MiAD: Mirage Atom Diffusion for De Novo Crystal Generation

2025-11-18

AI Summary

Existing diffusion models for crystal generation enforce a fixed number of atoms per structure, severely constraining sampling diversity and structural flexibility. To address this limitation, we propose the first equivariant joint diffusion framework capable of generating crystals with variable atom counts. Our method introduces end-to-end differentiable modeling of atomic existence states and a novel “phantom injection” mechanism that dynamically inserts or removes atoms during denoising, enabling concurrent optimization of both crystal geometry and atomic composition. This design significantly improves generation stability, uniqueness, and novelty (S.U.N.), achieving an 8.2% S.U.N. rate on the MP-20 dataset—2.5× higher than the state of the art. The framework preserves SE(3)-equivariance throughout the generative process and supports controllable compositional sampling. Code is publicly available.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

In recent years, diffusion-based models have demonstrated exceptional performance in searching for simultaneously stable, unique, and novel (S.U.N.) crystalline materials. However, most of these models don't have the ability to change the number of atoms in the crystal during the generation process, which limits the variability of model sampling trajectories. In this paper, we demonstrate the severity of this restriction and introduce a simple yet powerful technique, mirage infusion, which enables diffusion models to change the state of the atoms that make up the crystal from existent to non-existent (mirage) and vice versa. We show that this technique improves model quality by up to $ imes2.5$ compared to the same model without this modification. The resulting model, Mirage Atom Diffusion (MiAD), is an equivariant joint diffusion model for de novo crystal generation that is capable of altering the number of atoms during the generation process. MiAD achieves an $8.2%$ S.U.N. rate on the MP-20 dataset, which substantially exceeds existing state-of-the-art approaches. The source code can be found at href{https://github.com/andrey-okhotin/miad.git}{ exttt{github.com/andrey-okhotin/miad}}.

Problem

• Enabling diffusion models to vary atom counts during crystal generation
• Overcoming limitations in sampling variability for crystalline materials
• Improving generation of stable unique novel materials via mirage infusion

Innovation

• Mirage infusion enables atom existence state changes
• Equivariant joint diffusion alters atom count during generation
• Achieves 8.2% S.U.N. rate on MP-20 dataset