From kirving@seciu.uy Mon Oct 17 07:20:46 1994 Received: from seciu.uy for kirving@seciu.uy by www.ccl.net (8.6.9/930601.1506) id GAA19910; Mon, 17 Oct 1994 06:43:58 -0400 Received: by seciu.uy (Smail3.1.28.1 #27) id m0qwpTR-000lzsC; Mon, 17 Oct 94 07:39 GMT-3:00 Message-Id: Date: Mon, 17 Oct 94 07:39 GMT-3:00 From: kirving@seciu.uy (kenneth irving) To: CHEMISTRY@ccl.net Subject: Re: BSSE and DFT Hi! With respect to the question on BSSE and DFT: Martina Kieninger, Sandor Suhai and Istvan Mayer have had another paper on the subject accepted for Chem. Phys. Letters and should appear shortly. This paper reports the developing and coding of Mayer's Chemical Hamiltonian approach to correct BSSE in the framework of DFT. The application reported is, precisely, the water dimer. As I understand it, the author has also other work going on on this subject. I suggest you reach her(M.K.) directly at dok411@dkfz-heidelberg.de. Cheers. Oscar ------------------------------------- Oscar N. Ventura (oscar@bilbo.edu.uy) From bewilson@emn.com Mon Oct 17 11:20:47 1994 Received: from emngw1.emn.com for bewilson@emn.com by www.ccl.net (8.6.9/930601.1506) id LAA24858; Mon, 17 Oct 1994 11:19:18 -0400 Received: from stc150.kpt.emn.com by emngw1.emn.com with SMTP id AA13320 (5.67b/IDA-1.5 for ); Mon, 17 Oct 1994 11:10:32 -0400 Received: by stc150.kpt.emn.com (MX V4.1 VAX) id 13; Mon, 17 Oct 1994 11:18:43 EDT Date: Mon, 17 Oct 1994 11:18:42 EDT From: "Bruce E. Wilson (bewilson@emn.com), Eastman Chemical Company" To: chemistry@ccl.net Message-Id: <00986139.4F03D0E0.13@stc150.kpt.emn.com> Subject: Mac->DOS/Model(l)ing > Would anyone on the net know how to read MAC files on > a DOS/MSWindows PC? The files were created on a MAC computer > using MS WORD 5.x. Any pointers to commercial/public > domain software will be greatly appreciated. Many thanks. I have both a Mac and a Windows PC in my office and do this regularly. If you have access to a Mac with a 1.4 Mb Floppy (aka SuperDrive), it can write a PC floppy. Apple File Exchange can copy the Mac file to the PC disk (use the "no translation" option). I normally use AccessPC to make DOS disks look just like Mac disks. Note that you do have to rename the file into the xxxxxxxx.xxx format required by DOS. Word for Windows 2.x or Word for Dos 6.x will read the Mac Word 5.x document. I do find, however, that RTF translates a bit more reliably between Dos and Mac than either of the native formats, so if you can, have your collaborator save the files as RTF, rather than native format. If you don't have a Mac available, life is more complex (in more ways than one, but that's a whole 'nother religous debate ;-) ). At one point, there was a DOS product that enabled a DOS machine to read a Mac floppy. I don't know for sure, but I believe it was called MACMOUNT, or something such. Folks on info-mac would probably have a better recollection of such (info-mac@sumex-aim.stanford.edu). > modeling versus modelling. At one point, I learned that the rule in English was that you double a single final consonant when adding a suffix. However, my Word spelling dictionary says "modeled" and "modeling" are correct, and my American Heritage dictionary lists the one ell versions as the preferred spelling, with two ells as the secondary. Finally, the last two papers I published in ACS journals had my two ell spellings changed to one ell. Relatively conclusive evidence, I'd guess. Bruce Wilson (bewilson@emn.com) From acklin@chbs.CIBA.COM Mon Oct 17 12:21:46 1994 Received: from chsun.eunet.ch for acklin@chbs.CIBA.COM by www.ccl.net (8.6.9/930601.1506) id LAA25451; Mon, 17 Oct 1994 11:47:21 -0400 Received: from Ciba-GW.CHBS.CIBA.COM by chsun.eunet.ch (8.6.4/1.34) id QAA13250; Mon, 17 Oct 1994 16:22:57 +0100 Received: from Central.CHBS.CIBA.COM by Ciba-GW.CHBS.CIBA.COM (4.1/Ciba2.0-R1.0) id AA15225; Mon, 17 Oct 94 16:22:38 +0100 Received: from camm1.ph.chbs by Central.CHBS.CIBA.COM id AA13893; Mon, 17 Oct 1994 16:22:07 +0100 (5.65c8/Ciba2.0-C1.1) Received: by camm1.ph.chbs (931110.SGI/930416.SGI/camm.ph.chbs-1.0) for @cgchm.is.chbs:chemistry@ccl.net id AA22461; Mon, 17 Oct 94 16:24:29 +0100 Date: Mon, 17 Oct 94 16:24:29 +0100 From: acklin@chbs.CIBA.COM (Dr.Pierre Acklin) Message-Id: <9410171524.AA22461@camm1.ph.chbs> To: chemistry@ccl.net Subject: summary iron calculation Dear Netters, two weeks after the posting of my questions about calculating iron complexes with ab initio or semiempirical methods to the list, I'll summarise the answers I've got until now after the repetition of my original query. Many thanks to all of you who have replied. > I am trying to model iron complexes with round about 30 > heavy atoms. Introducing special angular and torsional constraints > into amber force field calculation in MacroModel leads to good > and reproducable geometry predictions. > Geometry predictions are ok as they allow to discriminate > between molecules which can form chelate complexes and those, which > can't. Nevertheless it would be very helpful for the project, if > some kind of energy or stability predictions would be possible. > This makes me ask you the following questions: > Is it possible to treat iron complexes with more than 30 heavy > atoms in an ab initio program like Gaussian in a reasonable time > on a SGI workstation? > Are there any basis sets and references therefore available to > handel iron in Gaussian92 or DFT on a better level than sto-3g? > How good are results of semi empirical calculations form Ampac > or Zindo in terrms of energy and geometry? > I'll be glad to summarise the responses for the list. ******************************************************************* Dear Dr. Acklin, concerning your CCL-questions about iron complexes, we can offer you the possibility to do your calculations on our CRAY C90 supercomputer here at the University of Stuttgart, Germany. The machine has 4 CPU's, 8 GBytes of main memory and much more than 100 GBytes of local disk space available. Of course this is not for free, so we charge some small amount per cpu-hour ( DM 300,- without service, DM 500,- including service, doing batch-calculations with normal priority) for industrial customers. Best regards, Heinz Poehlmann .-------------------------------------------------------------------------. | Dr. Heinz W. Poehlmann Regionales Rechenzentrum | | Manager Applications Department der Universitaet Stuttgart | | Computational Chemistry Support (University Computer Center) | | phone: (49)-711-685-5992 Allmandring 30 | | fax: (49)-711-6787626 D-70550 Stuttgart | | e-mail: poehlmann@rus.uni-stuttgart.de Germany | '-------------------------------------------------------------------------' ************************************************************************ Dear Pierre, it is always a question of time and precision...and symmetry of course! My experience with ab initio calculations in this field is that they have to be carried out at least at the MP2-level of theory (for first row transition metals, heavier give reasonable geometries already on the HF level!). MP2 usually increases computational time and hard disc require- ment. See recent papers by Gernot Frenking for ab initio calculations involving transition metals. I got the impression that DFT makes it slightly easier to handle larger TM complexes since electron correlation is always included while computational efford is comparable to HF calculations. There is a recent review by Tom Ziegler covering this topic: Tom Ziegler, Chem. Rev. 1991,91,651-657 I currently carry out DFT calculations on large metallocenes and the results come in... I have heard a lot about ZINDO, also on Iron complexes. Results looked good, though I cannot tell anything about it. Look for papers by Reinhold's group in (Leipzig, Halle, Magdeburg; sorry don't know exactly) and the original papers of course... Finally there are still Extended Hueckel calculations. Hoffmann in Munich is a good reference for these.... Hope this helps, John Lohrenz ----------- John Lohrenz Dept. of Chemistry Univerity of Calgary email: lohrenz@zinc.chem.ucalgary.ca *********************************************************************** Dear Pierre I have been doing calculations on organomettallic iron compounds of about 20 atoms (1 iron, 9-10 carbons, 2 oxygens, rest H). I was using STO-3G for geometry optimizations (there are a few problems with this of course) and something better for single points. > This makes me ask you the following questions: > > Is it possible to treat iron complexes with more than 30 heavy > atoms in an ab initio program like Gaussian in a reasonable time > on a SGI workstation? If you're interested in single point, this should be no problem, depending on how much correlation you want to include. In GAMESS a single point calculation (78 basis functions) took about 12 minutes on a SGI Indego 2 (100MHx R4000). You may want to use Effective Core Potentials (ECP's) in your system. > > Are there any basis sets and references therefore available to > handel iron in Gaussian92 or DFT on a better level than sto-3g? Yes you can check the papers by Marynick et al or use the predifined ones in the package. D. S. Marynick et al, Organometallics, 1991, 10, 2816-2823 and refs therin. > > How good are results of semi empirical calculations form Ampac > or Zindo in terrms of energy and geometry? I tried ZINDO calculation on my systems, but got very strange structures forming (eg carbonyls were pulled off the metal and incorporated into a alkyl chain, forming a small ring system). I have a feeling it works better for coordination complexes, but no experience myself. It should be remembered that ZINDO's primary aim was the description of electronic transitions in Metal-Ligand systems (See Michael Zerner's work on Porphrins etc). No experience with Ampac. Spartan, from Wavefunction inc is busy working on semi empirical parms for 1st row TM's. Good Luck Oliver ______________________________________________________________________ | Oliver Hill | | | Department of Chemistry | "The true scientist never | | University of Cape Town | loses the faculty of | | Rondebosch, 7700 | amazement. It is the essence | | SOUTH AFRICA | of his being." | | E-Mail:oliver@psipsy.uct.ac.za | - HANS SELYE | |--------------------------------------------------------------------| | Contact until 24 October 1994 | | | | E-Mail: semper@dircon.co.uk 1 Belgrave Place | | Tel. +44-71-235-8197 London SW1X 8BU | | Fax. +44-71-823-2112 UNITED KINGDOM | |____________________________________________________________________| ************************************************************************** Dear Pierre, A good basis for iron, and other first transition series, is Wachter's as implemented in HONDO8. Yes you can run upto 30 atoms, SCF direct, using Gaussian, hopefully in less than 12 hours. No, you cannot use AMPAC for iron but I believe SAM1 has recently been parameterized and other semi-empirical codes include ZINDO (from BIOSYM) or MNDO-d (from W. Thiel, Zurich), Good luck, John Waite (chem8@vax.york.ac.uk) PS I would be interested in other transition metal bases. ******************************************************************** Hi! I have been doing computations on iron porphyrins with ADF, the Amsterdam Density Functional program, on an INDIGO R4000 from SGI. The performance is pretty good and you get all the benefits of DFT computations. Namely, almost MP2 quality for energetics. Regarding DFT in G92, I had a bit of a bad experience since the guess for the density is based on a semiempirical HF computation, a very bad guess for my high spin system. Moreover, I could not converge on ROHF and use the density for G92/DFT. I also had to use the finegrid as default, otherwise, the numerical integration scheme would complaint. Hence, I left DFT/G92 behind, but I remind you that some of these problems (i.e. the numerical integration bit) have been addressed through upgrades. I am very happy with ADF. It has recently been upgraded to give thermochemistry data and many of the features that you find in G92. Nevertheless, ADF still has some way to go before it can compete with G92, particularly regarding the variety of functionals that you get with G92. Regarding basis sets, ADF comes with its own suite of basis sets since it uses STO. For G92 and similar programs you are better off using ECP's. Check out the work by CUNDARI who has done HF like computations with ECP in several transition metal complexes. I give you references below to some ECP's that CUNDARI has used successfully in studying the energetics and geometries of several tm complexes. Finally, I run with C4v or D4h symmetry using a triple Z basis set on the Fe and double Z with polarization on the light atoms. Single point computations on a single user R4000 SGI and ADF v. 1.0 take around 4-5 hrs. The newest version implements the DIIS algorithm so convergence is much faster. I think this will make a big difference. ************** The following references should provide you with very recent ECP's for most of the periodic table. The comments I have heard from those who have used them are rather favorable (comments limited to energetics and geometries, but recent work on spectroscopy by Garmer and Krauss is rather impressive): Stevens et. al. J. Chem.Phys. 81 (12), Pt. II, 1984 First and Second Row atoms Stevens et. al. Can. J. Chem. 70, 612, (1992) Third through Fifth row atoms. Some typos are known: Table 2: In RECP for second Vp-f component: -129.87594 should read -129.78594 Table 4: Basis for Fe Cp term for 1sp should be -0.007940, not -0.07940 asp term for 4sp should be 0.0410, not 0.410 Table 7: Result for Be Eval with core a should read -1.01214, not -0.01214 Cundari and Stevens J. Chem. Phys. 98 (7) 1993 Lanthanides I hope this is useful. These papers also discuss the variety of methods available to generate ECP's, relativistic effects and references to ECP's for DFT computations. Good Luck Gustavo A. Mercier, Jr. mercie@cumc.cornell.edu *********************************************************************** Dear Dr. Acklin, As you are aware (from your message) our AMPAC program includes the new SAM1 Hamiltonian which can treat iron systems. We are quite pleased with the results. If you would like to provide a test molecule or two, we would be happy to run it for you. (A set of cartesian coordinates would be best.) On another note, up until recently, your company owned a license to AMPAC. Heinrich Karfunkel elected to allow that license to lapse because he disagreed with our licensing scheme. Your New York research site still owns a license, so we would be happy to provide you with a second license at our very reduced rate if you find that you need the program in Basel. Andy Holder =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= DR. ANDREW HOLDER President, Semichem Semichem, Inc. || Internet Addr: aholder@vax1.umkc.edu 7128 Summit || Phone Number: (913) 268-3271 Shawnee, KS, 66216 || FAX Number: (913) 268-3445 =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= ========================================== Dr.Pierre Acklin CIBA-GEIGY AG K-136.P.13 CH-4002 Basel acklin@chbs.ciba.com ++41 61 696 23 62 ++41 61 696 23 75 ========================================== From berkley@wubs.wustl.edu Mon Oct 17 13:20:46 1994 Received: from ecl.wustl.edu for berkley@wubs.wustl.edu by www.ccl.net (8.6.9/930601.1506) id NAA27331; Mon, 17 Oct 1994 13:08:04 -0400 Received: from wubs.wustl.edu by ecl.wustl.edu (5.0/ECL-A1.27) id AA19208; Mon, 17 Oct 1994 12:08:03 -0500 Received: by wubs.wustl.edu; (5.65/1.1.8.2/26Aug94-0755AM) id AA02601; Mon, 17 Oct 1994 12:04:58 -0500 Date: Mon, 17 Oct 1994 12:04:58 -0500 From: "Mr. Berkley Shands" Message-Id: <9410171704.AA02601@wubs.wustl.edu> To: chemistry@ccl.net Subject: More benchmarks content-length: 5185 Receptor-III (TM - Tripos Associates) Release version V3.0 timings for the ACE series. Uniform Radial Increments at 0.25 Angstroms. Modified Mode Adaptive Sampling. Minimum scan factor 2.57 degrees, maximum scan factor 45.84 degrees, Average of 11.55. Uniform grid resolution of 0.25A. Fully automated setup and processing applied. Coordinates, torsions, and constraints calculated in double precision. The dataset consists of 45 MOPAC/AM1 minimized molecules having an IC50 > 6.0. All molecules pre-processed automatically for high energy conformer avoidance. Smallest molecule had 5 variable torsions (1e07 max conformers). Largest molecule had 19 variable torsions (1e27 max conformers). ******************************************************************************* Benchmarks of the ACE series of 45 molecules IC50 > 6.0 ******************************************************************************* Elapsed Child_CPU System_CPU Chargable Forks Ratio ======= ========= ========== ========= ===== ===== 1183 810.810 16.120 756.600 2 0.695 SGI Indigo 48Mb -O2 (R3K/33Mhz) 383 736.120 27.810 670.770 5 1.979 SGI 4d/380s 4/8 128Mb -O2 (R3K/33Mhz) 463 335.690 6.220 311.060 2 0.736 SGI Challenge M 96Mb -O2 (R4K/100Mhz) 498 345.930 8.200 320.140 2 0.709 SGI Indigo2 64Mb -O2 (R4K/50Mhz) 345 293.183 5.017 249.233 2 0.870 AXP 3K/300 Std CC -O2 32mb V3.0 311 263.700 4.233 222.033 2 0.870 AXP 3K/300 Gem CC -O4 32Mb V3.0 431 242.817 4.983 201.317 2 0.581 AXP 3K/400 Gem CC -O4 64Mb V2.0 418 242.267 10.983 200.150 2 0.590 AXP 3K/300 Gem CC -O4 32Mb V2.0 181 188.800 3.317 160.167 3 1.073 AXP 2100-5 2/4 Gem CC -O4 128Mb V2.0 113 100.533 1.300 84.333 2 0.906 AXP 3K/900 Gem CC -O4 512Mb 275Mhz V3.0 Notes: Street value for the above machines: AXP 3K/900 $45K, AXP 2100-500 (4cpu) $60K ($100K Loaded) AXP 3K/300 $4k (used) $8K (new) AXP 3K/400 $10K SGI Indigo $12K, Indigo2 $15K, Challenge M $20K SGI 4d/380 ($150K new) $15K today used. ALL machines should have at least 64Mb, with the AXP 2100-500 at 128Mb minimum. The greatly effects the SGI price upward. No SGI R8000 (TFP) timings are available, but at $500K, it doesn't seem worthwhile :-) All SGI machines ran IRIX 5.2 except the Indigo2 at 4.0.5 The "GEM" compiler is in save set DECC on the OSF/1 CD at V3.0. (cc -migrate; requires CMPDEVENH and OTABASE at V2.0) The "STD" compiler is the default "CC" for OSF/1. "Child_CPU" is the recorded total CPU for all forked processes, including overhead and I/O. "Elapsed" is the recorded elapsed time from command line parsing to the exit of the last child processes. "System_CPU" is the recorded UNIX system overhead from the times() function "Chargable" is the algorithmic chargable time from initial rotations to termination (child process CPU) ***************************************************************************** CPU benchmarking for current architectures September 1994. Software is an analytic ring closure Multiprocessor test suite "Elapsed" is the wall clock time. "Child_CPU" is the total CPU over all forks. "System_CPU" is measured system overhead. "Conformers" are stericly allowed conformers. (of the last molecule processed) "Forks" is the number of fork() system calls. "Ratio" is Child_CPU / Elapsed "Ovhd/Fork" is the System_CPU / Forks. pentcrys (cyclic pentapeptide with distance constaints) cyclooctane - cycloalkane with hydrogens. *** Single Threaded Application - Ringsearch *** DEC Alpha 3000/900 (275Mhz 512Mb OSF1 V3.0) Uniprocessor Elapsed Child_CPU System_CPU Conformers 455 452.650 0.533 1515 DEC alpha 2100 - 1/4 cpu (128 Mb OSF1 V3.0) Multiprocessor Elapsed Child_CPU System_CPU Conformers 731 729.700 0.100 1515 wubs2 DEC Alpha AXP 3000/300 (32MB OSF1 V3.0) Uniprocessor Elapsed Child_CPU System_CPU Conformers 1261 1243.317 1.967 1515 sgi 4d/380 1/8 cpus (128Mb IRIX 5.2) Multiprocessor Elapsed Child_CPU System_CPU Conformers 3738 3747.350 22.680 1515 ************ Multisearch (multi-threaded ringsearch) ***************** DEC Alpha 3000/900 (275Mhz 512Mb OSF1 V3.0) Uniprocessor Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 460 453.217 1.350 1515 353 0.985 0.003824 DEC alpha 2100 - 4/4 cpu (128 Mb OSF1 V3.0) Multiprocessor Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 195 753.717 1.833 1515 353 3.865 0.005194 sgi 4d/380 (128Mb IRIX 5.2) 4/8 cpus Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 973 3813.390 42.180 1515 353 3.919 0.119490 sgi 4d/380 (128Mb IRIX 5.2) 8/8 cpus Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 549 4012.360 54.070 1515 353 7.308 0.153173 *************** Comments ************** IRIX throughput dropped a great deal going from 4.0.5 to 5.2 >From 7.9x down to 7.3x :-( From A.W.J.Smith@reading.ac.uk Mon Oct 17 14:20:50 1994 Received: from mail2.rdg.ac.uk for A.W.J.Smith@reading.ac.uk by www.ccl.net (8.6.9/930601.1506) id OAA28919; Mon, 17 Oct 1994 14:20:28 -0400 Received: from suma1 (actually suma1.rdg.ac.uk) by suma2.rdg.ac.uk with SMTP - Local (PP); Mon, 17 Oct 1994 19:20:17 +0100 Date-Received: Mon, 17 Oct 94 19:20:12 BST Date: Mon, 17 Oct 1994 19:20:11 +0100 (BST) From: "Andy W. J. Smith" X-Sender: scrasmit@suma1 To: chemistry@ccl.net Subject: AMoO4 parameters Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Hi, Does any one know of any published parameters (LJ, Buckingham,shell etc) to model the crystal structures of the AMoO4 molybdate compounds.(where A=Co,Mn,Ni) Any suggestions welcomed. thanks, ANDY @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ Andrew W.J. Smith Department of Chemistry, University of Reading Whiteknights, Reading, Phone: 44 (0)734-875123 ex 7441 or 4195 Berkshire, Email: A.W.J.Smith@rdg.ac.uk RG6 2AD, UK. Fax : 44 (0)734- 311610 * email is most likely to get to me first * @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ From lagunez@redvax1.dgsca.unam.mx Mon Oct 17 16:20:48 1994 Received: from redvax1.dgsca.unam.mx for lagunez@redvax1.dgsca.unam.mx by www.ccl.net (8.6.9/930601.1506) id PAA01407; Mon, 17 Oct 1994 15:30:39 -0400 Message-Id: <199410171930.PAA01407@www.ccl.net> Received: from standard-input by redvax1.dgsca.unam.mx; Mon, 17 Oct 94 13:28:55 CST Date: Mon, 17 Oct 94 13:28:55 CST From: lagunez@redvax1.dgsca.unam.mx (Lagunez Otero Jaime-IQ) Sender: lagunez@redvax1.dgsca.unam.mx To: chemistry@ccl.net Subject: methylation I am interested in quantum chemical calsculations of DNA methylation - do you have any information? Thank-you. Jaime Lagunez, PhD From hinsenk@ERE.UMontreal.CA Mon Oct 17 17:20:49 1994 Received: from condor.CC.UMontreal.CA for hinsenk@ERE.UMontreal.CA by www.ccl.net (8.6.9/930601.1506) id QAA03599; Mon, 17 Oct 1994 16:31:19 -0400 Received: from eole.ERE.UMontreal.CA by condor.CC.UMontreal.CA with SMTP id AA14747 (5.65c/IDA-1.4.4 for CHEMISTRY@ccl.net); Mon, 17 Oct 1994 15:21:11 -0400 Received: from cyclone.ERE.UMontreal.CA by eole.ERE.UMontreal.CA (940406.SGI/5.17) id AA09487; Mon, 17 Oct 94 15:21:09 -0400 Received: by cyclone.ERE.UMontreal.CA (940406.SGI/5.17) id AA06848; Mon, 17 Oct 94 15:21:09 -0400 Date: Mon, 17 Oct 94 15:21:09 -0400 From: hinsenk@ERE.UMontreal.CA (Hinsen Konrad) Message-Id: <9410171921.AA06848@cyclone.ERE.UMontreal.CA> To: haney@netcom.com Cc: CHEMISTRY@ccl.net, amber@cgl.ucsf.edu, charmm-bbs@emperor.harvard.edu In-Reply-To: <199410142205.PAA18950@netcom6.netcom.com> (haney@netcom.com) Subject: Re: CCL:Help on SHAKE constraints I wish to ask about the use of constraints like SHAKE in molecular dynamics. What are the reasons that one would use SHAKE? What are the deficiencies of SHAKE? When SHAKE (or the like) is used and you experience problems: 1)What are the symptoms of "failure"?, 2)Under what conditions does this failure occure? and 3)What is the workaround for this failure? When doing MD simulations, you want to make your time step as big as possible. However, you must make them small enough to sample correctly the fastest motions in your system, which are typically bond vibrations. The idea behind constraints is to eliminate the fastest motions by simply fixing some internal coordinates (like bond lengths). Once they are fixed, there are no more fast motions, and you can increase you time steps. Of course this means that you are simulating a different system. Some physical quantities will change as a consequence of introducing constraints; an obvious example are bond length fluctuations, which become zero. But if the eliminated degrees of freedom are sufficiently well uncoupled from the remaining ones, you still get accurate results for the quantities you are really interested in. Some people dislike the use of constraints, arguing that since real molecules do not have these constraints, the simulation shouldn't have them either. To me, this is not at all obvious. Any MD simulation makes some rather drastic simplifying assumptions; the most obvious one is the use of classical mechanics instead of quantum mechanics. Whether or not to use constraints is simply part of a definite model, just as the choice of this or that force field. Whether a model is good or bad needs to be tried out. Unfortunately, there have been next to no systematic studies on the effect of constraints on the interesting quantities in real systems. It should be kept in mind that using a larger time step not only reduces cost, but also roundoff errors, so the results might iin some respect get better by using constraints. If you decide to use constraints, you have to get suitable equations of motion for the constrained system. There are two well-established methods from classical mechanics: 1) Add the constraints to the unconstrained equations of motion using Lagrange multipliers. 2) Introduce new coordinates that automatically satisfy the constraints. In theory both methods are equivalent, but of course not in numerical applications. If you follow the first approach, you have to solve the problem that the constraints are satisfied only for infinite time steps. To get them exactly satisfied, you have to solve the constraint equations explicitly for a finite time step. You then get a set of quadratic equations, which has to be solved iteratively. This is exactly what SHAKE does. And that also explains how it might fail: There is no way to prove that the iteration converges, and sometimes it doesn't. Most implementations simply stop after some number of iterations. I have been told that the "solution" is simply to make a few steps without SHAKE and then go on as if nothing had happened. The second approach has rarely been tried, probably it looks more complicated at first glance. However, it is possible to derive general equations of motion with just a few types of motion (translation and some kinds of rotation), and it is possible to let the computer find these coordinates automatically. The advantage is that you needn't worry about the exactness of your constraints. Another advantage is that you can treat a much wider range of constraints (rigid subgroups, planar molecules, etc.) Besides, the coordinates themselves may prove useful for analysis of the simulation results. On the other hand, there is a price to pay: you get velocity-dependent forces (Coriolis and centrifugal forces), and the implementation is a bit more complicated. If you are interested in this approach, have a look at the paper "Generalized Euler equations for linked rigid bodies", by G. Kneller and K. Hinsen, in the September issue of Phys. Rev. E. It contains the derivation of the general equations of motion. We will publish something on the implementation and some results soon. ------------------------------------------------------------------------------- Konrad Hinsen | E-Mail: hinsenk@ere.umontreal.ca Departement de Chimie | Tel.: +1-514-343-6111 ext. 3953 Universite de Montreal | Fax: +1-514-343-7586 C.P. 6128, succ. A | Deutsch/Esperanto/English/Nederlands/ Montreal (QC) H3C 3J7 | Francais (phase experimentale) ------------------------------------------------------------------------------- From bak@isadora.albany.edu Mon Oct 17 22:20:52 1994 Received: from UACSC2.ALBANY.EDU for bak@isadora.albany.edu by www.ccl.net (8.6.9/930601.1506) id VAA08281; Mon, 17 Oct 1994 21:26:06 -0400 Received: from isadora.albany.edu by UACSC2.ALBANY.EDU (IBM VM SMTP V2R2) with TCP; Mon, 17 Oct 94 21:23:17 EDT Received: by isadora.albany.edu (931110.SGI/920502.SGI) for @uacsc2.albany.edu:chemistry@ccl.net id AA20995; Mon, 17 Oct 94 21:25:49 -0400 From: bak@isadora.albany.edu (Brian A. Kell) Message-Id: <9410180125.AA20995@isadora.albany.edu> Subject: Re: CCL:More benchmarks To: berkley@wubs.wustl.edu (Mr. Berkley Shands) Date: Mon, 17 Oct 94 21:25:49 EDT Cc: chemistry@ccl.net In-Reply-To: <9410171704.AA02601@wubs.wustl.edu>; from "Mr. Berkley Shands" at Oct 17, 94 12:04 pm X-Mailer: ELM [version 2.3 PL11] > [...] > Street value for the above machines: > [...] > No SGI R8000 (TFP) timings are available, but at $500K, it doesn't seem > worthwhile :-) Excuse me? There are R8000 machines available for substantially less than $500K. An R8000-based Power Indigo2 is not too much more than an R4400-based Indigo2. brian ------------------------------------------------------------------ Brian Kell bak@biosym.com Scientific Support Manager, Eastern Region Tel. (518) 482-1436 Biosym Technologies, Inc. Fax. (518) 482-5433 ------------------------------------------------------------------