Lorentz Transformations

Newtonian mechanics answers the question "how do two observers in relative motion compare their coordinates?" with the Galilean transformation: leave the clock alone, and add velocities like arrows. It is intuitive, and it is wrong. The whole edifice collapses the moment you insist — with Einstein — that the same light pulse races away at exactly $c$ whether you chase it or flee from it. A transformation that keeps one speed fixed in every frame cannot leave time untouched; it must braid time and space together. The Lorentz transformations are precisely the linear coordinate changes that perform this braiding while holding one thing sacred: the spacetime interval $Δs² = -c²Δt² + Δx² + Δy² + Δz²$ . That single invariance is the seed of everything. Demanding it for all displacements forces the transformation matrix $Λ$ to satisfy one compact equation, $Λ^{\mathsf T}ηΛ = η$ — the algebraic heart of special relativity. Time dilation, length contraction, the relativity of simultaneity, the cosmic speed limit: none of them are separate postulates. They are corollaries of that one line. The solutions form a six-parameter group — three rotations, three boosts — and a boost turns out to be nothing more exotic than a rotation through an imaginary angle, a hyperbolic tilt of the axes toward the light cone they can never cross. This matters far beyond flat space. In general relativity, spacetime is curved, but at every event the tangent space is Minkowski, and physics in a freely-falling local frame is Lorentz-invariant. Special relativity is the local limit of the curved theory, and the residual freedom to rotate orthonormal frames — local Lorentz invariance — is the gauge symmetry that lets spinors and tetrads live in curved spacetime.

Key takeaways

A Lorentz transformation is a linear map $x'^μ = Λ^μ{}_ν x^ν$ between inertial frames that leaves the spacetime interval $Δs² = -c²Δt² + Δx² + Δy² + Δz²$ invariant.
Invariance is equivalent to the single matrix condition $Λ^{\mathsf T}ηΛ = η$ with $η = \mathrm{diag}(-1,1,1,1)$ ; the solutions form the Lorentz group $O(1,3)$ .
The group has four disconnected components; the physical, identity-connected piece $SO^{+}(1,3)$ (proper, orthochronous) is built from three rotations and three boosts.
A boost mixes time and space, $ct' = γ(ct - βx)$ , $x' = γ(x - β\,ct)$ , with $γ = 1/\sqrt{1-β²}$ — this one map contains dilation, contraction, and lost simultaneity.
Boosts are hyperbolic rotations by a rapidity $φ$ with $\tanh φ = β$ ; rapidities add linearly, so relativistic velocity addition is just $\tanh$ of a sum.
Key idea: a Lorentz transformation is any linear map $x'^μ = Λ^μ{}_ν x^ν$ obeying $Λ^{\mathsf T}ηΛ = η$ ; that one condition encodes every boost and rotation and guarantees the interval and the speed of light are identical in every inertial frame.

The core insight

A Lorentz transformation is any linear map $x'^μ = Λ^μ{}_ν\,x^ν$ that preserves the Minkowski metric, $Λ^{\mathsf T}ηΛ = η$ ; this single condition encodes boosts and rotations and guarantees the spacetime interval and the speed of light are the same in every inertial frame.

Where this lives today

Lorentz invariance is among the most stringently tested principles in all of physics, and probing where — if anywhere — it breaks is a live frontier. Modern "clock-comparison" and optical-cavity experiments (Michelson–Morley redux with lasers and atomic clocks) now bound any anisotropy of the speed of light at the level of parts in $10^{18}$ , and astrophysical tests watch high-energy photons and neutrinos race across billions of light-years to check whether the light barrier is truly energy-independent, as exact Lorentz symmetry demands. On the theory side, the Standard-Model Extension catalogues every way Lorentz violation could appear, motivated by quantum-gravity scenarios in which spacetime is granular at the Planck scale and the smooth Minkowski structure is only an approximation. Meanwhile the transformation does daily engineering work: particle physicists boost between lab and centre-of-momentum frames to extract collision energies at the LHC, and every relativistic-jet observation from a blazar or microquasar is decoded with Doppler and aberration formulas that are just the boost acting on a light four-vector. The humble matrix Lorentz wrote to rescue the ether now underwrites both the deepest tests of symmetry and the routine bookkeeping of high-energy physics.

The physical picture

Draw spacetime as a flat sheet with time running up and one space direction running across. In ordinary Euclidean geometry a rotation keeps $x² + y²$ fixed and sweeps points around a circle. A Lorentz boost keeps the Minkowski "length" $-c²t² + x²$ fixed — and because that invariant carries a minus sign, the level curves are hyperbolae, not circles. The "rotation" is therefore hyperbolic: as you boost harder, the time axis and the space axis both tilt toward the diagonal light line $x = ct$ , closing on it like a pair of scissors but never crossing. The speed of light is the asymptote no boost can reach, and $γ$ is a single number measuring how far the axes have tilted.

Figure

Minkowski diagram of a boost with β = 0.6: the primed axes ct′ and x′ tilt symmetrically toward the 45° light line x = ct by rapidity φ = 0.69. The green hyperbola Δs² = const is the locus a boost slides events along; the unit proper-time tick on ct′ sits at coordinate ct = γ = 1.25

The Galilean world is the $c \to \infty$ limit of this picture: push the light line flat and the cone opens into a horizontal plane of absolute simultaneity shared by all observers. At finite $c$ the cone closes to $45°$ , simultaneity fractures — events level in one frame slope in another — and the tilt of the axes becomes physical. The most elegant payoff is composition: stack two boosts in the same direction and their rapidities simply add, exactly as two planar rotations add their angles. The notorious relativistic velocity-addition law, which forbids any sum of sub-light speeds from reaching $c$ , is nothing but the statement that $\tanh$ of a finite sum is always less than one.

Figure

Velocity as the hyperbolic tangent of rapidity, β = tanh φ: rapidity (blue) is the natural, additive angle of a boost, but the physical speed β saturates at the light barrier β = 1 (red). Two collinear boosts of φ = 0.69 each (β = 0.6) add to φ = 1.39, giving β = 0.88, never reaching c

The core equations

Everything in special relativity hangs from one invariant. Before any transformation is written, the object that must survive it is the spacetime interval between two events — the Lorentzian analogue of a distance, with the crucial minus sign on the time part.

Δs² = η_{μν}\,Δx^μ Δx^ν = -c²Δt² + Δx² + Δy² + Δz² \qquad (1)

where: $Δs²$ — squared spacetime interval between two events (signature $-+++$ ); $η_{μν}$ — Minkowski metric $\mathrm{diag}(-1,1,1,1)$ ; $Δx^μ$ — coordinate separation with $x^0 = ct$ , $(x^1,x^2,x^3) = (x,y,z)$ ; $c$ — invariant speed of light. Physical reading: $Δs² < 0$ is timelike (a clock can be present at both events), $Δs² > 0$ is spacelike (no signal connects them), $Δs² = 0$ is a light ray. A Lorentz transformation is then defined to be any linear map that leaves equation (1) unchanged. Imposing that on the matrix $Λ$ collapses to a single relation — the defining equation of the whole group, the exact Lorentzian cousin of the orthogonality condition $R^{\mathsf T}R = 𝟙$ that defines rotations.

Λ^{\mathsf T}\,η\,Λ = η \qquad\Longleftrightarrow\qquad η_{μν}\,Λ^μ{}_ρ\,Λ^ν{}_σ = η_{ρσ} \qquad (2)

where: $Λ^μ{}_ν$ — transformation matrix, $x'^μ = Λ^μ{}_ν x^ν$ ; $Λ^{\mathsf T}$ — transpose; $η$ — the metric read as a matrix. Physical reading: the only maps allowed are those that preserve the metric, so the "distances" all observers agree on are exactly the Minkowski intervals. The single scalar that governs how strongly a boost tilts the axes is the Lorentz factor. It is the one number that appears in nearly every relativistic formula, growing without bound as $v$ approaches $c$ .

γ = \frac{1}{\sqrt{1 - v²/c²}} = \frac{1}{\sqrt{1-β²}}, \qquad β = \frac{v}{c} \qquad (3)

where: $γ$ — Lorentz factor, always $γ \geq 1$ ; $v$ — relative speed of the two inertial frames; $β = v/c$ — speed in units of $c$ . Physical reading: $γ$ is the stretch of moving clocks and the shrink of moving rods; at $β = 0.6$ , $γ = 1.25$ , and $γ \to \infty$ as $β \to 1$ . Specialising $Λ$ to relative motion along a single axis gives the workhorse of the subject — the boost. The transverse directions are untouched; only $t$ and $x$ mix.

\begin{aligned} ct' &= γ\,(ct - β x) \\ x' &= γ\,(x - β\, ct) \\ y' &= y, \quad z' = z \end{aligned} \qquad (4)

where: $(ct',x',y',z')$ — coordinates in the frame moving at $+v\hat{x}$ ; $β = v/c$ . Physical reading: the $-γβx$ piece of the time equation is the relativity of simultaneity — events at different $x$ but equal $t$ acquire different $t'$ — and it is the root of nearly every "paradox". Written as a matrix, the boost reveals its true nature. The velocity form and the rapidity form are the same object; the second makes the hyperbolic-rotation structure manifest.

Λ = \begin{pmatrix} γ & -γβ & 0 & 0 \\ -γβ & γ & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{pmatrix} = \begin{pmatrix} \cosh φ & -\sinh φ & 0 & 0 \\ -\sinh φ & \cosh φ & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{pmatrix} \qquad (5)

where: $φ$ — rapidity, defined by $\tanh φ = β$ , so $\cosh φ = γ$ and $\sinh φ = γβ$ . Physical reading: compare with the Euclidean rotation matrix built from $\cos$ / $\sin$ — the only change is trigonometric functions become hyperbolic, because the invariant is $-c²t²+x²$ instead of $x²+y²$ . The rapidity is the "natural angle" of a boost, and its defining relations are worth isolating. Its whole virtue is that it linearises composition.

\tanh φ = β = \frac{v}{c}, \qquad γ = \cosh φ, \qquad γβ = \sinh φ \qquad (6)

where: $φ$ — rapidity, ranging over $(-\infty , \infty )$ while $v$ ranges over $(-c, c)$ . Physical reading: because $\tanh$ maps the whole real line into $(-1,1)$ , no finite rapidity — hence no finite chain of boosts — ever reaches $β = 1$ . Two collinear boosts compose by adding rapidities, and unpacking that with the $\tanh$ addition formula reproduces the relativistic velocity-addition law.

φ_{\text{tot}} = φ_1 + φ_2 \quad\Longrightarrow\quad β_{\text{tot}} = \frac{β_1 + β_2}{1 + β_1 β_2} \qquad (7)

where: $φ_i$ — rapidity of each boost; $β_i$ — its velocity in units of $c$ . Physical reading: the denominator $1 + β_1β_2$ is the relativistic correction that keeps the sum below $c$ ; for small speeds it is $\approx 1$ and Galilean addition $v_{\text{tot}} = v_1 + v_2$ re-emerges. Finally, taking the determinant and the $00$ -component of equation (2) exposes that the Lorentz group is not connected — it splits into four topological pieces.

\det Λ = \pm 1, \qquad (Λ^0{}_0)² \geq 1 \qquad (8)

where: $\det Λ$ — $+1$ proper, $-1$ improper (contains a reflection); $Λ^0{}_0$ — $\geq +1$ orthochronous (preserves the arrow of time), $\leq -1$ non-orthochronous. Physical reading: only the proper orthochronous component $SO^{+}(1,3)$ is reachable by a continuous boost or rotation from rest — the rest require the discrete flips of parity or time reversal.

Derivation ADeriving the boost from the two postulates

The goal is to build the boost from scratch using nothing but three physical inputs — the homogeneity of spacetime (which forces linearity), the principle of relativity (which forces a symmetric, reciprocal form), and the invariance of $c$ (which fixes the last constant) — and then to pin the overall scale with the interval. Working in standard configuration, with the primed frame sliding at speed $v$ along the shared $x$ -axis and the origins coinciding at $t = t' = 0$ , we start from the most general linear ansatz and let each postulate remove one unknown until only $γ$ remains, itself fixed by demanding the interval be invariant.

Step 1. Write the most general linear map in the $t$ - $x$ plane, $x' = A x + B(ct)$ and $ct' = C x + D(ct)$ . Why linear: spacetime is homogeneous, so a free particle in uniform motion in one frame must be in uniform motion in the other — straight worldlines map to straight worldlines, and only linear maps through a common origin do that.

Step 2. The spatial origin of the primed frame, $x' = 0$ , traces $x = vt$ in the unprimed frame; substituting gives $Avt + Bct = 0$ , hence $B = -βA$ . Why: this simply encodes that the moving observer sits at rest at their own origin while drifting at $v$ through ours, tying $B$ to the velocity.

Step 3. By the principle of relativity the inverse map must have the identical form with $v \to -v$ ; this reciprocity forces the two diagonal stretch factors to be equal, $D = A ≡ γ$ . Why: neither frame is privileged, so the map from primed-to-unprimed cannot look structurally different from unprimed-to-primed — symmetry equates the coefficients.

Step 4. Demand that a light pulse obey $x = ct$ in both frames, i.e. $x = ct \Rightarrow x' = ct'$ ; feeding this into the ansatz with $B = -βA$ fixes the off-diagonal time coefficient to $C = -γβ$ . Why: this is where Einstein's second postulate enters — the constancy of $c$ is the one experimental fact that distinguishes Lorentz from Galileo.

Step 5. Impose interval invariance $-c²t'² + x'² = -c²t² + x²$ on the assembled map; the cross terms cancel automatically and the diagonal terms give $γ²(1 - β²) = 1$ , so $γ = 1/\sqrt{1-β²}$ , yielding $ct' = γ(ct - βx)$ and $x' = γ(x - β\,ct)$ . Why: the first four steps fixed the shape of the matrix; only the interval fixes its overall normalisation, delivering the familiar $γ$ . *Three quick checks confirm the result: setting $β \to 0$ returns $γ \to 1$ , $t' \to t$ , $x' \to x$ (the identity); a light ray $x = ct$ maps to $x' = ct'$ , preserving the speed of light; and $\det Λ = γ² - γ²β² = γ²(1-β²) = 1$ , confirming the transformation is proper.*

Go deeper · Derivation B · 1 minTime dilation and length contraction from one boost

Derivation BTime dilation and length contraction from one boost

The goal is to show that the two most famous kinematic effects are not independent postulates but two readings of the very same boost equations, the difference lying entirely in which pair of events each measurement compares. The subtlety — and the source of every apparent paradox — is that a clock measurement holds the spatial coordinate fixed in the clock's own frame, while a length measurement holds the time coordinate fixed in the observer's frame; those are different constraints, and reversing them is the classic error. We use the inverse boost $ct = γ(ct' + βx')$ , $x = γ(x' + β\,ct')$ throughout.

Step 1. For a clock at rest in $S'$ , two successive ticks happen at the same place, $Δx' = 0$ , and the elapsed primed time is the proper time, $Δτ = Δt'$ . Why: proper time is what a single carried clock reads, and a clock never moves relative to itself — that is exactly the $Δx' = 0$ condition.

Step 2. Transform to $S$ : $Δt = γ(Δt' + β\,Δx'/c) = γ\,Δt' = γ\,Δτ$ , so the moving clock's ticks are spread by the larger interval $γΔτ$ in the lab. Why: with $Δx' = 0$ the simultaneity term drops out and only the $γ$ stretch survives — moving clocks run slow, and the effect is purely $γ$ .

Step 3. For a rod at rest in $S'$ with rest length $L_0 = Δx'$ , the observer in $S$ must locate both ends at one instant, $Δt = 0$ . Why: a moving object's length is defined by simultaneous marking of its ends — measuring them at different times would let the rod slide between marks and give a meaningless answer.

Step 4. Use $Δx' = γ(Δx - β c\,Δt) = γ\,Δx$ with $Δt = 0$ , so the measured length is $L = Δx = L_0/γ$ , shorter than rest length. Why: here it is the space transform, evaluated at equal lab-time, that carries the $γ$ ; the rod contracts along the direction of motion by exactly the reciprocal factor. *As a concrete check: at $v = 0.6c$ , $γ = 1.25$ , so a moving clock's proper second reads $1.25\,\mathrm{s}$ of lab time, and a metre stick measures $0.80\,\mathrm{m}$ in the lab.*

Go deeper · Derivation C · 1 minThe four components of the Lorentz group

Derivation CThe four components of the Lorentz group

The goal is to prove that the defining condition $Λ^{\mathsf T}ηΛ = η$ does not describe one connected family of transformations but four disjoint sheets, separated by two discrete signs — the determinant and the sign of the time-time entry. Only one sheet, the proper orthochronous component containing the identity, is physically reachable by continuously accelerating a laboratory; the other three demand the discrete operations of spatial reflection, time reversal, or both. We extract the two invariants directly from the defining relation and count.

Step 1. Take the determinant of both sides of $Λ^{\mathsf T}ηΛ = η$ : since $\det η \neq 0$ , we get $(\det Λ)² = 1$ , so $\det Λ = \pm 1$ . Why: the determinant is a continuous function that can only take two isolated values here, so it can never change sign along a continuous path — it already cuts the group in two (proper vs. improper, the latter containing parity $P$ ).

Step 2. Read off the $00$ -component: $-(Λ^0{}_0)² + \sum_i (Λ^i{}_0)² = -1$ , hence $(Λ^0{}_0)² \geq 1$ , forcing $Λ^0{}_0 \geq +1$ or $Λ^0{}_0 \leq -1$ . Why: the forbidden gap $-1 < Λ^0{}_0 < 1$ is unbridgeable by any continuous motion, splitting each half again into orthochronous (arrow of time preserved) and non-orthochronous (time-reversed, containing $T$ ).

Step 3. Combine the two signs to get four sheets, $O(1,3) = SO^{+} ∪ P\,SO^{+} ∪ T\,SO^{+} ∪ PT\,SO^{+}$ ; the identity component is $SO^{+}(1,3)$ with $\det Λ = +1$ and $Λ^0{}_0 \geq +1$ . Why: this is the only physically realisable component — the proper orthochronous boosts and rotations — and it is a six-dimensional Lie group.

Step 4. Any element of $SO^{+}(1,3)$ is an exponential $Λ = \exp\!\big(\tfrac12 ω_{μν} M^{μν}\big)$ , with six antisymmetric parameters $ω_{μν}$ splitting into three rotation angles and three rapidities, and generators $M^{μν}$ spanning the Lie algebra $\mathfrak{so}(1,3)$ . Why: the exponential map builds every finite proper orthochronous transformation from infinitesimal ones, exhibiting the promised three-plus-three parameter count. *As a consistency note: $SO^{+}(1,3)$ is six-dimensional and is double-covered by $SL(2,\mathbb{C})$ , confirming the four-component structure and the six-parameter count.*

Limiting cases

Case	Result	Meaning
Low velocity (Galilean limit)	$v ≪ c \Rightarrow γ \to 1,\ t' \to t,\ x' \to x - vt$	Recovers the Galilean transformation and Newtonian absolute time; corrections are $O(v²/c²)$ .
Ultra-relativistic limit	$v \to c \Rightarrow γ \to \infty ,\ φ \to \infty$	Rapidity diverges, the boosted axes asymptote to the light line, and no inertial frame reaches $v = c$ .
Zero velocity (identity)	$v = 0 \Rightarrow Λ = 𝟙$	Frames at mutual rest differ at most by a rotation; the boost collapses to the identity map.
Light propagation	$x = ct \Rightarrow x' = ct'$	The light cone maps to itself — the speed of light is the same invariant $c$ in every frame, the defining demand.

Common trapTime dilation and length contraction are symmetric "who is really slow" statements

They are not. The factor $γ$ is never below $1$ , but that does not mean one observer is objectively slow or short while the other is objectively normal. Each effect compares a different pair of events. A clock is dilated only when its two ticks occur at the same place in its own frame ( $Δx' = 0$ ); a rod is contracted only when its two ends are marked simultaneously in the observing frame ( $Δt = 0$ ). Because "same place" and "same time" are themselves frame-dependent — the $-γβx$ simultaneity term in $t' = γ(t - vx/c²)$ — each observer legitimately finds the other's clock slow and rod short, with no contradiction. Forgetting the relativity of simultaneity built into that term is the root of nearly every twin, ladder, or pole-and-barn "paradox".

Worked examples

1 · Composing two boosts with rapidity (fully numeric)

A probe moves at $β_1 = 0.6$ relative to a station. It launches a drone forward at $β_2 = 0.6$ relative to itself. How fast is the drone relative to the station? Do it twice — the naive way and the rapidity way.

Step 1. Naive Galilean addition would give $β_1 + β_2 = 1.2$ , i.e. faster than light — impossible, and the tell that Galilean addition has failed. The relativistic law is equation (7): $β_{\text{tot}} = (β_1 + β_2)/(1 + β_1β_2)$ .

Step 2. Plug in: $β_{\text{tot}} = (0.6 + 0.6)/(1 + 0.6\cdot 0.6) = 1.2/1.36 = \mathbf{0.882}$ . The drone moves at $0.882c$ relative to the station — below $c$ , as it must be.

Step 3. Now the elegant route. Each boost has rapidity $φ = \tanh^{-1}(0.6) = \mathbf{0.6931}$ . Rapidities simply add: $φ_{\text{tot}} = 0.6931 + 0.6931 = \mathbf{1.3863}$ . Convert back: $β_{\text{tot}} = \tanh(1.3863) = \mathbf{0.882}$ — identical to Step 2, but obtained by plain addition rather than the awkward fraction.

Step 4. The lesson: rapidity is the additive coordinate of boosts. Even stacking a hundred boosts of $0.6c$ each just gives $φ = 69.3$ , and $\tanh(69.3)$ is indistinguishable from $1$ but never equals it — you approach $c$ asymptotically and never reach it.

2 · The atmospheric muon (fully numeric)

Cosmic-ray muons are created about $10\,\mathrm{km}$ up with a rest-frame mean lifetime $τ_0 = 2.2\,μ\mathrm{s}$ , travelling at $β = 0.998$ . Non-relativistically a muon covers only $c\,τ_0 \approx 0.66\,\mathrm{km}$ before decaying — it should never reach the ground. Resolve the puzzle from both frames.

Step 1. Lorentz factor: $γ = 1/\sqrt{1 - 0.998²} = 1/\sqrt{0.003996} = \mathbf{15.82}$ . This single number will do all the work, appearing as a time stretch in one frame and a length shrink in the other.

Step 2. Lab (ground) frame — time dilation. The moving muon's clock runs slow, so its lab-frame mean life is $γτ_0 = 15.82 \times 2.2\,μ\mathrm{s} = \mathbf{34.8\,μ\mathrm{s}}$ . In that time it travels $v\cdot γτ_0 = 0.998\cdot (3\times 10^8)\cdot (34.8\times 10^{-6}) \approx \mathbf{10.4\,km}$ — comfortably enough to reach sea level.

Step 3. Muon frame — length contraction. Here the muon lives only its ordinary $2.2\,μ\mathrm{s}$ , but the atmosphere rushes past contracted by $γ$ : the $10\,\mathrm{km}$ column is only $L_0/γ = 10/15.82 = \mathbf{0.63\,km}$ thick. Crossing $0.63\,\mathrm{km}$ at $0.998c$ takes $0.63\times 10^3/(0.998\cdot 3\times 10^8) \approx \mathbf{2.1\,μ\mathrm{s}}$ , about one lifetime.

Step 4. Both frames agree the muon arrives, but they attribute survival to different effects — dilation of the muon's clock in the lab, contraction of the atmosphere in the muon's frame. They are the same boost read two ways, and the observed muon flux at sea level confirms it daily.

Physics of the portrait

The transformation carries a Dutch physicist's name, yet he was not the one who understood what he had found — a fitting emblem for a formula that was written down to save an ether nobody would keep.

Origins · 1 minHendrik Lorentz and how this was discovered

Hendrik Lorentz

1853–1928 · electron theory and the electrodynamics of moving bodies

Hendrik Antoon Lorentz was the towering theorist of the generation just before Einstein, and the man whose mathematics made relativity possible without ever quite making relativity himself. Working to explain why the Michelson–Morley experiment stubbornly failed to detect Earth's motion through the luminiferous ether, Lorentz introduced by 1895 two remarkable devices: a "local time" $t' = t - vx/c²$ that varied from place to place, and — following FitzGerald — a contraction of moving bodies along their direction of travel. By 1904 he had assembled these into the full set of transformation equations that now bear his name, showing that Maxwell's equations kept their form under them. But for Lorentz these were computational fictions: real physical distortions of matter moving through a real, stationary ether, with "local time" a mere auxiliary variable, not the time any clock actually read. It fell to Poincaré, who in 1905 named the group after Lorentz and recognised its mathematical structure, and above all to Einstein, who that same year discarded the ether entirely and reinterpreted the very same equations as statements about space and time themselves, flowing from two simple postulates. Minkowski completed the arc in 1908 by revealing the transformations as rotations of a unified four-dimensional spacetime. Lorentz, generous and undogmatic, embraced the new physics he had unwittingly prepared, and remained until his death the revered elder statesman of theoretical physics — the calculation was his; the revolution was Einstein's.

History

1887

— Voigt writes transformations that leave the wave equation form-invariant

1889

— FitzGerald proposes length contraction to explain the Michelson–Morley null result

1895

— Lorentz introduces local time and the contraction hypothesis in his electron theory

1904

— Lorentz publishes the transformation equations in essentially their modern form

1905

— Einstein derives them from two postulates; Poincaré names them after Lorentz

1908

— Minkowski recasts them as rotations of four-dimensional spacetime

Going further

Texts. Misner, Thorne & Wheeler, Gravitation, ch. 2 · Wald, General Relativity, §1.2 · Carroll, Spacetime and Geometry, ch. 1 · Hartle, Gravity, ch. 4 · Schutz, A First Course in General Relativity, ch. 1–2. Key numerical anchors: direct multiplication confirms $Λ^{\mathsf T}ηΛ = η$ for the boost matrix (5); the identity $γ²(1-β²) = 1$ yields the rapidity identifications $γ = \cosh φ$ and $γβ = \sinh φ$ . Numerical values: $γ = 1.25$ at $β = 0.6$ ; $γ = 1.667$ at $β = 0.8$ , giving $Δt = 5\,\mathrm{s}$ for $Δτ = 3\,\mathrm{s}$ ; muon $γ = 15.82$ at $β = 0.998$ with $τ_0 = 2.2\,μ\mathrm{s}$ ; and the composition $0.6 ⊕ 0.6 = 0.882$ . For small $β$ , equation (7) reduces to Galilean addition $β_{\text{tot}} \approx β_1 + β_2$ .

Prerequisites. Postulates of Special Relativity · Spacetime Interval · Minkowski Metric and Signature. Leads to. Lorentz Boosts and Rapidity · Time Dilation · Length Contraction · Four-Vectors · Causal Structure and Light Cones.

Ask the tutor: Ask me to boost any pair of events by hand, resolve a specific "paradox" (twins, ladder-and-barn, pole-vaulter) event by event, or show why $Λ^{\mathsf T}ηΛ = η$ forces the boost matrix to look exactly the way it does.

Why must a Lorentz transformation be linear in the coordinates?

Because spacetime is homogeneous — no event is special. A free particle moving uniformly in one inertial frame must also move uniformly in another, so straight worldlines must map to straight worldlines. The only coordinate maps that send every straight line through the origin to another straight line (with origins coinciding) are linear, $x'^μ = Λ^μ{}_ν x^ν$ . If the map had quadratic or higher terms, uniform motion would look accelerated in the new frame, secretly singling out a preferred location or time and violating the relativity principle. Linearity is therefore not an assumption of convenience but a consequence of the equivalence of all inertial frames.

How is

Λ^{\mathsf T}ηΛ = η

different from the rotation condition

R^{\mathsf T}R = 𝟙

They are the same idea with a different metric. Rotations preserve the Euclidean quadratic form $x² + y² + z²$ , giving $R^{\mathsf T}𝟙R = 𝟙$ . Lorentz transformations preserve the Minkowski form $-c²t² + x² + y² + z²$ , so the identity is replaced by $η = \mathrm{diag}(-1,1,1,1)$ and the condition becomes $Λ^{\mathsf T}ηΛ = η$ . That single minus sign changes everything: the invariant "spheres" $x²+y²=\text{const}$ become invariant hyperbolae $-c²t²+x²=\text{const}$ , and the rotation angle, bounded and periodic, is replaced by the rapidity, unbounded and additive. Boosts are rotations through an imaginary angle.

What exactly is rapidity, and why bother with it?

Rapidity $φ$ is the "natural angle" of a boost, defined by $\tanh φ = β = v/c$ . Its payoff is that collinear boosts add their rapidities, $φ_{\text{tot}} = φ_1 + φ_2$ , exactly as planar rotations add their angles — whereas velocities combine through the messy fraction $(β_1+β_2)/(1+β_1β_2)$ . Because $\tanh$ maps the entire real line into $(-1,1)$ , any finite rapidity, however large, corresponds to a speed below $c$ : this makes the cosmic speed limit geometrically obvious. Rapidity also cleans up the boost matrix, which becomes $\cosh φ$ and $\sinh φ$ in perfect analogy to the $\cos$ and $\sin$ of a rotation. It is the additive, "flat" coordinate hiding inside the saturating variable $β$ .

Do time dilation and length contraction contradict each other since each observer sees the other's clock slow?

No — this apparent paradox dissolves once you track which events are compared. Time dilation and length contraction are statements about different pairs of events, and the definitions are not reciprocal in a naive way because "at the same time" is frame-dependent. When you say the other observer's clock is slow, you are comparing events at the same place in their frame; when they say the same of you, they compare events at the same place in their frame — different comparisons entirely. The relativity of simultaneity, the $-γβx$ term in the boost, is exactly the extra ingredient that reconciles the two symmetric-looking claims. Every "twin paradox" is resolved by carefully applying the boost to specific events rather than waving at "who moves".

Why can no chain of sub-light boosts ever reach the speed of light?

Because boosts add rapidities, not velocities. Each sub-light boost contributes a finite rapidity $φ_i = \tanh^{-1}β_i$ , and any finite sum $φ_{\text{tot}} = \sum φ_i$ is still finite. The resulting speed is $β_{\text{tot}} = \tanh φ_{\text{tot}}$ , and $\tanh$ approaches $1$ asymptotically but never attains it for finite argument. Concretely, stacking a hundred boosts of $0.6c$ each gives rapidity $\approx 69.3$ , and $\tanh(69.3)$ differs from $1$ only around the sixtieth decimal place — still short of $c$ . Reaching $β = 1$ would require infinite rapidity, hence infinitely many boosts or infinite energy. The light barrier is not a wall you crash into; it is a horizon you asymptote toward.

What survives a Lorentz transformation, and what does not?

The spacetime interval $Δs²$ survives — every inertial observer computes the same value — and with it the speed of light, the causal ordering of timelike-separated events, and the light-cone classification (timelike, spacelike, null). What does not survive are the separate time and space pieces: $Δt$ and $Δx$ individually change, which is precisely time dilation and length contraction. Simultaneity is frame-dependent, and so is which of two spacelike-separated events happens "first". The whole art of relativity is learning to phrase physics in terms of the invariants — intervals, proper times, four-vector magnitudes — rather than the frame-dependent components, so that predictions never depend on an arbitrary choice of observer.

In what sense do Lorentz transformations still matter in curved spacetime?

General relativity replaces flat Minkowski spacetime with a curved manifold, but curvature is a global feature — locally, spacetime is always flat. At every event the tangent space is Minkowski, and an observer in free fall there experiences exactly special relativity (the equivalence principle). One builds an orthonormal frame, a set of four tetrad vectors, at each event, and there remains the freedom to rotate or boost that frame by a position-dependent Lorentz transformation $Λ^a{}_b(x)$ without changing any physics. This local Lorentz invariance is a genuine gauge symmetry, and it is the machinery that lets spinor fields — electrons, quarks — be defined on curved spacetime, since spinors are representations of the Lorentz group, not of general coordinate transformations. Lorentz symmetry is thus the local skeleton of the curved theory.