One number sets the whole machine. The rarest channel worth seeing fixes the luminosity, and everything difficult about the LHC follows from that single requirement.
🎯 Why this matters
Every hard problem in these detectors descends from a physics decision rather than an engineering ambition. Pile-up exists because somebody wanted a channel with a branching ratio of 10⁻⁴, and twenty-five nanoseconds is the bill.§9.13 left a gap between 115 and 145 GeV that neither LEP nor the Tevatron could close, and identified what closing it needs: the rare decay channels, at and at . This section is the machine and the two detectors built to do that, and almost every number in it can be derived from that requirement.
The machine, from one constraint
The LHC had to fit the existing 27 km LEP tunnel — a civil-engineering investment the CERN member states had already made. Everything else follows.
| constraint↕ | what it forces↕ | the number↕ |
|---|---|---|
| the tunnel is fixed | the orbit radius is fixed, so raising the energy means raising the dipole field — superconductor technology pushed beyond the state of the art | T, TeV, TeV |
| T | the coils must have zero resistance at 11 850 A | cooled to 1.9 K by 120 t of liquid helium over 27 km, with 40 000 leak-tight seals |
| luminosity must be enormous | antiprotons cannot be made in the required intensity, so it is pp, not — and two same-charge beams need opposite field directions in the two pipes | 1232 dipoles, 14 m each, with the twin-bore magnetic design of Fig. 9.45 |
| the beams must be dense | transverse size squeezed at the crossing points | 200–300 μm around the ring, 16 μm at the intersections |
| and they carry energy | a lost beam would destroy the machine | 362 MJ per beam — enough to melt half a tonne of copper |
⚙️ Engineer’s bridge — two design arguments, each a single back-of-envelope calculation
The book does something unusual here: it derives the LHC’s two headline parameters rather than quoting them. Both arguments take three lines, and both start from the hardest case the machine had to cover — a 1 TeV Higgs.
The energy, from vector-boson fusion. At high the dominant production mode is §9.13’s VBF: a quark from each proton radiates a , and the two fuse. Each must carry about GeV, so the quark radiating it needs roughly twice that — 1 TeV. A valence quark carries on average 1/6 of the proton momentum (§6.2), so the proton needs about 6 TeV. The design is 7. That is where the number comes from.
The luminosity, from the rare channel. With fb at TeV and , asking for ten events in a running year of s fixes .
Both arguments have the same shape, and it is the shape of every systems-design calculation: start from the hardest requirement, propagate backwards through the chain, and read off the specification. You do not design for the typical case; you design for the case that would otherwise be impossible, and check that everything easier comes free.
The result is a machine whose energy was set by a particle that does not exist at that mass, and whose luminosity was set by a channel with a branching ratio of a thousandth. Both requirements turned out to be far more than needed — the Higgs is at 125 GeV — and that margin is why the discovery came in two years rather than twenty.
Where it breaks: both back-of-envelope arguments assumed the thing they were designed to find. The energy requirement was set by “whatever mass the Higgs turns out to have, up to ~1 TeV” and the luminosity by a branching ratio computed in the Standard Model — so a design driven by one channel’s rate is only as robust as that channel existing. Requirements estimates also bound what is necessary, never what is sufficient: neither number says anything about the trigger throwing the event away, the pile-up burying it, or the calorimeter resolution smearing the peak below significance, and all three were harder problems than the two the envelope solved.
the LHC, derived
import numpy as np
print("the energy argument (vector-boson fusion at the hardest mass, M_H = 1 TeV)")
print(" each fused W must carry about M_H/2 = 500 GeV")
print(" the quark radiating it needs roughly 2x = 1000 GeV")
print(" a valence quark carries ~1/6 of the proton")
print(" -> proton energy needed = 6000 GeV")
print(" design: 7000 GeV per beam. matches.")
sig, br, yr, N = 1e-38, 1e-3, 1e7, 10
print("\nthe luminosity argument, as the book sets it up")
print(f" sigma(H, M_H = 1 TeV) = 10 fb = {sig:.0e} cm^2")
print(f" BR(H -> ZZ -> 2l+ 2l-) ~ {br:.0e}")
print(f" a running year = {yr:.0e} s")
print(f" wanted = {N} events")
print(f" L = 10 / ({sig:.0e} x {br:.0e} x {yr:.0e}) = {N/(sig*br*yr):.1e} cm^-2 s^-1")
print("\n the book states 1e34. see the erratum -- and note that 1e34 with")
print(" these inputs gives ONE event per year, not ten.")
L, stot = 1e34, 110e-27
f_rev, nb = 299792458/26659, 2808
print("\nwhat 1e34 actually delivers, with sigma_tot = 110 mb at 14 TeV:")
print(f" interaction rate = {L*stot:.1e} per second")
print(f" bunch crossings = 2808 x {f_rev:.0f} Hz = {nb*f_rev:.2e} per second")
print(f" -> pile-up = {L*stot/(nb*f_rev):.1f} collisions per crossing book: 35")
print("\n (the book's 35 needs the TOTAL cross-section, not the inelastic")
print(" 80 mb, which would give 25. it is right as printed.)")
E = nb*1.15e11*7e12*1.602176634e-19
print("\nthe stored beam energy, and why it is terrifying")
print(f" 2808 bunches x 1.15e11 protons x 7 TeV = {E/1e6:.0f} MJ")
print( " melting 1 kg of copper takes 0.7 MJ")
print(f" -> one beam could melt {E/1e6/0.7:.0f} kg of copper")
print("\nbunch spacing")
print(f" 25 ns x c = {25e-9*299792458:.2f} m book: 7.5 m")the energy argument (vector-boson fusion at the hardest mass, M_H = 1 TeV) each fused W must carry about M_H/2 = 500 GeV the quark radiating it needs roughly 2x = 1000 GeV a valence quark carries ~1/6 of the proton -> proton energy needed = 6000 GeV design: 7000 GeV per beam. matches. the luminosity argument, as the book sets it up sigma(H, M_H = 1 TeV) = 10 fb = 1e-38 cm^2 BR(H -> ZZ -> 2l+ 2l-) ~ 1e-03 a running year = 1e+07 s wanted = 10 events L = 10 / (1e-38 x 1e-03 x 1e+07) = 1.0e+35 cm^-2 s^-1 the book states 1e34. see the erratum -- and note that 1e34 with these inputs gives ONE event per year, not ten. what 1e34 actually delivers, with sigma_tot = 110 mb at 14 TeV: interaction rate = 1.1e+09 per second bunch crossings = 2808 x 11245 Hz = 3.16e+07 per second -> pile-up = 34.8 collisions per crossing book: 35 (the book's 35 needs the TOTAL cross-section, not the inelastic 80 mb, which would give 25. it is right as printed.) the stored beam energy, and why it is terrifying 2808 bunches x 1.15e11 protons x 7 TeV = 362 MJ melting 1 kg of copper takes 0.7 MJ -> one beam could melt 517 kg of copper bunch spacing 25 ns x c = 7.49 m book: 7.5 m
Erratum — the luminosity calculation on p. 405
The design luminosity is derived as
Two problems. The third factor in the denominator should be the running year, s — the quantity the preceding sentence defines — and has been copied in from the answer. As printed the expression evaluates to .
And with the correct s it gives , not . Equivalently: at those inputs give one event per year, not the ten the sentence asks for. Nothing downstream depends on it — is the real design luminosity — but the arithmetic the reader is invited to follow does not close. Confirmed on the render of PDF p. 423.
Erratum — Run 2 was at 13 TeV, and the book says so on p. 416
Page 406 reads: “The machine restarted collisions at TeV in April 2015, delivering in the so-called Run-2 more than 150 fb⁻¹.”
Run 2 ran at 13 TeV. 13.6 TeV is the Run 3 energy, from 2022 — and the same paragraph correctly dates Run 3 to April 2022, so the number has been imported from one run into the other.
The book contradicts itself ten pages later: §9.16 on p. 416 says “In the subsequent data-taking period, Run-2, ATLAS and CMS collected (at TeV) an integrated luminosity of 150 fb⁻¹” — the same run, the same luminosity, the correct energy.
Confirmed on the render of PDF p. 424.
Pile-up, and the 25-nanosecond problem
At design luminosity there are about proton–proton interactions per second, arriving in bunch crossings 25 ns apart. Two consequences follow, and between them they dictate the detector technology.
Pile-up pile-up the several tens of additional proton–proton collisions occurring in the same bunch crossing as the one of interest. It is why LHC trackers use pixels rather than strips, and why choosing the right primary vertex is itself a measurement. defined in §9.14 — open in glossary . Roughly 35 separate collisions happen in the same crossing. The interesting one is buried among 34 others, and the detector must assign every track and every shower to the right vertex.
Speed. Twenty-five nanoseconds is 7.5 m of light travel — so the tracks from one crossing are still in the middle of the detector when the next crossing happens. A time projection chamber, which UA1 used and which drifts electrons for microseconds, is simply impossible. The trackers are silicon.
💡 What this really says — occupancy is the quantity that forced pixels
About a thousand charged tracks emerge from each crossing. The design question is not “can we record them” but what fraction of the detector elements contain a hit — the occupancy. If that fraction is high, tracks cannot be told apart and pattern recognition fails.
Occupancy is hits divided by channels, so there are two ways to reduce it and only one is available. You cannot reduce the hits. So you increase the channels, which means making each element smaller.
The innermost layers use pixels — two-dimensional elements 100–150 μm on a side — instead of micro-strips, which are narrow but long and therefore cover a large area each. The price is the channel count: 70 million in CMS.
An engineer will recognise the trade as spatial multiplexing under a collision constraint — the same reasoning behind cache associativity, hash-table sizing and frequency reuse in cellular networks. In every case the resource is distinguishability, the enemy is collisions between independent things landing in the same bin, and the fix is more bins, paid for in wiring.
And, as always, the wiring is the hard part: 70 million channels of low-noise electronics, inside the tracking volume, that must survive years of radiation.
🧅 ATLAS and CMS — one structure, two magnet philosophies
Pick a particle above to light up the layers it touches, and click any layer for what it measures. No single layer identifies anything — the pattern across all five does.
Bettini Figs. 9.46–9.49, redrawn. The layer order is the same as UA1's (§9.6) — tracker, electromagnetic calorimeter, hadron calorimeter, muon chambers — because the physics of what stops where has not changed. What differs between the two experiments is the magnet.
| subsystem↕ | ATLAS↕ | CMS↕ | the trade↕ |
|---|---|---|---|
| magnet | solenoid m, T, plus a huge air-core toroid for muons ( m, T, eight coils 26 m × 5 m) | one solenoid, m, T, containing both calorimeters | ATLAS buys muon resolution with a long lever arm in a weak field; CMS buys it with a strong field and a compact radius. The sagitta at 1 TeV is ≈ 0.5 mm either way |
| tracker | silicon pixels + strips | silicon pixels + strips, 70 M channels | both: = 2–3 % at 100 GeV — an order of magnitude better than LEP |
| ECAL | sampling: liquid argon with lead, in an accordion geometry for granularity | PbWO₄ crystals — 76 000 of them, faces 22 × 22 mm², Molière radius 21 mm | CMS chose ultimate energy resolution (crystals); ATLAS chose radiation hardness and longitudinal segmentation — which let it point photons back to their vertex |
| HCAL | iron–scintillator barrel, liquid argon end caps | brass–scintillator throughout | both hermetic to within 1° of the beams, because demands it |
| muons | toroid + chambers | four drift-chamber stations interleaved with the flux-return brass | both: at 1 TeV, and a di-muon mass resolution of 67 MeV at the ϒ |
Supplied — the section argues the technology from the timing and the numbers deserve an axis. 25 ns is not merely “fast”: it is shorter than the time light needs to cross the apparatus. A detector that integrates for microseconds would be summing hundreds of crossings, so a TPC — the workhorse of the previous generation, and the technology §1.13c spends pages on — is not slow by a margin that better electronics could close, it is out by two orders of magnitude.
That single constraint is why the LHC trackers are silicon, why they have 70 million channels, and why the whole readout is pipelined: at any instant the detector holds several events at different stages, which is a systems-engineering problem before it is a physics one.
Why the calorimeters decided the discovery
Both discovery channels are two-body invariant masses, so the resolution formula is the one that matters:
Bettini p. 409, from M = √(2E₁E₂(1 − cos θ)). The angular term is small in practice, so the mass resolution is HALF the energy resolution — which is why the calorimeters were where the effort went.
Every symbol, one at a time
Hover or tap a symbol above — it lights up in the equation and its meaning, units and type appear here.
The chain is worth stating in one line, because it explains the whole design. To see a 125 GeV Higgs you need a 1–2 GeV mass peak; that needs ~2 % energy resolution; that needs either crystals or a finely-segmented sampling calorimeter; and either way it needs the right vertex out of 35. Every one of those requirements traces back to a branching ratio of .
Finally the trigger, which is the least glamorous and most decisive part. About interactions per second, each producing roughly 1 Mb, is bits per second — nothing can record that. Dedicated fast electronics plus software reduce the rate to about 0.5 kHz before storage, a rejection factor of two million, and every event thrown away is thrown away forever.
🔑 If you remember only three things
-
A rare channel is a luminosity requirement wearing different clothes. Choosing what you must see chooses how many collisions per second you must survive.
-
Pile-up is the design point, not a nuisance. A thousand tracks per crossing is what running at the required luminosity means, and the detector is built around it.
-
The calorimeters decided the discovery. Both channels that found the particle are calorimetric, so the subsystem hardest to build was the one that mattered.
Where this goes next
The machine ran: 4.7 fb⁻¹ at 7 TeV in 2011, and 23 fb⁻¹ at 8 TeV in 2012 — after a false start in 2008, when a bad soldered joint between superconducting cables, one of tens of thousands that should each have had nano-ohm resistance, quenched a dipole and damaged the machine badly enough to cost more than a year.
§9.15 is what came out of those 28 fb⁻¹.
✅ Check yourself — the LHC and its detectors
0/6 answered · 0 correct
1.Where does the LHC's 7 TeV beam energy come from?
2.The book derives the design luminosity as 10/(10⁻³⁸ × 10⁻³ × 10³⁴) = 10³⁴ cm⁻²s⁻¹. What is wrong?
3.About a thousand tracks emerge from each crossing, and the innermost tracker layers use pixels rather than micro-strips. Why?
4.Why did the electromagnetic calorimeters decide the Higgs discovery?
5.ATLAS uses a 4.5 m air-core toroid at 0.6 T for muons; CMS uses a compact 3.8 T solenoid. Is one better?
6.The book says the machine 'restarted collisions at √s = 13.6 TeV in April 2015, delivering in the so-called Run-2 more than 150 fb⁻¹'. What is wrong?