You are on page 1of 93

Contents

1 Beyond the Independent Electron Approximation 3


1.1 Hartree Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Hartree-Fock Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Screening (General) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Coulomb Potential in Momentum Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Second Quantization 7
2.1 Slater Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Creation and Annihilation Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Occupation-Number Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Second-Quantized Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.5 Coherent States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Linear Response 13
3.1 Experimental Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Zero-Temperature Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 Finite-Temperature Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.4 Example: Noninteracting Electron Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4.1 Density-Density Correlation Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4.2 Current Density Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4.3 Current-Current Correlation Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4.4 Four-Vector Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4.5 Spin-Density Correlation Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4 Real-Time Correlation Functions 28


4.1 Real-Time Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Spectral Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2.1 Spectral Density Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2.3 Example: Noninteracting System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2.4 Real-Time Correlation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2.5 Generalization to a Complex Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

5 Imaginary-Time Correlation Functions 36


5.1 Periodicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.2 Imaginary-Time Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.3 Spectral Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.3.1 Imaginary-Time Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.3.2 Imaginary-Time Correlation Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.3.3 Analytic Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

6 Perturbation Theory 40
6.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.2 Wick’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.3 Diagrammatic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

7 Green’s Function Methods 44


7.1 Quasiparticles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
7.2 Introduction to Propagators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
7.3 Propagator Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7.4 Example: Scattering by an External Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.5 Example: Electron in an Impure Metal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.6 Formal Definition of Propagator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.7 Fermi Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.7.1 Particle-Hole Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.7.2 Example: Non-Interacting System of Electrons in an External Potential . . . . . . . . . . . . . 60

1
7.7.3 Interacting Fermi Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
7.7.4 Hartree and Hartree-Fock Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
7.7.5 Time-Ordered Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.7.6 Time-Ordering and Feynman Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.7.7 Dyson’s Equation and Self-Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7.7.8 Example: Imperfect Fermi Gas (Ladder Approximation) . . . . . . . . . . . . . . . . . . . . . . 77
7.8 High-density Electron Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.8.1 Random Phase Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.8.2 Evaluation of Π0 (q, ω) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.8.3 Static Screening Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
7.8.4 Self-Energy in RPA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.8.5 General Effective Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

2
1 Beyond the Independent Electron Approximation
(Following Ashcroft and Mermin, Chapter 17 - note Gaussian units)

1.1 Hartree Approximation


The independent electron approximation includes electron-ion interactions but ignores electron-electron interactions.
In fact, a more accurate calculation of the electronic properties of a metal should start with the Schrödinger equation
for the N -particle wavefunction of all N electrons in the metal, Ψ(r1 σ1 , r2 σ2 , . . . , rN σN ), where σi indicates the spin
coordinate (↑ or ↓) of the ith electron:
N
!
X ℏ2 2 X 1 1 X 1
− ∇i Ψ − Ze2 Ψ + e2 Ψ = EΨ
i=1
2m |r i − R| 2 |ri − ri |
R i̸=j

where the second term represents the interaction between the electrons and the ions fixed at points R, and the third
term represents electron-electron interactions.
Solving such an equation is too complex, so we try starting instead with the Schrödinger equation for a single
electron:
ℏ2 2
− ∇ ψi (r) + U (r)ψi (r) = εi ψi (r)
2m
where ψi (r) is the wavefunction for an electron occupying the energy level εi , and the potential energy U (r) is a sum
of electron-ion and electron-electron parts, i.e. U (r) = Ue-i (r) + Ue-e (r). The electron-ion potential energy is of the
same form as in the N -particle equation:
X 1
Ue-i (r) = −Ze2
|r − R|
R

For the electron-electron part, however, we make the simple assumption that one electron feels the effect of the
other electrons as a smooth distribution of negative charge of density
X
ρ(r) = −e |ψi (r)|2
i

where the summation is over all the other electrons. The electron-electron potential energy then takes the form
ρ(r′ )
Z
Ue-e (r) = −e dr′
|r − r′ |

Substituting for U (r) in the above Schrödinger equation we obtain


 
ℏ2 2 X 1 XZ |ψ j (r)| 2
− ∇ ψi (r) − Ze2 ψi (r) + e2 dr′  ψi (r) = εi ψi (r)
2m |r − R| j
|r − r′ |
R

The set of N such single-electron equations is known as the Hartree equations and they are solved self-consistently
by iteration, i.e. they are first solved using a trial form of Ue-e , the resulting wavefunctions are used to determine a
new form of Ue-e , and the process is then repeated until further iterations do not substantially change this. For this
reason the Hartree approximation is sometimes known as the self-consistent field approximation.

1.2 Hartree-Fock Approximation


The Hartree equations give the best approximation to the full N -electron wavefunction represented as a product of
single-electron wavefunctions:

Ψ(r1 σ1 , r2 σ2 , . . . , rN σN ) = ψ1 (r1 σ1 )ψ2 (r2 σ2 ) · · · ψN (rN σN )

The problem with this wavefunction, however, is that it is incompatible with the Pauli exclusion principle, which
requires that the sign of the wavefunction should change when any two of its arguments are interchanged, e.g.

Ψ(r1 σ1 , . . . , ri σi , . . . , rj σj , . . . , rN σN ) = −Ψ(r1 σ1 , . . . , rj σj , . . . , ri σi , . . . , rN σN )

3
The simplest way to incorporate this requirement is to replace the product wavefunction with a so-called Slater
determinant of single-electron wavefunctions:

ψ1 (r1 σ1 )
ψ1 (r2 σ2 ) ··· ψ1 (rN σN )
1 ψ2 (r1 σ1 ) ψ2 (r2 σ2 ) ··· ψ2 (rN σN )
Ψ(r1 σ1 , r2 σ2 , . . . , rN σN ) = √ .. .. .. ..
N ! . . . .

ψN (r1 σ1 ) ψN (r2 σ2 ) · · · ψN (rN σN )
The corresponding single-particle Schrödinger equation takes the form
ℏ2 2 XZ 1
− ∇ ψi (r) + Ue-i (r)ψi (r) + Ue-e (r)ψi (r) − e2 dr′ ψ ∗ (r′ )ψi (r′ )ψj (r) = εi ψi (r)
′| j
2m j
|r − r

where Ue-i (r) and Ue-e (r) are the Hartree potential energy terms. We also have an additional potential energy term
known as the exchange term (or Fock term), which results from the aforementioned requirement of antisymmetry
under particle exchange. The inclusion of this term is known as the Hartree-Fock approximation and the set of N
single-particle equations above are known as the Hartree-Fock equations.

1.3 Screening (General)


Consider a positively charged particle placed at a given position in a free electron gas. It will then attract electrons,
creating an excess of negative charge around it which screens its electric field. To study the effect of this screening
we start by introducing two electrostatic potentials. The first potential is due to the positively charged particle and
satisfies the following Poisson equation:
∇2 ϕext (r) = −4πρext (r)
where ϕext (r) is the potential at point r due to the particle’s charge density ρext (r) also at r. (Note that although the
LHS and RHS are specified at the same point r, a solution to the Poisson equation will give the potential ϕext (r) as an
integral in terms of the charge density at a different point, i.e. ρext (r′ ), so the equation contains all the information
on the potential at different distances from a charge.)
The second potential is the total potential produced by both the positively charged particle and its induced cloud
of screening electrons, and this satisfies another Poisson equation
∇2 ϕ(r) = −4πρ(r)
where ϕ(r) is the total potential and ρ(r) is the corresponding charge density given by
ρ(r) = ρext (r) + ρind (r)
where ρind (r) is the charge density induced in the electron gas by the positively charged particle.
By analogy with dielectrics, where the constitutive relation between fields D = ϵE can be generalized to the
non-local form Z
D = dr′ ϵ(r − r′ )E(r′ )

we can assume that ϕext and ϕ are linearly related as follows:


Z
ϕext (r) = dr′ ϵ(r − r′ )ϕ(r′ )

where we have also assumed that the electron gas is spatially uniform, so that ϵ(r, r′ ) = ϵ(r − r′ ).
We would now like to derive a similar relation between the potentials in Fourier space, so we begin by defining the
following pair of Fourier transforms:
Z Z
1
ϵ(q) = dr e−iq·r ϵ(r), ϵ(r) = dq eiq·r ϵ(q)
(2π)3
as well as a similar pair for both ϕext and ϕ. We now follow the derivation of the convolution theorem (see Arfken
p.811) to reexpress ϕext (r):
Z Z Z
ext ′ ′ ′ 1 ′ ′ ′
ϕ (r) = dr ϵ(r − r )ϕ(r ) = dr ϕ(r ) dq ϵ(q)eiq·(r−r )
(2π)3
Z Z  Z
1 ′ ′ iq·r′ iq·r′ 1 ′
= dq ϵ(q) dr ϕ(r )e e = dq ϵ(q)ϕ(q)eiq·r
(2π)3 (2π)3

4
Comparing this with the Fourier transform of ϕext (q), i.e.
Z
1
ϵ(r) = dq eiq·r ϵ(q)
(2π)3
we obtain the simple relation
ϕext (q) = ϵ(q)ϕ(q)
which can also be expressed in the form
ϕext (q)
ϕ(q) =
ϵ(q)
This asserts that the qth Fourier component of the total potential in the electron gas is equal to the qth Fourier
component of the external potential reduced by a factor of 1/ϵ(q).
We now assume that ρind and ϕ are linearly related by an equation similar to that used above for ϕext and ϕ:
Z
ρ (r) = dr′ χ(r − r′ )ϕ(r′ )
ind

where χ(r−r′ ) is known as a response function. (Note that in standard linear response theory, ρind is usually expressed
in terms of ϕext rather than ϕ.) Following the above procedure, this equation can be written in Fourier space as

ρind (q) = χ(q)ϕ(q)

The idea now is to relate ϵ, the quantity of physical interest, to χ, which can be calculated from a microscopic theory.
We proceed by taking the Fourier transforms of the two Poisson equations stated earlier - this is done by substituting
the expressions for the Fourier transforms of ϕext , ϕ, ρext and ρ obtained previously, and then evaluating the second
differential of the LHS and equating terms inside the integrals, i.e.
Z Z
dq q e ϕ (q) = 4π dq eiq·r ρext (q)
2 iq·r ext

Z Z
dq q 2 eiq·r ϕ(q) = 4π dq eiq·r ρ(q)

which leads to the following Poisson equations in Fourier space:

q 2 ϕext (q) = 4πρext (q)


q 2 ϕ(q) = 4πρ(q)

Substituting ρ(q) = ρext (q) + ρind (q) and ρind (q) = χ(q)ϕ(q) then yields

q2
(ϕ(q) − ϕext (q)) = χ(q)ϕ(q)

Rearranging for ϕ(q) and comparing with ϕ(q) = ϕext (q)/ϵ(q), we obtain the final result

4π 4π ρind (q)
ϵ(q) = 1 − 2
χ(q) = 1 − 2
q q ϕ(q)
Except for the assumption that the external charge is weak enough to produce only a linear response in the electron
gas, the preceding analysis is exact. To calculate χ, however it is necessary to make more serious approximations.

1.4 Coulomb Potential in Momentum Space


The Coulomb potential (energy) experienced by an electron in the field generated by another electron is given by

e2
V (r) =
r
To determine the corresponding form of the Coulomb potential in momentum space we take the Fourier transform:
Z
V (q) = d3 r V (r)e−iq·r

5
(Note that we are assuming infinite volume; in the case of a finite volume, Ω, there will be a factor of 1/Ω in front of
the integral, which will carry through to the final expression for V (q).)
Explicitly this yields Z 2π Z π Z ∞
e−iqr cos θ
V (q) = e2 dϕ dθ sin θ dr r2
0 0 0 r
The integral over ϕ gives a factor of 2π, while the integral over θ is carried out as follows:
Z π Z 1
−iqr cos θ 1  2 sin qr
dθ sin θe = du e−iqru = eiqr − e−iqr =
0 −1 iqr qr

This is oscillatory at infinity, which presents a problem. To overcome it, we replace the bare Coulomb potential with
the Yukawa potential :
e2
V (r) = e−λr , λ > 0
r
where the factor e−λr → 0 as r → 0, which ensures that the integrand also goes to zero in this limit. We then take
λ → 0 at the end of the calculation.
Repeating the above Fourier transform for the Yukawa potential then yields

2πe2 ∞  2πe2 ∞ h (−λ+iq)r


Z Z i
−λr −iqr
V (q) = dr e iqr
e −e = dr e − e−(λ+iq)r
iq 0 iq 0
2 −(λ+iq)r ∞
 (−λ+iq)r
4πe2

2πe e e
= + = 2
iq −λ + iq λ + iq 0 q + λ2

and hence, taking the limit λ → 0, we obtain the Coulomb potential in momentum space:

4πe2
V (q) =
q2
Taking the inverse transform of the Yukawa potential in momentum space, we get
Z
1
V (r) = d3 r V (r)eiq·r
(2π)3

which, after evaluating the angular integrals, yields



2e2
Z
q sin qr
V (r) = dq
πr 0 q 2 + λ2

Making use of the standard integral


Z ∞
x sin ax π
dx = e−aβ , a, β > 0
0 β 2 + x2 2

with a = r and β = λ then allows us to recover the real-space form

2e2 π −λr e2
V (r) = e = e−λr
πr 2 r

6
2 Second Quantization
2.1 Slater Determinants
We take as our single-particle wavefunctions the set of basis states labelled by the wavenumber k, which we write as

ϕk1↑ (r) = eik1 ·r (↑), ϕk1↓ (r) = eik1 ·r (↓), ϕk2↑ (r) = eik2 ·r (↑), . . .

The Slater determinant can then be expressed as


ik ·r
eik1 ·r2 (↑)2 eik1 ·rN (↑)N

ϕk1↑ (r1 ) ϕk1↑ (r2 ) ··· ϕk1↑ (rN ) e 1 1 (↑)1
ik ·r ···
eik1 ·r2 (↓)2 eik1 ·rN (↓)N

ϕk1↓ (r1 ) ϕk1↓ (r2 ) ··· ϕk1↓ (rN ) e 1 1 (↓)1
ik ·r ···
eik2 ·r2 (↑)2 eik2 ·rN (↑)N

ϕk2↑ (r1 ) ϕk2↑ (r2 ) ··· ϕk2↑ (rN ) e 2 1 (↑)1 ···
1 ϕk ↓ (r1 ) ϕk2↓ (r2 ) ··· 1
ϕk2↓ (rN ) = √ eik2 ·r1 (↓)1

eik2 ·r2 (↓)2 ··· eik2 ·rN (↓)N
Ψk1 k2 ···kN = √ 2

N ! .. .. .. .. N ! .. .. .. ..
. . . .
. . . .

ϕk ↑ (r1 ) ϕkN↑ (r2 ) · · · ϕkN↑ (rN ) eikN ·r1 (↑)1 eikN ·r2 (↑)2 ··· e ikN ·rN
(↑)N
N
ϕk ↓ (r1 ) ϕkN↓ (r2 ) · · · ϕkN↓ (rN ) eikN ·r1 (↓)1 eikN ·r2 (↓)2 ··· eikN ·rN (↓)N
N

Note that each column corresponds to a particular particle, while each row corresponds to a particular set of quantum
numbers. The latter means that we can represent the Slater determinant equivalently in abbreviated form as the state
vector
|ϕk1↑ ϕk1↓ ϕk2↑ ϕk2↓ · · · ϕkN↑ ϕkN↓ ⟩
The Pauli exclusion principle is enforced by the fact that if two electrons occupy the same state, i.e. have the same
set of quantum numbers, then two of the rows in the determinant become identical and the latter vanishes according
to the rules of determinants. Antisymmetry of the overall wavefunction is guaranteed by the fact that interchange of
two particle labels corresponds to exchanging two columns of the determinant, which from the rules of determinants
causes it to change sign. To illustrate this, consider the simplest Slater determinant representing two electrons in
states ϕk↑ (r) = eik·r (↑) and ϕk↓ (r) = eik·r (↓):

1 ϕ (r ) ϕk↑ (r2 ) 1 eik·r1 (↑)1 eik·r2 (↑)2
Ψk1 k2 = √ k↑ 1 = √ ik·r ik·r
2! ϕk↓ (r1 ) ϕk↓ (r2 ) 2! e 1 (↓)1 e 2 (↓)2

1 1 
= √ [ϕk↑ (r1 )ϕk↓ (r2 ) − ϕk↑ (r2 )ϕk↓ (r1 )] = √ eik·r1 (↑)1 eik·r2 (↓)2 − eik·r2 (↑)2 eik·r1 (↓)1

2! 2!
If we now interchange the two particle labels (1 ↔ 2) we have

1 eik·r2 (↑)2 eik·r1 (↑)1 1 
= √ eik·r2 (↑)2 eik·r1 (↓)1 − eik·r1 (↑)1 eik·r2 (↓)2 = −Ψk1 k2

(Ψk1 k2 )1↔2 = √ ik·r2 ik·r1
2! e (↓)2 e (↓) 1
2!
so that the overall wavefunction is antisymmetric as required.

2.2 Creation and Annihilation Operators


We introduce the second-quantized formalism by associating with each single-particle (fermion) state |ϕk ⟩ a creation
operator ĉ†k and annihilation operator ĉk (dropping the spin index for simplicity). Considering first the creation
operator, it is defined by
ĉ†k |ϕk1 ϕk2 · · · ϕkN ⟩ = |ϕk ϕk1 ϕk2 · · · ϕkN ⟩
i.e. the operator ĉ†k creates a fermion in the single-particle state |ϕk ⟩, thus adding a new row to the top of the Slater
determinant (as well as an extra column); the latter therefore becomes a (N +1)×(N +1) determinant. The properties
of the Slater determinant already noted above can also be seen in the second-quantized formalism: firstly, the action of
ĉ†k on a state vector yields zero if k coincides with any of the indices k1 , k2 , . . . , kN since we cannot create an fermion
in a state that is already occupied (Pauli exclusion principle); secondly, if we consider creating two extra fermions via

ĉ†k ĉ†k′ |ϕk1 ϕk2 · · · ϕkN ⟩ = ĉ†k |ϕk′ ϕk1 ϕk2 · · · ϕkN ⟩ = |ϕk ψk′ ϕk1 ϕk2 · · · ϕkN ⟩

7
and then interchanging the order in which they are created, we find that to obtain the original state function we
need to interchange the first two entries in the state function - this corresponds to swapping two rows of the Slater
determinant, and hence introduces an overall minus sign, thus confirming the antisymmetry of the state:

ĉ†k′ ĉ†k |ϕk1 ϕk2 · · · ϕkN ⟩ = |ϕk′ ϕk ϕk1 ϕk2 · · · ϕkN ⟩ = − |ϕk ϕk′ ϕk1 ϕk2 · · · ϕkN ⟩

Similarly, we define the annihilation operator by

ĉk |ϕk ϕk1 ϕk2 · · · ϕkN ⟩ = |ϕk1 ϕk2 · · · ϕkN ⟩

i.e. the operator ĉk annihilates a fermion in the single-particle state |ϕk ⟩. The annihilated state must be in the leftmost
position, i.e. it must be the first row in the Slater determinant, and if this is not the case it must be move to the
leftmost position by swapping positions with the intervening states, thus introducing a minus sign each time, i.e.

ĉk |ϕk1 ϕk ϕk2 · · · ϕkN ⟩ = −ĉk |ϕk ϕk1 ϕk2 · · · ϕkN ⟩ = − |ϕk1 ϕk2 · · · ϕkN ⟩

Finally, the state occupied by the fermion to be annihilated must be among the collection of states in the Slater
determinant, otherwise the action of the annihilation operator is defined to yield zero.

2.3 Occupation-Number Representation


Suppose that we have a complete set of single-particle states |ϕ1 ⟩ , |ϕ2 ⟩ , . . . that are ordered in some way, e.g. by
energy, ε1 ⩽ ε2 ⩽ . . . (where we have dropped the k’s and just use numerical subscripts). The Slater determinant lists
the occupied single-particle states, e.g. |ϕ1 ϕ3 ⟩ represents a configuration where one particle occupies |ϕ1 ⟩ and another
particle occupies |ϕ3 ⟩. We can represent this state equivalently by the state vector |1 0 1 0 0 · · ·⟩, which tells us that
the states |ϕ1 ⟩ and |ϕ3 ⟩ are occupied by one particle each, while all the other states are empty. In general, a state
of non-interacting particles, where n1 particles occupy |ϕ1 ⟩, n2 particles occupy |ϕ2 ⟩, and so on, may be represented
as |n1 n2 · · ·⟩ - this is known as the occupation-number representation. For fermions, ni =0 or 1, but for bosons, ni
can vary from 0 to N , the total number of particles. The vacuum state, with no particles at all, is written as |0⟩.
Finally, we note that in the occupation-number representation, the fermion creation and annihilation operators, ĉ†ν
and annihilation operator ĉν , act in the following way:

ĉ†ν |n1 n2 · · · nν · · ·⟩ = (−1)n1 +n2 +...+nν−1 |n1 n2 · · · nν +1 · · ·⟩


ĉν |n1 n2 · · · nν · · ·⟩ = (−1)n1 +n2 +...+nν−1 nν |n1 n2 · · · nν −1 · · ·⟩

The reason for the multiple prefactors of (−1) in the above expressions is that particles must be created/annihilated in
the leftmost position within the state vector (i.e. in the first row of the Slater determinant), so it is usually necessary to
shift the particle concerned right (after creation) or left (before annihilation) to/from its correct position by swapping
locations in the state vector, thus introducing one or more minus signs.
We also note a result that will prove very useful later:

ĉ†ν ĉν |n1 n2 · · · nν · · ·⟩ = nν |n1 n2 · · · nν · · ·⟩

2.4 Second-Quantized Operators


The Hamiltonian, Ĥ, for a system of N identical, interacting particles is generally the sum of a one-body operator, Ĥ0 ,
and a two-body operator, Ĥ ′ , i.e.
N N
X 1 X
Ĥ ≡ Ĥ0 + Ĥ ′ = h(ri ) + v(ri , rj )
i=1
2 i,j=1
(i̸=j)

where h(ri ) is a single-particle operator that depends on the coordinates of particle i (we ignore spin for now), and
v(ri , rj ) is a two-particle operator that depends on the coordinates of particles i and j.
These operators are first-quantized, and in order to second-quantize them we introduce a complete, orthonormal
set of single-particle states |ϕ1 ⟩ , |ϕ2 ⟩ , . . ., which also allows us to introduce the corresponding creation and annihilation
operators, ĉ†ν and ĉν . Considering first the one-body operator, Ĥ0 , this can then be expressed in second-quantized for
as follows: X X
Ĥ0 = ⟨ν|ĥ|ν ′ ⟩ ĉ†ν ĉν ′ = hνν ′ ĉ†ν ĉν ′
νν ′ νν ′

8
As an example, we consider the kinetic energy of a system of N electrons, for which the one-body operator in
first-quantized form is:
N N 
p2i ℏ2 2
X X 
T̂ = = − ∇
i=1
2m i=1
2m i
The second-quantized operator will depend on the basis set of single-particle states, and for this we choose the plane
waves
1
ϕk (r) ≡ ⟨r|k⟩ = √ eik·r
V
The operator can then we written as
ℏ2 2
 
∇ |k⟩ ĉ†k′ ĉk
X

T̂ = ⟨k | −
2m
kk′

and since
ℏ2 2 ℏ2 k 2

∇ |k⟩ = |k⟩ and ⟨k′ |k⟩ = δkk′
2m 2m
we can write the second-quantized form of the kinetic energy operator as
X ℏ2 k 2
ĉ†k ĉk ≡ εk ĉ†k ĉk
X
T̂ =
2m
k k

We now consider the effect of this operator on a many-body state by calculating the expectation value of the
kinetic energy for the state, say, |ϕk1 ϕk3 ⟩, which can be written alternatively in the occupation-number representation
as |1 0 1 · · ·⟩:

εk ⟨ϕk1 ϕk3 |ĉ†k ĉk |ϕk1 ϕk3 ⟩ = εk ⟨1 0 1 · · · |ĉ†k ĉk |1 0 1 · · ·⟩


X X
⟨T ⟩ =
k k

= εk1 ⟨1 0 1 · · · |ĉ†k1 ĉk1 |1 0 1 · · ·⟩ + εk2 ⟨1 0 1 · · · |ĉ†k2 ĉk2 |1 0 1 · · ·⟩ + εk3 ⟨1 0 1 · · · |ĉ†k3 ĉk3 |1 0 1 · · ·⟩ + . . .


= εk1 ⟨1 0 1 · · · |(1)|1 0 1 · · ·⟩ + εk2 ⟨1 0 1 · · · |(0)|1 0 1 · · ·⟩ + εk3 ⟨1 0 1 · · · |(1)|1 0 1 · · ·⟩ + . . .
= εk1 + εk3

Turning now to the two-body operator, Ĥ ′ , this can be expressed in second-quantized form in terms of the complete,
orthonormal set of single-particle states |ϕ1 ⟩ , |ϕ2 ⟩ , . . . as follows:
1 X
Ĥ ′ = ⟨ϕk ϕl |v̂|ϕm ϕn ⟩ ĉ†k ĉ†l ĉn ĉm
2
klmn

where, for example, |ϕm ϕn ⟩ = |ϕm ⟩ ⊗ |ϕn ⟩.

2.5 Coherent States


There are various physical situations in which a macroscopic number of quanta all occupy the same momentum state.
Examples include the laser, where a large number of photons in the same wave-vector are placed in a cavity, and a
superfluid, where a macroscopic number of Bose particles occupy a zero-momentum state. The best way to describe
such systems uses a set of basis states known as (bosonic) coherent states. A coherent state |α⟩ is defined as an
eigenstate of the annihilation operator, i.e.
â |α⟩ = α |α⟩
and is labelled by its particular (complex) eigenvalue α. An arbitrary state |α⟩ will, in general, be the sum of many
energy eigenstates |n⟩, each corresponding to a state containing n quanta (bosons):

X
|α⟩ = Cn |n⟩
n=0

Noting the operation of creation/annihilation operators for bosons:


√ √
↠|n⟩ = n + 1 |n + 1⟩ , â |n⟩ = n |n − 1⟩

9
the LHS of the eigenvalue equation becomes
∞ ∞
X X √ √ √
â |α⟩ = Cn â |n⟩ = Cn n |n − 1⟩ = C1 1 |0⟩ + C1 2 |1⟩ + . . .
n=0 n=1

since â |0⟩ = 0. Similarly, for the RHS of the eigenvalue equation we have

X
α |α⟩ = Cn α |n⟩ = C0 α |0⟩ + C1 α |1⟩ + . . .
n=0

Equating the coefficients of |n⟩ on both sides then gives Cn+1 n + 1 = αCn , i.e. the recursion relation
α
Cn+1 = √ Cn
n+1

and so we can write |α⟩ as


∞  
X α α α
|α⟩ = Cn |n⟩ = C0 |0⟩ + √ C0 |1⟩ + √ C0 √ |2⟩ + . . .
n=0
1 1 2

Using the general relation for creating bosonic states

(↠)n
|n⟩ = √ |0⟩
n!
the above becomes
α2 α2 † 2
 
α α
|α⟩ = C0 |0⟩ + C0 ↠|0⟩ + C0 (↠)2 |0⟩ + . . . = C0 1 + ↠+ (â ) + . . . |0⟩
1! 2! 1! 2!

We can normalize this using ⟨α|α⟩ = 1, i.e.

α∗ (α∗ )2 2 α2 † 2
  
α
|α⟩ = |C0 |2 ⟨0| 1 + â + (â) + . . . 1 + ↠+ (â ) + . . . |0⟩ = 1
1! 2! 1! 2!

Only even numbers of creation-annihilation


√ √ operator pair will yields non-zero terms, and in general ⟨0|(â)n (↠)n |0⟩
will contribute a factor of n! × n! = n!, which cancels with one of the factors of n! in the denominator. The result
is
|α|2 |α|4
 
2 |α|2
⟨α|α⟩ = |C0 |2 1 + + + . . . = |C0 |2 e|α| = 1 =⇒ C0 = e− 2
1! 2!
so that we can write our arbitrary coherent state as

α2 † 2
 
|α|2 α |α|2 †
|α⟩ = e− 2 1 + ↠+ (â ) + . . . |0⟩ = e− 2 eαâ |0⟩
1! 2!

The amplitude to find |α⟩ in the state |n⟩ with n quanta is given by
|α|2
ân |α|2 † e− 2 †
Cn = ⟨n|α⟩ = ⟨0| √ e− 2 eαâ |0⟩ = √ ⟨0|ân eαâ |0⟩
n! n!

The matrix element is only non-zero for the αn (↠)n /n! term of the exponential, so we have
|α|2
e− 2 α n
Cn = √ ⟨0|ân (↠)n |0⟩
n! n!
Using the general relation for creating bosonic states from earlier, together with the conjugate relation
ân
⟨n| = √ ⟨0|
n!

10
√ √
we find that ⟨0|ân (↠)n |0⟩ = n! n! ⟨n|n⟩ = n!, and so we are left with
|α|2
e− 2 α n
Cn = √
n!
This corresponds to a probability density
2 2
2e−|α| (α∗ )n αn e−|α| |α|2n
Pn = |Cn | = =
n! n!
which is a Poisson distribution, with the general form P (X = k) = λk e−λ /k! Since the average number of particles in
a coherent state is given by
⟨n̂⟩ = ⟨α|↠â|α⟩ = ⟨α|α∗ α|α⟩ = |α|2 ⟨α|α⟩ = |α|2
we see that specifying ⟨n̂⟩ determines the shape of the above probability density distribution.
The uncertainty in the number of particles, i.e. the standard deviation, is given by
p
∆n = ⟨n̂2 ⟩ − ⟨n̂⟩2

To determine this we need

⟨n̂2 ⟩ = ⟨α|↠â↠â|α⟩ = ⟨α|α∗ â↠α|α⟩


= |α|2 ⟨α|â↠|α⟩ = |α|2 ⟨α|1 + ↠â|α⟩ = |α|2 + |α|4

where we have used the commutation rule [â, ↠] = â↠− ↠â = 1. This yields
p
∆n = |α|2 + |α|4 − |α|4 = |α|

The fractional uncertainty is then given by


∆n |α| 1 1
= = =p
⟨n̂⟩ |α|2 |α| ⟨n̂⟩

which tends to zero as the average number of particles tends to infinity.


We now consider explicitly the effect of the number operator n̂ on an arbitrary coherent state |α⟩. We begin by
writing |α⟩ in the form
α2 α3
 
|α|2 α
|α⟩ = e− 2 |0⟩ + √ |1⟩ + √ |2⟩ + √ |3⟩ + . . .
1! 2! 3!
so that we have
α2 α3
 
|α|2 α
n̂ |α⟩ = ↠â |α⟩ = e− 2 ↠â |0⟩ + √ ↠â |1⟩ + √ ↠â |2⟩ + √ ↠â |3⟩ + . . .
1! 2! 3!
2 3
 
|α|2 α 2α 3α
= e− 2 0 + √ |1⟩ + √ |2⟩ + √ |3⟩ + . . .
1! 2! 3!
† 2
2
3α3 (↠)3
 
|α|2 α 2α (â )
= e− 2 0 + √ ↠+ √ √ + √ √ + . . . |0⟩
1! 2! 2! 3! 3!
2
α3
 
|α| 2
α α
= e− 2 0 + (1) ↠+ 2 (↠)2 + 3 (↠)3 + . . . |0⟩
1! 2! 3!

Hence, as expected, an arbitrary coherent state |α⟩ is not an eigenstate of the number operator, n̂, as such a state
does not contain a fixed number of particles.
Since the eigenvalue α that labels a particular coherent state is a complex number, we can express it in the form

α = |α|eiθ

where θ is a phase angle. To see how particle number and phase are related, we substitute the above in the definition
of |α⟩, which yields
|α|eiθ † |α|2 e2iθ † 2 |α|3 e3iθ † 3
 
|α|2
|α⟩ = e− 2 1+ â + (â ) + (â ) + . . . |0⟩
1! 2! 3!

11
Differentiating w.r.t. θ then gives

|α|eiθ † |α|2 e2iθ † 2 |α|3 e3iθ † 3


 
∂ |α|2
|α⟩ = e− 2 0+i â + 2i (â ) + 3i (â ) + . . . |0⟩
∂θ 1! 2! 3!
2 3
 
|α|2 α α α
= e− 2 0 + i ↠+ 2i (↠)2 + 3i (↠)3 + . . . |0⟩
1! 2! 3!

so that we have

−i |α⟩ = n̂ |α⟩
∂θ

This implies that just as we have conjugate variables momentum and position linked via, p̂ = −iℏ ∂x , we can also
identify the conjugate variables n and θ. These obey a kind of uncertainty relation, so that the more precisely we
know n, the less knowledge we have of θ, and vice versa.

12
3 Linear Response
3.1 Experimental Methods
Experimental condensed-matter physics can be roughly subdivided into three broad categories of analytical technique:
(1) Experiments probing thermodynamic coefficients, such as the specific heat, cV = ∂U/∂T , i.e. the rate of change
of the internal energy with a change in temperature; the magnetic susceptibility, χ = ∂M/∂H, i.e. the change of
magnetization in response to a static magnetic field; the isothermal compressibility, κ = −V −1 ∂V /∂p, i.e. the volume
change in response to the external pressure.
(2) Transport experiments, in which an applied generalized voltage (electrical, thermal or magnetic) gives rise to
a flow of current (electrical, thermal or spin), and the ratio of current to voltage defines a conductance, g = I/V .
(3) Spectroscopic experiments, in which a beam of particles, either massive (electrons, neutrons, muons, atoms,
etc.) or massless (photons), is directed onto a sample. The kinematic information about the source beam is contained
in the dispersion relation, (k, ωk), while the accessible information about the scattering process is contained in the
frequency-momentum distribution of the scattered particles, P (ω(k′ ), k′ ), as monitored by a detector.
From the above, we can see that the interaction of a many-body system with its environment is almost always
mediated by electromagnetic forces. Most experiments subject the system under consideration to some external
electromagnetic perturbation, near a point r′ at time t′ , and then measure the response of the system, near a point r
at time t. For example, if a metal is subjected to a weak electromagnetic field, we say that the scalar potential, ϕ(r, t),
couples to the local electronic charge density operator, ρ̂(r) = −en̂(r), where n̂(r) is the electron number density
operator, causing a disturbance that propagates to other parts of the system, while the vector potential, A(r), couples
to the current density, ĵ(r).

3.2 Zero-Temperature Formalism


Consider an interacting many-body system with a time-independent Hamiltonian, Ĥ. The state vector in the
Schrödinger picture satisfies the Schrödinger equation

iℏ |ΨS (t)⟩ = Ĥ |ΨS (t)⟩
∂t
which has the solution
|ΨS (t)⟩ = e−iĤt/ℏ |ΨS (0)⟩
Suppose that the system is subject to a perturbation due to an external field at t = t0 , so that we have an additional
time-dependent Hamiltonian Ĥ ex (t). (Note that even though this Hamiltonian is time-dependent, we still consider it
to be a Schrödinger-picture operator; for convenience, however, we omit the subscript ‘S’ on both this operator and
Ĥ.) For t > t0 we now have a new Schrödinger-picture state vector, |Ψ̄S (t)⟩, which satisfies the modified equation

∂ h i
iℏ |Ψ̄S (t)⟩ = Ĥ + Ĥ ex (t) |Ψ̄S (t)⟩
∂t
We seek a solution of the form
|Ψ̄S (t)⟩ = e−iĤt/ℏ Â(t) |ΨS (0)⟩
where the operator Â(t) satisfies the causal boundary condition

Â(t) = 1 for t ⩽ t0

which means that for t ⩽ t0 , we have |Ψ̄S (t)⟩ = |ΨS (t)⟩. Substituting the trial solution into the corresponding
Schrödinger equation yields
∂ h −iĤt/ℏ i h ih i
iℏ e Â(t) |ΨS (0)⟩ = Ĥ + Ĥ ex (t) e−iĤt/ℏ Â(t) |ΨS (0)⟩
" ∂t #
−iĤt/ℏ ∂ Â(t)
h i
−iĤt/ℏ
=⇒ iℏe + Ĥe Â(t) |ΨS (0)⟩ = Ĥe−iĤt/ℏ Â(t) + Ĥ ex (t)e−iĤt/ℏ Â(t) |ΨS (0)⟩
∂t
" #
−iĤt/ℏ ∂ Â(t)
h i
=⇒ iℏe |ΨS (0)⟩ = Ĥ ex (t)e−iĤt/ℏ Â(t) |ΨS (0)⟩
∂t

13
Multiplying on the left by eiĤt/ℏ , we obtain

∂ Â(t) h i
iℏ |ΨS (0)⟩ = eiĤt/ℏ Ĥ ex (t)e−iĤt/ℏ Â(t) |ΨS (0)⟩
∂t
Because |ΨS (0)⟩ is arbitrary, we can rewrite the above as the operator equation

∂ Â(t)
iℏ = eiĤt/ℏ Ĥ ex (t)e−iĤt/ℏ Â(t) ≡ ĤH
ex
(t)Â(t)
∂t
ex
where ĤH (t) is the Hamiltonian due to the external field in the Heisenberg picture.
The above differential equation can be converted to an integral equation as follows (where we have introduced a
new variable, t′ , so that the solution will be expressed in terms of t):

∂ Â(t′ ) ex ′
iℏ = ĤH (t )Â(t′ )
∂t′
Z Â(t) Z t
=⇒ iℏ dÂ(t′ ) = dt′ ĤH
ex ′
(t )Â(t′ )
Â(t0 ) t0
h i Z t
=⇒ iℏ Â(t) − Â(t0 ) = dt′ ĤH
ex ′
(t )Â(t′ )
t0

and noting that Â(t0 ) = 1, we have


Z t
Â(t) = 1 − iℏ−1 dt′ ĤH
ex ′
(t )Â(t′ )
t0

This is a Volterra equation of the second kind, which has the general form (see Hassani, pp.488-492):
Z x
u(x) = v(x) + λ dy K(x, y)u(y)
a

and can be solved iteratively. The first approximation to u(x) is obtained by substituting v(y) for u(y) in the integrand,
i.e. yields Z x
u1 (x) = v(x) + λ dy K(x, y)v(y)
a
This is then substituted into the integrand of the original equation to obtain the second approximation, and so on.
Returning to our integral equation, we find that v(x) ≡ 1 and u(y) ≡ 1, so that the first approximation to Â(t) is
given by Z t
Â1 (t) = 1 − iℏ−1 dt′ ĤH
ex ′
(t )
t0

and the exact solution is an infinite series:


Z t
Â(t) = 1 − iℏ−1 dt′ ĤH
ex ′
(t ) + . . .
t0

ex
Note that the causal boundary condition , Â(t) = 1 for t ⩽ t0 , is automatically satisfied because ĤH (t) = 0 for t < t0 .
The physical quantities, or dynamical variables, for our system are represented by Hermitian operators, and the
expectation value of a particular dynamical variables can be obtained from the matrix element of the corresponding
operator. The expectation value of the dynamical variable O(t), represented by the Schrödinger-picture operator
Ô(t) (which may depend explicitly on time), after the perturbation has been applied, can be expanded as a series in

14
ex
increasing powers of ĤH as follows:

⟨O(t)⟩ex ≡ ⟨Ψ̄S (t)|ÔS (t)|Ψ̄S (t)⟩


= ⟨ΨS (0)|† (t)eiĤt/ℏ ÔS (t)e−iĤt/ℏ Â(t)|ΨS (0)⟩
 Z t   Z t 
−1 ′ ex ′ −1 ′ ex ′
= ⟨ΨS (0)| 1 + iℏ dt ĤH (t ) + . . . ÔH (t) 1 + iℏ dt ĤH (t ) + . . . |ΨS (0)⟩
t0 t0
Z t
= ⟨ΨS (0)|ÔH (t)|ΨS (0)⟩ + iℏ−1 ⟨ΨS (0)| dt′ ĤH
ex ′
(t )ÔH (t) |ΨS (0)⟩
t0
Z t
−1
− iℏ ⟨ΨS (0)| dt′ ÔH (t)ĤH
ex ′
(t ) |ΨS (0)⟩ + . . .
t0
Z t h i
= ⟨ΨS (0)|ÔH (t)|ΨS (0)⟩ + iℏ−1 ⟨ΨS (0)| dt′ ĤH
ex ′
(t ), ÔH (t) |ΨS (0)⟩ + . . .
t0
Z t h i
= ⟨ΨH |ÔH (t)|ΨH ⟩ + iℏ−1 ⟨ΨH | ex ′
dt′ ĤH (t ), ÔH (t) |ΨH ⟩ + . . .
t0

where we have used the fact that |ΨS (0)⟩ = |ΨH (0)⟩ ≡ |ΨH ⟩. The first term on the RHS can be expressed in the
compact form
⟨O(t)⟩ ≡ ⟨ΨH |ÔH (t)|ΨH ⟩
ex
while the second term on the RHS is linear in ĤH , and if we choose to retain only these two terms and no higher-order
ones, we can define the linear response to the external perturbation via

δ⟨O(t)⟩ ≡ ⟨O(t)⟩ex − ⟨O(t)⟩


Z t h i
= iℏ−1 dt′ ⟨ΨH | ĤH
ex ′
(t ), ÔH (t) |ΨH ⟩
t0

If |ΨH ⟩ corresponds to the Heisenberg interacting ground state, |Ψ0 ⟩, the linear response of the ground-state expectation
value of the dynamical variable O(t) is given by
Z t h i
−1
δ⟨O(t)⟩0 = iℏ dt′ ⟨Ψ0 | ĤH
ex ′
(t ), ÔH (t) |Ψ0 ⟩
t0

If we introduce an operator density, ÔH (r, t), such that


Z
ÔH (t) = d3 r ÔH (r, t)

as well as an explicit expression for the external perturbation


Z
ex ′
ĤH (t ) = d3 r′ ÔH (r′ , t′ )F (r′ , t′ )

where F (r′ , t′ ) is known as a generalized force, then the linear response is given by
Z t Z n o
δ⟨O(r, t)⟩0 = iℏ−1 dt′ d3 r′ ⟨Ψ0 | ÔH (r′ , t′ )ÔH (r, t)F (r′ , t′ ) − ÔH (r, t)ÔH (r′ , t′ )F (r′ , t′ ) |Ψ0 ⟩
t0
Z t Z h i
= iℏ−1 dt′ d3 r′ ⟨Ψ0 | ÔH (r′ , t′ ), ÔH (r, t) |Ψ0 ⟩ F (r′ , t′ )
t0

If we define the retarded response function via


h i
C0R (r, t; r′ , t′ ) ≡ −θ(t − t′ ) ⟨Ψ0 | ÔH (r, t), ÔH (r′ , t′ ) |Ψ0 ⟩

then we can write Z ∞ Z


δ⟨O(r, t)⟩0 = ℏ−1 dt′ d3 r′ C0R (r, t; r′ , t′ )F (r′ , t′ )
−∞

15
where we are justified in changing the lower limit from t0 → −∞ because F (r′ , t′ ) = 0 for t′ < t0 , and the upper limit
from t → ∞ because C0R (r, t; r′ , t′ ) = 0 for t′ > t due to the step function. We can also introduce the generalized
susceptibility, χ0 (r, t; r′ , t′ ), defined via

χ(r, t; r′ , t′ ) = ℏ−1 C R (r, t; r′ , t′ )

so that we can write, alternatively


Z ∞ Z

δ⟨O(r, t)⟩0 = dt d3 r′ χ0 (r, t; r′ , t′ )F (r′ , t′ )
−∞

While F (r′ , t′ ) and δ⟨O(r, t)⟩0 are externally adjustable/observable, either as an experimental input/output, or as
parameters in the theory, the response function (i.e. the integral kernel), C R (r, t; r′ , t′ ), represents a purely intrinsic
part of the system. Response functions play an important role in connecting experiment and theory: experimentally,
they can be measured by relating the input F (r′ , t′ ) to the response δ⟨O(r, t)⟩0 , while theory attempts to predict the
response behaviour, ideally in a way that conforms with experimental observation.
If our system is translationally invariant, then C0R (r, t; r′ , t′ ) depends on r − r′ rather than r and r′ separately, and
because the Hamiltonian is time-independent, C0R (r, t; r′ , t′ ) also depends on t − t′ rather than t and t′ separately, so
we can write it as C0R (r − r′ , t − t′ ). It is also useful to introduce the Fourier transforms
Z Z
F (k, ω) ≡ dt d3 r e−ik·r eiωt F (r, t)
Z Z
δ⟨O(k, ω)⟩0 ≡ dt d3 r e−ik·r eiωt δ⟨O(r, t)⟩0
Z Z
′ ′
C0 (k, ω) ≡ d(t − t ) d3 (r − r′ ) e−ik·(r−r ) eiω(t−t ) C0R (r − r′ , t − t′ )
R ′

Starting with the transform δ⟨O(k, ω)⟩0 , and substituting our expression from earlier for δ⟨O(r, t)⟩0 into the RHS, we
obtain Z Z Z Z 
−1 3 −ik·r iωt ′ 3 ′ R ′ ′ ′ ′
δ⟨O(k, ω)⟩0 = ℏ dt d r e e dt d r C0 (r − r , t − t )F (r , t )

Introducing the new variables, r′′ = r − r′ and t′′ = t − t′ , and noting that d3 r′′ = d3 r and dt′′ = dt, we can
R R R R

rewrite this as
Z Z Z Z
′′ ′ ′′ ′
−1 ′′ 3 ′′ ′
δ⟨O(k, ω)⟩0 = ℏ dt d r dt d3 r′ e−ik·r e−ik·r eiωt eiωt C0R (r′′ , t′′ )F (r′ , t′ )

We can now carry out the integration over r′ and t′ to obtain


Z Z
′′ ′′
δ⟨O(k, ω)⟩0 = ℏ−1 dt′′ d3 r′′ e−ik·r eiωt C0R (r′′ , t′′ )F (k, ω)

followed by the integration over r′ and t′ , which yields

δ⟨O(k, ω)⟩0 = ℏ−1 C0R (k, ω)F (k, ω) = χ0 (k, ω)F (k, ω)

which shows that the system responds at the same wavevector and frequency as the perturbation.

3.3 Finite-Temperature Formalism


We now extend the preceding method to a system at a finite temperature, T , in equilibrium with a reservoir, with
which it can exchange heat and particles. In general, the system will no longer be found in its ground state, so we
assume that it is in a state |n, N ⟩, which is an exact (Heisenberg-picture) eigenstate of the unperturbed interacting
Hamiltonian, Ĥ, and the number operator, N̂ , i.e.

Ĥ |n, N ⟩ = En(N ) |n, N ⟩ and N̂ |n, N ⟩ = N |n, N ⟩

We consider an ensemble of systems, where the members of the ensemble may have different different numbers of
particles, N , as well as different energies, En(N ) , namely the grand canonical ensemble. The probability of finding the
unperturbed system in a particular N -particle state |n, N ⟩ with energy En(N ) is then given by
−1 −β(En(N ) −µN )
pn(N ) = ZG e

16
where XX
ZG = e−β(En(N ) −µN ) = tr e−β(Ĥ−µN̂ )
N n

If an operator, Ô(t), representing a physical quantity, O(t), acts on the system, its expectation value for the state
|n, N ⟩ is just the diagonal matrix element

⟨O(t)⟩n(N ) = ⟨n, N |ÔH (t)|n, N ⟩

If the probability of the system being found in the state |n, N ⟩ is pn(N ) , then the ensemble average of O(t) is given by
XX XX
−1
⟨O(t)⟩ ≡ pn(N ) ⟨O(t)⟩n(N ) = ZG e−β(En(N ) −µN ) ⟨n, N |ÔH (t)|n, N ⟩
N n N n
XX h i
−1 −β(Ĥ−µN̂ ) −1
= ZG ⟨n, N |e ÔH (t)|n, N ⟩ = ZG tr e−β(Ĥ−µN̂ ) ÔH (t)
N n

which is a kind of ‘double averaging’. If we define the (Heisenberg-picture) statistical operator for the grand canonical
ensemble via XX
−1 −β(Ĥ−µN̂ )
ρ̂H ≡ ZG e = pn(N ) |n, N ⟩ ⟨n, N |
N n

then we can express the ensemble average as


h i
⟨O(t)⟩ = tr ρ̂H ÔH (t)

Note that even though ρ̂H is a Heisenberg-picture operator, for an unperturbed system with time-independent Hamil-
tonian, Ĥ, it is also time-independent. This is because ρ̂H depends on the eigenstates {|n, N ⟩} and the probabilities
{pn(N ) }, both of which are time-independent for such a system. When we introduce a perturbation via a time-
ex
independent Hamiltonian, ĤH (t), however, the probabilities {pn(N ) } will change with time, so that ρ̂H also becomes
time-dependent and the expression for the ensemble average becomes
XX h i
⟨O(t)⟩ = pn(N ) (t) ⟨n, N |ÔH (t)|n, N ⟩ = tr ρ̂H (t)ÔH (t)
N n

In order to obtain the linear response of a system at finite temperature, we begin by making the key assumption
that the interval t − t0 , between the external perturbation being turned on and the measurement of the physical
quantity O(t) being made, is short enough that the probabilities {pn(N ) }, and hence also the statistical operator, ρ̂H ,
do not change. Based on this assumption, we can see that any change in the ensemble average ⟨O(t)⟩ must be due to
changes in the diagonal matrix elements alone, i.e.
XX
δ⟨O(t)⟩ = pn(N ) (t0 )δ ⟨n, N |ÔH (t)|n, N ⟩
N n

However, we have already derived an expression for the change in the matrix elements, before and after the perturbation
is turned on, for the T = 0 case:
Z t h i
−1
δ⟨O(t)⟩0 = iℏ dt′ ⟨Ψ0 | ĤH
ex ′
(t ), ÔH (t) |Ψ0 ⟩
t0

so it simply a matter of replacing |Ψ0 ⟩ with |n, N ⟩, which yields


XX  Z t h i 
δ⟨O(t)⟩ = pn(N ) (t0 ) iℏ−1 dt′ ⟨n, N | ĤH
ex ′
(t ), ÔH (t) |n, N ⟩
N n t0
Z t n h io
= iℏ−1 dt′ tr ρ̂0 ĤH
ex ′
(t ), ÔH (t)
t0

where ρ̂0 ≡ ρ̂H (t0 ).


ex ′
As in the zero-temperature case, if we assume that ĤH (t ) takes the form of a generalized force, F (r′ , t′ ) coupled
′ ′
to an operator density, ÔH (r , t ), i.e.
Z
ĤH (t ) = d3 r′ ÔH (r′ , t′ )F (r′ , t′ )
ex ′

17
then the linear response is given by
Z t Z n h io
δ⟨O(r, t)⟩ = iℏ−1 dt′ d3 r′ tr ρ̂0 ÔH (r′ , t′ ), ÔH (r, t) F (r′ , t′ )
t0

If we define a retarded response function and generalized susceptibility, evaluated in the grand canonical ensemble,
via
n h io
C R (r, t; r′ , t′ ) ≡ −iθ(t − t′ ) tr ρ̂0 ÔH (r, t), ÔH (r′ , t′ )
χ(r, t; r′ , t′ ) ≡ ℏ−1 C R (r, t; r′ , t′ )

then the above can be rewritten as


Z ∞ Z
δ⟨O(r, t)⟩ = ℏ−1 dt′ d3 r′ C R (r, t; r′ , t′ )F (r′ , t′ )
−∞
Z ∞ Z
= dt′ d3 r′ χ(r, t; r′ , t′ )F (r′ , t′ )
−∞

while in Fourier space we have

δ⟨O(k, ω)⟩ = ℏ−1 C R (k, ω)F (k, ω) = χ(k, ω)F (k, ω)

If the Schrödinger-picture operator, ÔH (r, t), commutes with the number operator, N̂ , which is usually the case,
then if we define a grand canonical Hamiltonian, K̂ ≡ Ĥ − µĤ, we can write

ÔH (r, t) = ei(Ĥ−µN̂ )t/ℏ ÔS (r, t)e−i(Ĥ−µN̂ )t/ℏ


= eiK̂t/ℏ ÔS (r, t)e−iK̂t/ℏ ≡ ÔK (r, t)

where ÔK (r, t) is a modified Heisenberg picture operator. The retarded response function then becomes
n h io
C R (r, t; r′ , t′ ) = −iθ(t − t′ ) tr ρ̂0 ÔK (r, t), ÔK (r′ , t′ )

It is straightforward to generalize the above formalism to the case of multiple operators, or components of a single
operator, which couple to the corresponding number of generalized forces. Dropping the subscript ‘K’ for now, we
assume that all operators are in the modified Heisenberg picture, and the unperturbed system is in an eigenstate of
the grand canonical Hamiltonian. We express the external perturbation via the Hamiltonian
Z
Ĥ (t ) = d3 r′ Ôi (r′ , t′ )Fi (r′ , t′ )
ex ′

where summation over i is implied. The linear response is then given by


Z t Z n h io
δ⟨Oi (r, t)⟩ = iℏ−1 dt′ d3 r′ tr ρ̂0 Ôj (r′ , t′ ), Ôi (r, t) Fj (r′ , t′ )
t0

We now define the retarded response function and generalized susceptibility


n h io
R
Cij (r, t; r′ , t′ ) ≡ −iθ(t − t′ ) tr ρ̂0 Ôi (r, t), Ôj (r′ , t′ )
χij (r, t; r′ , t′ ) ≡ ℏ−1 C R (r, t; r′ , t′ )

so that the linear response can be expressed as


Z ∞ Z
δ⟨Oi (r, t)⟩ = ℏ−1 dt′ d3 r′ Cij R
(r, t; r′ , t′ )Fj (r′ , t′ )
−∞
Z ∞ Z
= dt′ d3 r′ χij (r, t; r′ , t′ )Fj (r′ , t′ )
−∞

18
3.4 Example: Noninteracting Electron Gas
We now consider the response of a noninteracting electron gas to a general external electromagnetic field. In doing so,
we first note that the ‘external’ field introduced in the preceding sections corresponds in the electromagnetic case to
the total contribution (external plus induced), represented by the potentials ϕ(r, t) and A(r, t). These couple to the
induced charge density ρ̂(r, t) and current density Ĵ(r, t), respectively, and hence there are two different cases to be
considered: (1) density response, and (2) current response.

3.4.1 Density-Density Correlation Function


We first consider the response of a non-interacting electron gas at finite temperature to a scalar potential ϕ(r, t), which
couples to the (modified Heisenberg-picture) charge density operator ρ̂K (r, t) as follows:
Z Z
ĤK (t ) = d r ρ̂K (r , t )ϕ(r , t ) = −e d3 r′ n̂K (r′ , t′ )ϕ(r′ , t′ )
ex ′ 3 ′ ′ ′ ′ ′

where we have introduced the particle number density operator, n̂K (r, t). Using the formulation derived earlier, we
find that the linear response of the number density to the perturbation is given by
Z ∞ Z
−1 ′
δ⟨n(r, t)⟩ = −eℏ dt d3 r′ C R (r, t; r′ , t′ )ϕ(r′ , t′ )
−∞

where the retarded density-density correlation function is defined via

C R (r, t; r′ , t′ ) ≡ −iθ(t − t′ ) tr {ρ̂0 [n̂K (r, t), n̂K (r′ , t′ )]} ≡ −iθ(t − t′ )⟨[n̂K (r, t), n̂K (r′ , t′ )]⟩

in which ⟨· · · ⟩ denotes the ensemble average.


Since K̂ = Ĥ − µN̂ is time-independent and the system is translationally invariant, C R (r, t; r′ , t′ ) depends on r − r′
and t − t′ , and we can therefore expand it as a Fourier series in terms of a wavevector q:
X ′
C R (r, t; r′ , t′ ) = V −1 eiq·(r−r ) C R (q, t − t′ )
q

where V is the volume of the system. Similarly, we can decompose n̂K (r, t) and n̂K (r′ , t′ ) into Fourier components:
X X ′ ′
n̂K (r, t) = V −1 eiq·r n̂K (q, t), n̂K (r′ , t′ ) = V −1 eiq ·r n̂K (q′ , t′ )
q q′

so that we may write


XX ′ ′
C R (r, t; r′ , t′ ) = −iθ(t − t′ )V −2 eiq·r eiq ·r ⟨[n̂K (q, t), n̂K (q′ , t′ )]⟩
q q′
XX ′ ′ ′
= −iθ(t − t′ )V −2 eiq·(r−r ) ei(q+q )·r ⟨[n̂K (q, t), n̂K (q′ , t′ )]⟩
q q′

Since the RHS must depend on r − r′ , and not independently on r′ , it follows that q + q′ = 0, i.e. q′ = −q. Therefore
X ′
C R (r, t; r′ , t′ ) = −iθ(t − t′ )V −2 eiq·(r−r ) ⟨[n̂K (q, t), n̂K (−q, t′ )]⟩
q

Comparing the above with the Fourier expansion for C R (r, t; r′ , t′ ), we find that

C R (q, t − t′ ) = −iθ(t − t′ )V −1 ⟨[n̂K (q, t), n̂K (−q, t′ )]⟩

Taking as an example the operator n̂K (q, t), this can be expressed as follows (see Jishi, p.115):
X †
n̂K (q, t) = eiK̂t/ℏ n̂q e−iK̂t/ℏ = ĉkσ (t)ĉk+qσ (t)

So far, everything we have written applies equally well to an interacting electron gas, but for our case of a non-
interacting system we can make the substitution (see Jishi, p. 108):

ĉ†kσ (t) = ĉ†kσ eiϵ̄kσ t/ℏ , ĉk+qσ (t) = ĉk+qσ e−iϵ̄k+qσ t/ℏ

19
where ϵ̄kσ is the single-particle energy state relative to the chemical potential. This means that
X †
n̂K (q, t) = ĉkσ ĉk+qσ ei(ϵ̄kσ −ϵ̄k+qσ )t/ℏ

The retarded density-density correlation function for the non-interacting electron gas can then be written as (see
Jishi, p.116):

ei(ϵ̄kσ −ϵ̄k+qσ )t/ℏ ei(ϵ̄k′ σ′ −ϵ̄k′ −qσ′ )t /ℏ ⟨[ĉ†kσ ĉk+qσ , ĉ†k′ σ′ ĉk′ −qσ′ ]⟩
XX
C R,0 (q, t − t′ ) = −iθ(t − t′ )V −1
kσ k′ σ ′

Evaluating the commutator gives


 
⟨[ĉ†kσ ĉk+qσ , ĉ†k′ σ′ ĉk′ −qσ′ ]⟩ = ĉ†kσ ĉk′ −qσ′ − ĉ†k′ σ′ ĉk+qσ δσσ′ δk+q,k′

which yields for the correlation function



ei(ϵ̄kσ −ϵ̄k+qσ )(t−t )/ℏ ⟨ĉ†kσ ĉkσ − ĉ†k+qσ ĉk+qσ ⟩
X
C R,0 (q, t − t′ ) = −iθ(t − t′ )V −1

X ′
′ −1
= −iθ(t − t )V ei(ϵ̄kσ −ϵ̄k+qσ )(t−t )/ℏ (fkσ − fk+qσ )

where fkσ is the Fermi-Dirac distribution function.


The next step is to take the Fourier transform w.r.t. t (setting t′ = 0):
Z ∞ X Z ∞
−1
R,0
C (q, ω) = iωt R,0
dt e C (q, t) = −iV (fkσ − fk+qσ ) dt e[iω+(ϵ̄kσ −ϵ̄k+qσ )/ℏ]t
−∞ kσ 0
X Z ∞
−1
= −iV (fkσ − fk+qσ ) lim dt e[iω+(ϵ̄kσ −ϵ̄k+qσ )/ℏ+iη]t
η→0+ 0

X fkσ − fk+qσ
= V −1
ω + (ϵ̄kσ − ϵ̄k+qσ )/ℏ + i0+

The linear density response of the non-interacting system can then be expressed in Fourier space as
δ⟨n(q, ω)⟩0 = −eℏ−1 C R,0 (q, ω)ϕ(q, ω) = −eχ0 (q, ω)ϕ(q, ω)
with the generalized susceptibility given by
X fkσ − fk+qσ
χ0 (q, ω) = (ℏV )−1
ω + (ϵ̄kσ − ϵ̄k+qσ )/ℏ + i0+

This is known as the Lindhard function.

3.4.2 Current Density Operator


Before considering current response, we need to derive the form of the current density operator in quantum field
theory. Starting with the classical expression
j = −env
the contribution of electron i to the current density is
e
ji (r) = −eδ(r − ri )vi = − δ(r − ri )pi
m
where the particle number density is obtained from (see Jishi, p.49):
Z
ni (r) d3 r = 1 =⇒ ni (r) = δ(r − ri )

For N particles we therefore have


N N
X e X
n(r) = δ(r − ri ), j(r) = − δ(r − ri )pi
i=1
m i=1

20
When moving to quantum mechanics, we find that the following obvious choice for the corresponding (Schrödinger-
picture) current density operator:
N
e X
ĵ(r) = − δ(r − ri )p̂i
m i=1
is not Hermitian, so we need to use the ‘average’ (see Bransden and Joachain, p.96):
N
e X
ĵ(r) = − [p̂i δ(r − ri ) + δ(r − ri )p̂i ]
2m i=1

In the presence of a vector potential A(ri ), we have


e
p̂i → p̂i + A(ri )
c
so that the expression for the current density operator becomes
N
e X h e   e i
ĵ(r) = − p̂i + A(ri ) δ(r − ri ) + δ(r − ri ) p̂i + A(ri )
2m i=1 c c

In quantum field theory we second quantize spin-independent operators according to (see Jishi, pp.58-59):
N
X XZ
Ô(r) = ôi (r) → Ô(r) = Ψ̂†σ (r)ô(r)Ψ̂σ (r) d3 r
i=1 σ

where the field operator Ψ̂†σ (r)(Ψ̂σ (r)) creates (annihilates) a particle with spin projection σ (↑ or ↓) at position r.
The current density operator therefore becomes
Z
e X h e   e i
ĵ(r) = − Ψ̂†σ (r′ ) p̂ + A(r′ ) δ(r − r′ ) + δ(r − r′ ) p̂ + A(r′ ) Ψ̂σ (r′ ) d3 r′
2m σ c c

We consider each of the terms in turn, starting with


Z Z
e X † ′
h
′ ′
i
3 ′ ieℏ X h i
ĵ1 (r) = − Ψ̂σ (r ) p̂ δ(r − r )Ψ̂σ (r ) d r = Ψ̂†σ (r′ ) ∇r′ δ(r − r′ )Ψ̂σ (r′ ) d3 r′
2m σ 2m σ

Integrating by parts yields


 i∞ Z h 
ieℏ X h † ′ ′ ′ † ′
i
′ ′ 3 ′
ĵ1 (r) = Ψ̂σ (r )δ(r − r )Ψ̂σ (r ) − ∇r′ Ψ̂σ (r ) δ(r − r )Ψ̂σ (r ) d r
2m σ −∞

The first part on the RHS is zero, since Ψ̂†σ (r′ ), Ψ̂σ (r′ ) = 0 at r′ = ±∞, and so after evaluating the integral we have
ieℏ X  
ĵ1 (r) = − ∇r Ψ̂†σ (r) Ψ̂σ (r)
2m σ

Turning now to the second term, we have


Z
e X e 
ĵ2 (r) = − Ψ̂†σ (r′ ) A(r′ ) δ(r − r′ )Ψ̂σ (r′ )d3 r′
2m σ c

There is no operator to worry about this time, so integral simply yields

e2 X e2
ĵ2 (r) = − A(r)Ψ̂†σ (r)Ψ̂σ (r) = − A(r)n̂(r)
2mc σ 2mc

where the (Schrödinger-picture) number density operator is defined via


X
n̂(r) ≡ Ψ̂†σ (r)Ψ̂σ (r)
σ

21
The third term is similar to the first:
Z Z
e X † ′ ′ ′ 3 ′ ieℏ X
ĵ3 (r) = − Ψ̂σ (r ) δ(r − r )p̂Ψ̂σ (r )d r = Ψ̂†σ (r′ ) δ(r − r′ )∇r′ Ψ̂σ (r′ )d3 r′
2m σ 2m σ

although in this case the integral can be evaluated immediately to yield


ieℏ X †
ĵ3 (r) = Ψ̂ (r)∇r Ψ̂σ (r)
2m σ σ

while the fourth term is identical to the second, i.e. ĵ4 (r) = ĵ2 (r).
Putting all this together, we have
ieℏ X h †   i e2
ĵ(r) = Ψ̂σ (r)∇r Ψ̂σ (r) − ∇r Ψ̂†σ (r) Ψ̂σ (r) − A(r)n̂(r)
2m σ mc
≡ ĵP (r) + ĵD (r)

where ĵP (r) and ĵD (r) are defined as the paramagnetic current density and diamagnetic current density, respectively.
Another way to obtain the current density operator is to start from the expression for the Hamiltonian in the
presence of a vector potential, A (and with ϕ = 0):
XZ  
1  e
Ĥ tot = Ψ̂†σ (r) p̂ + A(r) Ψ̂σ (r) d3 r
σ
2m c
e2
Z  
X p̂ e e 2
= Ψ̂†σ (r) + p̂ · A(r) + A(r) · p̂ + 2
(A(r)) Ψ̂σ (r) d3 r
σ
2m 2mc 2mc 2mc
= Ĥ + Ĥ ex

where the perturbation Ĥ ex is given by


e2
XZ  
ex † e e 2
Ĥ = Ψ̂σ (r) p̂ · A(r) + A(r) · p̂ + (A(r)) Ψ̂σ (r) d3 r
σ
2mc 2mc 2mc2
X Z  ieℏ h e2 h
 i i 
= ∇r Ψ̂†σ (r) Ψ̂σ (r) − Ψ̂†σ (r)∇r Ψ̂σ (r) · A(r) + 2
A(r) Ψ̂†
σ (r) Ψ̂σ (r) · A(r) d3 r
σ
2mc mc
Z
1
=− ĵ(r) · A(r) d3 r
c

Hence, Ĥ ex is a functional of A(r), and so we can obtain ĵ(r) by functional differentiation:

δ Ĥ ex [A] 1 δ Ĥ ex [A]
= − ĵ(r) =⇒ ĵ(r) = −c
δA(r) c δA(r)

3.4.3 Current-Current Correlation Function


We now consider the response of a non-interacting electron gas at finite temperature to a vector potential A(r, t),
which couples to the (modified Heisenberg-picture) current density operator ĵK (r, t) as follows:
Z
ex ′ 1
ĤK (t ) = − d3 r′ ĵK (r′ , t′ ) · A(r′ , t′ )
c

where ĵK (r′ , t′ ) is obtained from the (Schrödinger-picture) current density operator derived above as follows:

ĵK (r, t) ≡ eiK̂t/ℏ ĵ(r) e−iK̂t/ℏ

We can also split ĵK (r′ , t′ ) into paramagnetic and diamagnetic parts:

ĵK (r, t) ≡ ĵP D


K (r, t) + ĵK (r, t)
ieℏ X h †   i e2 h i
= Ψ̂Kσ (r, t)∇r Ψ̂Kσ (r) − ∇r Ψ̂†Kσ (r, t) Ψ̂Kσ (r, t) − A(r, t) Ψ̂†Kσ (r, t)Ψ̂Kσ (r)
2m σ mc

22
where summation over the spin index σ is implied.
To determine the expectation value of the total current density in the presence of the vector potential A(r, t), we
consider the linear response (i.e. the terms linear in A(r, t)) of the diamagnetic and paramagnetic parts of ĵK (r, t)
separately:
(a) Diamagnetic term:
Z t Z
i ′
D
⟨j (r, t)⟩ex = ⟨jD
K (r, t)⟩ + dt d3 r′ ⟨[ĵD D ′ ′ ′ ′
K (r, t), ĵK (r , t ) · A(r , t )]⟩ + . . .
ℏc −∞

The only term on the RHS linear in A(r, t) is the first one, and hence to a first approximation we have

e2 ne2
⟨jD (r, t)⟩ex = ⟨jD
K (r, t)⟩ = − A(r, t)⟨Ψ̂†Kσ (r, t)Ψ̂Kσ (r)⟩ = − A(r, t)
mc mc
where n is the equilibrium electron number density. Note that this term is not identically zero, as we are not assuming
that A = 0; we are simply evaluating the expectation value of the operator ĵD
K in the equilibrium (i.e. A = 0) quantum
state.
(b) Paramagnetic term:
Z t Z
i
⟨jP (r, t)⟩ex = ⟨jP
K (r, t)⟩ + dt′ d3 r′ ⟨[ĵP P ′ ′ ′ ′
K (r, t), ĵK (r , t ) · A(r , t )]⟩ + . . .
ℏc −∞

In this case, the first term on the RHS is zero, while the second term on the RHS is the only one linear in A, and
hence to a first approximation we have
Z t Z
i
⟨jP (r, t)⟩ex = dt′ d3 r′ ⟨[ĵP P ′ ′ ′ ′
K (r, t), ĵK (r , t ) · A(r , t )]⟩
ℏc −∞

We are now in a position to write down the total current density in the presence of the vector potential, to first
order in A:
Z t
ne2
Z
i ′
D P
⟨j(r, t)⟩ex = ⟨j (r, t)⟩ex + ⟨j (r, t)⟩ex = − A(r, t) + dt d3 r′ ⟨[ĵP P ′ ′ ′ ′
K (r, t), ĵK (r , t ) · A(r , t )]⟩
mc ℏc −∞

The current density operator is a (three-component) vector operator, and we can express the above in component
notation as follows:
Z t
ne2
Z
i X
⟨ji (r, t)⟩ex = − Ai (r, t) + dt′ d3 r′ P
⟨[ĵKi P
(r, t), ĵKj (r′ , t′ )]⟩Aj (r′ , t′ )
mc ℏc −∞ j
2 Z t Z
ne 1 X R,P
=− Ai (r, t) − dt′ d3 r′ Cij (r, t; r′ , t′ )Aj (r′ , t′ )
mc ℏc −∞ j

ne2 1 t
Z Z X
=− Ai (r, t) − dt′ d3 r′ χP ′ ′ ′ ′
ij (r, t; r , t )Aj (r , t )
mc c −∞ j

where
R,P
Cij (r, t; r′ , t′ ) ≡ −iθ(t − t′ )⟨[ĵKi
P P
(r, t), ĵKj (r′ , t′ )]⟩ ≡ ℏχP ′ ′
ij (r, t; r , t )

For a translationally-invariant system we can introduce the Fourier transforms


Z Z
Ai (q, ω) ≡ dt d3 r e−iq·r eiωt Ai (r, t)
Z Z
⟨ji (q, ω)⟩ex ≡ dt d3 r e−iq·r eiωt ⟨ji (r, t)⟩ex
Z Z
R,P ′ ′ R,P
Cij (q, ω) ≡ d(t − t′ ) d3 (r − r′ ) e−iq·(r−r ) eiω(t−t ) Cij (r − r′ , t − t′ )
Z Z
′ ′
χij (q, ω) ≡ d(t − t ) d3 (r − r′ ) e−iq·(r−r ) eiω(t−t ) χP
P ′ ′ ′
ij (r − r , t − t )

23
and using the result from earlier we find that

ne2 X R,P
⟨ji (q, ω)⟩ex = − Ai (q, ω) − (ℏc)−1 Cij (q, ω)Aj (q, ω)
mc j
X  ne2 R,P

= − δij − (ℏc)−1 Cij (q, ω) Aj (q, ω)
j
mc
ne2
X  
−1 P
= − δij − c χij (q, ω) Aj (q, ω)
j
mc

and hence we can write


ne2
ji (q, ω) ≡ ⟨ji (q, ω)⟩ex = −c−1 χij (q, ω)Aj (q, ω), with χij (q, ω) ≡ δij + χP
ij (q, ω)
m
where repeated indices are assumed to be summed over.
To obtain an expression for the current density resulting from an applied electric field (and neglecting the induced
polarization field), we start by expressing the latter in terms of the vector potential as follows:

1 ∂A(q, ω) 1 ∂Ai (q, ω)


E(q, ω) = − , or Ei (q, ω) = −
c ∂t c ∂t
Expressing each side as an inverse Fourier transform then yields
Z Z Z Z 
1 ∂
dω d3 q e−iq·r eiωt Ei (q, ω) = − dω d3 q e−iq·r eiωt Ai (q, ω)
c ∂t
1 c
=⇒ Ei (q, ω) = − (−iω)Ai (q, ω), or Ai (q, ω) = Ei (q, ω)
c iω
Substituting in the expression for ji (q, ω) above then gives

1 ne2
 
ji (q, ω) = − δij + χP
ij (q, ω) Ej (q, ω)
iω m

which is simply Ohm’s law:


j(q, ω) = σ(q, ω)E(q, ω)
where the components of the conductivity tensor are given by

1 ne2
 
σij (q, ω) = − δij + χP
ij (q, ω)
iω m

This is known as the Kubo formula for the conductivity.


If we set r′ = 0, t′ = 0, the expression for the Fourier transform of the paramagnetic response function, χP
ij (q, ω),
is simplified somewhat, and the Kubo formula for the conductivity becomes

1 ne2
 Z Z 
σij (q, ω) = − δij − (iℏ)−1 dt d3 r e−iq·r eiωt θ(t)⟨[ĵKi
P P
(r, t), ĵKj (0, 0)]⟩
iω m

3.4.4 Four-Vector Notation


We start with the action for an electromagnetic field with a source term (dropping the hats over operators for now):
Z  
tot ex 4 1 µν 1 µ
S [A] ≡ S[A] + S [A] = d x − Fµν F − j Aµ
4 c

where j µ = (cρ, jP ) and Aµ = (ϕ, −A). However, by definition


Z
1
S [A] ≡ d4 x Lex =⇒ Lex = − j µ Aµ
ex
c

24
and
∂Lex 1
Hex ≡ π i ∂0 Ai − Lex = − Lex = j µ Aµ
∂(∂0 Ai ) c
since Lex is independent of ∂0 Ai . Finally, we can write
Z Z
3 ′ 1
ex
H ≡ d r H = ex
d 3 r ′ j µ Aµ
c
or, in terms of modified Heisenberg-picture operators:
Z
ex ′ 1 µ ′ ′
HK (t ) = d3 r ′ jK (r , t )Aµ (r′ , t′ )
c
Using the results from earlier, the linear response of the current density in four-vector notation is given by
Z t Z Z t Z
−1 ′ 3 ′ ex ′ µ i ′ µ
µ
δ⟨j (r, t)⟩ = iℏ dt d r ⟨[HK (t ), jK (r, t)]⟩ = − dt d3 r′ ⟨[jK ν
(r, t), jK (r′ , t′ )]⟩Aν (r′ , t′ )
−∞ ℏc −∞

Taking the µ = 0 component (and reintroducing the hats for operators) gives

ic t
Z Z
cδ⟨ρ(r, t)⟩ = − dt′ d3 r′ ⟨[ρ̂K (r, t), ρ̂K (r′ , t′ )]⟩ϕ(r′ , t′ )
ℏ −∞

which, after substituting ρ̂(r, t) = −en̂(r, t), yields

ie t
Z Z
δ⟨n(r, t)⟩ = dt′ d3 r′ ⟨[n̂K (r, t), n̂K (r′ , t′ )]⟩ϕ(r′ , t′ )
ℏ −∞
which agrees with the result for density-density correlations. The µ = 1, 2, 3 components, on the other hand, yield
Z t Z
i ′
P
δ⟨j (r, t)⟩ = dt d3 r′ ⟨[ĵP P ′ ′ ′ ′
K (r, t), ĵK (r , t )]⟩ · A(r , t )
ℏc −∞
so that
t
ne2
Z Z
i
⟨j(r, t)⟩ex = − A(r, t) + dt′ d3 r′ ⟨[ĵP P ′ ′ ′ ′
K (r, t), ĵK (r , t )]⟩ · A(r , t )
mc ℏc −∞

which agrees with the result for current-current correlations.

3.4.5 Spin-Density Correlation Function


We next consider the response of a non-interacting electron gas at finite temperature to a magnetic field, B(r, t),
applied at t = t0 . We ignore the orbital response of the electrons, and consider only the interaction of the magnetic
field with the spins of the electrons.
The magnetic moment operator of an electron is given by (in CGS units):
gs µb e
µ̂s = − Ŝ ≃ − Ŝ
ℏ mc
where gs ≃ 2 is the spin g-factor and µb = eℏ/2mc is the Bohr magneton. We shall take the above relation to be
exact, and hence write the Hamiltonian of the spin in an external magnetic field as
e
Ĥ ex = −µ̂s · B = Ŝ · B
mc
(Note that the usual factor of 1/2 resulting from Thomas precession is absent here.) In terms of modified Heisenberg
picture operators, we may express this as
Z
ex ′ e
ĤK (t ) = d3 r′ ŝK (r′ , t′ ) · B(r′ , t′ )
mc
where we have introduced the spin density, defined via
Z

ŜK (t ) ≡ d3 r′ ŝK (r′ , t′ )

25
If we introduce the magnetic moment density or magnetization, via
e
m̂K (r′ , t′ ) ≡ − ŝK (r′ , t′ )
mc
then the Hamiltonian becomes Z
ex ′
ĤK (t ) = − d3 r′ m̂K (r′ , t′ ) · B(r′ , t′ )

The linear response of the magnetization to the perturbation can now be expressed (in component notation) as:
Z t Z X
−1 ′
δ⟨mi (r, t)⟩ = iℏ dt d3 r ′ ⟨[m̂Ki (r, t), m̂Kj (r′ , t′ )]⟩Bj (r′ , t′ )
−∞ j
2 Z t Z
ie X
= dt′ d3 r′ ⟨[ŝKi (r, t), ŝKj (r′ , t′ )]⟩Bj (r′ , t′ )
ℏm2 c2 −∞ j
2 Z ∞ Z
e X
=− dt′ d3 r ′ R
Cij (r, t; r′ , t′ )Bj (r′ , t′ )
ℏm2 c2 −∞ j

where
R
Cij (r, t; r′ , t′ ) ≡ −iθ(t − t′ )⟨[ŝKi (r, t), ŝKj (r′ , t′ )]⟩
To determine the thermal average in the commutator above, we begin by noting that for N electrons at positions
r1 , r2 , . . . , rN , the spin-density operator (i.e. the spin per unit volume at r) is given by
N
X
ŝ(r) = δ(r − ri )Ŝi
i=1

where Ŝi is the spin operator for electron i. We now second-quantize the spin-density operator using a basis of
plane-wave states, {|kσ⟩}:
⟨kσ|δ(r − r′ )Ŝ|k′ σ ′ ⟩ ĉ†kσ ĉk′ σ′
XX
ŝ(r) =
kσ k′ σ ′

Noting that

Ŝ |kσ⟩ = σ̂ |kσ⟩
2
where σ̂ = (σ̂x , σ̂y , σ̂z ), and substituting
Z
1 ′
|kσ⟩ = V − 2 d3 r′ eik·r |σ⟩

we can express the spin-density operator as follows:


ℏ X XZ ′ ′ ′
ŝ(r) = ′
⟨σ|σ̂|σ ⟩ d3 r′ e−ik·r δ(r − r′ )eik ·r ĉ†kσ ĉk′ σ′
2V ′ ′
σσ kk

To eliminate k , we introduce the new variable q = k′ − k, and note that


Z

d3 r′ δ(r − r′ )eiq·r = eiq·r

which yields
ℏ X X iq·r
ŝ(r) = e ⟨σ|σ̂|σ ′ ⟩ ĉ†kσ ĉk+qσ′ = ŝ(r)
2V ′ σσ kq

Introducing the Fourier transform of ŝ(r) via


X
ŝ(r) = V −1 eiq·r ŝ(q)
q

we can then make the identification


ℏ XX
ŝ(q) = ⟨σ|σ̂|σ⟩ ĉ†kσ ĉk+qσ′
2 ′k σσ

26
Switching to the modified Heisenberg picture, we can express this as
ℏ XX
ŝK (q, t) = eiK̂t/ℏ ŝ(q)e−iK̂t/ℏ = ⟨σ|σ̂|σ ′ ⟩ ĉ†kσ (t)ĉk+qσ′ (t)
2 ′k σσ

Since we are assuming that our system is non-interacting, we can make the substitution:

ĉ†kσ (t) = ĉ†kσ eiϵ̄kσ t/ℏ , ĉk+qσ′ (t) = ĉk+qσ e−iϵ̄k+qσ′ t/ℏ

where ϵ̄kσ is the single-particle energy state relative to the chemical potential, and this yields the following expression
for ŝK (q, t):
ℏ XX
ŝK (q, t) = eiK̂t/ℏ ŝ(q)e−iK̂t/ℏ = ⟨σ|σ̂|σ ′ ⟩ ĉ†kσ ĉk+qσ′ ei(ϵ̄k −ϵ̄k+q )t/ℏ
2 ′ k σσ

We now follow the same procedure as in the density-density case to express the retarded correlation function as
follows:
R,0
Cij (q, t − t′ ) = −iθ(t − t′ )V −1 ⟨[ŝKi (q, t), ŝKj (−q, t′ )]⟩
iℏ2 X X
=− θ(t − t′ ) ⟨σ1 |σ̂i |σ1′ ⟩ ⟨σ2 |σ̂j |σ2′ ⟩
4V ′ ′
σ1 σ1 σ2 σ2
X i(ϵ̄k σ −ϵ̄ ′
′ )t/ℏ i(ϵ̄k2 σ2 −ϵ̄k −qσ ′ )t /ℏ
× e 1 1 k1 +qσ1 e 2 2 ⟨[ĉ†k1 σ1 ĉk1 +qσ1′ , ĉ†k2 σ2 ĉk2 −qσ2′ ]⟩
k1 k2

We then evaluate the commutator and ensemble average (see Jishi, p.119):

⟨[ĉ†k1 σ1 ĉk1 +qσ1′ , ĉ†k2 σ2 ĉk2 −qσ2′ ]⟩ = δk2 ,k1 +q δσ1′ σ2 ⟨ĉ†k1 σ1 ĉk1 σ2′ ⟩ − δk2 ,k1 +q δσ1 σ2′ ⟨ĉ†k1 +qσ2 ĉk1 +qσ1′ ⟩
= δk2 ,k1 +q δσ1′ σ2 δσ1 σ2′ ⟨ĉ†k1 ĉk1 ⟩ − δk2 ,k1 +q δσ1 σ2′ δσ1′ σ2 ⟨ĉ†k1 +q ĉk1 +q ⟩
= δk2 ,k1 +q δσ1′ σ2 δσ1 σ2′ (fk1 − fk1 +q )

where fk1 and fk1 +q are Fermi-Dirac distribution functions. Substituting in the expression for the retarded correlation
function, and setting k1 ≡ k, then yields

R,0 iℏ2 X X
Cij (q, t − t′ ) = − θ(t − t′ )δk2 ,k1 +q δσ1′ σ2 δσ1 σ2′ ⟨σ1 |σ̂i |σ1′ ⟩ ⟨σ2 |σ̂j |σ2′ ⟩
4V
σ1 σ1′ σ2 σ2′
X ′
× (fk1 − fk1 +q )ei(ϵ̄k1 −ϵ̄k1 +q )t/ℏ ei(ϵ̄k2 −ϵ̄k2 −q )t /ℏ
k1 k2

iℏ2 X X ′
=− θ(t − t′ ) ⟨σ1 |σ̂i |σ2 ⟩ ⟨σ2 |σ̂j |σ1 ⟩ (fk − fk+q )ei(ϵ̄k −ϵ̄k+q )t/ℏ ei(ϵ̄k+q −ϵ̄k )t /ℏ
4V σ σ 1 2 k
iℏ2 X X ′
=− θ(t − t′ ) ⟨σ1 |σ̂i |σ2 ⟩ ⟨σ2 |σ̂j |σ1 ⟩ (fk − fk+q )ei(ϵ̄k −ϵ̄k+q )(t−t )/ℏ
4V σ σ 1 2 k

Noting that (see Jishi, p.120): X


⟨σ1 |σ̂i |σ2 ⟩ ⟨σ2 |σ̂j |σ1 ⟩ = 2δij
σ1 σ2

we then have
R,0 iℏ2 X ′
Cij (q, t − t′ ) = − θ(t − t′ )δij (fk − fk+q )ei(ϵ̄k −ϵ̄k+q )(t−t )/ℏ
2V
k

The final step is to take the Fourier transform w.r.t. t (setting t′ = 0 and reintroducing the spin index via a factor
of 1/2):
R,0 ℏ2 X fkσ − fk+qσ
Cij (q, ω) = δij
2V ω + (ϵ̄kσ − ϵ̄k+qσ )/ℏ + i0+

27
The linear magnetization response of the non-interacting system can then be expressed in Fourier space as

e2 X R,0
δ⟨mi (q, ω)⟩0 = − C (q, ω)Bj (q, ω)
ℏm2 c2 j ij
e2 ℏ X fkσ − fk+qσ
=− 2 2
δij Bj (q, ω)
4m c V ω + (ϵ̄kσ − ϵ̄k+qσ )/ℏ + i0+

If we define the paramagnetic susceptibility, χP,0


ij (q, ω), as follows:

δ⟨mi (q, ω)⟩0 = χP,0


ij (q, ω)Bj (q, ω)

then we have
e2 ℏ X fkσ − fk+qσ X fkσ − fk+qσ
χP,0
ij (q, ω) = − 2 2
δij +
= −µ2b (ℏV )−1 δij
4m c V ω + (ϵ̄kσ − ϵ̄k+qσ )/ℏ + i0 ω + (ϵ̄kσ − ϵ̄k+qσ )/ℏ + i0+
kσ kσ

which contains the Lindhard function encountered earlier.

4 Real-Time Correlation Functions


Consider a system of interacting particles at temperature T and with a total Hamiltonian Ĥ. If the system is able to
exchange energy and particles with a reservoir, then we can think of it as a member of a grand canonical ensemble.
How consider an operator  that acts on the state space of the system and is represented in terms of particle creation
and annihilation operators ĉ† and ĉ, respectively.
In the Heisenberg picture we can define a time-dependent operator
′ ′
Â(t) = eiĤ t/ℏ
Âe−iĤ t/ℏ

where  is our original time-independent (Schrödinger picture) operator and Ĥ ′ = Ĥ − µN̂ , with µ the chemical
potential and N̂ the particle number operator.
Considering now any two operators  and B̂, we can define the real-time causal, or time-ordered, correlation
function
T
CAB (t, t′ ) = −i⟨T [Â(t)B̂(t′ )]⟩
where ⟨· · · ⟩ indicates the ensemble average of an operator (i.e. the expectation value of an operator over the ensemble
of states that the system can be in), given by
 ′

−1
⟨· · · ⟩ = ZG Tr e−β Ĥ · · ·


where β = 1/kB T and ZG = Tr(e−β Ĥ ) is the grand partition function.
In the expression for the correlation function, T [ÂB̂ · · · ] denotes the time-ordered product, which places two or more
operators in increasing time order from right to left, and introduces a minus sign each time two fermion operators are
interchanged: (
Â(t)B̂(t′ ) t > t′
T [Â(t)B̂(t′ )] =
±B̂(t′ )Â(t) t < t′

where the plus (minus) sign refers to the case when  and B̂ are boson (fermion) operators.
We can also define the retarded and advanced correlation functions as follows:
R
CAB (t, t′ ) = −iθ(t − t′ )⟨[Â(t), B̂(t′ )]± ⟩
A
CAB (t, t′ ) = +iθ(t − t′ )⟨[Â(t), B̂(t′ )]± ⟩

where ± indicates a commutator in the case of boson operators and an anti-commutator in the case of fermion
operators, and θ(t − t′ ) is the Heaviside unit step function defined as follows:
(
1 t > t′
θ(t − t′ ) =
0 t < t′

28
Finally, we define one more correlation function, which is important for analyzing experimental data:

CAB (t, t′ ) = ⟨Â(t)B̂(t′ )⟩

If the Hamiltonian Ĥ ′ is time-independent, all the correlation functions above depend on t − t′ rather than t and
t independently, so we can set t′ = 0 and consider them to be a function of t alone.

4.1 Real-Time Green’s Functions


Probably the most important correlation function is the Green’s function, expressed in terms of the field operators
Ψ̂†σ (r, t) and Ψ̂σ (r, t) that create and annihilate a particle at a position r with spin projection σ at a time t. The
causal, retarded and advanced Green’s functions are then given by

G(r, σ, t; r′ , σ ′ , t′ ) = −i⟨T [Ψ̂σ (r, t)Ψ̂†σ′ (r′ , t′ )]⟩


GR (r, σ, t; r′ , σ ′ , t′ ) = −iθ(t − t′ )⟨[Ψ̂σ (r, t), Ψ̂†σ′ (r′ , t′ )]∓ ⟩
GA (r, σ, t; r′ , σ ′ , t′ ) = +iθ(t′ − t)⟨[Ψ̂σ (r, t), Ψ̂†σ′ (r′ , t′ )]∓ ⟩

where ∓ indicates a commutator in the case of boson operators and an anti-commutator in the case of fermion
operators.
There are two other Green’s functions that are used in the theory of transport, the so-called greater and lesser
Green’s functions:

G> (r, σ, t; r′ , σ ′ , t′ ) = −i⟨Ψ̂σ (r, t)Ψ̂†σ′ (r′ , t′ )⟩


G< (r, σ, t; r′ , σ ′ , t′ ) = ∓i⟨Ψ̂†σ′ (r′ , t′ )Ψ̂σ (r, t)⟩

where ∓ indicates boson and fermion operators, respectively.


To see more explicitly the structure of the Green’s function, consider as an example the causal Green’s function,
assuming t > t′ :

−1
iG(r, σ, t; r′ , σ ′ , t′ ) = ⟨Ψ̂σ (r, t)Ψ̂†σ (r′ , t′ )⟩ = ZG Tr[e−β Ĥ Ψ̂σ (r, t)Ψ̂†σ (r′ , t′ )]
If we now introduce a complete set of states for the whole system {|n⟩}, then we can express the trace above as a
sum of matrix elements: X
Tr[· · · ] = ⟨n| · · · |n⟩
n

A suitable choice of states is the eigenstates of the total Hamiltonian Ĥ ′ , which will then automatically be eigenstates
of Ĥ and N̂ :
Ĥ ′ |n⟩ = (Ĥ − µN̂ )|n⟩ = En′ |n⟩ = (En + µNn )|n⟩
The expression for the Green’s function can then be expressed as
X ′ ′ ′ ′ ′ ′
−1
iG(r, σ, t; r′ , σ ′ , t′ ) = ZG e−βEn ⟨n|eiĤ t/ℏ Ψ̂σ (r)e−Ĥ (t−t )/ℏ Ψ̂†σ (r′ )e−Ĥ t /ℏ |n⟩
n

where the time dependence of the field operators has been shown explicitly. (See Jishi, pp.95-96) for an physical
interpretation of this expression for the Green’s function.)
As with the general correlation functions, if the Hamiltonian is time-independent the Green’s function depends only
on t − t′ , so we can set t′ = 0. If the Hamiltonian is also spin-independent, then we can assume that σ = σ ′ . Finally,
for a translationally invariant system the Green’s function is a function of r − r′ rather than r and r′ independently.
So, for example we can write the retarded Green’s function as

GR (r − r′ , σ, t) = −iθ(t)⟨[Ψ̂σ (r, t), Ψ̂†σ (r′ , 0)]± ⟩

We now introduce a complete set of single-particle momentum eigenstates {|k, σ⟩}, for which
1
ϕk,σ (r) = ⟨r|k, σ⟩ = √ eik·r |σ⟩
V

29
where V is the volume of the system and |σ⟩ = | ↑⟩ or | ↓⟩. We can then expand the field operators in terms of this
set of momentum eigenfunctions as follows (see Jishi p.57):
1 X −ik·r †
ϕ∗k (r)ĉ†k,σ (t) = √
X
Ψ̂†σ (r, t) = e ĉk,σ (t)
k
V k
X 1 X ik·r
Ψ̂σ (r, t) = ϕk (r)ĉk,σ (t) = √ e ĉk,σ (t)
k
V k

where the operators ĉ†k,σ and ĉk,σ create and annihilate a particle in the state |k, σ⟩.
Inserting the above into our retarded Green’s function we obtain
1 X ik·(r−r′ ) 1 X ik·(r−r′ ) R
GR (r − r′ , σ, t) = −iθ(t) e ⟨[ĉk,σ (t), ĉ†k,σ (0)]± ⟩ = e G (k, σ, t)
V V
k k

where GR (k, σ, t) is the Fourier transform of GR (r − r′ , σ, t). We can define Fourier transforms for all the different
Green’s functions:
G(k, σ, t) = −i⟨T [ĉk,σ (t)ĉ†k,σ (0)]⟩
GR (k, σ, t) = −iθ(t)⟨[ĉk,σ (t), ĉ†k,σ (0)]± ⟩
GA (k, σ, t) = +iθ(−t)⟨[ĉk,σ (t), ĉ†k,σ (0)]± ⟩
G> (k, σ, t) = −i⟨ĉk,σ (t), ĉ†k,σ (0)⟩
G< (k, σ, t) = ±i⟨ĉk† ,σ (0), ĉk,σ (t)⟩

4.2 Spectral Representation


4.2.1 Spectral Density Function
A spectral representation allows a Green’s function (or any general correlation function) to be expressed in terms of
the exact energy spectrum and eigenstates of the system. It also allows a real-time Green’s function to be obtained
from its imaginary-time counterpart.
Using the retarded Green’s function as an example, we start by expanding it as
GR (k, σ, t) = −iθ(t)⟨ĉk,σ (t)ĉ†k,σ (0)⟩ ± iθ(t)⟨ĉ†k,σ (0)ĉk,σ (t)⟩
Starting with the first term, we can write
′ ′ ′ ′ ′ ′
⟨ĉk,σ (t)ĉ†k,σ (0)⟩ = ZG t/ℏ † t/ℏ †
X
−1 −1
Tr[e−β Ĥ eiĤ t/ℏ
ĉk,σ e−iĤ ĉk,σ ] = ZG ⟨n|e−β Ĥ eiĤ t/ℏ
ĉk,σ e−iĤ ĉk,σ |n⟩
n


where {|n⟩} are the energy eigenstates P of the whole system (i.e. of the total Hamiltonian Ĥ ). To extract the time
dependence we insert the identity m |m⟩⟨m| = 1 to yield
′ ′ ′
⟨ĉk,σ (t)ĉ†k,σ (0)⟩ = ZG ⟨n|e−β Ĥ eiĤ t/ℏ ĉk,σ |m⟩⟨m|e−iĤ t/ℏ ĉ†k,σ |n⟩
X
−1

n,m
′ ′ ′
e−βEn e−i(Em −En )t/ℏ ⟨n|ĉk,σ |m⟩⟨m|ĉ†k,σ |n⟩
X
−1
= ZG
n,m
′ ′ ′
e−βEn e−i(Em −En )t/ℏ |⟨m|ĉ†k,σ |n⟩|2
X
−1
= ZG
n,m

Taking the Fourier transform we obtain the P-spectral function (which is the same as in Jishi apart from the minus
sign):
Z ∞

P (k, σ, ω ′ ) = ⟨ĉk,σ (t)ĉ†k,σ (0)⟩eiω t dt
−∞
Z ∞
′ ′ ′ ′
e−βEn |⟨m|ĉ†k,σ |n⟩|2
X
−1
= ZG ei(ω −(Em −En )/ℏ)t dt
n,m −∞

|⟨m|ĉ†k,σ |n⟩|2 δ(ω ′
X
−1 −βEn ′
= 2πZG e − (Em − En′ )/ℏ)
n,m

30
where the inverse Fourier transform, which we will use later, is given by
Z ∞
′ dω ′
⟨ĉk,σ (t)ĉ†k,σ (0)⟩ = P (k, σ, ω ′ )e−iω t
−∞ 2π

We proceed in the same way with the second term, which yields (see Jishi p.100):
′ ′ ′ ′ ′
⟨ĉ†k,σ (0)ĉk,σ (t)⟩ = ZG e−β(Em −En ) e−βEn e−i(Em −En )t/ℏ |⟨m|ĉ†k,σ |n⟩|2
X
−1

n,m

the Fourier transform of which is


Z ∞
′ ′ ′ ′
⟨ĉ†k,σ (0)ĉk,σ (t)⟩eiω t dt e−β(Em −En ) e−βEn |⟨m|ĉ†k,σ |n⟩|2 δ(ω ′ − (Em
X
−1 ′
= 2πZG − En′ )/ℏ)
−∞ n,m
−βℏω ′
= e P (k, σ, ω ′ )

where we have used the fact that the delta function is nonzero only for ω ′ = (Em ′
− En′ )/ℏ to replace (Em

− En′ ) with

ℏω .
The corresponding inverse transform is then
Z ∞
′ ′ dω ′
⟨ĉ†k,σ (0)ĉk,σ (t)⟩ = e−βℏω P (k, σ, ω ′ )e−iω t
−∞ 2π

We can now insert the two expressions for the inverse transforms above in the original expansion for the retarded
Green’s function to give Z ∞
′ ′ dω ′
R
G (k, σ, t) = −iθ(t) (1 ∓ e−βℏω )P (k, σ, ω ′ )e−iω t
−∞ 2π
The Fourier transform of GR (k, σ, t) is then given by
Z ∞ ∞ ∞
dω ′
Z Z
R R iωt −βℏω ′ ′ ′
G (k, σ, ω) = G (k, σ, t)e dt = −i (1 ∓ e )P (k, σ, ω ) ei(ω−ω )t dt
−∞ −∞ 2π 0

where we have used the fact that θ(t) = 0 for t < 0.


There is, however, a problem because the integral over t is oscillatory at t = ∞. To ensure that the integral
converges we therefore introduce a factor e−ηt so that

Z ∞ i(ω−ω ′ +iη)t
′ ′ e 1
ei(ω−ω )t dt = lim+ ei(ω−ω )t e−ηt dt = lim+ = i lim+

η→0 η→0 i(ω − ω ′ + iη) η→0 ω − ω ′ + iη
0
0

where η → 0+ below the limit indicates that η is allowed to go to zero from the positive direction.
So, we can now express GR (k, σ, ω) as follows:
∞ ′ ∞
(1 ∓ e−βℏω )P (k, σ, ω ′ ) dω ′ A(k, σ, ω ′ ) dω ′
Z Z
GR (k, σ, ω) = lim+ = lim
η→0 −∞ ω − ω ′ + iη 2π η→0+ −∞ ω − ω ′ + iη 2π

where ′
(1 ∓ e−βℏω )e−βEn |⟨m|ĉ†k,σ |n⟩|2 δ(ω − (Em
X
−1 ′
A(k, σ, ω) = 2πZG − En′ )/ℏ)
n,m

This is the so-called Lehmann spectral representation of the Green’s function, and A(k, σ, ω) is known as the spectral
density function.
It should be noted that both ω and ω ′ are real (ignoring the infinitesimal imaginary part), and that the Green’s
function has singularities (poles) at ω = ω ′ = (Em

− En′ )/ℏ. (If the volume of thePsystem is large enough the single-
particle energy spectrum can be considered to be continuous, the sum over states n,m in A(k, σ, ω) will be replaced
by an integral, and the poles will be replaced by a branch cut - see Fetter and Walecka p.78.) At T = 0, |n⟩ is
the ground state of the whole system (i.e. |n⟩ = |0⟩, En′ = 0) and |m⟩ are the excited states of the system with

corresponding energies Em . At finite temperatures, however, both |n⟩ and |m⟩ can be excited states.

31
4.2.2 Properties
The spectral density function is real, positive and satisfies the sum rule (normalization condition):
Z ∞

A(k, σ, ω) =1
−∞ 2π

We can also determine a further relation to the retarded Green’s function by using the following equality from the
theory of complex variables:
Z Z Z
f (x) f (x)
lim+ dx = P dx ∓ iπ f (x)δ(x) dx
η→0 x ± iη x

where x is real and P indicates the principal part of the integral. If we let f (x) = δ(x − a) then this simplifies to
1 1
lim = P ∓ iπδ(a)
η→0+ a ± iη a

For the retarded Green’s function we therefore have


Z ∞ Z ∞
A(k, σ, ω ′ ) dω ′ dω ′
GR (k, σ, ω) = P ′
− iπ A(k, σ, ω ′ )δ(ω − ω ′ )
−∞ ω − ω + iη 2π −∞ 2π
Z ∞
A(k, σ, ω ′ ) dω ′ i
= P ′
− A(k, σ, ω)
−∞ ω − ω + iη 2π 2

and so we arrive at the important relation

A(k, σ, ω) = −2 Im GR (k, σ, ω)

4.2.3 Example: Noninteracting System


We consider a system of noninteracting particles, for which the grand canonical Hamiltonian is given by (see Mattuck
pp.129, 342):
(εk,σ − µ)ĉ†k,σ ĉk,σ
X
Ĥ ′ = Ĥ − µN̂ =
k,σ

where εk,σ is the energy of the single-particle state |k, σ⟩. The Hamiltonian is therefore diagonal in the basis {|k, σ⟩}.
We have also used the fact that X X †
N̂ = n̂k,σ = ĉk,σ ĉk,σ
k,σ k,σ

where n̂k,σ is the number operator for the state |k, σ⟩.
The spectral density function (where the subscript zero indicates the noninteracting case) is given by

(1 ∓ e−βℏω )e−βEn |⟨m|ĉ†k,σ |n⟩|2 δ(ω − (Em
X
−1 ′
A0 (k, σ, ω) = 2πZG − En′ )/ℏ)
n,m

The matrix element in this expression is zero unless |m⟩ differs from |n⟩ by one extra particle in the state |k, σ⟩ (in
addition to the particles which may already be in this state), in which case Em − En = εk,σ and Nm = Nn + 1.
Therefore

Em − En′ = (Em − µNm ) − (En − µNn ) = Em − En − µ(Nm − Nn ) = Em − En − µ = εk,σ − µ

and we have

e−βEn ⟨n|ĉk,σ |m⟩⟨m|ĉ†k,σ |n⟩
X
−1
A0 (k, σ, ω) = 2πZG δ(ω − (εk,σ − µ)/ℏ)(1 ∓ e−βℏω )
n,m

e−βEn ⟨n|ĉk,σ ĉ†k,σ |n⟩
X
−1 −βℏω
= 2πZG δ(ω − (εk,σ − µ)/ℏ)(1 ∓ e )
n,m
P
where we have used the fact that m |m⟩⟨m| = 1.

32
−1 ′
Remembering the definition of the grand canonical ensemble average ⟨· · · ⟩ = ZG Tr(e−β Ĥ · · · ), we can express
the above as

A0 (k, σ, ω) = 2πδ(ω − (εk,σ − µ)/ℏ)(1 ∓ e−βℏω )⟨ĉk,σ ĉ†k,σ ⟩


= 2πδ(ω − (εk,σ − µ)/ℏ)(1 ∓ e−βℏω )⟨1 ± ĉ†k,σ ĉk,σ ⟩

where we have used the commutation relation [ĉk,σ , ĉ†k,σ ]∓ = 1, where ∓ refers to bosons/fermions.
The reason for making the above substitution becomes clear when we recall that ĉ†k,σ ĉk,σ = n̂k,σ , the number
operator for the state |k, σ⟩. Similarly, ⟨ĉ†k,σ ĉk,σ ⟩ = ⟨n̂k,σ ⟩, the average occupation number of the state |k, σ⟩ over the
ensemble, which is given by the Fermi-Dirac/Bose-Einstein distribution function
 −1
fk,σ = eβ(εk,σ −µ) ∓ 1

where ∓ refers to bosons/fermions. Hence we can write the spectral density function as

A0 (k, σ, ω) = 2πδ(ω − (εk,σ − µ)/ℏ)(1 ∓ e−βℏω )(1 ± fk,σ )

To simplify this, we note that the effect of the delta function is to replace ω with (εk,σ − µ)/ℏ, and after some
algebra we obtain the final form of the spectral density function for a system of noninteracting particles:

A0 (k, σ, ω) = 2πδ(ω − (εk,σ − µ)/ℏ)

Inserting this into the expression for GR (k, σ, ω) gives the noninteracting retarded Green’s function
1
GR
0 (k, σ, ω) =
ω − (εk,σ − µ)/ℏ) + i0+

where 0+ is shorthand for limη→0+ .


The spectral density function is singularly peaked at ω = εk,σ − µ, i.e. at the single-particle energy (measured
relative to the chemical potential) of the state |k, σ⟩. This is because, in the noninteracting case, the state ĉ†k,σ |n⟩
obtained by adding a single particle in |k, σ⟩ to the many-body state |n⟩ is also an energy eigenstate of the whole
system, and is therefore orthogonal to all other states except itself (i.e. for |m⟩ = ĉ†k,σ |n⟩). We say that the ‘spectral
weight’ of the state ĉ†k,σ |n⟩ is concentrated on a single eigenstate of the system.
However, in the interacting case, the state ĉ†k,σ |n⟩ will in general no longer be an energy eigenstate of the system,
so there is no reason to believe that it will be orthogonal to all states |m⟩ except one. The spectral weight of the state
ĉ†k,σ |n⟩ is then distributed over many (potentially a continuum) of states |m⟩.

4.2.4 Real-Time Correlation Functions


We now seek to determine the spectral representation for the more general case of the time-ordered (causal) correlation
function of any two operators  and B̂:
T
CAB (t) = −i⟨T [Â(t)B̂(0)]⟩ = −iθ(t)⟨Â(t)B̂(0)⟩ ∓ iθ(−t)⟨B̂(0)Â(t)⟩

Following the same procedure as before, we write the trace explicitly and extract the time dependence:
′ ′ ′ ′ ′ ′
−1 −1
T
CAB (t) = −iθ(t)ZG Tr[e−β Ĥ eiĤ t/ℏ Âe−iĤ t/ℏ B̂] ∓ iθ(−t)ZG Tr[e−β Ĥ eiĤ t/ℏ
B̂e−iĤ t/ℏ
Â]
X ′ ′ ′
−1
= −iθ(t)ZG e−βEn e−i(Em −En )t/ℏ ⟨n|Â|m⟩⟨m|B̂|n⟩
n,m
X ′ ′ ′
−1
∓ iθ(−t)ZG e−βEn e−i(En −Em )t/ℏ ⟨n|B̂|m⟩⟨m|Â|n⟩
n,m

Adjusting the second term by relabelling indices n ↔ m and reordering the matrix elements yields
X ′ ′
h ′ ′
i
−1
T
CAB (t) = −iZG ⟨n|Â|m⟩⟨m|B̂|n⟩e−i(Em −En )t/ℏ θ(t)e−βEn ± θ(−t)e−βEm
n,m

33
We now take the Fourier transform:
Z ∞
T T
CAB (ω) = CAB (t)eiωt dt
−∞
 Z ∞ Z 0 
X ′ ′ ′ ′ ′ ′
−1
= −iZG ⟨n|Â|m⟩⟨m|B̂|n⟩ e−βEn ei(ω−(Em −En )/ℏ)t dt ± e−βEm ei(ω−(Em −En )/ℏ)t dt
n,m 0 −∞

The first integral is evaluated as before, while for the second integral we obtain
Z 0 Z 0
′ ′ ′ ′ −i
ei(ω−(Em −En )/ℏ)t dt = lim ei(ω−(Em −En )/ℏ−iη)t dt = ′ − E ′ )/ℏ − i0+
−∞ η→0 +
−∞ ω − (Em n

We then obtain the final form


′ ′
!
T −1
X e−βEn e−βEm
CAB (ω) = ZG ⟨n|A|m⟩⟨m|B|n⟩ ′ ′

n,m
ω − (Em − En )/ℏ + i0+ ω − (Em − En′ )/ℏ − i0+

This is the Lehmann spectral representation of the time-ordered (causal) correlation function. Similar expressions
can be derived for the retarded and advanced correlation functions:
′ ′
R −1
X e−βEn ∓ e−βEm
CAB (ω) = ZG ⟨n|A|m⟩⟨m|B|n⟩ ′ − E ′ )/ℏ + i0+
n,m
ω − (Em n
′ ′
A −1
X e−βEn ∓ e−βEm
CAB (ω) = ZG ⟨n|A|m⟩⟨m|B|n⟩ ′ − E ′ )/ℏ − i0+
n,m
ω − (Em n

Making use of the identity from earlier (for real x):


1 1
= P ∓ iπδ(x)
x ± i0+ x
we can rewrite the retarded and advanced correlation functions as
′ ′
!
−1
X e−βEn ∓ e−βEm ′ ′
R
CAB (ω) = ZG ⟨n|A|m⟩⟨m|B|n⟩ P ′ ′
− iπ(e−βEn ∓ e−βEm )δ(ω − (Em

− En′ )/ℏ)
n,m
ω − (E m − En )/ℏ
′ ′
!
−1
X e−βEn ∓ e−βEm ′ ′
A
CAB (ω) = ZG ⟨n|A|m⟩⟨m|B|n⟩ P ′ − E ′ )/ℏ
+ iπ(e−βEn ∓ e−βEm )δ(ω − (Em

− En′ )/ℏ)
n,m
ω − (E m n

while the time-ordered (causal) correlation function becomes


( ′ ′
T −1
X e−βEn e−βEn
CAB (ω) = ZG ⟨n|A|m⟩⟨m|B|n⟩ P ′ − E ′ )/ℏ
∓ P ′ − E ′ )/ℏ
n,m
ω − (Em n ω − (Em n
)
′ ′
−βEn ′
−iπe δ(ω − (Em − En′ )/ℏ) + iπe −βEm
δ(ω − ′
(Em − En′ )/ℏ)
( ′ ′
)
−1
X e−βEn ∓ e−βEm ′ ′
= ZG ⟨n|A|m⟩⟨m|B|n⟩ P ′ − E ′ )/ℏ
− iπ(e−βEn ± e−βEm )δ(ω − (Em

− En′ )/ℏ)
n,m
ω − (E m n

If we now define the spectral density function as


X ′
−1
A(ω) = 2πZG ⟨n|A|m⟩⟨m|B|n⟩e−βEn (1 ∓ e−βℏω )δ(ω − (Em

− En′ )/ℏ)
n,m

then this allows us to express the advanced and retarded correlation functions as follows:
Z ∞ Z ∞
R A(ω ′ ) dω ′ A(ω ′ ) dω ′ i
CAB (ω) = ′ +
= P ′
− A(ω)
−∞ ω − ω + i0 2π −∞ ω − ω 2π 2
Z ∞ ′ ′ Z ∞ ′ ′
A A(ω ) dω A(ω ) dω i
CAB (ω) = ′ +
=P ′
+ A(ω)
−∞ ω − ω − i0 2π −∞ ω − ω 2π 2

34
and the time-ordered correlation function as
Z ∞
T A(ω ′ ) dω ′ i 1 ± e−βℏω
CAB (ω) = P ′
− A(ω)
−∞ ω − ω 2π 2 1 ∓ e−βℏω
Z ∞
A(ω ′ ) dω ′ i ∓1
=P ′ 2π
− tanh( 12 βℏω) A(ω)
−∞ ω − ω 2
−1 R −1 A
= 1 ∓ e−βℏω CAB (ω) + 1 ∓ eβℏω CAB (ω)

where we have used the fact that


1 1
1 1 (1 ∓ eβℏω ) − (1 ∓ e−βℏω ) (e 2 βℏω )2 − (e− 2 βℏω )2
− = = 1 1
1 ∓ e−βℏω 1 ∓ eβℏω (1 ∓ e−βℏω )(1 ∓ eβℏω ) (e 2 βℏω )2 + (e− 2 βℏω )2 ∓ 2
1 1 1 1 1 1
(e 2 βℏω + e− 2 βℏω )(e 2 βℏω − e− 2 βℏω ) e 2 βℏω ± e− 2 βℏω ∓1
= tanh( 21 βℏω)

= 1 1 = 1 1
(e 2 βℏω ∓ e− 2 βℏω )2 e 2 βℏω ∓ e− 2 βℏω
From the above we can see that for the real parts

A(ω ′ ) dω ′
Z
R A T
Re CAB (ω) = Re CAB (ω) = Re CAB (ω) = P
−∞ ω − ω ′ 2π

and for the imaginary parts


R
Im CAB (ω) = − 12 A(ω)
A
Im CAB (ω) = + 12 A(ω)
T 1 ∓1
Im CAB (ω) = − tanh( 21 βℏω) A(ω)
2
and by eliminating A(ω) these are related via
∞ T
(ω ′ ) dω ′
Z
±1 Im CAB
T
tanh( 21 βℏω ′ )

Re CAB (ω) = −P
−∞ ω − ω′ π

4.2.5 Generalization to a Complex Variable


In the preceding sections we have shown that the spectral representation of real-time correlation functions can be
obtained in two different forms: as an integral in terms of a spectral density function, or as a simple fraction. For
example, the retarded and advanced real-time Green’s functions can be expressed as
Z ∞
R A(k, σ, ω ′ ) dω ′
G (k, σ, z) = ′ + 2π
−∞ ω − ω + i0
Z ∞
A(k, σ, ω ) dω ′

GA (k, σ, z) = ′ + 2π
−∞ ω − ω − i0

and the more general retarded and advanced real-time correlation functions can be expressed as
′ ′
R −1
X e−βEn ∓ e−βEm
CAB (ω) = ZG ⟨n|A|m⟩⟨m|B|n⟩ ′ − E ′ )/ℏ + i0+
n,m
ω − (Em n
′ ′
A −1
X e−βEn ∓ e−βEm
CAB (ω) = ZG ⟨n|A|m⟩⟨m|B|n⟩ ′ − E ′ )/ℏ − i0+
n,m
ω − (Em n

with the two forms of expression being interchangeable.


These correlation functions are treated as functions of a real variable, ω. For reasons which will become apparent,
it is useful to generalize them to functions of a complex variable, z, so that the above treatment then applies to the
case z = ω, with ω restricted to the real axis. The retarded Green’s/correlation function then has singularities for
z = ω ′ = (Em ′
− En′ )/ℏ − i0+ slightly below the real axis, but is analytic in the entire upper complex half-plane (i.e.
for Im (z) ≥ 0), while the advanced Green’s/correlation function has singularities for z = ω ′ = (Em ′
− En′ )/ℏ + i0+
slightly above the real axis, but is analytic in the entire lower complex half-plane (i.e. for Im (z) ≤ 0). (The causal

35
Green’s/correlation function, by comparison, has singularities on either side of the real axis, which makes it harder to
analyze.)
If we now introduce a complex function
Z ∞
A(k, σ, ω ′ ) dω ′
Γ(k, σ, z) =
−∞ z − ω ′ 2π

which is analytic everywhere except for the real axis, then GR (k, σ, ω) and GA (k, σ, ω) can be considered to be limiting
cases of Γ(k, σ, z) as the complex variable z approaches the real axis (aside from an infinitesimal imaginary part) from
above and below, respectively , i.e.

GR (k, σ, ω) = Γ(k, σ, ω + i0+ ), GA (k, σ, ω) = Γ(k, σ, ω − i0+ )

Similarly, if we introduce a complex function


′ ′
−1
X e−βEn ∓ e−βEm
ΓAB (z) = ZG ⟨n|A|m⟩⟨m|B|n⟩ ′ − E ′ )/ℏ
n,m
z − (Em n

then C R (ω) and C A (ω) can be considered to be limiting cases of ΓAB (z) as z approaches the real axis from above and
below, respectively, i.e.
R
CAB (ω) = ΓAB (ω + i0+ ), A
CAB (ω) = ΓAB (ω − i0+ )

5 Imaginary-Time Correlation Functions


If we consider, for example, the real-time correlation function
′ ′ ′
−1
CAB (t) = ⟨Â(t)B̂(0)⟩ = ZG Tr[e−β Ĥ eiĤ t/ℏ
Âe−iĤ t/ℏ
B̂]

then to carry out a perturbation expansion would be very difficult due to the mismatch in the exponents, i.e. −β Ĥ ′ is
real, while ±iĤ ′ /ℏ is imaginary. However, if we replace it with τ (τ ∈ R), which is equivalent to introducing imaginary
time t = −iτ , then a perturbation expansion becomes possible. There is also a simple procedure for obtaining the
real-time correlation function from its imaginary-time counterpart.
We define the imaginary-time causal, or time-ordered, correlation function
T
CAB (τ, τ ′ ) = −⟨T [Â(τ )B̂(τ ′ )]⟩

where the Heisenberg (time-dependent) operators Â(t) and B̂(t) can be expressed in terms of boson or fermion creation
and annihilation operators, ĉ† and ĉ, and are related to the corresponding Schrödinger (time-independent) operators
via ′ ′ ′ ′
Â(τ ) = eĤ τ /ℏ Âe−Ĥ τ /ℏ , B̂(τ ′ ) = eĤ τ /ℏ B̂e−Ĥ τ /ℏ
As in the real-time case, for a time-independent Hamiltonian the imaginary-time correlation function will depend on
τ − τ ′ , and we are free to set τ ′ = 0 so that we may write
T
CAB (τ ) = −⟨T [Â(τ )B̂(0)]⟩

5.1 Periodicity
Consider now the case of τ > 0, so that
′ ′ ′
−1
T
CAB (τ ) = −⟨Â(τ )B̂(0)⟩ = −ZG Tr[e−β Ĥ eĤ τ /ℏ
Âe−Ĥ τ /ℏ
B̂]

Choosing as a basis set the eigenfunctions of Ĥ ′ , denoted by {|n⟩}, we can expand the trace to give
X ′ ′ ′
−1
T
CAB (τ ) = −ZG ⟨n|e−β Ĥ eĤ τ /ℏ Âe−Ĥ τ /ℏ B̂|n⟩
n
P
If we then extract the time dependence by inserting the identity m |m⟩⟨m| and combine the exponentials we obtain
X ′ ′
−1
T
CAB (τ ) = −ZG e(−βℏ+τ )En /ℏ ⟨n|Â|m⟩e−τ Em /ℏ ⟨m|B̂|n⟩
n,m

36
Since the eigenvalues En′ and Em ′
can be arbitrarily large positive quantities, we can only guarantee that the sum
will be finite if the exponents are negative. For the first exponent we have

(−βℏ + τ )En′ < 0 =⇒ −βℏ + τ < 0 =⇒ τ < βℏ

and for the second exponent



−τ Em < 0 =⇒ −τ < 0 =⇒ τ > 0
so the combined condition is that 0 < τ < βℏ. A similar procedure for the case of τ < 0 yields the condition
−βℏ < τ < 0, so the overall result is that τ is limited to the interval [−βℏ, βℏ].
Consider again the correlation function for τ > 0:
′ ′ ′
−1
T
CAB (τ ) = −⟨Â(τ )B̂(0)⟩ = −ZG Tr[e−β Ĥ eĤ τ /ℏ
Âe−Ĥ τ /ℏ
B̂]
′ ′
If we perform the following three steps: (1) move B̂ to the far right, (2) introduce 1 = eβ Ĥ e−β Ĥ at the far right, and

(3) move e−β Ĥ from the far right to the far left, then we are left with
′ ′ ′
−1
T
CAB (τ ) = −ZG Tr[e−β Ĥ B̂eĤ (τ −βℏ)/ℏ
Âe−Ĥ (τ −βℏ)/ℏ
]

−1
= −ZG Tr[e−β Ĥ B̂(0)Â(τ − βℏ)]
= −⟨B̂(0)Â(τ − βℏ)⟩

So, using our earlier result that τ ∈ [−βℏ, βℏ], if τ > 0 then τ − βℏ will be negative.
Now consider the correlation function
T
CAB (τ − βℏ) = −⟨T [Â(τ − βℏ)B̂(0)]⟩
T
This is evidently related to CAB (τ ) above, and to see how we note that for τ − βℏ < 0 the time-ordered product
becomes
T [Â(τ − βℏ)B̂(0)] = ±B̂(0)Â(τ − βℏ)
where ± indicates boson/fermion operators, so the correlation function becomes
T
CAB (τ − βℏ) = ∓⟨B̂(0)Â(τ − βℏ)B̂(0)⟩
T
Comparing this with the original correlation function, CAB (τ ), we find that
T
CAB (τ ) = ±cTAB (τ − βℏ)

Similarly, it can be shown that for τ < 0


T
CAB (τ ) = ±cTAB (τ + βℏ)
T
Thus, CAB (τ ) is periodic in τ , with period βℏ, in the range [−βℏ, βℏ]. We can therefore expand it in a Fourier series:

1 X T
T
CAB (τ ) = C (iωn )e−iωn τ
βℏ n=−∞ AB

where the ωn are known as the Matsubara frequencies.


T T
The fact that CAB (τ ) = ±CAB (τ − βℏ) implies that

e−iωn τ = ±e−iωn (τ −βℏ) =⇒ e−iωn βℏ = ±1

where ± indicates bosons/fermions. Therefore


 2nπ
 , n∈Z (bosons)
βℏ

ωn = (2n + 1)π

 , n∈Z (fermions)
βℏ
T
The Fourier coefficients CAB (iωn ), for both bosons and fermions, are given by

1 βℏ T
Z
T
CAB (iωn ) = C (τ )eiωn τ dτ
2 −βℏ AB

37
It is convenient to change the limits of integration by separating the integral into two parts:
"Z #
0 Z βℏ
T 1
CAB (iωn ) = C T (τ )eiωn τ dτ + T
CAB (τ )eiωn τ dτ
2 −βℏ AB 0

T T
Using the fact that CAB (τ ) = ±CAB (τ + βℏ) for τ < 0 and eiωn βℏ = ±1, followed by a change of variables τ → τ − βℏ,
the first term can be rewritten as
Z 0 Z βℏ
T
CAB (τ + βℏ)eiωn (τ +βℏ) dτ = T
CAB (τ )eiωn τ dτ
−βℏ 0

Therefore, we have for the Fourier coefficients


Z βℏ
T T
CAB (iωn ) = CAB (τ )eiωn τ dτ
0

5.2 Imaginary-Time Green’s Function


A special case of the imaginary-time correlation function is the imaginary-time Green’s function, defined as

G(r, σ, τ ; r′ , σ ′ , τ ′ ) = −⟨T [Ψ̂σ (r, τ )Ψ̂†σ′ (r′ , τ ′ )]⟩

where the τ -dependent field operators are given by


′ ′ ′ ′
Ψ̂†σ (r, τ ) = eĤ τ /ℏ
Ψ̂†σ (r)e−Ĥ τ /ℏ
, Ψ̂σ (r, τ ) = eĤ τ /ℏ
Ψ̂σ (r)e−Ĥ τ /ℏ

It is important to note that Ψ̂†σ (r) is not the adjoint of Ψ̂σ (r). As in the real-time case, we assume that Ĥ ′ is time-
and spin-independent and that the system is translationally invariant, so that the Green’s function can be written as
G(r − r′ , σ, τ ), where we have set τ ′ = 0.
Expanding the field operators in terms of particle creation and annihilation operators:
1 X −ik·r † 1 X ik·r
Ψ̂†σ (r, τ ) = √ e ĉk,σ (τ ), Ψ̂σ (r, τ ) = √ e ĉk,σ (τ )
V k V k

where V is the volume of the system, yields the following form of the Green’s function:
1 X ik·(r−r′ ) 1 X ik·(r−r′ )
G(r − r′ , σ, τ ) = − e ⟨T [ĉk,σ (τ )ĉ†k,σ (0)]⟩ = e G(k, σ, τ )
V V
k k


where G(k, σ, τ ) is the spatial Fourier transform of G(r − r , σ, τ ), given by

G(k, σ, τ ) = −⟨T [ĉk,σ (τ )ĉ†k,σ (0)]⟩

5.3 Spectral Representation


5.3.1 Imaginary-Time Green’s Function
We proceed as with the real-time causal Green’s function, except that due to the periodicity of τ we only need to
consider the range 0 ≤ τ ≤ βℏ. This means that we can drop the time-ordering and express G(k, σ, τ ) as
′ ′ ′
G(k, σ, τ ) = −⟨ĉk,σ (τ )ĉ†k,σ (0)⟩ = −ZG
−1
Tr[e−β Ĥ eĤ τ /ℏ ĉk,σ e−Ĥ τ /ℏ ĉ†k,σ ]
′ ′ ′
⟨n|ĉk,σ |m⟩⟨m|ĉ†k,σ |n⟩e−βEn e−(Em −En )τ /ℏ
X
−1
= −ZG
n,m
Z ∞

=− P (k, σ, ω)e−ωτ
−∞ 2π

where P (k, σ, ω) is the P-spectral function, which was defined earlier as



e−βEn |⟨m|ĉ†k,σ |n⟩|2 δ(ω − (Em
X
−1 ′
P (k, σ, ω) = 2πZG − En′ )/ℏ)
n,m

38
Using the expression for the Fourier coefficients from the previous section we have
Z βℏ Z ∞ Z βℏ

G(k, σ, iωn ) = G(k, σ, τ )eiωn τ dτ = − P (k, σ, ω) e(iωn −ω)τ dτ
0 −∞ 2π 0
∞ (iωn −ω)τ
βℏ ∞
(1 ∓ e−βℏω )P (k, σ, ω) dω
Z  Z
dω e
=− P (k, σ, ω) =
−∞ 2π iωn − ω 0 −∞ iωn − ω 2π

where we have made use of the fact that eiωn βℏ = ±1 for bosons/fermions. This can be simplified further if we recall
that ′
(1 ∓ e−βℏω )e−βEn |⟨m|ĉ†k,σ |n⟩|2 δ(ω − (Em
X
−1
A(k, σ, ω) = (1 ∓ e−βℏω )P (k, σ, ω) = 2πZG ′
− En′ )/ℏ)
n,m

which yields the spectral representation of the imaginary-time Green’s function:


Z ∞
A(k, σ, ω) dω
G(k, σ, iωn ) =
−∞ iωn − ω 2π


So, the imaginary-time Green’s function has singularities (poles) at the points iωn = ω = (Em − En′ )/ℏ along the
imaginary axis.

5.3.2 Imaginary-Time Correlation Function


We proceed as with the real-time causal correlation function, except that due to the periodicity of τ we only need to
T
consider the range 0 ≤ τ ≤ βℏ. This means that we can drop the time-ordering and express CAB = CAB as
′ ′ ′
−1
CAB (τ ) = −⟨Â(τ )B̂(0)⟩ = −ZG Tr[e−β Ĥ eĤ τ /ℏ Âe−Ĥ τ /ℏ B̂]
X ′ ′ ′
−1
= −ZG ⟨n|Â|m⟩⟨m|B̂|n⟩e−βEn e−(Em −En )τ /ℏ
n,m

Using the expression for the Fourier coefficients from the previous section we have
Z βℏ
CAB (iωn ) = CAB (τ )eiωn τ dτ
0
Z βℏ
X ′ ′ ′
−1 −βEn
= −ZG ⟨n|Â|m⟩⟨m|B̂|n⟩e e[iωn −(Em −En )/ℏ]τ dτ
n,m 0
X ′
h ′ ′
iβℏ
−1
= −ZG ⟨n|Â|m⟩⟨m|B̂|n⟩e−βEn e[iωn −(Em −En )/ℏ]τ
0
n,m

and noting the fact that eiωn βℏ = ±1 for bosons/fermions gives


′ ′
−1
X e−βEn ∓ e−βEm
CAB (iωn ) = ZG ⟨n|Â|m⟩⟨m|B̂|n⟩ ′ − E ′ )/ℏ
n,m
iωn − (Em n

which is the spectral representation of the imaginary-time causal correlation function.

5.3.3 Analytic Continuation


As in the case of the real-time Green’s/correlation function, we can express their imaginary-time counterparts as
limiting cases of complex functions. Recalling the function from earlier:
Z ∞
A(k, σ, ω ′ ) dω ′
Γ(k, σ, z) =
−∞ z − ω ′ 2π

which is analytic everywhere except for the real axis, the imaginary-time Green’s function can be considered as the
limiting case
G(k, σ, iωn ) = Γ(k, σ, iωn )

39
Similarly, for the complex function in the alternative representation
′ ′
−1
X e−βEn ∓ e−βEm
ΓAB (z) = ZG ⟨n|A|m⟩⟨m|B|n⟩ ′ − E ′ )/ℏ
n,m
z − (Em n

the imaginary-time correlation function can be considered as the limiting case

CAB (iωn ) = ΓAB (iωn )

As mentioned at the beginning of this section, a perturbation expansion can be carried out for an imaginary-time
correlation function where this may be impossible for the corresponding real-time correlation function. It is then
necessary to somehow relate the imaginary-time function to its real-time counterpart, and the complex function above
suggests a way to do this, as both the imaginary-time and real-time correlation functions are limiting cases of this
more general function.
For example, if we evaluate G(k, σ, iωn ) for all positive Matsubara frequencies ωn , we then know Γ(k, σ, z) at the
discrete set of points {iωn } on the positive imaginary axis. If the function Γ(k, σ, z) can be extended so that it is
analytic in the entire upper complex half-plane (Im z > 0), a process known as analytic continuation, then this region
will also include the limiting case when Γ(k, σ, z) coincides with the real-time retarded correlation function, which is
also analytic in this region, i.e. Γ(k, σ, ω + i0+ ) = GR (k, σ, ω). We are then justified in making the simple substitution
ωn → ω + i0+ in the imaginary-time correlation function to obtain its real-time counterpart.

6 Perturbation Theory
6.1 Preliminaries
For an interacting system of particles with Heisenberg-picture ground state |Ψ0 ⟩, we can define the single-particle
Green’s function (quasiparticle propagator) via

⟨Ψ0 |T [Ψ̂H (x, t)Ψ̂†H (x′ , t′ )]|Ψ0 ⟩


iG(x, t; x′ , t′ ) ≡
⟨Ψ0 |Ψ0 ⟩

For a Hamiltonian that can be expressed as the sum of two terms, i.e. Ĥ = Ĥ0 +Ĥ ′ , we can introduce the interaction
picture, in which states and operators are given in terms of their corresponding Schrödinger-picture counterparts as
follows:
|ΨI (t)⟩ ≡ eiĤ0 t |ΨS (t)⟩ , ÔI (t) ≡ eiĤ0 t ÔS e−iĤ0 t
We can also define a unitary time-evolution operator in the interaction picture:
∞  n
1 t
Z Z t
X i
Û (t, t0 ) ≡ − dt1 · · · dtn T [Ĥ ′ (t1 ) · · · Ĥ ′ (tn )]
n=0
ℏ n! t 0 t 0

such that
|ΨI (t)⟩ = Û (t, t0 ) |ΨI (t0 )⟩
and we also define
Ŝ ≡ Û (∞, −∞)
We now assume that the interaction between particles is switched on very slowly (adiabatically), so that it starts
from zero at t0 = −∞ and reaches its full strength at t = 0. Let |Φ0 ⟩ be the ground state of the system before the
interaction is switched on, i.e. the non-interacting ground state (which is an eigenstate of Ĥ0 ). Now gradually increase
the interaction until it reaches full strength, so that the system becomes fully interacting. We choose to define the
Heisenberg-picture interacting ground state |Ψ0 ⟩ (which is an eigenstate of the full Hamiltonian Ĥ) as equivalent to
the interaction-picture ground state at time t = 0, which in turn is related to the non-interacting ground state |Φ0 ⟩ at
t0 = −∞ as follows:
|Ψ0 ⟩ ≡ |ΨI (0)⟩ = Û (0, −∞) |ΨI (−∞)⟩ = Û (0, −∞) |Φ0 ⟩
The Heisenberg-picture state |Ψ0 ⟩ then does not change with time, while the interaction-picture state continues to
evolve for t > 0.

40
To evaluate the interacting Green’s function for physical systems, we use perturbation theory and expand in powers
of the interaction. Using the Gell-Mann-Low theorem, we can show (see Fetter and Walecka, pp. 83-85) that we can
write the Green’s function for the interacting system as the following perturbation series:
∞  n
1 ∞ ⟨Φ0 |T [Ĥ ′ (t1 ) · · · Ĥ ′ (tn )Ψ̂I (x, t)Ψ̂†I (x′ , t′ )]|Φ0 ⟩
Z Z ∞
X i
iG(x, t; x′ , t′ ) = − dt1 · · · dtn
n=0
ℏ n! −∞ −∞ ⟨Φ0 |Ŝ|Φ0 ⟩

Using the definition of a second-quantized operator, Ĥ ′ takes the following form:


1 ∞
Z Z ∞
Ĥ ′ = dx1 dx′1 Ψ̂†I (x1 )Ψ̂†I (x′1 )V (x1 , x′1 )Ψ̂I (x′1 )Ψ̂I (x1 )
2 −∞ −∞

where V (x1 , x′1 ) is the matrix element of the two-body interaction potential energy.
The zero-order (n = 0) contribution to the perturbation series is given by (dropping the subscript ‘I’ on operators
from this point on for clarity):
iG(0) (x, t; x′ , t′ ) = ⟨Φ0 |T [Ψ̂(x, t)Ψ̂† (x′ , t′ )]|Φ0 ⟩
or, in terms of the four-vector notation x ≡ (x, t), y ≡ (x′ , t′ ):

iG(0) (x, y) = ⟨Φ0 |T [Ψ̂(x)Ψ̂† (y)]|Φ0 ⟩

For the first-order (n = 0) contribution, we consider for now just the numerator of the above expression (indicated
by a tilde), and obtain
i 1 ∞
  Z
iG̃(1) (x, t; x′ , t′ ) = − dt1 dx1 dx′1 V (x1 , x′1 )
ℏ 2 −∞
× ⟨Φ0 |T [Ψ̂† (x1 , t1 )Ψ̂† (x′1 , t1 )Ψ̂(x′1 , t1 )Ψ̂(x1 , t1 )Ψ̂(x, t)Ψ̂† (x′ , t′ )]|Φ0 ⟩

To express the above in four-vector notation, we first introduce an extra time variable, t′1 , via a delta function:
i 1 ∞
  Z
′ ′
(1)
iG̃ (x, t; x , t ) = − dt1 dt′1 dx1 dx′1 V (x1 , x′1 )δ(t1 − t′1 )
ℏ 2 −∞
× ⟨Φ0 |T [Ψ̂† (x1 , t1 )Ψ̂† (x′1 , t′1 )Ψ̂(x′1 , t′1 )Ψ̂(x1 , t1 )Ψ̂(x, t)Ψ̂† (x′ , t′ )]|Φ0 ⟩

and then define


U (x1 , x′1 ) ≡ V (x1 , x′1 )δ(t1 − t′1 )
so that we can write:
i 1 ∞ 4
  Z
iG̃(1) (x, y) = − d x1 d4 x′1 U (x1 , x′1 ) ⟨Φ0 |T [Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y)]|Φ0 ⟩
ℏ 2 −∞

6.2 Wick’s Theorem


To proceed further, we need to be able to evaluate time-ordered expectation values of interaction-picture operators,
such as
⟨Φ0 |T [Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y)]|Φ0 ⟩
We begin by defining the contraction of interaction-picture operators, Â and B̂, via

ÂB̂ ≡ ⟨Φ0 |T [ÂB̂]|Φ0 ⟩

We can then apply Wick’s theorem, which says that a time-ordered expectation value is equal to the sum over all
possible contracted pairs of operators. As an example, consider

⟨Φ0 |T [ÂB̂ Ĉ D̂]|Φ0 ⟩ = ÂB̂ Ĉ D̂ + ÂB̂ Ĉ D̂ + ÂB̂ Ĉ D̂

We now need to rearrange the operators so that contracted pairs are adjacent to one another. For fermion operators,
changing the order of two operators introduces a factor of −1, i.e.

ÂB̂ Ĉ D̂ = −ÂĈ B̂ D̂

41
which applied to the above yields

⟨Φ0 |T [ÂB̂ Ĉ D̂]|Φ0 ⟩ = ÂB̂ Ĉ D̂ − ÂĈ B̂ D̂ + ÂD̂B̂ Ĉ


= ⟨Φ0 |T [ÂB̂]|Φ0 ⟩ ⟨Φ0 |T [Ĉ D̂]|Φ0 ⟩ − ⟨Φ0 |T [ÂĈ]|Φ0 ⟩ ⟨Φ0 |T [B̂ D̂]|Φ0 ⟩
+ ⟨Φ0 |T [ÂD̂]|Φ0 ⟩ ⟨Φ0 |T [B̂ Ĉ]|Φ0 ⟩
If we now apply this method to the time-ordered expectation value in the expression for iG̃(1) (x, y), we obtain
⟨Φ0 |T [Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y)]|Φ0 ⟩ =

Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y) + Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y)

+ Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y) + Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y)

+ Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y) + Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y)
and rearranging the operators yields
⟨Φ0 |T [Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y)]|Φ0 ⟩ =

Ψ̂† (x1 )Ψ̂(x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x)Ψ̂† (y) − Ψ̂† (x1 )Ψ̂(x′1 )Ψ̂† (x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y)

+ Ψ̂† (x1 )Ψ̂(x)Ψ̂† (x′1 )Ψ̂(x1 )Ψ̂(x′1 )Ψ̂† (y) − Ψ̂† (x1 )Ψ̂(x)Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂† (y)

+ Ψ̂† (x1 )Ψ̂(x′1 )Ψ̂† (x′1 )Ψ̂(x)Ψ̂(x1 )Ψ̂† (y) − Ψ̂† (x1 )Ψ̂(x1 )Ψ̂† (x′1 )Ψ̂(x)Ψ̂(x′1 )Ψ̂† (y)
which, after evaluating the contractions, gives
⟨Φ0 |T [Ψ̂† (x1 )Ψ̂† (x′1 )Ψ̂(x′1 )Ψ̂(x1 )Ψ̂(x)Ψ̂† (y)]|Φ0 ⟩ =
⟨Φ0 |T [Ψ̂(x1 )Ψ̂† (x1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x′1 )Ψ̂† (x′1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x)Ψ̂† (y)]|Φ0 ⟩
− ⟨Φ0 |T [Ψ̂(x′1 )Ψ̂† (x1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x1 )Ψ̂† (x′1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x)Ψ̂† (y)]|Φ0 ⟩
+ ⟨Φ0 |T [Ψ̂(x)Ψ̂† (x1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x1 )Ψ̂† (x′1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x′1 )Ψ̂† (y)]|Φ0 ⟩
− ⟨Φ0 |T [Ψ̂(x)Ψ̂† (x1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x′1 )Ψ̂† (x′1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x1 )Ψ̂† (y)]|Φ0 ⟩
+ ⟨Φ0 |T [Ψ̂(x′1 )Ψ̂† (x1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x)Ψ̂† (x′1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x1 )Ψ̂† (y)]|Φ0 ⟩
− ⟨Φ0 |T [Ψ̂(x1 )Ψ̂† (x1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x)Ψ̂† (x′1 )]|Φ0 ⟩ ⟨Φ0 |T [Ψ̂(x′1 )Ψ̂† (y)]|Φ0 ⟩
Finally, substituting into the expression for iG̃(1) (x, y) and replacing the time-ordered expectation values on the RHS
with the corresponding zero-order Green’s functions yields
i 1 ∞ 4
  Z
iG̃(1) (x, y) = − d x1 d4 x′1 U (x1 , x′1 )
ℏ 2 −∞
n h i
× iG(0) (x, y) iG(0) (x′1 , x′1 )iG(0) (x1 , x1 ) − iG(0) (x′1 , x1 )iG(0) (x1 , x′1 )
h i
+ iG(0) (x, x1 ) iG(0) (x1 , x′1 )iG(0) (x′1 , y) − iG(0) (x1 , y)iG(0) (x′1 , x′1 )
h io
+ iG(0) (x, x′1 ) iG(0) (x′1 , x1 )iG(0) (x1 , y) − iG(0) (x1 , y)iG(0) (x1 , x1 )

The question remains, how to interpret iG(0) (x1 , x1 ) and iG(0) (x′1 , x′1 ), since the time-ordered product is undefined
at equal times? The answer is that since these terms arise simply from contractions of two field operators within Ĥ ′ ,
in which Ψ̂† always stands to the left of Ψ̂, we can in some sense consider Ψ̂† as being applied ‘after’ Ψ̂, and hence
can assign it an infinitesimally later time. We then define the equal-time Green’s function as the limit
iG(0) (x1 , x1 ) = lim+ ⟨Φ0 |T [Ψ̂(x1 , t1 )Ψ̂† (x1 , t1 + δ)]|Φ0 ⟩
δ→0

= − lim+ ⟨Φ0 |Ψ̂† (x1 , t1 + δ)Ψ̂(x1 , t1 )|Φ0 ⟩ = − ⟨Φ0 |Ψ̂† (x1 )Ψ̂(x1 )|Φ0 ⟩
δ→0

42
6.3 Diagrammatic Analysis
Returning to the expression for iG̃(1) (x, y) above, if we rewrite it as

i 1 ∞ 4
  Z
iG̃(1) (x, y) = − d x1 d4 x′1 U (x1 , x′1 )
ℏ 2 −∞
(
× iG(0) (x, y)iG(0) (x′1 , x′1 )iG(0) (x1 , x1 ) − iG(0) (x, y)iG(0) (x′1 , x1 )iG(0) (x1 , x′1 )
(A) (B)

+ iG (0)
(x, x1 )iG (0)
(x1 , x′1 )iG(0) (x′1 , y) (0)
− iG (x, x1 )iG (0)
(x1 , y)iG(0) (x′1 , x′1 )
(C) (D)
)
+ iG (0)
(x, x′1 )iG(0) (x′1 , x1 )iG(0) (x1 , y) (0)
− iG (x, x′1 )iG(0) (x1 , y)iG(0) (x1 , x1 )
(E) (F )

then we can associate a picture, or Feynman diagram, with each of the terms on the RHS, as shown in Figure 1.

Figure 1: First-order contributions to iG̃(x, y).

43
7 Green’s Function Methods
7.1 Quasiparticles
Consider a many-body system of interacting fermions or bosons, such as an electron gas consisting of a large number
of electrons interacting by means of the Coulomb force (with a uniform positive charge background to preserve overall
neutrality). If an extra electron is added to such a system (which, in practice is done by exciting a single conduction
electron from below the Fermi level to above it), the repulsive Coulomb interaction means that it repels other electrons,
thus creating an ‘empty’ region of reduced negative charge around it which has the effect of ‘screening’ the effect of this
extra electron on other electrons in the gas. The bare electron plus this screening cloud that moves, or ‘propagates’
through the system can be considered as a quasiparticle, and two such quasiparticle interact only weakly because of this
shielding. Unlike a normal electron in the system, a quasiparticle has an ‘effective’ mass and resulting renormalized
energy, as well as a finite lifetime.
An alternative way of viewing this ‘empty’ region around the extra electron, which will be used in the particle-
hole formalism later, is to imagine the extra electron ‘lifting out’ other electrons in its vicinity and moving them
further away, thus creating positively-charged ‘holes’ in the charge distribution. The holes and lifted-out electrons
are constantly being destroyed by interactions with the extra electron and other electrons in the system, with new
holes and lifted-out electrons taking their place. We can visualize the sum of these microscopic processes as the extra
electron being surrounded by a cloud of constantly-changing holes and lifted-out electrons, which is another way to
picture the quasiparticle, and the one that will be the most useful in determining its properties.
Other examples of quasiparticles are a single conduction electron in a metal with a lattice of ions (the effect of
which gives rise to the familiar effective ‘band mass’ of electrons in metals), nucleons in a nucleus and elementary
excitations in a superconductor (so-called ‘bogolons’).

7.2 Introduction to Propagators


For the system described above, we can define the electron quasiparticle propagator or (retarded) Green’s function,
iG+ (r2 , r1 , t2 − t1 )t2 >t1 , as the probability amplitude that if we add an extra electron at a point r1 and time t1 to the
interacting system in its ground state, the system will be in its ground state with an extra electron at a point r2 at
(later) time t2 . In fact it is the sum of probability amplitudes for all the possible ways that the electron can propagate
through the system, allowing for interactions with other electrons along the way. (Note that we ignore the vacancy
left behind by the ‘extra’ electron, which is in fact promoted from below the Fermi surface to above, and assume that
the Fermi sea remains filled, which corresponds to the ground state of the system.)
More commonly, we actually consider the probability amplitude that if we add an extra electron in a single-particle
eigenstate ϕk1 (r) with wavevector (or what we shall call momentum) k1 at time t1 to the interacting system in its
ground state, the system will be in its ground state with an extra electron in an eigenstate ϕk2 (r) with momentum
k2 at a later time t2 . In this case the electron quasiparticle propagator is then denoted iG+ (k2 , k1 , t2 − t1 )t2 >t1 . For
t2 ⩽ t1 , we define iG+ (k2 , k1 , t2 − t1 )t2 ⩽t1 = 0.
The electron quasiparticle propagator can be expressed as a sum of terms, or perturbation series, with each term in
the series represented by a corresponding Feynman diagram. The first term in the series is just the free propagator, or
the probability amplitude for the electron to undergoes no interactions and be found in the eigenstate ϕk2 (r) = ϕk1 (r)
at time t2 . The quasiparticle propagator will actually have the same form as the free propagator, except for the

renormalization of the energy and the inclusion of an exponential decay factor e−(t −t)/τκ , where τκ is the lifetime
of the quasiparticle, and reflects the fact due to interactions, the quasiparticle is less likely to be in the eigenstate
ϕk2 (r) = ϕk1 (r) at time t2 than the free particle is, and the probability amplitude for this to be the case decreases as
the time interval (t2 − t1 ) increases.
We choose ϕk (r) as the eigenstates of the free-particle Hamiltonian, Ĥ0 , i.e.

Ĥ0 ϕk (r) = ϵk ϕk (r)

where
∇2r 1 ℏ2 k 2
Ĥ0 = − , ϕk (r) = √ eik·r , ϵk =
2m Ω 2m
with Ω the normalization volume.
We first determine an expression of the free propagator. We assume that at time t1 , the wavefunction of the free
particle is given by
ψ(r, t1 ) = ϕk1 (r)

44
At a later time t2 , the time-dependent Schrödinger equation gives the corresponding wavefunction as (with ℏ ≡ 1 for
clarity): X
ψ(r, t2 ) = ϕk1 (r)e−iϵk1 (t2 −t1 ) = aki ϕki (r)
i
where in the last step, we have expanded this new wavefunction in terms of the basis set of free-particle eigenfunctions
{ϕki (r)}, with corresponding probability amplitudes, {aki }.
The retarded free-particle Green’s function for propagation from a state ϕk1 (r) at time t1 to a state ϕk2 (r) at later
time t2 can then be obtained as
Z
iG+
0 (k 2 , k ,
1 2t − t 1 ) ≡ ak2 = d3 r ϕ∗k2 (r)ψ(r, t2 )
Z
= θ(t2 − t1 )e−iϵk1 (t2 −t1 ) d3 r ϕ∗k2 (r)ϕk1 (r)

= θ(t2 − t1 )e−iϵk1 (t2 −t1 ) δk2 k1


where θ(t2 − t1 ) is the unit step function, which ensures that the retarded Green’s function has the desired behaviour
of being non-zero only for t2 > t1 . The delta function gives zero unless k2 = k1 (≡ k), and so we can rewrite the above
(moving the factor of i over to the RHS) as simply
(
−iθ(t2 − t1 )e−iϵk (t2 −t1 ) , t2 ̸= t1
G+0 (k, t2 − t1 ) =
0, t2 = t1

(Note that although the probability amplitude G+ oscillates with time, the probability |G+ |2 = 1 at all times, i.e. the
free electron remains in the initial state indefinitely.)
We can Fourier transform the above expression in order to change from the time variable (t2 − t1 ) to a frequency
variable, or ‘energy parameter’, ω as follows:
Z ∞ Z ∞
iω(t2 −t1 )
G+0 (k, ω) = d(t 2 − t 1 ) G+
(k, t2 − t1 )e = −i d(t2 − t1 ) θ(t2 − t1 )e−iϵk (t2 −t1 ) eiω(t2 −t1 )
−∞ −∞
∞  i(ω−ϵk )(t2 −t1 ) ∞
ei(ω−ϵk )∞
Z
e 1
= −i d(t2 − t1 ) ei(ω−ϵk )(t2 −t1 ) = − = −
0 (ω − ϵk ) 0 (ω − ϵk ) (ω − ϵk )
There is a problem, however, as the function is not well-defined due to the exponential oscillating at infinity. To
overcome this, we introduce a factor of e−δ(t2 −t1 ) , where δ is a positive infinitesimal such that δ × ∞ = ∞, into the
propagator as follows:
−i(ϵk −iδ)(t2 −t1 )
G+0 (k, t2 − t1 ) = −iθ(t2 − t1 )e
For any finite value of (t2 − t1 ), we have δ × (t2 − t1 ) = 0, but for infinite (t2 − t1 ), we have δ × (t2 − t1 ) = ∞, so that
e−δ(t2 −t1 ) = 0 and hence G+ (k, t2 − t1 ) = 0. Fourier transforming with the presence of this extra factor then yields
1 ei(ω−ϵk +iδ)∞ 1
G+
0 (k, ω) = − =
ω − ϵk + iδ ω − ϵk + iδ ω − ϵk + iδ
The above modification has no physical significance, since (t2 − t1 ) is always finite in any experiment, but it does allow
us to work with well-defined integrals, and we can always take the limit δ → 0 at the end of any calculation.
We can see from this Fourier-transformed version of the free propagator that it has poles (singularities) infinites-
imally close to ω = ϵk , i.e. at the energy of the added electron in state ϕk . This is what makes the propagator so
useful: if we can obtain the Fourier transform of the quasiparticle propagator, we can find the energy levels of the
quasiparticle, which correspond to the excited states of the interacting many-body system.
Since we expect electron quasiparticles to act like free electrons, except for the fact that they have a new, renor-
malized energy, ϵ′k , instead of ϵk , together with a lifetime, τk , the quasiparticle propagator will have the same form
as the free propagator, except for the replacement of ϵk with ϵ′k and the inclusion of an exponential decay factor with
decay constant τk , i.e.
−iϵ′k (t2 −t1 ) −(t2 −t1 )
G+ qp (k, t2 − t1 ) = −iZk θ(t2 − t1 )e e
where the factor Zk ⩽ 1 ensures that the added electron satisfies the Pauli exclusion principle. This propagator has
the Fourier transform
Zk
G+qp (k, ω) =
ω − ϵk + iτk−1

which, as expected, has the same form as the Fourier transform of the free propagator, and hence yields the excited
states of the system via its poles, in addition to the quasiparticle lifetime.

45
7.3 Propagator Expansion
In mathematics, if we have a differential equation of the form

Lψ(x, t) = f (x, t)

where L is a linear differential operator that depends only on the differentials w.r.t. x and t, but not on these variables
explicitly, then the Green’s function, G, associated with this equation is the solution of

LG(x − x′ , t − t′ ) = δ(x − x′ )δ(t − t′ )

Since the Schrödinger equation for a free particle can be written as


 2 
∇ ∂
+i ψ(x, t) = 0
2m ∂t

then this is of the same form as the first equation above, just with f (x, t) = 0, and so the associated Green’s function
will obey the equation  2 
∇ ∂
+i G(x − x′ , t − t′ ) = δ(x − x′ )δ(t − t′ )
2m ∂t
The Fourier transform of G(x − x′ , t − t′ ) is given by

d3 k ik·(x−x′ )
Z
′ ′
G(x − x , t − t ) = e G(k, t − t′ )
(2π)3

and if we substitute this into the above equation, while noting that

d3 k ik·(x−x′ )
Z
δ(x − x) = e
(2π)3

we have
∇2 d3 k ik·(x−x′ )
 Z

+i e G(k, t − t′ ) = δ(x − x′ )δ(t − t′ )
2m ∂t (2π)3
d3 k ik·(x−x′ ) k2 d3 k ik·(x−x′ )
Z   Z
∂ ′
=⇒ e − + i G(k, t − t ) = e δ(t − t′ )
(2π)3 2m ∂t (2π)3
k2
 

=⇒ − +i G(k, t − t′ ) = δ(t − t′ )
2m ∂t

If we now use for G our expression for the free propagator from earlier (in terms of the new variables, t and t′ ),
namely
′ ′ −iϵk (t−t′ )
G+0 (k, t − t ) = −iθ(t − t )e

and substitute it in the above, then noting that


d
θ(y) = δ(y) and f (y)δ(y) = f (0)δ(y)
dy

we have (using the fact that ϵk = k 2 /2m):

k2 ik 2
 
∂ ′ ′ ′
− +i G(k, t − t′ ) = θ(t − t′ )e−iϵk (t−t ) − iϵk θ(t − t′ )e−iϵk (t−t ) + δ(t − t′ )e−iϵk (t−t )
2m ∂t 2m

= δ(t − t′ )e−iϵk (t−t ) = δ(t − t′ )

which shows that G+ ′


0 (k, t − t ) is indeed the Green’s function for the unperturbed Schrödinger equation in momentum
space.
We can now extend this treatment to the Schrödinger equation with a momentum-dependent perturbing potential,
V (∇) ≡ V (k):  2 
∇ ∂
+ i − V (∇) ψ(x, t) = 0
2m ∂t

46
so that the associated (quasiparticle) interacting propagator is the solution of the following equation in momentum
space:
k2
 

− + i − V (k) G+ (k, t − t′ ) = δ(t − t′ )
2m ∂t
(Note that G+ is not strictly a Green’s function, since the operator on the LHS is now nonlinear.)
The solution to the above may be written as
Z ∞
′ ′
+ +
G (k, t − t ) = G0 (k, t − t ) + dt′′ G+ ′′ + ′′ ′
0 (k, t − t )V (k)G (k, t − t )
−∞

which we can see by operating on both sides by −k 2 /2m + i∂/∂t :
Z ∞
k2 k2
    
∂ ∂
− +i G+ (k, t − t′ ) = − +i G+0 (k, t − t ′
) + dt′′
G+
0 (k, t − t ′′
)V (k)G+
(k, t′′
− t′
)
2m ∂t 2m ∂t −∞
Z ∞
= δ(t − t′ ) + dt′′ δ(t − t′′ )V (k)G+ (k, t′′ − t′ ) = δ(t − t′ ) + V (k)G+ (k, t − t′ )
−∞

which, when rearranged, is just the above equation for G+ (k, t − t′ ).


The solution can be expanded iteratively to yield
Z ∞
G+ (k, t − t′ ) = G+ ′
0 (k, t − t ) + dt′′ G+ ′′ + ′′
0 (k, t − t )V (k)G0 (k, t − t )

−∞
Z ∞Z ∞
+ dt′′ dt′′′ G+ ′′ + ′′ ′′′ + ′′′ ′
0 (k, t − t )V (k)G0 (k, t − t )V (k)G0 (k, t − t ) + . . .
−∞ −∞

This is a perturbation series in powers of V (k), and if we represent each term by a Feynman diagram, then we obtain
the Feynman diagram series for the quasiparticle propagator.

7.4 Example: Scattering by an External Potential


As an example of using the Feynman diagram series to evaluate the quasiparticle propagator, we consider scattering
of a single electron by an external momentum-dependent potential of the form

V (k) = M k 2 + Lk 4

Such potentials arise in the case of an electron in a magnetic field, as well as in nuclear physics.
We imagine that at time t1 an electron is introduced into the (in this case, initially empty) system in a state
ϕk1 (r) = Ω−1/2 exp(ik1 · r), and propagates through the system, being scattered zero, one or more times by the
external potential, V = M k 2 + Lk 4 . By definition, the quasiparticle propagator iG+ (k2 , k1 , t2 − t1 ) is the probability
amplitude that the electron will be in the state ϕk2 (r) = Ω−1/2 exp(ik2 · r) at time t2 , which is the sum of all the
different ways the electron can go through the system, starting in state ϕk1 and ending up in state ϕk2 .
The simplest way the electron can propagate through the system is freely, without interaction - the probability
amplitude for this is just the free propagator, iG+ 0 . Another way is to start in the state ϕk1 at time t1 , be scattered
by V into the state ϕk2 at time tV , then continue freely in state ϕk2 until time t2 . The amplitude for this second way
will be a product of the amplitudes for the independent processes it is composed of. The first of these independent
processes, i.e. free propagation from t1 to tV in a state ϕk1 , has probability amplitude iG+ 0 (k1 , tV − t1 ), while the
probability amplitude for the second independent process, i.e. scattering from the state ϕk1 to the state ϕk2 by the
potential V at time tV , can be obtained using time-dependent perturbation theory. The third independent process,
i.e. free propagation from tV to t2 in a state ϕk2 , has probability amplitude iG+ 0 (k2 , t2 − tV ).
The result from time-dependent perturbation theory that we need says that if a system at time t0 is in a superpo-
sition of states {ϕi }, with corresponding probability amplitudes {ai }, then under the influence of a perturbation V ,
the rate of change of a particular probability amplitude, say aj , at a later time t is given by

daj (t) X
= −i Vji aj ei(ϵj −ϵi )(t−t0 )
dt i

where Vji is the matrix element of V between states ϕj and ϕi .

47
For our process, we know that the initial state of the system is definitely ϕk1 , so there is only one term in the
summation with ck1 = 1. We are then interested in the system being in the final state ϕk2 , with corresponding
probability amplitude aj , at t = tV . The rate of change of the probability amplitude (or probability amplitude per
unit time) that the system undergoes a transition from ϕk1 to ϕk2 at time tM is therefore given by

dak2 (t)
= −iVk2 k1 ck1 ei(ϵk2 −ϵk1 )(tV −tV ) = −iVk2 k1
dt t=tV
Z Z
= −i d3 r ϕ∗k2 (r)ϕk1 (r) = −i d3 r ϕ∗k2 (r)(M k12 + Lk14 )ϕk1 (r) = −i(M k12 + Lk14 )δk2 k1

The Kronecker delta, δk2 k1 , shows that the above process conserves momentum, so that the particle still has the same
momentum after scattering and we are justified in setting k2 = k1 (≡ k).
To obtain the contribution of this simplest, first-order scattering process to the quasiparticle propagator, we need
to multiply the three factors listed above and then integrate over the time of scattering, tV , i.e.
Z ∞ !
+
 +  dak2 (t)  + 
iG1 (k, t2 − t1 ) = dtV iG0 (k, t2 − tV ) iG0 (k, tV − t1 )
−∞ dt
t=tV
Z ∞
=i dtV G+ +
0 (k, t2 − tV )Vkk G0 (k, tV − t1 )
−∞

where we can simply extend the limits of integration from (t1 , t2 ) to (−∞, ∞) because the free propagators are equal
to zero for tV < t1 and tV > t2 .
There are also second- and higher-order processes in which the electron scatters from V two or more times, and
takeing account of these too gives us the following series expansion for the quasiparticle propagator:
Z ∞
G+ (k, t2 − t1 ) = G+
0 (k, t2 − t1 ) + dtV G+ +
0 (k, t2 − tV )Vkk G0 (k, tV − t1 )
−∞
Z ∞Z ∞
+ dtV dt′V G+ ′ + ′ +
0 (k, t2 − tV )Vkk G0 (k, tV − tV )Vkk G0 (k, tV − t1 ) + . . .
−∞ −∞

We can convert this to an expression in (k, ω)-space if we defining the Fourier-transformed free propagator, G+
0 (k, ω),
via Z ∞
1 −iω(tj −ti )
G+0 (k, tj − ti ) = dω G+0 (k, ω)e
2π −∞
with a similar expression for the quasiparticle propagator, G+ (k, ω). This yields the (k, ω)-space version of the first
term on the RHS of the above series expansion immediately, while for the second term on the RHS we have:
Z ∞
dtV G+ +
0 (k, t2 − tV )Vkk G0 (k, tV − t1 )
−∞
Z ∞  Z ∞   Z ∞ 
1 −iω(t2 −tV ) 1 ′ ′ −iω ′ (tV −t1 )
= dtV dω G+ 0 (k, ω)e Vkk dω G+
0 (k, ω )e
−∞ 2π −∞ 2π −∞
Z ∞ Z ∞ Z ∞
1 ′ i(ω ′ t1 −ωt2 ) ′
= 2
dω dω ′ G+ +
0 (k, ω)Vkk G0 (k, ω )e dtV e−itV (ω −ω)
(2π) −∞ −∞ −∞
Z ∞
1 −iω(t2 −t1 )
= dω G+ +
0 (k, ω)Vkk G0 (k, ω)e
2π −∞

where we have used the fact that Z ∞


1 ′
dtV e−itV (ω −ω) = δ(ω ′ − ω)
2π −∞

Hence the series expansion can be expressed as


Z ∞
1
dω G+ (k, ω)e−iω(t2 −t1 ) =
2π −∞
Z ∞ Z ∞
1 −iω(t2 −t1 ) 1 −iω(t2 −t1 )
dω G+
0 (k, ω)e + dω G+ +
0 (k, ω)Vkk G0 (k, ω)e + ...
2π −∞ 2π −∞

48
or in other words:
G+ (k, ω) = G+ + +
0 (k, ω) + G0 (k, ω)Vkk G0 (k, ω) + . . .

We can now make a ‘dictionary’ to translate the above series into Feynman diagrams - see Figure 2, which shows
how to convert probability amplitudes to diagrams in both (k, t)-space and (k, ω)-space. The resulting Feynman

Figure 2: Diagram dictionary for potential scattering propagator.

diagram series is shown in Figure 3, and consists of the free propagator plus an infinite number of scattering terms,
which constitute the self-energy of the electron. As can be seen, it is possible to break the diagram series down into
a geometric series in terms of diagram components; the sum of the series can then be evaluated using the familiar
summation formula for such a series:
N
X a(1 − rN )
ari−1 =
i=1
1−r
Using the dictionary to translate back into probability amplitudes then yields
1
G+ (k, ω) =
(G+
0 ) −1 − V
kk

which is finite as long as we have |G+0 Vkk | < 1.


The diagrammatic method of summation is extremely powerful, but in this very simple case we could have just
evaluated the sum algebraically, i.e.

G+ (k, ω) = G+ + 2 + 3 2 + + + 2 2
 
0 + (G0 ) Vkk + (G0 ) Vkk + . . . = G0 1 + G0 Vkk + (G0 ) Vkk + . . .
G+ 1
= 0
+ = + −1 (|G+
0 Vkk | < 1)
1 − G0 Vkk (G0 ) − Vkk

We are now in a position to substitute for G+0 and Vkk to obtain an explicit expression for the quasiparticle
propagator, i.e.
1 1
G+ (k, ω) = =
ω − ϵk + iδ − Vkk ω − (ϵk + M k 2 + Lk 4 ) + iδ
Comparing this with the expression for the quasiparticle propagator derived earlier, we find that the renormalized
energy and quasiparticle lifetime are given by
1
ϵ′k = ϵk + M k 2 + Lk 4 , τk = =∞
δ
Hence, the interaction with Vkk has ‘clothed’ the electron, turning it from a free particle into a quasiparticle with a
modified energy dispersion law and an infinite lifetime.

49
Figure 3: Series expansion for quasiparticle propagator expressed in terms of Feynman diagrams, then factorized and
summed using formula for sum of geometric series.

It is worth noting that we can write the diagrammatic series in Figure 3 in a very useful alternative form - see
Figure 4 - which is more general, since it can also be used when the diagrams do not factor. Using our dictionary, this
translates in (k, ω)-space to
G+ (k, ω) = G+ + +
0 (k, ω) + G0 (k, ω)Vkk G (k, ω)
which in (k, t)-space becomes the expression derived in the previous Section:
Z ∞
G+ (k, t − t′ ) = G+
0 (k, t − t ′
) + dt′′ G+ ′′ + ′′ ′
0 (k, t − t )Vkk G (k, t − t )
−∞

We also note that the matrix element Vkk is an example of what is known as the proper or irreducible self-energy,
denoted by Σ(k, ω), since the circle that represents it in the Feynman diagrams cannot be reduced into a smaller part
by cutting any lines, unlike in the case of strings of multiple circles. This allows us to generalize the above equation
in (k, ω)-space
G+ (k, ω) = G+ + +
0 (k, ω) + G0 (k, ω)Σ(k, ω)G (k, ω)
which is known as Dyson’s equation, and will be discussed in detail later. Rearranging this equation yields the following
expression for the quasiparticle propagator in terms of the irreducible self-energy:
1
G+ (k, ω) =
ω − ϵk + iδ − Σ(k, ω)

50
Figure 4: Alternative way to express quasiparticle propagator in terms of Feynman diagrams.

which again does not rely on factorization of the Feynman diagram series, but is more general. Hence, to obtain
the quasiparticle propagator, we basically need to identify all irreducible self-energy parts of the series of Feynman
diagrams.

7.5 Example: Electron in an Impure Metal


We shall now apply the propagator method to a more realistic problem, namely an electron interacting with a set
of N identical, randomly-distributed impurity ions in a metal of volume Ω, where we assume for simplicity that
the regularly-arranged lattice ions in the metal have been removed. The Feynman diagram series for the electron
quasiparticle propagator in such a system, whereby the free electron may scatter multiple times from any of the N
impurity ions, is shown in Figure 5.

Figure 5: Feynman diagram series for electron scattering from impurity ions.

We assume that the potential for an ion at a point Ri will have the form W (r − Ri ), so that the matrix element
for a transition from a free-particle eigenstate ϕk1 to another eigenstate ϕk2 at ion i is given by
Z Z
i
−iVk2 k1 (Ri ) = −i d3 r ϕ∗k2 W (r − Ri )ϕk1 = − d3 r W (r − Ri )e−i(k2 −k1 )·r

Z
i i
= − e−i(k2 −k1 )·Ri d3 r W (r − Ri )e−i(k2 −k1 )·(r−Ri ) = − Wk2 k1 e−i(k2 −k1 )·Ri
Ω Ω

51
where we have introduced the Fourier transform of the potential via
Z

Wk2 k1 ≡ W (k2 − k1 ) = d3 r U (r′ )e−i(k2 −k1 )·r

We can now use the Feynman diagram series to write the algebraic expression for the quasiparticle propagator in
(k, ω)-space as follows:
"N #
X
+ +
+
G (k2 , k1 , ω) = G0 (k1 , ω)δk2 k1 + G0 (k2 , ω) Vk2 k1 (Ri ) G+
0 (k1 , ω)
i=1
" N
#
XX
+ G+
0 (k2 , ω) Vk2 l (Ri )G+
0 (l, ω)Vlk1 (Ri ) G+
0 (k1 , ω)
l i=1
 
XX N
X
+ G+
0 (k2 , ω)
 Vk2 l (Rj )G+
0 (l, ω) Vlk1 (Ri ) G+
0 (k1 , ω) + . . .
l j̸=i i=1

where we now need to sum over the intermediate momentum, l, since we are assuming that the potential does not
necessarily conserve momentum. (Note, however, that although the potential can change the momentum of the
electron, it does not change its energy, and hence ω is conserved along the electron propagator.) The tilde indicates
that the above expression is only for one particular arrangement of impurity coordinates, {Ri }, and for each different
arrangement we will get a different value of G+ . We need to consider an ensemble of all possible arrangements of
impurities, in which the coordinate of the ith impurity, Ri (and all the other impurities), is equally likely to be found
anywhere in the volume Ω. If we can calculate the average over the ensemble, or impurity average, of G+ , denoted
by ⟨G+ ⟩, then in the limit N → ∞ (with N/Ω constant), we are justified in assuming that this represents the actual
quasiparticle propagator.
The impurity average ⟨G+ ⟩ will be equal to the sum of the impurity averages of each term on the RHS of the above
perturbation expansion. For the first term, we have simply that ⟨G+ +
0 ⟩ = G0 , since the free propagator is independent
of Ri . For the second term, we can use the aforementioned fact to factorize out the free propagators, so that we have,
after substituting for the matrix element:
* "N # + *N +
+
X
+ + + Wk2 k1 X −i(k2 −k1 )·Ri
G0 (k2 , ω) Vk2 k1 (Ri ) G0 (k1 , ω) = G0 (k2 , ω)G0 (k1 , ω) e
i=1
Ω i=1

We can rewrite the last factor in the above as


*N + N D
X X E D E
−i(k2 −k1 )·Ri
e = e−i(k2 −k1 )·Ri = N e−i(k2 −k1 )·Ri
i=1 i=1

since each of the N terms is identical in form.


Since the probability of the ith impurity atom being located within the infinitesimal volume d3 Ri surrounding the
point Ri will be independent of Ri and equal to d3 Ri /Ω, we can define the impurity average of a physical quantity,
A, via Z Y 3
d Ri
⟨A⟩ = Â[{Ri }]
i

Applying this to the exponential function above, we are interested in only a single impurity at all possible values of
Ri , so that the impurity average simplifies to
D E Z d3 R 1
i −i(k2 −k1 )·Ri
e−i(k2 −k1 )·Ri = e = × Ωδk2 k1 = δk2 k1
Ω Ω
so that we have
* " N
# +
X N +
G+
0 (k2 , ω) Vk2 k1 (Ri ) G+
0 (k1 , ω) = G (k2 , ω)G+
0 (k1 , ω)Wk2 k1 δk2 k1
i=1
Ω 0

52
For the third term in the propagator expansion, which represents two successive scatterings from the same impurity,
the impurity average of the factor in squared brackets yields
*N + *N +
X
+
X X
+ Wk2 l Wlk1 X −i(k2 −l+l−k1 )·Ri
G0 (l, ω) Vk2 l (Ri )Vlk1 (Ri ) = G0 (l, ω) e
i=1
Ω2 i=1
l l
N X +
= 2 G0 (l, ω)Wk2 l Wlk1 δk2 k1

l

In the case of a macroscopic system, where the points l in momentum space are very close together (with density
Ω/(2π)3 ), we are justified in making the change from a sum to an integral as follows:
Z
X Ω
→ d3 l
(2π)3
l

so that the expression for the third term in the expansion becomes
*N +
d3 l + d3 l +
Z Z
X
+
X N N
G0 (l, ω) Vk2 l (Ri )Vlk1 (Ri ) = 3
G0 (l, ω)W k 2 l W lk 1 δ k2 k1 = G (l, ω)|Wlk1 |2
3 0
i=1
Ω (2π) Ω (2π)
l

For the fourth term in the propagator expansion, which represents two successive scatterings from two different
impurities (at all possible values of Ri and Rj ), the impurity average of the factor in squared brackets yields
*N + *N +
X XX X Wk2 l Wlk1 XX
−i(k2 −l)·Rj −i(l−k1 )·Ri
G+
0 (l, ω) Vk2 l (Rj )Vlk1 (Ri ) = G+
0 (l, ω) 2
e e
i=1 j̸=i
Ω i=1 j̸=i
l l
Z 3 Z 3
N (N − 1) X + d Ri d Rj −i(k2 −l)·Rj −i(l−k1 )·Ri
= G0 (l, ω)Wk2 l Wlk1 e e
Ω2 Ω Ω
l
 2 X
N
≃ G+ 0 (l, ω)Wk2 l Wlk1 δk2 l δlk1

l
 2
N
≃ G+ 2
0 (k1 , ω)Wk2 k1 δk2 k1

where we have assumed that the probability of the ith impurity atom being located within the infinitesimal volume
d3 Ri and the jthe impurity being simultaneously located within d3 Rj is equal to (d3 Ri /Ω)(d3 Rj /Ω), and also that
N is large enough such that N (N − 1) ≃ N 2 . (Note that the Kronecker deltas ensure that l can take only one value,
i.e. l = k1 = k2 , so we are free to choose either of these in the argument of G+ 0 (l, ω).)
We are now in a position to write down an expression for the quasiparticle propagator, G+ = ⟨G+ ⟩. Taking into
account the Kronecker deltas in each term, which enforce conservation of momentum between initial and final states,
we let k1 = k2 ≡ k to obtain
 2
d3 l +
Z
N + N + N
⟨G+ ⟩ ≃ G+0 + [G (k, ω)]2
W kk + [G (k, ω)]2
G (l, ω)|W lk |2
+ [G+ 3 2
0 (k, ω)] Wkk + . . .
Ω 0 Ω 0 (2π)3 0 Ω

We can also express this in terms of Feynman diagrams using the dictionary in Figure 6, where we introduce an empty
circle to show that Vkl (Ri ) has been replaced by Wkl (plus factor(s) of N/Ω) due to impurity averaging, and use
dotted lines to connect scattering events from the same impurity.

53
Figure 6: Diagram dictionary for impurity-averaged propagator.

This leads to the diagram series in Figure 7, which includes terms representing mixtures of multiple scatterings
from the same impurity and from different impurities. The first term on the RHS is just the free propagator, while
the remaining terms constitute the self-energy of the electron.

54
Figure 7: Feynman diagram series for impurity-averaged propagator.

The irreducible self-energy, on the other hand, which we denote by Σ(k, ω), is made up of parts such as the second,
third, fifth and eighth terms on the RHS, since these cannot be broken down into simpler units by cutting internal
lines. The complete series for Σ(k, ω) is shown in Figure 8.

55
Figure 8: Complete diagrammatic series for irreducible self-energy, Σ(k, ω).

In order to arrive at an approximation for the irreducible self-energy, we make the assumption that the most
important processes are single and double scattering by the same impurity, i.e. we consider only the contributions to
the irreducible self-energy from the first two terms in the series in Figure 8. (Since these terms are proportional to
N/Ω, this corresponds to assuming a low density of impurities.) This approximation, together with the expression for
the quasiparticle propagator obtained from Dyson’s equation, are shown diagrammatically in Figure 9).

Figure 9: Quasiparticle propagator expressed in terms of irreducible self-energy approximated as shown.

We can now use the dictionary to translate the expressions for the impurity-averaged quasiparticle propagator and

56
approximated irreducible self-energy by into algebraic form:
1 1
⟨G+ (k, ω)⟩ = =
(G+
0)
−1 − Σ(k, ω) ω − ϵ k + iδ − Σ(k, ω)
where
d3 l |Wlk |2
Z
N N
Σ(k, ω) = Wkk +
Ω Ω (2π)3 ω − ϵl + iδ
In order to find the quasiparticle energy and lifetime, we need to determine the pole of the quasiparticle propagator,
i.e. we need to solve the following for ω:
ω − ϵk + iδ − Σ(k, ω) = 0
Considering the real part of this equation and rearranging yields
d3 l |Wlk |2
Z
N N
ω = ϵk + Σ(k, ω) = ϵk + Wkk +
Ω Ω (2π)3 ω − ϵl + iδ
Note that ω appears on both sides of the above equation. Since we are assuming that W , and hence Σ(k, ω), is small
we can simply iterate. The zeroth-order approximation to ω is ω = ϵk , so to obtain the first-order approximation we
substitute this into the RHS as follows:
d3 l |Wlk |2
Z
N N
ω = ϵk + Wkk +
Ω Ω (2π)3 ϵk − ϵl + iδ
= ϵk + Σ(k, ϵk ) = ϵk + Re Σ(k, ϵk ) +i Im Σ(k, ϵk )
ϵ′k −τk−1

Hence, to determine the quasiparticle energy and lifetime we need to find the real and imaginary parts of Σ(k, ϵk ).
To start with, we note that the vector integral in the expression for Σ(k, ϵk ) can be expanded as follows:
Z Z Z Z
A(l, . . .) A(l, θ, ϕ, . . .)
d3 l = dϕ dθ sinθ dl l2
B(l, . . .) + iδ B(l, θ, ϕ, . . .) + iδ
In our case, B (i.e. the difference in energies, ϵk − ϵl ) is a function of l only, so if we let x = B(l), we can write
l = B −1 (x), which allows us to reexpress the integral over l as
[B −1 (x)]2 A[B −1 (x), . . .]
 −1
dB (x) [B −1 (x)]2 A[B −1 (x), . . .]
Z Z Z 
A(l, θ, ϕ, . . .)
dl l2 = dB −1 (x) = dx
B(l, θ, ϕ, . . .) + iδ x + iδ dx x + iδ
We then make use of the following identity from the theory of complex functions:
Z Z Z
f (x) f (x)
dx = P dx − iπ dx f (x)δ(x)
x + iδ x
where P denotes the Cauchy principal value and δ(x) on the RHS is a delta function, to rewrite the above as
 −1
dB (x) [B −1 (x)]2 A[B −1 (x), . . .]
Z 
dx =
dx x + iδ
 −1
dB (x) [B −1 (x)]2 A[B −1 (x), . . .]
Z  Z  −1 
dB (x)
P dx − iπ dx [B −1 (x)]2 A[B −1 (x), . . .]δ(x)
dx x dx
Converting back to an integral over l, and hence l, then yields the general equality
Z Z Z
3 A(l, . . .) 3 A(l, . . .)
d l =P d l − iπ d3 l A(l, . . .)δ(B(l, . . .))
B(l, . . .) + iδ B(l, . . .)
Applying this to the integral in the expression for Σ(k, ϵk ) then gives
d3 l |Wlk |2 d3 l |Wlk |2
Z Z Z
=P − iπ d3 l |Wlk |2 δ(ϵk − ϵl )
(2π)3 ϵk − ϵl + iδ (2π)3 ϵk − ϵl
and hence, taking the real and imaginary parts of this yields the following expressions for the quasiparticle energy and
(inverse) lifetime:
d3 l |Wlk |2
  Z
N N
ϵ′k = ϵk + Wkk + P
Ω Ω (2π)3 ϵk − ϵl
 Z
N
τk−1 = π d3 l |Wlk |2 δ(ϵk − ϵl )

57
7.6 Formal Definition of Propagator
For an interacting system of particles with Heisenberg-picture ground state |Ψ0 ⟩, we can define the single-particle
Green’s function (quasiparticle propagator) via

⟨Ψ0 |T [Ψ̂H (x, t)Ψ̂†H (x′ , t′ )]|Ψ0 ⟩


iG(x, t; x′ , t′ ) ≡
⟨Ψ0 |Ψ0 ⟩

For now we assume that the ground state is normalized, i.e. ⟨Ψ0 |Ψ0 ⟩ = 1, so that the above can be written as simply

iG(x, t; x′ , t′ ) ≡ ⟨Ψ0 |T [Ψ̂H (x, t)Ψ̂†H (x′ , t′ )]|Ψ0 ⟩

Express as sum of advanced and retarded functions, applicable to describing both particle and hole in particle-hole
picture. Go on to introduce this picture in Fermi systems and formal derivation of Feynman diagram expansion (see
below for both).

7.7 Fermi Systems


7.7.1 Particle-Hole Picture
In our treatment of Fermi systems so far, we have considered exciting a single electron from a state below the Fermi
surface to a state above it, and then studied how this electron interacts with the other electrons in the system via the
concept of a quasiparticle. However, we can also view things in a different, but completely equivalent way, by using
the so-called particle-hole picture. In this picture, the filled Fermi sea is taken as the ground state (vacuum), and the
aforementioned excitation of a single electron is instead viewed as the creation of a ‘particle’ above the Fermi surface
and a ‘hole’ below it - see Figure 10. (The ‘hole’ is analogous to the positron in Dirac’s theory, and it is taken as having
minus the energy, (vector) momentum and spin of the removed electron.) When considering the interacting system,
we now focus on how the ‘particle’ becomes clothed by other ‘particle-hole’ pairs, thus forming a quasi-‘particle’ that
can be described by a propagator and the Feynman diagram formalism as previously. (In what follows, we drop the
inverted commas and simply refer to particles, holes and quasiparticles, but have in mind the picture outlined above.)

Figure 10: Equivalence between electron excited from below the Fermi level to above in the normal picture, and
addition of a particle and hole to the Fermi vacuum in the particle-hole picture.

In terms of second quantization, the above change corresponds to making a following canonical transformation from
electron to particle/hole creation and annihilation operators. Under such a transformation we redefine the electron
creation operator via (suppressing the spin variable):
(
† â†k k > kF (particle)
ĉk =
b̂−k k < kF (hole)

i.e. creating an electron with momentum k either above or below the Fermi surface is equivalent to either creating
a particle with momentum k above the Fermi surface or annihilating a hole with momentum −k below the Fermi

58
surface. (It is important to note that the creation operator for an electron, etc. does not represent the process of
exciting the electron from one state to another; it simply represents the presence of the electron in a particular state.)
Similarly, we redefine the electron annihilation operator via:
(
âk k > kF (particle)
ĉk = †
b̂−k k < kF (hole)

i.e. annihilating an electron with momentum k either above or below the Fermi surface is equivalent to either
annihilating a particle with momentum k above the Fermi surface or creating a hole with momentum −k below the
Fermi surface. We can represent the above alternatively as

ĉ†k = θ(k − kF )â†k + θ(kF − k)b̂−k


ĉk = θ(k − kF )âk + θ(kF − k)b̂†−k

We are now in a position to express the Hamiltonian for the system, which describes the total energy of all the
electrons, in terms of these redefined operators. Starting with a non-interacting system in an external perturbing
potential, described by the first-quantized Hamiltonian
X p2 X
i
Ĥ = + V (ri )
i
2m i
Ĥ0 Ĥ1

we have for the unperturbed part of the Hamiltonian in second-quantized form (indicating just the magnitude of the
momenta in the operator indices, rather than the full vector momentum):
X †
ϵk â†k âk − ϵk b̂†k b̂k +
X X X
Ĥ0 = ϵk ĉk ĉk = ϵk
k k>kF k<kF k<kF
particles holes filled Fermi sea

The first line above corresponds to the situation in Figure 10(b), whereas the second line corresponds to that in Figure
10(d).
In the absence of particles or holes, the energy is simply that of the filled Fermi sea. Creating a hole lowers the
energy, whereas creating a particle raises the energy. If the total number of electrons is fixed, particles and holes
necessarily occur in pairs, and since each particle-hole pair has a net positive energy (because the momentum of the
particle is larger than that of the hole), the filled Fermi sea (i.e. with no particle-hole pairs) represents the ground
state.
Similarly, for the perturbed part of the Hamiltonian
X
Ĥ1 = Vmn ĉ†m ĉn
m,n
X X X X
= Vmn â†m ân + Vmn â†m b̂†n + Vmn b̂m ân + Vmn b̂m b̂†n
m,n>kF m>kF ,n<kF m<kF ,n>kF m,n<kF

Moving now to an interacting system in an external perturbing potential, described by the first-quantized Hamil-
tonian
X p2 X
i
Ĥ = + 21 V (ri − rj )
i
2m i,j
Ĥ0 Ĥ1

the unperturbed part of the Hamiltonian in second-quantized form is exactly the same as for the non-interacting
system. However, for the perturbed part of the Hamiltonian we now have

Vklmn ĉ†l ĉ†k ĉm ĉn


X
Ĥ1 = 12
k,l,m,n

Vklmn â†l â†k âm ân + Vklmn â†l â†k âm b̂†n + . . . +
X X X
= 1
2
1
2
1
2 Vklmn b̂l b̂k b̂†m b̂†n
k,l,m,n>kF k,l,m>kF ,n<kF k,l,m,n<kF

59
We now consider how to define propagators for particles and holes. The particle case is identical to that of the
electron, i.e. the quasiparticle propagator or retarded Green’s function, iG+ (k2 , k1 , t2 − t1 ), is simply the probability
amplitude that if we add a particle to the interacting system in an eigenstate ϕk1 (r) with momentum k1 at time t1 ,
it will be found in an eigenstate ϕk2 (r) with momentum k2 at a later time t2 , and is given by
(
+ −iθ(t2 − t1 )e−iϵk (t2 −t1 ) , t2 ̸= t1 , ϵk > ϵF
G0 (k, t2 − t1 ) =
0, t2 = t1

with Fourier transform


1
G+
0 (k, ω) = , ϵk > ϵF
ω − ϵk + iδ
To define a propagator for the hole, we begin by noting that the hole wavefunction, which is the same as that of
the removed electron, will contain an exponential decay factor e−i(−ϵk )t (ϵk < ϵF ), where −ϵk is the energy of the hole.
Since negative energies seem unphysical, we note that we can obtain the same value of the exponent if we assume
a positive energy and associate the sign change with the time factor instead, i.e. if we view the hole as an electron
moving backwards in time. Therefore, a hole moving forwards in time from t2 to t1 (where t2 < t1 ) is simply an
electron moving backwards in time from t1 to t2 .
We are now in a position to define the quasihole propagator or advanced Green’s function, iG− (k2 , k1 , t2 − t1 )t2 ⩽t1 ,
as (−1 ×) the probability amplitude that if we remove an electron in a single-particle eigenstate ϕk2 (r) at time t2 from
the interacting system in its ground state (i.e. add a hole in a state ϕk2 ), the system will be in its ground state with an
electron removed from the eigenstate ϕk1 (r) (i.e. with an added hole in the state ϕk2 ) at a later (or equal) time t1 . (The
factor of −1 is due to the fact that we are considering fermions). For t2 > t1 , we define iG+ (k2 , k1 , t2 − t1 )t2 >t1 = 0.
When drawing Feynman diagrams, propagators representing holes are shown with arrows pointing in the opposite
direction to those representing particles, to indicate that a hole moving forward in time is equivalent to an electron
moving backwards in time.
Following the method for the retarded propagator, the advanced free-particle Green’s function for propagation
from a state ϕk2 (r) at time t2 to a state ϕk1 (r) at later (or equal) time t1 can be obtained as
(
− iθ(t1 − t2 )e−iϵk (t2 −t1 ) , t2 ̸= t1 , ϵk < ϵF
G0 (k, t2 − t1 ) =
i, t2 = t1

where ϵk (> 0) is the energy of the removed electron. Fourier transforming the above expression yields
Z ∞ Z ∞
G−0 (k, ω) = d(t 2 − t 1 ) G−
(k, t2 − t 1 )e iω(t2 −t1 )
= i d(t2 − t1 ) θ(t1 − t2 )e−iϵk (t2 −t1 ) eiω(t2 −t1 )
−∞ −∞
Z −∞ Z ∞
= −i d(t1 − t2 ) θ(t1 − t2 )eiϵk (t1 −t2 ) e−iω(t1 −t2 ) = i d(t1 − t2 ) θ(t1 − t2 )eiϵk (t1 −t2 ) e−iω(t1 −t2 )
∞ −∞
∞  −i(ω−ϵk )(t1 −t2 ) ∞
e−i(ω−ϵk )∞
Z
e 1
=i d(t1 − t2 ) e−i(ω−ϵk )(t1 −t2 ) = − = −
0 (ω − ϵk ) 0 (ω − ϵk ) (ω − ϵk )

To deal with the exponential oscillating at infinity we introduce a factor of e−δ(t1 −t2 ) , where δ is a positive infinitesimal,
into the propagator as follows:
G−
0 (k, t2 − t1 ) = iθ(t1 − t2 )e
−i(ϵk +iδ)(t2 −t1 )

and Fourier transforming with the presence of this extra factor then yields
1 e−i(ω−ϵk −iδ)∞ 1
G−
0 (k, ω) = − = , ϵk < ϵF
ω − ϵk − iδ ω − ϵk − iδ ω − ϵk − iδ
This has the same form as the particle propagator, except for the opposite sign of the imaginary part of the denomi-
nator.

7.7.2 Example: Non-Interacting System of Electrons in an External Potential


As a first example of the application of the diagram technique to a many-body system of fermions within the particle-
hole picture, we consider a system of electrons in an external perturbing potential, V (r). We have already seen how
the contribution that this interaction makes to the Hamiltonian is expressed in the particle-hole picture, i.e.

Vkl â†k âl + Vkl â†k b̂†l + Vkl b̂k b̂†l


X X X X
Ĥ1 = Vkl b̂k âl +
k,l>kF k>kF ,l<kF k<kF ,l>kF k,l<kF

60
where the matrix element, or interaction amplitude, Vkl , is given by
Z
Vkl = d3 r ϕ∗k (r)V (r)ϕl (r)

The four terms in the interaction part of the Hamiltonian above therefore correspond to, respectively: (a) the potential
scattering a particle (above the Fermi surface) from state ϕl to state ϕk ; (b) the potential scattering an electron out of
state ϕl (where ϵl < ϵF ) into state ϕk (where ϵk > ϵF ), which translates in the particle-hole picture to simultaneously
creating a particle in ϕk and a hole in ϕl ; (c) the potential scattering an electron out of state ϕl (where ϵl > ϵF ) into
state ϕk (where ϵk < ϵF ), which translates in the particle-hole picture to simultaneously annihilating a particle in ϕl
and a hole in ϕk ; and (d) the potential scattering a hole (below the Fermi surface) from state ϕk to state ϕl . These
processes, together with the various propagators associated with particles and holes, are summarized in the dictionary
in Figure 11.

Figure 11: Diagram dictionary for many-electron system in external perturbing potential (Goldstone method).

We can now build up the diagrammatic series for the quasiparticle propagator by summing all the possible different
diagrams that can be built up using sequences of interaction dots connected by particle and hole lines, beginning in a
state with momentum k1 at time t1 and ending in a state with momentum k2 at time t2 (see Figure 12). The physical
significance of the first four terms in the series is as follows: (1) a particle enters the system in state k1 at time t1 ,
propagates freely, and leaves the system in the same state k1 (= k2 ) at time t2 ; (2) a particle enters the system in state
k1 at time t1 , is scattered by the potential at time t into state k2 , and continues propagating in this state until t2 ;
(3) a particle enters the system in state k1 at time t1 , is scattered by the potential at time t into state q, then again
at time t′ into state k2 , and continues propagating in this state until t2 ; and (4) a particle enters the system in state
k1 at time t1 ; then at time t′ the potential knocks an electron out of the state l into the state k2 , thus creating a
particle in k2 and a hole in l; then at time t the potential knocks the particle in k1 into the hole in l causing mutual
annihilation; and finally the particle in k2 continues propagating in this state until t2 .

61
Figure 12: Diagram series for quasiparticle propagator with external perturbing potential.

Using the dictionary in Figure 11, we can translate the diagram series into the following algebraic expansion for
the quasiparticle propagator in (k, t)-space:
Z ∞

G+ (k2 , k1 , t2 − t1 ) = G+
0 (k 1 , t − t )δ k1 k2 + dt G+ +
0 (k2 , t2 − t)Vk2 k1 G0 (k1 , t − t1 )
−∞
X Z ∞Z ∞
′ ′ ′
+ dt dt G+ + +
0 (k2 , t2 − t )Vk2 q G0 (q, t − t)Vqk1 G0 (k1 , t − t1 )
q>kF −∞ −∞
XZ ∞ Z ∞

+ dt dt′ G+ ′ ′ +
0 (k2 , t2 − t )Vk2 l G0 (l, t − t)Vlk1 G0 (k1 , t − t1 ) + . . .
l<kF −∞ −∞

or in (k, ω)-space (where ω is again conserved by the external potential):

G+ (k2 , k1 , ω) = G+ + +
0 (k1 , ω)δk1 k2 + G0 (k2 , ω)Vk2 k1 G0 (k1 , ω)
X X

+ G+ + +
0 (k2 , ω)Vk2 q G0 (q, ω)Vqk1 G0 (k1 , ω) + G+ +
0 (k2 , ω)Vk2 l G0 (l, ω)Vlk1 G0 (k1 , ω) + . . .
q>kF l<kF

The power of the diagram technique lies in the fact that we do not even need to write out the algebraic series
for the propagator - we simply decide which diagrams are most important and carry out a partial summation. In
the present case, we make the assumption that the potential is such that Vkm and Vmk are large (where we take
k1 = k2 ≡ k(> kF ), and m(< kF ) now labels the intermediate momentum, rather than l), while the other two matrix
elements are small, i.e. particle-hole pair creation and annihilation dominate. The quasiparticle propagator may then
be approximated by the partial summation represented by the diagrams in Figure 13.

Figure 13: Partial summation of diagram series for external perturbing potential.

62
Translating back into algebraic notation, we therefore have
1 1
G+ (k, ω) = − =
[G+
0 (k, ω)]
−1 − Vkm G0 (m, ω)Vmk (ω − ϵk + iδ) − |Vkm |2
(ω−ϵm −iδ)

where ϵk and ϵm are the energies of the unperturbed single-electron states above and below the Fermi level, respectively.
The poles of G+ (k, ω) tell us how these energy levels are modified by the external potential, i.e. they give the excited
state energies of the perturbed system, and are determined by solving the following equation for ω:

|Vkm |2
ω − ϵk − =0
ω − ϵm
which yields
ϵk + ϵm 1 1/2
ω= ± (ϵk − ϵm )2 + 4|Vkm |2
2 2
where the upper and lower signs correspond to the modified energies, ϵ′k and ϵ′m , respectively.

7.7.3 Interacting Fermi Systems


We now move on to consider a many-body system of interacting electrons, where this interaction takes place via a
two-body potential, V (ri − ri ). The contribution that this interaction makes to the Hamiltonian is expressed in the
particle-hole picture as

Vklmn â†l â†k âm ân + 21 Vklmn â†l â†k âm b̂†n + . . . + 21
X X X
Ĥ1 = 21 Vklmn b̂l b̂k b̂†m b̂†n
k,l,m,n>kF k,l,m>kF ,n<kF k,l,m,n<kF

where the matrix element, or interaction amplitude, Vklmn , is given by


Z Z
Vklmn = d3 r d3 r′ ϕ∗k (r)ϕ∗l (r′ )V (r − r′ )ϕm (r)ϕn (r′ ) = Vlknm

The terms in the Hamiltonian can be represented by the Feynman diagrams in Figure 14, and correspond to, respec-
tively: (a) ordinary scattering of two particles (above the Fermi surface) from states ϕm and ϕn to states ϕk and ϕl ;
(b) a particle in ϕm simultaneously creating a particle in ϕl and a hole in ϕn , with the original particle undergoing a
transition to state ϕk in the process; (c) two particle-hole pairs being spontaneously created out of the vacuum; and
(d) ordinary scattering of two holes (below the Fermi surface) from states ϕk and ϕl to states ϕm and ϕn .

Figure 14: Feynman diagrams corresponding to terms in interaction part of Hamiltonian.

Since we are interested in determining the single-particle propagator, we shall focus only on processes that involve
a single particle line entering and leaving a diagram, which means that we shall join up pairs of hole lines into
loops. Also, we shall concentrate only on scattering processes in which the particle enters the diagram with the same
momentum as it leaves with (which we label k) - this means that the matrix element will now be denoted Vklkl .
(Note, however, that the interaction does not in general conserve momentum.) A full treatment would consider all
the possible Feynman diagrams of different orders (i.e. with different numbers of interaction lines) that contribute to
the quasiparticle propagator, but for the moment we shall consider just first-order diagrams, which give the dominant
contribution to the perturbation series and yield some useful approximations.
The possible first-order processes are represented by the diagrams in Figure 15, which are known as bubble diagrams
and open oyster diagrams. The bubble diagrams represent what is known as forward scattering, and can be interpreted
as follows: a particle enters in a state with momentum k, knocks an electron out of a state with momentum l at time
t, thus creating a hole, then instantaneously knocks the electron back into the same state, thus annihilating the hole,
and then continues freely in a state with momentum k. Since the hole that is created effectively exists for zero time,

63
there is effectively no transfer of momentum and hence no accompanying particle. The open oyster diagrams represent
what is known as exchange scattering, and the process described is similar to that of the bubble diagrams, except
for the fact that when the incoming particle strikes the electron in the state with momentum l at time t and creates
the hole, it is instantaneously exchanged with the electron in state l, so that it is this exchanged particle that then
continues freely in the state with momentum k. (Note that diagrams (a) and (b), and (c) and (d), are simply mirror
images of one another, corresponding to the change r ↔ r′ , but must be included for completeness.) We now consider
how to evaluate the contribution to the quasiparticle propagator from bubble and open oyster diagrams.

Figure 15: First-order contributions to quasiparticle propagator, corresponding to forward and exchange scattering.

The bubble diagram in Figure 15(a) yields the following contribution to the quasiparticle propagator:

where we have integrated over the intermediate time t, and summed over the intermediate momentum l. The
factor of −1 in front comes from the fact that the diagram contains one ‘fermion loop’ (see below). In fact, an
additional factor of −1 occurs when we evaluate the equal-time propagator for the (non-propagating) fermion loop,
since iG−0 (l, t − t) = i × i = −1 from the definition of the advanced Green’s function above. Keeping this in mind, the
contribution of the bubble diagram in (k, ω)-space (where we assume that ω is conserved along with k from when the
particle enters to when it leaves the diagram) is obtain as:

Similarly, the reverse bubble diagram in Figure 15(b) yields:

64
However, since Vlklk = Vklkl , the contributions from the two diagrams are equal. In fact, this is an example of a
general rule, which says that given a particular diagram, if we form a new diagram from it by rotating one or more of
its interaction lines about the vertical axis by 180 degrees, then the new diagram has the same value as the original
one. Hence, we can omit all twisted and/or mirror image diagrams if we just multiply the contribution of the original
diagrams by a factor of 2.
Moving on to the open oyster diagrams in Figure 15(c) and (d), these yields the following total contribution to the
quasiparticle propagator in (k, ω)-space:

where the factor of 2 mentioned above has been included. The single factor of −1 at the end comes from evaluating
the equal-time propagator for the non-propagating exchange ‘half-loop’.
The rules for evaluating the first-order diagrams above, as well as higher-order diagrams (which are in general
known as self-energy diagrams), for the interacting many-electron system are tabulated in the dictionary in Figure 16.

65
Figure 16: Diagram dictionary for interacting many-electron system (Goldstone method).

As an example of the use of this, we evaluate the simplest second-order diagram, first in (k, t)-space:

and then in (k, ω)-space:

66
where we see that frequency (or energy parameter) ω, etc. is conserved at each vertex as well as momentum.

7.7.4 Hartree and Hartree-Fock Approximations


We now consider the simplest of all partial sum approximations for the propagator. We begin by assuming that the
interaction between electrons is dominated by forward-scattering processes, so that the interaction Vklmn is given by
Vklmn = δmk δnl Vklkl + . . .
This is known as the Hartree approximation, and in diagrammatic terms it means that the most important diagrams
contributing to the perturbation series will be those consisting of one or more bubble parts, as shown in Figure 17, so
that the quasiparticle propagator can be approximated by a partial summation over repeated bubbles. In algebraic
terms this corresponds to
1
iG+ (k, ω) = + −1
P
[iG0 (k, ω)] − [(−1) l<kF (−iVklkl (−1)]
or simply
1
G+ (k, ω) = P
ω − ϵk − l<kF Vklkl + iδ
P
where l<kF Vklkl is the self-energy of the electron. The above expression yields the quasiparticle energy and lifetime
as X 1
ϵ′k ϵk + Vklkl , τk = =∞
δ
l<kF

Figure 17: Partial summation of Feynman diagram series in Hartree approximation.

We can see that


P the quasiparticle propagator is identical in appearance to that for the impurity scattering case, with
Vkk replaced by l<kF Vklkl (and the circles in the diagram series replaced by bubbles). We can therefore interpret

67
P
−i l<kF Vklkl as the probability amplitude for a transition from a state ϕk to the same state ϕk caused by an effective
external potential, Veff , which can be determined from
X XZ Z
−i Vklkl = −i d3 r d3 r′ ϕ∗k (r)ϕ∗l (r′ )V (r − r′ )ϕk (r)ϕl (r′ )
l<kF l<kF
Z " #
XZ Z
= −i 3
d r ϕ∗k (r) 3 ′ ′ 2 ′
d r |ϕl (r )| V (r − r ) ϕk (r) = −i d3 r ϕ∗k (r)Veff ϕk (r)
l<kF

Since |ϕl (r′ )|2 is the density at a point r′ of an electron in the state ϕl below the Fermi surface, we see from the
summation over l that Veff is the average potential at a point r due to all the electrons in the Fermi sea, and hence
the Hartree approximation is a mean-field approximation.
The next stage is to assume that the interaction between electrons is dominated by both forward-scattering and
exchange-scattering processes, so that the interaction Vklmn is now given by
Vklmn = δmk δnl Vklkl + δml δnk Vkllk + . . .
This is known as the Hartree-Fock approximation, and in diagrammatic terms it means that we must now consider all
diagrams consisting of one or more bubble parts and/or one or more open oyster parts, as shown in Figure 18, which
leads to a partial summation over bubbles and open oysters. In algebraic terms this corresponds to
1
G+ (k, ω) = P
ω − ϵk − l<kF (Vklkl − Vkllk ) + iδ

which yields the quasiparticle energy and lifetime as


X
ϵ′k = ϵk + (Vklkl − Vkllk ), τk = ∞
l<kF

Figure 18: Partial summation of Feynman diagram series in Hartree-Fock approximation.

The Hartree and Hartree-Fock approximations have the limitation that they treat the effect of all the other electrons
on the test electron via a time-independent average (i.e. effective) potential, which leads to unphysical results such
as an infinite quasiparticle lifetime and zero effective mass. To capture the sense in which a cloud of virtual particles
follows the bare particle, we need to move from this static approach to a more dynamic approach, whereby we taken
into account what are known as correlations, which have the effect of screening the bare electron and reducing the
strength of the interaction between electrons. To do this, we need to take into account higher-order diagrams in the
perturbation series.

68
7.7.5 Time-Ordered Green’s Function
In the preceding Sections we defined the retarded and advanced Green’s functions, G+ (k, t2 − t1 ) and G− (k, t2 − t1 ),
where the former is only non-zero for t2 > t1 and ϵk > ϵF (and hence describes particles above the Fermi surface),
while the latter is only non-zero for t2 ⩽ t1 and ϵk < ϵF (and hence describes holes below the Fermi surface). We can
combine both of these into a single time-ordered Green’s function as follows:

G(k, t2 − t1 ) = G+ (k, t2 − t1 )t2 >t1 + G− (k, t2 − t1 )t2 ⩽t1

which reduces to the retarded or advanced function depending on the order of t1 and t2 .

In the free-particle case, we can use the expressions from earlier for G+ 0 and G0 to express the free-particle
time-ordered Green’s function in the form
(
−i[θ(t2 − t1 )θ(ϵk − ϵF )e−iϵk (t2 −t1 ) − θ(t1 − t2 )θ(ϵF − ϵk )e−iϵk (t2 −t1 ) ], t2 ̸= t1
G0 (k, t2 − t1 ) =
iθ(ϵF − ϵk ), t2 = t1

which, after taking the Fourier transform, yields the following in (k, ω)-space:

θ(ϵk − ϵF ) θ(ϵF − ϵk )
G(k, ω) = +
ω − ϵk + iδ ω − ϵk − iδ
which can be expressed more compactly as
1
G(k, ω) =
ω − ϵk + iδ sgn(ϵk − ϵF )

An even more compact form can be obtained by defining


(
+δ, ϵk > ϵF
δk =
−δ, ϵk < ϵF

which yields
1
G(k, ω) =
ω − ϵk + iδk
and this is the form that we shall use in what follows.

7.7.6 Time-Ordering and Feynman Diagrams


Returning to the case of a non-interacting many-electron system in an external perturbing potential considered earlier,
we note that diagrams are constructed using one or more directed lines, that is lines with arrows pointing up or down,
representing particle and hole propagators, respectively. If we attempt to draw all possible diagrams of each order
(i.e. with given numbers of ‘vertex’ or interaction dots) systematically in the diagram series for the quasiparticle
propagator, then we obtain something like Figure 19, which show all possible first-order diagrams and five of the
possible second-order diagrams. This is known as the Goldstone method of drawing the diagrams, and they are all
topologically different from one another (in the Goldstone sense) in that one cannot be distorted into another without
changing the vertical ordering (i.e. time-ordering) of the vertices and initial/final points. Hence, all topologically-
different ways of joining n vertices corresponds to all the physically-different ways that the electron can propagate
through the system, scattering n times.

69
Figure 19: Beginning of diagram series for quasiparticle propagator in case of non-interacting many-electron system
in external potential.

The above Goldstone method of drawing the diagrams can be simplified considerably if, instead of associating
the retarded propagator G+ with a particle line and the advanced propagator G− with a hole line, we associate

the time-ordered (free) propagator, G0 = G+ 0 + G0 , with every line. Then, in the integrals over intermediate times
we automatically get G0 (t − t) = G0 when t > t (particle line) and G0 (t′ − t) = G−
′ + ′ ′
0 when t < t (hole line).
As a consequence, it is no longer necessary to draw any hole lines, and we can represent several diagrams via one
single diagram of just directed lines with arrow between vertices and initial/final points directed upwards, since the
time-ordered propagator automatically takes care of all possible time orderings of these vertices and initial/final
points. Hence, the diagram series in Figure 19 reduces to the much-simplified series for the time-ordered quasiparticle
propagator G in Figure 20, which is much easier to sum over than the original series. This method of drawing diagrams,
in which time ordering of the various components is not important and we use time-ordered propagators, is known as
the Feynman method, and we note that the single diagram with two vertices in Figure 20 is topologically equivalent
in the Feynman sense to any possible time-ordering of the two vertices and the initial/final points.

Figure 20: Simplified diagram series for quasiparticle propagator in case of non-interacting many-electron system in
external potential.

In the case of an interacting many-electron system with no external potential, the Feynman method simplifies
things considerably. For example, if we consider just the second-order case, then some of the many possible diagrams
that can be drawn using the Goldstone method (i.e. with time-ordering) are shown in Figure 21.

70
Figure 21: Some of the possible second-order diagrams drawn using Goldstone method for quasiparticle propagator in
case of interacting many-electron system.

If we use the Feynman method, however, whereby the full time-ordered propagator is associated with each line and
time order has no significance, things are simplified considerably and we have just six distinct second-order diagrams
(see Figure 22). The upper part of the Figure shows the way of drawing the diagrams that we have used so far,
whereas the lower part of the Figure shows them as some authors represent them (with curved rather than straight
interaction lines), although both methods of representation are topologically equivalent in the Feynman sense.

71
Figure 22: All possible second-order diagrams drawn using Feynman method for quasiparticle propagator in case of
interacting many-electron system.

We are now in a position to write down the Feynman rules for drawing and evaluating diagrams using the Feynman
method:
(1) For an nth-order diagram, draw n horizontal interaction lines terminating in vertices and two external points.
(2) Join all vertices and external points to one another with directed propagator lines in all possible topologically-
distinct ways (in the Feynman sense), with one line entering and one line leaving each vertex, and a line leaving one
external point and a second line entering another, to produce all the unique diagrams of that particular order. (Two
diagrams are topologically distinct if, when they are visualized as being made of rubber bands, one cannot be deformed
into the other.)
(3) Label each propagator line and vertex with a momentum (e.g. k or q) and a frequency (e.g. ω or ϵ), such
that the sum of momenta and frequencies entering each vertex is equal to the sum of those leaving. Eliminate all
‘anomalous’ diagrams that do not conserve momentum, i.e. those which have a particle and a hole in the same state.
(4) Evaluate all diagrams using the dictionary in Figure 23.
Rules (1)-(3) yield the diagram series shown in Figure 24).

72
Figure 23: Diagram dictionary for interacting many-electron system (Feynman method).

73
Figure 24: Feynman diagram series for quasiparticle propagator in case of interacting many-electron system, con-
structed using Feynman rules.

There are several key differences between the dictionary in Figure 23 for the Feynman method of evaluating
diagrams and Figure 16 for the Goldstone method. Firstly, we now use time-ordered propagators rather than re-
tarded/advanced propagators, and secondly, when integrating over intermediate frequencies for various lines in a
diagram, we avoid treating the non-propagating lines (i.e. the fermion loop and exchange ‘half-loop’) as a special case
when translating them into functions by including a convergence factor exp(iϵ0+ ) (where 0+ is a positive infinitesimal
such that 0+ × ∞ = ∞), i.e.

The integral over the intermediate frequency ϵ then yields (via residues):
Z ∞ +
(
dϵ ieiϵ0 −1, l > kF
=
−∞ 2π ω − ϵl + iδl 0, l < kF

Applied to the example of the (first-order) open-oyster diagram, we then have (as in the previous Section):

74
7.7.7 Dyson’s Equation and Self-Energy
Returning to the diagrammatic series for the full time-ordered quasiparticle propagator in Figure 24, we can see that
it consists of repeated simple parts of diagrams (such as bubbles, open oysters and pair bubbles) branching off from
the main directed propagator line. We now introduce several definitions relating to components of diagrams, which
will prove useful later.
A self-energy part is any part of a diagram without external propagator lines that can be inserted into a particle
or hole line, e.g.

whereas a proper (or irreducible) self-energy part is a part of a diagram which cannot be split into two unconnected
self-energy parts by cutting a particle or hole line, e.g.

Parts which can be split, such as diagrams (3) and (5) in the list above, are known as improper or reducible.
In the Hartree approximation we summed over the repeated self-energy part known as a ‘bubble’, while in the
Hartree-Fock approximation we summed over the part known as an open ‘oyster’. In general, it is possible to sum
over repetitions of all irreducible self-energy parts, which then accounts for the entire diagrammatic series, i.e.

75
which we can represent as

where the irreducible self-energy is defined via

Note that in the summation above we restricted the sum to just repeated proper parts, since if we had summed
over repeated improper parts too, some diagrams would have been counted more than once.
In algebraic terms, the summation of the quasiparticle propagator expansion is given by
1
G(k, ω) =
ω − ϵk − Σ(k, ω) + iδk

where the irreducible self-energy equates to −iΣ(k, ω).


If we express the second line of the propagator expansion above in abbreviated algebraic form, we may reorder the
factors and expand as follows:

G = G0 1 + G0 Σ + G20 Σ2 + . . . = G0 + G0 ΣG0 + G0 ΣG0 ΣG0 + . . .


 

which can be represented diagrammatically, and further manipulated, as follows

76
The diagrammatic equation

or its algebraic equivalent


G(k, ω) = G0 (k, ω) + G0 (k, ω)Σ(k, ω)G(k, ω)
is known as Dyson’s equation for the full time-ordered propagator, and can be solved for G(k, ω) to yield the algebraic
expression for the quasiparticle propagator given above.

7.7.8 Example: Imperfect Fermi Gas (Ladder Approximation)


To illustrate the approximate evaluation of the irreducible self-energy for a realistic physical system, we take the
example of a low-density system of fermions interacting via a short-range but strong (i.e. ‘hard core’) repulsive
interaction. We assume that the range of the interaction is a, and the average distance between the fermions is r0 ,
and by ‘low-density’ we mean that r0 ≪ 1 (i.e. r0 ≫ a). Such a theory can be applied in a qualitative way to nuclear
matter, where a/r0 ∼ 1/3, provided that we neglect the attractive part of the nuclear potential. The low-density
criterion can also be expressed in terms of the Fermi momentum, kF : since the fermion density can be expressed as
either n = 1/r0 3 or n = kF /3π 2 , so that 1/r0 ∼ kF , we require that kF a ≪ 1.
For a strong repulsive potential the first-order irreducible self-energy, determined from the lowest-order (i.e. bubble
and open-oyster) diagrams and corresponding to the Hartree-Fock approximation, is totally inadequate, and we must
consider higher-order diagrams. We shall use Feynman diagrams with no time-ordering, which can apply equally to
particle or hole propagation, but for the present case we shall focus on the particle part of the Feynman propagator.
Since n ∼ kF , low density corresponds to small kF (i.e. a small Fermi sphere), which means that in the low-density case
the contribution from hole lines in the Feynman diagrams will be very small compared with those from the particle
lines. The dominant diagrams in the series will therefore be those with the least number of hole lines, which is one
hole line. With this restriction, we find that the sum of diagrams containing just one hole line is as follows:

77
Such diagrams, which here represent repeated particle-particle interactions (i.e. two-particle scattering), are known
in general as ladder diagrams, and a partial summation of the diagrammatic series based exclusively on such diagrams
is known as the ladder approximation. Note that the second diagram in the bottom row is topologically equivalent
to the more usual form of second-order exchange diagram, but has been twisted to match the higher-order diagrams
which emphasize the multiple particle-particle interactions.
The sum of ladder diagrams to all orders (without the hole line!) is known as the T-matrix in the Feynman method
(N.B. In the original formulation of the ladder approximation, for the problem of nuclear matter, the Goldstone method
was used and the sum was known as the K-matrix ). We represent the T-matrix by

where the arrows at the corners are not propagators, but simply indicate where the external propagators attach.
We assume an infinite system, so that the summations over intermediate momenta can be replaced by integrals, which
allows us to combine the momentum and frequency into a four-momentum, e.g. p1 .
The T-matrix is an example of an effective two-particle interaction, −iΓ(p1 , p2 ; p3 , p4 ), which can interpreted as
a generalized scattering amplitude. The T-matrix (which we denote here by −iΓ(p1 , p2 ; p3 , p4 )) obeys the following
integral equation, which is similar to Dyson’s equation:

78
or in algebraic form:
Z
−4
Γ(p1 , p2 ; p3 , p4 ) = Vp1 −p3 + i(2π) d4 q Vq G0 (p1 − q)G0 (p2 + q)Γ(p1 − q, p2 + q; p3 , p4 )

This is known as the Bethe-Saltpeter equation (in the ladder approximation). If this equation is expanded using
perturbation theory, assuming V is small, we obtain the sum of all ladder diagrams in the T-matrix above.
We now turn to the evaluation of the self-energy for our modified ladder diagram series with an extra hole line.
The first-order and second-order contributions are calculated from the following diagrams

79
which, using the Feynman rules, translates to
Z Z
+ +
−iΣ(1) (p) = (−1)(2π)−4 d4 k (−iV0 )[iG0 (k)]eiωk 0 + (2π)−4 d4 k (−iVk−p )[iG0 (k)]eiωk 0
Z Z
−4 4 iωk 0+ −4 +
= i(2π) d k V0 [iG0 (k)]e − i(2π) d4 k Vk−p [iG0 (k)]eiωk 0
Z Z
+ +
=⇒ Σ(1) (p) = −(2π)−4 d4 k V0 [iG0 (k)]eiωk 0 + (2π)−4 d4 k Vk−p [iG0 (k)]eiωk 0
Z Z
(2) −8
−iΣ (p) = (−1)(2π) d k [iG0 (k)] d4 q (−iVq )[iG0 (p − q)][iG0 (k + q)](−iV−q )
4

Z Z
+ (2π)−8 d4 k [iG0 (k)] d4 q (−iVq )[iG0 (p + q)][iG0 (k − q)](−iVk−q−p )
Z Z
−8
= −i(2π) d k G0 (k) d4 q Vq G0 (p − q)G0 (k + q)V−q
4

Z Z
+ i(2π)−8 d4 k G0 (k) d4 q Vq G0 (p + q)G0 (k − q)Vk−q−p
Z Z
(2) −8
=⇒ Σ (p) = (2π) d k G0 (k) d4 q Vq G0 (p − q)G0 (k + q)V−q
4

Z Z
− (2π)−8 d4 k G0 (k) d4 q Vq G0 (p + q)G0 (k − q)Vk−q−p

This process can be extended to higher orders, and it is found that the self-energy can be encapsulated by the
following sum of terms based on the K-matrix with the addition of a single hole line:

which can be expressed algebraically as


Z Z
−iΣ(p) = (−1)(2π)−4 d4 k G0 (k)Γ(p, k; p, k) + (2π)−4 d4 k G0 (k)Γ(k, p; p, k)
Z Z
=⇒ Σ(p) = −i(2π)−4 d4 k G0 (k)Γ(p, k; p, k) + i(2π)−4 d4 k G0 (k)Γ(k, p; p, k)

where the factor of −1 in front of the first term comes from the fermion loop, and we have defined the algebraic form
of the K-matrix via
Z
Γ(p1 , p2 ; p3 , p4 ) ≡ Vp1 −p3 + i(2π)−4 d4 q Vq G0 (p1 − q)G0 (p2 + q)Vp1 −q−p3 + . . .
Z
−4
=⇒ −iΓ(p1 , p2 ; p3 , p4 ) = −iVp1 −p3 + (2π) d4 q Vq G0 (p1 − q)G0 (p2 + q)Vp1 −q−p3 + . . .

which, it can be seen, is equivalent to the diagrammatic ladder series from earlier with with the addition of a single
hole line joining a pair of ingoing and outgoing momenta.
The problem of calculating the irreducible self-energy therefore reduces to determining Γ, which must be obtained
by solving the Bethe-Saltpeter equation - see Fetter and Walecka, pp.137-147, based on Galitskii (1958).

7.8 High-density Electron Gas


7.8.1 Random Phase Approximation
We now apply the Feynman diagram technique to the model system of a high-density (degenerate) gas of N electrons
moving against a ‘smeared-out’ positively-charged background of ions to maintain overall charge neutrality (the so-
called jellium model). At zero temperature, such a gas is characterized by a single parameter, rs , which approximates

80
to the average distance between electrons. In the high-density limit, which corresponds to 0 < rs < 1, the kinetic
energy of the electrons is much higher than their mutual potential energy, so we are justified in treating the latter as
a relatively small perturbation.
The Hamiltonian for the system described above is given by
X †
Vklmn ĉ†l ĉ†k ĉm ĉn + Ĥions + Ĥel-ion
X
Ĥ = ϵk ĉk ĉk + 21
k k,l,m,n

where we have used electron operators, rather than particle/hole operators, for clarity, and

4πe2
Vklmn ≡ Vq =
Ωq 2

is the repulsive inter-electron Coulomb potential expressed in momentum space (with Ω the system volume). It can
be shown (see Mattuck, pp.186-7) that the q = 0 part of Vq cancels out the last two terms in the Hamiltonian, which
allows us to write X †
Vq ĉ†n−q ĉ†m+q ĉm ĉn + Ĥions + Ĥel-ion
X
Ĥ = ϵk ĉk ĉk + 21
k q(̸=0),m,n

We now consider how the Feynman diagrams of increasing order in the interaction contribute to the self-energy
of a particle (or hole) in the system. A typical irreducible self-energy part is the following second-order pair-bubble
diagram, which can be evaluated as

If we take the limit of infinite volume, the sum over q changes to an integral as follows:
X Z Z
→ d q ∝ q 2 dq
3

and we can see that as q → 0, this diverges due to the factor of q 4 in the denominator. The fact that the second-order
contribution is divergent means that it is insufficient to truncate the perturbation expansion at second order, since
the energy of the electron gas would then be infinite which clearly does not correspond to reality. A look at diagrams
of other orders in fact reveals that most of the diagrams diverge, although as we shall see, this is not necessarily a
problem.
Among the diagrams for a particular order of perturbation, the most divergent diagram is the most important one.
Our approach will therefore be to classify the diagrams at each order in the interaction according to their degree of
divergence (i.e. the number of factors of q in the denominator of the integrand), select the most divergent diagram at
each order in the interaction, and sum only these most divergent diagrams.
Finally, we note that each term in the self-energy, represented by a particular diagram, is proportional to some
power of the small parameter rs , In fact, it can be shown (see Mattuck, p.188) that any term of order n in the
interaction ∼ rsn−2 .
We now have all the information we need to arrange the diagrams in the perturbation series according to both
degree of divergence and dependence on rs - see Figure 25. (Note that the first-order bubble diagram makes no

81
contribution to the self-energy since it has q = 0; in fact it is this diagram that is responsible for cancelling the
positive background.)

Figure 25: Contributions to self-energy of high-density electron gas of increasing orders, arranged according to degree
of divergence and dependence on rs .

In the Figure, the degree of divergence increases along the diagonals towards lower right, and taking into account
the above dependence of a diagram on rsn−2 , we can see that given two diagrams with the same degree of divergence
(i.e. along a diagonal), as rs → 0 in the high-density limit, the diagram of lower order in the interaction makes a
much larger contribution to the self-energy. Hence, the most important diagrams we need to consider are those of
lowest order in rs for each degree of divergence, i.e. the diagrams in the left-hand column of Figure 25. These consist
of one or more particle-hole pair-bubbles linked together in a ring, and are therefore known as ring diagrams. The
self-energy is then given by the partial sum

i.e. the sum over all diagrams of the repeated pair-bubble or ring type. This approximation for the self-energy is
known as the random phase approximation (RPA), and amazingly, this sum over an infinite number of infinite terms
can be carried out, and gives a finite result!
The sum over ring diagrams can be simplified by factoring out a free propagator as follows:

82
where the double interaction line represents the ‘effective interaction’ in the RPA, which can be interpreted as a
‘screened’ interaction between two particles and is given by

Diagrams such as (1)-(3) in the above are known as polarization insertions since they represent all the ways in
which the interaction causes the medium (in this case, the electron gas) to become virtually polarized via the formation
of particle-hole pairs, which form ‘virtual dipoles’. In fact, we can represent the above summation algebraically as

RPA Vq Vq
Veff (q, ω) = ≡ RPA
1 + Vq Π0 (q, ω) ε (q, ω)

where

defines a proper or irreducible polarization part, Π0 (q, ω).


RPA
The above expression for Veff has the form of an interaction between two charges in a dielectric, with

εRPA (q, ω) ≡ 1 + Vq Π0 (q, ω)

being the so-called generalized dielectric constant. The reason for this is that the dielectric properties of a medium arise
RPA
from the polarization of the medium by a field, and −iVeff is just the sum of diagrams representing the polarization
of the electron gas by the field of one of the electrons in the gas itself. Note that unlike the bare interaction Vq , the
RPA
effective interaction Veff (q, ω) depends on ω, so that it the latter is transformed to (q, t)-space, it will therefore
become a time-dependent interaction, a fact that is due to the inertia of the polarization charge. To calculate the
RPA
self-energy of a particle (or hole) in an electron gas within the RPA, we therefore need to determine Veff , which in
turn requires a knowledge of Π0 (q, ω) and Vq .

83
7.8.2 Evaluation of Π0 (q, ω)
The polarization part, Π0 (q, ω), is the simplest example of a polarization propagator, and we can evaluate it using the
Feynman rules as follows
d3 k
Z Z

−iΠ0 (q, ω) = 2 × (−1) 3
[iG0 (k + q, ϵ + ω)][iG0 (k, ϵ)]
(2π) 2π
d3 k
Z Z
dϵ i i
= −2 3
×
(2π) 2π ϵ + ω − ϵk+q + iδk+q ϵ − ϵk + iδk
where the factor of 2 comes from the sum over spins and the (−1) from the fermion loop, and we have assumed an
infinite volume so that the sum over momentum changes to an integral. (Note that this differs from the definition in
Fetter and Walecka by a factor of −1.)
The momentum integral above can actually be split into four parts, corresponding to whether |k| and/or |k + q|
are above/below the Fermi level. (This is a consequence of the fact that Feynman diagrams are not time-ordered,
so that the propagator lines can represent either particles or holes.) We shall consider the four cases separately, and
then evaluate the integral over the frequency, ϵ, for each of these. Although we are only interested in the value of
the integral along the entire real axis (i.e. for −∞ < ϵ < ∞), it is easiest to formulate it as a contour integral in the
ϵ-plane and for a semi-circular contour, since we can then use the residue theorem to help evaluate it.
The first case we shall consider is |k| > kF , |k + q| < kF , so that
Z ∞
d3 k
Z
dϵ i i
−iΠ0 (q, ω)(1) = −2 3
(2π) −∞ 2π ϵ + ω − ϵk+q − iδ ϵ − ϵk + iδ
|k|>kF ,
|k+q|<kF

The integral we need to evaluate is therefore


Z ∞
dϵ i i
−∞ 2π ϵ + ω − ϵk+q − iδ ϵ − ϵk + iδ
where the integrand has poles at ϵ = ϵk − iδ and ϵ = −ω + ϵk+q + iδ. We choose as our contour the following:

From the residue theorem, we know that


I m
X
f (z) dz = 2πi Res [f (z); z = zk ]
C k=1

where the residue for a single pole at z = z0 is defined via


Res [f (z); z = z0 ] = lim (z − z0 )f (z)
z→z0

Applying this to our case, we find that


I  
dϵ i i 1 i i
= 2πi Res ; ϵ = −ω + ϵk+q + iδ
C 2π ϵ + ω − ϵk+q − iδ ϵ − ϵk + iδ 2π ϵ + ω − ϵk+q − iδ ϵ − ϵk + iδ
 
i i
= 2πi lim
ϵ→−ω+ϵk+q +iδ 2π ϵ − ϵk + iδ
i
=
ω − ϵk+q + ϵk − iδ

84
where we have assumed that 2iδ = iδ, since δ is infinitesimally small.
We can also express the contour integral as the sum of the integral along the straight part of the contour and the
integral around the semicircular part. Making a change of variables in the latter using ϵ = Reiθ (so that dϵ = iReiθ dθ),
we can then write
I
dϵ i i
=
C 2π ϵ + ω − ϵk+q − iδ ϵ − ϵk + iδ
Z R Z π
dϵ i i dθ i i
+ iReiθ iθ iθ − ϵ + iδ
−R 2π ϵ + ω − ϵ k+q − iδ ϵ − ϵ k + iδ 0 2π Re + ω − ϵ k+q − iδ Re k

In the limit R → ∞, the second term on the RHS, which is proportional to 1/R, will go to zero, while the limits on
the first term on the RHS will go to ±∞. We recognize this latter term as the integral we set out to evaluate, so
equating to the other expression for the contour integral obtained via the residue theorem above, we have
Z ∞
dϵ i i i
=
−∞ 2π ϵ + ω − ϵ k+q − iδ ϵ − ϵk + iδ ω − ϵ k+q + ϵk − iδ

The next case we shall consider is |k| < kF , |k + q| > kF , so that


Z ∞
d3 k
Z
dϵ i i
−iΠ0 (q, ω)(2) = −2
(2π)3 −∞ 2π ϵ + ω − ϵk+q + iδ ϵ − ϵk − iδ
|k|<kF ,
|k+q|>kF

The integral we need to evaluate is therefore


Z ∞
dϵ i i
−∞ 2π ϵ + ω − ϵk+q + iδ ϵ − ϵk − iδ

where the integrand has poles at ϵ = ϵk + iδ and ϵ = −ω + ϵk+q − iδ. We choose as our contour the following:

Applying the residue theorem (with a minus sign on the RHS due to the fact that we are traversing the contour in
the clockwise direction) now yields:
I  
dϵ i i 1 i i
= −2πi Res ; ϵ = −ω + ϵk+q − iδ
C 2π ϵ + ω − ϵk+q + iδ ϵ − ϵk − iδ 2π ϵ + ω − ϵk+q + iδ ϵ − ϵk − iδ
 
i i
= −2πi lim
ϵ→−ω+ϵk+q −iδ 2π ϵ − ϵk − iδ
i
=−
ω − ϵk+q + ϵk + iδ

The final two cases to consider are |k|, |k + q| > kF and |k|, |k + q| < kF . Since we now have both poles either
above/below the real axis, we can choose the contour in the opposite half-plane so that it does not enclose any
poles, which means that the contour integral (and hence the frequency integral we are interested in) will be zero and
therefore makes no contribution to the total integral. (These cases correspond to propagation of two electrons/holes,
respectively, which is unphysical so the result is to be expected.)

85
Combining the two non-zero frequency integrals now enables us to express the polarization propagator as

d3 k
Z Z
dϵ i i
−iΠ0 (q, ω) = −2 3
×
(2π) 2π ϵ + ω − ϵk+q + iδk+q ϵ − ϵk + iδk
3
d3 k
Z Z
d k 1 1
= 2i 3
− 2i
(2π) ω − ϵk+q + ϵk + iδ (2π)3 ω − ϵk+q + ϵk − iδ
|k|<kF , |k|>kF ,
|k+q|>kF |k+q|<kF

or equivalently as

d3 k d3 k
Z Z
1 1
Π0 (q, ω) = −2 +2
(2π)3 ω − ϵk+q + ϵk + iδ (2π)3 ω − ϵk+q + ϵk − iδ
|k|<kF , |k|>kF ,
|k+q|>kF |k+q|<kF

We can make the region of integration of the second term the same as that of the first as follows. Firstly, we let
k + q = m, so that
d3 k d3 m
Z Z
1 1
=
(2π)3 ω − ϵk+q + ϵk − iδ (2π)3 ω − ϵm + ϵm−q − iδ
|k|>kF , |m−q|>kF ,
|k+q|<kF |m|<kF

Secondly, we let m = −k (with d3 m = d3 k), so that

d3 m d3 k
Z Z
1 1
3
= 3
(2π) ω − ϵm + ϵm−q − iδ (2π) ω − ϵ−k + ϵ−k−q − iδ
|m−q|>kF , |−k−q|>kF ,
|m|<kF |−k|<kF

However, since | − k| = |k| (with ϵ−k = ϵk ) and | − k − q| = |k + q| (with ϵ−k−q = ϵk+q ), the above becomes

d3 k d3 k
Z Z
1 1
=
(2π)3 ω − ϵ−k + ϵ−k−q − iδ (2π)3 ω − ϵk + ϵk+q − iδ
|−k−q|>kF , |k|<kF ,
|−k|<kF |k+q|>kF

which allows us to simplify the expression for Π0 (q, ω) to

d3 k
Z  
1 1
Π0 (q, ω) = −2 −
(2π)3 ω − ϵk+q + ϵk + iδ ω − ϵk + ϵk+q − iδ
|k|<kF ,
|k+q|>kF

Although it is possible to evaluate Π0 (q, ω) for arbitrary q, ω, the derivation is very involved (see Fetter and
Walecka, pp.158-163), and in what follows we shall only consider the limit q ≪ kF , which simplifies things considerably.
We proceed by expanding the integral over k as follows:
Z 2π Z 1 Z  
2 2 1 1
Π0 (q, ω) = − dϕ d(cos θ) dk k 1 − 1
(2π)3 0 −1 ω − 2m [2kq cos θ + q 2 ] + iδ ω + 2m [2kq cos θ + q 2 ] − iδ
|k|<kF ,
|k+q|>kF

where we have used the fact that d(cos θ) = − sin θ dθ, and substituted ϵk = k 2 /2m, etc. If we use the vector q to
define the z-direction, then we have x = cos θ, and we can define the region of integration for k as the following shaded
area:

86
where the Figure is cylindrically symmetric about the z-axis. The upper condition on the integral over k is simple,
and translates to k < kF . The lower condition requires a bit more thought, however. Since |k + q| has to be greater
than kF , the absolute minimum value k can take is kF − q, which corresponds to k pointing in the positive z-direction;
however, as θ increases, this minimum value varies towards kF , and in general is given by kF − q cos θ = kF − qx.
Hence, the lower condition translates to k > kF − qx. (The reason why the shaded area is confined to the top half of
the Fermi sphere is that for the condition |k + q| > kF to be satisfied in the bottom half, we would require |k| > kF ,
which is however forbidden by the upper condition on the integral.) The integral then becomes
Z 1 Z kF  
1 1 1
Π0 (q, ω) = − 2 dx dk k 2 1 − 1
2π 0 kF −qx ω − 2m [2kqx + q 2 ] + iδ ω + 2m [2kqx + q 2 ] − iδ
where we have evaluated the integral over ϕ and reset the limits on the integral over x to account for the fact that we
only need to consider the upper hemisphere of the Fermi sphere (i.e. 0 < θ < π/2).
If we now consider the aforementioned limit q ≪ kF , then this means that k will be restricted to just a thin shell
near the Fermi level, so to a first level of approximation we are justified in replacing k with kF q in the integrand and
setting q 2 to zero, which yields
Z 1 Z kF ( )
1 2 1 1
Π0 (q ≪ kF , ω) ≃ − 2 dx dk kF −
2π 0 kF −qx ω − kFmqx + iδ ω + kFmqx − iδ
Z 1 ( )
1 2 1 1
≃− 2 dx qxkF −
2π 0 ω − kFmqx + iδ ω + kFmqx − iδ

Defining ζ ≡ mω/kF q, this reduces to


Z 1  
mkF 1 1
Π0 (q, ω) ≃ − dx x −
2π 2 0 ζ − x + iδ ζ + x − iδ
Using the following identity from the theory of complex functions:
Z Z Z
f (x) f (x)
dx = P dx − iπ dx f (x)δ(x)
x + iδ x
we can rewrite the first part of the integral above as
Z 1 Z 1 Z 1
x x
dx =P dx − iπ dx xδ(ζ − x)
0 ζ − x + iδ 0 ζ −x 0

the real part of which is given by


Z 1 Z 1  
x ζ 1 1 − ζ
P dx = −P dx 1 + = [−x − ζ ln |x − ζ|]0 = −1 − ζ ln
0 ζ −x 0 x−ζ −ζ
while the imaginary part is given by
(
1
−iπζ,
Z
for 0 < ζ < 1
−iπ dx xδ(ζ − x) =
0 0, otherwise

87
whereas we can rewrite the second part of the integral above as
Z 1 Z 1 Z 1
x x
dx =P dx − iπ dx xδ(ζ + x)
0 ζ + x − iδ 0 ζ +x 0

the real part of which is given by


Z 1 Z 1  
x ζ 1 1 + ζ
P dx =P dx 1 − = [x − ζ ln |x + ζ|]0 = 1 − ζ ln
0 ζ +x 0 x+ζ ζ

while the imaginary part is given by


(
1
for − 1 < ζ < 0
Z
iπ|ζ|,
−iπ dx xδ(ζ + x) =
0 0, otherwise

Combining the two parts then yields


 
mkF ζ 1 + ζ π
Π0 (q ≪ kF , ω) ≃ 2 1 − ln + i |ζ|θ(1 − |ζ|)
π 2 1−ζ 2

7.8.3 Static Screening Limit


Returning to the simplified expression for Π0 (q, ω), namely

d3 k
Z  
1 1
Π0 (q, ω) = −2 −
(2π)3 ω − ϵk+q + ϵk + iδ ω − ϵk + ϵk+q − iδ
|k|<kF ,
|k+q|>kF

we can evaluate this for arbitrary q and ω = 0, where the latter condition corresponds to static screening (i.e. a
time-independent interaction between electrons). Applying this condition, and imposing the conditions on the integral
by means of step functions, we have

d3 k
Z  
1 1
Π0 (q, 0) = −2 θ(kF − |k|)[1 − θ(kF − |k + q|)] −
(2π)3 −ϵk+q + ϵk + iδ −ϵk + ϵk+q − iδ
Z 3
d k 1
= −4 θ(kF − |k|)[1 − θ(kF − |k + q|)]
(2π)3 ϵk − ϵk+q + iδ

Using the identity Z Z Z


f (x) f (x)
dx =P dx − iπ dx f (x)δ(x)
x + iδ x
then yields

d3 k
Z
1
Re Π0 (q, 0) = −4P θ(kF − |k|)[1 − θ(kF − |k + q|)]
(2π)3 ϵk − ϵk+q
d3 k
Z
Im Π0 (q, 0) = 4π θ(kF − |k|)[1 − θ(kF − |k + q|)]δ(ϵk − ϵk+q )
(2π)3

The imaginary part is identically zero, since it is impossible for ϵk to be equal to ϵk+q , since q ̸= 0. Expanding the
real part, we have

d3 k θ(kF − |k|) d3 k θ(kF − |k|)θ(kF − |k + q|)


Z Z
Re Π0 (q, 0) = −4P 3
− 4P
(2π) ϵk − ϵk+q (2π)3 ϵk − ϵk+q

Focusing on the second term, if we make the change of variables k′ = −(k + q), this leads to
Z 3 ′
d k θ(kF − |k′ + q|)θ(kF − |k′ |)
Z 3 ′
d k θ(kF − |k′ + q|)θ(kF − |k′ |)
Z 3 ′
d k θ(kF − |k′ + q|)θ(kF − |k′ |)
3
= 3
= −
(2π) ϵ−k′ −q − ϵ−k′ (2π) ϵk′ +q − ϵk′ (2π)3 ϵk′ − ϵk′ +q

88
which is just minus the contribution from +(k + q), and hence makes the second term zero. We are therefore left with
Z 2π Z 1 Z kF
d3 k θ(kF − |k|) k2
Z
4
Π0 (q, 0) = −4 = − dϕ d(cos θ) dk k +2kq cos θ+q 2
2
(2π)3 ϵk − ϵk+q (2π)3 0 k 2
−1 0 2m − 2m
m 1
Z kF
m kF
Z kF
k2
Z Z
k 2
1 m k + q/2
= 2 dx dk = 2 dk ln |kqx + q /2| 0 = 2 dk k ln
π −1 0 kqx + q 2 /2 π 0 q π q 0 k − q/2

where in the last step we have used the fact that ln | − x| = ln |x|. Splitting the logarithm into two terms and making
use of the standard integral
x 2 − b2
Z
1
dx x ln |x + b| = ln |x + b| − (x − b)2
2 4
we arrive at the final expression

q2
   
mkF kF 1 + q/2kF
Π0 (q, 0) = 1+ 1 − 2 ln

2π 2 q 4kF 1 − q/2kF

or, alternatively:
mkF 1 1 − x2 1 + x
 
q
Π0 (q, 0) = + ln 1 − x , x≡
2π 2 2 4x 2kF
Using the expression for the dielectric function in the RPA:

εRPA (q, ω) = 1 + Vq Π0 (q, ω)

with the momentum-space Coulomb potential, Vq = 4πe2 /q 2 , we obtain the so-called Lindhard dielectric function

4me2 kF 1 1 − x2 1 + x
 
RPA q
ε (q, 0) = 1 + 2
+ ln , x≡
πq 2 4x 1−x 2kF

In the Thomas-Fermi model, the above dielectric function is replaced by its value in the long-wavelength limit, i.e.

εTF (q, 0) = lim εRPA (q, 0)


q→0

Using the fact that


lim ln |1 + x| = x
x→0

so that
1 1 − x2 1 + x
 
lim + ln =1
x→0 2 4x 1 − x
we find that
4me2 kF
εTF (q, 0) =
πq 2
and hence the effective potential in this approximation is given by

TF Vq Vq 4πe2 4πe2
Veff (q) = = TF = 2 = 2
1 + Vq Π0 (q, 0) ε (q, 0) q (1 + 4me2 kF /πq 2 ) q 2 + qTF

where 1/2
4me2 kF

qTF ≡
π
is known as the Thomas-Fermi screening wavenumber. Notably, the above effective potential has the same form as
the Yukawa potential in momentum space, but with the physical quantity qTF replacing λ, which was introduced for
purely mathematical convenience. Hence, in real space the effective potential in the Thomas-Fermi approximation can
be expressed as
e2
TF
Veff (r) = e−qTF r
r
which is known as the screened Coulomb potential.

89
7.8.4 Self-Energy in RPA
We are now in a position to consider the self-energy of a particle (or hole) in an electron gas within the random phase
approximation, which is approximated by the following diagram

RPA
where the double interaction line represents the RPA effective interaction, Veff . Evaluating the diagram (in the
limit of infinite volume), we obtain

d3 q
Z Z
dγ +
−iΣRPA (k, ω) ≃ 3
RPA
[−iVeff (q, γ)][iG0 (k − q, ω − γ)]ei(ω−γ)0
(2π) 2π
+
3
4πe2 ei(ω−γ)0
Z Z
d q dγ
≃ ×
(2π)3 2π εRPA (q, γ) ω − γ − ϵk−q + iδk−q

Hence, despite the fact that all the diagrams (with the exception of the open oyster) contributing the sum are infinite,
the self-energy is finite, as shown by the fact that the q 2 in the integral cancels with the q 2 in the denominator. This
is due to the fact that unlike the bare momentum-space Coulomb potential, which diverges as q → 0, the effective
interaction, represented by the Thomas-Fermi potential as q → 0, remains finite in this limit due to the additional
term qTF in the denominator.
The expression for the self-energy in this limit becomes

d3 q 4πe2
Z  Z
TF dγ +
Σ (k, ω) ≃ 2 iG0 (k − q, ω − γ)ei(ω−γ)0
(2π)3 q 2 + qTF 2π
Z 3 ′
d3 q 4πe2 4πe2
Z  
d k
≃ 2 (−1) ≃ − 2
(2π)3 q 2 + qTF (2π)3 (k − k′ )2 + qTF
|k−q|<kF |k′ |<kF

In the Thomas-Fermi limit of the RPA, we therefore have a quasiparticle with energy
Z 3 ′
′ k2 d k 4πe2
ϵk = − 2
2m (2π)3 (k − k′ )2 + qTF
|k′ |<kF

7.8.5 General Effective Interaction


In the RPA we only considered the contribution of pair-bubble polarization insertions to the effective interaction,
and it is possible to generalize the concept of an effective interaction to include all possible polarization insertions as
follows:

We can sum this series in the same way that we summed the series for the quasiparticle propagator or the self-energy
earlier, in order to yield a Dyson-like equation.
As in the case of the self-energy, we begin by defining a polarization part as any diagram without external interaction
lines which may be inserted into an interaction line, e.g.

90
whereas we define a proper (or irreducible) polarization part as a polarization part which cannot be split into two
unconnected polarization parts by removing an interaction line, e.g.

Hence, diagrams (2) and (5) above are reducible polarization parts, whereas the other four are irreducible. (Note
that diagram (3) is an example of an exchange diagram, which is ignored in the RPA since in that approximation we
assume that the exchange interaction is much smaller than the direct interaction.)
We represent the sum over irreducible polarization parts by

from which, by substituting into the expansion for −iVeff and factoring, we can obtain the Dyson-like equation

which translates to the following algebraic form in (q, ω)-space:

Vq Vq
Veff (q, ω) = ≡
1 + Vq Π(q, ω) ε(q, ω)

Note that this is just a generalization of the RPA result derived earlier.
We can use the effective interaction to simplify considerably the appearance of the expansion for the irreducible
self-energy. For example, consider the second-order exchange diagram, which may be drawn in several equivalent ways:

91
In the expansion for the irreducible self-energy, there will be a subset of diagrams the same as the above, except
for the fact that one of the interaction lines in each (say the upper one) has been replaced by a polarization insertion,
and we can collect this subset in brackets as follows:

Within this subset, we can then factor out the part of the diagram that remains unchanged, namely the lower part,
to obtain

where, in the second line, we have used the expansion for the effective interaction in terms of polarization parts
presented at the very start of this Section. Similarly, we can sum over all possible polarization insertions in the lower
interaction line, and hence replace the bare interaction with the effective interaction in this lower part. In fact, we can
generalize this procedure and sum over all possible subsets in the series by replacing the bare interaction lines with
effective interactions in the first diagram of each subset to allow for the inclusion of polarization insertions, i.e.

92
which leads to the following greatly-simplified form of the expansion for the irreducible self-energy:

so that we now have a series in which no interaction lines contain any polarization parts, and bare interaction lines
have become ‘dressed’ to become effective interactions.
It is important to note that when drawing such diagrams with effective interactions, no polarization parts can
be included, since these have already been accounted for in the sum over the subset of diagrams in which the bare
interaction line was replaced with the effective interaction. Hence, a diagram such as

is strictly forbidden, since these diagrams have already been included in

93

You might also like