From chemistry-request@ccl.net Sat Jul 6 18:32:30 1991 Date: Sat, 6 Jul 91 14:56:01 PDT From: ross@zeno.mmwb.ucsf.EDU (Bill Ross) To: chemistry@ccl.net Subject: parallel mol mech Status: R I am collecting information on on parallel processing in molecular mechanics/dynamics simulation: algorithms, anecdotes and references. All architectures are of interest, including shared memory (e.g. Cray, Iris), SIMD (e.g. Connection Machine, MasPar) and MIMD (e.g. hypercube). I will summarize to the reflector. For example, here are two ways of allocating processors for nonbonded calcs on MIMD architectures (such as hypercubes): Each processor gets some atoms, and communicates atomic positions to all the other processors in a global broadcast or a series of message passings. How much trouble is it to get any savings from nonbonded cutoff, pairlist-style tracking in this scheme? Each processor is assigned to a volume so pairlisting is inherent; if the volume is sized to fit the cutoff, only neighbors need to communicate. Is it worth tweaking this for irregular systems? Has anyone derived the formulas for optimizing problem distribution based on per-processor memory, communication startup/message_size time, processing speed, degree of vectorization, number of atoms etc? Bill Ross From chemistry-request@ccl.net Sat Jul 6 20:11:17 1991 Date: Sat, 06 Jul 91 19:35:19 EST From: "Mr. Mingzuo Shen" Subject: requesting help on the COLUMBUS ab initio program system To: chemistry@ccl.net Status: R Hello, I have just become aware of the distribution of the COLUMBUS ab initio quantum chemistry program system, developed by Profs. I. Shavitt, R.M. Pitzer, R. Shepard and H. Lishka (spelling?) >from three prominent centers of ab initio studies. This whole program system is available on the internet via anonymous FTP! (ftp.tcg.anl.gov; which is Prof. R. Shepard's site) I have three questions regard this system. 1. The system directly supports a number of computer platforms. One of the platforms supported is Sun and I have compiled the whole thing without a single error. But when I ran the test case (CH2) I found that the sequential write subroutine seqwbf (in colib) causes Seqmentation Violation. I found that when the buffer length is 512 or 20 this subroutine is OK. However, in the program system a buffer of 2048 is used. My question: has anybody successfully compiled COLUMBUS on the Sun -- would you please tell me about this sequential writer error (via e-mail please)? The Q&A guy (from computer science) does not know about Fortran. 2. Sun computers are of course a little slow. I have access to IBM RS6000 computers, and has anybody compiled the COLUMBUS system on one of these -- are you willing to share more information on porting? 3. IBM RS6000 computers are also a little slow and at our group are always crowded by queued jobs. I am EXTREMELY interested in seeing the system running on an IBM 3090 running VM/CMS or AIX/370 (version 1.x I think). I wish anybody who has successfully implemented COLUMBUS on the IBM 3090 (CMS or AIX/370) will help me!!! I am a beginner at ab initio programs. There are at least some documentation aimed at programmers that go with the COLUMBUS system. I expect that if I succeed in compiling the COLUMBUS system, I will have better luck with compiling our own programs on the IBM 3090 (AIX/370). Thanks in advance. Sincerely, Mingzuo Shen From chemistry-request@ccl.net Sat Jul 6 22:43:00 1991 Date: Sat, 6 Jul 91 22:29:58 EDT From: Kevin Moore To: chemistry@ccl.net Subject: MNDO/AM1/PM3 parameters... Status: R I am looking for semi-empirical (MNDO/AM1/PM3) parameters for Ru and Re. Has anyone parameterized these atoms? Any information would be greatly appreciated. Kevin Moore North Carolina Supercomputing Center (919)248-1179 kmoore@ncsc.org