Symbolic Regression for Discovery of a DFT Functional - IPAM at UCLA

Explore symbolic regression for discovering DFT functionals in this 53-minute conference talk by Patrick Riley at IPAM's Learning and Emergence in Molecular Systems Workshop. Delve into the SyFES system, which combines AutoML concepts to find new DFT functionals, and learn how it rediscovered the B97 exchange functional and improved upon the wB97M-V functional. Gain insights into machine-guided searches for compact models in scientific domains, covering topics such as program operations, regularized evolution, DFT evaluation, self-consistent field calculations, and evolutionary algorithms. Compare approaches like Deep Blue vs AlphaGo and examine the impact of design choices in functional development. No prior knowledge of DFT is required to appreciate this demonstration of advanced machine learning techniques in computational chemistry.

Agenda
What is symbolic regression
Program operations
Parameters
Regularized Evolution
DFT Evaluation
DFT Setup
Problems
Selfconsistent field calculations
Decay interactions
How is this functional different
Evolutionary algorithms
Deep Blue vs Alphago
Did we just get lucky
Why didnt we get lucky
Selfconsistent calculation
The impact of reasonable choices
Conclusion