Accelerated Purification Using Generalized Nonpurifying Intermediate Functions for Large-Scale Self-Consistent Field...

Area of Science:

• Computational Chemistry
• Electronic Structure Theory
• Quantum Chemistry

Background:

• Purification methods are crucial for calculating idempotent density matrices in large-scale electronic structure calculations.
• Traditional recursive schemes (McWeeny, Holas) require many iterations due to initial errors in density matrix guesses.
• Existing methods rely on stability conditions with fixed points and vanishing derivatives at 0 and 1.

Purpose of the Study:

• To develop a more rapid method for obtaining converged density matrices.
• To reduce the computational cost associated with electronic structure calculations.
• To improve the efficiency of purification algorithms.

Main Methods:

• Relaxing stability conditions (fixed points, vanishing derivatives) in purification.
• Extending the recursive function space using generalized nonpurifying intermediate functions.
• Optimizing these generalized functions to approximate the step function.

Main Results:

• Achieved significantly faster convergence for density matrix calculations.
• Reduced purification cost by approximately 1.5 times compared to grand canonical purification.
• Demonstrated effectiveness on linear alkanes, diamondoid, and the protein endothelin.

Conclusions:

• The proposed method offers a substantial improvement in computational efficiency for electronic structure calculations.
• Removing traditional stability assumptions accelerates the convergence of density matrix purification.
• This technique is applicable to systems with small band gaps, such as proteins.