comparing EPA with full matrix element calculation

Is that not an issue of the form-factor?

Did you compare the different implementation of EPA available in the code?
How did you set your scales here?

One typically hidden cut is associated to the dynamical scale which cut any event with too low scale (since this is designed for collision with proton to keep the perturbative regime).

Cheers,

Olivier

Re: form factor -- yes I did check this as well by simulating e t > e t + ax (top quark mass happens to be similar to my nucleus, so the kinematics is similar) and the equivalent EPA process. Again the cross-section for the full ME calculation is much larger than the EPA one.

Re: different implementations. I've tried the standard one from MG 3.1.1 where the photon distribution comes from PhotonFlux.f; I also tried the iww = pdlabel1, and eva = pdlabel1 options MG 3.4.0, both giving similar cross-sectons (both ievo_eva options also give the same result).

For all the EPA calculations I've used
.true. = fixed_fac_scale1
and varied dsqrt_q2fact1 from 1 to 3 GeV (the cross-section is very weakly dependent on this, as expected)

By the way, my beam energy is low, 4 GeV and the target nucleus is at rest. Could the dynamical scale cut be the culprit here? if so, how can I change it? Thanks!

I would suggest to use fixed scale for all parameter (both beam and factorization scale)

Cheers,

Olivier

> On 12 May 2022, at 06:15, nblinov <email address hidden> wrote:
>
> Question #701752 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/701752
>
> Status: Answered => Open
>
> nblinov is still having a problem:
> Re: form factor -- yes I did check this as well by simulating e t > e t
> + ax (top quark mass happens to be similar to my nucleus, so the
> kinematics is similar) and the equivalent EPA process. Again the cross-
> section for the full ME calculation is much larger than the EPA one.
>
> Re: different implementations. I've tried the standard one from MG 3.1.1
> where the photon distribution comes from PhotonFlux.f; I also tried the
> iww = pdlabel1, and eva = pdlabel1 options MG 3.4.0, both giving similar
> cross-sectons (both ievo_eva options also give the same result).
>
> For all the EPA calculations I've used
> .true. = fixed_fac_scale1
> and varied dsqrt_q2fact1 from 1 to 3 GeV (the cross-section is very weakly dependent on this, as expected)
>
> By the way, my beam energy is low, 4 GeV and the target nucleus is at
> rest. Could the dynamical scale cut be the culprit here? if so, how can
> I change it? Thanks!
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

I just tried this as well, but no luck unfortunately.
The "eva" EPA implementation does yield a higher cross-section compared to "iww", but both are still much smaller than the direct calculation ( factors of ~ 24 and 36, respectively).

Here are the run card snippets for these runs:

EPA (a t > ax t):
mixed = pdlabel
eva = pdlabel1 ! PDF type for beam #1
none = pdlabel2 ! PDF type for beam #2
 False = fixed_ren_scale ! if .true. use fixed ren scale
 True = fixed_fac_scale1 ! if .true. use fixed fac scale for beam 1
 True = fixed_fac_scale2 ! if .true. use fixed fac scale for beam 2
 3. = scale ! fixed ren scale
 3. = dsqrt_q2fact1 ! fixed fact scale for pdf1
 3. = dsqrt_q2fact2 ! fixed fact scale for pdf2

(this requires we generate the process with "set group_subprocesses False" as noted in 2111.02442; eva -> iww for the other implementation does not require this)

Direct (e- t > e- t ax):
all cuts are default except
 0.01 = ptl ! minimum pt for the charged leptons
-1. = etal ! max rap for the charged leptons

These settings lead to cross-sections with reasonably small uncertainties (for certain processes involving my nucleus I needed to play with -- t_strategy output option to get stable results as discussed in the appendix of https://arxiv.org/abs/2102.00773; however the defaults work fine for this top quark toy example).

You still have
 False = fixed_ren_scale ! if .true. use fixed ren scale

Can you put that one to fixed too?

I have tried this as well, for both iww and eva implementations, along with variations of the scale parameter.
These changes didn't have an impact on the cross-section unfortunately.

Good, this means that you do not face an hidden cut (which was my worry).

Now I do not agree that the two computation should match since this is a LO computation.
If you do not put any cut on the electron, that cross-section should actually be infinite (and technically cancel by NkLO contribution).
While the EPA actually do a resummation as needed (i.e. they do not have the same accuracy).

Cheers,

Olivier