Apr 7, 2025
2:30pm - 2:45pm
Summit, Level 4, Room 422
Justin Tahmassebpur1,Boris Barron1,Brandon Li1,Peter Frazier1,Hector Abruna1,Tomas Arias1
Cornell University1
Justin Tahmassebpur1,Boris Barron1,Brandon Li1,Peter Frazier1,Hector Abruna1,Tomas Arias1
Cornell University1
The accelerated discovery of advanced materials is crucial for breakthroughs in energy conversion, catalysis, and other material domains including superconductivity. In this work, we introduce effective atom theory (EAT), a novel method that significantly enhances the speed of materials optimization and is readily implemented within an ab initio density-functional theory (DFT) framework. Traditional optimization methods rely on brute-force combinatorial searches or more advanced discrete algorithms such as Bayesian optimization, both of which scale exponentially with the number of atoms involved, rendering them computationally infeasible for complex systems. EAT addresses this challenge by allowing each atom to temporarily be represented as a continuous mixture of potential elements, enabling the use of far more efficient gradient-based optimization algorithms. This approach allows for exponential speed ups in navigating vast material design spaces. A key innovation of EAT is the "atomization" process we developed, which ensures that our final recommendations correspond to real elements when we identify the optimal material. We demonstrate the efficacy of EAT by designing high-entropy rock salt oxide (HERSO) electrocatalysts that optimize both the onset potential of the oxygen evolution reaction (OER) and the stability of the HERSO material in acidic and alkaline media under operational electrochemical conditions.