Benchmarks and Reliable DFT Results for Spin Gaps of Small Ligand Fe(II) Complexes

In two(!) previous posts, I criticized the work of Eunji Sim et al. on the so-called HF-DFT. See here, and here. Guess what, it is finally published after 6 months of peer-review, see here.

So today, I am officially reviewing the paper (my first time, please be gentle!).

Disclaimer: this is my personal opinion. The authors, your paper is accepted and published.

First and foremost, did the authors read my blog? Maybe yes, because they changed the title of the paper from “Benchmarks and reliable DFT results for spin-crossover complexes”, which is BS, to a better one “Benchmarks and Reliable DFT Results for Spin Gaps of Small Ligand Fe(II) Complexes”.

Being not a native speaker, I am aware that I’m not good at academic writing (that’s why I started this blog for practising). But I have to admit, the paper is poorly written.

A translation fail (from Chinese)

In this paper, the authors studied the spin gap of some small Fe(II) transition metal complexes [Fe(NCH)6]2+, [Fe(NH3)6]2+, [Fe(H2O)6]2+, [Fe(CO)6]2+, as well as one larger Fe(II) complex Fe-porphyrin-NO. The authors found that “standard quantum chemical methods (semilocal DFT and CCSD(T)) fail badly for the energy difference between their high- and low-spin states. Density-corrected DFT is both significantly more accurate and reliable and yields a consistent prediction for the Fe−Porphyrin complex.“

The author used several functionals, including LDA (SVWN5), GGA (PBE, BP86, BLYP), mGGA (TPSS), and hybrid (TPSSh, B3LYP, PBE0) functionals, and they stated that “All DFT approximations overstabilize the LS state relative to the HS state in a very systematic way.“. Well, this is a big slap in the face of quite some theoreticians, who usually assume that GGA functionals overstabilize low spin and hybrid functionals with more exact exchange overstabilize high spin. The results of Sim et al. lead me to a conclusion that we should use more exact exchange, maybe Becke’s half-and-half?

Batman does not approve B3LYP

Since it tooks 200k to 400k core hours(!) on 512 Intel Knight’s Landing nodes to do QMC for these systems, I cannot check the number myself. Moreover, I’m not an expert in the field to judge their QMC numbers. Thus, I believe their reference QMC results, which clearly show that CCSD(T) sucks badly. I therefore keep my promise as written in the previous post and decide to quit science.

Well, not that fast!

The authors studied Fe(II)-porphyrin-NO and wrote: “Figure 6 clearly demonstrates the tremendous reduction in variation among approximate energies when the HF density is consistently used and the unambiguous prediction that the LS state is lower by about 0.8 eV, in contradiction to self-consistent B3LYP or PBE0.” Now all conventional DFT functionals overstabilize the LS state as compared to HF-DFT. What the hell? It doesn’t make any sense. The authors even did not bother to explain this. Furthermore, Radon et al. estimated the (not really correct) “experimental” value for this complex (0.14 eV). Their CASPT2 result is also around 0.15 eV. So not only CCSD(T) is extremely wrong, but also multireference methods are horribly bad. We should all quit science.