Goal 1
Goal 1: Foundation AI/ML Models
Build DFT-training and experimental data-training sets for predictive bespoke and foundation MLIPs for reactive electrochemical interfaces with a diversity of microenvironments.
The State of the Art
Fuel-forming reactions depend on how atoms, molecules, and electrical charge organize and move at interfaces. First-principles calculations describe atomic interactions using quantum mechanics. Their computational cost limits the system sizes and timescales that can be studied. Machine-learning models accelerate simulations, yet predictions inherit errors in the calculations used for training. Many models also omit long-range electrical interactions or lack data for reactive interfaces. Predictive models require accurate reference calculations and direct tests against measurements on systems co-designed by computational and experimental teams.
LiSA’s Research
LiSA is developing machine-learning interatomic potentials- models that predict atomic energies and forces for molecular simulations at much lower computational cost than quantum calculations. We incorporate long-range interactions and electrical response to describe how ions, solvent, and electrode atoms reorganize under applied voltage. Models are being trained for interfaces between metals, semiconductors, electrolytes, polymers, catalysts, and protective layers. These simulations will reveal how evolving interfacial structure controls reaction rate and selectivity as well as degradation.
Computations are coupled closely with experiments. LiSA uses automated measurements that combine electrochemistry with spectroscopy and imaging during operation. These measurements generate high-quality and comparable scientific data across many interface compositions. The quantitative benchmarks and higher-accuracy calculations in Goal 2 test the descriptions of elementary processes. Discrepancies guide new calculations, measurements, and model refinement, establishing where predictions are reliable.
We connect atomistic simulations to models of collective electrolyte transport and chemical reaction networks, which describe connected sequences of reaction steps leading to efficient fuel production. This links molecular interactions to local composition, reaction rates, product selectivity, and system durability. Predictions will guide interface design in Goal 3 and coupled microenvironment prototypes in Goal 4, as those prototypes provide lab-scale tests of model accuracy.