Advanced QM Lecture Notes

2
Chapter 0
Introduction
This course will be concerned with quantum theory, the fundamental theory of nature.
Properties of condensed matter systems, of atoms, molecules and basically the world
surrounding us cannot be understood without resorting to quantum mechanics. The ad-
vent of quantum mechanics constitutes the most radical break from classical mechanics
as it was developed until the end of the 19th century. The fact that quantum mechanics
is an intrinsically statistical theory that does not allow for an interpretation in terms of
classical probability theory is one one of the features of quantum theory in which it is
departing from classical physics.
Building upon what we have learned in the rst course on quantum mechanics,
we will now be concerned with advanced topics. In this endeavour, we will spend a
lot of time with dealing with the situation in which many particles of the same kind
come together. This is for a good reason: After all, this is the situation commonly
encountered when thinking of actual materials, of solid bodies of systems are they
are commonly studied in statistical mechanics. We will encounter intriguing quantum
phenomena of conductors having no resistance, uids with no viscosity. We will have
a look at properties of interacting quantum systems that cannot be explained by just
looking at its parts. There, we will investigate some truly advanced topics, related to
tensor network descriptions of quantum many-body systems. Before coming to that,
however, we will quickly resume the basics of quantum mechanics as such. This will
be brief, but do not worry too much: It will all be content that has been treated in great
detail in the rst course anyway.
How to use these lecture notes: It is the point of these lecture notes to summarize
the content of the course. Important denitions or results will be highlighted using
boxes that look like this.
These lecture notes will be dynamically evolving along with the course and are not
set in stone. This comes along with the advantage that not even the course content is
fully set in stone, there is some exibility here. The course is not based on any specic
book. Having said that, here and there I am shamelessly copying content from other
sources, if I think this to be appropriate (but of course, credit is given then). The lecture
3
4 CHAPTER 0. INTRODUCTION
notes are strictly meant to be of good use concomitant with the lecture. They are no
draft for a book in the making, they are raw and incomplete. Still, they should fulll
a good purpose, I would think. Having said that, it will be important to read more
than one book along with the course. The subsequent list provides some guidance in
this respect. They are of a very different kind: While the book by Galindo/Pascual
is mathematically much more pedantic and precise, for better or worse, the one by
Schwabl is much more pragmatically minded.
Bibliography
[1] A. Galindo and P. Pascual, Quantum mechanics II (Springer, 1989).
[2] A. L. Fetter und J. D. Valecka, Quantum theory of many-particle systems

(McGraw-Hill, 1971).
[3] E. K. U. Gross und E. Runge, Vielteilchentheorie (Teubner, 1986).
[4] W. Nolting, Grundkurs theoretische Physik 7 (Springer, 2007).
[5] F. Schwabl, Fortgeschrittene Quantenmechanik (Springer, 2005).
5
Chapter 1
Elements of quantum theory
Every physical theory is supposed to make predictions on future measurement out-

comes when performing experiments with a well-dened physical system that is ini-
tially prepared in the same way. The predictions of quantum mechanics are of a sta-
tistical nature: The theory is utterly silent about specic measurement outcomes. It
will rather provide probabilities for obtaining certain outcomes. Conversely, to obtain
evidence into the correctness of a prediction, one needs to perform many experiments
under identical conditions. Then, by investigating relative frequencies of measurement
outcomes, one can estimate probabilities. At the heart of the formalism are notions of
expectation values. This is no shortcoming of the theory: This intrinsic randomness
is actually a deep structure element of quantum mechanics that is there to stay: Bell’s
theorem shows that there cannot be an underlying classical statistical picture that can
be held responsible to explain the randomness of quantum mechanics.
That is to say, we have to ask ourselves how to capture states – the collection of
information summarizing all information required to make future predictions – how
observables – the quantities that can be measured. Also, we will think about how
systems evolve in time.
1.1 Quantum states and observables
1.1.1 Pure quantum states
We start by considering quantum systems that are not composite, but that consist of
simple physical systems. Quantum systems are associated with a Hilbert space. A
complex Hilbert space is a vector space equipped with a scalar product that is complete
with respect to the norm induced by the its scalar product. The simplest conceivable
Hilbert space is that of a single spin. It is spanned by the two vectors |0〉 and |1〉 or | ↑〉
and | ↓〉: That is to say, the spin points up or down.
3
4 CHAPTER 1. ELEMENTS OF QUANTUM THEORY
State vectors: Pure quantum states are described by normalized state vectors |ψ〉 ∈
H from a complex Hilbert space.
This could be a superposition
|ψ〉 = α|0〉 + β|1〉 (1.1)
with complex α, β, normalized as
|α|2 + |β|2 = 1. (1.2)
This example already shows that spins cannot only be pointing up or down in quantum
mechanics. They can be in an arbitrary superposition of pointing up or down. The
associated Hilbert space is simply H = C2 .
The scalar product between two state vectors is written as 〈ψ|φ〉. Normalization
means that the standard vector norm takes a unit value, which in turn is equivalent with
〈ψ|ψ〉 = 1. (1.3)
The vector 〈ψ| ∈ H∗ is a dual vector. Jokingly referring to the term bracket, one also
calls dual vectors “bras” and vectors “kets”. Matrix elements of operators A take the
form 〈ψ|A|φ〉.
For every Hilbert space of a d-dimensional quantum system, so a system with d
“levels”, one can pick a basis
B = {|0〉, . . . , |d − 1〉}. (1.4)
In this basis, every state vector can be expressed as

d−1
∑
|ψ〉 = cj |j〉. (1.5)
j=0
The complex numbers c0 , . . . , cd−1 are called coefcients. The basis is normalized and
complete, which means that
〈j|k〉 = δj,k , (1.6)

d−1
∑
|j〉〈j| = . (1.7)
j=0
All this applies to so-called nite-dimensional quantum systems, where d is an integer.

The Hilbert space of a particle in the position representation is H = L2 (R): Pure states
of particles without spin in one spatial dimension are captured by by wave function.
State vectors |ψ〉 hence then belong to complex wave functions ψ that are normalized
as ∫
dxψ ∗ (x)ψ(x) = 1. (1.8)
1.1. QUANTUM STATES AND OBSERVABLES 5
Their scalar product is then

∫
dxφ∗ (x)ψ(x) = 〈φ|ψ〉. (1.9)
A good chunk of most elementary quantum mechanics lectures is concerned with the
dynamics of such wave functions, say, the scattering induced by a box potential. As
we know,
p(x) = |ψ(x)|2 (1.10)
is the probability density of nding a particle at the position x in a position measure-
ment. Here, this refers to a single spatial dimension, but we will come to more than
one spatial dimension in a second. In innite-dimensional Hilbert spaces, such as the
one reecting our situation, it is sufcient to have a separable Hilbert space. For the
situation at hand, the eigenfunctions of the harmonic oscillator
B = {|0〉, |1〉, . . . } (1.11)
provide such a countable (but innite) basis.
1.1.2 Observables
Quantities that can be measured as called, unsurprisingly, observables. The are associ-
ated with Hermitian operators A, meaning that
A = A† . (1.12)
Their eigenvalues (or rather spectral values, but let us not be too mathematically pedan-
tic at this point) are possible outcomes of (idealized projective) measurements. The fact
that observables are Hermitian implies the property that their eigenvalues (or spectral
values) are real, which is a nice feature if one wants to interpret them as measurement
outcomes.
Observables: Observable are Hermitian operations in a Hilbert space. Expectation

values of such observables for systems prepared in pure states are given as
〈A〉 = 〈ψ|A|ψ〉. (1.13)
As mentioned before, such expectation values make predictions about relative fre-
quencies in experiments. Observables can always be diagonalized. After all, unitary
operators are precisely those that map one orthonormal basis onto another one.
Diagonalization: Every Hermitian operator can be diagonalized in that there exists

a unitary operator U (satisfying U U † = U † U = ) and a diagonal matrix D so that
A = U DU † . (1.14)
That is to say, when expressed in the appropriate basis, every Hermitian operator
takes a diagonal form in the matrix representation. General operators in Hilbert spaces
can, needless to say, be expressed in this basis, as

d−1
∑
A= 〈j|A|k〉|j〉〈k|. (1.15)
j,k=0
Their trace is given by

d−1
∑
tr[A] = 〈j|A|j〉. (1.16)
j=0
The trace is independent of the choice of the basis, as a moment of thought reveals.
Operators in innite-dimensional Hilbert spaces do not need to have a trace. One can
still dene trace-class operators for which the sequence of traces of nets converges.
In what follows, we will not distinguish between operators and their matrix repre-
sentation, as is common in the literature. A kind of observable that takes a key role for
good reasons are the Pauli operators. They are dened as
[ ]
0 1
X = |0〉〈1| + |1〉〈0| = , (1.17)
1 0
[ ]
0 −i
Y = −i|0〉〈1| + i|1〉〈0| = , (1.18)
i 0
[ ]
1 0
Z = |0〉〈0| − |1〉〈1| = . (1.19)
0 −1
The unit operator
[ ]
1 0
= |0〉〈0| + |1〉〈1| = , (1.20)
0 1
is commonly included as a Pauli matrix. It is obvious how to compute their expectation
values. The expectation value of Z of a system prepared in |ψ〉 = α|0〉+β|1〉 is simply
〈ψ|Z|ψ〉 = |α|2 − |β|2 . (1.21)
Stern-Gerlach type experiments can be described like that.
1.1.3 Mixed quantum states

Let us imagine we have a single spin, associated with a Hilbert space H ' C2 . We now
throw a coin. In case of heads, we prepare the spin in |0〉, in case of tails, we prepare it
in |1〉. That is to say, with the classical probability 1/2 we have |0〉, and with classical
probability 1/2 we get |1〉. How do we capture this situation? Can we describe the
system by a state vector √
|+〉 = (|0〉 + |1〉)/ 2? (1.22)
Not quite. This is easy to see: In case of a σx measurement, we would always get the
same outcome. But this is different from the situation we encounter here. In fact, when
we make a measurement of σx , we would get both outcomes with equal probability. Or
√
|−〉 = (|0〉 − |1〉)/ 2? (1.23)
Again, this will not work, for the same reason. In fact, no state vector is associated
with such a situation, and for that, we need to generalize our concept of a quantum
state slightly: to density operators. This is, however, the most general quantum state in
standard quantum mechanics, and we will not have to generalize it any further.
In fact, the above situation is an instance of the situation where we prepare with
probability pj , j = 1, . . . , n, a system in a state vector |ψi 〉. Since we encounter a
probability distribution, we have
n
∑
pj = 1. (1.24)
j=1
Such a situation is sometimes referred to as a mixed ensemble. How do we incorporate

that? To start off from what we know, let us express rst pure states corresponding to
state vectors as a density operator.
Density operator of a pure state: A pure state associated with a state vector |ψ〉 ∈
H from some Hilbert space H is given by the density operator
ρ = |ψ〉〈ψ|. (1.25)
We immediately nd some properties of such an operator: We obviously have that
ρ = ρ† . (1.26)
Then,
tr(ρ) = 1. (1.27)
Finally, we have that
ρ ≥ 0, (1.28)
which means that all of its eigenvalues are non-negative, which is clearly the case, as
all the eigenvalues are given by 0 or 1, clearly non-negative numbers. We also have the
property that
tr(ρ2 ) = tr(|ψ〉〈ψ|ψ〉〈ψ|) = tr(|ψ〉〈ψ|) = tr(ρ) = 1. (1.29)
How to we compute expectation values from such a density operator? Well, we know
that for an observable
〈A〉 = 〈ψ|A|ψ〉. (1.30)
This we can equally well write as
〈A〉 = 〈ψ|A|ψ〉 = tr(A|ψ〉〈ψ|) = tr(Aρ). (1.31)
We have hence made use of one of the above rules, and have written expectation values
as a trace of the observable, multiplied with the density operator. A general density
operator is just extended by linearity from this denition.
Density operator of a mixed ensemble: Consider the situation of preparing |ψj 〉,

j = 1, . . . , n with probability pj . This is associated with a density operator
n
∑
ρ= pj |ψj 〉〈ψj |. (1.32)
j=1
Then, how do we compute expectation values from that? We have for observables
A that
∑n
〈A〉 = pj 〈ψj |A|ψj 〉, (1.33)
j=1
from the very denition of a mixed ensemble. This we can, however, also write as
n
∑
〈A〉 = pj tr(A|ψj 〉〈ψj |)
j=1
 
n
∑
= tr A pj |ψj 〉〈ψj |
j=1
= tr(Aρ). (1.34)
So again, expectation values are just computable as the trace of the density operator
multiplied with the observable. We now once more investigate properties of such a
density operator: We nd that again,
ρ = ρ† . (1.35)
In the same fashion as before, since now
n
∑
pj = 1, (1.36)
j=1
we also have that

tr(ρ) = 1. (1.37)
Finally, we have that
ρ ≥ 0, (1.38)
since all of the probabilities are positive, and a sum of positive operators is positive.
These are exactly the same properties as above, except from one that is now missing.
We no longer have tr(ρ2 ) = 1. In fact, this property is replaced by
 (
n n
)
∑ ∑
2
tr(ρ ) = tr  pj |ψj 〉〈ψj |  pk |ψk 〉〈ψk |
j=1 k=1
n
∑
= pj pk tr(|ψj 〉〈ψj |ψk 〉〈ψk |) ≤ 1
j,k=1
where we have bounded the scalar products between two arbitrary state vectors. We
have now arrived at the most general concept of a state in (standard) quantum mechan-
ics. This is surely worth a box:
Density operators: General states of quantum systems with Hilbert space H are
given by density operators ρ. Their properties are
ρ = ρ† (Hermicity), (1.39)
ρ ≥ 0 (Positivity), (1.40)
tr(ρ) = 1 (Normalization). (1.41)
Pure states are those density operators for which
tr(ρ) = 1, (1.42)
those can be represented by state vectors |ψ〉 ∈ H as
ρ = |ψ〉〈ψ|. (1.43)
Otherwise, if tr(ρ2 ) < 1, the state is called mixed. For observables, expectation
values are computed as 〈A〉 = tr(Aρ).
This is a good moment to discuss a number of examples. Let us go back to our

initial situation discussed at the beginning of the chapter, of preparing |0〉 or |1〉 with
equal probability. We can now easily associate this with a density operator
1 1
ρ= |0〉〈0| + |1〉〈1|. (1.44)
2 2
We can write this in matrix form – remember that operators and their matrix represen-
tation are identied with each other throughout the script
[ 1 ]
0
ρ= 2 1 . (1.45)
0 2
We have that
1 1 1
tr(ρ2 ) = + = < 1. (1.46)
4 4 2
2 2
This in fact the minimum value tr(ρ ) can take for a system with H ' C . The pure
state ρ = |0〉〈0| in turn is represented as
[ ]
1 0
ρ= , (1.47)
0 0
obviously satisfying tr(ρ) = 1. Generally, if we have probabilities p0 and p1 to prepare

|0〉〉 and |1〉, we have the density operator
[ ]
p0 0
ρ= . (1.48)
0 p1
But of course, we are not forced to take the standard basis. The situation of having
prepared |+〉 and |−〉 with equal probabilities is captured as
1 1
ρ= |+〉〈+| + |−〉〈−|. (1.49)
2 2
This is
1 1
ρ = ((|0〉 + |1〉)(〈0| + 〈1|)) + ((|0〉 − |1〉)(〈0| − 〈1|))
4 4
1 1
= |0〉〈0| + |1〉〈1|, (1.50)
2 2
with matrix representation

1
[ ]
0
ρ= 2
1 . (1.51)
0 2
There is a cute observation that is important at this point: There are many different ways
of preparing the same density operator! Since all expectation values of observables are
computed as
〈A〉 = tr(Aρ), (1.52)
we get exactly same same value for all observables in case of
n
∑ m
∑
ρ= pj |ψj 〉〈ψj | = qk |φk 〉〈φk |, (1.53)
j=1 k=1
even if all of the probabilities {pj } and {qk } as well as all state vectors {|ψj 〉} and
{|φk 〉} are different. In fact, now even n = m has to hold. What matters for all
outcomes in all experiments is the density operator, not the mixed ensemble we have
started with. Sometimes, people use notions of the kind, “the system is in some pure
state vector |ψj 〉, j = 1, . . . , n, we simply do not know which one”. Such reasoning
is not quite precise and can be plain wrong, in which case it is referred to as preferred
ensemble fallacy.
1.2 Measurement postulate

The measurement postulate has to give an answer to the following questions: What are
the outcomes of a measurement? What is the probability of obtaining this? What is
the state immediately after the measurement? The measurement postulate settles these
questions.
1.3. COMPOSITE QUANTUM SYSTEMS 11
Measurement postulate: Let A be an observable with spectral decomposition

∑
A= λ k Pk , (1.54)
k
where ∑
Pk = |ψk 〉〈ψk |. (1.55)
EVλk
That is, the Pk = Pk2 are the projections onto the eigenspaces to eigenvalue λk . The
possible measurement outcomes are λj . The probability of obtaining for obtaining
the outcome related to λk is
pk = tr[ρPk ]. (1.56)
The state immediately after the measurement is
Pk ρPk Pk ρPk
ρ′k = = . (1.57)
tr[Pk ρPk ] tr[ρPk ]
1.3 Composite quantum systems

How do we describe composite quantum systems in quantum theory? Clearly, the
formalism must have an answer to that. We think of a particle having several degrees
of freedom. Or we aim at describing several different particles at once. How do we
capture this situation? Composition of degrees of freedom is incorporated by the tensor
products in quantum mechanics. Let us assume that we have one degree of freedom
associated with a d1 -dimensional Hilbert space
H1 = span{|0〉, . . . , |d1 − 1〉}. (1.58)
We then consider another, second degree of freedom, coming along with a d2 -dimensional
Hilbert space
H2 = span{|0〉, . . . , |d2 − 1〉}. (1.59)
These spaces could, for example, capture all superpositions of two spin degrees of
freedom of two particles described by quantum mechanics. The Hilbert space of the
joint system is then given by the tensor product
H = H 1 ⊗ H2 . (1.60)
It is spanned by the orthonormal basis vectors
{|j〉 ⊗ |k〉 : j = 0, . . . , d1 − 1; k = 0, . . . , d2 − 1} . (1.61)
Such basis elements of tensor products are sometimes also written as
{|j, k〉 : j = 0, . . . , d1 − 1; k = 0, . . . , d2 − 1} . (1.62)
This looks more complicated than it is: While an arbitrary superposition of a state
vector from H1 can be written as
d∑
1 −1
|ψ1 〉 = αj |j〉 (1.63)

j=0
and an arbitrary superposition of a state vector from H2 is

d∑
2 −1
|ψ2 〉 = βj |j〉, (1.64)

j=0
an arbitrary state vector taken from the composite Hilbert space H = H1 ⊗ H2 is given
by
d∑1 −1 d∑
2 −1
|ψ〉 = γj,k |j〉 ⊗ |k〉, (1.65)

j=0 k=0
as a linear combination of all new basis vectors, with all γj,k ∈ C. If you think at this
point that it may be confusing that such general state vectors contain ones that are no
longer a product between the respective Hilbert spaces: Indeed, it is, and we will come
to the profound implications of this later. Again:
Composite quantum systems: The Hilbert space of the composite quantum sys-
tems the parts being associated with Hilbert spaces H1 and H2 is given by the
tensor product
H = H1 ⊗ H 2 . (1.66)
Similarly, an arbitrary linear operator can be decomposed as

∑
O= cj,k Aj ⊗ Bk , (1.67)
j,k
with operators {Aj } and {Bk } on H1 and B1 , respectively. The same is true, needless
to say, for more elaborate composite quantum systems having many parts. For example,
the Hilbert space of n particles of the same character (and associated with H each) is
given by
H ⊗ · · · ⊗ H = H⊗n . (1.68)
This is not such an alien situation in quantum physics: We should not forget that most
things we can see are quantum systems with more than 1023 individual parts. The parts
of composite quantum systems also do not have to be the same. In fact, on the rst
page of this script we have already encountered a composite quantum system, even if
at this point we had abstracted from this fact. The particle has a position and a spin
degree of freedom. So the joint Hilbert space capturing this situation is given by
H = L2 (R) ⊗ C2 . (1.69)
The Hilbert space of the three coordinates of the spatial degree of freedom of a single
particle is given by
H = L2 (R) ⊗ L2 (R) ⊗ L2 (R). (1.70)
1.4. TIME EVOLUTION 13
1.4 Time evolution
When we prepare a quantum system in some state ρ, how does it evolve in time? The
answer to this question is given by the Schrödinger equation. It is given in the form of
a differential equation.
Schrödinger equation: State vectors of pure states evolve in time according to
∂
i~ |ψ(t)〉 = H|ψ(t)〉, (1.71)
∂t
where H is the Hamiltonian. General quantum states evolve as
∂
i~ ρ(t) = [H, ρ(t)]. (1.72)
∂t
Making use of the unitary time evolution operator
U (t) = e−iHt/~ , (1.73)
valid for time-independent Hamiltonians, we can capture time evolution also as fol-
lows. Since it is unitary, it satises
U (t)U † (t) = U † (t)U (t) = . (1.74)
Obviously, U (0) = .
Time evolution: The time evolution of a closed quantum systems from time t1 to
t2 > t1 is captured by a time evolution of state vectors as
|ψ(t2 )〉 = U (t2 − t1 )|ψ(t1 )〉. (1.75)
General states, so density operators, evolve according to
ρ(t2 ) = U (t2 − t1 )ρ(t1 )U † (t2 − t1 ). (1.76)
This type of time evolution is referred to as time evolution in the Schrödinger pic-
ture, in which observables are kept constant and quantum states evolve. It can also
make sense to refer to a picture in which states following preparations are kept con-
stant and observables evolve. This picture is referred to as Heisenberg picture.
Evolution of observables in the Heisenberg picture: In the Heisenberg picture,

observable A evolve from time t1 to t2 > t1 as
A(t2 − t1 ) = U † (t2 − t1 )AU (t2 − t1 ). (1.77)
Of course, the predictions in both pictures are identical,

〈A(t)〉 = tr[Aρ(t)] = tr[A(U (t)ρU † (t))] = tr[(U † (t)AU (t))ρ]
= tr[A(t)ρ]. (1.78)
with t2 = t und t1 = 0 to simplify the notation.
1.5 Heisenberg’s uncertainty principle

This notion of outcomes of measurements being “uncertain” can be made more precise
in terms of the so-called uncertainty relation. There are few statements in quantum the-
ory, yet, that are so often misunderstood as the uncertainty relation. Recently, a lawyer
came to me and asked, well, is the uncertainty in adjudication not just a manifestation
of the Heisenberg principle, that “nothing is precisely dened”? Well, actually, no.
But except from the principle being overly referred to in urban slang, even within the
formalism there often is a kind of confusion coming along with this principle. I was
slightly confused when I rst heard of it in my rst lecture, and slightly irritated by
the fact that nobody seemed to have any desire to x the precise pre-factor on the right
hand side of the inequality. Later I understood that this is because there have been dif-
ferent readings used of the principle. We will mention three readings of Heisenberg’s
uncertainty principle, even if we discuss in detail only one.
The central object here is the mean square deviation of an observable from its ex-
pectation value, for a system initially prepared in |ψ〉,
(∆A)2 = 〈(A − 〈A〉)2 〉 = 〈A2 〉 − 〈A〉2 . (1.79)
We now consider this for two observables A and B. For simplicity of notation, let us
dene
C = A − 〈A〉, (1.80)
D = B − 〈B〉. (1.81)
Even if C and D are clearly Hermitian, the product CD is usually no longer Hermitian,
and the expectation value can be complex. Let us write
〈CD〉 = 〈ψ|CD|ψ〉 = z = x + iy, (1.82)
〈DC〉 = 〈ψ|(CD)† |ψ〉 = z ∗ = x − iy, (1.83)
with x, y ∈ R. That is to say,
〈[C, D]〉 = 〈ψ|[C, D]|ψ〉 = z − z ∗ = 2iy = 2i im(〈CD〉). (1.84)
1.5. HEISENBERG’S UNCERTAINTY PRINCIPLE 15
Taking the absolute value, we obtain
|〈[C, D]〉|2 ≤ 4|〈ψ|CD|ψ〉|2 . (1.85)
Of course, we have for the commutator
[C, D] = [A, B], (1.86)
as numbers always commute with each other. We now make use of the Cauchy-
Schwarz inequality – a useful inequality from linear algebra. This leads to
|〈ψ|CD|ψ〉|2 = |(〈ψ|C) · (D|ψ〉)|2

≤ ‖C|ψ〉‖2 · ‖D|ψ〉‖2
= 〈ψ|C 2 |ψ〉 · 〈ψ|D2 |ψ〉. (1.87)
We can now collect the terms and state the uncertainty relation:
Heisenberg’s uncertainty relation: Two observables A and B of a quantum sys-

tem prepared in |ψ〉 satisfy the uncertainty relation
1
∆A · ∆B ≥ |〈ψ|[A, B]|ψ〉|. (1.88)
2
That is to say, the product of the two uncertainties cannot take an arbitrarily small
value – unless the two observables A and B commute. Otherwise, the product of
the two uncertainties of A and B cannot be arbitrarily small. This is a remarkable
observation.
When applied to position and momentum, this principle reads
~
∆X · ∆P ≥ , (1.89)
2
which means that wave functions cannot be arbitrarily narrow both in position and mo-
mentum. This means that if we prepare many quantum systems and rst estimate ∆A
many times, such that we know it later with high statistical signicance, and then es-
timate ∆B: Then the product of the two uncertainties will always be larger than the
above given value. Quantum mechanics simply does not allow for any smaller uncer-
tainties. So in a way, one cannot “know the values of two non-commuting observables
at once”. Precisely, what is meant, however, is that one independently prepares the sys-
tems and measures either A or B, and then analyses the data. This is the rst reading
of the Heisenberg uncertainty relation.
This is also the derivation that most books on quantum mechanics offer. Interest-
ingly, the explanation given is quite often incompatible with the above derivation. This
is the notion of a measurement of one observable A makes the outcome of another
non-commuting observable B less certain. This is the second reading of the uncer-
tainty relation. We hence measure on the same system rst A, then B. If we “know the
value of A precisely, then the measurement of B will be a lot disturbed”. This is the
reading of the famous Heisenberg microscope: Heisenberg discussed how a measure-

ment of X disturbs later later measurements of the non-commuting observable P . This
is also true, but this is not quite what we have derived above (which made use of the
independent and identical distribution (i.i.d.) of the initial preparation). One can also
derive uncertainties for the Heisenberg uncertainty principle in this second reading, but
the pre-factor on the right hand side will be slightly different. A third reading which we
only briey mention is the one where tries to jointly measure two observables in a sin-
gle generalized measurement. We have to delay this discussion as we are at this point
not quite clear about what a generalized measurement is. Just for the records: Here one
aims at obtaining as much information as possible about A and B in a single run of a
measurement, which is then repeated many times. This again leads to an uncertainty
relation, but yet again with a different pre-factor. All these readings have the narrative
in common, however, that if the mean square deviation of one observable is small, it
cannot be small at the same time for a non-commuting observable. This is again one
of the profound consequences of non-commuting operators in quantum theory.
Chapter 2
Identical particles
2.1 Permutation group

In this section, we make a rst contact with quantum many-body systems. We consider
a physical system composed of N identical particles. “Identical” means that there is no
observable that could distinguish between them. In such a situation, the Hamiltonian
operator and in fact any other observable must be symmetric in the coordinates (that
is, in the position and spin degrees of freedom). In other words, if A is an arbitrary
observable, and P an arbitrary permutation from SN – so the permutation group with
N elements – then we will say that the particles are identical if
[A, P ] = 0 (2.1)
for all P ∈ SN . We will use the same symbol for the permutation and its representation
in Hilbert space. This is a very natural denition: If we could generate an observable
effect by simply permuting particles, then we could not call the particles indistinguish-
able, as we have found a way to distinguish them. For the moment being, this is only
a constraint to the involved Hilbert spaces and to their Hamiltonians (as H is also an
observable), but we will see in a minute what the consequences are. We will now more
closely consider this operator in Hilbert space.
Let ξj be the coordinates of the j-th particle, of N particles in total. Say, the state
vector of such a particle is from some Hilbert space H. This could in particular be the
position of the particle, a situation for which then H = L2 (R). Or, for a particle with
spin, the position and the third component of the spin, with Hilbert space
H = L2 (R) ⊗ C2 . (2.2)
A wave function is then simply a function with values
ψ(ξ) = ψ(x, σ), (2.3)
3
where x ∈ R and σ = {0, 1}. How does the entire Hilbert space then look like? Well,
we know that this is given by
H ⊗ · · · ⊗ H = H⊗N . (2.4)
5
6 CHAPTER 2. IDENTICAL PARTICLES
We will now investigate what happens to the wave function when we start permuting
particles.
If ψ(ξ1 , . . . , ξN ) is the wave function of the system, then the permutation operator
acts as
(P ψ)(ξ1 , . . . , ξN ) = ψ(ξP (1) , . . . , ξP (N ) ). (2.5)
It is clear that this really gives rise to a representation of the permutation group, as for
two permutations P1 , P2 we have that
((P2 P1 )ψ)(ξ1 , . . . , ξN ) = (P2 (P1 ψ))(ξ1 , . . . , ξN ). (2.6)
If the function ψ(ξ1 , . . . , ξN ) is a solution of the time independent Schrödinger equa-

tion,
(Hψ)(ξ1 , . . . , ξN ) = Eψ(ξ1 , . . . , ξN ), (2.7)
so is (P ψ)(ξ1 , . . . , ξN ) a solution with the same energy E, if
[H, P ] = 0. (2.8)
This is a consequence of the denition. Since for N particles, there exist N ! different
permutations (this is the cardinality of the symmetric group), we get in this way N !
different wave functions. Not all of them will be linearly independent, but usually we
will obtain a number of linearly independent solutions of the same energy. That is
to say, the energy values of Hamilton operators of identical particles will typically be
highly degenerate. This degeneracy is referred to as exchange degeneracy.
For an arbitrary observable A, we get
〈ψ|A|ψ〉 = 〈P ψ|A|P ψ〉. (2.9)
If ψ(ξ1 , . . . , ξN ; t0 ) is the wave function at time t0 and U (t − t0 ) is the time evolution

operator that for all permutations P satises
[U (t − t0 ), P ] = 0 (2.10)
(for obvious reasons, since the Hamiltonian P commutes with H)
U (t − t0 ) = e−i(t−t0 )H (2.11)
setting ~ = 1, then also
(P U (t − t0 )ψ)(ξ1 , . . . , ξN ; t0 ) = (U (t − t0 )P ψ)(ξ1 , . . . , ξN ; t0 ). (2.12)
This means that the permuted time evolved wave function is the same as the time
evolved permuted wave function. This means no less than that two wave functions
that are different only by a permutation, cannot be distinguished by any even sophis-
ticatedly chosen observable. This is hence a property that is preserved under time
evolution.
2.2. COMPLETELY SYMMETRIC AND ANTISYMMETRIC WAVE FUNCTIONS7
2.2 Completely symmetric and antisymmetric wave func-

tions
It makes a lot of sense to elucidate the situation for a small number of particles. For
one particle, the issue is void. Two particles are still quite boring in this respect, as
there are only completely symmetric and antisymmetric functions. Or, more precisely,
the Hilbert space decomposes int a direct sum of those components. Hence, the most
natural rst good example is the case of N = 3: In fact, for this we see all features we
need to know. In order to simplify the notation, we will write j for the coordinates ξj .
But it is important to stress that this is no number, but just a shortcut for a coordinate.
We have the wave function ψ(i, j, k), where i 6= j 6= l can take the values 1, 2, 3. From
these wave functions, one can construct the following functions,
1
ψS (1, 2, 3) = (ψ(1, 2, 3) + ψ(1, 3, 2) + ψ(2, 3, 1) + ψ(2, 1, 3)
61/2
+ ψ(3, 1, 2) + ψ(3, 2, 1)), (2.13)
1
ψA (1, 2, 3) = (ψ(1, 2, 3) − ψ(1, 3, 2) + ψ(2, 3, 1) − ψ(2, 1, 3)
61/2
+ ψ(3, 1, 2) − ψ(3, 2, 1)), (2.14)
as well as
1
ψM1 (1, 2, 3) = (2ψ(1, 2, 3) − ψ(1, 3, 2) + 2ψ(2, 1, 3) − ψ(2, 3, 1)
231/2
− ψ(3, 1, 2) − ψ(3, 2, 1)), (2.15)
1
ψM2 (1, 2, 3) = (ψ(1, 3, 2) − ψ(2, 3, 1) + ψ(3, 1, 2) − ψ(3, 2, 1)), (2.16)
2
1
ψM1′ (1, 2, 3) = (2ψ(1, 2, 3) − ψ(1, 3, 2) − 2ψ(2, 1, 3) − ψ(2, 3, 1)
231/2
− ψ(3, 1, 2) + ψ(3, 2, 1)), (2.17)
1
ψM2′ (1, 2, 3) = (ψ(1, 3, 2) + ψ(2, 3, 1) − ψ(3, 1, 2) − ψ(3, 2, 1)). (2.18)
2
The span of these six new wave functions is identical with that of the original functions,
needless to say. If ψ(1, 2, 3) already was completely symmetric, then only ψS (1, 2, 3)
would be non-vanishing. The pre-factors are only normalization constants. It is not
difcult to verify that
(P ψS )(1, 2, 3) = ψS (1, 2, 3) (2.19)
and
(P ψA )(1, 2, 3) = ±ψA (1, 2, 3), (2.20)
depending on whether the permutation P is even or odd (so can be generated from
an even or odd number of two-entry exchanges). Therefore, as the index indicates,
the functions ψS and ψA are completely symmetric or antisymmetric. There are also
the two two-dimensional subspaces spanned by Eqs. (2.15,2.16) and Eqs. (2.17,2.18),
respectively. They are again invariant under any permutation P ∈ S3 .
That is to say, the Hilbert space decomposes into a direct sum of four subspaces
H 3 = HS ⊕ H A ⊕ H M ⊕ H M ′ . (2.21)
This corresponds to a classication of the wave function according to complete sym-

metry in HS , antisymmetry HA , and mixed symmetry in HM ⊕ HM ′ . What is more,
the representations HM und HM ′ are equivalent. In this language, we also see why
N = 2 is a somewhat boring case: Here we simply have
H 2 = HS ⊕ H A . (2.22)
The general case of N particles is now clear from basic group theoretic considera-
tions. The Hilbert space decomposes as
H N = H S ⊕ H A ⊕ H S 1 ⊕ H S2 ⊕ . . . (2.23)
each of which is invariant under any permutation P ∈ SN . In physical system, only

the completely symmetric and antisymmetric wave functions can be realized. Why is
this so? An argument can be formulated as follows: Let {A1 , . . . , Ak } be a complete
set of compatible observables (that is, a complete set of observables that are mutually
commuting) and {a1 , . . . , ak } the corresponding eigenvalues of A1 , . . . , Ak . Then the
state vector |a1 , . . . , ak 〉 is unique, up to a phase. But if |a1 , . . . , ak 〉 ∈ HSi , then one
can nd a permutation P so that
P |a1 , . . . , ak 〉 6= eiα |a1 , . . . , ak 〉 (2.24)
for all α ∈ R, and still P |a1 , . . . , ak 〉 is an eigenvector of A1 , . . . , Ak with eigen-

values a1 , . . . , ak . This leads to a contradiction, however, as the subspaces are not
one-dimensional. This is the point: The invariant subspaces must be one-dimensional.
And, as a representation theoretic consideration shows (which we will not discuss here
in detail), this is true only for the completely symmetric and antisymmetric wave func-
tions. All other invariant subspaces are of higher dimension. This leads to an important
principle.
Symmetry principle: A state vector of a pure quantum state of a system of identi-

cal particles is eigner completely symmetric or antisymmetric under the exchange
of two particles.
One can also derive from considerations in relativistic quantum eld theory (which
we will not go into at this point), the following important spin statistics theorem.
Spin statistics theorem: The pure states of a system of identical particles is com-
pletely symmetric, if the spin is integer-values (bosons) and completely antisym-
metric, if the spin takes a half-integer value.
Their names originate from the boson and fermion statistics, to which we will come
to in a minute. It may be worth stressing – and this is a topic we will get to later – that
2.2. COMPLETELY SYMMETRIC AND ANTISYMMETRIC WAVE FUNCTIONS9
there are situations of a non-trivial topology that also allow for frationalized statistics.
Such a situation is encountered, for example, when charged quasi-particles with charge
q can strictly move in two spatial dimensions only, in a eld of magnetic ux Φ. In
this case, the exchange of such quasi-particles leads to a phase factor eiφ , where now
φ = qΦ/(~c). Only for φ = 0 and φ = π one arrives at the standard statistics.
Important examples of fermions are the usual suspects: Electrons, nuclei 3 He are
particles with spin 1/2 are hence fermions. Important examples for bosons are pho-
tons in the theory of quantized light, phonons in a body, as quantized excitations of
mechanical motion, and 4 He. They are all particles with spin 0 or 1. The fermionic
Pauli principle can be reduced to the antisymmetry of the wave function. It is a cute
observation that there are, by the way, additional constraints arising from the antisym-
metry of the wave function, except having at most one fermion per mode. In fact, the
single-body density operator is constrained to be contained in a polytope. We will later
understand what this means. 1
1 At this point, we have all reason in the world to be confused, and presumably this was an origin of
confusion even ealier. The universe consists of countless particles, in particular a large number of electrons.
If it was necessary, in the light of the previous considerations, to take them all into account and think of
completely antisymmetric wave functions, physics was doomed to failure: Physics is concerned with sys-
tems, not with describing the entire universe (which would be practically impossible anyway). If we could
ever only meaningfully talk about the entire universe, we would not get anywhere, if it was impossible to
meaningfully talk about subsystems. Fortunately, one can overcome this dilemma. In fact, to a very good
approximation, one can avoid such a radical description of the entire collection of electrons. Let us be more
specic: Let us assume we have two electrons that are sufciently far from each other, so that the overlap
of the position part and their interaction can be neglected. If ψ1 and ψ2 are the respective normalized wave
functions, the approximately normalized antisymmetric wave function is given by
1
ψ(ξ1 , ξ2 ) = √ (ψ1 (ξ1 )ψ2 (ξ2 ) − ψ1 (ξ2 )ψ2 (ξ1 )) (2.25)
2
(it is not strictly normalized for obvious reasons). We can now ask for the particle density in a point ξ =
(x, σ), that is, in a position in space and a spin component. This density is given by
∫ ∫
ρ(ξ) = dξ2 |ψ(ξ, ξ2 )|2 + dξ1 |ψ(ξ1 , ξ)|2 , (2.26)
where the integral is to be seen as an ordinary integral in space and a sum over the spin components. This
gives
( ∫ )
ρ(ξ) = |ψ1 (ξ)|2 + |ψ2 (ξ)|2 − 2re ψ1∗ (ξ)ψ2 (ξ) dξ1 ψ2∗ (ξ1 )ψ1 (ξ1 ) . (2.27)
Since ψ1 and ψ2 are highly localized in spatially disjoint regions R1 und R2 , the last term can be neglected,
and if ξ ∈ R1 ist, then
ρ(ξ) ≈ |ψ1 (ξ)|2 . (2.28)
The very same result we would have obtained if we had ignored the second electron altogether. Of course,
a similar reasoning holds in generality. The symmetrization and antisymmetrization has to be applied only
for particles that are relevant for a physical system. We do not have to include all other particles that do
not interact with a given particle. In the classical limit, particles are also highly localized in space, and the
inuence from symmetry can eb entirely neglected. Hence, we can feel safe again and continue with our
argument.
2.3 Non-interacting identical particles

Let us now again turn to systems of N identical particles. But now we do not consider
the structure of the underlying Hilbert space, but describe their Hamilton operator. This
Hamiltonian – assuming that the particles do not interact – takes the simple form
N N ( )
∑ ∑ 1 2
H0 = hj = pj + V j . (2.29)
j=1 j=1
2M
Here, all hj are the same, but act on the j-th particle only. For simplicity of notation,
we will assume that the spectrum of all hj is discrete, in order to avoid mathematical
ne print that is needed for continuous spectra. Let ψ1 , ψ2 , . . . be the normalized one
particle wave functions (or “orbitals”) of hj with eigenvalues E1 , E2 , . . . , that is, we
have
hj ψj (ξj ) = Ej ψj (ξj ). (2.30)
The wave function
ψ(ξ1 , . . . , ξN ) = ψ1 (ξ1 ) . . . ψN (ξN ) (2.31)
is then an eigenfunction of H0 with eigenvalue
E = E 1 + · · · + EN . (2.32)
In the light of the above considerations, the wave function can only be completely
antisymmetric or symmetric, depending on whether we are encountering bosons or
fermions.
2.3.1 Wave functions of fermions
Wave functions of fermions: The wave functions of fermionic systems are linear
combinations of antisymmetric functions of the form
1 ∑
ψN (ξ1 , . . . , ξN ) = 1/2
(−1)π(P ) ψ1 (ξP (1) ) . . . ψN (ξP (N ) ). (2.33)
(N !) P ∈S N
This expression can also be written as a determinant, as a so-called Slater determi-

nant of an N × N -matrix,
∣ ∣
∣ ψ1 (ξ1 ) . . . ψ1 (ξN ) ∣
∣ ∣
1 ∣ ψ2 (ξ1 ) . . . ψ2 (ξN ) ∣ .
∣ ∣
ψN (ξ1 , . . . , ξN ) = (2.34)
(N !)1/2 ∣∣ . . . ... ... ∣
∣
∣ ψN (ξ1 ) . . . ψN (ξN ) ∣
The exchange of two coordinates is reected by an exchange of two rows, leading

to a sign change of the determinant, concomitant with the desired antisymmetry. The
exchange of two orbitals correponds to the exchange of two columns, which again leads
to a sign change.
2.4. THE NON-INTERACTING GAS OF FERMIONS AND BOSONS 11
2.3.2 Wave functions of bosons

For bosons we arrive at a similar expression. In the sum (2.35), not all terms are
necessarily different. Instead, each term is shown N1 !N2 ! . . . many times.
Wave functions of bosons: The wave function of bosonic systems are linear com-
binations of symmetric functions of the form
1 ∑
ψN (ξ1 , . . . , ξN ) = ψ1 (ξP (1) ) . . . ψN (ξP (N ) ), (2.35)
(N !N1 !N2 . . . )1/2 P ∈S
N
where N1 , N2 , . . . is the number of particles in the orbitals labeled 1, 2, . . . , with

N1 + N2 + · · · = N .
It should be clear at this point that for both fermions and bosons, it is possible to
label the basis states merely by stating the occupation numbers (N1 , N2 , . . . ), where
N1 + N2 + · · · = N. (2.36)
For fermions these occupation numbers can be 0 or 1. For bosons, in contrast, any
non-negative integer is possible. The total energy of such a basis state is given by
E = N 1 E1 + N 2 E2 + . . . . (2.37)
Again, unsurprisingly, it depends only on the occupation numbers.
2.4 The non-interacting gas of fermions and bosons

A system of N identical particles, for which mutual interactions can be neglected, is
called ideal gas. We will briey discuss the situation here, assuming that in statistical
mechanics, this type of system will be discussed in more detail. For fermions, we know
that the occupation numbers Nj can only take the values 0 and 1. We assume that the
concepts of grand canonical ensembles are roughly clear. It is an equilibrium state in
quantum statistical mechanics under certain boundary conditions.
In the thermodynamic equilibrium state, the average occupation number of the or-
bital labeled j is for fermions given by
1
N̄j = , (2.38)
exp(−(µ − Ej )/(kB T )) + 1
which is the distribution function in the Fermi-Dirac-statistics. Here kB is the Boltz-

mann constant. The chemical potential µ, a real number, depends on the temperature
T and the total particle number N via
∑
N= (exp(−(µ − Ej )/(kB T )) + 1)−1 . (2.39)
j
Formally, µ takes the role of a Lagrange multiplier.

For bosons we obtain a very similar expression, only that now the occupation num-
bers are no longer constrained to be 0 or 1. It turns out that one only nds an equi-
librium value if exp((µ − Ej )/(kB T )) < 1 for all energies Ej . For this reason, the
chemical potential mu cannot be positive. One nds for the occupation numbers in
thermal equilibrium
1
N̄j = . (2.40)
exp((Ej − µ)/(kB T )) − 1
This is the distribution function of the ideal bosonic quantum gas, corresponding to the
Bose-Einstein-statistics The chemical potential, once again, it determined by
∑
N= (exp((Ej − µ)/(kB T )) − 1)−1 . (2.41)
j
From this, other statistical properties can be derived. Planck’s radiation formula, for
example, is an immediate consequence thereof. We will briey come back to it, without
anticipating too many topics of quantum statistical physics, which is a course in its own
right.
Chapter 3
Elements of second quantization
3.1 Preliminary remarks

We have seen that the basis vectors of a system of N identical particles is completely
described by the occupation numbers (N1 , N2 , . . . ). These occupation numbers tell us
how many particles are in each single-particle mode, and this is all that is needed to
capture the basis state. We can hence write the basis vectors in the simple form
|N1 , N2 , . . . 〉. (3.1)
In this context, it is helpful to grasp systems with the help of creation and annihilation
operators. This is a concept rather familiar from the rst course of quantum theory. We
will nevertheless repeat it here and go in more depth.
3.2 Second quantization for bosons

Somewhat pedantically, we will introduce the Fock space: This separable Hilbert space
is the direct sum of Hilbert spaces of xed particle number, that is
∞
⊕
H= H(N ) , (3.2)
N =0
where H(N ) , N ≥ 1, is the Hilbert space of physically realizable pure states of N

bosonic particles. H(0) = C is the one-dimensional vector space spanned by |ø〉
aufgespannt wird, the vacuum state vector. Note that in the literature it is common
to call both the space the Fock space as well as the one that is spanned by the vector
of a xed occupation number. Let us not get too much hung up and rather stress the
occupation number basis for bosons.
5
6 CHAPTER 3. ELEMENTS OF SECOND QUANTIZATION
Occupation numbers for bosons: The vectors
B = {|N1 , N2 , . . . 〉 : Nj ∈ N0 , j = 1, 2, . . . } (3.3)
form a basis of the Hilbert space of identical bosons.
This basis is orthonormal and complete, in that
〈N1′ , N2′ , . . . |N1 , N2 , . . . 〉 = δN1′ ,N1 δN2′ ,N2 . . . , (3.4)

∑
|N1 , N2 , . . . 〉〈N1 , N2 , . . . | = . (3.5)
N1 ,N2 ,...
Let us dene now the annihilation operator, which we have already encountered when
discussing the harmonic oscillator in the rst course.
Annihilation operator for bosons: The annihilation operator for mode j is an

operator in H, for which
√
bj |N1 , . . . , Nj , . . . 〉 = Nj |N1 , . . . , Nj − 1, . . . 〉 (3.6)
holds.
The– admittedly somehat unpleasant – name of the annihilation operator makes a

lot of sense: It suggests to remove a particle in mode labeled j. We know immediately
that
bj |0, 0, . . . 〉 = bj |ø〉 = 0. (3.7)
Note also that |ø〉 is not the null vector of the vector space. In the same way, we get via
looking at the adjoint the following expression that we dedicate a box to only because
we will so frequently make use of in the latter.
Creation operators for bosons: A creation operator for mode j is an operator in

H for which
b†j |N1 , . . . , Nj , . . . 〉 = Nj + 1|N1 , . . . , Nj + 1, . . . 〉.

√
(3.8)
In a bit of an ination of boxes we will consider the following important operator,

labeling the particle number in a given orbital or a mode:
3.2. SECOND QUANTIZATION FOR BOSONS 7
Number operator for bosons: The operator nj = b†j bj with action
n†j |N1 , . . . , Nj , . . . 〉 = Nj |N1 , . . . , Nj , . . . 〉 (3.9)
is called number operator.
It will happen that we give operators hats, yet most often not to render the notation
not unnecessarily clumsy. It should be clear at this point, however, that nj is an oper-
ator in H and not only a number, unlike Nj . The total particle number operator Die
Gesamtzahl der Teilchen is the operator
N
∑
nj . (3.10)
j=1
A system in a state with exactly N particles will hence will give rise to an expectation
value N for the total particle number operator. Obviously, the spectrum of this operator
is that of the non-negative numbers. It follows from the above denitions that bosonic
operators full the following commutation relations.
Bosonic commutation relations: The bosonic creation and annihilation operators

satisfy
[bj , b†k ] = δj,k , (3.11)
and
[bj , bk ] = [b†j , b†k ] = 0. (3.12)
As usual, for pairs of operators, the commutator is dened as
[A, B] = AB − BA. (3.13)
It is not uncommon to the dene the position and momentum operators

√
xj = (bj + bj )/ 2, (3.14)
√
pj = i(bj − bj )/ 2 (3.15)
We nd immediately
|N1 , N2 , . . . 〉 = (N1 !N2 ! . . . )−1/2 (b†1 )N1 (b†2 )N2 . . . |ø〉. (3.16)
The precise order of denitions and corollaries are not consistent in text books on
the subject. It should be clear that the bosonic commutation relations – the rules that
dene the bosonic algebra – are already dened via the above rules and give rise to the
bosonic operators up to a phase. We could have introduce rst the bosonic commutation
relations and would have ended up in the occupation number representation.1
3.3 Second quantization for fermions
3.3.1 Creation and annihilation operators for fermions

Let us start by dening the annihilation and creation operators for fermions. They are
dened as follows.
1 Let us here see how we can derive the occupation number representation from the bosonic commutation
relations. We are somewhat innocent here, but it should be clear that this can be made fully rigorous: We
will present the argument for one orbital or mode that is hence no longer necessary. Let us hence look at
an operator b for which [b, b† ] = 1 holds. b† b is also a Hermitian and positive operator. Its spectrum must
therefore be included in the non-negative real numbers. Let us now consider
[b† b, b] = −b. (3.17)
From Eq. (3.17) it follows that b applied to an eigenvector of b† b of eigenvalue N again delivers an eigen-
vector, the eigenvalue of which is reduced by 1,
b† b(b|N 〉) = b(b† b)|N 〉 + [b† b, b]|N 〉

= (N − 1)(b|N 〉). (3.18)
Iterated application of this rule must lead to a zero eigenvalue. Otherwise, we would arrive at a contradiction
with the observation that b† b is a positive operator. Hence 0 must be a (and the smallest) eigenvalue of the
number operator. In the same way, we nd
[b† b, b† ] = b† , (3.19)
so that b† is a creation operator that increases the eigenvalue of the number operator by 1. Hence, we arrive
at the conclusion that the spectrum of b† b are exactly the non-negative integers with eigenvectors {|N 〉}.
Similarly,
b|N 〉 = N 1/2 |N − 1〉, (3.20)

†
b |N 〉 = (N + 1)1/2 |N + 1〉, (3.21)
up to a global phase factor. This is actually the only freedom we have here, and without loss of generality,
we can set the phase to zero, since
〈N |b† b|N 〉 = N. (3.22)
The same argument can be applied to a multi-mode system, simply by adding a label j to the respective
operators, and noting that operators acting on different modes must commute. Slightly more precisely put,
the core question is whether every irreducible representation reecting (3.11) und (3.12) is unitary equivalent
to the Fock representation. There are some technicalities coming into play here, arising from the observation
that the operators bj and b†j are unbounded. For this reason, in mathematical physics, the commutation
relations are usually not dened directly in terms of bj und b†j , but for their exponentiated form, the so-
called Weyl operators. In this framework one indeed encounters under the additional assumption of the
existence of a vacuum a unique Fock representation up to a phase.
3.3. SECOND QUANTIZATION FOR FERMIONS 9
Annihilation and creation operators for fermions: An annihilation operator for

mode j is an operator, for which
∑j−1
Nk
fj |N1 , . . . , Nj , . . . 〉 = (−1) k=1 Nj |N1 , . . . , 1 − Nj , . . . 〉 (3.23)
holds true, where now Nj ∈ {0, 1}. For this reason, the fermionic creation operator
satises
∑j−1
fj† |N1 , . . . , Nj , . . . 〉 = (−1) k=1 Nk
(1 − Nj )|N1 , . . . , 1 − Nj . . . 〉. (3.24)
As before, the vacuum |ø〉 is mapped by all fj onto the null vector. Again, the –
now fermionic – number operator
nj = fj† fj (3.25)
has the same role and is given the same name as the equivalent of a bosonic opera-
tor. These are the familiar laws as they originate from the fermionic anticommutation
relations. Such anticommutation rules are given for pairs of operators as
{A, B} = AB + BA. (3.26)
Anticommutation relations for fermionic operators: For fermionic creation and

annihilation operators, the following familiar anticommutation rules hold true,
{fj , fk† } = δj,k , (3.27)
{fj , fk } = {fj† , fk† } = 0. (3.28)
As before, we also nd the analogous occupation number expressions for fermions.
Occupation numbers for fermions: The vectors
B = {|N1 , . . . , Nj , . . . 〉 = (f1† )N1 . . . (fj† )Nj . . . |ø〉, N1 , N2 , · · · ∈ {0, 1}}

(3.29)
form a basis of the basis of identical fermions.
Basically all of the above expressions are still valid, but with replacing the commu-
tation relations by anticommutation relations.2
2 The Fock representation is, by the way, again unique and dened essentially by the anticommutation
relation, if we postulate the existence of a vacuum and a fully occupied state.

3.3.2 Causality, superselection rules and Majorana fermions

Familiar? Wait a second. Eq. (3.23) contains the innocent looking phase
∑j−1
Nk
(−1) k=1 (3.30)
that contains all the particle numbers Nk of the modes to the “left” of orbital labeled j.
This expression might be surprising and confusing in more than one way:
• On the one hand, this expression depends on the order of fermions. But this
order is obviously completely up to us. Our predictions of physical properties
must not depend on how we order modes in our description.
• More disturbing is the property that one could in principle observe this phase,
leading to a erce violation of causality.
The latter is compelling: it is easy to see that one can generate situations involving
three modes, one, say, on Mars and two kept on Earth. Then the measurement of an
operator that involves an odd number of fermionic annihilation and creation operators
reveals by a phase whether a particle is present at Mars or not. But in this way, one can
do instantaneous communication, faster than light: One merely has to place a particle in
the mode on Mars or not. Such superluminal communication is not allowed in physics.
The resolution of this apparent dilemma is one that is stated in surprisingly few text
books, even though this issue often gives rise to confusion. In fact, the resolution is that
not all observables are allowed for fermionic operators. But only those that respect the
superselection rule of the parity of fermion number. Any legitimate fermionic operator
must commute with the parity operator The parity operator can be written as
2N
∏
P = iN cj (3.31)
j=1
in terms of the 2N Majorana fermions.
Majorana fermions: The Majorana fermions of N fermionic modes are dened as
c2j−1 = fj† + fj , (3.32)

c2j = (−i)(fj† − fj ) (3.33)
for j = 1, . . . , N . They are Hermitian, cj = c†j and full the fermionic anticom-
mutation relations
{cj , ck } = δj,k (3.34)
for j, k = 1, . . . , 2N .
These Majorana fermions are “half fermions” and are exciting in their own right.
They can be prepared in mesoscopic structures, showing topological features. We will
3.3. SECOND QUANTIZATION FOR FERMIONS 11
get back to that later in the course. In other words, the Hilbert space can be decomposed
into a direct sum
H = Heven ⊕ Hodd , (3.35)
involving terms containing an even and an odd number of fermions, respectively. No
physical interaction can mix the two terms and respects this direct sum. Again, without
this rule one can easily construct paradoxa of the above type.3 Obviously, for fermionic
annihilation operators, we always have
(fj† )2 = 0, (3.36)
as a manifestation of the fact that no two fermions can occupy the same orbital.
3.3.3 Single-body density operators and Pauli principles

As a nal remark, we briey mention single-body density operators for fermionic sys-
tems of n modes and N particles. For n fermionic modes and a state ρ, we can look at
the n × n matrix C with entries
Cj,k = tr(fj† fk ρ)/N
. (3.37)
This is basically a correlation function. We easily nd its properties. It is Hermitian by

denition, so that
C = C †. (3.38)
We also nd that it is positive semi-denite. This is slightly less obvious, but can be
seen by making a suitable fermionic mode transformation, so that
C ≥ 0, (3.39)
which means that the real eigenvalues of C are non-negative. Also
tr(C) = 1. (3.40)
That is to say, C has precisely the properties of a density operator. For this reason, it
is called one-body density operator. It is no density operator as a quantum state, of
course. But it captures the correlation structure of a fermionic system in a correlation
matrix that takes the form of a density operator. The familiar Pauli principle can now
be stated that the eigenvalues λ1 , . . . , λn satisfy
0 ≤ λj ≤ 1. (3.41)
3 There is no bosonic equivalent of the parity of fermion number. For massless bosons, such as photons,
one can perfectly

√ well create superpositions of particle numbers. For example, state vectors of the form
(|0〉 + |1〉)/ 2 can be prepared. Only if the particle have mass, the mass superselection rule applies, and
again no superposition is possible. Massive bosonic particles can hence not be prepared in a superposition
of different particle numbers.
This reects the fact that in each mode, there must be at least zero and at most one
fermions. It originates from the antisymmetry of the wave function of identical fermionic
particles. However, there are more constraints arising from the antisymmetry. There is
an entire family of generalized Pauli principles: These are general constraints any one-
body density operator must satisfy. They take the form of a simplex of the eigenvalues
λ1 , . . . , λn of C. It is interesting to see that these constraints can be completely charac-
terized. It is also worth stressing that the stability of matter is essentially a consequence
of the Pauli principle, and not of interactions.
Chapter 4
Field operators
4.1 Denition of eld operators
4.1.1 Bosonic eld operators
We will now get back to bosonic systems; in fact, for some courses to come, we will to
an extent oscillate between looking at bosonic and fermionic many-body systems. Let
us start from noting that the one-body basis can be chosen arbitrarily in principle. When
going from one set of bosonic operators to a new set, reecting a basis transformation
of this kind, this can be captured by a transformation
∑
∞
bgj = 〈gj |ψk 〉bk , (4.1)
k=1
∑∞
b†gj = 〈ψk |gj 〉b†k , (4.2)
k=1
dened by the scalar product between single particle state vectors. Going to a picture of
eld operators can be seen as a transformation of this kind. They are very much help-
ful. On a higher level, the situation at hand is as follows: In second quantization, one
merely states how many bosons are occupying what single-particle orbital. When we
would like to nd out how Hamiltonian interactions dened in the position representa-
tion manifest themselves, we would like to have some contact with the single-particle
wave functions. This connection is delivered by the eld operators. Without such eld
operators, it would not be obvious how to derive a Hamiltonian in second quantization
in terms of bosonic creation and annihilation operators. They often serve the purpose
of a vehicle: At the end of the day, the Hamiltonian does not necessarily contain the
eld operators any more, but they are helpful to get the right expressions in the rst
place. They are so important that they deserve a box.
5
6 CHAPTER 4. FIELD OPERATORS
Field operators: The bosonic eld operators are dened as

∫ ∫
bj = dξψj∗ (ξ)Ψ(ξ), b†j = dξψj (ξ)Ψ† (ξ), (4.3)
conversely,
∑ ∑
Ψ(ξ) = ψj (ξ)bj , Ψ† (ξ) = Ψ∗j (ξ)b†j , (4.4)
j j
and satisfy
[Ψ(ξ), Ψ(ξ ′ )] = [Ψ† (ξ), Ψ† (ξ ′ )] = 0, (4.5)
as well as
[Ψ(ξ), Ψ† (ξ ′ )] = δ(ξ − ξ ′ ). (4.6)
We can write state vectors in terms of eld operators as follows. If this looks a
bit odd at rst, please be patient, it will be used to formulate Hamiltonians of inter-
acting models in a neat form, and again, to relate the wave functions in the position
representation to a second quantized picture.
State vectors in terms of eld operators: A general symmetrized state vector

|ψN 〉 of N particles can be written as
∫
1
|ψN 〉 = √ dξ1 . . . dξN ψN (ξ1 , . . . , ξN )Ψ† (ξN ) . . . Ψ† (ξ1 )|ø〉. (4.7)
N!
The normalization of the state vectors follows from the symmetrized wave function.
We hence start from the vacuum that we successively ll up, rst by placing a particle
at ξ1 by means of Ψ† (ξ1 ), then with a particle at ξ2 using Ψ† (ξ2 ) and so on. The wave
function ψN species the amplitude of the respective term. Since the wave function is
symmetrized, we have for bosonic wave functions of identical particles
ψN (ξ1 , . . . , ξN ) = ψN (ξN , . . . , ξ1 ) (4.8)
and analogously for any other permutation of the coordinates, since these expression
deliver identical expressions under symmetrization. We now turn to an interesting
expression. The following is true.
4.1. DEFINITION OF FIELD OPERATORS 7
Action of eld operators on state vectors: When applying a bosonic eld operator
to a state vector, we get
√ ∫
Ψ(ξ)|ψN 〉 = N dξ1 . . . dξN −1 ψN (ξ1 , . . . , ξN −1 , ξ)Ψ† (ξN −1 ) . . . Ψ† (ξ1 )|ø〉.
(4.9)
This expression will be helpful in a moment. We would like to derive this formula
for two particles at rst, then it will become clear how to obtain general expressions of
this type. We have
∫
1
|ψ2 〉 = √ dξ1 dξ2 ψS (ξ1 , ξ2 )Ψ† (ξ1 )Ψ† (ξ2 )|ø〉. (4.10)
2
Therefore, making use of the commutation relations for bosonic eld operators, we get
∫
1
Ψ(ξ)|ψ2 〉 = √ dξ1 dξ2 ψ2 (ξ1 , ξ2 )Ψ(ξ)Ψ† (ξ1 )Ψ† (ξ2 )|ø〉
2
∫
1
= √ dξ1 dξ2 ψ2 (ξ1 , ξ2 )Ψ† (ξ1 )Ψ(ξ)Ψ† (ξ2 )|ø〉
2
∫
1
+ √ dξ2 ψ2 (ξ, ξ2 )Ψ† (ξ2 )|ø〉
2
∫
1
= √ dξ1 dξ2 ψ2 (ξ1 , ξ2 )Ψ† (ξ1 )Ψ† (ξ2 )Ψ(ξ)|ø〉
2
∫
1
+ √ dξ1 ψ2 (ξ1 , ξ)Ψ† (ξ1 )|ø〉
2
∫
1
+ √ dξ2 ψ2 (ξ, ξ2 )Ψ† (ξ2 )|ø〉. (4.11)
2
Since now ψ2 (ξ1 , ξ) = ψ2 (ξ, ξ1 ) for all ξ and ξ1 , since this function is symetric, and
since
Ψ(ξ)|ø〉 = 0 (4.12)
for all ξ, because annihilation operators acting on the vacuum always lead to the zero
vector, we have
√ ∫
Ψ(ξ)|ψ2 〉 = 2 dξ1 ψ2 (ξ1 , ξ)Ψ† (ξ1 )|ø〉. (4.13)
For wave functions in the position representation we arrive at

√
(Ψ(ξ)ψ2 )(ξ1 ) = 2ψ2 (ξ1 , ξ). (4.14)
We can argue in a similar fashion for N particles: This is left to the reader as an
exercise. But the idea should be clear: We again can exhange neighbouring terms
iteratively.
4.1.2 Fermionic eld operators

For fermionic eld operators there is not so much left to say. Let us briey state the
anticommutation relations for fermionic eld operators, when the coordinates are ξ =
(x, σ), as position and a third spin component.
Fermionic eld operators: The fermionic eld operators fulll the anticommuta-
tion relations
{Ψ(x, σ), Ψ† (x′ , σ ′ )} = δ(x − x′ )δσ,σ′ . (4.15)
Equipped with these expressions we will now turn to rst applications.
4.2 Hamiltonians of identical particles

4.2.1 Hamiltons in second quantization
Being prepared in this fashion, we can describe interacting bosonic systems consisting
of N constituents. The single particle state vectors will be denoted as {|ψk 〉} with
wave function ψk . The Hamilton operator can then be written as
(N ) (N )
H (N ) = H0 + H1 , (4.16)
(N )
∑
N
H0 = Fj , (4.17)
j=1
(N )
∑
N
H1 = Vj,k . (4.18)
k>j=1
Here F is a single particle operator, V a two particle operator. Fj only acts on the
particle labeled j, while Vj,k captures the interaction between particles labeled j and
k. In matrix form, for each of the Fj , we have
∑
F = 〈ψl |F |ψk 〉|ψl 〉〈ψk |. (4.19)
k,l
The matrix with elements
{〈ψl |F |ψk 〉 : k, l = 0, 1, . . . }, (4.20)
to state this again, is the matrix that expresses the single particle Hamiltonian F in the
single particle state vectors {|ψk 〉}. We will now try to write this Hamiltonian in terms
of creation and annihilation operators. As anticipated, eld operators will come to our
rescue.
4.2. HAMILTONIANS OF IDENTICAL PARTICLES 9
We will see that this Hamiltonian is the projection of
H = H 0 + H1
∑
= 〈ψj |F |ψk 〉b†j bk
j,k
1 ∑
( )
+ 〈ψj , ψi |V |ψk , ψl 〉 + 〈ψi , ψj |V |ψk , ψl 〉 b†j b†i bl bk (4.21)
4
i,j,k,l
onto the completely symmetric subspace of the N -particle Hilbert space. The inter-
action V is local, in that the position representation of the operator V satises des
Operators V
〈ξ1′ , ξ2′ |V |ξ1 , ξ2 〉 = V (ξ1 , ξ2 )δ(ξ1′ − ξ1 )δ(ξ2′ − ξ2 ). (4.22)
This is a very natural property of a point interaction. Of course, this interaction is sym-
metric, in that V (ξ1 , ξ2 ) = V (ξ2 , ξ1 ). This means that in the position representation,
we have
∫
(〈ψj , ψi |V |ψk , ψl 〉 + 〈ψi , ψj |V |ψk , ψl 〉) = dξ1 dξ2 ψj∗ (ξ1 )ψi (ξ2 )V (ξ1 , ξ2 )ψk (ξ1 )ψl (ξ2 )
∫
+ dξ1 dξ2 ψj∗ (ξ2 )ψi (ξ1 )V (ξ1 , ξ2 )ψk (ψ1 )ψl (ξ2 ).
(4.23)
From this, it should be clear how to proceed to the interaction term H1 , however.
In order to show that the two formulations of the Hamiltonian give rise to the same
expressions, we start from H0 . Admittedly, we leave it at that. It should be clear from
this argument how to treat H1 as well, which follows exactly the same logic. We will
more precisely show that the two expressions are the same when being computated for
arbitrary state vectors. For two arbitrary completely symmetric state vectors |ψN 〉 and
′
|ψN 〉 we nd
∑
′
〈ψN |H0 |ψN 〉 = ′
〈ψj |F |ψk 〉〈ψN |b†j bk |ψN 〉
j,k
∑
′
= 〈ψj |F |ψk 〉〈bj ψN |bk ψN 〉, (4.24)
j,k
where we merely have moved an annihilation operator into the dual vector. From the
dening expressions of bosonic eld operators, we nd
∫
bj = dξψj∗ (ξ)Ψ(ξ), (4.25)
and hence
∑∫
′
〈ψN |H0 |ψN 〉 = dξdξ ′ ψj∗ (ξ ′ )ψk (ξ)〈ψj |F |ψk 〉〈Ψ(ξ ′ )ψN |Ψ(ξ)ψN 〉. (4.26)
j,k
Now we can apply the rule Gl. (4.9) that we have established above. In this way, we
get the somewhat tedious expression
∑∫
′
〈ψN |H0 |ψN 〉 = N dξdξ ′ ψj∗ (ξ ′ )ψk (ξ)
j,k
× 〈ψj |F |ψk 〉dξ1 . . . dξN −1 dξ1′ . . . dξN

′
−1
′ ∗ ′ ′ ′
× (ψN ) (ξ1 , . . . , ξN −1 , ξ )ψN (ξ1 , . . . , ξN −1 , ξ)
† †
× 〈ø|Ψ(ξ1′ ) . . . Ψ(ξN
′
−1 )Ψ (ξN −1 ) . . . Ψ (ξ1 )|ø〉. (4.27)
Of course, and fortunately, most terms will disappear, as they act onto the vacuum.
And so we get
∑∫
′
〈ψN |H0 |ψN 〉 = N dξdξ ′ ψj∗ (ξ ′ )ψk (ξ)〈ψj |F |ψk 〉dξ1 . . . dξN −1
j,k
′ ∗
× (ψN ) (ξ1 , . . . , ξN −1 , ξ ′ )ψN (ξ1 , . . . , ξN −1 , ξ). (4.28)
′
Since the wave functions ψN and ψN are symmetric in their arguments, we obtain
nally
∑
N ∑
′
〈ψN |H0 |ψN 〉 = dξ1 . . . dξl . . . dξN dξl′ (4.29)
l=1 j,k
× ψj∗ (ξ ′ )ψk∗ (ξ)〈ψj |F |ψk 〉

′ ∗
× (ψN ) (ξ1 , . . . , ξl′ , . . . , ξN )ψN (ξ1 , . . . , ξl , . . . , ξN ).
This is what we intended to show. Let us remind ourselves that each term is
∑
F = 〈ψj |F |ψk 〉|ψj 〉〈ψk | (4.30)
j,k
so that indeed, ∑
′ ′
〈ψN |H0 |ψN 〉 = 〈ψN | Fj |ψN 〉. (4.31)
j
We have therefore shown that a Hamilton operator that is composed of single particle
terms can equally well be written in terms of creation and annihilation operators. This
is done in a way so that the coefcient matrix is nothing but the matrix form of the
single body problem. A very similar argument applies to the interacting term H1 ,
again that the two formulations are identical on the symmetric subspace. It should be
obvious that the specics of F and V do not matter here. We have learned something
very important here.
• One can write quantum many-body Hamiltonians in second quantization directly
in terms of bosonic creation and annihilation operators. In order to compute
Hamiltonians, one does not go back to rst quantization and compute expres-
sions in the original Hilbert space. Instead, one can directly remain in the occu-
pation number basis. The coefcients in the Hamiltonian are obtained by solving
a one-body and two-body problem once and for all. This is an immense simpli-
cation!
• This expression makes sense even if the particle number is not held constant.
Of course, we still have the freedom to pick the state vectors {|ψj 〉}. One usually
picks the energy eigenvectors of the single-particle Hamiltonian F with energy values
Ej . This means that the single-particle Hamiltonian then becomes diagonal as
∑ ∑
F = 〈ψl |F |ψl 〉|ψl 〉〈ψl | = El |ψl 〉〈ψl |. (4.32)
l l
In the literature, these expressions are often written as
{|ψk 〉 : k = 0, 1, . . . } = {|k〉 : k = 0, 1, . . . }. (4.33)
We will, unless there is the risk of confusion, often write the single-particle basis as
{|k〉}, even if they are not the eigenvectors of the harmonic oscillator, but some arbi-
trary single-particle Hamiltonian. In this notation,
∑
F = El |l〉〈l|. (4.34)
l
The resulting expression is so important that it deserves a box in its own right.
General Hamiltonian of bosonic identical particles in second quantization: A

general Hamiltonian with two-body interaction takes the form
H = H 0 + H1 (4.35)
∑ † 1 ∑ † †
= E j bj b j + (〈i, j|V |k, l〉 + 〈j, i|V |k, l〉) bj bi bl bk .
j
4
i,j,k,l
an.
This Hamiltonian has a simple interpretation. One can interpret the bosonic op-
erators as taking out bosons, letting them interact and putting them back again. The
amplitude by which this happens is dened by the single- and two-particle problem.
Specically for the interaction term, one takes out two bosons, lets them interact and
places them back. This is a very compact and simple form to capture interacting quan-
tum many-body problems. Again, to derive this form, one only has to solve the one-
and two-body problems once and for all, to get the appropriate coefcients. Once this
is done, one does not have to return to the microscopic interactions any more. We can
hence always think in terms of occupation numbers.
4.2.2 Hamiltonians expressed in eld operators

This Hamiltonian is call, as often eluded to already, a Hamiltonian in second quantiza-
tion. The eld operators Ψ and Ψ† are often regarded as “second quantized instances
of a wave function”. This does not mean that we now have formulated a new quantum
mechanics of any sort, or a renement of the “old” quantum mechanics we are more
familiar with: It is still the same quantum theory. It is just a neat, concise, and simple
formalism of capturing quantum many-body systems of identical particles. There is
some substance, however, to the claim of performing a “second quantization” of wave
functions, and we will hint at that in this subsection.
Let us look into that. We focus on the situation of bosonic identical particles in an
external potential V1 , interacting via a two-body interaction. In this situation, we have,
to start with,
P2
F = + V1 , (4.36)
2M
which leads for an arbitrary basis set {|ψk 〉}, using
∑
H0 = 〈ψj |F |ψk 〉b†j bk ,
j,k
to the expression
∫ ∑ ~2 ∆
( )
H0 = dξ ψj∗ (ξ) + V1 (ξ) ψk (ξ)b†j bk . (4.37)
j
2M
Turning to a picture of eld operators, this becomes

∫ ( 2 )
~ ∆
H0 = dξΨ† (ξ) − + V1 (ξ) Ψ(ξ). (4.38)
2M
Performing an analogous step for the interacting part of the Hamiltonian, we get the
form of a Hamiltonian of particles in a potential in second quantization, expressed in
eld operators as follows.
Hamiltonians in terms of eld operators:

∫ ( 2 )
~ ∆
H = dξΨ† (ξ) − + V1 (ξ) Ψ(ξ)
2M
∫
1
+ dξdξ Ψ (ξ )Ψ† (ξ)V2 (ξ, ξ ′ )Ψ(ξ)Ψ(ξ ′ ).
′ † ′
(4.39)
2
The rst term looks a bit like a single-body operators, despite the fact that it is
of course an expression involving eld operators. According to the denition of eld
operators, it is to be read as
∑
∆Ψ(ξ) = bj ∆(ψj (ξ)). (4.40)
j
The factor 1/2 originates from the indistinguishability of the congurations ξ, ξ ′ and
ξ ′ , ξ. The total particle number operator expressed in eld operators is
∫
dξΨ† (ξ)Ψ(ξ), (4.41)
the expectation value of which provides the total particle number. In a system consist-
ing of exactly N particles, one has
∫
dξ〈ψN |Ψ† (ξ)Ψ(ξ)|ψN 〉 = N. (4.42)
Let us be aware of the fact, however, that we can also allow for superpositions of
particle numbers, unless a super-selection rule prevents us from doing so. Hence, this
particle number can in principle be an arbitrary non-negative real number. It looks like
the particle density of a single-particle problem. But again, now Ψ is an operator and
not a complex values function.
The analogy to the “rst quantization” becomes even more transparent when con-
sidering the time evolution of eld operators in the Heisenberg picture. This may at rst
look a bit awkward, but let us not forget that any operator, including a eld operator,
follows some time evolution in the Heisenberg picture. We have
Ψ(ξ, t) = eiHt Ψ(ξ, 0)e−iHt (4.43)
for the Hamiltonian (4.39), again setting /~ = 1. Writing this in differential form, we
get the differential equation
( 2 )
∂ ~ ∆
i~ Ψ(ξ, t) = − + V1 (ξ) Ψ(ξ, t)
∂t 2M
∫
+ dξ ′ Ψ† (ξ ′ , t)V2 (ξ, ξ ′ )Ψ(ξ ′ , t)Ψ(ξ, t). (4.44)
This is an equation of motion taking the form of a non-linear Schrödinger equation, or

even a linear Schrödinger equation in the simple case of having no interaction what-
seoever. This is quickly shown. One starts from the Heisenberg equations of motion
∂
i Ψ(ξ, t) = −[H, Ψ(ξ, t)]
∂t
= −eiHt [H, Ψ(ξ, 0)]e−iHt , (4.45)
and makes use of the general property of commutators
[AB, C] = A[B, C] + [A, C]B. (4.46)
Again, we see the formal resemblance with the single-body picture of basic quantum
mechanics.
Chapter 5
Fermi gases
Fermi gases play a key role in physics. The most prominent example is that repre-
senting electrons in solid state systems, but also star models in astrophysics can be
described as Fermi gaes. With small variations, we can also capture atoms in this fash-
ion. For most of this chapter, we will think of a gas of atoms, but it should be clear that
the formalism as such smoothely carries over to any other kind of Fermi gas.
5.1 Preliminary remarks

We will start from reconsidering an ideal Fermi gas exhibiting no interaction, but this
time with more care. We will see that equipped with our new formalism, we can already
come to a number of interesting conclusions. Then we will turn to a model of a gas of
electrons featuring a Coulomb interaction. These considerations will also prepare us
well for our discussion of superconductivity.
In order to describe an ideal Fermi gas in a container, it is most convenient to make
use of periodic boundary conditions. This may not be the most natural of all bound-
ary conditions, but it renders Fourier analysis easily available. It should also be clear
that for large containers, the boundary conditions do not matter for bulk properties,
and hence this seems a meaningful choice. Hence, the following single-particle wave
functions seem most suitable,
1
ψp,σ (x, σ ′ ) = √ eipx/~ δσ,σ′ , (5.1)
V
where σ, σ ′ stands for the third component of a spin and
2π~
p= n, (5.2)
L
where n ∈ Z3 is a vector of integers. To be sure, this is the position-spin representation
of the wave function, which is also why we have the Kronecker delta to the right.
This quantization property follows from the boundary condition. The volume of the
container is V = L3 in three spatial dimension with length of each side of L. Product
5
6 CHAPTER 5. FERMI GASES
like px are to be read as standard scalar products. These eigenfunctions are orthonormal
as ∫
∗
dxψp,σ (x, σ ′ )ψp′ ,σ′′ (x, σ ′′′ ) = δp,p′ δσ,σ′ δσ′ ,σ′′ δσ′′ ,σ′′′ . (5.3)
Here and in the following, the coordinates ξ = (x, σ) are explicitly given in position
and spin component. The eld operators are now
∑ 1
Ψ(x, σ) = √ eipx/~ fp,σ , (5.4)
p V
where fp,σ denote the fermionic annihilation operators for momentum p and spin com-
ponent σ. The anti-commutation relations between the fermionic eld operators are
1 ∑ ip(x−x′ )/~
{Ψ(x, σ), Ψ† (x′ , σ ′ )} = δσ,σ′ e . (5.5)
V p
Now we have
1 ∑ ip(x−x′ )/~ ∑
e = δ(x − x′ + nL). (5.6)
V p n
Since we are interested what happens inside the box but not in its “replicas”, we nd
the standard fermionic anti-commutation relations.
It is worth hesitating at this point and to considering other quantities in the momen-
tum representation. The number density operator
∑
Ψ† (x, σ)Ψ(x, σ) (5.7)
σ
becomes under a Fourier transform to

∑∫
dxΨ† (x, σ)Ψ(x, σ)e−iqx (5.8)
σ
and – making use of Eq. (5.4) – this becomes

∑∫ 1 ∑ ∑ −ipx †
dx e fp,σ fk,σ e−iqx . (5.9)
V p
σ k
Exploiting orthonormality, this gives

∑∑
†
fp,σ fp+q . (5.10)
σ p
This is the Fourier transformed number density operator, different 

fromthe number
†
density operator in the momentum representation, which is given by σ p fp,σ fp,σ .
5.2. GROUND STATE OF THE IDEAL FERMI GAS 7
5.2 Ground state of the ideal Fermi gas

It is now the right moment to get back to the Hamiltonian of N non-interacting fermions
in a box, expressed in momentum space. This Hamiltonian is given by
∑ ∑ (~k)2 †
H= fk,σ fk,σ . (5.11)
2M
k σ
Obviously, we only have a kinetic term here, and do not encounter any terms reecting
a physical interaction or external potentials. The ground state is now the following: Es
handelt sich also lediglich im Terme der kinetischen Energie, ohne Wechselwirkungen
oder
Ground state of the ideal Fermi gas: The state vector of the ground state of the
non-interacting Fermi gas is given by
∏ ∏
†
|ψN 〉 = fp,σ |ø〉. (5.12)
p,|p|<kF σ
This is the state in which all basis vectors with wave number up to the Fermi level
are occupied: This is a consequence of the fact that there can only be a single fermion
present in each mode. The expectation value of the particle number operator in the
momentum representation is given by
{
† 1 |p| ≤ kF ,
np,σ = 〈ψN |fp,σ fp,σ |ψN 〉 = . (5.13)
0 |p| > kF
Here, the precise value of the Fermi level has not yet been specied. We still have to
determine it from the particle number. But this is very easy: For |p| > kF , we have
∏ ∏ †
fp,σ |ψN 〉 = fp,σ fp′ ,σ′ |ø〉
p′ ,|p′ |<kF σ ′
fp†′ ,σ′ fp,σ |ø〉 = 0.

∏ ∏
= (5.14)
p′ ,|p′ |<k F σ′
The total particle number is

∑ ∑
N= np,σ = 2 1
p,σ |p|≤kF
kF
dp V kF3
∫
= 2V = , (5.15)
0 (2π)3 3π 2
And hence1
3π 2 N
kF3 = . (5.18)
V
1 Here, we have used that
( )3 ∫
∑ ∑ ∆ L
f (k) = f (k) = dkf (k), (5.16)
k k
(2π/L)3 2π
This is the desired expression for the Fermi level in the momentum representation. The
Fermi energy is
(~kF )2
εF = , (5.19)
2M
as the energy associated with this Fermi level.
Now what is the particle density in the position representation inside the box? As a
function of x this becomes
∑ ∑ ∑ e−ipx eip′ x
〈ψN |Ψ† (x, σ)Ψ(x, σ)|ψN 〉 = †
〈ψN |fp,σ fp′ ,σ |ψN 〉
V
σ σ p,p′
∑ ∑ e−i(p−p′ )x
†
= δp,p′ 〈ψN |fp,σ fp′ ,σ |ψN 〉
V
σ p,p′
N
= . (5.20)
V
That is no surprise: The density is homogeneous (no point is distinguished), and simply
given by the average particle density inside the container with volume V .
The elementary excitations of such a ground state are constituted by taking a par-
ticle from the Fermi ball and to place is outside the Fermi ball. Such an excited state
vector is given by
|φ〉 = fk†1 ,σ2 fk1 ,σ1 |ψN 〉. (5.21)
A particle with spin σ1 and wave number k1 is hence exchanged by one with spin σ2
and wave number k2 . A missing electron has the effect of a positively charged hole
that in itself can be treated as a fermionic quasi-particle. In fact, we can dene
†
ak,σ = f−k,−σ , (5.22)
a†k,σ = f−k,−σ , (5.23)
and ne that these hole operators again full the anti-commutation relations and can
be seen as fermionic quasi-particles.
5.3 Correlations
5.3.1 Fermionic correlation functions
The fermionic correlation function is dened as follows. We can treat it as an auxiliary
quantity for the pair correlation function that we will treat in the subsequent section,
which is the actually more interesting quantity.
where
2π 3
( )
∆= (5.17)
L
is the volume element in k-space (and of course no differential operator), up to an approximation error that
can be made arbitrarily small.
5.3. CORRELATIONS 9
Correlation function: The correlation function of a eld operator in the ground

state
Gσ (x − x′ ) = 〈ψN |Ψ† (x, σ)Ψ(x′ , σ)|ψN 〉 (5.24)
provides the probability density that the annihilation of a particle at position x′ and
the creation of a particle at x gives again the original state.
This can at the same time be seen as a transition amplitude of the unnormalized
state vector2 Ψ(x′ , σ)|ψN 〉 – the one in which a particle has been omitted at x′ – into
the (equally unnormalized) state vector Ψ(x, σ)|ψN 〉 (where a particle is missing at x).
We will now consider such as correlation function for our free Fermi gas. We nd
∑ 1 ′ ′
Gσ (x − x′ ) = 〈ψN | e−ipx+ip x fp,σ
†
fp′ ,σ |ψN 〉
V
p,p ′
∑ ′
= e−ip(x−x ) 〈ψN |fp,σ
†
fp′ ,σ |ψN 〉
p
dp −ip(x−x′ )
∫
= e Θ(kF − p)
(2π)3
∫ kF ∫ 1
1 ′
= dpp2 dηeip|x−x |η , (5.26)
(2π)2 0 −1
where we have made use of polar coordinates and have substituted the cos. The inte-
gration over η gives rise to
eipd − e−ipx
(5.27)
ipd
with
d = |x − x′ |. (5.28)
Hence, we nd
∫ kF
1
Gσ (x − x′ ) = dpp sin(pd)
2π 2 d 0
1
= (sin(kF d) − kF d cos(kF d))
2π 2 d3
3N sin(kF d) − kF d cos(kF d)
= . (5.29)
2V (kF d)3
That is to say, the correlation function oscillates with a characteristic period of 1/kF .
This may be not so much of a surprise, as this is the only scale present in momentum
space. What is more, it decays to zero for large distances. The zeros in d are given by
the solutions of tan(x) = x.
2 We will in a minute nd that
N
〈ψN |Ψ† (x, σ)Ψ(x, σ)|ψN 〉 = . (5.25)
2V
5.3.2 Fermionic pair correlation function

Due to the Pauli principle, even non-interacting Fermions of the same spin are in a way
correlated with each other3 . The Pauli principle forbibs that two fermions occupy the
same orbital. This hard constraint is inherited by an interesting effect in the position
representation. They have an effective tendency to repel each other, so that the prob-
ability of nding them in the same place is small. The Coulomb interaction (different
from this), which is also repulsive, strengthens this effect even more. For the moment
being, we will consider non-interacting fermions, and will only later consider particles
under the additional effect of the Coulomb interaction.
There are several options to meaningfully characterized such correlations. One is
by means of the pair correlation function. Let us assume we remove a particle at x
from |ψN 〉, in order to get the N − 1 particle state vector
|φ(x, σ)〉 = Ψ(x, σ)|ψN 〉. (5.30)
Its density distribution is given by
〈φ(x, σ)|Ψ† (x′ , σ)Ψ(x′ , σ)|φ(x, σ)〉 = 〈ψN |Ψ† (x, σ)Ψ† (x′ , σ ′ )Ψ(x′ , σ ′ )Ψ(x, σ)|ψN 〉
 2
N
= gσ,σ′ (x − x′ ). (5.31)
2V
This expression denes the pair correlation function. It is the probability density to
nd a particle at x and at the same time also at x’. It can be helpful to express the pair
correlation function in Fourier space. This is given by
 2
N 1 ∑ ∑ −i(k−k′ )r−i(q−q′ )r′
gσ,σ′ (x − x′ ) = e
2V V2 ′ ′
k,k q,q
† †
× 〈ψN |fk,σ fq,σ ′ fq ′ ,σ fk ′ ,σ |ψN 〉. (5.32)
Let us consider two special cases separately.

• In the rst case, we have that σ 6= σ ′ . That is to say, the spin component is dif-
ferent. Then k = k ′ and q = q ′ must be true to get a non-vanishing contribution.
We nd
 2
N 1 NN
gσ,σ′ (x − x′ ) = , (5.33)
2V V2 2 2
and hence simply
gσ,σ′ (x − x′ ) = 1, (5.34)
completely regardless of the distance. This is a consequence of the rather obvious

observation that fermionic particles with a different spin are not subject to the
Pauli principle. Hence, they take no notice of each other.
3 Even though they are not entangled, as one would say in the precise wording of quantum information
theory.
5.4. SYSTEMS WITH COULOMB INTERACTION 11
• There is a second case, however. Here, we have that σ = σ ′ . Then there are in
turn two possibilities: Either k = k ′ , q = q ′ or k = q ′ , q = k ′ . We now nd that
† † † †
〈ψN |fk,σ fq,σ fq′ ,σ fk′ ,σ |ψN 〉 = δk,k′ δq,q′ 〈ψN |fk,σ fq,σ fq,σ fk,σ |ψN 〉
† † †
+ δk,q′ δq,k′ 〈ψN |fk,σ fq,σ fk,σ fq,σ |ψN 〉
= (δk,k′ δq,q′ − δk,q′ δq,k′ )
† †
× 〈ψN |fk,σ fk,σ fq,σ fq,σ |ψN 〉.
Since (fk,σ )2 = 0 it must be true at the same time that k 6= q, and so we get
2
N2

N
gσ,σ (x − x′ ) = − (Gσ (x − x′ ))2 . (5.35)
2V 4V 2
With the help of the above correlation function – and this is where it comes into
play – we nd in terms of the normalized distance e = kF |x − x′ | the expression
3 2
gσ,σ (x − x′ ) = 1 − (sin(e) − e cos(e)) . (5.36)
e6
Interestingly, the probability to nd two fermions with the same spin nearby with dis-
tances smaller than1/kF is small. The suppression of gσ,σ on these length spaces is
called exchange hole or correlation hole. If we already nd a particle at x , then the
conditioned probability to nd another particle at the same location is reduced.
This effect is a consequence of the fermionic anticommutation relations and no
effect resulting from a genuine interaction – even if the net effect is similar. Again,
fermions with the same spin act as if they intended to repel each other, even in the
absence of an interaction, but with the same effect. For bosons, this is different: They
tend to “bunch”. In the lecture, I have explained how this can be measured with thermal
light. In fact, the effectively repulsive effect for fermions plays a key role in physics:
Surely in the understanding of properties of materials – in fact even when showing
the stability of matter. But is also determines properties of stars, to name just two
examples.
5.4 Systems with Coulomb interaction
We now turn to describing fermionic systems exhibiting a Coulomb interaction. The

Hamiltonian of a system of fermions interacting with a Coulomb interaction is in sec-
ond quantization given by an expression of the following form:
Hamiltonian of electrons with Coulomb iteraction:

∑ ~2 k 2 †
H = fk,σ fk,σ
2M
k,σ
e2 4π †
f †′
∑
+ f fk′ ,σ′ fk,σ . (5.37)
2V q 2 k+q,σ k −q,σ
k,k′ ,q,σ,σ ′ ,q6=0
The rst term is the known kinetic term. The second one captures the interaction.
We have left out the term corresponding to q = 0. Why have we done that? To
start with, it would surely diverge. But we also do not have to take it into account,
making use of an argument of the following sort: This part is being compensated by
the positively charged background, e.g., by the positive ions in a solid state system. We
will now set out to derive this Hamiltonian from a microscopic model. Its Hamiltonian
is given by
H = HEl + HIon + HI , (5.38)
capturing the dynamics of the electrons, the positively charged background, and the
interactions. We will see that because of the long-ranged character of the Coulomb
interaction the individual terms will diverge in the limit of large systems: The sum,
however, is perfectly well-dened and non-divergent. In order to deal with these diver-
gences, it can be helpful to dene a cutoff µ > 0 that should not worry us too much.
We will in the last step investigate the limit of a small cutoff. In this way, we can more
easily discuss the terms individually.
The Hamiltonian of the positively charged ions is given by
1 2 n(x)n(x′ ) −µ|x−x′ |
∫
HIon = e dxdx′ e . (5.39)
2 |x − x′ |
Here, n(x) = N/V simply is constant density. µ > 0 is the mentioned cutoff. We
hence have
 2 ∫ ∞
1 N
HIon = e2 V 4π dsse−µs , (5.40)
2 V 0
obviously simply a number. The interactions of the electrons with the positively charged
background is
N
N e−µ(x−xj )
∑ ∫
HI = −e2 dx , (5.41)
j=1
V |x − xj |
where {xj : j = 1, . . . , N } are the positions of the electrons. These are in principe
one-body operators. But making use of translational invariance, we again nd that this
again gives rise to a number,
N 2 4π
HI = −e2 . (5.42)
V µ2
Finally, the actual Hamiltonian of interest of the electrons is

N N
∑ pj e2 ∑ e−µ|xj −xk |
HEl = + . (5.43)
j=1
2M 2 |xj − xk |
j,k=1,j6=k
The thermodynamic limit is the limit of N → ∞ and V → ∞, where N/V is held

constant. We begin with the terms of ions
1 2 N 2 4π
HIon = e . (5.44)
2 V µ2
In the limit µ → 0 this term diverges, which again only manifests that in the ther-
modynamic limit because of the long-ranged nature of the interaction every ion will
eventually interact with any other. Note that the interaction between ions and electrons
is given by Gl. (5.42). The Hamiltonian hence is
e2 2 −1
H=− N V 4πµ−2 + HEl , (5.45)
2
and we can again turn to the electronic contribution, the one we are actually interested
in. We will now express it in second quantization. The kinetic term is as usual
∑ ~2 k 2 †
T = fk,σ fk,σ . (5.46)
2M
k,σ
To identify the interaction term in second quantization is obviously somewhat more

tedious. The coordinates are ξ = (p, σ), with single-body wave functions
1
ψp,σ (x, σ ′ ) = √ eipr/~ δσ,σ′ . (5.47)
V
Let us recap what we need to compute the interaction Hamiltonian in second quantiza-
tion. These are, for the interaction contribution V2 of the Hamiltonian in rst quantiza-
tion, terms of the form
e2
∫
〈ψk1 ,σ1 , ψk2 ,σ2 |V2 |ψk3 ,σ3 , ψk4 ,σ4 〉 = dx1 dx2 e−ik1 x1 δσ1 ,σ3 e−ik2 x2 δσ2 ,σ4
V2
e−µ|x1 −x2 | ik3 x1 ik4 x2
× e e . (5.48)
|x1 − x2 |
Turning to new position coordinates of the absolute position x = x2 and the relative
distance y = x1 − x2 we get
e2
∫
〈ψk1 ,σ1 , ψk2 ,σ2 |V2 |ψk3 ,σ3 , ψk4 ,σ4 〉 = dxe−i(k1 +k2 −k3 −k4 )x
V2
e−µy
∫
× dyei(k3 −k1 )y δσ1 ,σ3 δσ2 ,σ4
y
e2 4π
= δσ ,σ δσ ,σ δk +k ,k +k .
V 1 3 2 4 1 2 3 4 (k1 − k3 )2 + µ2
(5.49)
The Kronecker deltas result from the orthogonality of the spin wave functions. There-
fore, the Hamiltonian becomes
1 e2 N 2 4π ∑ ~2 k 2 †
H = − + f fk,σ
2 V µ2 2M k,σ
k,σ
e2 ∑ ∑ ∑ ∑
+ δσ1 ,σ3 δσ2 ,σ4 δk1 +k2 ,k3 +k4
2V
k1 ,σ1 k2 ,σ2 k3 ,σ3 k4 ,σ4
4π
× f † f † fk ,σ fk ,σ . (5.50)
(k1 − k3 )2 + µ2 k1 ,σ1 k2 ,σ2 4 4 3 3
We will now see how we can get gid of µ in the Hamiltonian, making use of the charge
neutrality. It makes a lot of sense to again change the coordinates, after all, we are
perfectly free to make use of any coordinates we would like to use,
k1 = k + q, (5.51)
k2 = p − q, (5.52)
k3 = k, (5.53)
k4 = p. (5.54)
Now always k1 + k2 = k3 + k4 , and ~(k1 − k3 ) = ~q is the momentum that is being
transferred in a two-body interaction. In these coordinates, the last term of (5.50)
becomes
e2 ∑ ∑ 4π
f† f† fp,σ2 fk,σ1 , (5.55)
2V σ ,σ
q 2 + µ2 k+q,σ1 p−q,σ2
k,p,q 1 2
again using Kronecker deltas. It makes a lot of sense to chop this expression in parts,
specically in contributions with q 6= 0 and those with q = 0, that is,
′
e2 ∑ ∑ 4π
f† f† f f
2 + µ2 k+q,σ1 p−q,σ2 p,σ2 k,σ1
2V σ ,σ
q
k,p,q 1 2
∑ ∑ 4π †
+ f f † fp,σ2 fk,σ1 . (5.56)
σ ,σ
µ2 k,σ1 p,σ2
k,p 1 2
′
Here refers to the sum in which the term q = 0 has been omitted. The second term,
that is, the one belonging to q = 0, becomes
e2 4π ∑ †
fk,σ f † fp,σ2 fk,σ1
1 p,σ2
2V µ2
k,p,σ1 ,σ2
e2 4π ∑ 
†  † 
= fk,σ fk,σ1 fp,σ f
2 p,σ2
− δk,p δσ1 ,σ2
2V µ2 1
k,p,σ1 ,σ2
e2 4π ∑
= nk,σ1 (np,σ2 − δk,p δσ1 ,σ2 )
2V µ2
k,p,σ1 ,σ2
e2 4π  2 
= N −N , (5.57)
2V µ2
where we have used that we have exactly N particles: We can therefore replace the
number operators by the actual particle number, the respective eigenvalue of the particle
number. They become numbers, the leading terms of which, linear in N 2 , precisely
cancel each other. The term
e2 4π
− N (5.58)
2V µ2
That is to say, if we take the limits in the order of rst taking N, V → ∞ and then
µ → 0, then the we get exacty the above Hamiltonian of electrons with Coulomb
interaction. We are done.
5.4.1 Perturbation theory and stability of Fermi gases

The trouble is, there is no exact solution of this Hamiltonian: It is not a quadratic
polynomial in the fermionic operators and hence no “non-interacting Hamiltonian”.
This is no surprise, as it is set up to capture the Coulomb interaction. Let us remind
ourselves that the number of problems in quantum physics that have an exact solution is
pretty small: It consists of free fermions and bosons, some stabilizer and spin models,
and highly structured problems such as Bethe-solvable models. What we can make
use of, needless to say, are ideas of perturbation theory. We will now compute the rst
order correction of the energy density. We start from the ground state vector of free
Fermi gases,
(k )(k )
F F
† †
∏ ∏ ∏ ∏
†
|ψN 〉 = fp,σ |ø〉 = fp,↑ fp,↓ |ø〉 (5.59)
p≤kF σ p=0 p=0
We consider the kinetic terms as the leading term and treat the interaction as a small
perturbation, one that we treat in perturbation theory. The kinetic term already is diag-
onal, and hence we get
∑ ~2 k 2 †
E (0) = 〈ψN | fk,σ fk,σ |ψN 〉
2M
k,σ
~2 ∑
= θ(kF − k)
2M
k,σ
~2 V
∫
= 2 3
dkk 2 θ(kF − k)
2M (2π)
~2 V 1
= 4π k 5
M (2π)3 5 F
3~2 kF2
= N
10M
3
= εF N
5
 2/3
e2 1 3 9π
= 2
N. (5.60)
2a0 rs 5 4
So the zeroth order contribution to the energy that only includes the kinetic term be-
comes
e2 2.21
E (0) = N. (5.61)
2a0 rs2
Here we have made use of the fact that
N k3 3 3
= F2 = = . (5.62)
V 3π 4πr03 4πa30 rs3
Here, r0 is the radius of a ball with a volume that corresponds to the volume per particle.
This is not an uncommon type of argument. One treats the systems as if they were balls
lling a certain volume. The number
~2
a0 = (5.63)
M e2
is the Bohr-radius, and
a0
rs =
. (5.64)
r0
This number captures the density of the Fermi gas.
We can now turn to discussing the contribution of the interaction. In rst order of
perturbation theory the change in energy is just the expectation value of the perturbingg
Hamiltonian. We hence get
′
e2 4π †
fk†′ −q,σ′ fk′ ,σ′ fk,σ |ψN 〉.
∑
E (1) = 〈ψN |fk+q,σ (5.65)
2V q2
k,k′ ,q,σ,σ ′
′
The sum again means that the term with q = 0 has been omitted. The only term for
which each annihilation operator is compensated by a creation operator (as only then
we get a non-vanishing contribution), is linear in
† †
δσ,σ′ δk,k′ +q fk+q,σ fk,σ ′ fk+q,σ ′ fk,σ . (5.66)
This energy contribution hence becomes

′
e2 ∑ 4π
E (1) = − nk+q,σ nk,σ
2V q2
k,q,σ
′
e2 ∑ ∑ 4π
= − θ( kF − |q + k|)θ(fK − k)
2V σ q2
k,q
3πe2 V 1
∫ ∫
= − 6
dkθ(kF − k) dk ′ θ(kF − k ′ ). (5.67)
(2π) |k − k ′ |
This is an interesting expression. We compute here the overlap of two Fermi balls with
relative distance q. Now we have
4πe2 2e2
 
1 k
∫
− 3
dk ′ ′ 2
θ(kF − k) = − kF F , (5.68)
(2π) |k − k | π kF
where the function F : R+ → R+ is dened as

1 1 − y2
∣ ∣
∣1 + y ∣
F (y) = + log ∣∣ ∣. (5.69)
2 4y 1 − y∣
This means that the energy contribution becomes
e2 kF V kF2 − k 2
 ∣ ∣
dk ∣ kF + f ∣
∫
E (1) = − 3
1 + log ∣ ∣
k<kF (2π) 2kkF
π ∣ kF − f ∣
3 e2 kF
= −N
4 π
 1/3
e2 9π 3N
= −
2a0 rs 4 2π
e2 0.916
= − N (5.70)
2a0 rs
Thus, the rst two terms are in the perturbation theory for the energy density given by
e2 2.21 0.916
 
E
= 2
− + O (1) . (5.71)
V 2a0 rs rs
The rst term is that of the kinetic energy, the second the exchange term. This is an
important result in that it shows that the contribution of the potential energy becomes a
small perturbation when rs → 0, so in the limit of high density. The leading term in the
interaction energy of electron gases of high density can indeed be described in pertur-
bation theory. This is remarkable in the light of the facts that the interaction potential is
neither small (remember that the particles are close and hence the Coulomb interaction
strongest) nor short-ranged (but actually innitely-ranged). But still, in this regime the
interaction is found to be a small perturbation of the kinetic energy. The second term
that captures the exchange energy is negative. The microscopic Hamiltonian has con-
tributions of direct interaction and exchange terms. The direct term corresponds to the
term with q = 0, and cancels with
HIon + HI . (5.72)
All that remains, therefore, is the negative exchange energy. The system is stable. We
should not forget, however, that this stability and the cancellation of the q = 0 term is
a consequence of the presence of a positive background. If one would put negatively
charged electrons in a container, the system would be all but stable.
These expressions provide roughly the correct energy densities for the electron gas
of metals. Most importantly, it captures the right behaviour, in that the energy density
takes a minimum for some value of rs > 0 and one hence arrives at a stable bound
system.
4
4 Dernächste Term divergiert aber logarithmisch: Wir erhalten

E e2 
a + brs + crs2 log rs + drs2 + o(rs2 ) ,

= (5.73)
N a0 rs2
wobei a, b, c, d ∈ R Konstanten sind. Die genaue Berechung von c und d ist nicht einfach, aber in nichtnu-
merischer Weise mit elementaren Methoden möglich.
Another limit that we would like to briey discuss is the limit rs → ∞. This is
a limit that Wigner discussed rst [?]. One can show that in principle, even smaller
energy densities are possible if electrons are arranged in a Wigner crystal. One nds
for large rs the expression
e2
 
E 1.79 2.64
= − + 3/2 + O(rs5/2 ) , (5.74)
N 2a0 rs rs
an expression that provides a good approximation for rs  10. It is indeed true that
the Wigner crystal has a lower energy density as the uid. Then the kinetic energy
becomes negligible over the electrostatic energy of classical charged particles. Ions in
a Penning trap show a very similar behaviour.
5.4.2 Impact of the Coulomb interaction on electron levels

As a nal remark we will investigate the impact of the Coulomb interaction on the
energy levels
(~k)2
ε0 (k) = (5.75)
2M
of the electrons. To recap, the Hamiltonian of an electron gas with interaction reads
∑ ~2 k 2 †
H = fk,σ fk,σ
2M
k,σ
e2 4π †
f †′
∑
+ f fk′ ,σ′ fp,σ . (5.76)
2V q 2 p+q,σ k −q,σ
p,k′ ,q,σ,σ ′ ,q6=0
We will now ask the question how the impact of the interaction could possibly be.
We will do this in an unorthodox fashion, yet a simple one. We will consider a time
evolution and see what the inuence of interaction is. We will resort to an approximate
treatment of the time evolution of fermionic operators. Again, this may look a bit
weird, but quickly leads to a result.
We consider the evolution of fermionic operators in the Heisenberg picture, that is
 
∂ i ∑ ′ †
fk,σ (t) = ε0 (k )fk′ ,σ′ fk ,σ , fk,σ 
′ ′
∂t ~ ′ ′
k ,σ
i ∑
= − ε0 (k ′ ){fk†′ ,σ′ , fk,σ }fk†′ ,σ′ , (5.77)
~ ′ ′
k ,σ
which immediately leads to

∂ i
fk,σ (t) = − ε0 (k)fk,σ (t). (5.78)
∂t ~
We dene the time dependent correlation function Gk,σ as
†
Gk,σ (t) = 〈ψN |fk,σ (t)fk,σ |ψN 〉. (5.79)
†
The multiplication with fk,σ (t) delivers the evolution equation for Gk,σ as
∂ i
Gk,σ (t) = − ε0 (k)Gk,σ (t). (5.80)
∂t ~
It solution is
Gk,σ (t) = e−iε0 (k)t/~ (−nk,σ + 1), (5.81)
since
†
〈ψN |fk,σ (0)fk,σ (0)|ψN 〉 = −nk,σ + 1. (5.82)
The is the result without taking the Coulomb interaction into account. Doing so gives
rise to
 
∂ i 1 ∑ 4πe2 †
fk,σ (t) = ε0 (k)fk,σ − f f f
′ k+q,σ p,σ ′  . (5.83)
∂t ~ V q 2 p+q,σ
′ p,q6=0,σ
This expression results in the equation of motion of the correlation function as

∂ i
Gk,σ (t) = − ε0 (k)Gk,σ (t) (5.84)
∂t ~
2
1 ∑ 4πe † †

− 2
〈ψN |fp+q,σ ′ (t)fk+q,σ (t)fp,σ ′ (t)fk,σ (0)|ψN 〉 .
V q
p,q6=0,σ
′
An annoying observation is that on the right hand side, we no longer nd the corre-
laton function, but also terms that contain the correlation functions in higher orders.
We make use of the assumption that these expressions can be approximately written as
products of lower correlators. Such truncations of innite hierarchies are not uncom-
mon in physics (in fact, to mention that is the secret reason for including this chapter).
One basically makes a Gaussian assumption, referring to the feature that moments fac-
tor as if one had encountered a Gaussian distribution. This is good enough for our
purposes, but not exact (and a rigorously minded person will ask for bounds to the
errors made, which can be provided). We will here be not too pedantic, however. We
will therefore assume that
† †
〈ψN |fp+q,σ ′ (t)fk+q,σ (t)fp,σ ′ fk,σ (0)|ψN 〉
† †
≈ 〈ψN |fp+q,σ ′ (t)fk+q,σ (t)|ψN 〉〈ψN |fp,σ ′ (t)fk,σ (0)|ψN 〉
† †
= δσ,σ′ δp,k 〈ψN |fp+q,σ ′ (t)fp+q,σ (t)|ψN 〉〈ψN |fk,σ ′ (t)fk,σ (0)|ψN 〉 (5.85)
Under this assumption – and we will now remove the ≈ symbols and will make use
of equality signs, knowing that this is an approximation – we get the close equation of
motion
∂ i
Gk,σ (t) = − ε0 (k)Gk,σ (t) (5.86)
∂t ~
1 ∑ 4πe2 
− nk+q,σ Gk,σ (t).
V q2
q6=0
We can read off the following expression
(~k)2 1 ∑ 4πe2
ε(k) = − nk′ ,σ . (5.87)
2M V ′ |k − k ′ |
k
The change in ε is therefore
dk 4πe2
∫
∆ε(k) = − θ(kF − k ′ )
(2π)3 |k − k ′ |
e 2 kF
∫ 1
1
∫
= − dss2 dt 2
π 0 −1 k + s2 − 2kst
e 2 kF
∣ ∣
∣k + s∣
∫
= − dss log ∣∣ ∣
πk 0 k − s∣
2e2 kF
 
k
= − F , (5.88)
π kF
where F is exactly the above function that we had encountered in (5.69). The energy
levels are hence closer compared to the free Fermi gas.
Chapter 6
Bosonic systems and

superuidity
6.1 Densities and correlations
6.1.1 Density distribution for free bosons
Before coming to these exciting insights, we will, however, start from pretty dry basics.
We consider a system of N non-interacting bosons in the state vector
|ψN 〉 = |Np0 , Np1 , . . . 〉. (6.1)
The particle number density as a function of position x is
1 ∑ −i(kx+k′ x)
〈ψN |Ψ† (x)Ψ(x)|ψN 〉 = e 〈ψN |b†k bk |ψN 〉
V ′ k,k
N
= . (6.2)
V
This expression is manifestly independent of position, but this is no surprise in a system

that is translationally invariant.
6.1.2 Pair distribution function for non-interacting bosons
We can, in analogy to the fermionic case, again dene a pair distribution function, now
in the bosonic reading, without a spin. It is dened as follows.
5
6 CHAPTER 6. BOSONIC SYSTEMS AND SUPERFLUIDITY
Bosonic pair distribution function: For spinless bosons, it is dened as
N2
g(x − x′ ) = 〈ψN |Ψ† (x)Ψ† (x)Ψ(x′ )Ψ(x′ )|ψN 〉. (6.3)
V2
We also have
N2 1 ∑ ′ ′ ′ ′
g(x − x′ ) = e−ikx−iqx +iq x +ik x
V2 V2
k,k′ ,q,q ′
× 〈ψN |b†k b†q bq′ bk′ |ψN 〉. (6.4)

The expectation value gives a value different from zero only if k = k ′ and q = q ′ , or if
k = q ′ and q = k ′ . The case of k = q that has been impossible for fermions but that is
now possible, has to be considered separately. Hence, we have
〈ψN |b†k b†q bq′ bk′ |ψN 〉 = (1 − δk,q )
 
× δk,k′ δq,q′ 〈ψN |b†k b†q bq bk |ψN 〉 + δk,q δq,k′ 〈ψN |b†k b†q bk bq |ψN 〉
+ δk,q δk,k′ δq,q′ 〈ψN |b†k b†k bk bk |ψN 〉

= (1 − δk,q )
× (δk,k′ δq,q′ + δk,q′ δq,k′ ) nk nq + δk,q δk,k′ δq,q′ nk (nk − 1). (6.5)
We therefore arrive at the somewhat complicated looking expression
〈ψN |Ψ† (x)Ψ† (x′ )Ψ(x′ )Ψ(x)|ψN 〉
 
1 ∑ −i(k−q)(x−x′ )
∑
= (1 − δk,q )(1 + e )nk nq + nk (nk − 1)
V2
k,q k
( ∣ ∣2
1 ∑ ∑
2
∣∑
−ik(x−x′ )
∣ ∑
= nk nq − nk + ∣ e nk ∣ − n2k
∣ ∣
V2 ∣ ∣
k,q k k k
)
∑ ∑
2
+ nk − nk
k k
∣ ∣2
N2 1 ∣∑
−ik(x−x′ )
∣ 1 ∑
= + e nk ∣ − 2 nk (nk + 1). (6.6)
∣ ∣
V2 V V
∣
∣ ∣
k k
There is a positive second term, reminiscent of the fermionic situation. For the last
term, however, there is no fermionic equivalent, simply because one cannot have a
double occupation for fermions.
Let us now look at a couple of examples. If all bosons are taking the same momen-
tum state p0 , then (6.6) becomes
N2 N (N − 1)
g(x − x′ ) = . (6.7)
V2 V2
6.2. PHOTON CORRELATIONS 7
Now the pair distribution function is also constant. There are simply no correlations
The right hand side can be interpreted in a way that the probability to nd the rst
particle N/V ist, that to nd the second (N − 1)/V and so on.
The situation changes signicantly if particles are distributed over momentum val-
ues, say, following a Gaussian distribution
(2π)3 N −(k−k0 )2 /∆2

nk = √ e (6.8)
V ( π∆)3 )
with normalization
dp N
∫
nk = . (6.9)
(2π)3 V
Then the second term in the above expression becomes
dk −ik(x−x′ ) N
∫
2 ′ 2 ′
e nk = e−∆ (x−x ) /4 e−ik0 (x−x ) , (6.10)
(2π)3 V
up to a small error originating from the discrete integration, and for the third term
2 ∫
(2π)3 N

1 dk 3 1 dk −2(k−k0 )2 /∆2
∫
n = √ e
V (2π)3 k V V ( π∆)3 (2π)3
N2
≈ . (6.11)
V 3 ∆3
Holding the density N/V and the width of the distribution ∆ xed, then the third term
vanishes in the limit of large volumes V in (6.6) as 1/V . The pair distribution function
then becomes
N2 N2  2 ′

g(x − x′ ) = 2 1 + e−∆ (x−x )/2 . (6.12)
V2 V
The probability to nd bosons nearby, closer than a distance ∆−1 , is increased. This
is an interesting phenomenon. Because of the symmetry alone, but not due to genuine
interactions, bosons tend to stick together, to cluster or to “bunch”. The probability
to nd two bosons at exactly the same position is precisely twice as large as for large
distances. This is no contradiction to what has been said before. The density can
very well be constant, and at the same time the pair correlation functions indicates a
clustering.
6.2 Photon correlations

Photons constitute the prototypical example of non-interacting bosons. They are indeed
non-interacting to an extraordinarily good approximation. Making use of methods of
quantum optics, one can prepare sophisticated states of light modes. One can also mea-
sure photon correlations in Hanbury-Brown Twiss experiments, as we briey discuss
in the lecture, but not in this script.
6.3 Weakly interacting dilute Bose gases

6.3.1 Quantum uids and Bose Einstein condensation
Quantum degenerate Bose gases are physical systems that are still enjoying enormous
attention in present-day research. An important bosonic uid is He4 , exhibiting S = 0,
a kind of Bose gas that features little interaction compared to other Bose gases made
from heavier atoms. He4 is uid down to T = 0, turning to a superud state at T =
2.18K we come to later. The interaction of such Helium atoms is indeed small, but
non-negligible. It can be well described by a Lennard-Jones potential,
  
σ 12  σ 6
V (r) = 4ε − , (6.13)
r r
where ε = 1.411 × 10−15 erg und σ = 2.556 × 10−10 m. It captures a strong repulsion
at short distances and a mild attraction at large distances.
In relatively recent years, Bose Einstein condensation has moved into the focus
of signicant attention both in experimental as well as in theoretical work, starting
with the successful demonstration of Bose-Einstein-condensation of about 2000 87 Rb
atoms, followed by an experiment with 100000 7 Li atoms and several millions 23 Na
atoms. Today, the eld of ultra-cold atomic quantum gases is one of the fastest growing
research elds, giving rise to a number of exciting applications. We will later turn to
a specic one, that of ultra-cold atoms in optical lattices, giving rise to instances of
highly controlled quantum simulators. Before that, we will discuss in great detail the
situation without the presence of an optical lattice, in form of the Bogoliubov-theory
of weakly interacting Bose gases.
6.3.2 Bogoliubov theory of weakly interacting Bose gases

In the momentum representation we can write the Hamiltonian of interacting bosons as
∑ k2 † 1 ∑
H= b k bk + Vq b†k+q b†p−q bp bk . (6.14)
2M 2V
k k,p,q
In the following, we will make use of a series of approximations that are well justied
for dilute, weakly interacting Bose gases and give rise to a good model. To start with,
Vq is the Fourier transform of the interaction in the position representation
∫
Vq = dxe−iqx V (x). (6.15)
For low temperatures, Bose Einstein condensation takes place into the lowest mode
corresponding to k = 0 statt. Indeed, at zero temperature and in the absence of in-
teractions, so in case of the ideal Bose gas, this is the obvious ground state: Then the
mode corresponding to k = 0 is simply occupied N times. The bosons “condense”
in the same mode, which is perfectly possible for bosons, in contrast to the fermionic
situation.
6.3. WEAKLY INTERACTING DILUTE BOSE GASES 9
In analogy to the ideal Bose gas one should also expect that even for small interac-
tions, this mode is macroscopically occupied. By this we mean that
N0 = 〈ψN |a†0 a0 |ψN 〉 ≈ N (6.16)
us expected to hold true. The number of particles outside this mode, so outside the
condensate, is hence given by
N − N0  N0 ∼ N. (6.17)
This leads us to the rst approximation step: One can neglect all interactions outside
the k = 0-mode, simply as densities are too small. We can hence discuss interactions
exclusively of the k = 0 mode with those outside of it. Under these – indeed very mild
– assumptions, we can write the Hamiltonian as
∑ k2 † ′
1 1 ∑
H = bk bk + V0 b†0 b†0 b0 b0 + (V0 + Vk )b†0 b0 b†k bk
2M 2V V
k k
′
1 ∑  † † 
+ Vk bk b−k b0 b0 + b†0 b†0 bk b−k , (6.18)
2V
k
in a form
• in which we have neglected polynomials higher than third degree in bk .
Here V0 again
refers to the Fourier transform of the interaction (and not the volume).
′
The symbol k again referes to a sum that takes out the term k = 0. Of course, we
have that
√
b0 |N0 , . . . 〉 = N0 |N0 − 1, . . . 〉, (6.19)
b†0 |N0 , . . . 〉 = N0 + 1|N0 − 1, . . . 〉.
√
(6.20)
Now the following approximations are plausible. To highlight them, they are all em-
phasized in an itemized environment in this text.
• Since Since N0 is a large number, N0 ∼√1023 , both terms largely correspond to
the multiplication with the real number N0 .
• What is more, in comparison to N0 , the impact of the commutator [b0 , b†0 ] = 1 is
negligible, and hence the operators
b0 = b†0 = N0 ∈ R
√
(6.21)
can be replaced by real numbers.
This last step, needless to say, only makes sense for the mode k = 0. In this approxi-
mation, the Hamiltonian becomes
′
∑ k2 † 1 2
H = b bk + N V0
2M k 2V 0
k
′  
N0 ∑ 1
+ (V0 + Vk )b†k bk + Vk (b†k b†−k + bk b−k ) . (6.22)
V 2
k
Interestingly, we do not know the precise value of N0 ; we rather made some assump-
tions about it. It is determined by the N/V and implicitly by the interaction. We will
now write total particle number operator of the entire system as
′
∑
N0 + b†k bk , (6.23)
k
where were know that

′
∑
〈ψN |N0 + b†k bk |ψN 〉 = N. (6.24)
k
Hence, (6.23) can always be replaced by N . That is to say, by expanding this expres-
sion, we get terms of the form
′
V0 2 V0 2 N V0 ∑ †
N = N − bk bk
2V 0 2V V
k
′
V0 ∑ † †
+ bk bk bk ′ bk ′ . (6.25)
2V ′k,k
In this way, the Hamiltonian becomes

′ ′
∑ k2 † N∑
H = b k bk + Vk b†k bk
2M V
k k
N2
+ V0
2V
′
N ∑  
+ Vk b†k b†−k + bk b−k + H ′ , (6.26)
2V
k
where
• H ′ is again a Hamiltonian with more than four creation or annihilation operators.

Such terms are, however, linear in (n′ )2 , where
n′ = (N − N0 )/V (6.27)
is the density of those particles not contained in the condensate.
The Bogoliubov theory consists of this scheme of approximations. Let us briey review
this approximation scheme: It was key to the idea to single out the mode corresponding
to k = 0 and to replace the respective operators by numbers (which makes perfect
sense for large occupation numbers). Then we have neglected higher contributions to
interaction terms, which is a good approximation for small densities. Indeed, when
n′  N/V (6.28)
this is a good approximation, and in fact a very good approximation for weakly inter-
acting, highly dilute Bose gases. Since we have neglected H ′ , we have
′ ′
∑ k2 † N∑
H = bk bk + Vk b†k bk
2M V
k k
N2
+ V0
2V
′
N ∑  † † 
+ Vk bk b−k + bk b−k . (6.29)
2V
k
This is again a quadratic expression in bosonic operators in the Hamiltonian. Such

systems are called free systems, or non-interacting systems, or quasi-free systems, the
latter in particular in the mathematical literature. Such systems for which the poly-
nomial is a polynomial of second order in bosonic or fermionic operators are exactly
solvable. Indeed, they constitute some of the few solvable systems in physics. The
mindset here is to understand in what way one can make approximation schemes, until
one arrives at such a quadratic Hamiltonian problem. We will solve this model rst,
and later have a more systematic look at such problems.
Let us now make use of the transformation
bk = uk ak + vk a†−k , (6.30)
b†k = uk a†k + vk a−k , (6.31)
where uk , vk ∈ R, from one set of valid bosonic annihilation operators {bk } to a new
one {ak }. This is only a legitimate transformation to such a new set if the new operators
again satisfy
[ak , ak′ ] = 0, (6.32)

[a†k , a†k′ ] = 0, (6.33)
[ak , a†k′ ] = δk,k′ , (6.34)
which is nothing but1

u2k − vk2 = 1 (6.37)
for all k. The inverse transformation is then given by
ak = uk bk − vk b†−k (6.38)
as one easily nds. Such a transformation is a symplectic transformation. Note that it is

not a linear transformation of creation operators, but a mixed transformation that brings
1 The proof is rather obvious, as
[ak , ak′ ] = uk vk′ δk,−k′ + vk uk′ (−δk,−k′ ) = 0, (6.35)
and similarly
[ak , a†k′ ] = uk uk′ δk,k′ + vk vk′ (−δk,k′ ) = (u2k − vk2 )δk,k′ . (6.36)
together annihilation and creation operators. We will later identify the transformation
implemented here as a symplectic transformation.
Making use of a transformation that transforms the previous particles into “quasi-
particles”, we hence have
b†k bk = u2k a†k ak + vk2 a−k a†−k + uk vk (a†k a†−k + ak a−k ), (6.39)
† †
bk b−k = u2k a†k a†−k + vk2 ak a−k + uk vk (a†k ak + a−k a†−k ), (6.40)
bk b−k = u2k ak a−k + vk2 a†k a†−k + uk vk (a†−k a−k + ak a†k ). (6.41)
In these coordinates, the Hamiltonian is given by
1 2
H = N V0 (6.42)
2V
′
∑  k2 
N 
+ + Vk u2k a†k ak + vk2 ak a†k + uk vk (a†k a†−k + ak a−k )
2M V
k
′
N ∑  2   
+ Vk (uk + vk2 ) a†k a†−k + ak a−k + 2uk vk (a†k ak + ak a†k ) .
2V
k
We have not yet made of all the freedom we have, however: We still have the freedom
to let the non-diagonal terms vanish, without approximation. This means that (i) for all
k  2 
k N N
+ Vk u k v k + Vk (u2k + vk2 ) = 0 (6.43)
2M V 2V
should hold. Together with (ii) u2k − vk2 = 1 for all k from the preservation of the
commutation relations, we hence have a system of equations in {uk } und {vk }. This
allows us to identify the coefcients {uk } and {vk }. It is helpful to introduce the
following quantities talking – maybe unsurprisingly – the role of frequencies,
( 2 )1/2
k2 N
ωk = + Vk − (N Vk /V )2
2M V
( 2 )1/2
k2 N k 2 Vk
= + . (6.44)
2M VM
In this way, we nd the solution for the coefcients {uk } and {vk } as
 2 
k
ωk + 2M +NV Vk
2
uk = , (6.45)
2ω
 k2 
k
−ωk + 2M +N V Vk
2
vk = , (6.46)
2ωk
N Vk
uk v k = − . (6.47)
2V ωk
This nally gives for the Hamiltonian the following expression. Before spelling it out,
let us remind outselves that the last steps have been nothing but a way to decouple a
quadratic systems by means of a clever choice of coordinates. Such an approach is
always possible for quadratic systems, as we will see later. We therefore arrive at the
following simple Hamiltonian.
Hamiltonian of the weakly interacting Bose gas in the Bogoliubov approxima-

tion:
′  ′
N2 1 ∑ k2
 ∑
N
H = V0 − + Vk − ωk + ωk a†k ak . (6.48)
2V 2 2M V
k k
We can easily interpret this Hamiltonian. The rst term is a number, the ground
state energy. The second term is a collection of harmonic oscillators of different
frequency, reecting excitations. These excitations, generated by a†k , are also called
“quasi-particles”, in fact quasi-particles with wave number k. These creation operators
correspond to both creation and annihilation operators in the original coordinates, it is
important to emphasize.
The ground state of the systems with N particles is the vacuum with state vector
|ø〉, in these coordinates. Then we have
ak |ø〉 = 0 (6.49)
for all k. In this ground states, that is, at zero temperature in a language of statistical
physics, we can now determine the number of particles (not quasi-particles), referring
to the original bosonic operators {bk }. This number is given by
′
∑ ′
∑
N′ = 〈ø| b†k bk |ø〉 = 〈ø| vk2 ak a†k |ø〉
k k
′
∑
= vk2 , (6.50)
k
where we have made use of the connection between quasi-particles and particles. That
means: Without any interaction, all particles are condensed. When taking interactions
into account, it is still true that no quasi-particles are excited in the ground state. But
this now means that some original particles are outside the condensate.
We would now like to make another assumption, that is,
• about the specic interaction of the particles:
We make the plausible assumptions that bosons are weakly interaction via a contact
potential such as hard balls. This means that the potential is given by
V (x) = λδ(x). (6.51)

Making use of (6.44) as well as (6.45) and (6.47), we nd for the number density of
the particles outside the condensates2
N′ M 3/2
n′ = = (λN/V )3/2 . (6.52)
V 3π 2
The number of particles in the condensate is then
N0 = N − N ′ = N − n′ V. (6.53)
This is no longer N , but reduced due to the interaction. Equipped with these insights,
we can also revisit the ground state energy: In (6.48) it consists of a term that would
be the interaction energy, were all particles contained in the condensate, and another
negative term. Due to the occupation of bosonic states with k 6= 0 in the ground state
the kinetic energy is increased, while the potential energy is reduced.
6.3.3 Dispersion relation in the Bogoliubov theory

As mentioned before, the elementary excited states are those for which one applies
a†k to |ø〉: This gives quasi-particles with wave number k. Their energy is, as being
determined by the Hamiltonian, simply ωk . For small k we nd that ωk is according to
(6.44) essentially linear in k, and one gets
ωk ≈ ck, (6.54)
where  1/2
N V0
c= . (6.55)
VM
Excitations with long wave length are hence those with linear dispersion. c can be
seen as the speed of sound that determines the velocity with which excitations travel,
a quantity that can be related to the actual information propagation speed. For large k
one nds according to (6.44)
k2 N
ωk ≈ + Vk , (6.56)
2M V
a quadratic expression. This is the dispersion relation of free particles, the energy of
which is shifted by an average potential of N Vk /V . We will now assume – stretching
the above assumptions to an extent – that the interaction is sufciently strong so that
we have a local minimum of ωk as a function of k that is different from k = 0: The
function increases linearly, the decreases again, only to grow subsequently. Such a
behaviour is explained by the above dispersion relation.3
2 Here we encounter the subtlety that the density is not analytic in λ and simple perturbation theory does
not work.
3 Let us not be too pedantic here and be too worried about whether for such strong interactions the Bo-
goliubov theory may no longer deliver good approximations. More exact computations indeed conrm this
feature of the dispersion relation that we will accept as the proper dispersion relation of the type of system
at hand.
6.3.4 Superuidity
We have just seen that the dispersion relation of the quasi-particles of the weakly inter-
acting dilute Bose gas rst grows linearly and for large k grows quadratically. What is
more, we nd a local minimum of ωk as a function of k, different from zero. We will
now see that such a dispersion relation has a remarkable implication: Such systems
can be superuid. Superuidity is a remarkable phenomenon: They are uids without
any viscosity. They ow around objects, and items can be dragged through such uids
without losing velocity. It is a perfectly frictionless uid.
This behaviour is a consequence of the dispersion relation, for which
ωk
ink = vCrit > 0 (6.57)
k
holds true. Again, for small values of k the dispersion relation is linear. Such quasi-
particles are called phonons. Excitations with a k nearby the local minimum are called,
for historical reasons, rotons.4 We will now see how this dispersion relation gives
rise to superuidity. In order to see this, we will investigate a tube through which a
superuid is owing. We will look at this system in two reference frames:
• In the rst reference frame B, the tube is at rest and the uid moves with velocity
−v.
• In the other reference frame B ′ , the uid is at rest, and the tube moves with
velocity v.
Of course, both are perfectly legitimate references frame, related to each other by a
Galilei transformation.5
4 For superuid He4 this minimum is at k = 1.91 × 1010 m~. The effective mass of such rotons is
0
about 0.16 the mass of Helium, and they have an energy gap of ∆/k = 8.6K.
5 Here a brief reminder on Galilei transformations: Let us assume we have N Teilchen, the above velocity
is v and the particles have positions {xj } and momenta {pj }. Then we have in the above reference frames
xj = x′j − vt, (6.58)
pj = p′j − M v. (6.59)
This means that ∑ ∑
P = pj = (p′j − mv) = p′ − M v. (6.60)
j j
The energy transforms as (note that V is here the interaction, not the velocity)
∑ p2j ∑
E = + V (xj − xk )
j
2M
〈j,k〉
)2
∑M p′j ∑
(
= −v + V (x′j − x′k )
j
2 M
〈j,k〉
∑ (p′j )2 1 ∑
= − p′ v + M v2 + V (x′j − x′k )
j
M 2
〈j,k〉
1
= E ′ − p′ v + M v2 . (6.61)
2
This is precisely the transformation that we need, and it makes little difference whether the particles are
classical or quantum.
We now allow for a little bit more, namely that in B ′ the uid has energy E ′ and
momentum P ′ . The energy E in reference frame B and the momentum are given by
P = P ′ − M v, (6.62)
1
E = E ′ − P v + M v2 . (6.63)
2
Let us discuss why there is no friction up to a critical velocity. Friction means that
quasi-particles are being generated that dissipate energy and transfer this energy to
other, undirected degrees of freedom. If there are no quasi-particles, there is no friction
and dissipation. Viewed in reference frame B ′ , when encountering dissipation quasi-
particles will have to be generated that move with the tube. In the reference frame of
the tube it looks as if the uid was decelerated.
Of course, such excitations will only arise if they are energetically favoured, and
this insight is at the heart of the matter. Let us begin in the ground state at temperature
T = 0. In the reference frame B ′ , energy and momentum are given by
E′ = Eg′ , (6.64)
′
p = 0. (6.65)
In B, these expressions are
1
Eg = Eg′ + M v 2 (6.66)
2
p = −M v. (6.67)
If a quai-particle with momentum p = k (actually, p = ~k, but we have set ~ = 1) and
energy ωk is being generated, energy and momentum in B ′ are given by
E′ = Eg′ + ωk ,
p′ = k, (6.68)
and hence in B ′
1
E′ = Eg′ + ωk − kv + M v 2 ,
2
p′ = k − M v. (6.69)
The excitation energy in B equals, again with ~ = 1,
∆E = ωk − kv. (6.70)
For this reason, ∆E is the change of energy of the uid due to the generation of a
quasi-particle in reference frame B. Only if
∆E < 0 (6.71)
the uid will lessen its energy by the dissipation process. Of course, this means that,
seeing the problem as a one-dimensional problem, that
ωk
v> (6.72)
k
has to hold, such that energy is lost in the rst place, for this excitation labeled k. In
other words, the velocity has to be larger than a critical velocity.
What does the dispersion relation have to do with all this? Well, everything. The
thing is that if (6.57) holds true and a critical velocity vCrit exists, then there is no k for
v < vCrit (6.73)
that satises (6.72). Therefore, there cannot be excitations that are energetically fa-
voured, and hence there is no dissipation. The phenomenon of superuidity is hence a
consequence of a curious dispersion relation, in which the energy per wave number is
bounded from below. We have – within a framework of second quantization – hence
understood how quantum gases can ow entirely without friction and dissipation as a
genuine quantum effect. This is one of the most curious and intriguing macroscopic
quantum phenomena. For uid Helium, this effect occurs at temperatures T < 2.18K
and can well be observed. This is one of the core results of this course.
6.3.5 Phenomenology of superconductivity

Until now the discussion was limited to the situation of zero temperature T = 0. As
mentioned before, the temperature does not have to be strictly zero (a ctitious state
of affairs anyway) in order to arrive at a superuid. For temperatures different from
zero, superuids are captured rather well by the two uid model due to Tizsa: It is
a purely phenomenological model that regards such superuid as being composite of
a superuid component (the actual superuid) and a normal component (that for low
temperatures largely consists of phonons). The superuid component is assumed to
have
• zero entropy and
• can move freely through capillaries without any friction.
The normal component has a non-zero entropy, and also a viscosity different from zero.
Based on this crude model, a number of effects can be nicely explained.
• For example, the effect can be explained in which a small hole in a tank lled
with He II can serve as a “lter” for the superuid component. If one connects
two tanks via a thin capillary and puts some pressure onto one tank, the uid
will ow from one tank A to the other B. But then, because the superuid
component has no entropy, the entropy per mass in tank A will increae and that
in B will decrease. Hence, A will cool down, while B will heat up. This is
indeed observed, in a fascinating effect called mechanocaloric effect.
• A further quite spectacular effect is called fountain effect: It is basically the op-
posite effect in which a temperature gradient gives rise to a difference in pressure.
If one does it well, one can generated nice fountains, hence the name.
• Finally, there is an effect called second sound: If one generates a sound wave,
both the normal and the superuid component will usually oscillate, and hence
one will arrive at a sinusoidal modulation of the mass density. But now there
is also another type of oscillation conceivable in which both components oscil-
late against each other with an opposite phase. This would not be a sound wave
in the ordinary sense, as the mass density will remain constant in time. But it
would give rise to a modulation of the entropy density, for the above mentioned
reason. One could therefore hope that by means of clever local heating one can
generate such waves. Then the temperature gradient would not diffusively prop-
agate, but ballistically, like a wave, propagating with some speed of sound. This
could be called a sound wave of second sound, and indeed, it can be observed in
experiments.
Chapter 7
Superconductivity
Let us now have a look at another interesting and somewhat surprising phenomenon,
namely superconductivity (zero electrical resistance). Superconductivity can satis-
able be described within a microscopic theory, which is valid for T = 0 and small
temperatures. Until now, superconductivity at high temperatures remains to some ex-
tend unresolved.
7.1 Introductory thoughts

Superconductivity was rst described by Kamerling-Onnes in 1911. He found that
mercury, that was cooled under Tc = 4.15K, lost its resistance abruptly. For all tem-
peratures below that critical Tc , the same behavior was observed. Today there are
more such superconducting substances known, that completely loose their resistance
upon cooling under some appropriate critical temperature. One of the most astonish-
ing effects might be the existence of permanent currents, which was also found by
Kamerling-Omnes; consider a ring of superconducting material at some T > Tc ex-
posed to an orthogonal magnetic eld B. Now cool the ring to T < Tc and switch off
the magnetic eld. A current will be induced in the ring, that is identical to the current
that was there when B was on. Since there is no resistance, the current will stay as it is
for ever.
Another effect is the Meissner-Ochsenfeld-effect, which states, that in the interior
of a superconductor, the magnetic eld
B = 0. (7.1)
Superconductivity remained completely unclear for many years, until Gorter, Casimir
and London developed rst theories, which were macroscopic in the beginning (we are
planning to get back to this later). Herbert Fröhlich observed rst, that superconductiv-
ity in metals originates from interaction of electrons with phonons (the quasi particles
of lattice vibrations). That central insight led to a lot of research in microscopic theo-
ries, one of which was the BCS theory from Bardeen, Cooper and Schrieffer. It was the
5
6 CHAPTER 7. SUPERCONDUCTIVITY
rst satisfying theory, and – of course – formulated in second quantization. There is no

time to go more deeply into that, but let us understand the main ideas.
7.2 BCS theory

7.2.1 Microscopic Hamiltonian
The Hamiltonian of a gas of electrons (i.e. fermions) in a solid body of volume V reads
as follows.
Hamiltonian of electrons in solid body:
H = H 0 + H1 , (7.2)
∑ †
H0 = εk fk,σ fk,σ , (7.3)
k,σ
1 ∑ † †
H1 = − Vk,k′ fk,σ f−k,−σ f−k′ ,−σ fk′ ,σ . (7.4)
2V
k,k ,σ
′
†
As always fk,σ denotes the fermionic creation operator with wave number k and
spin component σ. Moreover
k2
εk = −µ (7.5)
2M
is the kinetic energy, corresponding to the wave number k, where M is the mass of an
electron and µ is the chemical potential, which is in a system of exactly N particles
(like a Lagrange or more precisely Kuhn Tucker multiplier) xed by
∑ †
〈 fk,σ fk,σ 〉 = N. (7.6)
k,σ
Vk,k′ is the effective interaction between the electrons in k-space, which obviously
satises
Vk,k′ = Vk′ ,k . (7.7)
7.2.2 Justication of the microscopic Hamiltonian

The central insight is the one from Cooper, supported by Fröhlich’s observations, who
noticed, that the known ground state of non interacting fermions collapses if one allows
a small attractive interaction between them. The idea can be roughly summarized as
follows.
• On its way through the solid body, the electron deforms the background of posi-
tively charged ions. Before, we considered only a classical and xed background,
but in a good microscopic theory we have to allow some movement in it.
7.2. BCS THEORY 7
• Hence the electron leaves a trace of a slightly higher density of positively charged
ions, which again leads to the attraction of another (negatively charged) electron.
This effect can be seen like an effective attraction between two electrons. The re-
pulsion among each other is compensated by the mechanism, we talked about in
the last chapter on Fermi gases. So the trace of deformation leads to an attraction.
• There is a rather subtle effect, namely that the described attractive electron-
electron-interaction is, as a result of the slower movement of ions, retarded com-
pared to the Coulomb interaction between them, which can lead to knowledge
about the range of the effective interaction.
Those thoughts should have given some ideas why the Hamiltonian is plausible (even
if we do not deliver a full derivation here). The original publication is also not entirely
concrete on these matters, but in the meantime, the picture of an effective fermionic
interaction mediated via bosonic systems has been understood rather well.
7.2.3 Bogoliubov transformation of the Hamiltonian

We will now try to nd the spectrum of the Hamiltonian. We can not nd an exact so-
lution, but by using a similar strategy as we did with superuids, we will nally end up
with a quadratic Hamiltonian. The idea is again to do a series of approximations which
are highly plausible, given the mindset laid out here, until one arrives at a quadratic
Hamiltonian that can be exactly solved. Needless to say, this can also be seen as an
invitation that it makes a lot of sense to revisit such derivations, and aim at bounding er-
rors made in approximations in a more rigorous mindset. To proceed, it will be helpful
to dene the following two new sets of operators {Ak } and {Bk }:
fk,1/2 = uk Ak + vk Bk† , (7.8)

f−k,−1/2 = uk B k − vk A†k , (7.9)
with {uk }, {vk } ∈ R and
uk = u−k , (7.10)
vk = v−k , (7.11)
u2k + vk2 = 1. (7.12)
This is, again, a Bogoliubov transformation, this time a fermionic one. It is easily found
that the inverse transformation reads
†
Ak = uk fk,1/2 − vk f−k,−1/2 , (7.13)
†
Bk = uk f−k,−1/2 + vk fk,1/2 . (7.14)
We would like the new coordinates to fulll the same anti-commutation relations as the
old ones, which are
{Ak , A†k′ } = δk,k′ , (7.15)

{Bk , Bk†′ } = δk,k′ . (7.16)
Inserting these new operators into the above given Hamiltonian and also using anti-
commutation relations to order annihilation operators, gives us the new Hamiltonian.
(Note that one calls an order of this kind normal order)
H = E0 + H0′ + H1′ + H2′ , (7.17)
where
∑ 1 ∑
E0 = 2 εk vk2 − Vk,k′ uk vk uk′ vk′ , (7.18)
V
k k,k
′
 
εp (u2p − vp2 ) + 2up vp
∑ ∑
H0′ = Vp,p′ up′ vp′ 
p
V ′ p
× (A†p Ap + Bp† Bp ), (7.19)

 
u2 − vp2 ∑
2εp up vp − p
∑
H1′ = Vp,p′ up′ vp′ 
p
V ′ p
× (A†p A†p + Bp Bp ). (7.20)
H2′ contains expressions involving higher order polynomials in Ap and Bp , which can
be neglected as they represent interactions between quasi particles. The nal result will
show that the ground state is the vacuum (in terms of quasi particles) and as such, at
low energies, there will be little interaction between them.
Now we can again use the freedom we have, to choose {uk } and {vk } such that
H1′ = 0, (7.21)
which will be the case if

u2p − vp2 ∑
2εp up vp = Vp,p′ up′ vp′ . (7.22)
V ′ p
We decide to pick
 1/2
1  εp
up = √ 1+ √  , (7.23)
2 ∆2p + ε2p
 1/2
1  εp
vp = √ 1− √  , (7.24)
2 ∆2 + ε2 p p
where ∆p is determined by the anti-commutation relations as
1 ∑ V ′∆ ′
∆p =  p,p p , (7.25)
2V ′ ∆ 2 (p′ ) + ε2
k′
k
7.2. BCS THEORY 9
Kicking out H1′ in this fashion was no approximation but just our freedom to choose
the pre-factors in the Bogoliubov transformation (which was of course a reason for
doing the transformation in the rst place). The new Hamiltonian then becomes
H = E0 + H0′ , (7.26)
with
∑ 1
E0 = √
p ∆2p + ε2p
 √  
× εp ∆2p + ε2p − εp − ∆2p /2 , (7.27)
and
∑√
H0′ = ∆2p + εp (A†p Ap + Bp† Bp ). (7.28)
p
This procedure, similar to the one we did for superuids, leaves us with a, not only
quadratic Hamiltonian, but it even takes the form of uncoupled harmonic oscillators and
is thus solved. We have followed a very similar logic here as before when discussing
superuidity.
7.2.4 Discussion of the interaction

Since this can be seen like a “rst date” with the ideas of superconductivity, we only
consider the case T = 0 and say µ = k02 /(2M ). All energies are thus expressed
in terms of the chemical potential, upon choosing εk . We will make the following
assumption about the interaction potential:
{
C, if |εk |, |ε′k | ≤ ωc ,
Vp,p′ = (7.29)
0, otherwise.
with C ∈ R being some constant.

Here ωc  p20 /(2M ). So the interaction occurs only between electrons with some
momentum in a small sphere around the momentum p0 . It is consistent with all said
above to assume {
∆, if |εp | ≤ ωc ,
∆p = (7.30)
0, otherwise.
In the limit of a large volume V , the constant ∆ can be found by solving the integral
equation
C 1
∫
∆=∆ dp √ , (7.31)
2(2π)3 ∆ + ε2
2
p
where one integrates over all k for which
|εk | ≤ ωc (7.32)
The integral becomes one dimensional:

∫ √p20 +2M ωc ( )−1/2
C 1
∆=∆ 2 √ dpp2 ∆2 + 2
(p2 − p20 )2 . (7.33)
2π p20 −2M ωc 4M
Introducing the new variables p = p0 + x and using that p20  2M ωc , we get

M ωc /p0 )−1/2
Cp20 p2
∫ (
∆≈∆ dx ∆2 + 02 x2 . (7.34)
4π 2 −M ωc /p0 M
And nally

Cp0 ∆2 + ωc2 + ωc
∆≈∆ log  . (7.35)
4π 2 ∆2 + ωc2 − ωc
This is a curious expression, since ∆ is contained in both sides of the equation.
7.2.5 The spectrum of the Hamiltonian

We found the spectrum of the Hamiltonian to be
∑√
E0 + ε2k + ∆2k (NA (k) + NB (k)), (7.36)
k
where NA (k) and NB (k) are the number of quasi particles A and B with wave number
k respectively. There are now two possible solutions to Eq.7.35, one is trivially ∆ = 0.
It corresponds to a state with neither any A nor B quasi particles and the energy is just
Egs = 0. (7.37)
Also p0 = pF and up (Eq.7.23) and vp (Eq.7.24) can be chosen to be
up = 1, vp = 0 for εp > 0 (7.38)
and
up = 0, vp = 1 for εp < 0, (7.39)
where ε >< 0 implies |p| >< pF . If |p| > pF , it follows
Ap = fp,1/2 , (7.40)
Bp = f−p,−1/2 . (7.41)
Those operators thus annihilate fermions outside the Fermi sphere. Otherwise, if |p| <
pF :
†
Ap = −f−p,−1/2 , (7.42)
†
Bp = fp,1/2 . (7.43)
7.2. BCS THEORY 11
As such, they ll up fermionic holes in the Fermi sphere. The ground state with energy
Egs = 0, is the one, where all states up to |p| ≤ pF are lled up and all states with
|p| > pF are empty. Exited states have some fermions that jumped up to |p| > pF . For
sufciently large systems, the spectrum is thus continuous. If C is positive and large
enough, that is, if there is enough attractive interaction between fermions of opposite
momentum and different spins near the Fermi level, there exists another solution for
Eq. (7.35).
Energy gap of the superconducting solution : We nd the solution
e−D/C
∆ = 2ωc , (7.44)
1 − e−2D/C
2
2π
D = , (7.45)
M p0
which gives rise to an energy gap of 2∆ between the ground and rst excited state.
This solution is called superconducting solution and the ground state energy is
Egs < 0. Hence it is the real ground state of the system. Fixing the particle number,
one nds an energy gap of 2∆ between the ground state energy and the rst excited
energy. In fact,
† †
A†p Ap − Bp† Bp = fp,1/2 fp,1/2 − f−p,−1/2 f−p,−1/2 . (7.46)
In the ground state one nds

† †
〈fp,σ fp,σ 〉 = 〈f−p,−σ f−p,−σ 〉. (7.47)
Which corresponds to electrons appearing in pairs; pairs with opposite k and spin, the
so called Cooper pairs:
(k, 1/2; −k, −1/2). (7.48)
It is energetically favourable for the electrons to come in such pairs, in fact, the ground
state is well described by the BCS ground state vector.
BCS ground state vector:

∏ † †

|ψBCS 〉 = uk + vk fk,1/2 f−k,−1/2 |ø〉. (7.49)
k
“Divorcing” such a pair generates two unpaired electrons and NA + NB = 2, such

that the rst excited energy is 2∆ above the ground state, and precisely that energy gap
is responsible for superconductivity, as too much energy is needed to generate such an
excitation. Thus no dissipation occurs.
Chapter 8
Lattice models and strongly

correlated systems
In the following we would like to turn towards lattice models. Those are very impor-
tant models in physics as they describe many systems in solid state physics, like the
structure of crystals, that determines the graph G = (V, E). There are many kinds of
lattices, in different dimensions, the simple chain in 1D being the easiest case. Lat-
tices are described by graphs, bonds and sites, where quantum particles sit at each site.
There is a natural metric dist(j, k), which describes the minimal number of bonds one
has to cross to get from site j to site k, the graph theoretical distance.
An important class of examples is constitutes by one-dimensional chains with sites
1, . . . , n, where on each site sits a quantum degree of freedom, like a spin with Hilbert
space C2 , or a bosonic degree of freedom H = L2 (R), or a fermion with spin, and
the distance is of course just dist(j, k) = |j − k|. Interactions in such systems are
usually local: that is, not all particles interact with all others but only with their nearest
neighbours or next nearest neighbours. The Hamiltonian of a local chain with nearest
neighbour interaction takes the following form.
Hamiltonian of local chain:

n
∑
H= hj,j+1 , (8.1)
j=1
where site n + 1 is again the rst site (periodic boundary conditions).
Strictly speaking, the summands are

⊗ ··· ⊗ ⊗ hj,j+1 ⊗ ⊗ ··· ⊗ , (8.2)
but the sites on which the Hamiltonian only acts in a trivial manner are usually ne-
glected; hj,j+1 acts (non trivially) only on site j and j + 1. They are of the form
hj,j+1 = uj + vj,j+1 , (8.3)
5
6 CHAPTER 8. LATTICE MODELS AND STRONGLY CORRELATED SYSTEMS
where uj is the onsite term and vj,j+1 is the interaction between sites.
We will have a look at some examples. Non interacting models that relate to
fermions by virtue of the Jordan-Wigner transformations, the XY-model, quantum crit-
icality and even strongly correlated systems.
8.1 Lattice models and spin chains

The subsequent list summarizes some paradigmatic models of this kind that are fre-
quently discussed in the literature.
• The harmonic chain is described by

1 ∑ 
H= pi Pi,j pj + xi Xi,j xj , (8.4)
2
i,j∈L
where X, P ∈ Rn×n are real, symmetric and positive matrices, which deter-
mine the coupling. The operators {xi } and {pi } fulll canonical commutation
relations
[xj , pk ] = iδj,k . (8.5)
√
The Hamiltonian in terms of bosonic operators bj = (xj + ipj )/ 2 for i =
1, . . . , n reads
1 ∑ †
b Ai,j bj + bi Ai,j b†j + bi Bi,j bj + b†i Bi,j b†j ,

H= (8.6)
2 i,j i
where
A = (X + P )/2, B = (X − P )/2. (8.7)
This model, e.g., captures lattice vibrations in solid bodies or discrete versions of
free elds in quantum eld theory. It is a non interacting model, as the Hamilto-
nian is quadratic in bosonic operators. If the interaction is translational invariant,
A and B are circulant matrices. Also, if there is only nearest neighbour interac-
tion, A and B are nonzero only at the diagonal and rst off diagonals.
• Fermionic chain:
n
1 ∑ † 
H= f Ai,j fj − fi Ai,j fj† + fi Bi,j fj − fi† Bi,j fj† (8.8)
2 i,j=1 i
For this Hamiltonian to be Hermitian
AT = A, B T = −B (8.9)
† √ † √
has to hold. Now xj = (fj + fj )/ 2 and pj = i(fj − fj )/ 2 are not posi-
tion and momentum operators but Majorana fermions. The energy gap between
ground and rst excited state is the smallest non zero singular value of A + B
8.2. XY MODEL AS A PARADIGMATIC EXACTLY SOLVABLE MODEL 7
• XY-chain: spin chains of that kind are important models. At each site sits a
C2 spin. The most popular and exactly solvable model is the XY-chain with a
transverse magnetic eld. The corresponding Hamiltonian reads
n   n
1∑ 1+γ 1−γ λ∑
H=− Xi Xi+1 + Yi Yi+1 − Zi , (8.10)
2 i=1 4 4 2 i=1
where 〈i, j〉 denotes summation over nearest neighbours. γ is some weird pa-
rameter, λ the extern magnetic eld. We are coming back to that model later.
• The Heisenberg model: is another important spin model and can be described by
n n
1∑ λ∑
H=− (Xi Xi+1 + Yi Yi+1 + Zi Zi+1 ) − Zi . (8.11)
2 i=1 2 i=1
which is not a free model anymore.

• Obviously, this is only the tip of the iceberg, and there are lots of further models
of a similar type.
8.2 XY Model as a paradigmatic exactly solvable model

The XY model is the easiest model that shows critical behaviour, quantum phase transi-
tions, scaling properties and degenerate ground states, while still being easily solvable.
The central insight is, that even though being a spin chain, the model can be translated
to free fermions.
8.2.1 Jordan-Wigner transformation

The corresponding transformation is the Jordan-Wigner transformation (Jordan-Wigner
transformation ), which states that a system of
• n fermionic modes and
• n spins
are isomorphic to each other. In second quantization a state of fermions in n modes
reads
|N1 , . . . , Nn 〉 = (f1† )N1 . . . (fn† )Nn |ø〉 (8.12)
with Nk ∈ {0, 1} for all k. This is nothing but a state vector in (C2 )⊗n , which looks
just like a chain of n spins, where each can be either up (0) or down (1).
The only difculty is to represent the spin operators (Oj ) by fermionic operators
(fj ). Fermionic operators anti-commute, whereas local spin operators do not. The key
question to be solved, hence, is to nd way to make sure that
 
〈ø|fN Mn
. . . f1M1 fj† fj (f1† )N1 . . . (fn† )Nn |ø〉 = 〈M1 , . . . , Mn |Oj |N1 , . . . , Nn 〉
(8.13)
8 CHAPTER 8. LATTICE MODELS AND STRONGLY CORRELATED SYSTEMS
holds? The answer is that spin operators are not locally represented. And the specic
transformation at the heart of this is given by the Jordan-Wigner transformation, which
is actually less of a transformation than a representation.
Jordan-Wigner transformation: Systems of n fermionic modes are isomorphic

to a system of n spins. For a xed order of fermionic modes one can transform
operators like
( j−1 )
fj† =
∏
Zj σj− , (8.14)
k=1
( j−1 )
∏
fj = Zj σj+ (8.15)
k=1
and inversely
Zj = 1 − 2fj† fj , (8.16)
j−1
(1 − 2fk† fk )fj ,
∏
σj+ = (8.17)
k=1
j−1
(1 − 2fk† fk )fj† .
∏
σj− = (8.18)
k=1
Here, as always
σj− = (Xj − iYj )/2, (8.19)

σj+ = (Xj + iYj )/2. (8.20)
Some remarks are in order at this point.
• Again, a core insight is that the Jordan-Wigner transformation is non local, and
necessarily so. Fermionic modes that have support only on few fermionic modes
can have support on almost the whole lattice in terms of spin operators. This is a
consequence of the anti-commutation-relation-problem mentioned above.
• Consequently the order of fermionic modes plays a role as the spin operator on
site j contains all fermionic modes to the left of j. As such, we have to x the
order in the beginning.
• Physical operators are only those, which do not depend on the order. See the rst
example below.
Two examples of operators under Jordan-Wigner transformation :
fj† fj = (1 − Zj )/2, (8.21)

for all j. So the number operator on mode j becomes the Pauli-Z operator in spin
representation. Also the non local string canceled out (because Zj2 = ), and it is thus
a physical operator. Say, for s > 0
 
j+s−1
†
∏
− +
fj fj+s = σj Zl  σj+s . (8.22)
l=j+1
Evidently fj† fj+s are not mapped to Zj Zj+s , as one might have expected, but there
is still a string between j and j + s. Hence the operator could have support on many
sites, which is a problem in the theoretical characterization of fermions as most models
depend on some locality of interactions. In the following, we will look at an example,
where this problem does not occur.
8.2.2 XY model as a fermionic model

Consider the XY model again, a chain of spins with nearest neighbour interaction.
Choosing the order such that nearest neighbours are always on neighbouring sites, one
gets a non interacting local fermionic model:
XY model as non interacting fermionic model: A Jordan-Wigner transformation

of the XY model gives a non interacting fermionic model of the form indicated in
Eq.8.8. With
Ai,i = λ, (8.23)
Ai,j = −1/2, if |i − j| = 1, (8.24)
γ
Bi,j = −Bj,i = , if |i − j| = 1 (8.25)
2
and else Aj,k = Bj,k = 0. This corresponds to a non interacting fermionic chain
Hamiltonian with nearest neighbour interaction, with A and B being circulant ma-
trices.
We chose periodic boundary conditions, which means site n + 1 is site 1 again. We

get:
n
1 ∑ † 
H = − fi fi+1 + h.c.
2 i=1
n n
γ∑ λ∑ †
+ (fi fi+1 + h.c.) + f fj , (8.26)
4 i=1 2 j=1 j
and in the isotropic case of γ = 0 simply

n n
1 ∑ †  λ∑
H = − f fi+1 + h.c. + f † fj . (8.27)
2 i=1 i 2 j=1 j
10CHAPTER 8. LATTICE MODELS AND STRONGLY CORRELATED SYSTEMS
The case γ = 0 is (not very intuitively) called XX model. No strings occur, so the
model is indeed local.
8.2.3 Spectrum and ground state

Let us be specic and take the value γ = 0 to be concise in what follows. We introduce
a new set of fermionic operators {a1 , . . . , an }, that are Fourier transforms of the old
fermionic operators and retain anti-commutation relation
n
1 ∑
ak = fj e−i2πkj/n , (8.28)
n1/2 j=1
for k = 1, . . . , n. The operation that maps the old set to the new one is indeed unitary
(a1 , . . . , an )T = U (f1 , . . . , fn )T , (8.29)
with U ∈ U (n). The transformed Hamiltonian reads

n
Λk a†k ak ,
∑
H= (8.30)
k=1
with the following spectrum.
Spectrum of the XX model:

 
λ 1 2πk
Λk = − cos . (8.31)
2 2 n
In the limit n → ∞ and with periodic boundary conditions one may also write
λ 1
Λφ = − cos φ, (8.32)
2 2
with π ∈ [0, 2π), by replacing 2πk/n with φ.
The discussion of this spectrum is easy but nevertheless shows rich behaviour.
• if λ < −1 then, for all k

Λk ≥ 0 (8.33)
so the ground state is the fermionic vacuum |ø〉 for which
fk |ø〉 = 0 for all k. (8.34)
mapped back to spins, the ground state is
|GS〉 = |0, 0, . . . , 0〉, (8.35)
a product state of fermions, all pointing in the same direction.

• Similarly for λ > 1 and all k one gets
Λk ≤ 0, (8.36)
and the ground state of the XX model in spin representation reads
|GS〉 = |1, 1, . . . , 1〉. (8.37)
Those two phases are called ferromagnetic phases. It is highly plausible, that
for a sufciently strong magnetic eld (|λ| > 1) all spins align to it. The only
surprise value might be, that there is no correlation whatsoever and the ground
state is a proper product state.
• It remains the case λ ∈ (−1, 1). Since there is a shift of sign in the Hamiltonian
one has to consider two cases for the ground state. One is:
fk |GS〉 = 0 for all k for which Λk > 0. (8.38)
And the other one:
fk† |GS〉 = 0 for all k for which Λk ≤ 0. (8.39)
We dene the Fermi level as

⌊  ⌋
n λ
kc = arccos . (8.40)
2π 2
It is easy to see that

{
< 0, for kc ≤ k ≤ 0 or n − 1 ≥ k ≥ b − kc ,
λk (8.41)
≥ 0, else.
(Here bxc denotes the oor function, which gives the largest integer that is smaller than
x.) So the ground state is the one, where the system is lled up to the Fermi level. The
ground state energy is
∑
Λk θ(−Λk ). (8.42)
k
8.2.4 Criticality and quantum phase transitions

There can be said a lot about critical systems but we will only discuss the ideas by
taking the above example (XX chain). Consider again the energy spectrum
λ 1
Λφ = − cos φ. (8.43)
2 2
with φ ∈ [0, 2π).
• If |λ| > 1, the spectrum has no zeros: as explained above, this leads to all spins
pointing either up or down (ferromagnetic phases). The rst excited state is the
one, where one fermion is taken out. To do so, the energy
∆E := min{Λφ : φ ∈ [0, 2π)}, (8.44)
is needed. And in that case

∆E > 0. (8.45)
therefor there is a nonzero energy gap between the ground state and rst excited
state. Such phases are called critical or gapped phases. In general, gapped
phases are those that have a nonzero energy gap between its ground state(s) and
its rst excited state in the thermodynamic limit.
• In contrast, for λ ∈ (−1, 1), there are zeros. So
∆E = 0. (8.46)
A system in that phase is at a critical point. In the thermodynamic limit, to go to

an excited state costs arbitrarily little energy.
Whether or not there is an energy gap makes a huge difference on physical grounds.
If there is one, one can show, that the ground state always exhibits some locality.
correlations decay exponentially in distance. For any observable Oj that has support
on j alone, and Oi that has support on i alone, one has
|〈Oi Oj 〉 − 〈Oi 〉〈Oj 〉| = |〈GS|Oi Oj |GS〉 − 〈GS|Oi |GS〉〈GS|Oj |GS〉|

∼ e−|i−j|/ξ , (8.47)
for an appropriate correlation length
ξ > 0, (8.48)
So if one measures at two sites, far enough from each other, the results will be un-
correlated. That this holds was “known” for many years. Maybe surprisingly, it was
rst properly proven with rigorous methods only in 2004, with methods referred to as
Lieb-Robinson theorems, bounds to group velocities in general lattice models (and then
using an ingenious trick of an integral in the complex plane).
The situation changes radically for critical models. Correlations do not decay expo-
nentially anymore, but following power laws. In fact. there is no length scale anymore
and the correlation function (Eq. 8.47) decays only algebraically. The ground state will
in a sense that can be made precise look the same on all scales. The eld of confor-
mal eld theory captures such physical systems that follow such scaling laws at critical
points.
The theory of universality studies what happens close to critical points λc (in our
case |λc | = 1). One nds
ξ ∼ |λ − λc |−ν , (8.49)
8.3. STRUCTURE OF NON INTERACTING BOSONS AND FERMIONS 13
with ν ∈ R being some critical exponent. And for the energy gap
∆E ∼ |λ − λc |s , (8.50)
with another critical exponent s. The universality hypothesis (which is a hypothesis
since is has not been proven yet) states that there are only nitely many critical ex-
ponents possible in nature, which is quite surprising. Models with different critical
exponents are collected in different universality classes. In our XX model we nd that
∆E = |λ − 1| (8.51)
so the critical exponent s = 1 and
1
ξ= , (8.52)
(λ − 1)1/2
so the critical exponent ν = 1/2.
Let us take the opportunity to be at this point a bit more precise about quantum
phase transitions in the rst place.
• The family of Hamiltonians
H(g) = H0 + gH1 , (8.53)
where g ∈ R is some parameter (like a magnetic eld strength) shows quantum
phase transition if there are non analyticities in the ground state energy for some
g = gc .
• One classies quantum phase transitions of rst and second order, where rst
order quantum phase transitions are indicated by kinks in the ground state energy
and second order quantum phase transitions by a smooth energy but kinks in the
rst derivative. The XX model is an example of such a second order quantum
phase transition.
8.3 Structure of non interacting bosons and fermions

It should be clear by now, that non interacting models play an important role in physics,
either exactly or at least in good approximation. Let’s have some general words on such
systems again.
8.3.1 Coordinate transformations

Let us at this point go more into detail what we precisely mean by linearly transforming
one set of bosonic or fermionic coordinates into a new one. We consider bosonic
(b1 , . . . , bn ) or fermionic (f1 , . . . , fn ) annihilation operators that commute or anti-
commute. From those we can dene
√
xj = (a†j + aj )/ 2, (8.54)
√
pj = i(a†j + aj )/ 2, (8.55)
where aj = bj , fj are the position and momentum operators, or the Majorana fermions,
respectively. So how can we transform them linearly, respecting the relevant commu-
tation or anti-commutation relations?
Symplectic transformations for bosons: The allowed linear transformations from

one set of canonical coordinates (x1 , . . . , xn , p1 , . . . , pn ) to a new such set that
satises the canonical commutation relations give rise to the so-called symplecic
transformations
(x1 , . . . , xn , p1 , . . . , pn ) 7→ S(x1 , . . . , xn , p1 , . . . , pn ). (8.56)
They satify
SσS T = σ, (8.57)
where σ is a matrix that embodies the canonical commutation relations (again
choosing ~ = 1) as [ ]
0
σ= . (8.58)
− 0
These transformations form a group, the real symplectic group S ∈ Sp(2n, R).
Some remarks are again in order here:1

• It is not difcult to see that those are precisely the linear transformations that
respect the canonical commutation relations.
• σ has the given form, since positions of one mode of course commute with the
momentum of another mode. There is only an entry in σ if a position coordinate
of one mode comes together with a momentum coordinate of the very same
mode.
• The same structure is encountered in classical statistical mechanics.
An important special case we have already encountered before: The case in which
the positions and momenta are all rotated orthogonally in the same fashion. Such
transformations are always symplectic. Formally, this means that
O ⊕ O ∈ Sp(2n, R) (8.59)
for all O ∈ O(n), where the latter denotes the real orthogonal group (the real unitary
transformations). This is no surprise: If we rotate momenta and positions in the same
way, then the commutation relations are preserved. 2 If should be clear that all trans-
formations that we have encountered as “Bogoliobov transformationens” are included
in this group, and are special cases of such symplectic transformations.
1 If
S ∈ Sp(2n, R), then also S −1 ∈ Sp(2n, R) and S T ∈ Sp(2n, R).
2 These transformations constitute a reducible representation of O(n), which in turn is a subgroup of
K(2n),
K(2n) = Sp(2n, R) ∩ O(2n), (8.60)
the maximum compact subgroup of Sp(2n, R). Such transformations are tremendously important in qaun-
tum optics, as they are those that can be generated with passive optical elements such as beam splitters,
mirrors, and phase shifters. Sometimes, a bit of group theory can be helpful.
8.3. STRUCTURE OF NON INTERACTING BOSONS AND FERMIONS 15
An important insight is that with such symplectic transformations, one can diag-
onalize strictly positive matrices. This is different from the “diagonalization” with
unitary transformations, but to the same effect. This insight is at the heart of notions
of decoupling harmonic chains, in normal mode decomposition. In condensed matter
physics and in quantum optics, this is very important. The statement goes as follows.
Williamson theorem: Every strictly positive 2n × 2n-Matrix M can be brought

into a diagonal form under symplectic transformations,
SM S T = diag(d1 , d1 , d2 , d2 , . . . , dn , dn ), (8.61)
where {dj } are the positive square roots of the eigenvalues of −σM σM . The are
called symplectic eigenvalues.
While the core statement is difcult to show, the latter statement is rather easy.
Note also that the symplectic eigenvalues are in general different from the ordinary
eigenvalues. How, now, can we transform fermionic operators? Well, in a somewhat
similar fashion.
Linear coordinate transformations for fermions: The allowed transformations

that linearly map (x1 , . . . , xn , p1 , . . . , pn ) into a new set of Majorana fermions, are
transformations of the form
(x1 , . . . , xn , p1 , . . . , pn ) 7→ O(x1 , . . . , xn , p1 , . . . , pn ), (8.62)
with O ∈ O(2n).
Interestingly, the operations O ⊕ O with orthogonal O ∈ SO(n) are contained in

both groups. In fact, in many practical settings, one can treat bosons and fermions in
an analogous fashion.
8.3.2 Covariance matrices

Non-interacting systems have another interesting property: Its ground states as well
as its thermal states are Gaussian states which are perfectly dened by means of their
rst and second moments (for fermions actually only the second moments). The rst
moments
mj = tr[ρrj ] (8.63)
are simply the expectation values of
(r1 , . . . , r2n ) = (x1 , . . . , xn , p1 , . . . , pn ). (8.64)
Again, for fermions, they are all zero. For bosons, they are often also zero: What
is more, one can often shift systems in phase space to make them zero. The second
moments can be embodied in covariance matrices.
Bosonic covariance matrices: The second moments of a bosonic state with van-
ishing rst moments can be embodied in the 2n × 2n symmetric covariance matrix
with entries
γj,k = tr[ρ(rj rk + rk rj )]. (8.65)
It is not only positive, but satises with
γ + iσ ≥ 0 (8.66)
the Heisenberg uncertainty principle.
The latter, of course, also implies that γ = γ T . Maybe in this form, the uncertainty
principle it is a bit in disguise. But a moment of thought reveals that this is the right
form. Using the above Williamson theorem, one can treat each mode separately. For
each such mode, we then have
[ ] [ ]
2〈x2 〉〈xp〉 + 〈px〉 0 1
γ + iσ = 2 +i ≥ 0, (8.67)
〈xp〉 + 〈px〉 2〈p 〉 −1 0
which is the common form of the uncertainty principle. Based on such covariance
matrices, one can compute all properties one can dream of, in particular quantities
such as entropies. Fermonic covariance matrices are dened as follows.
Fermionic covariance matrices: The 2n × 2n anti-symmetric fermionic covari-

ance matrices γ = −γ T are dened as
γj,k = 2itr[ρrj rk ] − iδj,k . (8.68)
It satises
iγ ≤ . (8.69)
Similar to bosons, one can “decouple fermions”, in a way that reminds of the nor-
mal mode decomposition. Simple as this may look, it is at the basis of a huge body of
material in condensed matter physics.
Normal form of fermionic covariance matrices: Every fermionic covariance ma-

trix γ ∈ R2n×2n can be broght into the form
n [ ]
⊕ 0 dj
OγOT = , (8.70)
−dj 0
j=1
with a suitable O ∈ SO(2n), where now dj ∈ [0, 1].
Again, say, tight binding models can be solved in this form. The above fermionic
Bogoliubov transformations are all special cases of this more general formalism.

Advanced QM Lecture Notes

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Advanced QM Lecture Notes

Uploaded by

Copyright:

Available Formats

2

[1] A. Galindo and P. Pascual, Quantum mechanics II (Springer, 1989).

[2] A. L. Fetter und J. D. Valecka, Quantum theory of many-particle systems

[5] F. Schwabl, Fortgeschrittene Quantenmechanik (Springer, 2005).

Elements of quantum theory

Every physical theory is supposed to make predictions on future measurement out-

1.1 Quantum states and observables

1.1.1 Pure quantum states

This could be a superposition

|ψ〉 = α|0〉 + β|1〉 (1.1)

with complex α, β, normalized as

|α|2 + |β|2 = 1. (1.2)

B = {|0〉, . . . , |d − 1〉}. (1.4)

In this basis, every state vector can be expressed as

〈j|k〉 = δj,k , (1.6)

All this applies to so-called nite-dimensional quantum systems, where d is an integer.

Their scalar product is then

Observables: Observable are Hermitian operations in a Hilbert space. Expectation

〈A〉 = 〈ψ|A|ψ〉. (1.13)

Diagonalization: Every Hermitian operator can be diagonalized in that there exists

can, needless to say, be expressed in this basis, as

Their trace is given by

1.1.3 Mixed quantum states

Such a situation is sometimes referred to as a mixed ensemble. How do we incorporate

We immediately nd some properties of such an operator: We obviously have that

tr(ρ2 ) = tr(|ψ〉〈ψ|ψ〉〈ψ|) = tr(|ψ〉〈ψ|) = tr(ρ) = 1. (1.29)

〈A〉 = 〈ψ|A|ψ〉 = tr(A|ψ〉〈ψ|) = tr(Aρ). (1.31)

Density operator of a mixed ensemble: Consider the situation of preparing |ψj 〉,

we also have that

Pure states are those density operators for which

those can be represented by state vectors |ψ〉 ∈ H as

This is a good moment to discuss a number of examples. Let us go back to our

obviously satisfying tr(ρ) = 1. Generally, if we have probabilities p0 and p1 to prepare

with matrix representation

we get exactly same same value for all observables in case of

1.2 Measurement postulate

Measurement postulate: Let A be an observable with spectral decomposition

1.3 Composite quantum systems

H1 = span{|0〉, . . . , |d1 − 1〉}. (1.58)

It is spanned by the orthonormal basis vectors

{|j〉 ⊗ |k〉 : j = 0, . . . , d1 − 1; k = 0, . . . , d2 − 1} . (1.61)

Such basis elements of tensor products are sometimes also written as

|ψ1 〉 = αj |j〉 (1.63)

and an arbitrary superposition of a state vector from H2 is

|ψ2 〉 = βj |j〉, (1.64)

|ψ〉 = γj,k |j〉 ⊗ |k〉, (1.65)

Similarly, an arbitrary linear operator can be decomposed as

1.4 Time evolution

Schrödinger equation: State vectors of pure states evolve in time according to

Making use of the unitary time evolution operator

U (t) = e−iHt/~ , (1.73)

U (t)U † (t) = U † (t)U (t) = . (1.74)

|ψ(t2 )〉 = U (t2 − t1 )|ψ(t1 )〉. (1.75)

General states, so density operators, evolve according to

ρ(t2 ) = U (t2 − t1 )ρ(t1 )U † (t2 − t1 ). (1.76)

Evolution of observables in the Heisenberg picture: In the Heisenberg picture,

A(t2 − t1 ) = U † (t2 − t1 )AU (t2 − t1 ). (1.77)

Of course, the predictions in both pictures are identical,

1.5 Heisenberg’s uncertainty principle

Taking the absolute value, we obtain

|〈[C, D]〉|2 ≤ 4|〈ψ|CD|ψ〉|2 . (1.85)

Of course, we have for the commutator

[C, D] = [A, B], (1.86)

|〈ψ|CD|ψ〉|2 = |(〈ψ|C) · (D|ψ〉)|2