From chemistry-request@server.ccl.net Thu Jun 13 03:14:00 2002 Received: from chem.ch.huji.ac.il ([132.64.90.35]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g5D7Dxj11037 for ; Thu, 13 Jun 2002 03:13:59 -0400 Received: from baikal (baikal.ch.huji.ac.il [132.64.98.52]) by chem.ch.huji.ac.il (8.9.3/8.9.3) with SMTP id KAA27217 for ; Thu, 13 Jun 2002 10:13:55 +0300 (IDT) Message-ID: <059401c212b2$311a21b0$34624084@ch.huji.ac.il> From: "David Danovich" To: Subject: G98: I want to save twi-eoectron integrals Date: Thu, 13 Jun 2002 10:13:25 +0200 MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="----=_NextPart_000_0591_01C212C2.F483F800" X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 6.00.2600.0000 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000 This is a multi-part message in MIME format. ------=_NextPart_000_0591_01C212C2.F483F800 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Dear all, When you are doing G98 calculations with cc-pVQZ basis set you are not = able to use scf=3Dnodirect keyword and not able to produce two electron = integrals int file. For some reason I need those integrals. = Unfortunately, no iop options for non standard job is suitable to help. = Is there any possibility to produce two-electron integrals file for = cc-pQZ basis in G98?=20 I will appreciate any help. Thanks David ____________________________________________________ Dr. David Danovich, Department of Organic Chemistry,=20 The Hebrew University, Givat-Ram Campus, 91904 Jerusalem, Israel=20 http://yfaat.ch.huji.ac.il/david.html, David.Danovich@huji.ac.il=20 FAX:(+972)-2-6585345, Phone:(+972)-2-6586934(w), (+972)-2-9995935(h) ------=_NextPart_000_0591_01C212C2.F483F800 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable

Dear all,

When you are doing G98 = calculations with=20 cc-pVQZ basis set you are not able to use scf=3Dnodirect keyword and not = able to=20 produce two electron integrals int file. For some reason I need those=20 integrals. Unfortunately, no iop options for non standard = job is=20 suitable to help.

Is there any possibility to produce = two-electron=20 integrals file for cc-pQZ basis in G98?

I will appreciate any = help.

Thanks David

____________________________________________________
Dr. = David=20 Danovich, Department of Organic Chemistry,
The Hebrew University, = Givat-Ram=20 Campus, 91904 Jerusalem, Israel
http://yfaat.ch.huji.ac.il= /david.html,=20 David.Danovich@huji.ac.il=20
FAX:(+972)-2-6585345, Phone:(+972)-2-6586934(w),=20 (+972)-2-9995935(h)

------=_NextPart_000_0591_01C212C2.F483F800-- From chemistry-request@server.ccl.net Thu Jun 13 01:32:23 2002 Received: from cpimssmtpu12.email.msn.com ([207.46.181.87]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g5D5WNj09860 for ; Thu, 13 Jun 2002 01:32:23 -0400 Received: from catchcafmsn ([68.129.127.202]) by cpimssmtpu12.email.msn.com with Microsoft SMTPSVC(5.0.2195.4905); Wed, 12 Jun 2002 22:31:30 -0700 Message-ID: <001f01c2129b$f3a06080$b07f8144@com> Reply-To: "Mauricio Cafiero" From: "Mauricio Cafiero" To: Subject: Maple 6 on Windows XP Date: Wed, 12 Jun 2002 22:34:05 -0700 X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 5.50.4807.1700 X-MimeOLE: Produced By Microsoft MimeOLE V5.50.4807.1700 X-OriginalArrivalTime: 13 Jun 2002 05:31:32.0550 (UTC) FILETIME=[939BFE60:01C2129B] hello: Does anyone know if Maple 6 will run in Windows XP? I would like to know before committing to buying the OS. Thanks. Mauricio "Squatch" Cafiero Doctoral Candidate: Theoretical and Computational Quantum Chemistry University Of Arizona From chemistry-request@server.ccl.net Thu Jun 13 04:24:09 2002 Received: from gems.uia.ac.be ([143.169.254.10]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g5D8O8j13000 for ; Thu, 13 Jun 2002 04:24:09 -0400 Received: from localhost (bswerts@localhost) by gems.uia.ac.be (8.12.3/8.12.3) with ESMTP id g5D8O3UU025185 for ; Thu, 13 Jun 2002 10:24:03 +0200 (MET DST) Date: Thu, 13 Jun 2002 10:24:03 +0200 (MET DST) From: X-Sender: To: CCL Subject: CHARMM benchmark outputs In-Reply-To: <200206101225.g5AC2eXR022372@gems.uia.ac.be> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Spam-Flag: NO X-Spam-Status: Not checked, trusted sender X-Scanned-By: MIMEDefang 2.9 Hi all, I tried to send the following message to the CHARMM mailing list, but it appears the server is not reachable. That's why I'll try on this list. > I'm using version c27b3 on an Alpha (running Tru64 Unix) and a PC (running > Redhat Linux). When comparing outputs from the testcases, I see slight > differences in the results. On the PC, when using the 2 different sets of > compiler options (gcc 2.96), I also see differences. > After a few thousand geometry optimization steps, the energy of the enzyme > I'm studying is off by a kcal... > > My question: where could I obtain the benchmark outputs? I need those for > alpha, gnu and possibly a third as another reference. > > Thanks a lot in advance. Ben --------------------------------------------------- Ben Swerts University of Antwerp - Dept. of Chemistry (C2.29) Universiteitsplein 1 B-2610 Wilrijk email: ben.swerts@ua.ac.be Belgium tel: +32-3-820.23.57 --------------------------------------------------- From chemistry-request@server.ccl.net Thu Jun 13 06:40:28 2002 Received: from salmon.maths.tcd.ie (mmdf@[134.226.81.11]) by server.ccl.net (8.11.6/8.11.0) with SMTP id g5DAeSj14818 for ; Thu, 13 Jun 2002 06:40:28 -0400 Received: from bell.maths.tcd.ie by salmon.maths.tcd.ie with SMTP id ; 13 Jun 2002 11:40:28 +0100 (BST) Date: Thu, 13 Jun 2002 11:40:28 +0100 (BST) From: Gemma Kinsella To: chemistry@ccl.net Subject: CCL: adding new atom types Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Hi all, I am adding a new atom type to Gromos and Charmm force fields and was wondering if anyone has any experience with this. Is there any way to obtain force constants using Gaussian 98? Any help would be greatly appreciated. Gemma From chemistry-request@server.ccl.net Thu Jun 13 06:49:52 2002 Received: from century.fen.bilkent.edu.tr ([139.179.97.100]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g5DAnoj14977 for ; Thu, 13 Jun 2002 06:49:51 -0400 Received: from fen.bilkent.edu.tr (fenII-121.fen.bilkent.edu.tr [139.179.97.239]) by century.fen.bilkent.edu.tr (8.9.1/8.9.1) with ESMTP id NAA08493 for ; Thu, 13 Jun 2002 13:46:01 +0300 (EET DST) Message-ID: <3D087958.8060103@fen.bilkent.edu.tr> Date: Thu, 13 Jun 2002 13:52:08 +0300 From: Ulrike Salzner Organization: Bilkent University User-Agent: Mozilla/5.0 (Windows; U; Win98; en-US; rv:0.9.4) Gecko/20011128 Netscape6/6.2.1 X-Accept-Language: en-us MIME-Version: 1.0 To: chemistry@ccl.net Subject: G98.memory - summary Content-Type: multipart/alternative; boundary="------------070603030809030003000302" --------------070603030809030003000302 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Dear all, thanks to everybody who responded to my request. Please find the original answers attached below. To summarize the essentials: First of all I interpreted the information of the "free" commond incorrectly. Linux uses most of the memory until another process requests it. The proper place to look for memory information is the second line starting with: -/+ buffers/chache. About using Gaussian: the most efficient procedure seems to be to estimate the memory a job requires and specify not much more than this amount using the "%mem" command. If too much memory is requested, even if it is less than the computer has, the performance will decrease. With more than one processors, the specified memory is per processor. Best regards, Ulrike Salzner Original message: Dear Compouter Experts, I am trying to run a big job (1260 basis functions) on a dual 933MHz PC. The operating system is RedHat 7.2, and there is 1 GB of memory. I have not specified memory in the default route since test jobs showed that the performance decreased considerably even if much less memory was requested than the machine has. Now the number of processors is reduced to 1 by G98 due of lack of memory. Checking with the top command, G98 is supposed to use only 5.8% of the memory. Gtop (memory usage resident) shows that there are three main users of memory: l502 (59756k), httpd (54312k), and nautilus (54780k). Sum of resident sizes is 260752k. In contrast the command "free" gives: Mem total: 1027928, used: 1022784, free 5144, shared 47864, buffers 100320 cached 771980 I do not understand where the memory is. Is there a way to tune the system better? Would it be better to not use a graphical interface? Or am I doing everything wrong? Any suggestions wil l be greatly appreciated. Sincerely, Ulrike Salzner > from Patrik Johansson: Dear Ulrike without knowing to many details there's the possibility that the Linux OS uses some considerable amount of the memory to speed up processes (like an extra cache in some way). This, however, will only happen if you do not specify the G98 job to use an extensive amount (like 850 if you really only need 450). So more memory doesn't always improve timing. If the run can be made with let's say 200Mb I usually specify like 350Mb and let the OS use the rest - this works for me. best regards /Patrik > from Laurence Cuffe: 1) I'm no expert! 2) performance should improve if you drop the graphic interface. 3) try using %Mem=250MB, this should allow a two processor job with both parts fitting into main memory, where they take 250MB each thus giving a total of 500 Mb. G98's memory usage can confuse red hat 7.2's memory management routines, running a 4 processor system with 4 g98 jobs each of which required less than 1/4 of the ava ilable ram I found that swap and memory management routines were taking 60% of one processors time, the solution was to provide more conservative memory allocations for each job. Another thing to note is that the memory requirement for a job with the following: %Nproc=4 %Mem=100MB is 100x4MB =400MB not 100MB in total as I had first supposed. I hope this helps All the best Laurence Cuffe > from Andreas Goetz: Dear Ulrike, I am not a computer expert nor am I using gaussian often. So I cannot tell much about gaussian. If you invoke "man httpd" you can inform yourself about httpd - Apache hypertext transfer protocol server. I don't know what else than calculations you are using the computer for, but I believe you don't want to shut down httpd. As far as I am informed, the memory which is given as buffers and cached evoking the command "free" is memory which is reserved by the system for speeding up certain pro cesses. This memory does not have to be used and can be freed in favour of a program which might request it. In other words: Linux will always occupy almost all the memory to cache instructions in the memory which might be required by programs. If a program like gaussian will need more physical memory, the system will reduce the cache memory and let gaussian use this part of the memory. The numbers which really should concern you are those in the line "-/+ buffers/cache" of the output from the "free" command. As a comfortable tool for monitoring system ressources you might also want to use "xosview". Kind regards, Andreas > from Fabien Fontaine: I guess most of the memory is cached, which mean that it is considered as used but actually it is like a buffer for future use. If you really want to know if you have memory problem, check if the swap memory is used. Regards, Fabien > from Serguei Patchkovskii: Ulrike, Try 'ipcs -ma' command - G aussian stores most of its data in a shared memory region. If the system has been up for a while, Gaussian tends to leave memory regions behind, making memory inaccessible. If you see any regions with 'nattch' of zero, it is usually safe to remove them with 'ipcrm'. This should free up some memory. Serguei > from Doug Fox: Prof. Salzner, Ignoring the question about tuning the system let me describe the way G98 allocates and uses memory. Gaussian uses dynamic allocation within a large block of memory which is allocated/released by each link. By default this block of memory is 6MW (48MB) long which is in reasonable agreement with your gtop output, some additional memory is used for code and local data. Within this block of memory some arrarys are fixed by the problem size, others can be adjusted to allow the calculation to succeed, sometimes at a cost of additional CPU due to recomputing integrals, but in each case there is a maximum about of disk wh ich is useful and a minimum below which it cannot be done. In the case of shared memory and a two processor system if the amount of memory is small enough that two+ passes would be needed to process all of the integrals the code drops down to 1 processor because that allows for a single long inner loop which is more efficient than the shorter loop over two processors. Remember that with two processors it would need to setup the loops twice, i.e. it nearly destroys the time saved by running on two processors. The fact that any given problem has a maximum effective memory is the reason for your observation that giving G98 an arbitrary large memory will slow down small problems. It takes time to process the allocation and deallocation of the memory so if it is not used it simply shows up as overhead. On the other hand giving the program more memory on a problem by problem basis, i.e. adding %mem=nnnmw at the top of the input, lets you increase memory for cases wh ich need it. There is a rough formula in chapter 4 of the G98 User's Reference which tells you the memory for a single processor taking account of problem size and the nature of the basis set and method. For a dual processor system you don't need to double all of this, just double the portion dependent on basis set and method. This is a very rough guide but it should be an over estimate and will let you use both processors. > from Cory Allen Rice: ULRIKE SALZNER ; YOU CAN USE THE %mem=XXXMb IN G98. THIS WILL GIVE YOU MORE MEMORY GO TO THE RUNNING OF A JOB ON G98. Corey A Rice Southern Illinois University at Edwardsville Chemistry Department corrice@siue.edu http://www.siue.edu/~corrice > from Bern Kallies: Hallo Ulrike, G98 will be compiled by default for a usage of 4mw (1 word = 8 byte, mega = 1024 kilo) of scratch memory. I think this is what you see in your memory usage statistics (59MB includes the size of the exe, its statically allocated memory and the dynamically allocated scratch memory of 32MB). Regarding your problem to get g98 running on more than 1 processor with your system, I'll try to explain the strategy of gaussian: When running parallel gaussian jobs (without LINDA), the master process of a gaussian link allocates scratch space (4mw by default, or as specified in the %mem card), and then it forks children which use this allocated scratch space as shared memory. In l502 scratch memory will be used mainly for handling integrals. When there is no possibility to hold a substantial amount of integrals in memory, they will be recomputed as needed. But some information about integrals must be stored in memory in any event. The size required for that depends on the number of basis functions, the maximum l value, the packing scheme of integrals and other t hings. There is no easy way to find out this amount. But what counts in your case is that every process stores these data private in the scratch area. Thatswhy more scratch space will be needed when running g98 in parallel. There exists an empirical rule for the amount of memory needed: mem(nproc) = mem(1proc) * (1 + 0.75*(nproc-1)) where nproc is the number of processors you want to use, and mem(1proc) is the minimal amount required to run a particular job on 1 processor. Gaussian checks if there is enough memory available when running in parallel. If it guesses that it will not be enough, then it reduces the number of processes until they fit in the available amount of memory, as in your case. So you should specify something greater than 4mw in your %mem card. How much? Is an interesting question. To my experience, gaussian becomes slower when it gets an amount of memory, which is larger than needed for a direct SCF, but which is not enough for incore SCF. When using abou t 120 basis functions, about 30mw are enough for incore. That means, using 8mw is slower than 4mw, with 16mw is slower than with 8mw, with 32mw is faster than with 4mw (switches to incore), with 64mw is as fast as with 16mw and so on. When using about 350 basis functions, one would need 2.7gw (about 20GB) for incore. Allocating only 2.6gw in the latter case would have the only effect that the job runs terribly long, longer than with 8mw. So in your case I would succesively try it with 8mw, 16mw, 32mw, until the job runs. When it runs already with 4mw on one processor, then 8mw would be enough according to the formula above. Hope this helps, best regards, Bernd > from Demetrio de Silva Filho: Hello Ulrike, There is something you must remember. The memory you specify is not what Gaussian will always use but the maximum memory Gaussian will use during the job. If you follow what is happening while you are running, you will see these numbers change, depending on the link. The problem with gaussian is : there are few parts of the program written in old fortran 77 that cannot alocate memory dynamicaly, i.e. as the process needs. So, you must specify in advance how much memory you have. Hope it helps! Demetrio FIlho --------------070603030809030003000302 Content-Type: text/html; charset=us-ascii Content-Transfer-Encoding: 7bit Dear all,

thanks to everybody who responded to my request. Please find the original answers attached below.
To summarize the essentials: First of all I interpreted the information of the "free" commond incorrectly. Linux uses most of the memory until another process requests it. The proper place to look for memory information is the second line starting with: -/+ buffers/chache.
About using Gaussian: the most efficient procedure seems to be to estimate the memory a job requires and specify not much more than this amount using the "%mem" command. If too much memory is requested, even if it is less than the computer has, the performance will decrease. With more than one processors, the specified memory is per processor.

Best regards,
Ulrike Salzner

Original message:

Dear Compouter Experts,

I am trying to run a big job (1260 basis functions) on a dual 933MHz PC.
The operating system is RedHat 7.2, and there is 1 GB of memory. I have
not specified memory in the default route since test jobs showed that the
performance decreased considerably even if much less memory was requested
than the machine has.  
Now the number of processors is reduced to 1 by G98 due of lack of memory.
Checking with the top command, G98 is supposed to use only 5.8% of the
memory. Gtop (memory usage resident) shows that there are three main
users of memory: l502 (59756k), httpd (54312k), and nautilus (54780k). Sum
of resident sizes is 260752k. In contrast the command "free" gives:
Mem total: 1027928, used: 1022784, free 5144, shared 47864, buffers 100320
cached 771980
I do not understand where the memory is. Is there a way to tune the
system better? Would it be better to not use a graphical interface? Or am
I doing everything wrong? Any suggestions wil

l be greatly appreciated.
Sincerely,
Ulrike Salzner

from Patrik Johansson:

Dear Ulrike

without knowing to many details there's the possibility that the Linux
OS uses some considerable amount of the memory to speed up processes
(like an extra cache in some way). This, however, will only happen if
you do not specify the G98 job to use an extensive amount (like 850 if
you really only need 450). So more memory doesn't always improve timing.
If the run can be made with let's say 200Mb I usually specify like 350Mb
and let the OS use the rest - this works for me.

best regards
/Patrik



from Laurence Cuffe:

1) I'm no expert!
2) performance should improve if you drop the graphic interface.
3) try using %Mem=250MB, this should allow a two processor job 
with both parts fitting into main memory, where they take 250MB 
each thus giving a total of 500 Mb.
G98's memory usage can confuse red hat 7.2's memory 
management routines, running a 4 processor system with 4 g98 
jobs each of which required less than 1/4 of the ava

ilable ram I 
found that swap and memory management routines were taking 
60% of one processors time, the solution was to provide more 
conservative memory allocations for each job.
Another thing to note is that the memory requirement for a job with 
the following:
%Nproc=4
%Mem=100MB
is 100x4MB =400MB not 100MB in total as I had first supposed.
I hope this helps
All the best
Laurence Cuffe


 
from Andreas Goetz:

Dear Ulrike,

I am not a computer expert nor am I using gaussian often.
So I cannot tell much about gaussian.

If you invoke "man httpd" you can inform yourself about httpd - Apache
hypertext transfer protocol server. I don't know what else than
calculations you are using the computer for, but I believe you don't want
to shut down httpd.

As far as I am informed, the memory which is given as buffers and cached
evoking the command "free" is memory which is reserved by the system for
speeding up certain pr
o
cesses. This memory does not have to be used and
can be freed in favour of a program which might request it.
In other words: Linux will always occupy almost all the memory to cache
instructions in the memory which might be required by programs. If a
program like gaussian will need more physical memory, the system will
reduce the cache memory and let gaussian use this part of the memory.
The numbers which really should concern you are those in the line
"-/+ buffers/cache" of the output from the "free" command.

As a comfortable tool for monitoring system ressources you might also want
to use "xosview".

Kind regards,
	Andreas


from Fabien Fontaine:

I guess most of the memory is cached, which mean that it is considered
as used but actually it is like a buffer for future use.
If you really want to know if you have memory problem, check if the swap
memory is used.

Regards,
Fabien



from Serguei Patchkovskii:


Ulrike,

Try 'ipcs -ma' command -
 G
aussian stores most of its data in a shared
memory region. If the system has been up for a while, Gaussian tends to
leave memory regions behind, making memory inaccessible. If you see any
regions with 'nattch' of zero, it is usually safe to remove them with
'ipcrm'. This should free up some memory.

Serguei




from Doug Fox:

  Prof. Salzner,

   Ignoring the question about tuning the system let me describe the
way G98 allocates and uses memory.  

   Gaussian uses dynamic allocation within a large block of memory which
is allocated/released by each link.  By default this block of memory is
6MW (48MB) long which is in reasonable agreement with your gtop output,
some additional memory is used for code and local data.  Within this block
of memory some arrarys are fixed by the problem size, others can be 
adjusted to allow the calculation to succeed, sometimes at a cost of
additional CPU due to recomputing integrals, but in each case there is a
maximum about of disk
 wh
ich is useful and a minimum below which it 
cannot be done.

   In the case of shared memory and a two processor system if the amount
of memory is small enough that two+ passes would be needed to process
all of the integrals the code drops down to 1 processor because that allows
for a single long inner loop which is more efficient than the shorter
loop over two processors.  Remember that with two processors it would 
need to setup the loops twice, i.e. it nearly destroys the time saved by
running on two processors.  

   The fact that any given problem has a maximum effective memory is the
reason for your observation that giving G98 an arbitrary large memory 
will slow down small problems.  It takes time to process the allocation
and deallocation of the memory so if it is not used it simply shows up as
overhead.  On the other hand giving the program more memory on a problem
by problem basis, i.e. adding %mem=nnnmw at the top of the input, lets
you increase memory for case
s wh
ich need it.  There is a rough formula
in chapter 4 of the G98 User's Reference which tells you the memory
for a single processor taking account of problem size and the nature of
the basis set and method.  For a dual processor system you don't need to 
double all of this, just double the portion dependent on basis set and
method.  This is a very rough guide but it should be an over estimate
and will let you use both processors.



from Cory Allen Rice:

ULRIKE SALZNER ;

YOU CAN USE THE %mem=XXXMb IN G98. THIS WILL GIVE YOU MORE MEMORY GO TO
THE RUNNING OF A JOB ON G98.

Corey A Rice
Southern Illinois University at Edwardsville
Chemistry Department
corrice@siue.edu
http://www.siue.edu/~corrice




from Bern Kallies:

Hallo Ulrike,
G98 will be compiled by default for a usage of 4mw (1 word = 8 byte,
me
ga = 
1024 kilo) of scratch memory. I think this is what you see in
your memory usage statistics (59MB includes the size of the exe, its
statically allocated memory and the dynamically allocated scratch memory
of 32MB).

Regarding your problem to get g98 running on more than 1 processor with
your system, I'll try to explain the strategy of gaussian:
When running parallel gaussian jobs (without LINDA), the master process
of a gaussian link allocates scratch space (4mw by default, or as
specified in the %mem card), and then it forks children which use this
allocated scratch space as shared memory. In l502 scratch memory will be
used mainly for handling integrals. When there is no possibility to hold
a substantial amount of integrals in memory, they will be recomputed as
needed. But some information about integrals must be stored in memory in
any event. The size required for that depends on the number of basis
functions, the maximum l value, the packing scheme of integrals and
o
ther t
hings. There is no easy way to find out this amount. But what
counts in your case is that every process stores these data private in
the scratch area. Thatswhy more scratch space will be needed when
running g98 in parallel. There exists an empirical rule for the amount
of memory needed:

mem(nproc) = mem(1proc) * (1 + 0.75*(nproc-1))

where nproc is the number of processors you want to use, and mem(1proc)
is the minimal amount required to run a particular job on 1 processor.
Gaussian checks if there is enough memory available when running in
parallel. If it guesses that it will not be enough, then it reduces the
number of processes until they fit in the available amount of memory, as
in your case. So you should specify something greater than 4mw in your
%mem card.
How much? Is an interesting question. To my experience, gaussian becomes
slower when it gets an amount of memory, which is larger than needed for
a direct SCF, but which is not enough for incore SCF. When usi
ng abou
t
120 basis functions, about 30mw are enough for incore. That means, using
8mw is slower than 4mw, with 16mw is slower than with 8mw, with 32mw is
faster than with 4mw (switches to incore), with 64mw is as fast as with
16mw and so on. When using about 350 basis functions, one would need
2.7gw (about 20GB) for incore. Allocating only 2.6gw in the latter case
would have the only effect that the job runs terribly long, longer than
with 8mw. So in your case I would succesively try it with 8mw, 16mw,
32mw, until the job runs. When it runs already with 4mw on one
processor, then 8mw would be enough according to the formula above.

Hope this helps,
best regards, Bernd


from Demetrio de Silva Filho:

Hello Ulrike,


There is something you must remember. The memory you specify

is not what Gaussian will always use but the maximum memory

Gaussian will use during the job. If you follow what is happening while

you are running, you will see these numbers
 change,
 depending on the

link.


The problem with gaussian is : there are few parts of the program written in

old fortran 77 that cannot alocate memory dynamicaly, i.e. as the process

needs. So, you must specify in advance how much memory you have.


Hope it helps!


Demetrio FIlho

--------------070603030809030003000302-- From chemistry-request@server.ccl.net Thu Jun 13 04:44:06 2002 Received: from ribotargets.com ([194.129.39.11]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g5D8i5j13263 for ; Thu, 13 Jun 2002 04:44:05 -0400 Received: from szilva (helo=localhost) by echo.ribotargets.com with local-esmtp (Exim 3.13 #1) id 17IQ9T-00031t-00; Thu, 13 Jun 2002 08:40:19 +0000 Date: Thu, 13 Jun 2002 08:40:19 +0000 (GMT) From: Szilveszter Juhos To: eric hu cc: chemistry@ccl.net Subject: Re: CCL:linux babel In-Reply-To: <20020612060627.19400.qmail@web13403.mail.yahoo.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII > Hi, I wonder if anyone can point me the place to > download a linux version BABEL. Source at: http://openbabel.sourceforge.net/ Szilva From chemistry-request@server.ccl.net Thu Jun 13 15:12:35 2002 Received: from web13401.mail.yahoo.com ([216.136.175.59]) by server.ccl.net (8.11.6/8.11.0) with SMTP id g5DJCYj24480 for ; Thu, 13 Jun 2002 15:12:34 -0400 Message-ID: <20020613191234.88940.qmail@web13401.mail.yahoo.com> Received: from [169.232.140.249] by web13401.mail.yahoo.com via HTTP; Thu, 13 Jun 2002 12:12:34 PDT Date: Thu, 13 Jun 2002 12:12:34 -0700 (PDT) From: eric hu Subject: thanks for info about linux babel To: chemistry@ccl.net MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Hi, I want to thank so many of you (about 20!) to send me various sources of babel downloads. In fact I also wanted to suggest a better organization of the ccl website especially the message archives to make searching easier. Eric __________________________________________________ Do You Yahoo!? Yahoo! - Official partner of 2002 FIFA World Cup http://fifaworldcup.yahoo.com From chemistry-request@server.ccl.net Thu Jun 13 16:57:53 2002 Received: (from ccl@localhost) by server.ccl.net (8.11.6/8.11.0) id g5DKvr726269 for chemistry@ccl.net; Thu, 13 Jun 2002 16:57:53 -0400 Message-ID: <001d01c2131f$ae132780$4bd8f143@anick000> From: "David J Anick" To: chemistry@ccl.net Subject: Thermal corrections Date: Thu, 13 Jun 2002 14:17:08 -0700 Dear CCLers, Having completed a Gaussian98 frequencies/ZPE calculation at an optimized geometry, I now find that I would like to know the thermal corrections to energy, enthalpy and Gibbs free energy at a different temperature (77 K) than the one that Gaussian98 automatically calculated (298 K). I don't want to re-run the calculation if possible because it took nearly a week. Can anyone tell me the formula(s) that Gaussian98 uses to compute these data? If it involves integrating Cp or Cp/T, how do I obtain Cp as a function of temperature? Thank you in advance, David Anick MD PhD David.Anick@gte.net From chemistry-request@server.ccl.net Thu Jun 13 15:58:40 2002 Received: from mail.isg.siue.edu ([146.163.5.4]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g5DJwej25367 for ; Thu, 13 Jun 2002 15:58:40 -0400 Received: from WEBSHIELD2.isg.siue.edu (webshield2.isg.siue.edu [146.163.5.150]) by mail.isg.siue.edu (8.9.1/8.9.1) with SMTP id OAA15410 for ; Thu, 13 Jun 2002 14:58:26 -0500 (CDT) Received: FROM mail.isg.siue.edu BY WEBSHIELD2.isg.siue.edu ; Thu Jun 13 14:58:25 2002 -0500 Received: from cougar.isg.siue.edu (cougar [146.163.5.29]) by mail.isg.siue.edu (8.9.1/8.9.1) with ESMTP id OAA15341 for ; Thu, 13 Jun 2002 14:58:21 -0500 (CDT) Received: (from nobody@localhost) by cougar.isg.siue.edu (8.9.1/8.9.1) id OAA02921; Thu, 13 Jun 2002 14:58:21 -0500 (CDT) Date: Thu, 13 Jun 2002 14:58:21 -0500 (CDT) From: corrice@siue.edu Message-Id: <200206131958.OAA02921@cougar.isg.siue.edu> X-Authentication-Warning: cougar.isg.siue.edu: nobody set sender to corrice@siue.edu using -f To: chemistry@ccl.net Reply-To: corrice@siue.edu MIME-Version: 1.0 Content-Type: multipart/mixed; boundary="-MOQ1023998300a18fb29236f6a2768ca7244fd9df10fb" User-Agent: IMP/PHP3 Imap webMail Program 2.0.11 Sender: corrice@siue.edu X-Originating-IP: 146.163.155.168 Subject: CI calculation for CuO This message is in MIME format. ---MOQ1023998300a18fb29236f6a2768ca7244fd9df10fb Content-Type: text/plain Content-Transfer-Encoding: 8bit TO WHOM THIS MAY CONCERN: I WAS WONDERING WHAT WAS WRONG WITH THE WAY THIS INPUT FILE WAS SET UP FOR G98. I HAVE BEEN HAVE PROBLEMS WITH THIS PARTICULAR ONE. I HAVE TRIED DIFFERENT KEYWORDS BUT NOTHING SEEMS TO BE WORKING. THE INPUT FILE IS ATTACHED. COREY RICE SIUE CHEMISTRY DEPT. CORRICE@SIUE.EDU ---MOQ1023998300a18fb29236f6a2768ca7244fd9df10fb Content-Type: text/plain; name="CuO_excited.gjf" Content-Transfer-Encoding: 8bit Content-Disposition: inline; filename="CuO_excited.gjf" %mem=210Mb #p UCIS=(maxcycle=3000,NStates=15)/6-311(d) TransitionDensities FormCheck=(All) Pop=(NPA,Regular) scf=(QC,maxcycle=5000) IOp(9/40=5) > from cuo 0 2 O 0 -1.013351 -0.285889 0.293335 Cu 0 1.013336 0.285889 -0.293320 ---MOQ1023998300a18fb29236f6a2768ca7244fd9df10fb-- From chemistry-request@server.ccl.net Thu Jun 13 16:53:09 2002 Received: from haney.hbond.com ([207.137.2.252]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g5DKr8j26167 for ; Thu, 13 Jun 2002 16:53:08 -0400 Received: (from haney@localhost) by haney.hbond.com (8.9.3/8.9.3) id NAA01766; Thu, 13 Jun 2002 13:55:09 -0700 From: "Dr. David N. Haney" Message-Id: <200206132055.NAA01766@haney.hbond.com> Subject: Re: CCL:linux babel To: erichu_linux@yahoo.com (eric hu) Date: Thu, 13 Jun 2002 13:55:09 -0700 (PDT) Cc: chemistry@ccl.net In-Reply-To: from "eric hu" at Jun 11, 2002 11:06:27 PM X-Mailer: ELM [version 2.5 PL5] MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Actually, Babel has been made into an open source library. You can compile it into a "BABEL" program or you can compile it into your own software. It is now called "OELIB" and is available for download > from OpenEye Scientific Software (http://www.eyesopen.com/download.html). > Hi, I wonder if anyone can point me the place to > download a linux version BABEL. -- ######### David N. Haney, Ph.D. ######### # Haney Associates Phone - 858-483-1197 # # 5455 Westknoll Dr. FAX - 858-483-1046 # # La Jolla, CA 92037 Email - haney@hbond.com # ################# #####################