Sherpa 2.2.2 - Cannot reproduce filtered parton-level quantities using status 3 particles
Hi Sherpa experts,
I've been looking into the efficiency of parton-level m_jj filtering, using a Z to nu nu sample generated with Sherpa 2.2.2 (for the ATLAS exotics VBF Higgs to invisible search). In particular, we want to understand how likely it is for events in the low slice to end up with a post-parton-shower mjj in the high slice, to understand if this is a feasible strategy.
The sample in question has five parton-level mjj slices, and was also filtered in max(HT, pTV) between 140 and 280.
I've been attempting to reproduce the filtered variable, the pre-parton-shower mjj. Our understanding was that this could be done in Sherpa 2.2.2+ by looking at the status 20 particles in an event. For events without status 20 partons, we could instead look at the status 3 partons.
This specific filter could then be reproduced by clustering the partons into parton-level jets, taking the two with the highest pT, and computing mjj from them. Unfortunately, when I tried to do this (using just the 0-500 slice), I found there were a fraction of events for which I could not reproduce the filter-- events that have a parton mjj above 500, and therefore should have been rejected by the filter if these were really the particles that it processed.
If I then split the sample into S events (with status 20 partons) and H events (those without), we see that using the status 20 partons does perfectly reproduce the filter. All the events have parton mjj < 500 GeV.
It's the remaining events, the "H" events that only have status 3 partons, where I don't seem to be able to reproduce the filter.
(On these plots, "Truth Mjj" is the dijet mass computed by looking at the post-shower truth jets, whereas parton mjj is computed from clustering the status 3 or 20 truth particles. In both cases, it's computed by taking the two highest pT jets and adding them together).
I presented this material at an ATLAS PMG V+Jets group meeting on August 1st. There are more details and plots in my slides from that meeting-- a few other interesting points:
- There are similar issues if we look at the distribution of max(HT, pTV).
- Most-- but not all-- of the events that have parton-level mjj > 500 are also negatively weighted.
- All five slices of the sample seem to have this problem.
- In the 0-500 slice, 40% of the events are S events with status 20 partons. (In the higher slices, the percentage of S events drops off significantly).
- There are never more than two status 20 partons in an event. Each such event has one additional status 3 parton, which Frank says comes from the beginnings of the parton shower.
This may imply that some extra process can happen between generating/filtering a H event and writing the status 3 particles to the HepMC output?
At the meeting, Frank asked me to open a Sherpa ticket to continue the discussion. Let me know if there's any other information I can provide or checks I can do that would be helpful!
Ben Rosser