BUTP–96/11

Efficient Simulation of Quantum Systems

by Quantum Computers
arXiv:quant-ph/9603026v2 14 Aug 1996

Christof Zalka
Institut für theoretische Physik, Universität Bern, Switzerland
March 25, 1996

Abstract
We show that the time evolution of the wave function of a quantum
mechanical many particle system can be implemented very efficiently on
a quantum computer. The computational cost of such a simulation is
comparable to the cost of a conventional simulation of the corresponding
classical system. We then sketch how results of interest, like the energy
spectrum of a system, can be obtained. We also indicate that ultimately
the simulation of quantum field theory might be possible on large quantum
computers.
We want to demonstrate that in principle various interesting things
can be done. Actual applications will have to be worked out in detail also
depending on what kind of quantum computer may be available one day...

1 Quantum Computers (QCs)

Quantum computers are still imaginary devices, but it is hoped that even-
tually the technical problems involved in their realization can be overcome
[1, 2, 3, 4, 5, 6, 7]. Quantum computers could solve some problems much faster
than conventional computers. Most prominently, Peter Shor (1994) has given
a ”quantum algorithm” for factoring large integers in polynomial time [8](see
also [2]).
A l-bit quantum computer may be thought of consisting of l two-state sys-
tems. Computations would be carried out by inducing unitary transformations
of a few at a time of these quantum-bits (qubits). This may be done by exterior
fields controlled from the outside. Upon observation at the end of the ”unitary
computation” the quantum computer would collapse into a state where each
qubit is either a 0 or a 1. The art of finding a ”quantum algorithm” is to
extract useful information from a few such runs.
The main technical problem in realizing QCs is, to prevent unwanted in-
teractions with the environment during the calculation. Such interactions can
cause decoherence, such that the QC is no longer in a pure quantum state. Here
we consider an idealized QC which doesn’t suffer from such problems.

1
2 Simulating Quantum Systems
General ideas about using specially designed quantum systems to simulate other
quantum systems have been published, e.g. by Feynman [9]. I present here an
actual implementation of the simulation of quantum mechanical many particle
systems on a general purpose QC 1 . For every degree of freedom of the system
we need a l-bit quantum register. This allows to store the whole (discretized)
wave function as amplitudes of the ”classical” states in the quantum computer.
Already to store so much information would be impossible for a conventional
computer. It is then shown that the time evolution for short time steps amounts
essentially to a Fourier transformation which can be carried out very efficiently
on a quantum computer [8, 10, 11].

2.1 Quantum Mechanical Particle in 1 Dimension

First we discretize the wave function and impose periodic boundary conditions:

an = ψ(n ∆x) an+N = an . (1)

After proper normalization we store these amplitudes in a l-bit quantum register:
N
X −1
|ψi = an |ni N = 2l . (2)
n=0
Where |ni is the basis state corresponding to the binary representation of the
number n. For short time steps ∆t the Greens function of the Schrödinger
equation is approximately:
2
m (x1 −x2 )
−iV (x1 )∆t
G(x1 , x2 , ∆t) = k ei 2 ∆t . (3)
Applying this to the amplitudes an is equivalent to transforming the basis states
with the inverse transformation. Using proper normalization to make the trans-
formation unitary, we get:
N −1
1 X −i m (n−n′ )2 ∆x2 +iV (n ∆x)∆t
|ni → √ e 2 ∆t |n′ i . (4)
N n′ =0
The crucial observation is that this corresponds to two diagonal unitary matrices
plus a discrete Fourier transformation:

N −1  
1 m n2 ∆x2
X nn′ ∆x2 m n′2 ∆x2
|ni → √ e−i 2 ∆t +iV (n ∆x)∆t
eim ∆t e−i 2 ∆t |n′ i . (5)
N n′ =0 | {z }
| {z } | {z }
diagonal diagonal
Fourier trafo
1 Independently and at practically the same time as me, Steven Wiesner [12] has published

a preprint concerning this, using a slightly different approach

2
To be able to use the fast Fourier transformation algorithm (FFT), we chose
the parameters such that:

∆x2 2π
m =A , A integer . (6)
∆t N
Provided the potential V has been made periodic (with period N), this also
makes the whole phase factor periodic in n and n′ , as is necessary for consistency.
It is rather straightforward to implement the FFT algorithm on a quantum
computer [8, 10, 11] such that it uses only about l2 /2 computational steps (local
unitary transformations). The diagonal unitary transformations of the type

|ni → eicF (n) |ni , (7)

can be done with the following succession of steps:

|n, 0i → |n, F (n)i ; |F (n)i → eicF (n) |F (n)i ; |n, F (n)i → |n, 0i .
(8)
In the first and last step the vectors represent two quantum registers. The first
step corresponds to parallel classical calculations. It can be carried out on a
quantum computer with a somewhat higher cost than just one such calculation
on a conventional computer (P. Shor [8]). As all transformations are unitary
they can just as well be inverted, so that the last step is also possible. The
second step is really a transformation of the form

|ni → eicn |ni , (9)

which can be carried out in a straightforward manner in l steps by transforming
each bit separately by an appropriate transformation. Note that of course all
these operations are carried out in parallel on all basis states which is why this
is often called ”quantum parallelism”.

2.2 Many Particles and Field Theory

The generalization to many particles is straightforward. For n particles in 3
dimensions we need 3n quantum registers. Besides the above steps we need
transformations corresponding to the coupling between particles and between
the different degrees of freedom of the same particle. They are diagonal unitary
transformations acting on several registers, e.g.:
′ ′′
|n, n′ , n′′ i → eicF (n,n ,n )
|n, n′ , n′′ i . (10)
Such a transformation can be carried out analogously to eq.(8). For identical
particles the initial state of the quantum computer has to be chosen symmet-
ric resp. antisymmetric. Also the generalization to more general Hamiltonian

3
operators than in the previous paragraph is possible (e.g. with coupling to a
magnetic field).
Also quantum field theory could be simulated, e.g. by discretizing the field as
in lattice gauge theory (for a standard text see e.g. [15]). For a bosonic quantum
field this would be equivalent to having 1 particle at each lattice point. At any
rate, applications like lattice QCD would require large quantum computers with
thousands of quantum bits, which are not going to be realizable soon.

2.3 Fermionic Field Theories

Fermionic quantum fields pose some problems, as their functional formulation
(path integrals, wave functionals,...) involves anticommuting Grassmann num-
bers. There is no wave functional in terms of the usual complex numbers, so
we have to look for a representation of the fermionic field operator-algebra on
some other Hilbert space. One possibility is the Fock space, where we have
for the n-particle sector a totally antisymmetric wave function of n variables.
Another possibility is to give the occupation number for each particle species at
each lattice point. For fermions this can only be 0 or 1 and thus requires only
1 qubit.
I expect, that further difficulties which may arise with quantum fields can
in principle be overcome, possibly along the same lines as it is done in lattice
QCD.

3 Other manipulations
3.1 Simulating a decay to obtain the ground state
Often one is interested in the ground state of a quantum system, be this the
ground state of a molecule, the vacuum of a field theory or a hadron stable under
the strong interaction. I propose here to obtain the ground state essentially in
the same way as it happens in reality, namely by letting some initial state decay.
Decays happen because the instable particle is coupled to some other quantum
system, like the electromagnetic field. It is important that this quantum system
has a lot of different energy eigenvalues such that it can absorb the energy from
various transitions in the decaying system. For the simulation of a quantum
computer we may take a collection of 2-level systems with energy gaps ∆E =
E0 2−n , n = 0 . . . l such that the whole system has practically a continuous
energy spectrum. The coupling to the quantum system of interest may be done
in various ways.
During the simulation of the decay, the auxiliary ”energy-drain” system
must periodically be reset to its ground state such that it is able to absorb more
energy. As the study of simple systems shows, this does not necessarily in every
step, but only in the average, lead to a reduction of the energy expectation

4
value of the system of interest. For actual implementations one may have to
think about an adequate ”resetting” strategy, e.g. depending on the state the
auxiliary system is found in upon observation just before each resetting. Also
one may want to reduce the coupling towards the end of the decay simulation
in order not to disturb the ground state of interest too much.
Actually it may be enough to simulate the decay until the few lowest lying
states (which are of interest) have a large amplitude. Then one can follow the
procedure of paragraph 4.2 to obtain one of the energy eigenstates.

3.2 Putting a wave function on a Quantum Computer

Here I sketch a method which in principle allows us to store a given wave function
as an initial state on a QC. I demonstrate this for the case of a scalar quantum
mechanical particle in 1 dimension. The generalization to multi dimensional
wave functions is then rather straightforward. Here I show only how the absolute
value of the wave function can be stored in a quantum register, as above I have
already shown how the complex phase can later be introduced.
So we want to carry out the following transformation on a l bit quantum
register which initially is in the state |0i:

l
2X −1  
L
|0i → k ψ n l |ni , where ψ(x + L) = ψ(x) , (11)
n=0
2

and k is an appropriate normalization constant. The idea now is to split the

norm of the initial state (which is 1, of course) l times such that in the end
we get the 2l contributions for the single basis states |ni.2 (Imagine a binary
tree with l levels and 2l leaves.) For this to work out correctly each split has to
yield the right ratio. For this we must have an algorithm which can compute
the integrals

Z k+1
L
2i
Ii,k = |ϕ(x)|2 dx , for k = 0 . . . 2i − 1 and i = 1 . . . l . (12)
k
L
2i

Now the first split is realized by a real O(2) rotation acting on the most signif-
icant qubit of the quantum register, thus:
p p
|0, . . .i → I1,0 |0, . . .i + I1,1 |1, . . .i . (13)
Such an O(2) rotation of a qubit can be carried out analogously to the rephasing
of an amplitude (see eq. (7)) once the rotation angle ϕ has been calculated:
2 Note that the norm of mutually orthogonal substates is a useful notion in quantum com-

puting because it is conserved under unitary transformations.

5
 p
sin ϕ = I1,1 . (14)
The subsequent calculations of angles and rotations are then carried out in
quantum parallelism, thus:

|bl , bl−1 , ...bl−i , 0, ...i → cos(ϕ) |bl , bl−1 , ...bl−i , 0, ...i+sin(ϕ) |bl , bl−1 , ...bl−i , 1, ...i ,
p (15)
with sin(ϕ) = Il−i+1,k and k = bl−i + 2 bl−i+1 + ... + 2i bl .

So this happens simultaneously for all values of the b’s (resp. of k).
The method given here to input a wave function to a QC is rather ambitious,
as essentially it evaluates the function at every (discretized) point x. In practice
a simpler sceme may do.

4 Obtaining Results of Interest

4.1 Measuring positions of particles, resp. field strengths
One can think of various ways how to obtain quantities of interest from a quan-
tum computer simulation of a quantum system. Say we have some state on the
QC which we want to analyze, like the vacuum or a 1-particle state of a field
theory. One can now make a measurement of some field strength simply by ob-
serving the corresponding quantum register. By repeatedly observing some field
values in different points one may in several runs of the QC get a statistical esti-
mate of n-point functions. In particular, one may obtain the correlation length.
To get n-point functions for space-time points with a timelike separation, one
has to continue the simulation of the time evolution between observations of
single quantum registers.

4.2 Measuring arbitrary observables

To study the measurement problem in quantum theory, J. von Neumann has
proposed an idealized simple interaction [13, 14] between the quantum system
and the part of the measuring apparatus that directly interacts with the system.
For convenience this part of the apparatus can be chosen to be equivalent to a
quantum mechanical particle in 1 dimension or, more precisely, its Hilbert space
is spanned by the basis vectors

|xi x real, and X̂|xi = x|xi . (16)

The time evolution during the measurement of an observable  shall then be
given by

6
 Û(t)|Ψa i|xi = |Ψa i|x + k · a · ti , with Â|Ψa i = a|Ψa i , (17)

where |Ψa i is the state of the system. One easily verifies, that the Hamilton
operator Ĥ = k P̂ Â leads to this time evolution. Here P̂ is the usual momentum
operator for the imaginary quantum mechanical particle, thus [X̂, P̂ ] = ih̄ .
Now we simply have to implement this time evolution on the QC. This can be
expected to work out for any reasonable observable represented by the hermitian
operator  which can be written in terms of the fundamental observables and
their conjugate momenta.
Then, after simulating this “1. stage von Neumann measuring process”, we
can observe the auxiliary quantum register and should find a number propor-
tional to an eigenvalue of Â. By repeating this, we get the spectrum of  and
the weight of the various Â-eigenstates in the original system state. Also after
each observation the system will be in an eigenstate of Â. Thus this is a way of
obtaining e.g. the energy eigenstates.

5 Conclusion
I have demonstrated that on a large enough quantum computer various quantum
theoretic quantities of interest could be calculated, which may be too hard to
compute by a conventional computer. This is in particular true for strongly
interacting field theories like QCD. However simulating quantum fields would
require large quantum computers with many thousand qubits, which we don’t
know whether they can ever be built. Furthermore, here I have assumed to have
an ideal QC which doesn’t suffer from decoherence and can carry out unitary
transformations with sufficient precision (as these are really analog operations).

References
[1] S. Lloyd, A Potentially Realizable Quantum Computer.
Science 261, 1569 (1993)
[2] I. L. Chuang et al., Quantum Computers, Factoring and Decoherence. LA-
UR-95-241, quant-ph/9503007
[3] I. L. Chuang and Y. Yamamoto, A Simple Quantum Computer. quant-
ph/9505011 (1995)
[4] I. L. Chuang et al., Effects of Loss and Decoherence on a Simple Quantum
Computer. quant-ph/9602018 (1996)
[5] A. Barenco, A Universal Two Bit Gate for Quantum Computation. quant-
ph/9505016 (1995)

7
 [6] A. Barenco et. al., Elementary Gates for Quantum Computation. quant-
ph/9503016 (1995)
[7] W.G. Unruh, Maintaining Coherence in Quantum Computers. hep-
th/9406058 (1994)
[8] P. Shor, Polynominal-Time Algorithms for Prime Factorization and Dis-
crete Logarithms on a Quantum Computer. In Proc. 35th Annual Sympo-
sium on Foundations of Computer Science. IEEE Press, pp 124-134, Nov.
1994, quant-ph/9508027
[9] R. Feynman, Internat. J. Theoret. Phys., 21, pp 467-488 (1982)
[10] D. Coppersmith, IBM Research Report RC 19642 (1994)
[11] A. Ekert and R. Jozsa (1995), Rev. Mod. Phys., to appear
[12] S. Wiesner, Simulations of Many-Body Quantum Systems by a Quantum
Computer. quant-ph/9603028 (1996)
[13] J. von Neumann, Mathematical Foundations of Quantum Mechanics,
Princeton University Press, Princeton (1955)
[14] B. S. DeWitt, in Physics Today, September 1970
[15] I. Montvay, G. Münster, Quantum Fields on a Lattice,
Cambridge University Press (1994)