In Quantum chemical calculations, it is not always the case that the more complex (correlated) the method or the basis set the more accurate the results.

For example, it's sometimes found that MP2 results better fit the experimental results than that of CI calculation.

In other instances diffuse functions negatively impacts the accuracy of the calculation.

Are there any general rules for the choice of a method and the basis sets?

What is the degree of accuracy of a calculation with no experimental reference ?


The experimental equivalent of your question would be: what kind and level of impurity am I going to accept in my experiment?

In terms of basis sets: in principle you keep increasing the complexity of your basis set until you reach convergence in terms of electronic energy. What you consider to be converged depends of course on your preferences.

With regard to the choice of method. Again, this depends on what you consider to be accurate enough. In heterogeneous catalysis, typically density functional theory calculations are performed using a PBE functional. This is accurate to about 5-15 kJ/mol which is fine in this field. This type of accuracy would be completely unacceptable in other fields. If high accuracy is required, people typically use post-Hartree-Fock methods such as MP2 or higher or CI.

My recommendation in picking a method and a basis set is the following:

  1. Decide what you consider accurate enough.
  2. Look into the literature if there are calculations being performed on the same or equivalent type of system and analyse their results.
  3. Experiment using different methods on a model system. Look how the energy converges with respect to basis set and methodology. Pick the method that you think is best, both in terms of computational speed and accuracy. (it is OK to pick a couple of methods)
  4. Perform the bulk of your calculations using this method.
  5. If you are going to report your calculation. Clearly explain the reasoning why you picked a certain method. The potential readers of your report really appreciate to hear your motivation.
  • $\begingroup$ thank you ifilot .... so in case of no experimental reference i have to choose an equivalent computed reference $\endgroup$
    – M.ghorab
    Dec 29 '15 at 9:41
  • 1
    $\begingroup$ Exactly. Just critically read through their results. Try to reproduce a couple of these results and then try to extrapolate their method to your system. $\endgroup$
    – Ivo Filot
    Dec 29 '15 at 10:38
  • 1
    $\begingroup$ I don't agree with the comparison to the impurity level. Less impurities is always better while in computational chemistry that kind of order might not exist, especially if you compare different density functionals. $\endgroup$
    – DSVA
    May 15 '17 at 20:35
  • 1
    $\begingroup$ To add onto this answer, you can always consider additional levels of theory (e.g. a few density functionals, for instance). Nobody would complain about providing more data. So, if you do this, feel free to publish whatever you run in the SI (explaining any anomalies in the actual text). I also would like to specifically agree with Point 5. More often than not, there is no justification for the level of theory used even if the authors had some underlying reason. It is extremely helpful for the reader if you briefly support your decision, even if it's just a matter of citing similar papers. $\endgroup$
    – Argon
    Aug 3 '17 at 23:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.