Two-rung ladder: $H_0$ from Tip of the Red Giant Branch and geometric anchors alone
Two-rung ladder: $H_0$ from Tip of the Red Giant Branch and geometric anchors alone
We infer the Hubble constant from the ~400 Extragalactic Distance Database hosts with a tip of the red giant branch (TRGB) distance, calibrated on the geometric anchors of the Large Magellanic Cloud and NGC 4258. We forward-model the tip magnitudes and host redshifts including magnitude-limited selection, sky exposure, and density-dependent galaxy bias, conditioning on each of the 80 Manticore-Local realisations of the local density and peculiar-velocity fields in turn. As most hosts lie at km/s, where peculiar velocities are important, the inference is…
Community Review
This is a community review of an externally published paper. The original authors retain all rights to their work. TOE-Share provides independent AI analysis — full content is available at the original source linked below.
3.7 out of 5. Audited on TheoryOfEverything.ai. Permanent record: https://theoryofeverything.ai/review-profile/paper/647fab59-1657-473a-94f7-335b592da884
AI Review Rating
Composite of the review dimensions below, on a 0–5 scale.
Consensus round triggered on 2 dimensions
Resolved: 1 - Still contested: 1
This is a well-constructed empirical distance-ladder paper testing whether a supernova-free TRGB ladder, conditioned on Manticore-Local density/velocity reconstructions, can arbitrate the Hubble tension. All specialists agree the honest, self-limiting conclusion (H0 cannot yet be pinned down because field-to-field reconstruction scatter of 5.2 km/s/Mpc dominates over the ~1 km/s/Mpc sensitivity to redshift-likelihood and smoothing choices) is well supported by the reported diagnostics, and the paper's candour about Neff=1.00 out of 80 realisations for its evidence-weighted headline (72.3±2.6 km/s/Mpc) is a genuine strength repeatedly praised across all three specialist tracks. The mathematical core—Eq. 5's r^2 Jacobian, the conjugate-Gaussian colour marginalisation (Eqs. 18-21), the Student-t redshift likelihood (Eq. 14), the field-marginalisation identities (Eqs. 29-31), and multiple numerical spot-checks (Maxwell prior conversion, Dirichlet fractional scatter, quadrature error propagation)—was independently verified by multiple math specialists and found sound.
However, the panel surfaced one load-bearing and one secondary mathematical concern that the reader should weigh explicitly. First, a HIGH-severity, source-verified risk flag (confirmed both algebraically and against the exact source expression by gpt-6-sol, and independently corroborated by two other math specialists) identifies that the selection-normalisation in Eqs. 25-28 treats the bright-end truncation as a soft cumulative-normal edge with width σ_sel=0.93 mag, even though Section 2 constructs the retained catalogue with a deterministic hard cut m_obs>22.1. At m_obs=22.0, the soft-edge model assigns roughly 44% detection probability to a magnitude that is excluded from the catalogue with certainty, meaning the detected-fraction denominator in Eq. 25 does not exactly normalise the likelihood for the data actually analysed. Appendix B's mock validation cannot test this specific mismatch because the mocks impose membership with the identical soft window. The panel's assessment of severity diverged: two specialists treated this as capping internal consistency and validity at 2/5, while others (claude-opus-5, DeepSeek) argued the effect is confined to a narrow magnitude range far from the operative faint-end roll-off (m_lim=24.07) and does not touch the paper's central claim (the 5.2 km/s/Mpc field-to-field scatter), landing at 3-4/5. This is a genuine, unresolved disagreement about magnitude of impact rather than about whether the mismatch exists, and readers should note both that the issue is real and that its quantitative effect on H0 has not been tested by the authors.
Second, a MEDIUM risk flag on Eq. 24's finite Simpson-quadrature evaluation of the distance-marginalised host likelihood notes the absence of any reported grid-refinement or endpoint-extension convergence test; since every host likelihood and hence the H0 posterior depends on this integral, this is a patchable but unaddressed numerical-rigor gap. A further concern raised by claude-opus-5 (not independently corroborated as an 'error' but worth flagging as a verification target) is that the Section 5.1 implied-RMS conversion σ_v·√(ν/(ν-2)) is evaluated at ν=2.59±0.48, close enough to the ν=2 singularity that the same formula at the -1σ point (ν≈2.11) would give a substantially larger RMS than quoted, with no uncertainty propagated through this nonlinear, near-singular map. This affects a secondary discussion point, not the headline H0 values.
On completeness, evidence, and communication the panel is largely aligned and positive: the sources specialists converge on 4-5/5, noting the fully worked generative model, extensive Appendix A priors, 1020-mock validation with a KS calibration check, and candid disclosure of untested residual selection effects (the 33-host redshift-availability cut near M81/Centaurus A, the posited rather than survey-derived selection function). The falsifiability score (4/5) reflects genuinely testable, quantitative claims (specific H0 values, Δlog Z model comparisons, a forecast that deeper TRGB data will reduce scatter below 5.2 km/s/Mpc), and novelty (3/5) reflects that the statistical architecture is inherited from the authors' own prior Cepheid-based framework (Stiskalek et al. 2026c/d), with the TRGB-specific contribution being the diagnostic finding that velocity-field variance rather than calibration dominates. Citation verification could not confirm most references, but no fabrication was found—these should be treated as identifiers needing confirmation, not integrity concerns.
This review was generated by AI for research and educational purposes. It is not a substitute for formal peer review. All analyses are advisory; publication decisions are based on numerical score thresholds.
The field-mixture definitions in Eqs. 29–32 are coherent, and the paper distinguishes the equal-weight stack from Bayesian field marginalization. However, the central inclusion event changes between Section 2's deterministic m_obs > 22.1 cut and Eq. 26's soft bright edge. Eq. 25 then conditions the H0 inference on the latter event without also conditioning on the former. Panel split 2, 2, 4, 4 across 4 math specialists. A consensus round did not resolve the disagreement, so the displayed score remains anchored to the conservative panel result.
The displayed mathematics is correct and largely reproducible. Eq. 5 follows from Eq. 4 by the stated spherical-coordinate change of variables with the delta-function angular likelihood, and the r^2 Jacobian is present. Eq. 28 is the standard Gaussian convolution with a difference of normal CDFs, giving the quadrature addition sqrt(sigma_sel^2 + sigma_m^2). Eqs. 20-21 are the correct conjugate-Gaussian posterior mean and variance for the latent colour. Eq. 14 has the correctly normalised Student-t form with scale sqrt(e_cz,i^2 + sigma_v^2) and reduces to the Gaussian as nu -> infinity. Eq. 30 follows from Eq. 29 with the 1/N_field cancellation, and Eq. 31 is Kish's ESS written in unnormalised form, algebraically identical to 1/sum w_k^2. Eqs. 34-36 correctly distinguish the log ratio of ensemble evidences from the mean per-field log-evidence ratio. Numerical spot-checks pass: the Student-t implied RMS sigma_v * sqrt(nu/(nu-2)) for (67, 2.59) gives 67 * sqrt(2.59/0.59) = 140.4, matching the quoted 141; for (57, 2.34), 57 * sqrt(2.34/0.34) = 149.6, matching 149. The Maxwell prior conversion a = mu/(2*sqrt(2/pi)) = 300/1.5958 = 188.0 km/s and mode sqrt(2)*a = 265.9 match the quoted 188 and 266. The Dirichlet fractional scatter sqrt((N_pix-1)/(kappa+1)) gives sqrt(11/49) = 0.474 at N_side=1 and sqrt(47/193) = 0.493 at N_side=2, both ~0.5 as claimed. Two genuine limitations keep this below 5. First, the soft bright-edge window in Eq. 26 assigns nonzero inclusion probability below the hard cut m_obs > 22.1 used to construct the sample, so the detected-fraction denominator in Eq. 25 does not normalise the likelihood for the actual retained sample; the effect on H0 has not been quantified, and the mock catalogues generated with the same soft window cannot test this mismatch. Second, the evidence-weighted headline is a Monte Carlo estimator of Eq. 29 evaluated at N_eff = 1.00 effective samples, so its sampling variance is uncontrolled; the paper is explicit about this and supplies the equal-weight stack as a bracket, but the 72.3 +/- 2.6 number is an estimator applied far outside its convergent regime. The finite Simpson grid for Eq. 24 lacks a reported convergence study, which is a patchable numerical-rigor issue. None of these is a fundamental analytic error or a circular derivation, so the score is 4 rather than lower. A consensus round resolved an earlier panel split before this score was finalized.
Empirical rubric applied. The paper makes specific, quantitative, testable claims: H0=68.4±5.9 (equal-weight) vs 72.3±2.6 (evidence-weighted) km/s/Mpc, field-to-field scatter of 5.2 km/s/Mpc, a velocity monopole shift of 5.4±3.4 km/s/Mpc, and explicit model-comparison evidences (Δlog Z) between competing likelihood/selection/smoothing choices. Crucially, the authors state clear falsification-relevant claims: that extending TRGB data beyond ~1000 km/s and improving local reconstructions should reduce the 5.2 km/s/Mpc scatter, and that a Student-t likelihood is decisively preferred over Gaussian by evidence — both testable with near-term surveys (e.g., JWST-extended TRGB catalogues, next-generation constrained reconstructions) or via reanalysis. The paper is unusually transparent about what would break its own central number (Neff=1.00 out of 80 realisations), effectively self-falsifying the precision of its headline evidence-weighted value. This is a strong exemplar of falsifiable, quantitative work; it falls short of 5 because the central H0 estimate itself is explicitly acknowledged as not yet capable of arbitrating the Hubble tension, i.e., not yet a decisive test of competing cosmological models, only of methodological choices within the ladder itself.
The paper is well organized with a logical progression from data selection, generative model, selection modelling, field marginalisation, to results and discussion. Notation is introduced before use and used consistently throughout the visible sections (e.g., r_i, c_i, Λ, δ(x), V(x), θ_q). The distinction between evidence-weighted and equal-weight treatments is clearly and repeatedly explained, and the physical reasoning behind discrepancies (Neff=1.00, dominant realisation) is stated plainly rather than buried in equations. Minor deductions: some equations appear truncated or fragmentary in the condensed view (likely an artifact of extraction rather than the source), and a graduate reader unfamiliar with Bayesian Origin Reconstruction from Galaxies or HEALPix conventions would need to consult cited works for full technical grounding, but the prose itself is clear and self-consistent throughout the exposed material.
The paper extends an existing hierarchical Bayesian, field-conditioned distance-ladder framework (previously applied to Cepheids by the same group, Stiskalek et al. 2026c/d) to the TRGB indicator, and combines it with the Manticore-Local density/velocity reconstruction and an explicit treatment of angular sky-exposure and magnitude-based selection specific to the archival EDD/TRGB catalogue. This is a meaningful and non-trivial extension — applying the same rigorous forward-modelling machinery to a stellar indicator with markedly different systematics (dust, metallicity, halo tracer) and a much lower typical redshift range, which surfaces a genuinely new finding: that for TRGB (unlike Cepheids) the local peculiar-velocity field variance, not calibration, dominates the error budget, and that the evidence-weighted stack collapses onto a single field realisation (Neff=1.00/80). This diagnostic result about ensemble marginalisation is a novel methodological insight. However, the core statistical architecture (generative model, selection modelling, field marginalisation via evidence weighting) is inherited wholesale from the authors' prior Cepheid paper, so the mechanism itself is not new — the novelty lies in application and in the sobering finding about reconstruction-driven bias, not in a new theoretical or inferential mechanism.
The paper is substantively complete for its stated empirical goal. It identifies the host and anchor samples and exclusions; defines a hierarchical likelihood linking the TRGB calibration, geometric anchors, host distances, redshifts, selection, galaxy bias, and reconstructed fields; states priors and numerical inference procedures; compares relevant model variants; and supplies both posterior-predictive and mock-recovery checks. It explicitly addresses important boundary/limiting cases, including the bright-end cut, Galactic-plane mask, Gaussian versus heavy-tailed redshift residuals, density smoothing, angular exposure, a velocity monopole, and a no-reconstruction control. It also carefully distinguishes the formal evidence-weighted field marginalisation from the conservative equal-weight diagnostic, and does not overclaim what either can establish when the evidence weights have Neff=1.00.
The remaining gaps are material but do not break the core argument. The archival selection is necessarily posited rather than derived from a survey design, and the authors explicitly acknowledge unmodelled dependence on host properties and do not test the non-random redshift-availability depletion near the M81 and Centaurus A groups. The posterior-predictive check is evaluated at the single evidence-dominant field, while the evidence-weighted result itself is acknowledged to be sensitive to finite reconstruction sampling. Finally, the mocks validate recovery only under matched generation and recovery assumptions, explicitly excluding field mismatch and real-catalogue selection misspecification. These limitations are candidly stated and appropriately constrain the conclusions, so they warrant a 4 rather than a 5.
Key Equations (2)
Combines cosmological expansion and line-of-sight peculiar velocity to predict each host redshift.
Evidence-weighted mixture of parameter posteriors conditioned on individual local density–velocity field realisations.
Other Equations (2)
Linear colour standardisation of the TRGB absolute magnitude.
Equal-weight stack of the reconstruction-conditioned posteriors, used to quantify field-to-field sensitivity.
Testable Predictions (2)
Extending TRGB distance measurements beyond the present local sample, together with improved reconstructions of the local density and velocity fields, is expected to reduce the field-to-field scatter in inferred H₀ below the 5.2 km s⁻¹ Mpc⁻¹ found here.
Falsifiable if: A deeper TRGB sample analysed with improved local-Universe reconstructions fails to reduce the reconstruction-to-reconstruction scatter in H₀ below 5.2 km s⁻¹ Mpc⁻¹.
For the current nearby TRGB host sample, the residual velocity distribution is strongly non-Gaussian and a Student-t redshift likelihood is preferred over a Gaussian likelihood by the model evidence.
Falsifiable if: A reanalysis or independent comparable sample yields no evidence preference for heavy-tailed residuals over a Gaussian likelihood.
Tags & Keywords
Keywords: Hubble constant, tip of the red giant branch, distance ladder, peculiar velocities, local density field, Bayesian hierarchical inference, Malmquist bias, Hubble tension
Full content is available at the original source:
arxiv.org/abs/2609.29996You Might Also Find Interesting
Semantically similar papers and frameworks on TOE-Share