From mckelvey@Kodak.COM Fri Feb 25 09:03:19 1994 Received: from Kodak.COM for mckelvey@Kodak.COM by www.ccl.net (8.6.4/930601.1506) id IAA05501; Fri, 25 Feb 1994 08:51:20 -0500 From: Received: from NTPROF.DECnet MAIL11D_V3 by Kodak.COM (5.61+/2.1-Eastman Kodak) id AA02908; Fri, 25 Feb 94 08:49:33 -0500 Date: Fri, 25 Feb 94 08:49:30 -0500 Reply-To: mckelvey@Kodak.COM Message-Id: <9402251349.AA02908@Kodak.COM> To: osc@Kodak.COM Subject: HONDO vs GAMES running parallel Does anyone have any timing comparisons for HONDO versus GAMESS running parallel on RS/6000's ? Will post summary of info rec'd. Thanks in advance.. John McKelvey Eastman Kodak Co. McKelvey@Kodak.Com From young@slater.cem.msu.edu Fri Feb 25 09:11:03 1994 Received: from msu.edu for young@slater.cem.msu.edu by www.ccl.net (8.6.4/930601.1506) id IAA05363; Fri, 25 Feb 1994 08:18:35 -0500 Received: from slater.cem.msu.edu by msu.edu (IBM VM SMTP V2R2) with TCP; Fri, 25 Feb 94 08:17:37 EST Received: by slater.cem.msu.edu (920330.SGI/911001.SGI) for @msu.edu:chemistry@ccl.net id AA23420; Fri, 25 Feb 94 08:19:40 -0500 Date: Fri, 25 Feb 94 08:19:40 -0500 From: young@slater.cem.msu.edu (Dave Young) Message-Id: <9402251319.AA23420@slater.cem.msu.edu> To: chemistry@ccl.net Subject: DFT cancellation of errors In response to the previous allusion to a cancellation of errors in DFT (Density Functional Theory). When the current generation of ab initio programs reports frequencies, they are reporting frequencies as predicted by the harmonic oscillator approximation. Since the real internuclear potential actually looks more like a Morse potential than a harmonic potential, these calculations tend to report frequencies that are too high by about 10%. In the DFT calculations, we have been seeing frequencies which are closer to the experimental frequencies than predicted by Hartree Fock, MPn or nCI. If the calculation were really getting the shape of the internuclear potential correct, we would be getting results approaching the Harmonic frequencies not the observed frequencies. Thus the cancellation of errors is that the error due to the harmonic oscillator approximation is cancelled by the error due to the wrong shape of the internuclear potential. Note that this cancellation is not perfectly consistent, but it does result in DFT frequencies being closer to observed than the other ab initio methods. A good reference, where frequencies are compared to both the observed and the harmonic frequencies is the paper by B. G. Johnson, P. M. W. Gill & J. A. Pople, J. Chem. Phys. 98 (7), page 5612, 1 Apr 93. Dave Young young@slater.cem.msu.edu youngdc@msucem From george@archimedes.cray.com Fri Feb 25 17:03:35 1994 Received: from cray.com for george@archimedes.cray.com by www.ccl.net (8.6.4/930601.1506) id QAA12341; Fri, 25 Feb 1994 16:40:35 -0500 Received: from archimedes.cray.com by cray.com (Bob mailer 1.2) id AA07192; Fri, 25 Feb 94 15:40:27 CST Received: by archimedes.cray.com (4.1/CRI-5.14) id AA03453; Fri, 25 Feb 94 15:40:41 CST Date: Fri, 25 Feb 94 15:40:41 CST From: george@archimedes.cray.com (George Fitzgerald) Message-Id: <9402252140.AA03453@archimedes.cray.com> To: chemistry@ccl.net Subject: CCL: Density Functional Theory Cc: George.Fitzgerald@cray.com Dear Netters, In response to Bob Williams' questions on DFT, I would like to make the following observations: > What is the relative speed of DFT optimization > and frequency calculations on a large molecule > at modest levels of theory > (compared to a similar RHF/4-31G job) The question of relative speed can not be answered without further discussion of different "flavors" of DFT. In the codes such as DGauss (from Cray) or deMon (from Montreal) the Coulomb term is replaced by a three-center integral based on Dunlap's charge fitting scheme. This gives a significant speedup relative to doing the exact 4-center Coulomb integration. However, not all DFT codes do this. I understand that both Gaussian's and CADPAC's DFT routines use 4-center Coulomb integrals. Since I do only 3-center Coulomb integrals in my work, I can only make the comparison between 3-center DFT and 4-center Hartree-Fock. In my experience, such DFT codes are at least one order of magnitude (10x) faster for SCF, gradient, & hessian calculations. > Also, we would like to know something about > the quality of the force matrix and frequencies > produced by modest level DFT calculations, > and about the transferability of scale factors. > That is: how does the "cancelation of errors" produced by > DFT compare to that from the 4-31G level. Less work exists comparing DFT force constants to experiment than Hartree-Fock vs. experiment. In my experience, the average error of DFT frequency (not force constant) is about 5% vs expt. The value typically quoted for Hartree-Fock 6-31G* is about 10%. However, the Hartree-Fock seem to be more consistent than the DFT. That is, Hartree-Fock scaled by .9 (or what ever your favorite scale factor is) are quite reliable. In my experience, there is no comparable, reliable scale factor for DFT computations, but given the relative speed of DFT, many users still find it extraordinarily useful. Be aware that at this point in time, the number of DFT results in the literature is tiny compared with the Hartree-Fock. With the recent advancements in analytic DFT 2nd derivative methods we can expect the DFTers to begin to catch up. It is also important to keep in mind that the 4-31G basis may not be suitable for DFT computations. This is certainly the case if one does charge fitting. We generally use basis sets which are comparable in size to 6-31G*, although we have re-optimized them for the DFT potential. George Fitzgerald Cray Research, Inc. 655E Lone Oak Dr. Eagan, MN 55121 george@cray.com From fpolicar@sun.cis.smu.edu Fri Feb 25 19:03:27 1994 Received: from sun.cis.smu.edu for fpolicar@sun.cis.smu.edu by www.ccl.net (8.6.4/930601.1506) id SAA13287; Fri, 25 Feb 1994 18:44:43 -0500 Received: by sun.cis.smu.edu (/\==/\ Smail3.1.28.1 #28.1) id ; Fri, 25 Feb 94 17:44 CST Date: Fri, 25 Feb 1994 17:44:22 -0600 (CST) From: Franco Policaro Subject: Re: CCL:subsribe To: shs348f cc: Compchem In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Signoff Or at least how do I do it? "Let's get ready to rumble!" ******************************** ++++++++ +++++++ * * * * + + + * G * W * * ++++ +++++++ * R * H * R * + + * E * I * E * + + * E * T * D * + + * N * E * * * * * * * * * * ******************************** fpolicar@sun.cis.smu.edu ____________________________________________________________________________