Luminosity & event rates

Everything a collider physicist wants to measure is a cross section — a number, with the dimensions of area, encoding how likely a given quantum process is when two particles meet. But a cross section is not something you read off a dial. What a detector does is count: it records interactions one at a time and tallies how many looked like a Higgs boson, a top-quark pair, or nothing new. Luminosity is what bridges the two. It is the single machine-supplied proportionality constant that turns the abstract area $\sigma$ into a countable event rate, $R=\mathcal{L}\,\sigma$ , and its time integral turns $\sigma$ into a total harvest, $N=\sigma\!\int\!\mathcal{L}\,\mathrm{d}t$ . Nothing about the physics you are hunting enters $\mathcal{L}$ : it is set entirely by how tightly and how often the beams are squeezed together at the interaction point. That division of labour is what makes luminosity so central and so often misread. The energy of a collider decides which processes are kinematically open and roughly how big their cross sections are; the luminosity decides how many you will ever see. A rare picobarn process is invisible at low luminosity and a precision measurement at high luminosity, at the very same beam energy. This article develops luminosity from three angles — as a counting rate ( $\mathcal{L}=R/\sigma$ ), as a beam-overlap integral fixed by bunch charge and transverse size ( $\mathcal{L}=f n_b N_1 N_2/4\pi\sigma_x\sigma_y$ ), and as the quantity whose integral, in inverse femtobarns, is the currency of a discovery. The same overlap that delivers events also delivers pile-up, so a luminous machine is always a negotiation between rate and clutter.

Key takeaways

The event rate of any process is $R=\mathcal{L}\,\sigma$ , and the total recorded is $N=\sigma\int\mathcal{L}\,\mathrm{d}t=\sigma\,\mathcal{L}_{\text{int}}$ — luminosity converts an area into a count.
Instantaneous luminosity $\mathcal{L}$ has units $\mathrm{cm^{-2}\,s^{-1}}$ ; integrated luminosity $\mathcal{L}_{\text{int}}$ is quoted in inverse barns, usually $\mathrm{fb^{-1}}$ ( $1\ \mathrm{fb^{-1}}=10^{39}\ \mathrm{cm^{-2}}$ ).
Luminosity is a property of the beams and machine only — bunch charge, bunch frequency, transverse beam size — never of the process studied.
For head-on Gaussian bunches, $\mathcal{L}=f n_b N_1 N_2/(4\pi\sigma_x\sigma_y)$ ; the LHC's design $10^{34}\ \mathrm{cm^{-2}\,s^{-1}}$ follows from $N\approx1.15\times10^{11}$ , $n_b=2808$ , $\sigma\approx17\ \mu\mathrm{m}$ .
The same overlap produces pile-up, $\mu=\mathcal{L}\,\sigma_{\text{inel}}/(n_b f)$ interactions per crossing — about 25 at LHC design, up to 200 at the HL-LHC.
Counting is Poisson, so a yield $N$ carries an irreducible uncertainty $\sqrt{N}$ — why rare physics demands enormous $\mathcal{L}_{\text{int}}$ .
Key idea: luminosity is the machine-only proportionality constant that turns a process's cross section into a countable event rate ( $R=\mathcal{L}\sigma$ ), so that discovery is a race to accumulate integrated luminosity against the smallness of $\sigma$ and the statistics of $\sqrt{N}$ .

The core insight

Luminosity $\mathcal{L}$ is the beam-and-machine property that multiplies a process cross section to give its event rate, $R=\mathcal{L}\sigma$ ; its time integral $\mathcal{L}_{\text{int}}=\int\mathcal{L}\,\mathrm{d}t$ fixes the total number of events $N=\sigma\,\mathcal{L}_{\text{int}}$ , making integrated luminosity the true currency of every collider measurement.

Where this lives today

Luminosity is the axis along which particle physics is now advancing hardest, because the energy frontier is expensive to push while the intensity frontier still has room. The High-Luminosity LHC, from the late 2020s, aims to deliver $3000\ \mathrm{fb^{-1}}$ — ten times the Run 1–3 dataset — by squeezing $\beta^{*}$ to about $15\ \mathrm{cm}$ and installing "crab cavities" that tilt the bunches to collide head-on despite a crossing angle, recovering $F\to1$ . The price is pile-up of $\mu\approx140$ – $200$ , driving new silicon trackers with tens of millions of channels and precision timing layers that separate pile-up vertices to tens of picoseconds. A different strategy, luminosity levelling, deliberately holds $\mathcal{L}$ below its peak by keeping the beams slightly separated early in a fill and closing the gap as they deplete — trading peak rate for flat, manageable pile-up; LHCb has run this way for years. At the intensity extreme sits SuperKEKB, the electron–positron B-factory in Japan, which in 2022 set the all-time record of $4.7\times10^{34}\ \mathrm{cm^{-2}\,s^{-1}}$ with a "nano-beam" scheme. Proposals for a Future Circular Collider quote their physics case almost entirely in ab⁻¹, because at fixed energy that number is the discovery.

The physical picture

Two counter-rotating beams are not continuous streams but trains of dense packets — bunches — each holding of order $10^{11}$ particles compressed into a sliver a few centimetres long and tens of microns wide. At the interaction point the optics squeeze each bunch to its smallest transverse size, and the two are steered through each other. Most particles miss; a handful, weighted by the overlap of the two transverse charge clouds, pass close enough to interact. Multiply that overlap by the particles in each bunch and how often bunches cross, and you have the luminosity — a rate of opportunities per unit area, independent of what those opportunities produce.

Figure

Two Gaussian bunches crossing at the interaction point: with N₁ = N₂ = 1.15×10¹¹ protons per bunch, n_b = 2808 colliding pairs, f = 11.2 kHz and round beams squeezed to σ_x = σ_y ≈ 17 μm, the overlap formula L = f n_b N₁N₂ ÷ (4π σ_x σ_y) yields the LHC design luminosity 1.0×10³⁴ cm⁻² s⁻¹

The luminosity is only half the story; the other half is how steeply the cross sections of interesting processes fall. The total inelastic cross section is tens of millibarns, while the processes worth building a laboratory for — top pairs, Higgs bosons, anything beyond the Standard Model — sit ten or eleven orders of magnitude below it. The second figure is the one a physicist actually reasons with: a staircase of cross sections at the 13 TeV LHC, each rung annotated with the events it yields for a Run-2 slice of $100\ \mathrm{fb^{-1}}$ — a picobarn process gives millions of events only because $\mathcal{L}_{\text{int}}$ is astronomically large.

Figure

The cross-section staircase at the 13 TeV LHC on a log scale, with the event yield each process gives for 100 fb⁻¹: inelastic ≈ 80 mb (8×10¹⁵ events), W ≈ 190 nb (1.9×10¹⁰), Z ≈ 60 nb (6×10⁹), top pairs ≈ 830 pb (8×10⁷), Higgs ≈ 55 pb (5.5×10⁶) — eleven decades in σ bridged by one number, L_int

The core equations

The whole subject rests on one proportionality: the instantaneous rate at which a process occurs is its cross section times the luminosity. Because $\sigma$ is a property of the physics and $\mathcal{L}$ of the machine, the equation separates "what nature does" from "how hard we push".

R=\frac{\mathrm{d}N}{\mathrm{d}t}=\mathcal{L}\,\sigma \qquad (1)

where: $R$ — event rate, $\mathrm{s^{-1}}$ ; $\mathcal{L}$ — instantaneous luminosity, $\mathrm{cm^{-2}\,s^{-1}}$ ; $\sigma$ — cross section, $\mathrm{cm^2}$ (or barns). A larger cross section or a brighter machine both raise the rate, and neither knows about the other. Beams degrade over a fill and machines stop for maintenance, so what a measurement accumulates is the time integral of the luminosity — the number physicists quote for how much data an experiment has, and it fixes the event count outright.

N=\sigma\int\mathcal{L}\,\mathrm{d}t=\sigma\,\mathcal{L}_{\text{int}},\qquad 1\ \mathrm{fb^{-1}}=10^{39}\ \mathrm{cm^{-2}} \qquad (2)

where: $N$ — total events recorded; $\mathcal{L}_{\text{int}}=\int\mathcal{L}\,\mathrm{d}t$ — integrated luminosity, an inverse area in $\mathrm{fb^{-1}}$ or $\mathrm{ab^{-1}}$ . Shortcut: $N=\sigma[\mathrm{fb}]\times\mathcal{L}_{\text{int}}[\mathrm{fb^{-1}}]$ , so a one-picobarn process over one inverse femtobarn gives exactly a thousand events. The simplest way to build a luminosity is the fixed-target geometry: a beam of known flux fired into a stationary block. The luminosity is then the beam flux times the number of target scattering centres per unit area.

\mathcal{L}=\Phi\,n_T\,\ell \qquad (3)

where: $\Phi$ — incident flux, $\mathrm{s^{-1}}$ ; $n_T$ — target number density, $\mathrm{cm^{-3}}$ ; $\ell$ — target thickness, $\mathrm{cm}$ . A dense, thick target buys enormous luminosity — why fixed-target experiments dominate rare-decay and neutrino physics — at a fixed, low centre-of-mass energy. At a collider there is no material target: each beam is the other's target, and $\mathcal{L}$ is set by how completely the two transverse charge distributions overlap. For head-on bunches with Gaussian widths $\sigma_x,\sigma_y$ , the overlap integral collapses to the master equation of collider design (derived below).

\mathcal{L}=\frac{f\,n_b\,N_1 N_2}{4\pi\,\sigma_x\sigma_y} \qquad (4)

where: $f$ — revolution frequency, $\mathrm{s^{-1}}$ ; $n_b$ — colliding bunch pairs; $N_1,N_2$ — particles per bunch; $\sigma_x,\sigma_y$ — transverse r.m.s. beam sizes at the interaction point. Every lever a builder can pull is here: more charge, more bunches, tighter focus. Accelerator physicists rarely control $\sigma$ directly; they control the beam emittance $\varepsilon_n$ (a conserved phase-space area) and the focusing $\beta^{*}$ , with $\sigma=\sqrt{\varepsilon_n\beta^{*}/\gamma}$ . Substituting turns equation (4) into the form used to specify a machine, with a factor $F\le1$ for the crossing angle.

\mathcal{L}=\frac{f\,n_b\,N^2\,\gamma}{4\pi\,\varepsilon_n\,\beta^{*}}\,F \qquad (5)

where: $\gamma$ — beam Lorentz factor; $\varepsilon_n$ — normalized transverse emittance, $\mathrm{m\cdot rad}$ ; $\beta^{*}$ — amplitude function at the interaction point, $\mathrm{m}$ ; $F$ — crossing-angle factor. Luminosity rises with beam brightness $N/\varepsilon_n$ and tighter focusing, which is why every upgrade is a fight to shrink $\beta^{*}$ . The same overlap that yields signal events also yields many soft inelastic collisions in every crossing. The mean of these — the pile-up — follows from dividing the total inelastic rate by the crossing rate.

\mu=\frac{\mathcal{L}\,\sigma_{\text{inel}}}{n_b\,f} \qquad (6)

where: $\mu$ — mean interactions per crossing (pile-up); $\sigma_{\text{inel}}$ — total inelastic cross section ( $\approx80\ \mathrm{mb}$ for $pp$ at 13 TeV); $n_b f$ — crossing rate. Pile-up scales with luminosity, so the brightness that delivers rare physics also buries each event under soft collisions.

Derivation AFrom cross section to counting rate

The goal of this first derivation is to establish equation (1) not as a definition but as a consequence, and to show in the same stroke that luminosity is a pure property of the beams — the flux of projectiles times the areal density of targets — with the cross section entering only as the area each target presents. We start fixed-target, obtain $\mathcal{L}=\Phi n_T\ell$ , then argue a collider is the same statement with each beam as the other's target.

Step 1. A beam delivers $\Phi$ particles per second onto a thin slab of thickness $\ell$ and target density $n_T$ . Each particle sees $n_T\ell$ scattering centres per unit transverse area, each of effective area $\sigma$ , so the probability one projectile interacts is $P=n_T\ell\,\sigma$ (for $P\ll1$ ). Why: a cross section is the effective area of a target seen by a projectile, so "probability of hitting" is "area covered by targets" times "area per target".

Step 2. Multiply by the particles arriving per second: $R=\Phi P=\Phi n_T\ell\,\sigma$ . Grouping the beam-and-target factors defines $\mathcal{L}\equiv\Phi n_T\ell$ , giving $R=\mathcal{L}\sigma$ — equation (1), with the fixed-target $\mathcal{L}$ of equation (3) as a by-product. Why: the rate factorizes into a machine part (opportunities per area per time) and a physics part (how likely each converts), and naming the machine part is what "luminosity" does.

Step 3. Pass to a collider. There is no slab; beam 2 supplies the targets beam 1 flies through, and vice versa. The role of $\Phi n_T\ell$ is taken over by the overlap of the two moving bunch densities, but the factorization is untouched: still $R=\mathcal{L}\sigma$ with $\mathcal{L}$ containing only beam quantities. Why: nothing in Step 2 required the targets to be at rest, only that their projected areal density is a beam property — so the counting law survives, and the next derivation computes the overlap.

Step 4. Integrate over a run: since $\sigma$ is constant, $N=\int R\,\mathrm{d}t=\sigma\,\mathcal{L}_{\text{int}}$ — equation (2). Why: the total harvest depends on the machine only through the one accumulated number $\mathcal{L}_{\text{int}}$ , which is why datasets are reported in inverse femtobarns, not seconds.

Go deeper · Derivation B · 1 minThe luminosity of two Gaussian bunches

Derivation BThe luminosity of two Gaussian bunches

This derivation computes the collider luminosity of equation (4) from first principles, taking two bunches with Gaussian transverse charge distributions and evaluating the overlap integral that measures how effectively they collide. The aim is to show that $1/4\pi\sigma_x\sigma_y$ is the exact result of convolving two normalized Gaussians, and to expose what a builder must do to raise it. We work per crossing, then multiply by $f n_b$ .

Step 1. Write each bunch's normalized transverse density as a product of Gaussians, $\rho_i(x,y)=\dfrac{1}{2\pi\sigma_{xi}\sigma_{yi}}\exp\!\left(-\dfrac{x^2}{2\sigma_{xi}^2}-\dfrac{y^2}{2\sigma_{yi}^2}\right)$ , with $\int\rho_i\,\mathrm{d}x\,\mathrm{d}y=1$ . The overlap is $\mathcal{O}=\int\rho_1\rho_2\,\mathrm{d}x\,\mathrm{d}y$ , weighted by the $N_1 N_2$ pairs. Why: the chance a particle from bunch 1 finds a partner in bunch 2 is proportional to how much their probability clouds coincide — the product integral of the two densities.

Step 2. The integral factorizes in $x$ and $y$ . Using $\int\exp(-a x^2/2)\,\mathrm{d}x=\sqrt{2\pi/a}$ with $a=1/\sigma_{x1}^2+1/\sigma_{x2}^2$ , the $x$ -integral gives $\sqrt{2\pi}\,\sigma_{x1}\sigma_{x2}/\sqrt{\sigma_{x1}^2+\sigma_{x2}^2}$ , likewise for $y$ ; cancelling prefactors leaves $\mathcal{O}=\dfrac{1}{2\pi\sqrt{\sigma_{x1}^2+\sigma_{x2}^2}\,\sqrt{\sigma_{y1}^2+\sigma_{y2}^2}}$ . Why: two Gaussians convolve with widths adding in quadrature, so the overlap is governed by $\Sigma_x=\sqrt{\sigma_{x1}^2+\sigma_{x2}^2}$ — the very quantities a van der Meer scan measures.

Step 3. Take equal beams, $\sigma_{xi}=\sigma_x$ , $\sigma_{yi}=\sigma_y$ : then $\sqrt{\sigma_{x1}^2+\sigma_{x2}^2}=\sqrt{2}\,\sigma_x$ and $\mathcal{O}=\dfrac{1}{2\pi\cdot2\sigma_x\sigma_y}=\dfrac{1}{4\pi\sigma_x\sigma_y}$ . Why: the famous $4\pi$ is $2\pi$ from the normalizations times a factor 2 from adding two equal widths in quadrature.

Step 4. Restore counts and rate: each crossing brings $N_1 N_2$ pairs into overlap $\mathcal{O}$ at $f n_b$ crossings per second, so $\mathcal{L}=f n_b N_1 N_2\,\mathcal{O}=\dfrac{f n_b N_1 N_2}{4\pi\sigma_x\sigma_y}$ — equation (4). Why: this exposes every lever — luminosity grows linearly in bunch number and frequency, quadratically in charge, inversely in area, which is why machines fight to shrink $\sigma$ .

Go deeper · Derivation C · 1 minPile-up as a Poisson process

Derivation CPile-up as a Poisson process

The object here is to derive the pile-up relation of equation (6) and to show that the number of simultaneous interactions in a crossing is Poisson-distributed with a mean set by luminosity, cross section, and crossing rate. This matters because pile-up is the price of luminosity: its mean $\mu=\mathcal{L}\sigma_{\text{inel}}/n_b f$ and Poisson fluctuations are what let an experiment plan detector granularity and triggering.

Step 1. The total inelastic rate is $R_{\text{inel}}=\mathcal{L}\,\sigma_{\text{inel}}$ from equation (1). Shared among crossings at rate $n_b f$ , the mean per crossing is $\mu=R_{\text{inel}}/(n_b f)=\mathcal{L}\sigma_{\text{inel}}/(n_b f)$ . Why: pile-up is just the inelastic rate divided over crossings per second — a bookkeeping quotient, not new physics.

Step 2. The distribution is Poisson: one crossing has $\sim10^{22}$ potential proton pairs, each with an independent tiny probability of interacting — the textbook Poisson limit, so $P(k)=\mu^k e^{-\mu}/k!$ . Why: the same "many chances, each unlikely" structure that fixes the rate fixes the statistics, so signal yields and pile-up alike fluctuate as Poisson variables.

Step 3. Read off consequences: the probability of a clean crossing is $P(0)=e^{-\mu}$ , negligible once $\mu\sim25$ ; and since a Poisson variable's variance equals its mean, any yield $N$ carries uncertainty $\sqrt{N}$ . Why: the $\sqrt{N}$ behind every significance estimate and the $e^{-\mu}$ suppression of clean events are one statistical fact — the reason rare physics needs both high luminosity and pile-up-robust detectors.

Limiting cases

Case	Result	Meaning
Fixed target	$\mathcal{L}=\Phi\,n_T\ell$	dense thick target buys huge $\mathcal{L}$ but fixes a low $\sqrt{s}$
Collider, round beams	$\mathcal{L}=f n_b N^2/4\pi\sigma^2$	equal $\sigma_x=\sigma_y$ ; luminosity $\propto N^2$ , $\propto1/\sigma^2$
Head-on vs crossing angle	$\mathcal{L}\to\mathcal{L}\,F$ , $F<1$	a crossing angle reduces overlap; crab cavities recover $F\to1$
Low luminosity (LEP-like)	$\mu\ll1$ , clean events	one interaction per crossing, no pile-up, precision electroweak
High luminosity (HL-LHC)	$\mu\sim140$ – $200$	pile-up dominates; needs timing detectors and fine granularity
Rare process	$N=\sigma\mathcal{L}_{\text{int}}$ , error $\sqrt{N}$	discovery is limited by accumulated $\mathcal{L}_{\text{int}}$ , not by energy

Common trapThe belief that luminosity depends on the process, or that higher energy means more events

Two conflations recur. First, luminosity is not a property of the reaction you study — it is fixed by the beams and machine, and the same $\mathcal{L}$ multiplies the Higgs and inelastic cross sections alike. You do not "get more luminosity for rare processes"; you get the same $\mathcal{L}$ and a smaller $\sigma$ , so fewer events. Second, raising the beam energy does not by itself deliver more events: energy sets which processes are open and roughly how large their cross sections are, but the event count is $\sigma\times\mathcal{L}_{\text{int}}$ , so a machine can sit at the energy frontier and still starve for statistics. Energy opens the door, luminosity fills the room — and never let "brighter beam" slide into "more energetic beam": a bright beam has many closely focused particles, not fast ones.

Worked examples

1 · Building the LHC design luminosity from beam parameters (fully worked)

The nominal LHC stores $n_b=2808$ bunches of $N=1.15\times10^{11}$ protons each, at $f=11.245\ \mathrm{kHz}$ (the $26.66\ \mathrm{km}$ ring), focused to round beams $\sigma\approx16.6\ \mu\mathrm{m}$ . Find the luminosity, then the Higgs rate given $\sigma_H\approx55\ \mathrm{pb}$ .

Step 1. Assemble the beam factor. With $N_1=N_2=N$ , using $(1.15\times10^{11})^2=1.3225\times10^{22}$ and $f n_b=3.158\times10^{7}$ : $f n_b N^2=4.18\times10^{29}\ \mathrm{s^{-1}}$ .

Step 2. Divide by the overlap area. For round beams $4\pi\sigma^2=4\pi(1.66\times10^{-3}\ \mathrm{cm})^2=3.46\times10^{-5}\ \mathrm{cm^2}$ , so $\mathcal{L}=4.18\times10^{29}/3.46\times10^{-5}=\mathbf{1.2\times10^{34}\ cm^{-2}\,s^{-1}}$ before the crossing-angle factor; with $F\approx0.84$ , the design $\mathbf{1.0\times10^{34}\ cm^{-2}\,s^{-1}}$ .

Step 3. Convert to a Higgs rate with equation (1). With $\sigma_H=55\ \mathrm{pb}=5.5\times10^{-35}\ \mathrm{cm^2}$ , $R=\mathcal{L}\sigma_H=(10^{34})(5.5\times10^{-35})=\mathbf{0.55\ Hz}$ — one Higgs every two seconds. Over a running year of $10^{7}\ \mathrm{s}$ that is $\mathcal{L}_{\text{int}}=100\ \mathrm{fb^{-1}}$ and $N_H=(55\ \mathrm{pb})(100\ \mathrm{fb^{-1}})\times10^{3}=\mathbf{5.5\times10^{6}\ Higgs\ bosons}$ . The design machine makes millions of Higgs a year from a picobarn cross section purely because $\mathcal{L}_{\text{int}}$ is enormous.

2 · Event yields and statistical reach at 140 fb⁻¹

ATLAS and CMS each recorded about $\mathcal{L}_{\text{int}}=140\ \mathrm{fb^{-1}}$ in Run 2. Using $\sigma_H=55\ \mathrm{pb}$ and $\sigma_{t\bar t}=830\ \mathrm{pb}$ , find the produced yields and the fractional statistical uncertainty.

Step 1. Apply the shortcut $N=\sigma[\mathrm{pb}]\times\mathcal{L}_{\text{int}}[\mathrm{fb^{-1}}]\times10^{3}$ . For the Higgs, $N_H=55\times140\times10^{3}=\mathbf{7.7\times10^{6}}$ ; for top pairs, $N_{t\bar t}=830\times140\times10^{3}=\mathbf{1.2\times10^{8}}$ .

Step 2. These are produced counts; the observable yield in a clean channel is smaller by branching ratio and efficiency. The four-lepton "golden" channel $H\to ZZ^{*}\to4\ell$ (branching fraction $\approx1.2\times10^{-4}$ ) gives $7.7\times10^{6}\times1.2\times10^{-4}\approx\mathbf{900}$ produced events — hundreds observed, enough for a mass peak.

Step 3. Statistics are Poisson, so the fractional uncertainty is $1/\sqrt{N}$ : on the full Higgs sample $\approx\mathbf{0.04\%}$ , but on the $\sim900$ -event golden channel $1/\sqrt{900}\approx\mathbf{3\%}$ . The rare channel, not total production, sets the statistical reach — exactly why the HL-LHC aims for $3000\ \mathrm{fb^{-1}}$ to shrink that $3\%$ toward the systematic floor.

3 · Pile-up at design luminosity

At $\mathcal{L}=10^{34}\ \mathrm{cm^{-2}\,s^{-1}}$ with $\sigma_{\text{inel}}\approx80\ \mathrm{mb}$ and crossing rate $n_b f=3.16\times10^{7}\ \mathrm{Hz}$ , how many interactions accompany each crossing?

Step 1. Total inelastic rate: $R_{\text{inel}}=\mathcal{L}\sigma_{\text{inel}}=(10^{34})(80\times10^{-27}\ \mathrm{cm^2})=8\times10^{8}\ \mathrm{Hz}$ , using $80\ \mathrm{mb}=8\times10^{-26}\ \mathrm{cm^2}$ .

Step 2. Share it over crossings: $\mu=R_{\text{inel}}/(n_b f)=8\times10^{8}/3.16\times10^{7}=\mathbf{25}$ . Every interesting collision at the design LHC sits atop roughly 25 soft inelastic collisions, and the chance of a pile-up-free crossing is $e^{-25}\approx10^{-11}$ — which is why pile-up mitigation, not avoidance, is the design philosophy.

Physics of the portrait

Luminosity would be mere bookkeeping if you could not measure it absolutely, and you cannot read it off the beam currents alone — the transverse overlap is what matters, and it is invisible. The method that made luminosity a calibrated number, and the cooling that made high-luminosity antiproton beams possible, both came from one engineer at CERN.

Origins · 1 minSimon van der Meer and how this was discovered

Simon van der Meer

1925–2011 · beam physics and luminosity calibration

Simon van der Meer was a Dutch accelerator engineer whose two inventions bracket the whole subject of this article. In 1968, at CERN's Intersecting Storage Rings, he asked how one could know the absolute luminosity of two colliding beams when the transverse overlap is invisible. His answer — the van der Meer scan — was disarmingly direct: sweep one beam across the other in small steps, record the interaction rate versus separation, and read off the effective widths $\Sigma_x,\Sigma_y$ from the curve. Since $\mathcal{L}=f n_b N_1 N_2/(2\pi\Sigma_x\Sigma_y)$ and the bunch charges are measured independently, the scan delivers luminosity to about one percent without ever knowing the individual beam sizes; every LHC experiment still calibrates this way. His second invention was even more consequential: stochastic cooling, a feedback scheme that senses the statistical spread of a beam and nudges it back toward the axis, shrinking its emittance and hence — by equation (5) — its luminosity. Cooling made it possible to accumulate and collide dense antiproton beams, turning the SPS into the SppS collider on which Carlo Rubbia's team discovered the W and Z bosons in 1983. Van der Meer shared the 1984 Nobel Prize with Rubbia for that work — a rare and fitting Nobel for an accelerator physicist: without his cooling there would have been no luminosity to find them with, and without his scan no one could have said how much.

History

1961

— Touschek builds AdA at Frascati, the first electron–positron storage ring, proving beams can be collided

1968

— van der Meer proposes the beam-separation scan to calibrate luminosity absolutely at the ISR

1971

— CERN's Intersecting Storage Rings, the first proton–proton collider, reaches luminosities near 10³¹

1981

— the SppS proton–antiproton collider starts, its luminosity enabled by stochastic cooling

2010

— the LHC begins high-energy collisions, its luminosity calibrated by van der Meer scans

Going further

Texts. Thomson, Modern Particle Physics, §1.3 (collider experiments and luminosity) · Griffiths, Introduction to Elementary Particles, 2nd ed., §6.1 · Halzen & Martin, Quarks and Leptons, Ch. 4 · Wille, The Physics of Particle Accelerators, Ch. 5 · Particle Data Group, Review of Particle Physics, "Accelerator physics of colliders".

Provenance. The counting law $R=\mathcal{L}\sigma$ and fixed-target form $\mathcal{L}=\Phi n_T\ell$ are standard (Thomson §1.3; Griffiths §6.1); the Gaussian-overlap luminosity and its emittance form are the machine-design standard (Wille Ch. 5; PDG accelerator review). LHC beam parameters ( $N=1.15\times10^{11}$ , $n_b=2808$ , $f=11.245\ \mathrm{kHz}$ , $\varepsilon_n=3.75\ \mu\mathrm{m}$ , $\beta^{*}=0.55\ \mathrm{m}$ , design $\mathcal{L}=10^{34}$ ) are LHC Design Report values; cross sections ( $\sigma_H\approx55\ \mathrm{pb}$ , $\sigma_{t\bar t}\approx830\ \mathrm{pb}$ , $\sigma_{\text{inel}}\approx80\ \mathrm{mb}$ at 13 TeV) are LHC Higgs Working Group and TOTEM/ATLAS values. A verifier should confirm: (1) equations (4)–(5) reproduce $\mathcal{L}\approx10^{34}$ ; (2) $\mu=\mathcal{L}\sigma_{\text{inel}}/n_b f\approx25$ ; (3) $N=\sigma[\mathrm{pb}]\times\mathcal{L}_{\text{int}}[\mathrm{fb^{-1}}]\times10^{3}$ ; (4) the overlap integral gives $1/4\pi\sigma_x\sigma_y$ for equal round beams.

Prerequisites. Interaction cross sections · Tracking detectors · Particle–matter interactions. Leads to. Collider experiments · Jets and fragmentation · Z resonance at LEP.

Ask the tutor: Want me to walk through a van der Meer scan step by step, or work out the integrated luminosity your favourite process needs for a five-sigma discovery?

What is the difference between instantaneous and integrated luminosity?

Instantaneous luminosity $\mathcal{L}$ is a rate, in $\mathrm{cm^{-2}\,s^{-1}}$ : how many collisions per second per unit cross section the machine delivers right now, rising and falling within a fill as the beams are squeezed and deplete. Integrated luminosity $\mathcal{L}_{\text{int}}=\int\mathcal{L}\,\mathrm{d}t$ is the accumulated total, in $\mathrm{fb^{-1}}$ , and it fixes how many events you have: $N=\sigma\mathcal{L}_{\text{int}}$ . Think of $\mathcal{L}$ as the flow from a tap and $\mathcal{L}_{\text{int}}$ as the water in the bucket: an analyst cares about the integral, a machine operator about the instantaneous value.

Does the luminosity depend on which process I am studying?

No, and this is the single most important thing to internalize. Luminosity is fixed entirely by the beams and machine — bunch charge, number of bunches, revolution frequency, transverse size — and one $\mathcal{L}$ applies to every process at once. The Higgs and inelastic cross sections are multiplied by the same $\mathcal{L}$ ; they differ only in $\sigma$ . You never gain luminosity by choosing a rarer target; you just get fewer events because $\sigma$ is smaller. The factorization $R=\mathcal{L}\sigma$ — machine on one side, physics on the other — is exactly what makes luminosity so useful.

How is the absolute luminosity actually measured?

Through a van der Meer scan. You cannot read $\mathcal{L}$ off the beam currents alone because the transverse overlap is invisible, so you sweep one beam across the other in small steps and record how the interaction rate changes with separation. The width of that curve gives the convolved beam sizes $\Sigma_x,\Sigma_y$ , and with the measured bunch charges the luminosity follows from $\mathcal{L}=f n_b N_1 N_2/(2\pi\Sigma_x\Sigma_y)$ . This calibrates a reference "luminometer" rate to about one percent, after which the experiment tracks luminosity by scaling it. Van der Meer devised the method at the ISR in 1968, and every LHC experiment still uses it.

What is pile-up and why is it a problem?

Pile-up is the presence of several proton–proton collisions in the same bunch crossing. Because the inelastic cross section is large ( $\sim80\ \mathrm{mb}$ ), a high-luminosity machine produces on average $\mu=\mathcal{L}\sigma_{\text{inel}}/n_b f$ per crossing — about 25 at LHC design, up to 200 at the HL-LHC. Only one is usually the hard collision you care about; the rest deposit extra tracks and energy clusters that overlap it, degrading momentum, missing-energy, and jet measurements. Experiments fight it with finely segmented detectors that resolve the many primary vertices, and increasingly with precision timing. Pile-up is the unavoidable shadow of luminosity: the brighter the beams, the more you must dig through.

Why do we need such enormous luminosities in the first place?

Because the processes worth studying are rare. The total inelastic cross section is tens of millibarns, but Higgs production is around 55 picobarns — nine orders of magnitude smaller — and many new-physics signatures smaller still. Since $N=\sigma\mathcal{L}_{\text{int}}$ , seeing even a few thousand of a picobarn process demands many $\mathrm{fb^{-1}}$ . Worse, the uncertainty on a yield is $\sqrt{N}$ , so precision improves only as $1/\sqrt{\mathcal{L}_{\text{int}}}$ : halving your error bars means quadrupling your dataset. That scaling is why machines chase ever-higher luminosity.

Which gives more luminosity, a fixed target or a collider?

A fixed target wins on raw luminosity, often by orders of magnitude, because a solid or liquid target packs an astronomically high density of scattering centres — why fixed-target experiments dominate high-statistics, rare-decay, and neutrino physics. The catch is energy: most of the beam energy goes into centre-of-mass motion and is unavailable for making particles, so $\sqrt{s}\propto\sqrt{E_{\text{beam}}}$ . A collider sacrifices luminosity — two thin beams overlap far less than a beam hitting a brick — to put both beams' full energy into the collision, $\sqrt{s}=2E_{\text{beam}}$ . The energy frontier belongs to colliders; the intensity frontier often to fixed targets.