The text below is maybe hard to read as it is taken off context; however, let me at least spend one
word about the landscape: we are talking about the CDF experiment, a proton-antiproton experiment that worked at the Fermilab laboratories near Chicago from 1985 to 2012. The collaboration is still active, by the way, but the data has ceased to pour in...
* * *
The online data acquisition system of CDF worked by selecting only the most significant collisions among the hundreds of thousand that took place every second in the core of the detector, and for events with jets, "most significant" is a synonym of "with jets of highest transverse energy (ET)." So you might imagine that the data one needed were those coming from a trigger selecting events with jets of high ET. "Below 100 GeV? Off to the trash bin"; "Above 100 GeV? Please follow this way to the data tape."
The trigger of CDF was a quite complicated thing because it made all efforts to not burn any bridges, as the collection of a single dataset with jets above a fixed threshold would have undoubtedly done. Instead, triggers with looser energy thresholds were also at work, to avoid entirely losing the information about those less dramatic but still useful collisions: there were jet triggers with thresholds of 70, 50, and even only 20 GeV, a very loose requirement.
The resulting datasets allowed CDF to perform calibrations, study backgrounds, and improve the detector modeling. But they were prescaled: only one every 10, or 100, or even 1000 events fulfilling the trigger requirements was actually sent to the output data stream. To explain how this worked we may consider a numerical example not too far off the real situation.
Let us imagine that you study the average rate of jet events at a given instantaneous luminosity, and you observe rates of 2000, 100, 20, and 5 events per second as the ET threshold is set at 20, 50, 70, and 100 GeV respectively. You are interested in jet events and you would be glad to get all of them, yet a frowning Hans Jensen, the thick-bearded leader of the Trigger group, tells you that you are allowed to store no more than 10 jet events per second. That is because other triggers, that select precious events containing electrons, muons, or other striking signatures, must also have a chance to write those events to tape, and the total output capability of the data acquisition system is of no more than 50 events a second!
So you decide to take all of the five-per second events with jets above 100 GeV, which are by far the most interesting ones. At that point your bandwidth budget allows you to record five more events per second, while events with jets still requiring a decision pour in at the terrifying rate of 2000 per second (a jet passing the 50, 70, or 100 GeV threshold of course also passes the 20-GeV one, so 2000 per second is both the rate of the 20-GeV trigger and the total jet rate).
What to do? The solution is that you use a prescale, a device which randomly saves one every N events passing a given trigger threshold, and dumps all others. You can for instance set the prescale factor of the Jet-70 trigger at one-in-ten (N=10): this will ensure that a rate of only two events per second will be written to tape, out of the 20 that pass the 70 GeV threshold. You can also set the prescale factor of the Jet-50 trigger at N=100, or one-in-a-hundred, so out of the 100 events per second with a jet above 50 GeV only one gets written to tape; and finally, you can set the prescale factor of the Jet-20 trigger at N=1000, for one-in-a-thousand, which ensures that of the 2000 such events per second only two get actually stored.
Let us take stock: every second the above arrangement of trigger thresholds and prescale factors yields on average five 100-GeV events, two 70-GeV events, one 50-GeV event, and two 20-GeV events, for a total of ten per second. There you have a recipe for using effectively the bandwidth granted by Jensen, while ensuring that you collect all the highest-energy jet events, as well as a representative sample of jets with smaller energies. In a 100,000 seconds store (a typical one) you get 200,000 events passing the Jet-20 trigger, 100,000 passing the Jet-50 trigger, and 200,000 passing the Jet-70 trigger. Those datasets are large enough to allow for the study of the characteristics of jets in each energy interval with a similar statistical power.
Note that you could have also gotten away with setting a single prescale factor of 400 on jets above the 20-GeV threshold and ignored altogether the Jet-50 and Jet-70 triggers, arguing that the Jet-20 selection automatically also collects jets that pass those higher thresholds. This would work in terms of bandwidth, as the initial rate of 2000 events per second prescaled by a factor of 400 would still yield the five events per second which you are allowed to store in addition to the five Jet-100 ones. However, the result of such alternative approach would be quite unbalanced: in the course of the 100,000 seconds store the collected sample of jets with ET above 20 GeV would be half a million, but of these only 25,000 and 5,000 would have jets above 50 and 70 GeV, respectively. Such a choice would enable you to measure very well the properties of low-energy jet events, but any study involving jets in the 50-100 GeV region would then suffer from very low statistics.
* * *
It is interesting to note that trigger prescales are used also at the LHC experiments nowadays. The random selection of a small fraction of the produced collisions is still a good way to create useful data samples for calibration studies!
word about the landscape: we are talking about the CDF experiment, a proton-antiproton experiment that worked at the Fermilab laboratories near Chicago from 1985 to 2012. The collaboration is still active, by the way, but the data has ceased to pour in...
* * *
The online data acquisition system of CDF worked by selecting only the most significant collisions among the hundreds of thousand that took place every second in the core of the detector, and for events with jets, "most significant" is a synonym of "with jets of highest transverse energy (ET)." So you might imagine that the data one needed were those coming from a trigger selecting events with jets of high ET. "Below 100 GeV? Off to the trash bin"; "Above 100 GeV? Please follow this way to the data tape."
The trigger of CDF was a quite complicated thing because it made all efforts to not burn any bridges, as the collection of a single dataset with jets above a fixed threshold would have undoubtedly done. Instead, triggers with looser energy thresholds were also at work, to avoid entirely losing the information about those less dramatic but still useful collisions: there were jet triggers with thresholds of 70, 50, and even only 20 GeV, a very loose requirement.
The resulting datasets allowed CDF to perform calibrations, study backgrounds, and improve the detector modeling. But they were prescaled: only one every 10, or 100, or even 1000 events fulfilling the trigger requirements was actually sent to the output data stream. To explain how this worked we may consider a numerical example not too far off the real situation.
Let us imagine that you study the average rate of jet events at a given instantaneous luminosity, and you observe rates of 2000, 100, 20, and 5 events per second as the ET threshold is set at 20, 50, 70, and 100 GeV respectively. You are interested in jet events and you would be glad to get all of them, yet a frowning Hans Jensen, the thick-bearded leader of the Trigger group, tells you that you are allowed to store no more than 10 jet events per second. That is because other triggers, that select precious events containing electrons, muons, or other striking signatures, must also have a chance to write those events to tape, and the total output capability of the data acquisition system is of no more than 50 events a second!
So you decide to take all of the five-per second events with jets above 100 GeV, which are by far the most interesting ones. At that point your bandwidth budget allows you to record five more events per second, while events with jets still requiring a decision pour in at the terrifying rate of 2000 per second (a jet passing the 50, 70, or 100 GeV threshold of course also passes the 20-GeV one, so 2000 per second is both the rate of the 20-GeV trigger and the total jet rate).
What to do? The solution is that you use a prescale, a device which randomly saves one every N events passing a given trigger threshold, and dumps all others. You can for instance set the prescale factor of the Jet-70 trigger at one-in-ten (N=10): this will ensure that a rate of only two events per second will be written to tape, out of the 20 that pass the 70 GeV threshold. You can also set the prescale factor of the Jet-50 trigger at N=100, or one-in-a-hundred, so out of the 100 events per second with a jet above 50 GeV only one gets written to tape; and finally, you can set the prescale factor of the Jet-20 trigger at N=1000, for one-in-a-thousand, which ensures that of the 2000 such events per second only two get actually stored.
Let us take stock: every second the above arrangement of trigger thresholds and prescale factors yields on average five 100-GeV events, two 70-GeV events, one 50-GeV event, and two 20-GeV events, for a total of ten per second. There you have a recipe for using effectively the bandwidth granted by Jensen, while ensuring that you collect all the highest-energy jet events, as well as a representative sample of jets with smaller energies. In a 100,000 seconds store (a typical one) you get 200,000 events passing the Jet-20 trigger, 100,000 passing the Jet-50 trigger, and 200,000 passing the Jet-70 trigger. Those datasets are large enough to allow for the study of the characteristics of jets in each energy interval with a similar statistical power.
Note that you could have also gotten away with setting a single prescale factor of 400 on jets above the 20-GeV threshold and ignored altogether the Jet-50 and Jet-70 triggers, arguing that the Jet-20 selection automatically also collects jets that pass those higher thresholds. This would work in terms of bandwidth, as the initial rate of 2000 events per second prescaled by a factor of 400 would still yield the five events per second which you are allowed to store in addition to the five Jet-100 ones. However, the result of such alternative approach would be quite unbalanced: in the course of the 100,000 seconds store the collected sample of jets with ET above 20 GeV would be half a million, but of these only 25,000 and 5,000 would have jets above 50 and 70 GeV, respectively. Such a choice would enable you to measure very well the properties of low-energy jet events, but any study involving jets in the 50-100 GeV region would then suffer from very low statistics.
* * *
It is interesting to note that trigger prescales are used also at the LHC experiments nowadays. The random selection of a small fraction of the produced collisions is still a good way to create useful data samples for calibration studies!
Comments