Generative Design of inorganic compounds using deep diffusion language models

Rongzhi Dong,Nihang Fu,dirisuriya M. D. Siriwardane,Jianjun Hu

2023-10-01

Abstract:Due to the vast chemical space, discovering materials with a specific function is challenging. Chemical formulas are obligated to conform to a set of exacting criteria such as charge neutrality, balanced electronegativity, synthesizability, and mechanical stability. In response to this formidable task, we introduce a deep learning-based generative model for material composition and structure design by learning and exploiting explicit and implicit chemical knowledge. Our pipeline first uses deep diffusion language models as the generator of compositions and then applies a template-based crystal structure prediction algorithm to predict their corresponding structures, which is then followed by structure relaxation using a universal graph neural network-based potential. The density functional theory (DFT) calculations of the formation energies and energy-above-the-hull analysis are used to validate new structures generated through our pipeline. Based on the DFT calculation results, six new materials, including Ti2HfO5, TaNbP, YMoN2, TaReO4, HfTiO2, and HfMnO2, with formation energy less than zero have been found. Remarkably, among these, four materials, namely Ti2$HfO5, TaNbP, YMoN2, and TaReO4, exhibit an e-above-hull energy of less than 0.3 eV. These findings have proved the effectiveness of our approach.

Materials Science,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to discover new materials with specific functions in the vast chemical space. The composition and structure of new materials must meet many strict conditions, such as charge neutrality, electronegativity balance, synthesizability and mechanical stability. Traditionally, the discovery of new substances depends on the experience of experts and fine - tuning of existing materials. However, this method has limitations in generating new chemical formula prototypes and can only generate novel combinations through element substitution. To overcome these challenges, the paper proposes a generative model based on a deep diffusion language model for material composition and structure design. This method first uses a deep diffusion language model as a composition generator, then applies a template - based crystal structure prediction algorithm to predict the corresponding structure, and finally further optimizes the structure through structure relaxation based on the potential energy of a general graph neural network. The validity of the newly generated structures is verified by density functional theory (DFT) calculations of formation energy and energy - above - convex - hull analysis. The paper shows that this method can effectively generate six new materials, four of which exhibit energy - above - convex - hull values below 0.3 eV, demonstrating the effectiveness of this method.

Generative Design of inorganic compounds using deep diffusion language models

Discovery of 2D materials using Transformer Network based Generative Design

MatterGen: a generative model for inorganic materials design

Crystal Composition Transformer: Self-Learning Neural Language Model for Generative and Tinkering Design of Materials

Computational Discovery of New 2D Materials Using Deep Learning Generative Models

Active learning based generative design for the discovery of wide bandgap materials

Physics guided deep learning for generative design of crystal materials with symmetry constraints

Generative Design of Crystal Structures by Point Cloud Representations and Diffusion Model

Structural Constraint Integration in Generative Model for Discovery of Quantum Material Candidates

Materials Transformers Language Models for Generative Materials Design: a benchmark study

Data-Driven Deep Generative Design of Stable Spintronic Materials

Generative adversarial networks (GAN) based efficient sampling of chemical composition space for inverse design of inorganic materials

Crystal Transformer: Self-learning neural language model for Generative and Tinkering Design of Materials

WyCryst: Wyckoff Inorganic Crystal Generator Framework

Deep Generative Models for Materials Discovery and Machine Learning-Accelerated Innovation

Generative Inverse Design of Crystal Structures via Diffusion Models with Transformers

High‐Throughput Discovery of Novel Cubic Crystal Materials Using Deep Generative Neural Networks

Generative Hierarchical Materials Search

Generative Design of Functional Metal Complexes Utilizing the Internal Knowledge of Large Language Models

Crystal Structure Generation Based On Material Properties