Advanced AI approaches for Electrolyte Discovery*
Session Number
1
Advisor(s)
Dr. Rajeev S. Assary, Argonne National Laboratory
Location
A115
Discipline
Chemistry
Start Date
15-4-2026 10:15 AM
End Date
15-4-2026 11:00 AM
Abstract
With the world increasingly relying on rechargeable batteries, the discovery of cheaper and more efficient batteries has grown in importance. It is impractical and expensive to discover new materials through repeated experiments sifting through possible electrolyte/solvent/additive candidates. Computational techniques are an alternative to this by calculating the properties between these molecules in realistic and accurate simulations. However, many of these methods including DFT calculations require large computational resources and time. Recently, developments in artificial intelligence (AI) and neural networks (NN) have allowed Machine Learning Interatomic Potentials (MLIPs) to become an accessible and realistic alternative to traditional calculations. By using the software Architechtor, we can generate various three dimensional complexes from the Simplified Molecular Input Line Entry System (SMILES). We subsequently use MLIPs to optimize the electronic structures to more accurately represent their equilibrium geometries and employ DFT calculations for final property predictions. Combining these methods, we can accurately predict and observe trends for thousands of molecules, reducing the cost of molecular discovery. This investigation allows us to benchmark the applicability of AI approaches to molecular simulations, and hence build agents to perform autonomous computational experiments that develop a library of fundamental knowledge to enable a priori electrolyte discovery.
Advanced AI approaches for Electrolyte Discovery*
A115
With the world increasingly relying on rechargeable batteries, the discovery of cheaper and more efficient batteries has grown in importance. It is impractical and expensive to discover new materials through repeated experiments sifting through possible electrolyte/solvent/additive candidates. Computational techniques are an alternative to this by calculating the properties between these molecules in realistic and accurate simulations. However, many of these methods including DFT calculations require large computational resources and time. Recently, developments in artificial intelligence (AI) and neural networks (NN) have allowed Machine Learning Interatomic Potentials (MLIPs) to become an accessible and realistic alternative to traditional calculations. By using the software Architechtor, we can generate various three dimensional complexes from the Simplified Molecular Input Line Entry System (SMILES). We subsequently use MLIPs to optimize the electronic structures to more accurately represent their equilibrium geometries and employ DFT calculations for final property predictions. Combining these methods, we can accurately predict and observe trends for thousands of molecules, reducing the cost of molecular discovery. This investigation allows us to benchmark the applicability of AI approaches to molecular simulations, and hence build agents to perform autonomous computational experiments that develop a library of fundamental knowledge to enable a priori electrolyte discovery.