Re: [AMBER-Developers] XMIN minimisation and coordinates

From: Ben Roberts <ben.roberts.geek.nz>
Date: Tue, 8 Mar 2011 16:45:55 -0500

Hi Istvan,

On 2/3/2011, at 12:21 a.m., istvan.kolossvary.hu wrote:

>> I believe that's not true. In sander, there's a CASE statement that
>> sends the code down one path if the value is DONE; another if the
>> value is CALCENRG, CALCGRAD or CALCBOTH; a third if the value is
>> CALCENRG_NEWNBL, CALCGRAD_NEWNBL or CALCBOTH_NEWNBL; and a fourth if
>> the value is CALCENRG_OLDNBL, CALCGRAD_OLDNBL or CALCBOTH_OLDNBL. (A
>> fifth option, the "default", triggers an ASSERT.)
>
> This might be something new, I must remember an older version, but
> then this should be fine.

OK, no trouble.

>> Have you any thoughts on my other question, specifically: under what
>> conditions are the lines starting "MIN:" printed out in xminC.c
>> (assuming the appropriate verbosity is turned on)? I want to be able
>> to capture the coordinates at those points - where a minimisation
>> iteration is recorded.
>
> If I understand your question correctly the answer is that "MIN: "
> lines are printed at each and every minimization step, at the end of
> line minimization and at that point the gradient is called with NEWNBL.

Right! Between your information and much mucking about with print statements, I think I now understand. In xminC.c, it seems that the energy is calculated - and, if verbosity is high enough, printed - during line minimisation. Then, sometime during line minimisation, the coordinates are changed (following the "line"?) but the "energy" is not. In lmod.f, the energy is printed (by calling report_min_progress) before gradient_calc is called, so it hasn't been updated to reflect the changed coordinates.

I think this can be fixed just by changing the order in which subroutines are called in lmod.f.

Thanks for your help!

Cheers,
Ben
-- 
For greater security, I support S/MIME encryption.





_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers


Received on Tue Mar 08 2011 - 14:00:04 PST
Custom Search