From owner-chemistry@ccl.net Wed Apr 10 11:54:00 2019 From: "Susi Lehtola susi.lehtola],[alumni.helsinki.fi" To: CCL Subject: CCL:G: G09-Excessive mixing of frozen core and valence orbitals Message-Id: <-53675-190410045039-4732-RT4871EC6WCohx0qHJ0UNQ++server.ccl.net> X-Original-From: Susi Lehtola Content-Language: en-US Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=utf-8; format=flowed Date: Wed, 10 Apr 2019 11:50:23 +0300 MIME-Version: 1.0 Sent to CCL by: Susi Lehtola [susi.lehtola*alumni.helsinki.fi] On 4/9/19 1:40 PM, Christoph Riplinger riplinger%faccts.de wrote: > Thanks to Susi for the detailed reply and explanation! > > I would like to additionally mention our recent study > > https://pubs.rsc.org/en/content/articlehtml/2017/cp/c7cp00836h (clip) > As a result of this study we now have more conservative frozen core > definitions in ORCA (which removes the above-mentioned large > deviations). Also, if "physical core" orbitals are found in the > valence region, while "physical valence" orbitals are found in the > core region, the corresponding MO pairs are swapped automatically for > the subsequent Post-SCF treatment. Dear Christoph, thanks for the citation. It appears you are following the recipe of Rassolov et al by using Mulliken charge analysis to classify MOs as core or valence. However, this method appears to require one to classify the functions in the basis set beforehand into core or valence, as was also done by Rassolov et al for small Pople-type basis sets. Unfortunately, this takes some work, and becomes complicated if one is not using "standard" basis sets - especially if they are fully uncontracted. A simple solution would be to employ a reference set of orbitals from atomic calculations, which are usually needed anyhow to start off SCF calculations: the Superposition of Atomic Densities (SAD) guess is the default in most quantum chemistry codes, including ORCA. Indeed, such a solution has been reported by Austin and coworkers in Theor. Chem. Acc. 107, 180 (2002). (Because Austin et al used Gaussian, they call SAD the Harris guess.) Such a procedure abstracts away the need for Mulliken charge analysis, as it only relies on projections onto the atomic orbitals, which are well-defined even for fully uncontracted basis sets or real-space methods. Susi PS. Talking about SCF guesses, I can't resist to mention my own recent work from last October https://arxiv.org/abs/1810.11659 which was recently published in JCTC https://pubs.acs.org/doi/10.1021/acs.jctc.8b01089 Surprisingly, even though the SAD guess has been around for a long time, and even though use of atomic potentials for estimating molecular orbitals was suggested already in the 1960s, it appears that people had either forgotten about the idea, or thought that it was already common knowledge / that it had already been published. In fact, a superposition of pretabulated atomic potentials can give you a better guess than the superposition of neutral atom densities, and the machinery developed for quadrature in density functional theory makes it possible to evaluate the matrix elements of (a superposition of) pretabulated potentials accurately. Furthermore, if one approximates the potentials obtained from atomic calculations with a Gaussian expansion, the necessary integrals can be evaluated analytically in Gaussian basis sets. Indeed, I found out only after my paper was published that this is exactly what the Dirac code has been doing since 2016 (Lucas Visscher implemented the approach). -- ------------------------------------------------------------------ Mr. Susi Lehtola, PhD Junior Fellow, Adjunct Professor susi.lehtola]![alumni.helsinki.fi University of Helsinki http://susilehtola.github.io/ Finland ------------------------------------------------------------------ Susi Lehtola, dosentti, FT tutkijatohtori susi.lehtola]![alumni.helsinki.fi Helsingin yliopisto http://susilehtola.github.io/ ------------------------------------------------------------------