Linear-scaling implementation of molecular electronic self-consistent field theory.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Traditional Self-Consistent Field (SCF) methods face computational bottlenecks with increasing molecular size.
• Diagonalization is a key bottleneck in conventional SCF calculations.
• Existing methods like DIIS can converge to saddle points instead of energy minima.

Purpose of the Study:

• To develop a linear-scaling SCF method that overcomes the limitations of traditional approaches.
• To enable accurate electronic structure calculations for large molecular systems (>1000 atoms).
• To provide a robust and reliable SCF convergence strategy.

Main Methods:

• Implementation of linear-scaling Hartree-Fock and Kohn-Sham theories.
• Utilized Roothaan-Hall (RH) energy function minimization with preconditioned conjugate-gradient (PCG) method.
• Introduced a level-shift parameter in RH Newton equations for rapid PCG convergence (LS-TRRH).
• Employed trust-region density-subspace minimization (TRDSM) for density averaging.
• Combined LS-TRRH and TRDSM into the linear-scaling trust-region SCF (LS-TRSCF) method.

Main Results:

• The LS-TRSCF method achieves linear scaling for large molecular systems.
• Demonstrated successful applications to molecules with over 1000 atoms.
• LS-TRSCF shows robust and smooth convergence, outperforming traditional SCF/DIIS.
• In specific cases, LS-TRSCF converged to a minimum while SCF/DIIS converged to a saddle point.

Conclusions:

• The developed LS-TRSCF method provides a significant advancement in computational chemistry.
• This method offers a reliable and efficient approach for electronic structure calculations of large systems.
• LS-TRSCF overcomes convergence issues inherent in traditional SCF/DIIS methods.