{"id":"762c1563-fb08-43a2-abd8-2f253ea4fe5b","arxiv_id":"2505.18260","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The authors conjecture an ETH-like thermalization for open quantum systems via Liouvillian stripes, with numerical support from random Liouvillians and a driven-dissipative spin chain.","lead":"This paper extends the eigenstate thermalization hypothesis (ETH) to open quantum systems by slicing the Liouvillian spectrum into vertical stripes and defining effective chaotic Hamiltonians within each stripe. Numerical evidence from random models and a driven-dissipative spin chain suggests that chaotic dissipative dynamics hides internal oscillations from local probes.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Stripe width is selected by maximizing the GOE diagnostic that defines the stripe, so the Wigner-Dyson statistics and the 1/D matrix-element scaling may be partly artifacts of the selection; robustness to stripe definition is untested.","rationale":"Both the reader and I identify the stripe-construction rule as the weakest point. I considered alternatives: the failure to test the full Ansatz (smooth f(E,omega,Gamma)), the missing overlap terms for non-Hermitian eigenvectors, and the small system sizes. None is as load-bearing, because each would weaken but not collapse the argument if fixed. The stripe selection is different: the conjecture is only meaningful for Liouvillians that admit a stripe with GOE-like level statistics, and the paper defines admit by optimizing the very statistic. This creates a circularity in the spectral part of the evidence. The matrix-element part is then evaluated only on the optimized partition, so the 1/D scaling could be a by-product of selecting eigenstates whose frequencies are maximally rigid (GOE-like). The integrable control is useful (r remains 0.386 for all d), but it does not rule out the possibility that, for chaotic spectra, the optimized width is special for matrix elements. The proposed test, varying the width and using a fixed-width or fixed-count rule, would resolve whether the result is robust. If the test passes, the paper's evidence is solid and the conditional verdict stands; if it fails, the conjecture remains unsubstantiated for generic Liouvillians. Since the reader has already requested that this selection bias be addressed, the verdict should remain CONDITIONAL, and no change is needed.","tokens_in":14661,"tokens_out":8144,"duration_ms":71498,"concrete_test":"Recompute the spin-chain ETH diagnostics at N=7,8,9 (Delta=0.8, h=1, parameters as in Fig. 2) for three fixed stripe widths d = dmax/2, dmax, 2 dmax, and for a width fixed by a sample-size-independent criterion (e.g., a fixed number of eigenvalues per stripe, N_str ~ 1000). For each width, measure r(d) with bootstrap confidence intervals, the scaling exponent of var_alpha alpha with D, and the Gaussianity (e.g., Kolmogorov-Smirnov p-value) of the off-diagonal Om_alpha beta. If 1/D scaling and Gaussianity persist within error bars across these choices, the post hoc optimization is not the source of the effect; if they degrade, the central numerical evidence for Eq. (6) is not independent of the stripe-selection rule.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The conjecture (Eq. 6) is stated for local superoperators acting on a Liouvillian stripe, but the stripe is not defined a priori. The Supplementary Information fixes dmax as the width that maximizes the 1D level-spacing ratio r (Eq. 12), i.e., the very diagnostic used to certify that the stripe is chaotic. Consequently, the observed GOE-like statistics within a stripe are, to an unknown degree, a property of the optimization rather than of the Liouvillian. Because the same dmax determines which eigenpairs are included in the matrix-element statistics, the reported 1/D scaling of the diagonal variance and the Gaussian off-diagonal distribution (Figs. 1 and 2) are conditioned on this data-dependent partition. The paper does not demonstrate robustness of the ETH signatures to the stripe construction: no results are shown for fixed widths, for widths away from dmax, or for alternative partition rules. For a conjecture whose domain of validity is defined by the existence of such stripes, this is the load-bearing gap: if the ETH-like behavior disappears under a reasonable, non-optimized stripe definition, the numerical support is an artifact of the selection rule.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a generalization of the eigenstate thermalization hypothesis (ETH) to open quantum systems described by Lindblad master equations. The central object is the 'Liouvillian stripe', a vertical slice of the complex Liouvillian spectrum at fixed decay rate, on which the imaginary parts of the eigenvalues are treated as the real spectrum of an effective pseudo-Hermitian Hamiltonian. The authors conjecture (Eq. 6) that matrix elements of local superoperators within a stripe follow an ETH-like Ansatz: a smooth diagonal term plus a Gaussian random part whose variance scales as the inverse of the Liouville-space dimension. Numerical support is presented for random Liouvillians and for a driven-dissipative XXZ spin chain, showing 1/D variance scaling and Gaussian off-diagonal statistics in chaotic (but not integrable) regimes, together with a dynamical consequence: suppression of coherent oscillations in local observables. The paper is framed as a conjecture with numerical evidence, not a proof.","tokens_in":14919,"tokens_out":6528,"duration_ms":51580,"significance":"If valid, the Liouvillian-ETH conjecture provides a conceptual bridge between non-Hermitian random matrix theory, quantum chaos, and thermalization in open systems, and it yields a concrete, falsifiable prediction for the scaling of superoperator matrix elements. The manuscript is explicitly honest about the conjectural status and ships reproducible numerical data via QuantumToolbox.jl. The negative control (integrable case) and the dynamical predictions add value. However, the evidence as presented is not yet at the standard needed to establish the conjecture: the stripe definition is data-dependent, the scaling tests are under-powered, and the distributional claims are not quantitatively validated.","major_comments":[{"comment":"The Liouvillian stripe is defined by the width dmax that maximizes the one-dimensional level-spacing ratio ⟨r⟩, the same diagnostic used to certify that the stripe has GOE-like spectral statistics. Because the identical selection rule determines which eigenpairs are included in the matrix-element statistics in Figs. 1 and 2, the reported 1/D variance scaling and Gaussian off-diagonal distributions are conditioned on a partition that has been optimized to look GOE-like. The paper does not test robustness to this choice: no results are shown for fixed widths, for widths away from dmax, or for alternative partition rules (e.g., based on density of states or a fixed fraction of the spectral span). Since the conjecture (Eq. 6) is stated for 'a Liouvillian stripe' without an a priori definition, this is a load-bearing gap: if the ETH signatures disappear under a reasonable, non-optimized stripe definition, the numerical support is an artifact of the selection rule.","section":"Supplementary Information, Sec. II (Eq. 12)"},{"comment":"The claim that the diagonal variance scales as 1/D is supported by only four system sizes for the random Liouvillian (D = 60, 80, 100, 120) and three for the spin chain (N = 7, 8, 9), with no error bars and no fitted exponent. The insets show data points compared visually to a 1/D line, but no power-law fit, confidence intervals, or statistical test are provided. Given that Eq. (6) makes a precise prediction for the exponent and the prefactor structure, the numerical evidence should include a quantitative fit (e.g., log-log regression with uncertainties) or additional system sizes; without this, the scaling claim is not established beyond qualitative inspection.","section":"Fig. 1(a) inset and Fig. 2(c) inset"},{"comment":"The Gaussianity of off-diagonal matrix elements is asserted from visual comparison of the empirical PDF with a Gaussian of the same variance, for a single energy window (|ω| = 10, δω = 0.2 in the random case; |ω| = 8, δω = 0.1 in the spin chain). The text states 'Similar results are obtained for different choices of ω and δω (not shown)', but no quantitative goodness-of-fit test (e.g., Kolmogorov-Smirnov, Anderson-Darling, or cumulant ratios) is reported. Since Eq. (6) predicts Gaussian R_αβ for all ω, a more systematic and quantitative distributional test is needed to support the conjecture.","section":"Fig. 1(b) and Fig. 2(d)"},{"comment":"Equation (6) uses δ_αβ for the diagonal term, whereas the non-Hermitian ETH recalled in Eq. (3) contains the overlap ⟨φ_α|φ_β⟩, which is not δ_αβ for non-orthogonal right eigenvectors. The authors do not specify whether the matrix elements O_αβ in Eq. (6) are computed in the biorthogonal basis (with ⟨l_α|r_β⟩ = δ_αβ) or in the right-eigenbasis alone. This distinction is essential for interpreting the numerics in Figs. 1 and 2 and for the correctness of the Ansatz. Please clarify the basis and normalization convention, and if the right-only basis is used, justify the replacement of the overlap with δ_αβ.","section":"Eq. (6) and preceding text"}],"minor_comments":[{"comment":"There is a typo in the phrase 'This interpretation is in contrst with' — 'contrst' should be 'contrast'.","section":"Introduction"},{"comment":"The notation eO(E) for the smooth function is nonstandard and confusing, as it may be misread as the exponential of an operator. Consider denoting it as O(E) or a distinct symbol such as Ō(E).","section":"Eq. (1)"},{"comment":"'The purple line referes to the integrable model' should read 'refers'.","section":"Fig. 3 caption"},{"comment":"The caption says 'Scaling of the stripe width dmax' but the main text in the same paragraph explains that for unbounded random matrices dmax remains constant while for matrices constrained to a box it scales polynomially; please make the caption consistent with the text.","section":"Supplementary Information, Fig. 4(d) caption"},{"comment":"The 'Note added' mentions Ref. [54] as an alternative definition of ETH in open systems, but the relation and contrast to the present framework are not discussed; adding a brief comparison would help the reader situate the contribution.","section":"Discussion and conclusions"}],"recommendation":"major_revision","confidential_remarks":"The paper presents an interesting conjecture with suggestive numerics, but the central strip-construction step is self-referential: the same ⟨r⟩ diagnostic used to select dmax also certifies the GOE character of the stripe and determines the eigenstates entering the matrix-element statistics. This is a fixable issue if the authors demonstrate robustness to the stripe definition and provide quantitative scaling and distributional tests. Given the explicit conjectural framing, I do not see a fatal error, but the evidence base is too thin for acceptance as is."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know about this paper: it's a genuine attempt to extend ETH to Lindbladians via a spectral \"Liouvillian stripe\" construction, and the numerical evidence is suggestive but the stripe selection rule is dangerously close to circular.\n\nThe Liouvillian stripe idea is new and clean: slice the complex spectrum into vertical strips of nearly constant decay rate, treat each strip's imaginary parts as eigenvalues of a pseudo-Hermitian Hamiltonian, then conjecture ETH for local superoperators on those eigenstates. Eq. (6) is a real conjecture, falsifiable, and the paper tests it on two very different systems — a random Liouvillian and a driven-dissipative XXZ chain with an impurity. The integrable/chaotic contrast is well drawn. The dynamical consequence — that chaotic dissipative dynamics suppresses coherent oscillations and leaves local response structureless — is a nice physical payoff and gives the conjecture practical teeth. The numerics show the expected 1/D variance scaling and Gaussian off-diagonal distributions.\n\nThe main soft spot is the stripe construction. The width dmax is chosen by maximizing the average level-spacing ratio <r> inside the stripe — the very diagnostic that certifies GOE-like statistics. That means the reported Wigner-Dyson behavior and, more importantly, the matrix-element statistics computed on that stripe are conditioned on a data-dependent partition. The paper doesn't test robustness: no fixed widths, no alternative definitions, no check that the ETH signatures survive a non-optimized stripe. Since the conjecture's domain is defined by the existence of such stripes, this is a real gap, not a cosmetic one. The numerical evidence is also thin: three or four system sizes, no error bars, no fitted exponents — the 1/D scaling is asserted from visual comparison. These are fixable with more numerics, but as it stands the claim \"demonstrates the validity of Liouvillian ETH\" overreaches; \"supports the conjecture\" would be accurate.\n\nThis paper is for people working on quantum chaos, open quantum systems, and thermalization. It will be a useful reference even if the conjecture is later modified.\n\nI'd send it to peer review. The idea is original and the stakes are high, but the referee should push hard on the stripe construction and demand robustness tests before the conjecture is accepted as established.","headline":"Original and worth engaging, but the stripe width is chosen by maximizing the very GOE diagnostic used to certify chaos — the numerical support for Liouvillian ETH is suggestive, not yet convincing.","tokens_in":15394,"tokens_out":2301,"would_cite":true,"duration_ms":17754,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81Q50","82C10","15B52","81S22"],"pacs":["03.65.Yz","05.30.-d","05.45.Mt"],"model":"deepseek-v4-flash","headline":"The paper extends the eigenstate thermalization hypothesis to open quantum systems, conjecturing that local superoperators on Liouvillian stripes obey ETH statistics with $1/\\mathcal{D}$ fluctuations and Gaussian off-diagonal elements.","keywords":["eigenstate thermalization hypothesis","open quantum systems","Liouvillian stripes","quantum chaos","random matrix theory","Lindblad master equation","thermalization","dissipative many-body systems"],"falsifier":"A decisive test would fix the stripe width by an a priori rule—say $d = c/\\mathcal{D}^\\alpha$ for some constants $c,\\alpha$—and then check in a chaotic driven-dissipative spin chain whether the diagonal variance still scales as $1/\\mathcal{D}$ and the off-diagonal elements remain Gaussian; if those signatures appear only at the specially tuned width that maximizes $\\langle r\\rangle$, the conjecture fails as a predictive statement.","tokens_in":14444,"feed_emoji":"⚛️","tokens_out":13147,"duration_ms":113214,"temperature":0.7,"pith_summary":"This paper claims that open quantum systems governed by a Lindblad equation can satisfy an eigenstate thermalization hypothesis (ETH) of their own. The proposed statement is that local superoperators, when restricted to a 'Liouvillian stripe'—a spectral slice of the non-Hermitian Liouvillian at nearly fixed decay rate—have matrix elements with the same structure as ETH for Hermitian Hamiltonians: a smooth diagonal part plus Gaussian off-diagonal fluctuations whose variance scales as $1/\\mathcal{D}$, where $\\mathcal{D}$ is the Liouville-space dimension. The authors substantiate this with exact diagonalization of random Liouvillians and of a driven-dissipative XXZ spin chain, comparing chaotic and integrable regimes. If the conjecture holds, it gives a statistical-mechanics foundation for thermalization in Markovian open systems and predicts that chaotic dissipation strips local probe responses of structure, which matters for controlling and measuring many-body open quantum devices.","feed_headline":"Open quantum systems get their own thermalization law","feed_subtitle":"New conjecture extends eigenstate thermalization to Lindblad dynamics and explains why probe signals lose structure.","key_machinery":"The Liouvillian stripe is the organizing object: a vertical slice of the complex spectrum obtained by fixing the real part of the eigenvalues to $\\Gamma$ within a narrow width $d_{\\max}$. Gauging away this common decay rate leaves purely imaginary eigenvalues $i\\Omega_\\alpha$, so each stripe can be reinterpreted as a pseudo-Hermitian effective Hamiltonian $\\hat{H}_{\\mathrm{eff}}(\\Gamma)$ with real eigenenergies; the stripe width is chosen by maximizing the one-dimensional level-spacing ratio $\\langle r\\rangle$ over $d$, which yields GOE-like statistics in chaotic systems and Poisson statistics in integrable ones. The conjectured identity of Eq. (6) is then the statement that local superoperators—coherent perturbations $\\hat{O}_{\\mathrm{coh}}\\hat{\\rho} = [\\hat{O},\\hat{\\rho}]$ and measurement superoperators $\\hat{O}_{\\mathrm{m}}\\hat{\\rho} = \\hat{O}\\hat{\\rho}\\hat{O}^\\dagger$—behave on these eigenstates like ETH operators on the eigenstates of a Hermitian chaotic Hamiltonian.","core_discovery":"The central discovery is a conjectured ETH for Liouvillians, stated in Eq. (6): within a stripe of nearly equal decay rate $\\Gamma$, the matrix elements of a local superoperator between right eigenstates read $O_{\\alpha\\beta} = \\mathcal{O}(E,\\Gamma)\\,\\delta_{\\alpha\\beta} + e^{-S(E)/2} f(E,\\omega,\\Gamma)\\,R_{\\alpha\\beta}$, with $R_{\\alpha\\beta}$ Gaussian random numbers. The paper reports numerical evidence that in chaotic systems the diagonal variance and the off-diagonal variance both decrease as $1/\\mathcal{D}$, that off-diagonal elements are Gaussian, and that the smooth functions $\\mathcal{O}$ and $f$ genuinely depend on the stripe decay rate. It also provides the contrasting integrable case, where the variances stay roughly constant and the off-diagonal distribution is non-Gaussian, and it connects the conjecture to dynamics by showing that stripe-restricted observables in the chaotic chain decay without coherent oscillations, while the integrable chain keeps oscillating because only few eigenstates participate.","pith_inferences":["Because the smooth functions $\\mathcal{O}(E,\\Gamma)$ and $f(E,\\omega,\\Gamma)$ depend on the stripe decay rate, the same local observable could thermalize differently in different dissipative environments, making decay rate an extra thermodynamic control parameter.","A fully predictive version of the conjecture requires a principled, system-size-dependent choice of stripe width; without one, the random-matrix signatures inside a stripe are partly a product of the construction.","Engineered dissipation could be used deliberately to erase coherent response features of a many-body system, suggesting a control strategy for hiding or protecting local information from readout.","Extending the free-probability formulation of Hermitian ETH to this superoperator setting would sharpen the conjecture from first and second moments to the full distribution of matrix elements."],"forward_implications":["In a chaotic open system, local observables decay toward the steady state with suppressed coherent oscillations, because the many frequencies inside a stripe interfere destructively.","Coherent or measurement-based probes of a chaotic dissipative system produce structureless response spectra, so individual Liouvillian eigenfrequencies and decay rates become hard to extract from local signals.","The $1/\\mathcal{D}$ scaling of diagonal and off-diagonal variances gives a practical, observable-based diagnostic of quantum chaos in dissipative many-body systems, supplementing spectral level statistics.","Integrable open systems are expected to violate the conjecture, keeping a finite variance and non-Gaussian off-diagonal distributions at large system size."],"supporting_citations":[{"why":"states the Hermitian ETH Ansatz that Eq. (6) generalizes to superoperators.","marker":"[8]"},{"why":"introduces the complex spacing ratio used to classify Liouvillian spectra and to select stripe widths.","marker":"[14]"},{"why":"establishes ETH for non-Hermitian Hamiltonians, the template for the stripe's pseudo-Hermitian Hamiltonian.","marker":"[22]"},{"why":"provides the random-matrix spectral statistics for regular versus chaotic dissipative motion.","marker":"[26]"},{"why":"supplies the universal spectra of random Lindblad operators used as the first testbed.","marker":"[30]"},{"why":"gives the random-Liouvillian construction and steady-state properties underlying the numerical ensemble.","marker":"[31]"},{"why":"shows the locally perturbed XXZ chain is chaotic and satisfies Hermitian ETH, motivating the spin-chain model.","marker":"[32]"},{"why":"provides the exact Bethe-ansatz spectrum identifying the integrable limit of the spin chain.","marker":"[34]"}],"fun_headline_variants":["Open quantum systems get their own ETH conjecture","Liouvillian stripes extend thermalization law to open systems","ETH for open quantum systems via Liouvillian spectral stripes","Thermalization in open systems: a new Liouvillian ETH","Quantum chaos and thermalization meet in open systems"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Liouvillian spectrum can be sliced into vertical stripes of nearly constant decay rate, with each stripe acting as a bona fide pseudo-Hermitian Hamiltonian with real energies; since the stripe width is chosen after the fact by maximizing the level-spacing ratio, the random-matrix statistics observed inside a stripe are partly selected by that choice rather than predicted independently.","fun_headline_variants_meta":{"raw":{"variants":["Open quantum systems get their own ETH conjecture","Liouvillian stripes extend thermalization law to open systems","ETH for open quantum systems via Liouvillian spectral stripes","Thermalization in open systems: a new Liouvillian ETH","Quantum chaos and thermalization meet in open systems"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000179,"raw_usage":{"total_tokens":1281,"prompt_tokens":907,"completion_tokens":374,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":523,"completion_tokens_details":{"reasoning_tokens":296}},"tokens_in":523,"tokens_out":374,"duration_ms":4023,"temperature":1.0,"reasoning_tokens":296,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:33:50.739032+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive test would fix the stripe width by an a priori rule—say $d = c/\\mathcal{D}^\\alpha$ for some constants $c,\\alpha$—and then check in a chaotic driven-dissipative spin chain whether the diagonal variance still scales as $1/\\mathcal{D}$ and the off-diagonal elements remain Gaussian; if those signatures appear only at the specially tuned width that maximizes $\\langle r\\rangle$, the conjecture fails as a predictive statement.","supporting_citations":[{"cited_title":"Singha Roy, S","cited_arxiv_id":null,"evidence_quote":"establishes ETH for non-Hermitian Hamiltonians, the template for the stripe's pseudo-Hermitian Hamiltonian."},{"cited_title":"Grobe, F","cited_arxiv_id":null,"evidence_quote":"provides the random-matrix spectral statistics for regular versus chaotic dissipative motion."},{"cited_title":"Denisov, T","cited_arxiv_id":null,"evidence_quote":"supplies the universal spectra of random Lindblad operators used as the first testbed."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the random-Liouvillian construction and steady-state properties underlying the numerical ensemble."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the exact Bethe-ansatz spectrum identifying the integrable limit of the spin chain."}],"review_version":1}