Geometrical Properties of Cumulant Expansions
Geometrical Properties of Cumulant Expansions
Geometrical Properties of Cumulant Expansions
Dresden, Germany
arXiv:cond-mat/9709044v1 4 Sep 1997
Abstract
in the Hilbert space. Cases are investigated where this integration can be
cumulant wave operator is derived. In the second part of the article we derive
I. INTRODUCTION
In statistical physics, quantum mechanics, quantum field theory and many other fields
of modern theoretical physics we constantly face objects of the type
Here A1 , A2 , ... are abstract elements, λ1 , λ2 , ... are numbers, h...i means some kind of
averaging and e(...) denotes a generalized exponential function.
1
Assuming the average of 1 to be nonzero, we conclude that F is an analytic function
of the λi in a vicinity of zero. Therefore, it has a Taylor expansion with respect to the
λi . This expansion is called cumulant expansion, the coefficients of the expansion are called
cumulants.
Cumulants (semiinvariants) have been known long before in mathematical statistics and
probability theory. Kubo [1] pioneered in demonstrating their usefulness in diverse branches
of physics. Considered as abstract objects cumulants form a special kind of noncommuting
algebra. In his original paper Kubo applied relations found in this algebra to a diverse
set of physical problems. In particular, he gave an elegant derivation of the Ursell-Mayer
expansion for classical and quantum gases that is usually obtained by much longer diagram
considerations, he rederived Goldstone perturbation theory for the ground-state of a Hamil-
ton operator, and investigated the motion of an oscillator with a random frequency. Kubo
stressed the idea that ”the generalized cumulant expansion provides us with a point of view
from which many existent methods in quantum mechanics and statistical physics can be
unified” [1].
when the ground-state of the unperturbed Hamiltonian is not a single particle one and
the usual diagrammatic theory based on Wick theorem either breaks down or becomes
very complicated. Also, an important feature is that any cumulant expression vanishes for
statistically independent processes. Thus the cumulant technique takes into account only
2
studying how the properties of objects change under transformations [7]. First, we find a
formula for an infinitesimal transformation of a cumulant, and then integrate it in order to
obtain a finite transformation.
This includes orthogonality relations for Ω, for the ground-state energy, and for expectation
values of operators and correlation functions. Then we give some new applications of the
cumulant technique.
First we show how improved short-time propagators for quantum Monte Carlo meth-
ods can be derived by using cumulant expansions and derive high-order approximations
for coordinate-momentum and coordinate-coordinate propagators. Then we consider the
foundation of the method of increments and extend this method to excited states of a
Hamilton operator. The method of increments has been proven a powerful and intuitive
way of taking into account electron correlations in solids [8], [9], [10], [11], [12], [13]. The
Bethe-Goldstone equations provide a justification for the method of increments, they are not
able to predict what part of the problem is neglected by performing an n-th order increment
calculation. In other words, we do not have an exact n-th order equation, such that the n-th
order increment equation would be an approximation to this equation. Also we should make
3
a remark that the n-th order Bethe-Goldstone equation is restricted to a vacuum taken as
the unperturbed state. Therefore, it is not applicable when we need to take into account
additional correlations starting from an already correlated state.
On the other hand Faddeev‘s equations for the 3-body problem are well known [15], [16]
and have been successfully applied in nuclear physics [17] and in solid-state theory [18],
[19], [20], [21]. Faddeev‘s equations are exact, this feature makes their application mathe-
matically controllable. We show that we are able to generate a hierarchy of Faddeev-like
cumulant equations. Taking the equation of the n-th order in the leading approximation
we exactly recover the n-th order increment expansion. We can see explicitly from the n-th
order cumulant Faddeev‘s equation what part is neglected by the increment approximation.
Therefore, we generally predict the validity and the precision of the n-th order increment
∂ ∂
hA1 ...AN ic = ... ln hx | eλ1 A1 ...eλN AN | yi. (2)
∂λ1 ∂λN λ1 =...λN =0
Cumulants posses a number of properties (see [2]). We briefly repeat them below. To
simplify notations we consider the case of a two-operator cumulant.
4
1. Linearity:
3. Operators A and B are called statistically independent with respect to x, y when the
hx | AB | yi = hx | A | yihx | B | yi, hx | yi = 1.
Then
hx | AB | yic ≡ 0.
4. Applying Definition 1 we can obtain explicit formulas for cumulants. For the case
hx | yi = 1 they read as
hx | A | yic = hx | A | yi
5. We need to distinguish the cumulant of two operators A and B and the cumulant of
an operator AB considered as an entity, when cumulants are evaluated. The latter one we
denote as h(AB)• i.
Also we need to distinguish between the number 1 and the unit operator 1. For instance
h1 · Aic = hAic and h1Aic ≡ 0.
We define formal powers as hAn ic = hA . . A}ic . Then the formal algebra of power series
| .{z
n
must be considered.
Let us formulate an important statement.
5
Statement 1 Let H be a Hamilton operator. Let ψ be an eigenvector of this operator with
an eigenvalue E. Then for any vector ψ0 , hψ0 | ψi =
6 0 and for any cumulant series A, not
including c-numbers the following relations hold
Proof:
The proof of (3) comes from the definition of a cumulant of one operator. Equation
(4) follows from the fact, that H acting on | ψi leads to a multiplication by a number and
therefore the logarithm in (2) factorizes.
Statement 2 Let
hψ0 | AH | ψic = 0
Then ψ is an eigenvector of H.
Proof:
6
B. Transformations of averaging vectors
Here we analyse the basic geometric properties of cumulants. We show that a transfor-
Statement 3 Let
Then
The proof follows from the definition of a cumulant. We present it for the case that X
is a one-operator cumulant.
∂
hψA | X | ψB ic = ln hψA | eλX | eǫS ψB′ i =
∂λ λ=0
∂ ∂ ∂
= ln hψA | eλX | ψB′ i + ǫ ln hψA | eλX eǫS | ψB′ i =
∂λ λ=0 ∂ǫ ǫ=0 ∂λ λ=0
c
then hψA | X | ψB ic = hψA′ | (1 + δS † )X | ψB i .
7
We conclude that an infinitesimal transformation of the averaging vector result in a
multiplication of the cumulant operator by 1 + δS. In the following we call this operator
infinitesimal cumulant wave operator.
The next step is going over to a finite transformation by integrating an infinitesimal one.
We consider the following construction:
a continuous path in Hilbert space between vectors ψB and ψC . Let us choose N + 1 points
ψ0 , ψ1 , ... ψN on this path such that ψ0 = ψC and ψN = ψB and N operators δSi such that
ψi = eδSi ψi−1 .
By assuming that the operators δSi are small and using (5) we obtain
N
Y
c
hψA | X | ψB i ≈ hψA | X (1 + δSi ) | ψC ic .
i=1
hψA | X | ψB ic = hψA | XΩ | ψC ic
N
Y
Ω = lim (1 + δSi ). (6)
N →∞
i=1
We note that there is a considerable freedom in the Construction 1 for choosing the
path between vectors ψB , ψC and therefore the operators δSi . Thus the operator Ω is not
defined uniquely. Two different representations for Ω differ by a cumulant operator with a
zero contribution to any cumulant expression (such operators can be obtained by integrating
over a closed path in Hilbert space and subtracting the number 1).
We also note that Ω always contains the number 1 as the first term of the formal series.
The formula for the finite transformation of the left averaging vector is obtained in
analogy to (6). In that case we have
8
c
hψA | X | ψB ic = hψC | Ω† X | ψB i . (7)
To complete the definition of the algebra of cumulant series we need to define the cu-
mulant of the number 1. By formally putting the number of lambdas in the Definition 1 to
zero we have
The usefulness of this definition comes from the point that the new object we just intro-
duced satisfies the general expression for an infinitesimal transformation of a cumulant. It
can be directly checked that
Therefore when doing transformations we need not to discriminate between the cumulant
of the number 1 and ”ordinary” cumulants.
D. Superoperators
∂
hx | D | yic = ln hx | eλD 1 | yi. (9)
∂λ λ=0
9
For some operator H we introduce the superoperator L: L = [H, ...].
Cumulants containing L have the following property ( [22])
c
hGLAic = hG(LA)• i . (10)
Proof:
The existence of G does not play any role in the following proof and so we omit G to
∂2
hLAic = ln hx | eλ1 L eλ2 A | yi
∂λ1 λ2 λ1 =λ2 =0
∂ ∂
= ln hx | (1 + λ1 L)(1 + λ2 A) | yi
∂λ1 ∂λ2 λ1 =λ2 =0
∂ ∂ ∂
= ln hx | (1 + λ1 λ2 LA) | yi = ln hx | (1 + λLA) | yi (11)
∂λ1 ∂λ2 λ1 =λ2 =0 ∂λ λ=0
∂ ∂
h(LA)· ic = ln hx | eλLA | yi = ln hx | (1 + λLA) | yi. (12)
∂λ λ=0 ∂λ λ=0
E. Finite Temperatures
The above results can be extended to finite temperatures. Here the averaging implies
taking a trace with some other operator (averaging operator). Therefore, we define the
cumulant of a product of N operators as:
10
∂ ∂
hA1 ...AN | Gic = ... ln Tr eλ1 A1 ...eλN AN G . (13)
∂λ1 ∂λN λ1 =...λN =0
Then
c
F = −β ln T r(e−βH ) = −βh1 | e−βH i
c c
= −βh1 | e−β(L+H1 ) e−βH0 i = −βhe−β(L+H1 ) | e−βH0 i . (14)
In the first step we have used the definition of a cumulant of the number 1 and then trans-
formed the averaging operator from e−β(L+H1 ) e−βH0 to e−βH0 .
c
Then δF = F − F0 = −βhe−β(L+H1 ) − 1 | e−βH0 i .
This expression is analogous to the closed loop expansion of the free energy (thermodynamic
potential) in field theory.
Here we use the results obtained in the previous section to describe properties of a
quantum system. Most of these results were obtained previously by using algebraic consid-
erations [2–6]. Here we rederive them from a geometrical point of view.
hΨ0 | HΩ | Ψ0 ic = E (15)
11
hΨ0 | AHΩ | Ψ0 ic = 0. (16)
Proof:
By taking equations (3), (4) and transforming the right averaging vector from Ψ to Ψ0
we immediately obtain (15), (16).
Using Statement 2 and transforming Ψ into Ψ0 we find that the opposite statement is
also true, i.e., if Ω transforms Ψ into Ψ0 and (16) holds for any A then Ψ is an eigenvector
of H.
Next we consider expectation values of different physical quantities.
c
hΨ | G | Ψi = hΨ0 | Ω† GΩ | Ψ0 i . (17)
Proof:
First we note that hΨ | G | Ψi = hΨ | G | Ψic . Transforming the right and left averaging
Then one of the possible forms for the wave operator transforming ΨB into ΨC is
Ω = eS . (18)
Proof:
12
We take δSi in (6) as:
δSi = S/N.
We have used a representation of the formal exponential series known from algebra. It is
possible to prove a more general statement:
Ω = eg(S) . (19)
δSi = g ′ (iS/N)S/N
It is known from the theory of series expansions (see for instance [23]) that this expression
tends to
Ω = exp [g(S)] .
The question arises under which conditions an analytic function f (S) can be represented
by eg(S) . The theorem for Weierstrass products [23] states that a necessary and sufficient
condition is that function f (s) does not have zeroes in the complex plane.
13
For instance, the function 1 + S has a zero in the complex plane and therefore cannot
be represented by eg(S) . Therefore, we cannot state that if ΨB = 1 + SΨC then Ω = 1 + S.
Let us now rederive an operator Ω transforming the unperturbed ground-state | ψ0 i
Here E is the ground-state energy of H. We assume that | ψi and | ψ0 i have nonzero overlap.
Because cumulants do not depend on the norm of the averaging vectors the prefactor in front
of | ψi is not important and we can write
Taking the Laplace transform with respect to the variable z and multiplying by z to get
the constant part of the expression we obtain
z 1 1
Ω = lim | ) = lim | 1 + H) = lim | 1 + V ). (23)
z→0 z−H z→0 z−H z→0 z−H
This expression coincides with the one [2]. In the last line we took into account the fact
that | ψ0 i is an eigenvector of H0 .
C. Wavefunction
Here we consider the question how to obtain the wavefunction from the Ω operator.
Statement 9 Let Ψ and Ψ0 be vectors of Hilbert space, and Ω be a cumulant wave operator
transforming Ψ to Ψ0 . Then there exists an expression for Ψ through Ψ0 and Ω that is be
derived below.
Proof:
14
We use the fact that
| Ψi
∇Ψ0 hΨ0 | 1 | Ψic = ∇Ψ0 lnhΨ0 | Ψi = =| Ψnorm i.
hΨ0 | Ψi
Here 1 is number 1 and | Ψnorm i means that | Ψi is normalized by the condition hΨ0 | Ψi = 1.
By transforming the left side from Ψ to Ψ0 we have:
Here lef t Ψ0 means that we should take the gradient only with respect to the left Ψ0 .
D. Green Functions
Rij (τ ) = hψ | ci eτ H c+
j e
−τ H
| ψi = hψ | ci eτ H c+
j e
−τ H
| ψi − hψ | ci | ψihψ | c+
j | ψi =
c c
hψ | ci eτ H c+
j e
−τ H
| ψi = hψ | ci eτ L c+
j | ψi . (26)
As a result we obtain the following expression for the Laplace transform of Rij
1
Rij (z) = (Ω | ( ci ) • c+
j | Ω). (27)
z−L
This result coincides with the one obtained in [6,2] by resummation of a power series.
In a similar way expressions for higher-order Green functions can be obtained.
15
E. Short-time propagators
The Diffusion Quantum Monte Carlo technique uses Feynman‘s path integration for the
the method used by Faddeev [27] to obtain the first-order propagator. First we calculate
the phase-space propagator, and then by taking the Fourier transform come back to the real
space.
The phase-space propagator is defined as [27]
h ci
F (p, q, t) = hp | e−iHt/h̄ | qi = exp hp | e−iHt/h̄ | qi . (28)
Consider H = p2 /2+V (q). The above expression allows us to evaluate an improved short-
time propagator up to any power of t. Let us calculate it to order t2 . Direct calculation
gives:
2 1
F (p, q, t) = (2π) −1/2
exp −ipqh̄ −1
− i(p /2 + V (q))th̄ −1
− (2ph̄V ′ (q) − V ′′ (q))t2 . (29)
4
By inserting (29) and evaluating the integral over p we obtain the short-time real space
propagator:
16
" 2 #
1 [q ′′ − q ′ + (1/2)t2 V ′ (q ′ )]
′ ′′
F (q , q , t) = q exp i − itV (q ′ )h̄−1 − 1/4t2 V ′′ (q ′ ) . (31)
(2πih̄T ) 2th̄
Let us discuss the physical meaning of the expression (31). We have two additional
terms in the propagator as compared to the usual leading order expression derived in [28].
The term (1/2)t2 V ′ (q ′ ) corresponds to the fact that the velocity of the particle following
a Feynman path is not constant, changing due to the force V ′ (q ′ ) and therefore leading
to a correction of the kinetic energy term. The meaning of the term 1/4t2 V ′′ (q ′ ) will be
considered later.
Let us use the above results to obtain a high-temperature expansion for the partition
function of a quantum system.
The partition function is defined as [28]:
Z
Z = T r exp(−βH) = hq | exp(−βH) | qidq. (32)
For small β we can use the expansion (31) setting t = −iβ. Then we have to order β 2 :
Z
1 h i
Z=q exp −βV (q) − (1/4)h̄2 β 2 V ′′ (q) dq. (33)
(2πh̄2 β)
To the first order in β we obtain a classical expression for a partition function, this
means that for high temperatures the average energy of the system is large and we can
neglect quantisation effects.
Let us discuss the meaning of the correction 1/4h̄2 β 2 V ′′ (q). Suppose the potential V (q)
has a very high but very narrow peak. Then the region of the peak will be prohibited in a
classical theory because V (q) is large. On the other hand, for the quantum case there will
be some probability for the system to be in the region of the peak due to tunnelling. For a
high and narrow peak V ′′ (q) will be large and negative. Now looking at the expression (33)
we see that 1/4h̄2 β 2 V ′′ (q) has then a sign opposite to βV (q) so the potential is effectively
smeared out. Therefore, we see that 1/4h̄2 β 2 V ′′ (q) term describes an effective smearing of
the potential due to the quantum tunnelling.
17
The expansion for a partition function obtained above is a high temperature expansion
(in powers of β). We note that this expansion does not coincide with the quasiclassical
expansion (in powers of h̄). Although both types of expansions give the classical formula
as the leading term, the higher-order terms are different. The expansion in powers of h̄ was
first obtained by Wigner [29]. To the second order in h̄ it can be written as:
Z
1 h i
Z=q exp −βV (q) − (1/12)h̄2 β 2 V ′′ (q) + (1/24)h̄2 β 3 V ′ (q)2 dq. (34)
(2πh̄2 β)
The formulas for improved short-time propagators can be useful to improve the speed
and the precision of numerical Quantum Monte Carlo calculations [25]. They could also
improve analytical calculations that make use of short-time propagators.
We remark that the above technique applies only when the short-time propagator has a
well-defined Fourier transform with respect to the coordinate. This is usually satisfied for
potentials approaching zero or a constant at infinite distance. This condition includes the
majority of potentials used for Monte Carlo calculations of quantum gases and liquids. The
case when the propagator cannot be Fourier transformed with respect to the coordinate is
subject to future considerations.
It happens frequently that we deal with a Hamiltonian consisting of different parts and
that these parts considered separately can be solved analytically or numerically.
The first example of such a situation is the3-body problem. Three particles interact with
each other by two-particle potentials. Considering each pair of particles separately we face
a 2-body problem which can be solved analytically. Then the question is how we can use
the information given by these solutions to construct an approximation to the solution of
18
The second example are quantum chemistry calculations where we need to find an ap-
proximate solution for a large molecule or a cluster having available numerical calculations
for its subparts, i.e., chemical bonds.
For the 3-body problem L. D. Faddeev invented coupled integral equations that connect
the unknown 3-body scattering matrix to 2-body scattering matrices that can be found
analytically [15], [16].
In quantum chemistry the method of increments is well known, that expresses the
H = H0 + H1 + H2 + H3 . (37)
T1 = S1 (1 + T2 + T3 )
T2 = S2 (1 + T1 + T3 )
(38)
T3 = S3 (1 + T1 + T2 )
S = T1 + T2 + T3 .
19
Proof:
and define
1 1 1
A1 = H1 , A2 = H2 , A3 = H3 . (40)
z − H0 z − H0 z − H0
Then
∞
(A1 + A2 + A3 )n =
P P P P
S= A1 . . . + A2 . . . + A3 . . . = T1 + T2 + T3 . (41)
n=1
We divided the terms in the power series into three groups, the terms in the first group
begin with A1 , the ones in the second group with A2 , in the third group with A3 . We call
these three sums T1 , T2 , T3 .
We rewrite T1 in the form
The first bracket defines S1 . The expressions for T2 , T3 are similar. This results in
Faddeev‘s equations (38).
In the original work of Faddeev S1 , S2 , S3 correspond to two-body scattering matri-
ces coming from three pairs of particles in the 3-body problem, and S corresponds to the
unknown 3-body scattering matrix.
We easily generalize our equations to the case when the perturbation has N parts, in
that case we have N equations of the type
X
Ti = Si 1 + Tj . (43)
j6=i
Assuming Si to be small, the equations can be solved by iterations. Then S is given by (41).
In the leading approximation we have
20
X
Ti = Si , S = Si . (44)
X
δE = E − E0 = δEi , (45)
Here E0 is the ground-state energy of H0 . The equation (45) gives the simplest example
exactly the N-th order increment method. Therefore Faddeev‘s equations and increment
methods are closely related.
As an example we derive the second-order Faddeev‘s equations. We again take a Hamil-
tonian consisting of H0 part and N perturbations
H = H0 + H1 + . . . + HN . (47)
But now we additionally assume that we can solve analytically or numerically all the Hamil-
tonians of the form H0 + Hi + Hj , i.e., that we know the scattering operators Sij for these
Hamiltonians.
Then we write again the expression for S,
∞
(A1 + . . . + AN )n =
P
S=
n=1
AnN = (48)
P P P P∞ P∞
A1 . . . A2 . . . + A2 . . . A3 . . . + A1 . . . A3 . . . + . . . + n=1 An1 + . . . n=1
P P∞
= 1≤i,,j≤N. Tij + n=1 Si .
This resummation is more sophisticated and needs explanations. First, we open brackets in
the sum in order to obtain a power series with respect to Ai ’s. For each term in this power
21
series we look for the first two letters in it, and regroup terms according to this property.
For example, T1,2 will be the sum of all terms in the power series having A1 and A2 as the
first two letters, so that ,e.g., the term A31 A22 A4 is included in T1,2 .
We are still left with terms containing only one letter like A42 . These terms are not
contained in any of Tij ’s so we sum them separately, this explains the result of (48).
Now we consider T1,2 :
X
P
T1,2 = (A1 A2 + A1 A22 + A21 A2 + A1 A2 A1 + ...) 1 + (i,j)6=(1,2)
Tij + i6=1,2 Si =
X P
(49)
= (S1,2 − S1 − S2 ) 1 + T +
(i,j)6=(1,2) ij i6=1,2 Si .
This result is obtained by considerations similar to the previous ones. The first bracket is
a sum of all possible terms consisting of A1 ‘s and A2 ‘s only. The second bracket is a sum
of all possible terms that do not have A1 and A2 as the first two letters. S1,2 denotes the
scattering operator for the Hamiltonian H0 + H1 + H2 .
The similar equations are obtained for all other Tij . Together with (48) these equations
give the second-order cumulant Faddeev‘s equations.
Taking again the leading approximation for the equations we have:
X X
S= Si + Kij (50)
i i,j
where
X X
δE = δEi + δEij . (52)
i ij
22
By continuing this procedure and taking the leading approximation to the next contribution
we end up with the third-order cumulant expression for S
X X X
S= Si + Ki,j + Lijk (54)
i i,j i,j,k
Again this expression agrees up to third-order with the energy expression of the increment
a a characteristic length we called the correlation length λ, then the omitted terms in the
Faddeev‘s equation will sum incoherently and can be neglected. Therefore we can terminate
an expansion of the correlation energy in terms of increments when the spatial extent of a
set of bonds is of the order of the correlation length. This is in accordance with the well
known idea, that the minimum size to which we can reduce a system not changing its macro-
scopic properties is dictated by the correlation length [30]. In order to formulate it more
rigorously we may state that the macroscopic properties of an infinite solid can be described
by an incremental expansion with increments taken up to the correlation length. Therefore
23
incremental methods will work efficiently for systems having a moderate correlation length,
like semiconductors [11].
In the following we formulate a method of increments for excited states. We begin with a
nondegenerate unperturbed Hamiltonian, so perturbations do not split excitation energies,
instead they only shift them.
We start from a formula derived in [4] for the excited states of a Hamiltonian:
Ei = hψi | HΩ | ψi ic . (56)
Here | ψi i is an excited state of H0 having an overlap of more than 1/2 with the true
Here δEi,1 , δEi,2 , δEi,3 are increments corresponding to the excited state i,
24
Let us consider as an example of an increment calculation the valence band calculations
done in [31].
In a first step self-consistent field CSCF calculations are done and the energy bands in
SCF approximation are obtained. Then the SCF states are transformed into Wannier states
(localized orbitals) and the residual part of the interaction Hres = H − HSCF is expressed
in terms of matrix elements between the localized bonds.
The increment method up to forth order was used to obtain the valence band. In this case
SCF hamiltonian plays the role of H0 and the matrix elements of the residual Hamiltonian
between pairs of localized bonds play role of perturbations. In the calculations increments
corresponding to NN (nearest neighbours) and NNN (next-nearest neighbours) were retained
and the effect of more distant correlations was approximated by introducing a polarizable
continuum.
The results have been shown to describe with a good accuracy the reduction of the
valence bonds of diamond due to correlations [31].
V. CONCLUSIONS
In this article we considered the question, how geometrical methods can be used to
formulate the theory of cumulant expansions. Cumulant geometry simplifies the proofs of
many expressions obtained before by cumulant algebra and makes them transparent.
Then we studied applications of the cumulant technique to short-time propagators and
to the method of increments. In both cases the use of cumulants is a natural choice. A
hierarchy of cumulant Faddeev-like equations was generated and a method of increments for
excited states was established.
Future directions of research will include diverse applications of the cumulant technique
to solid state theory. A particularly interesting question there is how properties of the
infinite solid can be related to properties of a finite cluster. This question is of outmost
importance for all numerical calculations.
25
VI. ACKNOLEDGMENTS
We would like to thank Prof. K. Becker, K. Fischer and Tran Mien Tien for stimulating
discussions and Tran Mien Tien for carefully reading and correcting the manuscript.
26
REFERENCES
[2] P. Fulde. Electron Correlations in Molecules and Solids. Springer Verlag, third edition,
1995.
[3] K.W. Becker and P. Fulde. Z.Phys.B - Condensed Matter, 72:423, 1988.
[4] T. Schork and P. Fulde. Int. Journal of Quantum Chemistry, 51:113, 1994.
[10] K. Doll, M. Dolg, P. Fulde, and H. Stoll. Phys. Rev. B, 52:4842, 1995.
[16] L. D. Faddeev and S. P. Merkuriev. Quantum Scattering Theory for Several Particle
Systems. Kluwer Academic Publishers, 1993.
27
[19] J. Igarashi. J. Phys. Soc. Jpn., 54:260, 1985.
[20] J. Igarashi. In: Core-Level Spectroscopy, edited by J. Kanamori and A. Kotani. Springer
Verlag, 1983.
[21] J. Igarashi, P. Unger, K. Hirai, and P. Fulde. Phys. Rev. B, 49:16181, 1994.
[25] M. Suzuki. In: Quantum Monte Carlo Methods, Proceedings of the 9-th Taniguchi In-
[26] L. S. Schulman. Techniques and applications of Path Integration. John Wiley and Sons,
1981.
[27] L. D. Faddeev. In: Methods in Field Theory, Les Houches Session XXVIII. North
Holland and World Scientific, 1975.
[31] J. Graefenstein, H. Stoll, and P. Fulde. Chem. Phys. Lett., 215:611, 1993.
28