Exploring chemical space with discrete, gradient, and hybrid optimization methods.

Area of Science:

• Computational chemistry
• Molecular modeling
• Chemical informatics

Background:

• Discovering molecules with specific properties is crucial for materials science and drug discovery.
• Existing optimization methods face challenges in efficiency and applicability to diverse chemical structures.

Purpose of the Study:

• To compare the cost and performance of discrete, gradient, and hybrid optimization methods.
• To identify robust strategies for inverse chemical design.
• To maximize the static first electronic hyperpolarizability of molecules.

Main Methods:

• Utilized a tight-binding model for molecular property calculations.
• Applied discrete branch and bound, gradient, and hybrid optimization techniques.
• Developed a hybrid discrete-gradient strategy based on linear combination of atomic potentials.

Main Results:

• Discrete branch and bound methods demonstrate robustness for inverse chemical design.
• The hybrid discrete-gradient method significantly enhances gradient-based approaches.
• The hybrid method outperforms dead-end elimination and rivals branch and bound and genetic algorithms.

Conclusions:

• Hybrid optimization strategies offer improved performance for molecular property optimization.
• Branch and bound methods are cost-effective for moderate-sized molecular design problems.
• The developed hybrid approach advances inverse chemical design capabilities.