Generative models for crystalline materials

2025-11-27

AI Summary

Crystal material inverse design faces critical challenges in accurately modeling structure–property relationships and ensuring synthetic feasibility of generated structures. Method: We propose an end-to-end crystal generation framework that integrates graph neural networks with advanced deep generative models (VAEs, GANs, and diffusion models), uniquely incorporating disorder/defect modeling, synthesis-aware constraints, and experimental verifiability assessment. Contribution/Results: We establish a multidimensional evaluation benchmark covering representation learning, generation quality, physical consistency, and synthesis orientation, and release an open-source, hardware–software co-designed toolchain. Experiments demonstrate substantial improvements in structural rationality, diversity, and experimentally validated synthetic success rate—advancing a new paradigm for rational, high-throughput discovery of high-performance functional materials.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Understanding structure-property relationships in materials is fundamental in condensed matter physics and materials science. Over the past few years, machine learning (ML) has emerged as a powerful tool for advancing this understanding and accelerating materials discovery. Early ML approaches primarily focused on constructing and screening large material spaces to identify promising candidates for various applications. More recently, research efforts have increasingly shifted toward generating crystal structures using end-to-end generative models. This review analyzes the current state of generative modeling for crystal structure prediction and extit{de novo} generation. It examines crystal representations, outlines the generative models used to design crystal structures, and evaluates their respective strengths and limitations. Furthermore, the review highlights experimental considerations for evaluating generated structures and provides recommendations for suitable existing software tools. Emerging topics, such as modeling disorder and defects, integration in advanced characterization, and incorporating synthetic feasibility constraints, are explored. Ultimately, this work aims to inform both experimental scientists looking to adapt suitable ML models to their specific circumstances and ML specialists seeking to understand the unique challenges related to inverse materials design and discovery.

Problem

• Generative models predict and design new crystalline materials
• Review examines crystal representations and generative model strengths
• Addresses experimental evaluation and software tools for materials discovery

Innovation

• Generative models design crystal structures end-to-end
• Review examines crystal representations and model strengths
• Explores disorder modeling and synthetic feasibility constraints