Quantum Computation and Quantum Information

Author(s): Yazhen Wang

Source: Statistical Science , August 2012, Vol. 27, No. 3 (August 2012), pp. 373-394
Published by: Institute of Mathematical Statistics

Stable URL: https://www.jstor.org/stable/41714771

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/41714771?seq=1&cid=pdf-
reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Institute of Mathematical Statistics is collaborating with JSTOR to digitize, preserve and

extend access to Statistical Science

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
Statistical Science
2012, Vol. 27, No. 3, 373-394
DOI: 10.1214/1 1-STS378
© Institute of Mathematical Statistics, 2012

Quantum Computation and Quantum

Information
Yazhen Wang

Abstract. Quantum computation and quantum information are of great cur-

rent interest in computer science, mathematics, physical sciences and engi-
neering. They will likely lead to a new wave of technological innovations in
communication, computation and cryptography. As the theory of quantum
physics is fundamentally stochastic, randomness and uncertainty are deeply
rooted in quantum computation, quantum simulation and quantum informa-
tion. Consequently quantum algorithms are random in nature, and quantum
simulation utilizes Monte Carlo techniques extensively. Thus statistics can
play an important role in quantum computation and quantum simulation,
which in turn offer great potential to revolutionize computational statistics.
While only pseudo-random numbers can be generated by classical comput-
ers, quantum computers are able to produce genuine random numbers; quan-
tum computers can exponentially or quadratically speed up median evalua-
tion, Monte Carlo integration and Markov chain simulation. This paper gives
a brief review on quantum computation, quantum simulation and quantum
information. We introduce the basic concepts of quantum computation and
quantum simulation and present quantum algorithms that are known to be
much faster than the available classic algorithms. We provide a statistical
framework for the analysis of quantum algorithms and quantum simulation.

Key words and phrases: Quantum algorithm, quantum bit (qubit), quan-
tum Fourier transform, quantum information, quantum mechanics, quantum
Monte Carlo, quantum probability, quantum simulation, quantum statistics.

1. INTRODUCTION approaches run up against the size limit. One possi-

ble way to get around the difficulties is to move to a
For decades computer hardware has grown in power new computing paradigm provided by quantum infor-
approximately according to Moore's law, which states mation science. Quantum information science is based
that the computer power doubles for constant on cost
the idea of using quantum devices to perform com-
roughly once every two years. However, because ofputation
the and manipulate and transmit information, in-
fundamental difficulties of size in conventional com- stead of electronic devices following the laws of clas-
puter technology, this dream run is ending. The con-sical physics, see Deutsch (1985), DiVincenzo (1995),
ventional approaches to the fabrication of computer Feynman (1981/82). Quantum mechanics and infor-
technology are to make electronic devices smaller and mation theory are two of the great scientific develop-
smaller in order to increase the computer power. As ments and technological revolutions in the 20th cen-
tury, and quantum information science is to marry the
the sizes of the electronic devices get close to the
two previously disparate fields and form a single uni-
atomic scale, quantum effects are starting to inter-
fying viewpoint. Quantum information science studies
fere in their functioning, and thus the conventional
the preparation and control of the quantum states of
physical systems for the purposes of information trans-
Yazhen Wang is Professor, Department of Statistics, mission and manipulation. It includes quantum compu-
University of Wisconsin-Madison, Madison, Wisconsin tation, quantum communication and quantum cryptog-
53706, USA (e-mail: [email protected]). raphy. This revolutionary field will enable a range of

373

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
374 Y. WANG

exotic new indevices to

biochemist be
eral biomolecules
agreement that quan
likely lead is
to to
theuse com
creati
solve its
problems time
that evol
coul
a classical for quantum
computer. Alr
mentary and
quantum creating
compu
to run puters
quantum built
algorith
lations. ulating
Intensive natur
research
the world large
to quantum
investigate
could lead in more
to an efficien
powe
tum tum
computers system
in the n
quantum scribed
information with an
a 21st while
century a classi
technolo
sic to
information store
and its q
classic
Since the tum
theory computer
of quan
tally systems
stochastic, that
randomn
rooted in computers.
quantum comp
tion. As a In this article we review the
result, concepts of quan-
quantum
ture in thetum computation
sense and introduce quantum
thatalgorithms th
with some and quantum simulation.
probabilities, The quantum algorithms are
known to be much faster than the available classi-
widely employed in qua
tics has an calimportant
algorithms. Statistical analyses of quantum algo- ro
tation, rithms and quantum simulation
quantum are provided. We give
simulat
On the a brief description
other hand, on quantum information.quan The ar-
simulation ticle
havesections start with presentations
tremend in broad brush-
computationalstrokes, followed statistics.
by specific discussions along with
A quantum some mathematical derivations if necessary.
system is The geninten-
and the tion
state is to give each
is topic first
matheman overview and then
vector in a general
some description and a precise
complex characterization.
complex It is recommended to focus on the requir
numbers qualitative discus-
tum state sions but skip the derivations for the
usually readers who would
grows
the system, like to get a quick
rather picture of quantum computation
than and
cal systems. quantum simulation.
As a conse
tial number The rest of of
the paper proceeds
bits as follows. Sectionof
2
puter to briefly introduces quantum
store the mechanics, quant
quantum prob-
computers ability
in and quantum
a statistics. Section
difficult 3 reviews basic
tum system. concepts of quantum
On computation
the and entanglement.
oth
tems are Section 4 illustrates
able to some store
widely known quantum al- an
number of gorithms
complexand provides a statistical framework
num for the
ulations andstudy of quantum algorithms. Section 5 presents quan-
calculations
tum simulation and discusses its statistical analysis.
tum information science
Section 6 gives a short description on quantum infor-
ing how to take advanta
mation theory. Section 7 features concluding remarks
tion hidden in the quan
and lists some open research problems.
the immense potential c
and molecules for the p
2. BRIEF BACKGROUND REVIEW ON QUANTUM
tation and processing in
THEORY
shown that quantum algo
gorithm and Shor's
Quantum mechanics has been applied to fact
everything
advantage under and inside the Sun,
over from chemical reaction and
known c
Contemporary superconductor to the structure
scientifi of DNA and nuclear
derstanding complex
fusion in stars. Although the significant difference be-qu

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 375

tween classical physics and quantum physics liesthat

dicate in the
the objects are column vectors or row vec-
quantum prediction of physical entity whentors the scale
in the vector space, respectively. Denote by super-
of observations becomes comparable to thescripts
atomic *, / or
and t the conjugate of a complex number,
sub-atomic scale, many macroscopic properties of sys-
the transpose of a vector or matrix, and conjugate trans-
pose operation,
tems can only be fully explained and understood by respectively. We define an inner prod-
quantum physics. The quantum world is uct on the vector space to be a function that takes as
extremely
strange, and quantum theory is completelyinput two vectors from the vector space and produces
counterin-
tuitive. Light waves behave like particles and a complex
particles number as output. For |m) and |u) in the
behave like waves (wave particle duality); matter can we denote their inner product by {u'v).
vector space,
go from one spot to another without moving The inner
throughproduct must satisfy (i) conjugate symmetry,
the intermediate space (quantum tunneling); (w|v) = ((i'|w})*; (ii) linearity in the second argument,
informa-
tion can be moved across a vast distance without trans- {u'v -t- u>) = {u'v) + (w|u>); (iii) positive-definiteness,
mitting it through the intervening space (quantum tele- (u'u) >0 with equality only for u = 0. For example,
portation). Quantum theory provides a mathematical C* has a natural inner product
description of wave particle duality and interaction of k

matter and energy. It describes the time evolutions of (u'v) = = (u',...,u*k)(v',...,vk)',

physical systems via wave functions. The wave func- j= 1
tions encapsulate the probabilities that particles are to where (w| = (u', . . . , Uk ) and |v) = (vi, . . . , Vk)'. An
be found in a given state at a given time. For exam- inner product induces a norm ||u|| = V( u'u ), and a
ple, the probability of finding a photon in some region distance ''u - v || between | u) and 'v). For the finite-
is the square of the modulus of a wave function, and, dimensional case, a Hilbert space H is simply a vector
since at some point the sum of two wave functions can space with an inner product.
be zero but neither wave function is zero, probabili- An operator A on H, denoted by A(|«)) for | u) e
ties appear to cancel out each other in a way totally H, is a function mapping from H to H that satisfies
unexpected from classical probability. The intrinsic A.(a'u) + b'v )) = aA(|w)) + ¿A(|v)) for any | u), |u) e
stochastic nature of quantum theory indicates a deep H and a,b e C. We can represent an operator through
connection between quantum mechanics and probabil- a matrix. Suppose that A is an operator on H and
ity. Since the main focus of this paper is on quan- ei , . . . , ek form an orthonormal basis in H. Then each
tum computation and quantum information, we give A(|ey)) G Tí and there exists a unique k x k matrix
a brief description of quantum theory in this section (aje) such that A(|e;)) = Y!í=' I *i)atj, j = 1, . . . , k.
to provide some quantum background for the purpose We will identify operator A with matrix (aje) and use
of reviewing quantum computation and quantum sim- A for both operator and matrix (aje). An operator A
ulation in subsequent sections. For further reading on
on H is said to be self-adjoint if its corresponding ma-
the subjects we recommend textbooks by Sakurai and trix A is Hermitian, that is, A = A1". We also refer to
Napolitano (2010) at the graduate level and Griffithsself-adjoint operators as Hermitian operators. An oper-
(2004) at the undergraduate level for quantum mechan-ator U is said to be unitary if its corresponding matrix
ics, Holevo (1982), Parthasarathy (1992) and Wang U is unitary, that is, UU^ = U^U = I. We say an oper-
(1994) for quantum probability and quantum stochas- ator A is semi-positive (or positive) definite if its cor-
tic processes, and Artiles, Gill and Gu{ä (2005) and responding matrix A is semi-positive (or positive) def-
Barndorff-Nielsen, Gill and Jupp (2003) for quantum inite, that is, (u'p'u) > 0 for |w) € H (or (u'p'u) > 0
statistics. for 'u) e H with equality only for | u) = 0). The trace
of an operator A, denoted by Tr(A), is defined to be
2.1 Hilbert Space and Operator
the trace of its corresponding matrix A = (ají), that is,
For the sake of simplicity we choose to work with
Tt(A) = E5=i ajj.
comparatively easy finite-dimensional situations. De-
note by C the set of all complex numbers. We start
2.2 Quantum System
with vector space in linear algebra. A simple exam- Quantum mechanics depicts phenomena at micro-
ple of vector space is C* consisting of all ¿-tuples of scopic level such as position and momentum of an indi-
complex numbers (z' , . . . , Zk)- The elements of a vec- vidual particle like an atom or electron, spin of an elec-
tor space are called vectors. As in quantum mechanics tron, detection of light photons, and the emission and
and quantum computation, we use Dirac notations | ) absorption of light by atoms. Unlike classical mechan-
(which is called ket) and ( | (which is called bra) to in- ics where physical entities like position and momentum

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 376 Y. WANG

can be measured
is motivated from the fact that the precisely
eigenvalues of self-
chanics is adjoint operators are real. Assume that an observable
intrinsically sto
only make X has a discrete spectrum with the following diagonal
probabilistic p
the form
measurements perfor
p
Quantum mechanics is m
a Hilbert (3) X=J^xaQ H
space a, and se
a= 1
A quantum system is com
state and the time evolution of the state. A state is where xa are real eigenvalues of X and Qa are
defined to be a unit vector in H. Let IVKO) be the corresponding one-dimensional projections onto
state of the quantum system at time t, which is also orthogonal eigenvectors of X. Consider such a
referred to as a wave function. The states 'ý(t')) andservable in the quantum system prepared in sta
Measure space (Í2, JF) is used to describe pos
I tyto)) at t' and t2 are connected through 'ir{h)) =
measurement outcomes of the observable, and th
U(/i, t2)'Ý(h)), where U(fi, ¿2) is a unitary operator
sult of the measurement is a random variable on (£2
depending only on time t' and t2- In fact, there exists a
with probability distribution Pp. We denote by X
self-adjoint operator H, which is known as the Hamil-
result of the measurement of observable X give
tonian of the quantum system, such that U(fi , ti) =
(3). Then X is a random variable taking value
exp[- í'H(í2 - íi)]. With Hamiltonian H, we may de-
{x',x2, . . . and under pure state | ý ), the prob
scribe the continuous time evolution of | ý(t)) by the
ity that measurement outcome xa occurs is defin
Schrödinger equation be

/"i' . 9IVK0) ni ; . / - 7
/"i' (1) i . - - - = H|VK0), ni ; i=v- . / 7 1.P(a) = Pp{X = xa)
Ot
= {t'QaW)=Tma't)m, a = 1,2,....
Alternatively a quantum system can be described by
With the probability we derive the expectation unde
a density operator (or density matrix). A density oper-
pure state 'ý),
ator p is an operator on 7i which (1) is self-adjoint;
(2) is semi-positive definite; (3) has unit trace [i.e., E^(X) = ^xaP(a) = ^xa{f IQalf )
a a
Tr(/o) = 1]. Following the convention in quantum in-
formation science, we reserve notation p for state, den- = (Ý'X'f)=Tr(X'Ý)(f').
sity operator or density matrix. A state is often classi- Note the difference between an observable X which is a
fied as a pure state or an ensemble of pure states. A pure Hermitian matrix and its measurement result X which
state is a unit vector ''fr) in H, which corresponds to a is a real- valued random variable.
density operator p = IVf)(1AK and an ensemble of pure Measuring observable X will alter the state of the
states corresponds to the case that the quantum system quantum system (Kiefer, 2004; von Neumann, 1955).
is in one of states 'ýj), j = 1, J, with probability If the quantum system is prepared with initial state 'ý),
Pj being in state |i (rj), and the corresponding density the state of the system after the measurement result xa
operator is defined to be

J
(4) %!ŽL
(2) p = J^pj'Ýj)(Ýj'-
j= 1 For an ensemble state with density operator
by (2), if the quantum state is | ýj), the probabil
See Griffiths (2004), Sakurai and Napolitano (2010)
result xa occurs is
and Shankar (1994).
P(a'j) = (Ýj'Qa'Ýj)=Tr(Qa'Ýj)(Ýj').
2.3 Quantum Probability
Applying the law of total probability, we obtain that
We can test the theory of quantum mechanics by under state p, the probability that xa occurs is equal to
checking its predictions with experiments of perform- j

ing measurements on quantum systems in the labora- P(a) = P„(X = xa) = J2 PjP(a'j)
tory. The usual quantum measurements are on observ- j= 1
ables such as position, momentum, spin, and so on, J

where an observable X is defined as a self-adjoint op- = Pj Tr(Qa'ýj)(ýj I) = Tr(Qfl/o).

erator on Hilbert space H. The observable definition j= 1

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 377

The expectation of X under state p, about probability distribution Pp of the measurements

p p and thus indirectly about density matrix p. In the liter-
ature of quantum physics, quantum tomography is re-
Ep[X] = £ X
a= 1 a=l ferred to as the reconstruction of the underlying den-
sity matrix p by probing identically prepared quantum
= tr(X/o),
systems from some different angles. Specifically, sup-
and variance pose that we perform measurements of observables on
identically prepared quantum systems in an unknown
Varp[X] = tr[X2/>] - (tr[X/>])2.
state p and obtain measurement results X i , . . . , Xn .
We may derive the density operator of the quan- Assume that p is known up to some unknown parame-
tum system after obtaining the measurement result xa ter 6; then X', ...,Xn are i.i.d. observations with dis-
by conditional probability arguments as follows. If the tributions Pp which depend on 6. This gives a quan-
quantum system is in pure state | ýj) before the mea- tum parametric statistical model. We may then define
surement, the quantum state after measurement result quantum likelihood and Fisher quantum information
xa has occurred is and establish quantum point estimation and quantum
hypothesis testing theory. Alternatively we may model
QM'}
p nonparametrically by assuming that p is an infinite
J ypR/) matrix and then use nonparametric methods to estimate
the
If the quantum state is p before the measurement, density matrix. For details see Artiles, Gill and
after
observing measurement outcome xa we have theGufafol-
(2005), Barndorff-Nielsen, Gill and Jupp (2003),
Butucea,
lowing ensemble of states: the quantum system is in Gujã and Artiles (2007) and Nussbaum and
pure state IV^) with probability P(j'a), where Bayes's
Szkola (2009).
theorem shows
3. QUANTUM COMPUTING CONCEPTS
P(j'a) = pjP(a'j)/P(a).
Unlike classical computers using transistors to
Thus after measurement xa the density operator for the crunch the ones and zeroes individually, quantum com-
ensemble state is given by puters can handle both one and zero simultaneously via
j what are known as superposition quantum states. A su-
perposition state is a state of matter which we may
Pa = J2 pU'a)'ýj)(Ýj'
7=1
think of as both one and zero at the same time. Quan-
tum computers use the strange superposition states and
quantum entanglements to do the trick of performing

f¿ Ul P(a'j) simultaneous calculations and extracting the calculated

results. The spooky phenomena of quantum entangle-
ment and superposition are the key that enables quan-
tum computers to be superfast and vastly outperform
classical computers.
See Holevo (1982), Parthasarathy
3.1 Quantum Bit (1992) and Sak
and Napolitano (2010).
Analogous to the fundamental concept of bit in clas-
2.4 Quantum Statistics sical computation and classical information, we have
its counterpart, quantum bit, in quantum computation
For a given quantum system, it is very important but
difficult to know its state. If we do not know in ad- and quantum information. Quantum bit is called qubit
for short. Just like a classical bit with state either 0 or
vance the state of the quantum system, we may in-
1, a qubit has states |0) and |1). However, the real dif-
fer the quantum state by the measurement results of
ference between a bit and a qubit is that besides states
some observables obtained from the quantum system
|0) and 1 1), a qubit may take the superposition states,
and show that a certain state has been created. In statis-
tical terminology, we want to estimate density matrix IVO =ao|0) +ai|l),
p based on measurements on an often large number where «o and a' are complex numbers and called am-
of systems which are identically prepared in the state plitudes satisfying |aol2 + |o;il2 = l- That is, the states
p. That is, after measuring observables on some identi- of a qubit are unit vectors in a two-dimensional com-
cal quantum systems, we can make statistical inference plex vector space, and states |0) and |1) consist of an

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 378 Y. WANG

orthonormal basis
where amplitudes ax are complexfor the
numbers satisfying
to as computational
l«ool2 + |«oil2 + l«iol2 + |«n|2 = 1. As in basis
the single
can examine it to determine whether it is in the state qubit case, when two qubits are measured we get result
0 or 1 . However, for a qubit we cannot determine its x being one of 00,01, 10, 11, with probability |ax|2.
state and find the values of «o and ai by examining it. Moreover, we may measure just the first qubit of the
The stochastic nature of quantum theory shows that we two-qubit system and obtain either the result 0, with
can measure a qubit and obtain either the result 0, with probability |aool2 + |«oi I2» or the result 1, with proba-
probability |ccol2> or the result 1, with probability 'a' '2. bility |ofiol2 + |of 1 1 12. As quantum measuring changes
Physical experiments have realized qubits as physical the quantum state, if the measurement result on the first
objects in different physical systems, such as the two qubit is 0, after the measurement the qubits are in the
states of an electron orbiting a single atom, the two dif- state
ferent polarizations of a photon, or the alignment of a
aoo|00) +aoi|01)
nuclear spin in a uniform magnetic field. Consider the P/ j •
case of atom model by corresponding |0) and |1) with y l«ool2
the so-called "ground" and "excited" states of the elec-
tron, respectively. As the atom is shined by light with A qubit is t
suitable energy and for a proper amount of time, the tum system
electron can be moved from the |0) state to the 1 1) state al complex v
and vice versa. Furthermore, by shortening the length specified by
of time shining the light on the atom, we may move the tially in b, i
electron initially in the state |0) to "halfway" between an enormou
|0) and 1 1), say, into a state (|0) + |1))/V2. with even a
Note that for qubit state | 'jf) the only measurable sources of e
quantities are the probabilities |aol2 and |ai|2; since a quantum s
'e'eax' 2 = la* I2, where x = 0, 1, i = V- Ï, and 0 is amplitudes a
a real number, from the viewpoint of the qubit mea- With 128 bit
surements, states eie''¡r) and 'ý) are identical. That is, 32 thousand
multiplying a qubit state by a global phase factor eld complex am
bears no observational consequence. available in f
Note the distinction between superposition states tem with b
and probability mixtures (or ensemble of pure states plex amplitu
defined in Section 2.1). Consider superposition (|0) + store all 250
|l»/>/2as a pure state. Its density matrix is given by puters. In pr
5CIO) + |1»«0| + (1|) hundred atom
of data and e
= 5 (|0) <0| + |1)(1|) + |(|0)(1| + |1)(0|), Quantum com
while the first term on the right-hand side of the above find ways to
equation corresponds to the ensemble of pure states |0) tional power
and 1 1), that is, a probabilistic mixture of states |0) and
3.2 Quantum Circuit Model
|1) with equal probability.
Similar to classic bits, we can define multiple As a classical computer is built from an electri-
qubits. The states of b qubits are unit vectors in a 2b- cal circuit consisting of wires for carrying informa-
dimensional complex vector space with 2b computa- tion around the circuit and logic gates for perform-
tional basis states of the form 'x'x2 - ■ -Xb), Xj = 0 or ing simple computational tasks, a quantum computer
1, j = 1, . . . , b. For example, the states of two qubits can be created from a quantum circuit with quantum
are unit vectors in a four-dimensional complex vec- gates to perform quantum computation and manipulate
tor space, with four computational basis states labeled quantum information. A number of physical systems
by 1 00), |01), 1 10) and |11). The computational ba- are being investigated for building quantum comput-
sis states 1 00), |01), 1 10) and 1 1 1) generate the four- ers. These include optical photon, optical cavity quan-
dimensional complex vector space, and the superposi- tum electrodynamics, ion traps, nuclear magnetic res-
tion states are all unit vector in the space with the forms
onance with molecules, quantum dots, and supercon-
IVO =aoo|00) +«01 101) +aio|10> +an|ll), ductors (Nielsen and Chuang, 2000). In fact, prim-

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 379

3.3 been
itive solid-state quantum processors have Entanglement
cre-
ated in research laboratories to run quantum algorithms
Quantum entanglement is one of the most
(DiCarlo et al., 2009; Johnson et al., 201 1; Mariantoni
bending creatures known to science. It is refe
et al., 2011; Sayrin et al., 2011). The circuit model
as the phenomenon that two qubits behave lik
is particularly important in quantum computation and
that are connected by an invisible wave to sha
quantum information, and a quantum computer is often
other's properties.
synonymous with the quantum circuit model. A quan-
tum circuit operates on b qubits for some3.3.1integer
Bellb.
states. Consider a quantum gate on
The state takes a form of |xi • • -Xb), with state
qubit space
basis states |00), |01), 1 10) and 1 1 1 ) that
being a 2b -dimensional complex Hilbert posed
space.ofWhen
a Hadamard gate on the first qubit and
x¡ = 0 or 1, states 'x'---xb) are the computational
followed ba-by a control-NOT gate. The output st
the
sis states of the quantum computer and often gate are
written as as follows:
|x), where x is the integer with binary representation
100) H- 111) |01) + 110)
i°°) ^ - vi - • |01>^- 7|- ' •
X' •■■Xb-

As a classical logic gate converts classical bits from

one form to another such as 0 - >• 1 and 1 -> 0, a quan-
tum gate operates on qubits. Quantum mechanics dic-V2
lio)^ |00»-J">, ui)^|01)-J10>.
V2
tates that quantum gates operating on b qubits are 2b
Physicists Bell, Ei
by 2b unitary matrices on the 2b -dimensional Hilbert
ered the amazing p
space. For example, a Hadamard gate on one qubit
are often referre
is the 2 x 2 unitary matrix that realizes the following
transformation:
EPR pairs (Bell, 1
1935). In general s
1A, 10) + ID 11X 10) -ID cannot be express
1A, 101 11X ^/I- ' are called entangle
not fully understo
Consider another important gate on two qubits which is
called control-NOT gate. It takes the two input qubits able properties.
as control qubit and target qubit, respectively, and the For the two-qubit
output target qubit of the gate retains the input target ,,x 101) -|1
qubit if the control qubit is |0) and is flipped if the con- 1« ,,x =
trol qubit is 1 1), that is,
and an observable
100) -H00), 101) -H01),
M - dXCÍ X ~l~ Cl y Ci y + d^Cf
|10) -> 111), 111) -►110).
where (ax,ay, az) is a real unit vector (i.e., aj + +
Generally for any single qubit unitary operation U, a
a' = 1), and crx, a y and <yz are Pauli matrices given by
control-U gate is a two-qubit gate, with one control
qubit and one target qubit. If the control qubit is |1),
U is applied to the target qubit; if the control qubit is
|0), the target qubit is left alone, that is,
m •■-(! ;)■ --(! í)-
|0)|0)-H0)|0), |0)|i)-H0)|i),
1 1 ) |0) -> 1 1 )U|0) , |1)|1)-H1)U|1). It is easy to show that M has eigenvalues ±1 for any
real unit vector (ax,ay,az). If measuring observable
If /(x) maps {0, ' }b onto {0, 1}, we define M a unitary
on each qubit of | '¡r), we will obtain a measure-
transformation U / that operates on b + 1 qubit
mentstate
result of +1 or -1. Surprisingly, no matter what
choice of (ax,ay,az), the measurement results on the
(6) |x,;y) ->• |x,y©/(x)),
two qubits are always opposite of each other, that is,
when
where x = x' ■ • • Xb with xj = 0 or 1 is the datatheregister,
first qubit measurement is -1, then the sec-
y = 0 or 1 is the target register, © denotes additional
ond qubit measurement will be +1, and vice versa.
modulo 2. If y - 0, after the transformation The two-qubit
U/, the system can be realized by the spins of
state of the last qubit is the value of /(x).
two particles, and the measurement of M is referred to

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 380 Y. WANG

as a real 0. Substituting the above expressions into

measurement of the en- spin
After the tangled state and ignoring a global phase factor el°
two-particle sy
state |f), (which has no two
the effects on measurement results;
particlesee Sec-
sures the tion 3.1), we obtain of
spin the fir
the spin of the second
|01) - 1 10) w I (p+(p-) - I <p-<p+)
measurement phenomeno
3J- ' - vi -
the (8)
entangled state 'ý), if
I <P+<P~) - I <P-<P+)
her spin measurement on
state of the •v/2
system immed
state so that the
From the definitions of | <p±) and Q±,second
Q+ <g>I|^>+^_) = p
state and Bob's spin
'<p+<p-),Q-®l'<p-<p+) = -'<P-(P+),Q+®''(P-<P+) = meas
cle gives definite result
0, and Q_ ®''(p+<p-) = 0. If the measurement result of -
referred toM on as
the first qubitanti-correl
is +1 (or -1), from (8) we obtain
ments (Neumann et
the post-measurement state of the two-qubit system as al., 2
2010). follows:
The mathematical arguments for the anti-correlation Q+ ®l'f)
phenomenon are as follows. The measurement of M on
''Q+®I'f)''
the first (or second) qubit of | ý ) corresponds to the spin
measurement of Alice's (or Bob's) particle along the _ i0Q+®l'<p+<p-) -Q+®l|y-i+)
(ax,ay,az) axis in the above two-particle spin model. V2IIQ+ <8)I|^)||
From Sections 2.3 and 3.1 we have that the two-qubit
system is described by the Bell state 'ý) in C4; mea- _e,g jy+y~| =el6' (p+(p-) ~ I <p+<p-)
suring M on the first qubit of 'ý) means performing
measurement on observable M<g>I in the Bell state 'xf/), (orQ_<g>I|f )/||Q_®I|f)|| =e,e'<p-<p+) ~ | <p-<p+)).
which alters the quantum state of the two-qubit system; Since

measuring M on the second qubit corresponds to mea-

I®M|<¡0+<p_) =I(g>Q+|<p+^>_) - I<g)Q-|iP+^-)
suring observable 1 0 M in the altered quantum state,
where I is the 2 by 2 identity matrix, and M <g> I and = -I
I ® M are matrix tensor products. I®]V%_<p+) = '<p-<p+),
Denote by '(p±) the two orthonormal eigenvectors of
M corresponding to eigenvalues ±1, respectively, and that is, the post-measurement state '<p+<p~) (or | <p-<p+))
let Q± be the respective projections onto the eigenvec- is the eigenvector of I ® M corresponding to eigen-
tors I <p±). Following (3)-(5) we have a diagonal repre- value - 1 (or +1), performing measurement on I (g» M
sentation M = Q+ - Q_ ; when we measure observable in the post-measurement state must always yield mea-
M on each qubit, the possible measurement results are surement result - 1 (or +1). Thus, the measurement re-
± 1 ; measuring M on the first qubit changes the state of sults of M on the two qubits of 'ý) are always opposite
to each other.
the two-qubit system, and after the measurement result
± 1 on the first qubit, the post-measurement state of the 3.3.2 Quantum teleportation. Quantum teleporta-
two-qubit system is Q± <g> I|i/0/IIQ± <8> IIV0II- Below tion is a process by which we can transfer the state
we will evaluate the post-measurement state and show of a qubit from one location to another, without trans-
that measuring I <g> M in the post-measurement state al- mitting it through the intervening space. We illustrate
ways yields measurement results opposite to the mea- the phenomenon as follows. Alice and Bob together
surement results on the first qubit. generated a Bell state long ago. Each took one qubit of
Since (|0), |1)) and ('<p+), '<P-)) are two bases for the Bell state when they split. Now they are far away
the one-qubit system in C2, then from each other. The mission for Alice is to deliver
a qubit 'ir) to Bob, while he is hiding, and she can
ai2]/l<P+)' only send classical information to Bob but does not
VU)/ L«21 «22 J V I
know the state of the qubit If). Quantum teleporta-
where (a j¿) forms a 2 x 2 unitary matrix with deter- tion is a way that Alice utilizes the entangled Bell state
minant equal to a phase factor e'e (i = V^T) for some to send a qubit of unknown state to Bob, with only a

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 381

small overhead of classical communication.in the statea |00) and Bob's qubit in state |f ). There-
Recently
breakthrough in quantum teleportation hasfore,
beenifmade
Alice's measurement result on her qubits is 00,
thendata
by successfully transferring complex quantum Bob's qubit will be in state |f ). Below is a list of
in-
Bob's four
stantaneously from one place to another, paving thepost-measurement states corresponding to
way for real-world applications of quantumthe results of Alice's measurements:
communi-
cations (Lee et al., 201 1).
00 - >■ ofo |0) - I- of 1 1 1 ) , 01 -> atoll) +t*i|0),
Here is how it works. Alice interacts the qubit 't¡r) to
10 -then
be teleported with her half of the Bell state, and > a0|0) - «i 1 1 ) , ll-»ao|l)-ai|0).
performs a measurement on the two interacted qubits
As Alice's measurement outcome on her qubits is one
to obtain one of four possible two-classical-bit results:
of 00, 01, 10 and 11, depending on her measurement
00, 01, 10 and 11. She sends the two-bit information
outcome Bob's qubit will be one of the above four pos-
via classical communication to Bob. Depending on Al-
sible states. Once Alice sends to Bob her two-classical-
ice's classical message, Bob performs one of bitfour op-
measurement outcome through a classical chan-
erations on his half of the Bell state. Surprisingly, the
nel, he applies appropriate quantum gates to his state
described procedure allows Bob to recover the original|f ). For example, if her measurement is
and recovers
state |f). 00, Bob's state is If), and he does not need to apply
Specifically assume that the state to be teleported is any quantum gate. If her measurement is 01, then Bob
If) = ao|0) + of 1 1 1 > , where ao and a' are unknown needs to apply a a x gate to his state aol 1) + a' |0) and
amplitudes. First, consider a three-qubit state yields |f ). If her measurement is 10, then applying a
a z gate to his state ao|0) - «i 1 1) Bob recovers |f ). If
, X ,,J00) + |11>
, l<Po) X = If) her measurement is 11, then Bob can fix up his state
V2
ao|l) - <*i|0) to recover | f ) by applying first a a x
gate and then a az gate. Here the a x and az gates are
= -Uao|0>(|00) + 111» +«i|l>(|00>
V2 defined by Pauli matrices ax and <rz given by (7). In
summary, according to Alice's measurement outcome,
where the first two qubits (on the left) belong to Alice,
applying some appropriate quantum gates to his qubit
and the third qubit to Bob. Note that Alice's second
Bob will recover the state |f ).
qubit and Bob's third qubit are from the entangled Bell
A few important remarks about quantum teleporta-
state. Second, Alice applies a control-NOT gate to her
tion are in the line. First, quantum teleportation does
qubits in |<po) and obtains
not involve any transfer of matter or energy. Alice's
particle has not been physically moved to Bob; only
'n) = ^[«0|0)(|00) + |ii)) + «i|i>(|io) + |oi»]. its state has been transferred. Second, after the tele-
Third, she applies a Hadamard gate to the first qubit in portation Bob's qubit will be on the teleported state,
'<p') and gets while Alice's qubit will become some undefined part
of an entangled state. In other words, what the telepor-
'<Pi) = 2t«o(|0) + |1))(|00) + |11>) tation does is that a qubit was destroyed in one place
+ ai(|0)-|l»(|10) + |01»]. but instantaneously resurrected in another. Teleporta-
tion does not copy any qubits, and hence is consistent
We regroup the terms of 'q>2) and rewrite it as follows:
with the no-cloning theorem (which forbids the cre-
ation of identical copies of an arbitrary unknown quan-
'n) = j[|00)(ao|0) +ai|l>) + |01)(ao|l) +ai|0))
tum state; see Wootters and Zurek, 1982). Third, in or-
+ |10)(a0|0)-ai|l)) der to teleportate a qubit, Alice has to inform Bob of
+ |ll)(a0|l>- ai|0»]. her measurement by sending him two classical bits of
information. These two classical bits do not carry com-
The new expression has four terms, and each term pletehas
information about the qubit being teleported. If
Alice's qubits in one of four possible states |00),the|01),
two bits are intercepted by an eavesdropper, he or
1 10) and 1 11), and Bob's qubit is in the state related to know exactly what Bob needs to do in order to
she may
recover the desired state. However, this information is
the original state IVO - If Alice performs a measurement
on her qubits and informs Bob of the measurement re-if the eavesdropper cannot interact with the en-
useless
sult, then his post-measurement state is completelytangled
de- particle in Bob's possession. Also the require-
ment of sending two bits of information via classical
termined. For example, the first term has Alice's qubits

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 382 Y. WANG

channel on the second qubit,

prevents quantumwhere ax and az ar
ting ces given by (7).faster
information Again Alice performs
th
on Xi or X2 while Bob measures X3 or X
3.3.3 inequality. Bell's T
tum expectations of X1X3, X2X3, X2X4
are designed to investiga
state IV") are calculated below:
glement effect in quantu
inequality. Over the pas
cal experiments EÝ(XyX3) = -j=, Eý(X2X3) = -±=,
on quant
to check the validity of
in some Ex¡,(X2X4) = E^(XiX4) = - - j=.
violation of the
pect, Grangier and Rog
vided overwhelming
Here the observable product is in the sense of tensor supp
inequality. The
product. experimen
Thus we obtain a value in the quantum frame-
as the proofwork for theof analog quantityquantum
on the left-hand side of
ism that no particle
the Bell's inequality (9) has d
sured and measuring a qu
neously influence
Eý(XiX3) + E* (X2X3) + E^(X2X4) another
ier and Roger - E^(XiX4) = (1981,
2V2, 19
Bell (1964), Clauser et al
sky and which exceeds 2 and hence violates
Rosen the Bell's inequal-
(1935). B
version of ity.
the In fact, the quantum Bell's
version of the Bell's inequal- ineq
Suppose ity is the
Xj, i Tsirelson's
= inequality
1, (Tsirelson, 1980)
2, which 3, 4,
taking shows
values that in any quantum±1.state p, Consid
with two people, Alice an
E„(XiX3) + E,(X2X3) + E(9(X2X4)
ice observes
(10) Xi or X2 w
Consider the -quantity
E^XiX^ < 2V2. X1
It is equal to
3.4 Quantum Parallelism
(Xt + X2)X3 + ( X2 - Xx)X4 = ±2 < 2.
Quantum computation has an amazing feature
Regardless of the distributions of X¡, taking expecta- termed as quantum parallelism, which may be heuristi-
tion on both sides of the above inequality we arrive at cally explained by the following oversimplifying de-
the famous Bell inequality, scription: a quantum computer can simultaneously
E(X 1X3) + E(X2X3) + E(X2X4) evaluate the whole range of a function fix) at many
different values of x.
(9)
-E(X 1X4) <2. For function f(x) with b bit input x = x' • Xb and
1 bit output f(x), we illustrate quantum parallel eval-
The violation of Bell's inequality demonstrates en-
uation of its values at many different x simultaneously
tanglement effect in quantum mechanics. In fact, quan-
as follows. First we apply b Hadamard gates to the first
tum experiments yield a quantum version of the in-
b qubits of |0 * - * 0) |0) to obtain
equality. Consider that a quantum system of two qubits
is prepared in a Bell state 101 + 11) |0) + |1),„, 1
l01> -11°) -R
W) s/2 ' x = x' ■ ■ ■ Xb, Xj = 0, 1 ,

Alice takes the first qubit of IVO while Bob gets its where the sum is over all possible 2b va
second qubit. Define four observables with eigenvalues ond, apply quantum circuit U/ defined
±1, obtained b + 1 qubit state to yield
Xi=(Tz, X2=Ox,
on the first qubit and ¿ E «!/«>•
oz + ox az - ax The quantum circuit with b Hadamard gates is ex-
= 7T~" X4=~vl~'
tremely efficient in producing an equal superposition

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 383

of all 2b computational basis states with only

that b
allgates;
problems solvable on a quantum computer are
and quantum parallelism enables simultaneous evalua-
solvable on a classical computer, and problems unde-
tion of the whole range of the function /, although
cidable by weclassical computers remain undecidable on
evidently evaluate / just once with single quantum
quantum cir-
computers, what makes quantum algorithms
cuit U / applied to the superposition state. exciting
To make is it
the faster speed that quantum algorithms
more clear we consider the case of b = 1 . Apply
mightcircuit
be able to achieve, compared to classical algo-
U/ to a superposition state as follows: rithms, for solving some tough problems. The well-
known quantum algorithms are Shor's factoring algo-
rithm and Grover's search algorithm. Shor's algorithm
and Grover's algorithm run, respectively, exponentially
|0/(0)) + |l/(l)> faster and quadratically faster than the best known clas-
V2 sical algorithms for the same tasks. Common tech-
niques used in quantum algorithms include quantum
One application of a single circuit U / results in a su-
perposition state whose two components contain infor- Fourier transform, phase estimation and quantum walk.
mation about both /( 0) and /(l), as if we have evalu- 4.1 Quantum Fourier Transform
ated f(x) at values 0 and 1 simultaneously. The quan-
The quantum Fourier transform is defined to be a lin-
tum parallelism is in contrast with classical parallelism,
where multiple circuits each built to compute one value ear transformation on n qubits that maps the computa-
of f(x) are executed simultaneously. Quantum paral- tional basis states |y), j = 0, 1, . . . , 2" - 1, to superpo-
sition states as follows:
lelism arises from superposition states. A superposi-
tion state has many components, each of which may
be thought of as a single argument to function f(x). L/)^4=2¿1*2*W2"lít>' i =
V 2" k=o
Because of quantum nature, a single circuit U / applied
once to the superposition state is actually performed on The inverse of quantum Fourier transform is given by
each of the components of the superposition, and the
whole range of the values of function f(x) is stored in
|t>- >-L2¿V2"W2"i;>.
the resulted outcome superposition state. V2 j=0
The quantum parallelism can be a potentially pow-
We use the binary representation to express the state
erful tool for computational statistics. For example,
j = j]2n~l + j22n~2 -'
Bayesian analysis often encounters the problems of
evaluating sums over 2b quantities, with b proportional and represent binary fraction j¿/2 + je
to sample size or the number of variables. For moder- jm/2m~t+1 as O.jeje+i ■•■jm, where
ate to large b, the evaluation of such sums is computa- Then the quantum Fourier transform
tionally prohibitive by classical computers (Vidakovic, 1 7172 • • • jn) has the following useful pro
tation:
1999). Because of the quantum parallelism, it is possi-
ble for quantum computers to perform such computing
tasks. 'j'h 2^72 (|0> + e27li0'in I i
4. QUANTUM ALGORITHMS
•(|0) + c2,r,'a-'"-1Ä|l))"-
■('0)+e27Ii0-iìh~j"'l)).
Quantum algorithms are described by quantum cir-
It can be easily checked from the prod
cuits that take input qubits and yield output measure-
ments for the solutions of the given problems. Astation
a that with quantum parallelism
Fourier transform can be realized as a q
classical algorithm is a step-by-step problem-solving
procedure, with each step performed on a classical with only O ( n 2) operations, while clas
computer, a quantum algorithm is a step-by-step pro-
Fourier transform requires 0(n2n) opera
cedure to solve a problem, with each step executed by
cessing 2" data, which indicates an expo
up (Nielsen and Chuang, 2000). Reali
a quantum computer. Although all classical algorithms
can also be carried out on a quantum computer, we exponential
re- saving accommodated by qu
fer to quantum algorithms as the algorithms that uti-
lelism requires clever measurement sch
ful examples include quantum phase e
lize essential quantum features such as quantum super-
position and quantum entanglement. While it is trueShor's algorithms for factoring and disc

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 384 Y. WANG

4.2 Phase Estimation O.771 í?2 ■■•rib is the first b bit representation in the bi-
nary expansion of (p , which satisfies 0 <<p - t]/2b <
Quantum algorithms are of random nature in
2-bthe
sense that they are able to produce correct answers only
Perform the inverse quantum Fourier transform on
with some probabilities. Consider quantum phase esti-
the first register given by (1 1), which is obtained in the
mation which provides the key to many quantum algo-
first stage results, and get
rithms. Assume that a unitary operator U has an eigen-
vector 'x) with eigenvalue e2ni<p . The phase <p of the
eigenvalue is unknown and the goal of the phase es- 1 £ e-2mUI2»e2ni^g)=2Y^ M¿h
timation algorithm is to estimate <p based on the as- M=o e=o

sumption that the state |x) can be prepared and the

where amplitud
controlled-U2J operations [see Section 3.2 for control
2b-'
gate] can be performed for suitable nonnegative inte-
gers j.
2b k= 0
The registers are used in phase estimation. The first
register consists of b qubits initially in the state |0). I / i _ e2xiQ.b<p-ii-l) '
The second register starts in the state |jc) and involves
= 2^ V 1 - e2ni(<p-r)l2b-t/2b) J '
enough qubits to store |x). The phase estimation pro-
cedure is performed in two stages. Assume that the result of the final measurement from

First, we apply Hadamard transform to the first reg- phase estimation is ij and dividing the result by 2b gives
ister and then controlled-U operations on the second <p = ř)/2b. Let Ç be the specified accuracy for the phase
register, with U raised to successive powers of 2, to ob- estimation procedure. By adding up Ijö^l2 with I being
tain the final state with the second register unchanged within Ç2b, we bound the probability that the obtained
and the first register given by <p is within Ç from <p:

p('v-<p'<o>p('n-ri'<ab-v
¿(|0>+*2*i2*-,«'|l>)
(11) • (|0) + e2ni2b~2<e'')) • • • (|0) + e2jti20,fi'l)) - 1 ~ 2(ç2b - 2) '
For € > 0, set

= 255k=0
S«2"**!«-
(13) i,= [l„g2(i)] + [log2(2+i)].
If <p is expressed exactly in b bits as <p = 0.<pi ■■■(pb,
(11) becomes Then P{'ip - <p' < Ç) > 1 - e, that is, with probabil-
ity at least 1 - e the phase estimation procedure can
J_(|0 )+e2^b'')) successfully produce čp within Ç from the true (p. See
Nielsen and Chuang (2000).
(12) ■('0) + e2ni()(('b-m''))--- 4.3 Statistical Analysis
. ('Q) + e2ni°4'm "(pb'l)),
The phase estimation algorithm requires b qubits for
which is the quantum Fourier transform of the product the first register to achieve [- log2 £] bit accuracy and
state '(pi<p2---<pb)- success probability 1 - e. With accuracy fixed, to in-
The second stage of phase estimation is to take the crease the success probability the required qubits
inverse quantum Fourier transform on the first register.
For <p = 0.<pi ■■■<pb, the output state from the second
stage is '<p'cp2 • • • <Pb), and a measurement in the com-
t~1-log2f+47ib-
putational basis yields (p'---(pb and dividing the mea- which grows at a very fast rate. For example, an in-
surement by 2b gives <p' - ■ ■ (pb/ 2b = 0.<p' ■ ■ ■ <pb = (p. crease in success probability from 90% to 99% re-
We obtain a perfect estimate of cp. quires eighteen times of qubit increase compared to the
Now we consider the case that <p cannot be expressed change from 80% to 90%.
exactly with a b bit binary expansion. Take 0 < t] < Quantum algorithms are of random nature in the
2b to be the integer that its binary fraction r¡/2b = sense that they often produce correct answers only with

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 385

certain probabilities. The success probabilities depend

that (p is within Ç from <p can be calculated from the bi-
upon the schemes of the algorithms as well nomialas probability
the con- as follows:
text of applications. Given a quantum algorithm for
P{'ip - <p' < f ) > P (more than n( 1 - 2a) number
solving a problem, a common practice is to repeatedly
run the quantum algorithm to achieve high probability of <pj are within Ç from <p)
of successfully obtaining a correct answer. Consider
that the phase estimation procedure is repeatedly run
n times to obtain results <p', ... ,<pn. Then cp', . . . ,<pn k=[n(l-2a)]-l V 7 = t (ï)a-ov-*. V 7
may be treated as i.i.d. random variables with each ¿pj
As n - > oo, ti€ approaches to infinity. The binomial
satisfying
probability can be approximated by resorting to a nor-
P('<pj-<p' > Ç) <e. mal approximation, yielding

We may statistically model <pj by the gross error model

(Huber and Ronchetti, 2009) as follows. Assume that ± (¡»Vi-ov-* V 7
k=[n( 1- 2a)]- 1 V 7
<Pj are independently and identically generated from
(1 - €)F(x) + e H (x ), where F(x) is the distribution ~ i _ = d
of the correct answers that are within Ç from true <p, ' ) = d I )'
and H(x) is the distribution of wrong answers that are
where 3>(.) is the standard norma
at least Ç away from true <p. Then <pj are correct with
tion. Since 2a - 6 > 0, as n incre
probability 1 - e and incorrect with probability e.
approaches to 1 exponentially fast
If the outcome result of the algorithm is verifiable to
cases together, we arrive at the fo
be a correct answer or not [as in the case of Shor's al-
gorithms for factoring and order-finding in Section 4.4 THEOREM 1 . Suppose that th
below], the obtained result from each run is checked to quantum algorithm obeys the gro
be a correct answer or not. Then the number of times with probability 1-6 it produces a
required to run the algorithm for obtaining a correct an- probability € it gives a wrong ans
swer follows a geometric distribution. Thus the proba- edly running the quantum algorit
bility that we obtain a correct answer in n repetitions is correct answer with probability a
equal to nentially fast in the number of repe

P (obtain a correct answer in n trials) For a quantum algorithm that pr

swer with probability 70% and a =
= 1 - />(no success in the n trials) = 1 - €n.
tain a correct answer with 0.999 p
Since €n goes to zero geometrically fast, we may run the algorithm five times and 20
choose a moderate e with fewer qubits to achieve very for the cases that the outcome results are verifiable and
high probability of successfully obtaining a correct not verifiable.

answer by repeatedly running the algorithm enough

times.
4.4 Factoring and Order-Finding Algorithms
On the other hand, if the outcome result is not ver- The factoring problem is to find all prime factors of
ifiable to be a correct answer or not [as in the case a given positive composite number such that the prod-
of phase estimation], careful analysis is needed to de- uct of these prime numbers is equal to the composite
sign ways for obtaining a correct answer with very number. Factoring is known to be a very hard problem
high probability. As wrong answers are far away from for classical computers. Shor (1994, 1997) developed
true <p, estimators like sample average of <p',...,ipn a quantum algorithm for the factoring problem that is
may not estimate <p well. We adopt a robust statisti- exponentially faster than the most efficient known clas-
cal method to estimate <p by a -trimmed mean ip, which sical factoring algorithm.
is defined as follows. Ordering ip' , . . . , <pn and then re- Shor's quantum algorithms work as follows. Math-
moving [ na ] largest and 'na ] smallest ones, we take ematically the factoring problem is equivalent to the
the average of the remaining <pj as a -trimmed mean, order-finding problem that for two positive integers x
where a is chosen to be greater than e/2. One exam- and N, x < N, with no common factors, find the small-
ple is the sample median of <p' , . . . , ¿pn . The probability est integer r such that dividing xr by TV we obtain a

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 386 Y. WANG

reminder 1
difficult,(Shor, 1997
though not impossible, to invert the encryp-
The quantum algorithm
tion transformation. When publishing the public key
quantum Bob keeps a matched secret key for for
algorithm easy inversion of
o
gorithm for order-findin
the encryption transformation and decryption of the re-
tion algorithm to
ceived information. the
One of unit
the most widely used pub-
lic key cryptosystems is the RSA cryptosystem, which
U|y) = |xy(modJV)).
is named after its creator Rivest, Shamir and Adleman
The (Menezes, van Oorschot and
eigenvectors ofVanstone,U 1996; Rivest
ar
Shamir and Adleman, 1978). RSA is built on the math-
, , 1 v-i ( ~2 7tisk' k ematical asymmetry of factoring: it is easy to multi-
, l"i) , = -7= 2^exP(
v ¡ r k=0 ' r / ply large prime numbers and obtain their product as
composite number but hard to find the prime factors
s = 0, 1 , . . . , r - 1 , i = '/- I,
of a given large composite number. RSA encryption
keeps
with corresponding eigenvalues exp(2nis/r). the large primes as a secret key and uses their
Using
the phase estimation algorithm we can obtain the to make a "public key." Because of its expo-
product
eigenvalues exp(27ns/r) with high accuracy and nential
thus complexity, tremendous efforts tried to break
find the order r with certain probability. the RSA system so far have resulted in vain, and there
While the quantum factoring algorithm canisaccom- a widespread belief that the RSA system is secure
plish the task of factoring an «-bit integer with op-any classical computer based attacks. As the
against
erations of order n2 log n log log n, the currentfactoring
best problem can be efficiently solved by Shor's
known classical algorithm requires operations quantum
of order factoring algorithm, a quantum computer can
exp (w1/3 log2/3 n) to factor an n-bit composite break
number the RSA system easily. Fortunately, while quan-
(Crandall and Pomeranče, 2001). Note that thetum num-mechanics takes away with one hand, it gives back
ber of operations required in the best classical with
algo-the other. A quantum procedure known as quan-
rithm grows exponentially in the size of the tum cryptography or quantum key distribution can do
number
key distribution so that the communication security
being factored. Because of the exponential complex-
ity, the factoring problem is generally regarded cannot
as anbe compromised. The idea is based on the quan-
intractable problem on classical computers. tum principle that observing a quantum system will
The factoring problem plays an important role disturb
in the system being observed. If there is an eaves-
cryptography. Cryptography is to enable two dropper
parties,during the transmission of the quantum key be-
Alice and Bob, to communicate privately, whiletween it
Alice and Bob, eavesdropping will disturb the
quantum communication channel that is used to estab-
is very difficult for the third parties to "eavesdrop"
on the contents of the communications. Examples in-key, and the disturbance will make eavesdrop-
lish the
ping
clude ATM cards, computer passwords, internet com- visible. Alice and Bob will throw away the com-
promised key and keep only the secured key for their
mences, clandestine meetings and military communi-
communication.
cations. Two cryptographic protocols used in the com-
munications are private key cryptosystem and public
4.5 Quantum Search Algorithm
key cryptosystem. A private key cryptosystem requires
the two communicating parties to share a privateSuppose
key.that you would like to find the name corre-
Alice uses the key to encrypt the information, sends
sponding to a given phone number in a telephone direc-
the encrypted information to Bob who uses thetory;key
or suppose that there are some locations in a given
to decrypt the received information. The severe
city draw-
you would like to visit and wish to find the shortest
back of the private key cryptosystem is that route passing through all the locations. If there are N
the par-
names
ties have to safeguard the key transmission from in the telephone directory or N possible routes
being
eavesdropped. A public key cryptosystem invented in
to pass through all the locations, search algorithms by
the 1970s requires no sharing secret key in advance.
classical computers usually require operations of or-
Bob publishes a "public key" available to the gen-
der N. One such simple classical algorithm is to check
eral public, and Alice uses the public key to encrypt
exhaustively all names to find a name matching with
information and sends the encrypted information
the giventophone number or to search all possible routes
Bob. The encryption transformation is specially cre-find the shortest route among all routes. How-
and then
ever, Grover (1996, 1997) developed a quantum search
ated such that with only the public key, it is extremely

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 387

algorithm that needs only operations of order */Ñ

solution to search problem with probability at least
to the
find a solution to the search problem. cos2 (0/2) > 1 - M/N.
The quantum search algorithm works as The number of iterations R depends on M, the
follows.
Suppose that the search space has N elements and
number of solutions. Since R < ti/ (29) and 9/2 >
the search problem has exactly M solutions.sin (0/2) = R < (n/4 )y/WjM. Typically,
Assume
M < N / 2. (For the silly case of M > N / 2,M we <žC N, 9 sin# 2 y/M/N , thus R & (n/4) •
either
search for the solution by doing random selection
y/N from
/M. We estimate the number of solutions by quan-
the search space or double the number of the tumelements
counting, which is to combine the Grover operator
in the search space by adding N extra non-solution el- estimation method. Under the basis | <f>)
with the phase
ements to the search space.) The algorithm and works by Grover operator has eigenvalues eld and
I cp) the
creating superposition state with Hadamard gate,
ei(2n-o) Applying the phase estimation method we
can estimate the eigenvalues and thus 9 with prescribed
j N-Í
i «=pEw. x=0
precision and probability, which in turn yields M . The
combination of the quantum counting and search pro-
cedure will find a solution of the search problem with
and then applying a so-called Grover iteration (or op-
erator) repeatedly. Set certain probability. Repeating the quantum search al-
gorithm will boost the probability and enable us to ob-
tain a solution to the search problem.
Quantum walk and quantum Markov chain are cur-
rently being investigated for new quantum search algo-
where the summations over x' and x" denote sums over
rithms and quantum speed-up of Markov chain based
all non-solutions and solutions, respectively. Then we
algorithms (Aharonov and Ta-Shma, 2003, Childs et
can express 'ý) as follows:
al., 2003; Childs, 2010; Tulsi, 2008; Shenvi, Kempe

= + /!>>•
and Whaley, 2003 and Szegedy, 2004). In Section 5 we
show that the quantum search algorithm can also be
viewed as a quantum simulation procedure.
The Grover operator is to perform two reflections, one
about the vector '<f>) and another about the vector 'ý). 5. QUANTUM SIMULATION
The two reflections together are a rotation with angle 9
in the two-dimensional space spanned by '<f>) and '<p), Quantum simulation is to intentionally and artifi-
where cially mimic a natural quantum dynamics, which is
hard to access, and analyze, by a computer-generated
In- m
cos(0/2) = J - - - . quantum system, which is easy to manipulate and in-
vestigate. It provides scientific means for simulating
After the rotation, the initial state | '¡r) = cos(0/2)|</>) + complex biological, chemical or physical systems in
%in(9/2)'<p) becomes state order to study and understand certain scientific phe-
nomena and evaluate hard-to-obtain quantities in the
cos(30/2)| </>) + sin(36>/2)|^).
systems. Examples in modern scientific studies include
Thus each application of the Grover operator is a ro- the estimation of dielectric constant, proton mass, and
tation with angle 9. The initial state | 'fr) has angle precise energy of molecular hydrogen, the study of su-
n ¡2 - 6/2 with I (p)' after the first rotation, the resulted perconductivity, the test of novel nano-materials, and
state has angle n/2 - 39/2 with '<p) ; and in general af- the design of new biomolecules.
ter the rth rotation, the resulted state has angle n/2 - To simulate a quantum system we need to solve the
(2 r + 1)9/2 with | <p). Repeatedly applying the Grover Schrödinger equation (1) which governs the dynamic
operator, we rotate the state vector near | <p). With the evolution of the system. For a typical Hamiltonian with
initial state 'ý) = cos(ö/2)|</>) + ûn(9 /2)'<p) , we need real particles the Schrödinger equation usually consists
to rotate through arceos -JM/N radians to transform of elliptical differential equations, each of which can
the state vector to 'tp). After R = arccos ( ^/M/N) / 9 = be easily simulated by a classical computer. However,
Oi+JN/M) times of applications of the Grover opera- the real challenge in simulating a quantum system is
tor, we rotate the state vector | ý ) to within an angle 9 ¡2 to solve the exponential number of such differential
of '(p). Performing measurements of the state yields a equations. For a quantum system of b qubits, its states

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 388 Y. WANG

have 2b Thus we can approximate e~liis by which

amplitudes. Toneeds to si
ior of b qubits
evaluate only each e~l}ieS.
evolving
equation, we Assumeneed
that the quantum system to starts at t =solv
0 with
equations. initial state |^(0)) to
Due and ends at final
the time t = 1. Forexpan
ber of differential
integer m, set tj = j/m, j = 0, 1, . . .equati
, m. The quan-
quantum systems
tum simulation is to apply approximation by U¿ of e~2,mcla
efficient. Classical simula
to evaluate (14) at tj iteratively and generate approxi-
feasible for the cases whe
mate solutions to I ir(tj)}. Denote by | ý(tj)) the state
are available to dramatical
at tj obtained from the quantum simulation as an ap-
ber of equations involved
proximation of the true state 'ý(tj)) at tj. Then for
excel in simulating
j = 1, phys
tant quantum systems for
classical Mtj)) =*-2,'HV(0-l)> =e~2m^mo)),
computers may
(17)
5.1 Simulate a
I Ý(tj)) =Vsmtj-i)) Quantum
= VJs'Ý(to)).
The key ofWhile classicalquantum
computers are inefficient in simulat- s
Schrödinger equation
ing general quantum systems, quantum computers can (1)

(14) 'Ý(t)) efficiently carry= e~iHt'Ý

out the quantum simulation proce-
dure and provide an exponential speedup for the quan-
Numerical evaluation of
tum simulation on classical computers. In spite of the
tonian H isinefficiency,
usually expon
classical computers are currently being
difficult to exponentiate.
used to simulate quantum systems in biochemistry and
merical analysis iswill be
material science. Quantum simulation to among use
imation, 1 - i'H<5, of e~
the important applications of quantum computers. See
yields unsatisfactory
Abrams and Lloyd (1997), Aspuru-Guzik, Dutoi and num
Many classes of Hamilto
Head-Gordon (2005), Bennett et al. (2002), Berry et
tations. For such
al. (2007), Freedman, Kitaev and Wang sparse
(2002), Jané et H
ficient evaluation of
al. (2003), Boghosian and Taylor (1998), Lloyd (1996), the
approximation. Nielsen and Chuang (2000), For and Zalka (1998). examp
physical systems involve
originate 5.2 Recast Quantum Search
from the Algorithm as fact t
Quantum Simulation
with increasing distance
ference in energy.
Grover's search algorithm discussedIn the
in Section 4.5
Hamiltonian of
is an important a
finding quantum
in quantum computation. It can
¿-dimensional space has t
be heuristically sketched as a quantum simulation by
L
writing down an explicit Hamiltonian H such that a
(15) H = 2£Hť, quantum system evolves from its initial state 'i(r) to |jc)
i= l
after some specified time, where x is a solution of the
where L is a polynomial in a+d, and each He acts on asearch problem. Of course the Hamiltonian H depends
small subsystem of size free from a and d. For exam-on the initial state 'ý) and solution jc. Suppose that | y)
is another state such that 'x) and |y) form an compu-
ple, the terms 1 1¿ are typically two-body interactions
tational basis, and 'ý) =a'x) + ß'y) for real a and ß
and one-body Hamiltonians. Hence e~,H(S are easy
to approximate, although e~'HS is very hard to eval- with a2 + ß2 = 1 . Define Hamiltonian
uate. Since H( and are non-commuting, e~'HS ^
H= |x)(x| + 'x[r)(f'=l + a(ß<rx+acrz),
g-iHiS . . . e-iHLs Applying a modification of the Trot-
ter formula (Kato, 1978; Trotter, 1959; Yu, 2001) we where ax and <rz are Pauli matrices defined in (7).
obtain Then

<ríHá = Ua + 0(á2), exp(- /Hi) IVO

where = e~"[cos(at)'ì//} - i sin (at)(ßarx +a<rz)'ý )]
(16) Us = [e'iHiS ■ ■ ■ e-iHLS][e~iHLS ■ ■ ■ e~iHlS]. = e~"[cos(at)'Ý) - i sin(ař)k)]-

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 389

Measuring the system at time t = it /{2a) yields the 6. QUANTUM INFORMATION

solution state |jc).
Classical information theory is centered on Shan-
5.3 Quantum Monte Carlo Simulation non's two coding theorems on noiseless and noisy
channels. The noiseless channel coding theorem quan-
Quantum theory is intrinsically stochastic and quan-
tifies the number of classical bits required to store
tum measurement outcome is random. As many natu-
information for transmission by Shannon entropy,
rally occurring quantum systems involve a large num-
while the noisy channel coding theorem quantifies the
ber of interacting particles, due to the computational
amount of information that can be reliably transmitted
complexity we are forced to utilize Monte Carlo tech-
niques in the simulations of such quantum systems. through a noisy channel by an error-correction coding
The combination of Monte Carlo methods with quan- scheme. The quantum analogs of Shannon entropy and
tum simulation makes it possible to obtain reliable Shannon noiseless coding theorem are von Neumann
quantifications of quantum phenomena and estimates entropy and Schumacher's noiseless channel coding
of quantum quantities. Such combination procedures theorem, respectively. The von Neumann entropy is de-
are often referred to as quantum Monte Carlo sim- fined to be S(p) = - tr(/olog/o). Schumacher's noise-
ulation (Nightingale and Umrigar, 1999; Rousseau, less channel coding theorem quantifies quantum re-
sources required to compress quantum states by von
2008). Consider the problem of estimating the follow-
ing quantity: Neumann entropy (Schumacher, 1995). Analogous to
Shannon's noisy channel coding theorem, a theorem
(18) 0 = Tr (Xp) = E(X), known as Holevo-Schumacher-Westmoreland theo-
rem canresult,
where X is an observable, X is its measurement be used to compute the product quantum
and p is the state of the quantum systemstate capacity
under whichfor some noisy channels (Holevo, 1998;
we perform the measurements and evaluate the quan- Westmoreland, 1997). However, com-
Schumacher and
tity e. munications over noisy quantum channels are much
As p is the true final state of the quantum system, we less understood than the classical counterpart. It is an
denote by p the final state of the quantum system ob- unsolved problem to determine quantum channel ca-
tained via quantum simulation. The quantum systems pacity or the amount of quantum information that can
are prepared in initial state p0, and we use the quan- be reliably transmitted over noisy quantum channels.
tum simulation procedure described above to simulate See Hayashi (2006) and Nielsen and Chuang (2000).
the evolutions of the systems from initial state p0 to In spite of the above similarity, there are intrinsic
final state p according to Schrödinger's equation (14) differences between classical information and quan-
with Hamiltonian H given by (15). We repeatedly per- tum information. Classical information can be distin-
form the measurements of such n identically simulated guished and copied. For example, we can identify dif-
quantum systems at the final state and obtain measure- ferent letters and produce an identical version of a
ment results X', ...,Xn. We estimate 9 defined in (18) digital image for back-up. However, quantum mechan-
by ics does not allow unknown quantum states to be dis-
« 1 A tinguished or copied exactly. For example, we cannot
(19) 0 « = ~Y,Xj- reliably distinguish between quantum states |0) and
nU (|0) + |l))/'/2. If we perform measurement for quan-
The target 9 given by (18) is defined under the true tum state |0), the measurement result will be 0 with
state p, while the simulated quantum system is under probability 1, while measuring quantum state (|0) +
approximate final state p which is close to p. The mea- 1 1))/V2 yields measurements 0 or 1 with equal proba-
surement results X' , . . . , Xn are obtained via quantum bility. A measurement result of 0 cannot tell the iden-
simulation from the quantum systems in the simulated tity of the quantum state being measured. A theorem
state p. Therefore, the Monte Carlo quantum estimator known as a no-cloning theorem states that unknown
9 in (19) involves both bias and variance. Wang (201 1) quantum states cannot be copied exactly (Wootters and
studied the quantum simulation procedure and investi- Zurek, 1982; Nielsen and Chuang, 2000).
gated the bias and variance of 9. The derived bias and As we discussed in Section 3.3, quantum entan-
variance results can be used to design optimal strategy glement plays a crucial role in strange quantum ef-
for the best utilization of computational resources to fects such as quantum teleportation, violation of Bell's
obtain the quantum Monte Carlo estimator. inequality, and superdense coding (Hayashi, 2006;

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
390 Y. WANG

Nielsen and an
Chuang,
error 2000
syndr
type of resource
what that
error,diff
if
tional observable
resources in has
classical
far from sponding
having a proje
general
tum entanglement but enc
Qo = |000)(0
far reveals the amazing pro
ture of Qi
entangled = |100)
states <10
an
between noisybit
quantum
flip on th
c
transformation. Consider q
reliable Q2
quantum = |010)(0
computat
tion processing.
bitQuantum
flip on the
in quantum computation
tion to Q3
protect = |001)
quantum <0
in
to quantum noise
bit and on
flip othe
th
tum gates. Classical inform
If one of three
achieve error-correction, b
syndrome wi
presents an obstacle to cop
responding to
and formulating a theory o
the first, secon
based on simple redundan
ample, if the f
glement comes to the resc
is IV) =ao|10
qubits but we can spread th
(ÝiQjiif) = 0
onto a highly entangled st
drome is 1. Al
(1995) first discovered the
ally causes cha
quantum error-correction c
of the constru
tion of one qubit onto a hig
surement does n
qubits. Over time several
to check that th
codes are proposed (Calderb
syndrome mea
et al., 1998;information
Steane, 1996
abo
error-correction codes can
not contain an
tion against quantum noise
tected, that is,
likely poses no fundamenta
about the amp
of large-scale quantum
erty is the com
gene
mation processing.
Step 2. The er
Here is how drome
quantum can erro
info
We consider thethe original
single qub
st
qubit ao|0) + a' | 1)
dicates is apassed
bit f
which flips the state
first of
qubit a qub
again
|1) to |0), each with +
ao|000) probab
oři 1
states |0) and 1 1)
error untouche
and doin
describe a bit
3 flip code
correspondtha t
quantum noise from
third the
qubit,bit
res
We encode states |0)
original and
state 1 w
encoded as |000)
one and
bit 1 1) in
flip as t
ao|0) + a' 1 1) is
that encoded
more thi
«i 1 1 1 1 ) . We
3 pass
p2 - 2each
p3, of
wh
independent ability
copy ofp the
of mb
that at mostflip
one channel.
qubit is f
T
ple two-step makes
error-correcti
the stor
recover the reliable. quantum
correct
Step 1. Perform
Nextawe measure
consi
structed observable
channel: and
a cal
pha

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 391

/>, changes a qubit state «o|0)+ ail 1) to ao|0) - ai|l),

over the available classical algorithms. We illustrate
and with probability 1 - p, leaves alone quantum
the qubit.
simulation procedure and Monte Carlo meth-
The following scheme is to turn the phaseods flip
in chan-
quantum simulation. As classical computation
nel into a bit flip channel. Let |+) = (|0) and
+ |l))/'/2
simulation are ubiquitous nowadays in statistics,
and |- ) = (|0) - |l))/-s/2 be a qubit basis. we
The phase
expect quantum computation and quantum sim-
flip channel leaves alone states |+) and |- ) with ulation will have a paramount role to play in mod-
prob-
ability 1 - p and changes |+) to |- ) and ern vicestatistics.
versa This paper exposes the topics to statis-
with probability p. In other words, the phase flip ticians and encourages more statisticians to work in
chan-
nel with respect to the basis |+) and |- ) acts thejust like
fields. There are many statistical issues in theoret-
a bit flip channel with respect to the basisical |0) research
and |1). as well as experimental work in quan-
Thus we encode |0) as |+ + +) and |1) as | tum computation, quantum simulation and quantum
protection against phase flip information.
errors.For Theexample, as measurement
operations data col-
for
encoding, error-detection and lected in quantum experiments
recovery are the require
same more as
and more
for the bit flip channel but sophisticated with respect statistical methods
to the for |+)
better estimation,
and
|- ) basis instead of the |0) and simulation
|1) and understanding, it is imperative to de-
basis.
Last we describe Shor error-correction code. It is velop good quantum statistics methods and quantum
a combination of the three-qubit phase flip and bit simulation procedures and study interrelationship and
flip codes. First use the phase flip code to encode mutual impact between quantum estimation and quan-
states |0) and 1 1 ) in three qubits, with |0) encoded astum simulation. Since quantum computation is intrinsi-
|+ + +) and 1 1) as I cally random, and quantum simulation employs Monte
bit flip code to encode each of Carlo these techniques,
qubits, as with
we point|+)
out in Section 4.3 and
encoded as (|000) + |lll))/'/2Wang and |- it
(2011), ) isencoded
important to asprovide sound statis-
tical
(1 000) - 1 1 1 l))/-s/2. The resulted nine-qubit code has methods for analyzing quantum algorithms and
codeworks as follows: quantum simulation in general and study high-order
1000) + 1111) |000) + 1111) |000) + 1111) approximations to exponentiate Hamiltonians and the

1 ' V2 V2 V2 ' efficiency of the resulted quantum simulation proce-

dures in particular. On the other hand, quantum com-
1000) -|111) |000) -|111) |000) -|111) putation and quantum simulation have great potential
1 V2 V2 y/2 ' to revolutionize computational statistics. Below are a
few cases in point.
With the mixture of both phase flip and bit flip codes,
the Shor error-correction code 1can . The protect against
"random numbers" bit by classical com-
generated
flip errors, phase flip errors, as well puters asare a combined numbers
pseudo-random bit in the sense
and phase flip errors on any single qubit. In fact it has
that they are produced by deterministic procedures
been shown that this simple quantum and can beerror-correction
exactly repeated and perfectly predicted
code can protect against the effects givenof the any
deterministic
completelyschemes and the initial
arbitrary errors on a single qubitseeds. (Shor, 1995).
On the contrary, superposition states enable
quantum computers to produce genuine random
7. CONCLUDING REMARKS numbers. For example, measuring (|0) + |l))/'/2
yields 0 and 1 with equal probability. In general
Quantum information science gains enormous atten-
we generate ¿»-bit binary random numbers x =
tion in computer science, mathematics, physical sci-
x' ■ ■ ■ x i, , Xj = 0, 1 as follows. Apply b Hadamard
ences and engineering, and several interdisciplinary
gates to b qubits of |0 • • • 0) to obtain
subfields are developing under the umbrella of quan-
tum information. This paper reviews quantum compu- |0) + |1) |0) + |1) 1 v
tation and quantum information from a statistical per- ~2
spective. We introduce concepts like qubits, quantum
gates and quantum circuits in quantum computation
x =x' '"X
and discuss quantum entanglement, quantum paral- where the
lelism and quantum error-correction in quantum com- and then m
putation and quantum information. We present ma- bit binary r
jor quantum algorithms and show their advantages probability

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
392 Y. WANG

random numbers
the physical
are ge
tum the
computers quantum
are able
numbers isolated
and performin ort
It is at
exciting the
to same
design t
number sible
generatorso that
and s
computations
simulation. Perhaps w
Monte posing
Carlo requir
simulation
sical coupling
computers. of th
2. It is The coupling
interesting to inv
ence refers to the loss of coherence between the com-
tum computation and
computationalponents of a quantum system or quantum superposi-
statistics
computers tion from
may the interaction of the quantum
be system with
muc its
puters forenvironment. It is
computing very crucial but challenging to con-
Moreover, trolquantum
a quantum system of qubits and correct the effectscom
out some of decoherence
computation in quantum computation and quantum
prohibitive information.by Quantum computing
classicalhas witnessed great
ples are asadvances in
follows:recent years, and quantum computers of a
(a) W
of Grover's handful of qubits
search and basic quantum communication
algo
tum
devices have been built in research
algorithms for
laboratories (seeim
Barz et al., 2012; Clarke and Wilhelm, 2008; DiCarlo
procedures. For exampl
et al., 2009; Johnson et al., 2011; Lee et al., 2011;
huge data set is to searc
Neumann et al., 2008), but there are technological hur-
separates the top and bo
dles in the development of a quantum computer of large
quantum algorithms ca
capacity. History shows that scientific innovations and
for calculating median a
technological surprises are a never-ending saga. It is
genuine random numbe
anticipated that quantum computers with a few dozen
evaluation, quantum co
of qubits will be built in near future. As we have dis-
cant advantages over cla
cussed in Section 3.1, such a quantum computer has
Carlo integration. For e
capacity of a classical supercomputer. We are very op-
gration in high dimens
timistic that someday quantum computers will be avail-
or quadratically faster
able for statisticians to crunch numbers. For the time o
on classical computers,
being, instead of waiting in the sidelines for that to
quantum computers to
happen, statisticians should get into the field of play.
statistical It is
computing
time for us to dive into this frontier research and ta
putation discussed in Se
work with scientists and engineers to speed up the ar-
research along these lin
rival of practical quantum computers. As a last note,
and Wu (1999) and Hei
in 2011 a Canadian company called D-Wave has sold
3. As quantum computat
the claimed first commercial quantum computer of 128
are ideal for simulating
qubits to the Lockheed-Martin corporation, despite the
like the Ising
D- Wave's quantummodel,
system being criticized as a black it
interplay between
box. Large scale quantum computers mayquan be years
chain Monte Carlo
away, but quantum computing is already here meth
as a sci-
potential entific
to speed
endeavor to provoke deep thoughts and integrateup
rithms. In fact,
profound it
questions in physics has
and computer science. be
based algorithms can of
certain Markov ACKNOWLEDGMENTS chain b
al., 201 1 ; Richter, 2007
Wang's research was supported in part by NSF Grant
and Abeyesinghe, 2008)
DMS-10-05635. He thanks editor David Madigan and
Finally we point
two anonymous out
referees for helpful comments and t
wonderful suggestions which
but it led to is
significant improvements in
diffic
with present technology
both substance and the presentation of the paper.

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 QUANTUM COMPUTATION AND QUANTUM INFORMATION 393

REFERENCES Clauser, J. F., Horne, M. A., Shimony, A. and Holt, R. A.

(1969). Proposed experiment to test local hidden- variable theo-
Abrams, D. S. and Lloyd, S. (1997). Simulation of many-
ries. Phys. Rev. Lett. 23 880-884.
body Fermi systems on a quantum computer. Phys. Rev. Lett.
Cory, D. G., Mass, W., Price, M., Knill, E., Laflamme, R.,
79 2586-2589.
Zurek, W. H., Havel, T. F. and Somaroo, S. S. (1998). Ex-
Aharonov, D. and Ta-Shma, A. (2003). Adiabatic quantum
perimental quantum error correction. Phys. Rev. Lett. 81 2152-
state generation and statistical zero knowledge. In Proceedings2155.
of the Thirty-Fifth Annual ACM Symposium on Theory of Com-
Crandall, R. and Pomeranče, C. (2001). Prime Numbers :
puting 20-29. ACM, New York (electronic). MR2 121066
A Computational Perspective. Springer, New York. MR 1821 158
Artiles, L. M., Gill, R. D. and Guja, M. I. (2005). An in-
Deutsch, D. (1985). Quantum theory, the Church-Turing prin-
vitation to quantum tomography. J. R. Stat. Soc. Ser. B Stat.
ciple and the universal quantum computer. Proc. R. Soc. Lond.
Methodol. 67 109-134. MR2136642
Ser. A 400 97-1 17. MR0801665
Aspect, A., Grangier, P. and Roger, G. (1981). Experimental DiCarlo, L., Chow, J. M., Gambetta, J. M., Bishop, L. S.,
tests of realistic local theories via Bell's theorem. Phys. Rev.
Johnson, B. R., Schuster, D. I., Majer, J., Blais, A.,
Lett. 47 460-463.
Frunzio, L., Girvin, S. M. and Schoelkopf, R. J. (2009).
Aspect, A., Grangier, P. and Roger, G. (1982a). Experimen- Demonstration of two-qubit algorithms with a superconducting
tal realization of Einstein-Podolsky-Rosen-Bohm Gedanken- quantum processor. Nature 460 240-244.
experiment: A new violation of Bell's inequalities. Phys. Rev.Di Vincenzo, D. P. (1995). Quantum computation. Science 270
Lett. 49 91-94.
255-261. MR1355956
Aspect, A., Grangier, P. and Roger, G. (1982b). Experimen-Einstein, A., Podolsky, B. and Rosen, N. (1935). Can
tal test of Bell's inequalities using time- varying analyzers. Phys. quantum-mechanical description of physical reality be consid-
Rev. Lett. 49 1804-1807.
ered complete? Phys. Rev. 47 777-780.
Aspuru-Guzik, A. D., Dutoi, P. J. L. and Head-Gordon,
Feynman, R. P. (1981/82). Simulating physics with computers.
M. (2005). Simulated quantum computation of molecular ener- Internat. J. Theoret. Phys. 21 467-488. MR065831 1
gies. Science 309 1704. Freedman, M. H., Kitaev, A. and Wang, Z. (2002). Simula-
Barndorff-Nielsen, O. E., Gill, R. D. and Jupp, P. E. tion of topological field theories by quantum computers. Comm.
(2003). On quantum statistical inference. J. R. Stat. Soc. Ser. B
Math. Phys. 227 587-603. MR1910832
Stat. Methodol. 65 775-816. MR2017871
Griffiths, D. J. (2004). Introduction to Quantum Mechanics, 2nd
Barz, S., Kashefi, E., Broadbent, A., Fitzsimons, J. F., ed. Benjamin Cummings, San Francisco, CA.
Zeilinger, A. and Walther, P. (2012). Demonstration of Grover, L. K. (1996). A fast quantum mechanical algorithm for
blind quantum computing. Science 335 303-308. MR29 19052 database search. In Proceedings of the Twenty-Eighth Annual
Bell, J. (1964). On the Einstein Podolsky Rosen paradox. PhysicsACM Symposium on the Theory of Computing (. Philadelphia ,
1 195-200.
PA, 1996) 212-219. ACM, New York. MR1427516
Bennett, C. H., Cirac, J. I., Leifer, M. S., Leung, D. W., Grover, L. K. (1997). Quantum mechanics helps in searching for
Linden, N., Popescu, S. and Vidal, G. (2002). Optimal a needle in a haystack. Phys. Rev. Lett. 79 325-328.
simulation of two-qubit Hamiltonians using general local op- Hayashi, M. (2006). Quantum Information : An Introduction.
erations. Phys. Rev. A (3) 66 012305. MR1929513 Springer, Berlin. Translated from the 2003 Japanese original.
Berry, D. W., Ahokas, G., Cleve, R. and Sanders, B. C. MR2228302
(2007). Efficient quantum algorithms for simulating sparse
Heinrich, S. (2003). From Monte Carlo to quantum computation.
Hamiltonians. Comm. Math. Phys. 270 359-371. MR2276450 Math. Comput. Simulation 62 219-230. MR 1988372
Boghosian, B. M. and Taylor, W. IV (1998). SimulatingHOLEVO, A. S. (1982). Probabilistic and Statistical Aspects of
quantum mechanics on a quantum computer. Phys. D 120 30- Quantum Theory. North-Holland Series in Statistics and Prob-
42. MR1 679863
ability 1. North-Holland, Amsterdam. Translated from the Rus-
Böhm, d. (1951). Quantum Theory. Prentice-Hall, Englewood sian by the author. MR0681693
Cliffs, NJ. Holevo, A. S. (1998). The capacity of the quantum channel with
Butucea, C., Gujä, M. and Artiles, L. (2007). Minimax and general signal states. IEEE Trans. Inform. Theory 44 269-273.
adaptive estimation of the Wigner function in quantum homo- MR 1486663
dyne tomography with noisy data. Ann. Statist. 35 465-494.
Huber, P. J. and Ronchetti, E. M. (2009). Robust Statistics,
MR2336856 2nd ed. Wiley, Hoboken, NJ. MR2488795
Calderbank, A. R. and Shor, P. W. (1996). Good quantum
Jané, E., Vidal, G., Dür, W., Zoller, P. and Cirac, J. I.
error-correcting codes exist. Phys. Rev. A 54 1098-1 105. (2003). Simulation of quantum dynamics with quantum optical
Childs, A. M. (2010). On the relationship between continuous-
systems. Quantum Inf. Comput. 3 15-37. MR1965173
Johnson, M. W., Amin, M. H. S., Gildert, S., Lanting, T.,
and discrete-time quantum walk. Comm. Math. Phys. 294 581-
603. MR2579466 Hamze, F., Dickson, N., Harris, R., Berkley, A. J., Jo-
Childs, A. M., Cleve, R., Deotto, E., Farhi, E., Gut- hansson, J., Bunyk, P., Chapple, E. M., Enderud, C.,
mann, S. and Spielman, D. A. (2003). Exponential algorith-
Hilton, J. P., Karimi, K., Ladizinsky, E., Ladizin-
mic speedup by quantum walk. In Proc. 35 th ACM Symposiumsky, N., Oh, T., Perminov, I., Rich, C., Thom, M.
on Theory of Computing 59-68. ACM Press, New York. C., Tolkacheva, E., Truncik, C. J. S., Uchaikin, S.,
Wang, J., Wilson, B., Rose, G. et al. (2011). Quantum an-
Clarke, J. and Wilhelm, F. K. (2008). Superconducting quan-
tum bits. Nature 453 1031-1042. nealing with manufactured spins. Nature 473 194-198.

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms
 394 Y. WANG

Kato, T. Sayrin, C., Trotter's

(1978). Dotsenko, I., Zhou, X., Peaudecerf, B., Ry-prod
of BARCZYK,
self-adjoint T., GLEYZES, S., ROUCHON, P., MlRRAHIMI, M.,
contraction sem
Analysis ( Amini, H., Brune, M., Raimond,
Essays Dedicated J. M. and Haroche, S. t
of His 10th (2011). Real-time quantum feedback prepares and
Birthday). stabilizes
Adv. i
Academic Press, New
photon number states. Nature 477 10376. York. M
Kiefer, C. Schumacher, B. (1995).
(2004). On Quantumthecoding. Phys. Rev. A (3) 51
interp
Copenhagen to2738-2747.
the MR1328824 present day
tion Schumacher,
291-299. B. and Westmoreland, M. D. (1997).
Springer, Sending
Berli
Lee, N., Benichi, H., Takeno, Y., Takeda, S., Webb, J., classical information via noisy quantum channels. Phys. Rev. A
Huntington, E. and Furusawa, A. (2011). Teleportation 56 131138.

of nonclassical wave packets of light. Science 332 330-333. Shankar, R. (1994). Principles of Quantum Mechanics , 2nd ed.
Lloyd, S. (1996). Universal quantum simulators. Science 273 Plenum Press, New York. MR1 343488
1073-1078. MR 1407944 Shenvi, N., Kempe, J. and Whaley, K. B. (2003). Quantum
Magniez, F., Nayak, A., Roland, J. and Santha, M. (2011). random- walk search algorithm. Phys. Rev. A 67 052307.
Search via quantum walk. SIAM J. Comput. 40 142-164. Shor, P. W. (1994). Algorithms for quantum computation: Dis-
MR2783206 crete logarithms and factoring. In 35 th Annual Symposium on
Mariantoni, M., Wang, H., Yamamoto, T., Neeley, M., Foundations of Computer Science ( Santa Fe , NM, 1994) 124 -
BialczakI, R. C., Chen, Y., Lenander, M., Lucero, E., 134. IEEE Comput. Soc. Press, Los Alamitos, CA. MR1489242
O'Connell, A. D., Sank, D., Weides, M., Wenner, J., Shor, P. W. (1995). Scheme for reducing decoherence in quantum
Yin, Y, Zhao, J., Korotkov, A. N., Cleland, A. N. computer memory. Phys. Rev. A 52 2493-2496.
and Martinis, J. M. (2011). Implementing the quantum Shor,
von P. W. (1997). Polynomial-time algorithms for prime factor-
Neumann architecture with superconducting circuits. Science 7ization and discrete logarithms on a quantum computer. SIAM
1208517. J. Comput. 26 1484-1509. MR1471990
Menezes, A., van Oorschot, P. C. and Vanstone, S. A. Steane, A.M. (1996a). Error correcting codes in quantum theory.
(1996). Handbook of Applied Cryptography. CRC Press, NewPhys. Rev. Lett. 11 793-797. MR1398854
York. Steane, A. (1996b). Multiple-particle interference and quantum
error correction. Proc. R. Soc. Lond. Ser. A 452 2551-2577.
Nayak, A. and Wu, F. (1999). The quantum query complexity
MR 142 1749
of approximating the median and related statistics. In Annual
Szegedy, M. (2004). Quantum speed-up of Markov chain based
ACM Symposium on Theory of Computing ( Atlanta , GA, 1999)
384-393. ACM, New York (electronic). MR1798059 algorithms. In Proc. 45th IEEE Symposium on Foundations of
Neumann, P., Mizuochi, N., Rempp, F., Hemmer, P., Computer Science 32-41. IEEE Computer Society Press, Los
Watanabe, H., Yamasaki, S., Jacques, V., Gaebel, T., Alamitos, CA.
Jelezko, F. and Wrachtrup, J. (2008). Multipartite entan-Trotter, H. F. (1959). On the product of semi-groups of opera-
glement among single spins in diamond. Science 320 1326-tors. Proc. Amer. Math. Soc. 10 545-551. MR0108732
1329. TSIRELSON, B. S. (1980). Quantum generalizations of Bell's in-
Nielsen, M. A. and Chuang, I. L. (2000). Quantum Computa- equality. Lett. Math. Phys. 4 93-100. MR0577178
tion and Quantum Information. Cambridge Univ. Press, Cam-Tulsi, A. (2008). Faster quantum- walk algorithm for the two-
bridge. MR 1796805 dimensional spatial search. Phys. Rev. A 78 012310.
Nightingale, M. P. and Umrigar, C. J., eds. (1999). Quan- VlDAKOVlC, B. (1999). Statistical Modeling by Wavelets. Wiley,
tum Monte Carlo Methods in Physics and Chemistry. NATO New York. MR1681904

Science Series C: Mathematical and Physical Sciences 525. VON Neumann, J. (1955). Mathematical Foundations of Quan-
Kluwer Academic, Dordrecht. MR 17 12250 tum Mechanics. Princeton Univ. Press, Princeton, NJ. Trans-
NUSSBAUM, M. and Szkola, A. (2009). The Chernoff lower lated by Robert T. Beyer. MR0066944
bound for symmetric quantum hypothesis testing. Ann. Statist. Wang, Y. Z. (1994). Quantum Gaussian processes. Acta Math.
37 1040-1057. MR2502660 Appl. Sin. {Engl. Ser.) 10 315-327. MR1310171
PARTHASARATHY, K. R. (1992). An Introduction to Quan- Wang, Y. (2011). Quantum Monte Carlo simulation. Ann. Appl.
tum Stochastic Calculus. Monographs in Mathematics 85. Stat. 5 669-683. MR2840170
Birkhäuser, Basel. MR 1164866 WOCJAN, P. and Abeyesinghe, A. (2008). Speed-up via quan-
Richter, P. C. (2007). Quantum speedup of classical mixing pro- tum sampling. arXiv:0804.4259v3[quant-ph].
cesses. Phys. Rev. A 76 042306. Wootters, W. K. and Zurek, W. H. (1982). A single quantum
Rivest, R. L., Shamir, A. and Adleman, L. (1978). A method cannot be cloned. Nature 299 802-803.
for obtaining digital signatures and public-key cryptosystems. Yu, A. B. (2001). Campbell-Hausdorff formula. In Encyclopae-
Comm. ACM 21 120-126. MR0700103 dia of Mathematics (M. Hazewinkel, ed.). Kluwer Academic,
Rousseau, V. G. (2008). Stochastic Green function algorithm. Dordrecht.

Phys. Rev. E (3) 77 056705. MR2495497 Zalka, C. (1998). Simulating a quantum systems on a quantum
Sakurai, J. J. and Napolitano, J. (2010). Modern Quantum computer. R. Soc. Lond. Philos. Trans. Ser. A Math. Phys. Eng.
Mechanics , 2nd ed. Addison- Wesley, Reading, MA. Sci. 454 313-322.

This content downloaded from

92.29.138.46 on Wed, 07 Jun 2023 20:16:59 +00:00
All use subject to https://about.jstor.org/terms