A complete new software environment for analysis,H1OO, has been developed, to com-plement the hardware upgrades for HERA II. The new software environment based on ROOT [109], written in C++ and utilising object oriented programming techniques, was designed and implemented over the course of the HERA luminosity upgrade project. A summary of the H1OO project can be found in [110].
Data Storage
The physics data scheme consists of three layers of storage. At the lowest level is the Object Data Store(ODS), which is completely equivalent to the DST. The ODS stores the same track, cluster and other detector-level information as the DST, al-beit represented as C++ objects. These are the classes H1Track, H1Cluster and H1Cell. In practice, to avoid duplication of information on disk, the ODS layer is usually created “on the fly” when accessing the DST. The DST contents are read and the ODS information for the event is created in memory. This has only a small performance penalty compared to accessing persistently stored ODS files. In this way, the transient ODS storage layer functions as an interface from the DST tapes to the H1OO software. The second level, the micro-ODS (µODS), allows fast access to particle level information. The µODS stores identified particle four vectors (for example electromagnetic particles, hadronic particle hypotheses) and associated in-formation. The third event layer, known as the H1 Analysis Tag (HAT), contains event level information such as the reconstructed vertex position, trigger information or kinematic quantities. The HAT is a flat tree, storing only simple variable types rather than collections of objects. This allows a fast pre-selection of events. At each state, the storage space needed per event decreases and the size of an event on the HAT is significantly smaller than on the ODS. The data storage levels are represented
9For most of the 2003-04e+pperiod ST57 had ToF veto requirements, as well. therefore it was used to monitor the CIP vetocondition only during that time. For the last part of the 2004 e+p and whole 2005e−ppart of running ST57 was loosened so that it contained no veto conditions, as given in the table above.
schematically in figure 3.28.
! #"%$&# (')
*
,+&
*
-.
/,01
2')2
34
"5 (')
*
6798;:=<?>
-.
@
+&&A
=$CBDB 4 " 1
EF
G
#"H
3
I'JK')
'LAJM"%B
4
('N
:7PORQ
:=STS UWV;XZY\[^]`_HabVcX
degf.hjilknmpopq r [ts=[tuvs(V;]xwzy\_C[t{
]|[tuHVcXt_&s,]~}uvsAabVcX
[
s(V;]`y^t[t_s=[;
Figure 3.28: A schematic overview of the data storage levels used by theH1OOProject.
Finally, the H1OOenvironment allows for further data layers to be added. The “nor-mal” layers of storage described above can be extended by specially filled trees con-taining user-defined information. These User Trees allow persistent storage of spe-cialised information. Reading User Trees is faster than reading and processing ODS (or POT raw data) information. User Trees allow for experimental extensions of basic H1OOobjects, filling of detector level information found only on DST (or in raw data) or more sophisticated physics finders used only for a small subset of analysis.
In this analysis, detailed information on the LAr trigger is used to evaluate the effi-ciency of the principal trigger element for high Q2 neutral current, LAr electron 1.
The list of big towers that fired for each selected NC event is stored in a User Tree for convenient offline analysis (see section section 6.8.1 for discussion of the results of this study).
Generic Analysis
While theH1OO project provides a common basis for analysis, with expert knowledge available to all users and the ready availability of all quantities suitable for analysis, there still remains a need to standardise common analysis tasks. Hence the need for a generic analysis framework to complement the H1OO software with high level anal-ysis tools. Tasks like event selection book-keeping, histogramming, determination of event weights and binning of kinematic variables are common to all kinds of physics analysis. The H1OO generic analysis framework, described in further detail in [111], provides tools to accomplish all these tasks. The steps required for analysis are fur-ther formalised with dynamically steerable 10 analysis objects. This allows users to quickly build in a simple way, or extend, an analysis code base comprised of objects representing the tasks necessary in the analysis.
The “Calculator” package is an important part of the generic analysis framework, which interfaces to the H1OO data. The Calculator is essentially a transient event layer, lying between the data storage and the high level analysis tools. As well as providing further information calculated from µODS/HAT variables, the Calculator allows the determination and combination of event weights for data and simulation.
It also provides a mechanism for the propagation of systematic uncertainties, e.g. in the scattered electron energy, through the re-calculation of the event kinematics. The effect of possible systematic mis-measurements can be accounted for by propagating a shift of these quantities by ±σ up to the final result. This is effective, and much less expensive in terms of CPU time and storage space than re-creating data files for each systematic shift.
The analysis presented in this document has been performed in the H1OO analysis framework using the generic analysis tools.
10The H1OO framework provides a steering mechanism, allowing run time behaviour to be set by text files.
Monte Carlo Simulation
To make a well controlled physics measurement, a detailed simulation of the physics processes and detector response is needed. For these purposes, stochastic techniques are commonly used. These techniques, which use random numbers and probability distributions to simulate physical processes are termed Monte Carlo (MC) methods.
In order to ensure that the statistical error of the simulated sample can be, to a good approximation, ignored, the simulated event samples are generally required to be several times the size of the data samples.
A cross section measurement requires corrections for acceptance, and an understand-ing of the influence of the resolution of the detector components on the final result.
These are difficult to determine directly from data due to the complex interplay of different detector effects.
Monte Carlo event simulations are also used in developing the data selection criteria.
The Monte Carlo programs can help to determine which variables are particularly useful for separating signal from background. In developing the selection criteria and determining the acceptance corrections, it is necessary that the Monte Carlo simula-tions accurately describe the data. In addition, many different Monte Carlo generator programs must be used, each describing a specific class of process. It also means that detailed simulations of the detector response to particles must be performed, in active as well as in inactive materials.
When the discrepancy of simulation from data is observed the MC simulation is adjusted to model the data behaviour. The Monte Carlo is also used to model an inevitable smearing of reconstructed variables due to finite detector resolutions. The efficiency of the selection criteria, the detector calibration and the resolution are de-termined directly from data and are implemented in MC.
4.1 Generation of DIS Events
Deep inelastic scattering processes are generated using the DJANGOMonte Carlo sim-ulation program [88], which is based on LEPTO [90] for the hard interaction and HERACLES [87] for single photon emission off the lepton line and virtual EW correc-tions. LEPTO combines O(αs) matrix elements with higher order QCD effects using the colour dipole model as implemented in ARIADNE [91]. The JETSET program is
used to simulate the hadronisation process [92].
The events are generated using the MRSH [81] PDFs as input. The simulated events are then re-weighted by the ratio of the DIS cross section using the NLO QCD fit of the HERA I data (H1 2000 PDF [85]) and the MRSH fit. This is equivalent to the usage of the H1 2000 PDF parameterization for the generated MC events.