Four-vectors & index notation

Particle physics is written in a language designed so that its sentences mean the same thing to every observer, no matter how fast they move. That language is the four-vector, with its grammar of upstairs and downstairs indices. It exists because of Einstein's single postulate — the speed of light is the same in all inertial frames — which forces the spacetime interval $c^2t^2-x^2-y^2-z^2$ to be invariant under the Lorentz transformation that carries one frame into another. A four-vector is any quartet $a^\mu=(a^0,a^1,a^2,a^3)$ that transforms exactly like the coordinates $x^\mu$ , and its defining virtue is that scalar products built from four-vectors come out identical in every frame. This article, resting on aq-pp-natural-units and the non-relativistic quantum mechanics review, fixes three pieces of machinery. First, the metric tensor $g_{\mu\nu}=\mathrm{diag}(1,-1,-1,-1)$ , which lowers a contravariant index $a^\mu$ into its covariant partner $a_\mu$ by flipping the sign of the spatial components — a bookkeeper for the minus signs of the $(+,-,-,-)$ signature. Second, the Einstein summation convention, in which a repeated up–down index is silently summed, so $a^\mu b_\mu$ means $a^0b^0-\mathbf{a}\cdot\mathbf{b}$ . Third, the payoff: the four-momentum $p^\mu=(E,\mathbf{p})$ , whose invariant square $p^2=E^2-\mathbf{p}^2=m^2$ is the same in the lab, the centre of mass, and any boosted frame. Master this notation and the kinematics of the rest of the course — Mandelstam variables, decay rates, cross sections — become one-line statements that are automatically frame-independent.

Key takeaways

Einstein's postulate makes the spacetime interval $(ct)^2-x^2-y^2-z^2$ invariant; the transformation preserving it is the Lorentz transformation, $a'^\mu=\Lambda^\mu{}_\nu a^\nu$ .
A contravariant four-vector $a^\mu=(a^0,\mathbf{a})$ has a covariant partner $a_\mu=(a^0,-\mathbf{a})$ ; the metric $g_{\mu\nu}=\mathrm{diag}(1,-1,-1,-1)$ raises and lowers indices.
A repeated up-down index is summed (Einstein convention), and every contraction $a^\mu b_\mu=a^0b^0-\mathbf{a}\cdot\mathbf{b}$ is a Lorentz scalar — the same in all frames.
Four-momentum $p^\mu=(E,\mathbf{p})$ obeys $p^2=E^2-\mathbf{p}^2=m^2$ ; this single invariant replaces separate energy and momentum bookkeeping.
The four-gradient $\partial_\mu=(\partial_t,\nabla)$ builds the d'Alembertian $\Box=\partial^\mu\partial_\mu=\partial_t^2-\nabla^2$ , the operator of the wave and Klein–Gordon equations.
Key idea: if you write a physical relation purely in terms of four-vector scalar products, it is automatically true in every inertial frame — no need to re-derive it after a Lorentz boost.

The core insight

A four-vector is any quartet that transforms like the spacetime coordinates, and the metric-weighted contraction of two four-vectors is a Lorentz invariant; write physics as such contractions and it holds, unchanged, in every frame.

Where this lives today

Four-vector notation is the working coordinate system of every high-energy experiment running now. At the LHC's Run 3, protons collide at $\sqrt s=13.6$ TeV, and every analysis begins by assembling the measured energies and momenta into four-vectors and contracting them. The 2012 discovery of the Higgs boson was, operationally, a histogram of the invariant mass $M^2=(p_1+p_2)^2$ of photon pairs and four-lepton systems: a bump at 125 GeV. The same reconstruction finds the $Z$ boson as a peak at 91.2 GeV in dimuon events, tags the $J/\psi$ and $\Upsilon$ , and hunts for new resonances as excesses on a mass spectrum. Mandelstam variables $s,t,u$ are the natural arguments of every scattering amplitude computed with Feynman diagrams, and the identity $s+t+u=\sum m_i^2$ is a routine check on those calculations. Beyond colliders, four-momentum bookkeeping underlies neutrino-oscillation kinematics, cosmic-ray air-shower reconstruction, and lattice-QCD hadron masses. Whenever someone writes "the missing transverse momentum" or "the amplitude as a function of $s$ and $t$ ," they are speaking the grammar this article defines.

The physical picture

Two observers in relative motion disagree about almost everything an event's coordinates measure — its time, its position, the length of a rod, the rate of a clock. Special relativity's discovery is that beneath all that disagreement one number survives untouched: the spacetime interval. The Lorentz transformation is precisely the change of coordinates that keeps $(ct)^2-x^2$ fixed while sliding time and space into each other. Boosting to a moving frame tilts the time and space axes toward the light cone like the closing blades of scissors, yet every event slides along a fixed hyperbola $(ct)^2-x^2=\text{const}$ that both frames agree on. The first figure pins one event to its invariant hyperbola in two frames at once.

Figure

A Minkowski spacetime diagram. Boosting to a frame moving at β = 0.5 tilts the time and space axes (blue) toward the 45° light cone, yet the event E stays on the invariant hyperbola (ct)² − x² = 16, so its interval s = 4 is the same in both frames

The four-vector formalism turns that geometric fact into an algebraic reflex. Instead of drawing hyperbolas we contract indices: any two four-vectors produce a scalar $a\cdot b=a^0b^0-\mathbf{a}\cdot\mathbf{b}$ that no boost can change. The most important instance is the four-momentum contracted with itself. A particle's energy $E$ and momentum $\mathbf{p}$ are frame-dependent — a fast frame sees them large, the rest frame sees $\mathbf{p}=0$ — but the combination $E^2-\mathbf{p}^2$ is frozen at $m^2$ , the mass squared. In the energy–momentum plane, every four-momentum of a particle lies on one hyperbola, the mass shell, and collider kinematics is the art of reading masses off these invariant curves. The second figure shows the proton's mass shell with a real number attached.

Figure

The mass shell E² − p² = m² for the proton (m = 0.938 GeV). Every four-momentum of the proton lies on this hyperbola whatever the frame; at p = 2 GeV the energy is E = √(p² + m²) = 2.209 GeV, and the dashed 45° asymptote E = p is the massless limit

The core equations

The starting point is the Lorentz transformation. For a boost with velocity $v=\beta c$ along the $z$ -axis, it mixes the time and $z$ coordinates through $\gamma$ while leaving $x$ and $y$ alone. Every four-vector, by definition, transforms with this same matrix $\Lambda^\mu{}_\nu$ .

a'^\mu=\Lambda^\mu{}_\nu\,a^\nu,\qquad \Lambda^\mu{}_\nu=\begin{pmatrix}\gamma&0&0&-\gamma\beta\\0&1&0&0\\0&0&1&0\\-\gamma\beta&0&0&\gamma\end{pmatrix},\qquad \gamma=\frac{1}{\sqrt{1-\beta^2}} \qquad (1)

where: $a^\mu$ — any contravariant four-vector; $\Lambda^\mu{}_\nu$ — the boost matrix (upper index rows, lower index columns); $\beta=v/c$ ; $\gamma\ge1$ . Reading it: $a'^0=\gamma(a^0-\beta a^3)$ and $a'^3=\gamma(a^3-\beta a^0)$ mix time and $z$ ; transverse components ride through unchanged. The metric tensor distinguishes the two flavours of index: it lowers a contravariant index by negating the spatial components, and its inverse $g^{\mu\nu}$ (numerically identical) raises them back. We use the $(+,-,-,-)$ signature throughout.

g_{\mu\nu}=g^{\mu\nu}=\mathrm{diag}(1,-1,-1,-1),\qquad a_\mu=g_{\mu\nu}a^\nu=(a^0,-\mathbf{a}),\qquad a^\mu=g^{\mu\nu}a_\nu \qquad (2)

where: $g_{\mu\nu}$ — metric tensor; $a^\mu=(a^0,\mathbf{a})$ — contravariant; $a_\mu=(a^0,-\mathbf{a})$ — covariant. Reading it: the metric only ever flips the sign of the three spatial entries, and $g^{\mu\lambda}g_{\lambda\nu}=\delta^\mu{}_\nu$ , so raising then lowering returns the original. The whole point of the up–down distinction is the scalar product. Contract a contravariant with a covariant index — a repeated up–down index is summed (Einstein convention) — and the result is a Lorentz invariant, the same in every frame.

a\cdot b\;\equiv\;a^\mu b_\mu=g_{\mu\nu}a^\mu b^\nu=a^0b^0-\mathbf{a}\cdot\mathbf{b} \qquad (3)

where: $a^\mu b_\mu$ — the metric-weighted contraction; the sum over $\mu=0,1,2,3$ is implied. Reading it: time multiplies with a plus, space with a minus; the case $a=b$ gives the invariant length-squared $a^2=(a^0)^2-\mathbf{a}^2$ , positive (timelike), negative (spacelike), or zero (lightlike). The four-momentum carries all of relativistic kinematics. Its time component is the energy, its space part the ordinary momentum, and its invariant square is the mass squared — the most-used equation in the subject.

p^\mu=(E,\mathbf{p}),\qquad p^2=p^\mu p_\mu=E^2-\mathbf{p}^2=m^2\;\;\Longleftrightarrow\;\;E=\sqrt{\mathbf{p}^2+m^2} \qquad (4)

where: $E$ — total energy; $\mathbf{p}$ — three-momentum; $m$ — rest mass ( $c=1$ ). Reading it: $E$ and $\mathbf{p}$ separately change under a boost, but $E^2-\mathbf{p}^2$ is locked at $m^2$ ; the rest frame ( $\mathbf{p}=0$ ) gives $E=m$ , the famous $E=mc^2$ . Derivatives also carry indices, with a sign subtlety. The four-gradient with a lower index, $\partial_\mu=\partial/\partial x^\mu$ , has all-plus spatial parts, while raising the index flips them — the opposite pattern to a coordinate. Contracting the two builds the d'Alembertian, the relativistic wave operator.

\partial_\mu=\frac{\partial}{\partial x^\mu}=(\partial_t,\nabla),\qquad \partial^\mu=(\partial_t,-\nabla),\qquad \Box\equiv\partial^\mu\partial_\mu=\partial_t^2-\nabla^2 \qquad (5)

where: $\partial_\mu$ — covariant four-gradient; $\partial^\mu=g^{\mu\nu}\partial_\nu$ — its contravariant form; $\Box$ — d'Alembertian. Reading it: the derivative index sits "downstairs" naturally (the coordinate $x^\mu$ is in the denominator), so $\partial_\mu$ transforms covariantly and $\Box$ is a Lorentz scalar — which is why $\Box\phi=0$ and the Klein–Gordon equation $(\Box+m^2)\phi=0$ hold in every frame. Finally, in a scattering process $1+2\to3+4$ , the invariant contractions of the four-momenta are packaged into the three Mandelstam variables, the frame-independent coordinates of a two-body reaction.

s=(p_1+p_2)^2,\quad t=(p_1-p_3)^2,\quad u=(p_1-p_4)^2,\qquad s+t+u=\sum_{i=1}^{4}m_i^2 \qquad (6)

where: $s$ — square of the total centre-of-mass energy; $t,u$ — momentum-transfer invariants; $m_i$ — the four external masses. Reading it: $\sqrt{s}$ is the energy available to make new particles, and $s+t+u=\sum m_i^2$ is a pure four-vector identity, derived below.

Derivation AThe invariant interval fixes the metric and every scalar product

The goal is to show, starting only from Einstein's postulate that the speed of light is frame-independent, that the spacetime interval must be invariant, that this invariance is exactly the condition $\Lambda^{\mathsf T}g\,\Lambda=g$ on the transformation matrix, and that this one matrix equation guarantees every scalar product $a^\mu b_\mu$ is a Lorentz invariant — turning a physical postulate into an algebraic identity that never has to be checked again.

Step 1. A light pulse leaving the origin obeys $c^2t^2-\mathbf{x}^2=0$ ; in a frame moving at velocity $v$ it must, by the postulate, also travel at $c$ , so $c^2t'^2-\mathbf{x}'^2=0$ . The two vanishing quadratics can differ only by an overall factor, and isotropy plus sensible composition of boosts forces that factor to one, so $c^2t^2-\mathbf{x}^2=c^2t'^2-\mathbf{x}'^2$ for the pulse and, by linearity, for any interval. Why: the constancy of $c$ is the whole physical input; everything downstream is the algebra of what respects it.

Step 2. Write the interval as a metric contraction, $s^2=g_{\mu\nu}x^\mu x^\nu$ ( $c=1$ ). Demanding $s^2$ be invariant under $x'^\mu=\Lambda^\mu{}_\nu x^\nu$ gives $g_{\alpha\beta}\Lambda^\alpha{}_\mu\Lambda^\beta{}_\nu x^\mu x^\nu\stackrel{!}{=}g_{\mu\nu}x^\mu x^\nu$ . Holding for all $x^\mu$ , the coefficients match: $g_{\alpha\beta}\Lambda^\alpha{}_\mu\Lambda^\beta{}_\nu=g_{\mu\nu}$ , i.e. $\Lambda^{\mathsf T}g\,\Lambda=g$ . Why: this single matrix condition IS the definition of a Lorentz transformation — the boost (1) satisfies it, and so does any rotation.

Step 3. Take two arbitrary four-vectors, each transforming as $a'^\mu=\Lambda^\mu{}_\nu a^\nu$ . Their scalar product is $a'\cdot b'=g_{\mu\nu}\Lambda^\mu{}_\alpha\Lambda^\nu{}_\beta a^\alpha b^\beta=g_{\alpha\beta}a^\alpha b^\beta=a\cdot b$ , using the identity from Step 2. Why: invariance was never special to coordinates — it follows for any pair of four-vectors from the metric-preserving property alone, which is why writing physics in contractions makes it frame-independent by construction.

Step 4. The covariant partner makes the contraction painless: define $b_\nu=g_{\mu\nu}b^\mu$ , and the scalar product collapses to $a^\mu b_\mu$ , a bare sum over a repeated up–down index with no explicit metric in sight — the lowered index has absorbed the minus signs. Why: this is the entire purpose of the two index positions — the metric is hidden inside the covariant vector so Lorentz invariance looks like an ordinary dot product, guarded only by the up–down pairing rule. *Note: the boost (1) explicitly satisfies $\Lambda^{\mathsf T}g\,\Lambda=g$ , which reproduces $t'^2-z'^2=t^2-z^2$ ; every contraction $a^\mu b_\mu$ is therefore frame-independent.*

Go deeper · Derivation B · 1 minRaising and lowering indices, and the sign flip of the four-gradient

Derivation BRaising and lowering indices, and the sign flip of the four-gradient

The aim here is to nail down the mechanics of the index calculus — how the metric moves indices up and down, why a contravariant $a^\mu=(a^0,\mathbf a)$ becomes the covariant $a_\mu=(a^0,-\mathbf a)$ , and why the four-gradient behaves in the mirror-image way, all-plus spatial signs when its index is down and all-minus when up. This reversed pattern is the commonest source of sign errors in the subject, so we derive it rather than assert it.

Step 1. Apply the metric componentwise: $a_0=g_{00}a^0=+a^0$ , while $a_1=g_{11}a^1=-a^1$ , and similarly $a_2=-a^2$ , $a_3=-a^3$ , so $a_\mu=(a^0,-\mathbf a)$ . Why: because $g$ is diagonal with a single plus in the time slot, lowering an index just flips the sign of the three spatial components — no mixing, no factors.

Step 2. The derivative operator $\partial_\mu\equiv\partial/\partial x^\mu$ has its index naturally down because $x^\mu$ sits in the denominator. The chain rule gives $\partial'_\mu=(\partial x^\nu/\partial x'^\mu)\,\partial_\nu=(\Lambda^{-1})^\nu{}_\mu\,\partial_\nu$ , the covariant (inverse-matrix) law. Why: this proves $\partial_\mu$ is a genuine covariant four-vector — it transforms with $\Lambda^{-1}$ as a lower-index object must, so contractions like $\partial_\mu a^\mu$ are automatically scalars.

Step 3. Write the components: $\partial_\mu=(\partial_t,\partial_x,\partial_y,\partial_z)=(\partial_t,\nabla)$ — all four with a plain plus. Raising the index with $g^{\mu\nu}$ gives $\partial^\mu=(\partial_t,-\nabla)$ . Why: this is the crucial reversal — the coordinate has $x^\mu=(t,+\mathbf x)$ upstairs and $x_\mu=(t,-\mathbf x)$ downstairs, but the derivative has $\partial_\mu=(\partial_t,+\nabla)$ downstairs and $\partial^\mu=(\partial_t,-\nabla)$ upstairs, the opposite placement of the minus signs.

Step 4. Contract to build the wave operator: $\Box=\partial^\mu\partial_\mu=(\partial_t)(\partial_t)+(-\nabla)\cdot(\nabla)=\partial_t^2-\nabla^2$ , a Lorentz scalar. Substituting $p^\mu\to i\partial^\mu$ into $p^2=m^2$ gives $-\Box\phi=m^2\phi$ , the Klein–Gordon equation $(\Box+m^2)\phi=0$ . Why: the covariance of $\partial_\mu$ guarantees the field equations of the whole course are frame-independent, and the sign bookkeeping is what makes $\Box$ come out as $\partial_t^2-\nabla^2$ rather than a Euclidean Laplacian. *In summary: $a_\mu=(a^0,-\mathbf a)$ , $\partial_\mu=(\partial_t,\nabla)$ , $\partial^\mu=(\partial_t,-\nabla)$ , $\Box=\partial_t^2-\nabla^2$ ; the derivative's sign pattern is the mirror of a coordinate's.*

Go deeper · Derivation C · 1 minThe mass shell and the Mandelstam identity

Derivation CThe mass shell and the Mandelstam identity

The purpose of this derivation is to extract the two most useful consequences of the four-momentum's invariance: first that $p^2=m^2$ holds in every frame, giving the energy–momentum relation $E=\sqrt{\mathbf p^2+m^2}$ , and second that the three Mandelstam invariants obey the closed sum rule $s+t+u=\sum_i m_i^2$ . Both follow from expanding scalar products under four-momentum conservation, and both are workhorses of collider kinematics.

Step 1. In its rest frame a particle has $p^\mu=(m,\mathbf 0)$ , so $p^2=m^2$ . Boost with (1): $p^\mu=(\gamma m,\gamma m\boldsymbol\beta)=(E,\mathbf p)$ , and $p^2=\gamma^2m^2-\gamma^2\beta^2m^2=\gamma^2m^2(1-\beta^2)=m^2$ . Why: $p^2$ is invariant from Derivation A, so evaluating it in the easiest frame (rest) fixes its value everywhere — the standard trick of computing an invariant where it is simplest.

Step 2. Rearranging $E^2-\mathbf p^2=m^2$ gives the dispersion relation $E=\sqrt{\mathbf p^2+m^2}$ , the relativistic replacement for $E=\mathbf p^2/2m$ . Two limits check it: $\mathbf p=0$ gives $E=m$ (rest energy); $m\to0$ gives $E=|\mathbf p|$ (a photon on its own light cone, $p^2=0$ ). Why: a correct invariant must reproduce both the rest energy and the massless light-cone condition, and the mass shell interpolates smoothly between them.

Step 3. For a reaction $1+2\to3+4$ with conservation $p_1+p_2=p_3+p_4$ , define $s=(p_1+p_2)^2$ , $t=(p_1-p_3)^2$ , $u=(p_1-p_4)^2$ . Expand using $p_i^2=m_i^2$ : $s=m_1^2+m_2^2+2p_1\cdot p_2$ , $t=m_1^2+m_3^2-2p_1\cdot p_3$ , $u=m_1^2+m_4^2-2p_1\cdot p_4$ . Why: each Mandelstam variable is a Lorentz scalar, so all three are the same in the lab and the centre of mass — which is why cross sections are quoted as functions of $s,t,u$ .

Step 4. Add the three: the mass terms give $3m_1^2+m_2^2+m_3^2+m_4^2$ , the cross terms $2p_1\cdot(p_2-p_3-p_4)$ . Conservation gives $p_2-p_3-p_4=-p_1$ , so the cross terms become $2p_1\cdot(-p_1)=-2m_1^2$ , and $3m_1^2-2m_1^2=m_1^2$ leaves $s+t+u=m_1^2+m_2^2+m_3^2+m_4^2$ . Why: only two Mandelstam variables are independent, and this identity is the exact statement of that redundancy — a free consistency check on any kinematic calculation. *In summary: $p^2=m^2$ holds in every frame; $E=\sqrt{\mathbf p^2+m^2}$ correctly reproduces $E=m$ in the rest frame and $E=|\mathbf p|$ in the massless limit; $s+t+u=\sum_i m_i^2$ follows by direct expansion under four-momentum conservation.*

Limiting cases

Case	Result	Meaning
Low velocity $\beta\ll1$	$\gamma\to1$ , boost $\to$ Galilean	reduces to Newtonian $t'=t$ , $z'=z-vt$ ; time and space stop mixing
Rest frame $\mathbf p=0$	$p^\mu=(m,\mathbf 0)$ , $p^2=m^2$	energy equals rest mass, $E=mc^2$ ; the invariant in its simplest form
Massless particle $m=0$	$p^2=0$ , $E=\lvert\mathbf p\rvert$	four-momentum is lightlike (null); the particle lives on its own light cone (the photon)
Ultra-relativistic $E\gg m$	$E\approx\lvert\mathbf p\rvert+m^2/2\lvert\mathbf p\rvert$	mass is a small correction; collider particles are nearly null
Timelike / spacelike / null $a^2$	$a^2>0$ / $a^2<0$ / $a^2=0$	invariant classification of a separation, the same in every frame

Common trapThe scalar product is a signed sum, not an ordinary dot product

The single most frequent error is to treat $a\cdot b$ as $a^0b^0+\mathbf a\cdot\mathbf b$ , a plain Euclidean dot product. It is not. The Minkowski metric carries a relative minus sign: $a\cdot b=a^\mu b_\mu=a^0b^0-\mathbf a\cdot\mathbf b$ , time-with-plus, space-with-minus. Forgetting this turns $p^2=E^2-\mathbf p^2=m^2$ into the meaningless $E^2+\mathbf p^2$ . A subtler trap is the four-gradient: unlike a coordinate, whose minus signs live in the covariant form $x_\mu=(t,-\mathbf x)$ , the derivative's minus signs live in the contravariant form $\partial^\mu=(\partial_t,-\nabla)$ , while $\partial_\mu=(\partial_t,+\nabla)$ — the opposite index position. The safe discipline is mechanical: always pair one up index with one down, and let the metric do the sign-flipping explicitly.

Worked examples

1 · Why colliders beat fixed targets (a real-numbers calculation)

Two protons ( $m_p=0.938$ GeV) collide head-on, each beam at $E=6500$ GeV. What centre-of-mass energy $\sqrt s$ results, and what beam energy would a fixed-target machine need to match it?

Step 1. For the head-on collider, take $p_1=(E,0,0,p)$ and $p_2=(E,0,0,-p)$ . Then $s=(p_1+p_2)^2=(2E,\mathbf 0)^2=4E^2$ , giving $\sqrt s=2E=\mathbf{13\,000\ GeV=13\ TeV}$ . All the beam energy is available in the centre of mass because the total momentum is zero.

Step 2. For a fixed target, $p_1=(E',0,0,p')$ hits a proton at rest, $p_2=(m_p,\mathbf 0)$ . Now $s=(p_1+p_2)^2=2m_p^2+2E'm_p\approx2E'm_p$ for $E'\gg m_p$ . The stationary target soaks up most of the momentum, so only a fraction of $E'$ becomes collision energy.

Step 3. To reach the same $s=(13\,000)^2=1.69\times10^8\ \mathrm{GeV}^2$ , invert: $E'=s/(2m_p)=1.69\times10^8/(2\times0.938)=\mathbf{9.0\times10^{7}\ GeV}$ — that is $9.0\times10^4$ TeV, or 90 PeV, roughly seven thousand times the LHC beam energy. A fixed-target machine would need a 90 PeV beam to do what two 6.5 TeV beams do; this is exactly why every energy-frontier machine is a collider. The whole argument is one line: $\sqrt s$ is a Lorentz invariant, computed most easily in each machine's natural frame.

2 · Reconstructing an invariant mass from decay products

A neutral particle decays to two photons of energy $E_1=E_2=62.5$ GeV that emerge back-to-back. What was the parent's mass? What if instead they made an opening angle $\theta=60^\circ$ ?

Step 1. Photons are massless, $p_i^2=0$ , so $E_i=\lvert\mathbf p_i\rvert$ . The parent four-momentum is $P=p_1+p_2$ , and its invariant mass is $M^2=P^2=p_1^2+p_2^2+2p_1\cdot p_2=2p_1\cdot p_2$ .

Step 2. Expand the contraction: $p_1\cdot p_2=E_1E_2-\mathbf p_1\cdot\mathbf p_2=E_1E_2-\lvert\mathbf p_1\rvert\lvert\mathbf p_2\rvert\cos\theta=E_1E_2(1-\cos\theta)$ , using $\lvert\mathbf p_i\rvert=E_i$ . Hence the master formula $M^2=2E_1E_2(1-\cos\theta)$ .

Step 3. Back-to-back means $\theta=180^\circ$ , $\cos\theta=-1$ , so $M^2=4E_1E_2=4(62.5)(62.5)=15\,625\ \mathrm{GeV}^2$ , giving $M=\mathbf{125\ GeV}$ — the mass of the Higgs boson, and exactly how it was found in the diphoton channel.

Step 4. At $\theta=60^\circ$ , $\cos\theta=0.5$ , so $M^2=2(62.5)(62.5)(1-0.5)=3906.25\ \mathrm{GeV}^2$ and $M=\mathbf{62.5\ GeV}$ . Same photon energies, half the mass — the invariant mass depends on the opening angle as well as the energies, which is why a detector must measure both directions accurately. The result $M$ is frame-independent: a boosted observer sees different $E_1,E_2,\theta$ but recovers the identical $M$ .

Physics of the portrait

The four-vector was the geometric reinterpretation that turned Einstein's postulate into a calculus — supplied by his own former mathematics teacher.

Origins · 1 minHermann Minkowski and how this was discovered

Hermann Minkowski

1864–1909 · the geometry of spacetime

Hermann Minkowski was a mathematician, not a physicist by trade — a prodigy who at eighteen shared the Grand Prix of the Paris Academy for work on quadratic forms, and who taught the young Albert Einstein at the Zurich Polytechnic. His decisive contribution came late and fast. In 1907–1908, three years after special relativity, Minkowski realised that Lorentz's transformations and Einstein's postulate had a purely geometric meaning: space and time are not separate stages but a single four-dimensional continuum whose "distance" is the invariant interval $c^2t^2-x^2-y^2-z^2$ . His 1908 Cologne address opened with the now-famous declaration that "henceforth space by itself, and time by itself, are doomed to fade away into mere shadows, and only a kind of union of the two will preserve an independent reality." There he introduced the four-dimensional vectors and metric bookkeeping this article is built on — what we now call four-vectors, the light cone, and the timelike/spacelike/lightlike classification of intervals. Einstein at first dismissed the reformulation as "superfluous learnedness," then reversed himself completely: Minkowski's spacetime became the scaffolding of general relativity a decade later, and the index calculus grew, through Sommerfeld, Ricci, and Levi-Civita, into the tensor language of all modern field theory. Minkowski did not live to see it — he died of appendicitis in January 1909, months after the Cologne lecture, at forty-four.

History

1904

— Lorentz writes the coordinate transformation that preserves the form of Maxwell's equations

1905

— Einstein derives the same transformation from the constancy of the speed of light

1906

— Poincaré shows the transformations form a group and highlights the invariant t² − x² − y² − z²

1908

— Minkowski recasts relativity as the geometry of four-dimensional spacetime

1910

— Sommerfeld systematizes the four-vector and tensor calculus of electrodynamics

Going further

Texts. Thomson, Modern Particle Physics, §2.2 (the Lorentz transformation, four-vectors and invariance, Mandelstam variables) · Griffiths, Introduction to Elementary Particles, 2nd ed., §3.2–3.4 (relativistic kinematics) · Halzen & Martin, Quarks and Leptons, §3.1–3.3 · Peskin & Schroeder, An Introduction to Quantum Field Theory, §1 and the conventions appendix (metric $(+,-,-,-)$ , index notation) · Schwartz, Quantum Field Theory and the Standard Model, §2 (Lorentz invariance and conventions). The metric convention $g_{\mu\nu}=\mathrm{diag}(1,-1,-1,-1)$ is the particle-physics standard; Misner–Thorne–Wheeler and much of general relativity use the opposite $(-,+,+,+)$ signature, which flips the sign of every scalar product and of $m^2$ . Natural units $c=\hbar=1$ throughout (see aq-pp-natural-units). Key results to keep in mind: the boost (1) satisfies $\Lambda^{\mathsf T}g\Lambda=g$ and preserves $t^2-z^2$ ; $p^2=m^2$ in both rest and boosted frames; $\sqrt s=2E=13$ TeV for the LHC head-on case and the fixed-target equivalent is $E'=9.0\times10^7$ GeV; the diphoton invariant mass formula gives $M=\sqrt{2E_1E_2(1-\cos\theta)}=125$ GeV back-to-back; and $s+t+u=\sum_i m_i^2$ .

Prerequisites. Natural Units · Non-Relativistic Quantum Mechanics Review. Leads to. Special-Relativistic Kinematics · Mandelstam Variables · Centre-of-Mass Frame · Klein–Gordon Equation.

Ask the tutor: Give me any decay or collision and I will walk you through building the four-momenta and contracting them into the invariant mass or Mandelstam variables step by step.

Why is there a minus sign between the time and space parts of the scalar product?

Because the invariant that survives a Lorentz boost is $c^2t^2-x^2-y^2-z^2$ , not the Euclidean $t^2+x^2+y^2+z^2$ . Light travelling at $c$ satisfies $c^2t^2=\mathbf x^2$ , so the combination that stays zero in every frame is the difference, and the metric $g_{\mu\nu}=\mathrm{diag}(1,-1,-1,-1)$ simply encodes which sign each component contributes. Geometrically, the minus separates a genuine light cone from a sphere; physically it distinguishes the past and future you can reach (timelike) from the elsewhere you cannot (spacelike). Every strange feature of relativity — time dilation, the impossibility of overtaking light — traces back to that one relative minus.

What is the practical difference between an upper and a lower index?

An upper (contravariant) index transforms like the coordinates, $a'^\mu=\Lambda^\mu{}_\nu a^\nu$ ; a lower (covariant) index transforms with the inverse matrix, $b'_\mu=(\Lambda^{-1})^\nu{}_\mu b_\nu$ . The metric relates them by flipping the sign of the spatial parts, $a_\mu=(a^0,-\mathbf a)$ . The practical rule is mechanical: you may only sum an index that appears once up and once down. Such a contraction $a^\mu b_\mu$ transforms with $\Lambda$ and $\Lambda^{-1}$ back to back, which cancel, leaving a Lorentz scalar. If you ever sum two upper or two lower indices, you have made an error — the result would not be frame-independent.

Why is the invariant mass so useful in a real collider experiment?

Because it is the one number about a decay that every observer agrees on, and it is computable from what the detector measures. When a heavy particle decays, its daughters fly off with energies and momenta that depend entirely on how the parent was moving. But the contraction $M^2=(\sum p_i)^2$ of the daughter four-momenta always returns the parent's rest mass. So you histogram $M$ over millions of events: a particle of definite mass piles up as a sharp peak while combinatorial background spreads out smoothly. That is literally how the Higgs (125 GeV), the $Z$ (91 GeV), and countless hadrons were found. The four-vector formalism guarantees the peak sits at the same place regardless of each event's boost.

Does the choice of metric signature (+,−,−,−) versus (−,+,+,+) matter?

It is a convention, but you must never mix the two within a calculation. Particle physics almost universally uses $(+,-,-,-)$ (the "mostly minus" metric), which makes $p^2=E^2-\mathbf p^2=+m^2$ , a tidy positive mass squared. Much of general relativity uses $(-,+,+,+)$ , which gives $p^2=-m^2$ and flips the sign of every scalar product. Physical predictions — masses, cross sections, decay rates — are identical either way, because signs cancel consistently. The danger is purely clerical: importing a formula from a mostly-plus textbook into a mostly-minus calculation without adjusting signs. Always check which convention a source uses; it is usually stated up front.

How does the Klein–Gordon equation drop out of this notation?

Through the operator substitution $p^\mu\to i\partial^\mu$ , the relativistic version of the quantum rule $E\to i\partial_t$ , $\mathbf p\to-i\nabla$ . Start from the invariant every free particle obeys, $p^2=m^2$ , and promote it to an operator on a field: $p^\mu p_\mu\to(i\partial^\mu)(i\partial_\mu)=-\Box$ . Setting this equal to $m^2$ gives $-\Box\phi=m^2\phi$ , or $(\Box+m^2)\phi=0$ — the Klein–Gordon equation. It is one line precisely because $\Box=\partial^\mu\partial_\mu$ is a Lorentz scalar, so the wave equation is automatically valid in every frame; without the index calculus you would have to check frame-independence by hand.

Why do we only need two of the three Mandelstam variables?

Because they are tied together by the identity $s+t+u=\sum_i m_i^2$ , proved above. A two-body reaction $1+2\to3+4$ has, after conservation and the mass-shell conditions, only two independent kinematic degrees of freedom — physically, the total energy and the scattering angle. The three variables $s,t,u$ package these symmetrically (each channel of a diagram depends on one of them), but the sum rule removes the redundancy: fix any two and the third is determined. This is also a live consistency check — if computed $s$ , $t$ , and $u$ do not add up to the sum of the four masses squared, there is an arithmetic error in the kinematics.