PRISMat: Policy-Driven, Permutation-Invariant Autoregressive Material Generation

TL;DR

PRISMat is a permutation-invariant, efficient autoregressive model that outperforms large language models in predicting material properties, enabling faster and more accurate high-throughput material discovery.

Contribution

We introduce PRISMat, a novel cost-effective, permutation-invariant model that surpasses LLMs in material generation tasks with reduced inference time.

Findings

PRISMat achieves lower mean absolute errors in property prediction tasks.

PRISMat outperforms LLMs despite faster inference.

The model reduces errors by up to 4 times compared to previous models.

Abstract

Rapid identification of candidate materials with target properties has become a key task in materials science. Machine learning has emerged as an alternative to physics-based simulation, offering a faster and cheaper way to filter materials based on their stability and other target properties, reducing the number of candidates that reach the costly synthesis stage. Recently, Large Language Models (LLMs) have been applied to this role, but these models are parameter-heavy and computationally expensive both during training and at inference time, making them unsuitable for high-throughput tasks. This inefficiency stems from both the large over-parameterization of language models and the difficulty of framing material generation as a sequence learning problem. In this paper, we present PRISMat, a cost-effective, permutation-invariant model, which addresses these limitations. We show that PRISMat, despite taking less time for inference, is able to outperform LLMs in generating crystal slabs conditioned on critical materials' surface properties. In targeted material discovery, we achieve mean absolute errors of 0.188 eV/A$^2$ and 2.79 eV for cleavage energy and work function tasks, respectively, reducing the error of the next best model by 4$\times$.