Modelling ligand exchange in metal complexes with machine learning potentials.

Area of Science:

• Computational Chemistry
• Physical Chemistry
• Materials Science

Background:

• Metal ions are crucial in catalysis and self-assembly, but accurately modeling their behavior in solution is difficult.
• Existing methods like force fields and ab initio calculations have limitations for metal-ligand complexes.

Purpose of the Study:

• To develop a computationally efficient strategy for modeling metal ions in diverse chemical environments.
• To train machine learning potentials (MLPs) for metal-ligand complexes in explicit solvents using MACE.

Main Methods:

• Utilized MACE, an equivariant message-passing neural network, to train MLPs.
• Investigated Mg2+ in water and Pd2+ in acetonitrile as model systems.
• Focused on equilibrium structures and ligand exchange dynamics.

Main Results:

• Trained MLPs accurately reproduced equilibrium structures, including coordination numbers and geometries.
• MLPs successfully modeled structural changes and free energy barriers for ligand exchange.
• Demonstrated accurate prediction of metal-ligand complex behavior in solution.

Conclusions:

• The developed strategy offers a computationally efficient method for modeling metal ions in solution.
• This approach facilitates the study of larger and more complex metal-containing systems.
• Enables advancements in understanding metal ions in biomolecules and supramolecular assemblies.