Minimum Energy Conical Intersection Optimization Using DFT/MRCI(2).

Area of Science:

• Computational chemistry
• Theoretical chemistry
• Quantum chemistry

Background:

• The combined density functional theory and multireference configuration interaction (DFT/MRCI) method offers computational efficiency and accuracy for electronic excited states.
• A key challenge is constructing smooth potential energy surfaces due to discontinuities from selected-CI procedures.

Purpose of the Study:

• To develop a method for learning smooth potential energy surfaces from DFT/MRCI calculations.
• To address discontinuities in potential energy surfaces, particularly at conical intersections.

Main Methods:

• Utilized Gaussian process regression to treat discontinuities as noise.
• Incorporated and optimized a white-noise kernel within the regression framework.
• Learned characteristic polynomial coefficient surfaces instead of adiabatic energies.

Main Results:

• Successfully learned smooth potential energy surfaces for molecules like ethylene, butadiene, and fulvene.
• Optimized minimum energy conical intersection geometries using the learned surfaces.
• Obtained structures and branching spaces comparable to ab initio MRCI results.

Conclusions:

• The Gaussian process regression approach provides a viable method for learning smooth DFT/MRCI(2) surfaces.
• This technique improves the simulation of electronic spectroscopies by handling surface discontinuities.
• The method demonstrates good agreement with high-level ab initio calculations.