2

Quantum mechanics: Hilbert space formalism

Classical mechanics can describe physical properties of macroscopic objects,

whereas quantum mechanics can describe physical properties at the micro-
scopic scale. Many quantum mechanical phenomena are counter-intuitive,
and researchers have developed mathematical models to explain these phe-
nomena. In particular, the Hilbert space model has been used successfully
to explain and predict behaviors of quantum systems. Quantum information
and quantum computation concern the use of quantum properties to store,
transmit, and process information. It is important to understand the basic
rules governing quantum systems. These basic rules are summarized as the
postulates of quantum mechanics, which cannot be proven theoretically but
they are justified through empirical facts and by numerous experiments. By
the Hilbert space model, one can use vectors and linear maps to give inter-
pretation to these postulates, and explain the counter-intuitive phenomena.
Furthermore, one can develop theory and design algorithms in quantum infor-
mation and quantum computing using mathematical tools if one has a good
understanding of the mathematical model of quantum mechanics. There are
different sets of postulates for quantum mechanics. Here we give one that
turns out to be most convenient in the study of quantum information and
quantum computation. We will use the matrix concepts presented in Chapter
1 to illustrate the postulates of quantum mechanics and explain how quan-
tum properties are used in quantum information and quantum computation
in our subsequent discussions. We recommend [1, 2, 3, 4, 5, 6] for a general
introduction to quantum mechanics.

2.1 Postulates of Quantum Mechanics

Quantum mechanics was discovered in early 20th century. In spite of its long
history, the interpretation of the wave function, which describes the state of a
quantum system, remains an open question. Here we adopt the most popular
one, called the Copenhagen interpretation.

A1 A pure state in quantum mechanics is represented in terms of a normal-

ized vector |ψ⟩ in a Hilbert space H (a complex vector space with an

31
32 QUANTUM COMPUTING

inner product): ⟨ψ|ψ⟩ = 1. Suppose two states |ψ1 ⟩ and |ψ2 ⟩ are physi-
cal states of the system. Then their linear superposition c1 |ψ1 ⟩ + c2 |ψ2 ⟩
(ck ∈ C) is also a possible state of the same system. This is called the
superposition principle.
A2 For any physical quantity (i.e., observable) a, such as energy, spin,
position and momentum, there exists a corresponding Hermitian oper-
ator A acting on the Hilbert space H. When we measure a in the state
|ψ⟩, we obtain one of the eigenvalues λj of the operator A and the state
undergoes an abrupt change to the corresponding eigenvector |λj ⟩. This
phenomenon is called the collapse of wave function. If we prepare
an ensemble of many identical state |ψ⟩ and measure a, the probability
of obtaining the outcome |λj ⟩ is pj = |⟨λj |ψ⟩|2 .
expand |ψ⟩ in terms of a complete orthonormal basis {|λj ⟩} as
Let us P
|ψ⟩ = k ck |λk ⟩, where ck = ⟨λk |ψ⟩ ∈ C is called the probability
amplitude, and consider ⟨ψ|A|ψ⟩. We find
X X X
⟨ψ|A|ψ⟩ = c∗j ck ⟨λj |A|λk ⟩ = c∗j ck λk δjk = λk |ck |2 .
j,k j,k k

Since |ck |2 = |⟨λk |ψ⟩|2 is the probability of observing λk , the above

identity shows that the expectation value of measurement outcome of a
is
⟨a⟩ = ⟨ψ|A|ψ⟩. (2.1)
Note that the average is over many measurements of a with respect
to many copies of |ψ⟩ and onlyPone outcomeP λj 2is obtained for each
measurement. Note also that k p k = k |ck | = 1 is guaranteed
because of the normalization condition.
A3 The time evolution of a state is governed by the Schrödinger equation
d|ψ⟩
iℏ = H|ψ⟩, (2.2)
dt
where ℏ is a physical constant known as the Planck constant and H
is a Hermitian operator (matrix) corresponding to the energy of the
system and is called the Hamiltonian.
Several comments are in order.
ˆ In Axiom A1, the phase of the vector may be chosen arbitrarily; |ψ⟩ in
fact represents the “ray” {eiα |ψ⟩ |α ∈ R}. This is called the ray rep-
resentation. In other words, we can totally ignore the overall phase of
a vector since it has no observable consequence. For example, neither
the probability |⟨λk |ψ⟩|2 nor the average ⟨ψ|A|ψ⟩ depends on the phase.
Note, however, that the relative phase of two different states is mean-
ingful. Although |⟨ϕ|eiα ψ⟩|2 is independent of α, |⟨ϕ|ψ1 + eiα ψ2 ⟩|2 does
depend on α.
Quantum mechanics: Hilbert space formalism 33

ˆ Axiom A2 is the most counter-intuitive quantum phenomenon which

puzzles people. Here, we explain its meaning in mathematical terms.
First, an expectation value of a can be viewed as a numerical quantity
assigned to a quantum system represented by a state |ψ⟩. Since |ψ⟩
and eiα |ψ⟩ represent the same state, we want to assign this value to
|ψ⟩⟨ψ|, which is independent of the phase α. (As we shall see, this is
indeed to the density operator representation of a quantum state.) It is
believed this process is linear on the linear space spanned by all density
operators. By Hilbert space theory, specifically, the Riesz representation
theorem of linear functional, there is a Hermitian operator A such that
the measured quantity is ⟨ψ|A|ψ⟩. Suppose H has dimension n, then
n
we can identify it with
Pn C . The Hermitian operator A has a spectral
decomposition A = j=1 λj |λj ⟩⟨λj |. One can expand |ψ⟩ in terms of
Pn
|λj ⟩ as |ψ⟩ = j=1 cj |λj ⟩, where cj = ⟨λj |ψ⟩ for j = 1, . . . , n. Then the
probability of observing λj upon measurement of a is |ci |2 , and therefore
the expectation value after many measurements is
X X X
⟨ψ|A|ψ⟩ = c∗r cs ⟨λr |A|λs ⟩ = c∗r cs λs δrs = λr |cr |2 .
r,s r,s r

This measurement is called the projective measurement. Any par-

ticular outcome λj will be found with the probability

|cj |2 = ⟨ψ|Pj |ψ⟩, (2.3)

where Pj = |λj ⟩⟨λj | is the projection operator, and the state immedi-
ately after the measurement is |λj ⟩ or equivalently

P |ψ⟩
p j , (2.4)
⟨ψ|Pj |ψ⟩
where the overall phase has been ignored.
The discussion is similar but more involved if H has an infinite dimen-
sion.
ˆ The Schrödinger equation (2.2) in Axiom A3 is formally solved to yield

|ψ(t)⟩ = e−iHt/ℏ |ψ(0)⟩, (2.5)

if the Hamiltonian H is time-independent, while

i t
 Z 
|ψ(t)⟩ = T exp − H(t)dt |ψ(0)⟩ (2.6)
ℏ 0
if H depends on t, where T is the time-ordering operator defined by

A(t1 )B(t2 ), t1 > t2
T [A(t1 )B(t2 )] = ,
B(t2 )A(t1 ), t2 ≥ t1
34 QUANTUM COMPUTING

for a product of two operators. Note that T is required since H(t1 ) and
H(t2 ) do not commute with each other in general if t1 ̸= t2 . General-
ization to products of more than two operators should be obvious. We
write Eqs. (2.5) and (2.6) as

i t
 Z 
|ψ(t)⟩ = U (t)|ψ(0)⟩, where U (t) = T exp − H(t)dt .
ℏ 0

The operator U (t) : |ψ(0)⟩ 7→ |ψ(t)⟩, which we call the time-

evolution operator, is unitary. Unitarity of U (t) guarantees that
the norm of |ψ(t)⟩ is conserved: ⟨ψ(t)|ψ(t)⟩ = ⟨ψ(0)|U † (t)U (t)|ψ(0)⟩ =
⟨ψ(0)|ψ(0)⟩ = 1.
Extensive use of unitary matrices is made in quantum information and
quantum computing. These matrices are implemented by manipulating
the Hamiltonian by making use of (2.6). Efficient implementation of
unitary matrices is an active area of quantum control theory.
It might be instructive to think about the scalar case and compare the
situation with the one variable differential equation x′ (t) = (−ig(t))x(t)
for a real-valued function g(t), for which time-ordering is not necessary.

2.2 Multipartite System

So far, we have assumed implicitly that the system is made of a single com-
ponent. Suppose a system is made of two components; one lives in a Hilbert
space H1 and the other in another Hilbert space H2 . A system composed of
two separate components is called bipartite. Then the system as a whole
lives in a Hilbert space H = H1 ⊗ H2 , whose general vector is written as
X
|ψ⟩ = cij |e1,i ⟩ ⊗ |e2,j ⟩, (2.7)
i,j

where {|ea,i ⟩} (a = 1, 2) is an orthonormal basis in Ha and i,j |cij |2 = 1.

P
A state |ψ⟩ ∈ H written as a tensor product of two vectors as |ψ⟩ = |ψ1 ⟩ ⊗
|ψ2 ⟩, (|ψa ⟩ ∈ Ha ) is called a separable state or a tensor product state.
A separable state admits a classical interpretation such as “The first system
is in the state|ψ1 ⟩,while the second system is in |ψ2 ⟩, and viewed as a vector
|ψ1 ⟩
of the form ”. It is clear such a set of vectors spans a linear space
|ψ2 ⟩
of dimension dimH1 + dimH2 . Note however that the total space H has
different dimensions since we find, by counting the number of coefficients in
(2.7), that dimH = dimH1 dimH2 . This number is considerably larger than
the dimension of the separable states when dimHa (a = 1, 2) are large. What
Quantum mechanics: Hilbert space formalism 35

are the missing states then? Suppose

 the spin
 states or a particle are denoted
1 0
and represented by | ↑⟩ = and | ↓⟩ = . Then the spin states for two
0 1
particles are
       
1 0 0 0
0 1 0 0
0 , | ↑⟩ ⊗ | ↓⟩ = 0 , | ↓⟩ ⊗ | ↑⟩ = 1 , | ↓⟩ ⊗ | ↓⟩ = 0 .
| ↑⟩ ⊗ | ↑⟩ =        

0 0 0 1

Consider a spin state

1
|ψ⟩ = √ (| ↑⟩ ⊗ | ↑⟩ + | ↓⟩ ⊗ | ↓⟩) (2.8)
2

of two separated electrons. Suppose |ψ⟩ may be decomposed as

|ψ⟩ = (c1 | ↑⟩ + c2 | ↓⟩) ⊗ (d1 | ↑⟩ + d2 | ↓⟩)

= c1 d1 | ↑⟩ ⊗ | ↑⟩ + c1 d2 | ↑⟩ ⊗ | ↓⟩ + c2 d1 | ↓⟩ ⊗ | ↑⟩ + c2 d2 | ↓⟩ ⊗ | ↓⟩.

However this decomposition is not possible since we must have

1
c1 d2 = c2 d1 = 0, c1 d1 = c2 d2 = √
2

simultaneously, and it is clear that the above equations have no common

solution. Therefore the state |ψ⟩ is not separable.
Such non-separable states are called entangled in quantum theory [9]. The
fact
dimH1 dimH2 ≫ dimH1 + dimH2
tells us that most states in a Hilbert space of a bipartite system are entangled
when the constituent Hilbert spaces are higher dimensional. These entangled
states refuse classical descriptions. Entanglement will be used extensively as
a powerful computational resource in quantum information processing and
quantum computation.
Suppose a bipartite state (2.7) is given. We are interested in when the state
is separable and when entangled.
In case H1 = Cm and H2 = Cn , we may choose B1 = {|e1,1 ⟩, . . . , |e1,m ⟩}
and B2 = {|e2,1 ⟩, . . . , |e2,n ⟩} to be the standard bases for Cm and Cn , i.e.,
|e1,1 ⟩, . . . , |e1,m ⟩ are the columns of Im and |e2,1 ⟩, . . . , |e2,n ⟩ are the columns
of In . Then
B = {|e1,r ⟩ ⊗ |e2,s ⟩ : 1 ≤ r ≤ m, 1 ≤ s ≤ n}
is the standard basis of Cmn = Cm ⊗ Cn . This can be seen by checking that
B is an orthonormal set with mn vectors.
36 QUANTUM COMPUTING

We can use the SVD to express |ψ⟩ = r,s crs |e1,r ⟩ ⊗ |e2,s ⟩ ∈ Cm ⊗ Cn in
P
a simpler form and detect whether |ψ⟩ is entangled or not as follows. For the
SVD of
Xk
(crs ) = U ΣV † = sj |uj ⟩⟨vj |,
j=1
∗
we can set |f1,j ⟩ = |uj ⟩ and |f2,j ⟩ = |vj ⟩ , the conjugate of |vj ⟩. Then

X k
X
|ψ⟩ = crs |e1,r ⟩ ⊗ |e2,s ⟩ = sj |f1,j ⟩ ⊗ |f2,j ⟩.
r,s j=1

Clearly, |ψ⟩ is a product state if and only if (crs ) has rank one, i.e., k = 1 and
s1 = 1.
EXAMPLE 2.2.1. Let {|e1,1 ⟩, |e1,2 ⟩, |e1,3 ⟩} and {|e2,1 ⟩, |e2,2 ⟩⟩} be the stan-
dard bases of C3 and C2 , and consider the bipartite state
1
|ψ⟩ = (|e1,1 ⟩|e2,1 ⟩ + |e1,1 ⟩|e2,2 ⟩ + i|e1,3 ⟩|e2,1 ⟩ + i|e1,3 ⟩|e2,2 ⟩) ∈ C2 ⊗ C3
2
whose coefficients form a matrix


11
1
C = 0 0.
2
i i

It is immediate that the matrix is rank one, and i,j |cij |2 = 1. So, |ψ⟩ is a
P
product state. We can find a simple representation of |ψ⟩ using the SVD of
C obtained in Example 1.6.2. We have

|ψ⟩ = |f1,1 ⟩|f2,1 ⟩,

where
3
X 1
|f1,1 ⟩ = Ui1 |e1,i ⟩ = √ (|e1,1 ⟩ + i|e1,3 ⟩)
i=1
2
and
2
X
∗ 1
|f2,1 ⟩ = Vj1 |e2,j ⟩ = √ (|e2,1 ⟩ + |e2,2 ⟩).
j=1
2

In general, we have the following.

PROPOSITION 2.2.2. Let H = H1 ⊗ H2 be the Hilbert space of a bipartite
system. Then a vector |ψ⟩ ∈ H admits the Schmidt decomposition
r
X X
|ψ⟩ = si |f1,i ⟩ ⊗ |f2,i ⟩ with s2j = 1, (2.9)
i=1 i
Quantum mechanics: Hilbert space formalism 37

where si > 0 are called the Schmidt coefficients and {|fa,i ⟩} is an orthonor-
mal set of Ha . The number r ∈ N is called the Schmidt number of |ψ⟩.

Proof. This is a direct consequence of SVD introduced in §1.6. Let |ψ⟩ be ex-
panded as in Eq. (2.7). Note that the coefficients cij form a dimH1 × dimH2
matrix C. We apply the SVD to obtain C = U ΣV † , where U and V are
unitary matrices and Σ is a matrix whose (j, j) elements are nonnegative real
numbers while all the other elements vanish. Now |ψ⟩ of Eq. (2.7) is put in
the form X
|ψ⟩ = Uik Σkl Vjl∗ |e1,i ⟩ ⊗ |e2,j ⟩.
i,j,k,ℓ
P P ∗
Now define |f1,k ⟩ = i Uik |e1,i ⟩ and |f2,k ⟩ = j Vjk |e2,j ⟩. Unitarity of U and
V guarantees that they are orthonormal vectors of H1 and H2 , respectively.
By noting that the (k, l) entry of Σ is sk δkl , we obtain
r
X
|ψ⟩ = sj |f1,j ⟩ ⊗ |f2,j ⟩,
j=1

P r is the number of positive (diagonal) elements in Σ. Since ⟨ψ|ψ⟩ =

where
1 = r,s |crs |2 , which is the trace of C † C, and therefore equal to the sum of
the eigenvalues of C † C, i.e., j s2j .
P

It follows from the above proposition that a bipartite state |ψ⟩ is separable
if and only if its Schmidt number r is 1.
Generalization to a system with more components, i.e., a multipartite
system, should be obvious. A system composed of N components has a
Hilbert space
H = H1 ⊗ H 2 ⊗ . . . ⊗ H N , (2.10)
where Ha is the Hilbert space to which the ath component belongs. Classifi-
cation of entanglement in a multipartite system is far from obvious, and an
analogue of the Schmidt decomposition is not known to date for N ≥ 3.∗

2.3 No-Cloning Theorem

We copy classical data almost every day. In fact, this is amongst the most
common functions with digital media. (Of course we should not copy media
that are copyright protected.) However, the situation is different in quantum
mechanics. The no-cloning theorem below asserts that it is impossible to
create an independent and identical copy of an arbitrary unknown quantum

∗ See, however, [10, 11].

38 QUANTUM COMPUTING

state. As shown in the proof, one cannot duplicate an unknown quantum

system with a quantum operation, which is a unitary transformation by (A3).
THEOREM 2.3.1. (Wootters and Zurek [19], Dieks [20]) An unknown
quantum system cannot be cloned by unitary transformations.
Proof. Suppose there would exist a unitary transformation U that makes a
clone of a quantum system. Namely, suppose U acts, for any state |φ⟩, as
U : |φ0⟩ → |φφ⟩. (2.11)
Let |φ⟩ and |ϕ⟩ be two states that are linearly independent. Then we should
have U |φ0⟩ = |φφ⟩ and U |ϕ0⟩ = |ϕϕ⟩ by definition. Then the action of U on
1
|ψ⟩ = √ (|φ⟩ + |ϕ⟩) yields
2
1 1
U |ψ0⟩ = √ (U |φ0⟩ + U |ϕ0⟩) = √ (|φφ⟩ + |ϕϕ⟩).
2 2
If U were a cloning transformation, we must also have
1
U |ψ0⟩ = |ψψ⟩ = (|φφ⟩ + |φϕ⟩ + |ϕφ⟩ + |ϕϕ⟩),
2
which contradicts the previous result. Therefore, there does not exist a unitary
cloning transformation.

The no-cloning theorem may be proved by using the special theory of rela-
tivity, which assumes no information can propagate faster than the speed of
light. [27]
Suppose Alice and Bob share a Bell state
1 1
|Ψ− ⟩ = √ (|0⟩|1⟩ − |1⟩|0⟩) = √ (|−⟩|+⟩ − |+⟩|−⟩). (2.12)
2 2
where |±⟩ = √12 (|0⟩ ± |1⟩). Readers are encouraged to verify the second
equality. Alice keeps the first qubit while Bob keeps the second. If Alice
wants to send Bob a bit “0”, she measures her qubit in {|0⟩, |1⟩} basis while
if she wants to send “1”, she employs {|+⟩, |−⟩} basis for her measurement.
Bob always measures his qubit in {|0⟩, |1⟩} basis.
After Alice’s measurment and before Bob’s measurment, Bob’s qubit is |0⟩
or |1⟩ if Alice sent “0” while it is |+⟩ or |−⟩ if Alice sent “1”.
Suppose Bob is able to clone his qubit. He makes many copies of his qubit
and measures them in {|0⟩, |1⟩} basis. If Alice sent “0”, Bob will obtain
0, 0, 0, . . . or 1, 1, 1, . . . while if she sent “1”, Bob will obtain approximately
50% of 0’s and 50% of 1’s. Suppose Bob received |±⟩ and made N clones, then
the probability of obtaining the same outcome is 1/2N −1 , which is negligible
if N is sufficiently large. Note that Bob obtains the bit Alice wanted to send
immediately after Alice’s measurement assuming it does not take long to clone
his qubit. This could happen even if Alice and Bob are separated many light
years apart, thus in contradiction with the special theory of relativity.
Quantum mechanics: Hilbert space formalism 39

2.4 Qubits
A (Boolean) bit assumes two distinct values, 0 and 1. Bits constitute the
building blocks of the classical information theory founded by C. Shannon.
Quantum information theory, on the other hand, is based on qubits.

2.4.1 Qubit and Bloch Sphere

A qubit is a (unit) vector in the vector space C2 , whose basis vectors are
denoted as    
1 0
|0⟩ = and |1⟩ = . (2.13)
0 1
What these vectors physically mean depends on the physical realization em-
ployed for quantum-information processing.
ˆ In some cases, |0⟩ stands for a horizontally polarized photon | ↔⟩, while
|1⟩ represents a vertically polarized photon | ↕⟩. Alternatively they
might correspond to photons polarized in different directions. For ex-
ample, |0⟩ may represent a polarization state
1
| ⟩ = √ (| ↔⟩ + | ↕⟩),
↔

2
while |1⟩ represents a state
1
| ⟩ = √ (| ↔⟩ − | ↕⟩).
↔

2
Note that if | ↔⟩ (| ↕⟩) corresponds to an eigenstate of σz with the eigen-
value +1 (−1), respectively, then | ⟩ (| ⟩) corresponds to an eigenstate
↔

of σx with the eigenvalue +1 (−1), respectively.

Similarly, the states
1 1
|σ + ⟩ = √ (| ↔⟩ + i| ↕⟩), |σ − ⟩ = √ (| ↔⟩ − i| ↕⟩)
2 2
correspond to the eigenstates of σy with the eigenvalues ±1 and repre-
sent circularly polarized photons.
ˆ They may represent spin states of an electron, |0⟩ = | ↑⟩ and |1⟩ =
| ↓⟩. Electrons are replaced by nuclei with spin-1/2 in NMR quantum
computing.
ˆ Truncated two states from many levels may also be employed as a qubit.
Take the ground state and the first excited state of ionic energy levels
or atomic energy levels, for example. We may assign |0⟩ to the ground
state and |1⟩ to the first excited state.
40 QUANTUM COMPUTING

In any case, we have to fix a set of basis vectors when we carry out quantum
information processing. All the physics should be described with respect to
this basis. In the following, the basis is written in an abstract form as {|0⟩, |1⟩},
unless otherwise stated.
Note that the third example of a qubit above suggests that a quantum
system with more than two states may be employed for information storage
and information processing. If a quantum system admits three different states,
it is called a qutrit, while if it takes d different states, it is called a qudit.
A spin S particle, for example, takes d = 2S + 1 spin states and works as a
qudit. The significance of qutrits and qudits in information processing is still
to be explored.
Qubits can be used as a Hardware random number generator or true
random number generator to generate cryptographic keys. (See [28] for
background and references for random number generation.) Based on the
quantum postulates, one can generate a zero-one sequence by measuring a
sequence of quibts in a state √12 (|0⟩ + |1⟩) one by one.
It is convenient to assume the vector |0⟩ corresponds to the classical value
0, while |1⟩ to 1 in quantum computation. Moreover it is possible for a qubit
to be in a superposition state:
|ψ⟩ = a|0⟩ + b|1⟩ with a, b ∈ C, |a|2 + |b|2 = 1. (2.14)
The fundamental requirement of quantum mechanics is that if we make mea-
surement on |ψ⟩ to see whether it is in |0⟩ or |1⟩, the outcome will be 0 (1) with
the probability |a|2 (|b|2 ), and the state immediately after the measurement
is |0⟩ (|1⟩).
Although a qubit may take infinitely many different states, it should be kept
in mind that we can extract from it as the same amount of information as that
of a classical bit. Information can be extracted only through measurements.
When we make measurement on a qubit, the state vector “collapses” to the
eigenvector that corresponds to the eigenvalue observed. Suppose that a spin
is in a state a|0⟩ + b|1⟩. If we observe that the z-component of the spin
is +1/2, the system immediately after the measurement is definitely in the
state |0⟩. This happens with probability ⟨ψ|0⟩⟨0|ψ⟩ = |a|2 . The outcome
of a measurement on a qubit is always one of the eigenvalues, which we call
abstractly 0 and 1, just like for a classical bit. As a result, if we can make
measurements of a large number of copies of unknown state, we may estimate
|a| and |b|, but we cannot estimate the coefficients a and b.
It is useful, for many purposes, to express a state of a single qubit graph-
ically. Note that every qubit |ψ⟩ = a|0⟩ + b|1⟩ corresponds to the rank one
orthogonal projection
 
1 1 + nz nx − iny 1
ρ = |ψ⟩⟨ψ| = = (I2 + nx σx + ny σy + nz σz ),
2 nx + iny 1 − nz 2
where σx , σy , σz are the Pauli matrices, (nx − iny )/2 = ab̄, nz = 2|a|2 − 1.
As we will see in the next chapter, the matrix ρ is a density matrix corre-
Quantum mechanics: Hilbert space formalism 41

sponding to a pure state. We use the notation in Proposition 1.5.8, and let
n̂ = (nx , ny , nz ) ∈ R3 be a unit vector, σ = (σx , σy , σz ), and
 
nz nx − iny
A = n̂ · σ = .
nx + iny −nz
Then
A = P1 − P2 with P1 = (I + A)/2 and P2 = (I − A)/2,
where |ψ⟩⟨ψ| = P1 is precisely the eigenprojection of the eigenvalue 1 of |ψ⟩⟨ψ|.
In particular, we may parametrize a one-qubit state |ψ⟩ with θ and ϕ as
θ θ
|ψ⟩ = |ψ(θ, ϕ)⟩ = cos |0⟩ + eiϕ sin |1⟩, θ ∈ [0, π], ϕ ∈ [0, 2π), (2.15)
2 2
where we take advantage of the freedom of the phase to make the coefficient
of |0⟩ real. Then
1
|ψ⟩⟨ψ| = (I + n̂ · σ)
2
with n̂ = (sin θ cos ϕ, sin θ sin ϕ, cos θ) corre-
sponding to the unique point n̂ on the unit
sphere in R3 using the polar co-ordinates. So,
there is a natural correspondence between a
unit vector n̂(θ, ϕ) and a state vector |ψ(θ, ϕ)⟩.
Namely, a state |ψ(θ, ϕ)⟩ is expressed as a unit
vector n̂(θ, ϕ) on the surface of the unit sphere,
The Bloch sphere
called the Bloch sphere.
Pr
A more general density matrix ρ can be constructed as ρ = j=1 pr |ψj ⟩⟨ψj |,
where |ψ1 ⟩⟨ψ1 |, . . . , |ψr ⟩⟨ψr | are rank one orthogonal projections and
p1 , . . . , pr are positive numbers summing up to one. Then ρ has the form
 
1 X
ρ= I+ uj σi  , (2.16)
2 j=x,y,z

where ui are components of a real vector u satisfying |u| ≤ 1. The Hermitian

matrix ux σx + uy σy + uz σz has trace 0 and determinant −|u|2 = −(u2x + u2y +
u2z ). So, it has eigenvalues ±|u|, and hence the eigenvalues of ρ are
1 1
λ+ = (1 + |u|) , λ− = (1 − |u|) (2.17)
2 2
and therefore non-negative. In case |u| = 1, the eigenvalue λ− vanishes and
rank ρ = 1. Therefore the surface of the unit ball corresponds to pure states.
The converse is also shown easily. The ball is called the Bloch ball, and
its boundary called the Bloch sphere. The vector u is also called the Bloch
vector. The normalized vector n̂ of the Bloch sphere is a special case of u
restricted in pure states.
42 QUANTUM COMPUTING

2.4.2 Multi-Qubit Systems and Entangled States

Let us consider a group of many (n) qubits next. Such a system behaves
quite differently from a classical one, and this difference gives a distinguishing
aspect to quantum information theory. An n-qubit system is often called a
(quantum) register in the context of quantum computing.
Consider a classical system made of several components. The state of this
system is completely determined by specifying the state of each component.
This is not the case for a quantum system. A quantum system made of
many components is not necessarily described by specifying the state of each
component as we have learned in §2.2.
As an example, let us consider an n-qubit register. Suppose we specify
the state of each qubit separately in analogy with a classical case. Each of
the qubits is then described by a two-dimensional complex vector of the form
ai |0⟩ + bi |1⟩, and we need 2n complex numbers {ai , bi }1≤i≤n to specify the
state. This corresponds the the tensor product state

(a1 |0⟩ + b1 |1⟩) ⊗ (a2 |0⟩ + b2 |1⟩) ⊗ . . . ⊗ (an |0⟩ + bn |1⟩)

introduced in §2.2. If the system is treated in a fully quantum-mechanical

way, however, we have to include superposition of such tensor product states,
which is not necessarily decomposable into a tensor product form. Such a state
is entangled (see §2.2). A general state vector of the register is represented
as
X
|ψ⟩ = ci1 i2 ...in |i1 ⟩ ⊗ |i2 ⟩ ⊗ . . . ⊗ |in ⟩
ik =0,1

and lives in a 2n -dimensional complex vector space. Note that 2n ≫ 2n

for a large number n. The ratio 2n /2n is ∼ 6.3 × 1027 for n = 100 and
∼ 5.4 × 10297 for n = 1000. These astronomical numbers tell us that most
quantum states in a Hilbert space with large n are entangled, i.e., they do not
have classical analogy which tensor product states have. Entangled states that
have no classical counterparts are extremely powerful resources for quantum
computation and quantum communication as we will show later.
Let us consider a system of two qubits for definiteness. The combined
system has a basis {|00⟩, |01⟩, |10⟩, |11⟩}. More generally, a basis for a system
of n qubits may be taken to be {|bn−1 bn−2 . . . b0 ⟩}, where bn−1 , bn−2 , . . . , b0 ∈
{0, 1}. It is also possible to express the basis in terms of the decimal system.
We write |x⟩, instead of |bn−1 bn−2 . . . b0 ⟩, where x = bn−1 2n−1 + bn−2 2n−2 +
. . . + b0 is the decimal expression of the binary number bn−1 bn−2 . . . b0 . Thus
the basis for a two-qubit system may be written also as {|0⟩, |1⟩, |2⟩, |3⟩} with
this decimal notation. Whether the binary system or the decimal system is
employed should be clear from the context. An n-qubit system has 2n =
exp(n ln 2) basis vectors.
Quantum mechanics: Hilbert space formalism 43

The set
1 1
{|Φ+ ⟩ = √ (|00⟩ + |11⟩), |Φ− ⟩ = √ (|00⟩ − |11⟩),
2 2
(2.18)
1 1
|Ψ+ ⟩ = √ (|01⟩ + |10⟩), |Ψ− ⟩ = √ (|01⟩ − |10⟩)}
2 2
is an orthonormal basis of a two-qubit system and is called the Bell basis.
Each vector is called the Bell state or the Bell vector. Note that all the
Bell states are entangled.
Among three-qubit entangled states, the following two states are important
for various reasons and hence deserve special names. The state
1
|GHZ⟩ = √ (|000⟩ + |111⟩) (2.19)
2
is called the Greenberger-Horne-Zeilinger state and is often abbreviated
as the GHZ state [21]. Another important three-qubit state is the W state
[22],
1
|W⟩ = √ (|100⟩ + |010⟩ + |001⟩). (2.20)
3
The GHZ state has a different entanglement pattern as the W state. If any
qubit in the GHZ state is measured, the resulting state is a separable state
while if any qubit in the W state is measured, there is a probability of 2/3 that
the resulting state is entangled. Accordingly. it is impossible to transform
the GHZ state to the W state by local operations and vice versa.

2.5 Measurement
Classical information theory is formulated independently of measurements of
the system under consideration. This is because the readout of the result
is always the same for anyone and at any time, provided that the system
processes the same information without error. This is completely different
in quantum information processing. Measurement is an essential part of the
theory as we see below.
By making a measurement on a system, we project the state vector to one of
the basis vectors that the measurement equipment defines.† Suppose we have
a state vector |ψ⟩ = a|0⟩ + b|1⟩ and measure it to see if it is in the state |0⟩
or |1⟩. Depending on the system, this means if a spin points up or down or a
photon is polarized horizontally or vertically, for example. The result is either

† This is called a projective measurement as was noted in Section 2.1.

44 QUANTUM COMPUTING

0 or 1. In the first case, the state “collapses” to |0⟩ while in the second case, to
|1⟩. We find, after many measurements, the probability of obtaining outcome
0 (1) is |a|2 (|b|2 ). Here, we may assume that the observable is associate with
the Hermitian operator M = λ0 |0⟩⟨0| + λ1 |1⟩⟨1|. Note that the eigenvalues λ0
and λ1 represent the readings produced by the apparatus if the states |0⟩ and
|1⟩ are measured, respectively. They may be determined or limited by the set
up of the measuring device.
To be more formal, P we construct a Hermitian matrix corresponding to the
observable M = m λm Pm such that the eigenprojection Pm will serve as
the measurement operator Pm such that the probability of obtaining the
outcome λm in the state |ψ⟩ is

p(m) = ⟨ψ|Pm |ψ⟩, (2.21)

and the state immediately after the measurement is

Pm |ψ⟩
|m⟩ = p . (2.22)
p(m)

In the above example, the measurement operators are nothing but projection
operators; P0 = |0⟩⟨0| and P1 = |1⟩⟨1|. In fact, we have

p(0) = ⟨ψ|P0 |ψ⟩ = ⟨ψ|0⟩⟨0|ψ⟩ = |a|2 ,

and
P |ψ⟩ a
p0 = |0⟩ ≃ |0⟩,
p(0) |a|

and similarly for the other case P1 . It should be noted that a quantum state
is defined up to a phase and hence a/|a| does not play any role. ‡
P many copies of a particular state |ψ⟩. If we measure
Suppose we are given
an observable M = m λm Pm in each of the copies, the expectation value of
M is given, in terms of the projection operators, by
X X
E(M ) = λm p(m) = λm ⟨ψ|Pm |ψ⟩
m m
X
= ⟨ψ| λm Pm |ψ⟩ = ⟨ψ|M |ψ⟩, (2.23)
m
P
where use has been made of the spectral decomposition M = m λm Pm . The
standard deviation of the measurement outcomes of M is given by
p p
∆(M ) = ⟨(M − ⟨M ⟩)2 ⟩ = ⟨M 2 ⟩ − ⟨M ⟩2 . (2.24)

‡ We will consider some more general measurement operators. One may see [18] if desired.
Quantum mechanics: Hilbert space formalism 45

Let us analyze measurements in a two-qubit system in some detail. An

arbitrary state is written as

|ψ⟩ = a|00⟩ + b|01⟩ + c|10⟩ + d|11⟩, |a|2 + |b|2 + |c|2 + |d|2 = 1,

where a, b, c, d ∈ C. We make a measurement of the first qubit with respect

to the basis {|0⟩, |1⟩}. To this end, we rewrite the state as

a|00⟩ + b|01⟩ + c|10⟩ + d|11⟩

= |0⟩ ⊗ (a|0⟩ + b|1⟩) + |1⟩ ⊗ (c|0⟩ + d|1⟩)
   
a b c d
= u|0⟩ ⊗ |0⟩ + |1⟩ + v|1⟩ ⊗ |0⟩ + |1⟩ ,
u u v v
p p
where u = |a|2 + |b|2 and v = |c|2 + |d|2 . The measurement operators
acting on the first qubit are

M0 = |0⟩⟨0| ⊗ I, M1 = |1⟩⟨1| ⊗ I. (2.25)

Note that we need to specify ⊗I explicitly since we are working in a two-

qubit Hilbert space C4 . Here we may assume that the Hermitian operator
corresponding to the observable is M = M0 − M1 , say. Upon a measurement
of the first qubit, we obtain 0 with the probability

⟨ψ|M0 |ψ⟩ = u2 = |a|2 + |b|2 ,

projecting the state to

 
M |ψ⟩ a b
p0 = |0⟩ ⊗ |0⟩ + |1⟩ ,
p(0) u u
2 2 2
while weobtain |1⟩ with
 the probability v = |c| + |d| , projecting the state
c d
to |1⟩ ⊗ |0⟩ + |1⟩ . Note that the state after the measurement has unit
v v
norm in both cases. The measurement of the second qubit can be carried out
similarly. Measurements on an n-qubit system can be carried out by repeating
one-qubit measurement n times.
In the two-qubit example above, the Hilbert space for the system is sepa-
rated into a direct sum of H0 , where the first qubit is in the state |0⟩, and H1 ,
where it is in |1⟩: H = H0 ⊕ H1 . An arbitrary two-qubit state |ψ⟩ is uniquely
decomposed into two vectors, each of which belongs to H0 or H1 as

(|0⟩⟨0| ⊗ I)|ψ⟩ ∈ H0 , (|1⟩⟨1| ⊗ I)|ψ⟩ ∈ H1 ,

where normalization has been ignored. More generally, a measurement of k

qubits in an n-qubit system yields 2k possible outcomes mi (1 ≤ i ≤ 2k ).
Accordingly, the 2n -dimensional Hilbert space of the system is separated into
the direct sum of mutually orthogonal subspaces Hm1 , Hm2 , . . . , Hm2k as H =
46 QUANTUM COMPUTING

Hm1 ⊕ Hm2 ⊕ . . . ⊕ Hm2k . When the result of the measurement of the k

qubits is mi , the state after the measurement is projected to the subspace
Hmi . It should be clear from the construction that each subspace Hmi has
dimension 2n /2k = 2n−k . The measurement device projects the state before
the measurement

|ψ⟩ = cm1 |ψm1 ⟩ + cm2 |ψm2 ⟩ + . . . + cm2k |ψm2k ⟩, (|ψmi ⟩ ∈ Hmi )

into one of the subspaces Hmi randomly with the probability |cmi |2 .
Measurement gives an alternative viewpoint to entangled states. A state
is not entangled if a measurement of a qubit does not affect the state of the
other qubits. Suppose the first qubit of the state
1
√ (|00⟩ + |11⟩)
2
was measured to be 0 (1). Then the outcome of the measurement of the
second qubit is definitely 0 (1). Therefore the measurement of the first qubit
affects the outcome of the measurement on the second qubit, which shows
that the initial state is an entangled state. In other words, there exists a
strong correlation between the two qubits. This correlation may be used for
information processing as will be shown later. In contrast with this, the state
1
√ (|00⟩ + |01⟩) is not entangled since it can be written as
2
1 1
√ (|00⟩ + |01⟩) = |0⟩ ⊗ √ (|0⟩ + |1⟩).
2 2
Regardless of the measurement of the second qubit, the measurement of the
first qubit definitely yields 0. Moreover, the second qubit is measured to be
0 (1) with the probability 1/2, independently of whether the first qubit is
measured or not.

2.5.1 Einstein-Podolsky-Rosen (EPR) Phenomenon

Einstein, Podolsky and Rosen (EPR) proposed a Gedanken experiment which,
at first glance, shows that an entangled state violates an axiom of the special
theory of relativity [23]. Suppose a particle source produces the so-called EPR
pair in the state
1
|Ψ− ⟩ = √ (|01⟩ − |10⟩)
2
and it sends the first qubit to Alice and the second to Bob, who may be
separated far away (see Fig. 2.1).§ Alice measures her qubit and obtains her
reading |0⟩ or |1⟩. Depending on her reading, the EPR state is projected to

§ Alice and Bob are names frequently used in information theory.

Quantum mechanics: Hilbert space formalism 47

FIGURE 2.1
EPR pair produced by a source in the middle. One qubit is sent to Alice and
the other to Bob.

|01⟩ (|10⟩), and Bob will definitely observe |1⟩ (|0⟩) in his measurement. The
change of the state

1
√ (|01⟩ − |10⟩) → |01⟩ or |10⟩ (2.26)
2

takes place instantaneously even when they are separated by a large distance.
It seems that Alice’s measurement propageted to Bob’s qubit instantaneously,
and it violates the special theory of relativity. This is the very point EPR pro-
posed to defeat quantum mechanics, which claims that nothing can propagate
faster than the speed of light.
Note, however, that nothing has propagated from Alice to Bob and vice
versa, upon Alice’s measurement. Clearly no energy has propagated. What
about information? It is impossible for Alice to control her and hence Bob’s
readings. Therefore it is impossible to use EPR pairs to send a sensible
message from Alice to Bob. If they could, the message would be sent in-
stantaneously, which certainly violates the special theory of relativity. If a
large number of EPR pairs are sent to Alice and Bob and they independently
measure their qubits, they will observe random sequences of 0 and 1. They
notice that their readings are strongly correlated only after they exchange
their sequences by means of classical communication, which can be done at
most with the speed of light.

2.5.2 Bell inequality

About 30 years after the EPR paper was published, an experiment test was
proposed to check whether the measurement of entangled pairs follow a cer-
tain predetermined rule imposed by Nature, or the postulate of quantum
mechanics.
Here is the proposed experiments. Suppose Charlie prepares an entangled
pair of qubits (photons or particles) and sends the first one to Alice and the
second one to Bob. Alice will apply one of her two measurement schemes,
say, Q and R, each will produce a measured value in {1, −1}. Bob will also
48 QUANTUM COMPUTING

apply one of his two measurement schemes, say, S and T , each will produce
a measured value in {1, −1}.
Let us consider

QS + RS + RT − QT = (Q + R)S + (R − Q)T.

Because R, Q ∈ {1, −1}, it follows that either (Q + R)S = 0 or (R − Q)T = 0.

As a result, QS + RS + RT − QT ∈ {2, −2}.
Suppose there is a hidden rule governing the measurement outcomes, and
p(q, r, s, t) is the probability that, before the measurements are performed, the
system is in the state (Q, R, S, T ) = (q, r, s, t). Then the expectation value
E(QS + RS + RT − QT ) = E(QS) + E(RS) + E(RT ) − E(QT ) satisfies
X
|E(QS + RS + RT − QT )| = p(q, r, s, t)|qs + rs + rt − qt|
(q,r,s,t)
X
≤ p(q, r, s, t) · 2 = 2.
(q,r,s,t)

So, we get the Bell inequality

|E(QS) + E(RS) + E(RT ) − E(QT )| ≤ 2. (2.27)

Suppose Charlie prepares an entangled state

1
|Ψ− ⟩ = √ (|01⟩ − |10⟩)
2
and gives Alice the first qubit, and Bob the second one. Alice uses the mea-
surement operators Q = σz and R = σx , and Bob uses the measurement
−1
operators S = √ 2
(σz + σx ) and T = √12 (σz − σx ). Then

1 1
E(QS) = ⟨Ψ− |Q ⊗ S|Ψ− ⟩ = √ , E(RS) = ⟨Ψ− |R ⊗ S|Ψ− ⟩ = √ ,
2 2
1 −1
E(RT ) = ⟨Ψ− |R ⊗ T |Ψ− ⟩ = √ , E(QT ) = ⟨Ψ− |Q ⊗ T |Ψ− ⟩ = √ ,
2 2
and hence √ √
E(QS + RS + RT − QT ) = 4/ 2 = 2 2. (2.28)
This equality clearly violates the Bell inequality.
To determine whether (2.27) or (2.28) is valid, Alice and Bob can estimate
E(QS) by performing measurements on many copies of |Ψ− ⟩, and record their
results. After the experiments, they can multiply their measurements when
they used the measurement schemes Q and S, respectively. Similarly, they
can estimate E(RS), E(RT ), E(QT ), so as to obtain an estimate of E(QS +
RS + RT − QT ).
Quantum mechanics: Hilbert space formalism 49

Experimental results showed strong support to (2.28). Hence, the EPR

proposal that there is a hidden rule governing the measurement results of
entangled pair was ruled out.
Note that measurement of a qubit in |Ψ− ⟩ disentangles the pair. As a result,
a tensor product state |0⟩|1⟩ or |1⟩|0⟩ satisfies the Bell inequality. This fact
will be used to detect eavesdroppers in quantum key distribution protocols.

2.6 Additional Topics

2.6.1 The uncertainty principle
Historically, a deep quantum mechanical property that differs from classical
mechanics is the Uncertainty Principle. We can use matrix theory to
explain the uncertainly principle as follows. Recall that the expectation value
of an observable associated with a Hermitian matrix A with respect to a
quantum state |ψ⟩ is ⟨A⟩ = ⟨ψ|A|ψ⟩. Let A and B be Hermitian operators
and |ψ⟩ be some quantum state on which A and B operate. As mentioned
before, if A corresponds to an observable, then the standard deviation of the
observable is
q
2
p
∆(A) = ⟨⟨ψ|(A − αI)2 |ψ⟩ = ⟨ψ|A2 |ψ⟩ − ⟨ψ|A|ψ⟩ ,

where α = ⟨ψ|A|ψ⟩. We have the following.

THEOREM 2.6.1. Let A, B ∈ Mn be Hermitian, |ψ⟩ ∈ Cn be a unit vector.
Then
∆(A)∆(B) ≥ |⟨ψ|(AB − BA)|ψ⟩|/2. (2.29)
Suppose α = ⟨ψ|A|ψ⟩ and β = ⟨ψ|B|ψ⟩. The equality holds in (2.29) if and
only if there is θ ∈ R such that
cos θ(A − αI)|ψ⟩ + i sin θ(B − βI)|ψ⟩ = |0⟩.

qProof. Letq  = A − αI and B̂ = B − βI. Note first that ∆(A)∆(B) =

⟨ψ|Â2 |ψ⟩ ⟨ψ|B̂ 2 |ψ⟩ and ⟨ψ|[A, B]|ψ⟩ = ⟨ψ|[Â, B̂]|ψ⟩. So, we only need to
show that 4⟨ψ|Â2 |ψ⟩⟨ψ|B̂ 2 |ψ⟩ ≥ |⟨ψ|[Â, B̂]|ψ⟩|2 . Note that the matrices

⟨ψ|Â2 |ψ⟩ ⟨ψ|ÂB̂|ψ⟩ ⟨ψ|Â2 |ψ⟩ −⟨ψ|B̂ Â|ψ⟩

   
C1 = and C2 =
⟨ψ|B̂ Â|ψ⟩ ⟨ψ|B̂ 2 |ψ⟩ −⟨ψ|ÂB̂|ψ⟩ ⟨ψ|B̂ 2 |ψ⟩
are positive semi-definite as proved by checking that all their principal minors
are nonnegative using the Cauchy-Schwartz inequality. Thus, C = C1 + C2 is
positive semi-definite and
4⟨ψ|Â2 |ψ⟩⟨ψ|B̂ 2 |ψ⟩ − |⟨ψ|[Â, B̂]|ψ⟩|2 = det(C) ≥ 0.
50 QUANTUM COMPUTING

The equality det(C) = 0 holds if and only if C is singular, equivalently, the

positive semi-definite matrices C1 and C2 are singular and share a common
null vector. Since C1 and C2 have the same trace, we see that
(1) C1 = C2 = (tr C1 )|u⟩⟨u| for some unit vector |u⟩ ∈ Cn , and
(2) ⟨ψ|ÂB̂|ψ⟩ = −⟨ψ|B̂ Â|ψ⟩, i.e., ⟨ψ|{Â, B̂}|ψ⟩ = 0.
Condition (1) implies det(C1 ) = 0, namely ⟨ψ|Â2 |ψ⟩⟨ψ|B̂ 2 |ψ⟩ = |⟨ψ|ÂB̂|ψ⟩|2 .
By the Cauchy-Schwartz inequality, Â|ψ⟩ and B̂|ψ⟩ are linearly dependent.
Condition (2) implies that ⟨ψ|ÂB̂|ψ⟩ ∈ iR. So, Â|ψ⟩ and iB̂|ψ⟩ are linearly
dependent over R. Thus, there is θ ∈ [0, 2π) such that cos θÂ|ψ⟩ + i sin θB̂|ψ⟩
is the zero vector. Conversely, if cos θÂ|ψ⟩ + i sin θB̂|ψ⟩ is the zero vector, one
readily checks that C1 = C2 and det(C1 + C2 ) = 0.

For example, if A = σx , B = σy , and |ψ⟩ = (1, 0)t , we have the equality.

ℏ d
Now, suppose A = Q and B = P ≡ ; see [26, Example 2.20]. Deduce
i dQ
from the above arguments that

ℏ
∆(Q)∆(P ) ≥ .
2
Note that AB − BA = γI with γ ̸= 0 can only happen for infinite dimensional
operators as Tr (AB − BA) = 0 for finite dimensional operators. The Uncer-
tainty principle in terms of standard deviation has been formulated first in [7]
and [8].

2.6.2 Schrödinger Picture and Heisenberg Picture

In the previous section, we have introduced the Schrödinger equation

d|ψ(t)⟩
iℏ = H|ψ(t)⟩
dt
where |ψ(t)⟩ = U (t)|ψ(0)⟩. The expectation value of an observable a at t is
given by ⟨ψ(t)|A|ψ(t)⟩. This description with time-dependent state |ψ(t)⟩ and
time-independent operator A is called the Schrödinger picture.
There is another equivalent description in which the state is time-
independent while the operator depends on time, called the Heisenberg
picture. Let
A(t) = U (t)† AU (t) (2.30)
and consider ⟨ψ(0)|A(t)|ψ(0)⟩. We find

⟨ψ(0)|A(t)|ψ(0)⟩ = ⟨ψ(0)|U (t)† AU (t)|ψ(0)⟩ = ⟨ψ(t)|A|ψ(t)⟩,

Quantum mechanics: Hilbert space formalism 51

which shows the two pictures give the same expectation value at any t. It is
also shown that any matrix element satisfies

⟨ϕ(t)|A|ψ(t)⟩ = ⟨ϕ(0)|A(t)|ψ(0)⟩

and hence two pictures are equivalent.

Now the dynamics of the system in the Heisenberg picture is carried by
operators. We find
d 1 i
A(t) = (iHA(t) − A(t)iH) = [H, A(t)]. (2.31)
dt ℏ ℏ
If, moreover, A has explicit time-dependence, we obtain
 
d ∂A i
A(t) = (t) + [H, A(t)], (2.32)
dt ∂t ℏ

where (∂A/∂t)(t) = eiHt/ℏ (∂A/∂t)e−iHt/ℏ .

In the Schrödinger picture, the state evolves with time. In contrast, observ-
ables, such as position and momentum, evolve with time in the Heisenberg
picture. In this sense, the Heisenberg picture depicts situation similar to the
classical dynamics. Let H be a Hamiltonian of a classical system and a(t) be
any physical quantity such as the coordinates or the momentum. Then the
time evolution of a(t) is described by the Hamilton’s equation of motion
da(t) ∂a(t)
= + [a(t), H]PB , (2.33)
dt ∂t
where [x, y]PB is the Poisson bracket defined by
X  ∂x ∂y ∂y ∂x

[x, y]PB = − , (2.34)
∂qk ∂pk ∂qk ∂pk
k

where {qk , pk } is the set of coordinates and corresponding momenta. Notice

the similarity between Eqs. (2.33) and (2.32).

2.6.3 Some Examples

We now give some examples to clarify the axioms introduced §2.1. They turn
out to have relevance to certain physical realizations of a quantum computer.
EXAMPLE 2.6.2. Let us consider a time-independent Hamiltonian
ℏ
H = − ωσx . (2.35)
2
Suppose the system is in the eigenstate of σz with the eigenvalue +1 at time
t = 0;  
1
|ψ(0)⟩ = .
0
52 QUANTUM COMPUTING

FIGURE 2.2
Probability P↑ (t) with which a spin is observed in the ↑-state and P↓ (t) ob-
served in the ↓-state.

The wave function |ψ(t)⟩ (t > 0) is then found from Eq. (2.5) to be
 ω 
|ψ(t)⟩ = exp i σx t |ψ(0)⟩. (2.36)
2

The matrix exponential function in this equation is evaluated with the help of
Eq. (1.23) and we find
!  !
cos ωt/2 i sin ωt/2 1 cos ωt/2
|ψ(t)⟩ = = . (2.37)
i sin ωt/2 cos ωt/2 0 i sin ωt/2

Suppose we measure the observable σz . Note that |ψ(t)⟩ is expanded in terms

of the eigenvectors of σz as

ω ω
|ψ(t)⟩ = cos t|σz = +1⟩ + i sin t|σz = −1⟩.
2 2

Therefore we find the spin is in the spin-up state with the probability P↑ (t) =
cos2 (ωt/2) and in the spin-down state with the probability P↓ (t) = sin2 (ωt/2)
as depicted in Fig. 2.2. Of course, the total probability is independent of time
since cos2 (ωt/2) + sin2 (ωt/2) = 1. This result is consistent with classical
spin dynamics. The Hamiltonian (2.35) depicts a spin under a magnetic field
along the x-axis. Our initial condition signifies that the spin points the z-
direction at t = 0. Then the spin starts precession around the x-axis, and the
z-component of the spin oscillates sinusoidally as is shown above.
Next let us take the initial state
 
1 1
|ψ(0)⟩ = √ ,
2 1
Quantum mechanics: Hilbert space formalism 53

which is an eigenvector of σx (and hence of the Hamiltonian) with the eigen-

value +1. We find |ψ(t)⟩ in this case as
eiωt/2 1
     
cos ωt/2 i sin ωt/2 1 1
|ψ(t)⟩ = √ = √ . (2.38)
i sin ωt/2 cos ωt/2 2 1 2 1
Therefore the state remains in its initial state at an arbitrary t > 0. This is an
expected result since the system at t = 0 is an eigenstate of the Hamiltonian.
Now let us formulate Example 2.6.2 (and also Exercise 2.1) in the most
general form. Consider a Hamiltonian
ℏ
H = − ω n̂ · σ, (2.39)
2
where n̂ is a unit vector in R3 . The time-evolution operator is readily ob-
tained, by making use of the result of Proposition 1.5.8, as
ω ω
U (t) = exp(−iHt/ℏ) = cos t I + i(n̂ · σ) sin t. (2.40)
2 2
Suppose the initial state is
 
1
|ψ(0)⟩ = ,
0
for example. Then we find
 
cos(ωt/2) + inz sin(ωt/2)
|ψ(t)⟩ = U (t)|ψ(0)⟩ = . (2.41)
i(nx + iny ) sin(ωt/2)
The reader should verify that |ψ(t)⟩ is normalized at any instant of time t > 0.
EXAMPLE 2.6.3. (Rabi oscillation) This example is often employed for a
quantum gate implementation. We will take the natural unit ℏ = 1 to simplify
our notation throughout this example. Let us consider a spin-1/2 particle in
a magnetic field along the z-axis, whose Hamiltonian is given by
ω0
H0 = − σ z . (2.42)
2
Suppose the particle is irradiated by an oscillating magnetic field of angular
frequency ω, which introduces transitions between two energy eigenstates of
H0 . Then the perturbed Hamiltonian is modeled as
0 eiωt −ω0 ω1 eiωt
   
ω0 ω1 1
H = − σz + = , (2.43)
2 2 e−iωt 0 2 ω1 e−iωt ω0
where ω1 > 0 is a parameter proportional to the amplitude of the oscillating
field. Let us evaluate the wave function |ψ(t)⟩ at time t > 0 assuming that the
system is in the ground state of the unperturbed Hamiltonian
 
1
|ψ(0)⟩ = (2.44)
0
54 QUANTUM COMPUTING

at t = 0. Note that we cannot simply exponentiate the Hamiltonian since it

is time-dependent. Surprisingly, however, the following trick makes it time-
independent. Let us consider the following unitary transformation:

|ϕ(t)⟩ = e−iωσz t/2 |ψ(t)⟩. (2.45)

A straightforward calculation shows that |ϕ(t)⟩ satisfies

d
i |ϕ(t)⟩ = H̃|ϕ(t)⟩, (2.46)
dt
where
 
−iωσz t/2 iωσz t/2 −iωσz t/2 d iωσz t/2 1 −ω0 + ω ω1
H̃ = e He − ie e =
dt 2 ω1 ω0 − ω
δ ω1
= − σz + σx (2.47)
2 2
is, in fact, time-independent. Here δ = ω0 − ω stands for the “detuning”
between ω and ω0 . Note that the Hamiltonian H̃ can be put into the form
(2.39) as  
∆ ω1 δ
q
H̃ = σx − σz , ∆ ≡ δ 2 + ω12 . (2.48)
2 ∆ ∆
Now it is easy to solve Eq. (2.46). The time evolution operator is obtained
using Eq. (2.40) as
 
∆t ω1 δ ∆t
Ũ (t) = cos I −i σx − σz sin
2 ∆ ∆ 2
∆t δ ∆t ω1 ∆t
 
cos + i sin −i sin
 2 ∆ 2 ∆ 2 
= . (2.49)
 ω1 ∆t ∆t δ ∆t 
−i sin cos − i sin
∆ 2 2 ∆ 2
The wave function |ϕ(t)⟩ with the initial condition |ϕ(0)⟩ = (1, 0)t is

∆t δ ∆t
 
cos + i sin
 2 ∆ 2 
|ϕ(t)⟩ = Ũ (t)|ϕ(0)⟩ =  . (2.50)
 ω1 ∆t 
−i sin
∆ 2
We find |ψ(t)⟩ from Eq. (2.45) as
 

iωt/2 ∆t δ ∆t
e cos + i sin
2 ∆ 2
|ψ(t)⟩ = eiωσz t/2 |ϕ(t)⟩ = 
 
. (2.51)
ω1 ∆t
 
−ie−iωt/2 sin
∆ 2
Quantum mechanics: Hilbert space formalism 55

Suppose the applied field is in resonance with the energy difference of two
levels, namely ω = ω0 . We obtain δ = 0 and ∆ = ω1 in this case. The wave
function |ψ(t)⟩ at later time t > 0 is
ω1 t
 
eiω0 t/2 cos
2
|ψ(t)⟩ = eiωσz t/2 |ϕ(t)⟩ = 
 
. (2.52)

−iω0 t/2 ω1 t 
−ie sin
2
The probability with which the system is found in the ground (excited) state
of H0 is given by

P0 = cos2 ω1 t/2 (P1 = sin2 ω1 t/2). (2.53)

This oscillatory behavior is called the Rabi oscillation. The frequency ω1 is

called the Rabi frequency, while ∆ in Eq. (2.48) is called the generalized
Rabi frequency

2.7 Notes and open problems

The EPR states and GHZ states are useful in the verification of the
Bell’s inequality and its generalization; see [26, Theorems 4.5 and 4.7].
These results show that the entanglement of quantum states cannot be ex-
plained by the hidden variable theory suggested in the EPR paper, and the
Copenhagen interpretation for entanglement is valid. Experiments to ver-
ify the Bell’s inequality and its generalizations have been performed. See
https://arxiv.org/pdf/quant-ph/0504166.pdf
Here are some open problems on quantum states.

1. Let {|x1 ⟩, . . . , |xk ⟩}, {|y1 ⟩, . . . , |yk ⟩} ⊆ Cn . It is known that there is a

unitary U ∈ Mn such that U |xj ⟩ = |yj ⟩ for all j if and only if (⟨xr |xs ⟩) =
(⟨yr |ys ⟩).
If n = n1 n2 , determine when there will be unitary U = U1 ⊗ U2 such
that U |xj ⟩ = |yj ⟩ for all j.

2. Suppose many identical copied of pure state ρ = |ψ⟩⟨ψ| ∈ Mn is given.

Using the orthogonal projection Pj = |ej ⟩⟨ej | for j = 1, . . . , n, corre-
sponding to the standard basis {|e1 ⟩, . . . , |en ⟩} for Cn , the measurement
can provide estimate for the diagonal entries of ρ. Suppose n = 2m so
that |ψ⟩ is an m qubit state. A unitary matrix of the form U1 ⊗ · · · ⊗ Um
with U1 , . . . , Um ∈ U(2) is called local unitary. In state tomography
problem, it is known that there are M = 3m local unitary matrices
56 QUANTUM COMPUTING

V1 , . . . , VM ∈ Mn such that the diagonal entries of Vj ρVj† can be used

to determine/reconstruct ρ.
Can we determine ρ using fewer than 3m local unitary matrices?
If we only want to distinguish pure states, find the smallest number of
local unitary matrices W1 , . . . , WK such that no two pure states ρ1 , ρ2
will have zero diagonal entries for all Wj (ρ1 − ρ2 )Wj† .

Exercises for Chapter 2

EXERCISE 2.1. Let us consider a Hamiltonian
ℏ
H = − ωσy . (2.54)
2
Suppose the initial state of the system is
 
0
|ψ(0)⟩ = . (2.55)
1

(1) Find the wave function |ψ(t)⟩ at later time t > 0.

(2) Find the probability for the system to have the outcome +1 upon measure-
ment of σz at t > 0.
(3) Find the probability for the system to have the outcome +1 upon measure-
ment of σx at t > 0.
EXERCISE 2.2. Prove that if B1 and B2 are orthonormal bases for Cm and
Cn , then B = {|u⟩ ⊗ |v⟩ : |u⟩ ∈ B1 , |v⟩ ∈ B2 } is an orthonormal basis for
Cmn = Cm ⊗ Cn .
EXERCISE 2.3. Suppose U is a cloning unitary transformation, such that

|Ψ⟩ ≡ U |ψ⟩|0⟩ = |ψ⟩|ψ⟩

|Φ⟩ ≡ U |ϕ⟩|0⟩ = |ϕ⟩|ϕ⟩

for arbitrary |ψ⟩ and |ϕ⟩.

(1) Write down ⟨Ψ|Φ⟩ in all possible ways.
(2) Show, by inspecting the result of (1), that such U does not exist.
EXERCISE 2.4. Let |ψ(θ, ϕ)⟩ be the state given by Eq. (2.15). Show that

⟨ψ(θ, ϕ)|σ|ψ(θ, ϕ)⟩ = (sin θ cos ϕ, sin θ sin ϕ, cos θ), (2.56)

where the left hand side is defined as

(⟨ψ(θ, ϕ)|σx |ψ(θ, ϕ)⟩, ⟨ψ(θ, ϕ)|σy |ψ(θ, ϕ)⟩, ⟨ψ(θ, ϕ)|σz |ψ(θ, ϕ)⟩).
Quantum mechanics: Hilbert space formalism 57

EXERCISE 2.5. Find the density matrix of a pure state (2.15) and write
it in the form of Eq. (2.16).

EXERCISE 2.6. Let ρ be given by Eq. (2.16). Show that

⟨σ⟩ = Tr (ρσ) = u. (2.57)

EXERCISE 2.7. The Bell basis is obtained from the binary basis {|00⟩, |01⟩,
|10⟩, |11⟩} by a unitary transformation. Write down the unitary transforma-
tion explicitly.

EXERCISE 2.8. Find the expectation value of σx ⊗ σz measured in each of

the Bell states.

EXERCISE 2.9. In many quantum algorithms, the result of an action of a

function f on x is encoded into the form
X X
Uf : N |x⟩|0⟩ 7→ N |x⟩|f (x)⟩,
x x

where |x⟩ stands for the tensor product state |bn−1 bn−2 . . . b0 ⟩ with x =
bn−1 2n−1 + bn−2 2n−2 + . . . + b0 and N is the normalization constant. The
first register is for the input x, while the second one is for the corresponding
output f (x). Note that Uf acts on all possible states simultaneously.
Let f (x) = ax mod N , where a and N are coprime, and consider the state
" 511
# 511
1 X 1 X
Uf √ |x⟩|0⟩ = √ |x⟩|ax mod N ⟩
512 x=0 512 x=0

with a = 6 and N = 91. Suppose the measurement of the first register results
in (1) x = 11, (2) x = 23 and (3) x = 35. What is the state immediately after
each measurement?

References
[1] P. A. M. Dirac, Principles of Quantum Mechanics (4th ed.), Clarendon
Press (1981).
[2] L. I. Shiff, Quantum Mechanics (3rd ed.), McGraw-Hill (1968).
[3] A. Messiah, Quantum Mechanics, Dover (2000).
[4] J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics (3rd Edi-
tion), Cambridge University Press, Cambridge (2020).
58 QUANTUM COMPUTING

[5] L. E. Ballentine, Quantum Mechanics, World Scientific, Singapore

(1998).
[6] A. Peres, Quantum Theory: Concepts and Methods, Springer (2006).
[7] E. H. Kennard, Z. Phys. 44, 326 (1927).
[8] H. P. Robertson, Phys. Rev. 34 163, (1929).
[9] R. Horodecki et al., eprint, quant-ph/0702225 (2007).
[10] A. Acı́n et al., Phys. Rev. Lett. 85, 1560 (2000).
[11] A. Acı́n et al., Phys. Rev. Lett. 87, 040401 (2001).
[12] A. Peres, Phys. Rev. Lett. 77, 1413 (1996).
[13] M. Horodecki et al., Phys. Lett. A 223, 1 (1996).
[14] G. Vidal, J. Mod. Opt. 47, 355 (2000).
[15] P. Horodecki, Phys. Lett. A 232, 333 (1997).
[16] R. Jozsa, J. Mod. Opt. 41, 2315 (1994).
[17] E. Riefell and W. Polak, ACM Computing Surveys (CSUR) 32, 300
(2000) and E. G. Rieffel and W. H. Polak, Quantum Computing: A
Gentle Introduction, The MIT Press (2011).
[18] M. A. Neilsen and I. L. Chuang, Quantum Computation and Quantum
Information, Cambridge University Press (2000).
[19] W. K. Wootters and W. H. Zurek, Nature 299, 802 (1982).
[20] D. Dieks, Phys. Lett. A 92, 271 (1982).
[21] D. M. Greenberger, M. A. Horne and A. Zeilinger, in ’Bell’s Theorem,
Quantum Theory, and Conceptions of the Universe, ed. M. Kafatos,
Kluwer, Dordrecht (1989). Also avilable as arXiv:0712.0921 [quant-ph].
[22] W. Dür, G. Vidal and J. I. Cirac, Phys. Rev. A 62, 062314 (2000).
[23] A. Einstein, B. Podolsky, N. Rosen, Phys. Rev. 41, 777 (1935).
[24] C. H. Bennett et al., Phys. Rev. A 59, 1070 (1999) and D. P. DiVincenzo,
D. W. Leung and B. M. Terhal, IEEE Trans. Info. Theory 48, 580
(2002). See also A. SaiToh, R. Rahimi and M. Nakahara, Phys. Rev. A
77, 052101 (2008).
[25] C. H. Bennett and G. Brassard, in Proc. IEEE Int. Conf. Comp., Systems
and Signal Processing 175 (1984).
[26] W. Scherer, Mathematics of Quantum Computing, An Introduction,
Springer (2019).
[27] https://en.wikipedia.org/wiki/Speed of light
[28] https://en.wikipedia.org/wiki/Random number generation