A firehose you cannot drink from
In the previous guide you saw how an onion of detector layers turns one collision into a rich snapshot — a few megabytes describing every track, every energy deposit, every clue. That is wonderful for one collision. The trouble is the sheer number of them. At the heart of a machine like the LHC, bunches of protons cross through each other about 40 million times every second, and at each crossing not one but dozens of proton pairs actually collide. The detector is, in effect, being asked to photograph the inside of an atom tens of millions of times a second, forever.
Do the brutal arithmetic. Forty million crossings a second, each producing roughly a megabyte of raw detector data, is something like forty terabytes per second. Run that for one day and you would need to write down more data than the entire written output of humanity, over and over. No disk array, no network, no budget on the planet can swallow that firehose. And here is the cruel part: you cannot pause the beams to catch up. The protons keep crossing whether you are ready or not, so any collision you fail to record in the moment it happens is gone for good. This single fact — more data than you can ever keep, arriving faster than you can ever store it — is what forces the existence of everything in this guide.
Why almost everything is worth throwing away
It would feel reckless to discard collisions — until you realise that nearly all of them are dull. The reason is the cross-section you met earlier: it is the effective target size of a process, and the bigger it is, the more often that process happens. The trouble is that the boring processes have enormous cross-sections while the prizes have minuscule ones. When two protons graze each other they almost always just spray out a mess of low-energy hadron jets through the ordinary strong force — events we already understand completely and have recorded billions of times. Making a Higgs boson, by contrast, has a cross-section smaller by a factor of roughly a few billion. The interesting physics is a vanishingly thin seam running through a mountain of the mundane.
So the goal is not to keep collisions; it is to keep needles and let the hay fall through your fingers. The needles betray themselves by carrying signatures the common junk almost never has: a stiff, high-momentum electron or muon flying out sideways from the beam, a pair of very energetic photons, a large amount of missing transverse energy hinting that something invisible escaped. None of these is a guarantee of new physics — they are just the rare features that rare processes tend to leave. The trigger's whole job is to recognise those features in the raw, unreconstructed data, fast enough to act before the next crossing overwrites everything.
The trigger is a ladder, not a single gate
You cannot fully reconstruct an event in the time between two crossings — there is barely time for light to cross the detector. So the trigger works in tiers, each one slower and smarter than the last, each throwing away most of what the previous tier passed along. Think of it as a series of ever-finer sieves: the first is crude but instant and shakes out the obvious junk, the last is nearly a full analysis but only ever has to look at a trickle. The reduction is staggering, and it is the engineering miracle that makes a collider possible at all.
- Level 1 (hardware, microseconds): custom electronics sitting right on the detector look only at coarse, fast signals — a big lump of energy in the calorimeter, a stiff track in the muon chambers. No software, no reconstruction; just dedicated circuits asking simple yes/no questions. This stage alone slashes the rate from 40 million down to about 100 thousand events a second.
- Higher level (software, milliseconds): a vast farm of ordinary computers now has time to run a fast, partial reconstruction on each surviving event — building rough tracks, matching an energy deposit to a track to confirm it is really an electron. With more time and more information, it discards most of what Level 1 let through, dropping the rate to perhaps a thousand events a second.
- Write to disk: only those final ~1000 events per second are deemed worth keeping. They flow into the trigger and data-acquisition system, get written to permanent storage, and become the dataset that physics analysis will actually live on. Everything else — more than 99.99 percent of all collisions — is gone, never having existed for science.
Notice the strategy in that ladder: spend cheap, fast decisions on the easy rejections, and save the expensive, careful decisions for the few events that already look promising. It is the same logic as an emergency room triaging patients — the word "trigger" and the word "triage" share that spirit. By the bottom of the ladder, a forty-terabyte-per-second torrent has been tamed into a stream a hundred thousand times thinner, and yet, if the menu of triggers was chosen well, almost every needle has been kept.
Pile-up: many collisions, one snapshot
There is a second twist that makes triggering harder than mere speed. To get enough rare collisions, you have to pack the beams so densely that each bunch crossing produces not one collision but many — often fifty or more proton pairs colliding within a fraction of a millimetre and a fraction of a nanosecond of each other. The detector cannot tell them apart in time; it records them as a single, hopelessly overlapping snapshot. This jumble of simultaneous collisions is called pile-up, and it is the price you pay for high collision rate.
Pile-up is corrosive. The one interesting collision you care about is buried under dozens of dull ones happening at the same instant, all dumping their own tracks and energy into the same picture. That stray energy smears your measurements and, worst of all, can fake a trigger signature — fifty boring collisions piled together can add up to a misleading lump of energy that looks, briefly, like one exciting event. Disentangling which track came from which collision point is a major part of why the higher trigger tiers and the later analysis are so much work. It is also why you want each collision to be as cleanly separated in space as the machine can manage.
The arithmetic of what survives
Let us tie the numbers together, because the arithmetic is the whole point. How many of a given process you collect over a year is set by two factors: its cross-section (how often nature makes it) and the integrated luminosity (how much total collision exposure the machine delivered, measured in inverse femtobarns). Multiply them and you get a count of events produced. But you only keep the ones your trigger fired on, so the count you can actually analyse is that product times the trigger's efficiency for your signature. Every percent of trigger efficiency you lose is a percent of your hard-won discovery you will never see.
events you can analyse = cross-section x integrated luminosity x trigger efficiency fewer events kept -> larger statistical uncertainty -> harder to claim a discovery
This is also why a thrown-away collision is not quite as wasteful as it first sounds. The events you discard are overwhelmingly the high-cross-section common processes you already have in billions; keeping more of them would add almost nothing. What you fight to keep is the rare seam. The art of trigger design is to set each threshold — how stiff a muon, how big an energy lump — just low enough to catch the signal you want, but high enough to reject the flood, so that your precious recording bandwidth is spent on needles and not on hay. Get that balance wrong in either direction and you either drown in data or miss the discovery.