{"id":"7243783c-fc7b-458f-9550-85ad63cc8cc9","arxiv_id":"2411.15959","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"The authors propose fitting the distribution of two-point products of CMB map pixels with the analytic PDF for a product of correlated Gaussians to recover the inflationary power spectrum, but provide no validation and the derivation contains apparent errors.","lead":"This paper proposes to separate the Gaussian (inflationary) part of CMB two-point correlations from non-Gaussian foregrounds by fitting the observed distribution of pixel-pair products to the analytic distribution of a product of two correlated Gaussian variables. The method is not demonstrated on data or simulations, and the analytic distribution appears to have normalization and scaling errors.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The core identifiability claim is unsupported and generically false: fitting the Gaussian product PDF to observed pixel-pair products cannot recover the inflationary-only correlation in the presence of foregrounds or Gaussian noise, and Eq. (8) is not a correct product PDF.","rationale":"The reader's verdict is REJECT with high confidence, primarily on soundness grounds. My stress-test agrees and identifies the same load-bearing weakness: the paper asserts, without proof or simulation, that fitting a Gaussian product distribution to the observed pixel-pair product distribution recovers the inflationary Gaussian correlation. That assertion is not merely unproven; it is generically false. The observed distribution is a convolution of three components, and a Gaussian template fit to a convolution does not isolate one component unless strong orthogonality conditions hold, which are neither stated nor demonstrated. A concrete failure mode is Gaussian noise: if the foreground f is Gaussian, the total field s is Gaussian, and the product distribution is exactly of Pgg form with a total variance that includes noise; the method cannot distinguish the inflationary Gaussian component from any other Gaussian component. This is a fundamental identifiability problem, not a technical detail. In addition, the analytic expression for Pgg in Eqs. (7)–(8) has clear normalization and scaling errors and, as written, diverges for negative x_r, so the template is invalid even in the clean case. The paper provides no simulations, no recovery tests, and no error analysis, so there is no empirical support that could rescue the claim. The reader's weakest_assumption correctly identifies the identifiability issue, and my reading does not change the REJECT verdict.","tokens_in":4691,"tokens_out":7027,"duration_ms":67060,"concrete_test":"Run a Monte Carlo at a single angular scale r. Generate a Gaussian CMB realization with known correlation ξr and zero noise; compute pixel-pair products, histogram them, and fit the histogram with the corrected product PDF (and with Eq. (8) as written). Repeat the same procedure after (a) adding Gaussian white noise of known variance and (b) adding a non-Gaussian foreground component such as a dust template. If the fitted ξr in the noiseless case deviates from the input beyond statistical error, the template is wrong. If the fit with Gaussian noise returns the total correlation rather than the input CMB ξr, or if the fit with foregrounds is biased by more than the statistical uncertainty, the central identifiability claim fails. This directly tests both the normalization of Eq. (8) and the ILC-like noise interpretation of Δ.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim requires that fitting the template Pgg(x_r, ξ_r) to the observed distribution P(x_r) of pixel-pair products returns the Gaussian correlation ξ_r of the inflationary signal. This claim is the load-bearing assumption, and it is both unproven and generically false. With s = g + f, the observed product PDF is a convolution of the Gaussian product distribution with foreground and cross-term distributions, as the paper's own Eq. (5) shows. Writing P(x_r) = Pgg(x_r) + Δ(x_r) in Eq. (6) is a tautology; it does not imply that Δ behaves as additive noise. For a fit of Pgg to P to recover ξ_r, Δ must be nearly orthogonal to ∂Pgg/∂ξ_r over the fitted range, which is not shown and is not generally true. If f is Gaussian—for instance, instrumental noise that is independent between pixels—then s itself is Gaussian with total variance ξ0 + σ² and correlation ξr; the fit would return the total Gaussian correlation, which includes noise and any other Gaussian component, so the inflationary-only ξr is not identifiable from P(x_r) alone. Separately, the analytic template in Eqs. (7)–(8) is incorrect: for unit-variance correlated normals the product PDF is (1/(π√(1−γ²))) exp(γx/(1−γ²)) K0(|x|/(1−γ²)), whereas Eq. (8) has a different coefficient, no ξ0 scaling, a factor of 1/4 in the exponent, and an integral that diverges for x_r < 0 because the absolute value is omitted. Thus even in a perfectly clean, noiseless map, the proposed fit cannot return the correct ξr.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript proposes a method for extracting the angular power spectrum component generated by a Gaussian inflationary field from CMB observations. The authors consider pixel-pair products x_r = s1 s2 at fixed angular separation r, derive an analytic probability distribution P_gg(x_r, ξ_r) for the case of a pure Gaussian signal, and claim that fitting this distribution to the observed distribution P(x_r) yields the correct two-point correlation ξ_r, from which the 'clean' power spectrum can be obtained. The method is asserted to separate Gaussian inflationary correlations from non-Gaussian foregrounds and lensing. No simulations or data applications are presented; the paper consists of a short derivation and a statement of intent.","tokens_in":5093,"tokens_out":6552,"duration_ms":54643,"significance":"The idea of using the full distribution of pixel-pair products as a statistic to separate Gaussian from non-Gaussian components is conceptually interesting and, if it worked, could provide a complementary CMB power-spectrum estimator. However, the paper's central derivation contains concrete errors: Eq. (7) is not a correctly normalized joint Gaussian PDF with the stated variance ξ0, and Eq. (8) is not a valid normalized product PDF (for γ=0 it integrates to 1/8 over positive x_r and diverges for negative x_r). Furthermore, the load-bearing identifiability assumption—that fitting P_gg(x_r, ξ_r) to the observed P(x_r) recovers the inflationary-only correlation—is stated without proof or simulation and is generically false when foregrounds include any Gaussian component. Because the method is not validated and the analytic foundation is incorrect, the paper does not establish its central claim.","major_comments":[{"comment":"Eq. (7) is not the joint probability density for two Gaussian variables with zero mean and variances ⟨g1²⟩=⟨g2²⟩=ξ0. The exponent should contain a factor 1/ξ0 inside the exponentials; as written, the density implicitly assumes ξ0=1, which contradicts the stated definitions. The prefactor is also wrong by a factor of 2: for unit variances the correct prefactor is 1/(2π√(1-γ²)), not 1/(4π√(1-γ²)). Consequently Eq. (7) is not normalized and cannot be the starting point for the derivation of Eq. (8).","section":"Section II, Eq. (7)"},{"comment":"The derived product PDF P_gg(x_r) is incorrect. For the special case γ=0 (independent unit-variance Gaussians), the standard product distribution is (1/π)K_0(|x_r|). Eq. (8) instead gives, for x_r>0, (1/(16π))K_0(x_r/4), which integrates to 1/8 over the positive real line and is therefore not normalized. For x_r<0 the integral in Eq. (8) diverges because the exponent becomes positive for t>γ. Even if the prefactor and scaling are corrected, the missing absolute value in the exponent is a fatal flaw. Since the entire fitting procedure in Section II relies on this template, the claimed recovery of ξ_r via Eq. (9) is not established.","section":"Section II, Eq. (8)"},{"comment":"The central identifiability claim—that fitting P_gg(x_r, ξ_r) to the observed distribution P(x_r) gives the correct value of ξ_r—is asserted without proof, simulation, or a precision analysis. Writing P(x_r)=P_gg(x_r)+Δ(x_r) is a tautology; treating Δ as 'noise' in an ILC-like fit requires that Δ be approximately orthogonal to ∂P_gg/∂ξ_r over the fitting region, which is neither shown nor generally true. A concrete counterexample: if the foreground f is Gaussian instrumental noise of variance σ², then s=g+f is itself Gaussian with total variance ξ0+σ² and correlation ξr, so the observed product distribution is exactly P_gg(x_r; ξ0+σ², ξr). Fitting the template returns the total Gaussian correlation, which includes the noise contribution; the inflationary-only ξr is not identifiable from P(x_r) alone. The authors need to provide either a rigorous argument or end-to-end simulations demonstrating that the fit isolates the inflationary component in the presence of foregrounds.","section":"Section II, Eqs. (5)-(6) and the paragraph 'To correctly fit...'"}],"minor_comments":[{"comment":"The title mentions Simons Observatory, but the paper contains no analysis of SO data or SO-specific forecasts; the only connection is a reference in the introduction. The scope should be clarified.","section":"Title/Introduction"},{"comment":"The notation in Eq. (5) is under-specified: the integral sign is written as 'R' and the integration variables y_r, z_r are introduced without explicit limits. The Jacobian factor 1/4 is not derived in the text.","section":"Section II, Eq. (5)"},{"comment":"The statement that for a negligibly small foreground the functions P_gf and P_ff 'actually turn into delta functions' is imprecise and is not used in the subsequent derivation; the limit is not formally justified.","section":"Section II, after Eq. (4)"},{"comment":"Figure 1 shows the proposed P_gg(x_r) for three values of ξ_r/ξ0, but no comparison to a histogram or Monte Carlo sample is provided, and the vertical dashed lines are not explained in the caption. Given that Eq. (8) is the object being plotted, a direct test against a Gaussian simulation would have been instructive.","section":"Figure 1"},{"comment":"The conclusions state that the method 'extracts the correct part of the power spectrum from observational data,' but no uncertainty quantification or validation on mock sky maps is presented. This overstates the demonstrated result.","section":"Section III"}],"recommendation":"reject","confidential_remarks":"The manuscript is very short and appears to be an early-stage research note. The mathematical errors in Eqs. (7) and (8) are straightforward to verify and are central to the method; the identifiability assumption is not just unproven but fails in a simple Gaussian-foreground case. I see no route to making the central claim sound without substantially reformulating the method and demonstrating it with simulations, which is beyond what a revision can achieve in its current scope. The paper may be of interest as a preliminary idea, but it is not ready for publication in a journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Thanks for the report. I agree with your rejection, and the stress-test note holds up on reading the paper. The genuinely new thing here is the suggestion to fit the full PDF of pixel-pair products rather than just the mean, and to skip map cleaning entirely. That is a legitimate statistical idea: product distributions of correlated Gaussians are known, and reading the two-point function off the shape of the distribution rather than its average is a real methodological novelty within CMB analysis. If it worked, it would complement ILC and delensing. So credit for the application.\n\nThe problem is that the central derivation is not close to correct. Eq. (7) is off by a factor of two in normalization and ignores the variance ξ0. Eq. (8) is not the product PDF: for correlated unit-variance Gaussians the correct expression is (1/π√(1−γ²)) exp(γx/(1−γ²)) K₀(|x|/(1−γ²)), whereas their Eq. (8) has a different prefactor, a factor of 4 in the exponent, and no absolute value—so the integral diverges for x<0 and the distribution is not normalized. That is load-bearing: Eq. (9) claims the mean is ξr, but the object in Eq. (8) doesn't integrate to 1, so Eq. (9) doesn't follow.\n\nEven if the PDF were correct, the identifiability argument is asserted, not proven. Eq. (6) is a tautology; Δ is simply the difference between the observed distribution and the Gaussian template, and nothing guarantees that fitting the template recovers the inflationary ξr rather than the total Gaussian correlation. If the foreground is independent Gaussian noise, the observed signal is itself Gaussian with variance ξ0+σ² and correlation ξr, so the fit will return the total Gaussian ξr—noise included. And if ξ0 is also unknown, the shape depends on two parameters, not one.\n\nThere is no simulation, no synthetic test, no error analysis, no foreground model. For a three-page methodological proposal that is fatal, not a minor gap. The citation pattern is fine; the ILC references are appropriate context.\n\nWho is this for? Someone working on CMB component separation might read it for the novelty of the distribution-fitting idea. But it is not ready for a referee as it stands. Recommendation: desk reject, with a suggestion to the authors to fix the product PDF, include the variance parameter, and demonstrate the fitting on realistic simulations before resubmission.","headline":"A novel distribution-fitting idea for CMB power spectrum extraction, but the central PDF is wrong and the identifiability claim is unsupported; desk reject.","tokens_in":5592,"tokens_out":3765,"would_cite":false,"duration_ms":33792,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that fitting an analytic one-parameter distribution to histograms of CMB pixel-pair products recovers the two-point correlation function—and hence the angular power spectrum—of the inflationary Gaussian component alone…","keywords":["Cosmic Microwave Background","angular power spectrum","Gaussian random field","primordial B-mode polarization","foreground cleaning","two-point correlation function","internal linear combination"],"falsifier":"Simulate a sky with a known Gaussian CMB component plus realistic polarized dust and lensing, build histograms of pixel-pair products at fixed separations, fit the analytic product distribution, and compare the recovered correlation values and their Legendre power spectrum with the known input; biased recovery whenever the foreground has nonzero mean or variance comparable to the signal would falsify the central claim.","tokens_in":4470,"feed_emoji":"🌌","tokens_out":8376,"duration_ms":70607,"temperature":0.7,"pith_summary":"This paper proposes a way to pull the inflationary signal out of CMB maps without ever cleaning the map itself. The authors observe that the product of two values of a Gaussian random field at fixed angular separation has an analytic one-parameter probability distribution, and that any foreground or lensing contamination shows up only as a deviation from that distribution. They argue that fitting the analytic distribution to the observed histogram of pixel-pair products recovers the true two-point correlation of the Gaussian component, whose Legendre transform is the foreground-free angular power spectrum. If correct, this would give experiments like Simons Observatory a direct statistical route to the primordial B-mode polarization produced by inflation in the early universe.","feed_headline":"A single PDF fit recovers inflation's CMB power spectrum","feed_subtitle":"The method reads the Gaussian-only correlation out of pixel-pair products, no clean map required.","key_machinery":"The load-bearing object is the analytic probability density function for the product of two correlated Gaussian variables separated by angle $r$: $P_{gg}(x_r, \\xi_r) = \\frac{1}{16\\pi\\sqrt{1-\\gamma^2}} \\int_1^\\infty \\frac{e^{-(t-\\gamma)x_r/(4(1-\\gamma^2))}}{\\sqrt{t^2-1}} \\, dt$, with $\\gamma = \\xi_r/\\xi_0$. Its shape is set entirely by $\\xi_r$, and its first moment equals $\\xi_r$, so a fit of this one-parameter family to the observed pixel-pair product distribution is, by construction, a measurement of the Gaussian-only two-point correlation. This function converts the problem of foreground separation from map-space cleaning into a one-parameter distribution fit.","core_discovery":"The paper's central claim is that the two-point correlation function of the Gaussian, inflationary component of the CMB can be read off from the probability distribution of products of pixel values, without cleaning the map. For a fixed angular separation $r$, the product $x_r = s_1 s_2$ of two Gaussian values has the analytically known distribution $P_{gg}(x_r, \\xi_r)$ whose only free parameter is the correlation $\\xi_r$. The observed distribution is this Gaussian-product distribution plus a deviation $\\Delta$ produced by foregrounds and lensing. The paper asserts that fitting $P_{gg}$ to the observed histogram, treating $\\Delta$ as noise in the manner of internal linear combination, returns the true $\\xi_r$, and that Legendre-decomposing these $\\xi_r$ values yields the power spectrum produced by inflation alone.","pith_inferences":["The paper leaves the estimation of the zero-lag variance $\\xi_0$ unexamined, but the template depends on it through $\\gamma = \\xi_r/\\xi_0$; an implementation would need to propagate map noise and foreground residuals into $\\xi_0$ before $\\xi_r$ can be trusted.","A direct numerical test is the natural next step: simulate a Gaussian CMB plus dust and lensing, fit $P_{gg}$ to pixel-pair histograms, and compare the recovered power spectrum with the input; this would settle whether the noise treatment of the foreground deviation holds.","The same product-distribution idea should transfer to other nearly Gaussian sky fields, such as CMB lensing convergence or cosmic infrared background maps, wherever non-Gaussian contaminants bias two-point statistics."],"forward_implications":["For each angular separation $r$, the fitted parameter $\\xi_r$ is the Gaussian-only two-point correlation, and its Legendre transform gives a foreground-cleaned angular power spectrum $C_\\ell$ for temperature or polarization without constructing a clean map.","Applied to polarization, the method targets the B-mode spectrum from primordial gravitational waves, the science goal of Simons Observatory, CMB-S4, and LiteBIRD.","The statistical power of the method grows with map resolution because the number of pixel pairs at separation $r$ scales as $N \\cdot r/h$, and next-generation experiments provide this large pair count.","The authors note the fit can be upgraded with modeled foreground two-point information using modified internal-linear-combination variants (cILC, MILC, LRM)."],"supporting_citations":[{"why":"Cited for the inflationary prediction that primordial perturbations are a Gaussian random field, the premise that makes $P_{gg}$ the correct template for the clean signal.","marker":"[10, 11]"},{"why":"WMAP results cited as confirming that observed CMB anisotropy is close to Gaussian, supporting the Gaussian-signal assumption.","marker":"[12, 13]"},{"why":"Planck results cited for the near-Gaussianity of CMB anisotropy, the same justification applied to the polarization signal.","marker":"[14–16]"},{"why":"The internal linear combination method whose treatment of the extra component as noise is borrowed for fitting $P_{gg}$ to the observed distribution.","marker":"[17]"},{"why":"Identifies Simons Observatory as a next-generation experiment whose high-resolution maps would provide the large number of pixel pairs the method requires.","marker":"[1]"},{"why":"Delensing techniques that the proposed method aims to complement, representing the standard approach to the lensing B-mode foreground it avoids modeling.","marker":"[6–9]"}],"fun_headline_variants":["No clean map needed to get inflation's CMB spectrum","PDF fit extracts Gaussian-only CMB power spectrum","Inflation's spectrum read from pixel-pair distribution","Single fit pulls inflation signal from uncleaned CMB","Gaussian component of CMB isolated by PDF fit"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method rests on the assumption that fitting the Gaussian product distribution to the observed pixel-pair histogram, treating all foreground and lensing deviations as noise, returns the true Gaussian correlation value without bias, and the paper states this without proof or simulation.","fun_headline_variants_meta":{"raw":{"variants":["No clean map needed to get inflation's CMB spectrum","PDF fit extracts Gaussian-only CMB power spectrum","Inflation's spectrum read from pixel-pair distribution","Single fit pulls inflation signal from uncleaned CMB","Gaussian component of CMB isolated by PDF fit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000181,"raw_usage":{"total_tokens":1234,"prompt_tokens":797,"completion_tokens":437,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":413,"completion_tokens_details":{"reasoning_tokens":361}},"tokens_in":413,"tokens_out":437,"duration_ms":4221,"temperature":1.0,"reasoning_tokens":361,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:45:09.341117+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a sky with a known Gaussian CMB component plus realistic polarized dust and lensing, build histograms of pixel-pair products at fixed separations, fit the analytic product distribution, and compare the recovered correlation values and their Legendre power spectrum with the known input; biased recovery whenever the foreground has nonzero mean or variance comparable to the signal would falsify the central claim.","supporting_citations":[],"review_version":1}