Using Quantum Computers for Quantum Simulation
aa r X i v : . [ qu a n t - ph ] N ov Submitted to
Entropy . Pages 1 - 43.
OPEN ACCESS entropy
ISSN 1099-4300
Review
Using Quantum Computers for Quantum Simulation
Katherine L. Brown ∗ , William J. Munro , and Vivien M. Kendon School of Physics and Astronomy, University of Leeds, Woodhouse Lane, Leeds LS2 9JT, UK National Institute of Informatics, 2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo, 101-8430, Japan NTT Basic Research Laboratories, NTT Corporation, 3-1 Morinosato-Wakamiya, Atsugi-shi,Kanagawa-ken 243-0198, Japan
Version November 17, 2010 submitted to
Entropy . Typeset by L A TEX using class file mdpi.cls
Abstract:
Numerical simulation of quantum systems is crucial to further our understand-ing of natural phenomena. Many systems of key interest and importance, in areas such assuperconducting materials and quantum chemistry, are thought to be described by modelswhich we cannot solve with sufficient accuracy, neither analytically nor numerically withclassical computers. Using a quantum computer to simulate such quantum systems has beenviewed as a key application of quantum computation from the very beginning of the field inthe 1980s. Moreover, useful results beyond the reach of classical computation are expectedto be accessible with fewer than a hundred qubits, making quantum simulation potentiallyone of the earliest practical applications of quantum computers. In this paper we survey thetheoretical and experimental development of quantum simulation using quantum computers,from the first ideas to the intense research efforts currently underway.
Keywords: quantum simulation; quantum computation; quantum information
Contents
I The Theory Behind Quantum Simulation 3 ∗ Corresponding author: [email protected] ersion November 17, 2010 submitted to
Entropy
II Experimental Implementation of Quantum Simulations 25
Entropy
11 Outlook 33
Part I
The Theory Behind Quantum Simulation
1. Introduction
The role of numerical simulation in science is to work out in detail what our mathematical modelsof physical systems predict. When the models become too difficult to solve by analytical techniques, ordetails are required for specific values of parameters, numerical computation can often fill the gap. Thisis only a practical option if the calculations required can be done efficiently with the resources avail-able. As most computational scientists know well, many calculations we would like to do require morecomputational power than we have. Running out of computational power is nearly ubiquitous whateveryou are working on, but for those working on quantum systems this happens for rather small systemsizes. Consequently, there are significant open problems in important areas, such as high temperaturesuperconductivity, where progress is slow because we cannot adequately test our models or use them tomake predictions.Simulating a fully general quantum system on a classical computer is possible only for very smallsystems, because of the exponential scaling of the Hilbert space with the size of the quantum system.To appreciate just how quickly this takes us beyond reasonable computational resources, consider theclassical memory required to store a fully general state | ψ n i of n qubits (two-state quantum systems).The Hilbert space for n qubits is spanned by n orthogonal states, labeled | j i with ≤ j < n . The n qubits can be in a superposition of all of them in different proportions, | ψ n i = n − X j =0 c j | j i (1)To store this description of the state in a classical computer, we need to store all of the complex numbers { c j } . Each requires two floating point numbers (real and imaginary parts). Using 32 bits (4 bytes) foreach floating point number, a quantum state of n = 27 qubits will require 1 Gbyte of memory – a newdesktop computer in 2010 probably has around 2 to 4 Gbyte of memory in total. Each additional qubitdoubles the memory, so 37 qubits would need a Terabyte of memory – a new desktop computer in 2010probably has a hard disk of this size. The time that would be required to perform any useful calculationon this size of data is actually what becomes the limiting factor. One of the largest simulations ofqubits on record [100] computed the evolution of 36 qubits in a quantum register using one Terabyte ofmemory, with multiple computers for the processing. Simulating more than 40 qubits in a fully generalersion November 17, 2010 submitted to Entropy superposition state is thus well beyond our current capabilities. Computational physicists can handlelarger systems if the model restricts the dynamics to only part of the full Hilbert space. Appropriatelydesigned methods then allow larger classical simulations to be performed [123]. However, any model isonly as good as its assumptions, and capping the size of the accessible part of the Hilbert space below orthogonal states for all system sizes is a severe restriction.The genius of Feynman in 1982 was to come up with an idea for how to circumvent the difficulties ofsimulating quantum systems classically [45]. The enormous Hilbert space of a general quantum state canbe encoded and stored efficiently on a quantum computer using the superpositions it has naturally. Thiswas the original inspiration for quantum computation, independently proposed also by Deutsch [37] afew years later. The low threshold for useful quantum simulations, of upwards of 36 or so qubits, meansit is widely expected to be one of the first practical applications of a quantum computer. Compared tothe millions of qubits needed for useful instances of other quantum algorithms, such as Shor’s algorithmfor factoring [104], this is a realistic goal for current experimental research to work towards. We willconsider the experimental challenges in the latter sections of this review, after we have laid out thetheoretical requirements.Although a quantum computer can efficiently store the quantum state under study, it is not a “dropin” replacement for a classical computer as far as the methods and results are concerned. A classicalsimulation of a quantum system gives us access to the full quantum state, i.e., all the n complex num-bers { c j } in equation (1). A quantum computer storing the same quantum state can in principle tellus no more than whether one of the { c j } is non-zero, if we directly measure the quantum state in thecomputational basis. As with all types of quantum algorithm, an extra step is required in the process-ing to concentrate the information we want into the register for the final measurement. Particularly forquantum simulation, amassing enough useful information also typically requires a significant number ofrepetitions of the simulation. Classical simulations of quantum systems are usually “strong simulations”[119,120] which provide the whole probability distribution, and we often need at least a significant partof this, e.g., for correlation functions, from a quantum simulation. If we ask only for sampling fromthe probability distribution, a “weak simulation”, then a wider class of quantum computations can besimulated efficiently classically, but may require repetition to provide useful results, just as the quantumcomputation would. Clearly, it is only worth using a quantum computer when neither strong nor weaksimulation can be performed efficiently classically, and these are the cases we are interested in for thisreview.As with all quantum algorithms, the three main steps, initialization, quantum processing, and data ex-traction (measurement) must all be performed efficiently to obtain a computation that is efficient overall.Efficient in this context will be taken to mean using resources that scale polynomially in the size of theproblem, although this isn’t always a reliable guide to what can be achieved in practice. For many quan-tum algorithms, the initial state of the computer is a simple and easy to prepare state, such as all qubitsset to zero. However, for a typical quantum simulation, the initial state we want is often an unknownstate that we are trying to find or characterise, such as the lowest energy state. The special techniquesrequired to deal with this are discussed in section 4. The second step is usually the time evolution of theHamiltonian. Classical simulations use a wide variety of methods, depending on the model being used,and the information being calculated. The same is true for quantum simulation, although the diversity isersion November 17, 2010 submitted to Entropy less developed, since we don’t have the possibility to actually use the proposed methods on real problemsyet and refine through practice. Significant inovation in classical simulation methods arose as a responseto practical problems encountered when theoretical methods were put to the test, and we can expect thesame to happen with quantum simulation. The main approach to time evolution using a universal quan-tum computer is described in section 2.1, in which the Lloyd method for evolving the Hamiltonian usingTrotterization is described. In section 5, further techniques are described, including the quantum versionof the pseudo-spectral method that converts between position and momentum space to evaluate differentterms in the Hamiltonian using the simplest representation for each, and quantum lattice gases, whichcan be used as a general differential equation solver in the same way that classical lattice gas and latticeBoltzmann methods are applied. It is also possible to take a direct approach, in which the Hamiltonianof the quantum simulator is controlled in such a way that it behaves like the one under study – an ideaalready well established in the Nuclear Magnetic Resonance (NMR) community. The relevant theory iscovered in section 2.3. The final step is data extraction. Of course, data extraction methods are dictatedby what we want to calculate, and this in turn affects the design of the whole algorithm, which is why itis most naturally discussed before initialization, in section 3.For classical simulation, we rarely use anything other than standard digital computers today. What-ever the problem, we map it onto the registers and standard gate operations available in a commercialcomputer (with the help of high level programming languages and compilers). The same approach toquantum simulation makes use of the quantum computer architectures proposed for universal quantumcomputation. The seminal work of Lloyd [77] gives the conditions under which quantum simulationscan be performed efficiently on a universal quantum computer. The subsequent development of quantumsimulation algorithms for general purpose quantum computers accounts for a major fraction of the theo-retical work in quantum simulation. However, special purpose computational modules are still used forclassical applications in many areas, such as fast real time control of experimental equipment, or videorendering on graphics cards to control displays, or even mundane tasks such as controlling a toaster, or ina digital alarm clock. A similar approach can also be used for quantum simulation. A quantum simulatoris a device which is designed to simulate a particular Hamiltonian, and it may not be capable of universalquantum computation. Nonetheless, a special purpose quantum simulator could still be fast and efficientfor the particular simulation it is built for. This would allow a useful device to be constructed before wehave the technology for universal quantum computers capable of the same computation. This is thus avery active area of current research. We describe a selection of these in the experimental sections 8 to10, which begins with its own overview in section 7.While we deal here strictly with quantum simulation of quantum systems, some of the methods de-scribed here, such as lattice gas automata, are applicable to a wider class of problems, which will bementioned as appropriate. A short review such as this must necessarily be brief and selective in thematerial covered from this broad and active field of research. In particular, the development of Hamil-tonian simulation applied to quantum algorithms begun by the seminal work of Aharonov and Ta-Shma[3] – which is worthy of a review in itself – is discussed only where there are implications for practicalapplications. Where choices had to be made, the focus has been on relevance to practical implementa-tion for solving real problems, and reference has been made to more detailed reviews of specific topics,where they already exist. The pace of development in this exciting field is such that it will in any caseersion November 17, 2010 submitted to
Entropy be important to refer to more recent publications to obtain a fully up to date picture of what has beenachieved.
2. Universal Quantum Simulation
The core processing task in quantum simulation will usually be the time evolution of a quantumsystem under a given Hamiltonian, | Ψ( t ) i = exp( i ˆ Ht ) | Ψ(0) i (2)Given the initial state | Ψ(0) i and the Hamiltonian ˆ H , which may itself be time dependent, calculatethe state of the system | Ψ( t ) i at time t . In many cases it is the properties of a system governed by theparticular Hamiltonian that are being sought, and pure quantum evolution is sufficient. For open quantumsystems where coupling to another system or environment plays a role, the appropriate master equationwill be used instead. In this section we will explore how accomplish the time evolution of a Hamiltonianefficiently, thereby explaining the basic theory underlying quantum simulation. Feynman’s seminal ideas [45] from 1982 were fleshed out by Lloyd in 1996, in his paper on universalquantum simulators [77]. While a quantum computer can clearly store the quantum state efficientlycompared with a classical computer, this is only half the problem. It is also crucial that the computationon this superposition state can be performed efficiently, the more so because classically we actually runout of computational power before we run out of memory to store the state. Lloyd notes that simplyby turning on and off the correct sequence of Hamiltonians, a system can be made to evolve accordingto any unitary operator. By decomposing the unitary operator into a sequence of standard quantumgates, Vartiainen et al [121] provide a method for doing this with a gate model quantum computer.However, an arbitrary unitary operator requires exponentially many parameters to specify it, so we don’tget an efficient algorithm overall. A unitary operator with an exponential number of parameters requiresexponential resources to simulate it in both the quantum and classical cases. Fortunately, (as Feynmanhad envisaged), any system that is consistent with special and general relativity evolves according tolocal interactions. All Hamiltonian evolutions ˆ H with only local interactions can be written in the form ˆ H = n X j =1 ˆ H j (3)where each of the n local Hamiltonians ˆ H j acts on a limited space containing at most ℓ of the total of N variables. By “local” we only require that ℓ remains fixed as N increases, we don’t require that the ℓ variables are actually spatially localised, allowing efficient simulation for many non-relativistic modelswith long-range interactions. The number of possible distinct terms ˆ H j in the decomposition of ˆ H isgiven by the binomial coefficient (cid:0) Nℓ (cid:1) < N ℓ /ℓ ! . Thus n < N ℓ /ℓ ! is polynomial in N . This is a generousupper bound in many practical cases: for Hamiltonians in which each system interacts with at most ℓ nearest neighbours, n ≃ N .ersion November 17, 2010 submitted to Entropy
In the same way that classical simulation of the time evolution of dynamical systems is often per-formed, the total simulation time t can be divided up into τ small discrete steps. Each step is approxi-mated using a Trotter-Suzuki [112,116] formula, exp { i ˆ Ht } = (cid:16) exp { i ˆ H t/τ } . . . exp { i ˆ H n t/τ } (cid:17) τ + X j ′ >j [ H j ′ , H j ] t / τ + higher order terms (4)The higher order term of order k is bounded by τ || ˆ Ht/τ || k sup /k ! , where || ˆ A || sup is the supremum, ormaximum expectation value, of the operator ˆ A over the states of interest. The total error is less than || τ { exp( i ˆ Ht/τ ) − − i ˆ Ht/τ }|| sup if just the first term in equation (4) is used to approximate exp( i ˆ Ht ) .By taking τ to be sufficiently large the error can be made as small as required. For a given error ǫ , fromthe second term in equation (4) we have ǫ ∝ t /τ . A first order Trotter-Suzuki simulation thus requires τ ∝ t /ǫ .Now we can check that the simulation scales efficiently in the number of operations required. Thesize of the most general Hamiltonian ˆ H j between ℓ variables depends on the dimensions of the individ-ual variables but will be bounded by a maximum size g . The Hamiltonians ˆ H and { ˆ H j } can be timedependent so long as g remains fixed. Simulating exp { i ˆ H j t/τ } requires g j operations where g j ≤ g isthe dimension of the variables involved in ˆ H j . In equation (4), each local operator ˆ H j is simulated τ times. Therefore, the total number of operations required for simulating exp { i ˆ Ht } is bounded by τ ng .Using τ ∝ t /ǫ , the number of operations Op Lloyd is given by Op Lloyd ∝ t ng /ǫ (5)The only dependence on the system size N is in n , and we already determined that n is polynomial in N , so the number of operations is indeed efficient by the criterion of polynomial scaling in the problemsize.The simulation method provided by Lloyd that we just described is straightforward but very general.Lloyd laid the groundwork for subsequent quantum simulation development, by providing conditions(local Hamiltonians) under which it will be possible in theory to carry out efficient quantum simulation,and describing an explicit method for doing this. After some further consideration of the way the errorsscale, the remainder of this section elaborates on exactly which Hamiltonians ˆ H q in a quantum simulatorcan efficiently simulate which other Hamiltonians ˆ H j in the system under study. Although Lloyd [77] explicitly notes that to keep the total error below ǫ , each operation must haveerror less than ǫ/ ( τ ng ) where n = poly ( N ) , he does not discuss the implications of this scaling as aninverse polynomial in N . For digital computation, we can improve the accuracy of our results by increas-ing the number of bits of precision we use. In turn, this increases the number of elementary (bitwise)operations required to process the data. To keep the errors below a chosen ǫ by the end of the computa-tion, we must have log (1 /ǫ ) accurate bits in our output register. The resources required to achieve thisin an efficient computation scale polynomially in log (1 /ǫ ) . In contrast, as already noted in equation(5), the resources required for quantum simulation are proportional to t ng /ǫ , so the dependence on ǫ is inverse, rather than log inverse.ersion November 17, 2010 submitted to Entropy
The consequences of this were first discussed by Brown et al [19], who point out that all the workon error correction for quantum computation assumes a logarithmic scaling of errors with the size ofthe computation, and they experimentally verify that the errors do indeed scale inversely for an NMRimplementation of quantum simulation. To correct these errors thus requires exponentially more oper-ations for quantum simulation than for a typical (binary encoded) quantum computation of similar sizeand precision. This is potentially a major issue, once quantum simulations reach large enough sizes tosolve useful problems. The time efficiency of the computation for any quantum simulation method willbe worsened due to the error correction overheads. This problem is mitigated somewhat because wemay not actually need such high precision for quantum simulation as we do for calculations involvingintegers, for example. However, Clark et al [33] conducted a resource analysis for a quantum simulationto find the ground state energy of the transverse Ising model performed on a circuit model quantum com-puter. They found that, even with modest precision, error correction requirements result in unfeasiblylong simulations for systems that would be possible to simulate if error correction weren’t necessary.One of the main reasons for this is the use of Trotterization, which entails a large number of steps τ eachcomposed of many operations with associated imperfections requiring error correction.Another consequence of the polynomial scaling of the errors, explored by Kendon et al [66], is thatanalogue (continuous variable) quantum computers may be equally suitable for quantum simulation,since they have this same error scaling for any computation they perform. This means they are usuallyconsidered only for small processing tasks as part of quantum communications networks, where the poorscaling is less of a problem. As Lloyd notes [77], the same methods as for discrete systems generalisedirectly onto continuous variable systems and Hamiltonians.On the other hand, this analysis doesn’t include potential savings that can be made when implement-ing the Lloyd method, such as by using parallel processing to compute simultaneously the terms inequation (3) that commute. The errors due to decoherence can also be exploited to simulate the effectsof noise on the system being studied, see section 5.4. Nonetheless, the unfavorable scaling of the errorcorrection requirements with system size in quantum simulation remains an under-appreciated issue forall implementation methods. Once Lloyd had shown that quantum simulation can be done efficiently overall, attention turned to theexplicit forms of the Hamiltonians, both the { ˆ H j } in the system to be simulated, and the { ˆ H q } availablein the quantum computer. Since universal quantum computation amounts to being able to implement anyunitary operation on the register of the quantum computer, this includes quantum simulation as a specialcase, i.e., the unitary operations derived from local Hamiltonians. Universal quantum computation is thussufficient for quantum simulation, but this leaves open the possibility that universal quantum simulationcould be performed equally efficiently with less powerful resources. There is also the important questionof how much the efficiency can be improved by exploiting the { ˆ H q } in the quantum computer directlyrather than through standard quantum gates.The natural idea that mapping directly between the { ˆ H j } and the { ˆ H q } should be the most efficientway to do quantum simulation resulted in a decade of research that has answered almost all the theoreticalquestions one can ask about exactly which Hamiltonians can simulate which other Hamiltonians, andersion November 17, 2010 submitted to Entropy how efficiently. The physically-motivated setting for much of this work is a quantum computer witha single, fixed interaction between the qubits, that can be turned on and off but not otherwise varied,along with arbitrary local control operations on each individual qubit. This is a reasonable abstractionof a typical quantum computer architecture: controlled interactions between qubits are usually hardand/or slow compared with rotating individual qubits. Since most non-trivial interaction Hamiltonianscan be used to do universal quantum computation, it follows they can generally simulate all others (ofthe same system size or smaller) as well. However, determining the optimal control sequences andresulting efficiency is computationally hard in the general case [130–132], which is not so practicalfor building actual universal quantum simulators. These results are thus important for the theoreticalunderstanding of the interconvertability of Hamiltonians, but for actual simulator design we will need tochoose Hamiltonians { ˆ H q } for which the control sequences can be obtained efficiently.Dodd et al [39], Bremner et al [16], and Nielsen et al [84] characterised non-trivial Hamiltonians asentangling Hamiltonians, in which every subsystem is coupled to every other subsystem either directlyor via intermediate subsystems. When the subsystems are qubits (two-state quantum systems), multi-qubit Hamiltonians involving an even number of qubits provide universal simulation, when combinedwith local unitary operations. Qubit Hamiltonians where the terms couple only odd numbers of qubitsare universal for the simulation of one fewer logical qubits (using a special encoding) [17]. When thesubsystems are qudits (quantum systems of dimension d ), any two-body qudit entangling Hamiltonianis universal, and efficiently so, when combined with local unitary operators [84]. This is a useful andilluminating approach because of the fundamental role played by entanglement in quantum informa-tion processing. Entanglement can only be generated by interaction (direct or indirect) between two (ormore) parties. The local unitaries and controls can thus only move the entanglement around, they cannotincrease it. These results show that generating enough entanglement can be completely separated fromthe task of shaping the exact form of the Hamiltonian. Further work on general Hamiltonian simulationhas been done by McKague et al [79] who have shown how to conduct a multipartite simulation usingjust a real Hilbert space. While not of practical importance, this is significant in relation to foundationalquestions. It follows from their work that Bell inequalities can be violated by quantum states restrictedto a real Hilbert space. Very recent work by Childs et al [30] fills in most of the remaining gaps in ourknowledge of the conditions under which two-qubit Hamiltonians are universal for approximating otherHamiltonians (equally applicable to both quantum simulation and computation). There are only threespecial types of two-qubit Hamiltonians that aren’t universal for simulating other two-qubit Hamiltoni-ans, and some of these are still universal for simulating Hamiltonians on more than two qubits. The other important question about using one Hamiltonian to simulate another is how efficiently itcan be done. The Lloyd method described in section 2.1 can be improved to bring the scaling with t down from quadratic, equation (5), to close to linear by using higher order terms from the Trotter-Suzukiexpansion [112]. This is close to optimal, because it is not possible to perform simulations in less thanlinear time, as Berry et al [10] prove. They provide a formula for the optimal number k opt of higher orderersion November 17, 2010 submitted to Entropy
10 of 43 terms to use, trading off extra operations per step τ for less steps due to the improved accuracy of eachstep, k opt = (cid:24) q log ( n || ˆ H || t/ǫ ) (cid:25) (6)where || ˆ H || is the spectral norm of ˆ H (equal to the magnitude of the largest eigenvalue for Hermitianmatrices). The corresponding optimal number of operations is bounded by Op Berry ≤ g n || ˆ H || t exp (cid:18) q ln 5 ln( n || ˆ H || t/ǫ ) (cid:19) (7)This is close to linear for large ( n || ˆ H || t ) . Recent work by Papageorgiou and Zhang [89] improves onBerry et al’s results, by explicitly incoporating the dependence on the norms of the largest and nextlargest of the ˆ H j in equation (3).Berry et al [10] also consider more general Hamiltonians, applicable more to quantum algorithmsthan quantum simulation. For a sparse Hamiltonian, i.e., with no more than a fixed number of nonzeroentries in each column of its matrix representation, and a black box function which provides one of theseentries when queried, they derive a bound on the number of calls to the black box function required tosimulate the Hamiltonian ˆ H . When || ˆ H || is bounded by a constant, the number of calls to obtain matrixelements scales as O ((log ∗ n ) t / k ) (8)where n is the number of qubits necessary to store a state from the Hilbert space on which ˆ H acts, and log ∗ n is the iterative log function, the number of times log has to be applied until the result is less thanor equal to one. This is a very slowly growing function, for practical values of n it will be less thanabout five. This scaling is thus almost optimal, since (as already noted) sub-linear time scaling is notpossible. These results apply to Hamiltonians where there is no tensor product structure, so generalisewhat simulations it is possible to perform efficiently. Child and Kothari [28,29] provide improved meth-ods for sparse Hamiltonians by decomposing them into sums where the graphs corresponding to thenon-zero entries are star graphs. They also prove a variety of cases where efficient simulation of non-sparse Hamiltonians is possible, using the method developed by Childs [27] to simulate Hamiltoniansusing quantum walks. These all involve conditions under which an efficient description of a non-sparseHamiltonian can be exploited to simulate it. While of key importance for the development of quantumalgorithms, these results don’t relate directly to simulating physical Hamiltonians.If we want to simulate bipartite (i.e., two-body) Hamiltonians { ˆ H (2) j } using only bipartite Hamiltoni-ans { ˆ H (2) q } , the control sequences can be efficiently determined [9,131,132]. D¨ur, Bremner and Briegel[41] provide detailed prescriptions for how to map higher-dimensional systems onto pairwise interactingqubits. They describe three techniques: using commutators between different ˆ H q to build up higher orderinteractions; graph state encodings; and teleportation-based methods. All methods incur a cost in termsof resources and sources of errors, which they also analyse in detail. The best choice of technique willdepend on the particular problem and type of quantum computer available.The complementary problem: given two-qubit Hamiltonians, how can higher dimensional qubitHamiltonians be approximated efficiently, was tackled by Bravyi et al [15]. They use perturbation theorygadgets to construct the higher order interactions, which can be viewed as a reverse process to standardersion November 17, 2010 submitted to Entropy
11 of 43 perturbation theory. The generic problem of ℓ -local Hamiltonians in an algorithmic setting is known to beNP-hard for finding the ground state energy, but Bravyi et al apply extra constraints to restrict the Hamil-tonians of both system and simulation to be physically realistic. Under these conditions, for many-bodyqubit Hamiltonians ˆ H = P j ˆ H ( ℓ ) j with a maximum of ℓ interactions per qubit, and where each qubitappears in only a constant number of the { ˆ H ( ℓ ) j } terms, Bravyi et al show that they can be simulatedusing two-body qubit Hamiltonians { ˆ H (2) q } with an absolute error given by nǫ || ˆ H ( ℓ ) j || sup ; where ǫ is theprecision, || ˆ H ( ℓ ) j || sup the largest norm of the local interactions and n is the number of qubits. For physicalHamiltonians, the ground state energy is proportional to n || ˆ H ( ℓ ) j || , allowing an efficient approximationof the ground state energy with arbitrarily small relative error ǫ .Two-qubit Hamiltonians { ˆ H (2) q } with local operations are a natural assumption for modelling a quan-tum computer, but so far we have only discussed the interaction Hamiltonian. Vidal and Cirac [124]consider the role of and requirements for the local operations in more detail, by adding ancillas-mediatedoperations to the available set of local operations. They compare this case with that of local operationsand classical communication (LOCC) only. For a two-body qubit Hamiltonian, the simulation can bedone with the same time efficiency, independent of whether ancillas are used, and this allows the problemof time optimality to be solved [9]. However, for other cases using ancillas gives some extra efficiency,and finding the time optimal sequence of operations is difficult. Further work on time optimality forthe two qubit case by Hammerer et al [56] and Haselgrove et al [57] proves that in most cases, a timeoptimal simulation requires an infinite number of infinitesimal time steps. Fortunately, they were alsoable to show that using finite time steps gives a simulation with very little extra time cost compared tothe optimal simulation. This is all good news for the practical feasibility of useful quantum simulation.The assumption of arbitrary efficient local operations and a fixed but switchable interaction is notexperimentally feasible in all proposed architectures. For example, NMR quantum computing has tocontend with the extra constraint that the interaction is always on. Turning it off when not required hasto be done by engineering time-reversed evolution using local operations. The NMR community hasthus developed practical solutions to many Hamiltonian simulation problems of converting one Hamil-tonian into another. In turn, much of this is based on pulse sequences originally developed in the 1980s.While liquid state NMR quantum computation is not scalable, it is an extremely useful test bed formost quantum computational tasks, including quantum simulation, and many of the results already men-tioned on universal Hamiltonian simulation owe their development to NMR theory [9,39,130]. Leung[74] gives explicit examples of how to do time reversed Hamiltonians for NMR quantum computation.Experimental aspects of NMR quantum simulation are covered in section 8.1.The assumption of arbitrary efficient local unitary control operations also may not be practical forrealistic experimental systems. This is a much bigger restriction than an always on interaction, and inthis case it may only be possible to simulate a restricted class of Hamiltonians. We cover some examplesin the relevant experimental sections.
3. Data extraction
So far, we have discussed in a fairly abstract way how to evolve a quantum state according to a givenHamiltonian. While the time evolution itself is illuminating in a classical simulation, where the fulldescription of the wavefunction is available at every time step, quantum simulation gives us only veryersion November 17, 2010 submitted to
Entropy
12 of 43 limited access to the process. We therefore have to design our simulation to provide the informationwe require efficiently. The first step is to manage our expectations: the whole wavefunction is an ex-ponential amount of information, but for an efficient simulation we can extract only polynomial-sizedresults. Nonetheless, this includes a wide range of properties of quantum systems that are both usefuland interesting, such as energy gaps [133]; eigenvalues and eigenvectors [2]; and correlation functions,expectation values and spectra of Hermitian operators [109]. These all use related methods, includingphase estimation or quantum Fourier transforms, to obtain the results. Brief details of each are givenbelow in sections 3.1 to 3.3.As will become clear, we may need to use the output of one simulation as the input to a furthersimulation, before we can obtain the results we want. The distinction between input and output is thussomewhat arbitrary, but since simulation algorithm design is driven by the desired end result, it makessense to discuss the most common outputs first.Of course, many other properties of the quantum simulation can be extracted using suitable mea-surements. Methods developed for experiments on quantum systems can be adapted for quantum sim-ulations, such as quantum process tomography [85] (though this has severe scaling problems beyond afew qubits), and the more efficient direct characterisation method of Mohseni and Lidar [81]. Recentadvances in developing polynomially efficient measurement processes, such as described by Emerson etal [44], are especially relevant. One well-studied case where a variety of other parameters are requiredfrom the simulation is quantum chaos, described in section 3.4.
One of the most important properties of an interacting quantum system is the energy gap between theground state and first excited state. To obtain this using quantum simulation, the system is prepared inan initial state that is a mixture of the ground and first excited state (see section 4.2). A time evolutionis then performed, which results in a phase difference between the two components that is directlyproportional to the energy gap. The standard phase estimation algorithm [35], which uses the quantumFourier transform, can then be used to extract this phase difference. The phase estimation algorithmrequires that the simulation (state preparation, evolution and measurement) is repeated a polynomialnumber of times to produce sufficient data to obtain the phase difference. An example, where this methodis described in detail for the BCS Hamiltonian, is given by Wu et al [133]. The phase difference can alsobe estimated by measuring the evolved state using any operator ˆ M such that h G | ˆ M | E i 6 = 0 . where | G i is the ground state and | E i the first excited state. Usually this will be satisfied for any operator thatdoes not commute with the Hamiltonian, giving useful experimental flexibility. A polynomial numberof measurements are made, for a range of different times. The outcomes can then be classically Fouriertransformed to obtain the spectrum, which will have peaks at both zero and the gap [19]. There will befurther peaks in the spectrum if the initial state was not prepared perfectly and had a proportion of higherstates mixed in. This is not a problem, provided the signal from the gap frequency can be distinguished,which in turn depends on the level of contamination with higher energy states. However, in the vicinityof a quantum phase transition, the gap will become exponentially small. It is then necessary to estimatethe gap for a range of values of the order parameter either side of the phase transition, to identify whenersion November 17, 2010 submitted to Entropy
13 of 43Figure 1.
A quantum circuit for measuring correlation functions, X is the Pauli σ x operator,U(t) is the time evolution of the system, and Hermitian operators ˆ A and ˆ B are operators(expressible as a sum of unitary operators) for which the correlation function is required.The inputs are a single qubit ancilla | a i prepared in the state ( | i + | i ) / √ and | ψ i , thestate of the quantum system for which the correlation function is required. h σ + i is theoutput obtained when the ancilla is measured in the σ + = σ x + σ y basis, which provides anestimate of the correlation function.it is shrinking below the precision of the simulation. This allows the location of the phase transition tobe determined, up to the simulation precision. Generalising from both the Lloyd method for the time evolution of Hamiltonians and the phase esti-mation method for finding energy gaps, Abrams and Lloyd [2] provided an algorithm for finding (someof) the eigenvalues and eigenvectors of any Hamiltonian ˆ H for which ˆ U = exp( i ˆ Ht/ ~ ) can be efficientlysimulated. Since ˆ U and ˆ H share the same eigenvalues and eigenvectors, we can equally well use ˆ U tofind them. Although we can only efficiently obtain a polynomial fraction of them, we are generally onlyinterested in a few, for example the lowest lying energy states.The Abrams-Lloyd scheme requires an approximate eigenvector | V a i , which must have an overlap |h V a | V i| with the actual eigenvector | V i that is not exponentially small. For low energy states, anapproximate adiabatic evolution could be used to prepare a suitable | V a i , see section 4.2. The algorithmworks by using an index register of m qubits initialised into a superposition of all numbers 0 to m − .The unitary ˆ U is then conditionally applied to the register containing | V a i a total of k times, where k is the number in the index register. The components of | V a i in the eigenbasis of ˆ U now each have adifferent phase and are entangled to a different index component. An inverse quantum Fourier transformtransfers the phases into the index register which is then measured. The outcome of the measurementyields one of the eigenvalues, while the other register now contains the corresponding eigenvector | V i .Although directly measuring | V i won’t yield much useful information, it can be used as the input toanother quantum simulation process to analyse its properties. Somma et al [109] provide detailed methods for extracting correlation functions, expectation valuesof Hermitian operators, and the spectrum of a Hermitian operator. A similar method is employed for allof these, we describe it for correlation functions. A circuit diagram is shown in figure 1. This circuit canersion November 17, 2010 submitted to
Entropy
14 of 43 compute correlation functions of the form C AB ( t ) = h ˆ U † ( t ) ˆ A ˆ U ( t ) ˆ B i (9)where ˆ U ( t ) is the time evolution of the system, and ˆ A and ˆ B are expressible as a sum of unitary operators.The single qubit ancilla | a i , initially in the state ( | i + | i ) / √ , is used to control the conditionalapplication of ˆ B and ˆ A † , between which the time evolution ˆ U ( t ) is performed. Measuring | a i thenprovides an estimate of the correlation function to one bit of accuracy. Repeating the computation willbuild up a more accurate estimate by combining all the outcomes. By replacing ˆ U ( t ) with the spacetranslation operator, spatial correlations instead of time correlations can be obtained. The attractions of quantum simulation caught the imagination of researchers in quantum chaos rela-tively early in the development of quantum computing. Even systems with only a few degrees of freedomand relatively simple Hamiltonians can exhibit chaotic behaviour [51]. However, classical simulationmethods are of limited use for studying quantum chaos, due to the exponentially growing Hilbert space.One of the first quantum chaotic systems for which an efficient quantum simulation scheme was providedis the quantum baker’s transformation. Schack [101] demonstrates that it is possible to approximate thismap as a sequence of simple quantum gates using discrete Fourier transforms. Brun and Schack [21]then showed that the quantum baker’s map is equivalent to a shift map and numerially simulated how itwould behave on a three qubit NMR quantum computer.While the time evolution methods for chaotic dynamics are straightforward, the important issue is howto extract useful information from the simulation. Using the kicked Harper model, L´evi and Georgeot[75] extended Schack’s Fourier transform method to obtain a range of characteristics of the behaviourin different regimes, with a polynomial speed up. Georgeot [49] discusses further methods to extractinformation but notes that most give only a polynomial increase in efficiency over classical algorithms.Since classical simulations of quantum chaos are generally exponentially costly, it is disappointing notto gain exponentially in efficiency in general with a quantum simulation. However, there are someuseful exceptions: methods for deciding whether a system is chaotic or regular using only one bit ofinformation have been developed by Poulin et al [97], and also for measuring the fidelity decay in anefficient manner [98]. A few other parameters, such as diffusion constants, may also turn out to beextractable with exponential improvement over classical simulation. A review of quantum simulationsapplied to quantum chaos is provided by Georgeot [50].
4. Initialization
As we saw in the previous section, a crucial step in extracting useful results from a quantum sim-ulation is starting from the right initial state. These will often be complex or unknown states, such asground states and Gibbs thermal states. Preparing the initial state is thus as important as the time evolu-tion, and significant research has gone into providing efficient methods. An arbitrary initial state takesexponentially many parameters to specify, see equation (1), and hence exponential time to prepare usingits description. We can thus only use states which have more efficient preparation procedures. Althoughersion November 17, 2010 submitted to
Entropy
15 of 43 preparing an unknown state sounds like it should be even harder than preparing a specific arbitrary state,when a simple property defining it is specified, there can be efficient methods to do this.
Where an explicit description is given for the initial state we require, it can be prepared using anymethod for preparing states for a quantum register. Soklakov and Schack [106,107] provide a methodusing Grover’s search algorithm, that is efficient provided the description of the state is suitably efficient.Plesch and Brukner [92] optimise general state preparation techniques to reduce the prefactor in therequired number of CNOT gates to close to the optimal value of one. Direct state preparation is thusfeasible for any efficiently and completely specified pure initial state. Poulin and Wocjan [95] analysethe efficiency of finding ground states with a quantum computer. This is known to be a QMA-completeproblem for k -local Hamiltonians (which have the form of equation (3) where the H j involve k ofthe variables, for k ≥ ). They provide a method based on Grover’s search, with some sophisticatederror reduction techniques, that gives a quadratic speed up over the best classical methods for findingeigenvalues of matrices. Their method is really a proof of the complexity of the problem in general ratherthan a practical method for particular cases of interest, which may not be as hard as the general case theytreat. Adiabatic quantum computing encodes the problem into the ground state of a quantum Hamiltonian.The computation takes place by evolving the Hamiltonian from one with an easy to prepare ground state ˆ H to the one with the desired solution ˆ H as the ground state, ˆ H ad = (1 − s ( t )) ˆ H + s ( t ) ˆ H (10)where the monotonically increasing function s ( t ) controls the rate of change, s (0) = 0 . This has to bedone slowly enough, to keep the system in the ground state throughout. Provided the gap between theground state and first excited state does not become exponentially small, “slowly enough” will requireonly polynomial time. Extensive discussion of quantum adiabatic state preparation from an algorithmicperspective, including other useful states that can be produced by this method, is given by Aharonov andTa Shma [3].The application to preparing ground states for quantum simulation was first suggested by Ortiz et al[88]. The potential issue is that finding ground states is in general a QMA-complete problem, whichimplies it may not be possible to do this efficiently for all cases of interest, that the gap will becomeexponentially small at some point in the evolution. In particular, we know the gap will become exponen-tially small if the evolution passes through a quantum phase transition. Since the study of quantum phasetransitions is one aspect of quantum many-body systems of interest for quantum simulation, this is notan academic problem, rather, it is likely to occur in practice. Being of crucial importance for adiabaticquantum computation, the question of how the time evolution scales near a phase transition has beenextensively studied. Recent work by Dziarmaga and Rams [42] on inhomogeneous quantum phase tran-sitions explains how in many cases of practical interest, disruption to the adiabatic evolution across theersion November 17, 2010 submitted to Entropy
16 of 43 phase transition can be avoided. An inhomogeneous phase transition is where the order parameter variesacross the system. Experimentally, this is very likely to happen to some extent, due to the difficulty ofcontrolling the driving mechanism perfectly, the strength of a magnetic field for example. Consequently,the phase change will also happen at slightly different times for different parts of the system, and therewill be boundaries between the different regions. Instead of being a global change, the phase transitionsweeps through the system, and the speed with which the boundary between the phases moves can beestimated. Provided this is slower than the timescale on which local transitions take place, this allowsthe region in the new phase to influence the transition of the nearby regions. The end result is that it ispossible to traverse the phase transition in polynomial time without ending up in an excited state, for afinite-sized system.Moreover, we don’t generally need to prepare a pure ground state for quantum simulation of suchsystems. The quantity we usually wish to estimate for a system with an unknown ground state is theenergy gap between the ground and first excited states. As described in section 3.1, this can be done byusing phase estimation applied to a coherent superposition of the ground state and first excited state. Sotraversing the adiabatic evolution only approximately, to allow a small probability of exciting the systemis in fact a useful state preparation method. And if we want to obtain the lowest eigenvalues and studythe corresponding eigenvectors of a Hamiltonian, again we only need a state with a significant proportionof the ground state as one component, see section 3.2.Oh [87] describes a refinement of the Abrams-Lloyd method for finding eigenvalues and eigenvectorsdescribed in section 3.2, in which the state preparation using the adiabatic method is run in parallelwith the phase estimation algorithm for estimating the ground state energy. This allows the groundstate energy to be extracted as a function of the coupling strength that is increased as the adiabaticevolution proceeds. Oh adds an extra constant energy term to the Hamiltonian, to tune the running time,and uses the Hellman-Feynman theorem to obtain the expectation value of the ground state observable.Boixo et al [14] prove that this and related methods using continuous measurement, as provided by thephase estimation algorithm run in parallel, improve the adiabatic state preparation. The running timeis inversely proportional to the spectral gap, so will only be efficient when the gap remains sufficientlylarge throughout the evolution.
Temperature dependent properties of matter are of key importance. To study these, efficiently prepar-ing thermal states for quantum simulation is crucial. The most obvious method to use is to actuallyequilibriate the quantum state to the required temperature, using a heat bath. Terhal and DiVincenzo[114] describe how this can be done with only a relatively small bath system, by periodically reinitial-izing the bath to the required temperature. The core of this algorithm begins by initializing the systemin the “all zero” state, | . . . ih . . . | and the bath in an equilibrium state of the required tempera-ture. The system and bath are then evolved for time t after which the bath is discarded and re-preparedin its equilibrium state. This last step is repeated a number of times, creating to a good approximationthe desired thermal initial state for subsequent simulation. Terhal and DiVincenzo don’t give explicitbounds on the running time of their method, though they do discuss reasons why they don’t expect itto be efficient in the general case. Recent results from Poulin and Wocjan [96] prove the upper boundersion November 17, 2010 submitted to Entropy
17 of 43 on the running time for thermalisation is D a , where a ≤ / is related to the Helmholtz free energyof the system, and D is the Hilbert space dimension. This thus confirms that Terhal and DiVincenzo’smethod may not be efficient in general. Poulin and Wocjan also provide a method for approximating thepartition function of a system with a running time proportional to the thermalisation time. The partitionfunction is useful because all other thermodynamic quantities of interest can be derived from it. So incases where their method can be performed efficiently, it may be prefered over the newly developedquantum Metropolis algorithm described next.The quantum Metropolis algorithm of Temme et al [113] is a method for efficiently sampling fromany Gibbs distribution. It is the quantum analogue of the classical Metropolis method. The process startsfrom a random energy eigenstate | Ψ i i of energy E i . This can be prepared efficiently by evolving fromany initial state with the Hamiltonian ˆ H , then using phase estimation to measure the energy and therebyproject into an eigenstate. The next step is to generate a new “nearby” energy eigenstate | Ψ j i of energy E j . This can be achieved via a local random unitary transformation such that | Ψ i i −→ P j c ij | Ψ j i with E j ∼ E i . Phase estimation is then used again to project into the state | Ψ j i and gives us E j . Wenow need to accept the new configuration with probability p ij = min[1 , exp( − β ( E i − E j ))] , where β is inverse temperature. Accepting is no problem, the state of the quantum registers are in the newenergey eigenstate | Ψ j i as required. The key development in this method is how to reject, which requiresreturning to the previous state | Ψ i i . By making a very limited measurement that determines only one bitof information (accept/reject), the coherent part of the phase estimation step can be reversed with highprobability; repeated application of the reversal steps can increase the probability as close to unity asrequired. Intermediate measurements in the process indicate when the reversal has succeeded, and theiteration can be terminated. The process is then repeated to obtain the next random energy state in thesequence. This efficiently samples from the thermal distribution for preparing the initial state, and canbe used for any type of quantum system, including fermions and bosons. Temme et al also prove thattheir algorithm correctly samples from degenerate subspaces efficiently.
5. Hamiltonian evolution
The Lloyd method of evolving the quantum state in time according to a given Hamiltonian, describedin section 2.1, is a simple form of numerical integration. There are a variety of other methods fortime evolution of the dynamics in classical simulation, some of which have been adapted for quantumsimulation. Like their classical counterparts, they provide significant advantages for particular types ofproblem. We describe two of these methods that are especially promising for quantum simulation: aquantum version of the pseudo-spectral method using quantum Fourier transforms, and quantum latticegas automata. Quantum chemistry has also developed a set of specialised simulation methods for whichwe describe some promising quantum counterparts in section 5.3. We would also like to be able tosimulate systems subject to noise or disturbance from an environment, open quantum systems. Somemethods for efficiently treating non-unitary evolution are described in section 5.4.ersion November 17, 2010 submitted to
Entropy
18 of 43
Fast Fourier transforms are employed extensively in classical computational methods, despite in-curring a significant computational cost. Their use can simplify the calculation in a wide diversity ofapplications. When employed for dynamical evolution, the pseudo-spectral method converts betweenreal space and Fourier space (position and momentum) representations. This allows terms to be evalu-ated in the most convenient representation, providing improvements in both the speed and accuracy ofthe simulation.The same motivations and advantages apply to quantum simulation. A quantum Fourier transformcan be implemented efficiently on a quantum computer for any quantum state [20,63]. Particles movingin external potentials often have Hamiltonians with terms that are diagonal in the position basis plusterms that are diagonal in the momentum basis. Evaluating these terms in their diagonal bases providesa major simplification to the computation. Wiesner [128] and Zalka [134,135] gave the first detaileddescriptions of this approach for particles moving in one spatial dimension, and showed that it can easilybe generalized to a many particle Schr¨odinger equation in three dimensions. To illustrate this, considerthe one-dimensional Schr¨odinger equation (with ~ = 1 ), i ∂∂t Ψ( x, t ) = (cid:18) − m ∇ + V ( x ) (cid:19) Ψ( x, t ) (11)for a particle in a potential V ( x ) . As would be done for a classical simulation this is first discretized sothe position is approximated on a line of N positions (with periodic boundary conditions) and spacing ∆ x . We can then write the wavefunction as | Ψ( n, t ) i = X n a n ( t ) | n i (12)where {| n i} , ≤ n < N are position basis states, and a n ( t ) is the amplitude to be at position n at time t . For small time steps ∆ t , the Green’s function to evolve from x to x in time ∆ t becomes G ( x , x , ∆ t ) = κ exp (cid:26) i m x − x ) ∆ t + iV ( x )∆ t (cid:27) (13)where κ is determined by the normalization. The transformation in terms of basis states is the inverse ofthis, G ′ ( n, n ′ , ∆ t ) | n i = 1 √ N X s n ′ exp (cid:26) − i m n − n ′ ) ∆ x ∆ t − iV ( n ∆ x )∆ t (cid:27) | n ′ i (14)Expanding the square, this becomes G ′ ( n, n ′ , ∆ t ) | n i = 1 √ N exp (cid:26) − i m n ∆ x ∆ t − iV ( n ∆ x )∆ t (cid:27) × X n ′ exp (cid:26) − im nn ′ ∆ x ∆ t (cid:27) (cid:18) exp (cid:26) − i m n ′ ∆ x ∆ t (cid:27)(cid:19) | n ′ i (15)The form of equation 15 is now two diagonal matrices with a Fourier transform between them, showinghow the pseudo-spectral method arises naturally from standard solution methods. Benenti and Strini [8]provide a pedagogical description of this method applied to a single particle, with quantitative analysisersion November 17, 2010 submitted to Entropy
19 of 43 of the number of elementary operations required for small simulations. They estimate that, for presentday capabilities of six to ten qubits, the number of operations required for a useful simulation is in thetens of thousands, which is many more than can currently be performed coherently. Nonetheless, theefficiency savings over the Lloyd method will still make this the preferred option whenever the terms inthe Hamiltonian are diagonal in convenient bases related by a Fourier transform.
Lattice gas automata and lattice-Boltzmann methods are widely used in classical simulation becausethey evolve using only local interactions, so can be adapted for efficient parallel processing. Despitesounding like abstract models of physical systems, these methods are best understood as sophisticatedtechniques to solve differential equations: the “gas” particles have nothing directly to do with the parti-cles in the system they are simulating. Instead, the lattice gas dynamics are shown to correspond to thedifferential equation being studied in the continuum limit of the lattice. Different equations are obtainedfrom different local lattice dynamics and lattice types. Typically, a face-centred cubic or body-centredcubic lattice is required, to ensure mixing of the particle momentum in different directions [47]. Succiand Benzi [111] developed a lattice Boltzmann method for classical simulation of quantum systems, andMeyer [80] applied lattice gas automata to many-particle Dirac systems. Boghosian and Taylor [13]built on this work to develop a fully quantum version of lattice gas automata, and showed that this can beefficiently implemented on a qubit-based quantum computer, for simulations of many interacting quan-tum particles in external potentials. This method can also be applied to the many-body Dirac equation(relativistic fermions) and gauge field theories, by suitably modifying the lattice gas dynamics, both arebriefly discussed by Boghosian and Taylor.To illustrate the concept, we describe a simple quantum lattice gas in one dimension. This can beencoded into two qubits per lattice site, one for the plus direction and the other for the minus direction.The states of the qubits represent | i for a particle present, and | i for no particle, with any superpositionbetween these allowed. Each time step consists of two operations, a “collision” operator that interacts thequbits at each lattice site, and a “propagation” operator that swaps the qubit states between neighboringlattice sites, according to the direction they represent. This is like a coined quantum walk dynamics,which is in fact a special case of lattice gas automata, and was shown to correspond to the Dirac equationin the continuum limit by Meyer [80]. Following Boghosian and Taylor [12], we take the time stepoperator S.C combining both collision C and propagation S to be S.C q ( x + 1 , t + 1) q ( x − , t + 1) ! = 12 − i − − i − − i i ! q ( x, t ) q ( x, t ) ! (16)where q and q are the states of the two qubits. Taking the continuum limit where the lattice spacingscales as ǫ while the time scales as ǫ gives ∂∂t q ( x, t ) = i ∂ ∂x q ( x, t ) (17)and a similar equation interchanging q and q . Hence for the sum, ∂∂t { q ( x, t ) + q ( x, t ) } = i ∂ ∂x { q ( x, t ) + q ( x, t ) } (18)ersion November 17, 2010 submitted to Entropy
20 of 43
The total amplitude ψ ( x, t ) = q ( x, t ) + q ( x, t ) thus satisfies a Schr¨odinger equation. It is a straightfor-ward generalisation to extend to higher dimensions and more particles, and to add interactions betweenparticles and external potentials, as explained in detail by Boghosian and Taylor [12]. Based on theutility of lattice gas automata methods for classical simulation, we can expect these corresponding quan-tum versions to prove highly practical and useful when sufficiently large quantum computers becomeavailable. Study of the dynamics and properties of molecular reactions is of basic interest in chemistry andrelated areas. Quantum effects are important at the level of molecular reactions, but exact calculationsbased on a full Schr¨odinger equation for all the electrons involved are beyond the capabilities of classi-cal computation, except for the smallest molecules. A hierarchy of approximation methods have beendeveloped, but more accurate calculations would be very useful. Aspuru-Guzik et al [5] study the ap-plication of quantum simulation to calculation of the energies of small molecules, demonstrating that aquantum computer can obtain the energies to a degree of precision greater than that required by chemistsfor understanding reaction dynamics, and better than standard classical methods. To do this, they adaptthe method of Abrams and Lloyd [2] for finding the eigenvalues of a Hamiltonian described in section3.2. The mapping of the description of the molecule to qubits is discussed in detail, to obtain an efficientrepresentation. In a direct mapping, the qubits are used to store the occupation numbers for the atomicorbitals: the Fock space of the molecule is mapped directly to the Hilbert space of the qubits. Thiscan be compacted by restricting to the subspace of occupied orbitals, i.e., fixed particle number, and afurther reduction in the number of qubits required is obtained by fixing the spin states of the electronsas well. By doing classical simulations of the quantum simulation for H O and LiH, they show in de-tail that these methods are feasible. For the simulations, they use the Hartree-Fock approximation forthe initial ground state. In some situations, however, this state has a vanishing overlap with the actualground state. This means it may not be suitable in the dissociation limit or in the limit of large systems.A more accurate approximation of the required ground state can be prepared using adiabatic evolution,see section 4.2. Aspuru-Guzik et al confirm numerically that this works efficiently for molecular hydro-gen. Data from experiments or classical simulations can be used to provide a good estimate of the gapduring the adiabatic evolution, and hence optimise the rate of transformation between the initial and finalHamiltonians.The Hartree-Fock wavefunctions used by Aspuru-Guzik et al are not suitable for excited states. Wanget al [126] propose using an initial state that is based on a multi-configurational self consistent field(MCSCF). These initial states are also suitable for strong interactions, since they avoid convergenceto unphysical states when the energy gap is small. In general, using MCSCF wavefunctions allowsan evolution that is faster and safer than using Hartree-Fock wavefunctions, so represents a significantimprovement.To calculate the properties of chemical reactions classically, the Born-Oppenheimer approximationis used for the electron dynamics. The same can be done for quantum simulations; however, Kassalet al [64] observe that, for all systems of more than four atoms, performing the exact computation ona quantum computer should be more efficient. They provide a detailed method for exact simulation ofersion November 17, 2010 submitted to
Entropy
21 of 43 atomic and molecular electronic wavefunctions, based on discretizing the position in space, and evolvingthe wavefunction using the QFT-based time evolution technique presented by Wiesner [128] and Zalka[134] described in section 5.1. Kassal et al discuss three approaches to simulating the interaction po-tentials and provide the initialisation procedures needed for each, along with techniques for determiningreaction probabilities, rate constants and state-to-state transition probabilities. These promising resultssuggest that quantum chemistry will feature prominently in future applications of quantum simulation.
Most real physical systems are subject to noise from their environment, so it is important to be ableto include this in quantum simulations. For many types of environmental decoherence, this can be doneas a straightforward extension to Lloyd’s basic simulation method [77] (described in section 2.1). Lloyddiscusses how to incorporate the most common types of environmental decoherence into the simulation.For uncorrelated noise, the appropriate superoperators can be used in place of the unitary operatorsin the time evolution, because these will also be local. Even for the worst case of correlated noise, theenvironment can be modeled by doubling the number of qubits and employing local Hamiltonians for theevolution of the environment and its coupling, as well as for the system. Techniques for the simulation ofopen quantum systems for a single qubit have been further refined and developed by Bacon et al [6], whoprovide a universal set of processes to simulate general Markovian dynamics on a single qubit. However,it is not known whether these results can be extended to include all Markovian dynamics in systems ofmore than two qubits, since it is no longer possible to write the dynamics in the same form as for the oneand two qubit cases.Better still, from the point of view of efficiency, is if the effects of noise can be included simply byusing the inevitable decoherence on the quantum computer itself. This will work provided the type ofdecoherence is sufficiently similar in both statistics and strength. Even where the aim is to simulateperfect unitary dynamics, small levels of imperfection due to noisy gates in the simulation may still betolerable, though the unfavorable scaling of precision with system size discussed in section 2.2 will limitthis to short simulations. Nonetheless, in contrast to the error correction necessary for digital quantumcomputations where precise numerical answers are required, a somewhat imperfect quantum simulationmay be adequate to provide us with a near perfect simulation of an open quantum system.
6. Fermions and bosons
Simulation of many-body systems of interacting fermions are among the most difficult to handle withclassical methods, because the change of sign when two identical fermions are exchanged prevents theconvergence of classical statistical methods, such as Monte Carlo sampling. This is known as the “signproblem”, and has limited effective simulation of fermionic many-body systems to small sizes that canbe treated without these approximations. Very recent work from Verstraete and Cirac [122] has openedup variational methods for fermionic systems, including relativistic field theories [55]. Nonetheless, thecomputational cost of accurate classical simulations is still high, and we have from Ortiz et al [88] ageneral proof that conducting a simulation of a fermionic system on a quantum computer can be doneefficiently and does not suffer from the sign problem. They also confirm that errors within the quantumersion November 17, 2010 submitted to
Entropy
22 of 43 computation don’t open a back door to the sign problem. This clears the way for developing detailedalgorithms for specific models of fermionic systems of particular interest. Some of the most importantopen questions a quantum computer of modest size could solve are models involving strongly interactingfermions, such as for high temperature superconductors.
One of the important fermionic models that has received detailed analysis is the Hubbard model, oneof the most basic microscopic descriptions of the behaviour of electrons in solids. Analytic solutionsare challenging, especially beyond one dimension, and while ferromagnetism is obtained for the rightparameter ranges, it is not known whether the basic Hubbard model produces superconductivity. Thedifficulties of classical simulations thus provide strong motivation for applying quantum simulation tothe Hubbard model. The Hubbard Hamiltonian ˆ H γV is ˆ H γV = − γ X h j,k i ,σ ˆ C † j,σ ˆ C k,σ + V X j ˆ n j, ↑ ˆ n j, ↓ (19)where ˆ C † j,σ ˆ C k,σ are the fermionic creation and annihilation operators, σ is the spin (up or down), ˆ n j, ↑ ˆ n j, ↓ are the number operators for up spin and down spin states at each site j , γ is the strength of the hoppingbetween sites, and V is the on site potential. Abrams and Lloyd [1] describe two different encodings ofthe system into the quantum simulator. An encoding using the second quantization is more natural sincethe first quantization encoding requires the antisymmetrization of the wavefunction “by hand”. However,when the number of particles being simulated is a lot lower than the available number of qubits, the firstquantization is more efficient. In second quantization, there are four possible states each site can be in:empty, one spin up, one spin down, and a pair of opposite spin. Two qubits per site are thus requiredto encode which of the four states each site is in. It is then a simple extension of the Lloyd method toevolve the state of the system according to the Hubbard Hamiltonian. Somma et al [109] describe how touse this method to find the energy spectrum of the Hubbard Hamiltonian for a fermionic lattice system.They perform a classical computer simulation of a quantum computer doing a quantum simulation, todemonstrate the feasibility of the quantum simulation. The Hubbard model is the natural Hamiltonian inoptical lattice schemes, so there has been considerable development towards special purpose simulatorsbased on atoms in optical lattices, these are discussed in section 9.2. Pairing Hamiltonians are an important class of models for many-body systems in which pairwiseinteractions are typically described using fermionic (or bosonic) creation and annihilation operators { c m , c † m } . Nucleons in larger atomic nuclei can be described by pairing Hamiltonians, and Bardeen,Copper and Schrieffer (BCS) [7] formulated a model of superconductivity as a a pairing Hamiltonian inthe 1950s. The BCS model of superconductivity is still not fully understood, so quantum simulationscould be useful to improve our knowledge of superconducting systems, especially for realistic materialswith imperfections and boundary effects. While the BCS ansatz is exact in the thermodynamic limit, itis not known how well it applies to small systems [69].ersion November 17, 2010 submitted to Entropy
23 of 43
The BCS Hamiltonian for a fully general system can be written H BCS = N X m =1 ǫ m c † m c m + c †− m c − m ) + N X m,l =1 V ml c † m c †− m c − l c l (20)where the parameters ǫ m and V ml specify the self energy of the m th mode and the interaction energy ofthe m th and l th modes respectively, while N is the total number of occupied modes (pairs of fermionswith opposite spin). Wu et al [133] developed a detailed method for quantum simulation of equation (20).The two terms in the BCS Hamiltonian do not commute, therefore the simulation method requires theuse of Trotterization (see section 2.1) so the two parts can be individually applied alternately. This meansthat any simulation on a universal quantum computer will require many operations to step through thetime evolution, which will stretch the experimentally available coherence times. Savings in the numberof operations are thus important, and recent work by Brown at al [18] adapting the method to a qubusarchitecture reduces the number of operations required in the general case from O ( N ) for NMR to O ( N ) for the qubus. Pairing Hamiltonians are used to describe many processes in condensed matterphysics and therefore a technique for simulating the BCS Hamiltonian should be adaptable to numerousother purposes. For simulation on qubit quantum computers (as opposed to special purpose quantum simulators), wefirst need an efficient mapping between the particles being simulated and the spin-1/2 algebra of thequbit systems. Somma et al [110] discuss in detail how to map physical particles onto spin-1/2 systems.For fermions there is a one-to-one mapping between the fermionic and spin-1/2 algebras. Particularlyin the second quantization this allows a simple mapping that can be generalised to all anyonic systemswhich obey the Pauli exclusion principle, or generalised versions of it. For bosonic systems there is nodirect mapping between the bosonic algebra and spin 1/2 algebra. Therefore Somma et al propose usinga direct mapping between the state of the two systems, provided there is a limit on the number of bosonsper state. This mapping is less efficient but allows simulations to be conducted on the bosonic systems.Systems of indistinguishable particles require special state preparation to ensure the resulting stateshave the correct symmetry. Ortiz et al [88] developed a method for fermions that was then adapted forbosons by Somma et al [110]. In general, a quantum system of N e fermions with an anti-symmetrizedwavefunction | Ψ e i can be written as a sum of Slater determinants | Φ α i| Ψ e i = n X α =1 a α | Φ α i (21)where n is an integer and P nα =1 | a α | = 1 . The individual Slater determinants can be prepared effi-ciently using unitary operations. Provided the desired state doesn’t require an exponential sum of Slaterdeterminants, with the help of n ancilla qubits it is possible to prepare the state n X α =1 a α | α i ⊗ | Φ α i (22)ersion November 17, 2010 submitted to Entropy
24 of 43 where | α i is a state of the ancilla with the α ’th qubit in state | i and the rest in state | i . A further registerof n ancillas is then used to convert the state so that there is a component with the original ancillas in theall zero state, n X α =1 a α | i a ⊗ | Φ α i (23)associated with the required state of the fermions. A measurement in the z -basis selects this outcome(all zeros) with a probability of /n . This means the preparation should be possible using an average of n trials.A general bosonic system can be written as a linear combination of product states. These productstates can be mapped onto spin states and then easily prepared by flipping the relevant spins. Oncethe bosonic system has been written as a linear combination of these states, a very similar preparationprocedure to the one for fermionic systems can be used [110].While the above method using Slater determinants is practical when working in second quantization,this isn’t always convenient for atomic and molecular systems. Ward et al [127] present a system for effi-ciently converting states prepared using Slater determinants in second quantization to a first quantizationrepresentation on a real space lattice. This can be used for both pure and mixed states. Lattice gauge theories are important in many areas of physics, and one of the most important examplesfrom a computational perspective is quantum chromodynamics (QCD). Classical QCD simulations areextremely computationally intensive, but very important for predicting the properties of fundamentalparticles. Providing more efficient quantum simulations would be very useful to advance the field. Thequantum lattice gas method developed by Boghosian and Taylor [12] (discussed in section 5.2) is suitablefor simulating lattice gauge theories, using similar methods to the lattice QCD simulations currentlyperformed classically, but with the benefit of a quantum speed up. Byrnes and Yamamoto [24] providea more general method. They map the desired Hamiltonian to one involving only Pauli operations andone and two qubit interactions. This is then suitable for any qubit-based universal quantum simulator.They focus on the U(1), SU(2) and SU(3) models, but their methods easily generalise to higher orderSU(N) theories. To conduct the simulation efficiently it is necessary to use a truncated version of themodel, to keep the number of qubits finite. They demonstrate that the number of operations requiredfor the time evolution and for the preparation of the necessary initial states are both efficient. To getresults inaccessible to classical computers, of the order of qubits will be required. Despite this, thealgorithm has advantages over classical techniques because the calculations are exact up to a cut off, andwith simple adaptions it can be extended to to simulate fermionic systems.Methods suitable for special purpose quantum simulators have been presented by Sch¨utzhold andMostame [103] and Tewari et al [115]. Sch¨utzhold and Mostame describe how to simulate the O(3)nonlinear σ -model, which is of interest to the condensed matter physics community as it applies to spinsystems. It also reproduces many of the key properties of QCD, although it is only a toy model in thiscontext. To conduct their simulation, Sch¨utzhold and Mostame propose using hollow spheres to trapelectrons, described in more detail in section 10.1. Tewari et al [115] focus specifically on compact U(1)lattice gauge theories that are appropriate for dipolar bosons in optical lattices. The basic Hamiltonian inersion November 17, 2010 submitted to Entropy
25 of 43 optical lattices is the Hubbard Hamiltonian, equation (19), but different choices of atom can enhance theHamiltonian with different nearest neighbour interactions. The specific example chosen by Tewari et alis chromium, which has a magnetic dipolar interaction that can provide the extra term in the Hamiltonian.The ratio of the two types of couplings (Hubbard and dipolar) can be varied over a wide range by tuningthe Hubbard interaction strength using Feshbach resonances. Further types of relativistic quantum fieldtheories that can be simulated by atoms in optical lattices are presented by Cirac et al [32].This concludes the theory part of our review, and provides a natural point to move over to considera-tion of the different physical architectures most suited to quantum simulation.
Part II
Experimental Implementation of QuantumSimulations
7. Overview
As we have seen, while algorithms for quantum simulation are interesting in their own right, thereal drive is towards actual implementations of a useful size to apply to problems we cannot solve withclassical computers. The theoretical studies show that quantum simulation can be done with a widevariety of methods and systems, giving plenty of choices for experimentalists to work with. Questionsremain around the viability of longer simulations, where errors may threaten the accuracy of the results,and long sequences of operations run beyond the available coherence times. As with quantum computingin general, the main challenge for scaling up is controlling the decoherence and correcting the errors thataccumulate from imperfect control operations. Detailed treatment of these issues is beyond the scopeof this review and well-covered elsewhere (see, for example, Devitt et al [38]). The extra concern forquantum simulation lies in the unfavorable scaling of errors with system size, as discussed in section 2.2.In section 2 we described how to obtain universal quantum simulation from particular sets of re-sources, mainly a fixed interaction with local unitary controls. Building a universal quantum simulatorwill allow us to efficiently simulate any quantum system that has a local or efficiently describable Hamil-tonian. On the other hand, the generality of universal simulation may not be necessary if the problem weare trying to solve has a specific Hamiltonian with properties or symmetries we can exploit to simplifythe simulation. If the Hamiltonian we want to simulate can be matched with a compatible interac-tion Hamiltonian in the quantum simulator, then there are are likely to be further efficiencies availablethrough simpler control sequences for converting one into the other. From the implementation perspec-tive, a special purpose simulator may be easier to build and operate, a big attraction in these early stagesof development. Most architectures for quantum computing are also suitable for universal quantum sim-ulation. However, the range of experimental possibilities is broader if we are willing to specialise tothe specific Hamiltonians in the quantum simulator. This allows more to be achieved with the samehardware, and is thus the most promising approach for the first useful quantum simulations.ersion November 17, 2010 submitted to
Entropy
26 of 43
Buluta and Nori [23] give a brief overview of quantum simulation that focuses on the various possiblearchitectures and what sort of algorithms these could be used for. There is broad overlap of relevantexperimental techniques with those for developing quantum computers in general, and many issues andconsiderations are common to all applications of quantum computers. In this paper, we concentrateon implementations that correspond to the theoretical aspects we have covered. Many experimentalimplementations of quantum simulation to date have been in NMR quantum computers. This is not ascalable architecture, but as a well-developed technology it provides an invaluable test bed for smallquantum computations. Optical schemes based on entangled photons from down-conversion have alsobeen used to implement a variety of small quantum simulations, but since photons don’t normally interactwith each other, they don’t provide a natural route to special purpose quantum simulators. We describethe lessons learned from these quantum simulations in section 8. We then turn to simulators built bytrapping arrays of ions, atoms, and electrons in sections 9 and 10. Most of these have applications bothas universal quantum simulators and for specific Hamiltonians, with promising experiments and rapidprogress being made with a number of specific configurations.
8. Proof-of-principle experiments
Some of the most advanced experimental tests of quantum computation have been performed usingtechnology that does not scale up beyond ten or so qubits. Nonetheless, the information gained fromthese experiments is invaluable for developing more scalable architectures. Many of the control tech-niques are directly transferable in the form of carefully crafted pulse sequences with enhanced resilienceto errors and imperfections. Observing the actual effects of decoherence on the fidelities is useful toincrease our understanding of the requirements for scaling up to longer sequences of operations.
Nuclear Magnetic Resonance is a highly developed technology that provides an adaptable toy sys-tem for quantum computing (see Jones [62] for a comprehensive review). A suitable molecule withatoms having various nuclear spins is prepared, often requiring chemical synthesis to substitute dif-ferent isotopes with the required spins. A solution of this molecule then provides an ensemble whichcan be collectively controlled by applied magnetic fields and radio frequency (rf) pulses. The nuclearspins of the different atomic species will in general have different resonant frequencies, allowing themto be addressed separately. Read out is provided by exploiting spin echo effects. Liquid state NMRisn’t considered to be scalable due to the difficulty of addressing individual qubits in larger molecules.Nonetheless, the relative ease with which quantum algorithms can be implemented for small systems hasmeant that many proof-of-principle experiments have been carried out using NMR. These are often ofonly the smallest non-trivial size, using as few as one or two qubits, but are still useful for developing andtesting the control sequences. The real advantage lies in the flexibility of applying gates through radiofrequency (RF) pulses. This allows NMR to outperform other test-bed systems such as optics, whereeach gate requires its own carefully aligned components on the bench. Since most quantum algorithmshave been tested in NMR by now, we select for discussion a few that bring out important points aboutthe experimental feasibility of quantum simulation in general.ersion November 17, 2010 submitted to
Entropy
27 of 43
Numerous groups have performed NMR quantum simulations of spin chains. The Heisenberg inter-action is already present in NMR in the form of the ZZ interaction ( X , Y , Z are used to denote thePauli spin operators). This allows more complex Heisenberg interactions to be simulated by using localunitary operations to rotate the spin between the X , Y and Z orientations. These simulations are thusa simple example of using one fixed Hamiltonian – ZZ in this case – to simulate another, as describedtheoretically in section 2. This allows the investigation of interesting properties of these spin chains suchas phase transitions [90,91], the propagation of excitons [67] and the evolution under particular interac-tions [136]. Peng et al [90] and Khitrin et al [67] found that the decoherence time of the system is oftentoo short to get meaningful results, even for these small simulations. The limited decoherence timeswere turned into an advantage by Alvarez et al [4], to study the effects on quantum information transferin spin chains. As expected, they were able to show that decoherence limits the distance over whichquantum information can be transferred, as well as limiting the time for which it can be transferred. Thisis an example of using the noise naturally present in the quantum computer to simulate the effects on thesystem under study, as described in section 5.4.Tseng et al [118] describe how to simulate a general three-body interaction using only the ZZ in-teraction present in NMR, and experimentally demonstrated a ZZZ interaction. This provided proof ofprinciple for extending the repertoire of NMR quantum simulation beyond two-body Hamiltonians, latercomprehensively generalised theoretically by D¨ur et al [41] (see section 2.3). Liu et al [76] demonstratedexperimentally that four-body interactions in a four qubit NMR quantum computer can be simulated towithin good agreement of their theoretical calculations.Pairing Hamiltonians (see section 6.2) are of particular importance for quantum simulation, with thefermionic systems they describe including superconductors and atomic nuclei. The long range interac-tions put simulation of general pairing systems beyond the reach of classical computers. Studies usingNMR have focused on the BCS Hamiltonian, equation (20), which is a pairing Hamiltonian with inter-actions composed of Pauli spin operators. However, because it consists of two non-commuting parts,these have to be implemented individually and then recombined using the Trotter-Suzuki formula, asdescribed in section 2.1. Wu et al. [133] provided a detailed discussion of how to make this efficient forNMR, and their method was implemented experimentally by Brown et al [19] on three qubits. As wellas their insightful comments on the scaling of errors in the simulation, discussed in section 2.2, they alsoadded artificial noise to their simulations to verify the scaling. This confirmed that simulation of largersystems will be challenging, due to the high number of operations required for the Trotter expansion,and correspondingly large error correction overheads.Negrevergne et al [83] simulate a many-body Fermi system that obeys the Fano-Anderson model, aring system with an impurity at the centre. This can be done with three NMR qubits, once the transla-tional symmetry in the ring has been taken into account and the fermion modes mapped to the qubits.To minimise problems with decoherence caused by running the system for a long time, Negrevergne de-signed and implemented an approximate refocusing scheme. This provided a scalable algorithm, whichcan be adapted to other architectures as more powerful quantum simulators are built.Although bosons are easier to simulate classically than fermions (because they don’t suffer from the“sign problem”) for quantum simulations they are harder, due to the unlimited size of the Hilbert space.The Hilbert space has to be artificially truncated, and this limits the accuracy. Simulations of a bosonicersion November 17, 2010 submitted to
Entropy
28 of 43 system have been carried out by Somaroo et al [108], who chose the truncated harmonic oscillator. Thelimitations due to the truncation are quite significant in a small NMR simulation, and scaling up wouldbe difficult, as a larger system would require small couplings within the NMR simulator that wouldseverely limit the time scale of the experiment. As with other simulations, the decoherence time limitsthe duration of the experiment, which in this case corresponds to the number of periods of the oscillatorwhich can be simulated.Du et al [40] have simulated molecular hydrogen in order to obtain its ground state energy. To dothis they use the algorithm presented by Aspuru-Guzik et al [5], described in section 5.3. This is animportant class of quantum simulations, because it turns out to be more efficient in the quantum caseto simulate the dynamics exactly, instead of following the approximations used to do these calculationsclassically. They thus offer the possibility of significant improvements for quantum chemistry, givena large enough quantum computer. With NMR systems, the simulations are limited to hydrogen, andwhile the decomposition of the molecular evolution operator scales efficiently, Du et al [40] are not surewhether the same is true of their adiabatic state preparation method. Nonetheless, this is an importantproof of principle for the method and application.
Linear optics, with qubits encoded in the photonic degrees of freedom, are an attractive option forquantum computing due to the relatively straightforward experimental requirements compared to archi-tectures requiring low temperatures and vacuum chambers. The main difficulty is obtaining a suitablenonlinear interaction, without which only regimes that can be simulated efficiently classically can bereached. Current experiments generally use less scalable techniques for generating the nonlinear opera-tion, such as using entangled pairs of photons from down-conversion in nonlinear crystals, or measure-ments with probabilistic outcomes, so the experiment has to be repeated until it succeeds.Lanyon et al [72] used the algorithm presented by Aspuru-Guzik et al [5] to simulate molecules. Thequbits were encoded in the polarisation of single photons, with linear optical elements and a nonlinearityobtained through projective measurements used to provide the necessary control. Ma et al [78] used thepolarisation states of four photons to simulate a spin system of four spin − / particles with arbitraryHeisenberg-type interactions between them. They used measurements to induce the interaction betweenthe spins, and were able to measure ground state energy and quantum correlations for the four spins.While photonic systems do not have an intrinsic Hamiltonian that is adaptable for special purposequantum simulation, they are expected to come into their own as universal quantum computers. There arestrong proposals for scalable architectures based on photonic systems [70,86] that can also be exploitedfor quantum simulation.
9. Atom trap and ion trap architectures
Among the architectures for quantum computing predicted to be the most scalable, qubits based onatoms or ions in trap systems are strongly favoured [73,102]. Locating the atoms or ions in a trap allowseach qubit to be distinguished, and in many cases individually controlled. Review of the many designsthat are under development is beyond the scope of this article; while any design for a quantum computerersion November 17, 2010 submitted to
Entropy
29 of 43 is also suitable for quantum simulation, we focus here on arrays of atoms or ions where the intrinsiccoupling between them can be exploited for quantum simulation.Trapped ions form a Coulomb crystal due to their mutual repulsion, which separates them sufficientlyto allow individual addressing by lasers. Coupling between them can be achieved via the vibrationalmodes of the trap, or mediated by the controlling lasers. Atoms in optical lattices formed by counter-propagating laser beams are one of the most promising recent developments. Once the problem ofloading a single atom into each trap was overcome by exploiting the Mott transition [54], the road wasclear for developing applications to quantum computing and quantum simulation. For comprehensivereviews of experimental trap advances, see Wineland [129] for ion trapping, and Bloch et al [11] for coldatoms.Jan´e et al [60] consider quantum simulation using both neutral atoms in an optical lattice and ionsstored in an array of micro traps. This allows them to compare the experimental resources required foreach scheme, as well as assessing the feasibility of using them as a universal quantum simulator. Atomsin optical lattices have the advantage that there is a high degree of parallelism in the manipulation ofthe qubits. The difficulty of individually addressing each atom, due to the trap spacing being of thesame order as the wavelength of the control lasers, can be circumvented in several ways. If the atoms arespaced more widely, so only every fifth or tenth trap is used, for example, then individual laser addressingcan be achieved. Applied fields that intersect at the target atom can also be used to shift the energy levelssuch that only the target atom is affected by the control laser. Jan´e et al conclude that both architecturesshould be suitable for quantum simulation.An alternative approach is to avoid addressing individual atoms altogether. Kraus et al [71] explorethe potential of simulations using only global single-particle and nearest neighbor interactions. This isa good approximation for atoms in optical lattices, and the three types of subsystem they consider –fermions, bosons, and spins – can be realised by choosing different atoms to trap in the optical latticeand tuning the lattice parameters to different regimes. They make the physically reasonable assumptionthat the interactions are short range and translationally invariant. They also apply an additional constraintof periodic boundary conditions, to simplify the analysis. Most physical systems have open rather thanperiodic boundary conditions, so their results may not be immediately applicable to experiments. Fora quadratic Hamiltonian acting on fermions or bosons in a cubic lattice, Kraus et al found that genericnearest neighbor interactions are universal for simulating any translationally invariant interaction whencombined with all on-site Hamiltonians (the equivalent of any local unitary) provided the interactionsacted along both the axes and diagonals of the cubic lattice (compare lattice gases, section 5.2). However,for spins in a cubic lattice, there is no set of nearest-neighbor interactions which is universal and noteven all next-to-nearest neighbor interactions could be simulated from nearest-neighbor interactions. Itis possible that different encodings to those used by Kraus et al could get around this restriction, but thefull capabilities of spin systems on a cubic lattice remains an open problem. Their results demonstratethat schemes which don’t provide individual addressability can still be useful for simulating a large classof Hamiltonians.Coupled cavity arrays are a more recent development, combining the advantages a cavity confers incontrolling an atom with the scalability of micro-fabricated arrays. While there is a trade off betweenthe relative advantages of the various available trapping architectures, with individual addressabilityersion November 17, 2010 submitted to
Entropy
30 of 43 and greater control resulting in systems with a poorer scaling in precision, each scheme has its ownadvantages and the experiments are still in the very early stages.
The greater degree of quantum control available for ions in traps, compared with atoms in optical lat-tices, means that research on using ion traps for simulating quantum systems is further developed. Clarket al [34] and Buluta and Hasegawa [22] present designs based on planar RF traps that are specificallygeared towards quantum simulations. They focus on producing a square lattice of trapped ions, but theirresults can be generalised to other shapes such as hexagonal lattices (useful for studying systems suchas magnetic frustration). Clark et al carried out experimental tests on single traps that allowed them toverify their numerical models of the scheme are accurate. They identify a possible difficulty when it isscaled to smaller ion-ion distances. As the ion spacing decreases, the secular frequency increases, whichmay make it difficult to achieve coupling strengths that are large relative to the decoherence rate.As with the simulations done with NMR computers, some of the earliest work on ion trap simulatorshas focused on the simulation of spin systems. Deng et al [36] and Porras and Cirac [93,94] discussthe application of trapped ions to simulate the Bose-Hubbard model, and Ising and Heisenberg interac-tions. This would allow the observation and analysis of the quantum phase transitions which occur inthese systems. They mention three different method for trapping ions that could be used to implementtheir simulation schemes. Arrays of micro ion traps and linear Paul traps use similar experimental con-figurations, although Paul traps allow a long range interaction that micro ion trap arrays don’t. Bothschemes are particularly suited to simulating an interaction of the form XYZ. Penning traps containingtwo-dimensional Coulomb crystals could also be used, and this would allow hexagonal lattices to beapplied to more complex simulations, such as magnetic frustration. Alternatively [94], the phonons inthe trapped ions can be viewed as the system for the simulation. Within the ion trap system phonons canneither be created nor destroyed, so it is possible to simulate systems such as Bose-Einstein condensates,which is more difficult using qubit systems.Friedenauer et al [46] have experimentally simulated a quantum phase transition in a spin systemusing two trapped ions. The system adiabatically traverses from the quantum paramagnetic regime to thequantum (anti)-ferromagnetic regime, with all the parameters controlled using lasers and RF fields. Toextract data over the full parameter range the experiment was repeated times, to obtain good statisticsfor the probability distributions. While the simulation method is scalable, involving global applicationof the control fields, it isn’t clear the data extraction methods are practical for larger simulations. Thiswork is significant for being one of the few detailed proof-of-concept experimental studies done in asystem other than NMR, and demonstrates the progress made in developing other architectures. InGerritsma et al [53], they simulate the Klein paradox, in which electrons tunnel more easily throughhigher barriers than low ones, by precisely tuning the parameters in their trapped ion system. Edwardset al [43] have simulated an Ising system with a transverse field using three trapped ions. They alter theHamiltonian adiabatically to study a wide range of ground state parameters, thereby mapping out themagnetic phase diagram. This system is scalable up to many tens of ions, which would reach regimescurrently inaccessible to classical computation, allowing behavior towards the thermodynamic limit tobe studied in detail for general and inhomogeneous spin systems.ersion November 17, 2010 submitted to Entropy
31 of 43
Proof-of-principle simulations have also been done with single ions. While less interesting thancoupled ions, because the coupled systems are where the Hilbert space scaling really favours quantumsimulations, these still test the controls and encoding required. For example, Gerritsma et al [52] sim-ulated the Dirac equation using a single trapped ion, to model a relativistic quantum particle. The highlevel of control the ion trap provides allows information about regimes and effects that are difficult tosimulate classically such as Zitterbewegung.
Atoms trapped in the standing waves created by counter-propagating lasers are one of the most excit-ing recent developments in quantum computing architectures. Their potential for the quantum simulationof many-body systems was obvious from the beginning, and has been studied by many groups since theinitial work of Jan´e et al [60]. Trotzky et al [117] compare optical lattice experimental data with theirown classical Monte Carlo simulations, to validate the optical lattice as a reliable model for quantumsimulations of ultra-cold strongly interacting Bose gases. They find good agreement for system sizes upto the limit of their simulations of × particles.The most promising way to use atoms in optical lattices for quantum simulation is as a special purposesimulator, taking advantage of the natural interactions between the atoms. This will allow larger systemsto be simulated well before this becomes possible with universal quantum computers. The followingthree examples illustrate the potential for thinking creatively when looking for the best methods to simu-late difficult systems or regimes. Johnson et al [61] discuss the natural occurrence of effective three-bodyand higher order interactions in two-body collisions between atoms in optical lattices. They use theseto explain experimental results showing higher-than-expected decoherence rates. Tuning these many-body interactions could be done using Feshbach resonance control or manipulating the lattice potentials,allowing them to be used for the simulation of effective field theories, see section 5.2. Ho et al [58]propose that simulating the repulsive Hubbard model is best done using the attractive Hubbard model,which should be easier to access experimentally. Mapping between different regimes in the same modelshould be simpler to implement, allowing access to results that are usually difficult experimentally. Aswith the trapped ion schemes, one of the most common subjects for simulation is many-body quantumphase transitions. Kinoshita et al [68] use rubidium-87 atoms trapped in a combination of two light traps.By altering the trap strengths, the interactions between the atoms can be controlled, allowing them tobehave like a one-dimensional Tonks-Girardeau gas through to a Bose-Einstein condensate. They findvery good agreement with theoretical predictions for a 1D Bose gas. This is a good example of a specialpurpose simulator, since there are no individual controls on the atoms, allowing only regimes dictated bythe globally controlled coupling to be realised. Optical lattices are not the only way to trap arrays of atoms. Coupled cavity arrays offer control overindividual atoms much more conveniently than with optical lattices. In coupled cavities the qubits arerepresented by either polaritons or hyperfine ground state levels, with the former allowing continuouscontrol, and the latter individual addressability. The cavities themselves are an artificial system grownersion November 17, 2010 submitted to
Entropy
32 of 43 on a microchip in which the qubits on the chip interact with the field mode of the cavity, and the cavitiesare coupled by the exchange of photons. A simulation of the Heisenberg model is generally one of theearliest proof-of-principle simulations for a new architecture, and Cho et al [31] propose a technique toallow these coupled cavity arrays to do this. Their method should apply generally to different physicalimplementations of micro cavities. Kay et al [65] and Chen et al [26] both discuss implementation of theHeisenberg model in specific coupled cavity architectures. They confirm that control over nearest andnext-nearest neighbour coupling can be achieved, but without short control pulses only global controlsare available. Schemes that give individual addressability need short control pulses to modify the intrin-sic interactions. These may necessitate the use of the Trotter approximation, making it more difficultto obtain high precision results in cavity arrays. Ivanov et al [59] look at exploiting the polaritons incouple cavity arrays to simulate phase transitions, in the same way as Porras and Cirac [94] considerusing phonons in ion traps. These proposals show the versatility and potential of coupled cavity arraysfor further development.
10. Electrons and excitons
While atoms and ions in arrays of traps are the most promising scalable architectures for quantumsimulation at present, electrons can also be controlled and trapped suitably for quantum simulation. Thiscan be done either by confining free electrons, or exploiting the electrons-hole pairs in quantum dots.Superconducting qubits harness collective states of electrons or quantized flux to form qubits from su-perconducting circuits with Josephson junctions. We briefly describe applications of these architecturesto quantum simulations that exploit their special features.
Spin lattices are arrays of electrons, where the spin of the electron is used as the qubit. Persuading theelectrons to line up in the required configuration can be done in various ways. Mostame and Sch¨utzhold[82] propose to trap electrons using pairs of gold spheres attached to a silicon substrate under a thin filmof helium. The electrons float on the surface of the helium and induce a charge on the spheres, whichgenerates a double well potential and hence traps the electrons. Mostame and Sch¨utzhold describe how touse this architecture to simulate an Ising spin chain, from which the generalisation to more complicatedmodels can easily be made. This model for trapping electrons is suggested to be more scalable than atomor ion traps. However, it may be difficult to realise experimentally, due to the precise controls needed,particularly in the thickness of the film of helium. Byrnes et al [25] propose to confine a 2D electrongas using surface acoustic waves to create an ‘egg-carton’ potential. The advantage that this system hasover optical lattices is that it produces long range interactions. It should therefore be more suitable forsimulating Hubbard dynamics, which originate from the long range Coulomb interaction. This schemewill allow observations of quantum phase transitions in systems of strongly correlated electrons as wellas the study of the metal-insulator transition.ersion November 17, 2010 submitted to
Entropy
33 of 43
The trapped electrons or holes in a semiconductor quantum dot can be exploited as qubits, withcontrol provided via gate electrodes or optical fields. Instead of focusing on just the qubit degrees offreedom, the whole quantum dot can be thought of as an artificial atom, which may thus make themsuitable to simulate chemical reactions. Quantum dots are now easy to make; the problem is to controltheir parameters and location so they can be used collectively in a predictable manner. Smirnov et al[105] discuss using the coupling of quantum dots to model bond formation. They consider one of thesimplest possible systems for proof of principle calculations, the interaction H + H → H + H , wherethe molecular bond between a pair of hydrogen atoms switches to a different pair. This can be simulatedwith a system of three coupled quantum dots, such as has been demonstrated experimentally [48,125].The high level of control in quantum dot systems will allow the detailed study of chemical reactions inconditions not available in real molecules. Superconducting architectures have been developing steadily although in general they are a few yearsbehind the atom and ion trap systems. As universal quantum computers they are equally suitable inprinciple for quantum simulation. Charge, phase and flux qubits can be constructed using Josephsonjunction superconducting circuits, with controls provided by a variety of externally applied fields.An ingenious proposal from Pritchard et al [99] describes how to use a systems of Josephson junctionsfor simulation of molecular collisions. The simulations are restricted to the single excitation subspace ofan n − qubit system, which requires only an n × n -dimensional Hamiltonian. In return for this subspacerestriction, the individual parameters in the Hamiltonian can be varied independently, providing a highlevel of generality to the simulation. They use a time-dependent rescaling of time to optimise the actualrun time of the simulation to minimise decoherence effects. They test their method in an experimentwith three tunable coupled phase qubits simulating a three-channel molecular collision between Na andHe. They study the fidelities achieved, and determine the relationship between the fidelity and length oftime the simulation is run for. Higher fidelities require longer simulation times, but this is independentof n , showing this aspect of the method is fully scalable.
11. Outlook
Quantum simulation is one of the primary short- to mid-term goals of many research groups focusingon quantum computation. The potential advances that even a modest quantum simulator would unleashare substantial in a broad range of fields, from material science (high temperature superconductors andmagnetic materials) to quantum chemistry. Quantum simulations are particular promising for simulatingfermionic many-body systems and their phase transitions, where the “sign problem” limits efficient clas-sical numerical approximation techniques. Larger quantum simulators could tackle problems in latticeQCD that currently consume a sizable fraction of scientific computing power, while quantum simulationsof quantum chemistry have wide-ranging applications reaching as far as the design of molecules for newdrugs. We have seen that the theoretical foundations have been laid quite comprehensively, providingdetailed methods for efficient quantum simulators, and calculations that confirm their viability.ersion November 17, 2010 submitted to
Entropy
34 of 43
One significant issue that remains to be fully addressed is the precision requirements for larger scalequantum simulations. Due to the one-to-one mapping between the Hilbert space of the system andthe Hilbert space of the quantum simulator, the resources required for a given precision scale inverselywith the precision. Compared with digital (classical and qubit) computations, this is exponentially morecostly. When combined with the long control sequences required by Trotterization, this threatens theviability of such simulations of even fairly modest size.Special purpose quantum simulators designed with similar Hamiltonians to the quantum system beingstudied are the front runners for actually performing a useful calculation beyond the reach of conven-tional computers. These come in many forms, matching the variety of common Hamiltonians describingphysical systems. Among the most developed and versatile, ion traps and atoms in optical lattices arecurrently in the lead, although micro-fabrication techniques are allowing more sophisticated solid statetrap arrays to catch up fast. Actual experimental systems capable of quantum simulations of a significantsize are still in the future, but the designs and proof-of-concept experiments already on the table providea strong base from which to progress on this exciting challenge.
Acknowledgments
We thank Clare Horsman for careful reading of the manuscript. KLB is supported by a UK EPSRCCASE studentship from Hewlett Packard. VMK is funded by a UK Royal Society University ResearchFellowship. WJM acknowledges part support from MEXT in Japan.
References
1. Daniel S. Abrams and Seth Lloyd. Simulation of many-body Fermi systems on a universal quantumcomputer.
Phys. Rev. Lett. , 79:2586–2589, 1997. doi: 10.1103/PhysRevLett.79.258. URLhttp://link.aps.org/abstract/PRL/v79/p2586.2. Daniel S. Abrams and Seth Lloyd. Quantum algorithm providing exponential speed increase forfinding eigenvalues and eigenvectors.
Phys. Rev. Lett. , 83(24):5162, December 1999. doi:10.1103/PhysRevLett.83.5162. URL http://prola.aps.org/abstract/PRL/v83/i24/p5162 1.3. Dorit Aharonov and Amnon Ta-Shma. Adiabatic quantum state generation and statistical zeroknowledge. In
Proceedings of the 35th Annual ACM Symposium on Theory of Computing, SanDiego, California, USA , pages 20–29, New York, USA, June 2003 2003. ACM Press. URLhttp://arxiv.org/abs/quant-ph/0301023.4. Gonzalo A. ´Alvarez and Dieter Suter. NMR quantum simulation of localization effects induced bydecoherence.
Phys. Rev. Lett. , 104(23):230403, Jun 2010. doi: 10.1103/PhysRevLett.104.230403.5. Alan Aspuru-Guzik, Anthony D. Dutoi, Peter J. Love, and Martin Head-Gordon. Simulated quan-tum computation of molecular energies.
Science
Phys. Rev. A , 64(6):062302, Nov2001. doi: 10.1103/PhysRevA.64.062302. URL http://dx.doi.org/10.1103/PhysRevA.64.062302.ersion November 17, 2010 submitted to
Entropy
35 of 43
7. J. Bardeen, L. N. Cooper, and J. R. Schrieffer. Theory of superconductivity.
Phys. Rev. , 108:1175,1957.8. Giuliano Benenti and Giuliano Strini. Quantum simulation of the single-particle Schr¨odingerequation.
American Journal of Physics , 76(7):657–662, 2008. URL http://link.aip.org/link/?AJP/76/657.9. C. H. Bennett, J. I. Cirac, M. S. Leifer, D. W. Leung, N. Linden, S. Popescu, and G. Vidal. Optimalsimulation of two-qubit Hamiltonians using general local operations.
Phys. Rev. A , 66(1):012305,July 2002. doi: 10.1103/PhysRevA.66.012305. URL http://dx.doi.org/10.1103%2FPhysRevA.66.012305.10. Dominic W. Berry, Graeme Ahokas, Richard Cleve, and Barry C. Sanders. Efficient quantumalgorithms for simulating sparse Hamiltonians.
Commun. Math. Phys. , 270:359–371, 2007. doi:10.1007/s00220-006-0150-x. URL http://springerlink.com/content/hk7484445j37r228/.11. Immanuel Bloch, Jean Dalibard, and Wilhelm Zwerger. Many-body physics with ultracold gases.
Rev. Mod. Phys. , 80(3):885–964, Jul 2008. doi: 10.1103/RevModPhys.80.885.12. Bruce M. Boghosian and Washington Taylor. Quantum lattice-gas model for the many-particleSchr¨odinger equation in d dimensions.
Phys. Rev. E , 57(1):54–66, January 1998. doi: 10.1103/PhysRevE.57.54. URL http://link.aps.org/abstract/PRE/v57/p54.13. Bruce M. Boghosian and Washington Taylor. Simulating quantum mechanics on a quantum com-puter.
Physica D , 120(1-2):30–42, September 1998. doi: 10.1016/S0167-2789(98)00042-6.14. S. Boixo, E. Knill, and R. D. Somma. Eigenpath Traversal by Phase Randomization.
QuantumInformation and Computation , 9:0833–0855, 2009. ArXiv:0903.1652v2[quant-ph].15. Sergey Bravyi, David P. DiVincenzo, Daniel Loss, and Barbara M. Terhal. Quantum simulationof many-body Hamiltonians using perturbation theory with bounded-strength interactions.
Phys.Rev. Lett. , 101(7):070503, Aug 2008. doi: 10.1103/PhysRevLett.101.070503. URL http://dx.doi.org/10.1103/PhysRevLett.101.070503.16. Michael J. Bremner, Christopher M. Dawson, Jennifer L. Dodd, Alexei Gilchrist, Aram W. Harrow,Duncan Mortimer, Michael A. Nielsen, and Tobia J. Oscorne. Practical scheme for quantumcomputation with any two-qubit entangling gate.
Phys. Rev. Lett. , 89:247902, 2002. URLhttp://prola.aps.org/abstract/PRL/v89/i24/e247902.17. Michael J. Bremner, Jennifer L. Dodd, Michael A. Nielsen, and Dave Bacon. Fungible dynamics:There are only two types of entangling multiple-qubit interactions.
Phys. Rev. A , 69:012313, 2004.18. K. L. Brown, S. De, V. M. Kendon, and W. J. Munro. Improving the efficiency of quantumsimulation by using a qubus quantum computer, 2010. URL http://arxiv.org/abs/1011.2984.19. Kenneth R. Brown, Robert J. Clark, and Issac L. Chuang. Limitations of quantum simulationexamined by a pairing Hamiltonian using nuclear magnetic resonance.
Phys. Rev. Lett. , 97:050504, 2006. URL http://link.aps.org/abstract/PRL/v97/e050504.20. Dan Browne. Efficient classical simulation of the semi-classical quantum Fourier transform.
NewJ. Phys. , 9:146, 2007.21. Todd A. Brun and R¨udiger Schack. Realizing the quantum Baker’s map on a NMR quantumcomputer.
Phys. Rev. A , 59(4):2649–2658, April 1999. doi: 10.1103/PhysRevA.59.2649. URLhttp://link.aps.org/abstract/PRA/v59/p2649.ersion November 17, 2010 submitted to
Entropy
36 of 43
22. I. M. Buluta and S. Hasegawa. Designing an ion trap for quantum simulation.
Quantum Informa-tion & Computation , 9(5-6):361–375, MAY 2009.23. Iulia Buluta and Franco Nori. Quantum simulators.
Science , 326(5949):108–111, October 2009.doi: 10.1126/science.1177838. URL http://dx.doi.org/10.1126/science.1177838.24. Tim Byrnes and Yoshihisa Yamamoto. Simulating lattice gauge theories on a quantum computer.
Phys. Rev. A , 73(2):022328, 2006. doi: 10.1103/PhysRevA.73.022328. URL http://link.aps.org/abstract/PRA/v73/e022328.25. Tim Byrnes, Patrik Recher, Na Y. Kim, Shoko Utsunomiya, and Yoshihisa Yamamoto. Quan-tum simulator for the Hubbard model with long-range coulomb interactions using surface acousticwaves.
Phys. Rev. Lett. , 99(1):016405, 2007. doi: 10.1103/PhysRevLett.99.016405. URLhttp://link.aps.org/abstract/PRL/v99/e01640.26. Zhi-Xin Chen, Zheng-Wei Zhou, Xingxiang Zhou, Xiang-Fa Zhou, and Guang-Can Guo. Quantumsimulation of Heisenberg spin chains with next-nearest-neighbor interactions in coupled cavities.
Phys. Rev. A , 81(2):022303, February 2010. ISSN 1050-2947. doi: 10.1103/PhysRevA.81.022303. URL http://dx.doi.org/10.1103/PhysRevA.81.022303.27. Andrew M. Childs. On the relationship between continuous- and discrete-time quantum walk.
Communications in Mathematical Physics , 294:581, 2010. URL doi:10.1007/s00220-009-0930-1.28. Andrew M. Childs and Robin Kothari. Limitations on the simulation of non-sparse hamiltonians.
Quantum Information & Computation , 10:669, 2010. URL http://arxiv.org/abs/0908.4398.29. Andrew M. Childs and Robin Kothari. Simulating sparse Hamiltonians with star decompositions.arXiv:1003.3683, 2010. URL http://arxiv.org/abs/1003.3683.30. Andrew M. Childs, Debbie Leung, Laura Mancinska, and Maris Ozols. Characterization of uni-versal two-qubit Hamiltonians.
Quantum Information and Computation , 11:19–39, January 2011.URL http://arxiv.org/abs/1004.1645.31. Jaeyoon Cho, Dimitris G. Angelakis, and Sougato Bose. Simulation of high-spin Heisenbergmodels in coupled cavities.
Phys. Rev. A , 78(6):062338, Dec 2008. doi: 10.1103/PhysRevA.78.062338. URL http://dx.doi.org/10.1103/PhysRevA.78.062338.32. J. I. Cirac, Paolo Maraner, and Jiannis K. Pachos. Cold atom simulation of interacting relativisticquantum field theories. arxiv:1006.2975, Jun 2010. URL http://arxiv.org/abs/1006.2975.33. Craig R. Clark, Kenneth R. Brown, Tzvetan S. Metodi, and Samuel D. Gasster. Resource require-ments for fault-tolerant quantum simulation: The transverse Ising model ground state.
Phys. Rev.A , 79:062314, Oct 2008. URL http://arxiv.org/abs/0810.5626.34. Robert J. Clark, Tongyan Lin, Kenneth R. Brown, and Isaac L. Chuang. A two-dimensionallattice ion trap for quantum simulation.
Journal of Applied Physics , 105(1):013114, 2009. doi:10.1063/1.3056227. URL http://link.aip.org/link/?JAP/105/013114/1.35. Richard Cleve, Artur Ekert, Chiara Macchiavello, and Michele Mosca. Quantum algorithms revis-ited.
Proc. Roy. Soc. London A , 454:339, 1998.36. X. L. Deng, D. Porras, and J. I. Cirac. Effective spin quantum phases in systems of trappedions.
Phys. Rev. A , 72(6):063407, 2005. doi: 10.1103/PhysRevA.72.063407. URL http://link.aps.org/abstract/PRA/v72/e063407.ersion November 17, 2010 submitted to
Entropy
37 of 43
37. D Deutsch. Quantum-theory, the Church-Turing principle and the universal quantum computer.
Proc. R. Soc. Lond. A , 400(1818):97–117, 1985.38. Simon J. Devitt, Kae Nemoto, and William J. Munro. The idiots guide to quantum error correction,2009. arXiv:0905.2794.39. Jennifer L. Dodd, Michael A. Nielsen, Michael J. Bremner, and Robert T. Thew. Universal quan-tum computation and simulation using any entangling Hamiltonian and local unitaries.
Phys. Rev.A , 65:040301, 2002.40. Jiangfeng Du, Nanyang Xu, Xinhua Peng, Pengfei Wang, Sanfeng Wu, and Dawei Lu. NMRimplementation of a molecular hydrogen quantum simulation with adiabatic state preparation.
Phys. Rev. Lett. , 104(3):030502, Jan 2010. doi: 10.1103/PhysRevLett.104.030502. URLhttp://dx.doi.org/10.1103/PhysRevLett.104.030502.41. Wolfgang D¨ur, Michael J. Bremner, and Hans J. Briegel. Quantum simulation of interactinghigh-dimensional systems:the influence of noise.
Phys. Rev. A , 78:052325, Jun 2008. URLhttp://arxiv.org/abs/0706.0154v1.42. Jacek Dziarmaga and Marek M. Rams. Adiabatic dynamics of an inhomogeneous quantum phasetransition: The case of z > dynamical exponent. New J. Phys. , 12:103002, 2010.43. E. E. Edwards, S. Korenblit, K. Kim, R. Islam, M.-S. Chang, J. K. Freericks, G.-D. Lin, L.-M.Duan, and C. Monroe. Quantum simulation and phase diagram of the transverse field ising modelwith three atomic spins.
Phys. Rev. B , 82:060412(R), 2010.44. Joseph Emerson, Marcus Silva, Osama Moussa, Colm Ryan, Martin Laforest, Jonathan Baugh,David G. Cory, and Raymond Laflamme. Symmetrized characterization of noisy quantum pro-cesses.
Science , 317(5846):1893–1896, 2007. doi: 10.1126/science.1145699.45. Richard P. Feynman. Simulating physics wih computers.
Internat. J. Theoret. Phys.
Nature Physics , 4(10):757–761, July 2008. ISSN 1745-2473. doi:10.1038/nphys1032. URL http://dx.doi.org/10.1038/nphys1032.47. U. Frisch, B. Hasslacher, and Y. Pomeau. Lattice-gas automata for the Navier-Stokes equation.
Phys. Rev. Lett. , 56:1505, 1986.48. L. Gaudreau, S. A. Studenikin, A. S. Sachrajda, P. Zawadzki, A. Kam, J. Lapointe, M. Korkusinski,and P. Hawrylak. Stability diagram of a few-electron triple dot.
Phys. Rev. Lett. , 97(3):036807,Jul 2006. doi: 10.1103/PhysRevLett.97.036807. URL http://dx.doi.org/10.1103/PhysRevLett.97.036807.49. B. Georgeot. Quantum computing of Poincar´e recurrences and periodic orbits.
Phys. Rev. A , 69(3):032301, March 2004. doi: 10.1103/PhysRevA.69.032301. URL http://prola.aps.org/abstract/PRA/v69/i3/e032301.50. B. Georgeot. Complexity of chaos and quantum computation.
Mathematical Structures in Com-puter Science , 17(06):1221–1263, 2007. URL http://journals.cambridge.org/action/displayAbstract?fromPage=online&aid=1444860.ersion November 17, 2010 submitted to
Entropy
38 of 43
51. B. Georgeot and D. L. Shepelyansky. Exponential gain in quantum computing of quantum chaosand localization.
Phys. Rev. Lett. , 86(13):2890, March 2001. doi: 10.1103/PhysRevLett.86.2890.URL http://prola.aps.org/abstract/PRL/v86/i13/p2890 1.52. R. Gerritsma, G. Kirchmair, F. Zahringer, E. Solano, R. Blatt, and C. F. Roos. Quantum simulationof the Dirac equation.
Nature , 463(7277):68–71, January 2010. ISSN 0028-0836. doi: 10.1038/nature08688. URL http://dx.doi.org/10.1038/nature08688.53. R. Gerritsma, B. Lanyon, G. Kirchmair, F. Z¨ahringer, C. Hempel, J. Casanova, J. J. Garc´ıa-Ripoll, E. Solano, R. Blatt, and C. F. Roos. Quantum simulation of the klein paradox, 2010.arXiv:1007.3683.54. Markus Greiner, Olaf Mandel, Tilman Esslinger, Theodor W. Hansch, and Immanuel Bloch. Quan-tum phase transition from a superfluid to a Mott insulator in a gas of ultracold atoms.
Nature , 415(6867):39–44, 2002. doi: 10.1038/415039a. URL http://dx.doi.org/10.1038/415039a.55. Jutho Haegeman, J. Ignacio Cirac, Tobias J. Osborne, Henri Verschelde, and Frank Verstraete.Applying the variational principle to (1+1)-dimensional quantum field theories. arXiv:1006.2409,2010. URL http://arxiv.org/abs/1006.2409v1.56. K. Hammerer, G. Vidal, and J. I Cirac. Characterization of nonlocal gates.
Phys. Rev. A , 66:062321, 2002.57. Henry L. Haselgrove, Michael A. Nielsen, and Tobias J. Osborne. Practicality of time-optimaltwo-qubit Hamiltonian simulation.
Phys. Rev. A , 68:042303, 2003. URL http://prola.aps.org/abstract/PRA/v68/i4/e042303.58. A. F. Ho, M. A. Cazalilla, and T. Giamarchi. Quantum simulation of the Hubbard model: Theattractive route.
Phys. Rev. A , 79(3):033620, Mar 2009. doi: 10.1103/PhysRevA.79.033620.URL http://dx.doi.org/10.1103/PhysRevA.79.033620.59. P. A. Ivanov, S. S. Ivanov, N. V. Vitanov, A. Mering, M. Fleischhauer, and K. Singer. Simulationof a quantum phase transition of polaritons with trapped ions.
Phys. Rev. A , 80(6):060301, Dec2009. doi: 10.1103/PhysRevA.80.060301. URL http://dx.doi.org/10.1103/PhysRevA.80.060301.60. E. Jan´e, G. Vidal, W. D¨ur, P. Zoller, and J. I. Cirac. Simuation of quantum dynamics with quantumoptical systems.
Quantum Information and Computation , 3:15–37, 2003.61. P. R. Johnson, E. Tiesinga, J. V. Porto, and C. J. Williams. Effective three-body interactions ofneutral bosons in optical lattices.
New J.Phys. , 11(9):093022, September 2009. ISSN 1367-2630.doi: 10.1088/1367-2630/11/9/093022. URL http://dx.doi.org/10.1088/1367-2630/11/9/093022.62. J. A. Jones. NMR quantum computation: A critical evaluation.
Fortschr. Phys. , 48(9-11):909–924,2000. doi: 10.1002/1521-3978(200009)48:9/11 h i Proc. Roy. Soc. Lon. Ser. A , 454(1969):323–337, 1998. doi: 10.1098/rspa.1998.0163. URL http://rspa.royalsocietypublishing.org/content/454/1969/323.abstract.64. Ivan Kassal, Stephen P. Jordan, Peter J. Love, Masoud Mohseni, and Al´an Aspuru-Guzik. Polynomial-time quantum algorithm for the simulation of chemical dynamics.
Proceedings of the NationalAcademy of Sciences , 105(48):18681–18686, December 2008. doi: 10.1073/pnas.0808245105.URL http://dx.doi.org/10.1073/pnas.0808245105.ersion November 17, 2010 submitted to
Entropy
39 of 43
65. A. Kay and D. G. Angelakis. Reproducing spin lattice models in strongly coupled atom-cavitysystems.
EPL , 84:20001, October 2008. doi: 10.1209/0295-5075/84/20001. URL http://dx.doi.org/10.1209/0295-5075/84/20001.66. Vivien M. Kendon, Kae Nemoto, and William J. Munro. Quantum analogue computing.
Phil. Trans. Roy. Soc.
Phys. Rev.A , 64(3):032306, 2001. doi: 10.1103/PhysRevA.64.032306. URL http://dx.doi.org/10.1103%2FPhysRevA.64.032306.68. Toshiya Kinoshita, Trevor Wenger, and David S. Weiss. Observation of a one-dimensional Tonks-Girardeau gas.
Science , 305(5687):1125–1128, August 2004. doi: 10.1126/science.1100700.URL http://dx.doi.org/10.1126/science.1100700.69. E. Knill, R. Laflamme, R. Martinez, and C.H. Tseng. An algorithmic benchmark for quantuminformation processing.
Nature , 404:368, 2000.70. Pieter Kok, W. J. Munro, Kae Nemoto, T. C. Ralph, Jonathan P. Dowling, and G. J. Milburn.Linear optical quantum computing with photonic qubits.
Rev. Mod. Phys. , 79(1):135, 2007. doi:10.1103/RevModPhys.79.135. URL http://dx.doi.org/10.1103/RevModPhys.79.135.71. Christina V. Kraus, Michael M. Wolf, and Ignacio J. Cirac. Quantum simulations under transla-tional symmetry.
Phys. Rev. A , 75(2):022303, 2007. doi: 10.1103/PhysRevA.75.022303.72. B. P. Lanyon, J. D. Whitfield, G. G. Gillett, M. E. Goggin, M. P. Almeida, I. Kassal, J. D. Biamonte,M. Mohseni, B. J. Powell, M. Barbieri, A. Aspuru-Guzik, and A. G. White. Towards quantumchemistry on a quantum computer.
Nature Chemistry , 2(2):106–111, January 2010. ISSN 1755-4330. doi: 10.1038/nchem.483. URL http://dx.doi.org/10.1038/nchem.483.73. D. R. Leibrandt, J. Labaziewicz, R. J. Clark, I. L. Chuang, R. Epstein, C. Ospelkaus, J. Wesenberg,J. Bollinger, D. Leibfried, D. Wineland, D. Stick, J. Sterk, C. Monroe, C. S. Pai, Y. Low, R. Frahm,and R. E. Slusher. Demonstration of a scalable, multiplexed ion trap for quantum informationprocessing.
Quantum Information & Computation , 9:0901, Apr 2009. URL http://arxiv.org/abs/0904.2599.74. Debbie Leung. Simulation and reversal of n-qubit Hamiltonians using Hadamard matrices.
Journalof Modern Optics , 49(8):1199–1217, 2002. doi: 10.1080/09500340110109674. URL http://dx.doi.org/10.1080/09500340110109674.75. B. L´evi and B. Georgeot. Quantum computation of a complex system: The kicked Harper model.
Phys. Rev. E , 70:056218, 2004.76. WenZhang Liu, JingFu Zhang, ZhiWei Deng, and GuiLu Long. Simulation of general three-bodyinteractions in a nuclear magnetic resonance ensemble quantum computer.
Science in China SeriesG , 51(8):1089–1096, August 2008. ISSN 1672-1799. doi: 10.1007/s11433-008-0117-3. URLhttp://dx.doi.org/10.1007/s11433-008-0117-3.77. Seth Lloyd. Universal quantum simulators.
Science , 273:1073–1078, 1996.78. X. Ma, B. Dakic, W. Naylor, A. Zeilinger, and P. Walther. Quantum simulation of a frustratedheisenberg spin system, 2010. arXiv:1008.4116.ersion November 17, 2010 submitted to
Entropy
40 of 43
79. Matthew McKague, Michele Mosca, and Nicolas Gisin. Simulating quantum systems using realHilbert spaces.
Phys. Rev. Lett. , 102(2):020505, Jan 2009. doi: 10.1103/PhysRevLett.102.020505.URL http://dx.doi.org/10.1103/PhysRevLett.102.020505.80. D A Meyer. From quantum cellular automata to quantum lattice gases.
J. Stat. Phys. , 85:551–574,1996.81. M. Mohseni and D. A. Lidar. Direct characterization of quantum dynamics.
Phys. Rev. Lett. , 97(17):170501, Oct 2006. doi: 10.1103/PhysRevLett.97.170501.82. Sarah Mostame and Ralf Sch¨utzhold. Quantum simulator for the Ising model with electronsfloating on a helium film.
Phys. Rev. Lett. , 101(22):220501, Nov 2008. doi: 10.1103/PhysRevLett.101.220501.83. C. Negrevergne, R. Somma, G. Ortiz, E. Knill, and R. Laflamme. Liquid-state NMR Simulationsof Quantum Many-Body Problems.
Phys. Rev. A , 71(3):032344, 2005. doi: 10.1103/PhysRevA.71.032344.84. Michael A. Nielsen, Michael J. Bremner, Jennifer L. Dodd, Andrew M. Childs, and Christo-pher M. Dawson. Universal simulation of Hamiltonian dynamics for quantum systems with finite-dimensional state spaces.
Phys. Rev. A , 66:022317, 2002. URL http://prola.aps.org/abstract/PRA/v66/i2/e022317.85. J. L. O’Brien, G. J. Pryde, A. Gilchrist, D. F. V. James, N. K. Langford, T. C. Ralph, and A. G.White. Quantum process tomography of a controlled-not gate.
Phys. Rev. Lett. , 93:080502, 2004.URL doi:10.1103/PhysRevLett.93.080502.86. Jeremy L. O’Brien, Akira Furusawa, and Jelena Vuckovic. Photonic quantum technologies.
NaturePhotonics , 3:687, 2009. URL doi:10.1038/nphoton.2009.229.87. Sangchul Oh. Quantum computational method of finding the ground-state energy and expectationvalues.
Phys. Rev. A , 77(1):012326, 2008. doi: 10.1103/PhysRevA.77.012326. URL http://link.aps.org/abstract/PRA/v77/e012326.88. G Ortiz, J. E. Gubernatis, and R. Laflamme. Quantum algorithms for fermionic simulations.
Phys.Rev. A. , 64:022319, 2001. doi: 10.1103/PhysRevA.64.022319. URL http://link.aps.org/abstract/PRA/v64/e022319.89. Anargyros Papageorgiou and Chi Zhang. On the efficiency of quantum algorithms for hamiltoniansimulation. arXiv:1005.1318, 2010. URL http://arXiv.org/abs/1005.1318.90. Xinhua Peng, Jiangfeng Du, and Dieter Suter. Quantum phase transition of ground-state entangle-ment in a Heisenberg spin chain simulated in an NMR quantum computer.
Phys. Rev. A , 71(1):012307, 2005. doi: 10.1103/PhysRevA.71.012307.91. Xinhua Peng, Jingfu Zhang, Jiangfeng Du, and Dieter Suter. Quantum simulation of a system withcompeting two- and three-body interactions.
Phys. Rev. Lett. , 103(14):140501, Sep 2009. doi:10.1103/PhysRevLett.103.140501. URL http://dx.doi.org/10.1103/PhysRevLett.103.140501.92. Martin Plesch and Caslav Brukner. Efficient quantum state preparation. arXiv:1003.5760, 2010.URL http://arxiv.org/abs/1003.5760.93. D. Porras and J. I. Cirac. Effective quantum spin systems with trapped ions.
Phys. Rev. Lett. , 92(20):207901, 2004. doi: 10.1103/PhysRevLett.92.207901. URL http://link.aps.org/abstract/PRL/v92/e207901.ersion November 17, 2010 submitted to
Entropy
41 of 43
94. D. Porras and J. I. Cirac. Bose-Einstein condensation and strong-correlation behavior of phononsin ion traps.
Phys. Rev. Lett. , 93(26):263602, Dec 2004. doi: 10.1103/PhysRevLett.93.263602.URL http://dx.doi.org/10.1103/PhysRevLett.93.263602.95. David Poulin and Pawel Wocjan. Preparing ground states of quantum many-body systems on aquantum computer.
Phys. Rev. Lett. , 102:130503, 2009. URL doi:10.1103/PhysRevLett.102.130503.96. David Poulin and Pawel Wocjan. Sampling from the thermal quantum gibbs state and evaluatingpartition functions with a quantum computer.
Phys. Rev. Lett. , 103:220502, 2009. URL http://arXiv.org/abs/0905.2199.97. David Poulin, Raymond Laflamme, G. J. Milburn, and Juan P. Paz. Testing integrability with asingle bit of quantum information.
Phys. Rev. A , 68(2):022302, 2003. doi: 10.1103/PhysRevA.68.022302. URL http://link.aps.org/abstract/PRA/v68/e022302.98. David Poulin, Robin Blume-Kohout, Raymond Laflamme, and Harold Ollivier. Exponentialspeedup with a single bit of quantum information: Measuring the average fidelity decay.
Phys.Rev. Lett. , 92(17):177906, April 2004. doi: 10.1103/PhysRevLett.92.177906. URL http://link.aps.org/abstract/PRL/v92/e177906.99. Emily J. Pritchett, Colin Benjamin, Andrei Galiautdinov, Michael R. Geller, Andrew T. Sorn-borger, Phillip C. Stancil, and John M. Martinis. Quantum simulation of molecular collisions withsuperconducting qubits, 2010. arXiv:1008.0701.100. K. De Raedt, K. Michielsen, H. De Raedt, B. Trieu, G. Arnold, M. Richter, Th. Lippert, H. Watan-abe, and N. Ito. Massive parallel quantum computer simulator.
Comp. Phys. Comm. , 176:121–136,2007.101. R¨udiger Schack. Using a quantum computer to investigate quantum chaos.
Phys. Rev. A , 57(3):1634, March 1998. doi: 10.1103/PhysRevA.57.1634. URL http://prola.aps.org/abstract/PRA/v57/i3/p1634 1.102. T. Schaetz, A. Friedenauer, H. Schmitz, L. Petersen, and S. Kahra. Towards (scalable) quan-tum simulations in ion traps.
Journal of Modern Optics
JETPLetters
SIAM Journal on Computing , 26(5):1484–1509, 1997. URL http://link.aip.org/link/?SMJ/26/1484.105. Smirnov, S. Savel’ev, L. G. Mourokh, and Franco Nori. Modelling chemical reactions usingsemiconductor quantum dots.
EPL , 80(6):67008, 2007. doi: 10.1209/0295-5075/80/67008. URLhttp://dx.doi.org/10.1209/0295-5075/80/67008.106. Andrei N. Soklakov and R¨udiger Schack. Efficient state preparation for a register of quantum bits.
Phys. Rev. A , 73(1):012307, Jan 2006. doi: 10.1103/PhysRevA.73.012307.ersion November 17, 2010 submitted to
Entropy
42 of 43
Optics and Spectroscopy , 99(2):211–217, 2005.doi: 10.1134/1.2034606.108. S. Somaroo, C. H. Tseng, T. F. Havel, R. Laflamme, and D. G. Cory. Quantum simulations on aquantum computer.
Phys. Rev. Lett. , 82(26):5381–5383, June 1999. doi: 10.1103/PhysRevLett.82.5381. URL http://dx.doi.org/10.1103%2FPhysRevLett.82.5381.109. R. Somma, G. Ortiz, J. E. Gubernatis, E. Knill, and R. Laflamme. Simulating physical phenomenaby quantum networks.
Phys. Rev. A , 65(4):042323, April 2002. doi: 10.1103/PhysRevA.65.042323. URL http://dx.doi.org/10.1103%2FPhysRevA.65.042323.110. Rolando D. Somma, Gerardo Ortiz, Emanuel H. Knill, and James Gubernatis. Quantum simula-tions of physics problems. In E. Donkor, A.R. Pirich, and H.E. Brandt, editors,
Quantum Informa-tion and Computation, Orlando, Florida, USA , volume 5105, pages 96–103. SPIE, 4 August 20032003. doi: 10.1117/12.487249. URL http://link.aip.org/link/?PSI/5105/96/1.111. S. Succi and R. Benzi. Lattice Boltzmann equation for quantum mechanics.
Physica D , 69:327–332, 1993.112. Masuo Suzuki. Improved Trotter-like formula.
Phys. Lett. A , 180(3):232 – 234, 1993. ISSN0375-9601. doi: DOI:10.1016/0375-9601(93)90701-Z.113. K. Temme, T. J. Osborne, K. G. Vollbrecht, D. Poulin, and F. Verstraete. Quantum Metropolissampling. arXiv:0911.3635, 2009. URL http://arxiv.org/abs/0911.3635.114. Barbara M. Terhal and David P. DiVincenzo. Problem of equilibration and the computationof correlation functions on a quantum computer.
Phys. Rev. A , 61:22301, 2000. URL http://arXiv.org/abs/quant-ph/9810063.115. Sumanta Tewari, V. W. Scarola, T. Senthil, and S. Das Sarma. Emergence of artificial photons inan optical lattice.
Phys. Rev. Lett. , 97:200401, 2006. URL doi:10.1103/PhysRevLett.97.200401.116. H. F. Trotter. On the product of semi-groups of operators.
Proc. Am. Math. Phys. , 10:545, 1959.117. S. Trotzky, L. Pollet, F. Gerbier, U. Schnorrberger, I. Bloch, N. V. Prokof’ev, B. Svistunov, andM. Troyer. Suppression of the critical temperature for superfluidity near the Mott transition:Validating a quantum simulator.
Nature Physics Online , 2010/10/03/online:1745–2481, May 2010.URL http://arxiv.org/abs/0905.4882.118. C. H. Tseng, S. Somaroo, Y. Sharf, E. Knill, R. Laflamme, T. F. Havel, and D. G. Cory. Quantumsimulation of a three-body-interaction Hamiltonian on an NMR quantum computer.
Phys. Rev.A , 61(1):012302, December 1999. doi: 10.1103/PhysRevA.61.012302. URL http://dx.doi.org/10.1103%2FPhysRevA.61.012302.119. M. Van den Nest. Classical Simulation of Quantum Computation, the Gottesman-Knill Theorem,and Slightly Beyond. arXiv:0811.0898, 2008. URL http://arxiv.org/abs/0811.0898.120. M. Van den Nest. Simulating Quantum Computers with Probabilistic Methods. arXiv:0911.1624,Nov 2009. URL http://arxiv.org/abs/0911.1624.121. Juha J. Vartiainen, Mikko Mottonen, and Martti M. Salomaa. Efficient decomposition of quantumgates.
Phys. Rev. Lett. , 92:177902, 2004. URL doi:10.1103/PhysRevLett.92.177902.122. F. Verstraete and J. I. Cirac. Continuous matrix product states for quantum fields.
Phys. Rev. Lett. ,104(19):190405, May 2010. doi: 10.1103/PhysRevLett.104.190405.ersion November 17, 2010 submitted to
Entropy
43 of 43
Phys. Rev. Lett. , 93:227205, 2004.124. G. Vidal and J. I. Cirac. Nonlocal Hamiltonian simulation assisted by local operations and classicalcommunication.
Phys. Rev. A , 66:022315, 2002.125. A. Vidan, R. M. Westervelt, M. Stopa, M. Hanson, and A. C. Gossard. Triple quantum dotcharging rectifier.
Applied Physics Letters , 85(16):3602–3604, 2004. doi: 10.1063/1.1807030.URL http://dx.doi.org/10.1063/1.1807030.126. Hefeng Wang, Sabre Kais, Alan Aspuru-Guzik, and Mark R. Hoffmann. Quantum algorithm forobtaining the energy spectrum of molecular systems.
Phys. Chem. Chem. Phys. , 10(35):5388–5393, September 2008. ISSN 1463-9076. doi: 10.1039/b804804e. URL http://dx.doi.org/10.1039/b804804e.127. Nicholas J. Ward, Ivan Kassal, and Al´an A. Guzik. Preparation of many-body states for quantumsimulation.
The Journal of Chemical Physics , 130(19):194105, May 2009. doi: 10.1063/1.3115177. URL http://dx.doi.org/10.1063/1.3115177.128. Stephen Wiesner. Simulations of many-body quantum systems by a quantum computer. arXiv:quant-ph/9603028v1, 1996.129. D J Wineland. Quantum information processing and quantum control with trapped atomic ions.
Physica Scripta , T137(T137):014007, 2009. URL http://stacks.iop.org/1402-4896/2009/i=T137/a=014007.130. P. Wocjan, D. Janzing, and T. Beth. Simulating arbitrary pair-interactions by a given Hamiltonian:Graph-theoretical bounds on the time-complexity.
Quantum Information & Quantum Computa-tion , 2(2):117, 2002.131. Pawel Wocjan, Martin Roetteler, Dominik Janzing, and Thomas Beth. Universal simulation ofHamiltonians using a finite set of control operations.
Quantum Information & Quantum Computa-tion , 2:133, Sep 2002. URL http://arxiv.org/abs/quant-ph/0109063.132. Pawel Wocjan, Martin R¨otteler, Dominik Janzing, and Thomas Beth. Simulating Hamiltonians inquantum networks: Efficient schemes and complexity bounds.
Phys. Rev. A , 65(4):042309, Mar2002. doi: 10.1103/PhysRevA.65.042309. URL http://dx.doi.org/10.1103/PhysRevA.65.042309.133. L.-A. Wu, M. S. Byrd, and D. A. Lidar. Polynomial-time simulation of pairing models on aquantum computer.
Phys. Rev. Lett. , 89:057904, 2002.134. Christof Zalka. Efficient simulation of quantum systems by quantum computers.
Forschr. Phys. ,46:877–879, 1998.135. Christof Zalka. Simulating quantum systems on a quantum computer.
Proc. R. Soc. Lond. A ,454:313–322, 1998.136. Jingfu Zhang, Gui L. Long, Wei Zhang, Zhiwei Deng, Wenzhang Liu, and Zhiheng Lu. Simulationof Heisenberg XY interactions and realization of a perfect state transfer in spin chains using liquidnuclear magnetic resonance.
Phys. Rev. A , 72(1):012331, Jul 2005. doi: 10.1103/PhysRevA.72.012331. URL http://dx.doi.org/10.1103/PhysRevA.72.012331.c (cid:13)
November 17, 2010 by the author; submitted to