{"id":"0faeecab-8be5-46ce-8e08-ac601df60621","arxiv_id":"1908.09836","paper_version":5,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"dVQE variationally computes non-equilibrium steady states of open quantum systems by minimizing the squared Liouvillian over a doubled-qubit ansatz.","lead":"The paper introduces dVQE, a hybrid quantum-classical algorithm that finds the steady state of an open quantum system by minimizing a cost function on a quantum circuit. It is a step toward simulating dissipative many-body systems on near-term quantum hardware.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The cost-to-observable guarantee is the weak link: footnote 56 has the inequality reversed, and the correct bound depends on the Liouvillian gap δ, so small cost need not imply accurate observables when δ is small.","rationale":"The reader's weakest assumption correctly identifies the unproven cost-to-observable link as the load-bearing issue. My stress-test agrees with that diagnosis and sharpens it: the only bound offered in the paper, footnote 56, is reversed, and the correct bound depends on the Liouvillian spectral gap δ, which can be very small. This is not an ad hominem or a disagreement with consensus; it is a correctness risk in the central promise. The core mathematical mapping via the Choi-Jamiolkowski isomorphism and the demonstrations on small systems are genuine supporting evidence, and the paper honestly labels the proof as open. However, because the algorithm's utility depends on a guarantee that is absent and, in the corrected form, requires a gap-dependent stopping criterion or an estimate of δ, the conditional verdict remains appropriate. I see no reason to move the verdict further, but the authors should correct footnote 56 and either prove the δ-dependent bound or state clearly that the method is reliable only when δ is known to be sufficiently large. The concrete test with scaled dissipation rates would provide quantitative evidence on whether the concern actually occurs in practice.","tokens_in":55466,"tokens_out":18551,"duration_ms":222242,"concrete_test":"Run the noiseless dVQE of Sec. IV on a fixed dissipative Ising chain (N=2 or 4) with all dissipation rates multiplied by η ∈ {1, 10⁻¹, 10⁻², 10⁻³}, so δ scales roughly with η. For each η, stop the optimization when ⟨L†L⟩ reaches a fixed threshold ε=10⁻⁶ and record the vector infidelity 1−f², the maximum error of local magnetizations relative to exact diagonalization, and min(λq). If the observable error grows as √ε/δ or if any optimized λq is negative, the concern lands: small cost does not by itself guarantee accurate observables. If errors stay negligible across all η and λq remain nonnegative, the practical premise is supported in this model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that minimizing ⟨L†L⟩ yields a state whose observables are close to those of the NESS. In Sec. III A this is asserted through an unspecified function f(ε), with the proof left as an open problem. The only quantitative support, footnote 56, states that for overlap f and δ the lowest nonzero eigenvalue of L†L, one has 1−f² ≥ ⟨L†L⟩/δ. The inequality direction is wrong: the correct identity is C ≥ δ(1−f²), equivalently 1−f² ≤ C/δ. As printed, the bound gives no upper bound on the state error at all. Even after correcting the direction, the bound controls only the vector-space overlap, not the trace-norm error of physical observables. A standard pseudoinverse argument gives |Tr[A(ρ−ρSS)]| ≤ ||A||√d ||ρ||F δ⁻¹√C for physical states, so the promised f(ε) must depend on δ. When δ is small (weak dissipation, near-critical or slow-relaxing regimes), a numerically small C can still correspond to O(1) observable errors. The problem is compounded by the positivity issue: the entangled ansatz is not explicitly restricted to nonnegative λq, and the observable sampling in Sec. III C presumes λq ≥ 0. The optimizer can therefore explore non-physical near-null directions, making the link between low cost and accurate observables even less secure. Thus the utility claim rests on an unproven, model-dependent gap condition.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes dVQE, a variational quantum-classical algorithm for the non-equilibrium steady state (NESS) of a Markovian open quantum system. The authors map the density matrix to a pure state on a doubled Hilbert space through the Choi-Jamiolkowski isomorphism, formulate the NESS condition as L†L|ρSS⟩ = 0, and use C(θ) = ⟨0|U†(θ)L†LU(θ)|0⟩ as a lower-bounded cost function. The ansatz U(θ) = [V(θv)⊗V*(θv)] CNOT D(θd) is designed to produce Hermitian operators, and observables are evaluated by sampling the diagonal distribution λ_q with an N-qubit circuit. Demonstrations include the dissipative Ising model on a Rigetti device (N = 1), noisy numerical simulations of N = 2 for the Ising and persistent-current models, and noiseless simulation for N = 8, with infidelities of order 10^-2. The authors state that a rigorous error bound between a small cost and accurate observables is an open problem.","tokens_in":55771,"tokens_out":7821,"duration_ms":82652,"significance":"The work extends VQE methodology to stationary states of open systems, an experimentally relevant and theoretically nontrivial task; the cost function is constructed directly from the Liouvillian without fitted constants, so the optimization is not circular. The inclusion of a real-device demonstration and a noiseless N = 8 simulation is a strength, as is the explicit discussion of the vector/matrix normalization difference. The main limitation is the unproven and, as written, incorrectly stated relation between cost and observable error, which affects the reliability of the algorithm's predictions in the regimes the paper targets.","major_comments":[{"comment":"The bound stated in footnote 56, 1−f² ≥ ⟨L†L⟩/δ, has the inequality reversed. Decomposing the normalized ansatz as |ψ⟩ = f|ρSS⟩ + √(1−f²)|φ⟩ with ⟨φ|ρSS⟩ = 0 gives ⟨L†L⟩ = (1−f²)⟨φ|L†L|φ⟩ ≥ δ(1−f²), hence 1−f² ≤ ⟨L†L⟩/δ. As printed, the footnote provides no upper bound on the state error, and even after correction the bound depends on δ, the smallest nonzero eigenvalue of L†L. Consequently the f(ε) assertion in the main text remains unsupported when δ is small; the authors should correct the footnote and either prove a δ-dependent bound or qualify the claim accordingly.","section":"III A (footnote 56)"},{"comment":"Equation (20), ⟨O⟩ = ∑_q λ_q ⟨q|V†OV|q⟩, omits the trace renormalization required by the vector representation. In Eq. (4) the vector |ρ⟩ carries coefficients ρ_ij/C with C² = ∑|ρ_ij|², so the operator reconstructed from the ansatz is ρ ∝ V D V† with D = diag(λ_q), and physical expectation values require division by Tr ρ = ∑_q λ_q (assuming real λ_q). The ansatz of Eq. (18) only fixes the L2 norm ∑|λ_q|² = 1, not the trace. The authors should specify the normalization used in the simulations and amend Eq. (20) (or state explicitly that λ_q have been renormalized).","section":"III C, Eq. (20)"},{"comment":"The entangled-type circuit for D(θd) in Fig. 2(a) can produce negative real amplitudes (or complex ones) for λ_q, so the state |ρθ⟩ is not guaranteed to correspond to a positive semidefinite operator; the text's statement that condition (II) is satisfied by restricting 'θv' appears to be a typo for θd, and no explicit restriction or penalty is given. Because Sec. III C uses λ_q as a probability distribution for sampling q, the optimizer may explore nonphysical states with negative 'probabilities' and reach low cost while remaining far from the NESS. The authors should impose or verify PSD conditions in the ansatz (e.g., angle ranges or a positivity penalty) and report whether the optimized states in the demonstrations satisfy them.","section":"III B and III C"}],"minor_comments":[{"comment":"In Sec. III B, the sentence 'condition (II) can be satisfied by imposing appropriate restriction on the parameters θv' should refer to θd; positivity of ρ = V D V† depends on D, not on the basis V.","section":"III B"},{"comment":"Appendix A compares the vector 2-norm and trace distance only for diagonal density matrices with diagonal perturbations; this is a narrow class and does not directly probe the cost-to-observable relation discussed in Sec. III A. A sentence acknowledging this limitation would avoid overgeneralization.","section":"Appendix A"},{"comment":"The captions of Figs. 4 and 5 could state more explicitly that filled circles are dVQE data and dotted/dashed lines are exact diagonalization, and the infidelity axis scale should be consistent between the two figures.","section":"Figs. 4 and 5"},{"comment":"Reference [56] as used in the main text might be better replaced with an explicit derivation in an appendix, since the corrected bound is central to the error discussion.","section":"III A"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within scope for the journal and the algorithmic idea is timely. The errors identified above are correctable, but the authors should be asked to verify the numerical normalization and positivity constraints in their reported data. The hardware demonstration is limited to N = 1, so the practical quantum advantage claim is modest."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"dVQE is worth a careful read. The core idea—vectorize the Lindblad equation, minimize ⟨L†L⟩ on a doubled-qubit ansatz that respects Hermiticity and trace—is a clean, natural extension of VQE to open systems, and I don't see it in prior work. The demonstrations are real: classical noisy simulations for N=2 and a Rigetti hardware run, plus a noiseless N=8 check. That is more than many method papers ship.\n\nThe main soft spot is exactly the one the paper admits: no proof that small cost implies small error in observables. Sec. III A states this as f(ε) and leaves the proof open; Appendix A only gives random-state numerics comparing vector 2-norm to trace distance. That is honest but it is the load-bearing gap for anyone who wants to use dVQE for actual predictions. The footnote claiming a bound is also wrong as printed: footnote 56 has 1−f² ≥ C/δ, and the correct direction is 1−f² ≤ C/δ. Even with that fixed, the bound depends on the Liouvillian gap δ. Near weak dissipation or slow relaxation, δ can be small and C can be tiny while observable errors stay O(1). So the utility claim should be understood as conditional on a gap condition.\n\nTwo smaller points. Eq. (20) writes ⟨O⟩ = Σ λ_q ⟨q|V†OV|q⟩ without the normalization constant from the vector representation; I think the intended λ_q are normalized, but the formula as stated is missing the trace renormalization. And the positivity of the ansatz is under-specified: if the RY angles are not restricted, the λ_q can be negative and the \"density matrix\" is unphysical. The text mentions restrictions but doesn't spell them out.\n\nNone of this kills the paper. The algorithm itself is coherent, the cost function is not circular (it is directly defined from the Liouvillian, no fitted parameters), and the numerics support the basic claim. The error bound is a known open problem, not a hidden contradiction. For a method paper this is a solid contribution to the variational quantum simulation literature.\n\nI would recommend sending it to peer review with the expectation that the authors fix the footnote, state the normalization and positivity assumptions explicitly, and maybe add a remark that the cost-to-observable guarantee is gap-dependent. Reader: anyone working on NISQ algorithms for open quantum systems. I'd use it as a citation for prior art in variational NESS approaches.","headline":"dVQE is a real new method for NESS on near-term devices, but the cost-to-observable guarantee is unproven and footnote 56 states a reversed bound.","tokens_in":56302,"tokens_out":3168,"would_cite":true,"duration_ms":34565,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","81S22"],"pacs":[],"model":"deepseek-v4-flash","headline":"A variational algorithm obtains non-equilibrium steady states of open quantum systems by minimizing the squared Liouvillian on doubled qubits.","keywords":["open quantum systems","non-equilibrium steady state","variational quantum eigensolver","quantum master equation","Liouvillian","quantum-classical hybrid algorithm","dissipative quantum Ising model","persistent current"],"falsifier":"On a small exactly solvable dissipative model, such as the two-site Ising chain with damping and dephasing, optimize dVQE to a very small cost and then compare a sensitive two-body observable such as the persistent current against exact diagonalization; if the observable error remains large while the cost is tiny, the assumed cost-to-observable bound fails.","tokens_in":55259,"feed_emoji":"⚛️","tokens_out":6471,"duration_ms":67303,"temperature":0.7,"pith_summary":"The paper proposes a quantum-classical hybrid algorithm, called dVQE, to compute the non-equilibrium steady state (NESS) of an open quantum system. It maps the density matrix to a pure state on twice as many qubits via the Choi–Jamiołkowski isomorphism, then minimizes the expectation value of $\\hat L^\\dagger \\hat L$, where $\\hat L$ is the Liouvillian in this doubled space. Because $\\hat L^\\dagger \\hat L$ is Hermitian and non-negative, its zero eigenstates are exactly the steady states. The authors demonstrate on the dissipative quantum Ising model and a reservoir-engineered persistent-current model that the optimized state reproduces physical observables with infidelity of order $10^{-2}$, both in noisy numerical simulation and on actual quantum hardware. The method matters because it extends variational eigensolving from closed to open systems, potentially reaching steady states of dissipative many-body models that are hard to simulate classically.","feed_headline":"A variational loop finds steady states of open quantum systems","feed_subtitle":"Doubling the qubits turns the search into a VQE-style cost minimization with infidelity around 1 percent.","key_machinery":"The load-bearing object is the cost function $\\langle \\hat L^\\dagger \\hat L\\rangle$ evaluated on the vectorized density matrix. Here $\\hat L$ is the Choi–Jamiołkowski image of the GKSL Liouvillian, a non-Hermitian operator whose kernel is the steady state; squaring with its adjoint turns the fixed-point problem into a Hermitian ground-state problem. The ansatz is the tensor-product structure $V(\\theta_v)\\otimes V^*(\\theta_v)$ applied after entangling physical and ancillary qubits through CNOT gates, with a diagonal circuit $\\vec D(\\theta_d)$ controlling the eigenvalue distribution. This structure guarantees that the resulting matrix is a valid density matrix while keeping the circuit compatible with near-term hardware. Around this core, the paper adds a measurement scheme that returns to the $N$-qubit representation for observables and a sequential-minimal-optimization update loop for the parameters.","core_discovery":"The central claim is that the NESS of a Markovian open quantum system can be obtained variationally by searching for the ground state of the non-negative Hermitian operator $\\hat L^\\dagger \\hat L$ in the vectorized (doubled-qubit) representation, instead of simulating the dissipative dynamics. The paper establishes the mapping $\\hat L|\\rho_{SS}\\rangle=0 \\Leftrightarrow \\hat L^\\dagger \\hat L|\\rho_{SS}\\rangle=0$, so the steady state is the zero-energy ground state of the cost operator. To keep the state physical, the ansatz decomposes $\\rho$ as $V D V^\\dagger$ and realizes $|\\rho_\\theta\\rangle=[V(\\theta_v)\\otimes V^*(\\theta_v)]\\prod_n \\mathrm{CNOT}_{n,n+N}\\vec D(\\theta_d)|0\\rangle$, which enforces Hermiticity and positive semidefiniteness by construction. Observables are then evaluated on $N$ qubits by sampling computational basis states $|q\\rangle$ with weights $\\lambda_q$ and measuring $V|q\\rangle$. The demonstrations target the dissipative Ising model and the persistent-current model, reaching state infidelities of order $10^{-2}$.","pith_inferences":["If a rigorous version of the $f(\\epsilon)$ bound exists, dVQE would certify observable errors directly from the measured cost, giving open-system simulation the same kind of guarantee that ground-state VQE has for energies.","The decoupled $\\tilde D$ ansatz makes the eigenvalue distribution $\\lambda_q$ explicit, so the algorithm could be extended to estimate entropic quantities such as the von Neumann entropy with the same sampling procedure, which the paper does not discuss.","Because the cost is a local Pauli sum when the Liouvillian is local, Pauli grouping and measurement-shot weighting should transfer from closed-system VQE; the paper leaves this as a future question.","The ansatz structure suggests a testable extension: for fixed circuit depth, compare entangled versus decoupled eigenvalue-distribution circuits on highly entropic steady states to identify where expressive power limits fidelity."],"forward_implications":["Any open quantum system whose Liouvillian is a sum of local terms inherits a local cost operator $\\hat L^\\dagger \\hat L$, so the algorithm has the same per-iteration resource scaling as ordinary VQE.","Reaching cost $\\langle \\hat L^\\dagger \\hat L\\rangle<\\epsilon$ gives the optimized state a guaranteed overlap with the true steady state through the gap bound $1-f^2\\ge \\langle \\hat L^\\dagger \\hat L\\rangle/\\delta$, so the optimization quality is directly quantifiable.","The same scheme applies to models with multiple steady states by borrowing excited-state VQE methods, as noted in the paper.","The measurement protocol keeps observable estimation at $O(1/\\epsilon^2)$ shots for a target precision $\\epsilon$, so the added cost of open-system simulation is the doubled-qubit ansatz rather than an exponential number of measurements."],"supporting_citations":[{"why":"Defines the variational-quantum-eigensolver strategy that dVQE extends from closed to open systems.","marker":"[15]"},{"why":"Introduces the vectorized (doubled-space) steady-state formulation and variational ansatz that dVQE adapts to NISQ circuits.","marker":"[27]"},{"why":"Supply the GKSL master equation whose Liouvillian becomes the cost operator.","marker":"[50, 51]"},{"why":"Gives the uniqueness condition under which the steady state can be treated as a non-degenerate ground state.","marker":"[55]"},{"why":"Provides the sequential minimal optimization routine used to update the variational parameters.","marker":"[49]"},{"why":"Defines the reservoir-engineered persistent-current model used as one of the demonstrations.","marker":"[65]"},{"why":"Documents the cloud quantum device and instruction set used for the hardware demonstration.","marker":"[48]"}],"fun_headline_variants":["Variational loop digs out open-system steady states","Doubled qubits unlock steady states of open quantum systems","Open quantum systems get a variational steady-state solver","New hybrid algorithm pins down non-equilibrium steady states","Quantum algorithm for open-system steady states via VQE"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The algorithm relies on the unproven premise that a small cost value $\\langle \\hat L^\\dagger \\hat L\\rangle$ forces small errors in every physical observable; the paper gives numerical evidence on random states but leaves the mathematical proof open.","fun_headline_variants_meta":{"raw":{"variants":["Variational loop digs out open-system steady states","Doubled qubits unlock steady states of open quantum systems","Open quantum systems get a variational steady-state solver","New hybrid algorithm pins down non-equilibrium steady states","Quantum algorithm for open-system steady states via VQE"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000189,"raw_usage":{"total_tokens":1324,"prompt_tokens":925,"completion_tokens":399,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":541,"completion_tokens_details":{"reasoning_tokens":324}},"tokens_in":541,"tokens_out":399,"duration_ms":4348,"temperature":1.0,"reasoning_tokens":324,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:59:54.111296+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On a small exactly solvable dissipative model, such as the two-site Ising chain with damping and dephasing, optimize dVQE to a very small cost and then compare a sensitive two-body observable such as the persistent current against exact diagonalization; if the observable error remains large while the cost is tiny, the assumed cost-to-observable bound fails.","supporting_citations":[{"cited_title":"Variational quantum algo- rithms for discovering hamiltonian spectra,","cited_arxiv_id":null,"evidence_quote":"Introduces the vectorized (doubled-space) steady-state formulation and variational ansatz that dVQE adapts to NISQ circuits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the uniqueness condition under which the steady state can be treated as a non-degenerate ground state."},{"cited_title":"Thermo- electric effects in nanoscale junctions,","cited_arxiv_id":null,"evidence_quote":"Provides the sequential minimal optimization routine used to update the variational parameters."},{"cited_title":"Verstraete, M","cited_arxiv_id":null,"evidence_quote":"Documents the cloud quantum device and instruction set used for the hardware demonstration."}],"review_version":1}