Professional Documents
Culture Documents
Lecture 1 Fermi Liquid Theory
Lecture 1 Fermi Liquid Theory
I.
SECOND QUANTIZATION
A.
Systems of many particles are conveniently described by the second quantization representation. Suppose that we have a systems of particles (bosons or fermions), each of which
can be in one of the states i , i = 1, 2... The many-body wavefunction can be written down
in the occupation number representation, which specifies the number of particles occupying
each of the quantum states. Dirac notation for the state vector
| . . . , Ni1,
, Ni+1 , . . .i.
i
|{z}
# of particles in state i
ai | . . . , Ni1, Ni , Ni+1 , . . .i =
ai | . . . , Ni1, Ni , Ni+1 , . . .i =
Ni | . . . , Ni1, Ni 1, Ni+1 , . . .i
For fermions,
ai | . . . , Ni1, Ni , Ni+1 , . . .i =
ai | . . . , Ni1, Ni , Ni+1 , . . .i =
Commutation relations
h
ai , aj
ai aj aj ai = ij
h
[ai , aj ] = ai , aj
-operators
A -operator
is defined as
2
= 0.
(r) =
i (r) ai
(1.1)
(r) =
i (r) ai ,
where i (r) is the wave function, i.e., a solution of the Shroedinger equation, of state i.
As we will be dealing with non-relativistic Shroedinger equation, it is convenient to choose
i (r) to be independent of the spin, and to assign an additional quantum number (spin
(r) operator changes to
projection) to the a- operators. Then the definition of the
(r) =
i (r) ai
(r) =
i (r) ai ,
Now the sum runs only over the orbital states. In most cases, the orbital states i (r) form
a complete basis. In this case,
h
(r) ,
(r0 )
i (r) j (r0 ) ai , aj
ij
= (r r ) .
i (r) j (r0 ) ij =
i (r) i (r0 )
ij
-operator creates a particle at point r with spin regardless of its orbital quantum state.
eipr ap
vol p
X
(r) = 1
eipr ap ,
vol p
(1.2)
XX
n
px
1 X
npx (r) ap ,
L z pz
where n is the Landau level index, pz is the wavevector along the field (z-direction), px is
auxiliary quantum number specifying the degeneracy of the quantum state, and
npx (r) = e
with `B =
ipz z ipx x
1
(y + px `2B )
exp
2`2B
1/4 2n n!`B
2!
Hn
y + px `2B
`B
h
c/eB being the magnetic length and Hn (x) being the Hermite polynomial of
order n [1].
C.
H =
XZ
1X
2
r
(r)
dd r
h
2 2
(r) +
2m
{z
(r)
(r)
U (r)
{z
kinetic energy
(r)
(r0 )
(r0 )
(r) .
dd r 0 V (r r0 )
{z
(1.3)
Notice that the form of the Hamiltonian does not depend on the statistics, which enters only
-operators [2]. Substituting expansion (1.2) into
through the commutation properties of the
(1.3) and introducing Fourier transforms V (q) =
we obtain the Hamiltonian in terms of a operators
H =
h
2 k2
1 X
ak ak +
vol q
| 2m{z }
kinetic energy
U (q) ak+q ak
|
{z
1 1 X X
+
V (q) ak+q apq ap ak
2 vol k,p,q
|
II.
{z
dd reiqr U (r) ,
GREENS FUNCTIONS
A.
Classical Physics
D2 G (rt, r0 t0 ) = (r r0 ) (t t0 ) .
t
4
(2.1)
G gives a probability to find a particle at point r in time t t0 after the particle started its
0 0
G (rt, r t ) =
2 (D (t t0 ))3/2
B.
"
(r r0 )2
exp
4D (t t0 )
In Quantum Physics, the Greens functions are introduced as expectation values of com -operators. For example, the single-particle time-ordered (causal) Greens
binations of the
function is defined as
(rt)
(r0 t0 )i.
G (rt, r0 t0 ) = ihT
T f (t1 ) g (t2 ) =
where the first definition is for fermions (the minus sign is due the interchange of two
fermions) and the second one is for bosons. The Heisenberg representation of an operator
is defined as
(rt) = eiHt
(r) eiHt
(rt) = eiHt
(r) eiHt ,
(r) and
(r) are the operators in the Shroedinger representation, defined by
where
(1.1).
Example 5 Heisenberg representation for free particles
H=
X
k
k ak aka
P
1 X ipr it Pk k ak ak
a a
it
(rt) = eiHt
k0 0 k0 k0 0 k0 0
ap e
(r) eiHt =
e e
vol p
(2.2)
X
X
1 X ipr
=
k0 ak0 0 ak0 0 + . . .
e
k ak ak + . . . ap 1 it
1 + it
vol p
k
k0 0
X
X
1
1
=
eipr (ap itp ap + . . .) =
ap ei(prp t)
vol p
vol p
ap ap .
p ap ap
where p = p is the energy measured from the chemical potential. This choice is
especially convenient for Fermi systems, where the energy above or below the filled Fermi
sea is the only real energy we are concerned with. The Heisenberg operators simply change
to
(rt) = eiH 0 t
(r) eiH 0 t .
vol p
(rt)
(00)i =
G (rt) = G (rt, 00) = ihT
1
i vol
1
i vol
(rt)
(00) for t > 0
i
(00)
(rt) for t < 0
i
pp0
p p
pp
1, for p < pF
0, for p > pF
G (rt) =
0, for x < 0
. Thus
ei(prp t) (p pF ) , t > 0
P i(prp t)
e
(p p) , t < 0
i vol
i vol
1, for x > 0
Ga (, p) =
dt
dd rei(tpr) G (rt)
= i (p pF )
dte
i(p )t
+ i (pF p)
dtei(p )t np .
To regularize the divergent integrals over time, we add infinitesimally small imaginary parts
to the arguments of the exponential
dtei(p )t
dtei(p +i)t =
p + i
0
0
i
dtei(p )t np
dtei(p i)t np =
,
p i
"
(p pF )
(pF p)
+
.
p + i p i
1
.
p + isgn (p pF )
7
(2.3)
Since we are interested in the properties of the Greens function near its pole = p =
p2 /2m p2F /2m, the sign of p pF coincides with that of . Then Eq.(2.3) can be re-written
as
Ga (, p) =
1
.
p + isgn
(2.4)
Without the term in the denominator, the Greens function is obviously just a solution of
the Shroedinger equation
= G I
HG
The term is very important as it brings in the information about the Fermi sea. Also, the
similarity between the quantum Greens functions and Greens functions of the linear classical
equations of motion exists only as long as we are considering free particles. Quantum Greens
functions are defined for interacting systems as well which formally correspond to a nonlinear
Shroedinger equation.
Example 7 Time-ordered Greens function for phonons. Displacement operator for an elastic continuum is a linear superposition of plane waves
1 i(qrq t)
1 X
bq e
+ bq ei(qrq t) ,
q
u
(rt) =
2q
vol q
where is the mass density of the material, bq (bq ) is the bosonic annihilation (creation)
operator, and q = sq is the dispersion of the acoustic sound mode (s is the speed of sound).
Electrons couple to deformations produced by phonons. A uniform deformation means displacement of the crystal as a whole, which obviously cannot have any physical consequences.
Therefore, the energy of the electron-phonon interaction can contain only the divergence of
u
. The corresponding Hamiltonian is
eph =
H
(r)
~ u
(r) = i
dd r
X
k,q
ak+q ak q u
q ,
q i(qrq t)
bq e
+ bq ei(qrq t)
2
D (, q) =
C.
XZ
(r) f (r)
(r)i.
dd rh
where the upper sign refers to bosons and the lower one to fermions. Therefore,
hf i = i
= i
XZ
(r0)
(r0 0)i |r0 r
dd r f (r) (i) h
XZ
(2.5)
where the limit r0 r is taken after f acts on G . For example, the number density
operator is
n
=
X
i
(r ri )
so that the single particle operator f is just (r r1 ) . According to (2.5), the expectation
value of n
, i.e., the observable number density, is
n h
ni = i
XZ
dd r (r r1 ) G (rt, r0 0) |t0;r0 r = i
G (r1 t, r1 0) |t0 .
For a translationally invariant system, it is convenient to express n via the Greens function
in the momentum space
n=
XZ
dd p
(2)d
d it
e G (, p) |t0 .
2
9
(2.6)
XZ
dd p
np ,
(2)d
where np is the occupation number. Comparing the two last results, we find that
np =
d it
e G (, p) |t0 .
2
(2.7)
1 ~
~ r0 G (rt, r0 0) |t0 ;r0 r
r
m
(2.8)
Time-ordered Greens functions are not analytic functions of frequency (cf. Eq.(2.4).
This poses some inconveniences in the applications. A more convenient type of Greens
functions are the retarded and advanced ones. They defined in the same way as retarded and
advanced Greens functions are defined in Classical Physics: the retarded Greens function
corresponds to perturbation that occurs at t = 0. The advanced one is time-reversed: it
corresponds to a perturbation that occurs at t = . In Quantum Physics, the retarded and
advanced Greens function are defined as
GR (rt, r0 t0 ) = i (t t0 ) h (rt) (r0 t0 ) (r0 t0 ) (rt)i
GA (rt, r0 t0 ) = i (t0 t) h (rt) (r0 t0 ) (r0 t0 ) (rt)i.
The first sign is for bosons and the second one is for fermions. By construction, GR = 0 for
t < t0 , i.e., the system responds to the perturbation applied at time t0 only at later times.
Similarly, GA = 0 for t > t0 .
In equilibrium, one can set t0 = 0 so that the arguments of the Greens functions become
GR (rt, r0 ) and GA (rt, r0 ) .
h
G (rr ) =
A
G (rr ) =
0
0
dteit GR (rt, r0 )
dteit GA (rt, r0 )
are analytic functions in the upper and lower halves of the complex planes. (To see this,
substitute = 0 + i00 into the exponentials and check the convergence of the integrals.)
h
Exercise 11 Show that on the real axis (00 = 0), GR (rr0 ) = GA (rr0 ) .
Exercise 12 Show that the retarded and advanced Greens functions of the free Fermi gas
are given by
GR,A (, p) =
where > 0.
1
,
p i
(2.9)
(2.10)
This helps to establish a useful relation between the density of states and the Greens
function. For free fermions, the density of states (per one spin orientation) is defined as
() =
dd p
( p ) .
(2)d
1 Z dd p
ImGR,A (, p ) .
(2)d
Even if fermions are not free, this relation still holds. This formula is very important in
application to tunneling experiments. Quantity
1
A (, p ) = ImGR,A (, p )
is called the spectral function. This is a very important characteristic of a many-body system.
For free fermions, A (, p ) is a function. This means that a fermion with momentum p
11
can be found in a state with one and only one energy = p2 /2m. If fermions interact with
each other or with phonons,etc., the function gets smeared. Quite often, it is replaced by
a Lorentzian
( p )
.
( p )2 + 2
Now a fermion with momentum p can be found with any energy . However, the probability
of having a given energy is peaked at = p2 /2m and falls off outside the interval around
this value. The width of this interval is determined by .
Example 13 Another advantage of the retarded and advanced Greens functions become
obvious when one considers bi-linear combinations of these functions integrated over the
momenta. Consider, for example
dd p R
R
d G (1 , p) G (2 , p)
(2)
ARR =
dp (p ) GR (1 , p) GR 2 , p),
where (p ) is the density of states per one spin orientation. Switching to the new variable
p = p F , we obtain for the integral
ARR =
dp (p + F ) GR (1 , p) GR 2 , p).
If |1 |, |2 | F , we can extend the low limit of the integral to and take the density of
states out of the integral
ARR = (F )
dp
1
= 0.
(1 p + i) (2 p + i)
dd p L
L
d G (1 , p) G (2 , p) = 0,
(2)
whereas
ALR = ARL
1
1
= 2i (F )
(1 p + i) (2 p i)
1 2 + i
1
= 2i (F )
.
1 2 i
ARL = (F )
dp
Notice that, since is infinitesimally small, 2 and can be considered as the same number.
12
E.
Matsubara representation
Many-body systems are to be studied at finite temperatures. None of the Greens functions introduced so far provides a convenient way to deal with finite temperatures. We
need yet another Greens function: the Matsubara one. The Matsubara Greens function is
defined similar to the time-ordered one, except for the time is imaginary
0 0
0 0
GM
(r, r ) = hhT (r ) (r )ii.
(r) eiH
(r ) = eH
(r ) = eH 0
(r) eH 0 ,
0 = H
N
is the Grand Canonical Hamiltonian and T is the tau-ordering
where H
operator, defined similar to the T operator in real time
T f (1 ) g (2 ) =
f (1 ) g (2 ) for 1 > 2
g (2 ) f (1 ) for 1 < 2
T f (1 ) g (2 ) =
f (1 ) g (2 ) for 1 > 2
,
g (2 ) f (1 ) for 1 < 2
Notice that the Matsubara operator is obtained from the Heisenberg one by performing a
Wick rotation: t = i.
Finally, the double brackets hh. . .ii mean taking the trace over quantum mechanical states
and thermodynamic averaging with statistical weights. Explicitly,
GM
0
(r0 0 ) ,
(r0 ) = e Tr T e H 0 (r )
(r, r ) = Tr T e (H ) (r )
0 0
(r0 ) e H 0 (r) eH 0 .
(r0 0)ii = e Tr e H 0
GM (r, r0 ) = hhT (r )
13
0
0
0
GM (r r0 ) = e Tr e H (r) eH e H
(r0 )
0
0
0 0
= e Tr e H eH ( +) (r) eH ( +) e H
(r0 )
(r0 )ii = GM (r + , r0 )
= hh (r, + )
Because of the (anti) periodicity, the Greens function can be expanded into the Fourier
series
GM (r, r0 ) =
X
1 X
ein GM (rr0 , n ) = T
ein GM (rr0 , n ) .
n=
n=
n = 2n/ = 2nT, n = 0, 1, 2 . . .
(2.11)
n = (2n + 1) / = (2n + 1) T, n = 0, 1, 2 . . .
(2.12)
For fermions,
GM (rr0 , n ) is defined at discrete points on the imaginary axis of the complex plane.
Exercise 16 Show that the Matsubara Greens function for a free Fermi gas is given by
GM (n , p) =
1
.
in p
(2.13)
Exercise 17 Show that the Matsubara Greens function of the phonon field is
D M (n , q) =
q2
.
n2 + q2
(2.14)
Remark 18 Matsubara functions are useful not only at finite temperatures but also at T =
0. In the limit of T 0, Matsubara frequencies in Eqs. (2.11) and (2.12) become continuous
variables, and the discrete Matsubara sums are replaced by the integrals
T
X
n
...
d
...
2
X
n
in
dd p M
G (n , p) | 0 .
(2)d
14
GM (n , p) .
To get a guess at what this expression should be equal to, lets first take the T 0 limit.
Then,
Z
d i p
d 1
=
2 i p
2 2 + p2
Z
Z
d
d 1
= i
.
p
2 2 + p2
2 2 + p2
P (T = 0) =
The first integral diverges logarithmically but, as we can always choose the symmetric upper
limits, it in fact equals to zero by parity. The second integral is elementary (just dont forget
that p may be of both signs, hence the result is not simply /p but rather / |p |), so that
1
P (T = 0) = sgnp .
2
For T = 0, this is nothing more than np 1/2. Indeed, for p < pF , p < 0 and np 1/2 =
1 1/2 = 1/2; for p > pF , p > 0 and np 1/2 = 0 1/2 = 1/2. The conjecture is then
that
np
X
1
GM (n , p)
=T
2
n
(2.15)
F.
Analytic continuation
Retarded and advanced Greens function are obtained by analytic continuation of the
Matsubara Greens function from the discrete points on the imaginary axis to the entire
complex planes. The details of the procedure are given in standard textbooks, see, e.g.,
Ref.[[3]]. The rule we need to know is
GR (, p) = GM (n , p) |in +i for n > 0
(2.16)
(2.17)
Comparing Eqs.(2.9) and (2.13), we see that this rule is indeed obeyed.
15
III.
FEYNMAN DIAGRAMS
I am going to skip all the details of the formal derivation of the diagrammatic technique
(adiabatic switching of the interaction, interaction representation, S matrix, etc.) and
just formulate the diagrammatic rules. Suppose that we develop the perturbation theory
for electron-phonon interaction. Graphically, the diagrams for the Greens function are
represented by the following sequence (see Fig.1). The solid line with an arrow corresponds
the free Greens function. e.g., Eq.(2.13), if we are using the Matsubara representation.
The wavy line is the phonon Greens function, Eq.(2.14), multiplied by the electron-phonon
coupling constant g. The double solid line corresponds to the exact (renormalized) Greens
function. The direction of the arrow corresponds to the direction of time. The process
depicted in Fig.1 corresponds to fixing the initial and final states as: initialfermion n , p,
no phonons; finalfermion n , p, no phonons. Electrons propagate through an elastic medium
interacting with virtual phonons. This is why all phonon lines are always inserted into the
electron lines. I grouped the diagrams by the order of the perturbation theory. The main
rule of the diagram technique is the energy and momenta are conserved at every vertex, i.e.,
in every act of interaction.
For electron-electron (or, more generally, fermion-fermion) interaction, the diagrams are
similar except for we have one additional element to first order: the tadpole diagram (diagram b) in Fig. 2). To second order, tadpoles proliferate and get mixed with other elements.
The wavy line in tadpoles carries zero momentum, and this is why tadpoles are absent for
the electron-phonon interaction: by construction, the electron-phonon vertex vanishes for
q 0. In general, fermion-fermion interaction does not have to vanish in this limit. However,
we can still get rid of tadpoles. The closed circle is a Greens function at coinciding spatial
and temporal arguments, i.e., the number density. Therefore, all tadpoles renormalized the
fermion density at fixed chemical potential. If we are dealing with Coulomb interaction in a
jellium model, the total charge density (electrons +ions) is equal to zero, so that tadpoles
are eliminated. For He, the total density is not equal to zero but we can collect all tadpoles
diagrams together and treat them separately. Once they are removed, the fermion-fermion
diagrams are the same as electron-phonon ones, except for the wavy line is the interaction
potential, V (q) .
Finally, the last important example of diagrammatic technique is the interaction of elec16
=
a)
+
p
b)
c)
p-q
+
k
k+q
d)
e)
p+q
k+q
FIG. 1: Perturbative expansion of the Greens function for the electron-phonon interaction.
17
=
b)
a)
+
p
d)
c)
p-q
+
k
e)
k+q
f)
g)
p+q
k+q
h)
i)
j)
k)
l)
m)
FIG. 2: Perturbative expansion of the Greens function for the fermion-fermion interaction.
18
FIG. 3: Perturbative expansion of the Greens function in the electron-impurity interaction. Each
circle corresponds to scattering by the exact (before averaging) impurity potential (3.1).
trons with static impurities. Imagine that we have N impurities scattered randomly over
the sample. Impurity #i is at point ri . Each of the impurities is described by the potential
U (r) . The total potential field on the electron is
W (r) =
Ni
X
i=1
U (r ri ) .
(3.1)
Corrections to the Greens function for a fixed realization of impurities are shown in Fig.3.
Now we can average the Greens function over realizations of disorder, i.e., over the locations
of impurities. To first order, we obtain the average impurity potential hW i. This can be
always added to the reference energy. The non-trivial element, the correlation function of
the impurity potential
hW (r) W (r 0 )i = h
X
i,j
U (r ri ) U (r 0 rj )i
shows up to second order. Averaging means integrating over positions of all impurities.
For i 6= j in the sum, the average is vanishingly small. Therefore, only diagonal elements
i = j survive in the sum. Also, all impurities contribute equally to the average, so one can
integrate over the position of the first impurity and multiply the result by Ni
hW (r) W (r 0 )i = h
X
i
U (r ri ) U (r 0 ri )i
dd q 0
dd q
i(q(rr1 )+q0 (r0 r1 ))
0
= Ni
d
d U (q) U (q ) e
(2)
(2)
Z
Z
d 0
d
d q
d q
0
0
i(qr+q0 r0 )
= ni
d
d (q + q ) U (q) U (q ) e
(2)
(2)
Z
d
d q
iq(rr0 )
= ni
d U (q) U (q) e
(2)
Z
d d r1
vol
19
+ ...
A.
Self-energy
Consider all diagrams, e.g., for the electron-phonon interaction, which cannot be cut into
two parts without cutting at least one interaction line. The first order diagram is of this
type. All second-order diagrams except for diagram (b) also belong to this type. The sum of
all such diagrams is depicted graphically by a circle, called the self-energy. The perturbation
series can be re-drawn using a circle as shown in Fig.5 By direct inspection one can see that
all diagrams are reproduced in this way. Algebraically, the series for G can be written as
G = G0 + G20 + G30 2 ...
This is equivalent to the Dyson equation
G = G0 + G0 G.
Solving this equation, we obtain
G=
1
.
G1
0
1
.
i p M (, p )
20
1
.
p R,A (, p )
(3.2)
Certainly, a solution of the Dyson equation is not very instructive as long as we have not
found , which can be done only in the approximate way. Nevertheless, this parameterization
allows to introduce such important concepts as the effective mass and damping of singleparticle states. We can also immediately specify the general conditions for the system to be
a Fermi liquid.
B.
The main premise of the Fermi-liquid theory is that the system of interacting fermions
behaves similarly to a system of weakly interacting quasi-particles. At the lowest energy
scales (as counted from the Fermi energy) these quasiparticles become free, and then the
similarity becomes complete. The masses of these quasiparticles are, generally speaking,
different from the bare masses. Interacting fermions will behave as effectively free particles
only if their Greens function have the same form as free-fermion Greens function. This
statement is equivalent to the requirement that the self-energy entering, e.g. Eq.(3.2) allows
for a expansion in both of its arguments, the linear terms being the most important ones
R (, p ) = 0 + p p + O 2 , p2 + . . .
(3.3)
The constant term (0 ) can be absorbed in a shift of the chemical potential. Doing so and
keeping only the leading terms, the Greens function can be written as
GR (, p ) =
1 +
Z
1+p
p + i
p 1+ + i
1
=
2
d Z
1
Zsgn
p .
=
2 i p
2
21
b)
a)
c)
d)
22
The discontinuity in np is
np=pF np=pF + = Z.
Vanishing Z means smearing of the Fermi surface.
Recalling that near the Fermi surface p = (p2 p2F )/2m (p pF ) vF , we see that p
contains a renormalized Fermi velocity
vF = vF
1 + p
.
1 +
(3.4)
1 +
pF
,
=m
vF
1 + p
(3.5)
where m is the bare mass. The Fermi momentum is the same in the interacting and free system. More generally, the volume of the Fermi surface is independent of the interaction. This
very non-trivial statement, known as the Luttinger theorem, can be understood from the
Pauli principle: for fixed number of particles, the volume of the Fermi surface is determined
only by this number.
Remark 23 Notice that this effective mass has nothing to do with band mass defined as
mb = (d2 p /dp2 )
23
m = (1 + ) m = 1 + a1 m
but now it is a constant but a function of as well. Furthermore, m diverges for 0.
This means that the specific heat coefficient C (T ) /T is proportinal to the T dependent mass
and diverges at T 0 as well
C (T ) /T T (1a) .
(3.6)
2 + 2 ()
p
For fixed , the width of the Lorentzian () = 2 varies quadratically with . For small
energies, () , the spectral function has a sharp peak. This is a typical behavior for the
Fermi liquid.
Definition 26 Condition () is usually taken as on operational definition of the
Fermi liquid. Notice that the quadratic behavior of is a sufficient but not necessary condition for the Fermi liquid. The quasiparticles are well-defined as long as b with b > 1.
Remark 27 The self-energy is an analytic function of . As for any analytic function, its
real and imaginary parts are related to each other via a Kramers-Kronig relations
ImR (0 )
1Z
d0 0
Z
1
ReR ()
ImR () =
.
d0 0
ReR () =
24
(3.7)
(3.8)
Because of these relations, the behaviors of real and imaginary parts of are correlated.
Here are some simple rules relevant for the non-Fermi liquid behavior
ReR () || ln || ImR () ||
ReR () sgn ||a ImR () ||a for 0 < a < 1.
The first case is known as marginal Fermi liquid. In this case, quasiparticles are marginally
defined ( ) but the effective mass diverges logaithmically. The second case tells as that
the exponent in the power-law divergence of the effective mass gives us the exponent of the
decay rate.
Remark 28 Notice that ReR () is an odd function of whereas ImR () is an even
function of . This is easy to understand. Indeed, ImR () gives the decay rate of singleparticle states. The rate must be the same for > 0 and < 0, therefore Im R () must be
even in . Now, knowing that ImR () is even, we obtain from Eq.(3.7)
R
0
R
0
1Z
1Z
0 Im ( )
0 Im ( )
=
= ReR () .
d
d
Re () =
0
0
+
R
25