WZ Born integration with Amegic in 2.1.1

[Frank Siegert]

I'm encountering a rather obscure Amegic bug when trying to integrate lllv at NLO with Amegic in 2.1.1. It can be simplified so much that it already happens with the Born contribution, albeit not for a simple LO run card. Strangely, the issue can be reproduced only if using two different processes (separately or both of them within one process group):

(run){
  BEAM_1=2212; BEAM_ENERGY_1=6500;
  BEAM_2=2212; BEAM_ENERGY_2=6500;
  SCALES VAR{sqr(90)};
  FRAGMENTATION=Off
  MI_HANDLER=None
}(run)

(processes){
  Process 93 93 -> 11 -11 11 -12;
  Order_EW 4;
  NLO_QCD_Mode Fixed_Order;
  NLO_QCD_Part B;
  End process;

  Process 93 93 -> 11 -11 13 -14;
  Order_EW 4;
  NLO_QCD_Mode Fixed_Order;
  NLO_QCD_Part B;
  End process;
}(processes)

(selector){
  Mass 11 -11 5.0 E_CMS
}(selector)

The integration of the first process runs without any problems and is compatible with Comix and also with Amegic's normal LO integration (i.e. without NLO_QCD_* lines). The second process though does converge badly and only to a very wrong cross section:

Process_Group::CalculateTotalXSec(): Calculate xs for '2_4__j__j__e-__e+__mu-__nu_mub__QCD(B)' (Amegic)
Starting the calculation at 14:54:02. Lean back and enjoy ... .
0.078064 pb +- ( 0.0394354 pb = 50.5168 % ) 5000 ( 5493 -> 91 % )
full optimization:  ( 1s (1s) elapsed / 41s (41s) left ) [14:54:02]
...
0.101602 pb +- ( 0.00292004 pb = 2.874 % ) 830000 ( 958952 -> 86.4 % )
integration time:  ( 1m 31s (1m 28s) elapsed / 8h 53m 24s (8h 35m 17s) left ) [14:55:33]
...

This compares to ~0.139 calculated without any problems both by Comix and Amegic LO.

Does anybody have an idea, why Amegic's B(VI) behaves so differently from the normal LO process, and why the one process would affect the other here? Maybe the integrators?

[Frank Krauss]

My suspicion is that this is because you do not apply any minimal mass cut on the [13 -14] pair. You sometimes pick up funny configurations where the intermediate W has no mass, and the pair is collinear. This leads to denominators and numerators going to 0 ... and a tricky cancellation; COMIX does this better than AMEGIC due to its structure. Can you repeat the exercise with a mass cut of, say, 1 GeV?

[Frank Siegert]

Thanks for the suggestion. I've just tried this by adding a 5 GeV cut also on the lv pairs but unfortunately it doesn't change the picture.

I can also add, that if I change the order of the two processes it's again the second integration which misbehaves. So there is no inherent problem in the different-flavour channel, but only that it's not the first process.

[Frank Siegert]

Actually, I just realise that this seems to be working fine on trunk. Maybe Stefan's recent Amegic library restructuring has helped to avoid some kind of accidental library name clashes? Now I'll just have to understand what we can/cannot do with 2.1.1 to generate the 4-lepton MEPS@NLO samples.

[Frank Siegert]

I can confirm that Stefan's library restructuring seems like the fix, since r24391 (including the changes from r24390) is the first revision where this works properly. Unfortunately I don't think this is something that can be done in 2.1.1 without a patch.