HPCRD Logo

Scalable Methods for Electronic Excitations and Optical Responses of Nanostructures

News

Motif-based Hessian Preconditioner

Computed structure for a 339-atom silicon quantum dot (Si235H104) using a new motif-based Hessian preconditioned CG for the solution of an ab initio atomic relaxation. The unpreconditioned CG method took 31 iterations, where each iteration used on average 140 minutes on 32 processors of the NERSC Seaborg system. The preconditioned CG method took 6 iterations, resulting in a total time savings of 1867 CPU hours per simulation. For details see Motif-based Hessian matrix for ab initio geometry optimization of nanostructures, Zhao, Wang, and Meza, Technical Report LBNL-59974 (2006).

 

 

 

 

This project is jointly funded by DOE Office of Science Basic Energy Sciences and Advanced Scientific Computing Research.

About LBNL | Privacy Policy | Contact Us