Phoenics: A Bayesian Optimizer for Chemistry.

Area of Science:

• Computational Chemistry
• Chemical Engineering
• Data Science

Background:

• Optimization is crucial in chemistry for identifying optimal experimental or computational conditions.
• Many real-world problems involve expensive or time-consuming objective evaluations, limiting traditional optimization approaches.
• Existing methods may struggle with complex, high-dimensional search spaces and limited data.

Purpose of the Study:

• To introduce Phoenics, a probabilistic global optimization algorithm designed for efficiently identifying optimal experimental or computational conditions.
• To address challenges in chemical optimization, such as limited evaluations and complex response surfaces.
• To provide a robust tool for accelerating scientific discovery in resource-constrained settings.

Main Methods:

• Phoenics integrates Bayesian optimization with Bayesian kernel density estimation.
• It employs a probabilistic approach to guide the search for optimal conditions.
• The algorithm facilitates efficient parallel search strategies, balancing exploration and exploitation.

Main Results:

• Phoenics effectively identifies target conditions even with limited objective evaluations.
• Benchmarks show Phoenics is less sensitive to response surface complexities than established algorithms.
• The algorithm successfully optimized the Oregonator case study, a nonlinear chemical reaction network.

Conclusions:

• Phoenics offers a powerful solution for rapid optimization of expensive-to-evaluate objective functions.
• It is particularly well-suited for applications in experimental design and computational chemistry.
• The algorithm accelerates the discovery of desired dynamic behaviors in complex chemical systems.