Efficient Eigensolver Methods For very large, sparse matrices using high performance computers
|Date/Time:||Wednesday, 06 Apr 2016 from 4:10 pm to 5:00 pm|
|Location:||Room 3 Physics Building|
|Actions:||Download iCal/vCal | Email Reminder|
In order to solve forefront problems in quantum many-body theory, we need algorithms tailored for today's high performance computers. Expressing the quantum many-body problem as a Hamiltonian matrix in a basis representation produces the challenge of solving a large sparse matrix eigenvalue problem. The real-world problem of solving ab initio nuclear physics problems requires (1) ability to efficiently evaluate and store the non-vanishing matrix elements for a complicated strong interaction with three-particle interactions and Coulomb contributions, and (2) an efficient method to solve for the lowest 10-100 eigenvalues and eigenvectors for matrix dimensions exceeding one billion basis states.
I will concentrate on an efficient eigensolver based on the Lanczos algorithm suitable for high performance computers with hundreds of thousands of cores.