It is known in computational chemistry, when computing e.g. electronic ground states energy and geometric paramters (bond lengths, angles, torsions etc.) of a given molecule using an ab initio or DFT methods with some basis sets, there are situations where a certain combination of method and basis will give good agreement with the experiment due to error contributions in that method/basis combinations having opposite sign of similar magnitude and thus cancel out.
This phenomenon is common in e.g. MP2 calculations of small molecules, where the electronic energy is usually in good agreement with experiment due to favourable error cancellations
However, not all calculations can benefit from error cancellations, and the presence of them will often mean they are quite sensitive to the nature of the molecule and the method/basis used to model them, making said method have less say on its reliability on similar molecules (e.g. with one ligand being replaced or rearranged in configuration)
Therefore, I wonder whether there are general guidelines in terms of what sequence of calculations to be carried out to test whether a good result is due to favourable error cancellation, or that the method faithfully reproduce the experimental result.
In particular, ideally if the error cancellation can be traced all the way down to the details and implementation of the theoretical method and basis set, such as the error cancellation in some molecule is due to the dipole interaction of a oxygen functional group with one of the aromatics, or that e.g. a polarisation function on oxygen atoms in that particular basis set contribute a term that is of opposite sign to another term in gaussian orbitals that model the delocalisation of electrons in an aromatic system.
Any good references that explores error hunting in computational chemistry and attempt at localising their sources?
(preferably in inorganic systems)