# MatterGen: a generative model for inorganic materials design

**Authors:** Claudio Zeni, Robert Pinsler, Daniel Zügner, Andrew Fowler, Matthew Horton, Xiang Fu, Sasha Shysheya, Jonathan Crabbé, Lixin Sun, Jake Smith, Ryota Tomioka, Tian Xie  
**arXiv:** [2312.03687](https://arxiv.org/abs/2312.03687)  
**Date:** 2023-12-06  
**Summary page:** https://www.summarizepaper.com/en/paper/2312.03687v1/

## Key points

- The ability to create functional materials with specific properties is crucial for advancements in technological applications such as energy storage, catalysis, and carbon capture.
- Generative models are a promising approach for designing novel materials based on desired property constraints.
- MatterGen is a groundbreaking generative model developed by a team of researchers led by Claudio Zeni, Robert Pinsler, and Daniel Zügner.
- MatterGen excels in generating stable and diverse inorganic materials across the periodic table using a unique diffusion-based generative process.
- MatterGen introduces adapter modules for fine-tuning towards specific property constraints using labeled datasets.
- Structures generated by MatterGen are significantly more likely to be both novel and stable compared to previous generative models.
- MatterGen successfully generates novel materials with tailored properties that meet predefined criteria after undergoing fine-tuning with adaptive modules.
- MatterGen showcases multi-property design capabilities by proposing structures with high magnetic density and low supply-chain risk chemical composition.

## Summary

In the field of materials design, the ability to create functional materials with specific properties is crucial for advancements in various technological applications such as energy storage, catalysis, and carbon capture. Generative models have emerged as a promising approach to designing novel materials by directly generating them based on desired property constraints. However, existing generative models often struggle to propose stable crystals or are limited in their ability to satisfy a wide range of property constraints. In response to these challenges, a team of researchers led by Claudio Zeni, Robert Pinsler, and Daniel Zügner has developed MatterGen - a groundbreaking generative model that excels in generating stable and diverse inorganic materials across the periodic table. What sets MatterGen apart is its unique diffusion-based generative process which refines atom types, coordinates, and the periodic lattice to produce crystalline structures with high stability and novelty. Moreover, MatterGen introduces adapter modules that enable fine-tuning towards specific property constraints using labeled datasets. This fine-tuning capability allows researchers to steer the generation process towards desired chemistry, symmetry, mechanical properties, electronic properties, magnetic properties and more. Compared to previous generative models, structures generated by MatterGen are significantly more likely to be both novel and stable while being much closer to the local energy minimum. After undergoing fine-tuning with MatterGen's adaptive modules,the model successfully generates novel materials with tailored properties that meet predefined criteria. Notably,MatterGen showcases its multi-property design capabilities by proposing structures that exhibit high magnetic density alongside a chemical composition with low supply-chain risk. The research team led by Claudio Zeni et al.believes that the exceptional quality of generated materials and the extensive capabilities of MatterGen represent a significant advancement towards establishing a universal generative model for materials design.With its ability to generate stable and diverse inorganic materials while meeting specific property constraints across various domains of application, MatterGen holds great promise for driving innovation in material science and engineering.