I am running one system containing two gold atoms in vacuum using mp2, mp3 method with changing the distance between two gold atoms. I tried many different basis set ( lanl2dz, lanl2mb, aug-cc-pvdz-pp, aug-cc-pvtz-pp, aug-cc-pvqz-pp). I got the lowest energy at 2.6 angstrom. After that upto 5.75-6.0 angstrom distance the energy is raising. Upto here it looks like a normal van der Waals energy plot. But after crossing 6 angstrom, again the total energy of the system is decreasing. But it should not be decreased. In case 15 angstrom distance separation b/w gold atoms, the energy is even lower than 2.6 angstrom distance b/w those atoms. According to the normal vdw energy plot, the energy graph should be flat after a certain distance. I can not understand what is the source of the problem here. I didn't observe this kind of problem in Hartree-Fock calculation.
Electron correlation methods you use are doomed to fail for large distances. In their derivation it is assumed that main features of a molecule are well described by a single reference (i.e. one determinant) and that contributions to the wave function from other determinants (which are singly-, doubly- etc. excited with respect to this reference) are small. This assumption is no longer valid for large distances. The models try to reproduce growing coefficients for determinants, which are formally n-excited, but become close in energy with the reference, and depending on the method used for the energy calculation they show different anomalies. Note that e.g. for MPn or CC we do not have the variational principle, so nothing prevents an ill-defined problem from going down with the energy. You can look at numerous papers dealing with problems of single-reference methods with a description of bond-breaking, e.g. this review (see fig.25 and fig.26), M. Musial, R.Bartlett, Rev.Mod.Phys.,70,291(2007) dealing with CC theory.
Electron correlation methods you use are doomed to fail for large distances. In their derivation it is assumed that main features of a molecule are well described by a single reference (i.e. one determinant) and that contributions to the wave function from other determinants (which are singly-, doubly- etc. excited with respect to this reference) are small. This assumption is no longer valid for large distances. The models try to reproduce growing coefficients for determinants, which are formally n-excited, but become close in energy with the reference, and depending on the method used for the energy calculation they show different anomalies. Note that e.g. for MPn or CC we do not have the variational principle, so nothing prevents an ill-defined problem from going down with the energy. You can look at numerous papers dealing with problems of single-reference methods with a description of bond-breaking, e.g. this review (see fig.25 and fig.26), M. Musial, R.Bartlett, Rev.Mod.Phys.,70,291(2007) dealing with CC theory.
Use multi-reference methods such as MRCC, MRCI or MRPT to get correct qualitative description during bond-breaking. You may try EOM-SF-CC as well. If you are not keen about quantitative accuracy, then SF-DFT or TD-DFT with suitable exchange-correlation functional (taking care of long-range interactions properly) may work in some cases.
- Badges
- Science topic
- Similar topics
- Chemistry
- Theoretical Chemistry