{"id":"b79779b3-764c-4676-b3df-c5439e800645","arxiv_id":"2607.26992","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":5,"one_line_summary":"For ~20 000 particles in 3D, D&R is ~1800× faster than dp3D with a 9% relative-density shortfall, while DEM reaches denser packings at much higher cost.","lead":"Four open-source sphere-packing tools were benchmarked for powder RVEs against an industrial MIM powder. Sequential dropping-and-rolling is ~1800× faster than DEM at ~20k particles for a ~9% density shortfall, yielding practical tool-selection rules.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"The reported 9% φ shortfall compares different physical ensembles (gravity-stable D&R vs isostatic jamming in dp3D), so it is not a pure algorithmic gap.","rationale":"The reader correctly flags φ_exp = 0.62 (Archimedes solid loading of a powder+binder feedstock) as an imperfect dry-pack target and still ACCEPTS because the idealization is disclosed and the speed/density tables are internally consistent. I agree the paper should stay ACCEPT: numbers in Tables 2–4 and Fig. 4 support the practical claim that sequential D&R is orders of magnitude cheaper for moderate φ, DEM reaches higher φ, and choice is application-driven. I only partially agree on the weakest link. φ_exp is a soft external anchor; a more load-bearing internal issue is that the 9% shortfall and the claim that all four tools meet (i)–(ii) by construction compare gravity-stable sequential packs to isostatically jammed DEM packs. That does not overturn the benchmark or the guidelines, but it means the shortfall is not a pure generator ranking. No replicate error bars, the excluded ρ=40 bimodal point, and LAMMPS tail truncation are real but secondary documentation limits already noted by the reader. Hybrid D&R\to dp3D is suggested in Sec. 4 yet not timed—the concrete test above would tighten the central claim without requiring new theory. Verdict remains ACCEPT / UNCHANGED.","tokens_in":18758,"tokens_out":933,"duration_ms":69499,"concrete_test":"On the same 133 µm IN718 PSD case, (1) run LAMMPS or dp3D in pure gravity-settle mode to kinetic-energy equilibrium with (W,μ)=(0,0) and with Lee2018 params, and (2) take the existing D&R pack as the initial state and apply dp3D isostatic compression to the same stop criterion. Report φ_B90 and wall time for gravity-DEM vs D&R and for hybrid D&R\to dp3D vs dp3D-from-scratch. If gravity-DEM φ lands within ~2–3% of D&R while jammed φ stays ~0.72, the 9% figure is mostly ensemble, not algorithm; if hybrid recovers dp3D_NANF φ at a large fraction of the 1800× saving, the paper’s own hybrid recommendation should be the headline result.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim (Table 4 / abstract: D&R ~9% φ short of dp3D at ~1800× lower cost for ~20k particles) treats φ as a method score under a shared (i)–(v) rubric. Criterion (ii) is defined as gravitational equilibrium (vertical CoM projection inside the contact convex hull). D&R and LAMMPS-gravity satisfy that by construction; dp3D is run in isostatic jamming under periodic BCs with no gravity (Sec. 2: domain contracts until ε̇v < ε̇v,stop). Frictionless jammed φ_max (dp3D_NANF = 0.723) is a different ensemble from a gravity-stable bed (D&R = 0.636). Contact-parameter sweeps already move dp3D by Δφ ~ 0.14 (Lee2018 0.584 vs NANF 0.723), comparable to the D&R–dp3D gap. Thus the headline shortfall partly reflects protocol physics (gravity settle vs isostatic jam), not only generator efficiency. Application guidelines remain useful, but ranking tools solely by φ under a unified (ii) overstates commensurability. The φ_exp = 0.62 feedstock anchor (reader’s weak point) is a related, disclosed idealization; the ensemble mismatch is more internal to the benchmark design.","agreement_with_reader":"partial"},"referee_report":{"model":"grok-4.5","summary":"The manuscript benchmarks four open-source sphere-packing generators (D&R, D&R-ME, LAMMPS in gravity mode, and dp3D in isostatic compression) against a five-criterion checklist for representative powder RVEs: hard-sphere non-overlap, gravitational equilibrium, PSD fidelity (Hellinger distance on number- and surface/volume-weighted histograms), relative density φ (via a clipped central 90% sub-box estimator), and wall-clock cost. Across 2D/3D lognormal (industrial MIM-grade IN718 PSD), 3D binary bimodal, and a 3D domain-size sweep, the authors report that DEM codes can reach the highest densities but are orders of magnitude slower; the headline quantitative claim is that for ~20 000 particles in 3D, D&R is ~9% short in φ relative to dp3D while running ~1800× faster. Application-driven selection guidelines follow from these idealised model packings, with φ also compared to an Archimedes feedstock solid loading φ_exp = 0.62.","tokens_in":19186,"tokens_out":1665,"duration_ms":58017,"significance":"A systematic, multi-configuration comparison of widely used open-source packing tools under matched domains and external metrics (prescribed industrial PSD, Hellinger fidelity, fixed-hardware timings, and a powder-specific density anchor) is genuinely useful for powder metallurgy, AM, and granular-physics communities that currently choose generators ad hoc. Strengths include: transparent reporting of both Lee2018 and NANF DEM contact settings; a bin-width-independent Hellinger metric on probability masses; the B90 clipped-volume estimator that enables fair sequential-vs-periodic comparison; empirical scaling exponents; and explicit speed/density trade-offs (Tables 2–4, Fig. 4). The work does not claim new packing physics, but the practical guidelines and the reproducible open-tool framing are a solid contribution if the ensemble and criterion-(ii) issues below are cleaned up.","major_comments":[{"comment":"Criterion (ii) is defined (§1) as gravitational equilibrium (vertical CoM projection inside the contact convex hull). Section 2 and the abstract state that all four tools meet (i)–(ii) “by construction,” yet dp3D is run in gravity-free isostatic jamming under periodic BCs (domain contracts until ė_v < ė_v,stop). That ensemble is mechanically equilibrated but not gravitationally settled. The unified (i)–(v) rubric and the headline φ ranking therefore mix distinct physical protocols. Please either (a) restate (ii) as mechanical/force balance and treat gravity-settle vs isostatic jam as separate application classes, or (b) restrict the “meets (ii) by construction” claim to D&R/D&R-ME/LAMMPS-gravity and flag dp3D as a densification/jamming reference. The 9% gap cannot be read as a pure generator-efficiency delta without this caveat.","section":"§1 criteria (i)–(ii); §2 dp3D jamming mode; Abstract"},{"comment":"The abstract and conclusion state that “the DEM codes reach the densest packings” and that D&R shows a “9% φ shortfall relative to dp3D” at ~1800× speedup. Table 4 shows this holds only for dp3D_NANF (φ = 0.723 vs D&R 0.636); dp3D_Lee2018 is the loosest packing in the table (φ = 0.584), below both D&R variants. Contact-parameter variation alone moves dp3D by Δφ ≈ 0.14, comparable to the reported D&R–dp3D gap. The main quantitative claim and the tool-selection guidelines must be stated conditionally on the DEM contact setting (and on gravity vs jamming), not as an unconditional DEM-vs-D&R density ordering.","section":"Abstract; Table 4; §4 3D lognormal; §5"},{"comment":"φ_exp = 0.62 is the MIM feedstock solid loading (powder + binder) by Archimedes (§3, Table 1). The text treats the binder as merely filling inter-particle voids and uses φ_exp as a powder-specific target for dry hard-sphere packings. Feedstock rheology and wet packing are not the same physics as the dry, cohesion/friction-modulated assemblies being scored; method ranking versus φ_exp can shift if a dry-bed or tomography density were used. This idealisation is partly disclosed but under-discussed relative to its role as the external density anchor. A short limitations paragraph (and, if available, any dry-poured or tomographic reference density) would keep the comparison honest without new campaigns.","section":"§3 Simulation parameters; Table 1; §4 3D lognormal discussion of φ_exp"}],"minor_comments":[{"comment":"LAMMPS large-particle truncation (r ≲ 11 µm vs r_max = 50 µm) is correctly identified as a pre-processing/discretisation limit, not a DEM contact failure. State earlier (methods or 2D results) that LAMMPS φ and H_V are only loosely comparable for this highly polydisperse PSD, so readers do not over-interpret Table 2/4 LAMMPS rows.","section":"§4 2D lognormal; §4 3D lognormal; Fig. 1(f), 3(f)"},{"comment":"D&R at ρ = 40 (Table 3, parenthesised φ = 0.644) is attributed to the fixed N_max = 200 stopping rule. A one-sentence sensitivity check (or a note that N_max was not re-tuned) would strengthen the bimodal trend discussion.","section":"§4 3D bimodal; Table 3"},{"comment":"Eq. (11) scaling exponents are useful; report the fitted range of N and R² (or equivalent) in the caption of Fig. 4(d) so the β values can be assessed.","section":"§4; Fig. 4(d); Eq. (11)"},{"comment":"Clarify that the 2D DEM runs are pseudo-2D (single-particle-layer thickness) when comparing native-2D D&R to LAMMPS/dp3D, so dimensionality artefacts are explicit.","section":"§4 2D lognormal"},{"comment":"Minor typography: abstract “approximatively”; consistent D&R vs DR hyphenation between abstract and body; ensure φ vs Phi is uniform in the abstract HTML/text.","section":"Abstract; passim"},{"comment":"The conclusion’s suggestion to validate against coordination number / g(r) from tomography is well taken; if any such descriptor is already computable from the generated packings, even a brief internal comparison among the four methods would add value without new experiments.","section":"§5"}],"recommendation":"major_revision","confidential_remarks":"The reader’s ACCEPT is understandable given the clean tables and practical intent; the skeptic’s ensemble point is real and should be fixed in revision, but it does not sink the paper. I do not see novelty or citation-pattern concerns. Scope fit for a materials/granular methods audience is good. Major revision is for reframing criterion (ii) and the headline φ claim, not for new large-scale campaigns."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The thing worth knowing is the concrete 3D number: for ~20k particles on the same MIM-grade PSD, D&R sits ~9% below dp3D in φ and runs ~1800× faster, with the ordering and scaling holding across a domain sweep. That is the claim people will actually use.\n\nWhat is new is not the generators—D&R, LAMMPS gravity, and dp3D jamming are prior art, including from this group—but the unified five-criterion frame, Hellinger on number and volume weights, the B90 clipped-volume density so sequential and periodic boxes can be compared, and the explicit hybrid suggestion (D&R then dp3D densify). Tables are consistent across 2D/3D lognormal, bimodal, and size sweep. Contact parameters come from Lee 2018 rather than being tuned to win; NANF bounds are shown separately. Circularity is low.\n\nSoft spots, in proportion. The stress-test lands: criterion (ii) is gravitational equilibrium, yet dp3D is isostatic compression under periodic BCs with no gravity. Frictionless jammed φ_max and a gravity-stable bed are different ensembles, and the Lee2018↔NANF swing on dp3D (~0.14) is the same order as the D&R–dp3D gap. So the headline “9% shortfall” is partly protocol physics, not pure algorithm efficiency. The paper still gives honest application guidelines; it just over-unifies φ under one rubric. Related and disclosed: φ_exp = 0.62 is feedstock solid loading (powder+binder), used as a dry-pack target—fair as an anchor, not identical physics. Minor: no replicate error bars, one bimodal point excluded for Nmax, LAMMPS large-tail truncation is an implementation bug they document. Math and citations look solid; no invented physics.\n\nThis is for people who build powder RVEs for sintering, HIP, or LPBF init and need a defensible tool choice. Not a theory paper. I would send it to referees; it is competent methods work with one framing caveat they should force into the open. Engage if you generate packings; skim the tables if you only consume them.","headline":"Useful head-to-head on packing generators with a real speed/density trade-off, slightly oversold as a single-score ranking because gravity settle and isostatic jam are different ensembles.","tokens_in":19863,"tokens_out":566,"would_cite":true,"duration_ms":17733,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"In 3D powder packings, dropping-and-rolling runs about 1800× faster than DEM at a 9% density shortfall, so the right open-source generator depends on the job.","keywords":["sphere packing","representative volume element","discrete element method","dropping and rolling","additive manufacturing","particle size distribution","relative density"],"falsifier":"Measure the dry packing fraction of the same MIM-grade powder, or extract density and contact statistics from high-resolution tomography of the actual bed, and check whether the tool ranking versus that measured density still matches the ranking versus 0.62.","tokens_in":19580,"feed_emoji":"⚙️","tokens_out":988,"duration_ms":41583,"temperature":0.7,"pith_summary":"Dense sphere packings are the usual starting geometry for simulations in powder metallurgy, additive manufacturing, and granular physics, yet the choice of generator is seldom backed by a common benchmark. This paper compares four open-source tools that already enforce non-overlapping particles in gravitational equilibrium: sequential dropping-and-rolling (D&R), its densified moving-enlarging variant, and two discrete-element codes (LAMMPS under gravity and dp3D under isostatic compression). Across 2D and 3D lognormal packs based on an industrial MIM-grade powder, a controlled 3D bimodal study, and a domain-size sweep, they score size-distribution fidelity with the bin-width-independent Hellinger distance and packing fraction against the feedstock solid loading of 0.62. DEM reaches the densest assemblies but costs more than three orders of magnitude in runtime; for roughly 20,000 particles in 3D, D&R is about 9% less dense than the densest DEM result while finishing about 1800 times faster. The outcome is application-driven selection guidance rather than a single preferred tool.","feed_headline":"Powder packings: 1800× faster at a 9% density cost","feed_subtitle":"Open-source tools rank by speed versus density, guiding which generator fits metallurgy and AM work.","key_machinery":"Five simultaneous criteria for a representative packing—no overlaps, gravitational equilibrium, PSD fidelity via Hellinger distance on number and surface/volume weights, relative density on a clipped central 90% sub-box, and compute cost—applied uniformly to D&R, D&R-ME, LAMMPS, and dp3D.","core_discovery":"Among open-source generators that satisfy hard-sphere and gravitational-equilibrium constraints by construction, DEM methods produce the densest packings while preserving the target size distribution, but sequential dropping-and-rolling recovers that distribution at a fraction of the cost—with a roughly 9% relative-density shortfall versus the densest DEM case at about 20,000 particles in 3D and an 1800× speedup—yielding concrete rules for choosing a tool by whether speed, a prescribed dense packing fraction, or strict size-distribution fidelity dominates the application.","pith_inferences":["Contact-network metrics from tomography (coordination, radial distribution) would likely reorder the tools even when bulk density matches, because sequential gravitational placement and isostatic jamming build different neighbourhoods.","The large-particle truncation seen in LAMMPS is a preprocessing discretisation limit, so any polydisperse DEM workflow that bins the size law upstream needs an independent tail check before trusting volume-weighted statistics.","For powder-bed fusion layers, process-specific spreading may dominate these idealised gravitational packs, so D&R’s speed is most useful as a cheap initialisation rather than a final bed surrogate."],"forward_implications":["Choose D&R for fast prototyping and large representative volumes when moderate density is enough.","Choose dp3D or LAMMPS when both a prescribed dense packing fraction and a faithful size distribution are required.","Choose D&R-ME when a small density gain over plain D&R is worth a slight size-distribution shift.","Seed dp3D compression from a finished D&R packing to reach high density cheaper than running DEM from scratch.","Score generators with Hellinger distance and clipped-sub-box density when boundary treatments differ."],"fun_headline_variants":["DR packings: 1800× faster, 9% density shortfall vs DEM","Open-source packers ranked: DEM densest, DR 1800× quicker","Which packing tool? Speed vs density trade-off at 20k particles","DEM hits densest Φ; DR recovers PSD 1800× faster at 9% cost","Tool guide: DR for speed, DEM for dense 3D powder packings"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"That the MIM feedstock solid loading of 0.62 is a fair powder-specific density target for dry hard-sphere gravitational packings, treating the binder as merely filling voids.","fun_headline_variants_meta":{"raw":{"variants":["DR packings: 1800× faster, 9% density shortfall vs DEM","Open-source packers ranked: DEM densest, DR 1800× quicker","Which packing tool? Speed vs density trade-off at 20k particles","DEM hits densest Φ; DR recovers PSD 1800× faster at 9% cost","Tool guide: DR for speed, DEM for dense 3D powder packings"]},"model":"grok-4.5","effort":"low","cost_usd":0.003784,"raw_usage":{"total_tokens":1243,"prompt_tokens":864,"num_sources_used":0,"completion_tokens":94,"cost_in_usd_ticks":37844000,"prompt_tokens_details":{"text_tokens":864,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":285,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":864,"tokens_out":94,"duration_ms":5519,"temperature":1.0,"reasoning_tokens":285,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-30T14:41:36.865937+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Measure the dry packing fraction of the same MIM-grade powder, or extract density and contact statistics from high-resolution tomography of the actual bed, and check whether the tool ranking versus that measured density still matches the ranking versus 0.62.","supporting_citations":[],"review_version":1}