Optimizing disorder with machine learning to harness phase synchronization

Area of Science:

• Nonlinear dynamics
• Complex systems
• Machine learning

Background:

• Disorder is typically viewed as detrimental to system coherence.
• However, specific conditions can enable disorder to promote synchronization.
• Understanding and controlling disorder-induced synchronization is crucial for complex dynamical networks.

Purpose of the Study:

• To develop a machine-learning framework for the inverse design of optimal-disorder configurations.
• To maximize phase synchronization in coupled driven nonlinear systems.
• To efficiently explore high-dimensional parameter spaces for synchronization optimization.

Main Methods:

• Utilized an array of forced and damped nonlinear pendulums with controlled disorder and noise.
• Trained a feedforward neural network (FNN) to predict synchronization strength (Shannon entropy index) from disorder parameters.
• Employed the FNN as a surrogate for computationally intensive stochastic differential equation simulations.

Main Results:

• The trained FNN accurately predicted synchronization strength based on disorder parameters.
• The framework efficiently identified optimal disorder configurations that maximize phase synchronization.
• Demonstrated that machine learning can capture and optimize disorder-induced synchronization.

Conclusions:

• Machine learning provides an effective tool for optimizing synchronization in complex dynamical networks.
• Inverse design using ML can identify beneficial roles of disorder in system coherence.
• This approach offers a computationally efficient alternative to traditional simulation methods.