Generating Multiscale Amorphous Molecular Structures Using Deep Learning: A Study in 2D.

Area of Science:

• Materials Science
• Computational Chemistry
• Biophysics

Background:

• Amorphous molecular assemblies are prevalent in nature and technology.
• Understanding their mesoscale properties requires detailed structural knowledge.
• Current simulation methods face scalability challenges with increasing system size.

Purpose of the Study:

• To develop a deep learning method for generating large-scale amorphous molecular assemblies.
• To overcome the computational limitations of traditional simulation techniques.
• To enable mesoscale simulations of disordered systems from nanoscale data.

Main Methods:

• An autoregressive deep learning model is employed.
• The method leverages finite-range structural correlations.
• Generation is performed from small-scale computational or experimental samples.

Main Results:

• The method successfully generates disordered molecular aggregates of arbitrary size.
• Performance was benchmarked on self-assembled nanoparticle aggregates.
• Atomistic resolution simulation of monolayer amorphous carbon was achieved.

Conclusions:

• The presented deep learning approach effectively bridges nanoscale and mesoscale simulation gaps.
• This method offers a scalable solution for studying amorphous materials.
• Enables precise control and understanding of amorphous material properties.