

QUANTUM

STATES

and V.G. ZELEVINSKY

of Nuclear Physics, 630090, Novosibirsk, USSR Received

22 August

I988

The statistical theory of spectra formulated in terms of random matrices is extended to unstable states. The energies and widths of these states are treated as real and imaginary parts of complex eigenvalues for an effective non-hermitian hamiltonian. Eigenvalue statistics are investigated under simple assumptions. If the coupling through common decay channels is weak we obtain a Wigner distribution for the level spacings and a Porter-Thomas one for the widths, with the only exception for spacings less than widths where level repulsion fades out. Meanwhile in the complex energy plane the repulsion of eigenvalues is quadratic in accordance with the T-noninvariant character of decaying systems. In the opposite case of strong coupling with the continuum, k short-lived states are formed (k is the number of open decay channels). These states accumulate almost the whole total width, the rest of the states becoming long-lived. Such a perestroika corresponds to separation of direct processes (a nuclear analogue of Dicke coherent superradiance). At small channel number, Ericson fluctuations of the cross sections are found to be suppressed. The one-channel case is considered in detail. The joint distribution of energies and widths is obtained. The average cross sections and density of unstable states are calculated.

Abstract:

1. Introduction Statistical

methods

are now providing

deep insight into the physics of real quantum

systems. They arise from structure uncertainties instance, disordered solids or atoms in random complexity systems Statistical

of the wave functions.

but, in some conditions, properties

with the dynamical

of a quantum stability

of systems under consideration (for external fields) as well as from the

The latter can take place not only in many-body even for a small number system were found

of phase space trajectories

of degrees

to be profoundly of a corresponding

of freedom. interrelated classical

system. One can schematically subdivide existing statistical approaches ‘) into global and local ones. Theories of the first kind deal with wide energy regions. The main statistical quantity here is the level density p,(E) for given energy E and other exact integrals of motion 1. It results in the coarse description, commonly using thermodynamical concepts. Actually, one should take an ensemble average over typical eigenstates of the given hamiltonian. The Bohr idea of a compound nucleus ‘) is an excellent example. In the local-type approach ‘> a sequence of closeiy spaced levels is considered and the main subjects of interest are correlations and ffuctuations of quantities relating 0375-9474/89/$03.50 @ Elsevier Science Publishers (North-Holland Physics Publishing Division)

B.V.

V. V. Sokolov,

to specific energy

As for the usual

it is of minor

statistical works

states.

importance

spectroscopy

of Wigner

were begun

V.C. Zelevinsky

dependence

563

of state properties

development

Dyson “) although

long before ‘)_ Usually

and statistics

here. This field can be naturally

“). Its extensive

‘) and

/ Dynamics

empirical

one considers

was initiated studies

on excitation referred

of local

the statistics

to as

by pioneering regularities

of energy

levels and

matrix elements of simple operators “). The progress in this area has provided us with an understanding of the difference between regular and stochastic quantum systems, as connected with quasiperiodic or ergodic behaviour of co~esponding classical systems 9-‘t). The random matrix theory 3,6,‘2) proved

to be an adequate

tool for investigating

local statistics of quantum states. This method implies a higher degree of abstractness in comparison with Gibbs thermodynamics and reveals the most profound properties determined by general features of dynamics (hermiticity, unitarity, 7’-invariance) rather than by specific details. Stochasticity of motion gives rise to the hypothesis of equivalence of bases in the space of states. It means that the distribution function of matrix elements of a hamiltonian is invariant with respect to orthogonal transformations. Thus, one obtains the gaussian orthogonal ensemble (GOE) of random real symmetrical matrices which gives a good description of stochastic motion in actual

systems

(for example,

isolated

neutron

resonances

in nuclei)

as well as in

model calculations. Lifting the T-invariance restriction one comes to the gaussian unitary ensemble (GUE) of complex hermitian matrices. As it is well known, GOE and CUE predict respectively linear and quadratic level repulsion at small spacings. The repulsion reflects the possibility of level crossing only on a manifold of zero measure in the space of random matrix elements. Actually, excited states of physical systems have a finite lifetime. An unstable state is created by an external field and decays afterwards via open channels. Strictly speaking, the above-mentioned standard gaussian ensembles are applicable to discrete stationary levels only. The influence of coupling with the continuum on the properties of internal states has not been su~ciently understood up to now. In fact, such effects have been considered only in some numerical calculations (see e.g. ref. 13)) of nuclear reactions. On physical grounds, one should expect that even in the case of separated resonances (T/L2 % 1) the level statistics will be modified due to instability.

In particular,

one can guess that the level repulsion

at small distances

=zT should be washed out. The remote classical analogue for this phenomenon is originating from non-crossing phase the possibility of intersection of “corridors” space trajectories of conservative motion as a result of dissipation. In the opposite case of overlapped resonances (r/D* 1) the whole pattern is set by the openness of a system, namely by external mixing via continuum 14). This situation is new for statistical spectroscopy. Since unstable states emerge as intermediate ones in various reactions, the complete statistical description calls for extra hypotheses concerning reaction amplitudes. Assuming that the decay of the internal state n into the channel a performs an

V. V. Sokolov,

564

analysis

of a complicated

its component probabilities Thomas

matched within

distribution

V.G. Zelevinsky

wave function to this channel,

/ Dynamics

and extracts

the modulus

one can also obtain

the scope of Wigner-Dyson 15) for neutron

and stabtics

widths

theory. of isolated

squared

the distribution

IAz12 of of decay

In such a way, the Porterresonances

arises

in good

agreement with experiment ‘1. In contrast to this approach, we use below independent statistical assumptions on internal states and decay amplitudes. For separated resonances nevertheless we come to the same Porter-Thomas width distribution. For many years, statistical reaction theory and statistical spectroscopy were being advanced practically independently. direction of a combined description

Only recently the progress was achieved in the of reaction amplitudes and cross section together system 16--‘R).Having in mind primarily investigation

with levels of an open quantum of unstable intermediate states we will pursue

essentially

the same aim of combined

description using methods generalizing those of traditional statistical spectroscopy. Such as approach leads to the problem of spectral properties of nonhermitian random matrices. Ensembles of those matrices provide generalization of gaussian ensembles of hermitian hamiltonians to unstable systems. We proceed from the general theory of resonance nuclear

reactions

“). In good

approximation reaction amplitudes can be represented as sums of pole terms in the complex energy plane. Such poles 8, = E, --iiF,, being the eigenvalues of the non-hermitian effective hamiltonian X correspond to unstable intermediate states with energies E, and widths r,. We adopt simple statistical hypotheses concerning matrix elements of hermitian and antihermitian parts of X. If coupling of internal states with the continuum is weak (small external mixing) and resonances do not overlap, r/D Q 1, our assumptions give the same results theory. The only significant difference is the disappearance small distances

as the Wigner-Dyson of level repulsion at

1Em - E,,] -Cmax {r,,, r,,}.

In the opposite, limiting case of strong external mixing, the results are governed by the algebraic structure of the anti-hermitian part of %. This structure follows, in fact, from the unitarity condition. Therefore its consequences are of a general nature. Due to this structure,

sharp

redist~bution

(“collectivization”)

of widths

occurs

at

r/D 5 1. As a result, k rapidly decaying states are formed (k is the number of open channels) reflecting segregation of fast “direct” processes. The rest of the intermediate states are long-lived and have small excitation cross sections. This discrimination of widths is consistent with the picture 20) of the two-step course of nuclear reactions in the region of overlapping resonances. The similar phenomenon-formation of a short-lived coherently decaying state in a system with almost degenerate levels - is known in quantum optics as Dicke superradiance. Such a state of N identical two-level atoms coupied via the common radiation field has a width r = A+ where y stands for the width of the excited level of a single atom. In sect. 2 we discuss the phenomenological non-he~itian effective hamiltonian X and its connection to the scattering matrix 3 arising from the general reaction theory. In particular, it is shown that the number of nonzero eigenvalues of the

V. V. Sokolov, V.G. Zelevinsky 1 Dynamics and statistics

anti-hermitian different

part

of 2

can not exceed

forms of the secular

sect. 3. In the case of strong state is demonstrated R-matrix

equation external

for one-channel

and compare

different

the number

for complex mixing,

scattering.

k of open

energies

the width

channels.

Two

8, are considered

accumulation

We introduce

representations

565

in

at a single

the phenomenological

of the scattering

matrix.

Sects. 4

and 5 are devoted to discussion of our statistical hypotheses concerning the distributions of matrix elements of hermitian and anti-hermitian parts of X The distributions of corresponding eigenvalues are analyzed. In sects. 4 and 6 the ensemble averages of R- and S-matrices are found as well as the averaged cross sections of scattering and absorption r,, is obtained of complex

processes. The joint distribution function for energies E, and widths in sect. 7 for the one-channel case. In particular, quadratic repulsion

energies

is revealed.

Due to the unitarity

condition,

residues

in the poles

of an amplitude for the one-channel scattering can be expressed in terms of complex energies 8, only. Therefore the statistics of residues follow from the level statistics in this case. Finally, the mean density of unstable states in the complex energy plane is found in sect. 8. Both cases of weak and strong examined.

2. Effective hamiltonian

external

mixing

are explicitly

and scattering matrix

Suppose a complex quantum system (for instance, a highly excited atomic nucleus) near an excitation energy E is considered. Let the given energy region contain N resonance states studied with reactions a + b (a, b = 1,2, . . . , k are open channels). In accordance with ref. 19), the scattering matrix

sha(E) = 8’” - iTb”(E) can be expressed in terms of amplitudes AZ connecting n (n = 1,2,. . . ) N) with a channel state a as follows:

(2.1) an internal

basis

state

(2.2) Here the N x N matrix

in the internal

space

X= H -tiW plays the role of the For simplicity, we do ing contribution can The anti-hermitian

(2.3)

effective hamiltonian; matrices H and W are both hermitian. not take into account pure potential scattering. The correspondbe introduced without difficulties. part W of (2.3) has the form W,,,,=

; A”,A:*, o=,

(2.4)

566

V. V. Sokolov,

V.G. Zelevinsky

where the sum goes over k open channels. (2.1) [ref. ‘“)I. Generally energy dependent.

speaking,

However,

is high in the energy matrix

that reaction

thresholds

Al and

amplitudes

within

elements energy

are negligible

far from this interval).

of S-matrix H,,

are

since the level density

in, the regular

this region

are sufficiently

matrix

is rather smooth,

we are interested

elements

and statistics

Eq. (2.4) assures the unitarity

this dependence

region

hamiltonian

/ Dynamics

variations

(one should Hen&,

of

assume

one can restrict

oneself to considering only the explicit pole energy dependence in (2.2). This means we can use the matrix % as an effective hamiltonian for studying local dynamical and statistical properties of an unstable system. For convenience, we introduce operator notations in the channel space marking N x k matrices

them with a caret. Using rectangular (2.2) and (2.4) in the compact form ~(E)=A.~;_L_A,

Separability

of W causes the important

A = {At} we write down eqs.

W=AA’.

for k < N the matrix

consequence:

W cannot

have more than k nonzero eigenvalues. Indeed, columns of the matrix A are N-dimensional vectors in the inner space. Provided that k < N these vectors span the k-dimensional subspace of the state space. One can choose the remaining N - k basis vectors to be orthogonal to the columns of the matrix A. Then it is clear that the rank of the matrix W is equal to k. It can be shown readily that the nonzero eigenvalues of W coincide with those of the k x k matrix ~=A+A

in the channel

space.

Its elements

X’” =I,,

(2.6) Ai*Az

are scalar

products

of k N-

dimensional vectors A”. For definiteness, we will consider below the T-invariant theory. As it is known, then the choice of channel states is feasible making the S-matrix symmetric. Without loss of generality, make

the inner

the effective

transposed

basis

hamiltonian

matrices).

As a result,

can also be chosen 5Y symmetric,

for the T-invariant

XT= 2

(the

AZ become

the amplitudes

superscipt

real, At = AT.

N eigenvectors

spond

r,!~“”(n = 1,2, . . . , N) of the non-hermitian hamiltonian to exponentially decaying states and the complex eigenvalues 8, = E, -$ir,,

where

(Z,),,

= S,,8,.

Symmetry

!JC&w-

Z! corre-

(2.7)

give their energies E, and widths r,,. With the notation composed of columns (Cl(“),one can write the diagonalizing Z=

case to T marks

)

V for the N x N matrix transformation

of % as (2.8)

of Z leads to pTW=l=C@T.

(2.9)

V. V. Sokolov,

Here second

the first equality

KG. Zelevinsky

expresses

condition.

X is non-hermitian.

of the existence

of common

of the eigenvalue

problem

and statistics

the orthonormalization

one is the completeness

hamiltonian

/ Dynamics

condition

At the same time

Thus eigenstates

decay channels.

567

whereas

!P’!P # 1 since the

$“” are non-orthogonal

Multiplying

the

the equation

by ?P’ from the left and the conjugate

because XV = !PX,

equation

by tY

from the right we get ~+W~d-~~~+~~=+(~-~+)W=-jty’W~=-iA”*A”T. Unlike

A, the tilded

(2.10)

amplitudes i=

A”, = 1 A&b’,“’ n

WTA,

(2.11)

are complex, Ai being the decay amplitude of an unstable eigenstate m into a channel a. Eq. (2.10) is nothing but the matrix form of the Bell-Steinberger relation 22). After diagonalization of resonance terms

of 2, the scattering

matrix

(2.2) takes the form of a sum

ha (2.12)

> Thus,

in adopted

approximation,

reaction

amplitudes

are meromorphic

functions

in the complex energy plane having poles in the points of eigenvalues of X. Residues in the poles are complex. It is essential for the interference of close resonances 23,24).

3. Secular equation and R-matrix. Complex

energies

(2.7) of intermediate

“Collectivization”

of widths

states are roots of the secular

equation

Det(g--X)=0. The structure

(2.4) of the operator

Let us introduce respectively,

the resolvents

G0(8)=(SS-~))‘, In what follows

(3.1)

W guarantees G’(8)

and

non-negativity G(8)

%‘=E-$iT.

G(8)=(%-X))‘,

we will also need the k x k matrix

of the widths

of the matrices

in the channel

I’,,.

H and

SY

(3.2)

space (3.3)

R( ZY)= ATGO( ‘@A

similar to the Wigner R-matrix of the nuclear reaction theory. Matrix elements of R( 8) have poles in the real points 8 = 8, of eigenvalues of the hermitian part H. Eq. (3.1) can be transformed as follows Det(g-X)=Det(SS_)Det[1+fiG”(8)W] = Det (SZ- H) exp {Tr log [ 1 +fiG’(

8)AAT]}.

(3.4)

V. V. Sokolov,

568

V.G. Zelevinsky

/ Dynamics

It is easy to verify that one can permute cyclically result one reduces the secular equation (3.1) to

and stafistics

the matrix

AT under

log. As a

det[l+$R(8)]=0, where

det stands

for the determinant

(3.5)

in the k-dimensional

channel

space.

In the

practically interesting case k < N, the new equation (3.5) is simpler than the original one (3.1). Using the separability of W we get the Dyson-type relation connecting resolvents (3.2) for stable and unstable systems c(g) Insertion

= cO($)-~iG0(8)~[l+~iR($)I-‘~~~0(8).

of eq. (3.6) into eqs. (2.5) and (2.1) gives f(E)

= ATG(E)A

g(E)

In addition,

(3.6)

the useful

formula

=

=

i(E)

1 +$iR(E)

(3.7)



1-$6(E) 1 +;iR(E)

(3.8)



for the trace of the Green

function

TrG(8)=TrG0(8)+$K(~) follows from (3.6) where the function

(3.9)

K( 8) is given by the trace (Tr) in the channel

space (3.10) As should

be the case, complex

of eq. (3.5). The contributions cancel out.

poles in eqs. (3.6)-(3.10) of real poles

coincide

with the roots

8 = E, from the two terms of eq. (3.9)

In the simplest one-channel case the R-matrix (3.3) turns out to be just a meromorphic function with positive residues (Wigner R-function) and the secular equation (3.5) takes the form l+$R(8)=l+$Tr[G”($)W]=0.

(3.11)

If 8, = E, -@, is a root of eq. (3.11) then 1 -$R( 8*) = 0, i.e. the one-channel S-matrix (3.8) vanishes at 8= 8 :. Taking into account the asymptotic condition lim E+m S(E) = 1 we obtain the factorized representation (3.12) used repeatedly in the literature [see for example ref. 26)]. This representation is equivalent to the sum (2.12) over poles, provided that the complex parameters 8, are roots of the secular equation (3.11). However, the expression (3.12) remains

V.V. Sokolov, KG. Zelevinsky / Dynamics and statistics

unitary

for arbitrary

unspecified

8,. If one takes,

parameters,

cease to be equivalent.

as in ref. 26), energies

then the two representations

En and widths

of the one-channel

The noted fact often slips away. Meanwhile

ation (3.12) with arbitrary

8, may become

incompatible

of W stressed in sect. 2. Such a non-equivalence is especially important ances when the anti-hermitian part W dominates.

569

r,

as

S-matrix

the parametriz-

with the algebraic

structure

in the case of overlapped resonIn the basis of eigenvectors of H

one has (3.13) Accordingly,

the secular

equation

(3.11) can be written

as a pair of coupled

real

equations Y”E” arc arc n (&ET;2+$2=1’ ” (E-E,)2+$2=E*

In the limiting

case of the completely

degenerate

immediately that there exists the unique with nonzero width

eigenstate

operator

H = E * 1, one obtains

of the effective

hamiltonian

r=Cy,=TrW=w n

5%’

(3.15)

accumulated over all individual widths y”. The remaining N - 1 states remain stable. A similar situation ruling out the arbitrary choice of widths r, always occurs if external mixing is strong. Indeed, it follows from (3.14) that if the summarized width w exceeds the energy range AE where levels a, are located, then the broad level with the width ri - w emerges whereas the remaining widths prove to be small. In the limits w/AE

s 1 and N > 1, assuming

the distribution

of E, to be homogeneous

and

symmetric around E = 0, one can obtain easily for the broad level E, = (llW)C”Y&” -O,~,=~[1-O((AE/w)~)].Asforthenarrowlevels(n=2,3,...,N), their energies E,, occur intermittently between neighbouring a, and their widths r,

- (AE/ w)~w/N complement r, to the total value w. For strong external mixing the more convenient form of the secular equation be obtained with the help of the orthogonal transformation diagonalizing

can the

anti-hermitian part W. Since in the one-channel case the matrix W has N - 1 degenerate zero eigenvalues, one can diagonalize in addition the (N - 1) x (N - 1) submatrix of the hermitian part H with the extra orthogonal transformation of the (N - 1)-dimensional subspace. As a result, eq. (3.11) takes the form %-h-lZ($)+$w=O,

(3.16)

where notations

h=H,,,

h, = HI, = Hv,

(3.17)

V.V. Sokolov, V.G. Zelevinsky / Dynamics and statistics

570

are introduced,

E, being

eigenvalues

At h, = 0 only one level is unstable the widths function

through

admixture

R (3.17)

analogous

continuum states. On the basis of the preceding

of the (N - 1)-dimensional having

a width

of the unstable

one. This mixing

to the function paragraph,

submatrix

w. The remaining

R (3.13)

the scattering

is described

describing amplitude

of H.

levels E, get mixing

by the with

is

W

T(E)=Tr[G(E)W]=

(3.18)

E-h-I?(E)+;iw’

Since l?(E) +oo at E + E, the amplitude (3.18) and the cross section zeroes coinciding with E,. Peaks of the cross section are disposed satisfying

have N- 1 at energies

the equation E-h-&E)=O,

(3.19)

which is nothing but the secular equation for eigenvalues of the hermitian part H. The R-function (3.13) has roots of eq. (3.19) as its poles*. Therefore, independently of the power of external mixing, the one-channel cross section has (neglecting the potential contribution) N peaks alternating with N - 1 zeroes**. However, only in the case of weak coupling via the continuum, the cross section peaks lie, according to (3.14) and (2.12), at the points E, = E, of the energies of unstable intermediate states and have near those points the Breit-Wigner shape with widths r,, = -rn. In the opposite case of strong coupling E,, = F,, (v = 2,3,. . . , N; see eq. (7.36) below) so that now the cross section minima rather than its maxima fall at energies of unstable states. The contributions of the broad and narrow resonances happen to be opposite in phase and cancel out. Outside the interval AE occupied by the narrow resonances,

the amplitude

(3.18) has the Breit-Wigner

shape

with the width

w.

One can conclude that the variables h, E,,, h,, and w are more adequate for parametrizing the energy dependence of the one-channel S-matrix then energies E, and widths r,, of intermediate unstable states. The latter are suitable for analysis of the time dependence

of the process.

These two ways of treatment

are complemen-

tary. We would like to note though that the pole parametrization (2.12) is universal for any channel number whereas the simple properties of the expression (3.18) have no analogues in the many-channel case. Thus, the strong coupling with continuum drives creating a short-lived state. Similarly, if there exist decay probabilities, k broad states are formed summed width. This phenomenon observed in

to sharp redistribution of widths k open channels with comparable absorbing practically the whole realistic calculations of specific

* We recall that the energy shifts occurring in the general theory of nuclear reactions are incorporated into H. l* Such a behaviour of the cross section is typical for the one-channel case only. If, apart from the elastic channel, some inelastic channels are opened the elastic cross section has no zeroes. This is the case also for inelastic cross sections at k 2 3 [ref. “)I. The cross section peaks do not coincide, in the general case, with R-matrix poles. These results follow from the many-channel analogue of eq. (3.18).

V. V. Sokolov,

nuclear

reactions

to a compound of separated at N>l Using

13) implies nucleus.

resonances

hearthevalue the term

VG. Zelevinsky

/ Dynamics

the predominance

Quite abrupt

“phase

to development

and statistics

571

of direct processes transition”

of broad

over relaxation

from the uniform

coherently

decaying

pattern

states occurs

w-AE,i.e.(y)=w/N-D=AE/N. “direct

process”

we actually

have

in mind

any

fast process

independently of its mechanism. The duration h/w of this process is essentially shorter not only than the recurrence time h/D but also than the time interval h/AE of fragmentation of the entrance state into more complicated configurations. On the other hand, we have shown that the lifetime of long-lived states is T(w/AE)h/D% factor w/AE. have

started

h/D. This time exceeds the delay time obtained The discrepance is due to the difference of input from the S-matrix

representation

(2.1),

in ref. “) by the assumptions. We

(2.2) based

on the general

reaction theory. In the one-channel case we have obtained from here eq. (3.18) as a consequence. Hence, the only parameter characterizing the instability of intermediate states is Tr W = w. Unlike this, the explicitly unitary form (3.12) of the S-matrix with arbitrary widths was used in ref. *“). As it has been stressed earlier, the two representations are not equivalent. The large lifetime r of compound states as compared with the recurrence time h/D implies that the system reaches thermal equilibrium. Elastic scattering

through

the single-particle

resonance

coupled

with the back-

ground of stable many-particle levels 29) can be considered as a particular example. The single-particle resonance is formed by the optical shell-model potential. The scattering amplitude found in ref. 29) coincides with (3.18), w being the width of the single-particle resonance. Generally, short-lived states are generated by the anti-hermitian part W of the effective hamiltonian. This part forms superpositions of internal states matched to channel wave functions weakly coupled with the internal

ones.

In fact, such a state is identical

to the doorway

state introduced

in

ref. 30). It is clear from (3.18) how the spreading width r’ of the doorway state arises. The influence of the ocean of long-lived states is described by the function k(E). If one replaces the summation over v in eq. (3.17) by integration, this function gets the imaginary

part corresponding

to the width r’ = 27r(ht)/

D - AE. Therefore,

in our case rt - w B r’ - AE. As it has been mentioned

in Introduction,

width collectivization

induced

by strong

coupling via common decay channels (“self-organization” observed in numerical calculations “)) can be considered as a nuclear analogue of Dicke coherent superradiance 2’). Similar effects were found in numerical simulation 3’) of a dissipative spin system.

4. Statistics of the hermitian part of the hamiltonian; In the preceding logical hamiltonian

averaging over GOE

sections we have discussed dynamical properties of a phenomenoX of an unstable system. Now we proceed by treating 2 as a

V. V. Sokolov,

572

member

of an ensemble

postulates

consistent

We consider independent.

V.G. Zelevinsky

of random

with general

the hermitian We suppose

/ Dynamics

matrices.

physical

We assume

N x N matrix

for matrix

H belongs

to the

and their distribution function Such assumptions are known

elements

N’N+‘l/nexp( Here the centre joint distribution

statistical

W parts of X to be statistically

H and anti-hermitian that a real symmetric

distribution

the simplest

principles.

GOE, i.e. matrix elements H,,, are not correlated invariant with respect to orthogonal transformations. to imply the gaussian

and statistics

is ‘)

of H, -$TrH’).

of the energy scale is placed at the origin. function of N eigenvalues .el, . . . , Ed

(4.1)

Eq. (4.1) leads to the

(4.2) The normalization

constant cN=2

in (4.2) is equal 32) to

N(N-1)/4

N(N+l)/4 1 _N ,[!*I.&)]-‘. ( a2 >

As in ref. I*) we have introduced

the scale factor

N in the exponent

(4.3) of eq. (4.1).

As a result, in the limit N -+OO the energy levels E, are located within the Nindependent interval (-a, a) the mean spacing being equal to D = 2a/ N. The essential feature of the GOE is the level correlation described by the preexponent factor in (4.2). The probability for two levels to coincide tends to zero in proportion to their spacing (so-called linear level repulsion). If the T-invariance is not supposed (the GUE case) this factor is replaced by the quadratic one “). Due to orthogonal invariance, the wave functions p”” of stationary

states may

be represented in the GOE by N-component unit vectors {cp’,“‘}spread isotropically over the hypersphere C,,, (cp!,!,‘))‘= 1. In the large-N limit, it gives the gaussian distribution of components cp’,“’with zero mean values and variances ((cpi,“‘)‘) = l/N. Postulating that amplitudes AZ have the same statistical properties as components cp’,“’one easily obtains

the xk-square

distribution

of decay widths where the number

of degrees of freedom k is equal to the channel number (Porter-Thomas distribution for k = 1). This procedure, however, has no serious justification so we prefer to derive the width distribution with the aid of independent statistical hypotheses concerning the anti-hermitian part W (see the next section). Computation of mean values of various quantities over the GOE is simplified in the limit of large N. In particular, in order to average the Green function G’(8) (3.2) one can expand it into a formal series about H and average each item with the use of the pair contraction formula (see eq. (4.1)) (4.4)

V. V. Sokofov, V.G. Zeleuinsky ,I Dynamics and statistics

573

At N + CO,only non-intersecting contractions are surviving 16) since there a trace in the internal space being proportional to N compensates the factor N-’ in (4.4). The summation of those terms gives* 1

(Go(Q)= ~-(u2/4~)Tr(Go(~)}’

(4.5)

From eq. (4.5) we get 33) g’(~)-~(TrG’(g))=~[i(-~~~,

(4.6)

where the sign is determined by the asymptotic condition g”( %‘)= l/ 8 at ‘%: + co. Using expression (4.6) we obtain the mean level density for a stable system p’(E)=

CS(E-E,) (. =

>

=iIm(TrG’(E-i0))

N--$&x2-E2 0(a2- E2)

(4.7)

(the semicricle Wigner rule “)). The analogous GOE - averaging the R-matrix (3.3) leads in the limit N+ co to (R(~P))GoE=(G’(~))~~~‘(~)JZ,

(4.8)

where the matrix ?? has been defined by eq. (2.4). In particular, in the one-channel case

(NWm=g”(~‘)~.

(4.9)

In the similar manner averaging the function I?( ‘8) (3.17), we find

For GOE-averaging the S-matrix connected with the R-matrix by the non-linear relation (3X), we note that at N + a3 contractions (4.4) of random matrix elements belonging to different R-matrices reduce the number of independent traces leading to contributions of higher order in l/N If the channel number k < N then the smallness of l/N can not be recompensed by a trace in the channel space arising when contracting two matrix elements of H from neighbouring R-matrices. Therefore, up to terms of order k/N, 1 -$i,q”( E)??

&%,E = 1 +4ig”(E)2. * This method

is applicable

also to h~miltonians

containing

a regular

(4.11) part along with a random

one.

V. V. Sokoiov, V.G. Zelevinsky / Dynamics and statisiics

574

As one

can

compound the original “direct”

see, the GOE-averaging states

so that the result

matrix

W. In other

has smoothed depends

words,

contributions

only on the matrix

the average

S-matrix

2

of long-lived rather

than

(4.11) describes

on fast

processes.

5. Statistics To investigate

unstable

of the anti-hermitian states of an open

part of the hamiltonian system,

one is in need of information

about Nk real (due to the T-invariance) matrix elements Ax of the total hermitian hamiltonian between internal (n) and channel (a) states as well as about internal matrix elements H,, . We will assume that the statistical properties of these quantities are similar

and can be described

As in (4.4) we have introduced

by gaussian

distributions.

here the scale factor (w)=(Tr

W)=C

l/N

So we suggest

that*

so that the average

va

trace (5.2)

a

is independent of N serving as an invariant characteristic parameter of decay channels under study. As in eq. (4.1), the gaussian distribution of amplitudes AP, follows practically uniquely from the statistical independence and orthogonal invariance in the internal space. Strictly

speaking,

the hypothesis

of statistical

independence

implies

ignoring

the

direct inelastic reactions. However, it is not too difficult to generalize the consideration to include these reactions in our scheme using the method of ref. 34). For simplicity, we suppose open channels to be equiprobable (TV = 7, (w) = kv). Then additional orthogonal invariance in the channel space arises resulting in the joint amplitude distribution function P(Af,

. . .,A:)=(s)““‘*exp(

-Ew)

(5.3)

depending on w = Tr W = C,, (A:)* only. The distribution (5.3) allows us to derive statistical properties of the matrix elements W,,,, (2.4). The diagonal elements W,,, = C, (AZ)2 = ‘y,, are positive, statistically independent and distributed by the Xk-square rule, (5.4) l In the paper by Weidenmiiller I’), it is supposed that the amplitude randomness takes place in the eigenbasis of the hermitian part H only. However, it is difficult to adjust this assumption to orthogonal invariance of the GOE of hermitian parts.

V. V. Sokolov, V.G. Zeievinsky / Dynamics and siatisiics

This derivation

needs

wave functions

(sect. 4). With the channel

tends to the gaussian

no identification

of decay

amplitudes

number

575

with components

increasing,

of

the expression

(5.4)

distribution

ok

-~(Y+l)‘)

= $&JXP{

(5.5)

with the mean value and the variance

(5.6) respectively. These equations were used frequently 35) in order to extract the effective number of open channels from experimental data. Such a procedure can be meaningful only if the is then a weak The trace w its distribution

widths are small and resonances are not overlapping. The matrix W perturbation and the y,,‘s indeed can be treated as resonance widths. of the matrix W is a sum of independent random quantities yn and function (Nk-I)/2 gk(W)

exp

=

(

at Nk s 1 turns

into gaussian

N --w 2rl

(5.7) )

one (5.8)

with the mean value (w) = kv and the small variance

As for off-diagonal 1W,,I

W,,,,, m # n, their

elements

only and can be expressed ~k(Wmn)=L

T(fk)

N

J;;q

(” IW,,,“~)‘kP”‘2 277

(WmJ=O, The analogous

distribution

distribution

function

in terms of the MacDonald

function

Kck-,),2

depends

function (;“I

wm,,~)

(w’,,)=+2=&w)2. of matrix

elements

for a transition

on

K,(Z), ,

(5.10)

(5.11) between

complicated states has been considered in ref. 8). of a certain row or column of W Unlike the original amplitudes AZ, elements are not statistically independent. Variables mutually independent and independent of y,, are the angles f3,, between N k-dimensional vectors A, with components A”, . Using such a representation we have W,j,,=(A,~A,)=Jy,y,cosfL.

(5.12)

576

V. V. Sokolov,

The distribution

of any angle

V.G. Zelevinsky

=’J;;

from the expression

and statistics

em,, is given by the formula

LP’k(e) ensuing

/ Dynamics

(k 2 2)

Qfk) T($(k-1))

for a solid

angle

(5.13)

sink-2e in a k-dimensional

space.

Since a

rotation in the channel space does not change a scalar product (5.12), the total number of random parameters specifying the matrix W is equal to Nk -+k( k - 1) for k s N. For k > N one should add the number i(k - N)(k - N - 1) of rotation generators for the complementary (k - N)-dimensional space getting as a result tN(N + 1) independent of k. The distribution of k non-trivial eigenvalues of the matrix W at k < N is of particular interest. As it has been mentioned in sect. 2 they coincide with k eigenvalues

y* of the matrix

2 = ATA (2.6) in the channel

space.

The distribution sought for can be obtained by the stochastic equation method 36,37). A stochastic process is constructed to describe evolution of the amplitude matrix A in fictitious time r, dA(r)= where the gaussian

random

force is defined

daz=O,

Independently

of initial

-A(T)

conditions,

da: da:

dT+dQ(r),

(5.14)

by

=$ ~“c?~~&,,,, dr.

the distribution

function

(5.15)

of amplitudes

AZ tends

to that of eq. (5.3) at r+ CO. The evolution of eigenvalues ya(r) is determined by (5.14) as well. Diagonalizing the matrix *(r+dr) with the help of perturbation theory with respect to the parameter (dr)1’2 one can get the drift and diffusion coefficients and then write down the Focker-Planck equation for the distribution function of eigenvalues. Its steady solution is (recall that all y’ > 0)

P( y’, . . .

)

yk) =

c,,

(~h/Y~-Yh~)(~YC)‘N-*“‘2exp(

where C, ya =Tr 2 =Tr W = w. The normalization constant methods developed in ref. “) and proves to be equal

-E;

y”),

(5.16)

can be calculated

by

-1

T(k+l)flr(;a)T(;(N-a-l))

I

.

(5.17)

It can be seen from (5.16) that (y”)= 7. When applied to the hermitian part H the stochastic equation method leads naturally to eq. (4.2). The distinction between eqs. (4.2) and (5.16) is caused by the specific form (2.6) of the matrix 2: its matrix elements are finite sums of products

V. V. Sokolov,

of normally

distributed

and the expression

V.G. Zelevinsky

random

variables.

/ Dynamics

and statistics

577

In the limit N + ~0 this difference

-1N 10 x If - yblexp-$z w-T*) 1 I

fades

(5.16) turns into

k(k+l)/4

B(y’,

2”“r(k+l)nr&)

. . .

1 rl

a

n a a outside the domain of compound resonances the two cross sections coincide being equal to (in units of (r/ k2)(2J + 1) where J is the angular momentum and k is the relative initial wave number) a,(E)

= o,.,.(E)

=

=

77* ~ [;(E+JE*-a2)]*++n2 x2

[$(E/a)+JE2/.2-1)]2+,x2’

(El> a.

(6.3)

The parameter x = T/a introduced in (6.3) is a measure of external mixing and resonance overlapping. In the far regions 1E I z=-a, the function (6.3) looks like the wings of a Breit-Wigner resonance with the broad width n (we omit the channel superscript for simplicity).

K V. Sokolov,

578

In

the compound

formation

V.G. Zelevinsky

resonance

/ Dynamics

region IEl< a, the projectile

of an intermediate

compound

system.

$xJlelastic

following

E2/a2+&*+

(6.4)

1)

one

x2

rr,,,.(E)= lwm-112= do not coincide,

can be captured

As a result, the total cross section

x2+xtil-E2/aZ

a,(E)=2(1-ReJS(E)))= and the shape

and statistics

(6.5)

+&E”/a*+$(x’+l)

their difference

rat,,(E) = 1 - kWH/*= being the cross section of absorption accordance with the qualitative analysis

xJ1 -

E*/a*

~XJI-E*/aZ+$(x2+

(6.6)

1)

due to compound nucleus formation. In of ref. *‘), the absorption cross section (6.6)

decreases when the degree of resonance overlap increases. Fig. 1 demonstrates the energy behaviour of the cross sections

(6.4)-(6.6).

shape

form symmetric

-

3.5

elastic

-3.0

and the total cross section

-2.5

-2.0

-+.5

f

-1.0

have the double-humped

The

1 I

-0.5

Cl

0.5

i.0

1.5

2.0

2.5

30

35

E/o

Fig. 1. Energy behaviour of average cross sections; solid lines 1 and 2 correspond to the shape elastic cross section and to the absorption cross section, respectively; the dashed line 3 corresponds to the total .cross section.

V. V. Sokolov,

V.G. Zelevinsky

with respect to the centre of a spectrum. the minimum

is in the centre.

The maxima

The relative

(+,(O) x2+1 _==~l-g,(a) x(x + 1)

/ Dynamics

drops

and starisiics

579

lie at the edges E = *a whereas

of the minima

1

us..%

x’

a,.,.(a) =(X+1)Z=

are

x2+ 1

2

(6.7)

l-2

The approximate equalities correspond to the strong mixing limit when the middle parts of resonance curves flatten. This flattening is the manifestation of the spreading width (see sect. 1).

7. Distribution of complex energies for an N-level one-channel system As we have noted in sect. 5, at k < N the matrix

W is specified

by Nk - qk( k - 1)

random parameters. Recollecting N eigenvalues of the hermitian part H we have altogether I= N( k + 1) -$k( k - 1) parameters on which the N complex energies are dependent. Being interested in the distribution of energies and widths we have to integrate over I-2N = (k - l)( N -ik) extra variables. In the exceptional case of k = 1 the number

1 of original

variables is just equal to the number 2 N of parameters of integration drops out and the problem becomes transformation and calculation of the jacobian. This one-channel

{E,,, T,,} so that the necessity

reduced

to variable

case will be considered below in detail. We use the secular equation (3.11) with the R-matrix with respect to levels and perform the transformation define characteristic polynomials of N x N matrices

in the form (3.3) symmetric {E,, -y,,}+ {E,,, I’,,}. Let us H, X = H -$W and Xt =

H+iiW:

Qo(h)=Det(A-H)=n(h-E,,);

(7.1)

n Q(h)=Det(A-X)=n(A-g,,), n Q(A)=Det(A-X+)=fl(A-%z).

(7.2)

” One can easily verify that due to eq. (3.13)

aQo(A 1

Q(A)=QdA)-$iC %,a~. m

The coefficients

AiN,” of polynomia Qo(A)=

Q(A)=

(7.3)

m

1s (7.1) and (7.2) defined

according

to

; (-l)“A’,N’(~)AN-n, n=O

; (-l)“A’,“‘(8)AN-“, n=O

Q(A)=

; II=0

(-l)“A$NN’(8*)AN-”

(7.4)

V. V. Sokolov,

580

are symmetric

V.G.

Zelevinsky

over their N variables

AhN’(&)= 1 )

AiN’

/ Dynamics

and linear

of coefficients

in each of them: A:N’(E) = 1 EmEn,. . . ) f?l ’ “lJ2J(E,-EJ2+:(r,+rJ2=

(7.30)

2

( > c

n

After integrating resonances 9v2,.

rn

(7.31)

-,r~=2r,~2r~+[(~~2r;)‘-~2r:].

over E, and r, one obtains

the distribution

function

for long-lived

(m, n = 2,3, . . . , N): . . , EN;

r2v..

. , rN)

T(N-4)

[I + Wd

c, miN-1/2

)I .

x exp

(7.32)

To estimate values of typical small widths it is convenient to introduce amplitudes Pn = a (n 3 2) assuming that they run over the whole real axis. Then dr, ;z+rJdB” where

G =C,“=, r,

dimensional density

= G(N-3)‘2 dG do,_,

and d ON-1 stands

for an infinitesimal

space with Pn as its coordinates.

(7.32) has a maximum

(7.33) solid angle of a (N - l)-

It is easy to show that the probability

at (7.34)

whereas the angle distribution in the space of Pn is nearly isotropic. Hence, the individual small widths are of order rnz2 - G,/( N - l), i.e., at large N, r,,, - n/z2N and the corresponding overlap parameter is r/D - (T/x2N)N/a = l/x < 1. As a consequence, long-lived states fall again in the small overlap regime, and, at spacings S,, < l/x, the energy distribution coincides with that of eq. (4.2).

V. V. Sokolov,

584

These results

are evidently

they can be derived The version

(3.16) obtained

study,

/ Dynamics

of dynamical

from the secular

for the case x * 1. Since, limit under

V.G. Zelevinsky

rather

equation

and statistics

than statistical

by diagonalization

of the matrix

due to eq. (4.4), (ht) = a2/4N

we can use the perturbational

%,=h-$w+

c”h -E, ht-$iw

origin.

Therefore

(sect. 3) straightforwardly. W is appropriate

= n2/4x2N

is small in the

formulae:

=h-$w+I?(h-fw),

(7.35)

“-2

h:

q,=&“+

Neglecting (h - .z~)~c u2 as compared (7.36) the large width

with

N

r,=w-w

vs2.

E,-h+$w’

(7.36)

w2= n2 we find from eqs. (7.35) and

ht

(7.37)

U;2(h_E,)2++W2=

as well as small ones (7.38) A similar picture has been discussed by Shapiro 2”) based on quite different arguments. In contrast to that paper, we are not in need of any assumptions concerning the preexistence of narrow resonances on the background of a broad one. The separation of width scales happens automatically by virtue of the structure of the matrix W which is dictated ultimately by the unitarity principle. It should be stressed that in the one-channel case the width rearrangement leaves no room for manifesting Ericson fluctuations 38). Indeed, at N> 1, the conditions necessary for the strong overlap of resonances cannot be implemented. To reveal such fluctuations

there

should

be k B 1 open

channels

when

k short-lived

states

with comparable widths and weakly correlated decay amplitudes arise. To investigate rigorously the cross section fluctuations in the framework of our approach, one should derive the distribution problem elsewhere.

of decay amplitudes

8. Average

density

An important quantity for an N-level complex energies %,, = En -$ir, (compare

(2.11). We hope to return to this

of unstable

unstable (4.7)),

states

system

= C6(E-E,)6(r-I’,) n >(

is the mean

density

of

(8.1)

V. V. Sokolov, V.G. Zeleoinsky / Dynamics and statistics

normalized

according

to dTp(E,

At x < 1 the anti-hermitian order

585

part

8, = E, - $y, . The density

r) = N.

(8.2)

W is a small perturbation

so that in the lowest

p (E, ZJ is then factorized,

P(E, r) = p’(E)%(r)

(8.3)

where p”(E) is the semicircular distribution (4.7) and C?‘,(r) stands Thomas distribution (5.4) with the mean value (r) = n/N.

for the Porter-

In the opposite case of x > 1, it is convenient to use the alternative version (3.16) of the secular equation. Eigenvalues of X can be found by perturbation theory (7.35), (7.36). Separating in eq. (8.1) the short-lived ones (n=~=2,3,..., N) we have

(n = 1) states from long-lived

p( E, ZJ = ;(Sc2’( ‘2’)- 8,)) +; C (S(*)( i!T- Z$,)) . Averaging over h, and E, in the first term (denoted at N + cc like in (4.9) resulting in

(8.4)

by p,( E, I’)) can be carried

pl(E,~)=~(6(2)[~-h+~iw-$a2g0(h-~iw)])h,W. Quantities

h and w are normally

distributed

out

(8.5)

with mean values

(h) = 0 and (w) = n

and variances (h2)= n2/2x2N and ((~-77)~)=277~/N so that we can expand the argument of the S-function in eq. (8.5) over h/w. In the real part, the corrections cancel out up to the second order inclusively result coincides with that of eq. (7.37). Hence,

whereas in the imaginary part the we obtain in this approximation

p,(EJ)=(cT(E-~)~(CW+W/X*)),,,,. Final

averaging

over h and w can be performed

(8.6) trivially

giving

N[C~(1-x-2)]2 v2( 1 - X-*)2

.

(8.7)

Within our accuracy, the energy and the width of the short-lived state are not correlated. This resonance is located near the centre of the region of levels coupled with a given channel. The whole energy region is covered by the resonance width. As for fluctuations of E and r they are relatively small, -l/m. The second term of eq. (8.4), p,(E, r), contains N - 1 identical contributions. Each of them is equal to

(8.8)

V. V. Sokolov,

586

V.G. Zelevinsky

/ Dynamics

and statistics

with notations I=

wE+hT

Now

it is obvious

f=



w+r

wr [a+(g’].

(8.9)

that (see eq. (4.7)) (8.10)

(8.11) The remaining The result is

averaging

over h and w can be done

p2(E,r)=(N-I)~&I~-E~B(u~-E~)

by the saddle-point

method.

(-gr) .

Jsexp

(8.12)

We keep here unity side by side with N to guarantee proper normalization. Thus, for N - 1 long-lived states the semicricle GOE rule (4.7) along with the PorterThomas distribution (7.38), (r) = 7/x2N.

are recovered. However, the latter has a small average width As we have emphasized earlier, overlap of long-lived resonances

is small, (r)/D - l/x. In the general case of arbitrary external mixing one can use the electrostatic analogy to reduce the calculation of the average density (8.1) to the more convenient averaging of the trace of the Green function (see ref. 39)), (8.13) The function (8.13) acquires the meaning of the two-dimensional electric field created by the charge distribution p( E, r). Then, by virtue of the Maxwell equation, (8.14) It allows to perform

the explicit

averaging

A,. The remaining

GOE-averaging

over the gaussian

is still a complicated

ensemble

of amplitudes

task.

9. Conclusion In this paper we have attempted to extend standard statistical spectroscopy of discrete levels to unstable states. Decaying systems are described by ensembles of random non-hermitian hamiltonians represented by N-dimensional matrices. Complex eigenvalues of these matrices determine energies E, and widths r, of states with a finite lifetime. Under simple statistical assumptions corresponding to completely stochastical motion, the eigenvalue distribution in the complex plane (E, r) is obtained. The one-channel case is discussed in detail.

V. V. Sokolov, KG. Zelevinsky / Dynamics and statistics

The instability repulsion

at distances

they repulse increasing

of states changes smaller

quadratically coupling

place. When matrix

significantly

than widths.

with respect

with the continuum, elements

the energy statistics,

As for complex

to the distance dynamical

of the anti-hermitian

587

removing

energies

8, = E, -$I’”

in the complex

restructuring

level

plane*.

At

of a system takes

part of an effective hamiltonian

become comparable with the spacings of eigenvalues of the hermitian part, a transition to a new regime occurs. In the new regime the picture is determined by the algebraic structure of the anti-hermitian part dictated by unitarity rather than by particular statistical hypotheses. As a result, decay probabilities turn out to be redistributed in such a way as to distinguish short-lived states (their number is equal to that of open decay channels) saturating approximately the whole summed width of levels. The remaining long-lived states are quite isolated and their statistics are essentially the same as those of stable states. Such a picture implies the coexistence of direct and compound processes resulting in a gross-structure of the cross section. A similar structure with several short-lived resonance states has been observed in numerical simulations of nuclear reactions 13). In this work formation of such a structure along with the state overlap increasing was referred to as “self-organization”. As one can see its mechanism is of a quite general to Dicke superradiance in quantum optics.

nature,

being analogous

In the one-channel case there is no room for manifestation of Ericson fluctuations. In principle, they might be observed if the channel number is large enough to generate a number of interfering resonances. This problem is under study. The main results of this paper have been published in ref. 4’). We thank Ingrid Rotter for stimulating discussions and making information on her results available to us prior to publication. We appreciate useful discussions with M.G. Urin;

we are grateful

also to F.M. Izrailev

for permanent

interest

to the

work and to L.F. Khailo for help with the calculations. In addition, one of us (V.S.) would like to mention fruitful discussions with V.L. Lyuboshitz, and the other (V.Z.), those with I.V. Krivoshey.

References 1) 2) 3) 4) 5) 6) 7) 8)

A. Bohr and B. Mottelson, Nuclear structure, vol. 1 (Benjamin, Massachusetts, 1969) N. Bohr, Nature 137 (1936) 344 Statistical theories of spectra: Fluctuations, ed. C.E. Porter (Academic Press, New York, S.T. Belyaev and V.G. Zelevinsky, Sov. Phys. Usp. 28 (1985) 853 E.P. Wigner, Ann. Math. 53 (1951) 36; 62 (1955) 248 F.J. Dyson, J. Math. Phys. 3 (1962) 140, 157, 166 1.1. Gurevich, Zh. Eksp. Teor. Fiz. 9 (1939) 1283 T.A. Brody et al., Rev. Mod. Phys. 53 (1981) 385

* This is equivalent to cubic repulsion if one uses polar coordinates in the (E, r)-plane. is confirmed by the numerical calculations “) for a kicked nonlinear top with dissipation.

1965)

Such a rule

588 9) 10) 11) 12) 13) 14)

15) 16) 17) 18) 19) 20) 21) 22) 23) 24) 25) 26) 27) 28) 29) 30) 31) 32) 33) 34) 35) 36) 37) 38) 39) 40) 41)

V. V. Sokolov,

V.G. Zelevinsky

/ Dynamics

and statistics

B.V. Chirikov, Phys. Reports 52 (1979) 263 M.V. Berry, Ann. of Phys. 131 (1981) 163 R. Pechukas, Phys. Rev. Lett. 52 (1984) 1 M.L. Mehta, Random matrices and the statistical theory of energy levels (Academic Press, New York, 1967) P. Kleinwlchter and I. Rotter, Phys. Rev. C32 (1985) 1742 I. Yu. Kobzarev, N.N. Nikolaev and L.B. Okun, Yad. Fiz. 10 (1969) 864; I. Yu. Kobzarev, Proc. of the VIII Winter School at the Leningrad Institute of Nuclear Physics. Part I., Leningrad (1972) p. 298 C.E. Porter and R.G. Thomas, Phys. Rev. 104 (1956) 483 D. Agassi, H.A. Weidenmiiller and G. Mantzouranis, Phys. Reports C22 (1975) 145 J.J.M. Verbaarschot, H.A. Weidenmiiller and M.R. Zirnbauer, Phys. Rev. Lett. 52 (1984) 1597; Phys. Reports 129 (1985) 367 H.A. Weidenmiiller, Ann. of Phys. 158 (1984) 120 C. Bloch, Nucl. Phys. 4 (1957) 503; H. Feschbach, Ann. of Phys. 5 (1958) 357 F.L. Friedman and V.F. Weisskopf, in: Niels Bohr and the development of physics (Pergamon, London, 1955) R.H. Dicke, Phys. Rev. 93 (1954) 99 J.S. Bell and J. Steinberger, Proc. Int. Conf. on elementary particles, Oxford (1965) V.G. Baryshevsky, V.L. Lyuboshitz and M.I. Podgoretzky, Zh. Exsp. Teor. Fiz. 51 (1969) 157 V.D. Kirilyuk, N.N. Nikolaev and L.B. Okun, Yad. Fiz. 10 (1969) 1081 E.P. Wigner and L. Eisenbud, Phys. Rev. 72 (1947) 29 V.L. Lyuboshitz, Communication of JINR, P2-5328, Dubna (1970); I.S. Shapiro, in: Problems of modern nuclear physics, Novosibirsk (Nauka, 1971) p. 273 E.P. Wigner, Proc. Nat]. Acad. Sci. (USA) 32 (1946) 302 V.L. Lyuboshitz, Proc. of the XIX Winter School at the Leningrad Institute of Nuclear Physics, Leningrad (1984) p. 33 M.G. Urin, Fiz. Elem. Chastits Yadra 8 (1977) 817 H. Feshbach, A.K. Kerman and R.H. Lemmer, Ann. of Phys. 41 (1967) 230 R. Grobe and F. Haake, Z. Phys. 868 (1987) 512 M.L. Mehta, Nucl. Phys. 18 (1960) 395 L.A. Pastur, Teor. Mat. Fiz. 10 (1972) 102 C.A. Engelbrecht and H.A. Weidenmiiller, Phys. Rev. C8 (1973) 859 L. Wilets, Theories of nuclear fission (Clarendon, Oxford, 1964) F.J. Dyson, J. Math. Phys. 3 (1962) 1191 S. Tanaka and S. Sugano, Phys. Rev. B34 (1986) 6880 T.E.O. Ericson, Ann. of Phys. 23 (1963) 390 V.V. Sokolov and V.G. Zelevinsky, preprint INPSS-79, Novosibirsk (1988) R. Grobe and F. Haake, preprint, Essen (1988) V.V. Sokolov and V.G. Zelevinsky, Phys. Lett. B202 (1988) 10