PaperlayerView exact revision on arXiv

Source-linked AI summary

Quantum Markovian Subsystems: Invariance, Attractivity, and Control

Francesco Ticozzi, Lorenza Viola

arXiv:0705.1372v2quant-ph

TL;DR

The paper asks how quantum information encoded in subsystems behaves under continuous-time Markovian noise. It provides linear-algebraic and Lyapunov-based characterizations of invariant, noiseless, and attractive subsystems, then applies them to feedback stabilization and noiseless-subspace generation. The results include explicit finite-dimensional stabilization and synthesis conditions, while feedback performance depends on detection and model assumptions.

  • Problem

    The paper addresses how to characterize and control subsystem behavior in continuous-time Markovian quantum systems under model and initialization uncertainties.

  • Method

    The paper combines linear-algebraic subsystem characterizations with Lyapunov stability techniques and output-feedback Markovian control strategies.

  • Results

    The paper obtains explicit characterizations and synthesis results for invariant, noiseless, attractive, and stabilizing subsystems in finite-dimensional Markovian systems.

  • Takeaways & Limitations

    Markovian output-feedback control offers a simpler alternative to Bayesian feedback while supporting pure-state stabilization and noiseless-subspace generation under stated conditions.

Abstract

from arXiv · show

We characterize the dynamical behavior of continuous-time, Markovian quantum systems with respect to a subsystem of interest. Markovian dynamics describes a wide class of open quantum systems of relevance to quantum information processing, subsystem encodings offering a general pathway to faithfully represent quantum information. We provide explicit linear-algebraic characterizations of the notion of invariant and noiseless subsystem for Markovian master equations, under different robustness assumptions for model-parameter and initial-state variations. The stronger concept of an attractive quantum subsystem is introduced, and sufficient existence conditions are identified based on Lyapunov's stability techniques. As a main control application, we address the potential of output-feedback Markovian control strategies for quantum pure state-stabilization and noiseless-subspace generation. In particular, explicit results for the synthesis of stabilizing semigroups and noiseless subspaces in finite-dimensional Markovian systems are obtained.

I. INTRODUCTION AND PRELIMINARIES

The paper frames quantum subsystems as physically meaningful carriers of information in open quantum systems and develops the Markovian dynamical framework used to analyze them.

  • Quantum subsystems support physical representations of quantum information while reducing errors and decoherence caused by environmental interactions.
  • Open-system evolution is modeled by trace-preserving completely positive maps, with Markovian dynamics forming a continuous-time quantum dynamical semigroup.
  • Complete positivity preserves positivity for arbitrary system purifications, including entangled states.
  • The framework extends discrete-time quantum operations to continuous-time dynamics through a forward composition law.
  • The semigroup generator has a Hamiltonian component and dissipative Lindblad terms describing non-unitary environmental effects.

D. Phenomenological Markovian models and robustness

The paper treats Markovian generators phenomenologically when microscopic derivation is impractical and introduces robustness notions for uncertainty in noise parameters.

  • Markovian generators are often assumed phenomenologically because complete microscopic system-environment knowledge is typically unavailable.
  • A two-level atom with decay rate γ has excited-state survival probability Pe(t) = e−γt under a single lowering-noise channel.
  • Physical constraints can reduce the relevant error generators and GKS matrix to a compact representation.
  • Robustness analysis tests whether invariance or attractive states can be guaranteed without fine-tuning noise parameters.
  • A-robustness requires a property for every admissible GKS matrix, whereas γ-robustness requires it only across admissible spectra.

II. THEORY

The theory formalizes subsystem encoding and derives necessary and sufficient Markovian invariance conditions, including stronger constraints under model-parameter robustness.

  • Quantum information is encoded in a subsystem when noise acts as identity on its logical factor while remaining confined to the designated block.
  • A subsystem is invariant when evolution preserves the prescribed initialization structure and induces quantum dynamical semigroups on its factors.
  • Theorem 2 characterizes Markovian invariance through conditions ensuring both confinement to the subsystem block and autonomous reduced evolution.
  • The block characterization requires vanishing leakage terms and, for each noise operator, identity action on either the subsystem or co-subsystem factor.
  • Robust invariance imposes independent Hamiltonian and dissipative constraints, preventing invariance from relying on fine-tuned cancellations.

C. Noiseless subsystems

Noiseless subsystems preserve encoded information through unitary evolution even when the surrounding state may evolve nontrivially.

  • Reduced-state initialization permits entangled subsystem states while requiring the relevant projected blocks to define the encoded state.
  • A noiseless subsystem is one whose reduced state evolves unitarily, independently of the system’s initial state outside the subsystem.
  • Perfectly initialized noiseless subsystems form a special case of invariant subsystems.
  • The noiseless-subsystem characterization separates unitary logical evolution from arbitrary co-subsystem dynamics.

T NSF

The paper gives explicit conditions for Markovian noiseless subsystems and their robustness to model-parameter variations. These results connect noiselessness to invariant-subsystem conditions and specialize them to decoherence-free subspaces.

  • Relation to invariance: Noiselessness is a specialization of invariance: with suitable initialization, an invariant subsystem whose subsystem factor evolves unitarily supports a noiseless subsystem.For noiseless subsystems, invariance is robust to initialization within the complementary factor.
  • Markovian noiseless subsystems: A Markovian noiseless subsystem is characterized by unitary evolution on its subsystem factor while allowing general Markovian dynamics on the complementary factor.The NSF block is driven by unitary dynamics on the noiseless factor and independent Markovian dynamics on the complementary factor.
  • Robust noiseless subsystems: A-robust and γ-robust noiseless subsystems require distinct block conditions for uncertainty in the reduced noise matrix and in the Lindblad coefficients, respectively.The conditions constrain the Hamiltonian and error-generator blocks, including vanishing coupling blocks and factorized noise actions.
  • Robust noiseless subsystems: A-robust noiseless subsystems can exist only when the allowed error generators do not span the entire operator algebra B(HI).Thus, robustness is restricted to a specified set of possible noise generators rather than arbitrary noise.
  • Decoherence-free subspaces: A γ-robust decoherence-free subspace is characterized by specialized generator conditions and equivalently by joint right-eigenvector conditions for each Lindblad operator and jump operator L†_kL_k.The decoherence-free-subspace results are obtained by taking the complementary factor to be one-dimensional.

D. Imperfect initialization

The paper introduces ρ-robustness to remove dependence on the initial state of the full system. For noiseless subsystems, this robustness imposes stronger decoupling conditions on the noise and Hamiltonian blocks.

  • Definition: ρ-robustness requires the desired subsystem property for every initial full-system state sharing the same reduced subsystem state.This extends robustness beyond initialization within the intended subsystem-complementary block.
  • Imperfectly initialized noiseless subsystems: An initialization-free Markovian noiseless subsystem requires the NSF block to be dynamically decoupled from the remaining blocks.The conditions include L_P,k = 0, L_Q,k = 0, H_P = 0, and L_NSF,k = I ⊗ M_k.
  • Trade-off: The stronger initialization robustness imposes tighter noise constraints, reducing the room for Hamiltonian compensation of the noise action.The paper notes that these conditions may be demanding to enforce in applications.

E. Attractive subsystems

The paper defines attractive subsystems as those that asymptotically recover the desired subsystem behavior from arbitrary initial states. It develops Lyapunov-based sufficient conditions and identifies incompatibilities and dimensional constraints.

  • Definition: An attractive subsystem asymptotically satisfies the target subsystem behavior for every initial state under the given family of trace-preserving completely positive maps.Attractivity can be viewed as asymptotic self-initialization that reabsorbs initialization errors.
  • Incompatibility with initialization-free noiselessness: A noiseless subsystem with a nontrivial remainder cannot be attractive when the coupling blocks satisfy L_P,k = L†_Q,k = 0 for every k.Under these conditions, the remainder-state trace is preserved, preventing convergence from arbitrary initial states.
  • Sufficient conditions: For a pure-factor invariant subsystem, a unique attractive state of the complementary generator is sufficient for subsystem attractivity.This is the relaxing-complement condition used in the general noiseless-subsystem construction.
  • Lyapunov conditions: For invariant subspaces, a continuously differentiable Lyapunov functional with nonincreasing derivative and W ∩ Z contained in the initialized set guarantees attractivity via Krasowskii-LaSalle invariance.The construction combines decay of the Lyapunov function with exclusion of non-initialized invariant limit states.
  • Scope condition: The sufficient attractivity condition for an invariant subspace can hold only when dim(HS) ≥ dim(HR).This dimensional restriction follows from rank considerations on the relevant coupling block.
  • General attractive subsystems: The general attractive-subsystem result combines Lyapunov convergence outside the target block with a relaxing complementary generator and the condition W ∩ Z ⊆ I_SF(HI).The complementary dynamics must have a unique attractive state, while the Lyapunov argument drives support into the subsystem-complementary block.

A. Quantum trajectories and Markovian output feedback

The paper models continuous quantum measurement through stochastic master equations and obtains deterministic Markovian dynamics by averaging over the measurement noise. It then uses instantaneous Hamiltonian feedback to formulate Markovian feedback master equations for stabilization and noiseless-subspace synthesis.

  • Quantum trajectories: Continuous monitoring produces a stochastic master equation whose conditional state follows trajectories driven by the measurement record and Wiener noise.The measurement operator determines the system-probe interaction, while η quantifies detection efficiency.
  • Unconditional dynamics: Dropping the martingale term from the stochastic master equation yields the deterministic unconditional quantum dynamical semigroup generator.The diffusion term represents the innovation component of the quantum filtering dynamics.
  • Markovian feedback: The paper assumes perfect detection by default and interprets instantaneous measurement-record feedback through an equivalent Itô formulation to preserve Markovian evolution.The feedback operation is composed with the measurement action and converted from an implicit Stratonovich description.
  • Control applications: The resulting Wiseman-Milburn feedback master equation is the controlled Markovian model used for state stabilization and noiseless-subspace synthesis.The paper explicitly targets these control problems for dynamics described by feedback master equations.

B. Control assumptions

The paper examines Markovian feedback design through specified measurement and feedback operators, while also considering complete Hamiltonian control and constant compensation. These assumptions enable noise-operator transformations and stabilization of states not invariant under uncontrolled dynamics.

  • B. Control assumptions: The standard feedback design specifies measurement and feedback operators, treating measurement strength and feedback gain as control parameters.
  • B. Control assumptions: Complete Hamiltonian control permits arbitrary feedback Hamiltonians and arbitrary constant Hamiltonian perturbations to the free Hamiltonian.The assumption is explicitly identified as physically demanding because weak-coupling derivations may constrain allowed Hamiltonian contributions.
  • B. Control assumptions: Transforming Lindblad operators can vary their trace and, under complete Hamiltonian control, compensate the induced Hamiltonian correction.For Hermitian Lindblad operators and real transformation parameters, the correction Hamiltonian vanishes.
  • B. Control assumptions: Constant Hamiltonian compensation can stabilize a state that is not invariant under uncontrolled dynamics without directly modifying the non-unitary part.In the example, Hc = σy makes ρd = diag(1, 0) invariant and attractive.

C. Pure-state preparation with Markovian feedback: Two-level systems

For two-level systems, the paper derives conditions under which a desired pure state is invariant and globally attractive, then uses feedback and Hamiltonian compensation to synthesize stabilization. Under complete Hamiltonian control, arbitrary desired pure states can be stabilized.

  • C. Pure-state preparation with Markovian feedback: Two-level systems: The paper’s approach first identifies Lindblad-equation constraints for a pure state to be an attractive equilibrium, rather than directly selecting measurement and feedback strengths.
  • C. Pure-state preparation with Markovian feedback: Two-level systems: A pure state ρd = diag(1, 0) is globally attractive and invariant exactly when the stated generator conditions hold and at least one coupling coefficient l_k,P is nonzero.The nonzero coupling condition excludes additional stationary diagonal states and supplies attraction.
  • C. Pure-state preparation with Markovian feedback: Two-level systems: Pure-state stabilization is stronger than unconditional stability because a pure target cannot be the convex average of distinct states, so each trajectory must converge to it almost surely.
  • C. Pure-state preparation with Markovian feedback: Two-level systems: Under complete Hamiltonian control, a feedback Hamiltonian and Hamiltonian compensation can stabilize an arbitrary desired pure state.The construction selects feedback so the measurement operator’s Hermitian part is not diagonal in the target basis.
  • C. Pure-state preparation with Markovian feedback: Two-level systems: The constructive analysis recovers that equatorial pure states are not stabilizable under the earlier control assumptions when they commute with the Hermitian part of M = σ+.

D. Extension to multi-level systems

The two-level stabilization results extend to finite-dimensional systems through structured measurement and feedback operators and a Lyapunov function. The resulting dynamics globally attracts the target pure state, while practical implementations require strong control and detection assumptions.

  • D. Extension to multi-level systems: For d-level systems, the target pure state is globally attractive and invariant under the specified measurement and Markovian feedback Hamiltonian structure.The construction uses nonzero coefficients m_i for i = 1, ..., d − 1.
  • D. Extension to multi-level systems: A Lyapunov function V_d(ρ) = tr(Ĥρ) proves global attraction of the target state in the multi-level construction.Its derivative is non-positive and vanishes only at ρ = ρd.
  • D. Extension to multi-level systems: The multi-level construction generalizes the two-level design, with the Lyapunov Hamiltonian using equally spaced renormalized energy gaps.The feedback and measurement operators play roles analogous to σy and σx.
  • D. Extension to multi-level systems: Initial-state estimation errors do not prevent the desired convergence feature for each initial state.
  • D. Extension to multi-level systems: The feedback strategy avoids a real-time state-estimation stage, unlike Bayesian feedback, which becomes rapidly prohibitive as target-system dimensionality grows.
  • D. Extension to multi-level systems: Markovian output feedback requires strong control capabilities and perfect detection, including infinite-bandwidth real-time feedback and accurate tuning of system and control parameters.

F. DFS synthesis with Markovian feedback

The paper shows that Markovian feedback can construct decoherence-free subspaces under a controllability assumption, including an explicit dimension guarantee and a constructive synthesis procedure. A coupled-qubit example demonstrates that feedback can make a noiseless subspace attractive, unlike noiseless subspaces already present in the uncontrolled noise operator.

  • Control constraint: The feedback loop can modify only the skew-Hermitian part of the measurement operator, constraining which non-unitary generators can be synthesized.This limitation motivates the question of whether decoherence-free or noiseless subspaces can nevertheless be generated by closed-loop control.
  • Theorem 5: A decoherence-free subspace of dimension at least p can be generated by Markovian feedback for every measurement operator M under the CHC assumption.Here p=d/2 for even d and p=(d+1)/2 for odd d.
  • Constructive synthesis: The synthesis constructs p vectors from paired eigenvectors of the Hermitian part of M, then chooses a feedback Hamiltonian F to eliminate the Q block.The paired vectors share the same restricted eigenvalue, and the feedback Hamiltonian makes the resulting subspace decoherence-free.
  • Limitations: The construction depends on both the CHC assumption and perfect monitoring, which are demanding experimentally and motivate weaker-condition analyses.The authors describe the proof as constructive but identify these assumptions as practical limitations.
  • Coupled-qubit example: The same noise operator has noiseless subspaces that are not attractive, so feedback-generated noiseless structure can offer an advantage over existing ones.The example contrasts the feedback-generated attractive subspace with noiseless subspaces already admitted by M.
Loading 0705.1372v2…