Dear colleagues,
Below are the details for the pchem seminar with Anatole von Lilienfeld
next Thursday, Jan 30th.
Best wishes,
-Martin
*Date:*
Thursday, Jan 30, 4:15pm, Pfizer (coffee & cookies at 4pm)
*Title:*" 'Quantum Machine': Supervised learning of Schrödinger's
equation
in chemical compound space"
*Abstract:*
Many of the most relevant chemical properties of matter depend explicitly
on atomistic details, rendering an atomistic resolution of any employed
simulation model mandatory. Alas, even when using high-performance
computing, brute force high-throughput screening of all the possible
compounds is beyond any capacity for all but the simplest systems and
properties due to the combinatorial nature of chemical compound space
(compositional, constitutional, and conformational isomers). Consequently,
when it comes to the computational design of properties or systems that
require first principles calculations, a successful optimization algorithm
must not only make a trade-off between sufficient accuracy of applied
models and computational speed, but must also aim for rapid convergence in
terms of number of compounds visited. I will discuss recent contributions
related to the former aspect. More specifically, we developed statistical
models of quantum mechanical observables based on supervised learning in
chemical space. Our results suggest that for those out-of-sample molecules
that lie in interpolating regimes of chemical space, a predictive accuracy
can be achieved that comes close to ``chemical accuracy'' (~1 kcal/mol),
highly sought-after in thermochemistry, at a fraction of the computational
cost.
Show replies by date