Semiempirical quantum chemical (SQC) methods are indispensable for performing many computational chemistry studies. They are often used for exploring large systems and carrying out large number of computations within reasonable amount of time.
My research is focused on the development of SQC methods based on the neglect of diatomic differential overlap (NDDO) integral approximation. In my presentation, I will first talk about our implementation of semiempirical unrestricted natural orbital–configuration interaction (UNO–CI) methods. They allow for black-box selection of active space orbitals, while being thousand times faster than common TD DFT techniques and often are as accurate or even better than TD DFT, which makes them useful for example in studies of organic photovoltaics materials. I will also show that semiempirical methods are useful for understanding unique electronic properties of carbon peopods used as ambipolar transistors and predicting reactivity of radicals via calculating unrestricted local electron affinities and ionization potentials.
Then I will talk about orthogonalization-corrected methods (OMx).We benchmarked these methods against huge collection of accurate reference data to identify their strengths and weaknesses for groundand excited-state properties, and specifically for noncovalent interactions, to compare with traditional MNDO-based methods and to identify the direction of future developments. Among the reference data, the W4-11 benchmark set proved to be very useful for both validation and development of new orthogonalization- and dispersion-corrected methods (ODMx). The ODMx methods have been carefully designed to be generally better than OMx methods for both ground- and excited-state properties and are as good as OMx for noncovalent interactions. The new methods have also more consistent formalism for calculating heats of formation.
Finally, I will talk about how machine learning can be used to improve the accuracy of semiempirical methods.