Muon calibration

31 March 2008


The architecture of MDT Calibration begins with the transfer of the events collected by the Level 2 Muon Trigger in the special Muon Calibration stream from the Point1 network to the three Calibration Centres in Ann Arbor, Rome and Munich. The Calibration procedure will also need to access information produced by the MDT DCS and by the Alignment System, which will be stored in COOL folders in the ATLAS Online Conditions Database (ATONR) at Point 1, and then automatically transferred to the ATLAS Offline Conditions Database ATLR. From here, the information will be replicated to the Calibration Centres using the ORACLE Streams technology. The Calibration Stream events will be processed at the local sites, and every step of the computation will be stored in a local ORACLE DB. The three local Calibration Databases will be replicated back at CERN, and merged in a unique DB. The newly computed MDT calibration constants will then be written on ATLR for offline use (e.g., by reconstruction) and propagated to the Online DB for use in the HLT chain.

Monitored Drift Tubes (MDTs) have been selected more than 10 years ago for the precision measurement of muon trajectories in most of the ATLAS Muon Spectrometer. Since then, in parallel to the titanic task of building 1200 MDT chambers within the tight mechanical specifications (and in time for ATLAS completion!), a team has been devoting its time to understand how to best calibrate these detectors.

It has been proven that MDTs can be reliably calibrated down to the requested precision (~20µm) using the muon tracks crossing the chambers.  The procedure, known under the name of autocalibration, exploits the geometrical arrangement of the tubes in staggered layers, which allows the computation of the residual of a given hit with respect to a track segment built using the other hits in the same chamber.

Finding the appropriate scaling law from the autocalibration of a single chamber to the application in the complex ATLAS environment was not straightforward. A considerable number of tracks is needed for a reliable calibration, and all environmental parameters (temperature, gas properties, magnetic field) must be taken into account in the computation. But the Muon Spectrometer spans an impressive volume: so large in fact that each one of the 350,000 MDTs sees on average only  (1/20000)th of the solid angle - and of the muons reaching the external stations.

The tight selection performed by the three levels of trigger does not make the task any easier, as only 40 Hz of muons are accepted and collected. At this rate, the MDT illumination in the ATLAS stream of recorded data would be too low to allow a reasonably fast calibration: we cannot afford to wait for days or weeks until we accumulate a sizeable sample, since we would not be able to follow possible time variations of the system. Last but not least, should we try to use the muon tracks prior to the level-2 trigger selection, we would simply be flooded by background hits and just collect unusable data.

Although it sounded much like a “mission impossible”, several MDT groups stood up to the challenge, and started endless brainstorming sessions at CERN, during a dedicated workshop in December 2004, in the cozy rooms of the Ringberg Castle in February 2005 oblivious to the snow storming outside, and on the sunny beach of Cetraro during the 10th ATLAS Muon Week, in July 2005. These were the key events where the calibration model (described in a recent communication submitted as public note) was shaped. Experts from the ATLAS Trigger, DAQ and Database groups, as well as from the IT Database Strategy and Deployment team participated actively to the design of the whole architecture, allowing us to proceed quickly on the chosen path.

The model is based upon a dedicated event source, produced directly by the second level muon trigger.  The level-2 muon trigger finds muon track candidates in the Region of Interest identified by the level-1: this information is normally used in conjunction with the findings of the other second level triggers in the standard ATLAS Data Flow. In order to build the special Muon Calibration Stream, all candidate muon tracks are collected in a specific stream. In addition, only a limited amount of information from the muon precision and trigger chambers is recorded, just what is strictly relevant for the reconstruction of muon tracks in the selected Region of Interest. The Muon Calibration Stream events are thus much smaller than typical ATLAS events, by at least a factor of 1500, so they can be extracted and recorded at a much higher rate (up to 12 kHz). Such a statistics makes it possible to calibrate the MDT detectors on a day by day basis and to update the calibration constants, whenever needed, within the 24 hours requested for the first reconstruction pass at Tier0.

The statistics accumulated by the Calibration Stream is huge (up to 100 million muons per day) and requires a non negligible but manageable amount of CPU; also, the expertise of the people who developed the calibration techniques is vital to ensure a smooth processing. However, the data size of the single events is tiny, thus implying a small bandwidth for the transfer of these samples over the network. This made it possible to process the Calibration Stream not directly at CERN, but at three specific Tier2s, called Calibration Centres: these are located in Ann Arbor, Munich and Rome, and the processing of the data will be under the direct control of groups experienced in the MDT calibration.

At the Calibration Centres, a dedicated software developed in the ATHENA framework is run to compute the calibration parameters. All the intermediate results of the calibration of the 350,000 MDTs need to be archived safely, while the final parameters have to be propagated back to the Conditions Database at CERN. ORACLETM is used both for archiving and stream replication across databases.

Some of the members of the MDT Calibration Community, pictured during a break in the last Muon Week - tired but satisfied to see the first results of their hard work.

The MDT calibration is a project involving expertise ranging from trigger to data acquisition, from data transfer to data processing, from C++ coding to database design and administration, from reconstruction to detector performance. Three years after its conception, the calibration model is finally being tested. First data were extracted from the second level trigger already in December 2007 but during the March 2008 Milestone Week 6 the Muon Calibration Stream was regularly extracted in all runs and distributed through the GRID to the Calibration Centres, where data arrival automatically triggered their processing. Nearly 15 million muon triggered cosmic events were successfully processed with only a few hours delay with respect to the data taking. Now that the input data flow has been validated, the Full Dress Rehearsal 2 scheduled for the next months will be devoted to the test of the output data flow with the transmission of calibration parameters from local ORACLE databases to a CERN central one, interfaced to the COOL Conditions Database.

Many people have been bringing their expertise and grain of salt in the many hours of work and heated discussions that made this project come true. It is a real pleasure to see that our long debated scheme is going smoothly through the process of testing all the individual ingredients – the Calibration stream extraction, the transfer to the Calibration Centres, data analysis at the remote farm, storage into the local ORACLE Databases and into the ATLAS Conditions Database. The forthcoming months will allow us to close the loop, from the extraction of the events, to the processing at the Calibration Sites, to the availability of  calibration constants in the Central COOL Conditions Database, and to give the final proof that we have built a fast, solid and reliable scheme  to allow ATLAS to get the most out of its Muon Spectrometer.

Domizia Orestano

Universita' di Roma Tre & INFN


Manuela Cirilli

University of Michigan