Professional Documents
Culture Documents
p1937 Ralston
p1937 Ralston
p1937 Ralston
b 1 aeb
The efficiency with which multi-qubit superpositions
couldbe created in a QC also extends to the com-
putation of functions. If we consider calculating a func-
tion that takes values on arguments from 0 to 2’ - we
could create a single argument value in the n-qubit
’,
register and then create thefunction’svalue ina second
register with appropriate quantum logic gates. But
using instead a superposition of all 2n values in the argu-
Controlled-controlled-NOT
a a
ment register, a single application of the same quantum
logic gates would create all 2”’ values of the function
in superposition in the second register. (In contrast, on
any conventional computer, even a parallel one, the
values would haveto be computed individually.) This
“quantum parallelism” apparently allows an exponen-
tial amount of computational work ina single quantum
operation, but if we are interested in the function
values themselves, this feature of a QC is not directly
useful because only a single value of the function is
Figure 1. A set of three universal reversible Boolean logic gates, obtained on measurement. However, for certain prob-
for which the input bit values (a, b, c) can be
recovered from the output values by a second lems where we are interested in some common prop-
application of the same logic gate. All of the arithmetic erty shared by all the values of the function, such as its
functions for conventional computation can be period, the power of quantum parallelism can be har-
constructed from this set of gates. The curved arrows
represent the qubit in one of its two possible states.
nessed using another nonclassical feature of quantum
physics: quantum interference. Just as two waves can
interfere constructively, when in-phase, to produce a
of a QC would arise from the ability to perform much stronger wave or destructively, when out of phase, to
more general quantum gate operations that are not cancel each other out, thesuperposition states within a
limited to theconventional Boolean operations of ordi- QC can be manipulated to produce constructive inter-
nary computers. ference at correct answers and destructive interference
at incorrect ones. The Fourier transformation is the
It is one of the most counterintuitive features of quan-
natural mathematical device for such manipulations,
tum physics that, at theatomic scale, material systems
and in 1994Shor showed how non-Boolean operations
can behave with wavelike properties and exhibit the
could be usedto perform a quantumFourier transform
phenomenon of superposition, so thata qubit in a
(QFT) operation, with exponentially fewer operations
superposition state has aspects of both 0 and 1 simul-
thanthe classicalfast Fourier transform. His QFT
taneously. Of course, once measured, aqubit in a super-
could be usedto find the period of a function efficiently
position state will always be found to have a definite
on a QC, and he related the problem of factoring an
0 or 1 value, with only the probability of each result
integer to a periodicity problem, showing that factor-
being predictable by quantum mechanics, and deter-
ing could be solved efficientlyon a quantumcomputer.
mined by the amount of each component in the super-
position. For example, quantum physics predicts that To put Shor’s algorithm in perspective, during 1994 a
a horizontally propagating photon prepared with its 129-digit number known as RSA- 129 was factored in
polarization oriented at 45” to the vertical will pass 8 months, requiring lo” instructions performed on
through avertically oriented polarizer with 50% proba- over 1,000 networked computers (see COOPERATIVE
bility and fail with 50% probability. (Superpositions are COMPUTING). But a QC with a clock speed of 100 MHz
also the origin of the intrinsically probabilistic nature of could have factored this number in a few seconds,
quantum computation.) In 1986 Deutsch realizedthat requiring a memory of about 2,000 qubits and IO9
a QC could exploit this distinctly nonclassicalphenom- quantum logic gates. Furthermore,the number of
QUANTUM COMPUTING 1495
quantum gates required to factor even larger numbers coupled together through an external drive in order to
grows as only a polynomial function of the number of perform logic-gate operations. Thirdly, there must be a
digits in the number being factored: a 250-digit number read-out method to determine the state of each qubit at
would only require about 10 times as many quantum the end of the computation. (In the quantum world
logic gates. In contrast, conventional factoring algo- measurement can be problematic.) At present, two
rithms have a muchfaster growth: a 250-digit number hardware schemes have been used to demonstrate
would require a factoring time of about lo7 years by the basic operations of a QC with existing technology:
extrapolation from the RSA-129 result. The power of the manipulation of nuclear spin qubits using nuclear
quantum parallelism becomes even clearer if we con- magnetic resonance (NMR) methods; and photon
sider trying to simulate Shor’s algorithm on a conven- polarization qubits manipulated with conventional
tional computer: for factoring RSA-129 wewould linear optical elements. With these systemsseveral
require 22,000memory bits. logic operations with up to three qubits have been
demonstrated, and even very simpleinstances of quan-
At present there are only a few algorithms known in tum algorithms, such as searching a four-element data-
which quantum parallelism could provide an advan- base using Grover’s algorithm have been performed.
tageover conventional computation. Shor has also
invented a quantumalgorithm to solve the discrete log- Although these systems will continue to be very useful
arithm problem: given a prime number, p , and integer for exploring the basics of quantum computation
g < p and another integer y < p , find the integer x , during the next few years, neither of them scales up
such that gx = ymodp. Like factoring, this problem is easily to more thanabout 10 qubits. To go beyond this
computationally intractable on conventional comput- limit will require the use of other technologies now
ers and forms the basis for another class of public-key under development, such as laser manipulation of elec-
cryptosystems. The quantum factoring and discrete tronic energylevelsin trapped ions and controlled
logarithm algorithms are particular instances of one quantum states of the electromagnetic field in micro-
class of quantum algorithms. Another classis based on cavities. Single quantum logic operations have been
Grover’s quantum algorithm that finds a “marked” demonstrated in both of these technologies. Estimates
element in an unstructured database more efficiently suggest that the trapped ion scheme should be par-
than aclassical computer. One of the most activeareas ticularly promising, with a computational potential of
of research in quantum computation is to understand 30 or more qubits and many thousands of logic opera-
the general nature of problems that are amenable to tions. Ion trap QCs should allow quantum computation
efficient solution on a QC. to be developed into the realm where simulation of
quantum computation on conventional computers is
A serious obstacle to practical quantum computation impossible. Even so, the computational capacity of an
is the propensity for qubit superpositions of 0 and 1 to ion trap QC would allow the factoring of only small
“decohere” into either 0 or 1. (This phenomenon of numbers. The long-term future of QC is therefore likely
decoherence is invoked to explain why macroscopic to involve solid-state schemes, although these are cur-
objects are never observed in quantum superposition rently at a much earlier stage of development than the
states.) However, over the past two years there have ion traps. A particularly promising silicon-based QC
been breakthroughs in generalizingconventional error concept, which might allow large-scale quantum com-
correction concepts to correct decoherence in a QC. putation and industrial fabrication techniques, was
A single logical bit would be encoded as the state of proposed in 1998. This scheme is now under active
several qubits and quantum gate operations used to development. However, the extrapolation to computa-
correct decoherence errors through a “majority vote.” tionally significant problems is so great that it will be
Moreover, these ideas have been shown to allow robust several years before the practical potential of quantum
or fault-tolerant quantum computation with the en- computation can be reliably assessed. Quantum com-
coded logical bits. It is now known that if a certain putation is not only a new computational paradigm, but
threshold precision per gate operation can be achieved, also opens up new perspectives on quantum physics.
quantum error correction would allow a QC to com- The future promises to be exciting forboth fields.
pute indefinitely.
1996. Ekert, A. K., and Josza, R. “Quantum Computation and Queueing theory was born in the early 1900s with the
Shor’s Factoring Algorithm,”Reviews of Modern Physics, 68, work of A. K. Erlang of the Copenhagen Telephone
733-753.
1996. Feynman,R. P. “The Feynman Lectures on Computation” Company, who derived severalimportant formulas for
(eds. A. J. G. Hey and R. W. Allen). Reading, MA: teletraffic engineering that still bear his name today.
Addison-Wesley. The range of applications has grown to include not only
1996. Grover,L. K. “A Fast Quantum Mechanical Algorithm for telecommunications and computer science, but also
Database Search,” in Proceedings of the 28th Annual ACM
Symposium on the Theory of Computing. New York: ACM. manufacturing, air traffic control, militarylogistics,
1998. Gershenfeld, N., and Chuang, I. L. “Quantum Computing design of themeparks, and many other areas that
with Molecules,” Scientific American, June, 66. involve service systems whose demands are random.
1999. Hey, A. J. G. (ed.) Feynman and Computation, Reading, Queueing theory is considered to be one of the stan-
MA: Perseus.
dard methodologies (together with linear program-
Richard J. Hughes ming, simulation, etc.) of operations research and
management science, and is standard fare in academic
programs in industrial engineering, manufacturing
QUEUEING THEORY engineering, etc., as well as in programs in telecom-
munications, computer engineering, and computer
science. There are dozens of books and thousands of
papers on queueing, and they continue to be published
at an ever-increasing rate. But, despite its apparent
simplicity (customers arrive, request service, and leave
A queue is a waiting line (like customers waiting at a
or wait until they getit), thesubject is one of depth and
supermarket checkout counter); queueingtheory is
subtlety. We w iillustrate this by briefly visiting some
l
the mathematical theory of waiting lines. More gener-
of the most important models, and describing along
ally, queueing theory is concerned with the mathemat-
the way some of the obvious features and some of the
icalmodeling and analysis of systems that provide
subtleties.
service to random demands. A queueing model is an
abstract description of such a system. Typically, a
queueing model represents (1) the system’s physical
Queueing Models
configuration, by specifying the number and arrange-
ment of the servers, which provide service to the cus- The essence of queueing theory is that it takes into
tomers, and (2) the stochastic (that is, probabilistic or account the randomness of the arrival process and the
statistical) nature of the demands, by specifying the randomness of the service process. The most common
variability in the arrival process and in the service assumption about the arrival process is that the
process. customer arrivals follow a Poisson process. One way
to describe a Poisson arrival process is to imagine that
For example, in the context of computer communica-
time is divided into small intervals of length AT.
tions, a communications channel might be a server,
Assume that in each interval either an arrival occurs
and the messages the customers; the (random) times at
(with probability X . AT, say, where theproportionality
which messages request the use of the channel would
constant X is the arrival rate) or it doesn’t, indepen-
be the arrival process, and the (random) lengths of
dently of the occurrence or non-occurrence of arrivals
service time that the messages hold the channel while
in other intervals. Finally, imagine that AT + 0 (that is,
being transmitted would constitute the service process.
take limits to pass from discrete time to continuous
Another example is a computer system where a pro-
time). Then the arrivals are said to follow a Poisson
grammer (customer) sitting at a terminal requests
process; and one of the properties of the Poisson
access to a CPU (server) for the processing of a trans-
process is that the times between arrivals (the inter-
action; both the arrival time of the request for access
arrival time)are exponentially distributed. (A random
and the amount of processing time requested are ran-
variable X is said to be exponentially distributed if its
dom. Then, the mathematical analysis of the models
distribution function F,(t) is given by F,(t) = 1 -
would yield formulas that presumably relate the phys-
for all t 2 0, where 1/X is the average value of X ) .
ical and stochastic parameters to certain performance
measures, such as average waiting time, server utili- One of the most important queueing models is the
zation, throughput, probability of buffer overflow, etc. Erlang loss model; it assumes that the arrivals follow a
The art of applied queueing theory is to construct a Poisson process, and that the blocked customers (those
model that is simple enough so that it yields to mathe- who find all servers busy) are cleaved (that is, they are
matical analysis, yet contains sufficient detail so that denied entry into the system, so the blocked customers
its performance measures reflect the behavior of the are lost) (Fig. 1). The fraction of arriving customers who
real system. find all the servers busy (the probability of blocking, or