In his popsci book, "Particle at the end of the universe", Sean Carroll says that the LHC, due to it's sheer information gathering capability, necessarily needs to completely discard most of the data that it gathers.

I was wondering how this was done, separating the "interesting" events from the rest.

Carroll says that between hardware and software techniques, only several hundred events per second, out of the many millions produced per second, are kept on record for full analysis.

Is it a pattern matching algorithm that is used to analyse the data, in which case do the people involved, both theorists and experimentalists, use this as a guide to what "odd" events would look like?

I'm sure I am second guessing the combined years of experience of thousands of theorists and experimentalists here. Also I certainly do not have the background to ask the question in specific terms.

But in general, if an exotic event produces a signature that is really new to us, such as dark matter or a member of a fourth generation of particles, how would we be sure we have not missed it, if we don't know what to look for in the first place?

Of course, the answer could be, say due to the energy output of the LHC, we have limits as to what is possible and we do actually know what to expect in any particular range. I just wonder is there a chance that we will miss something as important as the Higgs?


1 Answer 1


Events in high energy physics detectors that can't produce useful data, mostly because they are the result of soft scattering events, are discarded by multiple layers of trigger circuits. What these circuits do is prescribed by so called trigger menus, which are based on theoretical predictions about a large number of known and hypothetical physics event types.

As you correctly observe, if an exotic event should occur that is markedly different from any one of the event types that are covered by the trigger menus, then it may get lost in the trigger. To guard against a complete loss of such events there are pre-scaled trigger menus (like one in thousand, ten thousand or a million bunch crossings etc.) that let data stream trough that is less biased or completely unbiased by assumptions about event structure. One does have to make a tradeoff between post-trigger event rate and the number of these pre-scaled events, though, otherwise the data acquisition, storage and especially the data analysis pipelines will be swamped with most likely useless data. In the end this means that sufficiently rare exotic events may go unnoticed, after all.

If you want to see how complex all of this is in reality, I would suggest to take a peak at e.g. this ATLAS technical design report cern.ch/atlas-proj-hltdaqdcs-tdr/tdr-v1-r4/PDF/TDR-2up.pdf. That's just one of the top level documents. To get to know the system in detail you would have to read hundreds of papers and internal technical reports, of course.

  • $\begingroup$ Yes, and life is short. I know this question will have been considered by thems that know FAR more than I do, just that Carroll skipped over what to me was more deserving of a longer treatment that might have taught me some particle physics. That's popsci for you though, sometimes it makes you think for yourself and sometimes it leaves you frustrated, with just a glimmer of an answer. Thanks for that $\endgroup$
    – user81619
    Commented May 29, 2015 at 8:17
  • $\begingroup$ @AcidJazz: I can tell you that the design of these trigger menus was a major deal in the early and mid 1990s (based, of course, on decades of previous experience and theory), even while the detectors were still in the design phase. It's really not even possible to design a functional detector without understanding event structure, first. Thankfully every experiment can build on the physics learned from the previous ones, so most of this is actually very well established theory... but the amount of intellectual labor that goes into these calculations and simulations is absolutely stunning. $\endgroup$
    – CuriousOne
    Commented May 29, 2015 at 8:24
  • 1
    $\begingroup$ @AcidJazz In addition to CouriousOnes comments, consideration of what should be in the triggers, how to structure the triggering decisions and how many minimum- and no-bias events should be taken is subject to on-going consideration. moreover, parts of the trigger system run on FPGAs so substantial changes of kind are possible. The system in use these days are mindbogglingly flexible compared to the ones I leaned on less than two decades ago. $\endgroup$ Commented May 29, 2015 at 18:47
  • $\begingroup$ @dmckee It's a long, long way from Wilson and his cloud chamber photos. I asked how the triggers operated but to be honest, it seems like it would take the entire site to cover that process alone. It must be more than a little frustrating , although necessary, to let so much data go, knowing that one record might, just might , have something really new to tell us. $\endgroup$
    – user81619
    Commented May 29, 2015 at 19:36
  • $\begingroup$ @AcidJazz: dmckee gave excellent hints. Modern triggers are programmable trough tables, reconfigurable trough programmable hardware designs and some levels are programs running on server hardware. Triggers are evolving and are being upgraded. The interesting thing that happened at LHC, however, was that the initial estimates for how many events had to be processed were about ten times what they are processing, right now. Partly that was due to overzealous physics estimates. Even with 1995 technology we could have processed far more than we do today, but it's not deemed necessary. $\endgroup$
    – CuriousOne
    Commented May 29, 2015 at 20:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.