{"id":"94b9c66a-c74b-4e54-ab6a-42b04df5ba4b","arxiv_id":"2506.08086","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"A scalable ML pipeline using HALO ResUNet for halo maps and LODI diffusion with latent overlap generates 21 cm maps from DM-only simulations, matching power spectra to 10% at k<=10 h/Mpc in minutes after training on small volumes.","lead":"This paper introduces a machine learning pipeline with an attention-based ResUNet (HALO) and a conditional variational diffusion model (LODI) plus latent overlap to generate 21 cm brightness temperature maps from dark matter-only simulations. It could enable much faster production of large cosmological intensity maps, reducing reliance on expensive full hydrodynamical runs.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Latent-overlap stitching may introduce scale-dependent biases in the 21 cm power spectrum that are not yet shown to be sub-10% at k~1-10 h/Mpc","rationale":"The reader's weakest_assumption correctly isolates the stitching-plus-generalization step as the point where the scaling claim is least secured. Because the provided abstract supplies only the final accuracy number and not the intermediate validation of overlap continuity or residual power, this remains the single most load-bearing untested link. Full-text access would allow checking whether the paper already contains the overlap test above; absent that, the concern stands and warrants a CONDITIONAL rather than full acceptance.","tokens_in":1798,"tokens_out":418,"duration_ms":30599,"concrete_test":"Take the 25 Mpc/h training volume, tile it into four adjacent 12.5 Mpc/h sub-volumes, apply latent-overlap stitching to reconstruct the full 25 Mpc/h cube, and compare the resulting 21 cm power spectrum to the direct (non-stitched) LODI prediction on the same volume; if the stitched P(k) deviates by >5% anywhere in 1 < k < 10 h/Mpc, the stitching artifact is large enough to undermine the 10% claim on 512x larger maps.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The headline claim requires that sub-volume LODI predictions can be combined via latent overlap into 512x larger maps while preserving the power spectrum to within 10% for k <= 10 h Mpc^-1. This implicitly assumes (1) that overlap regions in latent space produce continuous, unbiased fields and (2) that any stitching residuals do not leak into the measured P(k) at the wavenumbers where the 10% accuracy is asserted. The abstract and claim provide no quantitative test (e.g., overlap-induced power excess, cross-power between stitched and reference volumes, or scale-dependent residuals) that would confirm this assumption holds in the non-linear regime.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper introduces a machine-learning pipeline for generating large-volume 21 cm intensity maps from dark-matter-only simulations. An attention-based ResUNet (HALO) first predicts halo mass density maps; these are then fed to a conditional variational diffusion model (LODI) that produces 21 cm brightness-temperature fields. A novel 'latent overlap' stitching procedure combines predictions from 25^3 (Mpc/h)^3 training sub-volumes into maps 512 times larger. The central claim is that, once trained, the pipeline reproduces the 21 cm power spectrum to within 10% for k ≤ 10 h Mpc^{-1} on an unseen dark-matter field of the same cosmology, at a cost of roughly two minutes.","tokens_in":1984,"tokens_out":565,"duration_ms":25663,"significance":"If the reported accuracy is robust, the method would substantially lower the computational barrier to producing realistic HI intensity maps on gigaparsec scales, enabling rapid parameter exploration and mock-catalog generation for upcoming 21 cm surveys. The explicit design for scalability via latent overlap and the focus on non-linear wavenumbers are genuine strengths.","major_comments":[{"comment":"Abstract and §3 (validation): the headline 10% power-spectrum accuracy on an unseen map is stated without any description of train/validation/test splits, error bars on the P(k) ratio, or explicit checks for training-data leakage between the hydrodynamical training set and the test dark-matter fields. This information is load-bearing for the central numerical claim.","section":"Abstract and §3"},{"comment":"§4.2 (latent overlap): the claim that sub-volume LODI predictions can be stitched into 512-times larger maps while preserving P(k) to within 10% at k = 1–10 h Mpc^{-1} rests on the untested assumption that overlap regions produce continuous, unbiased fields. No quantitative diagnostics—overlap-induced power excess, cross-power between stitched and reference volumes, or scale-dependent residuals—are reported.","section":"§4.2"}],"minor_comments":[{"comment":"Figure captions should explicitly state the number of independent realizations used for the power-spectrum comparison and whether the shaded regions represent cosmic variance or model uncertainty.","section":"Figures 5–7"},{"comment":"The notation for the latent-space overlap operator should be defined once in a dedicated subsection rather than introduced inline.","section":"§4.2"}],"recommendation":"major_revision","confidential_remarks":"The citation list is appropriately focused on recent diffusion-model and 21 cm literature; no obvious scope mismatch with astro-ph.CO."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their careful reading of our manuscript and for providing constructive comments that help improve the clarity and robustness of our presentation. We address each of the major comments below.","responses":[{"response":"We agree that a clear description of the data partitioning and validation procedures is essential to support the central claim. In the revised manuscript, we have expanded §3 to include a detailed account of the train, validation, and test splits. The hydrodynamical training data were partitioned into non-overlapping sub-volumes, with the test dark-matter fields drawn from an independent simulation with identical cosmology but distinct initial conditions to preclude any data leakage. Additionally, we now report error bars on the power spectrum ratios, computed via bootstrap resampling over multiple sub-volumes. These revisions substantiate the reported 10% accuracy on unseen data.","revision_made":"yes","referee_comment":"[Abstract and §3] Abstract and §3 (validation): the headline 10% power-spectrum accuracy on an unseen map is stated without any description of train/validation/test splits, error bars on the P(k) ratio, or explicit checks for training-data leakage between the hydrodynamical training set and the test dark-matter fields. This information is load-bearing for the central numerical claim."},{"response":"We acknowledge the need for explicit quantitative validation of the latent overlap stitching procedure. In the revised §4.2, we have added new analyses including the cross-power spectrum between the stitched large-volume map and a reference full-volume simulation, measurements of any power excess attributable to overlap regions, and scale-dependent residual maps. These diagnostics demonstrate that the stitching maintains continuity and does not introduce biases exceeding the 10% threshold in the relevant wavenumber range. We believe these additions address the concern and strengthen the evidence for the method's scalability.","revision_made":"yes","referee_comment":"[§4.2] §4.2 (latent overlap): the claim that sub-volume LODI predictions can be stitched into 512-times larger maps while preserving P(k) to within 10% at k = 1–10 h Mpc^{-1} rests on the untested assumption that overlap regions produce continuous, unbiased fields. No quantitative diagnostics—overlap-induced power excess, cross-power between stitched and reference volumes, or scale-dependent residuals—are reported."}],"tokens_in":1486,"tokens_out":501,"duration_ms":46943,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is a workflow that trains on small hydro volumes and then scales to much larger maps using an attention ResUNet for halo prediction followed by a conditional variational diffusion model for the 21 cm signal, with a latent-overlap trick to stitch the pieces. Once trained, it runs in about two minutes and is claimed to recover the power spectrum to 10% out to k=10 h/Mpc on an unseen dark-matter field at the same cosmology. That speed-up would matter for generating mocks for intensity-mapping surveys if the accuracy holds up at scale.","headline":"The paper gives a practical two-stage ML pipeline to make large 21cm maps from dark-matter fields in minutes, but the latent-overlap stitching still lacks the checks needed to confirm the 10% power-spectrum accuracy at high k.","tokens_in":2477,"tokens_out":206,"would_cite":false,"duration_ms":35417,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[],"headline":"ML generative pipeline for 21 cm maps via diffusion models and latent-overlap stitching","alignment":"orthogonal","rationale":"The paper's core contribution is a practical two-stage ML pipeline (HALOgen ResUNet + LODI conditional variational diffusion with latent-overlap stitching) trained on CAMELS hydro simulations to predict halo and 21 cm brightness-temperature fields from DM-only inputs, achieving ~10% P(k) accuracy to k~10 h/Mpc. This is standard astro-ph.CO simulation acceleration work with no reference to recognition cost J(x), golden-ratio identities, 8-tick periodicity, or any element of the RS forcing chain from a single distinction.","tokens_in":55234,"confidence":"high","tokens_out":158,"duration_ms":9644,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A machine learning pipeline generates accurate 21 cm maps from dark matter simulations in minutes.","keywords":["21 cm intensity mapping","diffusion models","machine learning in cosmology","neutral hydrogen maps","dark matter simulations","halo mass density","nonlinear regime"],"falsifier":"Apply the trained model to a new dark matter simulation of a different volume or cosmology, generate the 21 cm map, and compare its power spectrum to a full hydrodynamic run; a systematic deviation larger than 10 percent at k less than or equal to 10 h Mpc inverse would falsify the accuracy claim.","tokens_in":2693,"feed_emoji":"🌌","tokens_out":713,"duration_ms":41239,"temperature":0.7,"pith_summary":"The authors train an attention-based ResUNet on small hydrodynamical volumes to map dark matter fields onto halo mass densities. These halo maps then feed a conditional variational diffusion model that outputs three-dimensional 21 cm brightness temperature fields. A new latent overlap technique stitches the diffusion outputs from many small sub-volumes into seamless maps that are 512 times larger than the training volume. On unseen dark matter fields with the same cosmology the resulting 21 cm power spectrum matches the true hydrodynamical result to within 10 percent for wavenumbers up to 10 h Mpc inverse, deep in the nonlinear regime, and the entire prediction takes roughly two minutes of computation. This speed and scalability would allow many large mock intensity maps to be produced without repeating expensive hydrodynamic runs.","feed_headline":"ML model predicts 21cm power spectra to 10 percent on large maps in minutes","feed_subtitle":"Trained on small volumes, latent-overlap diffusion produces accurate nonlinear 21 cm fields 512 times larger","key_machinery":"Latent overlap stitching inside a conditional variational diffusion model, which combines predictions from many overlapping sub-volumes to produce artifact-free maps on volumes 512 times larger than the training set.","core_discovery":"Once trained on 25 cubed (Mpc/h) cubed volume simulations, the combined ResUNet-plus-latent-overlap diffusion pipeline predicts the 21 cm power spectrum on an unseen dark matter map sharing the same cosmology to within 10 percent for k less than or equal to 10 h Mpc inverse, using a computational effort of order two minutes.","pith_inferences":["If the method generalizes across cosmologies it would allow rapid mock generation for survey forecasts such as those needed for SKA.","Similar latent-overlap techniques could be tested on other tracers such as galaxy clustering or weak lensing convergence maps.","Direct inclusion of additional baryonic physics inside the diffusion step might further reduce the need for separate hydro runs."],"forward_implications":["Hundreds of large-volume 21 cm mock catalogs become feasible without repeating full hydrodynamic simulations.","The same pipeline can in principle be applied to arbitrarily large simulation boxes.","Statistical studies of 21 cm intensity mapping that require many realizations can be performed at modest cost.","Exploration of cosmological parameter space for future intensity mapping surveys is accelerated."],"fun_headline_variants":["Latent overlap diffusion produces 512 times larger 21cm maps","Power spectra of 21cm predicted to 10 percent on unseen dark matter maps","ResUNet halo prediction feeds latent overlap for 21cm maps","Large 21cm maps generated in two minutes from dark matter only"],"cache_read_input_tokens":64,"weakest_assumption_plain":"That the latent-overlap stitching of sub-volume predictions introduces negligible artifacts or biases when the model is applied to dark matter fields that share only the same cosmology but are 512 times larger.","fun_headline_variants_meta":{"raw":{"variants":["Latent overlap diffusion produces 512 times larger 21cm maps","Power spectra of 21cm predicted to 10 percent on unseen dark matter maps","ResUNet halo prediction feeds latent overlap for 21cm maps","Large 21cm maps generated in two minutes from dark matter only"]},"model":"grok-4.3","cost_usd":0.012305,"raw_usage":{"total_tokens":5297,"prompt_tokens":696,"num_sources_used":0,"completion_tokens":76,"cost_in_usd_ticks":123053000,"prompt_tokens_details":{"text_tokens":696,"audio_tokens":0,"image_tokens":0,"cached_tokens":64},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":4525,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":696,"tokens_out":76,"duration_ms":61326,"temperature":1.0,"reasoning_tokens":4525,"cache_read_input_tokens":64,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-22T13:46:39.345000+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Apply the trained model to a new dark matter simulation of a different volume or cosmology, generate the 21 cm map, and compare its power spectrum to a full hydrodynamic run; a systematic deviation larger than 10 percent at k less than or equal to 10 h Mpc inverse would falsify the accuracy claim.","supporting_citations":[],"review_version":1}