Efficient Phase Diagram Sampling by Active Learning.

Area of Science:

• Computational physics
• Materials science
• Machine learning

Background:

• Efficient phase diagram sampling is crucial for understanding material properties.
• Traditional methods like grid search are inefficient for complex systems, especially active matter.
• Existing advanced techniques may not generalize to non-equilibrium systems.

Purpose of the Study:

• To develop a more efficient method for phase diagram sampling.
• To reduce the number of sampled state points required for high-precision phase boundary determination.
• To overcome the limitations of grid search in simulating active matter systems.

Main Methods:

• Adoption of active learning techniques from machine learning.
• Gaussian process regression for interpolating sampled phase data.
• Acquisition functions to identify the most informative next state points.
• Generalization with asynchronous sampling and multi-replica uncertainty incorporation.

Main Results:

• Achieved significant reduction in sample size for phase diagram determination.
• Demonstrated a 5x sample efficiency improvement for a mixture of active and passive colloids.
• Successfully mapped a previously unstudied equilibrium phase boundary of quasicrystals with high precision.

Conclusions:

• The active learning framework dramatically accelerates phase diagram exploration.
• This method enhances the efficiency of simulations and experimental studies.
• Opens new avenues for large-scale phase diagram investigations across various scientific domains.