Velocity Composition and the Speed-of-Light Invariant

Run after a light pulse at $0.5c$ and it still recedes from you at $c$ . Run at $0.9c$ — still $c$ . At $0.999999c$ — still $c$ , to every digit you can name. Nothing you do to your own motion changes the number, and this is the fact that broke classical physics. The usual telling makes it an axiom: Einstein postulated the invariance of $c$ , and everything follows. But once you have the composition law

w = u \oplus v \equiv \frac{u+v}{1 + uv/c^2},

the invariance is no longer an axiom. It is a theorem about the arithmetic — and a rather startling one. Substitute $u = c$ and the whole expression collapses: $(c+v)/(1 + v/c) = c$ , identically, for every $v$ . The speed of light is a fixed point of the composition operation. That raises the question this article is named for, and which the arithmetic answers by itself: is it the only one? Could some other speed — perhaps far above $c$ , perhaps some strange number below it — also refuse to change when you compose it with anything? The answer is a two-line proof, and it says: no. Exactly $+c$ and $-c$ , and nothing else. The invariant speed is a structural constant of the composition law, forced by the demand that boosts form a group at all. And the group is the point — because closure, associativity and reciprocity, with no mention of light anywhere, already prove that some invariant speed must exist. Only experiment tells you it is the one light travels at.

Key takeaways

The collinear composition of velocities is written $\oplus$ : $u \oplus v = \dfrac{u+v}{1 + uv/c^2}$ . Its derivation from the Lorentz transformation belongs to Relativistic Velocity Addition; this article asks what structure it has. Signature $(+,-,-,-)$ ; $S$ is the lab, $S'$ moves at $+v\hat{x}$ .
$\oplus$ makes the open interval $(-c, c)$ an abelian group. Closure (nothing escapes), associativity, identity $0$ , inverse $-v$ , commutativity — all five, all provable, none assumed.
And it is a group you have already met. $\varphi = \operatorname{artanh}(v/c)$ is a group isomorphism $\big((-c,c),\ \oplus\big) \;\xrightarrow{\ \cong\ }\; (\mathbb{R},\ +)$ : it turns $\oplus$ into ordinary addition, exactly. The hyperbolic geometry behind it is owned by Rapidity and Hyperbolic Boosts.
Closure is not an inequality — it is a factorisation. $1 - \dfrac{(u \oplus v)^2}{c^2} = \dfrac{\left(1 - u^2/c^2\right)\left(1 - v^2/c^2\right)}{\left(1 + uv/c^2\right)^2}$ . Two positive factors upstairs, a square downstairs: subluminality is algebra, not luck.
Key idea — the uniqueness theorem, which is the article. Suppose some speed $w$ is unchanged by composing with EVERY subluminal $v$ : $w \oplus v = w$ for all $|v| ＜ c$ . Then $v\left(1 - w^2/c^2\right) = 0$ for all such $v$ , so $w^2 = c^2$ and $w = \pm c$ — no other value survives. The two fixed points are the two endpoints, they lie OUTSIDE the group, and they are the same two for every observer. That is the invariance of the speed of light, proved rather than postulated.

The core insight

A group law with a fixed point has a preferred speed built into it, whether or not anyone mentions light. Boosts must compose (do one, then another, and you have done a boost); composition must be associative and reversible. Those demands alone — plus isotropy — force the composition law into the $\oplus$ shape and hand you an invariant speed for free. Experiment then walks in and says: the invariant speed is $299\,792\,458\ \mathrm{m\,s^{-1}}$ , and light happens to travel at it.

Where this lives today

The group structure proved on this page is why particle physicists never compose velocities. They compose rapidities — the isomorphism turns a nonlinear composition into an addition, and an addition is something a computer, a detector and a human can all handle. Every longitudinal boost between the lab and the centre-of-momentum frame at a hadron collider is applied as a shift in rapidity, and the reason that works is Step 5 above. The uniqueness theorem is the sharpest available statement of what is being tested when experiments test Lorentz invariance. Modern searches do not ask "is light's speed constant?" — they ask "is there a single invariant speed, and do all massless species share it?" Gamma rays and gravitational waves from the neutron-star merger GW170817 arrived within $1.7$ seconds of each other after $130$ million light years, bounding the fractional difference between the speed of gravity and the speed of light to a few parts in $10^{15}$ . If two massless species travelled at different invariant speeds, the composition law would have two different fixed points — and, by the theorem above, two different Lorentz groups. The universe would not have one relativity; it would have two. That is why the bound is stated on the invariant speed, not on the speed of any particular thing.

The physical picture

Think of $\oplus$ as a warped ruler. On an ordinary ruler, adding is sliding: put $u$ and $v$ end to end, read off $u+v$ , and nothing prevents the total from running off the end. The relativistic ruler is made by taking the infinite real line — the rapidity line, on which sliding really is just addition — and squashing all of it into the finite open interval $(-c, c)$ via $v = c\tanh\varphi$ . Every property of $\oplus$ is a property of that squashing. Nothing escapes, because $\tanh$ never returns $\pm 1$ : the image of the whole real line is the open interval. Composition is addition in disguise, because the squashing map is a group isomorphism. And the ends are fixed, because they are the images of $\pm\infty$ — and adding a finite rapidity to infinity leaves infinity. That last fact is the invariance of $c$ , and in this picture it is almost too easy. The work of this article is to prove it without borrowing the rapidity picture — straight from the algebra of $\oplus$ — so that the conclusion does not depend on the very change of variables that makes it obvious.

Figure

PUNCHLINE — the composition law as a warped number line. TOP: rapidity φ, an ordinary infinite ruler on which boosts add. BOTTOM: velocity β, the SAME boosts pushed through β = tanh φ, crowding into the open interval (−1, 1). The map is a group isomorphism, so ⊕ is just + wearing a disguise. The two endpoints β = ±1 are the images of φ = ±∞: they are the FIXED POINTS of ⊕, and the uniqueness theorem says there are no others. Marked: 0.5 ⊕ 0.5 = 0.8, 0.9 ⊕ 0.9 = 0.994475, 0.99 ⊕ 0.99 = 0.999949 — and c ⊕ v = c for every v, always

The core equations

The law itself, and where it comes from. This article does not derive it — the derivation from the boost belongs to Relativistic Velocity Addition, which owns it. We take it as given and ask what it is.

w \;=\; u \oplus v \;=\; \frac{u+v}{1 + \dfrac{uv}{c^2}} \qquad (1)

where: $u$ = the object's velocity in $S'$ ; $v$ = the velocity of $S'$ relative to $S$ ; $w$ = the object's velocity in $S$ ; all collinear. Physical reading: this is the INVERSE direction ( $S' \to S$ ), written with $+$ signs, and it is the direction in which $\oplus$ reads naturally as a composition of two boosts. The chapter's FORWARD rule, $u' = (u-v)/(1-uv/c^2)$ , is the same operation with $v \to -v$ : $u' = u \oplus (-v)$ . Reciprocity is a group inverse in disguise — a statement we are about to make precise. The single identity that does all the structural work:

\boxed{\;1 - \frac{(u \oplus v)^2}{c^2} \;=\; \frac{\left(1 - \dfrac{u^2}{c^2}\right)\left(1 - \dfrac{v^2}{c^2}\right)}{\left(1 + \dfrac{uv}{c^2}\right)^2}\;} \qquad (2)

where: each bracket is $1/\gamma^2$ for its own velocity. Physical reading: read this as $\gamma$ -deficits multiplying. It is not an inequality dressed as an identity — it is an exact factorisation, and every structural fact below is squeezed out of it. If $|u|, |v| ＜ c$ both numerator factors are strictly positive and the denominator is a square, so the left side is strictly positive and $|u \oplus v| ＜ c$ : closure. If either input equals $c$ , its factor is exactly zero, so the left side is zero and $|u \oplus v| = c$ exactly: the fixed point. One identity, both theorems. And the change of variable that reveals what the group is:

\varphi \;=\; \operatorname{artanh}\!\left(\frac{v}{c}\right) = \tfrac12 \ln\frac{c+v}{c-v}, \qquad v = c\tanh\varphi, \qquad \varphi_{u \oplus v} = \varphi_u + \varphi_v \qquad (3)

where: $\varphi$ is the rapidity. Physical reading: the additivity in (3) is not this article's to prove — Rapidity and Hyperbolic Boosts and Rapidity own it, and they get it from the $\tanh$ addition formula and from the boost matrices respectively. What is this article's is the observation that (3) is exactly the statement that $\operatorname{artanh}(\cdot/c)$ is a group isomorphism, and that this fact classifies the group completely: it is $(\mathbb{R},+)$ , and there is nothing else to know about it.

The owned result: ⊕ is a group, and it has exactly two fixed points

The base article for this page asserted, in its summary, that $c$ is "the fixed point" of the composition law and that $\oplus$ is a group — and then proved neither, spending all three of its derivations on results that belong to its neighbours. Both claims are true. Both have short, complete proofs. Here they are.

Derivation A⊕ makes the open interval a group — and artanh is an isomorphism onto the real line

A binary operation is not a group law just because it looks like one. Five things must be checked: closure (the operation never leaves the set), associativity, an identity, inverses, and — for an abelian group — commutativity. Every one of the five is something a physicist already believes about boosts on physical grounds; the point of proving them is that they then stop being beliefs. Set $c = 1$ throughout (restore it by $v \to v/c$ ); the set is the open interval $I = (-1, 1)$ .

Step 1 — closure. Read identity (2). For $u, v \in I$ the numerator brackets $(1-u^2)$ and $(1-v^2)$ are strictly positive, and $(1+uv)^2$ is the square of a nonzero number ( $|uv| ＜ 1$ forces $1+uv ＞ 0$ ). Hence $1 - (u \oplus v)^2 ＞ 0$ , i.e. $u \oplus v \in I$ . ∎ Why: "you can't reach $c$ by adding speeds" is not a coincidence and not a limit argument. It is a factorisation: a product of two positive numbers is positive. Note also that the denominator can never vanish on $I \times I$ , so $\oplus$ is defined everywhere with no exceptions to police.

Step 2 — commutativity, identity, inverses. $u \oplus v = v \oplus u$ by inspection (numerator and denominator of (1) are symmetric). $u \oplus 0 = u$ , so $0$ is a two-sided identity. And $u \oplus (-u) = (u-u)/(1-u^2) = 0$ with $-u \in I$ , so every element has an inverse — its negative. Why: the inverse law is reciprocity — if $S'$ moves at $+v$ relative to $S$ , then $S$ moves at $-v$ relative to $S'$ , and doing both gets you home. That is the entire content of Inverse Lorentz Transformation and Reciprocity, restated as a group axiom. Note also the caveat hiding in commutativity: it holds for collinear boosts. Non-collinear boosts do not commute, and the failure is a rotation.

Step 4 — associativity, by brute force. Expand both bracketings over a common denominator:

(u \oplus v) \oplus w = \frac{\dfrac{u+v}{1+uv} + w}{1 + \dfrac{(u+v)w}{1+uv}} = \frac{u+v+w+uvw}{1 + uv + uw + vw},

and the same computation on $u \oplus (v \oplus w)$ gives $\dfrac{u+v+w+uvw}{1+uv+uw+vw}$ — the identical expression, and a manifestly symmetric one. ∎ Why: what came out is symmetric in all three arguments, which is stronger than associativity — the composition of any number of collinear boosts depends only on the multiset of velocities, not on the order or the bracketing. That is the physical fact that a chain of boosts has a well-defined total.

Step 5 — name the group. Define $\varphi = \operatorname{artanh}(u)$ , a bijection $I \to \mathbb{R}$ (strictly increasing, with $\operatorname{artanh}(\pm 1^{\mp}) = \pm\infty$ ). The $\tanh$ addition formula gives $\tanh(\varphi_1 + \varphi_2) = \dfrac{\tanh\varphi_1 + \tanh\varphi_2}{1 + \tanh\varphi_1\tanh\varphi_2} = u_1 \oplus u_2$ , so

\operatorname{artanh}(u_1 \oplus u_2) = \operatorname{artanh}(u_1) + \operatorname{artanh}(u_2).

A bijection that carries $\oplus$ to $+$ is precisely a group isomorphism, so $\big(I, \oplus\big) \cong (\mathbb{R}, +)$ . ∎ Why: and now the group is completely classified — it is the additive reals, the simplest continuous group there is, and every "strange" feature of relativistic velocity addition is the strangeness of the coordinate $\tanh$ , not of the physics. Associativity, which cost a paragraph of algebra in Step 4, is now free: addition is associative. This is what an isomorphism is for. *Verified in verify_aq-sr-velocity-composition-light-invariant.py: identity (2) is proved symbolically; closure is checked on $10^5$ random pairs and on the boundary; associativity is proved symbolically by simplifying $(u \oplus v) \oplus w - u \oplus (v \oplus w)$ to zero and independently verified numerically; the identity and inverse laws are checked symbolically; and $\operatorname{artanh}(u \oplus v) - \operatorname{artanh} u - \operatorname{artanh} v$ simplifies to zero, establishing the isomorphism.*

Go deeper · Derivation B · 3 minTHE UNIQUENESS THEOREM — c and −c are the ONLY fixed points of the composition law

Derivation BTHE UNIQUENESS THEOREM — c and −c are the ONLY fixed points of the composition law

Here is the result the article is named for, and which its source article asserted without ever proving. That $c$ is a fixed point is a one-line substitution and no surprise once seen. The real question is whether it is the fixed point — whether the composition law singles out one speed or merely leaves one alone among many. The answer is that the fixed set is exactly $\{+c, -c\}$ , and the proof takes two lines and uses no analysis, no limits and no physics. It converts Einstein's second postulate from an assumption into a corollary, which is worth stating carefully.

Step 1 — what "fixed point" must mean. A speed $w$ is invariant if every observer measures it to be $w$ . Observers differ by boosts, and a boost by $v$ acts on a velocity by $w \mapsto w \oplus v$ . So the precise statement is:

w \ \text{is invariant} \qquad\Longleftrightarrow\qquad w \oplus v = w \quad \text{for every } |v| ＜ c.

Why: note the quantifier — for every $v$ , not for some. A speed that happened to be unchanged by one particular boost would be a coincidence; a speed unchanged by all of them is a structural feature of the law. And note that $w$ is not required to lie in $(-c,c)$ : we are asking the question of every real number, precisely so that the answer is not smuggled in.

Step 2 — write it out and clear the denominator. The condition $w \oplus v = w$ reads

\frac{w+v}{1 + wv/c^2} = w \qquad\Longrightarrow\qquad w + v = w + \frac{w^2 v}{c^2} \qquad\Longrightarrow\qquad v\left(1 - \frac{w^2}{c^2}\right) = 0.

Why: the $w$ 's cancel — that is the whole trick, and it is why the theorem is two lines. What survives is a product of $v$ with a factor that does not depend on $v$ at all.

Step 3 — apply the quantifier. The equation must hold for every $|v| ＜ c$ , in particular for some $v \neq 0$ . A product of two numbers vanishes only if one of them does, and $v \neq 0$ , so

1 - \frac{w^2}{c^2} = 0 \qquad\Longrightarrow\qquad w^2 = c^2 \qquad\Longrightarrow\qquad \boxed{\,w = +c \quad\text{or}\quad w = -c\,}

and conversely both do satisfy $w \oplus v = w$ (substitute and cancel $(c+v)$ ). Hence the fixed set is exactly $\{+c, -c\}$ . ∎ Why: there is the second postulate. It has not been assumed anywhere on this page. It is a property of the arithmetic — and the arithmetic came from the Lorentz transformation, which came from the group structure of boosts. Two speeds, one in each direction, and no others of any magnitude: not $2c$ , not $10^{-3}c$ , not anything.

Step 4 — where the fixed points live. $\pm c$ are not elements of the group. The group is the open interval $(-c, c)$ ; the fixed points are its two boundary points, approached and never reached. In rapidity language they sit at $\varphi = \pm\infty$ , and " $c \oplus v = c$ " becomes " $\infty + \varphi_v = \infty$ ". Why: this is the sharpest kinematic statement of what light is. A light pulse is not a very fast particle; it is a boundary point of the velocity group — a different kind of object. It is also why no group element is fixed: for $|w| ＜ c$ , choosing $v = -w \neq 0$ gives $w \oplus v = 0 \neq w$ . Every subluminal speed is moved by some boost; exactly two speeds are moved by none, and no massive body can have them.

Step 5 — the converse, and what it costs. Any law of the form $u \oplus_K v = (u+v)/(1 + uv/K^2)$ is an abelian group on $(-K, K)$ with fixed points $\pm K$ — the proofs above never used the numerical value of $c$ . So the structure predicts an invariant speed, but not which. Letting $K \to \infty$ recovers the Galilean law $u \oplus v = u+v$ , a perfectly good group $(\mathbb{R},+)$ with no finite fixed point at all. Why: this is the honest boundary of what group theory buys, and it is what Ignatowski found in 1910 — see the portrait. Symmetry narrows the possibilities to a one-parameter family; a measurement picks the member. The parameter is $K = c = 299\,792\,458\ \mathrm{m\,s^{-1}}$ exactly. *Verified in verify_aq-sr-velocity-composition-light-invariant.py: sympy solves $w \oplus v = w$ for symbolic $w$ under the constraint that it hold for two distinct nonzero $v$ , returning exactly $\{+c, -c\}$ ; $c \oplus v - c$ simplifies to zero for symbolic $v$ ; a scan of $10^4$ candidate $w$ from $-3c$ to $+3c$ finds no other value fixed by both $v = 0.3c$ and $v = 0.7c$ ; and $w \oplus (-w) = 0$ confirms no interior fixed point exists.*

Limiting cases

Case	Result	Meaning
$\|u\|, \|v\| \ll c$	$u \oplus v \approx u + v$	Galileo. The correction is $-uv(u+v)/c^2$ , second order in the small quantities
$c \to \infty$	$u \oplus v = u + v$ exactly	the Galilean group $(\mathbb{R},+)$ ; no finite fixed point — the invariant speed has run off to infinity
$u = c$	$u \oplus v = c$	the fixed point, for every $v$ ; the second postulate
$u = v = 0.5c$	$0.8c$	not $c$ . The most quoted number in the subject
$u \to c^-$ , $v \to c^-$	$u \oplus v \to c^-$	the supremum is approached, never attained: $c$ is not in the group
$v = -u$	$u \oplus v = 0$	the group inverse; no interior element is a fixed point

Common trap⊕ is a group only for COLLINEAR boosts — and the failure is not small

Everything on this page is one-dimensional. The moment the two velocities are not parallel, three of the five properties above fail at once, and the wreckage is a real physical effect.

Commutativity fails. Boost along $x$ , then along $y$ , and you do not land where you would have the other way round.
The composed transformation is not a pure boost. It is a boost times a spatial rotation — the Thomas–Wigner rotation, owned by Thomas–Wigner Rotation. So the set of pure boosts is not closed, and is not a subgroup of the Lorentz group at all.
"Adding rapidity vectors" is wrong. Rapidity is additive along a line; it is not a vector obeying the parallelogram rule. Velocity space is hyperbolic, not Euclidean.

None of this touches the uniqueness theorem: $\pm c$ remain invariant in every direction, because a light ray boosted any way you like is still a light ray (Transverse Velocity Transformation). What breaks is the group structure of velocities, not the invariance of $c$ .

Common trapClosing speed is not relative speed, and it is allowed to exceed c

At the LHC two proton bunches approach each other at $0.999999991c$ each in the lab frame, so in the lab the gap between them shrinks at $2 \times 0.999999991c = 1.999999982c$ . There is nothing wrong with that. A closing speed is the rate of change of a coordinate difference measured by one observer with one clock; it is not any object's velocity, and no signal travels at it. It may legitimately be anything up to $2c$ . The relative speed — one proton's speed in the rest frame of the other — is a different quantity, and it is what $\oplus$ computes: $\beta \oplus \beta = 2\beta/(1+\beta^2)$ , strictly less than $c$ always, and for these protons $1 - 4.0 \times 10^{-17}$ in units of $c$ . The perennial "but they're closing at $2c$ !" objection is answered by: yes, and that is a statement about a coordinate difference, not a velocity. Only velocities obey $\oplus$ ; only velocities are bounded by $c$ .

Worked examples

1 · Two 7 TeV protons meet head-on

At the LHC each beam carries protons of energy $E = 7\ \mathrm{TeV}$ . In the lab they close on each other at nearly $2c$ . What speed does one proton measure for the other? And what fixed-target beam energy would be needed to reproduce the same collision?

Step 1 — identify. $\gamma = E/(m_pc^2) = 7000/0.938272 = \mathbf{7460.52}$ , so

\beta = \sqrt{1 - \gamma^{-2}} = \mathbf{0.999\,999\,991}.

Each proton has velocity $+\beta c$ and $-\beta c$ in the lab. To get one proton's view of the other we must compose, not subtract: $\beta_{\mathrm{rel}} = \beta \oplus \beta$ .

Step 2 — apply. The cleanest route is the $\gamma$ of the composed velocity, which the factorisation identity (2) hands us directly. Rearranged, (2) says $\gamma_{u \oplus v} = \gamma_u \gamma_v (1 + uv/c^2)$ , so with $u = v = \beta c$ :

\gamma_{\mathrm{rel}} = \gamma^2\left(1 + \beta^2\right) = \gamma^2\left(1 + 1 - \gamma^{-2}\right) = \boxed{\,2\gamma^2 - 1\,} = 2(7460.52)^2 - 1 = \mathbf{1.1132 \times 10^{8}}.

Step 3 — the boundary number. Convert back:

\beta_{\mathrm{rel}} = \sqrt{1 - \gamma_{\mathrm{rel}}^{-2}} \qquad\Longrightarrow\qquad 1 - \beta_{\mathrm{rel}} \approx \frac{1}{2\gamma_{\mathrm{rel}}^2} = \mathbf{4.0 \times 10^{-17}}.

So one proton sees the other approaching at $0.999\,999\,999\,999\,999\,96\,c$ — strictly below $c$ , by four parts in $10^{17}$ , even though the lab says they are closing at $1.999\,999\,982\,c$ . Closure held. It always does.

Step 4 — cash it out in energy. A proton with $\gamma_{\mathrm{rel}} = 1.1132 \times 10^8$ has energy $\gamma_{\mathrm{rel}} m_pc^2 = 1.0445 \times 10^{8}\ \mathrm{GeV} = \mathbf{104\ \mathrm{PeV}}$ — what you would have to build to do LHC physics with a fixed target. Cross-check by the invariant-mass route: $s = 2m_pc^2E_{\mathrm{lab}} + 2m_p^2c^4$ , and demanding $\sqrt{s} = 14\ \mathrm{TeV}$ gives $E_{\mathrm{lab}} = (s - 2m_p^2c^4)/(2m_pc^2) = 1.0445\times10^{8}\ \mathrm{GeV}$ ✓ — same number, entirely different road. The factor of $\sim 1.5\times10^4$ between $7\ \mathrm{TeV}$ and $104\ \mathrm{PeV}$ is why colliders exist.

2 · Chasing a photon, and then hunting for a second fixed point

A light pulse moves at $c$ . You chase it at $\beta = 0.99$ . What speed do you measure for it? Then: is there any other speed that behaves this way?

Step 1 — compose. Use the FORWARD rule ( $S \to S'$ , your frame moving at $+0.99c$ ), which is $u \oplus (-v)$ :

u' = \frac{c - 0.99c}{1 - (0.99c)(c)/c^2} = \frac{0.01c}{0.01} = \mathbf{c}.

Step 2 — push it. Try $\beta = 0.999999$ : $u' = (1-0.999999)c/(1-0.999999) = c$ . Numerator and denominator are the same number and cancel before you have a chance to take a limit. This is not "approximately $c$ ", and it is not a cancellation of small quantities — it is the exact identity $(c-v)/(1 - v/c) = c$ , valid for every $v \neq c$ .

Step 3 — the boundary number: look for a rival. Ask whether some other speed $w$ has the same property. Impose $w \oplus v = w$ for both $v = 0.3c$ and $v = 0.7c$ :

v = 0.3c: \quad w(1 + 0.3w/c) = w + 0.3c \;\Rightarrow\; 0.3\,w^2/c = 0.3c \;\Rightarrow\; w^2 = c^2.

v = 0.7c: \quad 0.7\,w^2/c = 0.7c \;\Rightarrow\; w^2 = c^2.

Both boosts give the same equation, and its only roots are $w = \pm c$ . A single boost is already enough to pin the fixed points, and every other boost agrees with it. There is no rival, at any magnitude — not above $c$ , not below it.

Step 4 — check the negative cases. $w = 0.8c$ : $0.8c \oplus 0.3c = 1.1c/1.24 = 0.887c \neq 0.8c$ . Moved. $w = 0$ : $0 \oplus 0.3c = 0.3c$ . Moved. $w = 2c$ (a hypothetical tachyon): $2c \oplus 0.3c = 2.3c/1.6 = 1.4375c \neq 2c$ . Moved — superluminal speeds are not invariant either, which is not often said. Exactly two speeds in the universe are unmoved by a boost, and light travels at one of them in each direction.

Figure

The uniqueness theorem, drawn. Plot the boost map w ↦ w ⊕ v against the diagonal w = w, for v = 0.5c (blue) and v = −0.8c (orange). A FIXED POINT is a crossing of the curve with the diagonal. Both curves cross the diagonal at exactly two places — w = +c and w = −c, the two corners of the box — and nowhere else, for any v. Subluminal speeds (green dots) are all MOVED: 0.8c goes to 0.928571c under v = 0.5c, and 0 goes to 0.5c. Superluminal w = 2c is moved too, to 1.4375c. Exactly two speeds are fixed, they are the endpoints, and they lie outside the group

Physics of the portrait

Origins · 2 minGroup theory and how this was discovered

Group theory

1910–1911 · Ignatowski's theorem · Frank and Rothe's proof · the invariant speed that symmetry gives you, and the one thing it does not

This card has no physicist on it, because the discovery belongs to a piece of mathematics. The question is: how much of relativity survives if you delete the light postulate? The answer, found by Wladimir Sergejewitsch von Ignatowski in 1910 (Physikalische Zeitschrift 11, 972), is: almost all of it. Assume only that the transformations between inertial frames form a group; that space is homogeneous (so the transformation is linear) and isotropic; and that the relation between two frames is reciprocal (if you see me at $+v$ , I see you at $-v$ ). No light, no electrodynamics, no second postulate. From these alone the transformation is forced into the Lorentz form — with one undetermined constant $K$ , of the dimensions of a speed. Philipp Frank and Hermann Rothe gave an improved derivation in 1911 (Annalen der Physik 34, 825), assuming the transformations form a homogeneous linear group with reciprocity. Here is the honest boundary, which matters as much as the theorem. The group structure proves that an invariant speed $K$ exists. It does not tell you its value, and it does not tell you that light travels at it. $K = \infty$ is perfectly consistent — that is Galilean relativity, whose group is $(\mathbb{R},+)$ with no finite fixed point. Ignatowski could not identify $K$ with $c$ ; the criticism is Pauli's, and it stands. Symmetry gives you a one-parameter family of possible kinematics; experiment gives you the parameter, and says $K$ is finite and equal to $299\,792\,458\ \mathrm{m\,s^{-1}}$ . So the modern reading of the second postulate is not "light is special" but "the invariant speed is finite, and light — being massless — travels at it." The postulate is a measurement dressed as an axiom; the group theory is what makes the measurement decisive.

History

1905

— Poincaré, in a note to the Comptes Rendus (140, 1504) read on 5 June, names the Lorentz transformation and is the first to prove that the transformations form a GROUP — the structural fact this article turns into a theorem.

1905

— Einstein derives the composition law from his two postulates and observes that composing two subluminal velocities never yields c or more.

1910

— Ignatowski shows that the relativity principle plus group structure, homogeneity, isotropy and reciprocity already force a Lorentz-type transformation with an undetermined invariant speed K — no light postulate required.

1911

— Frank and Rothe give an improved derivation, assuming the transformations form a homogeneous linear group; they too obtain K as a free constant.

1911

— Varicak recognises that relativistic velocity space is hyperbolic (Lobachevskian), the geometry behind the isomorphism artanh.

1921

— Pauli, in Relativitätstheorie, notes the limitation: the group-theoretic route cannot identify K with the speed of light. Experiment must.

1983

— The metre is redefined so that c = 299 792 458 m/s exactly; the invariant speed of the group law becomes a defined constant rather than a measured one.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §§13–14 · Taylor & Wheeler, Spacetime Physics, 2nd ed., ch. 3 · Griffiths, Introduction to Electrodynamics, 4th ed., §12.2. On the group-theoretic route: Ignatowski, Phys. Z. 11, 972 (1910); Frank & Rothe, Ann. Phys. 34, 825 (1911); a modern account is Lévy-Leblond, "One more derivation of the Lorentz transformation", Am. J. Phys. 44, 271 (1976).

Provenance. Signature $(+,-,-,-)$ ; boost convention as in Transverse Velocity Transformation ( $\oplus$ is written in the composition direction, which is the inverse rule). Authored from scratch: the base was a displaced payload. Its three derivation cards proved $c \oplus v = c$ , the subluminality identity, and the rapidity ceiling — all of which belong to Relativistic Velocity Addition and Rapidity and the Additivity of Velocities. Its own claims — that $\oplus$ is a group and that $c$ is the fixed point — were asserted in its summary and derived nowhere. Both are proved here. Every number recomputed in verify_aq-sr-velocity-composition-light-invariant.py: $0.5 \oplus 0.5 = 0.8$ , $0.8 \oplus 0.5 = 0.928571$ , $0.9 \oplus 0.9 = 0.994475$ , $0.99 \oplus 0.99 = 0.999949$ ; $\gamma = 7460.52$ , $\beta = 0.999999991$ , $\gamma_{\mathrm{rel}} = 2\gamma^2-1 = 1.1132\times10^8$ , $1-\beta_{\mathrm{rel}} = 4.0\times10^{-17}$ , $E_{\mathrm{lab}} = 104\ \mathrm{PeV}$ cross-checked against $\sqrt{s} = 14\ \mathrm{TeV}$ ; $2c \oplus 0.3c = 1.4375c$ ; and the group axioms, the isomorphism and the uniqueness theorem symbolically.

Builds on. Relativistic Velocity Addition · Lorentz Transformation Equations · Rapidity and Hyperbolic Boosts. Leads to. Rapidity and the Additivity of Velocities · Thomas–Wigner Rotation · Relativistic Aberration of Light.

If the composition law already proves c is invariant, why did Einstein need to postulate it?

Because the composition law is derived from the Lorentz transformation, which Einstein derived from the postulate. The logic is a circle unless you break in somewhere, and Einstein broke in at the postulate. This article shows the circle can be entered at a different door: assume only that boosts form a group, with homogeneity, isotropy and reciprocity, and you get the Lorentz transformation with an undetermined invariant speed $K$ — Ignatowski's theorem — then $\oplus$ , then the uniqueness theorem, then $\pm K$ as the only fixed points. The second postulate is what tells you $K$ is finite and equal to the speed of light, and that piece cannot come from symmetry. The postulate is not redundant; it is smaller than it looks. It fixes one number, not a whole kinematics.

Is c a fixed point, or a limit that is approached but never reached? These sound like different things.

They are different things, and both are true of different objects. For massive particles, $c$ is a supremum never attained: no element of the group $(-c,c)$ equals $c$ , and no finite chain of boosts gets there, because $\operatorname{artanh}$ of anything finite is finite. For light, $c$ is a genuine fixed point: $c \oplus v = c$ exactly, not in a limit. The two statements coexist because $\pm c$ are boundary points of the velocity group, not elements of it. A massive particle lives inside the interval and can be moved anywhere inside it by a suitable boost. A photon lives on the boundary and cannot be moved at all. That is the kinematic content of "massless".

Does the uniqueness theorem rule out tachyons?

Not directly, but it says something sharper than most people expect. The theorem says that if $w \oplus v = w$ for all $|v| ＜ c$ then $w = \pm c$ — and the derivation never assumed $|w| ＜ c$ , so a hypothetical superluminal $w$ was in the running and lost. Concretely, $2c \oplus 0.3c = 1.4375c \neq 2c$ : a tachyon's speed would be frame-dependent, and by choosing a boost you could make it anything you liked (including, for a suitable $v$ , infinite, and then negative — which is the tachyonic-antitelephone problem). So the theorem does not forbid superluminal speeds; it forbids invariant superluminal speeds, and thereby removes the only thing that could have made them respectable. There is exactly one invariant speed, and it is $c$ .

Why is the factorisation identity so much better than just saying "the denominator is bigger than 1"?

Because the denominator is not always bigger than $1$ — if $u$ and $v$ have opposite signs, $1 + uv/c^2 ＜ 1$ , and the hand-waving argument collapses precisely where you most want it (composing a forward and a backward boost). The identity $1 - (u\oplus v)^2/c^2 = (1-u^2/c^2)(1-v^2/c^2)/(1+uv/c^2)^2$ has no such gap: the numerator is a product of two strictly positive numbers whenever both inputs are subluminal, the denominator is a square, and the conclusion follows for every sign combination with no case analysis. It also does double duty: set either input to $c$ and the left side is exactly zero, which is the fixed point. One identity, closure and invariance both.

The group is isomorphic to (ℝ, +). Doesn't that mean relativity is secretly Galilean?

It means relativity is secretly linear — in the right variable. That is deep, but it is not Galilean. The Galilean group is also $(\mathbb{R},+)$ , but with the velocity itself as the group parameter, so velocities are unbounded and there is no finite invariant speed. In relativity the parameter is the rapidity, and velocity is $\tanh$ of it — a function whose range is the open interval. Same abstract group, different realisation on physical velocities, and the difference is exactly the two fixed points. Abstract isomorphism is not physical identity: what distinguishes the theories is not the group but its action on velocity.

What happens to the group when the velocities are not collinear?

It stops being a group, and the failure is the subject of Thomas–Wigner Rotation. Non-parallel boosts do not commute, and their composition is not a pure boost at all — it is a boost times a spatial rotation, so the set of pure boosts is not closed. Velocities in three dimensions therefore do not form a group under relativistic addition; they form a gyrogroup, in which associativity and commutativity hold only up to a rotation. The full Lorentz group (boosts and rotations) is perfectly closed — but it is not a group of velocities. The one-dimensional case is special precisely because there is one direction, and hence no rotation to generate.