Two variations of G2(MP2) theory which employ smaller basis sets in ev
aluating the quadratic configuration interaction [QCISD(T)] component
of the energy are presented. The first, G2(MP2,SVP), uses the split-va
lence plus polarization (SVP) 6-31G(d) basis, while the second, G2(MP2
,SV), uses the split-valence (SV) 6-31G basis. The methods are evaluat
ed on the basis of results for the set of 125 systems used for testing
G2 theory. The mean absolute deviation of G2(MP2,SVP) results from ex
perimental values is 1.63 kcal mol(-1) compared with 1.58 and 1.21 kca
l mol(-1) for G2(MP2) and G2, respectively. The G2(MP2,SVP) method thu
s provides results which are generally very similar in quality to thos
e obtained from G2(MP2) but at considerably reduced computational expe
nse. On the other hand, the mean absolute deviation of G2(MP2,SV) resu
lts from experiment is substantially larger (2.13 kcal mol(-1)). The G
2(MP2,SV) method exceeds the 2 kcal mol(-1) target accuracy of G2 theo
ry for an unacceptably large number of comparisons. (C) 1996 American
Institute of Physics.