AI & ML
A GPU Interior-Point Solver with Macro-Element Condensation for Equilibrium-Based Plasticity
arXiv cs.NA0 views
arXiv:2611.25386v1 Announce Type: new Abstract: Equilibrium-based conic formulations of elastoplasticity and limit analysis provide statically admissible stress fields but generate large Newton systems whose sparse factorization limits three-dimensional computations. This work develops and analyzes a macro-element condensation tailored to the Newton systems arising from primal--dual interior-point methods. Exploiting four-tetrahedron macro-elements, the method eliminates stress directions and macro-internal traction-continuity multipliers, yielding reduced systems that can be solved using positive-definite factorizations, with estimated reductions of approximately \(4\times\) in dimension and structural nonzeros relative to stress-only elimination. The validity of this elimination is established through global and macro-local nonsingularity conditions. Local nonsingularity is independent of the yield criterion in elastoplasticity, whereas in limit analysis it holds directly for Rankine and is ensured for von Mises through trace-bound augmentation. Numerical results show that, relative to stress-only elimination, the proposed condensation achieves CPU speedups of up to approximately \(2\times\) and reduces peak linear-solver memory by up to \(66.5\%\). We further develop a CUDA Fortran implementation that exploits uniform macro-local blocks through batched routines and uses cuDSS for the reduced sparse solve. This implementation achieves speedups of up to \(13.5\times\) over its CPU counterpart and up to \(87.5\times\) over MOSEK, while successfully solving the largest tested 3D problems that exceed the device-memory capacity of the GPU solver CuClarabel. For elastoplasticity, we also propose a preconditioned conjugate-gradient solver that reduces device-memory requirements relative to the direct GPU solver, at the expense of increased solution time.
Source:https://arxiv.org/abs/2611.25386PreviousVariational integrators for time-dependent Lagrangian systems on Lie groupsNextA Modified Quasi-Newton Method for Unconstrained Optimization with Nonpositive Curvature Detection and Structured Complexity
Related posts
More of the AI in Your Product May Run on the User's Own Device
0dev.to

So sicherst du einen MCP-Server mit codex sandbox ab
0dev.to

Perfiles de permisos de Codex: ejemplos de config.toml para ejecuciones seguras de agentes
0dev.to

GPTBot vs OAI-SearchBot: How to Configure robots.txt for AI Crawlers
0dev.to

ملفات تعريف أذونات Codex: أمثلة config.toml لتشغيل الوكلاء بأمان
0dev.to

Write a cover letter with AI and keep every fact yours: a nine-step loop
0dev.to