Shape Derivative-Informed Neural Operators with Application to Risk-Averse Shape Optimization

Xindi Gong, Dingcheng Luo, Thomas O’Leary-Roseberry, Ruanui Nicholson, Omar Ghattas · 2026

Read Shape Derivative-Informed Neural Operators with Application to Risk-Averse Shape Optimization by Xindi Gong, Dingcheng Luo, Thomas O’Leary-Roseberry, Ruanui Nicholson, Omar Ghattas (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-03211.txt.

Description

Xindi Gong and four others speed up shape design when the best shape must also be safe under uncertainty. Engineers optimizing a shape governed by partial differential equations must resample many uncertain cases for every candidate shape, which standard methods do very slowly. Neural surrogates are fast but often give unreliable sensitivities, the gradients an optimizer follows. Their 2026 approach, Shape-DINO, trains on derivatives deliberately.

Geometries are encoded by smooth mappings onto one fixed reference domain, and training teaches both the physical solution and its derivatives with respect to design and uncertainty. The authors prove error bounds tying surrogate accuracy to final optimization error. On three test problems, including flows around bodies in two and three dimensions, the method gave more trustworthy optima than plain surrogates. It ran orders of magnitude faster per evaluation, and needed far fewer costly simulations.

Catalog record

CREATORXindi Gong, Dingcheng Luo, Thomas O’Leary-Roseberry, Ruanui Nicholson, Omar Ghattas
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE245147
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.03211

Browse the full library · About the Tower of Babel