The goal of our shifts is to monitor production and transfer of ATLAS data. Let’s see, what this data IS.
Proton-antiproton collisions produce a number of particles. Some of these particles exist only for a very short time, decaying into other ones, with smaller mass and more stable.
For example, a pair of top quarks may be produced. Top quark, the heaviest quark of the Standard Model, has very short lifetime and almost immediately decays, producing much lighter bottom quark and W boson. W boson then decays into pair of light quarks or lepton-neutrino pair. These final particles (quarks, lepton and neutrino) should be somehow detected.
These particles pass through different detectors of ATLAS (tracker, calorimeters, etc…) interact with them, producing various responses, for example light flashes in scintillator of the tile calorimeter. Quarks produce cone-like “jets” that make traces in tracker and both electromagnetic and hadron calorimeter. Electrons leave much thinner trace in tracker and EM calorimeter. Muons can pass through larger amounts of matter, so they also produce signal in the most outer part of ATLAS, unreachable for electrons and jets. Neutrinos almost no interact with matter of the detector, and their presence, and further, energy and direction can be deduced from studying energy misbalance of the event.
Detector response then converted into electric pulses, digitized and passed to the DAQ and Trigger system. Trigger quickly analyzes information, coming from different detectors, trying to determine which particles may produce such combination of signals. If result of the analysis satisfies pre-defined requirements, information on the “Event” is stored for further analysis.
For example, if we interested in events with pair of top quarks, trigger can select events with two high-energy jets from b-quarks, two “softer” jets from lightest quark, electron or muon and large “missing” energy taken by the escaping neutrino.
Of course, research program of ATLAS is quite extensive, and physicists are interested in many different processes with different particles, having its typical “picture” in detector (called “Signature”). That’s why trigger simultaneously looks for many different “patterns”, called “menu”.
In the end, information on up to 200 selected events per second is being stored, resulting in up to one Petabyte (million gigabytes) of data per year. This vast amount of data then being reproduced, during this process events are being reconstructed and data files in format ESD (Event Summary Data) stored on storage sites in different parts of the globe. This data will be used further by physicists for their analysis. More compact data files in format AOD (Analysis Object Data), which is a summary of the reconstructed event, and contains sufficient information for common analyses, are also produced, with event size up to 100k instead of 500k for ESD.
Slide 5 So, the typical way from proton collision to reconstructed event is the following: particles produced in collision produce signals in ATLAS detectors, these signals are digitized and then parameters of particles are being reconstructed, giving us picture of physical process, maybe the one that we want to study
Of course, in real life everything is much more complicated, and we get parameters of the particles with significance errors, furthermore, furthermore during reconstruction different particles may be mixed up, or two particles passing close to each other may be considered as one.
And above all, collisions not started yet, so why do we have to bother now?
Before we start getting real data from the detector, we should be able to “digest” them. We need to develop algorithms of particle reconstruction, criteria for event selection and triggering. Also we should check efficiency of such algorithms.
This is usually done by using Monte-Carlo simulation.
Slide 6 The Monte-Carlo method has been developed in forties and was named after famous center of gambling, due to its random nature. Legends say that at early time scientists used gambling roulettes as source of random numbers.
On the picture you can see mechanical random number generator actually used by Russian physicists for calculating parameters of the reactor in the times where Bill Gates wasn’t even born, and the computers were monstrous novelties.
Monte-Carlo method uses repeated random sampling of input parameters with further computation of result due to known deterministic algorithms. Like when you play monopoly, the number of movements is determined by random number from the dice, but the result of you move determined by the filed you arrive.
The same come with particles and detectors. In proton-proton collisions particles appears randomly, with probability depending on the type of process. These particles may have random energies and initial directions of movement. But for particular type of the particle with particular energy and direction, we can to some extent predict its path and result of interaction with detector.
Simulation on ATLAS passes as following:
Special program-generator creates series of possible scenarios of particle collision and further particle conversions. For each even in the series it produces set of final particles with information where the particle move and what energy it has. Normally sets of events of certain type are being generated, for example, our favorite top quark pair production. Events also can be restricted to some “modes” of the decay.
There are number of different generator programs (PYTHIA, JIMMY), suitable for different type of processes, both common and exotic. You will further notice these names in the name of data files.
Generated events are then passed through a GEANT4 Simulation of the ATLAS Detector to produce GEANT4 Hits i.e. a record of where each particle traversed the detector and how much energy etc was deposited.
Next stage in data processing is Digitization, where the GEANT4 Hits from the simulation are “converted” into response of the detector to produce Digits, such as times and voltages, as produced in the Raw Data from the real detector.
At that point we have imitation of the data from the detector that can be feed to reconstruction program to produce ESDs and AODs similar to ones that we will use for real data. I the reconstruction process we get from the raw data Digits, such as times and voltages - tracks and energy deposits and further - particles.
Slide 7 Slide 8 Slide 6 Monte-Carlo data files contain extra information about “true” particles (from the first stage, the generation), so one can check efficiency of reconstruction algorithms by comparing reconstructed picture of the event with the one that was generated initially.
Of course, the real data will not contain this “truth” information.
Because of the time this process takes, especially the Simulation stage, it is unlikely that most users will produce many events themselves but will rely on centrally produced events.. There is a shortcut around the Full Chain for impatient users called Atlfast which provides a fast simulation of the whole chain by taking the generated events and smearing them to produce AOD directly. Atlfast can in fact take input from any of the event generator, simulation, digitization, or Event Summary Data (ESD) files.
Slide 9 As it was mentioned before, running Full Chain takes tremendous amount of time and computing resources, so production of Monte-Carlo data is done centrally using ATLAS distributed production system, using thousands computers around the world, combined into GRID network. This network will be described in one of the next lectures.
Slide 10 Monitoring of a Monte-Carlo production and transfer of data between parts of this network are the main goals of the ADCoS shifts. On the further lectures we will see how this production and data transfer is organized and what the actual flow of work of the shifter is.