Problem 1161

Summary:	Crashes due to exception in Bertini Cascade code
Product:	Geant4	Reporter:	banerjee <banerjee>
Component:	processes/hadronic/models/cascade	Assignee:	Michael Kelsey <kelsey>
Status:	RESOLVED LATER
Severity:	normal	CC:	John.Apostolakis, kelsey
Priority:	P3
Version:	9.4
Hardware:	All
OS:	All

Description banerjee 2011-01-19 09:37:18 CET

In CMS application, we are getting occasional crashes with Geant4.9.4 using the QGSP_FTFP_BERT(_EML) ohysics list. This happens after a printout from the Bertini cascade code

Energy conservation violated by -0.195996 GeV (-0.0848549)
Momentum conservation violated by 0.0848835 GeV/c (0.0848549)

from PostStepDoIt

Comment 1 Michael Kelsey 2011-01-19 18:14:37 CET

The "crash" in this case is a thrown G4Exception which terminates the job.  The full set of relevant error messages are:

  >>> G4CascadeInterface::ApplyYourself()
  has non-conserving cascade after 100 attempts.
  Energy conservation violated by -0.195996 GeV (-0.0848549)
  Momentum conservation violated by 0.0848835 GeV/c (0.0848549)

The point where this occurs (in 9.4) is G4CascadeInterface.cc:220, followed by a call to G4CascadeInterface::throwNonConservationFailure(), which generates the exception.

If this crash is reproducible, it would be very helpful to have the projectile and target information (type and four-momentum) which were passed to G4CascadeInterface::ApplyYourself().

Comment 2 John Apostolakis 2011-01-20 10:58:34 CET

Changed Assignee to Mike K.

Comment 3 Michael Kelsey 2011-01-20 19:50:09 CET

1)  Increased importance to P3/normal.

2)  Log file from original crash should contain (on G4cout) details of the interaction which caused the exception.  I've requested via e-mail more of the log file.  Lines subsequent to the "conservation violated by" messages should report the projectile kinematics, target nucleus, and cascade secondaries.

Comment 4 Michael Kelsey 2011-02-24 08:16:53 CET

We have not yet received any more details on this thrown exception.  The last information I had (a couple of weeks ago) was that the CMS folks were running a new series of jobs to reproduce it.  Without details on the interaction which caused the exception, I can neither isolate nor "fix" whatever problem might be present.  I'll mark this as "RESOLVED" for the moment; it can be reopened if additional details become available.