{"id":"dc99350a-fc3c-4f5b-8770-a0be9d1dd55e","arxiv_id":"2502.03187","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Low Mode Averaging improves lattice precision enough to resolve the rho resonance in the energy-smeared R-ratio at sigma=250 MeV.","lead":"This lattice QCD paper shows that Low Mode Averaging cuts the statistical error of a key correlation function by about a factor of 3.6 and yields a net speedup of about 5 for the same accuracy. The gain makes it possible to compute the R-ratio smeared with a 250 MeV Gaussian kernel, narrow enough to see the rho resonance.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"HLT systematic control at σ=250 MeV is not established; the rho-visibility claim rests on statistical errors alone.","rationale":"The reader's weakest assumption correctly identifies the absence of systematic-error control for HLT at σ=250 MeV as the key condition for the paper's central claim. My independent reading confirms that this is the most load-bearing concern: the LMA methodology itself is internally consistent and the statistical gains are demonstrated with explicit numbers, but the extension to a narrower smearing kernel is not backed by the stability checks that would make the 'rho visible' claim trustworthy. The disagreement with the reader is minimal; the concern is the same, and the appropriate verdict remains CONDITIONAL. I do not see an internal inconsistency or a fraudulent step; the paper is an honest preliminary proceedings contribution. The concrete test I propose would settle the concern by measuring the HLT bias at σ=250 MeV on synthetic data with realistic noise, which directly probes truncation and regularization systematics. If that test passes, the conditional can be lifted; if it fails, the rho-visibility claim would need to be softened or accompanied by a systematic error budget.","tokens_in":9543,"tokens_out":11204,"duration_ms":104436,"concrete_test":"Generate synthetic correlators from a known spectral function (e.g., a rho Breit-Wigner plus a smooth continuum), add noise matching the LMA-improved statistical errors on B64, and run the HLT pipeline exactly as in Fig. 5 at σ=250 MeV. Scan τ_max from half to twice the value used and scan λ across the stability plateau; compare the reconstructed R_σ(E) at E≈770 MeV with the exact smeared value. If the bias or the τ_max/λ spread exceeds the 3% statistical error, the claim that σ=250 MeV is systematically controlled fails. Run the same test at σ=440 MeV as a control to confirm the method's baseline.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central methodological claim is that LMA enables HLT spectral reconstruction at a Gaussian width of σ=250 MeV, making the ρ resonance visible. The LMA part is well supported: Eq. (13) gives an unbiased deflated estimator, and the measured gains (3.63(15) for TM, 3.57(16) for OS) are consistent across two regularizations, with an explicit cost model yielding a net speedup of about 5. The load-bearing weakness is the HLT extension to σ=250 MeV. Ref. [1] validated HLT for widths of 440–630 MeV; the present paper applies it at 250 MeV without a systematic error analysis for this new regime. The HLT estimator (Eqs. 15–17) has two controlled-but-quantified sources of bias: truncation of the exponential basis at τ_max in Eq. (15), and the choice of regularization parameter λ and weight w_n in Eqs. (16)–(17). A narrower Gaussian kernel makes the target function G_σ(E−ω)/ω^2 harder to represent in the finite exponential basis, so these systematics are expected to be more severe than at the previously validated widths. The paper shows no τ_max scan, no λ-stability study, and no synthetic-data test at σ=250 MeV. The agreement between TM and OS in Fig. 5 (right) is a useful cross-check, but both discretizations share the same HLT systematic. The reported 11%→3% error reduction is purely statistical; if the HLT bias at 250 MeV is comparable to or larger than the 3% statistical error, the claim that the ρ resonance is resolved is not established. The paper itself labels the results preliminary and blinded (Section 3, Fig. 5 caption) and defers details to Ref. [15], consistent with this missing systematic budget.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a proof-of-concept application of Low Mode Averaging (LMA) to the light-quark connected vector-vector correlator in the ETMC mixed-action setup, with the goal of enabling Hansen-Lupo-Tantalo (HLT) spectral reconstruction of the smeared R-ratio at a Gaussian width of sigma = 250 MeV. The LMA estimator is defined in Eq. (13), and the authors show that it reduces the statistical error of the correlator by a factor of 3.63(15) for TM and 3.57(16) for OS regularizations, with a net computational speedup of about 5. Using these improved correlators, they present preliminary, blinded results for the light connected contribution to the smeared R-ratio at sigma = 250 MeV, reporting that the relative error at E around 770 MeV drops from about 11% to 3%, which they interpret as enough to resolve the rho resonance.","tokens_in":9864,"tokens_out":5310,"duration_ms":48873,"significance":"If the HLT systematic errors at sigma = 250 MeV are under control, this paper would be a significant methodological advance: LMA makes spectral reconstruction at smaller Gaussian widths feasible, and the quantitative error-reduction benchmark against data from Ref. [11] is a concrete and reproducible result. The LMA part is well supported: Eq. (13) is a clean deflation identity, and the measured noise reductions and cost model give a net speedup of about 5. The HLT extension to sigma = 250 MeV, however, is the load-bearing novelty, and it is not yet backed by a systematic error analysis. The agreement between TM and OS is a useful cross-check but does not constrain the shared HLT bias. The paper is a preliminary proceedings contribution, and the central physics claim therefore remains conditional on the missing systematic validation.","major_comments":[{"comment":"The central claim that sigma = 250 MeV is sufficient to resolve the rho resonance is not yet supported by a systematic error analysis for the HLT reconstruction at this width. The approximation space in Eq. (15) is a truncated exponential basis, and the Tikhonov regularization in Eq. (17) introduces a bias; both become more severe as the Gaussian kernel narrows. Ref. [1] validated HLT for widths of 440-630 MeV, while the present paper uses sigma = 250 MeV without providing a tau_max scan, a lambda-stability study, or a synthetic-data test at this width. Consequently, the reported reduction from 11% to 3% at E ~ 770 MeV is purely statistical. The TM/OS agreement in Fig. 5 (right) is a useful cross-check, but it does not constrain the shared HLT systematic. I ask for an explicit HLT stability analysis or a synthetic-data validation at sigma = 250 MeV before the rho-resolution claim is made.","section":"Sec. 3, Eqs. (15)-(17), Fig. 5"},{"comment":"No systematic error budget is provided for the smeared R-ratio: there is no continuum extrapolation (only two lattice spacings, B64 and C80, are shown), no estimate of finite-volume, O(a), or isospin-breaking effects, and the results are blinded. As a result, the error bars in Fig. 5 do not represent the total uncertainty relevant for comparing with the rho peak, and the TM/OS agreement cannot be taken as evidence of continuum behavior. A proceedings paper may report preliminary status, but the phrase 'enough to appreciate the rho resonance' should be qualified as referring to statistical precision only, pending the systematic analysis.","section":"Sec. 3, Fig. 5"},{"comment":"The plotted quantity is the connected light-quark (u/d) contribution to the smeared R-ratio, not the full R-ratio defined in Eq. (2), which also receives disconnected and heavier-quark contributions. The text should consistently say 'light-quark connected contribution' when discussing the rho signal, and it should state how much of the rho peak is expected to arise from this partial contribution. Without this qualification, the claim that the full R-ratio has been resolved at sigma = 250 MeV is stronger than what the data show.","section":"Sec. 3, Fig. 5"}],"minor_comments":[{"comment":"The formula for the Gain contains a stray symbol 'vt' before the two ratios; this is likely a LaTeX artifact and should be removed.","section":"Eq. (14), Sec. 2.2"},{"comment":"The name 'Hansen-Lupo-Tantatlo' is misspelled; it should be 'Hansen-Lupo-Tantalo', matching the reference list and Sec. 2.3.","section":"Title and Abstract"},{"comment":"For the B64 ensemble the optimal value Neig = 400 is supported by Fig. 2, but Table 1 lists Neig = 530 for C80 and D96 without showing the corresponding optimization scans; the choice for the finer ensembles should be justified or described as a scaling assumption.","section":"Sec. 2.2, Table 1"},{"comment":"The notation for the weight function is inconsistent: the text says 'weight-functions w_n > 0', while Eq. (16) uses w_n(omega); the relation between the subscript n and the functional A_n should be stated explicitly.","section":"Sec. 2.3, Eq. (16)"},{"comment":"Ref. [8] is incomplete ('JHEP 04 (2004) .') and Ref. [15] is a preprint 'In preparation'; these entries should be completed where possible.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"This is a well-written proceedings contribution whose LMA methodological result is solid and quantitatively convincing. The main reservation is that the new physics claim at sigma = 250 MeV rests on statistical errors alone, with no HLT systematic validation at that width and no continuum extrapolation. I believe the manuscript can be made acceptable by adding an HLT stability analysis (for example a tau_max and lambda scan, or a synthetic-data test at sigma = 250 MeV) or by explicitly restricting the claims to 'statistical precision' and marking the rho-resolution statement as preliminary. The spelling of 'Tantalo' in the title and abstract should also be corrected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The LMA part is solid and genuinely useful; the sigma=250 MeV smeared R-ratio is an interesting preliminary, but the systematic control of HLT at that width is not yet demonstrated.\n\nThe new thing here is the combination of Low Mode Averaging with the HLT spectral reconstruction for the light vector-vector correlator, plus the quantitative LMA gain: a factor of about 3.6 in statistical error at essentially fixed cost, a net speedup of about 5. That is a real, useful result for lattice efforts on the hadronic vacuum polarization. The deflation estimator in Eq. (13) is standard and correctly presented, the gain study (Neig scan, stochastic source saturation) is careful, and the comparison between TM and OS regularizations is a good cross-check.\n\nThe main soft spot is exactly what the stress-test flags: HLT at sigma=250 MeV is not yet under systematic control. Ref [1] validated the method for widths 440-630 MeV; here it is applied at 250 MeV without a tau_max scan, a lambda stability study, or a synthetic-data test at that width. The TM/OS agreement is useful but both discretizations share the same HLT systematic. The quoted 11% to 3% error reduction is purely statistical, so the statement that the rho resonance is \"resolved\" is stronger than what is actually demonstrated. Since this is a preliminary, blinded proceedings paper with details deferred to Ref [15], this is not a fatal flaw; it is simply the condition that has to be met before the physics claim can be taken as final.\n\nOne smaller point: the title and abstract say \"R-ratio in isospin symmetric QCD,\" but the plotted quantity is the connected light-quark contribution. That should be made explicit in the final version.\n\nOverall: the LMA methodology is the most solid part, and the prospect of a 250 MeV smeared R-ratio is genuinely interesting. A serious referee should engage with it, mainly to push for the missing HLT systematics and for unblinded, continuum-extrapolated results. I would cite this for the LMA gain numbers, and I would not object to sending it to peer review as a proceedings or short paper.","headline":"The LMA gain is solid and useful; the sigma=250 MeV smeared R-ratio is interesting but the HLT systematic control at that width is not yet demonstrated.","tokens_in":10486,"tokens_out":3228,"would_cite":true,"duration_ms":29133,"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":"Combining Low Mode Averaging with the HLT spectral-density reconstruction reduces the statistical error of the light vector-vector correlator by a factor of about 3.6 and resolves the rho resonance in the smeared R-ratio at a Gaussian…","keywords":["lattice QCD","smeared R-ratio","Low Mode Averaging","spectral density reconstruction","vector-vector correlator","rho resonance","muon g-2","isospin symmetric QCD"],"falsifier":"Run the HLT analysis at $\\sigma = 250$ MeV with several truncation values $\\tau_{\\rm max}$ and with 400 versus 500 eigenvectors: if the reconstructed R-ratio at $E \\sim 770$ MeV moves by more than the quoted roughly 3% statistical error when $\\tau_{\\rm max}$ is increased, or if the two lattice regularizations no longer agree after unblinding, then the systematic control claim for the narrower width would be disproven.","tokens_in":9342,"feed_emoji":"⚛️","tokens_out":9383,"duration_ms":77652,"temperature":0.7,"pith_summary":"This paper is a methods demonstration in lattice QCD: it combines Low Mode Averaging (LMA) with the HLT spectral-density reconstruction to compute the light-quark connected contribution to the smeared R-ratio in isospin-symmetric QCD. The authors report that LMA reduces the statistical error of the vector-vector correlator by a factor of about 3.6 while increasing the cost by only about 2.7, giving a net speedup of about 5 to reach a fixed accuracy. With that gain they are able to reduce the Gaussian smearing width to about 250 MeV, roughly half the width of previous lattice extractions. At a center-of-mass energy near 770 MeV the relative error on the smeared R-ratio falls from about 11% to about 3%, which makes the rho resonance visible directly in the lattice data. The results are preliminary and blinded, but they establish a practical route to finer energy resolution in first-principles determinations of the R-ratio.","feed_headline":"Lattice noise cut 3.6x reveals rho resonance in R-ratio","feed_subtitle":"Low Mode Averaging turns an 11% error into 3% near 770 MeV, exposing the rho resonance.","key_machinery":"The machinery has two parts. Low Mode Averaging (LMA) deflates the Hermitian Dirac operator $Q_W = \\gamma_5 D_W$: its $N_v$ lowest eigenpairs are computed exactly and used to build an all-to-all infrared propagator, while the ultraviolet part is still estimated stochastically. Because the stochastic sources are time- and spin-diluted, the deflated and non-deflated two-point correlators differ only in their infrared part, so the exact infrared contribution can be added back without touching the ultraviolet noise. The second part is the HLT spectral-density reconstruction, which approximates the target smearing kernel by a short exponential sum $\\sum_\\tau g_\\tau e^{-a\\omega\\tau}$, fixing the coefficients $g_\\tau$ by minimizing a weighted $L^2$ distance to the desired Gaussian kernel subject to a covariance penalty. The correlator $C(t) = (12\\pi^2)^{-1}\\int d\\omega\\, e^{-\\omega t}\\,\\omega^2 R(\\omega)$ then supplies the smeared R-ratio. The gain in signal-to-noise is quantified by comparing deflated and non-deflated correlators with adjusted numbers of configurations and stochastic sources.","core_discovery":"The central claim is that deflating the low eigenmodes of the Dirac operator does not merely reduce noise in the correlator; it changes which physics can be extracted from the same lattice ensembles. In the mixed-action twisted-mass setup, the deflated two-point function is obtained from the stochastic correlator by removing the stochastically estimated infrared part and adding back the exact all-to-all infrared part. With around 400 eigenvectors and about 1000 time-diluted stochastic sources per configuration, the signal-to-noise ratio on the light vector-vector correlator improves by a factor of 3.63(15) and 3.57(16) for the two current regularizations, at a cost increase of 2.7(2), i.e. a net speedup of about 5. Feeding these correlators into the HLT spectral-density reconstruction yields the connected $u/d$ contribution to the Gaussian-smeared R-ratio $R^\\ell_\\sigma(E)$ with $\\sigma = 250$ MeV, where the relative error at $E \\sim 770$ MeV drops from roughly 11% to 3%, exposing the rho resonance. The same pipeline is applied at three lattice spacings (about 0.057, 0.068 and 0.080 fm) and the results from the two regularizations are mutually compatible, so the authors regard the rho signal as a genuine physics outcome rather than a discretization artifact.","pith_inferences":["Going beyond the paper: if the HLT truncation systematics are confirmed at 250 MeV, the same LMA plus HLT combination could plausibly reach widths near 150-200 MeV, where the omega and phi resonances, and possibly excited vector states, would come into view rather than just the rho.","The gain factor is expected to grow on larger physical volumes because the number of low Dirac modes scales with the spacetime volume, so on boxes larger than the 5.1-5.5 fm used here the statistical advantage of LMA should become even more pronounced.","Because LMA removes only the exactly-deflated infrared noise, its benefit should transfer to other bilinear channels, such as axial or scalar correlators, using the same eigenvector set and at no extra inversion cost; testing this on the same ensembles would be a direct extension.","A practical consequence for the muon g-2 program is that cutting the statistical error on the connected light-quark contribution by about 3.6 times at fixed cost would allow the disconnected and strange or charm contributions to be computed at comparable precision, which often dominate the error budget."],"forward_implications":["At $E \\sim 770$ MeV and $\\sigma = 250$ MeV, the relative error on the smeared R-ratio drops from about 11% to about 3%, making the rho resonance visible in the lattice data.","The signal-to-noise gain is 3.63(15) for one current regularization and 3.57(16) for the other; with a computational cost increase of about 2.7(2), the net speedup to reach a given accuracy is about 5(1).","The method is demonstrated at three lattice spacings (about 0.057, 0.068 and 0.080 fm) with volumes up to about 5.5 fm, so the smeared R-ratio can be computed with controlled discretization effects.","Results from the two lattice regularizations are compatible within the reached precision, allowing low-energy lattice artifacts specific to each regularization to be identified.","Combining LMA with HLT makes Gaussian widths of 250 MeV feasible, roughly half the previously accessible width, enabling a more direct comparison with phenomenological smeared R-ratio data."],"supporting_citations":[{"why":"Prior first-principles lattice computation of the smeared R-ratio at widths 440-630 MeV; the baseline result this paper extends toward 250 MeV and the source of the HLT analysis setup.","marker":"[1]"},{"why":"Introduces the HLT spectral-density reconstruction method used to extract the smeared R-ratio from Euclidean correlators.","marker":"[2]"},{"why":"Experimental R-ratio compilation whose Gaussian smearing provides the phenomenological comparison target.","marker":"[3]"},{"why":"Shows that low fermionic eigenmodes dominate long-distance correlators, the principle LMA exploits.","marker":"[7]"},{"why":"Provides the non-deflated vector-vector correlator data used as the no-LMA baseline in the gain comparison.","marker":"[11]"},{"why":"Documents the gauge ensembles at the three lattice spacings used in this study.","marker":"[16]"}],"fun_headline_variants":["Noise cut 3.6x exposes rho peak in smeared R-ratio","Low Mode Averaging reveals rho resonance at 770 MeV","Lattice error drops from 11% to 3%, rho appears","Deflated Dirac modes sharpen R-ratio, expose rho","R-ratio resonance visible after 3.6x noise cut"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the HLT reconstruction stays systematically under control at the narrower Gaussian width of 250 MeV, because the method had previously been validated only for widths of 440 to 630 MeV and this paper does not yet give an explicit error analysis for truncation or excited-state contamination at the new width.","fun_headline_variants_meta":{"raw":{"variants":["Noise cut 3.6x exposes rho peak in smeared R-ratio","Low Mode Averaging reveals rho resonance at 770 MeV","Lattice error drops from 11% to 3%, rho appears","Deflated Dirac modes sharpen R-ratio, expose rho","R-ratio resonance visible after 3.6x noise cut"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000275,"raw_usage":{"total_tokens":1652,"prompt_tokens":963,"completion_tokens":689,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":579,"completion_tokens_details":{"reasoning_tokens":593}},"tokens_in":579,"tokens_out":689,"duration_ms":6122,"temperature":1.0,"reasoning_tokens":593,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T05:35:00.166602+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the HLT analysis at $\\sigma = 250$ MeV with several truncation values $\\tau_{\\rm max}$ and with 400 versus 500 eigenvectors: if the reconstructed R-ratio at $E \\sim 770$ MeV moves by more than the quoted roughly 3% statistical error when $\\tau_{\\rm max}$ is increased, or if the two lattice regularizations no longer agree after unblinding, then the systematic control claim for the narrower width would be disproven.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows that low fermionic eigenmodes dominate long-distance correlators, the principle LMA exploits."},{"cited_title":"Alexandrou, S","cited_arxiv_id":null,"evidence_quote":"Provides the non-deflated vector-vector correlator data used as the no-LMA baseline in the gain comparison."},{"cited_title":"Alexandrou et al.,Status of the ETMC ensemble generation effort, PoS LATTICE2024 (2025) 429","cited_arxiv_id":null,"evidence_quote":"Documents the gauge ensembles at the three lattice spacings used in this study."}],"review_version":1}