That’s is all it takes, a little tiny error to mess completely with a three months project. I’m working with a third row metal’s complex, trying to describe its decomposition through radiation with visible light. At the beginning it seemed like a fairly straightforward thing to compute with careful optimizations using the ORCA¹ program with DFT, a good basis set, the Zeroth-Order Regular Approximation² (ZORA), and little else.
However, after a while I started to find odd results. Very odd ones. Trying to fix my perceived error I made test with different basis sets and functional to no avail. Finally, the previous week my supervisor told me that I was using the incorrect ZORA. I was surprised, to say the least. I looked quickly into the ORCA Inputs site and found this:
Important: Due to the use of the one-center approximation (see manual) in geometry optimizations, energies from a ZORA/DKH2 geometry optimization are not the same as those from a single-point calculation. Be careful not to mix energies from optimization and single-point relativistic jobs
So, yes, I was mixing numbers coming from two different sources, the single point energies are calculated using the normal ZORA, which doesn’t allow us to compute the gradient, while the final energy in an optimization run corresponds to the scaled ZORA variant³, where calculation of analytical gradient is very cheap. Thus, in conclusion, we have to do first an optimization and then get the single-point energy in a different run. It is quite simple, yes, sadly simple. With the new numbers the history is pretty similar to what we expect from chemical intuition. Let’s keep working and maybe we can have a draft for an article by the end of the spring.
1. F. Neese. The ORCA program system, Wiley Interdiscip. Rev.: Comput. Mol. Sci., 2, 73–78, 2012.
2. E. van Lenthe, E. J. Baerends, and J. G. Snijders. The Journal of Chemical Physics, 99(6):4597–4610, 1993.
3. E. van Lenthe, E. J. Baerends, and J. G. Snijders. J. Chem. Phys., 101:9783–9792, 1994.