From chemistry-request@ccl.net Wed Mar 6 09:14:26 1991 Date: Wed, 6 Mar 91 08:15 EST From: "DOUGLAS A. SMITH" Subject: consistency in semiempirical codes To: chemistry@ccl.net Status: RO Here is some food for thought and fuel for a fire regarding MOPAC 5.0 versus MOPAC 6.0. Those of you who read the DECnet news feed will know that Brent Bessler 'ported' MOPAC 6.0 to several platforms and released the code via an anonymous ftp site. We found that his code did NOT reproduce the MOPAC 5.0 output of the many test jobs available, and posted this to the same newsfeed. Brent later told me that he had not run all the test jobs. A lively discussion ensued in that forum, with Mike Peterson at U. Toronto telling tales of major woe when trying to port and test MOPAC 6.0 on his Apollo DN10000. As a result of 1) the above situation, and 2) the fact that I had recently been to the USAFA and gotten a version of 6.0 from Jimmie Stewart, Richard Counts at QCPE asked me to test his DECstation and VAX versions of MOPAC 6.0 for test job output and to update it where ever Jimmie had made bug fixes, etc. We are now doing that and should have the final product done by this weekend for the VAX, DECstation, and Apollo 68000-based platforms. This code will then be certified for release through QCPE. THE QUESTION. For an identical set of parameters, should MOPAC 6.0 reproduce the output on any given job that was generated by MOPAC 5.0? This of course would not be the case if there were bug fixes, but what about simply changes in the code (i.e. designed to make the code more efficient)? What about if a different optimization routine was used (MOPAC 6.0 has eigenvector following as the primary optimization routine rather than BFGS as in MOPAC 5.0 or DFP as in MOPAC 4.0 - yet the change from 4.0 to 5.0 did not change the output). Richard Counts claims, while admitting to being cynical, that since these are semiempirical methods he would not necessarily expect the results to match. But then comes another question. If the results don't match, which is (dare I say?) better? Or, given that we have 2+ years of research results using MOPAC 5.0, do we have to redo all of that work in order to switch over to 6.0 and be able to compare results? "Life's just full of tough choices, isn't it?" (Ursula the Sea Witch to Ariel, Dsiney's The Little Mermaid). Doug Smith Assistant Professor of Chemistry The University of Toledo Toledo, OH 43606-3390 voice: (419)537-2116 fax: (419)537-4033 FAX0236@UOFT02.UTOLEDO.EDU --- From chemistry-request@ccl.net Wed Mar 6 09:53:29 1991 Date: Wed, 6 Mar 91 08:42:12 EST From: balbes@osiris.rti.ORG Subject: Brookhaven PDB files via e-mail To: chemistry@ccl.net Status: RO For those who may not know, there is a way to get Brookhaven PDB files by e-mail. To get detailed information, send email to netserv@embl.bitnet. In the body of the message, put (make sure they are separate lines) : help help proteindata Further instructions will be returned by email, usually within 24 hours. Lisa %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% balbes@osiris.rti.org %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Lisa M. Balbes, Ph.D. "Life would be boring without problems." Research Triangle Institute P. O. Box 12194 919-541-6563 Work Research Triangle Park, NC 27709-2194 919-490-1888 Home %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% --- From chemistry-request@ccl.net Wed Mar 6 10:32:44 1991 Date: Wed, 6 Mar 91 09:52:00 EST From: rs0thp@rohmhaas.COM Subject: RE:consistency in semiempirical codes To: CHEMISTRY@ccl.net Status: RO >Doug Smith writes that should MOPAC 6.0 reproduce >the output on any given job that was generated by MOPAC 5.0? .. >[even if there are different optimizers...] and >If the results don't match, which >is (dare I say?) better? Or, given that we have 2+ years of research results >using MOPAC 5.0, do we have to redo all of that work in order to switch over >to 6.0 and be able to compare results? I have a home-converted version of MOPAC6 that I obtained from QCPE in late December 1990. My opinion is that MOPAC6 is fast, but not trustable. We have been extensively comparing MOPAC5 with MOPAC6, and have found results that first converged, and then MOPAC 6 carried on and generated results (geometry optimizations) that had 10kcal higher energies. (Not good when we are looking for 2-5kcal changes) My version of MOPAC6 runs on an IBM RS/6000, but I have tested the VAX version as well, and it generated the same variable results. I have heard that there might be dimensioning differences using 60 atoms or more, but so far my results have been with less than 50 atoms per molecule. So I hope that the bug(s) that were found were not just limited to large molecules. We are continuing with MOPAC 5. As for that same results with different optimizers, that is something that I do not expect. However, I would like the results to be in the same neighborhood (1-2 kcals) or I would think that the optimizer was not adequate for the problem. Using new code is always tricky, but MOPAC is too useful (and CPU time- consuming) to just use the old technology. -------------------------------------------------------------------- Sincerely, Thomas Pierce Computer Applications Research | rs0thp@rohmhaas.com Internet Bldg 64C | rs0thp@rohvm1 Bitnet Rohm and Haas Co. | (215)-785-8989 Voice Bristol, PA 19007 | (215)-781-4204 Fax < Artificial Intelligence is Better than None! - AAAI Slogan > --- From chemistry-request@ccl.net Wed Mar 6 15:16:52 1991 Date: Wed, 6 Mar 91 13:55:20 CST From: berry@rainbow.uchicago.edu Subject: Hurrah! To: chemistry@ccl.net Status: RO --- From mckelvey@Kodak.COM Wed Mar 6 15:46:08 1991 Date: Wed, 6 Mar 91 15:48:55 -0500 From: mckelvey@Kodak.COM To: osc@Kodak.COM Subject: remarks concerning mopac version X Status: RO I have followed the recent comments concerning MOPAC with interest. We have ported MOPAC V5 to many machines including IBM/3090, MIPS, SUN, DECSTATION 3100, PERSONAL IRIS, IBM RS6000( 320 and 530), ARDENT, STELLAR, FPS/464. We have had it ported to TEKTRONIX 4337, and DG AVION, and ported it ourselves to the AVALON board. These last three are based on the MOROROLA 88k chip, and all have the same GREEN-HILLS compiler. George Purvis at TEKTRONIX has ported it to the CACHE/MAC II , also using the 88K chip, using the ABSOFT compiler. For our benchmark we obtained the same answer to +/- .005 kcal for all IEEE machines and the IBM/3090. These answers differed from the VAX results (uvaxII,8600) if the default D_FLOAT compiler option was used; however, if G_FLOAT was used then all answers agreed. Please note also that there was an article by STEWART and BOYD pointing out the importance of the keyword PRECISE and CRAY vs VAX. (J. Comp Chem.) Also note that not all machines save variables locally after exiting, hence it may be necessary to explicitly save (however done) local variables. There is a "static" option on certain machines, e.g. MIPS base machines. We have MOPAC V6 running on the RS 6000. If users have copies from wherever, but without proper documentation, then that's when the fun starts. There are SEVERAL!! new geometry optimisation routines. Eigenvalue following is only one. Another is the GDIIS routine , by PULAY, and which is a varient on the DIIS (or PULAY) SCF extrapolation method in the program. The GDIIS method chases an extremum very hard and may not find the one you want. We have found the eigenvalue following (EF) routine ok. If you like neither and want FLEPO then specify NODIIS and don't use EF. The discussions here are related to ALGORITHMS, it seems to me and not semiempirical methods in general. It is not meaningful to say that these difficulties discussed in this forum are to be blamed on the semiemperical nature of things. I do believe that a contributing factor is the BLACK-BOX nature of things and too much blind use thereof. From AHOLDER@VAX1.UMKC.EDU Wed Mar 6 20:23:19 1991 Date: Wed, 6 Mar 91 19:23 CST From: AHOLDER@VAX1.UMKC.EDU Subject: How 'bout AMPAC? To: CHEMISTRY@ccl.net Status: RO As a former member of the Dewar group, I have always marveled at the apparent popularity of MOPAC over AMPAC. AMPAC is faster and has some extremely important additions that MOPAC does not incorporate. To be fair, MOPAC has things that AMPAC does not, but the lean in the direction of MOPAC puzzles me. How about it? Andy Holder U. of Missouri-Kansas City Dept. of Chemistry AHOLDER@UMKCVAX1