Machine learning many-body potentials for colloidal systems.

Area of Science:

• Computational physics
• Soft matter physics
• Machine learning applications

Background:

• Simulating colloidal suspensions with multiple length and time scales is computationally intensive.
• Microscopic species (ions, depletants) add complexity to mesoscopic particle simulations.

Purpose of the Study:

• To develop a computationally efficient machine learning (ML) approach for simulating colloidal suspensions.
• To integrate out microscopic degrees of freedom and model mesoscopic particles with effective potentials.

Main Methods:

• Utilized a machine learning approach to derive effective many-body potentials for mesoscopic particles.
• Fitted ML potentials using symmetry functions based on colloid coordinates.
• Applied the ML method to a colloid-polymer mixture system.

Main Results:

• The ML potentials were found to be effectively state-independent, enabling direct-coexistence simulations.
• Achieved a reduction in computational cost by several orders of magnitude compared to traditional methods.
• Accurately described the phase behavior and structure of the colloid-polymer mixture.

Conclusions:

• The ML approach offers a significant computational advantage for simulating complex colloidal systems.
• Effective many-body potentials derived via ML accurately capture system behavior, even with dominant many-body contributions.