(Georgia Institute of Technology, 2004-11)
Abrams, Micah L.; Sherrill, C. David
The C₂ molecule exhibits unusual bonding and several low-lying excited electronic states, making the prediction of its potential energy curves a challenging test for quantum chemical methods. We report full configuration interaction results for the X 1Σg+, B 1Δg, and B′ 1Σg+ states of C₂, which exactly solve the electronic Schrödinger equation within the space spanned by a 6-31G∗ basis set. Within the D2h subgroup used by most electronic structure programs, these states all have the same symmetry (1Ag), and all three states become energetically close for interatomic distances beyond 1.5 Å. The quality of several single-reference ab initio methods is assessed by comparison to the benchmark results. Unfortunately, even coupled-cluster theory through perturbative triples using an unrestricted Hartree–Fock reference exhibits large nonparallelity errors (>20 kcal mol−1) for the ground state. The excited states are not accurately modeled by any commonly used single-reference method, nor by configuration interaction including full quadruple substitutions. The present benchmarks will be helpful in assessing theoretical methods designed to break bonds in ground and excited electronic states.
(Georgia Institute of Technology, 2003-01)
Abrams, Micah L.; Sherrill, C. David
We compare several standard polarized double-zeta basis sets for use in full configuration interaction benchmark computations. The 6-31G**, DZP, cc-pVDZ, and Widmark–Malmqvist– Roos atomic natural orbital ~ANO! basis sets are assessed on the basis of their ability to provide accurate full configuration interaction spectroscopic constants for several small molecules. Even though highly correlated methods work best with larger basis sets, predicted spectroscopic constants are in good agreement with experiment; bond lengths and harmonic vibrational frequencies have average absolute errors no larger than 0.017 Å and 1.6%, respectively, for all but the ANO basis. For the molecules considered, 6-31G** gives the smallest average errors, while theANO basis set gives the largest. The use of variationally optimized basis sets and natural orbitals are also explored for improved benchmarking. Although optimized basis sets do not always improve predictions of molecular properties, taking a DZP-sized subset of the natural orbitals from a singles and doubles configuration interaction computation in a larger basis significantly improves results.