Henry Rzepa's Blog

This are just a few insights I have got from some of the talks I attended. As usual, this does not represent a report on the WATOC congress itself, but simply some aspects that caught my personal eye.

1. Frank Neese talked about his Bubblepole approximation for large molecules.[cite]10.1021/acs.jpca.4c07415[/cite] And he was not kidding – large. Lets say a DFT calculation at the Def2-TZVPP basis set level (often the level used in this blog). Thus Crambin + 500H₂O, which is 2142 atoms can not only be done at this basis set level (33,562 basis functions) but at the astonishing Def2-QZVPP level (rarely attempted here!) with 86,667 basis functions. But that is not the largest – he has also done unhydrated Crambin octamer (5132 atoms) with 116,904 basis functions using the Bubblepole method. Currently this method appears only in his ORCA code – and if I understood correctly they are still working on first and second derivatives. So it will be a little while longer before e.g. reaction transition states for such sizes appear, but probably not that long!
2. Martin Head Gordon is responsible for the highly regarded ωB97 set of DFT functionals (again used throughout this blog). Until now, the most recent of these, ωB97M(2) from 2019[cite]10.1063/1.5025226[/cite] had represented a significant advance in accuracy (let’s say reaction barrier heights) over the previous generations, this having a mean error of ~0.9 kcal/mol compared to 2-3 kcal/mol for earlier generations. At the conference he introduced a “Carefully Optimised and Appropriately Constrained Hybrid” or COACH functional. He introduced 17 constraints or exact conditions that an ideal functional should have and explained that COACH satisfied 12 of these (another relatively recent functional, SCAN satisfies all 17[cite]10.1021/acs.jpclett.0c02405[/cite]). Earlier functionals satisfy ~6 or less. For 7 selected properties, including barrier heights, the mean errors are around ½ to ⅓ of earlier functionals such as the veritable B3LYP+D4 dispersion. His concluding remarks suggested that DFT as such is nearing the ultimate limit of general purpose accuracy achievable by such procedures. I hope to be trying out e.g. COACH here in the next year or so.
3. Fritz Schaefer “threw the kitchen sink” at the small tetra-atomic fulminic acid, or HCNO, to try to answer the simple question – is it bent or linear?[cite]10.1021/jacs.4c13823[/cite] At the CBS (complete basis set) limit and the CCSDTQ(P) level of coupled cluster theory (wow!), the answer converges to the conclusion that it is linear! This level cannot be that far off an exact solution of the Schroedinger equation – and it agrees with experiment!
4. Oh, a general observation, machine learning permeates the entire congress.

This entry was posted on Wednesday, June 25th, 2025 at 12:32 pm and is filed under Interesting chemistry. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.