Machine Learning: A Route to High Accuracy Simulation of Complex Reacting Systems

Explore a virtual seminar on machine learning applications for high-accuracy simulations of complex reacting systems. Delve into the challenges of simulating chemistry-driven evolution in multiscale systems and discover how machine-learned interatomic models, particularly CHIMES, offer innovative solutions. Learn about CHIMES' capabilities, including rapid tuning of existing models through "delta learning" and its versatility in handling complex problems. Investigate the application of these techniques in understanding shockwave-driven nanocarbon synthesis and their potential for simulating large systems where quantum mechanics predictions fall short.

Syllabus

Intro
LAB OVERVIEW
MANY INTERESTING PROBLEMS INVOLVE CHEMISTRY-DRIVEN EVOLUTI INHERENTLY MULTISCALED SYSTEMS, BUT SIMULATING THEM IS HARD
WHAT IS A MACHINE LEARNED (ML) INTERATOMIC MODEL (IAM)?
OUR ML INTERATOMIC MODEL (CHIMES) PROVIDES A PATH FORWARD
WHAT CAN CHIMES DO?
CHIMES CAN RAPIDLY TUNE EXISTING MODELS USING MINIMAL TRAINING SETS, VIA "DELTA LEARNING"1-3
COMPLEX PROBLEMS REQUIRE MORE SOPHISTICATED FITTING FRAMEWORKS
CHIMES IS SURPRISINGLY VERSATILE!
QM DOESN'T PREDICT CLUSTERS, BUT WHAT ABOUT LARGE SYSTEMS?
UNDERSTANDING SHOCKWAVE-DRIVEN NANOCARBON SYNTHESIS