{"id":"8547aee3-64c1-4829-95dd-0bc4af3db84e","arxiv_id":"2506.13370","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"For degenerate spectra, thermalization of all initial states is equivalent to two conditions on observable averages inside each degenerate subspace; failure of the second condition marks symmetry-breaking equilibrium states.","lead":"This paper proposes an upgraded version of the eigenstate thermalization hypothesis for quantum systems whose energy levels are degenerate due to symmetries. It identifies two distinct ways thermalization can fail, one tied to missing chaos and one tied to symmetry breaking, and tests both on a small Bose-Hubbard model.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The closest-energy pairing used to build H_n assumes approximate degeneracy is exact; at finite N the r=1 component cannot contribute stationary order-parameter values, so the three-branch phase classification rests on an unverified dynamical equivalence.","rationale":"The reader's weakest assumption was the approximate-degeneracy pairing in Appendix C; my analysis confirms that as the load-bearing point. The theorem itself is internally consistent under exact degeneracy and the Srednicki off-diagonal assumption, so the central proof is not the main risk. The risk is the application to the Bose-Hubbard numerics: at finite N the r=1 state is not degenerate with the complex pair, and the infinite-time average of Î in the exact dynamics only retains matrix elements within the exactly degenerate pair. Thus the three-branch eigenvalues computed in the 3D subspace are not automatically stationary expectation values. This does not necessarily destroy the qualitative phase picture, because the exactly degenerate 2D subspace may already yield nonzero λ for the order parameters, but it does undermine the specific three-branch structure and the quantitative phase boundaries claimed from N=320 data. Since the reader already conditioned acceptance on justifying the pairing and providing scaling analysis, my concern does not move the verdict; it sharpens the required check by specifying an exact diagonal-ensemble comparison that would settle whether the approximate degeneracy is dynamically equivalent to an exact one.","tokens_in":15517,"tokens_out":19395,"duration_ms":206694,"concrete_test":"At E/N≈−4.2 and N=320, construct the three Appendix-C branch states |φ_j> (normalized eigenvectors of Î restricted to each triple). Compute their exact diagonal-ensemble average ⟨Î⟩_DE = Σ_{E_a=E_b} ⟨φ_j|E_a⟩⟨E_b|φ_j⟩⟨E_a|Î|E_b⟩ using the true Hamiltonian eigenbasis, and compare with the three eigenvalues λ_j; repeat for N=160,240,320. If ⟨Î⟩_DE does not approach the λ_j values as N grows (or if the three values collapse because only the exactly degenerate r=ω/r=ω² pair contributes), the closest-energy pairing is not dynamically equivalent to an exact degeneracy, and the reported phase boundaries and three-branch structure are unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The formal theorem in Appendix A requires H_n to be spanned by exactly degenerate Hamiltonian eigenstates; only then are the eigenvalues of O restricted to H_n stationary long-time averages. In the numerical construction (Appendix C), H_n is formed by taking the exactly degenerate r=e^{2πi/3} and r=e^{4πi/3} pair and adding the r=1 eigenstate with the closest energy. At every finite N these three states are not degenerate, so the 3×3 restriction of the order parameter is not an invariant subspace of the dynamics. In particular, for the imbalance Î, which changes the rotation quantum number, the only matrix elements that survive the infinite-time average are those within the exactly degenerate complex pair; coherences involving the r=1 state oscillate at the nonzero splitting Δ_n and average to zero. The paper provides no estimate of Δ_n, no check that Δ_n is small compared with the relevant dynamical scales, and no scaling test showing that the Appendix-C triples converge to exact degeneracy. If the r=1 contribution to the reported eigenvalues λ(I)_{n,α} is not dynamically stationary, the 'three branches' in Fig. 3(a) and the inferred symmetry-breaking phase regions are not established by the numerics.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a generalized eigenstate thermalization hypothesis (ETH) for isolated quantum systems with degenerate energy subspaces. It states that every sufficiently narrow initial state thermalizes if and only if two conditions hold for any physical observable: the trace of the observable over each degenerate subspace, divided by the subspace dimension, approaches the microcanonical average (Eq. (2)), and the spread of the observable's eigenvalues inside each subspace vanishes (Eq. (3)). The failure of Eq. (2) is attributed to non-chaotic behavior, while the failure of Eq. (3) for an order parameter is interpreted as symmetry breaking; if Eq. (3) fails on an entire spectral region, that region is identified as a symmetry-breaking phase. The proof is presented in Appendix A under a Srednicki-type exponential decay of off-diagonal matrix elements between different energy subspaces. The formalism is then applied numerically to the three-site Bose-Hubbard model with D3 symmetry, using exact diagonalization and classical trajectories to identify thermalizing, rotation-breaking, reflection-breaking, and mixed spectral regions.","tokens_in":15753,"tokens_out":12664,"duration_ms":128750,"significance":"If the proposed criteria are validated, the paper offers a clean, basis-independent framework for understanding thermalization in the presence of degeneracies and connects failure of the generalized ETH to symmetry breaking. The Appendix A theorem is simple and appears logically sound under its stated assumptions, and the model study is informative, with a useful classical-quantum comparison. The main value would be in providing a concrete diagnostic for where symmetry-breaking equilibrium states can arise. However, the phase-level claims go beyond what the theorem establishes, and the numerical construction of degenerate subspaces relies on an approximate degeneracy that is not validated; the current evidence is finite-size and does not yet justify the thermodynamic-limit phase classification.","major_comments":[{"comment":"The phase criterion 'a spectral region is a symmetry-breaking phase if and only if it has no subspaces fulfilling Eq. (3)' is not established by Appendix A. The backward direction of the theorem (Appendix A, Eqs. (A6)-(A7)) only proves that when Eq. (3) fails there exists at least one initial state, namely an eigenstate |~E_m,alpha>, whose long-time order-parameter average is non-thermal. It does not show that all, or typical, initial conditions in that spectral region equilibrate to one of the lambda branches; the statement that 'every initial condition equilibrates in one of these branches, or in a quantum superposition of them' requires an additional argument. As written, an initial state with comparable weights on several branches could have a vanishing long-time average even when each subspace violates Eq. (3). The numerical evidence in Fig. 3 concerns classical trajectories and eigenvalue distributions, not quantum quenches from generic narrow initial states.","section":"Main text, after Eq. (3): phase criterion"},{"comment":"The construction of the three-dimensional subspaces H_n is not justified by the theorem. The exactly degenerate pair with r=e^{2pi i/3} and r=e^{4pi i/3} is combined with the r=1 eigenstate whose energy is closest to that pair, but at any finite N this r=1 state is not degenerate with the pair. The theorem in Appendix A applies to exactly degenerate subspaces; for a nonzero splitting Delta_n, coherences between the r=1 state and the complex pair oscillate at frequency Delta_n/hbar and vanish in the infinite-time average, so the eigenvalues lambda(I)_{n,alpha} obtained from the 3x3 matrix are not stationary long-time averages. The manuscript provides no estimate of Delta_n, no comparison of Delta_n with relevant dynamical scales, and no scaling test showing that the triples converge to exact degeneracy in the thermodynamic limit. Therefore the three-branch structure in Fig. 3(a) and the inferred rotation-breaking phase in Fig. 1(b) are not established by the numerical data.","section":"Appendix C, 'Eigenvalues lambda(O)_n,alpha and traces T(O)_n'"},{"comment":"The phase classification is based on finite-size data without scaling analysis in the symmetry-breaking regions. Fig. 1 reports max_alpha |lambda(I)_{n,alpha}| and max_alpha |lambda(C)_{n,alpha}| for N=320 only, and Fig. 3 uses N=320; the finite-size scaling shown in Fig. 2 is limited to the thermalizing region. Since Eq. (3) is an asymptotic condition in the thermodynamic limit, observing that lambda is 'significantly different from zero' in some energy windows at N=320 does not distinguish a true symmetry-breaking phase from finite-size remnant order. The authors should provide scaling of the violations of Eq. (3) in each proposed phase, for example the maximum or the fraction of subspaces with |lambda| above a threshold as a function of N, or else state explicitly that the phase boundaries are finite-size observations rather than thermodynamic-limit conclusions.","section":"Figs. 1-3: finite-size evidence for phase regions"}],"minor_comments":[{"comment":"The proof assumes that the observable can be diagonalized within each degenerate subspace by a transformation that preserves an orthonormal eigenbasis; for the non-Hermitian but normal operator hat I this is valid, but the manuscript should state this, since Eq. (3) is applied to a non-Hermitian order parameter.","section":"Appendix A, Eq. (A1)"},{"comment":"The statement that full chaos 'immediately' implies Eq. (2) should be qualified: ETH per symmetry sector refers to the sector-resolved microcanonical average, and one must assume that the relevant sector averages coincide with the global O_ME(E) for the trace condition to follow. The numerical test in Fig. 1(a) is therefore important and should be emphasized.","section":"S1, after Eq. (2)"},{"comment":"The fitted exponent for T(h12) is -0.419, which is not very close to the Srednicki expectation of -1/2; the sentence calling this 'close to Srednicki's ansatz' should quantify the expected finite-size corrections or be softened.","section":"Fig. 2(a)"},{"comment":"The caption of Fig. 2(b)-(d) says the number of subspaces is counted for differences 'below a given arbitrary bound', while the text says 'larger than several bounds'; these should be made consistent.","section":"Fig. 2 caption and text"},{"comment":"The phrase 'we rule out remaining degeneracies by assuming that off-diagonal matrix elements between them fulfill Srednicki's ansatz' is ambiguous; it should say explicitly that the exponential decay assumption applies to matrix elements between different energy subspaces En != Em.","section":"Introduction, generalized ETH setup"}],"recommendation":"major_revision","confidential_remarks":"The core theorem is sound under its stated assumptions, and the paper is likely suitable for publication after the numerical construction is validated. The main risk is the closest-energy pairing in Appendix C: if the r=1 partner is not dynamically stationary, the reported phase regions do not follow from the numerics. This is fixable by reporting the splittings and scaling tests, or by weakening the phase claims to finite-size observations."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know: the two-condition generalized ETH (Eqs. 2-3) is a clean and useful formulation, and the Appendix A proof is right under the stated Srednicki ansatz. The numerical model is a nice illustration, but the phase-level conclusions rest on two assumptions the paper doesn't check.\n\nWhat's new: the iff characterization. For degenerate subspaces, thermalization of all narrow initial states requires both the trace condition (2) and the intra-subspace eigenvalue spread condition (3). That is a genuine generalization of ETH, cleanly separating two failure modes. The authors correctly credit the NATS literature (Refs 75-82) and their own prior work (Ref 74); the incremental contribution is the compact two-condition criterion and its application to a model with two noncommuting discrete symmetries. The proof is logically sound: the forward direction is straightforward, and the backward direction correctly constructs non-thermalizing initial states for each violated condition.\n\nMain soft spot is the construction of H_n in Appendix C. The theorem requires exact degeneracy within each subspace. The numerics build triples by taking the exactly degenerate r=e^{2πi/3} and r=e^{4πi/3} pair and adding the r=1 eigenstate with closest energy. At finite N these three states are not degenerate, and the paper gives no estimate of the splitting Δ_n, no scaling test, and no argument that Δ_n is dynamically irrelevant. For the imbalance I, coherences between the r=1 state and the complex pair oscillate at Δ_n and average to zero, so the eigenvalues of the 3x3 restriction are not obviously the stationary long-time values. The paper's own assumption \"for all practical purposes\" is plausible for large N in a symmetry-broken phase, but N=320 is not the thermodynamic limit, and the claim that no subspaces fulfill Eq. (3) is only as good as this pairing.\n\nSecond soft spot: the theorem's backward direction only guarantees the existence of non-thermalizing initial states when Eq. (3) fails. The paper then calls a whole spectral region a symmetry-breaking phase when Eq. (3) fails everywhere. That inference is stronger than the theorem: failure everywhere could still leave some eigenstates of the order parameter with zero eigenvalue, giving symmetric equilibrium states. The classical trajectories and the max-|λ| plots suggest all branches are nonzero in the fully broken region, but the paper does not show that every λ is bounded away from zero. So the phase-level \"iff\" is a conjecture supported by numerics, not a proven consequence.\n\nThese are not fatal to the theoretical result. The theorem stands; the numerical boundaries are provisional. The paper would be strengthened by reporting the typical splitting Δ_n as a function of N, showing the λ-eigenvalues converge to the exact-degeneracy values, and either proving or softening the phase-level claim.\n\nWho is this for? People working on ETH, thermalization, symmetry breaking in isolated systems, and quantum simulators with discrete symmetries. The generalized ETH is a useful organizing criterion. The numerical example is a good testbed but not a conclusive phase diagram.\n\nI would send this to a serious referee. The theorem deserves publication even if the numerical support requires revision. With scaling checks and a more careful phase claim, it would be solid.","headline":"A clean two-condition generalization of ETH with a sound proof, but the finite-size numerical support for the claimed symmetry-breaking phases relies on an unverified near-degeneracy assumption and an inference stronger than the theorem.","tokens_in":16290,"tokens_out":5737,"would_cite":true,"duration_ms":53668,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["05.30.-d","05.45.Mt","03.65.-w"],"model":"deepseek-v4-flash","headline":"Thermalization in degenerate quantum systems reduces to two conditions: observable averages over each subspace match microcanonical values, and intra-subspace eigenvalue spreads vanish; failure of the second on an order parameter produces…","keywords":["eigenstate thermalization hypothesis","degenerate energy subspaces","symmetry breaking","quantum thermalization","Bose-Hubbard model","order parameter","quantum chaos","discrete symmetries"],"falsifier":"For the Bose-Hubbard trimer at $N \\approx 320$, time-evolve a narrow energy shell initial state in the claimed thermalizing region ($-3.1 \\lesssim E/N \\lesssim -2.3$) and check whether the long-time average of $\\hat{h}_{12}/N$ equals the classical microcanonical value; a persistent deviation beyond the reported power-law width would falsify Eq. (2). Alternatively, repeat the eigenvalue extraction while keeping the $r=1$ sector separate from the degenerate pair and see whether the symmetry-breaking regions persist; if they disappear, the pairing assumption is load-bearing and the phase classification collapses.","tokens_in":15283,"feed_emoji":"⚛️","tokens_out":7518,"duration_ms":64452,"temperature":0.7,"pith_summary":"Thermalization of an isolated quantum system is usually attributed to the eigenstate thermalization hypothesis, which assumes non-degenerate spectra. This paper removes that assumption and states a sharper criterion: a system with degenerate energy subspaces thermalizes if and only if, for every observable, the average over each degenerate subspace equals the microcanonical average, and the observable's eigenvalues within each subspace have vanishing spread. The two failures are physically distinct: the first means the system is not chaotic and needs extra conserved quantities, while the second, for an order parameter, means initial states can relax to symmetry-broken equilibrium values; if the second failure covers an entire spectral region, that region is a symmetry-breaking phase. The authors prove the equivalence and illustrate it on a three-site Bose-Hubbard model, where the two conditions separate thermal, rotation-broken, reflection-broken, and mixed spectral regions without running full dynamics.","feed_headline":"Degenerate quantum systems thermalize under two conditions","feed_subtitle":"Failure of the second condition predicts where symmetry-broken equilibrium states and phases appear.","key_machinery":"The degenerate subspace $H_n$ and the restricted observable's eigenvalues $\\lambda^{(O)}_{n,\\alpha}$ form the central object. Each energy shell is decomposed into subspaces spanned by Hamiltonian eigenstates of equal energy, and the observable is diagonalized inside every subspace; the long-time average of the observable is then written in this adapted eigenbasis, where cross-subspace terms are killed by exponential decay, leaving only the $\\lambda$'s weighted by initial-state populations. This decomposition carries the argument: the trace condition controls the mean level, the spread condition controls the fluctuations, and order parameters are recognized by having zero trace but non-zero individual eigenvalues.","core_discovery":"The central claim is a generalization of the eigenstate thermalization hypothesis to Hamiltonians with degenerate energy subspaces. For any physical observable $\\hat O$, let $H_n$ be the subspace of eigenstates sharing energy $E_n$, define $T^{(O)}_n$ as the trace of $\\hat O$ restricted to $H_n$, and let $\\lambda^{(O)}_{n,\\alpha}$ be its eigenvalues there. Then every sufficiently narrow initial state thermalizes if and only if the trace per dimension approaches the microcanonical average (Eq. 2) and the eigenvalue spread inside each subspace vanishes (Eq. 3). Failure of the first condition signals non-chaotic dynamics and additional conserved quantities requiring a generalized Gibbs ensemble; failure of the second, applied to an order parameter, is exactly the appearance of symmetry-breaking equilibrium states, and when it covers a whole spectral region that region is a symmetry-breaking phase. The proof assumes only that off-diagonal matrix elements between different energy subspaces decay exponentially with system size, as in Srednicki's ansatz.","pith_inferences":["The same two-condition test could map symmetry-breaking phases in other lattice models with non-commuting discrete symmetries, such as spin chains with translation and reflection, using only static diagonalization rather than time evolution.","A practical finite-size diagnostic suggested by the paper: in candidate thermal regions the width of the intra-subspace eigenvalue distribution should decay at least as $N^{-1/2}$; regions where it saturates or decays more slowly are symmetry-breaking candidates.","If failure of the vanishing-spread condition occurs only for a vanishing fraction of subspaces, the formalism predicts scar-like long-lived oscillations, giving a sharper, quantitative definition of many-body scars in symmetry-degenerate systems.","The near-degeneracy pairing assumption could be tested directly by comparing the eigenvalue extraction against full diagonalization in the $r=1$ sector; agreement would confirm the phase boundaries, disagreement would shift them with system size."],"forward_implications":["In any system whose degenerate subspaces satisfy both conditions, every sufficiently narrow initial state equilibrates to the microcanonical ensemble, so thermalization can be certified by diagonalizing observables inside degenerate subspaces rather than simulating full dynamics.","Analyzing an order parameter's eigenvalues inside degenerate subspaces gives a direct spectral diagnostic for symmetry-breaking phases: a region where no subspace satisfies the vanishing-spread condition is an ordered phase.","Symmetry-breaking equilibrium states need not fill a whole phase: the Bose-Hubbard results show they can also appear as isolated subspaces inside a chaotic sea, playing a role similar to many-body scars.","Non-chaotic regions are flagged by failure of the trace condition, which in the model coincides with extra integrals of motion and signals the need for a generalized Gibbs ensemble description.","Both conditions reduce exactly to standard ETH when degeneracies are absent, so the formalism is a strict generalization of the usual mechanism rather than a separate one."],"supporting_citations":[{"why":"Supplies the exponential decay ansatz for off-diagonal matrix elements that the proof assumes between different energy subspaces.","marker":"[3]"},{"why":"Establishes the standard eigenstate thermalization framework that the new conditions reduce to in the non-degenerate limit.","marker":"[4]"},{"why":"Gives the order-parameter property (zero diagonal expectation in symmetric eigenstates) that makes the trace vanish for symmetry-breaking operators.","marker":"[58]"},{"why":"Provides the prior characterization of discrete symmetry-breaking equilibrium states that the new generalized-ETH criterion extends.","marker":"[74]"},{"why":"Gives the semi-classical level-spacing distribution used to measure the chaotic fraction in the numerical phase classification.","marker":"[92]"}],"fun_headline_variants":["Two conditions decide thermalization in degenerate quantum systems","Degeneracy unlocks symmetry-breaking phases via new ETH","Generalized ETH predicts when symmetry breaks in quantum systems","How degenerate spectra spawn symmetry-breaking phases","New ETH conditions for degenerate quantum thermalization"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The phase classification rests on treating near-degenerate states as exactly degenerate by pairing the $r=1$ eigenstate closest in energy to the exactly degenerate pair, and on assuming exponential decay of off-diagonal matrix elements between different subspaces; if either fails, the reported phase regions are not established.","fun_headline_variants_meta":{"raw":{"variants":["Two conditions decide thermalization in degenerate quantum systems","Degeneracy unlocks symmetry-breaking phases via new ETH","Generalized ETH predicts when symmetry breaks in quantum systems","How degenerate spectra spawn symmetry-breaking phases","New ETH conditions for degenerate quantum thermalization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000167,"raw_usage":{"total_tokens":1218,"prompt_tokens":865,"completion_tokens":353,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":481,"completion_tokens_details":{"reasoning_tokens":283}},"tokens_in":481,"tokens_out":353,"duration_ms":3377,"temperature":1.0,"reasoning_tokens":283,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:04:21.250229+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the Bose-Hubbard trimer at $N \\approx 320$, time-evolve a narrow energy shell initial state in the claimed thermalizing region ($-3.1 \\lesssim E/N \\lesssim -2.3$) and check whether the long-time average of $\\hat{h}_{12}/N$ equals the classical microcanonical value; a persistent deviation beyond the reported power-law width would falsify Eq. (2). Alternatively, repeat the eigenvalue extraction while keeping the $r=1$ sector separate from the degenerate pair and see whether the symmetry-breaking regions persist; if they disappear, the pairing assumption is load-bearing and the phase classification collapses.","supporting_citations":[{"cited_title":"Rela˜ no, Thermalization in an interacting spin sys- tem in the transition from integrability to chaos, J","cited_arxiv_id":null,"evidence_quote":"Gives the order-parameter property (zero diagonal expectation in symmetric eigenstates) that makes the trace vanish for symmetry-breaking operators."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the prior characterization of discrete symmetry-breaking equilibrium states that the new generalized-ETH criterion extends."}],"review_version":2}