Order- N tight-binding molecular dynamics on parallel computers

A recently proposed order- N method for tight-binding molecular dynamics is implemented on parallel machines. The Fortran code, written for the parallel computer CM-5, uses a general message passing library, being therefore highly portable. The main subjects of this implementation, which are matrix...

Full description

Saved in:
Bibliographic Details
Published inComputer physics communications Vol. 88; no. 2; pp. 173 - 185
Main Authors Itoh, Satoshi, Ordejón, Pablo, Martin, Richard M.
Format Journal Article
LanguageEnglish
Published Amsterdam Elsevier B.V 01.08.1995
Elsevier Science
Subjects
Computational techniques
Exact sciences and technology
Mathematical methods in physics
Molecular dynamics and particle methods
Physics
Tight binding approximation
Molecular dynamics
Theoretical study
Sparse matrix
Parallel computation
Wave functions
Performance
Online AccessGet full text
ISSN0010-4655
1879-2944
DOI10.1016/0010-4655(95)00031-A

Cover

More Information
Summary:A recently proposed order- N method for tight-binding molecular dynamics is implemented on parallel machines. The Fortran code, written for the parallel computer CM-5, uses a general message passing library, being therefore highly portable. The main subjects of this implementation, which are matrix partition and sparse matrix multiplication, are discussed in detail. We describe the assignment of work to each processor and the parallelization of sparse matrix multiplications. The performance of our parallel programs is analyzed by timing tests on a CM-5. It is shown that the program is both weakly and strongly scalable. With the present code, tight-binding molecular dynamics simulations in systems with up to 85184 carbon atoms are possible on a CM-5 with 512 processors.
ISSN:0010-4655
1879-2944
DOI:10.1016/0010-4655(95)00031-A