[gmx-users] error in g_cluster "diagonalization" method

[Mark Abraham]

On 20/06/2012 12:03 AM, Isaure Chauvot de Beauchêne wrote:
> Hi,
>
> I'm trying to cluster 200 structure of a 20 aas peptide, by the 
> g_cluster "diagonalization" method. There is my command line:
>
> *********************
> g_cluster -f all_1-200.pdb -s ../pdb/FinalGSAFold_1.pdb -clid -cl 
> clusters.pdb -method diagonalization -o rmsd-clust.xpm -g cluster.log
> *********************
>
> I obtain these lines:
>
> *********************
> The RMSD ranges from -nan to -nan nm
> Average RMSD is -nan
> Number of structures for matrix 200
> Energy of the matrix is -nan nm
> *********************
>
> If I run exactly the same job with changing for "gromos" method, I 
> obtain:
>
> *********************
> The RMSD ranges from 0.26113 to 1.22787 nm
> Average RMSD is 0.622773
> Number of structures for matrix 200
> Energy of the matrix is 41.58 nm
> ...etc
> *********************
>
> Are there some case in which the diagonalization method is not usable? 
> What could be the reason of this fail?
> I use VERSION 4.5.5.

Depending how GROMACS was configured, diagonalization may be using an 
external library that might not be (dynamically) linked correctly, 
leading to garbage results. Or your problem size is too large for your 
available memory (and you are somehow not segfaulting). Diagonalizing a 
200 x 200 matrix is not huge, but might not be possible on small 
systems. I suggest trying a much smaller number of structures and 
gradually increasing it to see what happens - or trying on a computer 
with more memory. If everything fails, then the linking hypothesis might 
be the best lead.

Mark