Efficient Machine Learning Prediction of Solvation Thermodynamics
T2025-212
The Need
Modeling solvent effects on catalytic surfaces is critical for designing industrial processes like biomass conversion, fuel synthesis, and electrocatalysis. Traditional multiscale simulations combining density functional theory (DFT) and molecular dynamics (MD) offer accuracy but are computationally intensive and time-consuming, limiting their scalability across diverse catalytic systems. There is a significant unmet need for a faster, cost-effective method that can predict hydration and solvation energies for surface-bound species with accuracy comparable to current gold-standard simulations.
The Technology
This innovation is a suite of machine learning (ML) models trained to predict hydration interaction energies, solvation energies, and solvation free energies for adsorbates at Pt(111) surfaces under aqueous conditions. The models are built using molecular descriptors and fingerprints derived from prior MD/DFT simulations, eliminating the need for new DFT calculations. These ML models achieve prediction accuracies (RMSE < 0.1 eV) comparable to multiscale modeling, enabling rapid, low-cost predictions for new catalytic adsorbates.
Commercial Applications
- Catalyst design for biomass reforming and Fischer–Tropsch synthesis
- Electrocatalyst screening for water splitting or fuel cells
- Solvent effect modeling in pharmaceutical surface adsorption
- Computational screening in heterogeneous catalysis research
- Accelerated materials discovery using AI-driven workflows
Benefits/Advantages
- High Accuracy: Predictive errors within the standard error of multiscale methods
- Computational Efficiency: Eliminates the need for multiscale modeling, and computes a solvation free energy in about 1/3 of the time
- Scalability: Easily extends to new adsorbates or solvents
- Interpretability: Feature importance analysis reveals physical drivers like hydrogen bonding
- Versatility: Can be integrated into existing catalyst screening pipelines