{"id":"a7e5cb27-1553-4419-82e3-836dd9f2c71d","arxiv_id":"2505.17262","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A model-independent combination of strong lensing and supernova data gives H0 = 70.55 ± 7.44 km/s/Mpc, consistent with both Planck and SH0ES within 1sigma.","lead":"The authors combine gravitational lens time delays, Einstein radius measurements, and Type Ia supernova distances to estimate the Hubble constant without assuming a cosmological model, finding H0 = 70.55 ± 7.44 km/s/Mpc. The result is a cross-check of the cosmic distance scale that agrees with both major camps in the Hubble tension, though its uncertainty is too large to resolve the dispute.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The GP reconstruction of D^Obs as a function of z_l alone is the load-bearing step: Eq. 9 inherits any mismatch between the 99-system sample and the seven TDCOSMO lenses in source redshift or mass-profile slope, so the central H0 value can be biased even if the CDDR step is exact.","rationale":"The reader's weakest-assumption identification is the same as mine: the algebraic core is clean, but the bridge from the 99-system Einstein-radius sample to the seven TDCOSMO lenses is an interpolation over lens redshift only. I agree with CONDITIONAL because the issue is testable and the paper is transparent about limitations. I did not raise a separately different concern because every other issue (sigma_int, dataset selection) is either downstream of this mismatch or is a caveat the authors already state. The concrete test is deliberately model-independent in spirit: instead of assuming a cosmology, it uses the same SGL sample to see whether the source-redshift coordinate changes the answer. If H0 is stable, the projection onto z_l is harmless and the published result stands; if not, the method needs a two-dimensional reconstruction or a matched subsample before it can claim a model-independent determination.","tokens_in":15720,"tokens_out":5910,"duration_ms":50651,"concrete_test":"Re-run the analysis with a two-dimensional Gaussian process over (z_l, z_s) using the same 99 systems, and evaluate the reconstructed D^Obs at each TDCOSMO lens's actual z_s (and, if posteriors are available, at its gamma_pl) before forming Eq. 9. If the resulting seven D_Al values differ from the published z_l-only values by more than the GP error in quadrature, or if the final H0 shifts by more than ~1σ, the z_l-only GP is the dominant source of bias and the central claim is not established. A simpler cross-check is to restrict the 99 systems to those whose z_s lies within a narrow window of each TDCOSMO source redshift and recompute the GP; stability of H0 under that restriction indicates the concern is minor.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Eq. 4 defines D^Obs = D_Als/D_As for each strong-lensing system through z_l, z_s and the power-law slope gamma. Section II D reconstructs D^Obs(z_l) with a one-dimensional Gaussian process trained on 99 systems, and Eq. 9 multiplies that reconstruction by the TDCOSMO time-delay distance of each of seven lenses. The product is D_Al only if the reconstructed value equals the distance ratio of the actual TDCOSMO lens. The 99-system sample has source redshifts 0.196 <= z_s <= 3.595 and a fixed gamma = 2.1, while each TDCOSMO system has its own z_s and modeled gamma_pl. If the seven TDCOSMO systems sit systematically away from the sample's average z_s or gamma at their lens redshifts, every D_Al in Eq. 9 is shifted in the same direction and H0 from Eq. 10 is biased. The paper's assertion that source redshifts are 'clustered' around the time-delay values is supported only by a citation to a CDDR test, not by an explicit comparison, and fixed gamma = 2.1 is not matched to the TDCOSMO gamma_pl values. The fitted sigma_int ~24% enlarges the error bars but cannot remove a common-mode offset, since it is added after the fact to force chi2_red ~1. The limitation statements about mass-profile anisotropy are real, but the z_l-only GP is an additional missing marginalization over z_s and gamma.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a cosmological-model-independent estimate of the Hubble constant by combining three datasets at the lens redshifts of seven TDCOSMO time-delay systems: (i) the Einstein-radius observable D^Obs = D_Als/D_As, reconstructed as a function of lens redshift by a Gaussian process trained on 99 strong-lensing systems from Cao et al. (2015); (ii) the time-delay angular diameter distance D^Obs_{A,Δt} of each of the seven TDCOSMO systems; and (iii) the unanchored luminosity distance Θ = H0 D_L from the Pantheon+ supernova sample, also reconstructed by a Gaussian process. Using the cosmic distance duality relation, the authors form seven estimates H0,i = Θ(z_l)/[(1+z_l)^2 D^Obs(z_l) D^Obs_{A,Δt}] and combine them with an MCMC likelihood that includes a fitted intrinsic scatter, obtaining H0 = 70.55 ± 7.44 km/s/Mpc (68% CL). The paper frames the result as a model-independent consistency check that lies between the Planck and SH0ES values without resolving the Hubble tension.","tokens_in":16017,"tokens_out":21059,"duration_ms":142921,"significance":"The core algebraic identity is sound and genuinely cosmology-independent: D^Obs is constructed to be independent of H0, and the product D^Obs · D^Obs_{A,Δt} cancels the source angular diameter distance to leave D_Al; the CDDR then converts the SNe unanchored luminosity distance into H0 per lens. This is a clever and nontrivial way to build an inverse distance ladder from strong lensing alone, and the use of the full Pantheon+ covariance matrix, including the a_B intercept uncertainty through Eq. (13), is careful. The paper is also commendably explicit about its limitations (mass-sheet degeneracy, anisotropy, no marginalization over mass-profile details) and does not overclaim about resolving the Hubble tension. The main value, if the concerns below are addressed, is a novel cross-check that is independent of any FLRW metric assumption. The current constraining power is modest (about 10.5% precision), and the unbiasedness of the central value depends on assumptions about the Gaussian-process reconstruction and the lens mass profiles that are not yet demonstrated.","major_comments":[{"comment":"The Gaussian process reconstructs D^Obs ≡ D_Als/D_As as a one-dimensional function of the lens redshift z_l, but this distance ratio also depends on the source redshift z_s and on the mass-profile slope γ. Equation (9) evaluates the population-mean curve at the z_l of each of the seven TDCOSMO lenses and multiplies it by the per-system time-delay distance; the product equals D_Al only if each TDCOSMO system is representative of the 99-system sample in both z_s and γ at that z_l. The justification given at the end of Section II.B — that the source redshifts of the Einstein-radius sample are 'clustered around those in time-delay systems' — is supported only by a citation to ref. [67], with no explicit comparison of the z_s distributions at matched lens redshifts. If, for example, the TDCOSMO systems have systematically higher (or lower) source redshifts than the average of the 99 systems at the same z_l, all seven D_Al values shift in the same direction and H0 is biased by a common-mode offset that the fitted σ_int cannot remove. I request an explicit (z_l, z_s, γ) comparison between the seven TDCOSMO systems and the 99-system sample, or alternatively a two-dimensional GP in (z_l, z_s) with a marginalization over γ.","section":"II.D and Eq. (9)"},{"comment":"The Einstein-radius observable is computed with the power-law slope fixed at γ = 2.1 for all 99 systems, while the TDCOSMO time-delay distances used in Eq. (9) are individually derived under power-law lens models with system-specific slopes γ_pl (Table 2 of ref. [43]). Since Eq. (4) depends on γ through f(γ) and the factor (θ_ap/θ_E)^{2−γ}, a TDCOSMO system whose γ_pl differs from 2.1 yields a D^Obs estimate that is inconsistent with the mass model used for its time-delay distance; the product in Eq. (9) is then not D_Al for that system. The paper does not quote the γ_pl values or test their consistency with 2.1. The analysis should marginalize over γ (for instance with the prior from the lens-sample measurements) and report the sensitivity of H0 to this choice.","section":"II.A, Eq. (4)"},{"comment":"The final uncertainty is dominated by a fitted intrinsic scatter σ_int ≈ 24% that is added to the covariance matrix specifically 'to obtain a χ_red ≈ 1'. This large scatter means that the seven H0 estimates from Eq. (10) are mutually inconsistent at the level of their propagated uncertainties; the quoted 68% interval H0 = 70.55 ± 7.44 km/s/Mpc is therefore determined by an ad hoc parameter rather than by the data, and a common-mode systematic would survive the enlarged error bars. The authors should report the seven individual estimates H0_i and the χ² obtained before adding σ_int, and should justify the 24% level as a physical scatter (for example, arising from the γ and environment distributions) rather than as a normalization fudge.","section":"III, σ_int"},{"comment":"As written, the covariance C_{H0} = C_{Θ_SNe} + C_{D^Obs} + C_{D^Obs_{A,Δt}} adds matrices for quantities of different physical dimensions (Θ has units of km/s, D^Obs is dimensionless, and D^Obs_{A,Δt} has units of Mpc), and it treats the multiplicative relation of Eq. (10) as if it were additive. The correct linearized propagation of H0 ∝ Θ/(D^Obs D^Obs_{A,Δt}) requires Jacobian-weighted covariance matrices, or equivalently the computation should be done in log-space. Please correct Eq. (17), or clarify that the code propagates the covariance of the product in Eqs. (9)-(10) and present the correct expression.","section":"III, Eq. (17)"},{"comment":"The Einstein-radius training set is selected by excluding the dataset of ref. [77] after inspecting its Gaussian-process reconstruction, which the authors report as showing 'bigger uncertainties... due to an unknown trend with redshift'. Because ref. [77] is stated to contain all systems of ref. [58], choosing the training sample on the basis of the reconstructed output is a post-hoc selection that can bias the GP mean function; the authors should either adopt a pre-defined selection criterion or demonstrate that the inferred H0 and its uncertainty are stable when the [77] data are included.","section":"II.D"}],"minor_comments":[{"comment":"Typo: 'Einstien Radius dataset' should read 'Einstein radius dataset'.","section":"II.D"},{"comment":"The right-panel axis label 'Θ(z)[km/s/Mpc]' has unclear units and normalization; since Θ = H0 D_L has units of km/s, please verify the units and explain the 10^6 axis scaling.","section":"Fig. 2"},{"comment":"Typo: 'C −1_H Est 0 is is the inverse' should read 'is the inverse'.","section":"III"},{"comment":"The seven TDCOSMO systems' values of (z_l, z_s, D^Obs_{A,Δt}, γ_pl) are not reproduced; given their central role in Eqs. (9)-(10), a table of these inputs would improve reproducibility and would allow readers to assess the z_s and γ comparisons requested above.","section":"II.B / ref. [43]"},{"comment":"The second term in Eq. (13) should be identified as (5σ_aB)^2 J, where J is the all-ones matrix, reflecting the common shift induced by the intercept uncertainty a_B.","section":"Eq. (13)"},{"comment":"The statement 'the observed quantity 4 is independent of the Hubble constant value' is awkward; it should cite Eq. (4) explicitly. Also, the aperture conversion to σ0 is described in one line; a brief equation or a pointer to Table 1 of ref. [58] would help.","section":"II.A"},{"comment":"Please state whether any of the seven TDCOSMO lenses are also members of the 99 Einstein-radius systems; if so, their Einstein-radius data enter both the GP training and the time-delay product, and this partial dependence should be discussed.","section":"II.D"},{"comment":"The 'Keywords' and 'PACS numbers' fields are empty and should be completed.","section":"Form"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of the journal and the algebraic construction is elegant. My recommendation of major revision is driven by the representativeness of the GP reconstruction and the role of the fitted σ_int, both of which affect the central value and uncertainty of H0. The authors have a number of closely related papers using the same distance-duality framework (refs. [52, 53, 56] and several other self-citations); the incremental element here is the combination with Einstein-radius GPs, which is genuinely new, but the overlap of datasets and methodology with refs. [52, 56] should be stated more explicitly for the reader. The heavy reliance on the group's own prior results is worth monitoring but does not by itself affect my verdict."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does one clean thing: it uses the CDDR to combine TDCOSMO time-delay distances, a GP reconstruction of the Einstein-radius ratio D_Als/D_As, and Pantheon+ SNe to get H0 = 70.55 ± 7.44. The algebraic step in Eq. 9, isolating D_Al as the product of the reconstructed distance ratio and the time-delay distance, is genuinely neat, and the authors are honest that this does not arbitrate the Hubble tension.\n\nThe strong points: the SNe covariance is propagated carefully, the CDDR anchoring is a legitimate inverse-distance-ladder move, and the paper clearly states its limitations regarding mass-sheet degeneracy and anisotropy. On a first read the algebra checks out.\n\nThe soft spots are real. The GP treats D^Obs as a function of z_l alone, but Eq. 4 shows it depends on z_s and the power-law slope gamma. The seven TDCOSMO systems have their own z_s and gamma_pl values; if they sit systematically off the 99-system sample's average at those lens redshifts, every D_Al is shifted in the same direction and H0 is biased. The paper's defense—that source redshifts are clustered—is a citation to a CDDR test, not an explicit comparison. Fixing gamma = 2.1 without marginalizing over the TDCOSMO gamma_pl values is another missing layer.\n\nThe error budget is also softer than it looks. The sigma_int ≈ 24% is added after the fact to force chi2_red ≈ 1. That inflates the quoted uncertainty but cannot remove a common-mode offset, because it is not a physical model of the scatter. And the choice of the [58] catalog over [77] was made after seeing the reconstructed trends; that is post-hoc selection, though the authors do flag the instability.\n\nNone of this makes the paper wrong, but it does mean the central value is a demonstration rather than a measurement. The paper is a useful methodological existence proof, and it will become more relevant with the next generation of lens surveys. The right referee will ask for a two-dimensional GP or a marginalization over z_s and gamma, and for a sigma_int that is fit jointly with H0 rather than retrofitted.\n\nI would send this to peer review. It is a fair, honest paper with a real kernel, and the flaws are identifiable and fixable. I would not cite the central value in my own work, but I would bring it to a reading group as a good case study of GP-based distance reconstruction.","headline":"A clean CDDR-based H0 combination with an honest error budget, but the GP step over D_Als/D_As ignores source-redshift and mass-profile dependence, so I read the central value as a consistency check rather than a determination.","tokens_in":16624,"tokens_out":2151,"would_cite":false,"duration_ms":15823,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Combining seven time-delay lens distances, 99 Einstein-radius systems, and Pantheon+ supernovae through the cosmic distance duality relation yields a model-independent $H_0 = 70.55 \\pm 7.44$ km/s/Mpc.","keywords":["Hubble constant","cosmic distance duality relation","strong gravitational lensing","time-delay cosmography","Einstein radius","Gaussian process regression","Pantheon+ supernovae","model-independent cosmology"],"falsifier":"Split the 99 Einstein-radius systems by source redshift or by image multiplicity, reconstruct $D^{\\mathrm{Obs}}(z_l)$ in each subgroup, and compare the curves at the seven TDCOSMO lens redshifts; a shift between subgroups larger than the reported ~24% intrinsic scatter would show that evaluating a single smoothed curve at $z_l$ misses source-dependent information and biases $H_0$. A sharper test: for a lens with multiple measured images or a known source redshift, require the GP prediction to match the actual $D^{\\mathrm{Obs}}$ value, not just the smoothed mean.","tokens_in":15434,"feed_emoji":"🔭","tokens_out":8592,"duration_ms":65608,"temperature":0.7,"pith_summary":"The paper attempts to determine the Hubble constant without assuming any cosmological model, by chaining three datasets together: seven time-delay lens distances from the TDCOSMO collaboration, 99 Einstein-radius lens systems, and the Pantheon+ supernova sample. It computes each lens's angular diameter distance as the product of a Gaussian-process reconstruction of the Einstein-radius ratio at that lens redshift and the measured time-delay distance. It then anchors the supernova unanchored luminosity distance to this angular diameter distance through the cosmic distance duality relation, obtaining $H_0 = 70.55 \\pm 7.44$ km/s/Mpc at 68% confidence. The authors present this as a consistency check that sits between the Planck and SH0ES values, not as a resolution of the Hubble tension.","feed_headline":"Strong lenses plus supernovae yield H0 = 70.6 ± 7.4 km/s/Mpc","feed_subtitle":"A cosmology-free distance-duality route chaining lensing and supernovae lands between the rival H0 values.","key_machinery":"The load-bearing identity is the product in Eq. (9): the observed time-delay angular diameter distance $D^{\\mathrm{Obs}}_{A,\\Delta t}(z_l,z_s)$, which is cosmology-independent, is multiplied by $D^{\\mathrm{Obs}}(z_l) \\equiv D_{Als}/D_{As}$, the Einstein-radius distance ratio reconstructed from 99 systems by Gaussian Process regression as a smooth function of the lens redshift $z_l$. Evaluating that reconstruction at each of the seven time-delay lens redshifts produces $D_A^l$. The same regression machinery, applied to Pantheon+ apparent magnitudes and converted with a calibrated intercept $a_B$, gives the unanchored luminosity distance $\\Theta_{\\mathrm{SNe}}(z_l)$; Eq. (10) joins the two through the cosmic distance duality relation. The lens mass model is fixed to a power-law profile with $\\gamma = 2.1$, and the GP covariance is a squared-exponential kernel.","core_discovery":"The central claim is that the product $D_A^l = D^{\\mathrm{Obs}}(z_l)\\,D^{\\mathrm{Obs}}_{A,\\Delta t}(z_l,z_s)$ gives a cosmology-independent angular diameter distance to each time-delay lens, and that the cosmic distance duality relation $D_L = (1+z)^2 D_A$ then turns unanchored supernova distances into a per-lens $H_0$ through $H_0 = \\Theta_{\\mathrm{SNe}}(z_l)/[(1+z_l)^2 D_A^l]$. Applied to the seven TDCOSMO systems, the seven model-free estimates combine in a MCMC posterior, with an added intrinsic scatter of about 24%, to give $H_0 = 70.55 \\pm 7.435$ km/s/Mpc at 68% confidence. The paper emphasizes that this value agrees within $1\\sigma$ with both the early-universe Planck value and the local SH0ES value, so the method is a viable independent consistency probe even though it does not resolve the Hubble tension.","pith_inferences":["Beyond the paper, the Gaussian-process assumption can be tested directly by splitting the 99 Einstein-radius systems by source redshift and comparing the reconstructed curves; a systematic offset at the seven TDCOSMO lens redshifts would propagate straight into $H_0$.","Beyond the paper, the same product identity could be crossed with absolute distance anchors other than supernovae, such as gravitational-wave standard sirens, to see whether the CDDR consistency holds across distance indicators.","Beyond the paper, the seven individual $H_0$ estimates should be published separately; if they show a redshift trend, that is an early warning of CDDR violation or residual lens-model systematics that the combined value hides."],"forward_implications":["If the central claim holds, every strong-lensing Einstein-radius system becomes a distance indicator: the GP-reconstructed $D^{\\mathrm{Obs}}(z)$ curve, once validated, can supply angular diameter distances at arbitrary lens redshifts without fixing a cosmology.","The same pipeline applied to larger lens samples, such as those expected from next-generation surveys, should sharpen the $H_0$ posterior directly, since the dominant uncertainty in the current result comes from only seven time-delay lenses.","The result's $1\\sigma$ agreement with both Planck and SH0ES means the inverse-distance-ladder route is currently a consistency check, not a tension discriminator; its real test will come from reducing the reported ~24% intrinsic scatter.","Because the method avoids anchoring to the CMB sound horizon or the distance ladder, it offers a third, independent rung for cross-checking $H_0$ determinations."],"supporting_citations":[{"why":"Supplies the seven TDCOSMO time-delay angular diameter distances and the power-law model table used in Eq. (8).","marker":"[43]"},{"why":"Provides the 99 Einstein-radius strong lensing systems whose $D^{\\mathrm{Obs}}$ values are reconstructed by Gaussian process.","marker":"[58]"},{"why":"Supplies the Pantheon+ supernova sample used to reconstruct unanchored luminosity distances.","marker":"[57]"},{"why":"Derives the CDDR-based formula (Eq. 2) that anchors $H_0$ to $D_A$ and $\\Theta_{\\mathrm{SNe}}$.","marker":"[50]"},{"why":"Provides the calibrated intercept $a_B$ that converts Pantheon+ apparent magnitudes into $\\Theta_{\\mathrm{SNe}}$.","marker":"[99]"},{"why":"Gaussian process regression implementation used to reconstruct $D^{\\mathrm{Obs}}(z_l)$ and $\\Theta_{\\mathrm{SNe}}(z_l)$.","marker":"[102]"},{"why":"Background theory of Gaussian process regression underlying the reconstructions.","marker":"[101]"}],"fun_headline_variants":["Model-free H0=70.6±7.4 from lensing+supernovae","Distance duality links lenses & SNe: H0=70.6±7.4","No-cosmology H0: 70.6±7.4 from seven lenses and SNe","Lensing+SNe, no model: H0=70.6±7.4, tension unchanged","H0=70.6±7.4: model-independent via distance duality"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Gaussian-process curve built from 99 Einstein-radius ratios, treated as a smooth function of lens redshift, correctly predicts $D_{Als}/D_{As}$ for each of the seven time-delay lenses, even though that ratio also depends on each system's source redshift and mass profile.","fun_headline_variants_meta":{"raw":{"variants":["Model-free H0=70.6±7.4 from lensing+supernovae","Distance duality links lenses & SNe: H0=70.6±7.4","No-cosmology H0: 70.6±7.4 from seven lenses and SNe","Lensing+SNe, no model: H0=70.6±7.4, tension unchanged","H0=70.6±7.4: model-independent via distance duality"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001166,"raw_usage":{"total_tokens":4865,"prompt_tokens":1026,"completion_tokens":3839,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":642,"completion_tokens_details":{"reasoning_tokens":3717}},"tokens_in":642,"tokens_out":3839,"duration_ms":23827,"temperature":1.0,"reasoning_tokens":3717,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:50:02.743108+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Split the 99 Einstein-radius systems by source redshift or by image multiplicity, reconstruct $D^{\\mathrm{Obs}}(z_l)$ in each subgroup, and compare the curves at the seven TDCOSMO lens redshifts; a shift between subgroups larger than the reported ~24% intrinsic scatter would show that evaluating a single smoothed curve at $z_l$ misses source-dependent information and biases $H_0$. A sharper test: for a lens with multiple measured images or a known source redshift, require the GP prediction to match the actual $D^{\\mathrm{Obs}}$ value, not just the smoothed mean.","supporting_citations":[{"cited_title":"A Hubble Constant Determination Through Quasar Time Delays and Type Ia Supernovae","cited_arxiv_id":"2503.06189","evidence_quote":"Provides the 99 Einstein-radius strong lensing systems whose $D^{\\mathrm{Obs}}$ values are reconstructed by Gaussian process."}],"review_version":1}