4.

QUANTUM COMPUTING - HILBERT SPACES -

2020

Jozef Gruska
Faculty of Informatics
Brno
Czech Republic

October 27, 2020

Quantum Computing 4 – Hilbert Space Basics, 2020

Chapter 3: HILBERT SPACE BASICS

ABSTRACT
Hilbert space is a mathematical
framework suitable for describing
concepts, principles, processes and laws
of the theory of quantum world called
(for historical reasons) quantum
mechanics, in general; and quantum
information processing and
communication (QIPC) in particular.
In this chapter those basics of Hilbert
space theory are introduced and
illustrated that play an important role
in QIPC.

Jozef Gruska October 27, 2020 1

Quantum Computing 4 – Hilbert Space Basics, 2020

QUANTUM SYSTEM = HILBERT SPACE

Hilbert space Hn is n-dimensional complex vector space with

scalar product (dot product)

φ1 ψ1
n φ ψ
hψ|φi = φiψi∗ of vectors |φi = ..2 , |ψi = ..2 ,
X

i=1 . .
φn ψn

norm of vectors r
||φ|| = |hφ|φi|
and the metric

dist(φ, ψ) = ||φ − ψ||.

This allows us to introduce on H a metric topology and such concepts as
continuity.
All n-dimensional Hilbert spaces are isomorphic. Their vectors of norm 1
are called pure quantum states¿ Their physical counterparts are n-level
quantum systems.

Jozef Gruska October 27, 2020 2

Quantum Computing 4 – Hilbert Space Basics, 2020

MORE ABOUT RELATIONS BETWEEN QUANTUM SYSTEMS A

Basic assumption With every quantum systems S

there is associated a Hilbert space HS , whose
dimension depends on the nature of the degree of
freedom being considered for the system.
Example: If only spin orientation of electrons (
spin-1/2 particles) is considered, then the
corresponding Hilbert space will be a two
dimensional Hilbert space H2.
However, if the position of electrons is of a
concern, which can be in any point of some space,
then the corresponding Hilbert space is usually
taken to be continuous and therefore infinite
dimensional.

Jozef Gruska October 27, 2020 3

Quantum Computing 4 – Hilbert Space Basics, 2020

BRA-KET NOTATION

Dirac introduced a very handy notation, so

called bra-ket notation, to deal with amplitudes,
quantum states and linear functionals f : H → C.

If ψ, φ ∈ H, then
hψ|φi — scalar product of ψ and φ
(an amplitude of going from φ to ψ).
|φi — ket-vector — an equivalent to φ
hψ| — bra-vector a linear functional on H (and a
dual vector to |φi)
such that hψ|(|φi) = hψ|φi
Example For states φ = (φ1, . . . , φn) and
ψ = (ψ1, . . . , ψn) we have
∗
  
 φ1   φ1 ψ1
 
∗ ∗ n ∗

..
|φi =   , hφ| = (φ1 , . . . , φn ); hφ|ψi =
 ... 
φ ψ ; |φihψ| =
  X 
i

i 

i=1 
φn φnψ1∗
  

Jozef Gruska October 27, 2020 4

Quantum Computing 4 – Hilbert Space Basics, 2020

GENERAL DEFINITION

Definition 0.1 An inner-product space H is a complex vector space,

equipped with an inner product h·|·i : H × H → C satisfying the following
axioms for any vectors φ, ψ, φ1, φ2 ∈ H, and any c1, c2 ∈ C.
hφ|ψi = hψ|φi∗,
hψ|ψi ≥ 0 and hψ|ψi = 0 if and only if ψ = 0,
hψ|c1φ1 + c2φ2i = c1hψ|φ1i + c2hψ|φ2i.1

The inner product introduces on H the norm (length)

r
||ψ|| = hψ|ψi
and the metric (Euclidean distance)
dist(φ, ψ) = ||φ − ψ||.
This allows us to introduce on H a metric topology and such concepts as
continuity.
Some basic properties of the norm:
• ||φ|| ≥ 0 for all φ ∈ H and ||φ|| = 0 if and only if φ = 0
• ||φ + ψ|| ≤ ||φ|| + ||ψ|| (triangle inequality);
• ||aφ|| = |a| ||φ||; (e) |hφ, ψi| ≤ ||φ|| ||ψ|| (Schwarz inequality).

Jozef Gruska October 27, 2020 5

Quantum Computing 4 – Hilbert Space Basics, 2020

COMPLETENESS, ISOMORPHISM and DUAL SPACE

Definition 0.2 An inner-product space H is called

complete, if for any sequence
{φi}∞
i=1 ,

with φi ∈ H, and with the property that

lim ||φi − φj || = 0,
i,j→∞

there is a unique element φ ∈ H such that

lim ||φ − φi|| = 0.
i→∞
A complete inner-product space is called a Hilbert space.
Two Hilbert spaces H1 and H2 are said to be isomorphic,
notation H1 ' H2, if the underlying vector spaces are
isomorphic and their isomorphism preserves the inner product.

Jozef Gruska October 27, 2020 6

Quantum Computing 4 – Hilbert Space Basics, 2020

BASIC EXAMPLES of HILBERT SPACES

Let us start with the two most important examples of Hilbert

spaces.
Example 0.3 Hilbert spaces l2(D) For any countable set D,
let l2(D) be the space of all complex valued functions on D
bounded by the so-called l2-norm, i.e.
 
1/2
∗
l2(D) = {x | x : D → C,  X
 x(i)x (i) 
 < ∞}2.
i∈D

We say that l2(D) is a Hilbert space with respect to the inner

product h · | · i : l2(D) × l2(D) → C, defined by
hx1|x2i = x∗1 (i)x2(i).
X

i∈D

Elements of l2(D) are usually called vectors (to be indexed by

elements of D). The notation l2 = l2(N) is usually used in the
case D = N.
2 ∗
x denotes the conjugate of the com x; i.e., x∗ = a − bi if x = a + bi, where a, b are real.

Jozef Gruska October 27, 2020 7

Quantum Computing 4 – Hilbert Space Basics, 2020

Example 0.4 Hilbert space L23 Let (a, b) be an interval, with

finite or infinite bounds, on the real axis. By L2((a, b)), or
simply L2, we denote the set of all complex valued functions
such that ab |f (x)|2 dx exists, equipped with the inner product
R

b
hf |gi = a f ∗(t)g(t) dt < ∞.
Z

If f and g are such that |f |2 and |g|2 are integrable functions

(with respect to Lebesgue measure) on (a, b), then so are
functions cf and f + g, for any complex number c, and
therefore L2 is a linear space.4
Surprisingly, for two Hilbert spaces introduced in the last
examples it holds
l2 ' L2
that is they are isomorphic (so-called Riesz-Fischer
Theorem.)
The Hilbert space corresponding to a simple harmonic
oscillator is L2 of all complex valued functions, each of which
is square integrable over the entire real line.

3
Hilbert studied spaces l2 and L2 , in his work on linear integral systems, and that is why von Neumann all spaces of such
types named as Hilbert spaces.
4
To be more precise L2 is to be the set of Lebesgue integrable functions on (a, b) and we do not consider as different a pair
of functions that differ only on a set of measure zero. In such a linear space the zero element is a function that is equal to zero
almost everywhere on (a, b).

Jozef Gruska October 27, 2020 8

Quantum Computing 4 – Hilbert Space Basics, 2020

DUAL HILBERT SPACE

The set of dual vectors of a Hilbert space H forms so called

Dual Hilbert space H∗. A dual vector hφ| to a vector |φi is
often denoted as |φi†.
If {|βii}ni=1 forms an orthogonal basis of a Hilbert space H,
then {hβi|}ni=1 forms an orthogonal basis of H∗.
If H is a Hilbert space, then the set of linear operators on H
forms again a Hilbert space, denoted usually L(H) and all
inner products of an orthogonal basis of H forms an
orthogonal basis of L(H). As a consequence if {βii}ni=1
form an orthogonal basis on H, then every linear operator O
on H can be expressed in the form
O= tn,m|βnihβm|
XX

n m
for some constants tn,m = hβn|O|βmi

Jozef Gruska October 27, 2020 9

Quantum Computing 4 – Hilbert Space Basics, 2020

ORTHOGONALITY of STATES
Two vectors |φi and |ψi are called orthogonal
vectors if hφ|ψi = 0.
Physically are fully distinguishable only
orthogonal vectors (states).
By a basis B of Hn we will understand any set of
n vectors |b1i, |b2i, . . . , |bni in Hn of the norm 1
which are mutually orthogonal.
Given a basis B, any vector |ψi from Hn can be
uniquely expressed in the form
n
|ψi = αi|bii.
X

i=1
A set S of vectors is called orthonormal if all
vectors of S have norm 1 and are mutually
orthogonal.
DefinitionA subspace G of a Hilbert space H is a
subset of H closed under addition and scalar
multiplication.
An important property of Hilbert spaces is their
decomposability into mutually orthogonal
subspaces. It holds:
Theorem For each closed subspace W of a
Hilbert space H there exists a unique subspace
Jozef Gruska October 27, 2020 10
Quantum Computing 4 – Hilbert Space Basics, 2020

W ⊥ such that hφ|ψi = 0, whenever φ ∈ W and

ψ ∈ W⊥
and
each ψ ∈ H can be uniquely expressed in the
form ψ = φ1 + φ2, with φ1 ∈ W and φ2 ∈ W ⊥. In
such a case we write H = W ⊕ W ⊥ and we say
that W and W ⊥ form an orthogonal
decomposition of H.
In a natural way we can make a generalization
of an orthogonal decomposition

H = W1 ⊕ W2 ⊕ . . . ⊕ Wn ,
of H into mutually orthogonal subspaces
W1, . . . , Wn such that each ψ ∈ H has a unique
representation as ψ = φ1 + φ2 + . . . + φn, with
φi ∈ Wi, 1 ≤ i ≤ n.

Jozef Gruska October 27, 2020 11

Quantum Computing 4 – Hilbert Space Basics, 2020

OPERATORS of HILBERT SPACES

A linear operator on a Hilbert space H is a

linear mapping A : H → H.
An application of a linear operator A to a vector
|ψi is denoted A|ψi.
A is also a linear operator of the dual Hilbert
space H ∗, mapping each linear functional hφ| of
the dual space to the linear functional, denoted
by hφ|A.
A linear operator A is called positive or
semi-definite, notation A ≥ 0, if hψ|Aψi ≥ 0 for
every |ψi ∈ H.

The norm ||A|| of a linear operator A is defined

as
||A|| = sup ||A|φi||.
||φ||=1
A linear operator is called bounded if its norm
is finite.

Jozef Gruska October 27, 2020 12

Quantum Computing 4 – Hilbert Space Basics, 2020

Projections have a special role among linear

operators.
If H = W1 ⊕ W2 is an orthogonal decomposition
of a Hilbert space H into subspaces W1 and W2,
then, as mentioned above, each ψ ∈ H has a
unique representation ψ = ψ1 + ψ2, where
ψ1 ∈ W1 and ψ2 ∈ W2.
PW1 (ψ) = ψ1 and PW2 (ψ) = ψ2
are called projections onto the subspaces W1
and W2, respectively.
Example If φ ∈ H for ||φ|| = 1 and a Hilbert
space H, then the operator defined by |φihφ| and
defined by
|φihφ|(|ψi) = hφ|ψi|φi
is a projection into the one-dimensional subspace
spanned by the vector |φi.

Jozef Gruska October 27, 2020 13

Quantum Computing 4 – Hilbert Space Basics, 2020

REPRESENTATIONS

If β = {θi}ni=1 is an orthonormal basis of a Hilbert

space H, then any state |φi has a unique
representation
n
|φi = hθi|φi|θii.
X

i=1
Each linear operator A of a countable Hilbert
space H with a basis B = {|θii| i ∈ I} can be
represented by a matrix, in general infinitely
dimensional, whose rows and columns are
labeled by elements of I and with
hθi|A|θj i = hθi|Aθj i in the i-th row and j-th
column. Such a matrix is said to be matrix
representation of A relative to the basis β.
Each operator A has also so-called outer-product
representation:
A = hθi|A|θj i|θiihθj |.
X

ij

(here we use for hθi|Aθj i Dirac’s notation

hθi|A|θj i).

Jozef Gruska October 27, 2020 14

Quantum Computing 4 – Hilbert Space Basics, 2020

TRACE of OPERATORS

Definition 0.5 Let H be an n-dimensional Hilbert space and B be an

orthogonal basis on H. The trace (operator) of a linear mapping
M : H → H is defined by
Tr(M ) = hφ|M |φi.
X

φ∈B

In addition, if A is the matrix representation of M in the basis B. Then

n
Tr(M ) = Tr(A) = aii,
X

i=1

where aii is the element of the matrix A at the position (i, i).

Properties of the trace operator:

• Trace of a linear mapping does not depend on the basis chosen.
• Tr(A + B) = Tr(A) + Tr(B).
• Tr(AB) = Tr(BA); Tr(A ⊗ B) = Tr(A)Tr(B).
• Tr(αA) = αTr(A).
• Tr(|ψihφ|) = hφ|ψi.

Jozef Gruska October 27, 2020 15

Quantum Computing 4 – Hilbert Space Basics, 2020

SELF-ADJOINT OPERATORS

Of special importance are adjoint and

self-adjoint operators.
The adjoint operator T ∗ to a bounded linear
operator T is an operator such that for any φ
and ψ ∈ H,
hψ|T φi = hT ∗ψ|φi.
An operator T is self-adjoint if T = T ∗.
Instead of hψ|T φi notation hψ|T |φi is used. Hence

hT ∗ψ|φi = hψ|T |φi = hψ|T φi.

Jozef Gruska October 27, 2020 16

Quantum Computing 4 – Hilbert Space Basics, 2020

SELF-ADJOINT OPERATORS — HERMITIAN MATRICES

To self-adjoint operators correspond Hermitian

matrices, i.e., matrices A such that A = A∗.
Theorem 0.6 Hermitian matrices have the
following properties.
1. All eigenvalues of a Hermitian matrix are
real.
2. The eigenvectors of an Hermitian matrix
corresponding to distinct eigenvalues are
orthogonal.

Proof of property 1: If Aφ = λφ, then

λ∗hφ|φi = hλφ|φi = hAφ|φi = hφ|Aφi = λhφ|φi.
hence λ∗ = λ.
Proof of property 2 Assume that λ 6= λ0,
Aφ = λφ, Aφ0 = λ0φ0. Since λ, λ0 are real, it holds
λ0hφ0|φi = hAφ0|φi = hφ0|Aφi = λhφ0|φi
and therefore hφ0|φi = 0.

Jozef Gruska October 27, 2020 17

Quantum Computing 4 – Hilbert Space Basics, 2020

SPECTRAL REPRESENTATION of SELF-ADJOINT OPERATORS

A self-adjoint operator A of a finite dimensional Hilbert

space H has the so-called spectral representation. If
λ1, . . . , λk are its distinct eigenvalues, then A can be
expressed in the form
k
A= λj Pj ,
X

j=1

where Pj is the projection operator into the subspace of H

spanned by the eigenvectors corresponding to λj .
In a special case when all eigenvalues are distinct and |φii is
the eigenstate/eigenvector corresponding to the eigenvalue
λi, then
n
A= λi|φiihφi|
X

i=1
Since PiPj = 0 for two different projections, it holds for any
polynomial p
k
p(A) = p(λi)Pi.
X

i=1
This is generalized to define for any function f : R → C by
k
f (A) = f (λi)Pi.
X

i=1

Jozef Gruska October 27, 2020 18

Quantum Computing 4 – Hilbert Space Basics, 2020

Example Pauli matrix σx has eigenvalues 1 and −1 and

corresponding eigenvectors are |00i = √12 (|0i + |1i) and
|10i = √12 (|0i − |1i). Since
1 1 1
− 12
   

|00ih00| =  2 2 |10ih10| =  2
   
1 1
− 12 1
 
 
2 2 2
and therefore
1 1 1
− 21
   

σx = 1  2 2 − 1  2
   
1 1
− 12 1
 
 
2 2 2
√ √ 1 1 √  1
− 21
   

σx = 1  12 2  + −1  2
 
1
− 12 1
 

2 2 2
that is

√ 1+i 1−i
 

σx = 

2 2 
.
1−i 1+i


2 2

Jozef Gruska October 27, 2020 19

Quantum Computing 4 – Hilbert Space Basics, 2020

SPECTRAL REPRESENTATION of UNITARY OPERATORS

Each self-adjoint operator A has spectral decomposition

A = Pnj=1 λj |φj ihφj | and therefore
n
eiA = eiλj |φj ihφj |
X

j=1

and therefore
n
(eiA)∗ = e−iλj |φj ihφj | = (eiA)−1
X

j=1

what implies that the matrix eiA is unitary.

We show now that each unitary matrix U = eiH for some
self-adjoint operator H. Indeed, if U is decomposed into a
“real and imaginary” part U = A + iB, then A and B are
both self-adjoint and have spectral decompositions
n
A= λi|φiihφi|
X

i=1
n
B= µi|φiihφi|
X

i=1
All eigenvalues of a unitary matrix have absolute value 1 and
self-adjoint matrices have eigenvalues real. Therefore, for
each j an θj ∈ [0, 2π] has to exists such that
λj + iµj = eiθj . Hence for
n
H= θj |φj ihφj |
X

j=1
Jozef Gruska October 27, 2020 20
Quantum Computing 4 – Hilbert Space Basics, 2020

we have
U = eiH .

 
1 1  √1
Example 1 Hadamard transform H =  has as


1 −1 2 

eigenvalues 1 and −1 and the corresponding eigenvectors are

 √   √ 
1  1 + 2  1  1 − 2 
φ1 = r √  φ−1 = r √ 
1 1
 
4+2 2 4−2 2
The spectral decomposition of H is then

H = 1 · |φ1ihφ1| + (−1)|φ−1ihφ−1|
and therefore H = eiA, where
A = 0 · |φ1ihφ1| + π|φ−1ihφ−1|
Example 2 It holds
   
cos θ sin θ  iAθ 0 iθ 
Rθ =  = e whereA = .
 
θ

sin θ cos θ −iθ 0
 

Jozef Gruska October 27, 2020 21

Quantum Computing 4 – Hilbert Space Basics, 2020

QUANTUM TIME EVOLUTION

It is natural to assume that for a Hilbert space Hn there is a

mapping Ut : Hn → Hn that depends on time t that maps
the initial state |φ0i into the state |φti in time t, that is that
|φti = Ut|φ0i.
It is also natural to put the following four requirements on
Ut :
1. Ut should map states into states, that is it should preserve
norm:
For each real t > 0 and each state |φi, ||Ut|φi|| = |||φi||.
2. Ut should be linear - what means that each basis state
should develop independently. Namely: for each basis
{βi}ni=1,
n n
Ut( ai|βii) = aiUt(|βii)
X X

i=1 i=1
.
3. Ut should be decomposable. Namely, for all t1 > 0, t2 > 0
Ut1+t2 = Ut1 Ut2 .
4. Evolution should be smooth. Namely, for each real t0
lim Ut|φ0i = lim |φti = |φt0 i.
t→t0 t→t0
Jozef Gruska October 27, 2020 22
Quantum Computing 4 – Hilbert Space Basics, 2020

Theorem If time evolution Ut satisfies the above four

conditions, then Ut has to be unitary and Ut = e−iH for
some self-adjoint operator H. Hence
∂φt
= −iHe−itH φ0 = −iHφt
∂t
what is known as the abstract Schrödinger equation.

Jozef Gruska October 27, 2020 23

Quantum Computing 4 – Hilbert Space Basics, 2020

ANOTHER VIEW OF QUANTUM DYNAMICS

The time evolution of any state of a closed quantum system

is unitary; i.e. the state |φ(t1)i of the system at time t1 is
related to the state |φ(t2)i of the system at a later time t2
by a unitary operator Ut1,t2 which depends on t1 and t2:
|φ(t2)i = Ut1,t2 |φ(t1).
moreover such an evolution is linear, i.e.
Ut1,t2 αi|φii = αiUt1,t2 |φii.
X X

t i

Jozef Gruska October 27, 2020 24

Quantum Computing 4 – Hilbert Space Basics, 2020

QUANTUM (PROJECTION) MEASUREMENT - OBSERVABLES

Informally, a quantum state |ψi is observed (measured) with respect to

an observable — a Hermitian matrix A which specifies a
decomposition of the Hilbert space into orthogonal subspaces (such that
each vector can be uniquely represented as a sum of vectors of these
subspaces) that are subspaces generated by eigenvectors corresponding
to different eigenvalues of the operator A.

There are two outcomes of a projection measurement of a state |φi:

1. Information into which subspace projection of |φi took place.
2. Resulting projection (a new quantum state) |φ0i.
The subspace into which projection is made is chosen randomly and
the corresponding probability is uniquely determined by the amplitudes at
the representation of |φi at the basis states of the subspace.

Jozef Gruska October 27, 2020 25

Quantum Computing 4 – Hilbert Space Basics, 2020

Namely, if
k
A= λiPi
X

i=1
for a self-adjoint operator A, where Pi are
projections into mutually orthogonal subspaces,
then by measuring (observing) a state |φi with
respect to A, this state collapses into the state
Pi|φi
s

hφ|Pi|φi
with probability
hφ|Piφi
and we also say that with the same probability the
value λi is observed in the classical world.

Jozef Gruska October 27, 2020 26

Quantum Computing 4 – Hilbert Space Basics, 2020

PROJECTION MEASUREMENT - OTHER VIEW

For any decomposition of a unitary

operator
I = Pi
X

i
into orthogonal projectors Pi there
exists a projective measurement that
outputs, if a state |φi is measured, as
the outcome an i with probability
Pr(i) = hφ|Pi|φi
and leaves the system in the state
Pi|φi
s

Pr(i)

Jozef Gruska October 27, 2020 27

Quantum Computing 4 – Hilbert Space Basics, 2020

EXAMPLE

 

0 1  
Observable σx =  has eigenvalues



1 0



{1, −1} and eigenvectors {|00i, |10i}.

A measurement with respect to the
observable σx is therefore measurement
with respect to the dual basis.
 

1 0  
Observable σz =  has eigenvalues



0 −1 



{1, −1} and eigenvectors {|0i, |1i}.

A measurement with respect to the
observable σx is therefore measurement
with respect to the standard
(computational) basis.

Jozef Gruska October 27, 2020 28

Quantum Computing 4 – Hilbert Space Basics, 2020

PROBABILITIES

To the key outcomes of quantum mechanics

belong rules for determining probabilities of the
outcomes of quantum measurements.
The classical outcome of a measurement of a
state |ψi with respect to an observable A is one
of the eigenvalues of A and quantum impact of
such a measurement is a “collapse” of |ψi into a
state |ψ 0i. In the measurement the eigenvalue λi
is obtained with probability
P r(λi) = ||Pi|ψi||2 = hψ|Pi|ψi,
and the new state |ψ 0i, into which |ψi collapses,
has the form

Pi|ψi
|ψ 0i = r .
hψ|Pi|ψi

Jozef Gruska October 27, 2020 29

Quantum Computing 4 – Hilbert Space Basics, 2020

ANOTHER VIEW of MEASUREMENT

Quantum measurement is described by a finite

set {Pm} of projectors acting on the state space
of the system being measured and such that
m Pm = I - the index m refers to the potential
P

classical outcomes of a measurement.

If a state |φi is measured with respect to {Pm},
then the result m occurs with probability
Pr(m) = hφ|Pm|φi
and if such a result occurs the state of the
system immediately after measurement is
Pm|φi Pm|φi
r =r .
hφ|Pm|φi Pr(m)
If α is a real number, we say that states |φi and
eiα|φi are equivalent, or equal up to a phase
factor. Two such states give the same
measurement statics, what follows from the
relations
heiαφ|Pm|eiαφi = e−iαeiαhφ|Pm|φi = hφ|Pm|φi.

Jozef Gruska October 27, 2020 30

Quantum Computing 4 – Hilbert Space Basics, 2020

IMPORTANT OBSERVATION

Quantum mechanics is a probabilistic

mathematical theory for describing the
physical world.
However, probability involved is not
probability of some dynamic variables
having a particular value in some state.
Rather, it represents the probability of
finding a particular value of a dynamical
variable if that dynamical variable is
measured.
Quantum mechanics says nothing about
values of dynamical variables when the
system is not subjected to any
measurement.

Jozef Gruska October 27, 2020 31

Quantum Computing 4 – Hilbert Space Basics, 2020

EXPECTATION VALUE

Let A be self-adjoint operator of a Hilbert space H , with

spectral decomposition
k
A= λiPλi .
X

i=1
The expectation value of A in the state ψ is defined by
k
expψ (A) = λiprobψ (λi)
X

i=1
k
= λihPλi ψ|Pλi ψi
X

i=1
k
= λihψ|Pλi ψi
X

i=1
k
= hψ| λiPλi |ψi
X

i=1
= hψ|Aψi
= hψ|APψ ψi = Tr(APψ ) = Tr(Pψ A).

Jozef Gruska October 27, 2020 32

Quantum Computing 4 – Hilbert Space Basics, 2020

MIXED STATES — DENSITY MATRICES

Pure states are fundamental objects for quantum mechanics

in the sense that the evolution of any closed quantum system
can be seen as a unitary evolution of pure states.
However, to deal with unisolated and composed quantum
systems the concept of mixed states is of importance.
A probability distribution {(pi, φi) | 1 ≤ i ≤ k} on pure
states {φi}ki=1, with probabilities 0 < pi ≤ 1, Pki=1 pi = 1 is
called a mixed state or mixture, and denoted by
[ψi = {(pi, φi) | 1 ≤ i ≤ k}. For example, a mixed state is
created if a source produces pure state |φii with probability
pi and Pki=1 pi = 1.
To each mixed state [ψi = {(pi, φi) | 1 ≤ i ≤ k}
corresponds a density operator
k
ρ[ψi = pi|φiihφi|.
X

i=1

Key observation Two mixed states with the same

density matrix are physically undistinguishable.

Jozef Gruska October 27, 2020 33

Quantum Computing 4 – Hilbert Space Basics, 2020

EXAMPLES

The density matrix corresponding to the mixed state

1 1
( , |0i) ⊕ ( , |1i)
2 2
has the form    
11 10 1
  (0, 1) +   (0, 1) = I.
2 0 2 1 2

For any pure one qubit state α|0i + β|1i, to the mixed state
1 1 1 1
( , α|0i + β|1i) ⊕ ( , α|0i − β|1i) ⊕ ( , β|0i + α|1i) ⊕ ( , β|0i − α|1i)
4 4 4 4
corresponds the density matrix
   
1α ∗ ∗ 1 α
 (α , β ) +

(α∗, −β ∗)
4 −β
  
4 β
   
1β  ∗ ∗ 1 β  ∗ ∗ 1
+  (β , α ) +  (β , −α ) = I
4 −α
 
4 α 2

Jozef Gruska October 27, 2020 34

Quantum Computing 4 – Hilbert Space Basics, 2020

REPRESENTATION of MIXED STATES

If ρ is a density matrix and in a basis

{βi}ni=1
ρ = {ρi,j }ni,j=1,
then
n
ρ= ρi,j |βiihβj |.
X

i,j=1
As a consequence, for any k, l,
hβk |ρ|βl i = ρk,l .

Jozef Gruska October 27, 2020 35

Quantum Computing 4 – Hilbert Space Basics, 2020

PROPERTIES of DENSITY MATRICES

1. Any density matrix ρ is Hermitian,

nonnegative, has only nonnegative
eigenvalues and Tr(ρ) = 1.
2. If ρ1, ρ2 are density matrices on a
Hilbert space H, then pρ1 + (1 − p)ρ2,
0 ≤ p ≤ 1 is a density matrix on H.
3. If ρ is a density matrix, then so is
the matrix ρT .
4. If ρ1 is a density matrix on a Hilbert
space H1 and ρ2 is a density matrix
on a Hilbert space H2, then ρ1 ⊗ ρ2 is
density matrix on the Hilbert space
H1 ⊗ H 2 .
5. A matrix ρ is a density matrix if it is
Hermitian, nonnegative and Trρ = 1.
6. If ρ2 = ρ for a density matrix ρ, then
ρ is a pure state, i.e. ρ = |φihφ| for a
pure state |φi.
Jozef Gruska October 27, 2020 36
Quantum Computing 4 – Hilbert Space Basics, 2020

DENSITY OPERATORS as STATES

We had a description of quantum (pure) states in

terms of vectors of the norm one of a Hilbert
space.
An alternative description is in terms of density
operators which is very useful in describing states
of subsystems of a composite quantum systems.
Quantum states are therefore often (mostly)
associated with density operators (positive trace 1
operators).

Jozef Gruska October 27, 2020 37

Quantum Computing 4 – Hilbert Space Basics, 2020

BLOCH VECTORS REPRESENTATION of MIXED

For qubits, any density operator ρ (matrix) can be

written uniquely in the form
1
ρ = [I + axσx + ay σy + az σz ]
2
where ai are real numbers such that for
i ∈ {x, y, z}, Pi |ai|2 ≤ 1 (because the set of
matrices {I, σx, σy , σz } form a basis) .
In short, ρ can be written as
1
ρ = [I + a · σ]
2
where a s a vector with components ai and the
notation r · σ means Pi aiσi.
Therefore there is one-to-one correspondence
between density operators for qubits and points of
the Bloch/Poincare sphere.
Example Totally mixed state {( 12 , |0i), ( 12 , |1i)k
that is identical with {( 21 , |0ih0|), ( 12 , |1ih1|)k
corresponds to the centre of the Bloch sphere.
Jozef Gruska October 27, 2020 38
Quantum Computing 4 – Hilbert Space Basics, 2020

EVOLUTION of MIXED STATES and DENSITY MA

If a unitary matrix U is applied to a

mixed state

{(p1, |φii), . . . , (pk , φk )}

with the density matrix
k
ρ= pi|φiihφi|
X

i=1
then the resulting mixed state is

{(p1, |U φii), . . . , (pk , U φk )}

and the corresponding density matrix is
k k
piU |φiihφi|U = U ( pi|φiihφi|)U † = U ρU †
†
X X

i=1 i=1

Jozef Gruska October 27, 2020 39

Quantum Computing 4 – Hilbert Space Basics, 2020

TRACING OUT OPERATION

One of the profound differences between the

quantum and classical systems lies in the
relation between a system and its subsystems.
As discussed below a state of a Hilbert space
H = HA ⊗ HB cannot be always decomposed into
states of its subspaces HA and HB . We also
cannot define any natural mapping from the
space of linear operators on H into the space of
linear operators on HA (or HB ).
However, density operators are much more
robust and that is also one reason for their
importance. A density operator ρ on H can be
“projected” into HA by the operation of tracing
out HB , to give the density operator (for finite
dimensional Hilbert spaces):
 

|φi hφ, ψ|ρ|φ , ψi hφ0|,

0
 
ρHA = T rHB (ρ) =
X  X 
 
|φi,|φ0 i∈BH
 
|ψi∈BH
A B

where BHA (BHB ) is an orthonormal basis of the

Hilbert space HA (HB ).

Jozef Gruska October 27, 2020 40

Quantum Computing 4 – Hilbert Space Basics, 2020

MEANING of TRACING OUT OPERATIONS

If dim(HA) = n, dim(HB ) = m, then ρ is an

nm × nm matrix which can be seen as an n × n
matrix consisting of m × m blocks ρij as follows:
 


ρ11 ... ρ1n 
 
 
 

ρ=




.. ... 



 
 
 
 
 
ρn1 ... ρnn
 

and in such a case

 
 Tr(ρ11 ) ... Tr(ρ1n) 

 
 
 

ρHA =




.. ... .. 



 
 
 
 
 
Tr(ρn1) ... tr(ρnn)
 

Jozef Gruska October 27, 2020 41

Quantum Computing 4 – Hilbert Space Basics, 2020

WHY SUCH a TRACING OUT OPERATION?

The following fact is the main mathematical

justification that strangely looking tracing out
operation has the proper physical meaning:

If H = HA ⊗ HB and ρ is a density matrix of H,

then ρA = Tr B (ρ) is the unique density matrix of
HA such that
Tr (ρA · O) = Tr (ρ · (O ⊗ I))
for each observable (Hermitian matrix) O of HA.

In other words the average value of the

measurement of ρA on HA with respect to the
observable O, on HA, equals the average value of
the measurement of ρ on H with respect to the
observable O ⊗ I on H.
Informally, one often says that in order to get
the density matrix of a subsystem, given the
density matrix of the whole system, one should
trace over the degrees of freedom of the rest of
the system.

Jozef Gruska October 27, 2020 42

Quantum Computing 4 – Hilbert Space Basics, 2020

TRACING OUT OPERATION

Perhaps the simplest way to introduce

tracing out operation is to say that it is
a linear operation such that for any
bipartite system A ⊗ B and any states
|φ1i and |φ2i of A and any states |ψ1i and
|ψ2i of B
TrB (|φ1ihφ2|⊗|ψ1ihψ2|) = |φ1ihφ2|Tr(|ψ1ihψ2|) = hψ2|

Jozef Gruska October 27, 2020 43

Quantum Computing 4 – Hilbert Space Basics, 2020

EXAMPLES

Example 1. For
1
ρ = (|00ih00| + |00ih11| + |11ih00| + |11ih11|)k
2
it holds
1
Tr(ρ) = (|0ih0| + |1ih1|
2

Jozef Gruska October 27, 2020 44

Quantum Computing 4 – Hilbert Space Basics, 2020

MEASUREMENT of PURE STATES

Let us assume that we are measuring (with respect to) the

observable X with spectral decomposition
X= λj |jihj|.
X

From the hermiticity of X it follows that the eigenvalues λj

are real. For simplicity we assume that eigenvalues are
nondegenerate (all different) and the corresponding
eigenvectors, {|ji}j , form an orthonormal basis. Then:
1. The projectors Pj = |jihj| span the entire Hilbert space,
j Pj = 1.
P

2. From the orthogonality of the basis states we have

PiPj = δij Pi. in particular, Pi2 = Pi, what implies that
eigenvalues of any projector are −1 and 1.
3. Each measurement of a state |φi with respect to X
yields, on a classical level, one of the eigenvalues λj .
4. (Collapse postulate) The quantum state of the system
immediately, after the measurement of |φi with respect to
X, is
Pj |φi
|φj i = r
hφ|Pj |φi
if the classical outcome is λj .
Jozef Gruska October 27, 2020 45
Quantum Computing 4 – Hilbert Space Basics, 2020

5. (Born’s rule) The probability that this particular outcome

is found as the measurement result is
pj = ||Pj |φi||2 = hφ|Pj2|φi = hφ|Pj |φi,
where we used the property 2.
6. If we perform the measurement as above, but we do not
record the result, ten the postmeasurement state can be
described by the density operator
ρ= pj |φj ihφj | = Pj |φihφ|Pj .
X X

j j

The above six rules (postulates) describe what happens to

the system during the measurement if it was initially in a
pure state.

Jozef Gruska October 27, 2020 46

Quantum Computing 4 – Hilbert Space Basics, 2020

MEASUREMENT of MIXED STATES

If the quantum system is initially in a mixed state

ρ, then the last three postulates are to be replaced
by their immediate generalisations:
1. The projectors Pj = |jihj| span the entire
Hilbert space, that is Pj Pj = 1.
2. From the orthogonality of the basis states we
have PiPj = δij Pi. in particular, Pi2 = Pi,
what implies that eigenvalues of any projector
are −1 and 1.
3. Each measurement of a state |φi with respect
to X yields, on the classical level, one of the
eigenvalues λj .
4. The quantum state of the system after the
measurement is
Pj ρPj Pj ρPj
ρj = =
Tr(Pj ρPj ) Tr(Pj ρ)
if the classical outcome is λj .
Jozef Gruska October 27, 2020 47
Quantum Computing 4 – Hilbert Space Basics, 2020

5. The probability that this particular outcome is

found as the measurement result is
pj = Tr(Pj ρPj ) = Tr(Pj2ρ) = Tr(Pj ρ)
where, again, we used the property 2.
6. If measurement is performed, but result is not
recorded, then the postmeasurement state can
be described by the density operator
ρ0 = λj Pj ρPj .
X

Jozef Gruska October 27, 2020 48

Quantum Computing 4 – Hilbert Space Basics, 2020

SUPEROPERATORS

• A superoperator (SO) is a linear

mapping/map on linear operators of a
Hilbert space.
• A positive superoperator (PSO) is a
superoperator that maps density
matrices into density matrices.
• A completely positive
superoperator/map (CPO) G is a PSO
such that, for all positive integer m,
G ⊗ Im is also a PSO, where Im is the
identity matrix.
CPO are exactly (all) physically allowed
transformations on density matrices.
Examples: encoders, decoders, quantum
channels, quantum measurements.

Jozef Gruska October 27, 2020 49

Quantum Computing 4 – Hilbert Space Basics, 2020

CPO — INFORMAL VIEW

Informally, the best way is to see a CPO

S applied to a state |φi of a Hilbert
space H as follows:
Take at first an auxiliary state (called
usually ancilla) of another Hilbert space
H 0, then apply a unitary operator U to
the state |φi ⊗ |ψi and, finally, discard
(trace out) the H 0-part of the resulting
state.

Jozef Gruska October 27, 2020 50

Quantum Computing 4 – Hilbert Space Basics, 2020

WHAT ARE QUANTUM OPERATIONS?

The main question we deal with in this section
is very fundamental.
What are physically realizable operations one
can perform (at least theoretically) on (mixed)
states (to get again (mixed) states )?
In closed quantum systems unitary operations
are actually the only quantum operations that
are available. Measurements are actually outside
of the closed quantum system framework, an
interface from the quantum to classical world,
but surely they are operations we should
consider as physically realizable.
Of the main importance are quantum operations
in open quantum systems. Actually, all actions
that are performed in/on quantum systems are
quantum operations: unitary operations ,
measurements, channel transmissions, flow of
time, noise impacts, ....
The concept of quantum operations is therefore
very general and very fundamental.
It is perhaps a bit surprising, but actually nice,
useful and natural, that we can actually study
and consider open quantum systems in the
Jozef Gruska October 27, 2020 51
Quantum Computing 4 – Hilbert Space Basics, 2020

framework of closed quantum systems. We can

consider as the basic setting that our (principal)
quantum system and its environment form a
closed quantum system in which we operate.
The requirement to consider only physically
realizable (at least theoretically) operation is, of
course, logical. As we shall see this question
has, in a sense and at least theoretically, clear
and simple answer. They are, as discussed later,
trace preserving completely positive linear
maps.

Jozef Gruska October 27, 2020 52

Quantum Computing 4 – Hilbert Space Basics, 2020

THREE APPROACHES

There are basically three main approaches to

define what are “physically realizable quantum
operations” (superoperators) E.
A physically motivated axiomatic approach says
that for a Hilbert space H we should consider as
physically realizable operations/maps
B(H) → B(H) which are consistent with the
(statistical) interpretation of quantum theory.
That is physically realisable quantum operations
aremaps that are linear (to preserve
superpositions), positive and trace preserving
(to map density operators to density operators)
and actually completely positive (to be sure that
if a superoperator is applied to a subsystem,
then the whole system is again in a quantum
state).
Jozef Gruska October 27, 2020 53
Quantum Computing 4 – Hilbert Space Basics, 2020

A pragmatic approach says that superoperators

are those operations that can be combined from
unitary operations, adding ancillas, performing
(non-selective) projective measurement and
discarding subsystems (ancillas), by performing
a tracing out operation.
A mathematical approach says that all basic
quantum operations: adding and discarding
quantum subsystems, unitary operations and
non-selective projective measurements are those
operations that have Kraus operator-sum
representation
k
ρ→ EiρEi†,
X

i=1
where so called Kraus operators Ei : H → H are
not necessarily Hermitian operators, but they
should be positive and should form a
“decomposition of the identity operator”, that is
it should hold:,
Pk †
i=1 Ei Ei = IH – so called completeness condition.
It is a consequence of the completeness
condition, and a property of trace operation,
that for any superoperator E holds
Tr(E(ρ)) = Tr( EiρEi†) = Tr( Ei†Eiρ) = Tr(( Ei†Ei)ρ) = Tr(ρ
X X X

i i i

Jozef Gruska October 27, 2020 54

Quantum Computing 4 – Hilbert Space Basics, 2020

STINESPRING DILATION THEOREM

So called Stinespring dilation theorem,

discussed below, says that each
superoperator can be realized in “one
big three-stage-step” : adding an ancilla,
performing a unitary operation on a
composed quantum system and, finally,
discarding the ancilla, see Figure 1
ρ Ε(ρ)
U
|φ><φ|

Figure 1: A Stinespring realization of a superoperator. In this view a superoperator E performs the mapping
E(ρ) = Tra (U (ρ × ρa )U † ), where ρa is the “initial state”, for example |φihφ| of an ancilla subsystem, U is a
unitary operation on composed system and, finally, a tracing out operation is performed.

Jozef Gruska October 27, 2020 55

Quantum Computing 4 – Hilbert Space Basics, 2020

POVM (GENERALIZED QUANTUM MEASUREMENT

Most general quantum observables

(measurements) , so called POVM
measurements, are given by a set {Ei}i of
positive operators 0 ≤ Ei ≤ I such that Pi Ei = I.
A measurement of a state ρ with respect to such
an observable provides i-th outcome with the
probability
Tr[ρEi].
The idea of POVM occurs naturally when we
consider projective measurements on copositive
systems. Indeed,the projective measurement on
the tensor product Hilbert space of subsystems
A and B may not remain projective on the
Hilbert space associated with A and may result
in a POVM on it.
By Neumark’s theorem a POVM measurement
on a Hilbert space can always be realized as
projective measurement in a larger Hilbert
space.

Jozef Gruska October 27, 2020 56

Quantum Computing 4 – Hilbert Space Basics, 2020

APPENDIX

Jozef Gruska October 27, 2020 57

Quantum Computing 4 – Hilbert Space Basics, 2020

SCHMIDT DECOMPOSITION THEOREM

If |ψi is a state in a bipartite Hilbert space HA ⊗ HB , and

{|αii}ni=1 is a basis of HA, {|βj i}mj=1 is a basis of HB , then
n m
|ψi = pij |αii ⊗ |βj i
X X

i=1 j=1

for some amplitudes pij .

Schmidt decomposition theorem says that |ψi can be
expressed also through a one-sum superposition and not only
through a two-sums superposition as above, what very often
makes considerations and proofs much simpler.
Theorem If |ψi is a vector of a bipartite Hilbert space
HA ⊗ HB , then there exists an orthogonal basis {|αii} of
HA and an orthogonal basis {|βj i|i} of HB and nonnegative
integers {pk } such that
X√
|ψi = pk |αk i ⊗ |βk i.
k
√
The coefficients pk are called Schmidt coefficients and
k = 1, ..., min{n, m}.

Jozef Gruska October 27, 2020 58

Quantum Computing 4 – Hilbert Space Basics, 2020

PARTIAL TRACE and SCHMIDT DECOMPOSITIO

If a state |ψi of a bipartite Hilbert space HA ⊗ HB has the

Schmidt decomposition
X√
|ψi = pk |αk i ⊗ |βk i.
k
where {|αii} and {|βj i} of HB are orthogonal bases of HA
and HB , then to trace out any of the subsystems is easy.
Indeed

Tr HB (|ψihψ|) = pk |αk ihαk |

X

k
and

Tr HA (|ψihψ|) = pk |βk ihβk |

X

Jozef Gruska October 27, 2020 59

Quantum Computing 4 – Hilbert Space Basics, 2020

PARTIAL TRACE - A USEFUL FACT

The following fact concerning tracing out

operation are often useful:
• If HA ⊗ HB is a bipartite system and ρ a state
on it, then an application of a unitary operation
U on A commute with operation of tracing out
system B. Namely
TrB ((U ⊗ I)ρ(U † ⊗ I)) = U (TrB ρ)U †.

Jozef Gruska October 27, 2020 60

Quantum Computing 4 – Hilbert Space Basics, 2020

MATHEMATICAL versus PHYSICAL NOTATION

If |φi is a vector of a Hilbert space H1 and |ψi of a Hilbert

space H2, then
|φi ⊗ |ψi = |φi|ψi
is a vector of H1 ⊗ H2.
If we want to be more precise about to which Hilbert space
vectors belong we specify them explicitly through indices as
follows, for example,
|φ1i|ψ2i.
In such a case for dual vectors mathematicians usually write
(|φ1i|ψ2i)† = hφ1|h|ψ2|.
However, physicists usually write
(|φ1i|ψ2i)† = hψ2|hφ1|.
If this convention is used, then we have
(|φi|ψi)†|αi|βi = hψ|hφ||αi|βi = hφ|αihψ|βi
Moreover it holds
(|φ1i ⊗ |ψ2i)(hα1| ⊗ |ψ2) = |α1ihα1| ⊗ |ψ2ihβ2|.

Jozef Gruska October 27, 2020 61