Breaking a bottleneck: Accurate extrapolation to "gold standard" CCSD(T) energies for large open shell organic radicals at reduced computational cost.

Abstract

Open Shell organic radicals are principal species involved in many diverse areas such as combustion, photochemistry, and polymer chemistry. Computational studies of such species with an accurate method like coupled-cluster with single and double and perturbative triple (CCSD(T)) may be restricted to systems of modest size due to the steep computational scaling of the method. Herein, we assess the accuracy of extrapolated CCSD(T) energies determined using the connectivity-based hierarchy (CBH) method on medium to large sized radicals. In our method, an MP2 calculation on the target radical is coupled with CCSD(T) energies of fragments determined uniquely by our hierarchy to perform accurate extrapolations. A careful assessment is done with a robust CBH-rad49 test set comprising of 49 diverse cyclic and acyclic radicals with a variety of functional groups. We demonstrate that the extrapolation method with CBH-2 or CBH-3 is sufficient to obtain sub-kcal accuracy. ROMP2 and PMP2 calculations with both Pople-style and Dunning-style basis-sets resulted in mean absolute errors for CCSD(T) extrapolation (full CCSD(T)-extrapolated CCSD(T)) within 0.5 kcal/mol. Further speedup for such CCSD(T) extrapolations are obtained with ROHF-based RI-MP2 calculations. Challenging systems with (a) high ring strain, (b) delocalized character, and (c) spin contamination are identified and analyzed in detail. Finally, we apply our extrapolation method on 10 larger radicals containing 10-15 heavy atoms, where accurate CCSD(T) energies are obtained at a fractional cost of full CCSD(T) calculations.