[Paper Review] MatterGPT: A Generative Transformer for Multi-Property Inverse Design of Solid-State Materials

Yan Chen, Xueru Wang+6 more|arXiv (Cornell University)|Aug 14, 2024

Manufacturing Process and OptimizationEngineering13 citations

TL;DR

MatterGPT trains a generative Transformer on next-token prediction to design de novo solid-state materials with targeted properties, including single and multi-property objectives, using SLICES crystal encoding.

ABSTRACT

Inverse design of solid-state materials with desired properties represents a formidable challenge in materials science. Although recent generative models have demonstrated potential, their adoption has been hindered by limitations such as inefficiency, architectural constraints and restricted open-source availability. The representation of crystal structures using the SLICES (Simplified Line-Input Crystal-Encoding System) notation as a string of characters enables the use of state-of-the-art natural language processing models, such as Transformers, for crystal design. Drawing inspiration from the success of GPT models in generating coherent text, we trained a generative Transformer on the next-token prediction task to generate solid-state materials with targeted properties. We demonstrate MatterGPT's capability to generate de novo crystal structures with targeted single properties, including both lattice-insensitive (formation energy) and lattice-sensitive (band gap) properties. Furthermore, we extend MatterGPT to simultaneously target multiple properties, addressing the complex challenge of multi-objective inverse design of crystals. Our approach showcases high validity, uniqueness, and novelty in generated structures, as well as the ability to generate materials with properties beyond the training data distribution. This work represents a significant step forward in computational materials discovery, offering a powerful and open tool for designing materials with tailored properties for various applications in energy, electronics, and beyond.

Motivation & Objective

• Address the inverse design challenge in solid-state materials.
• Leverage the SLICES notation to render crystal structures as text for NLP-style modeling.
• Train a generative Transformer on next-token prediction to generate structures with targeted properties.
• Demonstrate single-property and multi-property inverse design capabilities.
• Provide an open, scalable tool for computational materials discovery.

Proposed method

• Represent crystal structures as SLICES strings to enable NLP-modeling of crystals.
• Train a generative Transformer on the next-token-prediction task to generate crystal strings.
• Evaluate generated structures for validity, uniqueness, and novelty.
• Demonstrate targeting of lattice-insensitive (formation energy) and lattice-sensitive (band gap) properties.
• Extend the model to multi-property objective design for crystals.

Experimental results

Research questions

• RQ1 Can MatterGPT generate valid, unique, and novel crystal structures that meet targeted properties?
• RQ2 Can the model target single properties such as formation energy and band gap?
• RQ3 Can MatterGPT achieve multi-property inverse design for crystals?
• RQ4 How well does the approach generalize beyond the training data distribution?

Key findings

• The approach achieves high validity, uniqueness, and novelty in generated structures.
• MatterGPT can generate de novo crystal structures with targeted properties.
• The model can target both lattice-insensitive and lattice-sensitive properties (e.g., formation energy and band gap).
• The method extends to multi-property design, addressing multi-objective inverse design of crystals.
• The work indicates potential for generating materials with properties beyond the training data distribution.