{"id":"a394a4e5-0e6f-43a4-a7fc-9ff0d0e62b51","arxiv_id":"2412.17706","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A simplified Lindblad-based Gibbs state preparation protocol with local Pauli jumps is shown to mix polynomially under the eigenstate thermalization hypothesis, with numerical and noise-resilience analysis.","lead":"The authors present a simplified dissipative quantum algorithm for preparing Gibbs (thermal) states, using local Pauli jump operators and relying on the eigenstate thermalization hypothesis to guarantee fast convergence. The work includes analytical complexity bounds, numerical scaling studies in the mixed-field Ising model, and circuit simulations with depolarizing noise.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central polynomial-in-beta bounds rest on Assumption 1(b) that the ETH off-diagonal function f(E,nu) has support Theta(1/beta); this is not standard ETH and is never tested, so the efficiency claim is conditional on an unverified structural assumption.","rationale":"The reader's weakest_assumption identifies exactly this: the ETH frequency window Delta_RMT = Theta(1/beta) in Assumption 1(b) is load-bearing, and the numerics only probe a single beta value. I agree with that assessment. The concern is specific and technical: the conductance lower bound (Eq. 136) and the resulting spectral gap (Eqs. 144-146) would degrade exponentially in beta if the off-diagonal ETH function had a beta-independent width. The paper's stated motivation for Assumption 1(b) is insufficient: the normalization condition in Eq. (107) constrains the large-nu tail of f, not its support near nu=0. The filter function eta_nu does have width 1/beta, but the proof does not rely on the filter alone; it explicitly uses the support of f to restrict transitions to adjacent intervals and to control the concentration bounds in App. C.4. Therefore the theorem's polynomial beta scaling is not guaranteed for realistic ETH functions, and the numerical study cannot validate it because beta is fixed. This does not change the reader's CONDITIONAL verdict; it reinforces it. The authors should either prove the bounds using the filter width instead of f's support, or provide numerical/empirical evidence for Assumption 1(b) across a range of beta before claiming the ETH-based polynomial scaling.","tokens_in":65414,"tokens_out":9806,"duration_ms":98887,"concrete_test":"Numerically extract the ETH off-diagonal function for the mixed-field Ising model (30) at the chaotic point CH (h/J=1.0, m/J=0.4) using exact diagonalization for n=8, 10, 12. For local Pauli observables A (e.g., X_i and Z_i Z_{i+1}), compute the off-diagonal matrix elements A_ij = <E_i|A|E_j> in the bulk spectrum, bin them by E_ij and nu_ij, and estimate |f(E,nu)|^2 as Var(A_ij) * D(E_ij) where D(E) is the density of states. Then fit the width Delta_RMT of |f|^2 as a function of inverse temperature beta, varying beta J over 0.5, 1, 2, 4 (scaling J=1 and adjusting h, m accordingly). If Delta_RMT does not scale as Theta(1/beta) but saturates to an O(1) value, the conductance bound in Eq. (136) gains a factor e^{-2 beta Delta_RMT}, the gap in Eq. (144) becomes exponentially small in beta, and the mixing time bound in Thm. 2 cannot hold for large beta.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central theorem (Thm. 2, Eqs. 17-18) promises a mixing time polynomial in n and beta under the ETH. The proof requires Assumption 1(b) (App. C.1): the ETH off-diagonal function f(E,nu) is flat and supported on |nu| <= Delta_RMT = Theta(1/beta). This assumption is load-bearing in the conductance calculation: the transition rates (Eq. 129) are non-negligible only for adjacent energy intervals of width Delta_RMT, and the conductance lower bound (Eq. 136) contains a factor e^{-beta Delta_RMT} from detailed balance. If Delta_RMT were O(1) rather than Theta(1/beta), the spectral gap bound (Eq. 144) would become Omega(exp(-2 beta Delta_RMT) / poly(n)), and the mixing time bound would be exponential in beta. The paper's motivation (Eq. 107) only requires |f(nu)| to decay faster than e^{-beta nu/4} for large negative nu; it does not imply compact support of width 1/beta. Moreover, the numerical study fixes beta = (2J)^{-1} throughout, so the beta-dependence is never empirically probed. The filter eta_nu has width 1/beta by design, but the proof as written uses the support of f, not of eta, in the concentration bounds (Eqs. 162, 171) and in the conductance decomposition; without Assumption 1(b) the proof does not go through. Thus the claimed polynomial scaling with beta is conditional on a structural property of realistic chaotic Hamiltonians that is neither part of the standard ETH nor verified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a Lindblad-engineering algorithm for preparing Gibbs states of non-commuting Hamiltonians. The Lindbladian (3) uses filtered operator-Fourier-transform jump operators built from local Pauli strings, with coherent part G = H, and the authors argue that under the eigenstate thermalization hypothesis (ETH) the ETH-averaged Lindbladian is approximately sigma_beta-detailed-balanced, has spectral gap Omega(1/(n beta^2)), and gives a mixing time O(n beta^2 (beta ||H||_infinity + log(1/epsilon))) with high probability (Thm. 2, Eqs. 17-18), provided the number of jump operators is poly(n, beta). The circuit implementation uses a single ancilla and a randomized selection of one jump operator per step, with Hamiltonian-simulation time Theta(beta t_mix^2 / epsilon sqrt(log(beta t_mix / epsilon))) (Thm. 1) and a detailed error budget (Prop. 1). A second contribution is a noise-resilience analysis (Thm. 3): for stochastic noise, early-time errors are damped by the contractive Lindblad dynamics. The claims are complemented by numerics for the mixed-field Ising model (n = 3,...,8, beta = (2J)^{-1}), including mixing-time and steady-state-distance scalings, and by full circuit simulations with depolarizing noise.","tokens_in":65768,"tokens_out":22471,"duration_ms":191588,"significance":"If the central theorem held as stated, the protocol would be a practically relevant simplification over CKG-style Gibbs samplers, and the single-ancilla randomized implementation with local Pauli jump operators is a genuine resource reduction. Strengths: the analytical derivation is elaborate and mostly self-contained; the error analysis of the circuit implementation (truncation, discretization, Trotter, dilation) is careful; the paper ships code and compiled circuits on GitHub; the numerical studies give falsifiable predictions (polynomial mixing in n in the chaotic regime, exponent kappa approx 1.4; |A|-dependence of the steady-state deviation with kappa approx -0.2); and the noise-resilience bound (Thm. 3) formalizes an effect usually only discussed informally. However, the headline quantitative claims are conditional on a structural assumption on the ETH off-diagonal function whose beta-dependence is neither standard nor tested, and the central theorem contains internal inconsistencies in the stated beta- and epsilon-scalings that must be fixed before the resource counts are reproducible.","major_comments":[{"comment":"The polynomial-in-beta claim of Thm. 2 is conditional on Assumption 1(b), which is not a standard consequence of the ETH and is never probed in the numerics. The assumption is load-bearing: the conductance bounds in Eqs. (136)-(140) carry a factor e^{-beta Delta_RMT}, the gap bound (144) uses Delta_RMT^2 and the overlap Gamma, and the concentration estimates (162) and (171) use that f(nu) vanishes for |nu| > Delta_RMT. If the physical off-diagonal function for a chaotic Hamiltonian had support of width O(1) rather than Theta(1/beta), the same derivation would yield a gap Omega(e^{-O(beta)}/poly(n)) and a mixing time exponential in beta. The motivation in Eq. (107) only forces |f(nu)| to decay faster than e^{-beta nu/4} for large negative nu, which does not imply compact support of width 1/beta, so Assumption 1(b) is an added structural assumption, not a consequence of the ETH. All numerical studies fix beta = (2J)^{-1} (Sec. 5, App. D), so the assumed beta-dependence of Delta_RMT is never tested. I ask the authors to (i) state Thm. 2 in the abstract and introduction as conditional on Assumption 1(b), and (ii) provide a concrete test, e.g. extract |f(E,nu)| or its effective support width from eigenstate matrix elements of local observables in the chaotic lake of Eq. (30) at several beta, and/or scan t_mix as a function of beta at fixed n. These are well-posed and feasible checks; without them the claimed polynomial scaling in beta cannot be assessed.","section":"App. C.1, Assumption 1(b); Thm. 2"},{"comment":"The quantitative statement of Thm. 2 has internal inconsistencies in the beta- and epsilon-scalings. Eq. (18) (and Thm. 2, Eq. (102)) quotes the steady-state error as O(epsilon + n beta^2 |A|^{-1/2} (beta ||H||_infinity + log(1/epsilon))), but the proof at Eq. (192) yields n beta^3 |A|^{-1/2} (...): gap^{-1} = O(n beta^2) from Eq. (146) times the channel distance O(beta/|A|^{1/2}) of Eq. (178). The sufficient number of jump operators is quoted as Omega(n^2 beta^6 ||H||^2_infinity / epsilon) in the main text after Eq. (18), as Omega(n^2 beta^8 ||H||^2_infinity / epsilon) in Eq. (194), as Omega(n^2 beta^6) in Eq. (183), and as Omega(n^2 beta^8) in the main text after Eq. (17); these statements differ by powers of beta. In addition, plugging |A| = Omega(n^2 beta^6 ||H||^2_infinity / epsilon) into Eq. (18) leaves a residual of order sqrt(epsilon) log(1/epsilon)/(beta ||H||_infinity), so the stated choice does not yield an epsilon-close steady state; the derivation implies |A| = Omega(n^2 beta^6 (beta ||H||_infinity + log(1/epsilon))^2 / epsilon^2) (with the beta-exponent shifted by two if one follows Eq. (192)). These mismatches should be reconciled so that the resource counts of Thm. 2 are reproducible; the qualitative polynomial dependence survives the correction.","section":"Sec. 3.2, Eqs. (17)-(18); App. C.5, Eqs. (183), (192)-(194)"},{"comment":"The numerical support for the claimed polynomial mixing-time scaling is weaker than the text suggests. The fits in Fig. 4 use six points (n = 3,...,8), report no error bars or bootstrap intervals, and the fitted exponents vary appreciably with |A| (e.g., for CH, kappa approx 1.09-1.26 for the mixing time and kappa approx -0.67 to -1.04 for the gap). Moreover, the mixing-time estimate (31) is computed from a single initial state, the maximally mixed state, whereas the mixing time in Eq. (7) is worst-case over all initial states; the text acknowledges this, but the plotted polynomial scaling is then an initial-state-dependent quantity, and for the REG point the trend is explicitly non-monotonic in n (Sec. 5.2, App. D.2.2). I recommend reporting bootstrap uncertainties on kappa and, at least for the chaotic point CH, the worst-case t_hat_mix over several random initial states (e.g., the Haar-random states already used in Fig. 15); the initial-state-independent spectral-gap data in the bottom row of Fig. 4 is the more robust witness of the polynomial trend.","section":"Sec. 5.2, Eq. (31), Fig. 4"},{"comment":"The headline noise-resilience figures rest on an extrapolation whose hypotheses are only partially validated. The bound (35) requires the convergence inequality (33) to hold for all input states reachable under the noiseless dynamics (App. E.3), but B and alpha are fitted to a single initial state, rho(0) = I/2^n, for n = 3,...,8 (Sec. 6.3, App. E.4), and the fitted exponent alpha proportional to n^{-1.39} is quoted without uncertainty. Combined with the assumed gate model N_g = 50 n per step, this leads to the statement that at lambda_g = 10^{-8} Gibbs states of up to n = 100 qubits can be prepared with error approx 0.2. The text does flag the extrapolative nature of the estimate, but the phrasing under Fig. 6 reads as a concrete capability claim. I recommend reformulating that sentence as a heuristic extrapolation and, where feasible, testing the all-inputs condition on a few additional initial states at small n (the Haar-random states of Fig. 15 partially serve this purpose).","section":"Sec. 6.3, Eqs. (33)-(35), Fig. 6"}],"minor_comments":[{"comment":"The statement that the number of Bohr frequencies satisfies |B_H| <= 4n is incorrect: for an n-qubit Hamiltonian the set of pairwise energy differences can have up to 4^n distinct elements. The Theta scaling in Eq. (85) only needs log|B_H| = O(n), so the asymptotic conclusion is unaffected, but the stated bound should be corrected.","section":"App. B.3, Eq. (85)"},{"comment":"The theorem statement says 'with probability of no error incurring lambda', but the proof and Eq. (34) use lambda as the overall error probability; the two conventions should be aligned (error with probability lambda, no error with probability 1 - lambda).","section":"Thm. 3 (App. E)"},{"comment":"Typo: 'we ues a randomized simulation strategy' should read 'we use a randomized simulation strategy'.","section":"Sec. 5.1"},{"comment":"The statement that MCWF 'will likely require runtimes of around > 10^6 seconds' is an extrapolation beyond the plotted range of Fig. 20; stating the extrapolation method (e.g., a power-law fit in N_tra) would make the claim checkable.","section":"App. D.3.3, Fig. 20"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within scope for Quantum, and the authors are transparent about the main assumption; the derivation is elaborate and largely self-contained, and the released code and circuit data are valuable. My main concerns are (i) the gap between the abstract's 'shown to be efficient under the ETH' and the theorem's reliance on the untested Assumption 1(b); (ii) the internal beta/epsilon-scaling inconsistencies in Thm. 2 and its proof; and (iii) the extrapolative noise claims. None of these is a basis for rejection: they are fixable within the manuscript's scope by conditional restatement, a beta-scan, corrected scalings, and softer wording. The arXiv header indicates the paper has been accepted at Quantum; that does not affect my assessment of the present text."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, it is one of the more honest and thorough treatments of dissipative Gibbs preparation I have seen recently: the protocol (OFT Lindblad operators, single-ancilla dilation, randomized jump-operator selection, ETH-based removal of the CKG coherent term) is genuinely new, and the authors never hide that their efficiency claim is conditional on an ETH ansatz. Second, the central bound — polynomial mixing time in n and beta — rests on Assumption 1(b) in App. C.1: the ETH off-diagonal function f(E,\\nu) is flat and supported on |\\nu| <= Theta(1/beta). That is not part of the standard ETH, which only guarantees exponential decay of the trace-weight integral, and the paper does not check it numerically. The stress-test note is right that if the ETH frequency window were O(1) instead, the conductance lower bound would pick up e^{-2\\beta\\Delta_RMT} and the mixing time would become exponential in beta. The authors state the assumption clearly and flag it as a limitation, but the reader should not walk away thinking the beta-scaling is empirically demonstrated: all numerics fix beta = (2J)^{-1}, so the temperature dependence is never probed.\n\nWhat the paper does well is substantial. The spectral-gap analysis via reduction to a classical Markov chain and conductance is detailed, and the concentration argument bounding the distance between L and its ETH average is carefully executed. The numerics on the mixed-field Ising model are honest: the chaotic/regular distinction is based on fractal dimensions and level-spacing ratios, and the scaling plots show the data and fitted exponents, though without error bars on those exponents. The noise-resilience analysis — that early-time stochastic errors get damped by the contractive Lindblad dynamics — is a real and useful contribution, and the circuit simulations with local depolarizing noise validate the qualitative trade-off between algorithmic and noise errors. They also ship the compiled circuits and code, which counts.\n\nSoft spots, in proportion. The load-bearing Assumption 1(b) is the main one; it is not verified, and the beta-dependence is untested. The extrapolation of B and alpha from n <= 8 to n = 100 in the noise study is a phenomenological fit, and the paper says so, but it is easy to over-read. The algorithmic-error fit uses n = 5 only, so the four fitted coefficients are illustrative rather than predictive. None of these are fatal; the paper is explicitly conditional, and the qualitative claims — ETH helps, local jumps help, noise is damped — are supported.\n\nWho is this for? Anyone working on near-term demonstrations of dissipative Gibbs sampling, and theorists interested in ETH-based Lindblad engineering. It deserves serious refereeing; I would send it out and I would cite it for the protocol and the noise analysis, while being careful not to cite the polynomial beta-scaling as established for generic chaotic Hamiltonians.","headline":"A careful, useful Lindblad Gibbs-preparation paper whose polynomial mixing-time theorem is genuinely conditional on a non-standard and untested ETH frequency-window assumption, but which still earns a serious read.","tokens_in":66368,"tokens_out":1758,"would_cite":true,"duration_ms":21354,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["03.67.-a","03.65.Yz"],"model":"deepseek-v4-flash","headline":"Under the eigenstate thermalization hypothesis, a simple dissipative Lindblad circuit with local Pauli jump operators prepares Gibbs states in polynomial time with a single ancilla, and the dynamics tolerates stochastic noise.","keywords":["quantum Gibbs state preparation","Lindblad engineering","eigenstate thermalization hypothesis","mixing time","quantum detailed balance","noise resilience","single-ancilla protocol","mixed-field Ising model"],"falsifier":"Take the mixed-field Ising model at its chaotic parameter point, obtain its exact eigenstates for $n$ up to practical limits, and directly estimate the off-diagonal ETH function $|f(E,\\nu)|^2$ from matrix elements of local observables while sweeping $\\beta$ around $\\beta=(2J)^{-1}$. If the support in $\\nu$ is $O(1)$ rather than $\\Theta(1/\\beta)$, or if the measured mixing time grows faster than $n\\beta^2(\\beta\\|H\\|_\\infty+\\log(1/\\epsilon))$ as $\\beta$ increases, the central claim is falsified. The paper's numerics fix $\\beta=(2J)^{-1}$, so this temperature sweep is the decisive missing test.","tokens_in":65173,"feed_emoji":"⚛️","tokens_out":16216,"duration_ms":142092,"temperature":0.7,"pith_summary":"This paper proposes a simplified Lindblad-engineering algorithm for preparing Gibbs (thermal) states of a quantum Hamiltonian on a quantum computer, and proves that the simplification is efficient precisely in the regime where the Hamiltonian is quantum chaotic. The main claim is that if the Hamiltonian satisfies the eigenstate thermalization hypothesis (ETH), then the Lindbladian with coherent part $G=H$ and local Pauli jump operators is $\\sigma_\\beta$-detailed balanced on average, so its steady state is close to the target Gibbs state $\\sigma_\\beta=e^{-\\beta H}/\\operatorname{Tr}[e^{-\\beta H}]$. Under that condition the mixing time is polynomial in the number of qubits, $t_{\\rm mix} \\le O(n\\beta^2(\\beta\\|H\\|_\\infty+\\log(1/\\epsilon)))$ with high probability given a sufficiently large set of jump operators, and the steady-state error obeys $\\|\\rho_\\infty-\\sigma_\\beta\\|_1 \\le O(\\epsilon + n\\beta^2|\\mathcal A|^{-1/2}(\\beta\\|H\\|_\\infty+\\log(1/\\epsilon)))$. The circuit implementation uses a single ancilla and randomized single-jump-operator steps, with total Hamiltonian simulation time $\\Theta(\\beta t_{\\rm mix}^2 \\epsilon^{-1}\\sqrt{\\log(\\beta t_{\\rm mix}/\\epsilon)})$; the paper also proves and numerically demonstrates damping of stochastic noise errors, and verifies polynomial mixing on the chaotic regime of the mixed-field Ising model.","feed_headline":"Simple dissipative circuit prepares Gibbs states in polynomial time","feed_subtitle":"For chaotic systems, local Pauli jumps and one ancilla make thermal states fast and noise-tolerant.","key_machinery":"The central object is a filtered operator Fourier transform jump operator $L_a = \\int dt\\, g(t) e^{iHt} A_a e^{-iHt}$, with a Gaussian filter $g(t)$ whose frequency profile $\\eta_\\nu$ has width $\\Delta_E = \\sqrt{2/\\beta}$ and is peaked at negative frequencies, so it selects energy-lowering transitions on the scale $1/\\beta$. Under the ETH ansatz, the off-diagonal matrix elements of the local observable $A_a$ are random with variance $|f(E,\\nu)|^2/D(E)$, and Assumption 1 takes $f$ to be flat on a window of width $\\Delta_{\\rm RMT}=\\Theta(1/\\beta)$ overlapped by this filter. The ETH average makes the dissipative part of the Lindbladian detailed balanced with respect to $\\sigma_\\beta$ and kills the extra coherent correction needed in earlier exact constructions, reducing the averaged generator to a classical Markov chain on the energy spectrum. A conductance lower bound yields spectral gap $\\Omega(1/(n\\beta^2))$, and concentration bounds show a single realization stays within $O(\\beta/\\sqrt{|\\mathcal A|})$ of the averaged generator, which is what upgrades the averaged statement to the high-probability mixing-time and accuracy theorems.","core_discovery":"The paper's central discovery is that the ETH turns 'approximately detailed balanced' into 'detailed balanced on average' for a much simpler Lindbladian than previously used. Averaging over the random matrix elements in the ETH ansatz makes the decay term of the dissipator commute with the modular action of $\\sigma_\\beta$, and makes the coherent correction used in earlier exact-constructi ons constructions vanish; since $H$ commutes with $\\sigma_\\beta$, setting $G=H$ preserves the steady state. The ETH-averaged generator reduces to a classical Markov chain on the energy eigenstates, whose conductance gives a spectral gap $\\Omega(1/(n\\beta^2))$, and concentration bounds (using the flatness and narrow support of the off-diagonal ETH function $f(E,\\nu)$) show a single realization of the Lindbladian differs from the average by $O(\\beta/\\sqrt{|\\mathcal A|})$ with high probability. Theorem 2 then gives $t_{\\rm mix}\\le O(n\\beta^2(\\beta\\|H\\|_\\infty+\\log(1/\\epsilon)))$ and $\\|\\rho_\\infty-\\sigma_\\beta\\|_1\\le O(\\epsilon+n\\beta^2|\\mathcal A|^{-1/2}(\\beta\\|H\\|_\\infty+\\log(1/\\epsilon)))$; Theorem 1 gives the single-ancilla circuit cost $\\Theta(\\beta t_{\\rm mix}^2\\epsilon^{-1}\\sqrt{\\log(\\beta t_{\\rm mix}/\\epsilon)})$ in Hamiltonian simulation time. Numerics on the mixed-field Ising model at $\\beta=(2J)^{-1}$ show mixing times scaling roughly as $n^{1.2}$ to $n^{1.4}$ in the chaotic regime, with slower convergence in regular limits, and noisy circuit simulations match the predicted noise bounds.","pith_inferences":["Beyond the paper: the narrow frequency window $\\Delta_{\\rm RMT}=\\Theta(1/\\beta)$ is the empirical crux; measuring $|f(E,\\nu)|^2$ for a chaotic local Hamiltonian across a range of $\\beta$ would test whether the polynomial claims survive outside the single temperature $\\beta=(2J)^{-1}$ used in the numerics.","Beyond the paper: the observed steady-state error exponent ($\\kappa\\approx -0.2$ in $|\\mathcal A|$ for chaotic points, versus $-1/2$ in the bound) hints that a fixed number of jump operators may keep $\\|\\rho_\\infty-\\sigma_\\beta\\|_1$ under control as $n$ grows; that would be stronger than Theorem 2 states.","Beyond the paper: the coherent term $-i[H,\\rho]$ drops out of the analytical gap calculation but visibly improves convergence in the numerics, suggesting the design space (filter shape, dissipation strength, partial coherent corrections) contains parameters that could further shorten mixing times.","Beyond the paper: since the noise-resilience proof is for stochastic noise, applying randomized compiling to convert coherent hardware errors into stochastic channels would make the protocol's guarantees directly relevant to current devices."],"forward_implications":["Gibbs state preparation for non-integrable local Hamiltonians reduces to a single-ancilla circuit with total Hamiltonian simulation time $\\Theta(\\beta t_{\\rm mix}^2 \\epsilon^{-1}\\sqrt{\\log(\\beta t_{\\rm mix}/\\epsilon)})$, avoiding the block-encoding overhead of the known near-optimal Lindblad simulation.","The ETH assumption removes the need for the exactly detailed-balanced coherent term and for $n$-qubit unitary 2-design jump operators; local Pauli products suffice, which is what makes the circuit depths plausible for near-term hardware.","In the chaotic regime of the mixed-field Ising model the mixing time scales roughly as $n^{1.2}$ to $n^{1.4}$, close to the analytical polynomial bound, while regular (non-chaotic) limits mix much more slowly or with strong initial-state dependence.","Stochastic noise does not drive the state to the noise channel's fixed point: the bound $\\|\\tilde\\rho_\\infty-\\sigma_\\beta\\|_1\\le B\\lambda/(1-(1-\\lambda)e^{-\\alpha})$ shows early-time errors are damped by the dissipative dynamics, and depolarizing circuit simulations confirm the predicted trade-off between algorithmic error and noise.","Because $|\\mathcal A|$ does not enter the circuit complexity of the Lindblad simulation, the steady-state accuracy can be improved by adding jump operators without increasing circuit depth, up to the limits of the jump operator model."],"supporting_citations":[{"why":"Supplies the ETH ansatz for matrix elements that makes the Lindbladian detailed-balanced on average.","marker":"[48]"},{"why":"Introduces the filtered-operator-Fourier-transform Lindblad simulation algorithm that this protocol simplifies.","marker":"[32]"},{"why":"Constructs an exactly $\\sigma_\\beta$-detailed-balanced Lindbladian whose coherent correction the ETH renders unnecessary.","marker":"[33]"},{"why":"Provides the conductance calculation for the ETH-averaged generator used to lower-bound the spectral gap.","marker":"[60]"},{"why":"Establishes that random sampling of one jump operator per step approximates Lindblad evolution with error $O(1/M)$.","marker":"[38]"},{"why":"Uses a discrete set of operator Fourier transform jump operators, the construction adopted for the protocol.","marker":"[35]"},{"why":"Supplies the single-ancilla dilation used to implement each dissipative step in the quantum circuit.","marker":"[34]"},{"why":"Gives the bound turning the spectral gap into a mixing time, $t_{\\rm mix}\\le \\Delta_L^{-1}\\log(2\\|\\rho_\\infty^{-1/2}\\|_\\infty/\\epsilon)$.","marker":"[83]"}],"fun_headline_variants":["ETH harnessed for fast, noise-tolerant Gibbs state preparation","Polynomial-time Gibbs states via ETH Lindbladian","One ancilla, ETH: fast thermal states that fight noise","Simplified Lindblad scheme for Gibbs states, ETH-accelerated","Noise-tolerant, fast Gibbs states from a single-ancilla circuit"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on Assumption 1(b): the off-diagonal ETH function $f(E,\\nu)$ is flat and supported on a frequency window of width $\\Delta_{\\rm RMT}=\\Theta(1/\\beta)$ that the Gaussian filter of width $\\sqrt{2/\\beta}$ overlaps, so if real chaotic Hamiltonians have a wider ETH frequency window the conductance calculation and the polynomial mixing-time claim stop working.","fun_headline_variants_meta":{"raw":{"variants":["ETH harnessed for fast, noise-tolerant Gibbs state preparation","Polynomial-time Gibbs states via ETH Lindbladian","One ancilla, ETH: fast thermal states that fight noise","Simplified Lindblad scheme for Gibbs states, ETH-accelerated","Noise-tolerant, fast Gibbs states from a single-ancilla circuit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000711,"raw_usage":{"total_tokens":3269,"prompt_tokens":1081,"completion_tokens":2188,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":697,"completion_tokens_details":{"reasoning_tokens":2102}},"tokens_in":697,"tokens_out":2188,"duration_ms":13222,"temperature":1.0,"reasoning_tokens":2102,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T05:15:10.419488+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the mixed-field Ising model at its chaotic parameter point, obtain its exact eigenstates for $n$ up to practical limits, and directly estimate the off-diagonal ETH function $|f(E,\\nu)|^2$ from matrix elements of local observables while sweeping $\\beta$ around $\\beta=(2J)^{-1}$. If the support in $\\nu$ is $O(1)$ rather than $\\Theta(1/\\beta)$, or if the measured mixing time grows faster than $n\\beta^2(\\beta\\|H\\|_\\infty+\\log(1/\\epsilon))$ as $\\beta$ increases, the central claim is falsified. The paper's numerics fix $\\beta=(2J)^{-1}$, so this temperature sweep is the decisive missing test.","supporting_citations":[{"cited_title":"Quantum logarithmic Sobolev inequalities and rapid mixing","cited_arxiv_id":null,"evidence_quote":"Gives the bound turning the spectral gap into a mixing time, $t_{\\rm mix}\\le \\Delta_L^{-1}\\log(2\\|\\rho_\\infty^{-1/2}\\|_\\infty/\\epsilon)$."}],"review_version":1}