Data samples used

These samples all have the 30 micron cut in LAr.

• GJ1: trig1_misal1_mc12.008095.PythiaPhotonJet1_FIXED.recon.AOD.v12000604_tid008034
• GJ2: trig1_misal1_mc12.008096.PythiaPhotonJet2_FIXED.recon.AOD.v12000604_tid008033
• GJ3: trig1_misal1_mc12.008097.PythiaPhotonJet3_FIXED.recon.AOD.v12000604_tid008035
• GJ4: trig1_misal1_mc12.008098.PythiaPhotonJet4_FIXED.recon.AOD.v12000604_tid008032
• GJ5: trig1_misal1_mc12.008099.PythiaPhotonJet5_FIXED.recon.AOD.v12000604_tid008031
• GJ6: trig1_misal1_mc12.008078.PythiaPhotonJet6_FIXED.recon.AOD.v12000604_tid008030

I'm using Cone7 jets in truth and Cone7 tower jets in reco.

The above samples all have both ckin3 and ckin4 cuts. When we naively combine them, edge effects become clear. Here is the true balance, as a function of the average pT of the photon and thejet. The blue lines indicate the pT at which one sample ends and the next begins.

To avoid edge effects, we merged the samples using their relative cross sections. Events were killed events but never weighted or duplicated. So for example for the sample with pT>35 GeV, all available GJ2 events were used; the right amount of GJ3, GJ4, and GJ5 was added according to the ratio of cross sections. The resulting samples are similar to trigger samples in that they have a minimum pT cut but no maximum pT cut.

Here is an example of how three "trigger samples" were created and their integrated luminosities:

• pT>35: 149000 GJ2 + 16802 GJ3 + 1482 GJ4 + 92 GJ5 + 4 GJ6 = 9.1 pb-1
• pT>70: 100000 GJ3 + 8821 GJ4 + 547 GJ5 + 23 GJ6 = 47 pb-1
• pT>140: 100000 GJ4 + 6310 GJ5 + 262 GJ6 = 535 pb-1

The samples created this way can be combined by defining certain pT ranges where each sample is used, steering clear of edges. For example the pT>35 GeV sample is used in the range 40<pTavg<80 GeV; pT>70 GeV sample is used in 80<pTavg<160 Gev, etc when considering the balance. With this treatment, the true balance looks like this:

The statistics are reduced since the region close to a sample's turnon has to be avoided, but the edge effects are no longer statistically significant.

In all of the studies below, the three samples listed above (pT>35, pT>70, and pT>140) were combined with the pT>17 GeV sample as described and used. The higher pT samples weren't included because I was trying to restrict this study to a realistic integrated luminosity for early running (535 pb-1 might be too optimistic already). The full sample then looks like this:

Truth Balance

Photon selection in truth is based on the pdg ID; the hardest photon in the event is used for balancing. Selecting the truth jet is a little bit more complicated: there is always a truth jet formed around the photon and sometimes this is the hardest truth jet in the event. To avoid using this as the recoil jet in the balance, the hardest truth jet with dR from the photon greater than 0.7 is used. This is not meant to be a cut to reduce ISR; it is merely to avoid balancing the photon against a jet formed around it. Only jets with |eta|<2.5 are considered; this cut could be removed, but I would like to avoid the forward regions for now.

The balance is binned in average pT (0.5*(pTjet+pTgamma)), with constant bin sizes of 10 GeV for now.

The ratio between the pT of the photon and the pT of the truth jet for four pT bins is shown here:

There are significant tails due to ISR (shown elsewhere in the note). Because the tails aren't symmetrical, the most probable value of the histogram is estimated using a Gaussian fit. The Gaussian fit is done once to estimate the mean and the sigma, and on the second iteration the bounds of the fit are one sigma.

The balance as a function of pT:

Using the average pT as a reference (on the x-axis in the plot above) is a way of avoiding bias from ISR. But if the jets are miscalibrated it is difficult to understand that quantity. It might be more useful to just use the photon pT as a reference. For a certain bin of photon pT or average pT, the balance histogram looks like this:

The balance as a function of photon pT looks like this:

A cut on phi(gamma)-phi(jet) can help in avoiding biases and tails due to ISR. The same plots as above are shown again below with the photon as the reference, this time requiring that |pi-|phi(photon)-phi(jet)|| < 0.2. The balance is significantly improved.

A summary of the truth balance with different references is shown below with wider bins.

Photons in 12.0.6

The default photon selection used is "Photon" in HighPtView. This corresponds to a tuned selection by Sandrine LaPlace plus a slightly outdated version of a tool provided by Guillaume Unal for recovering converted photons. In addition the efficiency and balance were studied with a tight photon selection which is described below in the background section. The plots below are for the leading photon in each event; other photons are just soft photons within jets and don't matter to the balance. Here is the reconstruction efficiency with respect to truth as a function of pT and eta (pT>70 GeV sample):

In gamma-jet events, usually when the leading truth photon is not reconstructed, no other photon in the event is reconstructed either. Occasionally, however, a jet in the event fakes a photon. In this case the leading reconstructed photon is not the leading truth photon. For the samples being studied, this only occurred around 0.4% of the time.

The photon calibration for reconstructed photons is good to within 1% (This is true for both default and tight selections, though only the default is shown here):

I think the default photon trigger in 12.0.6 is not optimized. Probably the best place to get an efficiency curve is from the people studying the photon trigger for the direct photon note. Using the default Event Filter isolated gamma 60 GeV trigger, which should have no prescale, the efficiency curve looks like this for the default photon selection:

Jet reconstruction

Again jets with |eta|<2.5 are considered. We are working on a leading jet balance, so only the jets selected for balance with the photon are considered. The jet selection is simple: the hardest jet with dR(photon, jet)>0.7 is used.

Every event in the sample being considered had at least one reconstructed jet. Occasionally the leading reconstructed jet did not match the leading truth jet, meaning that a second jet in the event had higher pT in reconstruction. This happens more at lower pT where there often several truth jets with similar pTs. In the lowest pT sample (pT>17 GeV) it happened in around 15% of the events; this decreases to 4% for pT>35 GeV and 1.5% for pT>70 GeV.

The jet reconstruction efficiency for cone7 jets is shown below. Again this is only the efficiency for the jet selected for balance against the photon. The efficiency gets flat around 100 GeV.

The calibration for the leading jets in 12.0.6 is shown below. Here the ratio between the reconstructed jet pT and the matched true jet pT is shown. There is roughly a 1% miscalibration, relatively constant as a function of pT.

The calibration as a function of eta shows that the regions around the crack are not well-calibrated.

Reconstructed balance

The reconstructed balance with the photon pT as reference is shown below for the two different photon selections. A delta(phi) cut is applied. The truth balance is plotted for reference.

The balance as function of the jet eta is shown below. The red points are truth and the black points are reconstructed.

Since the crack regions have uncalibrated jets, the balance was compared using all jets with |eta|<2.5 and using only those jets in the central region with |eta|<1.0. There is a small change in the balance at the level of 1%.

Cone4 jets

The photon and jet balance will be affected by out of cone activity and underlying event; they roughly cancel with Cone7 jets. Doing the same balance study with narrower jets should allow us to understand these effects a bit better. Shown below is the balance with Cone4 jets; the truth balance is shown in red while the reconstructed balance is shown in black. The truth and reconstructed balance looks very similar, so the imperfect balance is not due to the jet calibration. Instead, it is probably due to losses outside of the cone.

Background study

QCD dijet and multijet events with one of the jets faking a photon should present a significant background to photon + jet events. The cross section for dijet production is ~3 orders of magnitude higher than that of gamma+jet production in the relevant pT range.

The standard dijet J1-J6 (5010-5015) were used to estimate the effect of the background on the photon + jet balance.

Isolation criteria in the photon selection improve the jet rejection, since photons in gamma-jet events should be relatively isolated. The calorimeter isolation variable used is etcone, which describes the transverse energy deposited in a cone of radius 0.45 around the photon in the calorimeter. Tracking isolation was exploited by requiring using nTrack, the number of tracks pointing to a jet of radius 0.7 around the photon. The distribution of etcone/pT(photon) and nTrack for photons in signal and background events is shown below.

The cuts used in the tight selection were: etcone/pT(photon)<0.05 and nTrack<3. As can be seen from the figure above, the etcone cut will have a low efficiency for signal events at low pT. It would be nice to use a more sophisticated cut to improve this efficiency, but the focus of this studies was jets higher than 100 GeV.

The photon efficiency for the two photon selections was shown above. The jet rejection is shown below for the default photon selection and the tight photon selection.

The reconstructed photon often has much lower pT than the jet faking it. Shown below is the fake photon pT as a function of the pT of the jet faking the photon.

The signal to background ratio is shown below for the default photon selection, the default photon selection with a delta(phi) cut, and the tight selection described above with a delta(phi) cut. The available statistics for the dijet sample are poor, so the uncertainty in the S/B ratio is large, but it is greater than 3 over most of the range. At pT < 50 GeV there is more background than signal, but the cuts have not been optimized for this region. The second plot shows the photon spectrum for the photon + jet and dijet samples with the tight photon selections and delta(phi) cut.

The balance in two large pT bins is shown below for signal and background events. The left plot is just signal events with the Gaussian fit for extracting the most probable value superimposed. On the right the background events are shown in red and the points with error bars show the sum of signal and background events. Signal and background are normalized to the expected number of events for 100 pb-1. A Gaussian fit is done on the total distribution. The MPV is very similar for the signal distribution and the signal+background distribution. But the fit to the signal+background is not a realistic estimate of the effect of the background on the balance, since the bins with background events have very large errors due to low statistics. Thus these bins will not affect the fit strongly. A more realistic estimate of the effect of the background would need to interpolate some kind of smooth distribution for the background balance to add to all bins, but there are not adequate statistics to try this.