Four-Vectors and Minkowski Spacetime

The last two articles produced a strange inventory: durations that stretch, lengths that shrink, simultaneity that tilts — each result correct, each derived by pushing coordinates through the boost by hand. Minkowski's 1908 insight was that all of it is ordinary geometry, done in the wrong number of dimensions. Merge time with space into a single four-dimensional arena, give it the indefinite metric $\eta_{\mu\nu}=\mathrm{diag}(+1,-1,-1,-1)$ , and the Lorentz transformation stops being a strange mixing rule: it is simply the group of rotations of this geometry, and every kinematic effect becomes a projection of invariant objects onto an observer's tilted axes. The workhorse concept is the four-vector: any quadruple of quantities that transforms between frames exactly as the coordinates $(ct,x,y,z)$ do. The definition sounds bureaucratic and is actually a machine for manufacturing physics. Once a quantity passes the transformation test, every inner product built from it with the metric is automatically the same number in every inertial frame — invariants, conservation laws, and frame-proof equations fall out wholesale. Differentiate the position four-vector by the invariant proper time and you get the four-velocity; multiply by mass and you get the four-momentum that unifies energy with momentum in the next article. For the mechanics track this is deliberately a working introduction: the metric, index bookkeeping kept light, the light cone, proper time, and the four-velocity — the exact toolkit the four-momentum, four-force, and relativistic Lagrangian articles consume. The full tensor machinery lives in its own track later.

Key takeaways

A four-vector is any quadruple $A^\mu=(A^0,A^1,A^2,A^3)$ that transforms under the same Lorentz matrix $\Lambda^\mu{}_\nu$ as the coordinates $x^\mu=(ct,x,y,z)$ ; passing this test is what earns a quantity frame-proof physics.
Minkowski spacetime is $\mathbb{R}^4$ with the indefinite metric $\eta_{\mu\nu}=\mathrm{diag}(+1,-1,-1,-1)$ ; the invariant interval is $\Delta s^2=\eta_{\mu\nu}\Delta x^\mu\Delta x^\nu=c^2\Delta t^2-|\Delta\mathbf{x}|^2$ .
The inner product $A\cdot B=\eta_{\mu\nu}A^\mu B^\nu$ is a Lorentz scalar; $A\cdot A$ classifies four-vectors as timelike ( $>0$ ), null ( $=0$ ), or spacelike ( $<0$ ), and the light cone of null directions is the invariant causal skeleton.
Indices lower and raise with the metric, $A_\mu=\eta_{\mu\nu}A^\nu$ ; contracting one upper with one lower index always produces an invariant — the entire point of the notation.
Proper time $\mathrm{d}\tau=\mathrm{d}t/\gamma$ is the invariant arc length of a worldline; differentiating by it preserves four-vector character, giving the four-velocity $U^\mu=\gamma(c,\mathbf{v})$ with fixed normalization $U\cdot U=c^2$ .
Key idea: Demand that the interval be invariant; the linear maps preserving it form the Lorentz group, and any quadruple transforming under those maps like the coordinates is a four-vector whose scalar contractions are automatically the same in every inertial frame.

The core insight

Demand that the interval $c^2t^2-\mathbf{x}^2$ be invariant; the linear maps that preserve it form the Lorentz group, and any quadruple transforming under those maps like $x^\mu$ is a four-vector whose scalar contractions are automatically the same in every inertial frame.

Where this lives today

Minkowski's geometry is the stage on which all of fundamental physics currently performs. Every quantum field theory calculation — every cross-section at the LHC, every anomalous-moment prediction — begins with fields defined on flat Minkowski spacetime, with equation (2) as the symmetry constraint that dictates which interaction terms are even allowed to exist. Collider software leans on the four-vector method of Example 2 daily: reconstruction code stores every jet, lepton and photon as a four-vector, and "invariant mass" — the frame-proof contraction $\sqrt{p\cdot p}$ of summed four-momenta — is the x-axis of essentially every discovery plot in particle physics, the Higgs bump at 125 GeV included. The geometry itself has become measurable infrastructure. Gravitational-wave interferometers treat incoming ripples as perturbations $\eta_{\mu\nu}+h_{\mu\nu}$ of the Minkowski metric — the 2026 observing runs of LIGO–Virgo–KAGRA fit hundreds of mergers against waveforms computed on exactly this background split. Relativistic geodesy does the inverse: optical clock networks map the geoid by measuring where proper time (5) accumulates fastest. And the light-cone causal structure has escaped physics entirely — "light-cone bounds" now name the provable limits on how fast correlations spread in quantum simulators and lattice models.

The physical picture

Euclidean geometry is the study of what rotations leave alone: lengths and angles built from $\delta_{ij}$ . Minkowski geometry is the same discipline with one sign flipped. A boost scrambles $ct$ and $x$ exactly as a rotation scrambles $x$ and $y$ — but it preserves $c^2t^2-x^2$ rather than $x^2+y^2$ , so its "circles" are hyperbolas and its invariant "length" can be positive, zero, or negative. That single sign is the entire mathematical content of relativistic kinematics; time dilation and length contraction are just the shadows an indefinite metric casts on tilted axes. The sign of $\Delta s^2$ is not bookkeeping — it is causal law. The null directions $\Delta s^2=0$ form a double cone at every event, and the cone is the same for every observer, because boosts preserve the interval. Inside the cone ( $\Delta s^2>0$ , timelike): a clock can travel between the events, their time order is absolute, causation is possible. Outside ( $\Delta s^2<0$ , spacelike): no signal connects them, and different frames disagree about which came first — harmlessly, since neither can influence the other. Geometry enforces causality without any extra postulate.

Figure

The light cone at an event — null directions form an invariant double cone separating the absolute future and past from the spacelike elsewhere; timelike, null, and spacelike separations carry frame-independent meaning

A four-vector is defined by how it transforms, not by what it "is." That litmus test is the source of all the power: if a physical quadruple changes between frames with the same $\Lambda$ as the coordinates, then every inner product you build from it is guaranteed observer-independent before you compute a single component. The strategy of relativistic mechanics is to hunt for four-vectors the way Newtonian mechanics hunts for conserved quantities — find one, and frame-proof equations follow for free. The final piece is a frame-proof parameter to differentiate by. Coordinate time $t$ belongs to one observer, but the proper time $\tau$ ticked by a clock riding the worldline is an invariant — the Minkowski analogue of arc length along a curve. Differentiating any four-vector with respect to $\tau$ therefore yields another four-vector; that closure property is what lets us build velocity, momentum, and force versions of the whole Newtonian toolkit, one $\mathrm{d}/\mathrm{d}\tau$ at a time.

Figure

A worldline parametrized by proper time — the clock carried along the curve ticks the invariant arc length, and the four-velocity is the unit tangent, always timelike, always normalized to c squared

The core equations

Fix the convention once for the whole track: signature $(+,-,-,-)$ , Greek indices $0..3$ , Latin $1..3$ , and the summation convention — a repeated upper–lower index pair is summed. The invariant interval between neighbouring events, in metric form:

\Delta s^2 = \eta_{\mu\nu}\,\Delta x^\mu \Delta x^\nu = c^2\Delta t^2 - \Delta x^2 - \Delta y^2 - \Delta z^2 \qquad (1)

where: $\Delta x^\mu=(c\Delta t,\Delta\mathbf{x})$ — the coordinate separation; $\eta_{\mu\nu}=\mathrm{diag}(+1,-1,-1,-1)$ — the Minkowski metric. This is the previous articles' invariant, now wearing its geometric uniform: the metric plays the role of the Euclidean dot product, and everything that was proved about the interval — invariance, causal classification — becomes a statement about this bilinear form. The inverse metric $\eta^{\mu\nu}$ is numerically identical and satisfies $\eta^{\mu\alpha}\eta_{\alpha\nu}=\delta^\mu{}_\nu$ . The transformations of relativity are then defined geometrically. A linear map $x'^\mu=\Lambda^\mu{}_\nu x^\nu$ is a Lorentz transformation precisely when it preserves the metric:

\eta_{\mu\nu}\,\Lambda^\mu{}_\alpha\,\Lambda^\nu{}_\beta = \eta_{\alpha\beta} \qquad (2)

where: $\Lambda^\mu{}_\nu$ — the transformation matrix (boosts, spatial rotations, and their products). Equation (2) is the exact analogue of $R^TR=\mathbb{1}$ for rotations — Derivation A shows it is forced by interval invariance. The solutions form the Lorentz group $O(3,1)$ ; the physically connected piece (no time reversal, no mirror flip) contains the boosts of the first article and all spatial rotations, and rapidity is its natural boost coordinate. Now the definition that names the article. Any quadruple $A^\mu$ is a four-vector if it transforms like the coordinates,

A'^\mu = \Lambda^\mu{}_\nu\,A^\nu \qquad (3)

where: $A^\mu=(A^0,\mathbf{A})$ — one time component and three space components, mixed by boosts exactly as $(ct,\mathbf{x})$ are. This is a test, and quantities fail it: $(c\,\mathrm{d}t,\mathrm{d}\mathbf{x})/\mathrm{d}t$ — the naive velocity quadruple — is not a four-vector because the denominator $\mathrm{d}t$ transforms too. Passing the test is what licenses every trick in the rest of the chapter. The payoff structure is the inner product. Lower an index with the metric, $A_\mu=\eta_{\mu\nu}A^\nu$ (which just flips the sign of the spatial components: $A_0=A^0$ , $A_i=-A^i$ ), and contract:

A\cdot B = \eta_{\mu\nu}A^\mu B^\nu = A^\mu B_\mu = A^0B^0 - \mathbf{A}\cdot\mathbf{B} \qquad (4)

where: $A\cdot B$ — a Lorentz scalar: the same number in every inertial frame, by (2) and (3) together. The self-product classifies: $A\cdot A>0$ timelike, $=0$ null, $<0$ spacelike, and no Lorentz transformation can change the class. When you meet an unfamiliar four-vector, computing its invariant square in the cleverest frame — usually a rest frame — is the single most-used manoeuvre in relativistic mechanics. To build dynamics we need an invariant parameter along a worldline. For a timelike path, the interval accumulated between neighbouring ticks defines the proper time:

c^2\,\mathrm{d}\tau^2 = \eta_{\mu\nu}\,\mathrm{d}x^\mu\mathrm{d}x^\nu \quad\Longrightarrow\quad \mathrm{d}\tau = \frac{\mathrm{d}t}{\gamma} \qquad (5)

where: $\mathrm{d}\tau$ — proper time element, the reading of the clock riding the worldline; $\gamma$ — the instantaneous Lorentz factor of the particle in whatever frame supplies $t$ . Proper time is to worldlines what arc length is to curves in Euclidean space, and it is the only natural invariant parameter — which is exactly why the relativistic Lagrangian article will build the action out of $\int\mathrm{d}\tau$ . Differentiate the position four-vector by proper time and four-vector character survives (the numerator transforms with $\Lambda$ , the denominator is invariant). The result is the four-velocity:

U^\mu = \frac{\mathrm{d}x^\mu}{\mathrm{d}\tau} = \gamma\,(c,\ \mathbf{v}),\qquad U\cdot U = c^2 \qquad (6)

where: $\mathbf{v}$ — the ordinary 3-velocity; $\gamma=(1-v^2/c^2)^{-1/2}$ . The normalization $U\cdot U=c^2$ (Derivation B) holds for every particle at every speed — the four-velocity is a unit timelike tangent in disguise. Its derivative, the four-acceleration $A^\mu=\mathrm{d}U^\mu/\mathrm{d}\tau$ , is always Minkowski-orthogonal to it, $A\cdot U=0$ : acceleration can only rotate the tangent vector, never change its invariant length. Two consumers of this machinery, stated here and developed in the next articles. Multiplying $U^\mu$ by the invariant mass $m$ gives the four-momentum,

p^\mu = m\,U^\mu = \left(\frac{E}{c},\ \mathbf{p}\right),\qquad p\cdot p = m^2c^2 \qquad (7)

where: $E=\gamma mc^2$ and $\mathbf{p}=\gamma m\mathbf{v}$ — energy and momentum revealed as time and space components of one four-vector, with the mass-shell relation $E^2=(pc)^2+(mc^2)^2$ as its invariant square (that story is the four-momentum article's). And derivatives assemble into the four-gradient $\partial_\mu=\left(\tfrac{1}{c}\partial_t,\ \nabla\right)$ , whose invariant contraction $\Box=\partial^\mu\partial_\mu=\tfrac{1}{c^2}\partial_t^2-\nabla^2$ is the d'Alembertian — precisely the operator of the one-dimensional wave equation article, now revealed as Lorentz-invariant, which is why the wave speed $c$ was frame-independent all along.

Object	Components	Invariant square	Role
position $x^\mu$	$(ct,\mathbf{x})$	$s^2$ (from origin)	the arena
displacement $\mathrm{d}x^\mu$	$(c\,\mathrm{d}t,\mathrm{d}\mathbf{x})$	$c^2\mathrm{d}\tau^2$	defines proper time
four-velocity $U^\mu$	$\gamma(c,\mathbf{v})$	$c^2$	unit tangent
four-momentum $p^\mu$	$(E/c,\mathbf{p})$	$m^2c^2$	next article
four-gradient $\partial_\mu$	$(\tfrac1c\partial_t,\nabla)$	$\Box$	field theory bridge

Derivation AThe Lorentz group from interval invariance

We show that requiring the interval to be invariant for all separations forces the metric-preservation condition (2) — establishing that the Lorentz transformations of the first article, plus spatial rotations, are the complete symmetry set of Minkowski geometry. The strategy is direct: substitute the general linear map into the primed interval, demand equality with the unprimed one for arbitrary displacements, and match coefficients of the quadratic form. Watch the "for arbitrary $\Delta x$ " step — that is what upgrades one equation into the full matrix condition.

Step 1. Let $\Delta x'^\mu = \Lambda^\mu{}_\alpha\,\Delta x^\alpha$ and expand the primed interval:

\Delta s'^2 = \eta_{\mu\nu}\,\Delta x'^\mu\,\Delta x'^\nu = \eta_{\mu\nu}\,\Lambda^\mu{}_\alpha\,\Lambda^\nu{}_\beta\,\Delta x^\alpha\,\Delta x^\beta

Step 2. Demand invariance, $\Delta s'^2 = \Delta s^2 = \eta_{\alpha\beta}\,\Delta x^\alpha\Delta x^\beta$ , for every choice of $\Delta x^\alpha$ :

\left(\eta_{\mu\nu}\,\Lambda^\mu{}_\alpha\,\Lambda^\nu{}_\beta - \eta_{\alpha\beta}\right)\Delta x^\alpha\,\Delta x^\beta = 0

Why arbitrariness matters: a quadratic form that vanishes for all arguments has vanishing symmetric coefficient matrix — pick $\Delta x$ along single axes and along pairs of axes to isolate each component. One invariance statement becomes ten independent matrix equations.

Step 3. Read off the condition:

\eta_{\mu\nu}\,\Lambda^\mu{}_\alpha\,\Lambda^\nu{}_\beta = \eta_{\alpha\beta}\qquad\text{i.e.}\qquad \Lambda^T\eta\,\Lambda = \eta

Step 4. Extract the group structure. Taking determinants: $(\det\Lambda)^2=1$ , so $\det\Lambda=\pm1$ — no volume change, possible mirror flips. Taking the $00$ component: $(\Lambda^0{}_0)^2 = 1 + \sum_i(\Lambda^i{}_0)^2 \ge 1$ , so $|\Lambda^0{}_0|\ge1$ — the future cone maps to the future ( $\Lambda^0{}_0\ge1$ ) or is time-reversed. The physical, connected subgroup has $\det\Lambda=+1$ and $\Lambda^0{}_0\ge1$ : boosts, rotations, and their products. Why this closes the loop with the first article: the standard boost matrix satisfies Step 3 identically — substitute and use $\gamma^2(1-\beta^2)=1$ — and pure spatial rotations satisfy it trivially since they preserve $\mathbf{x}^2$ and leave $t$ alone. The postulates and the geometry define the same transformation set. *Verified: for the x-boost, $\Lambda^T\eta\Lambda$ gives diagonal entries $\gamma^2-\gamma^2\beta^2=1$ (time), $\gamma^2\beta^2-\gamma^2=-1$ (x), off-diagonal $-\gamma^2\beta+\gamma^2\beta=0$ — equals $\eta$ ; matches Goldstein Ch. 7 and Landau–Lifshitz CTF §6.*

Go deeper · Derivation B · 1 minFour-velocity, its normalization, and the orthogonality of four-acceleration

Derivation BFour-velocity, its normalization, and the orthogonality of four-acceleration

We construct the four-velocity properly, prove its invariant length is always $c^2$ , and derive the orthogonality $A\cdot U=0$ that constrains relativistic dynamics. The strategy: the naive $\mathrm{d}x^\mu/\mathrm{d}t$ fails the four-vector test because $\mathrm{d}t$ is frame-dependent, so we divide by the invariant $\mathrm{d}\tau$ instead; normalization then follows from the definition of proper time itself, no algebra about motion required — watch how the result is really the statement $\mathrm{d}s^2 = c^2\mathrm{d}\tau^2$ read backwards.

Step 1. Define $U^\mu = \mathrm{d}x^\mu/\mathrm{d}\tau$ . Since $\mathrm{d}x^\mu$ transforms with $\Lambda$ and $\mathrm{d}\tau$ is a scalar, $U^\mu$ transforms with $\Lambda$ : it passes the four-vector test by construction.

Step 2. Express in lab components using $\mathrm{d}\tau=\mathrm{d}t/\gamma$ from (5):

U^\mu = \frac{\mathrm{d}t}{\mathrm{d}\tau}\,\frac{\mathrm{d}x^\mu}{\mathrm{d}t} = \gamma\,\frac{\mathrm{d}}{\mathrm{d}t}(ct,\ \mathbf{x}) = \gamma\,(c,\ \mathbf{v})

Why both factors of $\gamma$ matter: the time component $\gamma c$ says the particle advances through coordinate time faster than through proper time — time dilation; the space part $\gamma\mathbf{v}$ is the source of the "extra" $\gamma$ in relativistic momentum, as the next article exploits.

Step 3. Compute the invariant square with the metric:

U\cdot U = \gamma^2\left(c^2 - \mathbf{v}\cdot\mathbf{v}\right) = \gamma^2c^2\left(1-\frac{v^2}{c^2}\right) = c^2

Why this was inevitable: $U\cdot U = (\mathrm{d}x\cdot\mathrm{d}x)/\mathrm{d}\tau^2 = c^2\mathrm{d}\tau^2/\mathrm{d}\tau^2$ . The four-velocity is the unit timelike tangent to the worldline, scaled by $c$ ; "everything moves through spacetime at speed c" is this identity, not a new dynamical fact.

Step 4. Differentiate the constant $U\cdot U=c^2$ along the worldline:

\frac{\mathrm{d}}{\mathrm{d}\tau}\left(\eta_{\mu\nu}U^\mu U^\nu\right) = 2\,\eta_{\mu\nu}\,A^\mu U^\nu = 0 \qquad\Longrightarrow\qquad A\cdot U = 0

Step 5. Interpret: four-acceleration is always Minkowski-orthogonal to four-velocity. In the particle's instantaneous rest frame $U^\mu=(c,\mathbf{0})$ , so orthogonality forces $A^0=0$ and $A^\mu=(0,\mathbf{a}_{\mathrm{proper}})$ — the invariant square $A\cdot A = -a_{\mathrm{proper}}^2$ stores the accelerometer reading, tying back to the rapidity discussion of the first article. Verified: for a particle momentarily at rest, $\gamma=1$ , $U=(c,\mathbf{0})$ , $U\cdot U=c^2$ trivially; the orthogonality reproduces the known result that pure boosts change speed while $U\cdot U$ stays pinned; matches Morin Ch. 12 and Goldstein Ch. 7.

Go deeper · Derivation C · 1 minRelative velocity from an invariant contraction

Derivation CRelative velocity from an invariant contraction

We derive the workhorse trick of collider kinematics: the relative Lorentz factor between two particles is $\gamma_{\mathrm{rel}} = U_1\cdot U_2/c^2$ , computable in any frame without ever performing a boost. The strategy exhibits the four-vector method at its best — evaluate an invariant in the frame where it is obvious (one particle's rest frame), then equate it with the evaluation in the convenient frame (the lab). Velocity addition drops out as a corollary, closing the loop with the transformation-based derivation of the first article.

Step 1. Let particles 1 and 2 have four-velocities $U_1^\mu$ , $U_2^\mu$ . The contraction $U_1\cdot U_2$ is a Lorentz scalar: its value is frame-independent by construction.

Step 2. Evaluate in the rest frame of particle 1, where $U_1^\mu=(c,\mathbf{0})$ and $U_2^\mu=\gamma_{\mathrm{rel}}(c,\mathbf{v}_{\mathrm{rel}})$ :

U_1\cdot U_2 = c\cdot\gamma_{\mathrm{rel}}c - \mathbf{0}\cdot\gamma_{\mathrm{rel}}\mathbf{v}_{\mathrm{rel}} = \gamma_{\mathrm{rel}}\,c^2

Why this frame: an invariant may be computed in whichever frame makes it trivial; the rest frame of one particle turns the contraction into a definition of the relative gamma.

Step 3. Evaluate the same scalar in the lab, for head-on motion with speeds $\beta_1,\beta_2$ (opposite directions):

U_1\cdot U_2 = \gamma_1\gamma_2\left(c^2 - \mathbf{v}_1\cdot\mathbf{v}_2\right) = \gamma_1\gamma_2\,c^2\,(1+\beta_1\beta_2)

Step 4. Equate Steps 2 and 3:

\gamma_{\mathrm{rel}} = \gamma_1\gamma_2\,(1+\beta_1\beta_2)

Step 5. Extract the relative speed via $\beta_{\mathrm{rel}}=\sqrt{1-1/\gamma_{\mathrm{rel}}^2}$ — algebra reduces it to $\beta_{\mathrm{rel}} = (\beta_1+\beta_2)/(1+\beta_1\beta_2)$ , the velocity-addition law, now obtained without differentiating any transformation. Why practitioners prefer this route: no components were boosted, no simultaneity bookkeeping was risked; one contraction plus one smart frame choice replaced the entire kinematic apparatus. This is the pattern for every invariant-mass and threshold-energy calculation in the four-momentum article. *Verified: $\beta_1=\beta_2=0.5$ head-on: $\gamma_1=\gamma_2=1.1547$ , $\gamma_{\mathrm{rel}} = 1.3333\times1.25 = 1.6667 = 5/3$ , giving $\beta_{\mathrm{rel}}=0.8$ — matching $1.0/1.25$ from the addition law; matches Landau–Lifshitz CTF §12 exercises.*

Limiting cases

Case	Result	Meaning
Low speed, $\beta\ll1$	$U^\mu\to(c,\ \mathbf{v})$ , $\mathrm{d}\tau\to\mathrm{d}t$	Four-velocity's space part becomes ordinary velocity; Newtonian kinematics is the small-tilt limit of the tangent vector.
Null separation	$\Delta s^2=0$ in every frame	The light cone is invariant; no proper time accrues along light rays — photons have no rest frame and no four-velocity.
Spacelike separation	$\Delta s^2<0$ ; time order frame-dependent	A frame exists where the events are simultaneous; none exists where they coincide spatially. Causality survives because no signal crosses the cone.
Pure rotation	$\Lambda$ block-diagonal, $t'=t$	Rotations are Lorentz transformations with zero rapidity: Euclidean geometry embeds intact in the spatial slice.
$\gamma\to\infty$	$U^\mu\to$ null direction	The unit tangent tilts toward — never onto — the cone; infinite rapidity is the unreachable edge of the group.

Common trapSignature roulette — half the books flip every sign and the invariants keep their values only if you stay consistent

This track fixes $(+,-,-,-)$ : timelike squares are positive, $U\cdot U=+c^2$ , $p\cdot p=+m^2c^2$ . Much of the field-theory literature uses $(-,+,+,+)$ , where the same physical statements read $U\cdot U=-c^2$ and $p\cdot p=-m^2c^2$ . Neither is wrong; mixing them mid-calculation is. The classic failure: importing $E^2=p^2c^2+m^2c^4$ derived in one convention into a metric contraction written in the other, and "discovering" tachyonic masses. Check the signature on page one of any text you consult, translate all imported formulas before use, and never trust the sign of a squared four-vector you did not compute yourself in your own convention.

Worked examples

Example 1 — classify, transform, and check the invariant (confidence builder). In the lab, event P occurs at $t=0,\ x=0$ and event Q at $t=5.00\ \mu\mathrm{s},\ x=900\ \mathrm{m}$ (take $c=3.00\times10^8\ \mathrm{m/s}$ ). Classify the separation, find the proper time between them, and verify the interval in a frame moving at $\beta=0.6$ .

Step 1. Convert: $c\Delta t = (3.00\times10^8)(5.00\times10^{-6}) = 1500\ \mathrm{m}$ ; $\Delta x = 900\ \mathrm{m}$ . Interval: $\Delta s^2 = 1500^2 - 900^2 = 2.25\times10^6 - 8.1\times10^5 = 1.44\times10^6\ \mathrm{m}^2 > 0$ — timelike; causal contact is possible.

Step 2. Proper time: $c\Delta\tau = \sqrt{1.44\times10^6} = 1200\ \mathrm{m}$ , so $\Delta\tau = 1200/(3.00\times10^8) = 4.00\ \mu\mathrm{s}$ . A clock travelling uniformly from P to Q (at $v = 900/5.00\ \mu\mathrm{s} = 1.8\times10^8\ \mathrm{m/s} = 0.6c$ , $\gamma=1.25$ ) reads $5.00/1.25 = 4.00\ \mu\mathrm{s}$ ✓ — consistent both ways.

Step 3. Boost at $\beta=0.6$ , $\gamma=1.25$ : $c\Delta t' = \gamma(c\Delta t - \beta\Delta x) = 1.25(1500 - 540) = 1200\ \mathrm{m}$ ; $\Delta x' = \gamma(\Delta x - \beta\,c\Delta t) = 1.25(900-900) = 0$ .

Step 4. Check: $\Delta s'^2 = 1200^2 - 0^2 = 1.44\times10^6\ \mathrm{m}^2$ ✓. The separation is timelike with $\Delta\tau=4.00\ \mu\mathrm{s}$ , and the $\beta=0.6$ frame is precisely the travelling clock's rest frame — the events happen at the same place there, 4.00 μs apart.

Example 2 — head-on relative velocity by contraction (exam level). Two protons approach head-on, each at $\beta=0.900$ in the lab. Find the Lorentz factor and speed of one proton in the other's rest frame, using the invariant $U_1\cdot U_2$ , and cross-check with velocity addition.

Step 1. Each proton has $\gamma = 1/\sqrt{1-0.81} = 1/\sqrt{0.19} = 2.294$ .

Step 2. Four-velocities, motion along $\pm x$ : $U_1^\mu = \gamma(c,\ +0.9c,0,0)$ , $U_2^\mu = \gamma(c,\ -0.9c,0,0)$ . Contract with the metric: $U_1\cdot U_2 = \gamma^2\left(c^2 - (0.9c)(-0.9c)\right) = \gamma^2 c^2(1+0.81) = (5.263)(1.81)\,c^2 = 9.526\,c^2$ .

Step 3. By Derivation C, $\gamma_{\mathrm{rel}} = U_1\cdot U_2/c^2 = 9.53$ . Speed: $\beta_{\mathrm{rel}} = \sqrt{1 - 1/\gamma_{\mathrm{rel}}^2} = \sqrt{1 - 1/90.75} = \sqrt{0.98898} = 0.99448$ .

Step 4. Cross-check with the addition law: $\beta_{\mathrm{rel}} = (0.9+0.9)/(1+0.81) = 1.8/1.81 = 0.99448$ ✓. In either proton's frame the other arrives at $0.9945c$ with $\gamma_{\mathrm{rel}}=9.53$ — compare $\gamma=2.29$ in the lab: colliding beams buy an enormous relative energy, the arithmetic behind every collider design and the reason fixed-target machines fell out of fashion.

Origins · 1 minHermann Minkowski and how this was discovered

Hermann Minkowski

1864–1909 · Göttingen

Minkowski taught Einstein mathematics at ETH Zürich — and famously remembered him as a "lazy dog" who "never bothered about mathematics at all." Nine years later the lazy dog's relativity paper prompted Minkowski's revenge of the highest order: his 1908 Cologne lecture "Space and Time" opened with the most quoted sentence in physics after Newton's — "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." He supplied the four-dimensional formalism, the invariant interval as metric geometry, the light cone, the worldline, proper time as arc length. Einstein initially dismissed it as "superfluous learnedness" — then discovered that general relativity is unwritable without it. Minkowski died of appendicitis months after the Cologne lecture, at 44, reportedly lamenting the physics he would not live to see. His name is now on the flat spacetime that every quantum field inhabits.

History

1905

— Einstein's kinematics paper: the transformation is physical, but presented as algebra on clocks and rods

1907

— Minkowski begins reformulating electrodynamics on a four-dimensional manifold with an indefinite metric

1908

— The Cologne lecture "Space and Time" introduces spacetime, worldlines, proper time, and the light cone to physics

1909

— Minkowski dies; Sommerfeld and Laue systematize four-vector algebra into the working notation of the field

1916

— Einstein builds general relativity as the curved extension of Minkowski geometry, retracting his early skepticism

Going further

References. Morin, Introduction to Classical Mechanics, Ch. 12–13 · Goldstein, Classical Mechanics (3rd ed.), Ch. 7 · Landau & Lifshitz, The Classical Theory of Fields (Vol. 2), §§1–9 · Taylor & Wheeler, Spacetime Physics, Ch. 5–6 · Schutz, A First Course in General Relativity, Ch. 1–2

Provenance. Standard four-vector formalism following Landau–Lifshitz CTF §§1–9 and Goldstein Ch. 7. Verifier should recompute: (1) the boost matrix satisfies $\Lambda^T\eta\Lambda=\eta$ using $\gamma^2(1-\beta^2)=1$ ; (2) $U\cdot U=c^2$ from $\gamma^2(c^2-v^2)$ ; (3) Example 1 (1500² − 900² = 1200² at $\beta=0.6$ ; $\Delta\tau=4.00\ \mu$ s both routes); (4) Example 2 ( $\gamma_{\mathrm{rel}}=2.294^2\times1.81=9.53$ , $\beta_{\mathrm{rel}}=0.99448$ against $1.8/1.81$ ); (5) the $\Lambda^0{}_0\ge1$ and $\det\Lambda=\pm1$ group facts from the metric condition. No nonstandard numbers introduced.

Scope note. This is the mechanics track's working introduction — enough index machinery for four-momentum, four-force, and the relativistic Lagrangian. The full story — general tensors, the electromagnetic field tensor, Wigner rotation, representation theory of the Lorentz group — lives in the Special Relativity track (migrates later as its own subject).

Builds on: Special Relativity Postulates and the Lorentz Transformation · Relativistic Kinematics: Time Dilation, Length Contraction, and Velocity Addition Leads to: Four-Velocity, Four-Momentum, and Mass-Energy Equivalence · Lagrangian Density and the Continuum Limit · Euler-Lagrange Equations for Fields · Stress Tensor and Cauchy's Relation

Does the four-velocity normalization mean everything moves through spacetime at the speed of light?

As a mnemonic, yes; as physics, it is an identity about parametrization, not a speed. $U\cdot U = (\mathrm{d}x\cdot\mathrm{d}x)/\mathrm{d}\tau^2$ , and the numerator is $c^2\mathrm{d}\tau^2$ by the definition of proper time — so $c^2$ pops out tautologically, the way a unit tangent vector always has length one. The picture it licenses is still useful: at rest, all of $U$ points along time; moving, the tangent tilts, trading time-rate ( $\mathrm{d}t/\mathrm{d}\tau=\gamma$ grows, so proper time per coordinate time shrinks) for spatial components. Just don't ask the slogan to do dynamics — nothing accelerates "through time."

Why does light have no four-velocity? The article keeps stepping around photons.

Because the construction divides by $\mathrm{d}\tau$ , and along a null worldline $\mathrm{d}\tau=0$ identically — the "clock a photon carries" never ticks. There is no rest frame to normalize in, and no finite quadruple satisfies both "tangent to a null line" and "invariant length $c^2$ ." The cure is standard: parametrize null worldlines by an affine parameter $\lambda$ instead of $\tau$ , giving a null tangent $k^\mu$ with $k\cdot k=0$ ; for a photon the physical choice is the wave four-vector $(\omega/c,\mathbf{k})$ or equivalently $p^\mu=\hbar k^\mu$ , which is a perfectly good four-vector. Massless particles have four-momentum without four-velocity — the next article uses exactly that.

If the metric is just diag of plus one and three minus ones, why bother with upper and lower indices at all in special relativity?

In SR you could survive by writing $\eta$ explicitly everywhere — the index gymnastics automates the minus signs, nothing more. Three reasons to form the habit anyway. It makes invariance visible: any expression with all indices contracted in up–down pairs is a scalar, checkable at a glance before any computation. It catches errors: an equation with a free upper index on one side and lower on the other is wrong by inspection, like a dimensional mismatch. And it is the entry fee for what follows: in general relativity the metric becomes position-dependent and raising/lowering is no longer a sign flip; the notation you learn here transfers unchanged. Ten minutes of bookkeeping now, decades of free error-checking later.

Is time literally a fourth dimension identical to space, then?

No — and the metric says precisely how it differs. Minkowski geometry treats time as a fourth coordinate, but the signature $(+,-,-,-)$ singles it out: no Lorentz transformation mixes coordinates in a way that crosses the light cone, so timelike and spacelike directions are permanently different kinds of direction. You can rotate a metre stick from north-pointing to east-pointing; no boost rotates a worldline into a spacelike line. The correct slogan is that time and space are unified the way the components of one vector are unified — inseparable, jointly transforming, but not interchangeable. The single minus sign carries causality, the arrow between cause and effect, and the impossibility of standing still in time.

Which events can a frame make simultaneous, and which can it make co-located?

Read it off the interval's sign. Spacelike pair ( $\Delta s^2<0$ ): boosts can set $\Delta t'=0$ — a frame exists where they are simultaneous — but never $\Delta\mathbf{x}'=0$ ; their invariant proper distance $\sqrt{-\Delta s^2}$ is the separation in that frame. Timelike pair ( $\Delta s^2>0$ ): the reverse — some frame (the travelling clock's) sees them at the same place, separated by the proper time $\Delta\tau=\sqrt{\Delta s^2}/c$ , but no frame sees them simultaneous. Null pair: neither; every frame agrees a light ray exactly connects them. The classification is the complete answer to "what can boosts do" — they slide events along their hyperbola, and the hyperbola never leaves its causal class.

The tldr says contractions with one upper and one lower index are automatically invariant. Why does that work?

Because the two indices transform inversely to each other. An upper index picks up $\Lambda$ ; lowering with the metric and using the defining condition $\Lambda^T\eta\Lambda=\eta$ shows a lower index picks up the inverse matrix. In the contraction $A^\mu B_\mu$ the $\Lambda$ and $\Lambda^{-1}$ meet and cancel: $A'^\mu B'_\mu = \Lambda^\mu{}_\alpha A^\alpha\,(\Lambda^{-1})^\beta{}_\mu B_\beta = \delta^\beta{}_\alpha A^\alpha B_\beta = A^\alpha B_\alpha$ . That cancellation is the entire engineering of the notation: build any expression with matched index pairs and Lorentz invariance is guaranteed structurally, before numbers are inserted. It is the same reason $\mathbf{a}\cdot\mathbf{b}$ survives rotations in Euclidean space, generalized to an indefinite metric.

Are Newtonian 3-vectors the spatial parts of four-vectors? Force, for instance?

Sometimes, and the exceptions matter. Rotations sit inside the Lorentz group, so the spatial part of any four-vector is automatically a good 3-vector under rotations — but a given Newtonian 3-vector need not extend to a four-vector with its familiar components intact. Velocity extends only after re-parametrizing by proper time (picking up factors of $\gamma$ ); momentum extends as $\gamma m\mathbf{v}$ , not $m\mathbf{v}$ ; force extends as the four-force $\gamma(\mathbf{F}\cdot\mathbf{v}/c,\ \mathbf{F})$ , the subject of two articles hence. Some 3-vectors refuse entirely: the electric field $\mathbf{E}$ is not the spatial part of any four-vector — it shares a rank-2 tensor with $\mathbf{B}$ . The four-vector test, not habit, decides.

Why signature plus-minus-minus-minus here when field theory books often use the opposite?

Pure community convention, with real bookkeeping consequences. Particle physics and this mechanics track prefer $(+,-,-,-)$ because timelike quantities — proper time, $U\cdot U$ , $p\cdot p=m^2c^2$ — come out positive, pleasant for mechanics. General relativity and much modern field theory prefer $(-,+,+,+)$ because spatial geometry embeds with positive signs. All physics is identical; every squared four-vector, every " $\pm$ " in a Lagrangian mass term, flips between conventions. The professional habit: identify the signature within the first page of anything you read, and translate imported formulas immediately — as the warning box says, most "paradoxes" involving stray minus signs are two conventions colliding in one calculation.