From m10!frisch@uunet.UU.NET Thu Oct 15 17:28:03 1992 Date: Thu, 15 Oct 92 21:28:03 EDT From: m10!frisch@uunet.UU.NET (Michael Frisch) Subject: Re: Spin density calculations for benzene (fwd) To: chemistry@ccl.net On Oct 14 3:32pm, Christopher Cramer wrote: > expense of the empty orbital. Mike Frisch may weigh in to tell us how the > resultant wavefunction from G92 may be other than D6h if that symmetry is > imposed from the start, but this is one of those cases where symmetry would > clearly produce an unstable wavefunction in any case. and Dave Bernholdt wrote: I'm not Mike Frisch, but I probably weigh more :-) Gaussian, and most other quantum chemistry codes are limited to abelian point groups for practical reasons. The largest abelian subgroup of D6h is D2h, so that's what the calculation would have been done in. The HF wavefunction is forced to maintain the symmetry imposed by the _calculation_, but in this case it is not the symmetry of the nuclear framework. Therefore, the wavefunction (and hence the spin densities) will satisfy D2h symmetry, but not necessarily D6h. Some codes (I don't remember of Gaussian does or not) will analyze the symmetry of the wavefunction after an SCF calculation and compare it to the symmetry of the nuclear framework. This can be a very useful thing to check if you're in the habit of studying "challenging" systems. When studying Jahn-Teller systems, be careful about which potential surface you're on too. I'll try not to be too ponderous (just verbose) :-), but packages vary in how they handle symmetry, and I'd like to clear up what Gaussian does and what one can conclude about some of these cases. Gaussian does use abelian symmetry to speed up the calculation, but it does not apply any symmetry constraints on the orbitals. They normally come out symmetric because that represents the lowest energy solution. There are two ways of using symmetry to reduce the work of forming the Fock matrix (and computing the integrals, for direct SCF). In the method used by default, the Fock matrix is formed from the symmetry-unique integrals and then symmetrized. This is only valid if the density matrix has the molecular symmetry, and this is checked for by the program. The other method (using the operators of the point group to replicate the unique integrals while forming the Fock matrix) is available as an option, and does not impose any symmetry constraints on the wavefunction. Since Gaussian does not impose any symmetry on the solution, if the wavefunction/density comes out D2h for nuclear symmetry D6h, that IS telling you something about the physics of the molecule. Sometimes the wavefunction and density have the full molecular symmetry but there is a lower (broken) symmetry solution. In this case, the high symmetry wavefunction is a saddle point in orbital (Fock) space. This can be tested for via a stability calculation -- the stability matrix will have a negative eigenvalue corresponding to mixing occupied and virtual orbitals to break the symmetry. Gaussian does analyze the symmetries of the orbitals, which as Dave points out can be useful in understanding what's happening in a difficult system. That shows up the case of a highly symmetric system whose wavefunction has broken symmetry. However, in a careful study of these systems one also needs to test stability in order to recognize the opposite case -- the wavefunction stayed symmetric during the SCF cycles, but there is really a lower solution. In the case of a Jahn-Teller distortion, a set of degenerate orbitals is only partially occupied. The result in a Hartree-Fock calculation is that the degeneracy is split, because HF occupied orbitals are optimized for the neutral molecule, but the virtual orbitals "see" n+1 rather than n electrons. Consequently, the density is not symmetric, leading to non-symmetric populations and also forces which break symmetry, leading to a distorted structure if an unconstrained optimization is performed. There can't be a Jann-Teller distortion in a molecule with a non-degenerate point group such as C2v, but a similar effect can result from using an excited-state wavefunction. If the SCF proceedure happens to converge to a state of a different symmetry than the ground state, then a frequency calculation will show an imaginary frequency, which normally implies a saddle point on the nuclear potential energy surface. In the case of an excited state SCF solution, however, what is happening is that the CPHF procedure (which finds the derivatives of the density with respect to the nuclear coordinates) determines that if a distortion is made which breaks symmetry, then the lower energy (ground state) SCF solution can be mixed in, lowering the energy. This case can be distinguished from the usual (nuclear saddle point) case by a stability calculation. I think this is par of what Dave meant about being careful about what potential surface you're on, and that's good advice all the time, not just for Jahn-Teller systems. Finally, someone else suggested that for delocalized systems one can get spurious localized solutions using Hartree-Fock, and thus one needs MCSCF. This is true with RESTRICTED open-shell Hartree-Fock. However, UNRESTRICTED Hartree-Fock can and does usually give properly delocalized orbitals. The simplest example is allyl radical, for which ROHF gives a localized solution with one single and one double bond, and hence optimization with ROHF will give a non-symmetric system. UHF on the same system gives the proper C2v symmetry. There are certainly many systems which are not well described by a single determinant, but one can't make blanket generalizations (at least, not for UHF) based on whether the system has high symmetry. Mike Frisch ------- From arne@mango.mef.ki.se Fri Oct 16 12:09:47 1992 Date: Fri, 16 Oct 92 11:09:47 +0100 From: arne@mango.mef.ki.se (Arne Elofsson) To: CHEMISTRY@ccl.net Subject: RE: Nobel to theory. Dear netters. I think there might be some explanations for the reaseon why theorist not have received so many prizes. 1. The prize is not given to any theorist before the theory have been verified by experiements. (info from X Nobel comittee secretary in Medicine) 2. The nobel prize is given to maximum three persons. In theory I think it is quite common that major advances in comp.chem. involves more than three persons. But let's hope that theoretical chemistry in the future will be more rewarded. PS! I also think it might be difficult to differ between theoretical physics and chemistry. (What said that deGennes is a physist and not a chemist ??) It might just be that some of the theoretical chemistry is put on the board of physics. NOTE! This is my thought and nothing I know. arne From markm@portal.vpharm.com Fri Oct 16 04:28:01 1992 From: markm@portal.vpharm.com (Mark Murcko) Subject: X-plor running on a Stardent P3 ??? To: chemistry@ccl.net Date: Fri, 16 Oct 92 8:28:01 EDT Hello gang. We have ported Xplor to a two-processor Stardent Titan (P3) machine with 128 Mb of memory. It runs __slower__ on this machine than it does on an SGI Indigo, which has the same processor! Even worse, the Stardent has a vector co-processor which the Indigo lacks, so given the nature of the code (FFT's and non-bonds, mostly) you would expect the Stardent to be quite a lot faster. We are currently checking the obvious things, but we haven't found anything yet, so I thought I would ask the group whether anyone has any Stardent and Xplor experiences that might shed some light on the problem... Thanks! / Mark From mfrancl@cc.brynmawr.edu Fri Oct 16 06:20:49 1992 Date: Fri, 16 Oct 92 10:20:49 -0400 From: mfrancl@cc.brynmawr.edu (Francl Michelle M) To: chemistry@ccl.net Subject: G92 and the CI matrix Does anyone out there know how to tease the CI matrix out of G92? A colleague has a need for the off-diagonal terms -- and it seemed that it should be straightforward! Michelle M. Francl mfrancl@cc.brynmawr.edu From lim@omega.chem.yale.edu Fri Oct 16 07:21:10 1992 From: Dongchul Lim Subject: chelpg charge calculation in gaussian To: chemistry@ccl.net (Computational Chemistry) Date: Fri, 16 Oct 92 11:21:10 EDT (I sent the same question to help@gaussian.com. Gaussian people don't need to answer.) In the calculation of chelpg charges with Gaussian 92 program, as someone suggested, I increased the number of points to fit in chelpg charge calculation by stating "IOP(6/42=n)" in route card. I was able to increase n up to 3. Beyond that, the gaussian program died leaving the following message: (We're running IRIX 4.0.1 on SGI 4D/35 with 16Mb memory) Breneman (CHELPG) radii used for charge fitting Generate Potential Derived Charges using the Brenneman model, NDens= 4. Grid spacing= 0.189 Box extension= 2.800 NStep X,Y,Z= 51 53 39 Total possible points= 105417 Too many points to fit, reduce NDens. Error termination in Lnk1e. I wonder how many points are "too many". Also I want to know how I can get symmetric charge distribution for symmetric molecules. For example, the follwing molecule has C2v symmetry and I guess atomic charges must be symmetrical when a reasonable number of points to fit are used. Is there any other way to get symmetric charges other than increasing the number of points, e.g., by using sort of symmetry option? Thank you very much. -DCL ****************** cyclopentadiene.com *********************** %MEM=4000000 #N 6-31G* SCF=(TIGHT,DIRECT) NAME=LIM POP=CHELPG IOP(6/42=4) Cyclopentadiene (C2v) 6-31G*//3-21G Single Point Calculation 0 1 C C 1 R2 C 1 R2 2 A3 C 2 R4 1 A4 3 D4 0 C 3 R4 1 A4 2 D4 0 H 1 R6 3 A6 2 D6 0 H 1 R6 3 A6 2 -D6 0 H 2 R8 1 A8 4 D8 0 H 3 R8 1 A8 5 D8 0 H 5 R10 3 A10 4 D10 0 H 4 R10 2 A10 5 D10 0 Variables: R2 1.51920200 R4 1.32927300 R6 1.08656800 R8 1.06880300 R10 1.06908200 A3 102.10308600 A4 109.65329900 A6 111.61262822 A8 123.42611300 A10 126.56346320 D6 -119.34690620 Constants: D4 0.00000000 D8 180.00000000 D10 180.00000000 ************************************************************** ******************************************************************** * Dongchul Lim | Phone (203) 432-6288 * * Dept. of Chemistry, Yale Univ. | Email: lim@rani.chem.yale.edu * * 225 Prospect Street | (130.132.25.60) * * New Haven, CT 06511 | * ******************************************************************** From pat@mercury.aichem.arizona.edu Fri Oct 16 01:58:52 1992 Date: Fri, 16 Oct 92 08:58:52 -0700 From: pat@mercury.aichem.arizona.edu (Pat Walters) To: markm@portal.vpharm.com Subject: X-plor running on a Stardent P3 ??? Mark, I haven't had any experience with Xplor, but we found that Mopac runs faster on a DecStation 5000 (12.5 mips) than on a Titan (60 mips ?). Could it be that Stardent (Ardent, Kutbota Pacific, or whoever they are today) just doesn't write very good compilers ? _________________________________________________________________________ Pat Walters pat@mercury.aichem.arizona.edu Graduate Student Laboratory for Artificial Intelligence in Chemistry * Dept of Chemistry * * * University of Arizona, Tucson AZ 85721 * * * 602-621-6334 *** "At least its a dry heat" * _______________________________________________________*__________________ From mail Fri Oct 16 12:32:04 1992 Date: Fri, 16 Oct 1992 11:46:35 -0400 From: hyper!guo (Yufei Guo) To: chemistry@ccl.net Subject: Slater-Condon Constants In connection with some work I have been doing on Mike Zerner's implementation of the INDO approximation for transition metals, I have run up against a "lack of parameters" problem. Perhaps somebody in netland can help me. The original papers do not publish the Slater-Condon constants that come into the one-centre two-electron integrals for the second row transition elements. Nor do they have a reference to where they may be found. Despite considerable library work, I have not been able to track these down, and so I turn to the net as a last resort. Does anybody know of a public-domain source for these parameters? Yufei Guo From jle@world.std.com Fri Oct 16 13:07:37 1992 Date: Fri, 16 Oct 1992 17:07:37 -0400 From: jle@world.std.com (Joe M Leonard) To: chemistry@ccl.net Subject: Titan performance Since the Titan is a vector machine, taking non-vector code and telling all of it to run vectorized can result in slowing the code down. Assuming that you are running the latest software and compilers, and run a P3 (R3000) version of the processors cards, one should expect to see perfoemance equivalent to an IBM 530 (1 proc). If you need advice, contact your Kubota rep, or perhaps Doug Smith's research group at UToledo... Joe Leonard jle@world.std.com From zheng@violet.berkeley.edu Fri Oct 16 07:07:04 1992 Date: Fri, 16 Oct 92 14:07:04 -0700 From: zheng@violet.berkeley.edu To: chemistry@ccl.net Subject: Estimate activation barrier from one rate constant In a recent paper, A. Warshel reported that the activation barrier of enzymatic reactions can be calculated from one rate constant with a universal preexponential factor. Since I donot have any experience with with, could someone tell me how reliable is this approach? Another question is related to the EVB/FEP method. I would like to know when doing simulation on enzymatic reaction, whether the whole enzyme is included in the simulation or just few active site residuals. I would appreciate any comment. Please send to me directly. Yajun Zheng From m10!frisch@uunet.UU.NET Fri Oct 16 11:31:14 1992 Date: Fri, 16 Oct 92 15:31:14 EDT From: m10!frisch@uunet.UU.NET (Michael Frisch) Subject: Re: G92 and the CI matrix To: chemistry@ccl.net Does anyone out there know how to tease the CI matrix out of G92? A colleague has a need for the off-diagonal terms -- and it seemed that it should be straightforward! Michelle M. Francl mfrancl@cc.brynmawr.edu For CIS excited state calculations, the CI matrix (over single excitations) is formed if you do the calculation in the MO basis, which is the default. Set IOP33(9=6) to crank up the print and see the matrix. For CID and CISD ground state calculations, the CI matrix is never formed. Gaussian does an iterative diagonalization, forming the product of the CI Hamiltonian with a trial set of coefficients directly from two-electron integrals. (It uses some MO integrals and recomputes the AO integrals in doing different terms). So it's not straightforward to get the program to produce the matrix elements. Mike Frisch -------