{"id":"e0ee1211-bb23-4a07-8f20-d03a3210227f","arxiv_id":"2512.00382","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Bubble-basis projector QMC extracts the universal corner log-coefficient of Rényi entanglement entropy at (2+1)d Ising and Gaussian quantum critical points as 0.020(1) and 0.025(1), the latter matching the analytic value 0.02567.","lead":"A new quantum Monte Carlo method in a 'bubble basis' computes the tiny universal corner contribution to entanglement entropy at (2+1)-dimensional quantum critical points with an order-of-magnitude speedup. The authors validate it against the known free-theory value at the Gaussian tricritical point and then extract the Ising value, linking solvable and strongly-correlated regimes.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Gaussian fixed-point identification ignores acknowledged φ^6 log corrections; a small h_c shift could bias the s=0.025(1) benchmark.","rationale":"The reader's weakest-assumption analysis highlights the same load-bearing concern: the Gaussian fixed-point identification ignores the acknowledged φ^6 logarithmic corrections. This is the single point on which the central benchmark rests, because a misidentification of K_c or h_c would invalidate the s=0.025(1) match to the analytic 0.02567. The paper's empirical evidence (linear S_s vs ln L, clear contrast with Ising and first-order points) is suggestive but not sufficient, because a slightly off-critical point can mimic logarithmic growth over a limited L range. The proposed larger-L Binder-cumulant test directly probes whether the log corrections are large enough to shift h_c beyond the quoted uncertainty. If such a shift is absent, the benchmark stands; if present, the central claim must be conditionally accepted only after re-running the EE at the corrected point. Given the method's novelty and the otherwise careful presentation, CONDITIONAL (rather than REJECT) is the appropriate verdict, consistent with the reader's assessment.","tokens_in":23327,"tokens_out":18294,"duration_ms":184604,"concrete_test":"Measure the Binder cumulant at K=16.02 for linear sizes L=80, 96, and 128. If the apparent crossing point h_c(L) drifts beyond the quoted 67.57(4) or the pure ν=1/2 power-law collapse degrades, repeat the EE simulation at the log-corrected h_c and check whether the extracted s remains 0.025(1). If s shifts by more than 0.001, the Gaussian benchmark is not robust.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central benchmark s=0.025(1) at the Gaussian fixed point depends entirely on the identification K_c=16.02(6) and h_c=67.57(4). That identification uses a pure power-law Binder-cumulant collapse with ν=1/2 (Fig. 2(d-f)). The authors themselves acknowledge (Sec. II A) that the marginally irrelevant φ^6 operator generates logarithmic finite-size corrections at the upper critical dimension: 'yielding logarithmic finite-size corrections to critical exponents. Previous Monte Carlo simulations tried to detect the logarithmic corrections, but their status does not appear to be definite.' These corrections are not included in the FSS ansatz. If they are non-negligible for L≤64, the effective ν from a power-law fit can be biased away from 1/2, and the Binder crossing h_c can drift with L, so the simulated point may lie off the Gaussian fixed point. Because the EE simulation uses L_max=36 and S_s=s ln L + γ, a shift of h_c by even a fraction of the quoted error can produce a finite-size-dependent slope that mimics a different s (e.g., 0.025 vs 0.020). The straightness of S_s vs ln L in Fig. 1(b) is suggestive but not conclusive: a slightly off-critical system with a correlation length large compared to L can also produce a quasi-logarithmic S_s. The concern is purely about the argument's weakest link, not about the authors' intent; the method is otherwise carefully laid out, and the free-lattice analytic result in Sec. III A is a useful cross-check.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper introduces a 'bubble basis' projector quantum Monte Carlo method combined with the incremental SWAP algorithm and the subtracted-corner-entanglement-entropy technique to extract the universal corner logarithm in the second Rényi entropy of (2+1)d quantum critical points. The model is a square-lattice transverse-field Ising model with an additional four-spin interaction; tuning this interaction traces an Ising critical line to a tricritical point and then a first-order line. The authors report that at the tricritical point, identified as the Gaussian fixed point at K_c=16.02(6) and h_c=67.57(4), the corner coefficient is s=0.025(1), matching the analytic free-scalar value s=0.02567; along the Ising critical line they obtain s=0.020(1), and on the first-order line s=0.000(1). The algorithmic claim is that the bubble-basis formulation reduces the cost of incremental SWAP EE computations from O(m^2) to O(mP), where P is a state-dependent parameter scaling as a power of L.","tokens_in":23728,"tokens_out":5154,"duration_ms":56510,"significance":"If the central results hold, this is a substantial methodological and numerical advance: it would be the first QMC computation that reproduces an exact free-theory corner-entropy coefficient and then moves continuously to an interacting (2+1)d critical point within one framework, with precision sufficient to distinguish the Gaussian coefficient from the Ising coefficient at the 0.005 level. The free-theory benchmark calculation in Sec. III A is clearly described and reproducible, and the algorithmic complexity analysis is a genuine contribution. The paper is also careful to benchmark the bubble-basis PQMC against known TFIM critical exponents in the Supplemental Material. The main risk is not internal inconsistency of the algorithm but the identification and finite-size control of the Gaussian fixed point, on which the benchmark relies.","major_comments":[{"comment":"The identification K_c=16.02(6) and h_c=67.57(4) as the Gaussian fixed point is obtained from Binder-cumulant crossings and a pure power-law finite-size collapse with ν=1/2. In Sec. II A you explicitly state that φ^6 is marginally irrelevant at the upper critical dimension and yields logarithmic finite-size corrections. Those corrections are absent from the scaling ansatz used in Fig. 2(e). If they are non-negligible for L≤64, the crossing location h_c can drift with L and the effective ν from a power-law fit can be biased, so the EE simulation at 'K_c' may not be exactly at the Gaussian fixed point. Because the benchmark s=0.025(1) is the central anchor, please (i) repeat the Binder analysis with a log-corrected scaling form or otherwise bound the size of the log corrections for these sizes, and (ii) demonstrate that the extracted s is stable when h is varied within the quoted uncertain","section":"II A / II B, Fig. 2(d-f)"},{"comment":"The extrapolation protocol for extracting s is not defined in a falsifiable way. The text says the process is stopped 'once the two smallest values of 1/L_min converge' and the final slope is the average of the last two points. This introduces a model-selection step with no stated criterion; the claimed 0.001-level differences between s=0.020(1) and s=0.025(1) are precisely the quantity affected by the chosen L_min window. Please report the full s versus 1/L_min table for each K, state an explicit convergence rule, and provide an estimate of the systematic uncertainty due to the fitting window (e.g., bootstrap over windows or fits with explicit 1/L corrections). Without this, the quoted error bars are likely underestimated and the sharp separation of universal values is not fully established.","section":"II C, Fig. 1(b)-(h)"}],"minor_comments":[{"comment":"The main figure is captioned 'FIG. 1' but the text repeatedly refers to 'Fig. I (a)', 'Fig. I (b)', etc. The numbering should be made consistent.","section":"General"},{"comment":"Several typos: 'simulaition' in Sec. II, 'trail wavefunctions' in Sec. III B, 'Renyi' without accents, and 'convergeds' in Sec. II C. Please proofread.","section":"Throughout"},{"comment":"The variable s is defined as the coefficient for four 90-degree corners, but the text sometimes says 'universal corner log-coefficient' without specifying the number of corners. State explicitly in each place that the quoted values are for four 90-degree corners.","section":"Eq. (6) and Fig. 1"},{"comment":"The notation P is introduced as 'a slice-independent parameter' but later shown to depend on the phase and on L. Please clarify that P is the statistical average number of sites in the relevant link-bubbles, not an independent algorithmic parameter.","section":"Sec. III B"}],"recommendation":"major_revision","confidential_remarks":"The paper is strong and likely publishable after revision. The decisive issue is whether the Gaussian fixed-point identification and the extrapolation protocol can withstand scrutiny; both are central to the benchmark claim. I recommend major revision rather than rejection because the weaknesses are addressable with additional analysis and transparency."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Lee-Yeung et al. have a real algorithmic result: the bubble-basis PQMC lowers the cost of incremental SWAP from O(m^2) to O(mP), with P scaling roughly like L^0.84 at the QCP. That is a genuine improvement, and they integrate it with the subtracted-corner-entropy trick so the area law cancels inside the simulation. The free-field lattice calculation in Sec. III A is a useful, independently reproducible cross-check. The headline numbers — s=0.025(1) at the Gaussian point versus the analytic 0.02567, and s=0.020(1) at the Ising point — are the strongest part of the paper. The match at Gaussian is a credible external benchmark, not a fit.\n\nThe main soft spot is the identification of K_c. The FSS collapse at K=16.02 uses a pure power-law ansatz with nu=1/2, while the authors themselves note that the marginally irrelevant phi^6 operator generates log corrections at the upper critical dimension. If those logs are sizeable for L<=64, h_c could drift and the simulated point might sit off the Gaussian fixed point. The stress-test note is on target there. That said, the optimized nu comes out as 0.52(5), consistent with 0.5, and the S_s versus ln L data are straight; nothing in the paper shows the log corrections are actually large enough to move s by 0.005. So I would call it a legitimate caveat, not a refutation. A sensitivity run at h_c +/- delta, or a collapse that includes the log correction, would settle it.\n\nTwo smaller issues. The 1/L_min stopping rule (\"stop when the two smallest values converge\") is ad hoc; fine as a heuristic, but it should be justified or replaced by a more systematic criterion. And there is no code or data release. For a methods paper whose main claim is a faster, more precise algorithm, that is a real gap; reviewers should ask for it.\n\nThe complexity argument is empirical rather than proven, but the P scaling measurement is reasonable and the data structure explanation in the SM is convincing. The first-order points giving s=0.000(1) are a nice consistency check. The citation pattern is fair; the core algorithm citations are to their own earlier incremental SWAP work, which is appropriate.\n\nBottom line: this deserves a serious referee. The Gaussian benchmark is the kind of nontrivial cross-check the field needs, and the algorithm is genuinely useful even if the tricritical identification needs more work. I would recommend conditional acceptance: address the h_c sensitivity, tighten the extrapolation rule, and release the code.","headline":"Genuine algorithmic advance in QMC entanglement entropy, with a strong Gaussian benchmark; the main caveat is the FSS identification of the tricritical point, which is addressable but not fatal.","tokens_in":24193,"tokens_out":2542,"would_cite":true,"duration_ms":24963,"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":"A projector quantum Monte Carlo algorithm in a bubble basis computes the universal corner logarithmic term in (2+1)d Rényi entanglement entropy at both Ising and Gaussian critical points, yielding s=0.020(1) and s=0.025(1).","keywords":["universal corner entropy","Rényi entanglement entropy","projector quantum Monte Carlo","bubble basis","incremental SWAP","tricritical Ising","Gaussian fixed point","transverse-field Ising model"],"falsifier":"Perform the finite-size scaling collapse at K_c=16.02 including explicit logarithmic corrections, or push the subtracted-corner simulation to L>64 and check whether s continues to converge to 0.025(1) rather than drifting; a drift would show the Gaussian identification or the extrapolation was not yet asymptotic.","tokens_in":23221,"feed_emoji":"⚛️","tokens_out":4449,"duration_ms":41371,"temperature":0.7,"pith_summary":"The paper tries to establish that the subleading logarithmic corner term in entanglement entropy can be computed precisely at interacting (2+1)d quantum critical points, not just in free theories. It introduces a basis of 'bubble' states for projector QMC that lowers the cost of incremental swap sampling from O(m^2) to O(mP), and combines it with a subtracted-corner scheme that removes the area-law term during sampling. Applied to a transverse-field Ising model with a four-spin coupling, the method traces a line of Ising critical points to a Gaussian tricritical point. At the Gaussian point the measured corner coefficient s=0.025(1) matches the analytic value 0.02567; along the Ising line it converges to s=0.020(1). If correct, this supplies a benchmark separating universal entanglement data for two distinct (2+1)d universality classes and connects solvable free-theory results to strongly coupled critical points.","feed_headline":"Two critical points, two corner entropies: 0.020 and 0.025","feed_subtitle":"Bubble-basis QMC isolates the universal log term and matches the free-theory value 0.02567 at the Gaussian point.","key_machinery":"The central mechanism is the bubble-basis projector QMC update combined with an incremental SWAP estimator and subtracted corner entropy. A bubble state is a superposition in which each bubble is a connected cluster of sites with common spin orientation; operators H_J, H_h, and H_K act on bubble states without branching, and the overlap of two bubble states equals 2^{N_B}, where N_B counts link-bubbles. This makes the weight-ratio update cost O(P) per slice (P the average number of sites in the relevant link-bubbles) rather than O(m), so the full EE computation runs at O(mP) instead of O(m^2) in the sigma^z basis. That efficiency, together with computing S_s = S_{A1,2} - S_{A2,2} directly in","core_discovery":"The central claim is that the universal corner term in the second Rényi entropy—the coefficient s in S_2 = a L - s ln L + c for an entanglement region with four 90-degree corners—is now numerically accessible at strongly interacting (2+1)d quantum critical points. The authors construct a square-lattice transverse-field Ising model with a four-body K term, whose phase diagram contains an Ising transition line terminating at a Gaussian (tricritical Ising) point, followed by a first-order line. Using their bubble-basis projector QMC with the incremental swap estimator and subtracted-corner-entropy technique, they extract s=0.025(1) at the Gaussian point, in quantitative agreement with the free-","pith_inferences":["Editorial inference: The demonstrated ability to resolve 0.020 from 0.025 at the 0.001 level suggests corner-entropy coefficients could serve as a practical probe of universality class in numerical studies where critical exponents are ambiguous.","Editorial inference: The bubble-basis update may generalize beyond spin models to path-integral or fermionic QMC settings, where exponential observables such as Rényi negativity or free-energy differences suffer from the same variance problem.","Editorial inference: A direct check of the Gaussian classification at K_c by including logarithmic corrections in the finite-size scaling collapse would determine whether s=0.025(1) is truly the free-theory value or a close finite-size accident."],"forward_implications":["The universal corner coefficient for the (2+1)d Ising universality class is pinned at s=0.020(1), a sharper benchmark than previous estimates.","The Gaussian fixed-point value s=0.02567 is reproduced numerically, validating both the algorithm and the identification of the tricritical point.","The bubble-basis incremental-swap complexity O(mP) makes larger systems and more difficult entanglement observables accessible at interacting quantum critical points.","The vanishing s on the first-order line confirms the expected product-state behavior and provides a clean diagnostic for first-order transitions.","The same pipeline can be applied to other corner configurations, higher Rényi orders, or the von Neumann entropy at the same critical points."],"fun_headline_variants":["Gaussian corner entropy: 0.025(1) matches theory","Bubble-basis QMC nails universal corner term","Corner entropy at 2+1D QCPs: Ising and Gaussian","Precise corner log term from bubble-basis QMC"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the Gaussian classification of the tricritical point at K_c=16.02(6) is unaffected by the logarithmic corrections from the marginally irrelevant phi^6 operator at the system sizes studied; if those corrections are sizeable for L up to 64, the Gaussian labeling and hence the benchmark value s=0.025(1) would be compromised.","fun_headline_variants_meta":{"raw":{"variants":["Gaussian corner entropy: 0.025(1) matches theory","Bubble-basis QMC nails universal corner term","Corner entropy at 2+1D QCPs: Ising and Gaussian","Precise corner log term from bubble-basis QMC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000315,"raw_usage":{"total_tokens":1671,"prompt_tokens":843,"completion_tokens":828,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":587,"completion_tokens_details":{"reasoning_tokens":764}},"tokens_in":587,"tokens_out":828,"duration_ms":7478,"temperature":1.0,"reasoning_tokens":764,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T19:25:34.735140+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Perform the finite-size scaling collapse at K_c=16.02 including explicit logarithmic corrections, or push the subtracted-corner simulation to L>64 and check whether s continues to converge to 0.025(1) rather than drifting; a drift would show the Gaussian identification or the extrapolation was not yet asymptotic.","supporting_citations":[],"review_version":1}