{"id":"c074108a-a41f-4e97-a27c-fc66e2a4b63d","arxiv_id":"2608.04139","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Kinetic misalignment fragmentation drives the small-scale axion spectrum toward the same attractor as post-inflationary axions, largely erasing the memory of the initial misalignment angle and velocity.","lead":"Axion dark matter from kinetic misalignment can fragment into small ripples, and this paper uses very large computer simulations to follow those ripples until the modern era. It finds that for heavier axions, self-interactions erase the memory of the early universe, so different starting conditions can end up looking the same.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The amnesia claim rests on the FAT-regulated non-relativistic peak extraction; a regulator-induced bias in kNp would erase the attractor.","rationale":"The reader's weakest assumption identifies the FAT regulator, the small N=256^3 boxes, and the fitted Cfit as the main threats to the late-time peak-drift conclusion. My reading agrees: the strongest claim in the paper is precisely that distinct early histories converge to a common late-time spectrum, and the only way that convergence is demonstrated is through the NR simulations in which the physical peak is inferred after subtracting an artificial DAS component created by an ad hoc regulator. The paper is honest about the regulator's ad hoc nature and about the DAS width shifting the extracted peak, but it does not quantify how large such a shift can be in the units that matter for the attractor claim. The run-to-run stability tests vary parameters within the same regulator family; they do not test the regulator family itself. A dedicated higher-resolution run with varied μ_fat, and ideally with a different collapse treatment or with initial conditions taken directly from the relativistic snapshot, would settle whether the extracted kNp(ηc) is physical. The exponent mismatch with Ref. [24] (0.73 versus 0.5) strengthens the need for such a test, since the claimed 'compatibility within uncertainties' is exactly what a regulator bias could destroy. For these reasons I do not propose changing the reader's conditional verdict, but I would not accept the paper until this regulator dependence is quantified.","tokens_in":44168,"tokens_out":7590,"duration_ms":71961,"concrete_test":"Run one benchmark fA=10^10 GeV NR simulation at N=512^3 and N=1024^3 with the same box size and the same physical initial spectrum, varying the FAT parameter over μ_fat ∈ {1.5, 2, 2.5, 3, 4}, and extract kNp(ηc) with the same two-component fit. If the median kNp shifts by more than the 25%-FWHM band quoted in Fig. 4, or if the higher-resolution runs change the kNp/kJ exponent in Fig. S13 by more than 0.1, the amnesia claim is not robust to the regulator and resolution.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that after fragmentation the spectral peak at matter-radiation equality is set by fA and ΩA alone—is carried by the non-relativistic continuation of the peak drift. That continuation depends on two linked ad hoc choices. First, Eq. (S128) replaces mψ by μ_fat^2/(δx^2 mψ) only in the resummed Bessel potential, so the regulator creates artificial dense-axion-star (DAS) configurations while leaving the dilute Gross-Pitaevskii term unchanged. Second, the physical peak kNp is not measured directly: it is extracted from a two-component fit that subtracts a Gaussian DAS peak located near the grid scale (Fig. S9), after masking overdensities δΥ≥5. The paper's own Supplemental Material concedes that the finite width of the DAS mass distribution 'can shift the extracted peak' and that the regulator is not unique. Run-to-run stability across μ_fat and box size is reassuring, but all those runs share the same FAT class and the same fit procedure, so a common bias would not appear as run-to-run scatter. The comparison with Ref. [24] already shows an exponent mismatch (kNp/kJ ∝ fA^-0.73 here versus fA^-0.5 in Ref. [24]); a regulator-induced shift of even ~30% in kNp would change that comparison materially. The small N=256^3 boxes also mean the ultraviolet modes that feed the drift are only marginally resolved, so the whole 'freeze' at kNp(ηc→ηeq) could be a property of the regulated UV, not of the physical axion spectrum.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This Letter studies the nonlinear fate of pre-inflationary QCD axion dark matter in kinetic misalignment, treating the initial condition as a two-dimensional landscape in the initial angle Θ1 and velocity v. After mapping the zero-mode abundance and identifying a continuum of 'summit-landing' solutions, the authors perform large 3D relativistic lattice simulations of fragmentation for v≳100 and then switch to non-relativistic Gross-Pitaevskii simulations with a 'fattening' (FAT) regulator to follow the spectral peak through the QCD crossover toward matter-radiation equality. Their central claim is that for fA below about 2×10^10 GeV, strong self-interactions erase the early-history memory and drive the small-scale spectrum to the same attractor previously found for post-inflationary axions, with the peak position at matter-radiation equality determined only by fA and ΩA; this may extend to fA~5.7×10^10 GeV. The paper also reports that fragmentation reduces the comoving axion number by a velocity-dependent dilution factor D(v) and calibrates the velocity needed to reproduce the observed relic density.","tokens_in":44422,"tokens_out":5059,"duration_ms":50302,"significance":"If the central claim holds, the paper is significant: it would imply that kinetic misalignment and post-inflationary axion production produce equivalent small-scale dark matter structure for a wide range of axion masses, directly affecting minicluster and axion-star predictions. The work is also valuable for its extensive 3D simulations, explicit falsifiable scaling relations, and publicly available analysis tools in the companion repositories. The fragmentation dilution factor is supported by an independent lattice study, and the paper is commendably explicit about several numerical limitations. However, the central amnesia claim rests on the non-relativistic continuation of the peak drift, and that continuation depends on a fitted coefficient and an ad hoc regulator, so the significance is conditional on resolving the concerns below.","major_comments":[{"comment":"The predicted peak drift is not parameter-free. The text states that Eq. (4) is multiplied by a free coefficient Cfit and fitted to the late relativistic peak evolution (Cfit~0.5–1), so the reported agreement between the model and the relativistic simulations is partly by construction. Moreover, the NR runs are initialized with a spectrum peaked at the value extrapolated from the same linear/√(mA/R) model, so their subsequent evolution is not an independent test of the drift law. Please provide a posterior predictive check with Cfit fixed by the scattering calculation, or show explicitly how the attractor conclusion depends on the fitted coefficient.","section":"§4, Eq. (4)"},{"comment":"The FAT regulator replaces mψ only in the resummed Bessel potential, creating dense-axion-star (DAS) configurations at the grid scale, and the physical peak kNp is not measured directly: it is extracted from a two-component fit that subtracts a Gaussian DAS peak. The Supplemental Material itself concedes that the finite width of the DAS mass distribution can shift the extracted peak and that the regulator is not unique. All 15 runs per fA share the same FAT class and the same fitting procedure, so run-to-run stability cannot exclude a common bias; with N=256^3 the ultraviolet modes feeding the drift are only marginally resolved. A regulator-induced shift of even ~30% in kNp would materially change the comparison with Ref. [24], whose exponent is fA^-0.5 while this work finds fA^-0.73. Please validate the drift with an alternative regulator or a direct peak measurement without DAS subtraction.","section":"Supplemental Eq. (S128) and Fig. S9"},{"comment":"The paper explicitly states that for velocities leading to ρ0/ρ≲0.1 no clear convergence of ΩAh2 at η≳4η1 was found with feasible L and N, and that the quoted values 'potentially involve huge uncertainties.' This non-convergence weakens the treatment of the summit-landing branch, which is nevertheless presented as part of the misalignment landscape in Fig. 1 and used in the Discussion to argue that both Θ1 and v matter. The summit-landing power spectra in Fig. S6 also show features that depend on simulation parameters. The quantitative claims about this branch are therefore not supported by the present simulations.","section":"Supplemental Material, summit-landing section (text after Fig. S4)"},{"comment":"The dilution factor D(v) is a six-parameter fit to simulations at only two values of fA, and the quoted uncertainties are statistical only. This fit is then used to set the initial velocities of the large-scale runs whose spectra determine the attractor and the fA range over which amnesia holds. A systematic bias in D(v) would propagate into the initial velocities and hence into the reported kNp(fA) relation. Please quantify the systematic error, for example by varying the fit form and by including the independent lattice results of Ref. [51] in the calibration.","section":"Supplemental Eqs. (S110)–(S112) and Table S2"},{"comment":"The claimed equivalence to the post-inflationary attractor is not quantitatively demonstrated. The paper quotes kNp/kJ ~ 1.78(10^10 GeV/fA)^0.73 while Ref. [24] gives ~(10^10 GeV/fA)^0.5; these differ by a factor of order fA^0.23, which is about a factor 3 over the plotted decade. Given that this exponent is the quantitative basis for the statement that the spectra are 'equivalent within the uncertainties', the mismatch should be addressed explicitly, and the uncertainty on the exponent should be propagated into the conclusion.","section":"Discussion, comparison with Ref. [24]"}],"minor_comments":[{"comment":"The symbol PΘ is used in the figure but defined only through Eq. (S39); please define it directly in the caption as the integrated tail probability over initial angles.","section":"Fig. 2 caption"},{"comment":"The displayed exponents in Eq. (2) are typeset ambiguously (e.g., 'n+6 n+4'); please follow the notation of Supplemental Eq. (S21) to make the formula readable.","section":"Main text, Eq. (2)"},{"comment":"The phrase 'over the theory model∼√mA/R' is incomplete; please specify which curves correspond to the √(mA/R) prediction and which to the linear approximation.","section":"Fig. 4 caption"},{"comment":"The fitted sigmoid has a very steep slope n=20.0(9); a plot of the fit together with the residuals, or an alternative fit form, would help the reader judge the robustness of D(v).","section":"Supplemental Eq. (S110)"},{"comment":"The text says 'for fA > 5×10^9 GeV, where Ref. [24] reports results', but the comparison in Fig. S13 appears to rely on an extrapolation of Ref. [24]; please clarify the fA range in which the comparison is direct rather than extrapolated.","section":"Discussion"}],"recommendation":"major_revision","confidential_remarks":"The central claim is interesting and timely, but the paper currently leans on two coupled numerical choices: the fitted Cfit in Eq. (4) and the ad hoc FAT regulator in the NR continuation. If the authors can provide an alternative validation (e.g., a run with a different regulator or a direct measurement of the physical peak without DAS subtraction), I would support publication. As it stands, the amnesia conclusion is plausible but not yet robust enough for the main claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's what you need to know about arXiv:2608.04139. It does two things. First, it maps the pre-inflationary axion misalignment landscape in the (Theta_1, v) plane and identifies a continuum of \"summit-landing\" solutions interpolating between large misalignment and kinetic misalignment. That part is genuinely new, and the analytic treatment of the overshoot and pre-summit branches, with the logarithmic enhancement of Omega_A near the hilltop, is clean. Second, it follows kinetic-misalignment fragmentation on the lattice through the QCD crossover into the non-relativistic regime to matter-radiation equality, arguing for an \"amnesia\" attractor: for fA below about 2e10 GeV, the small-scale spectrum converges to the post-inflationary result of Ref. [24], with the spectral peak set by fA and Omega_A alone.\n\nCredit where it's due. The fragmentation dilution factor D(v) (0.2-0.3 at large v) agrees with the independent lattice study [51]. The onset of fragmentation at v~150 is reproduced by a Lambert-W analytic estimate, which is a nice cross-check. The authors ship their cosmology module and simulation tools, and the text is unusually candid about limitations, especially the non-convergence of the summit-landing runs.\n\nNow the soft spots, in proportion. The headline amnesia claim is less nailed down than the abstract suggests. Equation (4) is multiplied by a free coefficient Cfit fitted to the late relativistic peak, and the D(v) calibration sets the initial conditions of those same runs, so the absolute normalization of the predicted drift is calibrated, not predicted. The interesting outputs--the linear drift, the freeze after the crossover, and the fA-exponent--do come from the simulations, but the quoted exponent ratio (0.73 vs. 0.5 in Ref. [24]) shows the attractor agreement is not exact on their own numbers. Second, the load-bearing NR continuation uses the FAT regulator, and the physical peak is extracted by subtracting an artificial dense-axion-star Gaussian sitting just below the grid scale. The authors vary the regulator and box size and find stability, but every run shares the same regulator and fit procedure, so a common bias would not appear as scatter. The supplemental concedes the DAS width \"can shift the extracted peak,\" and with 256^3 boxes the UV is marginal; a shift of a few tens of percent in kNp would materially change the comparison with Ref. [24]. That concern is real, not manufactured. The summit-landing non-convergence is a separate weakness, but it doesn't carry the main claim.\n\nWho gets value: anyone working on axion miniclusters, kinetic misalignment, or small-scale structure probes of axion dark matter. It deserves a serious referee. My recommendation: accept the central fragmentation result as solid, and ask for a systematic error budget on the peak extraction, a convergence study of the summit-landing branch, and a quantified comparison with Ref. [24]'s exponent before the amnesia statement goes into the abstract.","headline":"The summit-landing map and fragmentation physics are new and solid, but the headline amnesia claim rests on calibrated, regulator-dependent peak extraction that still needs an error budget.","tokens_in":45017,"tokens_out":7777,"would_cite":true,"duration_ms":63759,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"For axion decay constants below about 2×10^10 GeV, kinetic-misalignment axions fragment and then self-interact into the same small-scale spectrum as post-inflationary axions, erasing the initial misalignment history by matter–radiation…","keywords":["QCD axion","kinetic misalignment","axion fragmentation","axion miniclusters","axion stars","misalignment landscape","lattice simulation","small-scale dark matter structure"],"falsifier":"Rerun the non-relativistic evolution without the fattening regulator, or with a range of $\\mu_{\\rm fat}$ and box sizes such as $N=512^3$ and $1024^3$, and check whether the fitted peak $k_p^N(\\eta_c)$ still follows the $30\\,\\mathrm{mpc}^{-1}(10^{10}\\,\\mathrm{GeV}/f_A)^{1.29}$ scaling; a regulator- or box-size-dependent peak would falsify the attractor claim. Alternatively, follow the relativistic simulation through the QCD crossover far enough to resolve axiton collapse directly and compare the resulting spectrum at matter–radiation equality with the post-inflationary attractor.","tokens_in":43888,"feed_emoji":"🌌","tokens_out":11886,"duration_ms":93281,"temperature":0.7,"pith_summary":"This paper tries to show that the QCD axion's small-scale dark-matter structure can be insensitive to how the axion was produced. In the pre-inflationary kinetic misalignment scenario, the axion starts with a large field velocity and traverses many potential barriers, so its evolution depends on both the initial angle $\\Theta_1$ and velocity $v$; the authors map this two-dimensional landscape and identify fine-tuned 'summit-landing' branches where the field stops near a potential maximum. Using lattice simulations that follow fragmentation through the QCD crossover into the non-relativistic regime, they find that nonlinear fragmentation reduces the relic abundance by a dilution factor $D\\sim 0.2$–$0.3$ at high velocity, and that strong self-interactions then drive the small-scale spectrum toward the same attractor previously found for post-inflationary axions. If this holds, for $f_A$ below about $2\\times10^{10}$ GeV the initial history is erased: kinetic misalignment and post-inflationary axions produce essentially equivalent small-scale structure at matter–radiation equality, and minicluster or axion-star observations cannot distinguish the two scenarios.","feed_headline":"Kinetic axion histories converge to one small-scale spectrum","feed_subtitle":"Fragmentation erases the axion's early history, so miniclusters may not reveal how it was made.","key_machinery":"The load-bearing mechanism is an attractor from marginal self-interaction scattering. Non-relativistic $2a\\to2a$ scattering with coupling $\\lambda=\\chi(T)/f_A^4$ redistributes axions toward higher momenta at a rate $\\Gamma$, and because $\\Gamma/H\\propto m_A/(k_p^2 R^2)$, the spectral peak grows until $\\Gamma/H\\sim1$, giving $k_p\\propto\\sqrt{m_A}/R$ during radiation domination. This is simulated with a two-stage lattice setup: a relativistic code runs through fragmentation until axitons shrink below the grid, then a non-relativistic envelope equation is evolved with the 'fattening' regulator, which replaces $m_\\psi$ by $\\mu_{\\rm fat}^2/(\\delta x^2 m_\\psi)$ only in the resummed potential and prevents unresolved wave-collapse into dense axion stars. The physical peak is extracted by masking the largest overdensities and fitting a two-component model that subtracts the artificial dense-axion-star peak. The paper also introduces the 'summit-landing' branch of the $(\\Theta_1,v)$ landscape, where the field approaches a potential maximum with small residual velocity; the long residence near the unstable maximum produces a large amplification of fluctuations and a logarithmically enhanced relic abundance.","core_discovery":"The central discovery is that the late-time axion spectrum is an attractor: after fragmentation, the spectrum—and especially its peak—is driven to a value set by $f_A$ and $\\Omega_A$ and largely independent of the previous history. The paper reports that the comoving number-spectrum peak settles at $k_p^N(\\eta_c\\to\\eta_{\\rm eq})\\sim 30\\,\\mathrm{mpc}^{-1}(10^{10}\\,\\mathrm{GeV}/f_A)^{1.29}$ for $f_A\\in(3\\times10^8,\\,10^{10})$ GeV, following the marginal-scattering trajectory $k_p\\propto\\sqrt{m_A}/R$ with a fitted coefficient $C_{\\rm fit}\\sim0.5$–$1$. The resulting density-fluctuation peak lies above the quantum Jeans wavenumber for $f_A\\lesssim6\\times10^{10}$ GeV, so gravitational collapse into miniclusters is strongly affected by quantum pressure. The paper argues that this convergence is robust for $f_A\\lesssim2\\times10^{10}$ GeV and may extend to $f_A\\sim5.7\\times10^{10}$ GeV; at larger $f_A$ self-interactions are too weak to erase the initial conditions, leaving observable memory of the misalignment landscape.","pith_inferences":["If the attractor is robust, axion small-scale structure becomes a function of $(f_A,\\Omega_A)$ alone; one could then derive the minicluster mass function directly from the attractor spectrum and translate any null detection of miniclusters into a constraint on $f_A$ independent of the production mechanism.","The summit-landing instability is a finite-time hilltop effect; applying the same lattice pipeline to axion-like particles with different temperature-dependent masses (different exponent $n$ and crossover timing) would predict which mass ranges retain initial-condition memory, a parameter scan the paper does not perform.","Because the FAT regulator is ad hoc, a natural extension is to compare the extracted peak against an independent relativistic simulation that resolves axiton collapse for one representative $f_A$, checking explicitly whether the two-component dense-axion-star subtraction is unbiased."],"forward_implications":["For $f_A\\lesssim2\\times10^{10}$ GeV, the spectra from kinetic misalignment and post-inflationary axions coincide at matter–radiation equality, so minicluster and axion-star observations cannot distinguish the two production histories.","The velocity-dependent dilution factor $D(v)\\sim0.2$–$0.3$ means homogeneous estimates overpredict the axion relic density by up to a factor of about four, and for $2.15\\times10^{10}\\lesssim f_A/\\mathrm{GeV}\\lesssim2.48\\times10^{10}$ two different velocities reproduce the observed abundance.","For $f_A\\lesssim6\\times10^{10}$ GeV the density-fluctuation peak lies above the quantum Jeans scale, so gravitational collapse into miniclusters is delayed and central densities are lowered relative to the standard matter–radiation-equality collapse.","At larger $f_A$ (roughly above $5.7\\times10^{10}$ GeV) self-interactions decouple before erasing the history, so the initial angle and velocity remain observable and different misalignment histories can produce different small-scale structures."],"supporting_citations":[{"why":"Supplies the post-inflationary axion spectrum attractor, the $2a\\to 2a$ scattering rate used to model peak drift, and the structure-formation analysis applied to the final spectra.","marker":"[24]"},{"why":"Provides the adiabatic axion initial conditions and the kinetic-fragmentation framework (tachyonic and parametric resonances, kination cutoff) that the simulations start from.","marker":"[21]"},{"why":"Gives the high-temperature lattice QCD topological susceptibility and thermal degrees of freedom used to set $m_A(T)$ and the QCD crossover timing.","marker":"[30]"},{"why":"Furnishes the cosmological tables and early-seeds simulation methodology for axion miniclusters that the relativistic runs follow.","marker":"[28]"},{"why":"Supplies the lattice code used for the relativistic field simulations throughout the paper.","marker":"[50]"},{"why":"Provides independent lattice results for QCD-axion kinetic fragmentation used to check the dilution factor $D(v)$.","marker":"[51]"},{"why":"Defines dense axion star configurations used as the regulated late-time objects in the non-relativistic simulations.","marker":"[53]"}],"fun_headline_variants":["Axion fragmentation wipes out early history, unifying spectra","Fragmentation erases axion's past: late-time spectra converge","Minicluster spectra forget their axion misalignment origins","Attractor spectrum: axion histories converge after fragmentation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The late-time conclusion rests on the assumption that the 'fattening' regulator used in the non-relativistic simulations—replacing the axion mass by a lattice-scale value only in the potential to halt wave collapse—does not bias the drift of the physical spectral peak, and that the small $256^3$ boxes resolve the relevant ultraviolet dynamics.","fun_headline_variants_meta":{"raw":{"variants":["Axion fragmentation wipes out early history, unifying spectra","Fragmentation erases axion's past: late-time spectra converge","Minicluster spectra forget their axion misalignment origins","Attractor spectrum: axion histories converge after fragmentation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00022,"raw_usage":{"total_tokens":1446,"prompt_tokens":946,"completion_tokens":500,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":562,"completion_tokens_details":{"reasoning_tokens":431}},"tokens_in":562,"tokens_out":500,"duration_ms":4755,"temperature":1.0,"reasoning_tokens":431,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:42:18.849505+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Rerun the non-relativistic evolution without the fattening regulator, or with a range of $\\mu_{\\rm fat}$ and box sizes such as $N=512^3$ and $1024^3$, and check whether the fitted peak $k_p^N(\\eta_c)$ still follows the $30\\,\\mathrm{mpc}^{-1}(10^{10}\\,\\mathrm{GeV}/f_A)^{1.29}$ scaling; a regulator- or box-size-dependent peak would falsify the attractor claim. Alternatively, follow the relativistic simulation through the QCD crossover far enough to resolve axiton collapse directly and compare the resulting spectrum at matter–radiation equality with the post-inflationary attractor.","supporting_citations":[{"cited_title":"Kinetic fragmentation of the QCD axion on the lattice","cited_arxiv_id":null,"evidence_quote":"Defines dense axion star configurations used as the regulated late-time objects in the non-relativistic simulations."}],"review_version":1}