{"id":"af23680c-77b0-4610-83de-7dc379561fb9","arxiv_id":"2412.03888","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A tree-level improved lattice formula combined with an over-improved gradient flow reproduces integer winding numbers for T^3 to SU(2) maps on coarse lattices.","lead":"The authors propose a numerical recipe for computing the winding number of a map from a 3D torus to unitary matrices using only lattice points. They show that a corrected lattice formula plus an over-improved gradient flow recovers integer winding numbers even on coarse grids.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The method's generality is the load-bearing unverified premise: negative-η flow stabilization is shown only on the one-parameter family (2.4), and the paper's own footnote 9 concedes escape from this family is possible.","rationale":"I found no internal inconsistency or numerical error in the reported computations; the η=1 discretization improves the L=10 values dramatically (Table 2), and the figures support the stated observations. The only load-bearing weakness is the extrapolation from the single test family to generic coarse or noisy configurations. This is the same assumption the Reader identified. The paper is honest about the limitation, explicitly in the Conclusion and in footnote 9, which is why the concern does not amount to rejection; it requires conditional acceptance pending a broader test. I therefore recommend no change to the Reader's CONDITIONAL verdict.","tokens_in":8883,"tokens_out":7553,"duration_ms":74619,"concrete_test":"Apply the same recipe (η=-20 gradient flow, η=1 measurement, L=10, ε∈[0,1] sitewise noise, flow times t=2 and t=10) to a second independent family of maps with analytically known winding numbers, e.g. the SU(2) maps used in Refs. [5,6] or a product of (2.4)-type maps with known additive winding numbers. Also include initial configurations near sector boundaries of (2.4), such as m=0.5, 1.5, 2.5, and 3.5, to test whether the attraction basins are robust. If the flowed W_lat reproduces the known integer in every sector without re-tuning η, the generality concern is settled; if any sector fails or requires a different η, the claim must be revised to state validation only for (2.4)-like maps.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central assertion is that the tree-level discretization (η=1) plus over-improved gradient flow (η<0) yields accurate W3 on coarse lattices. The load-bearing premise is that the negative-η lattice action (3.7) has local minima with correct attracting basins in every topological sector, so generic noisy configurations flow to the right integer. What is demonstrated is narrower: Fig. 5 evaluates S_lat only along the one-parameter family (2.4), showing local minima for η=-20 in the sectors probed. That is a stationary-point statement on a 1D slice of a high-dimensional configuration space; it says nothing about the size of the basin of attraction, behavior near sector boundaries, or behavior for maps unlike (2.4). The noisy-flow tests (Figs. 1 and 2) use only m=-1, 1, and 3, all deep inside sectors and all sharing the special structure of (2.4). Footnote 9 explicitly concedes that a configuration could slip away from the family under flow, and the Conclusion states that real applications require exploratory tuning of η. Since W_lat is not integer-valued on the lattice, flow can cross between sectors; stabilization is an empirical property of the chosen η and flow time, not a theorem. The advertised 'simple and versatile' method therefore rests on an unverified representative-ness assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a numerical method for approximating the winding number of a map from T^3 to U(N) when the map is only known on a cubic lattice. The method combines a tree-level improved lattice discretization of the W3 functional, Eqs. (2.2)-(2.3), with gradient-flow smearing generated by an over-improved lattice action, Eqs. (3.4)-(3.9). Using a one-parameter family of SU(2) maps with known winding numbers, the authors show that the eta=1 discretization converges quickly (Tables 1-2), that gradient flow with negative eta stabilizes W_lat against random perturbations (Figs. 1-2), and that the lattice action restricted to the family develops local minima in the corresponding sectors for negative eta (Fig. 5). The paper also explains the stabilization mechanism in terms of discretization errors in the flow action and notes that the method avoids diagonalization or interpolation of g(x).","tokens_in":9143,"tokens_out":6304,"duration_ms":57835,"significance":"If the results generalize beyond the tested family, the method is a useful and computationally light tool for lattice computations of Chern-Simons-type invariants. The paper is transparent in an exemplary way: the benchmark uses analytically known winding numbers, the code and numerical data are publicly available on GitHub, and the authors explicitly disclose that the choice of eta and flow time is exploratory rather than derived from first principles. The tree-level improvement is shown to reduce discretization errors by large factors on the tested family, and the over-improved flow provides clear numerical stabilization in the shown cases, which are genuine assets of the manuscript.","major_comments":[{"comment":"The load-bearing claim is that over-improved gradient flow stabilizes the lattice winding number for generic coarse or noisy configurations. The evidence, however, is restricted to the one-parameter family (2.4): Fig. 5 evaluates S_lat only along that slice, so it establishes local minima on a one-dimensional subspace of a high-dimensional configuration space, not the size of the basins of attraction. Footnote 9 explicitly concedes that a configuration could slip away from the family under flow. The noisy tests in Figs. 1 and 2 are also perturbations of the same family and cover only m=-1, 1, and 3. Because W_lat in Eq. (2.3) is not integer-valued on the lattice, the flow can cross between sectors, and stabilization is an empirical property of the chosen eta and t rather than a theorem. I would ask for at least one additional family of maps with known winding number, or a quantitative basin-size diagnostic, before the method is advertised as general; alternatively, the claims should be narrowed explicitly to the tested class.","section":"Sec. 3, Figs. 1-2 and 5; footnote 9"},{"comment":"The method requires two tuning parameters, the over-improvement coefficient eta and the flow time t, and no concrete criterion for choosing them is provided. Figures 1 and 2 are shown for eta=-10 and eta=-20 and for t up to 40 or 80, with t=2 used in Fig. 6, but nowhere is there a rule such as a plateau-time selection or an automatic stopping condition. The conclusion states that real applications require exploratory tuning of eta. Since the central practical promise is a 'simple and versatile' method, a concrete protocol for choosing (eta, t), or at least a sharper statement of how sensitive the result is to these choices, is needed for the method to be reproducible in new settings.","section":"Secs. 3-4"}],"minor_comments":[{"comment":"The displayed equation d_t g(t,x) = -sum_x (d^a_x S_lat)(d^a_x S_lat) is not what is meant; the context shows that the derivative of the action should be d_t S_lat = -sum_x (d^a_x S_lat)^2 <= 0. Please correct this typo, since the monotonic decrease of S_lat is used in the subsequent explanation.","section":"Eq. (3.6)"},{"comment":"The abstract and introduction state that the method is 'simple and versatile' and can be generalized to higher-dimensional tori, but the numerical support is only for d=3 and for the SU(2) family (2.4). If no further tests are added, please qualify these statements, for example by saying 'for the class of maps studied here'.","section":"Abstract and Sec. 1"},{"comment":"No comparison with the existing lattice winding-number algorithms of Refs. [5-8] is made. A brief numerical or conceptual comparison would help the reader assess the claimed advantage of the present method.","section":"Sec. 1, Refs. [5-8]"},{"comment":"The 15 trajectories per m value are plotted as separate lines, but no legend connects line style to m. A legend or distinctive line styles would improve readability.","section":"Figs. 1-2"},{"comment":"The piecewise winding-number formula for the family (2.4) is stated without derivation or citation; please add a reference or a short justification.","section":"Eq. (2.6)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is honest and well written, and the numerical evidence within the tested family is solid. My hesitation concerns external validity: the general claim of the method rests on a single one-parameter family, and the authors themselves accept that exploratory tuning is needed. This is the right kind of contribution for a rapid-communication venue if the claims are either strengthened with additional tests or weakened appropriately. I see no grounds for rejection, but the requested additions or caveats are necessary before the paper can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: Morikawa and Suzuki propose a two-part recipe for computing winding numbers of maps T^3 -> U(N) on coarse lattices: a tree-level improved discretization of W3 and a gradient flow driven by an over-improved lattice action with negative eta. The combination is new, and the demonstration on the one-parameter family (2.4) is convincing. The tree-level improvement alone gives dramatic error reduction (Tables 1–2), and the over-improved flow stabilizes the lattice winding number for eta = −20 in the noisy tests (Figs. 1–2). The code and data are public, which is good practice.\n\nThe honest soft spot is the one the authors themselves flag in footnote 9 and the conclusion: only one parametric family is tested. The explanation of stabilization via Fig. 5, showing S_lat(m) developing local minima for negative eta, is a 1D slice argument, not a proof of attracting basins in the full configuration space. So the claim that the method is “simple and versatile” rests on an unverified representativeness assumption. Also, no comparison with Refs. [5–8] methods, and eta and flow time are tuned. These are real limitations but proportionate: the paper doesn’t oversell; it says exploratory tuning is needed.\n\nMy take: this is a solid methods paper, worth refereeing and likely useful to practitioners. The central claim is supported for the tested family; the generality claim should be tempered, but that’s a revision-level concern, not a rejection-level one. I’d cite it if I worked on lattice topology numerics.","headline":"A clean, honest numerical methods paper: tree-level improved winding number plus over-improved gradient flow works on the tested one-parameter family, but generality is asserted more than demonstrated.","tokens_in":9620,"tokens_out":1037,"would_cite":true,"duration_ms":8704,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81T25","81T13"],"pacs":["11.15.Ha"],"model":"deepseek-v4-flash","headline":"A tree-level improved discretization plus an over-improved gradient flow reproduces integer winding numbers for maps from a 3D torus to SU(2) even on coarse lattices.","keywords":["winding number","3D torus","lattice discretization","gradient flow","over-improved action","SU(2) maps","topological charge","Chern-Simons theory"],"falsifier":"Take a different family of maps from $T^3$ to $SU(2)$ with known winding numbers, add noise comparable to the paper's, and check whether the over-improved flow with $\\eta=-20$ on an $L=10$ lattice produces a plateau of $W_3^{\\rm lat}$ at the correct integer over flow times $t\\simeq2$–$40$; if the plateau drifts, fails to reach an integer, or jumps between sectors, the claimed stabilization is not generic. A cheaper check is to compute $S_{\\rm lat}$ along a two-parameter family and see whether sectors with nonzero winding number lack a local minimum.","tokens_in":8659,"feed_emoji":"🌀","tokens_out":6768,"duration_ms":60283,"temperature":0.7,"pith_summary":"This paper proposes a numerical method for computing the winding number of a map from the 3D torus to the unitary group when the map is known only on lattice points. The authors claim that a tree-level improved discretization of the standard continuum formula, combined with a gradient flow generated by an over-improved lattice action, gives accurate integer winding numbers even on coarse lattices and in the presence of random noise. They demonstrate this on a one-parameter family of maps from $T^{3}$ to SU(2) with known winding numbers 0, 1, and -2, on lattices as small as L=10. If correct, the method offers a practical way to extract topological data from discrete configurations without diagonalizing or interpolating group elements, and it extends to higher-dimensional tori.","feed_headline":"Over-improved flow pins winding numbers on coarse lattices","feed_subtitle":"Tree-level improvement plus negative-η gradient flow reproduces W = 0, 1, −2 for SU(2) maps on L = 10 lattices.","key_machinery":"The central object is the combination $H(x,\\mu)$, a 'tree-level improved' finite-difference replacement for $g^{-1}\\partial_\\mu g$ that includes a tunable parameter $\\eta$; the improved discretization of the winding number is $W_3^{\\rm lat} = \\frac{1}{4^3\\,24\\pi^2}\\sum_{x,\\mu,\\nu,\\rho}\\epsilon_{\\mu\\nu\\rho}\\mathrm{tr}[H(x,\\mu)H(x,\\nu)H(x,\\rho)]$. The gradient flow is defined by $\\partial_t g(t,x) = -g(t,x)\\,\\partial^a_x S_{\\rm lat}T^a$ with $S_{\\rm lat}=-\\frac{1}{16}\\sum_{x,\\mu}\\mathrm{tr}H(x,\\mu)^2$, and is integrated by a Runge-Kutta scheme. The load-bearing mechanism is the dependence of $S_{\\rm lat}$ on $\\eta$: for negative, over-improved $\\eta$ the action as a function of the parameter $m$ acquires local minima inside each topological sector (Fig. 5), so the flow settles into the correct sector instead of sliding to the trivial map, while for $\\eta=0,1$ the action has no such minima and the flow erases the topological information.","core_discovery":"The paper claims that the continuum winding number $W_3=\\frac{1}{24\\pi^2}\\int_{T^3}\\mathrm{tr}(g^{-1}dg)^3$ can be approximated on a cubic lattice by replacing $g^{-1}\\partial_\\mu g$ with an improved difference operator $H(x,\\mu)$ (Eq. (2.2)) containing a parameter $\\eta$, and summing $\\epsilon_{\\mu\\nu\\rho}\\mathrm{tr}(HHH)$ over the lattice (Eq. (2.3)). Setting $\\eta=1$ removes the leading lattice-discretization error and already reproduces $W_3=0,1,-2$ accurately for $L\\gtrsim20$. For coarse or noisy configurations, flowing $g(x)$ by the gradient equation $\\partial_t g = -g\\,\\partial^a_x S_{\\rm lat}T^a$ with the over-improved action $S_{\\rm lat}=-\\frac{1}{16}\\sum_{x,\\mu}\\mathrm{tr}H(x,\\mu)^2$ (using negative $\\eta$ such as $-20$) stabilizes the lattice winding number around the correct integer; the paper shows that for negative $\\eta$ the lattice action develops local minima in each topological sector, pinning the flow. The method requires neither diagonalization of $g(x)$ nor interpolation of the map.","pith_inferences":["The mechanism suggests a practical tuning rule: choose $\\eta$ negative enough that the lattice action has local minima in every sector of interest for the given lattice size; the required $\\eta$ likely depends on $L$ and on the roughness of the configurations.","Because the stabilization comes from discretization error, the method's over-improved flow is complementary to standard improvement: on fine lattices the improved discretization suffices, while on coarse lattices the flow provides the stability, potentially making the pair useful for Monte Carlo samples with noisy topological charge.","One testable extension, not pursued in the paper, is to apply the method to configurations with multiple spatially separated topological structures, where pinning to a single integer is harder and the action landscape would need local minima in each sector separately."],"forward_implications":["With tree-level improvement ($\\eta=1$), the discretized winding number alone reaches essentially exact integers already at $L\\simeq20$ for the tested family.","For coarse lattices ($L=10$) with added random noise, an over-improved flow ($\\eta=-10$ or $-20$) holds the lattice winding number near the correct integer over a range of flow times; for example, $W_3^{\\rm lat}=-1.96797(4)$ at $t=2$ for $m=3$.","Standard or tree-level flow ($\\eta=0,1$) drives configurations toward the trivial sector, so the over-improved choice is essential for stabilization.","The same construction works for $U(N)$ and extends trivially to higher-dimensional tori, with light computational cost."],"supporting_citations":[{"why":"Supplies the Runge-Kutta integration scheme used to solve the lattice gradient flow equation.","marker":"[11]"},{"why":"Provides the over-improved-action idea that stabilizes topological sectors under gradient flow.","marker":"[12]"},{"why":"Underlies the action-as-a-function-of-moduli argument that explains sector stabilization.","marker":"[13]"},{"why":"Defines the one-parameter family of maps from T^3 to SU(2) whose known winding numbers serve as the test bed.","marker":"[5]"},{"why":"Offers the effective method for the first Chern number on a discrete lattice that motivates the improved local formula.","marker":"[9]"},{"why":"Supplies the numerical lattice gauge field implementation used for the flow integration.","marker":"[18]"}],"fun_headline_variants":["Negative-eta flow pins winding numbers on coarse lattices","Improved discretization yields winding numbers without diagonalization","Over-improved action stabilizes integer winding on L=10 grids","Gradient flow with over-improved action reproduces homotopy on 3D tori"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method's reliability rests on the untested premise that the single one-parameter family tested is representative: for sufficiently negative $\\eta$, the lattice action develops local minima in every topological sector for generic coarse or noisy configurations, not just for this family.","fun_headline_variants_meta":{"raw":{"variants":["Negative-eta flow pins winding numbers on coarse lattices","Improved discretization yields winding numbers without diagonalization","Over-improved action stabilizes integer winding on L=10 grids","Gradient flow with over-improved action reproduces homotopy on 3D tori"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000578,"raw_usage":{"total_tokens":2728,"prompt_tokens":951,"completion_tokens":1777,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":567,"completion_tokens_details":{"reasoning_tokens":1702}},"tokens_in":567,"tokens_out":1777,"duration_ms":15410,"temperature":1.0,"reasoning_tokens":1702,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T21:57:58.642465+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a different family of maps from $T^3$ to $SU(2)$ with known winding numbers, add noise comparable to the paper's, and check whether the over-improved flow with $\\eta=-20$ on an $L=10$ lattice produces a plateau of $W_3^{\\rm lat}$ at the correct integer over flow times $t\\simeq2$–$40$; if the plateau drifts, fails to reach an integer, or jumps between sectors, the claimed stabilization is not generic. A cheaper check is to compute $S_{\\rm lat}$ along a two-parameter family and see whether sectors with nonzero winding number lack a local minimum.","supporting_citations":[],"review_version":1}