Regensburg 2016 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 88: Frontiers of Electronic Structure Theory: Focus on Topology and Transport V
O 88.6: Talk
Thursday, March 10, 2016, 16:15–16:30, H24
CELL: a python package for cluster expansions with large parent cells — •Santiago Rigamonti1, Maria Troppenz1, Christopher Sutton2, Luca M. Ghiringhelli2, and Claudia Draxl1 — 1Humboldt-Universität zu Berlin — 2Fritz-Haber-Institut der Max-Planck-Gesellschaft
The discovery of new materials for applications in areas such as energy harvesting, relies more and more on the accurate theoretical description of complex structures with large unit cells. The properties of interest are often tuned by substitutional dopants. Due to the vast configurational dopant space, a wide-spread approach is the cluster expansion (CE) technique. Most available CE codes are designed for alloys based on small parent cells, with usually 1 to 4 atoms. For the many important materials with much larger parent cells such approaches can’t be applied. We devise an iterative scheme, based on efficient samplings of the configurational space, avoiding full structure enumerations. CELL consists of several modules that can be used independently, enabling to design CEs for specific purposes. Various CE schemes are available, offering ℓ2 and ℓ1 norms as penalization terms and different cross-validation strategies. Methods such as LASSO and split Bregman iteration are available for dealing with the ℓ1 norm (compressive sensing). Access to finite-temperature properties and the characterization of phase transitions is possible through the Wang-Landau and diffusive nested sampling modules. Examples are presented for type-I thermoelectric clathrates, with 46 sites in the parent cell.