Deep reinforcement learning for AgPd-based multimetallic nanoclusters: Accelerating global minimum discovery in...

Area of Science:

• Computational materials science
• Nanotechnology
• Chemical physics

Background:

• Multimetallic nanoclusters, especially high-entropy alloys, possess complex structures and energy landscapes.
• Traditional optimization methods struggle with the combinatorial complexity of these systems.

Purpose of the Study:

• To develop and apply a deep reinforcement learning (DRL) framework for predicting stable atomic arrangements in AgPd-based multimetallic nanoclusters.
• To validate DRL-predicted structures using rigorous computational methods.

Main Methods:

• Utilized a DRL agent with atom-centered symmetry functions and element-specific descriptors.
• Employed a hybrid action space for efficient navigation of the configuration space.
• Validated structures via density functional theory (DFT) for formation energies and ab initio molecular dynamics (AIMD) for thermal stability.

Main Results:

• Successfully predicted stable atomic configurations for AgPd-based multimetallic nanoclusters, including high-entropy alloys (e.g., Ag3Pd3Au3Pt3Cu3, Ag5Pd5Au5Pt5Cu5).
• DFT calculations confirmed the thermodynamic stability of DRL-generated structures.
• AIMD simulations demonstrated the thermal stability of high-entropy alloy nanoclusters at 300 K.

Conclusions:

• Deep reinforcement learning offers a powerful and efficient approach to overcome the challenges in predicting stable multimetallic nanocluster structures.
• This DRL framework accelerates the discovery and rational design of advanced nanomaterials.
• The study highlights the potential of AI in materials science for complex alloy systems.