From noy@tci005.uibk.ac.at Tue Nov 1 03:22:42 1994 Received: from uibk.ac.at for noy@tci005.uibk.ac.at by www.ccl.net (8.6.9/930601.1506) id DAA16076; Tue, 1 Nov 1994 03:14:47 -0500 Received: from tci005.uibk.ac.at by uibk.ac.at with SMTP id AA01197 (5.65c/IDA-1.4.4 for ); Tue, 1 Nov 1994 09:14:40 +0100 Received: by tci005.uibk.ac.at (AIX 3.2/UCB 5.64/CFIBK-2e.AIX) id AA11888; Tue, 1 Nov 1994 09:14:39 +0100 From: noy@tci005.uibk.ac.at (Noy) Message-Id: <9411010814.AA11888@tci005.uibk.ac.at> Subject: 100% parallel MD(2nd Summary) To: chemistry@ccl.net Date: Tue, 1 Nov 1994 09:14:39 +0100 (NFT) X-Mailer: ELM [version 2.4 PL23] Content-Type: text Content-Length: 24675 Dear fellows on the cyberspace, This is the summary of the responses I have got. Hope you enjoy it. Happy Halloween...... take care, Teerakiat ---------------------------- original post ------------------------- Dear cyberchemists, From the principle of molecular dynamics, velocity is reversible. Thus, if we assign a minus sign to all velocity of every particle in the simulation box ( at any equilibrated state ), the trajectory should be recoverd to the past correctly. This is not a new idea but I wonder if somebody has practiced it. For example if we simulate the phase space at time orgin t0 to the time t1, t0 ----------------> t1 (1 picosecond) Then, we revert velocity of all particles of the configuration at the time t0. Simulates it back to the past to the time t2 t0 ----------------> t2 (1 picosecond) Then we have the past trajectory. If we connect the two trajectories together, we will have t2 -------------------> t0 -----------------> t1 (2 picoseconds) Thus, we must revert the sign of velocity of the configurations >from the time t1 to the time t0. We will have the overall longer trajectory of the simulation. Given the run of two trajectories running in parallel on two computers, one can save time by the factor of 2. The dynamics and structural properties should be calculated >from the time origin t2 to the time t1. Does somebody have comments or suggestions on this ? Is it correct ? If this idea is practical, it would help to save time a lot. Imagine if I have to run MD for 10 ps to evaluate diffusion coefficient more precisely, I can divide the work into 2 jobs. One job is simulated on a computer. For another job, I convert the velocity and trace the phase space back to it past, have it run on another computer parallelly. Connect two jobs together by going back to the origin of the phase space in the far past time. Evaluate the time evolution and then I should get the results much faster. After all, the parallel efficiency is 100 % !!!!!!!! Thanks for your comments. best wishes, Teerakiat ---------------------------- end of original post ----------------------- ---------------------------- responses ----------------------------------- From: Richard A Caldwell Does the reversal risk numerical error? Small but numerous steps have roundoff error, and large steps have concomitant algorithmic inaccuracy. I am just wondering. dc ---------- From: gerson@VNET.IBM.COM From: Dennis J. Gerson, Ph.D. Technical Market Support; Chemical Sciences The parallel efficiency will NEVER be 100% because of setup time and shutdown time within the program (Amdahl's Law). The best you can do is to get less than 1% serial content in your code since you will need to have serial code to read in the starting information and write out the results (Amber is 97% parallel). Also, you must consider how you plan to partition the data and how the processes will communicate with each other. If the "scatter data and processes" and "gather data and processes" (either via fork/join on an SMP or send/receive on a DMP system) require more bandwidth than the communication bus can handle, then you reach the practical limit of parallelism. Remember, the algorithm may be 100% parallel in content, but the setup and shutdown are not, so the overall code is NOT 100% parallel. On top of this, the hardware/operating_system_software platform introduce constraints on parallel efficiency via bandwidth and compiler capability. Regards, Dennis Gerson --------------------------------------------------------------- IBM POWER Parallel Systems Division |Tele:(214)406-7267 Fax:7246 1503 LBJ Freeway MS/280750 |IBMLINK: USIB5MSK @ IBMMAIL Dallas TX 75234 USA |Email: gerson@vnet.ibm.com *** Forwarding note from SMTP2 --IINUS1 10/27/94 09:15 *** ========================================================================= ---------- From: "Gerald Loeffler" I see basically 2 problems with your suggestion: 1) Certain algorithmic tricks, like o cutting off long-range interactions beyond a certain radius o simulating with velocity scaling to keep the temperature constant are said to 'destroy reversibility'. As I understand it, this means that you can't go from state S1 to state S2, invert the velocities at state S2, go back the same time to state S1' and expect S1 = S1'. I'm really not sure if this spoils your idea: Phase space sampling is certainly not affected by this argument and should be perfectly compatible with your suggestion even when using cut-offs and velocity-scaling. Whether time correlations have an discontinuity at the point where you start simulating in two directions is not obvious. 2) The system has to be equilibrated, which it usually isn't when you start a MD. But this just means that you would have to simulate - say - 100ps as usual, and then apply your method. I think that the idea is very interesting, and would like to hear what other answers you got! Yours, Gerald -- Gerald Loeffler : gl@coil.mdy.univie.ac.at Institute for Theoretical Chemistry Theoretical Biochemistry Group University of Vienna Waehringerstrasse 17/Parterre A-1090 Wien, Austria ---------- From: DiveMaster Suffice it to say, I believe that this is the concept employed in parallel processing computers, although your particular application of "job-splitting" is equation (whatever you are doing) dependent. Brent Bettini NC School of Science & Math Durham, NC 27715 BETTINIC@ncssm-server.ncssm.edu ---------- From: case@scripps.edu (David Case) > > t0 ----------------> t2 (1 picosecond) > > Then we have the past trajectory. > If we connect the two trajectories together, we will have > > t2 -------------------> t0 -----------------> t1 (2 picoseconds) > Note that, because of round-off and integration errors (arising from a finite time step) with most integrators if you actually started at time t2, you would not reproduce the above trajectory. But that's probably not real important in most cases. After all, the parallel efficiency is > 100 % !!!!!!!! The idea is sound, but perhaps not worth eight explanation points :-). You do have to initialize two MD runs rather than one, so the actual efficiency is less than 100%. Plus you have to have a way to reverse the record of one of the trajectories, which can be cumbersome for long trajectories. But the main problem is that this "parallel" code only works on exactly two nodes. ...dave case ---------- From: claudio@itqb.unl.pt Dear Teerakiat I think that you are forgetting something in your scheme. The fact that the velocity is a function of the coordinates. In this way, these two steps that share the same absolute velocity, are only valid at one integration time step. After that, the coordinates of the system are no longer the same. Regards, Claudio ------------------------------------- Phone : (351-1) 442-6146 Ext. 237/341 Fax : (351-1) 442-8766 email : claudio@ctqb01.itqb.unl.pt Mailing address : Claudio Soares Instituto de Tecnologia Quimica e Biologica Rua da Quinta Grande 6 Apartado 127 2780 OEIRAS Portugal ---------- From: given@tiber.nist.gov (James A Given) MD trajectories are not reversible if they are longer than a few time steps. The reason is that the trajectories have nontrivial "mixing" properties and thus they lose information about the past. A theorist named Ronald Fox at Georgia Institute of Technology studied this question and wrote about it in Physical Review A (late 1970's ??). Do the experiment. It doesn't take very long. Run a trajectory. Reverse it. See what results. ---------- From: Ferenc.Molnar@chemie.uni-regensburg.de (Ferenc Molnar) Dear Teerakiat: Even if your integrator provides equations which are reversible in time, you should always worry about numerical precision. Try to check out if the integration from t1 with all velocities reversed leads to t0 and the same for t2 to t0. Otherwise your combined trajectory won't be correct. If this very nice idea turns out to be fruitfull, please let me know. Cheers, Ferenc Ferenc Molnar --------------------------------------------------------------------------- Institut fuer Physikalische und Theoretische Chemie - Lehrstuhl Prof. Dick - Tel.: (+49) 941 943-4466 /-4486 Universitaet Regensburg Fax.: (+49) 941 943-4488 Universitaetsstrasse 31 D-93053 Regensburg Deutschland / Germany --------------------------------------------------------------------------- EMail (SMTP): Ferenc.Molnar@chemie.uni-regensburg.de --------------------------------------------------------------------------- :-) ...THE PHYSICISTS HAVE MADE THEIR UNIVERSE, AND IF YOU DO NOT LIKE IT, YOU MUST MAKE YOUR OWN. -- E. R. HARRISON IN "COSMOLOGY" (1980) --------------------------------------------------------------------------- From: mutz@ifp.mat.ethz.ch Dear Klaus although I am quite new to the field of MD, I would like to share my thoughts on your posting with you. It is true that Newton's equation of motion are symmetric with respect to time, as they only contain second derivatives. However, these equations are most often modified in MD calculations. If one simulates a microcanonical ensemble (constant N,V and total energy of the system), there is no problem with your approach. But, and this is the more practical case, if you simulate a canonical ensemble (constant N,V and Temperature), you have to make sure that the temperature remains constant. Often, this is done by weak coupling of the system to a temperature bath. This introduces disspative terms into the equations of motion, which makes the reversal of the time direction impossible. Besides, if you are interested in longer trajectories to get better statistics of the properties you derive from your simulation, then you might alternatively start a second calculation from a different initial condition on a second machine. Kind regards, Marcel Utz. ------------------------------------------------------------------------------ Marcel Utz phone: +41 1 632 5672 Institute of Polymers fax: +41 1 632 1096 ETH-Zurich CNB E 98.2 CH-8092 Zurich, Switzerland email: Marcel.Utz@ifp.mat.ethz.ch ------------------------------------------------------------------------------ From: wallyr@netcom.com (Walter E. Reiher III) Noy: You wrote: >... > If we connect the two trajectories together, we will have > > t2 -------------------> t0 -----------------> t1 (2 picoseconds) > > Thus, we must revert the sign of velocity of the configurations > from the time t1 to the time t0. We will have the overall longer trajectory > of the simulation. Given the run of two trajectories running in parallel > on two computers, one can save time by the factor of 2. > The dynamics and structural properties should be calculated > from the time origin t2 to the time t1. > > Does somebody have comments or suggestions on this ? Is it > correct ? If this idea is practical, it would help to save time a lot. > Imagine if I have to run MD for 10 ps to evaluate diffusion coefficient >... Of course, you have to be certain that 't0' is NOT the beginning of your MD run, but the beginning of your "collection time", AFTER equilibration. Because of integration errors and rescaling of velocity for temperature equilibration, 'ts' (start of equibration) != 't2': ts -------------------> t0 then t2 <------------------- t0 t0 -----------------> t1 combine to give t2 -------------------> t0 -----------------> t1 (2 picoseconds) The idea makes sense to me, and I've heard people talk about doing this for years now, but I'm not sure if it's actually been done. This technique has, of course, been used to check the integration in MD runs: after equilibration, it should be possible to run 'forward' then 'backward' in time and get to where you began the 'collection' part (t0 == t2). ts -------------------> t0 t0 -----------------> t1 t2 <----------------- t1 For your application, a diffusion coefficient, it sounds reasonable to me; for other applications of MD, like conformational sampling/ equilibrium structure, it's better to use multiple starting structures and multiple parallel runs (where 'multiple' can be > 2). Good luck! Wally ======================================================================== Walter E. Reiher III, Ph.D. WallyR@netcom.com Consultant in Computational Chemistry P.O. Box 61056 voice 408-720-0240 Sunnyvale, CA 94088 fax 408-720-0378 --------------- From: K Bryson Hi Teerakiat, Took a while to work out what you were suggesting, but finally got it, and it's a very clever idea for parallising into two a single run. You would of course have to start with a state which had been equilibrated before hand, and this process could only be done in serial. After initial equilibration you would also have to perturb something or otherwise the backward time trajectory would just go back to your original starting ( unequilibrium ) state, well for classical dynamics anyway ( ignoring the effect of temperature adjustment, this may actually provide the perturbation that will make it produce a good equilibrium run backwards.) If the system was coupled to a heat bath then it wouldn't go back to the starting state (unequilibrated ), but this brings in the question whether the temperature coupling process is still valid for 'backward' trajectories in time ... certaining if it was temp. coupled your splitting process would not end up producing the same trajectory as a single 2t length trajectory starting from point t2, I think. So I think it's a good suggesting if you have an equilibrated state and you are running classical dynamics. Do not think it will work for temperature coupled systems, well, it wouldn't give you an identical path to the 2t length simulation, although it might be just as physically realistic for calculating time statistics data, not sure. What do you think ? Interesting idea though, Kev. www (o o) =======================---ooO-(_)-Ooo---===================================== K.Bryson email: kb7@tower.york.ac.uk Biophysics Group Tel : +44 904 430000 Extn. 2236 Physics Department Fax : +44 904 432214 University of York Heslington "Molecular modelling of DNA and its YORK, UK interaction with small molecules." YO1 5DD ============================================================================= From: sling@euclid.chem.washington.edu (Song Ling) You proposed that by reversing all momenta (velocities when they are proportional to momenta) you can connect two pieces of integrated trajectories together to form a longer one, unfortunately this is incorrect in general. Here is a simple example, the trajectory of a harmonic oscillator in phase space is a closed topological circle, and for an initial condition (x,p) in the first quadron (spell), when you reverse its momentum (velocity), you will end up in the 4th quadron, integrating the trajectory for a time dt, your first trajectory is still in the first quadron, while your another trajectory is still in the 4th quadron, the two pieces simply don't connect. The problem originated from that you didn't think about the trajectories as phase space objects (flows), you had a coordinate space picture. It is indeed possible to connect two pieces of trajectories to form a longer one in phase space, the way to do it is to integrate backward in time for -dt, and then integrate forward in time for dt, endding up a trajectory of "length" 2dt. Do not reverse the signs of the momenta (velocities) of your initial point. (You stay at the same point in phase space.) But this approach may be error prone depending on how many time steps you integrate your trajectory, if your system is very sensitive to initial conditions, the chances are after you integrate a trajectory from point A to point B, your integrate backward from B (with -dt) you may not end up in A. You seemed to be a little confused with the concept of "time reversal symmetry", to be safe let's say you have a conservative system (non- dissipative), the total energy is a constant of the motion, time reversal symmetry says that looking at the trajectories in phase space you don't find the arrow of time. In the harmonic oscillator example above, you have a topological circel in the phase space, you can parameterize it such that a point on it runs counterclockwise or clockwise, corresponding to running time forward or backward. If you have a damped harmonic oscillator, your trajectory spirals into the origin, the fixed point, now you have an arrow of time, past is mapped into the future. In short, you do not apply time reversal to join two pieces of trajectories together; you integrate forward and backward in time to do that. I don't know if that will save you computer time. ----------------- From: Gustavo Mercier On Thu, 27 Oct 1994, Noy wrote: ... > This is not a new idea but I wonder if somebody has > practiced it. For example if we simulate the phase space at time > orgin t0 to the time t1, > > t0 ----------------> t1 (1 picosecond) > > Then, we revert velocity of all particles of the configuration > at the time t0. Simulates it back to the past to the time t2 > > t0 ----------------> t2 (1 picosecond) > > Then we have the past trajectory. > If we connect the two trajectories together, we will have > > t2 -------------------> t0 -----------------> t1 (2 picoseconds) > > Thus, we must revert the sign of velocity of the configurations > from the time t1 to the time t0. We will have the overall longer trajectory > of the simulation. Given the run of two trajectories running in parallel > on two computers, one can save time by the factor of 2. > The dynamics and structural properties should be calculated > from the time origin t2 to the time t1. Hi! I think there is one problem. By following your prescription you fail to properly sample the phase space. In fact it looks like you sample the same phase space backwards and forwards, hence your computation of dynamic and structural properties will be off. The key is not the length of the simulation, but the sampling of the phase space. If you do so in one picosecond, a one picosecond trajectory will be adequate to compute the property of interest. I recommend an article by van Gunsteren and Berendsen (Angew. Chem. Int. Ed. Engl. 29, (1990) 992-1023. He argues that your simulations should not sample those degrees of freedom that are averaged within the time frame of the property of interest. Instead the effect of such averaged motions should be included through some potential of mean field. In this way you don't have to waste simulation time in sampling the phase space so that the proper averaging occurs. In other words, don't do md simulations if your interest is molecular brownian motions. Good Luck! Gustavo A. Mercier, Jr. mercie@cumc.cornell.edu ------------------- From: Martyn Winn Teerakiat, My feeling is that your scheme should work, provided you also reverse quantities such as the 3rd derivative of the positions (needed in a predictor-corrector algorithm). It should then be entirely equivalent to running the simulation backwards in time. With regard to Song Ling's objection, he is right that the two trajectories do not connect in phase space after the transformation p -> -p (to take his harmonic oscillator example). However, my understanding is that you would make the transformation -p -> p after the run, thus reconnecting the trajectories. Martyn -- ***************** Dr. Martyn Winn ************************************** \ / \ Institut fuer Theoretische Physik, Technische Universitaet Wien, / \ Wiedner Hauptstrasse 8-10, A-1040 Wien, Austria. / \____ ____________/ \ Tel: +43 1 58801 5678 / \ Fax: +43 1 5867760 / \ E-mail: winn@tph12.tuwien.ac.at / \ / -------------------------------------- -------------- From: sling@euclid.chem.washington.edu (Song Ling) Dr. Winn's reconnection scheme is definitely correct, and in principle it is equivalent to integrating backward as I suggested earlier. One needs the 3rd derivatives because predictor-corrector is not self-starting (unless one uses Runge-Kutta which is slow). Two questions remain, one is computational: Will the -dt, dt approach be more efficient than the "conventional" 2dt forward method? The other is "fundamental": Will the -dt, dt approach generate the "same" trajectory as the 2dt one if the system is sensitive to initial conditions? I'd like to learn about the answers. ----------------- From: Martyn Winn Two points: 1) My impression is that there is much work devoted to chaotic trajectories for small molecules (i.e. with a small no. of degrees of freedom) but that the issue is ignored for MD simulations of bulk systems. This is, of course, purely for reasons of convenience. 2) If one is in a chaotic regime, then I'm sure that the -dt, dt trajectory will differ from the 2dt trajectory, but taken individually each would be a perfectly valid trajectory. There would be no reason to prefer one over the other. In fact, a chaotic trajectory is surely ideal to satisfy the ergodic requirements of most MD simulations. I wonder if anyone is answering the original question about efficiency?! Martyn Winn -- ***************** Dr. Martyn Winn ************************************** \ / \ Institut fuer Theoretische Physik, Technische Universitaet Wien, / \ Wiedner Hauptstrasse 8-10, A-1040 Wien, Austria. / \____ / \ Tel: +43 1 58801 5678 / \ Fax: +43 1 5867760 / \ E-mail: winn@tph12.tuwien.ac.at / \ URL: http://tph.tuwien.ac.at/~winn/me.html / -------------------------------------------------- ---------------- From: Gustavo Mercier Hi! In light of recent messages, I reviewed the message I sent you. I realized that I made a mistake that invalidates my argument. You reverse the simulation at t0 to go to t2, not at t1! Hence, you would sample a different area of the phase space. Good Luck! Gustavo A. Mercier, Jr. mercie@cumc.cornell.edu ------------------- From: "Marais, Charles F. " > Will the > -dt, dt approach generate the "same" trajectory as the 2dt one if the > system is sensitive to initial conditions? This is very interesting. Has much work been done on chaos in MD ? Any pointers would be much appreciated. C Dr Charles Marais Department of Chemistry University of Cape Town Private Bag Rondebosch chuck@uctvax.uct.ac.za South Africa 7700 chuck@psipsy.uct.ac.za ------------------------------------------------------------------ From Klaus.Liedl@uibk.ac.at Tue Nov 1 06:22:42 1994 Received: from uibk.ac.at for Klaus.Liedl@uibk.ac.at by www.ccl.net (8.6.9/930601.1506) id FAA16659; Tue, 1 Nov 1994 05:24:39 -0500 Received: from tci002.uibk.ac.at by uibk.ac.at with SMTP id AA01771 (5.65c/IDA-1.4.4 for ); Tue, 1 Nov 1994 11:24:37 +0100 Received: by tci002.uibk.ac.at (AIX 3.2/UCB 5.64/CFIBK-2e.AIX) id AA14395; Tue, 1 Nov 1994 11:24:36 +0100 From: Klaus.Liedl@uibk.ac.at (Klaus R. Liedl) Message-Id: <9411011024.AA14395@tci002.uibk.ac.at> Subject: Re: "parallel" MD To: chemistry@ccl.net, noy@tci002.uibk.ac.at (Teerakiat Kerdcharoen, phone 507/5163) Date: Tue, 1 Nov 1994 11:24:35 +0100 (NFT) X-Mailer: ELM [version 2.4 PL23] Content-Type: text Content-Length: 1364 Dear Noy, Just a few hints to you: 1. It is trivial, that changing the sign of the time-coordinate is equivalent to changing the sign of the velocity coordinates in a CONSERVATIVE system (actually it is valid a little bit more general), as long as you remember that you have to sample with the "forward" velocities of your backward "simulation". 2. The problems you get with things like temperature bath, Cutoffs, ... are just the same as in the forward direction. (You do not get trajectories in either direction.) 3. There is nothing like a "equilibrium" in a microcanonical ensemble ... (The process you call equilibration is just a complicated methode to find a starting point in phase-space. It has nothing to do with a trajectory or a simulation.) 4. You do not need ONE trajectory for your sampling. You can sample over as many trajectories as you like. The only problem is to find starting points for your trajectories (cf. 3.). Even your methode in principle gives you ONE "MD-trajectory" (with the limitations of 2.) this is not the real point. The time you save is not the time for the actual simulation, but the time for searching a second starting point in phase space. Good luck Klaus -- (Klaus.Liedl@uibk.ac.at) -------------------------------------- LinuX the choice of the GNU-generation From winn@tph12.tuwien.ac.at Tue Nov 1 07:22:44 1994 Received: from tph12.tuwien.ac.at for winn@tph12.tuwien.ac.at by www.ccl.net (8.6.9/930601.1506) id GAA17359; Tue, 1 Nov 1994 06:59:32 -0500 Message-Id: <199411011159.GAA17359@www.ccl.net> Received: by tph12.tuwien.ac.at (1.37.109.4/16.2) id AA17830; Tue, 1 Nov 94 13:02:34 +0100 From: Martyn Winn Subject: CCL:100% parallel MD To: chemistry@ccl.net Date: Tue, 1 Nov 94 13:02:34 MEZ Mailer: Elm [revision: 70.85] Many thanks Robert for your interesting discussion. I have to admit to being slightly out of my depth here - chaos is not my field. However, I'd like to make one more point about this. > However, if you also want dynamic properties (autocorrelation > functions?), the distinction between a (-dt,dt) trajectory and a > (0,2t) trajectory could be relevant. I think this should be ironed > out before sweating out how to make the calculation efficient. > The (-dt,dt) and (0,2t) trajectories are distinct because of cumulation of errors, using the normal chaos argument. In fact, both are "wrong" because of this - neither represents an accurate trajectory. So again, why should one choose one over the other, EVEN when calculating dynamic properties? To put it another way, when calculating dynamic properties in a "normal" forward simulation, one is quite happy to use an incorrect trajectory - why not use a different incorrect trajectory? Even time correlation functions are averaged over phase space, so assuming ergodicity things should be OK. This seems relevant here: > The Lyapunov exponent for the > starting point in phase space will give the same "divergence rate" > whether one integrates asymmetrically (0,2t) or symmetrically (-t, t)... > either numerically generated trajectory is just as "wrong" > (or just as "right" depending on your point of view). Martyn -- ***************** Dr. Martyn Winn ************************************** \ / \ Institut fuer Theoretische Physik, Technische Universitaet Wien, / \ Wiedner Hauptstrasse 8-10, A-1040 Wien, Austria. / \____ ___/ \ Tel: +43 1 58801 5678 / \ Fax: +43 1 5867760 / \ E-mail: winn@tph12.tuwien.ac.at / \ URL: http://tph.tuwien.ac.at/~winn/me.html / ----------------------------------------------- From Jonathan.Connor@mailhost.mcc.ac.uk Tue Nov 1 07:26:37 1994 Received: from nessie.mcc.ac.uk for Jonathan.Connor@mailhost.mcc.ac.uk by www.ccl.net (8.6.9/930601.1506) id GAA16995; Tue, 1 Nov 1994 06:31:35 -0500 Message-Id: <199411011131.GAA16995@www.ccl.net> Received: from mailhost.mcc.ac.uk by mailhost.mcc.ac.uk id <16267-0@mailhost.mcc.ac.uk>; Tue, 1 Nov 1994 11:29:45 +0000 Subject: Theoretical Chemistry Days No. 2 To: CHEMISTRY@ccl.net Date: Tue, 1 Nov 1994 11:29:42 +0000 (GMT) X-Mailer: ELM [version 2.4 PL21] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 2123 From: Jonathan Connor Sender: Jonathan.Connor@mailhost.mcc.ac.uk Forwarded message: Delivery-Date: Fri, 28 Oct 1994 15:59:07 +0000 Message-Id: <14349.9410281539@tcibm2.mols.sussex.ac.uk> From: peterk@tc.mols.susx.ac.uk Subject: tcd2 To: J.N.L.Connor@manchester.ac.uk (Jonathan Connor) Date: Fri, 28 Oct 1994 15:39:06 +0000 (GMT) Reply-To: P.J.Knowles@susx.ac.uk Content-Type: text Content-Length: 1296 Royal Society of Chemistry: Theoretical Chemistry Group and Collaborative Computational Project No. 1 Theoretical Chemistry Days No. 2: Recent Advances in Electronic Structure A half-Day meeting to be held in the Department of Chemistry, University College, London on Wednesday 30th November 1994, >from 13.30 to 17.10. Programme: 1330 Chairman's Introduction J. N. L. Connor (University of Manchester) 1340 Keynote Lecture: Quantum-mechanical Macromolecular Modelling, I. H. Hillier (University of Manchester). 1430 Theoretical Modelling of Organic Reaction Mechanisms, I. H. Williams (University of Bath). 1455 Tea 1530 Plenary Lecture: The Multiconfigurational Approach to the Electronic Structure Problem, B. O. Roos (University of Lund) 1620 The Structure of Ionic Rare-Gas Clusters, P. J. Knowles (University of Sussex). 1645 Recent Advances with Density Functional Theory, N. C. Handy (University of Cambridge). 1710 Close Organiser: Peter Knowles, School of Chemistry and Molecular Sciences, University of Sussex; email P.J.Knowles@Sussex.ac.uk, telephone +44-1273-678624. Local organiser: Dr. Sally Price, University College London. Programme available on WWW as http://tcibm.mols.susx.ac.uk/tcd2/tcd2.html -- ****************************************************************************** Professor J.N.L. Connor, Phone(direct line): 061-275-4693 (national) Department of Chemistry, :+44-61-275-4693 (international) University of Manchester, Manchester M13 9PL, Phone(secretary): 061-275-4686 or 4600 England. Fax: 061-275-4734 or 4598 ****************************************************************************** From steve@chem.columbia.edu Tue Nov 1 09:22:47 1994 Received: from cucbs.chem.columbia.edu for steve@chem.columbia.edu by www.ccl.net (8.6.9/930601.1506) id JAA19860; Tue, 1 Nov 1994 09:07:24 -0500 Received: from amazon.chem.columbia.edu.columbia.edu (amazon.chem.columbia.edu) by cucbs.chem.columbia.edu with SMTP id AA10383 (5.65c/IDA-1.4.4 for ); Tue, 1 Nov 1994 09:11:42 -0500 Date: Tue, 1 Nov 1994 09:11:42 -0500 From: Steve Stuart Message-Id: <199411011411.AA10383@cucbs.chem.columbia.edu> Received: by amazon.chem.columbia.edu.columbia.edu (4.1/SMI-4.1) id AA27203; Tue, 1 Nov 94 09:11:32 EST To: chemistry@ccl.net Subject: Re: "parallel" MD Klaus.Liedl@uibk.ac.at (Klaus R. Liedl) wrote: >4. You do not need ONE trajectory for your sampling. You can sample > over as many trajectories as you like. The only problem is to find > starting points for your trajectories (cf. 3.). > Even your methode in principle gives you ONE "MD-trajectory" (with the > limitations of 2.) this is not the real point. > The time you save is not the time for the actual simulation, but the > time for searching a second starting point in phase space. and a few other people have made similar comments, stating that you could parallelize over N processors, if you could just find N different starting points. I just thought I'd point out that for many interesting properties (diffusion constant, residence times, etc) the the total length of the run can be more important than the number of runs averaged over. So it is important that the forward and backward "trajectories" have the same starting point. One other comment that interested me is K Bryson's note that the reverse run would begin to unequilibrate the system. (Now why didn't that occur to me...) Certainly true if you started from some ordered state and never adjusted the temperature, but any T-scaling would prevent this. Out of curiosity - does anyone know how reversible a non-T-scaling MD run actually is? (with/without cutoffs, SHAKE, whatever) -Steve Stuart steve@chem.columbia.edu echo "a'rfg cnf Crpv | har" | tr '[a-m][n-z] ' '[n-z][a-m]\012' | sort | tr "\012" " " ; echo " " From jmolmod@organik.uni-erlangen.de Tue Nov 1 09:27:58 1994 Received: from faui45.informatik.uni-erlangen.de for jmolmod@organik.uni-erlangen.de by www.ccl.net (8.6.9/930601.1506) id IAA18937; Tue, 1 Nov 1994 08:36:05 -0500 Received: from derioc1.organik.uni-erlangen.de by uni-erlangen.de with SMTP; id AA09355 (5.65c-6/7.3w-FAU); Tue, 1 Nov 1994 14:36:04 +0100 Received: by organik.uni-erlangen.de; id AA12120 (5.64/7.3n-FAU); Tue, 1 Nov 94 14:36:03 +0100 From: Journal of Molecular Modeling Message-Id: <9411011336.AA12120@derioc1.organik.uni-erlangen.de> Subject: Call for papers, J. of Molecular Modeling To: chemistry@ccl.net Date: Tue, 1 Nov 94 14:36:02 MET X-Mailer: ELM [version 2.3 PL11] *********************************** * * * Journal of Molecular Modeling * * * *********************************** As already announced the JOURNAL OF MOLECULAR MODELING will become available on January 1st 1995. As the first electronic journal in the field, the JOURNAL OF MOLECULAR MODELING is able to offer a unique combination of short publicationtimes and full color illustrations with no page charges and with subscription rates that are among the lowest available. Naturally, the success of the JOURNAL OF MOLECULAR MODELING depends heavily on being able to attract high quality papers from recognized experts. Knowing that they are among the subscribers of this list, we invite everybody, who is interested in, to contribute a paper on any subject (methodology or applications) within the molecular modeling field. The JOURNAL OF MOLECULAR MODELING is a fully refereed journal that will also be covered by the usual abstract services. It will be published initially on two FTP-servers (one in the US and one in Switzerland) and then made available both as a CD-rom and as a classical printed volume (from Springer, Heidelberg) at the end of each year. Graphical abstracts will also be generally accessible on WWW sites in Europe and the USA. Since the publication times (for full papers and reviews) will be significantly shorter then for conventional journals there is still time to submit papers to appear in January. Tim Clark, editor, Erlangen 1994. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^ Journal of Molecular Modeling ^ Tel: [+49](0)9131 / 85-2948 ^ ^ Computer-Chemie-Centrum ^ [+49](0)9131 / 85-6581 ^ ^ Universitaet Erlangen-Nuernberg ^ ^ ^ Naegelsbachstrasse 25 ^ Fax: [+49](0)9131 / 85-6565 ^ ^ D-91052 Erlangen ^ ^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^ email: jmolmod@organik.uni-erlangen.de ^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ From steve@carbo.cc.binghamton.edu Tue Nov 1 12:22:54 1994 Received: from carbo.cc.binghamton.edu for steve@carbo.cc.binghamton.edu by www.ccl.net (8.6.9/930601.1506) id LAA23074; Tue, 1 Nov 1994 11:32:46 -0500 Received: by carbo.cc.binghamton.edu (931110.SGI/931108.SGI.ANONFTP) for chemistry@ccl.net id AA17083; Tue, 1 Nov 94 11:17:43 -0500 Date: Tue, 1 Nov 1994 11:07:07 -0500 (EST) From: Steven Schafer Subject: Q: GAMESS Orbitals To: chemistry@ccl.net Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Is there a way to put 3P orbitals on Oxygen atoms in GAMESS? If there is no easy way to do it through GAMESS can the 3P orbitals be added by directly modifing the basis sets, and if so, where can I find the appropriate exponents? After all my questions have been asked, I would like to announce the Chemistry Mosaic site at the State University of New York at Binghamton. The site is still under construction, but I believe it be fairly good. The address to connect to is: http://chemiris.cc.binghamton.edu:8080 Any help with the afore mentioned questions would be of greatly appreciated. Thanks in advance, Steven Schafer S.U.N.Y. Binghamton Chemistry Department Binghamton, New York From theochem@ctc.com Tue Nov 1 14:23:06 1994 Received: from server1.ctc.com for theochem@ctc.com by www.ccl.net (8.6.9/930601.1506) id NAA25721; Tue, 1 Nov 1994 13:48:51 -0500 Received: by server1.ctc.com (5.65/DEC-Ultrix/4.3) id AA09988; Tue, 1 Nov 1994 13:48:35 -0500 Received: by pauling.ctc.com (931110.SGI/930416.SGI.AUTO) for @server1.ctc.com:chemistry@ccl.net id AA18256; Tue, 1 Nov 94 13:48:31 -0500 From: theochem@ctc.com (Douglas Smith) Message-Id: <9411011848.AA18256@pauling.ctc.com> Subject: predicting degradation products To: chemistry@ccl.net (Computational Chemistry List) Date: Tue, 1 Nov 1994 13:48:24 -0500 (EST) X-Mailer: ELM [version 2.4 PL22] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 2230 Dear Netters: I need to find one or more programs which can, given a single component or mixture of components (organic and inorganic), the temperature and pressure, calculate the products and percentages of the thermodynamic products of complete oxidative destruction as well as the energy used/released. I have two programs, BLAKE and McVEE, which do something of this sort. They rely on a library of formula and heats of formation, and can calculate the products under a variety of conditions (constant volume adiabatic, constant T and V, constant T and P). These are rather simplistic but may be good enough, if I can expand the libraries (such as by doing MOPAC/AMPAC calculations). But I want to know what else is out there. As an extension of this, what programs would users suggest I use if I want to examine the mechanism of oxidative destruction of organics and inorganics in the presence of molten and gaseous sulfur? My only idea to date is to perform MD calculations in a bath of sulfur at its melting point, using a quantum mechanical engine rather than a molecular mechanics engine to drive the simulation. Thus, I could look at arrays of atoms without requiring a specific bonding scheme. The only program I know of that will permit this is Hyperchem. Comments and suggestions are most welcome. As always, I will summarize to the net. Doug -- Douglas A. Smith Theoretical Chemist Concurrent Technologies Corporation 1450 Scalp Avenue Johnstown, PA 15904 voice: (814) 269-2545 fax: (814) 269-2798 email: theochem@ctc.com Stadnard Disclamur: All opinions, comments, mistakes, endorsements and odd noises are my own, not my employer's. +---------------------------------+---------------------------------+ | "The juvenile sea squirt wanders through the sea searching for | | a suitable rock or hunk of coral to cling to and make its home | | for life. For this task it has a rudimentary nervous system. | | When it finds its spot and takes root, it doesn't need its | | brain any more so it eats it. It's rather like getting tenure." | | --source unknown | +-------------------------------------------------------------------+ From ZUILHOF@CHEM.CHEM.ROCHESTER.EDU Tue Nov 1 17:22:51 1994 Received: from chem.chem.rochester.edu for ZUILHOF@CHEM.CHEM.ROCHESTER.EDU by www.ccl.net (8.6.9/930601.1506) id RAA00133; Tue, 1 Nov 1994 17:16:32 -0500 From: Received: from CHEM.CHEM.ROCHESTER.EDU by CHEM.CHEM.ROCHESTER.EDU (PMDF V4.2-14 #6313) id <01HIZ2TYP4AO000CHO@CHEM.CHEM.ROCHESTER.EDU>; Tue, 1 Nov 1994 18:13:20 EDT Date: Tue, 01 Nov 1994 18:13:20 -0400 (EDT) Subject: ROHF versus UHF in semiempirical calc.'s of radical cations To: chemistry@ccl.net Message-id: <01HIZ2TYPNKY000CHO@CHEM.CHEM.ROCHESTER.EDU> X-VMS-To: IN%"chemistry@ccl.net" X-VMS-Cc: ZUILHOF MIME-version: 1.0 Content-type: TEXT/PLAIN; CHARSET=US-ASCII Content-transfer-encoding: 7BIT Dear CCl'ers, Semiempirical programs such as MOPAC allow the properties of radicals and radical ions to be calculated with both the Restricted Open-shell Hartree-Fock method (ROHF) and the Unrestricted Hartree-Fock (UHF) methods. I want to calculate reactionpaths for nucleophilic attack on a series of radical cations. Does anyone know of 1) any apriori reasons why one method might be preferred over the other in such calculations? 2) literature data in which the performance of the ROHF and UHF methods for the study of radical cations is compared directly? Please report directly to me, and I'll summarize to the net. Thanks in advance, Han Zuilhof