{"id":"fb67f75e-8bcf-4184-8b35-e80ccd40da07","arxiv_id":"1908.07050","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A propagator sparsening algorithm that samples only one site per 4^3 block reproduces the ground-state masses and binding energies of hadrons and light nuclei from full lattice QCD correlators.","lead":"This paper tests a sparsening trick that throws away most spatial information in quark propagators used to build lattice QCD correlation functions for light nuclei, keeping only every fourth spatial site. The sparsened correlation functions give the same ground-state energies and binding energies as the full calculation, pointing to large savings in the most expensive step of these simulations.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The core estimator is validated at exactly one blocking factor (N=4) on one ensemble; the theoretical guarantee in Sec. II is overbroad because the sparse sink aliases momenta, and no N-scan shows the agreement is not a tuned coincidence.","rationale":"The reader's conditional verdict is appropriate: the paper gives a clean demonstration at one ensemble that sparsened and full correlators yield consistent ground-state energies and binding energies. The regressions, plateaus, and fitted energies support that. No machine-checked proof or independent code is provided, but the empirical checks are reasonable. My stress-test does not overturn the verdict; it sharpens the reason for conditionality. The weakest spot is not primarily the extrapolation to physical pion mass (where the coherence length is longer and the method is likely safer), but the absence of any variation of the single parameter N and the overbroad theoretical guarantee about preserving the spectrum. Since N is the knob that controls both speedup and bias, an N-scan is the decisive test. I therefore keep the conditional verdict and mark partial agreement with the reader, whose focus was on the single ensemble and physical-mass extrapolation.","tokens_in":17779,"tokens_out":20999,"duration_ms":245793,"concrete_test":"On the same 32^3 x 48 ensemble, re-run the pion and nucleon ground-state fits for sparse correlators with N=2, N=6, and N=8, using the same source averaging, fit ranges, and jackknife procedure; additionally fit the sparse pion correlator at n=(4,0,0). If the sparse-vs-full energy differences and uncertainty ratios remain within roughly one sigma across N, and the n=(4,0,0) fit returns E(n=4) rather than the aliased E(n=-4), the coherence-scale criterion and the Sec. II guarantee are supported. If energies drift with N or the large-momentum fit follows the aliased minimum, the method's applicability must be restricted to the tested regime.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The method's only free parameter is the blocking factor N, fixed to 4 to match (a m_pi/2)^-1 ≈ 3.4. All consistency claims in Tables II-IV are at this one value. Eq. (3) does not average over a block; it keeps one site per 4^3 block, so the sink operator is a momentum-space comb coupling to momenta p + 2 pi m / (N a). For the momenta tested, |n| <= sqrt(5), the intended p is the lowest state in that comb, which is why the late-time fits can agree. The Section II statement that 'any choice of interpolating operator with correct quantum numbers ... is guaranteed to preserve' physical observables is therefore too strong: for larger momenta the sparse correlator's lowest state would be an aliased momentum, and even for the tested momenta the agreement could reflect a cancellation between aliased excited-state contributions and the reduced overlap with physical excited states. Because no comparison with N=2, N=6, or N=8 is reported, the single N=4 result cannot be distinguished from a tuned sweet spot. The speedup claim also lacks a timing measurement, but the more load-bearing issue is that the estimator's validity outside this one parameter point is unestablished.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a 'sparsening' algorithm for lattice QCD correlation functions: quark propagators are retained only on a coarsened spatial sublattice (one site per N^3 block), and correlators are assembled from these sparsened propagators. Using one 32^3 x 48 Wilson-clover ensemble at m_pi ~ 806 MeV and a ~ 0.145 fm, with N = 4, the authors compare effective energies and single-exponential ground-state fits for the pion, rho, nucleon, Delta, and for two-nucleon, 3He, and 4He channels, finding full and sparsened results statistically consistent in central values and jackknife uncertainties. They further propose a hybrid estimator (Eq. (10)) that corrects modified excited-state couplings using a small number of full-propagator source locations, and claim O(10-100) speedups in the contraction stage.","tokens_in":17954,"tokens_out":8396,"duration_ms":90741,"significance":"If the consistency holds beyond the tested ensemble and blocking factor, the method is a practical and welcome tool: contraction costs in multi-hadron calculations are often dominant, and a reduction by roughly N^3 in the number of sink sites would be valuable. The paper's statistical analysis is careful: correlated single-exponential fits, jackknife errors, condition numbers, scatter regressions with R^2, and effective-mass plateaus are all reported, and the hybrid estimator is a sensible adaptation of all-mode-averaging ideas. The central caveats are that the advertised speedup is not measured and the consistency result rests on a single blocking factor at a single heavy-pion ensemble.","major_comments":[{"comment":"The statement that sparsening is 'guaranteed to preserve' physical observables because any interpolating operator with the correct quantum numbers is equally valid is too strong. The sparsened sink is not a definite-momentum operator: Eq. (3) couples to the momentum comb p + 2 pi m / (N a). For the momenta tested, |n| <= sqrt(5), p is indeed the lowest state in this comb, but for larger momenta the lowest state can be an aliased momentum, so the extracted ground state would not be the intended one. The consistency claim should either be restricted to the tested kinematic window or supplemented by a demonstration with larger momenta (or other N values) showing that aliased states do not contaminate the extracted energies.","section":"Section II, after Eq. (3)"},{"comment":"All consistency results are at the single blocking factor N = 4, chosen from the correlation-length estimate (a m_pi / 2)^{-1} ~ 3.4. Because N is the only free parameter in the method, the absence of an N-scan (e.g., N = 2, 6, 8) leaves open the possibility that the agreement is a tuned coincidence. A scan over N would show whether the plateau consistency and fit results vary smoothly with N and would establish robustness; without it, the claim that sparsening preserves ground-state energies is not supported beyond this one parameter point on this one ensemble.","section":"Tables II-IV and Figures 2-5"},{"comment":"The claim that the method enables O(10-100) fold speedups in the contraction stage is not supported by any timing measurement, operation count, or benchmark in the paper. The reduction in sink sites is a factor of N^3 = 64 for N = 4, but actual contraction speed depends on implementation, memory access, and the overhead of sparse data structures. Please report measured timings for the contractions used here, or rephrase the conclusion as an expected asymptotic speedup based on site reduction with appropriate caveats.","section":"Section IV (Conclusions)"}],"minor_comments":[{"comment":"The cost-effectiveness of the hybrid estimator is not quantified; please report the number of full-propagator source locations N_Delta needed to reach a given bias/error target, and state whether this is small compared with N_sparse for the ensemble studied.","section":"Section III B, Eq. (10)"},{"comment":"The word 'sophisticted' in the paragraph on blocking procedures should be corrected to 'sophisticated', and the terminology 'sparsified' in Section IV should be made consistent with the 'sparsened' used elsewhere.","section":"Section II"},{"comment":"The text states that the scale of correlations is supported by a numerical study in companion proceedings Ref. [60]; a brief summary of that study in the main text would make the choice of N = 4 more self-contained.","section":"Section III A 1"},{"comment":"The convergence of the hybrid estimator in N_Delta is shown only for sources on a single time slice; the authors note this in the text, but a short discussion of how the required N_Delta would scale with a more distributed source ensemble would help readers assess the practical cost reduction.","section":"Section III B, Figure 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid methods demonstration; the main revisions needed are a measured or properly qualified speedup claim and an N-dependence check (or a restriction of the claims to the tested regime). I see no issues with novelty or attribution. The paper fits the journal's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a careful read before accepting the abstract at face value. The empirical core is solid: on one 32^3x48 ensemble at m_pi ~ 800 MeV, they show that sparsening the sink propagators to one site per 4^3 spatial block reproduces the full-lattice ground-state energies and binding energies for pion, rho, nucleon, Delta, and the deuteron/dinucleon, 3He, and 4He, with statistical errors essentially unchanged. The regression slopes, effective-mass plateaus, and correlated single-exponential fits all line up. That is real work, carefully presented, and the hybrid estimator in Eq. (10) is a sensible adaptation of all-mode averaging to remove the modified excited-state overlaps they clearly identify.\n\nThe soft spots are two, and they sit in the introduction and conclusion more than in the data. First, the O(10-100) contraction-stage speedup is never measured. No timings, no operation counts, no comparison of the cost of building baryon blocks on full versus sparse propagators. The claim in the conclusions is an extrapolation. It may well be true, but it is not demonstrated here.\n\nSecond, the stress-test note about aliasing is on target. Eq. (3) is not a block average; it keeps one lattice site per block, so the sink operator couples to a comb of momenta p + 2 pi m/(N a). The statement in Section II that \"any choice of interpolating operator with correct quantum numbers ... is guaranteed to preserve physical observables\" is overbroad. For the tested momenta |n|^2 <= 5 the intended momentum is the lowest in the comb, so the late-time agreement is plausible. But for larger momenta the aliased state could become the lowest, and the argument as written would not cover that. The paper would be stronger with an N-scan (N=2,6,8) to show the agreement is not a sweet spot. That said, the absence of that scan is a validation gap, not a red flag; nothing in the data suggests tuning.\n\nAlso worth noting: everything is at one unphysical quark mass, one lattice spacing, one blocking factor. That limits generality, but the authors are mostly careful about this.\n\nVerdict: this deserves a serious referee. The algorithmic idea is genuinely useful for nuclear LQCD, the analysis is above the usual bar, and the gaps are fixable: add a timing measurement and an N-scan, soften the guarantee, and the paper is in good shape.","headline":"Solid, carefully analyzed algorithmic paper with a real consistency check at one parameter point, but the advertised speedup is unmeasured and the Section II guarantee is overbroad due to momentum aliasing.","tokens_in":18560,"tokens_out":2442,"would_cite":true,"duration_ms":25068,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["11.15.Ha","12.38.Gc"],"model":"deepseek-v4-flash","headline":"This paper claims that correlation functions built from quark propagators sampled on a coarsened spatial lattice—one site per 4^3 block—give ground-state masses and binding energies statistically indistinguishable from full propagator…","keywords":["lattice QCD","propagator sparsening","multi-hadron correlation functions","light nuclei","baryon blocks","excited-state contamination","all-mode averaging","contraction cost"],"falsifier":"Repeat the comparison of sparsened and full correlation functions on an ensemble with m_pi near the physical pion mass (about 140 MeV) and a finer lattice spacing, extracting the nucleon and deuteron ground-state energies from both; if the sparsened results differ from the full results by more than the combined statistical uncertainty, the method's central claim is refuted. A cheaper check is to increase the blocking factor on the existing ensemble until the effective masses disagree.","tokens_in":17532,"feed_emoji":"⚛️","tokens_out":5808,"duration_ms":51847,"temperature":0.7,"pith_summary":"This paper proposes a way to cut the cost of the most expensive step in lattice QCD calculations of light nuclei: assembling correlation functions from quark propagators. The idea is to define 'sparsened' propagators by keeping only the value at the first site of each spatial block, and to build correlation functions from these coarsened objects. On a single ensemble with pion mass around 800 MeV, the ground-state energies and binding energies extracted from sparsened correlation functions agree with those from full propagators, with no loss in statistical precision. The method also changes the overlap onto excited states, but a cheap correction removes that effect. If the result holds, it enables order-of-magnitude speedups in nuclear lattice calculations.","feed_headline":"Sparsened propagators give the same QCD spectra at 10-100x lower cost","feed_subtitle":"Ground-state masses and binding energies from coarse-blocked propagators match full-lattice results on a test ensemble.","key_machinery":"The central object is the sparsened propagator: on a lattice blocked by $N^{3}$ in the spatial directions, only the value at the first site of each block is retained, defining a propagator on the coarsened sublattice $\\tilde{\\Lambda}^3(N)$. Correlation functions are then built by summing only over this sublattice, which reduces the cost of the Fourier transforms and contractions that dominate multi-hadron calculations. The blocking scale N = 4 is chosen from the estimated spatial coherence length of hadronic two-point functions, $(a m_\\pi/2)^{-1} \\approx 3.4$ lattice units; the construction is justified by noting that sparsening is equivalent to replacing the sink interpolating operator.","core_discovery":"The central claim is that a simple spatial blocking prescription—uniformly blocking the lattice by a factor N = 4 and taking the propagator value at the first site of each block—produces a sparsened correlation function whose ground-state plateau, extracted energy, and jackknife uncertainty are consistent with those of the full correlation function, for pions, rho mesons, nucleons, $\\Delta$ baryons, and the NN, 3He, and 4He systems. The authors argue this is guaranteed at the level of the theory because sparsening only changes the sink interpolating operator, and any operator with the right quantum numbers probes the same finite-volume spectrum; the practical question is whether the overlaps and statistical noise are preserved. They find that excited-state couplings are modified at early Euclidean times, especially at higher momentum, but that a modified estimator with a small number of full-propagator correction sources removes this contamination.","pith_inferences":["If the coherence-length estimate scales inversely with the pion mass, then at physical quark mass the safe blocking factor could be larger than 4, making sparsening even more effective—but the short-distance structure of heavier nuclei may set a stricter limit.","The correction term in the modified estimator is essentially an all-mode-averaging correction applied in position space; combining sparsening with low-mode deflation of the Dirac operator could give a systematic hierarchy of approximations with controlled bias.","Sparsening could also be applied at the source rather than the sink, or used to compress stored propagators, reducing both memory and I/O, though the paper does not explore this.","A practical test for future ensembles is to check that the ratio of sparsened to full effective masses approaches unity within errors at every plateau time; if the residual deviates at larger momenta, the blocking scale must be reduced."],"forward_implications":["Contraction-stage cost for multi-nucleon systems drops by roughly the block volume factor, enabling calculations of heavier nuclei or larger variational bases at fixed computational budget.","The modified estimator with a small number of full-propagator sources removes the sparsening-induced excited-state contamination, so the method can be used in spectroscopy with large operator bases.","Because the allowed blocking factor grows as the lattice spacing shrinks relative to the hadronic scale, the speedup increases toward the continuum limit.","The same sparsening principle should apply to three-point functions and background-field calculations, where block construction currently dominates the cost."],"supporting_citations":[{"why":"Supplies the Wilson-clover ensemble, the lattice spacing determination, and the prior nuclear spectrum results used for comparison.","marker":"[9]"},{"why":"Provides the baryon-block construction and interpolating-operator weights used to build multi-hadron correlation functions.","marker":"[52]"},{"why":"Gives the factorization of the nucleon correlation function used to estimate the spatial coherence scale that sets the blocking factor.","marker":"[59]"},{"why":"Introduces the all-mode averaging technique that inspires the modified estimator used to remove excited-state contamination.","marker":"[61]"},{"why":"Establishes the baryon block algorithm whose contraction cost sparsening is designed to reduce.","marker":"[3]"},{"why":"Documents a computational implementation of baryon blocks, serving as a cost reference for the contraction stage.","marker":"[19]"}],"fun_headline_variants":["Sparsened propagators match full QCD spectra at lower cost","Blocked propagators give same nuclear ground states","Sparsening propagators: same spectra, 10-100x cheaper","Consistent QCD masses from coarse-blocked propagators"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that quark propagators are smooth enough that sampling one site per $4^{3}$ block preserves the ground-state signal; this has only been demonstrated on a single ensemble with a heavy pion mass, so it could fail at physical quark masses or finer lattice spacings.","fun_headline_variants_meta":{"raw":{"variants":["Sparsened propagators match full QCD spectra at lower cost","Blocked propagators give same nuclear ground states","Sparsening propagators: same spectra, 10-100x cheaper","Consistent QCD masses from coarse-blocked propagators"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000237,"raw_usage":{"total_tokens":1508,"prompt_tokens":948,"completion_tokens":560,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":564,"completion_tokens_details":{"reasoning_tokens":487}},"tokens_in":564,"tokens_out":560,"duration_ms":6128,"temperature":1.0,"reasoning_tokens":487,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:27:43.306953+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the comparison of sparsened and full correlation functions on an ensemble with m_pi near the physical pion mass (about 140 MeV) and a finer lattice spacing, extracting the nucleon and deuteron ground-state energies from both; if the sparsened results differ from the full results by more than the combined statistical uncertainty, the method's central claim is refuted. A cheaper check is to increase the blocking factor on the existing ensemble until the effective masses disagree.","supporting_citations":[{"cited_title":"L¨ uscher and P","cited_arxiv_id":null,"evidence_quote":"Introduces the all-mode averaging technique that inspires the modified estimator used to remove excited-state contamination."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the baryon block algorithm whose contraction cost sparsening is designed to reduce."},{"cited_title":"Calm Multi-Baryon Operators","cited_arxiv_id":"1710.05642","evidence_quote":"Documents a computational implementation of baryon blocks, serving as a cost reference for the contraction stage."}],"review_version":1}