Exercise Sheet 1: Basics of Quantum Information The-

ory
This exercise sheet tries to teach you some of the basics of quantum information theory.

Density Matrix Formalism

Introduction. There are multiple ways to approach a description of quantum mechanics from
the analytical point of view. Quantum mechanics courses usually rely mostly on the formalism of
quantum states, normally expressed as |ψ⟩, and the Schrödinger equation. For our purposes, we
will, however, mostly use the density matrix formulation of quantum mechanics, which allows for
a simpler treatment of probabilistic mixtures of quantum states. These arise for example when
a quantum system undergoes unwanted random interactions with an environment, introducing
“noise” to a quantum state.
The density matrix formulation starts from the following (incomplete) set of postulates:

(I) Each physical system is associated with a Hilbert space (H, ⟨·|·⟩). The (mixed) state of
a quantum system is described by a non-negative (also called positive semi-definite, this
means that all eigenvalues of the matrix are non-negative), self-adjoint linear operator
with unit trace, i.e. an element of1

D := {ρ ∈ L(H) | ρ = ρ† , ρ ≥ 0, Tr[ρ] = 1}. (1)

Here, ρ ≥ 0 is the notation we use to say that ρ is positive semi-definite.

(II) Observables are represented by Hermitian operators on H. The expectation value of an

observable A in the state ρ is given by ⟨A⟩ρ = Tr[Aρ].

(III) The time-evolution of the state of a quantum system satisfies

dρ
= −i[H, ρ],
dt
where H is the Hamiltonian, the observable associated to the total energy of the system.

2 P. Exercise 1 (Rank one projectors). Show that the set

P = {π ∈ L(H) | π = π † , π 2 = π, rank π = 1} (2)

of orthogonal projectors onto one-dimensional subspaces of H is a subset of D.

Solution
We need to show that any rank one orthogonal projectors π is positive semi-definite and
has trace 1. Since π 2 = π, the only eigenvalues of π are 0 and 1, as a matter of fact if
|λ⟩ is an eigenvector then λ|λ⟩ = π|λ⟩ = π 2 |λ⟩ = λ2 |λ⟩ hence λ2 = λ. This implies π
is positive semi-definite. Since π has rank 1, there can only be one eigenvalue 1, hence
Tr[π] = 1.

Most probably, you have originally learned another definition for quantum states in your
first quantum mechanics course. Namely, pure quantum states are rays of the Hilbert space
H. The rays of a Hilbert space are the equivalence classes of unit vector that only differ by
1
In quantum information theory, it will be sufficient to consider finite-dimensional Hilbert spaces most of the
time. A finite-dimensional Hilbert space is simply a vector space. In infinite dimension there are more subtleties,
but these do not concern us.
 a phase factor. In symbols, we have rays(H) = {|ψ⟩ ∈ H | ∥|ψ⟩∥22 = 1}/ ∼. The equivalence
relation |ψ⟩ ∼ |ϕ⟩ captures the fact that if there exist α ∈ R such that |ψ⟩ = eiα |ϕ⟩, then |ψ⟩
and |ϕ⟩ represent the same state. Often physicists tend to drop the equivalence relation and
talk about unit vectors as quantum states instead of rays.
5 P. Exercise 2 (Pure states). In this exercise, we will show that the set P of Eq. (2) is equivalent
to the set of pure quantum states and use this to derive the time-evolution in the density matrix
formalism from the pure state Schrödinger equation.

2 P. (a) Show that the mapping

[|ψ⟩] 7→ |ψ⟩⟨ψ| (3)

is a bijection between the set P defined in Eq. (2) and rays(H) irrespective of which
representative of [|ψ⟩] is chosen.
Solution
Let [|ψ⟩] ∈ rays(H) and |ψ⟩ be a representative of [|ψ⟩]. We can then map it to
|ψ⟩⟨ψ| ∈ P. This is well-defined since given another representative |ψ ′ ⟩ = eiα |ψ⟩
with α ∈ R we have |ψ ′ ⟩⟨ψ ′ | = eiα |ψ⟩⟨ψ|e−iα = |ψ⟩⟨ψ|. Let π ∈ P. The mapping is
inverted by choosing a normalized vector from the image of π. This choice is unique
up to complex phase which does not matter because of the equivalence relation.

1 P. (b) Show that the mapping of Eq. (3) is the correct one as it reproduces the same expectation
values for any observable A.
Solution
For any representative |ψ⟩ of the ray [|ψ⟩], we have that

⟨A⟩ = ⟨ψ|A|ψ⟩.

Using the same representative under the mapping of Eq. (3), we have

⟨A⟩ = Tr[A|ψ⟩⟨ψ|] = ⟨ψ|A|ψ⟩,

proving the desired result.

2 P. (c) Starting from the Schrödinger equation for pure states, i.e.

d
|ψ⟩ = −iH|ψ⟩ (4)
dt
derive the corresponding evolution equation for density matrices
dρ
= −i[H, ρ]. (5)
dt
Here [A, B] = AB − BA is the commutator of two matrices. (Hint: start by proving this
for ρ = π a pure state, then use linearity.)
 Solution
We start from ρ = π a pure state. As shown in the previous point, π = |ψ⟩⟨ψ| for
some time dependent normalized vector |ψ⟩(t). We have

dπ d d d
= (|ψ⟩⟨ψ|) = (|ψ⟩)⟨ψ| + |ψ⟩ (⟨ψ|).
dt dt dt dt
P
If this looks strange, you can verify that this is true by writing |ψ⟩(t) = i ψi (t)|i⟩,
where |i⟩ are time independent basis vectors, then you need only apply the product
rule to the coefficients ψi , which are just complex valued functions. Using the
Schroedinger equation we get
dπ
= −iH|ψ⟩⟨ψ| + i|ψ⟩⟨ψ|H = −i[H, π].
dt
Any density matrix ρ can be decomposedP as a sum of projectors by the spectral
theorem (see also Exercise 4 below), ρ = i ρi πi . Since the equation is linear, it
holds for any ρ.

We have seen that density matrices describe both mixed and pure quantum states. Let us
define the following function of the state
f : D → R, ρ 7→ Tr[ρ2 ]. (6)
Before we come to the next exercise, we will also introduce the so-called Hilbert-Schmidt inner
product, which endows the space of matrices with an inner product. It is defined as
⟨A, B⟩ := Tr[A† B]. (7)
As any proper inner product, it also obeys a Cauchy-Schwarz inequality:
Tr[A† B]2 ≤ Tr[A† A] Tr[B † B]. (8)
9 P. Exercise 3 (Purity).
3 P. (a) Show that d1 ≤ f (ρ) ≤ 1, where d is the dimension of the Hilbert space H. (Hint: Use the
Cauchy-Schwarz inequality)
Solution
Let {λi }di=1 be the eigenvalues of ρ. We then have that
d
X d
X
2
Tr[ρ ] = λ2i ≤ λi = 1,
i=1 i=1

where we used that for all 0 ≤ λ ≤ 1 we have that λ2 ≤ λ. This proves the upper
bound. For the lower bound, we observe that we can insert an identity and use
Cauchy-Schwarz to obtain

1 = Tr[ρ]2
= Tr[Iρ]2
≤ Tr[I2 ] Tr[ρ2 ]
= Tr[I]f (ρ)
= df (ρ),

which after rearranging yields f (ρ) ≥ d1 .

3 P. (b) Show that f (ρ) = 1 if and only if ρ is a pure state.
Solution
For ρ = |ψ⟩⟨ψ| a pure state, we have that

f (ρ) = Tr[|ψ⟩⟨ψ|ψ⟩⟨ψ|]
= Tr[|ψ⟩⟨ψ|]
= 1.

We recall from the last exercise the for {λi }di=1 the eigenvalues of ρ, we have that
d
X d
X
λ2i ≤ λi = 1.
i=1 i=1

The only way equality can be achieved is if for all i we have λi = λ2i , which can
only happen if λi ∈ {0, 1}. As there can only be a single 1 among the eigenvalues,
we conclude that ρ has to be a projector, and hence a pure state, to fulfill f (ρ) = 1

3 P. (c) What state attains the lower bound f (ρ) = d1 ? Argue that f (ρ) can be seen as a measure
of “purity” of the state ρ.
Solution
It is readily verified that ρ = d1 I is a valid quantum state that has f (ρ) = d1 . We can
find this state by revisiting the argument from exercise (a), considering that the ρ
that saturates (achieves the smallest value) in the Cauchy-Schwarz inequality must
be proportional to the identity I in analogy to the Cauchy-Schwarz inequality for
vectors.
This state is called the maximally mixed state, as it represents a mixture of all
possible pure states of a system in any basis and is thus the least pure state we could
have. Computing f (ρ) thus determines how close a state is to being pure, where
larger values signal larger purity and lower values a higher degree of mixedness.

5 P. Exercise 4 (Decompositions of mixed states).

2 P. (a) Show that every mixed state of a finite-dimensional quantum system can be written as a
convex decomposition of pure states.
Solution
Let ρ ∈ L(H) for some finite-dimensional Hilbert space H. Then ρ is in partic-
ular a self-adjoint linear operator, so by the spectral theorem, ρ has eigenvalues
λ1 , . . . , λd ∈ R with associated normalized eigenvectors |ψ1 ⟩, . . . , |ψd ⟩ ∈ H. In Dirac
bra-ket notation, this means
d
X
ρ= λi |ψi ⟩⟨ψi | .
i=1

As ρ ≥ 0, we have λi ≥ 0 for all 1 ≤ i ≤ d. Moreover, we have di=1 λi = Tr[ρ] = 1.

P
Thus, the λi are non-negative and sum to 1, so the above decomposition is in fact
a convex decomposition into pure states.

2 P. (b) Consider the following two (macroscopically different) preparation schemes of a large
number of polarised photons:
 Preparation A. For each photon we toss a fair coin. Depending on whether we get
head or tail, we prepare the photon to have either vertical or horizontal linear polarisation.
Preparation B. For each photon we toss a fair coin. Depending on whether we get
head or tail, we prepare the photon to have either left-handed or right-handed circular
polarisation.
Note: You can simply think of the polarization of the light as a binary variable and of the
polarization axis as a local basis. That is, the vertical and horizontal linear polarizations
may be identified with the |0⟩ and |1⟩ eigenstates of the Z operator. Likewise you may
interpret the left- and right handed circular polarizations as the |+⟩ = |0⟩+|1⟩
√
2
and |−⟩ =
|0⟩−|1⟩
√
2
eigenstates of the X operator.
(m) (m)
Write down the density matrices ρA and ρB describing the mixed quantum states
obtained after m rounds of Preparations A and B, respectively.
Solution
A single round of Preparation A produces the mixed state
1 1 1
ρA = |0⟩⟨0| + |1⟩⟨1| = 12 .
2 2 2
Similarly, a single round of Preparation B produces the mixed state
1 1 1
ρB = |+⟩⟨+| + |−⟩⟨−| = 12 = ρA .
2 2 2
As the different rounds are independent (since the coin tosses are independent), the
quantum states describing m rounds are given by
 ⊗m
(m) 1 1 (m)
ρA = 12 = m 12m = ρB .
2 2

1 P. (c) Use the result of (b) to argue that, having only access to the photons produced by the
preparation procedures in (b), we cannot distinguish whether Preparation A or Prepara-
tion B was used.
Solution
(m) (m)
According to our result in (b), no matter how large m is, we have ρA = ρB . Thus,
the m photons coming from the preparation procedures are described by exactly the
same mixed quantum state. Hence, there is no measurement that distinguishes the
two preparation procedures when given only access to the photons.

4 P. Bonus Exercise 1. Argue that if it were possible to distinguish Preparation A from Prepa-
ration B (from the previous exercise) by measuring the photons, then this could be used to
communicate a bit of information without actually sending any (classical or quantum) infor-
mation carrier. Which fundamental physical principle would this violate? (Hint: What is the
reduced state of the maximally entangled state |Ω⟩ = √12 (|00⟩ + |11⟩) = √12 (|++⟩ + |−−⟩)?)
 Solution
Protocol: EPR setting with Bell state |Ω⟩ = √12 (|00⟩ + |11⟩) = √12 (|++⟩ + |−−⟩)
Bob chooses a measurement setting, X or Z, and measures his half of the EPR state
without observing the measurement outcome. If Bob chooses to measure in the X-basis,
the post-measurement state on Alice’s side reads 21 (|+⟩⟨+| + |−⟩⟨−|). If Bob chooses to
measure in the Z-basis, the post-measurement state on Alice’s side reads 12 (|0⟩⟨0| + |1⟩⟨1|).
If Alice had a way of distinguishing the two preparation procedures from a single photon,
the two parties could have communicated a bit (encoded as {X, Z}) without Bob sending
any message to Alice. This would violate locality. More generally, if Alice had a way
of distinguishing the two preparation procedures from m photons, then Alice and Bob
could start from m maximally entangled states to communicate a single bit without Bob
sending any message to Alice, again violating locality.

Composite Quantum Systems

Next, we will see that the generalization to density matrices is a necessary one if we want to
study subsystems. Consider a bipartite system AB with Hilbert space H = CdA ⊗ CdB and
an observable that is only supported on the subsystem A as OA ⊗ IB . We will see that the
restriction to a subsystem is described by the partial trace: For a a linear operator (matrix)
M : H → H on the composite system AB, the partial trace with respect to the system B is
defined as
dB
X
TrB [M ] = (IA ⊗ ⟨j|B )M (IA ⊗ |j⟩B ), (9)
j=1

where {|j⟩B } is an arbitrary orthonormal basis for CdB (as with the trace, this definition is
independent of the particular choice of the basis). In quantum information theory, we usually
say “we trace out the system B”.
10 P. Exercise 5.
2 P. (a) As a technical prerequisite, prove that a self-adjoint operator is positive semi-definite, i.e.
has only non-negative eigenvalues, if and only if ⟨v|M |v⟩ ≥ 0 for all |v⟩.
Solution
If ⟨v|M |v⟩ ≥ 0 for all |v⟩, this especially holds for the eigenvectors of M , and
hence the eigenvalues are all non-negative, establishing one direction.P The other
direction is verified by direct computation by expanding both M = di=1 mi |i⟩⟨i|
and |v⟩ = di=1 vi |i⟩ in the eigenbasis of M . We then have
P

d
! d 
d
!
X X X
∗
⟨v|M |v⟩ = vi ⟨i|  mj |j⟩⟨j| vk |k⟩
i=1 j=1 k=1
d
X
= vi∗ vi mi
i=1
Xd
= |vi |2 mi .
i=1

As all |vi |2 ≥ 0 and by assumption mi ≥ 0, the desired result follows.

3 P. (b) Show that the partial trace of a state with respect to the system B (density operator) is
a valid state on the subsystem A.
 Solution
We have to verify three properties: self-adjointness, unit trace and positive-
semidefiniteness. We first observe that taking the adjoint is additive and hence:
 †
dB
X
(TrB (ρ))† =  (IA ⊗ ⟨j|)ρ(IA ⊗ |j⟩)
j=1
dB
X
= ((IA ⊗ ⟨j|)ρ(IA ⊗ |j⟩))†
j=1
dB
X
= (IA ⊗ ⟨j|)ρ† (IA ⊗ |j⟩)
j=1
dB
X
= (IA ⊗ ⟨j|)ρ(IA ⊗ |j⟩)
j=1

= TrB (ρ).

Next, we prove that the trace is preserved under the partial trace:
 
XdB
Tr(TrB (ρ)) = Tr  (IA ⊗ ⟨j|)ρ(IA ⊗ |j⟩)
j=1
dA
X dB
X
= ⟨i| (IA ⊗ ⟨j|)ρ(IA ⊗ |j⟩)|i⟩
i=1 j=1
dA X
X dB
= (⟨i| ⊗ ⟨j|)ρ(|i⟩ ⊗ |j⟩)
i=1 j=1

= Tr(ρ) = 1.

Here, the last line follows from observing that |i⟩⊗|j⟩ is by definition an orthonormal
basis for H. For positivity, we use consider the expectation value
X
⟨ψ|A TrB (ρ)|ψ⟩A = (⟨ψ|A ⊗ ⟨j|B )ρ(|ψ⟩A ⊗ |j⟩B ) ≥ 0
j

since ρ is positive semi-definite by assumption.

2 P. (c) Prove that for any state ρAB we have

Tr[ρAB (OA ⊗ IB )] = Tr[TrB [ρAB ]OA ]. (10)

for all observables OA . That is, the partial trace is the reduced state on the subsystem A.
 Solution
First notice that for any operator XAB , Tr[XAB ] = TrA [TrB [XAB ]]. Now, let |i⟩B
be a basis of HB , then
dB
X
TrB [ρAB (OA ⊗ IB )] = (IA ⊗ ⟨i|B )ρAB (OA ⊗ IB )(IA ⊗ |i⟩B )
i=1
XdB
= (IA ⊗ ⟨i|B )ρAB (OA ⊗ |i⟩B )
i=1
XdB
= (IA ⊗ ⟨i|B )ρAB (IA ⊗ |i⟩B )OA = TrB (ρAB )OA
i=1

where in the second to last equality we used (OA ⊗ IB )(IA ⊗ |i⟩B ) = (IA ⊗ |i⟩B )OA .
Here, (IA ⊗ |i⟩B ) can be seen as a map HA → HA ⊗ HB , |ψ⟩A 7→ |ψ⟩A ⊗ |i⟩B .

3 P. (d) Reduced states of pure states are not necessarily pure. Let dA = dB = d. Show that there
is no pure state |ψA ⟩⟨ψA | acting on A that satisfies

Tr[ρAB (OA ⊗ IB )] = Tr[|ψA ⟩⟨ψA |OA ] (11)

for ρAB = |ΩAB ⟩⟨ΩAB | and all observables OA . Here,

d
1 X
|Ω⟩ := d− 2 |j, j⟩ (12)
j=1

is the maximally entangled state.

Solution
The statement we need to prove is the following: For any pure state |ψA ⟩ acting on
A, there exists an observable OA such that

Tr[ρAB (OA ⊗ IB )] ̸= Tr[|ψA ⟩⟨ψA |OA ], (13)

for ρAB the maximally mixed state introduced in the exercise.

We show it suffices to pick OA = |ψA ⟩⟨ψA |.
The r.h.s. of Eq. (11) is easy to compute:

Tr[|ψA ⟩⟨ψA |ψA ⟩⟨ψA |] = |⟨ψA |ψA ⟩|2 = 1. (14)

For the l.h.s, we use Eq. (10). For full generality, we allow OA = |ϕ⟩⟨ϕ| be a projector
to any pure state on A:
 
d X d
1 X
Tr[ρAB (OA ⊗ IB )] = Tr  |j, j⟩⟨k, k|(|ϕ⟩⟨ϕ| ⊗ I) (15)
d
j=1 k=1
 !
d X d d
1 X X
= Tr  (|j⟩⟨k| ⊗ |j⟩⟨k|) |ϕ⟩⟨ϕ| ⊗ |i⟩⟨i|  (16)
d
j=1 k=1 i=1
 
d d X d
X 1 X
= Tr  |j⟩⟨k|ϕ⟩⟨ϕ| ⊗ ⟨i|j⟩⟨k|i⟩ (17)
d
i=1 j=1 k=1
d  
X 1
= Tr |i⟩⟨i|ϕ⟩⟨ϕ| (18)
d
i=1
d
1X
= |⟨i|ϕ⟩|2 (19)
d
i=1
1
= , (20)
d

where we exploited that for any orthonormal basis di=1 |⟨i|ϕ⟩|2 = ∥|ϕ⟩∥22 = 1. So
P
we see that the l.h.s. equals 1/d for any pure state projector. In particular, this is
also true for our choice OA = |ψA ⟩⟨ψA |.
In all, we saw that the l.h.s. is equal to 1/d, whereas the r.h.s. is equal to 1, thus
solving the exercise.

Total Points: 30 (+4)