Quantum Networks on Cubelike Graphs

Anna Bernasconi
Dipartimento di Informatica, Università degli Studi di Pisa, 56127 Pisa, Italy

Chris Godsil
Department of Combinatorics & Optimization, University of Waterloo, Waterloo N2L 3G1, ON Canada

Simone Severini
Institute for Quantum Computing and Department of Combinatorics & Optimization,
University of Waterloo, Waterloo N2L 3G1, ON Canada

Cubelike graphs are the Cayley graphs of the elementary abelian group Zn 2 (e.g., the hypercube
arXiv:0808.0510v1 [quant-ph] 4 Aug 2008

is a cubelike graph). We study perfect state transfer between two particles in quantum networks
modeled by a large class of cubelike graphs. This generalizes results of Christandl et al. [Phys. Rev.
Lett. 92, 187902 (2004)] and Facer et al. [Phys. Rev. A 92, 187902 (2008)].

I. INTRODUCTION have the hypercube as a spanning subgraph (for this rea-

son, these authors coined the term dressed hypercubes).
In view of applications like the distribution of crypto- Other questions related to quantum dynamics on hyper-
graphic keys [3, 11] or the communication between reg- cubes have been addressed in [2, 8, 21, 23, 26].
isters in quantum devices [6, 20], the study of natural The paper is organized as follows: Section 2 contains
evolution of permanently coupled spin networks has be- the necessary definitions and the statements of our re-
come increasingly important. A special case of interest sults. A proof will be given in Section 3. This is obtained
consists of homogenous networks of particles coupled by by diagonalizing the Hamiltonians with simple tools from
constant and fixed (nearest-neighbour) interactions. Fourier analysis on Zn2 .
An important feature of these networks is the possi-
bility of faithfully transferring a qubit between specific
particles without tuning the couplings or altering the net- II. SET UP AND RESULTS
work topology. This phenomenon is usually called perfect
state transfer (PST). Since quantum networks (and com-
munication networks in general) are naturally associated Let Zn2 be the additive abelian group (Z2 )×n . Each el-
to undirected graphs, there is a growing amount of liter- ement of Zn2 is represented as a binary vector of length n.
ature on the relation between graph-theoretic properties The zero vector 0 is made up of all 0’s. Let f : Zn2 → Z2
and properties that allow PST (see [13, 15, 17, 18, 28]). be a Boolean function on n variables and let Ωf = {w ∈
In the present paper we will give necessary and suffi- Zn2 | f (w) = 1}. Let d = |Ωf | be the number of vectors
cient conditions for PST in quantum networks modeled w ∈ Zn2 such that f (w) = 1. Finally, if w and v are two
by a large class of cubelike graphs. The vertices of a cube- binary vectors of the same length, then w ⊕ v denotes the
like graph are the binary n-vectors; two vertices u and vector obtained by computing their elementwise addition
v are adjacent if and only if their symmetric difference modulo 2, and wT v their scalar product.
belongs to a chosen set. Equivalently, cubelike graphs The Cayley graph X(Γ, T ) of a group Γ w.r.t. the set
on 2n vertices are the Cayley graphs of the elementary T ⊆ Γ (T = T −1 ) is the graph with vertex-set V (X) =
abelian group of order 2n [10, 24]. {Γ} and an edge {g, h} ∈ E(X), if there is s ∈ T such
Among cubelike graphs, the hypercube is arguably the that gs = h. The set T is also called Cayley set. The
most famous one, having many applications ranging from Cayley graphs of the form X(Zn2 , Ωf ) are called cubelike
switching theory to computer architecture, etc. (see, e.g., graphs. Some cubelike graphs are illustrated in Figure 1
[16, 22]). There are various and diverse results about below.
quantum dynamics on hypercubes. These are essentially Notice that this definition embraces every possible set
embraced by two areas: continuous-time quantum walks Ωf . When f is the characteristic function of the stan-
[19]; quantum communication in spin networks [7]. The dard generating set of Zn2 , the graph X(Zn2 , Ωf ) is called
common ingredient is the use of a Hamiltonian represent- hypercube. (For example, the leftmost graph in Figure
ing the adjacency structure of the graph. 1). The adjacency matrix of X(Zn2 , Ωf ) is the 2n × 2n
Concerning state transfer, Christandl et al. [9] have matrix
shown that networks modeled by hypercubes are capable X
of transporting qubits between pairs of antipodal nodes, Af = ρreg (w),
perfectly (i.e., with maximum fidelity) and in constant w∈Ωf
time. Facer et al. [12] generalized this observation, by
considering a family of cubelike graphs whose members where ρreg (x) is the regular (permutation) representation
 2

2. For tL= π/2, we have F (a, b) = 1 if a ⊕ b = u and

u = w∈Ωf w 6= 0.

As a simple consequence of this statement, we have

various ways to route information between any two nodes
of a network whose vertices correspond to the elements
of Zn2 . Let f be a Boolean function such Lthat Ωf =
{w1 , ..., wr } is a generating set of Zn2 . Let wi ∈Ωf wi =
w 6= 0. Let us define C = {w, w1 , ..., wr } and Ci = C\wi .
FIG. 1: Drawings of three nonisomorphic cubelike graphs on
Since the sum of the elements of Ci is nonzero, the Cayley
8 vertices.
graph X(Zn2 , Ci ) has PST between a and b such that
a ⊕ b = wi at time π/2. The simplest case arises when
we chose r = n and take the vectors w1 , ..., wr to be
of w ∈ Ωf . In particular, if w = w1 w2 · · · wn then the standard basis of Zn2 . Then X(Zn2 , C) is the folded
d-cube and, by using a suitable sequence of the graphs
n
O X(Zn2 , Ci ), we can arrange PST from the zero vector to
ρreg (w) = σxwi , any desired element of Zn2 .
i=1 For example, consider the case d = 3. We write w1 =
where σx is a Pauli matrix. It is clear that Af commutes (100), w2 = (010) and w3 = (001). Then w = (111).
with the adjacency matrix of any other cubelike graph, The graph X(Z32 , C) is illustrated in Figure 1 – left. Since
given that the group Zn2 is abelian. w1 = w2 ⊕w3 ⊕w, there is PST between 000 and w1 = 100
Now, let us choose a bijection between vertices of at time π/2, given that 000 ⊕ 100 = 100 (see Figure 2
X(Zn2 , Ωf ) and the elements of the standard basis – right). Also, there is PST for the pairs {010, 110},
|1i, |2i, ..., |N i of an Hilbert space H ∼ {001, 101} and {011, 111}. Notice that X(Zn2 , C1 ) is iso-
= CN , where
N = 2n . This is the usual space of n qubits. If we morphic to the 3-dimensional hypercube.
look at the single excitation case in the XY model, the
evolution of a network of spin 1/2 quantum mechanical
particles on the vertices of X(Zn2 , Ωf ) can be seen as in-
duced by the adjacency matrix Af , which then plays the
role of an Hamiltonian (for details see [12] or [9]). On
the light of this observation, given two vectors a, b ∈ Zn2 ,
the (unnormalized) transition amplitude between a and
b induced by Af is the expression

T (a, b) = hb|e−iAf t |ai (1)

X T T
= (−1)a w e−iλw t (−1)b w
w∈Zn
2
X T
= (−1)(a⊕b) w −iλw t
e , FIG. 2: Left: The graph X(Zn 2 , C), where C =
w∈Zn
2 {(100), (010), (001), (111)}. This graph is isomorphic to the
complete bipartite graph K4,4 . Right: The graph X(Zn 2 , C1 ),
where t ∈ R+ . The fidelity of state transfer between a where C1 = {(010), (001), (111)}. This graph turns out
and b is then F (a, b) = 21n |T (a, b)|. By definition, the to be isomorphic to the hypercube of dimension 3. Since
evolution under Af is periodic if there is t ∈ R+ such 100 = 010 ⊕ 001 ⊕ 111, there is PST between 000 and 100 at
that F (a, a) = 1 for every a ∈ Zn2 . Graph-theoretic prop- time π/2, given that 000 ⊕ 100 = 100. Also, there is PST for
erties responsible of periodic evolution (sometime also the pairs {010, 110}, {001, 101} and {011, 111}.
called perfect revival ) have been considered in the litera-
ture ([14, 15, 27, 28, 29]).
With the next proposition, we show that every network For generic dimension, the hypercube is X(Zn2 , E),
−→
modeled by a cubelike graph has a periodic evolution. where E = {(10...0), (010...0), ..., (0...01)} and Jn =
n
L
The period is π and it does not depend on the number w∈E w, the all-ones vector of length 2 . Since the Ham-
of vertices of the graph. Equivalently, it does not depend ming distance between two different elements a, b ∈ Zn2
on the dimension of H. is exactly their distance in X(Zn2 , E), we have PST be-
tween any two antipodal vertices of the hypercube, as it
Theorem 1 Let X(Zn2 , Ωf ) be a cubelike graph and let was already observed in [9] and [12]. Recall that the dis-
a, b ∈ Zn2 . tance between two vertices in a graph is the length of the
geodesic (equivalently, the shortest path) connecting the
1. For t = π, we have F (a, b) = 1 if and only if a = b. vertices. Two vertices are said to be antipodal if their
 3

distance is the diameter of the graph, i.e., the longest that, for any v ∈ Zn2 ,
among all the geodesics. Antipodal vertices in a Cayley X T X T
graphs are connected via a sequence of all elements of the λv = (−1)w v f (w) = (−1)w v
Cayley set. It remains as an open problem to verify that w∈Zn
2 w∈Ωf
whenever there is PST between two vertices of a cubelike X
wT v aT v
= (−1) + (−1)
graph then the vertices are antipodal.
w∈Ω′f

( w)T v
L
X T w∈Ω′
= (−1)w v
+ (−1) f

III. PROOF OF THE THEOREM w∈Ω′f

X T Y T
= (−1)w v
+ (−1)w v
.
The abstract Fourier transform of a Boolean function f w∈Ω′f w∈Ω′f
is the rational valued function f ∗ : Zn2 → Q which defines
the coefficients of f with respect to the orthonormal basis T
Let nv = |{w ∈ Ω′ | (−1)w v = −1}|, and pv = |{w ∈
of the functions T
Ω′ | (−1)w v = 1}|. Note that pv = |Ω′f | − nv = (d − 1) −
T nv . We have
Qw (x) = (−1)w x
,
λv = pv − nv + (−1)nv = d − 1 − 2nv + (−1)nv .
that is,
Thus, if nv is an even number, nv = 2kv , with kv ∈ N,
X T we get
f ∗ (w) = 2−n (−1)w x
f (x).
x∈Zn
2 λv = d − 1 − 4kv + 1 = d − 4kv .

Then Otherwise, if nv = 2kv − 1 is odd, we get

X T λv = d − 1 − 2(2kv − 1) − 1 = d − 4kv .
f (x) = (−1)w x ∗
f (w)
w∈Zn
2 Since, for all v, −d ≤ λv ≤ d, we finally get 0 ≤ kv ≤
⌊d/2⌋.
is the Fourier expansion of f . Note that the zero-order
Fourier coefficient is equal to the probability that the Lemma
L 3 Let f be a Boolean function such that
function takes the value 1, i.e., f ∗ (0) = 2dn , while the w∈Ωf w = u 6= 0.
other Fourier coefficients measure the correlation be-
1. If u 6∈ Ωf , then, for all v ∈ Zn2 ,
tween the function and the parity of subsets of its ar-
guments. 
d − 4kv , if uT v is even;
Using a vector-representation for the functions f and λv =
d − 4kv + 2, if uT v is odd;
f ∗ , and considering the natural ordering of the binary
vectors w ∈ Zn2 , one can derive a convenient matrix where d = |Ωf | and kv ∈ N, 0 ≤ kv ≤ ⌊ d+1
2 ⌋.
formulation for the transform pair: f = Hn f ∗ and
f ∗ = 21n Hn f , where Hn is the Hadamard transform ma- 2. If u ∈ Ωf , then, for all v ∈ Zn2 ,
trix. Given a function f : Zn2 → Z2 , the set Ωf defines the
if uT v is even;

d − 4kv ,
Cayley graph X(Zn2 , Ωf ), whose spectrum coincides, up λv =
d − 4kv − 2, if uT v is odd;
to a factor 2n , with the Fourier spectrum of the function:
1
2n Hn Af Hn = Df , where Af is the adjacency matrix of where d = |Ωf | and kv ∈ N, 0 ≤ kv ≤ ⌊ d−1
X(Zn2 , Ωf ) and Df is an 2n × 2n diagonal matrix. In par- 2 ⌋.

ticular, for w ∈ Zn2 , we have λw = 2n f ∗ (w). Theorem 1 Proof. (1.) Consider the function g such that Ωg =
needs the following two technical lemmas. Ωf ∪ {u}. Let µv denote the
L eigenvalues of the Cayley
graph associated to g. As w∈Ωg w = 0, from Lemma
Lemma 2 Let f be a Boolean function such that 2, we get
= 0. Then, for all v ∈ Zn2 ,
L
w∈Ωf w µv = |Ωg | − 4kv = d + 1 − 4kv ,

λv = d − 4kv , with 0 ≤ kv ≤ ⌊ d+1

2 ⌋. Now, observe that
X T X T
where d = |Ωf | and kv ∈ N, 0 ≤ kv ≤ ⌊d/2⌋. µv = (−1)w v g(w) = (−1)w v
w∈Zn
2 w∈Ωg
wT v uT v T
X
Proof. Let a ∈ Ωf , a 6= 0, and let Ω′f
= Ω \ {a}. Since = (−1) + (−1) = λv + (−1)u v
.
L L
w∈Ωf w = 0, we have that a = w∈Ω w. Now observe
′ w∈Ωf
f
 4
T
The thesis is verified, since w∈Zn (−1)(a⊕b) w = 2n if
P
Thus,
L 2
T T and only if a ⊕ b = 0 = w∈Ωf w, but here a 6= b. The
λv = µv − (−1)u v
= d + 1 − 4kv − (−1)u v
, L
remaining case is when w∈Ωf w = u 6= 0. Applying
and the thesis immediately follows. Lemma 3, we can write
(2.) Consider the function g such that Ωg = Ωf \ {u}. T
X
T (a, b) = (−1)(a⊕b) w e−i(d−4kw )π/2
Let µv denote the eigenvalues of theLCayley graph asso- w∈Zn
L 2
ciated to g. As u = w∈Ω f
w = w∈Ωg ⊕u, we have uT w even
T
L X
w∈Ωg w = 0. By applying Lemma 2, we obtain (−1)(a⊕b) w −i(d−4kw ±2)π/2
+ e
w∈Zn
2
µv = |Ωg | − 4kv = d − 1 − 4kv , uT w odd

with 0 ≤ kv ≤ ⌊ d−1
2 ⌋.
Now, as in (1.) observe that  X T
= e−idπ/2  (−1)(a⊕b) w ei2kw π
T T w∈Zn
X X
λv = (−1)w v f (w) = (−1)w v 2
uT w even
w∈Zn w∈Ωf

2
X T T T
(−1)w v
+ (−1)u v
= µv + (−1)u v
X T
= . − (−1)(a⊕b) w i2kw π 
e 
w∈Ωg w∈Zn
2
uT w odd

So we get X T T
= e−idπ/2 (−1)(a⊕b) w
(−1)u w
T w∈Zn
λv = d − 1 − 4kv + (−1)u v
, 2
X T
= e−idπ/2 (−1)(a⊕b⊕u) w
.
concluding the proof of the lemma. w∈Zn
2

Proof of Theorem 1. (1.) All eigenvalues λw are inte- T

The statement holds since w∈Zn (−1)(a⊕b⊕u) w = 2n if
P
gers with the same parity. In particular, they are all odd 2 L
if d = |Ωf | is odd, and all even, otherwise. Thus, by Eq. and only if a ⊕ b ⊕ u = 0, i.e., a ⊕ b = u = w∈Ωf w.
(1), for t = π we have
X T
T (a, b) = (−1)(a⊕b) w e−iλw π IV. CONCLUSION
w∈Zn
2
X T We have given a necessary and sufficient condition for
= (−1)(a⊕b) w
cos(λw π) PST in quantum networks modeled by a large class of
w∈Zn
2 cubelike graphs.
T
X L
= (−1)(a⊕b) w
(−1)λw A special case is left open: when w∈Ωf w = 0. Nu-
w∈Zn
2
merical evidence suggests that cubelike graphs with this
X T property do not allow PST.
= (−1)d (−1)(a⊕b) w
. An application of our result is a straightforward
w∈Zn
2 methodL to distinguish L if, for a Boolean function f , we
Thus the statement follows since have w∈Ωf w 6= 0 or w∈Ωf w = 0, when this promise
 n holds. By Theorem 1, we know that hb|e−iAf π/2 |ai = 1
X T 2 , a ⊕ b = 0, i.e., a = b; if a ⊕ b = w 6= 0. If we let evolve the system under
(−1)(a⊕b) w =
n
0, otherwise. the Hamiltonian Af for a time π/2, we then obtain the
w∈Z2
L state |bi by performing a von Newmann measurement
(2.) First, let us suppose that f is such that w∈Ωf w = w.r.t the standard basis of H. When n gets large, the
0. Then, applying Lemma 2, we can see that probability of observing |bi with distinct
L measurements
X T
tends to decrease exponentially if w∈Ωf w = 0.
T (a, b) = (−1)(a⊕b) w e−iλw π/2 On the other side, this problem is easy without the use
w∈Zn
2 of any quantum technique. Beyond this trivial applica-
X T tion, given the link between cubelike graphs and Boolean
= (−1)(a⊕b) w −i(d−4kw )π/2
e
functions, it is natural to ask weather quantum dynam-
w∈Zn
2
ics on these graphs can help in getting useful information
X T
= e−idπ/2 (−1)(a⊕b) w i2kw π
e about the corresponding functions.
w∈Zn
2
X T Acknowledgments. We thank Sougato Bose, Andrew
= e−idπ/2 (−1)(a⊕b) w
. Childs and Peter Høyer for helpful discussion. Part
w∈Zn
2 of this work was carried on during the conference
 5

“Graph Theory and Quantum Information: Emerging ical Physics (April 28 - May 2, 2008, Waterloo).
Connections”, held at Perimeter Institute for Theoret-

[1] W. Adamczak, K. Andrew, P. Hernberg, C. Tamon, A over distance-regular spin networks. arXiv:0709.0755v2
note on graphs resistant to quantum uniform mixing. [quant-ph]
arXiv:quant-ph/0308073v1 [18] M. A. Jafarizadeh, R. Sufiani, S. F. Taghavi, E. Barati,
[2] G. Alagic, A. Russell, Phys. Rev. A 72, 062304 (2005). Perfect transference of a d-level quantum state over
arXiv:quant-ph/0501169v5 pseudo-distance-regular networks. arXiv:0803.2334v1
[3] C. H. Bennett and G. Brassard, Quantum Cryptogra- [quant-ph]
phy: Public Key Distribution and Coin Tossing, Proc. [19] J. Kempe, Contemporary Physics, Vol. 44 (4), pp.307-
IEEE Int. Conf. Computers Systems and Signal Process- 327, 2003. arXiv:quant-ph/0303081v1
ing, Bangalore India, December 1984, pp 175-179. [20] D. Kielpinski, C. Monroe, and D. J. Wineland, Nature
[4] A. Bernasconi, Harmonic analysis and Boolean function 417, 709 (2002).
complexity, Calcolo 35 (1998), no. 3, 149–186. [21] H. Krovi, T. A. Brun, Phys. Rev. A 73, 032341 (2006).
[5] A. Bernasconi, B. Codenotti, Spectral analysis of arXiv:quant-ph/0510136v1
Boolean functions as a graph eigenvalue problem, IEEE [22] R. J. Lechner, Harmonic analysis of switching functions.
Trans. Comput. 48 (1999), no. 3, 345–351. In: Recent development in switching theory, London:
[6] B. B. Blinov, D. L. Moehring, L.-M. Duan, and C. Mon- Academic Press, 1971, pp. 122–229.
roe, Nature 428, 153 (2004). [23] P. Lo, S. Rajaram, D. Schepens, D. Sullivan, C. Ta-
[7] S. Bose, Contemporary Physics, Vol. 48 (1), pp. 13-30, mon, J. Ward, Mixing of Quantum Walk on Circu-
2007. arXiv:0802.1224v1 [cond-mat.other] lant Bunkbeds, Quantum Info. Comp., Vol. 6, No. 4&5
[8] A. M. Childs, E. Farhi, S. Gutmann, An example of the (2006), pages 370-381. arXiv:quant-ph/0509059v1
difference between quantum and classical random walks, [24] L. Lovász, Spectra of Graphs with Transitive Groups,
Quantum Inf. Process. 1, 35 (2002). quant-ph/0103020 Period. Math. Hungar., 6(2):191–195, 1975.
[9] M. Christandl, N. Datta, A. Ekert and A. J. [25] Y. Mansour, Learning Boolean functions via the Fourier
Landahl, Phys. Rev. Lett. 92, 187902 (2004). transform, Theoretical Advances in Neural Computing
arXiv:quant-ph/0309131v2 and Learning. Dordrecht: Kluwer Acad. 1994.
[10] L. Chu, Colouring Cayley Graphs, Master Thesis, Uni- [26] C. Moore and A. Russell, in Proc. RANDOM’02.
versity of Waterloo, 2004. arXiv:quant-ph/0104137v1
http://etd.uwaterloo.ca/etd/l4chu2005.pdf [27] O. Muelken, A. Blumen, Phys. Rev. E 71, 036128 (2005).
[11] A. K. Ekert, Phys. Rev. Lett. 67, 661 (1991). arXiv:quant-ph/0502004v1
[12] C. Facer, J. Twamley, J. D. Cresser, Phys. Rev. A 77, [28] N. Saxena, S. Severini, I. Shparlinski, Int.
012334 (2008). arXiv:0706.3821v1 [quant-ph] J. Quantum Information, 2007, v.5, 417-430.
[13] D. L. Feder, Perfect quantum state transfer with spinor arXiv:quant-ph/0703236v1
bosons on weighted graphs. arXiv:quant-ph/0606065v1 [29] A. Volta, O. Muelken, A. Blumen, J. Phys. A 39, 14997
[14] H. Gerhardt, J. Watrous, Proc. RANDOM’03. (2006). arXiv:quant-ph/0610212v1
arXiv:quant-ph/0305182v1 [30] G. M. Ziegler, Coloring Hamming graphs, optimal binary
[15] C. Godsil, Periodic Graphs. arXiv:0806.2074 [math.CO]. codes, and the 0/1-Borsuk problem in low dimensions.
[16] J. L. Hennessy and D. A. Patterson, Computer Architec- Computational discrete mathematics, 159–171, Lecture
ture: A Quantitative Approach, Third Edition, Morgan Notes in Comput. Sci., 2122, Springer, Berlin, 2001.
Kaufmann Publishers, Inc., 2003.
[17] M. A. Jafarizadeh, R. Sufiani, Perfect state transfer