From owner-chemistry@ccl.net Tue Jul 9 14:17:00 2024 From: "Carlos Silva Lopez carlos.silva*_*uvigo.gal" To: CCL Subject: CCL:G: How to proceed after a RHF to UHF instability calculation Message-Id: <-55191-240709135421-27096-5bfWgs+2+qaXZkfLGqB96A##server.ccl.net> X-Original-From: Carlos Silva Lopez Content-Language: en-US Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=UTF-8; format=flowed Date: Tue, 9 Jul 2024 19:54:02 +0200 MIME-Version: 1.0 Sent to CCL by: Carlos Silva Lopez [carlos.silva%x%uvigo.gal] Hi Gary, The protocol that you describe seems solid under the circumstances that you have chosen to use DFT (a single determinant method) to describe systems that show strong non-dynamic correlation. You just need to ensure that the KS-wavefunction solution of the SCF is the energy minimum, which you are already doing by checking its stability. I have done this myself a number of times, many of them comparing results with respect to MCSCF and MRMP methods, and it seems that DFT does a surprisingly good job handling these situations, but it is a slippery slope, so be cautious if any results you get seem suspicious. You may want to check the contributions by Dieter Cremer on the topic if you want to have a better idea on how DFT's self interaction error (SIE) is helping you here ans which may be the pitfalls that you may encounter in the way. Just an addition to your protocol, in very ill behaving cases I often run an initial stable=opt job first and then I read the obtained KS wfn as the initial guess of my calculation. In such cases I optimize using the quadratically convergent SCF procedure, because it has a lower tendency to symmetrize the orbitals and fall back into the RFH solution. In nightmarish cases I have needed to create custom made optimization protocols that run an stable=opt job at every optimization step so that I ensure that initial broken symmetry solution survives at every SCF being run until geometry convergence. The latter is done via a NonStd input, but it is something you will want to avoid almost at any cost: https://gaussian.com/progdev/?tabid=4#NonStd_keyword I hope this is of help to you, Best, Carlos On 7/4/24 14:50, Gary Breton gbreton,berry.edu wrote: > Sent to CCL by: "Gary Breton" [gbreton(_)berry.edu] > Hi everyone, > > I'm seeking confirmation that I am performing some calculations correctly. I > am running some calculations (i.e., geometry optimizations and frequency > calculations at BP86-D3/6-311G(d) which has been used before for similar > calculations) using Gaussian 16 on twisted alkenes. I am also interested in > obtaining singlet/triplet energy gaps. > > Input for optimizing the geometries and obtaining the frequencies assuming a > singlet state: > > # opt freq 6-311g(d) bp86 empiricaldispersion=gd3bj > > I then tested the stability of the final wavefunction for each of the > optimized geometries: > > # stable 6-311g(d) bp86 empiricaldispersion=gd3bj > > For some of the more "twisted" geometries, I obtain this result: > > The wavefunction has an RHF -> UHF instability > > Of course, this isn't surprising since I would expect diradical character to > begin to mix in as the nominal "p-orbitals" become poorly aligned. > > To "fix" the problem, I re-optimized and obtained frequencies using the > following, again assigning as a singlet: > > # opt freq 6-311g(d) guess=mix ubp86 empiricaldispersion=gd3bj > > The result was a considerable change in geometry (the C=C bond lengthened). > Again, I would expect this result if diradical character was mixed in. The > wavefunction remains a "singlet" as desired (I am now attributing it as the > lowest energy singlet state). > > I then ran a stability check: > > # stable 6-311g(d) guess=mix empiricaldispersion=gd3bj ubp86 > > and obtain: > > The wavefunction is stable under the perturbations considered. > > So, I *think* I've done this correctly. Before I extend this to a series of > related compounds, I was hoping for some reassurance that this is the correct > approach. If not, some direction as to how to proceed would be very welcome! > > Best, > > Gary Breton > Berry College> https://urldefense.com/v3/__http://www.ccl.net/cgi-bin/ccl/send_ccl_message__;!!D9dNQwwGXtA!TiLYPKurtv5zgB7gtzGhuJBreZjYdICnfYCXX_pmE7CC12ydz24vjTOVY3E42dcUBTwSLNGgFaSbEyolG3Mm7Qfo$> https://urldefense.com/v3/__http://www.ccl.net/cgi-bin/ccl/send_ccl_message__;!!D9dNQwwGXtA!TiLYPKurtv5zgB7gtzGhuJBreZjYdICnfYCXX_pmE7CC12ydz24vjTOVY3E42dcUBTwSLNGgFaSbEyolG3Mm7Qfo$> https://urldefense.com/v3/__http://www.ccl.net/chemistry/sub_unsub.shtml__;!!D9dNQwwGXtA!TiLYPKurtv5zgB7gtzGhuJBreZjYdICnfYCXX_pmE7CC12ydz24vjTOVY3E42dcUBTwSLNGgFaSbEyolGz2tnpAJ$ > > Before posting, check wait time at: https://urldefense.com/v3/__http://www.ccl.net__;!!D9dNQwwGXtA!TiLYPKurtv5zgB7gtzGhuJBreZjYdICnfYCXX_pmE7CC12ydz24vjTOVY3E42dcUBTwSLNGgFaSbEyolG-bQsSlx$ > > Job: https://urldefense.com/v3/__http://www.ccl.net/jobs__;!!D9dNQwwGXtA!TiLYPKurtv5zgB7gtzGhuJBreZjYdICnfYCXX_pmE7CC12ydz24vjTOVY3E42dcUBTwSLNGgFaSbEyolGzdzgVKM$ > Conferences: https://urldefense.com/v3/__http://server.ccl.net/chemistry/announcements/conferences/__;!!D9dNQwwGXtA!TiLYPKurtv5zgB7gtzGhuJBreZjYdICnfYCXX_pmE7CC12ydz24vjTOVY3E42dcUBTwSLNGgFaSbEyolG6CpEbvy$ > > Search Messages: https://urldefense.com/v3/__http://www.ccl.net/chemistry/searchccl/index.shtml__;!!D9dNQwwGXtA!TiLYPKurtv5zgB7gtzGhuJBreZjYdICnfYCXX_pmE7CC12ydz24vjTOVY3E42dcUBTwSLNGgFaSbEyolGyna7_eb$> https://urldefense.com/v3/__http://www.ccl.net/spammers.txt__;!!D9dNQwwGXtA!TiLYPKurtv5zgB7gtzGhuJBreZjYdICnfYCXX_pmE7CC12ydz24vjTOVY3E42dcUBTwSLNGgFaSbEyolG-i4jw7A$ > > RTFI: https://urldefense.com/v3/__http://www.ccl.net/chemistry/aboutccl/instructions/__;!!D9dNQwwGXtA!TiLYPKurtv5zgB7gtzGhuJBreZjYdICnfYCXX_pmE7CC12ydz24vjTOVY3E42dcUBTwSLNGgFaSbEyolGwDyNDge$ > > -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Dr. Carlos Silva Lopez Universidad de Vigo Phone : (+34) 986812632 Departmento de Quimica Organica Fax : (+34) 986812262 Campus Universitario e-mail: carlos.silva*|*uvigo.es 36310 Vigo, Spain http://webs.uvigo.es/csilval ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~