A quantum network stack and protocols for reliable entanglement-based networks

by   A. Pirker, et al.

We present a stack model for breaking down the complexity of entanglement-based quantum networks. We construct the quantum network stack in a hierarchical manner comprising several layers, similar to the classical network stack, and identify quantum networking devices operating on each of these layers. The layers responsibilities range from establishing point-to-point connectivity, over intra-network graph state generation, to inter-network routing of entanglement. In addition we propose several protocols operating on these layers. In particular, we extend the existing intra-network protocols for generating arbitrary graph states to ensure reliability inside a quantum network, where here reliability refers to the capability to compensate for devices failures. Furthermore, we propose a routing protocol for quantum routers which enables to generate arbitrary graph states across network boundaries. This protocol, in correspondence with classical routing protocols, can compensate dynamically for failures of routers, or even complete networks, by simply re-routing the given entanglement over alternative paths. We also consider how to connect quantum routers in a hierarchical manner to reduce complexity, as well as reliability issues arising in connecting these quantum networking devices.



There are no comments yet.


page 7

page 17

page 18


Quantum Internet Protocol Stack: a Comprehensive Survey

Classical Internet evolved exceptionally during the last five decades, f...

Modeling and Designing Routing Protocols in Quantum Networks

Quantum networks enable a number of important applications such as quant...

A P4 Data Plane for the Quantum Internet

The quantum technology revolution brings with it the promise of a quantu...

Quantum routing with fast reversals

We present methods for implementing arbitrary permutations of qubits und...

Decentralized Base-Graph Routing for the Quantum Internet

Quantum repeater networks are a fundamental of any future quantum Intern...

High fidelity GHZ generation within nearby nodes

Generating entanglement in a distributed scenario is a fundamental task ...

Quantum Transport Protocols for Distributed Quantum Computing

Quantum computing holds a great promise and this work proposes to use ne...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

I Introduction

Quantum networks lie at the heart of the success of future quantum information technologies. Possible applications of such networks, over even a global scale quantum internet Kimble (2008), are quantum key distribution protocols Shor and Preskill (2000); Renner (2005); Zhao and Yin (2014); Gottesman and Lo (2003); Lo (2001); Wang et al. (2015); Chen et al. (2016), quantum conference key agreement Xu et al. (2014); Sun et al. (2016a, b); Epping et al. (2017); Ribeiro et al. (2018), secure quantum channels Portmann (2017); Garg et al. (2017); Broadbent and Wainewright (2016), clock-synchronization techniques Jozsa et al. (2000); Komar et al. (2014) and distributed quantum computation Cirac et al. (1999) in general.

In principle there are mainly two approaches to construct quantum networks. On the one hand quantum networks could simply forward quantum information directly, which however needs to be protected against noise and decoherence using quantum error correcting codes Nielsen and Chuang (2010), and repeatedly refreshed at intermediate stations where error correction is performed Knill and Laflamme (1996); Zwerger et al. (2014); Muralidharan et al. (2014); Ewert et al. (2016). On the other hand, quantum networks may use a property which is only accessible in quantum physics, namely entanglement. Constructing quantum networks by using entanglement has one significant advantage compared to directly motivated approaches: The entanglement topology of a network, which determines in that case also the boundaries and ultimately the structure of a network, is completely independent of the underlying physical channel configuration. In particular, by characterizing quantum networks abstractly in terms of entangled quantum states allows for several interesting features which are not explicitly available in a direct approach, like e.g. creating shortcuts in a network on demand (by introducing an entangled state between parties) Schoute et al. (2016).

A crucial element to establish long-distance entanglement are quantum repeaters Briegel et al. (1998); Dür et al. (1999); Sangouard et al. (2009); Zwerger et al. (2018); Muralidharan et al. (2016); Munro et al. (2010); Azuma et al. (2015); Pirandola et al. (2017), and multiple proposals for repeater-based networks have been put forward Epping et al. (2016a); Hayashi (2007); Van Meter (2014); Pant et al. (2017); Das et al. (2017); Pant et al. (2017); Munro et al. (2015, 2008); Sangouard et al. (2011); Guha et al. (2015); Pirandola (2016); Meter and Touch (2013); Van Meter et al. (2013, 2009); Munro et al. (2010); Muralidharan et al. (2016) . Most schemes are based on bipartite entanglement, where Bell pairs are generated between nodes of the network. However, a future quantum network shall not be limited to the generation of Bell-pairs only Van Meter et al. (2011); Pirker et al. (2018); Cuquet and Calsamiglia (2012); Epping et al. (2016b), because many interesting applications require multipartite entangled quantum states. Therefore, the ultimate goal of quantum networks should be to enable their clients to share arbitrary entangled states to perform distributed quantum computational tasks. An important subclass of multipartite entangled states are so-called graph states Hein et al. (2004). Many protocols in quantum information theory rely on this class of states.

Here we consider entanglement-based quantum networks utilizing multipartite entangled states Pirker et al. (2018); Cuquet and Calsamiglia (2012); Epping et al. (2016b); Benjamin et al. (2006); Matsuzaki et al. (2010); Wallnöfer et al. (2016) which are capable to generate arbitrary graph states among clients. In general, we identify three successive phases in entanglement-based quantum networks: dynamic, static, and adaptive. In the dynamic phase, which is the first phase, the quantum network devices utilize the quantum channels to distribute entangled states among each other. Once this phase completes, the quantum network devices share certain entangled quantum states, which results in the static phase. In this phase, the quantum network devices store these entangled states for future requests locally. Finally, in the adaptive phase, the network devices manipulate and adapt the entangled states of the static phase. This might be caused either due to requests of clients in networks, but also due to failures of devices in a quantum network.

We follow the approach of Pirker et al. (2018) where a certain network state is stored in the static phase, and client requests to establish certain target (graph) states in the network are fulfilled by processing this network state using only local operations and classical communication (LOCC) in the adaptive phase. This has the advantage that requests can be fulfilled without delay, as the required resource states are pre-generated during the dynamic phase. In contrast, in complete dynamical networks requests are fulfilled by generating the required entanglement over physically available links on demand. This may be rather resource- and time consuming, and involves additional difficulties such as the so-called routing problem Van Meter et al. (2013); Schoute et al. (2016); Pant et al. (2017); Gyongyosi and Imre (2017, 2018, 2018); Pirandola (2016); Das et al. (2017); Hahn et al. (2018) where the goal is to determine a way of combining short-distance Bell-pairs to establish a long-distance Bell-pair in the most resource efficient way. In our approach, the problem is split into the generation of a universal network state (which also involves routing, but can be done prior to the request) in the dynamic phase, and its processing in the adaptive phase to establish desired target states. What all approaches have in common are two basic problems: (i) The complexity of how to organise quantum networking devices in a real network and how to systematically execute tasks in it; (ii) How entanglement can be established efficiently between networking devices in a dynamical manner. Despite the fact that some aspects have been addressed in recent works, it still remains unclear how the different techniques, ranging from the physical channel configuration, over the entanglement structure of a network to routing between quantum networks collaborate to enable for a feasible and tractable quantum network.

Classical computer networks tackle the complexity of transmitting bits between two nodes by breaking down the transmission into several layers of a stack model, the Open Systems Interconnection model (OSI model) Zimmermann (1980). In this model, information passes through seven layers, where each layer has a clear responsibility and adds additional descriptive information to the original message. Networking devices use this prepended information for various tasks. One of these layers is the network layer (layer three), which is responsible for logical addressing and routing in classical networks. Routing protocols for computer networks aim at determining a transmission path from a sender to a receiver by inspecting the descriptive information of the network layer. This task is accomplished by so-called routers, operating on layer three of the OSI model.

In this work we introduce a quantum network stack, which breaks down the complexity of entanglement-based quantum networks into several hierarchical layers of a stack model, similar to the OSI model. Each of the layers has a clear responsibility and can therefore be evolved independently in the future. We identify layers for ensuring connectivity at the lowest level (where quantum repeaters operate on), for generating graph states inside a network, but also for enabling inter-network graph state requests. Observe that in such a model, each layer uses its own set of protocols for accomplishing its associated responsibility.

After identifying the quantum network stack, we also present protocols which operate on the proposed layers. We start by proposing a protocol for the static phase in a quantum network for ensuring reliability. Then we discuss protocols tackling problems arising on connecting entanglement-based quantum networks via routers. For that purpose, we introduce the notion of a region, which is essentially a network of routers. The protocol we propose, which we refer to as routing protocol, operates in the adaptive phase and transforms a given entanglement structure between regions to a virtual network state among the requesting networks. This enables our network devices to fulfill arbitrarily distributed graph state requests in a straightforward manner. We also present protocols for the static phase of regions, especially to organize regions in a hierarchical manner and a technique to increase the reliability of regions. Finally, we define the term reachability for quantum networks. In such networks we say that a network (or network device) is quantum mechanically reachable if entanglement to the target can be established or is still present.

The main contributions of our work are as follows:

  • We introduce a quantum network stack for entanglement-based quantum networks and a classification of quantum network devices in accordance with this stack.

  • We identify techniques for quantum networks in the adaptive phase to guarantee intra-network reliability, which means that the network devices can deal with the failure of some networking devices without the need to re-establish entanglement.

  • We present a protocol which is capable of generating arbitrary graph states across network boundaries in the adaptive phase. We also find schemes to tackle the complexity arising in connecting quantum networks as well as reliability issues among regions, both crucial properties for the quantum internet.

  • We discuss reachability in and between entanglement-based quantum networks.

We outline the paper as follows. In Sec. II we review necessary background information as well as some previous works on quantum networks. Next we present our quantum network stack in Sec. III. Then, we discuss reliability for the link layer of our stack in Sec. IV and present protocols for the network layer in Sec. V. Some auxiliary protocols regarding reachability in quantum networks will be discussed in Sec. VI. We provide our conclusions in Sec. VII. Finally we present concrete, detailed examples of the routing protocol and of how the layers of the stack work together to generate a graph state in Appendix A and B respectively.

Ii Background

In the following we review relevant background information which we will use in the remainder of this paper. First, we recall how classical computer networks work and describe some classical routing protocols. Then, we introduce Dijkstra’s algorithm and Steiner trees on graphs. Next, we define graph states and the effects of applying single qubit unitaries and single qubit measurements on these states. Finally, we review former works on quantum networks which are of relevance for our work.

ii.1 Classical networks

Classical computer networks use the OSI stack model Zimmermann (1980). This model comprises the following seven layers:

  1. Physical layer: Responsible for the physical transport of bits. This corresponds to different physical transport channels such as wireless technologies (e.g. WIFI, Bluetooth, etc.) or ethernet.

  2. Data link layer: Responsible for managing the access to the physical layer of a network. In particular, protocols at this layer reduce collisions of data transmissions on the physical layer, due to several clients sending data at the same time in a network. This layer introduces hardware addresses, which are unique for every network interface.

  3. Network layer: Responsible for the logical decomposition of a network. Thereby, this layer defines what a logical network is, what its boundaries are and how routing between networks can be done. For example, the IP protocol ip (1981) introduces the notion of IP addresses on this layer, which identify network devices in a computer network. The network layer adds the IP addresses of the source and destination of a message to the packet, which routers then use to determine a transmission path towards the destination of the packet.

  4. Transport layer: Responsible for controlling the information flow on transporting information between two end nodes of a network. In order to distinguish between different applications/protocols which access the network, the transport layer introduces so-called ports. The destination and source port numbers are added to all packets passing through the transport layer. Furthermore, this layer also provides protocols, like e.g. the Transmission Control Protocol tcp (1981), which ensure reliable transmissions between or in networks.

  5. Session layer: Responsible for managing sessions in networks.

  6. Representation layer: Responsible for representing data appropriately.

  7. Application layer: Responsible for displaying the information to the end user (which may also be a computer program). This is the layer which is usually encountered by users of applications, e.g. email clients, web browser, games, etc. They use the lower layers to transport data through a network.

The third layer, i.e. the network layer, is of special interest in our context, since the information of this layer is used to route packets between two remote points in computer networks. In particular, a client hands over a packet to the network, and the network devices of layer three, which are also called routers, collaborate to determine a path to the recipient. Mainly there are two types of routing protocols: distance-based and link-state protocols. Distance-based protocols, like e.g. Routing Information Protocol (RIP) Malkin (1998), are not aware of the full network topology, whereas link-state protocols, e.g. Open Shortest Path First (OSPF) Moy (1998), internally keep a graph representation of the current network configuration. Link-state protocols typically use Dijkstra’s algorithm Dijkstra (1959) to determine the shortest path between two nodes.

ii.2 Dijkstra’s algorithm and Steiner trees

In this section we briefly recall two important algorithms in graph theory, which we will use here frequently.

The first algorithm is Dijkstra’s algorithm Dijkstra (1959). This algorithm is used to determine the shortest path between two vertices and of a graph . For that purpose it uses a cost function , which associates with every edge in the graph a certain cost. The algorithm is a greedy algorithm, which evaluates at each step whether a shorter path to a vertex is available. One may generalize Dijkstra’s algorithm by not only finding the shortest path to one particular but for finding a shortest path from to any vertex . We denote Dijkstra’s algorithm throughout this paper by where and .

The second algorithim we use in this work constructs an approxmiation of a so-called Steiner tree Hwang et al. (1992) on a graph. A Steiner tree on a graph is defined for a subset of vertices as a tree which connects the nodes of with minimal cost. The choice of the cost function depends on the discussed application. The tree is allowed to also contain vertices which are not an element of . If , the algorithm derives a minimal spanning tree.

An algorithm for approximating a Steiner tree for on with weighted edges is described in protocol 1.

0:  Set of nodes
0:  Starting node
3:  while  do
4:      … distances to
5:     for each in  do
7:     end for
11:  end while
12:  return  
Protocol 1 Steiner(, )

The problem of determining a Steiner tree has been shown to be NP-complete in the rectilinear case Garey and Johnson (1977).

ii.3 Graph states and GHZ states

Graph states Hein et al. (2004); Gühne et al. (2005); Hein et al. (2006); Tóth et al. (2006) are an important subclass of multipartite entangled quantum states. These states are associated with a classical graph where the vertices of correspond to qubits and the edges in indicate correlations among the vertices in . In particular, for a graph (where ) we define the graph state as the common eigenstate of the correlation operators


where and denote the Pauli matrices and the subscripts indicate on which qubit the Pauli operator acts on. We call two graph states and LU-equivalent if there exist unitaries such that .

In the following we discuss several important operations on graph states which we will use frequently here Hein et al. (2004, 2006). The first operation is local complementation, which acts just locally on the qubits of the graph state and transforms the graph state according to the following rule: If a local complementation at vertex is applied, then the subgraph induced by the neighbourhood of is inverted.

We further also require in this work Pauli measurements of qubits of a graph state. A measurement in the basis effectively removes the measured vertex from the graph, thereby also removing all the incident edges. Depending on the measurement outcome, some local Pauli corrections may have to be applied. A measurement in the basis corresponds to the following transformation of the graph state: First a local complementation at the measured qubit is done, followed by removing the corresponding qubit and all its incident edges. Again, depending on the outcome some local Pauli corrections may be required.

Finally we discuss GHZ states. An qubit GHZ state reads as


This state is LU equivalent to a fully connected graph or a star graph state, see e.g. Pirker et al. (2018). Here, we usually depict GHZ states using a star graph with a chosen root node. When we refer to a GHZ state, we may also refer to the LU equivalent graph state.

An interesting property of GHZ states is that if one combines two GHZ states via a Bell-measurement, then the result is again a GHZ state, up to local Pauli corrections. More specifically, by measuring a qubit of and a qubit of with a Bell-measurement leads (up to LU) to the state . We will use this property extensively in this work. In principle one can also use a different measurement setup to transform the states and to the state , see e.g. Wallnöfer et al. (2016).

ii.4 Quantum networks

There is extensive work on quantum networks relying on quantum repeaters Epping et al. (2016a); Hayashi (2007); Van Meter (2014); Pant et al. (2017); Das et al. (2017); Pant et al. (2017); Munro et al. (2015, 2008); Sangouard et al. (2011); Guha et al. (2015); Pirandola (2016); Meter and Touch (2013); Van Meter et al. (2013, 2009); Munro et al. (2010); Muralidharan et al. (2016). Most of these approaches have in common that they assume a network of quantum repeaters sharing Bell-pairs with each other.

However, in general, the goal of a quantum network should be to generate arbitrary states between remote clients rather than solely Bell-pairs. For many applications, it however suffices to be able to generate a specific class of states. Graph states Hein et al. (2004, 2006) play an important role in this respect, and the generation of arbitrary graph states among clients has been identified as a desireable goal for quantum networks Van Meter et al. (2011); Pirker et al. (2018). This is what we will require from our network in the following. For that purpose, several different approaches regarding the entanglement structure may be pursued which include: (i) The usage of a central master node which creates the state locally and teleports it to clients via Bell-pairs shared between the central none and all others; (ii) to establish pairwise entanglement between all of the network nodes first, followed by combining or merging it in an appropriate manner; (iii) generating the target state directly in a distributed manner by using multipartite states.

Approaches (i) and (ii) are far better understood than (iii), due to in-depth knowledge about bipartite entanglement and quantum repeater networks. Using a central master node which creates the requested graph state and teleports it to clients thereby consuming Bell-pairs basically suffices to generate any arbitrary entangled state between the clients. However, this approach has one significant drawback: If the central master nodes fails, then the whole network is down. This motivates a decentralized approach, leading to quantum repeater networks.

In a network of quantum repeaters sharing bipartite entanglement, depending on the requested target state between the clients of the network, the intermediate quantum repeaters employ entanglement distillation and swapping operations (or other kinds of repeater protocols) to establish the required long-distance Bell-pairs subject to merging to generate the target state. In order to establish this long-distance Bell-pairs, routing in the repeater network needs to be done.

Recently several quantum routing protocols for bipartite quantum repeater networks were presented Van Meter et al. (2013); Schoute et al. (2016); Caleffi (2017); Gyongyosi and Imre (2017, 2018); Pirandola (2016); Das et al. (2017). In this context, the goal of a routing protocol is to determine a way of combining short-distance Bell-pairs to establish a long-distance Bell-pair in the most resource efficient way. Ref. Van Meter et al. (2013) studies the application of Dijkstra’s algorithm to quantum repeater networks, where the cost associated with an edge in the repeater network, i.e. a small-scale Bell-pair, depends on several physically motivated parameters, e.g. Bell-pair generation rate, transmittance, etc. A routing algorithm for ring and sphere type network topologies of repeater networks has been proposed in Schoute et al. (2016). In Pant et al. (2017) a routing protocol for a two-dimensional cluster-type network relying on Bell-pairs was proposed. Another algorithm for optimal routing in an end-to-end setting was subject of study in Caleffi (2017). Routing using an entanglement-gradient in quantum networks was studied in Gyongyosi and Imre (2017). Ref. Gyongyosi and Imre (2018) constructs a so-called base graph which represents the optimal entanglement structure of a repeater network to determine optimal paths, and a method for adapting this graph, e.g. due to node failures, was studied in Gyongyosi and Imre (2018). In Pirandola (2016) lower and upper bounds on the end-to-end capacities in arbitrarily complex quantum networks for single and multipath routing strategies for a single sender and a single receiver, but also for multiple senders and multiple receivers ultimately sharing bipartite states among each other simultaneously were established. Finally, in Das et al. (2017) the routing of Bell-pairs in memory-free, two dimensional quantum network was investigated, where intermediate workstations either generate Bell-pairs or perform entanglement swapping, both in configurable directions, thereby achieving routing for establishing Bell-pairs in the network.

Routing protocols using multipartite quantum states received far less attention. Recently it was shown that routing on a cluster state, which is shared among network nodes, using local complementation and measurements in the basis provides an advantage compared to routing based on Bell-pairs Hahn et al. (2018). Here the main goal was to generate one or several Bell-pairs simultaneously from the cluster state. In addition, they show that by slighting modifying the protocol it can be used to generate GHZ states from the cluster state. An algorithm which closely relates to routing which uses multipartite states was presented in Dahlberg and Wehner (2018). In particular, the algorithm of Dahlberg and Wehner (2018) decides whether a certain stabilizer state (which includes graph states) can be transformed into another stabilizer state by using single qubit Clifford operations, single qubit Pauli measurements and classical communication. In Dahlberg et al. (2018) the complexity of such transformations between graph states were studied and it was shown that this task is in general NP-complete.

As discussed in Sec. I, we follow in this work mainly the entanglement-based multipartite approach to quantum networks as presented in Pirker et al. (2018). Therefore, we recall several concepts which were introduced there. Clients are assumed to be of minimal functionality (single qubit unitaries and single qubit measurements), and are connected to quantum network devices by sharing entanglement with them, for example in form of Bell-pairs. The network devices, i.e. routers and switches, use an internal multipartite quantum state, which we refer to as device state, to connect their clients. Finally, networking devices connect to each other by sharing multipartite entangled quantum states, referred to as the network state. This is illustrated in Fig. 1.

Figure 1: A concrete network example: Three switches (boxes with horizontal arrows) and a router (box with diagonal arrows) connect in a network via GHZ states of decreasing size (black lines indicate entanglement). Internally each of these networking devices use again GHZ states of decreasing size to connect three clients each. In this example, the clients connect via Bell-pairs to the network devices. Observe that the entanglement structure of the network is different from the physical channel configuration (orange tubes).

The aim of the network is to generate arbitrary graph states between its clients on demand. Because the clients have only minimal functionality, the quantum network devices have to carry out the generation of the target state. Since the target state is not known prior to a request, the state of the static phase, i.e. the network and device states, need to be such that any graph state can be generated from them using only LOCC, without generating additional entanglement. For that purpose, two different types of multipartite states for device and network states were identified: multiple copies of GHZ states of decreasing size, or partite, fully connected graph state with decoration qubits on each edge (decorated graph state) where corresponds to the number of nodes which shall connect. Observe that multiple copies of GHZ states of decreasing size are mandatory for the network state to enable for arbitrary graph states in the network, see Pirker et al. (2018) for a detailed discussion. Depending on the desired target states of the network, only a subset of these states may be required.

We also relate the work of Pirker et al. (2018) to the phases of Sec. I as follows: The static phase corresponds to the entanglement structure in the devices and across the network, i.e. the network states. These states are generated during the dynamic phase, where the networking devices utilize the quantum channels to distribute the required entanglement. One way of doing this is to use the quantum network configuration protocol (QNCP) as discussed in Pirker et al. (2018). In the adaptive phase, the clients request a graph state from the network, and the quantum network devices manipulate the states of the static phase in such a way, that the target state is established between requesting clients. For that purpose, the quantum networking devices apply controlled phase gates (CZ gates), measurements in the or basis, and Bell-measurements to connect the network and the device state.

Finally we also comment on why following a direct multipartite approach is indeed benefical compared to using Bell-pairs between network devices in our setting. If network devices would share only Bell-pairs among each other, then, depending on the requested graph states, the network devices have to apply more CZ gates, Bell-measurements and single qubit measurements in the or basis in contrast to a multipartite approach. Observe that if these operations are noisy, this will result in a state of smaller fidelity compared to directly using multipartite quantum states, see Wallnöfer et al. (2018). Furthermore, such a direct multipartite approach offers a storage advantage compared to bipartite schemes Pirker et al. (2018); Wallnöfer et al. (2018).

Most works on quantum networks, in the bi- and multipartite case, have in common that there is no clear notion of which quantum task has to be done by which node in the network at which time. In particular, how to organize and classify quantum networking devices depending on their (to-be-defined) capabilities is not fully known yet. However, similar issues arising in quantum repeater networks have e.g. been addressed in

Meter and Touch (2013), which resulted in a stack model for quantum repeater architectures.

Iii A quantum network stack

In classical computer networks, communication in a network follows the OSI layer model, see Sec. II.1 for more information. This model vertically breaks down the complexity of networks into several layers. Each of these layers takes data (in form of a packet) from the layer above, and passes it, after optionally prepending the packet by additional descriptive information, to the layer below.

In contrast to the classical network stack where descriptive information is added we assume that qubits of neighbouring layers in the quantum network stack can be accessed and combined.

Our proposal for a quantum network stack for quantum networking devices is depicted in Fig. 2.

Figure 2: Proposal for a quantum network stack comprising four layers. We identify a physical layer (channel configuration), a conncetivity layer (for ensuring connectivity in terms of high fidelity entangled states between network devices), a link layer (comprosing a single network by sharing a multipartite entangled network state) and a network layer (connecting quantum networks via routers).

Each layer has a specific goal, meaning that we break down the responsibilities in a quantum network vertically by clearly defining the objectives of a particular layer. One important feature of such a stack is that different layer can be evolved and studied independently. More specifically, changing protocols at higher layers of the stack does not imply changes to the lower layers, i.e. lower layers do not depend on concrete implementations of higher layers. As already shown in Fig. 2, depending on the layer a network device operates on, it has access to more or less layers of the quantum network stack.

In the following subsections we elaborate on each of these layers in detail. Before doing so we want to emphasize, that if a client of a quantum network has sufficient capabilities, then it may work even on top of the network layer of Fig. 2, which means that it can act as a router or switch in the network. Observe that in such a case, the protocols itself remain unchanged. Furthermore, a client can employ verification techniques and applications to the final states after graph state generation.

We provide a complete example of how the layers of our quantum network stack work together for a particular request in Appendix B.

iii.1 Layer 1 – Physical Layer

This layer corresponds to the quantum channels connecting the interacting quantum network devices, for example optical fibres or free space channels. It is responsible for forwarding qubits from one network devices to the other, without applying any error correction or distillation mechanisms. The setting of layer 1 is depicted in Fig. 3.

Figure 3: Setting of layer 1 and 2: At layer 1, quantun channels connect quantum networking devices. Neighbouring quantum networking devices operating on layer 2 utilize the quantum channels to create long-distance entangled quantum states. In general, the entanglement structure at layer 2 is independent of the physical channel configuration of layer 1.

iii.2 Layer 2 – Connectivity Layer

This layer tackles errors due to imperfections in the quantum channels of layer 1. On this layer, the techniques for establishing long-distance quantum communications reside. In particular, concrete technologies include quantum repeaters, bi- Briegel et al. (1998); Dür et al. (1999); Sangouard et al. (2009); Zwerger et al. (2018); Azuma et al. (2015); Epping et al. (2016a); Hayashi (2007); Van Meter (2014); Pant et al. (2017); Das et al. (2017); Pant et al. (2017); Munro et al. (2015, 2008); Sangouard et al. (2011); Guha et al. (2015); Pirandola (2016); Meter and Touch (2013); Van Meter et al. (2013, 2009); Munro et al. (2010); Muralidharan et al. (2016) or multipartite Wallnöfer et al. (2016, 2018), but also the direct transmission of encoded quantum states Knill and Laflamme (1996); Zwerger et al. (2014); Muralidharan et al. (2014); Ewert et al. (2016). Therefore the main purpose of devices operating on this layer is to enable point-to-point or point-to-multipoint connectivity without any notion of requests. This functionality is crucial for the dynamic phase of a quantum network, where network devices have to (re-)establish entanglement, i.e. network states, required in the static phase. Observe that, if quantum repeaters are in use, the entanglement structure in terms of Bell-pairs can be independent (or different) from the configuration of quantum channels at layer 1.

Higher layers, i.e. devices operating on a higher layer like e.g. switches or routers, utilize this layer to establish multipartite entangled quantum states within a network or between several independent networks, see Fig. 3.

The layer above, i.e. the link layer, is independent of the technology which this layer employs. It simply instructs this layer to perform certain tasks within the network, like e.g. establishing a long-distance Bell-pair or a GHZ state between other high-level networking devices. Such instructions may also involve several devices of this layer.

Since the layers above (link and network layer) are completely decoupled from the connectivity layer, devices at layer 2 can apply enhanced techniques, like e.g. finding optimal paths in quantum repeater networks (see Sec. II.4) by routing, without affecting upper layers of our stack.

iii.3 Layer 3 – Link layer

The link layer defines the boundaries of a quantum network in terms of an entangled, distributed, multipartite network state which the networking devices of a quantum network share in the static phase. This layer utilizes the connectivity layer to establish the entangled network state during the dynamic phase, which therefore enables also for long-distance quantum networks. Once the dynamic phase completes, the link layer devices (switches) share multipartite entangled states which comprise the network state, see Fig. 4. Thereby we end up in the static phase. We observe that the entanglement structure can be completely different from the underlying configuration of quantum channels and devices or protocols operating at layer 2. Concrete instances of networks may connect in the static phase via e.g. GHZ states or decorated graph states (see also Sec. II.4), as proposed in Pirker et al. (2018).

Figure 4: Setting of layer 3 and 4: The network devices of layer 3 request and combine the entangled states from layer 2 to create the network state (green nodes). Depending on client requests, this network state is consumed during graph state generation. On layer 4, quantum routers connect quantum networks via multipartite entangled quantum states (red nodes). In this figure, the quantum routers connect via GHZ network states.

The link layer orchestrates and coordinates the process of generating the network state, utilizing the connectivity layer, in the dynamic phase and is responsible for generating arbitrary graph states between clients of the network during the adaptive phase via a so-called linking protocol, see Fig. 2. The linking protocol is responsible for transforming the entangled network state and device-internal states using only LOCC to the requested graph state, which consumes the entanglement of the network state. This linking protocols depend on the network state. For example, for GHZ or decorated architectures, devices at this layer may invoke the linking protocols of Pirker et al. (2018) to create the requested graph state.

Furthermore, the link layer also has the capability to invoke entanglement distillation protocols for two-colorable graph states Dür et al. (2003); Aschauer et al. (2005); Dür and Briegel (2007) which ensure that the network state has a sufficiently high fidelity.

iii.4 Layer 4 – Network layer

The network layer is responsible for generating and manipulating inter-network entanglement to enable graph state requests spanning several different quantum networks. The network devices operating on this layer are quantum routers. They are connected with each other via multipartite entangled quantum states in so-called regions in the static phase, similar to quantum networks at the link layer. The corresponding states depend on the protocol of this layer. The overall setting is illustrated in Fig. 4.

Regions connect in the same fashion as quantum networks do. More specifically, quantum routers in the same region share a multipartite entangled quantum state with each other, like e.g. a GHZ network state, but in contrast to a switch, a quantum router may be part of several regions at the same time. Furthermore, a router may also be part of a quantum network of the link layer, thereby providing an entry point to that quantum network from the viewpoint of other networks. We outline available operations and protocols below, and discuss them in more depth in Sec. V for regions connecting via GHZ states. Observe that lower layers, like e.g. link or connectivity layer, are independent of the protocols and considerations of the network layer.

The network layer is responsible for enabling for graph state requests across network boundaries in the adaptive phase. Therefore, some sort of routing between different quantum networks needs to be done. More precisely, to enable for graph state requests across networks, a quantum routing protocol should establish a ”virtual network state” between the quantum networks which are part of a graph state request. The topology of the virtual network state will depend on the routing protocol. Note however that this only involves local manipulation of entangled states that are already present in the network from the static phase, and no generation of additional entanglement is required. Hence these requests can typically be fulfilled fast.

Once routing finishes, the routers use this ”virtual network state” of the network layer to establish a network state between the requesting network devices inside their respective networks. Routers achieve this by combining the virtual network state of layer 4 with the inner network state of layer 3 by local operations, for example in terms of Bell-measurements for GHZ states or controlled phase gates and measurements in the basis for decorated architecture. However, the output of the routing protocol creates a full network state among the requesting network devices. It might be necessary to transform the output state of routing into an appropriate form to combine it with the intra-network states. We note that the link layer itself is not involved in this routing process.

We discuss a routing protocol for regions using GHZ network states in Sec. V.1 and approaches for simplifying the complexity in regions and introducing reliability for connecting routers using GHZ network states in regions in Sec. V.2 and Sec. V.3 respectively.

Finally, the network layer can also invoke entanglement distillation protocols for two-colorable graph states from the set of auxiliary protocols at all layers, see Sec. III.5.

iii.5 Auxiliary protocols

As illustrated in Fig. 2, each layer has access to some auxiliary protocols. The network devices use these protocols to e.g. to (i) generate high-fidelity entangled quantum states, (ii) check whether network devices operating at the same layer are still reachable, (iii) perform entanglement swapping or merging, or (iv) employ techniques for error correction. Depending on the layer a quantum network device operates on it will use different subsets of the aforementioned protocol types.

The protocols of (i) are entanglement distillation protocols. The layers use these protocols to generate high-fidelity entangled quantum states by transforming several noisy input states to fewer, but more entangled copies. We can associate to each layer one class of entanglement distillation protocols. For example, layer 2 may use entanglement distillation protocols for Bell-pairs Bennett et al. (1996); Deutsch et al. (1996), whereas layer 3 and layer 4 need access to the entanglement distillation protocols for two-colorable graph states Dür et al. (2003); Aschauer et al. (2005); Dür and Briegel (2007).

The protocols of (ii) address reachability in quantum networks. In entanglement-based quantum networks, there are different forms of reachability. At a basic level, this is about the (classical) reachability of the corresponding network device. However, this is not sufficient as also the presence of the required entangled states needs to be ensured. We discuss this issue in detail in Sec. VI.

Entanglement swapping and merging, i.e. protocol type (iii), are operations that are crucial for repeater architectures, but also for the modification of entangled states. Entanglement swapping corresponds to a Bell-measurement which is applied to one qubit of two Bell-states each plus the classical communication of the measurement outcome. Such a measurement establishes again a Bell-pair between outer nodes and is usually used to generate long-distance Bell-pairs.

Finally, the techniques of (iv) correspond to quantum error correcting codes, which networking devices may use to tackle channel noise and loss which occurs during the dynamic phase, but also on storing qubits which are part of larger entangled states for a longer time in quantum memory in the static phase.

Additional auxiliary protocols may be added on demand.

Iv Layer 3 – Reliable state linking

In this section we discuss how to achieve reliability at the link layer using multipartite entanglement, which is of high importance for the static phase of a quantum network. The term reliability means in our case that parts of the entanglement structure in a quantum network remain intact, i.e. usable for other devices, if one network device disconnects without performing any further operation.

Before discussing the reliability of multipartite networks, we review its issues arising in bipartite networks using Bell-pairs. In this case, reliability highly depends on the topology of the distributed Bell-pairs.

For example, consider a quantum network with a central master node sharing Bell-states with all clients. Clearly, if this central master node disappears, all Bell-pairs are lost, and hence, no further communication is possible 111Such an effect may be tackled by introducing several master nodes, which implies that large quantum memories will be required at these master nodes. In a fully bipartite approach, where all quantum network devices connect to each other via Bell-states in a decentralized manner, and clients only connect to these network devices, we note that the problem of reliability disappears. The failure of a node only affects the entangled pairs the node is part of, all other Bell pairs remain undisturbed.

Here we consider the GHZ architecture within networks for our reliability protocols in quantum networks, see Pirker et al. (2018). Recall that in this architecture network devices which reside within the same network share multiple copies of GHZ states of decreasing size. More specifically, the network state connecting devices is, up to several copies of the states, of the form


We now illustrate why it may be a problem to directly use GHZ states without any further modification as network state. For that purpose, suppose one network device disconnects from the network without performing any disconnecting operation on his qubits of the network state. Such a disconnect corresponds to tracing out all the qubits of that particular network device. The state after tracing out any qubit of one GHZ state of (3) results in


which is a separable state. Therefore, losing one qubit due to a disconnect will destroy the entanglement between all other network devices which are part of that GHZ state. The situation is shown in Fig. 5. As depicted in Fig. 5, depending on which network device disconnects, we may even loose all network states. In particular, if one of the network devices which connect via (i.e. or ) disappears, all network states will be lost, since these network devices store one leaf of each GHZ state of (3).

Figure 5: GHZ states are very fragile: If one of the network devices disappears, then at least one GHZ state is completely lost (grey vertices – entanglement lost), as disconnecting corresponds to tracing out a qubit from a GHZ state which results in a separable state. The green vertices are not affected by the disconnect.

Because we cannot predict which network device will fail or disconnect, we have to find solutions which are able to deal with the disconnect of any of the network devices such that the functionality for the remaining system is preserved. Nevertheless, we find that schemes using multipartite entanglement are still more beneficial in terms of storage size compared to a full bipartite approach.

iv.1 Reliable state linking – Symmetrization

In general, several copies of each GHZ states in (3), which comprises the network state, are mandatory to enable for arbitrary graph state requests in a network.

The first solution we propose is to symmetrize the network state. In particular, we circularly shift the parties of the network w.r.t. to their assignment to leafs and roots of the GHZ states of Eq. (3). The situation is summarized in Fig. 6.

Figure 6: a) Static phase: The first solution to tackle device failures is to rotate the full network state, i.e. cyclically shift the parties. For devices cyclically shifting the network state times results in different configurations. More specifically, we split the copies of (3) into configurations, where each configuration is obtained by cyclically shifting the root-leaf assignment in (3). A further extension might be a full symmetrization among the network devices. b) Adaptive phase: If one network device disconnects, then there exists at least one configuration which ensures full connectivity among the remaining devices at least copies of the network state of (3) remain intact. But there exist also other configurations remaining partially intact.

Observe that by symmetrizing the root-leaf assignment in the state of (3), each network device stores several roots of each GHZ state where . We call the state obtained after one cyclic shifting step a configuration.

The crucial observation is that if one network device disconnects, there exists one configuration for which the disconnecting device holds the root of the largest GHZ state of (3). All other network devices connect in this configuration via the states where . Therefore, the disconnect of this network device only destroys the largest GHZ state in that configuration whereas all other states in this configuration remain intact. The situation is summarized in Fig. 6. In order to tackle arbitrarily losses, including failures of multiple nodes, we suggest to use a full symmetrization according to all possible permutations of network devices.

iv.2 Reliable state linking – Shielding

The second solution to ensure reliability in a quantum network is to introduce shielding qubits. More specifically, we place on each edge of the GHZ states in (3) one additional qubit (which we call the shielding qubit of that edge in the GHZ state), except the Bell-pair, of the network state. We consider the graph state corresponding to this decorated graph, where we use the star graph to represent the initial GHZ state. This shielding qubit belongs to the network device which holds the root of the respective GHZ state, see Fig. 7.

Figure 7: a) Static phase: The second solution to achieve reliability in a quantum network using entanglement in form of GHZ states is to introduce shield qubits. We place on each edge of every GHZ state (represented by a star graph) one qubit and consider the resulting graph state. Only the Bell pair needs no decoration, as entanglement is gone in case of failure of either of the two nodes. b) Adaptive phase: Because tracing out a qubit commutes with measurements on other qubits, the remaining devices just have to measure the shield qubits to the disconnected device in the basis. The remaining devices will still have a full network state.

The crucial observation for the shielded GHZ states of Fig. 7 is that, if a network device disconnects, which corresponds to a trace out of its qubits, the remaining network devices preserve their states by measuring all the shielding qubits to the disconnecting network device in the basis. This can easily be seen since the trace out operation commutes with the measurement of neighbouring qubits, which effectively decouples the part of the shielded GHZ network states corresponding to the disconnecting network device, see Fig. 7.

However, if no error occurs, the networking devices can reduce the shielded GHZ states to a GHZ state by measuring the shielding qubits in the basis, which establishes wires to other network devices, see e.g. Pirker et al. (2018). Observe that depending on the measurement outcomes, some Pauli corrections may be necessary.

We compare the number of qubits necessary in this shielded GHZ approach to a full bipartite solution, see beginning of Sec. IV, solely using Bell-pairs, because this scheme automatically ensures reliability in quantum networks.

In Pirker et al. (2018) it was shown that the number of qubits of a GHZ network state for a network of devices connecting clients respectively after expanding the network state to all connected clients is


We explain (5) as follows: Network device connects to the network devices via copies of the GHZ state . Each copy of that GHZ state corresponds to the adjacency of one client of network device to the clients located at the network devices . To take into account for all this adjacencies, the network devices expand each of the copies of the GHZ state to via Bell-measurements. We refer to this state also as expanded network state.

Recall that we decorate each edge of the GHZ network state once, and that device has copies of the state . Therefore, the total number of qubits which have to be stored including the shielding qubits is


For the number of qubits necessary in following a direct bipartite approach one finds (see Pirker et al. (2018)) that


qubits are required in total. We compare the number of qubits of the shielded GHZ network state and the bipartite approach for various scenarios in Table 1.

Table 1: In this table we compare the number of qubits which have to be stored in a direct bipartite approach solely using Bell-pairs to the shielded GHZ network state, and the GHZ network state without shielding with different number of clients and devices.

From Table 1 we find that even though it seems at first glance that shielding the GHZ network state will introduce a large overhead, it still results in better performance in terms of qubits to be stored compared to a direct bipartite approach. The reason for this is that we only place qubits on the edges of the network state before expansion, and not for the expanded network state, see paragraph below Eq. (5).

V Layer 4 – Region routing, Hierarchical regions and Reliable regions

In this section we discuss protocols operating on layer 4 of our quantum network stack and use GHZ network states to connect routers into regions. Recall that the purpose of this layer is to enable for inter-network graph state requests by LOCC.

We start by proposing a routing protocol for the adaptive phase in quantum networks for quantum routers in Sec. V.1, which may connect in a highly irregular manner. Next we introduce a way of reducing the size of network states appearing in the static phase for connecting routers in a region in Sec. V.2, which also reduces the complexity of regions. Finally, we discuss how to achieve reliability for connecting routers in regions in Sec. V.3.

v.1 Region routing

We start with a brief review of classical routing protocols. In classical routing, there exist protocols using metrics (like e.g. RIP Malkin (1998)) and so-called link state protocols (like e.g. OSPF Moy (1998)) - see Sec. II.1.

Protocols using metrics internally construct a so-called routing table. Each table entry is a key-value pair with key corresponding to a network address and values corresponding to distance and interface port of the router to which packets shall be forwarded.

Link-state protocols operate in a different manner. They internally construct a global view of the network topology, i.e. a weighted graph where the weight of an edge corresponds to the distance or cost between two nodes of the network. Depending on the destination IP address of an incoming packet, routers compute a minimal cost path through the network by using Dijkstra’s algorithm, see Sec. II.2. The protocol which we propose follows a similar approach as link-state protocols in classical networks.

Before we start with the protocol description we first recall that we abstract quantum networks via routers as in classical networks. The router provides, according to our stack of Sec. III, an entry point to a quantum network. For simplicity we assume that there is only one router in each network. Several routers in one network can be taken into account as follows: If there are two routers, then there exists at least one part of the network state which connects those routers. One of both routers can teleport all of its qubit which belong to another region to the other, thereby providing a single entry point to the network under consideration.

In Sec. III.4 we identified the goal of the network layer, and therefore also routing, as follows: Routing protocols in quantum networks should establish a virtual network state across routers, as this enables routers of networks to combine the virtual network state with the respective inner network states of each router to fulfill graph state requests across network boundaries. The situation is summarized in Fig. 8.

Figure 8: The goal of a quantum routing protocol: A routing protocol shall generate a ”virtual network state” between routers of different quantum networks which are involved in a request. After that the routers combine the virtual network state with the inner network state to transfer the entanglement to the requesting devices. This enables the networking devices to directly apply the graph state linking protocol of Layer 3 to complete the request.

Now we discuss a routing protocol which achieves the aforementioned goal. For that purpose we assume that routers of networks connect via GHZ network states in regions. Such a scenario corresponds to the case when a network administrator defines which routers shall connect in a region.

Such a configuration of routers in regions, and thereby also the configuration of network states, may be highly irregular. We stress that such a scenario is of high relevance for practical settings, as it enables network administrators to define network boundaries and which networks connect to each other in regions. The network administrator may have knowledge about the traffic which clients in quantum networks produce and tries to minimize the overall entanglement cost associated with network states.

The goal of the routing protocol is now as follows: Clients, possibly located in different networks, wish to generate a particular graph state. The aim of the region routing protocol is to establish a virtual network state across the routers of networks involved in a request, see Fig. 9.

Figure 9: Regions and how they connect: A region (indicated by dashed lines) connects routers (boxes) via a GHZ network state. A router may be part of several regions. If clients of specific networks request a graph state, the region routing protocol shall establish a virtual network state among the routers of these networks. This state can then be used to generate a network state between graph state requesting quantum network devices.

In order to generate such a virtual network state we make use of Dijkstra’s algorithm (which also some works on routing for quantum repeater networks use, e.g. Van Meter et al. (2013)) and the algorithm for determining a Steiner tree as discussed in Sec. II.2. We remark that Steiner trees have also been used in Bäuml and Azuma (2017) for deriving fundamental limitations on quantum broadcast channels. These algorithms require the definition of a cost function for the edges of the graph. For illustration purposes, we use the number of states the routing protocol consumes as cost, i.e. each edge in the graph has unit cost. We discuss more appropriate cost functions later on.

The input to the region routing protocol is a set of networks (or more precisely, the routers of the networks) which connect via regions. Each router corresponds to a vertex in the graph corresponding to the configuration of regions. Suppose a subset of the vertices request a graph state. Then we perform the following algorithm:

0:  Set of nodes , set of networks
1:  Transform the graph of vertices (networks) to a classical graph, where qubits belonging to the same network fusion into one vertex. We denote this graph by .
2:  while  do
3:     Select
5:     Generate with root at according to
7:  end while
Protocol 2 RegionRouting()

In step 1 of protocol 2 the routers transform the configuration of states connecting the regions to a classical graph by merging all qubits of a router to a single node. However, observe that the routers have to keep track of the states (which correspond to edges in the graph) the Steiner tree algorithm selects internally. The routers further optimize the consumption of states by minimizing the multipartite entanglement they select between regions for a request. An alternative approach for step 1 is to generate the classical graph which the Steiner tree algorithm requires by associating a qubit with a vertex in the classical graph, and creating an edge for every possible Bell-measurement which can, in principle, connect two regions.

Observe that the while-loop of protocol 2 creates one of the GHZ states of the virtual network state. The steps 3 - 5 create one part of the GHZ network state with root located at by using a Steiner tree between the remaining routers in . A complete example for the routing protocol is provided in Appendix A.

We remark that protocol 1 of Sec. II.2 to determine a Steiner tree requires the usage of a specific cost function. The cost function should take into account for the cost to generate and combine bi- or multipartite states at different layers. Channel noise as well as noise in local operations are relevant for the performance of entanglement distillation protocols and the combination of different states (e.g. via entanglement swapping or merging), and hence determine the cost. We leave a discussion for proper choice of cost function to future work, but remark that the usage of multipartite entangled states can also be beneficial in this respect Dür et al. (2003); Aschauer et al. (2005); Dür and Briegel (2007).

v.2 Hierarchical regions

One obvious way of connecting quantum networks in a region is to connect all in the same fashion as a quantum network connects its devices, i.e. in a single region. In that case, only measurements on the GHZ states connecting the networks are necessary to establish the state depicted in Fig. 8. However, such an approach has one serious drawback: The size of the network state will increase with the number of networks, i.e. the number of routers. In particular, to connect routers in a single region, the largest GHZ state connecting them is of size . In a practical realization, the size of GHZ states might however be limited. The reason is that GHZ states suffer from noise and decoherence, and are in fact more fragile with increasing size Dür and Briegel (2004); Hein et al. (2005). Also the approach discussed in the previous section might be impractical. The network administrator plays a key role, as he defines which routers shall connect in a region. This determines the topology of the network, and a proper knowledge of the underlying traffic is crucial for an efficient choice. However, if the the administrator does not have this knowledge prior to region design, or there are unexpected fluctuations, the topology might be inefficient.

The method we propose here is an automatic and efficient scheme for connecting routers into regions in a hierarchical manner. The key element is to use only GHZ states of limited size, and arrange the regions in a hierarchical manner. This avoids the problem of fragile, large GHZ states. In addition, regions can be arranged on demand, e.g. optimized w.r.t. expected traffic. We remark that such a hierarchic arrangement was also implicitly assumed in Wallnöfer et al. (2016); Pirker et al. (2018). The features of such hierarchical graphs and their properties in a network structure has recently been analyzed in detail in Bapat et al. (2018).

In our case, regions connect via a GHZ network state, and we fix the maximum number of routers which are part of such a region. This effectively limits the size of the GHZ network state connecting the routers. Regarding fragility, we remark that in fact a three qubit GHZ state can accept more local depolarizing noise per particle than a Bell pair - only for larger particle numbers there is an increased fragility Wallnöfer et al. (2016). The situation for the case of is depicted in Fig. 10.

Figure 10: The figure depicts the state for connecting networks in hierarchical regions for : The entire network of routers is broken down into smaller regions, sharing again a network state, with size of at most (green). One of each routers of a regions connects to the next hierarchical level again via a GHZ network state (red).

We can also view these hierarchical regions as substructures inside the network layer of the quantum networking stack. More specifically, the hierarchical layers may be considered as high layers in the stack. We call a router which connects regions at different hierarchical levels also a designated router (DR). These routers enable to reach routers/networks which are located at regions at different levels of the hierarchy.

The process of establishing and connecting routers in regions can be done automatically: If a new routers starts, it simply discovers all previously appeared routers classically. Then, the designated routers checks if the new router fits into any of the existing regions. If so, then the new router will be added to that region. If not, a new hierarchical level is created, thereby creating an ary tree of regions. Notice that in principle, one can choose the position of routers in the hierarchy also w.r.t. to certain parameters, like e.g. traffic. This enables one to place routers of networks which often demand requests in the same region, thereby optimizing the regions between quantum networks. We also observe that the complexity of the region routing protocol decreases due to such a hierarchy of regions.

v.3 Reliable regions

Finally we want to discuss the reliability of regions. In principle, since regions connect via GHZ network states, they will suffer from similar problems as quantum networks at layer 3 if a router disconnects.

The schemes we discussed in Sec. IV enable for compensating the fail of network devices in a network, leaving the remaining devices within the network with a functional network state.

However, in the setting of regions the situation is more involved, since the failure of one router at the boundary of a region will disable also all other routers of the same region to generate graph states to other regions.

Such an issue can be solved via symmetrization inside regions. In particular, suppose regions connect with a GHZ network state. Instead of distributing the qubits of the GHZ network states to one particular router of each region, we symmetrize the GHZ states inside the respective regions. For example, in the case of the regions we distribute the qubits of the largest GHZ to different routers inside the regions . In particular, for region , we can assign the qubits to different routers, in region , we can distribute the qubits to different routers and so on. The scheme is illustrated in Fig. 11.

Figure 11: Symmetrization of a network state between regions (shown for three qubit GHZ state): Symmetrization is done w.r.t. all possible permutations of routers residing in different regions. Observe that the inter-region entanglement is still of GHZ type.

This result in total in


different qubit GHZ states. Observe that the GHZ-type of entanglement between the regions is preserved by this symmetrization. The same procedure is also applied to the smaller GHZ states between regions.

This technique will introduce complexity to the region routing protocol and therefore, one may use only one specific permutation in regular operation mode, e.g. the configuration . All other permutations are only used in case of route failures. However, note that multiple copies of the network states between regions are required anyway, so this protocol does not introduce an additional cost or effort.

Vi Auxiliary – Reachability protocols

In this section we discuss protocols for checking whether entanglement connectivity between quantum networking devices is still present or not. Such a quantum ping is of high importance for several layers, which motivates us to put them as auxiliary protocols in our proposed quantum stack. In particular, at layer 2 of the quantum networking stack one may use quantum repeaters to establish entanglement between remote nodes. At layer 3, the switches of a network might also want to check whether the entanglement among each other is still present. Finally, routers could want to determine whether a network is reachable via some intermediate regions. For that purpose, first routing towards a network has to be done.

In classical networks, the reachability of network devices is achieved by so-called pings icm (1981). The ping mechanism is part of the Internet Control Messaging Protocol icm (1981) which operates above the Internet Protocol ip (1981) at layer 3 of the classical network stack. The protocol is rather simple: If a network device or an operator wants to check whether a destination is reachable, it sends an echo-request to the corresponding destination IP address. The networking devices determine a path to the destination by routing the packet. Once the echo-request reaches the destination, it replies to the echo-request with an echo-reply. The reply is routed back to the originator of the request, and therefore connectivity/reachability has been checked.

For quantum networks the situation is more involved, as the network relies on entanglement. In particular, a quantum ping protocol shall ensure not only the classical reachability of a device, but also that entanglement between networking devices is still present and usable for future requests. In general we identify three different levels of checking for reachability:

  1. Classical ping: This corresponds to the classical ICMP echo protocol.

  2. State verification: This level has the purpose to check whether entanglement is still available or not, where the local apparatus is trusted. Routers can achieve this by using techniques like entanglement witnesses, checking the correlation operators of graph states, or quantum state tomography.

  3. Device-independent state verification: This level aims at checking for entanglement without trusting the local apparatus of each network device, and is therefore the strongest level of reachability.

We remark that in general a non-destructive state verification without additional entanglement as resource is not possible. We can only locally measure (and hence destroy) a subset of the quantum states. Under the assumption the initial ensemble consists of independent copies, one can then make predictions about the state and quality of the remaining ensemble. For graph states, methods for state verification have been put forward, see e.g. Mckague (2014); McKague (2016); Markham and Krause (2018); Pál et al. (2014). These approaches are based on stabilizer measurements and their expected values, or tests of generalized Bell inequalities. The latter allows for device independent state verification.

Vii Conclusion

In this work we have introduced a stack model for quantum networks, which is especially well suited for networks relying on multipartite entanglement. The stack comprises several different, goal-oriented layers. Each of these layers has a clear responsibility, and therefore enables for a straightforward implementation. By introducing such a layered model, we have broken down the complexity of large-scale quantum networks into smaller, manageable pieces, where each layer can be studied independently.

Furthermore we have introduced several protocols at different layers. In particular, we have found a way for ensuring reliability inside and between different networks. We also defined the term routing between quantum networks connecting via multipartite quantum states. To solve this routing problem we presented a protocol which achieves the goal by using Steiner tree constructions.

It remains to be considered which cost function is appropriate for the Steiner tree construction. Furthermore, it might be interesting to find alternative multipartite quantum states for connecting devices in a network, and routers in regions.


This work was supported by the Austrian Science Fund (FWF): P28000-N27 and P30937-N27. We thank J. Wallnöfer for helpful discussions and comments on the manuscript.

Appendix A An example of region routing

We illustrate the region routing protocol 2 for a concrete example, which is shown in Fig. 12. In this example, four clients of four different networks want to share a graph state. The region routing protocol shall establish a virtual network state among the respective routers. The first step in the Steiner tree construction for the first router is shown in Fig. 12.

Figure 12: Clients of the networks want to generate a graph state. The configuration of routers (boxes) in regions (dashed lines) is shown at the top of the figure. There are four regions of size two, three regions of size three (one includes , the other one includes and ) and one of size four. Observe that networks may be part of several regions, like e.g. , which connect at one router, see the discussion above. On the bottom we depict the first step of the protocol: The protocol starts with , and selects to be the root of the first Steiner tree. Then, in the first step of the Steiner tree computation, is chosen to be closest network to , and the path is added to the Steiner tree (red edge).

After the first Steiner tree construction finishes, the algorithm transforms the Steiner tree to a GHZ state as follows: If the degree of the root vertex selected by the protocol is larger than one, it combines all qubits into one (by e.g. preparing a local GHZ state of size , and performing at most Bell-measurements), thereby again obtaining a GHZ state of size . All nodes which are not target networks, employ Bell-measurements to their qubits which have been selected for the Steiner tree. If a target node is not a terminal node, then it locally creates a GHZ state of size , keeps one qubit, and performs at most Bell-measurement of all qubits which are part of the Steiner. Observe that it may be necessary that some routers measure qubits in the basis to shape the GHZ states in regions.

In the next step, the protocol 2 removes from the set of target networks and we perform the same procedure to the remaining network nodes in . The final state after the routing protocol is depicted in Fig. 13.

Figure 13: The Steiner tree constructions for all GHZ network states of our example.In a) the state , in b) the state and c) a Bell-pair is generated. After the protocol finishes, the router share a full GHZ network state.

Appendix B How things combine: A full example of how the quantum network stack works

In this section we provide a full example of how the quantum network stack presented in this paper works. For that purpose we illustrate the responsibilities of each layer and the corresponding operations of the network devices.

We start with the example scenario depicted in Fig. 14.

Figure 14: The figure depicts the example which we use to illustrate how the quantum network stack works. In this example, a client of the switch in the blue network requests to share a four qubit cluster state with a client located at a switch in the yellow, and two clients located at two different switches in the red network.

As explained in the figure, a client located in the blue network wants to share a four qubit cluster state with clients located in the yellow and red network. The routers connect in regions via GHZ network states in an irregular manner, defined by a network administrator. Observe that the blue network connects to the red and yellow network via some intermediate networks.

The first step in processing the request is now to run the routing protocol of layer 4. The result of routing is a virtual network state between the routers of the blue, yellow and red network, see Fig. 15.

Figure 15: The situation after routing completes: The routers of the blue, yellow and red networks share a virtual network state.

After routing completes, the routers fulfill their second responsibility: They merge the virtual network state with their respective intra-network states to transfer the entanglement of the virtual network state to the requesting network devices inside their networks. This is done by performing Bell-measurements between the qubits of the network layer and the qubits of the link layer, which is depicted in Fig. 16.

Figure 16: How the routers combine the virtual network state and the respective local, intra-network states (we illustrate the principle for the largest GHZ state of the virtual network state): The double arrows indicate between which qubits the routers perform a Bell-measurement. Observe that such merging operations might also induce the creation of some local GHZ states for expanding states at the network boundary. However, the procedure will always result in a GHZ state spanning all network devices involved in the request.

Once this operations complete, the requesting switches share one copy of the GHZ network state among each other. This operation is now carried out several times to generate multiple copies of the GHZ states, where the number of iterations depends on the graph state request. Observe that such a full GHZ network state enables the devices to generate any arbitrary graph state among the requesting network devices by simply invoking the linking protocol of layer 3 (see Pirker et al. (2018) or Sec. III.3). The situation is shown in Fig. 17.

Figure 17: Once the routers have combined the entanglement of the virtual network state among the routers with the network states of their associated networks, the switches of different networks share a GHZ network state. Observe that we have depicted in this figure a full GHZ network state for illustration. In real world applications, only a subset of these states might be necessary to complete the graph state request.

Finally, as outlined above, the switches invoke the state linking protocol of layer 3 (i.e. the graph state generating protocol of the layer they are operating on) to complete the graph state request for the requesting clients. The final graph state is shown in Fig. 18.

Figure 18: The switches have completed the graph state request by generating the four qubit cluster state via the state linking protocol of layer 3. For that purpose they consume the network state as established in Fig. 17. Finally, the network devices teleport this state to the requesting clients.

After the graph state request finishes, entanglement has to be replenished in most cases (dynamic phase). For that purpose, the switches and routers which were involved in the request instruct the layer 2 devices, i.e. the quantum repeaters, to re-establish the required entanglement in the networks and regions. In doing so, the quantum network devices use the physical channel configuration of the network at layer 1 to send quantum states to neighbouring devices. For example, in the blue network it may suffice to re-establish Bell-pairs between the router and the switch. In contrast, in the red network the networking devices instruct the repeaters of the network to generate the required resource mandatory to generate the consumed GHZ state among these devices. Finally, the routers instruct the repeaters residing in regions to generate the entanglement which is required to re-establish the network states connecting routers of a region.