Question & Answers to TOP-16-004

Differential cross section measurements of single top quark production at 13 TeV

useful links

color code

  • Answers in blue are considered final.
  • Answers in orange require further communication between the questioner and the authors.
  • Answers in green mark additional studies that can be performed but do not affect the final outcome of the analysis
  • Answers in red still need work from our side.

Jeremy Andrea on ANv2 (26/01/16)

  • Table 2 : could you please specify which signal sample(s) is(are) used
    for data/MC comparisons ? What sample is used for W+jets ?
    • The default samples are now marked in ANv3. It is amc@nlo 4FS for t-channel, powheg 5FS for tW, powheg for ttbar, amc@nlo for wjets, amc@nlo for DY
  • Table 3 : to be removed ?
    • This is just a placeholder for 76X samples (removed in ANv3)
  • For trigger and muon selections, did you used SF provided by the POG
    already ?
    • these have been measured by TOP-16-003
  • Equ (2) : are the "c" correction factors to the Jet Energy Resolution
    already provided for 13 TeV by the JetMEt POG ?
  • Btagging SF : are the one derived from QCD (13 TeV) used ? Or the ones
    from ttbar events ?
    • mistagging of light jets is from QCD and efficiency of true b's is from ttbar
  • Table 6 : There are 3 single top samples, each one with a scale of 0.5.
    Does the "scale" account for difference in top and antitop
    cross-sections ?
    • no, this is just because there are two independent samples used so both need to be scaled by 50% to match the cross section. Each individual sample is already scaled to the cross section by default.
  • Line 336 and 338 : I'm wondering, it seems there is no correlations
    between top pT and eta(j') at all ?
    • no, there seems to be no correlation since j' is not used to reconstruct the top quark
  • Section 4.3 : one of the weak point of the approach is that, the QCD
    estimation is done after the training. Meaning that the optimization of
    the BDT training might not be optimal ? This doesn't affect the
    robustness of the BDT of course. Would an "rough" estimation of QCD,
    prior to the training, improve something ?
    • the antiisolated QCD MC is scaled down to 0.5% in order to prevent the BDT to train solely against QCD
  • Figures 4 to 7 : I suggest to rebin the QCD plots.
    • done
  • Section 5.1 : I'm not sure to understand how QCD is treated. Do I
    understand correctly that there is only one training to distinguish
    against all backgrounds ? There is not a QCD-dedicated training, right ?
    • yes, there is only one training which is mainly meant to separate t-channel from wjets and ttbar. QCD is just mixed-in to prevent residual events to peak at high BDT output values. QCD is rejected by cutting on mTW>50GeV.
  • Figure 19 : could there be a missing contribution in the tail of mTW ?
    Which W+jets sample you used ? MLM or aMC@NLO ? Is the MET distribution
    ok in that region ?
    • this may be due to the recoil model. There is a way to correct it but this will require more studies first. MET is fine except for the first bin that is due to a lack of QCD event from antiiso region. The effect is less dramatic in the other regions.
  • Figure 27 : I guess there is no MET-PHI correction applied ? Would it
    make any significant changes to the results ?
    • MET-phi is not yet applied since a recipe will only be available for 76X; update: miniaod in 76X seems to have MET-phi applied by default.
  • Line 531 : could you prepare a table, summarizing the estimated
    background yields and the related uncertainties ?
    • added in ANv3
  • Line 533 : what is the relative uncertainties ? How much ?
    • this was indented to be taken as 20%; now individual fits for each pT/y bin are performed

Pedro da Silva on ANv3 (28/01/16)


  • L49-50 Are MET filters strictly necessary for this analysis? To which degree it’s important to measure its efficiency?
    • filters are not strictly necessary since this analysis is not sensitive to high MET tails (>300 GeV). See here for efficiency
  • L129 I guess the muon contacts will ask you as well : why not complemeting with TkMu20 ? It does increase the efficiency.
    • measured trigger efficiencies for new muon iso working point for TOP-16-003 only includes IsoMu20; analysis is now restricted to only this.
  • L141 It would be good to motivate why is a tighter cut used wrt to the one proposed by the POG, following maybe Georgios’ presentation in the PAG meeting
    • done
  • L214-216 I guess at this point you’re still working on including trigger/id+iso/b-tagging etc. scale factors. When ready, don’t forget to document it smile
    • done
  • L239 is there any limit on how much you are allowed to modify MET to attain this condition? Do you have some control plot of how much you have to modify it for the signal and the backgrounds?
    • there is no constraint but from possible solutions which eliminate the sqrt term, the one with the smallest delta R to the original pmiss vector is chosen
  • L274 It would be good to comment on how you have arrived to the choice of these scale factors
    • clarified, it is just the yields from MC. Each sample is normalized to the process cross section. So if combining multiple samples, these need to be scaled down.
  • L330 what is the underlying sorting algorithm ? increasing value i guess so, from Eq.1
    • yes
  • L321-338 it’s an interesting approach. Could you add a couple of sentences on the gain of using this algorithm, with respect to computing a simple correlation factor?
    • it is to detect even non-linear correlations as stated in l323
  • L365 Was the q/g discriminator tried for the forward jet?
    • not available in 74X, in 76X it shows some deviation wrt data. more studies are needed so it will not be used in the preliminary analysis
  • Figure 15 - the non-isolated region still contains a lot of W+jets. That’s because it’s a simple inversion of the cut. Was it studied to use an even looser isolation inversion .e.g I_rel>20% to remove further the W+jets contamination?
    • this is taken as a systematic variation
  • Eq.19 yeah ok, it’s a possibility. You could have extended the likelihood (Eq.13) to accomodate the categories to which you fit the MT(W). e.g. L_{2j1t} = L_{2j1t,MT(W)<50) x L_{2j1t,MT(W)>50} where the first one fits the MT(W) distribution and the second one the BDT
    • this is unfortunately not possible with theta; therefore the trick with the artificial fit variable is used; technically it is the same as using an extended likelihood
  • L436 Does theta accomodate for the usage of uniform priors? Not that any significant change is expected, except maybe in the final uncertainty.
    • yes
  • Eq. 18 how do you set the initial QCD normalization? To the MC prediction or to the #events in the isolation sideband?
    • to the MC prediction; this may be suboptimal to start the fitting algorithm but the fit converges nonetheless
  • Fig. 17/18/19/20 it would be nice to see the DD shape directly on the left plot, as we know QCD MC is insufficient to describe correctly the pre-fit data
    • the unfitted DD QCD yield will completely overshoot the data because it is the yield expected in the antiiso region; it can be changed but wont look very informative
  • L498 even if traditional could you add their definitions. In particular I’m not acquainted with “event shape C”.
    • done
  • L521 which parton is taken as reference (status etc.)
    • updated to isLastCopy
  • In general for introduciton of Section 8 - do you have plans to repeat this at particle level?
    • only for a paper
  • L523 can you comment on the strategy followed to define the binning? Did you follow some quantile-based requirement or purity/stability requirements (e.g. Eq. 10/11)? In particular i would be curious to see a plot of purity/stability for these two variables and the binning chosen.
    • plots added
  • L589/590 This was for Madgraph Born level samples. For NLO samples, besides the QCD scale choice variations we consider variations of the Parton Shower scale.
    • added
  • Table 7 Although the post-fit nuisances are very close to the initial value their final uncertainy is very much reduced. Is this expected.
    • yes, the priors are only there to help the fit converge
  • Table 8 Taking out stat+lumi uncertainties it is nice to see, with the available systematics, that there is an overall substantial decrease the uncertainty wrt to the EA PAS-TOP-15-004. This should be highlighted in the text when the assessment of systematics is completed. The largest decrease is for JES. I guess with more data is partially constrained in the fit.
    • result will not be quoted anymore since TOP-16-003 is already out
  • L611 The pseudo-data was generated with Poisson fluctuations, or should one read Fig.54 as a closure test plot? It would be important to expand the discussion on how the systematic uncertainties are computed to yield the error bars shown in Fig. 54. Maybe statistics is not enough. Could be interesting to show also separately for top and anti-top as a way to further explore the sensitivity to PDFs. Maybe a differential ratio?
    • I plan it for the paper
  • Do you intend to do absolute cross section as well?
    • this analysis is not optimized to get a precise inclusive xsec measurement; also depends on TOP-16-003 => to be discussed
  • If you have some time it could be good to add also some assessment of the statistical coverage of the unfolding procedure. E.g. throwing pseudo-experiments and checking that the pulls in each bin have a gaussian distribution with with 1. After reading the PAS (L202-209) I understand that you did all of this, altough you didn’t yet document in the AN smile
    • done
  • L616-617 - let’s see when unblinded wink
    • ok

Suggestions, typos, etc.

  • L68 ascence -> absence
    • ok
  • L78 You can of course disclose the method at this point or leave it more vague as it’s an introduction “when inverting the muon isolation” -> “in a control region dominated by this background process”
    • ok
  • L81 maybe reference for BDT?
    • cite tmva
  • L87 maybe reference for TUnfold?
    • cite tunfold
  • L120 reconstructed muons or electrons -> muon or electron candidates
    • ok
  • L144 antiisolated -> anti-isolated
    • ok
  • L174 eta->\eta
    • ok
  • L181 with the -> of the muon flight direction
    • ok
  • L183 hardonizing -> hadronizing
    • ok
  • L202 does it really improve the resolution or is it mostly for consistency of the global energy scale changes and therefore of the recoil scale?
    • it is understandable that the resolution improves since MET is corrected by using some "better calibrated tracks" in the calculation that are inside the jets
  • Eq. 5/6/7/8 leave space before to recover line numbering
    • ok
  • L227 build ->built
    • ok
  • L236 neglection -> negligence
    • ok
  • L256 minic -> mimic
    • ok
  • L381 discriminat -> discriminant
    • ok
  • Table 7 nuciance -> nuisance
    • ok

Pedro da Silva on PASv0 (28/01/16)


  • L42/43 Is it the CPU consumption that makes it hard or the correct description of all the detector effects and pileup together with missing higher orders in the MC to describe how QCD multijets leak into the single top region ?
    • at first order it is CPU and memory, with the available MC for QCD, no modeling issue was observed (e.g. like the polarization angle in Wjets) that justifies the need for NLO MC
  • L105-109 Maybe this could be simplified - you’ll discuss the reconstruction of the top quark kinematics (essential for the quantities to be mesured) and the usage of a multivariate discriminator to reject further the background with respect to a simple cut-based analysis (this should be stressed, otherwise there is no point in using MVAs)
    • whole paragraph rephrased
  • L114/120 please comment if these rates for 2 solutions, complex solutions are similar for both signal and backgrounds and what is the rate observed in data. See also comment to the AN regarding the modification of MET: is there any restriction in how much you modify?
    • this is for signal; background can still be studied
  • L155-157 - I understand the optimization is still on-going. Please motivate the choice of this cut for the analysis.
    • ok
  • Section 4.1 at the end you should state how the top pT is defined (pT of the b+lepton+neutrino system) and how the rapidity is defined. In particular for the latter: is it rapidity or pseudo-rapidity. In the first case, which top mass do you use to compute it? 172.5 GeV or the reconstructed mass?
    • it is the rapidity of the reconstructed top quark y=0.5*ln[(E+pz)/(E-pz)]
  • Eq. 3 - suggest to remove. This is a technicality. See comments on the AN regarding combining likelihoods. Suggest also to rephrase until L161 to something like: “We perform a combined fit to different distributions depending on which category an event belongs to. For events with low MT(W) (<50 GeV), we make use of the MT(W) distribution, while for high MT(W) (>=50 GeV) we make use of the BDT shape.”
    • ok
  • L172 when 76x is available please comment on the agreement found
    • ok
  • I miss in Section 4 the following discussion: normalization factor to be used for Section 5 : i.e. report the xsec measured from the ML-fit and comment on its uncertainties and compatibility with the early analysis result (TOP-15-004)
    • ok
  • Table 1
    I’ll not be picky on these numbers because they are based on 74x. use an appropriate number of significant digits explain in the caption what is the nature of the uncertainties remove the nuisance parameters. The uncertainties are very small because the priors have overblown widths. You can comment in the text that W and Z backgrounds are pulled by a factor of 1.6 while the remaining backgrounds are compatible with the initial prediction. t-channel is measured to be 20% higher than the SM prediction, although compatible with uncertainty.
    • ok
  • L173 Normalized differential cross section measurement
    • better unfolding? because "Normalized differential cross section measurement" is the title
  • I miss in Section 5 the following discussions
    • how is parton level defined (maybe around L177)
      • added: after ISR/FSR (lastCopy)
    • how is the binning chosen for the migration matrices/unfolded spectra (L199/L201)
      • ok, less-curved expectation & purity/stability>50% for all bins
    • From the physics side: will you consider particle level defintion? will you
      • planned for paper
  • Figure 4 - Maybe the current figure is just temporary while the analysis is in 74x. But, I would prefer to see the figures after the BDT cut so that one sees the distributions which are going to be unfolded.
    • ok
  • Section 6 - are there more theory curves which can be added for comparison?
    • there is amc@nlo 4fs, 5fs, powheg & herwig
  • Do you have enough stats to report separately for top/anti-top, or eventually a ratio of the two?
    • not yet but planned for the paper
  • L222 based on what? i.e. quote in which generators the theory predictions tested are based on.

Suggestions, typos, etc.

  • Abstract: suggest to refere explicitely that these are normalized differential cross section measurements (unless you’ll update to include also absolute differential cross sections)
    • Incipit is now: "Normalized differential cross sections of single top quark production in $t$ channel in proton-proton collisions are measured as functions of the transverse momentum and the absolute value of the rapidity of the top quark."
  • top quarks decaying to muons in the final state -> top quarks decaying to final states containing one muon
    • ok
  • boosted decision tree -> multivariate discriminator (avoid jargon)
    • ok
  • L3 Tevatron at Fermilab -> Fermilab Tevatron
    • ok
  • L10/11 spell out 4/5
    • This is just one example of a difference between different signal models in MC.
  • L13 The measurement presented in this note is data of proton-proton collisions -> proton-proton collisions data recorded at -> recorded by
    • ok
  • L18 suggest to include a short description of how the PAS is organized
    • ok
  • L21 Avoid negative statements regarding the CMS detector operation
    • Rephrased as "Restricting the analysis to periods of time when all the CMS sub-detectors were fully operational and the solenoid was operated at 3.8 Tesla, this dataset corresponds to an integrated luminosity of $2.2$\fbinv."
  • L23 Samples of simulated -> Simulated samples of
    • ok
  • L25 Monte Carlo -> MC
    • ok
  • L28/L32 and similar did you use \POWHEG ?
    • Now yes.
  • L30 maybe quote the underlying event tune used for completion
    • ok
  • L50 that is however mostly not within the detector acceptance -> that often fails the selection criteria applied
    • this is not correct; a 2nd b-tagged jet is not required by the selection because it is not inside the detector acceptance in the first place
  • before Eq.1 “using the “Delta Beta” definition” -> as (for non CMS people this is jargon and also strage that it’s called DeltaBeta but there is no \Delta in Eq.1)
    • removed delta beta
  • Eq.1/Eq.2 leave space before equations to recover line numbers
    • ok
  • L61 deposited energy from displaced tracks that are associated to pileup -> of the \pt flux from reconstructed tracks which are associated to pileup vertices
    • ok
  • L99 signal and control regions are defined in which different processes dominate. -> we define signal and control regions, in which different processes dominate.
    • ok
  • L123 move “under the signal hypothesis” to the start of the sentence
    • reshuffled
  • L131 Background discrimination with a multivariate analysis
    • ok
  • L140 add TMVA citation or another appropriate source for BDTs and Boosting algorithms
  • L164 tigher -> tighter
    • ok
  • L171 subjected simultaneously to the fit -> are included in the fit
    • ok
  • Figure 3
    • full transverse -> transverse
      • ok
    • left figure MTW -> M_T(W)
      • ok
    • right figure BDT discriminant-\eta binned -> BDT
      • ok
  • Figures 2,3,4
    • all have 2.1/fb but in the text you state 2.2/fb (L22)
      • changed coherently to 2.3/fb (Moriond ref.)
    • remove (Madgraph) and MC stat. from the caption, for the latter explain in the caption how it is represented in the figure
      • ok; MC stat. -> Total syst.
    • remove or explain in the caption what is chi^2/KS
      • removed
    • explain in the caption what is represented in the bottom panels and what the shaded bands represent
      • ok
  • L174/175 - suggest to remove. I hope the reader knows when he reaches this point.
    • ok
  • L181 get -> recover
    • infer?
  • L191-196 Do we need to describe the unfolding algorithm given it’s already referenced? The only relevant part is the regularization and the strategy adopted for it. I suggest to remove these lines or simplify further.
    • reduced text
  • Figure 5: luminosity issue
    • changed
  • L214 four flavour scheme -> 4FS
    • ok
  • References [20] remove Technical Report
    • ok

Nadjieh Jafari on ANv2 (02/02/16)

  • L385: I did not understand what was the motivation to check/use the
    discretized |eta'| variable in the first place. Could you please explain a
    • with 76X + the new JECs, there can be still some residual mismodeling present especially at the JEC eta bin edges. So by using a discretized j' eta, I hope that the BDT will be less sensitive to these edge effects
  • In post-fit plots, do I understand correctly that |eta| > 3 is opened
    up again and the same SFs from the fit in |eta|< 3 are applied there?
    • yes; the problem here is that when fitting |eta| completely, even events in the barrel display an overall MC/data mismatch because the fit tries to recover the yield in the forward region. So by the approach taken here one can at least validate the barrel region
  • Fig. 18 post-fit fit variable 3j1t: Do we have an idea of what causes
    the BDT trend in data/MC? Is it with fwd region included? If yes, is it
    explicitly checked in |eta'| < 3? Otherwise, could it be due to ttbar?
    • it is the forward region; same problem in 2j0t
  • Could it be covered by ttbar systematic samples or ttbar with pythia6?
    • this mismodeling wrt powheg showed only up for events with njets>4 but the resulting differences between both samples on the final result can be evaluated
  • Fig. 34: I would say that we see a similar effect as 2j0t in 3j1t. In
    fact, data points are just at the edge of MC stat. uncertainties in 3j1t
    • yes; in 2j0t; both jets can be |eta|>3 that is why the mismodeling is even more enhanced here
  • On systematics, do I understand correctly that R matrix is changing
    per source (with tau and rho fixed to their nominal values)?
    • yes - the response matrix changes when a systematic influences the signal prediction; regularization (tau) is not fixed and rederived per systematic on the fly - however, its value changes only very minor
  • Not sure I understand what is the ML-fit uncertainty, (L553).
    If I understood correctly, the signal and background samples are scaled
    to the fit results before unfolding. Hence the background yields are
    obtained from data. Could you please elaborate on this a bit?
    • this is just the uncertainty on the yield as estimated by the fit. It may become more clear when added to the systematic table as (Wjets yield, ttbar yield, etc.)
  • L556: The additional uncertainty on relative contributions of
    backgrounds: Are they varied within the theoretical uncertainties of the
    corresponding cross sections? Are these variations taken as correlated?
    • these are not taken as correlated; atm, a conservative variation of 20% can be used or more aggressively the theoretical uncertainty.
    • update: this is not important anymore since individual fits per top pt/y bin are performed and the t-channel fit result + uncertainty are passed to the unfolding directly.

Preapproval comments (03/02/16)

  • What is the chi^2 to expect for uncorrelated?
    • a p-value should be computed for the compatibility with non-correlated hypothesis
      • done
    • can this test be performed for the BDT as well?
      • done
  • aren't correlations for backgrounds also important? background subtraction would be affected
    • add correlation matrix for background
      • done
    • the 2D linear correlations should be quoted as well for signal and background (typical TMVA product)
      • done
  • why doing all in a single fit?
    • sample is divided in two orthogonal categories, so that data is not fit data is not fitted twice. Important to account for systematics and correlations in an easy/proper way.
  • in the text, change the wording to say it is a combined fit (for clarity), no need to use definition of combined variable.
    • done
  • different binning on the right figure for the fit variable? data looks different! looks like the binning is different
    • updated with consistent binning
  • clarify the definition of parton level
    • use 'isLastCopy'
  • It would be important to quote the purity/stability for the binning chosen
    • done

ARC on PASv1 & ANv3 (12/02/16)

  • Up to the unfolding, the analysis is done in one bin. The unfolding is done in several bins. How are the statistical uncertainties in each bin on signal and backgrounds determined?
    • bins in the reconstructed data distribution have events N(reco) and uncertainty sigma(reco) = sqrt(N(reco))
    • the remaining background B(reco) is subtracted from the number of data events, N(bkg. sub.)=N(reco)-B(reco), whereas the absolute uncertainty stays the same under this transformation, sigma(bkg. sub.)=sigma(reco); an additional systematic uncertainty on the number B is assumed by varying each background component within its fit result
    • the resulting data distribution is unfolded; this applies the transformation N(unfolded)=R^{-1}N(bkg. sub.) and sigma(unfolded)=R^{-1}sigma(reco) with the inverted response matrix R
    • plots from these 3 steps are shown in the slides here
    • NOTE: when switching to the signal/background fits per bin, the number and uncertainty on the signal after background subtraction can be read of the fit result directly per bin
  • Could you clarify whether the cut at 0.3 in the BDT fit is performed pre- or post-fit? Could you show the top pt and y distributions as well as other control plots in the two regions (BDT <0.3 ; BDT >0.3) post-fit?
    • the fit uses the whole BDT range, a cut is applied to get the data shape in a signal-enhanced region for unfolding; plots of top pt/y have been added to the PAS
  • What happens if you cut on the BDT to enrich the signal, is the sensitivity to bg systematics reduced ?
    • background contamination and also their systematics do vary when scanning the BDT WP; need to be evaluated;
    • update: no cut necessary anymore since fits per unfolding bin are performed
  • How can we be sure that the features (efficiencies/resolution) of the kinematic reconstruction in data are well described by the MC? Can we look at efficiencies as fct of lepton and/or b-jet variables ?
    • need to be studied
  • what is the resolution in pt(top) and eta ? Would like to see purity and stability (request from pre-approval)
    • added to AN
  • Is it clear that the signal selection efficiency is controlled in all bins ? Would like to see BDT output distributions separately for each bin.
    • done
  • is it clear that there is no kinematic bias due to the choice of MVA input variables ?
    Would like to see scatter plots between pt and eta (resp.) vs input variables (in addition to AN figure 11).
    • done
  • Do you understand the trend in AN figures 12 and 13 ?
    • QCD is overestimated by 500% when taken from MC
  • What do we learn from AN figure 16 ?
    • the correlations between ttbar/wjets reduce (2j1t: rho=-91%) when including the 3j1t/3j2t control regions; notice also the similar BDT shape of the two => BDT alone cannot disentangle their yields
  • The background priors are 100% Gaussians (30% for ttbar). Could we gain by tightening them to e.g. 50%(10%) ?
    • switched to log-normal now and assuming unconstrained for t-channel, 10% for ttbar; 30% for WZjets (for AMC@NLO)
  • Should we include PS modeling uncertainty (pythia vs herwig) ?
    • added
  • AN figure 54: it looks like the fluctuations of the data points are too small for the size of the statistical uncertainties.
    Could we get results from a couple more pseudo-experiments ?
    • binning has be changed to decrease stat. uncertainty; bias and pull tests added to AN
  • How well does the MC describe the data when varying the jet energy scales within uncertainties?
    • to be studied

Towards paper

  • do you ultimately foresee to include the electron channel ?
    • only when a gain in precision is expected in the combination of both channels. Also note that the "simple" data-driven method to estimate the multijet event shape and yield by using antiisolated muons was not working very well in 8TeV analyses. An extended/different procedure needs to be developed and studied as well.
  • should we also do a measurement in fiducial region and at particle level (pt_mu, b etc. ) ?
    • for the paper, I plan to focus directly on fiducial measurements since certain observables (e.g. polarization angle) are not well defined anymore in ST t-channel 13TeV samples at hand


  • suggest to include in PAS bottom right figure in pre-app slides p31.
    • this is only a technical detail; in PAS is now 2j1t MTW (w/o cuts) & BDT (MTW>50). Also the PAS text only mentions a combined fit now which is achieved by the artificial variable
  • Compare data also with powheg/aMC@NLO and 4FS/5FS in final figures.
    • done

ARC/author meeting (19/02/16)


  • The event yields are increased substantially, and it is unclear whether this can be attributed to the change of the eta cut. It is unclear why the W+jets sample seems to have substantially more statistics. Should try out aMC@NLO sample instead of (currently used) Madgraph.
    • one likely explanation for the increase yields are the change of the tight b-tagging working point from 0.97 to 9.35.
    • amc@nlo high-stat. sample (200M) is the default now
  • PU reweighting (AN figure 21): Use of 69mb does not seem optimal.
    • was investigated; 69mb is optimal for 2j1t but not for 2j0t control region; since 2j0t is not used; the recommended 69mb is kept
  • mTW distribution (AN Figures 12,13,14 and PAS Figure 3): should wait for appropriate JEC before final assessment.
    • ok
  • systematic uncertainties (slides p7)
    • Do the t-channel scale uncertainties include PS-scale ?
      • switched to new 7 variation ME/PS scheme as described here
    • Could it be that this includes uncertainty on cross section, not just acceptance ?
      • excluded uncertainty on xsec as recommended
    • Also take scale unceratinties on W+jets into account ? A: Yes when moving to aMC@NLO. Will check if Madgraph has weights.
      • done
  • distributions after cut MtW >50 && BDT>0.25 (slides p8):
    • distribution as fct of pt is not very well described. This seems a serious issue for the unfolding.
      • more confident now after performing individual fits per top pT bin
    • bins as fct of eta show different systematics, esp. 2nd bin has large uncertainies (under investigation)
      • mitigated after analysis updates
    • first few bins as function of pt are consistent with zero. Rebin ?
      • binning scheme adapted

Timeline / next steps

(the following is obsolete since Moriond was failed)

  • For Moriond EWK green light is required in about 10 days.
  • One of the next analysis steps is to investigate separate fits to each of the analysis bins.
  • Target next AN and PAS versions incorporating as many comments and improvements as possible at the timescale end of next week (Feb 26).
  • In the mean time, start providing tentative answers to our previous comments on twiki

Andreas Meyer on PASv4 and ANv6 (29/03/16)


  • understand size of uncertainties in first bin in pt(top). It may be best to check the pre-fit distributions in this bin for different systematic variations.
    • the prefit distributions for the first pT bin can be found here: merged.pdf
  • ME-PS treatment of scale uncertainties. The uncertainty is HUGE. how big is the uncertainty for ME scale variation only? how big is the uncertainty due to PS scale variation?
    • a problem in the normalization (reweighting acts differently on positive and negative events) has been identified and now fixed.
  • how do pre-unfolding (PAS fig 4) and post-unfolding (PAS fig 5) results match ?
    • with the Q-scale normalization problem: the large uncertainty pulled the total mean towards the SM. Note: pseudo experiments are diced for each uncertainty and then the 16%,50%,84% quantiles are taken as the total uncertainty and mean per bin
    • without the Q-scale normalization problem (now fixed): same deviation is observed (data predicts harder pT spectrum) before and after unfolding
  • are you sure you really only changed the acceptance effect ? If you determine the uncertainties using pseudo-experiments for different varied samples, keeping the analysis (response matrix) fixed, then you wrongly pick up BOTH acceptance effects AND physics modelling. One correct way to determine the acceptance effect only is to re-run the analysis on data for different MC assumptions, i.e. (change the response matrix), but nothing else.
    • whenever a systematic influences the signal, the response matrix is varied as well. This is the case for generator modeling, hadronization modeling, b-tagging, mistagging, JES, JER, muon SF, PDF, PU, Q-scale t-channel, top mass.
  • what is the source of the spike in the fit variable e.g. for 3j2t sample around 60 (AN fig 113) ?
    • this comes from the overlap of MTW <-> BDT due to the binning. Note: the fit variable is only by coincidence sometimes nearly continuous at 50

PAS editorial

  • give typical size of systematic uncertainties
    • list major systematics
  • do we really need fig 1 ?
    • I think it is important to understand the difference between 4FS and 5FS
  • do we need table 1?
    • removed
  • suggest to state consistency with TOP-16-003 but dont quote result for inclusive cross section.
    • done

ARC on PASv5 & ANv7 (04/04/16)

We are still worried about the first bin in pt(top), and we assume that you are, too!.

In the first bin some of the "partial uncertainties" are significantly bigger than the total uncertainty. This seems to indicate instability.

  • could it be that a fluctuation of the W+jets background due to statistics pushes the result down (see e.g. fig 123a and b)?
    • the first bin is just very sensitive to the systematics because 1. less events are observed than predicted and 2. the selection efficiency is rather low which yields an amplification of the uncertainty by the unfolding
    • to test against statistical fluctuations the fit was performed with (left) newton-minimizer (default) fitted to BDT (default), (middle) minuit2 fitted to BDT (default), and (right) newton-minimizer (default) fitted to eta j'. The results are stastistically compatible whereas the eta j' fit result produces significantly larger uncertainties due to the less discrimination power of that variable.
    • unfolded_top_pt.pngunfolded_top_pt_minuit.pngunfolded_top_pt_jeta.png
  • would it make sense to use a coarser binning (e.g. to reduce statistical fluctuations) ?
    • a coarser binning will reduce sensitivity a lot since one needs a relatively fine binning in the signal dominated region that contains by construction only a few background events
  • would it make sense to use pseudo-data for the determination of the uncertainties (to remove the statistical factor from the systematic uncertainties)?
    • if there is a fluctuation in the systematics templates wrt to the nominal templates (=pseudo data), the sensitivity to the limited MC statistics will be still present. This is not true for the few systematics templates that are derived by reweighting only (e.g. PU, b-tagging, muon SF, PDF, ...)
  • it seems that pre and post unfolding results are not really consistent. data/mc are ~0.7 pre-unfolding and ~0.5 post-unfolding.
    • this can occur due to some asymmetric uncertainties (e.g. Q-scale, generator, hadronization model) which also shift the average expectation when combining all systematic shifts for the total uncertainty. The systematic uncertainties are combined by dicing the yield per bin according to the unfolded templates and summing the shifts from all systematics up. The yield for a pseudoexperiment is obtained as yield(PE) = Gaus(mean=0,sigma=+-JES)+Gaus(mean=0,sigma=+-JER)+...+yield(nominal). The total uncertainties and expectations are taken as the +-1 sigma and 50% quantiles over many PEs.
    • in pseudo code this can be written as (given up,down,nominal as shown in the systematic templates below per unfolding bin
    • toys = array(N)

      for i in N:

      for up, down in systematics:



      totalDown, totalMean, totalUp = quantiles(toys,15.8%,50%,84.2%)

  • We propose that you show the distributions of acceptance (N_gen,cuts/N_gen) and efficiency (N_rec,cuts/N_gen,cuts) in bins of pt and y. Their products are a projection of the response matrix. We expect to see that the acceptance in the first bin is very small. Too small ? (In TOP-14-004 the first bin had an acceptance that was a factor ~3 smaller than in other bins. After discussion, we decided to indicate this in the PAS (figure 3))
      • effPt.pngeffY.png
        • the acceptance is indeed 3 times lower in the 1st pT bin after mu+2j1t selection compared to the 2nd & 3rd bin. Not only the acceptance increases the uncertainty by also just the fact that less events are observed there

ARC (07/04/16)

  • We would like to ask you to document exactly how the uncertainties for the differential cross sections are combined (and how the final central value is constructed from the results using the nominal sample and systematic variations. Specifically we should try to understand table 11 on page 75 in AN v7.
    • see above
  • In bin 1, the various uncertainties tend to be larger than the total uncertainty? In bin 3, in contrast, the total uncertainty seems significantly larger than each of the partial uncertainties.
    • this is just an artifact since the rel. uncertainties for the individual result are referring to a different center value than the total uncertainty. In a new version, the rel. uncertainties will be replaced by the absolute ones which is less confusing
  • It would be instructive to see the plain result for some of the variations (or at the least for the nominal sample).
    • the following shows a few unfolded systematic templates. Most notable is that the first bin seems to be much more affected than the others.
  • Btag.pngEn.pngTopMass.pngQScaleWjets.png
  • Also, we would like to ask you to show us the acceptance in each bin (similar as for PAS TOP-14-004).
    • done, see above
  • The other question that came up is whether there is an explanation why aMC@NLO(5FS) has a significantly different shape than the other predictions. Do we have Powheg(5FS) for 13 TeV?
    • there is no powheg 5FS at least in MCM

ARC/author meeting (08/04/16)

approval conditions

  • improve the documentation about combination of systematic (and statistical) uncertainties and about the recalculation of central value w.r.t the result from the nominal MC sample. Please add the description to the AN and give a sufficient explanation also in the PAS.
    • added to PAS: "Pseudo experiments are preformed per unfolding bin by dicing Gaussian distributions per uncertainty source around the unfolded nominal template using its difference with respect to the unfolded systematic varied templates as standard deviation. The differential cross section (uncertainty) is taken as the 50\% (one standard deviation) quantile over many pseudo experiments."
  • show (in AN or during freeze on twiki) generator-level and reco-level distributions of t-channel single top for varied top mass (and possibly also for other dominant systematic uncertainties).
    • the following shows gen & reco level distributions of theoretical uncertainties of t-channel MC scaled to prediction.


gen_top_pt_QScaleTChannel.png reco_top_pt_QScaleTChannel.png
gen_top_pt_TopMass.png reco_top_pt_TopMass.png
    • the following shows reco level distributions of experimental uncertainties of t-channel MC scaled to prediction
    • reco_top_pt_En.pngreco_top_pt_Btag.pngreco_top_pt_PU.png
  • include information about the acceptance (esp. the fact that it is low in the first bin of pt(top)) in the PAS.
    • The overall acceptance of signal events per unfolding bin after the 2j1t event selection ranges from 1.5% to 2.5% with the exception of the first top quark pt (1.1%) and the last |y| (0.7%) bins.
  • We also suggest to specify in text of PAS (section 6) which uncertainties are dominant
    • added to PAS: "The sources of systematic uncertainties which have the largest impact on the differential cross section measurements are the renormalization and factorization scale choice of the signal process, the signal modeling, and the top quark mass."

Andreas Meyer on PASv6 (18/04/16)


  • line 47: wondering, would one gain in precision by adding the ratio as constraint ?
    • the t/tbar ratio is already constrained (taken from the MC) since no separate fits for t/tbar are performed
  • section 4.1.: once this analysis becomes a precision analysis, we will need to study the dependence of the kinematic reconstruction on the phase space region.
    • ok
  • line 157: why is the BDT trained w/o cut on mt(W) ?
    • to increase training statistics (mostly for QCD)
  • line 224: did you ever try to run the analysis with 4 instead of 8 reconstructed bins? I could imagine that that would gain stability of the fit and reduce fluctuations in the response matrix ...
    • this was not tried since it is recommended to have twice as many bins at reconstruction level compared to the generator level to stabilize the minimization in the TUnfold.
  • line 274: are you quoting the full 3 GeV variation as uncertainty ? In that case I am less worried about the size of the mass uncertainty. I think you could/should assume linearity and scale it down by a factor 3.
    • changed after freezing for approval. see modified here: 2016_04_24.pdf


  • Title suggest: "Measurement of the differential cross section for t-channel single-top-quark production at \sqrt{s} = 13 TeV "
    • ok
  • Abstract suggest:

A measurement is presented of differential cross sections for t-channel single-top-quark production in pp collisions at a center-of-mass energy of 13 TeV. The cross sections are measured as functions of the transverse momentum and the absolute value of the rapidity of the top quark.
The data were recorded in the year 2015 at the CMS Experiment and correspond to an integrated luminosity of 2.3 fbâ^'1.
A maximum-likelihood fit to a multivariate discriminator is used to infer the signal and background fractions from the data. Unfolding to parton level is performed. The measured cross sections are compared with theoretical predictions to next-to-leading order with matched pardon-showering as
implemented in Monte Carlo generators. Good agreement is found within currently large experimental uncertainties.

ok, small adaption:

A measurement is presented of differential cross sections for t-channel single-top-quark production in pp collisions at a center-of-mass energy of 13 TeV. The cross sections are measured as functions of the transverse momentum and the absolute value of the rapidity of the top quark.
The analyzed data were recorded in the year 2015 at the CMS Experiment and correspond to an integrated luminosity of 2.3 fbâ^'1.
A maximum-likelihood fit to a multivariate discriminator is used to infer the signal and background fractions from the data. Unfolding to parton level is performed. The measured cross sections are compared with theoretical predictions to next-to-leading order with matched pardon-showering as
implemented in Monte Carlo generators. Good agreement is found within currently large experimental uncertainties.

  • line 7: "accuracy" I'd say the accuracy is NLO+PS. How about "validity" ?
    • changed to validity but both may work since the question here is also if NLO+PS is accurate enough to describe the data
  • lines 10-12 and figure 1: Fine for me if you keep the figure, but can you make a statement why 4FS and 5FS are expected to be different for pt-top and y-top ? It seems that figure 5 supports the statement that 4FS and 5FS could have differences in the predictions. Do we understand that ? Maybe extend sentence, line 12: "this could contribute to differences in the predicted shapes of pt-top and y-top" or similar ?
  • line 101-103: suggest to shorted into one sentence and move to after line 90: "To avoid overlap of the jets with the selected muon candidate, jets are rejected if DeltaR (mu-jet) < 0.3."
    • changed to: Jets are rejected if DeltaR (mu-jet) < 0.3 to avoid overlap of the jets with the selected muon candidate.
  • line 106: How about: "... mistag rate of about 0.1%. This results in a tagging efficiency of about 50%". By quoting 49% we claim that we know this better than 1%.
    • ok
  • line 117: clumsy. Please improve for clarity and correct english.
    • changed to: Signal and control regions are defined using the number of selected jets and b-tags per event. The signal region is characterized by two jets and one b-tag (``2j1t''). A \wjets control region is defined for events with two non-tagged jets (``2j0t'') and a \ttbar control region is defined for events with three jets and one or two b-tags (``3j1t'', ``3j2t'').
  • line 171: should we add plots of the other input variables in the PAS or on the twiki ? Maybe most straightforward to have them in the PAS ?
    • the remaining input variables DeltaR (l-jet,b-jet) and Delta |eta|(b-jet, muon) are somewhat artificial, hence their distribution may not be very informative (apart from being discriminating signal/background). This is not the case for the shown variables, the forward jet |eta| and the reconstructed top quark mass, which show characteristic features of t-channel ST production. If requested, it can be added as supplementary material.
  • figure 3: Is the systematic uncertainty band post-fit ? I assume so. Should we write this somewhere (in the caption?).
    • added to plot captions: "The striped band denotes the total systematic uncertainty scaled to the fit results"
  • line 177: Now this is important for the reader to understand: Suggest "A binned maximum-likelihood fit (spell out here once more) is performed to the combined mt(W) and discriminator distributions, separately in each bin of the measurement. The fits yield the fractions of signal and background components ..."
    • slightly adapted: "Binned maximum-likelihood fits are performed to the combined \mtw and BDT discriminant distributions, separately in each bin of the measurement. The fits yield the fractions of signal and background contributions in data inclusively and differentially in intervals of the reconstructed top quark \pt and rapidity".
  • line 183: Suggest to remove "nuisance parameter" and just write "number of parameters in the fit".
    • ok
  • line 184: replace "the "top background"" by "tt/tW background" - no quotes .
    • ok
  • line 186: why log-normal and not Gaussian (or any other shape) ?
    • the priors are truncated to >0. Gaussian priors are not recommended by StatCom since their mean shifts through the truncation. Log-normal are by definition >0.
  • line 196: the DIFFERENTIAL cross
    • ok
  • line 199 and following: suggest something like: " these distortions are due to effects from selection efficiencies, detector resolution and kinematic reconstruction and are dependent on the kinematic phase space"
    • changed to: "These distortions are due to the selection efficiencies, detector resolution, and kinematic reconstruction which vary with the event kinematics"
  • line 214: suggest: "For the pt DISTRIBUTION the data display a somewhat harder spectrum than the expectation"
    • ok
  • line 215-217: Clumsy: Ok to repeat line 177 here (see above).
    • adapted to: "For the unfolding, the signal shape is estimated by performing separated ML fits to the combined \mtw and BDT discriminant distributions, in each bin of the measurement."
  • line 223: My understanding: "not biased" and "regularisation" is a contradiction in terms. Rephrase "For minimal bias" or "for acceptably small bias"
    • ok
  • line 232-233: do we need give these numbers. Suggest to remove this paragraph.
    • ok
  • line 237: write here something like : "For each systematic variation the whole analysis is repeated and fit templates and response matrix are replaced correspondingly."
    • changed to: "For each systematic variation the whole analysis is repeated with replaced fit templates and response matrices"
  • Rephrase "special care is taken ..." and remove "can change" by "change".
    • obsolete
  • lines 242-245: More elaborate, but still very hard to understand. Better put a sentence upfront that describes what you want to achieve.
    • ok
  • line 295: the list of dominant systematics begs the question why they are dominant. Esp. for the mass this is funny. Can we reduce the mass uncertainty by a factor 3 (see comment to line 274 above).
    • ok
  • line 300: Can we write here, or in the introduction, what's different between 4FS and 5FS ? (see lines 10-12 above).
    • extended introduction
  • line 310: add something like "within the experimental uncertainties which are currently still large".
    • ok


  • see suggested title for hyphens in "single-top-quark production" etc.
    • ok
  • suggest: 4FS and 5FS (and not "4 FS" or "5 FS")
    • ok
  • line 26: clumsy. Rephrase.
    • ok
  • line 27: Restricting the ANALYSED dataset
    • ok
  • line 38: ... contribution from top-quark-pair production
    • ok
  • line 39: For ALL these
    • ok
  • line 46/47:
    • single-top-quark
      • ok
    • ... using the HATHORv2.1 library.
      • ok
    • at least put a "respectively" somewhere. Why not is 136... for top quark and 81... for top-antiquark production ... "
      • rephrased
  • line 54: "... to simulate a sufficiently large number of events for this process." (remove "of reliable quantity").
    • ok
  • line 55: shape of THE data IN a multijet-EVENT-enriched sideband region
    • ok
  • line 58: single-top-quark
    • ok
  • line 65: ..recorded by a trigger requiring the presence...
    • ok
  • line 67: no new paragraph after this line
    • ok
  • line 68: citation at end of sentence (after algorithm)
    • ok
  • line 68/69: replace "it reconstructs ..." by "Single particles are reconstructed and identified by ..."
    • ok
  • line 72: remove "that is" and replace fulfils by fulfilling
    • ok
  • lines 79/80: can we remove the double-quotes around "isolated" and "anti-isolated" ?
    • ok
    • originates
  • line 84:
    • How about a colon after criteria ?
      • ok
    • typo: THEIR
      • ok
  • line 85: Loose electrons ARE REQUIRED ... and TO fulfil dedicated ...
    • ok
  • line 91: clumsy.
    • changed to: "The jet energy and resolution is corrected to account for the non-flat detector response in $\eta$ and $\pt$ of the jet."
  • line 97:
    • How about ... following CRITERIA ARE IMPOSED: (colon) the neutral ..."
      • ok
    • in the following lines replace "need to" by "are" or "are required" e.g. "... hadronic energy fractions are < 99% ..."
      • ok
  • line 100: "... should be > 0 and .... should be <99%"
    • ok
  • line 105: "... jets THAT ORIGINATE FROM the ..."
    • ok
  • line 110: remove "as", i.e. "... jets described above...."
    • ok
  • line 114: suggest to insert "i.e." -> "... variable, i.e. the transverse momentum ..."
    • ok
  • line 120 suggest: ... b-tagged (2j1t) (no quotes). Events from the 2j1t region are used for the", ie. start new sentence and remove "Only these "
    • sentences rephrased: "Events from this region are used for the differential cross section measurement"
  • line 121: measurement (singular)
    • ok
  • line 122: of which one or two -> one or two of which
    • rephrased
  • line 124-129: suggest remove this paragraph. Not necessary in a (short) PAS.
    • ok but may help as an overview of this long section
  • line 131: single-top-quark
    • ok
  • line 137: remove "third of" (redundant and sounds funny).
    • ok
  • line 140: justified BASED ON STUDIES OF
    • ok
  • line 141: "Under the hypothesis" sounds strange to me (english?)
    • removed
  • line 141/143: top-quark (hyphen)
    • ok
  • line 144: summed -> sum of
    • ok
  • line 157: with -> using
    • ok
  • line 185: modelled using a template of anti-isolated muons from events in data
    • ok
  • line 191: on -> of
    • ok
  • caption figure 3 and others:
    • suggest: the ratio of yields in data and simulation
      • ok
    • replace "striped" by "hatched" or "shaded". I'd take any bet that "striped" is no english
      • striped seems ok, similar words: "grey-striped", "cross-striped", "striped skirt", "striped ebony", ...
  • line 206: typo minimisation
    • ok
  • line 208: such AS to
    • ok
  • line 228: remove one "also"
    • ok
  • line 257: spell out "charm"
    • ok

Pedro da Silva on PASv7 (26/04/16)


  • L43 suggest to remove "under...69mb." This is a phenomenological parameter and it's hard to justify that 69mb is really a number corresponding to the total inelastic pp cross section.
    • ok
  • L59 there must be a latex thing going on here as t->bW\mu is a BSM decay of the top smile
    • ok, :-)
  • L93 in the forward region does the HF spacial granularity allow for >10 candidates? doesn't this requirement lead to a drop in efficiency?
    • in particular, this ID cut is the replacement for cutting on the neutral hadron fraction which is still mismodeled. JetMet claims an efficiency of >99% (see here)
  • Figure 4 y axis why <Events> is it really an average?
    • yes, it is normalized to the bin width; "<Events>" is the recommendation by PubCom although exact y-axis labeling rules are not 100% clearly stated in my opinion (see here)
  • L202/203 Not sure i understand this sentence: Is the shape derived from the fit? or are the signal yields derived from the fit performed in exclusive pT/y categories? Please clarify
    • slightly adapted: For the unfolding, no selection on the BDT discriminat is performed. Instead, the signal yield is estimated by performing separated ML fits to the combined \mtw and BDT discriminant distributions in exclusive top quark $\pt$ and $|y|$ bins of the measurement.
  • L207 Are these upper bounds really needed? What's the fraction of tops >2.4 in eta?
    • w/o these bounds, I need to deal somehow with the overflow which is technically/statistically a bit complicated. The fraction of signal events pT>300 GeV is ~1% and |y|>2.4 is ~6%.
  • L208/211 Would be good to start with a sentence motivating why these values are relevant for the unfolding stability/final correlations
    • changed to: The binning is chosen to yield less migrations between the reconstructed bins. This is quantified by the stability (purity) which are defined as...
  • L224 given you use the median, shouldn't you use the 16% and 84% quantiles to define the uncertainty? Or alternatively, are the distributions of the toys wildly non-gaussian such that the median is preferred over the mean?
    • the 1 sigma quantiles = 16%/84% are used. Indeed, the median is preferred because some systematics are asymmetric
  • L256-257 "reweighted histograms" is a funny expression, can you clarify what does it mean using alternative wording?
    • changed to: The uncertainty on the PDF is estimated by reweighting simulated events according to all variations of the NNPDF3.0 set
  • L258 Choice of the renormalization and factorization scales. Suggest to rephrase first sentences as this is jargon, related to the way CMS organizes its software.
    • ok
  • L267 in addition to [37] please quote TOP-16-008 and TOP-16-011 as these are 13 TeV measurements
    • ok
  • Regarding Section 7 could you add a table or at least a sentence where you give the typical numbers in % for the dominant sources of uncertainty
    • ok

type A


  • L1 suggest to move "is presented" to the end of the sentence
    • ok
  • L2 spell-out pp
    • ok
  • L10 suggest to remove "currently large experimental"
    • ok


  • L9 add (\pt) after "transverse momentum"
    • ok
  • L11/L12 move (4FS) and (5FS) close to "4 flavour" and "5 flavour" / differ -> differ respectively
    • ok
  • L15 start the sentence with "The analysis uses" and remove "are analyzed that were"
    • ok
  • L33/36/37/40/249/279/280 aMC@NLO -> MG5_aMC@NLO
    • ok
  • L47 space after Hathor
    • ok
  • L56 This data-driven modelling of the shape of multijet events -> The method used to measure the shape of multijet events in data
    • more elaborative: The method used to model the shape and estimate the yield of multijet events in data is detailed in Section...
  • L71/72 using the definition -> by
    • ok
  • L78 originates -> originate ?
    • ok
  • L81/85 This sentence is extremely long and complex. Suggest to simplify it adding a stop after "rejected." and then The identification criteria used to search for additional leptons are loosened and in addition the following requirements are applied: \pt>10(20) and \eta<2.5(2.5) for muons (electrons). Loose muons are furthermore required to have I_{rel}^{\mu}<20%.
    • ~ok
  • L191 as a minimization
    • ok
  • L215 within -> while attaining
    • ok
  • L234/237 quote BTV-15-001
    • ok
  • L279/280 "in the 4FS" or "in the 5FS" read better to me
    • ok
  • Figure 5 remove "\mu+jets" from the figures
    • ok
  • L286 single top-quark production via the t channel -> t-channel production of single top-quarks
    • ok
  • L289 suggest to start the sentence with "Within the experimental uncertainties," and add "based on NLO+PS generators in the 4FS or 5FS" after theoretical predictions remove "that are currently still large"
    • ok

Approval condition

  • Q: Why is the top pt well modelled in the control region? A: we have applied the top pt correction by default => make explicit in the PAS
    • ok

Andreas Meyer on PASv8 (30/04/16)


  • About the possible additional figures you answered: "If requested, it can be added as supplementary material." Personally I think the PAS is still short and I'd find it natural to include all relevant figures in it directly.
    • added the two other input variables
  • Line 205: I stumbled over the statement "taking into account the intrinsic kt of initial state partons" It is not clear to me what you are referring to. Is the term "intrinsic" really correct and appropriate ? Is it relevant for the exact definition of the parton-level top quark ?
    • see discussion in HN
    • intrinsic pT refers to the simulated (empirical) non-pertubative transverse momentum of the initial partons from the proton.

A type

  • figure captions: striped -> hatched
    • ok
  • General remark: You often make statements of the form: "the analysis focuses", "the measurement selects". I find these cumbersome, as the analysis and the measurement themselves are not actors. In most of those cases I would prefer passive voice. See suggestions below.
    • ok
  • abstract (last two lines): propose: ... order matched with .... generators. General agreement is found within uncertainties (replace good by general).
    • ok
  • line 12:
    • that -> which; remove: respectively; into -> in
    • ok
  • caption fig 1: Propose last sentence: Corresponding diagrams exist for single-top-antiquark production.
    • ok
  • line 17: Avoid active voice (and avoid suggesting that t->mubnu is a direct decay): "The measurement is restricted to events in which the top quark decays into a final state containing a b-quark, a muon and a neutrino"
    • ok
  • lines 20-25: suggest "the dataset is described in section 2" and similar for the other sentences.
    • ok
  • line 27: the measurement uses .. (see above).
    • ok
  • line 52: remove "order"
    • ok
  • line 63: propose: "... splitting. In most events this second b jet is outside of the detector acceptance because its transverse momentum is small." (in my ears b-jets are soft or have small transv. mom., but not soft transv. mom.).
    • ok
  • line 65: suggest to replace therefore by as a consequence
    • ok
  • lines 67 and 71: propose to avoid online and offline (jargon): write: line 67: The events are recorded .. line 71: For the final analysis the presence ...
    • ok; removed final
  • line 78: avoid active voice: "Events are selected ..."
    • ok
  • line 79: use here "muon candidates" once. Can then refer to "muons" w/o "candidates".
    • already in 71
  • line 88: suggest to insert "in order" before "to avoid"
    • ok
  • line 123: suggest to replace "for which" by "and".
    • ok
  • line 140: "amount of ... events" sounds funny to me. Events are countable, amount is not. Suggest to remove "events" and write ...the amount of remaining background is still very high ..."
    • ok
  • line 140: replace "therefore" by "For further separation" and rephrase the rest of the sentence. "Therefore" generally indicates a logical argument. In this case, IMO the term "for this purpose" or similar is more appropriate.
    • changed to: "To discriminate between signal and background events further..."
  • line 152: biased (missing ED)
    • ok
  • line 153: sentence is a bit lengthy, but ok. Maybe replace "since" by "as", ("since" seems a stronger causality, "as" is more a remark which is what is intended here).
    • ok
  • line 167: multijet EVENT yield
    • ok
  • line 171: , AND W+jets ...
    • ok
  • line 174: suggest to add ", respectively" at the end of the sentence after "yields".
    • ok
  • line 175: suggest "top-QUARK background yield AS"
    • ok
  • line 178: suggest, "In addition to providing better control OF the ttbar EVENT yield, THEIR INCLUSION is found to reduce the correlations between the estimated background yields.
    • ok
  • line 185: "distorted w.r.t parton-level counterpart is a bit hard to swallow for an experimentalist. Parton level is not an observable, and thus intrinsically distorted by itself". How about: "The reconstructed distributions are affected by detector resolution, selection efficiencies and kinematic reconstruction which lead to distortions with respect to the original event distributions. The size of these effects vary with the event kinematics."
    • ok
  • line 189: suggest to use same font for TUnfold as for PYTHIA.
    • ok
  • line 196: finish sentence after "Fig 4.", remove left column (right column).
    • ok
  • line 200: in the pt-distribution, esp. at pt<100 GeV, the data ...
    • ok
  • line 201: Propose: The signal yields are estimated from ML fits to the combined mT(W) and BDT discriminant distributions, separately in exclusive top quark pT and |y| bins of the measurement. For the fits no selection on the BDT discriminant is applied.
    • adapted to: "For the unfolding, the signal yields are estimated from ML fits to the combined \mtw and BDT discriminant distributions, separately in exclusive top quark \pt and $|y|$ bins of the measurement. No selection on the BDT discriminant is applied."
  • line 207: Very loose boundaries for the unfolding. Sounds strange. Maybe it is ok to remove this sentence.
    • not sure; events beyond this boundary are not considered -> still important to interpreted the last unfolding bins.
  • line 208: "... the binning is chosen such AS to minimise the migrations while retaining sensitivity to the shapes of the distributions. The stability (purity) are defined as the probabilities that the parton-level (reconstructed) values of an observable within a certain range also have their reconstructed (parton-level) counterparts in the same range. The size of the bins is chosen such that both quantities are larger than \sim 50%." Suggest to remove everything after "... in nearly ...." until "... 40%".
    • ok
  • line 215: no new paragraph here
    • ok
  • line 219: uncertainty (singular, really!
    • ok
  • line 221: How about the following "The central value and total uncertainty of the cross section measurement are calculated from the combination of all variations. For this purpose, multiple Gaussian distributions, one per uncertainty source, are diced around the unfolded nominal template with a width corresponding to the difference between the unfolded systematic variation and the unfolded nominal result. The distribution of the sum of the diced numbers is used to extract the final result. The central value is taken as the median and the total uncertainty as one standard deviation quantile.
    • ok; changed: "diced numbers" -> "diced yields"
  • line 227: ... systematics and THEIR impact on the measurement are described.
    • ok
  • line 233: ... data BY INVERSION OF ... the muon isolation CRITERION.
    • ok
  • line 238: b jet (not b jets).
    • ok
  • line 256: top-quark (hyphen)
    • ok
  • line 264: suggest present tense "have been" -> "are"
    • ok
  • line 276 top-quark mass (hyphen)
    • ok
  • line 280: add here "in-situ", i.e. "to the in-situ measured " to be clear that the total number is not taken from theory or similar.
    • ok
  • line 281: replace "these distributions" by "the data".
    • ok
  • line 285-287: suggest "In the first bin of top-quark transverse momentum, a low acceptance to select signal events and a large sensitivity to the systematic uncertainties lead to a large relative uncertainty, esp. in this bin.
    • adapted to: In the first top-quark $\pt$ bin, the low acceptance to select signal events and the large sensitivity to the systematic uncertainties leads to a large relative uncertainty which renders the observed deviation not yet significant.
  • line 289: no hyphen between top and quarks
    • ok

George Wei-Shu Hou on PASv9 (04/05/16)

  • l22 "Details ..." l24 "The ..." l25 "The ..." should all be lower case
    • ok
  • l135 "... be unambiguously defined"
    • ok
  • The only issue is l208: pT < 100 GeV seems rough and imprecise. Note that the bin goes up to 110 GeV, and the bin above is also in (milder) excess [to be true, though without reason, the binning of Fig. 4 top-right looks rather peculiar]. Furthermore, saying its harder below 100 GeV just seems vague and not good. Maybe one just says "around pT ~ 100 GeV ", or like in earlier versions, do not mention any number.
    • ok
  • BTW, I feel putting Fig. 4 on p. 7 would be better.
    • ok

Andreas Meyer on PASv9 (04/05/16)

  • usually comma before "respectively".
    • ok
  • lines 174 and following: "Therefore" in line 176 sounds "germanic" to me. Suggest to invert the order: "The fits are performed using a combined likelihood by taking the mT(W) distribution for events with mT(W) < 50 GeV, and the BDT discriminant otherwise. The shape of the mT(W) distribution is a powerful handle to estimate the multijet event yield whereas the BDT discriminant provides sensitivity to the other backgrounds and to the signal yields."
    • ok, adjusted the complete paragraph to avoid repetitions
  • line 208, I agree with George's comment above that a qualitative statement like "below a value of about 100 GeV " or similar would be better. The "<" sign implies that there is a clear value where the discrepancy kicks in.
    • ok
  • line 214: I have still not understood whether the statement about the intrinsic kt has anything to do with the definition of the top quark at parton level. I believe the parton level definition is independent of whether intrinsic kt is there. If intrinsic kt is mentioned, one should probably mention the use/need and maybe even its value. Propose to remove.
    • my understanding is that the whole scattering system is shifted by the intrinsic kt. The net pT of the sum of all final state particles is not 0 but exactly this kt. Hence the top quark pT gets very slightly smeared by this effect.
  • Line 214: The statement about the boundaries seems misleading. It could mean one of two things: 1) we do not quote values above 300 GeV and 2.4 resp. or 2) events above 300 and 2.4 at generator level are removed from the response matrix. Option 1) is obvious and goes w/o saying. Option 2) would be incorrect. If there is another reason that would justify the sentence, I would still like to remove the words "very loose".
    • it means that values above these boundaries are not quoted. Technically, there are not enough reconstructed events to justify an extrapolation into this regions. Since the figure caption does not mentioning the inclusion of the overflow, the sentence can be removed.
  • line 225: "attain" has a similar meaning as "obtain", "achieve" or "yield". I think you mean "retain..." or "keep the bias at a minimum".
    • ok
  • line 239: can we remove "conservative" ? Uncertainties should be (as) correct (as possible) and not conservative
    • conservative means here that 20% is used for simplicity and not the theoretical uncertainties of the cross sections. Also, using the theoretical cross section uncertainties may underestimate the actual uncertainty in this particular phase space.
  • lines 274 and following: up-variated ... etc. the verb "to variate" does not exist. Could write "down-varied" etc.
    • ok
  • line 291: no new paragraph.
    • ok
  • line 293: propose to end the sentence after "uncertainty" in order to avoid use of "render" and "yet". The term "yet" seems to imply that you have an expectation (which you should not). The term "render" implies that the relative uncertainty "does" something to the measurement. But the uncertainty is an intrinsic property of the measurement.
    • ok, but should one not provide an interpretation of the result?

Orso Iorio on PASv9 (04/05/16)

  • line 9: absolute rapidity -> do you mean absolute value of the rapidity y?
    • ok
  • Abstract vs line 19: In the abstract you use center; in the body centre, suggest to change in the abstract for consistency
    • ok
  • line 35: you state that aMCatNLO is used for other single-top channels, isn't tW done with POWHEG? Or am I misreading?
    • indeed
    • added: The simulation of single-top-quark production in the \tw channel uses \POWHEG version~1 interfaced with \PYTHIA.
  • line 235 : sources of systematics -> sources of systematic uncertainty
    • ok
  • line 296 : dataset -> isn't "data sample" better?
    • I like to keep dataset. I like to write "MC sample" because these are sampled from theory (can be arbitrary large) but dataset because it is somewhat fixed in size.
Topic attachments
I Attachment History Action Size Date Who Comment
PDFpdf 2016_04_24.pdf r2 r1 manage 351.5 K 2016-04-25 - 13:51 UnknownUser  
PNGpng Btag.png r1 manage 17.7 K 2016-04-08 - 14:18 UnknownUser  
PNGpng En.png r1 manage 17.0 K 2016-04-08 - 14:18 UnknownUser  
PNGpng QScaleWjets.png r1 manage 18.8 K 2016-04-08 - 14:19 UnknownUser  
PNGpng TopMass.png r2 r1 manage 17.8 K 2016-04-25 - 13:44 UnknownUser  
PNGpng effPt.png r1 manage 14.2 K 2016-04-08 - 13:52 UnknownUser  
PNGpng effY.png r1 manage 14.7 K 2016-04-08 - 13:52 UnknownUser  
PNGpng gen_top_pt_QScaleTChannel.png r1 manage 21.2 K 2016-04-25 - 13:31 UnknownUser  
PNGpng gen_top_pt_TopMass.png r1 manage 20.8 K 2016-04-25 - 13:34 UnknownUser  
PDFpdf merged.pdf r1 manage 375.1 K 2016-04-08 - 14:49 UnknownUser  
PNGpng reco_top_pt_Btag.png r1 manage 23.4 K 2016-04-25 - 13:40 UnknownUser  
PNGpng reco_top_pt_En.png r1 manage 22.4 K 2016-04-25 - 13:37 UnknownUser  
PNGpng reco_top_pt_PU.png r1 manage 22.5 K 2016-04-25 - 13:42 UnknownUser  
PNGpng reco_top_pt_QScaleTChannel.png r1 manage 23.9 K 2016-04-25 - 13:31 UnknownUser  
PNGpng reco_top_pt_TopMass.png r1 manage 23.0 K 2016-04-25 - 13:35 UnknownUser  
PNGpng unfolded_top_pt.png r1 manage 31.3 K 2016-04-25 - 13:27 UnknownUser  
PNGpng unfolded_top_pt_jeta.png r1 manage 31.5 K 2016-04-25 - 13:28 UnknownUser  
PNGpng unfolded_top_pt_minuit.png r1 manage 31.5 K 2016-04-25 - 13:28 UnknownUser  
PNGpng unfolded_top_pt_minuit_bdt.png r2 r1 manage 30.7 K 2016-04-18 - 12:17 UnknownUser  
PNGpng unfolded_top_pt_newton_bdt.png r2 r1 manage 30.5 K 2016-04-18 - 12:17 UnknownUser  
PNGpng unfolded_top_pt_newton_etaj.png r2 r1 manage 30.6 K 2016-04-18 - 12:17 UnknownUser  
Edit | Attach | Watch | Print version | History: r52 < r51 < r50 < r49 < r48 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r52 - 2016-05-05 - unknown
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Main All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright &© 2008-2021 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback