Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Template-free symbolic regression finds that CMB data prefer an inverse-k primordial oscillation over standard linear and log templates.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-11 11:22 UTC pith:4YR7BDNU

load-bearing objection Template-free SR recovers the same inverse-k form on Planck and SPA; the result is real but weak (~1σ) and conditioned on a fixed baseline cosmology. the 3 major comments →

arxiv 2607.04925 v1 pith:4YR7BDNU submitted 2026-07-06 astro-ph.CO gr-qchep-th

Inverse-k Primordial Oscillations from a Symbolic Regression Search

Ze-Yu Peng , Qing-Yu Lan , Yun-Song Piao This is my paper
classification astro-ph.CO gr-qchep-th
keywords primordial featuressymbolic regressionprimordial power spectruminverse-k oscillationCMBPlanckACTSPT-3G
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Oscillatory wiggles in the primordial power spectrum can be fingerprints of new early-universe physics, but they are usually hunted with pre-chosen templates. This paper instead lets symbolic regression invent analytic forms for those wiggles, scoring them only by how well they fit cosmic-microwave-background data. Both Planck alone and the combination of Planck with ACT and SPT-3G independently converge on the same simple expression: an inverse-k oscillation of the form cos(B/k) or sin(B/k) with B around 4 Mpc^{-1}. When that form is compared head-to-head with the usual linear and logarithmic templates, it improves the fit the most and shows a weak preference for a non-zero amplitude, while the standard templates remain consistent with zero. The result suggests that an open, machine-driven search can surface feature shapes that fixed-template analyses miss, and that those shapes may point to distinctive early expansion histories.

Core claim

Both Planck and the combined Planck+ACT+SPT-3G datasets independently select an inverse-k oscillation, cos(B/k) or sin(B/k) with B ≃ 4 Mpc^{-1}, as the leading low-complexity feature on the symbolic-regression Pareto front. On the combined data this inverse-k template yields the largest likelihood improvement over a featureless power-law spectrum (best-fit Δχ² ≃ −12.6) and a weak one-sigma preference for non-zero amplitude, whereas the conventional linear and logarithmic templates remain consistent with zero.

What carries the argument

Symbolic regression (PySR) that searches the space of analytic expressions for the feature f(k) multiplying the standard power-law primordial spectrum, using the CMB χ² as the sole objective while background and nuisance parameters are held fixed; the Pareto front of complexity versus fit improvement then isolates the inverse-k oscillation as the simplest competitive form.

Load-bearing premise

All cosmological background and nuisance parameters are locked to their best-fit power-law values during the search, so the recovered feature form is conditioned on that fixed baseline rather than being re-optimized jointly with the rest of the cosmology.

What would settle it

A full MCMC re-analysis that frees the background and nuisance parameters together with the inverse-k amplitude, frequency and phase, and checks whether the preference for non-zero amplitude and the superiority over linear and log templates survive on the same SPA data.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper performs a template-free search for oscillatory features in the primordial power spectrum using symbolic regression (PySR) with CMB χ² as the objective. Both Planck and the combined SPA (Planck+ACT+SPT-3G) datasets independently recover an inverse-k oscillation of the form cos(B/k) or sin(B/k) with B ≃ 4 Mpc^{-1} as the leading low-complexity expression (complexity ~8) on the Pareto front. The authors then promote this form to a three-parameter template and compare it via MCMC to the standard linear and logarithmic oscillation templates on SPA, finding that inverse-k yields the largest improvement (Δχ² ≃ -12.6) and a weak 1σ preference for nonzero amplitude, while the other templates remain consistent with zero. A heuristic link to standard-clocks signals with p = -1 is noted.

Significance. If the inverse-k preference survives joint cosmological and nuisance variation, the work supplies a concrete, data-driven candidate feature beyond the usual linear/log templates and demonstrates that symbolic regression can surface interpretable analytic forms in CMB cosmology without hand-crafted templates. The dual-dataset convergence of the C8 expressions, the residual plots matching ACT/SPT polarization structure, and the explicit Pareto-front tables are genuine strengths. The result is of interest for early-universe model building (NEC-violating or phantom-like phases) and for future small-scale CMB experiments, even though the present preference remains only weak.

major comments (3)
  1. Methods and Supplemental Material: the entire SR search freezes all background cosmological and nuisance parameters at their power-law best-fit values, so the χ² objective varies only f(k). The claim that this is harmless is imported from the authors’ prior template analysis [27] and is not re-validated inside the SR pipeline. Because the headline result is that both datasets independently select cos(B/k) (or sin(B/k)) as the leading low-complexity feature, a joint MCMC or profile-likelihood check that frees at least the main ΛCDM parameters (and key ACT/SPT nuisances) while retaining the recovered inverse-k form is load-bearing; without it the “template-free discovery” remains conditioned on a fixed baseline.
  2. Table III and footnote [44]: the three templates share the same flat prior ω_X ∈ [1,100], yet g_lin, g_log and g_inv map this interval onto physically inequivalent frequency ranges. The paper itself notes that the comparison is therefore heuristic. Because the ranking of inverse-k over linear/log (Table IV, Fig. 3) is a central claim, either re-run the MCMC with priors that equalize the effective frequency content or quantify how much of the Δχ² advantage is prior-driven.
  3. SR results and Supplemental Material (resolution estimate): higher-complexity Planck expressions containing 1/k^{2} terms are discarded as numerically unreliable because they oscillate faster than the CLASS sampling (k_per_decade = 400). The same sampling criterion should be applied quantitatively to the C8 inverse-k solutions that carry the main claim; an explicit statement of the maximum reliable B (or a higher-resolution re-evaluation of Eqs. 2–3) is needed to confirm that the recovered B ≃ 4 Mpc^{-1} is not itself resolution-limited at the lowest k that affect the multipoles shown in Fig. 2.
minor comments (5)
  1. Abstract and Introduction: the phrase “model-independent” is too strong; the search is template-free within a fixed operator set and a fixed baseline cosmology. Soften to “template-free within a defined hypothesis class”.
  2. Fig. 1 caption and axis label: “2 relative to the power-law PPS” appears truncated; restore “Δχ^{2}” for clarity.
  3. Tables I–II versus V–VI: the main-text tables list only selected Pareto points; a one-sentence pointer that the full fronts appear in the Supplemental Material would help readers.
  4. Discussion: the standard-clocks interpretation (p = -1) is presented as heuristic, which is appropriate, but a brief remark on whether the recovered amplitude and phase are consistent with existing NEC-violating model predictions would strengthen the paragraph.
  5. Eq. (1) and subsequent expressions: units of k (Mpc^{-1}) are stated once; repeating them in the captions of Tables I–II would avoid ambiguity for readers who start from the tables.

Circularity Check

1 steps flagged

Mild self-citation for the fixed-parameter approximation; the inverse-k form itself emerges from free SR search rather than by construction.

specific steps
  1. self citation load bearing [Methods / Comparison with standard templates paragraph; citation [27]]
    "It has been shown that the variation of the background and nuisance parameters has a negligible effect on the feature constraints [27]."

    The SR search and subsequent MCMC ranking that produce the headline inverse-k selection are performed with all cosmological and nuisance parameters fixed to the power-law best-fit. The assertion that this freeze does not alter the preferred analytic form or Δχ² ranking rests only on a citation to the same authors' prior paper; without that self-citation the fixed-baseline χ² objective could in principle reshape the residual that SR attributes to f(k)≃A cos(B/k). The citation is therefore load-bearing for treating the recovered form as fully data-driven, though it does not insert the inverse-k expression by construction.

full rationale

The paper's central result is that PySR, run independently on Planck and SPA with a shared operator set and CMB χ² loss, selects inverse-k oscillations (C8 expressions in Tables I–II / Eqs. 2–3) as the leading low-complexity feature; that form is then re-parameterized as a three-parameter template and ranked against linear/log templates via MCMC. This is not definitional circularity, fitted-input-as-prediction, uniqueness smuggling, or renaming of a known result: the analytic expression is discovered by search, not inserted. The only load-bearing self-citation is the claim that freezing background and nuisance parameters at the power-law best-fit (Methods; Supplemental Material) has negligible effect on feature constraints, justified solely by the authors' prior work [27]. That is a standard methodological approximation in the literature and does not force the inverse-k functional form itself; the SR Pareto fronts and subsequent Δχ² ranking remain independent content. Score 2 reflects one minor self-citation that is not load-bearing for the claimed discovery. No other circular steps are present.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 1 invented entities

The central claim is empirical (SR + MCMC on CMB likelihoods), not a derivation from first principles. It rests on the standard power-law PPS baseline, fixed best-fit cosmology during the search, the chosen SR operator set and complexity metric, CLASS sampling resolution, and three-parameter oscillatory templates with flat priors. No new fundamental entity is required for the data claim; the NEC/phantom early-phase reading is explicitly heuristic and not load-bearing for the fit results.

free parameters (5)
  • Inverse-k frequency B (SR C8) = ≃4.1–4.2 Mpc^{-1}
    Leading SR solutions fix B≃4.10 (Planck) and ≃4.23 (SPA) Mpc^{-1} inside cos(B/k) or sin(B/k); this scale is read off the data-driven expressions and is central to the claimed feature.
  • Feature amplitude A_X (template MCMC) = A_inv best-fit 0.0203; posterior 0.0163^{+0.0099}_{-0.0071}
    Amplitude of f_X(k)=A_X cos[ω_X g_X(k)+φ_X] is fitted on SPA; the weak nonzero preference for the inverse-k template is the main statistical claim.
  • Frequency ω_X and phase φ_X = e.g. inverse best-fit ω=83.32, φ=2.24
    Fitted for each of linear, log, and inverse templates under shared flat priors; ω_X is not physically equivalent across g_X(k) choices, which the paper notes makes the comparison heuristic.
  • Fixed ΛCDM + nuisance baseline = best-fit power-law values (not re-listed)
    All background and nuisance parameters held at power-law best-fit values during SR and template MCMC; not varied, but the entire residual search is conditioned on this fixed point.
  • SR complexity cutoff / Pareto selection (C8) = complexity 8 highlighted
    Authors treat complexity ~8 as the leading robust feature where both datasets converge; higher Planck complexities are discarded as numerically unreliable—an analysis choice that selects which expression is promoted to the ‘inverse-k’ claim.
axioms (6)
  • domain assumption Primordial spectrum is P_R(k)=P_{R,0}(k)[1+f(k)] with power-law P_{R,0} and k_*=0.05 Mpc^{-1}.
    Eq. (1); standard feature ansatz used as the search space for f(k).
  • ad hoc to paper CMB χ² from CLASS with fixed cosmology is a sufficient objective for discovering the preferred analytic f(k).
    Methods/Supplemental: SR loss is CMB χ² with parameters frozen; validity of freezing is imported from prior work rather than re-derived.
  • ad hoc to paper Operator set {+,-,×,/,^,exp,log,sin,cos,tanh}, no nested trig, max size 25, defines the hypothesis class.
    Supplemental SR setup; different operators could change which low-complexity forms appear on the Pareto front.
  • domain assumption CLASS sampling with 400 k-points per decade adequately resolves inverse-k phases for B≃4 over the main CMB-sensitive range.
    Supplemental resolution estimate; used to keep C8 solutions and discard faster 1/k² Planck expressions.
  • domain assumption A k^{1/p} oscillation can be generated by a massive field in a∝t^p (standard-clocks), so inverse-k suggests p=−1.
    Discussion cites [12,13]; authors call the link heuristic and not a full standard-clocks detection.
  • standard math Standard arithmetic/trigonometric identities and likelihood comparison under flat priors.
    Used throughout SR expression algebra and MCMC posterior construction.
invented entities (1)
  • Inverse-k primordial feature template f_inv∝cos(ω k_*/k + φ) no independent evidence
    purpose: Phenomenological form promoted from SR Pareto front and compared to linear/log templates as the preferred low-complexity oscillation.
    Not a new particle or field; a data-selected functional form. Independent evidence would require confirmation with free cosmology and future CMB/LSS data; currently supported only by the same datasets used to discover it.

pith-pipeline@v1.1.0-grok45 · 12760 in / 4389 out tokens · 37001 ms · 2026-07-11T11:22:10.136126+00:00 · methodology

0 comments
read the original abstract

Oscillatory features in the primordial power spectrum, potential signatures of new physics in the early universe, are usually searched for using fixed templates. In this work, we perform a template-free search for primordial features using symbolic regression. We find that both Planck and the combined Planck+ACT+SPT-3G datasets independently select an inverse-$k$ oscillation, $\cos(B/k)$ with $B\simeq4\,\mathrm{Mpc}^{-1}$, as the leading low-complexity feature. Comparing this inverse-$k$ template with standard linear and logarithmic oscillating templates, we find that it fits the data best, showing a weak preference for a non-zero amplitude. Our results show that symbolic regression as a powerful machine learning technique can provide an interpretable, model-independent approach to cosmological discovery.

Figures

Figures reproduced from arXiv: 2607.04925 by Qing-Yu Lan, Yun-Song Piao, Ze-Yu Peng.

Figure 1
Figure 1. Figure 1: FIG. 1. Improvement of the CMB likelihood along the SR [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. CMB residuals for SR expressions with complexity 8, relative to the SPA power-law baseline. We also show the [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Marginalized posteriors of the oscillation amplitude [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. CMB residuals of the best-fit linear, logarithmic and inverse- [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

53 extracted references · 43 linked inside Pith

  1. [1]

    A. H. Guth, Phys. Rev. D23, 347 (1981)

    1981

  2. [2]

    A. D. Linde, Phys. Lett. B108, 389 (1982)

    1982

  3. [3]

    Albrecht and P

    A. Albrecht and P. J. Steinhardt, Phys. Rev. Lett.48, 4 0.00 0.01 0.02 0.03 0.04 0.05 0.06 AX Inverse Linear Log FIG. 3. Marginalized posteriors of the oscillation amplitude AX for the linear, logarithmic and inverse-ktemplates on the SPA dataset. 1220 (1982)

    1982

  4. [4]

    A. A. Starobinsky, Phys. Lett. B91, 99 (1980)

    1980

  5. [5]

    Aghanimet al.(Planck), Astron

    N. Aghanimet al.(Planck), Astron. Astrophys.641, A6 (2020), [Erratum: Astron.Astrophys. 652, C4 (2021)], arXiv:1807.06209 [astro-ph.CO]

    Pith/arXiv arXiv 2020

  6. [6]

    A. A. Starobinsky, JETP Lett.55, 489 (1992)

    1992

  7. [7]

    J. A. Adams, B. Cresswell, and R. Easther, Phys. Rev. D64, 123514 (2001), arXiv:astro-ph/0102236

    Pith/arXiv arXiv 2001

  8. [8]

    X. Wang, B. Feng, M. Li, X.-L. Chen, and X. Zhang, Int. J. Mod. Phys. D14, 1347 (2005), arXiv:astro- ph/0209242

    arXiv 2005

  9. [9]

    Gong, JCAP07, 015 (2005), arXiv:astro- ph/0504383

    J.-O. Gong, JCAP07, 015 (2005), arXiv:astro- ph/0504383

    arXiv 2005

  10. [10]

    Slosaret al., Bull

    A. Slosaret al., Bull. Am. Astron. Soc.51, 98 (2019), arXiv:1903.09883 [astro-ph.CO]

    Pith/arXiv arXiv 2019

  11. [11]

    Ach´ ucarroet al., (2022), arXiv:2203.08128 [astro- ph.CO]

    A. Ach´ ucarroet al., (2022), arXiv:2203.08128 [astro- ph.CO]

    Pith/arXiv arXiv 2022

  12. [12]

    Chen and C

    X. Chen and C. Ringeval, JCAP08, 014 (2012), arXiv:1205.6085 [astro-ph.CO]

    Pith/arXiv arXiv 2012

  13. [13]

    X. Chen, A. Loeb, and Z.-Z. Xianyu, Phys. Rev. Lett. 122, 121301 (2019), arXiv:1809.02603 [astro-ph.CO]

    Pith/arXiv arXiv 2019

  14. [14]

    Easther, W

    R. Easther, W. H. Kinney, and H. Peiris, JCAP05, 009 (2005), arXiv:astro-ph/0412613

    Pith/arXiv arXiv 2005

  15. [15]

    P. D. Meerburg, D. N. Spergel, and B. D. Wandelt, Phys. Rev. D89, 063536 (2014), arXiv:1308.3704 [astro- ph.CO]

    Pith/arXiv arXiv 2014

  16. [16]

    P. D. Meerburg and D. N. Spergel, Phys. Rev. D89, 063537 (2014), arXiv:1308.3705 [astro-ph.CO]

    Pith/arXiv arXiv 2014

  17. [17]

    Peiris, R

    H. Peiris, R. Easther, and R. Flauger, JCAP09, 018 (2013), arXiv:1303.2616 [astro-ph.CO]

    Pith/arXiv arXiv 2013

  18. [18]

    Easther and R

    R. Easther and R. Flauger, JCAP02, 037 (2014), arXiv:1308.3736 [astro-ph.CO]

    Pith/arXiv arXiv 2014

  19. [19]

    M. Aich, D. K. Hazra, L. Sriramkumar, and T. Souradeep, Phys. Rev. D87, 083526 (2013), arXiv:1106.2798 [astro-ph.CO]

    Pith/arXiv arXiv 2013

  20. [20]

    Akramiet al.(Planck), Astron

    Y. Akramiet al.(Planck), Astron. Astrophys.641, A10 (2020), arXiv:1807.06211 [astro-ph.CO]

    Pith/arXiv arXiv 2020

  21. [21]

    Beutler, M

    F. Beutler, M. Biagetti, D. Green, A. Slosar, and B. Wal- lisch, Phys. Rev. Res.1, 033209 (2019), arXiv:1906.08758 [astro-ph.CO]

    Pith/arXiv arXiv 2019

  22. [22]

    Ballardini, F

    M. Ballardini, F. Finelli, F. Marulli, L. Moscardini, and A. Veropalumbo, Phys. Rev. D107, 043532 (2023), arXiv:2202.08819 [astro-ph.CO]

    Pith/arXiv arXiv 2023

  23. [23]

    Mergulh˜ ao, F

    T. Mergulh˜ ao, F. Beutler, and J. A. Peacock, JCAP08, 012 (2023), arXiv:2303.13946 [astro-ph.CO]

    Pith/arXiv arXiv 2023

  24. [24]

    Louiset al.(Atacama Cosmology Telescope), JCAP 11, 062 (2025), arXiv:2503.14452 [astro-ph.CO]

    T. Louiset al.(Atacama Cosmology Telescope), JCAP 11, 062 (2025), arXiv:2503.14452 [astro-ph.CO]

    Pith/arXiv arXiv 2025

  25. [25]

    Calabreseet al.(Atacama Cosmology Telescope), JCAP11, 063 (2025), arXiv:2503.14454 [astro-ph.CO]

    E. Calabreseet al.(Atacama Cosmology Telescope), JCAP11, 063 (2025), arXiv:2503.14454 [astro-ph.CO]

    Pith/arXiv arXiv 2025

  26. [26]

    Camphuiset al.(SPT-3G), Phys

    E. Camphuiset al.(SPT-3G), Phys. Rev. D113, 083504 (2026), arXiv:2506.20707 [astro-ph.CO]

    Pith/arXiv arXiv 2026

  27. [27]

    Peng and Y.-S

    Z.-Y. Peng and Y.-S. Piao, Phys. Rev. D113, 023544 (2026), arXiv:2507.17276 [astro-ph.CO]

    Pith/arXiv arXiv 2026

  28. [28]

    S. K. Nerval, R. Hlozek, H. T. Jense, and J. R. Bond, (2026), arXiv:2606.28310 [astro-ph.CO]

    Pith/arXiv arXiv 2026

  29. [29]

    Braglia, X

    M. Braglia, X. Chen, and D. K. Hazra, JCAP06, 005 (2021), arXiv:2103.03025 [astro-ph.CO]

    Pith/arXiv arXiv 2021

  30. [30]

    Braglia, X

    M. Braglia, X. Chen, and D. K. Hazra, Eur. Phys. J. C 82, 498 (2022), arXiv:2106.07546 [astro-ph.CO]

    Pith/arXiv arXiv 2022

  31. [31]

    Braglia, X

    M. Braglia, X. Chen, and D. K. Hazra, Phys. Rev. D 105, 103523 (2022), arXiv:2108.10110 [astro-ph.CO]

    Pith/arXiv arXiv 2022

  32. [32]

    Braglia, X

    M. Braglia, X. Chen, D. K. Hazra, and L. Pinol, JCAP 03, 014 (2023), arXiv:2210.07028 [astro-ph.CO]

    Pith/arXiv arXiv 2023

  33. [33]

    Petretti, M

    C. Petretti, M. Braglia, X. Chen, D. Kumar Hazra, and S. Paban, JCAP06, 035 (2025), arXiv:2411.03459 [astro- ph.CO]

    Pith/arXiv arXiv 2025

  34. [34]

    Schmidt and H

    M. Schmidt and H. Lipson, Science324, 81 (2009)

    2009

  35. [35]

    Interpretable machine learning for sci- ence with pysr and symbolicregression.jl,

    M. Cranmer, “Interpretable machine learning for sci- ence with pysr and symbolicregression.jl,” (2023), arXiv:2305.01582 [astro-ph.IM]

    Pith/arXiv arXiv 2023

  36. [36]

    Udrescu and M

    S.-M. Udrescu and M. Tegmark, Sci. Adv.6, eaay2631 (2020), arXiv:1905.11481 [physics.comp-ph]

    Pith/arXiv arXiv 2020

  37. [37]

    Sousa-Neto, C

    A. Sousa-Neto, C. Bengaly, J. E. Gonzalez, and J. Alcaniz, Phys. Dark Univ.50, 102108 (2025), arXiv:2502.10506 [astro-ph.CO]

    Pith/arXiv arXiv 2025

  38. [38]

    S. M. Koksbang and A. Heinesen, (2026), arXiv:2604.05822 [astro-ph.CO]

    Pith/arXiv arXiv 2026

  39. [39]

    D. J. Bartlett, H. Desmond, and P. G. Ferreira, IEEE Trans. Evol. Comput.28, 964 (2024), arXiv:2211.11461 [astro-ph.CO]

    Pith/arXiv arXiv 2024

  40. [40]

    D. Blas, J. Lesgourgues, and T. Tram, JCAP07, 034 (2011), arXiv:1104.2933 [astro-ph.CO]

    Pith/arXiv arXiv 2011

  41. [41]

    See Supplemental Material at [URL will be inserted by publisher] for details of the symbolic-regression setup, the resolution estimate, and the full Pareto-front results

  42. [42]

    Stahl, D

    C. Stahl, D. Werth, and V. Poulin, Phys. Rev. D111, 123514 (2025), arXiv:2502.02571 [astro-ph.CO]

    Pith/arXiv arXiv 2025

  43. [43]

    Calderon, T

    R. Calderon, T. Simon, A. Shafieloo, and D. K. Hazra, JCAP01, 057 (2026), arXiv:2504.06183 [astro-ph.CO]

    Pith/arXiv arXiv 2026

  44. [44]

    Note that the MCMC results depend on the choice of priors, in particular onω X, which is not equivalent across the three templates; the comparison should therefore be regarded as heuristic

  45. [45]

    Torrado and A

    J. Torrado and A. Lewis, JCAP05, 057 (2021), arXiv:2005.05290 [astro-ph.IM]

    Pith/arXiv arXiv 2021

  46. [46]

    Piao and Y.-Z

    Y.-S. Piao and Y.-Z. Zhang, Phys. Rev. D70, 063513 (2004), arXiv:astro-ph/0401231

    Pith/arXiv arXiv 2004

  47. [47]

    Baldi, F

    M. Baldi, F. Finelli, and S. Matarrese, Phys. Rev. D72, 083504 (2005), arXiv:astro-ph/0505552

    Pith/arXiv arXiv 2005

  48. [48]

    Cai and Y.-S

    Y. Cai and Y.-S. Piao, Phys. Rev. D103, 083521 (2021), 5 40 20 0 20 40 DTT [ K2] Inverse Linear Log Planck ACT SPT-3G 2 0 2 DTE [ K2] 500 1000 1500 2000 2500 0.5 0.0 0.5 DEE [ K2] FIG. 4. CMB residuals of the best-fit linear, logarithmic and inverse-ktemplates relative to the no-feature SPA baseline, compared with the Planck (cropped), ACT and SPT-3G data...

    Pith/arXiv arXiv 2021

  49. [49]

    G. Ye, M. Zhu, and Y. Cai, JHEP02, 008 (2024), arXiv:2312.10685 [gr-qc]

    Pith/arXiv arXiv 2024

  50. [50]

    Y. Cai, M. Zhu, and Y.-S. Piao, Phys. Rev. Lett.133, 021001 (2024), arXiv:2305.10933 [gr-qc]

    Pith/arXiv arXiv 2024

  51. [51]

    S. Pan, Y. Cai, and Y.-S. Piao, Eur. Phys. J. C84, 976 (2024), arXiv:2404.12655 [astro-ph.CO]

    Pith/arXiv arXiv 2024

  52. [52]

    Adeet al.(Simons Observatory), JCAP02, 056 (2019), arXiv:1808.07445 [astro-ph.CO]

    P. Adeet al.(Simons Observatory), JCAP02, 056 (2019), arXiv:1808.07445 [astro-ph.CO]

    Pith/arXiv arXiv 2019

  53. [53]

    Inverse-k Primordial Oscillations from a Symbolic Regression Search

    K. Abazajianet al., (2019), arXiv:1907.04473 [astro- ph.IM]. 6 Supplemental Material for “Inverse-k Primordial Oscillations from a Symbolic Regression Search” Details of the SR search We write the primordial power spectrum (PPS) as PR(k) =P R,0(k) [1 +f(k)] (6) andf(k) is the primordial feature we search for. We usekin Mpc −1 as the input variable ofPySR....

    Pith/arXiv arXiv 2019