From p.grootenhuis@organon.akzonobel.nl Mon Mar 4 10:47:58 1996 Received: from gatekeeper.oss.akzonobel.nl for p.grootenhuis@organon.akzonobel.nl by www.ccl.net (8.7.1/950822.1) id KAA08785; Mon, 4 Mar 1996 10:13:02 -0500 (EST) Received: (from mail@localhost) by gatekeeper.oss.akzonobel.nl (8.6.12/8.6.12) id QAA12003 for ; Mon, 4 Mar 1996 16:16:13 +0100 Received: from apou01.akzonobel.nl(145.49.90.59) by gatekeeper.oss.akzonobel.nl via smap (V1.3) id sma013477; Mon Mar 4 16:15:44 1996 Received: by apou01.akzonobel.nl (8.7.1/AKZONOBEL-WvdL/951013) id QAA01042; Mon, 4 Mar 1996 16:11:10 GMT Received: (from groot@localhost) by organon.akzonobel.nl (950511.SGI.8.6.12.PATCH526/8.6.12) id QAA27556 for chemistry@www.ccl.net; Mon, 4 Mar 1996 16:12:52 GMT From: Peter Grootenhuis Message-Id: <199603041612.QAA27556@organon.akzonobel.nl> Subject: MD of non-globular proteins To: chemistry@www.ccl.net Date: Mon, 4 Mar 1996 16:12:50 +0000 (WET) Name: P.Grootenhuis Organisation: NV Organon Phone: (+31)0412-661920 X-Mailer: ELM [version 2.4 PL24 ME8b] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit CCL-ers, I want to carry out molecular dynamics calculations on proteins with a non-globular shape in order to qualitatively assess the effect of certain mutations. However, both in gas-phase and aqueous simulations there is a rather strong tendency of the system to adopt conformations that result in more globular shapes. I believe this effect is driven by the nonbonded interactions that ultimately promote globular shapes. Questions: (1) Is anybody aware of (published) MD simulations of non-globular proteins ? (2) Does anybody has (preferably computationally inexpensive) suggestions on how to handle such a system ? I will summarize and post to the List. Thanks very much, Peter Grootenhuis ______________________________________________________________________________ Dr. Peter D.J. Grootenhuis | N.V. Organon / CMC Dept. RK2337 | Phone : +31-412-661920 P.O. Box 20 / 5340 BH Oss | Fax : +31-412-662539 The Netherlands | E-mail : p.grootenhuis@organon.akzonobel.nl _________________________________|____________________________________________ From owner-chemistry@ccl.net Mon Mar 4 15:44:15 1996 Received: from bedrock.ccl.net for owner-chemistry@ccl.net by www.ccl.net (8.7.1/950822.1) id OAA14383; Mon, 4 Mar 1996 14:44:39 -0500 (EST) Received: from hrz-sun1.hrz.uni-kassel.de for gdanitz@hrz.uni-kassel.de by bedrock.ccl.net (8.7.1/950822.1) id OAA18681; Mon, 4 Mar 1996 14:44:37 -0500 (EST) Received: from hrz-serv1.hrz.uni-kassel.de by hrz-sun1.hrz.uni-kassel.de (4.1/SMI-4.1) id AA19722; Mon, 4 Mar 96 20:45:47 +0100 Received: by hrz-serv1.hrz.uni-kassel.de (AIX 3.2/UCB 5.64/HRZ-GhK/HRZ-SERV1/pm) id AA48801; Mon, 4 Mar 1996 20:44:28 +0100 From: gdanitz@hrz.uni-kassel.de (Robert Gdanitz) Message-Id: <9603041944.AA48801@hrz-serv1.hrz.uni-kassel.de> Subject: Number crunching in 16-byte precision. To: chemistry@ccl.net (Computational Chemistry List) Date: Mon, 4 Mar 1996 20:44:27 +0100 (MEZ) +-----+ Robert J. Gdanitz email: gdanitz@hrz.uni-kassel.de | GhK | Gesamthochschule Kassel Tel.: +(49) 561-804-4556 | | Fachbereich 18 (Physik) Fax: +(49) 561-804-4006 +-----+ 34109 Kassel X-Mailer: ELM [version 2.4 PL24 PGP3 *ALPHA*] MIME-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 8bit Content-Length: 3014 Dear Netters, I think it is justified to say that most of us are quite satisfied with, as well the pre- cision (approx. 2.22E-16 in IEEE) as the performance of the usual "working precision" arithmetic (8-byte, real*8, "double-precision" is, if at all, not much slower than 4-byte, real(*4), "single precision") in our routinely calculations which, in some cases, may take considerable time. I have heard that once there was a time, when the working precision was real*4 or something in between real*4 and real*8 which e.g. frequently gave rise to problems with numerical stability in quantum chemical calculations and double precision was prohibitively slow, but this now is history. Anyway, there are still people who want to push things to the limit (like me) who do their calculations in a way, that even real*8 precision is not sufficient. And the problems do not arise because I need to have, say 16 or more figures accuracy in the final result, or because from the algorithms available to solve a certain problem, due to some reason, I choose the one most prone to numerical instabilities. What I actually want to do, is to solve the Schroedinger equation of some small chemical system (e.g. N2) to "chemical accuracy" (1 kcal/mol) which still has not yet been achieved all (even "small") cases (e.g. the De of N2). One could give e.g. MR-CI-(SD) or CC(SD)[T] a try, but the convergence of De with respect to the increase of the size of the basis set (n) is VERY slow (~ n^-3). Recently, I managed to combine Kutzelnigg's and Klopper's "r12"-method with MR-CI(SD), resp. MR-ACPF to get a convergence ~ n^7, but there are still problems to be solved... The r12-method uses explicitly correlated terms which, when rather diffuse orbitals are joined by r12, are highly redundant. On the other hand, I cannot easily remove these terms because then problems with e.g. proper dissociation or size-extensivity may arise. So I have to do some fiddling with the CI- as well as with the r12-ansatz, i.e. I have to re- move e.g. orbitals with high energy and r12-terms which are not necessary and I always have to check that I am indeed allowed to do so, which makes things somewhat tedious. OK, now that I have these problems with numerical stability, why do I not simply switch to real*16 arithmetic and be happy? Well, I found out that on our only computer where I can dump some GByte temporary data (it's a SGI Power Challenge) real*16 arithmetic, when used for e.g. a dot-product of two vectors, is 300 times slower than real*8. Since I cannot afford to wait one year to compute what takes one day in real*8, I'm in serious trouble. On the other hand, our IBMs, which degrade only by about 30 in the present case, do not have enough temporary disc space available. Anyway, if there is anybody out there who has experience doing large scale computations in real*16, please tell me your experience with computers & algorithms. As usual, I will summarize to the net. Thanks in advance, Robert Gdanitz From owner-chemistry@ccl.net Mon Mar 4 17:32:20 1996 Received: from bedrock.ccl.net for owner-chemistry@ccl.net by www.ccl.net (8.7.1/950822.1) id QAA16087; Mon, 4 Mar 1996 16:47:04 -0500 (EST) Received: from theory.tc.cornell.edu for jeanne@tc.cornell.edu by bedrock.ccl.net (8.7.1/950822.1) id QAA22374; Mon, 4 Mar 1996 16:47:00 -0500 (EST) Received: from [128.84.181.75] (JEANNE.TC.CORNELL.EDU [128.84.181.75]) by theory.tc.cornell.edu (8.6.9/8.6.6) with SMTP id PAA89014; Mon, 4 Mar 1996 15:43:09 -0500 Date: Mon, 4 Mar 1996 15:43:09 -0500 Message-Id: Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" To: jeanne@TC.Cornell.EDU From: jeanne@TC.Cornell.EDU (Jeanne C. Butler) Subject: Cornell Theory Center IBM SP Workshop Cc: reynolds@TC.Cornell.EDU Workshop on Parallel Programming on the IBM RS/6000 SP Sunday April 28 - Thursday May 2, 1996 Cornell Theory Center Cornell University Ithaca, NY The Cornell Theory Center (CTC), a nationally funded high performance computing center, is offering four days of lecture and laboratory sessions on parallel programming for the IBM SP. CTC's SP, which consists of 512 RISC processors connected by a high performance switch, is the largest of its kind in the world. The SP programming model is distributed memory. This workshop will offer sessions on the following topics: -Introduction to CTC's IBM SP and Parallel Programming -Introduction to Performance Issues on CTC's IBM SP -Parallel Programming Using the Message Passing Interface (MPI) Library -Parallel Programming Using High Performance FORTRAN (HPF) (presented through a case study of a tariff modeling program) All topics will be presented using a mix of lectures and programming exercises, giving participants hands-on experience with parallel programming. A full agenda is appended to this announcement. The "Introduction to CTC's IBM SP and Parallel Programming" sessions are intended for beginning parallel programmers and users who are new to CTC. They cover basic information for running on the IBM SP, as well as the fundamentals of parallel processing and distributed memory computing. Experienced CTC users may wish to skip these sessions. All others should take these sessions, as the remainder of the workshop assumes participants are familiar with this information. The "Introduction to Performance Issues on CTC's IBM SP" sessions are intended for both beginning and intermediate distributed memory programmers. Participants will become familiar with the basic concepts and tools for performance modeling, timing, profiling, and tracing on the IBM SP. Basic strategies for improving the performance of a program on the IBM SP will also be presented. The workshop sessions on "Parallel Programming Using the Message Passing Interface (MPI) Library" will focus on the MPI standard library, as implemented by IBM. These sessions will start with the six fundamental MPI procedure calls and will proceed to some more advanced issues, such as derived data types and persistent communication. The final sessions will present a case study of parallelizing a serial program using MPI. The sessions on "Parallel Programming Using High Performance FORTRAN (HPF)" will demonstrate parallelizing a serial program, using a case study approach. The program for the case study will be the same as that used for the MPI case study sessions. Performance analysis and optimization of a parallel program will also be demonstrated. (All trade names referenced are trademarks or registered trademarks of their respective companies.) REGISTRATION INFORMATION To apply, please complete the registration form, found at: http://www.tc.cornell.edu/Events/SP.Apr96.html An ASCII text registration form is available for FTP from: ftp.tc.cornell.edu Change to the pub directory and get file Apr96.workshop Send payment separately to arrive no later than March 25, 1996, to: Jeanne Butler Conference Assistant 427 Frank H. T. Rhodes Hall Ithaca, NY 14853-3801 Fees: per day full workshop Academic/Gov. $50 $200 Corporate $225 $795 CPP* $175 $595 (* members of Corporate Partnership Program) Make checks payable to Cornell University. Local applicants may charge the registration fee to the appropriate Cornell University account number. Registrations will not be acted upon until the payment arrives. Refunds will be made to those applicants not accepted to the workshop. Refunds cannot be made after an applicant is accepted. Course attendance is limited. Preference will be given to those who have already received a CTC allocation, to Corporate Partnership Program (CPP) members, and to those who have an application pending. It might be necessary to limit the number of attendees from any one research project. OTHER TRAINING OPPORTUNITIES Applicants should also be aware that CTC will be offering a Virtual Workshop (VW) over the summer months. The VW offers World Wide Web versions of most of the material covered in this workshop, and it includes interactive logins on the CTC IBM SP for completion of exercises. CTC staff members offer consulting support to the VW participants through e-mail and through the CTC MOO. More information on this offering will be posted through CTC's Education Calendar of Events on the World Wide Web. More information and a preview of the Virtual Workshop can be found at: http://www.tc.cornell.edu/Edu/VW/ For general information on current and future CTC workshops, go to: http://www.tc.cornell.edu/Edu/Upcoming/workshops.html SPRING 1996 IBM SP WORKSHOP AGENDA Sunday, April 28 (starting at 1:00 p.m.) Introduction to CTC IBM SP and Parallel Programming Registration and Logging In Introduction to the Workshop Introduction to the IBM SP at CTC Introduction to Parallel Processing Monday, April 29 Introduction to Parallel Programming, cont. Introduction to Distributed Memory Programming Parallel Program Design Introduction to Performance Issues on the CTC IBM SP Performance Basics Single-Processor Performance Tools Single-Processor Performance Considerations Tuesday, April 30 Parallel Programming with Message Passing Using MPI Introduction to MPI Basics of MPI Programming Point-to-Point Communication Collective Communication Advanced Topic: Derived Data Types Wednesday, May 1 Message Passing Using MPI, cont. Advanced Topic: Groups and Communicators Advanced Topic: Persistent Communication Advanced Topic: More on Collective Communication Parallel Processing Performance Tools Tariff Case Study: Introduction and MPI VERSION Part I: Problem Description and Serial Implementation Part II: MPI Implementation Tariff Case Study Using HPF Part III: HPF Introduction and Automatic Parallelization Part IV: Tuning Thursday, May 2 Tariff Case Study Using HPF, cont. Part V: F90 and HPF Directives Part VI: The HPF Data Model Part VII: Parallel Loops in HPF Part VIII: Summary and Conclusions (Concludes at 12:00 noon) (An open lab session will be offered during the afternoon.) ...end/sp.workshop.announce