Machine-learned potentials for solvation modeling.

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Physical Chemistry

Background:

• Solvent environments critically influence molecular properties, but first-principles modeling is computationally expensive.
• Accurate solvation modeling is essential for understanding chemical processes.

Purpose of the Study:

• To review the development and application of machine-learned potentials (MLPs) for solvation modeling.
• To provide a classification of MLPs based on their training targets, model types, and design choices.
• To discuss the integration of MLPs into existing solvation workflows.

Main Methods:

• Summarizing the theoretical basis of MLP-based energy and force predictions.
• Classifying MLPs by training targets, model architectures, descriptors, and training protocols.
• Reviewing case studies involving small molecules, interfaces, and reactive systems.

Main Results:

• MLPs provide first-principles accuracy at significantly reduced computational cost.
• MLPs can effectively model complex solvation effects like hydrogen bonding and polarization.
• This review categorizes various MLP approaches and their integration strategies.

Conclusions:

• MLPs are powerful tools for efficient and accurate solvation modeling.
• Future work should focus on developing transferable, robust, and physically grounded MLPs.
• MLPs are poised to revolutionize atomistic modeling of solvated systems.