Tailoring interactions between active nematic defects with reinforcement learning.

Area of Science:

• Active Matter Physics
• Soft Condensed Matter
• Non-equilibrium Systems

Background:

• Active nematics are complex systems exhibiting dynamic patterns and flows at the micron scale.
• Optical control of molecular motors allows manipulation of non-equilibrium activity fields.
• Engineering effective control protocols for active nematics is challenging due to intricate dynamics.

Purpose of the Study:

• To explore a model-free approach for controlling active nematic fields using reinforcement learning.
• To demonstrate the ability to engineer interactions between nematic defects.
• To assess the feasibility of using low-dimensional system observables for feedback control.

Main Methods:

• Utilized reinforcement learning, a machine learning technique, for trial-and-error exploration of system dynamics.
• Bypassed the need for accurate parameterization and explicit model representation of the active nematic.
• Applied reinforcement learning to control local activity fields and influence nematic defect interactions.

Main Results:

• Successfully demonstrated that local activity fields can induce effective interactions between nematic defects.
• Enabled nematic defects to follow designer dynamical laws through engineered activity fields.
• Showcased the sufficiency of low-dimensional system observables and actions for precise control.

Conclusions:

• Reinforcement learning provides a powerful, model-free strategy for controlling active nematic systems.
• This approach allows for the engineering of specific behaviors and interactions of nematic defects.
• The findings suggest the plausibility of implementing feedback control loops in experimental and biological systems.