Reliable thermochemical modeling of reaction mechanisms requires hybrid DFT or higher-level models as well as inclusion of environment, conformer, thermal, etc. effects. Quantum embedding, such as the Huzinaga-equation and projection-based models employed here, can make such computations more accessible by focusing the use of the more costly models to the atoms involved in forming and breaking the bonds or residing in interacting surfaces, etc. Here, we further accelerate these embedding computations by combining local approximations in the atomic orbital and auxiliary function space of the hybrid DFT part with a new in-core density fitting implementation optimized for multilayer DFT. The so introduced local embedded subsystem (LESS) framework, when increasing the size of the environment, leads to asymptotically constant cost for the hybrid DFT layer. We demonstrate on reaction and activation energies of practical homogeneous, heterogeneous and enzymatic catalysis reactions that the intrinsic accuracy of hybrid DFT is retained, with a few tenths of a kcal/mol error including all (embedding and local) approximations. Compared to the same complete (density fitted) hybrid DFT reference, the LESS hybrid DFT-in-GGA runtimes are 30-90 times faster on systems with up to 171-238 atoms. Achieving energetics with practically hybrid DFT quality and GGA cost is a significant step toward predictive thermochemistry including reliable sampling, dynamics, etc. as well as quantum environment effects.
Csóka et al. (Sun,) studied this question.