{"id":"ac87f688-0cf9-4f7b-9393-bf2817af6c7d","arxiv_id":"2607.24939","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":6,"one_line_summary":"A binning-based IC method yields the first N-body measurements of oscillating halo bias from cosmological collider bispectra, with mass- and assembly-dependent phases fit by peak-background-split theory.","lead":"Simulations show that cosmological-collider primordial signals produce clear, mass- and assembly-dependent oscillations in halo bias, matched by simple peak-background-split theory. The new initial-conditions method and public sims make these oscillatory signatures a practical, systematics-resistant target for large-scale structure surveys.","discovery_kind":"new_method","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"All halo-bias fits use a single, very large non-Gaussian amplitude (f_NL=500 with a ×4 oscillation boost); linearity of the response in f_NL is never tested, so the b_φ/b_σ match to universality could absorb O(f_NL²) contamination.","rationale":"The reader's flagged weakest assumption — the engineered QSFosc template with ×4 boost, hard Θ cut, and hand-tuned phase — is real but is openly acknowledged in §2.2 and scoped by the authors as a deliberate phenomenological choice to make the signal measurable; it limits generality/detectability extrapolation rather than the internal correctness of what is shown. I therefore only partially agree that it is the most load-bearing issue. The less-examined assumption is f_NL-linearity of the measured response, which underpins the b_φ/b_σ universality comparison that the conclusions lean on (\"existing calibrations of b_φ... can be utilized for collider PNGs\"). This is a correctness-risk item, not circularity: the inference is valid if linearity holds, and the authors' own pipeline makes the check cheap. I do not change the verdict: the multi-frequency, multi-mass consistency of the fits is strong internal evidence that any O(f_NL²) contamination is small, the method is validated on Local PNG (App. B), code and sims are public, and the claims are appropriately scoped. ACCEPT remains warranted; the proposed test would upgrade confidence in the Fig. 7 universality match from plausible to demonstrated.","tokens_in":18991,"tokens_out":2848,"duration_ms":95621,"concrete_test":"Using the public Aarambam pipeline and the same Gaussian seeds, run two additional ω=1.5 simulations at f_NL = +250 and f_NL = −250 (or +100). Measure b(k) via Eq. 2.14 and compare ∆b(k)/f_NL across the three amplitudes within bootstrap errors over the 10-realization ensemble. If the rescaled responses agree, linearity holds and Fig. 7 stands; if ∆b(500)/500 deviates systematically (e.g., >1σ in the oscillation trough near k_c), re-fit b_φ, b_σ in the linear regime and re-examine the universality comparison.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim — that Eq. 2.15 with b_φ, b_σ at (or near) their universality values accurately fits the measured oscillating bias — rests on comparing b(k) at f_NL=500 against f_NL=0 only. The response is assumed linear: ∆b(k) = 2 f_NL F_R/M (b_φ + 2b_σ ∂lnF_R/∂lnσ²). But the IC generator (Eq. 2.1) is a quadratic convolution, so the simulated field is a second-order non-Gaussian field whose halo-bias response generically contains O(f_NL²) terms (e.g., trispectrum-mediated contributions to P_hm and P_mm). The effective local amplitude in the squeezed limit is f_NL·[1+4cos(...)] ∈ [−1500, +2500] — far outside any regime where linear response has been validated in the literature (Local-PNG studies typically verify linearity with |f_NL| ≲ few hundred). The authors do check the one-loop power-spectrum correction (dlnP/df_NL ≲ 1e-5, §2.1), which mitigates the concern for P_mm, but no analogous check exists for the halo-bias response itself. Mitigating evidence: the same two coefficients fit three frequencies and six mass bins simultaneously (Figs. 2–6, 7), and a nonlinear contaminant would need to conspire across ω and M to preserve those shapes — so the risk is moderate, not severe. Still, with only 10 realizations and a single nonzero f_NL, the universality match in Fig. 7 (used to conclude Local-type b_φ calibrations transfer to collider PNGs) is the paper's least-secured inference.","agreement_with_reader":"partial"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The manuscript introduces a new initial-conditions method for N-body simulations with arbitrary primordial bispectra: rather than decomposing the mode-coupling kernel into a global separable basis, the kernel is represented by binned cubic-Lagrange interpolators in (k1, k2) times per-bin cubic polynomials in k3 (Eqs. 2.4–2.9), combined with the Wagner–Verde reduced-bispectrum kernel (Eq. 2.11) to control IR corrections to the power spectrum. The method is applied to a phenomenological \"QSFosc\" collider template (Eq. 2.13) at frequencies ω = 1.5, 3, 4.5. The paper presents the first measurements of oscillating, scale-dependent halo bias in such simulations (Figs. 2–4, 6), shows that the oscillation phase depends on halo mass and on Ṽmax-selected assembly bias, that higher primordial frequencies are suppressed by window-function averaging (Fig. 5), and that a two-parameter peak-background-split model (Eq. 2.15, with b_φ and b_σ) fits all cases with coefficients close to the universality values δ_c(b1−1) and 1 (Fig. 7). Appendix B validates the IC pipeline on local-type PNG against the known 1/k² bias, Paper I, and 2LPTPNG.","tokens_in":19452,"tokens_out":3444,"duration_ms":119261,"significance":"If the results hold, this is a valuable and timely contribution. Scale-dependent halo bias is the leading LSS observable for primordial non-Gaussianity, and collider-type oscillatory bispectra are a physically motivated target for DESI, Euclid, SPHEREx, and Rubin; until now there was no general simulation tool for bispectra with non-trivial squeezed-limit behavior. The paper ships several concrete strengths: a flexible IC generator that avoids basis-decomposition bottlenecks and is publicly released (Aarambam), a public simulation suite (Ulagam), an explicit end-to-end validation on local PNG against an independent codebase (2LPTPNG) and the authors' own Paper I, a quantified kernel-approximation residual (few-percent in the kernel, <1% in the predicted bias, §2.1), and a non-trivial consistency check in that the same b_φ, b_σ values fit three frequencies and six mass bins simultaneously (Figs. 2–7). The demonstration that the oscillation phase carries mass- and assembly-bias-dependent information not easily mimicked by survey systematics is a genuinely useful addition to the case for collider searches in LSS. The finding that Local-type b_φ calibrations appear to transfer to colli","major_comments":[{"comment":"All measurements use a single non-Gaussian amplitude, f_NL = 500 with an additional ×4 oscillation boost (Eq. 2.13), so the effective local amplitude in the squeezed limit spans roughly [−1500, +2500]. The theoretical model (Eq. 2.15) is linear in f_NL, but the IC generator (Eq. 2.1) is a quadratic convolution, so the simulated field is a second-order non-Gaussian field whose halo-bias response generically contains O(f_NL²) contributions (e.g., trispectrum-mediated terms in P_hm and P_mm). The authors check the one-loop power-spectrum correction (dlnP/df_NL ≲ 1e-5, §2.1), but no analogous check exists for the halo-bias response itself. This matters specifically for one of the paper's three headline conclusions: that the best-fit b_φ, b_σ match their universality values to 10–20% (Fig. 7) and therefore that Local-type b_φ calibrations transfer to collider PNGs (§3.3, §4). An O(f_NL²) cont","section":"§2.3, §3, Fig. 7 / Eq. (2.15)"}],"minor_comments":[{"comment":"The QSFosc template (Eq. 2.13) is engineered: a ×4 oscillation boost, a hard Heaviside cut at k3/k1 = 0.15, and a hand-chosen phase φ = 0.5, explicitly to avoid the exponential suppression of high-frequency oscillations in real collider Lagrangians. The paper is transparent about this (§2.2), but the Conclusions' detectability framing would benefit from one explicit sentence stating that the measured oscillation amplitudes do not map directly onto constraints for any specific inflationary model, and that the ω-dependence results (Fig. 5) are the relevant guide for realistic templates.","section":"§2.2, §4"},{"comment":"The claim that the oscillatory signal 'is not easily mimicked by known observational systematics' is plausible but asserted rather than demonstrated. A brief quantitative remark (e.g., what residual large-scale power modulation would be required to fake a coherent multi-cycle oscillation with the correct mass-dependent phase) or a softening of the language would make this more defensible.","section":"§4"},{"comment":"Lowest mass bin (M = 10^14 M⊙/h): halos are resolved with only ~50 particles, and the authors note a concentration-based selection contaminates the mass-selected sample (§2.3, §3.1). It would help to quantify the shot-noise contribution to P_hm in this bin and state explicitly whether the P_hm/P_mm ratio in Fig. 2 is shot-noise corrected; this is relevant to how much weight the lowest-mass panel should carry.","section":"§2.3, Fig. 2"},{"comment":"k_c is defined operationally via a cubic-spline fit and the location of the first trough (footnote 8). Since the factor-of-two shifts in k_c (mass- and Ṽmax-dependent) are among the paper's quantitative headline numbers, the robustness of k_c to the spline choice and to the k-binning should be stated, ideally with uncertainties.","section":"§3.1, footnote 8"},{"comment":"Fig. 7 caption refers to 'different models (columns)' but the figure appears to use panels/rows for models and Ṽmax selections; please check the caption against the actual layout. Also, b_σ errors are not visible in the description — please state how coefficient uncertainties are estimated (per-realization fits vs. bootstrap).","section":"Appendix A, Fig. 7"},{"comment":"The relation to Goldstein et al. (2025) is discussed in footnote 3, but given that work also simulated a collider signal, a sentence in §1 or §4 clarifying precisely what is 'first' here (full-shape ICs vs. squeezed-limit-only ICs; halo bias vs. the statistics they analyzed) would preempt confusion about the novelty claim.","section":"§1, footnote 3"},{"comment":"Eq. (2.16): the q-integration limits [1e-4, 20] h/Mpc and the µ-discretization are given, but it would be useful to state the convergence of F_R (and hence of the predicted oscillation phase, which is sensitive to the integrand's oscillatory cancellations) with respect to these choices, particularly for the ω = 4.5 model.","section":"§2.4, Eq. (2.16)"}],"recommendation":"minor_revision","confidential_remarks":"The paper is a competent and well-validated methodological contribution from a group with a strong track record on PNG simulations; the companion Papers I/II and the public code/sim releases support reproducibility. The only substantive vulnerability is the untested linearity of the halo-bias response at the very large effective f_NL used, which bears on the Fig. 7 universality conclusion but not on the existence or morphology of the oscillating bias itself. The \"first measurement\" novelty claim relative to Goldstein et al. (2025) is handled in the text and is defensible given the full-shape IC construction here. I do not think the single-f_NL issue requires more than one additional simulation set or an explicit caveat, hence minor rather than major revision."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The punchline is simple: they finally have N-body runs that show non-monotonic, oscillating scale-dependent bias from a collider-like squeezed bispectrum, and a plain peak-background-split formula (their Eq. 2.15) tracks the mass and assembly-bias dependence. That fills the gap they correctly flag relative to Dalal et al. 2008 and to Goldstein et al. 2025 (who stayed in the pure squeezed limit).\n\nWhat is actually new is the IC generator. Instead of another global basis expansion they bin the reduced bispectrum kernel and glue the cells with cubic Lagrange interpolators (Eqs. 2.4–2.10). Residuals are a few percent; the halo-bias prediction from the approximate kernel matches the true one at the percent level. They ship the code and the Ulagam snapshots. Appendix B recovers the classic local 1/k² bias and matches 2LPTPNG, so the pipeline is not vaporware.\n\nThe science results are clean within the stated setup. Mass shifts the oscillation scale roughly as R(M); concentration splits shift the phase by a comparable amount; higher ω averages the signal down inside the window. The same two coefficients b_φ, b_σ work across three frequencies, which is what you want if they really are sample properties.\n\nSoft spots, in proportion. The input is a phenomenological QSFosc shape: QSF base times a cosine that is hard-cut to k3/k1 < 0.15, amplitude boosted by 4, phase hand-chosen so the feature sits in linear k. Real high-frequency colliders are exponentially suppressed unless mixing is strong; they say so. All fits use a single enormous amplitude (f_NL = 500 plus the boost), so linearity of the bias response is assumed, not demonstrated. O(f_NL²) contamination could in principle be absorbed into the fitted b_φ, b_σ. The fact that one pair of coefficients fits six masses and three frequencies makes a pure conspiracy less likely, but it is still the least-secured inference. Lowest-mass bin is only ~50 particles; they flag the concentration selection.\n\nThis is for people who actually want to put collider templates into LSS pipelines or mock challenges. The math and citation pattern look solid; the data products are real. I would send it to referees and I would cite the IC method and the public sims. Bring it to reading group if anyone is thinking about DESI/SPHEREx PNG forecasts.","headline":"First real N-body detection of collider-style oscillating halo bias, with a usable IC method and public sims; the template is engineered and f_NL is huge, but the measurements themselves hold up.","tokens_in":20300,"tokens_out":627,"would_cite":true,"duration_ms":17757,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Cosmological collider signals leave measurable mass- and assembly-dependent oscillations in the halo bias of N-body simulations.","keywords":["cosmological colliders","primordial non-Gaussianity","halo bias","scale-dependent bias","peak-background split","N-body simulations","squeezed bispectrum","assembly bias"],"falsifier":"Generate an independent set of initial conditions from the same QSFosc template with a different separable approximation, run matching N-body boxes, and check whether the measured b+(k)/b1 oscillations and their mass- and Vmax-dependent phase shifts still agree with the peak-background-split curves at the percent level shown in the paper’s figures.","tokens_in":19991,"feed_emoji":"🌌","tokens_out":1025,"duration_ms":17340,"temperature":0.7,"pith_summary":"Early-universe particle physics can leave oscillating patterns in the three-point correlations of the primordial density field. This paper shows that those oscillations survive into the late universe as scale-dependent wiggles in how dark-matter halos cluster. The authors introduce a new way to generate simulation initial conditions that captures the squeezed-limit collider signal at percent-level accuracy without relying on traditional basis decompositions. In the resulting N-body runs they obtain the first direct measurements of oscillating halo bias. A factor-of-ten change in halo mass shifts the location of the wiggles by roughly a factor of two; higher primordial frequencies wash the late-time signal out because the oscillations average inside the halo window function; and assembly bias (concentration selection at fixed mass) further shifts the phase. A simple peak-background-split formula with two free response coefficients fits every case. Because the oscillations carry mass- and selection-dependent phase offsets that ordinary survey systematics are unlikely to fake, the authors argue the signature is a cleaner observational target than the classic local non-Gaussian 1/k^{2} bias.","feed_headline":"Collider signals leave mass-tuned wiggles in halo clustering","feed_subtitle":"Simulations show oscillating bias whose phase shifts with mass and concentration, harder to fake than classic PNG","key_machinery":"A binned, cubic-Lagrange interpolating decomposition of the mode-coupling kernel (Eq. 2.9–2.10) that keeps the FFT-based initial-condition integral separable while remaining accurate in the strongly squeezed limit; the resulting scale-dependent bias is then predicted by the peak-background-split expression (Eq. 2.15) involving the two response coefficients b_φ and b_σ.","core_discovery":"Primordial bispectra of cosmological-collider type imprint non-monotonic, oscillating scale-dependent halo bias that can be measured in N-body simulations; the amplitude and phase of those oscillations depend systematically on halo mass and on secondary properties such as concentration, higher primordial frequencies suppress the late-time amplitude by window averaging, and a two-parameter peak-background-split model accurately reproduces all of the measured behavior.","pith_inferences":["If the phase-shift pattern really cannot be mimicked by known systematics, multi-tracer or multi-mass analyses become natural self-calibration tools for f_NL b_φ versus f_NL b_σ.","The same binned-kernel method should immediately extend to other non-separable squeezed bispectra (e.g., equilateral or orthogonal shapes with oscillatory corrections) that previous basis decompositions struggled to capture.","Because the suppression is set by the halo window, lower-mass tracers or higher-redshift samples may reopen a window onto higher-frequency colliders that are washed out at z=0 cluster scales."],"forward_implications":["Existing b_φ calibrations developed for local-type non-Gaussianity can be reused for collider analyses because the same coefficients fit across frequencies once the sample is fixed.","Higher-frequency collider models are observationally harder: once the primordial oscillation period becomes shorter than the halo window, the late-time bias signal averages toward zero.","Mass- and assembly-dependent phase offsets supply an extra handle that can help separate a true collider signal from large-scale survey systematics.","The public initial-condition code and simulation suite allow existing large-scale-structure pipelines to add collider templates without new N-body campaigns."],"fun_headline_variants":["Collider bispectra imprint mass-tuned oscillations in halo bias","Simulations measure oscillating halo bias from cosmological colliders","Halo mass and concentration shift phase of collider bias wiggles","Higher collider frequencies suppress halo-bias oscillations via window averaging","Peak-background split recovers mass-dependent collider wiggles in halos"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The input bispectrum is a hand-engineered phenomenological template that multiplies a quasi-single-field shape by a boosted cosine only inside a hard squeezed cut, rather than a full Lagrangian collider model whose high-frequency oscillations are generically exponentially suppressed.","fun_headline_variants_meta":{"raw":{"variants":["Collider bispectra imprint mass-tuned oscillations in halo bias","Simulations measure oscillating halo bias from cosmological colliders","Halo mass and concentration shift phase of collider bias wiggles","Higher collider frequencies suppress halo-bias oscillations via window averaging","Peak-background split recovers mass-dependent collider wiggles in halos"]},"model":"grok-4.5","effort":"low","cost_usd":0.004723,"raw_usage":{"total_tokens":1401,"prompt_tokens":812,"num_sources_used":0,"completion_tokens":88,"cost_in_usd_ticks":47228000,"prompt_tokens_details":{"text_tokens":812,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":501,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":812,"tokens_out":88,"duration_ms":8842,"temperature":1.0,"reasoning_tokens":501,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T05:24:43.689641+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Generate an independent set of initial conditions from the same QSFosc template with a different separable approximation, run matching N-body boxes, and check whether the measured b+(k)/b1 oscillations and their mass- and Vmax-dependent phase shifts still agree with the peak-background-split curves at the percent level shown in the paper’s figures.","supporting_citations":[],"review_version":1}