Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Field Theory
Condensed matter physics is a branch of physics that investigates the physical phenomena associated
with the many-body interaction of materials in their “condensed” (i.e. liquid and solid) states. The objective
of this course is to apply quantum field theory to various modern topics of condensed matter physics.
Typically most condensed matter physics topics may be described in terms of non-relativistic quantum field
theory. However, for topics involving gauge theory (e.g. spin liquids, high-temperature superconductivity)
and topological field theory (e.g. fractional quantum-Hall states), knowledge of relativistic quantum field
theory becomes necessary. In the interest of time, we shall first focus on non-relativistic descriptions of the
many-body interactions in fermions and bosons, although the basics of relativistic quantum field theory will
be briefly reviewed before taking the relativistic quantum field theory to the non-relativistic limit. The
necessary relativistic quantum field concepts for gauge theory, fractional statistics and topological field
theory will be covered in the context of high-temperature superconductivity and fractional quantum Hall
states later in this course.
Throughout this course we shall mostly use natural units in which the Dirac symbol (which is the
Planck constant h divided by 2π) and the speed of light c are both set to 1, although at times we’ll restore
them for quantitative comparison with experiments. Thus, in natural units both time and space are treated on
the same footing, and mass is inversely proportional to length.
The scope of condensed matter physics has evolved and expanded significantly in recently years,
from traditional “solid state physics” that largely focuses on effective single-particle pictures in solids and
Landau symmetry-breaking theory of phase transitions, to a new arena encompassing a broad range of topics
from highly interdisciplinary research such as nano- and biophysical sciences, optical lattices and Bose-
Einstein condensation in laser-cooled atoms, quantum computation, to fundamental subjects such as gauge
theory, quantum orders and quantum phase transitions, strongly correlated electronic systems, fractional
statistics, spin liquids, topological field theory, string-net condensate for unification of fermions and bosons,
etc. In the development of modern condensed matter physics, quantum field theory (QFT) and group theory
have played essential roles in the description of many-body interactions, symmetries and symmetry-breaking.
Therefore, QFT complemented by group theory becomes the language of choice in the discussion of modern
condensed matter physics. For a recent overview of aspects of modern condensed matter physics, you may
refer to the article “A perspective on frontiers of modern condensed matter physics”, N.-C. Yeh, Bulletin of
Association of Asia Pacific Physical Societies (AAPPS), Vol. 18, No. 2, pg. 11--29 (2008); also available at
the URL http://www.cospa.ntu.edu.tw/aappsbulletin/data/18-2/11_aperspective.pdf.
Generally speaking, QFT is a confluence of quantum mechanics and special relativity, and it builds
on important concepts of group theory and symmetries so that it can deal with any space-time dimensions,
matter-energy interactions (including three out of four of the known fundamental forces in the universe:
strong, electromagnetic and weak interactions) for both fermionic and bosonic fields, as well as topological
order and fractional statistics. Specifically, most condensed matter physics topics can be described by non-
relativistic quantum field theory. Interestingly, however, connections of condensed matter physics to
relativistic quantum field theory can be found in various aspects of topological orders and objects, and also in
certain aspects of strongly correlated electronic systems, such as in fractional quantum Hall states and high-
temperature superconductivity.
There are typically a variety of approaches taken for dealing with QFT. The most common
approaches include the canonical quantization formalism and the path integral formalism. Both approaches
evoke the use of Feynman diagrams and are convenient for different purposes. In this course we shall adapt
both approaches depending on the circumstances.
The foundation of conventional condensed matter physics may be regarded as building on two
conceptual cornerstones: the Fermi liquid theory, and the Landau symmetry-breaking theory of phase
transitions. The Fermi liquid theory treats properties of electronic states in solids as perturbations of a ground
state consisting of filling the single-particle energy levels. The Landau theory for phase transitions of matter
classifies different phases of matter by their symmetries, so that phase transitions are associated with changes
in the symmetry of the state of matter. However, these cornerstones can no longer hold grounds when facing
the challenge of emerging phenomena in various modern topics of condensed matter physics. For instance,
some of the strongly correlated electronic systems, such as high temperature superconductors, quantum Hall
phenomena in two-dimensional electron gas, and “Luttinger liquid” in one-dimensional conducting systems,
all involve properties beyond the perturbative descriptions of conventional Fermi-liquid theory, and certain
aspects of the strong correlation in these systems actually resemble phenomena encountered in high-energy
physics. Similarly, conventional notions of broken symmetry associated with phase transitions are no longer
applicable to the depiction of systems involving topological orders and their phase transitions. Well known
examples include the fractional quantum Hall (FQH) systems and spin liquids, where transitions among
different topological orders can occur without changing the corresponding symmetries. Hence, new
foundations must be established to describe these emerging areas of condensed matter physics. This course is
intended to first familiarize students with the conceptual foundation and basic language of conventional
condensed matter physics, and then proceed to the introduction of a small set of representative new
developments in modern condensed matter physics.
The course is structured as follows. Part I gives an overview of modern condensed matter physics
and a quick review of the second quantization techniques, quantum dynamics, pictures of quantum dynamics
and time-dependent perturbation theory, and the basic notions of low-energy excitations in solids. In Part II
non-relativistic quantum field theory for many-body systems using the Green function techniques and
Feynman diagrams is introduced. Part III discusses two important approximations, the Hartree-Fock and the
random phase approximations, that are widely used to account for electron-electron interactions in solids. In
Part IV linear response theory for many-body systems responding to external fields is developed, and the
corresponding response functions known as the Kobu formalism is introduced. An important many-body
interaction between electrons and the background phonons are discussed in Part V. Part VI describes the
phenomenology, microscopic foundation and applications of the Fermi liquid theory. Part VII deals with a
special case of Fermi liquid theory for non-perturbative strong interactions of magnetic impurities with the
spins of conducting electrons, known as the Kondo effect. Part VIII discusses the limitation of Fermi liquid
theory and introduces a representative example for the breakdown of Fermi liquid theory, the Luttinger
liquids in one dimension. In Part IX we consider interacting bosons at zero and finite temperatures and
superfluidity. Part X provides an overview of modern developments in atomic molecular physics that enables
cold gases for the manifestation of many interesting condensed matter phenomena, including the Bose-
Einstein condensation (BEC) and optical lattices. Superconductivity, a special state of matter involving a
bosonic ground state and fermionic low-energy excitations, is studied in Part XI for the case of conventional
and heavy Fermion superconductors. Time permitting, Part XII will cover the concepts and applications of
superconducting devices based on Josephson junctions. Finally, Part XIII describes recent developments in
high-temperature superconductivity, including phenomenology and various attempts at establishing the
microscopic theory. The newly discovered superconductivity in iron pnictides and related compounds will
also be briefly discussed.
Second quantization is a useful tool for dealing with many-body phenomena in condensed matter
physics. The method was developed by P. A. M. Dirac in 1927 for photons in radiation theory, and later on
extended to fermions by E. Wigner and P. Jordon in 1928.
The key concept of 2nd quantization technique is to describe a system of particles in the “occupation
number” space, in contrast to the description of particles in terms of ordinary coordinate wave functions Ψ as
in the first quantization language. Specifically, the distinction between the languages of 1st and 2nd
quantization is summarized as follows:
To convert 1st quantization expressions to 2nd quantization, let’s consider the following Hamiltonian
for N-interacting particles:
1 N
∑ V ( ri , r j ) ,
N
H = ∑ T ( ri , ri ) + (I.2)
i =1 2 i , j ≠i
where T and V denote the kinetic and potential energies, respectively. The time-dependent many-body wave
function Ψ ( r1 , r2 , … rN , t ) satisfies the Schrödinger equation:
∂
i Ψ ( r1 , r2 , … rN , t ) = H Ψ ( r1 , r2 , … rN , t ) , (I.3)
∂t
and
Ψ ( r1 , r2 , … rN , t ) = ∑ c (k , … k , t )ϕ (r ) … ϕ (r ) ,
1 N k1 1 kN N (I.4)
{k1 kN }
where ϕk ( ri ) (i = 1, 2, … N) denote the single-particle wave function, k i represent the quantum numbers
i
(which need not be the wave vector unless specified), and c ( k 1 , … k N , t ) are coefficients satisfying the
following permutation conditions:
In the case of fermions, we must keep track of the sign change. We may define occupation numbers and
write a Slater determinant as follows:
ϕk ( r1 ) … ϕk ( rN )
1 1
≡ f ( n1 , … n∞ , t ) . (I.7)
ϕk ( r1 ) … ϕk ( rN )
N N
Ψ ( r1 , r2 , … rN , t ) = ∑ c (k , … k , t )ϕ (r ) … ϕ (r ) ,
1 N k1 1 kN N
{k1 … k N }
≡ ∑ f ( n1 , … n∞ , t ) Φ n1 ,…n∞ ( r1 , … , rN ) ,
{n1 ,…n∞ }
⎡⎛ N ! N !… ⎞1/ 2 ⎤
≡ ∑ f ( n1 , … n∞ , t ) ⎢⎜ 1 2
N!
⎟ ∑ ϕk1 ( r1 ) … ϕk N ( rN ) ⎥ (I.9)
{n1 ,…n∞ } ⎢⎣⎝ ⎠ {k1 k N } ⎥⎦
Next, we define a Hilbert space spanned by the basis vectors n1 , n2 , … n∞ and also define the
creation and annihilation operators ai† and ai that satisfy the anti-commutation relations for fermions:
{a , a } ≡ a a + a a = δ
†
i j
†
i j
†
j i ij ,
{a , a } = {a , a } = 0 .
i j
†
i
†
j (I.10)
For comparison, we note that the creation and annihilation operators bi† and bi for bosons satisfy the
following commutation relations:
Returning to the case of fermions, a basis vector n1 , n2 , … n∞ can now be written in terms of the creation
operators:
( ) (a ) … ( a∞† )
n1 † n2 n∞
n1 , n2 , … n∞ = a1† 2 0 , ( ni = 0 or 1) , (I.12)
ak n1 , … nk , … n∞ = 0, if nk = 0,
Σk
= ( −1) n1 , … , ( nk − 1) , … n∞ , if nk = 1, (I.13)
Σk
and ak† n1 , … nk , … n∞ = ( −1) n1 , … , ( nk + 1) , … n∞ , if nk = 0,
= 0, if nk = 1, (I.14)
with Σ k = n1 + n2 + … + nk −1 . We note that creation operator ak† is the Hermitian conjugate of the annihilation
operator ak . Equations (I.13) and (I.14) may be rewritten into the following equivalent expressions:
Σk
ak n1 , … nk , … n∞ = ( −1) nk n1 , … , ( nk − 1) , … n∞ , (I.15)
Σk
ak† n1 , … nk , … n∞ = ( −1) 1 − nk n1 , … , ( nk + 1) , … n∞ . (I.16)
ak† ak = nk . (I.17)
bk n1 , … nk , … n∞ = nk n1 , … , ( nk − 1) , … n∞ , (I.18)
bk† n1 , … nk , … n∞ = nk + 1 n1 , … , ( nk + 1) , … n∞ . (I.19)
To express the Schrödinger equation of fermions in a second-quantization form, we first rewrite the
Slator determinant in EQ. (I.8) into the following:
1
∑ ( −1) P ⎡⎣ϕk1 ( r1 ) ϕk 2 ( r2 ) … ϕk N ( rN ) ⎦⎤
p
(I.20)
N ! P∈S N
where P denotes the operator that permutes the order of electrons, SN denotes the permutation group for N
indistinguishable objects, and p is the number of permutations. Next, we define the wave function
Inserting the expressions of EQs. (I.8), (I.20) and (I.21) into EQ. (I.3), we obtain
∂Ψ ( t ) ∂f ( n1′, … n∞′ , t ) 1
∑ ∑ ( −1) P ⎣⎡ϕk1 ( r1 ) ϕk 2 ( r2 ) … ϕk N ( rN ) ⎤⎦ ,
p
i =i
∂t {n1′ , … n∞′ } ∂t N! P
1
= H ∑ f ( n1′, … n∞′ , t ) ∑ ( −1) P ⎡⎣ϕk1 ( r1 ) ϕk 2 ( r2 ) … ϕk N ( rN )⎤⎦ .
p
(I.22)
{n1′ , … n∞′ } N! P
We may multiply both sides of EQ. (I.22) by the conjugate of a particular Slator determinant that
corresponds to a specific set of occupation numbers ( n1 , … , n∞ ) , which yields
∂f ( n1 , … , n∞ , t ) 1
∑ f ( n1′, … n∞′ , t ) ∑ ( −1)
p + p′
i
∂t
=
N ! {n1′ , … n∞′ } P , P′
∫ dr1 … drN
× P′ ⎡⎣ϕk*1 ( r1 ) ϕk*2 ( r2 ) … ϕk* N ( rN ) ⎤⎦ H P ⎡⎣ϕk1 ( r1 ) ϕk 2 ( r2 ) … ϕk N ( rN ) ⎤⎦ , (I.23)
For simplicity, we first consider a Hamiltonian with kinetic energy only. That is, V ( ri , r j ) = 0 , and
H = ∑ T ( ri ), i = 1, 2, ,N . (I.24)
i
Since T ( ri ) is a one-particle operator, the set of occupation numbers {n1′, … , n∞′ } cannot differ from a given
set {n1 , … , n∞ } in EQ. (I.23) by more than two numbers. Thus, the right side of EQ. (I.23) can be simplified
into the following form if we assume that k < l and define k ( l ) as the kth (lth) state of which {n1 , … , n∞ }
has an occupation number nk (nl):
∑ ( −1) f ( n1 , … , nk − 1, … , nl + 1, … , n∞ , t ) ∫ dr ϕk*( r ) T ( r ) ϕl ( r )
Σ k +Σl
k ,l
≡ ∑ ( −1) f ( n1 , … , nk − 1, … , nl + 1, … , n∞ , t ) k T l .
Σ k +Σl
(I.25)
k ,l
Therefore the Schrödinger equation using EQ. (I.23) and EQ. (I.25) becomes:
∂ Ψ (t ) ∂f ( n1 , … n∞ , t )
i
∂t
= ∑ i
∂t
n1 , … n∞ ,
{n1 , … n∞ }
∑ ∑ f ( n , … , n − 1, … , n + 1, … , n , t ) k T l ( −1)
Σ k +Σl
= 1 k l ∞ n1 , … , n∞ ,
{n1 , n∞ } k ,l
= ∑ ∑ f ( n , … , n − 1, … , n + 1, … , n , t )
1 k l ∞ k T l ak† al n1 , … , nk − 1, … , nl + 1, … , n∞ ,
{n1 , … n∞ } k ,l
= ∑ k T l ak† al Ψ ( t ) . (I.26)
k ,l
We note that in EQ. (I.26) the sum over P and P′ give a factor N! that cancels the same factor in the
denominator. Consequently, from EQ. (I.26) we find that for H = ∑ i =1 T ( ri ) , its second-quantization form
N
is:
H = ∑ k T l ak† al . (I.27)
k ,l
Following similar procedures that lead to EQ. (I.27), the interaction term of the
Hamiltonian, V ( ri , r j ) , can be expressed by the second-quantization form (see Problem Set 1):
1 1
∑
2 k , l , s ,t
kl V st ak† al† at as ≡ ∑ ⎡ ∫ dr1dr2 ϕ k*( r1 ) ϕl*( r2 ) V ( r1 , r2 ) ϕ s ( r1 ) ϕt ( r2 ) ⎤ ak† al† at as . (I.28)
2 k , l , s ,t ⎣ ⎦
Hence, the second-quantization expression for a general Hamiltonian with both the kinetic and potential
energy terms is given by:
1
H = ∑ k T l ak† al + ∑ kl V st ak†al†at as . (I.29)
k ,l 2 k ,l , s ,t
ψ ( r ) = ∑ ϕk ( r ) ak , (I.30)
k
and ak is a fermion operator. Using EQ. (I.30), we may obtain the second-quantization expression of an
interaction Hamiltonian by writing the following:
1
H = ∫ dr ψ †( r ) T ( r ) ψ ( r ) + dr dr′ ψ †( r )ψ †( r′ ) V ( r, r′ ) ψ ( r′ )ψ ( r ) .
2∫
(I.31)
You can easily verify that EQ. (I.31) is entirely consistent with EQ. (I.29).
Now let’s consider applying the above formalism to an explicit example, the Coulomb interaction in
a degenerate electron gas. The Hamiltonian for N electrons of mass me and charge e is approximated by:
H e−e = ∑
N
pi2 e 2 N (
exp − μ ri − r j ),
i =1
+
2me 2
∑
i≠ j ri − r j
(I.32)
where μ is the screening coefficient of the electron gas, ri and rj are the positions of the electrons, and pi
denotes the momentum of the i-th electron. The field operator is given by
1 ik i r
ψ (r ) = ∑ e ησ akσ , (I.33)
k ,σ Ω
where k denotes the wave vector, σ is the spin quantum number, Ω is the volume, and
⎛1⎞ ⎛0⎞
η↑ = ⎜ ⎟ , η↓ = ⎜ ⎟ , (I.34)
⎝0⎠ ⎝1⎠
are the spinors. We also note that the following condition is satisfied:
∫ dr ψ ( r )ψ ( r ) = ∑σ a σ a σ = N .
† †
k k (I.35)
k,
Using EQ. (I.31) and EQ. (I.33), the Coulomb interaction term of the Hamiltonian in EQ. (I.32) becomes:
e2 ⎧⎪ − i ( k1 i r + k 2 i r ′ ) ⎛ e
− μ r −r′
⎞ i ( k 3 i r + k 4 i r ′)
∑ ∑ ∑ ∑ ⎨∫ dr ∫ dr′ e ⎜⎜ ⎟⎟ e
2Ω 2 k1σ1 k 2σ 2 k 3σ 3 k 4σ 4 ⎩⎪ ⎝ r − r′ ⎠
3 ⎛e
i qix
⎞ −μx 4π
∫ ⎜⎝ x ⎟⎠ e = q 2 + μ 2 .
d x (I.37)
e2 ⎛ 4π ⎞ †
∑∑∑ ⎜
2Ω kσ k ′σ ′ q ⎝ q 2 + μ 2 ⎠
†
⎟ ak −q ,σ ak ′+q ,σ ′ ak ′,σ ′ ak ,σ , (I.38)
The second-quantization expression for the total Hamiltonian of the degenerate electron gas is therefore:
⎛ 2k 2 ⎞ † e2 ⎛ 4π ⎞ †
H = ∑⎜ ⎟ k ,σ k ,σ
a a + ∑∑∑ ⎜ 2 a† a a .
2 ⎟ k −q ,σ k ′+ q ,σ ′ k ′,σ ′ k ,σ
a (I.39)
kσ ⎝ 2 me ⎠ 2Ω kσ k ′σ ′ q ⎝ q + μ ⎠
The physical significance of the interaction Hamiltonian can be readily seen from its second-
quantization expression if we compare EQ. (I.39) with the corresponding diagrammatic depictions of the
interaction in Figure I.2.1, where we have defined V(q) ≡ 4π/(q 2 + μ 2). Specifically, the Coulomb interaction
leads to two types of processes, electron-electron scattering and electron-hole pair creation and annihilation.
There is no spin flipping involved in the processes because the Hamiltonian does not include magnetic
interaction.
( k − q, σ ) ( k ′ + q, σ ′ ) ( k − q, σ ) ( k ′ + q, σ ′ )
V (q ) V (q )
(k,σ ) ( k ′, σ ′ ) (k,σ ) ( k ′, σ ′ )
Fig.I.2.1 Diagrammatic depictions of the Coulomb interaction in EQ. (I.39). Here holes refer to the
removal particles below the Fermi level.
Thus far we have only considered the second-quantization of fermionic systems. In the following we
examine an example involving a bosonic system.
⎡1 1 2⎤
U = ∫ dr ⎢ ε 0 E ( r ) + B (r ) ⎥ ≡ U E + U M .
2
(I.40)
⎣2 2 μ0 ⎦
The electric and magnetic fields E(r) and B(r) can be expressed in terms of the normal coordinate qα and the
canonical momentum pα ≡ qα as follows:
∂A ( r )
E (r ) = − , B ( r ) = ∇ × A( r ) ,
∂t
1/ 2
⎛ μ ⎞
A( r ) = c ⎜ 03 ⎟
⎝ 8π ⎠
∑
α
eˆα ∫ dk qα ( k , t ) e i k ir
, (α : polarization) ,
1/ 2
⎛ μ ⎞
B ( r ) = ic ⎜ 03 ⎟
⎝ 8π ⎠
∑
α
∫ dk ( k × eˆα ) qα ( k , t ) e
i k ir
,
1/ 2
⎛ μ ⎞
E ( r ) = −c ⎜ 03 ⎟
⎝ 8π ⎠
∑
α
eˆα ∫ dk qα ( k , t ) e i k ir
. (I.41)
Assuming ∇i A = 0 so that k i eˆα = 0 and qα* ( k , t ) = qα ( −k , t ) to ensure a real vector potential, we obtain
c2
UM =
2 ( 2π )
3 ∑ ∫ dk ( k × eˆα ) i ( k × eˆα ) qα ( k , t ) qα ( k , t ) ,
, ′
αα
′
*
′
c2 1
∑ ∫ dk ( k δαα ) qα ( k , t ) qα ( k , t ) = 2 ∑ ∫ dk qα ( k , t )
2
= 2
′
*
′ ω2 , (I.42)
2 ( 2π ) ( 2π )
3 3
αα ′ α
1 1
∑ ∫ dk qα ( k , t ) ∑ ∫ dk pα ( k , t )
2 2
UE = = . (I.43)
2 ( 2π ) 2 ( 2π )
3 3
α α
Note that in EQ. (I.42), we have used the dispersion relation ω = ck for free electromagnetic field. Hence,
we obtain the following expression for the Hamiltonian:
1
∑ dk ⎡ pα ( k , t ) + ω 2 qα ( k , t ) ⎤ .
2 2
H =
2 ( 2π )
3
α
∫ ⎣ ⎦
(I.44)
From EQ. (I.44) we note that qα and pα are in fact conjugates so that the corresponding Lagrangian L of EQ.
(I.44) yields pα = ∂L ∂qα = qα . Therefore qα and pα satisfy the following relations (by restoring ):
⎡⎣ qα* ( k ) , pα ′( k ′ ) ⎤⎦ = i δαα ′ δ ( k − k ′ ) ,
⎡⎣ qα ( k ) , qα ′( k ′ ) ⎤⎦ = ⎣⎡ pα ( k ) , pα ′( k ′ ) ⎦⎤ = 0 . (I.45)
To second-quantize EQ. (I.45), we introduce the photon creation and annihilation operators bᆠand
bα so that
⎡⎣bα ( k ) , bα† ′( k ′ ) ⎤⎦ = δαα ′ δ ( k − k ′ ) ,
⎡⎣bα ( k ) , bα ′( k ′ ) ⎤⎦ = ⎡⎣bα† ( k ) , bα† ′( k ′ ) ⎤⎦ = 0 . (I.46)
This leads to
1/ 2
⎛ω ⎞ i
bα ( k ) = ⎜ ⎟ qα ( k ) + pα* ( k ) ,
(2 ω )
1/ 2
⎝2 ⎠
1/ 2
⎛ω ⎞ i
bα ( k ) = ⎜ ⎟ qα* ( k ) −
†
pα ( k ) . (I.47)
(2 ω )
1/ 2
⎝2 ⎠
Or equivalently,
1/ 2
⎛ ⎞
qα ( k ) = ⎜ ⎟ ⎡⎣bα ( k ) + bα† ( −k ) ⎤⎦ ,
⎝ 2ω ⎠
1/ 2
1⎛ ω ⎞
pα ( k ) = ⎜ ⎟ ⎡⎣bα ( k ) − bα† ( −k ) ⎤⎦ . (I.48)
i⎝ 2 ⎠
Thus, we obtain
1 ⎡ 1⎤ 1 ⎡ 1⎤
H = ∑ ∫ dk ⎢⎣bα ( k ) bα ( k ) + 2 ⎥⎦
†
ωα ( k ) ≡ ∑ ∫ dk ⎢⎣nα ( k ) + 2 ⎥⎦ ωα ( k ) , (I.49)
( 2π ) ( 2π )
3 3
α α
We note that similar techniques may be applied to phonons, the normal modes of lattice vibrations,
except that the k-vectors are associated with the lattice vibrations, and that ωα ( k ) contains the dispersion
relation of phonon branches, in contrast to the dispersion relation ω = ck for the free electromagnetic field.
In this section we review three pictures (the Schrödinger, interaction, and Heisenberg pictures) of
quantum dynamics and their application to the time-dependent perturbation theory that will be essential for
our later development of the Green’s function and diagrammatic techniques.
We begin with general consideration for the solution to the following differential equation
∂
i Ψ( t ) = H ( t ) Ψ( t ) , (I.50)
∂t
where H (t) is a Hermitian operator, (which is usually the case when H(t) corresponds to an energy operator
of a physical system, although sometimes it is allowed to be non-Hermitian to represent dissipative systems).
To find the general solution to EQ. (I.50) without restricting to any specific picture, we first introduce a time
evolution operator U ( t , t0 ) , which relates the initial state Ψ ( t0 ) to the final state Ψ ( t ) by the expression:
Ψ ( t ) ≡ U ( t , t0 ) Ψ ( t0 ) . (I.51)
From EQs. (I.50) and (I.51), the time evolution operator satisfies the following differential equation and
initial condition:
∂
i U ( t , t0 ) = H ( t ) U ( t , t0 ) , (I.52)
∂t
U ( t0 , t 0 ) = 1 . (I.53)
Following EQs. (I.50) – (I.53) and assuming that H(t) is Hermitian, the evolution for the adjoint space is
given by:
∂
−i Ψ( t ) = Ψ( t ) H ( t ) , (I.54)
∂t
∂ †
−i U ( t , t 0 ) = U † ( t , t0 ) H ( t ) , (I.55)
∂t
U ( t0 , t0 ) = 1 .
†
(I.56)
∂ 1
⎡⎣U †( t , t0 ) U ( t , t0 ) ⎤⎦ = ⎡⎣ −U †( t , t0 ) H ( t ) U ( t , t0 ) + U †( t , t0 ) H ( t ) U ( t , t0 ) ⎤⎦ = 0 , (I.57)
∂t i
which, together with the initial conditions, implies that:
U †( t , t0 ) U ( t , t0 ) = 1 ⇒ U †( t , t0 ) = U −1( t , t0 ) = U ( t0 , t ) . (I.58)
In other words, the time evolution U-operator is unitary. In addition, we note that
U ( t1 , t2 ) U ( t2 , t3 ) = U ( t1 , t3 ) .
In the special case of H being a constant of time, the U-operator satisfies the group property.
Solving for quantum evolution is equivalent to finding the solution for the U-operator. From EQs.
(I.52) and (I.53), the time evolution operator satisfies the following integral equation:
i i
U ( t , t0 ) = U ( t0 , t0 ) − dt ′ H ( t ′ ) U ( t ′, t0 ) = 1 − dt ′ H ( t ′ ) U ( t ′, t0 ) .
t t
∫
t0 ∫
t0
(I.59)
⎛ −i ⎞ t ′ ⎛ −i ⎞ t ⎡ ⎛ −i ⎞ t ′ ⎤
U ( t , t0 ) = 1 + ⎜ ⎟ ∫t0 dt H ( t ′ ) U ( t ′, t0 ) = 1 + ⎜ ⎟ ∫t0 dt ′ H ( t ′ ) ⎢1 + ⎜ ⎟ ∫t0 dt ′′ H ( t ′′ ) U ( t ′′, t0 ) ⎥ ,
⎝ ⎠ ⎝ ⎠ ⎣ ⎝ ⎠ ⎦
2
⎛ −i ⎞ t ⎛ −i ⎞ t′
= 1 + ⎜ ⎟ ∫ dt ′ H ( t ′ ) + ⎜ ⎟ ∫ dt ′ H ( t ′ ) ∫ dt ′′ H ( t ′′ ) + … .
t
(I.60)
⎝ ⎠ t0 ⎝ ⎠ t0 t0
1 t
≡ ∫ dt ′∫ dt ′′ Tˆ ⎡⎣ H ( t ′ ) H ( t ′′ ) ⎤⎦ ,
t
(I.61)
2 0 t t 0
where we have introduced a time-ordering operator Tˆ that orders the operators in the bracket from left to
right with descending times, and the step function θ ( ti − t j ) is defined by:
θ ( ti − t j ) = 1 for ti > t j ,
=0 for ti < t j . (I.62)
The expression in EQ. (I.61) can be generalized to all terms in EQ. (I.60) by considering the following:
⎡ n ⎤
Tˆ ⎢ ∏ H ( ti ) ⎥ = Tˆ ⎡⎣ H ( t1 ) … H ( tn ) ⎤⎦
⎣ i =1 ⎦
= ∑ θ ( tσ (1) − tσ (2) ) … θ ( tσ ( n −1) − tσ ( n ) ) H ( tσ (1) ) … H ( tσ ( n ) )
σ ∈Sn
⎛ n −1
⎞
∑ ⎜⎝ ∏θ ( tσ − tσ (i +1) )∏ H ( tσ (i ) ) ⎟ ,
n
= (i ) (I.63)
σ∈Sn i =1 i =1 ⎠
where the summation is over Sn, the permutation group of n objects. Hence, we have
dtn Tˆ ⎡⎣H ( t1 ) … H ( tn ) ⎤⎦
t t
∫
t0
dt1 … ∫
t0
⎡ t ⎛ n −1 ⎞⎤
( σ ( i +1) ∏ H ( tσ ( i ) ) ⎟ ⎥
)
n
∑ n ⎜∏
t
= ⎢ ∫t0 1
dt … ∫ dt θ tσ (i ) − t
σ ∈Sn ⎣
t0
⎝ i =1 i =1 ⎠⎦
⎡ t ⎞⎤
( ) ⎛
( )
n −1 n
= ∑ ⎢ ∫ d tσ −1 (1) … ∫ d tσ −1 ( n ) ⎜ ∏ θ ( ti − ti +1 )∏ H ( ti ) ⎟ ⎥
t
σ ∈Sn ⎣ 0
t t0
⎝ i =1 i =1 ⎠⎦
⎡ t ⎛ n −1 n
⎞⎤
= ∑ ⎢ ∫ dt1 … ∫ dtn ⎜ ∏ θ ( ti − ti +1 )∏ H ( ti ) ⎟ ⎥
t
σ ∈Sn ⎣
t0 t0
⎝ i =1 i =1 ⎠⎦
⎡ t ⎛ n −1 n
⎞⎤
= n ! ⎢ ∫ dt1 … ∫ dtn ⎜ ∏ θ ( ti − ti +1 )∏ H ( ti ) ⎟ ⎥
t
⎣ t0 t0
⎝ i =1 i =1 ⎠⎦
= n ! ⎡ ∫ dt1 ∫ dt2 … ∫ dtn H ( t1 )… H ( tn ) ⎤ .
t t1 t n−1
(I.64)
⎢⎣ t0 t0 t0 ⎥⎦
From EQs. (I.60) and (I.64), we find that the U-operator takes the form:
⎛ −i ⎞ 1 t
∞ n
n =0 ⎝ ⎠ n ! t0 t0
⎧ ⎡ ⎛ −i ⎞ t ⎤⎫
= Tˆ ⎨exp ⎢ ⎜ ⎟ ∫ dt ′ H ( t ′ ) ⎥ ⎬ . (I.65)
⎩ ⎣ ⎝ ⎠ t0 ⎦⎭
In EQ. (I.65) we note that the term associated with n = 0 is simply 1. In the event that H is time independent,
the time-evolution operator takes a simple form U ( t , t0 ) = exp [ − iH ( t − t0 ) / ].
Having derived the above general solution for the differential equation EQ. (I.50), we now apply it to
special cases in the following.
Ψ S ( t ) = U ( t , t0 ) Ψ S ( t0 ) ,
− iH ( t -t0 ) /
→e Ψ S ( t0 ) if H is time independent. (I.67)
Thus, for a given solution Ψ S ( t0 ) at time t0, the unitary transformation in EQ. (I.67) generates the solution
to the Schrödinger equation at time t if H is time independent.
In dealing with many realistic physical problems, we are often interested in a Hamiltonian that
consists of two terms, one is a model Hamiltonian H0 typically chosen as a soluble term, and the other is the
“interaction” term H ′:
H = H0 + H ′. (I.68)
The state vectors of the interaction picture are defined according to the following form:
Ψ I ( t ) = U 0 ( 0, t ) Ψ S ( t ) , (I.69)
where U0 refers to the time-evolution operator of the Hamiltonian H0. In the event that H0 is a constant of
time, we have U 0 ( 0, t ) = exp ( iH 0 t / ) and the state vectors of the interaction picture are given by:
Ψ I (t ) = e ΨS (t ) .
iH 0 t /
(I.70)
Using EQs. (I.52) and (I.67) – (I.69), we obtain the state of motion of the interaction state vector:
∂ ∂U 0 ( 0, t ) ∂
i Ψ I (t ) = i Ψ S ( t ) + U 0 ( 0, t ) i ΨS (t ) ,
∂t ∂t ∂t
= − U 0 ( 0, t ) H 0 Ψ S ( t ) + U 0 ( 0, t ) H Ψ S ( t ) ,
= − U 0 ( 0, t ) H 0 U 0 ( t , 0 ) Ψ I ( t ) + U 0 ( 0, t ) ( H 0 + H ′ ) U 0 ( t , 0 ) Ψ I ( t ) ,
= U 0 ( 0, t ) H ′ U 0 ( t , 0 ) Ψ I ( t ) ,
≡ H I′( t ) Ψ I ( t ) . (I.71)
We note that the order of H0 and H ′ must be tracked carefully because in general they do not commute.
Equation (I.71) has essentially the same form as EQ. (I.50) except that it is associated with HI′(t) and
Ψ I ( t ) rather than H and Ψ ( t ) .
The definition of HI′(t) in EQ. (I.71) may be compared with the matrix element of any arbitrary
operator OS ( t ) in the Schrödinger picture:
Ψ ′S ( t ) OS ( t ) Ψ S ( t ) = Ψ′I ( t ) U 0 ( 0, t ) OS ( t ) U 0 ( t , 0 ) Ψ I ( t ) ≡ Ψ ′I ( t ) OI ( t ) Ψ I ( t ) . (I.72)
Therefore, HI′(t) is an operator in the interaction picture, and both the operators and the state vectors in the
interaction picture are dependent on time. Moreover, if Os is time independent, we find that
∂ ∂
i OI ( t ) = i ⎡U 0 ( 0, t ) OS U 0 ( t , 0 ) ⎤⎦ = U 0 ( 0, t ) ⎣⎡OS H 0 − H 0OS ⎦⎤ U 0 ( t , 0 ) ,
∂t ∂t ⎣
= ⎡⎣ U 0 ( 0, t ) OS U 0 ( t , 0 ) ⎤⎦ H 0 I − H 0 I ⎡⎣ U 0 ( 0, t ) OS U 0 ( t , 0 ) ⎤⎦ = ⎡⎣OI ( t ) , H 0 I ⎤⎦ . (I.73)
∂ ∂ ⎡ ∂OS ( t ) ⎤
i OI ( t ) = i ⎡⎣U 0 ( 0, t ) OS ( t ) U 0 ( t , 0 ) ⎤⎦ = U 0 ( 0, t ) ⎢OS ( t ) H 0 − H 0OS ( t ) + i ⎥ U 0 (t, 0) ,
∂t ∂t ⎣ ∂t ⎦
∂OS ( t )
= ⎡⎣ U 0 ( 0, t ) OS ( t ) U 0 ( t , 0 ) ⎤⎦ H 0 I − H 0 I ⎡⎣ U 0 ( 0, t ) OS ( t ) U 0 ( t , 0 ) ⎤⎦ + i U 0 ( 0, t ) U 0 (t, 0) ,
∂t
⎛ ∂O ⎞
≡ ⎡⎣OI ( t ) , H 0 I ⎤⎦ + i ⎜ S ⎟ . (I.74)
⎝ ∂t ⎠ I
Equation (I.73) can be readily applied to the creation and annihilation operators. For a representation
in which H0 is diagonal, we can express H0 in terms of the creation and annihilation operators ck† and ck as
follows:
H 0 = ∑ ωk ck† ck . (I.75)
k
Hence, the time dependence of the creation and annihilation operators ck†I ( t ) and ckI ( t ) in the interaction
picture can be determined by using EQs. (I.70) and (I.73):
∂
i ckI ( t ) = ⎡⎣ckI ( t ) , H 0 I ⎤⎦ = ei H 0 t / ⎡⎣ck , H 0 ⎤⎦ e −i H 0 t / = ωk ei H 0 t / ck e− i H 0 t / = ωk ckI ( t ) , (I.76)
∂t
which leads to a solution:
Since any operators in the Schrödinger picture can be expressed in terms of a complete set of operators ck†
and ck , the operators in the interaction picture can then be obtained by making the following substitutions:
Now we want to find an explicit solution to the state vector in EQ. (I.71). Evidently, EQ. (I.71) is
essentially the same as the general differential equation in EQ. (I.50) if we make the substitutions H(t) →
HI′(t) and Ψ ( t ) → Ψ I ( t ) . Therefore, we simply follow the same prescription as before by introducing a
unitary operator U I ( t , t0 ) that describes the time evolution of the state vector Ψ I ( t ) from time t0 to time t:
Ψ I ( t ) ≡ U I ( t , t 0 ) Ψ I ( t0 ) . (I.79)
Clearly U I ( t0 , t0 ) = 1 . Moreover, from EQs. (I.67) and (I.69), we can rewrite EQ. (I.79) into the following:
Ψ I ( t ) = U 0 ( 0, t ) Ψ S ( t ) = U 0 ( 0, t ) U ( t , t0 ) Ψ S ( t0 ) = U 0 ( 0, t ) U ( t , t0 ) U 0 ( t0 , 0 ) Ψ I ( t0 ) . (I.80)
Hence, we obtain an explicit expression for U I ( t , t0 ) by comparing EQs. (I.79) and (I.80):
U I ( t , t0 ) = U 0 ( 0, t ) U ( t , t0 ) U 0 ( t0 , 0 ) . (I.81)
The operator U I ( t , t0 ) is unitary and satisfies the same group property as U ( t , t0 ) . In addition, from EQs.
(I.71) and (I.79), we find that
∂ ∂
i Ψ I (t ) = i U I ( t , t0 ) Ψ I ( t0 ) = H I′ ( t ) Ψ I ( t ) = H I′ ( t ) U I ( t , t0 ) Ψ I ( t0 ) ,
∂t ∂t
∂
⇒ i U I ( t , t0 ) = H I′ ( t ) U I ( t , t0 ) . (I.82)
∂t
Consequently, similar to our previous derivation for the general time-evolution U-operator, in the interaction
picture U I ( t , t0 ) is given by:
i
U I ( t , t0 ) = 1 − dt ′ H I′ ( t ′ ) U I ( t ′, t0 )
t
∫ t0
⎛ −i ⎞ 1 t
∞ n
n =0 ⎝ ⎠ n ! t0 t0
⎧ ⎡ ⎛ −i ⎞ t ⎤⎫
= Tˆ ⎨exp ⎢ ⎜ ⎟ ∫ dt ′ H I′( t ′ ) ⎥ ⎬ . (I.83)
⎩ ⎣⎝ ⎠ t0
⎦⎭
The expressions in EQ. (I.83) will be important for our diagrammatic consideration later.
An immediate application of EQ. (I.83) is found in time-dependent perturbation theory. To the first-
order of time-dependent perturbation, the U-operator is approximated by:
⎛i⎞ t
U I ( t , t0 ) ≈ 1 − ⎜ ⎟ ∫ dt1 H I′( t1 ) , (I.84)
⎝ ⎠ t0
which is used to find the transition probabilities between the eigenstates of the unperturbed Hamiltonian H0.
Specifically, the transition probability Pi → f ( t ) from an initial eigenstate ϕi to a final eigenstate ϕ f of the
unperturbed Hamiltonian H0 due to a time-dependent perturbation Hamiltonian HI′(t) is given by the
formula:
2 2
Pi → f ( t ) = ϕ f U ( t , t0 ) ϕi = ϕ f U 0( t , 0 ) U I ( t , t0 ) U 0( 0, t0 ) ϕi ,
2 2
= ϕ f e − i H 0 t U I ( t , t 0 ) e i H 0 t0 ϕi = ϕ f U I ( t , t0 ) ϕi ,
2 2
⎛i⎞ t ⎛i⎞ t
≈ ϕ f 1 − ⎜ ⎟ ∫ dt1 H I′( t1 ) ϕi = ϕ f 1 − ⎜ ⎟ ∫ dt1 ei H 0 t1 H ′ e − i H 0 t1 ϕi ,
⎝ ⎠ t0 ⎝ ⎠ t0
2
⎛i⎞ t i (E −E ) t
= δ i f − ⎜ ⎟ ∫ dt1 e f i 1 ϕ f H ′ ϕi .
⎝ ⎠ t0
2
(i ≠ f ) 1 t i (E −E ) t
⎯⎯⎯ → 2 ∫ dt1 e f i 1 ϕ f H ′ ϕi . (I.85)
t0
In EQ. (I.85) we have taken the unperturbed Hamiltonian H0 as time-independent. The fifth line in EQ.
(I.85) is the familiar expression for the transition probability between two different eigenstates in the first-
order time-dependent perturbation theory.
Ψ H ( t ) ≡ U ( 0, t ) Ψ S ( t ) = Ψ S ( 0 ) , (I.86)
∂ ∂
i Ψ H (t ) = i Ψ S (0) = 0 , (I.87)
∂t ∂t
implying that the state vector in the Heisenberg picture is time independent. The operator in the Heisenberg
picture OH ( t ) can be related to the operator in the Schrödinger picture OS by considering the matrix element
of OS :
Ψ ′S ( t ) OS ( t ) Ψ S ( t ) = Ψ′H U ( 0, t ) OS ( t ) U ( t , 0 ) Ψ H ≡ Ψ ′H OH ( t ) Ψ H . (I.88)
From EQs. (I.87) and (I.88) we find that all the time dependence in the Heisenberg picture is ascribed to the
operator OH ( t ) whereas the corresponding state vector is time independent, in contrast to the situation in the
Schrödinger picture. We may also rewrite the operator OH ( t ) in terms of the operator in the interaction
picture by using EQs. (I.69) and (I.81):
OH ( t ) = U ( 0, t ) OS ( t ) U ( t , 0 ) = U I ( 0, t ) OS ( t ) U 0( t , 0 ) U I ( t , 0 ) U 0( 0, t ) ,
= U I ( 0, t ) ⎡⎣U 0( 0, t ) OS ( t ) U 0( t , 0 ) ⎤⎦ U I ( t , 0 ) = U I ( 0, t ) OI ( t ) U I ( t , 0 ) . (I.89)
Ψ H = Ψ S ( 0) = Ψ I (0) , (I.90)
and OS = OI ( 0 ) = OH ( 0 ) . (I.91)
In other words, all three pictures coincide at t = 0. Moreover, from EQs. (I.79) and (I.90), we obtain the
relation:
Ψ H = Ψ I ( 0 ) = U I ( 0, t0 ) Ψ I ( t0 ) , (I.92)
which implies that the eigenstate in the Heisenberg picture can be derived from the eigenstate at a given time
t0 in the interaction picture through the U I -operator.
Having introduced all three pictures and the time-evolution operators, we are now well equipped to
discuss time-dependent perturbation theory. A representative case is to consider adiabatically turning on an
interaction within a non-interacting system, assuming that the eigenstates and eigen-energies are known for
the non-interacting system described by H0. The Hamiltonian H for such a process is given by:
H =H0 +e
−ε t
H ′, (ε > 0 ) (I.93)
where ε is a small positive quantity. In general we take ε → 0+ in the end of our calculations. Clearly H →
H0 for t → ± ∞ , and H = H0 + H ′ at t = 0. Moreover, any physically significant result should not be
dependent on the exact choice of ε. Following EQ. (I.83), we introduce a new ε-dependent U I -operator, U I ε ,
so that we can derive a solution for the Schrödinger equation in the interaction picture. We have
Ψ I ( t ) ≡ U I ε ( t , t0 ) Ψ I ( t 0 ) , (I.94)
and
⎛ −i ⎞ 1 t
∞ n
dtn e ( 1
− ε t + … + tn ) ˆ
U I ε ( t , t0 ) = ∑ ⎜ ⎟ ∫ dt1 … ∫ T ⎡⎣H I′( t1 ) … H I′( tn ) ⎤⎦ .
t
(I.95)
n =0 ⎝ ⎠ n ! t0 t0
In the limit of t0 → −∞ where H → H0, we define a time-independent stationary eigenstate for H0 in the
Schrödinger picture as Φ 0 and the corresponding eigen-energy as E0, so that
H 0 Φ 0 = E0 Φ 0 . (I.96)
For the state vector in the interaction picture to be time independent and coincide with the solution for the
unperturbed Schrödinger equation in the t0 → −∞ limit, we have:
Ψ I ( t0 ) = Φ 0 = U 0( 0, t0 ) Ψ S ( t0 ) . (I.97)
Ψ S ( t 0 ) = U 0 ( t0 , 0 ) Φ 0 . (I.98)
As time increases from −∞, the interaction is turned on adiabatically for small ε and the state vector in the
interaction picture develops in time. At t = 0, the interaction reaches its full strength, and we have the
following relation
Ψ H = Ψ I ( 0 ) = U I ε ( 0, −∞ ) Ψ I ( −∞ ) = U I ε ( 0, −∞ ) Φ 0 . (I.99)
In other words, we can obtain an exact eigenstate of H from the eigenstate of the unperturbed Hamiltonian
H0 by using the ε-dependent U I -operator.
Next we want to find out whether we can still obtain physically meaningful results if we have ε →
0+. The answer to this question is proven by the following theorem of Gell-Mann and Low [M. Gell-Mann
and F. Low, Phys. Rev. 84, 350 (1951)].
U I ε ( 0, −∞ ) Φ 0 Ψ Iε ( 0 ) Ψε
≡ ≡ ,
Φ0 Ψ Iε ( 0) Φ0 Ψ Iε ( 0) Φ0 Ψε
then an exact eigenstate Ψ of the Hamiltonian H can be derived from the unperturbed ground state Φ 0
via the relation
Ψ I (0) U I ε ( 0, −∞ ) Φ 0 U I ε ( 0, −∞ ) Φ 0
Ψ ≡ = lim = lim , (I.100)
Φ0 Ψ I ( 0) ε →0 Φ 0 Ψ Iε ( 0 ) ε →0 Φ 0 U I ε ( 0, −∞ ) Φ 0
H Ψ =E Ψ , (I.101)
(H 0 − E0 ) Ψ ε = ( H 0 − E0 ) U I ε ( 0, −∞ ) Φ 0 = ⎣⎡H 0 , U I ε ( 0, −∞ ) ⎦⎤ Φ 0 . (I.102)
To evaluate the commutator in EQ. (I.102), we recall the expression for U I ε in EQ. (I.95) and so we first
examine the following commutator:
⎡H 0 , H I′( ti ) H I′( t j ) H I′( tk ) ⎤⎦ = ⎡⎣H 0 , H I′( ti ) ⎤⎦ H I′( t j ) … H I′( tk ) + H I′( ti ) ⎡⎣H 0 , H I′( t j ) ⎤⎦ … H I′( tk )
⎣
+H I′( ti ) H I′( t j ) … ⎣⎡H 0 , H I′( tk ) ⎦⎤ . (I.103)
Here H I′( ti ) H I′( t j ) H I′( tk ) represents an arbitrary time ordering of the n indices. From EQ. (I.73), we
have
∂
−i H I′( t ) = ⎡⎣H 0 , H I′( t ) ⎤⎦ , (I.104)
∂t
so that each commutator in EQ. (I.103) yields a time derivative of HI′(t) and EQ. (I.103) for all possible time
orderings becomes:
⎛ n ∂ ⎞
⎡H 0 , H I′( ti ) H I′( t j ) … H I′( tk ) ⎤ = ⎜ ∑
⎦ i ν =1 ∂t ⎟ ⎣H I ( ti ) H I ( t j ) … H I ( tk ) ⎦ .
⎡ ′ ′ ′ ⎤ (I.105)
⎣
⎝ ν ⎠
∂
(H 0 − E0 ) Ψ ε = −H ′ Ψ ε + i ε g
∂g
Ψε , (I.107)
so that
∂
(H − E0 ) Ψ ε = i ε g
∂g
Ψε . (I.108)
[Φ ]
−1
Multiplying EQ. (I.108) by 0
Ψε Φ 0 , we obtain
Φ 0 ( H − E0 ) Ψ ε Φ 0 ( E0 + H ′ − E0 ) Ψ ε Φ0 H ′ Ψε
= = ≡ ( E − E0 )
Φ0 Ψε Φ0 Ψε Φ0 Ψε
Φ 0 ∂ ∂g Ψ ε ∂
=i εg =i εg ln Φ 0 Ψ ε , (I.109)
Φ0 Ψε ∂g
where we have used the fact that ( ∂ ∂g ) Φ 0 = 0 . Moreover, from EQs. (I.108) and (I.109), we find
⎛ ∂ ⎞ Ψε Ψε ⎡ ∂ ⎤ Ψε
⎜ H − E0 − i ε g ⎟ = ⎢ i εg ln Φ 0 Ψ ε ⎥ = ( E − E0 ) , (I.110)
⎝ ∂g ⎠ Φ 0 Ψ ε Φ0 Ψε ⎣ ∂g ⎦ Φ0 Ψε
which yields
Ψε ∂ ⎡ Ψε ⎤ Ψ I (0)
(H − E ) Φ0 Ψε
=i εg ⎢ ⎥
∂g ⎣ Φ 0 Ψ ε ⎦
(
⎯⎯⎯)ε →0
→ (H − E ) Φ0 Ψ I ( 0)
= 0. (I.111)
Thus, we have proven EQ. (I.101) by taking ε → 0 in EQ. (I.111). We remark that while it is justified to take
the limit of ε → 0 in EQ. (I.111), the same cannot be trivially applied to EQ. (I.109), because the term
( ∂ ∂g ) ln Φ 0 Ψ ε in fact diverges in the ε → 0 limit, whereas the term ( ∂ ∂g ) ⎡⎣ Ψ ε Φ 0 Ψ ε ⎤⎦ in EQ.
(I.109) is finite for ε → 0.
Before closing our discussion of the time-dependent perturbation theory, we demonstrate in the
following that the time-independent perturbation theory is equivalent to a special case of the time-dependent
perturbation theory if we take a time-evolution operator U I (0, −∞ ) .
From EQ. (I.95), the operator U I (0, −∞) can be explicitly given by the following:
⎛ −i ⎞ 1 0
∞ n
dtn e ( 1
− ε t + … + tn ) ˆ
U I ε ( 0, −∞ ) = ∑ ⎜ ⎟
0
∫ dt1 … ∫ T ⎡⎣H I′( t1 ) … H I′( tn ) ⎤⎦ . (I.112)
n =0 ⎝ ⎠ n ! −∞ −∞
Thus, the lowest order term in the construction of the matrix element Φ f U I (0, −∞ ) Φ 0 between two
unperturbed eigenstates Φ 0 and Φ f is:
⎛ −i ⎞ 0 ⎛ −i ⎞ 0
Φ f U I 1 ( 0, −∞ ) Φ 0 = ⎜ ⎟ ∫ dt1 e 1 Φ f H I′( t1 ) Φ 0 = ⎜ ⎟ ∫ dt1 e 1 Φ f ei H 0 t1 / H ′ e − i H 0 t1 / Φ 0
−ε t −ε t
⎝ ⎠ −∞ ⎝ ⎠ −∞
⎛ −i ⎞ 0 i ( E − E − i ε ) t1 Φ f H ′ Φ0
= ⎜ ⎟ ∫ dt1 e f 0 Φ f H ′ Φ0 = − , (I.113)
⎝ ⎠ −∞ (E f − E0 − i ε )
where Ef and E0 are the eigenvalues of the unperturbed states. Equation (1.113) is consistent with the familiar
expression for the perturbation amplitude in the first-order time-independent perturbation theory.
Similarly, if we denote the intermediate state vectors by Φ p and use the identity ∑ p
Φp Φp = 1,
the second-order term in the construction of the matrix element Φ f U I 2 (0, −∞ ) Φ 0 becomes:
2
⎛ −i ⎞ 0
Φ f U I 2 ( 0, −∞ ) Φ 0 = ⎜ ⎟ ∫ dt1 ∫ dt2 e ( 1 2 )
t1 −ε t + t
⎝ ⎠ −∞ −∞
⎡ ⎤
× ⎢ ∑ Φ f ei H 0 t1 / H ′ e − i H 0 t1 / Φ p Φ p e i H 0 t2 / H ′ e − i H 0 t2 / Φ 0 ⎥ ,
⎣ p ⎦
⎛ +i ⎞ i ( E − E − i ε ) t1 i ( E − E − i ε ) t1 Φ f H ′ Φ p Φ p H ′ Φ0
= ⎜ ⎟ ∑ ∫ dt1 e f p
0
e p 0 ,
⎝ ⎠ p −∞
(E p − E0 − i ε )
Φ f H ′ Φ p Φ p H ′ Φ0
=∑ . (I.114)
p (E f − E0 − i 2 ε )( E p − E0 − i ε )
Evidently EQ. (I.114) is consistent with the second-order time-independent perturbation theory.
In comparison with the ordinary perturbation theory, the time-dependent formalism introduced in
this section is more convenient in various ways. For instance, the time-dependent formalism can better
handle the poles encountered in the perturbation theory because we may use the following relation to deal
with the results in EQs. (I.113) and (I.114):
1 ⎛1⎞
lim = P ⎜ ⎟ ± iπδ ( x ) , (I.115)
α →0 + ( x ∓ iα )
⎝ x⎠
where P(…) represents the principal value of the function inside the parenthesis. Another advantage is that
one can easily separate parts of the problem that are associated with the disconnected parts of the system.
This point will be verified in general terms when we introduce Feynman diagrams in Part II. For now, we use
a simple example in the following to illustrate this concept.
⎧ ⎡ i 0 ⎤⎫ ⎧ ⎡ i 0 i ( H a +H b ) t / − i ( H 0a + H 0b ) t / ⎤⎫
U I ( 0, −∞ ) = Tˆ ⎨exp ⎢ − ∫ dt V ( t ) ⎥ ⎬ = Tˆ ⎨exp ⎢ − ∫ dt e 0 0 ( Va + Vb ) e ⎥⎦ ⎬
⎩ ⎣ −∞
⎦⎭ ⎩ ⎣ −∞
⎭
⎧
⎩
⎡ i 0
⎣
(−∞
a 0
−∞
b b ⎤⎫
= Tˆ ⎨exp ⎢ − ∫ dt ei H 0 t / Va e −i H 0 t / + ∫ dt ei H 0 t / Vb e − i H 0 t / ⎥ ⎬
a
⎦⎭
)
= U Ia ( 0, −∞ ) U Ib ( 0, −∞ ) . (I.116)
Therefore the U I -operator can be written as the product of independent operators if the corresponding
Hamiltonian involves disconnected parts.
Generally speaking, the low energy excitations in solids may be described in terms of quantized units
of energy in various fields. These excitations often combine both aspects of waves and particles, and are
typically divided into bosonic and fermionic excitations. Interestingly, for two-dimensional electronic
systems, the low-energy excitations can be characteristic of neither bosons nor fermions. Rather, they obey
fractional or even non-abelian statistics. Such excitations are known as anyons, which will not be covered in
this course due to time constraints. In the following we list definitions of various representative low-energy
excitations in three-dimensional solids.
Quasiparticles: electrons dressed by the interactions with the electron gas in a metal -- (fermions)
Polarons: charged particles associated with the polarization field in ionic crystals -- (fermions)
Phonons: elastic excitations of the crystalline lattice -- (bosons)
Magnons: elementary excitations of electron spins coupled together by the exchange interactions in a
magnetic system -- (bosons)
Plasmons: collective Coulomb excitations of the electron gas in a metal -- (bosons)
Excitons: coupled electron-hole pairs associated with the dielectric polarization field -- (bosons)
As an example, we consider the collective elastic motion of an electron gas relative to the rigid
background of positive charge, known as the plasmons. The Hamiltonian density of the electron gas can be
expressed in terms of the displacement fields uα and the conjugate fields πα (α: polarization) as follows:
1 1 ∂u ∂uβ 1
H el = ∑ παπα + λ ∑ α
2 α β ∂r
+ ( ρe − ρ0 ) ϕ( r ) , (I.117)
2ne me α , α ∂rβ 2
where ne and ρe denote the volume density and the charge density of electrons, respectively; ρ0 is the uniform
positive charge background, λ represents the bulk modulus, and ϕ( r ) is the electrostatic potential.
The plasmons are associated with the longitudinal waves of electron density fluctuations that destroy
the local charge neutrality, thereby inducing a Coulomb restoring force. To understand the nature of
plasmons, we begin with the charge density fluctuations δρ due to local dilation of the electron gas:
∂ul ( r )
δρ ( r ) = ρe ( r ) − ρ0 ( r ) = −ne e , ( l: longitudinal polarization ) (I.118)
∂rl
and ∇ 2ϕ ( r ) = −4π δρ ( r ) . (I.119)
Consequently, the spatial integration of the third term in EQ. (I.117) becomes
1 i ( k + k ′) i r ⎛ ne e l ⎞ ⎛ −i 4π ne eQkl ′ ⎞
∫ 2 () () ∑
d r δρ r ϕ r =
k ,k ′
∫ d r e ⎜
⎝
− i
2
kQk ⎟ ⎜
⎠⎝ k′
⎟
⎠
= 2π ne2 e 2 ∑ Qkl Q−l k . (I.122)
k
∫ dr π α ( r ) π α ( r ) = ∑ k
Pkα P−αk . (I.123)
In addition, the elastic energy in the second term of EQ. (I.117) can be rewritten into the form
λ
∑∑
2 αβ
k
kα kβ Qα Q β
,
k −k . (I.124)
Consequently, for k eˆl and from EQs. (I.122) – (I.124), we obtain the Hamiltonian of the electron gas:
⎡ 1 1 ⎤
H el = ∫ dr H el = ∑ ⎢ Pk P− k + ( λ k 2 + 4π ne2 e 2 ) Qk Q− k ⎥ . (I.125)
k ⎣ 2ne me 2 ⎦
From EQ. (I.125), we may define the plasmon frequency ωk and the plasma frequency ωp as follows:
⎛ λ ⎞ 2 4π ne2 e 2 ⎛ λ ⎞ 2
ωk2 = ⎜ ⎟k + ≡⎜ ⎟ k + ωp ,
2
(I.126)
⎝ ne me ⎠ ne me ⎝ ne me ⎠
1/ 2
⎛ 4π ne2 e2 ⎞
ωp ≡ ⎜ ⎟ . (I.127)
⎝ ne me ⎠
In general the plasmon frequencies for collective electron excitations are much higher than the
phonon frequencies associated with the lattice vibration because of the much smaller mass density in the
former. More quantitatively, if we take a typical value of ne ~ 1023 cm−3 and me ~ 10−27 g, we obtain ωp ~ 1016
s−1. On the other hand, even if we take the maximum k-value in the crystal, i.e., kmax ~ 108 cm−1, the first
quantity in EQ. (I.126), [λ kmax
2
/( ne me )] , is still much smaller than ωp2 if we take a typical value λ ~ 109
dynes/cm2. Consequently, the plasmon excitations of the electron gas in a solid can be regarded as nearly
uniform.
1/ 2 1/ 2
⎛n mω ⎞ ⎛ 1 ⎞
b =⎜ e e k ⎟
†
Q −i⎜
*
⎟ Pk , (I.128)
⎝ 2ne me ωk ⎠
k k
⎝ 2 ⎠
⎛ 1⎞
H el = ∑ ωk ⎜ bk†bk + ⎟ . (I.129)
k ⎝ 2⎠
Equation (I.129) is analogous to EQ. (I.49), except that the dispersion relations associated with the bosonic
excitations differ in the two cases.
Further Readings
Review of aspects of modern condensed matter physics:
1. N.-C. Yeh: Bulletin of AAPPS, Vol. 18, No. 2, pg. 11--29 (2008)
Second quantization and quantum dynamics:
2. Fetter and Walecka, “Quantum Theory of Many-Particle Systems”: Sections 1 – 3 and 6.
3. Kittel, “Quantum Theory of Solids”: Chapters 1 and 2.
4. Schweber, “An Introduction to Relativistic Quantum Field Theory”, Chapter 6.
5. Merzbacher, “Quantum Mechanics”, Chapters 14, 21 and 22.
[ z − L ( r )] G ( r, r′; z ) = δ ( r − r′ ) , (II.1)
subject to certain homogeneous boundary conditions on the surface S of the domain Ω of r and r′.
Assuming that {φ }n
is the complete orthonormal set of eigenfunctions of L, subject to the same
conditions on the surface S as G(r, r′; z), and that {λn } is the eigenvalues, we may express the Green
function G as follows:
1 φ φ φ φ
G ( z) = = Σ′n n n + ∫ dn n n , (II.2)
( z − L) z − λn z − λn
or equivalently
φ ( r ) φn* ( r′ ) φ ( r ) φn* ( r ′ )
G ( r , r ′; z ) = Σ′n n + ∫ dn n , (II.3)
z − λn z − λn
where L ( r ) φn = λn φn , Σ′ n
denotes the sum over the eigenfunctions of the discrete spectrum, and ∫ dn is
an integration over the continuous spectrum. We also note that the following orthogonality relations:
φn φm = δ nm = ∫ d 3 r φn*( r′ ) φm( r ) , ∑ φ ( r ) φ ( r′ ) = δ ( r − r′ ) .
n n
*
n
(II.4)
Moreover, since L is hermitian, all eigenvalues λn’s are real, implying that the singularities of G(z) are all on
the real axis. For branch cuts of G(z), we may define G+(λ) and G−(λ) so that
G ± ( λ ) ≡ lim G ( λ ± is ) . (II.5)
s →0+
The simple poles and branch cut of G(z) on the λ-s complex plane are illustrated in Fig. II.1.1.
[ z − L( r ) ] u ( r ) = f ( r ) , (II.6)
where
u ( r ) = ∫ d 3 r ′ G ( r, r ′; z ) f ( r ′ ) if z ≠ λn ,
u ( r ) = ∫ d 3 r ′ G ( r , r ′; z ) f ( r′ ) + φ ( r ) if z = λ ∈ {branch cuts of G(z)},
and u(r), φ(r), G(r, r′; z), all satisfy the same boundary conditions on S, with φ(r) being the solution to
the homogeneous equation:
[ z − L( r ) ] φ ( r ) = 0 . (II.7)
Here we note that there is no solution to u(r) if z coincides with λn unless f (r) is orthogonal to all
eigenfunctions of λn.
s z-plane
Simple poles of G(z), Natural boundary of Branch cut of G(z),
discrete spectrum of L, G(z), continuous continuous spectrum
localized eigenstates. spectrum of L, localized of L, extended
eigenstates. eigenstates. λ
〈2〉 Obtaining information for the eigenfunctions and eigenvalues of L(r). That is, finding the poles of G(z),
which correspond to the discrete eigenvalues of L(r); the residues at the poles, which correspond to the
products of φn ( r ) φn* ( r ′ ) , provided that φn ( r ) are not degenerate; and the branch cuts of G(r, r′; z) along
the real axis, which correspond to the continuous spectrum of L(r).
〈3〉 Deriving the density of states N (λ) associated with the operator L(r):
From G*(r, r′; z) = G(r′, r; z*) and EQ. (II.5), we have G−(r, r′; z) = [G+(r′, r; z)]*. Consequently,
Re{G−(r, r′; z)} = Re{[G+(r, r′; z)]}, Im{G−(r, r′; z)} = − Im{[G+(r, r′; z)]}, and
G ( r , r ′; λ ) ≡ G + ( r , r ′; λ ) − G − ( r , r ′; λ ) = −2π i ∑ n δ ( λ − λn ) φn ( r ) φn* ( r ′ )
1 ⎛1⎞
lim = P ⎜ ⎟ ∓ iπδ ( x ) . (II.9)
s →0+ x ± is ⎝x⎠
⎡ φn ( r ) φn* ( r ) ⎤
G ± ( r , r; λ ) = P ⎢ ∑ n ⎥ ∓ π i ∑ n δ ( λ − λn ) φn ( r ) φn ( r ) ,
*
(II.10)
⎣ ( λ − λn ) ⎦
which yields the trace:
{ }
Tr G ± ( λ ) = ∫ d 3 r G ± ( r , r; λ )
⎡ 1 ⎤ ⎡ 1 ⎤
= P ⎢∑ n ⎥ ∓ iπ ∑ n δ ( λ − λn ) ≡ P ⎢ ∑ n ⎥ ∓ iπ N ( λ ) . (II.11)
⎣ ( λ − λn ) ⎦ ⎣ ( λ − λn ) ⎦
Therefore, ⎡⎣N ( λ ) d λ ⎤⎦ is the number of states in the interval [ λ , λ + d λ ] , and the quantity
ρ ( r; λ ) ≡ ∑ n δ ( λ − λn ) φn ( r ) φn* ( r ) (II.12)
N ( λ ) = ∫ d 3 r ρ ( r; λ ) . (II.13)
Consequently,
1
ρ ( r; λ ) = ∓
π
{
Im G ± ( r , r; λ ) , } (II.14)
1
and N (λ ) = ∓
π
{
Im Tr ⎡⎣G ± ( λ ) ⎤⎦ . } (II.15)
Furthermore,
φn ( r ) φn* ( r′ ) φn ( r ) φn* ( r ′ )
G ( r, r ′; z ) = ∑ n = ∫ d λ ∑ n δ ( λ − λn )
∞
,
z − λn −∞
z −λ
i ∞ G ( r , r ′; z )
=
2π ∫ −∞
dλ
z−λ
. (II.16)
〈4〉 Using G0 ( z ) = ( z − L0 ) and L1 to obtain information for the eigenfunctions and eigenvalues of L = L0 +
−1
L1. Specifically, if the operator L is represented by the total Hamiltonian H of the system, so that for H
= H 0 + H 1, G0 ( z ) = ( z − H 0 ) and G ( z ) = ( z − H )
−1 −1
. Hence,
1
G = G0 = G0 + G0 H 1G0 + G0H 1G0H 1G0 + …
1 − H 1G0
= G0 + G0 H 1G = G0 + GH 1G0 . (II.17)
We further note that in EQ. (II.17), the equivalence of G0 + G0 H 1G and G0 + GH 1G0 (corresponding to
the equivalence of G0 (1 − H 1G0 ) and (1 − G0H 1 ) G0 ) is generally true without requiring specific
−1 −1
Next, let’s consider time-dependent Green functions. The Green function associated with a first-
order (in time) partial differential equation of the form
⎡i ∂ ⎤
⎢⎣ C ∂t − L ( r ) ⎥⎦ φ ( r , t ) = 0 , (II.18)
⎡ i ∂ − L r ⎤ g r, t ; r ′, t ′ = δ r − r′ δ t − t ′
⎢⎣ C ∂t ( )⎥ ( ) ( ) ( ) (II.19)
⎦
subject to certain boundary conditions on the surface S of the domain Ω of r and r′; L(r) is a linear, hermitian
and time-independent operator with a complete set of eigenfunctions {φ (r)}; and C is a constant. If C is real,
it may be taken as positive without losing generality. In this case, g ( r , t ; r ′, t ′ ) is associated with a
Schrödinger type equation. On the other hand, if C is imaginary, g ( r , t ; r ′, t ′ ) is associated with a diffusion
type equation.
Generally for time-independent L(r), g ( r , t ; r ′, t ′ ) may be expressed in terms of the time difference
τ ≡ t − t ′ . In this case, the Fourier transform of g ( r, r ′;τ ) is g ( r, r ′; ω ) , which is directly related to the
time-independent Green function G ( r , r ′; z ) if we take z = ω/C and allow ω to be complex. Thus, we have
∞ dω ′ ⎛ ω ′ ⎞ − iω ′τ
g > ,< ( r, r′;τ ) = ∫ G + , − ⎜ r, r ′; ⎟e . (II.20)
−∞
2π ⎝ C⎠
The Green function associated with a second-order (in time) differential equation is defined as the
solution of
⎡ 1 ∂2 ⎤
⎢ − C 2 ∂t 2 − L ( r ) ⎥ g ( r , r′;τ ) = δ ( r − r ′ ) δ (τ ) . (II.21)
⎣ ⎦
The Fourier transform g ( r, r ′; ω ) in this case is related to the time-independent Green function G ( r , r′; z )
⎛ ω2 ⎞
by g ( r, r ′; ω ) = G ⎜ r, r ′; ⎟2
. The special case of L = −∇ 2 reduces to the wave equation.
⎝ C ⎠
ω-plane
>
g Re{ω}
ω-plane
g< Re{ω}
ω-plane
g Re{ω}
where g (τ ) ≡ g > (τ ) − g < (τ ) , with the corresponding integration paths in the ω-plane illustrated in Fig.
II.1.2.
⎡i ∂ ⎤
⎢⎣ C ∂t − L ( r ) ⎥⎦ ψ ( r , t ) = f ( r, t ) , (II.23)
with ψ ( r , t ) given by
ψ ( r , t ) = φ ( r, t ) + ∫ d 3 r ′∫ dt ′ g > ( r, r′; t − t ′ ) f ( r′, t ′ ) , (II.24)
〈3〉 Use g0(τ) and L1(r) to obtain information for the solution of
⎡i ∂ ⎤
⎢⎣ C ∂t − L ( r ) ⎥⎦ ψ ( r , t ) = 0 , (II.25)
where g0(τ) is the Green function of solution of L0(r), and L = L0 + L1, similar to what we have done in
EQ. (II.17).
Having reviewed some basic properties of Green functions, next we consider the effect of
temperature, in preparation for applying Green functions to the derivation of thermodynamic properties.
II.2. Temperature Dependent Quantum Field Theory in the Path Integral Formalism
Temperature is one of the most fundamental variables in the description of a condensed matter
physics system. In reality, temperature T can be naturally introduced into the zero-temperature quantum field
theory through a Wick rotation t = − itE if one identifies the inverse of temperature (β = 1/T) with the
imaginary time. This concept is best described in terms of path integral formalism. For completeness, before
we proceed with the description of temperature-dependent quantum field theory, it is useful to briefly review
the path integral formalism.
Let’s consider the propagation of a quantum system, governed by a Hamiltonian H, from a point qI
to a point qF in time t. The amplitude for the propagation is given by qF exp ( − iH t ) qI , where we have
used the Dirac bra and ket notations, and exp ( − iH t ) is a unitary operator. If we divide the time t into N
segments and define δt = t/N, and recall that |q〉 forms a complete set of states so that ∫ dq |q〉 〈q| = 1, we can
rewrite qF exp ( − iH t ) qI into the following:
qF | e − i H t | qI
⎛ N −1 ⎞
= ⎜ ∏ ∫ dq j ⎟ qF | e − i Hδ t | qN −1 qN −1 | e − i Hδ t | qN − 2 ... q2 | e − i Hδ t | q1 q1 | e − i Hδ t | qI . (II.26)
⎝ j =1 ⎠
⎛ p2 ⎞ dp ⎛ p2 ⎞ dp − iδ t ( p2 / 2 m )
q j +1 exp ⎜ −iδ t ⎟ qj = ∫ q j +1 exp ⎜ −iδ t ⎟ p p qj = ∫ e q j +1 p p q j
⎝ 2m ⎠ 2π ⎝ 2m ⎠ 2π
1
⎛ p2 ⎞ ⎛ −im ⎞ 2 ⎛ ⎛ m ⎞ ⎡ (q j +1 − q j ) ⎤ ⎞
2
⎟ exp ⎡⎣ip ( q j +1 − q j ) ⎤⎦ = ⎜
dp
=∫ exp ⎜ −iδ t ⎟ exp ⎜⎜ iδ t ⎜ ⎟ ⎢ ⎥ ⎟⎟ . (II.27)
2π ⎝ 2m ⎠ ⎝ 2πδ t ⎠ ⎝ 2 ⎠⎣ δ t ⎦ ⎠
⎝
t
where we have taken qI = q0 and qF = qN. In the δ t → 0 limit, [(qj+1−qj)/δt]2 → q 2 , and δt ∑j → ∫ dt . If we
0
further define the integral over paths as
N
⎛ − im ⎞ 2 ⎛ N −1 ⎞
∫ Dq(t ) = Nlim ⎜⎜ ⎟⎟ ⎜⎜ ∏ ∫ dq j ⎟⎟ , (II.30)
→∞ 2πδ t
⎝ ⎠ ⎝ j =1 ⎠
⎛ t 1 ⎞
qF | e − i H t | qI = ∫ Dq (t ′) exp ⎜ i ∫ dt ′ mq 2 ⎟ . (II.31)
⎝ 0 2 ⎠
The above result can be generalized to a Hamiltonian for a particle in a potential V (q), so that H =
p2
/(2 m) + V( q ), and
⎛ t ⎡1
⎝ 0
⎣2
⎤⎞
⎦⎠
t
0 (
qF | e − i H t | qI = ∫ Dq (t ′) exp ⎜ i ∫ dt ′ ⎢ mq 2 − V ( q ) ⎥ ⎟ = ∫ Dq (t ′) exp i ∫ dt ′L (q, q ) , ) (II.32)
In most of the problems that interest us, we want to consider the amplitude between an initial state |I〉
and a final state |F〉 rather than between the initial and the final positions |qI〉 and |qF〉. We therefore rewrite
EQ. (II.32) into the following form:
⎡N 1 ⎤
S (q ) = ∫ dt ′ ⎢ ∑ ma qa 2 − V (q1 , q2 ,..., qN ) ⎥ = ∫ dt ′ L [ qa (t ′), qa (t ′) ] ,
t t
(II.34)
0
⎣ a =1 2 ⎦ 0
and V(q1, q2,…, qN) is the potential energy, which includes the interaction energy among particles. We may
take the continuum limit so that the discreteness of particles is replaced by a four-dimensional variable x ≡ (t,
x), and the discrete coordinates and momentum of particles by the field ϕ (x) and its space-time derivatives
∂μϕ (x), where ∂μ ≡ (∂/∂t, ∂/∂xi) and the superscript index i runs through three spatial dimensions. The action
S is now given by a 4-dimensional integral over the Lagrangian density L(ϕ(x),∂μϕ (x)): S = ∫ d 4 x L (ϕ , ∂ μϕ) ,
so that Z = 〈0| exp(− iH t) |0〉 becomes (after restoring the Planck constant)
⎡i ⎤
Z = ∫ Dϕ exp ⎢ ∫ d 4 x L(ϕ , ∂ μϕ ) ⎥ = ∫ Dϕ exp ⎡⎣ iS (ϕ , ∂ μϕ ) ⎤⎦ . (II.35)
⎣ ⎦
Using the Euler-Lagrangian variational procedure by minimizing the action and also integration by parts:
⎡δ L δL ⎤ ⎡⎛ δ L δL ⎞ ⎤
δ S = ∫ d 4x ⎢ δϕ + δ ∂ μϕ ⎥ = ∫ d 4 x ⎢⎜⎜ − ∂μ ⎟ δϕ ⎥ = 0 ,
⎣⎢ δϕ δ ∂ μϕ ⎦⎥ ⎢⎣⎝ δϕ δ∂ μϕ ⎠⎟ ⎥⎦
Having established the path integral formalism, we consider a specific example, the Minkowskian
path integral for scalar field theory in d-dimensional spacetime at T = 0 with a Lagrangian density
L (ϕ ) ≡ ⎡⎣( ∂ϕ ) 2 ⎤⎦ − V (ϕ ) :
2
⎡i d ⎛1 ⎞⎤
Z = ∫ Dϕ exp ⎢ ∫ d x ⎜ ( ∂ϕ ) − V (ϕ ) ⎟ ⎥ ,
2
(II.37)
⎣ ⎝2 ⎠⎦
where ϕ represents the scalar field, and V (ϕ ) denotes the potential of the scalar field. Under a Wick rotation
t = − itE, EQ. (II.37) is converted into the Euclidean functional integral
⎡ 1 d ⎛1 ⎞⎤ ⎡ 1 ⎤
Z = ∫ Dϕ exp ⎢ − ∫ d E x ⎜ ( ∂ϕ ) + V (ϕ ) ⎟ ⎥ ≡ ∫ Dϕ exp ⎢ − S (ϕ ) ⎥ ,
2
(II.38)
⎣ ⎝ 2 ⎠⎦ ⎣ ⎦
⎡ 1 ⎤
Z = ∏ ∫ dqi exp [ − β H (q1 , q2 , , qN ) ] ↔ Z = ∫ Dϕ exp ⎢ − S (ϕ ) ⎥ . (II.39)
i ⎣ ⎦
Hence, the Euclidean quantum field theory in d-dimensional spacetime is consistent with classical statistical
mechanics in d-dimensional space.
Let’s now consider the situation in quantum statistical mechanics. The partition function for a system
described by the Hamiltonian H is
{ }
Z = Tr e − β H = ∑ n e − β H n .
n
(II.40)
If we replace the time t in the expression of F e − i H t I in EQ. (II.33) by (−iβ) and set I = F = n , we
reach the following expression:
{ } Dq exp ⎡ − ∫ dτ L ( q ) ⎤ ,
β
Z = Tr e − β H = ∫
PBC ⎣ 0 ⎦
n =∞
Dq exp ⎡ − ∫ dτ H (q ) ⎤ n
β
= ∑ n ∫PBC ⎣ 0 ⎦
(II.41)
n =−∞
where “PBC” refers to periodic boundary condition because of the trace operation, the Lagrangian L(q) =
½(dq/dτ)2 +V(q) under the change of variable from dt to (−idτ ) is precisely the Hamiltonian H in the
Euclidean time τ, and the integration over τ runs from 0 to β. Since we are taking the trace in EQ. (II.41),
the boundary condition q(0) = q(β) must be satisfied. Equation (II.41) can be readily extended to field theory
if we take H as the Hamiltonian of a quantum field theory in D-dimensional space (or equivalently, d =
(D+1)-dimensional spacetime) and consider the following correspondence:
q ↔ ϕ, L ( q ) ↔ ∫ d D x H (ϕ ) = H ,
where H (ϕ ) denotes the Hamiltonian density. Hence, EQ. (II.41) may be rewritten as follows:
{ }
β
Z = Tr e − β H = ∫ Dϕ exp ⎡ − ∫ dτ ∫ d D x H (ϕ ) ⎤ ,
PBC ⎣ 0 ⎦
∞
β
= ∑ n ∫ Dϕ exp ⎡ − ∫ dτ H (ϕ ) ⎤ n (II.42)
n =−∞
PBC ⎣ 0 ⎦
ϕ ( x, 0 ) = ϕ ( x, β ) (II.43)
is satisfied. The boundary condition in EQ. (II.43) requires that the Euclidean frequency ω can only take on
discrete imaginary values:
Equation (II.44) is satisfied for bosonic fields, and ωn is known as the Matsubara frequency. For fermionic
fields, on the other hand, the corresponding Matsubara frequency is given by
so that the bosonic field operator ϕ ( x ,τ ) is replaced by the Dirac spinor field operator ψ ( x ,τ ) . The
difference in the expressions of the Matsubara frequency between bosons and fermions is the result of
Grassmann algebra associated with the spinor field operators of fermions in the path integrals, which is not
explicitly discussed here but can be found in typical relativistic field theory texts.
The expressions given in EQs. (II.41) -- (II.43) suggest that a field theory at finite temperature is
equivalent to rotate it to Euclidean space and impose the boundary condition in EQ. (II.43). In other words, a
Euclidean quantum field theory in (D+1)-dimensional spacetime with 0 ≤ τ < β is equivalent to quantum
statistical mechanics in D-dimensional space. In the limit of zero temperature where β → ∞, EQ. (II.42)
becomes the standard Wick rotated quantum field theory over an infinite spacetime, as expected. Thus, the
analytic continuation between [exp(− iH t)] in quantum physics and [exp(−βH)] in thermal physics leads to
a remarkable result suggesting that the notion of temperature is in fact equivalent to cyclic imaginary time.
Hence, we are able to deal with finite-temperature perturbation theory and Feynman diagrams by using the
Feynman rules in T = 0 and then replacing the frequency ω by the Matsubara frequency ωn.
In the high temperature limit T → ∞, we find that the dominant contribution to the propagator is
from the term n = 0 because large n values correspond to fast oscillating terms whose amplitude effectively
vanish. The fact that only the term n = 0 dominates implies that the Feynman diagrams are effectively
evaluated in D-dimensional space. In other words, the Euclidean quantum field theory in d-dimensional
spacetime is equivalent to high-temperature quantum statistical physics in D-dimensional space. One
important application of quantum field theory at finite temperature is to cosmology, because the early
universe may be considered as a soup of elementary particles at high temperatures. Finite-temperature
quantum field theory is also very important to condensed matter physics, as we shall see in the subsequent
chapters of this summer course.
ψ ( r , t ) = ei H tψ ( r , 0 ) e − i H t = ∑ anψ n ( r ) e − iε nt ,
n
and H = ∑ε a an
†
n n
. (II.47)
n
As before, the creation and annihilation operators an† and an obey the following communication relations:
an an†′ ∓ an†′ an = δ nn′ , an an′ ∓ an′ an = 0 , an† an†′ ∓ an†′ an† = 0 , (II.48)
The Green function g (r, t ; r ′, t ′) = g ( x, x′) = g ( x − x′) has the physical significance of a single-
particle propagator. For an N-particle system, each propagator corresponds to the propagation of a particle
(
g ( x − x′) ≡ −i T ψ ( x)ψ † ( x′) ) , (II.49)
where the symbol A denotes thermal average of the quantity A over the grand statistical ensemble:
− β (ε − μ N )
∑ i A i exp ⎡⎣ − β ( ε i − μ Ni ) ⎦⎤ Tr ⎡⎣ Ae ⎤
⎦,
A ≡ i
= (II.50)
∑i ⎣
exp ⎡ − β ( ε i − μ N )
i ⎦⎤ ⎡
Tr ⎣e − β (ε − μ N )
⎤
⎦
N is the total number of particles, μ the chemical potential, { i } the common eigen-functions, and
( ) ( )
T ψ ( x)ψ † ( x′) = T ψ (r , t )ψ † (r ′, t ′) = ψ (r , t )ψ † (r ′, t ′), t > t ′;
= ±ψ † (r′, t ′)ψ (r , t ), t < t ′. (II.51)
In EQ. (II.51) the upper (lower) sign refers to bosons (fermions). It should be noted that the definition given
in EQ. (II.49) has not explicitly included the spin indices of the field operators. For simplicity we temporarily
neglect the spin indices, which is valid if we only deal with spin-independent interactions. We shall restore
the spin indices later for generality when we discuss the diagrammatic analysis.
It is convenient (as we shall see later) to define different forms of Green functions besides the one in
EQ. (II.49). The customary forms of Green functions in addition to EQ. (II.49) are listed below:
where the upper (lower) sign refers to bosons (fermions). In the zero temperature limit the thermal average of
the physical quantity A becomes:
A → 0 A0 as T → 0+ , (II.57)
In the presence of interactions, the Hamiltonian H (= K + V) of the system consists of both the
kinetic energy of particles K:
2 2 2
k2
K=
2m
3
(
∫ d r ∇ψ ∇ψ = −
†
) 2m
3 †
(
∫d r ψ ∇ψ = ∑
2
) 2m
ak† ak , (II.58)
k
Consequently, the Schrödinger’s equation expressed in terms of the field operators becomes
⎛ ∂ ∇ 2r ⎞
⎟ψ ( r, t ) = ⎡⎣ ∫ d r ′′ψ ( r′′, t )V ( r′′ − r )ψ ( r′′, t ) ⎤⎦ψ ( r, t ) , (II.60)
3 †
⎜i +
⎝ ∂t 2m ⎠
and the corresponding Green function g ( x, x′) will involve the two-particle Green function g2:
⎛ ∂ ∇r2 ⎞
⎟ g ( x, x′ ) = δ ( x − x′ ) ± i ∫ d x1 V ( r − r1 ) g 2( x, x1 ; x′, x1 ) t1 = t , (II.61)
4 +
⎜i +
⎝ ∂t 2m ⎠
where x1+ ≡ lim ( r1 , t1 + s ) , and the two-particle Green function is given by:
s →0 +
where the chronological operator T arranges the operators in chronological order so that the earliest time
appears on the far right and the latest time appears on the far left. In addition, for fermions only, a factor ±1
is introduced depending on whether the time-ordered product is an even or odd permutation of the original
ordering. From EQ. (II.61), we note that the pair interactions in EQ. (II.59) give rise to the added
contribution of the two-particle Green function to g ( x, x′) . Similarly, we expect that the differential equation
for g2 will involve the three-particle Green function g3, and so on. The n-particle Green function gn is defined
as:
Thus, the existence of interaction complicates the calculation of the Green function g ( x, x′) in an essential
way: While in the non-interacting case g ( x, x′) is determined by a single differential equation and proper
initial conditions, there is an infinite hierarchy of equations in the presence of pair interactions, each connects
a Green function of order n to one of order n + 1. Moreover, the chronological order plays an important role
in generalizing the Green functions to finite temperatures.
From the general equation of motion idak ( t ) / dt = [ ak ( t ) , H ] we have idak ( t ) / dt = ε k ak ( t ) for the non-
interacting system, which leads to
ak ( t ) = exp ( −iε k t ) ak . (II.64)
ψ ( x1 , t ) = ∑ k ak ( t ) ei k x = ∑ k ak e − i ε k t ei k x ,
1 1
(II.65)
(
Gvac ( x1 , t ) = −i vac T ψ ( x1 , t )ψ † ( 0, 0 ) vac , )
= −i vac T ( ∑ k ,k ′
ak ak†′ e − i ε k t ei k x 1
) vac ,
= −i ∑ k e
−iεk t
ei k x 1
for t > 0; (II.66)
=0 for t < 0. (II.67)
Hence, ( )
Gvac x1 , 0+ = −i ∑ k e
i k x1
= −iδ ( x1 ) , (II.68)
i ∞ ⎡⎛ k 2 ⎞⎤
and Gvac ( x1 , t > 0 ) = − ∫−∞ dk exp ⎢i ⎜ kx1 − t ⎟⎥ ,
2π ⎣⎝ 2m ⎠ ⎦
1/ 2
⎛ 2mπ ⎞ i ( m x12 ) / ( 2 t )
= e − i 3π / 4 ⎜ ⎟ e , (II.69)
⎝ t ⎠
which is consistent with the differential equation:
⎛ ∂ 2
∂2 ⎞
⎜ i + 2 ⎟
Gvac ( x1 , t ) = δ ( x1 ) δ ( t ) . (II.70)
⎝ ∂t 2m ∂x1 ⎠
Next, we consider the one-particle Green function for the ground state of non-interacting Fermi sea
in one dimension. (N.B.! In reality, one-dimensional fermions are necessarily strongly interacting and form a
non-Fermi liquid system known as the Luttinger liquid, which will be discussed in Part VIII. For simplicity,
we shall ignore this complication for now.) In this case, the ground state is a filled Fermi sea (with occupied
momentum ranging from –kF to kF) rather than vacuum. Therefore, the one-particle Green function
G0 ( x1 , t > 0 ) deals with a fermion propagating to an unfilled state, whereas G0 ( x1 , t < 0 ) concerns with a
fermion propagating from a filled state. Thus, we have
i ⎡ − kF i k x1 − i ( k t 2 m ) −i ( )⎤,
2 ∞ k 2 t 2m
G0 ( x1 , t > 0 ) = − ∫ dk e e + ∫ dk e i k x1 e (II.71)
2π ⎣⎢ −∞ kF ⎦⎥
i ⎡ kF i k x −i ( k t 2 m) ⎤
2
G0 ( x1 , t < 0 ) = − ∫ dk e 1 e , (II.72)
2π ⎣⎢ F
− k ⎦⎥
so that
i ⎡ sin ( k F x1 ) ⎤
(
G0 x1 , 0+ = − ) π ⎢⎣
π δ ( x1 ) − ⎥, (II.73)
x1 ⎦
sin ( k F x1 )
and (
G0 x1 , 0 − = i) π x1
. (II.74)
In the case of three-dimensional fermion systems, we may Fourier transform the Green functions into
the following:
1
G0 ( k , ω ) = ∫ dt e i ω t G0 ( k , t ) , G0 ( k , t ) = G0 ( k , ω ) .
2π ∫
−i ω t
⇔ dω e (II.76)
Consequently, the 3D non-interacting one-particle Green function G0 ( k , t ) in the ground state 0 of a filled
Fermi sea is given by
G0 ( k , t ) = −i 0 T (∑ k′ ∫ )
ak ′ ak† e − i ε k′ t d 3 r e − i k i r ei k i r 0
′
where 0 ak† ak 0 ≡ nk denotes the occupation number of the state k and nk is either 0 or 1 for fermions at T
= 0. Therefore, we may rewrite EQs. (II.77) and (II.78) into the following:
where θ ( t ) is the step function and α > 0. Moreover, from EQs. (II.76), (II.79) and (II.80), we have
1
G0 ( k , ω ) = lim . (II.81)
α →0 + ω − ε k + iα sgn ( ω − ε F )
In a real Fermi gas, we expect the quasiparticle excitations to only last for a lifetime ∼ ( Γ k ) so that
−1
The results given above for non-interacting fermions in the ground state (at T = 0) can be further
generalized to non-interacting bosons and fermions at finite temperatures (T > 0) if we replace the number
operator nk by the thermal average number operator ak† ak , which, for non-interacting particles with a
chemical potential μ, is given as follows:
1
ak† ak = f ∓ ( ε k ) = , (II.86)
exp [ β ( ε k − μ )] ∓ 1
where the minus (plus) sign in the denominator refers to boson (fermion) statistics. Thus, we can express
various forms of the Green functions defined according to EQs. (II.52) -- (II.56) into the following (with
R ≡ r − r′ , τ ≡ t − t ′ ):
Following the derivation given above, the other Green functions can be derived and summarized below:
g ( k , τ ) = −ie
− iε kτ
[1 ± f ( ε )] ,
∓ k
(τ > 0) ;
− iε kτ
= ∓ ie f∓ (ε k ) , (τ < 0) ; (II.89)
g ( k , τ ) = −ie
− iε kτ
; (II.90)
g R
( k ,τ ) = −iθ (τ ) e − iε kτ
; (II.91)
g A
( k ,τ ) = iθ ( −τ ) e .
− iε kτ
(II.92)
Here the upper (lower) sign refers to the case for bosons (fermions). Hence, from EQs. (II.89) and (II.76), the
non-interacting single-particle Green function at T > 0 is given by:
⎡ f∓( ε k ) 1 − f∓( ε k )⎤
g ( k , ω ) = lim ⎢ + ⎥. (II.93)
α →0 +
⎣ ω − ε k − iα ω − ε k + iα ⎦
Interestingly, we note that for non-interacting systems, the Green functions g , g R and g A do not involve
either temperature or chemical potential μ, and are therefore identical to those associated with a single
particle moving in vacuum. In contrast, the Green functions g > , g < and g involve information pertaining not
only to the motion of the added particle (or hole) but also to the state of the system as well. In addition, for
fermions in the limit of T → 0, we have μ = εF = kF2/(2m). Therefore g ( k , τ ) at T = 0 becomes
g ( k , τ ) = −ie
− iε kτ
θ ( k − kF ) , (τ > 0) ;
= ie
− iε kτ
θ ( kF − k ) , (τ < 0) . (II.94)
On the other hand, for bosons in the T → 0 limit, the situation is more complicated because of the
phenomenon known as the Bose condensation. We shall consider the case of Bose condensation later.
Next, we take the Fourier transformation with respect to time for the Green functions given above,
and obtain the following expressions:
g ( k , ω ) = −2π i δ ( ω − ε k ) ; (II.95)
⎡θ ( k − kF ) θ ( kF − k ) ⎤
g ( k , ω ) = lim ⎢ + ⎥,
s →0 +
⎣ ω − ε k
+ is ω − ε k
− is ⎦
⎡ 1 ⎤
= lim ⎢ ⎥, (II.98)
s →0 +
⎢⎣ ω − ε k + is ( sgn ( k − k F ) ) ⎥⎦
where sgn(x) = 1 for x > 0 and sgn(x) = −1 for x < 0. For convenience, we may define
1
G (k, z ) = , (II.99)
z − εk
g R ( k , ω ) = lim G( k , ω + is ) , (II.100)
s →0 +
g A ( k , ω ) = lim G ( k , ω − is ) , (II.101)
s →0+
To find the form of the Green function in the exact ground state 0 of an interacting Fermi gas, we
need to express the particle operators in the Heisenberg representation. That is,
ak ( t ) = e i H t ak e −i H t , (II.103)
and
(
g ( k , t ) = −i 0 T ak ( t ) ak† ( 0 ) 0 )
iH t −iH t †
= −i 0 e ak e a 0 ,
k
t > 0;
= i 0 ak† e i H t ak e − i H t 0 , t < 0. (II.104)
Assuming the exact ground state energy of the N-particle system is E0N so that H 0 = E0N 0 , we obtain
g ( k , t ) = −i 0 ak e − i H t ak† 0 e
N
, t > 0;
i E0 t
N
= i 0 ak† e i H t ak 0 e
− i E0 t
, t < 0. (II.105)
If we further denote the excited states of the system as so that represents the (N+1)-particle system
for t > 0 and the (N−1)-particle system for t < 0, we may rewrite EQ. (II.105) into the following form:
∂E
μ≡ ≈ E0N +1 − E0N ≈ E0N − E0N −1 , (II.107)
∂N
and
( ) ( ) (
− E N +1 − E0N = − ⎡⎣ E N +1 − E0N +1 + E0N +1 − E0N ⎤⎦ ≈ − ( ε + μ ) , ) (II.108)
(E N −1
) ( ) (
− E0N = E N −1 − E0N −1 + E0N −1 − E0N ≈ ( ε − μ ) , ) (II.109)
we obtain
2 −i (ε + μ ) t
g ( k , t ) = −i ∑ ak† 0 e , (t > 0);
2 i (ε − μ ) t
= i∑ ak 0 e , (t < 0). (II.110)
− i ( ε + μ ) t −α t i (ε − μ ) t +α t
= lim ⎡ −i ∫ dt e i ω t ∑ ⎤
∞ 2 0 2
†
a 0 e +i ∫ dt e i ω t ∑ ak 0 e
α →0 + ⎢
⎣ ⎥⎦
0 k −∞
d ω ′ ⎡ −i ∫ dt e i ω t ∑
∞ ∞ 2 − i ( ω ′ + μ ) t −α t
ak† 0 δ ( ω ′ − ε ) e
α →0 + ∫
= lim
0 ⎣ 0
i ω ′− μ ) t + α t ⎤
+ i ∫ dt e i ω t ∑ ak 0 δ ( ω ′ − ε ) e (
0 2
−∞ ⎦
∞
⎡∑ 2
ak† 0 δ ( ω ′ − ε ) ∑ ak 0 δ ( ω ′ − ε
2
) ⎤⎥
= lim ∫ dω ′ ⎢ +
α →0 + 0 ⎢
⎣
(ω − μ ) − ω ′ + iα (ω − μ ) + ω ′ − iα ⎥
⎦
∞ ⎡ ρ + ( k, ω′) ρ − ( k, ω′) ⎤
≡ lim ∫ d ω ⎢
′ + ⎥, (II.111)
⎣ ( ω − μ ) − ω ′ + iα ( ω − μ ) + ω ′ − iα ⎦
α →0 + 0
where ρ ± ( k , ω ′ ) are known as the spectral density functions. The expression of g ( k , ω ) in terms of the
spectral density functions given in EQ. (II.111) is known as the Lehmann representation.
Using EQ. (II.9) and the Lehmann representation in EQ. (II.111), we find that the imaginary part of
g ( k , ω ) becomes
Im [ g ( k , ω )] = −π ρ + ( k , ω − μ ) if ω > μ ;
= π ρ − (k, μ − ω ) if ω < μ . (II.112)
Thus, Im [ g ( k , ω )] changes sign at ω = μ because the spectral density functions are positive definite.
Furthermore, the real part of the Green function is related to Im [ g ( k , ω )] via the following expression:
⎧ ⎡ ρ + (k, ω′)
ρ − (k, ω′) ⎤ ⎫
Re [ g ( k , ω )] = P ⎨ ∫ d ω ′ ⎢
∞
⎥⎬ +
⎣ (ω − μ ) − ω ′ (ω − μ ) + ω ′ ⎦ ⎭
0
⎩
1 ⎪⎧ ∞ ⎡ − Im [ g ( k , ω ′ + μ )] Im [ g ( k , μ − ω ′ )] ⎤ ⎪⎫
= P ⎨∫ dω′ ⎢ ⎬
(ω − μ ) + ω ′ ⎥⎦ ⎭⎪
+
π ⎩⎪0 ⎣ (ω − μ ) − ω ′
1 ⎧ ∞ Im [ g ( k , ω ′ )] ⎫
1 ⎧ μ Im [ g ( k , ω ′ )] ⎫
=− P ⎨∫ dω′ ⎬ + P ⎨ ∫ dω ′ ⎬
π ⎩ μ (ω − ω ′ ) ⎭ π ⎩ −∞ (ω − ω ′ ) ⎭
1 ⎧ ∞ μ Im [ g ( k , ω ′ )] ⎫
= P ⎨ ⎡ ∫ dω ′ − ∫ dω ′⎤ ⎬. (II.113)
π ⎩ ⎣⎢ μ −∞ ⎦⎥ ( ω ′ − ω ) ⎭
The above consideration of ground state Green functions at T = 0 can be generalized to finite
temperatures T > 0 by introducing a statistical operator:
Given the definition of a statistical operator, we can express the Green functions in their finite-
temperature forms. For instance, the quantity g > ( k , t ) :
= −i ∑ ρ m m e iH t
ak e − i H t ak† m ,
m
− i (ε −ε m ) t
= −i ∑ ρ m e
2
ak† m . (II.115)
m
∞ ∞
∫ dt e iω t g > ( k , t ) = ∫ dt e iω t g > ( k , t ) θ ( t )
0 −∞
2 −i (ε −ε ) t −α t
= lim ⎡ −i ∫ dt e i ω t ∑ m ρ m ⎤
∞
ak† m e m
α →0 + ⎢⎣ 0 ⎥⎦
1
= ∑ ρm
2
ak† m
m ω − ( ε − ε m ) + iα
1
= ∑ ρ m m ak
2
m ω − ( ε − ε m ) + iα
If the temperature is not very high, it is reasonable to assume that the matrix elements ak m are
dominated by those with N m − N = 1 . Hence, we may consolidate the expressions in EQs. (II.116) and
(II.117), yielding
⎡ β ⎡ μ + ( ε −ε m ) ⎤⎦ ⎤
1 e ⎣
g (k, ω ) = ∑ ρ
2
ak m ⎢ + ⎥
m
⎣⎢ ω + ( ε − ε m ) + iα ω + ( ε − ε m ) − iα ⎦⎥
⎧⎪ ⎡ e β ⎣⎡ μ +(ε −ε m )⎦⎤ + 1 ⎤ ⎫⎪
= ∑ρ ak m
2
⎨P ⎢
ω + ( ε − ε m ) ⎥⎦
β ⎡ μ +(ε
⎥ + iπ e ⎣ ( −ε m ) ⎦⎤
)
− 1 δ [ω + ( ε − ε m )]⎬ . (II.118)
m
⎩⎪ ⎢⎣ ⎭⎪
Thus, we find that
β ( μ −ω )
g <( k , ω ) = e g >( k , ω ) , (II.119)
so that at T = 0,
1 ⎧ ⎡ β ( ω ′ − μ ) ⎤ Im [ g ( k , ω ′ )] ⎫
Re [ g ( k , ω )] =
∞
π
P⎨ ∫
−∞
d ω ′ coth ⎢
⎣ 2 ⎥ (ω′ − ω ) ⎬ .
⎦
(II.121)
⎩ ⎭
Given the properties of g < ( k , ω ) and g > ( k , ω ) , we may define a quantity A ( k , ω ) that can be
shown to represent the generalized density of states in the (k,ω)-space:
Using EQ. (II.122), we find that g < ( k , ω ) and g > ( k , ω ) can be rewritten as:
g > ( k , ω ) = −i A ( k , ω ) [1 ± f ∓ ( ω )] , (II.123)
→ −i A ( k , ω ) θ ( ω − μ ) for fermions at T → 0 ; (II.124)
g <( k , ω ) = ∓i A ( k , ω ) f∓ (ω ) , (II.125)
→ i A ( k, ω )θ ( μ − ω ) for fermions at T →0; (II.126)
As stated previously, the situation for bosons at T = 0 is more complicated due to the phenomenon of
Bose condensation, and therefore is not explicitly expressed in EQs. (II.124) and (II.126). We further note
that A ( k , ω ) is real, which is consistent with the fact that A ( k , ω ) is the generalized density of states of the
many-body system under consideration. Furthermore, A ( k , ω ) satisfies the sum rule:
dω
∫ 2π A ( k , ω ) = 1 , (II.128)
dω dω
∫ 2π A ( k , ω ) = i ∫ 2π ∫ dt e ⎡⎣ g ( k , t ) − g ( k , t )⎤⎦ ,
i ωt > <
∞
dω ′ A ( k , ω ′)
G (k, ω ) = ⌠
⎮ . (II.129)
⌡−∞ 2π ω − ω ′
The function G ( k , ω ) is analytic in the complex ω-plane and has singularities (branch cuts, in general) along
the portions of the real axis where A ( k , ω ) ≠ 0 .
s →0 +
−1
( )
From EQs. (II.100) -- (II.102) and (II.129), also using lim ( x + is ) = P x −1 − iπδ ( x ) , we obtain:
⎧ ∞ dω ′ A ( k , ω ) ⎫
Re ⎡⎣ g R ( k , ω ) ⎤⎦ = Re ⎡⎣ g A( k , ω ) ⎤⎦ = P ⎨⌠
⎮ ⎬, (II.130)
⎩⌡−∞ 2π ω − ω ′ ⎭
1
Im ⎡⎣ g R ( k , ω ) ⎤⎦ = − Im ⎡⎣ g A( k , ω ) ⎤⎦ = − A (k,ω ) . (II.131)
2
⎧ dω ′ A ( k , ω ) ⎫
∞
Re ⎣⎡ g ( k , ω ) ⎦⎤ = Re ⎣⎡ g R ( k , ω ) ⎦⎤ = Re ⎣⎡ g A( k , ω ) ⎦⎤ = P ⎨⌠
⎮ ⎬, (II.132)
⎩⌡−∞ 2π ω − ω ′ ⎭
1 ⎡ β (ω − μ ) ⎤
Im ⎣⎡ g ( k , ω ) ⎦⎤ = − A ( k , ω ) ⎡⎣1 ± 2 f ∓ ( ω ) ⎤⎦ = Im ⎡⎣ g R ( k , ω ) ⎤⎦ coth ⎢ ⎥ for bosons; (II.133)
2 ⎣ 2 ⎦
⎡ β (ω − μ ) ⎤
= Im ⎡⎣ g R ( k , ω ) ⎤⎦ tanh ⎢ ⎥ for fermions. (II.134)
⎣ 2 ⎦
From EQ. (II.134), it follows that a fermion system in the limit of T → 0 satisfies the relation
Im ⎡⎣ g ( k , ω ) ⎤⎦ = sgn ( ω − μ ) Im ⎡⎣ g R ( k , ω ) ⎤⎦ ,
Next, we investigate the relation of Green functions to various important physical properties in
condensed matter systems besides the density of states. For instance, consider the second quantization
formalism for a first-quantization one-particle operator F ( r ) :
F = ∫ d 3 r ψ † ( r ) F ( r )ψ ( r ) . (II.136)
Examples of operators F include the kinetic energy K with K ( r ) = −∇ 2 /(2m) , the total number of particles
N with N ( r ) = 1 , and the density operator n ( r ) = δ ( r − r0 ) . It follows from the definition of g < ( x, x′ ) that
the thermal average of F is given by:
More explicitly, using EQ. (II.137) we can obtain the thermal average of the kinetic energy:
⎡ ∇r2 ⎤
K = ± i lim ∫ d 3 r lim ⎢ − g ( r , t ; r ′, t ′ ) ⎥
r ′→ r t ′→t +
⎣ 2m ⎦
∞ ∞
dω k 2
dω k2
= ±i∑ ∫ g < (k, ω ) =∑∫ A ( k , ω ) f∓ (ω ) . (II.138)
k −∞
2π 2m k −∞
2π 2m
We can also obtain the Hamiltonian for an interacting many-body system. Recall the pair interaction
potential given in EQ. (II.60), which is reproduced below:
⎛ ∂ ∇2 ⎞
⎟ψ ( r, t ) = ⎡⎣ ∫ d r1 ψ ( r1 , t )V ( r − r1 )ψ ( r1 , t ) ⎤⎦ψ ( r, t ) ,
3 †
⎜i + (II.60)
⎝ ∂t 2m ⎠
1 ⎡⎛ ∂ ∂ ⎞ ⎤ 1
∫
3
d r ⎢⎜ i − i ⎟ψ ( r, t ′ )ψ ( r, t ) ⎥ =
†
K + V . (II.142)
4 ⎣⎝ ∂t ∂t ′ ⎠ ⎦ t ′=t 2
From EQs. (II.138) and (II.60), the thermal average of the interaction potential becomes
i ⎡⎛ ∂ ∇r 2 ⎞ ⎤ dω 1 ⎛ k2 ⎞
V =± ∫ d r ⎢⎜ i
3
+ ⎟ g ( x, x′ ) ⎥ = ∑∫ ⎜ ω − ⎟ A ( k, ω ) f∓ (ω ) . (II.143)
2 ⎣⎝ ∂t 2m ⎠ ⎦ r′→r , t ′→t + k 2π 2 ⎝ 2m ⎠
i ⎡⎛ ∂ ∇ r 2 ⎞ ⎤ dω 1 ⎛ k2 ⎞
H =±
2
∫ d r ⎢⎜⎝ i ∂t 2m ⎟⎠ g ( x, x )⎥
3
− ′ = ∑ ∫ 2π 2 ⎜ ω 2m ⎟ A ( k, ω ) f∓ (ω ) .
+ (II.144)
⎣ ⎦ r′→r , t′→t + k ⎝ ⎠
To obtain all other thermodynamic quantities, it is sufficient to calculate the grand canonical
partition function ZG = Tr{exp[−β (H−μN)]} as function of the volume Ω, the chemical potential μ, and the
temperature T = β −1. The grand canonical partition function ZG is directly related to the pressure P by the
thermodynamic equation
Z G = exp ( β Ω P ) , (II.145)
and the pressure can be expressed in terms of the density through the relation
∞
d 3k dω
d μ ′ [ n ( β , μ ′ )] = ∫ d μ ′∫
μ μ
P(β, μ) = ∫ ∫ 2π A ( k , ω ) f (ω ) , (II.146)
(2π )
−∞ −∞ 3 ∓
−∞
where we have inserted the expression given in EQ. (II.139) for the density, and we note that both A ( k , ω )
and f (ω) depend on the chemical potential μ, and f (ω) depends on the temperature.
However, in general the dependence on μ may not be easily attainable. Hence, it is more convenient
to take an alternative approach to calculate ZG: Using the general thermodynamic relation
∂H ∂P
= −Ω , (II.147)
∂α ∂α T ,μ
H = K +αV , (II.148)
with α = 0 representing the non-interacting system and α = 1 representing the actual interacting system, we
obtain from EQs. (II.147) and (II.148)
1 dω 1 ⎛ k2 ⎞
ln ( Z G ) = β P Ω = β P0 Ω − β ∫ dα ∑ ∫ ⎜ ω − ⎟ Aα ( k , ω ) f ∓ ( ω ) , (II.150)
0
k 2π 2 ⎝ 2m ⎠
where the subscript α in A ( k , ω ) denotes that Aα ( k , ω ) corresponds to the Hamiltonian in EQ. (II.148),
and P0 is the pressure for the non-interacting system.
As seen in EQs. (II.138), (II.139), (II.143), (II.144), (II.146) and (II.150), various thermodynamic
quantities that we have discussed so far all involve an integral of the following type:
dω
I = lim I (σ ) = lim ∫ 2π e
−ωσ
F ( ω ) A( k , ω ) f ∓ ( ω ) , (II.151)
σ →0 − σ →0 −
where F(ω) is a polynomial, and the term exp(−ωσ) is introduced to assist the computation of I in the
complex ω-plane. Noting that A ( k , ω ) = i g ( k , ω ) = i [ G ( k , ω + is )− G ( k , ω − is )] , we can express the
integration over ω in EQ. (II.151) by the sum of two integrands, one along the upper ω-plane and the other
along the lower ω-plane:
dω dω
I (σ ) = i ∫ e−ωσ F (ω ) G ( k , ω ) f ∓ (ω ) − i ∫ e−ωσ F ( ω ) G ( k , ω ) f ∓ ( ω ) , (II.152)
CR
2π CA
2π
where the paths CR and CA refer to the integration over the upper and lower half-planes, respectively, and the
subscripts R and A are consistent with the those used in the definitions of Green functions, denoting
respectively the “retarded” and “advanced” chronological orders. The paths CR and CA are chosen to avoid
the poles of f ∓ ( ω ) , which are imposed by EQs. (II.44) and (II.45) as discussed earlier. More specifically, the
poles zν are slightly different between the cases for bosons and fermions, and are given explicitly below:
iπν
zν = μ + ; ν = 2n for bosons,
β
ν = 2n − 1 for fermions, (II.153)
and n is positive for the path CR and negative for the path CA. Thus, for –β < σ < 0, EQ. (II.152) becomes
⎡ 1 ⎤ ⎡ 1 ⎤
I ( σ ) = I R (σ ) − I A ( σ ) = ⎢ ∓ ∑e − zν σ
F ( zν ) G ( k , zν ) ⎥ − ⎢ ± ∑ e F ( zν ) G ( k , zν ) ⎥
−z σ ν
⎣ β ν >0 ⎦ ⎣ β ν <0 ⎦
1
=∓
β
∑e − zν σ
F ( zν ) G ( k , zν ) ,
ν
1
and therefore I =∓
β
∑ F ( z ) G (k, z ) ,
ν ν (II.154)
ν
which suggests that the thermodynamic quantities of a many-body system can be obtained from the values of
G ( k , ω ) at the special points zν specified in EQ. (II.153). This result is convenient because generally
G ( k , ω ) is easier to calculate than A ( k , ω ) .
k2
A ( k , ω ) / 2π = ρ ( k , ω ) = δ ω − ε ( 0
k ), ε ≡ 0
k
2m
. (II.155)
Hence, we may interpret A ( k , ω ) / 2π as the generalized density of states per unit frequency and per unit
volume in k-space. In addition, from EQ. (II.125) g < ( k , ω ) can be interpreted (apart from a factor of ± i) as
the thermally averaged number of particles per unit volume in the ( k , ω ) -space. Similarly, from EQ.
(II.123) g ( k , ω ) can be interpreted (apart from a factor of ± i) as the thermally averaged number of states
>
(per unit volume in the ( k , ω ) -space) available for the addition of an extra particle to the system.
Having seen the importance of Green functions to the descriptions of various physical properties of
many-body systems at all temperatures, we briefly discuss the equation of state method used to calculate the
Green functions of an interacting many-body system, followed by more elaborate studies of the perturbative
diagrammatic method.
As mentioned earlier, the Green functions for an interacting many-body system obey a hierarchy of
equations, and the first of which in EQ. (II.61) is reproduced below:
⎛ ∂ ∇r2 ⎞
⎟ g ( x, x′ ) = δ ( x − x′ ) ± i ∫ d x1 V ( r − r1 ) g 2( x, x1 ; x′, x1 ) t = t , (II.61)
4 +
⎜i +
⎝ ∂t 2m ⎠ 1
In order to obtain an explicit solution, the hierarchy has to be terminated at some point by employing
approximate relation that connects gn with gn−1, gn−2, etc. The simplest approximation is to express g2 in terms
of g, which is equivalent to the Hartree approximation:
Equation (II.156) implies that the added two particles propagate independently, one from x1 to x1´ and the
other from x2 to x2´. The Hartree approximation does not satisfy the basic symmetry property that requires
(g2)2 to be invariant under the exchange of x1 ↔ x2 or x1´ ↔ x2´, and is therefore oversimplified.
Substituting EQ. (II.156) into EQ. (II.61) and using EQ. (II.139), we obtain
⎛ ∂ ∇r2 ⎞
⎟ g ( x, x′ ) = δ ( x − x′ ) ± ig ( x, x′ ) ∫ d r1 V ( r − r1 ) g ( x1 , x1 ) ,
3 +
⎜i +
⎝ ∂t 2m ⎠
= δ ( x − x′ ) ± ig ( x, x′ ) ∫ d 3 r1 V ( r − r1 ) n( r1 ) ≡ δ ( x − x′ ) + ig ( x, x′ ) Veff ( r ) , (II.157)
where we have introduced an effective one-body potential Veff ( r ) . Thus, we can rewrite EQ. (II.157) into the
following simple form:
⎛ ∂ ∇r2 ⎞
⎜i + − Veff (r ) ⎟ g ( x, x′ ) = δ ( x − x′ ) , (II.158)
⎝ ∂t 2m ⎠
which implies that the added particle (or hole) moves independently in the averaged potential Veff ( r ) created
by all particles in the many-body system. For a translational invariant system, the density n( r1 ) is a
constant n0. As a result, Veff ( r ) is a constant given by n0V0, where V0 = i ∫ d r V ( r ) . Consequently, EQ.
3
(II.158) becomes a simple differential equation, and its Fourier transformation yields:
⎛ k2 ⎞
⎜ ω − − n0V0 ⎟ g (k, ω ) = 1 . (II.159)
⎝ 2m ⎠
In reference to EQ. (II.135), we obtain the Green function solution for fermions at T = 0:
⎡ 1 ⎤
g ( k , ω ) = lim ⎢ ⎥. (II.160)
( ) 00 (
s → 0 + ω − k 2 / 2m − n V + is sgn ω − μ
⎣⎢ ) ⎦⎥
To obtain the Green function for T ≠ 0, we note that g (k , ω ) is not the limit of an analytic function,
so that we cannot arbitrarily continue EQ. (II.160) in the complex ω-plane without further analysis.
However, we can still use the “imaginary time” trick to obtain the Green function for T ≠ 0. The procedure is
as follows. Consider the casual Green function g (r1 , t1 ; r2 , t2 ) . If we substitute t1 by –iσ1 and t2 by –iσ2, the
time ordering by the σ-ordering, and confine σ1 and σ2 within the interval [0, β ], we have:
dω d 3k −iω ( t1 −t2 ) i k ⋅r
∞
g < ( r1 , t1; r2 , t2 ) = ∓i ∫ e A ( k , ω ) f ∓ (ω ) ,
−∞ 2π ∫ (2π )3
e (II.162)
dω d 3k −ωσ i k ⋅r
∞
g < ( r1 , −iσ1; r2 , −iσ 2 ) = ∓i ∫ e e A ( k , ω ) f ∓ (ω ) .
−∞ 2π ∫ (2π )3
(II.163)
d 3k i k ⋅r ⎡ ∓1 ⎤
g ( r, −iσ ) = ∓i ∫ e ⎢ ∑ e − zν σ g ( k , zν ) ⎥ , (II.164)
(2π ) 3
⎣β ν ⎦
⎡i ⎤
and g ( k , −iσ ) = ⎢
⎣β
∑
ν
e − zν σ
g ( k , zν ) ⎥ ,
⎦
(II.165)
β
g ( k , zν ) = −
i
dσ g ( k , −iσ ) e
2∫ β
zν σ
, (II.166)
−
Substituting EQ. (II.165) into EQ. (II.158) using the imaginary times, we obtain
⎛ k2 ⎞ 1
⎜ ν
z − − n0V0 ⎟ g ( k , zν ) = 1 → g ( k , zν ) = . (II.167)
⎝ 2m ⎠ zν − ( k / 2m ) − n0V0
2
The solution in EQ. (II.167) gives rise to EQ. (II.160) through analytic continuation.
In the Hartree approximation, we find that the average energy of the system, using EQs. (II.144) and
(II.167), is given by
1
H = ∑ ε k f ∓ ( ε k ) − Ω n02 V0 , (II.168)
k 2
which is equivalent to the free particle energy with an added energy n0V0 per particle, as expected.
An improvement over the Hartree approximation can be made by taking into account the symmetry
or anti-symmetry of g2 under the exchange of x1 ↔ x2 or x1´ ↔ x2´ while still keeping the added particles as
moving independently of each other. This improved approximation is known as the Hartree-Fock
approximation:
By following the same procedure outlined above for the Hartree approximation, we obtain within the
framework of the Hartree-Fock approximation the Green function:
1
g (k,ω ) = , (II.170)
ω − εk
where
k2 d 3k ′
εk = + n0V0 ± ∫ V ( k − k′) n ( k′) , (II.171)
2m (2π )3
and V ( k ) = ∫ d 3 r exp ( −i k ⋅ r ) V ( r ) is the Fourier transform of the potential V ( r ) . We note that the quasi-
particle energy εk in the Hartree-Fock approximation depends implicitly on the temperature through its
dependence on the thermal average of the density n ( k ′ ) .
Next, we want to evaluate the Green functions of interacting many-body systems using diagrammatic
analysis of perturbation theory. It is generally convenient to apply Wick’s theorem to evaluate the Green
functions in the perturbation expansion. In the following we revisit Wick’s theorem by introducing the
definitions of chronological ordering, normal ordering, and contractions.
Let’s assume that in the interaction picture the total Hamiltonian H can be expressed by H = H0 +
HI, where H0 denotes the unperturbed Hamiltonian and HI the interaction perturbation. Moreover, we
denote the exact interacting ground state 0 by Ψ 0 and the eigenstate of H0 by Φ 0 . We shall first prove
that the matrix element of a Heisenberg operator OH evaluated in the Heisenberg picture can be converted
into the interaction picture according to the following relation:
Ψ 0 OH ( t ) Ψ 0
ν
1 ∞
⎛ −i ⎞ 1 ∞
∑ ⎜⎝ ⎟⎠ ν ! ∫−∞ dt1 … ∫−∞ dtν
∞
= Φ0
Ψ0 Ψ0 Φ 0 U ε ( ∞, −∞ ) Φ 0 ν =0
− ε ( t1 +…+ tν )
×e T [ H I ( t1 ) H I ( tν ) OI ( t )] Φ 0 . (II.172)
Equation (II.172) can be further extended to converting the matrix element of two Heisenberg operators into
the interaction picture as follows:
Ψ 0 OH ( t ) OH ( t ′ ) Ψ 0
ν
1 ∞
⎛ −i ⎞ 1 ∞
∑ ⎜⎝ ⎟⎠ ν ! ∫−∞ dt1 … ∫−∞ dtν
∞
= Φ0
Ψ0 Ψ0 Φ 0 U ε ( ∞, −∞ ) Φ 0 ν =0
− ε ( t1 +…+ tν )
×e T [ H I ( t1 ) H I ( tν ) OI ( t ) OI ( t ′ )] Φ 0 . (II.173)
It is clear from EQs. (II.172) and (II.173) that the evaluation of Green functions in the interaction picture
requires performing the time ordering of products of field operators where Wick’s theorem becomes helpful.
To prove EQ. (II.172), recall the Gell-Mann and Low theorem in Part I that expresses the exact
ground state of an interacting system in the interaction picture:
Ψ0 U ε ( 0, ±∞ ) Φ 0
= . (II.174)
Φ0 Ψ0 Φ 0 U ε ( 0, ±∞ ) Φ 0
Φ 0 U ε ( 0, ∞ ) U ε ( 0, −∞ ) Φ 0 Φ 0 U ε ( ∞, −∞ ) Φ 0
†
Ψ0 Ψ0
= = ,
Φ 0 U ε ( 0, ∞ ) Φ 0 Φ 0 U ε ( 0, −∞ ) Φ 0 Φ 0 U ε ( 0, ∞ ) Φ 0 Φ 0 U ε ( 0, −∞ ) Φ 0
2 * *
Φ0 Ψ 0
1 Φ 0 U ε ( ∞, 0 ) Φ 0 Φ 0 U ε ( 0, −∞ ) Φ 0 1
⇒ = . (II.175)
Φ 0 U ε ( ∞, −∞ ) Φ 0
2
Ψ0 Ψ0 Φ0 Ψ 0
Similarly, the numerator on the left side of EQ. (II.172) can be rewritten as:
Φ 0 U ε ( ∞, 0 ) Φ 0 Φ 0 U ε ( 0, −∞ ) Φ 0
2
Φ0 Ψ0
= Φ 0 U ε ( ∞, 0 ) U ε ( 0, t ) OI ( t ) U ε ( t , 0 ) U ε ( 0, −∞ ) Φ 0
Φ 0 U ε ( ∞, 0 ) Φ 0 Φ 0 U ε ( 0, −∞ ) Φ 0
2
Φ0 Ψ 0
= Φ 0 U ε ( ∞, t ) OI ( t ) U ε ( t , −∞ ) Φ 0 . (II.176)
Φ 0 U ε ( ∞, 0 ) Φ 0 Φ 0 U ε ( 0, −∞ ) Φ 0
Ψ 0 OH ( t ) Ψ 0 Φ 0 U ε ( ∞, t ) OI ( t ) U ε ( t , −∞ ) Φ 0
= . (II.177)
Ψ0 Ψ0 Φ 0 U ε ( ∞, −∞ ) Φ 0
Next, we want to express the product of operators in the numerator on the right side of EQ. (II.177)
explicitly. Following the discussion in Part I for the time-evolution U-operators, we obtain
⎛ −i ⎞ 1 ∞
∞ n
− ε ( t +…+ tn )
U ε ( ∞, t ) OI ( t ) U ε ( t , −∞ ) = ∑ ⎜ T [ H I ( t1 ) H I ( tn ) ]
∞
n =0 ⎝
⎟
⎠ n!
∫t
dt1 … ∫ dtn e 1
t
⎛ −i ⎞ 1 t
∞ m
− ε ( t +…+ tm )
× OI ( t ) ∑ ⎜ ∫ dt1 … ∫ dtm e 1 T [ H I ( t1 ) H I ( tm )] .
t
⎟ (II.178)
m=0 ⎝ ⎠ m! −∞ −∞
On the other hand, we note that the numerator of the right side of EQ. (II.172) can be rewritten if we have ν
= n + m and if for the ν-th term in the sum we divide the integration variables into n factors with ti > t and m
factors with ti < t. Clearly there are a total of ν!/(n!m!) different ways of such divisions for each ν. Thus, the
product of operators on the right side of EQ. (II.172) becomes:
ν
∞
⎛ −i ⎞ 1 ∞ ∞ ν! ∞ − ε ( t1 + + tn )
∑ ⎜⎝ ⎟⎠ ν ! ∑∑ δν ,m+ n n !m ! ∫t dt1 T [ H I ( t1 ) H I ( tn )]
∞
ν =0 n =0 m =0
∫
t
dtn e
− ε ( t1 +…+ tm )
× OI ( t ) ∫ dt1 … ∫ dtm e T [ H I ( t1 ) H I ( tm )] ,
t t
(II.179)
−∞ −∞
which is essentially identical to EQ. (II.178) after we apply the Kronecker delta function. Thus, we have
proven EQ. (II.172). Similar procedure can be applied to prove EQ. (II.173), except that we need to partition
the integration variables into three distinct groups.
Now we are ready to evaluate Green functions in the interaction picture with the aid of EQ. (II.173).
That is, by identifying the operators with the fermion field operators in the interaction picture so that
OI ( t ) = ψ α ( x ) and OI ( t ′ ) = ψ β† ( x′ ) where α and β refer to the spin indices, we obtain the Green function in
the interaction picture as follows:
∞
⎛
ν
−i ⎞ 1 ∞ Φ 0 T ⎡⎣ H I ( t1 ) H I ( tν )ψ α ( x )ψ β† ( x′ ) ⎤⎦ Φ 0
igαβ ( x, x′ ) = ∑ ⎜ ⎟
∞
ν =0 ⎝
∫ dt1 … ∫−∞ dtν
⎠ ν ! −∞ Φ 0 U ε ( ∞, −∞ ) Φ 0
. (II.180)
U ( x1 , x2 ) = V ( r1 , r2 ) δ ( t1 − t2 ) , (II.181)
⎛ −i ⎞ 1 ∞ 4
i gαβ ( x, x′ ) = iGαβ ( x, x′ ) + ⎜ ⎟ ∑ ∫−∞ d x1 d x1′ U ( x1 , x1′ )λλ ′, μμ ′
0 4
⎝ ⎠ λλ ′, μμ ′ 2
× Φ 0 T ⎡⎣ψ λ† ( x1 )ψ μ† ( x1′ )ψ μ ′ ( x1′ )ψ λ ′ ( x1 )ψ α ( x )ψ β† ( x′ ) ⎤⎦ Φ 0 . (II.182)
It is clear from EQ. (II.182) that the evaluation of requires finding the expectation values in the non-
interacting ground state of time-ordered products of creation and annihilation operators of the following
form:
Φ 0 T ⎡⎣ψ † ψ ψ α ( x )ψ β† ( x′ ) ⎤⎦ Φ 0 . (II.183)
We’ll rely on Wick’s theorem to provide a general procedure for evaluating matrix elements of the form
given in EQ. (II.183).
Before stating Wick’s theorem in the context of time-ordered operators, we summarize below three
different definitions for products of field operators ( ABCD ) .
where p is the number of permutations of fermion operators needed to rearrange the product given on the left
side of EQ. (II.184) to agree with the order on the right side.
so that the field operators within a normal-ordered product either commute (for bosons) or anti-commute (for
fermions). We remark that the normal-ordering notation N ( ABCD ) in EQ. (II.185) is equivalent to the
notation : ABCD : given in some textbooks of relativistic quantum field theory.
The introduction of normal-ordered products of field operators is motivated by the fact that their
expectation values in the unperturbed ground state vanish identically. This result remains true even if the
product consists of all creation operators, because the expectation value in the unperturbed ground state of its
adjoint product vanishes identically. It is therefore convenient to reduce the time-ordering products into
combinations of normal-ordering products plus extra terms due to commutation or anti-commutation of
bosonic or fermionic operators. For this purpose, it is also convenient to decompose the field operator into a
destruction part ψ ( a ) that annihilates the non-interacting ground state and a creation part ψ ( c ) . Hence,
As an explicit example, we may consider the decomposition of the free fermion field of a many-body system:
⎡ i ( k i r −ωk t ) i ( k i r −ωk t ) ⎤
ψ ( x ) = Ω −1/ 2 ⎢ ∑ e ηλ akλ + ∑ e ηλ b−†kλ ⎥ ≡ ψ ( a ) ( x ) + ψ ( c ) ( x ) , (II.188)
⎣ k > kF ,λ k < k F ,λ ⎦
where akλ and b−†kλ are respectively the particle and hole operators for |k| > kF and |k| < kF, ηλ is the spinor
operator, and Ω is the volume of the system.
〈3〉 Contraction:
The contraction of two operators A and B is denoted by A•B• and is defined as follows:
A i B i = T ( AB ) − N ( AB ) . (II.189)
Given the definition in EQ. (II.189), it can be easily verified that all the following contractions of operators
vanish:
ψ ( a ) iψ ( c ) i = ψ ( a ) † iψ ( c ) † i = ψ ( a ) † iψ ( c ) i = ψ ( a ) iψ ( c ) † i = 0 . (II.190)
Therefore, most contractions are zero. In particular, the contraction of two creation parts or two annihilation
parts always vanish. The only non-zero contractions are given by the following:
=0 for t x < t y ;
ψ ( c ) ( x ) ψ ( c ) †( y ) i = 0
i
for t x > t y ,
= iG (0)
( x, y ) for t x < t y . (II.191)
We note that the Green function in EQ. (II.191) for free fermions is given by (with spin indices restored)
(0)
iGαβ ( x, y ) = δαβ Ω−1 ∑ k ei k i( x−y ) ⎡⎣θ ( t x − t y ) θ ( k − kF ) − θ ( t y − t x ) θ ( kF − k ) ⎤⎦ . (II.192)
Moreover, the contractions are c numbers in the occupation–number Hilbert space rather than operators. In
the case of a product of operators consisting of more than one pair of contraction, we denote the first pair by
a pair of single dots, the second pair by a pair of double dots, etc.
Having introduced the definitions of time-ordering, normal-ordering and contraction, we are ready to
prove the following basic lemma for Wick’s theorem.
〈4〉 Lemma:
For a normal-ordered product N ( AB XY ) , if Z is an operator labeled with time earlier than the
times of all operators in the normal-ordered product, then the following relation is satisfied:
N ( AB XY ) Z = N AB ( )
XY i Z i + N AB ( X iYZ i + )
(
+N AiB )
XYZ i + N ( AB XYZ ) . (II.193)
[Proof] We first note that for the special case of Z being an annihilation operator, we have T (UZ ) = N (UZ )
for any operator U in the normal-ordered product so that all contractions vanish and EQ. (II.193) is proven.
We may also assume that the operator product AB XY is already normal-ordered for simplicity. Otherwise
we can always reorder the operators on both sides of EQ. (II.193) so that the same signature factor occurs in
each term of EQ. (II.193) and therefore cancels identically. To further simplify the problem in the proof, we
may assume next that Z is a creation operator and AB XY are all annihilation operators, because if we can
prove this special case for EQ. (II.193), for situations with creation operators within the normal-ordered
product, we may simply multiply the creation operators to the left side of AB XY , and the resulting
additional contractions with Z are identically zero because the contractions of two creation operators are
always zero and EQ. (II.193) remains valid.
Now we may prove EQ. (II.193) by induction for the case of Z being a creation operator and
AB XY being all annihilation operators. First, we note that EQ. (II.193) apparently holds for two operators
because by definition,
AZ = T ( AZ ) = A i Z i + N ( AZ ) . (II.194)
Next, if we assume that EQ. (II.193) holds for n operators, we want to prove that it is also valid for (n+1)
operators. If we multiply to the left of the product of operators in EQ. (II.193) another annihilation operator
D that has a time later than that of Z, we obtain
DN ( AB XY ) Z = N ( DAB XY ) Z = N DAB ( )
XY i Z i + N DAB ( X iYZ i + )
(
+ N DA i B )
XYZ i + DN ( AB XYZ ) . (II.195)
(
= N D i AB )
XYZ i + N ( DAB XYZ ) . (II.196)
By inserting EQ. (II.196) into EQ. (II.195) we therefore prove EQ. (II.193). Now we are ready to state and
prove Wick’s theorem for product of field operators.
[Proof] We shall prove this theorem by induction. First, EQ. (II.197) apparently holds for two operators by
definition. That is,
( )
T ( AB ) = N ( AB ) + N A i B i = N ( AB ) + A i B i . (II.198)
Next, we assume that EQ. (II.197) holds for n operators and we multiply on the right of EQ. (II.197) by an
operator Ω with a time earlier than all other operators ABC XYZ . Thus, we find that
where in the second line of EQ. (II.199) we have used the lemma given by EQ. (II.193). Thus, we have
proven Wick’s theorem in EQ. (II.197) by induction.
Wick’s theorem is useful for evaluating ground-state averaged quantities of the form Φ 0 Φ0 ,
where all uncontracted normal-ordered products of operators vanish. Therefore, the Green function given in
EQ. (II.180) consists of all possible fully contracted terms of the operators in the interaction picture.
1
d r ∫ d r ′ψ α ( r )ψ β ( r′ ) V ( r − r′ )αα ′, ββ ′ ψ β ( r′ )ψ α ( r ) ,
2∫
† †
HI = 3 3
(II.200)
where we have generalized our consideration for fermion systems by restoring the spin indices α and β to the
field operators, interaction potential, and the Green functions. For simplicity, we have assumed in EQ.
(II.200) that the interaction potential U ( x, x′ )αα ′ , ββ ′ = V ( r − r ′ )αα ′ , ββ ′ δ ( t − t ′ ) only involves instant-time
interaction. The Green function is now defined as
Ψ 0 T ⎡⎣ψ H α ( x )ψ H† β ( y ) ⎤⎦ Ψ 0
igαβ ( x, y ) = , (II.201)
Ψ0 Ψ0
where α and β denote spin indices and all other notations remain the same as before.
While in general it is more convenient to compute Green functions in momentum and frequency
space, as we shall focus on later, it is instructive to consider Green functions in spacetime first to illustrate
the utility of Wick’s theorem. As an example, let us consider the first-order contributions to in EQ. (II.182)
for a pair interaction potential given by U ( x, x′ )λλ ′ , μμ ′ . In this case, we have three pairs of creation and
annihilation operators so that there are 6 fully contracted and non-vanishing products of field operators
between ψ and ψ†. Hence, to the first order EQ. (II.182) becomes:
⎛ −i ⎞ 1
i gαβ ( x, y ) = ⎜ ⎟ ∑ ∫ d x1 d x1′ U ( x1 , x1′ )λλ ′, μμ ′
4 4
⎝ ⎠ 2 λλ ′μμ ′
{
× iGαβ0 ( x, y ) ⎡⎣iGμ0′μ ( x1′, x1′ ) iGλ0′λ ( x1 , x1 ) − iGμ0′λ ( x1′, x1 ) iGλ0′μ ( x1 , x1′ ) ⎤⎦
(A) (B)
The corresponding Feynman diagrams are illustrated in Fig. II.10.1, where each interaction potential is
explicitly given by a wavy line connecting with two solid lines (i.e. two propagators G0’s). It is clear from
Fig. II.10.1 that graphs C and E are equivalent and similarly D and F are equivalent. Moreover, graphs A and
B are disconnected graphs, and are therefore not contributing to the total Green function in EQ. (II.180)
because they are identically cancelled by contributions from the denominator Φ 0 U ε ( ∞, −∞ ) Φ 0 . This fact
is consistent with our earlier discussion of the generating functional in relativistic quantum field theory,
where we have shown using the path integral formalism that the only relevant Feynman diagrams
contributing to the propagators are connected diagrams. We demonstrate in the following that the same
concept applies to the canonical formalism in the interaction picture.
The contribution of a disconnected graph to the Green function factors because it closes on itself.
Therefore, we may express the νth-order term of the numerator in EQ. (II.182) as a product of connected and
disconnected parts:
n+m
∞ ∞
⎛ −i ⎞ 1 ν!
( x, y ) = ∑∑ ⎜
∞ ∞
dtm Φ 0 T ⎡⎣ H I ( t1 ) H I ( tm )ψ α ( x )ψ β† ( y ) ⎤⎦ Φ 0
ν ! n !m ! ∫ ∫
(ν )
i gαβ ⎟ δν , m + n dt1 connected
n =0 m =0 ⎝ ⎠ −∞ −∞
dtν Φ 0 T [ H I ( tm +1 ) H I ( tν )] Φ 0 .
∞ ∞
× ∫ dtm +1 ∫ (II.203)
−∞ −∞
Summing over all νth-order terms, we find that EQ. (II.182) becomes
⎛ −i ⎞ 1 ∞
∞ m
i gαβ ( x, y ) = ∑ ⎜
∞
⎟ ∫ dt1 ∫ dtm Φ 0 T ⎡⎣ H I ( t1 ) H I ( tm )ψ α ( x )ψ β† ( y ) ⎤⎦ Φ 0 connected
m =0 ⎝ ⎠ m ! −∞ −∞
⎛ −i ⎞ 1 ∞
∞ n
×∑⎜ ⎟ dtn Φ 0 T [ H I ( t1 ) H I ( tn )] Φ 0 .
∞
n=0 ⎝
∫ dt1
⎠ n ! −∞
∫−∞
⎛ −i ⎞ 1 ∞
∞ m
=∑⎜
∞
⎟ ∫ dt1 ∫ dtm Φ 0 T ⎡⎣ H I ( t1 ) H I ( tm )ψ α ( x )ψ β† ( y ) ⎤⎦ Φ 0 connected
m=0 ⎝ ⎠ m ! −∞ −∞
× Φ 0 U ε ( ∞, −∞ ) Φ 0 , (II.204)
⎛ −i ⎞ 1 ∞
∞ m
igαβ ( x, y ) = ∑ ⎜ ⎟
∞
In this context, graphs A and B in Fig. II.10.1 do not contribute to first-order Green function.
It is also instructive to comment on a special case of a solid line closed on itself, which corresponds
to a Green function evaluated at the same time. While the time ordering of field operators is not well defined
in the equal Green function, the single-particle Green function gαβ ( x, x ) arises from a contraction of two
field operators within the interaction Hamiltonian HI, and the field operators appear in the form of
ψ β† ( x )ψ α ( x ) . Therefore, the Green function at equal times must be interpreted as:
0
iGαβ ( x, x ) = tlim
′→t +
Φ 0 T ⎡⎣ψ α ( r , t )ψ β† ( r , t ′ ) ⎤⎦ Φ 0
= − Φ 0 ψ β† ( r )ψ α ( r ) Φ 0
= − ( 2 s + 1) δ αβ n 0( r )
−1
(II.206)
for a system of spin-s fermions, where n0(r) denotes the particle density in the unperturbed ground state,
which need not be the same as the particle density n(r) in the interacting system. Thus, the terms denoted by
graphs D and F represent the lowest-order direct interaction with all the particles that constitute the non-
interacting ground state. On the other hand, the terms C and E represent the lowest-order exchange
interaction.
Based on the above discussions, we summarize the following Feynman rules in spacetime for the
nth-order contribution to the single-particle Green function gαβ ( x, y ) :
1) Draw all topologically distinct connected diagrams with n interaction lines U and (2n+1) direct Green
functions G0.
-- There are 2n-pairs of internal Green functions for n interaction lines and one external pair of Green
function. For the internal Green functions denoted by the set of variables ( x1 , x1′ ) ( xn , xn′ ) , there are n!
different possibilities of choosing the variables for each of these diagrams, and Wick’s theorem is used to
verify the enumeration.
( )
U ( x, y )λλ ′ , μμ ′ = V ( r − r ′ )λλ ′ , μμ ′ δ t x − t y ,
5) Integrate all internal variables ( x1 , x1′ ) ( xn , xn′ ) over space and time.
6) Associate a spin matrix product with each continuous fermion line as well as with the potentials at each
vertex.
⎣ ⎦⎣ ⎦ ⎣ ⎦
without changing sign and then followed by moving the last field operator over to the far left via an odd
number of interchanges, so that there is a net sign change.
-- The nth-order term contains an explicit numerical factor ( −i / )n . In addition, the (2n+1) contractions
of field operators contribute an additional factor ( i ) , so that the computation of g ( x, y ) involves a net
2 n +1
0
9) A Green function with equal time variables must be interpreted as Gαβ r, t ; r, t + . ( )
As an example, we apply the above Feynman rules to express the first-order contribution to G ( x, y ) .
There are two internal variables and two internal Green functions, so that we have
( x, y ) = ⎛⎜ ⎞⎟ ∫ d 4 x1 d 4 x1′ {( −1) Gαλ(0)( x, x1 ) U( x1 , x1′ )λλ ′,μμ ′ Gμ(0)′μ ( x1′, x1′ ) Gλ(0)′β ( x1 , y )
(1) i
g αβ
⎝ ⎠
(0)
+ Gαλ }
( x, x1 ) U ( x1 , x1′ )λλ ′,μμ ′ Gλ(0)′μ ( x1 , x1′ ) Gμ(0)′β ( x1′, y ) . (II.207)
λ μ
x y
λ′ μ′
Figure II.10.2 Illustration of the matrix indices for the interaction potential U ( x, y )λλ ′ , μμ ′ , where λ ,
λ′, μ and μ′ denote the spin indices.
g = +
k
k
k´ + k´ k−k´ +
k k
+ + + +
+ + +
+ +
Figure II.10.3 Feynman diagrams for g of zero-order (first line), first-order (second line), and
second-order (third, fourth and fifth lines).
Next, we consider the Feynman diagrams in momentum space. If we assume a uniform and isotropic
system, and also assume that the Hamiltonian is time independent so that
U ( x, x′ )λλ ′ , μμ ′ = V ( r − r ′ )λλ ′ , μμ ′ δ ( t − t ′ ) ,
and the Feynman rules in momentum space are in fact analogous to those in relativistic quantum field theory,
except that here we replace the relativistic propagators for the directed lines by the non-relativistic many-
body Green function
⎡ 1 ⎤ ⎡θ ( k − kF ) θ ( kF − k ) ⎤
0
Gαβ ( k , ω ) = δαβ αlim ⎢ ⎥ = δαβ αlim ⎢ + ⎥ (for T = 0).
→0+
⎣ ω − ε k + iα sgn ( ω − μ ) ⎦ → 0 + ω − ε + iα
⎣ k
ω − ε k − iα ⎦
and each G 0 ( k , ω ) line that either starts from and ending at the same point or is linked by the same
interaction line should be interpreted as being G0 <( k , ω ) where Im ⎡⎣G0 <( k , ω ) ⎤⎦ = 2π iδ ω − ε k0 θ ( k F − k ) . ( )
Also, as for Green function in the spacetime coordinates, we need to multiply a factor of (−1) for each closed
loop in the diagram. In addition, each interaction (wavy) line should be labeled by
iV ( q ) = i ∫ d 3 r V ( r ) exp ( −iq ⋅ r ) , where q ≡ k − k ′ is the momentum change involved. The diagrams
contributing to the Green function up to the second order are illustrated in Fig. II.10.3.
For the consideration of spin-1/2 particles, we distinguish in the following two possibilities for the
interaction potential. The first case is for spin-independent interaction, where the interaction potential has the
form of [1(1) 1(2)] in spin space, which represents the unit spin matrix with respect to both particles 1 and 2.
Therefore, we have
U ( q )λλ ′, μμ ′ = U ( q ) δ λλ ′ δ μμ ′ , (II.208)
On the other hand, if the interaction is spin dependent of the form σ (1) i σ ( 2 ) , we find
σ λκ i σκμ ′ = ⎡⎣( σ ) ⎤⎦ = 3δ λμ ′ ,
2
σ λλ ′ i σ μμ = 0 and (II.211)
λμ ′
According the aforementioned rules, the contributions from the first-order diagrams to the Green
function g (k) appropriate for a spin-independent interaction HI given in EQ. (II.200) become:
d 4k′ d 4k′
−i ∫ G0< ( k ′ ) V ( 0 ) G0( k ) G0( k ) + i ∫ G0( k ) G0( k ) G0<( k ′ ) V ( k − k ′ )
(2π ) 4 (2π ) 4
⎡ d 3k ′ d 3k ′ ⎤
= G02( k ) ⎢V ( 0 ) ∫ θ ( kF − k ′ ) − ∫ V ( k − k ′ ) θ ( kF − k ′ ) ⎥ . (II.213)
⎣ (2π ) 3
(2π ) 3
⎦
The aforementioned diagrammatic method provides descriptions for the fermionic causal Green
function g at T = 0 in terms of expansions of G0 and V. For finite temperatures, the corresponding Green
function has to be obtained through the “imaginary time” trick. Upon Fourier transformation to the (k, zν )-
space with zν denoting the poles in the ω-space, we have the following rules for obtaining the nth order
contribution to the function g (k, zν ) given in EQ. (II.166):
1) Draw all topologically distinct connected diagrams with n interaction lines, two external points, and
(2n+1) directed lines.
2) Label each line with a four-momentum ( k , zν ) ; conserve momentum and Im ( zν ) at each vertex.
1
3) For each directed line labeled with a four-momentum ( k , zν ) write a factor G 0( k , zν ) = .
zν − ε k
0
5) Integrate over all internal independent momenta k through ⎡ (2π ) −3 ∫ d 3 k ⎤ and sum over all internal
⎣ ⎦
independent discrete frequencies through ⎡⎣ β −1
∑ ν
⎤⎦
7) Whenever a directed line either closes on itself or is joined by the same interaction line, insert a
convergence factor exp (−zν σ ) with σ → 0−.
As an example, we calculate the contributions of the first-order diagrams shown in Fig. II.10.4.
k, zν
k, zν
k´, zν´ + k´, zν´ k−k´, zν− zν´
k, zν
k, zν
⎡ d 3k ′ 1 d 3k ′ 1 ⎤
G02( k , zν ) ⎢ ∓ V ( 0 ) ∫
(2π ) β3 ∑e − zν ′σ
G0( k ′, zν′ ) − ∫
(2π ) 3
V( k − k′)
β
∑e − zν ′σ
G0( k ′, zν′ ) ⎥ . (II.214)
⎣ ν′ ν′ ⎦
(
Using EQ. (II.154), A( k , ω ) = 2πδ ω − ε k0 , and the identities )
i ωn η
e β 2π n
lim ∑ =− β x
, ωn = for bosons; (II.215)
η →0 +
n iωn − x e −1 β
i ωnη
e β π (2 n + 1)
lim ∑ = β x
, ωn = for fermions; (II.216)
η →0 +
n iωn − x e +1 β
⎡ d 3k ′ d 3k ′ ⎤
G02( k , zν ) ⎢ V ( 0 ) ∫
(2π ) 3 ( )
f ∓ ε k0′ ± ∫
(2π ) 3 ( )
V ( k − k ′ ) f ∓ ε k0′ ⎥ . (II.217)
⎣ ⎦
where the integration along contour C can be analytically deformed into that along contour C′ because the
integral along Γ for |z| → ∞ vanishes for positive η.
We emphasize that the results outlined above are only applicable to normal fermion systems or
bosons at relatively high temperatures. These results must be modified for bosons at low temperatures and
for fermion systems that are not normal, such as in the case of superfluids and superconductors. We shall
leave those more complicated cases to Parts IX and XI later.
In general, it is rare that a small number of lowest-order diagrams would be a good approximation
for the Green function g. Therefore in practical calculations one either tries to obtain certain general results
without employing approximations, or, whenever specific results are sought, to find a class of diagrams such
that the contribution from this whole class is both calculable and dominant. However, it is not very common
that such ideal situation occurs.
z-plane
Γ Γ
C C
C′ C′
C′ C′
z=x
z = iωn
Γ Γ
Fig. II.10.5: Contour for evaluation of Matsubara frequency sums on the complex z-plane.
The general strategy in practical calculations is to reorganize the expansion through certain partial
summations in a graphical way. One may begin with a few simple diagrams (called “skeletons”), and then
perform all possible summations of lines and vertices (corresponding to “putting flesh to the skeletons”) to
obtain the Green function g. To facilitate such calculations, we introduce in the following a few definitions
and important relations. First, we define the self-energy Σ as the sum of contributions from all parts of the
diagrams that are connected to the rest by two directed lines, one in and the other out. This definition can be
explicitly expressed by the following relation:
g ( k , zν ) = G0 ( k , zν ) + G0 ( k , zν ) Σ ( k , zν ) G0 ( k , zν ) . (II.220)
g = G0 + G0 Σ* g = G0 + g Σ*G0 . (II.222)
The diagrammatic expression for EQ. (II.223) is illustrated in Fig. II.10.6 (b).
The proper self energy Σ* contains an infinite number of diagrams, which can be expressed in terms
of g and a quantity Γ, known as the vertex part. The vertex part is defined as the sum of the contributions
from all parts that are connected with four (two in and two out) directed G0 lines to the rest, which cannot be
decomposed into disconnected parts. The vertex part Γ is related to the two-particle Green function g2, as
shown in Fig. II.10.7 and also expressed explicitly below:
G0
(a)
= + Σ
g G0 G0
G0
(b)
= + Σ*
g G0 G
Figure II.10.6 (a) Relation between the Green function g (thick line), the unperturbed Green function
G0 (thin line), and the self energy Σ. (b) Relation between the Green function g (thick line), the
unperturbed Green function G0 (thin line), and the proper self energy Σ*.
g g
g2 = ± +
Γ
g g g g g g
Figure II.10.7. Diagrammatic expression for the relation between the two-particle Green function g2,
the Green function g, and the vertex part Γ.
From EQs. (II.222) and (II.223), it can be shown that the proper self energy Σ* can be calculated
from the four diagrams shown in Fig. II.10.8 (a), which result from “fleshing” the four skeleton diagrams in
Fig. II.10.8 (b) – The “fleshing” is done by replacing the G0-lines by g-lines and one of the two interaction
lines by the vertex part Γ. Moreover, we note that the diagrammatic expression given in Fig. II.10.8 (a) is in
fact consistent with EQ. (II.61) given by the equation of motion method. The basic expression for the Green
function in EQ. (II.222) or EQ. (II.223) together with the diagrams in Fig. II.10.8 (a) for the proper self
energy is known as the Dyson’s equation.
(a) Σ* = + + +
(b) + + +
Figure II.10.8 The proper self energy Σ* is consistent with the sum of the four diagrams in (a), which
are obtained by “fleshing” the four skeleton diagrams in (b).
However, the vertex part Γ generally cannot be expressed in terms of a closed form involving g and
Γ because the skeleton diagrams for Γ are infinite in number. A typical approach is to end the infinite
hierarchy of relations by approximately expressing Γ in a closed form involving g and Γ. The simplest of
such approximations is to set Γ = 0, which is equivalent to the Hartree-Fock approximation that we have
discussed earlier. The proper self energy in the Hartree-Fock approximation is given by:
d k′ 1 d k′
3 3
1
Σ*( k , zν ) = ∓ V (0) ∫
(2π ) β
3 ∑e − zv′σ
G0( k ′, zν′ ) − ∫
(2π )
3
V( k − k′)
β
∑′ e − zv′σ
G0( k ′, zν′ )
ν′ ν
d k′ d k′
4 4
= V (0) ∫ 4
A ( k ′, ω ′ ) f ∓ ( ω ′ ) ± ∫ 4
V ( k − k ′ ) A ( k ′, ω ′ ) f ∓ ( ω ′ ) = Σ*( k ) . (II.245)
(2π ) (2π )
= + +
(a) iV iV iV -iΠ iV
= +
(c) -iΠ*
iV iV iV iV
= +
Figure II.10.10 Diagrammatic expressions for the relation between (a) V , V, and Π; (b) V, Π and Π*;
(c)V , V, and Π* for fermions at T = 0. For T ≠ 0, the imaginary time case, iV , iV, iΠ
and iΠ* must be replaced by −V , −V, −Π and −Π*, respectively.
For a rotational invariant system, if we define the momentum transfer due to the interaction as
q = k − k ′ , the diagrammatic expressions in Fig. II.10.10 can be written as follows:
V (q ) = V (q ) + V 2 (q ) Π (q ) , (II.246)
* * * * *
Π (q ) = Π (q ) + Π (q ) V (q ) Π (q ) + = Π (q ) + Π ( q ) V ( q ) Π ( q ) . (II.247)
Hence, we obtain
Π * (q )
Π (q ) = , (II.248)
1 − V (q ) Π * ( q )
V (q , ω )
V (q , ω ) = , (II.251)
ε c (q , ω )
we can express the dielectric function in terms of the polarization Π and proper polarization Π*:
1 1
= 1 + V (q , ω ) Π (q , ω ) = . (II.252)
ε (q , ω )
c
1 − V ( q , ω ) Π * (q , ω )
The polarization Π defined here is the result of the dielectric response of a many-body system to the inter-
particle interactions, which we shall elaborate in Part III. For comparison, we note that similar polarization
tensor can be introduced in the case of relativistic quantum field theory for quantum fluctuations associated
with photon propagation in vacuum.
Further Readings
For classic references on the application of Green functions to condensed matter physics, you may consider
the following books:
• “Green Functions for Solid State Physicists”, S. Doniach and E. H. Sondheimer, Imperial College Press
(1998).
Despite the apparent simplicity of the Hartree-Fock approximation, self-consistent solutions to the
corresponding Dyson’s equation are already highly non-trivial. As a heuristic example for diagrammatic
analysis of many-body physics, we go over the necessary steps in the following to obtain the self-consistent
solutions to a many-body system under the Hartree-Fock approximation. This approximation is also widely
used in the calculations of electronic bandstructures of crystalline materials.
We begin with setting up a more generalized non-perturbed Hamiltonian H0 that includes an external
potential U ext ( r ) and a perturbation HI that incorporates the inter-particle interaction U ( x, x′ )λλ ′ , μμ ′ . For
simplicity, we assume that both U ext ( r ) and U ( x, x′ )λλ ′ , μμ ′ are spin-independent, so that
U ( x, x′ )λλ ′ , μμ ′ = V ( x − x′ ) δ ( t − t ′ ) δ λλ ′δ μμ ′ .
Hence, we have
⎡ ∇2 ⎤
H 0 = ∫ d 3 xψ α† ( x ) ⎢ − + U ext ( x ) ⎥ ψ α ( x ) , (III.1)
⎣ 2m ⎦
1
d x d x′ψ ( x )ψ ( x′ ) V ( x − x′ ) ψ ( x′ ) ψ ( x ) .
2∫
HI = 3 3 †
α
†
β β α (III.2)
The Green’s function can be obtained via the Dyson’s equation represented in Fig. II.10.6 (b):
where the Kronecker delta function for spin indices has been factored out. The proper self-energy under the
Hartree-Fock approximation (see Fig. II.10.9) is therefore given by
(
Σ*( x1 , x1′ ) = −i δ ( t1 − t1′ ) ⎡⎣δ ( x1 − x1′ )( 2 s + 1) ∫ d 3 x 2 V ( x1 − x 2 ) g x 2 , t2 ; x 2 , t2+ )
(
− V ( x1 − x1′ ) g x1 , t1 ; x1′ , t ⎤⎦
+
1 ) (III.4)
for spin-s fermions. The Fourier transformed Green’s function for the above time-independent H0 and HI
thus becomes:
+ iV ( x1 − x1′ )( 2π ) ∫ dω e
i ωη
g ( x1 , x1′ , ω )
−1
(III.6)
⎡ ∇2 ⎤
H 0 ϕ 0j ( x ) = ⎢ − + U ( x ) ⎥ ϕ 0j ( x ) = ε 0j ϕ 0j ( x ) , (III.7)
⎣ 2m ⎦
the unperturbed Green’s function becomes:
* − i ε ( t −t ′)
iG 0 ( xt , x′t ′ ) = ∑ ϕ 0j ( x ) ϕ 0j ( x′ ) e ⎡⎣θ ( t − t ′ ) Φ 0 a j a †j Φ 0 − θ ( t ′ − t ) Φ 0 a †j a j Φ 0 ⎤⎦
0
j
⎡⎣θ ( t − t ′ ) θ ( ε 0j − ε F0 ) − θ ( t ′ − t ) θ ( ε F0 − ε 0j ) ⎤⎦ ,
* − i ε 0j ( t − t ′ )
= ∑ ϕ ( x ) ϕ ( x′ ) e
0
j
0
j
(III.8)
j
where ε F0 is the Fermi energy of the unperturbed system. Therefore, the Fourier transform EQ. (III.8) is
⎡ θ ( ε 0j − ε F0 ) θ ( ε F0 − ε 0j ) ⎤
G ( x, y ; ω ) = ∑ ϕ ( x ) ϕ ( y ) ⎢
0 0 0 *
+ ⎥, (III.9)
⎣⎢ ω − ε j + iη ω − ε j − iη ⎦⎥
j 0 j 0
j
and the particle density in the unperturbed ground state is [from EQ. (II.140) except now restoring the spin
degrees of freedom]
( )
2
n 0( x ) = − i ( 2 s + 1)( 2π ) ∫ dω e
i ωη
G 0 ( x, x, ω ) = ( 2 s + 1) ∑ ϕ 0j ( x ) θ ε F0 − ε 0j ,
−1
(III.10)
j
N 0 = ∫ d 3 x n 0( x ) = ( 2 s + 1) ∑ θ ε F0 − ε 0j . ( ) (III.11)
j
Equations (III.5), (III.7) and (III.9) form a set of coupled equations for the self-consistent Green’s
function g. Noting that Σ* in EQ. (III.5) is not dependent on ω, we expect that the frequency dependence of g
is similar to that of G 0:
* θ (ε j − ε F ) θ (ε F − ε j ) ⎤
⎡
g ( x, y ; ω ) = ∑ ϕ j ( x ) ϕ j ( y ) ⎢ + ⎥, (III.12)
j ⎣⎢ ω − ε j + iη ω − ε j − iη ⎦⎥
where {ϕ ( x )}
j
denotes a complete set of single-particle wave functions with eigen-energies ε j and ε F is
the Fermi energy of the interacting system. Similar to EQs. (III.10) and (III.11), the particle density of the
interacting system is
( )
2
n ( x ) = ( 2 s + 1) ∑ ϕ j ( x ) θ ε F − ε j , (III.13)
j
N = N 0 = ∫ d 3 x n 0( x ) = ( 2 s + 1) ∑ θ ( ε F − ε j ) , (III.14)
j
which is the same as that of the unperturbed system because the interaction Hamiltonian HI conserves the
total number of particles.
Using EQ. (III.12), we can evaluate the proper self-energy in EQ. (III.6) and find:
∑ ϕ ( x ) θ (ε −ε j )
2
Σ ( x1 , x1′ ) = δ ( x1 − x1′ )( 2 s + 1) d x 2 V ( x1 − x 2 )
∫
* 3
j 2 F
j
− V ( x1 − x1′ ) ∑ ϕ j ( x1 ) ϕ j ( x1′ ) θ ( ε F − ε j ) .
*
(III.15)
j
combining EQs. (III.5) and (III.15) yields a non-linear integral equation for {ϕ ( x )} if we assume that
j
⎡ ∇12 ⎤
L1 = ω − ⎢ − + U ( x1 ) ⎥ = ω − H 0 , (III.16)
⎣ 2m ⎦
so that
L1G 0 ( x1 , x1′ ; ω ) = δ ( x1 − x1′ )
and
L1 g ( x1 , x1′ ; ω ) = δ ( x1 − x1′ ) + ∫ d 3 x 2 Σ*( x1 , x 2 ) g ( x 2 , x1′ ; ω ) . (III.17)
⎡ ∇12 ⎤ ⎡θ (ε j − ε F ) θ (ε F − ε j ) ⎤
⎢ω + − U ( x )
1 ⎥ ∑ ϕ ( x1 ) ϕ ( x1
′ )*
⎢ + ⎥
2m ⎣⎢ ω − ε j + iη ω − ε j − iη ⎦⎥
j j
⎣ ⎦ j
⎡θ (ε j − ε F ) θ (ε F − ε j ) ⎤
− ∫ d 3 x 2 Σ*( x1 , x 2 ) ∑ ϕ j ( x 2 ) ϕ j ( x1′ ) ⎢ ⎥ = δ ( x1 − x1′ ) ,
*
+ (III.18)
j
⎣⎢ ω − ε j + iη ω − ε j − iη ⎦⎥
Multiplying EQ. (III.18) by ϕ ( x1′ ) , integrating over x1′ , using the orthogonality of {ϕ j ( x1 )} , and taking the
η → 0 limit, we obtain the following Schrödinger-like equation for ϕ j ( x1 ) :
⎡ ∇12 ⎤
⎢ − 2m + U ( x1 ) ⎥ ϕ j ( x1 ) + ∫ d x 2 Σ ( x1 , x 2 ) ϕ j ( x 2 ) = ε jϕ j ( x1 ) .
3 *
(III.19)
⎣ ⎦
The physical significance of the proper self-energy Σ* becomes evident from EQ. (III.19), which is
effectively a static non-local potential. Given that Σ* is hermitian and independent of the index j, the
orthogonality of {ϕ j ( x1 )} remains. Moreover, from EQ. (III.15) we note that Σ* contains two terms, a direct
term proportional to the local particle density, and an exchange term involving non-local interaction of
particles.
in EQ. (III.19) to solve for {ϕ ( x )} . The newly obtained {ϕ ( x )} are then applied to EQ. (III.15) to compute
j j
*
Σ , the resulting proper self-energy inserted back to EQ. (III.19), and so on, until self-consistent solutions are
{ } { }
found for both ϕ j ( x ) and ε j . Evidently it is a complicated computation procedure for arbitrary potential
U ( x ) . Therefore, the Hartree-Fock solutions to the eigenfunctions, eigen-energies and bandstructures of
realistic materials are generally carried out with large computers.
Finally, recall that the ground state energy can be evaluated from Green’s functions, we obtain the
corresponding ground state energy for a given potential U ( x ) using Eqs. (II.144) and (III.19):
i dω ⎡ ∇12 ⎤
E0 = − ( 2s + 1) ∫ d 3 x1 xlim
′ →x ∫
e i ωη ⎢ω − + U ( x1 ) ⎥ g ( x1 , x1′ ; ω )
2 1 1 ( 2π ) ⎣ 2m ⎦
dω ⎧1 ⎡1 1 3 ⎤
= −i ( 2 s + 1) ∫ d 3 x1 ∑ ϕ j ( x1 )
*
⎨ ω ϕ j ( x1 ) + ⎢ ε jϕ j ( x1 ) − ∫ d x 2 Σ ( x1 , x 2 ) ϕ j ( x 2 ) ⎥
i ωη
∫ ( 2π ) e
*
j ⎩2 ⎣2 2 ⎦
⎡ θ ( ε j − ε F ) θ ( ε F − ε j ) ⎤ ⎫⎪
×⎢ + ⎥⎬
⎣⎢ ω − ε j + iη ω − ε j − iη ⎦⎥ ⎪⎭
1
(
= ( 2 s + 1) ∑ ε jθ ε F − ε j − ) 2
( 2s + 1) ∫ d 3 x1 d 3 x 2 ∑ ϕ j ( x1 ) Σ*( x1 , x 2 ) ϕ j ( x 2 ) θ ( ε F − ε j ) .
*
j j
1
(
= ( 2 s + 1) ∑ ε jθ ε F − ε j − ) 2
( 2s + 1) ∑ θ ( ε F − ε j ) θ ( ε F − ε j ′ )∫ d 3 x1 d 3 x 2 V ( x1 − x 2 )
j j j′
× ⎡⎢( 2 s + 1) ϕ j ( x1 ) ϕ j ′( x 2 ) − ϕ j ( x1 ) ϕ j ′( x1 ) ϕ j ′( x 2 ) ϕ j ( x 2 ) ⎤⎥
2 2 * *
(III.20)
⎣ ⎦
Equation (III.20) is the usual Hartree-Fock result, where the first term is the simple sum of the energies of all
occupied states. The second term involving the proper self-energy includes both the direct and exchange
interactions. In general for short-range potentials, the direct and exchange interactions are comparable in
magnitude, whereas for long-range potentials, the exchange interaction is much smaller than the direct
interaction. In the special case of a uniform system so that the potential U ( x ) vanishes, plane-wave solutions
satisfy the self-consistent equations, and the corresponding self-consistent single particle energy becomes:
ε k = ε k0 + Σ*( k ) , (III.21)
⎡ 1 ⎤
E0 = ( 2 s + 1) V ( 2π ) ∫ d k ⎢⎣ε − Σ*( k ) ⎥ θ ( k F − k )
−3 3
2 ⎦
k
⎡ 1 ⎤
= ( 2 s + 1) V ( 2π ) ∫ d k ⎢⎣ε k + 2 Σ ( k )⎥⎦ θ ( kF − k ) .
−3 3 0 *
(III.23)
Therefore, in a uniform system the proper self-energy provides a simple correction to the ground-state energy
of an interacting system. However, it should be kept in mind that the simple correction given in EQ. (III.23)
is no longer valid for non-uniform systems.
III.2. The Random Phase Approximation (RPA) for Degenerate Fermi Gas
Here we want to consider the diagrammatic representation of the random phase approximation that
describes the dielectric response of high-density (or, degenerate) electrons. The primary reason for
considering degenerate electrons (which is a valid approximation for good conductors) is that the potential
energy of such a system can be assumed to be nearly uniform, which greatly simplifies the diagrammatic
analysis despite the apparent difficulties in dealing with the long-range Coulomb interaction among
electrons. Specifically, we consider an idealized approximation for the electronic response in a good metal,
known as the “jellium model”, which assumes a high density (n) of electrons moving in a uniform
background of positive charge that ensures overall charge neutrality. The 1/r2 nature of the Coulomb
interaction gives rise to an average potential energy per particle that is proportional to n1/3; whereas the
average kinetic energy per particle is proportional to kF2/(2m) ∝ n2/3 as the result of Pauli principle, with kF
being the momentum at the Fermi level. Thus, in contrast to ordinary gases, the kinetic energy dominates
over the potential energy in the high-density electron system, and therefore the system behaves like a gas,
known as the Fermi gas. On the other hand, in the low-density limit, the potential energy dominates, and the
system becomes a solid, known as the Wigner crystal. For intermediate density, we have the Fermi liquid.
We may quantify the above descriptions by introducing a dimensionless quantity rs ≡ r0/aB, where r0
is defined as (4πr03/3)−1 = n = N/Ω, Ω is the volume of the system, and aB = 2 /(me2 ) is the Bohr radius. If
we define the field operators ψ = ∑ k akψ k and ψ † = ∑ k ak†ψ k* where ψ k denotes the eigen-state, and
introduce dimensionless quantities Ω´ ≡ Ω/r03, k´ ≡ kr0, p´ ≡ pr0, and q´ ≡ qr0, the Hamiltonian H of the
electron system becomes
e2 ⎡ 1 4π † ⎤
2 ⎢ ∑
k ′2 ak†′ ak ′ + s ∑
r
H = ak ′+ q′ ap†′−q′ ap′ ak ′ ⎥ , (III.24)
aB rs ⎣ 2 k ′ 2Ω′ k ′,p′,q′≠ 0 q′ 2
⎦
We note that only Coulomb interactions with q′ ≠ 0 are relevant. In the rs → 0 limit, the potential energy
becomes negligible, and therefore we can limit contributions from the potential energy only up to second-
order terms. Since diagrams with q′ = 0 do not contribute to the inter-particle interaction (because of charge
neutrality for the entire system), the remaining diagrams (see Fig. II.10.3) up to second order in Coulomb
interaction that contribute to the proper self-energy are given in Fig. III.2.1. (You may compare Fig. III.2.1
with Fig. II.10.8 to relate the higher-order terms involving the vertex contributions in the latter to the second
order terms in the former.)
The diagrammatic contributions to Σ1* , Σ*2a and Σ*2b are finite, whereas the diagram Σ*2r is in fact
troublesome and requires more careful analysis. Specifically, the diagram Σ*2r involves a polarization loop.
+ + +
Figure III.2.1 Diagrams relevant to the proper self-energy of degenerate Fermi gas up to second-order.
For simplicity we restrict our consideration to a spatially homogeneous system of particles with a
spin-independent interaction potential:
U 0( x, x′ )λλ ′ , μμ ′ = V ( x − x′ ) δ ( t − t ′ ) δ λλ ′ δ μμ ′ . (III.25)
The interaction energy evaluated relative to the exact ground state becomes
1
d 3 x d 3 x′ V ( x − x′ ) ψ α† ( x )ψ β† ( x′ )ψ β ( x′ )ψ α ( x )
2∫
V =
1
d x d x′ V ( x − x′ ) ⎡⎣ n( x ) n( x′ ) − δ ( x − x′ ) n( x ) ⎤⎦ ,
2∫
3 3
= (III.26)
where n( x ) ≡ ψ α† ( x )ψ α ( x ) is the particle density operator. If we introduce the density fluctuation operator:
n( x ) ≡ n( x ) − n ( x ) , (III.27)
1
d x d x′ V ( x − x′ ) ⎡⎣ n( x ) n( x′ ) + n( x ) n( x′ ) − δ ( x − x′ ) n( x ) ⎤⎦ ,
2∫
3 3
V = (III.28)
where we have used the identity n( x ) = 0 . Although EQ. (III.26) is generally complicated for an arbitrary
interacting system, it can be greatly simplified for a uniform system because the second and third terms in
EQ. (III.28) become trivial. In this special case, n( x ) = n0 = ( N / Ω ) is a constant, so we can concentrate
on the correlation function for the density fluctuation operator n( x ) n( x′ ) :
Ψ 0 T [ n( x ) n ( x ′ ) ] Ψ 0
i Π ( x , x′ ) ≡ = i Π ( x′, x ) . (III.29)
Ψ0 Ψ0
Similarly, we may define the density fluctuation operator for the non-interacting system as follows:
i Π 0( x, x′ ) ≡ Φ 0 T ⎡⎣ψ α† ( x )ψ β† ( x′ )ψ β ( x′ )ψ α ( x ) ⎤⎦ Φ 0
− Φ 0 ψ α† ( x )ψ α ( x ) Φ 0 Φ 0 ψ β† ( x′ )ψ β ( x′ ) Φ 0 = i Π 0( x′, x ) ,
0
= iGαα ( 0
x, x + iGββ) (
x′, x′+ − iGαβ
0
)
( x, x′ ) iGβα0 ( x′, x ) − n( x ) n( x′ ) , (III.30)
where we have employed Wick’s theorem in the last line of EQ. (III.30). For a uniform fermion system of
spin s, the first and the third terms in the last line of EQ. (III.30) exactly cancel, so we obtain
i Π 0( x, x′ ) = ( 2 s + 1) G 0( x, x′ ) G 0( x′, x ) . (III.31)
The physical meaning of Π0 is the lowest-order contribution to the density correlation function of the
interacting system, Π, which is also known as the polarization propagator. The diagrammatic representations
for Π0 in the coordinate space and in momentum space are respectively shown in Figs. III.2.2 (a) and III.2.2
(b). For a degenerate electron gas, the lowest-order polarization insertion Π0 is analogous to the creation and
subsequent annihilation of an electron-hole pair, which is consistent with the notion of density fluctuations in
an otherwise uniform system.
(a) x (b) q
k k+q
x′ q
Figure III.2.2 Lowest-order contribution Π0 to the density correlation function (a) in coordinate space,
and (b) in momentum space.
Therefore, following the convention in EQs. (II.246) and (II.247), we have the effective interaction:
and the corresponding Feynman diagrams in coordinate space are shown in Fig. III.2.3.
x1′ x′
V V
= + + ...
x x′ x x′
x x1
0
(V) (Π ) (V)
Having introduced the polarization propagator, we may rewrite the expectation value for the
interaction potential in a uniform system into:
1
d x d x′ V ( x − x′ ) ⎡⎣i Π ( x, t ; x′, t ′ ) − i Π ( x, t ; x′, t ′ ) ⎤⎦
2∫
3 3 0
+
1
d 3 x d 3 x′ V ( x − x′ ) ⎡⎣i Π ( x, t ; x′, t ′ ) − i Π 0( x, t ; x′, t ′ ) ⎤⎦ .
2∫
= Φ0 V Φ0 + (III.33)
Hence, the ground-state energy of the interacting system can be evaluated as follows:
1 1 dλ
2∫ ∫ d x d x′ λV ( x − x′ ) ⎡⎣i Π ( x, t; x′, t ′ ) − i Π ( x, t; x′, t ′ )⎤⎦
λ
E = Φ0 H Φ0 + 3 3 0
0
λ
= Φ 0 H Φ 0 + Ecorr , (III.34)
where Πλ denotes the renormalized polarization associated with an interaction strength λ, and the correlation
energy is given by
1 1 dλ
2∫ ∫ d x d x′ λV ( x − x′ ) ⎡⎣i Π ( x, t; x′, t ′ ) − i Π ( x, t; x′, t ′ )⎤⎦
3 3 λ 0
Ecorr =
0
λ
1 1 dλ
Ω ( 2π ) ∫ ∫d q λV ( q ) ⎡⎣i Π λ ( q, ω ) − i Π 0( q, ω ) ⎤⎦ ,
−4 4
= (III.35)
2 0
λ
We also note that the expression in EQ. (III.34) has been obtained through a “trick” by Pauli, which asserts
that the exact ground state energy (E) of an interacting system is related to the ground state energy (E0) of the
unperturbed non-interacting system by the following relation
1 dλ
E − E0 = ∫ Ψ 0( λ ) λ H I Ψ 0( λ ) , (III.37)
0
λ
d d Ψ 0( λ ) d Ψ 0( λ ) dH (λ )
E( λ ) = H ( λ ) Ψ 0( λ ) + Ψ 0( λ ) H ( λ ) + Ψ 0( λ ) Ψ 0( λ )
dλ dλ dλ dλ
d
= E( λ ) Ψ 0( λ ) Ψ 0( λ ) + Ψ 0( λ ) H I Ψ 0( λ ) = Ψ 0( λ ) H I Ψ 0( λ ) (III.38)
dλ
for 0 ≤ λ ≤ 1 , which leads to the solution for ( E − E0 ) through integration over the coupling constant λ.
Next, if we further incorporate Dyson’s equation that allows the following conversion through the
introduction of a proper polarization propagator Π*,
∫d
4
x1 V ( x, x1 ) Π ( x1 , x′ ) = ∫ d 4 x1 V ( x, x1 ) Π * ( x1 , x′ ) , (III.39)
i 1 dλ
Ω ( 2π ) ∫ ∫d q ⎡⎣V λ ( q ) Π *λ ( q ) − λV ( q ) Π 0( q ) ⎤⎦ ,
−4 4
Ecorr = (III.40)
2 0
λ
where V λ denotes the effective interaction under an interaction strength λ. Finally, using EQs. (II.251) and
(II.252), we find
V ( q ) Π *( q ) 1 − ε c( q ) 1
V ( q ) Π *( q ) = = = −1 . (III.41)
1 − V (q ) Π (q ) *
ε (q)
c
ε (q)
c
where ε c ( q ) is the dielectric constant, which can be determined empirically. Hence, EQ. (III.40)
Ecorr =
i
2
Ω ( 2π )
−4
∫
0
1 dλ
λ ∫d
4
{ −1
q ⎡⎣ε c ( q ) ⎤⎦ − 1 − λV ( q ) Π 0( q ) . } (III.42)
To proceed further, we expand the proper polarization propagator Π* in a perturbation series so that
Π * = Π *( 0 ) + Π *(1) + … where Π *( 0 ) = Π 0 . Thus, EQ. (II.250) can be rewritten into the following:
V ( q ) Π *( q )
V (q) Π (q ) =
*
= V Π* + V Π*V Π* + …
1 − V (q) Π (q ) *
= V ⎡⎣ Π *( 0) + Π *(1) ⎤⎦ + V Π *( 0)V Π *( 0 ) + …
= V Π 0 + V Π *(1) + V Π 0V Π 0 + … . (III.43)
The first-order proper polarization Π *(1) contains first-order interaction potential V, and the contributions are
shown in Fig. III.2.4 (a) – (e).
Figure III.2.4 All first-order diagrammatic contributions to the proper polarization propagator.
In reality, the diagrams in (a) and (e) do not contribute to the correlation energy because for a neutral
system we have V(0) = 0. Consequently, if we keep the leading terms to V 2, the correlation energy becomes
i 1 dλ 2
Ω ( 2π ) ∫ ∫d q ⎡⎣ λV ( q ) Π 0( q ) ⎤⎦ ,
−4 4
E2r = (III.45)
2 0
λ
i 1 dλ
Ω ( 2π ) ∫ ∫d q ⎡⎣ λV ( q ) Π *(1)b ,c ,d ( q ) ⎤⎦ .
−4
E2b ,c , d = 4
(III.46)
2 0
λ
Here the subscript “2” refers to second-order in V. It turns out that the contributions in EQ. (III.46) are finite,
which you may try to verify. On the other hand, the contribution in EQ. (III.45) diverges logarithmically
because E2r ∼ [V ( q )] ∝ q −4 . In fact, all terms of the form Enr in the perturbation expansion with n ≥ 2
2
diverge. Nonetheless, if we keep in EQ. (III.43) all the terms of the form Enr with n ≥ 2 such that
V Π0
*
( 0
V Π = V Π +V Π VΠ +… +V Π 0 0
) *
(1) +… =
1−V Π 0
+ V Π *(1) + …
we have
⎛ ∞
⎞
Ecorr = Er + E2b + E2c + E2d + … = ⎜
⎝
∑En=2
r
n ⎟ + E2 + E2 + E2 + … ,
⎠
b c d
(III.47)
where
∞
dλ ∞
∑ ∑ ⎡⎣λV ( q ) Π ( q )⎤⎦
i 1
Ω ( 2π ) ∫ ∫d q
−4 4 0 n
Er = Enr =
n=2 2 0
λ n=2
2
i ⎡⎣ λV ( q ) Π 0( q ) ⎤⎦
1 dλ
= Ω ( 2π ) ∫
λ ∫
−44
d q
2 0
1 − λV ( q ) Π 0( q )
i −4 1 d λ
≡ Ω ( 2π ) ∫ ∫ d 4 q λV ( q ) Π 0( q ) Vrλ ( q ) Π 0( q ) , (III.48)
2 0
λ
and
V (q) V (q)
Vr ( q ) ≡ ≡ . (III.49)
1 − V (q ) Π (q ) 0
ε r (q )
The “ring approximation” for the interaction potential given in EQ. (III.49) consists of an infinite series of
ring (i.e. polarization propagator) diagrams, with each ring approximated by its lowest-order term Π0, as
shown in Fig. III.2.5. The diagrammatic expressions for the correlation energy in EQ. (III.47) are illustrated
in Fig. III.2.6. These results based on summing over the ring diagrams are known as the random phase
approximation.
= + + + ...
Π 0( q ) = Π 0( q, q0 ) = −i ( 2 s + 1)( 2π ) ∫d k G 0( k ) G 0( k + q )
−4 4
( 2s + 1) 3 ⎡ θ ( k − kF ) θ ( kF − k ) ⎤ ⎡ θ ( k + q − kF ) θ ( kF − k + q ) ⎤
4 ∫
= −i d k ∫ dω ⎢ + ⎥ ⎢ + ⎥
( 2π ) ⎣ ω − ωk + iη ω − ωk − iη ⎦ ⎣ ( ω + q0 ) − ωk +q + iη ( ω + q0 ) − ωk +q − iη ⎦
2 ⎡θ ( k + q − kF )θ ( kF − k ) θ ( kF − k + q )θ ( k − kF ) ⎤
( 2π ) ∫ ⎣
3
= d k⎢ − ⎥. (III.50)
q0 + ωk − ωk + q + iη q0 + ωk − ωk + q − iη
3
⎦
In the last line of EQ. (III.50), we have assumed spin-1/2 particles and note that the frequency integral
reduces four terms in EQ. (III.50) down to two, because those with poles on the same side of the complex
frequency plane do not contribute to the integration. For non-interacting systems, the eigen-energies are
(
given by ωk = ε k0 = k 2 ( 2m ) , so that ωqk ≡ ωk + q − ωk = k i q + q 2 / 2 m . We may further simplify EQ. )
(III.50) with the change of variables k ′ → − ( k + q ) in the second term, so that
2 ⎡ 1 1 ⎤
Π 0( q, q0 ) = d k θ ( k + q − k )θ ( k − k )⎢
( 2π ) ∫
3
− ⎥. (III.51)
⎣ q0 + ωk − ωk +q + iη q0 + ωk + q − ωk − iη ⎦
3 F F
In the limit of q0 → ∞ , EQ. (III.51) yields Π 0( q, q0 ) ∼ O q0−2 ( ) so that the polarization propagator is an
even function of large q0. By employing the following relation for the step function
2 ⎡ 1 1 ⎤
{
Re Π 0( q, q0 ) = } P ∫ d 3 k ⎡⎣1 − θ ( k F − k + q ) ⎤⎦ θ ( k F − k ) ⎢ − ⎥
( 2π ) ⎣ q0 − ωqk q0 + ωqk ⎦
3
2 ⎡ 1 1 ⎤
= P ∫ d 3k θ ( kF − k ) ⎢ − ⎥
( 2π ) ⎣ q0 − ωqk q0 + ωqk ⎦
3
⎛ mq0 ⎞ 2mk F ⎡ 1 1 ⎤
⎜ν ≡ 2 ⇒ ⎟ = P ∫ d 3 k θ (1 − k ) ⎢ − ⎥, (III.53)
⎠ ( 2π ) ⎣⎢ν − qk cos ϑ − ( q / 2 ) ν + qk cos ϑ + ( q / 2 ) ⎦⎥
3 2 2
⎝ kF
where in the last line of EQ. (III.53) we measure all wave vectors in units of kF. We also note that the
following term in the first line of EQ. (III.53) vanishes by exchanging k+q and k so that:
⎡ 1 1 ⎤
P ∫ d 3k θ ( kF − k + q )θ ( kF − k ) ⎢ − ⎥
⎣ q0 − ωqk q0 + ωqk ⎦
⎡ 1 1 ⎤
= P ∫ d 3k θ ( kF − k )θ ( kF − k + q ) ⎢ − ⎥
⎣ q0 + ωqk q0 − ωqk ⎦
⎡ 1 1 ⎤
⇔ P ∫ d 3k θ ( kF − k + q )θ ( kF − k ) ⎢ − ⎥=0
⎣ q0 − ωqk q0 + ωqk ⎦
⎧ ⎛ν q ⎞ ⎛ν q ⎞ ⎫
⎪ 2 1+ ⎜ − ⎟ 2 1+ ⎜ + ⎟ ⎪
2mk F ⎪ 1 ⎡ ⎛ν q ⎞ ⎤ ⎝ q 2 ⎠ − 1 ⎡1 − ⎛ ν + q ⎞ ⎤ ln ⎝q 2⎠ ⎪
Re {Π 0( q,ν )} = 2 ⎨ −1 + ⎢1 − ⎜ − ⎟ ⎥ ln ⎢ ⎜ ⎟ ⎥ ⎬
4π ⎪ 2q ⎣⎢ ⎝ q 2 ⎠ ⎦⎥ ⎛ ν q ⎞ 2q ⎢⎣ ⎝ q 2 ⎠ ⎦⎥ ⎛ν q ⎞ ⎪
1− ⎜ − ⎟ 1− ⎜ + ⎟
⎪⎩ ⎝q 2⎠ ⎝ q 2 ⎠ ⎪⎭
(III.54)
Ecorr = + + +
Er = + + + ...
1
Im {Π 0( q, q0 )} = − ∫ d k θ ( k + q − k )θ ( k
3
− k ) ⎡⎣δ ( q0 − ωqk ) + δ ( q0 + ωqk ) ⎤⎦ . (III.55)
4π 2 F F
The physical significance of EQ. (III.55) is associated with the energy absorption required to move a particle
from under the Fermi sea ( k < k F ) to outside the Fermi sea ( k + q > k F ) via conservation of energy, and
the imaginary part of the polarization is directly proportional to the absorption probability for transferring the
four momentum (q, q0) to a free Fermi gas.
⎛ 1 ⎞
Im {Π 0( q,ν > 0 )} = −
mk F
∫ d k ⎡⎣θ ( k + q − 1) θ (1 − k )⎤⎦ δ ⎜⎝ν − k iq − 2 q
3 2
⎟. (III.56)
4π 2
⎠
The integration in EQ. (III.56) may be evaluated by considering two unit Fermi spheres with their centers
separated by |q|. We define the upper Fermi sphere as one determined by k < 1 and and lower Fermi sphere
as one associated with momentum transfer |q|. The non-trivial integration over k must ensure that k + q > 1
and that the integration area involves the intersection between the Fermi sphere k < 1 and a plane defined by
( ) (
the energy conservation condition ν = k i q + q 2 / 2 ≡ kq cos ϑ + q 2 / 2 . Hence, there are three distinct )
possibilities of q and ν for evaluating the integration in EQ. (III.56), as specified below.
1 1
1) q > 2, q2 + q ≥ ν ≥ q2 − q :
2 2
In this case, the two Fermi spheres do not intersect, and the integration becomes (taking z = cos ϑ )
1 ⎛ν ⎞
Im {Π 0( q,ν )} = −
mk F 1 1 q
2π ∫ν k 2 dk ∫−1 dz qk δ ⎜⎝ qk − 2k − z ⎟⎠
4π 2 q
−
q 2
mk F ⎡ ⎛ ν q ⎞ ⎤
2
=− 1
⎢ ⎜− − ⎟ ⎥. (III.57)
4π q ⎣⎢ ⎝ q 2 ⎠ ⎦⎥
1 1
2) q < 2, q2 + q ≥ ν ≥ q2 − q :
2 2
Here the two Fermi spheres intersect so that the intersection regime of the two Fermi spheres is forbidden.
However, the intersection area between the upper Fermi sphere with k < 1 and the plane is not affected as
long as ν ≥ ( q 2 / 2) − q . Consequently, the integration yields the same result as in case 1):
mk F ⎡ ⎛ ν q ⎞ ⎤
2
Im {Π ( q,ν )}
0
=− ⎢1 − ⎜ − ⎟ ⎥ . (III.58)
4π q ⎣⎢ ⎝ q 2 ⎠ ⎦⎥
1
3) q < 2, 0 ≤ ν ≤ q − q2 :
2
In this case, the two Fermi spheres intersect, and the intersecting plane passes through the forbidden
Fermi sphere at the bottom, so the allowed region of intersection becomes an annulus. The area of the
annulus has a minimum value of k allowed by the conservation of energy:
⎛q ν ⎞ ⎡ ⎛q ν ⎞ ⎤
2 2
k 2
min
= ⎜ − ⎟ + ⎢1 − ⎜ + ⎟ ⎥ = 1 − 2ν
⎝ 2 q ⎠ ⎢⎣ ⎝ 2 q ⎠ ⎥⎦
⎛ν ⎞
Im {Π 0( q,ν )} = −
mk F 1 k dk 1 q
2π ∫ 1/ 2 ∫−1 dz δ ⎜⎝ qk − 2k − z ⎟⎠ .
4π 2 (1− 2ν ) q
mk F mk ν
=− ⎡1 − (1 − 2ν ) ⎦⎤ = − F .
⎣ (III.59)
4π q 2π q
Given EQs. (III.54) and (III.57) – (III.59) for Π 0( q, q0 ) , we can obtain estimates for the correlation
energy, dielectric constant, and other useful physical properties under the random phase approximation
(RPA). For instance, it can be shown from EQs. (III.49) and (III.51) that the dielectric constant under the
RPA is given by
1 1
=
ε r ( q, q0 ) 1 − V ( q ) Π 0( q, q0 )
2
4π e 2 ⎧ 1 1 ⎫
= 1− ∑ Φn ∑c †
k −q k
c Φ0 ⎨ + ⎬, (III.60)
⎩ q0 + ωn 0 + iη − q0 + ωn 0 − iη ⎭
2
q n k
where Φ 0 and Φ n refer to the unperturbed ground and excited states, and ωn0 denotes the energy
difference between the states Φ 0 and Φ n . The sum over the expectation values of density fluctuations
associated with all excitations as given in EQ. (III.60) lends the notation of “random phase” and hence the
random phase approximation. Later in one of the problem sets you’ll be asked to consider various interesting
limits of the random phase approximation.
Finally, we summarize in the following a few limiting forms of the zero-order polarization propagator,
which are often useful for various applications:
{
Im Π 0( q, 0 ) = 0 , }
⎡ 1 ⎤
1− q ⎥
mk F ⎢ ⎞ 1⎛ 1
Re {Π ( q, 0 )} =
0
2 ⎢
−1 + ⎜ 1 − q ⎟ ln 2 2
⎥. (III.61)
2π ⎢ q⎝ 4 ⎠ 1
1+ q ⎥
⎣ 2 ⎦
{
Im Π 0( 0,ν ) = 0 , }
{ }
mk F
Re Π 0( q,ν ) ≈ 2
q2 for q → 0. (III.62)
3ν π
2
{
Im Π 0 ( q, qx ) = − } mk F x
2π
for q → 0, 0 ≤ x ≤ 1 ,
=0 for q → 0, x > 1
⎛ 1+ x ⎞
{
Re Π 0( q,ν ) } mk
≈ − F2
2π
⎜ 2 − x ln ⎟
1− x ⎠
for q → 0. (III.63)
⎝
Further Readings:
1. Fetter and Walecka, “Quantum Theory of Many-Particle Systems”, Sections 7 – 10, 13 – 15, 44 – 47.
2. Abrikosov, Gorkov, and Dzyaloshinski, “Methods of Quantum Field Theory in Statistical Physics”,
Chapters 2 and 3.
3. Doniach and Sondheimer, “Green’s Functions for Solid State Physicists”, Chapters 3, 5 and 6.
∂ Ψ S(t )
i = H 0 Ψ S(t ) , ( IV.1)
∂t
so that
−i H 0 t
ΨS(t ) = e Ψ S( 0) for t ≤ t0 . (IV.2)
Next, if we turn on a time-dependent external Hamiltonian Hex (t) at t = t0, the new state vector Ψ S ( t ) for
the modified Schrödinger equation
∂ Ψ S(t )
i = ⎡⎣H 0 + H ex
( t ) ⎤⎦ Ψ S(t ) (IV.3)
∂t
becomes
ΨS(t ) = e
−i H 0 t
A (t ) Ψ S(0) , (IV.4)
∂A ( t )
i =e
iH 0 t
H ex
( t ) e−i H 0t
A ( t ) ≡ H Iex ( t ) A ( t ) for t > t0 , (IV.5)
∂t
A(t ) = 1 for t ≤ t0 , (IV.6)
where H Iex ( t ) denotes the external Hamiltonian operator in the interaction picture. For convenience,
however, in the following we redefine H Iex ( t ) by H ( t ) with the understanding that it is in the interaction
ex
picture. We further note that our following consideration does not require H ex ( t ) commute with H0.
Noting that H ex
( t ≤ t0 ) = 0 and assuming small perturbation at t > t0 , we obtain the solution to A(t)
in EQs. (IV.3) and (IV.4) iteratively to the first order of H ex
(t ) :
A ( t ) = 1 − i ∫ dt ′ H (t′) + … .
t ex
(IV.7)
t0
ΨS(t ) = e
−i H 0 t
Ψ S( 0) − i e
−i H 0 t
( t′) Ψ S(0) + … .
t
∫t
ex
dt ′ H (IV.8)
0
= Ψ ′S ( 0 ) ⎡1 + i ∫
ex
( t ′ ) ⎤⎦ e i H 0 t OS ( t ) e −i H 0 t ⎡⎣1 − i ∫t dt ′ H ex ( t ′ ) ⎤⎦ Ψ S ( 0 ) + …
t t
dt ′ H
⎣ t0 0
= Ψ ′H ( 0 ) OH ( t ) Ψ H ( 0 ) + i Ψ ′H ( 0 ) ∫ ( t ′ ) ,OH ( t ) ⎤⎦ Ψ H ( 0) + … ,
t
dt ′ ⎡⎣H ex
(IV.9)
t0
δ O (t ) ≡ O (t ) ex
− O (t )
( t ′ ) ,OH ( t ) ⎤⎦
t
∫t dt ′ ⎡⎣H ex
= i Ψ0 Ψ0 . (IV.10)
0
As an explicit example, let us consider the linear response of a system with charge e per particle to
an external scalar potential ϕ ex ( x, t ) , which is turned on at t = t0. If the exact particle density operator in the
unperturbed system is denoted by n ( x, t ) , the external perturbation Hamiltonian becomes:
H ex
( t ) = ∫ d 3 x n ( x, t ) e ϕ ex ( x, t ) , (IV.11)
and the corresponding linear response of the system, following EQs. (IV.10) and (IV.11), is given by
δ n ( x, t ) = i ∫ dt ′∫ d 3 x′ eϕ ex ( x′, t ′ ) Ψ 0 ⎡⎣ n ( x′, t ′ ) , n ( x, t ) ⎤⎦ Ψ 0
t
t 0
= i ∫ dt ′∫ d 3 x′ eϕ ex ( x′, t ′ ) Ψ 0 ⎡⎣ n ( x′, t ′ ) , n ( x, t ) ⎤⎦ Ψ 0 ,
t
(IV.12)
t0
where in the second line we have used the definition in EQ. (III.27) for the density fluctuation function.
Next, we define the retarded density correlation function in analogy with that for the retarded Green’s
function so that
Ψ 0 [ n ( x ) , n ( x′ ) ] Ψ 0
iGnR ( x, x′ ) = θ ( t − t ′ ) , (IV.13)
Ψ0 Ψ0
which enforces the causal behavior of the linear response through the retarded density correlation function.
Here we remark that the association of linear response with the retarded Green function may be understood
in terms of causality.
Hence, the Fourier transformed equivalence to the linear response function of EQ. (IV.14)
δ n( k , ω ) = GnR ( k , ω ) eϕ ex ( k , ω ) . (IV.18)
From the relation given in EQ. (IV.18) we may identify the retarded correlation function GR as a generalized
susceptibility representing the linear response of a many-body system to an external perturbation.
IV.2. Kubo Formula for Dielectric Response, Conductivity and Magnetic Susceptibility
Having defined the general theory of linear response to external perturbation, we are ready to consider
several explicit examples.
H ex
( x, t ) = ∫ d 3 x′ n ( x, t ) V ( x − x′ ) nex ( x′, t ) , (IV.19)
where V ( x − x′ ) denotes the Coulomb potential. From EQs. (IV.18) and (IV.19), we obtain the induced
charge density in response to the external charge density:
nex ( q, ω )
δ n( q, ω ) = GnR ( q, ω ) V ( q ) nex ( q, ω ) ≡ − nex ( q, ω ) , (IV.20)
ε ( q, ω )
which leads to a dielectric constant
1 4π e 2
= 1 + GnR ( q, ω ) V ( q ) = 1 + GnR ( q, ω ) , (IV.21)
ε ( q, ω ) q
2
where GnR is defined according to EQ. (IV.13) and is therefore related to the polarization propagator. Thus, if
the external charge density is specified, we may use the corresponding Π0(q,ω) in Part II.8 to derive related
physical quantities such as the dielectric constant, the induced charge density, the total induced charge, etc.
Equation (II.559) represents the linear response of a degenerate electron gas to external charge density,
which is given in terms of the density correlation function and is also known as a generalized Kubo formula
for dielectric response.
ϕ ex ( x, t ) = ei q i xϕ0 δ ( t ) ⇒ ϕ ex ( k , ω ) = ( 2π ) ϕ0 δ ( k − q ) .
3
(IV.22)
d 3k dω
δ n( x, t ) = − e ∫ e −iωt Π R ( k , ω ) ϕ ex ( k , ω )
( 2π ) ∫ ( 2π )
ik ix
3
e
= − eϕ0 ei q i x ∫
dω
( 2π )
e −iωt Π R ( q, ω ) = − eϕ0 ei q i x ∫
dω
( 2π )
e−iωt ⎡⎣V ( q ) ⎤⎦
−1
{⎡⎣ε (q, ω )⎤⎦ − 1}
R −1
(IV.23)
and the poles of the integrand in EQ. (IV.23) represent the frequency and lifetime of the collective modes of
the degenerate electron gas.
ε rR ( q, ω ) = 1 − V ( q ) Π 0 R ( q, ω ) , (IV.24)
and the linear response function associated with the retarded density correlation function is
{ }
Π 0 R ( q, ω ) = Re Π 0( q, ω ) + i sgn ( ω ) Im Π 0( q, ω ) { }
d 3k ⎧
⎪ [1 − θ ( k F − k + q )]θ ( k F − k ) θ ( k F − k + q ) [1 − θ ( k F − k )] ⎫⎪
= 2∫ 3 ⎨
− ⎬
( 2π ) ⎩⎪ ω + ωk − ωk +q + iη ω + ωk − ωk +q + iη ⎭⎪
⎪θ ( k F − k + q ) − θ ( k F − k ) ⎫⎪
d 3k ⎧
= −2 ∫ ⎨ ⎬. (IV.25)
( 2π ) ⎩⎪
3
ω + ωk − ωk +q + iη ⎭⎪
Suppose that the poles of the integrand in EQ. (IV.23) are denoted by ω = Ω p − iγ p , we have
(
V ( q ) Π 0 R q, Ω p − iγ p = 1 . ) (IV.26)
In general, EQ. (IV.26) can only been solved numerically. However, in the special case of small
damping so that γ p Ω p , the real and imaginary parts of EQ. (IV.26) can be separated so that we obtain the
following conditions (see Problem Set 3) the following conditions are satisfied:
{ }
V ( q ) Re Π 0 R ( q, Ω p ) = V ( q ) Re Π 0( q, Ω p ) = 1 , { } (IV.27)
−1
⎡ ∂ Re Π 0 R ( q, ω ) ⎤
γ q = Im {Π ( q, Ω p )}
0R ⎢ { } ⎥
⎢ ∂ω ⎥
⎣ Ωp ⎦
−1
⎡ ∂ Re Π 0( q, ω )
= sgn ( Ω p ) Im {Π 0( q, Ω p )} ⎢
{ } ⎤⎥ . (IV.28)
⎢ ∂ω ⎥
⎣ Ωp ⎦
1/ 2
⎛ 4π ne 2 ⎞ ⎡ 9 ⎛ q 2ε F ⎞ ⎤ ⎡ 9 ⎛ q 2ε F ⎞ ⎤
Ωq = ± ⎜ ⎟ 1 +
⎢ 10 ⎜ 6π ne 2 ⎟ + ⎥ ≡ ±ω p ⎢
1+ ⎜ 2 ⎟
+ ⎥, (IV.29)
⎝ m ⎠ ⎣ ⎝ ⎠ ⎦ ⎣ 10 ⎝ 6π ne ⎠ ⎦
In EQ. (IV.20), we note that the total charge density is given by ntot = (nex/ε). Furthermore, we may
relate the dielectric screening response to the longitudinal conductivity σ ≡ δ J / E = ( J tot − J ex ) / E ,
where E denotes the electric field, δ J is the induced current in response to an external current Jex, and the
total current is related to the total charge density via the continuity equation
∂ntot ( x) q ⋅ δ J ( q, ω ) qσ E ( q, ω )
e + ∇ ⋅ J tot ( x ) = 0 ⇒ δ n ( q, ω ) = = . (IV.30)
∂t ωe ωe
Recall the Maxwell’s equation:
δ n( x) 1 − ε ( q, ω )
∇ ⋅ E = 4π entot ( x) = 4π e ⇒ δ n ( q, ω ) = iq E ( q, ω ) , (IV.31)
1− ε 4π e
so that EQs. (IV.30) and (IV.31) lead to a simple relation between the dielectric constant and the longitudinal
conductivity:
4πσ
ε ( q, ω ) = 1 + i . (IV.32)
ω
Historically the theory for electrical conduction prior to the development of many-body physics was based
on the semi-classical one-particle approach that assumes random scattering of single-electron wavefunctions
governed by the Boltzmann equation. With the introduction of Green’s function techniques and the
aforementioned linear response theory, we can derive the electrical conductivity of a system directly from the
current-current correlation function with only one simple assumption that the induced current is linear in the
applied voltage, as first discussed by Kubo.
To derive the Kubo formula for electrical conductivity, we consider a perturbation field Hex associated
with an applied vector potential A( x, t ) so that H = H0 + Hex and
H ex = − d x ⎡⎣ A( x ) i J ( x ) ⎤⎦ ,
∫
3
(IV.33)
where J(x) is the total current density operator (including the external and induced current densities):
1
J( x ) =
2m
∑ {[p i
− neA( x )] δ ( x − xi ) + δ ( x − xi ) [p i − neA( x )]}
i
e2
⎡ψ †( x ) ∇ψ ( x ) − ( ∇ψ †( x ) )ψ ( x ) ⎤⎦ − A( x )ψ †( x )ψ ( x ) .
e
=− (IV.34)
2im ⎣ m
compute the average of the current density operator in the presence of a specific external field, we use the
density matrix ρ ( x, t ) such that
J ( x, t ) = Tr { ρ ( x ) J ( x )} = Tr {ρ ( x, t ) J ( x, t )} , (IV.35)
where
∂ρ
ρ ( x, t ) ≡ ei H 0 t ρ ( x ) e − i H 0 t and i = [H 0 + H ex , ρ ] . (IV.36)
∂t
−∞
≈ ρ 0 − i ∫ dt ′ [H ex ( t ′ ) , ρ 0 ] ,
t
(IV.38)
−∞
where ρ0 is the equilibrium density matrix, which is only a function of H0, and we have assumed linear
response in arriving at the second line of EQ. (IV.38).
J ( x, t ) ≈ Tr {ρ 0 J ( x, t )} − i ∫ dt ′ Tr {[H ex ( t ′ ) , ρ 0 ]J ( x, t ′ )}
t
−∞
2
ne
A( x, t ) − i ∫ dt ′ Tr {[ J ( x, t ′ ) , H ex ( t ′ )] ρ 0 }
t
=−
m −∞
ne2
A( x, t ) − i ∫ dt ′ [ J ( x, t ′ ) , H ex ( t ′ )] 0 ,
t
≡− (IV.39)
m −∞
where we have used the following relations to arrive at the second line of EQ. (IV.39):
{
Tr ρ 0ψ †ψ ≡ n , } (IV.40)
Tr { ABC} = Tr { BCA} = Tr {CAB} , (IV.41)
and n is the number of electrons per unit volume. Finally, inserting the interaction picture of EQ. (IV.31) into
EQ. (IV.39), we obtain:
ne2 ∞
J α ( x, t ) = − Aα ( x, t ) − i ∫ dt ′∫ d 3 x′ θ ( t − t ′ ) ⎡⎣ J α ( x, t ) , J β ( x′, t ′ ) ⎤⎦ Aβ ( x′, t ′ ) ,
−∞ 0
m
∞ ⎧ ne 2
⎫
= − ∫ dt ′∫ d 3 x′ ⎨ δ αβ δ ( x − x′ ) δ ( t − t ′ ) + iθ ( t − t ′ ) ⎡⎣ Jα ( x, t ) , J β ( x′, t ′ ) ⎤⎦ ⎬ Aβ ( x′, t ′ )
−∞ 0
⎩m ⎭
which is the Kubo formula for transverse electrical conductivity in coordinate space, where α and β refer to
different Cartesian suffixes, the response function Rαβ is proportional to the conductivity, and we have
assumed transverse perturbation so that q i A( q ) = 0 and there is no need to worry about the internal fields
arising from induced charge density as in the previous case for dielectric response.
We may further Fourier transform EQ. (IV.42) and recall the relation between the electric field and
the vector potential and the definition of electrical conductivity σ αβ ( q, ω ) :
∂A ( x, t )
E ( x, t ) = − ⇒ E ( q, ω ) = iω A ( q, ω ) , (IV.43)
∂t
Jα ( q, ω ) = σ αβ ( q, ω ) Eβ ( q, ω ) , (IV.44)
ne2 1 0
J α ( q, ω ) = − Eα ( q, ω ) + ∫ dt ′ e −iωt ′ ⎡⎣ J α ( q, 0 ) , J β ( −q, t ′ ) ⎤⎦ Eβ ( q, ω ) , (IV.45)
iω m ω −∞ 0
and the final Kubo formula for electrical conductivity is therefore given by
ne2 1 0
σ αβ ( q, ω ) = − δαβ + ∫ dt ′ e −iωt′ ⎡⎣ Jα ( q, 0 ) , J β ( −q, t ′ ) ⎤⎦ . (IV.46)
iω m ω −∞ 0
The magnetic susceptibility of a many-body system can be obtained by considering the linear response of
spins to an applied magnetic field H(x,t). The external perturbation Hamiltonian may be written as:
H ex = − ∫ d 3 x H( x, t ) i m( x ) , (IV.47)
m ( x ) = ∑ δ ( x − x i ) σ ( xi ) , (IV.48)
i
and σ(xi) denotes the spin operator at the position xi, with the components of the vector σ(xi) represented by
the Pauli matrices.
mα ( x, t ) H
= mα ( x, t ) 0 + ∑ ∫ dt ′ ∫ d 3 x′ χαβ ( x − x′, t − t ′ ) H β ( x′, t ′ ) , (IV.49)
β
where the susceptibility tensor is given by the retarded correlation function of the magnetic moment operator:
If we further define
1
σ± ≡
2
(σ x
± iσ y ) , (IV.51)
then we have the transverse and longitudinal susceptibility χ−+ and χzz given by:
χ zz ( x − x′, t − t ′ ) = iθ ( t − t ′ ) [σ z ( x, t ) , σ z ( x′, t ′ )] = ∑ p ,q e
i q i ( x − x′ )
χ zz ( p, q; t − t ′ ) . (IV.53)
For isotropic or cubic media in the paramagnetic state (i.e., T > TCurie), χαβ is diagonal and isotropic from
symmetry consideration, and therefore
implying that χ−+ determines the magnetic susceptibility completely. On the other hand, for T < TCurie
χzz differs from χ−+ and must be calculated separately. In Problem Set 4 you are asked to consider the
transverse magnetic susceptibility in the generalized Hartree-Fock approximation.
In addition to the derivation of linear response functions, the aforementioned linear response theory
can be applied to the fluctuation-dissipation theorem. Consider a liquid or a gas under thermal equilibrium in
which the random impacts of molecules produce irregular driving forces. If the Brownian motion of particles
is driven by an applied force, the same molecular impacts produce frictional resistive forces that can be
described by certain macroscopic quantities. Since the random and systematic parts of the microscopic forces
have the same physical origin, we expect a mathematical relation between them. In its general form, this
relation is known as the fluctuation-dissipation theorem. [Ref.: R. Kubo, J. Phys. Soc. (Japan) 12, 570
(1957)] In quantum mechanical terms, the fluctuations of a system in thermal equilibrium may be described
by time correlation functions of the type 〈A(t)B(0)〉, where A and B are operators, or by the Fourier
transforms of these correlation functions that characterize the fluctuation spectrum. As we have discussed
previously, the linear response to a driving force, such as the electrical conductivity to an applied electric
field and the magnetic susceptibility to an applied magnetic field, is generally given by a function of the type
of the retarded Green’s function if we replace A and B by the field operators:
G R ( t ) = −i θ ( t ) ⎡⎣ A ( t ) , B ( 0 ) ⎤⎦ . (IV.55)
{ }
A(t ) B (0) = Z −1Tr e − β H eiH t Ae − iH t B = Z −1 ∑ n e − β H eiH t Ae − iH t B n ,
n
(IV.56)
If the Fourier transform of 〈A(t)B(0)〉 is denoted by J1(ω), which is also known as the spectral density
function associated with the time correlation function 〈A(t)B(0)〉, it can be shown (see Problem Set 4) that the
Fourier transform of the equivalent retarded Green’s function at T = β −1 has the following form:
dω ′ J1(ω ′ )
( ) (η = 0 ) .
∞
G R (ω ) = ∫ 1 − e − βω ′ , +
(IV.57)
−∞ 2π ω − ω ′ + iη
We note that for A and B being hermitian conjugates at different times, J1(ω) is real so that EQ. (IV.57) gives
2
Im ⎡⎣G R (ω ) ⎤⎦ = −
1
2
(1 − e− βω ) J (ω ) ,
1 ⇒ J1(ω ≠ 0 ) = −
1 − e − βω
Im ⎡⎣G R (ω ) ⎤⎦ , (IV.58)
which is a form of the fluctuation-dissipation theorem. In other words, the spectral response of the system
(i.e. fluctuations) to an external perturbation at t > 0 gives rise to dissipation that is manifested by the
imaginary part of the retarded Green’s function.
spin spin
= + + +…
down up
Having studied interacting electrons, we want to understand how electrons in a solid interact with the
background ions characterized by their quantized modes, the phonons. Diagrammatically, we can consider
the electron-phonon interaction as a vertex contribution to electrons. This interaction can lead to scattering of
electrons, which gives rise to electrical resistivity. On the other hand, under special circumstances, the
electron-phonon interaction can also lead to an effective attractive potential for electrons, thus giving rise to
Cooper pairing and conventional superconductivity.
Further Readings:
1. Fetter and Walecka, “Quantum Theory of Many-Particle Systems”, Sections 31 – 34, 44 – 47.
2. Abrikosov, Gorkov, and Dzyaloshinski, “Methods of Quantum Field Theory in Statistical Physics”,
Chapters 2 and 3.
3. Doniach and Sondheimer, “Green’s Functions for Solid State Physicists”, Chapters 3, 5 and 6.
⎛ ∂P ⎞
B ≡ −Ω ⎜ ⎟ , (V.1)
⎝ ∂Ω ⎠ S
where P and Ω denote the pressure and volume, respectively, and the subscript S refers to constant entropy.
In this sense, the ionic background is treated as if it were a uniform fluid.
Based on the aforementioned simplifications, we are ready to provide a complete description of the
longitudinal phonons. We first introduce the displacement vector d(x) of the medium that characterizes the
displacement of each point from its equilibrium position. The change in volume under deformation to lowest
order can be expressed in terms of the displacement vector as follows:
⎛ ∂d x ∂d y ∂d z ⎞
d Ω′ = dx′dy ′dz ′ = dx dy dz ⎜ 1 +
∂x
+
∂y
+
∂z
+ ⎟ ≈ d Ω (1 + ∇ i d ) . (V.2)
⎝ ⎠
Therefore, the change in density becomes
δρ δ n
= = −∇ i d , (V.3)
ρ 0 n0
where ρ 0 (n0) denotes the equilibrium mass density (particle density). Moreover, in an elastic medium free of
shear strength and vorticity, the displacement field satisfies the general condition
∫ any path
d i dl = 0 ⇔ ∇ ×d = 0. (V.4)
1 ⎡ ⎛ ∂d i ⎞⎛ ∂d i ⎞ ⎛ ∂d i ⎞⎛ ∂d i ⎞ ⎤
L= ∫d x ⎢ ρ0 ⎜
3
⎟⎜ ⎟ − B⎜ ⎟⎜ ⎟⎥ . (V.5)
2 ⎣⎢ ⎝ ∂t ⎠⎝ ∂t ⎠ ⎝ ∂x j ⎠⎝ ∂x j ⎠ ⎦⎥
From the Euler-Lagrangian equations and EQ. (V.5) we obtain the equation of motion
1 ∂ 2d
− ∇ 2d = 0 , (V.6)
u02 ∂t 2
which is a wave equation and u0 is the sound velocity given by u0 ≡ (B/ρ0)1/2. Also, the corresponding
Hamiltonian is
1 ∂d
x ⎡ ρ 0−1 π 2 + B ( ∇ i d ) ⎤ ,
2
where π ( x, t ) = ρ 0
2∫
H = d 3
. (V.7)
⎣ ⎦ ∂t
Introducing the normal mode expansions that satisfies the condition ∇ × d = 0 , we have
1/ 2
⎛ 1 ⎞ k
d ( x, t ) = −i ∑ ⎜ ⎟ ( bk ei k i x −i ωt − bk† e − i k i x +i ωt ) , (V.8)
k ⎝ 2 ρ 0ωk Ω ⎠ k
1/ 2
⎛ρω ⎞ k
π ( x, t ) = − ∑ ⎜ 0 k ⎟ ( bk ei k i x−iωt + bk†e−i k i x+iωt ) , (V.9)
k ⎝ 2 Ω ⎠ k
⎛ 1⎞
H = ∑ ωk ⎜ bk†bk + ⎟ , (V.11)
k ⎝ 2⎠
which represents a system of uncoupled harmonic oscillators. The derivation leading to EQ. (V.11) is
analogous to what you have seen in Part I for the quantization of electromagnetic fields in free space except
that the sound velocity in the case of phonons is determined by the material properties of the system under
consideration.
The Hamiltonian given in EQ. (V.11) provides the basis for investigating the thermodynamics and
statistical mechanics of the free phonon system. For instance, the total energy E of the system is given by
⎛ 1⎞ ⎛ 1 1⎞
E = ∑ ⎜ nk + ⎟ ωk = ∑ ⎜ + ⎟ ωk , (V.12)
k ⎝ 2⎠ k ⎝ exp ( βωk ) − 1 2⎠
where β = T−1. The momentum sum can be converted into energy integration by the following relation:
2
Ω Ω Ω ⎛ω⎞ ⎛ω⎞ Ω
∑⇒ ∫ d ⎜⎝ u ⎟⎠ 4π ⎜⎝ u ⎟⎠ = 2π 2u 3 ∫ dω ω ≡ ∫ dω g (ω ) ,
( 2π ) ∫ ( 2π ) ∫
d 3
k= dk 4π k 2
= 2
( 2π )
3 3 3
k 0 0 0
Ω
⇒ g (ω ) = ω 2 and ω = ωk = u0 k . (V.13)
2π u0
2 3
In a uniform medium there is no upper limit for the frequency. On the other hand, for a real crystal the
wavenumber of propagation cannot exceed the reciprocal lattice constant. Therefore, we may define the
upper bound for the phonon frequency as ωD, the Debye frequency, which satisfies the following relation
ωD ωD ω 2Ω ωD3 Ω 9 Nω 2
3N = ∫ g (ω ) dω = ∫ dω = ⇒ g (ω ) dω = dω , (V.14)
0 0
2π 2 u03 6π 2 u0 3 ωD3
because the total number of degrees of freedom in a crystal of N ions is 3N. We may further define the Debye
temperature by Θ D ≡ ( ωD k B ) if we restore and kB, and apply EQ. (V.13) to EQ. (V.12), so that the
energy associated with phonons becomes
3
⎛ T ⎞ ΘD T ⎛ u 3 ⎞ 9 Nk B Θ D
E = 9 Nk BT ⎜ ⎟
⎝ ΘD ⎠
∫ 0
du ⎜ ⎟+
⎝ e −1 ⎠
u
8
, (V.15)
where u ≡ ( ω k BT ) is a dimensionless variable, and we note that the chemical potential for the phonon
system is zero. Thus, the specific heat is given by
3
∂E ⎛ T ⎞ ΘD T u 4 eu
Cv = = 9 Nk B ⎜ ⎟ ∫ du . (V.16)
( eu − 1)
2
∂T ⎝ ΘD ⎠ 0
Equation (V.16) is the Debye theory of the specific heat, which has two limiting values:
3 3
⎛ T ⎞ ∞ u 4 eu 12π 4 ⎛ T ⎞
Cv = 9 Nk B ⎜ ⎟ ∫ du = Nk B ⎜ ⎟ , T → 0; (V.17)
( )
2
⎝ ΘD ⎠ 0
eu − 1 5 ⎝ ΘD ⎠
Cv = 3 Nk B , T → ∞. (V.18)
3
The low-temperature limit of the specific heat is known as the Debye T -law, and the high-temperature limit
is consistent with the classical description of equipartition of energy. The Debye theory of the specific heat
provides an excellent one-parameter (the Debye temperature) description for the specific heat of metals, and
therefore one can obtain the values of the Debye temperature for different metals by fitting their specific
data.
Having discussed the basic properties of non-interacting phonons, we want to develop field-theory
description for the phonon propagator and electron-phonon interaction. We first define the phonon field
operator by the following expression:
1/ 2
⎛ω ⎞
ϕ( x) = i ∑ ⎜ k ⎟ (b e
k
i k i x − i ωk t
)
− bk† e −i k i x +i ωk t . (V.19)
k ⎝ 2Ω ⎠
Restricting the above expression to longitudinal phonons in the Debye model so that the sum over k is
limited to |k| < kD = ωD/u0, the Debye wavelength, and noting that the phonon fields are real, we define the
phonon propagator D ( x, x′ ) as:
D ( x, x′ ) ≡ −i T [ϕ ( x ) ϕ ( x′ )] . (V.20)
Substituting the free-field operator EQ. (V.19) into EQ. (V.20) and noting that there are no phonons in the
ground state, we obtain the free phonon propagator:
⎛ ωk ⎞ i k i x −iωk t
i
D (0)( x ) = − ∑⎜
Ω ⎝
⎟e
2 ⎠
θ ( ω D − ωk ) for t > 0,
k
i ⎛ ω ⎞ −i k i x +iωk t
= − ∑⎜ k ⎟ e θ ( ω D − ωk ) for t < 0. (V.21)
Ω k ⎝ 2 ⎠
ωk ⎡ 1 1 ⎤ ωk2
D (0)( k , ω ) = ⎢ − ⎥ θ ( ω − ω ) = θ ( ω D − ωk ) . (V.22)
2 ⎣ ω − ωk + iδ ω + ωk − iδ ⎦ ω 2 − ωk2 + iδ
D k
Diagrammatically, the free phonon propagator, or equivalently the phonon Green function, is expressed by a
thin dashed line. Similar to the fermion Green functions, we may also define the retarded and advanced
phonon propagators in the coordinate representation as follows:
DR ( x, x′ ) ≡ −iθ ( t − t ′ ) [ϕ ( x ) , ϕ ( x′ ) ] , (V.23)
DA( x, x′ ) ≡ iθ ( t ′ − t ) [ϕ ( x ) , ϕ ( x′ )] . (V.24)
where δρ = − zen0∇ i d according to EQ. (V.3), and z denotes the valence of the ions in the crystal. Using
EQ. (V.7) and expressing the electron charge density in terms of fermion field operators
1
ψ (x) = ∑ ei k i x ηλ akλ , (V.26)
kλ Ω1/ 2
1/ 2
⎛ ωk ⎞
⎟ U 0( q ) θ ( ωD − ωq ) ( ak +q ,λ ak ,λ bq + ak ,λ ak +q ,λ bq ) ,
zn0
= ∑∑ ⎜
u0 kλ q ⎝ 2 ρ 0 Ω ⎠
† † †
(V.27)
where U0(q) denotes the bare Coulomb potential. However, our previous analysis of the degenerate electron
gas reveals that the bare Coulomb interaction becomes modified by summing over the ring diagrams due to
many-body interaction in the electron gas. Hence, we may replace the bare Coulomb interaction potential
U0(q) by the effective static Coulomb interaction in the Thomas-Fermi approximation, Ur(q), which yields
4π e 2 4π e 2 4k F
U r(q ) = ≈ because q 2
k F and qTF ≡ . (V.28)
2
q +q 2
TF
q 2
TF
π a0
u0 ( ρ 0 )
1/ 2 2
q TF
≡ γ ∫ d 3 x ψ α† ( x )ψ α ( x ) ϕ ( x ) , (V.29)
where
4π ze 2 n0 zπ 2 n0
γ ≡ = (V.30)
u0 ( ρ 0 )
1/ 2 2
qTF mk F B1/ 2
is the electron-phonon coupling constant. It is worth noting that γ is dependent on the electron mass and the
bulk modulus, but is independent of the ion mass.
Diagrammatically, the effect of electron-phonon interaction may be treated in a way similar to the
results derived previously for Coulomb interactions among fermions if we simply replace the wavy lines for
Coulomb interactions with dashed lines for the phonon propagators. Therefore, each vertex now represents
the occurrence of electron-phonon interaction, as shown in Fig. V.2.1.
k+q
q
Fig. V.2.1 Illustration of the basic electron-phonon vertex, where solid lines represent the electron
propagators and the dashed line indicates the phonon propagator.
Specifically, the electron-electron interaction potential V(x1 − x2) mediated by the electron-phonon interaction
can be expressed by the following:
V ( x1 − x2 ) ⇒ γ 2 D (0) ( x1 − x2 ) , (V.32)
iD(0)(q)
−iγ −iγ
Fig. V.2.2 Illustration of the electron-electron interaction mediated by the electron-phonon interaction
with a coupling coefficient γ.
Given the electron-phonon coupling, we consider the correction to the non-interacting phonons. The
first non-vanishing correction occurs in the second order with respect to the interaction Hel-ph, as shown in
Fig. V.2.3 (a) – (b), and the corresponding expressions are
−γ 2i ∫ d 4 x1 d 4 x2 D (0) ( x − x1 ) Gαβ
(0)
( x1 − x2 ) Gβα( 0)( x2 − x1 ) D (0)( x2 − x′ ) for (a), (V.33)
However, the disconnected graph in Fig. V.2.3 in fact does not contribute to the electron-electron
interaction potential, which may be understood by the following simple consideration. If we inspect in EQ.
(V.34) the integration of D (0)( x − x1 ) over x1, we find that
D (0)( x − x1 ) ∼ T ⎡⎣ϕ ( x ) ∇ x1 i d( x1 ) ⎤⎦ ∼ ∇ x1 i T [ϕ ( x ) d( x1 )] ,
so that
∫d
3
x1 D (0)( x − x1 ) ∼ ∫ d 3 x1 ∇ x1 i T [ϕ ( x ) d( x1 )] ∼ ∫d
2
x1 nˆ i T [ϕ ( x ) d( x1 )] → 0 .
(a) (b)
Fig. V.2.3 Second-order corrections to the phonon propagator: (a) connected polarization correction;
(b) disconnected correction, which can be shown to vanish identically.
We summarize in the following the general rules used to calculate the corrections of order 2n (with
respect to Hel-ph) to the electron and phonon Green functions:
3) Integrate over the coordinates of all vertices and sum over the corresponding spin variables;
4) Multiply the resulting expression by γ 2 n i n ( −1) , where F is the number of closed loops formed by the
F
fermion G(0)-lines.
As an example, consider the diagram in Fig. V.2.4. Using the rules given above, we obtain the
corresponding expression for the diagram as follows:
γ 4 ∫ d 4 x1 d 4 x2 d 4 x3 d 4 x4 D (0) ( x − x1 ) Gγ(0)
1γ 2
( x1 − x2 ) D (0)( x2 − x3 ) Gγ(0)2γ 4 ( x2 − x4 )
× Gγ( 40)γ 3 ( x4 − x3 ) Gγ(0)
3γ 1
( x3 − x1 ) D ( 0)( x4 − x′ ) . (V.35)
ωq2
V ( q, ω ) = γ 2
θ (ωD − ωq ) . (V.36)
ω 2 − ωq2 + iδ
V ( x) = − γ 2 δ ( x) . (V.37)
On the other hand, for energy transfer greater than all of the relevant phonon modes, ω > ωD, we find from
EQ. (V.36) that
V ( q, ω > ω D ) > 0 , (V.38)
implying that the interaction is always repulsive. Therefore, the equivalent interaction potential between two
electrons can only be attractive if both electrons lie within an energy shell ωD below the Fermi surface and
become excited to unoccupied states within an energy shell ωD above the Fermi surface. (N.B.! for most
metals ωD ~ 102 meV and the Fermi energy εF ~ 1 eV so that ωD << εF). This attractive interaction between
particles near the Fermi surface as a consequence of electron-phonon interaction has important implications
on the occurrence of superconductivity in metallic systems.
Finally, we summarize the general rules used to calculate the corrections of order 2n (with respect to
Hel-ph) to the electron and phonon Green functions in momentum space:
1
G (0)( k ) = lim ,
δ →0 + ω − ( ε k − μ ) + iδ sgn ( ε k − μ )
4) Multiply the resulting expression by γ 2 n ( 2π ) i n ( −1) , where F is the number of closed loops formed
−4 n F
( )
−1
G = G ( 0) + G (0) Σ*G ⇔ G −1 = G (0) − Σ* . (V.39)
The simplest diagram for the self-energy correction due to electron-phonon coupling is given by the diagram
in Fig. V.3.1 (a), which is the skeleton diagram for the generalized diagram of Σ* expressed by the graph in
Fig. V.3.1 (b). The shaded area on the right of the graph in Fig. V.3.1 (b) represents all diagrams with three
external points, one associated with a phonon and two with electrons. This shaded triangle is the vertex part,
and is denoted by Γ( k , k − q; q ) . Thus, the proper self-energy can be explicitly given by the vertex part and
the electron and phonon Green functions:
d 4q
Σ *( k ) = i ∫ G ( k − q ) D ( q ) Γ( k − q , k ; q ) , (V.40)
( 2π )
4
(a) (b)
Fig. V.3.1 Corrections to the proper self-energy of the electron Green function due to electron-phonon
coupling: (a) the skeleton diagram for the lowest order correction; (b) the generalized
diagram consisting of the vertex part.
Substituting EQ. (V.40) into EQ. (V.39), we obtain the Dyson’s equation for the electron Green
function:
d 4q
[ ( k )] ( ) ∫
ω − ε − μ G k − i G ( k − q ) D ( q ) Γ( k − q , k ; q ) G ( k ) = 1 . (V.41)
( 2π )
4
Similarly, the proper self-energy part of the phonon propagator due to electron-phonon coupling is
associated with the polarization insertion, and is denoted by Π*. The corresponding skeleton diagram and the
generalized diagram containing the vertex part are shown in Fig. V.3.2 (a) – (b), and the Dyson’s equation
for the phonon Green function is given by
( )
−1
D = D (0) + D (0) Π * D ⇔ D −1 = D (0) − Π* . (V.42)
d 4q
Π *( k ) = −2i ∫ G ( q ) G ( q − k ) Γ( q , q − k ; k ) , (V.43)
( 2π )
4
where the factor 2 takes care of the spin degeneracies of electrons. Inserting EQ. (V.43) into EQ. (V.42), we
obtain an explicit form for the Dyson’s equation of phonons:
d 4q
ωk−2 ( ω 2 − ωk2 ) D( k ) + 2i ∫ G ( q ) G ( q − k ) Γ( q , q − k ; k ) D ( k ) = 1 . (V.44)
( 2π )
4
(a) (b)
Fig. V.3.2 Corrections to the proper self-energy of the phonon Green function due to electron-phonon
coupling: (a) the skeleton diagram for the lowest order correction; (b) the generalized
diagram consisting of the vertex part.
In general, Dyson’s equation can be obtained directly from the equations of motion for the
Heisenberg operators of fermions, which you will prove in Problem Set 5. The final result for a fermion
system under an interaction Hamiltonian Hint is:
⎡ ∂ ∇ 2x ⎤
⎢i ∂t + 2m + μ ⎥ Gαβ ( x, x′ ) = δ ( x − x′ ) δαβ −i 0 T ⎡⎣ψ H α ( x ) , H int ⎤⎦ψ H β ( x′ ) 0 .
†
(V.45)
⎣ ⎦
Taking Hint = Hel-ph as given in EQ. (V.29), we find that the last term in EQ. (V.45) becomes
with a set of Feynman diagrams with one external phonon line and two external electron lines, The simplest
of these diagrams is shown in Fig. V.3.3 (a), which corresponds to the following quantity
The generalized diagram for EQ. (V.47) is shown in Fig. V.3.3 (b), which corresponds to the following
expression:
Gαβ ( x1 , x2 ; x3 ) = δαβ G ( x1 , x2 ; x3 )
= − δ αβ ∫ d 4 x1′ d 4 x2′ d 4 x3′ G ( x1 − x1′ ) G ( x2′ − x2 ) D( x3′ − x3 ) Γ( x1′ , x2′ ; x3′ ) . (V.49)
Γ( k , k − q; q )( 2π ) δ ( k − k ′ − q ) = ∫ d 4 x1 d 4 x2 d 4 x3 Γ( x1 , x2 ; x3 ) e
4 − ikx1 + ik ′x2 + iqx3
. (V.50)
G ( k , k − q ; q ) = −G ( k ) G ( k − q ) D ( q ) Γ ( k , k − q ; q ) . (V.51)
(a) (b)
Fig. V.3.3 The Feynman diagrams associated with electron phonon interaction: (a) the first-order
perturbation theory approximation; (b) the generalized expression containing all diagrams
associated with the vertex part.
To calculate the vertex part Γ under electron-phonon interaction, we need to find all compact
diagrams and associate the corresponding analytical expressions with them according to the rules prescribed
earlier. Keeping in mind that the electron-phonon vertex part consists of three external points associated with
one phonon and two electrons, we find that the diagrams consist of contributions from terms of order γ, γ 3,
γ 5, etc. Some representative examples of such contributions up to order γ 7 are illustrated in Fig. V.3.4, and a
complete set of contributions up to order γ 5 are given in Fig. V.3.5.
Fig. V.3.4 Representative vertex contributions due to the electron-phonon interaction: (a) to order γ 3;
(b) to order γ 5; and (c) to order γ 7.
The vertex contributions exemplified in Figs.V.3.4 and V.3.5 seem tedious and difficult to compute exactly.
Interestingly, however, the exact proper vertex can be greatly simplified by a remarkable theorem known as
the Migdal theorem, which states that the exact vertex in the electron-phonon system satisfies a simple
relation:
⎧⎪ ⎡⎛ m ⎞1/ 2 ⎤ ⎫⎪
Γ = γ ⎨1 + O ⎢⎜ ⎟ ⎥⎬ , (V.52)
⎪⎩ ⎣ ⎝ M ⎠ ⎦ ⎭⎪
where (m/M) is the ratio of the electron mass to the ion mass. Therefore, EQ. (V.52) allows us to replace the
vertex by the point value γ to an excellent approximation. Here we only mention this important result without
going through the proof of the theorem. For details of the proof, you may consult the original paper by A. B.
Migdal, Sov. Phys.-JETP 7, 996 (1958). In addition, you may refer to Section 47 of the book by Fetter &
Walecka for proof to the second-order vertex correction.
Γ = + + + +
+ + + +
+ + + O(γ 7)
The aforementioned discussion suggests that attractive electron-phonon interaction can always occur
in metals, which in turn implies (according to the BCS theory, a topic of our later discussion) that
superconductivity will always occur at sufficiently low temperatures, i.e. for T < Tc where Tc denotes a
critical temperature. On the other hand, in the “normal state” of the metal where Tc << T << ωD, our
derivation of the electron and phonon Green functions must be modified to incorporate the finite temperature
effect. In the following section we discuss the finite-temperature Green functions for electron-phonon
interaction.
To deal with problems at finite temperatures, we need to introduce temperature Green function. As
discussed in Part II.2, we may transcribe the Green functions using the “imaginary time” trick so that we
write i t ⇒ τ , and τ varies from 0 to T −1. The temperature Green function is therefore defined as
{(
gαβ ( x1 , τ 1 ; x 2 , τ 2 ) = −Tr e
ΩG + μ N − H ) / T
Tτ ⎡⎣ψ H α ( x1 ,τ 1 )ψ H β ( x 2 ,τ 2 ) ⎤⎦ , }
= − Tτ ⎡⎣ψ H α ( x1 , τ 1 )ψ H β ( x 2 , τ 2 ) ⎤⎦ , (V.53)
where ΩG is the grand potential, Tτ is the imaginary time ordering operator, and the Heisenberg field
operators ψ H and ψ H for electrons and ϕ H for phonons at T ≠ 0 are given by:
(H − μ N ) τ − (H − μ N ) τ
ψ H α ( x, τ ) = e ψα(x) e , (V.54)
( H − μ N )τ − (H − μ N ) τ
ψ H α ( x, τ ) = e ψ α† ( x ) e , (V.55)
ϕ H ( x, τ ) = e Hτ
ϕ(x) e −H τ
. (V.56)
We note that the field operators ψ H and ψ H are no longer Hermitian conjugates of each other. Moreover,
given that the temperature Green function is a function of the imaginary time difference τ 1 − τ 2 = τ , it is
defined in the interval from –1/T to 1/T. On the other hand, from EQ. (V.53) we have
⎛ 1 1⎞
g (τ < 0 ) = − g ⎜ 0 < τ + < ⎟ for fermions, (V.57)
⎝ T T⎠
⎛ 1 1⎞
= g⎜0 <τ + < ⎟ for bosons, (V.58)
⎝ T T⎠
⎛ 1⎞
D (τ < 0 ) = D ⎜ τ + ⎟. (V.59)
⎝ T⎠
In addition, noting that the phonon temperature Green function is real, we have
Next, we consider the temperature Green functions for free particles, which play an important role in
the perturbation theory. In the case of free particles, the total Hamiltonian H is equal to the non-interacting
Hamiltonian H0, and the field operators in EQs. (V.54) and (V.56) are given by
1 1
ψ α ( x1 ) =
Ω
∑ ak α ei k i x ,
1
1 1
ψ β† ( x 2 ) =
Ω
∑ ak† β e − i k i x .
2
2 2
(V.61)
k1 k2
Consequently, we obtain
Tr {e( }
1 ΩG 0 + μ N − H 0 ) / T (H 0 − μ N ) τ
ak1α e ( 0 ) ak†2 β
− H −μ N τ
(0)
gαβ ( x1 , x 2 ;τ > 0 ) = −
Ω
∑ e
i ( k i x −k i x )
1 1 2 2
e
k1 , k 2
=−
1
Ω
∑e
k1 , k 2
i ( k1 i x1 −k 2 i x 2 ) − ⎣⎡ε 0( k1 ) − μ ⎦⎤τ
Tr e( { ΩG 0 + μ N − H 0 ) / T
ak1α ak†2 β }
1 i k i ( x1 − x 2 ) − ⎡⎣ε 0( k ) − μ ⎤⎦τ
⇒ (0)
gαβ ( x1 − x 2 ;τ > 0 ) = −δαβ
Ω
∑e akα ak†α . (V.62)
k
The quantity akα ak†α can be expressed in terms of the equilibrium occupation number n(k) at a finite
temperature T:
{ }
−1
⎡ε 0( k ) − μ ⎤⎦ / T
akα ak†α = 1 − n( k ) , n( k ) = e ⎣ +1 for fermions,
n( k ) = {e − 1}
−1
⎡⎣ε 0( k ) − μ ⎤⎦ / T
akα ak†α = 1 + n( k ) , for bosons. (V.63)
Taking the volume Ω to infinity, we may rewrite the free-particle temperature Green function into:
1
( x;τ > 0 ) = −δαβ i k i x − ⎡⎣ε 0( k ) − μ ⎤⎦ τ
[1 ∓ n( k )] ,
( 2π ) ∫
(0) 3
gαβ d ke
3
(V.64)
⎛ 1⎞
(0)
gαβ ( x;τ < 0 ) = ∓ gαβ(0) ⎜ x;τ + ⎟
⎝ T⎠
1 i k i x − ⎡⎣ε 0( k ) − μ ⎤⎦ τ
n( k ) .
( 2π ) ∫
= ± δαβ d ke 3
3
(V.65)
Similarly, the phonon Green function may be derived from the free-phonon field operator
1/ 2
⎛ ωq ⎞
ϕ(x) = ∑ ⎜ ⎟
Ω q ⎝ 2 ⎠
i
(b eq
iqix
− bq†e− i q i x ) , (V.66)
so that
∫ d q {[ N ( q ) + 1] e },
1 i q i x −ω0( q ) τ i q i x +ω0( q ) τ
D (0) ( x,τ ) = − 3
+ N (q) e (V.67)
2 ( 2π )
3
where
{ }
−1
N ( q ) = e 0( ) − 1
ω q /T
.
Diagrammatically, we may apply similar technique of Feynman rules for T = 0 to the particle and
phonon temperature Green functions. In coordinate space, the following formulae are satisfied:
Tτ ⎡⎣ψ α ( x1 ,τ 1 )ψ β ( x 2 , τ 2 ) ⎤⎦ = − gαβ
(0)
( x1− x 2 ;τ 1− τ 2 ) , (V.68)
Tτ ⎡⎣ψ β ( x 2 , τ 2 )ψ α ( x1 ,τ 1 ) ⎤⎦ = ± gαβ
(0)
( x1− x 2 ;τ 1− τ 2 ) , (+: fermions, −: bosons) (V.69)
Tτ [ϕ ( x1 ,τ 1 ) ϕ ( x 2 , τ 2 )] = −D (0) ( x1− x 2 ;τ 1− τ 2 ) . (V.70)
we can calculate the second-order correction to the phonon Green function in ways similar to the case for T =
0, and we find
γ 2 ∫ d 4 z1 d 4 z2 D (0) ( x − z1 ) gαβ
(0)
( z1 − z2 ) g βα(0)( z2 − z1 ) D (0)( z2 − x′ ) , (V.72)
which is similar to the result in EQ. (V.33) for the diagram shown in Fig. V.2.3 (a).
The general rules for calculating the corrections of order 2n (with respect to Hel-ph) to the electron
and phonon Green functions at T > 0 and in coordinate space may be summarized as follows:
3) Integrate over the coordinates of all vertices with respect to both x and τ, and sum over the corresponding
spin variables;
In reality, the aforementioned diagrammatic techniques in coordinate space are not practical for
calculating Green function corrections at finite temperatures, because the imaginary time varies from 0 to 1/T
rather than to infinity. The situation can be much simplified by expanding all quantities depending on τ in
Fourier series relative to the imaginary time difference τ. Noting that τ is defined in an interval between −1/T
and 1/T, we obtain
g (τ ) = T ∑ e g ( ωn ) ,
− iωnτ
(V.73)
n
1 1/ T
g ( ωn ) = g (τ ),
2∫
iωnτ
and dτ e ωn = nπ T . (V.74)
−1/ T
Using EQs. (V.57) and (V.58), we may rewrite EQ. (V.74) into the following:
1 1/ T 1 0
g ( ωn ) = g (τ ) + g (τ )
2∫ 2∫
iωnτ iωnτ
dτ e dτ e
0 −1/ T
1 1/ T 1 1/ T ⎛ 1⎞
∫ g (τ ) + ∫
iωnτ iωnτ iωn / T
= dτ e dτ e e g ⎜τ + ⎟
2 0
2 0
⎝ T⎠
1
( )∫
1/ T
g (τ )
ω /T iωnτ
= 1∓ e dτ e
i n
2 0
ωn = ( 2n + 1) π T for fermions,
1/ T
=∫ g (τ ),
iωnτ
dτ e (V.75)
0
1/ T
=∫ g (τ ),
iωnτ
dτ e ωn = 2nπ T for bosons. (V.76)
0
(0)
gαβ ( k;τ > 0 ) = −δαβ e − ⎡⎣ε 0( k ) − μ ⎤⎦ τ
[1 − n( k )] , (V.77)
so that
1/ T − ⎡⎣ε 0( k ) − μ ⎤⎦ τ
(0)
gαβ ( k; ωn ) = −δαβ [1 − n( k )] ∫ 0
dτ e
iωnτ
e ,
1
= , ωn = ( 2n + 1) π T for fermions, (V.78)
iωn − ε 0( k ) + μ
1
= , ωn = 2nπ T for bosons. (V.79)
iωn − ε 0( k ) + μ
ωk2
D (0) ( k ; ωn ) = − , ωn = 2nπ T . (V.80)
ωn2 + ωk2
In general, there are even numbers of fermion lines at each vertex, so that the integration over τ
involves the following integral
1
dτ e ∑ n = ∑ω
1/ T iτ ω
∫0
T
for n
= 0. (V.81)
=0 for ∑ ω n
≠ 0.
Consequently, the condition for summing over imaginary time in temperature Green functions is similar to
that for conservation of energy at a vertex in the case of Green functions at T = 0.
(1)
δ gαβ ( x − x′ ) = − ∫ d 4 z1 d 4 z2 gαγ(0)1 ( x − z1 ) gγ(0)1γ 2 ( z1 − z2 ) V ( z1 − z2 ) gγ(0)2 β ( z2 − x′ ) . (V.82)
1 1/ T − i k i ( x − x′) + iω (τ −τ )
δ g (1)( k , ωn ) = ∫ d ( x − x′ ) ∫ d (τ x − τ x′ ) δ g (1)( x − x′ ) e n x x′
, (V.83)
2 −1/ T
T
V ( x, τ ) = ∑ ∫ d q eiq i x−iω
3 n 4τ
V ( q, ωn 4 ) . (V.84)
( 2π )
3
ωn 4
Noting that
∞
T ∑ ei 2 nπ Tτ = δ (τ ) , (V.85)
n =−∞
we have
δαβ T V ( k − k 1 , ω n − ω n1 )
(1)
δ gαβ ( k , ωn ) = − ∑∫d k 3
iωn1 − ε 0( k 1 ) + μ
, (V.87)
2
( 2π )
1
⎡⎣iωn − ε 0( k ) + μ ⎤⎦
3
ω n1
and the diagrammatic representation for EQ. (V.87) is shown in Fig. V.4.1 (b).
Similarly, the correction to the temperature Green function shown in Fig. V.4.1 (c) is given by
i ωn1τ
δαβ T e
± 2 (
V 0, 0 )( 2 s + 1) ∑ ∫ d 3k 1 iωn1 − ε 0( k 1 ) + μ
, (V.88)
⎡⎣iωn − ε 0( k ) + μ ⎤⎦ ( 2π )
3
ωn1
where the plus (negative) sign corresponds to the correction to fermions (bosons).
As another example, the generalized two-particle interaction that involves the vertex contribution as
illustrated in Fig. V.4.1 (d) is given by:
1 T 1
− ∑∫d k 3
Γ αγ(0),γβ ( k , ωn , k 1 , ωn1 ; k 1 , ωn1 , k , ωn )
iωn1 − ε 0( k 1 ) + μ
, (V.89)
( 2π )
2 1
⎡⎣iωn − ε 0( k ) + μ ⎤⎦
3
ωn1
x z1 z2 x′ (k; ωn) (k1; ωn1) (k; ωn) (k; ωn) (k; ωn) (k; ωn) Γ (k; ωn)
Figure V.4.1 (a) Temperature Green function correction associated with two-particle interaction to
first order in coordinate space; (b) similar diagram to (a) in momentum space; (c) another
first-order diagram in momentum space; (d) generalized vertex correction to the temperature
Green function.
Next, we consider the case for electron-phonon interaction. As discussed before, only contributions
associated with even order of Hel-ph are non-zero. For an arbitrary diagram of order 2n in Hel-ph, there are
(2n+1) internal electron lines, n internal phonon lines, and 2n vertices. Hence, there are n independent
integrations. The general rules for calculating a diagram of order 2n relative to Hel-ph are summarized below:
1
g (0)( k ) =
iωn − ε 0( k ) + μ
and with two solid external lines associate a quantity
δαβ
2
.
⎡⎣iωn − ε 0( k ) + μ ⎤⎦
( 2π ) 3n
For example, we consider the second-order correction to the phonon propagator at T > 0, as shown in
Fig. V.4.2. Applying the general rules outlined above, we obtain
− ε 0 ( k ) + μ i ( ω n1 − ω n ) − ε 0 ( k − q ) + μ
. (V.90)
⎣ ω n + ωq ⎦ ( 2π )
3
ωn1 n1
(k; ωn1)
Figure V.4.2 Second-order correction to the phonon propagator at T > 0, in momentum space.
In general, the correction to the temperature Green function g can be obtained from the expression
for the correction to the Green function G at T = 0 by replacing the frequencies ω in G by iωn and changing
all integrals over ω to sums over ωn:
1
∫ dω ⇒ iT ∑ (V.91)
2π ωn
Finally, we note that the Dyson’s equations for the temperature Green functions of electrons and
phonons are similar to those for the Green functions at T = 0. The Dyson’s equations for corrections
associated with electron-phonon interactions are given below and also illustrated in Figs. V.4.3 (a) and (b)
for electrons and phonons, respectively:
γT
−1
gαβ ( k , ωn ) = [iωn − ε 0( k ) + μ ]δαβ + ∑∫d k 3
gαβ ( k1 , ωn1 ) D ( k 1 − k , ωn1 − ωn ) Γ ( k , k 1 ; ωn , ωn1 )
( 2π )
3 1
ωn1
(V.92)
γT
D −1( q, ωn ) = −ωq−2 ⎡⎣ωn2 + ωq2 ⎤⎦ − ∑∫d k 3
gαβ ( k1 , ωn1 ) g βα ( k 1 − q, ωn1 − ωn ) Γ ( k 1 , k 1 − q; ωn1 , ωn1 − ωn )
( 2π )
3 1
ωn1
(V.93)
(a)
= +
(b)
= +
Fig. V.4.3 Dyson’s equations for temperature Green functions of (a) electrons and (b) phonons.
Further Readings:
1. Fetter and Walecka, “Quantum Theory of Many-Particle Systems”, Sections 7 – 10, 13 – 15, 44 – 47.
2. Abrikosov, Gorkov, and Dzyaloshinski, “Methods of Quantum Field Theory in Statistical Physics”,
Chapters 2 and 3.
3. Doniach and Sondheimer, “Green Functions for Solid State Physicists”, Chapters 3, 5 and 6.
In Part VI we want to investigate basic concepts and important physical predictions of the Fermi liquid
theory. We first restrict to perturbative consideration of small momentum/energy transfers relative to the
Fermi momentum and Fermi energy. This approximation may be justified by noting that most interesting
electronic properties of metals are determined by electrons near the Fermi level, and large momentum/energy
transfers of electrons and holes near the Fermi level are generally quenched due to the Pauli exclusion
principle. Part III is structured as follows. In Part VI.1 we begin with an overview of the Fermi liquid theory,
including the basic assumptions and their general consequences. In Part VI.2 we develop rigorous formalism
for studying the vertex contribution in the case of small momentum transfer, followed by explicit predictions
of physical relations of Fermi liquid systems in Part VI.3. The Kondo effect, a celebrated phenomenon
associated with the Fermi liquid response to localized magnetic impurities, will be investigated in Part VII as
a special case study for many-body interactions in a Fermi liquid. It is further noted that the Kondo effect can
be extended beyond interactions of conduction electrons with single magnetic impurities to the Kondo lattice
problem encountered in heavy fermions, as well as to nanoelectronic systems such as the quantum dots and
carbon nanotubes. The limitation of Fermi liquid theory is discussed in Part VIII, followed by studies of an
exemplifying non-Fermi liquid system, the Luttinger liquid theory in one dimension.
To examine the excitations of interacting fermions in the context of the Fermi liquid theory, we first
establish the description for excitations in an ideal Fermi gas. At T = 0, the occupied states of an ideal Fermi
gas form a Fermi sphere of a radius kF in the momentum space, where kF is given by the volume density n ≡
(N/Ω) of fermions of spin ½:
2 ⎛4 3 ⎞
⎜ π k F ⎟ = n ⇒ k F = ( 3π n ) .
2 1/ 3
(VI.1)
( 2π )
3
⎝3 ⎠
In the excited state of the Fermi gas, the momentum distribution of fermions will be different from that in the
ground state, and the low-energy excitations consist of removing fermions from within the Fermi sphere to
outside of the Fermi sphere. It is clear that each process involves creating a pair of “quasiparticles” with one
particle having momentum |k| > kF and one hole having momentum |k| < kF. Therefore, the energy of a
particle-like excitation is given by
2
k k F2 kF
ξk = − ≈ ( k − k F ) ≡ vF ( k − k F ) , (VI.2)
2m 2m m
Similar to the excitations of the Fermi gas, in Landau’s theory the excitations of the Fermi liquid also
occur in pairs, with particle-like excitations of momentum |k| > kF and hole-like excitations of momentum |k|
< kF. However, there are also important differences between the Fermi gas and Fermi liquid. That is, the
excitations of a Fermi liquid interact among themselves, and the most significant manifestation of such
interactions is in the case of superfluid Fermi liquids such as 3He. Specifically, in a Fermi gas the low energy
excitations are gapless and are expected to have zero total momentum. On the other hand, the existence of
superfluidity implies that there is a critical velocity for excitations, so that a finite total momentum for the
excitations can occur below a critical value without incurring energy dissipation, and there exists an energy
gap in the spectrum for excitations. Thus, certain interactions among the quasiparticles in the superfluid
Fermi liquid can lead to a spectrum fundamentally different from that for a Fermi gas. For the time being, we
only concentrate on the properties of normal Fermi liquids.
To incorporate the interaction among quasiparticles in a Fermi liquid, Landau’s theory assumes that
the interaction can be described by a self-consistent field acting on one quasiparticle due to the presence of
all other quasiparticles. Therefore, the energy of the excitations is no longer the simple sum of the energies of
all quasiparticles. Rather, it becomes a functional of the quasiparticle distribution function. If the spin indices
are taken into account, the variation of the total energy per unit volume (E/Ω) under a varying density
distribution δnkσ is given by:
⎛E⎞ ⎪⎧ d 3 k ⎪⎫ ⎪⎧ d 3k ⎪⎫
3 ( k ,αβ k , βα ) ⎬
δ ⎜ ⎟ = ∑ αβ ⎨ ∫ ε δ n = Tr ⎨∫ 3 ( k
ε δ nk ) ⎬ , (VI.4)
⎝Ω⎠ ⎪⎩ ( 2π ) ⎪⎭ ⎪⎩ ( 2π ) ⎪⎭
where εk denotes the energy of the quasiparticles. For simplicity, we will drop the summation notation for
spin indices (such as α, β…) in the following expressions, with the understanding that repeated subscripts
indicate summations over the subscripts. We also note that the variations in nk is subjected to the following
condition
d 3k
∫ ( 2π )3 δ nk = 0 (VI.5)
because all excitations come in pairs, and the total number of particle-like excitations is equal to that of hole-
like excitations. Furthermore, we note that the definition of quasiparticle energy in EQ. (VI.4) implies that
their equilibrium distribution function is actually a Fermi distribution. This point can be proven by
considering the following expression for the entropy S:
⎛S⎞ ⎧⎪ d 3 k ⎫⎪
⎜ ⎟ = − Tr ⎨ ∫ 3⎣ k k⎡ n ln n + (1 − n ) ln (1 − n ) ⎤
k ⎦⎬
. (VI.6)
⎩⎪ ( 2π )
k
⎝Ω⎠ ⎭⎪
The conditions for maximizing the entropy requires that the total number and energy of the particles be
conserved, so that δ N = 0, δ E = 0 and Tδ (S/Ω) = δ (E/Ω). From EQs. (VI.4) and (VI.6), we find
ε k δ nk = −T [δ nk ln nk − δ nk ln (1 − nk )] ,
which, combined with the condition lim T →0 nk ≡ lim T →0 n ( ε k < μ ) = 1 where μ is the chemical potential,
leads to the distribution function
1
nk ≡ n ( ε k ) = , (VI.7)
exp [( ε k − μ ) T ] + 1
where εk is in fact a function of nk, so that EQ. (VI.7) is a complicated expression for nk.
Including the spin index σ and noting that the quasiparticle energy εkσ also depends on temperature
besides being a functional of nkσ, we may express εkσ by the formula:
⎧⎪ d 3 k ′ ⎫⎪
ε kσ = ε k(0)σ + Tr ⎨ ∫ f ( k , σ ; k ′, σ ′ ) δ nk ′σ ′ ⎬ , (VI.8)
⎪⎩ ( 2π )
3
⎪⎭
where ε k( 0)σ is the equilibrium energy of quasiparticles at T = 0, and f ( k , σ ; k ′, σ ′ ) is the Fermi liquid
function, which is a matrix depending on the momentum and spin operators of two interacting quasiparticles.
A more detailed expression for EQ. (VI.8) is to rewrite it as follows:
d 3k ′
ε k ,αβ = ε k ,αβ + ∫
(0)
fαβ ,γδ ( k , k ′ ) δ nk ′,δγ , (VI.9)
( 2π )
3
We shall show in Part VI.2 that the Fermi liquid function is associated with the forward scattering amplitude
of two quasiparticles and can be evaluated from the vertex contribution. We also note that in the absence of
magnetic fields, the equilibrium energy of quasiparticles at T = 0, ε k( 0) , can be expressed as
kF
ξk = ε k(0) − ε F ≈ ( k − kF ) ≡ v ( k − kF ) , (VI.10)
m*
where m* is the effective mass related to the Fermi liquid function, and v is the velocity of quasiparticles. To
see how m* is related to f ( k , σ ; k ′, σ ′ ) , we follow Landau’s argument by considering the momentum density
of the Fermi liquid, which is equal to the mass flow of quasiparticles:
d 3k d 3k d 3k
2∫ k nk = 2m ∫ v nk = 2m ∫ (∇ kε k ) n . (VI.11)
( 2π ) ( 2π ) ( 2π )
3 3 3 k
If we vary EQ. (VI.11) relative to nk and use EQ. (VI.8) for the variation of quasiparticle energy, we find by
explicitly expressing the spin indices the following relation:
d 3k ⎛ k ⎞ d 3k ⎧⎪ d 3 k d 3k ′ ⎫⎪
∫ ( 2π )3 ⎜⎝ m ⎟⎠αβ k ,βα ∫ ( 2π )3 ( k k ,αβ ) k ,βα ⎪⎨∫ ( 2π )3 ∫ ( 2π )3 ∇ k′ fαβ ,γδ ( k , k ′ ) nk′,δγ δ nk ,βα ⎪⎬
δ n = ∇ ε δ n +
⎩ ⎭
3 ⎧⎪ d k 3 3
d k′ ⎫⎪
=∫
d k
( 2π )
( )
∇ k ε k ,αβ δ nk , βα − ⎨ ∫ 3∫ 3 αβ ,γδ (
f ( )
k ′, k ) ∇ k ′ nk ′,δγ δ nk , βα ⎬
⎪⎩ ( 2π ) ( 2π )
3
⎭⎪
⎛k⎞ ⎧⎪ d 3 k ′ ⎫⎪
(
⇒ ⎜ ⎟ = ∇ k ε k ,αβ − ⎨ ∫ ) 3 αβ ,γδ (
f (
k ′, k ) ∇ k ′ nk ′,δγ ⎬ . ) (VI.12)
⎝ m ⎠αβ ⎪⎩ ( 2π ) ⎪⎭
The last line in EQ. (VI.12) is obtained because δn is arbitrary. Equation (VI.12) can be rewritten in terms of
traces, which yields
k ⎪⎧ d 3 k ′ ⎪⎫
⇒ = Trσ {( ∇ k ε kσ )} − Trσ Trσ ′ ⎨ ∫ f ( k , σ ; k ′, σ ′ ) ( ∇ k ′ nk ′σ ′ ) ⎬ . (VI.13)
⎪⎩ ( 2π )
3
m ⎭⎪
Moreover, at T = 0 the quasiparticle energy is approximately given by EQ. (VI.10), and we also have
k′
∇ k ′ nk ′ ≈ − δ ( k′ − kF ) for T → 0. (VI.14)
k′
Inserting EQs. (VI.10) and (VI.14) into EQ. (VI.13) and noting that f ( k , σ ; k ′, σ ′ ) depends only on the angle
θ between k and k ′ , we obtain for k = k ′ = k F the following relation for the effective mass:
1 1 ⎧⎪ 1 ⎫⎪
3 ∫ d Ω f ( n, θ , σ , σ ) cos θ ⎬ ,
= − k F Trσ Trσ ′ ⎨ ˆ ′ (VI.15)
⎪⎩ ( 2π )
*
m m ⎭⎪
Empirically the effective mass m* can be determined from measurements of the electronic specific
heat. Using EQ. (VI.4), the electronic specific heat (i.e. electronic heat capacity per volume) to the lowest
order in T becomes (after restoring the Boltzmann constant kB):
⎛ ∂ ( E Ω) ⎞ ⎪⎧ d 3 k ⎛ ∂ nk ⎞ ⎪⎫ 1 * 2 π2
C v ≡⎜ ⎟ = 2 ⎨∫ ε
3 k ⎜ ⎟ ⎬ ≈ m k k T = N ( ε F ) k B2T , (VI.16)
∂ ⎩⎪ ( 2π ) ⎝ ∂ ⎠ N ⎭⎪ 3 3
F B
⎝ T ⎠ N ,Ω T
where N (εF) denotes the density of states at the Fermi level. It is interesting to note that the electronic
specific heat is linearly proportional to T, whereas that of phonons is proportional to T3 at low temperatures.
To derive the result shown in EQ. (VI.16), we first rewrite the integration in EQ. (VI.4) over
momentum into integration over energy as follows:
d 3k
∫ ( 2π )3 ⇒ ∫ N (ε ) dε ,
and we further note that for any function of energy ψ (ε), the following relation holds:
∂n ∂n
d ε ψ ( ε ) n( ε ) = [ Ψ ( ε ) n( ε )]0 − ∫ d ε Ψ ( ε )
∞ ∞ ∞
∫ = − ∫ d ε Ψ(ε )
∞
, (VI.17)
0 0
∂ε 0
∂ε
where
ε
Ψ(ε ) ≡ ∫ ψ (ε ′) dε ′ ,
0
1
= Ψ ( μ ) + ( ε − μ ) Ψ ′( μ ) + (ε − μ ) Ψ ′′( μ ) + … .
2
(VI.18)
2
In EQ. (VI.17) we have used the fact that Ψ(0) = 0 and n (∞) = 0. Therefore, by inserting EQ. (VI.18) into
EQ. (VI.17), we obtain
∞ ⎛ ∂nε ⎞
∞ ∞ ⎛ ∂nε ⎞
∫ d ε ψ ( ε ) n( ε ) = Ψ ( μ ) ∫ d ε ⎜ − ⎟ + Ψ ′( μ ) ∫0 d ε ( ε − μ ) ⎜ − ⎟ +…
0 0
⎝ ∂ε ⎠ ⎝ ∂ε ⎠
∞
∂nΨ ∞
1 ∞ n ⎛ ∂nε ⎞ ⎛ ∂ n Ψ ⎞
≡ ∑ Fn ∑ ( )
∂ε n μ n = 0 n !∫0
≡ d ε ε − μ ⎜ − ⎟⎜ n ⎟
n =0 ⎝ ∂ε ⎠ ⎝ ∂ε ⎠ μ
(k T ) exp [( ε − μ ) k BT ] ⎛ ∂ n Ψ ⎞
n n
∞
∞ ⎛ε −μ ⎞⎛ε −μ ⎞
=∑ ∫
B
⎛ μ ⎞d ⎜ ⎟⎜ ⎟ 2 ⎜ n ⎟
n =0 n! −⎜ ⎟
⎝ k BT ⎠ ⎝ k BT ⎠ ⎝ k BT ⎠ (1 + exp [( ε − μ ) k BT ]) ⎝ ∂ε ⎠ μ
(k T )
n
∞
⎛ ∂nΨ ⎞ ∞ zn
≈∑ ⎜ ∂ε n ⎟ ∫−∞ ( because k BT μ)
B
dz
n =0 n! ⎝ ⎠μ (1 + e z )(1 + e− z )
∞
⎛ ∂ 2n Ψ ⎞
= ∑ 2c2 n ( k BT ) ⎜
2n
2n ⎟
, (VI.19)
n =0 ⎝ ∂ε ⎠ μ
where the coefficients c2n in the last line of EQ. (VI.19) can be evaluated as summable series. Specifically,
the term 2c2 is equal to π2/6. Consequently, we have
∞ μ π2 ⎡ ∂ψ ( ε ) ⎤
∫ d ε ψ ( ε ) n( ε ) = ∫ d ε ψ ( ε ) + ( k BT ) 2 ⎢ ⎥ +… . (VI.20)
0 0
6 ⎣ ∂ε ⎦ μ
Now if we define ψ (ε) = εN(ε) in EQ. (VI.20), we find the following expression for the energy:
π2 ∂
( k BT ) ⎡⎢ {N ( ε ) ε }⎤⎥
∞ μ
ε = ∫0 d ε N ( ε ) ε n( ε ) = ∫0 d ε N ( ε ) ε +
2
+… . (VI.21)
6 ⎣ ∂ε ⎦μ
Similarly, if we take ψ (ε) = N(ε), we may relate the Fermi energy εF to the chemical potential μ by the
following expression:
∞ εF μ π2 ⎡ ∂N ( ε ) ⎤
d ε N ( ε ) n( ε ) = ∫ d ε N ( ε ) = ∫ d ε N ( ε ) + ( k BT ) ⎢
2
∫
0 0 0
6 ⎣ ∂ε ⎥⎦ μ
+… , (VI.22)
so that
μ εF π2 ⎡ ∂N ( ε ) ⎤
∫ dε N (ε ) − ∫ dε N (ε ) ≈ ( μ − ε F ) N (ε F ) ≈ − ( k BT ) 2 ⎢ ⎥
0 0
6 ⎣ ∂ε ⎦ ε F
π2 ∂ π2 2 ⎡ 1 ∂N ( ε ) ⎤
( k BT ) ⎢⎡ lnN ( ε ) ⎥⎤ = ε F − ( k BT ) ⎢
2
⇒ μ ≈ εF − ⎥. (VI.23)
6 ⎣ ∂ε ⎦ε F 6 ⎢⎣ N ( ε F ) ∂ε εF ⎥
⎦
∂ε 2 2
⎛ d μ ⎞ π kB ⎡ ∂N ( ε ) ⎤
Cv = = N (μ) μ ⎜ ⎟ + T ⎢N ( ε ) + μ +…
∂T ⎝ dT ⎠ 3 ⎣ ∂ε ⎥⎦ μ
π 2 k B2 ⎡⎛ d μ ⎞ π 2 k B2T ∂N ( ε ) ⎤
= T N ( ε F ) + N ( μ ) μ ⎢⎜ ⎟+ ⎥ +…
3 ⎣⎝ dT ⎠ 3N ( ε ) ∂ε ⎦ μ
π 2 k B2T 1
≈ N (ε F ) = m* k F k B2T . (VI.24)
3 3
where we have used EQ. (VI.23) in the deriving the last line of EQ. (VI.24).
Similarly, we may consider the spin susceptibility of a Fermi liquid by the following consideration. In
the presence of an external magnetic field H, the quasiparticle energy is modified by the amount due to spin-
field coupling:
σ = − μ B σ i H + ∑ f ( k , σ ; k , σ ) δ nk ′σ ′ .
δε kσ = − μ B* σ i H ≡ δε k(1)σ + δε k(2) ′ ′ (VI.25)
k ′σ ′
Assuming that in the weak field limit the Fermi liquid function can be expressed in terms a spin-independent
part and a spin-dependent part:
f ( k , σ ; k ′, σ ′ ) = f 0 ( k , k ′ ) + ( σ i σ′ ) f1 ( k , k ′ ) . (VI.26)
δ nk ′σ ′
− μ B* σ i H = − μ B σ i H + ∑ f ( k , σ ; k ′, σ ′ ) δε k ′σ ′
k ′σ ′ δε k ′σ ′
δn
= − μ B σ i H + ∑ f ( k , σ ; k ′, σ ′ ) k ′σ ′ − μ B* σ′ i H ,
δε k ′σ ′
( )
k ′σ ′
2 δ nk ′ ( ε ′ )
⇒ μ B* = μ B + μ B* ∑ f1 ( k , k ′ ) , (VI.27)
3 k′ δε ′
because the term associated with f 0 vanishes after summing over σ ′. For T → 0, the derivative of the density
distribution function in EQ. (VI.27) can be replaced by −δ ( ε F − ε ) and the amplitude of k and k′ can be
replaced by kF. Therefore, we may write f1 ( k , k ′ ) = φ ( k F , θ ) with θ being the angle between k and k′, and
EQ. (VI.27) reduces to
μB 2
μ B* = , where s≡ N (ε F ) φ ( kF ,θ ) , (VI.28)
(1 + s ) 3
N ( ε F ) denotes the density of states at the Fermi level, and φ ( k F , θ ) represents averaging over the Fermi
surface. Finally, the magnetic susceptibility is given by
χ P0 μ B2 N ( ε F )
χP = ≡ , (VI.29)
(1 + s ) (1 + s )
where χ P0 is the Pauli spin susceptibility for free electron spins.
In general, our description of the Fermi liquid theory is based on the notion of well-defined
quasiparticles, so that not only the total number of quasiparticles is conserved, but also the number of
quasiparticles in a given momentum direction. Hence, a Fermi liquid has an infinite number of conserved
quantities, which implies that we can develop a hydrodynamic theory for a Fermi liquid and that a Fermi
liquid contains many bosonic modes, one mode for each conserved quantity. For hydrodynamic descriptions
of a Fermi liquid in two dimensions, you may consult Chapter 5 of “Quantum Field Theory of Many-Body
Systems” by Xiao-Gang Wen.
Now that we have established the phenomenology of Fermi liquid theory, we want to further
investigate the physical origin of the Fermi liquid function and to build the foundation of Fermi liquid theory
on quantum field theory.
As mentioned in the previous section, the Fermi liquid function contains information of quasiparticle
interaction. It is natural to consider the type of interaction due to two quasiparticle scattering, which is
diagrammatically associated with the vertex contribution.
Let us begin with the diagrammatic descriptions for quasiparticles. The particle-like excitations may be
considered as a pole of the retarded Green’s function GR in the lower half-plane near the positive real half-
line of the complex energy variable ε, whereas the hole-like excitations correspond to a pole of the advanced
Green’s function GA in the upper half-plane near the negative real half-line of the complex variable ε.
Therefore, the Green’s function for quasiparticles G can be expressed by the following form
⎡ a ⎤
G ( k , ε ) = lim ⎢ ⎥ (VI.30)
⎣ ε − v ( k − k F ) + iη sgn ( k − k F ) ⎦
η →0 +
for (|k|, ε) near (kF, 0), v = kF/m*, and a is a coefficient between 0 and 1, which will be determined a bit later.
Next, we consider the properties of the vertex part Γ ( k1 , k2 ; k3 , k4 ) . Specifically, we are interested in the
behavior of small momentum and energy transfer so that k1 is near k3 and k2 is near k4. We therefore
introduce the notation
Γ ( k1 , k2 ; k1 + κ , k2 − κ ) ≡ Γ ( k1 , k2 ; κ ) , (VI.31)
where the energy-momentum transfer k = (κ, ε) is a small four-vector so that |κ| << kF and ε << μ. The vertex
part can be expressed in terms of the sum over the ladder diagrams as shown in Fig. VI.2.1.
k1 + κ k2 − κ k1 + κ k2 − κ k1 + κ k2 − κ k1 + κ k2 − κ k1 + κ k2 − κ
q+κ q+κ
γ δ γ δ γ η δ(1) γ δ γ η δ
Γ = Γ(1) + Γ(1)
ξ
Γ +… = Γ(1) + Γ(1)
ξ
Γ
α β α β α β α β α β
q q
k1 k2 k1 k2 k1 k2 k1 k2 k1 k2
Fig. VI.2.1 The ladder-diagram contributions to the two-particle scattering vertex Γ. Here Γ(1) denotes
the vertex contribution that does not involve singularities upon κ → 0.
It is worth mentioning that there are in fact three different types of ladder diagrams associated with the
vertex contributions, as illustrated in Fig. VI.2.2. However, the first two types illustrated in Figs. VI.2.2 (a) –
(b) nothing special happens as κ → 0, whereas the poles of the two Green’s functions of Fig. VI.2.2 (c),
which is equivalent to Fig. VI.2.1, come together as κ → 0, leading to the appearance of singularities in Γ.
More explicitly, we can express the sum in Fig. VI.2.1 as an integral equation:
d 4q
Γαβ ,γδ ( k1 , k2 ; κ ) = Γαβ
(1)
,γδ ( k1 , k 2 ; κ ) − i ∫
(1)
Γαξ ,γη ( k1 , − q; κ ) G ( q ) G ( q + κ ) Γξβ ,ηδ ( q , k 2 ; κ ) , (VI.32)
( 2π )
4
where Γ(1) denotes the vertex contribution that does not yield singularities upon taking κ → 0, although Γ(1)
must contain pertinent momentum transfer to satisfy the diagrammatic representations. The integral in EQ.
(VI.32) consists of the contribution from regions far away from the point (|q| = kF, ε = 0) and that from the
neighborhood of this point, and the latter contribution determines the singularities of the expression. For
small momentum transfer κ, we may treat the neighborhood of the point (|q| = kF, ε = 0) to be small, so that
the only significant contribution to the integral comes from the part of the contour going around the poles of
the Green’s functions. Given that the poles of the two Green’s functions are very close to each other, we may
assume that other quantities in the integrand are slowly varying, and the only non-trivial contributions from
the poles can only occur if the poles lie on the opposite side of the real axis. Thus, the only possibilities for
the condition to be satisfied are |q| < kF, |q + κ| > kF or |q| > kF, |q + κ| < kF. For small κ, we must have |q| ≈
kF and ε = 0. Therefore, the product of the Green’s functions G(q) G(q+κ) may be replaced by Aδ(ε)δ(|q|−kF),
where the coefficient A is determined by integrating G(q) G(q+κ) relative to |q| and ε, which yields:
i 2π a 2 v iκ
A= , (VI.33)
v ω − v iκ
where v directs along q and |v| = kF/m*. Therefore, we can express the product G(q) G(q+κ) as follows:
i 2π a 2 viκ
G( q ) G( q + κ ) = δ ( ε ) δ ( q − k F ) + ϕ ( q ) ≡ i Φ( q ) + ϕ ( q ) , (VI.34)
v ω − viκ
where ϕ ( q ) represents the regular part of G(q) G(q+κ) that is important only in the integral far from the
point (|q| = kF, ε = 0).
Fig. VI.2.2 Three different types of ladder diagrams for the two-particle scattering vertex Γ:
, βη ( k1 , − q; − k1 − k 2 ) , Γξγ ,ηδ ( q , − k1 − κ ; − k1 − k 2 ) ; (b) Γαξ ,δη ( k1 , − q; k 2 − k1 − κ ) ,
(1) (1)
(a) Γαξ
Γξβ ,ηγ ( q, k2 ; k 2 − k1 − κ ) ; (c) Γαξ
(1)
,γη ( k1 , − q; κ ) , Γξβ ,ηδ ( q , k 2 ; κ ) .
The limit of the Green’s function product in EQ. (VI.32) and that of the vertex Γ as κ, ω → 0 depend
on the ratio of |κ| to ω. We first consider the limit ω → 0 and |κ| /ω → 0 so that i Φ( q ) → 0 from EQ.
ω
(VI.34), and denote the vertex in this limit by Γ . Thus, EQ. (VI.32) can be rewritten into
d 4q
,γδ ( k1 , k 2 ; κ ) = Γαβ ,γδ ( k1 , k 2 ; κ ) − i ∫ ,γη ( k1 , − q; κ ) ϕ ( q ) Γξβ ,ηδ ( q , k 2 ; κ ) .
ω (1) (1) ω
Γαβ Γαξ (VI.35)
( 2π )
4
To eliminate Γ(1) from EQ. (VI.35), we use the following short-hand expressions to rewrite EQs. (VI.32) and
(VI.35):
Γ = Γ (1) − i Γ (1) ( i Φ + ϕ ) Γ , (VI.36)
ω (1) (1) ω
Γ =Γ −iΓ ϕ Γ , (VI.37)
where the products are interpreted as integrals. From EQ. (VI.37) we obtain
( )
−1
Γ ω = 1 + i Γ (1)ϕ Γ (1) , (VI.38)
Γ = Γ ω + Γ ω ΦΓ , (VI.39)
a 2 k F2 viκ
Γαβ ,γδ ( k1 , k2 ; κ ) = Γαβ
ω
,γδ ( k1 , k 2 ; κ ) + Γαξ ,γη ( k1 , − q; κ ) Γξβ ,ηδ ( q, k 2 ; κ ) .
ω
( 2π ) v ∫ ω − v i κ
dΩ
3
(VI.40)
Next, we consider the other limiting case where |κ| → 0 and ω/|κ| → 0, and we denote the vertex
κ κ ω
contribution in this limit by Γ . Using EQ. (VI.40), we obtain the following relation between Γ and Γ :
a 2 k F2
,γδ ( k1 , k 2 ; κ ) = Γαβ ,γδ ( k1 , k 2 ; κ ) − d Ω Γαξ ,γη ( k1 , − q; κ ) Γξβ ,ηδ ( q, k2 ; κ ) .
κ ω ω κ
( 2π ) v ∫
Γαβ 3
(VI.41)
Now we are ready to investigate the poles of the vertex Γ ( k1 , k2 ; κ ) for small κ and ω. Noting that near the
pole Γ ( k1 , k2 ; κ ) Γω ( k1 , k2 ; κ ) , we may neglect the first term Γ in the right-hand side of EQ. (VI.41).
ω
Furthermore, by inspecting the second term in EQ. (VI.41) and Fig. VI.2.1, we find that near the pole,
ω
where χ and χ′ are two functions. Inserting EQ. (VI.42) into EQ. (VI.41) and neglecting the first term Γ ,
′ ( k2 ; κ ) on both sides, and obtain the following:
we can cancel the common term χ βδ
a 2 k F2
χαγ ( k1 ; κ → 0 ) ≈ − d Ω Γαξ ,γη ( k , − q; κ → 0 ) χξη ( q; κ → 0 ) .
ω
( 2π ) v ∫
3 1
(VI.43)
a 2 k F2 viκ
χαγ ( k1 ; κ ) ≈ Γαξ ,γη ( k1 , − q; κ ) χξη ( q; κ ) .
ω
( 2π ) v ∫ ω − v i κ
dΩ
3
(VI.44)
If we define
niκ
vαγ ( n ) = χαγ ( q; κ ) , (VI.45)
ω − niκ
where n denotes the normal vector along k1, EQ. (VI.44) can be rewritten into:
k F2
(ω − n i κ ) vαγ ( n ) = n i κ d Ω a Γαξ ,γη ( n, l ) vξη ( l ) .
ω
( 2π ) ∫
2
3
(VI.46)
Here l denotes the normal vector along q for small κ. We’ll show in Part VI.3 that EQ. (VI.46) is essentially
the equation for zero sound and spin waves, which is natural because the poles of the vertex determine the
ω
acoustic excitations of the Fermi liquid. Moreover, the quantity a2 Γ is in fact the Fermi liquid function f
κ
introduced in Part VI.1, and its relation to a2Γ through EQ. (VI.41) implies that the Fermi liquid function is
related to the forward scattering amplitude of two quasiparticles. To understand the last point, we note that
the two-particle Green’s function consists of two parts, one corresponds to the free motion of two particles
and the other corresponds to scattering of the particles by each other:
,γδ ( k1 , k 2 ; k3 , k 4 ) = ( 2π ) G ( k1 ) G ( k 2 ) ⎡
⎣δ ( k1 − k3 ) δαγ δ βδ − δ ( k1 − k4 ) δ βγ δαδ ⎤⎦
(2) 4 (0) (0)
Gαβ
+iG (0)( k1 ) G (0) ( k2 ) G (0)( k3 ) G (0)( k4 ) Γαβ
′ ,γδ ( k1 , k2 ; k3 , k4 ) . (VI.47)
We have also mentioned in the beginning of this section that near the poles, the Green’s functions of
quasiparticles have a form very similar to the free-particle Green’s function except a factor a. Therefore,
from EQ. (VI.47) we find that a 2 Γαβ′ ,γδ ( k1 , k2 ; k3 , k4 ) corresponds to the amplitude of scattering two
quasiparticles. In particular, in the limit of ω = 0, a 2 Γκ ( k1 , k2 ; k3 , k4 ) corresponds to the forward scattering
amplitude. Consequently, given a scattering mechanism for quasiparticles, we can in principle calculate the
vertex contribution and establish the effective mass and various properties of a Fermi liquid system. In Part
II.10 we have investigated a specific case of electron-phonon scattering, which is one of the fundamental
processes in condensed matter physics. Another special case related to resonant scattering of electrons due to
many-body interactions with magnetic impurities will be considered in Part VII.
Having established the physical significance of the vertex contribution, we derive in Part VI.3
several basic relations of Fermi liquid theory and investigate the bosonic excitations (zero sound and spin
waves) in Fermi liquid systems.
To derive basic relations of Fermi liquid theory involving the Green’s function of quasiparticles, we
need to investigate how the two-particle Green’s function varies under a small perturbation field. We shall
consider in the following four different types of small perturbations that lead to four relations between the
Green’s function of interacting quasiparticles and the vertex contributions.
Let’s first consider a perturbation in the form of an external time-dependent field δU(t). The
corresponding interacting Hamiltonian is therefore given by
If we express the Green’s function in the interaction picture with respect to Hint and expanding the Green’s
function in a power series in δU(t) and then keeping terms up to the first order in δU(t), we find that the
changes in the Green’s function under the perturbation becomes:
δ Gαβ ( x, x′ ) = − ∫ d 4 y δ U ( t y ) ⎡⎣ T (ψ Hα ( x )ψ H† γ ( y )ψ H γ ( y )ψ H† β ( x′ ) ) − T (ψ H α ( x )ψ H† β ( x′ ) ) ψ H† γ ( y )ψ H γ ( y ) ⎤⎦
(VI.49)
where ψH are the Heisenberg operators of the interacting particles in the absence of the external perturbation
field δU(t). Comparing EQ. (VI.49) with the two-particle Green’s function, we may rewrite δGαβ into:
δ Gαβ ( x, x′ ) = δαβ ∫ d 4 y δ U ( t y ) [G ( x − y ) G ( y − x′ )]
( )∫d
− i∫ d 4 y δU t y 4
x1d 4 x2 d 4 x3 d 4 x4 G ( x − x1 ) G ( y − x2 ) G ( x3 − x′ ) G ( x4 − y ) Γαγ , βγ ( x1 , x2 ; x3 , x4 ) .
(VI.50)
The diagrams equivalent to EQ. (VI.50) is shown in Fig. VI.3.1. The Fourier transformation of EQ. (VI.50)
becomes
δ Gαβ = δ αβ G ( k ) δ U ( ω ) G ( k + κ1 )
d 4q
− i G ( k ) G ( k + κ1 ) ∫ G ( q ) δ U ( ω ) G ( q + κ1 ) Γαγ , βγ ( k , q; κ1 ) , (VI.51)
( 2π )
4
where κ1 ≡ ( 0, ω ) . Furthermore, we note that the field δU(t) has no effect on the spin of the particles, so that
δGαβ is proportional to δαβ, so we can take the trace of EQ. (VI.51) and rewrite it into:
δ G = G ( k ) δ U ( ω ) G ( k + κ1 )
1 d 4q
− i G ( k ) G ( k + κ1 ) Γαβ ,αβ ( k , q; κ1 ) G ( q ) δ U ( ω ) G ( q + κ1 ) .
2 ∫ ( 2π )
4
(VI.52)
x′,β x′,β
δU x3 x4 δU
δ Gαβ ( x, x′ ) = y,γ +
x1 x2 y,γ
x,α x,α
Figure VI.3.1 Diagrammatic expression for δ Gαβ ( x, x′ ) under an external perturbation field δU, to
the first order in δU.
On the other hand, in the limit of δU → constant, the interaction Hamiltonian becomes
H int = δ U ∫ d 3 x ψ α† ( x ) ψ α ( x ) = δ U Nˆ , (VI.53)
where N̂ is the total number operator of the system. Consequently, the Green’s function of the perturbed
( ) − i δ U t −t ′
system can be obtained by simply multiplying the unperturbed Green’s function by e , which
corresponds to shifting the energy of the system by −δU. In other words, we have ( δ G δ U ) → − ( ∂G ∂ε )
for ω → 0. Therefore, from EQ. (VI.52) we find
∂G ⎡ i d 4q ω ⎤
= − {G ( k )} ⎢1 − ∫
2
Γαβ ,αβ ( k , q ) {G 2 ( q )} ⎥ , (VI.54)
∂ε ⎣⎢ 2 ( 2π )
ω 4 ω
ω →0 ⎦⎥
lim G ( k ) G ( k + κ1 ) ≡ {G 2 ( k )} = ϕ ( k )
ω →0 ω
by using EQ. (VI.34). Writing G(k) in the form of EQ. (VI.30) and dividing EQ. (VI.54) by − G 2 ( k ) { }ω
, we
obtain the first basic formula:
∂G −1 ( k ) 1 d 4q
∂ε
=
a
= 1−
i
2 ∫ ( 2π )
4
ω
Γαβ {
,αβ ( k , q ) G ( q )
2
} ω
. (VI.55)
The second equation can be obtained by considering an external perturbation due to a small charge
variation δ e in the presence of a vector potential A that is homogeneous in space and constant in time. The
change in the Hamiltonian due to this perturbation is therefore given by the following:
δe
H int = − ∫ d x ψ ( x ) ( kˆ iA( x ) ) ψ ( x ) ,
3 †
α α (VI.56)
m
where k denotes the momentum operator. Defining κ 2 ≡ ( κ , 0 ) and following similar steps leading to EQ.
(VI.55), we have
δe ˆ
δ G = −G ( k )
m
( k iA ) G( k + κ 2 )
i d 4q δe
+ G( k ) G( k + κ 2 ) Γαβ ,αβ ( k , q; κ 2 ) G ( q ) ( qi A ) G( q + κ 2 ) ,
2 ∫ ( 2π )
4
(VI.57)
m
d 4q
∇ k G −1 ( k ) = −
v
a
=−
k
ma*
=−
k
m
+
i
2 ∫ ( 2π )
4
κ
Γαβ ,αβ ( k , q )
q
m
{G ( q )}
2
κ
, (VI.58)
where lim G ( k ) G ( k + κ 2 ) ≡ {G 2 ( k )} .
κ →0 κ
The third equation can be obtained by considering the variation in the Green’s function when the
system as a whole moves with a small and slowly varying velocity δ u(t). In this case, the Hamiltonian of the
system is modified by adding the following:
δ H = − δ u i K = − δ u i ∫ d 3 x ψ α† ( x ) k ψ α ( x ) , (VI.59)
where K is the total momentum operator of the system. Therefore, the variation in the Green’s function is
δ G = − G ( k ) ( k i δ u )( ω ) G ( k + κ1 )
i d 4q
+ G ( k ) G ( k + κ1 ) Γαβ ,αβ ( k , q; κ1 ) G ( q ) ( q i δ u )( ω ) G ( q + κ1 ) ,
2 ∫ ( 2π )
4
(VI.60)
where κ1 ≡ ( 0, ω ) as defined before. In the ω → 0 limit, the change in the Green’s function is consistent with
a transformation to a coordinate that moves with a constant velocity δ u. Therefore, the energy change is δε =
( k i δ u ) and δ G = ( ∂G ∂ε ) k i δ u , so that from EQ. (VI.60) we find that by dividing − G 2 ( k ) , we have { }ω
∂G −1 d 4q
k
∂ε
=
k
a
=k−
i
2 ∫ ( 2π )
4
ω
Γαβ {
,αβ ( k , q ) q G ( q )
2
}ω . (VI.61)
Finally, the fourth equation can be derived by considering the change in the Green’s function under the
influence of a small field δU(x) constant in time and weakly homogeneous in space. The change in the
Green’s function is given by
δ G = G( k ) δ U ( κ ) G( k + κ 2 )
i d 4q
− G( k ) G( k + κ 2 ) Γαβ ,αβ ( k , q; κ 2 ) G ( q ) δ U ( κ ) G ( q + κ 2 ) ,
2 ∫ ( 2π )
4
(VI.62)
∂G −1 d 4q
∂μ
= 1−
i
2 ∫ ( 2π )
4
κ
Γαβ {
,αβ ( k , q ) G ( q )
2
} κ
. (VI.63)
Using EQs. (VI.41), (VI.55), (VI.58), (VI.61) and (VI.63), we are ready to derive the basic relations of
the Fermi liquid theory. In the following we consider the effective mass expression, the relation between the
Fermi momentum and the particle density, and the acoustic excitations.
[Effective mass]
a 2 k F2
,αβ ( k , q ) = Γαβ ,αβ ( k , q ) − d Ω Γαξ ,αη ( k , − q ) Γξβ ,ηβ ( q, q′ )
( 2π ) v ∫
κ ω ω κ
Γαβ ′ ′ 3
a 2 k F2
= ω
Γαβ ,αβ ( k , q′ ) − ∫ d Ω Γαβ ,αβ ( k , − q ) Γα ′β ′,α ′β ′( q, q′ )
ω κ
(VI.64)
2 ( 2π ) v
3
d 4 q′ q′
−
ma
k
*
+
k
m
=
i
2 ∫ ( 2π )
4
κ
Γαβ ,αβ ( k , q )
′
m
{G ( q′)}2
κ
,
d 4 q′ q′ a 2 k F2 ⎡i d 4 q′ ′ ⎤
=
i
2 ∫ ( 2π )
ω
Γαβ ,αβ ( k , q )
′ { G 2 ( q′ ) } −
2 ( 2π ) v
∫ d Ω Γαβ ,αβ ( k , − q ) ⎢
ω
∫ Γακ ′β ′,α ′β ′( q, q′ ) {G 2 ( q′ )} ⎥
q
⎢ 2 ( 2π )
4 κ 3 4 κ
m ⎣ m ⎦⎥
4
a 2 k F2 ⎛ q⎞
=
i d q
2 ∫ ( 2π )
4
ω
Γαβ ,αβ ( k , q )
q
m
{G ( q )}
2
κ
−
2 ( 2π ) v
∫ d Ω Γαβ ,αβ ( k , − q ) ⎜⎝ − m*a + m ⎟⎠ .
3
ω q
(VI.65)
i 2π a 2
{
G2 (q) }κ
=−
v
δ (ε ) δ ( q − kF ) + G 2 ( q ) . { } ω
(VI.66)
Inserting EQ. (VI.66) into EQ. (VI.65) and then using EQ. (VI.61), also taking the limit |k| → kF and ε → 0,
defining k iq ≡ k F2 cos θ , and noting that v = k/m*, we obtain:
kF kF i ⎛ 2π ia 2 m* ⎞ k F2 d Ω ω k cos θ ⎛ k F k F ⎞
− + = ⎜− ⎟∫ Γαβ ,αβ (θ ) F +⎜− + ⎟
m* a ⎠ ( 2π )
4
m 2⎝ kF m ⎝ ma m ⎠
a 2 k F m* ⎛ k F cos θ k F cos θ ⎞
∫ d Ω Γαβ ,αβ (θ ) ⎜⎝ −
ω
− + ⎟
2 ( 2π )
3 *
ma m ⎠
1 1 kF
∫ d Ω a Γαβ ,αβ (θ ) cos θ .
2 ω
⇒ = + (VI.67)
2 ( 2π )
* 3
m m
Comparing EQ. (VI.67) with EQ. (VI.15), we find the following relation between the vertex contribution and
the Fermi-liquid function:
1/ 3
⎛ N⎞
k F = ⎜ 3π 2 ⎟ , (VI.1)
⎝ Ω⎠
for a Fermi liquid under finite quasiparticle interaction. The validity of this relation is one of the most
essential properties of the Fermi liquid theory, because it confirms Landau’s assumption that interacting
quasiparticles still preserve the same states as the non-interacting Fermi gas. Noting that the Green’s function
can be related to the density of states, we begin by considering the Green’s function expression in EQ.
(VI.30) near the pole for |k| → kF and ε → 0. Given that the coefficient a, the velocity v and the Fermi
momentum kF are all functions of the chemical potential, we have
∂G G da G2 dv G2 dk F G2 dk F
= − ( k − kF ) − v ≈− v ,
∂μ a dμ a dμ a dμ a dμ
dk F ⎛ ∂G −1 ⎞
⇒ v = a⎜ ⎟ . (VI.69)
dμ ⎝ ∂μ ⎠ε =0, k = k F
Substituting EQ. (VI.63) into EQ. (VI.69) and using EQ. (VI.64) to express Γκ, we obtain
d 4q 1 a 2 k F2 ⎛ v dk F ⎞
( k , q ) {G ( q )}κ −
v dk F i
d Ω Γαβ ,αβ ( k , q ) ⎜
2 ∫ ( 2π ) 2 ( 2π ) v ∫
ω 2 ω
= 1− Γαβ ,αβ − 1⎟ . (VI.70)
a dμ 4 3
⎝ a dμ ⎠
−1
v dk F⎧⎪ k F2 ⎫⎪
= ⎨1 + ∫ ⎣
d Ω ⎡ a 2 ω
Γ αβ ,αβ ( k , q ) ⎤
⎦⎬ .
a dμ ⎩ ⎪ 2 ( 2π ) v
3
⎪⎭
⎛ ∂G −1 ⎞ d 4q κ
Γαβ ,αβ ( k , q ) {G 2 ( q )}
i
=⎜ ⎟ = 1− ∫ (VI.71)
⎝ ∂μ ⎠ε =0, k = kF 2 ( 2π )
4 κ
Next, we note that the total number density of particles (N/Ω) is related to the Green’s function by the
following relation
N
= ψ H† α ( x )ψ H α ( x ) = −i lim x′→x ,t ′→t + Gαα ( x − x′ ) = −2i lim x′→ x ,t ′→t + G ( x − x′ )
Ω
d 3k d ω
= −2i limt →0+ ∫ G ( k , ω ) eiωt . (VI.72)
( 2π )
4
d ( N Ω) d 4 k ∂G ( k ) d 4 k ⎡ ∂G ( k ) ⎤
−1
= −2i ∫ = 2i ∫ ⎥ {G ( k )}κ .
2
⎢ (VI.73)
dμ ( 2π ) ∂μ ( 2π ) ∂μ
4 4
⎣ ⎦
Taking EQ. (VI.63) and using EQ. (VI.41) for Γκ in EQ. (VI.73), we have
d ( N Ω) d 4k d 4k d 4q
= 2i ∫ {G 2( k ) } +∫ {G ( k )} { }
,αβ ( k , q ) G ( q )
( 2π ) ∫ ( 2π )
2 ω 2
Γαβ
dμ ( 2π )
4 κ 4 4 κ κ
d 4k d 4q ⎛ a 2 k F2 ⎞
+∫ ⎟⎟ ∫ d Ω {G ( k )}κ Γαβ ,αβ ( k , − q ) Γα ′β ′,α ′β ′( q, q′ ) {G ( q )}κ . (VI.74)
( 2π ) ∫ ( 2π )
2 ω κ 2
4 4 ⎜−
⎜ 2 ( 2π ) 3
⎝ v⎠
⎛ v dk F ⎞ d 4q κ
Γα ′β ′,α ′β ′( k , q ) {G 2 ( q )} ,
i
⎜ a dμ ⎟
⎝
− 1
⎠
= − ∫
2 ( 2π )
4 κ
(VI.75)
d ( N Ω) d 4k d 4k d 4q
= 2i ∫ {G 2( k ) } +∫ {G ( k )} { }
,αβ ( k , q ) G ( q )
( 2π ) ∫ ( 2π )
2 ω 2
Γαβ
dμ ( 2π )
4 κ 4 4 κ κ
ia 2 k F2 d 4k ⎛ v dk F ⎞
d Ω {G ( k )} ,αβ ( k , − q ) ⎜
( 2π ) v ∫ ( 2π ) ∫
2 ω
− Γαβ − 1⎟ . (VI.76)
⎝ a dμ ⎠
3 4 κ
d 4k d 4k 8π a 2 k F2
2i ∫ {
G 2( k ) }κ = 2i ∫ { G 2( k ) }ω + ; (VI.77)
( 2π ) ( 2π ) ( 2π )
4 4 3
v
d 4k d 4q
∫ ( 2π ) ∫ ( 2π ) {G ( k )} ,αβ ( k , − q ) G ( q ) { }
2 ω 2
4 4
Γαβ
κ κ
d 4k d 4q ⎛ i 4π k F2 a 2 ⎞ d 4 q ω
=∫ {G ( k )}ω Γαβω αβ ( k , −q ) {G ( q )}ω − ⎜⎜ Γαβ ,αβ ( k , − q ) {G 2( q )}
( 2π ) ∫ ( 2π )
⎟⎟ ∫
2 2
,
⎝( ) v ⎠ ( )
4 4 3 4 ω
2π 2π
⎛ i 4π k F2 a 2 ⎞ d 4 k ω ⎛ k F4 a 4 ⎞
−⎜
⎜ ( 2π )3 v ⎟ ∫ ( 2π ) 4 αβ ,αβ
⎟ Γ ( k , − q ) { G 2
( k )}ω
− ⎜ 16π 5 v 2 ⎟ ∫ d Ω Γαβ ,αβ ( k , − q ) ;
ω
(VI.78)
⎝ ⎠ ⎝ ⎠
2 2 4
⎛ v dk F ⎞
d Ω {G 2( k )} Γαβ
ia k F d k
− ∫ 4 ∫
ω
,αβ ( k , −q ) ⎜ − 1⎟
( 2π ) v ( 2π ) ⎝ a dμ ⎠
3 κ
ia 2 k F2 d 4k
}ω Γαβω αβ ( k , −q ) ⎛⎜ av dk ⎞ k a ⎛ v dk ⎞ 4 4
=− ∫ ( 2π ) 4 ∫ {
d Ω G 2( k ) , − 1⎟ − ∫
ω
d Ω Γαβ αβ ( k , − q ) ⎜
F
− 1⎟ .
F
,
F
( 2π ) v ⎝ d μ ⎠ 16π v ⎝ a dμ ⎠
5 2
3
(VI.79)
Inserting EQs. (VI.77) – (VI.79) into EQ. (VI.76), we find cancellations of terms between EQ. (VI.78) and
EQ. (VI.79). By further applying EQ. (VI.55) to the sum of the three terms, we arrive at
d ( N Ω) d 4k d 4k d 4q 8π a 2 k F2
= 2i ∫ { G 2( k ) } +∫ 4 {
G 2( k ) }ω {
,αβ ( k , q ) G ( q ) }ω
( 2π ) ∫ ( 2π )
ω 2
Γαβ +
dμ ( 2π ) ( 2π )
4 ω 4 3
v
8π k F2 ( a − 1) dk F k a 4 4
⎛ v dk F ⎞
∫ d Ω Γαβ ,αβ ( k , −q ) ⎜⎝ a d μ ⎟⎠ .
ω
− v − F
(VI.81)
( 2π ) v d μ
3 5 2
16π v
The first two terms in EQ. (VI.81) can be simplified by applying EQ. (VI.55), which yields
d 4k ⎡ d 4q ω ⎤
∫ ( 2π )4 { }ω ⎢ ∫ 4 Γαβ ,αβ ( k , q ) {G 2( q )}ω ⎥
G 2( k )
⎢⎣
2i +
( 2π ) ⎥⎦
d 4k ⎡ ⎛ ∂G −1 ⎞ ⎤ d 4 k ∂G
4 { }ω ⎢
=∫ G 2
( k ) 2i + 2 i ⎜ ∂ε − 1 ⎟⎥ = −2i ∫ ( 2π )4 ∂ε
( 2π ) ⎣ ⎝ ⎠⎦
d 3k
= −2i ∫ ⎡⎣G ( ε = ∞ ) − G ( ε = −∞ ) ⎤⎦ = 0 .
( 2π )
4
We may also use EQ. (VI.72) to express the last term of EQ. (VI.81). Finally, we obtain
d ( N Ω) 8π k F2 dk F
= . (VI.82)
( 2π )
3
dμ dμ
N 8π k F3 k F3
= = .
3 ( 2π )
3
Ω 3π 2
Thus, we have proven that EQ. (VI.1) holds under finite quasiparticle interaction.
Let us begin with the consideration of first sound. To find the sound velocity, we need to know the
compressibility of the Fermi liquid, which is given by ( ∂μ ∂N ) . Noting that the chemical potential μ is only
a function of (N/Ω), we have
∂μ Ω 2 ∂P 1 ∂P
=− = .
∂N 2
N ∂Ω N ∂ ( N Ω)
∂P ∂P 1 ∂μ
u2 = = = N . (VI.83)
∂ρ ∂ ( mN Ω ) m ∂N
Now we may use the relation derived in EQ. (VI.81), which yields
d ( N Ω)
−1
1 ⎛ ∂μ ⎞ 1⎛ N ⎞ 8π k F3 1 8π k F2 ⎛ dk F ⎞
= ⎜ ⎟ = ⎜ ⎟= = 3 ⎜
v ⎟
Ω ⎝ mu 2 ⎠ 3 ( 2π ) mu 2 ( 2π ) v ⎝ d μ ⎠
3
dμ Ω ⎝ ∂N ⎠
−1
8π k F2 m* ⎛ k F2 ⎞
3 ∫
2 ω
= ⎜⎜ 1 + dΩ a Γ ⎟ , (VI.84)
( 2π ) k F ⎝ 2 ( 2π ) v
3 ⎟
⎠
where we have used EQ. (VI.71) to derive the second line of EQ. (VI.84). Hence, we find the velocity of first
sound can be related to the vertex contributions as follows:
⎛
k F2 k F m* 2 ω
⎞ k F2 1 ⎛ kF ⎞
3
2 ω
3 ∫ ⎜ ⎟ ∫ dΩ a Γ .
u2 = * ⎜
1 + d Ω a Γ ⎟⎟ = + (VI.85)
⎜
3mm ⎝ 2 ( 2π ) *
⎠ 3mm 6m ⎝ 2π ⎠
Next, we consider the propagation of high-frequency acoustic excitations, the zero sound. We shall
take a semi-classical description for the zero sound in the following, and leave the rigorous field theory
derivation to you in a problem set.
In the presence of acoustic excitations, the fermion density distribution function n exhibits spatial and
temporal variations that deviate from the equilibrium value nF. We may express the distribution function as:
n = nF + δ n with δ n ~ exp [i ( k ir − ωt )] .
∂n ⎛δn ⎞ ∂n
+ ∇ r n i ∇ k ε − ∇ k n i ∇ rε ≈ O ⎜ ⎟
∂t ⎝τ ⎠ ∂t
∂n
⇒ + ∇ r n i ∇ k ε − ∇ k n i ∇ rε ≈ 0 . (VI.87)
∂t
where f is the Fermi liquid function as defined before. Inserting EQ. (VI.88) into EQ. (VI.87), and noting that
v = ∇ k ε and ( ∂n ∂ε )ε ≈ − δ ( ε − μ ) , we find the following relation for δn:
F
∂n d 3k ′
( v i k − ω )δ n − ( v i k ) Trσ ′ ∫ f δ n′ = 0 . (VI.89)
∂ε ( 2π )
3
εF
k F m*
(vik −ω)ϒ + (vik) Trσ ′ ∫ d Ω′ f ϒ′ = 0 . (VI.90)
( 2π )
3
In Problem Set 5 you will be asked to find the explicit form for the function ϒ in terms of the Green’s
functions and vertex contributions in the small momentum transfer limit.
It is clear from EQ. (VI.90) that the zero sound propagation differs from ordinary sound because the
density distribution function in the former is no longer isotropic. We also remark that the solutions to EQ.
(VI.90) represent the zero sound modes, and the Fermi surface under zero sound propagation does not remain
spherical, with changes in the Fermi surface determined by the function ϒ. The microscopic origin for the
zero sound excitations in Fermi liquids may be attributed to the short-range repulsive interaction of fermions
at low temperatures. Hence, the properties of Fermi liquids may differ for different systems such as liquid
3
He (known as a neutral Fermi liquid), metals, superconductors, and ferromagnetic materials. The possibility
of having sound waves propagating at T = 0 implies that these modes are bosonic in nature.
Finally, we remark that zero sound of electrons in metals cannot occur with varying the electron
density alone. Otherwise there would be uncompensated electric charge, leading to very large energies
required for such excitations. The zero-sound propagation associated with variations of electron densities in
metals must be accompanied by changes in the crystalline lattice to ensure charge neutrality. Consequently,
electron-phonon interactions are essential for the occurrence of high-frequency acoustic excitations in metals,
and we have considered such interactions in Part V.
So far we have been primarily concerned with perturbative effects on Fermi liquids. In Part VII, we
investigate an interesting example of the Fermi liquid theory in the presence of localized magnetic moments.
Although the conduction electrons may still be considered as weakly interacting among themselves, the
many-body interaction of the localized magnetic moments with the spins of conduction electrons is in fact
strong and non-perturbative, leading to the celebrated Kondo effect, a strongly correlated phenomenon
associated with the Fermi liquid theory.
Further Readings
Subsequently, it was realized that the Kondo effect was not limited to metallic systems with dilute
magnetic impurities. For instance, in late 1970’s, a class of compounds involving magnetic f-electrons in
their crystalline structures was found to have very large electronic specific heat corresponding to
approximately 1000 times effective masses at the Fermi level as predicted by bandstructure calculations.
These heavy fermion systems can be described as Kondo lattices, and the origin of the heavy masses may be
attributed to the strong many-body interaction with magnetic moments. In late 1990’s and early 2000’s,
modern nanotechnology led to further realizations of the Kondo effect in such systems as quantum dots,
carbon nanotubes, and artificial atomic structures constructed by scanning tunneling microscopy. To date the
“Kondo physics” remains an active and well defined testing ground for new numerical and analytical tools
that have been developed to investigate other challenging many-body problems. In this section, we describe
how the localized magnetic moment interacts with conduction electrons through the slave-boson techniques
that bosonize the action of the interacting system.
jmk
( )
jm
( )
kσ
( ) ( )
H = ∑ vkm ei k i Rj d †jm ckσ + h.c. + ξ 0f ∑ d †jm d jm + ∑ ξk ck†σ ckσ + U ∑ nd jm nd jm′ ,
jmm′
(VII.1)
where d †jm creates an electron in the m’th orbital of Ce at the j-th site, ck†σ is the creation operator for the
Bloch state of conduction electrons, vkm is the hybridization energy parameter between the electron of Ce and
the conduction electrons, ξ 0f is the energy of the f-orbital electron of Ce measured relative to the Fermi level,
ξk is the conduction electron energy measured relative to the Fermi level, and U is the Coulomb repulsion
energy for two electrons on the same site of Ce.
Generally the f-band of the cerium compounds is narrow relative to the on-site Coulomb repulsion
energy U. Therefore, we may take the U → ∞ limit in cerium compounds so that double occupancy of the f-
orbital is forbidden. To implement the constraint of no double occupancy, we apply the slave boson
technique to the f-orbital electrons by expressing the electron operators in terms of the following product:
d †jm = b j f jm
†
, d jm = b †j f jm , (VII.2)
†
where b j denotes a “slave boson” (or a “holon”) operator and f jm is a “pseudo fermion” (or a “spinon”)
operator. We note that the physical significance of the slave boson operator b†j ( b j ) corresponds to the
creation (annihilation) of the f 0 state on the j-th site. Under the U → ∞ limit, we may consider the f-electron
operator d †jm = b j f jm
†
as creating a fermion state from an initially empty state, which is equivalent to
annihilating a slave boson and thus the operator b j . Similarly, the annihilation of an f-electron leads to the
creation of a slave boson, leading to d jm = b†j f jm . Furthermore, we note that the chemical potential that
couples to real particle density operators d †jm d jm can only couple to f jm
†
f jm but not to b†j b j in the slave-
boson formalism.
Next, we impose the following constraint to enforce the condition of no double occupancy:
⎛ † ⎞
⎜ b j b j + ∑ f jm f jm ⎟ ψ j = ψ j .
†
(VII.3)
⎝ m ⎠
Therefore, ψ j includes both the f 0-state for which b†j b j = 1 and the f 1-state for which ∑ m f jm† f jm = 1 ,
but it excludes all other higher occupancy states because of the condition U → ∞. The introduction of the
slave boson operator effectively incorporates the many-body interaction of conduction electrons with the
localized f-electron because it imposes partial occupancy of the f-electron as the result of its dynamic
coupling with conduction electrons near the Fermi level, as illustrated in Fig. VII.1.1.
To proceed further, we consider the mean-field approximation in the limit of large degeneracy, so that the
hybridization parameter vkm is replaced by a mean value
v
vkm → ≡v , (VII.4)
N
where N denotes the degree of degeneracy in the total angular momentum J, N = (2J + 1). Therefore, the
Feynman diagrams for the propagator self-energy may be classified in powers of 1/N, depending on the
number of vertices and the sum over the f-state degeneracy in the diagram. In the N → ∞ limit, we may
replace the operators bj by a mean-field value
2
z ≡ b†j b j = b , (VII.5)
which represents the average occupancy of the f 0-state and also renormalizes the hybridization parameter v .
We therefore rewrite the Hamiltonian as
H = H0 +H ′, (VII.6)
where
⎛ ⎞
( )
H 0 = ∑ ξ k ck†σ ckσ + ξ 0f ∑ f jm
†
( )
f jm + v z1/ 2 ∑ f jm
†
(
ckσ + h.c. + λ ∑ ⎜ z + ) ∑f †
jm f jm − 1 ⎟ (VII.7)
kσ jm jmk j ⎝ m ⎠
and (
H ′ = v ∑ δ b j f jm
†
ckσ + h.c. , ) δ bj ≡ bj − b . (VII.8)
j
We note that λ in EQ. (VII.7) is a variational parameter that enforces the U → ∞ limit. The term H ′ given
in EQ. (VII.8) corresponds to fluctuations of the bosonic field, which is generally small. In the following we
shall restrict our discussion to mean-field theory only so that the fluctuations term is ignored.
(a) Initial state Virtual state Final state (b) Density of states
TK
U
Energy
Δ0
ε 0f
Fig. VII.1.1 (a) A simplified picture of the Kondo effect based on the Anderson model: an magnetic
impurity level at an energy ε 0f below the Fermi surface is initially occupied by one spin-up electron,
and the on-site Coulomb repulsion energy to doubly occupy the same site is denoted by U, which is
much larger than both ε 0f and the linewidth of the impurity state Δ0. The electron in the impurity
level may temporarily tunnel into an empty state near the Fermi level, (which corresponds to the
“virtual state”), and then may be replaced by another spin-down electron near the Fermi level,
thereby effectively flipping the spin in the impurity level, as shown in the “final state”. (b)
Schematic illustration of the electronic density of states (horizontal axis) vs. energy (vertical axis)
plot associated with the Kondo effect, indicating that an initial broad energy state of the magnetic
impurity can result in a sharp resonant peak (of linewidth ~ TK) near the Fermi level after
hybridization with conduction electrons due to the Kondo resonance.
Equations (VII.7) and (VII.8) can be applied to both the single-impurity Kondo problem and the
Kondo lattice problem. In the single-impurity case, we have
( ) (
H 0 = ∑ ξk ck†σ ckσ + v z1/ 2 ∑ f m† ckσ + h.c. + ξ 0f + λ) ( )∑( f †
m fm ) + λ ( z − 1) ,
kσ mk m
( ) ( ) (
≡ ∑ ξ k ck†σ ckσ + ∑ ξ rf f m† f m + b v ∑ f m† ckσ + h.c. + ξ rf − ξ 0f ) ( )( b 2
)
−1 , (VII.9)
kσ m mk
where we have defined a renormalized single impurity energy ξ fr ≡ ξ 0f + λ , which has been shifted by λ( )
relative to the bare impurity energy level ξ 0f due to many-body interactions.
To find the mean-field solution to the ground state energy, we minimize the Hamiltonian in EQ.
(VII.9) relative to λ and z. Recall the Hellman-Feynman theorem in Part II:
∂ ∂H ( λ )
ψ (λ ) H (λ ) ψ (λ ) = ψ (λ ) ψ (λ )
∂λ ∂λ
where ψ ( λ ) is the ground state of H ( λ ) , we obtain from ( ∂H ∂λ ) = 0 and EQ. (VII.9) the following
constraint equation:
= 1 − ∑ f m† f m ≡ 1 − ∑ n fm ,
2
z= b (VII.10)
m m
The action S of the single-impurity plus conduction-electron system according to the Hamiltonian
EQ. (VII.9) is related by the partition function Z by the following relation
(
Z = ∫ Dc † Dc Df † Df Db† Db δ Q ,1 e S , ) (VII.12)
where the Kronecker delta function δ Q ,1 is expressed by the following form (which is also known as “the
Abrikosov’s trick”):
1 π β
δ Q ,1 ≡ ∏τ ∫ d λ exp ⎡⎢ −i ∫ dτ λ (τ )( Q − 1) ⎤⎥ , (VII.13)
2π − π ⎣ 0 ⎦
and Q ≡ b†b + ∑ f m† f m . (VII.14)
m
Therefore, we find that the action for temperature T−1 = β is given by:
⎧ ⎫
( ) ( )
β
S = − ∫ dτ ⎨∑ ck†σ ( ∂τ + ξ k ) ckσ + ∑ f m† ∂τ + ξ 0f f m + v ∑ b f m† ckσ + ck†σ f m b† + b†( ∂τ b ) ⎬ . (VII.15)
0
⎩ kσ m mk ⎭
Moreover, using the U → ∞ constraints expressed in EQs. (VII.13) and (VII.14), we have
β
S ≡ S − i ∫ dτ λ (τ )( Q − 1) , (VII.16)
0
⎧
( )
β
= − ∫ dτ ⎨b† ( ∂τ + iλ (τ ) ) b + ∑ ck†σ ( ∂τ + ξk ) ckσ + ∑ f m† ∂τ + ξ 0f + iλ (τ ) f m
0
⎩ kσ m
+ v ∑ ( b f m†ckσ + ck†σ f m b† ) − iλ (τ ) .
mk
} (VII.17)
If we further shift the particle operators in the action S by the following relations:
1
Gk0(ξ ) = , (VII.19)
∂τ + ξk
⎧ ⎛ ⎛ ⎞⎞ ⎫
( )
β −1
S = − ∫ dτ ⎨b† ( ∂τ + iλ (τ ) ) b + ∑ ck†σ Gk0 ckσ + ∑ f m† ⎜ ∂τ + ξ 0f + iλ (τ ) − v 2 ⎜ b ∑ Gk0 b† ⎟ ⎟ f m − iλ (τ ) ⎬ .
0
⎩ kσ m ⎝ ⎝ k ⎠⎠ ⎭
(VII.20)
Hence, we can integrate out shifted c-operators in EQ. (VII.18) and obtain (see Problem Set 6)
Z = ∫ Dλ Db † Db e ( ),
Seff
(VII.21)
{ } ∑ ln (1 + e− βξ ) + Tr ( ln G ) ,
β
where Seff = − ∫ dτ b† ( ∂τ + iλ ) b − iλ (τ ) + k −1
f (VII.22)
0
k
and
⎛ ⎞
( )
Tr ln G −f 1 ≡ ∑∑ ln ⎜ ∂τ + ξ 0f + iλ (τ ) − v 2 b (τ ) ∑ Gk0 b†(τ ) ⎟ . (VII.23)
m ωn ⎝ k ⎠
Here Gk0 and G f denote the Green’s functions of the conduction electrons and f-electron, respectively. We
further note that in EQ. (VII.22) we have used the identity:
(
Tr ln ⎡ Gk0
⎣
( )
−1
)
⎤ ≡ ∑ ln ( −iω + ξ ) ≡ ln (1 + e − βξk ) ,
⎦ ωn n k ωn ≡ ( 2n + 1) π T .
Therefore, we have effectively bosonized our problem according to EQ. (VII.21). The bosonization is non-
perturbative, and its procedure involves the hybridization of the conduction electrons and f-electron through
the “shift in the c-operators”, as expressed in EQ. (VII.18). Next, we solve the problem in the mean-field
limit so that both b and λ are τ-independent and are simply c-numbers. Hence, we have b (τ ) → b = z1/ 2
and iλ (τ ) → λ , and the mean-field action becomes:
MF
Seff = βλ (1 − z ) + ∑σ ln (1 + e
k
− βξ k
) + ∑ ln ⎛⎜⎝ −iω
ωn
n + ξ 0f + λ + zv 2 ∑
1⎞
⎟,
k iωn − ξ k ⎠
⎛ ⎞
∑ ln (1 + e ) 1
+ ∑ ln ⎜ −iωn + ξ rf + zv 2 ∑
− βξ k
≡ βλ (1 − z ) + 2 ⎟. (VII.24)
k ωn ⎝ k iωn − ξ k ⎠
Here we remark that the derivation of EQ. (VII.24) does not involve any perturbation in the diagrammatic
expansion. It is simply obtained by assuming the saddle-point solution to the total action.
To proceed further, we assume that the momentum sum in EQ. (VII.24) is over the conduction band
of a bandwidth 2W. Therefore, we have
1 ∞ N (ε ) 1
∑ iω ≈ N (0) ∫ ( ξ = ε − EF )
W
= ∫ dε dξ ,
k n − ξk −∞ iωn − ξ −W iωn − ξ
⎛ iωn − W ⎞ ⎛ ωn + iW ⎞
= −N ( 0 ) ln ⎜ ⎟ = −N ( 0 ) ln ⎜ ⎟,
⎝ iωn + W ⎠ ⎝ ωn − iW ⎠
⎛ z eiϕ ⎞ −1 ⎛ W ⎞
≡ −N ( 0 ) ln ⎜ − iϕ ⎟ = −2iϕ N ( 0 ) = −2iN ( 0 ) tan ⎜ ⎟, (VII.25)
⎝ ze ⎠ ⎝ ωn ⎠
where we have assumed that the Fermi level is at the center of the conduction band, and that the density of
states of the conduction electrons is nearly a constant N(0). For W >> ωn, which corresponds to low
temperatures and a broad bandwidth limit, EQ. (VII.25) becomes
1
∑ iω − ξk
≈ −iπ N ( 0 ) sgn ( ωn ) . (VII.26)
k n
Hence, the mean-field action in EQ. (VII.24) implies that the f-electron energy level has shifted from ξ 0f for
the non-constrained case (U → 0) to ξ rf for the no-double-occupancy limit (U → ∞), and that the f-electron
energy linewidth evolves from Δ 0 ≡ π v 2N ( 0 ) in the U → 0 limit to Δ ≡ π zv 2 N ( 0 ) = z Δ 0 in the U → ∞
limit. (See Problem Set 6 for more details.) Given EQ. (VII.10), we note that z ≤ 1 and z << 1 when
ξ 0f W , implying that the many-body interaction gives rise to a much sharper resonance peak of a narrow
linewidth Δ ≡ zΔ0 near the Fermi level, as schematically shown in Fig. VII.1.1.
Given EQs. (VII.24) and (VII.26), the mean-field action may be simplified into (see Problem Set 6):
⎛ Δ ⎞
∑ ln (1 + e ) + 2β ∫
− βξ k 1
= βλ (1 − z ) + 2
W
dξ tan −1 ⎜ , (VII.27)
⎜ ξ r − ξ ⎟⎟
MF
Seff βξ
k
−W e +1 ⎝ f ⎠
where we have used the following identity (which you are asked to prove in Problem Set 6):
( ( ) ) ≡ 2∑ ln ( −iω
Tr ln G −f 1 n + ξ rf − iΔ sgn ( ωn ) , )
ωn
⎛ −β ⎞ 1
= 2⎜ ⎟ ∫ dz βξk ln ( − z + ξ rf − iΔ sgn ( ωn ) ) ,
⎝ 2π i ⎠ ( e + 1)
⎛ −2 β ⎞ W 1
=⎜
⎝ 2π i
⎟ ∫−W d ξ βξ
⎠ e + 1
( ln ( −ξ + ξ fr − iΔ ) − ln ( −ξ + ξ fr + iΔ ) ) ,
2β W 1 ⎛ Δ ⎞
= ∫ d ξ βξ tan −1 ⎜ r , ωn ≡ ( 2n + 1) π T .
π −W e +1 ⎜ ξ − ξ ⎟⎟
⎝ f ⎠
At T = 0, we find that the integration over the energy bandwidth 2W is reduced to integration from –W to 0
because there is no electron occupation above the Fermi level ξ = 0. Hence, the third term in EQ. (VII.27)
becomes:
⎛ Δ ⎞ Δ ⎛ (W + ξ rf ) + Δ 2 ⎞
2
⎛ Δ ⎞ ⎛ Δ ⎞ r
⎟⎟ = (W + ξ f ) tan ⎜⎜
0
∫−W ξ −1 −1 −1
⎜ ⎟
tan ⎜⎜ r ⎟ − ξ f tan ⎜⎜ ξ r ⎟⎟ + 2 ln ⎜ ξ r 2 + Δ 2 ⎟
r ⎟
r
d
⎝ ξ f −ξ ⎝ ( f)
⎠ ⎝ W + ξ f ⎠ ⎝ f ⎠ ⎠
⎛Δ ⎞ Δ ⎛ W 2 + Δ2 ⎞
≈ Δ − ξ rf tan −1 ⎜
⎜ ⎟⎟ + ln ⎜ r 2 ⎟ if W ( Δ, ξ ) .
r
(VII.28)
⎝ξf 2 ⎜ ( ) 2 ⎟
r f
⎠ ⎝ f ξ + Δ ⎠
Therefore, upon minimizing the action relative to Δ and ξ rf so that the term solely involving conduction
electrons in EQ. (VII.27) drops out, we arrive at the following relations:
∂Seff
MF ⎛ (ξ rf )2 + Δ 2 ⎞
Δ0
=0 → ln ⎜ ⎟ = λ ≡ ξ 0f − ξ rf , (VII.29)
∂Δ π ⎜ W2 ⎟
⎝ ⎠
∂Seff
MF
2 −1
⎛Δ⎞ Δ
= 0 → tan ⎜⎜ r ⎟⎟ = 1 − = 1 − z = n fm . (VII.30)
∂ξ f
r
π ⎝ξf ⎠ Δ0
For a bare f-electron energy level ξ 0f much lower than the bottom of the conduction band, i.e., ξ 0f W , we
expect that the f-electron level is nearly half-filled so that n fm → 1− and tan −1 ( Δ ξ rf ) ≈ (π 2 ) . Hence, we
have Δ ξ rf , ξ rf → 0 , z 1 , and Δ 0 Δ ≡ z Δ 0 . In other words, we find an enormous renormalization of
the f-electron level from ξ 0f to ξ rf → 0 due to strong interaction U → ∞.
Using EQ. (VII.29) and the fact that ξ rf → 0 , we may define the Kondo temperature TK as follows:
⎛ (ξ rf )2 + Δ 2 ⎞ 2Δ 0
Δ0 ⎛ Δ ⎞ 2Δ 0 ⎛ k BTK ⎞
ln ⎜ ⎟ = ξ 0f − ξ rf , → ln ⎜ ⎟≡ ln ⎜ 0
⎟ ≈ − ξf . (VII.31)
π ⎜ W 2
⎟ π ⎝W ⎠ π ⎝ W ⎠
⎝ ⎠
−π ξ 0f 2 Δ0
Δ = k BTK = W e . (VII.32)
Due to the exponential dependence on the parameters, TK may vary from <~ 1 K to 102 K. The Kondo
temperature sets an important energy scale below which the strong hybridization of the conduction electrons
and the f-electron leads to resonant scattering, thereby giving rise to increasing resistivity below TK in metals
with magnetic impurities, as illustrated in Fig. VII.1.1 (a). In other words, for T < TK the low-energy
excitations at the Fermi surface become heavy-fermions and are more f-electron like. This effect becomes
more significant with decreasing temperature, leading to increasing resistivity. Moreover, because of the
large onsite Coulomb repulsion, conduction electrons cannot occupy the f-orbit until the f-electron hops off
to the Fermi surface, creating a slave boson, and then a succeeding electron can hop on again. This delay (as
illustrated in Fig. VII.1.1) is the reason why the resonance energy width is reduced from Δ0 to zΔ0 ≡ Δ << Δ0.
As mentioned earlier, many of the single Kondo impurity properties are also found in heavy fermion
materials that are pure crystalline intermetallic compounds, such as CuAl3, CeCu6, CeCu2Si2, and UPt3.
Newns and Read [Advances in Physics 36, 799 (1987)] have considered a simplified model that captures the
essence of the Kondo lattice. Specifically, they consider an idealized SU(N) lattice model in which the
degeneracy of the conduction bands is taken to be identical to that of the f-levels. Returning to the mean-field
Hamiltonian of the Anderson lattice model in EQ. (VII.7) and assuming that vmσ ( k ) = v δ mσ , we have
⎛ ⎞
( ) ( ) (
H MF = ∑ ξk ck†σ ckσ + ξ 0f ∑ f j†σ f jσ + v z1/ 2 ∑ fk†σ ckσ + h.c. + λ ∑ ⎜ z + ) ∑ †
f σ f σ − 1⎟ .
j j (VII.33)
kσ jσ kσ j ⎝ σ ⎠
Similar to the procedure that we have employed to investigate the problem of single Kondo impurities, we
minimize the Hamiltonian relative to λ and z, and obtain the constraint equation at each site:
z = 1− ∑
σ
†
fσfσ,
j j (VII.34)
Hence, EQ. (VII.33) represents a set of N degenerate hybridized bands (N being the total number of lattices
in the sample), and its energy spectrum can be derived by diagonalizing the mean-field Hamiltonian in EQ.
(VII.33) and using EQ. (VII.35), which yields (see Problem Set 6):
(ξ )±
2
r
+ ξk ⎛ ξ rf − ξk ⎞
ξ ± 2
f
= ⎜ ⎟ + zv . (VII.36)
2 2
k
⎝ ⎠
The spectrum in EQ. (VII.36) implies a narrow hybridized energy gap. Furthermore, the hybridized density
of states is given by:
dξ
( ) (
N (ξ ) = ∑ ⎡⎣δ ξ − ξk+ + δ ε − ξk− ⎤⎦ = d ξ N ( ξ ) ) ∫ dξ
δ ( ξ − ξk± )
k
⎛ 1 ξ rf − ξ ⎞
≈ N (0) ⎜ ± ⎟, (VII.37)
⎝2 ξk± ⎠
where N(0) denotes the density of states at the Fermi level in the absence of hybridization. From EQ.
(VII.37) we find that the hybridized density of states has strongly enhanced peaks on each side of the gap,
and the effective mass is also much enhanced at the Fermi level, leading to the heavy-fermion behavior.
Finally, we comment on an issue associated with changes in the symmetry of the Hamiltonian under
bosonization of a Kondo lattice. Specifically, the original Hamiltonian is invariant under the transformation:
c jσ → e iθ c jσ , (VII.38)
where θ is site invariant. In other words, EQ. (VII.38) implies that the Hamiltonian is invariant under a
global U(1) transformation. On the other hand, in the slave-boson formalism, we have the constraint
b†j b j + ∑
σ
†
f σ f σ =1
j j (VII.39)
at each site j. Therefore, the new Hamiltonian is invariant under the following transformation
iθ iθ j
bj → e j bj , f jσ → e f jσ , (VII.40)
where θ j is site-dependent. Equation (VII.40) implies that the Hamiltonian in the slave-boson formalism has
a local rather than a global U(1) gauge symmetry.
With the advances of modern technology in electronic device fabrication, a system with a localized
spin embedded between metal leads can be created artificially in a semiconductor quantum dot device, which
is a transistor-like device also known as an artificial atom because its electronic properties resembles those of
a real atom. The total number of electrons within the quantum dot can be controlled by a gate voltage. If the
number of electrons confined in the dot is odd, then the conductance measured between the two leads
increases due to the Kondo effect at low temperatures, because of strong coupling between the conduction
electrons in the leads and the odd spins in the dot. This temperature dependence associated with odd spins in
quantum dots is opposite to that found in the case of Kondo impurities in metals, as shown in Fig. VII.4.1 (a)
– (b). The reason is that electrons are forced to travel through the dot without having other electrical path
around it, and for conductance to occur between the leads and the dot, availability of electronic states in the
dot is essential. Therefore, hybridization of the conduction electrons in the leads with those in the quantum
dot due to the Kondo resonance enhances the conductance, and the degree of such hybridization increases
with decreasing temperature. In this context, when the quantum dot contains an even number of electrons, the
total spins add up to zero, so that the conductance continuously decreases with decreasing temperature. In
contrast, for magnetic impurities in metals, conduction electrons are scattered by the Kondo resonant states
that are induced by magnetic impurities, and the degree of scattering increases with decreasing temperature,
so does the resistivity. Overall, other than the temperature dependence of the conductivity, quantum dots
with odd electrons follow the same Kondo physics as magnetic impurities in metals, and the ratio of the
resistance relative to the zero-temperature resistance for all Kondo systems, R(T)/R(0), follow the same
temperature dependent function f(T/TK) determined by a single parameter TK.
Generally speaking, quantum dots have provided new opportunities to control the Kondo effect
empirically, and have expanded the research scope in regimes previously inaccessible with magnetic
impurities. For instance, the Kondo effect can also occur for magnetic impurities and quantum dots with a
total spin of 1 or higher. In the case of magnetic impurities, the spin is controlled by the electronic structure
of the system, whereas in quantum dots, the spin of the quantum dots can be controlled more easily by
applying a modest magnetic field (~ 1 Tesla), which is sufficient to force a transition in the dot between a
singlet (S = 0) and a triplet (S = 1) state. For comparison, the same type of transition can also occur in real
atoms in principle, but the field required would be on the order of 106 Tesla, which is completely
inaccessible in a real lab. The availability of such singlet-triplet transitions in quantum dots leads to
interesting new degrees of degeneracy because of the combined spin and orbital effects. Thus, new many-
body effects as the result of the degeneracy and exchange interactions can be studied in quantum dots.
2e2/h
Kondo Odd
Resistance
Conductance
effect spins
TK : 1 ~ 10 K
TK : 0.1 ~ 1 K
Typical
metal
Superconductor
Even spins
Tc
Temperature Temperature
Fig. VII.4.1 (a) As the temperature of a metallic system is lowered, a typical metal exhibits decreasing
resistance because of reduced electron-phonon scattering, and the resistance eventually saturates
at a finite value with T → 0 due to non-magnetic impurities. Some metals may undergo a
superconducting transition and reach zero resistance below a critical temperature Tc. However,
for metals containing magnetic impurities, the resistance may reach a minimum at the Kondo
temperature TK, and then rise again with decreasing temperature due to resonant scattering of
conduction electrons by the magnetic impurities. (b) In the case of quantum dot systems, the
localized spins on the dot may be controlled by a gate voltage. If the number of electrons
confined in the dot is odd, then the conductance between the two electrical leads connecting to
the dot increases with decreasing temperature for T < TK due to the Kondo effect at low
temperatures. On the other hand, for even electrons where the total number of spins adds up to
zero, the conductance continuously decreases with decreasing temperature.
The invention of the scanning tunneling microscopy (STM) in 1981 has enabled imaging,
characterization and manipulation of matter down to atomic precision. Since 1998, STM has been used to
image and construct magnetic impurities directly on the surface of noble metals, which is a new experimental
approach in contrast to previous research that only inferred the role of the Kondo effect indirectly from bulk
measurements of the resistivity and magnetic susceptibility. In addition to observing how one magnetic atom
such as cobalt or manganese on the surface of copper or gold may influence the surrounding electronic
density of states of the host metal, how multiple magnetic atoms interacting among each other through the
host metal as a function of separation have also been investigated. Moreover, artificial structures such as an
elliptical “stadium” consisting of cobalt atoms surrounding one cobalt atom in one of the two foci of the
stadium have been constructed on a copper surface by the STM technique [See H. C. Manoharan, C. P. Lutz
and D. M. Eigler, Nature 403, 512 (2000)]. It is found that the density of states of the cobalt atom at the focal
point reveals Kondo resonance. In particular, Kondo resonance is also observed at the other focal point
where no cobalt atom is present. This mirror image of the Kondo resonance is referred to as a “quantum
mirage”. In addition to studying the Kondo effect on metallic surfaces, STM has been used to deposit cobalt
clusters on carbon nanotubes and to study their effect on the electronic states of the nanotubes [J. Nygard, D.
H. Cobden and P. E. Lindelof, Nature 408, 342 (2000)].
The availability of scanning tunneling microscopy and quantum dot devices has enabled new
dimensions for studying the Kondo effect from different perspectives and unprecedented control. Some of
the new experimental capabilities have their counterparts in the conventional magnetic-impurity-in-metal
systems, and others are unique to the artificial nanostructures.
Investigations into the Kondo effect are far from complete. Among the important issues concerns the
so-called Kondo cloud, which refers to the conduction electrons involved in the spin-flip processes in VII.1.1
that combine to build the Kondo resonance. The Kondo cloud consists of all electrons that have interacted
with the same magnetic impurity. Since each electron of the Kondo cloud contains information about the
same impurity, they are effectively correlated.
The most fundamental issue to address for research on the Kondo effect is to understand the time
evolution of the many-body quantum state. Experimental challenges to achieve this understanding will
involve devising means to measure and control the Kondo cloud with proper time resolution. The Kondo
cloud also provides a possible mechanism to investigate the interactions between magnetic impurities. For
instance, we would like to understand how two many-body states formed around two separated localized
magnetic moments merge. Additionally, a well controlled study of interacting localized spins could provide
us with a new perspective on extended Kondo systems, such as spin glasses. The basic technology for
fabricating interacting Kondo systems now exists. Therefore, we may be cautiously optimistic that better
understandings can be derived from new development in experiments.
Further Readings
VIII.1. Limitations of Fermi Liquid Theory & the Tomonaga-Luttinger Liquid Theory
In our previous discussion we have noted various signatures of the Fermi liquid theory, including the
effective mass description, the effective magnetic moment for spin susceptibility, zero sound propagation,
and the linear temperature dependence in the electronic specific heat. Another important signature commonly
used to distinguish whether a system exhibits Fermi liquid behavior is the quasiparticle residue (Zk) at the
Fermi level, which refers to the magnitude of discontinuity in the quasiparticle momentum distribution
function at the Fermi momentum kF. This discontinuity is associated with well defined eigenvalues in the
quasiparticle spectra, and therefore a Fermi liquid generally has a discontinuity in the momentum distribution
function at the Fermi level.
To understand the physical significance of the quasiparticle residue Zk, we consider an interacting
fermion system with a complex proper self-energy Σ* = Σ*R + iΣ*I , where the imaginary part of the proper
self-energy represents the damping rate of quasiparticles. If we assume Σ*I Σ*R and solve for the pole of
the Green function from the Dyson’s equation:
ω = ε k ≡ ε k + Σ*R ( k , ε k ) . (VIII.2)
On the other hand, the solution to first order in Σ*I can be found by defining ω = ε k + ω1 and by performing
Taylor’s expansion of Σ*R in EQ. (VIII.1), which yields:
∂Σ*R ( k , ε k ) ∂Σ*R ( k , ε k )
ω − ⎡⎣ε k + Σ ( k , ε k ) ⎤⎦ − ω1
*
− iΣ ( k , ε k ) = ω1 − ω1
*
− iΣ*I ( k , ε k ) = 0 ,
∂ω ∂ω
R I
i Σ*I ( k , ε k )
⇒ ω1 = . (VIII.3)
(
1 − ∂Σ*R ∂ω )
Therefore, under finite quasiparticle interactions as characterized by the proper self-energy Σ*, the pole of the
Green function G is given by
i Σ*I ( k , ε k ) i
ω = ε k + ω1 = ε k + ≡ εk − , (VIII.4)
(
1 − ∂Σ *
R
∂ω ) τk
where τk has the physical meaning of the quasiparticle lifetime. Hence, the Green function becomes
1 1
G( k , ω ) = =
ω − (ε k + Σ *
R ) − ω ( ∂Σ
1
*
R
∂ω − iΣ) *
I
ω (1 − ∂Σ ∂ω ) − ε k (1 − ∂Σ*R ∂ω ) − iΣ*I
*
R
=
(
1 1 − ∂Σ*R ∂ω )≡ Zk
=
Z k (ω − ε k )
−i
Z kτ k
−1
. (VIII.5)
(ω − ε ) + iτ
k
−1
k (ω − ε ) + iτ
k
−1
k (ω − ε ) k
2
+ τ k−2 (ω − ε ) k
2
+ τ k−2
1 1 Z kτ k−1 1
A( k , ω ) ≡ − Im [G ( k , ω )] = − Im [Gincoh ( k , ω )] . (VIII.6)
π π ( ω − ε k ) + τ k−2 2
π
∫ dω A( k , ω ) = 1 = Z k
+ (1 − Z k ) . (VIII.7)
pole incoherent part
Schematically, the quasiparticle spectra for the non-interacting and interacting systems are depicted
in Fig. VIII.1.1 (a) – (b), and the corresponding ground-state density distribution functions
nkσ ≡ 0 ak†σ akσ 0 are shown in Fig. VIII.1.2 (a) – (b). We note that the peak height of A(k,ω) at ω = ε k is
equal to ( Z kτ k π ) according to EQ. (VIII.6), and there is a discontinuity of a magnitude Zk in nkσ at k = kF.
Consequently, a fermion system is considered to be a Fermi liquid if empirically there is a peak in A(k,ω)
and a finite discontinuity in nkσ at kF, implying that Z k ≠ 0 .
(a) (b)
A(k,ω) A(k,ω)
Zkτk /π
ω ω
εk εk
Fig. VIII.1.1 The spectral density function A(k,ω) for (a) non-interacting and (b) interacting fermion
systems. The presence of interaction has the effect of shifting the pole and reducing the
peak height Zkτk at the pole energy.
There are occasions when more than one solution to a given proper self-energy Σ*(k,ω) exists for
each k. In this case, there is no one-to-one correspondence between the ground state and the excitation state.
However, as long as the pole energy ε k is unique and if the imaginary parts Σ I τ k−1 so that the multiple
solutions cannot be simultaneously accessed during the characteristic damping time τk, the fermion system
can still be considered as a Fermi liquid.
(a) (b)
nkσ nkσ
1
Zk
0 k k
kF kF
Fig. VIII.1.2 The fermion density distribution function nkσ for (a) non-interacting and (b) interacting
systems at T = 0. The presence of interaction results in a reduced discontinuity Zk at the
Fermi momentum kF.
(a) (b)
εk ky
εF
kx
−π/a 2kF π/a
k
− kF 0 kF
Fig. VIII.1.3 Nested Fermi surfaces at the wave vector q = 2 kF for (a) a one-dimensional conductor
and (b) a two-dimensional fermion system with a square Fermi surface, with a
representing the square lattice constant. Both systems are candidates for non-Fermi
liquid behavior.
On the other hand, systems with perfectly “nested” Fermi surfaces generally exhibit non-Fermi
liquid behavior. One of the most celebrated examples is the case of one-dimensional conductors. Consider
the dispersion relation of a one-dimensional conductor shown in Fig. VIII.1.3 (a). There are only two points
on the Fermi surface, and the wave vector q = 2 kF connects points on the entire Fermi surface, so that the
system is perfectly nested, and there is diverging response of the one-dimensional conductor at q = 2 kF.
Moreover, it can be shown that there are strong interactions between the left-moving and right-moving
wavefunctions, leading to bosonic excitations that involve spin-charge separation and are therefore
fundamentally different from quasiparticle excitations of Fermi liquids. In the language of the effective field
theory, the effective action of the system after bosonization involves two coupling coefficients g1 and g2 that
contain mixtures of spin and charge degrees of freedom for the left-moving and right-moving wavefunctions.
It is found that the Fermi liquid fixed point (g1, g2) = (0,0) is unstable in the renormalization group flow
interaction strengths and spin-charge separation is found to be a stable fixed point, as schematically
illustrated in Fig. VIII.1.4. We shall discuss the Luttinger liquid theory in the end of Part VIII.
g2
(g , g )
*
1
*
2
g1
(0,0)
Fig. VIII.1.4. The “flow patterns” in the renormalization group diagram of a one-dimensional
conductor, where g1 and g2 are coupling constants for the charge and spin mixtures of
left-moving and right-moving wavefunctions, (0,0) represents the (unstable) Fermi
liquid fixed point, and ( g1* , g 2* ) is the (stable) Luttinger liquid fixed point.
Finally, we note that the fractional quantum Hall (FQH) states associated with strongly correlated
electronic configurations of two-dimensional electron gas under large magnetic fields also deviate in a
fundamental way from the Fermi liquid theory. The FQH states generally involve complex ground states and
the excitations consist of quasiparticles that acquire fractional charges and obey fractional statistics. In the
following, we give a general derivation for the Luttinger liquid theory (also known as the Tomonaga-
Luttinger liquid theory), which is one of the celebrated examples of non-Fermi liquid systems.
To derive the Tomonaga-Luttinger liquid theory for one-dimensional conductor, we begin with the
Hamiltonian for one dimensional electrons:
where ti j denotes the hopping energy coefficient between site i and site j. We are primarily interested in the
low-energy physics close to the Fermi surface, so we shall restrict our discussion to linearized systems
around the Fermi surface. In this context, we consider the dispersion relation of the one-dimensional
conductor as depicted in Fig. VIII.1.3 (a), which corresponds to a simplified system with the hopping
coefficient restricted to the nearest neighbor hopping t ≡ ti ,i ±1 so that ε k = −2t cos ( ka ) inEQ. (VIII.8) and a
is the lattice constant. Therefore, the Hamiltonian for the linearized motions of particles around the Fermi
surface becomes:
−α k − k F −α − k − kF
H 0 = ∑ vF ( k − k F ) c1†k σ c1kσ e + ∑ vF ( −k − k F ) c2†k σ c2 kσ e , (VIII.9)
kσ kσ
where vF = ( ∂ε k ∂k )k = k is the Fermi velocity, c1kσ ( c2kσ ) refers to the operator for right- (left-) moving
F
waves, and α is a cutoff on spectrum. We may rewrite EQ. (VIII.9) into a form in coordinate space by the
following transformation:
∞ dk
ψ 1σ ( x ) = ∫ ei k x c1kσ (right-moving)
−∞
2π
∞ dk
ψ 2σ ( x ) = ∫ e − i k x c2 kσ (left-moving)
−∞
2π
so that we have
σ
∫ 0
σ
∫
0
≡ vF ∑ dx ψ σ† ( x ) A 0ψ σ ( x ) ,
L
σ
∫
0
(VIII.10)
⎛ −i ∂ x − k F 0 ⎞ ⎛ ψ 1σ ( x ) ⎞
where A0 ≡ ⎜ ⎟, ψσ ( x) ≡ ⎜ ⎟,
⎝ 0 i∂ x − k F ⎠ ⎝ψ 2σ ( x ) ⎠
and L is the length of the one-dimensional conductor.
Next, we want to include interactions among the particles, and the simplest interactions are:
σσ
g11 ( x − y ) ρ1σ ( x ) ρ1σ ′( y ) ,
′ σσ
g12 ( x − y ) ρ1σ ( x ) ρ 2σ ′( y ) ,
′ σσ
g 22 ( x − y ) ρ 2σ ( x ) ρ 2σ ′( y ) ,
′
involves the “head-on collisions” of right- and left-moving particles, so is of interest to us. Thus, we may
σσ ′ σσ ′ σσ ′
ignore contributions from the coefficients g11 and g 22 by adding an interaction term involving only g12 :
H = H0 +∑ ( x − y ) ρ1σ ( x ) ρ 2σ ′( y ) .
L L
∫ ∫
σσ ′
dx dy g12 (VIII.11)
0 0
σσ ′
To begin, we first assume a simplified case with all particles being spin polarized so that we can drop
the spin indices for now. This is equivalent to considering a spinless fermion system. We further assume that
only on-site interaction is relevant, so that g12 (x−y) = gδ (x−y). Hence,EQ. (VIII.11) becomes:
dxψ †( x ) H 0 ψ ( x ) + g dx ρ1( x ) ρ 2( x ) .
L L
H = ∫ 0 ∫
0
(VIII.12)
Z = ∫ Dψ 1† Dψ 1 Dψ 2† Dψ 2 e
(
S ψ 1† ,ψ 1 ,ψ 2† ,ψ 2 ) , (VIII.13)
( ) ∫
β β
dx ρ1( x ) ρ 2( x ) ≡ S 0 + S I
L L
S =− ∫ 0
dτ ∫ 0
dx ψ †∂τψ + vFψ † A 0ψ − g
0
dτ ∫ 0
β L ⎛ ∂τ − ivF ∂ x − vF k F ⎞ 0
=− ∫0
dτ
⎝
∫
0 0
dxψ † ⎜ ⎟ψ + S I
∂τ + ivF ∂ x − vF k F ⎠
β L ⎛ D1 0 ⎞
≡ − ∫ dτ ∫ dxψ † ⎜ ⎟ψ + S I . (VIII.14)
0 0
⎝ 0 D2 ⎠
(
exp ( S I ) = exp − g ∫ dτ ∫ dx ρ1 ρ 2 ,
β
0
L
0 )
which can be evaluated through the Hubbard-Strotonovich transformation by introducing two auxiliary
boson fields ϕ and ϕ* so that we obtain the following identity:
⎡ β L ⎞⎤ ⎛1
∫ Dϕ Dϕ exp ⎢ − ∫ dτ ∫ dx ⎜ ϕ *ϕ + iϕ * ρ1 + iϕρ 2 ⎟ ⎥
*
0 0
⎣ ⎝g ⎠⎦
⎡ β 1 ⎤
( )
β
(ϕ + ig ρ1 ) − g ∫0 dτ ∫0 dx ρ1 ρ 2 ⎥
L L
= ∫ Dϕ * Dϕ exp ⎢ − ∫ dτ ∫ dx ϕ * + ig ρ 2
0 0
⎣ g ⎦
⎡ β 1 ⎤ β
ϕ ⎥ exp ⎡ − g ∫ dτ ∫ dx ρ1 ρ 2 ⎤
L L
= ∫ Dϕ * Dϕ exp ⎢ − ∫ dτ ∫ dx ϕ *
⎣ 0 0 g ⎦ ⎣ 0 0 ⎦
=⎜
⎛πg ⎞
⎟
⎝ βL ⎠ 0
β
0(
L
exp − g ∫ dτ ∫ dx ρ1 ρ 2 ) (VIII.15)
n/2
⎛ n ⎞ ⎛ 1 ⎞ n ⎛ 2π ⎞ ⎛ λi λ j ⎞
∫ ⎜⎝ ∏ dyi⎟ exp ⎜ −
⎝ 2
y C y
i ij j + λi j⎟=∏ ⎜
y ⎜
⎠ i =1 ⎝ Ci j
⎟⎟ exp ⎜⎜ − ⎟⎟
. (VIII.16)
i =1 ⎠ ⎠ ⎝ 2Ci j ⎠
Substituting EQ. (VIII.15) into EQ. (VIII.14) and ignore the constant coefficient (π g/βL), we rewrite the
action into the following form:
β L ⎡⎛ D1 0 ⎞ ⎛ϕ* 0 ⎞⎤ 1 β L
∫ ∫ dxψ † ⎢⎜ ⎟ ⎥ψ − ∫0 dτ ∫0 dx ϕ ϕ .
*
S =− dτ ⎟+i⎜ (VIII.17)
0 0
⎣⎝ 0 D2 ⎠ ⎝0 ϕ ⎠⎦ g
Integrating out the fermion sector in EQ. (VIII.17) exactly, we find that the partition function becomes
Z = ∫ Dϕ * Dϕ Dψ † Dψ e
(
S ϕ * ,ϕ ,ψ † ,ψ ) = Dϕ * Dϕ e S ff (ϕ* ,ϕ ) ,
∫ (VIII.18)
e
1
( ) { ( )} + Tr {ln (D
β
+ iϕ )} .
L
g∫ ∫
S eff ϕ * , ϕ = − dτ dx ϕ *ϕ + Tr ln D1 + iϕ * 2 (VIII.19)
0 0
In EQ. (VIII.19) we have used the identity detM = exp (Tr {ln M}), where M is a matrix. Furthermore, the
positive sign associated with the matrix in EQ. (VIII.19) is the result of Grassman algebra involved in
integrating out the fermion operators, as discussed in Part II.5. Therefore, we have effectively bosonized the
action. We remark here that the concept of bosonization is also well known in relativistic field theory in the
operator formalism, including abelian bosonization that maps fermions to the soliton sector of the sine-
Gordon model [S. Coleman, Phys. Rev. D 11, 2088 (1975)] and non-abelian bosonization [A. M. Polyakov
and P. Wiegmann, Phys. Lett. 131B, 121 (1983); E. Witten, Commun. Math. Phys. 92, 455 (1984)].
ϕ * ⇒ ϕ1 and ϕ ⇒ ϕ2
S0 = ∑ i =1 Tr ⎣⎡ ln (Di ) ⎦⎤ .
2
(VIII.20)
⎛ 1 0 ⎞ ⎛ϕ* 0 ⎞ −1
A ≡ vF A 0 + ∂τ ⎜ ⎟ +i⎜ ⎟ ≡ −G , (VIII.21)
⎝ 0 1⎠ ⎝ 0 ϕ ⎠
⎛ ψ 1ψ 1† 0 ⎞ ⎛ G1 0 ⎞
where G = −⎜ ⎟≡⎜ , (VIII.22)
⎜ 0 ψ 2ψ 2† ⎟ ⎝ 0 G2 ⎟⎠
⎝ ⎠
G1 = − ( −ivF ∂ x + ∂τ − vF k F + iϕ1 ) ,
−1
and (VIII.23)
G2 = − ( ivF ∂ x + ∂τ − vF k F + iϕ 2 ) ,
−1
(VIII.24)
we find that the Green functions G1 and G2 satisfy the following equations:
If we take ϕ1 = ϕ 2 = 0 , we obtain the free fermion propagators G10 and G20 in a frozen background of boson
fields, which satisfy the following equations:
1 1
G10( k , ωn ) = = , (VIII.29)
iωn − vF ( k − k F ) iωn − ε 1( k )
1 1
G20( k , ωn ) = = . (VIII.30)
iωn − vF ( − k − k F ) iωn − ε 2( k )
Restoring the spectral cutoff α, the free fermion Green function G10( x,τ ) becomes
i k x − i ωn τ
∞ dk 1 e
G10( x,τ ) = ∫
−∞ 2π β
∑ iω − vF ( k − k F )
e
−α ( k − k F )
ωn n
− i ωn τ
∞ dk 1 e
=∫
−∞ 2π
ei k x e
−α ( k − k F )
β
∑ iω − vF ( k − k F )
ωn n
( )
f vF ( k − k F ) exp ⎡⎣ − vF ( k − k F )τ ⎤⎦
dk − vF ( k − k F )τ 0 dk
e(
kF −α k − k F i k + k F ) x −α k
(for β vF k F 1) ≈∫ ei k x e e =∫ e e − vF k τ
−∞ 2π −∞ 2π
ei k F x 1
= , (VIII.31)
2π ix − vFτ + α
and
ei k F x 1 ei k F x 1
G10( x, t ) = i = . (τ ↔ it ) (VIII.32)
2π ix − ivF t + α 2π x − vF t − iα
The pole x = vF t in EQ. (VIII.32) corresponds to classical right-moving waves. Similarly, the free fermion
Green function for left-moving waves is
e − i kF x 1 e− i kF x 1
G20 ( x, τ ) = and G20 ( x, t ) = − , (τ ↔ it ) . (VIII.33)
2π −ix − vFτ − α 2π x + vF t + iα
Next, with finite interactions so that ϕ j ≠ 0, ( j = 1, 2 ) , the Green functions satisfy the equations
(D j )
+ iϕ j G j ( x,τ ; x′, τ ′ ) = −δ ( x − x′ ) δ (τ − τ ′ ) , ( j = 1, 2 ) . (VIII.34)
Hence, for Gj0 being the free fermion propagator, the Green function Gj with ϕ j ≠ 0 becomes
( ) ( ).
( )
− i Φ j x ,τ ;ϕ j + i Φ j x′,τ ′;ϕ j
G j x,τ ; x′, τ ′; ϕ j = G j 0 ( x − x′,τ − τ ′ ) e (VIII.36)
Inserting EQ. (VIII.36) back to the differential equations in EQ. (VIII.34) and using EQs. (VIII.27) and
(VIII.28), we obtain the differential equations for the phases Φj of the Green functions Gj:
1 dk
∑∫
∞
Φ1( x, τ ) = Φ1( k , ωn ) e
i k x − i ωn τ
, (VIII.39)
β ωn
−∞
2π
1 dk
∑∫
∞
ϕ1( x, τ ) = ϕ1( k , ωn ) ei k x −i ωn τ , (VIII.40)
β ωn
−∞
2π
we find
ϕ1( k , ωn ) ϕ 2( k , ωn )
Φ1( k , ωn ) = − , Φ 2( k , ωn ) = − . (VIII.41)
iωn − vF k iωn + vF k
The physical significance of the phases Φj is that they specify the Green functions Gj in the bosonic
field configuration (ϕ1,ϕ2). It is worth noting that in reality the zero interaction solutions to the Green
functions must break down because in one dimension, fluctuations and interactions are very strong. Thus, the
physical observable Green functions are given by:
(ϕ*, ϕ )
∫ Dϕ Dϕ G e
* Seff
G ϕ ≡
j
≡ Zϕ−1 ∫ Dϕ * Dϕ G j e
Seff (ϕ*, ϕ ) , (VIII.42)
j
S (ϕ , ϕ ) *
∫ Dϕ Dϕ e
* eff
S (ϕ , ϕ ) *
Zϕ ≡ ∫ Dϕ Dϕ e * eff
, (VIII.43)
( )
and S eff ϕ * , ϕ satisfies EQ. (VIII.19):
1
( ) { ( )} + Tr {ln (D
β
+ iϕ )} .
L
g∫ ∫
S eff ϕ * , ϕ = − dτ dx ϕ *ϕ + Tr ln D1 + iϕ * 2 (VIII.19)
0 0
2
S 0 = ∑ Tr ln (D j ) . { } (VIII.44)
j =1
( )
To evaluate S eff ϕ * , ϕ , we first consider the terms ∑ j
{ }
Tr ln (D j + iϕ j ) . Noting that (D j + iϕ j )
( )
is related to the Green function G j x,τ ; x′,τ ′; ϕ j via EQ. (VIII.34), and that taking the trace corresponds to
evaluating ln (D j + iϕ j ) in the x′ → x and τ ′ → τ limit, we write
⎛ D j + iϕ j ⎞
ln (D j + iϕ j ) = ln (D j ) + ln ⎜ ⎟,
⎝ Dj ⎠
⎛ D j + iϕ j ⎞ dλ
= − iϕ j ∫ d λ G j ( x,τ ; x′→ x, τ ′→τ ; ϕ j λ ) .
1 1
and ln ⎜ ⎟ = iϕ j ∫0 (VIII.45)
⎝ Dj ⎠ D j + iϕ j λ 0
Taking τ ′ = τ and x′ = x + η where η → 0+ and using EQs. (VIII.31) and (VIII.33), we have
± i k F ( x − x′)
e 1 e ∓ i k Fη 1
G j 0( x,τ ; x′= x + η ,τ ) = ± =∓ =∓ , (VIII.46)
2π i ( x − x′ ) 2π iη 2π iη
where the upper (lower) sign corresponds to j = 1 (j = 2). Moreover, the phase difference in EQ. (VIII.36)
becomes
∂Φ j
−i Φ j ( x,τ ; ϕ j λ ) + i Φ j ( x + η ,τ ; ϕ j λ ) = iη . (VIII.47)
∂x
Consequently, by inserting EQs. (VIII.46) and (VIII.47) into EQ. (VIII.36) and symmetrizing η, we obtain:
( )
Collecting all terms for S eff ϕ * , ϕ = S eff (ϕ1 , ϕ 2 ) and using EQs. (VIII.41), (VIII.45), (VIII.48) and
(VIII.49), we find that
1 ⎧ ⎛ D1 + iϕ * ⎞ ⎫ ⎧ ⎛ D + iϕ ⎞ ⎫
( )
β L
∫
S eff ϕ * , ϕ − S0 = −
g 0 ∫ 0
dτ dx ϕ *ϕ + Tr ⎨ln ⎜
⎩ ⎝ D1 ⎠ ⎭
⎟ ⎬ + Tr ⎨ln ⎜ 2
⎩ ⎝ D2 ⎠ ⎭
⎟⎬
1 β i 1 β ⎛ ∂Φ1 ∂Φ 2 ⎞
= − ∫ dτ ∫ dx ϕ1ϕ 2 +
L
∫ d λ ∫ dτ ∫ dx ⎜ λϕ1
L
− λϕ 2 ⎟
g 0 0 2π 0 0 0
⎝ ∂x ∂x ⎠
1 β ⎡1 dk i ( k x −ω τ ) ⎤ ⎡ 1 dk i ( k x −ω τ ) ⎤
= − ∫ dτ ∫ dx ⎢ ∑ ∫ 1 ϕ1( k1 , ωn1 ) e 1 n1 ⎥ ⎢ ∑ ∫ 2 ϕ 2( k2 , ωn 2 ) e 2 n 2 ⎥
L
g 0 0
⎣ β ωn1 2π ⎦ ⎣ β ωn 2 2π ⎦
iω ′ ⎝
n1
n1
− vF k1′ ⎠
⎟e ⎥
⎦
1
1
i β ⎡1 dk ⎤⎡1 dk ′ ⎛ ϕ ( k ′ , ω′ ) ⎞ ⎤
∑ ∫ 2π2 ϕ2( k2 , ωn 2 ) ei ( − k2 x−ωn 2τ ) ⎥ ⎢ β ∑′ ∫ 2π2 ( −ik2′ ) ⎜ − iω2 ′ 2+ v nk2 ′ ⎟ ei ( − k2 x −ωn 2τ ) ⎥
L
4π ∫ ∫
′ ′
+ dτ dx ⎢
0 0
⎣β ωn 2 ⎦ ⎣ ωn 2 ⎝ n2 F 2 ⎠ ⎦
1 1 1 1 dk ⎡ ϕ ( k , ω ) ϕ ( − k , −ω ) ϕ ( k , ω ) ϕ ( − k , −ω ) ⎤
[ϕ ( k , ω ) ϕ ( −k , −ω )] +
dk
=−
g β
∑ ∫ 2π
1
4π β
n ∑ ∫2
2π
k ⎢
iω − v k
n
−
iω + v k
⎥
1 n 1 n 2 n 2 n
ωn ωn ⎣ n F n F ⎦
1 1 1
[ϕ ( k , ω ) ϕ ( −k , −ω )] − ∑ ∫ A ( k , ω ) ϕ ( k , ω ) ϕ ( −k , −ω )
dk dk
≡−
g β
∑ ∫ 2π
1
β
n
2π
2 n 1 n 1 n 1 n
ωn ωn
1 dk
−
β
∑ ∫ 2π
A ( k , ω ) ϕ ( k , ω ) ϕ ( − k , −ω ) ,
2 n 2 n 2 n
(VIII.50)
ωn
(k 4π ) (k 4π )
where A1( k , ωn ) = − and A2( k , ωn ) = . (VIII.51)
iωn − vF k iωn + vF k
In addition, from EQ. (VIII.41) the phase in EQ. (VIII.42) can be expressed by
1 dk ⎡ ei ( k x −ωn τ ) − ei ( k x′−ωn τ ′) ⎤
i Φ1( x,τ ; ϕ1 ) − i Φ1( x′, τ ′; ϕ1 ) = −i
β
∑ ∫ 2π ϕ ( k , ω ) 1 n ⎢
iωn − vF k
⎥
ωn ⎢⎣ ⎥⎦
1 dk
≡ −i
β
∑ ∫ 2π
ϕ ( k , ω ) J ( k , ω ; x, τ , x′, τ ′ ) ,
1 n 1 n
ωn
1 dk 1
= −i
β
∑ ∫ 2π 2
⎡⎣ϕ1( k , ωn ) J1( k , ωn ; x, τ , x′, τ ′ ) + ϕ1( − k , −ωn ) J1*( k , ωn ; x,τ , x′,τ ′ ) ⎤⎦ .(VIII.52)
ωn
Hence, from EQs. (VIII.42) and (VIII.50) – (VIII.52) also noting that ϕ1( − k , −ωn ) = ϕ 2( k , ωn ) , the
expectation value of G1 becomes
1 ⎡ 1 dk ⎧ 1 ⎫ ⎤
= ∫ Dϕ1Dϕ2 G10 exp ⎢⎣ − β ∑ ∫ 2π ⎨⎩ϕ 1
ϕ 2 + ϕ1 A1ϕ 2 + ϕ 2 A2ϕ1 +
i
2
(ϕ J
1 1
+ ϕ 2 J1* ) ⎬ + S0 ⎥ . (VIII.53)
Zϕ ω n
g ⎭ ⎦
1
Introducing D −1 ≡ + A1 + A2 , the exponent in G1 ϕ can be rewritten into
g
1 dk ⎧⎛ i ⎞ ⎛ i ⎞ D ⎫
∑ ∫ 2π ⎨⎜⎝ ϕ
2
S0 − 1
+ DJ1* ⎟ D −1 ⎜ ϕ 2 + DJ1 ⎟ + J1 ⎬ ,
β ωn ⎩ 2 ⎠ ⎝ 2 ⎠ 4 ⎭
1 ⎛ 1 dk D ⎞
∑ ∫ 2π J1 ⎟ ≡ G10 ( x − x′, τ − τ ′ ) exp ⎡⎣ −Q( x, τ , x′,τ ′ ) ⎤⎦ ,
2
G1 ϕ = G10 Zϕ exp⎜ − (VIII.54)
Zϕ ⎝ β ωn 4 ⎠
where
1 1 ⎡ 2 ⎤
2 (
1 − cos [ k ( x − x′ ) − ω (τ − τ ′ )]) ⎥
dk dk
∑ ∫ 2π ∑ ∫ 2π
2
Q( x, τ , x′, τ ′ ) ≡ D J1 ≡ D⎢ (VIII.55)
4β 4β ⎢⎣ ωn + ( vF k )
2 n
ωn ωn ⎥⎦
is analogous to the Debye-Waller factor in Bragg diffraction due to thermal fluctuations. In addition, from
EQ. (VIII.51) we have
1 k ⎛1 1 ⎞ 1 k ⎛ 2v F k ⎞
D −1 = − ⎜ − ⎟ = − ⎜⎜ ⎟
g 4π ⎝ iωn − vF k iωn + vF k ⎠ g 4π ( iωn ) − ( vF k ) ⎟
2 2
⎝ ⎠
1 ⎛ ( iωn ) − ( vF k ) − ( vF k ) ( g 2π vF ) ⎞ 1 ⎛ ( iωn ) − ( vF k ) (1 + 2 g ) ⎞
2 2 2 2 2
= ⎜ ⎟≡ ⎜ ⎟
g⎜ ( ) ( ) ⎟ g⎜ ( ) ( ) ⎟
2 2 2 2
⎝ iω n
− v F
k ⎠ ⎝ iω n
− v F
k ⎠
1 ⎛ ( iωn ) − ( uk ) ⎞
2 2
≡ ⎜ ⎟, (VIII.56)
g ⎜ ( iωn ) − ( vF k ) ⎟
2 2
⎝ ⎠
We further note that the physical significance of u corresponds to the velocity of the bosonic mode.
⎧ 1
dk ⎪ ⎛ 1 ⎞ ⎪⎫
1
⎟ ⎬ [1 − cos ( k x )]
g
Q ( x, 0 ) = − ⎨ ∑⎜
2 ∫ 2π ⎪ 2ε β ω iω − εk
−
iωn + ε k ⎠ ⎭
⎩ k n ⎝ n ⎪
⎧ 1 ⎛
dk ⎪ 1 1 ⎞ ⎪⎫
⎜ βε k − − βε k ⎟ ⎬ [1 − cos ( k x )] ,
g
2 ∫ 2π ⎪ 2ε
= ⎨ (VIII.58)
⎩ k ⎝ e −1 e − 1 ⎠⎭ ⎪
where uk ≡ εk, and we have used one of the following relations (for bosons):
1 e
iωnη
1 ⎛ π ( 2n + 1) ⎞
lim
β
∑ iω −x
= βx
+1
⎜ for fermions, ωn = β ⎟,
η →0 +
ωn n
e ⎝ ⎠
iωnη
1 e 1 ⎛ 2π n ⎞
lim
β
∑ iω −x
=− βx
−1
⎜ for bosons, ωn = β ⎟ . (VIII.59)
η →0 +
ωn n
e ⎝ ⎠
At T = 0, the first term in EQ. (VIII.58) contributes to k < 0 and the second term contributes to k > 0, so that
g⎧ 0 dk ( −1) ∞ dk 1 ⎫
Q ( x, 0 ) = ⎨ ∫−∞ [1 − cos ( k x )] + ∫0 [1 − cos ( k x )]⎬
2 ⎩ 2π 2uk 2π 2uk ⎭
g 2 ⎧ ∞ ⎡1 − cos ( k x ) ⎤ −Λk ⎫
=
4π 2u ⎩ ∫0
⎨ dk ⎢ ⎥ e ⎬ ≡ 2ν H ( x ) , (VIII.60)
⎣ k ⎦ ⎭
where
⎧ ∞ ⎡1 − cos ( k x ) ⎤ −Λk ⎫ g
H ( x ) ≡ ⎨ ∫ dk ⎢ ⎥e ⎬ and ν≡ ,
0
⎩ ⎣ k ⎦ ⎭ 8π u
∂H ( x )
∂x
= {∫
0
∞
dk sin ( kx ) e −Λk = ⎜ } ⎛ x ⎞
2 2 ⎟
⎝ x +Λ ⎠
⇒ H ( x) =
1
2
ln ⎜
⎛ x2 + Λ2 ⎞
⎝ Λ
2 ⎟,
⎠
we obtain
ν
⎛ x2 + Λ2 ⎞
Q( x, 0 ) = ln ⎜ 2 ⎟ . (VIII.61)
⎝ Λ ⎠
Therefore,
i k F ( x − x′ ) ν
e ⎛ Λ2 ⎞
G1 ( x − x′, 0 ) = ⎜⎜ ⎟ . (VIII.62)
ϕ
2π [i ( x − x′ ) + α ] ⎝ ( x − x′ ) + Λ ⎟⎠
2 2
In the absence of interaction so that g = 0 and ν = 0, EQ. (VIII.62) recovers the form of the free propagator.
The extra factor in EQ. (VIII.62) in the presence of finite interaction represents the particle-hole “cloud” that
an electron drags along as it moves. The cloud is associated with bosonic excitations that prevent coherent
motion of electrons in a one-dimensional system. Specifically, the propagator G1 ϕ reflects the singular
nature of the forward scattering in one dimension while the propagator G2 ϕ
represents that of the back
scattering.
Next, we relax the equal time propagator condition and consider a more general function Q( x,τ ≠ 0 )
so that from EQ. (VIII.57)
dk ⎧
⎪ 1 ⎛ 1 1 ⎞ ⎫⎪
⎟ ⎬ [1 − cos ( k x − ωnτ )]
g
Q ( x, τ ) = −
2u ∫ 2π ⎩
⎨ ∑ ⎜ iω −
⎪ 2k β ωn ⎝ n
− uk iωn + uk ⎠ ⎭⎪
g 1
{−n ( uk ) [1 − cos ( k x + iukτ )] + n ( −uk ) [1 − cos ( k x − iukτ )]} ,
∞ dk
2u 2π ∫
=− B B (VIII.63)
−∞ k
Defining X ± ≡ x ± iuτ and introducing a cutoff momentum Λ−1, we rewrite EQ. (VIII.63) into the following:
g 1
e −Λk {[1 − cos ( k X + )] + [1 − cos ( kX − )]}
∞ dk
Q ( x, τ ) =
2u 2π ∫ 0 k
g ⎡ ( X +2 + Λ 2 )( X −2 + Λ 2 ) ⎤ ⎡ ( X +2 + Λ 2 )( X −2 + Λ 2 ) ⎤
= ln ⎢ ⎥ = ν ln ⎢ ⎥. (VIII.64)
8π u ⎣⎢ Λ4 ⎦⎥ ⎢⎣ Λ4 ⎦⎥
Hence, the corresponding Green function becomes
ν
⎡ Λ4 ⎤
( Gν )ϕ = Gν 0 ⎢ 2 2 ⎥
, (VIII.65)
⎢⎣ ( X + + Λ )( X − + Λ ) ⎥⎦
2 2
which contains branch-cut singularities for finite ν and is the same for both left- and right-moving modes.
To examine whether the Fermi surface discontinuity (Zk) survives under finite interaction in one-
dimension, we want to evaluate the particle distribution function in momentum space, nk, which is directly
related to the Fourier transform of the equal time Green function G(x,0):
nk = 1 − ∫ dx e −i k x G ( x, 0 ) . (VIII.66)
ei ( kF − k ) x ⎛ Λ 2 ⎞ sgn ( k − k F ) ∞ sin ( k − k F x ) ⎛ Λ 2 ⎞
ν ν
−i ∞
G ( x, 0 ) =
∫ dx e 2π ∫−∞ ∫0 dx
−i k x
dx ⎜ 2 2 ⎟
= ⎜ x2 + Λ2 ⎟
x ⎝ x +Λ ⎠ π x ⎝ ⎠
sgn ( k − k F )
=
π
{Λ k − k F
a1 (ν , Λ k − k F ) + ( Λ k − k F )
2ν
a2 (ν , Λ k − k F )} , (VIII.67)
where a1 and a2 are smooth functions of ν and Λ k − k F . [For further details, you may refer to D. K. D. Lee
and Y. Chen, J. Phys. A: Math. Gen. 21, 4155 (1988).] In the event that Λ k − k F 1 and 2ν < 1, EQ.
(VIII.67) may be approximated by
sgn ( k − k F )
(Λ k − k )
2ν
∫ dx e
−i k x
G ( x, 0 ) ≈ a2 (ν , Λ k − k F ) .
π
F
Consequently, although nk is non-analytic at kF, Zk = 0 at kF and the system is not a Fermi liquid. Therefore,
Luttinger liquid is a quantum field in a different universality class from the Fermi liquid.
The non-Fermi liquid behavior of Luttinger liquid is due to large quantum fluctuations of bosons that
destroy the coherent propagation of fermions. In general, in three dimensions, quantum fluctuations are
unimportant at all temperatures, and thermal fluctuations are important near a phase transition temperature (T
~ Tc). In two dimensions, quantum fluctuations are still unimportant at all temperatures, whereas thermal
fluctuations are important at all finite temperatures T > 0. In one dimension, however, both quantum and
thermal fluctuations are always important at all temperatures T ≥ 0 , and the quantum fluctuations destroy
Fermi liquid characteristics. On the other hand, we have shown that under interaction in one dimension, there
is a free bosonic mode propagating coherently with a velocity u = vF (1 + 2 g ) . This bosonic mode in the
1/ 2
spinless case is associated with plasmons. On the other hand, the bosonic excitations in one dimension
become even more interesting when we restore spin indices in the Hamiltonian. In the following we shall
show that in a Luttinger liquid we have spin and charge separation.
The total Hamiltonian (H) consists of the non-interaction Hamiltonian (H0) and the interacting
Hamiltonian (HI), and HI includes different interaction strengths among parallel and anti-parallel spins:
2
H 0 = ∫ dx ∑∑ψ †j σ ( x )( ∓ ivF ∂ x − vF k F )ψ j σ ( x ) ,
L
(VIII.68)
0
j =1 σ
( ) ( )
L
H I = ∫ dx g ρ1↑ ρ 2↑ + ρ1↓ ρ 2↓ + g ⊥ ρ1↑ ρ 2↓ + ρ1↓ ρ 2↑ . (VIII.69)
0
1
gρ =
2
(g + g⊥ ,) (VIII.72)
1
gσ =
2
(g − g⊥ ,) (VIII.73)
( ) ∑
L L
H I = ∫ dx g ρ ρ1 ρ 2 + gσ σ 1σ 2 = ∫ dx gνν 1ν 2 , (VIII.74)
0 0
ν = ρ ,σ
⎡ 1 β ⎤ ⎡ 1 β ⎛ ϕν*ϕν ⎞⎤
∏ ∫0 dτ ∫ dx gνν 1ν 2 ⎥ = ∏ ∫ Dϕν* Dϕν exp ⎢ − ∫ dτ ∫
L L
exp ⎢ − dx ⎜ + iϕν*ν 1 + iϕνν 2 ⎟ ⎥ . (VIII.75)
ν = ρ ,σ ⎣ β 0
⎦ ν = ρ ,σ ⎣ β
0 0
⎝ gν ⎠⎦
Hence, we need separate spin and charge bosons to decouple the interaction given in EQ. (VIII.75). Similar
to our previous discussion of spinless fermions, the right and left-moving Green functions are related to the
non-interacting Green functions by the following expression:
(
G j = G j 0 exp ⎡⎣ −i Φ j ρ + σ Φ j σ − Φ′j ρ − σ Φ′j σ ⎤⎦ , ) (VIII.76)
1 dk ϕ jν ( k , ων )
∑ ∫ 2π e (
i ± k x −ων τ )
Φ jν ( x, τ ) = − . (VIII.77)
β ων iων ∓ vF k
1 β ϕν*ϕν 1 β ⎛ ∂Φ1ν ∂Φ 2ν ⎞
Seff − S0 = − ∑ ∑
β∫
dτ ∫ dx
4π ∫
dτ ∫ dx ⎜ ϕ1ν
L L
+ − ϕ 2ν ⎟ (VIII.78)
ν = ρ ,σ
0 0
gν ν = ρ ,σ
0 0
⎝ ∂x ∂x ⎠
so that we may treat the spin and charge terms separately. Fourier transforming each sector as in the spinless
case, we find that the second term in EQ. (VIII.78) is
1 dk
−
β
∑ ∫ 2π ⎡⎣ Dρ−1ϕ1ρ ( k , ωn ) ϕ1ρ ( − k , −ωn ) + Dσ−1ϕ1σ ( k , ωn ) ϕ1σ ( − k , −ωn ) ⎤⎦ , (VIII.79)
ωn
where
⎞ 1 ⎛ ( iων ) − ( uν k ) ⎞
2 2
−1 1 k ⎛ 1 1
Dν ≡ − ⎜ − ⎟≡ ⎜ 2 ⎟
, (VIII.80)
gν 4π ⎝ iων − vF k iων + vF k ⎠ gν ⎜⎝ ( iων ) − ( vF k ) ⎟⎠
2
gν
and uν2 ≡ vF2 (1 + 2 gν ) , gν ≡ . (VIII.81)
4π vF
In general, the velocities of the charge and spin sectors are different if the corresponding interaction
strengths are different. That is, uσ ≠ u ρ if gσ ≠ g ρ , implying that the coherent propagation of the spin and
charge sectors become separate in a Luttinger liquid. We also note that the exact ground state of a Luttinger
liquid depends on the specific interaction coefficients gν for a given model. Some known examples of the
ground state and the corresponding field operators include the following:
Moreover, the correlation function R(x) of a Luttinger liquid is of the general form:
( )
− ⎡⎣ 2 − α O K ρ , Kσ ⎤⎦
R ( x ) ∼ O †( x ) O ( 0 ) ∼ x , (VIII.82)
where the Kρ (> 0) and Kσ (> 0) are the correlation length exponents for charge and spin sectors, respectively,
and the exponent 0 < αO < 2 is specified by the field operator O and is also dependent on the interaction
coefficients gν of a given model. We further note that the dependence of the correlation length given inEQ.
(VIII.82) implies that there is only quasi-long range order in Luttinger liquids. In addition, there are simple
relations between the exponent αO and the correlation exponents Kρ and Kσ. For instance, the following
relations hold for CDW, SDW, SS and TS:
Generally, for a given model Hamiltonian, we only need to compute one dimensionless number Kρ to
know the character of the ground state of a Luttinger liquid. For example, if we find that Kρ > 1, then we
have αSS > αCDW, implying that the singlet superconductivity state is more stable than the charge density
wave. On the other hand, for Kρ < 1 we find αSS < αCDW, so that the charge density wave becomes a more
favorable ground state than singlet superconductivity. In the interest of time, we shall not proceed further
here with detail derivations of specific Luttinger liquids.
Further Readings
The reason why the usual form of perturbation theory does not apply to bosons can be understood by
the following consideration. For a non-interacting system of N bosons, the ground state consists of all
particles in the lowest energy mode and is given by the expression:
Φ 0( N ) = N , 0, 0, … . (IX.1)
Specifically, if we confine the non-interacting bosons within a volume Ω with periodic boundary conditions,
the ground state acquires zero momentum. Now if we apply the creation and annihilation operators a0† and a0
to the ground state, we find the following relations:
a0 Φ 0( N ) = N 1/ 2 N − 1, 0, 0,… = N 1/ 2 Φ 0( N − 1) ,
a0† Φ 0( N ) = ( N + 1) N + 1, 0, 0, … = ( N + 1) Φ 0( N + 1) .
1/ 2 1/ 2
(IX.2)
Clearly neither operators can annihilate the ground state, so that our prescriptions for separating operators
into purely creation and destruction parts in the case fermions fail completely for bosons, which makes the
application of Wick’s theorem to bosons much more complicated. The way around this problem encountered
with the definitions of a0† and a0 is to introduce new operators as follows:
⎡⎣ξ 0 , ξ 0† ⎤⎦ = Ω −1 , (IX.4)
1/ 2
⎛N⎞
ξ 0 Φ 0( N ) =⎜ ⎟ Φ 0( N − 1) , (IX.5)
⎝Ω⎠
1/ 2
⎛ N +1⎞
ξ 0† Φ 0( N ) = ⎜ ⎟ Φ 0( N + 1) . (IX.6)
⎝ Ω ⎠
The expressions given in EQs. (IX.4) – (IX.6) are based on an implicit assumption that the bosons
form an ideal bose gas. In the presence of interactions among the bosons, we expect that the occupation
number of the zero-momentum mode is reduced to N0 < N, so that the ground-state expectation value
becomes
Nonetheless, the Bogoliubov replacement is still applicable to the interacting system as long as N0 is still a
significant fraction of N. Hence, we may define the boson field operator as follows:
where the operator ϕ (x) has no zero-momentum components, and ξ0 is a c-number. The separation of the
boson field operator into two parts clearly modifies the Hamiltonian in a fundamental way. Specifically, the
potential energy can be expressed in terms of ϕ (x) and ξ0 (= n01/ 2 ):
7
1
Vˆ = ∫ d 3 x d 3 x′ ψ †( x )ψ †( x′ ) V ( x − x′ )ψ ( x′ )ψ ( x ) ≡ E0 + ∑ Vˆj , (IX.9)
2 j =1
where the eight distinct parts in EQ. (IX.9) are given by:
1
E0 = n02 ∫ d 3 x d 3 x′ V ( x − x′ ) , (IX.10)
2
1
Vˆ1 = n0 ∫ d 3 x d 3 x′ V ( x − x′ ) ϕ ( x′ ) ϕ ( x ) , (IX.11)
2
1
Vˆ2 = n0 ∫ d 3 x d 3 x′ ϕ †( x ) ϕ †( x′ ) V ( x − x′ ) , (IX.12)
2
⎛1 ⎞
Vˆ3 = 2 ⎜ n0 ⎟ ∫ d 3 x d 3 x′ ϕ †( x′ ) V ( x − x′ ) ϕ ( x ) , (IX.13)
⎝2 ⎠
⎛1 ⎞
Vˆ4 = 2 ⎜ n0 ⎟ ∫ d 3 x d 3 x′ ϕ †( x ) V ( x − x′ ) ϕ ( x ) , (IX.14)
⎝2 ⎠
⎛1 ⎞
Vˆ5 = 2 ⎜ n1/0 2 ⎟ ∫ d 3 x d 3 x′ ϕ †( x ) ϕ †( x′ ) V ( x − x′ ) ϕ ( x ) , (IX.15)
⎝2 ⎠
⎛1 ⎞
Vˆ6 = 2 ⎜ n01/ 2 ⎟ ∫ d 3 x d 3 x′ ϕ †( x ) V ( x − x′ ) ϕ ( x′ ) ϕ ( x ) , (IX.16)
⎝2 ⎠
1
Vˆ7 = ∫ d 3 x d 3 x′ ϕ †( x ) ϕ †( x′ ) V ( x − x′ ) ϕ ( x′ ) ϕ ( x ) . (IX.17)
2
We note that the interaction potential does not involve any terms associated with a single particle out of the
condensate because such terms vanish identically. For instance, we find that
Moreover, we note that E0 given in EQ. (IX.10) is taken as a c-number in the Bogoliubov prescription:
1 1
E0 = Ω −1 N 02 V ( 0 ) = Ω n02 V ( 0 ) . (IX.19)
2 2
Therefore, in the presence of interaction the contribution from E0 only effectively shifts the zero of energy
and is without any operator character.
The different processes associated with the eight terms given in EQs. (IX.10) – (IX.17) are shown in
Fig. IX.1.1, where a solid line (for either ϕ† or ϕ) denotes a particle not in the condensate, a dashed line (for
either ξ 0† or ξ0 (= n01/ 2 )) denotes a particle belonging to the condensate, and a wavy line denotes the interaction
potential V.
Figure IX.1.1 Diagrams for processes contained in the interaction potential Vˆ for bosons.
Although the use of EQ. (IX.10) based on the Bogoliubov replacement removes the problem
associated with the zero-momentum state, we still need to address the issue that the total number of particles
N under interaction is no longer a constant of motion. That is, if we treat the number of particles in the
condensate N0 as a c-number, then the number operator N̂ does not commute with the total Hamiltonian H:
⎡ 7
⎤
⎡⎣H , Nˆ ⎦⎤ = ⎢ Kˆ + E0 + ∑ Vˆj , Nˆ ⎥ ≠ 0 , (IX.20)
⎣ j =1 ⎦
and K̂ denotes the kinetic energy contribution to the Hamiltonian. This situation is problematic and we need
to return to the original Hamiltonian H in which a0† and a0 are still operators so that H commutes with N̂ .
We may reformulate the entire problem by defining an alternative hermitian operator:
Hˆ ≡ H − μ Nˆ , (IX.22)
which has a complete set of normalized eigenvectors Ψ j and eigenvalues Hj such that
Hˆ Ψ j = H j Ψ j . (IX.23)
The hermitian operator Ĥ clearly commutes with N̂ if H commutes with N̂ , so that for a given N, the
normalized ground state Ψ 0 corresponds to the lowest eigenvalue of Ĥ , which yields:
Hˆ Ψ 0( N ) = H ( μ , Ω, N ) Ψ 0( N ) = [ E ( Ω, N ) − μ N ] Ψ 0( N ) . (IX.24)
Noting that EQ. (IX.24) holds for any value of μ (so that μ can be treated as an effective Lagrange
multiplier) and recalling the thermodynamic relation
∂E ( Ω, N )
μ= , (IX.25)
∂N Ω
we find that H ( μ , Ω, N ) in EQ. (IX.24) is indeed consistent with the absolute minimum eigenvalue of the
( )
hermitian operator Ĥ because ∂Hˆ ∂N Ψ 0( N ) = 0 . In fact, we may use EQ. (IX.25) as a relation to
eliminate the variable N in terms of μ and Ω. Moreover, we note that the expectation value Ψ 0 Ĥ Ψ 0
≡ H G ( T = 0, Ω, μ ) corresponds to the minimum value of the thermodynamic potential at T = 0 for fixed μ
and Ω.
The introduction of the thermodynamic potential has the advantage of treating non-conserved particles
under finite interaction. Hence, for interacting bosons we use the Bogoliubov prescription that replaces the
operators ξ 0 and ξ 0† by the c-number n01/ 2 and rewrite N̂ and Ĥ as follows:
Nˆ = N 0 + ∑ ak† ak ≡ N 0 + Nˆ ′ , (IX.26)
k ≠0
⎛ ⎞ ⎛ ⎞
Hˆ = E − μ N = ⎜ E0 + ∑ V j + ∑ ε k0 ak† ak ⎟ − μ ⎜ N 0 + ∑ ak† ak ⎟
⎝ j k ≠0 ⎠ ⎝ k ≠0 ⎠
⎛ ⎞
(
= ⎜ E0 − μ N 0 + ∑ V j ⎟ + ∑ ε k0 − μ ak† ak )
⎝ j ⎠ k ≠0
≡ ( E0 − μ N 0 ) + Hˆ ′ . (IX.27)
Furthermore, we note that the application of any annihilation operator ak with k ≠ 0 to the non-interacting
ground state Φ 0 ≡ 0 as defined in EQ. (IX.1) always leads to zero. Therefore, we may take the non-
interacting ground state 0 as the effective “vacuum” for the interacting bose system, and Wick’s theorem is
henceforth applicable.
However, it should be noted that the use of the Bogoliubov prescription leads to final expressions of
the relevant physical quantities all containing an extra parameter N0, the number of bosons in the condensate.
To get around this problem, we may determine N0 by considering the equilibrium state of the bosonic system
at constant temperature (T), volume (Ω) and chemical potential (μ), because the thermodynamic potential of
a system is minimized for constant (T, Ω, μ). In other words, the condition of thermodynamic equilibrium
yields the following relation for the thermodynamic potential:
∂ Ψ 0 Hˆ Ψ 0 ⎡ ∂H G (T = 0, Ω, μ , N 0 ) ⎤
=⎢ ⎥ = 0, (IX.28)
∂N 0 Ω,μ ⎣ ∂N 0 ⎦ Ω,μ
Next, we want to find the exact single-particle Green’s function for bosons based on the Heisenberg
picture associated with the operator Ĥ ′ introduced in EQ. (IX.27):
7
Hˆ ′ = ∫ d 3 x ϕ †( x ) ⎡⎣ Kˆ − μ ⎤⎦ ϕ ( x ) + ∑ V j , (IX.29)
j =1
where K̂ corresponds to the kinetic energy contribution. From EQs. (IX.27) and (IX.10) – (IX.17), we note
that the operator Ĥ is hermitian and therefore has a complete set of eigenfunctions. We may denote the
ground state eigenfunction of Ĥ by O ≡ Ψ 0 , and the Heisenberg picture of the field operator ψ ( x )
becomes
ψ Hˆ ( x, t ) ≡ ei H tψ ( x ) e −i H t = ei H ′tψ ( x ) e − i H ′t
ˆ ˆ ˆ ˆ
= ei H t [ξ 0 + ϕ ( x )] e − i H t = ξ 0 + ϕ Hˆ ( x ) = n1/
0 + ϕ Hˆ ( x ) ,
ˆ′ ˆ′ 2
(IX.30)
where we have used the notation x for four-dimensional spacetime (x,t). The single-particle Green’s function
G(x,y) is therefore given by the following expression:
O T ⎡⎣ψ Hˆ ( x )ψ H†ˆ ( y ) ⎤⎦ O
iG ( x, y ) ≡
O O
=
( )
O T ⎡⎣( ξ 0 + ϕ Hˆ ( x ) ) ξ 0 + ϕ H†ˆ ( y ) ⎤⎦ O
O O
O ϕ Hˆ ( x ) + ϕ H†ˆ ( y ) O O T ⎡⎣ϕ Hˆ ( x ) ϕ H†ˆ ( y ) ⎤⎦ O
= n0 + n01/ 2 +
O O O O
O T ⎡⎣ϕ Hˆ ( x ) ϕ H†ˆ ( y ) ⎤⎦ O
= n0 + , (IX.31)
O O
where we have used the fact that ak ≠ 0 O = 0 and O ak† ≠ 0 O = 0 , the latter is the result of orthogonality of
momentum states. Hence, from EQ. (IX.31) we have
iG ( x, y ) = n0 + iG ′( x, y ) , (IX.32)
O T ⎡⎣ϕ Hˆ ( x ) ϕ H†ˆ ( y ) ⎤⎦ O
where iG ′( x, y ) ≡ (IX.33)
O O
is apparently associated with the non-condensate. Moreover, with the usual definition of Fourier transforms,
we may express the expectation values of the number operator and the kinetic energy using EQ. (IX.33) as
follows:
Ω q2
K = lim ∫ dω d q
3
iG ′( ω , q ) eiωη . (IX.35)
( 2π )
4
η →0 + 2m
It is clear from EQ. (IX.35) that the stationary condensate with zero momentum does not contribute to the
kinetic energy. Finally, with the definition of Ĥ ′ in EQ. (IX.30), we have the following relation
∂ϕ Hˆ ( x, t )
i = ⎡⎣ϕ Hˆ ( x, t ) , Hˆ ′ ⎤⎦ . (IX.36)
∂t
Having established the basic formalism for bosons, we are now ready to discuss the corresponding
perturbation theory, diagrammatic analysis and Feynman rules in Part IX.2.
Given the boson Green’s function in EQ. (IX.33) and the definition of the following operator
Hˆ 0 ≡ ( E0 − μ N 0 ) + Hˆ 0′ , (IX.37)
where
Ĥ 0′ ≡ ∫ d 3 x ϕ †( x ) [ K − μ ]ϕ ( x ) , (IX.38)
OI ( t ) = e
i Hˆ 0 t − i Hˆ 0 t i Hˆ 0′ t − i Hˆ 0′ t
OS e =e OS e , (IX.39)
where OS denotes the operator in the Schrödinger’s picture. Hence, we may relate the Heisenberg picture to
the interaction picture via the U-operator:
− i Hˆ ′ ( t −t0 )
U ( t , t0 ) = ei H 0′ t e
ˆ − i Hˆ 0′ t0
e , (IX.40)
∂U ( t , t0 )
i
∂t
=e
i Hˆ 0′ t
( Hˆ ′ − Hˆ ′ ) e 0
− i Hˆ 0′ t
U ( t , t0 )
⎛ 7 ⎞ − i Hˆ 0′ t
⎜ ∑V j ⎟ e U ( t , t0 ) ≡ e 0 Hˆ I e 0 U ( t , t0 )
i Hˆ 0′ t i Hˆ ′ t − i Hˆ ′ t
=e
⎝ j =1 ⎠
= Hˆ I ( t ) U ( t , t0 ) . (IX.41)
Furthermore, we note that the operator Ĥ 0 given in EQ. (IX.37) commutes with the number operator N̂ :
⎡ Hˆ 0 , Nˆ ⎤ = 0 , (IX.42)
⎣ ⎦
so that the non-interacting ground state 0 of Ĥ 0 is a state with a definite number of particles. In addition,
for all momentum k,
ak 0 = 0 , (IX.43)
so that 0 can be taken as the non-interacting “vacuum” state. Hence, we may apply perturbation theory to
the boson system. Following the Gell-Mann and Low theorem, we find that the interacting ground state O
and the non-interacting ground state 0 can be related by the expression
O U ( 0, ±∞ ) 0
= . (IX.44)
0O 0 U ( 0, ±∞ ) 0
⎛
∞ ν
−i ⎞ 1 ∞ 0 T ⎡⎣ Hˆ I ( t1 ) Hˆ I ( tν ) ϕ I ( x ) ϕ I† ( y ) ⎤⎦ 0
iG′( x, y ) = ∑ ⎜ ⎟
∞
ν =0 ⎝
∫ dt1 … ∫−∞ dtν
⎠ ν ! −∞ 0 U ε ( ∞, −∞ ) 0
which is the Green’s function for a free Bose gas, and the corresponding Fourier transform of EQ. (IX.46) is
1
G ′ (0) ( ω , q ) = lim . (IX.47)
η →0 + ω − ωq + μ + iη
From EQ. (IX.46), we note that G ′ (0) ( x, y ) vanishes if ty > tx, implying that the Green’s function of a free
Bose gas only propagates forward in time. In other words, the Bose gas cannot support backward
propagation in time (or equivalently, hole propagation), in contrast to the situation encountered in fermion
systems.
The analysis of EQ. (IX.45) is mostly consistent with that for fermions except the following features
of the Feynman’s rules in coordinate space for bosons:
1) The total number of lines (solid and dashed) going into a Feynman diagram must equal that coming out,
and there is a factor n01/ 2 for each dashed line entering or leaving each vertex.
2) For the connected graphs, the ν-th order involves ν operators Hˆ I so that there are ν! different ways of
arranging them in the Feynman diagrams. This factor exactly cancels the factor (ν!)−1 in EQ. (IX.45).
3) The terms Vˆ1 , Vˆ2 and Vˆ7 are symmetric under the interchange of variables x and x ′, whereas Vˆ3 , Vˆ4 , Vˆ5
and Vˆ are not. Consequently there is an extra factor of 2 associated with Vˆ , Vˆ or Vˆ , which exactly
6 1 2 7
cancels the factor of (1/2) in EQs. (IX.11), (IX.12) and (IX.17), and every distinct Feynman diagram
associated with Vˆj (j = 1, 2, … 7) need be counted only once.
4) Each ν-th order diagram in the G ′ expansion acquires a factor of (i)ν (−i)C, where C is the number of
condensate factors n0 appearing in the diagram.
5) The fact that backward propagation in time is absent in bose systems helps eliminate large classes of
diagrams. For instance, there are no contractions within Vˆj (j = 1, 2, … 7) because they are already normal
ordered according to EQs. (IX.11) – (IX.17). In addition, there are no contributions from any G ′(0) free
particle line either closes on itself or joined by the same interaction. Furthermore, any diagram containing
G ′(0) free particle lines running in opposite directions does not contribute to the boson propagator. [N.B.!
The vanishing contribution associated with the close loops or pairs of opposite particle lines only applies
to the free particle lines, and is not applicable to the exact particle lines G ′ .]
For Feynman rules in momentum space, a factor of n01/ 2 is assigned to each dashed line as in the case
of coordinate space. In addition, there is an overall factor of (i)ν (−i)C(2π)−4(ν−C) for the ν-th order
As an example, we consider the first-order contribution G ′(1) . We find that the terms Vˆ1 , Vˆ2 , Vˆ5 and
Vˆ6 in Hˆ I do not contribute to G ′(1) because they do not conserve the number of k ≠ 0 particles. The term Vˆ7
also does not contribute because the only possible diagrams containing Vˆ would involve holes. Thus, only
7
Vˆ3 and Vˆ4 are left in the first-order correction, and the Feynman rules yield
and the corresponding diagrams are shown in Fig.IX.2.1. Similarly, the second-order contributions to
G ′( 2) are illustrated in Fig. IX.2.2 (a) – (g), where the diagrams from (a) through (e) are of order n02 and
those from (f) and (g) are of order n0. We note that the absence of diagrams with only non-condensate lines is
the result of no hole propagation.
(a ) (b)
Although terms like Vˆ1 , Vˆ2 , Vˆ5 and Vˆ6 do not contribute to the first-order contribution G ′(1) (or
equivalently, second-order in G ′(0) ), it is important to note that in higher order corrections to the Green
function G ′ , those terms that do not conserve the number of non-condensate particles may enter. As we shall
discuss later, Vˆ1 and Vˆ2 will enter into terms that are third-order in G ′(0) . The key point to keep in mind is
that the total number of particles is conserved, whereas the condensate of bosons provides a sink and source
that allows non-conservation of non-condensate particles.
Therefore, in contrast to the case of fermions, the particle lines of bosons need not run continuously
through a diagram. If the proper self-energy for a bosonic system is defined as a part of a Feynman diagram
that connects to the rest of the diagram by two non-condensate particle lines, we find that there are three
distinct proper self-energies, as illustrated in Fig. IX.2.3. The first term Σ11 *
( p ) consists of one incoming
particle line and one outgoing particle line, similar to the proper-self energy of fermions. The second and
third terms are new features associated with bosons, the former Σ12 *
( p ) consists of two outgoing particle
lines and two incoming condensate lines, and the latter Σ*21( p ) contains two incoming particle lines and two
outgoing condensate lines.
Σ11
*
( p) Σ12
*
( p) Σ*21( p )
p −p p
With the introduction of two additional proper self-energies, we must also correspondingly introduce
two additional Green’s functions G12′ and G21′ , known as the anomalous Green’s functions. Thus, the Dyson’s
equations for the non-condensate Green’s functions in momentum space are given as follows:
′ ( p ) = G ′(0)( p ) Σ12
G12 *
( p ) G′( − p ) + G′(0)( p ) Σ11* ( p ) G12′ ( p ) , (IX.50)
The diagrammatic representations corresponding to EQs. (IX.49) –(IX.51) are shown in Fig. IX.2.4 (a) – (c).
We note that the direction of momentum flow given in Fig. IX.2.4. is determined by four-momentum
conservation. For instance, we find that an equivalent expression to EQ. (IX.49) is given by
(a) p
G′(0)( p ) p G′(0)( p ) p
(b) p
G′(0)( p ) p G′(0)( p ) p
−p
(c)
G′(0)( − p )− p G′(0)( − p ) − p
Σ*21( p ) + Σ11
*
(− p)
′ ( p)
G21 =
−p
G′( p ) p ′ ( p)
G21
p
p
Fig. IX.2.4 Dyson’s equations for non-condensate Green’s functions G ′( p ) , G12′ ( p ) and G21
′ ( p) .
The anomalous Green’s functions G12′ and G21′ can be defined in terms of Heisenberg field operators
as follows:
O T ⎡⎣ϕ Hˆ ( x ) ϕ Hˆ ( y ) ⎤⎦ O
′ ( x, y ) ≡
iG12 , (IX.53)
O O
O T ⎡⎣ϕ H†ˆ ( x ) ϕ H†ˆ ( y ) ⎤⎦ O
′ ( x, y ) ≡
iG21 . (IX.54)
O O
′ ( x, y ) = G12
G12 ′ ( y, x ) , ′ ( x, y ) = G21
G21 ′ ( y, x ) , (IX.55)
so that the Fourier transforms of the anomalous Green’s functions are even functions of the four-momentum:
′ ( p ) = G12
G12 ′ (− p) , ′ ( p ) = G21
G21 ′ (− p) . (IX.56)
⎛ ϕ Hˆ ( x ) ⎞
Φ Hˆ ( x ) = ⎜ ⎟. (IX.57)
ϕ †
⎝ H ⎠ˆ ( x )
Using EQ. (IX.57), we may define a (2×2) matrix Green’s function G ′ ( x, y ) :
O T ⎡⎣ Φ Hˆ ( x ) Φ †Hˆ ( y ) ⎤⎦ O
iG ′( x, y ) ≡ , (IX.58)
O O
where the off diagonal elements of G ′ ( x, y ) are given in EQs. (IX.53) and (IX.54). We further identify the
diagonal elements by the relations:
′ ( x, y ) = G ′( x, y ) ,
G11 ′ ( x, y ) = G ′( y , x ) .
G22 (IX.59)
Whence, we rewrite the Dyson’s equations for different Green’s functions into a single matrix equation:
⎛ Σ* ( x, y ) Σ12
*
( x, y ) ⎞
where Σ*( x, y ) = ⎜ *11 ⎟, (IX.61)
⎝ Σ 21( x, y ) Σ11( y, x ) ⎠
*
⎛ G ′(0) ( x, y ) 0⎞
and G ′(0)( x, y ) = ⎜ ⎟. (IX.62)
⎝ 0 G′ ( y, x ) ⎠
( 0)
In the case of a uniform medium, EQ. (IX.60) can be Fourier transformed into
⎛ G ′( p ) ′ ( p) ⎞
G12
where G ′( p ) = ⎜ , (IX.64)
′ ( p ) G ′( − p ) ⎟⎠
⎝ G21
⎛ G ′(0)( p ) 0 ⎞
G ′(0)( p ) = ⎜ ⎟, (IX.65)
⎝ 0 G′ ( − p ) ⎠
( 0)
⎛ Σ* ( p ) Σ12 *
( p) ⎞
Σ*( p ) = ⎜ *11 ⎟. (IX.66)
⎝ Σ 21( p ) Σ *
11 ( − p ) ⎠
To solve EQ. (IX.63), it is more convenient to invert the matrix so that the Dyson’s equation for the
Green’s function matrix becomes:
−1 −1
⎡⎣G ′( p ) ⎤⎦ = ⎡⎣G ′(0)( p ) ⎤⎦ − Σ*( p ) , (IX.67)
⎛ ⎡G ′(0) ( p ) ⎤ −1 0 ⎞
⎣
−1 ⎦
⎡⎣G ′(0)( p ) ⎤⎦ = ⎜ ⎟, (IX.69)
⎜ 0
−1 ⎟
⎡⎣G ′ ( − p ) ⎤⎦ ⎠
(0)
⎝
D( p ) ≡ [ G12 ′ ( p ) − G ′( p ) G ′( − p )] .
−1
′ ( p ) G21 (IX.70)
Using EQ. (IX.47) for the matrix elements in EQ. (IX.69), we obtain
( )
−1
G ′( − p ) = − ⎡ G ′(0) ( p ) *
− Σ11( p ) ⎤ D( p ) = − ⎡⎣ω − ω p + μ − Σ11* ( p ) ⎤⎦ D( p ) , (IX.71)
⎣ ⎦
( )
−1
G ′( p ) = − ⎡ G ′(0) ( − p ) *
− Σ11( − p ) ⎤ D( p ) = − ⎡⎣ −ω − ω p + μ − Σ11* ( − p ) ⎤⎦ D( p ) , (IX.72)
⎣ ⎦
′ ( p ) = − Σ12
−G12 *
( p ) D( p ) , (IX.73)
′ ( p ) = − Σ*21( p ) D( p ) .
−G21 (IX.74)
1 1
S( p) ≡ ⎡⎣ Σ11* ( p ) + Σ11* ( − p ) ⎤⎦ , A( p ) ≡ ⎡⎣ Σ11* ( p ) − Σ11* ( − p ) ⎤⎦ . (IX.76)
2 2
Hence, EQs. (IX.71) – (IX.76) provide the general expressions for the Green’s functions in terms of given
proper self-energies, eigen-energies and chemical potential of the boson system.
The non-condensate Green’s function for bosons can also be expressed in terms of the Lehmann
spectral representation, following similar procedures used for deriving the corresponding representation for
fermions. In Problem Set 1 you are asked to prove the Lehmann spectral representation given below for
interacting bosons:
⎡ O Φ( 0 ) j , p j , p Φ †( 0 ) O O Φ † ( 0 ) j , −p j , − p Φ ( 0 ) O ⎤
G ′( p ) = Ω lim
η →0+
∑ ⎢ ω − ( H − H ) + iη −
ω + ( − ) − η
⎥, (IX.77)
j ⎢
⎣ j p 00 H j , − p H 00 i ⎥⎦
where the states j , p refer to those eigen-states with ( N ± 1) particles if there are N particles in the O
state, and the momentum operator P̂ and the operator Ĥ yield the relations:
Pˆ j , p = p j , p , Hˆ j , p = H jp j , p . (IX.78)
From the spectral representation in EQ. (IX.77), we note that the residues of G ′( p ) are real. On the other
′ ( p ) and G21
hand, it can be shown that the residues of G12 ′ ( p ) are complex conjugates of each other.
Having derived the general expressions for the Green’s functions in EQs. (IX.71) – (IX.76), we are
ready to apply the formalism to real physical systems of specified interaction potentials. In the following
section, we consider an example of a weakly interacting homogeneous bose gas so that we may keep to the
lower-order terms of the interaction.
Consider a weakly interacting Bose gas whose interaction potential V ( x ) has a well-defined Fourier
transform V ( p ) . The non-condensate Green’s function may be kept to the first order because of the weak
interaction, so that the corresponding proper self-energy to lowest order can be derived from EQ. (IX.48):
where we have kept to the expansion to second order in G ′(0)( p ) so that the contribution from the term
associated with Σ12
*
( p ) that leads to third and higher orders in G′(0)( p ) is neglected in EQ. (IX.79). We note
that EQ. (IX.80) implies that the lowest order proper self-energy Σ11
*
( p ) is only dependent on momentum p
and is independent of energy.
′ and G21
Applying the Feynman rules to the anomalous Green’s functions G12 ′ as we have done for
G ′ in Part IX.2, we obtain all non-vanishing first and second-order contributions shown in Fig. IX.3.1.
′ in Fig. IX.3.1 (a) with Vˆ1 and Vˆ2 in Fig. IX.1.1, we
′ and G21
Comparing the first order contribution to G12
find that Vˆ contributes to G ′(1) whereas Vˆ contributes to G ′(1) . Therefore, we obtain the lowest order
1 21 2 12
contribution to Σ *
12 ( p ) and Σ ( p ) :
*
21
Σ12
*
( p ) = Σ*21( p ) ≈ n0 V ( p ) . (IX.81)
From EQs. (IX.71) – (IX.75) it is clear that we must determine the chemical potential μ to achieve a
full description of the Green’s functions of the interacting Bose system. To obtain the chemical potential, we
use EQs. (IX.27) and (IX.28), so that
⎛ ∂H ⎞ ∂ ∂Vˆ
⎜ ⎟ = O Hˆ O = O −μ O = 0,
⎝ ∂N 0 ⎠Ω , μ ∂N 0 ∂N 0
∂Vˆ
⇒ μ= O O . (IX.82)
∂N 0
In the interaction representation, the corrections to the chemical potential μ to all orders may be obtained by
rewriting EQ. (IX.82) into the following expression:
μ=
ν =0 ⎣ 0⎦
,
ν
∞
⎛ −i ⎞ 1 ∞ dt … ∞ dt 0 T ⎡ Hˆ t
∑ ⎜⎝ ⎟⎠ ν ! ∫−∞ 1 ∫−∞ ν ⎣ I ( 1 ) H I ( tν ) ⎤⎦ 0
ˆ
ν =0
ν
∞
⎛ −i ⎞ 1 ∞ ⎡ ∂Vˆ ⎤
= ∑⎜ ⎟
∞
p −p p −p p −p p −p p −p p −p
(g) G21′(1) (h) G21′( 2) (i) G21′( 2) (j) G21′( 2) (k) G21′( 2) (l) G21′( 2)
p −p p −p p −p p −p p −p p −p
∂Vˆ ∂ ⎡ 7
⎤
μ= 0
∂N 0
0 = 0 ⎢
∂N 0 ⎣
E0 + ∑ Vˆj ⎥ 0
j =1 ⎦
= 2 N 0−1 E0 + N 0−1 0 Vˆ1 + Vˆ2 + Vˆ3 + Vˆ4 0 + ( 2 N 0 ) 0 Vˆ5 + Vˆ6 0
−1
= n0 V ( 0 ) , (IX.84)
where we have used EQs. (IX.10) – (IX.17) in EQ. (IX.84), and we note that all matrix elements in EQ.
(IX.84) vanish because they are normal ordered. Thus, we obtain from EQs. (IX.80), (IX.81) and (IX.84):
μ = Σ11
*
( p ) − Σ12* ( p ) . (IX.85)
Having specified the proper self-energies and the chemical potential for the weakly interacting
bosons in EQs. (IX.81) and (IX.85) and noting that ωp = ε p0 = p ( 2m ) , we may rewrite EQs. (IX.72) –
2
ω + ε p0 + n0V ( p ) ω + ε p0 + n0V ( p )
G ′( p ) = ≡
ω 2 − ⎡⎣ε p0 + n0V ( p ) ⎤⎦ + [ n0V ( p )] ω 2 − Ep2
2 2
up2 vp2
≡ − , (IX.86)
ω − Ep + iη ω + Ep − iη
− n0V ( p )
′ ( p ) = G21
G12 ′ ( p) =
ω 2 − Ep2
−up vp up vp
≡ + , (IX.87)
ω − Ep + iη ω + Ep − iη
where
{ } {(ε ) }
1/ 2 1/ 2
Ep ≡ ⎡⎣ε p0 + n0V ( p ) ⎤⎦ − [ n0V ( p )] 0 2
+ 2ε p0 n0V ( p )
2 2
= p , (IX.88)
1/ 2
⎧1 1⎫
up ≡ ⎨ Ep−1 ⎣⎡ε p0 + n0V ( p ) ⎦⎤ + ⎬ , (IX.89)
⎩2 2⎭
1/ 2
⎧1 1⎫
and vp ≡ ⎨ Ep−1 ⎣⎡ε p0 + n0V ( p ) ⎦⎤ − ⎬ . (IX.90)
⎩2 2⎭
From EQs. (IX.89) and (IX.90), we note that in the V ( p ) → 0 limit, up → 1 and vp → 0 so that the Green’s
−1
functions become G ′( p ) → ⎡⎣ω − ε p0 + iη ⎤⎦ ′ ( p ) = G21
and G12 ′ ( p ) → 0 , recovering the free boson limit.
The poles in the Green’s functions thus derived provide information for the excitation spectrum of
the weakly interacting Bose system. In the following we consider the dispersion relation of Ep in two extreme
limits. First, in the long wavelength limit we find that the dispersion relation for Ep in EQ. (IX.88) reduces to
⎡ n V( 0) ⎤
{(ε ) + 2ε }
1/ 2
1/ 2
0 2
≈ {2ε ( p )}
1/ 2
Ep = p
0
p n0V (p) 0
p n0V ≈ p⎢ 0 ⎥ for p → 0 . (IX.91)
⎣ m ⎦
The linear dispersion relation in EQ. (IX.91) is phonon-like. Therefore, from EQ. (IX.91) we may define a
characteristic velocity
⎡ n V( 0) ⎤
1/ 2
u≡⎢ 0 ⎥ (IX.92)
⎣ m ⎦
which resembles the sound velocity. We further note that the results given above for the weakly interacting
Bose system are only well defined for repulsive interaction V(0) > 0.
Next, we consider the behavior of Ep for large momentum, which depends on the functional form of
V(p). If we assume that V(p) approaches a constant for |p| < (1/r0), which is equivalent to a repulsive
interaction potential of a range r0, and if we further assume
1
n0V ( 0 ) ≡ , (IX.93)
2mr02
we find
{(ε ) + 2ε n V(p )}
1/ 2
0 2
≈ ε p0 + n0V ( p ) 2mn0V ( 0 ) .
0 2
Ep = p p 0 for p (IX.94)
Comparing EQ. (IX.94) with EQ. (IX.91), we find that with increasing |p| the dispersion relation of Ep
changes from linear to quadratic near p ≈ [ 2mn0V ( 0 )]
1/ 2
, and the dispersion relation given in EQ. (IX.94)
for the large momentum limit may be interpreted as the quasiparticle energy consisting of the non-condensate
kinetic energy and the potential energy due to interaction between particles in the non-condensate and those
in the condensate.
To evaluate the depletion of condensate from the weak interaction prescribed in this section, we
employ EQs. (IX.34) and(IX.86) such that
1
d ω d p iG ′( ω , p ) e
( 2π ) ∫
iωη
n = n0 + lim 4
3
η →0 +
1
( 2π ) ∫
2
= n0 + d p vp .
3
3
(IX.95)
Therefore, vp2 in EQ. (IX.95) may be interpreted as the momentum distribution of non-condensate particles
in the ground state due to interaction, and we note that in the presence of interaction, the condition n0 < n
−1
always holds. Moreover, from EQ. (IX.90) we find that vp2 ∝ p in the long wavelength limit, implying the
macroscopic occupation of the zero momentum state. In the limit of V ( p ) → 0 , we find that Ep → ε p0 and
vp2 → 0 so that n0 → n and the system recovers the behavior of an ideal Bose gas.
Up to this point we have limited our consideration to uniform interacting bosons at T = 0. In Part
IX.4 we generalize the formalism for bosons to finite temperature and non-uniform conditions.
In this section we generalize our previous zero-temperature formalism to finite temperatures. However,
we note that the condensate of interacting bosons generally undergoes a phase transition at a finite critical
temperature Tc > 0, which leads to complicated behavior in the vicinity of Tc. For simplicity, we only restrict
our consideration to the temperature range of 0 < T << Tc, and leave the derivation of finite superconducting
phase transition temperatures associated with charged superfluidity to Part XI. Hence, we consider the grand
canonical Hamiltonian of an assembly of bosons:
⎡ ∇2 ⎤ 1
Hˆ = ∫ d 3 x ψ †( x ) ⎢ − x − μ ⎥ψ ( x ) + ∫ d 3 x d 3 x′ ψ †( x )ψ †( x′ ) V ( x − x′ )ψ ( x′ )ψ ( x ) . (IX.96)
⎣ 2m ⎦ 2
At T = 0, recall that the field operator of a uniform Bose system under the Bogoliubov approximation is
given by ψ ( x ) → ξ 0 + ϕ ( x ) , where we have the expectation values
O ψ ( x ) O = ξ0 and O ϕ( x) O = 0 .
For T > 0, we define the ensemble average of the field operator as follows:
Ψ ( x ) ≡ ψ ( x ) = Tr [ ρˆ Gψ ( x )] = Z G−1 Tr ⎢ e
⎡ (
− β H − μ Nˆ )ψ ( x )⎤ , (IX.97)
⎣ ⎥⎦
Z G ≡ Tr ⎢e
⎡ (
− β H − μ Nˆ )⎤.
⎣ ⎥⎦
Therefore, for a finite ensemble average in the thermodynamic limit, the deviation of the field operator from
the ensemble average is defined by:
ϕ ( x ) ≡ ψ ( x ) − ψ ( x ) = ψ ( x ) − Ψ( x ) . (IX.98)
We may treat Ψ ( x ) as a c-number, which is effectively the condensate wave function at T > 0 and has the
physical significance of the order parameter.
Next, we consider a weakly interacting Bose system where most particles are in the condensate, so
that ϕ ( x ) may be considered as a small correction to Ψ ( x ) . Thus, EQ. (IX.96) may be expanded in powers
of ϕ and ϕ†, and to the linear and quadratic terms of ϕ and ϕ† we have
Hˆ = Hˆ 0 + Hˆ 1 + Hˆ ′ , (IX.99)
where
⎡ ∇2 ⎤ 1
Hˆ 0 = ∫ d 3 x Ψ *( x ) ⎢ − x − μ ⎥ Ψ ( x ) + ∫ d 3 x d 3 x′ V ( x − x′ ) Ψ ( x ) Ψ ( x′ ) ,
2 2
(IX.100)
⎣ 2m ⎦ 2
⎡ ∇ 2
2⎤
Hˆ 1 = ∫ d 3 x ϕ †( x ) ⎢ − x − μ + ∫ d 3 x′ V ( x − x′ ) Ψ ( x′ ) ⎥ Ψ ( x )
⎣ 2m ⎦
⎡ ∇ 2x ⎤
+ ∫ d 3x ⎢− − μ + ∫ d 3 x ′ V ( x − x ′ ) Ψ ( x′ ) ⎥ Ψ * ( x ) ϕ ( x ) ,
2
(IX.101)
⎣ 2m ⎦
⎡ ∇ ⎤
2
Hˆ ′ = ∫ d 3 x ϕ †( x ) ⎢ − x − μ ⎥ ϕ ( x ) + ∫ d 3 x d 3 x′ V ( x − x′ ) ⎡ Ψ ( x′ ) ϕ †( x ) ϕ ( x )
2
⎣ 2m ⎦ ⎣
1 1 ⎤
+Ψ * ( x ) Ψ ( x′ ) ϕ †( x′ ) ϕ ( x ) + Ψ * ( x ) Ψ * ( x′ ) ϕ ( x′ ) ϕ ( x ) + ϕ †( x ) ϕ †( x′ ) Ψ ( x′ ) Ψ ( x ) ⎥ . (IX.102)
2 2 ⎦
⎧⎪⎛ ∇ 2x ⎞ 2⎫
⎪
⎨⎜ − − μ ⎟ + ∫ d 3 x′ V ( x − x′ ) Ψ ( x′ ) ⎬ Ψ ( x ) = 0 . (IX.103)
⎪⎩⎝ 2m ⎠ ⎪⎭
Hˆ → Hˆ eff = Hˆ 0 + Hˆ ′ , (IX.104)
Equation (IX.103) may be considered as a self-consistent Hartree equation for the condensate wave function
that renders the effective Hamiltonian quadratic in the non-condensate field operator. For self consistency,
we may further define an effective statistical operator based on EQs. (IX.103) and (IX.104):
− β Hˆ eff − β Hˆ eff
ρˆ eff = e Tr ⎡e ⎤, (IX.105)
⎣ ⎦
so that EQ. (IX.97) for the condensate wave function becomes:
∂ϕ Hˆ ( x, τ )
= ⎡⎣ Hˆ , ϕ Hˆ ( x,τ ) ⎤⎦ = ⎡⎣ Hˆ ′, ϕ Hˆ ( x,τ ) ⎤⎦
∂τ
⎡ ∇2 ⎤
= ⎢ x + μ − ∫ d 3 x′ V ( x − x′ ) Ψ ( x′ ) ⎥ ϕ Hˆ ( x,τ )
2
⎣ 2m ⎦
− ∫ d 3 x′ V ( x − x′ ) ⎡⎣ϕ H†ˆ ( x′,τ ) Ψ ( x′ ) + Ψ *( x′ ) ϕ Hˆ ( x′,τ ) ⎤⎦ Ψ ( x ) , (IX.108)
∂ϕ H†ˆ ( x, τ ) ⎡ ∇2 ⎤
= ⎢ − x − μ + ∫ d 3 x′ V ( x − x′ ) Ψ ( x′ ) ⎥ ϕ H†ˆ ( x, τ )
2
∂τ ⎣ 2m ⎦
+ ∫ d 3 x′ V ( x − x′ ) ⎡⎣ϕ H†ˆ ( x′,τ ) Ψ ( x′ ) + Ψ *( x′ ) ϕ Hˆ ( x′, τ ) ⎤⎦ Ψ * ( x ) , (IX.109)
≡ −Ψ ( x ) Ψ *( x′ ) + g ′( x, τ ; x′,τ ′ ) ,
where
g ′( x, τ ; x′,τ ′ ) ≡ − Tτ ⎡⎣ϕ Hˆ ( x, τ ) ϕ H†ˆ ( x′, τ ′ ) ⎤⎦ (IX.111)
refers to the non-condensate part of the Green’s function. In addition, we may express the mean value of the
total particle density n(x) and the condensate particle density n0(x) using the Green’s functions so that
∂ ⎡ ∂ϕ Hˆ ( x, τ ) † ⎤
g ′( x,τ ; x′, τ ′ ) = − Tτ ⎢ ϕ Hˆ ( x′, τ ′ ) ⎥ − δ (τ − τ ′ ) ⎡⎣ϕ Hˆ ( x, τ ) , ϕ H†ˆ ( x′, τ ′ ) ⎤⎦
∂τ ⎣ ∂τ ⎦
⎡ ∂ϕ Hˆ ( x, τ ) † ⎤
= − Tτ ⎢ ϕ Hˆ ( x′, τ ′ ) ⎥ − δ (τ − τ ′ ) δ ( x − x′ ) , (IX.114)
⎣ ∂τ ⎦
and then multiply EQ. (IX.108) to its right side of by ϕ H†ˆ ( x′, τ ′ ) and compare the result with EQ. (IX.114).
Thus, we obtain
⎧ ∂ ∇ 2x 2⎫
⎨− + + μ − ∫ d 3 x′′ V ( x − x′′ ) Ψ ( x′′ ) ⎬ g ′( x, τ ; x′,τ ′ )
⎩ ∂τ 2m ⎭
− ∫ d x′′ V ( x − x′′ ) Ψ ( x ) ⎡⎣ g 21
3
′ ( x′′, τ ; x′,τ ′ ) Ψ ( x′′ ) + Ψ *( x′′ ) g ′( x′′, τ ; x′, τ ′ ) ⎤⎦ = δ (τ − τ ′ ) δ ( x − x′ ) , (IX.115)
Next, if we multiply EQ. (IX.109) to its right side of by ϕ H†ˆ ( x′, τ ′ ) and follow similar steps that lead to EQ.
(IX.115), we obtain the other equation for the anomalous Green’s function:
⎧ ∂ ∇ 2x 2⎫
⎨ + + μ − ∫ d 3 x′′ V ( x − x′′ ) Ψ ( x′′ ) ⎬ g 21 ′ ( x,τ ; x′, τ ′ )
⎩ ∂τ 2m ⎭
− ∫ d 3 x′′ V ( x − x′′ ) Ψ * ( x ) ⎡⎣ g 21
′ ( x′′,τ ; x′,τ ′ ) Ψ ( x′′ ) + Ψ *( x′′ ) g ′( x′′, τ ; x′,τ ′ ) ⎤⎦ = 0 . (IX.117)
Thus, the equations of motion in EQs. (IX.115) and (IX.117) together with the self-consistent equation EQ.
(IX.103) for the mean value of the condensate completely determine the behavior of finite-temperature
interacting bosons.
We may Fourier transform the thermal Green’s functions into the form of Matsubara frequencies:
g ′( x, τ ; x′,τ ′ ) = β −1 ∑ e
− iωn (τ −τ ′)
g ′( x, x′; ωn ) , (IX.118)
n
′ ( x,τ ; x′, τ ′ ) = β −1 ∑ e
g 21
− iωn (τ −τ ′)
′ ( x, x′; ωn ) ,
g 21 (IX.119)
n
⎧ ∇ 2x 2⎫
⎨iωn + + μ − ∫ d 3 x′′ V ( x − x′′ ) Ψ ( x′′ ) ⎬ g ′( x, x′; ωn )
⎩ 2m ⎭
− ∫ d x′′ V ( x − x′′ ) Ψ ( x ) ⎡⎣ g 21
3
′ ( x′′, x′; ωn ) Ψ ( x′′ ) + Ψ *( x′′ ) g ′( x′′, x′; ωn ) ⎤⎦ = δ ( x − x′ ) , (IX.120)
⎧ ∇ 2x 2⎫
−
⎨ n iω + + μ − ∫ d 3 x′′ V ( x − x′′ ) Ψ ( x′′ ) ⎬ g 21 ′ ( x, x′; ωn )
⎩ 2m ⎭
− ∫ d 3 x′′ V ( x − x′′ ) Ψ * ( x ) ⎡⎣ g 21
′ ( x′′, x′; ωn ) Ψ ( x′′ ) + Ψ *( x′′ ) g ′( x′′, x′; ωn ) ⎤⎦ = 0 . (IX.121)
Although in principle the coupled equations in EQs. (IX.120) and (IX.121) can be solved for a given
condensate wave function Ψ ( x ) that satisfies EQ. (IX.103), the solutions can be quite complex. In the
following we only consider the behavior of two simple cases of interacting bosons.
[Uniform condensate]
We first consider a uniform and stationary system of bosons at T > 0. The condensate wave function
can therefore be treated as a temperature-dependent constant:
Ψ ( x ) ≡ [ n0( T )]
1/ 2
. (IX.122)
μ = n0( T ) V ( k = 0 ) . (IX.123)
For a time-independent system, we also Fourier transform the Green’s functions into the following:
1 ik ⋅ ( x − x ′ )
g ′( x − x′; ωn ) = g ′( k ; ωn ) ,
( 2π ) ∫
3
3
d ke (IX.124)
1 ik i ( x − x ′ )
′ ( x − x′; ωn ) = ′ ( k ; ωn ) ,
( 2π ) ∫
3
g 21 d ke
3
g 21 (IX.125)
{iω − ε − n V ( k )} g ′( k , ω ) − n V ( k ) g ′ ( k , ω ) = 1 ,
n
0
k 0 n 0 21 n
(IX.126)
{−iω − ε − n V ( k )} g ′ ( k , ω ) − n V ( k ) g ′( k , ω ) = 0 .
n
0
k 0 21 n 0 n
(IX.127)
uk2 vk2
g ′( k , ωn ) = − , (IX.128)
iωn − Ek iωn + Ek
⎡ 1 1 ⎤
′ ( k , ωn ) = −uk vk ⎢
g12 − ⎥, (IX.129)
⎣ iωn − Ek iωn + Ek ⎦
where Ek , uk and vk have the same expressions as those given in EQs. (IX.88) – (IX.90). Therefore, for the
uniform Bose system at relatively low temperatures, we may identify Ek as the single-particle excitation
energy through the usual analytic continuation to real frequency. Similar to our previous discussion for
weakly interacting bose system, in the long wavelength limit we find the excitation spectrum is phonon-like:
⎡ n (T ) V ( 0 ) ⎤
1/ 2
Ek ≈ u k ≡ ⎢ 0 ⎥ k for k → 0 , (IX.130)
⎣ m ⎦
and V(0) must be positive. Moreover, Ek > 0 for all k ≠ 0 , which has an important consequence of a finite
critical velocity (known as the Landau critical velocity) and superfluidity as opposed to the ideal Bose gas
situation. We shall discuss more about superfluidity in Part IX.5.
Using the equal-time Green’s function given in EQ. (IX.128), we obtain the non-condensate particle
density as a function of T via the relation in EQ. (IX.34) and the substitution ( 2π ) ∫ dω ⇒ iβ Σω
−1 −1
n
:
1 ⎛ uk2 vk2 ⎞
= ∫d k⎜ +
3
β Ek − β Ek ⎟. (IX.131)
( 2π )
3
⎝e −1 1− e ⎠
From EQ. (IX.95) we have n′( 0 ) , so that the thermal effect on depleting the density of the condensate is
estimated according to the following:
m
n0( T ) = n0( 0 ) − ( k BT )
2
for T → 0 (IX.134)
12u
1/ 2
Ψ( T )
⎡ n (T ) ⎤ m
≈ ( k BT )
2
⇒ 1− = 1− ⎢ 0 ⎥ for T → 0 . (IX.135)
Ψ( 0 ) ⎣ n0( 0 ) ⎦ 24un0( 0 )
Consequently, we find that the condensate wave function Ψ decreases with increasing T according to EQ.
(IX.135).
Next, we consider a spatially non-uniform Bose system where Ψ(x) must be determined from EQ.
(IX.103). Suppose that Ψ(x) satisfies a local U(1) gauge symmetry so that
Ψ( x ) = F ( x ) e ( ) ,
iθ x
(IX.136)
where the functions F(x) and θ (x) are both real. Noting that the total particle current density is related to the
equal-time Green’s function by the following expression:
⎡( ∇ − ∇ ′ ) g ( x, τ ; x′,τ + ) ⎤⎦
i
j( x) = , (IX.137)
2m ⎣ x = x′
−i 1 n ( x)
j0( x ) = ⎡⎣ Ψ ∗( x )∇Ψ( x ) − Ψ ( x )∇Ψ *( x ) ⎤⎦ = ⎡⎣ F( x ) ⎤⎦ ∇θ ( x ) = 0 ∇θ ( x ) ,
2
(IX.138)
2m m m
1
v 0( x ) = ∇θ ( x ) , (IX.139)
m
and ∇ i v 0( x ) = 0 from EQ. (IX.139) so that v 0( x ) is irrotational. If Ψ(x) is single-valued, then the line
integral of the velocity around a closed path in the condensate yields the quantization of circulation in units
of ( 2π / m ) if is restored:
⎛ 2π ⎞
∫d i v 0( x ) = i ∇θ ( x ) = n ⎜
m∫
d ⎟. (n: integer) (IX.140)
⎝ m ⎠
so that the self-consistent equation EQ. (IX.103) for the condensate wavefunction becomes
⎛ ∇ 2x ⎞
+ μ ⎟ Ψ( x ) − γ Ψ( x ) Ψ( x ) = 0 .
2
⎜ (IX.142)
⎝ 2m ⎠
From EQs. (IX.136) and (IX.142) we obtain the equations for the real and imaginary parts F(x) and θ (x):
⎧1 ⎫
⎡⎣ F( x ) ⎤⎦ ∇θ ( x ) ⎬ = ∇ i j0( x ) = 0 ,
2
∇i⎨ (IX.143)
⎩m ⎭
[ F( x ) ] ∇ 2 F( x ) 1 ⎡ ∇θ ( x ) ⎤ [ F( x ) ] ∇ 2 F( x )
2 2 2
μ γ γ 1
= − + = − + v02 . (IX.144)
m m 2 m F( x )
2
2 ⎢⎣ m ⎥⎦ m 2 m F( x ) 2
2
Equation (IX.143) is the continuity equation for the condensate, whereas EQ. (IX.144) is the quantum analog
of Bernoulli’s equation for steady flow of the condensate.
Ψ ( x ) = n1/
0 f ( z) ,
2
(z ≥0) (IX.145)
where f (z) is a real function, f (z) → 1 for z → ∞, and f (z) satisfies the following relation from EQ. (IX.142):
1 d2 f μ
+ f − f3 =0. (IX.146)
2mn0γ dz 2
n0γ
From the asymptotic behavior of f (z), EQ. (IX.146) yields the relation for the chemical potential for z → ∞:
μ = n0 γ , (IX.147)
and the characteristic length scale ξ can be obtained from EQs. (IX.146) and (IX.147) so that
1/ 2
⎛ 1 ⎞
ξ =⎜ ⎟ , (IX.148)
⎝ 2mn0γ ⎠
⎛ z ⎞
f ( z ) = tanh ⎜ ⎟. (IX.149)
⎝ 2ξ ⎠
As a second example, we consider another special form of EQ. (IX.136) that represents an unbound
condensate of the form:
Ψ ( x ) = n1/
0 f (r) e ,
2 iθ
(IX.150)
where (r,θ) represents the plane-polar coordinates, f (r) is real, and f (r) → 1 for r → ∞. The form in EQ.
(IX.150) corresponds to a singly quantized vortex of circulation ( 2π / m ) , which can be easily seen from
the condensate velocity by inserting EQ. (IX.150) into EQ. (IX.139):
v 0( x ) = θˆ , (IX.151)
mr
and we have restored in EQ. (IX.151). To find the solution for f (r), we insert EQ. (IX.150) into EQ.
(IX.142) and obtain
1 ⎛1 d r d − 1 ⎞ f + μ f − f 3 =0.
⎜ ⎟ (IX.152)
2mn0γ ⎝ r dr dr r 2 ⎠ n0γ
The asymptotic behavior for f (r) together with EQ. (IX.152) leads to the same chemical potential as that
given in EQ. (IX.147). Using the characteristic length ξ defined by EQ. (IX.148) and taking ζ ≡ (r / ξ ) , we
rewrite EQ. (IX.152) into the following:
d2 f 1 df 1
+ − f + f − f 3 = 0. (IX.153)
dζ 2
ζ dζ ζ2
The solution to EQ. (IX.153) takes different forms for ζ << 1 and ζ >> 1. Specifically, we find that
( )
−1
f ( ζ ) ∼ 1 − 2ζ 2 for ζ >> 1. (IX.155)
Furthermore, we note that EQ. (IX.154) implies that the condensate particle density decreases to zero in a
region of radius ξ, and this region may be interpreted as the vortex core. We further remark that the
vanishing condensate particle density inside the vortex core is the result of large circulating velocities
associated with the particles in the vortex, which suppress the condensate density to zero. This phenomenon
does not only only occur in neutral superfluids but can also exist in charged superfluids, the superconductors.
The element helium has two stable isotopes, one is bosonic 4He and the other fermionic 3He. Both
isotopes are known to remain liquid down to absolute zero without solidification under low pressure, because
of the combined effect of a weak inter-atomic attraction and their small masses that lead to strong zero-point
oscillations. Therefore, liquid 4He and 3He are known as quantum liquids, and they do not solidify until very
large pressures, approximately 25 atm for 4He and 30 atm for 3He. Moreover, the mass densities of both
liquids are low and are pressure dependent. Under saturated vapor, the density is approximately 0.145 g/cm3
for 4He and 0.081 g/cm3 for 3He. Schematics of the pressure (P) versus temperature (T) phase diagram for
3
He and 4He are shown in Fig. IX.5.1.
The two quantum liquids differ fundamentally because of their quantum statistics. Specifically, in the
case of fermions the Pauli exclusion principle tends to keep particles apart, so that at low temperatures the
many-body state associated with finite inter-particle interactions only involves mixtures of eigenfunctions
from the unoccupied states with |k| > kF. In contrast, bosons tend to occupy the same single-particle state at
low temperatures so that the introduction of inter-particle interactions leads to a many-body state
preferentially mixing eigenfunctions of already occupied states, thus enhancing the many-body effect. In the
context of interacting bosons in Part IX, we only focus our discussion on the bosonic superfluid 4He in this
section.
3
He
4
P He
(atm)
40
Tλ(P)
20 He I
He II
0 2 4 6 T (K)
Fig. IX.5.1 Comparison of the pressure (P) versus temperature (T) phase diagram of 3He and 4He.
Here we review some of the important experimental phenomena associated with liquid 4He and relate
these phenomena to various descriptions for interacting bosons described in Part IX.1 – IX.4.
Under saturated vapor pressure, liquid 4He undergoes a second order phase transition at the lambda point
Tλ = 2.17 K, below which the liquid becomes a He II phase that exhibits various novel properties different
from the He I phase above Tλ. Specifically, He II can flow through narrow channels without exhibiting any
pressure drop if the velocity is below a temperature dependent critical value, implying no viscosity. In
addition, the thermal conductivity of He II essentially diverges so that it is generally under the isothermal
condition although the overall temperature of He II can be changed by either varying the vapor pressure or
direct injection of heat. As illustrated in Fig. IX.5.1, the lambda transition temperature decreases with
increasing external pressure (P) before the liquid solidifies, and the lambda transition line Tλ (P) belongs to
Although the lambda transition is associated with the onset of superfluidity, in reality the He II phase does
not behave like a pure superfluid. For instance, while He II can flow through narrow channels without
pressure drop, direct measurements of the He II viscosity reveal results comparable to those of the He I
phase, which imply normal fluid behavior. The paradoxical phenomena were explained by Tisza and by
Landau [L. Tisza, Nature 141, 913 (1938); L. D. Landau, J. Phys. (USSR) 5, 71 (1941); J. Phys. (USSR) 11,
91 (1947)] with a two-fluid model, in which He II is considered as a liquid constituting a mixture of two
interpenetrating components, one is the superfluid of density ρs and velocity vs, and the other is the normal
fluid of density ρn and velocity vn. Therefore, the total density of He II is given by ρ (T) = ρs (T) + ρn (T), and
ρn (T → 0) → 0, ρs (T → Tλ) → 0. The superfluid component is assumed to be non-viscous and irrotational,
so that
∇ × vs = 0 , (IX.156)
whereas the normal fluid component is assumed to be viscous. The two-fluid model has been well verified
experimentally, and it is conceptually analogous to our earlier consideration of interacting bosons in EQ.
(IX.95), where the total number of bosons can be separated into that in the condensate and that in the non-
condensate. However, it should be noted that EQ. (IX.156) neglects the possibility of vortices. In the
presence of vortices, quantized circulations are present, which leads to singularities in the superfluid velocity.
3. Second sound:
The presence of two components in the He II phase naturally allows an additional degree of freedom for
sound wave propagation. That is, besides the standard first sound that involves both the superfluid and
normal fluid components moving together, there exists second sound in which the two fluid components are
moving along the opposite direction. The second sound propagation transmits periodic variations in the ratio
of (ρs/ρn) so that it represents a temperature and entropy wave. The existence of the second sound is a novel
feature of a Bose-Einstein condensed (BEC) system: the state of the BEC system must be specified by the
velocity of the superfluid in addition to the necessary variables for describing the excitations. We further note
that the first and second sound velocities depend on temperature and the interparticle interaction.
In the following we discuss the sound propagation in the two-fluid model more explicitly. We shall
restrict our consideration to the hydrodynamic limit so that sufficient collisions of particles allow the system
to reach local thermodynamic equilibrium. Given the mass density ρ (= ρs + ρn) and the mass current density
j (=ρsvs + ρnvn), the conservation law reads
∂ρ
+ ∇i j = 0 . (IX.157)
∂t
∂j ∂
= ( ρ s v s + ρ n v n ) = −∇p . (IX.158)
∂t ∂t
Using EQs. (IX.157) and (IX.158), we obtain the following relation between the mass density and the
pressure:
∂2ρ
2
− ∇2 p = 0 . (IX.159)
∂t
Here we note that the pressure in fact depends on both density and temperature in thermal equilibrium.
To determine the frequencies of the two sound velocities in a superfluid, we need to derive a second
relation associated with the acceleration of the superfluid. This relation may be obtained by considering the
variations of entropy in the superfluid as a function of the temperature. To understand how entropy varies in
a superfluid, we first note that the entropy in a superfluid must be carried by the normal fluid component, and
that in the absence of dissipation, the entropy is conserved. Hence, we have the following conservation
equation for the entropy per unit mass s ≡ S ( NM ) :
∂ (ρs)
+ ∇i ( ρ s v n ) = 0 . (IX.160)
∂t
∂s ∂ρ
ρ +s + ρ s∇i ( v n ) = 0 . (IX.161)
∂t ∂t
Combining EQs. (IX.161) and (IX.157), we obtain
∂s ρs ∂2s ρs ⎡ ∂ ⎤
=s ∇i ( v s − v n ) ⇒ =s ∇i ⎢ ( v s − v n )⎥ . (IX.162)
∂t ρ ∂t 2
ρ ⎣ ∂t ⎦
In the second part of EQ. (IX.162) we have employed the fact that ∂ ( s ρ s ρ ) ∂t = 0 .
To proceed further we need to find how superfluid velocity is related to various thermodynamic quantities
such as μ, T and p. In the absence of external potentials the change in the superfluid velocity vs for bosons of
mass m is due to variations in the chemical potential μ:
∂ vs
m = −∇μ . (IX.163)
∂t
Under local thermodynamic equilibrium, small changes in μ for a system of N particles are dependent on
pressure and temperature according to the Gibbs-Duhem relation:
N d μ = Ω dp − S dT , (IX.164)
where Ω and S denote the volume and entropy of the system, respectively. Given that the mass density
ρ ≡ Nm Ω , we may rewrite EQ. (IX.164) as follows:
∂ ( vn − vs ) ρ
= −s ∇T , (IX.166)
∂t ρn
where we have neglected terms associated with variations in the density. Combining EQs. (IX.166) and
(IX.162), we find the second relation between density and temperature variations:
∂2s ρs 2 2
= s ∇T. (IX.167)
∂t 2
ρn
Next, we consider collective modes of the superfluid with spatial and temporal dependence given by the
factor exp [i ( q i r − ωt )] . Under small variations in ρ and T, we may rewrite EQs. (IX.159) and (IX.167) into
a set of coupled equations:
⎡⎛ ∂p ⎞ ⎛ ∂p ⎞ ⎤ ⎡ 2 2 ⎛ ∂p ⎞ ⎤ 2 ⎛ ∂p ⎞
ω 2δρ − q 2 ⎢⎜ ⎟ δρ + ⎜ ⎟ δ T ⎥ = ⎢ω − q ⎜ ⎟ ⎥ δρ − q ⎜ ⎟ δ T = 0 , (IX.168)
⎣ ⎝ ∂ρ ⎠ T ⎝ ∂T ⎠ ρ ⎦ ⎣ ⎝ ∂ρ ⎠ T ⎦ ⎝ ∂T ⎠ ρ
⎡⎛ ∂s ⎞ ⎛ ∂s ⎞ ⎤ 2 ⎛ ρs ⎞ 2 2 ⎛ ∂s ⎞ ⎡ 2 ⎛ ∂s ⎞ 2 ⎛ ρs ⎞ 2 ⎤
ω 2 ⎢⎜ ⎟ δρ + ⎜ ⎟ δ T ⎥ − q ⎜ ⎟ s δ T = ω ⎜ ⎟ δρ + ⎢ω ⎜ ⎟ − q ⎜ ⎟ s ⎥ δ T = 0 , (IX.169)
⎣⎝ ∂ρ ⎠T ⎝ ∂T ⎠ ρ ⎦ ⎝ ρn ⎠ ⎝ ∂ρ ⎠T ⎣ ⎝ ∂T ⎠ ρ ⎝ ρn ⎠ ⎦
Noting that the specific heat per unit mass at a constant volume is defined as
⎛ ∂s ⎞
cv = T ⎜ ⎟ , (IX.170)
⎝ ∂T ⎠ ρ
⎛ ∂p ⎞ ⎛ ∂S ⎞ 2 ⎛ ∂s ⎞
⎜ ⎟ =⎜ ⎟ = −ρ ⎜ ⎟ , (IX.171)
⎝ ∂T ⎠ ρ ⎝ ∂Ω ⎠T ⎝ ∂ρ ⎠T
we find that the sound velocities in the superfluid may be obtained by the condition for non-trivial solutions
to the coupled equations in EQs. (IX.168) and (IX.169). In other words, if we define u ≡ ( ω q ) , the non-
trivial condition for sound progapation satisfies the following quadratic equation for u2:
(u 2
− c12 )( u 2
)
− c22 − u 2 c32 = 0 , (IX.172)
Thus, the first and second sound velocities for superfuid are given by the solutions to EQ. (IX.172):
Given the solutions in EQ. (IX.174), the exact values of the first and second sound velocities of a BEC
system depend on the specific model of our consideration. Here we examine a specific example, the weakly
interacting Bose gas at low temperatures. In this limit, the ground state energy is E0 = N 2V ( 0 ) ( 2Ω ) so that
the pressure p is given by:
∂E0 N2 1 ρ 2V ( 0 )
p=− = V( 0) = n 2V ( 0 ) = . (IX.175)
∂Ω 2Ω 2
2 2m 2
From EQ. (IX.173), we have c12 = nV ( 0 ) m , c22 = nV ( 0 ) ( 3m ) , and c32 = 0 . Hence, the first sound velocity
according to EQ. (IX.174) is
nV ( 0 )
u12 = c12 = , (IX.176)
m
whereas the second sound velocity is given by
nV ( 0 )
u22 = c22 = . (IX.177)
3m
In this special case, first sound is a pure wave of density modulations, which is a long-wavelength excitation
mode of the condensate. On the other hand, the second sound corresponds to a variation in the density of
excitations, which is a pure temperature wave without variations in the total particle density.
4. Critical velocities:
1 1
Mv 2 = Mv′2 + ε p ; Mv = Mv′ + p . (IX.178)
2 2
so the lowest value of v that satisfies EQ. (IX.179) occurs for v p , which corresponds to a critical velocity
vc that is the minimum velocity required to incur elementary excitations and energy dissipation in the moving
body:
⎛ p εp ⎞
vc = min ⎜ + ⎟. (IX.180)
⎝ 2M p ⎠
In the limit of M → ∞, the critical velocity vc is determined by the minimum of (ε p p ) . If the low-energy
excitations are primarily phonons as given in EQ. (IX.91) so that ε p = u p where u is the first sound
velocity given in EQ. (IX.92), then the critical velocity is identical to the sound velocity, implying that no
excitations can occur for velocities smaller than u so that the critical velocity is vc = u. If the excitation
spectrum is gapped as given by EQ. (IX.94), then the critical velocity is dependent on the magnitude of the
energy gap and is always finite. On the other hand, if the excitation spectrum exhibits a free boson dispersion
2
relation so that ε p ∝ p , then the critical velocity is zero so that no superfluidity can exist. This finding
confirms that ideal Bose gas in the condensate is not a superfluid. In other words, interaction among bosons
is essential for the occurrence of superfluidity.
5. Quantized circulations:
As discussed in Part IX.4, quantized circulations can exist in the condensate of interacting bosons with the
velocity field given by EQ. (IX.151), v s ( x ) = ( mr ) θˆ , so that the circulations are quantized in units of
2π
κ≡ = ∫ d i v s ( x ) = ∫ dS nˆ i ⎡⎣∇ × v s ( x ) ⎤⎦ = 0.997 × 10 −3 cm 2 / sec , (IX.181)
m C
where dS denotes the surface area enclosed by the closed path C, n̂ is the corresponding unit vector, and m is
the atomic mass of 4He. The existence of quantized vortex loops in He II was first predicted independently
by Onsager and Feynman, and has been confirmed empirically. Moreover, the energy per unit length of a
vortex line may be estimated by the following relation:
2
⎛ κ ⎞ ⎛ ρ sκ ⎞ dr
2
1 1
Ev = ∫ d r = ∫ d r ρs ⎜
2 ⎝ 2π r ⎠ ⎝ 4π ⎠ ∫ r
2
ρ 2
s vs
2
=
⎟ ⎜ ⎟
2
ρ sκ 2 ⎛ R ⎞
≈ ln ⎜ ⎟ , (IX.182)
4π ⎝ξ ⎠
where R represents an upper cutoff, which may be taken as the lateral characteristic dimension of the Bose
system under consideration, and ξ is the lower cutoff comparable to the radius of the vortex core.
Finally, we discuss Landau’s quasiparticle model and compare the model with realistic liquid 4He.
The specific heat of He II at low temperatures (T << Tλ) is found to vary as T 3. To explain this
empirical finding, Landau interpreted that the low-lying excitations of He II consist of weakly interacting
phonons near zero momentum |p| with a linear dispersion relation and gapped roton excitations with a
quadratic momentum dependence at larger momentum and a dip near p0:
ε phonon( p ) = u p , (IX.183)
(p −p )
2
ε roton ( p )
0
=Δ+ , (IX.184)
2mroton
where the empirically determined parameters include the first sound velocity u = 238 cm/sec, the roton
energy gap (Δ/kB) ≈ 8.6 K, the dip momentum k0 = ( p0 ) ≈ 0.19 nm −1 , and the roton mass 0.16 mHe. The
schematic of empirically determined low-energy excitations of He II is illustrated in Fig. IX.5.2, and we note
that Landau’s model described by EQs. (IX.183) and (IX.184) is consistent with EQs. (IX.91) and (IX.94) in
Part IX.4.
According to Landau’s quasiparticle model of He II, the total Helmholtz free energy F(T,Ω) of a
system of He II arises from the thermally induced quasiparticles. As discussed previously in Part IX.4, the
number of quasiparticles (Nqp) is not conserved so that it must be determined by the following relation:
⎛ ∂F ⎞
⎜ ⎟ =0. (IX.185)
⎝ ∂N qp ⎠Ω ,T
Moreover, if the total free energy consists of contributions solely from phonons and rotons, we may express
F and Nqp as follows:
F = Fphonon + Froton , N qp = N phonon + N roton . (IX.186)
F ( T , Ω ) = k BT
Ω
( 2π ) ∫
d3
3
p ln 1 − e ( − βε p
) (IX.187)
Ω
( )
−1
N qp ( T , Ω ) =
βε p
( 2π ) ∫
d3
3
p e −1 (IX.188)
and assuming that the relations in EQs. (IX.187) and (IX.188) apply separately to all momenta, we obtain
ζ ( 3)
3
⎛k T ⎞
N phonon = Ω⎜ B ⎟ , (IX.189)
π2 ⎝ u ⎠
2Ω p02
N roton = ( mroton k BT )1/ 2 e−Δ k T , B
(IX.190)
( 2π )
3/ 2
3
π2 ⎛ k BT ⎞
Fphonon ( T , Ω ) = − Ω k BT ⎜ ⎟ , (IX.191)
90 ⎝ u ⎠
Froton (T , Ω ) = − k BT N roton . (IX.192)
From EQs. (IX.191) and (IX.192) and the definitions of the entropy S and heat capacity Cv:
⎛ ∂F ⎞ ⎛ ∂S ⎞
S = −⎜ ⎟ , Cv = −T ⎜ ⎟ , (IX.193)
⎝ ∂T ⎠Ω ⎝ ∂T ⎠Ω
we find that the heat capacity for phonons has a T 3 temperature dependence, whereas the heat capacity for
rotons decreases exponentially with decreasing T.
[εk/kB]
(K)
16
k0
0 0.8 1.6 2.4 k (angstrom−1)
Next, consider the situation of quasiparticles moving with a mean drift velocity v relative to the rest
frame of the superfluid. The total momentum carried by the moving quasiparticles is given by
Ω p Ω
d p f (ε p − p • v ) p
( 2π ) ∫ exp ⎡⎣ β ( ε p − p • v ) ⎤⎦ − 1 ( 2π ) ∫
P= 3
d p 3
≡ 3
3
Ω ⎡ ∂f ( ε p ) ⎤
( )
( 2π ) ∫
= d p ⎢ f3
ε − p • v + ⎥p
∂ε
3 p
⎢⎣ p ⎥⎦
Ω ∞ ⎡ ∂f ( ε p ) ⎤ 4
⎥ p ≡ ρ n(T ) v ,
6π 2 ∫0
≈ v d p ⎢− (IX.194)
⎣⎢ ∂ε p ⎦⎥
because the total momentum is only associated with the normal fluid, i.e. the non-condensate. From the
expression for the normal fluid density
Ω ∞ ⎡ ∂f ( ε p ) ⎤ 4
ρ n(T ) = ∫ d p ⎢− ⎥p , (IX.195)
6π 2 ⎣⎢ ∂ε p ⎦⎥
0
we obtain
4
2π 2 ⎛ k BT ⎞
( ρ ) phonon = ⎜ ⎟ , (IX.196)
45u ⎝ u ⎠
n
p02 N roton
( ρ n )roton = . (IX.197)
3k BT Ω
which are consistent with experimental results. To obtain the results in EQs. (IX.196) and (IX.197), we have
used the identities
∞ xν −1dx ∞
∫0 = Γ(ν ) ζ (ν ) , Γ(ν ) ≡ ∫ tν −1e −t dt , (IX.198)
ex −1 0
Based on Landau’s quasiparticle model, we may estimate the critical velocity for He II. If the
quasiparticle energy were associated with pure phonons, the critical velocity would be the same as the sound
velocity of He II, which is vc = u ≈ 238 m / sec . On the other hand, if the roton excitations are included, we
find that vc ≈ ( Δ p0 ) ≈ 60 m / sec . Therefore, in He II the critical velocity appears to be roton-limited. In
reality, however, the critical velocity determined from measurements of the breakdown of superfluid flow in
narrow channels is much smaller than 60 m/sec and is attributed to the creation of quantized vortices in
moving superfluid. Overall, Landau’s quasiparticle model captures a number of essential features of He II.
Some quantitative discrepancies between the model and the real physical system are largely due to missing
elements of topological excitations in the model. You may recall that our earlier derivation for the low-
energy excitations (including the phonon mode and the gapped excitations consistent with the rotons) of the
bosonic condensate was purely based on diagrammatic analysis. Since topological excitations cannot be
derived from diagrammatic consideration, it is not surprising that the possibility of vortices was missing in
the picture.
Having discussed interacting fermions and bosons, we are ready to begin the investigation of
superconductivity, which is a physical state involving bosonic condensate of Cooper pairs and fermionic
quasiparticles as the corresponding low-energy excitations. Many phenomena in superconductivity are
similar to those in superfluidity, and superconductivity is often modeled as charged superfluidity. However,
we shall take a slight detour to visit some interesting recent developments in atomic and molecular physics
that involve many-body physics of bosons, including the Bose-Einstein condensation in dilute cold gases.
Further Readings:
Interacting bosons:
1. Fetter and Walecka, “Quantum Theory of Many-Particle Systems”, Chapters 6, 10, and 14.
2. Abrikosov, Gorkov, and Dzyaloshinski, “Methods of Quantum Field Theory in Statistical Physics”,
Chapter 5.
X.1. Introduction
The development of laser techniques to trap atoms and to cool them to ultralow temperatures since
1986 has ushered a new era of research that employs experimental techniques traditionally in atomic and
molecular physics to investigate topics of many-body interactions in condensed matter physics. A milestone
in this regard was the experimental manifestation of Bose-Einstein condensation (BEC) in trapped diluted
gases of alkali atoms of rubidium, sodium and lithium in 1995. Further technical developments lead to
condensation of bosons consisting of Fermion pairs with varying pairing strengths tuned by a trapping
magnetic field. Rotating the superfluid of such Fermion pairs reveals vortices as in superfluid helium and
superconductors. In addition, refined laser cooling techniques have led to the capability of trapping atoms to
either a three-dimensional or a two-dimensional optical lattice potential. This new capability enables a
controlled means of simulating realistic condensed matter systems with parameters such as the lattice
constant, periodic potential and particle-particle interaction strength all tunable under controlled optical
setups.
One major advantage of the cold gas clouds for investigations of quantum phenomena is that the
mean-field approximation is generally valid because the scattering length is typically much smaller than the
interparticle separation, in sharp contrast to strongly correlated condensed matter systems such as liquid 4He.
To date, in addition to observation of the Bose-Einstein condensation in bosons and verification of the
Bardeen-Cooper-Schrieffer (BCS) superconducting transitions in fermions, a number of fundamental
properties of condensed matter physics, such as the bandstructures and the two-dimensional superfluid and
one-dimensional Mott insulator systems have been demonstrated in the optical lattices. The application of
experimental techniques from atomic and molecular physics to the investigation of quantum many-body
physics in condensed matter has sparked much excitement and opened up new opportunities for research,
particularly in the area of controlled quantum dynamics that was not easily accessible in realistic condensed
matter systems.
In Part X, we only focus on the basic concepts and phenomena of BEC in cold atoms without getting
into extensive discussions of this active interdisciplinary research field. You may refer to the references
listed in the end of this chapter for more information related to research of cold atoms and optical lattices.
Hence, we have the relation TBEC ~ n 2 / 3 . Considering that the typical density of atoms in liquids and solids is
of order 1022 cm−3 while the typical particle density at the center of a Bose-Einstein condensed atomic clouds
ranges from 1013 cm−3 to 1015 cm−3, we expect the TBEC values for dilute gases to be typically 105 ~ 106 times
lower than that for a typical liquid or solid system. Therefore, the temperature must be of order 10−5 K or
lower in order to observe quantum phenomena in dilute gases. Such low temperatures cannot be achieved by
cryogenic techniques, and only became possible after the breakthrough development of laser cooling.
Specifically, particles were first trapped and cooled from room temperature to ~ 10 μK by laser beams in six
directions – front and back, left and right, up and down. The laser beams were subsequently turned off and
the cold atoms were magnetically trapped by the Zeeman interaction of electron spins with an
inhomogeneous magnetic field. The atoms in the magnetic trap were further cooled by evaporation, finally
reaching ~ 100 nK.
To generalize our discussion for trapped non-interacting bosons, we consider a semi-classical limit in
which the consecutive energy separations among single-particle states are sufficiently smaller than the
temperature so that we may treat the energy spectrum as a continuum. For a trapped atomic cloud in
thermodynamic equilibrium, the total number N of particles is conserved so that the chemical potential μ
must be included in the Bose distribution function f 0(ε) ≡ [exp(ε−μ)/(kBT)−1]−1, where ε denotes the energy
of the single-particle state in a specific trapping potential, and ε is a function of N and T. If the lowest single-
particle state εmin is taken to be 0, the chemical potential μ becomes 0 in the condensate (T < TBEC) and μ < 0
for T > TBEC. If we further define the fugacity of the system by the quantity ζ = exp[μ/(kBT)], we find that ζ =
1 for T < TBEC and ζ < 1 for T > TBEC.
Next, we consider the density of states N (ε) ≡ dN(ε)/dε of a bosonic system, where N(ε) denotes the
total number of states that depends on the trapping potential. For simplicity, we assume that all particles are
in one particular internal (spin) state. For a special case of three-dimensional free bosons in a volume Ω, the
total number of states N (ε) is related to the energy εp = p2/(2m) by the expression
4π ( 2mε ) 21/ 2 ( mε )
3 3/ 2 3/ 2
4π ⎛ p ⎞
N (ε ) = Ω ⎜ ⎟ =Ω =Ω , (X.2)
3 ( 2π )
3
3 ⎝ 2π ⎠ 3π 2 3
d N (ε ) Ωm3/ 2
N (ε ) = = 1/ 2 2 3
ε 1/ 2 . (X.3)
dε 2 π
For a d-dimensional free bosonic system, EQ. (X.3) may be generalized to yield the energy dependence N(ε)
∝ ε (d/2 − 1). Therefore, the density of states of free particles is independent of energy in two dimensions.
As another example, we consider a particle in the following anisotropic harmonic oscillator potential
1
V (r ) =
2
(
m ω12 x 2 + ω22 y 2 + ω32 z 2 . ) (X.4)
where nj are integers and nj ≥ 0. For sufficiently high energies, we may neglect the zero-point motion, so that
the total number of states becomes
1 ε ε − ε1 ε − ε1 − ε 2 ε3
N (ε ) = 3 ∫ d ε1 ∫ dε 2 ∫ dε 3 = , (X.6)
ω1ω2ω3 0 0 0 6 3ω1ω2ω3
We may generalize EQs. (X.5) – (X.7) to a d-dimensional harmonic-oscillator potential, and we find that the
density of states in the latter case is given by
ε d −1
N (ε ) = . (X.8)
( d − 1) !∏ i =1 ωi
d
The two examples discussed above suggest that the density of states generally varies as a power of
the energy and may be expressed by the following general relation:
N ( ε ) = cα ε α −1 , (X.9)
where cα and α are constants. Using the definition in EQ. (X.9), we have α = 3/2 for three-dimensional gas
confined by rigid walls and α = 3 for a three-dimensional harmonic oscillator potential.
Given the density of states, we can compute the total number of particles in the excited states:
∞
N ex = ∫ d ε N ( ε ) f 0( ε ) . (X.10)
0
At the BEC temperature, the chemical potential is μ = 0 and all particles are in the excited states. That is,
∞ 1 ⎛ ε ⎞
N = N ex ( TBEC , μ = 0 ) = ∫ d ε N ( ε ) ε ( k BTBEC )
, ⎜x ≡ ⎟
0
e −1 ⎝ k BTBEC ⎠
∞ x
= cα ( k BTBEC )
α
= cα Γ(α ) ζ (α )( k BTBEC ) ,
α
∫0 dx
e −1
x
(X.11)
where the Gamma function Γ(α) and the Riemann zeta function ζ(α) are defined as follows:
∞
Γ(α ) ≡ ∫ dx xα −1e − x , (X.12)
0
ζ (α ) ≡ ∑ n =1 n −α .
∞
(X.13)
For convenience in later discussions, we list in Table X.2.1 the Γ(α) and ζ(α) functions for selected α values.
Table X.2.1: The Gamma function Γ and the Riemann zeta function ζ for selected values of α.
α Γ(α) ζ(α)
1 1 ∞
3/2 π1/2/2 = 0.886 2.612
2 1 π2/6 = 1.645
5/2 3π1/2/4 = 1.329 1.341
3 2 1.202
7/2 15π1/2/8 = 3.323 1.127
4 6 π4/90 = 1.082
From EQ. (X.11), we obtain the following general expression for the BEC temperature of a system if
the density of states is given by EQ. (X.9):
N 1/ α
k BTBEC = . (X.14)
[ c Γ( α ) ζ (α ) ]
1/ α
α
Let’s consider several special cases. We have found earlier that the density of states associated with a
uniform Bose gas in a three-dimensional box of volume Ω corresponds to α = 3/2, and the coefficient c3/2 is
given in EQ. (X.3). Therefore, from EQ. (X.14) we obtain
2π ⎛ 2 2/3
n ⎞ ⎛ 2 2/3
n ⎞
k BTBEC = 2/3 ⎜ ⎟ ≈ 3.31⎜ ⎟, (X.15)
[ ζ ( 3 / 2 )] ⎝ m ⎠ ⎝ m ⎠
which is consistent with our earlier simple estimate in EQ. (X.1) within a numerical factor. Next, we note
that the BEC temperature for a uniform Bose gas in a two-dimensional box is 0 according to Table X.2.1 and
EQ. (X.14) for α = 1. On the other hand, a two-dimensional Bose gas can condense at a finite temperature if
it is confined in a two-dimensional harmonic-oscillator potential where α = 2.
Knowing the density of states, we may also derive the temperature dependence of Nex(T) as well as
the condensate fraction N0(T) = N − Nex(T) at T < TBEC where μ = 0. From EQ. (X.11), we have
∞ 1
N ex ( T < TBEC ) = ∫ d ε N ( ε ) = cα Γ(α ) ζ (α )( k BT ) ,
α
0 ε ( k BT )
e −1
α
⎛ T ⎞
= N⎜ ⎟ , (X.16)
⎝ TBEC ⎠
⎡ ⎛ T ⎞α ⎤
N 0( T < TBEC ) = N ⎢1 − ⎜ ⎟ ⎥. (X.17)
⎢⎣ ⎝ TBEC ⎠ ⎥⎦
∞ ε α +1
E ( T < TBEC ) = cα ∫ d ε ε α −1 ε ( k BT )
= cα Γ(α + 1) ζ (α + 1)( k BT )
0
e −1
ζ (α + 1) T α +1
= Nk Bα , (X.18)
ζ (α ) TBEC
α
where we have used the identity Γ(α + 1) = α Γ(α ) . From EQ. (X.18), the specific heat becomes
∂E E
C≡ = (α + 1)
∂T T
α
ζ (α + 1) ⎛ T ⎞
= Nk Bα (α + 1) ⎜ ⎟ . (X.19)
ζ (α ) ⎝ TBEC ⎠
If we use the alternative definition for specific heat, C ≡ T ( ∂S ∂T ) , where S is the entropy, we find that
C (α + 1) E
S= =
α α T
α
ζ (α + 1) ⎛ T ⎞
= Nk B (α + 1) ⎜ ⎟ . (X.20)
ζ (α ) ⎝ TBEC ⎠
∞ 1
N = cα ∫ d ε ε α −1 (ε − μ ) ( k T ) , (X.21)
0
e B
−1
∞
cα ∫ d ε ε α −1 ⎡⎣ e(
μ −ε ) ( k BT )
+ e ( ) ( B ) ⎤⎦ if ( ε − μ ) ( k BT )
2 μ −ε k T
1. (X.22)
0
∞ 1
E = cα ∫ d ε ε α (ε − μ ) ( k T ) , (X.23)
0
e B
−1
∞
cα ∫ d ε ε α ⎡⎣ e( ) ( B ) + e ( ) ( B ) ⎤⎦ if ( ε − μ ) ( k BT )
μ −ε k T 2 μ −ε k T
1. (X.24)
0
We may eliminate the fugacity exp [μ/(kBT)] by comparing the expressions for N and E, so that in the high
temperature limit, the following relation holds:
ζ (α ) ⎛ T ⎞
α
E
1 − α +1 ⎜ BEC ⎟ . (X.25)
α Nk BT 2 ⎝ T ⎠
Consequently, the specific heat at T > TBEC becomes
ζ (α ) ⎛ TBEC ⎞ ⎤
α
⎡
C α Nk B ⎢1 + (α − 1) α +1 ⎜ ⎟ ⎥. (X.26)
⎣⎢ 2 ⎝ T ⎠ ⎦⎥
⎛ ∂E ⎞ ⎛ ∂E ⎞ ⎛ ∂E ⎞
δE =⎜ ⎟ δ T + ⎜ ⎟ δμ = ⎜ ⎟ δ T + α N δμ . (X.27)
⎝ ∂T ⎠ μ ⎝ ∂μ ⎠T ⎝ ∂T ⎠ μ
∂μ
(
+
ΔC = C TBEC ) (−
− C TBEC =αN ) ∂T +
. (X.28)
T =TBEC
−1
⎛ ∂μ ⎞ ⎛ ∂μ ⎞ ⎛ ∂N ⎞ ⎛ ∂N ⎞ ⎛ ∂N ⎞
⎜ ⎟ = −⎜ ⎟ ⎜ ⎟ = −⎜ ⎟ ⎜ ⎟ , (X.29)
⎝ ∂T ⎠ N ⎝ ∂N ⎠T ⎝ ∂T ⎠ μ ⎝ ∂T ⎠ μ ⎝ ∂μ ⎠T
⎛ ∂N ⎞ ⎛ ∂N ⎞
dN = ⎜ ⎟ δT + ⎜ ⎟ δμ = 0 . (X.30)
⎝ ∂T ⎠ μ ⎝ ∂μ ⎠T
⎛ ∂N ⎞ ζ (α − 1) ⎛ N ⎞
⎜ ∂μ ⎟ = ζ α ⎜ k T ⎟, (X.31)
⎝ ⎠T ( ) ⎝ B BEC ⎠
and
⎛ ∂N ⎞ ⎛ N ⎞
⎜ ⎟ =α⎜ ⎟. (X.32)
⎝ ∂T ⎠ μ ⎝ TBEC ⎠
Hence, EQ. (X.29) becomes
⎛ ∂μ ⎞ ζ (α )
⎜ ⎟ = −α k B . (X.33)
⎝ ∂T ⎠ N ζ (α − 1)
ζ (α )
μ −α k B ( T − TBEC ) , (X.34)
ζ (α − 1)
ζ (α )
ΔC −α 2 Nk B . (X.35)
ζ (α − 1)
4.0
3.0
α=3
(C/αNkB)
2.0 2
1.0 3/2
0
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
(T/TBEC)
Fig. X.2.1: The specific heat C of bosons as a function of the reduced temperature (T/TBEC) for
different α values, in units of αNkB,
Next, we discuss important physical properties of realistic atoms that are employed in the studies of
dilute cold gases.
Neutral atoms consist of equal numbers of protons and electrons. Therefore, a charge-neutral bosonic
atom must have an even number of neutrons because the total number of spins of a boson must be an integer.
In the case of alkali atoms, their atomic numbers are odd, so that they are bosons if the neutron number N is
even and are fermions if the neutron number is odd. In Table I.6.2 we list the proton number Z, the neutron
number N and the nuclear spin I of isotopes of alkali atoms. We note that to date successful demonstrations
of BEC have been found mostly in alkali atoms with (S = ½, I = 3/2), including 7Li, 23Na, 87Rb. Additionally,
BEC have been observed in 85Rb with (S = ½, I = 5/2), in 1H with (S = ½, I = 5/2), and in metastable state of
4
He with (S = 1, I = 0).
The ground-state electronic structure of alkali atoms consists of close shells of core electrons plus
one electron in a higher energy outer s-orbit. Therefore, the total electronic angular momentum J of the
valence electron is equal to the electronic spin S = ½ because the orbital angular momentum L = 0. The
Table X.3.1: The proton number Z, neutron number N and nuclear spin number I of hydrogen
the isotopes of some alkali atoms.
Isotope Z N I Statistics
1
H 1 0 1/2 bosons
6
Li 3 3 1 fermions
7
Li 3 4 3/2 bosons
23
Na 11 12 3/2 bosons
39
K 19 20 3/2 bosons
40
K 19 21 4 fermions
41
K 19 22 3/2 bosons
85
Rb 37 48 5/2 bosons
87
Rb 37 50 3/2 bosons
133
Cs 55 78 7/2 bosons
In the absence of an external magnetic field, the atomic energy levels are split by the hyperfine
interaction given by and the following coupling Hamiltonian:
H h f = A IiJ , (X.36)
where A is a constant, and I and J represent the nuclear spin operator and the electronic angular momentum
operator, respectively. The total angular momentum F for the atom is given by
F =I+J. (X.37)
Therefore, the hyperfine interaction may be expressed in terms of the quantum numbers I, J, and F:
1
IiJ = [ F ( F + 1) − I ( I + 1) − J ( J + 1)] . (X.38)
2
For the alkali atoms, we have J = S = ½ and F = I ± ½ so that the hyperfine splitting in zero magnetic-field is
A ⎡⎛ 1 ⎞⎛ 3⎞ ⎛ 1 ⎞⎛ 1 ⎞⎤ ⎛ 1⎞
ΔE h f = ⎜ I + ⎟ ⎜ I + ⎟ − ⎜ I − ⎟ ⎜ I + ⎟ = A⎜ I + ⎟ . (X.39)
2 ⎢⎣⎝ 2 ⎠⎝ 2⎠ ⎝ 2 ⎠⎝ 2 ⎠ ⎥⎦ ⎝ 2⎠
To find an explicit expression for the coefficient A, we consider first-order perturbation theory of the
magnetic dipole interaction between the outer s-electron and the nucleus. Given the definitions of the Bohr
magneton μ B = e ( 2me ) and the nuclear magneton μ N = e ( 2m ) , where me and mp are the masses of
p
the electron and the proton, respectively, the hyperfine splitting may be approximated by the following (in SI
units):
μ0 16π ⎛ μ ⎞⎛ 1 ⎞ 2
ΔE h f = μ B ⎜ NL ⎟ ⎜ I + ⎟ ψ ( 0 ) . (X.40)
4π 3 ⎝ I ⎠⎝ 2⎠
In EQ. (X.40) the quantity ψ ( 0 ) denotes the valence s-electron wave function at the nucleus, and μNL is the
magnetic moment of the nucleus, which is of order of μN. Therefore, we expect that for a multiple-electron
atom, the hyperfine interaction would scale up with increasing electrons. However, the core electrons have
significant screen effects on the charge of the nucleus so that the outermost electron experiences a reduced
nuclear charge and ΔEhf increases much slower than (Z/n)3.
Next, we turn on an external magnetic field B and examine how the energy levels of alkali atoms
evolve with magnetic field. The Hamiltonian involving the interaction between an external magnetic field
and the spin degrees of freedom becomes:
H spin = A I i J + CJ z + DI z , (X.41)
where we have assumed that the magnetic field points along the z-axis so that B = Bzˆ , and the constants C
and D are associated with the Zeeman energies of the electronic and nuclear spins, respectively:
C = g μB B , (X.42)
D = − ( μ NL I ) B . (X.43)
In general the magnitude of D is much smaller than C because |D/C| ~ me/mp ~ 10−3. Therefore, for most
practical purposes we may neglect the term associated with the coefficient D.
As an explicit example, we consider the relatively common case for an alkali atom with quantum
numbers (S = ½, I = 3/2). In the presence of an external magnetic field, we have eight states mI , mJ
associated with the Hamiltonian Hspin in EQ. (X.41). Given that Hspin conserves the z-component of the total
angular momentum, it only couples states with a fixed sum of mI + mJ. Based on this constraint, we may
diagonalize Hspin in a basis consisting of the eight states of mI , mJ where mI = 3/2, 1/2, −1/2, −3/2 and mJ
= 1/2, −1/2. We first note that the states 3 2 ,1 2 and −3 2 , −1 2 clearly do not mix with other states. On
the other hand, states are mixed in pairs like mI , −1 2 and mI − 1,1 2 for mI − 1 2 = ±1, 0 . Therefore, we
obtain the energies of these states as follows.
⎛ 3 1 ⎞ 3 1 3 3 1
E⎜ , ⎟ = A+ C + D ≈ A+ C , (X.44)
⎝ 2 2 ⎠ 4 2 2 4 2
⎛ 3 1 ⎞ 3 1 3 3 1
E⎜ − ,− ⎟ = A− C − D ≈ A− C , (X.45)
⎝ 2 2 ⎠ 4 2 2 4 2
⎛ 3 1 3 ⎞ 3
⎜− A− C + D ⎟ A
⎜ 4 2 2 2 ⎟ (X.46)
⎜ 3 1 1 1 ⎟
⎜ A A+ C + D⎟
⎝ 2 4 2 2 ⎠
and obtain the following eigen energies for this pair of states:
A 3 1 ⎛ 1 3 1⎛ C⎞ ⎞
2
E=− +D± 2
A + ( A + C − D) 2
≈ A⎜ − ± + ⎜1 + ⎟ ⎟ . (X.47)
4 4 4 ⎜ 4 4 4⎝ A⎠ ⎟
⎝ ⎠
Similarly, for the states −3 2 ,1 2 and −1 2 , − 1 2 , we diagonalize the corresponding (2×2) matrix:
⎛ 3 1 3 ⎞ 3
⎜− A+ C − D ⎟ A
⎜ 4 2 2 2 ⎟ (X.48)
⎜ 3 1 1 1 ⎟
⎜ A A− C − D⎟
⎝ 2 4 2 2 ⎠
and obtain the following eigen energies for this pair of states:
A 3 1 ⎛ 1 3 1⎛ C⎞ ⎞
2
Finally, for the states 1 2 , − 1 2 and −1 2 ,1 2 , we diagonalize the corresponding (2×2) matrix:
⎛ 1 1 1 ⎞
⎜− 4 A+ 2C − 2 D A ⎟
⎜ ⎟ (X.50)
⎜⎜ 1 1 1 ⎟
A − A− C + D⎟
⎝ 4 2 2 ⎠
A 1 ⎛ 1 1⎛C⎞ ⎞
2
(C − D )
2
E=− ± A2 + ≈ A⎜ − ± 1+ ⎜ ⎟ ⎟ . (X.51)
4 4 ⎜ 4 4⎝ A⎠ ⎟
⎝ ⎠
In the low field limit, the solutions given in EQs. (X.44), (X.45), (X.47), (X.49) and (X.51) converge to two
hyperfine levels at E = 3A/4 and –A/4. On the other hand, in the high magnetic field limit, we find that all
eigen-energies become approximately equal to ± μ B B .
For most experiments on alkali atoms, the magnetic fields are relatively small so that the energy
associated with a state of the quantum number F , mF may be expressed by the following (to the first order
in magnetic field):
E ( F , mF ) = E ( F ) + mF g F μ B B , (X.52)
where E ( F ) is the energy for B = 0 and gF is the Landé g factor. In the case of F = I + 1/2, the electronic
spin aligns with the total spin so that the Landé g factor is negative. Consequently the state with mF = F, also
known as the doubly polarized state mI = I , mJ = 1 2 , has the highest energy. On the other hand, for F = I
− 1/2 the electronic spin is antiparallel to the total spin so that the Landé g factor is positive. In this case the
state mF = −F, also known as the maximally stretched state F = I − 1 2 , mF = − ( I − 1 2 ) , has the highest
energy. These two states are important because they have negative magnetic moments and are known as the
low-field seekers that may be magnetically trapped around a local minimum in the magnetic field.
The aforementioned example for alkali atoms with the quantum numbers (S = ½, I = 3/2) can be
easily generalized for alkali atoms with other quantum numbers and also for hydrogen atoms. In the next
section we briefly discuss some of the primary methods used for trapping and cooling alkali atoms.
The principle of magnetic trapping is based on the Zeeman effect: the energy of an atomic state
depends on the magnetic field, so that an atom in a spatially inhomogeneous magnetic field will experience a
spatially varying potential. In analogy to the discussion that leads to EQ. (X.52) in the previous subsection,
the energy Ei of a particular state i in a magnetic field B may be generally written as
Ei = Ai − μi B , (X.53)
where Ai is a constant, and μi denotes the effective magnetic moment of the state. If the magnetic moment is
positive, the atom experiences a force that tends to drive it towards regions of higher fields. On the other
hand, if the magnetic moment is negative, the force experienced by the atom tends to drive it towards lower
fields. Therefore, states with positive magnetic moments are referred to as high-field seekers, whereas those
with negative magnetic moments are low-field seekers. According to EQ. (X.53), it is apparent that the
energy depth of the magnetic traps is determined by the Zeeman energy μiB.
To construct a magnetic trap one must design the magnetic field configurations to achieve a local
extreme to attract either the high-field or low-field seekers. However, a local maximum in |B| has been
proven impossible in regions without electrical currents. [See the theorem by W. H. Wing in Prog. Quantum
Electronics 8, 181 (1984)]. Thus, only low-field seekers with negative magnetic moments can be trapped by
magnetic fields with a local minimum in |B|. Generally, field configurations with a local |B| minimum may
be further divided into two classes. One class involves the minimum being zero, and the other has a finite
minimum field.
A simple design that yields a magnetic field configuration vanishing at some point can be achieved
by pair of opposed Helmholtz coils. This quadrupole configuration yields linearly varying magnetic fields
with distance in all directions. If we define z as the direction of axial symmetry and denote the magnetic field
gradients along the x and y axes by b, the condition ∇iB = 0 for an origin chosen at the location where B = 0
yields the spatial dependence:
B = bx xˆ + by yˆ − 2bz zˆ . (X.54)
( )
1/ 2
The magnitude of the field configuration in EQ. (X.54) satisfies B = b x 2 + y 2 + 4 z 2 , which varies
linearly with distance from the minimum with an anisotropic slope depending on the direction.
In reality the quadrupole magnetic trap suffers from a major disadvantage because the approach has
implicitly assumed that the atoms remain in the quantum state throughout the trap. This assumption would
have been a good approximation had the field experienced by an atom changed slowly with time. However, a
moving atom experiences a time-dependent magnetic field, which will induce transitions between different
states. In particular, for very small magnetic fields, the hyperfine energy separations become vanishingly
small so that atoms in low-field seeking states can make transitions to high-field seeking states, thereby being
ejected from the trap. Consequently, the quadrupole trap effectively has a “hole” near the node of the field
configuration, which limits the time for which atoms may be stored in the trap.
There are a number of ways to circumvent the problem. One straightforward approach is to work
with a magnetic field configuration of a non-zero minimum. The other approach is to apply a laser field in
the region of the node so that the resulting radiation forces repel atoms from the vicinity of the hole. Another
method is to “plug the hole” in the trap by applying an oscillating bias magnetic field so that the time-
averaged magnitude of the magnetic field never vanishes anywhere in the trap. The last scheme for a
modified quadrupole trap is known as the time-averaged orbiting potential (TOP) trap with the following
instantaneous field:
where b0 and ω are the magnitude and angular frequency of the oscillating magnetic field. From EQ. (X.55)
we obtain the time-averaged field
b2
B
t
= b0 +
4b0
(x 2
)
+ y 2 + 8z 2 , (X.56)
which never vanishes anywhere so that there are no problems with the presence of a “hole” in the trap.
Moreover, from EQs. (X.53) and (X.56), we find that the angular frequencies for atomic motion in the three
principle directions are:
b2
ω x2 = ω y2 = − μi , (X.57)
2mb0
4b 2
ω z2 = 8ω x2 = − μi . (X.58)
mb0
Thus, the magnetic trap is consistent with a three-dimensional anisotropic harmonic-oscillator potential.
H ′ = − d iE = −E ( d iεˆ ) , (X.59)
where d is the electric dipole moment operator of the atom. For a time-dependent electric field with a
frequency ω, E ( ω , t ) = Eω e − i ω t + E −ω ei ω t , we illustrate the diagrammatic representation for its second-order
contributions to the ground state energy of an atom in Fig. X.4.1. Here g and e denote the atomic
ground state and excited state, respectively. Noting that the electric field is real so that the condition
E −ω = Eω∗ is satisfied, we find from the diagrammatic expression that the ground-state energy shift ΔEg is
given by
⎡ 1 1 ⎤
ΔE g = ∑ e ⎢ g d iEω e e d iE −ω g + g d iE −ω e e d iEω g ⎥
⎣ Eg − Ee + ω E g − Ee − ω ⎦
⎛ 1 1 ⎞
= ∑ e e d iεˆ g
2 2
⎜⎜ + ⎟⎟ Eω
⎝ E g − Ee + ω E g − Ee − ω ⎠
1
= − α (ω ) E ( r, t )
2 2
= − α ( ω ) Eω , (X.60)
2 t
ω −ω −ω ω
g e g g e g
⎛ 1 1 ⎞
α ( ω ) = ∑ e e diεˆ g
2
⎜⎜ + ⎟⎟
⎝ Ee − Eg + ω Ee − Eg − ω ⎠
= 2∑ e e d iεˆ g
2 (E e − Eg ) . (X.61)
(E )
2
−( ω)
2
e − Eg
If we further incorporate a realistic finite lifetime 1/Γe for the excited state, the polarizability becomes a
complex quantity:
2
e diεˆ g
α (ω ) ≈ ≡ α ′ ( ω ) + iα ′′ ( ω ) (X.63)
Ee − i Γ e 2 − E g − ω
so that
α ′ (ω ) ≈
(E e − Eg − ω ) e d iεˆ g
2
, (X.64)
( )
2
+ ( Γe 2)
2
Ee − E g − ω
2
( Γ e 2 ) e d iεˆ g
α ′′ (ω ) ≈ . (X.65)
(E ) +(
2
Γe 2 )
2
e − Eg − ω
From EQs. (X.61) and (X.63), we find that the energy shift is also a complex quantity given by
ΔEg = Vg − i Γ g 2 , (X.66)
where Γg corresponds to the rate of loss of atoms from the ground state. We may further define the detuning
( ) ( )
frequency δ so that δ ≡ ω − Ee − E g ≡ ω − ωeg . If we introduce the Rabi frequency ΩR:
Ω R = e d iEω g , (X.67)
the ground-state energy shift Vg and the rate of loss of atoms Γg in EQ. (X.66) become
1 δ Ω 2R
Vg = − α ′( ω ) E ( r, t )
2
= , (X.68)
δ 2 + ( Γe 2)
2
2 t
Γ g = α ′′( ω ) E ( r, t )
2
=
(Γ e
2 ) Ω 2R
. (X.69)
δ 2 + ( Γe 2 )
2
t
Conceptually the ground-state energy shift of an atom due to its interaction with the radiation field
may be viewed as an effective potential Vg in which the atom moves. Therefore, for spatially varying time-
averaged electric field, there is a force experienced by an atom, which is known as the dipole force:
1
Fdipole = − ∇Vg ( r ) = α ′( ω ) ∇ E ( r, t ) .
2
(X.70)
2 t
We further note that the polarizability changes sign at the transition frequency according to EQ. (X.64).
Therefore, at low frequencies the polarizability is positive and the force is towards the regions of higher
electric field. On the other hand, for frequencies above the resonance frequency, the polarizability becomes
In addition to the dipole force due to the energy-level shifts associated with virtual transitions
between atomic states, there is another force due to real transitions. That is, the force is the consequence of
the momentum of a photon being imparted to or removed from an atom during the absorption or emission
process. In this context, if we know the rate of loss of atoms from the ground state Γg and if the electric field
is a traveling wave with a wave vector q, we may compute the force due to absorption of photons by the
following relation:
Frad = qΓ g . (X.71)
We remark that both forces described by EQs. (X.70) and (X.71) play an important role in laser cooling.
Next, we discuss laser cooling by considering two oppositely directed laser beams of the same
frequency ω and intensity. Assuming that the lasers are slightly red detuned from a resonant frequency ωeg,
we find that the photon absorption rate (dNph/dt) is equal to the rate of loss of atoms from the ground state Γg
so that
dN ph
= Γ g ≡ π Ω 2R L( ω ) , (X.72)
dt
where L( ω ) ≡
( Γ e 2π ) (X.73)
( )
2
ω − ωeg + ( Γ e 2 )
2
is a Lorentzian function in frequency. Now suppose that an atomic cloud is moving to the right with a
velocity vz and that the laser beams are pointing along the left and right directions, we obtain two different
phonon absorption rates for the two laser beams due to Doppler shifts:
R
dN ph
= π Ω 2R L( ω − vz q ) , (X.74)
dt
L
dN ph
= π Ω 2R L( ω + vz q ) , (X.75)
dt
where q ≡ (ω/c) is the photon momentum. Consequently, there is a net frictional force (dpz/dt) on the right
moving atom, which is given by:
⎛ dN ph
R
dN phL
⎞
⎟⎟ = qπ Ω R [ L( ω − vz q ) − L( ω + vz q )]
dpz 2
≡ −γ vz = q ⎜ −
⎜
dt ⎝ dt dt ⎠
⎡ dL( ω ) ⎤
− ⎢ 2 q 2π Ω 2R vz . (X.76)
⎣ d ω ⎥⎦
In the second line of EQ. (X.76) we have made the assumption that the Doppler shift is small relative to both
the linewidth Γe and the detuning frequency δ.
From EQ. (X.76) we obtain a characteristic brake time τ fric for a moving atom:
1 1 dpz γ ⎡ 2 q 2π Ω 2R dL( ω ) ⎤
≡− = ⎢ ⎥. (X.77)
τ fric pz dt m ⎣ m dω ⎦
Apparently the brake time τfric can be very short for narrow lines according to EQs. (X.73) and (X.77). In
other words, the frictional coefficient γ can be very large for large (dL/dω) in the case of narrow lines. The
configuration of oppositely directed laser beams is referred to as optical molasses.
To estimate the achievable lowest kinetic energies of the atoms subjected to optical molasses, we
recognize that the absorption and emission of photons results in heating of atoms. Therefore, the steady-state
kinetic energy of an atom subjected to optical molasses is determined by balancing the rate of heating with
that of cooling, and the latter is provided by the frictional force. To determine the rate of heating, we
consider the random walk of an atom in the small velocity limit so that the Doppler shifts may be neglected.
For simplicity, we further assume a one-dimensional problem so that the direction of random walk aligns
with the laser beams. The momentum diffusion coefficient (Dp) associated with the random walk is given by
the rate of change in the mean-square momentum of the atom, d〈pz2〉/dt, and the latter is due to absorption
and emission of photons induced by both laser beams. That is,
= 2 ⎡( q ) 2π Ω 2R L(ω ) ⎤ = 4 ( q ) π Ω 2R L( ω ) .
2 2
(X.78)
⎣ ⎦
where we have used the relation in EQ. (X.73). Quantitatively, we note that for sodium atoms, the value of
Γ e is 480 μK. Consequently, our estimate of the minimum temperature achievable by laser cooling in
sodium is 240 μK according to EQ. (X.80).
As mentioned earlier, in addition to the cooling effects, radiation pressure may give rise to
confinement of atoms in space. One of the atomic trapping mechanisms is the magneto-optical trap (MOT)
that consists of a combination of laser beams and a spatially varying magnetic field. The principle of
operation is based on the fact that atomic energy levels are dependent on the spatially varying magnetic field
so that the radiation pressure is also spatially dependent. A standard three-dimensional MOT consists of six
laser beams (three pairs of counter-propagating beams along three principle axes) and a quadrupole magnetic
trap. The use of MOT’s is universal among experiments of cold alkali atoms because they are capable of
trapping atoms at the same time cooling them.
Therefore, we have seen that the interaction of atoms with laser beams may lead to both optical
trapping and laser cooling by means of the polarization of atoms and the Doppler effect on atoms. It is
interesting to note that the theoretically predicted temperature ( Γ e ) / 2 achievable by optical cooling was in
fact significantly higher than what was ultimately reached by experiments. This finding led to the discovery
of new mechanisms of cooling atoms to temperatures corresponding to thermal energy on the order of the
recoil energy (Er), which refers to the energy imparted to an atom of mass m at rest through the absorption of
a photon with momentum q :
( q)
2
Er = . (X.81)
2m
The temperature corresponding to the recoil energy is generally on the order of 0.1 ~ 1 μK, which is two to
four orders of magnitude lower than the temperatures achievable by the Doppler mechanism discussed
earlier. The underlying mechanisms that led to further cooling of alkali atoms are due to two features
unaccounted for by the Doppler mechanism. One feature is associated with the degenerate ground states in
the absence of magnetic field. These degenerate states become split if the atom is subjected to circularly
polarized radiation, and the population of different sub-levels may be controlled via optical pumping. The
other feature is due to the spatially inhomogeneous potential induced by counter-propagating laser beams. In
particular, by properly choosing the polarizations of the counter-propagating laser beams, one may achieve a
periodic potential for atoms, which is known as an optical lattice. The potential may evolve periodically with
positions from being linearly polarized to circularly polarized. When an atom moves into a red-detuned
circularly polarized potential with an increasing radiation-induced potential energy, it will lose its kinetic
energy by virtue of energy conservation and at the same time it has the tendency to be optically pumped from
a higher-energy sub-level to a lower-energy sub-level if the spontaneous emission rate is faster than the rate
of the radiation-induced reverse process. On the other hand, if an atom moves into a circularly polarized
potential with a decreasing radiation-induced potential energy, it will gain kinetic energy while at the same
time its rate of being optically pumped into a higher-energy sub-level will be reduced. Because of the
correlation between the pumping rate and the energy shifts, for atoms with a finite thermal spread of
velocities, there is a net tendency for them to lose kinetic energy irrespective of their direction of motion, and
optical pumping for red-detuned radiation also tends to repopulate the low-energy sub-levels at any point in
space. These processes are repeated as atoms continue to experience the spatially varying radiation potential,
leading to continuous cooling of atoms. The combined mechanisms are referred to as Sisyphus cooling by
Dalibard and Cohen-Tannoudji because it resembles the eternal punishment of Sisyphus in the Greek myth,
in which Sisyphus was condemned to repeatedly push a heavy rock up a steep hill. For more detailed analysis
of the Sisyphus cooling process, you may refer to the paper by J. Dalibard and C. Cohen-Tannoudji in J. Opt.
Soc. Am. B 6, 2023 (1989).
While the temperatures achieved by laser cooling were very low, they were still not sufficiently cold
to produce BEC in the trapped dilute gases. For all experiments successfully produced BEC of alkali atoms
to date, evaporative cooling after laser cooling was necessary. The basic principle of evaporative cooling is
to allow particles with energies higher than the average particle energy of a system to escape so that the
remaining particles are cooled. This process may be realized by applying radio-frequency (rf) radiation to
trapped atoms to induce spin-flip transitions so that the initial low-field seekers become high-field seekers,
thereby being expelled from the trap. Given that the resonant frequency for atoms in a trap is spatially
inhomogeneous, as atoms are lost from the trap and cooling proceeds, we may adjust the frequency of the rf-
radiation steadily to allow loss of atoms with lower and lower energies. More extensive discussions of the
evaporative cooling may be found in the review article by W. Ketterle and N. J. van Druten in Adv. At. Mol.
Opt. Phys. 37, 181 (1996).
In this subsection we consider the structure of the BEC state in the presence of interaction. We shall
focus on the zero-temperature properties of the non-uniform Bose gas system with a scattering length a much
smaller than the average inter-particle separation n−1/3. This ground state of trapped bosons may be described
by a mean-field approximation known as the Gross-Pitaevskii equation.
Our starting point is that the effective interaction between two low-energy particles may be described
by a contact interaction U0δ (r−r'), where r and r' denote the positions of the two particles. This contact
interaction potential is a constant in the momentum space and is directly related to the scattering length a:
2
4π
U0 = a. (X.82)
m
Next, we adopt the mean-field approximation for the many-body system, and we assume that the wave
function is a symmetrized product of single-particle wave functions. In the fully condensed state, all bosons
are in the same single-particle ground state, φ (r), and the single-particle wave function is normalized
according to the standard expression:
2
∫ dr φ ( r ) =1. (X.83)
Therefore, the wave function of the N-particle system may be written as:
N
Ψ ( r1 , r2 , … , rN ) = ∏ φ ( ri ) . (X.84)
i =1
In the presence of an external potential Vex(r), the effective Hamiltonian H for the system becomes:
⎡ pi2 ⎤
( )
N
H = ∑⎢ + Vex ( ri ) ⎥ + U 0 ∑ δ ri − r j . (X.85)
i =1 ⎣ 2m ⎦ i< j
From EQs. (X.84) and (X.85), we obtain the energy of the many-body system:
⎡ 2 2 2 ( N − 1) 4⎤
E = N ∫ dr ⎢ ∇φ ( r ) + Vex ( r ) φ ( r ) + U0 φ(r ) ⎥ . (X.86)
⎣ 2m 2 ⎦
In the large N limit, it is convenient to introduce the wave function of the condensate:
ψ ( r ) ≡ N 1/ 2φ ( r ) (X.87)
and the energy of the system in EQ. (X.86) may be rewritten into
2
⎡ 2 2 1 4⎤
E (ψ ) = ∫ dr ⎢ ∇ψ ( r ) + Vex ( r ) ψ ( r ) + U 0 ψ ( r ) ⎥ . (X.89)
⎣ 2m 2 ⎦
is conserved. The constraint is satisfied if we take δ E − μ δ N = 0 , where μ denotes the chemical potential.
This procedure is equivalent to minimizing the quantity (E − μN) at a constant chemical potential μ. Thus, by
taking the variations of (E − μN) to 0 with respect to ψ*(r) in EQ. (X.89), we obtain the time-independent
Gross-Pitaevskii equation:
2
2
− ∇ 2ψ ( r ) + Vex ( r )ψ ( r ) + U 0 ψ ( r ) ψ ( r ) = μψ ( r ) . (X.91)
2m
It is interesting to note that EQ. (X.91) is analogous to EQ. (IX.103) that we have derived earlier for the
order parameter of the condensate in Part IX.4 if we specify the contact interaction U 0δ ( r − r ′ ) for the
interaction potential and also include an external potential into EQ. (IX.103). Moreover, we find that for a
uniform Bose gas, the Gross-Pitaevskii equation yields a chemical potential μ = U 0 ψ ( r ) = nU 0 , where n ≡
2
(N/Ω) and Ω is the volume of the Bose gas. This chemical potential μ is consistent with the ground-state
energy E = Ω n 2U 0 2 derived from EQ. (X.89) for a uniform Bose gas because we have μ = ∂E ∂N = nU 0
from EQ. (X.89).
To investigate the ground-state properties of bosons in a trap, we may specify the trapping potential
Vex(r) in EQ. (X.90) and then examine the resulting solutions. For simplicity, we consider in the following a
specific example of an anisotropic harmonic oscillator trap
1
Vex ( r ) = Vex ( x, y, z ) =
2
(
m ω12 x 2 + ω22 y 2 + ω32 z 2 .) (X.92)
The solutions to the Gross-Pitaevskii equation may be obtained numerically by inserting the explicit form of
the trapping potential in EQ. (X.92) into EQ. (X.91). However, it is physically more instructive to consider
some analytical solutions under different approximations. We examine below three specific examples.
Our first example is a variational calculation based on a Gaussian trial function for ψ (r). This
approach is motivated by the fact that the lowest-energy single-particle state in the absence of interparticle
interactions acquires the following wave function:
1 ⎡ 1 ⎛ x2 y 2 z 2 ⎞⎤
φ0 ( r ) = exp ⎢ − ⎜ 2 + 2 + 2 ⎟ ⎥ , (X.93)
π 3/ 4 ( a1a2 a3 )
1/ 2
⎣ 2 ⎝ a1 a2 a3 ⎠ ⎦
particle interactions so that the dimensions of the cold gas cloud become modified. We adopt the same form
in EQ. (X.93) as the trial wave function for ψ (r):
N 1/ 2 ⎡ 1 ⎛ x2 y 2 z 2 ⎞⎤
ψ (r ) = exp ⎢ − ⎜ 2 + 2 + 2 ⎟ ⎥ , (X.94)
π 3/ 4 ( b1b2b3 )
1/ 2
⎣ 2 ⎝ b1 b2 b3 ⎠ ⎦
We may further simplify EQ. (X.95) by considering the strong interaction limit so that the interaction energy
per particle is large compared with ωi for all ω i. Thus, we may neglect the kinetic energy term (X.e. the
first term) in EQ. (X.95) when minimizing the energy relative to bi. Using EQ. (X.82) to replace U0 by the
scattering length a and the following definitions for the characteristic length a and the oscillator frequency
ω:
1/ 2
⎛ ⎞
ω ≡ ( ω1ω2ω3 ) ,
1/ 3
a ≡⎜ ⎟ , (X.96)
⎝ mω ⎠
we obtain the variational parameters
1/10 1/ 5 1/ 2 1/10 1/ 5
⎛2⎞ ⎛ Na ⎞ ⎛ ω ⎞ ⎛2⎞ ⎛ Na ⎞ ⎛ ω ⎞
bi = ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ =⎜ ⎟ ⎜ ⎟ ⎜ ⎟a. (X.97)
⎝π ⎠ ⎝ a ⎠ ⎝ ωi ⎠ ⎝π ⎠ ⎝ a ⎠ ⎝ ωi ⎠
Here we note that bi increases with increasing repulsive interaction U0, which implies that the dimensions of
the gas cloud expands with increasing repulsive interaction. Inserting EQ. (X.97) into EQ. (X.95), we find
that the leading contributions to the energy per particle (E/N) is given by:
1/ 5 2/5
E 5 ⎛ 2 ⎞ ⎛ Na ⎞
= ⎜ ⎟ ⎜ ⎟ ω. (X.98)
N 4 ⎝π ⎠ ⎝ a ⎠
From EQ. (X.98) we note that the energy per particle of the interacting Bose gas is larger than that of the
non-interacting system by approximately a factor of ( Na a )
2/5
.
Next, we consider the second case known as the Thomas-Fermi approximation. As seen in the
previous example, in the limit of large N and for repulsive interactions, it is justifiable to neglect the kinetic
energy because it is much smaller than the potential energy. A better approximation for the condensate wave
function in this limit is to solve the Gross-Pitaevskii equation directly by dropping off the kinetic term.
Hence, EQ. (X.91) becomes
⎡Vex ( r ) + U 0 ψ ( r ) 2 ⎤ψ ( r ) = μψ ( r ) . (X.99)
⎣ ⎦
n( r ) = ψ ( r ) = [ μ − Vex ( r )] U 0 ,
2
(X.100)
which implies that the wave function is only non-trivial if μ = ⎡⎣Vex ( r ) + n( r ) U 0 ⎤⎦ ≥ Vex ( r ) and that ψ = 0 if
μ < Vex ( r ) . Therefore, the boundary of the cold gas cloud occurs at Vex ( r ) = μ , and the chemical potential
required adding a particle inside the cloud is equal to the sum of the external potential and the contribution
from interactions, n( r ) U 0 . For an anisotropic harmonic oscillator potential given in EQ. (X.92), we obtain
the following relation between N and μ by using EQs. (X.90) and (X.100):
3/ 2
8π ⎛ 2 μ ⎞ ⎛ μ ⎞
N= ⎜ ⎟ ⎜ ⎟, (X.101)
15 ⎝ mω 2 ⎠ ⎝ U0 ⎠
so that
2/5
152 / 5 ⎛ Na ⎞
μ= ⎜ ⎟ ω. (X.102)
2 ⎝ a ⎠
Given μ = ( ∂E ∂N ) and μ ∝ N 2 / 5 from EQ. (X.102), we find that the energy per particle is
2/5
E 57 / 532 / 5 ⎛ Na ⎞
5
= μ= ⎜ ⎟ ω. (X.103)
N 7 14 ⎝ a ⎠
This energy for the Thomas-Fermi approximation is slightly smaller (by about 8%) than that given in EQ.
(X.98) from the variational calculations.
Now we turn to a special situation for N particles trapped in a three-dimensional isotropic harmonic-
oscillator potential so that Vex ( r ) = mω02 r 2 2 . The Gross-Pitaevskii equation for the ground-state wave
function is
⎡ 2
d ⎛ 2 d ⎞ 1 4π 2 a 2⎤
⎢ − 2 ⎜ r ⎟ + mω 2 2
0 r + ψ ( r ) ⎥ψ ( r ) = μψ ( r ) . (X.104)
⎣ 2mr dr ⎝ dr ⎠ 2 m ⎦
Defining ρ ( r ) ≡ rψ ( r ) , we have
⎡ 2
d 2ρ 1 4π 2 a ⎤
⎢ 2mr 2 dr 2 2 0 ( ) 2
− + mω 2
r 2
− R 2
ρ ( r ) + 2
ρ ( r ) ρ ( r )⎥ = 0 , (X.105)
⎣ mr ⎦
where R is the range of the harmonic oscillator potential so that μ = mω02 R 2 2 . In the Thomas-Fermi limit,
the solution to EQ. (X.105) becomes
1/ 2 1/ 2
⎡ m 2ω02 2 2 ⎤ ⎡ 1 ⎤
2 (
ρ(r ) = r ⎢ R − r )⎥ =r⎢ 4 (
R 2 − r 2 ) ⎥ = rψ ( r ) , (X.106)
⎣ 8π a ⎦ ⎣ 8π a a ⎦
which vanishes at r = R.
2
2
2
= μ = U0 ψ 0 , (X.108)
2mξ
so that
2
1
ξ2 = = , (X.109)
2mnU 0 8π na
⎛ z ⎞
ψ ( z ) = ψ 0 tanh ⎜ ⎟, (X.110)
⎝ 2ξ ⎠
which is essentially identical to what we have derived in EQ. (IX.149) for superfluid. The coherence length
in EQ. (X.109) is also similar to that in EQ. (IX.148). Therefore, the Gross-Pitaevskii equation for a
condensate in a trapping potential reveals many features that are consistent with our earlier findings from
diagrammatic approximations.
At this point we shall not proceed further with discussion of bosons in the forms of dilute cold gases.
You may consult the references listed below for additional in-depth information about the current status of
research in cold gases.
Further Readings:
1. S. Chu and C. Cohen-Tannoudji, and W. D. Phillips: Rev. Mod. Phys. 70, 685 (1998).
2. M. H. Anderson et al.: Science 269, 198 (1995).
3. K. B. Davis et al.: Phys. Rev. Lett. 75, 3969 (1995).
4. C. C. Bradly et al.: Phys. Rev. Lett. 75, 1687 (1995); Phys. Rev. Lett. 78, 985 (1997).
5. J. Dalibard and C. Cohen-Tannoudji, J. Opt. Soc. Am. B 6, 2023 (1989).
6. W. Ketterle and N. J. van Druten, Adv. At. Mol. Opt. Phys. 37, 181 (1996).
7. C. J. Pethick & H. Smith, “Bose-Einstein Condensation in Dilute Gases”, Cambridge University Press,
Cambridge, UK (2002).
Tc(K)
HgBa2Ca2Cu3Ox [> 1 GPa]
150 (1992)
HgBa2Ca2Cu3Ox [1 atm]
Tl2Ba2Ca2Cu3Ox macroscopic
120 theory of
superfluidity &
superconductivity
90 YBa2Cu3O7−δ (2003)
liquid
nitrogen LaO1-xFxFeAs
(2008)
60
discovery of BCS MgB2
superconductivity theory La2-xBaxCuOy (2001)
Nb3Ge (1986)
30
Nb high-temperature
liquid (1911) Pb Josephson
helium Hg superconductivity
effect
0
1900 1920 1940 1960 1980 2000
Year
Fig. XI.1.1 Summary of the time evolution of maximum superconducting transition temperature (Tc) and
various significant events in the research development of superconductivity.
This Chapter is structured as follows. We begin in Part XI.1 with a review of phenomenology of
superconductivity, including the experimental signatures and phenomenological theories of London-Pippard
and Landau-Ginzburg. In Part XI.2 we first establish the theoretical foundation for the formation of Cooper
The occurrence of superconductivity is accompanied by several special phenomena that are regarded
as the signatures of superconductivity. These signatures include zero resistance, Meissner effect, persistent
currents and fluxoid quantization, magnetic critical fields, macroscopic phase coherence and quantum
interference (due to Josephson pair tunneling effect), and gapped excitations.
Fig.XI.1.2 Vanishing electrical resistance (R) of a superconductor below the critical temperature Tc,
provided that the electrical current (I) passing through the superconductor does not exceed a
temperature-dependent critical current Ic (T).
Although a very simple signature in its manifestation, a rigorous proof for zero resistive response of
superconductors to an electromagnetic field in the dc limit is by no means trivial. Specifically, Mattis and
Bardeen [D. C. Mattis and J. Bardeen, Phys. Rev. 111, 412 (1958)] have shown, using first-order time-
dependent perturbation theory for a perturbative dipole field, that the frequency (ω) dependent complex
H ex = e st ∑ ω , j H ω ( r j ) eiωt = ∑ j A( r j , t ) i p j (XI.1)
⎛ σ 1S ⎞ 2 ∞ 1 −Δ
⎜ ⎟ = ∫Δ dE [ f ( E ) − f ( E + ω )] g ( E ) + ∫Δ−ω dE [1 − 2 f ( E + ω )] g ( E ) , (XI.2)
⎝ σN ⎠ ω ω
⎛ σ 2S ⎞ 1 Δ [1 − 2 f ( E + ω )] ( E + Δ + ω E ) 2 2
⎜ ⎟ = ∫Δ−ω ,( −Δ ) dE 2 , (XI.3)
( ) ⎣(
1/ 2
⎝ σN ⎠ ω Δ − E 2 1/ 2
⎡ E + ω )
2
− Δ 2
⎤
⎦
f ( E ) ≡ ⎡⎣1 + e βE −1
⎤⎦ , g( E ) ≡
( E + Δ + ωE )2 2
. (XI.4)
( E − Δ ) ⎡⎣( E + ω ) − Δ
1/ 2
2 1/ 2
2 2 2
⎤
⎦
In analogy to the two-fluid model of superfluidity, the real part of the conductivity σ1S(ω) may be associated
with the response of the “normal fluid” component of the superconductor to external fields, which is
dissipative in nature; whereas the imaginary component σ2S(ω) may be considered as the “superfluid”
response.
Although we shall not go through the detailed derivations of the results in EQs. (XI.2) – (XI.4) and
only refer you to the original paper by Mattis and Bardeen for details, we remark that the approach is
essentially the same as the Kubo formalism for the linear response of electrical currents to an external vector
potential A, and the vector potential is subject to the London gauge condition ∇i A = 0 , which implies that
there are no external charges in the sample and that electrical neutrality be maintained throughout. The
electromagnetic fields are thus expressed by E = − ( ∂A ∂t ) c and H = ∇ × A . Next, to derive the complex
conductivity, we consider the relationship between the current density j ( r , t ) and the vector potential. For
wave functions Ψ k ( r, t ) in the presence of an external field, we may express them in terms of unperturbed
wave functions and unperturbed eigen-energies ε k by the following (taking = 1):
Ψ k ( r, t ) = ⎡⎣ψ k ( r ) + ∑ k ′ ak ′k ( t )ψ k ′( r ) ⎤⎦ e − iε k t , (XI.5)
where the coefficients in the first-order time-dependent perturbation theory are given by
k′ Hω k
ak ′k ( t ) = lim s →0+ e st ∑ ω , (XI.6)
ε k − ε k ′ − ( ω − is )
k ′ Hω k
ak∗ ′k ( t ) = lim s →0+ − e st ∑ ω , (XI.7)
ε k − ε k ′ + ( ω − is )
where the condition H ω∗ = H ω has been imposed because of reality. Thus, the expression for the current
density is given by
e2
j ( r, t ) = −
e
∑ ( ) ∑ fk A( r , t ) Ψ k ( r , t )
2
fk Ψ ∗k ∇Ψ k − Ψ k ∇Ψ ∗k −
2mi k mc k
ne 2
∑ k ,k ′ ( )
e
=− fk ψ k∗ ∇ψ k ′ ak ′k + ak∗′kψ k∗′∇ψ k − ( c.c.) − A( r , t ) , (XI.8)
2mi mc
where fk ≡ f ( ε k ) denotes the Fermi-Dirac function, and the condition for conservation of carrier density n
is satisfied:
n = ∑ k fkψ k∗ψ k . (XI.9)
In general the expression in EQ. (XI.8) is applicable to both normal metals and superconductors. To
proceed further with the calculations for the conductivity, we must evaluate the density matrix
ρε = ψ k∗( r )ψ k ( r ) , (XI.10)
where 〈…〉 represents an average over an energy shell. The density matrix may be replaced by the equal-time
Green function for normal metals. For superconductors, the formation of Cooper pairs leads to bosonic Green
functions so that additional coherence factors as well as anomalous Green functions must be included. We
shall discuss the Green function expressions for superconductors in Part XI.2.
Returning to the complex conductivity derived by by Mattis & Bardeen, we note that at T = 0, the
probability for occupation of quasiparticles becomes zero because f (E) = 0, so that σ1S(ω) and σ2S(ω) in EQs.
(XI.2) and (XI.3) can be simplified into:
⎛ σ 1S ⎞ ⎛ 2Δ ⎞ 4Δ ⎛ 2Δ − ω ⎞
⎜ ⎟ = ⎜1 + ⎟E ( x ) − K ( x ) , ⎜ x ≡ 2Δ + ω , ω ≥ 2Δ ⎟ (XI.11)
⎝ σN ⎠ ⎝ ω ⎠ ω ⎝ ⎠
⎛ σ 2S
⎜
⎝ σN
⎞ 1 ⎛ 2Δ ⎞
⎟ = ⎜1 +
⎠ 2⎝ ω ⎠
1 ⎛ 2Δ ⎞
⎟ E ( x′ ) − ⎜ 1 − ⎟ K ( x′ ) ,
2⎝ ω ⎠
( x′ ≡ 1 − x2 ) (XI.12)
1 −1/ 2
( )
1 π /2
K ( x ) ≡ ∫ dt =∫ dθ 1 − x sin 2θ . ( 0 ≤ x ≤ 1) (XI.14)
(1 − xt ) (1 − t )
0 2
1/ 2 2
1/ 2 0
In the low-frequency limit ω << Δ, the imaginary part of the conductivity (σ2S) that corresponds to
the superfluid response becomes:
⎛ σ 2S ⎞ πΔ ⎛ Δ ⎞
⎜ ⎟= tanh ⎜ ⎟. (ω << Δ) (XI.15)
⎝ σN ⎠ ω ⎝ 2k BT ⎠
In Fig. XI.1.3 we illustrate the frequency dependence of the complex conductivity at T = 0, and we
note that σ1S completely vanishes for 0 < ω < Δ, implying no dissipative excitations can occur below the
superconducting gap. In contrast, the superfluid response σ2S diverges with decreasing frequency the ω → 0
limit, indicating infinite conductivity and therefore zero resistivity in the dc limit.
⎛ σS ⎞ ⎛ σ 2S ⎞ πΔ
⎜ ⎟ ⎜ ⎟∼
⎝σN ⎠ ⎝ σN ⎠ ω
⎛ σ 1S ⎞
⎜ ⎟
⎝ σN ⎠
(ω Δ )
0 1
Fig.XI.1.3 Schematic illustration for the frequency dependence of the complex conductivity σS = σ1S − iσ2S
in a superconductor at T = 0.
In fact, the frequency dependence of σ1S(ω) and σ2S(ω) are related by the Kramers-Kronig relations:
2 ∞ ω ′σ 2 S ( ω ′ )
σ 1S ( ω ) = ∫0 dω ′ + constant , (XI.18)
π ω ′2 − ω 2
2ω ∞ σ 1S ( ω ′ )
σ 2 S (ω ) = − ∫0 dω′ . (XI.19)
π ω ′2 − ω 2
Moreover, the sum rule for the oscillator strength [R. Kubo, J. Phys. Soc. Japan 12, 570 (1957)] asserts that
∞ ∞ π ne 2
∫0 d ω σ 1S ( ω ) = ∫ d ω σ 1 N ( ω ) = , (XI.20)
0 2m
where n is the total number of carriers in the system. Consequently, at T = 0 the “missing area” under σ1S(ω)
in Fig. XI.1.3 must appear as a delta-function at ω = 0 in order to recover the total oscillator strength.
Therefore both σ1S and σ2S diverge in the dc limit, which is consistent with complete disappearance of dc
resistivity in superconductors.
The perfect diamagnetism in superconductors is known as the Meissner effect, which was first
discovered by W. Meissner and R. Ochsenfeld in 1933. Unlike zero resistance, this phenomenon can be
easily proven theoretically by considering the following phenomenological descriptions of F. London and H.
London in Proc. Roy. Soc. (London) A147, 71 (1935).
Suppose the carrier density in the superconducting state of a pure metal is ns so that the supercurrent
density Js(r) induced by a local magnetic field h(r) can be expressed in terms of ns and the drift velocity v(r)
of the carriers:
J s ( r ) = ns e v( r ) . (XI.21)
Fig.XI.1.4 Schematics of the Meissner effect in a superconductor, showing complete expulsion of magnetic
flux in the interior of a superconductor at T < Tc if the applied magnetic field H is less than a
critical field Hc. The magnetic induction in the interior of the superconductor is B = 0 due to the
diamagnetic response of the superconductor that induces a screening current over a surface layer
of thickness λ, where λ denotes the magnetic penetration depth. In contrast, for T > Tc the
superconductor becomes normal and therefore can no longer screen the external magnetic field.
The total free energy of the superconductor of volume Ω is therefore given by:
F = (∫
Ω
)
d 3r FS + Ekinetic + Emag , (XI.22)
where FS denotes condensate energy density at rest, Ekinetic is the kinetic energy associated with the induced
currents, and Emag is the magnetic energy due to the presence of local magnetic fields. Assuming a parabolic
conduction band for the electrons, we find that
⎡1 ⎤
Ekinetic = ∫ d 3r ⎢ nS m v( r ) ⎥ ,
2
(XI.23)
Ω
⎣2 ⎦
h( r )
2
Emag = ∫ d r 3
. (XI.24)
Ω
8π
If we further assume that both Js(r) and h(r) are slowly varying function of r, and recall the Maxwell’s
equation
∇ × h( r ) = 4π J S( r ) ⇒ ∇ × h( r ) = 4π nS e v( r ) , (XI.25)
⎡1 ⎛ 1 ⎞
2
2⎤ 1
∇ × h( r ) ⎥ ≡ d 3r ⎡λL2 ∇ × h( r ) ⎤ ,
2
Ekinetic = ∫ d r ⎢ nS m ⎜
3
⎟ ∫ (XI.26)
Ω
⎢⎣ 2 ⎝ 4π nS e ⎠ ⎥⎦ 8π
Ω ⎣ ⎦
1
d 3r ⎡ h( r ) + λL2 ∇ × h( r ) ⎤ .
2 2
F = ∫ d 3r FS +
8π ∫
(XI.28)
Ω Ω ⎣ ⎦
Minimizing the free energy F relative to the magnetic field h(r), we obtain:
1
d 3r ⎡⎣h( r ) i δ h + λL2 ( ∇ × h( r ) ) i ( ∇ × δ h ) ⎤⎦
4π ∫
δF =
Ω
1
d 3r ⎡⎣h( r ) + λL2∇ × ( ∇ × h( r ) ) ⎤⎦ i δ h = 0
4π ∫
=
Ω
⇒ h( r ) + λL2∇ × ( ∇ × h( r ) ) = 0 . (XI.29)
1
∇ 2 h( r ) = h( r ) , (XI.30)
λL2
whereas the other London equation has been given in EQ. (XI.25).
Using the London equation, we are ready to prove the Meissner effect in superconductors. Consider
a simple example of a semi-infinite superconducting sample occupying the space z > 0. If we turn on a
magnetic field, the penetration of h(r) into the superconductor has two possibilities. In the first case, if the
magnetic field is parallel to the z-axis, we have
∂h
∇ i h( z ) = 0 ⇒ =0 ⇒ h = constant
∂z
⇒ ∇ × h( z ) = 0 ⇒ JS = 0 ⇒ h = 0.
This result implies that a magnetic field normal to the surface of a bulk superconductor cannot penetrate into
the superconductor. However, we note that this finding is only true for type-I superconductors or for type-II
superconductors with applied fields smaller than the lower critical field Hc1. If an external field larger than
Hc1 but smaller than the upper critical field Hc2 is applied normal to the surface of a bulk type-II
superconductor, vortex arrays with quantized magnetic fluxoid can nucleate inside the superconductor,
following the theory of Abrikosov.
dh( z ) d 2 h( z )
∇ × h( z ) = yˆ = 4π J S ⇒ ∇ × ( ∇ × h( z ) ) = − 2
xˆ = −λL−2 h( z ) xˆ
dz dz
⇒ h( z ) = h( z ) xˆ = h( 0 ) exp ⎣⎡ − ( z λL ) ⎦⎤ xˆ . (XI.31)
Therefore, the magnetic field strength decays exponentially into the bulk of the superconductor over a length
comparable to λL, consistent with the observed Meissner effect.
The London penetration depth λL associated with the London theory is entirely determined by the
superconducting carrier density ns and fundamental constants m and e, and is therefore independent of certain
material parameters such as the concentration of impurities or the doping level in alloys. In reality, however,
the empirically determined magnetic penetration depth is found to be a sensitive function of material
properties such as the mean free path l. To improve the theoretical situation, Pippard proposed to generalize
the local theory of London to a non-local theory [A. B. Pippard, Proc. Roy. Soc. (London) A216, 527
(1953)]. Specifically, the London equation in EQ. (XI.29) can be further simplified if we choose the London
gauge for the vector potential
nS e 2 3 ( r − r′ ) [( r − r′ ) i A( r′ )]
JS(r ) = −
− r −r′ ζ
∫ d r′
3
e , (XI.34)
m 4πξ 0 r − r′
4
1 1 1
where ≡ + . (XI.35)
ζ ξ0 l
This non-local expression in EQ. (XI.34) is analogous to the non-local current response of a normal metal to
an electric field first derived by Chambers:
3σ ( r − r′ ) [( r − r′ ) iE( r′ )]
J( r ) =
− r −r ′ l
4π l ∫
d r′ 3
e , (XI.36)
r − r′
4
In addition to the Pippard non-local current response to the vector potential in EQ. (XI.34), we
further note the vector potential consists of both contributions from external magnetic fields and induced
currents, so that we must solve EQ. (XI.34) simultaneously with the Maxwell’s equations. The exact
n e2 3 ( r − r′ )μ ( r − r′ )ν
⎡⎣ J S ( r ) ⎤⎦ μ ≈ − S [ A( r )]ν ∫ d 3r′ e
− r −r ′ ζ
m 4πξ 0 r − r′
4
2
nS e 1 ∞
=− [ A( r )]ν δ μν ∫0 dR e − R ζ
m ξ0
nS e 2 ⎛ l ⎞
=− ⎜ ⎟ [ A( r )]μ (for ζ << λ), (XI.37)
m ⎝ ξ0 + l ⎠
or equivalently,
nS e 2 ⎛l ⎞ 1 ⎛ l ⎞
JS(r ) = − ⎜ ⎟ A( r ) = − ⎜ ⎟ A( r )
m ⎝ ξ0 + l ⎠ 4πλL2 ⎝ ξ 0 + l ⎠
1
≡− A( r ) for ζ << λ. (XI.38)
4πλ 2
Consequently, we obtain the generalized penetration depth in the local limit as:
1/ 2
⎛ξ +l ⎞
λ = λL ⎜ 0 ⎟ for ζ << λ. (XI.39)
⎝ l ⎠
We refer superconductors that satisfy the condition ζ << λ to the London superconductors, and for a pure
London superconductor with ξ0 << l (which is also known as the “clean” limit), we find λ → λL so that the
original London theory is recovered. We also note that ζ is temperature independent while λ increases with
increasing T and diverges at Tc. Therefore, all superconductors become London superconductors near Tc,
regardless of their low-temperature behavior. We also remark that the statement of a diverging penetration
depth with temperature near Tc is based on the two-fluid model: Noting that the superconducting carrier
density nS is related to the London penetration depth λL by nS ( T ) ∝ λL−2( T ) and the temperature dependence
of the superfluid density in the two-fluid model is given by nS ( T ) ∝ [1 − ( T Tc ) ] , we find that
4
λL ( T ) ∝ [1 − ( T Tc ) ]−1/ 2 , which diverges at Tc. We may assume that the temperature dependence of λL ( T )
4
and that of λ ( T ) are comparable, so that the same type of divergence occurs. In either case, we may
consider λ−2(T) as a quantity directly proportional to the effective superconducting carrier density, so that the
vanishing superfluidity at Tc directly translates to a diverging magnetic penetration depth.
On the other hand, if we consider the non-local limit where ζ >> λ and the superconductors are
known as the Pippard superconductors, a general solution becomes difficult to find, and we must specify the
model and the relevant boundary conditions. For a translational invariant system, the general approach is to
Fourier transform EQ. (XI.34), so that the momentum-dependent current density becomes:
If a current is induced in a superconductor and if the superconductor is kept at T << Tc, the current
will persist without decay as long as it is smaller than the critical current. This phenomenon is analogous to
the absence of viscosity in superfluid flow if the flow velocity is below the critical velocity. In the case of
superconductors, the magnitude of critical current density Jc at T << Tc can be estimated by the following
consideration. If a uniform supercurrent of velocity vS is induced in a superconductor, the quasiparticle
energy becomes shifted by p i v S , where p is the momentum of the quasiparticles. If the velocity exceeds a
critical value
Δ
vS* ≡ 0 , (pF: Fermi momentum) (XI.41)
pF
where Δ0 is the magnitude of the superconducting energy gap in the T → 0 limit, then the superconducting
gap vanishes, so that the critical current determined by the critical velocity in EQ. (XI.41) is known as the
depairing current. Below the depairing current, all carriers contribute to the non-dissipative supercurrent so
that J S ∝ vS . Once vS exceeds critical value vS* , quasiparticle excitations occur at zero energy, so that the
supercurrent JS drops precipitously.
The occurrence of persistent currents leads to a phenomenon known as the fluxoid quantization, and
both persistent currents and fluxoid quantization have led to a wide range of applications, including the
fabrication of superconducting magnets and superconductor-based detectors for a variety of high-precision
measurements. Specifically, a fluxoid Φ′ in a superconductor under a magnetic field h is defined as:
Φ′ ≡ ∫ dS i h + 4π ∫ d i λL2 J S ≡ Φ + ∫ d i ( 4πλL J ) .
2
S
(XI.42)
Obviously Φ′ = 0 if we choose a path that encloses neither any hole nor any normal region, because in this
case the magnetic field inside the superconductor is entirely associated with the induced current because the
London equations yield the relation
(
h = −4π ∇ × λL2 J S ) ⇒ (
Φ = ∫ dS i h = − ∫ d i 4πλL2 J S , ) (XI.43)
so that EQ. (XI.42) vanishes identically. On the other hand, if Φ′ is chosen to enclose a hole or a normal
region so that the flux is determined by the vector potential A where h ≡ ∇ × A , the fluxoid becomes
∫ d i ( A + 4πλL J ) = e ∫ d i ( mv
1
Φ′ = 2
S S
+ eA ) , (XI.44)
1 1
d i ( mv + eA ) = ∫ d i ( 2mv + 2eA )
e∫
Φ′ = S S
2e
d i ( m* v )
1 1
=
2e ∫ S
+ e* A =
2e ∫d ip
1
= ( 2π n ) ≡ nΦ 0 , (n: integer) (XI.45)
2e
where we have taken the Cooper pairing of electrons into consideration. Hence, the fluxoid has a unique
constant value for all contours enclosing any given hole in a superconductor, and the quantity Φ0 is defined
as the flux quantum:
2π
Φ0 ≡ = 2.07 × 10−7 Gauss-cm 2 = 2.07 × 10−15 Tesla-m 2 . (XI.46)
2e
Fig.XI.1.5 Manifestation of fluxoid quantization and persistent currents in a thin superconducting ring: An
external magnetic field H is first applied to a thin superconducting ring at T > Tc. The ring is
subsequently cooled to T < Tc so that a current is induced in the ring. The external field H is then
reduced to 0, and the current persists in the ring while enclosing an integer number of flux quanta
nΦ0, where Φ 0 = 2.07 × 10−15 Tesla-m 2 .
Empirically, the fluxoid quantization was first verified by the Little-Parks experiment, which
measured the resistive transition of a thin-wall superconducting cylinder in an axial magnetic field. From the
suppression in the transition temperature due to a small magnetic field, the induced supercurrent velocity can
be determined. Noting that the total flux through the superconducting cylinder was entirely determined by
the geometry of the cylinder and the applied magnetic field, the fluxoid quantization condition in EQ. (XI.45)
would result in periodically varying supercurrent velocity amd therefore periodically varying transition
temperature with magnetic field, which was indeed confirmed by the Little-Parks experiment [W. A. Little
and R. D. Parks, Phys. Rev. Lett. 9, 9 (1962); Phys. Rev. 133, A93 (1964)].
A full theoretical description for the experimental results of Little and Parks requires using the
Landau-Ginzburg theory, which is suitable for modeling properties of a system near phase transitions where
the order parameter is small so can be expanded perturbatively in the free energy expression. Here we briefly
introduce the Landau-Ginzburg theory in the context of deriving the critical current density near Tc, which
differs from the low-temperature behavior discussed earlier.
The Landau-Ginzburg (L-G) theory provides macroscopic description for systems near phase
transitions with the introduction of a complex order parameter ψ ( r ) = ψ ( r ) e ( ) . Here we restrict our
iϕ r
consideration to superconductors. Assuming that the order parameter is relatively small and that both ψ ( r )
and A( r ) vary slowly in space, we express the L-G free energy density of the superconductor (fS) in powers
of ψ ( r ) and ∇ψ ( r ) as follows:
2 2
2 2
2 β
1 ⎛∇ * ⎞ 4 h
f S = f N0 + α ψ + ψ + * ⎜
− e A ⎟ψ + , (XI.47)
2 2m ⎝ i ⎠ 8π
where h(r) is the local magnetic field, and fN0 is the normal-state free energy density of the system in the
absence of magnetic fields. If ψ ( r ) = 0 , we recover the normal-state free energy density fN:
2
h
f N ( T ) = f N0( T ) + , (XI.48)
8π
γe
and f N 0( T ) = f N0( 0 ) − T2 , (XI.49)
2
if we restrict our consideration to the electronic contributions of the system, and γe is the electronic specific
heat coefficient.
In general the coefficients α and β are temperature dependent and can be related to fundamental
physical properties of the system. To appreciate the physical significance of the coefficients, we consider the
simple case for the absence of external fields and gradients. From EQs. (XI.47) and (XI.48) we obtain:
2 β 4
fS − f N = α ψ + ψ . (XI.50)
2
From EQ. (XI.50) we note that the coefficient β must be positive. Otherwise the order parameter could be
arbitrarily large, which would be unphysical. Given β > 0, there are two distinct cases represented by EQ.
(XI.50), depending on the sign of α. In the case of α > 0, ( f S − f N ) is minimized if the order parameter
vanishes. We may associate this case with the normal state at T > Tc. On the other hand, for α < 0, ( f S − f N )
becomes minimized at
ψ = ( α β ) = ⎡⎣ −α ( T < Tc ) β ⎤⎦ ≡ ψ ∞2 ,
2
(XI.51)
α 2( T ) H c2( T )
( fS − fN ) min
=−
2β
≡−
8π
. (XI.52)
where we have defined the thermodynamic critical field Hc(T) in terms of the coefficients α and β, and we
note that Hc(T>Tc) = 0. The condition α < 0 therefore corresponds to the superconducting state at T < Tc
because the free energy of the superconducting state becomes smaller than that of the normal state with a
finite magnitude of the order parameter. Thus, the system gains a condensation energy of the magnitude
H c2 ( 8π ) . The free energy density for the aforementioned two cases is illustrated in Fig. XI.1.6. Hence, we
find that the coefficient α must change sign at Tc. To the lowest order approximation, we may write
⎛ T − Tc ⎞
α (T ) = α 0 ⎜ ⎟ (α0 > 0). (XI.53)
⎝ Tc ⎠
Therefore, from EQs. (XI.51) and (XI.53) we find that the order parameter ψ follows the temperature
dependence ψ ∝ (1 − t ) near Tc, where we have defined the reduced temperature t ≡ ( T Tc ) . Similarly, we
1/ 2
find from EQs. (XI.52) and (XI.53) that the thermodynamic critical field follows the temperature dependence
H c ( T ) ∝ (1 − t ) .
( fS − fN ) ( fS − fN )
α α
−
β β
0 ψ 0 ψ
⎛ α2 ⎞
α >0 (T > Tc ) ⎜ 2β ⎟ α <0 (T < Tc )
⎝ ⎠
Fig.XI.1.6 The Landau-Ginzburg free energy of a homogeneous superconductor in the absence of gradients
and external magnetic fields. The left panel corresponds to the normal state with α > 0 and the
right panel corresponds to the superconducting state with α < 0.
Next, we turn on an external magnetic field and introduce spatial inhomogeneity so that the fourth term in
EQ. (XI.47) becomes
2
1 ⎛∇ * ⎞ 1 ⎡
( ) ψ ⎤,
2
(∇ ψ )
2 2
⎜ − e A ⎟ψ = + ∇ϕ − e * A (XI.54)
2 m* ⎝ i ⎠ 2 m* ⎣ ⎦
where the first term on the right side of EQ. (XI.54) represents the extra energy associated with the gradient
of the amplitude, as in a domain wall, and the second term is the kinetic energy associated with the
( ) ( 2m ) .
2
supercurrent. In the London gauge, ∇ϕ = 0 , so that the kinetic energy density becomes e* A ψ *
In the spirit of London theory, we may relate the kinetic energy associated with the magnetic field-induced
supercurrents to an effective penetration depth λeff via the following expression:
(e A )
*
2
2 1 2 m*
ψ ≡ A ⇒ λeff2 = (XI.55)
8πλeff2 ( )
2
2 m* 4π e* ψ
2
Hence, the effective penetration depth λeff is inversely proportional to the magnitude of the superconducting
order parameter, and its temperature dependence satisfies λeff ∝ (1 − t ) near Tc. Therefore, the coefficients
−1/ 2
α and β in the L-G theory can be expressed in terms of two measurable physical quantities, Hc(T) and λeff(T):
e*2 2e 2
α (T ) = − H c2( T ) λeff2 ( T ) = − H c2 (T ) λeff2 ( T ) , (XI.56)
m* m
4π e*4 16e 4
β (T ) = H c2( T ) λeff4 ( T ) = H c2( T ) λeff4 ( T ) , (XI.57)
m* 2 m2
In addition to the characteristic length λeff, we may define another important characteristic length ξ
from L-G theory. Consider the L-G free energy density in the absence of magnetic field. The corresponding
differential L-G equation is
∇ 2ψ ( 2m ) = αψ + βψ 3 . (XI.58)
where ξ ≡ (1/2m|α|)1/2 is known as the Landau-Ginzburg coherence length, which is characteristic of the
spatial variation of ψ. This point can be understood by taking F(x) = 1 + g(x) so that the differential equation
has a solution g(x) ≈ exp[−21/2 (|x|/ξ)]. This result is consistent with our findings associated with the neutral
superfluid and cold gases in Part IX and Part X.
As a heuristic example of applying the L-G theory to realistic problems, we consider the derivation
of the critical current density for a thin superconducting sample, which may be either a thin ring as depicted
in Fig. XI.1.5 or a thin film. In this case, the amplitude of the order parameter is essentially homogeneous
throughout the entire sample so that ∇ ψ = 0 . The supercurrent density is therefore given by
( 2e )
2
JS =
e
mi
(ψ ∇ψ −ψ ∇ψ ) −
*
* *
m *
ψ A
2
(ψ ( r ) = ψ e ( ) )
iϕ r
2e
( ∇ϕ − 2eA ) = 2e ψ
2 2
= *
ψ vS . (XI.60)
m
The L-G free energy density of the thin sample under a finite magnetic field h is
2
β 1 h
( ∇ϕ − 2eA )
2 4 2 2
f S = f N0 + α ψ + ψ + ψ +
2 2 m* 8π
2
2 β 4 m *
2 h
= f N0 + α ψ + ψ + vS2 ψ + . (XI.61)
2 2 8π
2
Minimizing the free energy in EQ. (XI.61) relative to ψ , we obtain
∂f S 2 m*
2
=α +β ψ + vS2 = 0
∂ψ 2
⎛ −α ⎞ ⎛ m vS ⎞ ⎛ m*vS2 ⎞
* 2
2
⇒ ψ =⎜ ⎟⎜ 1 − ⎟ = ψ 2
∞ ⎜1 − ⎟, (XI.62)
⎝ β ⎠⎝ 2 α ⎠ ⎝ 2α ⎠
which implies that the order parameter becomes suppressed by the increasing supercurrent velocity. From
EQs. (XI.54) and (XI.56), the supercurrent density is given by
2 ⎛ m*vS2 ⎞
J S = 2e ψ v S = 2 eψ ∞2 ⎜ 1 − ⎟ vS , (XI.63)
⎝ 2α ⎠
which has a maximum at ( ∂J S ∂vS ) = 0 . Therefore, the critical current density Jc at vS2 = 2 α ( 3m ) is
*
1/ 2
4 ⎛2α ⎞ H c(T )
J c = eψ ⎜ * ⎟
2
∞ = , (XI.64)
3 ⎝ 3m ⎠ 3 6π λeff ( T )
where we have used EQs. (XI.51), (XI.56) and (XI.57) in EQ. (XI.64).
(a) JS (b) JS
−
T → Tc T Tc −
Jc
Fig.XI.1.7 Schematic dependence of superconducting critical current density on the supercurrent velocity
for (a) T → Tc− and (b) T Tc .
We note that the supercurrent density given in EQ. (XI.63) differs from the low-temperature
depairing current density because at low temperatures the supercurrent velocity does not suppress the
superconducting order parameter until the velocity is very close to the depairing current density, whereas
near Tc the increase in vS suppresses the order parameter, so the overall supercurrent density first increases
with vS for small vS, reaching a peak, and eventually decreases and vanishes when vS becomes too large, as
shown in Fig. XI.1.7.
The different response of type-I and type-II superconductors to magnetic fields may be understood in
terms of the energetic consideration, which we elucidate in the following using an example of the domain
wall energy of a semi-infinite superconductor occupying the space at x > 0. We consider two extreme cases
where the Landau-Ginzburg parameter κ ≡ ( λ ξ ) is either much smaller or much greater than 1. Suppose
that an external magnetic field H = H c yˆ is applied to the surface of the superconductor, the proper boundary
conditions are:
ψ =0 and h ( x ) = H c yˆ for x → −∞ ,
ψ =ψ∞ and h ( x) = 0 for x → +∞ ,
where h(x) denotes the local magnetic field. To calculate the domain-wall energy parameter Γ defined as
∞
Γ ≡ ∫ dx ( g SH − f S 0 ) , (XI.67)
−∞
where gSH is the Gibbs free energy density of the superconductor under a constant applied field H, and fS0 is
the Helmholtz free energy density of the superconductor in zero field. Noting that the Gibbs free energy
density and the Helmholtz free energy density satisfy the following relations:
hH H c2
g= f − , f N 0 − f S0 = , (XI.68)
4π 8π
and for a constant applied field H, the normal and superconducting state Gibbs free energies GN and GS are
H2
GN = Ω S f N0 − ( Ω S + Ω ext ) , (ΩS: sample volume) (XI.69)
8π
H2
GS = Ω S f S0 − Ω ext , (Ωext: external volume) (XI.70)
8π
so that
H2 ⎛ H c2 − H 2 ⎞
( )
GN − GS = Ω S f N0 − f S0 − Ω S
8π
= ΩS ⎜ ⎟, (XI.71)
⎝ 8π ⎠
and GN = GS at H = Hc. Thus, EQ. (XI.67) becomes
∞ ∞ ⎛ hH H c2 ⎞
Γ ≡ ∫ dx ( g SH − f S 0 ) = ∫ dx ⎜ f SH − − fN 0 + ⎟
−∞ −∞
⎝ 4π 8π ⎠
∞ ⎛ β 4 1 ⎛∇ * ⎞
2
h
2
h Hc H c2 ⎞
= ∫ dx ⎜ f N0 + α ψ + ψ +
2
⎜ ⎜ − e A ⎟ψ + − − fN 0 + ⎟
−∞
⎝ 2 2 m* ⎝ i ⎠ 8π 4π 8π ⎟⎠
⎛ ( h − Hc ) ⎞
2 2
∞ β 4 1 ⎛∇ * ⎞
= ∫ dx ⎜ α ψ + ψ + ⎟.
2
⎜ ⎜ − e A ⎟ψ + ⎟
(XI.72)
−∞
2 2 m* ⎝ i ⎠ 8π
⎝ ⎠
(a) (b)
(c)
Fig.XI.1.8 Magnetic field (H) versus temperature (T) phase diagrams for (a) type-I superconductors and (b)
type-II superconductors; (c) schematics of the vortex lattice pattern in type-II superconductors.
If we minimize the L-G free energy density in EQ. (XI.47) relative to the order parameter, we obtain
the L-G differential equation:
2
1 ⎛∇
2 ⎞
αψ + β ψ ψ + * ⎜ − e* A ⎟ ψ = 0 , (XI.73)
2m ⎝ i ⎠
∞ ⎧⎪ 1 ⎛∇ * ⎞
2
⎫⎪
∫−∞ ⎨⎪α ψ + β ψ + 2m* ⎜⎝ i − e A ⎟⎠ψ
2 4
dx ⎬ = 0. (XI.74)
⎩ ⎭⎪
∞ ⎛ β 4 ( h − H c )2 ⎞ ∞ ⎛ H 2 ψ 4 ( h − H c )2 ⎞ H 2
Γ = ∫ dx ⎜ − ψ + ⎟ = ∫ dx ⎜ − c 4 + ⎟≡ c δ , (XI.75)
−∞ ⎜ 2 8π ⎟ −∞ ⎜ 8π ψ ∞ 8π ⎟ 8π
⎝ ⎠ ⎝ ⎠
so that
∞ ⎛ ψ4 ⎛ h ⎞ ⎞
2
δ ≡ ∫ dx ⎜ − + 1− ⎟. (XI.76)
−∞ ⎜ ψ ∞ 4 ⎜⎝ H c ⎟⎠ ⎟
⎝ ⎠
The first term in EQ. (XI.76) represents the negative condensation energy and the second term is
associated with the positive diamagnetic energy. To see whether magnetic field penetration is energetically
favorable, we want to determine the sign and magnitude of the domain-wall energy parameter δ by solving
the spatial dependence of ψ(x) and h(x) self-consistently. For δ > 0, the superconductor prefers complete
expulsion of magnetic flux. In contrast, for δ < 0 magnetic flux penetration into the superconductor helps
lower the domain wall energy. Numerically, it can be shown that
4 2
δ= ξ ≈ 1.89ξ if κ 1, (XI.77)
3
8 ( 2 − 1)
δ =− λ ≈ −1.104λ if κ 1, (XI.78)
3
The results in EQs. (XI.77) and (XI.78) can be approximated by δ ∼ ( ξ − λ ) . More precisely, it can be
shown that the crossover of δ from positive to negative occurs at κ = 1 2 . Therefore, the superconductor
is type-I if κ < 1 2 and becomes type-II if κ > 1 2 . The order parameter and local field profiles near
the superconductor/vacuum domain wall for type-I and type-II superconductors are depicted in Fig. XI.1.9.
To solve for the magnetic critical fields in superconductors using the L-G theory, the standard
procedure is as follows. The spatial distribution of h(r) is first determined using the London equation and
proper boundary conditions. This local field is then inserted into the L-G free energy, and the Gibbs free
energy is minimized to yield an expression for |ψ |2 as a function of h(r). Finally, the critical field is
determined by finding the magnetic field that gives |ψ | = 0. However, the solutions for general geometries
and material parameters must be found numerically. In Problem Set 11 you will be asked to employ the
aforementioned prescription to find the parallel critical field of a type-I superconducting slab as a function of
the slab thickness.
(a) (b)
Hc λ Hc ξ
ψ∞ ψ∞
h(x) ψ ( x)
ψ ( x)
h(x)
ξ λ
κ 1 (Type-I) κ 1 (Type-II)
Fig.XI.1.9 The spatial distribution of the order parameter |ψ| and local field h(r) around the domain wall of
a superconductor/vacuum interface with an external field applied parallel to the sample surface:
(a) for a type-I superconductor, and (b) for a type-II superconductor.
As mentioned earlier, in the Abrikosov vortex lattice there is repulsive interaction among vortices,
leading to finite bulk, shear and tilt elastic moduli in the vortex lattice as in real crystalline structures. To
understand how the repulsive interaction arises, we recall the solution to the order parameter of a superfluid
in the presence of a vortex and we express the order parameter in the L-G equation by the following:
ψ ( r ) = ψ ∞ F ( r ) e iθ . (XI.79)
This phase choice for the order parameter fixes the gauge choice for the vector potential A so that
⎛1⎞ r
A( r ) = A( r ) θˆ = ⎜ ⎟ ∫ r ′dr ′ h( r ′ ) θˆ . (XI.80)
⎝r⎠ 0
Near the center of the vortex, the vector potential in EQ. (XI.80) becomes
⎛1⎞ r h( 0 ) ˆ
A( r ) = lim ⎜ ⎟ ∫ r ′dr ′ h( r ′ ) θˆ ≈ rθ, (XI.81)
r →0+ ⎝ r ⎠ 0 2
Φ0
∫ r →∞ d l iA( r ) = Φ 0 ≡ 2π rA∞ ⇒ A ( r → ∞ ) ≡ A∞ =
2π r
. (XI.82)
Inserting EQ. (XI.79) into the L-G differential equation in EQ. (XI.73), we find that the function F(r)
satisfies the following relation:
⎡⎛ 1 2π A ⎞2 1 d ⎛ df ⎞ ⎤
F − F − ξ ⎢⎜ −
3 2
⎟ F− ⎜ r ⎟⎥ = 0 , (XI.83)
⎣⎢⎝ r Φ 0 ⎠ 2 dr ⎝ dr ⎠ ⎦⎥
⎛ν r ⎞
F ( r ) ≈ tanh ⎜ ⎟, (ν: a constant ~ 1) (XI.84)
⎝ξ ⎠
Therefore, for extreme type-II superconductors, we may assume that except a core region of radius ~ ξ, F(r)
can be treated like a constant ~ 1, and the London equations govern the fields and currents so that we have
the following relation between h(r) and JS(r):
4πλ 2∇ × J S + h = Φ 0 δ 2 D ( r ) zˆ . (XI.85)
λ 2∇ × ( ∇ × h ) + h = Φ 0 δ 2 D ( r ) zˆ . (XI.86)
1 Φ0
∇ 2h − h=− δ 2 D ( r ) zˆ , (XI.87)
λ 2
λ2
Φ0 ⎛r⎞
h( r ) = K 0⎜ ⎟ zˆ , (XI.88)
2πλ 2
⎝ξ ⎠
where K0 is the zeroth-order Hankel function of imaginary arguments, and the asymptotic expressions are
given by
1/ 2
Φ0 ⎛ π λ ⎞ ⎛ r⎞
h( r ) → ⎜ ⎟ exp⎜ − ⎟ zˆ (r → ∞) , (XI.89)
2πλ 2 ⎝ 2 r ⎠ ⎝ ξ⎠
Φ0 ⎛ λ ⎞
h( r ) ≈ 2 ⎜
ln + 0.12 ⎟ zˆ (ξ r λ) . (XI.90)
2πλ ⎝ r ⎠
The spatial dependence of h(r) and F(r) around a vortex in an extreme type-II superconductor is
schematically illustrated in Fig. XI.1.10. We further note that it costs finite energy to create a vortex, and the
energy required per unit length, known as the vortex line tension ε1, is given by:
dS ( h ),
1
8π ∫
2
ε1 = 2
+ λ2 ∇ ×h (XI.91)
1 λ2
=
8π
(
∫ dS h + λ ∇ × ∇ × h i h +
2
) 8π ∫ d l i( h × ∇ × h )
1 λ2
=
8π ∫ dS h Φ 0 δ 2 D( r ) + 8π ∫ d li( h × ∇ × h ) (XI.92)
2
Φ0 ⎛λ⎞ ⎛ Φ ⎞
Φ0 Φ0
≈ h( ξ ) ≈ ln ⎜ ⎟ = ⎜ 0 ⎟ ln κ . (XI.93)
8π 8π 2πλ 2
⎝ ξ ⎠ ⎝ 4πλ ⎠
ξ
h(r)
|ψ |
r
λ
Fig.XI.1.10 The spatial variation of the superconducting order parameter |ψ (r)| and the local magnetic
field h(r) near an isolated vortex in an extreme type-II superconductor with λ >> ξ.
Now suppose we create two vortices and then investigate how they interact with each other as we
bring them to position r1 and r2. The total field h(r) is given by
h( r ) = h1 ( r ) + h 2 ( r ) = ⎡⎣ h ( r − r1 ) + h ( r − r2 ) ⎤⎦ zˆ . (XI.94)
Using EQs. (XI.93) and (XI.94), the total free energy increase per unit length ΔF due to the creation of two
vortices at r1 and r2 becomes:
Φ0
ΔF =
8π
⎣⎡ h1 ( r1 ) + h1 ( r2 ) + h2 ( r1 ) + h2 ( r2 ) ⎦⎤
⎡ Φ0 ⎤ ⎡Φ ⎤
= 2⎢ h1 ( r1 ) ⎥ + 2 ⎢ 0 h1 ( r2 ) ⎥ ≈ 2ε1 + F12 , (XI.95)
⎣ 8π ⎦ ⎣ 8π ⎦
The first term in EQ. (XI.95) is simply the sum of the two individual line energies of the vortices, whereas
the second term represents the interaction energy:
Φ0 Φ 02 ⎛ r12 ⎞
F12 = h1 ( r2 ) = K 0⎜ ⎟, r12 ≡ r1 − r2 . (XI.96)
4π 8π λ
2 2
⎝λ ⎠
Clearly this vortex-vortex interaction given in EQ. (XI.96) is repulsive. Hence, we expect elastic interactions
due to the mutually repulsive forces among periodic structures of vortices, which resemble the situation in a
real crystal.
The occurrence of periodic vortex structures in type-II superconductors, known as the Abrikosov
vortex lattice, was first derived rigorously by A. A. Abrikosov in Soviet Phys. JETP 5, 1174 (1957). The
underlying theoretical approach begins with solving the Landau-Ginzburg differential equations at the upper
critical field Hc2 so that the vector potential A is homogeneous throughout the bulk superconductor. For an
applied field along the z-axis, we may use the gauge A = H 0 x yˆ ≈ H c 2 x yˆ . By neglecting the non-linear term
2
β ψ ψ in the Landau-Ginzburg differential equations initially, the superconducting order parameter takes
the following form:
ψ ( x, y , z ) = e
i k y y − ( x − x0 )
e
2
( 2ξ ) .
2
(XI.97)
∞ ⎡ 1 ⎛ nΦ 0 ⎞ ⎤
2
ψ ( x, y , z ) = ∑ cn e i nk y
exp ⎢ − 2 ⎜ x − k⎟ ⎥. (XI.98)
n =−∞
⎣⎢ 2ξ ⎝ 2π H 0 ⎠ ⎦⎥
The above generalized solution can be inserted into the Landau-Ginzburg differential equations with the non-
2
linear term β ψ ψ included, so that solutions for H 0 < H c 2 can be found perturbatively. Further details of
the derivation may be found in Abrikosov’s original paper.
Finally, we remark that the vortex phase diagrams depicted in Fig. XI.1.8 are in fact only applicable
to conventional superconductors. In the case of extreme type-II high-temperature superconductors where λ
>> ξ, the combined effects of strong quantum, thermal and disorder fluctuations together with the large
electronic anisotropy give rise to complicated vortex phase diagrams that differ from those of typical type-II
superconductors. We shall briefly touch upon the issues of unconventional vortex phase diagrams in Part IXI.
For N particles in the superconducting ground state (or equivalently, N* = N/2 pairs in the ground
state) characterized by an order parameter ψ = ψ e iϕ , the superconducting phase ϕ and the total number of
condensate particles N may be considered as the semi-classical canonically conjugate variables of the
Hamiltonian H, so that the equations of motion are given by:
∂ϕ ∂H ∂F
=− = −2 = −2 μ , (XI.99)
∂t ∂N ∂N *
∂N * ∂H ∂F
=− = , (XI.100)
∂t ∂ϕ ∂ϕ
Suppose that we bring a second superconductor close to the first superconductor considered above to
form a superconductor-insulator-superconductor (S-I-S) junction where the insulating barrier is very thin, we
may define the phase coherence between the two superconductors as Δϕ ≡ (ϕ 2 − ϕ1 ) so that
∂ ( Δϕ )
= 2 ( μ1 − μ 2 ) . ( Δϕ : modulo 2π ) (XI.101)
∂t
∂ ( Δϕ )
= ω ( t ) = 2eV , (XI.102)
∂t
Fig.XI.1.11 Schematics of the superconducting quantum interference phenomenon involving two Josephson
junctions: (a) the basic structure of a SQUID consists of two Josephson junctions connected to a
magnetic flux pickup loop of an area A. An input current I splits into two paths, with I1 passing
through one Josephson junction and I2 through the other. The two currents are recombined after
going around the pickup loop. The combined current is sent through a resistor for current-voltage (I-
V) measurements; (b) measurements of the I-V characteristics reveal periodic dependence on the
total magnetic flux Φ = HA enclosed by the pickup loop, where H is the applied magnetic field, and
the periodicity is in units of the flux quantum Φ0. The finite current at zero voltage corresponds to
the Josephson pair tunneling current; (c) as the SQUID is biased by a constant current I′ to operate in
a resistive mode, the resulting voltage exhibits periodic oscillations with the applied magnetic field,
and the period in field is determined by (Φ0/A).
Next, we consider the pair current tunneling across the S-I-S junction. If the number of condensate
pairs in the left superconductor is N1* and that in the right superconductor is N 2* , we find that the pair current
I12 from left to right satisfies the following relation:
Consequently, the pair current I12, also known as the Josephson current, is related to a phase dependent
energy F12( Δϕ ) . Noting that F12( Δϕ ) is a periodic even function of Δϕ because ψ and ψ* have the same
eigen-energy, we may express F12( Δϕ ) in a cosine Fourier series in Δϕ:
∑ν cν [cos ( Δϕ )]
ν
F12( Δϕ ) = . (XI.104)
Taking the leading term in F12( Δϕ ) , we have ⎡⎣ ∂F12 ∂ ( Δϕ ) ⎤⎦ ∼ sin ( Δϕ ) , so that according to EQ. (XI.103)
we obtain the Josephson current relation:
I S = I c sin ( Δϕ ) , (XI.105)
where Ic is a material and temperature-dependent critical current of the junction, and we have replaced I12 by
a more general notation IS. The Josephson effect and the macroscopic quantum interference phenomena in
superconducting junctions are the foundation for applications of many superconducting devices. Examples of
some of the superconducting devices will be considered in Part XII.
(
Ekk ′ = k − k ′
2 2
) ( 2m ) . (k: electron momentum; k′: hole momentum)
The energy Ekk′ can be made arbitrarily small, so that there are numerous low-energy excitations, which
gives rise to relatively large electronic specific heat
so that there is substantially large dissipation associated with the large degree of low-energy excitations if a
normal metal is subject to electromagnetic waves, ultrasonic waves, nuclear spin resonance, etc. In contrast,
for an s-wave superconductor (i.e. a superconductor with an energy gap isotropic in momentum space) the
creation of electron-hole pairs can only take place if
Consequently, the electronic specific heat at low temperatures becomes much suppressed:
Moreover, absorption of electromagnetic energy is possible only if the photon energy exceeds 2Δ. Similarly,
the ultrasonic attenuation that involves absorption of phonons via collision with pre-existing excitations is
However, we note that the existence of a finite energy gap is not a sufficient condition for
superconductivity, because superconductors may become gapless either for specific momenta in systems
with unconventional pairing symmetry, such as in certain families of the heavy-Fermion and high-
temperature superconductors or in dirty superconductors with the superconducting energy gap suppressed by
disorder or magnetic fields. The essential aspects of superconductivity are the occurrence of Cooper pairing
as well as the existence of macroscopic quantum coherence. The manifestation of quantum coherence in
superconductors clearly distinguishes them from other gapped systems such as semiconductors.
The occurrence of superconductivity relies on the formation of Cooper pairs so that the ground state of
an otherwise fermion system becomes a bosonic condensate. In this section we first demonstrate that any
infinitesimal attractive interaction between two electrons in the presence of a Fermi sea leads to a stable
bound pair known as the Cooper pair [L. N. Cooper, Phys. Rev. 104, 1189 (1956)]. The original non-
interacting ground state (i.e. the filled Fermi sea) becomes unstable against the pair formation, and the finite
binding energy of the Cooper pair provides a qualitative explanation for the presence of an energy gap in the
excitation spectrum of superconductors. One possible origin for the attractive interaction between a pair of
electrons is the electron-phonon interaction. This microscopic pairing mechanism can successfully account
for the isotope effect and various thermodynamic and electrodynamic properties of simple superconductors
[J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 108, 1175 (1957)]. We shall discuss the BCS
theory in depth using the thermal Green function approach by Gorkov [L. P. Gorkov, Sov. Phys. – JETP 7,
505 (1958)]. It should be noted, however, that in a variety of superconductors, including the heavy-fermion
superconductors and the high-temperature superconducting cuprates, the pairing mechanism need not be
mediated by phonons. In particular, the microscopic theory of BCS cannot provide a consistent account for
many complex and seemingly conflicting phenomena among different families of cuprate superconductors.
The quest for the underlying pairing mechanism for high-temperature superconducting cuprates remains one
of the greatest challenges in modern condensed matter physics.
[Cooper pairing]
Here we focus our discussion of the formation of Cooper pairing on a simpler situation of adding two
electrons to a filled Fermi sea and investigate how to obtain the pair binding energy when an attractive
interaction is turned on between the two electrons outside the Fermi sea. We may also consider an alternative
situation where an attractive interaction is introduced between two electrons inside the Fermi Sea. The latter
consideration must involve virtual interactions with states outside of the Fermi sea because all states within
are already filled, and such consideration is more tedious in the formalism. In the interest of time we shall
only outline the concepts of the latter case and refer you to references for more details.
Let us assume that a bound pair of electrons is added to the Fermi sea at T = 0. If the added bound
pair does not interact with the electrons in the Fermi Sea except obeying the Pauli exclusion principle, we
may construct a two-electron wavefunction for the bound pair, and the lowest energy state of the
wavefunction has zero total momentum relative to the Fermi sea. Anticipating an attractive potential, the
(α α ) ,
1 2
(β β ).
1 2
where αi and βi refer to the spin-up and spin-down states of the i-th electron.
For simplicity, we only consider the singlet situation in the following discussion, which in fact
represents the majority of superconductors. The Schrödinger’s equation for the two-electron system with an
electron-phonon interaction potential V is given by
(H 0
+ V )ψ 0 = Eψ 0 , (XI.110)
where
H 0ψ 0 = 2∑ k > k gk ε k cos [k i ( r1 − r2 )] (α1 β 2 − α 2 β1 ) , (XI.111)
F
(H 0
+ V )ψ 0 = ∑ k > k g k
F
{( 2ε k + V ) cos [k i ( r1 − r2 )] (α1 β 2 − α 2 β1 ) , } (XI.112)
⎧ 1 ⎫
Egk = ∑ k ′ > k gk ′ ⎨ 2ε k ′δ kk ′ + V ( r )⎬
i ( k ′−k ) i r
⇒ ∫d re
3
F
⎩ Ω ⎭
≡ ∑ k ′ > k gk ′ {2ε k ′δ kk ′ + Vkk ′ } , (XI.113)
F
⇒ ( E − 2ε k ) gk = ∑ k′ >k F
g k ′Vkk ′ . (XI.114)
The matrix elements Vkk ′ defined in EQ. (XI.113) characterize the strength of the potential for
scattering a pair of electrons from ( k ′, −k ′ ) to ( k , −k ) . Thus, the above derivation implies that a bound pair
exists if we can find a set of gk that satisfies EQ. (XI.114), or equivalently,
1
( E − 2ε k ) ∑ k′ >k
gk = gk ′Vkk ′ . (XI.115)
F
where V > 0 and ωD is the cutoff energy associated with the Debye frequency, the upper cutoff frequency for
phonons. Inserting EQ. (XI.116) into EQ. (XI.115), we find
so that
1 1 N (ε )
= ∑ k >k = ∫ dε , (XI.118)
V F
( 2ε k − E ) ( 2ε − E )
where N(ε) is the density of states. If ωD << εF, we may further simplify EQ. (XI.118) by making the
assumption that N(ε) ≈ N(εF). Hence, we find the relation
1 ε F +ω D N (ε ) 1 ⎡ 2ε F − E + 2ω D ⎤
=∫ dε = N ( ε F ) ln ⎢ ⎥. (XI.119)
V εF ( 2ε − E ) 2 ⎣ 2ε F − E ⎦
We note that EQ. (XI.119) always holds for any small values of V as long as the condition ( 2ε F − E ) > 0 is
satisfied, which implies that the eigen-energy of the pair is smaller than the sum of the two free particle
energies and therefore a bound state is formed. In other words, the Fermi sea becomes unstable against the
formation of a Cooper pair because solutions to gk in EQ. (XI.115) can always be found.
In the weak coupling limit, we have the condition N ( ε ) V 1 , so that from EQ. (XI.119) we
obtain:
⎤ ⎡ 2
2ε F − E + 2ω D = ( 2ε F − E ) exp ⎢ ⎥
⎣ N (ε F )V ⎦
⎡ ⎛ 2 ⎞ ⎤ ⎛ 2 ⎞
⇒ 2ωD = ( 2ε F − E ) ⎢ exp⎜ ⎟ − 1⎥ ≈ ( 2ε F − E ) exp⎜ ⎟,
⎣ ⎝ N (ε F )V ⎠ ⎦ ⎝ N (ε F )V ⎠
⎛ −2 ⎞
⇒ E ≈ 2ε F − 2ωD exp⎜ ⎟. (XI.120)
⎝ N (ε F )V ⎠
Equation (XI.120) implies that there is a bound state with a negative energy with respect to the Fermi
surface. If we rewrite EQ. (XI.120) into E ≡ 2ε F − Δ 0 , we obtain an energy gap:
⎛ ⎞
−2
Δ 0 ≈ 2ω D exp⎜ ⎟. (XI.121)
⎝ N (ε F )V ⎠
For convenience, we may measure all energy scales relative to the Fermi level, so that we define
ξk = ε k − ε F , E ′ = 2ε F − E , (XI.122)
which yields
gk ′
gk = V ∑ k ′ >k . (XI.123)
F
( 2ξk + E ′ )
Clearly the weighting factor ( 2ξk + E ′ )
−1
in EQ. (XI.123) has its maximum at ξk = 0 because both E ′ ≥ 0
and ξ k ≥ 0 . Thus, the electron states with a range of energy above εF are strongly involved in forming the
bound state.
1
ψ ( r1 , r2 ) = ϕ0( r1 , r2 ) + ∑ λ ≠ 0 ϕλ ( r1 , r2 ) ϕλ V ψ ( r1 , r2 ) , (XI.124)
E − Eλ
where ψ satisfies the Schrödinger’s equation for the interacting two-electron system ( H 0 + V )ψ = Eψ , ϕ λ
are the eigenfunctions of the non-perturbed Hamiltonian H0 so that H 0 ϕλ = Eλϕλ , and the states λ ≠ 0 refer
to the excited states outside of the filled Fermi sea. In a homogeneous system, the total momentum of the
two-electron system is conserved, and therefore the two-electron wavefunction can be expressed in terms of
a product of the center-of-mass wavefunction and the relative wavefunction
where we have introduced the following definitions in EQ. (XI.125) and assumed that the momenta of the
two electrons are k1 and k2:
1
P = k1 + k 2 k≡
2
(k 1
− k2 ) (k 1
> kF , k 2 > kF )
R≡
1
2
(r + r )
1 2 (
r ≡ r1 − r2 . )
If we further define the eigen-energy of the Schrödinger’s equation E as the sum of the kinetic energy of the
center-of-mass and an effective eigen-value κ2 so that
2
κ2 P
E= + , (XI.126)
m 4m
d 3q m
ψ P ,k ( r ) = e i k i r + ∫ ei q i r q V ψ P ,k , (XI.127)
( 2π )
3
Γ
κ − q2
2
1 1
where P ± k < kF Γ≡ P ± q > kF
2 2
κ 2 − q2
and = Ω −1 k V ψ P ,k . (XI.128)
m
It can be shown that the solution to the Bethe-Goldstone equation (see, for example, Section 36 of Fetter &
Walecka) also yields an eigen-energy lower than the non-interacting two-particle energy, implying a bound
pair, and the resulting energy gap for P = 0 is approximately
k F2 ⎡ 4π 2 ⎤ ⎡ 2 ⎤
Δ 0 ≡ ( 2ε F − E ) ≈ exp ⎢ − ⎥ = 2ε F exp ⎢ − ⎥, (XI.129)
m ⎣ mk FV ⎦ ⎣ N (ε F )V ⎦
where we have used the relation N ( ε F ) = mk F (2m) for the free electron gas.
1) The bound energy of the Cooper pair given in either EQ. (XI.121) or EQ. (XI.129) has an essential
singularity in the attractive potential V, and therefore cannot be obtained with perturbation theory.
2) The largest bounding energy (or the maximum superconducting gap) is obtained if the total momentum P
of the Cooper pair is zero. In principle, Cooper pairing can still be retained up to a finite momentum.
3) The occurrence of a bound pair for an arbitrarily small attractive interaction potential V is crucially
dependent on the presence of a Fermi sea, because in the absence of the medium we have kF → 0, or
equivalently N (εF) → 0, and therefore Δ0 → 0 according to EQs. (XI.121) and (XI.129). In other words,
a stable bound pair cannot exist in vacuum under an arbitrarily small attractive potential.
4) The condition for forming a bound pair as given in EQ. (XI.118) can be rewritten as I ( E ′ ) = 1 where
N (ξ )
I ( E ′ ) ≡ V ∫ dξ ( ξ ≡ ε − ε F , E ′ ≡ 2ε F − E ). (XI.130)
( 2ξ + E ′ )
It is clear from EQ. (XI.130) that for all E ′ > 0 we have I ( E ′ ) < I ( 0 ) . Given that I ( 0 ) is a large
number if V is finite, we can always find a value of E ′ > 0 associated with the condition I ( E ′ ) = 1 . In
other words, a bound-pair state can always be found, which implies instability of the Fermi sea against
the formation of a bound pair.
To derive the microscopic theory for superconductivity, we begin with the consideration of a model
Hamiltonian that includes an on-site attractive electron-phonon interaction:
⎧ 1 ⎡∇ ⎤
2
⎫ 1
Hˆ = Hˆ 0 + Vˆ = ∫ d 3 x ψ α† ( x ) ⎨ ⎢ − eA ( x ) ⎥ − μ ⎬ψ α ( x ) − γ ∫ d 3 x ψ α† ( x )ψ β† ( x )ψ β ( x )ψ α ( x ) , (XI.131)
⎩ 2m ⎣ i ⎦ ⎭ 2
where γ > 0. To solve the model Hamiltonian, we generalize the Hartree-Fock approximation as follows:
Vˆ ≈ −γ ∫ d 3 x ⎡ ψ α† ( x )ψ α ( x ) ψ β† ( x )ψ β ( x ) − ψ α† ( x )ψ β ( x ) ψ β† ( x )ψ α ( x ) ⎤
⎣ ⎦
1
− γ ∫ d 3 x ⎡ ψ α† ( x )ψ β† ( x ) ψ β ( x )ψ α ( x ) + ψ α† ( x )ψ β† ( x ) ψ β ( x )ψ α ( x ) ⎤
2 ⎣ ⎦
1
≡ VˆHF − γ ∫ d 3 x ⎡ ψ α† ( x )ψ β† ( x ) ψ β ( x )ψ α ( x ) + ψ α† ( x )ψ β† ( x ) ψ β ( x )ψ α ( x ) ⎤ . (XI.132)
2 ⎣ ⎦
The approximation given in EQ. (XI.132) no longer conserves the number of particles, which may be
justified by considering the condensate of Cooper pairs as a reservoir of bosons. We further note that the
which is the basis for the BCS theory. This theory is self-consistent because the ensemble average given in
the angular brackets are evaluated with Ĥ eff so that
∂ψ H ↑ ⎡ 1 ⎛∇ ⎞
2
⎤
⎟ − μ ⎥ψ H ↑ − γ ψ ↑ψ ↓ ψ H ↓ ,
†
= −⎢ ⎜ − eA (XI.136)
∂τ ⎣ 2m ⎝ i ⎠ ⎦
∂ψ H† ↓ ⎡ 1 ⎛ ∇ ⎤ †
2
⎞
⎜ − − eA ⎟ − μ ⎥ψ H ↓ − γ ψ ↓ψ ↑ ψ H ↑ .
† †
=⎢ (XI.137)
∂τ ⎣ 2 m ⎝ i ⎠ ⎦
( )
Δ ( x ) ≡ γ F x, τ + ; x,τ = −γ ψ ↑( x )ψ ↓( x ) = γ ψ ↓( x )ψ ↑( x ) , (XI.141)
∂ ⎡∂ ⎤
∂τ
{
G ( x, τ ; x′,τ ′ ) = −δ (τ − τ ′ ) ψ H ↑( x,τ ) ,ψ H† ↑( x′,τ ′ ) } − Tτ ⎢ ψ H ↑( x,τ )ψ H† ↑( x′, τ ′ ) ⎥
⎣ ∂τ ⎦
⎡ 1 ⎛∇ ⎞
2
⎤
= −δ (τ − τ ′ ) δ ( x − x′ ) − ⎢ ⎜ − eA ⎟ − μ ⎥ G ( x, τ ; x′,τ ′ ) + γ ψ ↑ψ ↓ F ( x,τ ; x′, τ ′ ) ,
†
⎣ 2m ⎝ i ⎠ ⎦
⎡ ∂ 1 ⎛∇ ⎞
2
⎤
⎜ − eA( x ) ⎟ + μ ⎥ G ( x, τ ; x′,τ ′ ) + Δ ( x ) F ( x,τ ; x′,τ ′ ) = δ (τ − τ ′ ) δ ( x − x′ ) .
†
⇒ ⎢− − (XI.142)
⎣ ∂τ 2m ⎝ i ⎠ ⎦
⎡ ∂ 1 ⎛∇ ⎞
2
⎤
⎢− − ⎜ − eA( x ) ⎟ + μ ⎥ F ( x,τ ; x′, τ ′ ) = Δ ( x ) G ( x, τ ; x′,τ ′ ) , (XI.143)
⎣ ∂τ 2m ⎝ i ⎠ ⎦
⎡∂ 1 ⎤ †
⎢⎣ ∂τ − 2m ( i ∇ − eA( x ) ) + μ ⎥⎦ F ( x, τ ; x ,τ ) = Δ ( x ) G ( x, τ ; x ,τ ) .
2
′ ′ *
′ ′ (XI.144)
We may consider EQs. (XI.142) and (XI.144) as a pair of coupled equations for G and F † , which are known
as the Gorkov equations. Alternatively, we may simplify the expressions in EQs. (XI.142) – (XI.144) in a
matrix form. Specifically, we introduce a two-component field operator:
⎛ψ Hˆ ( x ) ⎞
Ψ Hˆ ( x ) = ⎜ ⎟, x ≡ ( x, τ ) . (XI.145)
⎝ψ Hˆ ( x ) ⎠
†
Using EQs. (XI.138) – (XI.140) and (XI.145), we may define a (2×2) matrix Green function G ( x, x′ ) :
G ( x, x′ ) ≡ − Tτ ⎡⎣ Ψ Hˆ ( x ) Ψ †Hˆ ( x′ ) ⎤⎦
⎡ G ( x, x ′ ) F ( x, x ′ ) ⎤
= ⎢ † ⎥. (XI.146)
⎣ F ( x, x′ ) −G ( x′, x ) ⎦
Therefore, the corresponding equations of motion become
Dx G ( x, x′ ) = 1 δ ( x − x′ ) , (XI.147)
⎛ ∂ 1 ⎛∇ ⎞
2
⎞
⎜− − ⎜ − eA( x ) ⎟ + μ Δ( x ) ⎟
⎜ ∂τ 2m ⎝ i ⎠ ⎟.
where Dx ≡ (XI.148)
⎜ ∂ 1 ⎟
Δ*( x ) ( i∇ − eA( x ) ) − μ ⎟
2
⎜ − +
⎝ ∂τ 2m ⎠
Having derived the equations of motion for the single-particle and anomalous Green functions, we
can find the solutions to these Green functions and discuss the corresponding physical properties. We first
consider a simple situation of a time-independent Hamiltonian, so that the Green functions only depend on
τ − τ ′. We may Fourier transform the Green functions
G ( x, τ ; x′, τ ′ ) = β −1 ∑ e ν
− iω (τ −τ ′)
G ( x, x′; ων ) , (XI.149)
ν
F †( x, τ ; x′,τ ′ ) = β −1 ∑ e ν
− iω (τ −τ ′)
F †( x, x′; ων ) , (XI.150)
ν
where the Matsubara frequency for fermions is given by ων = ( 2ν + 1) πβ −1 . Therefore, the Gorkov equations
EQs. (XI.142) and (XI.144) become
⎡ 1 ⎛∇ ⎞
2
⎤
⎢iων − ⎜ − eA( x ) ⎟ + μ ⎥ G ( x, x′; ων ) + Δ ( x ) F ( x, x′; ων ) = δ ( x − x′ ) ,
†
(XI.151)
⎣ 2 m ⎝ i ⎠ ⎦
⎡ 1 ⎤ †
⎢⎣ −iων − 2m ( i∇ − eA( x ) ) + μ ⎥⎦ F ( x, x ; ων ) − Δ ( x ) G ( x, x ; ων ) = 0 ,
2
′ *
′ (XI.152)
which must be solved together with the self-consistent condition in EQ. (XI.141):
γ
( )
Δ*( x ) ≡ γ F † x, τ + ; x,τ = −γ ψ ↓†( x )ψ ↑†( x ) = lim ∑ e − iων η F †( x, x; ων ) ,
β η →0 + ν
(XI.153)
In general, the coupled equations in EQs. (XI.151) – (XI.153) are augmented by Maxwell’s equation that
relates the local field h = ∇ × A to the supercurrent and any other external currents used to generate the
applied field. The solutions generally require numerical analysis, so we shall only consider limiting cases in
the following.
If we assume an infinite bulk superconductor in zero field, the thermal Green functions become
translationally invariant so that EQs. (XI.151) – (XI.153) are rewritten into the following:
⎡ ∇2 ⎤
⎢ ν 2m + μ ⎥ G ( x − x′; ων ) + Δ F ( x − x′; ων ) = δ ( x − x′ ) ,
ω †
i + (XI.154)
⎣ ⎦
⎡ ∇2 ⎤ †
⎢ −iων + 2m + μ ⎥ F ( x − x′; ων ) − Δ G ( x − x′; ων ) = 0 ,
*
(XI.155)
⎣ ⎦
γ
Δ* = lim ∑ e − iων η F †( x = 0; ων ) . (XI.156)
β →0 + ν
η
d 3k d 3k
G ( x, ων ) = ∫ ei k i xG ( k , ων ) , F †( x, ων ) = ∫ ei k i x F †( k , ων ) , (XI.157)
( 2π ) ( 2π )
3 3
we find that EQs. (XI.154) and (XI.155) take the following simple forms:
⎡ ⎛ k2 ⎞⎤
ω
⎢ ν ⎜
i − − μ ⎟ ⎥ G ( k ; ων ) + Δ F †( k ; ων ) ≡ [iων − ξk ]G ( k ; ων ) + Δ F †( k ; ων ) = 1 , (XI.158)
⎣ ⎝ 2m ⎠⎦
⎡ ⎛ k2 ⎞⎤
−
⎢ ν ⎜
iω − − μ ⎟ ⎥ F †( k ; ων ) − Δ*G ( k ; ων ) ≡ [ −iων − ξk ] F †( k ; ων ) − Δ*G ( k ; ων ) = 0 . (XI.159)
⎣ ⎝ 2m ⎠⎦
− ( iων + ξ k )
G ( k ; ων ) = 2
, (XI.160)
ων2 + ξk2 + Δ
Δ*
F ( k ; ων ) =
†
2
. (XI.161)
ων2 + ξk2 + Δ
In the absence of an applied field, the superconducting gap Δ can be taken as real without loss of
generality. Therefore, F †( k ; ων ) = F ( k ; ων ) and EQs. (XI.160) and (XI.161) become
uk2 vk2
G ( k ; ων ) = + , (XI.162)
iων − Ek iων + Ek
⎡ 1 1⎤
F †( k ; ων ) = F ( k ; ων ) = −uk vk ⎢ + ⎥, (XI.163)
⎣ iων − Ek iων + Ek ⎦
where
(
Ek ≡ ξk2 + Δ )
2 1/ 2
, (XI.164)
Δ
uk vk = , (XI.165)
2 Ek
1⎛ ξk ⎞
vk2 = 1 − uk2 = ⎜1 − ⎟ . (XI.166)
2 ⎝ Ek ⎠
Here Ek denotes the quasiparticle energy and vk2 represents the distribution function for quasiparticles, as
illustrated in Fig. XI.2.1. Additionally, by inserting EQ. (XI.161) into the self-consistent expression in EQ.
(XI.156) for the present uniform medium, we obtain
γ γ d 3k Δ
Δ= lim ∑ e − iων η F †( x = 0; ων ) = ∑∫ . (XI.167)
β η →0 + ν β ( 2π ) ων + Ek2
3 2
ν
1
uk2
vk2
μ εk
Δ
Fig.XI.2.1 The dependence of the distribution functions vk2 = 1 − uk2 ( ) and uk2 of superconductors on the
normal state energy ε k , assuming T = 0. The occupied state distribution function vk2 exhibits a
smooth energy spread on the order of the superconducting gap Δ around the chemical potential
μ. This behavior differs from the Fermi liquid theory where a discontinuity in the distribution
function, known as the quasiparticle residue Zk, exists at the Fermi level.
d 3k dξ ⎛ (ξ 2 + Δ 2 )1/ 2 ⎞
1 ⎛1 ⎞ ωD
1=γ ∫ tanh ⎜ β Ek ⎟ ≈ N ( ε F ) γ ∫ tanh ⎜ ⎟, (XI.168)
( 2π ) (ξ 2 + Δ 2 ) ⎜ 2 k BT ⎟
1/ 2
⎝2 ⎠
3 0
2 Ek
⎝ ⎠
where N ( ε F ) denotes the density of states at the Fermi level. Therefore, this superconducting gap equation
takes the following form in extreme temperature limits:
ωD dξ
1 ≈ N (ε F ) γ ∫0 (ξ , (T → 0 ) (XI.169)
)
1/ 2
2
+ Δ2
ωD dξ ⎛ ξ ⎞
1 ≈ N (ε F ) γ ∫0 ξ
tanh ⎜ ⎟. (T → T )c
−
(XI.170)
⎝ 2k BTc ⎠
The temperature dependence of the superconducting gap can be obtained by numerically solving for Δ(T) in
EQ. (XI.168), although simpler analytical forms may be derived for special cases such as the weak coupling
limit, which you will be asked to consider in Problem Set 2. The limiting behavior of Δ(T) satisfies the
following temperature dependence:
⎛ Δ0 ⎞
Δ ( T ) ≈ Δ 0 − ( 2πΔ 0 k BT ) (T Tc )
1/ 2
exp ⎜ − ⎟, (XI.171)
⎝ k BT ⎠
1/ 2 1/ 2 1/ 2
⎛ 8 ⎞ ⎛ T ⎞ ⎛ T ⎞
Δ ( T ) ≈ k BTcπ ⎜ ⎟ ⎜1 − ⎟ ≈ 3.06k BTc ⎜ 1 − ⎟ , (Tc − T Tc ) (XI.172)
⎝ 7ζ ( 3) ⎠ ⎝ Tc ⎠ ⎝ Tc ⎠
To find an expression for Tc, we note that ω D 2k BTc so that the integral in EQ. (XI.170) becomes
ωD dξ ⎛ ξ ⎞ ωD 2 kBT dz ωD 2 k BT ω D 2 k BT
∫0 tanh ⎜ ⎟ = ∫0 tanh ( z ) = ln ( z ) tanh ( z ) 0 −∫ dz ln ( z ) sech 2( z )
ξ ⎝ 2k BTc ⎠ z 0
⎛ ω ⎞ ∞ ⎛ ω ⎞ ⎛ 4e γ 0 ⎞ ⎛ 2eγ 0 ω D ⎞
≈ ln ⎜ D ⎟ − ∫ dz ln ( z ) sech 2( z ) ≈ ln ⎜ D ⎟ + ln ⎜ ⎟ = ln ⎜ π k T ⎟ ( γ 0 ≈ 0.5772 ) , (XI.173)
⎝ 2k BTc ⎠ 0 ⎝ 2k BTc ⎠ ⎝ π ⎠ ⎝ B c ⎠
so that
γ0
2e ⎡ ⎛ 1 ⎞⎤ ⎡ ⎛ 1 ⎞⎤
k BTc ≈ ωD exp ⎢ − ⎜ ⎟ ⎥ ≈ 1.13 ω D exp ⎢ − ⎜ ⎟⎥ . (XI.174)
π ⎣ ⎝ N (ε F ) γ ⎠⎦ ⎣ ⎝ N (ε F ) γ ⎠⎦
Therefore, we find that the superconducting transition temperature increases with the increasing electronic
density of states at the Fermi level N ( ε F ) , electron-phonon coupling strength γ, and the Debye frequency
ωD of phonons. From EQs. (XI.121) and (XI.174), we also note that
Δ0 π
= γ0
≈ 1.76 , (XI.175)
k BTc e
Next, we compare the results of BCS theory with thermodynamic functions of superconductors. We
first consider the change in the thermodynamic potential between the superconducting state (ΩS) and the
normal state (ΩN) as the result of an attractive electron-phonon interaction Hamiltonian Hˆ e - ph :
dλ 1 γ dγ ′
d 3 x γ ′ ψ α† ( x )ψ β† ( x )ψ β ( x )ψ α ( x )
1
ΩS − Ω N = ∫
0
λ
λ Hˆ e − ph = −
2 ∫0 γ′ ∫
2
γ ⎛1⎞
∫ d x Δ( x )
2
≈ −∫ dγ ′ ⎜ ⎟ 3
, (XI.176)
0
⎝γ′⎠
where we have used the definition of Δ in EQ. (XI.153). For a uniform system, EQ. (XI.176) can be
simplified into the following expression:
2
⎛1⎞ 2 γ
ΩS − Ω N = − Ω ∫ ⎟ Δ dγ ′ ⎜
0
⎝γ′⎠
γ d (1/ γ ′ ) 2 Δ d (1/ γ )
( Δ′ ) ,
2
= Ω ∫ dγ ′ Δ = Ω ∫ d Δ′ (XI.177)
0 dγ ′ 0 dΔ ′
where in the last line we have changed variables from the electron-phonon coupling coefficient γ to the
superconducting gap Δ because they are related by EQ. (XI.168). Inserting EQ. (XI.168) into EQ. (XI.177),
we obtain
ΩS − Ω N ωD Δ ∂ ⎧1 ⎛ β E′ ⎞⎫
= N (ε F ) ∫0 ∫0 d Δ′ ( Δ′ )
2
dξ ⎨ tanh ⎜ ⎟⎬
Ω ∂Δ′ ⎩ E ′ ⎝ 2 ⎠⎭
ωD ⎡ Δ2 ⎛ βE ⎞ Δ Δ′ ⎛ β E′ ⎞⎤
= N (ε F ) ∫0 d ξ ⎢ tanh ⎜ ⎟ − 2 ∫0 d Δ′ ′ tanh ⎜ ⎟⎥
⎣E ⎝ 2 ⎠ E ⎝ 2 ⎠⎦
Δ2 ⎡ ωD Δ Δ′ ⎛ β E′ ⎞⎤
= − ⎢ 2N ( ε F ) ∫ d ξ ∫ d Δ′ tanh ⎜ ⎟
γ ⎣ 0 0 E′ ⎝ 2 ⎠ ⎥⎦
Δ2 ⎡ ωD ⎛ β E′ ⎞⎤
− ⎢ 2N ( ε F ) ∫ d ξ ∫ dE ′ tanh ⎜
E
= ⎟
γ ⎣ 0 0
⎝ 2 ⎠ ⎥⎦
Δ2 ⎡ 4N ( ε F ) ωD ⎛ cosh ( β E 2 ) ⎞ ⎤
= −⎢ ∫ d ξ ln ⎜ ⎟⎥
γ ⎣ β ⎝ cosh ( βξ 2 ) ⎠ ⎦
0
Δ2 4N ( ε F ) ⎡ 1 ⎤ 4N ( ε F )
( ) ( )
ωD ωD
= − ∫0 d ξ ⎢ ln 1 + e − β E + β ( E − ξ )⎥ + ∫0 d ξ ln 1 + e − βξ . (XI.178)
γ β ⎣ 2 ⎦ β
At low temperatures where βωD 1 , the last term of EQ. (XI.178) may be approximated by
4N ( ε F ) 4N ( ε F ) 4N ( ε F ) π 2 π2
( ) ( )
ωD ∞
∫0 ∫0 = N (ε F ) ( k BT )
2
d ξ ln 1 + e − βξ ≈ d ξ ln 1 + e − βξ = . (XI.179)
β β β 2
12 3
ωD ⎡1 ⎛ 2ω ⎞⎤
−2N ( ε F ) ∫ d ξ ( E − ξ ) ≈ −N ( ε F ) ⎢ Δ 2 + Δ 2 ln ⎜ D ⎟⎥
0
⎣2 ⎝ Δ ⎠⎦
⎡1 ⎛ 2ω ⎞ ⎛ Δ ⎞⎤ ⎡1 Δ2 ⎛ Δ ⎞⎤
= −N ( ε F ) ⎢ Δ 2 + Δ 2 ln ⎜ D ⎟ + Δ 2 ln ⎜ 0 ⎟ ⎥ = −N ( ε F ) ⎢ Δ 2 + + Δ 2 ln ⎜ 0 ⎟ ⎥
⎣2 ⎝ Δ0 ⎠ ⎝ Δ ⎠⎦ ⎣2 N (ε F ) γ ⎝ Δ ⎠⎦
1 Δ2 ⎛Δ ⎞
= − N (ε F ) Δ2 − − N ( ε F ) Δ 2 ln ⎜ 0 ⎟, (XI.180)
2 γ ⎝ Δ ⎠
where we have used EQ. (XI.121) with γ = V/2 in the second line of EQ. (XI.180) and Δ0 refers to the
superconducting gap at T = 0. Inserting EQs. (XI.179) and (XI.180) into EQ. (XI.178), we obtain
ΩS − Ω N Δ2⎡ ⎛ Δ0 ⎞⎤ π2
N ( ε F ) ⎢1 + 2 ln ⎜ ⎟ ⎥ + N ( ε F ) ( k BT ) − 4k BT N ( ε F ) ∫ d ξ ln (1 + e − β E ) . (XI.181)
2 ωD
≈−
Ω 2 ⎣ ⎝ Δ ⎠⎦ 3 0
ΩS − Ω N Δ 02
=− N (ε F ) . (XI.182)
Ω 2
Moreover, from EQ. (XI.181) and the relation between the thermodynamic potential and the free energy, we
arrive at the following for the free energy in the low temperature limit:
FS − FN Δ2 π2
N (ε F ) + N (ε F ) ( k BT )
2
≈− . (XI.183)
Ω 2 3
To find the specific heat in the T → 0 limit, we approximate the last term in EQ. (XI.181) by setting
ωD → ∞ and Δ ≈ Δ 0 so that
ωD
∫0 ( )
∞
{
d ξ ln 1 + e − β E ≈ ∫ d ξ ln 1 + exp ⎡ − β ξ 2 + Δ 02
0 ⎣ ( )
1/ 2
}
⎤ ≈ ∞ d ξ exp ⎡ − β (ξ 2 + Δ 02 )1/ 2 ⎤
⎦ ∫0 ⎣ ⎦
⎛ 1ξ2 ⎞ βξ 2
− βΔ0 ⎜⎜1+ ⎟ − 1
e − βΔ0 ( 2Δ 0πβ −1 ) .
∞ 2 Δ 02 ⎟⎠ − βΔ0 ∞ 1/ 2
≈ ∫ dξ e ∫0
2 Δ0
⎝
=e dξ e = (XI.184)
0 2
Therefore,
ΩS Ω N (T = 0 ) Δ 02
Ω
≈
Ω
−
2
N ( ε F ) − 2N ( ε F ) e
− β Δ0
( 2π Δ β ) 0
−3 1/ 2
, (XI.185)
To find the specific heat near Tc where βΔ 1 , we consider the expression for thermodynamic
potential in EQ. (XI.177) and express the term (1/γ) using EQ. (XI.167) as follows:
1 1 ωD 1 2 ωD 1
= N (ε F ) ∫ dξ ∑ω = N (ε F ) ∫ dξ ∑ ω2 +
γ β −ωD
ν ν
2
+E 2
k
β 0
ν ν (ξ 2
+ Δ2 )
2 ωD
⎡ 1 Δ2 ⎤
= N (ε F ) ∫ dξ ∑ ⎢ − + ⎥. (XI.187)
β ν ⎢ ων + ξ ( )
2 2
⎥
0 2 2
⎣ ων
2
+ ξ ⎦
Taking derivative of EQ. (XI.187) relative to Δ and inserting the result into EQ. (XI.177), we obtain
ΩS − Ω N N (ε F ) Δ4 ωD 1 N (ε F ) Δ4 ∞ 1
≈− ∫0 dξ ∑ ≈− ∑∫ dξ
Ω β (ω ) β (ω )
2 2
+ξ2 +ξ2
2 0 2
ν ν
ν ν
N (ε F ) Δ4 π 1 N (ε F ) Δ4 β 2 ∞
1
=−
β
∑ 4 ω3
ν
= −
2π 2
∑ ( 2ν + 1)
ν=0
3
ν
N ( ε F ) Δ 4 β 2 7ζ ( 3)
=−
2π 2 8
2
8 1⎛ T ⎞
N ( ε F )( π k BTc )
2
=− ⎜1 − ⎟ , (XI.188)
7ζ ( 3) 2⎝ Tc ⎠
where we have used EQ. (XI.172) in the last line of EQ. (XI.188). Hence, the specific heat difference at Tc
becomes
CS − C N 8
Ω
≈
7ζ ( 3)
N ( ε F ) π 2 k B2Tc . ( (XI.189) )
Tc
Noting that
2π 2
(k T ) ,
CN
≈ N (ε F ) 2
(XI.190)
Ω
B
3
we find that
CS − C N 12
≈ ≈ 1.43 . (XI.191)
CN Tc
7ζ ( 3)
A schematic illustration of the electronic specific heat of a superconductor is shown in Fig. XI.2.2. The
discontinuity ratio as given in EQ. (XI.191) is supposedly universal and independent of material properties,
which is in reasonable agreement with empirical findings in simple conventional superconductors.
CN
CS
T
0 Tc
Fig.XI.2.2 Schematic illustration of the electronic specific heat (C) of a BCS superconductor as a function
of temperature (T), showing exponentially decreasing specific heat with decreasing temperature
at T → 0, a linear temperature dependence above Tc, and a discontinuity at Tc.
Overall, BCS theory has been successful in explaining the thermodynamic and electrodynamic
properties of conventional superconductors where superconductivity is a well defined ground state mediated
by the electron-phonon interaction. However, in the case of strongly correlated superconductors with a
ground state not uniquely defined by superconductivity, such as in high-temperature superconducting
cuprates and some of the heavy fermion superconductors, the predictability power of BCS theory diminishes,
and a new paradigm of microscopic theory is needed to account for the pairing mechanism and the
anomalous low-energy excitations.
One of the most powerful tools in the investigation of the pairing symmetry and low-energy
excitations of novel superconductors is the quasiparticle tunneling spectroscopy, particularly the scanning
tunneling spectroscopy (STS) that provides high spatial and energy resolution, and the angle-resolved photo-
emission spectroscopy (ARPES) that provides complementary momentum and energy resolution. In the
following section, we investigate the theory of quasiparticle tunneling in superconductors and discuss how to
apply the theory to investigation of the pairing symmetry of superconductors. We assume that the ground
state of the system under consideration consists of nothing but superconductivity, and for simplicity we also
assume infinite quasiparticle lifetime and neglect any scattering of quasiparticles by impurities and bosonic
excitations such as phonons. A more general consideration of the quasiparticle excitation spectra in more
complicated superconductors (such as the high-temperature superconducting cuprates with coexistence of
competing orders besides superconductivity in the ground state) will require the Green function techniques.
In Part XIII we shall investigate specific cases of the low-energy excitations in cuprate superconductors by
explicitly incorporating more than one order parameter in the ground state and by permitting finite lifetimes
for the low-energy quasiparticle excitations.
In this section we focus on the physics of quasiparticle tunneling through junctions. The reason for
devoting an entire section to the discussion of quasiparticle tunneling is because it is a powerful tool for
characterizing various microscopic properties of superconductors such as the quasiparticle density of states,
superconducting energy gap, pairing symmetry, impurity response, and possible competing orders coexisting
with superconductivity. This process excludes the coherent tunneling of Cooper pairs so that the problem is
simplified significantly, because we may effectively treat the superconductor as a gapped electronic system
like a semiconductor, except that the superconducting density of states is very different from that of a
semiconductor. In reality, however, quasiparticles still contain coherent information associated with
superconductivity so that the transmission and reflection of quasiparticles at the interfaces of a
superconductor with other materials can retain information associated with the global phase coherence of the
superconductor. This point can be understood by the following consideration.
The coefficients uk and vk that we introduced in EQs. (XI.165) and (XI.166) in fact contain the
coherent information of superconductivity. In particular, the expression in EQ. (XI.165) indicates that the
product of uk and vk acquires the same phase as that of the superconducting order parameter Δ. Alternatively,
we note that the BCS ground state ψ G may be expressed in terms of the coefficients uk, vk and the phase ϕ
of the superconducting order parameter:
ψG = ∏ ( uk + vk ck†↑c−†k ↓ ) 0 = ∏ ( uk )
+ vk eiϕ ck†↑ c−†k ↓ 0 , (XI.192)
k = k1 , k M k = k1 , k M
where 0 represents vacuum, and ck† denotes the particle creation operator. Here the BCS ground state is
constructed from the BCS wavefunction for N-electrons forming N/2 pairs over M available k-values:
where ki and kl denote the first and the last of the occupied states among the M available k-values. Given that
there are
M! 1020
~ 10
[ M − ( N 2 ) ] !( N 2 ) !
different ways of choosing the (N/2) states for pair occupancy and an equal number of the coefficient
g ( k i , … , k l ) to determine, where M > (N/2) ~ 1020, we may treat the problem with the mean-field approach,
and we are only concerned with the mean value of N being fixed. The form given in EQ. (XI.192) therefore
implies that the probability of the pair ( k ↑, −k ↓ ) being occupied is vk , and the probability for it being
2
The particle creation and annihilation operators can be expressed through the Bogoliubov
transformation into quasiparticle operators α k and β k with uk and vk as coefficients:
where the transformation is canonical because the quasiparticle operators satisfy the following relations:
{α †
k, α k†′ } = {α k , α k ′ } = {β k† , β k†′ } = {βk , βk ′ } = {α k† , βk†′ } = {α k , βk ′ } = 0 . (XI.197)
Therefore, the superconducting coherence information is retained in the quasiparticle operators via EQs.
(XI.194) and (XI.165).
If we use the operator γk to represent in general both quasiparticle operators αk and βk, we find that
there is one-to-one correspondence between the quasiparticle excitation spectrum Ek (γk) and the normal
particle excitation spectrum ξk (ck), as illustrated in Fig. XI.3.1 (a). Consequently, we have
N S ( E ) dE = N N ( ξ ) d ξ , (XI.198)
dξ N S( E ) N S( E ) E
⇒ = ≈ = , (E > Δ); (XI.199)
dE N N (ξ ) N (ε F ) E − Δ2
2
= 0, (E < Δ).
where NS (E) and NN (ξ) in EQ. (XI.198) denote the density of states (DOS) in the superconducting and
normal states, respectively:
⎛ ξk ⎞ ⎛ ξk ⎞
N S ( E ) = ∑ k 2vk2 = ∑ k ⎜1 − ⎟ = ∑ ⎜1 −
k⎜
⎟, (XI.200)
S
⎝ Ek ⎠ ξk + Δ 2
2 ⎟
⎝ ⎠
⎛ ξ ⎞ Ω
N N ( ξ ) = ∑ k 2vk2 = ∑ k ⎜1 − k ⎟ =
kF 3
3 ∫0
d k, (XI.201)
⎝ ξk ⎠ ( 2π )
N
and the corresponding DOS for a BCS superconductor of s-wave pairing symmetry (i.e. with an isotropic
superconducting gap) is illustrated in Fig. XI.3.1 (b).
(a) (b)
Ek NS(Ek)/
NN(0)
NS(Ek)
Δ 1
NN(ξk)
μ εk −Δ Δ Ek
Fig. XI.3.1: (a) The one-to-one correspondence between the excitation spectrum of the normal (ξ ≡ ε − μ)
and the superconducting states. (b) The normalized density of states (DOS) of a BCS superconductor with s-
wave pairing symmetry at T << Tc.
Having established the physical significance of the coefficients uk and vk and the quasiparticle
operators γk, we are ready to describe the quasiparticle wavefunctions across a tunneling junction, which are
essential for establishing the theory of quasiparticle tunneling. To deal with quasiparticle tunneling
spectroscopy, we must consider the situation of spatially varying pairing potential Δ(x), chemical potential
μ(x) and Hartree potential V(x). Although in Part XI.2 we have primarily focused on the discussion of
uniform superconductors, we note that the Gorkov equations based on the Green function formalism are
generally applicable to superconductors with spatially varying properties. In fact, we may rewrite the Gorkov
equations into an equivalent formalism in the mean-field approximation, known as the Bogoliubov de-
Gennes (BdG) equations:
⎡ ∇2 ⎤ ∂uk ( x, t )
⎢ − 2m + V ( x ) − μ ( x ) ⎥ uk ( x, t ) + Δ( x ) vk ( x, t ) = i ∂t , (XI.202)
⎣ ⎦
⎡ ∇2 ⎤ ∂v ( x, t )
− ⎢− + V ( x ) − μ ( x ) ⎥ vk ( x, t ) + Δ* ( x ) uk ( x, t ) = i k . (XI.203)
⎣ 2m ⎦ ∂t
If we consider regions of the superconductor far away from its interface with other materials (e.g. an
insulating layer, a normal metal, or vacuum), we may treat Δ(x), μ(x) and V(x) as constants. In this case, the
solutions to the BdG equations are time-independent plane waves. Hence, we may express uk(x,t) and vk(x,t)
as follows:
uk ( x, t ) = uk ei k i x −iEk t , vk ( x, t ) = vk ei k i x −iEk t . (XI.204)
⎡ k2 ⎤
⎢ 2m − μ ( x ) ⎥ uk + Δvk = Ek uk , (XI.205)
⎣ ⎦
⎡ k2 ⎤
−⎢ − μ ( x ) ⎥ vk + Δuk = Ek vk . (XI.206)
⎣ 2m ⎦
Consequently, there are four k values associated with each energy Ek, where
k ±2
= μ ± Ek2 − Δ 2 . (XI.207)
2m
In EQ. (XI.207) we refer k+ to electron-like and k− to hole-like quasiparticles. The rationale for the definitions
can be understood by comparing the following relations with Fig. XI.2.1:
1⎛ Ek2 − Δ 2 ⎞ 1 1⎛ Ek2 − Δ 2 ⎞ 1
uk2+ = ⎜1 + ⎟ ≡ uk20 > , vk2+ = ⎜1 − ⎟ ≡ vk20 < ; (XI.208)
2⎜ Ek ⎟ 2 2⎜ Ek ⎟ 2
⎝ ⎠ ⎝ ⎠
1⎛ Ek2 − Δ 2 ⎞ 1 1⎛ Ek2 − Δ 2 ⎞ 1
uk2− = ⎜1 − ⎟ ≡ vk20 < , vk2− = ⎜1 + ⎟ ≡ uk20 > . (XI.209)
2⎜ Ek ⎟ 2 2⎜ Ek ⎟ 2
⎝ ⎠ ⎝ ⎠
⎡ u k ( x, t ) ⎤
ψk = ⎢ ⎥, (XI.210)
⎣ vk ( x, t ) ⎦
⎡ ⎤
⎢ 1⎛ Ek2 − Δ 2 ⎞
⎜1 ± ⎟⎥
⎡u± k± ( x ) e −iEk t ⎤ ⎢ 2⎜ ⎟⎥
Ek
± ik ± i x ⎢ ⎝ ⎠ ⎥ −iEk t
ψ ±S k± =⎢ ⎥ = e e . (XI.211)
⎢⎣ v± k± ( x ) e −iEk t ⎥⎦ ⎢ ⎥
1⎛ Ek − Δ ⎞ ⎥
2 2
⎢ ⎜1 ∓ ⎟
⎢ 2⎜ Ek ⎟⎥
⎣ ⎝ ⎠⎦
Similarly, we may consider the normal state far away from the interface of a junction so that Δ(x) =
V(x) = 0 and we find for each Ek there are two k values:
k ±2
= μ ± Ek , (XI.212)
2m
⎡u± k+ ( x ) e −iEk t ⎤
± ik + i x ⎡ 1 ⎤ − iEk t
ψ ± k+
N
=⎢ ⎥ = e ⎢0 ⎥ e , (electron branch) (XI.213)
⎢⎣ v± k+ ( x ) e −iEk t ⎥⎦ ⎣ ⎦
⎡u± k− ( x ) e −iEk t ⎤ ⎡0 ⎤
ψ ± k−
N
=⎢ ⎥ = e ± ik − i x ⎢ ⎥ e −iEk t , (hole branch) (XI.214)
⎢⎣ v± k− ( x ) e − iEk t
⎥⎦ ⎣1 ⎦
k +2
Ek = −μ ( k+ > kF ) , (XI.215)
2m
k−2
Ek = μ − ( k− < kF ) . (XI.216)
2m
Having established the quasiparticle wavefunctions for the superconducting and normal states, we
are ready to evaluate the quasiparticle tunneling spectroscopy [i.e. the quasiparticle tunneling conductance
(or tunneling current) versus energy (or voltage) spectroscopy] in a normal metal-insulator-superconductors
(N-I-S) junction. There are two typical approaches to address this problem. One is the transfer Hamiltonian
formalism following the original work by Bardeen [J. Bardeen, Phys. Rev. Lett. 6, 57 (1961)] that considers a
tunnel junction as consisting of two electrodes separated by a tunnel barrier and assumes that the left and
right electrodes can be described nearly independently by the Hamiltonians HL and HR in the ground state
except that they are weakly coupled by a perturbation Hamiltonian HT. The quasiparticle tunneling is then
treated as the excited states in which particles are transferred from one electrode to the other according to the
tunneling Hamiltonian HT. The advantage of the transfer Hamiltonian formalism is that it is conceptually
An alternative approach that amends the problem encountered in the transfer Hamiltonian formalism
was first developed by Blonder, Tinkham and Klapwijk [G. E. Blonder, M. Tinkham, and T. M. Klapwijk,
Phys. Rev. B 25, 4515 (1982)] for quasiparticle tunneling through one-dimensional N-I-S junctions, which is
known as the BTK theory. The BTK theory views the quasiparticle tunneling process as a scattering process
and solves the BdG equation where the superconducting order parameter Δ(x) serves as a position dependent
off-diagonal potential and the interface as a diagonal delta-function potential with a variable barrier height
Hδ(x). The original BTK theory was developed as a one-dimensional problem for quasiparticle tunneling into
s-wave superconductors. It was recently generalized by Hu and by Tanaka & Kashiwaya for two-dimensional
interfaces with unconventional superconductors of other pairing symmetries [C. R. Hu, Phys. Rev. Lett. 72,
1526 (1994); Y. Tanaka and S. Kashiwaya, Phys. Rev. Lett. 74, 3451 (1995); S. Kashiwaya and Y. Tanaka,
Phys. Rev. B 53, 2667 (1996)]. We shall consider both models in this section.
We first consider a simple case of quasiparticle tunneling in an N-I-S junction where the
superconductor is of s-wave pairing symmetry and the tunneling process is assumed to be one dimensional,
following the original theory by Blonder, Tinkham and Klapwijk [G. E. Blonder, M. Tinkham, and T. M.
Klapwijk, Phys. Rev. B 25, 4515 (1982)]. In this model, the interface is represented by a repulsive delta-
function potential U ( x ) = H δ ( 0 ) and the tunneling is restricted to elastic processes. There are only four
allowed processes if we impose the constraint that an incident particle with a positive group velocity can only
result in either transmission with a positive group velocity or reflection with a negative group velocity. These
four processes are illustrated in Fig. XI.3.2. Specifically, an incident electron from the normal electrode can
be Andreev-reflected as a hole (accompanied by a Cooper pair transmission into the superconductor),
normal-reflected as an electron, transmitted as an electron-like quasiparticle (ELQ), or transmitted as a hole-
like quasiparticle (HLQ). Therefore, for electrons traveling from N to S across the junction, the
wavefunctions in the normal and superconducting electrodes may be represented by the following forms:
The boundary conditions needed to solve for the coefficients a, b, c, and d in EQ. (XI.217) are
simply the continuity of wavefunctions and the conservation of momentum at x = 0:
ψ N ( 0) = ψ S ( 0) = ψ ( 0) , (XI.218)
1 dψ ( 0 ) 1 dψ ( 0 )
S N
− = Hψ ( 0 ) . (XI.219)
2m dx 2m dx
Hence, for a given incident electron energy E, we obtain the probability for Andreev reflection A(E) and that
for normal reflection B(E) as follows:
A( E ) = a ( E ) =
2 u02 v02
, B(E) = b(E) =
(u 2
2
0 ) (
− v02 Z 2 1 + Z 2 ), ( E > Δ) (XI.220)
γ 2
γ 2
Δ2
A( E ) = a ( E ) = B (E) = b(E) = 1− A , ( E < Δ)
2 2
, (XI.221)
(
E 2 + Δ 2 − E 2 1 + 2Z )( )
2 2
E E
Electron Hole Hole Electron Hole-like Electron-like
branch branch branch branch quasiparticles quasiparticles
B A D C
Δ
k
+ − − +
−kF kF −k −kF −k k kF k
Fig. XI.3.2: Schematic illustrations of the energy versus momentum diagrams for the four allowed processes
associated with an incident electron at the N-S interface x = 0: the Andreev reflected hole (A), the normal
reflected electron (B), the transmitted electron-like quasiparticle (C), and the transmitted hole-like
quasiparticle (D). The open (closed) circles represent holes (electrons), and the arrows point in the direction
of the corresponding group velocity.
dI
∝ ⎡1 + a( E ) − b( E ) ⎤
2 2
dV ⎣ ⎦
for a range of barrier heights Z has been calculated and illustrated in Fig. XI.3.3. We note that in the Z → 0
limit the conductance within the superconducting gap (E < Δ) is nearly doubled because most of the electrons
are Andreev reflected at the N-S interface, so that for each Andreev reflected hole a Cooper pair of electrons
are transmitted across the interface. On the other hand, in the high barrier limit (Z >>1), the tunneling
conductance given by the BTK formalism is essentially the same as that given by the transfer Hamiltonian to
Fig. XI.3.3: Differential tunneling conductance versus biased voltage (eV) for various normalized barrier
heights Z at T = 0, showing enhanced sub-gap conductance associated with complete Andreev reflection at Z
= 0 and completely gapped behavior in the tunneling limit with Z >> 1. [Adapted from Fig. 7 of Blonder,
Tinkham and Klpawijk (1982)]
It should be noted that the BTK theory is a mean-field theory based on the BdG equations, which
does not consider effects associated with either finite quasiparticle lifetimes or many-body interactions such
as quasiparticle coupling with bosonic excitations. In the following we generalize the original BTK theory
for one-dimensional quasiparticle tunneling between a normal metal and an s-wave superconductor to two-
dimensional tunnel junctions involving unconventional superconductors. However, we remark that the
generalized BTK theory remains a mean-field approximation. To incorporate fluctuations or many-body
effects such as quasiparticle coupling with phonons or spin excitations, one must return to the Green function
formalism and find an appropriate self-energy that models the effects under consideration.
The process of Andreev reflection at the N-S interface in the low-barrier limit was first realized by
Andreev when he considered an electron (or a hole) of energy E approaching an N-S interface from the N-
region. If E < |Δ|, the electron (hole) would be reflected as a hole (an electron), provided that the length scale
for |Δ| variation was much larger than the Fermi wavelength. The Andreev reflection process not only plays
an important role in the quasiparticle tunneling across conventional N-S junctions, but is also essential in the
formation of a novel zero-energy surface state for superconductors of unconventional pairing symmetries.
The novel zero-energy surface state in the N-S junctions of unconventional superconductors is manifested by
a zero-bias conductance peak (ZBCP) in the tunneling conductance spectra. [C. R. Hu, Phys. Rev. Lett. 72,
1526 (1994); Y. Tanaka and S. Kashiwaya, Phys. Rev. Lett. 74, 3451 (1995); S. Kashiwaya and Y. Tanaka,
The generalize BTK theory begins with the BdG equations for an inhomogeneous superconductor
with anisotropic pairing potential:
E f ( r1 ) = H 0( r1 ) f ( r1 ) + ∫ dr2 Δ( r1 , r2 ) g ( r2 ) ,
E g ( r1 ) = −H 0( r1 ) g ( r1 ) + ∫ dr2 Δ( r1 , r2 ) f ( r2 ) , (XI.224)
∇r21
where ( )
H 0 r1 = −
2m
− μ + U r1 ,( )
μ denotes the chemical potential, U ( r1 ) is the Hartree potential, Δ( r1 , r2 ) represents the pairing potential,
and we have introduced the two-component column vector to represent the quasiparticle wavefunctions:
⎡ f ( r1 ) ⎤
Ψ( r1 ) = ⎢ ⎥. (XI.225)
⎣ g ( r1 ) ⎦
We may rewrite the pairing potential in terms of the relative coordinates r = r1 – r2 and the center-of-
mass R = (r1 + r2)/2 so that Δ( r1 , r2 ) ≡ Δ( r, R ) . The Fourier transform of the pairing potential becomes:
⎛ k k ⎞
Δ( k , R ) = ∫ dr e − ik ir Δ( r , R ) ≡ Δ( γˆ , R ) , ⎜ γˆ ≡ ≈ ⎟ . (XI.226)
⎝ k kF ⎠
Therefore, in the quasi-classical approximation, Δ( k , R ) represents the pairing potential that quasiparticles
of momentum k experienced at position R. Next, by introducing two envelope functions u γˆ , r1 ( ) and
( )
v γˆ , r1 to factor away the fast atomic-scale oscillations, we rewrite the column vector in EQ. (XI.225) into
the following form:
⎡ f ( r1 ) ⎤ i kF γˆ ir1 ⎡u ( γˆ , r1 ) ⎤
Ψ ( r1 ) = ⎢ ⎥=e ⎢ ⎥. (XI.227)
⎣ g ( r1 ) ⎦ ⎣⎢ v ( γˆ , r1 ) ⎦⎥
Using EQs. (XI.226) and (XI.227), the BdG equations is recast as follows:
( )
Eu γˆ , r1 = −i
kF
m
γˆ i∇r u ( γˆ , r1 ) + Δ( γˆ , r1 ) v( γˆ , r1 ) ,
( )
Ev γˆ , r1 = i
kF
m
γˆ i∇r v ( γˆ , r1 ) + Δ* ( γˆ , r1 ) u ( γˆ , r1 ) , (XI.228)
B γˆS C
γˆN′
γˆN γˆS′
A D
Fig. XI.3.4: Schematic diagram of the transmission and reflection processes at the N-I-S interface, adapted
from Fig. 1(b) of Kashiwaya et al. (1996). For an electron incident from the normal metal with an angle of
incident θN relative to the N/S interface, it can be Andreev reflected as a hole (A), normal reflected as an
electron with an angle θN (B), transmitted as electron-like quasiparticles (ELQ) with an angle of refraction θS
(C), or transmitted as hole-like quasiparticles (HLQ) with an angle of refraction θS (D). The inset shows the
alignment of the order parameter of a dx2-y2 superconductor relative to the N-I-S interface, where a(x) and
b(y) denote the crystalline axes.
Following similar derivations in the original BTK theory, we consider electrons incident from the
normal metal side with an energy E and an angle of incidence θN relative to the N/S interface, as shown in
Fig. XI.3.4. There are four possible trajectories for the incident electrons. They can be Andreev reflected as
holes with an angle θN relative to the N/S interface (A), normal reflected as electrons with an angle of
reflection θ N′ equal to θN (B), transmitted as electron-like quasiparticles (ELQ’s) with an angle of refraction
θS (C), or transmitted as hole-like quasiparticles (HLQ’s) with an angle of refraction θ S′ equal to θS (D). We
denote the unit vector of the momentum of the incident electrons in the normal metal as γˆN , of the reflected
electrons as γˆN′ , of the ELQ in the superconductor as γˆS , and of the HLQ as γˆS′ . Noting that the pairing
potential in the normal metal is identically zero so that Δ ( γˆN , R ) = Δ ( γˆN′ , R ) = 0 , we can solve for the
wavefunctions on the N side and obtain:
where a(E) and b(E) are the amplitudes for Andreev and normal reflection, respectively, and kFN is the Fermi
momentum in the normal electrode. Noting that typically k FN ( Em k FN ) , we rewrite EQ. (XI.229) into:
In the superconducting electrode, the pairing potentials experienced by the ELQ’s and HLQ’s are
respectively given by
iφ (γˆS )
Δ ( γˆS , R ) = Δ ( γˆS , R ) e ≡ Δ + eiφ+ (ELQ), (XI.231)
iφ (γˆS′ )
Δ ( γˆS′ , R ) = Δ ( γˆS′ , R ) e ≡ Δ − eiφ− (HLQ). (XI.232)
Therefore, the wavefunction for the transmitted electron-like and hole-like quasiparticles is given by
⎡ 2 ⎤
⎢ E + E 2 − Δ+ ⎥
i ⎛⎜ m E 2 − Δ + 2 k FS ⎞⎟ γˆS i r ⎢ 2E ⎥
i k FS γˆS i r
Ψ S ( r ) = c( E ) e e ⎝ ⎠
⎢ ⎥
⎢ E − E 2 − Δ+
2
⎥
⎢ e −iφ+ ⎥
⎣ 2E ⎦
⎡ 2 ⎤
⎢ eiφ− E − E − Δ −
2
⎥
− i ⎛⎜ m E 2 − Δ − 2 k FS ⎞⎟ γˆS′ i r ⎢ 2E ⎥
i k FS γˆS′ i r
+ d(E)e e ⎝ ⎠
⎢ ⎥, (XI.233)
⎢ E + E 2 − Δ−
2
⎥
⎢ ⎥
⎣ 2E ⎦
where kFS is the Fermi momentum in the superconducting electrode. Noting that k FS m ( E 2− Δ±
2
)
k FS ,
we can simplify EQ. (XI.233) into
⎡ 2 ⎤ ⎡ 2 ⎤
⎢ E + E 2 − Δ+ ⎥ ⎢ eiφ− E − E − Δ −
2
⎥
i k γˆ i r ⎢ 2E ⎥ i k FS γˆS′ i r ⎢ 2E ⎥
Ψ S ( r ) ≈ c( E ) e FS S ⎢ ⎥ + d(E)e ⎢ ⎥. (XI.234)
⎢ 2
⎥ ⎢ 2
⎥
− iφ+ E − E − Δ + E + E 2 − Δ−
2
⎢e ⎥ ⎢ ⎥
⎣ 2E ⎦ ⎣ 2E ⎦
To obtain the tunneling conductance across the N-I-S interface, we model the insulating interface as
a delta-function Hartree potential U ( r ) = H δ ( r ) , and solve for the coefficients a(E), b(E), c(E), and d(E) by
considering the same boundary conditions as given in EQs. (XI.218) and (XI.219), and also taking into
account the momentum conservation parallel to the interface, we obtain the tunneling conductance at T = 0:
1 + σ N Γ + + (σ N − 1) Γ + Γ −
2 2
σ S ( E , θ N ) ≡ σ N ⎡1 + a( E , θ N ) − b( E , θ N ) ⎤ = σ N
2 2
, (XI.235)
⎣ ⎦ i (φ− − φ+ )
2
1 + (σ N − 1) Γ + Γ − e
where
⎛ k FS cos θ S ⎞
2 2 4⎜ ⎟
E − E − Δ± ⎝ k FN cos θ N ⎠ mH
Γ± = , σN = , Z= . (XI.236)
Δ±
2 2
⎛ k FS cos θ S ⎞ ⎛ Z ⎞ k FN
⎜ 1 + k cos θ ⎟ + 4 ⎜ cos θ ⎟
⎝ FN N ⎠ ⎝ N ⎠
In an s-wave superconductor where Δ + = Δ − = Δ , EQ. (XI.235) reduces to the expressions in EQs. (XI.220)
– (XI.224).
Given EQ. (XI.235), we are ready to calculate the directional tunneling conductance spectra of
superconductors with unconventional pairing symmetries and to compare the results with experimental data
from realistic scanning tunneling spectroscopy (STS). In general the transverse momentum distribution for
the incident particles is limited and can be approximated by a finite tunneling cone β. (The limited transverse
momentum distribution can be understood in the context of WKB approximation, to be discussed later in this
section under the transfer Hamiltonian formalism). Thus, the resulting tunneling conductance reads:
dI (
− θ2 β2 ).
(V ) ∝ ∫ dθ ⎡⎣1 + a( E , θ ) − b( E , θ ) ⎤ e
2 2
(XI.237)
dV ⎦
We may further simplify the tunneling conductance by taking the approximation k FS ≈ k FN so that
θ N ≈ θ S ≈ θ . In the STS configurations, the experiments typically operate in the tunneling limit with a large
effective barrier height. Taking these factors into consideration, we can compute the quasiparticle tunneling
spectra of unconventional superconductors for given quasiparticle energy E, the incident angle θN relative to
the N-I-S interface, the alignment of the superconducting order parameter relative to the interface α (see the
inset of Fig. XI.3.4), and the superconducting pairing potential Δ(k). Examples of the superconducting
pairing potentials and the corresponding directional quasiparticle tunneling spectroscopy for s-wave and dx2-
y2 superconductors are illustrated in Fig. XI.3.5. Specifically, the pairing potential Δ(k) of a dx2-y2-wave
superconductor can be expressed by Δ ( k ) = Δ (θ k ) ≈ Δ d cos ( 2θk ) , where θ k is the angle between the
quasiparticle wave-vector k and the anti-nodal direction {100} of the dx2-y2-wave superconductor. Therefore,
for an N-I-S junction as shown in Fig. XI.3.4, the pairing potentials Δ+ and Δ− experienced by the ELQ’s and
HLQ’s become
where α denotes the angle between the anti-nodal direction {100} of the dx2-y2-wave superconductor relative
to the N/S interface normal. Hence, for quasiparticles tunneling along the {100} direction of a dx2-y2-wave
superconductor, we have (θ S − α ) = 0 , and the tunneling spectrum exhibits a U-shape feature similar to that
of an s-wave superconductor. For quasiparticles tunneling along the {110} nodal direction, we have
(θ S − α ) = π 4 and Δ + = −Δ − , implying that the pairing potential experienced by ELQ’s and HLQ’s are
opposite in sign, which is a specially interesting case with the tunneling spectrum exhibiting a sharp peak at
the zero bias, known as the zero bias conductance peak (ZBCP), which is associated with the Andreev bound
states to be discussed in more details below. In the event of tunneling along the c-axis {001} of a dx2-y2-wave
superconductor, the incident quasiparticles sample over all possible pairing potentials so that the tunneling
cone completely opens up ( β ≈ ∞ ). We find that Δ + = Δ − , and the tunneling spectrum exhibit a V-shape
feature, as shown in Fig. XI.3.5.
(a)
(b)
Fig. XI.3.5: Comparison of the pairing potential and the quasiparticle tunneling spectra of (a) s-wave
and (b) dx2-y2-wave superconductors, where Δ0 denotes the maximum pairing potential for a given
superconductor.
1 + (σ N − 1) Γ + Γ − e ( − + ) ≈ 1 − Γ + Γ − e ( − + ) = 0 .
i φ −φ i φ −φ
(XI.239)
Given that φ− − φ+ = π and Δ + = Δ − for tunneling along the {110} direction, we find that the bound state
with a diverging conductance can only occur at E = 0 according to EQs. (XI.236) and (XI.239). Physically,
the formation of the Andreev bound state at the nodal surface of an unconventional superconductor can be
visualized by capping the surface with a thin layer of normal metal of thickness dN, as illustrated in Fig.
XI.3.6: The trajectories of the quasiparticles initially incident from vacuum can form closed loops within the
thin normal layer through two Andreev reflections at the N/S interface and two specular reflections at the
surface. Applying the quantization condition that the phase shift is a multiple of 2π for a bound quasiparticle
along a closed path of the classical trajectory, and taking into account that the phase shifts picked up by the
two Andreev reflections are different because of the different pairing potentials experienced by the ELQ’s
and HLQ’s, we obtain the following relation that determines the bound state energy Eb:
⎛ Δ + − Eb2 ⎞
2 ⎛ Δ − − Eb2 ⎞
2
− tan ⎜
−1
⎟ − tan ⎜
−1
⎟ − (φ+ − φ− ) + 2 ⎛⎜ md N Eb ⎞
⎟ = 2nπ , (XI.240)
⎜ Eb ⎟ ⎜ Eb ⎟ ⎝ kF ⎠
⎝ ⎠ ⎝ ⎠
where n is an integer. For tunneling along the {110} direction of a dx2-y2-wave superconductor we have
φ− − φ+ = π and Δ + = −Δ − so that EQ. (XI.240) becomes
2
⎛ md N Eb ⎛ 1⎞ ⎞ Δ 0 − Eb2
tan ⎜ − ⎜ n − ⎟π ⎟ = , ( Δ+ = Δ− ≡ Δ0 ) . (XI.241)
⎝ kF ⎝ 2⎠ ⎠ Eb
Clearly EQ. (XI.241) is satisfied for Eb = 0 regardless of the thickness of dN, so that the zero-energy
states are always bound and give rise to the ZBCP. One may think of the ZBCP as the consequence of
constructive interferences between the bound electrons and holes at the nodal surface of a superconductor
with an unconventional pairing symmetry.
Fig. XI.3.6: Main panel: Schematic illustration for the formation of the surface bound states in an N-S
bilayer system: The quasiparticles follow closed paths through two Andreev reflections at the N/S interface
and two specular reflections at the surface. These bound states are stable even if the thickness of the N-layer
becomes zero. The right panel shows the pair potential for a dx2-y2-wave superconductor. [Adapted from Fig.
7(a) of Kashiwaya and Tanaka (2000)]
Fig. XI.3.7: Normalized conductance spectra taken on an optimally doped hole-type cuprate
superconducting YBa2Cu3O6.95 (YBCO) single crystal (with Tc = 93 K) at T = 4.2 K, using a cryogenic
temperature scanning tunneling microscope (STM). The data are given in open circles, whereas theoretical
fittings obtained by using the generalized BTK model and the listed fitting parameters are represented by the
solid lines. (a) Main panel: An STM tunnel junction (large Z) along the nodal {110} direction of the YBCO
sample, showing a pronounced ZBCP that agrees with a dx2-y2-wave superconductor with a maximum gap of
27 meV. Left inset: An STM point-contact junction (small Z), showing consistent results with a
predominantly dx2-y2-wave superconductor. Right inset: Mixed-symmetry simulations for the tunnel junction,
showing splitting of the ZBCP had there been an imaginary component in the pairing potential. (b) Main
panel: An STM point-contact junction (small Z) along the anti-nodal {100} direction, showing enhanced
conductance with the superconducting gap due to Andreev reflection. Left inset: An STM tunnel junction
(large Z) along {100} direction, showing a superposition of a gapped tunneling spectrum together with a
ZBCP, probably due to imperfect alignment of the quasiparticle momentum relative to the {100} surface.
Right inset: Mixed-symmetry simulations for the point-contact junction. Both sets of data along {110} and
{100} indicate predominant dx2-y2-wave pairing symmetry, although a small real s-wave component for (dx2-
y2 + s)-wave pairing cannot be ruled out. On the other hand, imaginary s-wave component for (dx2-y2 + is)-
wave appears unlikely based on comparison of the BTK simulations with experimental data. [Adapted from
Figs. 1 and 2 of Wei, Yeh, Garrigus and Strasik, Phys. Rev. Lett. 81, 2542 (1998).]
Generally speaking, the pairing potential of unconventional superconductors need not be of pure s-,
p-, d-, or f-wave symmetry. Rather, the pairing potential may consist of mixed symmetries, such as (dx2-y2 +
s), (dx2-y2 + is), (dx2-y2 + idxy), (px − ipy), etc. According to the generalized BTK theory, the directional
quasiparticle tunneling spectra for superconductors of different pairing symmetries usually exhibit significant
differences, so that the pairing symmetry of superconductors can be determined from the STS experiments.
Experimental verification of predominantly dx2-y2-wave superconductivity in some of the cuprate
superconductors [e.g. under- and optimally doped hole-type cuprates YBa2Cu3O7−δ (YBCO)] and mixed
symmetries in some other cuprate superconductors [e.g. (dx2-y2 + s)-wave pairing in overdoped hole-type
cuprates (Y0.7Ca0.3)Ba2Cu3O7−δ (Ca-YBCO)] have been made with the STS measurements as exemplified in
Figs. XI.3.7 and XI.3.8 together with the comparisons with the generalized BTK analysis.
Finally, in the end of Part XI.3, we briefly review the approach involving the transfer Hamiltonian
formalism and apply the formalism to N-I-N and N-I-S tunnel junctions.
Fig. XI.3.8: Spatially resolved (dINS/dV) vs. V tunneling spectra taken at T = 4.2 K on (a) optimally doped
YBCO (Tc = 93 K) along {110} direction; (b) underdoped YBCO (Tc = 60 K) along {100} direction; and (c)
overdoped Ca-YBCO (Tc = 79 K) along {001} direction. Right panels: Upper two panels compare
normalized experimental (dINS/dV) vs. V tunneling spectra with simulated results using the generalized BTK
theory. The tunneling spectra were taken on taken on underdoped (upper) and optimally doped (middle)
YBCO single crystals along the {100} direction. The lower panel illustrates BTK simulations for the {001}
tunneling spectra of superconductors of different pairing symmetries, which verifies that the tunneling
spectra of the overdoped YBCO system along {001} direction are consistent with the (dx2-y2 + s)-wave
pairing symmetry. In other word, the pairing potential appears to evolve from predominantly
Δ ( k ) ≈ Δ d cos ( 2θ k ) in the underdoped region into Δ ( k ) = Δ d cos ( 2θ k ) + Δ s in the overdoped region.
[Adapted from Figs. 1 and 2 of Yeh et al., Phys. Rev. Lett. 87, 087003 (2001).]
Here we adapt the transfer Hamiltonian formalism following the original work by Bardeen [J.
Bardeen, Phys. Rev. Lett. 6, 57 (1961)]. We consider a tunnel junction consists of two electrodes separated
by a tunnel barrier. Suppose that the left and right electrodes can be described nearly independently by the
Hamiltonians HL and HR in the ground state, and that they are weakly coupled by a perturbation
Hamiltonian HT so that the excited states correspond to particles being transferred from one electrode to the
other according to the tunneling Hamiltonian HT. We may express the uncoupled Hamiltonians for the left
and right electrodes as:
{ }
where cp and cq† are the particle operators, cp , cq† = 0 , ε p and ε q are the single-particle eigen-energies, μL
and μR are chemical potentials, and Nˆ L and Nˆ R are the number operators. In addition, the tunneling
Hamiltonian, also known as the transfer Hamiltonian, is given by
(
H T = ∑ Tp ,q cp†cq + cq†cp ,
p ,q
) (XI.244)
where Tp,q denotes the tunneling matrix element that represents the probability amplitude for transferring a
particle across the insulating barrier. Microscopically, Tp,q is determined by the quantum mechanical current
density operator evaluated within the barrier:
1 1
d x (ψ p ∇ ψ q − ψ q ∇ ψ p ) = − ds i (ψ p ∇ψ q − ψ q ∇ψ p ) ,
2m ∫ 2m ∫
* * * *
Tp ,q = − 3 2 2
(XI.245)
Given the tunneling Hamiltonian, the total tunneling current I under a biased voltage V across the
junction can be determined by the Fermi golden rule:
( ) ( ) ( )
2
I (V ) = 4π e∑ Tp ,q ⎡⎣ f ξp − f ξq ⎤⎦ δ ξp − ξ q + eV , (XI.246)
p ,q
provided that the non-interacting single-particle approximation is valid, and f ( ξ ) = [1 + exp ( − βξ )] denotes
−1
the Fermi function. The expression in EQ. (XI.246) can be generalized for interacting systems by including
(
many-body effects in terms of the spectral functions AL p, ωp and AR q, ωq of the two electrodes so that ) ( )
d ωp d ωq
( ) ( ) ( ) ( )( )
2
I (V ) = 4π e∑ Tp ,q ∫ ∫ ⎡ f ωp − f ωq ⎤⎦ AL p, ωp AR q, ωq δ ωp − ωq + eV . (XI.247)
p ,q 2π 2π ⎣
(
For non-interacting systems, AL p, ωp = 2πδ ωp − ξp ) ( ) ( ) ( )
and AR q, ωq = 2πδ ωq − ξq so that EQ. (XI.247)
reduces to EQ. (XI.246) as expected.
∂ξp ∂ξq
( ) δ ( pt − qt ) ,
2 2
Tp ,q ≈ D ξp , pt (XI.248)
∂p l ∂q l
( ) (
D ξp , pt ≡ exp ⎡⎣ − ∫ dx κ x, ξp , p t ⎤⎦ , ) (XI.249)
where the subscripts l and t refer to the longitudinal and transverse components of the momentum,
respectively, D is the transmission coefficient, κ is the wave-vector inside the tunneling barrier U(x,V), and
the tunneling barrier is a function of the biased voltage V. If the longitudinal quasiparticle energy is much
smaller than the barrier height, i.e., ξp − pt2 2m ( )
U ( x, V ) , the transmission coefficient D ξp , p t can be ( )
approximated by a constant. Thererfore, EQs. (XI.246) and (XI.248) yield the following result
( ) (
I (V ) = 4π e∑ ∫ d ξ p D( ξ p , p t ) ⎡⎣ f ξ p − f ξp + eV ⎤⎦ ∝ D ) ∫ dξ [ f (ξ ) − f (ξ + eV )] ,
2 2
(XI.251)
pt
which leads to ohmic behavior I (V ) ∝ D eV for the N-I-N junction in the low voltage limit.
2
In contrast to the N-I-N tunnel junctions that contain convoluted information of the bandstructures of
the two electrodes, the tunneling current across the normal metal-insulator-superconductor (N-I-S) junctions
in fact exhibit a relatively simple form:
∞
I (V ) ∝ ∫ d ξ D
−∞
2
[ f (ξ ) − f (ξ + eV )] N (ξ + eV ) ,
S
(XI.252)
where N S is the density of states in the superconducting electrode. Therefore, the low-temperature tunneling
conductance derived from EQ. (XI.252) can provide direct information for the density of states N S :
dI (V ) ∞ ⎡ ∂f ( ξ + eV ) ⎤
∫−∞ dξ ⎢− ⎥ N S ( ξ + eV ) ∝ D N S ( eV ) ,
2 2
∝ D (XI.253)
dV ⎣ ∂ ( eV ) ⎦
provided that D varies little with energy. The result in EQ. (XI.253) has been widely used to determine the
microscopic properties of superconductors by means of quasiparticle tunneling spectroscopy in N-I-S
junctions.
To justify the result given in EQ. (XI.252), we consider the total Hamiltonian H of an N-I-S
junction, H = HN + HS + HT, where
d ˆ
I = Iˆ = e N S = −ie ⎡⎣ Nˆ S , H T ⎤⎦ = −ie ∑ Tp ,q cp†,σ cq ,σ − Tp*,q cq†,σ cp ,σ , (XI.255)
dt p ,q ,σ
where the angle brackets represent the ensemble average. In the linear response limit we keep to the first
order in H T so that EQ. (XI.255) is approximated by
∞
I = e2 ∑ ∫−∞ dt ′θ ( t − t ′ ) ⎡⎣H T ( t ′ ) , Tp ,q cp†,σ ( t ) cq ,σ ( t ) − Tp*,q cq†,σ ( t ) cp ,σ ( t ) ⎤⎦ , (XI.256)
p ,q ,σ
where the particle operators cp†,σ ( t ) , cp ,σ ( t ) , cq†,σ ( t ) and cq ,σ ( t ) are defined in the interaction picture. If we
further rewrite the Bogoliubov quasiparticle operators given in EQ. (XI.194) by Josephson’s definitions for
quasiparticles and quasiholes and with the introduction of pair creation and annihilation operators S† and S,
we find that the quasiparticle and quasihole operators are given by
Consequently, the particle operators can be expressed in terms of the quasiparticle and quasihole operators:
For a bias voltage applied across the N-I-S junction, we have μ N − μ S = eV , and the unperturbed
Hamiltonian and operators in the interaction picture are given by:
(
H S = μ S Nˆ S + ∑ Eq γ e†,qσ γ e,qσ + γ h†,qσ γ h ,qσ ;
q ,σ
) (XI.261)
i (ξp + μ N ) t †
cp†σ ( t ) = e
iH N t † − iH N t
cpσ e =e cpσ , (XI.262)
− i (ξp + μ N ) t
cpσ ( t ) = e
iH N t − iH N t
cpσ e =e cpσ ; (XI.263)
i ( Eq + μS ) t † − i ( Eq − μS ) t
cq† ↑( t ) = uqγ e†,q↑( t ) + vqγ h ,q↓( t ) = uq e γ e,q↑ + vq e γ h ,q ↓ ; (XI.264)
− i ( Eq + μS ) t i ( Eq − μS ) t †
cq↑( t ) = uqγ e ,q↑( t ) + vqγ h†,q↓( t ) = uq e γ e,q↑ + vq e γ h ,q ↓ ; (XI.265)
i ( Eq + μS ) t † − i ( Eq − μS ) t
cq† ↓( t ) = uqγ e†,q↓( t ) − vqγ h ,q↑( t ) = uq e γ e,q↓ − vq e γ h ,q ↑ ; (XI.266)
− i ( Eq + μS ) t i ( Eq − μS ) t †
cq↓( t ) = uqγ e ,q↓( t ) − vqγ h†,q↑( t ) = uq e γ e,q↓ − vq e γ h ,q ↑ . (XI.267)
Inserting EQs. (XI.262) –(XI.267) into EQ. (XI.256) and using the following identities:
∞
∫ dt ′ θ ( t − t ′ ) ⎡⎣H T ( t ′ ) , Tp ,q cp†,σ ( t ) cq ,σ ( t ) − Tp*,q cq†,σ ( t ) cp ,σ ( t ) ⎤⎦
−∞
( )
vq2 ⎡⎣1 − f ξp ⎤⎦ ⎡⎣1 − f Eq ⎤⎦ ⎫
⎪ 2 ( )
+i ⎬ Tp ,q . (XI.268)
ξp + Eq + eV − iδ ⎪ ⎭
1 ⎛ 1 ⎞
=P ⎜ ⎟ ∓ iπδ ( x − x0 ) ,
x − x0 ± iδ ⎝ x − x0 ⎠
we find that the principal components in EQ. (XI.268) all cancel out, and we are left with the delta function
terms in the tunneling current:
{u ⎡⎣ f (ξ ) − f ( E )⎤⎦ δ (ξ ) ( ) ( ) ( )} . (XI.269)
2
I = 4π e∑ Tp ,q 2
q p p p − Ep + eV − vq2 ⎡⎣1 − f ξp − f Ep ⎤⎦ δ ξ p + Ep + eV
p ,q
We further note that for each state q+ with Eq + and uq + , there is another state q− with Eq − = Eq + and
2 2 2 2 2 2 2 2
ξq − = −ξq + so that uq + + uq − = uq + + vq + = 1 and vq + + vq − = vq + + uq + = 1 . Moreover, both q+
and q− are near the same point on the Fermi surface so that Tp ,q + ≈ Tp ,q − . Therefore, the coherence factors
uq2 and vq2 cancel out so that the tunneling current in EQ. (XI.269) becomes
{⎡⎣ f (ξ ) − f ( E )⎤⎦ δ (ξ ) ( ) ( ) ( )} .
2
I = 4π e∑ Tp ,q p p p − Ep + eV − ⎡⎣1 − f ξp − f Ep ⎤⎦ δ ξp + Ep + eV (XI.270)
p ,q
∂ξp ∂ξq
( ) δ ( pt − qt ) ,
2 2
Tp ,q ≈ D ξp , pt (XI.271)
∂p l ∂q l
∞ ∞ ⎛ ∂ξq ⎞
I (V ) ∝ D N S ( 0 ) ∫ d ξ p ∫ dEq ⎜
2
−∞ −∞ ⎜ ∂E ⎟⎟
⎝ q⎠
× {⎡⎣ f ( ξ p ) − f ( Ep ) ⎤⎦ δ ( ξ p − Ep + eV ) − ⎡⎣1 − f ( ξp ) − f ( Ep ) ⎤⎦ δ ( ξp + Ep + eV )} , (XI.272)
where N S ( 0 ) is the density of states of the superconducting electrode near the Fermi level. Equation
(XI.272) can be further simplified by noting that
⎛ ∂ξq ⎞ N S ( Eq )
⎜⎜ ⎟⎟ ≈ and ( ) (
1 − f −ξ p − eV = f ξp + eV . ) (XI.273)
⎝ ∂Eq ⎠ N S(0)
{
d ξp N S ( ξ p + eV ) ⎡⎣ f ( ξ p ) − f ( ξ p + eV ) ⎤⎦ − N S ( −ξ p − eV ) ⎡⎣1 − f ( ξ p ) − f ( −ξ p − eV ) ⎤⎦ }
∞
I (V ) ∝ D ∫
2
−∞
{N (ξ }
+ eV ) ⎡⎣ f (ξ p ) − f (ξ p + eV ) ⎤⎦ − N S ( ξ p + eV ) ⎡⎣ f ( ξ p + eV ) − f ( ξ p ) ⎤⎦
∞
∫
2
= D d ξp S p
−∞
d ξ {N S ( ξ + eV ) [ f (ξ ) − f ( ξ + eV )]} .
∞
∫
2
=2 D (XI.274)
−∞
Therefore, we obtain the same simple expression as EQ. (XI.252) for the tunneling current through an N-I-S
junction. Given EQ. (XI.274), we can deduce the density of states of a superconductor by studying the
differential tunneling conductance of the corresponding N-I-S junction. We also remark that the (rather non-
trivial) dropout of the coherence factors gives rise to a rather simple semiconductor-like description for the
tunneling currents across the N-I-S junctions. However, the “semiconductor model” is only valid in the
tunneling (or, equivalently, high-barrier) limit and cannot capture the concept of Andreev reflection for low
barrier junctions, and is therefore not suitable for studying the quasiparticle tunneling spectroscopy of
superconductors with unconventional pairing symmetries.
[Heavy-fermion superconductors]
In 1961, P. W. Anderson and P. Morel generalized the BCS theory for s-wave superconductors to
include the consideration of non-zero angular momentum and the physical consequences of such BCS
pairing [“Generalized Bardeen-Cooper-Schrieffer States and the proposed low-temperature phases of 3He”,
Phys. Rev. 123, 1911 (1961)]. Subsequently superfluidity was discovered in 3He in 1972, and many
experimental results shortly after this discovery have established 3He as p-waves spin-triplet superfluids that
exhibit three superfluid phases with broken time-reversal symmetry.
In general, we note that the pairing amplitude for a superconductor takes the form:
Δαβ ( k F ) ∼ ck F α c− k F β , (XI.275)
where α and β refer to the spin labels of the quasiparticles, and equal time pairing has been assumed. In
conventional superconductors with s-wave pairing symmetry, we find that
( )αβ ,
Δαβ ( k F ) ∼ Δ 0( k F ) iσ y (XI.276)
where Δ 0( k F ) is a complex amplitude that breaks the U(1) gauge symmetry. On the other hand,
unconventional pairing occurs when the pairing amplitude spontaneously breaks one or more symmetries of
the normal state besides the U(1) gauge symmetry. That is, for a symmetry operation R ∈ G where the full
symmetry group G of the system is given by
T represents the time-reversal symmetry, and Gspin and Gspace respectively denote the symmetry groups of
the spin and orbit wavefunctions of the system, we find that unconventional pairing occurs if
( )
R ⊗ Δ R −1 ⊗ k F ≠ Δ ( k F ) . Moreover, the fermion statistics of quasiparticles in superconductors requires
the pairing amplitude to obey the antisymmetric condition so that Δαβ ( k F ) = −Δ βα ( −k F ) .
In the case of negligible spin-orbit interaction, the normal state system is invariant under rotation so
that Gspin = SO(3), and the α and β labels refer to the spin indices of the quasiparticle states near the Fermi
level. On the other hand, for systems with strong spin-orbit interaction, such as in the case of heavy-fermion
superconductors, the α and β labels in EQ. (XI.275) are no longer the same as the eigenvalues of the spin
operators for electrons. However, in this case the Kramers degeneracy in the absence of magnetic field still
guarantees that each k-state is two-fold degenerate. Consequently, for strong spin-orbit interaction we may
consider the α and β labels as pseudo-spin quantum numbers that take on two possible values.
To date, all superconductors exhibiting unconventional pairing have inversion symmetry in their
crystalline structures. Consequently, the pairing interaction that drives the superconducting transition
necessarily decomposes into even- and odd-parity sectors, and Δαβ ( k F ) for any of these superconductors
with a single primary order parameter can therefore be expressed as follows:
( )αβ + Δ ( k ) i ( iσ σ )αβ ,
Δαβ ( k F ) = Δ 0( k F ) iσ y F y (XI.278)
where σ denote the Pauli matrices, and Δ 0( k F ) = Δ 0( −k F ) for even parity and Δ ( k F ) = − Δ ( −k F ) for odd
parity. Furthermore, the general form of the order parameter can be expressed by
Δ 0( k F ) = ∑ Γ ∑ i ηi(Γ) YΓ,i ( k ) ,
even dΓ
F
(singlet, S = 0) (XI.279)
Δ ( k F ) = ∑ Γ ∑ i ηi( ) YΓ , i ( k ) ,
odd d Γ
Γ
F
(triplet, S = 1) (XI.280)
where Γ denotes an irreducible representation of the point group, dΓ is the dimension of a given irreducible
representation Γ, ηi( ) is the coefficient associated with the basis function YΓ,i (kF) of the ith partner function
Γ
belonging to an even-parity representation Γ, and YΓ,i (kF) denotes the basis function of the ith partner
function belonging to an odd-parity representation Γ. Based on EQ. (XI.280) and the experimental evidence
Among the best studied heavy-fermion superconductors, the UPt3 compound has been one of the
model systems that clearly reveal multiple superconducting phases and has been under intense theoretical
investigation. In the following we describe the primary experimental phenomena and some theoretical
development of the UPt3 compound as an example of triplet heavy-fermion superconductors.
The crystalline structure of UPt3 belongs to the point group D6h, and its physical properties are
generally consistent with the characteristics of a heavy-fermion system below a temperature T* ~ 10 K.
Specifically, the resistivity ρ, specific heat Cv, and magnetic susceptibility χ of UPt3 satisfy the following
temperature dependence:
ρ = ρ 0 + AT 2 , ρ 0 ≈ 0.1 μ Ω − cm ,
Cv = γ T , (γ γ 0 ) = ( m* me ) ≈ 500 ! (Thus the name “heavy-fermion”)
(χ χ 0 ) ≈ constant ~ ( m* me ) .
Superconductivity in UPt3 was first discovered in 1984 when a resistive transition was observed at Tc = 526
mK with a very sharp transition width of δTc ~ 1 mK. The existence of two superconducting phases at H = 0
was confirmed by specific heat measurements, which revealed two distinct transitions Tc and Tc2 separated by
~ 50 mK, as illustrated in Fig. XI.4.1(a). Additionally, ultrasonic attenuation in magnetic fields at low
temperatures also exhibited two distinct characteristic fields H* and Hc2, as schematically shown in Fig.
XI.4.1(b). By collecting information from experiments under varying H and T, an interesting H vs. T phase
diagram emerges, revealing three different flux phases that meet at a tetra-critical point for all orientations of
the magnetic field relative to the sample c-axis (which is normal to the hexagonal two-dimensional lattice),
as shown in Fig. XI.4.2.
Fig. XI.4.1 (a) Schematic of the zero-field specific heat (Cv) versus temperature (T) data in UPt3. The
pronounced steps with discontinuities ΔC1 and ΔC2 occur at two phase transition temperatures
at Tc and Tc2, where (Tc − Tc2) / Tc ~ 10% and (ΔC2 / Tc2) / (ΔC1 / Tc) ~ 1.25. (b) Schematic of the
relative ultrasound attenuation (αS / αN) versus magnetic field (H) data taken at T = 50 mK,
where the subscripts S and N refer to superconductivity and normal state, respectively. The
presence of two characteristic fields H* and Hc2 indicates three distinct flux phases in UPt3.
Δ 0( k F ) = η1Y1 ( k F ) + η2Y2 ( k F )
H
η ≡ (η1 ,η2 )
Fig. XI.4.2 Schematic of the magnetic field (H) versus temperature (T) phase diagram of UPt3.
Among the E-representation models, the most successful ones in explaining the superconducting
properties of UPt3 are based on E1g and E2u:
In the absence of spin-orbit coupling, the dimensionality of the odd-parity E-representation is three times
larger than that of the corresponding spin-singlet E-representation. Therefore, the order parameter transforms
according to the following:
(
Δ 0( k F ) → Rspin Δ 0 R orbit
−1
)
kF , where R orbit ∈ [ D6 h ]orbit and Rspin ∈ SU (2)spin . (XI.281)
Thus, the enlarged symmetry group for the normal state without spin-orbit coupling is:
In reality, the presence of strong spin-orbit coupling in the uranium-based heavy-fermion metals reduces the
symmetry group by allowing only joint rotations of the spin and orbit degrees of freedom. Therefore, the
(
Δ ( k F ) → R Δ R −1 k F , ) where R ∈ [ D6 h ]spin-orbit . (XI.283)
Moreover, the full symmetry group in the normal state in the presence of strong spin-orbit coupling becomes
In the 1990’s a series of empirical facts have led to strong support for E2u-pairing symmetry in UPt3.
Hence, the system is a spin-triplet superconductor, which requires consideration of the relation between the
orbital magnetic moment and the spin-degree of freedom of the pairs. There are two commonly discussed
special cases. In the first case, we consider the (pseudo) spin-triplet order parameter being factorized into a
single (pseudo) spin-vector and an orbital component, so that
Δ ( k F ) = d Δ 0( k F ) , (XI.285)
where d is a real unit vector, and the notation Δ 0( k F ) is an odd-parity orbital function. We note that
generally the vector d defines the axis along which the pairs have zero spin projection. In other words,
d i S ( k F ) = 0 . Hence, for d ẑ (with ẑ being a unit vector in spin space), from EQs. (XI.278) and (XI.285)
we find that Δ ↑↑ = Δ ↓↓ = 0 and Δ ↑↓ = Δ ↓↑ = Δ ( k F ) . On the other hand, if d ⊥ ẑ , then the same pairing
state, now described in terms of a different choice of the quantization axis, is given by Δαα = Δ ββ = Δ ( k F )
where α ⊥ ẑ and β ⊥ ẑ and Δαβ = Δ βα = 0 . That is, the pairing state is described as “equal-spin” pairing in
an “easy-plane”. In the second special case, one considers a complex unit vector d, so that the spin
components of the order parameter spontaneously break the time-reversal symmetry from the condition
d i S (k F ) = 0 .
More generally, the order parameter Δ is complex so that Δ × Δ * ≠ 0 , and it varies over the Fermi
surface. These states are non-unitary, because the square of the spin-matrix representation of the order
parameter is not proportional to the unit spin matrix. That is,
2
⎡⎣ Δ † Δ ⎤⎦αβ = Δ δαβ + i ⎡⎣ Δ × Δ ∗ i σ ⎤⎦αβ . (XI.286)
Consequently, the spin degeneracy of the excitation spectrum is lifted, and the quasiparticle energy depends
on the local pair spin at kF, where S pair ( k F ) ~ i Δ ( k F ) × Δ ( k F ) .
∗
In fact, paramagnetism can serve as an important probe for the spin structure of the superconducting
order parameter, particularly as an experimental signature to differentiate even- and odd-parity
superconductivity. In the case of odd-parity spin-triplet superconductors, the transition temperature, energy
gap, and many other superconducting properties can depend strongly on the orientation of the magnetic field
relative to the spin-quantization axis of the Cooper pairs. Let’s consider the following two examples. First, if
the Cooper pairs form spin-singlets, then the Zeeman energy that favors an unequal spin population has pair-
breaking effect for all field directions. Second, if Cooper pairs form spin-triplets, we may consider the
special case with a real d-vector. If d ẑ , then the non-vanishing order parameter for the pair ( k F , −k F ) can
be a pure “opposite-spin state” with ↑↓ + ↓↑ . In this case, an external magnetic field H ẑ is pair-
breaking for all kF, whereas H ⊥ ẑ is not pair-breaking because Δ ↑↑ = Δ ↓↓ = 0 already. Similarly, if d ⊥ ẑ ,
then a magnetic field along an “equal-spin-pairing” (ESP) direction can easily polarize the pairs and
minimize the Zeeman energy by alternating the relative number of ↑↑ and ↓↓ pairs with essentially no
loss in the condensation energy. Therefore, in general a magnetic field with H ⊥ d( k F ) for all kF is not pair-
breaking. However, if H d( k F ) , then the field is pair-breaking for the pairs ( k F , −k F ) , similar to the case
of spin-singlet pairing.
A1 g 1 1
⎣( )
Im ⎡ k x + ik y ⎤
6
A2 g ⎦ 1
k z Im ⎡( k x + ik y ) ⎤
3
B1 g 1
⎣ ⎦
k z Re ⎡( k x + ik y ) ⎤
3
B2 g 1
⎣ ⎦
⎛ kx ⎞
E1 g kz ⎜ ⎟ 2
⎝ ky ⎠
E2 g ⎛ k x2 − k y2 ⎞ 2
⎜ ⎟
⎝ 2k x k y ⎠
A1 u zˆ k z 1
⎣( ) ⎤⎦
zˆ k z Im ⎡ k x + ik y
6
A2 u 1
zˆ Im ⎡( k x + ik y ) ⎤
3
B1 u ⎣ ⎦ 1
zˆ Re ⎡( k x + ik y ) ⎤
3
B2 u ⎣ ⎦ 1
⎛ x⎞
k
E1 u zˆ ⎜ ⎟ 2
⎝ ky ⎠
⎛ k x2 − k y2 ⎞
E2 u zˆ k z ⎜ ⎟ 2
⎝ 2k x k y ⎠
In the case of UPt3, the Hc2 measurements reveal paramagnetic effect for H ĉ and no paramagnetic
effect for H ⊥ ĉ . Knowing that the order parameter is odd-parity spin-triplet state, we can conclude that
d ĉ in UPt3. That is, the orbital magnetic moment points along the crystalline c-axis, and the spins of the
Cooper pairs lie in the basal plane. For the E2u representation, we may express Δ ( k F ) as
( )
Δ ( k F ) = zˆ ⎡⎣ k x2 − k y2 η1 + 2 k x k y η 2 ⎤⎦ k Z . ( ) (XI.287)
It is also worth noting that significant differences exist in odd-parity superconductors, depending on
whether there is weak or strong spin-orbit coupling in the pairing channel. For an ESP state, in the absence of
spin-orbit coupling, the d-vector will orient itself perpendicular to the magnetic field in order to minimize the
Zeeman energy. Consequently, the measured spin susceptibility in the low-field limit will not show
dependence on the field-orientation for T < Tc. On the other hand, if there is crystalline anisotropy and strong
spin-orbit coupling, then a rotation of d under an applied field will cost energy on the order of Tc. Thus, spin-
orbit coupling tends to select preferred directions for d in the crystal, so that the orientation of an applied
magnetic field can be used to probe the spin structure of the order parameter directly.
Finally, we discuss the Landau-Ginzburg theory for triplet superconductors by considering the
special case of the E2u-pairing in UPt3. It should be noted, however, that similar analysis described below can
be done for E1g, E1u and E2g representations.
We already know that the pairing state in UPt3 is odd-parity triplet ESP state with strong spin-orbit
coupling, so that d ĉ and
Δαβ ( k F ) = Δ 0 ( k F ) ⎡⎣d i iσ σ y ⎤⎦ ( ) αβ . (XI.288)
⎛ 0 Δ 0( k F ) ⎞
Δ (k F ) = ⎜ ⎟, (XI.289)
⎝ Δ 0( k F ) 0 ⎠
where the two-dimensional orbital state is given by
In the case of strong spin-orbit coupling, the terms in the Landau-Ginzburg functional F must be
invariant under the symmetry group G = [ D6 h ]spin-orbit ⊗ T ⊗ U (1) . The form of F is governed by the
linearly independent invariants that are constructed from fourth-order products of the form
∑ bi j k l ηi η j ηk*ηl* (XI.291)
i , j , k ,l
∑ κ i j k l ( Diη j )( Dkηl )
*
, where Di = ∂ i + 2ieAi . (XI.292)
i , j , k ,l
Thus, we may express F in the following general form with (i,j) = (1,2):
{
F [ η, A ] = ∫ d 3r α ( T ) ηiη∗ + β1 ηi η∗ ( )
2
+ β 2 ηiη +
2
8π
1
b
2
( )( D η ) + κ 2 ( Diηi ) D jη j ( ) ( )( D η ) ( )( D η ) } ,
* * * *
+ κ1 Diη j i j + κ 3 Diη j j i + κ 4 Dzη j z j (XI.293)
δ F [ η, A ] δ F [ η, A ]
=0 and =0. (XI.294)
δη *
i δ Ai
Thus, from EQs. (XI.293) and (XI.294), we obtain the following two conditions:
form the basis for studying the H-vs.-T phase diagram, vortices and related magnetic properties of triplet
superconductors. Specifically, there are two possible homogeneous states that are dependent on the sign of
β2. For ( − β1 ) < β 2 < 0 , the equilibrium order parameter η = η0 x̂ (or any of the six degenerate states
obtained by rotation) breaks rotational symmetry in the basal plane, but still preserves the time-reversal
symmetry. For β 2 > 0 , the order parameter retains full rotational symmetry (provided that each rotation is
combined with a proper gauge transformation), but spontaneously breaks the time-reversal symmetry. This
case corresponds to a doubly-degenerate equilibrium state, with an order parameter of the form
1 1 α
η+ = η0 ( x + iy ) ≡ ( x + iy ) [or η− = η*+ ]. (XI.298)
2 2 β1
To derive better understanding of the phases associated with different parameters in the Landau-
Ginzburg functional F, we consider the simple case of zero-field solutions for a homogeneous system, so
that all gradient terms can be neglected. In this case, we find that EQ. (XI.293) becomes:
{
F [ η] = ∫ d 3r α ( T ) ηi η∗ + β1 ηi η∗ ( )
2
+ β 2 ηi η
2
}. (XI.299)
We may express η in terms of a global phase θ (r) and an internal, relative phase ζ (r), so that
iζ / 2
⎛ η1 ⎞ iθ ⎛ η1 e ⎞
η=⎜ ⎟=e ⎜ −iζ / 2 ⎟
. (XI.300)
⎝η2 ⎠ ⎝ η2 e ⎠
The global phase θ (r) is associated with broken U(1) symmetry and pS ~ ∇θ , whereas ζ (r) is associated
with the broken time-reversal symmetry. Inserting these two phases into EQ. (XI.299), we rewrite F into
{
F [ η] = ∫ d 3r α ( T ) ηiη∗ + β1 ηi η∗( )
2
+ β 2 ηiη
2
}. (XI.301)
Minimizing the free energy, we find two sets of stable solutions (see Fig.XI.4.3), depending on the sign of
β2, and we note that β1 is always > 0.
which implies that there is broken rotational symmetry from G = [ D6 h ]spin-orbit ⊗ T ⊗ U (1) in the normal
state to G ′ = [ D2 h ]spin-orbit ⊗ T in the superconducting state.
β2 Stable solutions in
Regions I & II at
II T < Tc (α < 0)
(Unstable)
β1
I
β1 + β 2 = 0
Fig. XI.4.3 Schematic of the stable solutions for triplet-pairing superconductivity in the (β1,β2)-plane.
1 ⎛1⎞ 1 ⎛1⎞
η+ = η 0 ⎜ ⎟ and η− = η 0 ⎜ ⎟, (XI.303)
2 ⎝i⎠ 2 ⎝ −i ⎠
which implies that there is broken time-reversal symmetry from G = [ D6 h ]spin-orbit ⊗ T ⊗ U (1) in the normal
state to G ′ = [ D6 h ]spin-orbit in the superconducting state. Therefore, in this phase the ab-plane isotropy is
preserved, and the order parameter exhibits two-fold degeneracy:
1
Δ + ( k F ) = η0 ⎡⎣( k x2 − k y2 ) + i 2 ( k x k y ) ⎤⎦ k Z , (XI.304)
2
1
Δ −( k F ) = η0 ⎡( k x2 − k y2 ) − i 2 ( k x k y ) ⎦⎤ kZ ,
⎣ (XI.305)
2
M orb = 2e ( κ 2 − κ 3 ) Im ⎡⎣ η × η∗ ⎤⎦ , (XI.306)
Δ+ (d ẑ )
Δ− (d − ẑ )
Fig. XI.4.4 Degenerate order parameters Δ+ and Δ− for the stable solutions in Case XI.
In principle, one can extend the above formalism to more complicated situations with finite magnetic
fields and/or order parameter gradients. We shall not pursue further here. For more details on specific
properties and phases associated with triplet superconductors, you may refer to some of the references listed
below. However, before closing it is worth noting an interesting development associated with a heavy-
fermion superconductor Sr2RuO4. Although not yet rigorously comfirmed, this superconductor is believed to
have a novel pairing symmetry of (px + ipy) based on a number of theoretical and indirect experimental
evidences. This specific pairing symmetry, if proven existent, can acquire novel “zero-mode” low-energy
excitations in the vortex-core states. Specifically, if we consider 2n well separated vortices in a (px + ipy)-
pairing superconductor, there will be n zero modes for “Majorana fermions”, so that there are a total of 2n
degenerate states for even fermion number and 2n−1 degenerate states for odd fermion number. Such
degeneracy is crucial for the occurrence of non-abelian statistics in topological quantum computation.
Further Readings:
Conventional superconductivity:
1. J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 108, 1175 (1957).
2. D. C. Mattis and J. Bardeen, Phys. Rev. 111, 412 (1958).
3. L. P. Gorkov, Sov. Phys. JETP 7, 505 (1958).
4. A. A. Abrikosov, Soviet Phys. JETP 5, 1174 (1957).
5. A. L. Fetter and J. D. Walecka, “Quantum Theory of Many-Particle Systems”, Chapters 10 and 13.
6. Abrikosov, Gorkov, and Dzyaloshinski, “Methods of Quantum Field Theory in Statistical Physics”,
Chapter 5.
7. P. de Gennes, “Superconductivity of Metals and Alloys”, Perseus Books (1999).
8. M. Tinkham, “Introduction to Superconductivity”, Dover Publications, Inc.
9. J. R. Schrieffer, “Theory of Superconductivity”, Benjamin, New York (1964).
History of superconductivity:
30. J. Matricon and G. Waysand, “The Cold Wars: A History of Superconductivity”, translated from the
French by C. Glashausser, Rutgers University press (2003). Also see the corresponding book review by
N.-C. Yeh, Physics in Perspectives 7, 259 – 261 (2006).