457255 Comparing Molecules and Solids Across Structural and Alchemical Space

Thursday, November 17, 2016: 12:42 PM
Yosemite A (Hilton San Francisco Union Square)
Sandip De, Material Science, EPFL, lausanne, Switzerland

Evaluating the (dis)similarity of crystalline, disordered and molecular

compounds is a critical step in the development of algorithms to classify

structures, search chemical space for better compounds and materials,

and drive the next generation of machine-learning algorithms for pre-
dicting the stability and properties of atomic systems. In recent years

several strategies have been designed [1-3] to compare atomic coor-
dination environments. In particular, the Smooth Overlap of Atomic

Positions has emerged as a natural framework to obtain translation, ro-
tation and permutation-invariant descriptors of atomic environments,

driven by the design of various classes of machine-learned inter-atomic

potentials. Here we will present few examples showcasing how one can

construct a Sketchmap[4-6] representation of databases of both molec-
ular and bulk structures, using (dis)similarity definitions based on such

local descriptors that can treat alchemical and structural complexity

within a unified framework.

[1] A. P. Bart ok, et al, Phys. Rev. B88, 054104(2013) [2] Ali Sadeghi

et al, J. Chem. Phys. 139, 184118 (2013) [3] Sandip De et al, Phys.

Rev. Lett. 112, 083401(2014) [4] G. A. Tribello et al,Proc. Acad. Natl.

Sci. U.S.A. 109 5196 (2012) [5] M. Ceriotti et al, Proc. Acad. Natl.

Sci. U.S.A. 108 13023 (2011) [6] M. Ceriotti et al, J. Chem. Theory

Comput. 9 1521 (2013)

Extended Abstract: File Not Uploaded