{"id":"c0a1c536-cfc3-43b0-b11b-589ee33e08ec","arxiv_id":"2507.00284","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"An effective-field-theory forward model reproduces the Lyman-alpha forest field from a hydrodynamic simulation at percent level, down to a few megaparsecs, using the same initial conditions.","lead":"This paper builds an analytic model that generates the Lyman-alpha forest, the absorption pattern in quasar spectra, from the same initial conditions used in hydrodynamic simulations, and shows it matches a simulation's flux maps at the few-percent level. If the model generalizes, cosmologists could analyze DESI quasar data at the field level instead of using lossy summary statistics, potentially extracting more cosmological information.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Transfer-function universality is asserted but not tested: all validations are in-sample on the same Sherwood snapshot, so the field-level predictive claim for other realizations/redshifts is unestablished.","rationale":"I read the paper in good faith and find the in-sample field-level agreement genuinely impressive: matching amplitudes and phases of Fourier modes using a physically motivated EFT expansion is a nontrivial success. The model clearly outperforms the linear-bias forward model, and the error power spectrum exhibits the expected white-noise behavior on large scales with a controlled rise near k ~ 0.6 h/Mpc. However, the paper's central scientific promise—an analytic forward model applicable to 'arbitrarily large volumes' and enabling field-level inference from DESI—rests on the claim that the calibrated transfer functions are universal, i.e., that they remain valid for other initial-condition realizations without re-fitting. The paper does not test this on any independent realization or simulation; all comparisons use the same snapshot used for calibration. This is precisely the reader's weakest_assumption, and I agree that it is the most load-bearing gap. The problem is not that EFT universality is false in principle—it is a standard and well-motivated assumption in galaxy clustering—but that the transfer functions here are fitted from a single simulation, and the required precision (percent-level, down to a few Mpc) is high enough that an empirical check is necessary. The paper provides no evidence that the fitted functions are smooth, stable, or transferable; it does not even show the transfer functions themselves or their estimated uncertainties. The 'count-in-cell' statistics claim is also not directly demonstrated in the main text (only smoothed PDF moments in the Supplement), but that is secondary to the universality issue. My recommendation is therefore to maintain the CONDITIONAL verdict: the method is promising and the in-sample results are strong, but the headline claim of field-level predictability for DESI requires the proposed out-of-sample test before it can be accepted as established. I do not see grounds for REJECT, since the in-sample phase matching is a legitimate and demanding test; nor do I see grounds for ACCEPT, given that the universality assumption is untested. UNCHANGED reflects my agreement that the reader's CONDITIONAL verdict is the right one.","tokens_in":19362,"tokens_out":8784,"duration_ms":92257,"concrete_test":"Calibrate the transfer functions exactly as in the paper on the Sherwood 160^3 (h^-1 Mpc)^3 z=2.8 snapshot. Then, without any re-fitting, apply them to (a) a different hydrodynamical simulation with the same cosmology and redshift but a different initial-condition seed (e.g., another Sherwood realization or a comparable simulation like ACCEL2 at z=2.8), and (b) a Sherwood snapshot at a different redshift (e.g., z=2.2 or z=3.4). Compute the out-of-sample error power spectrum Perr,out and cross-correlation coefficient r_cc between the predicted and true flux fields. If Perr,out is substantially larger than the in-sample Perr from Fig. 1 (e.g., exceeds 5% of the true power at k <= 0.6 h/Mpc) or r_cc drops significantly, the universality assumption fails and the headline claim is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the EFT forward model, once its transfer functions are calibrated, predicts the Ly-alpha field for arbitrary initial-condition realizations at percent-level accuracy, enabling field-level inference from DESI. The Forward model section states: 'Once the transfer functions are calibrated, they can be applied to other realizations... The universality of EFT guarantees that the large-scale distribution of the resulting Ly-alpha field will be indistinguishable from a full hydrodynamical simulation run with the same initial field realization.' This universality is the load-bearing bridge from the Sherwood fit to a predictive tool. However, every validation in the paper (Figs. 1-3, S1-S3, Tables S1-S2) is performed on the same snapshot used to calibrate the transfer functions and noise parameters: the transfer functions are estimated by cross-correlating the simulated flux field with the initial density field of that same realization, and the reported Perr, power spectra, PDFs, and moments are all evaluated on that same realization. No hold-out realization, independent simulation, or different redshift is tested. The 'universality of EFT' is a plausible principle, but the transfer functions are non-perturbative fitted objects; whether they remain valid at the claimed percent level on new initial conditions is an empirical question. The in-sample agreement could partly reflect overfitting to the specific realization's phases and noise, especially near k ~ 0.6 h/Mpc where Perr starts to grow. Additionally, the paper calibrates only at z=2.8, while DESI observes a redshift range, and no argument or test shows how the transfer functions evolve. Without an out-of-sample test, the scientific claim that this model 'enables field-level inference from Ly-alpha forest data' is not established. The count-in-cell claim is also not directly demonstrated in the main text, but the universality gap is the more central issue.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This Letter presents an analytic, perturbative forward model for the cosmological Lyman-α forest flux-decrement field, based on an EFT bias expansion with line-of-sight operators and k- and μ-dependent transfer functions calibrated against the Sherwood hydrodynamic simulation. The model is evaluated on the same Sherwood snapshot at z=2.8 used for calibration; the authors compare power spectra, one-point PDFs and moments, and Lyα–halo cross-spectra, reporting percent-level agreement down to k ≈ 0.6 h/Mpc and claiming accurate count-in-cell statistics. They argue that the calibrated transfer functions are universal and can be applied to other initial-condition realizations, enabling field-level inference for DESI.","tokens_in":19772,"tokens_out":5105,"duration_ms":58003,"significance":"The paper addresses an important problem: current Lyα analyses use only two-point statistics, and a validated field-level forward model would enable optimal inference and simulation-based priors. The analytic construction is novel and the in-sample agreement is impressive; the paper also provides a first quantification of the Lyα stochastic noise power spectrum. However, because the transfer functions and noise parameters are fitted to the same snapshot used for validation, the reported accuracy is partly by construction, and the universality that would make the model predictive is asserted rather than demonstrated. With an out-of-sample test, this would be a strong contribution; without one, the central predictive claim is not yet established.","major_comments":[{"comment":"The central predictive claim is not tested out of sample. In Eq. (5) the linear transfer function is defined as the cross-spectrum of the target field with the initial density of the same realization, and the full set of transfer functions in Eq. (10) is chosen to minimize the same-snapshot Perr; the noise parameters in Eq. (8) are then fitted to that residual. All validation figures (Figs. 1–3 and S1–S3, Tables S1–S2) use the same Sherwood snapshot. The statement that 'the universality of EFT guarantees' applicability to other realizations is an assertion, not a proof, especially since these transfer functions are non-perturbative fitted objects. The authors should demonstrate transfer-function universality on at least one independent realization, a different redshift, or a different simulation before claiming predictive field-level modeling. This is load-bearing for the abstract's claim and for the proposed DESI applications.","section":"Forward model (Eqs. 5 and 10) and Results"},{"comment":"The reported noise parameters n0, α1, α2 are extracted from the residual of the calibration snapshot. Because the transfer functions were chosen to minimize Perr on that same snapshot, the residual is the minimum achievable in-sample error; it is not independent evidence about the true stochasticity of the Lyα field. Calling these the 'first estimates of stochasticity' is therefore premature. A split-sample or cross-realization check is needed to determine whether the fitted noise level is physical or partly an artifact of overfitting to the phases of one realization.","section":"Model error power spectrum (Eq. 8)"},{"comment":"The abstract and Summary claim that the model reproduces 'count-in-cell statistics at the percent level,' but the only one-point statistics shown are PDFs and moments (variance, skewness, kurtosis) of the flux field smoothed with Gaussian kernels (Fig. S1, Table S1). No count-in-cell variance or related discrete-cell statistic is presented. Either provide the actual count-in-cell measurements or revise the claim to state that the model reproduces one-point statistics of the smoothed field.","section":"Abstract, Summary, and Supplemental Table S1"}],"minor_comments":[{"comment":"The text says 'refereed to as the error power spectrum'; 'refereed' should be 'referred'.","section":"Model error power spectrum"},{"comment":"The definition δF = F/F(z) − 1 uses F(z) for both the mean transmission and the field; please use an overbar or a different symbol for the mean to avoid ambiguity.","section":"Forward model"},{"comment":"Measured power spectra are shown without error bars; for a single realization, please include sample-variance estimates or state explicitly why they are omitted.","section":"Figures 1 and 3"},{"comment":"The terminology 'count-in-cell' vs 'counts-in-cells' is inconsistent between the abstract and the Summary; standardize after revising the relevant claim.","section":"Summary"}],"recommendation":"major_revision","confidential_remarks":"The in-sample validation is the main issue; I would suggest asking for at least one out-of-sample test, such as an independent Sherwood realization, a different redshift, or a different hydrodynamic simulation. If such a test is provided and passes, the paper could be acceptable. The count-in-cell claim should also be reconciled with the actual statistics shown."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is the first analytic EFT forward model for the Ly-alpha forest at the field level, and it works impressively in-sample. The paper deserves careful peer review, but the headline claim that it enables field-level inference from DESI rests on a universality assumption that is not yet tested.\n\nWhat's new: they adapt the field-level EFT framework developed for galaxies and HI to the Ly-alpha flux field, with the line-of-sight operators (eta, delta*eta, eta^2, etc.) that the problem's symmetries demand, plus IR-resummed Zel'dovich shifts. The linear model fails badly; the cubic-order model with transfer functions reproduces the Sherwood power spectrum at percent level to k~0.6 h/Mpc, and the Perr is small and roughly white on large scales. They also get the Ly-alpha-halo cross-correlation right. The field-level comparison of phases is genuinely more stringent than matching summary statistics, and the residual maps are informative.\n\nThe caveat, as the stress-test note says, is that every validation is on the same snapshot used to calibrate the transfer functions and noise parameters. The transfer functions are cross-spectrum fits to that realization; the noise parameters are fitted to the residuals of the same simulation. So the percent-level match is partly by construction. The paper explicitly asserts that once calibrated, the transfer functions apply to other realizations, with EFT universality as the justification. That may well be true—the same logic works for galaxies—but it is an empirical claim, and the paper does not test it on a different simulation, redshift, cosmology, or even a different realization of the same box. That is the load-bearing step for any DESI application.\n\nMinor points: the count-in-cell claim in the abstract is supported only indirectly by moments of smoothed fields; I don't see a direct counts-in-cells comparison. No code or data are released, which limits reproducibility. The noise interpretation (inherited from halos) is suggestive but not fully nailed down.\n\nBottom line: the technical machinery is solid and the in-sample demonstration is valuable as a proof of concept. The authors are credible and the citation pattern is appropriate. For a referee, I'd ask for an out-of-sample test on a second Sherwood snapshot or another hydro simulation before accepting the field-level predictive claim. Still, this is exactly the kind of paper that deserves a serious referee rather than a desk reject.","headline":"A well-executed first field-level EFT forward model for the Ly-alpha forest, with percent-level in-sample accuracy; the missing out-of-sample test is the main thing separating a proof-of-concept from a predictive tool.","tokens_in":20429,"tokens_out":2677,"would_cite":true,"duration_ms":28478,"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":"An analytic, perturbative forward model predicts the Lyman-alpha forest flux fluctuation field from cosmological initial conditions, and reproduces simulated flux statistics at the percent level down to scales of a few megaparsecs.","keywords":["Lyman-alpha forest","field-level inference","effective field theory","forward model","intergalactic medium","large-scale structure","transmitted flux fluctuations","stochastic noise power spectrum"],"falsifier":"Apply the calibrated transfer functions to a second hydrodynamic simulation with different initial conditions or to a different redshift snapshot of the same simulation suite, and measure the per-mode error spectrum $P_{\\rm err}(k,\\mu)$; if the flat percent-level error spectrum from the calibration snapshot is not recovered, the claimed universality is falsified.","tokens_in":19164,"feed_emoji":"🔭","tokens_out":11275,"duration_ms":111581,"temperature":0.7,"pith_summary":"Cosmological analyses of the Lyman-alpha forest currently compress quasar spectra into two-point statistics, throwing away phase information. This paper develops an analytic, perturbative forward model that predicts the whole transmitted-flux fluctuation field from a specified set of initial conditions, so that every Fourier mode's amplitude and phase can be compared. Calibrated on one hydrodynamic simulation snapshot and evaluated on the same initial conditions, the model reproduces the flux power spectrum, the flux-halo cross-spectrum, and count-in-cell statistics at the percent level down to scales of a few megaparsecs. The payoff is that Lyman-alpha data from current and planned surveys could be analyzed at the field level, with simulation-based priors, rather than through lossy summary statistics.","feed_headline":"Lyman-alpha forest reproduced at percent level by analytic model","feed_subtitle":"The analytic model matches simulated flux, halo cross-correlations, and counts-in-cells down to a few Mpc.","key_machinery":"The machinery is the effective-field-theory bias expansion for the Lyman-$\\alpha$ forest, with each bias parameter promoted to a momentum-dependent transfer function $\\beta_O(k,\\mu)$ fitted by cross-correlating the simulated flux with each operator. The operator set includes the linear matter density, the line-of-sight velocity gradient $\\eta = \\partial_\\parallel v_\\parallel/(aH)$, quadratic density and velocity terms, and additional line-of-sight operators such as $\\eta^2$ and $(KK)_\\parallel$; operators are orthogonalized so each transfer function is independent, and all operators are shifted by first-order Lagrangian displacements to resum long-wavelength motions. The diagnostic is the error power spectrum $P_{\\rm err}(k,\\mu) = \\langle |\\delta_F^{\\rm truth} - \\delta_F^{\\rm model}|^2\\rangle$, which the effective field theory predicts to be white with small scale-dependent corrections; the paper measures the leading noise amplitude $n_0 \\approx 0.18\\,[h^{-1}\\,{\\rm Mpc}]^3$.","core_discovery":"The central claim is that the Lyman-$\\alpha$ transmitted-flux fluctuation field can be predicted analytically at the field level from the initial dark matter density field, not just in its summary statistics. With an effective-field-theory bias expansion carried to cubic order, including line-of-sight dependent operators, and with each bias parameter promoted to a scale- and angle-dependent transfer function, the model matches the simulated flux field in both amplitude and phase. Concretely, the paper reports that the flux power spectrum is reproduced at the five percent level up to $k \\approx 0.6\\,h\\,{\\rm Mpc}^{-1}$, the counts-in-cell distribution down to cell radii of $1{-}2\\,h^{-1}\\,{\\rm Mpc}$, and the Lyman-$\\alpha$-halo cross-spectrum up to $k \\approx 1\\,h\\,{\\rm Mpc}^{-1}$. It also provides a first estimate of the stochastic noise power spectrum of the three-dimensional flux field, finding a noise floor $n_0 \\approx 0.18\\,[h^{-1}\\,{\\rm Mpc}]^3$ that appears to be inherited from dark matter halos.","pith_inferences":["If the universality claim survives an independent test, the calibrated transfer functions could be applied to initial conditions inferred from other cosmological probes, enabling a joint field-level likelihood across surveys.","The flattening of the noise power spectrum near $k \\approx 0.6\\,h\\,{\\rm Mpc}^{-1}$ suggests the stochastic component might be absorbed by running the effective-field-theory expansion against the full nonlinear matter field from an N-body simulation; the paper flags exactly this as future work.","A direct extension would be to calibrate the transfer functions on simulations with different IGM thermal histories, since the current validation uses a single photoionization and temperature state.","The same transfer-function formalism could be adapted to Lyman-alpha tomographic maps from high-redshift galaxies, trading line-of-sight resolution for much larger survey volume."],"forward_implications":["Lyman-alpha forest data from DESI can be analyzed by field-level inference, extracting information beyond the power spectrum without running large-volume hydrodynamic simulations.","Simulation-based priors for Lyman-alpha effective-field-theory parameters can be built from the analytic model, removing sample variance from parameter measurements.","The model can generate large-volume Lyman-alpha mocks for covariance matrices needed in joint Lyman-alpha-quasar analyses.","The measured stochastic noise floor sets the irreducible error budget for cosmological inference from the forest.","The failure of the linear bias model at the field level shows that higher-order bias operators are mandatory for any field-level analysis of the flux."],"supporting_citations":[{"why":"It supplies the hydrodynamic simulation snapshot, initial conditions, and halo catalog used to validate the forward model.","marker":"[83]"},{"why":"It establishes the field-level forward-model and transfer-function approach for redshift-space tracers that this paper adapts to the Lyman-alpha forest.","marker":"[75]"},{"why":"It provides the operator orthogonalization and transfer-function framework on which the model construction relies.","marker":"[86]"},{"why":"It gives the effective-field-theory bias expansion for the Lyman-alpha forest up to cubic order in redshift space.","marker":"[81]"},{"why":"It supplies the linear relation between flux fluctuations, dark matter density, and the line-of-sight velocity gradient that the model extends.","marker":"[84]"},{"why":"It provides the effective-field-theory model for Lyman-alpha-halo cross-correlations used for comparison with the combined forward model.","marker":"[99]"}],"fun_headline_variants":["Analytic Lyman-alpha model matches simulations to percent level","Field-level Lyman-alpha forest from analytic forward model","Lyman-alpha forest predictions at percent accuracy without simulations","Analytic model predicts Lyman-alpha forest field from initial conditions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The transfer functions calibrated on the single hydrodynamic simulation snapshot are universal, meaning the same fitted numerical coefficients apply to other initial-condition realizations and redshifts without re-fitting; the paper asserts this on effective-field-theory grounds but does not test it on an independent simulation.","fun_headline_variants_meta":{"raw":{"variants":["Analytic Lyman-alpha model matches simulations to percent level","Field-level Lyman-alpha forest from analytic forward model","Lyman-alpha forest predictions at percent accuracy without simulations","Analytic model predicts Lyman-alpha forest field from initial conditions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000689,"raw_usage":{"total_tokens":3136,"prompt_tokens":977,"completion_tokens":2159,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":593,"completion_tokens_details":{"reasoning_tokens":2093}},"tokens_in":593,"tokens_out":2159,"duration_ms":18706,"temperature":1.0,"reasoning_tokens":2093,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T21:20:12.488285+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Apply the calibrated transfer functions to a second hydrodynamic simulation with different initial conditions or to a different redshift snapshot of the same simulation suite, and measure the per-mode error spectrum $P_{\\rm err}(k,\\mu)$; if the flat percent-level error spectrum from the calibration snapshot is not recovered, the claimed universality is falsified.","supporting_citations":[],"review_version":1}