Lorentz Invariance and Invariants

Two events. One observer, at rest in a laboratory, records them as separated by $5$ seconds and $3$ light-seconds. A second observer, cruising past at $0.6c$ , records the same two events as separated by $4$ seconds and — nothing at all. Zero distance. In her frame the two events happen at the same place, one after the other, like two ticks of a clock she is carrying. Every number has changed: the elapsed time, the spatial separation, even the fact that the events happened "somewhere else". And yet if each observer forms the combination $c^2\Delta t^2 - \Delta x^2$ , they both get $16$ — exactly, and so would any other observer at any other speed. The coordinates are theirs. The number $16$ belongs to the events. That number is the interval, the first and most important of the Lorentz invariants: the quantities on which all inertial observers are compelled to agree. The compulsion is not an accident of the algebra. It is built into the definition of what a change of inertial frame is — a Lorentz transformation is, by construction, a linear map that leaves the metric $\eta$ alone, so everything assembled purely out of $\eta$ and four-vectors is stuck with the same value in every frame. Felix Klein had said the general thing decades before relativity existed: a geometry is the invariant theory of a transformation group. Minkowski spacetime is the geometry whose group is the Lorentz group, and this article is about what that group leaves behind.

Key takeaways

The signature here is $(+,-,-,-)$ : $\Delta s^2 = c^2\Delta t^2 - \Delta x^2 - \Delta y^2 - \Delta z^2 = \eta_{\mu\nu}\Delta x^\mu \Delta x^\nu$ , so timelike means $s^2 > 0$ and spacelike means $s^2 < 0$ . (The AllQFT base for this slug is written in the opposite convention and has this classification backwards; every sign below is re-derived.)
A Lorentz transformation is defined as a linear map with $\Lambda^{\mathsf T}\eta\,\Lambda = \eta$ . Invariance is then not a discovery but a tautology: any full contraction of four-vectors with $\eta$ has the same value in every inertial frame.
These maps form a group — closed under composition, with the explicit inverse $\Lambda^{-1} = \eta\,\Lambda^{\mathsf T}\eta$ . Klein's Erlangen framing applies exactly: Minkowski geometry is this group's invariant theory.
Owned result: $\Lambda^{\mathsf T}\eta\,\Lambda = \eta$ is a symmetric matrix equation, so it imposes exactly $10$ independent conditions on the $16$ entries of $\Lambda$ . $16 - 10 = 6$ : the Lorentz group is exactly six-dimensional — three boosts and three rotations, and nothing else.
Key idea: an invariant is a scalar with every index contracted. Components are opinions; contractions are facts. Compute an invariant in whichever frame is easiest — rest, co-locating, simultaneity — and the answer binds everywhere.

The core insight

There is a sharp test for whether a quantity is Lorentz-invariant, and it is purely syntactic: are all the indices contracted? $\Delta t$ has a hidden free index (it is the $0$ -component of a four-vector) and is not invariant. $\Delta s^2 = \eta_{\mu\nu}\Delta x^\mu\Delta x^\nu$ has none and is. $U\cdot U$ , $p\cdot p$ , $k\cdot x$ — all fully contracted, all invariant. This is why relativists write in index notation: the notation makes invariance visible, and makes it nearly impossible to write a frame-dependent statement by accident.

Where this lives today

Lorentz invariance is the most severely tested proposition in physics, and it is tested precisely because everything else is built on it. Rotating optical cavities compare the speed of light along perpendicular directions and bound any anisotropy at the $10^{-18}$ level; clock-comparison experiments bound preferred-frame effects in nuclear spin precession. The Standard-Model Extension is an entire framework built to parametrise the ways Lorentz invariance could fail, so that experiments can quote a limit on each one. So far every coefficient is consistent with zero. The invariance is also, quietly, a computational engine. Every collider result quoted as " $\sqrt{s} = 13\ \mathrm{TeV}$ " is a Lorentz invariant — the norm of the total four-momentum — chosen precisely because it does not depend on which frame you sit in. Computing in the convenient frame and trusting the answer everywhere is not a shortcut; it is the working content of the symmetry. And the symmetry does not stop at flat spacetime. In general relativity the metric $g_{\mu\nu}(x)$ varies from point to point, but at each event a freely falling observer sees $\eta_{\mu\nu}$ , and the local physics is the physics of this article. Lorentz invariance is demoted from a global symmetry to a local one — and in that demotion, the whole of gravitation is hiding.

The physical picture

In Euclidean geometry you already know how this works. Rotate your axes and the coordinates $(x,y)$ of a point both change; $x^2 + y^2$ does not. Nobody finds this mysterious, and nobody thinks the rotation "did something" to the point. It did something to the description. Minkowski spacetime works the same way, with one sign changed and one enormous consequence. Boosts mix time into space and space into time, so $\Delta t$ and $\Delta x$ both change. What survives is $c^2\Delta t^2 - \Delta x^2$ , the "radius squared" of a geometry whose circles are hyperbolae. A boost slides an event along an invariant hyperbola exactly as a rotation slides a point along a circle, and for exactly the same reason: it is the isometry group of the metric. The definition is worth stating in the right order, because the logic is often taught backwards. We do not observe that boosts preserve the interval and then congratulate ourselves. We define a Lorentz transformation to be a linear map $\Lambda$ with $\Lambda^{\mathsf T}\eta\,\Lambda = \eta$ , and the invariance of every $\eta$ -contraction follows in one line. The physical content sits entirely in the claim that the maps relating inertial frames are of this kind — which is what the light postulate buys you.

Figure

PUNCHLINE — the invariant ledger. Two event pairs, one boost of β = 0.6 (γ = 1.25). TIMELIKE (cΔt, Δx) = (5, 3) is co-located at (4, 0): s² = +16 both times, and the 4 is the proper time. SPACELIKE (3, 5) is made simultaneous by the SAME boost at (0, 4): s² = −16 both times, and the 4 is the proper length. Every coordinate moved. The invariant did not

Read the ledger. The timelike pair, boosted, becomes co-located, so the elapsed time in that frame is the proper time, $\sqrt{+16} = 4$ . The spacelike pair, boosted by the same $0.6c$ , becomes simultaneous, so the separation in that frame is the proper length, $\sqrt{-(-16)} = 4$ . The two cases are mirror images, which is why one number, $\beta = 3/5$ , does both jobs. And notice what the invariant is for. It is not a curiosity; it is a licence to be lazy. Compute $s^2$ in whichever frame makes the arithmetic trivial, and the answer binds every other frame. That habit — pick the easy frame, contract, stop worrying — is most of practical relativity.

The core equations

The whole subject rests on one definition, from which everything else is a corollary.

\Lambda^{\mathsf T}\eta\,\Lambda = \eta \qquad\Longleftrightarrow\qquad \eta_{\mu\nu}\,\Lambda^{\mu}{}_{\alpha}\,\Lambda^{\nu}{}_{\beta} = \eta_{\alpha\beta}, \qquad \eta = \mathrm{diag}(+1,-1,-1,-1) \qquad (1)

where: $\Lambda$ — a linear map on spacetime, $x'^\mu = \Lambda^\mu{}_\nu x^\nu$ . Physical reading: this is the definition of a Lorentz transformation. The physics is the claim that changes of inertial frame are maps of this kind; the mathematics is everything that follows. Note $\eta^2 = \mathbb{1}$ , so the condition rearranges to give the inverse in closed form: $\Lambda^{-1} = \eta\,\Lambda^{\mathsf T}\eta$ . The interval is the norm this metric induces on displacements:

\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 (2)

where: $\Delta x^\mu = (c\Delta t, \Delta \vec{x}\,)$ . Physical reading: timelike $\Delta s^2 > 0$ , null $\Delta s^2 = 0$ , spacelike $\Delta s^2 < 0$ , and because $\Delta s^2$ is invariant, that classification is absolute — no observer can boost a timelike separation into a spacelike one. (In $(-,+,+,+)$ all three signs flip; the base article for this slug is written in that convention, and its classification card is, for us, exactly backwards.) Invariance is then one line of matrix algebra, and it applies to any pair of four-vectors:

A'\cdot B' \;=\; \eta_{\mu\nu}\Lambda^{\mu}{}_{\alpha}\Lambda^{\nu}{}_{\beta}\,A^{\alpha}B^{\beta} \;=\; \eta_{\alpha\beta}\,A^{\alpha}B^{\beta} \;=\; A\cdot B \qquad (3)

where: $A\cdot B = A^0B^0 - \vec{A}\cdot\vec{B}$ . Physical reading: substitute (1) into the middle and the two $\Lambda$ 's vanish. That is the entire proof, and it is why every Lorentz invariant in physics is an instance of the same identity. The in-band catalogue is short and complete:

\Delta s^2, \qquad c^2\tau^2 = \Delta s^2\big|_{\text{timelike}}, \qquad U\cdot U = +c^2, \qquad \varphi = k\cdot x = \omega t - \vec{k}\cdot\vec{x} \qquad (4)

where: $\tau$ — proper time; $U^\mu$ — four-velocity; $k^\mu = (\omega/c, \vec{k}\,)$ — the wave four-vector. Physical reading: each is a full contraction, hence a number every inertial observer computes identically. $U\cdot U = +c^2$ in this signature, not $-c^2$ ; and the phase is $+\omega t - \vec{k}\cdot\vec{x}$ , not $-\omega t + \vec{k}\cdot\vec{x}$ — the base prints both with the other convention's signs. The phase is the most physical of the four: it counts wave crests, and crests cannot depend on who is counting.

Where the invariants come from

Derivation AThe Lorentz transformations are a group, and the geometry is its invariant theory

Before counting invariants it is worth establishing what kind of object we are dealing with, because the answer sets the shape of the whole theory. The set of maps preserving $\eta$ is not merely a collection: it is a group, closed under composition and inversion, with an inverse available in closed form. That fact licenses Klein's programme — the idea, stated in 1872, decades ahead of any physics that needed it, that a geometry simply is the invariant theory of a transformation group. Minkowski geometry is the case $G = O(1,3)$ .

Step 1. Identity. $\mathbb{1}^{\mathsf T}\eta\,\mathbb{1} = \eta$ , trivially. Why: the group must contain "do nothing" — the statement that a frame is related to itself by no boost at all.

Step 2. Closure. If $\Lambda_1$ and $\Lambda_2$ both satisfy (1), then $(\Lambda_1\Lambda_2)^{\mathsf T}\eta\,(\Lambda_1\Lambda_2) = \Lambda_2^{\mathsf T}\big(\Lambda_1^{\mathsf T}\eta\,\Lambda_1\big)\Lambda_2 = \Lambda_2^{\mathsf T}\eta\,\Lambda_2 = \eta$ . Why: the inner bracket collapses first, then the outer one. Composing two changes of frame is a change of frame — physically obvious, and here a two-line proof.

Step 3. Inverses, explicitly. Multiply (1) on the left by $\eta$ and use $\eta^2 = \mathbb{1}$ : $\;\eta\,\Lambda^{\mathsf T}\eta\,\Lambda = \mathbb{1}$ , so $\boxed{\Lambda^{-1} = \eta\,\Lambda^{\mathsf T}\eta}$ . For the standard boost this returns the boost with $\beta \to -\beta$ , which is reciprocity. Why: no matrix inversion is ever needed in relativity — transpose, sandwich with $\eta$ , done. It is the Minkowski analogue of $R^{-1} = R^{\mathsf T}$ for rotations, with the metric doing the sign bookkeeping.

Step 4. Conclude the Erlangen statement. Because $G$ is a group and $\eta$ is the object it preserves, the quantities all observers agree on are exactly the ones built from $\eta$ and four-vectors with every index contracted. "Minkowski geometry" means the invariant theory of $G$ . Why: this is not a metaphor but Klein's definition, applied to a group he lived just long enough to recognise as a geometry's group — and it is why relativists speak of "the geometry of spacetime" rather than "the effects of motion on rulers and clocks". *Verified symbolically in verify_aq-sr-lorentz-invariance.py: the boost satisfies $\Lambda^{\mathsf T}\eta\Lambda = \eta$ ; closure holds for a product of two boosts of arbitrary $\beta_1, \beta_2$ ; $\eta\Lambda^{\mathsf T}\eta$ inverts $\Lambda$ and equals the boost with $\beta \to -\beta$ ; and $A'\cdot B' = A\cdot B$ for symbolic $A$ , $B$ .*

Go deeper · Derivation B · 2 minTen equations, sixteen unknowns, and a six-dimensional group

Derivation BTen equations, sixteen unknowns, and a six-dimensional group

Here is the article's owned structural result. A general linear map on four-dimensional spacetime has $16$ free entries. The condition $\Lambda^{\mathsf T}\eta\,\Lambda = \eta$ looks like $16$ equations — it is an equality of $4\times4$ matrices — and if it were, the group would be zero-dimensional and there would be no boosts at all. It is not $16$ . It is exactly $10$ , for a reason that takes one line to see and that most textbooks state without proving. The surviving $6$ parameters are the three boost velocities and the three rotation angles, and that is why the Lorentz group has the size it has.

Step 1. Notice that the left-hand side is automatically symmetric. For any matrix $\Lambda$ , $\big(\Lambda^{\mathsf T}\eta\,\Lambda\big)^{\mathsf T} = \Lambda^{\mathsf T}\eta^{\mathsf T}\Lambda = \Lambda^{\mathsf T}\eta\,\Lambda$ , since $\eta$ is symmetric. Why: this is the crux, and it is a property of the expression, not of $\Lambda$ . The equation cannot impose $16$ independent conditions, because $6$ of its $16$ component-equations are copies of others.

Step 2. Count. A symmetric $4\times4$ matrix equation has one condition per entry of the upper triangle: $4 + 3 + 2 + 1 = 10$ independent scalar equations. Why: the entry $(0,1)$ and the entry $(1,0)$ say the same thing.

Step 3. Subtract. Sixteen unknowns minus ten independent constraints leaves $\boxed{6}$ free parameters, so the Lorentz group is a six-dimensional continuous group — and this is a count, not an observation. Why: the six was not put in by hand. It fell out of the fact that $\eta$ is a symmetric $4\times4$ matrix — that is, out of the dimension of spacetime.

Step 4. Verify the count by linearising. Write $\Lambda = \mathbb{1} + X$ for infinitesimal $X$ and keep first order: the condition becomes $X^{\mathsf T}\eta + \eta X = 0$ , again symmetric, again $10$ conditions on $16$ entries. Solving the resulting $10\times16$ system gives a solution space of dimension exactly $6$ — the rank is $10$ , so the ten equations really are independent. Why: the linearised problem is the Lie algebra $\mathfrak{so}(1,3)$ , whose dimension is the group's. This turns a counting argument into a machine-checkable nullspace computation.

Step 5. Exhibit the six. The condition $X^{\mathsf T}\eta + \eta X = 0$ says precisely that $\eta X$ is antisymmetric, and antisymmetric $4\times4$ matrices form a $6$ -dimensional space. The six generators split into three symmetric boost generators $K_i$ (mixing time with $x$ , $y$ , $z$ ) and three antisymmetric rotation generators $J_i$ (mixing the spatial axes). Exponentiating $\varphi K_1$ gives the standard boost with $\gamma = \cosh\varphi$ , $\gamma\beta = \sinh\varphi$ ; exponentiating $\theta J_3$ gives an ordinary rotation. Why: $6 = 3 + 3$ is the whole content. Three ways to change your velocity, three ways to turn your axes, and — because the count is exact — <strong>nothing else</strong>. There is no seventh continuous symmetry of Minkowski spacetime waiting to be found. *Verified symbolically in verify_aq-sr-lorentz-invariance.py: $\Lambda^{\mathsf T}\eta\Lambda$ is symmetric for a fully general $16$ -symbol $\Lambda$ ; the linearised system $X^{\mathsf T}\eta + \eta X = 0$ has rank $10$ and nullity exactly $6$ ; the six generators $K_1, K_2, K_3, J_1, J_2, J_3$ each satisfy the linearised condition and are linearly independent (rank $6$ ); and $\exp(\varphi K_1)$ , $\exp(\theta J_3)$ each satisfy $\Lambda^{\mathsf T}\eta\Lambda = \eta$ .*

Go deeper · Derivation C · 2 minCounting wave crests proves the wave four-vector is a four-vector

Derivation CCounting wave crests proves the wave four-vector is a four-vector

The last derivation looks like sleight of hand the first time you meet it and is in fact one of the most powerful tools in the subject: the quotient theorem. It lets you conclude that an object is a four-vector without ever computing how it transforms — you simply observe that its contraction with a known four-vector is a frame-independent number, and the transformation law is forced. The example is the phase of a plane wave, and the physical anchor is as concrete as physics gets: crests are things you can count on your fingers.

Step 1. Establish the physical invariant. Consider a plane wave and two events on its path. Between them some definite number of crests passes — a count, obtained by watching and tallying, and it cannot depend on who does the tallying. Hence the phase $\varphi$ , which is $2\pi$ times the crest count, is a Lorentz scalar. Why: this is the only physics in the argument. Everything after it is algebra.

Step 2. Write the phase as a contraction. For a plane wave $\varphi = \omega t - \vec{k}\cdot\vec{x}$ . Assemble $k^\mu \equiv (\omega/c, \vec{k}\,)$ and note $\eta_{\mu\nu}k^\mu x^\nu = (\omega/c)(ct) - \vec{k}\cdot\vec{x} = \varphi$ . Why: the metric's minus signs are exactly what is needed to turn the familiar phase into a contraction. In $(-,+,+,+)$ we would have got $-\omega t + \vec{k}\cdot\vec{x}$ — which is what the base for this slug prints: the same physics with the opposite overall sign.

Step 3. Apply the quotient theorem. $x^\mu$ is a four-vector (it is the archetype), and we have just argued that $k_\mu x^\mu$ is a scalar — for every choice of the two events, i.e. for every $x^\mu$ . If $k^\mu$ transformed by anything other than $\Lambda$ , some $x^\mu$ would exist for which the contraction changed. There is none. Therefore $\boxed{k^\mu \text{ is a four-vector}}$ . Why: this is a real theorem, not a hand-wave. "Contracts with an arbitrary four-vector to give a scalar" is logically equivalent to "is a four-vector", and it is usually far easier to check.

Step 4. Collect the corollaries. Since $k^\mu$ is a four-vector, $k\cdot k$ is an invariant; for light $\omega = c|\vec{k}|$ , so $k\cdot k = (\omega/c)^2 - |\vec{k}|^2 = 0$ . The wave four-vector of light is null, in every frame — the constancy of $c$ , restated once more. Boosting $k^\mu$ shifts $\omega$ : that shift is the relativistic Doppler effect, treated in its own article and not derived here. Why: relativistic optics is downstream of one observation about counting crests. *Verified symbolically in verify_aq-sr-lorentz-invariance.py: $k\cdot x = \omega t - kx$ in our signature; $k\cdot k = 0$ when $\omega = ck$ ; and numerically, a $600\ \mathrm{nm}$ wave between $(0,0)$ and $(2.0\ \mathrm{ns},\ 0.300\ \mathrm{m})$ gives exactly $500\,000$ crests, and $500\,000$ again after a $\beta = 0.6$ boost, although $t$ , $x$ , $\omega$ and $k$ all change.*

Figure

Why the group is exactly six-dimensional. Λ has 16 entries, but ΛᵀηΛ = η equates matrices that are ALREADY symmetric — so its 6 lower-triangle equations are duplicates and impose nothing new. 16 − 10 = 6 free parameters: three boost generators K₁, K₂, K₃ (symmetric) and three rotation generators J₁, J₂, J₃ (antisymmetric). Nothing else

Limiting cases

Case	Result	Meaning
Boost velocity $\to 0$	$\Lambda \to \mathrm{diag}(1, R)$	the rotation subgroup $SO(3)$ , preserving $x^2+y^2+z^2$
$v/c \to 0$	$\Delta s^2 \approx c^2\Delta t^2$	$\Delta t$ becomes effectively invariant — Newton's absolute time
Null separation	$\Delta s^2 = 0$ in every frame	the constancy of $c$ , restated as an invariance
$A\cdot A > 0$ / $=0$ / $<0$	timelike / null / spacelike	an absolute classification: no boost can change it
Free index left over	not invariant	$\Delta t$ , $\Delta x$ , $E$ , $\vec{p}$ all transform; only full contractions do not
Curved spacetime	$g_{\mu\nu}(x) \to \eta_{\mu\nu}$ locally	Lorentz invariance survives as a local statement — the equivalence principle

Common trapThe signature flips every sign in this article — read the metric first

This article uses $(+,-,-,-)$ : $\Delta s^2 = c^2\Delta t^2 - \Delta \vec{x}^2$ , timelike $> 0$ , spacelike $< 0$ , $U\cdot U = +c^2$ , $p\cdot p = +m^2c^2$ , phase $= \omega t - \vec{k}\cdot\vec{x}$ . The AllQFT base is written entirely in $(-,+,+,+)$ and prints $\eta = \mathrm{diag}(-1,1,1,1)$ , $u\cdot u = -c^2$ , $p\cdot p = -m^2c^2$ , a phase of $-\omega t + \vec{k}\cdot\vec{x}$ , and a card reading "timelike $\Delta s^2 < 0$ , spacelike $> 0$ ". Every one of those is correct there and wrong here. No physics depends on the choice; every intermediate sign does. The expensive failure mode is copying a formula from one convention into a calculation done in the other.

Worked examples

1 · The invariant ledger, with numbers

Two events in the lab frame $S$ are separated by $\Delta t = 5\ \mathrm{s}$ and $\Delta x = 3$ light-seconds. A second pair is separated by $\Delta t = 3\ \mathrm{s}$ and $\Delta x = 5$ light-seconds. Boost both by $\beta = 0.6$ and audit what survives.

Step 1 — identify. Work in units where $c = 1$ : times in seconds, distances in light-seconds. The invariant is $s^2 = (c\Delta t)^2 - \Delta x^2$ , and the boost is $c\Delta t' = \gamma(c\Delta t - \beta\,\Delta x)$ , $\Delta x' = \gamma(\Delta x - \beta\,c\Delta t)$ , with $\gamma = 1/\sqrt{1 - 0.36} = 1.25$ .

Step 2 — the timelike pair. $s^2 = 5^2 - 3^2 = +16 > 0$ : timelike, so the events are causally connectable. Boosting: $c\Delta t' = 1.25(5 - 1.8) = \mathbf{4}$ and $\Delta x' = 1.25(3 - 3) = \mathbf{0}$ . In $S'$ the two events happen at the same place. Check: $s'^2 = 4^2 - 0^2 = +16$ . Unchanged.

Step 3 — read the boundary number. $\Delta x' = 0$ means $S'$ is the frame in which one clock is present at both events, so the $4$ seconds it records is the proper time: $c\tau = \sqrt{+s^2} = 4$ . The invariant did not merely survive the boost — it told us in advance what the co-locating observer would read. And the required speed, $\beta = \Delta x/(c\Delta t) = 3/5$ , is below $1$ precisely because $s^2 > 0$ .

Step 4 — the spacelike pair. $s^2 = 3^2 - 5^2 = -16 < 0$ : spacelike, so no signal connects them. Boost by the same $\beta = 0.6$ : $c\Delta t' = 1.25(3 - 3) = \mathbf{0}$ and $\Delta x' = 1.25(5 - 1.8) = \mathbf{4}$ . In $S'$ the two events are simultaneous. Check: $s'^2 = 0 - 16 = -16$ . Unchanged again.

Step 5 — the boundary number, mirrored. $\Delta t' = 0$ means $S'$ is the frame in which one ruler spans both events, so the $4$ light-seconds it measures is the proper length: $L_0 = \sqrt{-s^2} = 4$ . Swap $\Delta t \leftrightarrow \Delta x$ and you swap co-location for simultaneity, proper time for proper length — which is why one and the same $\beta = 3/5$ does both jobs.

Step 6 — audit. Four of the coordinates changed under the boost ( $5 \to 4$ , $3 \to 0$ , $3 \to 0$ , $5 \to 4$ ); the two invariants did not ( $+16$ , $-16$ ). Every quantity with a free index moved; every fully contracted quantity stayed put. That is the whole rule, in arithmetic.

2 · Half a million crests, counted twice

A $600\ \mathrm{nm}$ light wave travels along $+x$ . Between the event $(t,x) = (0,0)$ and the event $(2.0\ \mathrm{ns},\ 0.300\ \mathrm{m})$ , how many crests pass? Now recount in a frame moving at $\beta = 0.6$ .

Step 1 — identify. The crest count is $\varphi/2\pi$ with $\varphi = \omega t - kx = k\cdot x$ , a full contraction of two four-vectors, hence invariant. So we already know the answer will not change. The point is to watch it not change while everything else does.

Step 2 — set up the wave. With $c = 3.00\times10^8\ \mathrm{m/s}$ and $\lambda = 600\ \mathrm{nm}$ : $k = 2\pi/\lambda = 1.047\times10^{7}\ \mathrm{rad/m}$ and $\omega = 2\pi c/\lambda = 3.1416\times10^{15}\ \mathrm{rad/s}$ . Note $\omega = ck$ , so $k\cdot k = (\omega/c)^2 - k^2 = 0$ : the wave four-vector is null, as light's must be.

Step 3 — count in the lab. $\varphi = (3.1416\times10^{15})(2.0\times10^{-9}) - (1.047\times10^{7})(0.300) = 6.2832\times10^{6} - 3.1416\times10^{6} = 3.1416\times10^{6}\ \mathrm{rad}$ , so $\varphi/2\pi$ is exactly $500\,000$ crests. (Sanity check: the wave oscillates $\omega t/2\pi = 10^6$ times in $2\ \mathrm{ns}$ , and the observation point sits $x/\lambda = 5\times10^5$ wavelengths downstream, so $10^6 - 5\times10^5$ crests pass.)

Step 4 — boost everything. With $\beta = 0.6$ , $\gamma = 1.25$ : the events move to $t' = \gamma(t - \beta x/c) = 1.25(2.0 - 0.6)\ \mathrm{ns} = 1.75\ \mathrm{ns}$ and $x' = \gamma(x - \beta ct) = 1.25(0.300 - 0.360)\ \mathrm{m} = -0.075\ \mathrm{m}$ — the second event is now behind the first. The wave moves too: $\omega' = \gamma(\omega - \beta ck) = 0.5\,\omega$ and $k' = 0.5\,k$ . (That $0.5$ is the longitudinal Doppler factor; here it is only an intermediate, and it is the subject of The Relativistic Doppler Effect.)

Step 5 — recount. $\varphi' = (1.5708\times10^{15})(1.75\times10^{-9}) - (5.236\times10^{6})(-0.075) = 2.7489\times10^{6} + 3.927\times10^{5} = 3.1416\times10^{6}\ \mathrm{rad}$ : exactly $500\,000$ crests again.

Step 6 — the boundary number. Four independent quantities changed — $t$ , $x$ , $\omega$ , $k$ — and one of them changed sign. The count did not move by a single crest, because a crest is a thing, and things do not care about coordinates. That is what a Lorentz invariant is, stated without a single index.

Physics of the portrait

Relativity's deepest structural idea was not invented by a physicist and did not arrive with relativity. It arrived in 1872, in a mathematician's inaugural programme at Erlangen, answering a question nobody had yet asked about spacetime: what is a geometry?

Origins · 2 minFelix Klein and how this was discovered

Felix Klein

1849–1925 · Erlangen · Munich · Leipzig · Göttingen

Christian Felix Klein was twenty-three when, on taking up a professorship at Erlangen, he circulated the pamphlet "Vergleichende Betrachtungen über neuere geometrische Forschungen" — the Erlangen Programme of 1872. Its thesis was audacious in its simplicity: a geometry is not a list of axioms about points and lines but a group of transformations together with the quantities that group leaves invariant. Euclidean geometry is the invariant theory of the rigid motions; affine, projective and the non-Euclidean geometries are the invariant theories of larger or smaller groups. The programme reorganised nineteenth-century mathematics and made "which group?" the first question to ask of any geometric structure. Klein spent most of his career at Göttingen, where he built the world's leading mathematical institute and — critically for this story — recruited Hilbert and Minkowski. When Minkowski recast relativity as spacetime geometry in 1908, the Erlangen framing fitted as if made to measure, and Klein said so: lecturing to the Göttingen Mathematical Society on 10 May 1910, in work published as "Über die geometrischen Grundlagen der Lorentzgruppe" (Jahresbericht der DMV 19, 281–300; reprinted in Physikalische Zeitschrift 12, 1911), he read special relativity as precisely the invariant theory of the Lorentz group. Ill health from 1883 had turned him from front-line research towards teaching, editing and organising; but the idea he had at twenty-three is why a modern physicist, handed a new theory, asks first what its symmetry group is.

History

1872

— Felix Klein circulates the Erlangen Programme: a geometry is the invariant theory of a group.

1904

— Lorentz publishes the transformation equations in the form bearing his name.

1905

— Poincaré, in a June note to the Comptes Rendus, names the Lorentz transformation and recognises that the transformations form a group.

1906

— Poincaré's Palermo memoir (received July 1905) identifies the invariant quadratic form and reads the transformations as rotations of a 4D space.

1908

— Minkowski's Cologne lecture "Raum und Zeit" recasts relativity as geometry with an indefinite metric.

1910

— Klein, lecturing to the Göttingen Mathematical Society on 10 May, reads relativity as the invariant theory of the Lorentz group (Jahresber. DMV 19, 281).

1910

— Sommerfeld's four-dimensional vector algebra makes invariant contractions routine.

1939

— Wigner classifies the representations of the Poincaré group: mass and spin are the only two invariant labels a particle can carry.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §§2, 5 · Landau & Lifshitz, The Classical Theory of Fields, §§1–6 · Weinberg, Gravitation and Cosmology, ch. 2 · Klein, "Über die geometrischen Grundlagen der Lorentzgruppe", Jahresbericht der DMV 19, 281–300 (1910) · Klein, Vergleichende Betrachtungen über neuere geometrische Forschungen (Erlangen, 1872).

Provenance. Signature $(+,-,-,-)$ ; the AllQFT base for this slug is written entirely in $(-,+,+,+)$ and every sign has been re-derived here. All results are recomputed in verify_aq-sr-lorentz-invariance.py: group closure and $\Lambda^{-1} = \eta\Lambda^{\mathsf T}\eta$ ; the symmetry of $\Lambda^{\mathsf T}\eta\Lambda$ for a fully general $16$ -symbol matrix; rank $10$ and nullity $6$ of the linearised system; six explicit, independent generators; the ledger at $\beta = 0.6$ ; and the $600\ \mathrm{nm}$ crest count of exactly $500\,000$ in both frames.

Scope. The base also carries the pair-production threshold, the Mandelstam invariants, and the field invariants $E^2 - c^2B^2$ and $\vec{E}\cdot\vec{B}$ . All are genuine Lorentz invariants, but all need relativistic dynamics or the field tensor, which arrive in later bands. They are named here as forward pointers and are not derived: see Threshold Energy and Particle Production, Mandelstam Variables, Invariants of the Electromagnetic Field.

Builds on. The Minkowski Metric · The Spacetime Interval · Four-Vectors. Leads to. Rapidity · Hyperbolic Geometry of Boosts · Causal Structure of Spacetime.

Is the interval invariant because we chose a clever definition, or is it a fact about nature?

Both, in the right order. Given the definition $\Lambda^{\mathsf T}\eta\,\Lambda = \eta$ , invariance is a one-line theorem with no physical content. The physics is the empirical claim that the maps relating real inertial frames are of this kind — and that is what the light postulate, and behind it a century of null results, are evidence for. If a preferred frame were found tomorrow, $\eta$ would not stop being preserved by $O(1,3)$ ; it would simply stop being the metric of the world we live in.

Why does the count come out ten and not sixteen? I still find this suspicious.

Write out the matrix equation and look at row $0$ , column $1$ ; then at row $1$ , column $0$ . Because $\Lambda^{\mathsf T}\eta\Lambda$ is symmetric — for any $\Lambda$ , since $\eta$ is — those two entries are the same expression, so setting both to $\eta_{01} = \eta_{10} = 0$ is one condition, not two. There are six such duplicated off-diagonal pairs, and $16 - 6 = 10$ . The suspicion is healthy, which is why the verification script also solves the linearised system and confirms rank $10$ , nullity $6$ .

If energy is not invariant, how can conservation of energy be a relativistic law?

Because conservation is a statement made within a frame, and it holds in every frame separately. What relativity adds is that $(E/c, \vec{p}\,)$ are conserved together, as a four-vector — so if total four-momentum is conserved in one frame it is conserved in all, since both sides of $\sum p^\mu_{\text{in}} = \sum p^\mu_{\text{out}}$ carry the same $\Lambda$ . Energy alone is conserved but frame-dependent; the norm of the total four-momentum is frame-independent but not a conservation law. Collision kinematics is the art of using both facts at once.

Are there Lorentz invariants that are not built from four-vectors?

Yes — from tensors of higher rank, contracted the same way. The electromagnetic field tensor $F^{\mu\nu}$ yields $F_{\mu\nu}F^{\mu\nu}$ and $\epsilon^{\mu\nu\alpha\beta}F_{\mu\nu}F_{\alpha\beta}$ , which reduce up to constants to $E^2 - c^2B^2$ and $\vec{E}\cdot\vec{B}$ . These lie outside this article's band and are treated where the field tensor is introduced. The rule, however, is identical: contract every index with the metric, and what is left is a scalar.

Does the six-dimensional count include translations?

No, and the distinction matters. The Lorentz group proper — boosts and rotations — is six-dimensional and consists of linear maps fixing the origin. Adding the four spacetime translations gives the ten-dimensional Poincaré group, the full symmetry group of Minkowski spacetime; the extra four parameters say that no event is privileged. Wigner's 1939 classification of its representations is what tells us a relativistic particle carries exactly two invariant labels — mass and spin — and that is about as deep as "which quantities are invariant?" gets.

The base article says timelike separations have Δs² < 0. Which of you is right?

Both, because we are answering different questions. In $(-,+,+,+)$ — Misner–Thorne–Wheeler's convention, and the one the base is written in — timelike separations do have $\Delta s^2 < 0$ . In $(+,-,-,-)$ — Rindler's, particle physics's, and this corpus's — they have $\Delta s^2 > 0$ . The two differ by an overall sign and nothing else, and no measurable quantity distinguishes them. What is not harmless is mixing them: substitute a formula from one into a calculation done in the other and you get an imaginary proper time, a negative rest mass, or a causality violation — and the algebra will not warn you. Read the metric before you read the equation.