2023 212C Lectures
2023 212C Lectures
Spring 2023
Lecturer: McGreevy
These lecture notes live here. Please email corrections to mcgreevy at physics dot
ucsd dot edu.
1
Contents
0.1 Introductory remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.2 Possible topics (not in order) . . . . . . . . . . . . . . . . . . . . . . . . 4
0.3 Sources and acknowledgement . . . . . . . . . . . . . . . . . . . . . . . 5
0.4 Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Interference phenomena 70
2.1 Path integrals, briefly . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.2 Wave mechanics of charged particles . . . . . . . . . . . . . . . . . . . 77
2.3 Integer Quantum Hall Effect . . . . . . . . . . . . . . . . . . . . . . . . 85
3 A simple magnet 92
2
7.1 Self-consistent mean-field theories . . . . . . . . . . . . . . . . . . . . . 163
7.2 Problems involving more than one nucleus . . . . . . . . . . . . . . . . 170
3
0.1 Introductory remarks
By now you know some quantum mechanics (QM), I hope. But it is a vast subject
with many applications. There are many directions in which we can generalize from
your previous encounters with QM.
One direction is toward situations where things are faster. This leads to “relativistic
quantum mechanics,” which inevitably means quantum field theory. That’s the subject
of Physics 215 and so I will try to stay away from it to the extent possible.
In this course, we will approach a different frontier: situations where things are
many. That is: what happens when we put together a whole bunch of quantum de-
grees of freedom. (As we will see, sometimes this also means quantum field theory!)
Sometimes ‘a whole bunch’ will mean just a big finite number, like the number of elec-
trons in an atom of large atomic number, or a very big finite number, like the number
of electrons in a chunk of solid. Sometimes, we’ll be interested in the thermodynamic
limit, where the number of degrees of freedom becomes parametrically large. This is
because new (emergent) phenomena happen in this limit.
There is some flexibility about what subjects we’ll discuss this quarter. Please tell
me about topics of interest to you.
Physics phenomena:
Phonons
Radiation (photons, Casimir effect)
Atoms
Molecules
Matter and radiation (photoelectric effect, scattering, lasers)
More interesting states of matter (superfluids and superconductors, magnetic insu-
lators, quantum Hall phases)
Relativistic quantum mechanics
Techniques:
How to describe an indefinite number of particles (‘second quantization’).
Hartree-Fock, Thomas-Fermi.
Born-Oppenheimer, Berry phases.
4
Mean field theory of various states of matter
Coherent state path integrals?
Fermi liquid theory?
We’ll learn about topics from these list in parallel.
• Pines and Nozieres, The Theory of Quantum Liquids (two volumes), Perseus
Advanced Book Program, 1966.
Some sources I recommend for some of the many topics that we could have studied
but did not are:
5
0.4 Conventions
The convention that repeated indices are summed is always in effect unless otherwise
indicated.
d is the number of space dimensions, D is the number of spacetime dimensions (it’s
bigger!).
A consequence of the fact that english and math are written from left to right is
that time goes to the left.
h dk
A useful generalization of the shorthand ~ ≡ 2π is d̄k ≡ 2π . I will also write
d d (d)
/δ (q) ≡ (2π) δ (q). I will try to be consistent about writing Fourier transforms as
dd k ik·x ˜
Z Z
e f (k) ≡ d̄d k eik·x f˜(k) ≡ f (x).
(2π)d
6
1 An indefinite number of identical particles
So far in Physics 212, you’ve mainly been discussing the quantum theory of particles,
where the number of particles is fixed (and quite small). There are processes in nature
where the number of particles can change – for example photons are created during
atomic transitions. How do we describe this kind of thing? The answer is sometimes
called quantum field theory or sometimes ‘second quantization’ (a terrible name), and
it is a crucial part of modern physics. We will discover it in a perhaps-unexpected
place, momentarily (§1.3).
This discussion also follows logically in that one’s study of QM begins (or at least
should begin) by discussing a single qubit, and then learning to make composite quan-
tum systems, which basically means two qbits. Here we will take the next step of
studying composite quantum systems with infinitely many components.
Another side-benefit of being able to describe an indefinite number of particles
is that we can also describe superpositions between states with different numbers of
particles. This is required to describe a superfluid.
[Le Bellac 11.1] The simple harmonic oscillator is ubiquitous in physics, not just because
it is exactly solvable both classically and quantum mechanically, but because it arises
as the leading approximation to any system near a stable equilibrium. Expanding a
potential V (x) in one dimension about its minimum at x0 , Taylor’s theorem says
1
V (x) = V (x0 ) + 0 + V 00 (x0 )(x − x0 )2 + ...
2
and we can often ignore the ... (aka anharmonic terms) for systems that are near their
equilibrium configuration. WLOG setting x0 ≡ 0, we are led to study
p2
1 2 2 ~ω 2 2
† 1
H= + mω x = P + Q = ~ω a a +
2m 2 2 2
with
1 1
a ≡ √ (Q + iP) , a† ≡ √ (Q − iP) .
2 2
Here I’ve defined these new operators to hide the annoying constant factors:
mω 1/2 1/2
1
Q≡ x, P ≡ p.
~ m~ω
7
The number operator N ≡ a† a is hermitian and satisfies
So a and a† are, respectively, lowering and raising operators for the number operator.
The eigenvalues of the number operator have to be positive, since
which means that for n = 0 we have a |n = 0i = 0. If it isn’t zero (i.e. if n ≥ 1), a |ni
is also an eigenvector of N with eigenvalue n − 1. It has to stop somewhere! So the
eigenstates of N (and hence of H = ~ω N + 21 ) are
n
|0i , |1i ≡ a† |0i , ..., |ni = cn a† |0i ...
where we must choose cn to normalize these states. The answer that gives hn|ni = 1
is cn = √1n! .
As a warmup, consider the following two (quantum) mechanical systems, each consist-
ing of two point masses.
In case A, we have two masses connected by springs and attached to the walls of the
box. In case B, the two masses are only connected to each other. In each case, we can
write the Hamiltonian as
p21 p22
H= + + V (q) (1.1)
2m 2m
where V (q) is a quadratic function in q1 , q2 . In case A:
1 2 1 2 1 2 1 2 −1 q1
VA (q) = κ(q2 − q1 ) + κq1 + κq2 = κ(q1 , q2 ) (1.2)
2 2 2 2 −1 2 q2
and in case B,
2 1 −1 q1
VB (q) = κ(q2 − q1 ) = κ(q1 , q2 ) . (1.3)
−1 1 q2
8
Recall from mechanics that to solve either of these problems, we can decompose the
motion into normal modes. In this case, the normal modes are just q± = √12 (q1 ± q2 ),
which describe the two particles moving together, or the two particles
moving op-
s1
positely. More mechanically, the eigenvectors of the matrix = s1 2 − σ x are
1s
1 1
√ , with respective eigenvalues s ∓ 1. So
2 ±1
q+ 1 1 1 q1 q1 1 1 1 q+
=√ ⇔ =√ . (1.4)
q− 2 1 −1 q2 q2 2 1 −1 q−
[Le Bellac section 11.3] Let’s think about a crystalline solid. The specific heat of solids
(how much do you have to heat it up to change its temperature by a given amount)
was a mystery before QM. The first decent (QM) model was due to Einstein, where
he supposed that the position of each atom is a (independent) quantum harmonic
oscillator with frequency ω. This correctly predicts that the specific heat decreases as
the temperature is lowered, but is very crude. Obviously the atoms interact: that’s
why they make a nice crystal pattern, and that’s why there are sound waves, as we
will see. By treating the elasticity of the solid quantum mechanically, we are going to
9
discover quantum field theory. One immediate benefit of this will be a framework for
quantum mechanics where particles can be created and annihilated.
As a toy model of a one-dimensional crystalline solid, let’s consider a chain of point
masses m, each connected to its neighbors by springs with spring constant κ. When
in equilibrium, the masses form a regular one-dimensional crystal lattice of equally-
spaced mass points. Now let qn denote the displacement of the nth mass from its
equilibrium position xn and let pn be the corresponding momentum. Assume there
are N masses and impose periodic boundary conditions: qn+N = qn . The equilibrium
positions themselves are
xn = na, n = 1, 2...N
where a is the lattice spacing. The Hamiltonian for the collection of masses is:
N 2
X p n1 2
H= + κ (qn − qn−1 ) + λq3 . (1.6)
n=1
2m 2
I’ve include a token anharmonic term λq3 to remind us that we are leaving stuff out;
for example we might worry whether we could use this model to describe melting. Now
set λ = 0 because we are going to study small deviations from q = 0.
This hamiltonian above describes a collection of coupled oscillators, with a matrix
of spring constants V = kab qa qb . If we diagonalize the matrix of spring constants, we
will have a description in terms of decoupled oscillators, called normal modes.
Notice that the hamiltonian commutes with the shift operation
10
Hence these are eigenvectors of kab as well.
A nice way to summarize this statement is the following slogan: Since our system
has (discrete) translation invariance, the normal modes are labelled by a wavenumber
k1:
N N
1 X ikxn 1 X ikxn
qk = √ e qn , pk = √ e pn ,
N n=1 N n=1
(Notice that in the previous expressions I didn’t use boldface; that’s because this step
is really just classical physics. Note the awkward (but here, inevitable) fact that we’ll
have (field) momentum operators pk labelled by a wavenumber aka momentum.)
2π(N − 1)
0≤k≤ .
Na
Because of the periodicity in k, we can equivalently label the set of wavenumbers by:
2π π π
0<k≤ or − <k≤ .
a a a
This range of independent values of the wavenumber in a lattice model is called the
Brillouin zone. There is some convention for choosing a fundamental domain which
prefers the last one but I haven’t found a reason to care about this.
Summary: Because the system is in a box (periodic), k-space is discrete. Because
the system is on a lattice, k-space is periodic. There are N oscillator modes altogether.
P
When I write k below, it is a sum over these N values.
1
The inverse transformation is:
2π/a 2π/a
1 X −ikxn 1 X −ikxn
qn = √ e qk , pn = √ e pk .
N k>0 N k>0
11
So the whole hamiltonian is a bunch of decoupled oscillators,
labelled by these funny wave numbers:
X pk p−k 1
2
H= + mωk qk q−k (1.7)
k
2m 2
This is called the dispersion relation – it says how fast a mode of a given wavenum-
ber propagates; its deviation from linearity in k says how quickly a wave packet will
disperse.
To verify (1.7), let’s look more carefully at the expression for the normal modes:
N
1 X ikxn X 2π
qk = √ e qn ≡ Ukn qn , k = j , j = 1 . . . N.
N n=1 n
N a
Notice that this T is an operator acting on the labels of the quantum states – it’s
a classical operator. So don’t be surprised that we can also Fourier transform the
12
momenta : X
pk = Ukn pn .
n
So far we’ve actually solved a classical problem of finding normal modes of these
coupled oscillators. The world is quantum mechanical so let’s remember that our
variables are quantum operators now, and figure out what the quantum Hamiltonian
is for the normal modes. The kinetic term in the Hamiltonian is
X XX X X
p2n = U−kn U−k0 n pk pk0 = δk,−k0 pk pk0 = pk p−k .
n n k,k0 k,k0 k
Sound waves. Where does this frequency ωk matter physically? For one thing,
consider the Heisenberg equation of motion for the deviation of one spring:
pn
i∂t qn = [qn , H] = , i∂t pn = [pn , H]
m
Combining these gives:
(You will recognize this as just Newton’s equation which would be valid classically.)
In terms of the fourier-mode operators, these equations decouple:
13
Plugging in a fourier ansatz in time as well qk (t) = ω e−iωt qk,ω turns this into an
P
2 |k|a
algebraic equation which says ω 2 = ωk2 = 2κ m
sin 2 for the allowed modes (we
used a trigonometric identity here). We see that (the classical version of) this system
describes waves:
k1/a
0 = ω 2 − ωk2 qk,ω ' ω 2 − vs2 k 2 qk,ω .
The result for small k is the fourier transform of the wave equation:
∂t2 − vs2 ∂x2 q(x, t) = 0 .
(1.9)
vs is the speed of propagation of the waves, in this case the speed of sound. Comparing
to the dispersion relation (1.8), we have found
r
∂ωk κ
vs = |k→0 = a .
∂k m
The description we are about to give is a description of quantum sound waves. (Below
we will also give a description of quantum light waves.) [End of Lecture 1]
Here is a slight generalization, by way of recap. Consider the more general class of
Hamiltonians
X p2 1X
n
H= + κnm qn qm .
n
2m n 2 nm
We allowed the masses to vary, and we made a whole matrix of spring constants. Notice
that only the symmetric part 12 (κnm + κmn ) of this matrix appears in the Hamiltonian.
Also κ must be real so that H is hermitian. To simplify our lives we can redefine
variables
√ √ κnm
Qn ≡ mn qn , Pn = pn / mn , Vnm ≡ √
mn mm
in terms of which !
1 X X
H= Pn2 + Vnm Qn Qm .
2 n nm
where I assumed that all the eigenvalues of V are non-negative – otherwise the system
is unstable. U is unitary (actually it could be taken to be an orthogonal matrix):
X X
Uαn U † nβ = δαβ , U † nβ Uβm = δnm .
(1.11)
n α
14
To take advantage of this, we make the change of variables to the normal modes
X
Q̃α = Uαn Qn .
n
P
Similarly, we define P̃α = n Uαn Pn .
Now let’s look at what this does to the terms in H:
X X XX X
Pn2 = Pn† Pm = Uαn U † nβ Pα† Pβ = Pα† Pα .
n n αβ
|n {z } α
(1.11)
= δαβ
X X XX X
Vnm Q†n Qm = Uαn Vnm U † Q†α Qβ = ωα2 Q†α Qα .
Vnm Qn Qm = mβ
nm nm αβ
|nm {z } α
(1.10)
2
= δαβ ωα
1. First, it is local in the sense that only nearby springs couple to each other, so κnm
is only nonzero when n and m are close together.
3. Finally, it has a certain additional symmetry that also has a right to be called
translation invariance from the microscopic point of view. This is the sym-
metry under qn → qn + , which shifts each mass by the same amount. This
doesn’t stretch the springs at all, and so is a symmetry of H. More precisely, this
15
symmetry arises because the potential depends on qn only through differences,
V = V (qn − qn−1 ).
Notice that the transformation on the field qn is non-linear, in the sense that
the field shifts under the symmetry (as opposed to a linear transformation like
q → Rq). This is related to the fact that the transformation takes one minimum-
energy configuration to another; in this sense, the symmetry is said to be spon-
taneously broken. An important general consequence of a non-linearly realized
continuous symmetry is the presence of a massless mode, a Goldstone mode.
We’ve already seen that there is such a massless mode in this system. This is an
example of Goldstone’s theorem. It’s a bit mysterious in this example, but will
become clearer later.
Notice that when k = 0, ωk = 0. We are going to have to treat this mode specially;
there is a lot of physics in it.
QM. So far the fact that quantumly [qn , pn0 ] = i~δnn0 1 hasn’t mattered in our
analysis (go back and check). For the Fourier modes, this implies the commutator
X X
[qk , pk0 ] = Ukn Uk0 n0 [qn , p0n ] = i~1 Ukn Uk0 n = i~δk,−k0 1. (1.12)
n,n0 n
where Ukn = √1 eikxn is again the N × N unitary matrix realizing the discrete Fourier
N
kernel.
Actually, the Hamiltonian in terms of the normal modes (1.7) is not quite fully
diagonalized – it still mixes modes with +k and −k. To make the final step to decouple
the modes with k and −k, introduce the annihilation and creation operators
r r
~ †
1 ~mωk
For k 6= 0: qk = ak + a−k , pk = ak − a†−k .
2mωk i 2
To preserve (1.12), they must satisfy
16
Phonons. The groundstate is
It has energy ~ωk above the groundstate. (This is an eigenstate because [Nk , a†k ] = a†k ,
where Nk = a†k ak .) The excitation is called a phonon with momentum ~k. This is
what in single-particle quantum mechanics we would have called “|ki” – a state of a
single particle in a momentum eigenstate; we can make a state with one phonon in a
position eigenstate by taking superpositions:
X X
|one phonon at position xi = eikx |one phonon with momentum ~ki = eikx a†k |0i .
k k
E − E0 = ωk + ωk0 . (1.13)
Notice that since [ak , ak0 ] = 0 for k 6= k 0 , we have |k, k 0 i = |k 0 , ki – only the collection
of occupied modes matters, there is no way to specify which particle is in which mode.
So this construction allows us to describe situations where the number of particles
P
N = k Nk can vary! That is, we can now describe dynamical processes in which the
number of particles change. This is a huge departure from the description of quantum
mechanics where the hilbert space of two particles is a tensor product of the hilbert
space of each. How can we act with an operator which enlarges the hilbert space?? We
just figured out how to do it.
We can specify basis states for this Hilbert space
nk1 nk2
a†k1 a†k2
p p · · · |0i = |{nk1 , nk2 , ...}i
(n1 )! (n2 )!
by a collection of occupation numbers nk , eigenvalues of the number operator for each
P
normal mode. The energy of this state above the groundstate is k ~ωk , simply the
sum of the phonon energies. There are no interactions between them.
17
Notice that in this description it is manifest that phonons have no identity. We
only keep track of how many of them there are and what is their momentum. They
cannot be distinguished. Also notice that we can have as many as we want in the same
mode – nk can be any non-negative integer. Phonons are an example of bosons.
It is worth putting together the final relation between the ‘position operator’ and
the phonon annihilation and creation operators:
r
~ X 1 ikx 1
qn = √ e ak + e−ikx a†k + √ q0 (1.14)
2N m k ωk N
The items in red are the ways in which p and q differ; they can all be understood from
the relation p = mq̇. To see this, use the Heisenberg equations of motion, q̇n = i[H, qn ].
Gaplessness. The state a†k |0i has energy ~ωk above the groundstate. In a box of
size L = N a, the smallest energy phonon excitation has k1 = 2π
L
and energy
1 L→∞
∆E = ~ωk1 ∼ → 0. (1.15)
L
(Note that here I am taking L → ∞ to implement the thermodynamic limit of infinitely
many degrees of freedom; the lattice spacing can remain finite for this purpose – it is
not a continuum limit.) So this system is gapless in the sense that the gap between
the first excited state and the groundstate goes to zero in the thermodynamic limit
L → ∞. Gaplessness of a many-body system is something that requires explanation.
It isn’t even possible for a system with finitely many degrees of freedom.
Why does it happen here? Goldstone2 : the system has a symmetry under qn →
qn + for all n. If everyone moves to the left three feet, none of the springs are stretched.
This is the dance enacted by the k = 0 mode. If nearly everyone moves nearly three
feet to the left, the springs will only be stretched a little; hence the modes with small
k have small ω. Hence, in the thermodynamic limit where k can be arbitrarily small,
the energy gap ∆E goes to zero.
2
I should probably give an account of Goldstone’s theorem here. The relevant statement for our
purposes is: if the groundstate is not invariant under a symmetry of H, then it forms a multiplet of
that symmetry. Moreover, if the groundstate is not invariant under a continuous symmetry of H, then
the spectrum is gapless.
18
Tower of States: Now I will say a few words about the zeromode, which is
horribly mistreated in most textbook discussions of this system that I’ve seen. There
is no potential at all for this mode – it drops out of the (qn − qn+1 )2 terms. It just has
a kinetic term, which we can think of as the center-of-mass energy of the system. How
much energy does it cost to excite this mode? Notice that if everyone moves to the left
by a, the system comes back to itself (I am assuming that the masses are themselves
indistinguishable particles): |{qn }i ≡ |{qn + a}i are the same state. In terms of the
k = 0 mode, this is
N N
!
1 X 1 X √
q0 = √ qn e−i0xn ≡ √ qn + N a , i .e. q0 ≡ q0 + N a.
N n=1 N n=1
This means that the wavefunction for the zeromode must satisfy
√ 2πZ
eip0 q0 = eip0 (q0 + N a)
=⇒ p0 ∈ √
Na
and the first excited state has energy
2
p20
1 1 2π
|p0 = √2π = .
2m Na 2 Nm a
This is a victory at least in the sense that we expect the center of mass of the system to
have an intertial mass N m. Notice that the spacing of these states depends differently
on the parameters than that of the ones from the nonzero-k phonon states.
But actually this phenomenon is ubiquitous: it happens whenever we take a sys-
tem that breaks a continous symmetry (here: a solid breaks continuous translation
invariance)3 and put it in finite volume, i.e. depart from the thermodynamic limit. In
particular, in finite volume the zeromode associated with a conserved quantity (here
the momentum) produces a tower of states with a different level-spacing (as a function
3
The fine print:
1. Actually it’s important that the order parameter doesn’t commute with the Hamiltonian; the
exception is ferromagnets, where the order parameter is the total spin itself, which is a conserved
quantity and therefore can be nonzero even in finite volume. So the tower is collapsed at zero
in that case.
2. Actually, a one-dimensional mattress of oscillators will not spontaneously break continuous
translation symmetry even in infinite volume. This is a consequence of the Hohenberg-Coleman-
Mermin-Wagner theorem: the positions of the atoms still fluctuate too much, even when there
are infinitely many of them in a row; more than one dimension is required to have the crystal
really sit still. You’ll see the effects of these fluctuations on the problem set when you study
the Debye-Waller factors. This does not vitiate our conclusions above at all.
19
of system size L = N a) than the particle excitations (1.15). (It is sometimes called the
Anderson Tower of States in the study of magnetism or the Rotator spectrum in lattice
gauge theory). In this case, both towers go like 1/N , but this is a coincidence. In other
1
cases the tower from the zeromode is more closely spaced (it goes like volume ∼ L1d ∼ N1 )
than the particle momentum tower (which goes like L1 ∼ N 1/d 1
(or maybe L12 )), so the
tower of states from the zeromode is usually much closer together, and in the thermo-
dynamic limit L → ∞, they combine to form the degenerate vacua associated with
spontaneous symmetry breaking. 4
So you see that we have constructed an approximation to the Fock space of bosonic
particles from a system with finitely many degrees of freedom per unit volume (here,
length), and in fact finitely many degrees of freedom altogether, since we kept the IR
regulator L finite. It is worth pausing to appreciate this: we’ve been forced to discover
a framework for quantum systems in which particles can be created and annihilated,
very different from the old-fashioned point of view where we have a fixed Hilbert space
for each particle.
Now for some experimental verification of all this hard work:
Heat capacity of (insulating) solids: phonons are real. The simplest demon-
stration that phonons are real is the dramatic decrease at low temperatures of the heat
capacity of insulating solids. At high temperatures, the equipartition theorem of clas-
sical thermodynamics correctly predicts that the energy of the solid from the lattice
vibrations should be T times the number of atoms, so the pacity, CV = ∂T E should be
independent of T . At low temperatures T < ΘD , this is wrong. ΘD is the temperature
scale associated with the frequencies of the lattice vibrations (say the maximum of the
curve ωk above).
The Mössbauer effect: phonons are real. A nice example where we can see
the importance of the tower of states and of the quantization of phonon number is
the Mössbauer effect: when scattering high-energy photons off a solid, there is a finite
4
The definitive discussion of this subject can be found in the last few pages of P. Anderson, Concepts
in Solids.
20
amplitude for scattering events which produce zero phonons. This means that all the
momentum transfer goes to the center of mass mode, which has negligible recoil as
N → ∞, since its inertial mass is N m. This allows for very sharp absorption lines,
which if the atom were in free space would be washed out (i.e. broadened to a width
2
Erecoil = (p2m
γ)
.) by the random recoils of the individual atoms (as depicted in the comic
strip below).
The nuclei of the atoms in a solid have various energy levels; when hit with a γ-ray
photon, these nuclei can experience transitions from the groundstate to some excited
energy level. If an excited nucleus somewhere in the lattice gets hit by a very energetic
photon (a γ-ray) of some very specific energy Eγ = ∆E ≡ Eexcited − E0 , the nucleus
can absorb and re-emit that photon. The resulting sharp resonant absorption lines at
Eγ = ∆E are indeed observed.
This sounds simple, but here is a mystery about this: Consider a nucleus alone
in space in the excited state, after it gets hit by a photon. The photon carried a
momentum pγ = Eγ /c. Momentum is conserved, and it must be made up by some
recoil of the absorbing nucleus. When it emits a photon again, it needn’t do so in
the same direction. This means that the nucleus remains in motion with momentum
∆~p = p~1 − p~2 . But if some of its energy ∆E = Eexcited − E0 goes to kinetic energy of
recoil, not all of that energy can go to the final photon, and the emitted photon energy
2 p)2 2
will be less than Eγ by Erecoil = ∆p2M
max
. This can be as big as Erecoil = (2~
2M
= (2Eγ/c)
2M
(in
the case of scattering by angle π). So instead of a sharp absorption line, it seems that
2
we should see a broad bump of width (Eγ/c) M
. But we do see a sharp line!
The solution of the puzzle is phonons: for a nucleus in a lattice, its recoil means
that the springs are stretched – it must excite a lattice vibration, it must create some
phonons. But there is a nonzero probability for it to create zero phonons. In this case,
the momentum conservation is made up by an acceleration of the whole solid, which is
p2
very massive, and therefore does not recoil very much at all (it loses only energy 2NγM ).
21
This allows for very sharp resonance lines. In turn, this effect has allowed for some
very high-precision measurements.
The different widths in these cartoon absorption spectra don’t do justice to the relative
factor of N .
[End of Lecture 2]
An essentially similar effect makes it possible to get precise peaks from scattering
of X-rays off of a solid (Bragg scattering) – there is a finite amplitude for the scattering
to occur without exciting any phonons.
This is actually a remarkable thing: although solids seem ordinary to us because
we encounter them frequently, the rigidity of solids is a quantum mechanical emergent
phenomenon. You can elastically scatter photons off of a solid only because the atoms
making up the solid participate in this collective behavior wherein the whole solid acts
like a single quantum object!
where Ni,f and Li,f are initial and final states of the nucleus and lattice, respectively.
Hint is the perturbing hamiltonian by which the transition can occur. This is Fermi’s
golden rule. Because the nuclear forces are such high-energy things, we can ignore
the density of states of the final states, and we can assume that the transition matrix
element factorizes:
W (Ni , Li → Nf , Lf ) ∝ | hLf | HL |Li i |2 ,
where we’ve factored out some nuclear stuff that we don’t care about right now into
the ∝.
22
The requirements of translation invariance and Galilean invariance (i.e. momentum
is conserved, and the transition for a moving observer shouldn’t depend on the velocity
of the observer) require that
~
HL = AeiK·~x + h.c.
where ~K ~ is the momentum of the emitted gamma ray (a c-number), and ~x is the
center-of-mass position of the nucleus in question. 5 But, in the 1d case at least, we
have an expression for x in terms of the phonon creation operators:
X
xn = na + qn = na + Nk eikna ak + e−ikna a†k , (1.16)
k
q
where a is the lattice spacing and Nk = ~
2mN ωk
.
Now the amplitude for emitting no phonons is the ‘vacuum-persistence amplitude’,
i.e. the amplitude for |Li i = |0i to stay that way:
PMössbauer ∝ | h0| eiK(na+qn ) |0i |2 .
Now it is an exercise in harmonic oscillator physics to get a function out of this. A
useful general fact is that for harmonic oscillators (and any gaussian theory)
eiKq = e− 2 K hq i .
1 2 2
23
For future reference, these effects of fluctuations of the lattice on photon scattering
are called Debye-Waller factors.
Two final comments, to make connections with future excitement in this course:
I’ve made a big deal about the regulators here. One reason we care about them
is if we remove them (N → ∞, a → 0) and ask bad questions, we’ll get infinity. For
example, we could think about the vacuum energy E0 = 21 k ~ωk . There is physics in
P
24
Interactions. Many aspects of the above discussion are special to the fact that
our hamiltonian was quadratic in the operators. Certainly our ability to completely
solve the system is. Notice that the number of phonons of each momentum Nk ≡ a†k ak
is conserved for each k. But if we add generic cubic and quartic terms in q (or if we
P
couple our atoms to the photon field) even the total number of phonons k Nk will
no longer be a conserved quantity7 . So a description of such particles which forced us
to fix their number wouldn’t be so great.
P
For example, think about expanding a general cubic term nml λnml qn qm ql in os-
cillators. It has terms like
a†k1 a†k2 a†k3 , a†k1 a†k2 ak3 , a†k1 ak2 ak3 , ak1 ak2 ak3
all of which change the number of phonons. A quartic terms will also contain terms
like a† a† aa which preserve the number of phonons, but describe an interaction between
them, where they exchange momentum.
Above, we have stumbled upon an example of a quantum field theory (so far, a scalar
field in one spatial dimension). Let me explain in terms of the action.
The action for our collection of oscillators is
Z !
X1
S[q] = dt mn q̇n2 − V ({q})
n
2
with V ({q}) = n 12 κ (qn+1 − qn )2 . Now let’s try to take the continuum limit a →
P
(q 1 , q 2 ) → (cos θq 1 , sin θq 2 ).
We can reorganize this as a complex field Φ = q 1 + iq 2 on which the symmetry acts by Φ → eiθ Φ.
25
We now have
(qn − qn−1 )2 ' a2 (∂x q)2 |x=na
Now the action becomes:
Z Z Z Z
1 2 2 2 2 4
S[q] = dt dx µ (∂t q) − µvs (∂x q) − rq − uq − ... ≡ dt dxL (1.17)
2
where I’ve introduced some parameters µ, vs , r, u determined from m, κ... in some ways
that we needn’t worry about. L is the Lagrangian density whose integral over space is
R
the Lagrangian L = dxL.
The equation of motion is
δS
0= = −µq̈ − µvs2 ∂x2 q − rq − 2uq 3 − ...
δq(x, t)
π(x)2
X Z Z
2 2 2 4
H= (pn q̇n − Ln ) = dx (π(x)q̇(x) − L) = dx + µvs (∂x q(x)) + rq + uq + ... .
n
2µ
(1.18)
(Note that I suppress the dependence of all the fields on t just so it doesn’t get ugly,
not because it isn’t there.)
If we were feeling fancy, we could now talk more about the field operator
s
~ X 1 ikx
q(x) = √ e ak + e−ikx a†k
2µL k ωk
(π(x) is the quantum operator associated with the field-momentum π above.) Notice
that the position along the chain x here is just a label on the fields, not a quantum
26
operator. The point of the mode operators a is that in terms of them, the Hamiltonian
is a sum of independent terms
X † 1
H= ~ωk ak ak + .
k
2
The field q is called a scalar field because it doesn’t have any indices decorating it.
This is to be distinguished from the Maxwell field, which is a vector field, and which
is our next subject. (Note that vibrations of a crystal in three dimensions actually do
involve vector indices. We will omit this complication from our discussion.)
The lattice spacing a and the size of the box N a in the discussion above are playing
very specific roles in regularizing our 1-dimensional scalar field theory. The lattice
spacing a implies a maximum wavenumber or shortest wavelength and so is called an
“ultraviolet (UV) cutoff”, because the UV is the short-wavelength end of the visible
light spectrum. The size of the box N a implies a maximum wavelength mode which
fits in the box and so is called an “infrared (IR) cutoff”.
There is one annoying thing I should mention about the infinite-volume limit. In
P
that limit, the wavenumbers become continuous, and the k wants to become an
1
P L→∞ R R dk
integral. The correct replacement is L k d̄k ≡ 2π . Note that the units work
L→∞
out. We will also want to replace the Kronecker delta in [ak , a†k0 ] = δk,k0 2πδ(k−k 0 )
with a Dirac delta function. Here the units√don’t work. This last step requires a change
in normalization of the ak s by a factor of L.
Continuum (free) scalar field theory in d + 1 dimensions
Notice that these continuum expressions are easy to generalize to scalar field theory
in any number of dimensions: The action is
Z
d 1 2 1 2~ ~
S[φ] = dtd x µφ̇ − µvs ∇φ · ∇φ − V (φ)
2 2
and the Hamiltonian is
π(x)2 1 2 ~
Z
H= d
d x ~
+ µvs ∇φ · ∇φ + V (φ)
2µ 2
with π = µq̇. Again, think of qn (t) ≡ φ(xn , t), pn (t) ≡ π(xn , t) as defining the fields
via their values at the grid points. The equations of motion are
δS ∂V
0= = −µφ̈ + µvs2 ∇2 φ + .
δφ(t, ~x) ∂φ
Let’s think about the special case where V (φ) = 12 m2 φ2 , so the model is gaussian,
and the equation of motion is linear in φ. A translation invariant linear problem is
27
~ ~
√1 e−ik·~x φk , π(x) = √1 e−ik·~x πk , this is
P P
solved by Fourier transforms: φ(x) = Ld k Ld k
X 1 1
2 2 2
H= πk π−k + µvs k + m φk φ−k
k
2µ 2
this is
X † 1
H= ~ωk ak ak +
k
2
where
ωk2 = vs2 k 2 + m2 .
This is morally the same equation as our starting point for each ball on springs:
[End of Lecture 3]
28
1.5 Quantum light: Photons
The quantization of the Maxwell field is logically very similar to the preceding dis-
cussion. There are just a few complications from its several polarizations, and from
the fact that quantum mechanics means that the vector potential is real and necessary
(whereas classically it is just a convenience).
Maxwell’s equations are :
~ ·B
∇ ~ = 0, ∇
~ ×E
~ = −∂t B,
~ (1.19)
~ ·E
∇ ~ = ρ/ε0 , ~ ×B
c2 ∇ ~ = ∂t E
~ + ~j/ε0 . (1.20)
~ and B
The first two equations (1.19) are constraints on E ~ which mean that their com-
ponents are not independent. This is annoying for trying to treat them quantumly. To
get around this we introduce potentials which determine the fields by taking derivatives
and which automatically solve the constraints (1.19):
~ = −∇Φ
E ~ − ∂t A,
~ ~ =∇
B ~ × A.
~
In the absence of sources ρ = 0 = ~j, we can also set Φ = 0. In this gauge, Ampere’s
law becomes
~ × ∇
c2 ∇ ~ ×A~ = c2 ∇~ · ∇~ ·A
~ − c2 ∇ 2 A~ = −∂ 2 A~ i.e. ∂ 2 A
~ − c2 ∇2 A
~=0.
t t
This wave equation is different from our scalar wave equation (1.9) in three ways:
0 = ~k · A(k)
~
29
Recall that the energy density of a configuration of Maxwell fields is u = 0 ~ 2 2 ~2
E +c B .
2
So the quantum Hamlitonian is
Z ε Z 2
ε0 3
~ +c B
2 2 ~ =
2 0 3
2
~ + ∇ ~ ×A ~
H= dr E dr ∂t A . (1.21)
2 2
Here E~ = −∂t A ~ plays the role of field momentum π(x) in (1.18), and B ~ =∇ ~ ×A
~ plays
the role of the spatial derivative ∂x q. We immediately see that we can quantize this
system just like for the scalar case (by analogy, replacing L → L3 , µ → 0 ) by expanding
the quantum Maxwell field in terms of independent creation and annihilation operators:
r
~
XX ~ i~k·~
r † ? −i~k·~
r
A(~r) = 3
a ~
e
~k,s s (k̂)e + a ~
e
~k,s s (k̂)e .
s=1,2
20 ωk L
~k
We’re going to need to know how this evolves in time (for example, because we’ll
need to construct E ~ = −∂t A).~ One way to do this is to solve the Heisenberg evo-
~ i ~
lution equation ∂t A = ~ [H, A] (using the fact that H is time-independent) to write
~ r, t) = e−iHt/~ A(~
A(~ ~ r)eiHt/~ . Alternatively, we can just write the general positive-
energy, transverse solution of the Maxwell equation (two polarizations for each ~k), as
~
a superposition of each such solution eik·~r−iωk t~es (k̂) (where k̂ · ~es = 0 guarantees trans-
verseness and ωk2 = c2 k 2 solves the Maxwell equations for A); ~ then replace the constant
coefficient in front of each solution with its own annihilation operator, a~k,s satisfying
[aks , a†k0 s ] = δkk0 δss0 . (1.22)
The result is
r
~ r, t) =
XX ~ i~k·~
r−iωk t † ? −i~k·~
r+iωk t
A(~ a~ ~
e s ( k̂)e + a ~
e
~k,s s ( k̂)e .
s=1,2
20 ωk L3 k,s
~k
8
I should say a bit more about the polarization vectors, ~es . They conspire to make it so that there
are only two independent states for each ~k and they are transverse ~k · ~es (k̂) = 0, so s = 1, 2. The
polarization vectors for a given k can be chosen to satisfy a completeness relation:
X
esi (k̂)e?sj (k̂) = δij − k̂i k̂j . (1.23)
s
30
~
where I’ve highlighted in red the places where these expressions differ from that for A.
It seems like the canonical commutator should generalize according to
? ?
[φ(x), π(x0 )] = i~δ(x − x0 ) [Ai (~r), Ej (~r0 )] = −i~δ 3 (~r − ~r0 )δij /ε0
where i, j = 1..3 are spatial indices. This is not quite true. If we plug in the expressions
above and use (1.22) and (1.23) we find instead
Z
0 3 i~k·(~ r0 )
r−~
[Ai (~r), Ej (~r )] = −i~ d̄ k e δij − k̂i k̂j /ε0
As a check, note that using this Hamiltonian and the canonical commutator, we can
reproduce Maxwell’s equations:
∂t2 A ~ = − i [H, E]
~ = −∂t E ~ = c2 ∇
~ 2 A.
~
~
The last step is a bit nontrivial (see the homework).
Plugging these expressions into the Hamiltonian (1.21), we can write it in terms of
these oscillator modes (which create and annihilate photons). As for the scalar field,
the definitions of these modes were designed to make this simple: It is:
X † 1
H= ~ωk a~k,s a~k,s + .
2
~k,s
The groundstate is the vacuum |0i annihilated by all the annihilation operators
aks |0i = 0. Notice that the vacuum energy is
Z
1X 1 3
E0 ≡ h0| H |0i = ~ωk = L d̄3 k~ck. (1.24)
2 2
~k,s
P
The fact that in the continuum k is no longer a finite sum might be something
to worry about. We will see below in §1.6 that this vacuum energy has physical
consequences.
The first excited states have the form
a†k,s |0i
and represent a single photon of momentum k in the polarization state s. Note that
the polarization represents a double-degeneracy of each k-state. We can make states
of two photons a†k,s a†k0 ,s0 |0i = a†k0 ,s0 a†k,s |0i. Photons are identical bosons. Again we can
make a basis for the full Hilbert space by specifying the occupation numbers of each
mode nk1 nk2
† †
ak1 ,s1 ak2 ,s2
p p · · · |0i = |{nk1 ,s1 , nk2 ,s2 , ...}i .
(n1 )! (n2 )!
31
Consolidation of understanding
So far in this chapter, we have studied systems of increasing complexity: the simple
harmonic oscillator, a scalar field, and the EM field. They all have the same structure,
in the following sense.
In the following, Here ReA ≡ 12 A + A† as usual. The normalization constant in
q
finite volume V is Nk = 12 2mω~ k V .
1 2 1 2 2 † 1
HSHO = p + mω q = ~ω a a +
2m 2 2
[q, p] = i~ =⇒ [a, a† ] = 1.
q = ReN a, p = mImωN a.
Z
1 2 1 2 X † 1
H1d scalar = dx π + µc (∂x φ)2 = ~ωk ak ak +
2µ 2 k
2
! !
X X
φ(x) = Re Nk eikx ak , π(x) = µIm ωk Nk eikx ak .
k k
0 ~ 2 0 c2 ~ 2
Z
3
X † 1
HEM = dx E + B = ~ωk aks aks +
2 2 k,s=1,2
2
?
[aks , a†k0 s0 ] = ~δk,k0 δss0 ⇔ [Ai (x), Ej (x0 )] = i~δ 3 (x − x0 )δij .
! !
~ ~
X X
~
A(x) = Re ~
Nk eik·~x aks~es (k̂) , E(x) = µIm ωk Nk eik·~x aks~es (k̂) .
k k
32
Mössbauer more microscopically
As an example where you can see the photon machinery in action, we return to our
discussion of the Mössbauer effect. Now we can answer in more detail the question:
where did that Hint that we used in the Mössbauer effect come from? This requires
information from all of the previous subsections: it involves both phonons and photons.
A more microscopic description of the transition rate would include the radiation field,
too:
W (Ni , Li , Ri → Nf , Lf , Rf ) ∝ | hf | Hint |ii |2 ,
where now Ni,f , Li,f , Ri,f denote initial and final states of the nucleus (and electrons),
lattice and radiation field, respectively. We are working in a big hilbert space H =
HN ⊗ HL ⊗ HR .
Again we ignore the nuclear stuff:
If the photon gets absorbed, the final state of the radiation field is the vacuum, no
photons:
hRf | = R h0|
which is annihilated by the photon annihilation operators: aK,s |0iR = 0 and hence
†
R h0| aK,s = 0. The initial state is one photon of momentum K and some polarization
s:
|Ri i = |K, siR = a†K,s |0iR .
What is the interaction hamiltonian Hint ? As you’ll remind yourself on the home-
work, the hamiltonian for a charged particle (such as an ion in the solid) in an EM
field is
1 ~
2 1 p2
p2 + epA + eAp + e2 A2 =
H1 = ~p + eA(x) = + Hint .
2m 2m 2m
Think of e as small, so that we may treat Hint as a perturbation. Here we should use
our expression above for the quantized photon field:
X
A(x) ∼ aK eiKx + a†K e−iKx .
K
The catch here is that we have to evaluate this at the location of the ion, which means
that the x appearing in the argument is an operator, x!
The final term in H1 proportional to A2 changes the number of photons by an
even number, and so doesn’t participate in the process we are talking about where one
33
photon is absorbed. So we just need to think about the middle terms with one A. The
p is pretty innocuous:
peiKx + eiKx p = (p + K) eiKx
and we’ll just focus on the second term. So the matrix element is:
[End of Lecture 4]
34
1.6 Casimir effect: vacuum energy is real
[Le Bellac, 11.5.12 page 399; A. Zee, Quantum Field Theory in a Nutshell] This sub-
section has two purposes. One is to show that the 21 ~ω energy of the vacuum of the
quantum harmonic oscillator is real. Sometimes we can get rid of it by choosing the
zero of energy (which doesn’t matter unless we are studying dynamical gravity). But
it is meaningful if we can vary ω (or the collection of ωs if we have many oscillators as
for the radiation field) and compare the difference.
The other purpose is to give an object lesson in asking the right questions. In
physics, the right question is often a question which can be answered by an experiment,
at least in principle. The answers to such questions are less sensitive to our silly
theoretical prejudices, e.g. about what happens to physics at very short distances.
In the context of the bunch of oscillators making up the radiation field, we can
change the spectrum of frequencies of these oscillators {ωk } by putting it in a box and
varying the size of the box. In particular, two parallel conducting plates separated by
some distance d experience an attractive force from the change in the vacuum energy
of the EM field resulting from their presence. The plates put boundary conditions on
the field, and therefore on which normal modes are present.
To avoid some complications of E&M which are not essential for our point here,
we’re going to make two simplifications:
To avoid the problem of changing the boundary conditions outside the plates we
use the following device with three plates:
| ← d → | ←− L−d −→ |
(We will consider L d, so we don’t really care about the far right plate.) The
‘perfectly conducting’ plates impose the boundary condition that our scalar field q(x)
vanishes there. The normal modes of the scalar field q(x) in the left cavity are then
35
The vacuum energy between the outer plates is the sum of the vacuum energies of
the two cavities
E0 (d) = f (d) + f (L − d)
where ∞ ∞
1 X π X ?!?!!
f (d) = ~ ωj = ~c j = ∞.
2 j=1 2d j=1
We have done something wrong. Our crime is hubris: we assumed that we knew what
the modes of arbitrarily large mode number k (arbitrarily short wavelength, arbitrarily
high frequency) are doing, and in particular we assumed that they cared about our silly
plates. In fact, no metal in existence can put boundary conditions on the modes of
large enough frequency – those modes don’t care about d. The reason a conductor puts
boundary conditions on the EM field is that the electrons move around to compensate
for an applied field, but there is a limit on how fast the electrons can move (e.g. the
speed of light). The resulting cutoff frequency is called the plasma frequency but we
don’t actually need to know about all these details. To parametrize our ignorance of
what the high-frequency modes do, we must cut off (or regularize) the contribution of
the high-frequency modes. Let’s call modes with ωj π/a high frequency, where here
a is some short time9 .
Then one possible model of the microphysics, which makes the sum both finite and
doable, is realized by making the replacement:
∞
π X −aωj /π
f (d) f (a, d) = ~c e j
2d j=1
∞
!
π~c X
= − ∂a e−aj/d
2 j=1
| {z }
1
= −1
1−e−a/d
π~c ea/d
= +
2d (ea/d − 1)2
ad πd π πa2
' ~ − + + ... (1.25)
2a2
|{z} 24d 480d 3
→∞ as a→0
Answers that don’t depend on a have a chance of being meaningful. The thing we can
measure is the force:
36
π
π π 2 π 2
= −~c + + O(a ) − + + O(a )
2a2 24d2 2a2 24 (L − d)2
a→0 π~c 1 1
= − −
24 d2 (L − d)2
Ld π~c
= − (1 + O(d/L)) . (1.26)
24d2
This is an attractive force between the plates.
The analogous force between real conducting plates, caused by the change of bound-
ary conditions on the electromagnetic field, has been measured.
The string theorists will tell you that ∞ 1
P
j=1 j = − 12 , and our calculation above
agrees with them in some sense. But what this foolishness means is that if we compute
something which is not dependent on the cutoff we have to get the same answer no
matter what cutoff we use. Notice that it is crucial to ask the right questions.
An important question is to what extent could we have picked a different cutoff
function (instead of e−πω/a ) and gotten the same answer for the physics. This interest-
ing question is answered affirmatively in Zee’s wonderful book, 2d edition, section I.9
(available electronically here!).
A comment about possible physical applications of the calculation we actually did:
you could ask me whether there is such a thing as a Casimir force due to the vacuum
fluctuations of phonons. Certainly it’s true that the boundary of a chunk of solid
puts boundary conditions on the phonon modes, which change when we change the
size of the solid. The problem with the idea that this might produce a measurable
force (which would lead the solid to want to shrink) is that it is hard to distinguish
the ‘phonon vacuum energy’ from the rest of the energy of formation of the solid,
that is, the energy difference between the crystalline configuration of the atoms and
the configuration when they are all infinitely separated. Certainly the latter is not
well-described in the harmonic approximation (λ = 0 in (1.6)).
37
wave vectors are then
~k = 2πnx 2πny πn
, ,
Lx Ly L
with nx , ny , n integers.
We have to do a bit of E&M here. Assume the plates are perfect conductors (this
is where the hubris about the high-frequency modes enters). This means that the
transverse component of the electric field must vanish at the surface. Instead of plane
waves in z, we get standing waves: Ex,y (z) ∝ sin (nπz/L) .
The frequencies of the associated standing waves are then
r
π 2 n2 ~ 2
ωn (~k) = c + k , n = 0, 1, 2...
L2
~ ∝ ẑ) for n = 0.
Also, there is only one polarization state (E
So the zero-point energy is
0
~ X
E0 (L) = 2 ωn (~k)
2
n,~k
Now you can imagine introducing a regulator like the one we used above, and replacing
0 0
~
X X
· e−aωn (k)/π ·
n,~k n,~k
and doing the sums and integrals and extracting the small-a behavior. The result is
π2
α = − 2·5! , an attractive force. It was measured in 1998.
38
1.7 Identical particles
where here π denotes a permutation on N objects (12...N ) → (π1 π2 ...πN ). This means
that the wavefunction (P = |Ψ|2 ) must be preserved up to a phase:
where a is some phase (recall: multiplying the whole wavefunction by a phase does not
change the state). Switching them back gives back the first state10 :
39
so a2 = 1. There are two solutions: a = +1 and a = −1 and the two classes of
particles associated with these two choices are called respectively bosons and fermions.
For bosons, the wavefunction is simply permutation invariant:
ΨB (1 · · · N ) = ΨB (π1 · · · πN ).
For fermions, the sign of the wavefunction changes under an odd permutation:
ΨF (1 · · · N ) = (−1)π ΨF (π1 · · · πN ).
An odd permutation is one that can be made from an odd number of transpositions.
(For example (123) → (213) is odd, but (123) → (231) is even.)
Note that the Hilbert space of N indistinguishable particles is therefore not quite
a tensor product of the Hilbert spaces of the individual particles. In fact, both the
bosonic states and the fermionic states fit easily into H1⊗N . If D ≡ dim H1 , then
D! D
dim HF = =
N !(D − N )! N
(N + D − 1)! N +D−1
dim HB = =
N !(D − 1)! N
and in fact dim HF + dim HB ≤ dim H1⊗N = DN (for N > 1).
different from going around the other way. This leads to the possibility of anyons.
One way to see that something is different in D = 2 + 1 is to think about the path integral for, say
two particles. In the special case of 2 + 1 dimensions, the worldlines of the particles can braid around
each other; undoing a braid requires a collision between the particles. In more dimensions, the paths
can be unlinked without any collision. This means that in D = 2 + 1, the sum over paths can be
divided up into topological sectors, labelled by the number of times n the particles wind around each
other. Nothing can stop us from weighting the contributions of paths in different sectors differently:
X
Z= einθ Zn
n
where Zn is the sum over all paths in the sector labelled n. The phase θ determines the statistics of
the particles.
There is an even more dramatic thing that can happen, again only in D = 2 + 1. There can be a
degenerate subspace of the Hilbert space associated with the presence of two anyons. Exchanging the
two particles, or moving them around each other, can then result in not just a phase, but a whole
unitary operator acting within this degenerate subspace. Since two such unitaries generally do not
commute, such particles are called non-abelian anyons.
40
How then can we ever discuss the wavefunction of a single electron?? Don’t we
have to antisymmetrize it with the state of all the other electrons in the world?
Suppose there were only two electrons in the world, and the two single-particle
states involved were ψH (x) and ψT (x) (‘H’ for ‘here’ and ‘T’ for ‘there’). The
full state is
1
Ψ(x1 , x2 ) = √ (ψH (x1 )ψT (x2 ) − ψT (x1 )ψH (x2 )) . (1.27)
2
What is the probability of finding an electron somewhere around here at location
x?
Z Z
P (an electron at x) = dx2 |Ψ(x, x2 )| + dx1 |Ψ(x1 , x)|2
2
(1.28)
Z Z
= |ψH (x)|2 dy|ΨT (y)|2 + |ψT (x)|2 dy|ΨH (y)|2
Z
? ?
− 2ReψH (x)ψT (x) dyψH (y)ψT (y) . (1.29)
But suppose we know that the two states don’t overlap: ψT (x)ψH (x) = 0 for
all x. Then the last interference term vanishes, and one or the other of the first
two terms vanishes wherever we look, and by normalization of ψT and ψH , the
other reduces to the formula we would have used if we didn’t know about this
antisymmetrization stuff.
Conclusion: you only need to worry about antisymmetrizing states that overlap.
are allowed states of bosons and fermions, respectively. In the second-quantized for-
malism, we only need to answer: how many particles are in each single-particle state?
Then the states (1.30) would be denoted
|nα = 1, nβ = 1i
with no further fuss. The second-quantized language is like Newspeak, the language in
Orwell’s 1984 in which it is not possible to formulate heretical statements. But in a
good way.
41
A second advantage of the second-quantized language is that the single-particle
language quickly becomes unwieldy as the the number of particles grows. Above in
(1.30) I wrote a state of two identical bosons or fermions. Its wavefunction, in position
space, say, is
ΨB/F (x1 , x2 ) = uα (x1 )uβ (x2 ) ± uβ (x1 )uα (x2 ).
Here each uα (x) ≡ hx|αi is a single-particle wavefunction. Suppose I had three parti-
cles. What does the wavefunction look like if I fill three levels α, β, γ? Here’s a trick
for the Fermi case:
ΨF (x1 , x2 , x3 ) = det M (1.31)
where
uα (x1 ) uα (x2 ) uα (x3 )
M ≡ uβ (x1 ) uβ (x2 ) uβ (x3 ) .
uγ (x1 ) uγ (x2 ) uγ (x3 )
The structure of the determinant
X
det M = (−1)π M1π1 · · · MN πN
π∈SN
makes it automatic that this is antisymmetric under interchange of the position labels,
ΨF (x1 , x2 , x3 ) = −ΨF (x2 , x1 , x3 ). The wavefunction (1.31) is called a Slater deter-
minant. For N fermions the state associated to a collection of single-particle levels
(α ···α )
{α1 · · · αN } is ΨF 1 N (x1 · · · xN ) = detij uαi (xj ).
To make the wavefunction for bosons made from these same single-particle states,
just remove all the minus signs involved in constructing the determinant:
X
ΨB (x1 · · · xN ) = M1π1 · · · MN πN ≡ Per(M ),
π∈SN
which is called the permanent of M . Now the αi don’t all have to be different.
An immediate consequence of the minus sign under exchange of fermion labels is
the Pauli exclusion principle:
No two fermions can occupy the same single-particle state. The ground state of a
collection of (non-interacting) fermions is therefore quite interesting, since we must find
a different single-particle state in which to put each of our fermions. This has many
dramatic consequences, including the periodic table of elements, and the distinction
between metals and insulators that we discuss next. [End of Lecture 5]
42
1.7.1 Band structure: metals and insulators
Now we will say some words about fermions in periodic potentials. This will allow us
to quickly understand a way to distinguish metals and insulators. Then we’ll translate
to the second-quantized language.
Consider a collection of fermions that care about each other only because of the
Pauli principle – the hamiltonian is a sum of terms involving the fermions one at a
time. Suppose that each fermion (call them electrons) can live at one of N sites in a one
dimensional crystal, and can hop from one site to the next by some tunneling process.
Further suppose that each site involves several (we’ll say two for simplicity) atomic
orbitals (or spin states), so the one-particle hilbert space is H1 = span{|ni ⊗ |αi , n =
1...N, α = 0, 1}. We’ll suppose that each electron is governed by the hamiltonian
X X
H1 = −t (|nihn + 1| + |n + 1ihn|) ⊗ 1 + |nihn| ⊗ |1ih1| ≡ Ht + H .
n n
The first term allows the electrons to hop. The second term says that one of the orbitals
has lower energy than the other (|0i is the ground state and |1i has energy > 0). For
example, if α is a spin label, this could be the result of a Zeeman field of strength .
What is the spectrum of H1 , the hamiltonian of one electron hopping in this solid?
The two terms commute [Ht , H ] = 0, so we can simultaneously diagonalize them.
Let’s assume periodic boundary conditions for simplicity. Notice that the problem has
a discrete translation invariance, which should be familiar by now. Moreover, it’s just
two copies of the problem on the problem set; the eigenstates are eigenstates of the
momentum
1 X ikna
|ki ⊗ |αi = √ e |ni ⊗ |αi α = 0, 1
N n
with eigenvalues11
εα (k) = −2t cos ka + δα,1 . (1.32)
43
Here the allowed bands of energies are highlighted:
> 4t), this means we just fill the bottom band: Filled
44
(‘conduction’) band: This costs energy ∆E = − 4t which
ter and eventually (for < 4t) they are filled in the ground state.
The name for the energy of the last filled level is the Fermi energy εF . (and the name
of its momentum is the Fermi momentum, kF ). 12 Now the first excited state of the
N -electron system is achieved by a very small change – we can stay in the same band.
The energy cost to excite the system from the ground state is of order ∆k∂k ε|εF ∼ N1
(where ∆k = N2πa ) which goes to zero as N → ∞. There is no energy gap. When the
Fermi energy is in the middle of a band, the system is a metal.
Groundstates of such macroscopic (N → ∞) systems can be called states of matter.
This sort of sharp distinction between possible behaviors of states of matter – like
whether there is an energy gap – is the central preoccupation of condensed matter
physics.
By the way, what we have really shown here is that when the Fermi energy is in
the middle of the band, the system has very low-energy excitations. The fact that it
actually conducts electricity is also a consequence of quantum mechanics. It happens
because the wavefunctions of these low-energy excitations are extended across the ma-
terial – they are (quasi-)momentum (k) eigenstates. This means that they can carry
a current across the sample, Imψk? ∂x ψk 6= 0. Notice that this picture departs dra-
matically from the classical (Drude) picture of charge transport by a charged particle
bouncing around in the lattice. Notice that the form of the lattice is already built into
the wavefunctions ψk ! (At this point we actually have the opposite problem that the
answer we would compute for the resistance of such a metal is zero. To get the right
12
Partly because we chose such a simple example for our Hamiltonian, we find several (4) places
where the bands cross the Fermi energy – several Fermi momenta.
45
finite answer we would need to include some form of disorder in the lattice, or phonons,
and/or interactions between the electrons.)
Fermionic operators. What is the analog of the description of the many-body
system in terms of mode operators ak that we had for phonons and photons? We can
introduce operators that create and annihilate electrons just like we did before:
Notice that they are labelled by a position n and an orbital label α = 0, 1. Our boson
creation operators satisfied the algebra 1 = [a, a† ] = aa† − a† a (for each mode) and
this led to boson statistics. We need somehow to prevent two electrons from being in
the same state. We can accomplish the Pauli principle simply by demanding that
c2 = 0
2 2
for each mode, i.e. cnα = 0 = c†nα . It’s just zero. More generally, in order to
make sure that the wavefunction is antisymmetric under interchange,
1 = {c, c† } ≡ cc† + c† c.
This is called an anticommutator. With the labels the algebra should be:
This last statement is weird: fermion operators, even if they create particles that are
far apart, do not commute. Rather they anticommute. Fermions are weird in this
sense. (We saw above however that this does not mean that you have to worry about
all the electrons in the world when thinking about the physics of the electrons you’re
interested in.)
c2 = 0, {c, c† } = 1
The two states of the qubit just describe the presence or absence of the fermion in this
state. Exercise: we can rewrite the Pauli operators as
1
σ x = c + c† , σ y = c − c† , σ z = 2c† c − 1.
i
46
These operators satisfy the Pauli algebra, σ i σ j = iijk σ k . Also note that σ + =
c† , σ − = c and the number operator c† c (whose eigenvalues are 0 and 1) is c† c =
1 z
2
σ + 1.
with εα (k) given above in (1.32). The ground state of n such electrons is
Y
|ground statei = c†αk |0i .
n (k, α) with the smallest α (k)
where the single-particle wavefunctions are ukβ (rα) = √1N eikr δαβ . (The dependence on
α, β is so simple because of our simple choice of Hamiltonian.) [End of Lecture 6]
47
1.8 Second quantization, from scratch
Above, we discovered the second quantized description in an organic way. Let’s back
up and consider the formalism more generally on its own terms.
Creation operators for general one-particle states. Given a one-particle state
with momentum-space wavefunction ϕ(k) ≡ hk|ϕi, let
X †
a† (ϕ) ≡ ak ϕ(k). (1.34)
k
How do we write the corresponding operator on the many-body Hilbert space? In the
first-quantized language, a basis for the states with n particles is of the form
X
|u1 · · · uN i = ζ π |uπ1 i ⊗ |uπ2 i ⊗ · · · ⊗ |uπn i
π
where ui are some single-particle states (they could be the same basis as α, β), π ∈ Sn
is a permutation, and again ζ = ± for B/F. Here by (−1)π I mean −1 if π is an odd
permutation and +1 otherwise. On such a state, we want A to act as
X
A |u1 · · · uN i = ζ π (A1 |uπ1 i) |uπ2 · · · uπn i + |uπ1 i (A1 |uπ2 i) |uπ3 · · · uπn i + · · ·
| {z }
π P
= αβ Aαβ |αihβ|uπ1 i
(1.35)
48
The effect of the operation |αihβ| is to replace the state β with the state α. On the
many-body Hilbert space, this happens wherever the A1 acts. But this is precisely
what is accomplished by the many-body operator
a† (α)a(β)
– it destroys a particle in state β and creates a particle in state α. Notice that this is a
bosonic object, no matter what are the statistics of the as. If there is no such particle
to destroy, it just gives zero. Therefore,
X
A= a† (α)a(β)Aαβ .
αβ
For example, we can write the kinetic energy of a collection of free particles in this
p2
language. The single-particle operator is just H1 = 2m . On n particles, we would write
n
X p2i
Hn = .
i=1
2m
k 2
In this last expression, p = ~k is just a number – 2m are the matrix elements of Aαβ in
this example, which happens to be diagonal in the k basis of the one-particle Hilbert
space. Notice that the fixed-particle-number operators above can be written as
Hn = Πn HΠn
Above, I claimed without much proof that the state we get by filling single-particle
levels and (anti)symmetrizing is a Slater determinant or permanent. Let me clarify
that statement. Consider an n-particle state
1 X σ
|α1 · · · αn i = √ ζ |ασ1 i ⊗ · · · ⊗ |ασn i
n! σ∈Sn
where the αi are arbitrary 1-particle states. Then the overlap between any two such
49
states is
1 X π σ
hα1 · · · αn |β1 · · · βn i = ζ ζ (hασ1 | · · · hασn |) (|βπ1 i ⊗ · · · ⊗ |βπn i) (1.36)
n! σ,π
1 X π σ
= ζ ζ hασ1 |βπ1 i · · · hασn |βπn i (1.37)
n! σ,π
1 X πσ−1
= ζ α1 |βπσ−1 (1) · · · αn |βπσ−1 (n) (1.38)
n! σ,π
X
= ζ ρ hα1 |βρ1 i · · · hαn |βρn i (1.39)
ρ=πσ −1
hα1 |β1 i · · · hα1 |βn i
= ... ..
.
.. (1.40)
.
hαn |β1 i · · · hαn |βn i ζ
respectively. In the tricky step after (1.37) we permuted the factors in the product by
−1 −1
the permutation σ (they are just numbers after all), and used ζ π ζ σ = ζ π ζ σ = ζ πσ .
Particles in free space. [Baym, chapter 19] Let’s focus for a bit on the special
but important example of particles (bosons or fermions) living in a rectangular periodic
box of volume V = Lx Ly Lz in the continuum. Then a useful basis of single-particle
states is the plane waves
ei~p·~r 2πni
up (r) = hr|pi = √ , pi = , ni ∈ Z, i = x, y, z.
V Li
We can include a spin label s =↑, ↓, too. So a†ps adds a particle with momentum p
and spin s; the amplitude for the particle to be at r is up (r). And aps removes such a
particle. These particles are bosons or fermions according to the algebra of the as
Now consider X
ψs† (r) ≡ u?p (r)a†ps .
p
This adds a particle in a particular superposition of plane waves. In the state ψs† (r) |0i,
the amplitude for finding the particle at position r0 is13 (using the completeness relation
13
More precisely, including the spin label, the amplitude to find a particle at position at r0 with
50
P
for our single-particle basis 1 1 = p |pihp|)
X
hr0 | ψ † (r) |0i = u?p (r)up (r0 ) = δ(r − r0 ).
p
(The δ function on the RHS here is a d-dimensional δ-function, δ d (r − r0 ) but I’ll omit
the superscript d to avoid clutter.) So we can say that ψ † (r) adds a particle at position
r. (Notice that this is a special case of the general formula (1.34); if we wanted to be
uniform, we would call it a† (r).) The ψ(r) are called “field operators,” just like φ in
our discussion of scalar field theory or Aµ (x) in our discussion of electrodynamics in
vacuum. A big difference is that now ψ may be a fermionic object (if ζ = −1). What
I mean by this is that they satisfy
X
† 0 † 0 † †
ψs (r)ψs0 (r ) − ζψs0 (r )ψs (r) = ? 0
up (r)up0 (r )δss0 aps ap0 s0 − ζap0 s0 aps = δss0 δ(r − r0 ).
pp0 | {z }
=δpp0 δss0
(1.41)
0 0 † † 0 † 0 †
ψ(r)ψ(r ) − ζψ(r )ψ(r) = 0, ψ (r)ψ (r ) − ζψ (r )ψ (r) = 0. (1.42)
When adding or creating particles at different points, the order doesn’t matter (up to a
sign for fermions). But at the same point, it matters if there is a particle there already.
For example, if there is no particle there then ψ † ψ = 0, but ψψ † 6= 0.
Now we can make a position eigenstate of n particles (let’s forget about spin for a
moment):
1
|r1 r2 · · · rn i = √ ψ † (rn ) · · · ψ † (r2 )ψ † (r1 ) |0i (1.43)
n!
= ζ |r2 r1 · · · rn i .
51
do
1
ψ(r) |r1 · · · rn i = √ ψ(r)ψ † (rn ) · · · ψ † (r2 )ψ † (r1 ) |0i (1.44)
n!
1
δ(r − rn ) + ζψ † (rn )ψ(r) ψ † (rn−1 ) · · · ψ † (r1 ) |0i
=√ (1.45)
n!
= ··· (1.46)
1
= √ δ(r − rn ) |r1 · · · rn−1 i + ζδ(r − rn−1 ) |r1 · · · rn−2 rn i + · · · ζ n−1 δ(r − r1 ) |r2 · · · rn i .
n
(1.47)
So this is only nonzero if the position r at which we try to remove the particle is one
of the positions ri of the particles that are present. If it is, then we get a perfectly
(anti)symmetrized n − 1-particle state.
A similar calculation tells us the inner product between these states:
δnn0 X π
hr10 · · · rn0 |r1 · · · rn i = ζ δ(r1 − rπ0 1 )δ(r2 − rπ0 2 ) · · · δ(rn − rπ0 n ) .
n! π∈S
n
Warning: Actually the state (1.43) is not normalized. Consider for example n = 2
with r1 6= r2 , where
1 (1.41)(1.42) 1
hr1 r2 |r1 r2 i = h0| ψ(r2 ) ψ(r1 )ψ † (r1 ) ψ † (r2 ) |0i = . (1.48)
2 | {z } 2
=ζψ † (r1 )ψ(r1 )+1
The reason is basically that |r1 r2 i and |r2 r1 i are not different and each count as half a
state in the resolution of the identity we’ll find below – they aren’t different states! If we
wanted to, we could normalize this state to one and try to sum only over r1 ≤ r2 . This
requires putting an order on the points and is annoying. Feynman (in his wonderful
Lectures on Statistical
√ Mechanics, §6.7), for example, chooses to normalize his states
without the n! and put a 1/n! in his resolution of the identity. However, his states
are then not normalized when e.g. r1 = r2 .
How to write the n-particle state |Ψi whose wavefunction of the n particles is
Ψ(r1 · · · rn ) ? This is easy:
X
|Ψi = Ψ(r1 · · · rn ) |r1 · · · rn i .
r1 ···rn
Notice that this state is correctly (anti)symmetrized, even if the function Ψ(r1 · · · rn )
is not; the un-(anti)symmetric part simply disappears. The amplitude in this state for
52
observing the particles at r10 · · · rn0 is
X
hr10 · · · rn0 |Ψi = Ψ(r1 · · · rn ) hr10 · · · rn0 |r1 · · · rn i (1.49)
r1 ···rn
1 X π
= ζ Ψ(rπ0 1 , rπ0 2 · · · rπ0 n ) (1.50)
n! π
if Ψ is (anti)symmetric: = Ψ(r10 · · · rn0 ). (1.51)
(The overlap with an n0 -particle state for n 6= n0 is zero because an extra annihilation
operator will hit |0i or an extra creation operator will hit h0|.) Since this is true for all
hΦ|, we can just erase the Φ and get
X
|Ψi = |r1 · · · rn ihr1 · · · rn |Ψi.
r1 ···rn
Operators on the Fock space. Ignore spin for a moment. I claim that ρ(r) ≡
†
ψ (r)ψ(r) is the density of particles at r. The idea is simply: try to remove one, and
if you find one there, put it back (otherwise you get zero). Here is a more formal
53
demonstration of this claim. If Ψ, Ψ0 are n-particle states,
~ ipr ),
we have (using p~eipr = −i∇e
~j(r) = 1 ψ † (r)∇ψ(r)
~ ~ † (r) ψ(r)
− ∇ψ
2mi
and the spin density is
~
X ~σss0
S(r) = ψs† (r) ψs0 (r),
ss0
2
54
where ~σ are the Pauli matrices.
Now I can explain the origin of the weird and misleading name ‘second quan-
tization’. The above expressions bear some visual similarity with the corresponding
single-particle expressions, were I to mistake the field operator ψ(r) for a single-particle
wavefunction. So it looks as if we’ve taken the single-particle QM expressions for vari-
ous observables and in each of them replaced the wavefunction with an operator. That’s
not actually what we did!
The many body hamiltonian for a collection of non-interacting particles in a poten-
p2
tial (for which the one-body hamiltonian is H = 2m + V (r)) is then
!
X ∇ψ ~ † · ∇ψ
~
†
H= + ψ (r)ψ(r)V (r) . (1.58)
r
2m
Why do I call this non-interacting? Well, notice that each of these operators we’ve
written is of the form ψ † ψ – whenever it removes a particle, it puts it right back. An
interaction is something involving more powers of ψ, so that the doings of one particle
have an influence on those of another.
Interactions. [Feynman §6.8] For example, we might wish to consider a 2-body
interaction V (2) (xi , yi ), which acts within the 2-particle Hilbert space as
1 XX
V (2) = |x, yihx, y|V (2) (x, y).
2 x y
e2
An example to think about is the Coulomb potential, V (2) (x, y) = |x−y|
. We want to
find an operator that acts on an n-particle state as
1 X (2)
V̂ |r1 · · · rn i = V (ri , rj ) |r1 · · · rn i . (1.59)
2 ij
where ρ(x) ≡ ψ † (x)ψ(x) is the particle density operator. This is almost correct. How-
ever
ρ(x)ρ(y) = ψx† ψx ψy† ψy = ζψx† ψy† ψx ψy + δxy ψx† ψy = ψx† ψy† ψy ψx + δxy ρ(x)
it is not normal-ordered. Normal-ordered means all annihilation operators are to the
right of all creation operators. This is a good feature because it will mean that there
is no part of the operator that is secretly a single-particle operator. The correct V̂
1 X (2)
V̂ = V̂guess − V (x, x)ρ(x)
2 x
55
subtracts off this ‘self-energy’ term.
Let’s check that this definition
1X † †
V̂ = ψ ψ ψy ψx V (2) (x, y)
2 xy x y
All of the signs cancel out when putting the particles back in this order. Therefore
1 X (2) 1 X (2)
V (x, y)ψx† ψy† ψy ψx |r1 · · · rn i = V (ri , rj ) |r1 · · · rn i
2 xy 2 i6=j
as we asked for. Notice that this is a true interaction, in the sense that it vanishes if
the number of particles is less than two.
We will learn more about interacting many-body systems. There’s actually quite a
bit more to say about non-interacting systems first.
[still Baym, chapter 19] Let’s return to the groundstate of a Fermi gas. Letting the
fermions move wherever they want, and not just on particular orbitals can be achieved
14
Beware the conventions for the order of the operators in |r1 · · · rn i ∝ ψ(rn )† · · · ψ(r1 )† |0i. With
this convention, the operator ψx has to move through n − i − 1 of its friends to get to ψ(ri )† .
56
by taking the continuum limit of the model we studied in §1.7.1. This involves two
steps: take N → ∞, so the momentum space becomes continuous, and focus on low
p2
energies – Taylor expand about the minimum of the band, where (p) = 2m + O(p4 ),
and ignore the p4 terms.
Alternatively, we could just start from scratch, using the non-interacting many-
body Hamiltonian (1.58), and set V = 0. Either way, the single-particle levels are
p2
(p) = 2m , and in the groundstate |Ψ0 i we fill them from the bottom until we run out
of fermions. The occupation numbers are
(
1, |p| < pF
nps ≡ hΦ0 | a†ps aps |Φ0 i = . (1.65)
0, |p| > pF
57
Now recall the definition of pressure15 : at zero temperature,
dE0
P =− |N . (1.70)
dV
This gives
2
2/3 !
3 N 3π N 3 2 1 5/3 2 E0
P = −∂V |N =+ N (3π 2 )2/3 V −5/3 = .
5 2m V 5 3 2m 3V
where nps is the groundstate occupation number (1.65). Indeed the density is uniform,
and therefore equal to the average density.
Now consider the similar but more interesting quantity, sometimes called the one-
particle density matrix, which depends on two positions:
15
If you like, you can think about it thermodynamically,
dE = T dS − P dV + µdN.
58
Gs (r − r0 ) ≡ Φ0 |ψs† (r)ψs (r0 )|Φ0 (1.71)
1 X −ip·(r−r0 )
= e nps (1.72)
V p
Z pF
0
= d̄d pe−ip·(r−r ) (1.73)
0
Z pF Z 1
d=3 1 0
= 2
dpp 2
dµe−ip|r−r |µ (1.74)
4π 0
| −1 {z }
sin p|r−r 0 |
=2 p|r−r 0 |
3n sin x − x cos x
= (1.75)
2 x3
Of course this is zero if there is no such particle at r. Then in this state, we ask about
16
A number of people have asked me about the physical significance of Gs (r). One answer is that
in some systems it can be measured by angle-resolved photoemission. This is an experiment where
we kick a particle out of the system (for example by hitting it with a high-energy photon with known
momentum and energy k, E), then we measure the momentum and energy p, ω of the particle we
kicked out. The intensity as a function of (k − p, E − ω) is the imaginary part of the Fourier transform
of G. You can see this by using Fermi’s golden rule.
59
the expected density at r0 (which is the probability to find a particle at r017 ):
hΦ0 | a†ps a†qs0 aq0 s0 ap0 s |Φ0 i = δpp0 δqq0 s2 hΦ0 | a†ps aps a†qs0 aqs0 |Φ0 i = δpp0 δqq0 nps nqs0
Therefore, for s 6= s0 ,
2 2
0 2 1 X 2 1
gss0 (r − r ) = 2
nps nqs0 = Ns Ns0 = 1,
n V pq n V2
since Ns = N2 for each spin. This is the classical behavior of non-interacting particles
– the locations of particles of different spins are uncorrelated.
Now consider s = s0 . Then there are two ways to put back what we took out:
p = p0 , q = q 0 or p = q 0 , q = p0 . So in this case
hΦ0 | a†ps a†qs aq0 s ap0 s |Φ0 i = δpp0 δqq0 a†ps a†qs aqs aps + δpq0 δqp0 a†ps a†qs aps aqs (1.83)
= (δpp0 δqq0 − δpq0 δqp0 ) a†ps a†ps aqs aqs . (1.84)
| {z }
=nps nqs
17
Actually this step is maybe not obvious. Here’s the idea: if Φ is an N -particle state, then
X
ProbΦ (particle at r0 ) = N | hr0 r2 · · · rN |Φi |2 (1.76)
r2 ···rN
!
X
= N hΦ| |r0 r2 · · · rn ihr0 r2 · · · rN | |Φi (1.77)
r2 ···rN
N X
= hΦ| ψ † (r0 ) |r2 · · · rN ihr2 · · · rN | ψ(r0 ) |Φi (1.78)
N r2 ···rN
| {z }
=1 N −1
The factor of N in the first line is because the particle we find could be any of the N particles (as in
R
(1.28)). Note that drProbΦ (r) = N , but this is OK, since finding particles at different places are
not exclusive events! (Note that if r is continuous then by ‘Prob’ I mean probability density.)
60
In the underbrace we used {aps , a†qs } = δp,q . But the contribution from p = q cancels
between the two terms by Fermi statistics. Therefore for like spins,
n 2 1 X 0
n 2
2
gss (r − r0 ) = 2 1 − e−i(p−q)·(r−r ) nps nqs = − (Gs (r − r0 )) (1.85)
2 V pq 2
and therefore
9
gss (r − r0 ) = 1 − 6
(sin x − x cos x)2 , x ≡ pF |r − r0 |,
x
which is plotted below:
In the right plot, we zoom in to show the oscillations with period π/pF . The key point,
though, is that there is a big trough around r = r0 which is a consequence of the Pauli
principle combined with the kinetic energy wanting to make the wavefunctions smooth.
Compare this plot to the analogous plot for a perfect solid. In that case g(r) would
have peaks whenever |r| = a lattice vector, and zero otherwise. This is because the
location of one atom determines all the rest. Now consider a liquid made of particles
with short-range repulsion. Then again g(r) = 0 when r < R where R is the hard core
radius of the particles. There will be a peak in g(r) just past R, since the particles are
trying to be close together in a liquid. After some wiggles, the answer will approach the
uniform answer because the correlations between the positions are only short-ranged.
It generally looks like the plot above! Fermions achieve such correlations without any
interactions.
There is more to say about correlation functions of the form we’ve been discussing, to
illustrate the contrast between bosons and fermions. But first an important general
comment which will also establish some notation.
Non-interacting/Gaussian/Free systems. So far, we have been studying a
very special class of quantum many-body systems, where the Hamiltonian is of the
61
form
D
X
H= a†i aj hij (i, j = 1 · · · D = dim H1 ) (1.86)
ij
The calculation we did of the pair correlation function is a special case of a very useful
fact about correlation functions of free fermions (but not free bosons, as we’ll see).
18
Notice that uiα ≡ uα (i) is a unitary matrix since the eigenvectors are orthonormal.
62
Consider correlation functions in a state |ψi of the form (1.87), which is an eigen-
state of our free Hamiltonian (not necessarily the groundstate). Here ax ≡ ψ(x) is just
another name for the field operator, but could be in an arbitrary basis; α labels the
eigenbasis of hxy .
X X
Gψ (x, y) = hψ| a†x ay |ψi = hψ| a†α aβ |ψi u?α (x)uβ (y) = u?α (x)uα (y).
αβ α∈ψ
| {z }
=δαβ nα
This is only nonzero if we put back what we take out, so α = β. What I mean by
α ∈ ψ is that the state α is one of the single-particle states {1 · · · n} that are occupied
in the state |ψi.
Now a more general 2-particle correlator:
X
C ≡ hψ| a†x a†x0 ay ay0 |ψi = hψ| a†α a†β aγ aδ |ψi u?α (x)u?β (x0 )uγ (y)uδ (y 0 ).
αβγδ
To evaluate it, we expand the field operators in terms of the single-particle eigenstates.
To be nonzero, again we must put back what we took out, so we need α = γ, β = δ or
α = δ, β = γ, and for fermions we must have α 6= β. Therefore
X
C= hu1 · · · un | a†α a†β aα aβ |u1 · · · un i u?α (x)u?β (x0 )uα (y)uβ (y 0 ) (1.88)
α6=β
+ hu1 · · · un | a†α a†β aβ aα |u1 · · · un i u?α (x)u?β (x0 )uβ (y)uα (y 0 ) (1.89)
X
−u?α (x)uα (y)u?β (x0 )uα (y 0 ) + u?α (x)uα (y 0 )u?β (x0 )uα (y)
=
α6=β∈ψ
! ! ! !
X X X X
=− u?α (x)uα (y) u?β (x0 )uβ (y 0 ) + u?α (x)uα (y 0 ) u?β (x0 )uβ (y)
α∈ψ β∈ψ α∈ψ β∈ψ
(1.90)
D E D E D E
= a†x ay0 a†x0 ay − a†x ay ψ
a†x0 ay0 (1.91)
ψ ψ ψ
where in the step (1.89) we used the sneaky trick of adding back in the term with
α = β because it cancels between the two terms.
This was a complicated calculation, but now that we’ve done it, the structure of
the answer is simple to understand and to generalize to arbitrary n-point functions:
A correlation function of fermion operators (in a state of the form ψ) is equal to the
sum of contractions, weighted by a (−1)# of crossings .
63
A contraction is just a way of pairing up the creation and annihilation operators in the
correlator: draw a line connecting
D E each pair ij, and replace the pair by its two-point
†
function in the same state, ai aj . By the number of crossings, I mean the number
ψ
of times these lines cross each other on the piece of paper; this is the number of times
we must move fermionic operators through each other.
In this example, we have
* + * +
D E
a†x a†x0 ay ay0 = a†x a†x0 ay ay0 + a†x a†x0 ay ay0 .
ψ
ψ ψ
The second contraction involves one crossing, and so comes with a (−1), and this
reproduces (1.91).
This is a generalization to fermions of the familiar idea that in a gaussian ensemble,
all correlations are products of two-point functions (‘covariances’).
In what sense is the state |ψi gaussian? Well, correlations in the state |ψi, are the
same as
hψ|O|ψi = trOρ (1.92)
where ρ = limβ→∞ e−βH /Z, and H is a quadratic operator.
Now let’s ask the same questions about a collection of bosons, again in a state of the
form
1
|ψi ≡ Q p b† (u1 ) · · · b† (un ) |0i , where b(uα ) |0i = 0. (1.93)
ψ
α n α !
To emphasize the difference, I’ve called the bosonic mode operators b. In the case of
bosons, some of the uα in the state |ψi may be the same; I’ll call nψα the occupation
number of the mode α in the state ψ, i.e.
n!
|0i. Recall that 0! = 1.
First let’s think about two-point functions:
X X
b†x by ψ ≡ hψ| b†x by |ψi = u?α (x)uβ (y) hψ| b†α bβ |ψi = u?α (x)uα (y)nψα .
αβ α
| {z }
(1.94)
= δαβ nψ
α
64
This is just like the fermion case above, with the only difference that nα can be larger
than one.
Let’s do the general case first:
X
B ≡ hψ| b†x b†x0 by by0 |ψi = hψ| b†α b†β bγ bδ |ψi u?α (x)u?β (x0 )uγ (y)uδ (y 0 ) . (1.95)
αβγδ
Now there are three ways to put back what we took out:
(i) α = β = γ = δ
(ii) α 6= β, α = γ, β = δ
(iii) α 6= β, α = δ, β = γ. This leads to three terms
X 2 2
B= hψ| b†α bα |ψi u?α (x)u?α (x0 )uα (y)uα (y 0 ) (i)
α
| {z }
=nψ ψ
α (nα −1)
X
+ hψ| b†α b†β bα bβ |ψi u?α (x)u?β (x0 )uα (y)uβ (y 0 ) (ii)
α6=β
| {z }
=nψ ψ
α nβ
X
+ hψ| b†α b†β bβ bα |ψi u?α (x)u?β (x0 )uβ (y)uα (y 0 ) (iii) .
α6=β
| {z }
=nψ ψ
α nβ
Therefore
! ! ! !
X X X X
B= nα u?α (x)uα (y) nβ u?β (x0 )uβ (y 0 ) + nα u?α (x)uα (y 0 ) nβ u?β (x0 )uβ (y)
α β α β
| {z }| {z } | {z }| {z }
b†x by
D E
=h iψ = b†x by0 =hb†x0 by i
D E
= b†x0 by0 ψ ψ
ψ
X
− nα (nα + 1)u?α (x)u?α (x0 )uα (y)uα (y 0 ). (1.97)
α
The first line is from terms (ii) and (iii), and the second line is from (i) plus the second
term in (1.96). The first line would be the result from the naive bosonic version of
Wick’s theorem. The second line is an extra term which is not present for fermions. So
Wick’s theorem is not true for a general permanent of bosons. An important exception
correlations in the vacuum, where all the nα = 0; then the extra term vanishes. (This
special case comes up a lot in relativistic QFT.)
Pair correlator for (spinless) bosons. A special case of the expression B =
B(x, y, x0 , y 0 ) is the pair correlation function, when the eigenmodes (for which so far
65
we’ve been using the abstract labels α, β... ) are momentum eigenstates. So consider a
state of (non-interacting) spinless bosons in free space, in a box,
By the same argument as above, the pair correlator (proportional to the probability of
finding a particle at y given that we already found one at x) is
The labelling of terms is the same as in the calculation above. The 1 − δpq s are there
to avoid overcounting. So
!
1 X X X X X
n2 gB (x − y) = 2 np nq + np e−ip·(x−y) nq e−iq·(y−x) − np (np + 1)
V p q p q p
(1.101)
2
1 X 1 X
= n2 + np e−ip·(x−y) − 2 np (np + 1). (1.102)
V p V p
Comparing this expression with the result for fermions (1.85), there are two differences.
The oscillating (second) term has a plus sign for bosons, and there is an extra additive
constant, the correction relative to the Wick result.
To see some physical consequences, consider two example states:
1. In the groundstate of free bosons, all N bosons are in the same (the lowest-energy)
single-particle level, |Φi = |np0 = N i. Then we get
N (N + 1) N (N − 1)
n2 gB (x − y) = n2 + n2 − 2
= .
V V2
66
Notice that if we take V → ∞ at fixed density n, we can ignore the correction to
Wick. Later we will see some dramatic physical consequences of such states.
density n):
Z
1 −ip·(x−y) 2 2
2
gB (x − y) ' 2 n + | d̄ pnp e3
| = 1 + e−(x−y) /α .
n
Independent of the width α of the beam, the value at x = y is twice the value for
distinguishable particles. Bosons tend to clump together!
Ingredients:
67
Recipe: The arrangement of detectors and beam-splitter is
shown at right. Measure (by sampling) the probability of
observing a photon with detector 2 at time t + τ given that
we have seen a photon with detector 1 at time t:
Prob observe a photon at detector 2 at time t + τ observe a photon at detector 1 at time t
∝ I1 (t)I2 (t + τ ).
Here I1,2 are the intensities of the beam at the two detectors. τ is a time delay that
we get to pick. When collecting samples over some time interval, the thing we actually
measure is the time average
Z
I1 I2 ≡ dtI1 (t)I2 (t + τ ) = gB (cτ ).
I claim that this will take the form of the pair correlator we calculated above for a
beam of bosons.
This is because light is a beam of bosons.
In case you don’t believe that argument, here is a more direct calculation, in a
model that also includes some information about the source. Suppose there are two
sources (for example, two locations on the star) A and B, each of which emits a plane
wave of wavenumber kA/B with a random phase. This random phase accounts for the
fact that the photons are coming from totally uncorrelated, incoherent sources.
The amplitudes of light at the two detectors
are then
Ii = |α|2 + |β|2
because the random phases α, β depend on time and so any term not of the form |α|2
or |β|2 averages to zero.
68
Consider instead
2
I1 I2 = |A1 A2 |2 = α2 eikA (r1A +r2A ) + β 2 eikB (r1B +r2B ) + αβ eikA r1A +ikB r2B + eikB r1B +ikA r2A
.
Some words: the α2 term is the amplitude for both photons to come from A, the β 2 is
where both come from B. The interesting αβ term is where one photon comes from A
and one comes from B; since photons are indistinguishable bosons, there are two ways
for this to happen (A → 1, B → 2 or A → 2, B → 1), and these two paths interfere.
Now time-average:
2
I1 I2 = |α|4 + |β|4 + |αβ|2 eikA r1A +ikB r2B + eikB r1B +ikA r2A (1.104)
2 2
= I1 · I2 + 2|α| |β| cos (kB (r1B − r2B ) − kA (r1A − r2A )) (1.105)
r1A −r2A 'r1B −r2B ≡r
' I1 · I2 + 2|α|2 |β|2 cos ((kB − kA )r) . (1.106)
In the last step, we considered the case where the two sources are close together com-
pared to their separation from the detectors, as in the case of parts of a distant star.
Suppose that the sources emit plane waves αk eik·r with amplitude |αk | = |βk |, with
2
probability p(k) = ce−αk /2 (normalization fixes c). Then we should average over the
colors in the beam:
Z Z Z Z
I1 I2 2 /2−αk 2 /2
−αkA 1 i(kB −kA )·r 1 2
3 3
d̄ kA p(kA ) d̄ kB p(kB ) =c 2 3 3
d̄ kA d̄ kB e B 1 + Ree = 1+ e−r /α
I1 · I2 4 2
as promised.
More generally, if instead of two discrete sources, we have a distribution of sources
ρ(r), the result is
Z 2
I1 I2 ~ ~
=1+ d3 rρ(r)ei(k1 −k2 )·~r .
I1 · I2
The fact that this is (absolute value of the) the Fourier transform of the source distri-
bution means that we can learn about the source distribution!
Notice that we could have replaced this whole discussion with a classical one. In-
~ ⊥ . The
stead of amplitudes in (1.103) I could have written electric field strengths E
calculations are continuously connected because Bose statistics allows for coherent
states.
This beautiful idea was first used by Hanbury-Brown and Twiss to measure the
sizes of very distant stars by varying the distance between two radio telescopes. A nice
review of its many applications (including in high energy physics and in condensed
matter) is this one by Baym. This method has been used to measure, for example, the
size of viruses. [End of Lecture 10]
69
2 Interference phenomena
Some of the things I’ve said above are unified by the description of quantum mechanics
in terms of path integral. It is the quantum mechanical analog of Lagrangian mechanics.
It describes, for example, the propagator, as a sum of phases, each associated with a
i R
possible path. The phase for a path is e ~ S[path] , where S = dtL is the action.
Why should there be such a representation? It is the logical conclusion of the
double-slit experiment. A wall with a double-slit restricts a particle to pass through
one of two locations. The conclusion of the experiment is that the amplitude is the
sum of the contributions of the two paths, each of which is a complex number.
Now suppose that instead of the detector, we put another wall with two slits. To
find the probability amplitude for a detector placed after the second wall, we would
have two sums, one over the path through first wall, and another over the path through
the second wall. You see that we can keep adding walls.
Now suppose instead of two slits, each wall has three slits. Then there would be
three terms in each sum.
Now imagine adding more walls, each with more holes.
Finally, imagine that the world is totally full of walls and that all the walls are
totally full of holes. Even if there is no wall, we must sum over the paths. 19
The more precise derivation of this statement is straightforward, in that it involves
inserting lots of resolutions of the identity, but it is a little annoying and takes most of
a lecture. Here are some payoffs:
• The fact that the vector potential term in the action is geometric makes the A-B
effect very natural.
• It gives a straightforward way to derive the propagator for the harmonic oscillator.
• In quantum systems with many degrees of freedom, such as quantum field theo-
ries, the path integral is more useful, especially as an organizing device.
• Path integrals, with fewer factors of i, are also useful in statistical physics. The
derivation from quantum mechanics gives a useful perspective on them.
19
This wonderful device is due to Feynman. Take a look at Volume III of the Feynman Lectures.
A more elaborate treatment appears in Feynman and Hibbs, Quantum Mechanics and Path Integrals.
70
Propagator. Recall that the time evolution operator is the unitary U(tb , ta ) that
satisfies U(ta , ta ) = 1 and
For a free particle, V = 0, this is diffusion in imaginary time, and is solved by Fourier
transform:
r
~2 ∂x2 m im(x−xa )2
i~∂t K(x, t; xa , ta ) = − K(x, t; xa , ta ) =⇒ K(x, t; xa , ta ) = e 2t
2m 2πit
(I set ~ = 1 in the last expression.)
Since waiting a bit and waiting some more is the same as waiting a lot, the time
evolution operators satisfy a nice composition property:
for any t (which need not even be in between tb , ta ). If H is time independent, this
just comes from adding the exponents. But we can stick a resolution of the identity in
between the two factors on the RHS:
Z
U(tb , ta ) = dxU(tb , t)|xihx|U(t, ta )
71
and make a sandwich of the whole thing:
Z
K(xb , tb ; xa , ta ) = dxK(xb , tb ; x, t)K(x, t; xa , ta ).
This is the operation of placing a wall full of holes along the path of the particle.
Path integral representation of propagator. Now do this N − 1 times: divide
−ta
the interval into N equal segments of (tiny) size ∆t = tb N :
Z
K(xb , tb ; xa , ta ) = dx1 · · · dxN −1 K(xb , tb ; xN −1 , tN −1 ) · · · K(x1 , t1 ; xa , ta ).
with L(t, x, ẋ) = 12 mẋ2 − V (x). We see that the true role in life of ~ is to make up
the units of action in this expression. The first expression here defines what we mean
R
by the path integral √ [dx]. It includes horrible things which are not differentiable at
all (they have ∆x ∼ ∆t, not ∆x ∼ ∆t); when this matters, you need to use the
discretization to make it well-defined.
Stationary phase and classical mechanics. The path integral explains the
special role of configurations that solve the equations of motion. Recall that
Z Z s
N 00 2 2π
I = dx e−N f (x) = e−N f (x0 ) dye− 2 f (x0 )y +··· ' e−N f (x0 )
N f 00 (x0 )
where f 0 (x0 ) = 0 and y ≡ x − x0 . This is called saddle point approximation, and it
holds because other contributions are suppressed by e−N ∆f . If instead we have
Z Z s
N 00 2 2πi
I˜ = dx eiN f (x) = e−iN f (x0 ) dyei 2 f (x0 )y +··· ' e−iN f (x0 )
N f 00 (x0 )
72
where f 0 (x0 ) = 0 and y ≡ x − x0 , this is called stationary phase. It holds because other
contributions add destructively. With more variables:
Z s
iN f (x1 ···xN ) iN f (x) π
dx1 · · · dxN e 'e det .
ij iN ∂i ∂j f
where
0 = ∂xj f (x)|x=x ∀j. (2.2)
The path integral is just of the previous form with ~ = N1 and f (x) = S[x] =
R
dtL(t, x, ẋ), the action functional. The square brackets indicate that the argument x
is a whole function, rather than just a finite collection of numbers, but basically it just
means that N is big. The saddle point condition (2.2) can be rewritten in the fancy
notation:
δ
0= S[x] , ∀t. (2.3)
δx(t)
What is this when S[x] = ds 12 mẋ2 (s) − V (x(s)) ? The key ingredient is the ‘fun-
R
73
Now plugging in xj = xj + yj ,
(xj − xj−1 )2 = (x(tj ) − x(tj−1 ))2 +(yj − yj−1 )2 + terms linear in δxj
| {z }
→S[x]
we can ignore the terms linear in δx because they are zero by definition of the classical
path. Z
m N/2 i P
S[x]
K(xb , tb ; xa , ta ) = e ~ dy1 · · · dyN −1 ei jk yj Mjk yk
2πi∆t
| {z }
N −1
2
= √π
det(−iM )
follows from the block structure of the matrix. For u = 0, it’s just dN = 2dN −1 − dN −2 .
We need two initial conditions to solve this two-step recursion: det M1 = 2 cosh u → 2
74
for u → 0. det M2 = 4 cosh2 u − 1 → 3 for u → 0. The solution for u = 0 is simple:
dN −1 = dN −2 + 1, hence dN −1 = N . (For general u, it is dN = sinh(N +1)u
sinh u
.)
m N/2 i N −1
S[x] π 2
K(xb , tb ; xa , ta ) = e~ p
2πi∆t det (−iM )
m N/2 i m −(N −1)/2
= e ~ S[x] N −1/2
r2πi∆t 2πi∆t
m i
= e ~ S[x] . (2.5)
2π(tb − ta )i
Recall that coherent states have a |αi = α |αi. The coherent-state representation
of the propagator is
K(αb , tb ; αa , ta ) ≡ hαb | U(tb , ta ) |αa i .
They resolve the identity by
Z
dReαdImα
1= |αi hα|
π
and their overlaps are
1
hα|α0 i = e− 2 (|α|
2 +|α0 |2
)+α? α0 .
This is useful if H acts nicely on the coherent states, as is the case when it can be
written in terms of creation and destruction operators. For simplicity consider the (cop
out!) case of the SHO:
†
K(αj , tj ; αj−1 , tj−1 ) = hαj | e−i∆tωa a |αj−1 i
?
= e−i∆tωαj αj−1 hαj |αj−1 i + O(∆t2 )
75
−i∆tωα?j αj +O(∆t2 ) 1 2 1 2 ?
=e exp − |αj | − |αj−1 | + αj αj−1 (2.6)
2 2
| {z }
? ?
= 12 α?j (αj − αj−1 ) + 12 αj−1 −αj−1 + αj
| {z } | {z }
=α̇∆t =−α̇? ∆t
−1 2
Z NY
d αj i Rttb dt( 2i (α? α̇−αα̇? )−ωα? α)
K(αb , tb ; αa , ta ) = e a
π
Z j=1 R
= [dα]ei dtL(α,α̇) (2.7)
with L = pẋ − H.
For the special case of the SHO, recall that we know the answer since coherent
states remain coherent under time evolution by the SHO hamiltonian:
†
U(t) |αi = Nα U(t)ea α U† (t) U(t) |0i = e−iωt/2 αe−iωt .
| {z } | {z }
† αe−iωt =e−iωt/2 |0i
=ea
So
2 /2−|α|2 /2+α? αe−iωt
K(αb , tb ; α, 0) = hαb | U(t) |αi = e−iωt/2 αb |αe−iωt = e−iωt/2 e−|αb | b .
The situation when this comes into its own is when we have a collection of SHOs:
X
H= ~ωk a†k ak + · · ·
k
76
R
Doing the same thing to prepare h0|, making a sandwich of eq(0)·k = e dτ q(τ )·kδ(τ ) , and
taking T → ∞ we can forget about the arbitrary states at the end, and we arrive at
Z
1 R∞
h0| Ô(q, p) |0i = [Dq]e− −∞ dτ L(q(τ ),q̇(τ ) O(q(0), mq̇(0)). (2.8)
Z
On the LHS Ô is some operator; on the RHS O is just a function of the integration
variables. Z is a normalization factor chosen so that h0|0i = 1.
For example, consider the case of our chain of balls on springs:
Z Y
1
eiKq
0
= dqi e−qi Dij qj eiKq0 (2.9)
Z i
−qi Dij qj
RQ
with Z = i dqi e . Here i, j are discrete time labels, and Dij is the matrix
which discretizes the action. Repeated indices are summed.
Notice that in this way we can find a path integral representation for the ther-
mal partition function tre−βH : simply impose periodic boundary conditions on the
configurations over which we sum: q(τ ) = q(τ + β):
I Rβ
−βH
Z = tre = [Dq]q(0)=q(β) e− 0 dτ L . (2.10)
A fancy way of describing this is to say that putting the system at finite temperature
is accomplished by making the imaginary time direction into a circle.
When we talked about the radiation field I mentioned that one motivation for intro-
~ such that
ducing the electromagnetic potentials, Φ, A,
E ~ − 1 ∂t A,
~ = −∇Φ ~ ~ =∇
B ~ × A,
~
c
is that they are required to discuss the quantum mechanics of a charged particle moving
in an electromagnetic field (E,~ B).
~
~→A
~ − ∇λ,
~ 1
A Φ → Φ − ∂t λ. (2.11)
c
These expressions are nicely packaged in terms of the four-vector potential Aµ ≡
~ cΦ)µ → Aµ − ∂µ λ, but we will not consider any relativistic situation here. This
(A,
ambiguity may make it seem like the vector potential is not real. Some of the vector
77
~ · d~x is not changed by the gauge
H
potential is indeed an artifice, but notice that C A
transformation (2.11). We will now see how this quantity arises physically, because of
quantum mechanics.
It is useful to think about the Lagrangian description of the charged particle, which
has
1 e
L = mẋ2 − eΦ − ~x˙ · A~. (2.12)
2 R c
Notice that the extra term in the action S = dtL from the vector potential is geo-
metric, in the sense that dt~x˙ · A
~ = d~x · A
~ depends only on the path of the particle,
R R
Now we turn to showing that the vector potential (or at least its integral around
closed curves) is real, via the Aharonov-Bohm effect.
78
We send quantum mechanical charged particles from the source at left to the de-
tector at right. They cannot enter the solenoid, but are otherwise free to proceed as
~ 6= 0. But we will show that the field
they like. So they never move in a region where B
affects them.
The path integral is very helpful here: We can group the possible paths (over which
we integrate to find e.g. the amplitude Z for the particle to hit the detector) into two
disconnected components: they either go above the solenoid or below it20 :
Z Z Z
Z = [dx]... = [dx]u ... + [dx]d ...
i R
The ... here is e ~ S[x] with S[x] the action for a charged particle above, S = dtL,
with L from (2.12). In particular it has a term
e 2 e 2~ e t2 dxi
Z Z Z
SA = A≡ A(x) · d~x = Ai dt
c 1 c 1 c t1 dt
~ ×A
The vector potential is curl-free ∇ ~ = 0 in the region outside the solenoid – this
is what it means that B~ = 0. This means that the line integral SA is path-independent.
21
R
For all the paths that go above the solenoid, the phase factor resulting from the A
term in the action is the same:
e
R
ei ~c Cu A
for some fixed reference path Cu (in the figure). Similarly, all the paths below give a
e
R
i ~c Cd A
phase factor e . But these phase factors from above and below are not the same:
if we deform the path through the solenoid, we have to go through the region with
B~ 6= 0.
20
Actually there are also paths that wind around the solenoid some number of times. These have
very large action and we will ignore them here.
21
Recall that the difference between two paths is
Z Z Z Z Z
A~ · d~x − ~ · d~x =
A A= ~ ×A
∇ ~ · d~a = ~ · d~a
B
C1 C2 C1 −C2 D D
~
by Stokes’ theorem, where D is the region bounded by the paths. So as long as the B-field vanishes
in D, we can change the path without changing the phase.
79
where Zu,d don’t depend on the EM field at all.
But now
Z Z Z Z
A− A = A= ~ · d~a = Bz A ≡ Φ .
B
Cu Cd Cu −Cd inside
Next, let’s consider a double-slit experiment, where a particle may follow two pos-
sible paths from a source to a detector, which we’ll call u(pper) and l(ower). The
intensity at position x on the detector is
where ψu/l (x) is the amplitude for the case where the l/u path is blocked.
Now here’s the real thing. Consider a charged particle traversing the same double-
slit experiment, where now the (completely impenetrable) wall between the slits con-
tains a (perfect) solenoid, with
(
~ = B ž, inside
B .
0, outside
80
The Schrödinger equation is
1 ~ − e A(x)
~
2
− i~∂t ψ(x, t) = −i~∇ ψ(x, t) + V (x)ψ(x, t) (2.15)
2m c
~ A
where V includes the walls of the barrier, and ∇× ~ = B.
~ Notice that the wavefunction
vanishes wherever the magnetic field is nonzero (since the walls are impenetrable (V =
∞ inside the walls)) and the magnetic field vanishes wherever the wavefunction is
nonzero (since the solenoid is perfect (B = 0 outside the solenoid)).
Now, suppose we know the solution u0 for A~ = 0:
1 2
~ + 0 u0 (x, t) + V (x)u0 (x, t) .
− i~∂t u0 (x, t) = −i~∇ (2.16)
2m
This might be quite difficult to find, but we don’t need to find it. I claim that the
solution to (2.15) is
i e x d~ ~
R
`A(`)
ψ = e ~c x0 u0 ≡ eg u0 .
Notice the similarity with (2.14). To check this, let’s differentiate:
~ e ~ FTC
~
g ie ~ g e ~ g
−i~∇ − A ψ(x) = −i~ ∇u0 e − i~u0 A e − Au0 e .
c ~c c
Doing it again:
2
1 ~ e ~ 2 g 1 ~
−i~∇ − A (u0 e ) = −i~∇ u0 eg .
2m c 2m
Rx
Notice that we didn’t have to choose a path in A.
The combination D ≡ −i~∇ ~ − eA ~ is sometimes called a covariant derivative,
c
because it preserves the gauge transformation properties of the function it acts on:
~ ~
~
ie
H
ie ~
R
A(`)·d` A(`)·d`
=e ~c Cu uu0 (x) + ul0 (x)e ~c Cl −Cu
. (2.17)
81
~ over a closed
The relative phase between the two terms is changed by the integral of A
path: I Z
e ~ ~ Stokes e ~ ×A ~ · d~a ≡ e ΦB (S) .
A(`) · d` = ∇
~c Cl −Cu ~c S | {z } |{z}
2
~c
~
=B ≡ňd x
Here S is any surface bounded by ∂S = Cl − Cu , and the magnetic flux through the
surface is ΦB (S) = BA in the situation above.
Notice that ΦB is gauge invariant. And this is good because the measured result
depends on it: The probability of seeing the particle hit the location x is
e 2
Px = |ψ(x)|2 = uu0 + ul ei ~c ΦB
– if we turn the knob controlling the current through the solenoid, the interference
pattern will shift! (Notice that the pattern returns to itself when ~ce ΦB changes by 2π.)
[End of Lecture 12]
The observability of this kind of phase was used brilliantly by Dirac to show that
if magnetic monopoles exist with magnetic charge g, then the combination eg is quan-
tized.
[from Commins] A magnetic monopole is a point source of magnetic field, around which
~ = g Ř
B
R2
~ · d~a = 4πg. We can learn quite a lot by supposing
H
where g is the magnetic charge, B
one of these exists.
Consider also an electric charge, separated by −~a from the location of the monopole,
so
~ = e ř,
E ~ ≡ ~r − ~a.
R
r2
~ is not parallel to B,
If neither moves, there is no force on either. But notice that E ~ so
there is energy and momentum in the fields; the momentum density (in cgs) is
1 ~ ~
℘
~= E×B
4πc
1 ge ~
= ~r × |{z}
R
4πc R3 r3
r−~a
~
1 ge
=− ~r × ~a (2.18)
4πc R3 r3
82
The angular momentum density about the origin is
~j = ~r × ℘ 1 ge
~= ~r × (~a × ~r) .
4πc r3 R3
The total angular momentum is
Z
~=
S ~jd3 v ;
But angular momentum is quantized in units of ~2 . (You learned this in the ‘boot-
strap’ discussion of 212A.) This means that S = egc = ~2 n, n ∈ Z. Therefore electric
~c
charges only come in integer multiples of 2g .
Here is another point of view on this result, which is called Dirac quantization.
~ ·B
Since ∇ ~ = 0 away from R = 0, we can still find a vector potential, though it will
be singular at R = 0.
~ = g Ř = ∇
B ~ ×A
~= 1
(∂θ (sin θAϕ ) − ∂ϕ Aθ ) Ř + . . .
R 2 R sin θ
g 1
=⇒ 2
= (∂θ (sin θAϕ ) − ∂ϕ Aθ )
R R sin θ
We can set Aθ = 0 by azimuthal symmetry, so we need
g
sin θ = ∂θ (sin θAϕ (R, θ))
R
83
Integrating gives
cos θ K(R)g g
−g + = sin θAϕ =⇒ Aϕ = (K(R) − cos θ) .
R R R sin θ
~ ×A
(Vanishing of the other components of ∇ ~ will tell us that K(R) = k is constant.)
Even for R 6= 0, this solution Aϕ will be singular when sin θ = 0, i.e. at the poles
θ = 0, π. We can avoid this by choosing k:
g 2 sin2 2θ
(+) g 1 − cos θ g θ
k = +1 : Aϕ = = θ θ
= tan
R sin θ R 2 sin 2 cos 2 R 2
g 2 cos2 2θ
(−) g −1 − cos θ g θ
k = −1 : Aϕ = =− = cot
R sin θ R 2 sin 2θ cos 2θ R 2
84
2.3 Integer Quantum Hall Effect
eB eB
where Dz = ∂z − 4
z̄, Dz̄ = ∂z̄ + 4
z. An intermediate step here, which we’ll use
later is:
Lz = i (y∂x − x∂y ) = z∂z − z̄∂z̄ (2.21)
the angular momentum of the particle.
85
When ~ = c = 1, a magnetic field has dimensions of length−2 :
c~ 1
`2B ≡ ∝ , (2.22)
eB B
`B is called the magnetic length.
Now I’ll do something mysterious-seeming that will solve this problem. On the
homework you’ll see that (in a different gauge) it can be written as a single harmonic
|z|2
−
4`2
oscillator. Let ϕ ≡ e B . Then it is useful to observe that if eB > 0,
1
H=ϕ − D̃z D̃z̄ + D̃z̄ D̃z ϕ−1 , (2.23)
m
where
eB eB eB
D̃z = ϕ−1 Dz ϕ = ∂z − z̄ − z̄ = ∂z − z̄,
4 4 2
D̃z̄ = ϕ−1 Dz̄ ϕ = ∂z̄ . (2.24)
Here f (z) is an arbitrary holomorphic function, meaning that it depends only on z and
not on z̄, and it is well-behaved. [End of Lecture 13]
There are many such functions! A nice basis of them is the polynomials:
|z|2
−
(0) m 4`2
ψm (z) =z e B (2.27)
is a lowest Landau level (LLL) state with angular momentum (2.21) equal to m. It has
√
support in a ring centered at the origin, with a maximum at rm = 2m`B :
|z|2
(0) m log z− 4`
|ψm (z)| ∝ e B . (2.28)
86
2
The orbit of such an electron has area πrm = 2π`2B m and therefore in a constant
= m hc
R 2
magnetic field it encloses m flux quanta: m B · da = Bπrm e
.
Therefore, there is one state in the LLL for each flux quantum. This means that
if the number of electrons per flux quantum ν = 1, the free fermion groundstate will
completely fill the lowest Landau level, and not at all involve states in the second
Landau level. There’s an energy gap ~ωc . The groundstate wavefunction takes a
beautiful form:
1 1 ···
Y z1 z2 ··· Y PN 2 2
(0)
Ψν=1 (z1 · · · zN ) = det ψm (zj ) = (−1) N
ϕ(zi ) = (zi −zj )e− i=1 |zi | /(4`B ) .
ij i z12 z22 ···
i i<j
.. .. ...
. .
(2.29)
Now, there’s something slightly awkward about what I said above: how can we
have a uniform magnetic field everywhere, and one electron per flux quantum, and a
finite number N of electrons? We could use periodic boundary conditions, but then we
couldn’t use polynomials z n . In real life, what happens is that the Hall droplet has a
boundary. Suppose that the electrons are confined to some region of space (where B is
uniform, so the analysis above applies) by a smoothly-varying (electrostatic) potential
V (r). We can write the full hamiltonian as
and think of V (r) as a perturbation. This perturbs the energies of the LLL states
above by
1
Em ≈ ~ωc + V (rm ). (2.31)
2
√
But here rm = 2m`B . The excitations with small m lie in the bulk and have energy
gap ~ωc . But for large enough m, rm reaches the boundary of the sample, where the
state m + 1 is unfilled, and we can make an excitation by moving the particle from m
to m + 1.
87
The angular momentum m of an excitation at radius rm is m = rm k, where k
is its linear momentum in the angular direction, k = m/rm . So we can eliminate
√ √
m = 2k`B and regard (2.31) as a dispersion relation
1
Em ≈ ~ωc + V (2`2B k) (2.32)
2
for the edge excitations. The energy above the groundstate of the excitation is δω =
2`2B V 0 δk. This is an excitation with velocity 2`2B V 0 , and there is no excitation with the
opposite velocity. It is a gapless chiral edge excitation. And it will be there no matter
what the potential is.
Using this picture we can also calculate the Hall conductivity. Let’s think about a
Hall droplet which is a strip extended in the y direction. In this geometry it’s nicer
to use Landau gauge, where Ay = Bx (as you’ll do on the homework) so that py is
a conserved quantity. These states are labelled by the Landau level index n and the
y-momentum k. The state labelled k is centered at x = −k`2B . (This relation is just like
our relation for rm above.) The groundstate looks like this [figure from David Tong]:
We’ll apply an electric field in the x direction. Then the state looks like this:
88
Robustness of quantum Hall plateaux. The value of σxy eh2 is measured to be
an integer to many decimal places. But we have not yet explained why it keeps the
same value of B is varied away from the value where the filling fraction is exactly an
integer. The fact that there is an energy gap suggests that not much can happen,
because getting a different state would require some involvement of higher Landau
levels. But in fact the extreme precision which with the Hall conductivity remains an
integer requires one more ingredient. Ironically, it is actually a consequence of disorder.
Imagine that the electrons are moving in a smooth but random electrostatic potential
(this is what I mean by disorder). Our calculation above shows that classically electrons
move along equipotential contours. As we vary the filling, the Hall fluid of electrons
is like a rising sea, filling up more of the landscape. One effect of the disorder is to
broaden each Landau level into a band of energies. The states at the top and bottom
of each of these bands are associated with orbits that go around minima or maxima of
the potential – these orbitals are localized, and do not contribute to the conductivity.
Only special contours, which are in the middle of the bands, are extended across the
sample. The key point is that only when the fermi level passes through one of these
extended states does the Hall conductivity change.
General arguments about quantum Hall states. Actually there is a very
2
general argument that a state with nonzero Hall conductivity σxy = ν eh and an energy
gap in the bulk must have gapless edge modes. The idea is to consider a cylinder
geometry.
Z Z Z I Z
~ Faraday
~ · d~` jr =σxy Eϕ c
Φ0 = ∆Φ = dt∂t d~a · B = −c dt E = − dtjr
hole C σxy
| {z }
=∆Q
where C is a curve going around the hole. We conclude that an amount of charge
Φ0
∆Q = σxy = νe (2.34)
c
(e is the charge of the electron) is transferred from one edge of the cylinder to the other.
89
Moreover, the initial and final Hamiltonians are related by a gauge gransformation:
H(Φ = 0) ∼
= H(Φ = 2π).
R
They have the same spectrum. Moreover, the work done on the system is IdΦ ∝
2
dt dΦ
R
dt
which goes to zero in the thermodynamic limit (make both inner and outer
radius of the cylinder large), if our process is adiabatic. Therefore the initial and final
states must be degenerate in the thermodynamic limit.
We’ve identified two different states related by the flux threading whose energies
become degenerate in the thermodynamic limit. Since we’ve assumed the bulk is
gapped, we conclude that there must be gapless edge states.
These two states differ in that an amount of charge ν has been moved from one
boundary to the other. If charge is carried only by electrons, localized objects with
integer charge, we conclude Z 3 ν = ∆Q/e = σxy , the Hall conductivity must be an
2
integer in units of eh .
Amazingly, the innocent-looking assumption “if charge is carried only by electrons”
can be violated. There are fractional quantum Hall states, where the bulk is gapped,
2
but σxy = ν eh with ν a fraction! The loophole is that these states have excitations
that carry fractional charge – in such a medium, the electron splits apart. Such states
cannot be understood without interactions. For more on this subject, you might try
reading chapter 2 here. For more on quantum Hall states in general, I recommend
David Tong’s notes.
[End of Lecture 14]
90
Part 2:
Hard Problems
I have already emphasized that so far we’ve been studying very special many-body
systems that can be solved by diagonalizing a one-particle Hamiltonian. Now it is time
to look at the rest of the world. The problem is that such Hamiltonians are hard to
solve. They are hard to solve numerically, because they involve matrices of size V N
where N is the number of particles and V is the volume of space. This grows very
quickly with N and the best we can do with exact diagonalization is usually on the
order of N ∼ 101 particles and V ∼ 101 lattice sites.
In general, they are also hard to solve analytically. There are a few examples where
various tricks allow us to solve exactly, mostly only in d = 1. Otherwise we must use
our wits. That is, we have to analyze extreme limits of couplings, where the problem
simplifies, and we have to use mean field theory when we can. And we have to come
up with new ideas. To illustrate these techniques in action, we will spend a little time
on an example which is not obviously made of particles hopping around. Rather it is
a model of a magnet.
91
3 A simple magnet
The Hilbert space is simply ⊗x H 1 where H 1 = span{|↑i , |↓i} is the Hilbert space of a
2 2
single qbit, that is, a spin-one-half. Acting on each of these Hilbert spaces is a set of
Pauli operators, which I will denote Xx , Yx , Zx . Their matrix elements are the Pauli
matrices. More precisely, e.g. by Xx I mean the operator that acts as the identity on
all the factors except for the spin at x, on which it interchanges up and down
Xx = 1 ⊗ · · · ⊗ X ⊗ · · · ⊗ 1
Hardcore bosons. In case you are bothered by the apparent left turn our discus-
sion has taken, let me show you that this same Hilbert space can actually be realized as
that of many particles. Relabel the two states at each site |↑i ≡ |1i , |↓i ≡ |0i. Regard
these as eigenstates of a number operator
That is, each site may contain either zero particles (like the Fock vacuum) or one
particle. The operators
1
σx± ≡ (Xx ± iYx )
2
respectively create a particle from |0ix or destroy a particle if it is present at x. So
these particles are like fermions in that there can only be one or zero in each orbital.
However, they commute at different sites
rather than anticommute, so they are not actually fermionic operators. These particles
are called hard-core bosons. Even if we write a Hamiltonian which is quadratic in σxα ,
it will not be exactly solvable, because the creation and annihilation operators do not
satisfy the canonical algebra. Another warning about this representation for present
purposes: the Hamiltonian we’ll study, (3.2), will not conserve particle number.
Another way to think about how such a Hilbert space might arise is if we had a
bunch of spin-half electrons, each of which got stuck, for some reason, at one of the sites
92
x. ‘Stuck’ means that their position degrees of freedom are frozen out, that is, it costs
too much energy to get them to move. But their spin degeneracy is still there. These
spins can then interact with each other. This is one way in which insulating magnets
happen, called a Mott insulator. Later we’ll make this description more quantitative.
Arrange the sites x into a lattice. Denote by hxyi a pair of neighboring sites, which
share a link of the lattice. The Hamiltonian we will consider is called the transverse
field Ising model:
X X
HTFIM = −J gXx + Zx Zy . (3.2)
x hxyi
Some of the things we say next will be true in one or more spatial dimensions.
Notice that J has units of energy; we could choose units where it’s 1. In 1d (or
on bipartite lattices), the sign of J does not matter for determining what state of
matter we realize: if J < 0, we can relabel our operators: Z̃j = (−1)j Zj and turn an
antiferromagnetic interaction into a ferromagnetic one. So let’s assume g, J > 0.
Competition. One reason that this model is interesting is because of the com-
petition between the two terms: the Xj term wants each spin (independently of any
others) to be in the state |→ij which satisfies
1
Xj |→ij = |→ij . |→ij = √ |↑ij + |↓ij .
2
In conflict with this are the desires of −Zj Zj+1 , which is made happy (i.e. smaller) by
the more cooperative states |↑j ↑j+1 i , or |↓j ↓j+1 i. In fact, it would be just as happy
about any linear combination of these a |↑j ↑j+1 i + b |↓j ↓j+1 i and we’ll come back to
this point.
Compromise? Another model that looks like it might have some form of compe-
tition is X X π
Hboring = cos θ Zj + sin θ Xj , θ ∈ [0, ]
j j
2
Why is this one boring? Notice that we can continuously interpolate between the states
enjoyed by these two terms: the groundstate of H1 = cos θZ + sin θX is
θ θ
|θi = cos |↑i + sin |↓i
2 2
– as we vary θ from 0 to π/2 we just smoothly rotate from |↑z i to |↑x i.
How do we know the same thing can’t happen in the transverse-field Ising chain?
Symmetry. The Ising model has a G = Z2 symmetry which acts by Zj → SZj S† =
−Zj , Xj → SXj S† = +Xj , where the unitary S commutes with HTFIM : SHTFIM S† =
93
HTFIM . More explicitly, S = x Xx . It is a Z2 symmetry in the sense that S2 = 1
Q
(since X2 = 1). The difference with Hboring is that HTFIM has two phases in which G
is realized differently on the groundstate.
g = ∞ : First, let’s take g so big that we may ignore the ZZ ferromagnetic term,
so X
Hg→∞ = − Xj .
j
(The basic idea of this discussion will apply in any dimension, on any lattice.) Since
all terms commute, the groundstate is the simultaneous groundstate of each term:
Notice that this state preserves the symmetry in the sense that S |gsi = |gsi. Such a
symmetry-preserving groundstate is called a paramagnet.
g = 0 : Begin with g = 0.
X
H0 = −J Zj Zj+1
j
has groundstates
|+i ≡ |↑↑ · · · ↑i , |−i ≡ |↓↓ · · · ↓i ,
or any linear combination. Note that the states |±i are not symmetric: S |±i = |∓i,
and so we are tempted to declare that the symmetry is broken by the groundstate.
E 1
± ≡ √ (|+i ± |−i)
2
are symmetric – they are S eigenstates, so S maps them to themselves up to a phase.
It gets worse: In fact, in finite volume (finiteEnumber of sites of our chain), with g 6= 0,
|+i and |−i are not eigenstates, and + is the groundstate. BUT:
1. The two states |+i and |−i only mix at order N in perturbation theory in g, since
P
we have to flip all N spins using the perturbing hamiltonian ∆H = −gJ j Xj
to get from one to the other. The tunneling amplitude is therefore
N →∞
T ∼ g N h−| X1 X2 · · · XN |+i → 0.
94
2. There’s a reason for the symbol I used to denote the symmetric states: at large N ,
these ‘cat states’ are superpositions of macroscopically distinct quantum states.
Such things don’t happen, because of decoherence: if even a single dust particle
in the room measures the spin of a single one of the spins, it measures the value
of the whole chain. In general, this happens very rapidly.
P
3. Imagine we add a small symmetry-breaking perturbation: ∆H = − j hZj ;
this splits the degeneracy between |+i and |−i. If h > 0, |+i is for sure the
groundstate. Consider preparing the system with a tiny h > 0 and then setting
h = 0 after it settles down. If we do this to a finite system, N < ∞, it will be in
an excited state of the h = 0 Hamiltonian, since |+i will not be stationary (it will
have a nonzero amplitude to tunnel into |−i). But if we take the thermodynamic
limit before taking h → 0, it will stay in the state we put it in with the ‘training
field’ h. So beware that there is a singularity of our expressions (with physical
significance) that means that the limits do not commute:
lim lim Z 6= lim lim Z.
N →∞ h→0 h→0 N →∞
The conclusion of this brief discussion is that spontaneous symmetry breaking actu-
ally happens in the N → ∞ limit. At finite N , |+i and |−i are approximate eigenstates
which become a better approximation as N → ∞.
This state of a Z2 -symmetric system that spontaneously breaks the Z2 symmetry
is called a ferromagnet.
95
[End of Lecture 15]
When g is not infinite, we can learn a lot from (1st order) degenerate perturbation
theory in the ferrmomagnetic term. The key information is the matrix elements of the
perturbing hamiltonian between the degenerate manifold of states. Using the fact that
Zj |→i = |←i , so,
Zj Zj+1 |→j ←j+1 i = |←j →j+1 i
X
hn ± 1| Zj Zj+1 |ni = 1,
j
the ferromagnetic term hops the spin flip by one site. Within the degenerate subspace,
it acts as
Heff |ni = −J (|n + 1i + |n − 1i) + (E0 + 2gJ) |ni .
It is a kinetic, or ‘hopping’ term for the spin flip.
Let’s see what this does to the spectrum. Assume periodic boundary conditions
and N sites total. Again this is a translation invariant problem (in fact essentially the
same one we’ve studied several times before), which we can solve by Fourer transform:
(
1 X −ikxj xj ≡ ja,
|ni ≡ √ e |ki ,
N j k = 2πm
Na
, m = 1..N .
|ji = Zj |gs∞ i .
And it is annihilated by the annihilation operator Zj – you can’t have two spin flips
at the same location! These particles are their own antiparticles.
P
The number of such particles is counted by the operator j (−Xj ). The number
of particles is only conserved modulo two, however.
96
What happens as g gets smaller? The gap to creating
a spin flip at large g looks like 2J(g − 1). If we take this
formula seriously, we predict that at g = 1 it costs zero
energy to create spin flips: they should condense in the
vacuum. An excitation is said to be condensed in a state
|ψi if hψ|O|ψi 6= 0, where O is the operator that creates the excitation. Condensing
spin flips means hZi = 6 0, so the state is magnetized. (We shouldn’t take this g = 1
prediction too seriously because it’s just first order in perturbation theory, but it turns
out to be exactly right.)
It’s possible to develop some more evidence for this picture and understanding of
the physics of the paramagnetic phase in the Ising chain by doing more perturbation
theory, and including states with two spin flips. Notice that for a state with two
spin-flip particles, the total momentum k no longer uniquely determines the energy,
since the two spin-flips can have a relative momentum; this means that there is a two-
particle continuum of states, once we have enough energy to make two spin flips. For
more on this, see e.g. Sachdev (2d ed) §5.2.2. In particular the two spin-flips can form
boundstates, which means there are two-particle states with energies slightly below
2∆.
g 1 Now let’s consider excitations of the ferromagnet, about the state |+i =
P
|↑↑ · · · ↑i . It is an eigenstate of H0 = −J j Zj Zj+1 and its (groundstate) energy is
E0 = −JN . We can make an excitation by flipping one spin:
|· · · ↑ ↑ ↑ · ↓ · ↑ ↑ ↑ · · · i
This makes two bonds unhappy, and costs 2J + 2J = 4J. But once we make it, there
are many such states: the hamiltonian is the same amount of unhappy if we also flip
the next one.
|· · · ↑ ↑ ↑ · ↓ ↓ · ↑ ↑ · · · i
The actual elementary excitation is a domain wall (or kink), which only costs 2J.
The domain wall should be regarded as living between the sites. It is not entirely a
local object, since with periodic boundary conditions, we must make two, which can
then move independently. To create two of them far apart, we must change the state
of many spins.
At g = 0 the domain walls are localized in the sense that a domain wall at a fixed
position is an energy eigenstate (just like the spinons at g = ∞), with the same energy
P
for any position. But now the paramagnetic term − j gXj is a kinetic term for the
domain walls:
Xj+1 |· · · ↑↑↑j · ↓j+1 ↓↓ · · · i = |· · · ↑↑↑j ↑j+1 · ↓j+2 ↓ · · · i .
| {z } | {z }
j̄ =|j̄+1i
97
Just like in our g 1 discussion, acting on a state with an even number of well-
separated domain walls
(Heff − E0 ) |j̄i = −gJ (|j̄ + 1i + |j̄ − 1i) + 2J |j̄i
where the diagonal term is the energy cost of one domain wall at rest. Again this is
diagonalized in k-space with energy
one dwall (k) = 2J(1 − g cos ka)
Again, this calculation is almost ridiculously successful at
predicting the location of the phase transition:
g→1
∆DW = 2J(1 − g) → 0.
We predict that for g > 1, the domain walls condense; this produces a paramagnetic
state.
Notice that although our discussion of the paramagnetic state g 1 can be applied
in any d ≥ 1, the physics of domain walls is very dimension-dependent.
So far we have used perturbation theory about g = 0 and g = ∞ to delicately inch our
way toward the interior of the phase diagram of the TFIM in one dimension (3.2). Here
we take a plunge and try to guess the groundstate for all g. The nice thing about trying
to guess the groundstate is the Rayleigh-Schrödinger hedge: the energy expectation in
any state is an upper bound for the groundstate energy; minimizing the energy within
a class of guesses is called a ‘variational approach’.
The name for the particular guess we’ll make is ‘mean field theory’, which means
that we completely ignore entanglement between different sites, and suppose that the
state is a product state
|MFTi = |ψ1 i ⊗ |ψ2 i · · · |ψj i · · · ...
If we further assume translational invariance then the state at every site is the same
and we have one bloch sphere to minimize over for each g:
θ iϕ/2 θ −iϕ/2
|ňi = ⊗j |↑ň ij = ⊗j cos e |→i + sin e |←i .
2 2 j
(Here θ is the angle ň makes with the x axis, and ϕ is the azimuthal angle in the yz
plane, from the z-axis.) To evaluate the energy expectation in this state, we only need
to know single-qbit expectations:
h↑ň | X |↑ň i = cos θ, h↑ň | Z |↑ň i = sin θ cos ϕ.
98
So the energy expectation is
z
E(θ, ϕ) = −N J sin2 θ cos2 ϕ + g cos θ
2
where z is the number of neighbors of each site of the lattice (z = 2 for the 1d chain).
This is extremized when ϕ = 0, π and when
0 = ∂θ E = N J sin θ (z cos θ − g) .
(where I have set g = gc = z except in the crucial quadratic term). This has minima
at
√
hZj i = sin θ ' θ = ± gc − g . (3.4)
The energy behaves like
1
2z (gc − g)2 , g < gc
EM F T (g)
= −g +
NJ
0, g ≥ gc
99
fluctuations), and so are other properties, such as the exponent in (3.4), which should
be 1/8 instead of 1/2.
Another point of view on mean field theory is: let’s think about the experience of
the world of a single spin. Much like a human during lockdown, its experience of the
world is mediated by the behavior of its neighbors. We can write the Hamiltonian as
!!
X X 1 X
H = −J g Xi + Zi Zj .
i i
2 j, neighbors of i
P
From the point of view of Zi , the object 12 j, neighbors of i Zj ≡ hi just acts like
an effective longitudinal magnetic field. A mean field theory arises by replacing the
neighboring operators by their expectation values:
X X
HM F = −J g Xi + (Zi hZj i + hZi i Zj − hZi i hZj i) .
i hiji
Now this hamiltonian has the form of Hboring above, whose groundstate is just all
the spins pointing in some direction θ(hZi) between x̂ and ẑ. This becomes slightly
nontrivial if we demand that hZi comes out right when computed in the resulting
groundstate. This self-consistency condition is
zJm
m ≡ hZi = p
(zJm)2 + (gJ)2
q
2
(where z is the number of neighbors of each site). This says m = ± 1 − gz2 , repro-
ducing our (approximate!) answer for the critical coupling g = z. (We would find the
same condition by minimizing the groundstate energy over m.)
Notice that if we let the magnetization m in the mean field ansatz depend on space,
we would get not just a Landau-Ginzburg potential, but a Landau-Ginzburg functional
of m, involving derivative terms, like
X Z
mi mj ' dd x∇m~ · ∇m~ + ··· . (3.5)
hiji
But you can see that, as long as the sign in front is positive, we minimize the energy by
taking m to be constant. (If the sign is the other way, we will make instead a density
wave, where the sign of the magnetization oscillates in space, spontaneously breaking
translation invariance. An antiferromagnet is an extreme example of this where the
period of oscillation is a single lattice spacing, and where the approximation on the
RHS of (3.5) is not valid.)
100
3.0.2 Correlation functions and long-range order
A useful set of physical (observable! as we’ll discuss in a moment) quantities are the
correlation functions of the spins:
the sum over positions is over r ∈ a{1...N }, and (assuming periodic boundary condi-
tions again) the argument q ∈ N2πa {1...N }.
Correlations in the quantum Ising chain. At g = ∞, all the way in the
paramagnetic phase,
h0| Zj Zl |0i |g=∞ = δjl
– different sites are totally uncorrelated. At finite but large g 1 we have instead:
|xj −xl |a
h0| Zj Zl |0i |g1 ' e−|xj −xl |/ξ (3.6)
eikx
X 1 Z
h0| Zj Zl |0i ' eika(j−l) ' d̄d k
2ωk 2ωk
k
2
k
where ωk = ∆ + 2M is the dispersion relation for the spin flips and x = a(j − l). This integral is
√
doable and the answer goes like e− 2M ∆|x| for large x. Notice that the long-wavelength behavior of
ωk is appropriate for large x, since then small k dominates the integral.
101
More simply, if we train the system appropriately (as we discussed in point 3), then
we’ll have
h0| Zj |0i = ±N0 (g).
N0 is the order parameter, the magnetization. Another way to state the issue about
the order of limits of h → 0, N → ∞: in a finite system the magnetization is zero:
Consider the limit where N → ∞ so that the allowed momenta fill in a continuous
Brillouin zone q ∈ 2π
N
{1..N } → (0, π] ' (−π, π]. So the inverse Fourier transform is
Z π
dq
C(R) = S(q)eiqR .
−π 2π
where ‘regular’ means terms which are smooth at q = 0. This singularity is a sharp
signature of the broken symmetry.
We can learn something by considering a spectral representation of S(q), obtained
by resolving the identity by energy eigenstates |ni:
1 1 X
S(q) = h0| Zq Z−q |0i = | hn| Zq |0i |2 . (3.9)
V V n
e−iqr Zr = Z†−q .
P
Here I used Zq ≡ r
A sum rule: C(R = 0) = h0| Zr Zr |0i = h0|0i = 1. But the (inverse) Fourier
transform is Z π
dq
1 = C(R = 0) = S(q) .
−π 2π
102
This is some short-distance information which remembers that the objects we’re looking
at square to one. Given the spectral interpretation, this sum rule means that as we vary
parameters, the spectral weight can move around, but it can’t go away. In particular,
when the system orders as in (3.8), the 2πN02 contribution at q = 0 has to come from
somewhere.
Dynamical structure factor.
C(r, t) ≡ h0| Zr (t)Z0 (0) |0i .
This reduces to C(r) if we set t = 0. It is useful to think of Zr (t) in Heisenberg
representation: Zr (t) = e+iHt Zr e−iHt . Its Fourier transform in time and space is the
dynamical structure factor:
X Z ∞
−iqr
S(q, ω) = e dte+iωt C(r, t).
r −∞
R
Notice that S(q) = d̄ωS(q, ω) (not S(ω = 0)). The spectral representation of the
dynamical structure factor is:
1 X
S(q, ω) = | hn| Zq |0i |2 2πδ(ω + E0 − En ) . (3.10)
V n
103
V is the volume of space.
Now let’s interpret this in terms of a transition rate in a scattering experiment, via
Fermi’s Golden Rule. Recall that the FGR says that the transition rate is
X
Γ= | hF | Hpert |Ii |2 ;
F
more generally, if we are uncertain about the initial state, so it is described by a density
P
matrix, ρ = I pI |IihI|, then
X
Γ= pI | hF | Hpert |Ii |2 .
F,I
In the problem above, the initial state is |n0 i with probability pn0 = e−En0 /T /Z
given by the Boltzmann distribution. (In the T → 0 limit, we return to (3.10), where
the initial state is the groundstate with probability 1.) The perturbing hamiltonian is
linear in the spin operator Hpert ∝ Z(q). With these assumptions, the transition rate
per unit time is exactly S(q, ω). This explains the connection between the dynamical
structure factor and experiments involving scattering of neutrons: a neutron couples to
the spins via its magnetic dipole moment, which couples linearly to the spin operator;
ω, q are the change in energy and momentum of the system. The transition probability
(at zero temperature) is proportional to something like:
D E
|hn| ⊗ neutron, final(E − ω, ~k − ~q, ẑ) ~Sneutron · σ
~ |0i ⊗ neutron, initial(E, ~k, ẑ) |2 .
(3.11)
The (known) neutron states allow us to determine q, ω. You could ask me how we do
an experiment that couples just to Z, and not all of σ ~ . We can do this if we can choose
the initial and final spin states of the neutrons to be polarized along ẑ, as indicated in
(3.11).
Given our understanding of the spectrum at large and small g, we have a prediction
for the behavior of S(q, ω): at large g, the lowest-energy excitation is a single spin
flip, which conveniently is created by Z, and so contributes to S(q, ω). But it only
contributes if q and ω are related in the right way, namely ω = q . This means that there
will be a sharp peak along this curve: for ω in the right range, S(q, ω) = Zδ(ω − q ).
For larger ω, we access the multi-particle continuum. In fact, Z |0i only includes states
with an odd number of particles, so the continuum in S(q, ω) starts near 3∆. What
happens at small g? At small g, the spin flip is not stable – it can decay into a pair of
domain walls, which are free to move apart. Although these domain walls behave in
many ways like particles, the operator Zj cannot create a single domain wall. (In fact
the domain wall creation operator is made of a string of Xs, since we must flip all the
104
spins on one side of the wall.) So the spectral function of Z has only a continuous mess
at small g – a multiparticle continuum of domain walls, no sharp delta-function peak.
People24 have done this experiment on a system that is well-described by the TFIM
(CoNb2 O6 ). It can be approximated as a collection of decoupled chains of spins with
nearest-neighbor ferromagnetic Ising interactions. The transverse field term can be
produced by applying a (surprise) transverse magnetic field, and hence g can be varied.
Indeed at large g one sees a sharp single-particle peak, and at small g there is a
featureless mess, i.e. a two-particle continuum. Here is a summary of the results of the
experiment (from Ashvin Vishwanath):
A further feature of these experiments that I must mention is: if we apply a longitudinal
P
field, ∆H = j hz Zj , we confine the domain walls, in the following sense. This term
means that a region of down-spins costs an energy proportional to its length. This
is an interaction energy between the domain walls at the edges of the region which is
linear in their separation, a constant attractive force between them. Their spectrum
of boundstates is extremely interesting and represents a physical appearance of E8
symmetry.
Confession. Actually it turns out that the TFIM in one dimension is one of the
examples that can be solved exactly by cleverness. The cleverness is called the Jordan-
Wigner transformation, and it relates the spin system to a model of fermions. And
the hamiltonian is actually gaussian in those fermions! Actually, the discussion around
(3.1) is the special case of it in d = 0, i.e. at a single site. But this is a story for another
time (namely, the homework).
24
Coldea et al, Science 327 (2010) 177.
105
4 Interacting bosons
[Leggett, Quantum Liquids] Recall that the groundstate of a collection of N non-
interacting bosons is
N
b†p0
√ |0i
N!
where p0 labels the mode with the lowest single-particle energy: (p0 ) is the minimum
of (p). The crucial feature of this state is that there is a macroscopic occupation of
a single single-particle orbital. Let us tentatively call this phenomenon Bose-Einstein
Condensation (BEC).
And recall briefly from statistical mechanics that this phenomenon survives to finite
T > 0. For 0 ≤ T ≤ Tc , the accounting of all N (still non-interacting) particles satisfies
Z ∞
ρ()d
N = N0 (T ) + N
0 eβ − 1
P
where ρ() = i δ( − i ) is the density of single-particle energy levels ({i } is the
spectrum of the single-particle hamiltonian h). Tc is the value of T for which N0 (T )
goes to zero. If ρ() ∝ α−1 then
α
T
N0 (T ) = N 1 − , T < Tc .
Tc
So what? Let us focus for a moment on question 3. The answer is associated with a
collection of phenomena which are called superfluidity or superconductivity. The two
names refer to two separate cases: the latter applies when the boson that is condensing
106
is charged under electromagnetism. Its condensation then has some important conse-
quences for the behavior of the electric and magnetic fields and currents. The former
case is when the boson is neutral, such as a 4 He atom. What are these phenomena?
(We are going to indulge for the next several paragraphs in phenomenology – I will
tell you some things that have been seen in experiments, and we’ll come back and
understand as many of them as possible as a consequence of the BEC phenomenon.)
Let us place our bosons in a ring-shaped container which can
rotate. We can think of this as simply an external single-
particle potential
tity with units of frequency, which since it has an ~ is like a quantum unit of angular
velocity. In terms of it, the angular momentum of the particles at angular frequency
ω is
ω
L = Icl ω = N ~ .
ωc
We will consider two different but related experiments.
(1) Rotate the fluid. Do this just by stirring it: start by rotating the container at
ω ωc . In fact, we can start stirring it when it’s in the ‘normal phase’ at T > Tc .
(In the case when the fluid is charged, we can do this by turning on a magnetic field,
i.e. threading magnetic flux through the hole in the ring; by Faraday, this creates an
EMF which will rotate the particles.) After it gets going at frequency ω, cool it below
Tc .
If it were a normal system, the rotational motion would be transient. It gets
damped by viscous drag with the walls of the container (in the case of a neutral fluid),
or electrical resistance (in the case of a charged fluid). In the case of a superfluid, the
rotation persists. In some cases it persists indefinitely. [End of Lecture 17]
Moreover, the angular momentum is given by
L = fs (T )Icl ω̃ (4.2)
ω
where (get ready) ω̃ is the nearest integer to ωc
when the system cools below Tc . fs (T )
107
is a function of temperature that is a property of the material. It satisfies
(
1, T → 0
fs (T ) → .
0, T → Tc
(4.2) therefore means that if we vary the temperature adiabatically, we can vary the
angular momentum of the fluid reversibly.
(2) Consider first the neutral case. For the second experiment, we rotate the con-
tainer. This means the hamiltonian is
N
X
Hω = H − ω ~ =H−ω
~ ·L ~· ~ri × p~i
i=1
where p~i is the lab-frame momentum of the ith particle, and H is the hamiltonian
before we did any rotating.
A normal fluid will rotate along with the container, with L ~ ∼ Icl ω
~ . In a superfluid
(as long as the rotation isn’t too fast, ω < ωc /2), the angular momentum will instead
satisfy
~ = (1 − fs (T ))Icl ω
L ~, (4.3)
with the same function fs as above. It is as if the fraction fs of the fluid simply
refuses to participate, and doesn’t contribute to the moment of inertia. Some names:
fn (T ) ≡ 1 − fs (T ) is sometimes called the ‘normal fluid fraction’. This behavior (4.3)
is called ‘non-classical rotational inertia’ or the London effect (after its theoretical
predictor) or the Hess-Fairbanks effect (after its experimental discoverers).
Comparison between (1) and (2): Notice that (1) is not an equilibrium phenomenon.
In fact, the groundstate of a system of non-relativistic particles cannot have angular
momentum L > N ~/2. Suppose itPdid and call the wavefunction Ψ. Then consider the
N
state with wavefunction Ψ0 ≡ e−i i=1 θi Ψ where θi is the angular position of the ith
particle. This state has hΨ0 | V |Ψ0 i = hΨ| V |Ψi where V is any terms in the hamiltonian
P p2
not involving derivatives. And if K = i 2mi is the kinetic term, we have
1 ~2
hΨ0 | K |Ψ0 i = hΨ| K |Ψi − ωc L + Icl ωc2 = hΨ| K |Ψi − (L − N/2) .
2 mR2
If L > N ~/2 this decreases the total hHi which by the variational theorem contradicts
our assumption that Ψ was the groundstate. Recall that we assumed that the initial
frequency of rotation was ω ' ω̃ ωc which means L 21 N ~. On the other hand,
the flow can last for 1015 years by some estimates. It is a very long-lived metastable
state. In contrast (2) is a statement about the actual groundstate of Hω .
108
The analog of (2) for charged fluids is the Meissner effect. And the analog of
rotating the container is applying a magnetic field. This replaces K + V with
N
1 X 2 X
~ i) +
X
HA = p~i − eA(r U (ri ) + V (rij ) .
2m i=1 i i<j
~
For example, consider eA(r) = m~ω × ~r for ω constant, and compare to Hω above. The
~ ~j = 0, with ρ(r) = e P δ d (r −ri ))
microscopic current operator (participating in ρ̇+ ∇· i
is
N
~j(r) = δHA = e
X
(pi − eAi ) δ d (r − ri ).
~
δ A(r) m i=1
~
~j(r) = −Λ(T )A(r).
ne2
This is called the London equation. Λ(T ) = m? s
f (T ) where fs (T ) is as above.
Its consequence is that a superconductor is diamagnetic: If we plug the London
~ ×B
equation into Maxwell’s equations (i.e. Faraday’s law) c2 ∇ ~ = J~ + ∂t E
~ in the way
we would usually derive the wave equation for lightwaves in vacuum, we find that A ~
satisfies
~ + c2 ∇
−∂t2 Ax ~ ×∇ ~ ×A ~ = −ΛA. ~
~ ·A
If we work in ∇ ~ = 0 gauge, this becomes the massive wave equation. This has the
consequence that if we apply a static magnetic field to such a system, it will fall off
inside the sample like √
B ∼ e−x Λ
◦
where the penetration depth √c is something like 100 A.
Λ
Notice that the results of (both versions of) experiment (2) are intrinsically quantum
mechanical: it is possible to prove that neither rotating the container, nor applying
a uniform magnetic field can change the equilibrium properties of a classical system.
(This is the Bohr-van Leeuwen theorem. For details see Appendix 1A of Leggett’s
book.)
109
More generally, suppose we are given a distribution of such wavefunctions Ψs , with
probability ps , that is, a density matrix
X
ρ= ps |Ψs ihΨs |.
s
Consider the one-particle density matrix (the reduced density matrix for a single par-
ticle, it doesn’t matter which one):
X X 1 1
ρ1 (r, r0 ) = ps Ψ?s (r, r2 · · · rN )Ψs (r0 , r2 · · · rN ) = ψ † (r)ψ(r0 ) = trρψ † (r)ψ(r0 ).
s r2 ···rN
N N
This matrix (in the r, r0 indices) is hermitian ρ1 (r, r0 ) = ρ?1 (r0 , r), i.e. ρ1 = ρ†1 , and so
it can be diagonalized: X
N ρ1 (r, r0 ) = Ni χ?i (r)χi (r0 ).
i
R
Here χi are orthonormal eigenfunctions of ρ1 ( dd rχ?i (r)χj (r) = δij ), and Ni /N are its
eigenvalues. (If the Ψs depend on a parameter, such as time, then so do χi , Ni .)
I put a factor of N in front so that the normalization of the density matrix (assuming
P
trρ = 1) trρ1 = 1 implies that i Ni = N . We can regard these eigenvalues Ni as
occupation numbers! The χi are an ON basis of single-particle states. If Ψ is of the
form b†1 · · · b†N |0i of a gaussian state, then they are the single-particle orbitals that we
filled, and the Ni are a histogram of their fillings.
If the Ni are all of order-one in N : Ni = O(N 0 ), we call this a normal state. If any
Ni = O(N ) is macroscopic, we call it BEC. (If exactly one is macroscopic, it is called
‘simple’ BEC. I will assume this always below for simplicity. There are some interesting
things to say about the other case.) What I mean by saying something is = O(N α )
is an asymptotic statement about a family of systems with different N : if limN →∞ Nx
is finite, then x = O(N ); if it is zero (but limN →∞ x is finite) then x = O(N 0 ). For
α
example, for N non-interacting bosons at T < Tc , limN →∞ NN0 = 1 − TTc is finite, so
this agrees with our previous definition of BEC in this case.
√
Order parameter. Suppose N0 = O(N ). Then let Ψ(r) ≡ N0 χ0 (r) ≡ |Ψ(r)|eiϕ(r) .
This is the (superfluid) order parameter. It satisfies r |Ψ|2 = N0 . The density and
P
Let
~jc (r) ~~
~vs ≡ = ∇ϕ(r).
ρc (r) m
110
This is a quantity with dimensions of velocity, called the superfluid velocity. Two
properties follow from the last expression:
• Consider one particle in a box, say in one dimension. The lowest-energy state
has wavefunction r
2 πx
ψ1 (x) = sin ≡ eiϕ1 |ψ1 |. (4.6)
L L
Nothing stops us from defining analogously the velocity
~~
v1 ≡ ∇ϕ1 . (4.7)
m
This velocity (4.7) also satisfies (4.4) and (4.5). However: clearly it is just zero,
since ψ1 is real and positive. And if we actually tried to measure the velocity of
the particle (p/m = i~∂m
x π~
) we’ll find ± mL each time with probability 21 . It is a
highly fluctuating, very quantum, quantity with big fluctuations.
• At the other end of things, consider a normal fluid of many particles. Here too
we can define a velocity
!
~jtotal 2~
P
~ i N i |χ i | ∇ϕ i
~vhydro = = P 2
.
ρtotal m j Nj |χj |
111
In stark contrast to (4.7), this quantity is classical in the sense that its fluctuations
go like √1N (I am appealing to the central limit theorem). However: it does not
satisfy (4.4) or (4.5) (unless it is a simple BEC, Ni = N δij !).
The superfluid velocity has the best of both limits: it has small fluctuations (because
of the macroscopic occupation of a single level), i.e. it behaves classically. For example,
consider N free particles in the state (4.6). This has ∆vs ∼ hv i
√ s , like a random walk
N
with N steps.
On the other hand, it is quantum in the sense of exhibiting quantized vorticity. To
reveal this property, we can put a hole in the box. I mean, we can consider a ring-
shaped geometry as above. For simplicity, consider the limit where the ring is very
narrow. The action of a particle fixed to move on such a ring of radius R is
Z
1 2 2 2
S = dt mR θ̇ + ωmR θ̇ .
2
The second term is the effect of rotating the ring with frequency ω: ω~ ×L~ = ωmR2 θ̇.
Notice that this is a proof of the Bohr-van-Leeuwen theorem quoted above: this term
is a total derivative, which therefore does not affect the equations of motion, or any
other classical physics.
It does however make a big difference quantum mechanically, as you know from the
homework. The associated single-particle hamiltonian is
~2
H= (Π − Iω)2 , I ≡ mR2 , [θ, Π] = i~.
2I
The eigenstates are einϕ with n ∈ Z so that it is single-valued under θ ≡ θ + 2π. Acting
on such a state, and writing H in terms of the quantum of angular velocity ωc ≡ ~I
above, the spectrum is
2
~ωc ω
En = n− .
2 ωc
Therefore, the groundstate is when π/~ = n is the
closest integer to ω/ωc , as promised above. Notice
that something special happens when ω → ωc /2 –
there is a degeneracy in the spectrum since two val-
ues of n are equally close to ωc .
If we put N bosons in this state, then the phase of the condensate wavefunction is
~ = n~ ϕ̂. Notice that
ϕ = nθ, and then ~vs = m~ ∇ϕ mR
I
n~ nh
~vs · d~` = 2π =
ring m m
112
in agreement with quantization of vorticity.
We haven’t yet explained all of the super-flow phenomena, but already can see a
basic reason for the robustness of the flow: The only way to get rid of it is to decrease
the vorticity. But the integer n cannot change smoothly, it can only jump.
Relation to long-range order. Put the system in a very big box. Now consider
the behavior of the single-particle density matrix when its arguments are far apart:
lim
0
N ρ1 (r, r0 ) = Ψ? (r)Ψ(r0 ) + N ρ̃1 (r, r0 ).
|r−r |→∞
The second term, i6=0 Ni χ?i (r)χi (r0 ) goes to zero for large |r − r0 |. The non-vanishing
P
of the Ψ? Ψ term is another way to extract the order parameter. This is called off-
diagonal long-range order (in this classic paper by CN Yang), in contradistinction to,
for example, the order associated with a solid, which is visible in the diagonal matrix
elements of the one-particle density matrix, ρ1 (r, r) = n(r), i.e. the density.
[End of Lecture 18]
Relation to symmetry breaking. Yet another way to think about the order
parameter is in terms of the field operator ψ(r) (which we’ve often been calling br for
bosons):
Ψ(r) = hgs|ψ(r)|gsi .
In this sense, given my definition above that a particle condenses when there is an ex-
pectation value for its creation (or annihilation) operator, Ψ represents the condensate
of bosons. This is consistent with the previous expression, because in the limit when
the two operators are arbitrarily far apart, only the groundstate propagates between
them:
|r−r0 |→∞
N ρ1 (r, r0 ) = hgs| ψ † (r)ψ(r0 ) |gsi → gs|ψ † (r)|gs hgs|ψ(r)|gsi .
Think of this like the equation for the classical electric field:
D E
~ classical (r) = E(r)
E ~
for example evaluated in a coherent state of the radiation field, a |zi = z |zi.
There is, however, an important distinction from the case of the EM field. For
P †
photons, E → eiα E is not a symmetry
D E [H, k ak ak ] 6= 0, and photons can come and
go as they like. This means that E ~ 6= 0 need break no symmetries (besides spatial
X
symmetries like rotations). But suppose that [H, ψr† ψr ] = 0. Here [N, ψ] = −ψ (ψ
| r {z }
≡N
113
removes a particle). The existence of this conserved charge N, by Noether’s theorem,
means a symmetry Uθ = e−iθN , which acts on ψ by
where adN (O) ≡ [N, O] is the adjoint action by N. The last equation follows by Taylor
expansion.
But now hα| ψ |αi =
6 0 means that the state |αi is not invariant under the particle-
number symmetry U. This is spontaneous symmetry breaking.
There is a problem with this picture, however. Acting with ψ changes the number
of particles:
ψ(r) |n-particle statei = |(n − 1)-particle statei .
P
This means that unless |αi = n αn |ni has an indefinite number of particles, i.e. is a
superposition of states with different number of particles, we will find
which is a source and a sink for particles. Adding this term is a bit like coupling the
system to a big reservoir of particles, except that we are not including the effects of
the entanglement with the reservoir, since we’re still treating the system as being in a
pure state. Again the order of limits between λ → 0 and N → ∞ will matter.
A good practical way to think about the situation is: the zero-temperature grand-
canonical ensemble is a convenient theoretical device. In this ensemble the U(1) sym-
metry is spontaneously broken. And in this ensemble, the fluctuations of the particle
number are negligibly small ∆N/ hN i ∼ √1 so don’t worry about it. If N were not
hN i
macroscopic, you would have to worry.
(Question 1) [Feynman, Statistical Mechanics §11.3, Leggett] Now we can ask whether
the BEC phenomenon survives interactions. We will approach this question in several
114
ways. For definiteness, let’s think about an interacting system of bosons governed by
a Hamiltonian such as
N
X p2i X
H= + V (ri − rj ). (4.8)
i=1
2m i<j
We will take V (r) to be short-ranged and strongly repulsive – it keeps the bosons from
sitting on top of each other. They behave like hard spheres of some radius we’ll call a.
A good physical realization to keep in mind is 4 He.
What is the groundstate wavefunction Φ0 (r1 · · · rN ) of (4.8)? This is a Hard Ques-
tion. But there are a few important things we can say about it without much work.
Because there are no complex numbers involved besides Φ0 , Φ?0 satisfies the same
equation. So Φ0 + Φ?0 , the real part, also satisfies it. (If the real part happens to
be zero, just take the imaginary part.) In fact this argument applies to all the
eigenstates of H, not just the groundstate. (It is a consequence of time-reversal
symmetry of H. More about that later if we have time.)
115
Suppose it were otherwise. For purposes of drawing,
let Φ(ξ) = Φ0 (r1 · · · ξ · · · rN ) so ξ is the coordinate
of one of the particles, it doesn’t matter which one.
The state with wavefunction |Φ| has the same val-
ues of (∂ξ Φ)2 , (∂ri Φ)2 , Φ2 as Φ0 . Because H only
involves first derivatives, this means that the energy
expectation for |Φ|,
R (∇Φ)
~ 2 2
R
|Φ|H|Φ| 2m
+VΦ 25
E= R = R
|Φ|2 Φ2
• Φ0 (r1 , r2 · · · rN ) ' 0 if |r1 − r2 | < a the size of the hard spheres determined by
where V becomes large and positive. But furthermore in the groundstate the
gradients of Φ0 must be small. That means that Φ0 must also be small if |r1 − r2 |
is just a little bigger. So in a state where the average space per particle is not
much bigger than a, the configurations where Φ0 is appreciable will have the
particles about equally spaced.
I will mention here a form of the wavefunction that builds in all these demands:
P Y
Φ = e− ij f (rij ) = F (rij )
ij
25
You might worry that this argument assumes that the slope of Φ(ξ) is nonzero at the putative
zero-crossing. However, if Φ(ξ) satisfies the schrodinger equation −∂ξ2 Φ + (V − E)Φ = 0 then if there
were some point ξ0 where both Φ(ξ0 ) = 0 and ∂ξ Φ(ξ0 ) = 0, then these two initial conditions would
determine Φ(ξ) = 0 everywhere. So this rules out a double zero Φ(ξ) ∼ ξ 2 , or a zero-crossing with
zero slope like Φ(ξ) ∼ ξ 3 . This argument comes from this thesis.
116
= 0, r < a
r→a a
with F (r) → 0 . An example of such a function is 1 − rij
(for rij > a). You
r→∞
→ 1
could imagine using this as a trial wavefunction, but it is not so easy to do the integrals.
Mean field theory in free space. One thing we can do is mean field theory. Here
is a simple mean field ansatz: put every
√ particle in a single-particle state χ(r), so that
the order-parameter field is Ψ(r) = N χ(r). Now evaluate the expectation value of
the Hamiltonian. We will try to find the orbital χ that minimizes the expected energy.
The answer will be a functional of Ψ which is invariant under (space-independent)
shifts of the phase,
Ψ → eiα Ψ. (4.9)
It will inevitably have the form
Z
~ ? · ∇Ψ
E[Ψ] = dd r r|Ψ|2 + u|Ψ|4 + ρs ∇Ψ ~ + ··· .
where the · · · is terms with more gradients or more powers of |Ψ|2 . Notice the similarity
to the Landau-Ginzburg energy for a magnet; the only difference is that instead of a
Z2 symmetry, we have a U(1) symmetry (4.9).
As long as ρs > 0, we lower the energy by making Ψ(r)
constant in space. As in the case of the magnet, there are
two possible phases. One is where r > 0, in which case the
minimum occurs when Ψ = 0. In this case, our ansatz isn’t
so good – the particles don’t want to macroscopically occupy
a single state. Notice that this solution is invariant under the
U(1) symmetry (4.9). In that sense, such a state preserves
the U(1) symmetry.
117
so short-ranged that it can be approximated as a delta-function:
V (rij ) = U0 δ d (rij ).
And let’s focus for simplicity on the case of N = 2 particles. We’ll make a variational
ansatz, i.e. a trial wavefunction Ψ(r1 , r2 ) = Ψ(r2 , r1 ). Its interaction energy is
Z Z Z Z
d d d
Eint = hΨ| Hint |Ψi = U0 d r1 d r2 δ (r1 −r2 )|Ψ(r1 , r2 )| = U0 d r|Ψ(r, r)| = U0 dd rρ1 (r, r).
2 d 2
1. Put both particles in the same single-particle state: Ψ(r1 , r2 ) = χ(r1 )χ(r2 ). This
gives Z
Eint = U0 dd r|χ(r)|4 .
2. Put the two particles in different, orthogonal states: Ψ(r1 , r2 ) = (χ1 (r1 )χ2 (r2 ) +
√
χ2 (r1 )χ1 (r2 ))/ 2. This gives
Z Z
d 1
Eint = U0 d r |χ1 | |χ2 | × 4 = 2U0 dd r|χ1 |2 |χ2 |2 .
2 2
(4.10)
2
First notice that (4.10) is twice the answer we would get for distinguishable particles.
Furthermore, consider the case where |χ1 (r)| = |χ2 (r)|, such as for plane waves. In
this case, Eint (2) = 2Eint (1) !
The generalization for N 2 particles is
Z
1 X
Eint ≈ U0 Ni Nj (2 − δij ) dd r|χi (r)|2 |χj (r)|2 (4.11)
2 ij
where Ni is the occupation number of level i26 . You see that, when the interaction is
repulsive U0 > 0, there is a big energy gain to be found by clumping the particles into
26
A more precise expression is
Z Z
1 X X
Eint ≈ U0 2 Ni Nj dd r|χi (r)|2 |χj (r)|2 + Ni (Ni − 1) dd r|χi (r)|4 .
2 i
i6=j
dd r. This is called
R
In this expression, we are still ignoring some terms that average to zero in the
118
the same single-particle level, as compared to dividing them up into multiple levels.
For example, for N 1,
Eint (Ni = 1) ' 4/3Eint (N1 = N/2, N2 = N/2, Ni>2 = 0) = 2Eint (N1 = N, Ni>1 = 0).
Eint(n)
1.8
Suppose we divide up the particles into n equal groups. Then,
2 Pn
assuming plane waves, Eint (n) = U20 Nn
1.6
ij (2 − δij ) = 1.4
U0 1
2
2 − n
, as plotted at right. 1.2
2 4 6 8 10 n
That is: repulsive interactions favor simple BEC over either the normal state or non-
simple BEC. The opposite is true for attractive interactions. [End of Lecture 19]
We’ll focus on repulsive interactions U > 0. Here ni ≡ b†i bi is the number of bosons
at site i, and n̄ > 0 is a fixed number, a coupling constant. We will work in the grand
canonical ensemble, so n̄ behaves like a chemical potential, specifying the number of
particles on average.
the Hartree-Fock approximation. Let me give some explanation of what’s involved. Write
1X D E
hHint i = Vijkl b†i b†j bk bl (4.12)
2
ijkl
where Vijkl is the potential written in terms of the single-particle χi states (could be plane waves):
Z
Vijkl = dd r1 dd r2 χi (r1 )χj (r2 )V (r1 , r2 )χ?j (r2 )χ?i (r1 ) . (4.13)
The matrix element is of the form we studied earlier. The failure of Wick’s theorem for bosons goes
away in the thermodynamic limit, so let’s ignore it. Then, for i 6= j,
D E D ED E D ED E
b†i b†j bk bl ' b†i bj b†j bk + ζ b†i bk b†j bl ' Ni Nj (δil δjl + ζδik δil ) .
2
On the other hand, if i = j, b†i bk bl = Ni (Ni − 1)δik δkl . This gives
1X X
hHint i ' (Vij,ij + ζVij,ji ) Ni Nj + Vii,ii Ni (Ni − 1).
2 i
i6=j
Taking all the Ni to be large, this gives the expression (4.11) when V is a contact potential.
119
As in the case of the TFIM, the two terms Ht and Hu do not commute and make
differing demands on the low-energy states of the system. Again we’ll make progress
by considering limits of the couplings. Unlike the TFIM, this problem has two dimen-
sionless couplings to play with: t/U and n̄.
√
t/U 1, n̄ 1 : Let b†i = ni eiϕi . Here ni and ϕi are the number and phase
operators. If they satisfy
[ϕi , nj ] = −iδij (4.14)
then I claim that [bi , b†j ] = δij as usual. The idea is
√ √ √ √
[b, b† ] = n [eiϕ , n] e−iϕ + n [ n, e−iϕ ] eiϕ = 1.
| {z } | {z }
iϕ −iϕ
= 2e√n = e2√n
We used the chain rule, e.g. [ϕ, f (n)] = −if 0 (n). This works best when hni is large, so
we don’t divide by zero. Notice that (4.14) implies that there is an uncertainty relation
between particle number n and the phase variable ϕ, they are conjugate variables like
position and momentum.
In terms of these operators, the hamiltonian is
X√ √ X
HBH = −t ni ei(ϕi −ϕj ) nj + h.c. + U (ni − n̄)2 .
hiji i
When t U , we start by ignoring the U term. The cosine wants neighboring phases
to be equal ϕi = ϕj , meaning ϕi ' ϕj for all ij. This means
D E √
b†i ' n̄ eiϕi 6= 0.
This is a BEC.
What are the elementary excitations about this BEC state? When the fluctuations
of ϕi are small (so that we can ignore the periodicity), we can expand the cosine about
its minimum, cos(ϕi − ϕj ) ' 1 − 21 (ϕi − ϕj )2 + · · · , to find (dropping an additive
constant)
X X
HBH ' tn̄ (ϕi − ϕj )2 + U (∆ni )2 , [∆ni , ϕj ] = −iδij .
hiji i
120
Where have we seen such a thing before? This has exactly the form of the harmonic
lattice
X p2 1 X
H= i
+ mω02 (qi − qj )2
i
2m 2
hiji
1 1
with the replacements p → ∆n, q → ϕ, U → 2m , 2 mω02 → tn̄. This means in particular
that there is a phonon-like state (called the phonon) which is gapless and has a linear
dispersion relation,
ka1
ωk = ω0 | sin(ka/2)| ' ω0 a|k|/2
√
(in the case of a chain) with ω0 = 2 tn̄U . Notice
√ that without the interaction term,
there is no dispersion, the sound velocity vs = tn̄U a goes to zero.
Another virtue of the symmetry-breaking interpretation of BEC is that we can
interpret the phonon here as a Goldstone mode for the spontaneously-broken U(1)
symmetry, which acts by ϕi → ϕi + α: rotating all the phases ϕi costs no energy;
therefore making a long-wavelength variation in ϕi will cost a small energy.
t/U 1 : This discussion will be valid for any n̄, even say n̄ = 1. If t = 0,
HBH = U i (ni − n̄)2 wants ni = N , the integer closest to n̄. This is unique, unless
P
which is a simple product state where we put N bosons at each site (in the figure, we
have N = 1). The first excited state is where we take one of the bosons from site x
and put it in site y, so that there are only N − 1 bosons at site x and an extra N + 1
at site y:
This state has energy above the groundstate by an amount proportional to U , inde-
pendent of system size – there is a gap. This has two consequences: it means that
small t can’t dramatically change the groundstate, it will just broaden the degeneracy
of the gapped excited states into a band, as in the TFIM. And it means that this state
is an insulator – it costs a finite amount of energy to move the particles around by
121
even a small amount. It is a new kind of insulator, compared the band insulator of
fermions that we discovered above. It is called a Mott insulator. In some sense it is
much simpler and less quantum mechanical than the band insulator. It is just like a
traffic jam – the particles can’t move because there are other particles in the way, and
they don’t want to sit on top of each other.
We saw above that the number operator ni at each site is conjugate to the phase
operator ϕi , like q and p. The Mott insulator groundstate (4.15) is an eigenstate of
ni . By the usual logic of Heisenberg uncertaintyD principle,
E this means that the value
†
of ϕi is maximally uncertain. This means that bi ∝ eiϕi = 0, since the variable
ϕi wanders all around between 0 and 2π. So the U(1) symmetry is not broken in the
Mott insulator phase, as we knew it couldn’t be because there is a gap, and hence no
Goldstone boson.
This means that at some intermediate value of t/U , something dramatic has to
happen, because the superfluid-ordered gapless state at t U cannot continuously
turn into the boring, dead Mott insulator state at t U . One way to understand
this is that the former breaks the U(1) symmetry and the latter does not. The drama
could be a direct phase transition between them; it could also in principle be some
intervening more interesting phase of matter.
1
n̄ = N + : If n̄ = N + 21 , the story is more interesting. Now there are 2N
2
degenerate groundstates, associated with the choice between N and N + 1 particles
at each site. A small perturbation by Ht can make a big difference in splitting this
degeneracy. The degeneracy is exactly like a spin system, under the identification
|↓ix = |N particles at xi , |↑ix = |N + 1 particles at xi. Under this identification,
Si+ = P b†i P, Si− = P bi P are the raising and lowering operators. Here P is the pro-
jector onto the degenerate subspace. (Si+ )2 = 0, while (b† )2 is not zero; but acting
with b† twice inevitably removes one from the degenerate subspace. The perturbing
Hamiltonian is
X † X
P Ht P = P −t bi bj + h.c. P = −t Si+ Sj− + h.c.
hiji hiji
What does this operator do? In the boson language, it hops the particles around (as
122
long as they don’t go on top of each other), as in the figure.
In the spin language, it’s a ferromagnetic interaction (because of the minus sign) like
you studied on the homework, which makes the spins want to align in the xy plane.
The groundstates are ⊗x |→ix and its images under spin rotations about the z axis:
Siz
P
eiθ i ⊗x |→ix .
D E
This state has Si+ 6 0. In the boson language, this is a superfluid state, b†i 6= 0.
=
123
With only a linearly dispersing mode at low energies, how-
ever, there is a critical velocity below which a non-relativistic
chunk of fluid cannot give up any momentum [Landau]: con-
serving momentum M~v = M~v 0 + ~~k says the change in en-
ergy (which must be negative for this to happen on its own)
is (eliminate v 0 = v − ~k/M ):
1 1 (~k)2 (~k)2
M (v 0 )2 +~ω(k)− M v 2 = −~|k|v+ +~ω(k) = (−v+vs )|k|+ .
2 2 2M 2M
For small k, this is only negative when v > vs .
You can ask: an ordinary liquid also has a linearly dispersing sound mode; why
doesn’t Landau’s argument mean that it has superfluid flow? The answer is that it has
other modes with softer dispersion (so more contribution at low energies), in particular,
we can excite a single-particle state which has ω ∝ k 2 .
Landau’s argument actually overestimates the critical velocity. Actually, the linearly-
dispersing mode in Helium doesn’t stay linear forever, but rather the dispersion relation
has a minimum at finite k; the associated excitation is called a roton. A good way to
think about this minimum is that it is trying to get to ω = 0, in which case it would
describe solidification. More precisely, it would describe an instability of the density
at that wavevector. The momentum space then becomes periodic; this new zero would
be the image under an inverse lattice vector of the zero at k = 0! I’ll say more about
the roton minimum below.
The absence of other low-lying modes in the superfluid can be traced directly to
the indistinguishable nature of bosons. Here are two points of view.
(1) First consider the following comparison between the number of states of bosons
and distinguishable particles. The problem of placing N particles in s single-particle
states i = 1..s is like putting N objects in s boxes.
124
First consider the case when the objects are distin-
guishable and take s = 2. Then
N N
N
X N X
2 = ≡ W (m)
m=0
m m=0
N
where W (m) = = m!(NN−m)!
!
is the number of
m
ways of placing m of the objects in to the first box
(and N − m in the other. This function W (m) is
highly-peaked about a typical configuration where
m ∼ N/2; the extreme configurations where all the
particles are in one box or another W (m ∼ 0) ∼
W (m ∼ N ) W (m ∼ N/2) are hugely suppressed.
The generalization to s orbitals is
X N!
sN = Qs .
{mi } i=1 mi !
This means that even an infinitesimal energetic preference for one such state (say
where all bosons are in the same single-particle state, such as we’ve seen above) will
lead that state to completely win.
Notice that this mechanism breaks down when s N and the distinction between
bosons and distinguishable particles goes away, because they can be distinguished by
their single-particle label. This is a way to predict Tc , by estimating the number of
accessible orbitals s as the number of states with < kB T .
(2) Now here is a more subtle but more informative argument due to Feynman.
(His papers on this subject are fun to read.)
125
Let’s return to the discussion of the form of the wavefunction
from the beginning of this subsection. We want to under-
stand why there are no low-lying excitations besides phonons.
What is a phonon excitation in this language? A phonon is a
compression wave. It’s a state involving small displacements
of each atom, as a result of which the density varies. For the
low-energy phonons, the variation has a large wavelength.
At right I compare a large-amplitude configuration in the
groundstate, and a large-amplitude configuration in a phonon
state of definite ~k ∝ x̂ (with a wavelength of order the size
of the box I’ve drawn).
Consider the wavefunction of the first excited state Φ1 . Again it can be taken to
be real. If our system were in one dimension, it would have to have “a single node”.
That is, for each coordinate ψ(ξ) ≡ Φ1 (ξ, x2 · · · xN ), the space can be divided into
two regions of ξ, one where ψ(ξ) > 0, and one where ψ(ξ) < 0. This is required by
R
0 = hΦ1 |Φ0 i = Φ1 Φ0 .
Now suppose that Φ1 is not a phonon state. This means that it has to be orthogonal
to all the phonon states also. But that means its sign has to be different in configura-
tions which differ by long-wavelength changes of the density. In order for the sign of
Φ1 to vary slowly (required in order for Φ1 to have low energy), it must change sign
when the atoms are moved by a large distance.
126
Let’s try to identify the configuration B where Φ1
is most negative and compare to configuration A.
Making Φ1 low-energy requires making A and B as
different as possible. How to do this? Suppose for
a moment the particles were distinguishable. Take
some atoms from the left of configuration A and put
them on the right. This leaves a hole on the left. To
fix that, we have to take some other atoms from the
right and put them on the left.
Now you can see why Bose statistics comes in.
The atoms are all the same. We can accomplish
the same rearrangement simply by rearranging the
atoms on the left and right individually by small
amounts. But either that’s a phonon (if the sign
doesn’t change between these configurations) or it’s
a high energy state (if the sign does change). There’s
nothing else.
127
Excitation spectrum. [Feynman, Statistical Mechanics, §11.5] I can’t resist say-
ing a little bit more about the spectrum of excitations. Following Feynman, we take a
variational approach (now called the single-mode approximation). We make an ansatz
for the excited-state wavefunction of the form:
X
ψ= f (ri )Φ0 ≡ F Φ0 . (4.16)
i
The idea is we want to excite one of the particles by multiplying by f (ri ) (and the rest
of the particles are just doing what they would do in the groundstate), but we have
to symmetrize, so we sum over i. And we can use the variational principle for excited
states. One way to think about it is that this state has a different momentum from
the groundstate, and the variational principle works within each momentum sector.
A little bit of work shows that
R dN P ~2
d r |∇~ i ψ|2 + V |ψ|2 P ~2 R dN ~
i 2m i 2m d r|∇i F |2 Φ20
hψ| H |ψi = R = E 0 + R ≡ E0 + .
ddN r|ψ|2 ddN r|F |2 Φ20
(4.17)
Here HΦ0 = E0 Φ0 and the trick is to integrate by parts enough to use this equation.
P
So far this works for general F ; now we set F = i f (ri ). The variational quantity
is ≡ N
D
. The numerator is
X ~2 Z ~2
Z
N = ~ 2 2 dN
|∇i f (ri )| Φ0 (r1 · · · rN )d r = N ~ 1 f (r1 )|2 Φ2 (r1 · · · rN )ddN r.
|∇ 0
j
2m 2m
(4.18)
Now we use
Z
1 1 1 1
dd r2 · · · dd rN Φ20 (r1 · · · rN ) = ρ1 (r1 , r1 ) = ψ † (r1 )ψ(r1 ) = hρ(r1 )i = ρ0 =
N N N V
where ρ1 is the one-particle density matrix and we used the fact that the groundstate
is translation invariant, so the density is equal to the average density ρ0 = N/V . Then
~2
Z
N = ρ0 |∇f~ (r)|2 dd r. (4.19)
2m
128
The denominator involves the two-particle density matrix.
XZ Z
D= d rf (ri )f (rj )Φ0 = ρ0 dd r1 dd r2 f ? (r1 )f (r2 )g(r12 )
dN ? 2
ij
(4.20)
where
XZ
ρ0 g(r12 ) ≡ ddN r0 Φ20 δ d (ri0 − r1 )δ d (rj0 − r2 ). (4.21)
ij
This is an integrodifferential equation, a bit scary, but it’s linear (if we treat as a
constant). And it’s translation invariant. So can you guess what the solution is? Trying
the ansatz f (r) = eik·r , we learn that
~2 k 2
(k) = (4.25)
2mS(k)
where Z
S(k) = dd rg(r)e−ik·r (4.26)
129
That’s something that can be measured by neutron scatter-
ing. It’s just a property of the groundstate, in terms of which
we’ve found an expression for the excitation spectrum (an up-
per bound at each k, actually). On general grounds, S(k) has
a peak at about k ∼ 2π r0
. This is just because the particles are
approximately equally spaced at a distance r0 , so g(r) has a
peak at r0 (and a smaller one at nr0 , for n = 2, 3 · · · ; the
form of g(r) is similar to our free-boson calculation, except
that it vanishes for r < a) as in the figure above. This means
that (k) has a minimum there! That’s the roton minimum.
ψk = ρk Φ0
in particular, that the denominator is N S(k) = hΦ0 | ρ†k ρk |Φ0 i. And in fact S(k) ∼ k
at small k for a good reason, as we’ll see below.
The energy spectrum can be measured by inelastic neutron scattering (where one
keeps track of how much energy the neutrons lose) and it looks a lot like this function
~2 k2
2mS(k)
(which gives a rigorous upper bound on the right answer). Landau had earlier
predicted a spectrum like this based on phenomenological considerations. It is roton
excitations which make up the ‘normal fluid fraction’ that leads to dissipation in the
130
experiments described above. They are responsible for the fact that the critical velocity
is smaller than vs . And they can be seen to change the specific heat from the T 3 con-
tribution of phonons once the temperature is big enough that they are not Boltzmann
suppressed, i.e. kB T ∼ (kroton ). (This was the origin of Landau’s suggestion.)
The roton minimum is a signature of incipient solidification. As we saw, the maxi-
mum in S(k) arises because of a peak in the probability of finding the atoms equally-
spaced. If it indeed formed a solid with spacing a, there would be a pole in S(k) at
k = 2π
a
, and hence a zero of the dispersion. This is just the image of k = 0 under an
inverse lattice vector – the momentum space becomes periodic.
Here is some poetry motivating the ansatz (4.16). Consider again the configuration
A where ψ is most positive, and B where ψ is most negative. To keep the energy small,
we want these configurations to be maximally different. But we saw above that it is
no use moving the atoms far away. Really the best we can do is move each of them
by about half their separation – the atoms in configuration B should be in between
those in configuration A. So we want to choose f (r) = 1 if r is occupied by an atom
in configuration A, and f (r) = −1 if r is occupied by an atom in configuration B, and
smoothly interpolate in between.
[Pines and Nozieres, The Theory of Quantum Liquids] What is the origin of the
name Single-Mode Approximation (SMA)? Actually one can arrive at the same conclu-
sion about the spectrum (though without learning about the wavefunction) by assuming
a particular form of the spectrum, and hence of the dynamical structure factor (which
we discussed earlier in the context of the TFIM) for the density:
X
S(k, ω) = δ(ω − ωj )| hΦj | ρk |Φ0 i |2 (4.28)
j
We can combine this information with various Inviolable Facts about the structure
131
factor called sum rules. Later we will see why they are true. The two that we’ll need
are
N ~2 k 2
Z
f -sum rule : dωωS(k, ω) = (4.31)
2m
Z
S(k, ω) N
compressibility sum rule : lim dω = (4.32)
k→0 ω 2mvs2
where vs is the speed of sound. To see why these might be useful here, observe that
the variational energy above is exactly
~2 k 2
dωωS(k, ω) (4.31),(4.30) N2m
R
k = R = .
dωS(k, ω) N S(k)
But forget for a moment the variational calculation, and just plug in the SMA
expression to the two sum rules. They give
k2 1 S(k)
= S(k)k , 2
= lim . (4.33)
2m 2mvs k→0 k
~ k 2 2
The first equation reproduces our dispersion relation k = 2mS(k) . Plugging this into
the second relation then says
k→0 |k|
S(k) → .
2mvs
So indeed we can guarantee that S(k) is linear at small k as promised. Notice the
consistency check that for small k this implies
k→0
k → vs |k|
so that indeed vs is the speed of the sound mode. The moral reason for the f sum
rule is the conservation of particle number, which is also the symmetry that’s broken
to produce the phonon as a Goldstone mode.
Finally, what is the origin of the name ‘roton’ ? I believe it
was introduced by Landau. Is it a good name? f (r) = eik·r
is a wave moving from an A site to a B site. In Feynman’s
paper linked above, he describes A and B as differing by the
motion of a small group of particles around a ring, as in the
figure at right (from his paper).
Feynman says this is like a small vortex ring. I’m not sure if this is a correct picture.
An attempt to make it more correct is the following. Imagine making a wavepacket
centered on the value of k at the minimum of the dispersion, so that the excitation is
localized in space. The fact that it is a minimum means ∂K = 0 – the group velocity
vanishes, so the excitation doesn’t move. On the other hand, the particle current in
132
the region of the excitation is nonzero. This means that our ansatz cannot actually
solve the Schrödinger problem, which implies current conservation. All of this is some
words to motivate an improvement of the ansatz, something like:
X P
ψ= eik·ri + j h(rij ) Φ0 (4.34)
i
which can be chosen to conserve current. This is called a backflow improvement, since
it describes the particles circulating back around from B to A. Obviously, we can find
a better state if we generalize the trial wavefunction. This class of wavefunctions does
seem to give a good improvement. I’m not sure about the words that go with it.
If you want to read more about this, a nice resource is this interesting article by
Pines.
Part of the reason to devote some effort to this line of thought is that it plays an
important role in our understanding of the excitation spectra of quantum Hall states.
That is a subject for another time.
133
5 Interacting fermions
[We ran out of time to talk about the topics that follow.]
[Feynman §7.1] We’ve seen several examples of interacting spin systems, where I just
wrote down a Hamiltonian in terms of (tensor products of) Pauli matrices. I explained
a bit what might have happened to the position degrees of freedom of the electrons.
But where does such a spin interaction come from? One might think it has something
to do with magnetic dipole-dipole interactions, but actually in condensed matter there
is a larger contribution.
with V (r1 , r2 ) = V (r2 , r1 ). Suppose that ψα and ψβ are eigenstates of h0 with en-
ergies α and β . The hamiltonian H has a symmetry under interchange of the two
particles, so the groundstate must be an eigenstate of this operation, either symmetric
or antisymmetric. With an eye towards considering identical fermions, consider the
symmetric and antisymmetric states
√
ΨS/A ≡ (Ψαβ ± Ψβα )/ 2, Ψαβ (r1 , r2 ) ≡ ψα (r1 )ψβ (r2 ).
Let’s use these as trial wavefunctions for H. Their expected energies are
Z Z
ES/A ≡ ΨS/A H ΨS/A = α + β + d r1 dd r2 Ψ?S/A (r1 , r2 )V (r1 , r2 )ΨS/A (r1 , r2 )
d
(5.1)
Z Z Z Z
1
= α + β + Ψ?αβ V
Ψαβ + Ψβα ± Ψ?βα V Ψ?αβ V Ψβα ± Ψ?βα V Ψαβ
2 1,2 1,2 1,2 1,2
Z Z
?
= α + β + Ψαβ V Ψαβ ± Ψ?βα V Ψαβ
1,2 1,2
≡ α + β + I ∓ J . (5.2)
134
Now consider the same Hamiltonian for spinful fermions:
X Z Z
H= aασ aασ α + d r1 dd r2 V (r1 , r2 )ψr†1 ,σ ψr†2 ,σ0 ψr2 ,σ0 ψr1 ,σ .
† d
Notice that the interactions are completely spin-independent. Again we’ll focus on two
orbitals α, β. The states of two fermions are completely antisymmetric (AS) under
interchange of the two particles:
But there are four such states. The total antisymmetry can be accomplished in two
ways: symmetric in space, and AS in spin, or AS in space, symmetric in spin. The
symmetric combination of two spin-half particles is a triplet, spin s = 1 (three states),
while the AS combination is the singlet, spin s = 0 (one state). We can write these
states as
1 1
|Si = √ a†ασ a†βσ0 εσσ0 |0i = √ a†α↑ a†β↓ − a†α↓ a†β↑ |0i
2 2
1
|A, 1i = a†α↑ a†β↑ |0i , |A, 0i = √ a†α↑ a†β↓ + a†α↓ a†β↑ |0i , |A, −1i = a†α↓ a†β↓ |0i .
2
Their wavefunctions are, for example (using (1.34): a†ασ = r a†rσ ψα (r))
P
1 √
hr1 σ, r2 σ 0 |Si ≡ √ h0| ar1 σ ar2 σ0 |Si = ΨS (r1 , r2 ) (δσ↑ δσ0 ↓ − δσ↓ δσ0 ↑ ) / 2.
2
1 √
hr1 σ, r2 σ 0 |A, 0i ≡ √ h0| ar1 σ ar2 σ0 |A, 0i = ΨA (r1 , r2 ) (δσ↑ δσ0 ↓ + δσ↓ δσ0 ↑ ) / 2.
2
Since the Hamiltonian is spin-independent, the variational energies of these states
only depend on the orbital (position) part. But the symmetry of the states is precisely
correlated with the total spin: the spatially-symmetric state has total spin Casimir
S2 = S~ ·S~ = s(s + 1) = 0, while the spatially-AS state has S 2 = 2. Here S~=S ~1 + S
~2
is the total spin. Therefore the effective hamiltonian on the spin degrees of freedom is
s(s + 1) ~ ·S
S ~
Heff = α + β + I ∓ J = ES + (EA − ES ) = ES + 2J ~1 · S
= 2J S ~2 + const
2 2
where in the last step we used
~ ·S
S ~ = (S~1 + S
~2 )2 = 2S
~1 · S
~2 + S 2 + S 2 = 2S ~2 + 3 .
~1 · S
1 2
2
Thus a spin-symmetric interaction combines with the Pauli principle to produce a spin-
spin interaction. Notice that the “exchange energy” J is so called because of its origin
in (5.2). It depends on the overlaps of the single-particle states. In a solid, it turns
out to be much bigger than the direct dipole-dipole interaction between the magnetic
moments.
135
5.2 Ordering in Fermi systems.
[Leggett, §2.4] So the spins in an insulating magnet are the electron spins. This means
that the spin operator is really
~r = 1 ψrσ
S †
~σσσ0 ψrσ0
2
– it is really a fermion bilinear operator. When the magnetization is nonzero m =
hS a i 6= 0, it means that there is an expectation value for a fermion bilinear, which is
a bosonic operator. It is not quite a pair of fermions which condenses, since one of
the operators is an annihilation operator, but this is an example of a fermionic system
where there is long-range order (in this case diagonal long-range order).
Another example is helium. Above I said a collection of helium atoms was a good
example to keep in mind in our discussion of bosons with short-range repulsion. But a
helium atom is only approximately a bosonic particle; more microscopically it is made
up of electrons and a helium nucleus, at least some of which are fermions. There is a
more microscopic density matrix for these particles, which cannot have ODLRO of ρ1 .
On the other hand, clearly the BEC phenomenon happens.
What is the analog of ρ1 , by which we gave a non-weak-coupling definition of BEC?
Well, the idea is simply that reduced density matrices for a larger collection of particles
can have a macroscopic eigenvalue. The next case beyond ρ1 is the 2-particle density
matrix:
X X X
ρ2 (r1 σ1 , r2 σ2 ; r10 σ10 , r20 σ20 ) ≡ ps Ψ?s (r1 σ1 , r2 σ2 , r3 σ3 · · · rN , σN )Ψs (r10 σ10 , r20 σ20 , r3 σ3 · · · rN , σN )
s σ3 ···σN r3 ···rN
(5.3)
= ζρ2 (r2 σ2 , r1 σ1 ; r10 σ10 , r20 σ20 ).
A few properties of ρ2 :
136
• Not every matrix with these properties actually arises as the reduced density
matrix of some state.
137
5.3 Instabilities of a Fermi surface to repulsive interactions
Consider X X
H = −t c†xσ cyσ + h.c. + U (nx − 1)2 = Ht + HU (5.4)
hxyi,σ x
k = −2t(cos kx + cos ky );
138
case, we have to go to second order, because the perturbing hamiltonian always takes
us out of the degenerate subspace. Second-order degenerate perturbation theory says
that the matrix elements of the effective hamiltonian (for |ai , |bi ∈ X, the degenerate
subspace) are
X ha| ∆H|nihn|∆H |bi
ha| Heff |bi = − .
En − EX
n∈X
/
The crucial ingredient can be seen by considering just two sites. A sketch of the
processes involved are:
Notice that the whole calculation is SU(2)-spin-rotation symmetric. The result, as you
will find on the homework, is
4t2 X ~ ~
Heff = + Sx · Sy .
U
hxyi
D E
~S~x=(x,y) = (−1)x+y ẑ
or its images under SU(2) spin rotations. (Note that if the lattice is frustrated, it is
much more difficult to determine the groundstate of the AFM Hamiltonian, and the
result can be a much more interesting quantum state, such as a spin liquid. In this
terminology, the idea is that a magnetic state such as the Neel state involves ordering
of the spins, and is therefore like a spin solid. So solid is to liquid as magnet is to spin
liquid.)
Now, what about when U/t is not infinite but t is not zero? To learn something
about what happens there, we can do mean field theory.
139
Mean field theory of square-lattice Hubbard model at half-filling. To get
a hint of what is a good mean-field ansatz, observe that at each site
2
(n − 1)2 = 1 − c† σ z c = 1 − (n↑ − n↓ )2 .
The easiest way to see this is to observe that the BHS is diagonal in the number basis,
and simply evaluate both sides in all four possible states of a single site:
n↑ n↓ 00 01 10 11
LHS 1 0 0 1
RHS 1 0 0 1
Notice that both are smaller when there is a single electron at the site, n = n↑ +n↓ = 1.
2
So we can interpret large U as favoring 1 = c† σ z c , which can happen for either value
of Sz = c† σ z c = ±1.
To develop a mean field theory, we can make the following replacement in the
Hamiltonian:
2 2
c† σ z c 2 c† σ z c c† σ z c − c† σ z c .
Why do we need to subtract the constant hSz i2 ? It matters if we want to treat hSz i
as the variational parameter, and determine it by minimizing hHi. You can check that
in the case of the TFIM it is required in order to reproduce the result of our product
state variational calculation. Another very useful way to think about this is that we
σ2
are using the Hubbard-Stratonovich trick: replace U (Sz )2 by −Sz σ + 2U where σ is an
z
auxiliary bosonic variable. The equations of motion for σ are σ = U S , and plugging
back into the Hamiltonian gives back our previous Hamiltonian. If instead we ignore
the fluctuations of σ, we arrive at the mean-field hamiltonian.
For our trial state, we choose one with
D E
c~†x σ z c~x = hSz (~x)i = M (−1)x+y , ~x = (x, y),
since we had such luck with the AFM solution at large U . We’ll choose M to produce
the best energy expectation. To do so we must find the groundstate energy of
X † X
HMF = ckσ ckσ k − 2U M (−1)x+y c†x σ x cx + U M 2 V
kσ x
where V is the number of sites of the lattice. Fortunately (not an accident), this is an
Easy Problem, i.e. HMF is quadratic in canonical fermion operators. The idea here is
that we will choose as our variational state the groundstate of a gaussian problem (the
fermionic analog of a product state).
140
The AFM ordering doubles the unit cell and therefore halves the Brillouin zone
(we’re using (−1)x+y = ei(π,π)·(x,y) ):
X 0
X
(−1)x+y c†x σ z cx = c†k σ z ck+Q + h.c., Q
~ ≡ (π, π)
x k
in units with lattice spacing a = 1. Let’s use 0k to denote a sum over half the previous
P
Brillouin zone (BZ). This folding of the BZ means that where previously we had two
bands (one for each spin), we now have four states at each value of k, a 4 × 4 matrix.
Fortunately it is a very simple 4 × 4 matrix:
0
!
z
X −2U M σ c
HMF = c†k , c†k+Q k
z
k
+ U M 2 V.
k
−2U M σ k+Q ck+Q
−2U M σ z
k
h(k) ≡ = k Z − 2U M σ z X
−2U M σ z −k
are q
± 2k + (2U M )2 ≡ ±Ek
independent of the sign of σ z . (Recall that the eigenvalues of the 2 × 2 matrix h =
p
h0 1 + hx X + hy Y + hz Z are h0 ± h2x + h2y + h2z .)
In terms of the creation and annihilation operators for the normal modes, d±σ (k)27 ,
0
X
HMF = Ek d†+σ (k)d+σ (k) − Ek d†−σ (k)d−σ (k) + V U M 2 .
k,σ
27
We actually don’t need the form of these normal modes for our purposes here, but for completeness,
they are
z
d+σ (k) = −v(k)σσσ 0 cσ 0 (k) + u(k)cσ (k + Q) (5.6)
z
d−σ (k) = u(k)cσ (k) + v(k)σσσ 0 cσ 0 (k + Q) (5.7)
−vσ z u
where is the matrix which diagonalizes h(k) above, and
u vσ z
r
1 k
q
uk = √ 1− , vk = 1 − u2k .
2 Ek
141
The groundstate at half-filling is
Y
|MF gsi = d†−σ (k) 0̃
k
where 0̃ is the vacuum with d±σ 0̃ = 0 – it is not the state with no c-electrons,
which instead satisfies c(k) |0i = 0!! This state involves completely filling the bottom
band – it is a band insulator – and has energy
0
X
E0 (M ) = −2 E(k) + M 2 U V. (5.8)
k
We would like to solve this condition for the variational parameter M , the amount of
AFM order. In the thermodynamic limit, the condition is
d̄d k
Z Z
dg()
1 = 4U p = 4U
BZ0 4U 2 M 2 + 2k 4U M 2 + 2
2
where Z
g() = d̄d kδ( − k )
BZ0
is the density of states at energy . The dominant contribution comes from near the
Fermi surface (where the denominator is smallest), where we can treat g() ' g(F ) as
a constant. So we can approximate the integral as
Z t
d t
1 ' 4U g(F ) √ = 8U g(F ) log (5.10)
−t 2 + 4U 2 M 2 2|U M |
whose solution is
t − 8g(1 )U
|M | '
e F . (5.11)
U
Notice that as U → 0+ , M vanishes (along with all of its U -derivatives). (5.11) is not
a formula we could ever get in perturbation theory in U .
The next figure is a cartoon of what happened. We started with a Fermi surface
(top); the spin density wave hS z i = M eiπ(x+y) broke translation invariance and folded
the BZ in half (middle); this allows the two parts of the Fermi surface to hybridize,
leading to a new bandstructure (bottom) which has a lower groundstate energy. (By
142
‘hybridize’ I just mean the usual thing that happens when two levels with the same
quantum numbers try to cross – there is an avoided crossing28 .)
A simpler, lower-dimensional example where the cartoon is a
precise description is in the 1d case, where this is called the
spin-Peierls instability of 1d metals. The ordinary Peierls
instability of 1d metals involves instead of spin-ordering, a
condensation of phonons which enlarges the unit cell, and
produces an energy gap in the electronic spectrum, which
lowers the energy of the filled bands. (See the homework and
notes from discussion section.)
The role of half-filling here is to make it so the integrals
behave as in the 1d case. The crucial phenomenon is that
different patches of the Fermi surface are separated by the
ordering wavevector Q (k = −k+Q ), as we used in (5.5). If
not for this property, there would have been an extra term
in h = h0 1 + hx X + hz Z proportional
p to the identity, and we
would have had E(k) = h0 ± hx + h2z , which would remove
2
But this is not the most generic 2 × 2 hermitian matrix with this property.
The most generic such matrix is instead
1 1
h = λ1 + λ − Z + hx X + hy Y = h0 1 + ~h · ~σ .
2 2
143
superconductor, such as La2 CuO4 . The copper and oxygen atoms form planes of square
lattices on which the electrons can hop. The undoped material is indeed an antiferro-
magnet. Increasing the doping leads eventually to the destruction of the AFM order
and the formation of a superconducting state.
Next we’ll see what happens with attractive interactions. I claim that the result is
superconductivity – BEC of electron pairs b† (x) = c†↑ c†↓ .
But aren’t the Coulomb interactions between electrons repulsive, since they have like
charge? Electrons in a solid also interact with each other through the phonon modes.
The idea is familiar to anyone who has sat on a squishy mattress at the same time
as someone else: each person makes an indentation in the mattress – some excitation
of (many!) phonons. That indentation is then a potential well into which the other
person is attracted.
To give a slightly more quantitative account of this phonon-mediated attraction
between electrons, consider
X † X †
He-ph = −t ci cj + h.c. + g ci cj uij + H(u).
hiji hiji
Here we can think of uij ≡ qi − qj as representing the phonon modes, and we can take
H(u) as the balls-and-springs model with which we began the course. We will regard
g as a small parameter. The groundstate at g = 0 is just
We treat g as a perturbation:
X X
∆H = g c†i cj uij = g(q)c†p−q cp a†q + h.c.
ij p,q
144
Consider an initial and final state with two electrons
and no phonons
|k1 , k2 i 7→ |k1 − q, k1 + qi
X
∆Heff = Vq,k1 ,k2 c†k1 −q c†k2 +q ck1 ck2 .
k1 ,k2 ,q
There are many other points of view from which to reach this conclusion. Perhaps
the easiest is using path integrals. Very sketchily, the basic idea is
Z
~ 2
[Dq]eiS[c]+i ((q̇) −(∇q) −gqc c) = eiS[c]+ c cDc c
2 †
R RR † †
where D = hqqi.
So the conclusion is that in some metals, there is a net attraction between electrons.
For a more complete discussion of the competition between the screened Coulomb force
and the phonon-mediated attraction, I recommend §5.2 of Leggett’s book.
Cooper problem. [Baym, chapter 8] Now we will examine a simple model which
demonstrates that in the presence of attractive interactions, electrons near a Fermi
145
surface will pair up. The model, due to Cooper, is a brutal simplification of the actual
situation. We consider two opposite-spin electrons above an inert Fermi sea:
X X
|ψi ≡ ak1 ,k2 ψk†1 ↑ ψk†2 ↓ |FSi ≡ ak1 k2 |k1 k2 i .
k1 ,k2 k1 k2
We only include (attractive) interactions between these two electrons, and ignore their
interactions with the electrons filling the Fermi sea, and the interactions between the
electrons filling the Fermi sea. The result will be a boundstate even for arbitrarily
weak interactions. Later we’ll come back and discuss the shortcomings of this model
and its conclusions.
The Schrodinger problem H |ψi = E |ψi is
X
Eak1 k2 = (k1 + k2 )ak1 k2 + hk1 k2 | V |k10 k20 i ak10 k20 . (5.12)
k10 k20
The 0 on the sum restricts the integration region to kF < |K/2 ± k 0 | < ka . Dividing
the BHS by E − k1 − k2 and summing 0k gives
P
0 0 0
X v0 X 1 X
ak (K) = − ak0 (K)
k
V k E − k1 − k2 k0
P0 P0
but now k ak (K) = k0 ak0 (K) is a nonzero constant and therefore
0
v0 X 1
1=− . (5.14)
V k E − k1 − k2
146
Following Baym, we can solve this equation
graphically. Let
0
1 X 1
Φ(E) ≡ .
V k E − k1 − k2
The function Φ(E) has poles at every possible energy of a two-particle state in the
free problem, (k1 ) + (k2 ) (for k1,2 in the range where the interaction is nonzero). The
crossings of Φ(E) = − v10 that occur at E > 2F are just like the corresponding states
of the unperturbed system. These states become a continuum as V → ∞. But you
can see that there is one crossing that occurs for E < 2F , at E = Eb ≡ 2F − 2∆.
This is a boundstate, below the 2-particle continuum, which wouldn’t exist if not for
the attractive interaction.
Let’s focus on K = 0, which will give the largest binding energy. Then k1 = k, k2 =
−k are on opposite sides of the Fermi surface.
ka
d̄d k a
Z Z
dg()
ΦK=0 (E) = =
kF E − 2k F E − 2
where g() ≡ d̄d kδ(k − ) is our friend the density of states at single-particle energy
R
, and a ≡ ka . Near the Fermi surface again we can approximate g() ' g(F ) as a
constant, and we find
Z a
d E<2F g(F ) 2a − E ! 1
ΦK=0 (E) ' g(F ) = − log =− .
F E − 2 2 2F − E v0
The solution for the binding energy is
a − F − 2
∆= 2 ' D e v0 g(F )
e v0 g(F ) − 1
where D is the Debye energy, roughly the maximum energy of a phonon excitation,
and we used the fact that v0 g(F ) is small because the coupling is weak. Again this is
non-perturbative in the coupling v0 .
1
The momentum-space wavefunction of the boundstate is ak (K) ∝ E−k1 −k2
, so in
position space it is
0
iK· 1 2 2 1
r +r X eik·(r1 −r2 )
ψ(r1 , r2 ) ∼ e .
V k E − k1 − k2
147
In the thermodynamic limit, the relative wavefunction is
Z 0
eik·r12
d K=0 sin kF |r12 | |r12 |
d̄ k ' sin ,
E − k1 − k2 |r12 | ξ
where ξ = 2km∆
F
is the size of the pair. Notice that the state is a relative s-wave, i.e. it
is spherically symmetric. This is made possible by the fact that the two electrons have
opposite spin.
In an actual metal, the role of our two special electrons is played by every electron
near the Fermi surface. They all want to form pairs with their antipodal partners (and
less so with other electrons), and condense. If we apply an electric field, the Fermi
surface shifts and they’ll want instead to form pairs with definite nonzero K. This is
a supercurrent.
This actually happens to electrons in metals, in the fermionic isotope of helium 3 He
at T < 0.0026◦ K, and to nucleons in the core of a neutron star.
What role is played by the Fermi surface? If we redo the analysis with kF= 0,
√
the conclusion is very different. Instead Φ(E) = πm2 −ka + 2 m∆ arctan 2√km∆
a
. In
order for this to hit −1/v0 , we need
1 ka m
< 2
v0 π
– an interaction strength above a nonzero threshold value is required in order to create
a boundstate. In the case with a Fermi surface, an arbitrarily weak interaction suffices.
A warning: in a real metal, the Fermi sea electrons are not inert! This leads to an
important modification of the outcome. Instead of a boundstate at negative E − 2F ,
instead one finds an instability – a mode whose energy has a positive imaginary part.
The instability is to the formation of a condensate of Cooper pairs. More on this next.
148
5.5 Instabilities of a Fermi surface to attractive interactions
This looks very much like our picture of a condensate of the bosons created by
Cooper pairs. (This is not quite a canonical boson, but it can still condense.) Notice
that we’ve made no assumptions about the filling, µ is arbitrary.
To make progress away from U = −∞ let’s rewrite the interaction to develop a
mean field theory that incorporates this bose condensation of pairs.
the second term is a shift of the chemical potential which we absorb into µ, and the
third term is a constant which we ignore. The remaining actual interaction is
2
U n2x = U c†x↑ cx↑ + c†x↓ cx↓ (5.16)
= 2U c†x↑ cx↓ c†x↓ cx↓ = −2U c†x↑ c†x↓ cx↓ cx↓ (5.17)
149
since c2 = 0. Now we approximate the hamiltonian by
X X D † † E D E D ED E
HMF = c†kσ ckσ (k − µ) − 2U c c c c
x↑ x↓ x↑ x↓ + c† †
c
x↑ x↓ c c
x↑ x↓ − c† †
c
x↑ x↓ c c
x↑ x↓
k x | {z } | {z }
≡∆ =−∆?
(5.18)
X X
= c†kσ ckσ (k − µ) − 2U ∆cx↑ cx↓ + ∆? c†x↓ c†x↑ − 2V U |∆|2 (5.19)
k x
X
= c†kσ ckσ (k − µ) + 2U ∆? c†k↑ c†−k↓ + ∆ck↑ c−k↓ − 2U |∆|2 V. (5.20)
k
In the last step we made the ansatz that ∆ is uniform in space, figuring that this
will minimize the energy. Notice
D the Eweird sign
D in E front of the |∆|2 term; this is a
?
consequence of fermi statistics cx↑ cx↓ = − c†x↑ c†x↓ = −∆? . Notice that H had a
symmetry under c → eiθ c, associated with particle number conservation. HMF does
not have this symmetry, only a Z2 subgroup c → −c. The way to think about this
is: we are using HMF as a way to construct a trial wavefunction for H (in particular
its groundstate). Based on the above discussion, we expect the groundstate of H to
spontaneously break this U(1) symmetry, so we are building this into HMF .
Solution of HMF . HMF is quadratic in canonical fermion operators, so we should
be able to solve it. It has an unfamiliar ingredient, namely c† c† terms. The name for
the general strategy for getting rid of such terms is called a Bogoliubov transformation.
The idea, as you saw on the homework, is to introduce new creation and annihilation
operators which mix the cs and c† s in terms of which the Hamiltonian does not have
such terms. Here a particle-hole transformation on just the down spins does the job:
dk↓ ≡ c†−k↓ =⇒ d†k↓ = c−k↓ (5.21)
dk↑ ≡ c+k↑ =⇒ d†k↑ = c†k↑ . (5.22)
You can check these are also canonical fermion operators. In terms of them,
X X X X †
HMF = (k − µ) d†k↑ dk↑ − (k − µ) d†k↑ dk↑ + (k − µ) + 2U dk↑ dk↓ ∆? + h.c. − 2U V |∆|2
k k k k
!
− µ 2∆? U d
† † k
= dk↑ , dk↓ k↑
− U |∆|2 V + const (5.23)
2∆U −(k − µ) dk↓
(where by constant, I mean independent p of ∆). The matrix is σ z (−µ)+σ x (−2U Re∆)+
σ y (−2U Im∆) whose eigenvalues are ± (k − µ)2 + 4U 2 |∆|2 . Therefore (suppose we’re
at less than half-filling, so we partially fill the bottom band)
kF
X p
E0MF = +|U ||∆| V − 2
(k − µ)2 + 4U 2 |∆|2 .
k
150
In the thermodynamic limit, the energy density is
Z kF p Z µ p
MF 2 d 2
E0 /V = |U ||∆| − d̄ k (k − µ)2 + 4U 2 |∆|2 = |U ||∆| − dg() ( − µ)2 + 4U 2 |∆|2 .
We would like to find the gaussian state |MFi which minimizes the expectation value
P D † E
hMF| H |MFi (at fixed average particle number N = i ci ci . By gaussian state, I
mean a Slater determinant α d†α 0̃ , where the dα are some collection of modes, to
Q
be determined.
A useful device for accomplishing this is to define an auxiliary quadratic Hamilto-
nian whose groundstate is |MFi. In this case, we can consider
X †
X † †
2
HMF = T − µi ciσ ciσ + ∆i ci↑ ci↓ + hc − |∆i | .
i i
So the thing we want to extremize over the choice of ds and occupation numbers is
!
XD † E
F ≡ hMF| H |MFi − λ ci ci − N − T S. (5.26)
i
151
I include the T S term in the free energy to emphasize that the same strategy works
at finite temperature; let’s set T = 0 from now on. Because |MFi is a gaussian state,
we can compute (5.26) using Wick’s theorem (all expectation values below are in the
state |MFi:
!
XD † E X D † † E D E D ED E D ED E
F = hTi−λ ciσ ciσ − N +V ci↑ ci↓ ci↓ ci↑ + c†i↑ ci↑ c†i↓ ci↓ − c†i↑ ci↓ c†i↓ ci↑ .
i i
(5.27)
D |MFi
In this expression I actually haven’t assumed that E is translation invariant. But
†
let’s assume time-reversal invariance, so that c↑ c↓ = 0 and we can drop the last
term.
Now consider varying (5.27) with respect to the choice of modes and filling:
X D † † E D E D
† †
ED E
†
δF = δ hTi − λδ cσ cσ + V ci↑ ci↓ δ ci↓ ci↑ + δ ci↑ ci↓ ci↓ ci↑ (5.28)
i
D ED E D E D E
+δ c†i↑ ci↑ † † †
ci↓ ci↓ + ci↑ ci↑ δ ci↓ ci↓ .
(5.29)
But now consider the free energy associated with the mean field hamiltonian:
!
XD † E
FMF = hMF| HMF |MFi − λ ci ci − N − T S.
i
Since |MFi is the groundstate of HMF with the right particle number, the variation is
zero:
X D E X D E
δFMF = δ hTi − (µi + λ)δ c†iσ ciσ + ∆i δ c†i↑ c†i↓ + hc . (5.30)
i i
These are the self-consistency conditions we should impose. Notice that the terms
(5.29) lead to a renormalization of the chemical potential due to the interactions.
152
6 Linear response
[Pines and Nozieres, chapter 2] What happens in an experiment? Pretty inevitably,
what happens involves two steps (1) the experimenter pokes at the system somehow.
This means changing the hamiltonian by H = H0 + V (t) for some perturbation V (t).
(2) the experimenter measures the response of the system.
Here we make two further assumptions. First, that the system starts in equilibrium
before the experimenter shows up. This means either the groundstate or the canonical
ensemble at temperature T .
Second, we assume that the poking is weak enough that the response is linear in the
perturbation. This means that the response is entirely a property of the unperturbed
system. In particular, it will be determined by equilibrium correlation functions, which
as we’ve seen and will see in more detail below are related to the low-lying spectrum
of the system.
A good simple example to keep in mind (during the more abstract bits below) is
the response of an RLC circuit to an applied voltage. As in that example, the response
can come in two varieties: dissipative and reactive (or reversible).
So suppose we kick the system by an operator OB . What this means is that the
source is a time dependent perturbation to the Hamiltonian, H = H0 + V (t):
Z
V (t) = dD−1 xφB (t, x)OB (x) .
The initial state is ρ0 = e−βH0 /Z or ρ0 = |Φ0 ihΦ0 |, the projector onto theR groundstate.
t 0 0
In the last step we are using interaction picture, where U (t) = T e−i V (t )dt , and
OA (t, x) = eiH0 t OA (x)e−iH0 t is the unperturbed evolution.
I think I should explain this step a bit more. First, in the Schrödinger picture, only
the states evolve: i∂t ρ(t) = [H, ρ(t)] so that ρ(t) = UH (t)ρ0 UH−1 (t), with UH = e−iHt .
The expectation value we want is
trρ(t)OA (x) = trρ0 UH−1 (t)OA (x)UH (t) = trρ0 U −1 (t)OA (x, t)U (t)
153
where at the last step we wrote
This last expression is the interaction picture evolution of the operators – they evolve
by the unperturbed Hamiltonian. The first expression in (6.1) defines the interaction
picture evolution operator U = UH U0−1 . Using i∂t UH = UH H, i∂t U0 = H0 U0 , we have
where V (t) is the evolution of V by the unperturbed hamiltonian. What is the solution
of i∂t U = V (t)U ? I claim, by the Fundamental Theorem of Calculus, that
Z t
U (t) = U (0) − i dt1 V (t1 )U (t1 ) (6.2)
0
Z t Z t1
(6.2)
= U (0) − i dt1 V (t1 ) U (0) − i dt2 V (t2 )U (t2 ) (6.3)
0 0
= ··· (6.4)
X∞ Z t Z t1 Z tn−1
n
= (−i) dt1 dt2 · · · dtn V (t1 )V (t2 ) · · · V (tn )U (0). (6.5)
n=0 0 0 0
In the first step I used the fact that the operators are already time ordered (this followed
from the differential equation we are solving, since the V always acts from the left). In
the second step we used the fact that the time-ordered integrand doesn’t change if we
permute the labels on the times. So we can just average over the n! possible orderings
of n times. If we pull out the time-ordering symbol, this is an exponential series:
Rt 0 0
−i dt V (t )
U (t, ti ) = T e ti .
154
Now linearize in the small perturbation:
Z t
δhOA i(t, x) = −iTr ρ0 dt0 [OA (t, x), δH(t0 )]
Z Z t
D−1 0
= −i d x dt0 h[OA (t, x), OB (t0 , x0 )]iφB (t0 , x0 )
Z
= dD x0 GR 0
OA OB (x, x )φB (x )
0
The retarded Green’s function (or response function) for two observables OA and OB is
GR
OA OB (t, x) ≡ −iθ(t)h[OA (t, x), OB (0, 0)]i (6.7)
where θ(t) = 1 for t > 0, else zero, is the usual Heaviside function. The time evolution
of the operators here is by the unperturbed Hamiltonian O(t, x) = eiH0 t O(x)e−iH0 t ,
as above. So we care about the retarded Green’s function because it determines what
hOA i does if we kick the system via OB .
Now Fourier transform:
δhOA i(ω, k) = GR
OA OB (ω, k)δφB (ω, k) (6.8)
where Z
GR
OA OB (ω, k) = −i dD−1 xdt eiωt−ik·x θ(t)h[OA (t, x), OB (0, 0)]i .
Note that if we want to think about the perturbations in Fourier space, e.g. for a
single mode Z
V = dd x eiq·x−iωt ϕ(q, ω)OB (q, ω)
so that the perturbation goes away at early times. This is required so that we are
actually describing a system that begins in its groundstate.
Linear response, an example.
perturbation: an external electric field, Ex = iωAx
R
couples via δH = dd xAx J x where J is the electric current. That is, OB = Jx .
response: also the electric current, OA = Jx . Then let’s look at (6.8) in this
case. It’s safe to assume hJiE=0 = 0. (Actually this can be rigorously proved in some
circumstances and goes by the name of Bloch’s (other) theorem.) So (6.8) takes the
form
Ex
hJ(ω, k)i = GR R
JJ (ω, k)Ax (ω, k) = GJJ (ω, k)
iω
155
Compare this expression to Ohm’s law: J = σE, which defines the conductivity σ.
(Really it is a tensor since J and E are both vectors.)
GR
JJ (ω, k)
=⇒ Kubo formula: σ(ω, k) =
iω
As a second example about which we’ll say a bit more, consider the case where
OA = OB = ρ, the density. Then
XZ
V = d̄ωρ†q ϕ(q, ω)e−iωt + h.c.
q
where ϕ(q, ω) is the fourier transform of some scalar potential. For example, if we
shoot at the system a heavy probe particle with velocity V and position R; then
ϕ(q, ω) = 2πVq e−iq·R δ(ω − q · V ). If we assume that the perturbation is small enough,
then everything is linear and each q, ω is independent of all the others, and the response
is
δ hρ(q, ω)i = GRρρ (q, ω)ϕ(q, ω).
156
rule (proved on the homework) we learn that at high energy the response function is
fixed Z ∞
ω→∞ 2 N q2
R
Gρρ (q, ω) → 2 ω 0 dω 0 S(q, ω 0 ) = ,
ω 0 mω 2
independent of any dynamics.
A common response to a perturbation localized in time is exponential decay: e−γt , γ >
0. What does this mean for GR (q, ω)? A pole in the LHP at ω = ωR − iγt. Because
of (6.9) our definition of GR actually defines it directly only in the UHP, but the ex-
pression (6.10) allows us to analytically continue to the whole complex plane (minus
the singularities, which determine the spectrum). The locations of the poles in the
integrand of (6.10) determine the response. It has poles at ω = ±ωn − iη. As we’ve
seen, the spectrum often is described by some low-lying single-particle states below a
multi-particle continuum. The continuum happens when the poles bunch up together
and produce a branch cut in GR .
The fact that GR is analytic (no singularities) in the UHP is equivalent to causality:
GR (q, t) = 0 if t < 0 – the response comes after the cause. If one computes GR somehow
and finds a pole in the UHP frequency plane, it means an assumption was violated.
The thing it usually means is that the putative groundstate is unstable, and the pole
in the UHP represents the instability which (initially) grows exponentially in time.
Just as for the response of a linear circuit to an external voltage source, the real
and imaginary parts of GR have very different interpretations. The real part is the
reactive, reversible part, while the imaginary part describes dissipation (like resistors,
which come with a single time derivative, and are therefore π/2 out of phase with the
source). Using the fact that
1 1
lim =P − iπδ(x − a)
η→0 x − a + iη x−a
where P means principal part, (6.11) implies
Z ∞
2ω 0
R 0 0
ReGρρ (q, ω) = dω S(q, ω )P (6.12)
0 ω 2 − (ω 0 )2
ImGR
ρρ (q, ω) = −π (S(q, ω) − S(q, −ω)) . (6.13)
You can see that ReGR is even in ω, while ImGR is odd. The imaginary part of GR
is basically just S(q, ω), but arranged in such a way that −ωImGR (q, ω) > 0. This
positivity is a powerful thing. For one thing, it follows from this that the average
work done on the system is positive – in equilibrium, there is only damping, never
anti-damping.
Because of the analyticity in the UHP, the real and imaginary parts of GR are
related by Kramers-Kronig dispersion relations.
157
Compressibility and the associated sum rule. Consider now ω = 0 and q
small (compared to the inverse separation of particles). This is a perturbation by a
static force field which is smoothly varying in space. The force felt by each particle is
~ which is
−∇ϕ
F(r) = −iqϕ(q, 0)eiq·r + h.c..
Now think about the problem macroscopically. Such a (small) force leads to a change
in the density δρ, which in turn leads to a change in the pressure of the fluid
δρ(r)
δP (r) = . (6.14)
κN
This last expression is the definition of the compressibility κ of the material. In equilib-
rium, the force associated with the induced pressure gradients cancel out the external
force, and therefore
~
0 = −∇δP + NF
from which we conclude
Comparing with our expression for δρ in terms of GR , we learn that in the regime
where this hydrodynamic description is valid
q→0 N
GR 2
ρρ (q, 0) → −N κ = − . (6.15)
mvs2
In the last step we used the fact that the macroscopic description knows about sound
waves30 .
30
Here’s a reminder about how this happens. It involves only two ingredients: (1) the continuity
equation for the number density, which says particles do not disappear
~ · J~ = ρ̇ + ∇
0 = ρ̇ + ∇ ~ · (ρ~u) . (6.16)
∂x p 1
∂t u = − =− ∂x s.
mρ mκN
In the second step, we used the definition (6.14) of the compressibility κ and the chain rule. Taking
158
Combining (6.15) with the relation between GR and S (6.11) we learn the long-
heralded compressibility sum rule
N 1 1
= − lim GR R R
2 ρρ (q, 0) = − lim ReGρρ (q, 0) + iImGρρ (q, 0) (6.18)
2mvs 2 q→0 2 q→0
Z ∞
2ω 0
1 0 0
= − lim dω S(q, ω )P (6.19)
2 q→0 0 0 − ω 02
Z ∞
S(q, ω)
= lim dω (6.20)
q→0 0 ω
(the third step uses (6.12), and (6.13) to see that for static perturbations, ImGR
vanishes).
159
7 Atoms, Molecules, Solids
[Commins, Tong]
Atoms. There are three kinds of atoms.
The key to our understanding of Everything Else is the Central Force Approxima-
tion (CFA). It is just like mean field theory: the idea is to approximate the force on a
given electron (from the nucleus and all of the other electrons) by some average force,
which is also spherically-symmetric about the nucleus. There are several methods to
figure out what we should use for this effective potential, which are the subject of §7.1.
A piece of physics which immediately falls out of this approximation is screening.
The presence of the N − 1 other electrons of charge −e will screen the electric field
of the charge +Ze nucleus. The form of the resulting potential must look like Ze r
for
very small r (when no other electrons are enclosed in the sphere of radius r), and like
(Z−(N −1))e
r
for very large r (when all the others are enclosed). It can be written in the
form
Z(r)e2
Veff = − ,
r
where Z(r) starts at Z at r = 0 and decays to Z − N + 1 at large r. We’ll see that
screening is a key ingredient in the structure of the periodic table.
Hydrogen reminder. Since we’re going to work with spherically-symmetric po-
tentials which often fall off like 1/r, hydrogen wavefunctions will be useful. Recall that
p2 2
the eigenstates of H = 2m − Zer are labelled by three quantum numbers (n, l, m) where
n = 1, 2, 3..., l = 0 . . . n, m = −l, −l + 1 . . . l − 1, l. l, m are orbital angular momentum
quantum numbers. And we use this dumb spectroscopic notation that l = 0 is called s,
l = 1 is called p, l = 2 is called d, l = 3 is called f . The energy spectrum only depends
Z2 2
on n, En = − 2n 2 (in atomic units me = 1 = e ) – it is completely independent of
the angular momentum, so there is a huge degeneracy (broken by small effects such as
relativity and coupling to nuclear spin). The groundstate energy (of the (1,0,0) state)
2
in the right units is − Z2 .
Z > 1 atoms. A good approximation to the hamiltonian of all atoms is
N
X p2i X X −Z 1
H= + Vnucl (ri ) − Vint (rij ), Vnucl (r) = , Vint (r) = .
i=1
2m i i<j
|r| |r|
160
(It ignores relativistic corrections, nuclear spin, spin-orbit couplings, all of which we
understand well, in the case of hydrogen at least, but which give small effects.) There
is no dependence on the electron spin.
Helium ground state. To begin let’s think about helium, Z = 2. The ground-
state, to zeroth order in the interactions, is
s
Z 3 −Zr/a0
Ψ(r1 , r2 ) = ψ100 (r1 )ψ100 (r2 ), with ψ100 (r) = e . (7.1)
πa30
Can I put the two electrons in the same orbital? Yes because they carry spin: I can
do this as long as the spin wavefunction is antisymmetric. So the actual wavefunction
(0)
is what I wrote above times the spin singlet. This has E0 = 2 (−Z 2 /2) |Z=2 = −4 in
atomic units. The experimental result is −2.903 in these units.
We can improve the result by a little perturbation theory in the ee interaction term:
(1) 1 5 5
∆E0 = h100| h100| |100i |100i = Z = .
r12 8 4
I do not want to spend time talking about this integral, but you can do it using the
partial-wave expansion
∞ `
1 1 X r<
= P` (cos θ)
|r12 | r> `=0 r>
where θ is the angle between ~r1 and ~r2 , and r> = max(|r1 |, |r2 |).
We can do even better using the variational method. A good idea is to let Z vary.
This is a way to allow for screening. Instead of (7.1), we’ll try ψ(r) ∝ e−λr/a0 , and
treat λ as a variational parameter. We find (by exactly the same calculation which I
just skipped over)
5
hHiλ = Z 2 − 2Zλ + λ
8
5
which is minimized by λmin = Z − 16 with
2
5
hHiλmin = − Z − = −2.85 (7.2)
16
161
which is not bad at all. This method does even better for larger-Z ions with two
electrons31 . The lesson from this calculation is that the charge of the nucleus is screened
by the other electrons.
Helium excited states. There are also some lessons to be learned from the
excited states of helium. The orbital wavefunction must involve one electron in the
(100) state (call it α) and one in one of the (2lm) states, of which there are 4 (1 state
with l = 0 and 3 with l = 1), which we’ll call β. So, including spin, there are actually
16 degenerate states. Fortunately, they are organized by symmetries – the states with
different l cannot be mixed with each other by the rotation-invariant hamiltonian. And
H is completely independent of the spin.
As in our discussion of the exchange interaction, we can organize these states into
symmetric and antisymmetric orbitals ΨS/A (r1 r2 ) = √12 (ψα (r1 )ψβ (r2 ) ± ψβ (r1 )ψα (r2 )).
To make a fermion state, the spin wavefunction must have the opposite symmetry
property, so the allowed states are: ΨA ⊗ triplet , ΨS ⊗ singlet. Again the hamiltonian
is spin-independent, so just as before the variational energies are just
ES/A = α + β + I ± J
where I used the fact that these wavefunctions are real. I claim that J ≥ 0. Recall from
~ 2 φ = −4πρ is solved by φ(r) = ρ(r0 )d30 r0 . If we let 4πρ(r1 ) = ψα (r1 )ψβ (r1 ),
R
E&M that ∇ |r−r |
then Z Z Z 2
3 ~ 2 IBP ~
J = d rφ(r)4πρ(r) = − φ∇ φ = ∇φ ≥ 0.
The intuition is that the AS wavefunction has lower interaction energy, since ΨA (r1 =
1
r2 ) = 0 so it avoids the region where |r1 −r 2|
is large. So the triplets have lower energy
than the singlets.
31
What about the smaller-Z ion? That is, let’s try to add a second electron to hydrogen, to get H− .
The calculation (7.2) predicts hHi = −.528, which is larger than the groundstate energy of neutral
H (E = -1/2) plus the electron at infinity (E=0), which gives E = −.5. However, there actually is a
(weakly bound) boundstate of H− , with E = −.528. A better variational ansatz, which incorporates
the fact that only the inner electron does screening, is of the form
where r> = max(r1 , r2 ), r< = min(r1 , r2 ), and both s and λ are variational parameters. For Z = 1,
this gives hHi = −.506. This calculation proves the existence of such a boundstate, since it gives an
upper bound on the energy. Thanks to Jiashu Han for asking about this.
162
But what about 2s versus 2p? Well it turns out that (I ± J)2s < (I ± J)2p . So the
2s is lower energy than 2p (just like in the periodic table).
I will not dwell on why this is true because the correct ex-
planation for aufbau is not here. The correct explanation is
that the larger-l hydrogen orbitals have less weight near the
origin. This means that the effective potential they see is
more highly screened, and therefore they are less bound. As
you can see at right, this statement is a little subtle. The top
figure shows the squares of the n = 6 functions. But what
really matters is the measure, which has an extra factor of
r2 ; this is plotted in the bottom figure, and there the maxima
actually move in as l increases. However, it is still true that
the support near the origin decreases as l increases.
This is a subject which seems fancy and which involves lots of names of people, but
the names really each refer to a very simple, sometimes quite dumb, variational ansatz
for the wavefunction.
Hartree. The Hartree ansatz is just a product state: Ψ = ψα1 (r1 ) · · · ψαN (rN ).
This is often a ridiculous thing to do. It ignores Fermi statistics! We could at least
pacify Pauli by taking αi 6= αj . But often it’s not so bad (as we saw above) because
we can tensor it with an appropriate spin wavefunction. Using the CFA, we could
label these orbitals by hydrogen quantum numbers (n, m, l), but I’ll continue to use
the composite index α (which includes spin).
The idea is that we treat the ψs themselves (the whole single-particle wavefunction)
as variational parameters. The variational energy is
N Z
~2 ~ ψα? (r)ψα? j (r0 )ψαj (r0 )ψαi (r)
XZ Z
X Z
hHi = 3
dr 2 2
|∇ψαi | − |ψαi (r)| + d r d3 r0 i
3
0|
.
i=1
2m r i<j
|r − r
To keep the states normalized, the thing we can vary freely should have Lagrange
multipliers:
X Z
2
F [ψ] = hHi − i |ψi | − 1 .
i
163
Its variation is
! δF
0= (7.3)
δψα? i
!
2 2
3 0 |ψαj (r)|
~ ~2 Z XZ
= − ∇ − + dr ψαi (r) − i ψαi (r). (7.4)
2m r j6=i
|r − r0 |
This looks just like the Schrödinger equation, with energy eigenvalue = i , and
Z
Veff (r) = − + Uαi (r)
r
with
2
3 0 |ψαj (r)|
XZ
U αi = dr (7.5)
j6=i
|r − r0 |
representing the electrostatic repulsion of the N − 1 other electrons. Notice that it
depends on the ψαj s! It’s hard to know where to start to solve these horrible coupled
equations. But actually there’s a strategy that works pretty well.
This procedure will actually converge, given a reasonable starting point. The output
is a set of i and a set of optimal occupied single particle states ψαi . The variational
P
estimate for the groundstate energy is not quite just i i , but rather
X X Z Z |ψαi (r)|2 |ψαj (r0 )|2
E0 ≤ hHi = i − .
i j6=i
|r − r0 |
Hartree-Fock. This improvement, due to Slater and Fock actually starts with an
allowed electron wavefunction:
|Ψi = a†α1 · · · a†αN |0i
where α = (n, l, m, σ) is a composite label for all the quantum numbers. That is, the
ansatz for the wavefunction is a Slater determinant
ψα1 (r1 ) · · · ψα1 (rN )
1 .. .. ..
Ψ(r1 σ1 · · · rN σN ) = √ det . .
. .
N!
ψαN (r1 ) · · · ψαN (rN )
164
By just the kind of calculation we’ve been doing all quarter, the energy expectation is
0 0
3 0 ψαi (r)ψαj (r )ψαi (r )ψαj (r)
XZ Z
3
hHi = hHiHartree − dr dr δσi σj .
i<j
|r − r0 |
This extra term Jij ≥ 0 is exactly the exchange integral we encountered above. It
only appears when the electrons involved have the same spin, since otherwise they are
distinguishable. Its physical origin is in the form of the pair correlator for free fermions
(1.85) (again, only of the same spin)– because of Fermi statistics, the electrons avoid
each other. This means that each electron is surrounded by an ‘exchange hole,’ which
lowers its Coulomb energy. By the same argument as above Jij ≥ 0. So this term
lowers the energy when the spins are aligned. In the context of materials, this leads to
(Slater) ferromagnetism.
This rule in the context of chemistry is named after Hund. It makes predictions
for the binding energies as a function of Z. In particular, when a shell is half-filled,
all of the electrons in the valence shell can have the same spin, and benefit from the
exchange energy term. Adding one more electron requires it to have the opposite spin,
and there should therefore be a decrease in the binding energy.
This is borne out by the
plot at right of the first
ionization energies (the
energy cost to remove one
electron) as a function of
Z, which shows e.g. a
step between nitrogen and
oxygen where the 2p or-
bital goes from half-filled
(N is [Be]2p3 ) to one-
more-than-half-filled (O is
[Be]2p4 ). There is a simi-
lar step between phospho-
rus and sulfur during the
filling of the 3p shell.
Incidentally, I made this plot with Mathematica, which knows all of this information
through the command ‘ElementData’.
Again we can treat the ψs as variational parameters. The Hartree-Fock equations
165
(obtained by varying the energy expectation value plus Lagrange multipliers) is
X Z !
! δ
0= hHi − i |ψi |2 − 1 (7.6)
δψα? i i
~2 ~ 2 Z ψα? (r0 )ψαi (r0 )
X Z
= − ∇ − + Uαi (r) ψαi (r) − δσi σj d3 r0 j 0|
ψαj (r) − i ψαi (r).
2m r j
|r − r
(7.7)
The extra term relative to the Hartree equations (7.4) (associated with the exchange
term in the energy) is like a nonlocal potential. It represents an attractive charge
density of electrons with the same spin.
To emphasize the unity of methods between atomic physics and all of our previous
discussion of many-body systems, let’s re-derive the Hartree-Fock(-Slater) equations
(7.7) using an auxiliary (quadratic) mean-field Hamiltonian, as explained more sys-
tematically in §5.6. The actual hamiltonian is
!
X ~2
∇ X
H= †
ar − + Vext (r) ar + V (r, r0 )a†r a†r0 ar0 ar ≡ T + Vext + V,
r
2m r<r0
where here Vext is the potential from the nucleus, and V (r, r0 ) is the electron-electron
interaction. We want to find the gaussian state α d†α 0̃ with the lowest expected
Q
This contains an in-general non-local hopping term Γ, which we will see is required by
the exchange term. The potentails Ur , Γrr0 are to be determined.
All expectation values below are evaluated in the groundstate of HMF , so we can
use Wick’s theorem:
X D E D ED E
0 † † † † †
hHi = hTi + hVext i + V (rr ) ar ar ar0 ar0 − ar ar0 ar0 ar
r<r0
166
We want this to vanish in order to give the best trial wavefunction. On the other and,
since |MFi is its groundstate, the variation of hHMF i is zero:
X X D E
0 = δ hHMF i = δ hTi + δ hVext i + U (r)δ a†r ar + Γrr0 δ a†r ar0 .
r rr0
Therefore
X D E
U (r) = V rr0 a†r0 ar0 (7.8)
r0
D E
Γrr0 = Vrr0 a†r0 ar . (7.9)
The HFS equations (7.7) are just the eigenvalue equations for HMF with the potentials
determined by these self-consistency conditions.
p3F
Z
T =0
n=2 d̄3 pf (p) =
3π 2
and the kinetic energy density is
p2 5/3 4/3 2
Z
T =0 3 π ~ 5/3
u = 2 d̄3 p f (p) = n .
2m π me
167
µ + V (r). So the idea is just to replace µ → µ + V (r) in the expression for the phase-
space density (7.10). It is an assumption of local equilibrium, generally a synonym for
hydrodynamics. Local equilibrium means that nearby patches of fluid are in equilib-
rium, including chemical equilibrium. But chemical equilibrium means the µ is the
same. Therefore, µ is constant, and in our replacement
p2F p2F (r)
µ = F = → µ≡ − eΦ(r)
2m 2m
we see that we must let pF = pF (r) the Fermi momentum vary with position. This
gives (assuming spherical symmetry of the applied potential and everything else)
Z pF (r)=2m√eφ+µ
(2m)3/2
n(r) = 2p2 dp4π = 2
(eΦ(r) + µ)3/2 .
0 3π
So this is a relation between the density and the electrostatic potential. There is
a second relation, namely Gauss’ law. For example, in the case of the electron cloud
around a nucleus
~ 2 Φ = en(r) − Zeδ 3 (r).
∇
Defining eΦ0 ≡ eΦ + µ, we have
~ 2 Φ0 = eαΦ3/2
∇ 0
(2m)3/2
(here α ≡ 3π 2
) with the boundary conditions that near the nucleus, the only charge
r→0
Ze
is the nuclear charge: Φ0 (r) → 4πr . (The boundary condition at r → ∞ specifies the
number of electrons.) By non-dimensionalizing this equation, we can learn that the
sizes of atoms go like Z −1/3 . To see how, look at Commins §12.4.
To slick up our understanding of the TF approximation, note that we can get the
above equations by minimizing the functional
Z
3
F [n] = E0 [n] − µ d rn(r) − N .
Its variation is
δF [n] 5
0= = cn2/3 (r) − µ − eΦ(r), with ∇2 Φ(r) = en(r) − Zeδ 3 (r).
δn(r) 3
168
Thomas-Fermi screening. As a tiny hint of the applications of this perspective,
let us see what it says about screening. Let us imagine applying an infinitesimal extra
electrostatic potential δφ to our system. It satisfies Gauss’ law:
On the RHS here is some recognition that the charges will move around in response to
the applied field. (And I am writing an equation for just the additional potential, so
we do not include on the RHS the equilibrium distribution, or e.g. the nuclear charge,
only the induced charge.) If δφ is small, we may use linear response:
where χ is the compressibility (defined by this relation). Now use the TF approximation
(with α as above):
e−qTF r
δφ(r) ∼
r
2
with qTF = 4πe2 |χ0 |. It says that as a result of screening, in this approximation, the
Coulomb law is replaced by an exponentially decaying potential, and hence force. The
correct answer is not quite so extreme and more interesting. In particular, in a metal,
it behaves like cosr2k3 F r .
Warnings. These methods are widely used and very successful. The Hartree-Fock
approximation makes precise what people mean when they speak of filling hydrogen
orbitals in an interacting system. However, the true groundstate need not be a single
Slater determinant (the fermion analog of a product state): it will almost certainly be a
complicated superposition of such states, meaning that the electrons will be entangled.
But, approximating the effect of all the other electrons as a central potential is clearly
a truncation. Also, we are ignoring spin-orbit coupling, which grows like Z 2 along
columns of the periodic table.
169
7.2 Problems involving more than one nucleus
So far in this chapter we’ve been discussing ways to deal with the interactions between
two or more electrons orbiting a single nucleus of charge Z. We’ve mostly focussed on
neutral atoms, but the same methods apply if the number of electrons is not equal to
Z. The periodic table is an important part of chemistry. Another important part is
chemical bonding, whereby one atom sticks to another. We could understand something
about this, too, if we had time.
Born-Oppenheimer approximation. Nuclei are much (at least 2000 times)
heavier than electrons. This fact is what allows us to make progress on problems with
more than one nucleus, such as molecules (a few nuclei) or solids (many many nuclei).
The idea is: first treat the positions R of the nuclei as fixed. Solve for the groundstate
energy of the electron system in the potential Vext (r) created by the nuclei, E0 (R).
Now minimize this over R to find where the nuclei want to be.
We can check a posteriori that this was a good approximation using dimensional
analysis and the uncertainty principle. The electronic coordinates satisfy ∆x∆p ∼ 1;
an atom has size ∆x ∼ a0 , so ∆p ∼ 1/a0 , and the electronic energies go like Ee ∼
∆p2 ~2
2me
= 2ma 2 = 1 (in atomic units, naturally). How much energy does it cost to excite
0
the nuclei? Near the minimum of E0 (R) it is a harmonic oscillator, so Evib ∼ M ω 2 x2 .
Dissociation happens when x ∼ 1 and costs energy ∼ Ee ∼ 1 ∼ M ω 2 . Therefore the
splittings between (SHO) vibrational levels go like ∆Evib = ~ω ∼ m
p e
M
. What about
2
`(`+1)
rotational motion? Erot = ~ M R02
where R0 is the optimal separation between nuclei,
also of order R0 ∼ a0 = 1. So the rotational splittings have ∆Erot ∼ mM
e
, which is of
−3
order 10 or smaller.
So, given Vext (r), we want to find the groundstate of
Z
H = T + V + dd rVext (r)n(r)
P p2 P
where T = i 2mi and V = i<j Vint (rij ). Notice that the first two terms are the same
for all of these problems, wherever the nuclei are. And the input external potential (by
which all these problems differ) only enters multiplying the density n(r). We usually
think of doing this by finding the groundstate wavefunction Φ0 , and then determining
the density through X
†
n(r) = nσ (r) = hΦ0 | ψrσ ψrσ |Φ0 i . (7.15)
σ
Weirdly, it is also true that the density determines the groundstate wavefunction, and
hence the groundstate energy. That is, different Vext necessarily give rise to different
groundstate electron densities n(r). The proof (due to Kohn) is sweet: Suppose oth-
0
erwise, i.e. suppose Vext and Vext (with groundstates Φ0 , Φ00 ) lead to the same n(r).
170
Assuming the groundstate is nondegenerate, the variational theorem says
Z
E0 = hΦ0 | H |Φ0 i < hΦ0 | H |Φ0 i = hΦ0 | H |Φ0 i+hΦ0 | (V − V ) |Φ0 i = E0 + n0 (Vext − Vext
0 0 0 0 0 0 0 0 0 0
).
But the same is true with the primes and unprimes switched:
Z
0 0
E0 < E0 + n (Vext − Vext ) .
which forbids n = n0 .
This means that the groundstate energy E0 and the many-body wavefunction
Φ0 (r1 · · · rN ) are purely functionals of n(r), a single function of d variables. Appre-
ciate for a moment how crazy thispis. (A case where it is manifestly true is when there
is only one particle: then ψ(r) = n(r), and we lose no information since ψ(r) has no
nodes.)
Moreover, the n which minimizes the energy expectation is the correct one. Here’s
why: suppose n(r) arises from some state Ψ[n] by (7.15). Then
Z Z
E0 [n] = hΨ[n]| (T + V) |Ψ[n]i + Vext n ≡ Eint [n] + Vext n. (7.16)
where n0 is the density in the true groundstate. Notice that the minimization of (7.16)
then describes a Legendre transform of a fixed, universal functional Eint [n], which is
the same form any atom or molecule or solid (the information about which only enters
through Vext ). It is like a philosopher’s stone. This realization leads to the develop-
ment called Density Functional Theory, which proceeds further by approximating the
functional in various ways. The simplest is the Thomas-Fermi approximation, where
2 R R n(r)n(r 0 )
Eint [n] = c n5/3 (r)d3 r + e2
R
|r−r0 |
as in (7.11). The next simplest possibility is
called the local density approximation, developed by Kohn and Lu Sham, with spec-
tacular success in determining electronic structure for many situations32 .
32
For more, try Chapter 15 of the book by Girvin and Yang, Modern Condensed Matter Physics.
I found this history of Kohn’s path towards the discovery of DFT quite compelling. This paper by
Banks also provides a well-encapsulated introduction.
171