[phenixbb] phenixbb Digest, Vol 79, Issue 4

[Nathaniel Echols]

On Thu, Jun 14, 2012 at 7:10 AM, CPMAS Chen <cpmasmit at gmail.com> wrote:
> I thought I am the only one have this problem.
> I have encountered the same problem with the newest version of phenix.
> RuntimeError: Estimated memory allocation for cubicles exceeds
> max_number_of_bytes:
>   This may be due to unreasonable parameters:
>     cubicle_edge=5
>     space_span=(2696.14,2555.98,2650.54)
>     n_cubicles=(540,512,531)
>     max_number_of_bytes=1073741824

Several other users have reported the same problem.  Pavel thinks the
underlying cause (which should be fixable) is that the minimizer is
exploding the structure during real-space refinement, and throwing
atoms several unit cells away.  The part that's crashing is trying to
divide the atoms into boxes for calculating the nonbonded interactions
more efficiently.  If the structure covers too much physical space, it
ends up with too many boxes.  (Regarding Kay's posting, I think I was
confused about the expected size of "unsigned long" on a 64-bit
platform - which does not in fact change.  Presumably I was confusing
this with "size_t", which should change.)

Those of you who encounter this problem, it seems to be associated
with these two parameters:

refinement.ncs.torsion.fix_outliers = Auto
refinement.ncs.torison.check_rotamer_consistency = Auto

If you set these to False, it should run okay.

-Nat