Speaker
Description
For Run III foreseen to start in 2020, the LHCb experiment will run at an instantaneous luminosity of 2x10^33 cm-2s-1 with a fully software based trigger. A major upgrade of the detector and of data acquisition system will allow having a full readout at the collision rate of 40 MHz. LHCb is planning to have the same Run II strategy with a real-time alignment and calibration procedure and an offline-like quality reconstruction in the latest stage of the software trigger. This strategy includes the alignment of the full tracking system (both for the vertex detector and all the tracking stations) evaluated in few minutes at the beginning of each fill and the complete calibration of the PID sub-detectors for each run that corresponds to a maximum of 1 hour of data taking. The reconstruction used in Run I was optimized to fit the time constrained required by the software trigger: 45 ms and 650 ms for the first and second trigger stage.
In Run III, a total time budget of 13 ms event to take the trigger decision is foreseen.
This implies a large gain in speed in the reconstruction to be achieved. Different approaches are under study to mach this challenging goal, e.g. using parallelism in the CPU and GPU, use of machine learning to veto bad events on an early stage, optimization on different architectures.
The Run II strategy ends to have as output of the second stage of the software trigger the same quality performance as in the “offline” processing. This results in the possibility to perform analyses directly on the output of the trigger without requiring an offline reconstruction. Saving only the interesting information of the selected events reduces significantly the event size down to a factor of 10. Thus an equivalent factor higher rate of signals can be exploited in physics analyses with the same resources. During Run II, the event format has been made more flexible, which has allowed to satisfy more physics analysis requirements. In Run III, LHCb plans to use this same strategy for all the analyses with abundant signals to record the events in a reduced format that can be fed directly to the physics analysis.
The importance and the challenging of this strategy is discussed. We illustrate the operational and physics performance of the real-time alignment and calibration, the overview of the reconstruction and the real-time analysis model in Run2. Plans and different approaches under study for Run III are also presented.
