Quantum Information

(ELEC-C9440)
Lecture 1-2

Lauri Ylinen1

Aalto University

Spring 2024

1
Slides courtesy of Arttu Pönni, Matti Raasakka, with some modifications by LY
Quantum information

Quantum information and computation is the study of information

processing tasks that can be accomplished by using quantum systems.
Important questions:
▶ How much classical information can be transmitted over a
quantum channel?
▶ How about quantum information instead?
▶ Or how might noise affect the channel’s capacity?
Why do we need physics for studying information? Because information
is physical:
▶ Information is encoded in physical systems
▶ It is processed by physical devices acting on those systems
▶ Measurement in general disturbs the system and affects
subsequent measurements (uncertainty principle)
▶ Erasure of information requires energy (Landauer’s principle)
▶ Perfect copying of information is forbidden by quantum mechanics
(no-cloning theorem)
▶ Etc.
Quantum computation

Quantum computers are machines which execute quantum circuits

(made up of wires and quantum logic gates) to process information.
Why bother with quantum computation when we already have classical
computers?
▶ There are problems that can be solved efficiently on a quantum
computer but not on a classical computer (e.g. Shor’s algorithm
for integer factorization)
▶ If you want to simulate quantum mechanics efficiently, your
computer needs to be quantum as well! Important applications in
chemistry
▶ Some day energy efficiency of quantum computers might be better
than traditional computers?
Technology for building quantum computers still has a long way to go,
but the rate of progress is fast.
In the meantime we can
▶ use the small quantum computers that are available
▶ use simulators on classical computers to study quantum algorithms
▶ develop the theory
 Linear algebra

Lightning fast recap of necessary algebra

Notation
Vectors and operators

Typically we will use the Dirac notation in which vectors are denoted by
|v ⟩, but sometimes it is useful to write out the components explicitly
w.r.t. some basis. For example, if V = C2 , then
 
c1
|v ⟩ = = c1 |v1 ⟩ + c2 |v2 ⟩ ,
c2

where c1 and c2 are the components of |v ⟩ in the basis |v1 ⟩ = (1, 0)T ,
|v2 ⟩ = (0, 1)T . Linear operators act on these vectors as
!
X X
A |v ⟩ = A ci |vi ⟩ = ci A |vi ⟩ .
i i

A linear operator A can be written as a matrix in some basis on V .

Then the action of A on vectors is given by the usual matrix-vector
multiplication
    
a1,1 a1,2 c1 a1,1 c1 + a1,2 c2
A |v ⟩ = = .
a2,1 a2,2 c2 a2,1 c1 + a2,2 c2
| {z } | {z }
A |v ⟩
Tensor products
We will use the tensor product to combine smaller vector spaces into
larger ones. Let V and W be vector spaces with bases {|i⟩}ni=1 and
{|j⟩}m
j=1 , respectively. Then V ⊗ W is a vector space such that:

1. Vectors {|i⟩ ⊗ |j⟩} form a basis of V ⊗ W .

2. The tensor product ⊗ is bilinear: For c ∈ C, v , ve ∈ V , and
e ∈ W the following hold:
w, w

(c |v ⟩) ⊗ |w ⟩ = c(|v ⟩ ⊗ |w ⟩),
|v ⟩ ⊗ (c |w ⟩) = c(|v ⟩ ⊗ |w ⟩),
(|v ⟩ + |e
v ⟩) ⊗ |w ⟩ = |v ⟩ ⊗ |w ⟩ + |e
v ⟩ ⊗ |w ⟩ ,
|v ⟩ ⊗ (|w ⟩ + |w
e ⟩) = |v ⟩ ⊗ |w ⟩ + |v ⟩ ⊗ |w
e⟩ .

If V and W are inner product spaces and {|i⟩}ni=1 and {|j⟩}m

j=1 are
orthonormal bases, then V ⊗ W has an inner product so that {|i⟩ ⊗ |j⟩}
is an orthonormal basis.
Often we use shorthand notation |ij⟩ = |i, j⟩ = |i⟩ ⊗ |j⟩.
For example if n = m = 2, then a vector |ψ⟩ ∈ V ⊗ W can be written as

|ψ⟩ = c00 |00⟩ + c01 |01⟩ + c10 |10⟩ + c11 |11⟩ ,

where cij ∈ C are scalars.

Component notation for tensor products

Let us still consider the case n = m = 2. We order the basis {|ij⟩}2i,j=1

in the lexicographic order:

|00⟩ , |01⟩ , |10⟩ , |11⟩ .

In the component notation, the tensor product of vectors can be

calculated as
    
c ac
 a d   ad 
   
a c
⊗ =   =   .
b d  c   bc 
b
d bd

The basis vectors |ij⟩ = |i⟩ ⊗ |j⟩ would be

       
1 0 0 0
0
 , |01⟩ = 1 , |10⟩ = 0 , |11⟩ = 0 .
     
|00⟩ = 0 0 1 0
0 0 0 1
Tensor products of operators

Tensor products of operators work similarly. Consider the operators

A : V → V and B : W → W . Their tensor product operates as

(A ⊗ B)(|v ⟩ ⊗ |w ⟩) = (A |v ⟩) ⊗ (B |w ⟩) .

If A and B have the matrix representations

   
a1,1 a1,2 b1,1 b1,2
A= , B= ,
a2,1 a2,2 b2,1 b2,2

then the operator A ⊗ B could be written

    
b1,1 b1,2 b1,1 b1,2
a1,1 b2,1 b2,2 a1,2
b2,1 b2,2 

A⊗B =  
 b1,1 b1,2 b1,1 b1,2 
a2,1 a2,2
b2,1 b2,2 b2,1 b2,2
 
a1,1 b1,1 a1,1 b1,2 a1,2 b1,1 a1,2 b1,2
a1,1 b2,1 a1,1 b2,2 a1,2 b2,1 a1,2 b2,2 
= a2,1 b1,1 a2,1 b1,2 a2,2 b1,1 a2,2 b1,2  .


a2,1 b2,1 a2,1 b2,2 a2,2 b2,1 a2,2 b2,2

Eigenvalues and vectors
A scalar λ ∈ C is an eigenvalue of an operator A : V → V if there
exists a non-zero vector |v ⟩ such that

A |v ⟩ = λ |v ⟩ .

One can solve for the eigenvalues by noticing that for nonzero |v ⟩

A |v ⟩ = λ |v ⟩ ⇐⇒ (A − λI) |v ⟩ = 0 ⇒ det(A − λI) = 0 .

The rightmost equation is known as the characteristic equation and it

can be solved for the eigenvalues λ. Then the corresponding
eigenvectors can be found for each eigenvalue λi by solving the linear
system of equations A |vi ⟩ = λi |vi ⟩ for the components of |vi ⟩.
Spectral decomposition: Any normal operator (A† A = AA† ) is diagonal
w.r.t. some orthogonal basis. Any such operator A can be written
X
A= λi |vi ⟩⟨vi | ,
i

where {|vi ⟩} is an orthonormal basis consisting of eigenvectors of A (vi

is an eigenvector corresponding to eigenvalue λi ). Here |vi ⟩⟨vi | is the
linear operator V → V defined by

|vi ⟩⟨vi | (|ψ⟩) = ⟨vi |ψ⟩ |vi ⟩

Operator functions

Later we will encounter expressions such as log A, where A is a linear

operator, so we will need a way to define what it means to apply a
function to an operator. In general, for any function on complex
numbers

f : C 7→ C

we can define a corresponding function on normal operators. The

definition uses the spectral
P decomposition of A: If the spectral
decomposition of A is i λi |vi ⟩⟨vi |, then
X
f (A) ≡ f (λi ) |vi ⟩⟨vi | .
i

So in practice to apply a function f to an operator A, one will first

compute the eigenvalues λi and form an orthonormal basis of
eigenvectors {|vi ⟩} of V , and then apply the complex function using
the above definition.
 Quantum mechanics

Postulates, quantum bits, superdense coding

Postulate 1: States

Postulate (States)
Associated to any isolated physical system is a complex vector
space (a Hilbert space H) known as the state space. The system
is completely described by a unit vector |ψ⟩ ∈ H

The states are normalized to unit vectors: ⟨ψ|ψ⟩ = 1

We often expand states in some orthogonal basis {|i⟩} s.t. ⟨i|j⟩ = δij
X
|ψ⟩ = ai |i⟩
i

Overall phase has no physical significance: |ψ⟩ and e iφ |ψ⟩ are the
same physical state for every φ ∈ R. Relative phases are however
physically significant: the states a |ψ⟩ + b |ϕ⟩ and a |ψ⟩ + e iφ b |ϕ⟩ are
physically different.
Postulate 2: Evolution of states

Postulate (Evolution of states)

The evolution of a closed quantum system is described by an
unitary transformation. That is, if |ψt ⟩ and |ψt ′ ⟩ are the states of
a system at two different times, they are related by

|ψt ′ ⟩ = U |ψt ⟩ ,

where U is a unitary operator.

In our use cases we can always represent U as a matrix. Remember

that a matrix is unitary if U † U = I, or in other words U −1 = U † . A
common way to state this postulate is to give the Schrödinger equation

d |ψ⟩
iℏ = H |ψ⟩ ,
dt
which gives the evolution of our quantum state in terms of its
Hamiltonian H.
Postulate 3: Quantum measurement

Postulate (Quantum measurement)

Any measurement is described by a collection of measurement
operators {Mm } acting on the state space, one operator for each
possible measurement outcome m. The probability of outcome m
is
†
p(m) = ⟨ψ| Mm Mm |ψ⟩

and the state collapses to

Mm |ψ⟩
|ψ⟩ 7→ p .
p(m)

These measurement operators satisfy the completeness relation

P †
m Mm Mm = I which guarantees that probabilities sum to one
!
X X X
† †
p(m) = ⟨ψ| Mm Mm |ψ⟩ = ⟨ψ| Mm Mm |ψ⟩ = 1 .
m m m
Postulate 3: Quantum measurement

We often talk about measurements in terms of observables.

Observables represent properties of physical systems which can be
measured (e.g. position, momentum, charge, etc). Mathematically,
observables are Hermitian/self-adjoint operators (A† = A). By the
spectral decomposition of A,
X
A= m |m⟩⟨m| ,
m

where |m⟩ is an eigenstate of A with eigenvalue m. This observable

corresponds to the measurement operators Mm = |m⟩⟨m|. Observables
make it easy to compute expected values of its corresponding
measurement:
X X
E(A) = mp(m) = m ⟨ψ|m⟩ ⟨m|ψ⟩
m m
!
X
= ⟨ψ| m |m⟩⟨m| |ψ⟩ = ⟨ψ|A|ψ⟩ .
m
Postulate 4: Composite systems

Postulate (Composite systems)

The state space of a composite system is the tensor product of
the state spaces of its components.

So, if we have n systems, with the ith system prepared in the state |ψi ⟩,
then the joint system is in the product state |ψ1 ⟩ ⊗ |ψ2 ⟩ ⊗ . . . ⊗ |ψn ⟩.
Accordingly, if we have operators Ui each acting on the ith subsystem,
their joint action is described by the operator

U = U1 ⊗ U2 ⊗ . . . ⊗ Un .

Note that in the case of states, often one omits the ⊗-sign:

|ψ1 ⟩ ⊗ |ψ2 ⟩ ⊗ · · · ⊗ |ψn ⟩ ≡ |ψ1 ⟩ |ψ2 ⟩ . . . |ψn ⟩ .

Quantum bits

Just as bits are the fundamental building blocks of classical

computation and information, quantum bits (qubits) are the analogous
concept in quantum computation and information. A qubit is a unit
vector in the two-dimensional Hilbert space C2 . In terms of the
computational basis states {|0⟩ , |1⟩} the state of a general qubit is

|ψ⟩ = α |0⟩ + β |1⟩ , |α|2 + |β|2 = 1

⟨0|0⟩ = ⟨1|1⟩ = 1 , ⟨0|1⟩ = 0 .

Whereas a bit has two states (0 and 1), the qubit has infinitely many
states since any linear combination of |0⟩ and |1⟩ yields a valid
quantum state.
One often measures qubits in the computational basis, that is, uses the
measurement operators

M0 = |0⟩⟨0| M1 = |1⟩⟨1| .

According to Postulate 3, measurement would give 0 with probability

|α|2 and 1 with probability |β|2 .
Quantum bits - Bloch sphere

A general qubit state has two complex parameters or four real

parameters. Because of the normalization condition |α|2 + |β|2 = 1,
only 3 of those parameters are free. Let’s then parametrize our state
with
θ θ
α = e iγ cos β = e i(γ+φ) sin .
2 2
This parametrization automatically satisfies the normalization
condition. Then our qubit is
 
θ θ
|ψ⟩ = e iγ cos |0⟩ + e iφ sin |1⟩ .
2 2

Notice that e iγ is a global phase and therefore isn’t physically

significant. Therefore the physically distinct qubit states can be written
in terms of two real parameters, θ and φ:

θ θ
|ψ⟩ = cos |0⟩ + e iφ sin |1⟩ .
2 2
Quantum bits - Bloch sphere

The Bloch sphere is useful for visualizing the state of a qubit. We

obtain the Bloch sphere by interpreting the parameters θ and φ as
coordinates on a sphere.

θ = 0 7→ |0⟩
θ = π 7→ |1⟩
π 1
θ= , ϕ = 0 7→ √ (|0⟩ + |1⟩)
2 2
π 1
θ = , ϕ = π 7→ √ (|0⟩ − |1⟩)
2 2
π π 1
θ = ,ϕ = 7→ √ (|0⟩ + i |1⟩)
2 2 2
π 3π 1
θ = ,ϕ = 7→ √ (|0⟩ − i |1⟩)
2 2 2
We will see that the states along the x, y , and z -axes corresponds to
eigenstates of the Pauli X , Y , and Z -matrices.
Multiple qubits
Now consider a system of 2 qubits. By Postulate 4, the state space is
now 4 dimensional and is spanned by the computational basis states
|00⟩, |01⟩, |10⟩, and |11⟩. Now a general 2 qubit state can be written

|ψ⟩ = α00 |00⟩ + α01 |01⟩ + α10 |10⟩ + α11 |11⟩ ,

where |01⟩ = |0⟩ ⊗ |1⟩ etc. This can easily be generalized for n qubits.
n
The state space would be C2 ⊗ C2 ⊗ · · · ⊗ C2 ∼= C2 with states
n
2X −1
|ψ⟩ = αx |x⟩ ,
x=0

where we have used the following useful shorthand for computational

basis states: consider a basis state |x1 ⟩ ⊗ |x2 ⟩ ⊗ · · · ⊗ |xn ⟩, where each
xi ∈ {0, 1}. This is the binary representation of the integer

x = x1 2n−1 + x2 2n−2 + · · · + xn 20 .

Now if we define

|x⟩ ≡ |x1 ⟩ ⊗ |x2 ⟩ ⊗ · · · ⊗ |xn ⟩ ,

we can write down multi-qubit basis states very easily. For example,
|6251⟩ instead of |1100001101011⟩.
Distinguishability of quantum states
A consequence of Postulate 3 is that quantum states can’t necessarily
be distinguished from each other with perfect accuracy. This is an
important difference between classical and quantum information.
Example: You are given a quantum state |ψ⟩ with the promise that it is
either |ψ0 ⟩ = |0⟩ or |ψ1 ⟩ = |1⟩. Can you determine which it is? Yes you
can, because the two possibilities are orthogonal ⟨ψ0 |ψ1 ⟩ = 0. You
would make the measurement

M0 = |0⟩⟨0| M1 = |1⟩⟨1|

and you would get the outcome 0 if and only if the state was
|ψ⟩ = |ψ0 ⟩. √
Example: Consider the same problem with |ψ0 ⟩ = (|0⟩ + |1⟩)/ 2 and
|ψ1 ⟩ = |1⟩. Now if you make the same measurement you would get

1 1
If |ψ⟩ = |ψ0 ⟩ : p(0) = ⟨ψ0 |0⟩ ⟨0|ψ0 ⟩ = p(1) = ⟨ψ0 |1⟩ ⟨1|ψ0 ⟩ =
2 2

If |ψ⟩ = |ψ1 ⟩ : p(0) = ⟨ψ1 |0⟩ ⟨0|ψ1 ⟩ = 0 p(1) = ⟨ψ1 |1⟩ ⟨1|ψ1 ⟩ = 1 ,

therefore if your measurement gave the result 0, then you know you
were given ψ0 . However if the result is 1, then you can’t know which
state you had.
Example: superdense coding

One bit can be encoded in a qubit, one can always encode

x1 x2 . . . xn 7→ |x1 x2 . . . xn ⟩ for n bits encoded in n qubits. Qubits can
however store even more information with superdense coding.

Alice and Bob share a Bell state Φ+ . Bell states are

1 1
Φ+ = √ (|00⟩ + |11⟩) Φ− = √ (|00⟩ − |11⟩)
2 2
+ 1 − 1
Ψ = √ (|01⟩ + |10⟩) Ψ = √ (|01⟩ − |10⟩) .
2 2
These are all orthogonal and such that Alice can transform the initial
state to each one at will.
Example: superdense coding

Suppose Alice wants to send two bits of information x1 x2 . In

superdense coding she would do the following:
1. If x1 = 1, then bit flip |0⟩ ↔ |1⟩
2. If x2 = 1, then phase flip |0⟩ 7→ |0⟩ and |1⟩ 7→ − |1⟩.
3. If x1 = x2 = 1, then do both: first bit flip and then phase flip
4. Otherwise do nothing
In other words the two-qubit state after Alice’s manipulation is

x1 x2 = 00 : Φ+ x1 x2 = 10 : Ψ+
−
x1 x2 = 01 : Φ x1 x2 = 11 : Ψ− .

Then Alice mails her qubit to Bob. Finally Bob can measure both
qubits and determine which Bell state he has and therefore which
two-bit string Alice wanted to send. Note that distinguishing between
the possible states is always possible with perfect fidelity since they are
orthogonal.
Quantum computation

Quantum circuits, quantum gates

Quantum circuits

As stated by Postulate 2, quantum states evolve unitarily. Since we are

interested in manipulating qubits, we need a useful way to describe
unitary transformations acting on qubits. Quantum circuits give a
pictorial representation of unitary operators which consist of:
1. Wires representing individual qubits
2. Boxes representing simple unitary operators
3. Symbols representing measurement
For example,
Reversibility

Important difference to classical computation is that quantum circuits

are reversible (if no measurements are made). It means that no
information is lost during computation. For example, consider the
classical AND-operation which maps two bits to one bit
a b a AND b
0 0 0
0 1 0
1 0 0
1 1 1
Only looking at the output bit, one can’t deduce what was the input:
irreversible and information is lost. A quantum circuit can always be
reversed by reading it backwards and replacing gates with their
conjugates:
▶ Read the circuit right-to-left
▶ Replace each gate U 7→ U †
This produces a circuit which perfectly undoes the effect of the original
circuit.
Hilbert space size, simulation by classical computers

The state space of an n-qubit system is

n
C ⊗ C ⊗ ··· ⊗ C ∼ C2 ,
| {z }=
n times

which has dimension 2n . The straightforward way to store the state of

an n-qubit system is to store the 2n amplitudes of the computational
basis states, but even for n = 500 the number of amplitudes is more
than the number of atoms in the universe!
Common one-qubit gates

All matrices are in the computational Z -basis.

Common two-qubit gates

All matrices are in the computational Z -basis.

Matrix representation of quantum gates
Quantum circuits are read from left to right. Note that operator
expressions work differently: rightmost operator acts first! For example,
the circuit

would correspond to the operator

U = SHX
U |q0 ⟩ = SHX |q0 ⟩ .

When we for example “apply X on the 2nd qubit”

we mean that we act on our qubits with:

U =I⊗X ⊗I
Examples of gate actions

Let’s see a few examples of how different gates act on qubits.

Another useful example is the action of H ⊗n on |0⟩⊗n :

 ⊗3
1
H ⊗3 |0⟩⊗3 = √ (|0⟩ + |1⟩)
2
1
= 3/2 (|000⟩ + |001⟩ + |010⟩ + |011⟩
2
+ |100⟩ + |101⟩ + |110⟩ + |111⟩) .

For n qubits, one can use Hadamards to easily create an equal

superposition of all computational basis states.
Pauli matrices
The Pauli matrices often occur in quantum algorithms. They are
   
1 0 0 1
σ0 ≡ I ≡ , σ1 ≡ σx ≡ X ≡
0 1 1 0
   
0 −i 1 0
σ2 ≡ σy ≡ Y ≡ , σ3 ≡ σz ≡ Z = .
i 0 0 −1

They are both Hermitian (U † = U), unitary (U † U = I). All of them

(except the identity) are traceless (tr σa = 0) and have eigenvalues ±1.
Their eigenvectors are
1 1
X : |+⟩ = √ (|0⟩ + |1⟩) , |−⟩ = √ (|0⟩ − |1⟩)
2 2
1 1
Y : |i⟩ = √ (|0⟩ + i |1⟩) , |−i⟩ = √ (|0⟩ − i |1⟩)
2 2
Z : |0⟩ , |1⟩ .

They form a basis in the space of 2x2 matrices, so any 2x2 matrix can
be written as

H = a0 σ0 + a1 σ1 + a2 σ2 + a3 σ3 ,

for some coefficients ai ∈ C.

Reordering gates, linearity

There is a certain amount of freedom in reordering gates in quantum

circuits. For example, all of these circuits are equivalent

because

(X ⊗ I)(I ⊗ Y ) = X ⊗ Y = (I ⊗ Y )(X ⊗ I) .

In other words, the X and Y gates commute because they act on

different wires/qubits.
Notice that in general we need to only specify the action of a gate on
the computational basis states. If this is known, then we automatically
know how the gate acts on arbitrary superpositions
n n
2X −1 2X −1
U αx |x⟩ = αx U |x⟩ .
x=0 x=0
Controlled gates

In traditional computation, flow control structures are vitally important:

for example, the IF-statement. The same is true in quantum
computation. Controlled operations is how this is accomplished in
quantum computing. A general controlled unitary is drawn

The first qubit is the control qubit and the second is the target qubit.
The action of the above circuit is

|c⟩ ⊗ |t⟩ 7→ |c⟩ ⊗ U c |t⟩ ,

that is, the one-qubit unitary U is applied on the target qubit |t⟩ if and
only if the control qubit is set |c⟩ = |1⟩.
Controlled gates
One of the most common controlled gates is the
controlled-X /controlled-NOT gate.

The traditional notation ⊕ refers to the fact that if the control qubit is
set, then the second qubit is combined with the first by the
XOR-operation
|c⟩ ⊗ |t⟩ 7→ |c⟩ ⊗ |c ⊕ t⟩ .
In other words, the target is flipped if the control is set. Another
important controlled gate is the controlled-Z:

The notation is symmetric because in this case it doesn’t matter which

wire is the control and which is the target because a minus-sign is
produced only when both qubits are set

c
|c⟩ ⊗ |t⟩ 7→ |c⟩ ⊗ Z c |t⟩ = |c⟩ ⊗ (−1)t |t⟩ = (−1)ct |c⟩ ⊗ |t⟩ .
Controlled gates

In general, we can have unitaries controlled on any number of qubits:

Now the unitary U is applied on the last 3 qubits only if all of the first
4 qubits are set

|x1 x2 . . . xn ⟩ |ψ⟩ 7→ |x1 x2 . . . xn ⟩ U x1 x2 ...xn |ψ⟩ .

Universal quantum computation

Arbitrary classical programs can be built from just a few basic gates:
e.g. AND, OR, and NOT -gates. One might wonder if quantum
computation is similar in this regard: yes it is. Different quantum
programs correspond to different unitary operators U, so the
appropriate question is how many and what kind of gates we need to
build up any unitary transformation on qubits. Like in the classical
case, different choices exist but one is

Universal gates: CNOT, H, S, T

Universality means that any unitary operator U can be approximated

with arbitrary accuracy by using a finite sequence of universal gates.
Now we’ll state two important facts about universal circuits without
proof:
▶ Solovay-Kitaev theorem: Any single-qubit unitary U can be
accurately and efficiently approximated with universal gates (for
accuracy ϵ, circuit length ∼ poly (log (1/ϵ)))
▶ There exist unitary transformations U on n-qubits which cannot be
efficiently approximated with universal gates (circuit length ∼ 4n )
Quantum circuit model of computation

Now we can summarize the core elements of quantum computation

1. State space: The quantum circuit operates on n qubits which form
a 2n -dimensional complex Hilbert space. The states |x1 x2 . . . xn ⟩
where xi ∈ {0, 1} are known as computational basis states.
2. Ability to prepare computational basis states: we may assume
that any computational basis state can be prepared by the
quantum computer efficiently.
3. Ability to realize universal gates: universal gates can be applied
on any qubits. For example, two-qubit gates such as the CNOT
can be applied to any two qubits.
4. Ability to measure in computational basis: useful information can
be extracted from the quantum computer when it is measured. We
may assume the ability of the computer to measure any qubit in
the computational basis.
The density operator

Mixed states and pure states

Mixed states

Up to this point we have assumed that we know the state vector |ψ⟩ of
the system which contains the complete description (as per Postulate 1)
of our physical system. Often however, we encounter situations where
our knowledge of the state is probabilistic. Consider a couple common
situations in our context of quantum information/computation
▶ A quantum computer is prepared to some initial state |0⟩. During
the computation, we perform deliberate actions to manipulate this
state in a way of our choosing. However, the manipulation
mechanism has finite precision: we end up with a probability
distribution of final states.
▶ The isolation of the qubits and their environment is not perfect:
the qubits become entangled with the environment and since we
are not observing the environment, our description of the qubits is
imprecise.
When we know the state vector precisely, we say that the state is pure.
If we don’t know the state exactly, the state is mixed and we use
density operators in their description.
Ensembles of quantum states

Suppose a quantum state is in a state |ψi ⟩ with probability pP i . We call

{pi , |ψi ⟩} an ensemble of pure states. Clearly we must have i pi = 1.
Consider some measurement {Mm }. We would like a description of this
statistical mixture which gives the probability for the outcome m
X X
†
p(m) = pi p(m|i) = pi ⟨ψi | Mm Mm |ψi ⟩ ,
i i

so a weighted mixture of probabilities in each state |ψi ⟩. Similarly, we

would like the expectation value of an observable A to be
X
⟨A⟩ = pi ⟨ψi |A|ψi ⟩ .
i

Such a description is given by density operators.

Density operators

A density operator ρ is an operator on the state space which we use to

describe a quantum state. It satisfies the following properties:
1. Hermiticity/self-adjointness: ρ† = ρ
2. Unit trace: tr ρ = 1
3. Non-negativity: ⟨ψ|ρ|ψ⟩ ≥ 0 for all |ψ⟩
4. ρ2 = ρ if state is pure, ρ2 ̸= ρ if state is mixed
A pure state can be written ρ = |ψ⟩⟨ψ| for some |ψ⟩. Since ρ is
Hermitian, its eigenvalues are real. Proof: let |ψ⟩ be an eigenstate of
Hermitian A with eigenvalue λ. Note that ⟨ψ|A|ψ⟩ = λ ⟨ψ|ψ⟩ = λ.
Therefore λ∗ = ⟨ψ|A|ψ⟩∗ = ⟨ψ|A† |ψ⟩ = ⟨ψ|A|ψ⟩ = λ, which implies
that λ is real. Using the spectral decomposition, we can always write
X
ρ= pi |ψi ⟩⟨ψi | .
i

This can be interpreted as a ensemble of quantum states {pi , |ψi ⟩}.

The trace
P condition guarantees that probabilities sum up to one:
tr ρ = i pi = 1.
Density operators

The positivity condition is satisfied if all pi ≥ 0. Proof: let |ψ⟩ be an

arbitrary pure
P quantum state. Then
⟨ψ|ρ|ψ⟩ = i pi ⟨ψ|ψi ⟩ ⟨ψi |ψ⟩ = i pi | ⟨ψi |ψ⟩ |2 ≥ 0 if each pi ≥ 0.
P
Remember that the trace of a matrix is the sum of the values on its
P If {|ψi ⟩} is an
diagonal. Porthonormal basis, then
tr ρ = i ⟨ψi |ρ|ψi ⟩ = i pi .
Example: Is the matrix
 
1/2 −1/2
ρ=
−1/2 1/2

a density matrix? It is clearly Hermitian ρ† = ρ and its trace is

tr ρ = 1/2 + 1/2 = 1. It is non-negative if its eigenvalues are
non-negative. The characteristic equation gives
 2
1 1
−λ − = 0 → λ = {0, 1} ,
2 4

so ρ is also positive and therefore a valid density matrix. Furthermore,

it is a pure state because it only has one non-zero eigenvalue.
Postulates for density operators

The postulates of quantum mechanics can be reformulated to use

density operators. The postulates are as follows:
1. A physical system is completely described by a density operator
acting on the state space of the system.
2. A closed quantum system evolves unitarily

ρt ′ = Uρt U † ,

for two times t ′ > t.

3. A measurement is described by a collection of measurement
operators {Mm }. The probability of outcome m is
 
†
p(m) = tr Mm Mm ρ

†
and the state after the measurement collapses ρ 7→ Mm ρMm /p(m).
The expectation value of an observable A is ⟨A⟩ = tr (Aρ).
4. If a joint system consists of n parts, each prepared to state ρi ,
then their joint system is in the state ρ1 ⊗ ρ2 ⊗ . . . ⊗ ρn .
Postulates for density operators
Let’s check that the measurement rule gives the expected probabilities.
The Pdensity operator corresponding to the ensemble {pi , |ψi ⟩} is
ρ = i pi |ψi ⟩⟨ψi |. Suppose our measurement operators are {Mm }.
According to the 3rd postulate, the probability of outcome m is
!
X X  
† †
p(m) = tr Mm Mm pi |ψi ⟩⟨ψi | = pi tr Mm Mm |ψi ⟩⟨ψi |
i i
X X X X
† †
= pi ψj Mm Mm |ψi ⟩ ψi ψj = pi ψj Mm Mm |ψi ⟩ δij
i j i j

X X
†
= pi ⟨ψi |Mm Mm |ψi ⟩ = pi p(m|i) .
i i

This is exactly the ensemble average we wanted. How about observable

expectation values:
X X
⟨A⟩ = tr (Aρ) = pi tr(A |ψi ⟩⟨ψi |) = pi ⟨ψi |A|ψi ⟩ ,
i i

again the correct weighted average. Therefore the measurement

probabilities work intuitively.
Density operators on the Bloch sphere
Previously, we saw that the Bloch sphere is useful for visualizing
one-qubit pure states. Let’s generalize this for arbitrary one-qubit
mixed states. The density matrix can be written
 
a b
ρ= ∗ .
b d

Because ρ† = ρ, we must have a, d ∈ R. Because tr ρ = a + d = 1, we

define a = (1 + r3 )/2 and d = (1 − r3 )/2. Also, let b = (r1 − ir2 )/2 for
real ri . Then
 
1 1 + r3 r1 − ir2 1 1
ρ= = (I + r1 σ1 + r2 σ2 + r3 σ3 ) = (I + ⃗r ·⃗
σ) ,
2 r1 + ir2 1 − r3 2 2

where σi are the Pauli matrices. Non-negativity of ρ gives a condition

on |⃗r |. After a straightforward calculation, the eigenvalues of ρ are
(1 ± |⃗ r |)/2. This means that in order for ρ to be positive, we must have
|⃗
r | ≤ 1. Therefore, any qubit state can be written

1
ρ= r ·⃗
(I + ⃗ σ)
2
and ⃗
r is known as the Bloch vector.
Density operators on the Bloch sphere

Another straightforward computation shows that ρ2 = ρ if and only if

|⃗
r | = 1. Then we can conclude:

▶ |⃗
r | = 1 corresponds to pure ρ
(surface of the Bloch sphere)
▶ |⃗
r | = 0 corresponds to the
completely mixed state
ρ = I/2 (origin)
Reminder: Trace of an operator

The trace is a vital operation in quantum information.

Definition (Trace)
The trace tr A or a operator A : H 7→ H is defined
X X
tr A = ⟨i|A|i⟩ = Aii ,
i i

where {|i⟩} is an orthonormal basis for H.

The trace is independent of the orthonormal basis you use to calculate
it. To see this, consider two different orthogonal bases {|i⟩} and { ϕj }:
 
X X X X
tr A = ⟨i|A|i⟩ = ⟨i| A ϕj ϕj  |i⟩ = ϕj i ⟨i| A ϕj
i i j i,j

!
X X X
= ϕj |i⟩⟨i| A ϕj = ϕj A ϕj ,
j i j
P P
where we used the completeness relation I = i |i⟩⟨i| = j ϕj ϕj .
Reminder: Trace of an operator

Other basic properties of the trace:

▶ Linearity: tr(zA + wB) = z tr A + w tr B, for z, w ∈ C
▶ Cyclic property: tr(AB) = tr(BA)
This simple relation will be useful when working with density operators.
Consider any two unit vectors |ψ⟩ and |ϕ⟩. Then the trace of |ψ⟩⟨ϕ| is
X X
tr |ψ⟩⟨ϕ| = ⟨i|ψ⟩ ⟨ϕ|i⟩ = ⟨ϕ|i⟩ ⟨i|ψ⟩
i i
!
X
= ⟨ϕ| |i⟩⟨i| |ψ⟩ = ⟨ϕ| I |ψ⟩
i
= ⟨ϕ|ψ⟩ .

The cyclic property also implies the useful fact that the trace of an
operator is invariant under unitary similarity transformations
A 7→ UAU † because
   
tr UAU † = tr U † UA = tr (IA) = tr A .
Ensemble interpretation of density operators
It is often useful
P to think of density matrices as ensembles of quantum
states: ρ = i pi |ψi ⟩⟨ψi | would correspond to the ensemble where the
state |ψi ⟩ occurs with probability pi . It is however important to notice
that this interpretation is far from unique. In fact there is infinitely
many different ensembles that give rise to the same density matrix ρ
(NC Theorem 2.6).
Example: You might say that
3 1
ρ= |0⟩ + |1⟩⟨1|
4 4
describes a system where the state is |0⟩ with probability 3/4 and |1⟩
with probability 1/4. Consider the ensemble
r r r r
3 1 3 1
|a⟩ = |0⟩ + |1⟩ , |b⟩ = |0⟩ − |1⟩ ,
4 4 4 4
where both states have probability 1/2. The corresponding ensemble is

1 1 3 1
ρ= |a⟩⟨a| + |b⟩⟨b| = |0⟩⟨0| + |1⟩⟨1| ,
2 2 4 4
exactly the same as the first ensemble!
Reduced density operator

We often need to study subsystems of larger quantum systems. An

important tool for this is the reduced density matrix and the partial
trace. Suppose we have two physical systems A and B. We denote their
joint state by ρAB . Now say we are interested only on A or only have
access to A. The density operator which describes the subsystem A is

ρA = trB ρAB .

The operator trB is the partial trace and we say that we “trace
out/over B”. The definition of the partial trace is

trB : HA ⊗ HB 7→ HA
trB (|a1 ⟩⟨a2 | ⊗ |b1 ⟩⟨b2 |) ≡ |a1 ⟩⟨a2 | tr(|b1 ⟩⟨b2 |) ,

for any |a1 ⟩ , |a2 ⟩ ∈ HA and |b1 ⟩ , |b2 ⟩ ∈ HB . The trace on the RHS is
the usual trace in HB :

tr (|b1 ⟩⟨b2 |) = ⟨b1 |b2 ⟩ .

Reduced density operator

Example: Let ρAB = ρA ⊗ ρB be the state of a joint system AB. If we

trace over B we get the state trB (ρ√AB ) = ρA .
Example: Let |ψ⟩ = (|00⟩ + |11⟩)/ 2 ∈ HAB . The density matrix is
  
|00⟩ + |11⟩ ⟨00| + ⟨11|
ρ = |ψ⟩⟨ψ| = √ √
2 2
|00⟩⟨00| + |11⟩⟨00| + |00⟩⟨11| + |11⟩⟨11|
= .
2
The reduced density matrix is then

|0⟩⟨0| tr (|0⟩⟨0|) + |1⟩⟨0| tr (|1⟩⟨0|) + |0⟩⟨1| tr (|0⟩⟨1|) + |1⟩⟨1| tr (|1⟩⟨1|)

ρA =
2

|0⟩⟨0| + |1⟩⟨1| 1
= = I.
2 2
It is interesting to note that the joint system is in a definite state |ψ⟩
but still if we look only at A the state is a coin flip between the
orthogonal |0⟩ and |1⟩. Later we will see that because ρA is
proportional to I, the joint qubit state is maximally entangled.
Reduced density operator

Example: Sometimes it is more convenient to work in component

notation, e.g. when calculating on a computer. So let’s do the same
example with explicit components. We have

T √
|ψ⟩ = 1 0 0 1 / 2. Then
   
1 1 0 0 1
1 0
1 0 0 0 0
ρ = |ψ⟩⟨ψ| =   1 0 0 1 =   .
2 0 2 0 0 0 0
1 1 0 0 1

The partial trace is then

ρA = (I ⊗ ⟨0|)ρ(I ⊗ |0⟩) + (I ⊗ ⟨1|)ρ(I ⊗ |1⟩)

     
1 0
1 0 1
= ⊗ 1 0 ρ ⊗
0 1 0 1 0
     
1 0
1 0 0
+ ⊗ 0 1 ρ ⊗
0 1 0 1 1
Reduced density matrix

   
  1 0   0 0
1 0 0 0 0 0
+ 0 1 0 0 1 0
ρA = ρ
 ρ
 
0 0 1 0 0 1 0 0 0 1 0 0
0 0 0 1
 
1 1 0 1
= = I.
2 0 1 2

The result is of course the same. Sometimes you’ll find it more

convenient to work with bras/kets and sometimes with explicit vector
components.
Example: quantum teleportation
Quantum teleportation is a protocol for sending an unknown qubit
state between two parties only using classical communication and a Bell
state.

Suppose Alice (first 2 qubits) wants to send a state |ψ⟩ = α |0⟩ + β |1⟩
to Bob (last qubit). The protocol is shown in the above Figure.
Initially, we have
1
|ψ0 ⟩ = |ψ⟩ |β00 ⟩ = √ (α |0⟩ (|00⟩ + |11⟩) + β |1⟩ (|00⟩ + |11⟩)) .
2
After the CNOT on Alice’s qubits, the state is
1
|ψ1 ⟩ = |ψ⟩ |β00 ⟩ = √ (α |0⟩ (|00⟩ + |11⟩) + β |1⟩ (|10⟩ + |01⟩)) .
2
Example: quantum teleportation

After the Hadamard the state is

1
|ψ2 ⟩ = (α(|0⟩ + |1⟩)(|00⟩ + |11⟩) + β(|0⟩ − |1⟩)(|10⟩ + |01⟩))
2
1
= |00⟩ (α |0⟩ + β |1⟩) + |01⟩ (α |1⟩ + β |0⟩)
2


+ |10⟩ (α |0⟩ − β |1⟩) + |11⟩ (α |1⟩ − β |0⟩) .

Next Alice measures her qubits in the computational basis. We can see
that if the result is 00, then Bob’s state is |ψ⟩ = α |0⟩ + β |1⟩. Other
measurement results cause Bob to have different states, each of which
can be transformed to |ψ⟩ if Bob knows Alice’s measurement result!
Example: quantum teleportation

Depending on Alice’s measurement results, Bob’s state collapses to

00 7→ |ψ3 ⟩ = α |0⟩ + β |1⟩

01 7→ |ψ3 ⟩ = α |1⟩ + β |0⟩
10 7→ |ψ3 ⟩ = α |0⟩ − β |1⟩
11 7→ |ψ3 ⟩ = α |1⟩ − β |0⟩ .

Finally, Alice sends her measurement result M1 M2 to Bob, who applies

Z M1 X M2 on his qubit. Then his qubit is in state |ψ4 ⟩ = |ψ⟩. So, using
only a shared Bell state and classical communication, Alice managed to
send Bob an unknown qubit state.
Example: quantum teleportation

So what do reduced density matrices have to do with any of this?

Using this formalism we can answer the glaring question: it looks like
even before Bob learns Alice’s measurement results, his qubit carries
information about |ψ⟩. Did we just transmit information faster than the
speed of light? Luckily, the answer is no.
Consider the joint state before Alice’s measurement:
1
ρ = |ψ2 ⟩⟨ψ2 | = |00⟩⟨00| (α |0⟩ + β |1⟩)(α∗ ⟨0| + β ∗ ⟨1|)
4
+ |01⟩⟨01| (α |1⟩ + β |0⟩)(α∗ ⟨1| + β ∗ ⟨0|)
+ |10⟩⟨10| (α |0⟩ − β |1⟩)(α∗ ⟨0| − β ∗ ⟨1|)
+ |11⟩⟨11| (α |1⟩ − β |0⟩)(α∗ ⟨1| − β ∗ ⟨0|) .


Example: quantum teleportation

To study this situation from Bob’s perspective, we compute his reduced

density matrix. That is, we trace out Alice’s qubits
1
(|α|2 + |β|2 ) |0⟩⟨0| + (|α|2 + |β|2 ) |1⟩⟨1|


ρB = trA ρ =
2
1
= (|0⟩⟨0| + |1⟩⟨1|) = I/2 .
2
So, Bob has the maximally mixed state which doesn’t depend in any
way on the teleported state |ψ⟩ (α, β do not appear in ρB ). The
quantum information is effectively teleported at the speed of light: only
after Bob learns Alice’s measurement results can he extract |ψ⟩.
Therefore causality is preserved.