Black-Box Optimization for Automated Discovery.

Area of Science:

• Chemistry and Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Chemical and materials discovery traditionally relies on expert knowledge and iterative experimentation.
• Black-box optimization offers a powerful framework to formalize and accelerate this discovery process.
• Machine learning algorithms have recently shown success in various chemistry applications.

Purpose of the Study:

• To provide an overview of recent advancements in automated discovery, design, and optimization using black-box optimization in chemistry.
• To highlight the diverse applications and algorithms relevant to chemical research.

Main Methods:

• Review of recent studies applying black-box optimization algorithms to chemical and materials science problems.
• Discussion of algorithms including Bayesian optimization, reinforcement learning, active learning, quantum annealing, best-arm identification, and gray-box optimization.
• Exploration of the integration of machine learning with computational simulations and experimental methods.

Main Results:

• Successful applications demonstrated in designing photofunctional molecules, medical drugs, thermal emission materials, and solid electrolytes.
• Discovery of new material phases for solar cells through automated optimization.
• Identification of key algorithms and methodologies for effective application in chemical discovery.

Conclusions:

• Black-box optimization and related machine learning techniques are transforming chemical discovery and design.
• High-quality data and advancements in laboratory automation are crucial for the future success of these methods.
• Automated discovery algorithms are expected to reach or exceed human performance in specific scientific domains in the near future.