{"id":"7afa6c81-d035-461f-a70e-f2b330df6b8b","arxiv_id":"2411.13998","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":1,"one_line_summary":"A bond-weight impurity method for BWTRG computes higher-order moments in 2D Ising and Potts models with improved accuracy and efficiency.","lead":"This paper presents a computational technique that places 'impurities' on bond weights to compute magnetization, energy, and higher moments inside tensor-network simulations of 2D magnets. The method is more accurate than the standard tensor renormalization group and gives precise critical-temperature estimates using finite-size scaling.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Efficiency claim vs MPS rests on BWTRG exponent κ/ν≈4.0, but the same paper cites free-energy-error fits giving κ≈2.2 for BWTRG; if the T_c exponent is pre-asymptotic or estimator-specific, the asymptotic time-complexity advantage evaporates.","rationale":"The manuscript provides a clear algorithmic construction (Sec. II.B), exact-model benchmarks for energy and magnetization (Figs. 4–6), and an interesting finite-size-scaling analysis using X1 (Fig. 7). The impurity update equations (8)–(9) are straightforward and the Ising/Potts results are consistent with known exact values. The strongest part of the paper is the demonstration that BWTRG with k=-1/2 outperforms TRG for moments. However, the headline sentence in the abstract—that BWTRG is more efficient than MPS in estimating T_c—is supported only by an asymptotic exponent comparison. That comparison is fragile because the BWTRG exponent κ/ν≈4.0 used in the time-complexity argument is not reconciled with the free-energy-error-derived κ≈2.2 for the same hyperparameter, cited in the same section. For ν=1 these imply δT_c exponents of 4.0 and 2.2 respectively. Since Eq. (33) is the basis of the comparison, this internal tension is load-bearing. If the true asymptotic exponent is closer to 2.2, the time exponents reverse (0.44 vs 0.68) and the abstract claim is wrong. A direct numerical check at larger χ and with simultaneous free-energy measurement can settle this. The reader's earlier CONDITIONAL verdict remains appropriate; no source code or wall-clock data are provided, and our concern sharpens the condition.","tokens_in":13876,"tokens_out":12064,"duration_ms":111581,"concrete_test":"Perform BWTRG (k=-1/2) at χ=64,128,256 on the Ising model, and from the same runs extract both the free energy f(χ) and T_c(χ) (via the X1 jump). Fit δf∼χ^{-2κ_f} and δT_c∼χ^{-κ_T/ν} with ν=1. If κ_f≈2.2 but κ_T≈4.0, the X1-jump estimator does not obey the simple scaling relation (33), and the t^{-0.80} efficiency comparison lacks support. Alternatively, if both give κ≈4.0, the free-energy-derived κ=2.2 is itself inconsistent and needs re-examination.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline efficiency conclusion is drawn in Sec. III.B by comparing δT_c ∼ χ^{-κ/ν} exponents: BWTRG (k=-1/2) is fitted as κ/ν≃4.0 (Fig. 10), so with O(χ^5) cost the error scales as t^{-0.80}, versus MPS/VUMPS with κ_CFT/ν≃2.034 and O(χ^3) cost giving t^{-0.68}. This comparison is only valid if the fitted exponent is the asymptotic κ/ν entering Eq. (33). The paper itself cites free-energy-error fits δf∼χ^{-2κ} giving κ=2.2 for BWTRG and κ=2.0 for HOTRG [13,37]. Since for the Ising universality class ν=1, Eq. (33) then predicts δT_c∼χ^{-2.2}, not χ^{-4.0}. The factor ~1.8 discrepancy is not addressed. It may arise because the X1-jump estimator has additional correction-to-scaling exponents, or because χ=32–128 is pre-asymptotic with oscillatory δT_c (Fig. 9). If the true asymptotic κ/ν is ~2.2, the time exponents become 2.2/5≈0.44 for BWTRG versus 2.034/3≈0.68 for MPS, reversing the stated conclusion. No wall-clock measurements are provided to corroborate the asymptotic estimate.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a multi-impurity method for the bond-weighted tensor renormalization group (BWTRG), in which bond weights are replaced by impurity matrices in a triad tensor network to compute higher-order moments of physical observables. The method is formulated at O(chi^5) cost, and the authors benchmark it on the square-lattice Ising model and the five-state Potts model, reporting better accuracy than conventional TRG. They further use finite-size scaling of the dimensionless quantity X1 to extract critical exponents and the critical temperature with relative error below 10^-7. The paper's central efficiency claim is that BWTRG with the optimal hyperparameter k = -1/2 is more efficient in computational time than MPS-based approaches for estimating the critical temperature, based on a fitted exponent kappa/nu ~ 4.0 combined with the O(chi^5) scaling.","tokens_in":14203,"tokens_out":9212,"duration_ms":83652,"significance":"If the efficiency claim holds, the method is a valuable practical advance: it extends impurity techniques to a TRG-type algorithm with O(chi^5) cost, avoids the spreading of impurities through tensor decompositions, and delivers accurate higher-order moments in 2D classical models. The explicit update equations, the benchmarks against exact Onsager/Baxter results, and the open data repository are clear strengths. The main caveat is that the headline efficiency conclusion relies on an asymptotic exponent fitted over a narrow bond-dimension range and is not corroborated by wall-clock timings; the paper's own free-energy-error fits suggest a different exponent, so the central claim needs either additional support or qualification.","major_comments":[{"comment":"The abstract's efficiency claim rests on the fitted exponent kappa/nu ~ 4.0 for BWTRG with k = -1/2, combined with Eq. (33) to obtain an error scaling t^{-0.80} versus t^{-0.68} for MPS. However, the paper itself cites free-energy-error fits giving kappa = 2.2 for the same BWTRG method and kappa = 2.0 for HOTRG (Refs. [13,37]). Since nu = 1 for the 2D Ising universality class, Eq. (33) would then predict delta T_c ~ chi^{-2.2}, not chi^{-4.0}. This factor-of-1.8 discrepancy is not addressed. If the asymptotic kappa/nu is about 2.2, the time exponents become 2.2/5 = 0.44 for BWTRG versus 2.034/3 = 0.68 for MPS, reversing the stated conclusion. The authors should either provide direct wall-clock timings, extend the fits to larger chi with an explicit treatment of corrections to scaling, or substantially qualify the claim in the abstract.","section":"Sec. III.B, Eq. (33), Fig. 10"},{"comment":"The delta T_c data in Fig. 9 oscillate as a function of chi, and the fits are performed over only a factor-of-4 range (chi = 32 to 128). No wall-clock timings are reported. Because the efficiency comparison is derived from the asymptotic slope of delta T_c versus chi, the narrow, oscillatory data do not establish that the fitted exponent is the asymptotic kappa/nu entering Eq. (33). A direct timing measurement, or a demonstration that the exponent is stable under corrections to scaling, is needed before the computational-efficiency claim can be accepted as stated.","section":"Sec. III.B, Fig. 9"}],"minor_comments":[{"comment":"The definition of the latent heat contains a typo: 'L = E(T_c+0) - L(T_c-0)' should read 'L = E(T_c+0) - E(T_c-0)'.","section":"Sec. II.D, Eq. (19)"},{"comment":"The introduction states that the last section is devoted to discussion and conclusion, but the final section is titled only 'Conclusions' and contains no separate discussion; this sentence should be updated.","section":"Sec. I"},{"comment":"For reproducibility, the authors provide a data repository, but not the implementation code; providing the code used for the BWTRG impurity updates and the FSS analysis would be helpful, especially because the efficiency claim depends on implementation details.","section":"Sec. III.A"},{"comment":"The statement that the relative error in the estimated critical temperature is less than 10^-7 would be more informative if the actual numerical value of delta T_c for chi = 128 were quoted, since Fig. 9 uses a logarithmic scale and the precise value is not easily read off.","section":"Sec. III.B"}],"recommendation":"major_revision","confidential_remarks":"The algorithmic contribution is sound and the numerical benchmarks are convincing, but the efficiency claim in the abstract is stronger than the evidence presented. I would be willing to accept after the authors either provide direct timings or significantly qualify the claim, and after the discrepancy between the fitted kappa/nu and the free-energy-based exponent is addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the multi-impurity construction for BWTRG is a genuine, useful extension, and the numerical benchmarks are solid. The efficiency claim against MPS is shakier than the rest of the paper, because it rests on an exponent that conflicts with another exponent quoted in the same paper.\n\nThe new thing is replacing bond weights with impurity matrices in a triad network. That solves the impurity-spread problem that blocked TRG-based moment calculations, and it runs at O(chi^5), which is why they can push chi=150 for the 5-state Potts model. The derivations in Section II.B are careful, and the Ising and Potts results are consistent with exact solutions and CFT values. The FSS extraction of Tc to 10^-7 is impressive. The observation that kappa/nu varies with hyperparameter k is a useful addition to the BWTRG literature.\n\nThe soft spot is the head-to-head efficiency comparison with MPS. The authors fit delta Tc ~ chi^{-kappa/nu} and get kappa/nu ≈ 4.0 for BWTRG at k=-1/2. But they also cite free-energy-error fits giving kappa ≈ 2.2 for the same method. Since nu=1 for Ising, those two numbers disagree by almost a factor of two. The scaling relation delta Tc ~ xi^{-1/nu} is standard, so if kappa is really ~2.2, the time exponent becomes 2.2/5 ≈ 0.44, versus 2.034/3 ≈ 0.68 for VUMPS — which would reverse their conclusion. They never address this discrepancy. It may be that the delta Tc estimator has different correction-to-scaling behavior, or that chi=32–128 is pre-asymptotic (the oscillations in Fig. 9 suggest they are fitting an envelope). But they don't say that, and the abstract's efficiency claim is exactly the kind of thing a casual reader will take away.\n\nThat said, the paper's central contribution — the impurity method itself — doesn't depend on the MPS comparison. Even if the efficiency claim needs qualification, the method stands as a practical tool. The paper ships no code, only data, so independent reproduction requires reimplementation, which is a moderate barrier but not a flaw in the algorithm.\n\nThis paper is for people who run TRG-family computations in 2D statistical mechanics or lattice gauge theory. It deserves a serious referee. I'd send it out, with a request that the authors either reconcile the two kappa estimates or soften the efficiency claim.","headline":"Solid algorithmic extension with an over-reaching efficiency claim that the authors should be pushed to reconcile.","tokens_in":14732,"tokens_out":4135,"would_cite":true,"duration_ms":30301,"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":"This paper claims that replacing bond weights with impurity matrices lets BWTRG compute higher-order moments, and that at $k=-1/2$ it estimates the Ising critical temperature with relative error below $10^{-7}$ more efficiently than…","keywords":["bond-weighted tensor renormalization group","multi-impurity method","higher-order moments","finite-size scaling","Binder parameter","Ising model","Potts model","critical temperature"],"falsifier":"Compute $\\langle m^2\\rangle$ and $\\langle m^4\\rangle$ for the Ising model at $T=2.26$ with the proposed fixed-isometry impurity update at $\\chi=64$ and $\\chi=128$, then repeat with isometries recomputed from the SVD of each impurity-inserted network; if the two moment estimates differ by more than the truncation error, or if the estimated $T_c$ shifts by more than $10^{-7}$, the shared impurity-free isometry choice is the limiting approximation rather than a harmless simplification.","tokens_in":13664,"feed_emoji":"🧲","tokens_out":11550,"duration_ms":107741,"temperature":0.7,"pith_summary":"This paper proposes a way to compute higher-order moments of physical quantities, such as magnetization and energy, inside the bond-weighted tensor renormalization group (BWTRG) by replacing bond weights with impurity matrices rather than replacing tensors. The authors argue that this keeps the isometric tensors shared across all moments, so the cost stays $O(\\chi^5)$ while conventional HOTRG-based impurity computations cost $O(\\chi^7)$. On the square-lattice Ising and five-state Potts models, the method gives energy and magnetization curves much closer to exact results than standard TRG, and it locates the Ising critical temperature with relative error below $10^{-7}$. The paper also shows that a dimensionless fixed-point quantity $X_1$ obeys the same finite-size scaling relation as the Binder parameter and takes its predicted CFT value at criticality. The central efficiency claim is that at the optimal hyperparameter $k=-1/2$, BWTRG estimates the critical temperature faster than MPS-based methods once computational cost is accounted for.","feed_headline":"Impurity trick sharpens critical-temperature estimates to 1e-7","feed_subtitle":"Replacing bond weights with impurity matrices computes magnetization moments faster than MPS and beats its cost scaling.","key_machinery":"The load-bearing object is the bond-weighted triad tensor network: a square lattice of three-index isometric tensors $A_i$ at plaquette corners connected by diagonal bond weights $\\sigma_j$ (inner) and $\\tau_j$ (outer). The mechanism is the substitution rule of Eqs. (6)-(9): the renormalized inner impurity matrix is $\\tilde S_j = \\tau_j^{-k'} T_j \\tau_j^{-k'}$, with $k'=(1-k)/2$, and the renormalized outer impurity matrix is the weighted average, over all placements of $n$ impurities, of a multilinear map $R(M_1,M_2,M_3,M_4)$ built from the current tensors and weights. Because the isometries and bond weights come from the same impurity-free SVD for every impurity configuration, the systematic summation over impurity positions carries all moments through the renormalization at $O(\\chi^5)$ cost; the hyperparameter $k$ tunes the accuracy and is set to $-1/2$.","core_discovery":"The central claim is that in a bond-weighted triad tensor network, an impurity matrix inserted at the position of a bond weight can represent a physical observable, and that systematic summation over impurity positions yields renormalized impurity matrices for moments of any order. Because the isometric tensors are computed once from the impurity-free singular value decomposition and shared by all moments, multiple impurities can be handled without repeated decompositions, making the algorithm $O(\\chi^5)$. Numerically, for the Ising model at $\\chi=128$ the method reproduces energy and magnetization consistent with exact results, improves on TRG across the hyperparameter $k$, and gives critical exponents and $T_c$ estimates that agree with exact values to 0.39% for $1/\\nu$, 2.2% for $2\\beta/\\nu$, and $10^{-7}$ for $T_c$. For the five-state Potts model at $\\chi=150$, the jumps in energy and magnetization at the first-order transition are substantially better estimated than by TRG. The paper further claims that the exponent $\\kappa/\\nu$ controlling $\\delta T_c \\sim \\chi^{-\\kappa/\\nu}$ varies continuously with $k$ and reaches about 4.0 at $k=-1/2$, larger than the CFT value for MPS, and that after comparing $O(\\chi^5)$ with $O(\\chi^3)$ cost the BWTRG error decays as $t^{-0.80}$ versus $t^{-0.68}$ for VUMPS.","pith_inferences":["The bond-weight-substitution idea should transfer directly to BWTRG variants on other geometries and to Grassmann tensor networks for fermions, where the same $O(\\chi^5)$ moment calculation would apply; the paper does not test these cases.","Because the isometries are shared by all moments, the fourth-order Binder cumulant and higher reduced cumulants can be extracted from the same run; their $\\chi$-convergence would be a sharp test of whether the fixed-isometry approximation biases moments.","The measured $\\kappa/\\nu \\simeq 4.0$ at $k=-1/2$, compared with $\\kappa_{\\mathrm{CFT}}=2.034$ for MPS, suggests that the effective correlation length in BWTRG is not governed by the same finite-entanglement mechanism; identifying that mechanism could turn $k$ into a tunable knob for critical-point calculations.","Since the isometries ignore temperature and field dependence, the energy and magnetization errors exceed the free-energy error; a hybrid using environment-tensor-improved isometries while keeping the impurity update could reduce those errors without leaving $O(\\chi^5)$ scaling."],"forward_implications":["Moments of arbitrary order, for example the Binder ratio $\\langle m^4\\rangle/\\langle m^2\\rangle^2$, are available from one BWTRG run at $O(\\chi^5)$ cost, allowing bond dimensions of 128-150 that were impractical with the earlier $O(\\chi^7)$ HOTRG impurity method.","At $k=-1/2$, the energy and magnetization of the Ising and five-state Potts models are consistently more accurate than TRG at equal bond dimension, with the relative error crossing the exact value near the optimal hyperparameter.","The dimensionless fixed-point quantity $X_1$ satisfies the same $g(L^{1/\\nu}t)$ scaling form as the Binder parameter and takes its CFT value at criticality, so it can be used to locate $T_c$ and $\\nu$ without fitting the magnetization amplitude.","The estimated critical temperature follows $\\delta T_c \\propto \\chi^{-\\kappa/\\nu}$ with $\\kappa/\\nu\\simeq 4.0$ at $k=-1/2$; after accounting for the $O(\\chi^5)$ cost, the error decays as $t^{-0.80}$, beating the $t^{-0.68}$ of VUMPS.","BWTRG at $\\chi=128$ avoids the rank reduction caused by the corner double-line redundancy up to $L=2^{25}$, whereas TRG at the same bond dimension rank-reduces at $L=2^{20}$, so the reported scaling is governed by finite size rather than finite entanglement."],"supporting_citations":[{"why":"Introduces TRG, the baseline accuracy that the proposed impurity method is shown to exceed.","marker":"[1]"},{"why":"Introduces BWTRG and identifies $k=-1/2$ as the optimal hyperparameter that the impurity update inherits.","marker":"[13]"},{"why":"Provides the earlier multi-impurity HOTRG method, establishing the $O(\\chi^7)$ baseline and the higher-order-moment formalism this paper extends to BWTRG.","marker":"[17]"},{"why":"Supplies exact latent-heat values for the five-state Potts transition used to judge the accuracy of the impurity method.","marker":"[26]"},{"why":"Introduces the Binder parameter whose finite-size scaling relation is the model for $X_1$.","marker":"[28]"},{"why":"Defines the dimensionless fixed-point quantity $X_1$ whose scaling form and CFT universal value anchor the critical-temperature analysis.","marker":"[30]"},{"why":"Supplies the VUMPS algorithm used as the MPS-based comparison for critical-temperature estimation.","marker":"[33]"},{"why":"Gives the finite-entanglement scaling exponent $\\kappa_\\mathrm{CFT}$ for matrix product states that underlies the efficiency comparison.","marker":"[35]"},{"why":"Reports the free-energy scaling exponents for HOTRG and BWTRG that the paper compares with its measured $\\kappa/\\nu$.","marker":"[37]"}],"fun_headline_variants":["Multi-impurity trick sharpens tensor renormalization","Impurity matrices boost BWTRG precision and speed","Multi-impurity method beats TRG for Ising, Potts","Higher-order moments via multi-impurity BWTRG","Impurity insertion refines critical temperature estimates"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that the isometric tensors obtained from the impurity-free singular value decomposition remain accurate when inserted into impurity networks, so that fixed truncation does not systematically bias the computed higher-order moments.","fun_headline_variants_meta":{"raw":{"variants":["Multi-impurity trick sharpens tensor renormalization","Impurity matrices boost BWTRG precision and speed","Multi-impurity method beats TRG for Ising, Potts","Higher-order moments via multi-impurity BWTRG","Impurity insertion refines critical temperature estimates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00023,"raw_usage":{"total_tokens":1526,"prompt_tokens":1031,"completion_tokens":495,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":647,"completion_tokens_details":{"reasoning_tokens":415}},"tokens_in":647,"tokens_out":495,"duration_ms":5672,"temperature":1.0,"reasoning_tokens":415,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:39:30.099744+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $\\langle m^2\\rangle$ and $\\langle m^4\\rangle$ for the Ising model at $T=2.26$ with the proposed fixed-isometry impurity update at $\\chi=64$ and $\\chi=128$, then repeat with isometries recomputed from the SVD of each impurity-inserted network; if the two moment estimates differ by more than the truncation error, or if the estimated $T_c$ shifts by more than $10^{-7}$, the shared impurity-free isometry choice is the limiting approximation rather than a harmless simplification.","supporting_citations":[{"cited_title":"Levin and C","cited_arxiv_id":null,"evidence_quote":"Introduces TRG, the baseline accuracy that the proposed impurity method is shown to exceed."},{"cited_title":"Homma, T","cited_arxiv_id":null,"evidence_quote":"Introduces BWTRG and identifies $k=-1/2$ as the optimal hyperparameter that the impurity update inherits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the earlier multi-impurity HOTRG method, establishing the $O(\\chi^7)$ baseline and the higher-order-moment formalism this paper extends to BWTRG."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the Binder parameter whose finite-size scaling relation is the model for $X_1$."},{"cited_title":"Binder, Critical properties from Monte Carlo Coarse graining and renormalization, Phys","cited_arxiv_id":null,"evidence_quote":"Defines the dimensionless fixed-point quantity $X_1$ whose scaling form and CFT universal value anchor the critical-temperature analysis."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the VUMPS algorithm used as the MPS-based comparison for critical-temperature estimation."},{"cited_title":"Vanderstraeten, J","cited_arxiv_id":null,"evidence_quote":"Gives the finite-entanglement scaling exponent $\\kappa_\\mathrm{CFT}$ for matrix product states that underlies the efficiency comparison."},{"cited_title":"Pirvu, G","cited_arxiv_id":null,"evidence_quote":"Reports the free-energy scaling exponents for HOTRG and BWTRG that the paper compares with its measured $\\kappa/\\nu$."}],"review_version":1}