[gmx-users] Bug in g_confrms !

[Berk Hess]

>From: Marc Baaden <baaden at smplinux.de>
>Reply-To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>To: gmx-users at gromacs.org
>Subject: [gmx-users] Bug in g_confrms !
>Date: Tue, 23 Aug 2005 19:57:18 +0200
>
>
>Hi,
>
>we just found a bug in g_confrms. We have not yet spotted the origin of
>the bug in the source code, but we can reproduce clearly wrong results.
>The problems start from version 3.2.1, version 3.1.4 is ok. For those who
>want to check, a simple test case is available here (just do make after
>untarring):
>
>http://www.shaman.ibpc.fr/rmsbug.tar
>
>It seems like this happens when the number of atoms are different. The
>example is just fitting 4 atoms on themselves, by randomly adding either
>one atom, or one in two (eg fitting 5 on 4 atoms or 8 on 4). The option
>-nomw amplifies the error. Below some numbers to show that the deviation
>can be significant.
>
>If anybody knows how to fix this, thanks in advance. Meanwhile, you should
>probably revert back to gmx 314 for doing g_confrms type analysis.

I found the problem (my fault as well actually...).

In calc_rm_cm in gmx_confrms.c it says:
m = atoms->atom[i].m;
whereas this should be:
m = atoms->atom[index[i]].m;

Thus when the index does not consist of the first n atoms
the weighting for the translational fit will not be correct.

Thanks for reporting this,

Berk.