Iterative diagonalization of symmetric matrices in mixed precision and its application to electronic structure calculations

  title={Iterative diagonalization of symmetric matrices in mixed precision and its application to electronic structure calculations},
  author={Eiji Tsuchida and Yoong‐Kee Choe},
  journal={Comput. Phys. Commun.},

Figures and Tables from this paper

Mixed precision algorithms in numerical linear algebra
This survey treats a broad range of mixed precision algorithms in numerical linear algebra, both direct and iterative, for problems including matrix multiplication, matrix factorization, linear systems, least squares, eigenvalue decomposition and singular value decomposition.
Fast, scalable and accurate finite-element based ab initio calculations using mixed precision computing: 46 PFLOPS simulation of a metallic dislocation system
An unprecedented sustained performance of 46 PFLOPS (27.8% peak FP64 performance) is demonstrated on a dislocation system in Magnesium containing 105,080 electrons using 3,800 GPU nodes of Summit supercomputer, which is the highest performance to-date among DFT codes.
An efficient time-stepping scheme for ab initio molecular dynamics simulations
If the Verlet method is used in conjunction with pre- and postprocessing, the accuracy of the time integration is significantly improved with only a small computational overhead and the structural properties obtained are found to be sufficiently accurate even for large time steps close to the stability limit.
An adaptive finite-element method for large-scale ab initio molecular dynamics simulations.
The current status of the finite-element method for large-scale atomistic simulations based on the density-functional theory is presented, and ab initio molecular dynamics simulations on sulfonated poly( 4-phenoxybenzoyl-1,4-phenylene) (SPPBP), which is a typical example of polymer electrolyte membranes for fuel cells are presented.
Stabilization ofAb InitioMolecular Dynamics Simulations at Large Time Steps
The Verlet method is still widely used to integrate the equations of motion in ab initio molecular dynamics simulations. We show that the stability limit of the Verlet method may be significantly
Effect of Elevated Temperatures on the States of Water and Their Correlation with the Proton Conductivity of Nafion
The results indicated that the amount of randomly arranged water gradually increased and created more H-bonded water networks in Nafion at above 60% RH, and from the deconvolution of the O–H bending band, it was found that the volume fraction fi (i=each deconvoluted band) of H-magnifying water for elevated temperatures increased remarkably higher than for 60 °C.


Exploiting Mixed Precision for Computing Eigenvalues of Symmetric Matrices and Singular Values
The use of higher precision preconditioning for the symmetric eigenvalue problem and the singular value problem of general non–structured non–graded matrices are discussed. The matrix Q from the
Towards grid-based OÑNÖ density-functional theory methods: Optimized nonorthogonal orbitals and multigrid acceleration
We have formulated and implemented a real-space ab initio method for electronic structure calculations in terms of nonorthogonal orbitals defined on a grid. A multigrid preconditioner is used to
Accelerating scientific computations with mixed precision algorithms
Finite element methods in ab initio electronic structure calculations
The construction and properties of the required FE bases and their use in the self-consistent solution of the Kohn–Sham equations of density functional theory are reviewed.
Limited-Memory Reduced-Hessian Methods for Large-Scale Unconstrained Optimization
Numerical results from a set of unconstrained problems in the CUTE test collection indicate that the proposed reduced-Hessian method is competitive with the limited-memory codes L-BFGS and L- BFGS-B.
An Efficient Algorithm for Electronic-Structure Calculations
We show how to adapt the quasi-Newton method to the electronic-structure calculations using systematic basis sets. Our implementation requires less iterations than the conjugate gradient method,
Efficiency of algorithms for Kohn-Sham density functional theory