PHYSICAL REVIEW RESEARCH 2, 033347 (2020)
Measurement-induced steering of quantum systems
Sthitadhi Roy ,1,2,* J. T. Chalker ,1,† I. V. Gornyi ,3,4,‡ and Yuval Gefen5,§ 1Rudolf Peierls Centre for Theoretical Physics, Clarendon Laboratory, Oxford University, Parks Road, Oxford OX1 3PU, United Kingdom 2Physical and Theoretical Chemistry, Oxford University, South Parks Road, Oxford OX1 3QZ, United Kingdom 3Institute for Quantum Materials and Technologies, Karlsruhe Institute of Technology, 76021 Karlsruhe, Germany 4A. F. Ioffe Physico-Technical Institute of the Russian Academy of Sciences, 194021 St. Petersburg, Russia 5Department of Condensed Matter Physics, Weizmann Institute of Science, 7610001 Rehovot, Israel
(Received 7 January 2020; revised 6 August 2020; accepted 10 August 2020; published 1 September 2020)
We set out a general protocol for steering the state of a quantum system from an arbitrary initial state toward a chosen target state by coupling it to auxiliary quantum degrees of freedom. The protocol requires multiple repetitions of an elementary step: During each step, the system evolves for a fixed time while coupled to auxiliary degrees of freedom (which we term “detector qubits”) that have been prepared in a specified initial state. The detectors are discarded at the end of the step, or equivalently, their state is determined by a projective measurement with an unbiased average over all outcomes. The steering harnesses backaction of the detector qubits on the system, arising from entanglement generated during the coupled evolution. We establish principles for the design of the system-detector coupling that ensure steering of a desired form. We illustrate our general ideas using both few-body examples (including a pair of spins-1/2 steered to the singlet state) and a many-body example (a spin-1 chain steered to the Affleck-Kennedy-Lieb-Tasaki state). We study the continuous time limit in our approach and discuss similarities to (and differences from) drive-and-dissipation protocols for quantum state engineering. Our protocols are amenable to implementations using present-day technology. Obvious extensions of our analysis include engineering of other many-body phases in one and higher spatial dimensions, adiabatic manipulations of the target states, and the incorporation of active error correction steps.
DOI: 10.1103/PhysRevResearch.2.033347
I. INTRODUCTION rapidly toward zero with increasing system size. With the second approach, the temperature scale required to completely There are many circumstances in which one would like to eliminate thermal excitations from a system also decreases initialize a many-body quantum system in a specified state: toward zero with size. In principle, if the initial state of the Examples of current interest range from quantum information system is precisely known, a further possibility is to act on the processing to studies of nonequilibrium dynamics. Two stan- state with an appropriately chosen perturbation for a precise dard approaches for preparing quantum states are suggested interval of time, so that it evolves into the target state; this, by the laws of quantum mechanics and statistical mechanics. however, requires extreme fine-tuning in a large system. One of these is to make projective measurements of a set of Our aim in the following is to establish a class of protocols observables represented by commuting operators that fully for quantum state preparation that improve on both projective specify the target state. Alternatively, if the target state is measurement and thermal contact with a heat bath. A protocol the ground state of the Hamiltonian for the system, it can of the type we describe will, in an ideal implementation, be reached by putting the system in thermal contact with steer a system from an arbitrary initial state to the target a heat bath that is at a sufficiently low temperature. Both state, with guaranteed success. It does so by coupling the approaches have disadvantages, especially for a system with quantum system of interest to external detector qubits (auxil- a large number of degrees of freedom: In the first approach, iary quantum degrees of freedom) that have been prepared in a general initial state is not an eigenstate of the measurement specified initial states, then evolving the coupled system under operator and hence the measurement outcome is probabilistic. standard unitary quantum dynamics for a fixed time interval, The probability that the target state is reached decreases and finally decoupling and discarding the detector qubits. Multiple repetitions of this process using freshly prepared detector qubits on each occasion, coupled to the system with *[email protected] a suitable interaction Hamiltonian, produce the outcome we †[email protected] ‡ require. In this paper, we illustrate our general approach both [email protected] / §[email protected] for small systems consisting of one or two spin-1 2 degrees of freedom and for a macroscopic many-body system. Published by the American Physical Society under the terms of the This protocol and possible generalizations have the fun- Creative Commons Attribution 4.0 International license. Further damental concepts of quantum entanglement and quantum distribution of this work must maintain attribution to the author(s) measurements as essential ingredients and possess conceptual and the published article’s title, journal citation, and DOI. links to the theory of open quantum systems. First, the effect
2643-1564/2020/2(3)/033347(20) 033347-1 Published by the American Physical Society ROY, CHALKER, GORNYI, AND GEFEN PHYSICAL REVIEW RESEARCH 2, 033347 (2020) of a coupling between the system and detector qubits is to time-continuum limit maps the dynamics of the density matrix generate entanglement between their states. Second, focusing of the system to a Lindblad equation. For part of the analysis on the behavior of the system after discarding the detector in our work, we exploit this formal connection between our qubits, this entanglement induces steering of its quantum steering protocol and Lindblad dynamics to gain insight into state, of a kind first discussed by Schrödinger in the context of the steady state and the rate of approach to it. the Einstein-Podolsky-Rosen (EPR) paradox [1,2]. A physical The emergence of Lindblad dynamics suggests a compar- setting for the decoupling and discarding of the detectors is ison of our protocol with proposals for preparing nontriv- that of performing strong measurements on these qubits and ial many-body states in open quantum systems, where the then taking an unbiased average over the outcomes. Indeed, dissipative environment is posited to be Markovian and is Schrödinger’s original formulation of steering [3]wasin treated within the framework of Lindblad dynamics [17–26]. terms of a sophisticated experimenter performing suitable The challenge in that context is to find suitable Lindblad measurements on one of the parts of a bipartite system to jump operators that can be cast in terms of physical systems drive the other part to a state, chosen by the experimenter, participating in the dynamical process. Our measurement- with nonzero probability. More recently, the scope of the term induced steering protocol has some technical similarities with quantum steering has been narrowed to the impossibility of this framework, as well as conceptual distinctions from it. local hidden-state models describing the ensemble of states With regard to the former, if the detector qubits we discuss that a system can take upon measurement of another system are viewed as an environment, this environment is Markovian entangled with it [4–6]. We, however, continue to use the term by construction, since the detector qubits are prepared afresh in its broader sense. at the start of every cycle. Hence, the map governing the To set out in more detail the links between our protocol evolution of the density matrix of the system has a rep- for preparing or steering a quantum state and discussions of resentation in terms of Kraus operators, which ultimately quantum measurement, it is useful to recall the distinction leads to a Lindblad equation in the time-continuum limit between strong or projective measurements and the notion of [27,28]. weak quantum measurements [7–13]. The former destroy the An obvious advantage of our protocol is that the jump coherent quantum dynamics of the system. By contrast, under operators in the emergent Lindblad equation are uniquely and the latter, auxiliary degrees of freedom are coupled weakly automatically fixed by the details of the system-detector cou- to the system and projective measurements are performed on pling Hamiltonian, thus facilitating “on-demand” engineering these detector qubits after decoupling them from the system. of Lindbladians. On the other hand, since our protocol is No matter how weak the coupling, a measurement of this fundamentally a microscopic measurement protocol, certain kind unavoidably impacts the system through its backaction, requirements can easily be relaxed to go far beyond what is even after the detector qubits are decoupled from the system. describable within the standard Lindblad framework. We note Instead of viewing the measurement-induced disturbance on earlier works in the context of quantum control, discussing the system’s state as a handicap, it may be considered a re- manipulation of quantum states via repeated interactions with source for quantum control and manipulation. The backaction designed environments [23,29,30]; much of this, however, has of measuring the state of detectors entangled with the system focused on the formal aspects of the theory or on applications can be harnessed to control the system’s evolution and steer restricted to few-body systems. its state toward a desired target state: This constitutes the Another advantage of a measurement-based protocol over central message of our work. In principle, one could make coupling with an environment is that, whereas the latter sim- use of the measurement outcomes to determine whether the ply gets entangled with the system, a detector qubit can be state of the system has been successfully steered in a realiza- read out, yielding information on the system state which could tion of the experiment and discard it if this is not the case. In be further used to accelerate convergence to the desired state. the simplest version of such postselected protocols, the proba- Obvious examples include the use of postselected protocols bility of success then goes down exponentially with the length or using the measurement outcomes to implement active of the outcome sequence or the number of detectors. This error correction and enhance the blind measurement protocols suppression is also likely to be exponentially strong in system [31,32]. size for a many-body system. To avoid such suppression, To summarize, our measurement-based steering protocol our aim is to establish protocols within a setting where the differs fundamentally from and has various advantages over outcomes are averaged over in an unbiased fashion, and hence related protocols for manipulating and stabilizing nontrivial there is no loss of probability. We refer to processes of this quantum states. First, the measurements in our case need not type as blind measurements or nonselective measurements. be weak: The detectors may be strongly coupled to the system. The concept of weak measurements allows for the no- Second, in an appropriate time-continuum limit, the dynamics tion of continuous measurements [14–16]. The latter can be is described by a Lindblad equation, but rather than introduc- viewed as a discrete sequence of weak measurements in the ing a Markovian reservoir or jump operators “by hand,” bath limit of vanishing time interval for each weak measurement. coupling is engineered by the measurement operators, which This results in a sequence of measurement outcomes defining fix the jump operators uniquely. Third, unlike unstructured en- a quantum trajectory. The equation for the density matrix of vironments or baths, the detector qubits can be read out, with the system averaged over the trajectories is governed by a the outcomes exploited to further the efficiency and scope of Lindblad equation. In our case, the steering protocol consists our steering on platforms employing postselected protocols or of a discrete sequence of measurements on detector qubits not using active decision-making protocols based on the readout necessarily coupled weakly to the system. A suitably defined sequence. Finally, we note the nontriviality of applying our
033347-2 MEASUREMENT-INDUCED STEERING OF QUANTUM … PHYSICAL REVIEW RESEARCH 2, 033347 (2020) measurement-based steering to many-body systems using an Equations (1) and (2) describe the discrete time-evolution extensive number of auxiliary degrees of freedom. map for the density matrix of the system under the steering dynamics. Structure of the paper (iv) The detector qubits are reprepared in their initial states and the above steps are repeated. The rest of the paper is structured as follows. We start with an overview in Sec. II, where we introduce the basic ingredients of the steering protocol and state the main results B. Steering inequalities of the work. Section III presents the details of the steering As we would like the system to get steered toward the | protocol. The guiding principle for designing the protocol target state, denoted by s⊕ (with corresponding density ρ is introduced and derived in Sec. III A,followedbythe matrix s⊕ ), the dynamics induced by Hs-d should satisfy the derivation of the effective Lindblad dynamics in Sec. III B. steering inequality The workings and advantage of the approach is exemplified |ρ (t + δt )| |ρ (t )| , ∀ t, (3) via the simple case of a spin-1/2 pair steered toward their s⊕ s s⊕ s⊕ s s⊕ singlet state in Sec. III C. We then turn to a many-body with the inequality ideally becoming an equality only if ρ = ρ quantum system, a spin-1 chain, which we steer to the Affleck- s(t ) s⊕ . Kennedy-Lieb-Tasaki (AKLT) ground state [33,34] in Sec. IV. Note that this inequality is very strong. When combined As a building block of the protocol, the steering of a spin-1 with the condition on equality, they ensure that the system pair is discussed in Sec. IV A, followed by the numerical is steered to the target state irrespective of its initial state. treatment of the spin-1 chain in Sec. IV B. We close with In principle, one could envisage weaker steering inequalities. discussions of possible experimental implementations of the One example is protocol and future directions in Sec. V. |ρ | = lim s⊕ s(t ) s⊕ 1(4) t→∞ II. OVERVIEW and a still weaker one is |ρ | > |ρ | . The central result of this work is a general formalism lim s⊕ s(t ) s⊕ s⊕ s(0) s⊕ (5) t→∞ for measurement-induced steering of a many-body quantum system toward a nontrivial target state, by repeatedly coupling We present in Sec. III A a general strategy for designing and decoupling a set of auxiliary degrees of freedom inter- system-detector couplings so that the strongest of these forms, spersed with unitary dynamics of the composite system. The (3), holds. protocol makes use of the entanglement generated between the system and the auxiliary degrees of freedom to steer the C. Guiding principle for choice of system-detector state. The auxiliary degrees of freedom are simple quantum coupling Hamiltonian systems with small Hilbert-space dimensions such that they With the protocol fixed, it remains to find a guiding are easy to prepare in a desired initial state; in this work, principle for the choice of the Hamiltonian that governs the / we consider a set of decoupled detector qubits (spins-1 2) evolution of the system-detector composite. As a first step, initially polarized along a given direction. consider summing both side of (3) over ρs(t ) representing all − δ possible pure states. If (3) is obeyed, then W ≡ e iHs-d t must A. Steering protocol satisfy ρ ⊗ † ⊗ ρ ρ ⊗ ρ . The steering protocol can be described as a sequence of Trs,d[W ( d 1s)W (1d s⊕ )] Trs,d[ d s⊕ ] (6) discrete steering events each of which consists of the follow- ing steps: In the special case where the Hilbert spaces of the system H H (i) The detector qubits are prepared in a initial given state, ( s) and the detector ( d) have the same dimension, we see which does not depend on the state of the system. We denote from this that for the inequality to be as strong as possible, W ρ the initial state of the detector qubits as | and the density should be the swap operator between these two spaces and d d ρ should be the image of s⊕ under this swap. More generally, matrix corresponding to this initial state as ρd. (ii) The system is then coupled to the detector qubits and a Hamiltonian which contains direct products of operators (n) H the composite system evolves unitarily under a Hamiltonian Od in the detectors’ subspace d that rotate the detectors from their initial state to an orthogonal subspace and operators for some time. Denoting the density matrix of the system at (n) U in the system’s subspace Hs that rotate the system to the time t as ρs(t ) and the system-detector Hamiltonian as Hs-d, s the state of the system-detector composite after evolution for target state manifold from an orthogonal subspace will steer an interval δt is given by the system toward the target state. Such a Hamiltonian has the form − δ δ ρ + δ = iHs-d t ρ ⊗ ρ iHs-d t . s-d(t t ) e d s(t )e (1) = (n) | | ⊗ (n) + . ., Hs-d Od d d Us H c (7) (iii) The detector qubits are then decoupled from the n system. Formally, we may trace out the detector degrees of (n) where n labels the detector qubit. Since O connects the state freedom to obtain the density matrix of the system at time d | |O(n)| = t + δt, d to its orthogonal subspace, it satisfies d d d 0. Additionally, we assert the following properties for the system ρs(t + δt ) = Trdρs-d(t + δt ). (2) operators:
033347-3 ROY, CHALKER, GORNYI, AND GEFEN PHYSICAL REVIEW RESEARCH 2, 033347 (2020)
(n) (n) | = (i) Us annihilates the target state, Us s⊕ 0. detector qubit, there is nothing special about this choice; see (n) (n) (n)† | =| (ii) Us is normalized such that Us Us s⊕ s⊕ . Appendix A. (n) (m) (iii) For different n and m,ifUs and Us share spatial support then U (m)U (n)† = 0. If they do not share a spatial s s E. Many-body states and multiple detectors support, they automatically commute. To ensure that equality in Eq. (3) always implies ρs(t ) = In the context of many-body systems, the subspace orthog- ρ | s⊕ it is necessary that s⊕ is coupled to every state in its onal to the target is exponentially large in the system size. (n)† complement by at least one operator Us . In our study of Hence it may appear that the system-detector coupling Hamil- steering to the AKLT state, we observe guaranteed conver- tonian of the form in Eq. (7) entails an exponentially large { (n)†} gence to the target state despite the set of operators Us number of detectors, or nonlocal system-detector coupling not connecting the AKLT state to all the other states in the Hamiltonians, or both. For implementation of the protocol Hilbert space. However, monotonicity is not ensured and, in to be feasible, we would like to have at most an extensive principle, can depend on the particular measure of distance number of detectors, with couplings that are local. Although from the target state. We discuss this issue in detail in Secs. these constraints appear rather restrictive, they can be satisfied III A and IV B and Appendix D. for versions of our protocol that allow steering to a large class of strongly correlated quantum states which are eigenstates D. Toy example with a spin-1/2 of unfrustrated local projector Hamiltonians. One can steer to such states by locally steering different parts of the system to / As a simple example, consider a single spin-1 2 which we the appropriate eigenstate of the corresponding local projector wish to steer to the fully polarized state along the z direction, part of the Hamiltonian. Since the effective Hilbert dimension | =|↑ | so that s⊕ s . The subspace orthogonal to s⊕ has of a local part of the system is finite, one can steer locally † | =|↓ only a single state: Us s⊕ s . Hence, a single detector with a finite number of detectors, and hence the total number qubit is sufficient to steer the state. Without loss of generality, of detectors required scales linearly with system size, and | =|↑ let us choose it to be initialized as d d . Naturally, not exponentially. Moreover, the couplings are manifestly | =|↓ we also have Od d d .FollowingEq.(7), the system- local. detector coupling Hamiltonian can then be written as The particular example we consider in detail is the Affleck- Kennedy-Lieb-Tasaki (AKLT) state of a one-dimensional Hs-d = J(|↓d ↑d|) ⊗ (|↑s ↓s|) + H.c. (8) spin-1 chain [33,34]. The AKLT state is a valence bond state = σ −σ + + σ +σ − , J( d s d s ) such that on each bond between two neighboring spins-1, σ ± there is no projection on the total spin-2 sector. Thus, one where s(d) denote the Pauli raising and lowering operators for the system and detector qubits. can steer the spin-1 chain by locally steering each bond out At time t, the state of the system can be generally written of the total spin-2 sector, and the uniqueness of the ground state of the AKLT chain (with periodic boundary conditions) as ρs(t ) = [I2 + s(t ) · σs]/2, where s(t ) is a classical three- component vector. The state of the system spin-1/2 at time guarantees steering to the AKLT state. Steering each bond t + δt can then be obtained by evolving the combined state requires only a finite number of detectors due to the finite + σ z ⊗ ρ / dimension of the total spin-2 subspace for a pair of spins-1 of the system and the detector, (I2 d ) s(t ) 2, with the unitary operator W , and tracing over the detector. This yields and hence the total number of detectors needed to steer a chain the relation for s(t + δt ) is only extensive in system size. Further examples of states which satisfy the above criteria t δ sx (t + δt ) = cos(Jδt )sx (t ) = cos t (Jδt )sx (0), include matrix product states, projected pair entangled states,
t the Laughlin state of a fractional quantum Hall system [36], + δ = δ = δt δ , sy(t t ) cos(J t )sy(t ) cos (J t )sy(0) the ground state of Kitaev’s toric code [37], and of recent 2 2 interest, fermionic symmetry-protected topologically ordered sz(t + δt ) = 1 − cos (Jδt ) + cos (Jδt )sz(t ) eigenstates of full commuting projector Hamiltonians [38,39]. 2t 2t δ δ = 1 − cos t (Jδt ) + cos t (Jδt )sz(0). (9) An obvious potential concern arises when our protocol is used to steer many-body systems to eigenstates of local The above equation explicitly shows that projector Hamiltonians, because in the cases of interest a lim s(t ) = (0, 0, 1), (10) given quantum degree of freedom appears in more than one t→∞ projector. In particular, in such a scenario it is not guaranteed (n) (m) irrespective of the initial conditions and that the limit is that two system operators Us and Us acting on such a approached exponentially in time. shared degree of freedom, while steering their corresponding This example illustrates how a system-detector coupling parts of the system locally, satisfy the conditions listed at Hamiltonian of the form in Eq. (7) leads to the satisfaction the end of Sec. II C. Steering toward the target space of one of the desired steering inequality, Eq. (3), for generic initial projector may therefore undo, at least partially, the effect states of the system. This guarantees successful steering to the of steering to the target space of a different projector with target state from arbitrary initial states. In turn, the example which the degree of freedom in question is shared. Despite shows how weak measurements can produce control and these complications, we show that our approach enables us to manipulation that projective measurements cannot [35]. select zero-energy eigenstates of local projector Hamiltonians We remark that although in the above example the target as target states and unique stationary states of our dynamics. state of the system qubit is the same as the initial state of the We illustrate this numerically for the AKLT chain, showing
033347-4 MEASUREMENT-INDUCED STEERING OF QUANTUM … PHYSICAL REVIEW RESEARCH 2, 033347 (2020) steering to the ground state with guaranteed success from We next show that the system is steered toward the target arbitrary initial states. state at every discrete steering event. For brevity, we show here the derivation with a single detector qubit and state the results for multiple detector qubits. Details of the derivation F. Steady states and rate of approach for the latter are presented in Appendix B. Equations (1) and (2) show that the time evolution of the To proceed, we find it most convenient to partition the density matrix of the system is governed by a linear map. composite Hilbert space of the detectors and the system Hd ⊗ On general grounds, the largest eigenvalue magnitude for this Hs into two subspaces, which we denote as Dd ⊗ Hs and map is unity. If the associated eigenvector is unique, this Dd ⊗ Hs, where Dd is the one-dimensional subspace spanned ensures that there exists a unique steady state. The logarithm by | d and Dd ⊕ Dd = Hd. In this convention, the density of the next eigenvalue then encodes the rate of approach in matrix ρd ⊗ ρs(t ) can be represented as time to this steady state. Ideally, we would like the rate to D D d d be finite in the thermodynamic limit so that the timescale }D ρ (t ) 0 d required to steer the many-body state arbitrarily close to the ρ ⊗ ρ (t )= s . (12) d s 00 target state does not diverge with system size. }Dd In the context of the AKLT state, we find that the magni- tude of these eigenvalues is most conveniently studied in the Additionally, the system-detector coupling Hamiltonian, time-continuum limit (δt → 0) of the map in Eq. (2) where Eq. (7), takes the form D D an effective Lindblad equation emerges for the equation of d d ρ }D motion of s with the jump operators being fixed by the 0 U † d system-detectors coupling Hamiltonian H . Unit eigenvalue Hs-d = . (13) s-d U 0 }D for the discrete map corresponds to zero eigenvalue for the d Lindbladian operator. If the corresponding Lindbladian eigen- The unitary time-evolution operator generated by the system- vector is unique, it is the stationary state of the steering detector coupling Hamiltonian is protocol. Successful steering implies that the steady state of ∞ − δ 2k the Lindbladian is the target state. Moreover, the gap in the −iH δt ( i t ) e s-d = spectrum of the Lindbladian between the zero eigenvalue (2k)! k=0 and the rest of the eigenvalues determines the rate of the of − δ approach of the system to the target state. In the case of the (U †U )k i t U †(UU†)k × 2k+1 . (14) AKLT state, analysis of the size dependence of the numer- −iδt † k † k + U (U U ) (UU ) ically obtained gap in the effective Lindbladian’s spectrum 2k 1 indicates that it stays finite in the thermodynamic limit. Following the steering map of Eq. (2), applying the above unitary matrix to the density matrix of Eq. (12), and taking the trace of the detector, one obtains the time-evolved density III. FORMALISM OF QUANTUM STATE STEERING matrix of the system as In this section, we describe the quantum steering protocol (−iδt )2k (iδt )2l in detail. We show explicitly that a system-detector coupling ρ (t + δt ) = [(U †U )kρ (t )(U †U )l ] s (2k)!(2l)! s Hamiltonian of the form given in Eq. (7) leads to dynamics k,l that satisfy the steering inequality, Eq. (3), on rather general (−iδt )2k+1(iδt )2l+1 grounds. Additionally, we make the connection to an effective + (2k + 1)!(2l + 1)! Lindblad equation that describes the time-continuum limit of k,l the steering map, Eq. (2). † k † † l × [U (U U ) ρs(t )U (UU ) ]. (15) | = † | =| A. General derivation of steering inequality Using Us s⊕ 0 and UsUs s⊕ s⊕ , the diagonal ma- from discrete time map trix element of the time-evolved density matrix corresponding to the target state can be obtained from Eq. (15)as Let us discuss in some generality the map governing the |ρ + δ | discrete time evolution of the system density matrix, Eq. (2), s⊕ s(t t ) s⊕ with the system-detector coupling Hamiltonian Hs-d of Eq. (7). = |ρ | + | ρ †| 2 δ . ρ = ρ =| | s⊕ s(t ) s⊕ s⊕ Us s(t )Us s⊕ sin ( t ) We will first show that s s⊕ s⊕ s⊕ is indeed a stationary state of the map and then show that the system is Q ρ progressively steered toward the s⊕ after each steering event. (16) That ρ is a stationary state of the map is easily shown by s⊕ Note Q is a diagonal matrix element of a valid density matrix noting that with H from Eq. (7), H (ρ ⊗ ρ ) = 0, which s-d s-d d s⊕ ρ (t ) which ensures that it is non-negative. The change in the naturally implies s diagonal element of the density matrix corresponding to the − δ δ e iHs-d t ρ ⊗ ρ eiHs-d t = ρ ⊗ ρ . (11) target state is proportional to the support of the density matrix d s⊕ d s⊕ | on the subspace orthogonal to s⊕ . Hence, the inequality Upon taking the trace over the detectors, one recovers that in Eq. (3) becomes an equality when ρs(t ) has no remaining − δ δ iHs-d t ρ ⊗ ρ iHs-d t = ρ ρ | Trd[e d s⊕ e ] s⊕ and hence s⊕ is a station- support on the subspace orthogonal to s⊕ —in other words, ary state of the map. ρs(t ) has reached its target form. This concludes the derivation
033347-5 ROY, CHALKER, GORNYI, AND GEFEN PHYSICAL REVIEW RESEARCH 2, 033347 (2020) of the steering inequality, Eq. (3), which in turn shows that the for the density matrix of the form system is steered toward the target state progressively at each ∂t ρs = L[ρs] steering event. In the case of multiple detectors, the equation for the time- 1 =−i[H ,ρ ] + L ρ L† − {L†L ,ρ } , (20) evolved density matrix has the form as in Eq. (16)butQ is s s i s i 2 i i s given by (see Appendix B for derivation) i where H is the intrinsic Hamiltonian of the system and s the L sarethequantum jump operators. They are fixed in Q = |U (n)ρ (t )U (n)†| , (17) i s⊕ s s s s⊕ our derivation by the form of the system-detector coupling n Hamiltonian. As in Sec. III A, we sketch the derivation with a single which is again a sum of diagonal elements of a density matrix detector and state the result for multiple detectors. To proceed and hence is non-negative. Also, it is the total support of the with the derivation, we again partition the composite Hilbert | density matrix on the subspace orthogonal to s⊕ spanned space and represent ρ ⊗ ρ (t )asin(12). The general system- { (n)† | } | d s by the set of states Us s⊕ . Provided s⊕ is connected (n)† detector coupling Hamiltonian can be represented by a matrix to every state in its complement by an operator Us , then of the form = ρ = ρ Q 0 implies s(t ) s⊕ . D D For a many-body system, if a sufficiently large number d d }D of operators is employed that the target state is connected to V U † d H = . (21) all others by the operators U (n)†, monotonic convergence to s-d UV s }Dd the target state is straightforwardly guaranteed. The required number of operators may, however, be exponentially large in Expanding the map in Eq. (2) to second order in δt one obtains system size. By contrast, we are interested in steering a many- ρ (t + δt ) − ρ (t ) 1 s s = i V,ρ t − V, V,ρ t δt body system using only an extensive number of operators, δ [ s( )] [ [ s( )]] each coupled locally to the system. It can then happen that t 2 = ρ 1 Q 0forsome s(t ) at a given step, but that further steps + Uρ (t )U † − {U †U,ρ (t )} δt. successfully steer the system to the target. In Sec. IV B,we s 2 s show how in a such a setting with extensive, local steering (22) operators a spin-1 chain is steered uniquely to the AKLT √ state. In this case, we also examine whether or not steering Taking the limit δt → 0 with U˜ = U δt while requiring is monotonic. We show that this may depend on the choices ||V || ∼ O(1) and ||U˜ || ∼ O(1) [40], one obtains a time- of initial state and of the measure of the distance between the continuum equation of motion time-evolving state and the target state; see Appendix D for ∂ ρ =− ,ρ + ˜ ρ ˜ † − 1 { ˜ † ˜ ,ρ } . t s(t ) i[V s(t )] U s(t )U 2 U U s(t ) details. We will quantify the distance of ρs(t ) from the target ρ (23) state, s⊕ , via the Frobenius and trace norms, denoted as DF and D1 respectively. The are defined as Comparing with Eq. (20), it is trivial to see that the equation ρ of motion for s is indeed of the Lindblad form with the jump operator given by U˜ . The general Hamiltonian in Eq. (21) D (t ) = Tr[ρ (t ) − ρ ]2, (18a) F s s⊕ reduces to the specific one in Eq. (7) with the choice of operators V = V = 0. Hence, the effective Lindblad equation = ρ − ρ 2 / . D1(t ) Tr[ ( s(t ) s⊕ ) ] 2 (18b) which describes the dynamics of the steering protocol in the time-continuum limit consists only of the jump operators. In Additionally, since the AKLT ground state is an unique the case of multiple detectors, the Lindblad equation obtained zero-energy ground state of the AKLT Hamiltonian (with has multiple jump operators periodic boundary conditions), another good measure for the † 1 † distance is the energy of the state measured with respect to the ∂t ρs =−i[V,ρs(t )] + U˜nρs(t )U˜ − {U˜ U˜n,ρs(t )} . n 2 n AKLT Hamiltonian n (24)
EAKLT(t ) = Tr[HAKLTρs(t )]. (19) A few comments regarding the Lindblad equation are in order which will eventually prove useful in the analysis of the concrete examples we discuss. Note that the Lindblad ρ D B. Effective Lindblad dynamics equation is linear in s. Hence, if the Hs -dimensional density matrix ρs(t ) is unravelled as a supervector, ρs(t ), of dimension We now show that the time-continuum limit of the map in 2 DH , the time evolution is simply generated by the superoper- Eq. (2) leads to a Lindblad equation for the dynamics of the s ator corresponding to the Lindbladian density matrix of the system. The exposition of the conceptual ←→ connection between measurement-induced quantum steering ρ (t ) = exp( L t ) · ρ (0). (25) and the dynamics being described by the Lindblad equation, s s often associated with dissipative dynamics, is an important The steady-state manifold of the dynamics is then given by result of this work. The Lindblad equation is a master equation the manifold of states corresponding to zero eigenvalues of
033347-6 MEASUREMENT-INDUCED STEERING OF QUANTUM … PHYSICAL REVIEW RESEARCH 2, 033347 (2020)
←→ L Jt =0.0 Jt =0.1 Jt =1.0 Jt =10.0 . Hence, the uniqueness of the steady state can be deter- 1 mined by studying the degeneracy of the zero eigenvalues. S0 (a1) (a2) (a3) (a4) ←→ T Moreover, the gap in the spectrum of L between the zero + eigenvalue and the rest of the eigenvalues, {v}, defined as T0 T = minv=0{|Re[v]|} determines the nature of the approach − 0 to the steady state. For a finite in the thermodynamic limit, S0 T+ T0 T− S0 T+ T0 T− S0 T+ T0 T− S0 T+ T0 T− the system approaches the steady state exponentially fast in 1.0 (b) 8 (c) 100 (d) time with a rate . −1 0.8 10 6 In some of the examples we study, the Lindblad equation S0|ρs(t)|S0 ) −2
. δt 10 that arises has a special form. Specifically, if there exists a 0 6 T+|ρs t |T+ ( ) ]( choice of basis {|α } such that all the jump operators are of v 4 −3 √ T0|ρs(t)|T0 10 γ |α β| 0.4 Re[ the form αβ , then the matrix elements of the density T |ρ t |T − − s( ) − −4 D 2 10 1 matrix in the same basis follow a master equation given by 0.2 δt distance from singlet DF Δ( ) −5 10 ∼ e−Δt 1 0.0 0 ∂t ρs,αβ = δαβ γανρs,νν − ρs,αβ (γνα + γνβ ), (26) −1 0 1 eigenvalue index ν 2 ν 10 10 Jt 10 0 Jt25 50 which directly implies that the off-diagonal elements of FIG. 1. Steering of a pair of spins-1/2 to the singlet state using the density matrix decay exponentially in time. Moreover, three detector qubits coupled via the Hamiltonian in Eq. (28). (a) The the equations for the diagonal elements do not involve the density matrix of the system, ρs, in the basis spanned by the singlet off-diagonal elements and hence follow a classical master and the three triplets as a color map (for the absolute values of the equation. matrix elements) at different times t. It shows the decaying support on the Stot = 2 diagonal elements. (b) The evolution of the diagonal C. Steering a pair of spins-1/2 elements of ρs(t ) shows the steering to the singlet. (c) The spectrum We next illustrate these ideas using a two-spin problem. of the corresponding Lindbladian with a unique zero eigenvalue Consider a pair of spins-1/2, acted on by the set of Pauli (steady state) and a finite gap . (d) The distance of ρs(t ) from the {σ μ} {σ μ} singlet state |S S | as measured via the matrix norms in Eq. (18) matrices, 1 and 2 , which we would√ like to steer to 0 0 the singlet state |S =(|↑↓ − |↓↑ )/ 2. The Hilbert space decays exponentially in time at a rate is correctly given by . For the 0 = δ = . of two spins-1/2 is four-dimensional with the subspace or- plots, J 1and t 0 1. thogonal to |S0 spanned by the three triplet√ states, |T+ = Results for the evolution of the two-spin system with the |↑↑ , |T− =|↓↓ , and |T =(|↑↓ + |↓↑ )/ 2. Hence, the 0 above steering protocol are shown in Fig. 1. All the elements simplest steering protocol uses three detector qubits, each for of density matrix in the basis spanned by the singlet and the steering the state out of one of the triplet states onto the singlet three triplet states decay to zero exponentially in time except state. At the start of every steering event, each of the detector for the diagonal element corresponding to the singlet, which qubits is prepared in a pure state fully polarized along the approaches unity. This shows that the system is indeed steered positive z axis such that to the singlet state exponentially in time. z z z I2 + σ I2 + σ I2 + σ Turning to the effective Lindblad dynamics corresponding ρ = d1 ⊗ d2 ⊗ d3 , (27) d 2 2 2 to the protocol, the three jump operators of the√ Lindblad μ equation can be read off from Eq. (28)asL = JU δt.The where {σ } denotes the set of Pauli matrices for the ith i i ←→ di spectrum of the so-obtained Lindbladian superoperator, L , detector. For this choice of ρd, the system-detector coupling Hamiltonian motivated from the form in Eq. (7) can be written is shown in Fig. 1(c). The zero eigenvalue is nondegenerate, as implying that the singlet state is the unique steady state of the dynamics. Moreover, the distance of ρ (t ) from the singlet 3 s − state obtained from the exact dynamics and measured via the Hs-d = J (σ ρd ⊗ Ui + H.c.), (28) di Frobenius or the trace norm [Eq. (18)] decays exponentially = i 1 in time with a rate given by × δt, which is in excellent with agreement to the numerically obtained gap in the spectrum of the Lindbladian between the zero eigenvalue and the rest 1 z − − z =| +|= + σ σ − σ + σ , U1 S0 T 3 1 1 2 1 1 2 (29a) of the eigenvalues. This shows that the effective Lindblad 2 2 equation is a valid description of the dynamics. =| |= 1 − σ z σ + − σ + − σ z , Note further that the form of the jump operators in Eq. (29) U2 S0 T− 3 1 1 2 1 1 2 (29b) 2 2 leads the equation of the density matrix elements in the 1 σ z + σ z basis spanned by the singlet and the triplets to be of the U =|S T |= σ +σ − − σ −σ + + 1 2 . (29c) 3 0 0 2 1 2 1 2 2 form in Eq. (26). This explains the exponential decay of the off-diagonal elements. Additionally, the master equation † = = Note from the above equations that UiUj 0fori j for the diagonal elements possesses only the loss term for the as each of the triplet states are orthogonal to each other triplet states and only the gain terms for the singlet, resulting and also to the target singlet state, and hence the conditions the exponential decay and growth of the former and latter asserted for the system operators in Sec. II C are satisfied. respectively.
033347-7 ROY, CHALKER, GORNYI, AND GEFEN PHYSICAL REVIEW RESEARCH 2, 033347 (2020)
Before concluding the section, it is worth mentioning that P ( , +1) tot = where Stot=2 is a projector onto the five-dimensional S a protocol for steering a pair of spins-1/2 to the singlet using 2 subspace for the pair of spins-1 at sites and + 1. The only a single detector qubit can also be envisaged. Such a form of HAKLT in Eq. (32) implies that the ground state has the protocol involves steering the system out of any one of the special property that each bond has zero weight in the Stot = 2 triplet states onto the singlet state while also acting on it with sector. Hence, it is natural to imagine that a spin-1 chain can an intrinsic Hamiltonian which has matrix elements connect- be steered to the AKLT ground state by locally steering each ing the other two triplet states to the first one. Consequently, bond out of the Stot = 2 sector. Moreover, since the Stot = 2 the Hamiltonian unitarily rotates the states within the triplet sector of each bond is finite dimensional, such a steering sector and weight from the triplet sector keeps leaking into protocol would satisfy our physically motivated constraints: the singlet, eventually leading to the pure singlet state as the locality and only an extensive number of detectors. steady state. For details, see Appendix C. In Sec. IV A, we discuss the basic building block of the steering protocol, namely steering a pair of spins-1 out of the IV. SPIN-1 CHAIN Stot = 2 sector. In Sec. IV B, we use this building block to steer a spin-1 chain to the AKLT state. Let us now discuss the steering of a quantum many-body system, namely that of the spin-1 chain to the AKLT ground A. Pair of spins-1 state. The AKLT Hamiltonian is N Consider steering on the bond between a pair of spins-1 tot 1 1 1 2 labeled as and + 1. The S , + = 2 subspace is five HAKLT = + Sˆ · Sˆ +1 + (Sˆ · Sˆ +1) , (30) ( 1) = 3 2 6 dimensional and hence the simplest protocol involves five 1 tot = detector qubits, one to steer out of each of the S( , +1) 2 where Sˆ denotes spin-1 operators at site and the site = tot = states. Note that we only wish to steer out of the S( , +1) 2 N + 1 is identified with = 1 to impose periodic boundary subspace and there are no restrictions on what precise state(s) conditions. tot = tot = in the S( , +1) 1 and S( , +1) 0 subspaces we steer onto. It is useful to set up the notation for the rest of the section This offers a lot of freedom in choosing the steering protocol. z here. The projectors onto the three eigenstates of Sˆ (with This could be exploited to construct the simplest system- ± ± 0 eigenvalues 1 and 0 respectively), P and P ,aregivenby detector coupling Hamiltonian, or from a practical point of view, the one that it is easiest to implement. In the following, ± = z 2 ± z / 0 = − + − −. P S S 2; P I3 P P (31) we choose a particular protocol which steers the state onto the tot = tot,z The total spin on a bond between the sites and + 1 will S( , +1) 1 subspace, keeping the sign of S( , +1) the same. tot = + At the start of each steering event, the detector qubits [41] be denoted by S( , +1) S S +1. In what follows, it will be often convenient to use the simultaneous eigenstates of are prepared in a polarized pure state ˆ tot 2 ˆtot,z | tot, tot,z (S( , +1)) and (S( , +1)), denoted as S S ( , +1), where 5 + σ z I2 ( , +1) tot tot,z , + d S takes values 0, 1, and 2, and S takes all integer values ρ( 1) =⊗ i , (33) − tot tot,z tot d 2 in the range S S S . Additionally, we denote the i=1 AKLT ground state as |AKLT and the corresponding density ρ =| | matrix as AKLT AKLT AKLT . and the system-detector coupling Hamiltonian is of the form The AKLT Hamiltonian with periodic boundaries has a unique valence-bond ground state which is most easily un- 5 ( , +1) − ( , +1) ( , +1) derstood by noting that HAKLT can be expressed as a sum of = σ ρ ⊗ + . ., Hs-d J ( , +1) d Ui H c (34) di local projectors i=1 N = P ( , +1), HAKLT Stot=2 (32) where =1 ( , +1) = | , , | = 1 − − − + + , U1 ( 1 1 2 2 )( , +1) [(S S +1)P P +1] (35a) 2 − + + − ( , +1) 1 S S +1 + 0 S S +1 0 + U = (|1, 1 2, 1|) , + = I − P P + − I − P P , (35b) 2 ( 1) 2 9 2 1 9 2 +1 − + + − 1 (S S )2 (S S )2 ( , +1) = | , , | = √ − +1 + − − − +1 − + + + − − − + 0 0 , U ( 1 0 2 0 )( , +1) I9 P P + I9 P P + (S S + S S + )P P + 3 2 3 4 1 4 1 1 1 1 (35c) + − − + ( , +1) 1 S S +1 − 0 S S +1 0 − U = (|1, −1 2, −1|) , + = I − P P + − I − P P , (35d) 4 ( 1) 2 9 2 1 9 2 +1
( , +1) 1 + + − − U = (|1, −1 2, −2|) , + = [(S − S )P P ]. (35e) 5 ( 1) 2 +1 +1
033347-8 MEASUREMENT-INDUCED STEERING OF QUANTUM … PHYSICAL REVIEW RESEARCH 2, 033347 (2020)
Jt =0.1 Jt =1.0 Jt =10.0 Jt = 100.0 An important point to notice is for a pair of spins-1, the 1 (a1) (a2) (a3) (a4) above dynamics does not have a unique steady state unlike tot S =2 the case exemplified in Sec. III C. This is due to the fact that in the case of the spins-1, the steering is not onto a Stot =1 particular pure state but rather onto a subspace of states, and Stot =0 0 the specific steady state of the dynamics depends on the initial =1 =2 =0 =1 =2 =0 =1 =2 =0 =0 =1 =2 condition of the system. The nonuniqueness of the steady state tot tot tot tot tot tot tot tot tot tot tot tot S S S S S S S S S S S S for the pair of spins-1 is also manifested in the corresponding
(b) ,z |2, 2 |2, 0 |2, −2 |1, 0 |0, 0 Lindbladian having a degenerate zero-eigenvalue manifold. tot |2, 1 |2, −1 |1, 1 |1, −1 ,S 0.4 In the case of a spin-1 chain, as we will discuss in the next tot
S subsection, the uniqueness of the AKLT ground state leads | s ρ | to the uniqueness of the steady state reached under dynamics ,z 0.2 tot obtained by extending the approach described above to all the ,S
tot bonds of the chain. S