Required accuracy in spinor construction too high?

[Frank Siegert]

When integrating ttbar+3jets with Comix I get the following error message:

Process_Group::CalculateTotalXSec(): Calculate xs for '2_9__j__j__t[W+[lepton__neutrino]__b]__tb[W-[lepton__neutrino]__bb]__j__j__j' (Comix)
Starting the calculation. Lean back and enjoy ... .
Spinor<Scalar>::Construct(): \sqrt{p^+p^-} = 1.78401 vs. |p_\perp| = 1.784, rel. diff. -7.69789e-06.
Sherpa: Spinor<Scalar>::Construct throws fatal error: 
   Cannot construct massive two-component spinor.

It doesn't make a difference if I set NUM_ACCURACY=1e-4. The run card is

(run){
  BEAM_1=2212; BEAM_ENERGY_1=3500;
  BEAM_2=2212; BEAM_ENERGY_2=3500;
  ME_SIGNAL_GENERATOR = Comix
  WRITE_MAPPING_FILE=3
  MASSIVE[15]=1
}(run)

(processes){
  Process : 93 93 ->  6[a] -6[b] 93 93 93
  DecayOS : 6[a] -> 24[c] 5
  DecayOS : -6[b] -> -24[d] -5
  DecayOS : 24[c] -> 90 91
  DecayOS : -24[d] -> 90 91
  End process
}(processes)

(selector){
  METS sqr(20/E_CMS)
}(selector)

Is it the METS selector which is failing to identify a divergence here or is the numerical accuracy simply too low?

[Stefan Hoeche]

The accuracy in Spinor construction is changed to rpa.gen.Accu() with trunk r16771. This allows to use NUM_ACCURACY here to circumvent the problem.

[Stefan Hoeche]

Milestone rel-1.3.0 deleted