Shower scale in single top t channel

Dear experts,

I am generating the single top t-channel at NLO

p p > t b~ j $$ w+ w- [QCD]

Now, for the official sample production of the CMS collaboration we have hit a point where we might need advise and help from authors.

Let me briefly explain the situation:

We use a customized setscales.f (https://github.com/cms-sw/genproductions/blob/master/bin/MadGraph5_aMCatNLO/cards/production/13TeV/st_tch_4f_ckm_hwpp_NLO/st_tch_4f_ckm_hwpp_NLO_setscales.f)

to take the spectator b kinematics into account.

The nominal sample is showered with Pythia8, and we do not encounter any problems with showering there.
However, we also want a systematic sample showered with Herwig++. (i.e. just exchanging PYTHIA8 with HERWIGPP in the run_card.dat).

When we try and shower the so produced aMCatNLO events with H++, we sooner or later hit an event where Herwig++ jumps into an infinite loop and does not proceed to the next event!

After carefully investigating when this happens, we found out that in all such cases the starting scale of the shower is exactly 3.
I looked into the code, and it seems the scale is set in the set_shower_scale routine of the fks_singular.f, where the following final check is performed

SCALUP(iFKS)=max(SCALUP(iFKS),3d0)

This means, before this, the SCALUP(iFKS) was smaller than 3.

Now, I have basically three questions:

1) (Probably rather a question to the H++ authors, but who knows ...) Do you know why Herwig++ is jumping into an infinite loop (it seems it dies in endless LHAPDF calls) with such events?

2) Do you know why, when generating 1M events with the run_card setting PYTHIA8 and HERWIGPP respectively, there are

326 events for PYTHIA8 where the shower scale is exactly 3

but

4674 events for HERWIG++ where this is the case? (Sometimes the problem occurs only after many many thousand events of the 1M, so there must be events with a shower scale of 3 where Herwig doesnt have problems!)

3) How exactly is the shower scale determined? For the leading order code of madgraph, it is always equivalent to the renormalization scale, but this is not the case for NLO ... otherwise one would never get to a 3 here, because the b mass is entering the renormalization scale formula, so this scale is at least 4.7 GeV. (and I have checked this with a minimal example of p p > t t~ as well that the shower scale and ren. scale are only identical when using the LO code, but not at NLO).

and maybe as a fourth question:

What would be the way you'd suggest to proceed? Like this, at the moment we just cannot produce aMCatNLO + Herwig++ events for single top in CMS ... The Powheg t channel in the 4fs works with Herwig++. There is no such an event where Herwig++ jumps into the infinite loop! ...

Thanks in advance for the help, in particular on behalf of CMS!

Benedikt

PS: All the necessary information besides the setscales.f can be found in:

https://github.com/cms-sw/genproductions/tree/master/bin/MadGraph5_aMCatNLO/cards/production/13TeV/st_tch_4f_ckm_hwpp_NLO