Dark showers kick off meeting

(15 October 2020)

Indico (including Zoom recording): https://indico.cern.ch/event/965219/

Intro (Suchita Kulkarni)

Primarily concentrate on SU(N) theories with hadronic final states: study the signature/theory space; what tools are needed?

Initial wishlist: comparison of benchmarks currently used @ LHC, which theory maximises signature overlap

Actions: centralise in public repo: model files, generator settings, validation plots; understand signatures, DM implications, commonalities across models

Github: https://github.com/dark-showers-snowmass21

Mailing list: dark-showers-snowmass21

Timeline: whitepaper (draft: May, submission: end July)

• Stefan Prestel: action items; making plots – useful to have some common plotting tool/code for analysis somewhere public

• Suchita: some preferences for Rivet / delphes / etc may already exist, so the framework may differ, but scripts used should be in repo so that people could build on available code

• Caterina Doglioni: github repository is where this should happen so that people can use other’s tools. Important to have documentation and we can then see how different groups tackle the problem, and eventually congregate around one analysis tool.

Finding dark showers at the LHC (Elias Bernreuther)

~10 dark mesons per event / most at Etmiss aligned with a jet (from rho_d decay)=> semi-visible jet; no public result from LHC on small dphi(met,jet) yet

Can we use ML to ID these jets? S-v jet images look very similar to QCD

• Stephen Prestel: Are jets made of all particles or just charged particles? 

• Elias: made from all particles

• Stephen: Why do the semi-visible jets look the same as light QCD jets? 

• Elias: what is missing compared to top jets is the multiprong structure

• Sarah Eno: jet structure depends on dark meson mass; if they were heavier, it would likely be more different

• Elias: these are for meson mass of 5 GeV, but we’ve also looked up to 50 GeV or so and we find similar results

• Steven Lowette: Why are the images asymmetric? 

• Elias: because of pre-processing - this is not just the average image, but the images are rotated so that the constituents with the higher pT are on the upper right for example

Dynamic graph convolutional NN outperforms other architectures in tagging them – can use mixed samples to avoid training on specific DS parameters

LLP generic prediction in strongly interacting DS (can lead to DV+MET, important to probe small impact param)

• Benedikt: monojet used, but for semi-visible wouldn't be more QCD BG 

• Elias: dependence on parameters, interesting to study tagging, but difficult to include mismeasured jets as a theorist

• Sarah: DV - how is the limit shown comparing to CMS emerging jet search? (https://arxiv.org/abs/1810.10069)

• Elias: difference in the model as bifundamental mediator (2 q + 2 dark q final state); haven’t looked yet, but doubt there would be good sensitivity because of that

• Felix Kahlhoefer: we typically get 1 DV here while emerging jets require a lot of activity

• Sarah: (Discussion here about comparison of CMS emerging jet search and Elias’s presentation puts in perspective the need of comparing models)

• James Beacham: there is a Z’ here, which is different from the emerging jet 

• Stefan Prestel: what does the point cloud learns that makes it more powerful (fine details?) 

• Elias: (missed...) reconstructing the mass is the main point

• Steven: when comparing monojet with a tag and you get a factor 20, is it comparing no-tag to tag and with the graph method?

• Elias: yes

• Steven: dependence on the model… maybe some uncertainties on that, so the factor would likely be smaller…

• Suchita: maybe we should think more about machine learning in the LOI work. 

• Kari Folan DiPetrillo: DV analysis from ATLAS? Surprised as that requires 5 tracks

• Elias: yes, in the majority of events get one displaced vertex

Dark showers and LLP community (James Beacham)

First LHC LLP workshop in 2016 and many others since, with discussions on dark showers, included in the LLP community white paper (chapter 7)

Focus more signature-based, not so much on benchmarks/models

Production through H, Z’, bifundamental, … ; shower and hadronisation depends on coupling, gauge group (jets, soft radiation patterns…); decays back to SM with wide range of lifetime, spectra…

Looked at Pythia HV parameter effects (from jets to SUEP), but also vincia: basically the validity of e.g. pythia8 hidden valley module is not completely tested (does it reproduce QCD where it should?)

Dark shower session at the November LLP workshop: https://indico.cern.ch/event/922632/

• Sarah: paper from the LUND group on HV (from original release of HV model) - are they enough?

• Stefan: validation will be necessary, as these were exploratory, it was not clear if it would be something people would actually use; worries me as a pythia developer

• James: it was indeed my understanding that validation was still needed (reproduction of SM QCD) and updated significantly in 2017 when running coupling was added centrally

• Sarah: that in particular, I verified

• Stefan: no guarantee the coupling would run as in QCD, so it was deliberate not to include the running coupling at the beginning; need people to do validation

• Suchita: good that we have a group with developers, theorists and experimentalists, so this would benefit all to work on that

Round table 

Stefan’s slides on Pythia

- Dark production

Switching to simply dark quark production i.e. hard process is fairly straightforward

If hard jets are not QCD dominated, dealing with multi-jet events within pythia is not straightforward 

SUEPs or soft-bombs can’t be dealt with via les Houches interface

If we use code as a group, we should consolidate it. Pythia-contrib system (like Fastjet-contrib). Simple enough that allows you to upload stuff on top of Pythia.

• Sarah: we are hacking into Pythia to try and do this kind of thing (https://arxiv.org/abs/1803.08080 ). So it would be something that we need to discuss with the developers as we need to also modify some inner workings in the code; https://arxiv.org/abs/2009.08981 uses the KK code in Pythia to do soft bombs (need to check) 

• James: Simon Knapen is working with Cari Cesarotti et al for this, so it would be interesting to have Simon et al in this discussion as well

• Stefan: if there are such tricks, maybe can be used instead of a contrib, need to look at it; can involve Simon

- Dark showers

Available in collinear approx, the dark gluon radiation would not reproduce the gluon radiation, but dark shower is quite configurable, but the config has a steep learning curve - was more case-by-case at the beginning

Helpful to move to a config through SLHA cards? Need to understand how to prioritize the work

• Stephen: what kind of flexibility do we need? Why do we hack code? It is quick, but a lot of time put into hooks, etc - you should contact the developers to make your needs understood

• Sarah: in this particular case, the model was lacking flexibility

• Suchita: we would like to go around and ask everybody what their plans would be - we will send a template around on the mailing list, including wishes eg for extensions of pythia we would like to have

• Stefan: very valuable to help us prioritize

- Dark hadronization

If people need to understand / tweak this more, it would be useful to know how. Eg we don’t have glueballs - if the model is going towards dark baryons and glueballs, right now we can’t simulate it.

• Stephen: could “soft-bomb” like models be handled using the LHE interface?   Has anyone tried this?   It works for quantum black holes.

-

Suchita: we should meet again in about 3-4 weeks. Feel free to use the mailing list in the meantime.