A Study On The Reliability Of Quqntum Chemistry For Calculating Molecular Structures

Generating reliable structures and energies of a molecular system has been a central role or is the main use of computational chemistry. It is realized that the reliability of optimization of molecular geometry is the basic guarantee of the reliability of all the accurate calculations. The exploration based on unreliable method is meaningless. There are many theoretical methods about molecular geometrical optimization, but so far it is not sure which kind of theoretical methods used to calculate the structures and properties are more reliable and applicable. For discussing the reliability of theoretical methods, some systematic comparisons must be used. The evaluation of the reliability of a quantum chemical method needs a comparison of theoretically computed structures with the corresponding experimental ones.A molecular structure is closely related to its energies and other molecular properties. The optimization of its geometry to attain minimum potential energy is prerequisite for the calculation of any property of a molecule (finding the equilibrium structure of the molecule). There were a very large number of researches which compute the geometries or structures in the past. However, the number of molecules (Max=184), the types of elements, the common molecules before the fourth period, and the number of computational methods and basis sets that have been studied in these references are limited. So the conclusion and the application scope is relatively limited.This paper embarks on the reliability of the quantum chemical methods for computing molecular structures. The reliability of the quantum chemical methods computing the structures is studied systematically and comprehensively by Gaussian Software. In this study, the calculation of more than one thousand molecules is carried out by more than one hundred computational methods/basis sets. At the same time, these results are compared with more accurate experimental parameters. In this paper, fifteen methods/seven basis sets (or 105 methods) are used,MP2,MP2(Full),B3LYP,B3PW91,B3P86,B98,LSDA,XaPL,VSXC,HCTH,M06,HFS,B2PLYP or mPW2PLYPD/6-31G*,6-31G**,6-311G**,6-31+G*,6-311++G**,cc-pVDZ or aug-cc-pVDZ. All the computations are carried out by using Gaussian 03 and/or Gaussian 09 programs. It is found that there are 305 molecules, more than 20% of all the computed molecules (1353), whose computed structures are unusual or problematic (i.e. more than 40% of the 105 methods show a relative errors larger than 0.04A for a bond length or 4°for a bond angle), and there exist 140 (10.35%) molecules whose bond length(s) or bond angle(s) fail to be calculated (i.e., there are at least 84 methods whose deviations are larger than 0.04A or 4°). When the mean absolute deviations (MADs) of these methods are calculated with these problematic molecules, the MADs of bond length are between 0.026A (B3P86/6-311++G**) and 0.050A (HFS/aug-cc-pVDZ), and the MADs of bond angle are between 2.3°(B3P86/aug-cc-pVDZ, B98/cc-pVDZ, B98/aug-cc-pVDZ, or B3PW91/aug-cc-pVDZ) and 3.3°(HFS/6-31+G*). The results of HFS, VSXC, XaPL, B3LYP, LSDA, B2PLYP, B98, and HCTH methods are comparatively unreliable. Then the calculated structures are not always reliable or accurate.