CCL: Case Studies of QM Computational Chemistry in Reactivity

 Sent to CCL by: Susi Lehtola [susi.lehtola*]
 On 09/10/2015 09:58 AM, Tom Albright talbright1234-$ wrote:
 Dear Susi - you have completely missed my point. So I do a
 calculation of the transition state for some reaction at the
 aug-cc-pCVTZ level with CCSD(T) or some double functional with
 dispersion and get a barrier of 40.35 kcal/mol. Great, so the fact
 that this reaction does not experimentally occur is in perfect
 agreement with my theoretical results. So this is the end of the
 story, right? Wrong! The point is why is the barrier so large - the
 conclusion should not be that one is reasonably assured that the
 numerology matches what we know is true (ie experiment).
And so have you. Sure, I'd trust the value 40.35 kcal/mol given by a CCSD(T)/aug-cc-pCVTZ calculation to be sufficiently close to the true result (assuming the system is close to single reference).
But, this does not say at all whether B3LYP/6-31G* would also give you a reasonably accurate answer; it might be anything from -60 to 100 kcal/mol [exaggeration for sake of argument]. As has already been said, a lot of the success of the B3LYP/6-31G* model relies on fortunate cancellations of errors between the small basis set and the B3LYP functional, but this may not be the case in all systems, or especially all properties.
Now, if you know that B3LYP/6-31G* is good enough to yield accurate results for your system and property (compared to higher level ab initio data[*]), then all is fine. However, the past successes of B3LYP/6-31G* have made a good part of the applications community blind to all of the advances made in the past 20+ years. Depending on the problem, one might get much more accurate results with only a trivial amount of more resources.
 [*] Straight comparison of a low-level chemical model to experiment may
     lead you astray, since the experimental data might not have been
     analyzed correctly, or the theoretical values ignore some relevant
     chemistry (say, zero point or solvation effects).
 Mr. Susi Lehtola, PhD             Chemist Postdoctoral Fellow
 susi.lehtola!A!   Lawrence Berkeley National Laboratory  USA