High-level ab initio energy divergences between theoretical optimized and experimental geometries

High-level ab initio, QCISD(T) / 6-311 + G(3 df, 2p), energy from the least accurate theoretical bond distance was calculated and compared with that from the experimental geometry on each of the thirty four diatomic molecules having a relative error (r(opt.) - r(exp.))/r(exp.) of theoretical bond length greater than 2%. These molecules were chosen from all of the sixty first- and second-row diatomic inorganic species collected in the 77th CRC Handbook of Chemistry and Physics involved in our previous comparison of systematic geometry optimizations. It was found, unexpectedly, that QCISD(T) / 6-311 + G(3 df, 2p) calculations did not result in the lowest energy with the experimental bond distance in a few cases, which implies that QCISD(T) / 6-311 + G(3 df, 2p) method has some deficiencies in the point of view of "high-level ab initio". Most of the QCISD (T) / 6-311 + G(3 df, 2p) energy divergences between the two different geometries were less than 4.2 kJ . mol(-1) with six exceptions with BN, CN, CP, N-2, O-2 and SiN molecules and especially on CN radical, which resulted in an error as large as 15.8 kJ . mol(-1) at the MP2 (full) / 6-311G(2 d, p) geometry. For CN, CP and SiN radicals, the energy errors were also from the poor geometry of MP2 optimizations. It was also found that the larger the basis sets used the poorer the geometries were for these three radicals at MP2 level. For the ground state of BN((3)Pi) radical, the experimental bond length 128. 1 pm maybe in doubt and has to be re-examined. For N-2 and O-2, MP2(full) gradient optimization at the basis sets of valence triple-zeta, 6-311G(d, p), improved the geometries as well as the high-level QCISD(T) energies compared with what at the valence double-zeta, 6-310(d, p), basis sets.