{"id":"4f993bcd-3b0c-4e81-abcc-6aabbe38c22d","arxiv_id":"2509.02185","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":4,"one_line_summary":"A tensor renormalization group algorithm computes entanglement entropy for arbitrary single-interval subsystems and reproduces c=0.49997(8) in the 2D Ising model.","lead":"This paper introduces a tensor-network algorithm that computes entanglement entropy for a single interval of any size in one-dimensional quantum systems. It validates the method on the critical Ising model, recovering the expected central charge c=0.5 to four decimal places.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The trimmed network ρ̃_A is never shown to be a positive semidefinite density matrix; the von Neumann logarithm and the isometry-trimming identity require PSD, so the central-charge result may be evaluating an ill-defined quantum entropy.","rationale":"The paper's strongest claim is that the TRG algorithm computes entanglement entropy for arbitrary single-interval subsystems, validated by reproducing c = 0.5 in the critical Ising model. For that claim to hold, the quantity being computed must actually be a von Neumann entropy. The weakest point is therefore the unverified positivity of the approximate reduced density matrix: both the definition S_A = −Tr(ρ_A log ρ_A) and the trimming identity of Sec. 2.2 require ρ̃_A to be PSD. This is an internal-consistency issue, not a disagreement with any external consensus, and it is exactly the assumption the reader identified. The numerical agreement with CFT is real evidence, but it does not by itself rule out the possibility that HOTRG truncation introduces small negative eigenvalues that are ignored by the diagonalization. The proposed spectral check is decisive and inexpensive. Other concerns—post hoc fit ranges, no released code—are real but secondary; they do not change the verdict. Because the reader already returned a moderate-confidence conditional verdict and this check is precisely the missing validation, no adjustment to that verdict is needed.","tokens_in":10401,"tokens_out":7597,"duration_ms":95256,"concrete_test":"Reproduce the L=1024, α=16, D=96, ℓ=512 run. After the final coarse-graining step, form the trimmed matrix ρ̃_A and compute its full eigenspectrum. Report the minimum eigenvalue, the number and total weight of negative eigenvalues, and compare S_A computed from the full spectrum with S_A computed from the positive-semidefinite part (negative eigenvalues projected away). Also verify Hermiticity by computing ||ρ̃_A − ρ̃_A†||_F. If the minimum eigenvalue is negative at a level that changes S_A by more than the quoted error, the central claim is not supported; if the spectrum is nonnegative and Hermiticity holds to numerical precision, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central derivation in Sec. 2.2 evaluates S_A = −Tr(ρ̃_A log ρ̃_A) after HOTRG truncation and isometry trimming. Two load-bearing assumptions are unstated: (i) the object ρ_A built from the classical Ising tensor (3.1) is a legitimate density matrix—Hermitian, positive semidefinite, trace 1—and (ii) the HOTRG truncations preserve positive semidefiniteness, so that the final ρ̃_A is still a density matrix. The paper itself places 'density matrix' in quotes in Sec. 3.1 and never checks Hermiticity or positivity of ρ̃_A. The identity used to discard isometries, Tr(U ρ̃ U† log(U ρ̃ U†)) = Tr(ρ̃ log ρ̃), is valid for an isometric U only when ρ̃ is PSD; otherwise the matrix logarithm is multivalued or undefined and the von Neumann entropy formula has no meaning. Tensor-network truncations that discard small singular values are not guaranteed to preserve positivity. If a small number of negative eigenvalues appear and are silently discarded during diagonalization, the quoted c = 0.49997(8) could be an accidental agreement rather than a demonstration that the method computes entanglement entropy. Thus the central claim is not fully secured until the spectrum of ρ̃_A is shown to be nonnegative.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a tensor renormalization group (TRG) method for computing the entanglement entropy of a single-interval subsystem in a one-dimensional quantum system. The reduced density matrix is represented as a (1+1)-dimensional tensor network, and the authors develop a HOTRG-based coarse-graining procedure in which the isometries that do not affect the entropy are trimmed away, leaving a smaller matrix whose von Neumann entropy approximates the desired entanglement entropy. The method is designed for arbitrary interval size, not just half-space, with the final matrix size controlled by the Hamming weight of the binary representation of the interval length. The method is tested on the two-dimensional classical Ising model at criticality, and the extracted central charge c=0.49997(8) at D=96 agrees with the CFT prediction c=0.5, with consistent results for D=64,80 and for fixed ratios x=ℓ/L.","tokens_in":10774,"tokens_out":6777,"duration_ms":89458,"significance":"If the method is sound, it offers a potentially valuable sign-problem-free route to entanglement entropy without the replica trick, extending previous TRG-based work that was largely limited to half-space partitions. The binary-trimming construction is conceptually elegant and the numerical agreement with c=0.5 across several bond dimensions and subsystem ratios is encouraging. The paper also gives a clear computational-cost estimate in terms of the Hamming weight. However, the central claim depends on the positivity of the object whose von Neumann entropy is evaluated, and this is not established; the numerical validation also relies on fit ranges and error estimates that are not fully standard. These issues need to be addressed before the method can be considered reliably demonstrated.","major_comments":[{"comment":"The derivation uses the identity Tr(U ρ̃ U† log(U ρ̃ U†)) = Tr(ρ̃ log ρ̃) to discard isometries, and evaluates S_A as −Tr(ρ̃_A log ρ̃_A). This is valid only when ρ̃_A is Hermitian positive semidefinite. The paper never verifies positivity of the initial \"density matrix\" (the quotes in §3.1 are telling) or its preservation under HOTRG truncation and the trimming step. For a closed network the classical Ising tensor (3.1) gives positive Boltzmann weights, but the reduced density matrix with open indices, and especially the truncated/trimmed network, need not be PSD. If any negative eigenvalues appear, the von Neumann entropy is not well defined, the trace identity fails, and the numerical agreement with c=0.5 could be accidental. The authors should report the spectrum of ρ̃_A for the parameters used in Table 1, or prove that positivity is preserved by their coarse-graining/trimming. This i","section":"§2.2 and §3.1"},{"comment":"The error estimate for the central charge is obtained by fixing k1 to its fitted central value and taking the maximal deviation of solutions of Eq. (3.2) for individual ℓ. This ignores the covariance between c and k1 and the uncertainty in the choice of fit ranges (7≤ℓ≤768 and the plateau 16≤xL≤128 are selected from the data). The reported errors such as 0.49997(8) are therefore not a reliable measure of the method's accuracy. A standard fitting procedure with full covariance, or a bootstrap over the fit range, should be reported. This is important because the central validation claim is the agreement with c=0.5.","section":"§3.2"}],"minor_comments":[{"comment":"Typo: \"subsysmtem\" should be \"subsystem\".","section":"§2.1"},{"comment":"Typo: \"T able 1\" should be \"Table 1\".","section":"Table 1"},{"comment":"The labels and the binary-path markings in Fig. 5 are hard to read at normal print size; please use larger fonts and clearer arrows.","section":"Fig. 5"},{"comment":"In the definition of the effective central charge, it would help to state explicitly that L is fixed and that the formula uses S_A(L,ℓ) and S_A(L,ℓ′). The notation is clear from context but should be spelled out.","section":"Eq. (3.3)"},{"comment":"The paper calls the object in Fig. 1b a \"density matrix\" in quotation marks. This is confusing because the numerical test uses the isotropic classical Ising model. The authors should either justify the terminology or use a neutral term such as \"tensor network state\" and clarify the relation to a genuine quantum density matrix.","section":"§3.1"}],"recommendation":"major_revision","confidential_remarks":"The main concern is the positivity issue in §2.2: the method's output is a von Neumann entropy only if the trimmed tensor network is a positive semidefinite density matrix. This is not checked. The numerical agreement with c=0.5 is suggestive but not conclusive if the logarithm is taken of a matrix with negative eigenvalues. The paper is otherwise well within the scope of hep-lat/TRG, and the construction is original and potentially publishable, but the authors need to provide a spectral check or a positivity argument, and to strengthen the error analysis in the fits."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look. The main algorithmic idea—using HOTRG isometries to trim the coarse-grained reduced density matrix and encoding the interval length in binary—is new as far as I know, and it directly removes the half-space restriction of [25–28]. The derivation in Sec. 2.2 is coherent; the isometry identity is correct when the operator it is applied to is positive semidefinite, and the trimmed network is genuinely simpler (D^2×D^2 instead of D^3×D^3 for the examples).\n\nThe numerical test is well designed: it checks convergence in the temporal extent α, extracts the central charge both by varying ℓ at fixed L and by varying L at fixed x = ℓ/L, and gets c = 0.49997(8) at D=96 with plateau values around 0.5001. That is strong evidence that the algorithm computes what it claims.\n\nThe soft spots are mostly things the paper should have checked rather than flaws in the idea. The main one is the positivity of ρ_A. The paper itself puts “density matrix” in quotes in Sec. 3.1, and a matrix with nonnegative entries is not automatically positive semidefinite. The trace identity and the von Neumann logarithm only make sense for a PSD operator. There may be a good argument from the even-power Ising transfer matrix that ρ_A is PSD, but the paper does not give it, and no spectrum of ρ̃_A is reported. If small negative eigenvalues are present, the central-charge agreement could be accidental. I would not call this fatal, but it is load-bearing enough that the referee should ask for the eigenvalue check.\n\nSecond, the fit ranges are chosen after looking at the effective central-charge plateaus, and the error is estimated from the maximal deviation rather than from a systematic fit procedure. That is post hoc and probably optimistic. Minor: no code or data is released, so a reimplementation is needed to verify the claims.\n\nThe citation pattern is fine: prior half-space TRG work is credited, and the new binary trimming is clearly distinguished. For lattice/TRG people who want a sign-problem-free, replica-free route to entanglement entropy, this is a useful tool. It deserves proper peer review; the right referee report would ask for the positivity check, a more defensible fit-range analysis, and ideally code.","headline":"A genuinely new binary-trimming construction for TRG entanglement entropy with a convincing Ising central-charge test, though the paper never checks positivity of the effective density matrix.","tokens_in":11217,"tokens_out":4406,"would_cite":true,"duration_ms":55110,"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 tensor renormalization group method computes entanglement entropy for single intervals of arbitrary size, giving central charge c=0.49997(8) for the critical Ising model.","keywords":["tensor renormalization group","entanglement entropy","reduced density matrix","HOTRG","central charge","Ising model","von Neumann entropy","tensor network"],"falsifier":"Take a small lattice, such as L=16 or 32, where the reduced density matrix can be formed exactly before coarse-graining, run the HOTRG plus trimming procedure, and inspect the final matrix. If any eigenvalue of the trimmed matrix is negative, or if its von Neumann entropy differs from the exact value by more than the truncation error, the method's central claim fails.","tokens_in":10340,"feed_emoji":"🔗","tokens_out":5905,"duration_ms":63323,"temperature":0.7,"pith_summary":"The paper proposes a tensor renormalization group (TRG) method to compute the entanglement entropy of a one-dimensional quantum system for a single-interval subsystem of any size, not just the half-space cases studied before. The key move is to represent the reduced density matrix as a two-dimensional tensor network, coarse-grain it with the higher-order TRG algorithm, and then trim away isometry factors that do not affect the von Neumann entropy, leaving a much smaller matrix. The structure of the leftover network is set by the binary expansion of the interval length, so the final matrix has size set by the Hamming weight of that length. Tested at the critical point of the two-dimensional Ising model, the method reproduces the conformal-field-theory formula and gives central charge c=0.49997(8) at bond dimension D=96, consistent with the exact c=1/2. If it works generally, it offers a way to compute entanglement entropy in theories where Monte Carlo methods face sign problems.","feed_headline":"Tensor-network method computes entanglement entropy for any interval","feed_subtitle":"By trimming coarse-grained tensors, the method reproduces the Ising model's central charge c=0.49997(8).","key_machinery":"The central object is the trimmed tensor network ρ~A, obtained from the HOTRG coarse-grained representation of the reduced density matrix by eliminating isometry pairs UU† whose contraction is the identity. It carries the argument because it shrinks a D^ℓ × D^ℓ matrix to D^h × D^h with h the Hamming weight of the binary expansion of ℓ, while preserving the von Neumann entropy. The load-bearing identity is Tr(UρU† log(UρU†)) = Tr(ρ log ρ), valid when U†U=I, which licenses the trimming.","core_discovery":"The paper claims that for a one-dimensional quantum system, the reduced density matrix of a single-interval subsystem can be coarse-grained with HOTRG and then reduced to a much smaller 'trimmed' tensor network whose von Neumann entropy equals the original entanglement entropy up to truncation error. The arrangement of the leftover isometries is controlled by the binary expansion of the interval length ℓ, and the final matrix has dimension set by the Hamming weight of ℓ. Testing this on the two-dimensional Ising model at criticality, the method reproduces the CFT scaling formula and yields c=0.49997(8) at D=96, matching c=1/2.","pith_inferences":["A natural next test is to apply the method to a theory without a known CFT prediction, where the extracted central charge could be cross-checked against other methods; the paper validates only the Ising case.","Because the trimming relies on the algebraic identity for the von Neumann entropy, the method implicitly assumes the approximate reduced density matrix is positive semidefinite; checking the eigenvalue spectrum after truncation would make the method's range of validity concrete.","For very long intervals with many 1 bits in their binary expansion, the cost exponent grows with Hamming weight, so choosing subsystem sizes whose binary form is sparse could make larger or higher-dimensional computations practical.","The binary-box structure suggests similar trimming could produce inexpensive estimators for Rényi entropies or mutual information by replacing the logarithm with other functions of the reduced density matrix."],"forward_implications":["Computing entanglement entropy no longer requires the replica trick and the n→1 extrapolation; the entropy is evaluated directly from the approximate reduced density matrix.","Because TRG is free from sign problems, the method is a candidate for entanglement-entropy computations in theories where Monte Carlo methods struggle.","For intervals whose length has a small Hamming weight, such as powers of two, the extra cost is modest, keeping large-system simulations feasible.","The same construction extends to higher-dimensional quantum systems for hyperrectangular subsystems, with (d−1)-dimensional isometries replacing the scalar ones.","The central charge extracted at criticality gives a direct numerical check of universal CFT predictions in lattice models."],"supporting_citations":[{"why":"Supplies the HOTRG coarse-graining algorithm whose spatial isometries the trimmed network keeps.","marker":"[21]"},{"why":"Provides the quantum-classical mapping identifying the 1D quantum Ising model with the 2D classical Ising model used in the numerical test.","marker":"[29]"},{"why":"Gives the conformal-field-theory formula for entanglement entropy used to fit the central charge.","marker":"[30]"},{"why":"Earlier TRG computation of entanglement entropy for half-space subsystems, the case this paper extends to arbitrary interval sizes.","marker":"[25]"},{"why":"Foundational TRG algorithm that showed tensor renormalization can evaluate partition functions of 2D lattice models.","marker":"[17]"}],"fun_headline_variants":["Tensor renormalization computes entanglement entropy","Entanglement entropy via coarse-grained tensor networks","TRG method yields Ising central charge with high precision","Arbitrary-interval entropy from tensor renormalization","Entanglement entropy from trimmed tensor networks"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The method treats the approximate reduced density matrix as a genuine quantum state; if its eigenvalues are not all non-negative, the logarithm in the entropy is not defined.","fun_headline_variants_meta":{"raw":{"variants":["Tensor renormalization computes entanglement entropy","Entanglement entropy via coarse-grained tensor networks","TRG method yields Ising central charge with high precision","Arbitrary-interval entropy from tensor renormalization","Entanglement entropy from trimmed tensor networks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000252,"raw_usage":{"total_tokens":1340,"prompt_tokens":627,"completion_tokens":713,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":371,"completion_tokens_details":{"reasoning_tokens":643}},"tokens_in":371,"tokens_out":713,"duration_ms":8043,"temperature":1.0,"reasoning_tokens":643,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T11:47:30.178634+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small lattice, such as L=16 or 32, where the reduced density matrix can be formed exactly before coarse-graining, run the HOTRG plus trimming procedure, and inspect the final matrix. If any eigenvalue of the trimmed matrix is negative, or if its von Neumann entropy differs from the exact value by more than the truncation error, the method's central claim fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the quantum-classical mapping identifying the 1D quantum Ising model with the 2D classical Ising model used in the numerical test."},{"cited_title":"Doubling of Entanglement Spectrum in Tensor Renormalization Group","cited_arxiv_id":"1306.6829","evidence_quote":"Earlier TRG computation of entanglement entropy for half-space subsystems, the case this paper extends to arbitrary interval sizes."}],"review_version":1}