Probability distribution of unweighted events

Hi,

You have many contribution to the final weight
1) the matrix-element --you can access it be using the momenta from the lhe file and use standalone mode to evaluate it--
2) the pdf --you can access it if you have use_syst=True in your run_card, then the value used for x and the scale are written in the file and you can call the pdf as you want.
All the rest are related to the phase-space
3) the multi-channel weight:
  - madgraph split the integration in multiple channel and depending of the channel that generate your events, it will be associated to a different weight. If we sum those weights for all the channel the sum equals one (it is related that in principle the same event could have been generated by any channel)
 - The information of which channel was used to generate that particular event is not include in the lhe file in general -- even if it is sometimes possible to know from the lhe information (presence of resonances particles in the lhe and/or from the color-flow)
4) The jacobian of the change of variable
We perform a bunch of change of variable and all those are associated to jacobian which contributes to the wieghts. So you need to know which change of variable are used for each channel to get the weights.
5) The MC grid, the code use an adaptative MC method to increase the efficiency of the un-weighting.
In principle you have all the information here to know that information --if you know which channel was used--
Since the grid are written on disk. But this is quite complex (extremelly complex). --and you need to be able to inverse all the change of variable discussed in 4 in order to know in which bin you are to get that weight.
6) the cuts on the phase-space. --this is a simple 0/1 weight--

My guess is that you are comparing at fix scale computation, while MG5aMC, is using a dynamical scale,
which leads to distorsion since we change the value of the pdf/matrix-element. (which is more accurate compare to not running the scale in experiment like LHC, since you can absorb some log --theoretical error- in that way)

Cheers,

Olivier

> On 17 Jan 2019, at 09:34, Ponf <email address hidden> wrote:
>
> New question #677857 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/677857
>
> Hi,
>
> as the title says I'm looking for the probability distribution that unweighted events follow.
>
> If, for example, I generate events for u u~ > t t~ and look at the invariant mass distribution of the tops, the distribution is slightly different from what I get when I generate events myself using the (same) PDFs and the differential cross section from the PDG.
> I would have expected that the events are distributed following the product of PDFs and differential cross section but that is apparently not the case.
> Are there any other (weight) factors and if so, is there a way I can access them in an LHE file?
>
> Cheers!
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Thanks Olivier,

does MG5aMC still use a dynamical scale although I set

True = fixed_ren_scale ! if .true. use fixed ren scale
True = fixed_fac_scale ! if .true. use fixed fac scale

in the run_card.dat?

Cheers

Thanks Olivier Mattelaer, that solved my question.