No permission to view Atlas.AtlasPublicTopic

BackgroundEstimationforInclusiveSUSYSearches

Document and Abstract

---

o ATL-PHYS-PUB-2009-077 document in CDS

---

o We discuss a new data-driven estimation technique, denoted {em Tiles Method}, for Standard Model (SM) background in inclusive SUSY searches. The Tiles Method translates prior knowledge on the SM distributions of discriminating variables in a two or higher dimensional grid into an estimate of the abundances of SM and beyond-SM events. Depending on the grid granularity, the abundances are calculated by solving a system of linear equations or by minimising a log-likelihood function. The method does not rely on assumptions on background dominance for particular values of the variables, nor does it require iterations. Correlations between the variables are fully taken into account for SM events, while they are assumed to vanish for beyond-SM events. Systematic effects due to uncertainties in the simulated prior distributions are investigated. Results for various mSUGRA scenarios are presented.

Plots and Tables

Figure 1 SUSY (SU3) and SM background distributions of events selected according to the cuts described in Table 2. From top left to bottom right: Meff, Emiss, MT , lepton pT. The cut MT > 100 GeV was dropped for the MT distribution.
Figure 2 Transverse mass (MT ) versus effective mass (Meff) distributions for simulated SM background events (left) and SUSY SU3 events (right). Indicated by the capital letters are the 2×2 tiles determined by the cross borders along Meff = 800 GeV and MT = 100 GeV. The correlation coefficients are 6.6% (SM) and 10.7% (SU3).
Figure 3 Demonstration of 8×8 tiles setup. Transverse mass (MT ) versus effective mass (Meff) distributions for simulated SM background events (left) and SUSY SU3 events (right). Each tile is denoted by a tuple (i, j), e.g., the number of observed events in the top left tile is called N1,1.
Figure 4 In 5000 toy experiments, each representing 1 fb^−1 of SM background and SU3 signal, the signal yield is fitted using the Tiles method for different tile configurations: 2×2 (top left, top right shows the corresponding pull distributions), 8×8 (middle left), and 12×12 (middle right). Open/black circles and the solid/blue lines represent results from the toy experiment and Gaussian fits, respectively. Filled/black circles and the dashed/blue lines represent the same, but for experiments with variable correlations turned off for signal. The mean chi^2 provides a measure for the goodness-of-fit. The bottom two plots show the evolution of the Gaussian s (left) and absolute bias w.r.t. true MC (right) versus the tile configuration.
Figure 5 Distributions of Meff (left) and MT (right) for SM background decomposed into W+jets (open/black circles), tt →ℓn ℓn (filled/red squares), and tt →ℓn ℓn (filled/blue triangles) contributions. Both plots are after the one-lepton event selection (the MT requirement has been dropped in the MT plot).
Figure 6 Stability test for the Tiles method using 2×2 tiles. The estimated number of SM background events (filled/red squares) is plotted versus the Meff tile boundary for SU3 (top) and SU4 signal (bottom plots). The true value is indicated by the dotted line. The signal correlations are turned off for SU4 in the bottom right plot.

Table 1 Definition of global event variables.
Table 2 Event selection for the one-lepton search channel.
Table 3 Summary of results obtained for several SUSY benchmark models [4] using the tile configuration 8×8. For all SUx points, the statistical error (sigma^S) and bias (Delta N^S) relative to the true number of signal events are given. Mean and sigma are obtained from Gaussian fits to the toy distributions (cf. Fig. 4).
Table 4 Relative systematic errors on the signal yield for 8×8 tiles and various SUSY benchmark models. Also given are the statistical errors and the biases due to signal correlations obtained from toy experiments. Due to the nature of the systematic studies we have done, these numbers should not be taken as the final systematic uncertainties we would expect from this background estimation technique, but should give an idea of how sensitive the method is to some of the main systematic effects.

---

Major updates:
-- TillEifert - 2009-09-16

Responsible: TillEifert
Last reviewed by: Never reviewed