et al.
et al.
et al.
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
REPORTS<br />
792<br />
overhead that precludes efficient univers<strong>al</strong> quantum<br />
computation.<br />
We consider gener<strong>al</strong>izations of quantum w<strong>al</strong>k<br />
with many interacting w<strong>al</strong>kers and show that multiparticle<br />
quantum w<strong>al</strong>k is univers<strong>al</strong> for quantum<br />
computation using a graph of polynomi<strong>al</strong> size.<br />
We present efficient univers<strong>al</strong> constructions based<br />
on the Bose-Hubbard model, fermions with nearestneighbor<br />
interactions, and distinguishable particles<br />
with nearest-neighbor interactions. We <strong>al</strong>so<br />
show that <strong>al</strong>most any interaction gives rise to univers<strong>al</strong>ity<br />
when using indistinguishable particles.<br />
Because our graphs are exponenti<strong>al</strong>ly sm<strong>al</strong>ler<br />
than those in reference (7), our construction is<br />
amenable to experiment<strong>al</strong> implementation using<br />
an architecture where vertices of the graph occupy<br />
different spati<strong>al</strong> locations (<strong>al</strong>though we leave<br />
issues of fault tolerance for future work). Current<br />
two-particle experiments involve only noninteracting<br />
bosons (11–14), but other experiments<br />
could re<strong>al</strong>ize interactions. Specific<strong>al</strong>ly, a Bose-<br />
Hubbard model of the type we consider could<br />
natur<strong>al</strong>ly be re<strong>al</strong>ized in a vari<strong>et</strong>y of systems, including<br />
tradition<strong>al</strong> nonlinear optics (15), neutr<strong>al</strong><br />
atoms in optic<strong>al</strong> lattices (16, 17), or photons in arrays<br />
of superconducting qubits (18).<br />
Multiparticle quantum w<strong>al</strong>k has been considered<br />
as an <strong>al</strong>gorithmic tool for solving graph isomorphism<br />
(19), <strong>al</strong>though this technique has known<br />
limitations (20). Other previous work has focused<br />
on two-particle quantum w<strong>al</strong>k (11–14, 21–25)and<br />
multiparticle quantum w<strong>al</strong>k without interactions<br />
(11–14, 21, 22, 26). Here, we consider multiparticle<br />
quantum w<strong>al</strong>ks with interactions, which<br />
seem to be required to achieve computation<strong>al</strong> univers<strong>al</strong>ity<br />
(27, 28).<br />
It has been shown that systems of interacting<br />
particles on a lattice can perform univers<strong>al</strong> computation<br />
with a time-dependent Hamiltonian<br />
(29, 30). However, because of the time dependence,<br />
such a system should not be considered a<br />
quantum w<strong>al</strong>k. In our scheme, the Hamiltonian<br />
is time-independent and the computation is encoded<br />
entirely in the unweighted graph on which<br />
the particles interact.<br />
In a multiparticle quantum w<strong>al</strong>k, particles (either<br />
distinguishable or indistinguishable) interact<br />
in a loc<strong>al</strong> manner on a simple graph G with vertex<br />
s<strong>et</strong> V(G) and edge s<strong>et</strong> E(G). The Hilbert space<br />
for m distinguishable particles on G is spanned<br />
by the basis<br />
fji1,…,im〉 : i1,…,im ∈ VðGÞg ð1Þ<br />
where iw is the location of the wth particle. A<br />
continuous-time multiparticle quantum w<strong>al</strong>k of m<br />
distinguishable particles on G is generated by a<br />
time-independent Hamiltonian<br />
H ðmÞ<br />
G<br />
¼ ∑m<br />
w¼1<br />
∑<br />
ði, jÞ∈EðGÞ<br />
ðji〉〈 jj w þjj〉〈ij w Þþ<br />
∑ Uijðn% i; n% jÞ ð2Þ<br />
i, j∈V ðGÞ<br />
where the subscript w indicates that the operator<br />
acts nontrivi<strong>al</strong>ly only on the location register for<br />
the wth particle. Here, the operators n% i and n% j<br />
count the numbers of particles at vertices i and j,<br />
respectively (explicitly, n% i ¼ ∑ m w¼1 ji〉〈ij w ).<br />
The first term of Eq. 2 moves particles b<strong>et</strong>ween<br />
adjacent vertices, and the second term is an<br />
interaction b<strong>et</strong>ween particles. We assume that Uij<br />
is zero whenever one of its arguments ev<strong>al</strong>uates<br />
to zero and that Uii is zero if there is only one<br />
particle at vertex i (thus, the Hamiltonian for a<br />
single particle reduces to that of a standard quantum<br />
w<strong>al</strong>k). We <strong>al</strong>so assume that the interaction<br />
Uij only depends on the distance b<strong>et</strong>ween i and j<br />
and has a constant range. Fin<strong>al</strong>ly, we assume that<br />
the norm of each term Uij is upper-bounded by a<br />
polynomi<strong>al</strong> in m.<br />
Indistinguishable particles can be represented<br />
in the basis specified in Eq. 1 as states that are<br />
either symm<strong>et</strong>ric (for bosons) or antisymm<strong>et</strong>ric<br />
(for fermions) under the interchange of any two<br />
particles. The Hamiltonian preserves both the symm<strong>et</strong>ric<br />
and antisymm<strong>et</strong>ric subspaces and, restricted<br />
to the appropriate subspace, generates a quantum<br />
w<strong>al</strong>k of m bosons or m fermions on G.<br />
This framework includes well-known interacting<br />
many-body systems defined on graphs. For example,<br />
it includes the Bose-Hubbard model, with<br />
interaction Uijðn% i,n% jÞ¼ðU=2Þdi, jn% iðn% i − 1Þ. It<br />
<strong>al</strong>so includes systems with nearest-neighbor interactions,<br />
such as with Uijðn% i,n% jÞ¼Ud ði, jÞ∈EðGÞn% in% j.<br />
The dynamics of our scheme can be understood<br />
by considering scattering events involving<br />
only one or two particles and can be an<strong>al</strong>yzed<br />
using a discr<strong>et</strong>e version of scattering theory (31).<br />
We first discuss single-particle scattering.<br />
Consider a single-particle quantum w<strong>al</strong>k on<br />
an infinite graph G obtained by attaching a semiinfinite<br />
path to each of N chosen vertices of an<br />
arbitrary (N + m)–vertex graph G % (Fig. 1A). A<br />
particle is initi<strong>al</strong>ly prepared in a state that moves<br />
(under Schrödinger time evolution) toward the<br />
A (4, 1)<br />
(3, 1)<br />
(2, 1)<br />
(4, 2)<br />
(1, 1) (3, 2)<br />
G (1, 2)<br />
(2, 2)<br />
(1,N)<br />
(2,N)<br />
(3,N)<br />
(4,N)<br />
B<br />
subgraph G % <strong>al</strong>ong one of the semi-infinite paths.<br />
After scattering through the subgraph, the particle<br />
moves away from G % in superposition <strong>al</strong>ong the<br />
semi-infinite paths. To understand this process, we<br />
discuss the scattering eigenstates of the Hamil-<br />
tonian H ð1Þ<br />
G .<br />
Given the graph G % , for each momentum<br />
k ∈ ð−p,0Þ and path j ∈ f1,2…,Ng, there is a<br />
scattering state jscjðkÞ〉 with amplitudes<br />
〈x,qjscjðkÞ〉 ¼ e −ikx dqj þ e ikx SqjðkÞ ð3Þ<br />
on the semi-infinite paths [with the labeling (x,q)<br />
of vertices on the paths as in Fig. 1A], where S(k)<br />
is an N by N unitary matrix c<strong>al</strong>led the S-matrix<br />
[see section S1 of (32)]. The state jscjðkÞ〉 is an<br />
eigenstate of H ð1Þ<br />
G with energy 2 cos k.<br />
A wave pack<strong>et</strong> is a norm<strong>al</strong>ized state with<br />
most of its amplitude on scattering states with<br />
momenta close to some particular v<strong>al</strong>ue. The scattering<br />
state jscjðkÞ〉 gives us information about<br />
how a wave pack<strong>et</strong> with momenta near k located<br />
on path j scatters from G % . The wave pack<strong>et</strong><br />
initi<strong>al</strong>ly moves toward the graph with speed<br />
j dE<br />
dk j¼j2sinkj. After scattering, the amplitude<br />
associated with finding the wave pack<strong>et</strong> on path q<br />
is Sqj(k). This picture of the scattering process<br />
remains v<strong>al</strong>id when the finite extent of the wave<br />
pack<strong>et</strong>s is taken into account and when the infinite<br />
paths are truncated to be long but finite<br />
[section S5 of (32)].<br />
We now consider scattering of two indistinguishable<br />
particles on an infinite path. Translation<br />
symm<strong>et</strong>ry makes this system easier to an<strong>al</strong>yze<br />
than more gener<strong>al</strong> two-particle quantum w<strong>al</strong>ks<br />
(33). Consider two indistinguishable particles<br />
initi<strong>al</strong>ly prepared in spati<strong>al</strong>ly separated wave pack<strong>et</strong>s<br />
moving toward each other <strong>al</strong>ong a path with<br />
momenta k1 ∈ ð−p; 0Þ and k2 ∈ ð0; pÞ. After<br />
scattering, the particles move apart with momenta<br />
k<br />
Fig. 1. (A) The infinite graph G. The vertices labeled (1, j) belongtoG%. (B) Encodedj0〉 (left) andj1〉<br />
(right).<br />
A<br />
0in<br />
0out<br />
B<br />
0in<br />
1in 1out 1in<br />
1out 1in<br />
Fig. 2. (A)Phasegateat–p/4. (B) Basis-changing gate at –p/4. (C) Hadamard gate at –p/2.<br />
0out<br />
15 FEBRUARY 2013 VOL 339 SCIENCE www.sciencemag.org<br />
C<br />
0in<br />
k<br />
0out<br />
1out<br />
Downloaded from<br />
www.sciencemag.org on February 14, 2013