Learning PDE to Model Self-Organization of Matter.

Area of Science:

• Materials Science
• Laser Physics
• Computational Science

Background:

• Femtosecond laser-induced nanopatterns on Nickel have applications in optics, microbiology, and medicine.
• The formation of these patterns is partially explained by self-organization hydrodynamic processes.
• Exploring the vast parameter space for laser patterning is experimentally challenging due to data scarcity.

Purpose of the Study:

• To develop a machine learning (ML) framework for predicting novel laser-induced nanopatterns.
• To integrate physical knowledge, specifically the Swift-Hohenberg (SH) partial differential equation (PDE), into the ML model.
• To address the challenge of limited experimental data by learning with few data points.

Main Methods:

• Utilized ML to predict novel nanopatterns by incorporating the SH PDE.
• Developed a framework to learn from limited data without initial conditions, leveraging a PDE solver.
• Created a second-order pseudospectral solver for the SH equation balancing accuracy and speed.

Main Results:

• Successfully predicted new nanopatterns that align well with experimental data.
• Demonstrated that initial conditions can be disregarded for self-organization processes, simplifying the problem to PDE parameters.
• Identified relationships between pattern features, enabling design constraints and iterative experimental data acquisition.

Conclusions:

• The proposed ML framework effectively predicts novel nanopatterns by integrating physical laws.
• The method simplifies pattern discovery by focusing on PDE parameters, overcoming experimental data limitations.
• The study highlights the limitations of the current SH model and suggests avenues for physical model improvement.