PAL 2.0: a physics-driven bayesian optimization framework for material discovery.

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Efficient discovery of advanced functional materials is hindered by vast combinatorial spaces and the high cost of characterization.
• Current search methods often rely on expert knowledge, potentially overlooking high-performing materials in unexplored regions.
• There is a critical need for computational algorithms capable of efficiently navigating large parameter spaces for materials discovery.

Purpose of the Study:

• To introduce PAL 2.0, a novel computational framework for efficient materials discovery.
• To combine physics-based surrogate models with Bayesian optimization for accelerated materials design.
• To demonstrate the efficacy of PAL 2.0 across diverse material systems.

Main Methods:

• PAL 2.0 integrates XGBoost and Neural Networks to generate physics-based hypotheses, serving as priors for a Gaussian process model.
• Bayesian optimization is employed to efficiently search the material design space guided by the physics-based priors.
• The approach was validated on three distinct material test cases: photovoltaic perovskites, perovskite-solvent systems for solution processing, and organic thermoelectric semiconductors.

Main Results:

• PAL 2.0 demonstrated superior efficiency in searching the material design space compared to state-of-the-art methods.
• The physics-based surrogate models within PAL 2.0 exhibited lower prediction errors for unseen material compositions.
• The framework successfully identified optimal candidates in the tested material discovery scenarios.

Conclusions:

• PAL 2.0 offers a significant advancement in computational materials discovery, particularly for data-scarce scenarios.
• The combination of physics-based priors and Bayesian optimization provides a powerful and efficient approach to materials design.
• This method accelerates the discovery of next-generation functional materials for energy, health, and sustainability applications.