[molpro-user] MULTI RESTART

[Dr Seth OLSEN]

Hi David,

It seems that at one point I saw in the manual an option to save
orbitals at every iteration.  Right now I can't seem to find it.  It
might be worth looking for.  In addition, perhaps it would work to
explicitly use the NATORB directive (if you are not already), perhaps
saving to a different record.  

Alternatively, if the orbitals are in fact being saved, you should be
able to access them via MATROP after your mcscf exceeds its iterations.
 If you can't access them at the specified record, than this is
informative and implies that they aren't being dumped.  If they are
accessible, than you can output them to an ascii file and read them in
explicitly in the next job.  This is an often-overlooked feature of
molpro that I find quite useful (since the wfn file is binary, it is
system-dependent - so if you want to restart on another architecture for
whatever reason than you have to do this).  The orbitals can be saved
using the following deck (assuming orbitals are stored in 2140.2).

matrop
load,orbitals,orb,2140.2
write,orbitals,output_file.orb

they can then be read in before mcscf in the next job with:

matrop
read,orbitals,orbitals
include,output_file.com,1
save,orbitals,2140.2,orbitals


Note that if you use this option it helps to use 'mcscf' to call the
multi program, rather than 'multi'.  This is because 'multi' is also a
matrop command - your job will fail with a bunch of 'illegal command'
errors b/c the program doesn't realize you're calling the multi program.

Hope this helps.

Cheers,

Seth


ccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccms

Dr Seth Olsen, PhD
Postdoctoral Fellow, Biomolecular Modeling Group
Centre for Computational Molecular Science
Chemistry Building,
The University of Queensland
Qld 4072, Brisbane, Australia

tel (617) 33653732
fax (617) 33654623
email: s.olsen1 at uq.edu.au
Web: www.ccms.uq.edu.au 

ccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccms




----- Original Message -----
From: David Ambrosek <ambrosek at chemie.fu-berlin.de>
Date: Monday, July 25, 2005 10:59 pm
Subject: Re: [molpro-user] MULTI RESTART
> Hello again,
> 
> You are right, it would be easier to just set START as follows:
> 
> start,2140.2
> 
> for example. This was also tried but doesnt work if you send in 
> your 
> second job. You see I have no problem using start if it is within 
> the same 
> job. But the moment my job ends and I need/want to start from 
> specific 
> orbitals then it just doesnt seem to work. So I want to verify that 
> you 
> use START even for resent jobs?
> 
> 
> I am just wanting to get my gs wavefunction for my system. If the 
> orbitals 
> are good, then I want to increase to more states. After I have 
> these 
> states then I would like to run caspt2 on them etc. So it is always 
> building on the step before, but if I can not save these orbitals 
> properly 
> and then 
> recall them properly, then it is just a lot of wasted time!
> 
> Can you expand on this topic,
> thanks again,
> 
> On Mon, 25 Jul 2005, Dr Seth OLSEN wrote:
> 
> >
> > Hi David,
> >
> > If I read your technique correctly, your problem is that you need to
> > refer to the VALUE of reforb, not the string 'reforb'.  This is done
> > as in bash, etc. by using a dollar sign in front of the variable
> > name.  In your case, the input in your second job might work 
> better if
> > you use:
> >
> > start,$reforb
> >
> > This might work.  I usually just put the record in w/o using the
> > intermediate variable, though, so I can't guarantee it from 
> firsthand> experience.
> >
> > Cheers,
> >
> > Seth
> >
> > ccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccms
> >
> > Dr Seth Olsen, PhD
> > Postdoctoral Fellow, Biomolecular Modeling Group
> > Centre for Computational Molecular Science
> > Chemistry Building,
> > The University of Queensland
> > Qld 4072, Brisbane, Australia
> >
> > tel (617) 33653732
> > fax (617) 33654623
> > email: s.olsen1 at uq.edu.au
> > Web: www.ccms.uq.edu.au
> >
> > ccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccms
> >
> >
> >
> >
> > ----- Original Message -----
> > From: David Ambrosek <ambrosek at chemie.fu-berlin.de>
> > Date: Monday, July 25, 2005 8:10 pm
> > Subject: [molpro-user] MULTI RESTART
> >> Perhaps a very naive question. I have run a MULTI job on a very
> >> large
> >> molecule and after 10 iterations he has NOT converged...but is
> >> close! Now
> >> I want to restart the multi program at the point where he left off
> >> and I
> >> do not know how to do this. Of course I have consulted the manual
> >> and it
> >> led me to try a few things, that unfortunately did not work.
> >>
> >> Let me shortly summarize:
> >>
> >> 1) Using the restart command:
> >>
> >> By using the restart command, which according to the manual is
> >> redundant,
> >> molpro should start from the most recent data files. Unfortunately
> >> when
> >> multi starts he automatically starts from orbitals that are
> >> generated from
> >> atomic densities instead of using the orbitals that he just
> >> calculated.
> >> 2) Using a variable like reforb
> >>
> >> 1st job:	reforb=2140.2
> >>
> >> 2nd job:  	start,reforb
> >>
> >> Still as in case one multi starts from atomic densities instead of
> >> the
> >> orbitals I assign!
> >>
> >> 3) Within the same script:
> >>
> >> In the same script I can control how many iterations are done in
> >> multi and
> >> then I can restart multi directly after giving also a new number
> >> of
> >> iterations and he does this successfully.
> >>
> >> Question: I do not know in advance if all the jobs are going to
> >> run to
> >> convergence. If they do not, how do I restart the multi job
> >> without having
> >> to start from point zero?
> >>
> >> Thanks in advance for any help,
> >> David
> >>
> >>
> >>
> > 
>
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~> ~~~
> >> David Ambrosek
> >> ambrosek at chemie.fu-berlin.de
> >> Freie Universitaet Berlin                      Tel: (++)49 (+)30
> >> 838 53348
> >> Institut fuer Chemie
> >> Room 35.21
> >> Takustr. 3
> >> D-14195 Berlin
> >>
> > 
>
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~> ~~~
> >>
> >
>