Machine-learned force field development from DFT data

Molecular dynamics simulations based on Density functional theory is an effective way to study systems where polarization plays an important role or chemical bond breaking and formation are involved. However, the computational cost of treating  a large system ( having more than few hundred atoms) with explicit atoms is currently prohibitively expensive and beyond the routine practice. In order to solve this bottleneck, we are currently using machine learning techniques (artificial neural network) to learn the forces and energies generated by appropriate density functional theory. When properly trained, the machine-learned force field can produce the accuracy of the DFT functional but provide a few-order of magnitude faster way of calculating energy and forces. This faster calculation can afford simulating complex molecular phenomena on a quantum level that was not possible to explore previously.

Figure Courtesy 

Understanding solid-liquid interfaces relevant to basic energy science

Solid -liquid interfaces play an important role in many chemical systems e.g. in electrochemistry and gas hydrate formation. Electrode-electrolyte interface plays a critical role in the design of energy storage devices. Concentrated electrolytes ( e.g. ionic liquids) are promising electrolytes in Li and beyond-Li ion batteries. We employ machine learning force field and statistical mechanical tools to understand the structure and dynamics of graphite and metal electrode with ionic liquids and leverage that understanding to design new electrolytes with improved performance.

Figure Courtesy 

Designing intermetallic alloy nano-particles for electro-catalysis

Metal nanoparticles are promising electrocatalysts for many electrochemical reactions e.g.  oxygen reduction reaction, hydrogen evolution reaction, and CO2 reduction reaction. Platinum-group metals are the most commonly used electrocatalysts which are scarce and costly. Intermetallic nanoparticles alloy where  low cost transition metals can be mixed with platinum-group metal is a promising new technology to reduce the use of costly metals. We employ machine learning, quantum mechanics and statistical mechanics to understand molecular level details and to build structure-activity correlation in intermetallic nanoparticles alloy and consequently pave the way of designing new electrocatalysts.

Figure Courtesy