2021 IEEE International Conference on Quantum Computing and Engineering (QCE)

A quantum walk control plane for distributed

quantum computing in quantum networks
2021 IEEE International Conference on Quantum Computing and Engineering (QCE) | 978-1-6654-1691-7/21/$31.00 ©2021 IEEE | DOI: 10.1109/QCE52317.2021.00048

Matheus Guedes de Andrade1 , Wenhan Dai1, 2 , Saikat Guha3 , and Don Towsley1
1
College of Information and Computer Science, University of Massachusetts Amherst
2
Quantum Photonics Laboratory, Massachusetts Institute of Technology
3
College of Optical Sciences, University of Arizona

Abstract—Quantum networks are complex systems alone [3]–[5]. Distributed quantum computing becomes
formed by the interaction among quantum processors even more interesting in the noise intermediate scale
through quantum channels. Analogous to classical com- quantum machines (NISQ) scenario where there is a
puter networks, quantum networks allow for the distribu-
tion of quantum computation among quantum comput- clear tradeoff between the size of quantum comput-
ers. In this work, we describe a quantum walk protocol ers, in terms of number of qubits, and the fidelity
to perform distributed quantum computing in a quantum of quantum operations, given the fact that physical
network. The protocol uses a quantum walk as a quantum separation directly reduces cross talk among qubits [6].
control signal to perform distributed quantum operations. When the quantum network scenario is considered, the
We consider a generalization of the discrete-time coined
quantum walk model that accounts for the interaction complexity of distributed quantum computing extends
between a quantum walker system in the network graph in at least two dimensions. First, physical quantum
with quantum registers inside the network nodes. The channels have a well known depleting effect in the
protocol logically captures distributed quantum com- exchange of quantum data, e.g the exponential decrease
puting, abstracting hardware implementation and the in channel entanglement rate with distance [7]. Second,
transmission of quantum information through channels.
Control signal transmission is mapped to the propagation there is a demand for a quantum network protocol
of the walker system across the network, while interac- capable of performing a desired distributed quantum
tions between the control layer and the quantum registers operation while accounting for network connectivity.
are embedded into the application of coin operators. Generic quantum computation with qubits in distinct
We demonstrate how to use the quantum walker system quantum processors demands either the application of
to perform a distributed CNOT operation, which shows
the universality of the protocol for distributed quantum remote controlled gates [8] or the continuous exchange
computing. Furthermore, we apply the protocol to the of quantum information. For both cases, a network
task of entanglement distribution in a quantum network. protocol is necessary to orchestrate the communication
between nodes that are not directly connected with one
another.
I. I NTRODUCTION
One challenge in the design of a control protocol is
Quantum networking is an innovative, multidisci- the need for it to be agnostic to hardware implementa-
plinary field of research that promises revolutionary tions. There is a plethora of physical systems suited for
improvements in communications, enabling tasks and quantum computation under investigation, supercon-
applications that are impossible to achieve with the ducting qubits [9], trapped ions [10], [11] and Silicon-
exclusive exchange of classical information [1], [2]. vacancy color centers in diamond [12], [13] to name a
Similar to a classical computer network, a quantum few. In addition, there is a diverse investigation in the
network is a distributed system composed of quantum architectural description of quantum interconnecting
computers and quantum repeaters that exchange quan- devices capable of exchanging quantum information
tum information across physical channels. Among ap- encoded in distinct quantum physical quantities [14].
plications supported by quantum networks, distributed This diverse ecosystem of quantum network tech-
quantum computing is of particular interest as it lever- nologies indicates that distributed quantum computing
ages the power of interconnected quantum computers network protocols need to abstract physical implemen-
to create a virtual quantum machine with process- tations of quantum switches and network connectivity
ing capabilities that surpass its physical constituents while maintaining universality requirements.

978-1-6654-1691-7/21/$31.00 ©2021 IEEE 313

DOI 10.1109/QCE52317.2021.00048
 The goal of this article is to propose a control nodes. Furthermore, it is generic in the sense that
protocol for distributed quantum computing based on it abstracts hardware implementation and channel
discrete time quantum walks [15]. Quantum walks are transmissions, while being well-defined for any
universal for quantum computing [16]–[18] and have network topology
been successfully employed in the quantum network • We demonstrate how the protocol can be used
scenario to perform perfect state transfer (PST) be- to recover the behavior of entanglement distribu-
tween network nodes [19], [20], teleportation [21], [22] tion protocols previously described in the litera-
and quantum key distribution (QKD) [23]. Previous ture [25], [26].
works describe ways of distributing entanglement be- The description of the protocol is carried in the
tween nodes on a quantum network using the coin logical setting under the assumption that quantum error
space of the walker to propel entanglement generation correction processes ensure unitary operations for both
between qubits [20], [21]. In addition, the formalism control and data qubits.
of quantum walks with multiple coins enabled the The remainder of this article is structured as follows.
description of an entanglement routing protocol, which In Section 2, we present the mathematical background
interprets qubits within a network node as vertices needed for the description of the quantum walk pro-
of an abstract graph used by a quantum walker to tocol. We describe the quantum walk protocol and
generate entanglement [24]. In spite of their relevance, demonstrate its universality in Section 3. The descrip-
the quantum walk approaches defined in the literature tion is extended to case of multiple walkers in Section
are suited to particular network structures and quantum 4. In section 5, we apply the protocol to recover
operations. In particular, they consider the case of the behavior of entanglement distribution protocols.
regular lattices, describing walker dynamics on regular Finally, the manuscript is concluded in Section 6.
structures and do not address how the quantum walk
can be used to perform generic distributed quantum op- II. BACKGROUND
erators. In this context, this work adds to the literature
Consider the graph G = (V, E). Let δ(v) denote the
of both quantum walks and quantum networks with
set of neighbors of vertex v ∈ V and d(v) = |δ(v)|
the description of a quantum network control protocol
denote the degree of v. Let Δ(u, v) denote the hop-
that can be applied to arbitrary graphs and perform
distance between u and v in G. Throughout this work
universal quantum computing in a quantum network.
we refer to the inverse of a binary string x ∈ {0, 1}∗
A. Contributions as x. A quantum network is a set of quantum hosts
(quantum processors) interconnected by a set of quan-
The contributions of this article are three-fold. tum channels that allow for the exchange of quantum
• We propose a quantum walk protocol for dis- information [2]. A host is either a quantum repeater, a
tributed quantum computing in a quantum net- quantum router or a quantum computer with a fixed
work. The protocol uses a quantum walker system number of qubits, which performs generic quantum
as a quantum control signal to perform computa- operations. A quantum network can be represented as
tions among quantum processors that are physi- a symmetric directed graph N = (V, E). Each node
cally separated. We assume that each processor v ∈ V represents a quantum host that has a set Mv of
dedicates part of its internal quantum register to qubits that can be processed together at any time and a
represent the walker control signal and describe set Nv of qubits that is used to exchange quantum in-
how the control subsystem interacts with the data formation with nodes in its neighborhood δ(v) through
subsystem. The interaction between data and con- a set of quantum channels. More precisely, each edge
trol is specified by unitary operations that nodes (u, v) ∈ E represents a quantum channel connecting
need to implement in order to realize the quantum the qubits in Nu and Nv which can interact through
walk control plane. operations mediated by the channel. We will refer
• We show the universality of the protocol by to Nv and Mv as the networking and data  registers
describing how a 2-qubit controlled X (CNOT) of node
 v, respectively. The sets M = v Mv and
operation between qubits in distinct nodes of the N = v Nv are respectively referred to as the network
network can be performed with the quantum walk. control plane and the network data plane.
The protocol is universal in the sense that it allows This network model separates the qubit registers
for any quantum operation in the Hilbert space in the nodes into control and data registers. This
formed by all qubits in the data subsystem of the choice differs from previous works where the set of
qubits N that can interact with quantum channels distribution protocols that At represents channel en-
are considered alone [25], [26]. Note that considering tanglement protocols performed in all channels of
only networking qubits suffices to define entanglement the network and Bt represents either entanglement
routing protocols. Nonetheless, it is straightforward for swapping operations or GHZ projections performed
the goal of describing a quantum network control plane independently in multiple nodes. The LOCC formalism
protocol to consider the separation between control and has proven useful since it allowed for the derivation of
data registers in the nodes. fundamental bounds for entanglement rate distribution
[27]. In this article, however, our focus is to address
A. Quantum network protocols network protocols for the transmission of quantum
Consider the system formed by two quantum pro- information in the logical perspective. In particular, we
cessors u and v connected by a channel and their exploit the representation in (3) considering A and B
respective qubits. A local operation is a quantum as unitary operators provided by the network instead of
transformation represented as a separable operator of generic superoperators and perform the analysis in the
the form state vector formalism. It is worth emphasizing that the
unitarity assumption for operators A and B translates
O = Ou ⊗ Ov , (1) to the assumption that quantum error correction is
provided by the network.
where Ou and Ov acts on the state space of the qubits
at processors u and v, respectively. A local operation B. Quantum walks on graphs
assisted by two-way classical communication (LOCC) There are many ways to define a quantum walk
is a local operation that depends on classical informa- on a graph and this article focuses on the discrete-
tion exchanged between nodes [7], e.g the unitaries time coined quantum walk model. Given a symmetric
of quantum teleportation. The classical information is directed graph G = (V, E), a coined quantum walk on
used to select what operators are to be applied to the G is a process of unitary evolution on the Hilbert space
system, which is embedded in the formalism by av- HG = HV ⊗ HC formed by the edges of the graph,
eraging non-unitary measurement operators into trace- where HV codifies vertices and HC is the coin space
preserving operators. A quantum network protocol for of the walker codifying the degrees of freedom the
N is an algorithm that operates on the qubits at the walker can move on. More precisely, every (v, u) ∈ E
nodes, transforming their joint state by an LOCC Λ̄t defines a basis vector |v, c for HG through a mapping
as between the edges incident to v with the set of degrees
of freedom Cv = {0, 1, . . . , d(v)−1}. As a convention,
ρ(t) = Λ̄t ρ(0)Λ̄†t , (2)
we map the degrees of freedom of the walker at a given
where t is the number of rounds in the protocol, ρ vertex v to the order of labels of its neighbors, such
is a density operator defined on the Hilbert space that |v, c refers to the edge (v, u), where u is the c-
HM formed by all qubits on the network, ρ(0) is a th smallest label in δ(v). Later, cvu will be used to
density matrix where registers in distinct refer to the degree of freedom of v that represents the
tnodes are edge (v, u) and cv to represent the self-loop (v, v). The
in a separable state. We express Λ̄t = k=0 Λk as 
a product of other LOCCs each applied at different generic state of the walker |Ψ(t) = ψ(v, c, t) |v, c
time steps [27]. It suffices for ρ(0) to be a separable is a superposition of the edges of G and the walker
density operator because the preparation of any non- evolution is defined as
separable state between registers in different network |Ψ(t + 1) = S(t)C(t) |Ψ(t) , (4)
nodes can be represented by an LOCC. In fact, it is also
possible to model a protocol as a sequence of external where C and S are respectively referred to as the coin
(mediated by channels) and internal (local to vertices) and shift operators. The coin is a unitary operator of
time-dependent superoperators At and Bt such that the the form

state of the network is described as C(t) = |vv| ⊗ Cv (t), (5)
v
ρ(t + 1) = Bt [At [ρ(t)]], (3)
where Cv : HC → HC . The shift can be defined as any
where ρ is a density matrix characterizing the joint permutation operator on the edges of the graph that
state of all network memories at discrete time t. As maps an edge (v1 , u) to an edge (u, v2 ). This mapping
an example, it is usually the case for entanglement of edges represents a permutation between states |v1 , c
and |u, c , where u is the c-th neighbor of v1 . Two shift applies the operator Ku (t) on qubits in v if, and only
operators are used throughout this work: the identity if, the walker has a non-zero wavefunction component
operator, which is a trivial permutation of the edges, in v. Note that Uv (t) is the extension of Kv (t), which
and the flip-flop shift operator given by acts in HMv , to HM with the identity operator. Note
  that the operator described in (7) extends the definition
Sf = |v, cvu u, cuv | , (6) in (5) to perform unitary operations in the data qubit
v∈V u∈δ(v)
space controlled by vertex position.
which applies, for every (v, u) ∈ E, the permutation The second law of interaction defines controlled
(v, u) → (u, v). Sf reverses edges in the walker operations between the data qubits and the coin space
wavefunction and is well defined for every symmetric of the walker. In particular, we consider the case where
directed graph. The label of nodes and degrees of only a single data qubit in a given node is used to
freedom are numerical and expressed as binary strings. control a unitary operation in the coin space of the
In this setting, the degree of freedom c represents walker. Let
the bit-wise negation of label c. Note that cvu is not  
Uvq (t) = Uvqs (t) ⊗ |ss| Iq  , (9)
necessarily cuv .
s∈{0,1} q  =q
III. Q UANTUM WALK NETWORK CONTROL PLANE be a unitary operator acting on the coin space of vertex
A direct way of controlling operations in a quantum v controlled by the qubit q ∈ Mv , defined for the joint
network with a quantum walk is to consider a cou- space HC ⊗HM . The complete interaction has the form
pled system between a walker and the qubits in the 
nodes, using supersposition in the walker system to O(t) = |vv| ⊗ Uvq (t). (10)
v
implement controlled unitary operations. We describe
this joint system in the logical setting. Let N  = The operators correspond to a distributed implemen-
(V, E ∪ {(v, v), ∀v ∈ V }) be the graph obtained by tation of a quantum walk system in the network. In
adding self-loops to a network N . HW = HV ⊗ HC essence, each node v contributes with Nv to describe
denotes the space of a walker system on N  and the space HW in HN . Thus, (7) and (10) are operations
Hg = HW ⊗ HM denotes the global Hilbert space controlled by the states of HN that represent walker
spanned by the systems. The walker system is assumed position since |v is implemented by an entangled state
to be implemented in the networking control plane, among qubits in N . This encapsulates the necessary
such that HW is a subspace of HN . Since the Hilbert entanglement required for distributed controlled opera-
space of the walker system represents an edge of N  tions between the qubits of the network. As previously
as a basis vector, the dimension of the Hilbert space mentioned, we assume quantum error correction and
HN must be at least |E| for HW to be a subspace. We describe the walker protocol in terms of logical qubits
consider Hg as a subspace of HN ⊗ HM . and operations. In the context of error correcting codes,
a single logical qubit is implemented by a set of
A. Control-data interactions physical qubits [28], [29]. An implementation of the
We prescribe two laws of interaction between the walker demands O(log(|E|)) logical qubits in N since
quantum walk system and the data qubits in the net- each edge of N  is a basis vector of HW and the di-
work. The first generalizes a coin operator to the global mension of the space spanned by qubits is exponential
Hilbert space Hg following in the number of qubits. For practical implementations,
 network models that consider physical networking
C(t) = |vv| ⊗ Cv (t) ⊗ Uv (t), (7) qubits in the order of O(|E|) have exponentially many
v
physical qubits as the required number of logical qubits
where Cv (t) : HC → HC is a unitary operator on the for the walker, which is interesting for the purpose of
coin space of the walker and Uv (t) : HM → HM is quantum error correction.
a unitary operator on the data space of the network
written as B. Universal distributed quantum computing
 The interaction behavior described by (7) and (10)
Uv (t) = Kv (t) I Mv , (8) suffices for universal quantum computing. Our protocol
u=v
trivially allows for the application of any separable
where IMv is the identity operator for the Hilbert space qubit operator in the network nodes, since there is no
spanned by all qubits of vertex v. Essentially, C(t) need for the interaction between data qubits and the
walker in this case. Thus, we demonstrate universality C. Protocol execution in the network
by demonstrating how a CNOT gate between any pair
of qubits a and b, respectively at arbitrary nodes A and In the context of quantum network protocols, the
B, can be performed using the quantum walk control operators defined by (7) and (10) do not capture the
plane. In fact, the demonstration works for any two- propagation of control information across the network.
qubit controlled operation by substituting X with the The propagation of quantum control information be-
desired single-qubit operator. Without loss of gener- tween neighbors in the network is embedded in the
ality, assume that a is the control qubit. To simplify formalism through the walker’s shift operator. The
notation, we only define the operators C(t) in (7) and propagation in the case of non-neighboring nodes
O(t) in (10) for the subspaces spanned by the qubits needs both coin and shift operators as defined in (5)
in nodes A and B, considering undefined operators and (6). Note that generic computation on the data
to be identities, and omit the qubits in M \ {a, b}. qubits can be performed with (7) and (10) given states
The subscripts following degrees of freedom inside a in the superposition (15), although it is not possible to
ket specifies edges while the ones following qubits are evolve the walker to that superposition from |A, cA 
considered as indices, e.g |A, cA , 0a , 1b  ∈ HW ⊗ H4 without shift operators.
represents the walker in the self-loop (A, A), a in the We define coin and shift operators that evolve the
|0 state and b in the |1 state, where H4 is the Hilbert global state from (13) to (15). The quantum walk
space spanned by 2 qubits. evolution is restricted to neighbor locality, such that,
We consider the initial state of the walker to be to have the state given in (15) at time t, all of the
|A, cA , which corresponds to the self-loop in A. Thus, wavefunction at time t − 1 must exclusively be a
the global system is described by the state vector superposition of edges incident to A, B and its neigh-
bors. There are many ways to define coin and shift
|Ψ(0) = |A, cA  ⊗ (α |0a  + β |1a ) ⊗ |Ψb  , (11) operators with this desired behavior and we consider
the case where the quantum walk traverses a single
where |Ψb  is the state of b. The interaction operator O
path connecting A and B. Some auxiliary definitions
described in (10) is applied with the CNOT operation
and assumptions are required to describe the operators
UAa = ICA ⊗ |00| + XCA ⊗ |11| , (12) in context. Let p be a minimum path of the network
connecting A and B with hop distance Δ(A, B). We
which trivially generates the entangled state assume that every node knows the network topology
and that classical information can be transmitted across
O |Ψ(0) = (α |A, cA , 0 + β |A, cA , 1) ⊗ |Ψb  . the nodes. Recall that the edges of N  are mapped to
(13) walker states following the relation
This entangled state can be propagated in the network
through the evolution of the walker system such that (v, u) → |v, cvu  , (16)
the state after propagation becomes
and that the self-loop of node v is mapped to the degree
|Ψ(t) = (α |A, cA , 0 + β |B, cB , 1) ⊗ |Ψb  . (14)
of freedom cv , for all v. We refer to the edge that
We refer to the propagation of the walker as the connects A to its neighbor in p as |A, cpA  and the
transmission of the control signal through the network, reverse edge that connects B to its precedent vertex
which will be further explained in detail. The last step in p as |B, cpB . This notation is depicted in Figure
refers to the application of the extended coin operator 1 for a 2D-grid network. If some quantum operation
defined in (7), with Kb = X and all other operators is performed between A and B at a given time t, the
defined as identity. The final state obtained is propagation process starts at time t − Δ(A, B).
  The extended shift operator used to route informa-
|Ψ(t) = α |A, cA , 0, Ψb  + β B, cB , 1, Ψb , (15) tion is fixed and given by
which clearly shows the application of a CNOT con-
trolled by a with b as a target. S = Sf ⊗ IM , (17)
Note that the state obtained in (15) is entangled with
the walker subsystem. We will later demonstrate how where Sf is defined in (6). The coin operator definition
to separate the data qubits from the walker. is also time-independent, although it depends on the
 B states |A, cA  and |A, cpA  that are entangled with a.
The flip-flop shift allows the wavefunction in |A, cpA 
|B, cp
B |w, cwB 
to propagate along p and ensures that the |A, cA 
B w component remains in the superposition throughout
|w, cwv  |u, cvw  protocol execution. Each coin flip routes information
u v w v
on nodes internally, ensuring propagation. The net
effect of Δ(A, B) successive applications of SC is the
|u, cvu  |v, cuv 
superposition specified in (15).
A u

|A, cp
D. Separating data and control
|v, cuA  A

A
The state prescribed in (15) is an entangled state
|A, cA  between control and data. This implies that a partial
(a) Path p connecting A and B. (b) Map trace operation in the walker system does not leave the
between edges state of a and b as it should be if the operation was
and vectors.
performed without the walker. In order to overcome
Fig. 1: Notation for edges exemplified in a grid graph. this problem, the walker evolution is reversed after
Consider that A and B are two nodes connected in a the controlled operation takes place by applying the
2D grid network. (a) p is a minimum path connecting A inverses of the unitary operators used for propagation.
and B with hop-distance 4 traversed by the walker. (b) Since the coin and shift operators considered are per-
Each edge on the path corresponds to a vector in HW , mutation operators, they are Hermitian unitaries and,
which appear in the walker wavefunction throughout thus, are their own inverses. It takes Δ(A, B) time
movement. The degrees of freedom are defined such steps to reverse the walker back to A and to transform
that |x, cxy  represents edge (x, y). As an example, the joint state of the system to the form
the flip-flop operator specified in (6) maps |v, cvu  →
|u, cuv , while the operator Cu defined in terms of (18) α |A, cA , 0a , Ψb  + β |A, cA , 1a , Ψb  . (19)
maps |u, cuA  → |u, cuv .
It should be clear that an extended coin operator with
interaction of the form given in (7) and a measurement
minimum path p chosen. All operators Cv in (7) have on the walker produces one of the following separable
the form states

Cvp = |c1 c2 | + |c2 c1 | + |cc| (18) |A, cA  ⊗ (α |0a , Ψb  + β |1a , Ψb ),
(20)
c∈Cv |A, cA  ⊗ (α |0a , Ψb  − β |1a , Ψb ),
c=c1 ,c2

where c1 and c2 refer to the degrees of freedom that with equal probability. It is straightforward from the
represent the edges incident to v in p. Thus, we define separability between the walker system and the qubits
Cvp specifying c1 and c2 for the vertices of interest. a and b that the state after a partial trace operation
The unitary CA has c1 = cA and c2 = cpA . Let u and in the walker system is now equivalent to a CNOT
w be the neighbors of v ∈ p \ {A, B} on the path. Cv gate, up to a single-qubit Z gate conditioned on the
has c1 = cvu and c2 = cvw , representing a permutation measurement outcome.
between the edges (v, u) and (v, w) in p. Finally, the Note that moving the walker backwards is only one
operator CB has c1 = cpB and c2 = cB . It suffices to set of several ways to achieve a separable state between the
Kv (t) = IMv in (8) to perform the desired controlled walker and the qubits a and b. In a nutshell, any walker
operation between a and b, although it is possible to dynamics that concentrates the walker’s part of the
perform operations controlled by a on the qubits in the wavefunction into a single network node suffices for
intermediate nodes as the walker moves by choosing this purpose, e.g propagating the walker wavefunction
Kv (t) accordingly. from both A and B to an intermediate node. Under the
The overall behavior of the walker is straightforward assumption of quantum error correction, this backward
and is illustrated in Figure 2. It is assumed that the propagation can occur simultaneously with any further
operator O defined in (13) is applied to the system at quantum operation that nodes A and B may perform
the beginning of execution. The first application of the on the qubits and does not impact the latency of the
coin operator in A creates a superposition between the protocol.
 B B B B B

|B, cp
B

w w w w w

u u u u u

|v, cvu 

v v v v v

|A, cA
p |v, cvA 
|A, cA 
z A A A A A
|A, cA  |A, cA  |A, cA  |A, cA  |A, cA 
(a) Initial state. (b) Coin in A. (c) Flip-flop shift (d) Coins in path. (e) Final state.
Fig. 2: Protocol execution in a 5-by-5 grid with a quantum walk through a single path. Dark edges depict
vectors which have non-zero wavefunction component in a given step. (a) The initial state of execution is the
state generated by the application of the controlled operation demonstrated in (13). The node z is shown for
illustration purposes and is not for propagation. (b) The first coin flip permutes the wavefunction into edges
(A, A) and (A, v). (c) The flip-flop shift exchanges edge (A, u) with edge (u, A), moving the walker while
mapping the self-loop edge to itself. (d) After the first coin flip, all subsequent coin operators work as shift
operators inside a node, mapping degrees of freedom in order to propel the walker towards B. (e) After Δ(A, B)
steps, the final wavefunction is a uniform superposition between edges (A, A) and (B, w), which can be used
to perform an operation controlled by qubit a located in A with target qubit b located in B.

IV. P ROPAGATION OF MULTIPLE CONTROL SIGNALS Since the memories affected by walker j are unique,
for every j ∈ {0, . . . , T − 1}, each walker can be
We demonstrated in the previous section how the
seen as a parallel control signal working on distinct
quantum walk can propagate through a particular path
qubits. Note that, even though the walker evolution is
of the network connecting nodes A and B. This is
performed by separable extended coin operators and
easily extended to the traversal of multiple paths as we
shifts, the application of the controlled operation O
will show next. In this setting, k walkers can be used
described in (10) before propagation implies that the
to simultaneously perform operations controlled by a
walkers are maximally entangled with each other and
with k target qubits bj located in nodes Bj , for j ∈
with qubit a. Furthermore, the flip-flop shift operator
{0, . . . , k−1}. This protocol execution translates to the
for each walker does not depend on a particular path
parallel application of k + 1 1-qubit gates controlled
and the joint operator for all walkers is described by
by a in terms of distributed quantum computation.
The analysis carried for a single walker system
extends to this case by considering a set of paths P k−1

such that, for every j, the path pj ∈ P connects A and S P
= S, (22)
Bj . The path pj defines extended coin operators for the j=0
j-th walker following (18). By carefully choosing the
qubits controlled by each walker, so that each walker
interacts with a unique set of qubits, it is possible with S given in (17).
to route k = |P| walkers concurrently. This is a
In addition to simultaneous 2-qubit control opera-
trivial extension to the single walker case because the
tions, it is trivial to use multiple walkers departing
separability of individual coins for each walker yield
from A to perform controlled operations in the nodes
the k-walker coin operator for the set of paths P to be
Bj , where the target qubit bj in Bj is controlled by
of the form
qubit aj in A. The fundamental difference for this case
k−1
 is that the O operator described in (13) is controlled by
C P (t) = C pj (t). (21) qubit aj when applied to the j-th walker, generating a
j=0 separable state among the walkers.
 V. Q UANTUM WALK PROPAGATION AND the protocol assumes a network model where every
ENTANGLEMENT DISTRIBUTION quantum channel in the network is mapped in a 1-to-
As described previously, the operators defined in 1 fashion with pairs of qubits in the nodes incident
(7) and (10) allow for universal distributed quantum to that channel. The logical behavior of this protocol
computation among nodes of the network. In this sense, is recovered by sending one quantum walk from A
the movement of the walker introduces entanglement through each path pj such that, once it arrives at B,
across nodes. Given this is the case, it is suitable to every qubit incident to the quantum channels in the
consider entanglement distribution protocols in terms path are in the same GHZ state. Thus, consider that
of the quantum walk control plane. In this section, we all qubits in the network start in the |0 state. Initialize
j
apply the quantum walk control protocol to recover the k walkers in state |A, cA . The operator CA (0) for the
behavior of two entanglement distribution protocols. In j-th walker can be chosen as any unitary that maps
particular, the recovered protocols distribute maximally 1  p 
entangled Greenberger-Horne-Zeillinger (GHZ) states |A, cA  → √ (|A, cA  + A, cAj ), (25)
2
across network nodes. We demonstrate, for both pro- j
tocols, how the entanglement introduced by the walker while the interaction operator KA (0) is taken to be
can be used to produce the same entangled states. the identity operator. The state of the system after the
Superposition of the walker wavefunction allows for application of the extended operator C(0) is
the creation of a Bell state between a and b as follows.  1 p
C(0) |Ψ(0) = √ ( |A, cA  + |A, cAj ). (26)
Assume that the joint state of the system is at time t 2
j
described by
1 Protocol execution is performed by the operators de-
|Ψ(t) = √ (|A, cA , 0a , 0b  + |B, cB , 0a , 0b ), (23) fined in (22) and (21) as specified in the previous
2 section, although the operators Kv are now defined
which is a separable state between the walker and the to generate entanglement in the nodes as the walker
qubits a and b. By setting UV = Xv ⊗I and CV (t) = I passes by. Let qvj0 and qvj1 denote the pair of qubits
for V ∈ {A, B} in (7), the extended coin operator C(t) inside node v ∈ pj \ {A, B} that need to be entangled
is together by walker j. The operator Uvj in equation (8)
1 is defined for walker j by specifying
C(t) |Ψ(t) = √ (|A, cA , 1a , 0b  + |B, cB , 0a , 1b ), 
2
(24) Kv (t) = Xqj0 ⊗ Xqj1 Iq (27)
q=qvj0 ,qvj1
which is an entangled state between a and b. The j j
entanglement between a, b and the walker system can for all t, performing an X gate on qubits qv0 and qv1
be removed by transmitting the walker back to A as controlled by the position of the j-th walker. Since
explained in Section III-D. Note that only operators of paths are all edge disjoint, every qubit in the graph
the form prescribed in (7) are needed because there is spanned by P interacts with exactly one walker under
no control dependent on the state of either a or b. the application of the generalized coin operator. For
A GHZ generalizes Bell states to multiple qubits. walker j, the interaction operator with qubit aj is given
A k-qubit GHZ projection is a von-Neumann mea- by
surement operation in the GHZ basis and a Bell state Xaj ⊗ I, when t = 1,
KA (t) = (28)
measurement (BSM) is a 2-qubit GHZ projection. Iaj ⊗ I, when t = 0,
A. Multi-path entanglement distribution with BSMs where the identity operator after the tensor product
We start with the multi-path protocol defined in [25]. symbol acts on all qubits in A except aj . Considering
Assuming global link state knowledge, it suffices to that uj is the first neighbor of A in the path pj , the
choose the same paths that a complete round of the application of the evolution operator S(0)C(0) gives
protocol uses to propagate entanglement. Consider that,  1
for such round, the set of paths P = {p0 , . . . , pk−1 } is |Ψ(1) = √ ( |A, cA  + |uj , cuj A ), (29)
j
2
used. In this protocol, every path pj used to distribute
entanglement yields a set of 2(|pj |−1) qubits in a GHZ where the data qubits are omitted for simplicity.
state, which comes from the net effect of performing S(1)C(1) generates an entangled state between aj and
|pj | − 2 BSMs in the intermediate nodes. Moreover, the qubits quj0 and quj1 and moves the walker to
the neighbors that are 2 hops away from A. Every
application of S(t)C(t) for t > 1 increases by two the
number of qubits entangled with walker j such that,
when t = Δ(A, B), the walker j is in an entangled
state with the qubits in pj , aj and bj . Figure 3(a)
depicts protocol execution for a 5-by-5 grid network.
In this case, P is formed by two shortest-paths in the
B
grid, i.e K = 2, and the entire evolution of the walkers
takes 4 steps.
B. Distributing GHZ states across network nodes
GHZ projectors are used in [26] to improve the
multi-path BSM protocol described in [25]. The logical A
behavior of the GHZ distribution protocol can be
recovered in the quantum walk protocol by slightly
modifying the description presented for the multi-path
BSM protocol. In this protocol, every qubit in a node (a) Multi-path propagation.
used to distribute entanglement is part of the final
GHZ state achieved. This is in clear contrast with the
BSM multi-path protocol, where each path defines an
independent GHZ state. The first modification is that, B
instead of preparing separable walkers with coins that
maps states following the form given in (25), the k u2
walkers need to be entangled in the state
1  p 
√ (|A, cA , . . . , A, cA  + A, cpA0 , . . . , A, cAj ). (30)
2 A

Note that this behavior can be obtained with an inter-

action operator of the form (10) and a coin operator in u1
the form specified by (18). Secondly, instead of rout-
ing walkers through edge disjoint paths, the walkers (b) Spanning tree propagation.
traverse a tree of the network. More concretely, let T Fig. 3: Recovering protocol behavior with quantum
be a spanning tree of the lattice, rooted at A. k is now walks. Protocols that rely on GHZ projections apply
the number of leaves in T and P is the set of paths entanglement swapping to distribute entanglement be-
in the tree connecting A with leaves. For all vertices tween A and B. Entanglement swapping is non-unitary,
v, the interaction operator between the walker and the although the overall effect of applying a GHZ projec-
data qubits is tion is a maximally entangled state between qubits.
 1 The quantum walk approach recovers this behavior by
Kv = XqvWv
, (31) generating GHZ states unitarily. (a) A multi-path BSM
qv
protocol is recovered by sending separable walkers
where Wv is the number of walkers that pass through into each edge-disjoint path used in a round of the
v, i.e the number of leaves connected to A by v in protocol, denoted in the figure by shaded edges. Each
the tree. Note that Kv is the same for every walker. path forms an independent GHZ state between all
The exponent W1v is required because more than one qubits in the path. (b) The GHZ projection protocol
walker may pass through the same vertex, and operate is recovered by sending entangled quantum walks to
on the same set of qubits. Since paths are taken from each leaf li of a spanning tree of the network rooted
a spanning tree, each walker can take a particular time in A, represented in the figure by green vertices. All
interval to traverse its path. Thus, the time necessary qubits in the network are, in this case, in a GHZ state.
to entangle all of the necessary qubit in the network
and remove the entanglement between the walkers
and the data qubits is on the order of O(maxj |pj |),
where |pj | denotes the hop distance of path pj . Again,
protocol execution for the generic GHZ projector case exclusively with quantum information. In this setting,
is exemplified for a 5-by-5 grid in Figure 3(b). In this nodes would transmit a quantum state containing both
case, k = 7 and the whole process takes 7 time steps the state of the walker and the target node to which
to complete, since 7 is the size of the largest path in control information must be transmitted to.
the tree. For nodes u1 and u2 , the value of W is 1 and Acknowledgments—This research was supported in
3 respectively. part by the NSF grant CNS-1955834, and NSF-ERC
Center for Quantum Networks grant EEC-1941583.
VI. C ONCLUSION
The quantum walk protocol proposed in this article R EFERENCES
provides a logical description for a network control [1] H Jeff Kimble. The quantum internet. Nature,
plane capable of performing universal distributed quan- 453(7198):1023–1030, 2008.
tum computing. The description abstracts the imple- [2] Stephanie Wehner, David Elkouss, and Ronald Hanson. Quan-
tum internet: A vision for the road ahead. Science, 362(6412),
mentation of the quantum walker system, as well as the 2018.
implementation of quantum operations in the network [3] Robert Beals, Stephen Brierley, Oliver Gray, Aram W Harrow,
nodes. It considers that quantum error correction yields Samuel Kutin, Noah Linden, Dan Shepherd, and Mark Stather.
Efficient distributed quantum computing. Proceedings of the
the application of perfect unitary operators. The key Royal Society A: Mathematical, Physical and Engineering
idea that the protocol builds upon is the use of a Sciences, 469(2153):20120686, 2013.
quantum walker system as a quantum control signal [4] Angela Sara Cacciapuoti, Marcello Caleffi, Francesco Tafuri,
Francesco Saverio Cataliotti, Stefano Gherardini, and Giuseppe
that propagates through the network one hop a time. In Bianchi. Quantum internet: networking challenges in dis-
spite of abstracting physical implementations, the prop- tributed quantum computing. IEEE Network, 34(1):137–143,
agation of the walker stipulates latency constraints for 2019.
[5] Laszlo Gyongyosi and Sandor Imre. Scalable distributed gate-
the protocol. A generic controlled operation between model quantum computers. Scientific reports, 11(1):1–28,
a qubit in node A with a qubit in node B demands 2021.
O(Δ(A, B)) steps of walker evolution. In the con- [6] Rodney Van Meter and Simon J Devitt. The path to scalable
distributed quantum computing. Computer, 49(9):31–42, 2016.
text of a possible physical realization of such control [7] Stefano Pirandola, Riccardo Laurenza, Carlo Ottaviani, and
system, this latency constraints translates directly to Leonardo Banchi. Fundamental limits of repeaterless quantum
the physical distance between nodes A and B. The communications. Nature communications, 8(1):1–15, 2017.
[8] Kevin S Chou, Jacob Z Blumoff, Christopher S Wang, Philip C
description of the protocol in the logical setting also Reinhold, Christopher J Axline, Yvonne Y Gao, Luigi Frunzio,
masks the effects of walker propagation in the fidelity MH Devoret, Liang Jiang, and RJ Schoelkopf. Deterministic
of distributed operations. When considering imperfect teleportation of a quantum gate between two logical qubits.
Nature, 561(7723):368–373, 2018.
operators, the fidelity of the final outcome is bounded
[9] Frank Arute, Kunal Arya, Ryan Babbush, Dave Bacon,
by the fidelity of the coin and shift operators in the Joseph C Bardin, Rami Barends, Rupak Biswas, Sergio Boixo,
quantum walk system. Throughout this manuscript, the Fernando GSL Brandao, David A Buell, et al. Quantum
propagation of a quantum walk across the network supremacy using a programmable superconducting processor.
Nature, 574(7779):505–510, 2019.
depended on a particular path of the network to be [10] David Kielpinski, Chris Monroe, and David J Wineland. Ar-
traversed. This has the clear requirement that every chitecture for a large-scale ion-trap quantum computer. Nature,
node in the network knows the network topology 417(6890):709–711, 2002.
[11] JM Pino, JM Dreiling, C Figgatt, JP Gaebler, SA Moses,
and that the operators used can be defined by the MS Allman, CH Baldwin, M Foss-Feig, D Hayes, K Mayer,
exchange of classical messages between the nodes. The et al. Demonstration of the trapped-ion quantum ccd computer
protocol described was used to represent entanglement architecture. Nature, 592(7853):209–213, 2021.
[12] Pieter-Jan Stas, Bartholomeus Machielse, David Levonian, Ralf
distribution protocols defined in the literature in terms Riedinger, Mihir Bhaskar, Can Knaut, Erik Knall, Daniel
of quantum control information exchanged between the Assumpcao, Rivka Bekenstein, Yan Qi Huan, et al. High-
nodes in the network. This result highlights connec- fidelity quantum memory for the silicon-vacancy defect in
diamond. Bulletin of the American Physical Society, 2021.
tions between the proposed protocol and entanglement [13] Jun-Feng Wang, Fei-Fei Yan, Qiang Li, Zheng-Hao Liu, Jin-
distribution protocols. Ming Cui, Zhao-Di Liu, Adam Gali, Jin-Shi Xu, Chuan-
There are two clear directions for future work con- Feng Li, and Guang-Can Guo. Robust coherent control of
solid-state spin qubits using anti-stokes excitation. Nature
sidering our results. The first relates to the investi- Communications, 12(1):1–9, 2021.
gation of a network implementation for a quantum [14] David Awschalom, Karl K Berggren, Hannes Bernien, Sunil
walk system. Such implementation would allow for a Bhave, Lincoln D Carr, Paul Davids, Sophia E Economou, Dirk
Englund, Andrei Faraon, Martin Fejer, et al. Development of
realistic characterization of quantities like fidelity and quantum interconnects (quics) for next-generation information
latency. The second point is the description of control technologies. PRX Quantum, 2(1):017002, 2021.
[15] Dorit Aharonov, Andris Ambainis, Julia Kempe, and Umesh
Vazirani. Quantum walks on graphs. In Proceedings of the
thirty-third annual ACM symposium on Theory of computing,
pages 50–59, 2001.
[16] Andrew M Childs. Universal computation by quantum walk.
Physical review letters, 102(18):180501, 2009.
[17] Andrew M. Childs, David Gosset, and Zak Webb. Univer-
sal computation by multiparticle quantum walk. Science,
339(6121):791–794, 2013.
[18] Neil B Lovett, Sally Cooper, Matthew Everitt, Matthew Tre-
vers, and Viv Kendon. Universal quantum computation us-
ing the discrete-time quantum walk. Physical Review A,
81(4):042330, 2010.
[19] Thomas Nitsche, Fabian Elster, Jaroslav Novotný, Aurél
Gábris, Igor Jex, Sonja Barkhofen, and Christine Silberhorn.
Quantum walks with dynamical control: graph engineering,
initial state preparation and state transfer. New Journal of
Physics, 18(6):063017, jun 2016.
[20] Xiang Zhan, Hao Qin, Zhi-hao Bian, Jian Li, and Peng
Xue. Perfect state transfer and efficient quantum routing:
A discrete-time quantum-walk approach. Physical Review A,
90(1):012331, 2014.
[21] Heng-Ji Li, Xiu-Bo Chen, Ya-Lan Wang, Yan-Yan Hou, and
Jian Li. A new kind of flexible quantum teleportation of
an arbitrary multi-qubit state by multi-walker quantum walks.
Quantum Information Processing, 18(9):266, 2019.
[22] Yu Wang, Yun Shang, and Peng Xue. Generalized telepor-
tation by quantum walks. Quantum Information Processing,
16(9):221, 2017.
[23] Chrysoula Vlachou, Walter Krawec, Paulo Mateus, Nikola
Paunković, and André Souto. Quantum key distribution with
quantum walks. Quantum Information Processing, 17(11):1–
37, 2018.
[24] Meng Li and Yun Shang. Entangled state generation
via quantum walks with multiple coins. arXiv preprint
arXiv:2011.01643, 2020.
[25] Mihir Pant, Hari Krovi, Don Towsley, Leandros Tassiulas,
Liang Jiang, Prithwish Basu, Dirk Englund, and Saikat Guha.
Routing entanglement in the quantum internet. npj Quantum
Information, 5(1):1–9, 2019.
[26] Ashlesha Patil, Mihir Pant, Dirk Englund, Don Towsley, and
Saikat Guha. Entanglement generation in a quantum network
at distance-independent rate, 2020.
[27] Stefano Pirandola. End-to-end capacities of a quantum commu-
nication network. Communications Physics, 2(1):1–10, 2019.
[28] Emanuel Knill and Raymond Laflamme. Theory of quantum
error-correcting codes. Phys. Rev. A, 55:900–911, Feb 1997.
[29] Joschka Roffe. Quantum error correction: an introductory
guide. Contemporary Physics, 60(3):226–245, 2019.