Crystal Generation using the Fully Differentiable Pipeline and Latent Space Optimization

TL;DR

This paper introduces a fully differentiable pipeline combining a symmetry-conditioned variational autoencoder with a differentiable objective for efficient crystal structure generation, enabling targeted material design.

Contribution

It presents a novel dual-level optimization strategy and GPU-accelerated implementation for generating complex crystal structures with specified local environments.

Findings

Fivefold speed improvement over previous CPU implementation

Effective escape from local minima in structure optimization

Scalable approach for multi-component and multi-environment systems

Abstract

We present a materials generation framework that couples a symmetry-conditioned variational autoencoder (CVAE) with a differentiable SO(3) power spectrum objective to steer candidates toward a specified local environment under the crystallographic constraints. In particular, we implement a fully differentiable pipeline to enable batch-wise optimization on both direct and latent crystallographic representations. Using the GPU acceleration, this implementation achieves about fivefold speed compared to our previous CPU workflow, while yielding comparable outcomes. In addition, we introduce the optimization strategy that alternatively performs optimization on the direct and latent crystal representations. This dual-level relaxation approach can effectively escape local minima defined by different objective gradients, thus increasing the success rate of generating complex structures satisfying the target local environments. This framework can be extended to systems consisting of multi-components and multi-environments, providing a scalable route to generate material structures with the target local environment.