ATLAS Spam Filter
Posted by Monica Dunford on 15 Nov 2007 at 09:38 am
I’m beginning to think that the phrase ‘combined TileCal/Level-one trigger tests’ is synonymous with the phrase ‘TileCal is experiencing cooling failures’. The TileCal cooling system is designed to keep the electronics at an acceptable operating temperature. If the cooling fails, then the power to the system must be shut down to prevent damage to the electronics. It is not that TileCal experiences cooling failures often, but it seem we always experience cooling failures when doing combined tests with the level-one. And no cooling means no power which means no combined testing for you today. Seriously, you can set your clock by it. The first time it happens, you’re frustrated. The second time you’re laughing. The third time you’re suspicious. The fourth time, you’re convinced you are the butt of some cosmic joke.
I am exaggerating of course. Usually we recover from the failure rather quickly and are able to continue the tests. But the correlation is uncanny….
The level-one trigger is like ATLAS’ spam filter. The beam will collide inside ATLAS approximately 40 million times per second. We can’t possibly store all of that data to disk, nor would we want to. Most of those 40 million per second events aren’t very interesting. Rather I should say, they aren’t AS interesting. They are ‘old physics’, physics we have studied before. We are interested in ‘new physics’, physics we have never seen before. Here is an example:
This is a simulation of what a supersymmetry (SUSY) event might look like in the detector. (Much more complicated then the cosmic data we are taking now.) SUSY, like other theories of new physics, predicts certain types of events that will be produced at the LHC. Typically these events involve lots of particles, with lots of energy, flying everywhere as seen in the picture. The goal of the level-one trigger is to sift through those 40 million events per second, find the interesting one like SUSY and ditch the not-so-interesting ones. And it has to cut the event rate down by a factor of 500, meaning for every event that the level-one accepts, it has rejected 500 events. There are additional layers to the trigger. Once an event passes the level-one, it must also pass the ‘high level triggers’ before being written to disk. The final rate of events being stored for analysis is approximately 100-200 events per second. Imagine that. For every email in your inbox, there are 200,000 deleted as spam.
The Tile Calorimeter as well as the electromagnetic calorimeter (called the Liquid Argon Calorimeter) plays a critical role in the level-one trigger decision. In this picture, TileCal is the orange sections (the upper and lower row of orange) and the liquid argon is the gray sections (plus the two orange sections in the center on the left and right). The level-one makes a decision based on the amount of energy deposited in small regions of the calorimeters. SUSY events are predicted to have very large energy deposits. The purpose of combined tests between TileCal and the level-one is to calibrate the electronics’ signal in voltage and convert that to the amount of energy deposited in the calorimeter. If the electronics aren’t working properly or the calibration is incorrect, the trigger might delete as spam the very events that you are interested in studying. The calibration is a long process and will take many months but it is crucial to get right. Because once the events are rejected, we can’t get them back. There will be more tests on Monday. The cooling gods allowing of course.
[...] Dunford, at the US/LHC Blog, has a great metaphor: thinking of the “trigger” in a particle detector as a spam [...]
SUSY, like other theories of new physics, predicts certain types of events that will be produced at the LHC.
Slightly offtopic question: Are there any possible tests or observations at the LHC which could provide evidence for SUSY, but which do not actually involve the production of a superpartner? Asked another way, is it possible that the LHC could prove SUSY even under the scenario that all the superpartners turn out to be heavier than the LHC range?
Slightly more on topic question: Some of the things I have read about particle colliders strongly give the impression that data analysis with these things is done not so much by taking individual events seriously, but rather by measuring a large number of events, and then statistically analyzing them to see how often certain kinds of events occur relative to others. If this is the case, then does this become more difficult when you are filtering entire kinds of events in this way? Is there any worry about whether the set of events recorded is statistically representative, or whether the filters are introducing some kind of bias? Or am I just confused about how statistics are used in collider data analysis, or is this for some reason not a problem for the LHC…?
Yes, it might be possible to provide some indication of new physics even if the new particles are heavier than the LHC energies. This can be done by trying to accurately measure certain branching ratios in decays and then compare the measurement to the Standard Model theoretical prediction. This is referred to as a precision measurement experiment (compared to a rare process experiment where you are searching for a tiny number of signal events in a huge number of background events) For the precision measurement experiment, heavy particles can affect the branching ratio in the higher orders of the theoretical calculation, coming into play via ‘loop-diagrams’ (see http://en.wikipedia.org/wiki/One-loop_Feynman_diagram). A precision measurement experiment is very, very hard to do, from the perspective of both the experimentalist doing the measurement and theorist making the calculation. If there was a significant experimental and theoretical disagreement, we might be able claim ‘discovery’ of new physics but we would not claim discovery of seeing for example a SUSY chargino until we had direct evidence of that particle (actually producing the heavy particle and observing its decay). The measurement of the muon magnetic moment at the E821 experiment at Brookhaven (http://www.g-2.bnl.gov/) is an excellent example of a precision measurement experiment. And here for example (http://arxiv.org/pdf/0710.2429) is an article discussing how SUSY can resolve the disagreement.
As for your comments about any bias in the trigger, you are absolutely correct. The trigger can have two major effects on any analysis. First, you need to understand the trigger’s overall efficiency. For example there might have been 1000 SUSY events produced in the detector but the trigger is only 90% efficient in accepting those events, therefore you only observe 900. Second, you need to understand if there is any bias introduced by the trigger. In other words, whether the trigger is systematically not accepting certain types of SUSY events. We can study these efficiency and biases using a very detailed Monte Carlo simulation of the trigger and detector. And we validate this simulation with the actual data using well-measured physics channels such as W, Z decays and QCD processes. As most SUSY models predict very energetic and multiple jets of particles, the trigger is expected to be very efficient and non-biased. But these are statements that we have to confirm before we can hope to publish.
Very interesting, thank you!
[...] produced in it. Out of the 40 million collision events that would occur, a lot of work is done to filter out “boring” and “well-known” events which our current theories of physics can already explain quite well, so that data for [...]