Truncation of small matrix elements based on the Euclidean norm for blocked data structures.

Area of Science:

• Computational Quantum Chemistry
• Electronic Structure Theory

Background:

• Large-scale Hartree-Fock (HF) and Kohn-Sham (KS) calculations require efficient methods to manage large matrices.
• Enforcing matrix sparsity is crucial for computational feasibility, but must be balanced with controlling calculation errors.

Purpose of the Study:

• To present novel methods for removing small symmetric matrix elements based on the Euclidean norm of the error matrix.
• To enable error control in the occupied subspace during large-scale electronic structure calculations.

Main Methods:

• Truncation schemes based on the unitary-invariant Euclidean norm of the error matrix.
• Repetitive application of the Lanczos method to compute Euclidean norms of candidate error matrices.
• Utilization of Ritz value convergence patterns to minimize the number of Lanczos iterations.

Main Results:

• The Euclidean norm is demonstrated to be a suitable unitary-invariant metric for error control in large systems.
• The proposed truncation schemes effectively reduce matrix size while maintaining controlled error levels.
• Ritz value convergence accelerates the computation by reducing the required Lanczos iterations.

Conclusions:

• The developed methods provide an efficient strategy for sparse matrix generation in large-scale HF and KS calculations.
• Error control is maintained through the use of the Euclidean norm, ensuring the reliability of truncated matrices.
• The combination of Euclidean norm truncation and Lanczos iteration offers a computationally advantageous approach.