Geometry Optimization: A Comparison of Different Open-Source Geometry Optimizers.

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Quantum Chemistry

Background:

• Accurate molecular geometry is crucial for understanding chemical properties and reactions.
• Efficient geometry optimization algorithms are essential for computational chemistry.
• Evaluating open-source tools is vital for accessibility and reproducibility.

Purpose of the Study:

• To compare the performance of various open-source geometry optimization codes interfaced with the QUantum Interaction Computational Kernel (QUICK).
• To assess the impact of coordinate systems (internal vs. Cartesian) on optimization efficiency.
• To evaluate a machine learning approach (Gaussian process regression) for energy minimization.

Main Methods:

• Energy minimizations performed on the Baker test set of 30 organic molecules.
• Comparison of geometry optimization codes including geomeTRIC, DL-FIND, and ASE (with Berny and Sella algorithms).
• Utilized QUICK for gradient and energy calculations, with Hartree-Fock theory and 6-31G** basis set for machine learning evaluation.

Main Results:

• Internal coordinates consistently required fewer optimization steps than Cartesian coordinates.
• The choice of initial Hessian and update method significantly affected convergence rates.
• Gaussian process regression (GPR) models showed superior performance with internal coordinates.
• ASE/Berny and ASE/Sella demonstrated faster convergence compared to other tested methods.

Conclusions:

• Internal coordinates are recommended for faster geometry optimization.
• ASE/Berny and ASE/Sella are suitable for large-scale optimization tasks.
• QUICK integrated with DL-FIND offers a viable single-executable solution for optimization.