{"id":"6dab9e1c-f88a-46c0-bef0-a2923ddcd20d","arxiv_id":"2411.10395","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A new localised basis for lattice QCD distillation, generated by a unitary-flow orthogonalisation, enables sparse importance-sampled contractions of baryon correlation functions.","lead":"Lattice QCD calculations of hadron properties need expensive tensor contractions over quark distillation modes, with costs growing rapidly as more quark fields are added. This paper builds a localised basis for the distillation space and uses it in an importance-sampled contraction scheme, with first numerical tests on nucleon, delta, and N-pi correlators.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Locality of the flow-orthogonalised basis is shown on a single 4^3-source, 32^3-volume ensemble; the central efficiency claim collapses if the overlap matrix A0†A0 is not near-identity at other volumes or lattice spacings, but no test or bound is provided.","rationale":"The paper's mathematical core is sound: the Hansen-Hurwitz estimator is unbiased, and the numerical tests in Sec. IV confirm agreement with exact distillation. The novel contribution is the localised basis, and every subsequent efficiency statement depends on it. I focus on locality rather than the perambulator-flatness assumption because even a non-flat perambulator leaves an unbiased estimator and could in principle be accommodated by modifying the sampling distribution, whereas a non-local basis removes the sparsity that is the entire source of the claimed speed-up. The analysis shows the flow's output is the polar factor of A0, so the question is quantitative: how close is A0 to unitary? The paper's empirical demonstration on one ensemble is real evidence but does not establish the scaling claim for larger volumes or finer lattices, especially since the coarse-grid density in physical units is the control parameter. The proposed test directly measures the locality of W and the sparsity of elementals, which are the quantities that enter the cost model. If the test passes, the CONDITIONAL verdict can be upgraded; if it fails, the method needs additional machinery (e.g., overlapping grids or a different orthogonalisation) before the central claim is supported. This is consistent with the paper's own caveat that more work is needed to build a robust algorithm (Sec. VI).","tokens_in":18490,"tokens_out":13097,"duration_ms":137242,"concrete_test":"Compute A0 = V†Q and the singular values of A0 (or the norm of A0†A0 - I) for at least two additional settings: (i) a finer lattice spacing at the same physical volume and same physical source separation, and (ii) a larger volume with 6^3 or 8^3 coarse-grid sources at the same physical spacing. For each, measure the exponential width of the ensemble-averaged ρ(i)(x) defined in Eq. (11) and the fraction of the baryon elemental L1 norm contained in the same-site blocks. If the width grows by more than about 20% or the same-site fraction drops below about 50%, the locality assumption underlying the cost model fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that W = V lim A(s) is local because the flow in Eq. (8) preserves the equivariance of Eq. (9). But the flow only changes the Hermitian factor: writing A = U H, Eq. (8) gives dH/ds = H(1 - H^2) and dU/ds = 0, so the final U is exactly the polar factor of A0 = V†Q. Locality of W = VU is therefore equivalent to A0 being close to unitary, i.e. A0†A0 ≈ I. That holds only when the coarse source grid is well separated compared with the smearing width of the low Laplacian modes. Fig. 1 verifies this for one ensemble (32^3 time-slices, 4^3 sources); Sec. II provides no test at other volumes, lattice spacings, or grid densities, and no analytic bound on A0†A0 - I. If the overlap matrix is not near-identity, the polar factor is a nontrivial unitary mixing that delocalises W, the elementals in Figs. 2-3 become dense, and the occupancy/cost estimates of Sec. V (Eqs. 50-52, Figs. 9-10) are invalid because |X(A)| and the marginal sets no longer saturate at small ns. The unbiasedness of the Hansen-Hurwitz estimator (Sec. III) is unaffected, but the claimed reduction from O(nD^{d+1}) to a sparse cost is not.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a new orthonormal basis for distillation space in lattice QCD, constructed by applying the distillation projection to a coarse grid of point sources and then orthogonalising via a flow equation whose fixed points are unitary matrices. The authors argue that this basis preserves the locality of the point sources, making hadronic operators (elementals) sparse in distillation-space index structure. They introduce a Hansen-Hurwitz importance-sampling scheme that exploits this sparsity to estimate two-point correlation functions, with algorithms for sparse tensor contractions, and they test the method on nucleon, Delta, and N-pi-Delta correlators on a single lattice ensemble, comparing stochastic estimates with exact distillation contractions. The paper also analyses the expected cost scaling of the method and extrapolates to a hypothetical tetraquark calculation.","tokens_in":18812,"tokens_out":6060,"duration_ms":58749,"significance":"If the locality of the new basis holds generally, the method has the potential to reduce the computational cost of multi-quark correlation functions (baryons, tetraquarks), which currently limit the reach of distillation-based spectroscopy. The mathematical foundation is sound: the Hansen-Hurwitz estimator is unbiased by construction, the algorithms are clearly specified, and the numerical tests show agreement with exact distillation contractions. The paper is also commendably honest about its limitations and distinguishes measured results from extrapolations. However, the practical efficiency gain is not directly measured, and the locality argument rests on a single ensemble, so the significance of the work is conditional on broader validation.","major_comments":[{"comment":"The claim that the flow preserves locality is demonstrated only on a single 32^3 lattice with a 4^3 source grid. As the stress-test note correctly observes, writing A = U H shows that the flow changes only the Hermitian factor (dU/ds = 0), so the final basis W = V U is local only to the extent that A0^dagger A0 is close to the identity. No bound on ||A0^dagger A0 - I|| and no tests at other volumes, lattice spacings, or grid densities are provided. Since the sparsity of the elementals in Figs. 2-3 and the occupancy/cost estimates in Sec. V (Eqs. 50-52, Figs. 9-10) rely on this locality, this is a load-bearing assumption that requires broader evidence.","section":"II, Eqs. (8)-(10), Fig. 1"},{"comment":"The paper's stated goal is to optimise the computation, but no wall-clock timings or total-cost comparison are reported. The cost coefficients zeta_k count only the sparse-matrix multiply operations in the contraction; the overhead of drawing samples, building the index sets, and generating the perambulators is not included. The predicted one-order-of-magnitude speedup for tetraquarks (Sec. V B, Fig. 10) relies on the untested scaling assumption M ≈ sqrt(ns/nD) and on extrapolating the nucleon marginal probabilities to tetraquark operators. To substantiate the efficiency claim, the authors should report either timings or a complete cost model that accounts for all algorithmic components.","section":"V, Sec. IV"},{"comment":"The factorisation of the sampling probabilities assumes that the perambulator entries depend weakly on the distillation-space indices, stated as \"dense structure is observed approximately for large time-separations\" but with no supporting data shown. This assumption is load-bearing because the whole scheme separates source and sink sampling. The N-pi-Delta results in Fig. 7 show substantial sampling noise even at early time separations, suggesting that this assumption can fail in practice. A quantitative test of the assumption (for example, the distribution of |tau_ab| across index pairs, or a comparison of the factorised probabilities with the optimal ones of Eq. (25)) is needed to justify the general applicability of the method.","section":"III A, Eq. (26)"}],"minor_comments":[{"comment":"The Gaussian fit in the right panel is presented without any stated purpose; since the fit parameters are not used in the construction of the basis, it would be helpful to state explicitly that the fit is illustrative only.","section":"II, Fig. 1"},{"comment":"The middle panels of Fig. 6 show ratios of standard errors, but the vertical axis labels are missing from the figure as displayed; adding explicit axis labels or a legend would improve readability.","section":"IV, Fig. 6"},{"comment":"The expected-occupancy formula assumes sampling with replacement; this is stated in Algorithm 1, but it would be helpful to repeat the assumption in Sec. V when defining the occupancy expressions.","section":"V, Eq. (50)"},{"comment":"The symbol Phi is used both for the meson elemental matrices (Eq. 15) and for the baryon operator projections (Eq. 17); the text distinguishes them, but a casual reader may be confused. Consider using separate symbols for the two types of objects.","section":"II, Eq. (15), IV"}],"recommendation":"major_revision","confidential_remarks":"The paper is a serious methods contribution from an experienced collaboration, and the unbiasedness of the estimator is not in question. The main risk is that the efficiency claim rests on the locality of the flow-orthogonalised basis, which is demonstrated on one ensemble only, and on cost estimates that do not include all relevant components. I would encourage the editor to request either additional ensemble tests (varying volume/lattice spacing/grid density) or at least a quantitative diagnostic of the near-unitarity of A0, together with a more complete cost model. The tetraquark speedup estimate should be clearly labelled as speculative until the sample-size scaling M is tested."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a solid methodological paper from the Hadron Spectrum group. The genuinely new piece is a locality-preserving orthonormal basis for distillation space, built by flowing a non-unitary overlap matrix to its polar factor, plus a Hansen-Hurwitz importance sampler that exploits the resulting sparsity of hadron elementals. The math is clean, and the numerical tests on nucleon, delta, and N-pi-delta correlators confirm unbiasedness against exact distillation, with sampling noise sub-leading for the nucleon. The authors are appropriately cautious: they explicitly say they make no precise performance claims, and they flag the main unvalidated assumptions.\n\nThe strongest part is the flow orthogonalization. It is simple, equivariant, and demonstrably preserves locality on the one ensemble tested (32^3 volume, 4^3 source grid). The sparse contraction algorithms are careful, and the occupancy/cost analysis gives a useful way to think about scaling. Credit where due: the paper is honest, well-written, and the estimator is unbiased for any choice of probabilities.\n\nNow the soft spots. First, the locality claim rests on a single ensemble. The stress-test note is right: the flow only changes the Hermitian factor, so the final basis is local only if A0^\\dagger A0 is close to the identity. That may hold more generally, but there is no test across volumes, lattice spacings, or grid densities, and no analytic bound. If the basis delocalises, the sparsity vanishes and the cost savings evaporate, while the estimator stays unbiased but pointless. This is the main risk, and it currently looks like an empirical observation rather than a property with a known domain of validity.\n\nSecond, there are no wall-clock timings and no direct comparison with existing stochastic LapH plus dilution. The tetraquark speedup is a crude extrapolation from nucleon marginal probabilities; the authors label it as such, but it is still a guess. Third, the factorised sampling probabilities assume the perambulator magnitude is roughly index-independent; the paper says this is \"observed approximately\" but provides no supporting data. That is a minor-to-moderate gap, not a fatal one.\n\nNone of these issues are fatal. This is a promising first report, not a finished production method. The right reader is a lattice QCD practitioner working on baryon or tetraquark spectroscopy; they will find the basis construction and the sparse contraction formalism genuinely useful. I would send it to a serious referee. A good referee should push for (a) locality tests on at least one more volume or lattice spacing, or a quantitative criterion for when A0^\\dagger A0 is near identity, and (b) at least one honest timing comparison against standard stochastic LapH with dilution. The paper deserves revision, not rejection.","headline":"A promising locality-based acceleration for distillation contractions, with clean estimator math and honest tests, but the central efficiency claim is conditional on a single ensemble and comes without wall-clock timings.","tokens_in":19354,"tokens_out":2862,"would_cite":true,"duration_ms":28541,"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":"A flow-orthogonalised distillation basis makes hadron correlation functions effectively sparse, and unbiased stochastic sampling of the sparse entries reproduces exact contractions; in the nucleon case the added noise is sub-leading to…","keywords":["lattice QCD","distillation","quark smearing","hadron correlation functions","Hansen-Hurwitz estimator","importance sampling","baryon operators","localised basis"],"falsifier":"Compute the flow-orthogonalised basis on a second ensemble with a different lattice spacing or volume and measure the fraction of the total baryon elemental weight contained in same-site blocks; if that fraction drops toward the dense-case value, the cost advantage disappears. Separately, compare the variance of the estimator using the paper's factorised probabilities with the variance using probabilities that include perambulator magnitudes; a large gap would show the factorised choice is the limiting factor.","tokens_in":18265,"feed_emoji":"⚛️","tokens_out":7477,"duration_ms":73979,"temperature":0.7,"pith_summary":"Distillation is the standard lattice-QCD way to smear quark fields into smooth modes, but its cost grows steeply for operators with several quarks. The paper tries to change that by replacing the usual Laplace-eigenvector basis of distillation space with an orthonormal basis whose vectors sit near sites of a coarse spatial grid. The orthonormalisation is done by a gradient flow whose fixed points are unitary matrices and which, unlike sequential subtraction-based orthogonalisation, leaves the locality intact. In this basis, hadron operators become sparse tensors, and a Hansen-Hurwitz importance-sampling estimator draws only the large entries of the contraction. The first tests show the estimator is unbiased and, for the nucleon, adds less noise than the statistical gauge noise, so the bottleneck cost of baryon and tetraquark spectroscopy could drop substantially.","feed_headline":"Localised distillation basis makes baryon sums sparse","feed_subtitle":"Flow-orthogonalised smearing plus importance sampling reproduces exact nucleon correlators with little added noise.","key_machinery":"Distillation projects quark fields on a time-slice onto the space spanned by low-lying eigenmodes of the gauge-covariant Laplace operator. The new machinery is the localised orthonormal basis: point sources on a coarse grid are smeared by the distillation projector, and the resulting spanning set $A_0$ is driven to a unitary fixed point by the gradient flow $dA/ds = (I - AA^\\dagger)A$, preserving the locality that sequential orthogonalisation would destroy. In this basis the distillation-space matrices called elementals, e.g. baryon tensors $\\phi^B_{ijk}$, have large entries only when the anchor sites coincide, making them sparse. The stochastic machinery is the Hansen-Hurwitz estimator, which draws index tuples with probability proportional to their magnitudes; to keep samples reusable the paper factorises probabilities into independent source and sink factors and averages over time slices and gauge configurations. Sparse contraction algorithms then build temporaries using only sampled index sets, scaling like $O(|X^{(A,r-1)}|\\,|X^{(B,d-r)}|)$ rather than the full $O(n_D^{d+1})$.","core_discovery":"The paper's central claim is that a basis change alone, from the delocalised Laplace eigenvectors to the flow-generated local basis $W = VU$, can make hadron correlation-function contractions sparse enough for efficient stochastic evaluation. The claim is backed by the numerical observation that elementals such as the baryon tensor $\\phi^B_{ijk}$ are large only when the anchor points of the three basis vectors coincide, and by the construction of an unbiased Hansen-Hurwitz estimator that draws index tuples with factorised, time-averaged probabilities. Tested on nucleon, $\\Delta$, and $N\\pi\\to\\Delta$ two-point functions on a single ensemble, the estimator reproduces the exact contraction; for the nucleon the sampling contribution to the error is sub-leading to the gauge noise. The paper also claims the method scales with the occupancy of the sampled index sets rather than the full size of distillation space, and extrapolates from the baryon results to a ten-fold cost reduction for a compact tetraquark correlator.","pith_inferences":["A cheap pre-production diagnostic, not reported in the paper, would be to measure the same-site weight fraction of baryon elementals on ensembles with different lattice spacing, volume, and boundary conditions; the whole cost argument depends on that fraction staying high.","The factorised probabilities drop the perambulator's index dependence, so a natural extension is a two-stage sampler that first draws large elemental entries and then corrects for large perambulator entries; the paper's estimator would then be closer to the variance-optimal Hansen-Hurwitz choice.","The phase noise seen in $N\\pi\\to\\Delta$ suggests that variance reduction should target complex-phase cancellations, for example by splitting the sum into real and imaginary parts or using a dense control variate.","If the locality holds at finer lattice spacings, the same basis could make larger distillation spaces affordable, which would help scattering and multi-hadron calculations that currently run at small $n_D$."],"forward_implications":["Nucleon correlators can be estimated with $n_s = 5\\times 10^4$ samples with sampling noise below gauge noise; Delta and $N\\pi\\to\\Delta$ correlators need $n_s = 8\\times 10^5$ and remain noisier, with phase cancellations hurting the cross-channel case.","The sparse contraction algorithm preserves the $O(n_D^{d+1})$ sequential-contraction scaling in the large-sample limit, with the central temporary tensors as the dominant cost.","Using the measured nucleon occupancies, a compact tetraquark calculation is estimated to need roughly $8\\times 10^5$ samples and to run about ten times faster than full distillation.","Increasing the physical volume at fixed grid spacing should improve the sparsity and hence the speed-up, because the number of coarse sites grows while the overlapping large entries do not.","The construction generalises to arbitrary tensor contractions with quark-line permutations through the marginal indicator functions of the paper's Eq. (38)."],"supporting_citations":[{"why":"Supplies the distillation framework whose operator and perambulator notation the paper builds on.","marker":"[1]"},{"why":"Provides the baryon operator construction and spin-flavour structure used in the test correlators.","marker":"[15]"},{"why":"Establishes the stochastic random-vector estimators in distillation space that the paper aims to improve on.","marker":"[19]"},{"why":"Introduces dilution schemes that reduce noise in those estimators and define the comparison point.","marker":"[20]"},{"why":"Supplies the Hansen-Hurwitz estimator that is the core sampling mechanism.","marker":"[21]"},{"why":"Provides the gauge and fermion actions defining the ensemble used for the numerical tests.","marker":"[22]"}],"fun_headline_variants":["Local basis sparsifies hadron correlators for stochastic sampling","Flow-localised distillation boosts stochastic efficiency in lattice QCD","Sparse local basis cuts stochastic cost for hadron sums","Locality in distillation space yields sparse baryon sums","Localised basis turns hadron correlators into sparse stochastic sums"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method's advantage rests on the unproven assumption that the flow-orthogonalised basis remains localised (elementals stay sparse) on other lattice spacings, volumes, and boundary conditions, and that perambulator entries do not vary strongly across distillation indices; if either fails, the sparse sampling loses its edge.","fun_headline_variants_meta":{"raw":{"variants":["Local basis sparsifies hadron correlators for stochastic sampling","Flow-localised distillation boosts stochastic efficiency in lattice QCD","Sparse local basis cuts stochastic cost for hadron sums","Locality in distillation space yields sparse baryon sums","Localised basis turns hadron correlators into sparse stochastic sums"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000243,"raw_usage":{"total_tokens":1459,"prompt_tokens":806,"completion_tokens":653,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":422,"completion_tokens_details":{"reasoning_tokens":571}},"tokens_in":422,"tokens_out":653,"duration_ms":7058,"temperature":1.0,"reasoning_tokens":571,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:40:34.634234+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the flow-orthogonalised basis on a second ensemble with a different lattice spacing or volume and measure the fraction of the total baryon elemental weight contained in same-site blocks; if that fraction drops toward the dense-case value, the cost advantage disappears. Separately, compare the variance of the estimator using the paper's factorised probabilities with the variance using probabilities that include perambulator magnitudes; a large gap would show the factorised choice is the limiting factor.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the gauge and fermion actions defining the ensemble used for the numerical tests."}],"review_version":1}