{"id":"586edbee-9efc-47bc-b9d8-ebc723b3c82e","arxiv_id":"2507.16745","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Grid-based initial conditions imprint roughly one-percent alignment artifacts on simulated halo shapes, with a redshift-dependent sign flip whose origin is not yet pinned down.","lead":"Cosmological simulations that start dark matter particles on a Cartesian grid leave a small but statistically significant imprint on the shapes of the halos they form. The effect is roughly one percent, so it is easy to overlook, but it can bias precision studies of galaxy alignments and cosmic structure morphology.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Glass control does not isolate the pre-IC variable: Kun differs from every grid run in cosmology, box size, resolution, code, and halo finder, so the causal claim that artifacts arise solely from grid-type initial conditions is not established.","rationale":"The reader's conditional verdict already identifies the same primary weakness: the single glass control, Kun, is confounded by cosmology, box size, resolution, code, and halo finder. My stress-test concurs. The detection itself has reasonable support from multiple grid simulations and from particle tracing, which shows initial alignment with the Cartesian axes; the concern is specifically about the causal attribution and the universality of the low-redshift sign. Since MDPL2 shows an opposite sign at z<1, the abstract's general statement that halos preferentially orient away from the axes at low redshift is not robust without an explanation of the MDPL2 reversal. A controlled grid-vs-glass pair would settle whether the effect is genuinely due to the pre-IC choice. Until then, the paper should be read as reporting a plausible but not fully isolated artifact, exactly matching the conditional verdict.","tokens_in":12823,"tokens_out":3943,"duration_ms":46865,"concrete_test":"Run a matched pair of simulations with the same code (e.g., Gadget-4), identical box size, particle number, cosmology, transfer function, random seed, and initial displacement field, changing only the pre-IC load from grid to glass. Apply the same FoF finder (b=0.2) and compute the same inertia-tensor major-axis anisotropy at z=0, 0.5, 1, and 2. If the grid run reproduces the away-from-axes signal and the glass run is null, the causal claim is supported; if both show the signal, or neither does, the current evidence does not isolate the pre-IC variable.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that grid pre-initial conditions are the cause of the halo-shape alignment, not merely a confounded correlate. In Sec. 3.4 the paper concludes that the patterns in CosmicGrowth, ELUCID, Uchuu1000Pl18, and MDPL2 arise solely from grid-type initial conditions because the single glass-run Kun shows no signal. But Kun differs from the grid simulations in essentially every other relevant variable: cosmology (Om=0.3111 vs 0.268-0.307), box size (1000 vs 500-600 h^-1 Mpc), particle mass (2.96e9 vs 3.09e8-1.51e9 Msun/h), force code (Gadget-4 vs Jing/Gadget-2/GreeM), and FoF implementation. The null result in Kun therefore cannot isolate grid vs glass. A cosmology with weaker anisotropic clustering, or a different force solver, could suppress a grid artifact, while a physical/cosmology-dependent alignment could masquerade as a numerical one. The headline pattern is also not universal: MDPL2 in Sec. 3.1 shows the opposite sign for z<1. Unless the pre-IC is varied with all other settings fixed, the causal attribution is unsupported and the abstract overstates universality.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper examines whether the choice of pre-initial conditions (grid vs. glass particle loads) leaves detectable artifacts in the shapes of dark matter halos in cosmological N-body simulations. Using halo catalogs from five simulations (CosmicGrowth, ELUCID, Uchuu1000Pl18, MDPL2, and Kun fiducial), the authors construct smoothed HEALPix maps of halo major-axis orientations and measure the excess probability of alignment with the simulation-box Cartesian axes. They report a redshift-dependent signal: at low redshift (z<2), halo shapes in most grid-initialized simulations preferentially avoid the Cartesian axes, while the constituent particles of present-day halos preferentially align with those axes at early times. Filament orientations also show a persistent grid-axis alignment. A single glass-initialized simulation (Kun) shows no such signal. The paper proposes a qualitative explanation involving the averaging of progenitor alignments during mergers.","tokens_in":13087,"tokens_out":4117,"duration_ms":46309,"significance":"If the causal attribution is correct, the result is relevant for precision analyses of halo and galaxy shapes, intrinsic alignments, and mock catalog construction, where even a ~1% box-axis-dependent bias could matter for percent-level statistics. The paper has notable strengths: it uses multiple independent public simulations, checks two different force solvers and two halo finders, and supports the main detection with Poisson error bars in the excess-probability measurement. The time-tracing of proto-halo particles and the filament analysis are informative. However, the central causal claim that the patterns arise solely from grid-type initial conditions is not established by the current control, and the behavior in MDPL2 shows an opposite sign, so the claimed universality is not yet supported.","major_comments":[{"comment":"The conclusion that the alignment patterns in CosmicGrowth, ELUCID, Uchuu1000Pl18, and MDPL2 \"arise solely from grid-type initial conditions\" is not supported by the Kun control as presented. Kun differs from every grid simulation in cosmology (Omega_m=0.3111 versus 0.258-0.3089), box size (1000 h^-1 Mpc versus 500-600 h^-1 Mpc), particle mass (2.96e9 versus 3.09e8-1.51e9 Msun/h), simulation code (Gadget-4 versus Jing/Gadget-2/GreeM), and in some comparisons halo finder. A null result under this combination of simultaneous changes cannot isolate the pre-initial-condition type. A matched control with identical code, cosmology, resolution, and halo finder, or at least a grid-initialized run from the Kun suite, is needed before the abstract's causal statement is justified.","section":"Sec. 3.4, Table 1"},{"comment":"The abstract and conclusions present the low-redshift avoidance of Cartesian axes as the generic grid-initialization signature, but MDPL2, also grid-initialized, shows the opposite sign at z<1. This is acknowledged in Sec. 3.1 and attributed to a possible difference in the initial-condition generator, but no supporting test is provided. Without a comparison of initial-condition generators (e.g., Ginnungagap versus NGenIC or 2LPTic) on an otherwise identical setup, the claimed universality of the artifact pattern is not established, and the interpretation of the low-z pattern as the characteristic grid-load signature remains open.","section":"Sec. 3.1, Fig. 3; Sec. 4"},{"comment":"The statistical significance of the map-level signal is not quantified. The sigma values quoted in the figures are the normalized amplitudes of maps smoothed with a 10-degree Gaussian kernel, but no null-hypothesis test is performed on the smoothed maps, the smoothing scale is described only as \"optimized,\" and no correction is made for the large number of pixels and the three tested axes. In Fig. 7, the error bars are Poisson errors on the binned cosine distribution; they do not capture the covariance between bins, between the three Cartesian axes, or the uncertainty in the map normalization. A robust significance estimate, for example from shuffled or un-smoothed maps, is needed to support the claim that the effect is statistically significant at the map level.","section":"Sec. 2.4, Figs. 1-7"},{"comment":"The proposed explanation for the low-redshift sign flip is illustrative rather than tested. The Gaussian-averaging example is not fitted to the measured excess probabilities, and the merger-tree categorization that would test the hypothesis is deferred to future work. The manuscript should state more explicitly that this is a qualitative conjecture, not a quantitative explanation, and should not present the zero-crossing argument in Sec. 3.5 as independent confirmation of the merger mechanism.","section":"Sec. 3.6"}],"minor_comments":[{"comment":"There are typos in this section: \"condtions\" should be \"conditions\" and \"signifcance\" should be \"significance.\"","section":"Sec. 3.1"},{"comment":"The description of the Gaussian smoothing would be clearer if the authors specified whether the 10-degree scale is the FWHM or the standard deviation of the kernel.","section":"Sec. 2.4"},{"comment":"The right panel of Fig. 7 is described as tracing particles from z=6 to z=0, but the same panel is also used to claim a consistent zero-crossing; a quantitative criterion for the zero-crossing (e.g., where the excess probability crosses unity) would help the reader assess this claim.","section":"Fig. 7"},{"comment":"The statement that halo particles \"initially prefer to align with filaments, and then align out of filaments at low redshift\" is an inference from the separate filament and halo measurements; a direct measurement of the angle between halo major axes and the nearest filament orientation would make this connection quantitative.","section":"Sec. 3.3"},{"comment":"The stacking procedure for the three Cartesian axes is described in words but not shown as an equation; writing out the stacked estimator would make the definition of the reported 1%, 0.8%, and 0.5% amplitudes unambiguous.","section":"Sec. 3.5"}],"recommendation":"major_revision","confidential_remarks":"The empirical detection is suggestive and worth pursuing, but the causal claim is overreaching relative to the control design. Since the authors have access to the Kun simulation suite, a grid-loaded Kun run, or a small set of matched grid/glass pairs, would be a natural and feasible way to strengthen the paper. The MDPL2 sign reversal should be addressed explicitly in the abstract or conclusions, rather than only as a puzzling aside. The paper is within the scope of the journal, and the topic is of current interest to the simulation and intrinsic-alignment communities."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the quick take: the paper credibly reports a small (~1%) redshift-dependent alignment of halo shapes with the simulation box axes in grid-initialized runs, and it deserves a proper referee; but the abstract's causal claim and universality outrun the actual evidence.\n\nWhat is genuinely new is the sign flip. Proto-halo particles start aligned with the Cartesian axes, and by z<2 the halo major axes prefer to point away from the axes—a pattern that is not in the cited pre-IC literature (void artifacts, power-spectrum differences, high-z tidal response, halo property variations). Tracing the particle positions back in time (Fig 4) is a nice move, showing the early alignment persisting and then reversing. Running the same statistic across four grid simulations with two force schemes and two halo finders is the right instinct, and the paper is honestly upfront that the amplitude is too small to matter for most cosmological studies, with pairwise contamination suppressed to about 0.01%.\n\nThe soft spots are in the causal attribution, not in the raw detection. The single glass control, Kun, differs from every grid run in cosmology, box size, mass resolution, code, and halo finder. A null in Kun cannot isolate grid versus glass, so the sentence in Sec 3.4 that the patterns 'arise solely from grid-type initial conditions' is not supported. The stress-test note is right on this. Just as important, MDPL2 shows the opposite sign at z<1; the authors acknowledge it and hypothesize an IC-generator peculiarity, but that is untested, so the low-z 'away from axes' pattern is not established as universal.\n\nSmaller quibbles: the 10-degree smoothing is tuned to maximize the signal, with no map-level covariance behind the significance claims, only Poisson errors on the 1D excess. The Sec 3.6 flipping mechanism is a toy model the authors themselves call overly simplistic, and no code or catalogs are released, which slows independent checks.\n\nWho this is for: simulation methodologists and anyone building intrinsic-alignment or shape mocks. The central cautionary finding—that grid pre-ICs leave a small box-aligned imprint in halo shapes—is credible. I'd send this to a serious referee and expect a revise rather than an accept: either vary the pre-IC with all other settings fixed, or soften the language from 'solely' to 'consistent with.' If I worked on shape statistics I would cite it as a cautionary floor.","headline":"Credible detection of a small, sign-flipping box-axis artifact in halo shapes from grid initial conditions, but the glass control does not isolate grid-vs-glass and MDPL2 undercuts the universality claim.","tokens_in":13578,"tokens_out":4551,"would_cite":true,"duration_ms":46089,"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":"This paper establishes that grid-type initial conditions used to set up cosmological simulations imprint a statistically significant, redshift-dependent orientation bias on dark matter halo shapes, flipping from box-axis alignment at high…","keywords":["cosmological N-body simulations","pre-initial conditions","grid-type initial conditions","glass-type initial conditions","halo shapes","halo alignment","numerical artifacts","intrinsic alignment"],"falsifier":"Run the same halo-shape orientation analysis on paired simulations that share identical cosmology, box size, resolution, code, and halo finder, differing only in grid versus glass pre-initial loading; if the grid run shows no ~1% axis-avoidance at z<2 or the glass run shows a similar pattern, the causal claim fails.","tokens_in":12605,"feed_emoji":"🔭","tokens_out":8153,"duration_ms":76709,"temperature":0.7,"pith_summary":"Cosmological simulations often start dark matter particles on a simple Cartesian grid, an approximation usually assumed to wash out by the time halos form. This paper argues that the grid leaves a small but real mark: the major axes of dark matter halos in grid-initialized simulations show a statistically significant ~1% preference to point away from the simulation box's Cartesian axes at redshifts z<2, even though the same particles began with the opposite preference, aligned with those axes. The flip is traced through time and appears across multiple simulation codes and halo finders, while a glass-initialized comparison simulation shows no such signal. If right, the finding means orientation-sensitive statistics from grid-based simulations carry a subtle numerical bias, negligible for many studies but relevant for precision alignment work. The paper leaves the mechanism of the sign flip as a hypothesis for future work.","feed_headline":"Grid-initialized simulations leave a ~1% imprint on halo shapes","feed_subtitle":"A statistically significant 1% bias at z<2 links halo shapes to the simulation grid.","key_machinery":"The analysis rests on the inertia tensor of each halo's particle positions: eigen-decomposition yields the halo major axis, and the distribution of these axes is binned on HEALPix sky maps and smoothed with a 10-degree Gaussian kernel to expose directional preferences. The preference is quantified as an excess probability of the cosine angle between the halo axis and the box's Cartesian axes. To connect the late-time pattern to the initial condition, the paper traces halo particles back to their earlier redshifts, and identifies filament orientations via a persistent-homology-based filament finder applied to halo catalogs. The proposed flipping mechanism is illustrated with a simple Gaussian-variable average model, where combining two axis-aligned Gaussian preferences shifts the resultant to an off-axis direction, though the authors call this model overly simplistic.","core_discovery":"The paper's central claim is that grid-type pre-initial conditions produce statistically significant artificial alignment in halo shapes, with the sign of the effect flipping between early and late times. At z<2, halo major axes preferentially avoid the simulation box's Cartesian axes, at an amplitude of about 1% with significance around -3σ; tracing the particles back shows that at high redshift the same axes preferentially align with the box axes. The amplitude is small enough to be negligible for most cosmological statistics, but the paper demonstrates the effect persists across P3M and TreePM codes, FoF and Rockstar halo finders, and several simulation suites, and is absent in a glass-initialized simulation. The authors propose that the reversal reflects the cumulative alignment artifacts of merging progenitors, though they note the explanation is preliminary and one simulation (MDPL2) shows an inverted pattern.","pith_inferences":["If the grid artifact is real, precision weak-lensing intrinsic alignment measurements using simulated shape catalogs may need a box-axis correction; the paper itself notes the effect on pairwise alignments would be suppressed to ~0.01%, but single-direction statistics could still be biased.","The reversed sign in MDPL2 hints that a property of the specific grid construction—such as grid offset, particle ordering in the displacement field, or the 1LPT vs 2LPT perturbation step—controls the sign; swapping grid offsets in a controlled run could pin down the mechanism.","The minor axes show stronger early-time artifacts than major axes, suggesting shape estimators that weight the minor axis (e.g., reduced inertia tensor) may change the measured magnitude; this is directly testable on the same halos.","The proposed merger-history mechanism predicts that low-redshift halos split by last major merger time should show opposite alignment signs; constructing merger trees from the same simulations would settle the hypothesis without new runs."],"forward_implications":["Halo shape and galaxy intrinsic alignment catalogs from grid-initialized simulations should be checked for a ~1% box-axis-dependent orientation bias at z<2.","Because filaments in grid simulations show a persistent alignment with the box axes, cosmic-web orientation studies are also susceptible to this numerical imprint.","Glass-type pre-initial conditions appear preferable for anisotropic statistics, while isotropic statistics remain largely unaffected.","The persistence of the artifact across P3M and TreePM codes and FoF and Rockstar halo finders indicates it is a property of the grid load itself, not of any single pipeline.","The sign flip from initial alignment to late-time avoidance contradicts the common assumption that virialization erases all memory of the initial particle load in halo shapes."],"supporting_citations":[{"why":"Introduces the grid-based initial particle loads whose imprint the paper claims to detect in halo shapes.","marker":"[16]"},{"why":"Defines the glass-like pre-initial distribution used by the Kun simulation as the artifact-free comparison.","marker":"[17]"},{"why":"Documents persistent grid imprints in void regions, showing grid patterns can survive to late times.","marker":"[18]"},{"why":"Reports less than 1% differences in matter power spectrum and halo mass function between grid and glass initial conditions, the assumption this paper challenges.","marker":"[22]"},{"why":"Found an enhanced anisotropic tidal response at z>=9 for grid loads, motivating the search for anisotropic artifacts.","marker":"[23]"},{"why":"Demonstrated measurable effects of the pre-initial condition choice on halo properties, supporting the plausibility of shape artifacts.","marker":"[27]"},{"why":"Describes the CosmicGrowth simulation suite that provides the primary grid-initialized halo sample in this work.","marker":"[28]"},{"why":"Gives the few-percent-level intrinsic alignment measurements used to contextualize the ~1% artifact amplitude.","marker":"[36]"}],"fun_headline_variants":["Halo shapes betray their simulation grid at 1% level","Grid initialization flips halo alignment in simulations","Simulation grid imprints halo shapes with 1% bias","Halo axes dodge box axes due to grid artifacts","1% grid artifact flips halo shape alignment"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The conclusion that these alignment patterns arise solely from grid-type initial conditions depends on one glass-initialized control simulation that differs from the grid runs in cosmology, box size, resolution, simulation code, and halo finder.","fun_headline_variants_meta":{"raw":{"variants":["Halo shapes betray their simulation grid at 1% level","Grid initialization flips halo alignment in simulations","Simulation grid imprints halo shapes with 1% bias","Halo axes dodge box axes due to grid artifacts","1% grid artifact flips halo shape alignment"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000237,"raw_usage":{"total_tokens":1446,"prompt_tokens":825,"completion_tokens":621,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":441,"completion_tokens_details":{"reasoning_tokens":543}},"tokens_in":441,"tokens_out":621,"duration_ms":7323,"temperature":1.0,"reasoning_tokens":543,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:02:33.756142+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same halo-shape orientation analysis on paired simulations that share identical cosmology, box size, resolution, code, and halo finder, differing only in grid versus glass pre-initial loading; if the grid run shows no ~1% axis-avoidance at z<2 or the glass run shows a similar pattern, the causal claim fails.","supporting_citations":[{"cited_title":"Efstathiou, M","cited_arxiv_id":null,"evidence_quote":"Introduces the grid-based initial particle loads whose imprint the paper claims to detect in halo shapes."},{"cited_title":"A comparison of the evolution of the density field in perturbation theory and numerical simulations - II Counts in cells analysis","cited_arxiv_id":"astro-ph/9408057","evidence_quote":"Documents persistent grid imprints in void regions, showing grid patterns can survive to late times."},{"cited_title":"Effects of the initial conditions on cosmological $N$-body simulations","cited_arxiv_id":"1401.6180","evidence_quote":"Reports less than 1% differences in matter power spectrum and halo mass function between grid and glass initial conditions, the assumption this paper challenges."},{"cited_title":"Impacts of pre-initial conditions on anisotropic separate universe simulations: a boosted tidal response in the epoch of reionization","cited_arxiv_id":"2007.08727","evidence_quote":"Found an enhanced anisotropic tidal response at z>=9 for grid loads, motivating the search for anisotropic artifacts."},{"cited_title":"Numerical convergence of pre-initial conditions on dark matter halo properties","cited_arxiv_id":"2109.02904","evidence_quote":"Demonstrated measurable effects of the pre-initial condition choice on halo properties, supporting the plausibility of shape artifacts."},{"cited_title":"Intrinsic alignment of simulated galaxies in the cosmic web: implications for weak lensing surveys","cited_arxiv_id":"1406.4668","evidence_quote":"Gives the few-percent-level intrinsic alignment measurements used to contextualize the ~1% artifact amplitude."}],"review_version":1}