From owner-chemistry@ccl.net Mon Aug 6 14:12:00 2018 From: "Grigoriy Zhurko reg_zhurko[a]chemcraftprog.com" To: CCL Subject: CCL: PBE for inorganic chemistry Message-Id: <-53421-180806140912-731-GF/XkOYaB1chqkaUnM2DJg]-[server.ccl.net> X-Original-From: Grigoriy Zhurko Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=us-ascii Date: Mon, 6 Aug 2018 21:10:33 +0400 MIME-Version: 1.0 Sent to CCL by: Grigoriy Zhurko [reg_zhurko-,-chemcraftprog.com] Hello, >> It's not just bad form for a journal, it is not a legitimate justification for using a KS-DFT functional. A popularity contest is not science. Why would you cite a popularity contest instead of scientific evidence? Is a popularity contest more valuable than empirical evidence published in countless review articles one can easily google? The problems of the poll can be similar to the problems of opinion polls or exit polls ? it is well known that these polls sometimes improperly predict the results of elections (9 examples in Wikipedia). As far as I can see, the main problem of such polls is that it is impossible to provide a really representative sample (only a minority of people want to take part in such polls, and the opinion of these people can be quite different from the opinion of an average statistical voter). I suppose that an important advantage of the DFT poll compared to citation analysis is that the poll includes the ?hate? option in addition to the ?like? option (i.e., the poll gathers both ?likes? and ?dislikes? while a citation analysis is based on counting ?likes? and ?reposts? only). In the cited blog it was mentioned that the poll correctly reflected the rise of the wB97X-D functional, as well as the rise of the PBE0 functional. So, if double hybrid functionals are as good as some people in this group have mentioned, the poll will reflect this within next several years. The accuracy of calculations for some properties may be more dependent on the basis set, or a technique for treating relativistic effects (DK, ZORA or ECP), rather than the DFT functional chosen. For example, the accuracy of computations for compounds with H-bonds may depend significantly on whether diffuse functions have been used or not. If one computes a system with H-bonds via B3LYP/cc-pVTZ, while another one computes it via wB97X-D/aug-cc-pVTZ, they can come to a conclusion that the wB97X-D functional is better suited for computations involving H-bonds (as stated in the 2017 poll), while the influence of the basis set augmentation can become the main source of this conclusion. It is unclear whether 5d / 6p elements are referred to in the poll as ?Transition metals? / ?Main group elements? or ?Relativistic elements? or both. The term ?Relativistic elements? does not look consistent, since in some cases even for light elements the relativistic effects (e.g., zero-field splittings) are important. I?d like to suggest dividing all elements into 3 groups (main group (s and p elements), d elements, f elements), and distinguishing a separate property ? ?Relativistic effects?. It seems for me that the size of the poll PDF (6 pages) is very small and it should be supplemented with the description of the poll methodology. It would be nice to ensure that only experienced DFT users were allowed to vote in this poll. Moreover, I would be glad to know how many people voted ?like? and ?hate? for a particular property. Anyway, I?d like to wish Mr. Swart success with further development of his project. Grigoriy Zhurko https://chemcraftprog.com/