Mathis Dubau

Mathis Dubau · Research

Forward electrons

DNN-based calibration and identification of forward electrons for Run-4, with the new variables of the ITk and the HGTD up to |η|<4.0|\eta| < 4.0.

The forward region with the ITk

The replacement of the Inner Detector by the all-silicon ITk, together with the timing information from the HGTD, opens the forward region to track-based electron reconstruction for the first time in ATLAS. The work presented here exploited these new handles to develop a dedicated calibration and identification chain for forward electrons in the high-pile-up environment of the HL-LHC, extending the coverage out to |η|=4.0\lvert\eta\rvert= 4.0.

Layout of the ITk and HGTD with lines of constant pseudorapidity up to 4
Scheme of the thesis · download

Calibration

The pTp_\mathrm{T}-calibration was addressed with a DNN regression combining the cluster moments, the layer energy fractions, the tracker hits, the track–calorimeter matching variables, and the HGTD timing into a single calibrated estimator pTDNNp_\mathrm{T}^\mathrm{DNN}. The track-only measurement is unusable in the forward region owing to the very short lever arm, while the calorimeter-only and combined measurements suffer from a sizeable bias and broad tails, in particular in the forward calorimeter region. The DNN instead recovers a response centred on unity in all three |η|\lvert\eta\rvert bins, removing the bias, with a resolution equal to that of the calorimeter-based methods in the two inner bins and markedly better in the forward calorimeter region, where their tails are largest.

Thesis, Figure 5.14(c)(ii). Ratio of reconstructed to true pTp_\mathrm{T} (3.2<|η|<4.03.2 < \lvert\eta\rvert< 4.0)

Identification

For the identification, a second DNN classifier was trained together with a pTp_\mathrm{T}-decorrelation technique, in which the cluster moments are replaced by the residuals of a linear regression against pTDNNp_\mathrm{T}^\mathrm{DNN}. This provides working points with a flat signal efficiency across the full pTp_\mathrm{T} range and a substantially better background rejection than the legacy LH and cluster-moment selections, which were not designed for the stringent conditions expected in Run-4. The decorrelation further enables a smooth and bounded extrapolation of the working-point thresholds beyond the training range, down to 5GeV5\,\mathrm{GeV} and up to 500GeV500\,\mathrm{GeV}. The complete chain, comprising the calibration, the identification, and the working-point definition, was implemented in Athena.

Thesis, Figure 5.20(c)(ii). ROC curve (3.2<|η|<4.03.2 < \lvert\eta\rvert< 4.0)
More: electron reconstruction and calibration in ATLAS
Flowchart of electron and photon reconstruction
Scheme of the thesis · download
Flowchart of the electron and photon energy calibration chain
Scheme of the thesis · download