{"id":"d0fb367d-fd5b-4baf-b91e-9dfdc9346006","arxiv_id":"2605.31425","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"A mixed-precision ripALM method on GPUs solves large-scale minimum enclosing ball problems faster than CPU geometric software and general conic solvers while maintaining high accuracy.","lead":"The paper develops a mixed-precision GPU framework using a relative inexact proximal augmented Lagrangian method to compute the minimum enclosing ball for large collections of balls. Smart generalists might read it to understand practical ways to accelerate geometric optimization in data-heavy applications like clustering or robotics.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's identification of the screening-plus-check step as the weakest assumption aligns directly with the only part of the argument that is not purely algebraic or GPU-parallelism-based. Because the full manuscript was stated to be inaccessible to the reader and no contradictory detail appears in the abstract, the UNVERDICTED verdict with low confidence is appropriate; no new load-bearing flaw is visible.","tokens_in":1707,"tokens_out":317,"duration_ms":14353,"concrete_test":"Re-run the largest reported instances (e.g., those with >10^6 balls) while logging the exact number of reintroduced balls and total outer iterations of the screening-refinement loop; if reintroductions remain below 1% of the screened set and total runtime stays within the claimed speedup factor, the mixed-precision correctness claim holds for the tested regime.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on the mixed-precision ripALM strategy (low-precision screening of boundary balls, high-precision refinement on the reduced set, and a posteriori feasibility check on the original problem). The abstract states that the check detects and reintroduces any violated discarded balls, and experiments report high accuracy plus speedups. No internal inconsistency, hidden assumption in the SOCP reformulation, or failure mode in the separable GPU updates is apparent from the given description. The reader's weakest assumption correctly flags the empirical reliability of screening, but the provided text supplies no counter-evidence that would falsify the claim.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript develops a mixed-precision GPU-oriented optimization framework for the minimum enclosing ball problem over a collection of balls. It reformulates the problem as an equivalent second-order cone program and solves it via a relative inexact proximal augmented Lagrangian method (ripALM) whose inner subproblems are solved only to a progress-dependent relative accuracy. The proximal augmented Lagrangian admits a constraint-wise separable structure that permits efficient parallel GPU evaluation of the objective, gradient, generalized Hessian, and multiplier updates. A mixed-precision reduction strategy is introduced in which a low-precision ripALM run screens for balls near the approximate boundary, a high-precision ripALM run refines the reduced problem, and an a-posteriori feasibility check on the original instance detects and reintroduces any violated discarded balls. Numerical experiments are reported to show that both ripALM and its mixed-precision variant attain high accuracy while delivering substantial speedups relative to CPU-based geometric software and general-purpose conic solvers on large-scale instances.","tokens_in":1814,"tokens_out":565,"duration_ms":23944,"significance":"If the reported performance and accuracy claims are substantiated by detailed, reproducible experiments, the work would constitute a meaningful practical advance for large-scale minimum-enclosing-ball computations. The combination of an SOCP reformulation, relative inexact proximal ALM, GPU-friendly separability, and a controlled mixed-precision screening strategy offers a concrete route to scaling beyond the reach of existing CPU geometric and conic solvers. The absence of free parameters or self-referential definitions in the algorithmic construction is a positive feature.","major_comments":[{"comment":"Abstract and Numerical Experiments section: the central claim that ripALM and mixed-precision ripALM achieve high accuracy and are substantially faster than the tested baselines rests on numerical experiments, yet the manuscript supplies no information on the dimensions and cardinalities of the test instances, the precise error metrics employed, the implementation details or versions of the CPU-based geometric software and conic solvers used as baselines, or any statistical controls (multiple runs, variance, or timing methodology). This omission leaves the performance assertions with limited verifiable support.","section":"Abstract and Numerical Experiments"},{"comment":"Mixed-precision reduction strategy (described in the abstract and the algorithmic development): the reliability of the low-precision screening step in correctly identifying balls near the approximate boundary, such that the subsequent high-precision solve plus a-posteriori feasibility check recovers a valid solution without excessive reintroductions, is load-bearing for the claimed efficiency gains. The manuscript provides no quantitative characterization of how often or under what conditions the check triggers reintroductions, nor any worst-case or probabilistic guarantee on the screening quality.","section":"Mixed-precision reduction strategy"}],"minor_comments":[],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments and positive overall assessment. We address the two major comments below and will revise the manuscript accordingly to improve clarity and verifiability.","responses":[{"response":"We agree that the manuscript currently lacks these experimental details, which limits the ability to fully verify the claims. In the revised version we will expand the Numerical Experiments section (and update the abstract if needed) to explicitly report: the dimensions and cardinalities of all test instances; the precise error metrics (e.g., relative duality gap, feasibility violation, and accuracy tolerances); the exact versions, configurations, and implementation details of all baseline solvers; and the statistical protocol including number of runs, variance reporting, and timing methodology.","revision_made":"yes","referee_comment":"[Abstract and Numerical Experiments] Abstract and Numerical Experiments section: the central claim that ripALM and mixed-precision ripALM achieve high accuracy and are substantially faster than the tested baselines rests on numerical experiments, yet the manuscript supplies no information on the dimensions and cardinalities of the test instances, the precise error metrics employed, the implementation details or versions of the CPU-based geometric software and conic solvers used as baselines, or any statistical controls (multiple runs, variance, or timing methodology). This omission leaves the performance assertions with limited verifiable support."},{"response":"The a-posteriori feasibility check on the original instance is intended to guarantee that any discarded balls violating the approximate solution are reintroduced, thereby ensuring correctness independent of screening quality. We acknowledge that the current manuscript does not provide quantitative data on reintroduction frequency or conditions. In the revision we will add empirical statistics from the reported experiments (e.g., average and maximum number of reintroductions per instance class) and discuss observed conditions under which the screening performs well. We do not provide worst-case or probabilistic guarantees, as the screening step is a practical heuristic; the check serves as the correctness safeguard.","revision_made":"yes","referee_comment":"[Mixed-precision reduction strategy] Mixed-precision reduction strategy (described in the abstract and the algorithmic development): the reliability of the low-precision screening step in correctly identifying balls near the approximate boundary, such that the subsequent high-precision solve plus a-posteriori feasibility check recovers a valid solution without excessive reintroductions, is load-bearing for the claimed efficiency gains. The manuscript provides no quantitative characterization of how often or under what conditions the check triggers reintroductions, nor any worst-case or probabilistic guarantee on the screening quality."}],"tokens_in":1532,"tokens_out":533,"duration_ms":22747,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is a GPU-oriented mixed-precision ripALM solver for the minimum enclosing ball problem via its SOCP reformulation. The approach uses low-precision runs to screen balls near the boundary, refines the reduced set in high precision, and applies an a posteriori check on the full problem to recover any discarded boundary balls.\n\nWhat stands out as new is the concrete combination of ripALM's separable proximal updates with this screening-plus-verification layer and the explicit mapping of objective, gradient, Hessian, and multiplier steps to parallel GPU operations followed by reductions. The design keeps low precision only for the cheap screening and warm-start phase while enforcing final feasibility on the original instance.\n\nThe separable structure is a genuine fit for GPU parallelism, and restricting high-precision work to the reduced set is a reasonable efficiency move for large collections of balls. The a posteriori check is a straightforward safeguard.\n\nThe soft spot is the experimental support. The abstract states that the method achieves high accuracy and substantial speedups versus CPU geometric codes and general conic solvers on large instances, yet supplies no concrete numbers on instance sizes, error measures, reintroduction rates from the check, or direct GPU baselines. Without those details it is difficult to assess how often the screening step works cleanly or how consistent the gains are.\n\nThis is aimed at researchers who need fast solvers for geometric problems that appear in clustering, packing, or robotics and who are comfortable with GPU convex optimization. It is worth sending to a serious referee because the algorithmic pieces are standard and the performance claims are testable; the main request would be fuller experimental reporting and some quantification of screening reliability.","headline":"Mixed-precision ripALM on GPU for MEB gives a sensible practical speed-up route but rests on thin experimental detail so far.","tokens_in":2297,"tokens_out":397,"would_cite":false,"duration_ms":21633,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A mixed-precision GPU framework solves large minimum enclosing ball problems faster than CPU solvers while keeping high accuracy.","keywords":["minimum enclosing ball","mixed precision","GPU acceleration","second-order cone programming","proximal augmented Lagrangian","large-scale optimization","inexact methods"],"falsifier":"On a large test instance the mixed-precision run produces a final ball that leaves many originally discarded balls outside after the feasibility check, or the reported accuracy drops below the claimed level.","tokens_in":2595,"feed_emoji":"","tokens_out":657,"duration_ms":20077,"temperature":0.7,"pith_summary":"The paper develops a GPU-oriented method for the minimum enclosing ball of a collection of balls by first rewriting the task as an equivalent second-order cone program. It then applies a relative inexact proximal augmented Lagrangian method whose updates separate across balls, allowing direct parallel evaluation and reduction on the GPU. A mixed-precision tactic runs a cheap low-precision pass to screen for balls near the boundary, solves the smaller problem at high precision, and finally checks feasibility against the full original set to reintroduce any missed balls. Numerical tests indicate the resulting ripALM and mixed-precision versions reach the target accuracy yet run substantially quicker than existing CPU geometric codes and general conic solvers on large instances. If the screening step works as intended, the approach lets users tackle bigger enclosing-ball problems without needing full high-precision work on every ball.","feed_headline":"Mixed-precision GPU speeds large minimum enclosing ball solves","feed_subtitle":"Low-precision screening shrinks the problem before high-precision refinement and feasibility check, outperforming CPU solvers on big instanc","key_machinery":"A relative-type inexact proximal augmented Lagrangian method (ripALM) equipped with a mixed-precision reduction that screens with low precision, refines the reduced problem at high precision, and applies an a-posteriori feasibility check on the original data.","core_discovery":"The central claim is that ripALM and mixed-precision ripALM achieve high accuracy and are substantially faster than the tested CPU-based geometric software and general-purpose conic solvers on large-scale instances.","pith_inferences":["The same screening-plus-refinement pattern could be tried on other large convex problems whose constraints separate across data points.","Memory savings from discarding most balls early may become useful when the input set exceeds GPU memory limits.","One could test whether the number of reintroduced balls stays small across different distributions of input ball radii and centers."],"forward_implications":["The separable structure of the proximal augmented Lagrangian permits efficient parallel GPU maps followed by reductions.","Low precision is used only for screening and warm-starting while final feasibility is enforced on the original problem.","The method reaches high accuracy on large instances.","It runs substantially faster than CPU geometric software and general conic solvers."],"fun_headline_variants":["Mixed-precision GPU ripALM speeds large enclosing ball solves","ripALM mixed-precision on GPU beats CPU on big ball problems","GPU mixed ripALM accelerates large minimum enclosing ball tasks","Mixed-precision ripALM outperforms CPU for large ball enclosures on GPU"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The low-precision screening step can reliably identify balls near the approximate boundary so that the subsequent high-precision solve on the reduced set plus a posteriori feasibility check recovers a valid solution to the original problem without excessive reintroductions.","fun_headline_variants_meta":{"raw":{"variants":["Mixed-precision GPU ripALM speeds large enclosing ball solves","ripALM mixed-precision on GPU beats CPU on big ball problems","GPU mixed ripALM accelerates large minimum enclosing ball tasks","Mixed-precision ripALM outperforms CPU for large ball enclosures on GPU"]},"model":"grok-4.3","cost_usd":0.005044,"raw_usage":{"total_tokens":2429,"prompt_tokens":608,"num_sources_used":0,"completion_tokens":65,"cost_in_usd_ticks":50437000,"prompt_tokens_details":{"text_tokens":608,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1756,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":608,"tokens_out":65,"duration_ms":12676,"temperature":1.0,"reasoning_tokens":1756,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-28T21:36:04.337720+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"On a large test instance the mixed-precision run produces a final ball that leaves many originally discarded balls outside after the feasibility check, or the reported accuracy drops below the claimed level.","supporting_citations":[],"review_version":1}