I implemented a cut set (#6) similar to cut set 1 except using a cut on the likelihood of each cluster. The conditions are

• met>20 GeV
• ht>100 GeV
• at least 1 track/pi0 cluster with likelihood > 0.4

The numbers are:

Compared to cut set 1, i.e. pT(cluster)>10 GeV w/ no likelihood cut, we have reduced the signal by 8.6% and the background by 40.6%.

–Dan

Following Carla’s work deriving a likelihood for jet based identification, I have implemented a similar scheme for track/pi0 clusters. I make a series of parameterized likelihood for y(i)=B(i)/S(i), where i is a particular variable, S is the signal Higgs sample, and B is the combined W+jets sample. Then I multiply all the y’s together as uncorrelated likelihood ratios and create the final likelihood as L=1/(1+y). The plots of each y are given below. Continue reading Track/pi0 Cluster Likelihood Ratios →

Below are 3 correlation matrices for track/pi0 cluster variables.  The matrices are formed from

1. The signal sample.
2. The W+jets sample.
3. The difference between the signal and W+jets correlation matrices.

Notice that the signal sample matrix is more anti-correlated across all variables than the W+jets sample.  All variables exhibit |rho|>10% for all combinations, i.e. there are no highly uncorrelated pairs of variables.

–Dan

I have implemented a cut set based on this one except with the additional requirements that the event contains a soft electron and that the charge correlation for clusters containing an electron be negative. The requirements are as follows:

• met>20 GeV
• ht>100 GeV
• at least 1 soft electron
• at least 1 track/pi0 cluster with pT>10 GeV and N(live)<7.
• if the soft electron is inside a cluster, qCorr<0.

The expected signal and background for these cuts are

I have also made tables of N(live), N(tracks), N(pi0s) for the following cases:

• Event contains a soft electron but it is not inside the cluster. Table is here.
• Event contains a soft electron and it is inside the cluster; no cut on the charge correlation. Table is here.
• Event contains a soft electron and it is inside the cluster and qCorr<0. Table is here.

The 5 most significant N(live), N(tracks), N(pi0s) bins for the 3 cases are listed below.

Soft Electron Outside Cluster

Soft Electron Inside Cluster; no cut on qCorr

Soft Electron Inside Cluster; qCorr<0

Notice that in all of the cases, the most significant bins are with N(live)<4. Also notice that the cut on charge correlation cuts 29% and 44% of the background in the (2,2,0) and (2,2,1) bins, respectively, while barely taking any signal away at all. This is intuitive in the 2 track bin, since one of the tracks must be the soft electron and therefore the other one is a 1-prong tau decay. Since the taus are coming from a neutral a0, the 2 track bin is the most pure for this cut. --Dan

Here is a table that breaks down the number of signal and background events w.r.t. N(live), N(tracks), and N(pi0s). N(live) is as defined before. N(tracks) and N(pi0s) are the tracks and pi0s in a given cluster with the given N(live). The most signal-like bins are:

Notice that the most “signal-like” entries are in N(live)=2 and N(live)=3.

I want to do 2 more things with this table:

1. Change the counting of N(tracks) to count the soft tracks in the cluster cone that do not pass threshold. I did a similar thing earlier when calculating N(live).
2. Require the presence of a soft electron inside the cluster and recalculate the table. I expect clusters containing a soft electron to be more signal-like. We might be able to get further separation for those SE-tagged clusters by looking at N(live), N(track), N(pi0).

–Dan

I implemented a cut set similar to Cut Set 1, shown here .

The cuts are as follows:

• tight central lepton pT>20 GeV
• Raw Met>20 GeV
• Raw Ht>100 GeV
• at least 1 track/pi0 cluster with
• N(live)<7
• pT>10 GeV

The results of this cut set compared to the cut set 1 (the same except for the N(live) cut) are

We have reduced the signal by 3.8% and the background by 19%.  Clearly, this cut is a keeper.

In a previous post, I defined the variable N(live) for track/pi0 clusters.  I was initially using tracks that were owned by the cluster, i.e. tracks that passed the seed and shoulder pT thresholds.  I wanted to see what effect these thresholds have on the N(live) distribution.

The plot below compares the N(live) distribution using owned tracks (in red) and all tracks within the dR threshold of the cluster, in this case dR=0.5, w.r.t. the seed track.  This plot was made on the signal MC sample and required that a generator level tau must be present inside the cluster.

As expected, when counting all tracks the mean of the distribution increases.  The increase is relatively small for these tau-tagged signal MC clusters; the mean goes from 2.52 to 2.89.  We didn’t expect this to be a huge effect for clusters coming from taus, as they should have relatively few tracks.  The effect is more dramatic when comparing to the W+jet background, as seen in the plot below.

We achieve much better separation in the N(live)=1,2,3 bins when using all tracks in the event to calculate it.
–Dan

Define a new variable, N(live), which we expect to discriminate between jets and di-taus. The definition is as follows:

• Extrapolate tracks in the track/pi0 cluster to the calorimeter
• Count the number of towers that have a track from the cluster extrapolating to the OR a pi0 identified in them AND have total energy > 100 MeV. This number is N(live).

Below is the plot of N(live):

The signal sample has more probability to have 1-3 towers live compared to W+jets. ttbar tends to have more “live” towers in the cluster than the other samples.

Next we look in each bin where the Higgs signal is dominant, N(live)=1,2,3

daDefine a new variable, N(live), which we expect to discriminate between jets and di-taus. The definition is as follows:

• Extrapolate tracks in the track/pi0 cluster to the calorimeter
• Count the number of towers that have a track from the cluster extrapolating to the OR a pi0 identified in them AND have total energy > 100 MeV. This number is N(live).

Below is the plot of N(live):

The signal sample has more probability to have 1-3 towers live compared to W+jets. ttbar tends to have more “live” towers in the cluster than the other samples.

Next we look in each bin where the Higgs signal is dominant, N(live)=1,2,3

It looks like we can cut clusters have large number of tracks and pi0s in all the above cases. The cut should depend on the N(live) multiplicity.

We have defined 3 cut sets

Cut Set 1

• met>20 GeV
• hT>100 GeV
• N(TP0 with pT>10)>0

Cut Set 2

• met>20 GeV
• N(jets)>1
• Et(jet 1, L5)>15 GeV
• Et(jet 2, L5)>5 GeV

Cut Set 3

• met>20 GeV
• hT>100 GeV
• N(track/pi0 clusters)>1
• pT(leading track pi0 cluster)>15 GeV
• pT(sub-leading track pi0 cluster)>10 GeV

The table below shows the W+jets and signal expectations for these 3 cut sets.

The following selections are used to create track/pi0 clusters:

• standalone CES pi0’s with correction for track overlap, pT>500 MeV
• seed track with pT>2.5 GeV
• if pT(seed)<10 GeV use low pT track selection (defined below)
• if 10<pT(seed)<20GeV use OR of high and low pT track selection.
• if pT(seed)>20 GeV use high pT track selection
• seed track cannot be an identified tight central electron or muon
• |z0(seed)|<60 cm
• |d0(seed)|<0.2 cm
• 9 cm<zCES(seed)<230 cm
• pT(shoulder)>2.0 GeV
• N(COT hits total for shoulder track)>5

Low pT track selection

• N(COT Axial hits)>19
• N(COT Stereo hits)>19
• N(COT Axial segments w/ 6 hits)+ N(COT Stereo segments w/ 6 hits)>1

Hi pT track selection

• N(COT axial segments w/ 5 hits)>2
• N(COT stereo segments w/ 5 hits)>1

Signal/Background Comparison Plots

Met:

pT(TP0):