{"id":"2e557f13-bcae-473e-bcb0-f1967790cba0","arxiv_id":"2506.16689","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Tikhonov regularization with L-curve parameter choice stabilizes the ill-conditioned limited inverse Fourier transform in LaMET and recovers pion quasi distribution amplitudes consistent with physics-driven lambda-extrapolation.","lead":"This paper applies a standard mathematical remedy, Tikhonov regularization, to the unstable Fourier inversion used in lattice QCD calculations of hadron structure. It reports stable reconstructions of pion distribution amplitudes that agree with an existing physics-guided method.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'moderately tractable' classification is unsupported: the paper's own singular-value ratio (σ1/σ81 ≈ 10^17) is consistent with exponential-in-index decay, not the asserted slower-than-exponential decay, so the central spectral claim needs a quantitative test.","rationale":"The reader's weakest-assumption analysis correctly identifies the unproved singular-value decay rate as the load-bearing support for the paper's 'moderately tractable' classification. I agree with that identification and sharpen it: the paper's own quoted spectrum is not merely unproved but numerically consistent with exponential-in-index decay, which is the standard signature of severe ill-posedness for an integral operator with an analytic kernel. This matters because the abstract's novelty claim is precisely the classification into a distinct, moderately tractable class; if the decay is exponential, that classification is false even though Tikhonov regularization can still produce useful reconstructions. The paper has real independent strengths: the uniqueness proof for continuum limited Fourier data is correct, the toy-model noise studies are well designed, and the consistency with λ-extrapolation on real lattice data is a useful cross-check. However, the central framing and the 'first-principles uncertainty quantification without ansatz-based assumptions' language overreach: the L2 penalty is itself a prior assumption, and the uniqueness theorem applies to data on a continuum interval, not to the finite discrete lattice samples actually inverted. These weaknesses are addressable, so the reader's CONDITIONAL verdict is appropriate. The proposed check on the singular-value spectrum will settle whether the 'moderately tractable' classification can be maintained.","tokens_in":26941,"tokens_out":14277,"duration_ms":168311,"concrete_test":"Compute the singular value spectra of K_re and of the full complex K for the parameters in Eqs. (12)-(13), and for the toy-model grids, using SVD in high precision. Fit log10 σ_i versus index i over all resolvable indices and compare an exponential-in-index model (constant slope) against polynomial decay models using information criteria. If the log-linear slope is approximately -0.49 (as implied by σ1/σ81 ≈ 10^17) and holds over a decade or more, the Sec. II.B statement that the singular values 'decay more slowly than exponentially' is false, and the problem should be reclassified as severely ill-posed or the 'moderately tractable' claim should be removed.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The abstract's central claim is that the limited DFT inversion in LaMET belongs to a class of 'moderately tractable ill-posed problems' with 'distinct spectral properties' separating it from more severely unstable inverse problems. The only quantitative basis for this separation is the unproved assertion in Sec. II.B that the singular values of K_re 'decay more slowly than exponentially.' The paper's own numbers do not support that assertion: with n=81, σ1=0.3545 and σn=3.8×10^-18, so σn/σ1 ≈ 10^-17. A log-linear fit over 80 index steps gives a slope of about -ln(10^17)/80 ≈ -0.49, i.e., the spectrum is consistent with geometric (exponential-in-index) decay. This is exactly the spectral signature of a Fredholm integral operator with an analytic kernel, which inverse problem theory classifies as severely, not mildly, ill-posed. The paper provides no proof, no reference, and no fitted decay law for the 'slower than exponential' claim; the Picard-criterion discussion merely restates the instability rather than establishing the rate. If the decay is exponential, the 'moderately tractable' classification collapses, and the abstract's 'distinct spectral properties' claim is contradicted by the paper's own data. Tikhonov regularization may still stabilize the inversion, but the central framing and novelty claim of the paper would then be unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes the limited discrete Fourier inversion problem that arises in the large-momentum effective theory (LaMET) when reconstructing momentum-space partonic distributions from finite-range, discrete coordinate-space lattice data. It shows that the continuous inverse Fourier transform on a bounded support is unique (with a proof in Appendix A), but that the discretized inversion is numerically unstable, as quantified by a singular value condition number of order 10^17. The authors propose Tikhonov regularization with the L-curve method for selecting the regularization parameter, demonstrate its effectiveness on toy models with uncorrelated and correlated noise, and apply it to lattice QCD data for the pion quasi distribution amplitude from the Lattice Parton Collaboration. The reconstructed quasi-DA is reported to be consistent with the previously used lambda-extrapolation method. The paper claims that the limited Fourier inversion in LaMET falls into a class of 'moderately tractable' ill-posed problems, distinguished by slow (slower than exponential) singular value decay.","tokens_in":27206,"tokens_out":9271,"duration_ms":88722,"significance":"If the claims are correct, the paper would provide a mathematically principled alternative to physics-driven extrapolation for partonic distributions in LaMET, with the potential to quantify uncertainties without model assumptions. The paper contains a self-contained uniqueness proof, a clean toy-model validation with a known ground truth, and a real-data application. However, the central classification of the problem as 'moderately tractable' is not substantiated: the reported singular value ratio is fully consistent with exponential (geometric) decay, which is typically associated with severely ill-posed problems. The uncertainty quantification is also less first-principles than claimed, since the L-curve is a heuristic and the bootstrap procedure does not propagate the regularization parameter choice. The core demonstration, that Tikhonov regularization stabilizes the inversion, is plausible and valuable, but the overreaching framing and the absence of a spectral analysis prevent acceptance in the current form.","major_comments":[{"comment":"The statement that the singular values of K_re 'decay more slowly than exponentially' is load-bearing for the paper's classification of the problem as 'moderately tractable,' but it is not substantiated. The paper gives sigma_1 = 0.3545 and sigma_n = 3.8e-18 for n = 81, yielding sigma_n/sigma_1 ~ 1e-17. Over the 80 index steps, this ratio is perfectly consistent with geometric decay sigma_i ~ r^i with r ~ 0.61, i.e., exponential decay in the index. Algebraic (slower-than-exponential) decay would give a far milder condition number. Please provide a quantitative spectral analysis (e.g., the full singular value spectrum on a log scale, or an asymptotic bound) or remove the classification claim from the abstract and Sec. II.B.","section":"Sec. II.B (after Eq. (14))"},{"comment":"The uniqueness theorem is proved for data given on a continuum interval [lambda_min, lambda_max], whereas the practical inversion uses finitely many discrete values of lambda. For finite discrete data, the inverse problem is generally underdetermined and uniqueness does not hold as stated. The manuscript should explicitly state that the existence/uniqueness results apply to the continuous idealization, and should discuss the relation between this idealization and the finite-dimensional discrete problem actually solved by the regularization.","section":"Sec. II.B and Appendix A"},{"comment":"The claim of 'first-principles uncertainty quantification' is not supported by the described methodology. The regularization parameter alpha is selected per bootstrap sample via the L-curve criterion, and the reported alpha values span three orders of magnitude (10^-6 to 10^-4). The final uncertainty band shown in Fig. 12 does not appear to include the systematic component from the ambiguity in alpha selection. Please describe how the uncertainty band is constructed, and provide a sensitivity analysis of the final profile to the alpha selection rule (e.g., by comparing L-curve-selected alpha with other criteria such as the discrepancy principle).","section":"Sec. IV, Eq. (27) and Fig. 9"},{"comment":"The agreement between the Tikhonov reconstruction and the lambda-extrapolation method is presented as confirmation of reliability and even of 'well-posedness under optimal conditions.' However, both methods use the same lattice data and share overlapping authorship/collaboration (LPC); the comparison is a consistency check rather than an independent validation. The text should be moderated and should state this limitation explicitly.","section":"Sec. IV, Fig. 12 and text"}],"minor_comments":[{"comment":"The phrase 'The reconstructed solutions is consistent' should read 'The reconstructed solutions are consistent.'","section":"Abstract"},{"comment":"Please define X_nx and Lambda_nlambda explicitly as vectors and clarify the matrix construction; the current notation using sets and an outer product is ambiguous.","section":"Eqs. (7)-(8)"},{"comment":"The approximation (sigma_1 + alpha)/(sigma_n + alpha) ~ sigma_1/alpha requires sigma_n << alpha << sigma_1; please state this condition or revise the heuristic.","section":"Eq. (18)"},{"comment":"There are several typos: 'maskes' should be 'masks', 'achiev' should be 'achieves', and 'the uncertainty in lattice data do not grow linearly' should read 'the uncertainties in lattice data do not grow linearly.'","section":"Sec. III and Sec. V"},{"comment":"The proof is correct but uses nonstandard notation (C[x_min,x_max]); consider referring to C([x_min,x_max]) and citing a standard result for the density of polynomials in L2.","section":"Appendix A"},{"comment":"The statement that the inverse problem satisfies 'existence' is not fully established; the Paley-Wiener argument applies to the true solution, not to arbitrary data. Please clarify the meaning of existence in the Hadamard sense.","section":"Sec. I and Sec. II.B"}],"recommendation":"major_revision","confidential_remarks":"The main concern is the unsubstantiated spectral classification, which is central to the abstract's novelty claim. I recommend asking the authors to provide a rigorous singular value analysis or to remove the 'moderately tractable' claim. There is also a heavy reliance on the authors' own collaborative work (Refs. [112], [163], [175]), and the lambda-extrapolation comparison is not independent. The paper would be acceptable after these points are addressed, assuming the spectral question is resolved in the authors' favor or the claim is removed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a competent application of textbook inverse-problem machinery to a real lattice QCD pain point, and the demonstration on the LPC pion quasi-DA data is genuinely useful. The uniqueness proof in Appendix A is correct, the toy-model noiseless inversion is a valid sanity check, and the raw-SVD versus Tikhonov comparison makes the instability vivid. If I worked on LaMET extractions, I would want this in the literature.\n\nThe main soft spot is exactly what the stress-test flags: the paper's central classification of the LaMET DFT inversion as 'moderately tractable' rests on the claim in Sec. II.B that the singular values decay more slowly than exponentially, and nothing in the paper establishes that. The numbers shown, sigma_1/sigma_81 ~ 1e17 over 80 indices, are consistent with geometric (exponential-in-index) decay. The distinction from 'more severely unstable' spectral-function problems is therefore not supported by the evidence presented. That does not kill the paper—Tikhonov works for exponentially ill-posed problems too—but it does mean the abstract overclaims. The fix is straightforward: fit the decay law, report log-linear plots, or state the classification more modestly.\n\nOther soft spots are minor in comparison. The instability analysis uses only the real part of the DFT matrix; fine as an illustration, but the actual inversion problem is complex, and the SVD of Re(K) may not characterize the full operator. The L-curve is a heuristic; calling the result 'first-principles uncertainty quantification without ansatz-based assumptions' overstates what a penalty-based regularizer does. No code or data are shipped, and the validation against lambda-extrapolation uses overlapping authors and data, so the cross-check is less independent than it looks. The reference list is broad and fits the topic; the self-citations point to the authors' own prior inverse-problem work and to the LPC data sources, which are the relevant places to cite.\n\nBottom line: a solid, useful paper with a load-bearing spectral claim that needs either proof or softening. Worth a serious referee, and I would send it back for a revision that addresses the decay-rate assertion. It should be in the literature because it gives practitioners a working recipe and a concrete baseline, but not with the current 'moderately tractable / distinct spectral properties' framing.","headline":"Useful, workmanlike application of Tikhonov regularization to LaMET quasi-DA inversion, but the 'moderately tractable' spectral classification is asserted rather than shown.","tokens_in":27786,"tokens_out":2518,"would_cite":false,"duration_ms":27371,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65F22","65R30"],"pacs":["12.38.Gc"],"model":"deepseek-v4-flash","headline":"This paper claims that the limited inverse discrete Fourier transform in the LaMET framework is a moderately ill-posed inverse problem—solutions exist and are unique, but stability fails—and that Tikhonov regularization with L-curve…","keywords":["inverse problems","Tikhonov regularization","L-curve method","limited discrete Fourier transform","large-momentum effective theory","quasi distribution amplitude","lattice QCD","ill-posedness"],"falsifier":"Compute the singular values $\\sigma_i$ of the real DFT matrix $K_{\\mathrm{re}}$ for the paper's parameters ($x \\in [-2,3]$, $\\delta x = 0.01$; $\\lambda \\in [-20,20]$, $\\delta \\lambda = 0.5$) and plot $\\log \\sigma_i$ against the index $i$. A straight-line (linear) decay in this log-linear plot means geometric, i.e., exponential-in-index, decay—consistent with $\\sigma_1/\\sigma_{81} \\approx 10^{17}$ giving a per-index ratio near $(10^{17})^{-1/80}\\approx 0.61$—which directly contradicts the claim of slower-than-exponential decay. The classification of the problem as moderate rather than severe would then need to be revised.","tokens_in":26697,"feed_emoji":"🧮","tokens_out":11469,"duration_ms":96628,"temperature":0.7,"pith_summary":"This paper asks whether the limited, noisy coordinate-space data available from lattice QCD can be inverted to recover a parton distribution inside the large-momentum effective theory (LaMET). It argues that the inversion is ill-posed in the strict Hadamard sense—existence and uniqueness hold, but tiny input noise is amplified by a factor of about $10^{17}$—yet that this ill-posedness is mild enough to be tamed by Tikhonov regularization. The paper demonstrates the method on synthetic toy models with both uncorrelated and correlated noise, and then on real lattice QCD data for the pion quasi distribution amplitude, where the regularized result agrees with the physics-driven $\\lambda$-extrapolation method. If the claims hold, LaMET practitioners can invert limited Fourier data with first-principles uncertainty quantification, without fitting ansatz functions.","feed_headline":"Tikhonov tames the ill-posed Fourier step in LaMET","feed_subtitle":"The inversion is unstable at factor ~10^17, but L-curve regularization restores it and matches λ-extrapolation.","key_machinery":"The load-bearing object is the discrete Fourier matrix $K$ with entries $\\exp(i x \\lambda)$ on a finite grid, whose singular value decomposition $K = U \\Sigma V^T$ exposes the instability: the formal solution $f = \\sum_i (u_i^T g / \\sigma_i) v_i$ divides by singular values as small as $10^{-18}$. Tikhonov regularization replaces that division with the normal equation $(K^\\dagger K + \\alpha I) f^\\delta_\\alpha = K^\\dagger g^\\delta$, adding $\\alpha > 0$ to each singular value so that the noise-amplifying factors $\\sigma_i^{-1}$ are controlled; the L-curve method fixes $\\alpha$ by maximizing the curvature of the $\\log\\|K f - g^\\delta\\|^2$ versus $\\log\\|f\\|^2$ tradeoff. Uniqueness of the limited Fourier transform is proved in the appendix by analytic continuation together with the Weierstrass approximation theorem.","core_discovery":"The central claim is that the limited inverse discrete Fourier transform underlying LaMET is an ill-posed inverse problem of moderate severity. It satisfies the first two Hadamard criteria: the Paley-Wiener theorem gives existence for compactly supported parton distributions, and an appendix proves uniqueness on $L^2$ of a finite interval. Stability fails because the singular values of the DFT matrix decay to about $10^{-18}$, giving a condition number of order $10^{17}$ that amplifies input errors massively. Tikhonov regularization, with the regularization parameter chosen by the L-curve criterion, is shown to turn this unstable inversion into a stable optimization problem; regularized solutions reproduce the true profile in toy models and produce a pion quasi distribution amplitude from real lattice data consistent with $\\lambda$-extrapolation. The paper additionally claims the singular spectrum decays more slowly than exponentially, placing this problem in a moderately tractable class distinct from severely ill-posed problems such as spectral-function reconstruction from Euclidean correlators.","pith_inferences":["The 'slower than exponential' singular-value decay that underpins the moderate-tractability classification is not demonstrated quantitatively; the quoted ratio $\\sigma_1/\\sigma_{81} \\approx 10^{17}$ is compatible with geometric decay in the index, which would put the problem in the severe class even though Tikhonov regularization might still work.","The same L-curve-regularized inversion could be applied as a cross-check to other LaMET observables (PDFs, GPDs, TMDs) that also provide only limited, noisy $g(\\lambda)$ data.","The agreement between Tikhonov and $\\lambda$-extrapolation could be turned into a quantitative diagnostic: plotting the difference between the two reconstructions as a function of $\\lambda_{\\max}$ and data precision would reveal where regularization bias or extrapolation systematic errors dominate.","Because the penalty term is $\\|f\\|^2_{\\ell^2}$, the method favours smooth solutions and may smooth away genuine endpoint or kink structures; combining it with operator-product-expansion endpoint constraints is a natural next step."],"forward_implications":["The LaMET Fourier inversion satisfies existence and uniqueness, so the only obstruction is stability, and Tikhonov regularization restores well-posedness.","L-curve-selected Tikhonov regularization provides quantified uncertainties for the pion quasi distribution amplitude without any ansatz-based functional form.","The regularized reconstruction from real lattice data is statistically consistent with the $\\lambda$-extrapolation result, cross-validating the two independent methods.","Classifying the problem as moderately tractable separates it from severely ill-posed lattice QCD inversions (such as spectral-function reconstruction), indicating that standard regularization tools suffice here.","As the input error approaches zero, the regularized solution converges to the true distribution, giving a formal guarantee behind the numerical reconstructions."],"supporting_citations":[{"why":"Supplies the real lattice QCD bootstrap ensembles for the pion quasi DA used in the real-data application.","marker":"[112]"},{"why":"Raises the concern that LaMET inherits ill-posedness, which the paper addresses by classifying the severity of the Fourier inversion.","marker":"[170]"},{"why":"Provides the physics-driven $\\lambda$-extrapolation method whose pion quasi DA result is compared with the Tikhonov reconstruction.","marker":"[163]"},{"why":"Sets the Hadamard well-posedness criteria and the mathematical theory of inverse problems used to frame the analysis.","marker":"[181]"},{"why":"Supplies the standard classification of ill-posed problems by singular-value decay that underlies the moderate-versus-severe distinction.","marker":"[182]"},{"why":"Gives the Tikhonov regularization and L-curve methodology plus the convergence guarantee used for the stable inversion.","marker":"[188]"},{"why":"Defines the L-curve criterion for selecting the regularization parameter $\\alpha$.","marker":"[193]"},{"why":"Establishes via the Paley-Wiener theorem the existence condition for compactly supported parton distributions.","marker":"[190]"}],"fun_headline_variants":["Tikhonov fixes unstable Fourier inversion in LaMET","Tikhonov tames exponential instability in LaMET","LaMET's Fourier inversion stabilized by Tikhonov","Regularizing the unstable DFT: a LaMET solution","Stability restored in LaMET via Tikhonov regularization"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The classification of the LaMET inversion as 'moderately tractable' rests on the unproven assertion in Sec. II.B that the DFT matrix singular values decay more slowly than exponentially; the paper's own quoted spectrum ($\\sigma_1/\\sigma_{81} \\approx 10^{17}$) is consistent with exponential decay, so if the decay is actually exponential or faster, the problem would belong to the more severe class and the claimed distinction would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Tikhonov fixes unstable Fourier inversion in LaMET","Tikhonov tames exponential instability in LaMET","LaMET's Fourier inversion stabilized by Tikhonov","Regularizing the unstable DFT: a LaMET solution","Stability restored in LaMET via Tikhonov regularization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000923,"raw_usage":{"total_tokens":3946,"prompt_tokens":925,"completion_tokens":3021,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":541,"completion_tokens_details":{"reasoning_tokens":2942}},"tokens_in":541,"tokens_out":3021,"duration_ms":22445,"temperature":1.0,"reasoning_tokens":2942,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:21:34.161748+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the singular values $\\sigma_i$ of the real DFT matrix $K_{\\mathrm{re}}$ for the paper's parameters ($x \\in [-2,3]$, $\\delta x = 0.01$; $\\lambda \\in [-20,20]$, $\\delta \\lambda = 0.5$) and plot $\\log \\sigma_i$ against the index $i$. A straight-line (linear) decay in this log-linear plot means geometric, i.e., exponential-in-index, decay—consistent with $\\sigma_1/\\sigma_{81} \\approx 10^{17}$ giving a per-index ratio near $(10^{17})^{-1/80}\\approx 0.61$—which directly contradicts the claim of slower-than-exponential decay. The classification of the problem as moderate rather than severe would then need to be revised.","supporting_citations":[],"review_version":2}