Parton showers & hadronisation

A quark born in a hard collision lives a strange double life. While the momentum scale is high and the strong coupling $\alpha_S$ is small, it behaves almost like a free particle that occasionally radiates a gluon — exactly as an accelerated electron radiates a photon — and this radiation is calculable with perturbative QCD. But the coupling is not constant: as the radiated partons drift to lower momentum scales, $\alpha_S$ climbs from $\sim0.12$ at the electroweak scale toward order unity near $1$ GeV, emission becomes copious, and the perturbative language collapses. What emerges is never a free quark — colour confinement forbids it — but a spray of colour-neutral hadrons collimated around the original parton directions: a jet. Two distinct pieces of physics bridge the quark and the jet. The first is the parton shower: a probabilistic cascade of $1\to2$ branchings ( $q\to qg$ , $g\to gg$ , $g\to q\bar q$ ) driven by the collinear and soft singularities of QCD, organised by the DGLAP splitting functions and resummed to all orders by the Sudakov form factor. The second is hadronisation: the genuinely non-perturbative step in which the low-scale partons are dressed into pions, kaons, and nucleons, modelled by the Lund string or the cluster mechanism because we cannot yet compute it from first principles. Together they turn a two-line Feynman diagram into the hundred-track events an LHC detector records.

Key takeaways

A high-energy parton radiates a cascade of gluons — the parton shower — governed by the DGLAP splitting functions $P_{q\to qg}$ , $P_{g\to gg}$ , $P_{g\to q\bar q}$ with colour factors $C_F=\tfrac43$ , $C_A=3$ , $T_R=\tfrac12$ .
Splitting has a collinear singularity $\mathrm{d}\theta^2/\theta^2$ and a soft singularity $\mathrm{d}E/E$ ; together they give the double logarithm $\alpha_S\ln^2$ that dominates every high-energy event.
The Sudakov form factor $\Delta(t_1,t_2)=\exp[-\int(\mathrm{d}t/t)\int\mathrm{d}z\,(\alpha_S/2\pi)P(z)]$ is the probability of no emission between two scales — it resums the logarithms and makes the shower a unitary Markov chain.
The shower stops at a cutoff $Q_0\approx1$ GeV where $\alpha_S$ becomes large; below it hadronisation takes over — a non-perturbative process modelled, not derived.
The Lund string model treats the colour field as a flux tube of tension $\kappa\approx1$ GeV/fm $\approx0.2$ GeV $^2$ (a force of $\sim14$ tonnes); the string stretches, stores energy $\kappa r$ , and snaps into new $q\bar q$ pairs.
Key idea: the shower turns one hard parton into many via calculable soft-and-collinear radiation, and hadronisation then binds the low-scale partons into the colourless jets we observe — the first step perturbative and universal, the second non-perturbative and modelled.

The core insight

A hard parton multiplies into a shower of softer partons through the calculable collinear-and-soft radiation encoded in the DGLAP splitting functions and resummed by the Sudakov form factor; below a cutoff of about $1$ GeV the coupling grows too strong to compute, and non-perturbative hadronisation binds the partons into the colour-neutral jets of hadrons that detectors record.

Where this lives today

Parton showers and hadronisation are the beating heart of collider phenomenology. Every LHC measurement — a Higgs coupling, a top mass, a supersymmetry search — is compared against events simulated by a general-purpose Monte Carlo (PYTHIA, HERWIG, SHERPA) that chains a hard matrix element to a Sudakov-ordered shower and a hadronisation model. The frontier is matching and merging: schemes like POWHEG, MC@NLO, and CKKW/MLM graft next-to-leading-order (now NNLO) matrix elements onto the shower without double-counting, so hard jets and soft collinear radiation are described accurately at once. In parallel, analytic resummation to next-to-leading-log underpins precision extractions of $\alpha_S$ , and jet substructure uses the angular-ordered, soft-drop structure of showers to tell a boosted $W$ , top, or Higgs jet from QCD background. Open problems stay non-perturbative: hadronisation is modelled not derived, colour reconnection and the underlying event carry stubborn uncertainties, and jet quenching in the heavy-ion quark–gluon plasma bridges this machinery to deconfined matter.

The physical picture

Follow a single quark outward from the hard vertex. At the highest scale it is far off-shell and sheds that virtuality by branching; each daughter is closer to on-shell and branches again until the cascade cools to the $\sim1$ GeV scale where the coupling is order one. Because both the collinear ( $\theta\to0$ ) and soft ( $E\to0$ ) regions are enhanced, most branchings hug their parent and most gluons are soft — which is why the debris stays collimated into jets rather than spraying isotropically. The first figure traces the cascade and pins the number that drives it: the running of $\alpha_S$ .

Figure

A quark and antiquark from the hard vertex each seed an angular-ordered cascade of gluons; the coupling runs from α_s ≈ 0.12 at Q ≈ M_Z = 91 GeV up to α_s ≈ 0.4 near the shower cutoff Q₀ ≈ 1 GeV, so emissions multiply as the scale falls until hadronisation binds the partons into two jets

Where does the shower end and confinement begin? There is no sharp line — the handover happens over the same $\sim1$ GeV region where $\alpha_S$ turns order-one — but every model needs a picture of what confinement does. The most successful is the Lund string: the colour field between a separating $q\bar q$ pair does not spread like a Coulomb field but collapses into a narrow flux tube of constant tension, so the potential energy grows linearly with separation. Pull hard enough and it is cheaper to snap the string by materialising a new $q\bar q$ pair than to keep stretching it — and that snapping, repeated down the string, is hadronisation. The second figure plots the linear potential and marks the tension that sets the scale of the strong force.

Figure

Linear confinement: the quark–antiquark potential rises as V(r) = κr with string tension κ ≈ 1 GeV/fm ≈ 0.2 GeV² — a constant force of about 14 tonnes. When the stored energy reaches 2m_q ≈ 0.66 GeV at r ≈ 0.66 fm the string snaps into a fresh qq̄ pair, so quarks are never freed, only re-clothed into hadrons

The core equations

The whole shower rests on one fact about gauge-theory radiation: when a parton $a$ splits into $b,c$ nearly collinearly, the cross section factorises into the hard cross section for producing $a$ times a universal branching probability. That probability carries a collinear singularity $\mathrm{d}\theta^2/\theta^2$ and a splitting function $P_{a\to bc}(z)$ depending only on the momentum fraction $z$ .

\mathrm{d}\sigma_{n+1} \simeq \mathrm{d}\sigma_n\,\frac{\alpha_S}{2\pi}\,\frac{\mathrm{d}t}{t}\,P_{a\to bc}(z)\,\mathrm{d}z \qquad (1)

where: $\mathrm{d}\sigma_n$ — cross section for the $n$ -parton state before branching; $t$ — the ordering variable, e.g. virtuality $q^2$ , transverse momentum $k_T^2$ , or angle $\theta^2$ ; $z$ — fraction of the parent's momentum taken by one daughter; $P_{a\to bc}(z)$ — DGLAP splitting function; $\alpha_S=\alpha_S(t)$ . Reading it: each extra parton costs one power of $\alpha_S$ but can win it back with a large logarithm $\int\mathrm{d}t/t=\ln$ , so the useful expansion parameter is $\alpha_S\ln$ , not $\alpha_S$ . The splitting functions are fixed by the QCD vertices and the colour algebra — the same universal functions that drive the scale evolution of parton distributions in deep inelastic scattering.

P_{q\to qg}(z)=C_F\frac{1+z^2}{1-z},\quad P_{g\to gg}(z)=2C_A\!\left[\frac{z}{1-z}+\frac{1-z}{z}+z(1-z)\right],\quad P_{g\to q\bar q}(z)=T_R\big[z^2+(1-z)^2\big] \qquad (2)

where: $C_F=\tfrac{N^2-1}{2N}=\tfrac43$ — quark–gluon colour factor; $C_A=N=3$ — gluon self-coupling factor; $T_R=\tfrac12$ — quark-loop normalisation. Reading it: $P_{q\to qg}$ blows up as $z\to1$ (soft gluon), $P_{g\to gg}$ blows up at both ends, and $P_{g\to q\bar q}$ is finite — gluons split into soft gluons far more readily than into quark pairs, so showers are gluon-rich. Combining the collinear pole of (1) with the soft ( $z\to1$ ) pole of $P_{q\to qg}$ gives the emission its characteristic double differential — one logarithm from the angle, one from the energy.

\mathrm{d}\mathcal{P}_{q\to qg}\simeq \frac{2\alpha_S C_F}{\pi}\,\frac{\mathrm{d}\theta}{\theta}\,\frac{\mathrm{d}E_g}{E_g} \qquad (3)

where: $\theta$ — opening angle of the emitted gluon; $E_g$ — its energy; $C_F=\tfrac43$ . Reading it: integrating both logarithms over a shower spanning $Q$ down to $Q_0$ produces the leading double logarithm $\sim\alpha_S\ln^2(Q/Q_0)$ . This is large ( $\alpha_S\ln^2\sim1$ ) even when $\alpha_S$ is small — precisely why fixed-order perturbation theory is not enough and the logarithms must be resummed. Resummation is done by the Sudakov form factor: the probability that a parton evolves from $t_1$ down to $t_2$ without branching. It exponentiates the single-emission rate and is the engine of every shower Monte Carlo.

\Delta(t_1,t_2)=\exp\!\left[-\int_{t_2}^{t_1}\frac{\mathrm{d}t}{t}\int_{z_-}^{z_+}\mathrm{d}z\,\frac{\alpha_S(t)}{2\pi}\,P(z)\right] \qquad (4)

where: $t_1>t_2$ — upper and lower scales; $z_\pm$ — kinematic limits regulating the soft pole; the exponent is the mean number of emissions between the two scales. Reading it: because $\Delta$ is a genuine probability ( $0<\Delta\le1$ ) the shower is unitary, and the next branching scale is generated by solving $\Delta(t_1,t)=\mathcal{R}$ for a uniform random number $\mathcal{R}$ — turning the cascade into a Markov chain. The shower runs $\alpha_S$ at each branching, and the running — recalled from aq-pp-running-alpha-s — both organises the cascade and terminates it.

\alpha_S(Q^2)=\frac{\alpha_S(M_Z^2)}{1+\dfrac{b_0}{2\pi}\,\alpha_S(M_Z^2)\,\ln(Q/M_Z)},\qquad b_0=11-\tfrac23 n_f \qquad (5)

where: $b_0$ — one-loop beta coefficient, positive because $11$ (gluon self-interaction) beats $\tfrac23 n_f$ (quark screening) — the origin of asymptotic freedom; $n_f$ — active flavours. Reading it: $\alpha_S$ shrinks at large $Q$ but grows toward $1$ as $Q\to\Lambda_{\rm QCD}\sim0.2$ GeV, forcing a cutoff $Q_0\approx1$ GeV below which perturbation theory is abandoned. Below the cutoff, hadronisation takes over. In the Lund string picture the confining dynamics is captured by a single number, the string tension, through a linear potential.

V(r)=\kappa\,r,\qquad \kappa\approx1\ \text{GeV/fm}\approx0.2\ \text{GeV}^2 \qquad (6)

where: $r$ — separation of the colour charges; $\kappa$ — string tension, the energy stored per unit length of flux tube. Reading it: a constant force (the derivative of a linear potential) means the field never weakens with distance — the hallmark of confinement — so the energy to isolate a quark is infinite; long before that, the string breaks into hadrons.

Derivation AThe collinear limit and the DGLAP splitting functions

The goal is to show why a nearly-collinear branching factorises off the hard process, producing the universal form $\mathrm{d}\sigma_{n+1}=\mathrm{d}\sigma_n\,(\alpha_S/2\pi)\,(\mathrm{d}t/t)\,P(z)\,\mathrm{d}z$ , and to expose where the collinear and soft singularities come from. We start from a single real-emission matrix element, take the small-angle limit, and read off the splitting function as the residue of the collinear pole — the object that becomes the DGLAP kernel.

Step 1. Consider $q\to q g$ attached to any hard process. The gluon propagator introduces a factor whose denominator is the virtuality of the internal line, $t=(p_q+p_g)^2\approx z(1-z)E^2\theta^2$ for small angle $\theta$ , with $z$ the quark's momentum fraction after emission. Why: the propagator $1/t$ diverges as the daughters become collinear ( $\theta\to0$ ), so the whole enhancement of QCD radiation is controlled by this single virtuality.

Step 2. The squared matrix element, integrated over azimuth and transverse momentum, organises into the hard piece times $\dfrac{\alpha_S}{2\pi}\dfrac{\mathrm{d}t}{t}\,P_{q\to qg}(z)\,\mathrm{d}z$ , with $P_{q\to qg}(z)=C_F\dfrac{1+z^2}{1-z}$ , and $\mathrm{d}t/t$ integrates to a logarithm. Why: this factorisation states that a collinear parton "doesn't know" about the hard process that made it — the branching is universal, the same for any production channel.

Step 3. Locate the two singularities: the collinear pole $\mathrm{d}t/t\to\mathrm{d}\theta^2/\theta^2$ as $\theta\to0$ , and the soft pole $1/(1-z)$ in $P_{q\to qg}$ as $z\to1$ (gluon energy $E_g=(1-z)E\to0$ ). Both are logarithmic, and their product is the double logarithm of (3). Why: the shower exists to organise these two logarithms, because each factor of $\alpha_S$ they multiply is compensated by a large $\ln$ and must be summed to all orders.

Step 4. The divergences are not physical: real and virtual contributions cancel for infrared-and-collinear-safe observables (the Kinoshita–Lee–Nauenberg theorem), and the Sudakov factor of (4) is the resummed virtual piece taming the real-emission poles. Why: this makes the shower predictive — individual branchings are singular and unobservable, but the resummed sum gives finite cross sections for well-defined observables.

Go deeper · Derivation B · 1 minThe Sudakov form factor as an exponentiated no-emission probability

Derivation BThe Sudakov form factor as an exponentiated no-emission probability

The aim is to derive the Sudakov form factor $\Delta(t_1,t_2)$ — the probability that a parton falls from scale $t_1$ to $t_2$ emitting nothing — and to show it exponentiates the single-branching rate, making the shower a unitary Markov process whose branching scales can be sampled with a random number. We treat branchings as independent events in the evolution variable and integrate the survival probability, the same logic as radioactive decay.

Step 1. Write the differential branching probability per unit evolution from (1) as $\mathrm{d}\mathcal{P}=\dfrac{\mathrm{d}t}{t}\displaystyle\int\mathrm{d}z\,\dfrac{\alpha_S}{2\pi}P(z)\equiv\dfrac{\mathrm{d}t}{t}\,\bar P(t)$ , the mean number of branchings in $[t,t+\mathrm{d}t]$ . Why: casting the branching as a rate per unit "evolution time" $\ln t$ lets us treat the cascade as independent random events.

Step 2. Demand the parton survive every slice between $t_1$ and $t_2$ without branching. The survival probability is the product $\prod(1-\mathrm{d}\mathcal{P})$ , which in the continuum limit becomes $\Delta(t_1,t_2)=\exp\!\big[-\int_{t_2}^{t_1}(\mathrm{d}t/t)\,\bar P(t)\big]$ . Why: identical mathematics to the survival probability $e^{-\int\Gamma\,\mathrm{d}t}$ of an unstable state — non-emission is "nothing happened yet" over every infinitesimal step.

Step 3. Check unitarity: the probability that the first branching happens at $t$ is $-\mathrm{d}\Delta(t_1,t)/\mathrm{d}t=\Delta(t_1,t)\,\bar P(t)/t$ , and integrating over all $t$ plus the no-branch probability $\Delta(t_1,t_0)$ gives exactly $1$ . Why: unitarity means the shower conserves probability — redistributing the cross section among final states rather than changing the total — which lets it dress a fixed-order calculation without double-counting.

Step 4. Turn it into an algorithm: given a uniform random number $\mathcal{R}\in(0,1)$ , solve $\Delta(t_1,t)=\mathcal{R}$ for the next branching scale $t$ ; if it falls below the cutoff $t_0=Q_0^2$ , stop and hand the parton to hadronisation. Why: this is literally how PYTHIA, HERWIG, and SHERPA generate events — the Sudakov factor is the line of code that decides when the next gluon comes out.

Go deeper · Derivation C · 1 minLinear confinement and string breaking in the Lund model

Derivation CLinear confinement and string breaking in the Lund model

The purpose is to connect the linear potential $V(r)=\kappa r$ to the actual mechanism of hadron production — the breaking of a colour string by quantum pair creation — and to motivate the Lund symmetric fragmentation function. We take the flux-tube picture as given (its justification is lattice QCD), compute when breaking becomes favourable, and borrow the Schwinger tunnelling result for the transverse-momentum and flavour spectra.

Step 1. Model the colour field of a separating $q\bar q$ pair as a one-dimensional flux tube of uniform energy density $\kappa$ . The stored energy is $V(r)=\kappa r$ , giving a constant attractive force $F=-\mathrm{d}V/\mathrm{d}r=-\kappa$ . Why: a constant restoring force is the essence of confinement — unlike a Coulomb field it never dies off, so the cost of full separation is infinite and free quarks cannot exist.

Step 2. As $r$ grows the string stores more energy; once $\kappa r$ exceeds roughly $2m_{q,\perp}$ it can break by materialising a new $q\bar q$ pair, converting field energy into two shorter strings — two hadrons-in-waiting. For light quarks ( $m\sim0.3$ GeV) this happens at $r\sim0.66$ fm, the size of a hadron. Why: this is hadronisation in one line — the string doesn't stretch to infinity, it repeatedly snaps at the femtometre scale.

Step 3. The pair is created by quantum tunnelling through the linear potential, so the probability carries a Schwinger-like Gaussian suppression $\mathrm{d}\mathcal{P}\propto\exp(-\pi m_\perp^2/\kappa)$ with $m_\perp^2=m^2+p_\perp^2$ . This predicts both a limited, roughly Gaussian $p_\perp$ spectrum and a strangeness suppression — heavier $s\bar s$ pairs made less often than $u\bar u$ or $d\bar d$ . Why: one tunnelling formula explains two measured facts — hadrons with $p_\perp\lesssim0.4$ GeV off the string axis, and kaons rarer than pions.

Step 4. Assign the longitudinal momentum with the Lund symmetric fragmentation function $f(z)\propto z^{-1}(1-z)^a\exp(-b\,m_\perp^2/z)$ , the unique choice for which the result does not depend on which end of the string you fragment first. Why: demanding left–right symmetry fixes the functional form up to two tuned parameters $a,b$ , giving real predictive power rather than freedom.

Limiting cases

Case	Result	Meaning
Soft limit $z\to1$	$P_{q\to qg}\to\dfrac{2C_F}{1-z}$	infrared divergence — soft gluons emitted copiously; regulated by $z_+$ and by virtual/Sudakov cancellation
Collinear limit $\theta\to0$	$\mathrm{d}\sigma\propto\dfrac{\mathrm{d}\theta^2}{\theta^2}$	branchings hug the parent direction — this is why radiation collimates into jets
High scale $Q\gg\Lambda_{\rm QCD}$	$\alpha_S\ll1$ , $\Delta\to1$	perturbative shower; few, calculable emissions; the domain of matrix-element accuracy
Cutoff scale $Q\to Q_0\sim1$ GeV	$\alpha_S\sim0.4$ – $0.5$	shower stops; hadronisation model takes over; boundary is a convention, not physics
Large $N_c$	colour flow is planar; each gluon = $q\bar q$ colour line	justifies the string/cluster picture — every parton sits on a unique colour chain
Successive emissions	angular ordering $\theta_1>\theta_2>\cdots$	colour coherence: soft gluons see only net colour charge, suppressing wide-angle radiation

Common trapThe parton shower is a real, unique sequence of physical particle emissions.

It is not. The shower is a probabilistic, leading-logarithm approximation whose internal branchings are neither unique nor directly observable. Three things are conventions, not facts of nature: the ordering variable (virtuality, $k_T$ , or angle — no choice is "correct"), the cutoff $Q_0$ (moving it just shifts work between shower and hadronisation), and the individual histories (a given final state can be reached by many branching trees, and quantum mechanics forbids asking which "really" happened). What is predicted, and what you may compare with data, are infrared-and-collinear-safe observables — jet rates, thrust, event shapes — insensitive to whether a parton emits an extra soft or collinear gluon. Treating the shower as a filmstrip of real particles is the same error as treating individual terms of a perturbative series as physical.

Worked examples

1 · The string tension, in tonnes and femtometres (fully worked)

The Lund string has tension $\kappa\approx1$ GeV/fm. Express it as a force in everyday units, and estimate how far a string can stretch before it snaps into a light $q\bar q$ pair (constituent mass $m_q\approx0.33$ GeV).

Step 1. A tension is an energy per unit length — a force. With $1\ \text{GeV}=1.602\times10^{-10}$ J and $1\ \text{fm}=10^{-15}$ m,

\kappa=1\ \frac{\text{GeV}}{\text{fm}}=\frac{1.602\times10^{-10}\ \text{J}}{10^{-15}\ \text{m}}=1.60\times10^{5}\ \text{N}.

Dividing by $g=9.81\ \text{m/s}^2$ gives $\approx1.6\times10^4$ kg: the force confining a quark is about 14 tonnes, constant at all separations.

Step 2. In natural units (for lattice cross-checks), multiply by $\hbar c=0.197$ GeV·fm: $\kappa=(1\ \text{GeV/fm})(0.197\ \text{GeV}\,\cdot\,\text{fm})=\mathbf{0.197\ GeV^2}\approx0.2$ GeV $^2$ , the standard value.

Step 3. The string breaks once its stored energy $\kappa r$ can supply the rest energy of a new pair, $\kappa r\gtrsim2m_q$ . Solving,

r\approx\frac{2m_q}{\kappa}=\frac{2(0.33\ \text{GeV})}{1\ \text{GeV/fm}}=\mathbf{0.66\ fm}.

This is exactly the femtometre scale of a hadron — the string never stretches macroscopically; it fragments at the size of the hadrons it produces, which is why isolated quarks are never seen.

2 · The coupling across the shower, and why it must stop (fully worked)

Using the one-loop running (5) with $\alpha_S(M_Z)=0.118$ at $M_Z=91.2$ GeV, estimate $\alpha_S$ at $10$ GeV and near the shower cutoff $1$ GeV. What does the result say about where the shower ends?

Step 1. With $n_f=5$ , $b_0=11-\tfrac23(5)=\tfrac{23}{3}=7.67$ , so $\tfrac{b_0}{2\pi}=1.22$ . Invert (5): $\dfrac{1}{\alpha_S(Q)}=8.47+1.22\ln\!\dfrac{Q}{91.2}$ .

Step 2. At $Q=10$ GeV: $\ln(10/91.2)=-2.21$ , so $1/\alpha_S=8.47-2.70=5.77$ and $\alpha_S(10\ \text{GeV})=\mathbf{0.173}$ — already half again its $M_Z$ value; emissions get cheaper as the cascade descends.

Step 3. Continuing toward $Q=1$ GeV (crossing the $b$ and $c$ thresholds, where $n_f$ drops to $3$ and the running steepens) gives $\alpha_S(1\ \text{GeV})\approx\mathbf{0.4}$ – $\mathbf{0.5}$ — of order unity. The perturbative shower is now meaningless and the partons must be handed to a hadronisation model. The cutoff $Q_0\approx1$ GeV is not arbitrary: it is set by where asymptotic freedom runs out.

3 · How often does a hard quark emit nothing? (Sudakov estimate)

Estimate the probability that a quark evolves from $Q=100$ GeV to $Q_0=1$ GeV with no gluon emission, using the leading double logarithm $\Delta\approx\exp[-\tfrac{\alpha_S C_F}{2\pi}\ln^2(Q^2/Q_0^2)]$ .

Step 1. The logarithm is $\ln(Q^2/Q_0^2)=\ln(10^4)=9.21$ , so $\ln^2=84.8$ . With $\alpha_S\approx0.12$ and $C_F=\tfrac43$ the coefficient is $\tfrac{0.12\times1.333}{2\pi}=0.0255$ , giving exponent $0.0255\times84.8=2.16$ .

Step 2. Hence $\Delta\approx e^{-2.16}=\mathbf{0.12}$ — only a 1-in-9 chance of no emission at all. A high-energy quark almost always radiates, usually many times, which is why one hard parton materialises as a multi-particle jet, not a lone track. (The crude estimate overstates the suppression, but emission is the rule.)

Physics of the portrait

The parton shower plus string hadronisation is not just a theory but running software every collider physicist depends on — and its longest-serving architect turned the equations above into the generator the LHC was designed around.

Origins · 1 minTorbjörn Sjöstrand and how this was discovered

Torbjörn Sjöstrand

b. 1957 · parton showers and Lund string hadronisation (PYTHIA)

Torbjörn Sjöstrand, more than any other single person, made the abstract machinery of parton showers and hadronisation into a working tool. As a young researcher in the Lund group around 1980 he helped build the string fragmentation model — the 1983 paper of Andersson, Gustafson, Ingelman and Sjöstrand remains among the most cited in the field — and wrote the code, JETSET and later PYTHIA, that implemented it. PYTHIA welds the hard matrix element, a Sudakov-driven parton shower, the Lund string, and hadron decays into a single event generator, and for four decades it has been the default simulation of essentially every $e^+e^-$ , $ep$ , and hadron-collider experiment, including ATLAS and CMS. His deeper contributions run through the whole chain: dipole and $p_T$ -ordered showers, multiple parton interactions and the underlying event, and colour reconnection. The lesson of his career is that in QCD the gap between a correct formula and a usable prediction is enormous, and closing it is itself first-rate physics. When an LHC background is quoted "from Monte Carlo," that Monte Carlo is, more often than not, his.

History

1974

— Wilson's lattice formulation of QCD demonstrates linear quark confinement

1977

— Altarelli and Parisi recast scaling violation as parton splitting P(z)

1979

— three-jet events at PETRA (TASSO, MARK-J, PLUTO, JADE) reveal the gluon

1983

— Andersson, Gustafson, Ingelman and Sjöstrand publish the Lund string model

1988

— coherent angular-ordered showering is built into the HERWIG generator

Going further

Texts. Ellis, Stirling & Webber, QCD and Collider Physics, Ch. 5–6 (parton showers, fragmentation) · Peskin & Schroeder, An Introduction to QFT, §17 (QCD, DGLAP) · Halzen & Martin, Quarks and Leptons, Ch. 10–11 · Thomson, Modern Particle Physics, §10.5–10.6 · Schwartz, QFT and the Standard Model, §32 (infrared divergences) · Andersson, The Lund Model (Cambridge) · Griffiths, Introduction to Elementary Particles, §9. Key numerical checks: $\kappa=1$ GeV/fm $=1.60\times10^5$ N $\approx14$ tonnes $=0.197$ GeV $^2$ ; string break length $2m_q/\kappa\approx0.66$ fm; $\alpha_S(10\ \text{GeV})\approx0.17$ and $\alpha_S(1\ \text{GeV})\approx0.4$ – $0.5$ from one-loop running; the leading-double-log Sudakov $\Delta\approx e^{-2.16}\approx0.12$ for $Q=100$ GeV, $Q_0=1$ GeV. Splitting-function forms follow Ellis–Stirling–Webber; colour factors are exact SU(3) values.

Prerequisites. Running of $\alpha_S$ · Asymptotic Freedom · $e^+e^-\to\mu^+\mu^-$ · Colour in QCD · QCD Vertices. Leads to. Jets & Fragmentation · Hadron Collisions · Running of $\alpha_S$ .

Ask the tutor: Want to see how the Sudakov form factor becomes the random-number algorithm PYTHIA uses to generate the next gluon?

What is the difference between a parton shower and a fixed-order calculation?

A fixed-order calculation evaluates the matrix element for a definite number of partons to a given power of $\alpha_S$ . It is accurate for hard, well-separated emissions but breaks down when partons become soft or collinear, because each such configuration carries a large logarithm that a finite-order series cannot sum. The parton shower does the opposite: it approximates the matrix element in exactly the soft-and-collinear region but resums the leading logarithms to all orders through the Sudakov form factor, generating an arbitrary number of emissions. Neither is complete alone — the shower misses hard, wide-angle radiation and the fixed-order result misses the resummation — which is why modern predictions match the two together (POWHEG, MC@NLO).

Why does the shower need a cutoff Q₀, and does its value matter?

The shower must stop because $\alpha_S$ grows toward $1$ as the scale falls, so the perturbative expansion in $\alpha_S\ln$ loses meaning below about $1$ GeV. The cutoff $Q_0\approx1$ GeV marks where you hand the partons to a hadronisation model, and its precise value is a convention: lower it and the shower does more of the work, raise it and the model does more. Because physical observables cannot depend on where you draw an artificial line, the two stages are tuned together so predictions are stable against reasonable changes in $Q_0$ . A residual $Q_0$ -dependence is a genuine theoretical uncertainty.

What is angular ordering, and why does it happen?

Angular ordering means successive emissions have decreasing opening angles, $\theta_1>\theta_2>\cdots$ . It is a consequence of colour coherence — quantum interference between amplitudes where a soft gluon is emitted from different partons. A long-wavelength gluon cannot resolve the small separation between a recently-branched pair; it sees only their combined colour charge, so it can only be radiated at angles larger than the branching that produced them. Wide-angle soft radiation is thereby suppressed and the cascade fills a cone of shrinking angles. Getting coherence right (HERWIG's angular-ordered shower, PYTHIA's vetoed $k_T$ -ordered shower) is essential to describe particle multiplicities and jet shapes.

The Lund string and the cluster model give similar results — is either "correct"?

Neither is derived from QCD; both are physically-motivated models constrained by lattice input and tuned to data, and both work well. The Lund string treats confinement as a linear flux tube that fragments by $q\bar q$ pair creation (PYTHIA); the cluster model exploits preconfinement — after the shower, colour-connected partons form low-mass clusters that decay into hadrons (HERWIG). They differ in detail — the string predicts specific $p_\perp$ and flavour correlations, the cluster model has fewer parameters — and comparing where they disagree estimates hadronisation uncertainty. That two different pictures both fit reflects local parton-hadron duality: gross hadronic distributions follow the parton-level ones.

What is local parton-hadron duality?

Local parton-hadron duality (LPHD) is the empirical observation that hadron distributions in a jet closely track the parton distributions computed at the shower cutoff, up to an overall normalisation — hadronisation reshuffles momentum only locally. Its most striking success is the hump-backed plateau: plot the hadron count versus $\xi=\ln(1/x)$ and you get not a monotonic rise but a hump peaked at intermediate $\xi$ , because coherence (angular ordering) suppresses very soft particles. Perturbative QCD predicts the peak position and its energy dependence, and measured spectra follow the parton prediction with only a constant rescaling — direct evidence that the perturbative cascade, not the details of confinement, sets the shape.

Why can't hadronisation be computed from first principles in QCD?

Because it lives at momentum scales around $\Lambda_{\rm QCD}\sim0.2$ GeV where $\alpha_S$ is of order one, so the perturbative expansion — the only systematic analytic tool — does not converge. Lattice QCD computes static quantities like the string tension and the hadron spectrum beautifully, but the real-time, many-body dynamics of a stretching, breaking colour field turning into dozens of hadrons is far beyond current lattice methods. So hadronisation is bridged by models whose forms are motivated by QCD symmetries and lattice input but whose parameters are fitted to data. Deriving it from the QCD Lagrangian remains a genuine open problem.