Deutsch-Jozsa algorithm as a test of quantum computation

David Collins, K. W. Kim and W. C. Holton

Department of Electrical and Computer Engineering, Box 7911, 232 Daniels Hall, North Carolina
State University, Raleigh, North Carolina 27695-7911
(September 7, 2018)

Abstract
arXiv:quant-ph/9807012v1 3 Jul 1998

A redundancy in the existing Deutsch-Jozsa quantum algorithm is removed

and a refined algorithm, which reduces the size of the register and simplifies
the function evaluation, is proposed. The refined version allows a simpler
analysis of the use of entanglement between the qubits in the algorithm and
provides criteria for deciding when the Deutsch-Jozsa algorithm constitutes a
meaningful test of quantum computation.
03.67.Lx, 03.65.-w

Typeset using REVTEX

1
 Quantum computation has emerged in the past decade as a potentially very powerful
way to solve certain problems. The idea is to store information in states of a quantum
system, manipulate these via unitary transformations and extract useful information from
the resulting state. The use of key features of quantum mechanics such as superposition of
states and quantum entanglement enables exponential speedups in the solution of certain
problems. Within this framework computational schemes, algorithms, and error correction
have been developed [1,2]. However, practical implementation, which requires precise control
of quantum systems, remains beset by difficulties [3]. Despite skepticism about the feasibility
of quantum computation several groups claim to have demonstrated experimentally the
operation of quantum gates [4], error correction [5] and simple algorithms [6–8].
To date most experimental work on quantum algorithms has been directed toward imple-
mentations of the Deutsch-Jozsa algorithm, which provides a fertile ground for illustrating
the key features of quantum computation [9–13]. Our purpose is to investigate the Deutsch-
Jozsa algorithm to determine when it is a meaningful test of quantum computation. We
shall focus on the algorithm’s use of entanglement, which gives quantum computation its
power [13].
The Deutsch problem [9,12] considers certain global properties of functions on N-bit bi-
nary numbers. Denote the set of all such numbers by XN := {xN xN −1 . . . x1 x0 | xm = 0, 1}.
A function f : XN → {0, 1} is called balanced if the number of times it returns 0 is equal to
the number of times it returns 1 as the argument ranges over XN . The Deutsch problem is
to take a given function which is known to be either balanced or constant and to determine
which type it is. A classical algorithm which would answer this with certainty would require
that f be evaluated for 2N −1 + 1 values of its argument and thus grows exponentially with
input size. The Deutsch problem may be solved by a quantum algorithm [6,7,9,12] which
only requires one “evaluation” of f . This requires: (1) A well defined physical system,
called the total register, to be used for storing and retrieving information. (2) A sequence
of unitary transformations to be enacted on the total register in such a way as to produce
an answer to the problem.
The total register in the existing algorithm consists of an N qubit control register, which
is generally used for storing the function arguments, plus a one qubit function register, which
is used for function evaluation. The Hilbert space for the control register will be denoted
Hc and that for the function register Hf . Thus the total register in the existing algorithm
is

Htot = Hc ⊗ Hf . (1)

Basis elements for the control register will be denoted

|xic ≡ |xN −1 . . . x0 ic ≡ |xN −1 ic . . . |x0 ic (2)

where {|xm ic | xm = 0, 1} are orthonormal basis sets for distinguishable two-state systems.
Similarly basis elements for the function register are denoted |yif where y = 0, 1. Note
that equation (2) provides a one-to-one correspondence between XN (function arguments)
and basis elements of the control register. Thus any physical implementation of the existing
algorithm, such as those accomplished using an NMR quantum computer [6,7], requires a
two-state system for the function register in addition to those for the control register.

2
 The existing state-of-the-art quantum algorithm (see the articles by Cleve et. al. [12]
and Jones and Mosca [7]) for the Deutsch problem [6,7,12] makes use of the total register
given in equation (1), and follows the scheme illustrated in figure 1. The two gates used are:
1. A Hadamard transformation applied to each qubit of the control register,
Ĥtot := Ĥ ⊗ . . . ⊗ Ĥ, where
!
1 1 1
Ĥ := √ . (3)
2 1 −1

This transforms the input state |0 . . . 0ic into an equal superposition over all basis
elements given in equation (2), thereby preparing the way for a simultaneous evaluation
of f over all possible arguments.

2. The f -controlled-NOT gate, Ûf −c−N , which performs the function evaluation, whose
operation on an orthonormal basis of Htot is defined as

Ûf −c−N |xic |yif := |xic |y ⊕ f (x)if (4)

and is extended linearly to all elements of Htot .

The final step is evaluation of the expectation value of |0 . . . 0ic h0 . . . 0|c on the control
register. If the result is 0 then f is balanced and if it is 1 then f is constant. It is easily
shown that this algorithm exhibits polynomial growth in input size.
Clearly the only point at which entanglement between any of the qubits of the total
register could occur is during the function
h evaluation.
i However, it is also apparent that the
function register is in the state √12 |0if − |1if before and after this step. In fact it is easily
shown that
1 h i 1 h i
Ûf −c−N |xic √ |0if − |1if = (−1)f (x) |xic √ |0if − |1if (5)
2 2
which implies
h that if the
i function register of the input state is restricted to the subspace
spanned by |0if − |1if then there is no entanglement between the control and the function
registers in the output of the Ûf −c−N . Thus there is no need for any coupling between the
two-state systems comprising the control register and that comprising the function register.
In the existing algorithm the function register is completely redundant.
This suggests that the algorithm should be modified by eliminating the function register
while retaining the control register of the previous algorithm. Thus the total register is Hc .
Equation (5) suggests that the function evaluation can be carried out via the f -controlled
gate whose operation on the basis elements of the control register is defined as

Ûf |xic := (−1)f (x) |xic (6)

and which is extended linearly to all elements of Hc . Indeed it is easily seen that in the
existing version of the algorithm the effect of Ûf −c−N is identical to that of Ûf ⊗ Iˆf where
Iˆf is the identity on Hf . Note that this is invalid for the most general element of Htot in
the existing algorithm but is true whenever the input state of the function register is in the

3
 h i
subspace spanned by |0if − |1if . Clearly Ûf satisfies the requirement that a gate must
be a unitary operator. The refined algorithm follows a similar pattern of operations as the
existing algorithm. A schematic form is provided in figure 2. The refined algorithm requires
one qubit fewer than the existing algorithm. Consequently physical implementation requires
one fewer two-state system. Thus the experimental demonstrations using NMR systems [6,7]
could have been carried out using a single spin one-half system instead of the two coupled
spin one-half systems that were actually used.
Typically quantum computers achieve their efficiency by utilizing entanglement between
the various qubits of the total register [13]. The above discussion shows that for both the ex-
isting and refined Deutsch-Jozsa algorithms the only possibility for entanglement is amongst
the qubits of the control register through the action of the f -controlled gate. Furthermore,
whether or not entanglement occurs depends on the form of f . For example, equation (6)
shows that if f is constant then Ûf = ±Iˆc and it does not cause any entanglement between
the qubits of the control register. The need for entanglement for any balanced function
requires a straightforward but tedious analysis of Ûf ; for some balanced functions it causes
entanglement and for others not. We shall demand that for a given value of N the quan-
tum computer (i. e. physical system) must be capable of solving the Deutsch problem for
all admissible functions. We thus investigate whether a given qubit of the control regis-
ter remains unentangled (after operation of Ûf ) from the remaining qubits for all possible
balanced functions. If so it can remain uncoupled from the rest of the register and can be
implemented on a completely isolated quantum system.
For N = 1 entanglement is clearly not an issue. For N = 2 it can be shown that (from
here onwards the arguments of the f are expressed in decimal form)
Ûf = Û1 ⊗ Û0 (7)

where Ûm operates on the |xm ic qubit and, with respect to the basis {|0ic , |1ic },
!
1 0
Û1 = f (0)+f (2) (8)
0 (−1)
and
!
f (0) 1 0
Û0 = (−1) f (0)+f (1) (9)
0 (−1)
where we have used the fact that for any balanced function
(−1)f (3) = (−1)f (0)+f (1)+f (2) . (10)
Note that Û1 and Û0 are unitary. Thus for N = 2 the f -controlled gate does not cause
entanglement between the qubits of the control register.
Now consider N > 2 and assume that the |x0 ic qubit is not entangled by the f -controlled
gate. Thus
Ûf = Û ′ ⊗ Û0 (11)

where Û ′ operates on the qubits |xN −1 . . . x1 ic and Û0 operates on |x0 ic . It is easily shown
that, with respect to the basis {|0ic , |1ic },

4
  
a1 0 . . . 0 0
 0 a2 . . . 0 0 
 

Û ′ =  ... .. ..
 
. . (12)
 

 
 0 0 . . . a N−1 0 
 (2 −1) 
0 0 ... 0 a(2N−1 )

where am 6= 0 and
!
b1 0
Û0 = (13)
0 b2

where bm 6= 0. This requires relationships of the form (expressing the arguments of the
function in decimal form):

(−1)f (0) = a1 b1
(−1)f (1) = a1 b2
(−1)f (2) = a2 b1
(−1)f (3) = a2 b2 (14)

etc., which implies (−1)f (3) = (−1)f (0)+f (1)+f (2) etc.. For N > 2 there clearly exist balanced
functions for which this is invalid and for these the f -controlled gate entangles the |x0 ic
qubit with the rest of the control register. In this fashion it can be shown that for N > 2
none of the control register qubits are always unentangled with the others (for all possible
balanced functions).
Physically this implies that for N = 1 or 2 the Deutsch-Jozsa algorithm may be carried
out by using uncoupled two-state systems. However, for N > 2 there is always some balanced
function which requires that a given qubit of the control register be coupled to the remaining
qubits. Thus meaningful tests of the Deutsch-Jozsa algorithm occur if and only if N > 2.
Note that the previous NMR demonstrations were conducted for the case where N = 1 [6,7].
This leaves the question of the utility of quantum computation for the cases in which N ≤
2, where the quantum algorithm still appears to answer the problem with just one function
“evaluation” as opposed to 2N −1 +1 as required by the classical algorithm. It has usually been
N −1
assumed that the best classical solution is to inspect and compare  the first 2 +1 elements
N
in the following representation of f : f (0), f (1), . . . , f (2 − 1) . However, the problem may
be solved by checking the parity
 (even or odd) of the first 2N −1 elements in the following 
alternative representation: f (0) + (f 1), f (0) + f (2), . . . , f (0) + f (2N − 1), f (0) − f (1) . If
they are all even then f is constant and if any is odd f is balanced. Thus for N = 1
the problem is answered by checking the parity of f (0) + f (1) which requires only one
“evaluation”. For N = 2 the problem is answered by checking the parities of f (0) + f (1) and
f (0) + f (2). Inspection of equations (7)-(9) shows that the quantum algorithm decides these
using the |x0 i qubit (for f (0)+f (1)) and the |x1 i qubit (for f (0)+f (2)) independently. Thus
the “one function evaluation” is essentially two simultaneous evaluations on independent one-
qubit computers, each of which carries out part of the classical algorithm. The apparent
gain made by the quantum computer has occured only because the number of computers
has been doubled; the method of solution is essentially classical and the problem can be

5
solved just as easily with two classical computers. Thus for N ≤ 2 the quantum algorithm
solves the Deutsch problem in a classical way.
To conclude we have shown that it is possible to simplify the existing quantum algorithm
for the Deutsch problem by eliminating the function register and redefining the function
evaluation in terms of the f -controlled gate. We also showed that in the existing algorithm
there is no entanglement between the control and function registers. In the refined quantum
algorithm which we have presented, entanglement occurs between the qubits of the register
only when N > 2. Thus in order to utilize the full power of quantum computation as applied
to the Deutsch problem, an implementation for N > 2 is necessary.
This work was supported, in part, by the Defense Advanced Research Project Agency
and the Office of Naval Research. We would also like to thank Gary Sanders for his help in
preparing this article.

6
 REFERENCES
[1] A. Ekert and R. Jozsa, Rev. Mod. Phys. 68, 733 (1996).
[2] C. Bennett, Phys. Today, Oct 1995, 24 (1995).
[3] J. Preskill, Proc. R. Soc. London, Ser. A 454, 469 (1998).
[4] C. Monroe, D.M. Meekhof, B. E. King, W. M. Itano and D. J. Wineland, Phys. Rev.
Lett. 75, 4714 (1995)
[5] D. G. Cory, W. Mass, M. Price, E. Knill, R. Laflamme, W. H. Zurek, T. F. Havel and
S. S. Somaroo, Report No. quant-ph/9802018.
[6] I. L. Chuang, L. M. K. Vandersypen, X. Zhou, D. W. Leung and S. Lloyd, Nature 393,
143 (1998).
[7] J. A. Jones and M. Mosca, Report No. quant-ph/9801027.
[8] I. L. Chuang, N. Gershenfeld and M. Kubinec, Phys. Rev. Lett. 80, 3408 (1998)
[9] D. Deutsch and R. Jozsa, Proc. R. Soc. London, Ser. A, 439, 553 (1992).
[10] I. L. Chuang and Y. Yamamoto, Phys. Rev. A 52, 3489 (1995)
[11] V. Vedral and M B. Plenio, Report No. quant-ph/9802065.
[12] R. Cleve, A. Ekert, C. Macciavello and M.Mosca, Proc. R. Soc. London, Ser. A 454,
339 (1998).
[13] R.Jozsa, Report No. quant-ph/9707034.

7
 FIGURES
FIG. 1. The existing Deutsch-Jozsa algorithm. Ĥtot represents a Hadamard transformation ap-
plied to each qubit of the control register. Ûf −c−N represents the operation of the f -controlled-NOT
gate. Note that here x.y := xN −1 yN −1 + . . . x0 y0 . Summation is over all elements of XN .

FIG. 2. The refined Deutsch-Jozsa algorithm. Notation is the same as in figure 1 with the
f -controlled gate Ûf replacing the f -controlled-NOT gate.

8
Input Output
f(x) + (x.y)
|0...0>c Σ (-1) |y> c
^ ^ x,y
H tot H tot

^
U f-c-N
|0> f - |1> f
|0> f - |1> f

Figure 1
Input Output
f(x) + (x.y)
|0...0>c Σ (-1) |y> c
^ ^ ^ x,y
H tot Uf H tot

Figure 2