Co-PiLOT: Constrained Physics-Informed Latent Optimization for Target-Driven Inverse Design

2026-09-29

AI Summary

This study addresses the challenges of high-dimensional non-Euclidean search, complex feasibility encoding, and prohibitive simulation costs in the inverse design of physical systems by proposing MERIDIAN, an active optimization framework. This method constructs a generative latent space using a Vision Transformer encoder coupled with diffusion or rectified flow decoders, and integrates uncertainty quantification, failure-aware prediction, and manifold trust-region mechanisms to enable physics-informed black-box optimization. Evaluated on magnesium alloy microstructure design, MERIDIAN reduces objective error by 3%–22% compared to baselines within a limited budget of only 160 simulations. These results demonstrate that the proposed framework significantly enhances inverse design efficiency in data-scarce scenarios, offering a scalable solution for computationally expensive physical system optimization.

Abstract

Inverse design of physical systems (molecules, devices, microstructures) often reduces to optimizing a high-dimensional structure against an expensive black-box simulator. Direct search is difficult because the space is non-Euclidean, feasibility is hard to encode, and each evaluation is expensive. We present Co-PiLOT, a latent optimization approach that maps candidates through a generative encoder-decoder, uses the decoder as a learned validity prior, and searches the latent space with physics-informed black-box optimization. The framework is applied on the inverse design of magnesium alloy microstructure/texture. We develop a vision transformer based-encoder; paired with latent diffusion, diffusion transformer and rectified-flow transformer-based decoders on $\sim80{,}000$ EBSD-derived microstructure dataset to learn a minimal bottleneck, $z$. The ViT-FMDiT model ($z$=$768$) reconstructs high-fidelity microstructure images (FID $27.86$, MS-SSIM $0.178$), which our self-segmenting orientation codec converts into input grids for crystal plasticity solver. Finally, we introduce MERIDIAN, an active latent optimizer driven by deep-kernel Gaussian-process uncertainty, failure-aware feasibility prediction, manifold-aware trust regions, and target-aware acquisition. Within a budget of $160$ simulations, the ViT-FMDiT and MERIDIAN combination yields the best target-driven objective score, reducing the relative target error by $3$--$22\%$ against seven baselines (DANTE, TuRBO, BAxUS, CMA-ES, DDOM, SEIKO, DDPO) on the same decoder.

Problem

• Inverse design
• Black-box optimization
• High-dimensional space
• Non-Euclidean space
• Expensive evaluation

Innovation

• Inverse Design
• Latent Optimization
• Physics-Informed
• Vision Transformer
• Gaussian Process