Mandelstam variables

Every two-body collision in particle physics — an electron meeting a positron, a proton striking a proton, a photon scattering off a quark — is a story told in four four-momenta: $p_1$ and $p_2$ coming in, $p_3$ and $p_4$ going out. The trouble is that those sixteen energy and momentum components look completely different in the laboratory, in the centre-of-mass frame, and in any boosted frame in between. Mandelstam's insight, borrowed from the deep fact that four-vector scalar products are Lorentz invariant, was to package all the kinematics into just three numbers that every observer agrees on: $s=(p_1+p_2)^2$ , $t=(p_1-p_3)^2$ , and $u=(p_1-p_4)^2$ . Compute them once, in whatever frame is convenient, and they are fixed for good. Three consequences organize this entire article. First, the three variables are not independent: four-momentum conservation forces $s+t+u=m_1^2+m_2^2+m_3^2+m_4^2$ , so only two of them can be chosen freely, and a full $2\to2$ process lives on a two-dimensional plane. Second, each variable is the squared four-momentum $q^2$ carried by the virtual particle in one of the three single-exchange Feynman diagrams — the $s$ -channel annihilation, the $t$ -channel exchange, and the crossed $u$ -channel — which is why cross-section formulas are written in $s$ , $t$ , and $u$ rather than in lab energies. Third, $\sqrt{s}$ is precisely the total energy available in the centre-of-mass frame, the number that decides whether a collision has enough reach to create a new particle. The same analytic function of $s$ , $t$ , and $u$ describes three physically distinct reactions at once — the content of crossing symmetry — which is where the story reaches all the way to the modern scattering-amplitudes program.

Key takeaways

For any $2\to2$ process $1+2\to3+4$ there are three Lorentz-invariant Mandelstam variables: $s=(p_1+p_2)^2$ , $t=(p_1-p_3)^2$ , $u=(p_1-p_4)^2$ .
They are built from four-vector scalar products, so they are manifestly frame independent — compute them once and every observer agrees.
They are not independent: four-momentum conservation pins them to $s+t+u=m_1^2+m_2^2+m_3^2+m_4^2$ .
$\sqrt{s}$ is the total centre-of-mass energy — the number that sets the mass reach of a collision — while $t$ and $u$ measure squared momentum transfer and are negative for physical scattering.
Each variable is the propagator's $q^2$ in the $s$ -, $t$ -, or $u$ -channel diagram; the $u$ -channel is only physically distinct when two final particles are identical.
Key idea: three frame-independent numbers — $s$ , $t$ , $u$ — capture all the kinematics of a two-to-two collision, one of them is the squared centre-of-mass energy, and each is the squared momentum carried by the exchanged particle in a corresponding Feynman diagram.

The core insight

Three frame-independent numbers $s$ , $t$ , $u$ capture all the kinematics of a two-to-two collision; they obey the single constraint $s+t+u=\sum_i m_i^2$ , $\sqrt{s}$ is the total centre-of-mass energy, and each variable is the squared four-momentum of the exchanged particle in a corresponding Feynman diagram.

Where this lives today

Mandelstam variables are the everyday language of every collider analysis. A cross section is quoted as $\mathrm{d}\sigma/\mathrm{d}t$ or as a function of $s$ ; the $s$ -channel resonance shape $1/(s-M^2)$ is exactly what LEP scanned across the $Z$ pole at $\sqrt{s}\approx91$ GeV to weigh the $Z$ boson and count the number of light neutrino species, and what the LHC now scans for new heavy states. Forward-peaked $t$ -channel exchange is what makes Bhabha scattering the luminosity monitor at $e^+e^-$ machines and what dominates elastic proton–proton scattering measured by TOTEM and ATLAS-ALFA. Three-body decays are displayed on Dalitz plots whose axes are Mandelstam-like invariant masses $m_{ij}^2=(p_i+p_j)^2$ , and modern amplitude analyses at LHCb and BaBar/Belle fit resonance structure directly in these variables. At the theory frontier, the same analyticity and crossing that Mandelstam pioneered underpin the "scattering amplitudes" revolution — on-shell recursion, the modern S-matrix bootstrap, and dispersive bounds on effective field theories — all of which are formulated on the $s$ – $t$ – $u$ plane he drew. The LHC's Run 3 now collides protons at $\sqrt{s}=13.6$ TeV, and the number that headlines every such machine is, quite literally, $\sqrt{s}$ .

The physical picture

Picture the collision not as particles moving through space but as a diagram: two lines enter, something virtual is exchanged, two lines leave. There are exactly three ways to wire two incoming and two outgoing lines through a single internal propagator, and each way defines one Mandelstam variable as the squared four-momentum flowing through that propagator. In the $s$ -channel the two incoming particles fuse into one virtual object that later splits — annihilation — and $s$ is that object's squared mass. In the $t$ -channel a quantum is tossed between the two lines as they pass, and $t$ is the squared four-momentum of the toss. The $u$ -channel is the same exchange with the two final particles swapped, and it becomes a genuinely separate diagram only when those final particles are identical. The first figure lays the three topologies side by side.

Figure

The three single-exchange topologies of a two-to-two collision. The virtual particle carries squared four-momentum q² equal to s, t or u respectively; in the massless high-energy limit t = −(s divided by 2)(1 − cos θ) and u = −(s divided by 2)(1 + cos θ), so the three invariants sum to zero.

Because the three variables carry a single constraint, $s+t+u=\sum_i m_i^2$ , a whole collision process does not need three numbers — two suffice, and the process traces out a region in a two-dimensional plane. Fixing $\sqrt{s}$ (the collision energy) and sweeping the centre-of-mass scattering angle $\theta^*$ walks a point along a line of constant $s$ : at $\theta^*=0$ the momentum transfer vanishes ( $t=0$ , grazing forward scattering) and at $\theta^*=180^\circ$ it is maximal. The second figure plots the invariants for a concrete case — muon-pair production at a $\sqrt{s}=10$ GeV collider — and shows the sum rule holding at every angle.

Figure

Mandelstam invariants versus centre-of-mass scattering angle for e⁺e⁻ → µ⁺µ⁻ at √s = 10 GeV so s = 100 GeV². As θ sweeps forward to backward, t runs from 0 down to −s and u from −s up to 0, while their sum stays pinned so that s + t + u = 0; at θ = 60° the invariants are t = −25 GeV² and u = −75 GeV².

The core equations

The three invariants are defined directly from the external four-momenta as squared four-vectors, using the metric convention in which $p^2=E^2-|\vec p|^2=m^2$ for a physical particle. Nothing here depends on a frame — that is the whole point.

s=(p_1+p_2)^2,\qquad t=(p_1-p_3)^2,\qquad u=(p_1-p_4)^2 \qquad (1)

where: $p_1,p_2$ — incoming four-momenta; $p_3,p_4$ — outgoing four-momenta; the square means the Lorentz scalar $p\cdot p$ . Each is a scalar product of four-vectors, hence invariant under any boost or rotation. Read $s$ as "annihilation energy squared", $t$ and $u$ as "squared momentum transfer". Conservation of four-momentum, $p_1+p_2=p_3+p_4$ , immediately supplies a second, equally valid form for each variable in terms of the other pair of particles. These identities are used constantly to switch between initial- and final-state expressions.

s=(p_3+p_4)^2,\qquad t=(p_2-p_4)^2,\qquad u=(p_2-p_3)^2 \qquad (2)

where: the right-hand sides follow by substituting $p_1+p_2=p_3+p_4$ . Reading: $s$ is symmetric between the initial and final pair; $t$ links particle 1 to 3 (and 2 to 4); $u$ links 1 to 4 (and 2 to 3). The three variables are not independent. Adding the three definitions and using conservation collapses them onto a single constraint — the sum rule — leaving only two free invariants for a full $2\to2$ process.

s+t+u=m_1^2+m_2^2+m_3^2+m_4^2 \qquad (3)

where: $m_i$ — rest mass of external particle $i$ . The right-hand side is a fixed number for a given reaction, so once any two invariants are known the third is determined. For all-massless or ultra-relativistic scattering the sum is simply zero. In the centre-of-mass frame the spatial momenta cancel, so $\sqrt{s}$ is exactly the total energy, and the common momentum of each particle follows from the Källén (triangle) function $\lambda$ . This is the number that decides what a collision can create.

\sqrt{s}=E_1^*+E_2^*=E_{\text{cm}},\qquad p^*=\frac{\sqrt{\lambda(s,m_1^2,m_2^2)}}{2\sqrt{s}},\qquad \lambda(x,y,z)=x^2+y^2+z^2-2xy-2yz-2zx \qquad (4)

where: $E_i^*$ — energy of particle $i$ in the centre-of-mass frame; $p^*$ — magnitude of each incoming momentum there; $\lambda$ — Källén function. Reading: a real physical collision requires $s\ge(m_1+m_2)^2$ (the threshold), which is exactly where $\lambda\ge0$ and $p^*$ is real. With $\sqrt{s}$ and the centre-of-mass scattering angle $\theta^*$ chosen, the remaining invariants follow. In the high-energy limit where all masses are negligible the relations are especially clean, and they make the sum rule visible at a glance.

t=-\frac{s}{2}\left(1-\cos\theta^*\right)=-s\,\sin^2\!\frac{\theta^*}{2},\qquad u=-\frac{s}{2}\left(1+\cos\theta^*\right)=-s\,\cos^2\!\frac{\theta^*}{2} \qquad (5)

where: $\theta^*$ — scattering angle in the centre-of-mass frame (massless limit). Reading: $t$ and $u$ are both $\le0$ for physical scattering; forward scattering ( $\theta^*\to0$ ) gives $t\to0$ , and $t+u=-s$ reproduces $s+t+u=0$ . Finally, the collision energy $\sqrt{s}$ is delivered very differently by a fixed-target machine and by a collider — the reason every high-energy frontier machine collides two beams head-on rather than firing one at a wall.

s_{\text{fixed}}=m_1^2+m_2^2+2m_2E_{\text{lab}}\;\xrightarrow{\ \text{high }E\ }\;2m_2E_{\text{lab}},\qquad s_{\text{coll}}=(E_1+E_2)^2\;\xrightarrow{\ \text{symmetric}\ }\;4E^2 \qquad (6)

where: $E_{\text{lab}}$ — beam energy hitting a target of mass $m_2$ at rest; $E$ — energy of each beam in a symmetric collider. Reading: fixed-target $\sqrt{s}$ grows only as $\sqrt{E_{\text{lab}}}$ , but collider $\sqrt{s}=2E$ grows linearly — a decisive advantage at high energy.

Derivation AThe invariant sum rule

The goal is to prove that the three Mandelstam variables are bound by one exact constraint, $s+t+u=m_1^2+m_2^2+m_3^2+m_4^2$ , using nothing but the definitions and four-momentum conservation. Establishing this shows that a two-to-two process has only two independent invariants, explains why cross sections are drawn on a two-dimensional plane, and gives the single most useful consistency check in all of relativistic kinematics — if your $s$ , $t$ , $u$ do not add to the sum of squared masses, you have made an error.

Step 1. Write each variable expanded, using $p_i^2=m_i^2$ : $s=(p_1+p_2)^2=m_1^2+m_2^2+2\,p_1\!\cdot\!p_2$ , $t=(p_1-p_3)^2=m_1^2+m_3^2-2\,p_1\!\cdot\!p_3$ , and $u=(p_1-p_4)^2=m_1^2+m_4^2-2\,p_1\!\cdot\!p_4$ . Why: expanding each squared four-vector separates the fixed mass terms from the scalar products that must be made to cancel.

Step 2. Add the three lines. The mass terms give $3m_1^2+m_2^2+m_3^2+m_4^2$ , and the scalar products collect into $2\,p_1\!\cdot\!(p_2-p_3-p_4)$ . Why: grouping all the $p_1\!\cdot\!p_j$ terms exposes a single combination of the other momenta that conservation will simplify.

Step 3. Impose conservation $p_1+p_2=p_3+p_4$ , which rearranges to $p_2-p_3-p_4=-p_1$ . Substituting gives $2\,p_1\!\cdot\!(-p_1)=-2m_1^2$ . Why: this is the only physics input beyond the definitions — momentum balance turns the leftover cross terms into a pure mass term.

Step 4. Combine: the $3m_1^2$ from the mass terms and the $-2m_1^2$ from the cross terms leave $m_1^2$ , so $s+t+u=m_1^2+m_2^2+m_3^2+m_4^2$ . Why: the extra copies of $m_1^2$ cancel cleanly, leaving the symmetric result that treats all four particles on the same footing even though the derivation singled out particle 1. *As a concrete check: for e⁺e⁻ → µ⁺µ⁻ at √s = 10 GeV, θ = 60°, the sum s + t + u = 100 − 25 − 75 = 0 = Σmᵢ² in the massless limit.

Go deeper · Derivation B · 1 minWhy √s is the centre-of-mass energy and where the threshold sits

Derivation BWhy √s is the centre-of-mass energy and where the threshold sits

The aim is to show that $\sqrt{s}$ is literally the total energy in the centre-of-mass frame and that demanding this energy be real and above rest mass produces the reaction threshold $s\ge(m_1+m_2)^2$ . This is what makes $s$ the single most important invariant in collider design: it is the budget of energy available to manufacture new mass, and the derivation makes precise the intuition that you cannot create a particle heavier than $\sqrt{s}$ .

Step 1. Evaluate $s=(p_1+p_2)^2$ in the frame where the total three-momentum vanishes, $\vec p_1^*+\vec p_2^*=0$ . Then $p_1+p_2=(E_1^*+E_2^*,\vec 0)$ , so $s=(E_1^*+E_2^*)^2$ and $\sqrt{s}=E_1^*+E_2^*=E_{\text{cm}}$ . Why: because $s$ is invariant we may compute it in the most convenient frame, and in the centre-of-mass frame the spatial part drops out and leaves a pure energy.

Step 2. Solve for the common momentum. With $\vec p_1^*=-\vec p_2^*\equiv\vec p^*$ and $E_i^*=\sqrt{m_i^2+p^{*2}}$ , imposing $E_1^*+E_2^*=\sqrt{s}$ and eliminating the energies gives $p^*=\sqrt{\lambda(s,m_1^2,m_2^2)}/(2\sqrt{s})$ with the Källén function $\lambda$ . Why: this packages the two energy constraints into one tidy expression whose reality condition is the physics we are after.

Step 3. Read off the threshold. The momentum $p^*$ is real only when $\lambda(s,m_1^2,m_2^2)\ge0$ , i.e. $s\ge(m_1+m_2)^2$ ; more generally to produce a final state of total mass $M_f$ one needs $s\ge M_f^2$ . Why: a negative $\lambda$ would give an imaginary momentum, which is unphysical, so the boundary $\lambda=0$ is exactly the minimum energy at which the reaction can proceed.

Step 4. Interpret. The lightest a final state can be is the sum of its particles' rest masses, so the minimum $\sqrt{s}$ to create particles of total mass $M_f$ is $\sqrt{s}=M_f$ ; any excess appears as final-state kinetic energy. Why: this is the working rule behind every "we need a machine of energy X to make particle Y" statement in the field. *As a concrete check: p̄ production p + p → p + p + p + p̄ needs √s ≥ 4mₚ, giving a fixed-target beam energy of exactly 7mₚ (worked below).*

Go deeper · Derivation C · 1 mint and u as squared momentum transfer and the scattering angle

Derivation Ct and u as squared momentum transfer and the scattering angle

The purpose here is to connect the abstract invariant $t$ to the concrete picture of a particle being deflected — to show that $-t$ is the squared momentum transferred in the collision and that, in the centre-of-mass frame, $t$ and $u$ are fixed by the scattering angle $\theta^*$ through $t=-\tfrac{s}{2}(1-\cos\theta^*)$ in the high-energy limit. This is what lets experimenters read a differential cross section $\mathrm{d}\sigma/\mathrm{d}t$ as an angular distribution and see forward ( $t$ -channel) peaks for what they are.

Step 1. Expand $t=(p_1-p_3)^2=m_1^2+m_3^2-2\,p_1\!\cdot\!p_3$ and evaluate the dot product in the centre-of-mass frame, where $p_1\!\cdot\!p_3=E_1^*E_3^*-|\vec p_1^*|\,|\vec p_3^*|\cos\theta^*$ and $\theta^*$ is the angle between the incoming and outgoing particle 1/3. Why: $t$ pairs the "same" particle before and after (1 with 3), so its dot product naturally carries the deflection angle.

Step 2. Take the high-energy limit where all masses are negligible, so $E^*=|\vec p^*|=\tfrac12\sqrt{s}$ for every leg. Then $p_1\!\cdot\!p_3=\tfrac{s}{4}(1-\cos\theta^*)$ and $t=-2\,p_1\!\cdot\!p_3=-\tfrac{s}{2}(1-\cos\theta^*)=-s\sin^2(\theta^*/2)$ . Why: dropping the masses isolates the pure angular dependence and yields the clean half-angle form used throughout QED and QCD scattering.

Step 3. Repeat for $u=(p_1-p_4)^2$ , which pairs particle 1 with 4 (scattered through the supplementary angle), giving $u=-\tfrac{s}{2}(1+\cos\theta^*)=-s\cos^2(\theta^*/2)$ , and confirm $t+u=-s$ . Why: $u$ is the same construction rotated by $\theta^*\to\pi-\theta^*$ , so it grows as $t$ shrinks and their sum reproduces the sum rule.

Step 4. Interpret the sign and size. Both $t$ and $u$ are negative (spacelike) for physical scattering, and $-t$ equals the squared three-momentum transfer in the centre-of-mass frame; small $-t$ means a soft, forward glance, large $-t$ a hard, wide-angle knock. Why: this is why $t$ -channel exchange of a light boson produces the sharp forward peak seen in Rutherford, Mott, and Bhabha scattering — the propagator $1/t$ blows up as $t\to0$ . *As a concrete check: at θ = 60°, √s = 10 GeV, the equations give t = −25 GeV² and u = −75 GeV², and −t = s sin²30° = 100 × 0.25 = 25 GeV².

Limiting cases

Case	Result	Meaning
Massless / ultra-relativistic	$s+t+u=0$	rest masses negligible; $t=-s\sin^2(\theta^/2)$ , $u=-s\cos^2(\theta^/2)$
Forward scattering	$t\to0$ , $u\to-s$	grazing deflection; $t$ -channel propagator $1/t$ diverges, giving the forward peak
Backward scattering	$t\to-s$ , $u\to0$	maximal deflection; roles of $t$ and $u$ swap
Threshold	$s=(m_1+m_2)^2$ , $p^*=0$	just enough energy to create the final state; particles produced at rest in the c.m. frame
$s$ -channel resonance	$s=M^2$	propagator $1/(s-M^2)$ peaks; a real particle of mass $M$ is formed (e.g. the Z pole)
Fixed target, high $E$	$s\approx2m_2E_{\text{lab}}$	$\sqrt{s}$ grows only as $\sqrt{E_{\text{lab}}}$ — the reason colliders win

Common trapThe u-channel is always a separate process

It is not. The $u$ -channel diagram is only physically distinct from the $t$ -channel when the two final-state particles are identical (or related by the exchange, as in Møller scattering $e^-e^-\to e^-e^-$ or Compton scattering where the crossed photon leg matters). When the final particles are different — say $e^+e^-\to\mu^+\mu^-$ — there is no ambiguity about which vertex emits which particle, and the $u$ -channel simply does not contribute a separate amplitude. Writing down a spurious $u$ -channel diagram for distinguishable final states double-counts. Equally, do not confuse "there are three Mandelstam variables" with "there are always three diagrams": every $2\to2$ process has the three invariants $s$ , $t$ , $u$ as kinematic quantities, but which channels contribute dynamically depends on the interaction and on whether the external particles are identical.

Worked examples

1 · The antiproton threshold on the Bevatron (fully worked)

To make an antiproton you must conserve baryon number, so the cleanest reaction on a stationary hydrogen target is $p+p\to p+p+p+\bar p$ — three protons and one antiproton in the final state. What beam energy does a fixed-target machine need? Take $m_p=0.938$ GeV.

Step 1. Find the minimum $\sqrt{s}$ . The lightest the four-body final state can be is the sum of rest masses, $4m_p$ , so the threshold is $\sqrt{s}=4m_p$ and $s_{\text{thr}}=(4m_p)^2=16m_p^2$ . This is the mass-reach requirement from Derivation 2.

Step 2. Relate $s$ to the beam energy for a stationary target using equation (6): $s=m_p^2+m_p^2+2m_pE_{\text{lab}}=2m_p^2+2m_pE_{\text{lab}}$ .

Step 3. Set the two equal: $2m_p^2+2m_pE_{\text{lab}}=16m_p^2$ , so $2m_pE_{\text{lab}}=14m_p^2$ and $E_{\text{lab}}=7m_p$ . Numerically $E_{\text{lab}}=7\times0.938=\mathbf{6.57\ GeV}$ total, i.e. a kinetic energy of $T=E_{\text{lab}}-m_p=6m_p=\mathbf{5.63\ GeV}$ .

Step 4. Interpret. Creating a single antiproton of mass $0.938$ GeV costs six times its rest energy in beam kinetic energy, because momentum conservation forces the heavy four-proton final state to keep moving forward and carry away kinetic energy. This is precisely why the Berkeley Bevatron was designed to reach $6.2$ GeV — and it discovered the antiproton in 1955. The threshold beam energy is 7mₚ ≈ 6.57 GeV (T ≈ 5.63 GeV).

2 · Muon-pair production: reading off s, t, u

In $e^+e^-\to\mu^+\mu^-$ at a symmetric collider with beam energy $5$ GeV, so $\sqrt{s}=10$ GeV, a muon comes out at centre-of-mass angle $\theta^*=60^\circ$ . Find $s$ , $t$ , $u$ . (Neglect $m_e$ and $m_\mu$ against $10$ GeV.)

Step 1. The centre-of-mass energy squared is $s=(2E)^2=(10\ \text{GeV})^2=\mathbf{100\ GeV^2}$ .

Step 2. Use the massless angle relation (5): $t=-\tfrac{s}{2}(1-\cos60^\circ)=-\tfrac{100}{2}(1-0.5)=-50\times0.5=\mathbf{-25\ GeV^2}$ .

Step 3. Likewise $u=-\tfrac{s}{2}(1+\cos60^\circ)=-50\times1.5=\mathbf{-75\ GeV^2}$ .

Step 4. Check the sum rule: $s+t+u=100-25-75=0$ , as required for massless external particles. Note $t$ and $u$ are both negative (physical, spacelike momentum transfer), and here the final muons are distinguishable from the initial electrons, so only the $s$ -channel photon actually contributes — the $u$ -channel is a bookkeeping variable, not a diagram. s = 100, t = −25, u = −75 GeV², summing to zero.

3 · Why the LHC is not a fixed-target machine

The LHC collides two $6.5$ TeV proton beams, $\sqrt{s}=13$ TeV. What beam energy would a fixed-target proton machine need to reach the same $\sqrt{s}$ ?

Step 1. The collider delivers $s_{\text{coll}}=(2E)^2=(13\ \text{TeV})^2=169\ \text{TeV}^2=1.69\times10^8\ \text{GeV}^2$ .

Step 2. For a fixed target, equation (6) gives $s\approx2m_pE_{\text{lab}}$ , so $E_{\text{lab}}=s/(2m_p)=1.69\times10^8/(2\times0.938)\ \text{GeV}$ .

Step 3. Evaluate: $E_{\text{lab}}=9.0\times10^7\ \text{GeV}\approx\mathbf{90\ PeV}$ — about seven thousand times the LHC's own beam energy, and far beyond any conceivable magnet ring. A fixed-target machine would need a 90 PeV beam to match the LHC's 13 TeV; colliding beams turn a linear-in-E advantage into the whole high-energy frontier.

Physics of the portrait

The Mandelstam variables are named for the physicist who, in a single 1958 paper, saw that the analytic structure of a scattering amplitude in these invariants was as physical as the particles themselves — that one function of $s$ , $t$ , $u$ could encode three reactions and their bound states at once.

Origins · 1 minStanley Mandelstam and how this was discovered

Stanley Mandelstam

1928–2016 · S-matrix theory and relativistic scattering kinematics

Stanley Mandelstam was born in Johannesburg, South Africa, in 1928, studied at Witwatersrand and Cambridge, and spent most of his career at the University of California, Berkeley. In 1958, working on pion–nucleon scattering, he introduced the invariants $s$ , $t$ , $u$ and — more consequentially — the Mandelstam representation, a double dispersion relation expressing the scattering amplitude as an analytic function of two of these variables at once. The representation made crossing symmetry concrete: the same analytic function, continued from one region of the $s$ – $t$ – $u$ plane to another, describes three physically distinct processes (for example $e^+e^-\to\mu^+\mu^-$ , $e^-\mu^-\to e^-\mu^-$ , and their crossed relatives), with resonances in one channel showing up as forces in another. This analyticity-and-crossing program, developed with Geoffrey Chew and others into the "S-matrix" and bootstrap approach of the 1960s, sought to determine strong-interaction amplitudes from consistency alone, without a Lagrangian — and it fed directly into Regge theory and, through the Veneziano amplitude, into the birth of string theory. Mandelstam himself made lasting contributions across that whole arc: light-cone quantization of gauge and string theories, the Mandelstam–Leibbrandt prescription, and an influential argument for the perturbative finiteness of superstring amplitudes. He died in Berkeley in 2016. The variables that bear his name are the simplest and most durable piece of a career spent insisting that the analytic behaviour of amplitudes is where the physics lives.

History

1943

— Heisenberg proposes the S-matrix as the primary observable of scattering theory

1954

— Gell-Mann and Goldberger formulate crossing symmetry relating distinct reaction channels

1955

— the Bevatron reaches 6.2 GeV and Chamberlain and Segrè discover the antiproton

1958

— Mandelstam introduces s, t, u and the double dispersion representation

1959

— Regge analyses complex angular momentum, later mapped onto the Mandelstam plane

1961

— Chew and Frautschi build the bootstrap program on Mandelstam analyticity

Going further

Texts. Thomson, Modern Particle Physics, §2.2 (four-vectors and Mandelstam variables) · Griffiths, Introduction to Elementary Particles, 2nd ed., Ch. 3 (relativistic kinematics) · Halzen & Martin, Quarks and Leptons, Ch. 4 · Peskin & Schroeder, An Introduction to Quantum Field Theory, §5.4 (crossing and Mandelstam variables) · Schwartz, Quantum Field Theory and the Standard Model, Ch. 7 · Particle Data Group, Review of Particle Physics, "Kinematics" section. The invariants and the double-dispersion representation are due to S. Mandelstam, Phys. Rev. 112, 1344 (1958). The sum rule $s+t+u=\sum_i m_i^2$ , the centre-of-mass relation $\sqrt{s}=E_{\text{cm}}$ , the Källén momentum $p^*=\sqrt{\lambda}/2\sqrt{s}$ , and the massless angle relations are standard and re-derived above. The key numeric results are: the antiproton threshold $E_{\text{lab}}=7m_p\approx6.57$ GeV, the muon-pair values $(s,t,u)=(100,-25,-75)$ GeV², and the LHC-versus-fixed-target comparison ( $13$ TeV $\leftrightarrow$ $90$ PeV).

Prerequisites. Four-vector notation · Special-relativity kinematics · Centre-of-mass and lab frames. Leads to. Feynman diagrams and virtual particles · Interaction cross sections · Resonances and the Breit–Wigner shape.

Ask the tutor: Give me any two-body collision — beam, target or colliding energies, and a scattering angle — and I will walk you through computing $s$ , $t$ , and $u$ , checking the sum rule, and reading off whether the reaction is even allowed.

Why are there exactly three Mandelstam variables and not more?

Because a $2\to2$ process has four external four-momenta, giving in principle several possible pairwise invariants, but four-momentum conservation and the on-shell conditions $p_i^2=m_i^2$ cut the count down. The three combinations $s=(p_1+p_2)^2$ , $t=(p_1-p_3)^2$ , $u=(p_1-p_4)^2$ exhaust the ways to group the momenta, and the sum rule $s+t+u=\sum_i m_i^2$ then removes one degree of freedom, leaving only two independent invariants. So "three variables, one constraint" is the exact accounting: enough to name all three channels, but only two are ever free.

Is √s the same as the beam energy?

Only for a symmetric collider, and even then with a factor. In a symmetric collider where each beam has energy $E$ , the head-on collision gives $\sqrt{s}=2E$ , so two $6.5$ TeV beams make $\sqrt{s}=13$ TeV. In a fixed-target setup the target is at rest and $\sqrt{s}\approx\sqrt{2m_2E_{\text{lab}}}$ , which is far smaller — a $6.5$ TeV beam on a proton target reaches only about $110$ GeV. The invariant $\sqrt{s}$ is the physically meaningful "collision energy," and quoting it rather than a beam energy is what lets you compare machines of different design on equal footing.

Why are t and u negative for real scattering?

Because they measure squared momentum transfer, which is spacelike. Take $t=(p_1-p_3)^2$ : for elastic-like scattering the two particles have similar energies but different directions, so $p_1-p_3$ is dominated by its spatial part and its square is negative in the mostly-minus metric. Concretely, the massless relation $t=-s\sin^2(\theta^*/2)\le0$ makes this explicit — $t$ is zero only for exactly forward scattering and reaches $-s$ for backscattering. A positive ("timelike") value of $t$ would instead correspond to the crossed reaction, where $t$ plays the role of an annihilation energy — which is the whole idea of crossing symmetry.

What is crossing symmetry, in one paragraph?

It is the statement that a single analytic amplitude $\mathcal{M}(s,t,u)$ describes several related reactions at once, depending on which region of the $s$ – $t$ – $u$ plane you evaluate it in. Turning an incoming particle into an outgoing antiparticle (and vice versa) maps the $s$ -channel process $1+2\to3+4$ into the $t$ -channel process $1+\bar3\to\bar2+4$ and the $u$ -channel process $1+\bar4\to\bar2+3$ . Physically distinct experiments — say muon-pair production and electron–muon scattering — are thus values of the same function continued to different Mandelstam regions. Mandelstam's dispersion representation was the tool that made this continuation rigorous, and it remains central to the modern amplitudes program.

When does the u-channel actually matter?

When the two final-state particles are identical or exchange-related, so that swapping them gives a genuinely different diagram. In Møller scattering $e^-e^-\to e^-e^-$ the two outgoing electrons are indistinguishable, so both the $t$ -channel and the $u$ -channel photon-exchange diagrams contribute and must be added (with a relative minus sign from Fermi statistics). In Bhabha scattering $e^+e^-\to e^+e^-$ the $s$ - and $t$ -channels appear; in Compton scattering the $s$ - and $u$ -channels do. But for a process with all-distinct final particles like $e^+e^-\to\mu^+\mu^-$ , only the $s$ -channel is dynamically present, and $u$ is merely a kinematic label satisfying the sum rule.

How do Mandelstam variables connect to Feynman propagators and resonances?

Each internal line in a tree diagram carries a virtual four-momentum $q$ whose square is one of the Mandelstam variables, and the propagator contributes a factor $\sim1/(q^2-M^2)$ where $M$ is the exchanged particle's mass. In the $s$ -channel, $q^2=s$ , so when $\sqrt{s}$ is tuned to $M$ the propagator blows up and the cross section shows a resonance peak — this is literally how the $Z$ boson and the $J/\psi$ were mapped out. In the $t$ -channel, $q^2=t\le0$ never hits a physical mass, so instead of a peak you get the smooth forward rise $\sim1/t$ familiar from Rutherford scattering. The variables thus tell you immediately whether a channel resonates or merely forward-peaks.