A trust-region augmented Hessian implementation for state-specific and state-averaged CASSCF wave functions

Area of Science:

• Quantum Chemistry
• Computational Chemistry
• Theoretical Chemistry

Background:

• Complete Active Space Self-Consistent Field (CASSCF) is crucial for accurately describing electron correlation in molecules.
• Existing CASSCF methods can face convergence challenges, especially for complex systems.
• State-specific (SS) and state-averaged (SA) CASSCF approaches are widely used but require efficient algorithms.

Purpose of the Study:

• To present a novel one-step second-order converger for SS and SA CASSCF wave functions.
• To enhance the robustness and reliability of CASSCF calculations.
• To enable accurate electronic structure calculations for larger and more complex molecular systems.

Main Methods:

• Implementation of a Trust-Region Augmented Hessian (TRAH) algorithm for robust convergence.
• Utilizing exponential parameterization of variational configuration parameters with a nonredundant orthogonal complement basis.
• Developing an integral-direct approach using intermediates in atomic-orbital or molecular-orbital bases.
• Integration with efficient integral decomposition techniques (e.g., resolution-of-the-identity).

Main Results:

• The TRAH-CASSCF algorithm achieves robust convergence for both SS and SA CASSCF wave functions.
• The method demonstrates reliable convergence to true minima, outperforming first-order methods when they fail.
• Efficient handling of large molecular systems, demonstrated by calculations on a 231-atom Ni(II) complex.
• Competitive runtime performance compared to other state-of-the-art second-order algorithms.

Conclusions:

• The TRAH-CASSCF method provides a reliable and efficient approach for electronic structure calculations.
• This advancement facilitates the study of complex chemical systems previously intractable with standard methods.
• The TRAH algorithm offers improved convergence properties, particularly in challenging cases.