Quantum Consensus: an Overview Arxiv:2101.04192V1 [Quant-Ph] 11 Jan 2021

Quantum Consensus: an overview

Marco Marcozzi1 and Leonardo Mostarda1

1Computer Science Division, University of Camerino, I-62032 Camerino (MC), Italy

Abstract We review the literature about reaching agreement in quantum networks, also called quantum consensus. After a brief introduction to the key feature of quantum computing, allowing the reader with no quantum theory background to have minimal tools to understand, we report a formal deﬁnition of quantum consensus and the protocols proposed. Proposals are classiﬁed according to the quantum feature used to achieve agreement.

Introduction scenario, the study of quantum distributed systems is fundamental for the correct The advent of Blockchain technology and, functioning of such a network [3]. To ensure more in general, of the Distributed Ledger agreement in a quantum network, thus, it is Technologies (DLTs), has attracted and vital the development of quantum consensus focused the attention on the study of Dis- protocols: from a formal deﬁnition, to actual tributed systems. It has been revised the proposals. Consensus problem, with new proposals being analysed and tested, because of the necessity of agreement on the publication Our paper is structured as follows: an of block, to ensure integrity, and to handle introduction to computational complexity faulty nodes in blockchains [1]. theory will show the importance of quantum arXiv:2101.04192v1 [quant-ph] 11 Jan 2021 Analogously, the increasing interest in computation, then a brief presentation about quantum computation is leading to a deeper quantum computing theory will give an understanding of quantum distributed sys- insight about quantum formalism to a reader tems, and eventually it will be implemented unfamiliar with quantum mechanics, ﬁnally a world wide quantum network, already an overview about quantum consensus is know as Quantum internet [2] - the quantum shown. counterpart of the World Wide Web. In this

1 Computation theory was born and it is breaking discover of the Shor’s algorithm, developing following the evolution of Inform- that allows to solve in polynomial time ation Technology: from the first computers, the factoring and the discrete logarithms through the transistors, arriving to the problem [9]. Shor’s algorithm sparked a today’s supercomputers and the enormous tremendous interest in quantum computers. amount of data produced every day [4]. The reason lies with the fact that Shor’s Even though the computational power of algorithm can theoretically break many of our devices nowadays is unbelievable high, the cryptosystems in use today [10]. if compared with the devices from a few years ago, classical computers seem not to Therefore, the discover of quantum al- be fit for the next generation of problems in gorithms solving some problems believed almost any field: chemistry, biology, medical to be computationally hard, triggered a technology, cryptography, optimisation, rephrase of the computational complexity finance, etc. [5] theory. Because of those and other applications, A problem is said to be computationally in last forty years, the possibility to build easy, hence it belongs to the computational a computer based on quantum mechanics class P, if it can be solved by a classical is focusing the attention and efforts of computer in polynomial time, respect to researchers, governments and companies. the number of bits needed to describe the The history of Quantum Computation problem. A problem is said to be computa- started in 1980, when Paul Benioff proposed tionally hard if the required resources (time) the first quantum mechanical model of for solving it rises super polynomially fast a computer [6]. The next year, Richard (often exponentially) with the input size. Feynman gave a talk at the First Conference Such problems belongs to the computational on the Physics of Computation, in which he complexity class NP (non-deterministic stated that quantum mechanical phenomena polynomial time). Regarding those problems cannot be efficiently simulated on a classical classes as sets, it is "strongly believed" that computer, proposing a basic model for a P⊂NP. "Strongly believed" means that quantum computer [7]. nowadays it is only know that P⊆NP, but it is strong the trust in the fact that NP Since then, a lot of discoveries have problems cannot be solved in polynomial been done. Among the most important time by a classical computer. ones, there are the description of the first Universal Quantum Turing machine, i.e. In this framework quantum computers play the definition of the first universal quantum their role. Shor’s algorithm (and other al- computer, that can simulate any other gorithms, e.g. Deutsch’s algorithm [11]) quantum Turing machine with at most a demonstrated that certain problems in the polynomial slowdown [8], and the ground- NP set could be solved efficiently (i.e. in

2 polynomial time) using a quantum computer. 2019, Google and NASA claimed to have It leads to a new complexity class, BQP achieved and proved the quantum suprem- (bounded-error quantum polynomial time), acy [13], [14]. The result was impressive, even that includes all the problems in P and some though there was scepticism, because some problems in NP (it is still not clear how researchers pointed out that the problem many). solved by this quantum annealer (not an uni- In Figure 1 the relationship between the dif- versal quantum Turing machine) could have ferent complexity classes is shown. been solved by a classical supercomputer in a comparable amount of time; moreover, since the machine is not a quantum computer, strictly speaking, they believe it is improper to address the achievement of quantum supremacy in this particular context [14].

Quantum Computing

Computers are physical objects, and computations are physical processes. (David Deutsch)

This sentence reveals a deep connection between information theory, computer science and physics. Even though information theory can be stated in a completely math- Figure 1: The relationship between classical ematical fashion, i.e. the physical support and quantum complexity classes. While it is used can be omitted, computation theory strongly assumed that BQP is larger than and algorithms need to rely on a physical P and encompasses some problems in NP, it theory. It means that, as classical computers remains unclear how the classes are exactly uses Maxwell’s electromagnetism in a Newto- related. Figure from [12]. nian framework, a quantum computer needs quantum physics and related phenomena In this context, scientists started talking to work. A brief review, based on [12], about the possible "quantum supremacy" for about quantum mechanics is presented in what concerns computation. It was conjec- the following. tured that indeed quantum computers would solve hard problems eﬃciently, but an ex- Quantum mechanics is an axiomatic perimental proof was needed. In October theory used as a mathematical framework

3 for the development of physical theories. It two vectors v, w, the expressions for the inner doesn’t hold any information about physical product hv|wi and the outer product |vihw|. laws, but it provides a conceptual connection Using bra-ket notation, therefore, a qubit can between the mathematical formalism and be defined as the physical world. As an axiomatic theory, then, quantum mechanics relies on some |Ψi = α |0i + β |1i , (1) axioms, or postulates. For the purposes where α, β are complex numbers satisfying of this review, those postulates will be not 2 2 the normalisation condition |α| + |β| = 1, presented rigorously. Therefore, naïvely, 1 0 |0i = |1i = postulates of quantum mechanics define: the and 0 , 1 are two vectors, in system, as a complex vector space (i.e. an principle arbitrary: in this case the computa- state of the system Hilbert space called ), tional basis states (connected with the logical evolves how a quantum mechanical state , "0" and "1") were used. and the measurement operations. In Figure 2 there is a graphical repres- For computational purposes, the minimum entation of a qubit, commonly called Bloch quantum state is represented by a vector in sphere. The normalisation condition guaran- a bi-dimensional complex space. Such a vec- tees that the vector has unitary magnitude qubit tor is called . It is important to point and, consequently, two angles are needed to out that there exist computational paradigms fully describe an arbitrary quantum state. d where a -dimensional Hilbert space is used: Therefore, a convenient way to express a in such spaces the unit of information is called quantum state is qudit. θ θ A qubit is built on the classical concept of |Ψi = cos |0i + eiφ sin |1i . (2) bit and it is the smallest information resource 2 2 for quantum computers. While a classical bit where θ, φ are angles on the Bloch sphere can be either 0 or 1, the qubit can be in a and |0i , |1i are vectors of the computational superposition of both at the same time. This basis. feature is intrinsic in the properties of the Hil- bert space: any combination of vectors in the No-cloning Theorem Hilbert space is a vector that belongs to the same Hilbert space. The no-cloning theorem [15] states that it A convenient way of expressing vectors in cannot exist an unitary operation able to cre- an Hilbert space is the Dirac notation, also ate an identical copy of an arbitrary state. known as bra-ket notation. In such notation It means that, if we have an unknown ar- a vector v, called ket, is represented as |vi, bitrary state |ψi1 ∈ H1 and a "blank" state while hv|, called bra, is its Hermitian adjoint. |ei2 ∈ H2, with H = H1 = H2, it is im- Using this formalism, it comes natural, given possible to find an unitary transformation U

4 but using as basis states the tensor product of the two qubits composing the system |00i , |01i , |10i , |11i. As a result, any arbitrary two-qubit state can be written as

|Ψi = α00 |00i + α01 |01i + α10 |10i + α11 |11i , (4) where αij are normalised complex amp- litudes and |abi = |abi1,2 = |ai1 ⊗ |bi2 is the tensor product of two different qubits |ai1 and |bi2 (for simplicity, in the tensor product indices can be omitted). In fact, due to superposition principle, when two, or more, qubits are interacting with each other, entanglement can occur. Figure 2: A qubit can be described (because Entanglement is an notably fascinating of the normalisation condition) by means of phenomenon, and a point of strength for only two angles θ, φ. It can be noticed that quantum computation. In a formal way, a North pole of the sphere represents the state state |Ψi is entangled if it cannot be ex- |0i and South pole the state |1i, while any pressed as a tensor product of its individual other state |Ψi is a superposition of the two vectors |Ψi= 6 |ai ⊗ |bi. states. Figure from [12]. Entanglement between qubits leads to a correlation in the physical quantities to be on H ⊗ H such that measured. When one qubit is measured in a certain basis, the measurement outcomes U |ψi |ei = |ψi |ψi 1 2 1 2 (3) measuring the other qubits are correlated. It is an important result because it negates Moreover, this correlation holds for any the possibility to copy an arbitrary quantum distance between the qubits. It has been state, differently from the classical case where the non-locality of entanglement that was information can be copied indefinitely. criticised by Einstein-Podolsky-Rosen in a 1935 paper (the so called EPR para- Entanglement dox) [16], but later it was experimentally proven [17], [18], [19] that quantum mechan- We already discussed that an arbitrary single ical states are indeed non-local. qubit state can be expressed as a superposition of the basis states. We can ap- It is convenient to introduce some spe- ply the same reasoning for a pair of qubits, cial entangled states, which are maximally

5 entangled (in the sense of entanglement en- To better understand the differences and, tropy). thus, define how quantum consensus may be Bell’s states stated, let us discuss first about what is classical consensus. ± |00i ± |11i Ψ = √ (5) 2 Classical Consensus ± |01i ± |10i Φ = √ (6) 2 In classical distributed computing, when dealing with concurrent processes, state ma- Greenberger-Horne-Zeilinger (GHZ) states chine replication, multi-agent systems and re- ⊗N ⊗N lated paradigms, the goal is to achieve an |0i + |1i |GHZi = √ , N > 2, (7) overall system reliability, despite the presence 2 of a number of faulty processes. and W-states Roughly, there are two main properties that a reliable distributed system has to satisfy: 1 |W i = √ (|100 ... 0i + |010 ... 0i + N (8) • Liveness: summarised as "something + ... + |000 ... 1i), good will eventually occur"; where N is the number of the qubits compos- • Safety: summarised as "something bad ing the system. will never happen". More formally, as stated in [20], for a con- Quantum Consensus sensus protocol it is required that, given a common input, the state related to each sub- n After this not exhaustive introduction to system xi(0) ∈ R is evolving (according to concepts of Quantum Computation, we are certain network policies and dynamics) to- approaching the main topic of our review: ward a non-trivial output state xi(t) at an quantum consensus. Because of the pe- arbitrary time t ∈ (0, ∞). The output state culiar features of Quantum Computation of each subsystem, except for a "small" set (e.g. entanglement, stochastic measurement of faulty processes, converges to a configura- outcomes, no-cloning theorem, etc.), it is tion in which subsystems have the same out- not easy or completely meaningful to define put: the consensus state xi(t) = xj(t) ∀i, j. quantum consensus based on the classical Moreover, if the value of the consensus state counterpart. In general, a definition of con- is the average of the initial states, the system sensus for quantum systems has to take into is said to reach an average consensus. account the fact that a quantum network, An other way of defining consensus is to however it may be built, is intrinsically dif- search for invariance with respect to subsys- ferent respect to a classical network. tems permutations. Given the set of all the

6 possible permutation operations, it is suffi- Quantum Consensus Algorithms cient to check that all the pairwise permutations Pπ are invariants of the system, i.e. Quantum protocols and algorithms to Pπx = x. achieve consensus over a quantum network have been proposed by researchers. A categorisation of such algorithms leads to the identification of four categories, based Defining Quantum Consensus on the quantum mechanical feature used to reach consensus: state invariance respect The definition of quantum consensus may be, to permutations, correlations due to en- in principle, based on its classical counter- tangled states, state evolution by means of part: even though, a classical probabilistic quantum measurements, and by means of consensus must be considered because of the Quantum Key Distribution (QKD) protocols. inherent stochastic nature of quantum measurement. Symmetric-state Consensus In [20], authors made a systematic study on consensus in quantum networks. They G. Shi et al. in three different pa- define four classes of consensus, namely σ- pers [21], [22], [23] present the convergence expectation consensus, reduced state con- of the state of a quantum network to a sensus, symmetric state consensus, and single consensus symmetric-state. They use an σ-measurement consensus, are basing their underlying Lindblad master equation [24] to definitions on symmetries and invariants of describe the state evolution of the quantum the system. As stated by the authors, those network with continuous-time swapping definitions may work also with classical ran- operators. Authors also prove that quantum dom variables or probability distributions of consensus of n qubits naturally defines a the state values, and it is worthy to point consensus process on an induced classical out that those definitions actually reflect graph with 22n nodes. Such a mapping allows the spirit of the classical consensus. They to study quantum consensus convergence also found hierarchies for their definitions of by means of already existing results for quantum consensus, and the implications de- consensus on classical networks. riving from them. Finally, they discuss how to detect consensus in quantum networks. L. Mazzarella et al., in [20], propose and They showed that studying symmetries in a analyse a quantum gossip-type algorithm quantum network is not trivial, because there that asymptotically converges to symmetric- is the possibility of having entanglement as state consensus states, while preserving the correlation between qubits: due to entangle- expectation of any permutation invariant ment, it becomes not obvious to check the global observable. effect of permutations on the full state.

7 R. Takeuchi and K. Tsumura study classical consensus: if the measurement of distributed feedback control of quantum net- a local observable quantity gives a certain works with local quantum observation and outcome, any subsequent measurement in feedback [25]. Authors prove that a quantum all the other subsystem will return the same consensus algorithm makes quantum states outcome. converge to a symmetric-state consensus from arbitrary initial states preserving Entanglement-based Consensus purity; authors also show that quantum consensus algorithms can generate a W-state A widely discussed and studied consensus entanglement. algorithm for classical networks is the one derived from a problem in distributed com- A work by S. Jafarizadeh [26] studies the puting know as The Byzantine Generals optimisation of the convergence rate of the Problem [28]. This problem deals with quantum consensus algorithm over quantum the presence of faulty participants in the network with N qudits. The model used consensus process, with the possibility that relies on the discrete-time evolution of the not only those participants are defective, but consensus algorithm. In this framework, they may also act deliberately against the author proves that the convergence rate of achievement of consensus. the algorithm depends on the value d of the Both those problems have been studied in qudits. a quantum framework, demonstrating that quantum properties can enhance the classical F. Ticozzi introduces two algorithms giv- results, and even solve problems otherwise ing both an improvement to the gossip-like believed unsolvable. consensus and a new dynamic for quantum consensus [27]. The author analyses the lim- M. Ben-Or and A. Hassidim presented its of the symmetric-state consensus states, an algorithm for a fast quantum Byzantine pointing out that this kind of algorithms agreement [29]. Authors prove that, using do not reach consensus strictly speaking, their algorithm, a network equipped with but they reach consensus on the statistical both quantum and classical channels can properties of the variables of interest. In reach Byzantine agreement in O(1) expected this scheme, no algorithm can attain ac- communication rounds against a strong full tual consensus on the output of each local information, dynamic adversary, tolerating measurement. The ﬁrst proposed algorithm up to the optimal t < n/3 faulty players in improves the existing gossip-type dynamics, the synchronous setting, and up to t < n/4 as in general it attains a purer output state faulty players for asynchronous systems. while still guaranteeing symmetric-state consensus. The second algorithm, instead, L. Helm proposed an algorithm solving guarantees a result closer to the idea of the same problem [30], with a particular

8 focus on addressing the FLP impossibility connected to the large amount of messages in of distributed consensus in an asynchronous a centralised solution, authors also develop a setting [31]. In particular, author claims that distributed Pairwise Qubit Projection (PQP) his algorithm can solve the consensus prob- algorithm, demonstrating that a quantum lem in a completely asynchronous setting, hybrid network almost surely converges to a without the need of classical communication. consensus status for each qubit. In [32], authors question if is it possible to reach consensus using the protocol by [30], QKD-based Consensus since that quantum algorithm cannot provide deterministic agreement and validity. X. Sun et al. proposed a quantum communication protocol to achieve Byzantine agree- Measurement-based Consensus ment among multiple parties without entanglement [34]. The protocol relies on the un- As prescribed by the postulates of quantum conditional security given by Quantum Key mechanics, the operation of measuring a Distribution (QKD). Sequences of correlated quantum state determines the collapse of the numbers shared between semi-honest distrib- quantum state in an eigenstate of the oper- utors of quantum keys is the solution on ator involved into the measurement. Loosely which authors base their proposal. Further speaking, this means that a quantum state, development of this protocol could consider a when it has been measured, evolves into a low-dimensional entanglement to replace the new state determined by the measurement need of semi-honest participants to the key itself. distribution in the protocol.

With this premise, consensus for a quantum hybrid network model is proposed Discussion and analysed in [33]. G. Shi et al. are considering, as a quantum hybrid network, In this paper we present an overview on a network consisting of a number of nodes quantum consensus, i.e. reaching agreement each holding a qubit, while communication in quantum networks. Our goal is to in- for reaching consensus is performed over troduce readers with no knowledge about classical channels. The proposed protocol quantum mechanics to the state of the art drives such a quantum hybrid network to a on quantum consensus. We gave some crude consensus in the sense that all qubits reach a tools to understand quantum formalism common state. This is achieved performing important in quantum information and com- measurements on qubits, thus outcomes puting. Then, we reported the literature on are exchanged between nodes by means the topic, dividing the discussion in two main of messages via classical communication subtopics: deﬁnition of quantum consensus channels. In order to address the issues and proposed protocols. In presenting the

9 proposed protocols, we classified them into [5] Sejuti Das. Top Applications Of four groups depending on the feature used Quantum Computing Everyone Should to reach consensus, namely: symmetric-state Know About, March 2020. consensus, entanglement-based consensus, measurement-based Consensus, and QKD- [6] Paul Benioff. The computer as a phys- based consensus. ical system: A microscopic quantum mechanical Hamiltonian model of com- It is important to notice that the literature puters as represented by Turing ma- Journal of Statistical Physics about quantum consensus is still scarce, chines. , especially if compared to the amount of 22:563–591, 1980. works on quantum distributed systems and [7] Richard P. Feynman. Simulating physics networks. with computers. International Journal of Theoretical Physics, 21:467–488, 1982. Further study may be related to the ap- plication of consensus in the (proposed) [8] David Deutsch. Quantum theory, the quantum distributed ledgers, prototypes of Church–Turing principle and the univer- the quantum blockchains. sal quantum computer. Proc. R. Soc. Lond. A., 400, 1985. References [9] Peter W. Shor. Polynomial-time algorithms for prime factorization and dis- [1] Dylan Yaga, Peter Mell, Nik Roby, crete logarithms on a quantum com- and Karen Scarfone. Blockchain puter. SIAM J. Comput., 26:1484–1509, technology overview. arXiv preprint 1994. arXiv:1906.11078, 2019. [10] Daniel J Bernstein. Introduction to post-quantum cryptography. In [2] H Jeff Kimble. The quantum internet. Post-quantum cryptography, pages 1–14. Nature, 453(7198):1023–1030, 2008. Springer, 2009.

[3] Vasil S Denchev and Gopal Panduran- [11] David Deutsch and Richard Jozsa. gan. Distributed quantum computing: Rapid solution of problems by quantum A new frontier in distributed systems or computation. Proceedings of the science ﬁction? ACM SIGACT News, Royal Society of London. Series A: 39(3):77–95, 2008. Mathematical and Physical Sciences, 439(1907):553–558, 1992. [4] Nicholas Metropolis. History of computing in the twentieth century. Elsevier, [12] Michael A. Nielsen and Isaac L. Chuang. 2014. Quantum computation and quantum in-

10 formation. Cambridge Univ. Press, 10th networks: Symmetry from gossip in- anniversary edition, 2010. teractions. IEEE Transactions on Automatic Control, 60(1):158–172, [13] F. Arute, K. Arya, R. Babbush, 2015. et al. Quantum supremacy using a pro- grammable superconducting processor. [21] Guodong Shi, Daoyi Dong, Ian R Nature, 574:505–510, 2019. Petersen, and Karl Henrik Johansson. Consensus of quantum networks with [14] Edwin Pednault, John Gunnels, continuous-time Markovian dynamics. Dmitri Maslov, and Jay Gam- In Proceeding of the 11th World Con- betta. On “Quantum Supremacy”. gress on Intelligent Control and Auto- IBM Research Blog, 2019. https: mation, pages 307–312. IEEE, 2014. //www.ibm.com/blogs/research/ 2019/10/on-quantum-supremacy/. [22] Guodong Shi, Shuangshuang Fu, and Ian R. Petersen. Reaching quantum con- [15] William K Wootters and Wojciech H sensus with directed links: Missing sym- Zurek. A single quantum cannot be metry and switching interactions, 2015. cloned. Nature, 299(5886):802–803, 1982. [23] Guodong Shi, Daoyi Dong, Ian R Petersen, and Karl Henrik Johansson. [16] A. Einstein, B. Podolsky, and N. Rosen. Reaching a quantum consensus: Master Can quantum-mechanical description of equations that generate symmetrization physical reality be considered complete? and synchronization. IEEE Transac- Phys. Rev., 47:777–780, May 1935. tions on Automatic Control, 61(2):374– [17] Bas Hensen et al. Loophole-free Bell 387, 2015. inequality violation using electron spins [24] Goran Lindblad. On the generators of separated by 1.3 kilometres. Nature, quantum dynamical semigroups. Com- 526:682–686, 2015. munications in Mathematical Physics, [18] Marissa Giustina et al. Signiﬁcant- 48(2):119–130, 1976. loophole-free test of Bell’s theorem with [25] Reiji Takeuchi and Koji Tsumura. Dis- entangled photons. Phys. Rev. Lett., tributed feedback control of quantum 115:250401, Dec 2015. networks. IFAC-PapersOnLine, [19] Lynden K. Shalm et al. Strong loophole- 49(22):309 – 314, 2016. free test of local realism. Phys. Rev. Lett., 115:250402, Dec 2015. [26] Saber Jafarizadeh. Optimizing the convergence rate of the quantum consensus: [20] L. Mazzarella, A. Sarlette, and A discrete-time model. Automatica, F. Ticozzi. Consensus for quantum 73:237 – 247, 2016.

11 [27] Francesco Ticozzi. Symmetrizing quantum dynamics beyond gossip-type algorithms. Automatica, 74:38 – 46, 2016. [28] Leslie Lamport, Robert Shostak, and Marshall Pease. The Byzantine Generals Problem. ACM Trans. Program. Lang. Syst., 4(3):382–401, July 1982. [29] Michael Ben-Or and Avinatan Hassidim. Fast quantum Byzantine agreement. In Proceedings of the thirty-seventh annual ACM symposium on Theory of computing, pages 481–485, 2005. [30] Louis K Helm. Quantum distributed consensus. In PODC, page 445, 2008. [31] Michael J Fischer, Nancy A Lynch, and Michael S Paterson. Impossibility of distributed consensus with one faulty process. Journal of the ACM (JACM), 32(2):374–382, 1985. [32] Wojciech Golab and Hao Tan. A closer look at quantum distributed consensus. In Proceedings of the 32nd ACM Sym- posium on Parallelism in Algorithms and Architectures, pages 539–541, 2020. [33] Guodong Shi, Bo Li, Zibo Miao, Peter M Dower, and Matthew R James. Reaching agreement in quantum hybrid networks. Scientiﬁc Reports, 7(1):1–9, 2017. [34] Xin Sun, Piotr Kulicki, and Mirek Sopek. Multi-party quantum Byzantine agreement without entanglement. arXiv preprint arXiv:2003.09120, 2020.