From predicting to learning dissipation from pair correlations of active liquids

Area of Science:

• Soft Matter Physics
• Statistical Mechanics
• Materials Science

Background:

• Active systems, driven by non-conservative forces, exhibit unique behaviors and structures.
• A key challenge is linking the static structure of these systems to energy dissipation.
• Understanding this connection is crucial for designing novel active materials.

Purpose of the Study:

• To analytically and computationally connect the static structure of active matter to energy dissipation.
• To develop a predictive framework for dissipation based on particle correlations.
• To explore the relationship between structure and dissipation, especially near nonequilibrium phase transitions.

Main Methods:

• Utilized liquid-state theories and machine learning tools.
• Analytically demonstrated the relationship between dissipation and pair correlations in isotropic active matter.
• Extended a nonequilibrium mean-field framework to predict dissipation from pair correlations.
• Constructed a neural network to map static configurations to dissipation rates.

Main Results:

• Established a close relationship between dissipation and pair correlations when driving forces act as an active temperature.
• Revealed a robust analytic relation between dissipation and structure, valid even near nonequilibrium phase transitions.
• Developed a predictive theory applicable to strongly interacting particles far from equilibrium.
• Successfully trained a neural network to predict dissipation from static configurations without prior dynamic knowledge.

Conclusions:

• The study provides a novel theoretical framework linking static structure to energy dissipation in active matter.
• The findings offer new perspectives on the interplay between dissipation and self-organization in nonequilibrium systems.
• This work facilitates the rational design of active materials by providing predictive tools for energy dissipation.