Entwicklung quantenchemischer Methoden
In our group, the large-scale quantum chemistry program ORCA is developed. ORCA is a highly-efficient, flexible and user friendly quantum chemistry program that is intensely used by a quickly growing user community of about 15,000 researchers worldwide. Its features are fully described elsewhere[2].
ORCA features all common standard functionality involving density functional theory (DFT), correlated single- (CCSD(T)) and multireference (MR-CI, SORCI, NEVPT2) ab initio wavefunction methods, as well as semi-empirical methods. ORCA is particularly well suited for the calculation of molecular spectra and is widely used by spectroscopists in various areas of research ranging from solid state chemistry to pharmacology.
One obvious goal of the theoretical method development is to enhance the efficiency and accuracy of theoretical methods and thereby push the boundaries of what is possible with computational chemistry. In recent years, a particular focus has been the development and application of low-order scaling electron correlation methods. This has led to the domain-based local pair natural orbital (DLPNO) family of methods that we regard as ‘breakthrough’ technology in the application of ab initio quantum mechanics to chemistry.[3] In a nutshell, DLPNO methods recover about 99.9% of the canonical correlation energy but the computational effort scales linearly with system size and with a sufficiently low prefactor to treat molecules with hundreds of atoms on standard hardware.
The DLPNO family presently spans DLPNO-MP2[3f] and DLPNO-CCSD(T)[3a, 3h, 3i] in the single-reference case as well as DLPNO-NEVPT2 and DLPNO-Mk-MRCC in the multireference case. DLPNO is based on the powerful concept of SparseMaps that we have developed in order to simplify the complex task of implementing high-level, linear scaling quantum chemical methods.[3f] Methods have been developed for closed-[3a, 3h, 3i] and open-[4] shell systems, molecular properties,[3e] excited states[3d] and explicit correlation.[3b]
While the single-reference problem appears to be well under control based on the DLPNO concepts, much more work remains to be done in the multi-reference case. The latter methods are instrumental to treat complicated multiplet problems, bond breaking phenomena and magnetic properties. We have intensely worked on several aspects of this very challenging problem. First of all, DLPNO concepts have been used to achieve a linear scaling DLPNO-NEVPT2 method that provides 99.9% of the correlation energy reliably. Starting from there, several new developments are: a) the ability to treat much larger reference spaces with fifty and more orbitals using the new iterative configuration expansion (ICE-CI) algorithm (a variant of the CIPSI method[5]), b) a differential correlation dressed CAS (DCD-CAS) method to address the limitations of the internal contraction scheme, c) internally contracted multi-reference configuration interaction (FIC-MRCI)[6] and coupled cluster (FIC-MRCC) approaches to achieve higher accuracy than NEVPT2. d) a smoother approach to the one-particle basis set limit using explicit correlation and e) the incorporation of relativistic effects using quasi-degenerate perturbation theory.
These challenging developments are greatly aided by new technology to directly and automatically implement complex theories using an automatic code generator (ORCA-AGE) that reduces development times from years to days.[7]