Robustness of the Quantum Search

Algorithm
B. Pablo-Norman and M. Ruiz-Altaba

Abstract. We find exact results for Grover’s quantum search algorithm and
analyze its behavior under noisy situations when no quantum correction codes
are available. We compute how the algorithm slows down: it is still better than
a classical one, provided the noise is smaller than some bound, which we also
compute.

I INTRODUCTION

High reaction rates at hadronic colliders call for new ideas in triggering.
Neural networks have been implemented in this setting with some success.
We wish to explore the possibility of using quantum algorithms (presumably
in classical computers) for this purpose. One crucial drawback of quantum al-
gorithms implemented in quantum computers is their extreme dependence on
the exact complex phase between various states. Quantum correcting codes
can deal with simple situations, preventing to a large extent the loss of quan-
tum coherence. Nevertheless, in an implementation on a classical computer,
one would have to worry about the finite precision of these machines.
We thus analyze in this paper how a random gaussian noise, added to the
output at each step of the algorithm, afects the recently proposed quantum
search algorithm, i.e. a quantum procedure for finding a number in a phone
book.

II GROVER’S QUANTUM SEARCH ALGORITHM

Classically, the only way to find a number in a random phone book (one
not ordered alphabetically) is to search it entry by entry, checking each time
whether it is the searched one. If it is, the search is over. If it is not, then con-
tinue. Thus, any classical algorithm (whether deterministic or probabilistic)
will find the wanted number after N/2 steps, on the average. But this is only
true when a classical algorithm√is used.
 Recently, Grover found a quantum
algorithm that requires only O N steps [1–4]. How does it work?
 Suppose there are N = 2n entries in the phone book. Each of them can be
represented by a quantum mechanical state of n spin-1/2 particles. Start with
the initial state
 
1
√  1 

u0 = 1/ N 
 .. .
 (1)
 . 
1

Assume, for notational simplicity and without loss of generality, that the entry
we are looking for is represented by the state |↓↓ . . . ↓i, let B be the unitary
transformation whose only action is invert the phase of the desired component,
which in this case is
 
−1 0 · · · 0
 . 
 0 1 0 .. 
 
B= .. . (2)
 . 
 . 0 .. 0 
0 0 ··· 1
The algorithm consists on the repeated action of the unitary transformation
X = DB, where D is the diffusion matrix. Explicitly:
 N

−1 + 2
1 ··· 1
 .. 
2  −1 1− N
1 . 
 
X = DB =  ..
2 . (3)
N  .. 
 . 1 . 1 
−1 1 1 1 − N2

After m iterations of the unitary transformation X on u0 the quantum state

will be
 
Am
 
 Bm 
um = X m u0 = 
 .. ,
 (4)
 . 
Bm
where the amplitudes are given by the recursion formula
! ! ! ! !
Am+1 1− 2
2− 2
Am Am √1
m N
= −2
N N
2 =S =S √1
, (5)
Bm+1 N
1− N
Bm Bm N

The trick is that S can be diagonalized very easily, with eigenvalues e±iϕ
such that cos ϕ = 1 − N1 and therefore
1  √ 
Am = √ cos (mϕ) + N − 1 sin mϕ (6)
N
 !
1 1
Bm = √ cos (mϕ) − √ sin mϕ (7)
N N −1

Thus, the probability of finding the searched for state is:

1  √ 2
Pm = cos (mϕ) + N − 1 sin mϕ , (8)
N
with the change of variable ϕ = 2θ, Pm can be written as [4]:

Pm = sin2 (θ (2m + 1)) , (9)

π
√
This implies that Pm is periodic in m with period ' 2
N, and reaches its
maxima at

θ (2m + 1) = nπ, n integer, (10)

For large N, the first maximum simplifies

√
π N
m∼ . (11)
4
This exact √result agrees with Grover’s, in the sense of the existence of a
number m < N , such that after m iterations of the algorithm, if we measure
the state of the system, we will find the searched one with a probability of at
least 0.5.

III NOISY QUANTUM SEARCH ALGORITHM

Like all experimental devices, quantum computers will be subject to noise.
Let us assume that the same Gaussian noise is present at each step of the
algorithm, i.e. each time the unitary X matrix is applied. Even though some
quantum correction codes have been developed [5,6], it is known that these
codes work only if the noise is small enough. Moreover, it is not known whether
these codes are subject to noise themselves, and if they are, whether they can
still be useful. So, for simplicity, we assume that no quantum correction code
is available, and study the effect of Gaussian noise in Grover’s algorithm.
First, let us investigate the maximum noise the algorithm can put up with
before it loses periodicity and, worse, the searched for amplitude is no longer
enhanced. The size of the white noise is characterized by the standard devi-
ation σ of its normal distribution. In numerical experiments, we found that
the maximum noise the algorithm can allow before it breaks down is:
4
σmax ' . (12)
3N
Unfortunately, the amount of noise that the algorithm can handle is very small
for large databases.
Secondly, assume that the noise is smaller than σmax . What happens to the
number m of steps needed to reach Pm ' 1, i.e. to find almost certainly the
searched–for state? Let us focus on the limiting case when the algoritm still
works, with noise given by σmax . In another numerical experiment we found
that
3
m(σmax )
max ' 2N 4 . (13)

Thus, the exponent of N increases from 12 to 34 . The algorithm slows down

but is still faster than a classical one (this is all for large N).
Recently, Grover’s algorithm with N = 4 has been succesfully implemented
experimentally [7]. Our explicit results 12 and 13 are evaluated for large N,
so they do not apply to this case. Still, we can compute exactly the effect of
white noise on the speed and robustness of the algorithm. These results will
be presented elsewhere.

IV CONCLUSIONS
√ 
Grover’s quantum search algorithm requires O N steps, for a large
database with N entries. It thus improves any classical algorithm, needing
O(N)
 3 steps.
 Nevertheless, if noise is present, the algorithm slows down to
O N 4 steps, before breaking down completely. This breakdown occurs when
the width of the white noise reaches 43 N −1 . For large N, consequently, the
algorithm can withstand very little noise.
Acknowledgements. This work is supported in part by CONACYT
25504-E, DGAPA-UNAM IN103997. B.P.N. enjoys a scholarship from CONA-
CYT.

REFERENCES
1. L.K. Grover, Quantum computers can search rapidily by using almost any trans-
formation, Phys. Rev. Lett. vol. 80 (1998) 4329-4332.
2. L.K. Grover, Quantum Mechanics helps in searching for a needle in a haystack,
Phys. Rev. Letters, vol. 78 (2) (1997) 325-328..
3. D. Pyo, J. Kim, Quantum Database Searching by a Simple Query, quant-
ph/9708005.
4. M. Boyer, G. Brassard, P. Hoeyer / A. Tapp, Tight bounds on quantum searching,
Proc., PhysComp 1996..
5. A. Steane, Multi-particle interference and quantum error correction, Proc. R.
Soc. Lond. A 452 (1996) 2551-2577.
6. A.R. Calderbank and P.W. Shor, Good quantum error correction codes exist,
Phys. Rev. A 54 (1996) 1098-1105.
7. I.L. Chuang, Lieren, M.K. Vandersypen, Xin-lan Zhou, D.W. Leung, S. Loyd,
Experimental realization of a quantum algorithm, quant-ph/9801037.