[phenixbb] MemoryError in phenix.refine

[Nathaniel Echols]

On Mon, Sep 5, 2011 at 9:13 AM, Tanner, John J. <TannerJJ at missouri.edu> wrote:
> I get the following error at the end of a phenix.refine run.  The pdb file
> is produced, but the mtz file is empty.  The resolution is 1.15 A and the
> cell is
>
> CRYST1   84.808   93.803  132.080  90.00  90.00  90.00 P 21 21 21
>
> The version is
>
>  PHENIX: Python-based Hierarchical ENvironment for Integrated Xtallography
>  Version: 1.7.1
>  Release tag: 743
>  Platform: intel-linux-2.6 redhat-e5.6
>
> I've seen this sort of thread on the bb previously.  Is there a fix for
> this, other than using refmac?

The recommended fix, I'm afraid, is to install a 64-bit OS (and
probably more physical memory as well).  Alternately, if you turn off
output of maps/map coefficients, it will skip the step that is
breaking here (you can use phenix.maps to create the maps later).
There are probably several places in the code where we handle memory
inefficiently, but debugging these issues is a longer-term project
since it's a slow process.  If you're willing to share the model and
data with us (or alert us when it gets published), this might make it
easier to isolate problems.

A more general recommendation: when encountering a MemoryError crash
in phenix.refine, the first thing to check is whether you have rotamer
fitting or solvent update enabled - these both require maps, which
tend to be fairly large and poorly optimized for memory use.  (This
isn't what's breaking in your case, however.)

thanks,
Nat