Measurements at accelerators

Almost everything we know about the fundamental constituents of matter — the quarks, the leptons, the force carriers, the Higgs — was learned by accelerating charged particles to enormous energy, smashing them together, and counting what comes out. An accelerator is not a microscope you look through; it is an instrument you read by tallying. Two numbers decide what any machine can do. The first is the centre-of-mass energy $\sqrt s$ , recalled from the prerequisite frame article: it fixes the mass scale of what can be created, since you cannot make a $W$ , $Z$ , or Higgs unless $\sqrt s$ exceeds the sum of the final-state masses. The second is the instantaneous luminosity $\mathcal L$ , which fixes how many collisions happen per second — and so how long you must wait for a process rare enough to be interesting. These figures split accelerators into two families. A fixed-target machine drives a beam into a stationary block of matter; momentum conservation forces the debris forward carrying most of the beam's energy as useless bulk motion, so the reach $\sqrt s\approx\sqrt{2m_2E}$ grows only as the square root of the beam energy. A collider brings two beams into head-on collision, so the laboratory is the centre-of-mass frame and the full $\sqrt s=2E$ is available — which is why the energy frontier is built from colliding beams. The rest of the subject is a special bookkeeping: a measured event count $N$ becomes a fundamental cross section $\sigma$ through $N=\sigma\int\mathcal L\,\mathrm dt$ , the equation that turns a tally on a detector into a number a theorist can predict.

Key takeaways

Two figures of merit rule every accelerator: the centre-of-mass energy $\sqrt s$ (sets which particles can be made) and the luminosity $\mathcal L$ (sets how often collisions happen).
Fixed-target machines waste energy on forward recoil, so $\sqrt s\approx\sqrt{2m_2E}$ grows only as $\sqrt E$ ; colliders put the whole $2E$ into the centre of mass — why the field runs on colliding beams.
Measurements reduce to counting: the yield is $N=\sigma\int\mathcal L\,\mathrm dt=\sigma\,\mathcal L_{\rm int}$ , so integrated luminosity converts counts into cross sections.
Absolute luminosity is hard to know, so cross sections are measured relative to a reference process, $\sigma=\sigma_{\rm ref}\,(N/\varepsilon)/(N_{\rm ref}/\varepsilon_{\rm ref})$ .
Cross sections are quoted in barns ( $1\ \text{barn}=10^{-24}\,\mathrm{cm^2}$ ) and span twelve orders of magnitude at one machine — from $\sim100$ mb inelastic $pp$ to $\sim50$ pb for the Higgs.
Key idea: accelerator physics is measurement by counting — you multiply a cross section by the integrated luminosity to predict a yield, and you invert the same relation, against a reference process, to extract a cross section from the tally your detector records.

The core insight

Accelerator measurements reduce to counting events: the yield is the cross section times the integrated luminosity, $N=\sigma\int\mathcal L\,\mathrm dt$ , so the physics rests on knowing at what energy the collisions occurred ( $\sqrt s$ ) and how many of them there were ( $\mathcal L$ ).

Where this lives today

The two figures of merit headline every machine. When the LHC quotes " $\sqrt s=13.6\ \mathrm{TeV}$ " it states the invariant of equation (1) and " $140\ \mathrm{fb^{-1}}$ " states the $\mathcal L_{\rm int}$ of equation (2) — and every published cross section is the ratio (3), normalised against a reference and corrected for efficiency. The HL-LHC upgrade is entirely a luminosity project, aiming for $3000\ \mathrm{fb^{-1}}$ to accumulate enough rare events such as Higgs pair production, at the cost of pileup near 200 collisions per crossing that drives timing detectors separating vertices in time as well as space. The subsystems that count the right events are the other half: silicon microvertex detectors with $\sim10\ \mu\mathrm{m}$ resolution tag $b$ quarks through the displaced vertex a $B$ hadron makes after a few millimetres of flight ( $\beta\gamma c\tau\approx2.6\ \mathrm{mm}$ for a $30\ \mathrm{GeV}$ $B$ ), and the asymmetric $B$ -factories PEP-II and KEKB tuned their beam energies to boost those decays into a measurable distance. Fixed-target physics thrives too — its forward boost is a feature for neutrino beams like DUNE's LBNF, and $ep$ machines like HERA mapped the proton by deep-inelastic scattering. Every one is the same bookkeeping: pick the geometry, know your $\sqrt s$ , measure your $\mathcal L$ , and count.

The physical picture

Start with the geometry, because it decides everything downstream. Fire a beam into a target that just sits there and the products cannot stop — the incoming momentum must go somewhere, so the debris is thrown forward as a boosted jet, and only the sliver left over in the centre-of-mass frame can make new mass. Bring two beams together head-on and there is no net momentum to carry away: the whole collision energy is available. The first figure sets the two geometries side by side with the numbers that decided the architecture of modern physics.

Figure

The two accelerator geometries at the same 7 TeV beam energy. A fixed target leaves the reach at √s = √(2 m_p E) = 114.6 GeV because the debris is thrown forward carrying most of the energy; colliding two beams head-on makes the laboratory the centre-of-mass frame and delivers the full √s = 2E = 14 TeV — a factor of 122 more reach from the identical beam

Having the energy is necessary but not sufficient — the collisions must also be plentiful, because the processes worth studying are rare. At the LHC an inelastic proton-proton scatter happens $\sim10^{9}$ times a second while a Higgs appears about once a second, a part in a billion; luminosity is the only lever on how fast the rare events accumulate. The second figure is the landscape a physicist plans around: the cross sections of standard LHC processes, spanning nearly twelve orders of magnitude.

Figure

Standard-Model cross sections at the LHC near √s = 13 TeV, on a logarithmic scale in picobarns. From the inelastic proton-proton total of about 75 mb down to Higgs production near 55 pb, the rates span nearly twelve orders of magnitude — which is why a rare process demands enormous integrated luminosity to accumulate a usable sample

The core equations

The first figure of merit is the energy that can be turned into mass, recalled from the centre-of-mass article. For a beam of energy $E_1$ on a target of mass $m_2$ at rest the invariant $s$ is linear in the beam energy; two equal head-on beams put the entire $2E$ into the centre of mass. The gap between a square root and a straight line is the whole case for colliders.

\underbrace{s=m_1^2+m_2^2+2E_1m_2}_{\text{fixed target}}\;\Rightarrow\;\sqrt s\approx\sqrt{2m_2E_1}\;;\qquad \underbrace{\sqrt s=2E}_{\text{symmetric collider}} \qquad (1)

where: $s$ — squared centre-of-mass energy, a Lorentz invariant; $E_1$ — beam energy; $m_2$ — target rest mass; $E$ — energy of each beam in a symmetric collider. Natural units ( $c=1$ , energies in GeV) throughout. Reading it: the fixed-target reach $\sqrt s\propto\sqrt{E_1}$ crawls upward, so doubling the beam buys only $\sqrt2$ ; the collider's $\sqrt s=2E$ grows linearly, wasting nothing. Everything a detector measures is ultimately a count of events, tied to physics by the luminosity. Instantaneous luminosity is defined as the proportionality constant between a process's rate and its cross section; integrating over a run gives the yield in terms of the integrated luminosity.

R=\frac{\mathrm dN}{\mathrm dt}=\mathcal L\,\sigma,\qquad N=\sigma\!\int\!\mathcal L\,\mathrm dt=\sigma\,\mathcal L_{\rm int} \qquad (2)

where: $R$ — event rate (per second); $\mathcal L$ — instantaneous luminosity ( $\mathrm{cm^{-2}\,s^{-1}}$ ); $\sigma$ — cross section; $\mathcal L_{\rm int}=\int\mathcal L\,\mathrm dt$ — integrated luminosity ( $\mathrm{cm^{-2}}$ , quoted as $\mathrm{fb^{-1}}$ ). Reading it: $\mathcal L$ is a property of the machine and $\sigma$ of the physics; their product is what you observe, so to predict how many you make, multiply cross section by integrated luminosity. In principle $\mathcal L$ follows from beam parameters, but transverse profiles are never known well enough for a precise absolute number. The escape is to measure every cross section relative to a reference process of known cross section, so the poorly-known luminosity cancels.

\sigma=\frac{N/\varepsilon}{\mathcal L_{\rm int}},\qquad \mathcal L_{\rm int}=\frac{N_{\rm ref}}{\sigma_{\rm ref}\,\varepsilon_{\rm ref}}\;\Rightarrow\;\sigma=\sigma_{\rm ref}\,\frac{N/\varepsilon}{N_{\rm ref}/\varepsilon_{\rm ref}} \qquad (3)

where: $N$ — background-subtracted signal count; $\varepsilon$ — detection efficiency (acceptance $\times$ reconstruction $\times$ selection); $N_{\rm ref},\sigma_{\rm ref},\varepsilon_{\rm ref}$ — the same for the reference. Reading it: because both processes share the same integrated luminosity, it drops out of the ratio, so the systematic uncertainty on $\mathcal L$ never enters the measured cross section. Where does $\mathcal L$ come from at the machine level? From how densely the beams are packed when they cross. For two bunched Gaussian beams meeting head-on it is set by the bunch populations, bunch count, revolution frequency, and transverse focusing. The detailed derivation is owned by the luminosity article and recalled by name.

\mathcal L=f\,n_b\,\frac{N_1N_2}{4\pi\,\sigma_x\sigma_y} \qquad (4)

where: $f$ — revolution frequency; $n_b$ — bunches per beam; $N_1,N_2$ — particles per bunch; $\sigma_x,\sigma_y$ — RMS transverse beam sizes at the interaction point. Reading it: luminosity rewards more particles ( $N_1N_2$ ), more bunches, and tighter focusing (small $\sigma_x\sigma_y$ ) — the three knobs a machine physicist turns to buy rare events. Finally, the unit. Cross sections carry the dimensions of area, and the natural unit is the barn — roughly the cross-sectional area of a heavy nucleus. Its inverse sub-multiples give the shortcut for turning a cross section and an integrated luminosity straight into an event count.

1\ \text{barn}=10^{-28}\ \mathrm{m^2}=10^{-24}\ \mathrm{cm^2},\qquad N=\sigma[\mathrm{pb}]\times\mathcal L_{\rm int}[\mathrm{fb^{-1}}]\times10^{3} \qquad (5)

where: $\mathrm{pb}=10^{-12}\,\text{barn}$ (picobarn); $\mathrm{fb^{-1}}=10^{15}\,\text{barn}^{-1}$ ; the factor $10^{3}$ is the arithmetic of the two prefixes. Reading it: a $1$ pb process in $1\ \mathrm{fb^{-1}}$ yields $10^{3}$ events — the shortcut every experimentalist carries.

Derivation AFrom beam energy to physics reach — fixed target versus colliding beams

The goal is to make quantitative the single most consequential fact about accelerator design: a stationary target squanders almost all of a beam's energy on bulk forward motion, so a fixed-target machine's reach climbs only as the square root of its beam energy, whereas colliding two beams head-on makes the full sum available. We take the invariant $s$ recalled from the centre-of-mass article, evaluate it in each geometry, take the high-energy limit, and read off the scaling that pushed the field toward colliders.

Step 1. For a beam $p_1=(E_1,\mathbf p_1)$ striking a target at rest, $p_2=(m_2,\mathbf 0)$ , the invariant is $s=(p_1+p_2)^2=m_1^2+m_2^2+2E_1m_2$ , since $p_1\!\cdot\!p_2$ collapses to $E_1m_2$ when the target does not move. Why: with $\mathbf p_2=0$ the cross term keeps only the energy component, leaving $s$ strictly linear in $E_1$ .

Step 2. At high energy $E_1\gg m_1,m_2$ the mass terms are negligible, so $s\approx2m_2E_1$ and $\sqrt s\approx\sqrt{2m_2E_1}$ . Doubling $E_1$ multiplies $\sqrt s$ by only $\sqrt2\approx1.41$ ; to double the reach you must quadruple the beam. Why: the centre-of-mass frame is dragged forward with the debris, and the energy locked in that bulk motion — most of $E_1$ — can never become new mass.

Step 3. For a symmetric collider, two equal beams meet head-on: $E_1=E_2=E$ and $\mathbf p_1=-\mathbf p_2$ , so the total three-momentum vanishes and the laboratory is the centre-of-mass frame. Then $s=(2E)^2$ and $\sqrt s=2E$ , growing linearly with beam energy. Why: with nothing drifting, no energy is spent on bulk motion, and the entire $2E$ is on the table for particle creation.

Step 4. Put in a real beam energy. At $E=7\ \mathrm{TeV}$ a collider reaches $\sqrt s=14\ \mathrm{TeV}$ ; the same beam on a stationary proton reaches only $\sqrt s=\sqrt{2(0.938)(7000)}\ \mathrm{GeV}=114.6\ \mathrm{GeV}$ , smaller by a factor of $122$ . To match $14\ \mathrm{TeV}$ fixed-target you would need a beam of $s/2m_2\approx1.0\times10^{8}\ \mathrm{GeV}$ , an accelerator far larger than the Earth. Why: this hundred-fold gap, drawn in the first figure, is why the energy frontier abandoned fixed targets for colliding beams. *Verified: $\sqrt s=114.6$ GeV (7 TeV beam, fixed proton target) versus $14$ TeV (collider), ratio $122$ ; the numbers reproduce Thomson Eqs. (1.4)–(1.5).*

Go deeper · Derivation B · 1 minTurning an event count into a cross section

Derivation BTurning an event count into a cross section

The aim is to convert what a detector records — a raw tally of events — into the frame-independent number a theorist predicts, the cross section, and to explain why that conversion runs not through first-principles beam parameters but against a reference process of known rate. We start from luminosity as the proportionality between rate and cross section, integrate over a run, fold in efficiency and background, and invert to isolate $\sigma$ in a form where the poorly-known luminosity cancels.

Step 1. Luminosity is defined by $R=\mathcal L\,\sigma$ : the rate of a process equals its cross section times a machine factor with the dimensions of inverse-area-per-time. Integrate over the data-taking period to get the yield produced, $N_{\rm prod}=\sigma\int\mathcal L\,\mathrm dt=\sigma\,\mathcal L_{\rm int}$ . Why: $\sigma$ (physics) and $\mathcal L$ (machine) enter only as a product, so a count constrains their product and nothing finer without extra input.

Step 2. A detector misses some produced events and records some non-signal ones, so $N_{\rm obs}=\varepsilon\,\sigma\,\mathcal L_{\rm int}+B$ , with efficiency $\varepsilon$ (acceptance $\times$ reconstruction $\times$ selection) and estimated background $B$ . Solving, $\sigma=(N_{\rm obs}-B)/(\varepsilon\,\mathcal L_{\rm int})\equiv N/(\varepsilon\,\mathcal L_{\rm int})$ . Why: acceptance and backgrounds are unavoidable, and both must be corrected before a count means a cross section.

Step 3. The remaining unknown $\mathcal L_{\rm int}$ cannot be computed precisely from equation (4) because $\sigma_x,\sigma_y$ are imperfectly known. So measure it from a reference process of known cross section in the same beams: $N_{\rm ref}=\varepsilon_{\rm ref}\,\sigma_{\rm ref}\,\mathcal L_{\rm int}$ , hence $\mathcal L_{\rm int}=N_{\rm ref}/(\sigma_{\rm ref}\,\varepsilon_{\rm ref})$ . Why: a process you can calculate — small-angle Bhabha scattering at LEP, or $Z$ production at the LHC — becomes a built-in luminosity meter.

Step 4. Substitute to eliminate $\mathcal L_{\rm int}$ : $\sigma=\sigma_{\rm ref}\,(N/\varepsilon)/(N_{\rm ref}/\varepsilon_{\rm ref})$ . Luminosity has cancelled, so its large systematic uncertainty never touches the ratio; what remains is the reference cross section and the two efficiencies. Why: a ratio inherits only the precision of the reference and the relative efficiency — which is why relative measurements dominate accelerator physics. *Verified: with $\sigma_{\rm ref}=40$ nb, $N=10^4$ , $N_{\rm ref}=5\times10^5$ and equal efficiencies, $\sigma=40\times(10^4/5\times10^5)=0.80$ nb, independent of $\mathcal L_{\rm int}$ .*

Go deeper · Derivation C · 1 minThe luminosity of two bunched beams

Derivation CThe luminosity of two bunched beams

The purpose is to show where the luminosity in the counting formula comes from — the microscopic collision rate of two crossing beams — so the abstract $\mathcal L$ becomes a set of knobs the accelerator physicist turns, and to connect it to the pileup the same number forces on the detector. We count how often particles in one bunch pass through the transverse area of the oncoming bunch, assemble the standard Gaussian-beam result, and evaluate it for the LHC. The full treatment is owned by the luminosity article and recalled here by name.

Step 1. One bunch of $N_1$ particles sweeps through an oncoming bunch of $N_2$ particles spread over effective transverse area $A$ . A target particle presents cross section $\sigma$ , so a beam particle interacts with probability $N_2\sigma/A$ , giving $N_1N_2\sigma/A$ collisions per crossing. Why: cross section is the effective area for a hit, so the count is (particles) $\times$ (targets per area) $\times$ (area presented).

Step 2. Bunches cross repeatedly: with $n_b$ bunches revolving at frequency $f$ , the crossing rate is $f\,n_b$ , so $R=f\,n_b\,N_1N_2\sigma/A\equiv\mathcal L\,\sigma$ . This identifies $\mathcal L=f\,n_b\,N_1N_2/A$ , and for Gaussian transverse profiles $A=4\pi\sigma_x\sigma_y$ . Why: matching the microscopic rate to $R=\mathcal L\sigma$ isolates the luminosity as a pure machine quantity, independent of the process $\sigma$ .

Step 3. Insert LHC-scale numbers — $N_1=N_2\approx1.1\times10^{11}$ , $n_b\approx2800$ , $f\approx1.1\times10^{4}\ \mathrm{Hz}$ , a micron-scale spot — and $\mathcal L$ lands near $10^{34}\ \mathrm{cm^{-2}\,s^{-1}}$ , the design value. Why: this is the number quoted for a machine, engineered upward by squeezing the beams and packing more protons per bunch.

Step 4. The same $\mathcal L$ forces pileup. With $\sigma_{\rm inel}\approx80\ \mathrm{mb}=8\times10^{-26}\ \mathrm{cm^2}$ , the inelastic rate is $R=\mathcal L\,\sigma_{\rm inel}\approx(2\times10^{34})(8\times10^{-26})=1.6\times10^{9}\ \mathrm{s^{-1}}$ ; dividing by $40\ \mathrm{MHz}$ gives $\sim40$ separate $pp$ interactions per crossing. Why: high luminosity is not free — it piles many collisions onto each event, and untangling them is a central challenge for the detectors. *Verified: $\mathcal L\sigma_{\rm inel}/f_{\rm cross}=(2\times10^{34})(8\times10^{-26})/(4\times10^{7})=40$ interactions per crossing.*

Limiting cases

Case	Result	Meaning
Fixed target, high energy	$\sqrt s\approx\sqrt{2m_2E_1}$	reach grows only as $\sqrt{E_1}$ — the collider motivation
Symmetric collider	$\sqrt s=2E$	lab is the CM frame; full beam energy available, linear growth
Rare process	$N=\sigma\,\mathcal L_{\rm int}$ small	needs large $\mathcal L_{\rm int}$ ; sets the run length for discovery
Reference normalisation	$\sigma=\sigma_{\rm ref}\,N/N_{\rm ref}$	luminosity cancels; precision set by the reference
High luminosity	$\mu=\mathcal L\,\sigma_{\rm inel}/f_{\rm cross}$	many collisions per crossing (pileup) to disentangle
Threshold	$\sqrt s=\sum_f m_f$	minimum energy to create the final state

Common trapDoubling a fixed-target machine's beam energy doubles its reach

It does not. What sets the reach — the heaviest particle you can make — is $\sqrt s$ , and for a fixed target $\sqrt s\approx\sqrt{2m_2E_1}$ grows only as the square root of the beam energy: double $E_1$ and $\sqrt s$ rises by a mere $\sqrt2\approx1.41$ , so to double $\sqrt s$ you must quadruple the beam. This is why the field abandoned fixed targets at the energy frontier: in a symmetric collider $\sqrt s=2E$ grows linearly, every extra GeV a full extra 2 GeV of reach. The confusion comes from equating "beam energy" with "energy available to make particles" — but in the lab the products inherit the beam's momentum and stream forward, carrying most of that energy as bulk motion that can never become mass. Only the centre-of-mass slice $\sqrt s$ is convertible. (A companion trap: high luminosity is not automatically good news — it brings pileup, tens of overlapping collisions per crossing, and a detector that cannot resolve them loses more than the extra rate gained.)

Worked examples

1 · Counting Higgs bosons at the LHC (fully worked)

The Higgs production cross section at $\sqrt s=13\ \mathrm{TeV}$ is about $\sigma_H\approx55\ \mathrm{pb}$ , and LHC Run 2 delivered $\mathcal L_{\rm int}\approx140\ \mathrm{fb^{-1}}$ . How many Higgs bosons were produced, and at what rate when $\mathcal L=2\times10^{34}\ \mathrm{cm^{-2}\,s^{-1}}$ ?

Step 1. Yield from equation (2) via the shortcut $N=\sigma[\mathrm{pb}]\times\mathcal L_{\rm int}[\mathrm{fb^{-1}}]\times10^{3}=55\times140\times10^{3}=7.7\times10^{6}$ — about 7.7 million Higgs bosons.

Step 2. Instantaneous rate from $R=\mathcal L\,\sigma$ . With $\sigma_H=55\ \mathrm{pb}=5.5\times10^{-35}\ \mathrm{cm^2}$ , $R=(2\times10^{34})(5.5\times10^{-35})=\mathbf{1.1\ Higgs\ per\ second}$ .

Step 3. Why so few are seen. Only a small fraction land in a clean channel: the $H\to\gamma\gamma$ branching ratio is $\sim0.23\%$ , so $7.7\times10^{6}\times0.0023\approx1.8\times10^{4}$ diphoton decays are produced, of which a few thousand survive selection — enough for a sharp peak, but only because the luminosity was enormous.

Step 4. Interpret. This is the division of labour in equation (2): the machine supplied $\mathcal L_{\rm int}$ , the physics $\sigma_H$ , and only their product — folded with branching ratios and efficiency — decides whether a signal is observable. 7.7 million produced, ~1 per second, distilled to a few thousand analysed events.

2 · Pileup from luminosity (fully worked)

At $\mathcal L=2\times10^{34}\ \mathrm{cm^{-2}\,s^{-1}}$ with $\sigma_{\rm inel}\approx80\ \mathrm{mb}$ and a $40\ \mathrm{MHz}$ bunch-crossing frequency, how many separate collisions occur in each crossing?

Step 1. Total inelastic rate. With $\sigma_{\rm inel}=80\ \mathrm{mb}=8\times10^{-26}\ \mathrm{cm^2}$ , $R=\mathcal L\,\sigma_{\rm inel}=(2\times10^{34})(8\times10^{-26})=\mathbf{1.6\times10^{9}\ interactions/s}$ .

Step 2. Interactions per crossing. Divide by $f_{\rm cross}=40\ \mathrm{MHz}=4\times10^{7}\ \mathrm{s^{-1}}$ : $\mu=R/f_{\rm cross}=1.6\times10^{9}/4\times10^{7}=\mathbf{40}$ .

Step 3. What it means. Every bunch crossing brings about 40 independent proton-proton collisions on top of one another — a hard-scatter event of interest buried among 39 soft "pileup" vertices within a few centimetres along the beam.

Step 4. Why it drives detector design. Resolving these overlapping vertices is why silicon trackers with tens-of-micron resolution are indispensable — they separate the primary vertices along the beam so the physics event can be picked out. About 40 collisions per crossing, the price of $10^{34}$ luminosity, and the reason for high-granularity tracking.

Physics of the portrait

The accelerator is the instrument that made particle physics an experimental science, and it has a founder whose entire career was the conviction that you learn what matter is made of by building a machine to break it apart and counting the pieces. That founder was Ernest Lawrence.

Origins · 1 minErnest Lawrence and how this was discovered

Ernest Lawrence

1901–1958 · the cyclotron and the accelerator as the instrument of particle physics

Ernest Orlando Lawrence, born in South Dakota in 1901, changed how physics is done by inventing a way to reach high energy without a high-voltage terminal. His insight, conceived in 1929 from a diagram in a paper by Rolf Widerøe, was resonant acceleration: a charged particle spiralling in a magnetic field can be pushed twice per orbit by an alternating voltage, and because the orbital period is independent of energy for a non-relativistic particle, the same oscillator stays in step turn after turn, adding energy on every pass. The first cyclotron, built with M. Stanley Livingston in 1931, was small enough to hold in a hand and reached 80 keV; within a few years the machines grew to a metre across and passed tens of MeV. Lawrence did more than build a device — he built the model of modern experimental physics: the large, centrally-funded laboratory (the Berkeley Radiation Laboratory) organising physicists, engineers, and chemists around an accelerator, a template every major laboratory has followed. He won the 1939 Nobel Prize, and element 103, lawrencium, and the Lawrence Berkeley and Livermore laboratories carry his name. Every synchrotron and storage ring since — the Tevatron, LEP, the LHC — descends from the resonant-acceleration principle Lawrence first made work, and every measurement at them is a tally of the kind his cyclotrons pioneered.

History

1929

— Lawrence conceives resonant acceleration, the cyclotron principle

1932

— Cockcroft and Walton induce the first accelerator nuclear reaction, splitting lithium

1952

— Courant, Livingston and Snyder invent strong (alternating-gradient) focusing

1971

— CERN's Intersecting Storage Rings collide protons on protons, the first hadron collider

1983

— Stochastic cooling delivers the luminosity that lets UA1 and UA2 find the W and Z

Going further

Texts. Thomson, Modern Particle Physics, §1.4, Eqs. (1.4)–(1.5) and Table 1.4 · Griffiths, Introduction to Elementary Particles, 2nd ed., §6.1 (cross sections and luminosity) · Halzen & Martin, Quarks and Leptons, §4.3 · Peskin & Schroeder, An Introduction to Quantum Field Theory, §4.5 · Schwartz, QFT and the Standard Model, Ch. 5 · Particle Data Group, Review of Particle Physics (accelerator parameters).

Provenance. The invariant $s$ and $\sqrt s=2E$ (collider) versus $\sqrt s\approx\sqrt{2m_2E_1}$ (fixed target) are recalled from aq-pp-centre-of-mass-frame and aq-pp-mandelstam-variables; natural units from aq-pp-natural-units. The luminosity formula $\mathcal L=f\,n_b\,N_1N_2/4\pi\sigma_x\sigma_y$ is developed in full in aq-pp-luminosity-event-rates (forward reference), and the cross section is owned by aq-pp-interaction-cross-sections. A verifier should confirm: (1) fixed-target $\sqrt s=114.6$ GeV for a 7 TeV beam versus 14 TeV collider, ratio 122; (2) $N=\sigma\,\mathcal L_{\rm int}=7.7\times10^{6}$ Higgs for $55$ pb and $140\ \mathrm{fb^{-1}}$ ; (3) pileup $\mathcal L\sigma_{\rm inel}/f_{\rm cross}=40$ ; (4) $\sigma=\sigma_{\rm ref}N/N_{\rm ref}=0.80$ nb. $m_p=0.938$ GeV and the LHC cross sections are standard PDG values.

Prerequisites. Centre-of-mass & lab frames · Mandelstam variables · Interaction cross sections. Leads to. Luminosity & event rates · Collider experiments · Z resonance at LEP.

Ask the tutor: Ask me to work out the reach, event yield, or pileup of any machine — give me a beam energy, a cross section, and a luminosity, and we will turn it into counts.

What is the difference between a fixed-target experiment and a collider?

In a fixed-target experiment a single beam strikes a stationary block of matter; in a collider two beams meet head-on. The distinction is energy bookkeeping. When a beam hits a stationary target, momentum conservation forces the products to fly forward, carrying most of the beam energy away as bulk motion that can never become mass — so the useful energy $\sqrt s\approx\sqrt{2m_2E_1}$ grows only as the square root of the beam energy. In a symmetric collider the total momentum is already zero, the laboratory is the centre-of-mass frame, and the full $\sqrt s=2E$ is available. That is why every energy-frontier machine of the last half-century is a collider.

What exactly is luminosity, and why is it so important?

Luminosity $\mathcal L$ is the proportionality constant between the rate of a process and its cross section: $R=\mathcal L\,\sigma$ , in units of $\mathrm{cm^{-2}\,s^{-1}}$ . It is a property of the machine, not the physics — how many particles you push through how small a spot how often. It matters because the processes worth studying are rare: a Higgs appears about once per billion LHC collisions, so the only way to a usable sample is to make collisions plentiful. The integrated luminosity $\mathcal L_{\rm int}=\int\mathcal L\,\mathrm dt$ , quoted in inverse femtobarns, is what multiplies a cross section to give a count — and two machines at the same energy can differ by orders of magnitude in reach purely through it.

Why not just compute the luminosity from the beam parameters?

In principle you can: equation (4) gives $\mathcal L$ from the bunch populations, bunch count, revolution frequency, and transverse beam sizes. In practice the transverse profiles $\sigma_x,\sigma_y$ at the collision point are not known precisely enough — they depend on the optics and drift during a fill — so the absolute luminosity carries a sizeable systematic uncertainty. Rather than let that pollute every measurement, experiments measure cross sections relative to a reference process of independently known cross section; the luminosity then cancels, and the answer inherits only the reference's precision. Dedicated van der Meer scans, which sweep the beams across each other to map their overlap, calibrate the absolute luminosity when it is genuinely needed.

What is a barn, and why is a cross section measured in units of area?

A cross section is the effective target area a process presents — the bigger it is, the more likely the interaction — so it genuinely has the dimensions of area. The natural unit is the barn, $1\ \text{barn}=10^{-28}\ \mathrm{m^2}=10^{-24}\ \mathrm{cm^2}$ , roughly the area of a uranium nucleus; the name is wartime slang, a nucleus being "as big as a barn" to a slow neutron. Collider cross sections are far smaller, so the practical units are sub-multiples: millibarn, microbarn, nanobarn, picobarn (pb, $10^{-12}$ barn), femtobarn. Integrated luminosity is quoted in the inverse of these, so that $\sigma\times\mathcal L_{\rm int}$ is a pure count: $1$ pb in $1\ \mathrm{fb^{-1}}$ yields $1000$ events.

What is pileup, and why is high luminosity a mixed blessing?

Pileup is the presence of many separate proton-proton collisions in one bunch crossing. Because luminosity is high, the inelastic rate $\mathcal L\,\sigma_{\rm inel}$ vastly exceeds the bunch-crossing rate, so each crossing contains not one but tens of overlapping collisions — about 40 at LHC Run 2, up to 200 at the coming HL-LHC. Only one is usually the "hard" event of interest; the rest are soft background piled on top. The detector must disentangle them, which is why high-resolution silicon tracking (separating vertices along the beam) and, increasingly, precision timing (separating them in time) are essential. More luminosity buys more rare events but also more clutter, a balance struck by detector design.

How does an experiment know it is counting the right events — for example, a b quark?

By exploiting a distinctive signature and a subsystem tuned to it. A hadron containing a $b$ quark lives long enough ( $c\tau\approx0.5\ \mathrm{mm}$ ) that, once boosted, it travels a few resolvable millimetres before decaying, producing a secondary vertex displaced from the primary collision point. A silicon microvertex detector, with hit resolution of order $10\ \mu\mathrm{m}$ , reconstructs tracks precisely enough to see that displacement and tag the event. The count $N$ in the cross-section formula is always a count of events passing such a selection, and the efficiency $\varepsilon$ folds in how often a genuine event is correctly tagged — turning raw collisions into a clean, identified sample is the whole art of the detector.