Rethinking Gradient-Based Methods: Multi-Property Materials Design Beyond Differentiable Targets
Akihiro Fujii, Yoshitaka Ushiku, Koji Shimizu, Anh Khoa Augustin Lu, Satoshi Watanabe
TL;DR
SMOACS introduces a gradient-based framework for multi-property crystal design that directly optimizes structure-space candidates using pretrained property predictors while enforcing non-differentiable constraints through oxidation-number masks and template-based initialization. It enables simultaneous optimization over multiple properties and structural constraints, preserving crystal-family identity and scaling to large systems such as 135-atom perovskites, and it integrates predictors from different datasets without retraining. Across diverse targets, SMOACS outperforms generative-model and Bayesian-optimization baselines and is validated with DFT, demonstrating practical viability for constrained inverse design in materials science. The approach offers a flexible, model-agnostic pathway to discovering high-performance materials under complex constraints, with potential gains from active learning and larger, more diverse datasets to improve predictor generalization.
Abstract
Gradient-based methods offer a simple, efficient strategy for materials design by directly optimizing candidates using gradients from pretrained property predictors. However, their use in crystal structure optimization is hindered by two key challenges: handling non-differentiable constraints, such as charge neutrality and structural fidelity, and susceptibility to poor local minima. We revisit and extend the gradient-based methods to address these issues. We propose Simultaneous Multi-property Optimization using Adaptive Crystal Synthesizer (SMOACS), which integrates oxidation-number masks and template-based initialization to enforce non-differentiable constraints, avoid poor local minima, and flexibly incorporate additional constraints without retraining. SMOACS enables multi-property optimization. including exceptional targets such as high-temperature superconductivity, and scales to large crystal systems, both persistent challenges for generative models, even those enhanced with gradient-based guidance from property predictors. In experiments on five target properties and three datasets, SMOACS outperforms generative models and Bayesian optimization methods, successfully designing 135-atom perovskite structures that satisfy multiple property targets and constraints, a task at which the other methods fail entirely.