Inverse Design of Amorphous Materials With Targeted Properties

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning

Background:

• Amorphous materials like glasses are crucial for energy storage, nonlinear optics, and catalysis.
• Their disordered structure offers vast design potential but is challenging to model.
• Current inverse design methods are less developed for amorphous materials due to data limitations and simulation cell size requirements.

Purpose of the Study:

• To propose and validate an inverse design method for generating amorphous material structures.
• To address the challenges of diffusion models in creating relaxed amorphous material configurations.
• To introduce new amorphous material datasets for framework evaluation and future research.

Main Methods:

• Developed AMDEN (Amorphous Material DEnoising Network), a diffusion model-based framework.
• Introduced an energy-based AMDEN variant incorporating Hamiltonian Monte Carlo refinement.
• Created diverse amorphous material datasets with varying properties and compositions.

Main Results:

• Demonstrated the effectiveness of the AMDEN framework in generating amorphous material structures.
• Showcased the capability of the energy-based variant to produce relaxed, low-energy configurations.
• Provided new datasets to facilitate further research in amorphous material design.

Conclusions:

• AMDEN offers a promising approach for the inverse design of amorphous materials.
• The integration of Hamiltonian Monte Carlo refinement is key to generating realistic, relaxed structures.
• The developed datasets will accelerate the discovery and design of novel amorphous materials.