Exploring Jet Substructure in Semi-visible jets

Dear referee,

We thank you for your review, and apologise for the delay in responding. One of the authors got busy in start of the term activities, other was moving countries to start her MCNet short-term student ship.

We completely agree with your assessment, this is meant to be a first exploratory study, far from the last word on this exciting area. Your suggestions definitely improve the quality of the paper. Please find our responses inline for your suggested changes:

( 1) Add investigation of more backgrounds: - Z+jets (Z->tautau). - W+jets (W->lnu, l=e,mu,tau)

We have expanded the text by adding a paragraph discussing the subdominant backgrounds:

Apart from the multijet process, which is the dominant background, $W/Z$ + jets processes can contribute to the background. However, the processes with one or more leptons can be almost completely rejected by vetoing events with leptons. The $W \rightarrow \tau (\textrm{had})\nu$ and $Z \rightarrow \nu\nu$ processes result in irreducible backgrounds, but since the large radius jets will still come from quarks or gluons, the considerations for multijet events still hold.

(2) Investigate the impact of using Madgraph+Pythia8 for the multijet background, with up to 4 partons in the final state). Ideally, could just change the background model to use Madgraph+Pythia directly.

We have generated events with Madgaph (upto 4 partons in fina state, as suggested) +Pythia8, using CKKWL matching, to check if conclusions change, which do not. We attach the plots here for your consideration, and in text we added a line.

Please see the attachment.

(3) Add a discussion about experimental considerations, given that current MET+jets searches impose a >cut on dphi(MET,jet)>0.4 to remove the impact of mis-measured jets. The goal is to explain how >experimenters can avoid being affected by mismeasured jets if searching for semi-visible jets at the LHC.

We added:

It must be noted though, that while multijet events at particle level mostly have low values of missing transverse momentum, at detector level, due to mis-measurement of jet energy and direction, a large fraction of events acquire large values of missing transverse momentum. This is essentially an irreducible background. Another possible origin missing transverse momentum close to jets arise the jet area includes dead calorimeter cells. In experimental searches, this is typically accounted for by removing jets when the angular separation of jets and missing transverse momentum direction is less than $\Delta \phi < 0.4$. For a signature like this, that requirement is clearly unusable, but events where jet areas include dead calorimeter cells can be discarded, and in ATLAS this is seen to be a small fraction of events~\cite{Aad:2020aze}.

(4) Improve the theory section: - add a discussion to explain exactly what is generated for the signal, and include Feynman diagrams as >appropriate. Similarly, need to fully explain the background process physics, rather than just referring to "HardQCD" >process. - improve the discussion of the model in the theory section, including a full description of all >parameters and focussing on aspects that affect this particular analysis (rather than general >considerations of effective lagrangians)

We expanded the text (as best as us experimentalists can do!), and added Feynman diagrams as requested.

(5) Some aesthetics: - why not use a subscript "d" for all dark mesons. Makes it easier to read.

Done

• can the legend labelling be improved? i.e. in fig 4 Signal, r_inv=0.5 is far more meaningful than Sig50.

Done

• C2 and D2 seem to be identical. Needs fixing. Why do you need to discuss D2 at all? there are no plots >of D2 in the paper.

We think it adds some information to say D2 was seen to be similar, so kept that.

The revised draft has been submitted to arXiv, it should appear Tuesday (16th March) morning.

Best regards, Sukanya and Deepak

Attachment:

ObsComp.pdf

We thank Tilman for this review, and suggesting ways how the physics message and the presentation can be improved. We are glad that he feel that the paper explores a new area, which is exactly what authors intended, even though we completely agree that this is a first look at this, which should serve as a baseline for further work. We must also admit that we really like the Scipost experience!

Report The paper makes for an interesting contribution to an active and relevant field. It is not a break-through, but it is definitely worth publishing. It also defines a very nice starting point for future studies, because it is very well documented. Requested changes 1- for the model, please give a Lagrangian or some appropriate definition in the beginning;

Since the model was not ours to begin with, we thought it's best to contact Tim Cohen, and he suggested that apart from the description we have listed in pg2, for further clarification we can mention that when we refer to a “SU(2)_D” gauge theory with two fermionic states, this is just a canonical Lagrangian for fermions with covariant derivatives. Does this answer the question you had, or maybe we misinterpreted your comment?

2- along the same line, Sec. 4.1 includes too much code-related slang for most readers to follow;

We have tried to clean this up with the aid of a table and shorthand names for the parameters.

3- where is \lambda defined, before it appears in the last paragraph on p.3?

It has been briefly defined in the 2nd paragraph of the introduction. Are you looking for something else?

4- in the discussion of Fig.4 the authors talk about the leading/sub-leading svj, what does this mean in terms of physics?

Ideally we should get 2 svj along with normal jets. However if the svj are exactly back to back, there is no possibility of getting real/large values of MET. But, wrt svj typically we have noticed that usually the subleading jet turns out to be the svj, as can be seen in the adjacent cartoon, and we get a significant MET contribution to balance out the momentum conservation. [Please see attached cartoon {svj-topo.png}]

5- on p.5 the authors start a paragraph with `Among these observables...', and I am not sure I see where this argument is coming from;

Rephrased as “Among these observables, we have primarily focused on ECF2, LHA and tau21, and tau32 for this study.”

6- at the very end of Sec. 3 the authors mention detector effects. For low-level variables we have found that detector effects can make a huge difference, why is this different here. Please document as well;

Absolutely. Since the study used Rivet, it was straightforward for us to implement the detector smearing proposed in [1]. While that made the differences between signal and background slightly less pronounced, the main conclusions were not affected. We added a line in text: Smearing of the substructure variables makes the peaks somewhat diffused and the difference between the signal and background slightly less pronounced. [1] https://arxiv.org/abs/1910.01637

7- the discussion of ISR and ME jets on p.8 is very interesting, but I have to admit that I do not understand the physics reason for the different impact, please explain that or elaborate. Could that be related to heavy-flavor jets?

We were also surprised by the observation. While HF jets can definitely contribute, our guess is that ISR jets are more isotropic so can overlap with SVJ, while ME jets are more well separated. We added this in the text.

8- little thing, DH in Fig.8 is not defined (as far as I can tell);

Defined in Fig8 caption now.

9- please cite some of the high-level or low-level analyses which were done before Ref.[31]. Not sure how many there were, but we certainly did one with our autoencoder, also referenced in Ref.[31].

We have added the relevant references from ref 31, in the conclusions paragraph.

Attachment:

We thank the referee for this thorough review of the paper (specially during the holidays), and suggesting ways how the physics message and the presentation can be improved. We are glad that they feel that the paper explores a new area, which is exactly what authors intended, even though we completely agree that this is a first look at this, which should serve as a baseline for further work.

Please find our responses inline.

Requested changes

1. On the face of it, heavy flavour jets (especially charm), which more often contain neutrinos, would be a background of concern, but they are not discussed, or studied (unless g --> ccbar included in the shower? if so maybe this would be the dominant source anyway). The authors should either study them, or else explain their rationale for neglecting them, and estimate the impact.

That is an interesting point. We used “standard” Pythia shower: HardQCD:all = on, which includes HardQCD:gg2ccbar as mentioned in Pythia manual: http://home.thep.lu.se/~torbjorn/pythia82html/QCDProcesses.html

We definitely agree that a LL generator like Pythia would probably not model the kinematics correctly. In order to convince ourselves, we generated just gg->ccbar events, and the MET distribution is attached is slightly steeper, but JSS observables are almost identical.

So we think it is reasonable to say while semileptonic decays of charm can be a dominant background, with a MET requirement of 200 GeV, we can reduce it significantly. We added: Pythia8 simulation does take into account the effect of heavy flavour jets created by gluon splitting.

1. Is the background only 2->2 QCD, or are 4 jet MLM matched configurations included? Please clarify/comment on the impact.

It is 2->2 QCD by pure Pythia8. In order to address this concern, we generated background events with MG5+Pythia8, MLM matched, which is how the signal was generated. A comparison with current background distributions showed no significant difference, which we think is expected as the we are using a fairly large jet pT bin, and the shower is mostly responsible for generating the substructure.

We added: As we are mostly concerned with the substructure of the jets, the lack of higher order matrix element in background simulation is not a concern.

1. On p5 the authors state that it makes no difference whether all quark or all gluon initiated jets are used. This is surprising. Can you give more detail on what evidence led to this conclusion?

Indeed, we were surprised as well, and in fact this is one of the first things we checked. We are afraid we used Pythia as well, using HardQCD:gg2gg HardQCD:qqbar2gg And HardQCD:gg2qqbar HardQCD:qq2qq In each case. We are however aware that gluon jet modelling is less constrained in most generators including Pythia [1].

[1] https://arxiv.org/abs/1704.03878

1. The paragraph at the top of p6 is an example of confusing grammar, especially the sentence with "actually not the case". What is really meant here? Can you write it more clearly?

Rephrased as: “Therefore, if the svj is more multi-pronged than two-pronged, then these two classes of observables can appear to show contradictory characteristics, i.e. tau21 will indicate that the svj is at least two pronged, whereas ECF2 will state it is not two-pronged.”

1. Please provide more detail/evidence on the "quick check" of detector effects.

Absolutely. Since the study used Rivet, it was straightforward for us to implement the detector smearing proposed in [2]. While that made the differences between signal and background slightly less pronounced, the main conclusions were not affected. We added a line in text: Smearing of the substructure variables makes the peaks somewhat diffused and the difference between the signal and background slightly less pronounced. [2] https://arxiv.org/abs/1910.01637

1. Related to 5, was pile up considered? Can you comment on its potential impact?

We are afraid not, however we did use trimmed large-radius jets, which should mitigate the effect of soft-radiation to a certain extent. It is a bit challenging to simulate pileup at particle level, but a priori we would think it will have the same effect of detector smearing, i.e “diffuse” the distributions somewhat, but would not dramatically shift the peaks.

1. I think I understood the paragraph at the top of page 8, but only after reading the later section. It is very unclear, can you please try to clarify it? This would seem to be quite critical to the overall conclusion.

Rephrased as: “Considering a background-like event selection, along with the \rinv$=0$ condition, if the substructure of the signal jets resemble that of the background jets, then that would give us some confidence that the difference seen for non-zero \rinv values, as seen before, are caused not only by the model specifications but also involve the effects owing to the dark hadrons.”

1. The second sentence of the conclusions seems to say that by choosing specific model parameters you can reduce the model dependence. I think I can work out what is intended, but as written it doesn't really make sense. Please clarify.

Rephrased as: “ We demonstrated that specific hidden valley parameter configurations can reduce the dark shower model dependent features of the signal jets.”

Thanks again, Sukanya and Deepak