Multivariate Analysis in the Reconstruction of Photon/Electron Energies in the CMS

A new semi-parametric multivariate regression was used to improve the energy reconstruction in the CMS electromagnetic calorimeter. The method is based on the generation of boosted decision trees by optimizing the parameters of the double crystal ball function fitted to the ratio of the raw to generated energies of simulated photons and electrons. The full training was done on half the electrons with generated transverse momenta p$_{T}\geq$ 16 GeV in the barrel and corrections were applied to subsets of the remaining events. The dependence of the means and widths of the resulting distributions on p$_{T}$ was deduced. The corrected reconstructed energies peak close 1 for p$_{T}$ values down to 16 GeV. It was found that fixing $\alpha$ of the double crystal ball function in the training improves its performance.