Lecture 5: The Non-Equilibrium Green Function

Method
Joachim Keller and Mark Jarrell
April 26, 2011
Contents
1 Review of Equilibrium Green function formalism 3
1.1 Equilibrium Green functions . . . . . . . . . . . . . . . . 3
1.2 Perturbation theory for equilibrium Green functions . . . 4
2 Introduction to Non-Equilibrium Green Function For-
malism 8
2.1 Green functions with time dependent perturbation . . . . 8
2.2 Dyson equation . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Langreth rules, Keldysh formalism . . . . . . . . . . . . 11
3 Kinetic equations: The Boltzmann Equation 15
3.1 The Conventional Derivation . . . . . . . . . . . . . . . . 15
3.2 Towards Transport in the Green Function Approach . . . 17
1
3.3 Derivation of kinetic equations . . . . . . . . . . . . . . . 19
3.4 Gradient expansion . . . . . . . . . . . . . . . . . . . . . 22
3.5 Self energy from impurity scattering . . . . . . . . . . . . 25
2
1 Review of Equilibrium Green function formal-
ism
In this section, we will review the Equilibrium Green function formalism
and use it to introduce some of the nomenclature which we will require
for the non-equilibrium formalism. We will introduce a time-ordered
Green function on the special contour C
i
. This will the basis for the
following discussion of non-equilibrium Green functions.
1.1 Equilibrium Green functions
In a non-equilibrium theory the distribution function f will become
an independent quantity. Therefore, in addition to the retarded and
advanced Green functions, we need the correlation functions
G
>
(x
1
, t
1
; x
2
, t
2
) = i(x
1
, t
1
)

(x
2
, t
2
)) (1)
G
<
(x
1
, t
1
; x
2
, t
2
) = +i

(x
2
, t
2
)(x
1
, t
1
)) (2)
They are related to the time-ordered Green function
G
c
(x
1
, t
1
;x
2
, t
2
) = iT
t
(x
1
, t
1
)

(x
2
, t
2
))
= i(t
1
t
2
)(x
1
, t
1
)

(x
2
, t
2
)) +i(t
2
t
1
)

(x
2
, t
2
)(x
1
, t
1
))
= (t
1
t
2
)(G
>
(x
1
, t
1
; x
2
, t
2
) + (t
2
t
1
)G
<
(x
1
, t
1
; x
2
, t
2
)
Note that the retarded Green function can also be expressed by the two
correlation functions
G
r
(x
1
, t
1
; x
2
, t
2
) = (t
1
t
2
)(G
>
(x
1
, t
1
; x
2
, t
2
) G
<
(x
1
, t
1
; x
2
, t
2
)) (3)
3
G
a
(x
1
, t
1
; x
2
, t
2
) = (t
2
t
1
)(G
>
(x
1
, t
1
; x
2
, t
2
)G
<
(x
1
, t
1
; x
2
, t
2
)) (4)
and G
r
G
a
= G
>
G
<
.
In thermal equilibrium these functions depend only on the time-
dierence t = t
1
t
2
. Introducing a Fourier transform (here we suppress
the dependence on variables x
1
, x
2
for the moment)
G() =
_
+

e
it
G(t)dt (5)
one obtains the important relation between the two correlation func-
tions
G
>
() = e

G
<
() (6)
and the spectral function
A() = i[G
r
() G
a
()] = i[G
>
() G
<
()] (7)
we nd
G
<
() = if()A(), G
>
= i(1 f())A() (8)
where f() = 1/ (exp() + 1) is the Fermi function.
1.2 Perturbation theory for equilibrium Green functions
In order to calculate the Green function with help of a perturbation
theory we split the Hamiltonian into H = H
0
+ V where H
0
describes
a non-interacting electron system. Going over to the interaction repre-
sentation the unitary operator for the time evolution between times t
0
4
t
0
t
t
t
t
t
1
1
2
2
0
t
t
t
t
1
1
2
2
<
>
Figure 1: The contour C used for the perturbation theory in the time-evolution
and t becomes
e
iH(tt
0
)
= e
itH
0
S(t, t
0
)e
iH
0
t
0
, S(t, t
0
) = T exp
_
i
_
t
t
0
V
I
(t
/
)dt
/
_
(9)
where O
I
(t) = exp(iH
0
t)Oexp(iH
0
t). Here the time t
0
is some arbi-
trary reference time. We can set t
0
= 0. However, for later use in the
case of a time dependent interaction it has to be chosen earlier than
the switch-on time of the time-dependent interaction (i.e., t
0
).
For xed time t
1
, t
2
we then obtain
(t
1
)

(t
2
)) = e
iH
0
t
0
S(t
0
, t
1
)
I
(t
1
)S(t
1
, t
2
)

I
(t
2
)S(t
2
, t
0
)e
iH
0
t
0
)
(10)
which can also be written as
(t
1
)

(t
2
)) = e
iH
0
t
0
T
C
S
C

I
(t
1
)

I
(t
2
)e
iH
0
t
0
), S
C
= T exp
_
i
_
C
V
I
()d
_
(11)
where C is the contour shown in Fig.1 and the time ordering operator
T orders the interaction operators contained in S in the direction of
the contour at the right places before, between, and behind the fermion
operators at the xed times t
1
, t
2
.
5
t t
t
t
1 0
2
0
- i
Figure 2: The contour C
i
used for the perturbation theory including expansion of the
statistical operator
Finally we have to expand also the statistical operator, which can
be expressed by
e
H
= e
i(t
0
i)H
0
S(t
0
i, t
0
)e
it
0
H
0
(12)
Then
T
C
i (t
1
)

(t
2
)) =
T
C
i S
C
i
I
(t
1
)

I
(t
2
))
0
T
C
i S
C
i )
0
(13)
where now C
i
is the contour shown in Fig. 2 and )
0
means a thermal
average performed with the statistical operator
0
= e
H
0
/Z
0
.
The time ordering contained in S means that a perturbation theory
for the Green function can be formulated easily only for a Green func-
tion with time-ordered operators. Therefore we dene a time ordered
6
t
t
0
0
- i
C
C
1
2
Figure 3: Contour C
i
used for the perturbation expansion of the time-ordered Green
function
Green function on the contour C
i
:
G(
1
,
2
) =
_

_
i(
1
)

(
2
)) for
1
>
c

2
+i

(
2
)(
1
)) for
1
<
c

2
(14)
where we distinguish between times
i
on dierent parts of the contour
(see Fig. 3).
Then for G the following perturbation expansion holds:
G(
1
,
2
) = i
T
C
i S
C
i
I
(
1
)

I
(
2
))
0
T
C
i S
C
i )
0
(15)
It is related to the real-time Green functions introduced above by
G(
1
,
2
) =
_

_
G
c
(t
1
, t
2
) for
1
,
2
C
1
G
>
(t
1
, t
2
) for
1
C
2
,
2
C
1
G
<
(t
1
, t
2
) for
1
C
1
,
2
C
2
G
c
(t
1
, t
2
) for
1
,
2
C
2
(16)
7
The last being an antitime-ordered Green function. The retarded and
advanced Green functions can be obtained from G
>
and G
<
in the usual
way.
In thermal equilibrium it is possible to simplify this contour consid-
erably: As in thermal equilibrium all Green functions depend only on
the time dierence we can replace C
i
by a contour C
th
which runs from
t
0
to t
0
i, where we can set the real times two zero without loss
of generality. One then denes a time-ordered Green function on the
imaginary axis. The physical retarded and advanced Green functions
are then obtained by an analytical continuation in frequency space.
This is the Matsubara technique.
2 Introduction to Non-Equilibrium Green Func-
tion Formalism
2.1 Green functions with time dependent perturbation
A non-equilibrium situation is obtained if we add to the Hamiltonian H
of the system a time-dependent perturbation H
/
t
describing for instance
the interaction with an external time-dependent eld:
H
t
= H +H
/
t
(17)
with H = H
0
+ V . We assume that the time dependent interaction
H
/
t
vanishes for times t < t
0
. The time dependent interaction can be
treated by a perturbation expansion in a similar way as the internal
8
Figure 4: Contours C
i
and C used for the double perturbation expansion of the
non-equilibrium time-ordered Green function
interaction. The nal result for the Green function can be written as a
double expansion:
G(
1
,
2
) = i
T
C
i S
C
i S
C

0
(
1
)

0
(
2
))
0
T
C
i S
C
i S
C
)
0
(18)
where
S
C
i = exp
_
i
_
C
i
dV
I
()
_
(19)
S
C
= exp
_
i
_
C
dH
/
,I
()
_
(20)
and both interaction operators are Heisenberg operators constructed
with H
0
. Note that H
/
,0
contains an implicit time dependence through
the external eld. Here the contour C starts and ends at t
0
, while the
contour C
i
starts at t
0
and ends at t
0
i.
The principal signicance of this result is that for the time-ordered
Green function on the contour C
i
the usual perturbation expansion
in form of Feynman graphs as in the case of equilibrium Green func-
tion can be applied. In particular if the Hamiltonian H
0
is bilinear in
Fermionic operators Wicks theorem can be applied. It states that in
each order of the perturbation theory the result for the Green func-
9
tion can be expressed by products of single particle Green functions of
the non-interacting system. The same holds for the statistical opera-
tor. Therefore the denominator cancels against non-connected graphs
in the numerator, leaving us with an expansion containing only con-
nected graphs. For these, nally, a Dyson equation can be formulated,
which will be the basis of the following discussion.
2.2 Dyson equation
In order to be specic let us assume a time-dependent interaction of
the form
H
/
t
=
_
dx

(x)U(x, t)(x) (21)

then the time-ordered Green function on the contour fullls the follow-
ing Dyson equation
G(1, 2) = G
0
(1, 2) +
_
d3G
0
(1, 2)U(3)G(3, 2) (22)
+
_
d3
_
d4G
0
(1, 3)(3, 4)G(4, 2)
where (i) is a short-hand notation for (x
i
,
i
) and
_
di =
_
dx
_
C
i
d
i
.
We did not specify the internal interaction further. It is contained in
the (irreducible) self-energy . The self-energy can be calculated and
expressed by single-particle Green functions in any specic case. In
principle the time-integrations have to be performed on the contour C
i
.
A simplication occurs if we put t
0
. Then the contribution from
the path t
0
to t
0
i can be neglected, if the internal interactions wash
10
t
t
C
C
1
1
2
2
2

Figure 5: Relation between real times and times on the dierent branches of the
Keldysh contour
out the memory of initial conditions. Then for the internal integrations
we may use the contour C (Fig. 5) extending from to + (part
C
1
) and back (part C
2
). This will be called the Keldysh contour in the
following.
There still remains the problem of expressing the integrations of
products of dierent functions by integrations over the usual time axis
< t < +. This will be the task of the following section.
2.3 Langreth rules, Keldysh formalism
In the Dyson equations appear integrations over the contour of products
of two and three functions. Let us start with the investigation of a
folding of two functions of the form (here we suppress the dependence
on the spatial variables)
C(
1
,
2
) =
_
C
d
3
A(
1
,
3
)B(
3
,
2
) (23)
Our aim is to express this integral on C by integrals over the simple
time-axis < t < +. The result depends on the choice of variables
11
on the contour. If we are interested, for instance, in C
<
(t
1
, t
2
) we place
as in Fig. 5
1
on C
1
and
2
on C
2
and evaluate the integrals on the
contour. Then we obtain:
C
<
(t
1
, t
2
) =
_
+

dt
_
A
r
(t
1
, t)B
<
(t, t
2
) +A
<
(t
1
, t)B
a
(t, t
2
)

(24)
For a derivation of this result we split the integrals:
C
<
(t
1
, t
2
) =
_
C
1
dA(
1
, )B(,
2
) +
_
C
2
dA(
1
, )B(,
2
) (25)
On C
1
the function B is a function B
<
while A = A
c
. On C
2
the
function A is a function A
>
while B = B
c
. Then for the rst integral
we obtain
_
C
1
dA(
1
, )B(,
2
) =
_
+

dtA
c
(t
1
, t)B
<
(t, t
2
) (26)
=
_
t
1

dtA
>
(t
1
, t)B
<
(t.t
2
) +
_
+
t
1
dtA
<
(t
1
, t)B
>
(t, t
2
)
For the second integral we nd
_
C
2
dA(
1
, )B(,
2
) =
_

+
dtA
>
(t
1
, t)B
c
(t, t
2
) (27)
=
_
t
2

dtA
>
(t
1
, t)B
>
(t, t
2
)
_
+
t
2
dtA
>
(t
1
, t)B
<
(t, t
2
)
Using the relations A
r
(t
1
, t
2
) = (t
1
t
2
)(G
>
(t
1
, t
2
) G
<
(t
1
, t
2
)) and
A
a
(t
1
, t
2
) = (t
2
t
1
)(G
>
(t
1
, t
2
) G
<
(t
1
, t
2
)) we can show the equiv-
alence of the two results.
In a similar way we can show
C
>
(t
1
, t
2
) =
_
+

dt
_
A
r
(t
1
, t)B
>
(t, t
2
) +A
>
(t
1
, t)B
a
(t, t
2
)

(28)
12
and
C
r
(t
1
, t
2
) =
_
+

dtA
r
(t
1
, t)B
r
(t, t
2
), C
a
(t
1
, t
2
) =
_
+

dtA
a
(t
1
, t)B
a
(t, t
2
)
(29)
In the Dyson equation we also encounter a folding of the form
C(
1
,
2
) =
_
C
dA(
1
, )U()B(,
2
) (30)
this is transformed into
C
<
(t
1
, t
2
) =
_
+

dt
_
A
r
(t
1
, t)U(t)B
<
(t, t
2
) +A
<
(t
1
, t)U(t)B
r
(t, t
2
)

(31)
For a product of three functions we obtain similar rules, for instance,
let
D(
1
,
2
) =
_
C
d
3
d
4
A(
1
,
3
)B(
3
,
4
)C(
4
,
2
) (32)
then
D
<
=
_
dt
_
A
r
B
r
C
<
+A
r
B
<
C
a
+A
<
B
a
C
a

(33)
The rules for foldings of products can also be cast into matrix multi-
plication form (Keldysh matrices) by writing:

A =
_
_
A
r
A
K
0 A
a
_
_
and

U = U
3
(34)
where A
K
(1, 1
/
) = A
>
(1, 1
/
) +A
<
(1, 1
/
). Then in the above example

D =

A

B

C (35)
The rules needed to calculate the other self energy terms can be found
in the review by Rammer and Smith.
13
As an example let us study the linear response of the density of
free conduction electrons in a scalar potential. Here the self-energy is
absent and we have the equation:
G(x
1
,
1
; x
2
,
2
) =
_
C
d
3
_
dx
3
G
0
(x
1
,
1
; x
3
,
3
)U(x
3
,
3
)G
0
(x
3
,
3
; x
2
,
2
)
(36)
The density is directly described by the function n(x, t) = iG
<
(x, t; x, t).
For this quantity we obtain
n(x, t) =
_
dx
3
_
+

dt
3
_
G
r
0
(x x
3
, t t
3
)U(x
3
, t
3
)G
<
0
(x
3
x, t
3
t)
G
<
0
(x x
3
, t t
3
)U(x
3
, t
3
)G
a
0
(x
3
x, t
3
t)

(37)
where we already have used that the equilibrium Green functions de-
pend only on the dierence of arguments.
For a time-dependence of the scalar potential of the form
U(x, t) = U exp(iq x it) (38)
and going over to a Fourier representation
G
0
(x, t) =
1
N

k
_
d
2
G
0
(k, ) exp(ik x it) (39)
we nd
n(x, t) = iU exp(iq x it)
1
N

k
_
d
2
_
G
r
0
(k +q, + )G
<
0
(k, )
G
<
0
(k +q, + )G
a
0
(k, )

Now we use the equilibrium relation:

G
<
(k, ) = if()A(k, ) = 2if()(
k
+), G
r,a
(k, ) =
1

k
+ i
(40)
14
L
V
e j
E
defect
Figure 6: Electronic transport due to an applied eld c, is limited by inelastic colli-
sions with lattice defects and phonons.
and obtain
n(x, t) =
1
N

k
f(
k
) f(
k+q
)
+
k

k+q
+i
U(x, t) (41)
Here we recover the well-known Lindhard function.
3 Kinetic equations: The Boltzmann Equation
3.1 The Conventional Derivation
The nonequilibrium (but steady-state) situation of an electronic current
in a metal driven by an external eld is described by the Boltzmann
equation. This diers from the situation of a system in equilibrium
in that a constant deterministic current diers from random particle
number uctuations due to coupling to a heat and particle bath. Away
from equilibrium (c ,= 0) the distribution function may depend upon
15
t - dt
r - dt
dr
dt
k - dt
dk
dt
r
k
t
hk = -eE

Figure 7: In lieu of scattering, particles ow without decay

r and t as well as k (or E(k)). Nevertheless, when c = 0 we expect the
distribution function of the particles in V to return to
f
0
(k) = f(r, k, t)[
c=0
=
1
e
(E(k)E
F
)
+ 1
(42)
To derive a form for f(r, k, t), we will consider length scales larger
than atomic distances, but smaller than distances in which the eld
changes signicantly. In this way the system is considered essentially
homogeneous with any inhomogeneity driven by the external eld. Now
imagine that there is no scattering (no defects, phonons), then since
electrons are conserved
f(r, k, t) = f
_
r vdt, k +
ec
h
dt, t dt
_
(43)
Now consider defects and phonons (See Fig. 8) which can scatter
a quasiparticle in one state at r v dt and time t dt, to another
at r and time t, so that f(r, k, t) ,= f
_
r vdt, k +
ec
h
dt, tdt
_
. We will
express this scattering by adding a term.
f(r, k, t) = f
_
r v dt, k +ec
dt
h
, t dt
_
+
_
f
t
_
S
dt (44)
16
Figure 8: Scattering leads to quasiparticle decay.
For small dt we may expand
f(r, k, t) = f(r, k, t) v
r
f +ec
k
f
h

f
t
+
_
f
t
_
S
(45)
or
f
t
+v
r
f
e
h
c
k
f =
_
f
t
_
S
Boltzmann Equation (46)
3.2 Towards Transport in the Green Function Approach
We can give a more microscopic derivation of the Boltzmann equation
employing the Green function approach. As above, we will assume
that U(x, t) is slowly dependent on x and t. Then it is expected that
also e.g. the density n(x, t) = iG
<
(x, t; x, t) slowly depends on these
variables. We therefore introduce the so-called Wigner coordinates
T = (t
1
+t
2
)/2, t = t
1
t
2
, X = (x
1
+x
2
)/2, x = x
1
x
2
(47)
and express the Green functions G(x
1
, t
1
; x
2
, t
2
) by these coordinates
G(x, t, X, T) := G(X+
x
2
, T +
t
2
; X
x
2
, T
t
2
) (48)
In equilibrium G(x, t, X, T) depends only on the dierence variables x
and t and falls o or oscillates rapidly in these variables, while in non-
equilibrium the dependence in X and T is induced by the perturbation
U(X, T).
17
In equilibrium we have introduced a Fourier transform with respect
to the dierence variables
G
<
(k, ) =
_
dt
_
d
3
xe
ikx+it
G
<
(x, t) (49)
Previously in this lecture we found that the function G
<
(k, ) can be
written as
G
<
(k, ) = iA(k, )f() (50)
where f() is the Fermi function and A(k, ) = 2ImG
r
(k, ) is the
spectral function, and
G
r
(k, ) =
1

k
+ (k, )
(51)
In the absence of internal interactions ( = 0) we have A(k, ) =
2(
k
+).
In the presence of a perturbation U(X, T) a similar representation
holds:
G
<
(k, , X, T) = iA(k, , X, T)F(k, , X, T) (52)
with
A(k, , X, T) = 2ImG
r
(k, , X, T) (53)
and a generalized distribution function F(k, , X, T).
In the quasiparticle approximation we assume that the electrons
dont decay (or are very long lived) and that they have a well dened
dispersion (k). In this case F does not depend on separately and
can be replaced by a simpler distribution function f(k, X, T). For this
18
function we obtain the Boltzmann equation which is of the form
_
f
T
+
(k)
k
f
X

U(X, T)
X
f
k
_
= If (54)
with a scattering term If, which in the case of impurity scattering
is given by
If = 2c

p
[V (k p)[
2
(
k

p
)(f(k, X, T) f(p, X, T)) (55)
In the remainder of this lecture we plan to show how this quasiclassical
Boltzmann equation can be derived within the frame of a more general
non-equilibrium theory.
3.3 Derivation of kinetic equations
We start from the Dyson equation in integral form on the Keldysh
contour:
G(1, 2) = G
0
(1, 2) +
_
d3G
0
(1, 3)U(3)G(3, 2)
+
_
d3
_
d4G
0
(1, 3)(3, 4)G(4, 2) (56)
or
G(1, 2) = G
0
(1, 2) +
_
d3G(1, 3)U(3)G
0
(3, 2)
+
_
d3
_
d4G(1, 3)(3, 4)G
0
(4, 2) (57)
19
From this we obtain for the retarded Green function on the time-axis
G
r
(x
1
, t
1
; x
2
, t
2
) = G
r
0
(x
1
, t
1
; x
2
, t
2
)
+
_
dt
3
_
dx
3
G
r
0
(x
1
, t
1
; x
3
, t
3
)U(x
3
, t
3
)G
r
(x
3
, t
3
; x
2
, t
2
)
_
dt
3
_
dt
4
_
dx
3
_
dx
4
G
r
0
(x
1
, t
1
; x
3
, t
3
)
r
(x
3
, t
3
; x
4
, t
4
)G
r
(x
4
, t
4
; x
2
, t
2
)
or using an abbreviated notation
G
r
= G
r
0
+G
r
0
UG
r
+G
r
0

r
G
r
(58)
where denotes a convolution in space and time with the same argu-
ments as above.
For the lesser function G
<
we get a more complicated equation:
G
<
= G
<
0
+G
r
0
UG
<
+G
<
UG
a
+G
r
0

r
G
<
+G
r
0

<
G
a
+G
<
0

a
G
a
(59)
Similar equations are obtained if we start from the other form of the
Dyson equation.
We now will transform the Dyson integral equation into a Dyson
equation in dierential form. Note that for non-interacting conduction
electrons we have the following equation of motion for the Fermionic
operator
i
(x, t)
t
= H
0
(x)(x, t) (60)
Therefore we obtain for the Green functions the equations of motion
_
i

t
1
H
0
(x
1
)
_
G
r,a
0
(x
1
, t
1
; x
2
, t
2
) = (t
1
t
2
) (61)
20
_
i

t
1
H
0
(x
1
)
_
G
<
0
(x
1
, t
1
; x
2
, t
2
) = 0 (62)
where H
0
(x) is the Hamiltonian for non-interacting electrons. For free
conduction electrons in a constant potential H
0
(x) =
2
x
/2m. The
-function in the equation of motion for G
r,a
0
is due to the step-function
in the denition. Similar equations hold if we dierentiate with respect
to the second time
_
i

t
2
H
0
(x
2
)
_
G
r,a
0
(x
1
, t
1
; x
2
, t
2
) = (t
1
t
2
) (63)
_
i

t
2
H
0
(x
2
)
_
G
<
0
(x
1
, t
1
; x
2
, t
2
) = 0 (64)
Then we obtain the equations of motion for G
r
:
_
i

t
1
H
0
(x
1
) U(x
1
, t
1
)
_
G
r
0
(x
1
, t
1
; x
2
, t
2
) = (t
1
t
2
)
+
_
+

dt
_
d
3
x
r
(x
1
t
1
; x, t)G
r
(x, t; x
2
, t
2
)
and a similar equation for G
a
, while for G
<
(or G
>
) we obtain the two
equations
_
i

t
1
H
0
(x
1
) U(x
1
, t
1
)
_
G
<
0
(x
1
, t
1
; x
2
, t
2
) = (65)
_
dt
_
d
3
x
r
(x
1
t
1
; x, t)G
<
(x, t; x
2
, t
2
) +
_
dt
_
d
3
x
<
(x
1
t
1
; x, t)G
a
(x, t; x
2
, t
2
)
_
i

t
2
H
0
(x
2
) U(x
2
, t
2
)
_
G
<
0
(x
1
, t
1
; x
2
, t
2
) = (66)
_
dt
_
d
3
xG
r
(x
1
, t
1
; x, t)
<
(x, t; x
2
, t
2
) +
_
dt
_
d
3
xG
<
(x
1
, t
1
; x, t)
a
(x, t; x
2
, t
2
)
21
In principle this is closed set of equations for the Green functions as soon
as one has a well-dened expression for the self-energies. However, due
to the many variables these equations of motion are very complicated.
3.4 Gradient expansion
Now we introduce the Wigner coordinates
T = (t
1
+t
2
)/2, t = t
1
t
2
, X = (x
1
+x
2
)/2, x = x
1
x
2
(67)
and express the Green functions by these coordinates
G(x, t; X, T) := G(X+
x
2
, T +
t
2
; X
x
2
, T
t
2
) (68)
Note that

T
G(x, t; X, T) =
_

t
1
+

t
2
_
G(x
1
, t
1
; x
2
, t
2
) (69)
therefore a kinetic equation with respect to T is obtained if we subtract
the two equations 65 and 66
_
i

t
1
+i

t
2
H
0
(x
1
) +H(x
2
) U(x
1
, t
1
) +U(x
2
, t
2
)
_
G
<
(x
1
, t
1
; x
2
, t
t
)
=
r
G
<
+
<
G
a
G
r

<
G
<

a
(70)
where the products on r.h.s. of Eq. 70 are convolutions in space and
time. If the variation of U is suciently slow we can make a gradient
expansion
U(x
1
, t
1
) U(x
2
, t
2
) x
U(X, T)
X
+t
U(X, T)
T
(71)
22
This expansion (the factors t and x) can be better handled if we go over
to a Fourier representation with respect to the fast dierence variables
x, t:
A(k, , X, T) =
_
dt
_
d
3
xe
ikx+it
A(x, t, X, T) (72)
then e.g.
_
dt
_
d
3
xe
ikx+it
xA(x, t, X, T) = i

k
A(k, , X, T) (73)
and the Fourier transform of the l.h.s. of Eq. 70 becomes
_
i

t
1
+i

t
2
H
0
(x
1
) +H(x
2
) U(x
1
, t
1
) +U(x
2
, t
2
)
_
G
<
(x
1
, t
1
; x
2
, t
t
)
i
_

T
+
(k)
k

X

U(X, T)
X

k
+
U(X, T)
T

_
G
>
(k, , X, T)
where we have slightly generalized the Fourier transform of the Hamil-
tonian H
0
into a dispersion (k). Here we encounter already all the
terms, except for the last term, which are contained in the driving
term of the Boltzmann equation. Dening
L :=
_

T
+
(k)
k

X

U(X, T)
X

k
+
U(X, T)
T

_
(74)
the equation of motion for the Green function G
<
now reads
iLG
<
(k, ; X, T) =
r
G
<
+
<
G
a
G
r

<
G
<

a
(75)
Before we discuss the r.h.s. of Eq. 75, let us introduce an ansatz for
the function G
<
which is similar in structure as in equilibrium:
G
<
(k, , X, T) = iA(k, , X, T)F(k, , X, T) (76)
23
where iA = G
r
G
a
is the spectral function and F is a generalized
distribution function. Within the gradient expansion the retarded and
advanced Green function are easily calculated from Eq. 3 and the cor-
responding equations of motion
G
r
(k, , X, T) =
1

k
+ U(X, T) (k, , X, T)
(77)
A(k, , X, T) = 2ImG
r
(k, , X, T) (78)
This result is remarkably simple. In order to obtain the Boltzmann
equation we make a further approximation by neglecting the self-energy
in the retarded Green function. Then we obtain the quasi-particle
approximation:
A
0
(k, , X, T)) = 2(
k
+ U(X, T)) (79)
Of course, in this approximation the frequency of the distribution
function F(k, , X, T) is no longer an independent variable. therefore
in the quasiparticle approximation
G
<
(k, , T, X) = A
0
(k, , X, T)f(k, X, T) (80)
Note that A
0
(as any function y(
k
+ U(X, T)) is an eigen-
function of the dierential operator of the Boltzmann equation:
LA
0
(k, , X, T)) = 0 (81)
Therefore
iLG
<
(k, , X, T) = A
0
Lf(k, X, T) (82)
24
Now we turn to the scattering term, the r.h.s. of Eq. 75. The gradient
expansion of convolutions can be performed systematically. For
C(x
1
, t
1
; x
2
, t
2
) =
_
dx
/
dt
/
A(x
1
, t
1
; x
/
, t
/
)B(x
/
, t
/
; x
2
, t
2
) (83)
one obtains
C(k, , X, T) =A(k, , X, T)B(k, , X, T)

i
2
_
A
T
B

B
T

A
X
B
k
+
A
k
B
X
_
Therefore, in the lowest approximation we write

r
G
<
+
<
G
a
G
r

<
G
<

a
(k, , X, T) =
=
_
G
<
(
r

a
) (G
r
G
a
)
<
_
(k, , X, T)
where we have replaced the convolutions by simple products. Then we
obtain
A
0
Lf(k, X, T) =
_
G
<
(
r

a
) (G
r
G
a
)
<
_
(k, , X, T) (84)
3.5 Self energy from impurity scattering
In the case of scattering from randomly distributed impurity scattering
with an impurity scattering potential V (q) the self energy is given in
Born approximation

r
(k, , X, T) = c

p
[V (k p)[
2
G
r
(p, , X, T) (85)
This holds in a similar way also for the other Green functions. This form
of the self-energy can be obtained from the corresponding Feynman
25
graph expansion which holds on the Keldysh contour. Inserting now
the ansatz for the G
<
function into the scattering term we nd
A
0
(k, , X, T)Lf(k, X, T) = c

p
A
0
(k, , X, T)A
0
(p, , X, T)
[V (k p)[
2
(f(k, X, T) f(p, X, T))
Finally we integrate over to obtain
Lf(k, X, T) = 2c

p
[V (k p)[
2
(
k

p
)(f(k, X, T) f(p, X, T))
(86)
This is the classical Boltzmann equation. The rst term on the r.h.s.
is the out-scattering term, the second term is the in-scattering term.
The importance of this result is the validity of this equation for the
distribution function also far from equilibrium. Their is no explicit ref-
erence to the temperature in the equation. It is contained only in the
initial conditions, where f(k, X, T) = f
0
(
k
) is the Fermi function.
The -function which expresses energy conservation is due to the quasi-
particle approximation of the spectrum function. Due to this approx-
imation typical quantum eects which are contained in the frequency
dependence of the distribution function get lost. There seems to be an
inconsistency in the treatment of the self energy: we have neglected
the self energy in the spectral function completely, while keeping it (in
lowest order in the gradient expansion) in the scattering term. This
defect can be cured partially by separating from the self energy those
terms which lead to a renormalization of the single-particle excitation
26
spectrum and those which lead to scattering.
Copies of these notes, supporting materials and links are available
at http://www.physics.uc.edu/jarrell/Green . Much of the material
came from standard texts and urls linked on the site.
27