General comments
- My biggest concern is the following.
In section 4 we give three categories of models in the three subsections. where subsection
4.3 contains "SPS+DPS" samples, and pointing out that PYTHIA and KATIE allow
for two 2->2 events.
Ok, but all MC generators we include in the hadronization step use PYTHIA or
HERWIG (or cascade). The tunes used in these models, and very explicitly in PYTHIA, include
MPIs, ie we include in fact for all of these generators
MPIs (and thus DPS). Or is the
DPS defined to be a different/additional process than what is covered by
MPI effects?
If so, I missed that...
For example if you combine
MadGraph with PYTHIA CP5 tune you always have
MPIs in this
combined result. The inclusion of
MPI is controlled by (the input of) Pythia via this
parameter:flag
PartonLevel:MPI (default = on)
i. e. this is the Master switch for multiparton interactions; on/off = true/false
IFF we did switch that off in these generator configurations, and thus the
MPIs, then of
course we should write that explicitly in eg section 4.2 or 4.1... But I don't think
we switched
MPIs off.
What it means is that combining the DPS templates from data with eg
MadGraph or
PYTHIA MCs is not a combination of DPS with pure SPS, as the "SPS" is contaminated
by
MPIs through these tunes used. I think that is diluting the message from Fig14.
as these intrinsic
MPI effects in the "SPS" MCs have an effect on all these
sigma_eff extractions. We do not discuss that.
So, my major comment, assuming that you did as I described above is correct, is not
explained well at all into paper, and should be done better.
Below I will come back in the details to a number of instances were we say something
in the paper on the DPS extraction. Of course if I missed something important in my
discussion above then these comments further down may loose their importance.
Yes you are correct. We assumed wrongfully that the MPI contribution from the UE tunes was small. What we found in the paper v17 is the amount of DPS necessary on top of the tune. In order to make qualitative statements about the total amount of DPS, we have performed the same extraction of the DPS cross section, but with a criterion in place that should leave out events with hard MPI jets originating from the UE description. Our criterion omits events from the sample if a parton originating from an additional interaction (status code 33) has a transverse momentum exceeding 20 GeV. This way we remove all events that could bias out results. We have updated the section on the extraction and the results in section 8.4 accordingly. - Just two week ago we released another paper on DPS studies, which is in CWR, but we make no
contact with that analysis. Are these analyses fully disjunct and the other recent one
has no bearing at all on the discussion in this paper?
Yes you are correct. We assumed wrongfully that the MPI contribution
from the UE tunes was small. What we found in the paper v17 is the
amount of DPS necessary on top of the tune. In order to make qualitative
statements about the total amount of DPS, we have performed the same
extraction of the DPS cross section, but with a criterion in place that
should leave out events with hard MPI jets originating from the UE
description. Our criterion omits events from the sample if a parton
originating from an additional interaction (status code 33) has a
transverse momentum exceeding 20 GeV. This way we remove all events that
could bias out results. We have updated the section on the extraction
and the results in section 8.4 accordingly."
While I added these lines starting at line 419 in the paper:
"Similar results have been obtained for the predictions of the \DptS
observable from the DPS tune in \cite{CMS:2021wfx}. Deviations between a
model employing a similar DPS tune (CDPSTP8S1 -WJ) and the data of
10-20\% were found in the Z+jets final state."
- The DPS cross section formula looks in eqn(1) looks rather simple, perhaps even a bit
naive, at Born level, as been the form we have been using for many years... Was there no
further phenomenological thought in the community if this is not too simplistic a view,
and has some better/evolved suggestions? (no worry if the answer is 'no'
)
Of course using what we did allows us to put different results on the
same plot (Fig 14) but I am doubtful it is very meaningful, which is supported by
the spread go he extracted numbers.
We have to do with the current version of thepocket formula. Strides are being made towards more complex models, but nothing that is developed up to a point that it is applicable in our study.
Details
- line 12: "..call for further studies of the strong interaction "
Does this really follow from the discussion above, they way it is written?
Replaced with "..., are not as well understood and, therefore, require and suggest further studies of the strong interaction."
- line 16: "A second approach"? -> "A different process"?
Changed accordingly.
- line 32 ".. resulting in a strong increase of the gluon density.." actually
it is the other way around: we have more interactions because the gluon density
is increasing strongly at low-x (as measured in e.g. electron-proton scattering
)
Rewritten as "...: at higher energy, smaller values of the proton’s momentum fraction carried by the partons can be probed, because of a strong increase of the gluon density, resulting in a largerprobability for DPS."
- line 49: do we see visible effects/changes by using this custom made calibration,
instead of the out of the box CMS numbers? Have these new calibrations been vetted by
the JETMET group?
Yes, large differences were observed, i.e. when examining the resolution of the jet pT ( (pT,gen - pT,det)/pT,gen ), the distribution was not centered around 1 but deviated from it, indicating that there was something wrong with the standard JEC. The JEC we derived have all been checked and approved by the JETMET group.
- line 57: What I found confusing first time around is the use of "inclusive
single-jet events as mentioned here. In fact I think what you mean is that these
are events triggered by a single jet trigger, but in fact they are (mostly) di-jet events,
so that when you merge two of these events you get (in most cases) a 4-jet event,
whereby the two 2-jet events come from a separate production vertex.
I guess when you know the analysis it is obvious but for a fresh reader it
takes a moment to get used to. :). I recommend therefore that one should spell
this out explicitly front he start.
We left the comment there as is, however, we tried to give a more extended and clear picture in section 6 where we detail the procedure more explicitly.
- line 138: ...rather old PDFs here from 2002 :(
But I guess we take here what we have...?
Yes, that is the case.
- line 170: Are you using a merging scheme here of the PS and ME for NLO programs,
like FXFX? With the LO programs we explicitly gave that we used the MLM scheme so if we
use one here as well, we should also mention that for the NLO and give a reference.
It is an MLM matching scheme as well. It is now stated explixitly in the paper.
- line 171: "2019.02.25" is that really the version number? So I guess the
version and the date are the same
?
Yes, it is like this.
- line 173:" for which the real phase space is build afterwards"
looks like slang. What does it mean exactly?
In the papers on dijet and trijet production in Powheg the same language is used. They calculate the Born cross section and afterwards apply a "multi-channel" technique to approach the singular regions. We felt that if we would mention the multichannel technique that we would have to explain it as well, making the text even more technical when it is besides the main point of the paper.
- line 188: what are the most important expected consequences for having on- and off-shell
incoming partons? For the off-shell ones we have to use TMD, and for the
on-shell we use kT integrated PDFs. But do we expect effects on the hadronic final state
phenomenology? If so this would be of interest for the reader to spell that out.
We added "In the case of the latter, the initial partons are generated with a non-zero intrinsic k T which can alter the momentum balance of the jets, yielding different topologies and correlations between the jets compared to on-shell production."
- line 188: "two-off shell" -> "two off-shell"
Also check on- and off-shell in the paper. I saw sometimes we forget to put the
"hyphen".
Changed accordingly.
- line 193: "For two the" -> "For the two"
Changed accordingly.
- KATIE: the program is combined with HERWIG or PYTHIA.
In lines 190-196: do we combine it such that NO DPS interactions are generated
or do they have the default ones from the CP5 and CH3 tunes? (see general comment)
We have the default ones from the CP5 and CH3 tune.
- line 205: Do I understand that this is the same sample as discussed in line 133, or is
there a difference? I believe it is the same but the repetition is somewhat confusing
(particularly since you mention for the next item only that this one has already
been mentioned before)
It is the same, we clarified it to: "A PYTHIA 8.240 sample is generated with the CP5 tune and the NNPDF2.3 NNLO PDFs. It is the aforementioned PYTHIA 8 sample to which a DPS contribution is added"
- line 222: So here we are using pre-scales for the low threshold triggers used?
What are these prescale factors?
We added the scentence: "...| η | < 4.7. As the triggers have been prescaled, they are used in disjoint p T ranges. Offline requirements..."
The prescales range from 1 to 25 depemnding on the trigger.
- line 235-236: This was not clear to me reading that here but I guess the
explanation of for is given explicitly in line 262, correct?
Yes, it is given later, we had to mention it here but explaining it fully in seciton 5 did not seem optimal to us.
- line 300: Recommend you say here how far down in pT the out of the box CMS calibration
works well, and that we need to bring it down to 20
GeV in this analysis.
Changed the scentence to: "The low pileup, low jet \PT data sample used in this analysis necessitates a dedicated JES calibration since the out of the box JES corrections factors did not hold up below $~50$ GeV."
- line 307: despite this effort on the calibration of the JER effects the uncertainty of the
cross sections seems still very large... Wouldn't we have expected to do better?
It is in line with previous results, the uncertainty rises rather sharp below ~40 GeV.
- line 358: "DPS pocket formula"
you just mean eqn (2)? Or did you have something
special in mind with "pocket"?
Changed accordingly.
- line 398: "...possible need for DPS" (see my large general comment and assuming
I did not overlook something in that comment): do we mean here "more DPS"? These models
all have
MPI'sat some level, no?
Yes, they need more DPS, changed accordingly.
- line 411: also all the KATIE cross sections are too high in region II. We don't
seem to mention that in the text.
Changed to "The \MGvATNLO LO samples and all the NLO samples predict cross sections that are roughly in agreement with the cross sections obtained from data in \regioni, but are larger than those from data in \regionii, as are all the \KATIE cross sections in the same region."
- Figure 6 legend: "PBMTD" is not directly related to the section 4.2 PDF terminology used.
(one can make a good guess of course, but just for definiteness). To be fixed.
We have defined MRW and PBTMD in section 4.2 now.
- line 436: "leave room for a DPS.."? They all have some DPS (see my general comment)
Perhaps "leave room for more DPS.."?
Changed accordingly througout the section.
- line 440: We don't study that here but the large DY cross sections could also be the
result of the fact that we the QCD approximation we are working with is not the right
one for such a measurement, (eg BFKL vs DGLAP although this may go in the wrong
direction here..)
Yes, but discussing it would take us too far of topic.
- line 448-450 The
DeltaS distribution in figure 9 shows in the bottom two plots
shows that many the models have essentially ratios with values below 1. Given that
we normalise on the high deltaS bin and the fact that the cross section in
table 3 predicted by the models are all larger than the data, this means that the
shape of the
DeltaS distribution for these 5 models is very different from the
data. Maybe something to take note of in the text?
We have extended to "The \DS distributions again show a more robust behavior with respect to the parton shower implementation. The LO \MGvATNLO samples leave less room for an additional DPS contribution compared to the NLO models. Due to the sole use of a $2\to 4$ matrix element, the \PT spectra are far too hard, resulting in a large overestimation of the slope for all the \KATIE models."
- line 460: This seems to be a pattern: all models we have discussed in this paper
give a too high cross section. This is a strong message. Are we sure nothing could have
gone wrong extracting this data cross section number for this region?
I imagine that has been given some thought?
Or do we have a hypothesis what it can mean?
When comparing litterature, similar studies all find the same. It is tied to the collinear factorization that most models use. It is known to fail at low pT. If one looks at KaTie off-shell (based on kt-factorization), one sees that it performs better all over the line. We made sure te check multiple times and perform cross checks.
- line 462: are these really pure SPS samples? (in view of my general comment above)
We removed the "pure".
- line 480: actually the P8 models don't seem to do to badly for the
DeltaS distribution...
This is already discussed in the text.
- line 490: I somehow failed to see why we drop that PYTHIA tune here at this stage...
what do I miss?
These are old tunes, we have the more recent ones CP5 and CH3, they are tuned to 13 TeV data as the CUET tunes are not. We only had to use these tunes in the analaysis to perform the unfolding. Both tunes did not yield any meaningfull results, therefore we opted to leave them out.
- line 493: not only this
CDPSTP8S1 tune contains DPS (see my general comment above)
The whole section has been overhauled.
- line 502 and caption of Fig 13: "DPS component in..." what do we mean exactly with that?
We did not define that before. Did we 'extract' the pure DPS part with this model
With KATIE this is probably straight forward, but how did we extract it from
PYTHIA? I recommend that we explain that in the text.
In section 4 we explain that Pythia8 and Katie can generate pure DPS samples. Explaining it agiain seems redundant. However, we did change "the DPS component" to "the pure DPS sample"
- line 513: "15" -> "Fig. 15"
Changed accordingly.
- line 523 and figure 14: the extracted values sigma_eff cover a very large range with
relative small uncertainties, casting somehow doubt that we are measuring a well
defined quantity.
The section has been overhauled.
- line 539: does this not show that indeed these 2 ->2 models have
MPIs included
already due to their tunes, as I mentioned in the general comments?
The section has been overhauled.
- fig 14: A few observations here:
The spread or unknown systematics seems very large. It shows that we cannot extract
a model independent measurement of sigma_eff with this technique as applied
here, from the 4 jet event data.
The section has been overhauled.
The former measurements in the 90's: can we really put these on the same plot? I assume
that the analyses at the time were much more naive and used different methods and tools
(certainly no MC programs of the same level as we have now)
We went through thepapers and yes, there are differences, expecially in the technicality of the programs, but we did not encounter anything that would clearly suggest that the results are not comparible.
Secondly and more provocatively (if I were a referee of this paper): what did we learn
in these last 30 years? You are not obliged to answer that (to me, but dono if the
referee will question it
)
In the case of the inclusive four-jet scenario, we have found that our models do still need to improved further before any more significant conclusions can be drawn. For example, the NLO 2->2 models form outliers compared to the other models (see updated results) or we need a matrix element mixing/matching in the off-shell samples, which look the most promising when looking at the correlations between the jets.
- line 555: The J/psi measurement is a very precise measurement with a systematic
error of ~1mb. In view of the high level discussion in this paper and the reported
results: Is the J/psi method really much more accurate/reliable?
We should comment on that.
Yes, the error is different, which is due to the machinery that they have available at the LHCb.
- Depending on which of my comments survive
this may affect a rewording of some
of the conclusions in the summary.
Conclusions have been revised accordingly.
- ref [51] this is submitted to the arXiv now as 2011.03422 . add the arXiv number.
Changed accordingly.