{"id":"c7ad2299-7779-46c2-8f32-cc2d0719fbb3","arxiv_id":"2501.05404","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Repeated gauge-covariant averaging (blocking) before decimation reproduces unsparsened two- and three-point correlation functions more faithfully than plain decimation, but with larger statistical uncertainties and a poorly controlled endpoint comparison.","lead":"This paper tests a way to make lattice QCD calculations cheaper by averaging neighboring quark-propagator sites before deleting most of the grid, called blocking plus decimation. It finds that repeating the averaging step many times preserves hadron energies and form factors best on the small test lattice, at the price of larger statistical errors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'many blocking steps' claim is confounded: n=1/n=5 runs block only the sink while n=20 blocks source+sink, and Sec. IV C even contradicts the figure captions about which runs are symmetric.","rationale":"The reader identified the core confound: n=1 and n=5 are sink-only while n=20 is source+sink, so the improvement attributed to 'many times' is not cleanly separated from symmetric blocking. I agree, and I want to stress that this is not a minor presentation issue but a direct threat to the abstract's central claim. The paper's own Sec. IV C adds an internal contradiction by stating that symmetric blocking was studied for F1, F5, and F20, while the figure captions say n=1 and n=5 were sink-only. If the Sec. IV C statement is true, then the figures omit the controlled comparison; if it is false, the central comparison lacks a control. Either way, the evidence as presented does not establish that sequential application many times is the cause of improved fidelity. I also note that the free-propagator appendix and the blocking-to-smearing equivalence are independent and sound; they support the general mechanism that blocking acts like smearing, but they do not resolve the n versus source/sink confound. The correct response is not rejection: the paper contains a plausible and useful numerical study, and the confound is addressable by a straightforward rerun or by reporting existing symmetric n=1/n=5 data. Hence CONDITIONAL seems the right verdict, consistent with the reader. I mark agreement as partial because I add the internal inconsistency and emphasize that the paper may already contain the needed controlled data that was simply not shown.","tokens_in":20402,"tokens_out":4024,"duration_ms":39293,"concrete_test":"Recompute the n=1, n=5, and n=20 comparisons at α=β=1 under two protocols on the same 24³×48 ensemble: (i) sink-only blocking and (ii) symmetric source+sink blocking, for decimation factors s=2 and s=6, and evaluate the metrics M1/M2 and the effective-energy plateaus shown in Figs. 5 and 6. If symmetric n=1 matches symmetric n=20, the improvement is due to symmetric blocking rather than sequential many-step averaging; if only sink-only n=1 matches symmetric n=20, the sequential-many-times claim survives. This check also resolves the Sec. IV C / Fig. 5 caption contradiction about which runs are symmetric.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central assertion is that sparsening is most effective when weighted covariant-averaging is sequentially applied many times. The data used to support this claim, however, do not isolate the number of blocking steps. In Figs. 5, 6, 7, and 8, the n=1 and n=5 cases block only the sink, while n=20 blocks both source and sink (as stated in the figure captions). Thus the improvement seen for n=20 could be caused by symmetric source+sink smearing of the interpolating operators rather than by sequential application of blocking. This is a known effect: the authors themselves note that repeated blocking reduces excited-state contamination 'in-line with previous results for operator smearing.' Because the undecimated n=20 correlator is itself smeared at both source and sink, the apparent improved agreement between decimated and undecimated results at n=20 may be an artifact of comparing differently smeared operators, not evidence that more blocking steps preserve long-distance information after decimation. Additionally, Sec. IV C states that 'For α = β = 1, we study symmetric blocking of the source and sink using F1, F5, and F20,' which directly contradicts the figure captions that n=1 and n=5 are sink-only. If symmetric n=1/n=5 data exist, they are not shown in the comparison figures; if they do not, the central comparison is not a controlled test of the number of blocking steps. This is load-bearing because the headline recommendation requires varying n at fixed source/sink treatment.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a general scheme for quark-propagator sparsening that combines gauge-covariant nearest- and next-to-nearest-neighbor blocking with spatial decimation, and studies the resulting pion, proton, and Delta two-point functions and pion/proton three-point functions on a 24^3 x 48 Wilson-clover ensemble. The central claim, stated in the abstract and reiterated in Sec. VII, is that sparsening is most effective in reproducing unsparsened correlation functions when weighted covariant averaging is sequentially applied many times. The paper also derives an analytic equivalence between blocking and smeared interpolating operators (Sec. II C) and includes a free-propagator consistency check (Appendix A).","tokens_in":1342,"tokens_out":1447,"duration_ms":58007,"significance":"If the central claim is correct, the work has clear practical value: it would justify larger decimation factors for sparsened propagators, reducing storage and Wick-contraction costs in multi-nucleon and many-body lattice QCD calculations. The analytic relation in Sec. II C between gauge-covariant blocking and smeared interpolating operators is a clean and useful contribution, and the free-propagator study in Appendix A is a sensible independent toy model that supports the qualitative dependence on alpha and beta seen in the QCD results. However, as detailed below, the numerical evidence for the headline claim about the number of blocking steps is confounded, so the quantitative conclusion is not yet established.","major_comments":[{"comment":"The comparison that underpins the abstract claim is confounded. In Fig. 5 (and similarly in Figs. 6, 7, 8, and 9), the n=1 and n=5 runs block only the sink, while the n=20 runs block both the source and the sink. The improved agreement between decimated and unsparsened results at n=20 could therefore be caused by symmetric source-sink smearing rather than by the number of sequential blocking steps. The text in Sec. IV C states 'For α = β = 1, we study symmetric blocking of the source and sink using F1, F5, and F20', which directly contradicts the figure captions that n=1 and n=5 are sink-only. If symmetric n=1 and n=5 data exist, they should be included in the comparison figures; if they do not, the comparison does not isolate the effect of varying n. This is load-bearing because the paper's central claim is precisely that 'many' blocking steps are most effective.","section":"Sec. IV C; Fig. 5 and Figs. 6-9"},{"comment":"Even if the source/sink asymmetry were fixed, the reference 'unsparsened' correlator is not the undecimated correlator at the same blocking level. The pink band in Fig. 5 appears to denote the n=0, s=1 result, so the distance between the n=20 decimated curve and this band is compared with the distance between the n=1 decimated curve and the same n=0 band. Because increasing n itself reduces excited-state contamination (as the paper acknowledges, 'in-line with previous results for operator smearing'), the improved agreement at n=20 may simply reflect the known smearing effect on the undecimated correlator rather than enhanced preservation of long-distance information after decimation. A controlled measure would compare the decimated n-step correlator against the undecimated n-step correlator for each n, or otherwise report M1 with the reference correlator at the same blocking level. This point is central to the interpretation of Figs. 5-9 and should be addressed.","section":"Sec. IV C, Eq. (25), Fig. 5"},{"comment":"There is an internal inconsistency about which runs are symmetric. The caption of Table I states that the quoted uncertainties are 'for sparsening both the source and sink with α = β = 1' and lists n=1 and n=5 entries, while the caption of Fig. 6 states that 'in the case of n = 1 and 5, we block only the sink but for n = 20, we block both the source and sink'. The manuscript must specify unambiguously which runs are sink-only and which are symmetric; if Table I includes symmetric n=1 and n=5 runs, the corresponding effective-energy comparisons should be shown, and if it does not, the table caption must be corrected. This inconsistency affects the reproducibility of the reported results.","section":"Table I; Fig. 6 caption"}],"minor_comments":[{"comment":"The caption labels the second panel as 'Left' when it should be 'Right'.","section":"Fig. 5 caption"},{"comment":"In the final term of Eq. (5), the factor U†_{μ2}(x) appears where a gauge link attached to the source coordinate y would be expected; please check the placement of the arguments in this term.","section":"Eq. (5), last line"},{"comment":"The sentence '⟨π†(p,t)π(p,0)⟩⟩' contains a duplicated angle bracket from the PDF typesetting; this should be corrected.","section":"Sec. II A, text after Eq. (1)"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the scope of the journal and the analytic equivalence in Sec. II C is a solid contribution. The numerical evidence for the headline claim, however, is not yet a controlled test of the number of blocking steps. The fix is straightforward in principle: show symmetric n=1 and n=5 comparisons or restrict the claim appropriately. I do not see a need to reject; the issue is fixable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read it. The genuinely new piece is the gauge-covariant nearest- and next-to-nearest-neighbor blocking with sequential iteration, applied to baryon two-point functions and vector-current three-point form factors. Section II C, showing that blocking the propagator is equivalent to using a smeared interpolating operator, is clean and worth keeping. The free-propagator check in Appendix A is a sensible independent toy model and it supports the qualitative trend that repeated blocking helps for positive couplings.\n\nThe soft spot is the headline claim. The abstract says sparsening is most effective 'when weighted covariant-averaging is sequentially applied many times,' but the main comparison does not isolate the number of steps. In Figs. 5, 6, 7, and 8, the n=1 and n=5 runs block only the sink, while n=20 runs block both source and sink. The improvement at n=20 could be due to symmetric smearing of both operators rather than to sequential blocking per se. The authors themselves note that repeated blocking reduces excited-state contamination 'in-line with previous results for operator smearing,' so the confound is not hypothetical. To make the claim stick, they need a controlled comparison with symmetric blocking at n=1 and n=5, or a clear statement of what data exist.\n\nThere is also a direct internal contradiction: Sec. IV C says 'For α = β = 1, we study symmetric blocking of the source and sink using F1, F5, and F20,' while the figure captions say n=1 and n=5 are sink-only. That needs fixing in revision.\n\nMinor points: the results are from a single 24^3 x 48 ensemble at unphysical quark mass with 100 configurations, the key comparison figures lack error bars on the n-dependence, and there are no public code or data artifacts. None of these are fatal; they are addressable.\n\nBottom line: the method is worth developing and the analytic part is solid, but the central conclusion is currently unproven. This deserves a serious referee who will ask for a controlled n comparison. I'd send it to review rather than desk-reject, but the revision should be significant.","headline":"A useful sparsening extension with a clean analytic core, but the central 'many steps are best' claim is confounded by mixing sink-only and source+sink blocking.","tokens_in":21260,"tokens_out":3074,"would_cite":false,"duration_ms":26429,"reading_group":"maybe","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 sequentially applying gauge-covariant averaging to quark propagators before decimating them preserves pion, proton, and Delta correlation functions and form-factor ratios far better than plain decimation, allowing…","keywords":["lattice QCD","propagator sparsening","gauge-covariant blocking","decimation","hadron correlation functions","form factors","Wick contractions"],"falsifier":"Compute the same correlation functions with symmetric source-sink blocking at n = 1, n = 2, and n = 5 at fixed decimation, and compare against n = 20; if one symmetric step already reproduces the n = 20 fidelity, then the number of sequential steps is not the cause. A second check would be a controlled scan of n = 10, 20, and 40 to see whether fidelity improves monotonically with step count.","tokens_in":20201,"feed_emoji":"⚛️","tokens_out":7993,"duration_ms":74463,"temperature":0.7,"pith_summary":"This paper asks how aggressively a quark propagator can be compressed before hadron physics is lost, and tests a compression recipe: instead of simply discarding lattice sites, first average each site with its gauge-transported nearest and next-to-nearest neighbors, and repeat that averaging many times. The claim is that sequential weighted covariant averaging is the most effective sparsening studied here for reproducing unsparsened two- and three-point correlation functions, including ground-state energies and vector-current form-factor ratios. If true, lattice QCD calculations can store smaller propagators and pay less for Wick contractions without sacrificing the long-distance, low-energy information they need. The paper also documents a trade-off: more averaging steps suppress excited states but add statistical noise.","feed_headline":"20 blocking steps preserve hadron data after heavy decimation","feed_subtitle":"Sequential weighted averaging before decimation keeps low-energy hadron physics intact, enabling smaller propagators.","key_machinery":"The central object is the blocking map F in Eq. (5), which replaces a propagator S(x|y) by a gauge-covariant average of S over nearest-neighbor (weight $\\alpha$) and next-to-nearest-neighbor (weight $\\beta$) lattice sites, transporting quark fields with gauge links so the result transforms correctly under gauge rotations. Applying F n times builds paths of length up to n between source and sink; for $\\alpha$ = $\\beta$ = 1 and n = 20, the averaged operator corresponds to a strongly smeared interpolating operator. This is what carries the argument: blocking before decimation is equivalent to constructing correlation functions with smeared operators, and sequential repetition systematically suppresses the higher-momentum modes that plain decimation would otherwise leave in the correlation function. The paper measures fidelity with three metrics, M1, M2, and M3, comparing sparsened to unsparsened effective energies and noise, and uses the ratio R^h, with a square-root-of-two correction at the maximal momentum q = pi/s, to extract form-factor information.","core_discovery":"The paper's central claim is that weighted covariant-averaging applied sequentially many times before decimation preserves low-energy physics better than plain decimation at the same decimation factor. On a $24^{3}$ x 48 lattice with an improved clover fermion action, the authors compute pion, proton, and $\\Delta$ two-point functions and pion and proton vector-current three-point functions, sparsened by factors s = 2, 3, 4, 6, 8, 12 with one, five, or twenty blocking steps at couplings $\\alpha$ = $\\beta$ = 1. They report that twenty blocking steps make the sparsened correlation functions agree with their unsparsened versions at earlier times and allow larger decimation factors, at the price of increased statistical uncertainties; the extracted ground-state energies and the ratio R^h used for form factors remain consistent with the unsparsened results. This is an extension claim: not a new law of nature, but evidence that a particular sparsening construction inherits the robust-energy result of decimation while improving excited-state control.","pith_inferences":["The paper's data do not isolate the number of blocking steps from the source-sink symmetry of the blocking, so a direct test with symmetric n = 1 blocking versus n = 20 blocking would settle whether sequential averaging itself is the active ingredient.","If the step-count effect survives that control, tuning alpha and beta separately for each hadron and momentum is a natural optimization, since the reported metrics show a broad region of good fidelity near alpha = beta = 1.","The same sequential covariant averaging could be applied to the gauge field or to other field objects, potentially moving storage savings earlier in the pipeline; the paper notes the construction generalizes but does not test it.","A practical adoption would require a floating-point cost benchmark: for multi-baryon Wick contractions, the extra blocking steps must cost less than the contraction savings they enable, which this paper does not quantify."],"forward_implications":["With sequential blocking at alpha = beta = 1 and n = 20, decimation factors up to at least s = 12 keep extracted hadron energies consistent with unsparsened values.","Sparsened three-point functions reproduce the vector-current ratio R^h across several momenta and operator insertion times, which carries over to form-factor extractions.","Because storage and Wick-contraction costs scale with the number of retained sites, larger usable decimation factors translate directly into smaller propagators and cheaper later stages of multi-hadron calculations.","The optimum sparsening choice is a three-way balance: more blocking steps suppress excited states, but increase statistical noise, so the best choice depends on the statistics budget and the target observable."],"supporting_citations":[{"why":"Supplies the plain-decimation baseline this work extends; its result that decimation robustly reproduces ground-state energies but adds excited-state contamination is the starting point.","marker":"[18]"},{"why":"Supplies the random-sampling sparsening baseline for two- and three-point meson correlation functions that the covariant-averaging construction generalizes.","marker":"[19]"},{"why":"Motivates the application of propagator sparsening by showing its use to reduce computationally intensive momentum projections in dibaryon correlation functions.","marker":"[20]"},{"why":"Motivates the cost savings by using random sparsening to reduce an O(V^2) electromagnetic contribution to the pion mass splitting.","marker":"[24]"},{"why":"Establishes the smearing correspondence used to interpret sequential blocking, since iterative smearing of that type matches the blocking map at alpha = 1, beta = 0.","marker":"[26]"}],"fun_headline_variants":["20 blocking steps allow larger QCD decimation factors","Sequential averaging preserves low-energy hadron physics","Sparsening gains from repeated weighted averaging"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The comparison that leads to the claim that many sequential averaging steps are the key mixes two changes at once: n = 20 blocks both source and sink, while n = 1 and n = 5 block only the sink, so if symmetric source-sink blocking alone explains the improvement, the central claim about many steps is not established.","fun_headline_variants_meta":{"raw":{"variants":["20 blocking steps allow larger QCD decimation factors","Sequential averaging preserves low-energy hadron physics","Sparsening gains from repeated weighted averaging"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001396,"raw_usage":{"total_tokens":5641,"prompt_tokens":931,"completion_tokens":4710,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":547,"completion_tokens_details":{"reasoning_tokens":4662}},"tokens_in":547,"tokens_out":4710,"duration_ms":29929,"temperature":1.0,"reasoning_tokens":4662,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T21:14:00.405426+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the same correlation functions with symmetric source-sink blocking at n = 1, n = 2, and n = 5 at fixed decimation, and compare against n = 20; if one symmetric step already reproduces the n = 20 fidelity, then the number of sequential steps is not the cause. A second check would be a controlled scan of n = 10, 20, and 40 to see whether fidelity improves monotonically with step count.","supporting_citations":[{"cited_title":"The theoretical background and properties of perfect actions","cited_arxiv_id":"hep-lat/9803027","evidence_quote":"Motivates the application of propagator sparsening by showing its use to reduce computationally intensive momentum projections in dibaryon correlation functions."}],"review_version":1}