{"id":"7d9da271-9a2b-44a5-90d7-e8e58511d3ba","arxiv_id":"2507.01399","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Proves a stability theorem for recovering early-universe perturbations in a shell near the last scattering surface from partial ISW line integral data, with a 2D numerical demonstration.","lead":"This mathematics paper proves that gravitational perturbations near the last scattering surface can be stably reconstructed from the integrated Sachs-Wolfe effect in the cosmic microwave background, using only observations made near Earth. The authors prove a stability theorem for this tomographic inverse problem and test several regularization methods on a simplified 2D simulation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 5.1's proof is invalid: the Fourier-analyticity step uses a v-dependent function, so the uniqueness input for Theorem 2.1 is not established.","rationale":"The reader identifies the uniqueness step as load-bearing, and I agree that it is the critical point; however, the precise failure is earlier than the finite-speed transition. Lemma 5.1's Fourier-slice proof does not work because A_v is v-dependent: the analyticity argument requires one function vanishing on an open set in frequency space, not a family of functions vanishing on different hyperplanes. The stated region in (5.2) is also inconsistent with the visibility calculation, which further undermines the proof of Theorem 2.1 as written. The proposed SVD check on the paper's own setup would at least test the theorem's uniqueness statement in the discrete model. The conditional verdict is therefore unchanged: the central claim is plausible but not established until Lemma 5.1 is repaired or replaced.","tokens_in":24760,"tokens_out":35510,"duration_ms":399853,"concrete_test":"Using the paper's own 2D discrete model (Section 6, T=2, 51×51 grid, 7×7 and 3×3 detector grids), assemble A=HS with columns restricted to pixels in the annulus 1<|x|<3 and compute the SVD of A. If a numerically zero singular value appears while the number of rows exceeds the number of ring pixels, and persists under mesh refinement, Theorem 2.1 has a concrete counterexample in the discrete model. If no zero singular value appears, the theorem may still be true, but the proof still requires replacing Lemma 5.1 by a valid uniqueness argument.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 2.1 hinges on Lemma 5.1, but Lemma 5.1 is not proved and as stated is likely false. In (5.3) the function A(t,x)=χ(t)φ(x−tv+Tv)u(t,x) depends on v; the equation \\hat A(−v·ξ,ξ)=0 holds for each fixed v only on the hyperplane τ=−v·ξ. Because A_v changes with v, the union of these hyperplanes over v∈S^2 cannot be used to conclude that a single analytic \\hat A vanishes on an open set. Thus the displayed conclusion u=0 on (5.2) does not follow. Moreover, the displayed region (5.2) is wrong: the visibility condition for a ray through x at time t is ||x|−(T−t)|<1, i.e. T−t−1<|x|<T−t+1, so the inequality T−t≤|x| omits the inner shell T−t−1<|x|<T−t. Even if Lemma 5.1 were repaired, the contradiction argument in Section 5 would need a valid uniqueness input; without it, the lower-order terms in (5.1) cannot be removed.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the inverse problem of recovering the initial Cauchy data (f1,f2) of a scalar perturbation solving a wave equation with lower-order terms from the light ray transform Lu of the solution, where the transform is observed on a unit ball at the final time t=T. The main result, Theorem 2.1, claims stable and unique recovery of data supported in the shell R_{T-1,T+1}, with the H^{s+2} norm of Lu controlling the H^{s+1} and H^s norms of f1 and f2. The proof combines a microlocal inversion based on backprojection (Sections 3-4) with a compactness argument (Section 5) that requires a uniqueness lemma for the light ray transform. The paper also contains a numerical study for a 2D model comparing several regularization approaches.","tokens_in":24888,"tokens_out":23404,"duration_ms":253019,"significance":"If Theorem 2.1 were fully established, the paper would be a significant contribution to cosmological X-ray tomography: it would provide a parameter-free stability estimate for partial-data recovery from the integrated Sachs-Wolfe effect, extending the full-data results of Vasy and Wang. The microlocal construction in Sections 3-4 is substantial and plausible, and the numerical experiments with synthetic data illustrate the potential of regularized inversion. However, the central stability theorem currently rests on a uniqueness lemma that is false as stated and whose proof is invalid, so the significance is conditional on repairing the uniqueness argument.","major_comments":[{"comment":"The lemma as stated is false for arbitrary u in H^s. Take u(t,x)=a(t) with a in C_c^∞(0,T) and ∫_0^T a(t)dt=0; then Lu(y,v)=0 for all (y,v), but u is not zero on the set (5.2). Moreover, the proof does not establish the stated conclusion: in (5.3), the function A(t,x)=χ(t)φ(x−tv+Tv)u(t,x) depends on v, so the identity \\hat A_v(−v·ξ,ξ)=0 is for a different function for each v. An entire function can vanish on the hyperplane τ=−v·ξ without being identically zero, and the union of these hyperplanes over v does not give a fixed analytic function vanishing on an open set. Thus the claim u=0 on (5.2) is not proved.","section":"Section 5, Lemma 5.1"},{"comment":"The displayed visibility region is incorrect. A point (t,x) lies on a null ray through B at time T exactly when there exists v∈S² with |x−(t−T)v|<1, i.e., when T−t−1<|x|<T−t+1 (for T−t>1). The region T−t≤|x|<T−t+1 in (5.2) omits the inner shell T−t−1<|x|<T−t. At t=0 this means the lemma would only constrain T≤|x|<T+1, not the full support R_{T−1,T+1} of the data in Theorem 2.1, so the contradiction argument cannot force f1=f2=0.","section":"Section 5, Eq. (5.2)"},{"comment":"The passage from the a priori estimate (5.1) to the final estimate (2.5) depends entirely on removing the terms C∥f1∥_{H^s}+C∥f2∥_{H^{s−1}} by the compactness argument. Since that removal uses Lemma 5.1, and the lemma is neither true as stated nor proved for solutions of (2.4), the central stability estimate is not established. The proof must supply a valid unique continuation result for solutions of the wave equation with lower-order terms, or replace the compactness argument.","section":"Section 5, proof of Theorem 2.1"},{"comment":"The stationary-phase assignments appear to be interchanged for the E+ contribution. The critical point v=+ξ/|ξ| yields the phase (x−y)·ξ+(2t−T−ϵ)|ξ|, while v=−ξ/|ξ| yields (x−y)·ξ+(T+ϵ)|ξ|; in (3.7) the phase factors e^{i(T+ϵ)|ξ|} and e^{-i(T+ϵ)|ξ|}e^{i2t|ξ|} are assigned to b+_{1,1} and b+_{1,2} respectively, while the symbols in (3.8) use the opposite arguments ±ξ/|ξ|. This inconsistency should be resolved, since the canonical relations and principal symbols from Lemma 3.1 feed into the parametrix construction in Section 4.","section":"Section 3, Eqs. (3.7)-(3.8)"}],"minor_comments":[{"comment":"The paper should state the assumption T>1, or explicitly invoke the convention R_{a,b}=B_b for a≤0, because the proof selects ϵ with the support contained in R_{T−1+2ϵ,T+1−2ϵ}.","section":"Section 2, Theorem 2.1"},{"comment":"There are apparent typos in the parametrix definitions; for example W+ is introduced as a parametrix of the f2 coefficient but is then used to isolate f1, and (4.11) appears to isolate f1 instead of f2. These need to be corrected and checked.","section":"Section 4, Eqs. (4.10)-(4.14)"},{"comment":"If the intended uniqueness statement is only for solutions of (2.4), the lemma should say so and the proof must use the equation; the present proof never uses it, and the counterexample in my major comment shows the unrestricted statement is false.","section":"Section 5, Lemma 5.1"},{"comment":"The numerical experiments solve a 2D discrete inverse problem and do not directly validate the 3D stability estimate (2.5); the conclusions in Section 7 should describe the numerics as heuristic support rather than as a demonstration of Theorem 2.1.","section":"Section 6"},{"comment":"The caption refers to visible structures 'in the region given by Theorem 2.1' for a 2D example; since the theorem is stated for R^3, the caption should clarify the relation.","section":"Figure 1"}],"recommendation":"major_revision","confidential_remarks":"The paper's main idea is attractive and the microlocal inversion is plausible, but the proof of Theorem 2.1 has a serious gap in the uniqueness lemma. In my view this is fixable in a revision, but the revision will need to add a genuine unique continuation argument for the wave equation with lower-order terms; a counterexample shows the current lemma cannot be repaired as a statement about arbitrary functions. The numerical section is useful but secondary. Scope is appropriate for the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The first thing you should know: the microlocal part is real work and the numerical study is a useful proof of concept, but Lemma 5.1 is not proved, and Theorem 2.1 leans on it. The paper is worth refereeing, not because the proof currently closes, but because the gap looks repairable and the problem is timely.\n\nWhat is genuinely new: the partial-data stability estimate extending Vasy-Wang's full-data result, the weighted-backprojection construction in Sections 3-4, and the clean FIO decomposition of the backprojected light ray transform. The numerical section is honest: it uses synthetic 2D data, reports conditioning of the forward map, and compares LSQR, FISTA, and an edge-preserving IGMRF method. No fitted parameters, no benchmark tuning; self-citations are legitimate because the full-data results are prior independent work. That part stands.\n\nThe soft spot is load-bearing. Lemma 5.1 tries to show that Lu=0 forces u=0 on the visible shell. The Fourier-slice computation defines A_v(t,x)=chi(t)phi(x-tv+Tv)u(t,x), which depends on v. From Lu=0 you only get Ahat_v(-v·xi,xi)=0 for each fixed v on the hyperplane tau=-v·xi. Since A_v changes with v, the union of these hyperplanes does not give one analytic function vanishing on an open set; the conclusion that A is identically zero does not follow. The stated region (5.2) is also wrong: the visibility condition is T-t-1≤|x|≤T-t+1, not T-t≤|x|<T+1-t. At t=0, the omitted inner shell is exactly the inner part of the Cauchy data the theorem claims to recover. Without a valid uniqueness input, the compactness argument in Section 5 cannot remove the lower-order terms, and (2.5) is only established modulo smoother terms.\n\nEverything else is in proportion: the physical ISW-to-transform reduction is quick and the scalar-field simplification is acknowledged; the numerics are 2D and code-adjacent rather than code. Those are minor. The gap in Lemma 5.1 is the thing a referee must resolve.\n\nRecommendation: send to peer review. A serious referee should ask for a repaired uniqueness argument or a different proof of the compactness step, and a corrected visibility region. If that comes back clean, the paper is a solid contribution. I would not cite the stability estimate in its current form.","headline":"Partial-data ISW tomography with real microlocal content, but Lemma 5.1 has a gap that leaves Theorem 2.1 unproved as written.","tokens_in":25523,"tokens_out":11954,"would_cite":false,"duration_ms":157586,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35R30","35L05","53C65","35S30"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that initial gravitational perturbations in a shell around the last-scattering surface can be stably recovered from integrated Sachs-Wolfe observations made near the observer.","keywords":["integrated Sachs-Wolfe effect","cosmological X-ray tomography","light ray transform","inverse problems","Fourier integral operators","partial data","wave equation","regularization"],"falsifier":"Run the paper's numerical discretization on initial data supported in the shell $R_{T-1,T+1}$ for a small detector grid such as $7\\times7$ and inspect the singular values of the forward matrix restricted to that subspace; singular values at machine precision would indicate a nontrivial numerical null space, contradicting the claimed stability. Alternatively, attempt to construct nonzero compactly supported $(f_1,f_2)$ in the shell whose wave solution satisfies $Lu=0$ on $B\\times S^2$; any such pair directly refutes Theorem 2.1.","tokens_in":24415,"feed_emoji":"🌌","tokens_out":6632,"duration_ms":69436,"temperature":0.7,"pith_summary":"The paper sets out to answer a practical question: can the state of the early universe be reconstructed from cosmic microwave background (CMB) measurements made only near the Earth, through the integrated Sachs-Wolfe (ISW) effect? It models the problem as a partial-data tomographic inverse problem for a scalar wave equation describing gravitational perturbations. The central claim is Theorem 2.1: initial data supported in the shell $R_{T-1,T+1}$ — a layer of thickness two just inside the backward light cone of the observer region — are uniquely and stably determined by the light ray transform measured on a unit ball at the final time. The result matters because it shows the observable-universe shell can in principle be recovered from line-of-sight ISW integrals, despite the transform itself being severely ill-posed. Numerical experiments in a simplified two-dimensional universe illustrate that with regularization the shell structure is recoverable.","feed_headline":"CMB data near Earth can be inverted to rebuild the early universe","feed_subtitle":"A tomography proof shows the shell of initial gravitational perturbations is stably recoverable from ISW observations.","key_machinery":"The load-bearing construction is the microlocal analysis of the backprojection of the light ray transform. The paper uses a modified transform $L_\\epsilon = \\varphi_\\epsilon L$ with cutoff $\\varphi_\\epsilon$ supported in $B_{1-\\epsilon}$ to avoid boundary artifacts, and two backprojections: $I h(x)=\\int_{S^2} h(x,v)\\,dv$ and a weighted version $I^\\varphi h(x)=\\int_{S^2}\\varphi(v)h(x,v)\\,dv$ with $\\varphi$ chosen so that $\\varphi(\\xi/|\\xi|)\\neq \\varphi(-\\xi/|\\xi|)$. Stationary phase decomposes $I L_\\epsilon E^\\pm_j$ into Fourier integral operators with canonical relations $C_a = \\{(x,\\xi;y,\\eta): \\xi=\\eta,\\ x=y-a\\xi/|\\xi|\\}$; combining the two backprojections cancels the leading-order terms and yields, via parametrices $Q^\\pm, W^\\pm, S$, explicit inversion formulas for $f_2$ and $f_1$ modulo regularizing operators.","core_discovery":"On its own terms, the paper proves a stability estimate for the cosmological X-ray transform with partial data. For the wave equation $P u=0$ with $P = \\Box + a_0 \\partial_t + \\sum a_j D_j + b$, Theorem 2.1 states that if $f_1 \\in H^{s+1}(\\mathbb{R}^3)$ and $f_2 \\in H^s(\\mathbb{R}^3)$ are compactly supported in the spherical shell $R_{T-1,T+1}$, then the observations $Lu$ on $B \\times S^2$ determine $f_1, f_2$ uniquely and\n$$\\|f_1\\|_{$H^{{s+1}}$} + \\|f_2\\|_{H^s} \\leq C \\|Lu\\|_{$H^{{s+2}}$},$$\nwith $C$ uniform over data supported in a fixed compact subset of the shell. This is the first stable recovery result for the ISW problem in the realistic setting where observations are confined to a small region near the Earth rather than a full Cauchy surface. The paper also constructs an explicit microlocal inversion, formulas (4.16) and (4.17), that recovers $f_2$ and then $f_1$ up to smoother terms, and it reports two-dimensional numerical reconstructions showing the shell structure emerging from simulated noisy ISW data.","pith_inferences":["Editorial inference: the same stationary-phase decomposition should yield a precise visible wave-front set for data supported in the whole ball $B_{T+1}$; writing it out would turn the remark after Theorem 2.1 into a full microlocal uniqueness theorem.","Editorial inference: the physical model, not the ray transform alone, is what stabilizes the inversion; this suggests that other causal PDE constraints, such as sound-speed wave equations for fluids, could stabilize otherwise ill-posed ISW inversions similarly.","Editorial inference: the unproved unique-continuation step could be tested numerically by computing the singular value spectrum of the discretized forward operator restricted to shell-supported initial data; a nontrivial numerical null space would indicate the stability estimate needs modification."],"forward_implications":["Initial data in the shell $R_{T-1,T+1}$ can be recovered with a quantified loss of three Sobolev derivatives: data in $H^{s+1}\\times H^s$ are controlled by observations in $H^{s+2}$.","The stability estimate is uniform over compact families of data, so small metric perturbations of the background can be handled by the same argument.","In the partial-data regime the discrete forward problem is severely ill-posed, with condition numbers growing rapidly as the detector grid shrinks, so regularization is essential; the paper's experiments with sparsity and edge-preserving priors demonstrate this.","The microlocal inversion formulas can in principle characterize which wave-front directions of data supported in the full ball $B_{T+1}$ are recoverable, extending the result beyond the shell."],"supporting_citations":[{"why":"Supplies the full-data stable recovery result and the backprojection strategy that this paper adapts to partial data.","marker":"[28]"},{"why":"Introduced the name and initial mathematical formulation of cosmological X-ray tomography.","marker":"[11]"},{"why":"Shows that time-like singularities of the perturbation are recoverable from the ISW transform; the partial-data microlocal analysis builds on this.","marker":"[19]"},{"why":"Provides the numerical discretization of the light ray transform and a numerical demonstration of the ill-posedness that motivates regularization.","marker":"[5]"},{"why":"Gives the Fourier integral operator parametrix construction for wave propagators used to represent solutions.","marker":"[27]"},{"why":"Provides the Fourier integral operator calculus and Sobolev mapping properties used in the microlocal inversion and stability estimates.","marker":"[14]"},{"why":"Derives the integrated Sachs-Wolfe formula expressing redshift as an integral of metric perturbations.","marker":"[25]"},{"why":"Derives the linearized scalar-perturbation equation (1.2) that the paper uses as the physical model.","marker":"[23]"}],"fun_headline_variants":["Cosmic CT: stable inversion of the ISW effect","See the early universe via CMB tomography","Rebuild the primeval cosmos from ISW data","ISW tomography: from photons to initial structure","A stable X-ray transform for the cosmos"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof's load-bearing premise is the unproved unique-continuation step that $Lu=0$ forces the initial data supported in the shell $R_{T-1,T+1}$ to vanish; if that implication fails, the compactness argument collapses and the stability estimate may only hold modulo smoother terms.","fun_headline_variants_meta":{"raw":{"variants":["Cosmic CT: stable inversion of the ISW effect","See the early universe via CMB tomography","Rebuild the primeval cosmos from ISW data","ISW tomography: from photons to initial structure","A stable X-ray transform for the cosmos"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000445,"raw_usage":{"total_tokens":2247,"prompt_tokens":941,"completion_tokens":1306,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":557,"completion_tokens_details":{"reasoning_tokens":1234}},"tokens_in":557,"tokens_out":1306,"duration_ms":11525,"temperature":1.0,"reasoning_tokens":1234,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T20:57:51.044111+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the paper's numerical discretization on initial data supported in the shell $R_{T-1,T+1}$ for a small detector grid such as $7\\times7$ and inspect the singular values of the forward matrix restricted to that subspace; singular values at machine precision would indicate a nontrivial numerical null space, contradicting the claimed stability. Alternatively, attempt to construct nonzero compactly supported $(f_1,f_2)$ in the shell whose wave solution satisfies $Lu=0$ on $B\\times S^2$; any such pair directly refutes Theorem 2.1.","supporting_citations":[{"cited_title":"Vasy and Y","cited_arxiv_id":null,"evidence_quote":"Supplies the full-data stable recovery result and the backprojection strategy that this paper adapts to partial data."},{"cited_title":"Guillemin","cited_arxiv_id":null,"evidence_quote":"Introduced the name and initial mathematical formulation of cosmological X-ray tomography."},{"cited_title":"Lassas, L","cited_arxiv_id":null,"evidence_quote":"Shows that time-like singularities of the perturbation are recoverable from the ISW transform; the partial-data microlocal analysis builds on this."},{"cited_title":"Chung, L","cited_arxiv_id":null,"evidence_quote":"Provides the numerical discretization of the light ray transform and a numerical demonstration of the ill-posedness that motivates regularization."},{"cited_title":"Tr` eves.Introduction to pseudodifferential and Fourier integral operators Volume 2: Fourier integral operators, volume 2","cited_arxiv_id":null,"evidence_quote":"Gives the Fourier integral operator parametrix construction for wave propagators used to represent solutions."},{"cited_title":"H¨ ormander.The analysis of linear partial differential operators IV: Fourier integral operators","cited_arxiv_id":null,"evidence_quote":"Provides the Fourier integral operator calculus and Sobolev mapping properties used in the microlocal inversion and stability estimates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Derives the integrated Sachs-Wolfe formula expressing redshift as an integral of metric perturbations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Derives the linearized scalar-perturbation equation (1.2) that the paper uses as the physical model."}],"review_version":1}