4 Commits

Author	SHA1	Message	Date
wassname	48d4c1044a	refactor(pinn): extract heat exchanger specifics to refs/ ... Moved heat-exchanger-specific content from pinn/SKILL.md to pinn/refs/heat_exchanger.md: complexity ladder table, known failure modes (U->0, counterflow signs), property mappings (REFPROP/PCHIP), multi-episode training. PINN skill is now domain-agnostic. pinn/SKILL.md reduced from 4961w to 4274w (~14%).	2026-03-06 13:39:53 +08:00
wassname	463c8fdbbc	fix: apply Gemini review fixes (device kwarg, gradcheck requires_grad, torch prefix) ... Review: Gemini 3.1 Pro approved. 3 fixes applied: - pinn/SKILL.md: PchipFunction torch.tensor missing device=h.device (GPU crash) - SKILL.md: gradcheck needs .requires_grad_(True) on doubled inputs - SKILL.md: loss surface pseudocode now has torch. prefix + indexing='ij'	2026-03-06 12:15:37 +08:00
wassname	2db012dd2c	docs(pinn): add Wang 2021 and Rathore 2024 evidence files	2026-03-06 12:12:51 +08:00
wassname	a90624b36d	feat(pinn): add pinn/ sub-skill with SKILL.md and evidence ... SKILL.md: 478-line PINN training best practices (complexity ladder, nondim, architecture, optimization, loss design, sampling, property mappings, ConFIG, domain decomposition). docs/evidence/: 6 files -- krishnapriyan2021, sukumar2022, wang2022 causal, wang2022+2023 expert guides, Brunton youtube transcripts. Missing evidence (to fetch): Wang 2001.04536 (gradient pathologies), Rathore 2402.01868 (ICML loss landscape). Author: wassname (https://github.com/wassname)	2026-03-06 11:48:41 +08:00