Reweighting in the particle mass?

This is a plot comparing the dedicated simulations with reweighted ones.

https://www.dropbox.com/s/smtf58lgeumrich/NLO_fatjetpt.png?dl=0

For the couplings it is okay, but the larger the deviation in the mass from the nominal one, the more obscure the results get. They do into the opposite direction, in fact. ;(

Benedikt

I think Olivier Mattelaer is back now and will be able to answer more precisely to your questions.

However, I believe that reweighting any kind of mass is not possible, simply because their impact on shape, i.e. on kinematics is drastic and therefore you basically lose all efficiency in your reweighting procedure.

This is obvious in the case of the a mediator mass where the sharp Breit-Wigner shape makes it clear that moving the mediator mass by just a couple times the width will send you to a region basically completely unpopulated (since there, in the original hypothesis samples, the cross-section is suppressed by several orders of magnitude and you have no events).

For the case of final particle masses, I believe that you still have a very strong impact on kinematics, clearly strong enough to spoil your reweighting procedure.
You can convince yourself of this by looking at the analytical expression of 2 > 2 process via s-channel resonances where you vary the mass of the final states; it significantly impacts your distribution of \hat(s).

I think Olivier Mattelaer is back now and will be able to answer more precisely to your questions.

However, I believe that reweighting any kind of mass is not possible, simply because their impact on shape, i.e. on kinematics is drastic and therefore you basically lose all efficiency in your reweighting procedure.

This is obvious in the case of the a mediator mass where the sharp Breit-Wigner shape makes it clear that moving the mediator mass by just a couple times the width will send you to a region basically completely unpopulated (since there, in the original hypothesis samples, the cross-section is suppressed by several orders of magnitude and you have no events).

For the case of final particle masses, I believe that you still have a very strong impact on kinematics, clearly strong enough to spoil your reweighting procedure.
You can convince yourself of this by looking at the analytical expression of 2 > 2 process via s-channel resonances where you vary the mass of the final states; it significantly impacts your distribution of \hat(s).

Hi Benedikt,

Sorry, the last part of my previous answer is incorrect. In a 2>2 process via an close-to-onshell s-channel resonances, varying the small mas of the decay products is not going to change much, you are right (as underlined by the fact that your width does not change by much.

I think that what happens instead is that your reweigthing weight 'rewgt' is, for each event 'evt' extracted from your sample:

rewgt(evt) = new_wgt(evt) / orig_wgt(evt)

Now the event kinematics is the same in the numerator and denominator on the r.h.s since the reweigthing procedure does not change the kinematic of each event individually.
This means that the numerator new_wgt computes the matrix element with a xd mass of 300 GeV in your example but the final state momenta are still at p^2 = 10 GeV.
This cannot lead to consistent results since our framework can only compute onshell matrix elements; in other words the polarization vectors used to compute the new_weight will not make sense since they do not describe onshell particles in this case.

So the conclusion is that you really cannot use the reweighting procedure to change any mass, but for the final states ones you are considering here, the explanation as to why differs from the usual one that applies to propagator masses for example.

Hi Valentin,

Thanks a lot for the detailed words. I think I begin to understand; however I have one more question - what exactly do you mean with "only on-shell matrix elements" in this case? 2x300 GeV is still smaller than the mediator mass, and also I have tried to reweight the other way around - from 300 to 10 GeV, and it wasn't working either.

Thanks again,
Benedikt

Sorry, I was not being clear. I was talking about the *external particles* (your dark matter xd) being on-shell vs off-shell.

If your original hypothesis (the one you used to generate events) has M_xd = 10 GeV and you then reweight it with an hypothesis where M_xd = 300 GeV, then the reweighting procedure MG5aMC will try to evaluate the matrix element with the internal xd mass parameter set to 300 GeV but still with external final state momenta with p^2 = 10 GeV (hence xd is then off-shell; that's what I meant).
This is of course inconsistent and the reason why reweighting cannot work in this case.

Thanks Olivier Mattelaer, that solved my question.