Pseudodiagonalization Method for Accelerating Nonlinear Subspace Diagonalization in Density Functional Theory.

Area of Science:

• Computational physics and chemistry
• Materials science

Background:

• Density functional theory (DFT) relies on self-consistent field (SCF) iterations.
• Each SCF step involves solving a linear eigenvalue problem to update Kohn-Sham orbitals.

Purpose of the Study:

• To adapt and apply the pseudodiagonalization method to finite-temperature and metallic systems.
• To improve the efficiency of SCF calculations in DFT.

Main Methods:

• Pseudodiagonalization approximates the solution to the linear eigenvalue problem using Jacobi rotations.
• This method leverages previous SCF iteration results for faster convergence.
• Applied to subspace eigenvalue problems in Chebyshev-filtered subspace iteration.

Main Results:

• Pseudodiagonalization effectively handles partially occupied orbitals in metallic systems.
• The method demonstrates rapid approximate solutions, especially near SCF convergence.
• Achieved comparable SCF convergence rates to exact diagonalization in tests.

Conclusions:

• Pseudodiagonalization is a viable and efficient alternative to exact diagonalization for SCF calculations.
• It offers significant speedups for metallic and finite-temperature systems.
• This technique enhances the computational feasibility of DFT studies.