AMBER Archive (2008)

Subject: Re: AMBER: Test failure with Amber 10

From: Robert Duke (rduke_at_email.unc.edu)
Date: Wed Nov 26 2008 - 11:28:14 CST


Hi folks,
It is possible that there will be some slight differences on the amoeba test
when running pmemd vs. when running sander. All the test comparison files
were generated on sander. Now here's the deal. In the course of doing the
pmemd.amba development a couple of years ago, I found some relatively minor
problems in the sander implementation and fixed them in pmemd. I pushed
this info back to the sander folks, but I am not certain that all the
changes were made. I also commented that there would need to be a change in
the tests, but the ball got dropped on this in the release. This was sort
of a matter of coordination problems, and at the time of the amber 10
release, I had not done any work on amoeba for something like 18 months.
BUT pmemd.amba is both the fast (relatively speaking...) and correct
implementation of the amoeba forcefield in amber. Determining what the
differences are in sander and fixing them has been a pretty low priority for
me, as it is sort of nuts to use sander for amoeba work, given the
performance; the amoeba forcefield will never be particularly fast due to
how polarization is handled as well as costs associated with the multipole
code, and this limits the profit in further work on this code from my
perspective. If enough folks get disgruntled, I can probably be pushed to
go back and figure out what is going on with sander and the tests, but it
really needs to turn out to be a problem for people. So far, pmemd.amba
usage has been limited, but I have received no complaints.
Regards - Bob Duke

----- Original Message -----
From: "Ross Walker" <ross_at_rosswalker.co.uk>
To: <amber_at_scripps.edu>
Sent: Wednesday, November 26, 2008 11:30 AM
Subject: RE: AMBER: Test failure with Amber 10

> Hi Kevin,
>
>> run with the PMEMD implementation of Sander (stated on line 6 of
>> mdout.tip5p.save) which outputs some additional information including
>> the fields highlighted in the test difference.
>
> This shouldn't matter. Bob has done a LOT of work on PMEMD to make it
> fully
> compatible with sander and results should be identical although when
> running
> in parallel different summation orders can lead to round off differences.
> However a test case run with PMEMD should pass when compared using dacdif
> against a test case save file generated with sander and vice versa. Indeed
> the tip5p test case passes on my machine with both sander and pmemd so I
> think what you are seeing is just your compiler rounding things in a
> different way that as Dave says for a very short run can give marked
> differences in the RMS.
>
> All the best
> Ross
>
>
> /\
> \/
> |\oss Walker
>
> | Assistant Research Professor |
> | San Diego Supercomputer Center |
> | Tel: +1 858 822 0854 | EMail:- ross_at_rosswalker.co.uk |
> | http://www.rosswalker.co.uk | PGP Key available on request |
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
> be read every day, and should not be used for urgent or sensitive issues.
>
>
>
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
> to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
      to majordomo_at_scripps.edu