{"id":"4ca88796-6b02-4115-a71c-8b5037784158","arxiv_id":"2507.14732","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Tensor network simulations of the non-Hermitian 5-state Potts model reveal the predicted spiral RG flow and a boundary spectrum matching complex conformal field theory.","lead":"This paper uses tensor network simulations to study a non-Hermitian version of the 5-state Potts model, refining its complex critical point and observing a spiral pattern in its renormalization group flow. It also matches the model's entanglement spectrum to predictions from complex conformal field theory, supporting the idea that weakly first-order transitions hide complex fixed points.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The paper's key premise—that SVD truncation of |ψ_R⟩ controls the error of the non-Hermitian MPS—is explicitly unquantified in SI Sec. E; every headline result (λ_c, RG spiral, entanglement spectrum) inherits this uncontrolled approximation.","rationale":"The reader's weakest_assumption matches mine, so agreement is agree. I considered the circularity of using theoretical Δ_th to locate λ_c as the main alternative; it is real but less load-bearing because SI Sec. B reports a data collapse at the conjugate pair of fixed points that does not use the conformal dimensions, providing a partially independent check. The MPS truncation issue, by contrast, is explicitly marked as unresolved in SI Sec. E and affects the entire chain of results: λ_c, g_epsilon' spiral, and entanglement spectrum. The abstract's 'up to L=28' discrepancy and the known slow convergence of imaginary parts are secondary. The proposed check is a direct benchmark with algorithms designed for non-Hermitian tensor networks, using the same quantities and system sizes. Passing it would make the central claim credible; failing it would reduce the claims to the ED-accessible regime. Since the reader already issued a CONDITIONAL verdict and this concern is addressable by an external benchmark, no verdict change is needed.","tokens_in":14131,"tokens_out":6347,"duration_ms":78809,"concrete_test":"Run the same protocol with a biorthonormal non-Hermitian algorithm (e.g., biorthonormal-block DMRG, PRL 135, 106502, or the fidelity algorithm, PRB 105, 205125) at L=16 and L=24, χ=400, at λ_c=0.0788+0.0603i. Compare: (1) ground-state energy E0; (2) the σ and ε gaps used in Eq. (4); (3) the fifteen largest eigenvalues of ρ_RL per Z5 sector (Fig. 4); and (4) the fixed-point estimate from the same subleading 1/L^2 FSS. If the two methods agree within the ED-checked tolerance for L=16 and the L=24 shifts are below the finite-size scatter of Fig. 3, the MPS-truncation concern is resolved. If λ_c moves by more than about 1e-4 in either Re or Im, or the entanglement spectrum reorders, the paper's claim that unadapted DMRG/QPA accurately captures the complex fixed point is not established at the quoted precision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing assumption is that an MPS obtained by truncating the right eigenvector |ψ_R⟩ accurately represents the non-Hermitian ground state. The Tensor Networks section simply states 'we assume that the Schmidt coefficients decay fast enough', and SI Sec. E concedes: 'Due to the observed disconnect between singular and eigenvalues, it remains unknown how much entanglement is actually being neglected by the MPS approximation.' This matters because ρ_RL = |ψ_L⟩⟨ψ_R|/⟨ψ_R|ψ_L⟩ is not positive semidefinite, so the standard DMRG singular-value truncation has no variational lower bound and no error estimate. ED agreement at L=12 and bond-dimension saturation at L=24 bracket the accessible regime, but the refined value λ_c is extracted from fits through L=24 and the entanglement spectrum uses L up to 64, where no independent algorithm validates the truncation. SI Sec. E explicitly says the approximation is expected to be worse for excited states, and the QPA excited states feed directly into the energy gaps used to determine λ_c and the conformal data. Thus all three headline deliverables inherit an uncontrolled error. This is not an accusation of dishonesty; it is an unverified premise that the paper itself flags.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies the non-Hermitian deformation of the 5-state Potts model proposed by Tang et al. and claims that standard MPS methods (DMRG and QPA), despite the breakdown of the variational principle, faithfully capture the complex fixed point. The authors determine a refined critical coupling λ_c = 0.0788 + 0.0603i by finite-size scaling of conformal dimensions, extract bulk conformal data, observe a spiral RG flow of the coupling g_ε′ in agreement with the CCFT prediction, and reconstruct a boundary entanglement spectrum matching the free-free conformal boundary condition. The paper includes an extensive supplementary discussion of error sources and a data-collapse cross-check of the fixed point.","tokens_in":14429,"tokens_out":3520,"duration_ms":40978,"significance":"If the central claims are correct, this is a valuable demonstration that tensor network algorithms can simulate non-Hermitian deformations of weakly first-order transitions at complex fixed points, going beyond exact diagonalization and providing numerical evidence for the CCFT scenario of the 5-state Potts model. The paper is unusually candid in documenting the limitations of the non-Hermitian MPS approximation, and the data-collapse check in Supplement B is a partial independent validation. However, the headline results inherit two load-bearing issues: the uncontrolled truncation of the non-Hermitian density matrix and the use of theory-derived conformal data in the fixed-point determination while simultaneously comparing extracted conformal data with those same inputs.","major_comments":[{"comment":"The central computational assumption is that truncating the right eigenvector |ψ_R⟩ by its singular values yields a controlled approximation to the non-Hermitian ground state. The manuscript itself states in Sec. E that \"due to the observed disconnect between singular and eigenvalues, it remains unknown how much entanglement is actually being neglected by the MPS approximation.\" Because the variational principle is invalid, DMRG provides no error bound, and the ED comparison at L=12 and bond-dimension saturation at L=24 do not validate the L=64 entanglement data or the QPA excited states used in the RG flow. This affects all three headline claims. The authors should quantify the truncation error (for example by comparing against a biorthonormal DMRG variant or by presenting convergence in the entanglement spectrum with respect to discarded weight in ρ_RL), or explicitly restrict claims to system sizes where an independent validation exists.","section":"Tensor networks / Supp. Sec. E"},{"comment":"The fixed-point search uses theoretical scaling dimensions and OPE coefficients as input: Eq. (4) defines δΔ_n relative to Δ_th^n, and Eq. (5) minimizes J(v,g_ε′) built from the theoretical C_ϕϕε′. The extracted conformal data in Table I are then compared with those same theoretical values. This is a consistency check, not an independent verification. The data collapse in Supp. Sec. B is a genuinely independent confirmation for the σ operator, which strengthens the paper, but it is not used for the main quoted precision and does not cover the other operators. Please reframe the conformal-data section accordingly and, if possible, provide a quantitative measure of the degree of independence between the input theory and the extracted numbers.","section":"Conformal data / Eq. (4) / Supp. Sec. B"},{"comment":"The paper reports that Im(Δ_ε) scales as 1/L^1.05 rather than the expected 1/L^2 and states that \"it remains unclear whether this observation has a deeper meaning or if it is due to the same simulation artifacts.\" Nevertheless, Table I lists Im(Δ_ε) = −0.2245(1)i fitted with 1/L^2. This unresolved discrepancy means the quoted precision of the imaginary part of ε is not supported by the presented scaling analysis. Either fit with the observed power law and state the resulting uncertainty, or identify the origin of the anomalous exponent.","section":"Fig. 3 / Table I"},{"comment":"The identification ρ_RL = e^{−2πK_C} is assumed rather than derived, and the paper acknowledges that ρ_RL is not positive semi-definite. The agreement with the free-free boundary CCFT is argued from degeneracies and the first two fitted levels of the zero sector; however, the imaginary parts converge more slowly and no error estimates are given for the extracted levels. Please provide a quantitative comparison (for example, residuals for all identified towers) and discuss how the non-positive semi-definiteness of ρ_RL affects the logarithm in Eq. (7).","section":"Complex entanglement spectrum / Eq. (7)"}],"minor_comments":[{"comment":"The symbols C and C are introduced only through the phrase \"fixed points C and C\"; since the conjugate fixed point is denoted in several places by an overline, this notation should be defined explicitly at first use.","section":"Introduction"},{"comment":"In the sentence \"The best subleading scaling for the real and imaginary part of the six most relevant operators is visible in Fig. 3,\" the six operators are not enumerated in the main text; please list them explicitly.","section":"Conformal data"},{"comment":"The statement \"we use χ=400 and χ=600 for the ground state and excitation calculations\" should specify which value applies to which algorithm and whether the same bond dimensions are used for the L=64 entanglement spectrum.","section":"Tensor networks"},{"comment":"The phrase \"the maximal reach of ED methods L∼12 barely captures the right physics due to the logarithmic finite-size effects\" would benefit from a reference or a quantitative definition of the logarithmic corrections.","section":"Conformal data"},{"comment":"Equation (S.5) is cited in the main text before the supplementary material is introduced; please ensure the cross-referencing is clear to the reader.","section":"Conformal data / Supplementary"}],"recommendation":"major_revision","confidential_remarks":"The paper is honest about its limitations, which is commendable, but the uncontrolled MPS truncation and the circularity of the fixed-point determination are load-bearing issues that should be resolved or substantially reframed before publication. The data-collapse section in Supplement B is an asset and should be moved more centrally if journal space permits, since it provides the most independent evidence for the critical value."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. The paper's main deliverables are a refined critical coupling λ_c = 0.0788 + 0.0603i for the non-Hermitian deformed 5-state Potts model, the first direct lattice observation of the spiral RG flow of g_ε′ on a 1+1D chain at system sizes beyond exact diagonalization, and an entanglement spectrum that matches the free-free boundary CCFT spectrum with the expected degeneracies. The supplemental material is unusually honest about the method's limitations, which is a real point in its favor.\n\nThe strongest part is the convergence evidence. DMRG and QPA energies track ED to tolerance at L=12, bond-dimension saturation is shown at L=24, and the entanglement spectrum is stable with χ. The data collapse in Fig. S2 is also a genuinely independent check: it fixes λ_c without using the conformal data that go into the cost function. That mitigates the circularity concern you raised, at least for the σ operator.\n\nThe soft spots are real but, I think, not disqualifying. The MPS truncation issue is the most serious. The paper itself says in Sec. E that it 'remains unknown how much entanglement is actually being neglected,' and that uncertainty is load-bearing for all three headline results. But the paper doesn't ignore it; it documents the disconnect between singular values and eigenvalues, explains why the ordering property makes the approximation useful, and flags that excited states are worse. So it's an unquantified systematic uncertainty, not a demonstrated failure. A referee should ask for a quantitative error bound or at least a more careful study of the imaginary parts.\n\nThe circularity is partially there: the fixed-point search uses theoretical scaling dimensions and OPE coefficients as input. But the data collapse is a solid counterweight. The unresolved scaling of Im(Δ_ε) (1/L^2 vs. 1/L^1.05) is minor and acknowledged.\n\nOne small inconsistency: the abstract says L=28, but the text says the ground-state simulations are reliable up to L=24, with L=64 only for the entanglement spectrum. That should be fixed.\n\nOverall: a careful numerical paper that moves the subfield forward, with honest limitations. It deserves a serious referee. I'd be willing to cite it for the refined λ_c and the first spiral-flow observation.","headline":"A careful numerical study that delivers the first lattice spiral-flow observation and a refined critical coupling, with honest limitations that a referee should push on but that do not sink the paper.","tokens_in":14935,"tokens_out":2305,"would_cite":true,"duration_ms":26331,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Ordinary tensor-network algorithms can simulate the non-Hermitian 5-state Potts model at its complex fixed point, yielding the spiral RG flow and a boundary CCFT spectrum.","keywords":["non-Hermitian quantum spin chains","5-state Potts model","complex conformal field theory","complex fixed points","matrix product states","DMRG","renormalization group flow","entanglement spectrum"],"falsifier":"A direct test would be to compute, at $\\lambda_c$ and $L=24$, the full eigenvalue spectrum of the reduced density matrix $\\rho_{RL}^A$ (using ED for $L\\lesssim 12$ and a much larger bond dimension on longer chains) and compare it with the singular-value ordering used in the MPS truncation: if the largest discarded singular value corresponds to an eigenvalue whose real or imaginary part is comparable to the smallest level retained in the fits, the MPS ground state is missing essential entanglement. Alternatively, if the fitted minimum of the cost function $J(v,g_{\\epsilon'})$ stops moving toward $\\lambda_c = 0.0788 + 0.0603i$ as $L$ grows beyond 24 and instead reverses its $1/L^2$ drift, the refined critical point is an artifact of finite-size extrapolation.","tokens_in":13914,"feed_emoji":"🌀","tokens_out":13249,"duration_ms":138557,"temperature":0.7,"pith_summary":"The quantum 5-state Potts model's weakly first-order transition is believed to be controlled by a pair of complex conformal field theory (CCFT) fixed points that cannot be reached by tuning real couplings alone. To realize them on the lattice, one adds a non-Hermitian term with a complex coefficient $\\lambda$, which destroys the variational principle that normally guarantees tensor-network accuracy. This paper argues that in the 'walking' regime, where RG flow slows and the non-Hermitian deformation remains small, ordinary DMRG and quasiparticle-ansatz methods still converge to the correct ground and excited states. With systems up to $L=28$, the authors refine the critical point to $\\lambda_c = 0.0788 + 0.0603i$, observe the theoretically predicted spiral flow of the running coupling $g_{\\epsilon'}$, and recover an entanglement spectrum that matches the free-free boundary CCFT. If correct, this turns complex fixed points from an analytic curiosity into a numerically accessible window on weakly first-order transitions.","feed_headline":"Tensor networks trace the spiral RG flow of the 5-state Potts model","feed_subtitle":"Tensor-network runs at L = 28 refine the critical coupling and match the boundary CCFT entanglement spectrum.","key_machinery":"The workhorse is the running coupling $g_{\\epsilon'}$ of the leading perturbing operator in the CCFT description. It is recovered from finite-size energy data by a cost function built from conformal perturbation theory, $\\delta E_{\\phi} = 2\\pi g_{\\epsilon'} C_{\\phi\\phi\\epsilon'}$, with the descendant shift fixed by the scaling dimension $\\Delta_{\\epsilon'}$. Plotting $g_{\\epsilon'}$ as the chain grows shows the spiral RG flow. For entanglement, the central object is the non-Hermitian density matrix $\\rho_{RL} = |\\psi_L\\rangle\\langle\\psi_R|/\\langle\\psi_R|\\psi_L\\rangle$ and the associated reduced density matrix $\\rho^A_{RL}$ built from the singular value decompositions of the right and left ground states; the paper proposes that its entanglement Hamiltonian satisfies $\\rho_{RL} = e^{-2\\pi K_C}$. The MPS ansatz's singular values serve as the truncation criterion, justified here by the smallness of the non-Hermitian deformation.","core_discovery":"On the paper's terms, the discovery is that the small non-Hermiticity of the deformed 5-state Potts Hamiltonian makes the MPS singular-value truncation nearly optimal even though the variational principle is invalid, so the unadapted DMRG and QPA reach algorithmic tolerance ($10^{-6}$) for the ground state at $\\lambda_c = 0.0788 + 0.0603i$. With the refined coupling, subleading finite-size scaling gives central charge $c = 1.1375 - 0.0211i$ and conformal dimensions consistent with the theoretical CCFT data; the cost-function fit yields the logarithmic-spiral RG flow of $g_{\\epsilon'}$ around the two conjugate fixed points; and the entanglement Hamiltonian extracted from $\\rho_{RL}^A$ reproduces the free-free boundary CCFT operator content (towers with degeneracies 1, 4, and 11). The failure modes are acknowledged: imaginary parts converge more slowly, the density matrix is not positive semi-definite, and the MPS ordering property is only approximate.","pith_inferences":["A practical criterion suggested by the paper: unadapted MPS methods should work when the imaginary part of the complex fixed-point data is small compared with the real part; more strongly non-Hermitian CCFTs (such as O(N>2) sigma models) should require biorthonormal or fidelity-based truncation.","The same entanglement-spectrum diagnostic could be applied to other candidate weakly first-order transitions, such as the Néel–valence-bond-solid transition, where characteristic boundary degeneracies could reveal a complex fixed point even without an explicit non-Hermitian lattice deformation.","The anomalous ~$1/L^{1.05}$ scaling of Im $\\Delta_\\epsilon$ is a candidate systematic artifact: checking whether it survives in a biorthonormal DMRG calculation at fixed $\\lambda_c$ would separate a true OPE effect from the MPS truncation bias in imaginary entanglement."],"forward_implications":["Standard DMRG and QPA can be used for other slightly non-Hermitian CCFT lattice models without biorthonormal adaptations, with errors controlled by algorithmic tolerance.","The refined $\\lambda_c = 0.0788 + 0.0603i$ improves the predicted conformal data beyond exact diagonalization at the previous estimate, because subleading $1/L^2$ finite-size effects are included.","The observed spiral flow of $g_{\\epsilon'}$ in a 1+1D lattice system confirms the theoretical picture of complex fixed points as the origin of walking and quasi-critical behavior.","The entanglement spectrum matching the free-free boundary CCFT (degeneracies 1, 4, 11) supports the proposed relation $\\rho_{RL} = e^{-2\\pi K_C}$ for complex fixed points.","Weakly first-order transitions with hidden complex fixed points become directly addressable by tensor networks, not only through analytic continuation of theory space."],"supporting_citations":[{"why":"Supplies the theoretical CCFT data and the one-loop prediction that the RG flow of the running coupling is a logarithmic spiral.","marker":"[9]"},{"why":"Provides the non-Hermitian lattice deformation H1 and the earlier exact-diagonalization estimate of the critical coupling that this work refines.","marker":"[11]"},{"why":"Gives the free-free boundary CCFT operator content and degeneracies used to match the entanglement spectrum.","marker":"[31]"},{"why":"Defines the non-Hermitian density matrix and complex entanglement entropy framework used for the CCFT.","marker":"[22]"},{"why":"Provides the construction of the reduced density matrix from left and right MPS via rotation matrices.","marker":"[35]"},{"why":"Provides the finite-size scaling relation between energies, central charge, scaling dimensions, and speed of light used for conformal data extraction.","marker":"[26]"},{"why":"Supplies the conformal perturbation theory energy shifts entering the cost function for the coupling g_epsilon'.","marker":"[28]"}],"fun_headline_variants":["Spiral RG flow of non-Hermitian Potts model via tensor networks","Non-Hermitian Potts spiral flow and entanglement spectrum","Tensor nets capture spiral RG and entanglement in non-Hermitian Potts","Spiral renormalization flow in non-Hermitian Potts via tensor networks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the MPS truncation based on the singular values of the right ground-state eigenvector preserves the ordering of the true non-Hermitian entanglement spectrum well enough to neglect the discarded states, even though the variational principle no longer guarantees this and the paper flags in Supplementary Section E that the disconnect between singular values and eigenvalues leaves the neglected entanglement unknown.","fun_headline_variants_meta":{"raw":{"variants":["Spiral RG flow of non-Hermitian Potts model via tensor networks","Non-Hermitian Potts spiral flow and entanglement spectrum","Tensor nets capture spiral RG and entanglement in non-Hermitian Potts","Spiral renormalization flow in non-Hermitian Potts via tensor networks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000507,"raw_usage":{"total_tokens":2458,"prompt_tokens":920,"completion_tokens":1538,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":536,"completion_tokens_details":{"reasoning_tokens":1459}},"tokens_in":536,"tokens_out":1538,"duration_ms":12050,"temperature":1.0,"reasoning_tokens":1459,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:49:24.891675+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test would be to compute, at $\\lambda_c$ and $L=24$, the full eigenvalue spectrum of the reduced density matrix $\\rho_{RL}^A$ (using ED for $L\\lesssim 12$ and a much larger bond dimension on longer chains) and compare it with the singular-value ordering used in the MPS truncation: if the largest discarded singular value corresponds to an eigenvalue whose real or imaginary part is comparable to the smallest level retained in the fits, the MPS ground state is missing essential entanglement. Alternatively, if the fitted minimum of the cost function $J(v,g_{\\epsilon'})$ stops moving toward $\\lambda_c = 0.0788 + 0.0603i$ as $L$ grows beyond 24 and instead reverses its $1/L^2$ drift, the refined critical point is an artifact of finite-size extrapolation.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the construction of the reduced density matrix from left and right MPS via rotation matrices."},{"cited_title":"Affleck, Phys","cited_arxiv_id":null,"evidence_quote":"Provides the finite-size scaling relation between energies, central charge, scaling dimensions, and speed of light used for conformal data extraction."}],"review_version":1}