From owner-chemistry@ccl.net Tue Dec 12 06:43:09 1995 Received: from bedrock.ccl.net for owner-chemistry@ccl.net by www.ccl.net (8.6.10/950822.1) id GAA08927; Tue, 12 Dec 1995 06:05:38 -0500 Received: from ds for uviqplco@cesga.es by bedrock.ccl.net (8.7.1/950822.1) id GAA29314; Tue, 12 Dec 1995 06:05:10 -0500 (EST) Received: by ds (4.1/1.34) id AA02073; Tue, 12 Dec 95 12:03:04 +0100 Date: Tue, 12 Dec 1995 12:03:04 +0100 (MET) From: Luis Carballeira Oca~na To: Computational Chemistry List Subject: Help to recover lost data Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Hello netters, During a maintenance procedure on an IBM RISC/6000 530H working with AIX 3.2.5 a problem occurred and both the logical volume (/dev/hd1) and its associated /home filesystem have been removed. So, we have lost data of ab initio calculations that we haven't on a backup tape. Does anybody know any procedure to recover these data? Thanks for any help. Luis Carballeira. Quimica Computacional. Lab. Quimica Fisica. Facultad de Ciencias. Apdo. 874. Vigo. Spain. e-mail: uviqplco@ds.cesga.es Tel 34 86 812306 FAX 34 86 812382 From uviqplco@cesga.es Tue Dec 12 07:22:40 1995 Received: from ds for uviqplco@cesga.es by www.ccl.net (8.6.10/950822.1) id HAA09474; Tue, 12 Dec 1995 07:05:37 -0500 Received: by ds (4.1/1.34) id AA02889; Tue, 12 Dec 95 13:03:00 +0100 Date: Tue, 12 Dec 1995 13:02:59 +0100 (MET) From: Luis Carballeira Oca~na To: Computational Chemistry List Subject: Help to recover lost data Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Hello netters, During a maintenance procedure on an IBM RISC/6000 530H working with AIX 3.2.5 a problem occurred and both the logical volume (/dev/hd1) and its associated /home filesystem have been removed. So, we have lost data of ab initio calculations that we haven't on a backup tape. Does anybody know any procedure to recover these data? Thanks for any help. Luis Carballeira. Quimica Computacional. Lab. Quimica Fisica. Facultad de Ciencias. Apdo. 874. Vigo. Spain. e-mail: uviqplco@ds.cesga.es Tel 34 86 812306 FAX 34 86 812382 From gonda@nobel.upjs.sk Tue Dec 12 08:13:09 1995 Received: from nobel.upjs.sk for gonda@nobel.upjs.sk by www.ccl.net (8.6.10/950822.1) id HAA09942; Tue, 12 Dec 1995 07:58:07 -0500 From: Received: (from gonda@localhost) by nobel.upjs.sk (8.6.11/8.6.9) id NAA05032; Tue, 12 Dec 1995 13:30:21 +0100 Date: Tue, 12 Dec 1995 13:30:20 +0100 (GMT+0100) To: "David J. States" cc: chemistry@www.ccl.net, states@ibc.WUStL.EDU Subject: Re: CCL:Call for Papers - Intelligent Systems for Molecular Biology '96 In-Reply-To: <30C76E28.30B@ibc.wustl.edu> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII subscribe From VARNAI@ch.bme.hu Tue Dec 12 13:28:13 1995 Received: from ns.bme.hu for VARNAI@ch.bme.hu by www.ccl.net (8.6.10/950822.1) id NAA17376; Tue, 12 Dec 1995 13:25:45 -0500 Received: from ch.bme.hu (ch.bme.hu [152.66.64.1]) by ns.bme.hu (8.6.12/8.6.12) with SMTP id TAA15508 for ; Tue, 12 Dec 1995 19:25:39 +0100 Received: from CH/TEMPQ by ch.bme.hu (Mercury 1.12); Tue, 12 Dec 95 19:24:41 +1100 Received: from TEMPQ by CH (Mercury 1.12); Tue, 12 Dec 95 19:24:24 +1100 From: "Varnai Peter" To: chemistry@www.ccl.net Date: Tue, 12 Dec 1995 19:24:15 GMT+100 Subject: summary of CAS questions X-Confirm-Reading-To: "Varnai Peter" X-pmrqc: 1 Priority: normal X-mailer: Pegasus Mail v3.22 Message-ID: <69FA130961@ch.bme.hu> Dear CClers, A little late, but I send You the summary. My thanks to those (Frank Jensen, Mathias Bock, Doug Fox) who sent me their comments. Yes, it is not an easy question how to select the active space of an e.g. transition state. In my view there are some methods, but none of them is always applicable: 1, visualization (what is the chemical problem) 2, orbitals weights in CI calculations 3, occupancy number of the orbitals. Important to check the orbitals of the structures around your initial structure to create a consistent active space! Be careful, sometimes occ. number or weights tell you chemically not important orbitals eg pi-orbitals in a sigmatropic shift (hard to answer whether it is a mistake to include those beyond the essential orbitals; whether CAS is ONLY for the non-dynamic part of the correlation energy etc). Gaussian 94 is indeed have a new routine which seems to converge better, but allow you to create nonsense as well. But, don't lose the chk file, hard to recover it! In case somebody has new ideas, those are always welcome! Here follows the original letter and the answers. Again, thank You! Peter Varnai ??????????????????????????????????????????????????????????????? I would need Your help to understand the practical tricks in CASSCF calculations. 1, One needs to define an active space. -At this point quantum chemists start to think as the conventional organic chemist,in localized orbitals. Here those orbitals needed -in view of a chemist- that are involved in a eg. reaction (forming/breaking orbitals). -A more exact answer would be that those orbitals needed which are important in "excitation" to give configurations that contribute to the total energy significantly. So: -according to the first approach one would visualize the MOs. Here he/she would realize that in MO theory the orbitals are spread all over the molecule, so the needed eg C-C bond will appear in at least 3 MOs. Which one should be used then? Let's use all (unfortunately, the valence MOs usually can't be all involved in the active space)? Let's localize the MOs with standard methods? Pick one random and what happens (the one is better than the other if using it the SCF energy is lower)? -according to the more "mathematical" approach let's calculate a CISD wavefunction [in the absence of the full CI :(] and involve the orbitals which are involved in large coefficient (say 0.03<) determinants? And what if the HF reference function was not a good approximation and some important determinants are missed? -If I am right CAS should be used as the correction to the non- dynamic electron correlation (through the the right geometry and wavefunction). What happens then when CISD tells you that an orbital energetically important though not needed in view of the reaction, description of the radical etc. it would be some parts of the dynamic correlation effect then?! 2, CASSCF in Gaussian 92 for me (and I guess for several others) stopped with the error message: rotation larger than 45 degree... If I understand well either a smaller step in wavefunction optimization would have been desirable or the active space was not well chosen and a non-selected orbital was crucial in the active space. Yet, I haven't got error message like this with the G94 routine, only warnings that large rotations are scaled and something like: I J= 12 26. I don't really understand the meaning of these numbers (i,if these are MO numberings, I have to say that eg MO- 12 was not selected to be in the active space, ii, in MOLCAS for example the user can define only the number and symmetry of the active orbitals, the exact MO not, how will then the routine pick the right orbital? iii, is it allowed to change orbitals of the active space during SCF). 3, In case one loses the chk file and would like to keep it for further use, is the following way is correct to create the same chk file? 1, Calculate the HF-MOs at the starting geometry once again and picking the same MOs 2, converging the CAS wavefunction as it was done in the previous calculation and 3, reading in the wavefunction (to have the same active space) and having the converged CAS-geometry (not to redo the whole previous calculation). If it is correct can anyone tell me why I did not reproduce the same stationary point, but locating an other one (higher in energy). (As I know this method is used when one tries to find the right active space in a supermolecule by isolating it to fragments). If You had the patience and read my letter through, and You can help me in using CAS smoothly: THANK YOU! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Varnai, one of the 'modern' ways of selecting the active MOs in a CASSCF is from the natural orbital occupation numbers, i.e. first do a MP2 or CISD and diagonalize the density matrix to giv ethe natural orbitals. Then select the orbitals which have occupation number 'significantly' different from 0 or 2. How far away depends on your problem (and hardware!) but typically orbitals with occ. >0.05 and < 1.95 must be included, and >0.03 and < 1.97 is desireable. Since CASSCF wave functions often are used for many different geometries (like a reaction), you need to do this at several points, and select a 'consistent' space, i.e. the smallest which include all the 'important' orbitals at each geometry. In some cases it will be difficult to make a consistent choice. So, as you can se from the 'quotes', CASSCF orbital selection is not easy. Frank !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Hallo Peter, your CASSCF questions were quite familar to me, unfortunetaly I am not able to answer them totally, maybe the specific G9x answers will come from help@gaussian.com if you havenot already done so. First of all, I should prefer to use several CI-methods to look for the most important configurations to be included. You can decide whether static correlation is included at a minimum of cost. In G92 CASSCF/6-31G gave nice results, even for open-shell molecules. The choice of orbitals was crucial. Sometimes I used sophisticated pre-runs to get starting orbitals. But this is no garanty to converge even from a converged ground state to reach convergence for an excited state. G94 will not care as much as G92, so don't get slippy ! Running small basis sets, gives an easy initial choice, but it happened not to give SCF-convergence, until another starting geometry was given. Good starting geometries were given by UHF/3-21G or RMP2/6-31G*. The open-shell UHF/ROHF gives you another riddle, as the UHF-functions don#t give reliable wave functions. The use of CISD seems necessary to annihilate spin contamination. Also try an RHF-guess first by changing charge and multiplicity. I hope a summary of answer will go to net. If not, I would appreciate a copy. Thanks . Mathias Brock !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Peter, The choice of active space is the most challenging part of getting a CASSCF right. WE have changed the code with G94 to make failures like "rotation larger than 45 degrees" go away but with an intellectual cost. The meaning of this error, as I think we have discussed, is to switch a pair of orbitals changing the active space. To avoid this G94 damps this rotation to keep it from switching and force a solution within the chosen active space. The reason behind not adapting MolCAS's behavior is that you most often want to study a particular state or reaction with a CAS wavefunction. This may well not be the lowest energy solution and if the active space is allowed to change you may well end up with the wrong state or with the wrong active space for the reaction of interest. For example if you are looking at formaldehyde loosing H2 the lowest energy solution is to include the pi orbitals but these play no part in proper dissociation to H2 + CO. The penalty for not allowing the switch is a poor initial guess may end up not describing the orbital you wanted or an ill chosen guess can include orbitals which don't interact with the orbitals that define/control the reaction/state of interest. Thus for G94 what you want to do is pick your active space as well as possible and then examine the result of a single point calculation. Make sure that there are not orbitals with double or zero occupancy. Use #P to get this printed out. If there are then the eigenvector for that state, key back to the list of configurations printed earlier, can let you see which orbitals are likely candidates either to be removed or switched. The goal is to make the solution match your state of interest not simply a lowest energy solution. Restarting can be a bit of a problem. First, you may well have moved to a part of the surface where the RHF solution is a lousy representation. Using STABLE=OPT to make sure that you get the best single determinant is worthwhile. Your plan of starting from the initial orbitals and then leaping to the final structure seems reasonable as well. There are changes to the orbitals as the optimization progresses. These changes may make the difference between the "excited" state you found and the state you optimized to. Is your case small enough that you can try the CASSCF(QC) option, requires writting AO integrals to disk, this may well be able to get past the problem. Doug Fox help@gaussian.com !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! From lal@chem.ufl.edu Tue Dec 12 16:28:16 1995 Received: from mailey for lal@chem.ufl.edu by www.ccl.net (8.6.10/950822.1) id QAA19421; Tue, 12 Dec 1995 16:25:23 -0500 Received: from homey. by mailey (SMI-8.6/4.11) id QAA03493; Tue, 12 Dec 1995 16:23:08 -0500 Received: by homey. (SMI-8.6/SMI-SVR4) id QAA00930; Tue, 12 Dec 1995 16:25:26 -0500 Date: Tue, 12 Dec 1995 16:25:25 -0500 (EST) From: lucian To: chemistry@www.ccl.net Subject: DELETE MY ADDRESS Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Dear CCL: Please remove me from your mailing list. Thanks. Lucian