AMBER Archive (2003)

Subject: Fw: AMBER: Questions on Phi, Psi angles produced by pmemd/sander7

From: Robert Duke (rduke_at_email.unc.edu)
Date: Sat Aug 23 2003 - 22:44:06 CDT


----- Original Message -----
From: "Robert Duke" <rduke_at_email.unc.edu>
To: "Chen, Ya" <ya.chen_at_RoswellPark.org>
Sent: Saturday, August 23, 2003 12:36 PM
Subject: Re: AMBER: Questions on Phi, Psi angles produced by pmemd/sander7

> Chen -
> Sounds good. I would point out for everyone, though, that in doing
> comparisons between pmemd and sander, what we will have the most luck with
> in comparing results is comparing sander 6 and pmemd in 6 mode. I say
this
> because while pmemd 7 mode is meant to mimic sander 7 to the extent
> practically possible, there are some "under-the-covers" differences in
> sander 7, mostly involved with extra points support, that I didn't track
> because I don't support extra points. Thus, Tom D. changed some stuff in
7
> that may or may not be significant. I first encountered this when we
> noticed a change in virials values for crosslinked molecules in sander 7;
> turns out that that was due to a change in 7 in the computation of
molecular
> and atomic virials; Tom regards neither the 6 or 7 implementation as "more
> correct", they are just different (basically in whether they treat two
> molecules crosslinked by an additional bond as 1 molecule or 2, I
believe).
> Where the 1-4NB calcs get done is different; I don't know if there are
> subtle differences in the results. So what I am saying is that there MAY
be
> some differences between sander 6 and sander 7 that have not been sorted
> out, so that muddies up comparisons between pmemd in 7 mode and sander 7.
> Pmemd is mostly a fast implementation of sander 6, with additional support
> for some sander 7 features, the primary one being sander 7-style step
> integration and default variables. If you do get significantly different
> numbers again, and they are not simply attributable to trajectory sampling
> issues, then I think it would be really good to see what sorts of results
> you get out of sander 6.
> Regards - Bob
>
> ----- Original Message -----
> From: "Chen, Ya" <ya.chen_at_RoswellPark.org>
> To: <rduke_at_email.unc.edu>
> Sent: Saturday, August 23, 2003 11:32 AM
> Subject: RE: AMBER: Questions on Phi, Psi angles produced by pmemd/sander7
>
>
> > Bob,
> >
> > Thanks for the quick reply.
> > I didn't know bugfix42 was significant and didn't patch it yet. So it
> could be the reason why RMSD from sander7 at the beginning was big: the
> structure wasn't optimal.
> > And indeed, I used Amber6 mode with pmemd, which wasn't patched either.
> > I just loaded the newly patched pmemd with a new job, amber7_compat=1.
> I'll let you know the results on Monday.
> >
> > Regards,
> > Chen
> >
> > -----Original Message-----
> > From: Robert Duke [mailto:rduke_at_email.unc.edu]
> > Sent: Saturday, August 23, 2003 10:42 AM
> > To: amber_at_scripps.edu
> > Subject: Re: AMBER: Questions on Phi, Psi angles produced by
pmemd/sander7
> >
> >
> > Chen -
> > I did some additional checks of the source, and still don't see problems
> in terms of sander 6 - pmemd 6 conformance. A dump of force array
> components (ie., force vectors for each individual atom) attributable to
the
> two executions of the dihedral code in the first step are identical to 5
> places after the decimal for a 20943 atom problem (the relevant part being
> 795 protein atoms). This typically constitutes at least 6digit precision,
> and I did not check for higher levels of conformance. For my code,
> dihedrals are handled the same way in sander 6 mode and sander 7 mode.
For
> sander, I think the mechanisms are supposed to be the same. However, the
> dihedral code in 7 has been modifed; specifically the 1-4 NB code has been
> pulled out of the dihedral routine and placed in the ewald code. Tom
Darden
> would have to comment on whether there are any differences in how this
code
> behaves (ie., a subtle difference between sander 6 and 7 is possible here,
> and pmemd would track the 6 behaviour). Regards - Bob
> >
> >
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu