Physics Dept - MtSE Joint Seminar

December 12, Thursday (*SPECIAL DAY*)

Machine Learning with Schrödinger Equation

Dr. Guangqi Li

Dept. of Chemistry, Columbia Univ.

(Materials Physics, Host: Ken Ahn)

*TALK*: Tiernan 409, 11:30am - 12:30pm

(* SPECIAL TIME/ROOM and NO TEA TIME *)

*LUNCH*: Tiernan 406, 12:30pm - 3pm

Machine learning (ML) is a method of data analysis, with the powerful ability in predicting. It had become overwhelming, in atomistic simulation and electronic property predictions. Recently, ML had been utilized in quantum system, Density Function Theory (DFT), molecular dynamics, and even in predicting the reaction performance for catalysis, and the protein-ligand interaction. So far, two large databases had been set up. One includes 134 kilo molecules with their quantum chemistry structures and properties. Another includes 20 million calculated off-equilibrium conformations for 57462 small organic molecules. These databases were obtained from DFT. Due to the self-interaction term (electron interacts with itself in mathematical equation), DFT is well-known for its error. The hypothesis of Localized Orbital Correction (via adding extra operators to remove the error) had been proposed to the numerical as atomic energies, ionization potential, and the 3d electron in transition metals. Combining this hypothesis, the new obtained databases will have the high accuracy when compared to the experiment or the benchmark quantum chemical calculations.