1

Cluster State Quantum Computing

CIS:410/510 Final Report, Spring 2016

Dileep V. Reddy, Mayra Amezcua, Zach Schmidt

[email protected], [email protected], [email protected]

Abstract— Any quantum computation can be performed via se- A. Preparation of linear cluster state
quences of one-qubit measurements on a specific type of initially entan-
gled state – the cluster state. Each computational step is a projective Intuitively, a cluster state can be thought of as a graph
measurement that destroys a quantum state, leaving a final state that where every vertex represents a qubit, and every edge rep-
relies on the outcomes of earlier computations. The model of interest resents the application of a Cz gate to both adjacent vertices.
is the one-way quantum computer which is based on this measurement
scheme. This paper will present background regarding computation us-
ing only measurements, a brief introduction into the preparation of clus-
ter states, a discussion of one way quantum computers (1WQC), and the
computational power of various configurations of a 1WQC, and will end
with an overview of physical implementations.

I. BACKGROUND
Over the past few decades, advances in science and tech-
nology have greatly contributed to the development of mod-
ern computers. While these computers are efficient and con-
venient for everyday needs, they fail at certain computational
tasks. Instead, quantum computers promise faster large scale
factorization and database searches that are intractable for
their classical counterparts. The first quantum computer de-
signs were based off of classical models; sequences of one-
and multi-qubit gate operations are performed on chosen Fig. 1. Figure from [3], showing representative 2D cluster shapes. The ver-
quantum bits and a final measurement would convert quan- tices are qubits with integer indices, and the edges indicate entanglement
connectivity between select neighbors.
tum information into classical bits. A new model, proposed
by Briegel and Raussendorf [1], demonstrates that quantum A cluster state can be represented as a graph G = (N, E),
computation can be achieved by using single qubit measure- where the n ∈ N is a qubit and e ∈ E is the application of a
ments as computational steps. This so-called cluster model Controlled-Z (Cz ) gate, where:
or one-way quantum computer (1WQC) relies on an entan-
gled state of a large number of qubits or cluster state as the
 
1 0 0 0
resource. The fascinating feature about 1WQC is that they 0 1 0 0 
have no classical analogues and probe into new territory in Cz = 0 0 1 0 

regards to entanglement and measurements. 0 0 0 −1
II. C LUSTER S TATES A linear cluster state is one where degree(n) ≤ 2∀n ∈ N .
Consider a set of qubits C labeled by an integer index, that
are distributed in some lattice such that every qubit can be
said to have adjacent neighbors. For these to collectively 1 2 3 4
form a cluster state, their quantum mechanical state would 4-Node Linear Cluster State
be characterized by the set of eigenvalue equations [2],
A method to prepare such a cluster state is given in [4],
Ka |ΦiC = κ |ΦiC (1) consisting of “cascading” Cz gates on n qubits as follows:
(a) (γ)
N
for a family of operators Ka = X γ∈Γ(a) Z , a ∈ C, 1 2 3
|+i1 •
where Γ(a) is the set of indices of all qubits in the “adjacent
neighborhood” of a. The matrix X (a) is used to denote an |+i2 CZ •
X operation on qubit-a, and so on. The eigenvalue κ = ±1
|+i3 CZ •
is determined by the specific occupation pattern of the neigh-
boring sites. |+i4 CZ
 2

We can then analyze the state of the qubits at each of the It is important to emphasize that the order in which the Cz
dotted lines: gates are applied to grow the cluster state is irrelevant, as all
of these pair-wise operations commute. This feature will be
1:
exploited later when discussing parallelizability.
|0i1 |+i2 + |1i1 |−i2
 
√ |+i3 |+i4
2 III. T HE E FFECTS OF M EASUREMENT ON A C LUSTER
|+i1 |0i2 + |−i1 |1i2
 
≡ √ |+i3 |+i4
S TATE
2
As is clear from the form of the expressions of all cluster
2: states illustrated thus far, measuring any node in the com-
|+i1 |0i2 |+i3 + |−i1 |1i2 |−i3
 
√ |+i4 putational basis severs it from the remaining graph by cut-
2 ting all of it’s edges with it’s neighboring nodes. Should the
|0i1 |+i2 + |1i1 |−i2 |0i1 |−i2 + |1i1 |+i2
 
≡ |0i3 + |1i3 |+i4 outcome of said measurement be 1, then a Z gate/transform
2 2 gets applied to all of it’s erstwhile neighbors in the leftover
3: cluster state. Thus, a large cluster state can be arbitrarily

|+i1 |0i2 |+i3 + |−i1 |1i2 |−i3

trimmed, split, and/or reshaped by removing qubits from the
|0i4
2 cluster. This is accomplished by measuring the target qubit in
|+i1 |0i2 |−i3 + |−i1 |1i2 |+i3
 
+ |1i4 the computational basis, and performing appropriate unitary
2
rotations on its former neighbors based on the measurement
(|+i1 |0i2 + |−i1 |1i2 ) |0i3 |+i4 + (|+i1 |0i2 − |−i1 |1i2 ) |1i3 |−i4
≡ outcome.
2
The effect of an X-measurement (i.e., a computational ba-
The action of the Cz gate in the computational basis can
sis measurement following a Hadamard transformation) on
be seen to be |x, yi → (−1)xy |x, yi. Cluster states of ar-
any node of the cluster state is much more involved. This is
bitrary shape and connectivity can similarly be prepared via
best illustrated when demonstrating the use of a linear clus-
the recursive use of the Hadamard gate and two-qubit fusion
ter state as a wire for quantum information. For this exer-
operations [5], [3].
cise, we start with a linear cluster state with three nodes (la-
B. Preparation of T-shaped cluster state beled 1, 2, and 3). A single qubit of quantum information
|ψi = α |0i + β |1i is stored in a physical qubit labeled 0 as
A cluster state without the limitation on the degree of a illustrated below.
node allows us to build nonlinear cluster states:

1 2 3 |ψi0

0 1 2 3
4
(0,1) (0) (1) (2)
Gate Cz , followed by measurements MX , MX , & MX .
4-Node T-Shaped Cluster State

The circuit creating this cluster state will look as follows: To teleport the state |ψi to physical qubit number 3, we
must first supply the quantum information to the “wire.” This
|+i1 • is achieved by applying a Cz gate between physical qubits 0
|+i2 CZ • • and 1. Using |LCi123 to denote the linear cluster state, we
have
|+i3 CZ
|+i4 CZ Cz(0,1) |ψi0 ⊗ |LCi123
The state of the qubits after the application of the first two 1
Cz gates is identical to the linear case, and the state after the = √ [α |0i0 |+i1 |0i2 |+i3 + β |1i0 |−i1 |0i2 |+i3
2
last Cz (at the dotted line) is given by:
α |0i0 |−i1 |1i2 |−i3 + β |1i0 |+i1 |1i2 |−i3 ]

|+i1 |0i2 |+i3 |+i4 + |−i1 |1i2 |−i3 |−i4 Following this, we perform X-measurements on physi-
√ cal qubits 0, 1, and 2 in that order. Let us denote an X-
2 (j)
1

|+i1 |0i2 |+i3 + |−i1 |1i2 |−i3
 measurement operation on the j th -node with MX , and let
≡√ √ |0i4 the outcome of any measurement on the same node be mj .
2 2
   Then, the end result of these operations is the state
|+i1 |0i2 |+i3 − |−i1 |1i2 |−i3
+ √ |1i4
2 X m2 Z m1 X m0 H |ψi3 , mj ∈ {0, 1}.
 3

The quantum information |ψi has successfully been tele- ray of entangled qubits, information propagates horizontally
ported to physical qubit 3, up to application of Pauli opera- through a row of qubits while vertical qubit neighbors are
tors depending on the measurement outcomes. Since the left- used for two-qubit gates. Similarly, three-dimensional clus-
over/extra Pauli operators do not commute, these measure- ters can be used to implement topologically protected gates
ments had to have been carried out in a specific order. How- [7], where the gate function only depends upon the way “con-
(2,3) nected defects” are wound around one another, but not on the
ever, the operation Cz , which was employed to grow the
(0,1)
linear cluster state, commutes with Cz , as well as mea- details of their shape. This degree of freedom affords the de-
(0/1)
surements MX . Thus, further links to the chain can be sign some fault tolerance.
grown as earlier links are being subjected to measurements. The basic principle of cluster-state quantum computa-
This aids in parallelizability, as well as physical implementa- tion is to effectively enact arbitrary quantum circuits onto
tion of cluster state quantum computing schemes. qubits storing quantum information by performing single-
Briegel and Raussendorf show that any quantum logic cir- qubit transformations and measurements on a pre-formed
cuit can be implemented on a cluster state, which demon- cluster state whose graph representation bears topological
strates universality of the proposed scheme [1]. Nielsen [6] similarities to the circuit in question. These measurements
extended this result to no longer require coherent dynamics, result in destruction of the node-qubits of the cluster state,
instead relying on a method to teleport quantum gates, and he and hence are irreversible. The outcomes of these measure-
provided a concise algorithm to accomplish this. ments need to be tracked or fed forward to influence future
operations along certain layers of the cluster, as will be illus-
IV. O NE -WAY Q UANTUM C OMPUTATION trated later in this article.

All quantum computation schemes may be characterized

by some combination of state preparation, unitary trans- A E
formation of said states, and measurements on the same. C J
Human-usable computational tasks necessarily require both
B D G
input and final output to be classical information. The classi-
F
cal input information can influence the quantum computation
K
in choice of initial states, the choice of unitary transforms
Arbitrary quantum circuit involving unitary operations on 3 qubits.
(i.e., algorithm), and the choice of measurement bases. The
output is always a classical function of the measurement out-
comes. In typical models for quantum computation, entire
algorithms are implemented as a sequence of unitary transfor-
mations on a prepared quantum state (stored in qubits) of size
appropriate to the problem, with a round of measurements
as the final step. In such models, the unitary transformation
stage is completely reversible. The splitting of the effective
unitary matrix into sequential steps can be arbitrary and en-
tirely dependent on physical hardware limitations. There is Single-qubit measurements on this cluster state is equivalent
no correspondence with “computational steps” or “clock cy-
to the topologically similar circuit above.
cles” in the classical sense, as the quantum state of the com-
puter in the midst of the unitary stages is inaccessible for di-
agnosis or debugging purposes. Any leakage of information A. Gates through teleportation
into computer memory or environment constitutes decoher- Quantum teleportation is a procedure by which quantum
ence, and will introduce errors in the computation. information can be transferred from one point to another via
One-way quantum computation, on the other hand, re- two classical bits of information if the sender and received
volves around single qubit measurements as a progression previously shared an entangled state. It is useful for quantum
of computational steps. Measurements are a crucial compo- computation however, this approach uses an entangled state
nent to quantum information processing because they irre- as a resource but if that state has some error then teleporta-
versibly destroy a quantum state. Entanglement, on the other tion fails. Gottesman and Chuang [8] first showed that if the
hand, will ensure that the state of the final qubit relies on the entangled state |ψi can be replaced by U |ψi, such that U
outcomes of preceding measurements. Given a cluster state, is a non-trivial quantum operation. The corresponding output
a series one-qubit measurements can be performed at each U |αi is in the initial state |ψi but with additional single-qubit
qubit to implement a quantum gate [4]. The unidirectionality Pauli operations X, Y , or Z. By simply reversing the Pauli
of cluster state computation is inherent, due to the fact that operators the original state can be reconstructed. This tele-
quantum information cannot be accurately recovered once a portation scheme can be useful in applying other gates that
measurement has been made. Consider a two-dimensional ar- are non-trivial.
 4

1 α1
−iα iα
!
HZα = √
1 e 2 e2 X 0 |ψ4 i = |ψ5 i = √ cos ( )e−iα2 /2 |+i3
−iα iα 2 2
2 e 2 −e 2
i α1 iα2 /2
− √ sin ( )e |−i3
As demonstrated in the previous section, by applying 2 2
single-qubit gates we can teleport one state from one side of a
cluster state to the other. In this section we demonstrate how The output of the circuit is X m2 HZ±α2 X m1 HZα1 |+i3 ,
to apply the HZα gate through teleportation. where m1 and m2 are the outputs for the first and sec-
ond qubits. Analyzing this output a bit further we see that
HX m1 = Z m1 H and Z±α2 X m1 = X m1 Zα2 . Using this
|+i1 • HZα1 the state can be rewritten as:

|+i2 CZ • HZ±α2 X m2 Z m1 HZα2 HZα1 |+i3

|+i3 CZ This is equivalent to the output of the conventional single-

qubit quantum circuit up to a known Pauli matrix. Thus it is
In this example we will apply two consecutive HZα gates easy to model gate application through teleportation.
to our cluster to simulate two gates on a single qubit. We We observe that in the circuit given below the two high-
will demonstrate that the outcome is the same except for an lighted boxes are both one-bit teleportations. Because the Cz
additional two Pauli operations. For this problem we prepare gate commutes with HZα we can perform one teleportation
a three node cluster state given by: procedure and then the second one or vice-versa. This is ad-
vantageous because it means that we can build the cluster as
Cz(1,2) Cz(2,3) |+i1 |+i2 |+i3 = |ψ1 i we go.
1
= √ |0i1 (|0i2 |+i3 + |1i2 |−i3 )
2
1 |+i1 • HZα1
+ √ |1i1 (|0i2 |+i3 − |1i2 |−i3 )
2
|+i2 CZ • HZ±α2
First we will apply HZα1 and measure in the computa-
tional basis. |+i3 CZ
e−iα1 /2
 
|0i2 |+i3 + |1i2 |−i3
HZα1 |ψ1 i = |+i1 √
2 2 B. Applying a 2-qubit gate via 2D cluster state
iα1 /2
 
e |0i2 |+i3 − |1i2 |−i3
+ |−i1 √ = |ψ2 i The ability to implement gates of the form HZα , and any
2 2 2-qubit controlled-phase gate, and to prepare arrays of |+i
states as inputs, constitute a set of resources that is univer-
For convenience we assume that we measure the state |0i. sal for quantum computation [9]. In this subsection, we will
demonstrate the use of two-dimensional cluster states to im-
1 plement a Cz gate between two input kets bearing quantum
X 0 |ψ2 i = √ cos (α1 /2) |0i2 |+i3 information |φA iA |φB iB encoded in physical qubits labeled
2
‘A’ and ‘B.’ Here, |φj i := αj |0i+βj |1i. For the scheme, we
i
− √ sin (α1 /2) |1i2 |3i3 = |ψ3 i will use an I-shaped cluster states with six nodes, as shown in
2 the figure.
We apply one final HZα gate.

HZ±α2 |ψ3 i = |ψ4 i |φA i A 1 2 3

1  α1 −iα2 /2 α1 
√ |0i2 cos ( )e |+i3 − i sin ( )eiα2 /2 |−i3
2 2 2
1  α1 α1  |φB i B 5 4 6
+ √ |1i2 cos ( )e−iα2 /2 |+i3 + i sin ( )eiα2 /2 |−i3
2 2 2
(A,1) (B,5)
The final output state we get is: Apply Cz and Cz to input quantum information into cluster state.
 5

The procedure would be to first entangle the input quantum C. Commutations and parallelizability
information qubits into the cluster state via the application of
Cz gates. Then we perform single qubit measurements on The previous subsections demonstrated that although mea-
all qubits but numbers 3 and 6. Using the Z-measurement surements in a specific layer (an ordered set of sequentially
property of nodes on cluster states, we can pick node 4 as connected nodes) of a cluster state need to occur in a specific
a representative anchor and write down the total state of the order, operations on different layers commute, and can there-
current system as fore be interspersed. This, combined with the ability to delay
the growth of the cluster state to occur just before the mea-
surements catch up with the remaining nodes, offers several
1 possibilities for process parallelization.
√ [|LCi123 |0i4 |+i5 |+i6 + Z2 |LCi123 |0i4 |−i5 |−i6 ]
2
⊗ |φA iA |φB iB ,

where Z2 is being applied on qubit 2.

The single qubit measurements have to be performed
(A) (1) (2) (B) (5) (4)
in the order (MX , MX , MX ) and (MX , MB , MX ).
These two subsets are parallelizable, and one could per-
(A) (B)
form the measurements {MX , MX } in parallel, and then
(1) (5)
{MX , MX }, and so on. But here, we will introduce the
concept of layers of measurement, by performing all bottom
row operations at once (including input entanglement), and
then working on the top row.
(B,5)
Enacting the gate Cz on the initial state, followed by
(B) (5) (4)
measurements MX , MB , MX yields random outcomes Fig. 2. The controlled-phase operations commute with unitaries and mea-
mB , m5 , m4 ∈ {0, 1} and gives us the state surements on other parts of the cluster state. This allows one to conserve
and reuse physical resources, as well as maintain coherence on smaller
cluster sizes at any given time.
(2,6)
X6m4 Z6m5 X6mB CZ |LCi123 |φB i6 ⊗ |φA iA .
Indeed, in the 2-qubit gate example, the quantum informa-
Note that even though the original cluster state had an edge tion in one of the qubits was put into the system first, followed
between nodes 2 and 4 in the graph representation, a measure- by some measurement-based teleportation, and the other in-
ment on qubit 4 did not sever the link between node 6 and the put qubit was brought in later. We effectively managed to en-
linear chain |LCi123 , due to the measurements being in the X tangle the two qubits using only single-qubit measurements.
basis. Now, if we follow the above operations by application
(A,1) (A) (1) (2)
of the CZ gate, and the measurements MX , MX , MX
in that order, we get the random results mA , m1 , m2 ∈ {0, 1}
and the final state

(6,3)
X3m2 Z3m1 X3mA H3 X6m4 Z6m5 X6mB H6 CX |φA i3 |φB i6 ,

(j,k)
where CX is the controlled-NOT gate. All single-qubit op-
erations on qubit 3 commute with single-qubit operations on
(6,3)
qubit 6. Using the fact that Hj2 = Ij , as well as CX ≡
(3,6)
(H3 ⊗ I6 )CZ (H3 ⊗ I6 ), the final state is equivalently

Fig. 3. X-measurements do not sever “vertical” links despite destruction of

(3,6) qubits. This allows different linear layers to be processed in any order.
X3m2 X6m4 Z3m1 Z6m5 X3mA X6m4 H6 CZ H3 |φA i3 |φB i6 ,

which is the desired result, up to overall Pauli transformations This seems to indicate that all multi-qubit interactions can
on the individual qubits. Thus, we have proven that a univer- be pre-computed by setting the topological layout of the
sal set of quantum operations can be implemented using the graph of the cluster state before any of the quantum infor-
cluster state model. mation has even been introduced into the system.
 6

Fig. 4. Figure from [10]. Topologically protected gates as realized in three-dimensional cluster states. (a) An individual encoded CNOT gate with control c
and target t. (b) The preparation of logical qubits in the {|0i, |1i} (i.e., Z-) and the {|±i} (i.e., X-) bases. The line-like structures are connected defect
sites (nodes measured in the Z-basis, denoted by set ‘D’) embedded in a 3D lattice cluster state, surrounded by sites belonging to set ‘V’ (measured in
the X-basis). The gate function only depends upon the way the defect lines are wound around one another.

Fig. 5. Figure from [10]. (a) Another version of the CNOT gate topologically equivalent to that shown in figure 4(a). (b) Equivalent circuit, representing a
CNOT gate between the control and target qubit.

D. 3D cluster states and topological fault tolerance V. C OMPUTATIONAL P OWER AND C OMPLEXITY
The spacial layout of the graph representation of the clus-
While 2D cluster states are sufficient for fully realizing
ter state plays a role in the computational power of that state.
any quantum computation, Raussendorf et al [10] have ex-
If a cluster state can be prepared linearly via the cascading Cz
ploited a correspondence between quantum gates, quantum
technique mentioned above, it can be represented as a “one-
correlations, and surfaces to propose a topological model
dimensional” graph (i.e., some graph G = (V, E), ∀v ∈ V ,
for cluster-state quantum computation. The method affords
deg(v) ≤ 2). Operations on a linearly prepared cluster
them a (fault-tolerance) threshold estimate of 0.75% for each
state can be efficiently simulated on a classical computer in
source in an error model with preparation, gate, storage and
O(n logc (1/n)), where n is the initial number of qubits, and
measurement errors, with a poly-logarithmic multiplicative
c is the cost of floating point multiplication [12]. Though the
overhead in the circuit size. While a full detailed description
author consequently dismisses linearly prepared cluster states
of this approach is beyond the scope of this article, here we
as a substrate for quantum computation, it would be interest-
present a brief overview.
ing to know which class of problems they would be able to
In their scheme, a 3D cluster state consisting of a lattice solve.
of qubits is ‘carved out’ with 1D line-like defects via Z-
measurements on connected physical qubits. This results in a A. Gate array reductions
nontrivial cluster topology in which a fault-tolerant quantum With only a bit of construction, it can be seen that the clus-
circuit is embedded (figure 4). These line defects essentially ter state model is polynomially reducible to the gate array
simulate anyons [11], and the ways they knot and loop around model, and the converse is also true. To see this, we first need
each other effectively obey non-Abelian braiding statistics, to create a definition of the standard gate array model:
allowing one to simulate gates (figure 5). The fault tolerance
comes from the topological invariance of the structures to lo- 1. All measurements take place at the end of the circuit
cal perturbations to the line defects (such as specific path and 2. All measurements take place in the computational basis
length). One of the dimensions of this 3D representation can
be mapped to ‘simulated time,’ which is a necessary facet of We can now generalize this definition to allow measure-
all computation that maps inputs to outputs. This mapping ments along the way with subsequent choices of gates and
can be literal when using physical 2D cluster states. measurements being allowed to depend on earlier measure-
 7

ment outcomes. Intuitively, what we will do is add an ancilla A.2 Reducing gate array to cluster state
bit to all measurements which take place before the end, and To see how an arbitrary gate array can be converted to the
perform controlled operations with this ancilla to return to the cluster state model, we first need some universal gate set G
standard definition. Specifically, for all measurements in the (for information on what this set might contain, see [13]). For
{U |0i , U |1i} basis, we add an ancillary qubit A, (initially each gate within this set, we can specify some number which
in state |0i) and replace the measurement with an application represents the number of measurements that a 1WQC would
of U † to B followed by applying CX to BA as shown: have to perform to get the same outcome. These numbers
can then be used to partially order a set, allowing us to pick
K, which represents the maximum number of measurements
B required to simulate any single gate from the set. From this,
we can see that even if some circuit contained only this “most
Fig. 6. Some non-standard circuit expensive” gate, the number of additional qubits and gates
would be a factor of K (polynomial).

B U† • B. Quantum layers
The above reduction strategy provides a nice way of cat-
|0i X egorizing a given quantum algorithm into its classical and
quantum parts:
Fig. 7. An equivalent standard circuit • The classical parts are those that are done serially (i.e. the
decision of future gates based on measurement outcomes)
Now, any future gates that depended on the measurement • The quantum parts are those that can be done in parallel
outcome are replaced by a corresponding controlled opera- From this, we can see that the cluster state model exempli-
tion, controlled by the state of A. It is therefore clear that this fies both parts, but the gate array model only does “quantum
process converts any non-standard circuit to the standard gate parts”. With this idea, we can define the notion of layers by
array model. saying that a quantum process with K layers is one where K
gates are operating in parallel. The above suggests the fol-
A.1 Reducing cluster state to gate array
lowing conjecture [9]:
The above technique shows how a cluster state circuit is Conjecture: Any polynomial time quantum algorithm can be
converted into an equivalent (standard) gate array. In addi- implemented with only O(log n) quantum layers interspersed
tion to building the required cluster state using an array of with polynomial time classical computations.
CZ gates acting on |+i states, we introduce an ancilla A for This conjecture remains unproven in general, but it has
each 1-qubit measurement. For each M (θ) measurement we been shown to hold for Shor’s algorithm [14].
introduce an extra gate W † (θ) which transforms the M (θ)
basis to the standard basis. C. Cluster graphs as an analysis tool
Outside of the physical implementation considerations,
cluster state model isomorphisms offer a new analysis tool
|+i • M (θ)
that disentangles the quintessential influence of quantum for-
malism on the complexity class of various algorithms.
|+i Z We have already shown in section IV, how any quantum
algorithm that is expressible via a quantum gate-array cir-
Fig. 8. A two qubit cluster state cuit can equivalently be computed via single-qubit operations
and measurements on a cluster state represented by a graph
whose connectivity is topologically similar to the circuit di-
|0i X agram. This equivalence is purely geometric and is indepen-
dent of the specific gates being applied (those are determined
|+i • W † (θ) • by the choice of measurements on the cluster state). This
allows us to reduce entire classes of algorithms to specific
|+i Z types of graphs with designated input and output nodes for
state-preparation and final-result measurements. We conjec-
Fig. 9. An equivalent cluster state with ancilla ture that the size of this graph has a bearing on the compu-
tational time for the entire class of algorithms. Furthermore,
Thus, to convert the measurement-based cluster state the multi-dimensional connectivity of said graphs embodies
model to the gate array model requires polynomially more entangling operations, and serves to explicitly quantify any
gates and qubits (one per measurement). gains in complexity quantum methods offer over classical
 8

ones. And finally, the connectivity and feed-forward paths rations. The durations are dictated by the natural Hamilto-
allows one to define modular operations that are independent nian frequencies of the energy states being coupled, and the
of each other, and can help exploit all avenues for paralleliza- pulses are designated names such as π-pulses in the literature,
tion more effectively. depending on their effect in the Hilbert space.

VI. P HYSICAL IMPLEMENTATIONS

Although the chief inspiration for cluster-state quantum
computation was the ability to enact gates via teleportation,
another motivation proved to be the extant experimental re-
alizations of material qubits arranged in 2D arrays in physi-
cal space; be they cold atoms in optical lattices [15], or 2D
ion traps [16], or other stationary qubits embedded in mate-
rial substrates (quantum dots, superconducting qubits). The
geometry of such systems encourages designs involving pro-
gramming a quantum circuit into the device by literally “etch-
ing” the circuit diagram onto the 2D qubit array.
Given the prevalence of experiments on 2D lattices of Fig. 10. Figure from [17]. Two coupled quantum dots with one valence
physical qubits, it is natural to ask if it were possible to anneal electron per dot (both confined to the xy plane). Their spins are denoted
by S~1 and S~2 . The magnetic field B is along the z-axis and the electric
such a system in an imposed, time-invariant potential such
field E is along the x-axis. The interaction term of the Hamiltonian
that the natural ground state of the corresponding Hamilto- ~1 · S
~2 , where the exchange coupling j between
reads Hs (t) = J(t)S
nian would be a desired cluster state. Nielsen [12] provides the spins is a function of B, E, and the interdot distance 2a.
an argument for why this is not possible for systems that
have only two-body interactions (up to first order). The “dis- A common feature of all such physical systems is the pro-
tance” between any cluster state and the energy eigenstates of fusion of degrees of freedom available for manipulation. The
a Hamiltonian with two-body interactions of the connection- storage of quantum information typically occurs within well
intended qubit pairs can be shown to be bounded below by chosen, two-level subspaces of the full Hilbert space, which
a constant independent of anything except the Hilbert space are local (non-interacting) and relatively decoherence free.
dimension (i.e., the cluster size). All multi-qubit operations, however, occur in higher-energy,
nonlocal regions of the Hilbert space. The storage subspaces
A. Two qubit operations on non-photonic matter qubits and the interaction subspaces are temporarily coupled to each
The above argument however, does not forbid the applica- other via externally applied potentials for the necessary dura-
tion of time-varying interaction potentials in order to facili- tion for specific types of interactions to occur.
tate inter-qubit entangling operations. In matter qubits that
have physical proximity, this can be achieved by applying ex- B. Nondeterministic two-qubit gates with linear optics
ternal electric and magnetic fields. For example, two neigh- Matter qubits are promising candidates for quantum com-
boring quantum dots with one valence electron each, can be puting but they are subject to decoherence. Photons, on the
coupled via transverse electric and magnetic fields to imple- other hand, do no suffer from decoherence effects because
ment a quantum XOR gate via an exchange mechanism [17]. they interact weakly with each other. However, multi-qubit
Figure 10 illustrates the essential mechanism. The coupling operations on photons are much harder because of this weak
fields are pulsed such that the net integral effect of the inter- interaction. A new approach fuses photon states into entan-
action results in a “swap” operator being applied. The tem- gled states via projective measurements. With this method,
poral duration of the unitary time-evolution operator associ- nondeterministic multi-qubit gates can be applied to photons.
ated with the interaction naturally depends on the physical KLM [20] showed that this method can be used for efficient
parameters in question, namely, the quantum dot size, mutual quantum computation using linear optical elements only. It
distance between them, peak field strength, etc. can be carried over to preparation of cluster states.
For trapped ions and neutral atoms, two-qubit gates can be The second section of this paper discussed how to construct
enacted by coupling internal degrees of freedom in individual a T-shaped cluster state by entangling qubits 2 and 4. A lin-
qubits (usually electron energy levels) with multi-qubit vibra- ear cluster state can be grown on qubit 4 to create a 2D clus-
tional modes. For ions [18], the interaction is mediated by the ter state. However, the Cz -gate cannot be easily applied to
Coulomb potential, whereas in neutral atoms [19], this could photons with high fidelity. Browne and Rudolph [5] first pro-
be due to dipole-dipole interactions between atoms excited to posed a scheme to build L-shaped clusters from polarization-
low-lying Rydberg states in constant electric fields. The state encoded photonic qubits using two different types of “fu-
read-in, single-qubit manipulation, and state read-out are all sion” measurements. This relies on the fact that polarization-
carried out using coherent laser pulses of set power and du- entangled two-photon states are identical to two-node clus-
 9

fusion between photons in paths 2 and 3. The final entangled

state was measured using photodetectors and verified with the
three particle Bell inequality.

Fig. 11. Figure from [5]. (a) Connceting the end qubits of two linear cluster
states using type-I fusion. (b) Middle qubits are fused to create a cross
in the cluster state.

Fig. 13. Figure from [21]. Two pairs of entangled photons are created
via spontaneous parametric down conversion. Opposite pairs are sent
through polarizing beam splitters to do type-I fusion and then detected
to verify the state. HWP, Half-wave plate; PBS, polarizing beamsplitter;
BBO, beta barium borate crystal.

C. Continuous-variable cluster states

While single-photons have sufficient degrees of freedom to
Fig. 12. Figure from [3]. (a) Hamdamard gate applied to qubits 2 and 3. (b) afford storage of quantum information, the difficultly of caus-
The operations prepare this box state which entangles qubits 1 and 4. (c)
ing multiple photons to efficiently interact with each other
Once the box is formed a measurement is made on qubit 2 to break the
bonds and create the L-shaped structure starting from a linear chain. limits their potential. However, the interaction barrier can be
circumvented by relying on cavities, and the quadrature space
of optical modes for information storage. Squeezed coherent
ter states up to local unitaries, and are a readily preparable states of light [22] are specific linear combinations of photon-
resource in quantum optics. The L-shaped structures would number states in a designated set of optical modes. In quadra-
provide an easy way to construct 2D arrays. ture space, they are expressed as Gaussian functions with well
In 2006, Gilbert et. al. [3] proposed a simpler scheme defined axes of squeezing. The widths of the major and minor
which used local unitaries and type-I fusion operations alone, axes refer to the quantum uncertainty in two canonical (non-
to create a two-dimensional cluster state. By only utilizing commuting) quadrature expectation values. The uncertainties
type-I fusion Gilbert et. al. demonstrated that the cost of are saturated to the Heisenberg limit, but the asymmetry re-
creating an entangled state was reduced from 34 bonds to 2 sults in one quadrature uncertainty having been drastically
bonds. Type-I fusion takes the end qubits of two separate lin- reduced at the expense of the other.
ear clusters and makes a projective measurement. If the oper- Quadratures make for a perfectly acceptable Hilbert space
ations succeeds (with probability 0.5, subject to efficient pho- of infinite dimension. They can be entangled, and by ex-
ton detection and number resolution), the two clusters would tension, can be used as a quantum resource to teleport other
be joined into a single linear chain with (n+m−1) qubits. If continuous variable states. Multiple optical modes are gener-
the fusion operation were to fail, the size of each participat- ally squeezed via a process of spontaneous parametric down
ing cluster would be reduced by one qubit (or split, if fusion conversion (SPDC), where in a coherent laser pump field
was attempted in the middle), and the fusion will have to be of a higher frequency interacts with a bulk, nonlinear ma-
reattempted. terial (typically a down-conversion χ(2) crystal or crystalline
The first photonic cluster state was created by Zhang et. waveguide) to produce pairs of correlated photons at a lower
al. in 2005 [21]. This three qubit linear cluster was created frequencies. If this process were to occur inside an optical
by generating pairs of entangled photons and using Type-I cavity whose modes are resonant at the emitted field frequen-
 10

Fig. 14. Figure from [23]. Yokoyama et al use (a) two separate optical parametric oscillators (OPOs) as sources of squeezed pulses of light, and mix them
in an unbalanced Mach-Zehnder interferometer to produce (b) a long, dual-rail style squeezed state. 50:50 BS, balanced beamsplitter; HD, homodyne
detector; LO, local oscillator.

cies, then the interaction is no longer perturbative, and will broadband SPDC process can squeeze many of them in pairs.
result in significant squeezing. Such sources are called opti- By using two pumps at different polarizations and slightly
cal parametric oscillators (OPO). offset frequencies, Chen et al [24] managed to generate the
same lengthy cluster state, with the time-multiplexed axis
having been replaced by a discrete frequency variable index
in some spiral magnitude order (figute 15).
Continuous variable cluster states cannot be used for quan-
tum computation if all our manipulation and measurement
operations are so-called “Gaussian” in nature [25]. For op-
tical quadrature squeezed cluster states, measurement in the
photon-number basis is a non-Gaussian operation, and might
be sufficient to break the constraint. Such optical cluster
states, however, are by definition travelling at the speed of
light, and barring a sophisticated quantum memory solution,
will require computational steps to occur on the fly using ul-
trafast homodyne detection setups.

VII. C ONCLUSIONS AND ACKNOWLEDGMENTS

Fig. 15. Figure from [24]. Chen et al pumped an OPO with two pumps with In this article, we have defined quantum cluster states, and
perpendicular polarizations (Z and Y) and offset frequencies to couple have motivated their use for performing universal quantum
th squeezing of intertwined pairs of (a) optical cavity-resonant modes. computation using single-qubit measurements. We briefly re-
These are then mixed on a single beamsplitter to produce (b) a dual-rail
style long cluster state. The horizontal axis here is the cavity resonant
viewed their preparation, and the effects of measurements on
frequency label ordered in a spiral permutation. them. We demonstrated their equivalence to traditional gate
array models of quantum computation and enumerated cer-
Multiple optical modes (defined by beam-momentum vec- tain advantages, namely, parallelization, and use as an anal-
tors) may then be deterministically entangled in quadra- ysis tool for complexity studies. We further presented some
ture space via use of linear optical elements such as simple experimental implementations and caveats in various physi-
beamsplitters. Yokoyama et al [23] exploited this to cre- cal systems.
ate a lengthy continuous variable cluster state encoded in We would like to thank Prof. Xiaodi Wu from the de-
quadratue space (figure 14). An optical cavity has multiple partment of computer and informatin science, University of
resonant modes at different frequencies, and a sufficiently Oregon, for offering the course on quantum information, and
 11

helping us grasp the basic concepts in this field. We thank optical multiqubit cluster states from bell states,” Physical Review A,
Robert Rogers, Abe Pauls, Thomas Sylwester, Raleigh Fos- vol. 73, no. 2, pp. 022330, 2006.
[22] A. I. Lvovsky, “Squeezed light,” arXiv preprint quant-ph/1401.4118,
ter, and Bijan Shahir for helpful discussions during our sem- 2001.
inar presentations on this topic. All factual errors in this [23] S. Yohoyama, R. Ukai, S. C. Armstrong, C. Sornphiphatphong, T. Kaji,
course project report are entirely our own, and should have S. Suzuki, J. Yoshikawa, H. Yonezawa, N. C. Menicucci, and A. Furu-
sawa, “Ultra-large-scale continuous-variable cluster states multiplexed
no bearing on the reputations of others not listed as authors. in the time domain,” Nature Photonics, vol. 7, pp. 982–986, 2013.
[24] Moran Chen, Nicolas C. Menicucci, and Olivier Pfister, “Experimen-
R EFERENCES tal realization of multipartite entanglement of 60 modes of a quantum
optical frequency comb,” Phys. Rev. Lett., vol. 112, pp. 120505, Mar
[1] R. Raussendorf and H. J. Briegel, “Quantum computing via measure- 2014.
ments only,” eprint arXiv:quant-ph/0010033, Oct. 2000. [25] Stephen D. Bartlett, Barry C. Sanders, Samuel L. Braunstein, and Kae
[2] Hans J. Briegel and Robert Raussendorf, “Persistent entanglement in Nemoto, “Efficient classical simulation of continuous variable quan-
arrays of interacting particles,” Physical Review Letters, vol. 86, no. 5, tum information processes,” Phys. Rev. Lett., vol. 88, pp. 097904, Feb
pp. 910, 2001. 2002.
[3] Gerald Gilbert, Michael Hamrick, and Yaakov S. Weinstein, “Efficient
construction of photonic quantum-computational clusters,” Phys. Rev.
A, vol. 73, pp. 064303, Jun 2006.
[4] Philippe Jorrand and Simon Perdrix, “Unifying quantum computation
with projective measurements only and one-way quantum computa-
tion,” in Moscow, Russia. International Society for Optics and Photon-
ics, 2005, pp. 44–51.
[5] Daniel E. Browne and Terry Rudolph, “Resource-efficient linear opti-
cal quantum computation,” Phys. Rev. Lett., vol. 95, pp. 010501, Jun
2005.
[6] M. A. Nielsen, “Universal quantum computation using only projec-
tive measurement, quantum memory, and preparation of the| 0> state,”
arXiv preprint quant-ph/0108020, 2001.
[7] R. Raussendorf, J. Harrington, and K. Goyal, “Topological fault-
tolerance in cluster state quantum computation,” New Journal of
Physics, vol. 9, no. 6, pp. 199, 2007.
[8] Daniel Gottesman and Isaac L. Chuang, “Demonstrating the viability
of universal quantum computation using teleportation and single-qubit
operations,” Nature, vol. 402, no. 6760, pp. 390–393, 1999.
[9] Richard Jozsa, “An introduction to measurement based quantum com-
putation,” NATO Science Series, III: Computer and Systems Sciences.
Quantum Information Processing-From Theory to Experiment, vol.
199, pp. 137–158, 2006.
[10] R Raussendorf, J Harrington, and K Goyal, “Topological fault-
tolerance in cluster state quantum computation,” New Journal of
Physics, vol. 9, no. 6, pp. 199, 2007.
[11] Chetan Nayak, Steven H. Simon, Ady Stern, Michael Freedman, and
Sankar Das Sarma, “Non-abelian anyons and topological quantum
computation,” Rev. Mod. Phys., vol. 80, pp. 1083–1159, Sep 2008.
[12] Michael A. Nielsen, “Cluster-state quantum computation,” Reports on
Mathematical Physics, vol. 57, no. 1, pp. 147–161, 2006.
[13] Yaoyun Shi, “Both toffoli and controlled-not need little help to do
universal quantum computation,” arXiv preprint quant-ph/0205115,
2002.
[14] Richard Cleve and John Watrous, “Fast parallel circuits for the quan-
tum fourier transform,” in Foundations of Computer Science, 2000.
Proceedings. 41st Annual Symposium on. IEEE, 2000, pp. 526–536.
[15] Olaf Mandel, Markus Greiner, Artur Widera, Tim Rom, Theodor W.
Hänsch, and Immanuel Bloch, “Coherent transport of neutral atoms in
spin-dependent optical lattice potentials,” Phys. Rev. Lett., vol. 91, pp.
010407, Jul 2003.
[16] W. K. Hensinger, S. Olmschenk, D. Stick, D. Hucul, M. Yeo, M. Acton,
L. Deslauriers, C. Monroe, and J. Rabchuk, “T-junction ion trap array
for two-dimensional ion shuttling, storage, and manipulation,” Applied
Physics Letters, vol. 88, no. 3, 2006.
[17] Guido Burkard, Daniel Loss, and David P. DiVincenzo, “Coupled
quantum dots as quantum gates,” Phys. Rev. B, vol. 59, pp. 2070–2078,
Jan 1999.
[18] J. I. Cirac and P. Zoller, “Quantum computations with cold trapped
ions,” Phys. Rev. Lett., vol. 74, pp. 4091–4094, May 1995.
[19] D. Jaksch, J. I. Cirac, P. Zoller, S. L. Rolston, R. Côté, and M. D. Lukin,
“Fast quantum gates for neutral atoms,” Phys. Rev. Lett., vol. 85, pp.
2208–2211, Sep 2000.
[20] E. Knill, R. Laflamme, and Milburn G. J., “A scheme for efficient
quantum computation with linear optics,” Nature, vol. 409, pp. 46–52,
2000.
[21] An-Ning Zhang, Chao-Yang Lu, Xiao-Qi Zhou, Yu-Ao Chen, Zhi
Zhao, Tao Yang, and Jian-Wei Pan, “Experimental construction of