Why a collider cannot keep every collision

Several views of the same collision

The layers around a collision point make different observations. Inner tracking detectors record positions along charged-particle paths. A magnetic field bends those paths, so calibrated positions and a measured field provide a momentum scale. Several hits must agree with one trajectory before they support that interpretation.

Farther out, calorimeters absorb energy and turn ionization or light in their materials into electrical signals. Electrons and photons often make concentrated electromagnetic showers. Hadrons produce more complicated deposits. Outer muon stations look for particles that have crossed substantial material and still leave hits. Together, the layers identify patterns worth a closer look.

The electronics initially supplies channel addresses, digital amplitudes and timestamps. A large number can reflect a stronger signal, a different gain or a shifted pedestal. Known inputs, blank signals and independent check points establish the scale. Aligning the channel times helps determine which hits belong to the same passage.

Each observation has a limit. A very slight bend can become indistinguishable from a straight path at finite position resolution. Near the noise level, a signal can remain unresolved on either side of a threshold. The available evidence sets the useful detail of the selection that follows.

Sources: [1][2][3]

Data needs somewhere to wait

The first trigger level examines relatively coarse information, such as a concentration of calorimeter energy or a pattern across outer detector stations. It uses features that can be calculated promptly. Accepted candidates proceed to software processing, where finer positions, signal shapes and consistency checks become available.

A decision takes time. Front-end electronics holds the corresponding data while waiting for instructions to continue the readout. That temporary store has a finite number of places and a finite retention time. A decision can arrive after the useful data has lost its place.

The next wait is for a processor. An event with many hits or a complicated reconstruction requires more work. An accepted result can then wait at the output buffer. A large packet occupies the link longer than a small packet. The processor limits how quickly an event can be examined; the link limits how quickly its bytes can leave.

Imagine two accepted packets reaching one transmitter. The first occupies it until its last byte leaves. The second needs its own waiting place during that interval. Making the first packet smaller can let both finish, even though neither particle signal changed.

Completion therefore matters as much as acceptance. When an observation window closes, an event may still be processing or only partly transmitted. Keeping that pending state separates unfinished work from a definite loss. It also lets an operator account for every arrived event and every admitted byte.

Sources: [1][2]

Selection leaves a shape in the sample

A higher energy or momentum threshold tends to favor harder events. Looking only at the saved sample can confuse that preference with the underlying collision distribution. An independent control selection gives the trigger under study an opportunity to fail on known probes.

Experiments can use another selection to identify a suitable probe sample, then count how often a probe satisfies the trigger being tested. The resulting fraction depends on how those probes were chosen, which momenta and directions they cover, and which readout and calibration were used. A finite sample also gives a statistical interval. Zero observed passes still leaves a nonzero upper limit on the probability.

Resource competition adds another selection. On fixed signals, a tighter threshold admits fewer individual candidates. It can also free a waiting place, allowing a different packet that was previously blocked to finish. The final saved set depends on the full arrival sequence and the available resources.

Replaying one fixed readout through several programmes separates these effects. Small changes reveal which candidates fail a signal requirement and which lose a place while waiting. Independent probes then test the selection within their stated preparation. The sample delivered to offline analysis has a path that can be examined, measured and checked.

Sources: [2][4][5]