avoid that mirroring is reset by the plugin

[oliviermattelaer]

Hi @valassi,

This is a PR to fix the issue that
p p > t t~ returns the wrong cross-section (compare to pure madevent case: (see #678)

The issue that I found is that the plugin reset the "mirror" flag to False and since this is done "before" the creation of the madevent metadata file, the madevent loose track of such symmetry (and therefore of a part of the cross-section).
This explains why g g > t t~ was not impacted but that q q > t t~ was wrong by a factor of two.

The question is why did you implement such line (and what is going to break by removing such line).
To avoid too much side effect, I have just removed them for the madevent+cpp/cuda mode?
But we likely need to take a more drastic option here (but want your opinion on it)

Olivier

[oliviermattelaer]

let's merge this and then see if we break something (release first fix later strategy)

[oliviermattelaer]

This patch fixes the issue for the fortran mode but not for the cuda/cpp mode.
In the cpp mode, all the events written on disk have (IMIRROR=1)
and the reason for that is clear when running the ../madevent_cpp < input_app.txt

 selecting           1           1           1
 selecting           1           1           1
 selecting           1           1           1
 selecting           1           1           1
 selecting           1           1           2
WARNING! flagging abnormal ME for ievt=0
WARNING! flagging abnormal ME for ievt=1
WARNING! flagging abnormal ME for ievt=2
WARNING! flagging abnormal ME for ievt=3
WARNING! flagging abnormal ME for ievt=4
WARNING! flagging abnormal ME for ievt=5
WARNING! flagging abnormal ME for ievt=6
WARNING! flagging abnormal ME for ievt=7
WARNING! flagging abnormal ME for ievt=8
WARNING! flagging abnormal ME for ievt=9
 Error in matrix element
 Error in matrix element
 Error in matrix element
 Error in matrix element
 Error in matrix element
 Error in matrix element
 Error in matrix element
 Error in matrix element
 Error in matrix element
 Error in matrix element
 selecting           1           1           2
WARNING! flagging abnormal ME for ievt=0

So all event with imirror=2 have a cpp issue and returns 0 (and therefore cross-section is ~.5 the correct value and events are biased in the way they are written.

[oliviermattelaer]

I guess that the issue is that the generated code for the amplitude is

      // Wavefunction(s) for diagram number 1
#if not( defined __CUDACC__ and defined MGONGPU_TEST_DIVERGENCE )
      imzxxx<M_ACCESS, W_ACCESS>( momenta, cHel[ihel][0], +1, w_fp[0], 0 ); // NB: imzxxx o\
nly uses pz
#else
      if( ( blockDim.x * blockIdx.x + threadIdx.x ) % 2 == 0 )
        ipzxxx<M_ACCESS, W_ACCESS>( momenta, cHel[ihel][0], +1, w_fp[0], 0 ); // NB: imzxxx\
 only uses pz
      else
        ixxxxx<M_ACCESS, W_ACCESS>( momenta, 0, cHel[ihel][0], +1, w_fp[0], 0 );
#endif

      omzxxx<M_ACCESS, W_ACCESS>( momenta, cHel[ihel][1], -1, w_fp[1], 1 ); // NB: opzxxx o\
nly uses pz

which due to the "mz" and "pz" assumes the sign of the z direction of the events.
This is problematic when using mirroring since those assumptions are wrong.

• weirdly using ixxxxxx fixes the issue.

[valassi]

Hi @oliviermattelaer thanks for asking the review :-)
I had a look, in my tests I do not see the benefits (but I vaguely understand you may see a factor 2x in the xsec in your own tests), but I also see no problems. Or better, there were some hickups, but you fixed them in a later commit, so that's sll good for me.