whizard is hosted by Hepforge, IPPP Durham

Opened 11 years ago

Closed 11 years ago

#611 closed defect (fixed)

NaN Efficiency

Reported by: msekulla Owned by: kilian
Priority: P0 Milestone: v2.2.0
Component: core Version: 2.2.0beta
Severity: blocker Keywords:
Cc:

Description

Efficiency becomes NaN for pr, pr -> H, Z, j, j

Output: |=============================================================================| | It Calls Integral[fb] Error[fb] Err[%] Acc Eff[%] Chi2 N[It] | |=============================================================================| Warning: ! vamp: Function identically zero in channel 5

1 4964 5.9075341E+01 2.89E+01 48.86 34.42* NaN 2 4964 1.1352075E+02 6.16E+01 54.27 38.24 NaN 3 4964 5.3036817E+01 1.79E+01 33.75 23.78* NaN 4 4964 4.9702273E+01 1.43E+01 28.68 20.21* NaN 5 4964 1.8684995E+02 1.19E+02 63.66 44.85 NaN

Attached Sindarin File includes only important Commands.

Attachments (1)

higgs_strahlung3.sin (381 bytes) - added by msekulla 11 years ago.

Download all attachments as: .zip

Change History (6)

Changed 11 years ago by msekulla

Attachment: higgs_strahlung3.sin added

comment:1 Changed 11 years ago by Juergen Reuter

Damn, who is taking care of this one? :(

comment:2 in reply to:  1 Changed 11 years ago by kilian

Priority: P3P0
Severity: normalblocker
Status: newassigned
Version: 2.2.0beta

Replying to jr_reuter:

Damn, who is taking care of this one? :(

Me. We talked about this before submitting the ticket. I think I know what's wrong.

comment:3 Changed 11 years ago by kilian

WTF? I can't reproduce this.

Happens w/ gfortran 4.8.1, default optimization, OpenMP enabled (but 1 thread).

No problem w/ gfortran 4.8.0.

comment:4 Changed 11 years ago by kilian

reproduced. I had to switch on openmp. Also, the numbers fluctuate slightly between runs, which does not happen w/o openmp.

comment:5 Changed 11 years ago by kilian

Resolution: fixed
Status: assignedclosed

After looking in wrong places for hours, found the culprit: the openmp loop in the phase space evaluation was broken. Occasionally, the output is garbage.

The bug was probably present in 2.1 as well. This piece of code is apparently unchanged.

Fixed in r5419. Also see comment in #521.

Note: See TracTickets for help on using tickets.