Gershom (Jan M.L.) Martin
Department of Organic Chemistry
Weizmann Institute of Science
7610001 Reḥovot, Israel
Video is available only for registered users.
While density functional theory has made great strides, even the best exchange-correlation functionals are about one order of magnitude less accurate than can be achieved using modern wavefunction ab initio techniques. The latter have a well-defined road map for refinement in accuracy; however, their steep computational cost scaling with system size limits their use to relatively small molecules. While some applications (e.g., atmospheric chemistry, fine thermochemistry) demand such levels of accuracy, a perhaps more important application is the creation of benchmark datasets for the parametrization and validation of density functional, reactive force fields, and other lower-cost methods.
Using the case of total atomization energies, we will discuss the breakdown of molecular binding energies into their constituent components, as well as the optimal convergence strategy for each. By such “layered approximations” as implemented in the Weizmann-n series of thermochemistry protocols [1,2,3] (and its ‘competitor’, the HEAT approach ), CPU times and memory requirements can be drastically reduced versus brute-force approaches. The introduction of explicitly correlated coupled cluster theory brings still larger molecules within reach, as long as non dynamical correlation effects are not too important. (See  for a discussion of static correlation diagnostics.)
We will illustrate some of the concepts using the W4-11 atomization energy benchmark , the DBH24 barrier heights benchmark , the HFREQ27 vibrational frequencies benchmark , and several recent benchmarks for noncovalent interactions such as the S66x8 set of biomolecule dimer potentials, , conformational energies of the proteinogenic amino acids, , and water clusters .