[gmx-developers] Problem with g_covar diagonalizing large covariance matrices

[Berk Hess]

Hi,

In addition, unless you are only looking at bound ions, the ions will diffuse around and dominate the covariance, which will give random, uninteresting and results.

Cheers,
Berk

On Oct 27, 2014 8:48 PM, Mark Abraham <mark.j.abraham at gmail.com> wrote:
>
>
>
> On Mon, Oct 27, 2014 at 2:19 PM, Mendez Giraldez, Raul <rmendez at email.unc.edu> wrote:
>>
>> Dear Gromacs developers,
>>
>> I am trying to compute PCA on a 46464x46464 matrix (15488 atoms). g_covar produces segmentation fault. We thought it was a lack of physical memory, but when our computer cluster system administrator, Shubin Liu, looked into the system logs, found that the maximum RAM used by the program was 16 Gb, while I had booked up to 800 Gb of RAM. Shubin realized that the actual problem arises at the eigensolver call within covar routine. So it looks like eigensolver is unable to handle big matrices, but not because of lack of physical memory (is it based on LAPACK libraries, still 32-bit coded?).
>
>
> I've no idea if there are any limits, but you can choose the LAPACK implementation with which to link GROMACS (see install guide), so you should find out what you've been using (e.g. inspect CMakeCache.txt in your build tree), and explore alternatives.
>
>> I am using Gromacs 4.6.3.
>>
>> Do you know an alternative to that? The reason I want to do that is to have some ion atoms in addition to my protein and see how they move along the low frequency modes (large variance PCs).
>
>
> How many of your protein atoms do you really need for that analysis? I presume you're already excluding the hydrogen atoms, since they're not doing much at low frequencies. Even if you think you really do need them, I suggest you start with something small that works.
>
> Mark
>  
>>
>>
>> Thank you so much,
>>
>> Raul
>>
>> Raul Mendez Giraldez, PhD
>> Dokholyan Lab
>> Dept. Biophysics & Biochemistry 
>> Genetic Medicine
>> University of North Carolina
>> 120 Mason Farm Road
>> Chapel Hill, NC 27599
>> US
>>
>> --
>> Gromacs Developers mailing list
>>
>> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before posting!
>>
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers or send a mail to gmx-developers-request at gromacs.org.
>
>