ARC-STAR: Auditable Post-Hoc Correction for PDE Foundation Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Chengze, Wei, Lingwei, Sun, Li, Lv, Hongbo, Yang, Jie, Zhang, Hanrong, Zheng, Kening, Huang, Wei-Chieh, Ma, Enze, Yu, Philip S.
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918520770527232
author Li, Chengze
Wei, Lingwei
Sun, Li
Lv, Hongbo
Yang, Jie
Zhang, Hanrong
Zheng, Kening
Huang, Wei-Chieh
Ma, Enze
Yu, Philip S.
author_facet Li, Chengze
Wei, Lingwei
Sun, Li
Lv, Hongbo
Yang, Jie
Zhang, Hanrong
Zheng, Kening
Huang, Wei-Chieh
Ma, Enze
Yu, Philip S.
contents Partial differential equation (PDE) foundation models are pretrained networks that forecast how physical fields like velocity and pressure evolve from a single reusable solver. On unfamiliar flows their predictions drift step by step, errors concentrate in a few regions, yet retraining destabilizes the network and uniform post-hoc correction overlooks this spatial concentration. To address this, we propose a frozen-solver post-hoc correction framework, Adaptive Risk-Calibrated Spatial Triage for Auditable Refinement (ARC-STAR). ARC-STAR organizes correction into three stages: a global corrector removes broad solver bias, a blockwise local refiner cleans the post-global residual, and, at deployment, a label-free score routes refinement to high-risk blocks under a compute budget. The framework is designed to be (i) frozen-host, preserving the pretrained solver without fine-tuning; (ii) auditable, with global and local stages trained and evaluated separately for measurable contributions; and (iii) budget-aware, using a blockwise interface that either refines the full field or routes limited compute to high-risk regions. Across five flow benchmarks spanning ten regime cells, ARC-STAR is the only method that cuts velocity rollout error by at least 36x over raw Poseidon on every cell. The global stage reduces raw host error by 91-99%, and the local stage further reduces the remaining post-global residual by up to 94.4%.
format Preprint
id arxiv_https___arxiv_org_abs_2605_22222
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ARC-STAR: Auditable Post-Hoc Correction for PDE Foundation Models
Li, Chengze
Wei, Lingwei
Sun, Li
Lv, Hongbo
Yang, Jie
Zhang, Hanrong
Zheng, Kening
Huang, Wei-Chieh
Ma, Enze
Yu, Philip S.
Machine Learning
Partial differential equation (PDE) foundation models are pretrained networks that forecast how physical fields like velocity and pressure evolve from a single reusable solver. On unfamiliar flows their predictions drift step by step, errors concentrate in a few regions, yet retraining destabilizes the network and uniform post-hoc correction overlooks this spatial concentration. To address this, we propose a frozen-solver post-hoc correction framework, Adaptive Risk-Calibrated Spatial Triage for Auditable Refinement (ARC-STAR). ARC-STAR organizes correction into three stages: a global corrector removes broad solver bias, a blockwise local refiner cleans the post-global residual, and, at deployment, a label-free score routes refinement to high-risk blocks under a compute budget. The framework is designed to be (i) frozen-host, preserving the pretrained solver without fine-tuning; (ii) auditable, with global and local stages trained and evaluated separately for measurable contributions; and (iii) budget-aware, using a blockwise interface that either refines the full field or routes limited compute to high-risk regions. Across five flow benchmarks spanning ten regime cells, ARC-STAR is the only method that cuts velocity rollout error by at least 36x over raw Poseidon on every cell. The global stage reduces raw host error by 91-99%, and the local stage further reduces the remaining post-global residual by up to 94.4%.
title ARC-STAR: Auditable Post-Hoc Correction for PDE Foundation Models
topic Machine Learning
url https://arxiv.org/abs/2605.22222