Four-Vectors

Take a single arrow and draw it in spacetime rather than in space. Two observers gliding past each other at constant velocity will slice that spacetime into "now" and "later" differently, so they will read off different lists of four numbers for the very same arrow — yet it is one arrow, and everything physical about it must be something they both agree on. A four-vector is the object built to make this precise: a quantity whose four components shuffle between temporal and spatial parts under a change of inertial frame in exactly the way the coordinates $(ct,x,y,z)$ do, through the Lorentz transformation. That transformation law — not the mere fact of "having four numbers" — is the whole definition, and it is what turns a bookkeeping convenience into a geometric object. The payoff is enormous and immediate. Any equation written as an equality between four-vectors, $A^\mu=B^\mu$ , automatically holds in every inertial frame, because both sides are dragged along by the same matrix $\Lambda^\mu{}_\nu$ . This is manifest Lorentz covariance, and it is the reason relativistic mechanics, electromagnetism, and fluid dynamics are all cleanest in four-vector language. Attached to every four-vector is a single frame-independent number, its squared norm $A\cdot A=\eta_{\mu\nu}A^\mu A^\nu$ , formed with the Minkowski metric that Lorentz transformations leave untouched. The sign of that number sorts four-vectors into timelike, null, and spacelike — the same trichotomy that carves spacetime into the interior, surface, and exterior of the light cone. Position, four-velocity, four-momentum, four-current, and four-potential are all instances, and their invariant norms hand back the constancy of $u\cdot u=-c^2$ , the mass-shell relation $E^2=(pc)^2+(mc^2)^2$ , and charge conservation.

Key takeaways

A four-vector is an ordered quadruple $A^\mu=(A^0,A^1,A^2,A^3)$ that transforms under a change of inertial frame exactly like the spacetime coordinates, $A'^\mu=\Lambda^\mu{}_\nu A^\nu$ — the transformation law is the definition, not the four-ness.
Its individual components are frame-dependent, but the arrow itself is a genuine geometric object; different observers merely list it in different bases.
The inner product $A\cdot B=\eta_{\mu\nu}A^\mu B^\nu$ built with the Minkowski metric is a Lorentz invariant — the same number in every frame — and generalizes the spacetime interval.
Its sign splits four-vectors into timelike ( $A\cdot A<0$ ), null ( $A\cdot A=0$ ), and spacelike ( $A\cdot A>0$ ) in the $-+++$ convention, a classification no Lorentz transformation can change.
Promoting physical quantities to four-vectors — position, velocity, momentum, current, potential — bakes in the unification of space and time and makes their laws covariant for free.
Key idea: a four-vector is defined by how it transforms, $A'^\mu=\Lambda^\mu{}_\nu A^\nu$ , the same linear Lorentz map that acts on the coordinate displacement $\Delta x^\mu$ ; invariants, index raising and lowering, and covariant laws all follow from this one rule together with the Minkowski metric.

The core insight

A four-vector is defined by how it transforms: $A'^\mu=\Lambda^\mu{}_\nu A^\nu$ , the same linear Lorentz transformation that acts on the coordinate displacement $\Delta x^\mu$ . Everything else — invariants, index gymnastics, covariant laws — follows from this single rule together with the Minkowski metric.

Where this lives today

Four-vectors are not a historical curiosity; they are the daily working language of experimental physics. Every event recorded at the Large Hadron Collider is reconstructed as a set of four-momenta, and the invariant mass $M^2c^4=(\sum p^\mu)\cdot(\sum p^\mu)$ of a collection of decay products is exactly how a resonance is found — the 2012 Higgs discovery is a bump in the four-momentum invariant mass of photon pairs and four-lepton final states, the same algebra as worked example 2. In astrophysics the null four-vector of light, boosted between the source frame and the observer frame, delivers relativistic Doppler shifts and aberration in one stroke, and it is the wave four-vector $k^\mu$ that redshifts photons from distant galaxies. And conceptually, the four-vector is the seed of everything that follows in general relativity: it is the rank-1 tensor, and promoting the flat metric $\eta_{\mu\nu}$ to a position-dependent $g_{\mu\nu}(x)$ while keeping four-vectors as the objects living in each tangent space is, in one sentence, the passage from special to general relativity.

The physical picture

Think of a four-vector as one arrow drawn in spacetime. Different inertial observers cut spacetime into space and time along differently tilted axes, so they disagree about the arrow's component list — but they are all pointing at the same arrow. The Lorentz transformation is nothing more than the dictionary that translates one observer's four numbers into another's, and the requirement that it be the same dictionary the coordinates obey is precisely what promotes a bare quadruple to a four-vector. The invariant length of that arrow, measured with the Minkowski metric, is something every observer computes to the same value. It plays the role ordinary length plays for arrows in Euclidean space, with one twist: the minus sign in the time slot means some non-zero arrows have zero length. Those are the null vectors, and they point along the light cone — the four-momentum of a photon is the standard example. The sign of the squared length is a permanent label: a boost can rotate a timelike arrow so its components change wildly, but it can never turn it spacelike.

Figure

One arrow, two frames. Observers S and S′ read different components but agree on the invariant norm A·A = −c² for any four-velocity

Because a boost mixes the time and space slots, a quantity that is purely temporal for one observer acquires a spatial part for another. The rest energy of a particle at rest is all time-component; set that particle moving and the same four-momentum sprouts a momentum three-vector. Four-vectors keep this bookkeeping automatically, so a law phrased between them never has to be rewritten when you change frame. The second figure shows the invariant that governs this particular arrow: the four-momentum lives on a hyperbola — the mass shell — whose vertex is the rest energy and whose asymptotes are the null lines of massless particles.

Figure

The mass shell of a proton. E² = (pc)² + (mc²)² is a hyperbola with vertex at the rest energy mc² = 938 MeV and asymptotes E = ±pc; at v = 0.8c the proton sits at pc = 1251 MeV, E = 1564 MeV

Finally, the two ways of listing an arrow. Contravariant components $A^\mu$ and covariant components $A_\mu$ are the same arrow measured against two natural rulers; the metric converts between them, and contracting one raised against one lowered index yields the frame-independent numbers that carry the physics.

The core equations

The definition is a transformation law, not a formula for the components. A four-vector is any object whose components in a new inertial frame are obtained from the old ones by the same matrix that carries the coordinate displacement $\Delta x^\mu$ across frames. Everything else in this article is an application of this one rule.

A'^{\mu} = \Lambda^{\mu}{}_{\nu}\, A^{\nu} \qquad (1)

where: $A^\mu$ — contravariant components in the original frame, $\mu=0,1,2,3$ ; $A'^\mu$ — components in the new (primed) inertial frame; $\Lambda^\mu{}_\nu$ — the Lorentz transformation matrix relating the two frames. Physical reading: the components are just the arrow's shadow on one observer's axes, and $\Lambda$ rotates the shadow when you rotate the axes in spacetime. The Minkowski metric supplies the one operation Euclidean geometry takes for granted — an inner product — but with a crucial minus sign in the time slot. Contract two four-vectors against $\eta_{\mu\nu}$ and you get a scalar that, as the next derivation shows, is identical in every inertial frame.

A\cdot B = \eta_{\mu\nu}\,A^{\mu}B^{\nu} = -A^0B^0 + A^1B^1 + A^2B^2 + A^3B^3 \qquad (2)

where: $\eta_{\mu\nu}=\mathrm{diag}(-1,+1,+1,+1)$ — the Minkowski metric; $A^\mu,B^\nu$ — contravariant components of two four-vectors. Physical reading: the single minus sign is the whole of special relativity's geometry; it is why time and space are unified but not interchangeable. The metric also lets you trade a raised index for a lowered one. The lowered-index object $A_\mu$ is a covariant vector — a one-form living in the dual space — and it is the correct partner to contract against a contravariant vector.

A_{\mu} = \eta_{\mu\nu}\,A^{\nu},\qquad A^{\mu} = \eta^{\mu\nu}\,A_{\nu} \qquad (3)

where: $A_\mu$ — covariant components; $\eta^{\mu\nu}$ — the inverse metric, satisfying $\eta^{\mu\alpha}\eta_{\alpha\nu}=\delta^\mu{}_\nu$ ; $\delta^\mu{}_\nu$ — the Kronecker delta. Physical reading: in the $-+++$ convention lowering the index flips the sign of only the time component, $A_0=-A^0$ , so $A\cdot A=A_\mu A^\mu$ reproduces equation (2). The sign of a four-vector's own squared norm is a Lorentz-invariant label. It sorts every arrow into one of three causal classes, mirroring the interior, surface, and exterior of the light cone.

A\cdot A = \eta_{\mu\nu}A^{\mu}A^{\nu} \;\begin{cases} <0 & \text{timelike}\\ =0 & \text{null}\\ >0 & \text{spacelike}\end{cases} \qquad (4)

where: $A\cdot A$ — the invariant squared norm. Physical reading: timelike arrows connect cause to effect and can be the worldline tangent of a massive particle; null arrows track light; spacelike arrows separate events no signal can join. The first genuinely physical four-vector is the four-velocity, the proper-time derivative of position. Differentiating with respect to the invariant proper time $\tau$ — rather than a frame-dependent coordinate time — is exactly what makes the result transform correctly.

u^{\mu} = \frac{\mathrm{d}x^{\mu}}{\mathrm{d}\tau} = \gamma\,(c,\;\mathbf{v}),\qquad u\cdot u = -c^{2} \qquad (5)

where: $\tau$ — proper time along the worldline; $\gamma=(1-v^2/c^2)^{-1/2}$ — the Lorentz factor; $\mathbf{v}=\mathrm{d}\mathbf{x}/\mathrm{d}t$ — the ordinary three-velocity. Physical reading: $u^\mu$ is the unit tangent to the worldline (up to the factor $c$ ); its norm is frozen at $-c^2$ no matter how fast the particle moves. Multiplying the four-velocity by the invariant mass yields the four-momentum, whose time and space parts are energy and momentum. Its invariant norm is the mass shell — the relativistic dispersion relation that particle physicists live by.

p^{\mu} = m\,u^{\mu} = \left(\tfrac{E}{c},\;\mathbf{p}\right),\qquad p\cdot p = -m^{2}c^{2}\;\Rightarrow\; E^2=(pc)^2+(mc^2)^2 \qquad (6)

where: $m$ — invariant (rest) mass; $E=\gamma mc^2$ — relativistic energy; $\mathbf{p}=\gamma m\mathbf{v}$ — relativistic three-momentum. Physical reading: energy and momentum are one object seen from two angles, and their frame-independent combination is the mass; massless particles ( $m=0$ ) live on the null cone with $E=pc$ . Electromagnetism supplies two more. Charge density and current combine into the four-current, whose vanishing four-divergence is charge conservation; the scalar and vector potentials combine into the four-potential.

J^{\mu} = (c\rho,\;\mathbf{J}),\quad \partial_{\mu}J^{\mu}=0;\qquad A^{\mu} = \left(\tfrac{\phi}{c},\;\mathbf{A}\right) \qquad (7)

where: $\rho$ — charge density; $\mathbf{J}$ — three-current density; $\partial_\mu=(\tfrac1c\partial_t,\nabla)$ — the four-gradient; $\phi$ — electrostatic potential; $\mathbf{A}$ — magnetic vector potential. Physical reading: the continuity equation, Maxwell's equations, and gauge freedom all become manifestly frame-independent once written between these four-vectors.

Derivation AWhy an equation between four-vectors holds in every frame

The single most important fact about four-vectors is that their inner product is a Lorentz scalar — the same number in all inertial frames — and the entire usefulness of the formalism rests on it. If the inner product is invariant, then so is any norm, any angle, and any equation of the form $A^\mu=B^\mu$ : both sides ride the same matrix, so equality in one frame forces equality in all. The goal here is to prove invariance directly from the transformation law and the one defining property of a Lorentz matrix, and then to read off the covariance principle as an immediate corollary. Nothing beyond linear algebra is needed.

Step 1. Insert the transformation law (1) for both vectors into the inner product: $A'\cdot B'=\eta_{\mu\nu}\Lambda^\mu{}_\alpha\Lambda^\nu{}_\beta A^\alpha B^\beta$ . Why: this expresses the primed-frame product entirely in terms of unprimed components, isolating the whole frame-dependence into the pair of $\Lambda$ factors sandwiching the metric.

Step 2. Use the defining property of a Lorentz transformation, $\eta_{\mu\nu}\Lambda^\mu{}_\alpha\Lambda^\nu{}_\beta=\eta_{\alpha\beta}$ — this is exactly the statement that $\Lambda$ preserves the metric, the four-dimensional analogue of an orthogonal matrix preserving lengths. Why: it is this preservation, and only this, that lets the two $\Lambda$ 's collapse back into a bare metric; a general linear map would not.

Step 3. Substitute to collapse the expression: $A'\cdot B'=\eta_{\alpha\beta}A^\alpha B^\beta=A\cdot B$ . The primed and unprimed products are literally the same number. Why: this is the invariance we wanted; the norm $A\cdot A$ , being the special case $B=A$ , is likewise a scalar, so causal type is frame-independent.

Step 4. Read off covariance: if $A^\mu=B^\mu$ in one frame, then $A'^\mu=\Lambda^\mu{}_\nu A^\nu=\Lambda^\mu{}_\nu B^\nu=B'^\mu$ in every other. Why: a law built as an equality of four-vectors needs no rewriting under a change of observer — this is manifest Lorentz covariance, the reason the whole language exists.

Go deeper · Derivation B · 1 minThe four-velocity norm and the energy–momentum relation

Derivation BThe four-velocity norm and the energy–momentum relation

Differentiating position with respect to proper time rather than coordinate time is a deliberate choice, and it has a striking consequence: the resulting four-velocity has a norm that is completely independent of the motion, fixed forever at $-c^2$ . Scaling by the invariant mass then turns that fixed norm into the mass-shell relation $E^2=(pc)^2+(mc^2)^2$ , the single most-used equation in relativistic kinematics. The goal is to derive both from the metric and the chain rule alone, showing how a purely geometric normalization becomes the physical dispersion relation connecting a particle's energy, momentum, and mass.

Step 1. Convert the proper-time derivative to a coordinate-time derivative with the chain rule: $u^\mu=\dfrac{\mathrm{d}x^\mu}{\mathrm{d}\tau}=\dfrac{\mathrm{d}t}{\mathrm{d}\tau}\dfrac{\mathrm{d}x^\mu}{\mathrm{d}t}=\gamma(c,\mathbf{v})$ , using $\mathrm{d}t/\mathrm{d}\tau=\gamma$ . Why: proper time and coordinate time differ precisely by the factor $\gamma$ , and it is this factor that will make the norm collapse to a constant.

Step 2. Feed the components into the metric: $u\cdot u=\gamma^2(-c^2+\mathbf{v}\cdot\mathbf{v})=\gamma^2 c^2(-1+v^2/c^2)$ . Why: the minus sign on the time slot pits the temporal part against the spatial part, and the shared $\gamma^2$ is about to cancel the velocity dependence entirely.

Step 3. Simplify using $\gamma^2(1-v^2/c^2)=1$ , giving $u\cdot u=-c^2$ for any speed. Why: the normalization is geometric, not dynamical — $u^\mu$ is the unit tangent to the worldline, so a "faster" particle has a differently-tilted, not longer, velocity arrow.

Step 4. Scale by the mass: $p\cdot p=m^2(u\cdot u)=-m^2c^2$ . Writing the same norm in components, $p\cdot p=-E^2/c^2+\mathbf{p}\cdot\mathbf{p}$ , and equating the two forms gives $E^2=(pc)^2+(mc^2)^2$ . Why: one invariant, computed two ways, is the entire content of the energy–momentum relation, valid in every frame and reducing to $E=mc^2$ at rest and $E=pc$ for massless particles.

Go deeper · Derivation C · 1 minA boost is a hyperbolic rotation in rapidity

Derivation CA boost is a hyperbolic rotation in rapidity

A rotation in the plane preserves circles; a boost preserves hyperbolae. Making this analogy exact means rewriting the boost matrix in terms of a single parameter, the rapidity $\varphi$ , defined by $\tanh\varphi=\beta$ . In that variable a boost along $x$ acts on the $(A^0,A^1)$ pair as a hyperbolic rotation — the same algebra as an ordinary rotation, with cosines and sines replaced by hyperbolic cosines and sines. The goal is to derive that form, verify it preserves the Minkowski norm exactly as an ordinary rotation preserves Euclidean length, and expose the payoff: rapidities of successive collinear boosts simply add, which ordinary velocities emphatically do not.

Step 1. Start from the standard boost with relative speed $v$ along $x$ : $A'^0=\gamma(A^0-\beta A^1)$ and $A'^1=\gamma(A^1-\beta A^0)$ , with the transverse components $A^2,A^3$ untouched. Why: only the $(t,x)$ plane containing the boost direction is affected, so the problem reduces to a two-by-two transformation.

Step 2. Substitute the rapidity identities $\gamma=\cosh\varphi$ and $\gamma\beta=\sinh\varphi$ , which follow from $\tanh\varphi=\beta$ together with $\cosh^2\varphi-\sinh^2\varphi=1$ . Why: this repackages two frame-dependent quantities, $\gamma$ and $\beta$ , into a single additive angle-like parameter.

Step 3. The boost becomes $A'^0=\cosh\varphi\,A^0-\sinh\varphi\,A^1$ and $A'^1=-\sinh\varphi\,A^0+\cosh\varphi\,A^1$ — a rotation through an imaginary angle in the $(t,x)$ plane. Why: side by side with an ordinary rotation, the only change is hyperbolic functions for circular ones, which is the geometric meaning of the minus sign in the metric.

Step 4. Check norm preservation: $-(A'^0)^2+(A'^1)^2=-(A^0)^2+(A^1)^2$ using $\cosh^2\varphi-\sinh^2\varphi=1$ , confirming the map is a Lorentz transformation. Successive collinear boosts add rapidities, $\varphi_{\rm tot}=\varphi_1+\varphi_2$ . Why: additivity is what makes rapidity the natural coordinate on the boost group and turns the awkward velocity-addition formula into simple addition.

Limiting cases

Case	Result	Meaning
Galilean / low-velocity limit	$\Lambda^\mu{}_\nu\to$ Galilean shear $+\,\mathcal{O}(v^2/c^2)$	As $v/c\to0$ the boost degenerates; time becomes absolute and the four-vector splits into an invariant time part and an ordinary 3-vector.
Null (lightlike) vectors	$A\cdot A=0,\ A^\mu\neq0$	Non-zero arrows of zero length exist and point along the light cone; a photon's four-momentum is the canonical case, $E=pc$ .
Pure spatial rotation	$A'^0=A^0,\ A'^i=R^i{}_jA^j,\ R\in SO(3)$	Rotations are Lorentz transformations that fix the time slot; the spatial part rotates as a familiar 3-vector.
Massless particle	$m\to0$ : $p\cdot p\to0$ , $E=pc$	The mass shell degenerates onto its own null-line asymptotes; four-momentum becomes null.
Signature convention	$-+++$ versus $+---$	Every norm flips overall sign between the mostly-plus (MTW, Wald, Carroll) and mostly-minus (much of particle physics) conventions; the invariant content is identical — always state your metric.

Common trapAny list of four numbers is a four-vector

It is not. The defining property is the transformation law $A'^\mu=\Lambda^\mu{}_\nu A^\nu$ , and most four-number stacks fail it. Pairing an energy with an unrelated three-vector, or stacking a charge with a velocity, produces a quadruple that does not transform as a unit and is therefore not a four-vector. The classic trap is the electromagnetic field: the electric field $\mathbf{E}$ and magnetic field $\mathbf{B}$ each have three components, but $(\mathbf{E},\mathbf{B})$ is not two four-vectors — under a boost the two fields rotate into each other as the six independent components of the rank-2 field-strength tensor $F^{\mu\nu}$ , not as vectors. Two further hazards ride alongside: sign conventions (a timelike vector has negative norm in $-+++$ but positive norm in $+---$ , so always state your signature), and index placement (never contract two raised or two lowered indices without a metric between them — $A^\mu B^\mu$ is meaningless, $A^\mu B_\mu$ is the scalar).

Worked examples

1 · Four-velocity and four-momentum of a fast proton (fully numeric)

A proton (rest energy $mc^2=938.27$ MeV) moves along the $x$ -axis at $v=0.8c$ . Find its four-velocity components and norm, its energy and momentum, and verify the mass shell.

Step 1. Lorentz factor: $\gamma=(1-0.8^2)^{-1/2}=(1-0.64)^{-1/2}=(0.36)^{-1/2}=1/0.6=\mathbf{5/3}$ . This single number scales everything that follows.

Step 2. Four-velocity: $u^\mu=\gamma(c,\mathbf{v})=\tfrac53(c,\,0.8c,\,0,\,0)=(1.667c,\,1.333c,\,0,\,0)$ . Its norm is $u\cdot u=-(1.667c)^2+(1.333c)^2=(-2.778+1.778)c^2=\mathbf{-c^2}$ , independent of the speed, exactly as the normalization demands.

Step 3. Energy and momentum: $E=\gamma mc^2=\tfrac53(938.27)=\mathbf{1563.8\ \text{MeV}}$ , and $pc=\gamma\beta mc^2=\tfrac53(0.8)(938.27)=\mathbf{1251.0\ \text{MeV}}$ .

Step 4. Mass-shell check: $(pc)^2+(mc^2)^2=1251.0^2+938.27^2=1{,}565{,}074+880{,}351=2{,}445{,}425\ \text{MeV}^2$ , whose square root is $1563.8$ MeV $=E$ . The invariant recombines to the mass: $E^2-(pc)^2=(938.27\ \text{MeV})^2$ , so the proton sits exactly on its mass shell, the amber-to-red point on the second figure.

2 · Two photons and the invariant mass (fully numeric)

In electron–positron annihilation at rest the system decays into two photons. Each carries energy $0.511$ MeV. Show each photon four-momentum is null, yet the pair has a timelike total with invariant mass equal to the original rest mass.

Step 1. Back-to-back geometry: the photons fly along $\pm x$ , so $p_1^\mu=(0.511,\,+0.511,\,0,\,0)$ MeV $/c$ and $p_2^\mu=(0.511,\,-0.511,\,0,\,0)$ MeV $/c$ (using $E=pc$ for light).

Step 2. Each is null: $p_1\cdot p_1=-(0.511)^2+(0.511)^2=\mathbf{0}$ , and likewise $p_2\cdot p_2=0$ . Neither photon alone has any mass.

Step 3. Add them: $P^\mu=p_1^\mu+p_2^\mu=(1.022,\,0,\,0,\,0)$ MeV $/c$ . The spatial parts cancel; the total is purely temporal.

Step 4. Invariant mass of the pair: $-P\cdot P=(1.022)^2-0=(1.022\ \text{MeV}/c)^2$ , so $Mc^2=\mathbf{1.022\ \text{MeV}}=2m_ec^2$ , twice the electron rest energy $0.511$ MeV. Two null four-vectors summed to a timelike one whose norm reproduces the initial rest mass — the sum of massless things can carry mass, and four-momentum conservation makes it exact.

3 · Adding boosts by adding rapidities

A frame $S'$ moves at $\beta_1=0.6$ relative to $S$ ; a third frame $S''$ moves at $\beta_2=0.6$ relative to $S'$ , in the same direction. Find the speed of $S''$ relative to $S$ two ways.

Step 1. Rapidity of each boost: $\varphi=\operatorname{artanh}(0.6)=\tfrac12\ln\!\frac{1+0.6}{1-0.6}=\tfrac12\ln4=\mathbf{0.6931}$ .

Step 2. Rapidities add for collinear boosts: $\varphi_{\rm tot}=0.6931+0.6931=\mathbf{1.3863}$ .

Step 3. Convert back to a velocity: $\beta_{\rm tot}=\tanh(1.3863)=\mathbf{0.8824}$ , i.e. $v=0.8824c$ — safely below $c$ , never reaching it however many boosts you stack.

Step 4. Cross-check with the velocity-addition formula: $\beta_{\rm tot}=\dfrac{\beta_1+\beta_2}{1+\beta_1\beta_2}=\dfrac{0.6+0.6}{1+0.36}=\dfrac{1.2}{1.36}=\mathbf{0.8824}$ . The two agree, because velocity addition is the addition formula for hyperbolic tangents — the rapidity is the additive coordinate on the boost group.

Physics of the portrait

The four-vector is the object that turned Einstein's kinematics into geometry. The person who saw that spacetime was the natural stage, and that its natural inhabitants transform as a single arrow, was Einstein's own former mathematics teacher.

Origins · 1 minHermann Minkowski and how this was discovered

Hermann Minkowski

1864–1909 · the geometry of spacetime and the four-vector

Hermann Minkowski was a mathematician, not a physicist, and for most of his career he worked on number theory and the geometry of numbers — his "convex body" theorems are still standard. He had taught the young Einstein mathematics at the ETH in Zürich and once described him, unfairly, as a "lazy dog." Yet it was Minkowski who, within three years of the 1905 paper, recognized that special relativity was not really about clocks and rulers but about the geometry of a four-dimensional continuum. In his 1907–1908 work on the electrodynamics of moving bodies he assembled the four-current, the four-potential, and the field-strength tensor, and on 21 September 1908 he opened his Cologne lecture "Raum und Zeit" 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." That union is Minkowski spacetime; its arrows are four-vectors; its invariant length is the metric norm. Minkowski died suddenly of appendicitis in January 1909, only months after the lecture, never seeing how completely his geometric language would take over. Einstein at first dismissed the reformulation as "superfluous learnedness," then adopted it wholesale — without Minkowski's four-dimensional geometry the leap to curved spacetime and general relativity would have had no vocabulary to be written in.

History

1905

— Einstein's special relativity makes the Lorentz transformation the law relating inertial frames

1906

— Poincaré writes the transformation as a rotation in a four-dimensional space with coordinate ict

1908

— Minkowski's "Raum und Zeit" introduces spacetime and the four-vector as its natural object

1909

— Minkowski dies; his four-dimensional program is carried on by Sommerfeld, Laue, and Born

1910

— Sommerfeld fixes the modern four-vector (Vierervektor) and six-vector calculus and notation

Going further

Texts. Misner, Thorne & Wheeler, Gravitation, ch. 2 (the geometry of four-vectors) · Wald, General Relativity, ch. 1 and appendix A · Carroll, Spacetime and Geometry, ch. 1 · Hartle, Gravity, ch. 5 · Schutz, A First Course in General Relativity, ch. 2. All equations and conventions use the $-+++$ metric. Key relations to keep in mind: the metric-preservation identity $\eta_{\mu\nu}\Lambda^\mu{}_\alpha\Lambda^\nu{}_\beta=\eta_{\alpha\beta}$ ; the four-velocity norm $u\cdot u=-c^2$ ; the mass shell $E^2=(pc)^2+(mc^2)^2$ ; and the rapidity boost matrix identity $\cosh^2\varphi-\sinh^2\varphi=1$ . The three worked examples anchor the numerics: a proton at $v=0.8c$ gives $\gamma=5/3$ , $E=1563.8$ MeV, $pc=1251.0$ MeV; two $0.511$ -MeV photons back-to-back yield invariant mass $1.022$ MeV; and $\operatorname{artanh}(0.6)=0.6931$ , so two such collinear boosts give $\tanh(1.3863)=0.8824$ , matching $(0.6+0.6)/(1+0.36)$ from the velocity-addition formula.

Builds on: Lorentz Transformations · Minkowski Spacetime · Minkowski Metric and Signature · Spacetime Interval. Leads to: Proper Time · Causal Structure and Light Cones · Lorentz Boosts and Rapidity.

Ask the tutor: ask me to boost any four-vector you like into a moving frame and watch its components scramble while its norm stays put.

What is the difference between a four-vector and just four numbers?

The difference is a transformation law. Four numbers become a four-vector only if, when you change to another inertial frame, they reshuffle according to $A'^\mu=\Lambda^\mu{}_\nu A^\nu$ — the same Lorentz matrix that acts on the spacetime coordinates. Most quadruples fail this test: stack an energy with an unrelated three-velocity and the result does not transform as a unit, so it is not a four-vector. The reason the distinction matters is covariance: only genuine four-vectors let you write laws that automatically hold in every frame. A four-vector is a geometric arrow whose reality is frame-independent; four arbitrary numbers are just a table.

Why is the four-velocity norm always −c², even for a photon-fast particle?

Because $u^\mu=\mathrm{d}x^\mu/\mathrm{d}\tau$ is differentiated with respect to proper time, the invariant clock carried by the particle itself. The chain rule inserts a factor $\gamma$ , and when you compute the Minkowski norm the algebra collapses: $u\cdot u=\gamma^2(-c^2+v^2)=-c^2$ for any $v<c$ . Geometrically, $u^\mu$ is the unit tangent to the worldline (scaled by $c$ ), so speeding up tilts the velocity arrow toward the light cone without lengthening it. The one exception is light itself: a photon has no rest frame and no proper time, so it has no four-velocity — only a null four-momentum with $p\cdot p=0$ .

What do covariant and contravariant indices actually mean?

They are two ways of listing the same arrow. Contravariant components $A^\mu$ (upper index) are the coordinates of the arrow in the usual basis; covariant components $A_\mu$ (lower index) are what you get after measuring the arrow against the metric, and they live in the dual space of one-forms. In the $-+++$ convention lowering an index flips the sign of the time component only, $A_0=-A^0$ , leaving the spatial parts alone. The point of keeping both is contraction: a scalar comes only from pairing one upper with one lower index, $A_\mu B^\mu$ . Summing two uppers or two lowers is not frame-independent and signals a mistake — the metric must always sit between them.

Does the sign convention of the metric change any physics?

No — it changes only bookkeeping. In the mostly-plus convention $-+++$ (favoured by MTW, Wald, Carroll, and most of the relativity community) a timelike vector has negative norm and $u\cdot u=-c^2$ ; in the mostly-minus convention $+---$ (common in particle physics) every norm flips overall sign, so $u\cdot u=+c^2$ and a timelike vector has positive norm. Every measurable quantity — proper time, energy, the mass shell — comes out identical, because the invariant content is convention-independent. The only rule is consistency: pick one signature and state it, since a stray sign from mixing conventions is one of the most common errors in relativistic calculations.

How can two massless photons add up to something with mass?

Because invariant mass is a property of the total four-momentum, not the sum of the individual masses. Each photon is null, $p\cdot p=0$ , but the norm is not additive: $(p_1+p_2)\cdot(p_1+p_2)=p_1\cdot p_1+p_2\cdot p_2+2\,p_1\cdot p_2=2\,p_1\cdot p_2$ , and for two photons that are not parallel this cross term is non-zero. Physically, the spatial momenta can partly or wholly cancel while the energies add, leaving a timelike total. In electron–positron annihilation the two photons fly apart back-to-back, their momenta cancel completely, and the invariant mass of the pair equals the $1.022$ MeV rest energy you started with — mass is conserved even though neither photon has any.

Why bother with four-vectors instead of just using the Lorentz transformation directly?

Because four-vectors let you stop transforming and start reasoning invariantly. Once a law is written as an equality of four-vectors, it is guaranteed true in every frame without ever applying $\Lambda$ — you get covariance for free. Better, you can extract answers by computing invariants in whichever frame is easiest: the mass shell $E^2=(pc)^2+(mc^2)^2$ is fastest read off in the rest frame, then used everywhere. This is why relativistic collisions are solved with Mandelstam invariants and why the whole apparatus scales up to the tensors of general relativity — four-vectors trade repeated coordinate juggling for a one-time geometric insight.