[phenixbb] reduce hangs
michael.strickler at yale.edu
Tue Dec 3 13:18:33 PST 2013
On 12/03/2013 09:45 AM, Jeff Headd wrote:
> Hi Andreas,
> We have seen in the past some cases that cause Reduce to go into an
> infinite loop, which I thought had been corrected, but perhaps your
> model has found a corner-case that I have not considered. NCS vs. no NCS
> is not the direct problem, it likely has to do with a particular set of
> coordinate positions that you are reaching during refinement which
> causes the hydrogen position determination to end up in either an
> infinite loop, or just an optimization that isn't tractable in a
> reasonable amount of time. Hydrogens are added for both NQH flips and
> for clashscore calculation if your input model does not have them.
To provide an additional datapoint, several of our users have
experienced similar hangs with reduce. Tests of one of the jobs reach
"Exporting results," and then reduce runs until killed. This is the
case even for rigid-body refinement. The hangup occurs under Linux with
Phenix 1.8.4-1496 and with Phenix 1.8.3-1479 (subsequently tried for
diagnostic purposes). Some users have no such problem running the
latest Phenix release; I have not yet determined conclusively what
distinguishes between them, but the problem job used for testing
involves large molecules with numerous NCS groups at 2.8 Angstroms.
If users fall back to Phenix 1.8.2-1309 (our default until recently),
the problem is reported to go away, which I have independently confirmed
for the test case.
Michael Strickler, Ph.D.
Richards Center for Structural Biology
260 Whitney Ave, 355C JWG
PO Box 208114
New Haven, CT 06520-8114
More information about the phenixbb