10 - H1 - Desy

86 Calibration and tuning
• Background BG.
The third sample used to study the quality of MC simulation of the background
clusters is selected by using photon candidates with low isolation z < 0.9 and no
cut on the cluster radius of the photon candidate. In this phasespace, selected events
are purely neutral hadron background.
a. b. c.
e
p
γ
γ
e e
e
p p
p
e
p
e
p
γ
Figure 7.1: Feynman diagrams illustrating the Bethe-Heitler events (a and b) and deeply
virtual Compton scattering (c).
All the samples are statistically independent of the selection used for final results determination
and provide a clean selection of electromagnetic clusters. The evaluation of
the systematic uncertainty follows the method developed in [66]. MC cluster shape v is
distorted by a stretching factor k:
v ′ (k) = v · (1 + k ), (7.1)
100
and its distribution is compared to the distribution measured in the data. An example
of the comparison is presented in figure 7.2, where the FLF MC distribution in wheel
CB1 is stretched against the data distribution measured in the BG sample. One may see
that for negative k the MC distribution is shifted towards too low FLF values, while for
positive k in the opposite direction. For k close to zero, both distributions are comparable,
being at it best for k ≈ 2.
For each factor k the χ 2 can be calculated between the data histogram and the stretched
MC distribution:
χ 2 = ∑ (b D i − b MC
i ) 2 /(σD,i 2 + σ2 MC,i ), (7.2)
i
where b D i and b MC
i denote the data and MC histogram content for bin i, σ its error and the
sum runs over all non-empty histogram bins. The number of non-empty bins correspond
to the number of degrees of freedom (NDF).
Figure 7.2 shows the χ 2 (k)/NDF dependence for all six shower shape variables in CB1
wheel as studied with the BG sample. A clear minima close to k = 0 can be observed
in each case. The min(χ 2 ) is used for the correction applied to the MC simulation of the
cluster shapes, while min(χ 2 )±NDF for the evaluation of its uncertainty. In case of DVCS
and BH samples, error deduced by the χ 2 method is dominated by statistical error of both
samples, which leads to an overestimated uncertainty of the cluster shapes description.