W4‐17: A diverse and high‐confidence dataset of atomization energies for benchmarking high‐level electronic structure methods

Abstract

Atomization reactions are among the most challenging tests for electronic structure methods. We use the first-principles Weizmann-4 (W4) computational thermochemistry protocol to generate the W4-17 dataset of 200 total atomization energies (TAEs) with 3σ confidence intervals of 1 kJ mol . W4-17 is an extension of the earlier W4-11 dataset; it includes first- and second-row molecules and radicals with up to eight non-hydrogen atoms. These cover a broad spectrum of bonding situations and multireference character, and as such are an excellent benchmark for the parameterization and validation of highly accurate ab initio methods (e.g., CCSD(T) composite procedures) and double-hybrid density functional theory (DHDFT) methods. The W4-17 dataset contains two subsets (i) a non-multireference subset of 183 systems characterized by dynamical or moderate nondynamical correlation effects (denoted W4-17-nonMR) and (ii) a highly multireference subset of 17 systems (W4-17-MR). We use these databases to evaluate the performance of a wide range of CCSD(T) composite procedures (e.g., G4, G4(MP2), G4(MP2)-6X, ROG4(MP2)-6X, CBS-QB3, ROCBS-QB3, CBS-APNO, ccCA-PS3, W1, W2, W1-F12, W2-F12, W1X-1, and W2X) and DHDFT methods (e.g., B2-PLYP, B2GP-PLYP, B2K-PLYP, DSD-BLYP, DSD-PBEP86, PWPB95, ωB97X-2(LP), and ωB97X-2(TQZ)). © 2017 Wiley Periodicals, Inc.

Multistep Reaction Pathway for CO2 Reduction on Hydride‐Capped Si Nanosheets

“…All free energies in Figure are for the experimental conditions of 443 K and 1.8 atm; these free energies are calculated by adding WMS electronic energies to the zero‐point vibrational energies (ZPEs) and thermal contributions calculated by revM06/MG3 S; we employed a scale factor (developed by a previously published protocol) of 0.968 for the revM06/MG3 S ZPEs. The WMS method is a wave function method developed recently by us, and it shows an averaged mean unsigned error of 0.02 eV for the atomization energies in the W4‐17 database . All the revM06 calculations were carried out using a locally modified Gaussian program incorporating the MN‐GFM module .…”

ACCDB: A collection of chemistry databases for broad computational purposes

“…W4‐17 is an extension of two previous databases—namely W4‐08, and W4‐11—developed by Martin's and Karton's groups. W4‐17 includes 203 total atomization energies of 215 first‐ and second‐row molecules and radicals with up to eight non‐hydrogen atoms.…”

“…ACCDB includes five different databases from five different sources: MGCDB84, [21] GMTKN, [22,23] Minnesota, [24,25] DP284, [26,27] and W4-17. [12] The data from these sources cover primarily main group elements. Some transition metals (TMs) are present in the Minnesota database, but we significantly expand this number by introducing here a database of reactions involving first-, second-, and third-row TMs, as well as elements from the second transition.…”

“…The modern use of computational chemistry databases containing a large number of high‐level calculated data is not limited to the parametrization and assessment of semi‐empirical composite or DFT methods: other interesting applications include validation of high‐accuracy ab initio methods, benchmarking of software or hardware, and the application of modern data mining techniques to chemistry, such as artificial intelligence and machine learning …”

“…It is objectively true that more parameters in the xc functionals require more data in the training set (a statistical necessity to avoid over-training), but the availability of larger training sets does not necessarily translate into more parametrized functionals, nor into more parameters in a functional. [11] The modern use of computational chemistry databases containing a large number of high-level calculated data is not limited to the parametrization and assessment of semi-empirical composite or DFT methods: other interesting applications include validation of high-accuracy ab initio methods, [12] benchmarking of software or hardware, [13][14][15] and the application of modern data mining techniques to chemistry, such as artificial intelligence and machine learning. [16] In light of this increasing importance of large sets of calculated data with high accuracy in chemistry, we present here a collection of almost 45k unique reference data points (with 37k of them presented here for the first time), all at a level of theory significantly higher than DFT.…”