pith. sign in

arxiv: 2506.11299 · v2 · submitted 2025-06-12 · 🌌 astro-ph.CO · gr-qc

Geometric Cosmology models: statistical analysis with observational data

Mat\'ias Leizerovich , Luisa G. Jaime , Susana J. Landau , Gustavo Arciniega This is my paper

Pith reviewed 2026-05-06 17:54 UTC · model claude-opus-4-7

classification 🌌 astro-ph.CO gr-qc PACS 98.80.-k04.50.Kd98.80.Es
keywords modified gravityhigher-curvature gravitygeometric inflationGILA modelcosmological parameter estimationcosmic chronometersType Ia supernovaeglobular cluster ages
0
0 comments X

The pith

Adding an exponential tower of higher-curvature terms to gravity yields three late-time cosmologies, two of which are ruled out by current data and the third of which fits but is statistically disfavored compared to ΛCDM.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper asks whether a gravity theory built from an infinite tower of higher-curvature invariants — without any cosmological constant or scalar field — can reproduce the late-time expansion history we actually observe. The authors restrict the resummed Friedmann function F(H) to an exponential form, which produces three distinct cosmological cases depending on how the leading Ricci term enters. They confront each case with Type Ia supernovae, cosmic chronometer measurements of H(z), and a 12.7 Gyr lower bound on the age of the Universe drawn from old globular clusters. Two of the three families fail outright; only three GILA configurations (with curvature exponents r=3 and s∈{4,5,6}) survive all three datasets simultaneously. Even these survivors lose to ΛCDM in standard model-comparison criteria by a wide margin (ΔAIC ≈ 37–39). A second contribution is methodological: equation stiffness and the irregular geometry of the age-prior region force the authors to replace MCMC with a direct grid-based Monte Carlo sampler, providing a template for testing other modified-gravity models against age constraints.

Core claim

The authors take a gravity theory built by adding an infinite tower of higher-curvature invariants to the Einstein-Hilbert action, restrict to the case where that tower sums to exponential functions of the Hubble rate, and read off three distinct late-time cosmologies: GILA, GR-deformation, and a non-GR-contribution case. They then confront each with Type Ia supernovae, cosmic chronometers, and a lower bound on the age of the Universe set by the oldest globular clusters. The GR-deformation family and several non-GR cases are ruled out — either because supernova and chronometer data prefer inconsistent parameter regions, or because the predicted Universe is younger than 12.7 Gyr. Only three G

What carries the argument

A modified Friedmann equation 3F(H)=κρ in which F(H) is the resummation of an infinite series of higher-curvature contributions. Choosing the series to converge exponentially produces three distinct late-time forms — depending on whether the linear Ricci coefficient α₁ vanishes (GILA), lies in (-1,0) (GR-deformation), or equals -1 (non-GR contribution). Confronting these forms with combined supernova, chronometer, and globular-cluster age data via a custom Monte Carlo grid sampler (replacing MCMC because of equation stiffness and the awkward geometry of the age prior) is what actually does the discriminating work.

If this is right

  • The exponential resummation is not enough on its own — alternative analytic forms of F(H) need to be searched if the higher-curvature program is to remain a serious competitor to ΛCDM at late times.
  • Any modified-gravity model whose preferred parameter region predicts a Universe younger than ~12.7 Gyr is in trouble before any supernova or chronometer fit is even performed; globular-cluster ages deserve to be a routine prior, not an afterthought.
  • The GR-deformation case — where the effective Newton constant is rescaled by (1+α₁) — is incompatible with cosmological age data even at α₁ as small as 10⁻⁵, sharpening solar-system bounds on this kind of deformation.
  • The grid-based Monte Carlo methodology developed here is a reusable template for testing other stiff or awkwardly-prior'd modified-gravity scenarios where MCMC chains fail to converge.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The 38-unit ΔAIC gap is large enough that no reasonable reweighting of the globular-cluster prior would close it; the surviving GILA cases are 'allowed' but not competitive, which suggests the exponential ansatz is the wrong functional form rather than the wrong parameter values.
  • Because the radiation density and early-time coupling λ are fixed by hand, this analysis cannot detect any modified-gravity signature that lives primarily at recombination — combining with CMB likelihoods would be the real test of the framework.
  • The methodology of treating an age-of-Universe lower bound as a hard uniform prior, rather than folding stellar-age uncertainties into a Gaussian, is conservative for ruling models in but aggressive for ruling models out; propagating the ~0.5 Gyr uncertainty would likely soften several of the exclusions.
  • The fact that all three viable GILA survivors share r=3 hints that the lowest allowed curvature order is doing most of the dynamical work, with s controlling only the sharpness of the late-time turn-on — a redundancy worth exploiting in future model-building.

Load-bearing premise

That fixing the early-time coupling and the late-time energy scale by hand, and choosing the curvature series specifically so that it sums to exponentials, leaves a fair sample of the theory rather than a narrow slice — so that "ruled out" really means the model class is dead, not just this configuration of it.

What would settle it

A direct numerical demonstration that one of the surviving GILA cases (e.g. r=3, s=5) achieves a total χ² within the AIC/BIC threshold of ΛCDM on the same combined SNIa + cosmic chronometer + globular-cluster data, or conversely an independent age-of-Universe lower bound from globular clusters that shifts the threshold by more than ~1 Gyr in either direction, would directly settle whether the surviving GILA slice is genuinely viable or has been rescued only by the chosen 12.7 Gyr cutoff.

read the original abstract

Although the standard cosmological model is capable of explaining most current observational data, it faces some theoretical and observational issues. This is the main motivation for exploring alternative cosmological models. In this paper, we focus on a novel proposal that consists in adding an infinite tower of higher-order curvature invariants to the usual Einstein-Hilbert action. We obtain the late-time background evolution for three families of models that can be obtained from this proposal. We use recent data from Cosmic Chronometers and type Ia supernovae to test the late-time predictions of our models. In addition, we consider estimations from the Age of the Older Globular Clusters to constrain our models. While some of the studied cases are ruled out by the data, we show that there are particular cases of the GILA model that can explain current data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

5 major / 8 minor

Summary. The authors test three families of "Geometric Cosmology" (GC) models — GILA, GR-deformation, and non-GR contribution — built from an infinite tower of higher-curvature invariants whose Friedmann-equation kernel F(H) is taken in an exponential convergent form. They restrict to late-time behaviour by setting λ=0, fix the late-time energy scale L̃ (or β) by hand, and confront the resulting one- or two-parameter (β, H₀, M_abs) families with Pantheon+SH0ES, cosmic chronometers, and a hard 12.7 Gyr lower bound on the age of the Universe motivated by the oldest globular clusters. Because of stiffness in the ODE for H(z) and a non-trivial prior region from the AoU cut, the parameter space is sampled on a 100³ grid (Monte Carlo, not MCMC). They conclude that the GR-deformation family and the non-GR-contribution family considered are excluded by the data, that three GILA cases (r=3, s∈{4,5,6}) survive all data, and that ΛCDM is nonetheless statistically preferred over those surviving GILA cases with ΔAIC≈37–39 and comparable ΔBIC.

Significance. The paper makes a defensible attempt at a problem that has not been done systematically before for this family of theories: combining recent SNIa, CC, and globular-cluster age constraints on a higher-curvature gravity ansatz that admits a closed-form F(H). Three concrete contributions are worth noting: (i) the explicit derivation of the late-time background equations for the GILA, GR-deformation, and non-GR-contribution families (Sec. 2 and Appendix A); (ii) a careful and honest treatment of the AoU prior, including the choice to use a uniform prior with a hard cut rather than a Gaussian, with footnote 8 correctly noting that propriety of the posterior — not the prior — is what matters; and (iii) public release of code (GC-MC) and data, allowing reproduction. The results themselves are negative-leaning (ΛCDM preferred), but the paper is upfront about that and frames the work as establishing a methodology and ruling out one specific functional choice of F(H), which is a legitimate framing. The astrophysical importance is modest — no surviving model becomes preferred — but the methodological scaffolding (in particular how to integrate the globular-cluster age bound into a Bayesian co

major comments (5)
  1. [Sec. 4 (Methodology), and Table 2] The headline quantitative result — ΔAIC ≈ 37–39 in favour of ΛCDM for the three surviving GILA cases — is computed with L̃ fixed at 0.90 H₀⁻¹ by hand. With λ=0 and L̃ fixed, GILA and ΛCDM have the same number of free parameters (M_abs, H₀, and one of β or ω_m), so ΔAIC reduces to Δχ². A Δχ² ≈ 37 between best-fit GILA and best-fit ΛCDM on CC+PPS is a very large number for models that the authors describe as 'viable'; it warrants a profile likelihood χ²(L̃) = min_{β,H₀,M_abs} χ² evaluated over a reasonable range (e.g. L̃ ∈ [0.5, 1.5]) to demonstrate whether L̃=0.90 is near the χ² minimum or whether the apparent disfavouring of GILA is largely an artefact of the hand-chosen scale. The justification for fixing L̃ — a 'correlation between L̃ and β' from the series expansion (Eqs. A33, A47–A48) — is a statement about a flat direction in the prior, not a justification for fixing a parameter whe
  2. [Sec. 4, step 8 and footnote 11] The criterion used to rule out a family on the basis of the AoU prior — that χ²_min over surviving grid points must lie inside the 68.3% confidence region in the full 3D parameter space, equivalent to the 98.3% level in 1D — is unusual and load-bearing for the exclusion of GR-deformation and several non-GR-contribution cases. The text labels this 'conservative when ruling out a family', but in fact the criterion mixes a frequentist Δχ² threshold with the dimensionality of the nuisance space. Please justify this choice, ideally by comparing with the Bayesian alternative — i.e. computing the Bayes factor or evidence ratio with the AoU>12.7 Gyr prior multiplied in — since the authors otherwise frame the analysis in Bayesian language. Reporting min χ²/d.o.f. on the AoU-restricted grid for each rejected family would also help the reader gauge how marginally these families fail.
  3. [Sec. 3.3 and Sec. 5.1] The 12.7 Gyr lower bound on the age of the Universe is constructed from a 12.2 Gyr GC age plus a 0.5 Gyr star-formation buffer, both with non-negligible uncertainty. The authors assert robustness over AoU_th ∈ [12, 13] Gyr but do not show this — please add a figure or table demonstrating how the surviving (r,s) and the constraints in Table 4 shift over that range, since some of the model exclusions in Table 3 may not survive the lower end. More broadly, treating an inherently noisy astrophysical quantity as a hard step-function prior is sharp; a soft cut (e.g. an erf or one-sided Gaussian beyond the threshold) would be a more honest representation of the underlying uncertainty and is straightforward to implement in the same Monte Carlo framework.
  4. [Sec. 4 and Table 2] Ten (r,s) pairs are sampled in the GILA case (Table 2) and three are kept; the rest are dropped via the AoU cut. The model-comparison numbers in Table 2 do not appear to penalise this discrete model selection. Either (i) (r,s) should be treated as discrete model labels and the comparison repeated using a marginalised evidence over the (r,s) grid, or (ii) the text should state explicitly that ΔAIC/ΔBIC are conditional on a post-hoc choice of (r,s), with the implication that the true penalty against GILA is at least as large as quoted. Currently the wording in Sec. 5.1 ('we were able to estimate confidence intervals for the free parameters') reads as if (r,s) were free physical parameters that were constrained, when in fact they were enumerated and selected.
  5. [Sec. 4, choice of Monte Carlo over MCMC] The motivation for abandoning MCMC — 'stiffness' of Eq. (17) and 'complex geometry of the allowed prior region' — is plausible but not demonstrated. A 100³ uniform grid in 3D is feasible here, but the same conclusion would follow more transparently from a nested-sampling run (e.g. dynesty/MultiNest), which handles hard prior boundaries and stiff likelihoods routinely and additionally yields the Bayesian evidence directly, sidestepping the criticism in major comment 2 above. At minimum, the authors should report the typical CPU time per H(z) integration that drove the choice, and verify that the grid-level marginalised posteriors agree with at least one independent sampler on a tractable subspace (e.g. ΛCDM, where MCMC obviously works).
minor comments (8)
  1. [Eq. (17)] The numerator on the RHS is κ(ρ_tot+P_tot), with ρ_tot = ρ_r + ρ_m. Please state explicitly that pressureless matter contributes only via ρ_m and that the geometric late-time term resides in F'(H) on the LHS — readers may otherwise expect a 'dark energy' term in the source.
  2. [Sec. 4, definition of ω_m] ω_m is defined as κ/(3·100²) ρ_m,0 and then used in Eq. (28) as F(H₀)/100² − ω_r. The convention that 100 here means 100 km/s/Mpc rather than the dimensionless h should be stated, to avoid ambiguity with the more common ω_m = Ω_m h².
  3. [Table 4] The 68%/95% intervals for β in the (r,s)=(3,5) row are +1.2/−2.0 (68%) and +3.1/−2.8 (95%), with a mean 3.72 and a prior range [0,12]. Please indicate whether the lower 95% bound touches the prior edge β=0 (i.e. whether GILA → GR is excluded by the data) and quote the corresponding Δχ² at β=0.
  4. [Fig. 1 and Fig. 2] Fig. 1 shows a 2D heat map at fixed M_abs but the methodology applies in 3D; the caption could note this explicitly. Fig. 2 (panels a–d) is hard to parse; consider labelling the 'surviving region A' more clearly and explaining what the contour levels in panels (b) and (d) represent after the cut.
  5. [Abstract / Sec. 6] The phrasing 'two entire families of GC models … are ruled out' (abstract) is stronger than the body warrants: only specific (r,s) and only with the s≥2, β=10⁻⁵ (GR-def) or β=1 (non-GR) slices were tested. Please soften to 'the families with the choices of exponents and energy scales considered here'.
  6. [Appendix A] Equations (A47)–(A48) impose ξ̃(j·q) = λ^j L^{2jq}/j! by hand to obtain the exponential convergence; this is the Taylor coefficient of an exponential by construction. A short remark that this is a choice (not derived from any field-theory consistency requirement) would help the reader assess how special the exponential ansatz is.
  7. [Sec. 3.1] The treatment of nuisance parameters (α, β_SN, γ) being fixed to ΛCDM-derived values is justified by appeal to Refs. [27,28]; one sentence noting that those references covered f(R) and MOG, not GILA-type higher-curvature models, would be appropriate.
  8. [Throughout] Several sentences need light copy-editing, e.g. 'cannot cannot meet the globular cluster constraint' (Sec. 4, step 8 description); 'the data are not able to constrain H₀ beyond what is imposed by the prior. However, the data are informative enough to constrain higher values of H₀' (Sec. 5.1) is internally contradictory as written.

Simulated Author's Rebuttal

5 responses · 1 unresolved

We thank the referee for a careful and constructive report. The five major comments converge on three legitimate methodological weaknesses in the present manuscript: (i) the dependence of our headline model-comparison numbers on a hand-fixed late-time scale L̃ and on a post-hoc enumeration of the integer exponents (r,s); (ii) the use of a non-standard hybrid frequentist/Bayesian criterion to rule out entire families, in tension with our otherwise Bayesian framing; and (iii) insufficient demonstration of robustness to the choice of the AoU threshold and to the sampler. We accept all three concerns and will address them in revision through (a) a profile likelihood in L̃ for the surviving GILA cases, (b) a fully Bayesian recomputation of the model comparison via nested sampling and explicit Bayes factors with the AoU prior built in, (c) an AoU_th-robustness table and a soft (erf-shaped) AoU prior as the primary analysis, (d) explicit treatment of (r,s) as discrete model labels with marginalised evidence, and (e) a nested-sampling cross-check of the grid-based posteriors. We retain the framing of the paper as a methodology contribution and an exclusion of one specific functional form of F(H), but we will state clearly where the revised quantitative conclusions differ from the present version, including the possibility that some currently excluded families are rehabilitated under softer or lower AoU bounds.

read point-by-point responses

  1. Referee: ΔAIC≈37–39 in favour of ΛCDM is computed with L̃ fixed at 0.90 H₀⁻¹ by hand. With λ=0 and L̃ fixed, GILA and ΛCDM have the same number of free parameters, so ΔAIC reduces to Δχ². A profile likelihood χ²(L̃)=min_{β,H₀,M_abs} χ² over e.g. L̃∈[0.5,1.5] is needed to determine whether L̃=0.90 is near the χ² minimum or whether the disfavouring is an artefact of the hand-chosen scale. The 'L̃–β correlation' argument is a flat-direction-in-the-prior statement, not a justification for fixing a parameter.

    Authors: We agree this is a fair concern and that our justification for fixing L̃ was insufficiently precise. The choice L̃=0.90 H₀⁻¹ was motivated physically (the late-time modification should manifest near the present horizon scale, L̃∼1) and follows the convention of Refs. [17,18], but the referee is correct that this does not by itself address whether L̃=0.90 sits near the χ² minimum. For the revised version we will add a profile-likelihood analysis χ²_prof(L̃)=min_{β,H₀,M_abs} χ²_{CC+PPS}(L̃,β,H₀,M_abs) for the three surviving GILA cases (r=3, s∈{4,5,6}) over L̃∈[0.5,1.5], with the AoU cut imposed, and present the resulting curve as a new figure together with an updated Δχ² (and hence ΔAIC) at the L̃ that minimises the profile. We note in advance, however, that the L̃–β degeneracy seen in the series expansion (Eqs. A33, A47–A48) is not merely a prior-volume effect: it is an algebraic redundancy in the parametrisation of F(H), so portions of the L̃ direction are reabsorbed into β at fixed F(H). We will make this distinction explicit and, where the profile shows a genuinely lower χ² at a different L̃, we will quote the corresponding revised Δχ²/ΔAIC and update the conclusions of Sec. 5.1 and Table 2 accordingly. If the headline ΔAIC drops materially, we will revise the qualitative claim that ΛCDM is preferred. revision: yes

  2. Referee: The step-8 criterion (χ²_min on AoU-surviving grid must lie inside the 68.3% region in 3D, equivalent to 98.3% in 1D) is unusual and load-bearing for the exclusions. It mixes a frequentist Δχ² threshold with the dimensionality of nuisance space. Justify, ideally by computing the Bayes factor with the AoU>12.7 Gyr prior, since the analysis is otherwise framed in Bayesian language. Reporting min χ²/d.o.f. on the AoU-restricted grid for each rejected family would help.

    Authors: We accept that the rationale for the step-8 criterion was underdocumented and that the framing was inconsistently Bayesian/frequentist. Our intent was conservative — to rule out a family only when no point compatible with the AoU prior lies within an extended (effectively 3σ in 1D) confidence region of the unrestricted CC+PPS posterior — but we agree the construction is non-standard. In revision we will: (i) tabulate min χ²/d.o.f. on the AoU-restricted grid for every (r,s) pair in the GILA, GR-deformation, and non-GR-contribution families considered, so the reader can see how marginally each rejected case fails; (ii) recompute the model selection in fully Bayesian terms by evaluating the evidence Z = ∫ L(θ)π(θ)Θ(AoU−AoU_th) dθ for each family and reporting Bayes factors B = Z_family/Z_ΛCDM, using nested sampling on the same parameter grid (see our reply to comment 5); and (iii) retain the Δχ²-based criterion only as an auxiliary diagnostic, with the Bayes-factor results as the primary basis for exclusion. If a previously rejected family is rehabilitated by the Bayes-factor analysis we will revise Tables 2–3 and the conclusions accordingly. revision: yes

  3. Referee: The 12.7 Gyr AoU bound (12.2 Gyr GC age + 0.5 Gyr buffer) has non-negligible uncertainty. The claim of robustness over AoU_th∈[12,13] Gyr is asserted but not shown; please add a figure/table showing how surviving (r,s) and Table 4 constraints shift over that range. Treating a noisy astrophysical quantity as a hard step prior is sharp; a soft cut (erf or one-sided Gaussian) would be more honest and is straightforward to implement.

    Authors: Both points are well taken. We will add to Sec. 5 a table summarising, for AoU_th∈{12.0, 12.2, 12.5, 12.7, 13.0} Gyr, (a) which (r,s) pairs in each family survive, and (b) the resulting 68%/95% intervals on (β, H₀, M_abs) for the GILA cases that remain viable. We expect — but will verify — that some currently rejected GR-deformation or non-GR-contribution cases are recovered at the lower end of the range; if so, we will state this explicitly rather than continuing to claim the families are excluded. We also agree that a hard step is a sharp idealisation. In revision we will repeat the analysis with a soft prior of the form π(AoU) ∝ ½[1+erf((AoU−μ)/√2 σ)] with μ=12.7 Gyr and σ chosen to encode the dominant systematic uncertainty in the GC age determination (we will adopt σ≈0.4 Gyr, motivated by the spread of recent GC age estimates and the uncertainty in the star-formation buffer; this choice will be stated and varied as a robustness check). The MC-grid framework already accommodates an arbitrary prior weight, so this is a one-line modification of the code, and we will release the updated GC-MC version alongside the paper. revision: yes

  4. Referee: Ten (r,s) pairs are sampled in GILA and three kept via the AoU cut; the model-comparison numbers in Table 2 do not penalise this discrete model selection. Either treat (r,s) as discrete model labels and marginalise evidence over the grid, or state explicitly that ΔAIC/ΔBIC are conditional on a post-hoc choice of (r,s). The wording 'we were able to estimate confidence intervals for the free parameters' reads as if (r,s) were continuous physical parameters.

    Authors: The referee is correct: (r,s) are integer model labels enumerated by hand, not continuously varying parameters that we constrained, and the current Table 2 ΔAIC/ΔBIC values are conditional on the post-hoc selection. In revision we will (i) explicitly relabel each (r,s) pair as a distinct model M_{r,s} and present a marginalised log-evidence ln Z̄ = ln Σ_{r,s} Z_{r,s}π(r,s) under a flat prior over the enumerated grid, with the corresponding global Bayes factor against ΛCDM; (ii) add a sentence in Sec. 4 and the caption of Table 2 stating that the per-(r,s) ΔAIC/ΔBIC are conditional and that the true model-level penalty is at least the quoted value plus 2 ln N_models for AIC-type comparisons across the discrete grid; and (iii) reword the Sec. 5.1 statement so that it reads 'for the three surviving (r,s) cases we estimate confidence intervals on (β, H₀, M_abs)', avoiding the implication that (r,s) themselves were inferred from the data. revision: yes

  5. Referee: Motivation for abandoning MCMC ('stiffness' and 'complex prior geometry') is plausible but not demonstrated. A nested-sampling run (dynesty/MultiNest) would handle hard prior boundaries and stiff likelihoods routinely and yield evidence directly. At minimum report typical CPU time per H(z) integration and verify that grid-level marginalised posteriors agree with an independent sampler on a tractable subspace (e.g. ΛCDM).

    Authors: We accept this. In revision we will (i) report the typical wall-clock cost of a single H(z) integration in the stiff regions of GILA parameter space (in our current implementation it is dominated by adaptive-step ODE evaluations near the exponential terms in F'(H) and ranges from ∼10 ms in benign regions to >1 s near the stiffness boundary, which is what made a million-step MCMC impractical for the full enumerated grid), and document the cases where standard MCMC failed to converge with concrete diagnostics (Gelman–Rubin, autocorrelation length). (ii) We will rerun the analysis with a nested sampler (we plan to use dynesty, which natively supports hard prior boundaries via the prior-transform interface). This addresses major comments 2 and 4 simultaneously by yielding ln Z directly. (iii) As a cross-check, we will compare grid-based marginalised 1D posteriors against the nested-sampling output on the ΛCDM subspace and on the (r,s)=(3,5) GILA case, and include the comparison plot in an appendix. The grid will be retained as an exploratory tool but the headline numbers in the revision will come from nested sampling. revision: yes

standing simulated objections not resolved
  • The five major comments are all addressable through additional analysis (profile likelihood in L̃, Bayes-factor recomputation, AoU robustness table and soft prior, marginalisation over the discrete (r,s) grid, and a nested-sampling cross-check). We anticipate that some of these revisions may quantitatively weaken the current exclusions of the GR-deformation and non-GR-contribution families and may reduce the headline ΔAIC against GILA; we cannot pre-commit to the qualitative conclusions surviving until the revised analysis is complete.

Circularity Check

1 steps flagged

No significant circularity: the GC→data comparison is an honest, if methodologically restricted, fit; concerns about fixing L̃ and (r,s) are robustness issues, not circular reasoning.

specific steps
  1. ansatz smuggled in via citation [Section 2 (after Eq. 4) and Section 4 ('the energy scale ˜L = 0.90 ... motivated by ... analyses previously carried out for the GILA model in Refs. [18] and [17]')]
    "This particular value of ˜L is motivated by the requirement that the effects of the theory manifest at late times; therefore, the associated energy scale is fixed close to its present value (˜L∼1). This choice is based on the analyses previously carried out for the GILA model in Refs. [18] and [17]."

    The fixed value L̃ = 0.90 H_0^{-1} and the exponential form of F(H) are adopted by reference to prior work by overlapping authors, where they are themselves ansatz choices rather than derived. This is mild — the paper is transparent that it is testing one functional form — but it means the ΔAIC ≈ 37–39 number is conditional on choices imported from self-citation rather than profiled over. Not load-bearing for circularity (the data fit is real), but worth noting.

full rationale

The paper's derivation chain is: (i) postulate an action with an infinite tower of higher-curvature invariants; (ii) adopt an exponential convergent form for F(H) (an ansatz, acknowledged as such and motivated by prior work); (iii) derive a modified Friedmann equation; (iv) fit free parameters (M_abs, H_0, β or L̃) to CC + PPS SNIa data with a globular-cluster age prior; (v) report χ², AIC/BIC vs ΛCDM. None of these steps redefines the data being predicted, renames a fitted parameter as a prediction, or invokes a self-citation as a load-bearing uniqueness claim. The exponential ansatz is attributed to Jaime & Arciniega [18] and Arciniega et al. [19], some of whom are co-authors, but the paper is explicit that this is one choice among possibilities ("the exponential convergent function F(H) is the one that has been most explored") and frames the conclusions as applying only to this functional form. No uniqueness theorem is imported. The skeptic's load-bearing attack — that fixing L̃ = 0.90 H_0^{-1} by hand inflates ΔAIC ≈ 37–39 because L̃ is not profiled over — is a legitimate concern about whether the comparison is fair, but it is a methodological/robustness criticism, not circularity. Fixing a parameter by prior choice and then computing Δχ² does not make the prediction equal to its input; it just narrows the model space being tested. The paper's justification for fixing L̃ (degeneracy with β from the series structure, Eqs. A33, A47–A48) is weak as a justification for AIC accounting, but the reported numbers are not tautological. Similarly, the post-hoc selection over (r,s) pairs and the 12.7 Gyr AoU bound (with embedded 0.5 Gyr star-formation buffer) affect which models are ruled out, but the rule is applied externally, not in a way that reduces "model passes" to "model fits." The one minor self-citation pattern is the adoption of the exponential F(H) and the L̃ ≈ 1 motivation by reference to Refs. [17, 18] by overlapping author sets. This is normal scientific genealogy and is not load-bearing for the statistical conclusions, which rest on external CC, SNIa, and globular-cluster data. Score: 1.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper inherits a previously proposed gravity theory and tests it. The load-bearing additions specific to this work are (a) the exponential F(H) ansatz with hand-chosen exponent pairs, (b) several parameters fixed by fiat to make the analysis tractable (λ=0, L̃=0.90, β=10⁻⁵ depending on subfamily), and (c) the 12.7 Gyr age-of-universe lower bound that drives several exclusions.

pith-pipeline@v0.9.0 · 9467 in / 6668 out tokens · 101040 ms · 2026-05-06T17:54:51.932108+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.