sherpa is hosted by Hepforge, IPPP Durham
close Warning: Can't synchronize with repository "(default)" (/hepforge/svn/sherpa does not appear to be a Subversion repository.). Look in the Trac log for more information.
Modify ↓

Opened 8 years ago

Closed 7 years ago

#370 closed defect (fixed)

Crash with "Invalid amlitude" in V+gamma

Reported by: Johannes Krause Owned by: Stefan Hoeche
Priority: minor Milestone: rel-2.3.0
Component: ME Generator Version: 2.2.4
Keywords: Cc:

Description

I observe some rare crashes in Sherpa for Z+gamma / W+gamma production at MEPS@NLO. The error message is

Sherpa: CS_Gamma::TrialWeight() throws fatal error: Invalid amplitude for ...

This seems to be related to a "zero matrix element" in MCATNLO/Main/CS_Gamma::SingleWeight(). The first issue is that in line 168 zero is returned but in line 121 minus one is expected. However, fixing this does not help since in both cases the weight map "ws" is not filled. This finally causes the "invalid amplitude" crash.

As a workaround one can comment out the "return 0" statement in line 168, when doing so the weight becomes infinity and a new event is requested. It is not clear to me why the matrix element can be zero and if this is a real problem.

The used run card and the complete event output are attached.

Attachments (2)

invalid_amplitude.log​ (168.0 KB) - added by Johannes Krause 8 years ago.
log file
Run.test​ (962 bytes) - added by Johannes Krause 8 years ago.
run card

Download all attachments as: .zip

Change History (6)

Changed 8 years ago by Johannes Krause

Attachment: invalid_amplitude.log​ added

log file

Changed 8 years ago by Johannes Krause

Attachment: Run.test​ added

run card

comment:1 Changed 8 years ago by Enrico Bothmann

I've just got the same crash with rel-2-2-3 for MC@NLO single-top production after a few thousand events: 

Sherpa: CS_Gamma::TrialWeight throws fatal error:
   Invalid amplitude for 2_3__G__u__t__d__bb

However, with rel-2-2-2 it worked fine. Although the versions are fortunately not very different, I have not spotted anything obvious with a diff between the two branches.

UPDATE: The crash appeared also with rel-2-2-2, only much later (after 200k events).

Last edited 8 years ago by Enrico Bothmann (previous) (diff)

comment:2 Changed 7 years ago by Frank Siegert

Version: 2.2.1 → 2.2.4

I also observe this rare crash in diphoton production and have put two log files showing this behaviour in Sherpa 2.2.0 and 2.2.4 into:

Last edited 7 years ago by Frank Siegert (previous) (diff)

comment:3 Changed 7 years ago by Frank Siegert

We should discuss two options:

  1. If it is Ok that the matrix element from Amegic is zero, then CS_Gamma has to be able to cope with that.
  1. Or we accept that a NaN appears and the event is being retried.

In both cases we need to remove the fatal_error if me=0.

Last edited 7 years ago by anonymous (previous) (diff)

comment:4 Changed 7 years ago by Frank Siegert

Resolution: → fixed
Status: new → closed

This is fixed in r31342 on rel-2-2-4 (was already fixed on trunk) and will become available in Sherpa >=2.2.5.

Modify Ticket

Change Properties
Action
as closed The owner will remain Stefan Hoeche.

Add Comment

E-mail address and name can be saved in the Preferences.

Attachments ↑ Description ↑
 
Note: See TracTickets for help on using tickets.