{"id":"b886881d-ff17-4966-93bf-5e46ee23ddb1","arxiv_id":"1908.03997","paper_version":5,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A commutativity-graph method produces Lieb-Robinson bounds whose velocities scale as sqrt(S) or sqrt(N) rather than linearly in the local Hilbert space dimension, and stay finite in several classical large-spin limits.","lead":"This paper introduces a graphical method, the commutativity graph, that yields much tighter Lieb-Robinson bounds for locally interacting quantum systems by exploiting which Hamiltonian terms commute. The new bounds give velocities that scale better with local Hilbert space dimension and spatial dimension than all previous bounds, with implications for quantum dynamics and numerical error bounds.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (59)'s enlarged Hamiltonian for spin-S Heisenberg sums only over a<b, omitting terms needed for the spin decomposition; as written it is not permutation-invariant and the transfer of the LR bound to the physical subspace fails.","rationale":"After reviewing the paper, the reader's weakest-assumption (transfer from enlarged Hilbert space) is the right place to look, but the difficulty is not leakage or norm factors: for the intended permutation-invariant enlarged Hamiltonian, the symmetric subspace is invariant and the transfer argument is sound, and extra norm factors enter only the constant C, not the LR velocity. The concrete defect is that Eq. (59) as printed sums over a<b. This is internally inconsistent with the decomposition preceding it and with the differential equations that follow, which effectively sum over all ordered pairs. I therefore partially agree with the reader: the transfer step is the load-bearing point, but the specific failure mode is a wrong summation range in the defining Hamiltonian, not an unproven invariance assertion. The concern is concrete and easily settled by checking S=1/2. A correction to Eq. (59) will very likely preserve the headline scaling, so the conditional verdict stands. Other flagged issues (Jordan-form bound in Eq. (29), sketched Theorem 1, small-t belief) are repairable or secondary: the Jordan-form polynomial prefactor does not affect the exponential velocity, and the small-t 'tightest possible' claim is not needed for the central scaling results. Thus the appropriate verdict remains CONDITIONAL (unchanged), contingent on the authors correcting Eq. (59) and confirming the subsequent summation.","tokens_in":35761,"tokens_out":17596,"duration_ms":181909,"concrete_test":"Instantiate S=1/2 and S=1: for S=1/2, the sum 1≤a<b≤2S in Eq. (59) is empty, yielding an enlarged Hamiltonian of zero, which cannot equal the physical Heisenberg Hamiltonian (nonzero exchange). More generally, evaluate the commutator norm transferred from the enlarged space to the physical triplet subspace for S=1 with the Hamiltonian as written; if the enlarged Hamiltonian does not map the symmetric subspace to itself, recompute the LR bound using the corrected Hamiltonian with the sum over all ordered pairs a,b=1,...,2S and verify that Eq. (63) (or its corrected version) still yields a finite velocity as S→∞.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central large-S Heisenberg claim (Eq. (63), finite v as S→∞) relies on decomposing spin-S operators into 2S spin-1/2 Pauli operators and deriving the LR bound in the enlarged Hilbert space. The transfer to the physical subspace (Sec. V A 2, after Eq. (53)) is valid only if the enlarged Hamiltonian leaves the symmetric subspace invariant. But the enlarged Hamiltonian written in Eq. (59) sums over 1≤a<b≤2S, which does not reproduce the physical Hamiltonian: the full spin product S^α_i S^α_j = (1/4)Σ_{a,b} σ^α_{i,a}σ^α_{j,b} requires all ordered pairs. As written, the Hamiltonian is not permutation-invariant at each site, so the physical (fully symmetric) subspace is not invariant; the norm of the enlarged-space commutator then need not dominate the physical commutator, and the bound does not transfer. The inconsistency is visible already for S=1/2 (2S=1), where Eq. (59) gives an empty sum and hence H=0 for a nonzero physical model, and for S=1, where only the a=1,b=2 terms are kept. The subsequent differential equations Eqs. (60)-(61) and the factor 2(2-1/S) in Eq. (61) appear to be derived from the Hamiltonian with the sum over all a,b, so the most plausible reading is a typo. Nevertheless, as the text stands, the derivation of the flagship Heisenberg result contains a false intermediate statement, and Eq. (63) cannot be accepted without correcting Eq. (59) and re-deriving the sum.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a method for obtaining Lieb-Robinson bounds that exploit the commutativity structure of the Hamiltonian. The Hamiltonian terms are represented as vertices of a commutativity graph, and the unequal-time commutator is bounded by solutions of linear differential equations on that graph. In translation-invariant systems the equations are solved by Fourier transform, and the Lieb-Robinson velocity is extracted from the maximum eigenfrequency of the Fourier-transformed matrix. The method is applied to the transverse-field Ising model, the spin-S Heisenberg XYZ model, the truncated Bose-Hubbard model, the SU(N) Fermi-Hubbard model, and Wen's quantum rotor model. The central claimed improvements are qualitative: finite velocity as S goes to infinity in the Heisenberg model, sqrt(N) rather than N growth in the SU(N) Hubbard model, sqrt(d) rather than d growth in large spatial dimension, and finite velocities for certain perturbed commuting models. The authors also use the improved bounds to derive tighter exponential clustering results and correlation-length bounds.","tokens_in":36118,"tokens_out":9092,"duration_ms":108926,"significance":"If the proofs are correct, the paper gives a substantial and general advance in Lieb-Robinson bounds. The qualitative improvements in the large-S, large-N, and large-d limits address long-standing limitations of previous bounds, and the claims are supported by independent exact checks at free-fermion and transverse-field-Ising points. A notable strength is that no parameters are fitted: the constants X_y and Z_y come from closed-form minimizations, and the tightness checks compare against exact Bessel-function solutions and the exact TFIM correlation length. The general commutativity-graph framework is likely to be useful beyond the specific examples treated. However, two load-bearing technical points need to be repaired before the central claims can be accepted as stated.","major_comments":[{"comment":"The inequality |[e^{H(i kappa) t}]_{alpha beta}| <= c_kappa e^{omega_m(i kappa) t} with c_kappa independent of t does not follow from diagonalizing H(i kappa) into Jordan canonical form. A nontrivial Jordan block of size m gives a factor t^{m-1} multiplying e^{omega_m t}, so in general c_kappa must be allowed to depend on t polynomially. Unless H(i kappa) is proved diagonalizable for every model used in the paper, Eq. (29) as stated is not a theorem. The velocity formula Eq. (31) is probably salvageable because a polynomial prefactor is subexponential, but the proof needs to be amended, for example by replacing c_kappa with a polynomial in t or by proving diagonalizability in the cases where the formula is applied.","section":"Section III, Eq. (28)"},{"comment":"The enlarged Hamiltonian in Eq. (59) is not the spin-S Heisenberg Hamiltonian obtained from the decomposition S^alpha = (1/2) sum_a sigma^alpha_a. Substitution into Eq. (58) gives H = (1/(2S)) sum_{<ij>, 1<=a,b<=2S} (J_x X^{ab}_{ij} + J_y Y^{ab}_{ij} + J_z Z^{ab}_{ij}) with a sum over all ordered color pairs. The restriction to 1<=a<b<=2S in Eq. (59) omits the diagonal color terms and half of the off-diagonal terms; for S=1/2 the sum is empty and the Hamiltonian vanishes, and for S=1 it does not reproduce the physical model. Because the fully symmetric subspace is invariant only for the all-pairs Hamiltonian, the transfer argument given after Eq. (53) does not apply to Eq. (59) as written. The differential equations Eqs. (60)-(61) appear to have been derived from the all-pairs Hamiltonian, so the most plausible reading is that Eq. (59) contains a typographical error, but this is a load-bearing false intermediate statement: Eq. (63) cannot be accepted until Eq. (59) is corrected and the sums leading to Eqs. (60)-(61) are re-derived from the corrected Hamiltonian.","section":"Section V B, Eq. (59)"},{"comment":"The claim that a Lieb-Robinson bound in the enlarged Hilbert space 'automatically gives a bound on operators acting on the physical Hilbert space' is asserted but not proved. Individual color Pauli operators sigma^alpha_{i,a} do not preserve the fully symmetric subspace; it is only the summed operators such as sum_a sigma^alpha_{i,a} that act within it. The authors should state and prove the transfer lemma explicitly: if H_enh leaves the physical subspace invariant and the physical operators are obtained by projection from the enlarged-space operators, then the physical commutator norm is bounded by the enlarged-space commutator norm. Without such a lemma, the large-S transverse-field Ising and Heisenberg results are not fully established even after Eq. (59) is corrected.","section":"Section V A 2, after Eq. (53)"}],"minor_comments":[{"comment":"The notation sum_{1<=|sigma|<=N} in Eq. (75) is confusing; it appears to sum over the 2N Majorana indices sigma, but the range should be defined explicitly.","section":"Section V D, Eq. (75)"},{"comment":"The caption and table layout mix analytic expressions and numerical values without making clear which rows correspond to which method of elimination; a sentence explaining the row structure would improve readability.","section":"Table I"},{"comment":"The proof of the arbitrary-graph bound uses the inequality n! >= (n/e)^n but does not state the resulting constants for small n; the conclusion is correct, but a brief comment on how the n<d_{ij} terms are absorbed would make the argument easier to follow.","section":"Section IV, Eq. (39)"},{"comment":"The Lambert W function is introduced correctly, but the two regimes in Eq. (103) are not derived in full detail; a short derivation or reference would help the reader verify the crossover at Delta = u.","section":"Section VII, Eq. (103)"}],"recommendation":"major_revision","confidential_remarks":"This is a potentially strong paper whose central claims are interesting and largely well supported by independent exact checks. The false statement in Eq. (59) appears to be a typo rather than a fundamental flaw, and the Jordan-form issue in Eq. (28) is fixable without changing the velocity formulas. I would recommend major revision rather than rejection, provided the authors correct Eq. (59), add the missing transfer lemma, and amend the matrix-exponential bound or prove diagonalizability in the cases used. The paper's scope and methods are well matched to the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe one thing to know: the commutativity graph is a genuinely new tool for Lieb-Robinson bounds, and it produces velocity scalings that previous methods cannot reach. It deserves a serious referee, but the large-S Heisenberg section contains an equation that is simply wrong as written, and that needs fixing before the headline claim in that model is accepted.\n\nWhat is actually new: encoding the commutation relations of Hamiltonian terms as a graph and bounding unequal-time commutators by the solution of linear differential equations on that graph. That is a real abstraction, not a routine extension. Concretely, it gives v ~ sqrt(d) for the TFIM, v ~ sqrt(N) for SU(N) Fermi-Hubbard, a finite velocity for spin-S Heisenberg as S→∞, and an order-of-magnitude improvement in standard spin-1/2 cases. The Fourier integral representation and the superexponential tail are nice, and the checks against exact free-fermion and TFIM solutions provide genuine evidence that the method is not just variational looseness.\n\nThe soft spots, in proportion. The most serious is Eq. (59). The enlarged Hamiltonian for spin-S Heisenberg sums only over 1≤a<b≤2S, but the spin decomposition S^alpha = 1/2 sum_a sigma^alpha_a requires the product S_i^alpha S_j^alpha to contain all ordered pairs (a,b) across the two sites. As written, the Hamiltonian is not permutation-invariant, so the physical symmetric subspace is not invariant; the transfer argument fails. The inconsistency is visible already for S=1/2, where the sum is empty and the Hamiltonian vanishes. This looks like a typo—Eqs. (60)-(61) appear to be derived from the full ordered-pair sum—but it is a false intermediate statement in the flagship derivation. Eq. (63) cannot be accepted until Eq. (59) is corrected and the factors re-derived.\n\nThe other concerns are smaller. The bound in Eq. (29) uses a matrix exponential estimate that assumes diagonalizability; the Jordan-form issue is acknowledged but not handled. Theorem 1 is a sketch rather than a proof. And the claim that the minimal Clifford decomposition gives the tightest small-t exponent is an expectation, not a theorem. None of these undercut the core method; they are places where the paper overreaches. The citation pattern looks fair, and the dependence on Ref. [53] is properly noted.\n\nWho this is for: people working on Lieb-Robinson bounds, quantum dynamics, and numerical error bounds for local simulation. It is a methods paper with broad applications. I would bring it to a reading group and would cite it if I worked in this area.\n\nRecommendation: send to peer review. The central method is solid and the results are important, but the Heisenberg derivation needs a serious correction, and the referees should check the enlarged-space decomposition carefully.","headline":"The commutativity graph is a real new tool for LR bounds, but the large-S Heisenberg derivation contains a false Hamiltonian (Eq. (59)) that must be fixed before the flagship finite-velocity claim is taken as proven.","tokens_in":36663,"tokens_out":4056,"would_cite":true,"duration_ms":42500,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["03.65.-w","05.30.-d","75.10.Jm"],"model":"deepseek-v4-flash","headline":"A commutativity-graph method tightens Lieb-Robinson bounds, making light-cone speeds finite where older bounds diverged.","keywords":["Lieb-Robinson bound","light cone","commutativity graph","information propagation","transverse-field Ising model","Heisenberg model","Fermi-Hubbard model","quantum rotor model"],"falsifier":"Examine whether the enlarged-space Heisenberg evolution generated by Eq. (54) maps the fully symmetric subspace to itself for all times; a single explicit counterexample — such as a symmetric initial two-spin state evolving into a state with nonzero antisymmetric component, or a physical operator whose norm grows under projection — would invalidate the S→∞ transfer. Alternatively, compute the exact or high-precision numerical Lieb-Robinson velocity of the S=2 or S=3 Heisenberg XYZ chain and check whether it approaches the finite bound 4dJ_m X_{1−1/(2Sd)} or instead grows with S.","tokens_in":35487,"feed_emoji":"⚛️","tokens_out":6030,"duration_ms":62226,"temperature":0.7,"pith_summary":"This paper introduces a method for deriving Lieb-Robinson bounds that uses the commutation relations among Hamiltonian terms, not just interaction strengths. The central claim is that this 'commutativity graph' method produces light-cone velocities that are dramatically tighter than all previous bounds, and in several limits removes divergences that old bounds exhibited when the local Hilbert space became large. For example, the spin-S Heisenberg XYZ model is shown to have a finite Lieb-Robinson velocity as S→∞, the SU(N) Fermi-Hubbard model has velocity growing like √N instead of N, and the d-dimensional transverse-field Ising model has velocity growing like √d instead of d. The paper also proves bounds with superexponential spatial tails and short-time exponents that match exact results, and uses the improved speeds to tighten ground-state correlation length bounds in gapped systems.","feed_headline":"Quantum light-cone speeds stop diverging as spin grows","feed_subtitle":"Commutativity graphs tighten Lieb-Robinson bounds for Heisenberg, Hubbard, Ising, and rotor models.","key_machinery":"The commutativity graph: a graph whose vertices are the unit-norm local Hermitian operators gamma_i in a chosen decomposition of the Hamiltonian, with an edge between i and j exactly when [gamma_i, gamma_j] ≠ 0. The argument's engine is Eq. (10), a first-order linear differential equation for the norm of the commutator [gamma_i(t), B(0)], whose coefficients are the coupling strengths on the graph edges; its solution upper-bounds the commutator via a Gronwall inequality. In translation-invariant systems the equation diagonalizes in Fourier space, and the bound on the Green's function becomes an integral whose large-momentum behavior yields the speed formula v_LR ≤ min_kappa omega_m(i kappa)/kappa. A power-series solution in arbitrary graphs gives the superexponential bound C (u|t|/d_XY)^{d_XY}. The paper also uses a minimal Clifford decomposition (every pair of terms commutes or anticommutes) to saturate the triangle inequality, and a graph-reduction step that removes mutually commuting terms with large coefficients by rotating them away unitarily.","core_discovery":"The paper's central discovery is that the speed of information propagation in a locally interacting quantum system is controlled by the largest eigenvalue of a linear operator built from the Hamiltonian's commutativity graph. Each local term of the Hamiltonian becomes a vertex, and edges connect non-commuting terms. The norm of an unequal-time commutator is bounded by the solution of a first-order linear differential equation on this graph; in translation-invariant systems the solution is a Fourier integral, and the Lieb-Robinson velocity is the minimum over imaginary momentum shifts of omega_m(i kappa)/kappa. Applying this recipe to concrete models yields the scalings summarized in the abstract and Table II: finite velocity at large S for Heisenberg XYZ, v ∝ √N for SU(N) Fermi-Hubbard, v ∝ √d for TFIM, and similar improvements for truncated Bose-Hubbard and Wen's rotor model. A separate graph-reduction step removes large commuting coupling constants, giving finite speeds in large-J, large-h, and large-U limits and a theorem that perturbed exactly solvable models have velocity at most linear in the perturbation strength.","pith_inferences":["The same machinery should yield model-specific power-law LR bounds for commuting long-range interactions such as Coulomb, dipolar, or Rydberg interactions, where the paper only sketches the generalization; if the large-commuting-block advantage persists, light cones in those systems could be much tighter than current generic bounds.","Because the differential equations are linear and few in number, the method can be used as a numerical routine to certify error bars in small quantum simulations, not just as an analytic proof tool.","The finite large-S velocity for Heisenberg XYZ suggests that cold-atom or ion-trap simulators using large effective spins should exhibit light-cone spreading at a speed nearly independent of S, a prediction that could be checked in experiment.","The criterion that a semiclassical approximation predicting a speed above the bound is invalid could be turned into a general diagnostic for mean-field and truncated-Wigner methods."],"forward_implications":["The spin-S Heisenberg XYZ model has an upper bound on its Lieb-Robinson velocity that remains finite as S→∞, so information propagation speed does not diverge with local Hilbert space size.","The SU(N) Fermi-Hubbard model has LR velocity growing like √(NdUJ) rather than linearly in N, so the large-N classical limit has finite speed.","The d-dimensional transverse-field Ising model has velocity growing like √(dJh), the first sublinear-in-dimension LR bound; combined with large-coupling eliminations it is bounded by min{2X0√(dJh), 4X_{d−1/d}dJ, 4X0dh}.","The bounds have superexponential tails (u|t|/d_XY)^{d_XY} and tight short-time exponents, improving the large-distance and early-time behavior over previous bounds.","For perturbed exactly solvable commuting models such as the toric code with a small local perturbation, the LR velocity vanishes linearly with perturbation strength, and the ground-state correlation length vanishes at least as 1/|ln J| as the perturbation goes to zero."],"supporting_citations":[{"why":"The original Lieb-Robinson bound establishing finite group velocity in quantum spin systems; the object this paper tightens.","marker":"[1]"},{"why":"Supplies the previous best LR velocities and methods that this paper improves on, including O(S), O(N), and O(d) scalings.","marker":"[2]"},{"why":"Hastings-Koma derivation of exponential decay of correlations from LR bounds; the proof template replaced in Sec. VII.","marker":"[14]"},{"why":"A special LR bound for a truncated rotor model that motivated the commutativity-graph idea.","marker":"[53]"},{"why":"Generalized Gronwall inequality used to convert integral inequalities into the linear differential equations.","marker":"[63]"},{"why":"Semiclassical large-S Heisenberg dispersion used to benchmark the finite large-S velocity.","marker":"[68]"},{"why":"A graph-theoretic operator growth bound for the 1D TFIM that is compared against and combined with the present method.","marker":"[77]"},{"why":"The rotor model derivation of emergent light speed v=√(2gJ), used as the semiclassical benchmark in Sec. V E.","marker":"[72]"},{"why":"Exact ground-state solution of the 1D TFIM used to benchmark the correlation-length bound.","marker":"[81]"}],"fun_headline_variants":["Graph eigenvalues cap LR speed in many models","Heisenberg spin no longer boosts light cone speed","Hubbard velocity √N instead of N in LR bound","Tighter LR bound: spin speed from S to constant","Commutativity graph tightens Lieb-Robinson velocity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"For the large-spin results the paper decomposes each spin-S operator into 2S spin-1/2 Pauli operators and asserts that bounds derived in the enlarged Hilbert space transfer unchanged to the physical symmetric subspace; this transfer is stated without a detailed proof, and if the enlarged evolution leaked out of the physical subspace or introduced extra norm factors, the claimed finite large-S velocity would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Graph eigenvalues cap LR speed in many models","Heisenberg spin no longer boosts light cone speed","Hubbard velocity √N instead of N in LR bound","Tighter LR bound: spin speed from S to constant","Commutativity graph tightens Lieb-Robinson velocity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000881,"raw_usage":{"total_tokens":3860,"prompt_tokens":1051,"completion_tokens":2809,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":667,"completion_tokens_details":{"reasoning_tokens":2731}},"tokens_in":667,"tokens_out":2809,"duration_ms":31209,"temperature":1.0,"reasoning_tokens":2731,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:56:12.947642+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Examine whether the enlarged-space Heisenberg evolution generated by Eq. (54) maps the fully symmetric subspace to itself for all times; a single explicit counterexample — such as a symmetric initial two-spin state evolving into a state with nonzero antisymmetric component, or a physical operator whose norm grows under projection — would invalidate the S→∞ transfer. Alternatively, compute the exact or high-precision numerical Lieb-Robinson velocity of the S=2 or S=3 Heisenberg XYZ chain and check whether it approaches the finite bound 4dJ_m X_{1−1/(2Sd)} or instead grows with S.","supporting_citations":[{"cited_title":"Lieb– robinson bounds for open quantum systems with long- ranged interactions,","cited_arxiv_id":null,"evidence_quote":"A special LR bound for a truncated rotor model that motivated the commutativity-graph idea."},{"cited_title":"Linear generaliza- tions of Gr¨ onwall’s inequality,","cited_arxiv_id":null,"evidence_quote":"Generalized Gronwall inequality used to convert integral inequalities into the linear differential equations."},{"cited_title":"The bond-algebraic approach to dualities,","cited_arxiv_id":null,"evidence_quote":"Semiclassical large-S Heisenberg dispersion used to benchmark the finite large-S velocity."},{"cited_title":"Quantum ether: photons and electrons from a rotor model,","cited_arxiv_id":null,"evidence_quote":"A graph-theoretic operator growth bound for the 1D TFIM that is compared against and combined with the present method."},{"cited_title":"Artificial light and quantum order in systems of screened dipoles,","cited_arxiv_id":null,"evidence_quote":"The rotor model derivation of emergent light speed v=√(2gJ), used as the semiclassical benchmark in Sec. V E."},{"cited_title":"Local stabilizer codes in three dimen- sions without string logical operators,","cited_arxiv_id":null,"evidence_quote":"Exact ground-state solution of the 1D TFIM used to benchmark the correlation-length bound."}],"review_version":1}