{"id":"1a35a3ba-d287-4b2b-8631-8085e87fdef6","arxiv_id":"1909.01516","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For frustration-free Hamiltonians on any finite-dimensional lattice, the minimum spectral gap of any rectangular region is O(γ + 1/t²) where t is the shortest side length, improving previous thresholds.","lead":"This paper proves a new quantitative bound on how small the local energy gap of a frustration-free quantum spin system must be when the full system is gapless. The bound scales as one over the square of the region size, which is optimal up to a dimension-dependent constant and improves previous results in all dimensions greater than one.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proof of Claim E.1 is only a sketch: absorbing nonlocal layer factors into S/T blocks is unjustified, so the O(1/t^2) threshold is not fully established.","rationale":"The reader identified Claim E.1 as the weakest assumption, and I agree that it is the most load-bearing concern. The central asymptotic claim likely survives if the missing proof is supplied, but as written the paper does not establish the bridge between the coarse-grained detectability operator and the global detectability operator. The absorption argument counts layer factors rather than controlling their noncommutation, and the cited AAG19 claim is not reproduced. The apparent numerical discrepancy '103 vs 800' mentioned in the reader's rationale appears to be a rendering artifact: the manuscript likely means 10^3 L^2 g^2, not 103, since Theorem C.1 uses 200 L^2 g^2 6^D and the proof yields 800 < 1000. Therefore I do not treat that as a substantive objection. The verdict should remain CONDITIONAL: the paper should be accepted only if the proof of Claim E.1 is completed or a fully detailed proof from AAG19 is incorporated.","tokens_in":13476,"tokens_out":31521,"duration_ms":302595,"concrete_test":"Re-derive Eq. (15) for the minimal nontrivial case L=2, q=1: expand both sides of (1-Q_S)(1-P_2)(1-P_1)(1-P_2)(1-Q_T) = (1-Q_S)(1-Q_T) using 1-P_i = (1-Q_i)+(Q_i-P_i), with Q_S and Q_T specified as projectors orthogonal to the ground spaces of the interval Hamiltonians h_S and h_T. Determine whether all cross terms vanish by the structure of h_S and h_T. If they do, repeat the calculation for L>2; if not, Claim E.1 is false and the main theorem is not proven. Alternatively, run a small brute-force search over two noncommuting two-local projectors P_1,P_2 with a common frustration-free ground state, placing their supports across an S/T boundary, and compare both sides of Eq. (15) on a complete basis.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Appendix E, Claim E.1 is the load-bearing step. Lemma B.5 and therefore Theorem B.1 require the identity DL(t) = (∏(1-Q_Sk)) F(DL(H)^† DL(H)) (∏(1-Q_Tk)). The proof given is a sketch: DL(H)^† DL(H) is expanded as a product of q(2L-2)+1 layer factors DL_α, and because this number is < floor(t/4) while each S/T pair overlaps in floor(t/4) sites, the factors are said to be 'absorbed' into the S and T projectors. This inference is not justified. Each DL_α = ∏_{i,j∈T_α}(1-P_ij) is a global product over an entire layer, not a local operator; to absorb it one must move it past the noncommuting projectors (1-Q_S), (1-Q_T) and past factors from other layers. The overlap bound only supplies spatial room, not commutation. In the L=2 case the identity would require (1-Q_S)(1-P_2)(1-P_1)(1-P_2)(1-Q_T) = (1-Q_S)(1-Q_T), and a direct expansion using 1-P_i = (1-Q_i)+(Q_i-P_i) has cross terms such as (1-Q_S)(Q_T-P_2)(Q_S-P_1)(1-Q_T) that are not shown to vanish. Unless these terms cancel by the specific structure of coarse-grained projectors, Lemma B.5 collapses, and with it the O(1/t^2) threshold in Theorems B.1 and C.1. The citation to [AAG19, Claim B.1] is not a substitute in this manuscript, which promises a full proof but gives only an outline.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies frustration-free local Hamiltonians on finite-dimensional lattices and proves a local spectral gap threshold. The main result (Theorem C.1) states that for a frustration-free Hamiltonian on a D-dimensional lattice with spectral gap γ, the minimum spectral gap γ(t_1,...,t_D) over all hyper-rectangular regions of side lengths t_q satisfies γ(t_1,...,t_D) ≤ 6^D γ + 200 L^2 g^2 6^D / min_q t_q^2, with constants depending on the lattice dimension and the interaction's locality and layer structure. The proof uses a one-dimensional reduction: a coarse-grained Hamiltonian of overlapping intervals (S and T regions), an estimate on low-degree Chebyshev polynomials, and the detectability lemma. A recursive dimensional-reduction argument then extends the one-dimensional bound to D dimensions. The paper also presents a lower-bound example (parallel copies of the Heisenberg ferromagnet chain) showing that the quadratic decay in the shortest side length is optimal up to dimension-dependent constants, and an example showing that a version with an average over regions would not hold.","tokens_in":13829,"tokens_out":15372,"duration_ms":134527,"significance":"If fully established, this result is a significant step in the finite-size criteria literature for frustration-free Hamiltonians: it improves the prior local gap thresholds for D-dimensional lattices (from O(log^2 t / t) in [KL18] and O(1/t) in [Lem19a]) to the optimal O(1/t^2) scaling, with a simple proof technique based on coarse-grained Hamiltonians and the detectability lemma. The lower-bound example correctly demonstrates that the scaling in the shortest side length cannot be improved. However, the manuscript as written is not complete: the key identity behind the main theorem is only sketched, and there is an arithmetic error in the derivation of the constant in Theorem B.1. These issues must be fixed before the claims can be accepted.","major_comments":[{"comment":"The derivation of the constant in Theorem B.1 is incorrect. The displayed bound γ(¯H(t)) ≥ t^2γ/(400 L^2 g^2 + 3 t^2γ) combined with Eq. (8), i.e., 2γ/γ(t) ≥ γ(¯H(t)), gives γ(t) ≤ 800 L^2 g^2 / t^2 + 6γ, not the stated 103 L^2 g^2 / t^2 + 6γ. The constant 103 appears in the theorem statement and is then propagated through the recursion in Theorem C.1. Since the paper explicitly says the constants are not optimized, this error is repairable, but as written the proof does not establish the stated numerical bound.","section":"B.3"},{"comment":"Claim E.1 is load-bearing for Lemma B.5 and hence for Theorems B.1 and C.1, but its proof is only an outline. The claim that all layer operators DL_α can be 'absorbed' into the S and T projectors because the overlap between adjacent S and T sets is at least ⌊t/4⌋ is not a valid inference: the DL_α are products over entire layers and do not commute with the projectors (1-Q_Sk) and (1-Q_Tk), so spatial overlap alone does not imply the identity in Eq. (15). For example, in the L=2 case with F(x)=x, the identity would require (1-Q_S)(1-P_2)(1-P_1)(1-P_2)(1-Q_T) = (1-Q_S)(1-Q_T), which is not a formal consequence of the assumptions. The author should provide a complete proof of Claim E.1 or replace the upper bound in Lemma B.5 with a fully rigorous argument; a citation to [AAG19, Claim B.1] is not sufficient in a manuscript that promises a proof for completeness.","section":"E, Claim E.1"}],"minor_comments":[{"comment":"The condition in Theorem C.1 is garbled: 'Suppose 264DL<t s<n s/ 5' should presumably be a condition of the form t_s > c_D L and t_s < n_s/5 for an explicit dimension-dependent constant c_D. Please correct the typesetting.","section":"C.1"},{"comment":"In the proof of Theorem C.1, the base-case display '10 3 2Lg' appears to be a misprint for '103 L^2 g^2', and similar notation errors make the constant bookkeeping hard to follow.","section":"C.1"},{"comment":"The lower bound in Lemma B.5 is stated to follow from Lemma B.4, but the connection is not made explicit: the coarse-grained Hamiltonian has two layers (the S projectors and the T projectors), so a short explanation of why Lemma B.4 applies would improve readability.","section":"B.5"},{"comment":"In the two-dimensional proof outline, the sentence 'The overall additive factor is O(1/t_1^2 + 1/t_1^2)' should read O(1/t_1^2 + 1/t_2^2); the repeated t_1 is a typo.","section":"2.1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript relies heavily on a claim cited from an unpublished preprint [AAG19] for the central estimate; the editor may wish to verify that the full proof will be made available in the revised version or in a companion publication. The numerical discrepancy in Section B.3 and the incomplete proof of Claim E.1 are the main obstacles to acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper proves the local gap threshold O(γ + 1/min t_q^2) for frustration-free Hamiltonians on any finite-dimensional lattice, improving on the prior O(1/t) and O(log^2 t / t) bounds. That is a genuine step forward, and the parallel-Heisenberg-chain example shows the t^{-2} scaling is optimal. The recursive column/row decomposition is clean and works without translation invariance, and the coarse-grained Hamiltonian + detectability lemma framework is appropriate.\n\nThe soft spots are two. First, the substitution in B.3 gives 800 L^2 g^2 / t^2, not 103. That is a concrete arithmetic error, but it only affects constants, not scaling.\n\nSecond, Claim E.1 is load-bearing and is only sketched. The stress-test objection lands: the proof says the layer operators are 'absorbed' into the S/T projectors because of spatial overlap, but overlap gives room, not commutation. The L=2 expansion with cross terms that don't obviously vanish is a fair challenge. The author cites [AAG19, Claim B.1], but this manuscript promises a full proof and gives an outline. If the claim is standard in the literature, a precise citation would fix it; if not, the proof needs to be written out. This is the main reason I would not accept the paper as is.\n\nI also note the paper is honest about non-optimized constants and gives a correct lower-bound example. The central argument seems plausible and the result is likely correct, but the proof is incomplete at a critical point.\n\nRecommendation: send it to peer review. The result is important enough to deserve referee time, and the referee should ask for a complete proof of Claim E.1 and a correction of the constant.","headline":"Improves the local gap threshold to optimal O(1/t^2) in any finite dimension, but the main proof leans on a sketched claim that needs a real proof.","tokens_in":14394,"tokens_out":16094,"would_cite":true,"duration_ms":145260,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Local spectral gaps of gapless spin lattices must shrink quadratically in the shortest side length.","keywords":["spectral gap","frustration-free Hamiltonian","local gap threshold","coarse-grained Hamiltonian","detectability lemma","quantum spin lattice","gapless systems","Chebyshev polynomials"],"falsifier":"A direct falsifier would be a family of translationally invariant frustration-free local Hamiltonians on a $D$-dimensional lattice, gapless as the system grows, for which the minimum spectral gap over $t \\times \\dots \\times t$ regions decays faster than $1/t^2$, for example $\\Theta(1/t^{5/2})$; this would break the claimed optimal additive term in Theorem C.1. A more surgical check is to test the polynomial-insertion identity (Equation 15) numerically for small $t$ just above $8L^2$: any failure there would invalidate the chain of inequalities that produces the quadratic threshold.","tokens_in":13255,"feed_emoji":"🧲","tokens_out":12223,"duration_ms":119814,"temperature":0.7,"pith_summary":"This paper establishes a tight quantitative link between the global spectral gap of a frustration-free local Hamiltonian (one whose ground state is annihilated by each local interaction) on a fixed-dimensional lattice and the spectral gaps of the same Hamiltonian restricted to finite hyper-rectangular blocks. The main theorem says the minimum local gap $\\gamma(t_1,\\dots,t_D)$ is bounded above by a constant, depending on dimension and interaction structure, times the global gap $\\gamma$, plus a term of order $1/\\min_q t_q^2$. For a gapless system, $\\gamma \\to 0$, so every $t_1 \\times \\dots \\times t_D$ block must have a gap at most $O(1/\\min_q t_q^2)$; the paper argues this quadratic decay is unavoidable, up to dimension-dependent constants, via parallel chains of a simple ferromagnetic model. This sharpens earlier local-gap thresholds in several dimensions and matters because local gaps are a practical route to proving global spectral gaps, which in turn govern correlation decay and phase structure.","feed_headline":"Gapless spin lattices: local gap falls quadratically in size","feed_subtitle":"New theorem ties finite-region gaps to the global gap in any dimension; the 1/t² rate is optimal up to constants.","key_machinery":"The load-bearing object is the coarse-grained Hamiltonian. On a chain, the lattice is divided into overlapping blocks $S_k$ and $T_k$ of length $t$, and $\\hat{H}(t) = \\sum_k (Q_{S_k} + Q_{T_k})$ is formed, where each $Q$ is the projector onto excited states of the block-restricted Hamiltonian; $\\hat{H}(t)$ has the same ground space as the original Hamiltonian. The proof couples three tools: the detectability lemma, which controls powers of the layered product of local projectors on excited states; a low-degree Chebyshev 'step' polynomial, which shrinks the excited spectrum enough to yield $\\gamma(\\hat{H}(t)) \\ge \\frac{t^2\\gamma}{400L^2g^2 + 3t^2\\gamma}$; and the inequality $\\gamma(\\hat{H}(t)) \\le 2\\gamma/\\gamma(t)$. Combining these gives the one-dimensional bound $\\gamma(t) \\le 103 L^2 g^2/t^2 + 6\\gamma$. Higher dimensions are handled by viewing a slab Hamiltonian as a chain of 'column' Hamiltonians and applying the one-dimensional argument recursively, once per dimension.","core_discovery":"The central discovery is Theorem C.1: for a frustration-free Hamiltonian $H = \\sum_\\alpha P_\\alpha$ made of local projectors on a $D$-dimensional lattice whose interaction graph has degree $g$ and whose terms can be partitioned into $L$ commuting layers, the minimum spectral gap over hyper-rectangles of side lengths $t_1,\\dots,t_D$ satisfies $\\gamma(t_1,\\dots,t_D) \\le 6D\\gamma + 200 L^2 g^2 6^D / \\min_q t_q^2$, provided each side length is larger than a constant depending on $L$ and $g$. In the gapless limit this gives $\\gamma(t,\\dots,t) = O(1/t^2)$, and the example of many independent Heisenberg ferromagnetic chains shows the $1/t^2$ term is necessary even for translationally invariant systems, so the dependence on $\\min_q t_q^2$ is optimal up to the dimension-dependent constant. The theorem applies to open and periodic boundary conditions and does not require translation invariance.","pith_inferences":["A fully rigorous proof of the absorption identity for all admissible $t$ would make the quadratic threshold unconditional for arbitrary local interactions; numerical checks on small chains could reveal whether the identity is exact or only approximate outside the sketched overlap regime.","The same machinery might yield quantitative decay-of-correlation or entanglement bounds in gapless frustration-free systems, since it already converts global gap information into local spectral information with sharp scaling.","If the paper's conjecture for isotropic translationally invariant Hamiltonians is correct, the relevant quantity becomes the inverse-squared diameter $1/\\sum_q t_q^2$, so long thin regions would be far less constrained than the shortest-side bound suggests; columnar quasi-one-dimensional models could test this.","The exponential dependence of the constant on dimension is an artifact of the recursive column-row decomposition, and a direct $D$-dimensional coarse-graining argument could plausibly reduce it to polynomial in $D$."],"forward_implications":["Any gapless frustration-free Hamiltonian on a fixed-dimensional lattice must have local spectral gaps over side-length $t$ regions that are at most $O(1/t^2)$; no such model can keep those region gaps at $\\Theta(1/t)$.","For finite systems, the theorem gives a finite-size criterion: if every $t_1 \\times \\dots \\times t_D$ region has a gap larger than the stated threshold, the global Hamiltonian is guaranteed gapped.","The bound holds without translation invariance and for open and periodic boundary conditions, so it applies to boundary-modified and disordered frustration-free systems.","Since the $1/\\min_q t_q^2$ rate is optimal, further improvement in this type of threshold can only come from the dimension-dependent constant or from additional symmetry of the Hamiltonian."],"supporting_citations":[{"why":"Establishes the original one-dimensional inequality between global and local spectral gaps and the local-gap-threshold phenomenon that this paper generalizes.","marker":"[Kna88]"},{"why":"Supplies the tighter one-dimensional and two-dimensional square-lattice thresholds and the parallel-Heisenberg-ferromagnet example showing the quadratic rate is necessary.","marker":"[GM16]"},{"why":"Introduces the coarse-grained Hamiltonian and the detectability-lemma converse that form the proof's bridge between global and local gaps.","marker":"[AAV16]"},{"why":"Provides the detectability lemma bounding the layered product operator's norm on excited states.","marker":"[AALV09]"},{"why":"Underlies the converse detectability lemma that yields the quadratic lower bound on the coarse-grained gap.","marker":"[Gao15]"},{"why":"Adapted for Claim E.1's polynomial-insertion step and the low-degree Chebyshev shrinking estimate.","marker":"[GH16]"},{"why":"Supplies the absorption argument showing powers of the detectability operator can be moved past block projectors.","marker":"[AAG19]"},{"why":"Contributes the key inequality $\\gamma(\\hat{H}(t)) \\le 2\\gamma/\\gamma(t)$ connecting coarse-grained and local gaps.","marker":"[Gos19]"},{"why":"Gives the prior $O(1/t)$ threshold in $D$ dimensions that the theorem improves for large regions.","marker":"[Lem19a]"},{"why":"Gives the alternative martingale-detectability approach with $O(\\log^2 t/t)$ threshold, the main comparison point.","marker":"[KL18]"}],"fun_headline_variants":["Local gap falls as 1/t², optimal up to dimension constant","Sharp bound: spectral gap of regions falls as 1/t² in any dimension","1/t² decay for local gaps is optimal up to dimension constant","Optimal quadratic local-gap threshold for frustration-free lattices","Local spectral gap bound: 1/t² rate, optimal up to dimension"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof's load-bearing step is the assertion that a low-degree polynomial of the full detectability operator can be inserted between the block projectors of the coarse-grained Hamiltonian without changing the operator; this absorption step is justified in the paper only by an outline and depends on the overlap between the two block families being at least a quarter of the block length.","fun_headline_variants_meta":{"raw":{"variants":["Local gap falls as 1/t², optimal up to dimension constant","Sharp bound: spectral gap of regions falls as 1/t² in any dimension","1/t² decay for local gaps is optimal up to dimension constant","Optimal quadratic local-gap threshold for frustration-free lattices","Local spectral gap bound: 1/t² rate, optimal up to dimension"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000989,"raw_usage":{"total_tokens":4182,"prompt_tokens":925,"completion_tokens":3257,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":541,"completion_tokens_details":{"reasoning_tokens":3161}},"tokens_in":541,"tokens_out":3257,"duration_ms":24154,"temperature":1.0,"reasoning_tokens":3161,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:16:05.183476+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct falsifier would be a family of translationally invariant frustration-free local Hamiltonians on a $D$-dimensional lattice, gapless as the system grows, for which the minimum spectral gap over $t \\times \\dots \\times t$ regions decays faster than $1/t^2$, for example $\\Theta(1/t^{5/2})$; this would break the claimed optimal additive term in Theorem C.1. A more surgical check is to test the polynomial-insertion identity (Equation 15) numerically for small $t$ just above $8L^2$: any failure there would invalidate the chain of inequalities that produces the quadratic threshold.","supporting_citations":[],"review_version":1}