Relativistic Velocity Addition Formula

Fire a probe forward at $0.6c$ from a ship already doing $0.6c$ . Galileo says $1.2c$ . The universe says $\mathbf{0.8824\,c}$ — and it is not being coy about a speed limit, it is doing arithmetic. The rule is u' = \frac{u - v}{1 - uv/c^2}, and the interesting character is the denominator. It looks like a correction bolted on to make the answer behave; every textbook that presents it that way leaves the reader with the impression that relativity has a fudge factor. It does not. The denominator is a $\gamma$ -factor in disguise — specifically, $1 - uv/c^2 = \gamma_{u'}/(\gamma_u\gamma_v)$ — and once you see that, the whole thing collapses into one sentence: velocity is a ratio, and ratios do not add. Here is the sentence unpacked. The four-velocity $U^{\mu} = \gamma_u(c,\ \vec{u})$ transforms linearly: $U'^{\mu} = \Lambda^{\mu}{}_{\nu}U^{\nu}$ , exactly like every other four-vector, with no strangeness anywhere. But the ordinary velocity is not a four-vector — it is the quotient $u = c\,U^1/U^0$ . A quotient of two linear functions is a fractional-linear function, and that is the entire content of the Einstein rule. The nonlinearity was never in the physics. It was in the division. That reframing pays for itself immediately. Taking the time component of the linear map gives $\gamma_{u'} = \gamma_u\gamma_v(1 - uv/c^2)$ — a product of finite positive numbers — and therefore $\gamma_{u'}$ is finite, and therefore $|u'| < c$ . Subluminality stops being a numerical coincidence and becomes a factorisation. No composition of sub-luminal velocities can ever reach $c$ , for the same reason that a product of positive numbers cannot be zero.

Key takeaways

The collinear rule, in the chapter's locked boost convention $x' = \gamma(x - vt)$ : u'_{\parallel} = \frac{u - v}{1 - uv/c^2}, \qquad\text{inverse}\qquad u = \frac{u' + v}{1 + u'v/c^2} The inverse is the forward rule with $v \to -v$ and the primes swapped — that, and nothing more.
Galileo's $u' = u - v$ is the $uv/c^2 \to 0$ limit. The first correction is $u' \approx (u-v)(1 + uv/c^2)$ .
The owned result: velocity is a ratio, and that is the whole story. The four-velocity $U^{\mu} = \gamma_u(c, \vec{u})$ transforms linearly under $\Lambda$ ; ordinary velocity is the quotient $u = c\,U^1/U^0$ . A quotient of linear maps is fractional-linear — so the Einstein rule is not a new law, it is a division.
Reading off the time component of that linear map gives the identity the base never derives: \boxed{\gamma_{u'} = \gamma_u\,\gamma_v\left(1 - \frac{uv}{c^2}\right)} The denominator IS $\gamma_{u'}/(\gamma_u\gamma_v)$ . It is not a fudge factor.
Hence the $\gamma$ -deficit identity $1 - u'^2/c^2 = (1-u^2/c^2)(1-v^2/c^2)/(1-uv/c^2)^2$ , whose right-hand side is a product of positives over a positive square. Subluminality is a factorisation, not an accident.
Numbers: $0.6c \oplus 0.6c = \mathbf{0.8824c}$ (with $\gamma = 1.25 \times 1.25 \times 1.36 = \mathbf{2.125}$ , exactly); $0.9c \oplus 0.9c = \mathbf{0.9945c}$ ; $0.99999c \oplus 0.99999c = \mathbf{0.9999999999c}$ . The ceiling is approached and never touched.
Key idea: velocities do not add because velocity is a quotient of two components of a four-vector that does add. The nonlinearity is division, not physics — and the denominator is the boosted $\gamma$ .

The core insight

Nothing in relativity says velocities "obey a strange addition law". What it says is that four-velocities transform by a matrix, linearly and boringly, and that the thing we call velocity is the ratio of two of that four-vector's components. Divide a linear map by a linear map and you get a fractional-linear map — the Einstein rule, on the nose. The denominator $1 - uv/c^2$ is then not a mystery but a measurement: it is exactly $\gamma_{u'}/(\gamma_u\gamma_v)$ , and its strict positivity for $|u|, |v| < c$ is the reason $c$ can never be reached.

Where this lives today

Every relativistic collision calculation lives on this rule. When two LHC beams meet head-on at $\gamma \approx 7000$ each, the lab closing rate is very nearly $2c$ — a number that appears in no invariant and means nothing physical. What sets the available energy is the velocity of one beam in the rest frame of the other, which the composition rule returns as a speed just below $c$ and a $\gamma$ of roughly $10^8$ . Confusing the two is not a subtle error; it is the difference between the right centre-of-mass energy and a meaningless one. The same arithmetic decodes apparent superluminal motion in astrophysical jets, where blobs ejected from active galactic nuclei appear to cross the sky faster than light: the true velocity, recovered by composing the jet's motion with the light-travel-time geometry, is always sub-luminal, and the illusion is an aberration effect (Relativistic Aberration). And the rule's deepest structural descendant is the observation this article has made precise — that velocity is a projective coordinate on the four-velocity. That is why velocities compose fractionally, why rapidity (which is the logarithm of the projective coordinate's cross-ratio) composes additively, and why the whole subject has the structure of a one-parameter group rather than a vector space. The additivity itself is owned by Rapidity as an Additive Velocity Parameter; the group structure by Velocity Composition and the Invariance of Light.

[//]: # (q2)

The physical picture

The Galilean rule $u' = u - v$ is a statement about a sum of displacements over a common time. Its hidden premise is that the two frames share that time. Relativity removes the premise, and the removal is not gentle: the moving observer's clocks are dilated and desynchronised, and both effects enter the ratio $\Delta x'/\Delta t'$ . Watch where they go. Under the boost $x' = \gamma(x - vt)$ , $t' = \gamma(t - vx/c^2)$ :

u' = \frac{\Delta x'}{\Delta t'} = \frac{\gamma(\Delta x - v\,\Delta t)}{\gamma\!\left(\Delta t - \dfrac{v}{c^2}\Delta x\right)}.

The numerator is exactly the Galilean answer. The denominator is the correction — and it comes entirely from the $-vx/c^2$ term in the time transformation, i.e. from the relativity of simultaneity. The $\gamma$ cancels between numerator and denominator, which is why time dilation alone never produces the velocity rule and why the rule looks so unlike $\gamma$ -physics. It is a simultaneity effect wearing a division sign. That cancellation is also a clue, and it is the clue this article follows. The $\gamma$ did not vanish — it went somewhere. It went into the four-velocity's time component, and finding it there turns the whole rule inside out. Meanwhile the numbers do what numbers do:

$u$	$v$	Galileo	Einstein
$0.6c$	$-0.6c$	$1.2c$	$\mathbf{0.8824c}$
$0.9c$	$-0.9c$	$1.8c$	$\mathbf{0.9945c}$
$0.99999c$	$-0.99999c$	$\approx 2c$	$\mathbf{0.9999999999c}$
$c$	anything	$c - v$	$\mathbf{c}$

The last row is the point of the exercise. Light comes out at $c$ whatever you subtract, and it does so not because a rule was chosen to make it so, but because $c$ is what the denominator is built to protect.

Interactive Simulation

Relativistic Velocity Addition — add two velocities with the sliders; the result (u+v)/(1+uv/c²) never exceeds c, unlike the Galilean sum that overshoots.

Figure

Velocity composition against the Galilean straight line. HORIZONTAL: β₁ ⊕ β₂ for equal inputs, β from 0 to 1. Galileo's answer (dashed) is the straight line 2β, which crosses c at β = 0.5 and keeps going. Einstein's answer (solid) is 2β∕(1 + β²), which rises, bends, and asymptotes to c without ever reaching it. Marked: 0.6 ⊕ 0.6 = 0.8824 where Galileo says 1.2; 0.9 ⊕ 0.9 = 0.9945 where Galileo says 1.8; 0.99999 ⊕ 0.99999 = 0.9999999999 where Galileo says ≈ 2. The gap between the two curves is not a correction — it is the denominator 1 + β², and the denominator is γ-arithmetic

The core equations

\boxed{\;u'_{\parallel} = \frac{u - v}{1 - \dfrac{uv}{c^2}}\;} \qquad\qquad u = \frac{u' + v}{1 + \dfrac{u'v}{c^2}} \qquad (1)

where: $S$ is the lab; $S'$ moves at $+v\hat{x}$ relative to $S$ ; $u$ is the object's velocity in $S$ and $u'$ in $S'$ , both parallel to the boost. Physical reading: the left form is the chapter's forward rule, matching the locked boost $x' = \gamma(x - vt)$ (owned by The Lorentz Transformation Equations). The right form is its inverse — obtained by $v \to -v$ with the primes swapped, and by nothing else. Mixing the two is the commonest sign error in the subject; the inverse is what you want when you are adding velocities ("ship plus probe").

U^{\mu} = \gamma_u\,(c,\ \vec{u}\,), \qquad U\cdot U = \eta_{\mu\nu}U^{\mu}U^{\nu} = \gamma_u^2(c^2 - u^2) = +\,c^2 \qquad (2)

where: $\gamma_u = 1/\sqrt{1 - u^2/c^2}$ ; $\eta_{\mu\nu} = \mathrm{diag}(+1,-1,-1,-1)$ , the chapter's locked signature. Physical reading: the four-velocity is a genuine four-vector and transforms linearly, $U'^{\mu} = \Lambda^{\mu}{}_{\nu}U^{\nu}$ . Its norm is $+c^2$ — positive, in this timelike-positive signature; in $(-,+,+,+)$ it would read $-c^2$ , and every sign in this article would flip with it. Owned by Four-Vectors; used here as the engine.

\boxed{\;\gamma_{u'} = \gamma_u\,\gamma_v\left(1 - \frac{uv}{c^2}\right)\;} \qquad (3)

where: the composition law for the Lorentz factors themselves. Physical reading: this is the article's owned result and the identity that dissolves the mystery. Rearranged, it says $1 - uv/c^2 = \gamma_{u'}/(\gamma_u\gamma_v)$ — the denominator of eq (1) is the boosted $\gamma$ , divided by the two you started with. It is the time component of $U'^{\mu} = \Lambda^{\mu}{}_{\nu}U^{\nu}$ , and nothing else. The base states only its inverse form, and derives neither.

1 - \frac{u'^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 (4)

where: the $\gamma$ -deficit identity — eq (3) squared and inverted. Physical reading: the right-hand side is a product of two strictly positive numbers over a positive square, whenever $|u|, |v| < c$ . So the left-hand side is strictly positive, so $|u'| < c$ , always. Subluminality is an algebraic factorisation, not a numerical coincidence — and the fact that $c$ is the unique speed with this property is the theorem of Velocity Composition and the Invariance of Light, which owns it.

Deriving the structure

Derivation AThe collinear rule, from the Lorentz transformation and a division

The rule is one line of algebra once you accept that velocity is a ratio of coordinate differences and that both the numerator and the denominator transform. What makes it worth writing out slowly is watching where the $\gamma$ goes — because it cancels, and its cancellation is the thing that makes the rule look mysterious. The next derivation finds it again.

Step 1. Transform both differentials with the chapter's locked boost $x' = \gamma(x - vt)$ , $t' = \gamma(t - vx/c^2)$ : \mathrm{d}x' = \gamma\,(\mathrm{d}x - v\,\mathrm{d}t), \qquad \mathrm{d}t' = \gamma\left(\mathrm{d}t - \frac{v}{c^2}\,\mathrm{d}x\right). Why: the transformation is linear, so it applies to differences and differentials exactly as it does to coordinates. Note that the time equation carries a spatial term — that is the relativity of simultaneity, and it is the whole source of the effect.

Step 2. Divide, and watch $\gamma$ leave. u' = \frac{\mathrm{d}x'}{\mathrm{d}t'} = \frac{\gamma(\mathrm{d}x - v\,\mathrm{d}t)}{\gamma\left(\mathrm{d}t - \frac{v}{c^2}\mathrm{d}x\right)} = \frac{\dfrac{\mathrm{d}x}{\mathrm{d}t} - v}{1 - \dfrac{v}{c^2}\dfrac{\mathrm{d}x}{\mathrm{d}t}} = \boxed{\;\frac{u - v}{1 - uv/c^2}\;} Why: the common factor $\gamma$ divides out of a ratio, which is why the velocity rule contains no $\gamma$ at all — a fact that puzzles everyone who has just learned that $\gamma$ is what relativity is about. Dividing through by $\mathrm{d}t$ in the last step is what turns coordinate differentials into velocities.

Step 3. Read the inverse off, honestly. $S$ moves at $-v$ relative to $S'$ , so swapping the roles of the frames means $v \to -v$ with the primes exchanged: u = \frac{u' + v}{1 + u'v/c^2}. Why: this is reciprocity, not a new derivation, and it is the form you use to add velocities — ship at $v$ , probe at $u'$ relative to the ship, probe at $u$ in the lab. The base writes the forward form; several neighbouring treatments write the inverse. They are the same rule and must never be mixed. *Verified in verify_aq-sr-relativistic-velocity-addition.py: the rule is re-derived symbolically from the boost matrix; forward and inverse are checked to be exact inverses of one another for $\beta \in \{0.1, 0.6, 0.9, 0.99\}$ ; and $\gamma$ is confirmed absent from the final expression.*

The owned result: velocity is a ratio, and that is the whole story

Everything strange about the Einstein rule comes from one unremarkable fact: velocity is not a four-vector. The four-velocity is, and it transforms linearly, with no fractional-linear anything in sight. Ordinary velocity is the quotient of two of its components — and a quotient of linear maps is fractional-linear, necessarily and unavoidably. Chase that through and the denominator stops being a fudge factor and becomes a $\gamma$ .

Go deeper · Derivation B · 2 minThe four-velocity transforms linearly; the denominator is the boosted gamma

Derivation BThe four-velocity transforms linearly; the denominator is the boosted gamma

Take the four-velocity, boost it with the same $\Lambda$ used everywhere else in the chapter, and read the two components separately instead of dividing them straight away. The time component alone gives an identity the base only asserts, and it is the identity that makes subluminality obvious.

Step 1. Set up, in the locked signature $(+,-,-,-)$ . For an object moving at $u$ along $\hat{x}$ in $S$ : U^{\mu} = \gamma_u\,(c,\ u,\ 0,\ 0), \qquad U\cdot U = \gamma_u^2(c^2 - u^2) = +c^2. Why: the norm is $+c^2$ — positive, timelike-positive signature — and it is invariant, so whatever $\Lambda$ does to the components, $U'\cdot U' = +c^2$ too. That single fact will turn out to be the subluminality theorem in disguise.

Step 2. Boost it — linearly. With $\beta_v = v/c$ and the locked $\Lambda$ : U'^0 = \gamma_v\big(U^0 - \beta_v U^1\big), \qquad U'^1 = \gamma_v\big(U^1 - \beta_v U^0\big). Why: this is an ordinary matrix acting on an ordinary vector. Nothing fractional, nothing nonlinear, no special rule for velocities — the machinery is the same $\Lambda^{\mu}{}_{\nu}$ that acts on $x^{\mu}$ , owned by The Lorentz Transformation Equations.

Step 3. Substitute, and take the time component alone. U'^0 = \gamma_v\left(\gamma_u c - \frac{v}{c}\,\gamma_u u\right) = \gamma_u\gamma_v\,c\left(1 - \frac{uv}{c^2}\right). But by definition $U'^0 = \gamma_{u'}c$ , where $u'$ is the object's speed in $S'$ . Therefore \boxed{\;\gamma_{u'} = \gamma_u\,\gamma_v\left(1 - \frac{uv}{c^2}\right)\;} Why: there is the denominator. It was never a correction term — it is $\gamma_{u'}/(\gamma_u\gamma_v)$ , the amount by which the boosted Lorentz factor falls short of the naive product. The base prints this identity (in its inverse form) as an equation card and never derives it; here it drops out of the time component of a matrix multiplication.

Step 4. Now the space component, and the division. U'^1 = \gamma_v\big(\gamma_u u - \beta_v \gamma_u c\big) = \gamma_u\gamma_v\,(u - v), \qquad\text{and}\qquad U'^1 = \gamma_{u'}u'. Divide the space component by the time component — that is all the Einstein rule is: u' = c\,\frac{U'^1}{U'^0} = c\cdot\frac{\gamma_u\gamma_v(u-v)}{\gamma_u\gamma_v c\left(1 - uv/c^2\right)} = \frac{u-v}{1 - uv/c^2}. Why: the $\gamma_u\gamma_v$ that cancelled so mysteriously in the first derivation is now visible on both sides of the fraction, and you can see exactly what it was. The rule is a quotient of two linear maps, and that is the only reason it is nonlinear. Velocity is a projective coordinate on the four-velocity, nothing more exotic than that.

Step 5. Harvest subluminality — as a factorisation. Take eq (3), invert, square, and use $1/\gamma^2 = 1 - u^2/c^2$ : 1 - \frac{u'^2}{c^2} = \frac{1}{\gamma_{u'}^2} = \frac{1}{\gamma_u^2\gamma_v^2\left(1 - uv/c^2\right)^2} = \frac{\left(1-\frac{u^2}{c^2}\right)\left(1-\frac{v^2}{c^2}\right)}{\left(1 - \frac{uv}{c^2}\right)^2}. For $|u|, |v| < c$ the two brackets are strictly positive and the denominator is a nonzero square (indeed $|uv/c^2| < 1$ , so it never vanishes). A product of positives over a positive square is positive — hence $1 - u'^2/c^2 > 0$ , hence $|u'| < c$ . Why: this is the payoff. No composition of sub-luminal velocities can reach $c$ , for the same reason that a product of positive numbers cannot be zero. It is not that the formula was cleverly designed to respect a speed limit; it is that $\gamma_{u'}$ is a product of finite positive factors and therefore finite. And it is the same statement as $U'\cdot U' = +c^2$ from Step 1: a timelike four-vector stays timelike, because $\Lambda$ preserves the metric. That $c$ is the unique speed with this fixed-point property is the theorem of Velocity Composition and the Invariance of Light — stated here, proved there. *Verified in verify_aq-sr-relativistic-velocity-addition.py: eq (3) is proved symbolically from the boost matrix acting on $U^{\mu}$ ; eq (4) is proved symbolically by direct algebra AND obtained independently as $1/\gamma_{u'}^2$ ; $U\cdot U = +c^2$ is asserted invariant under the boost; $1 - uv/c^2 > 0$ is proved for all $|u|,|v| < c$ ; and $|u'| < c$ is confirmed over $10^5$ random sub-luminal pairs, including $u = v = 0.999999999$ .*

Figure

PUNCHLINE — where the γ went. LEFT: the four-velocity Uᵘ = γᵤ(c, u) is an ordinary four-vector and the boost acts on it LINEARLY — an ordinary matrix on an ordinary vector, no strangeness anywhere. Its TIME component gives γ₍ᵤ′₎ = γᵤ γᵥ (1 − uv∕c²), and its SPACE component gives γ₍ᵤ′₎u′ = γᵤ γᵥ (u − v). RIGHT: ordinary velocity is the QUOTIENT u = c U¹∕U⁰ — divide the two and the γᵤγᵥ cancels, leaving the Einstein rule. The rule is fractional-linear ONLY because it is a division. And the denominator you were told was a correction factor is exactly γ₍ᵤ′₎ ∕ (γᵤ γᵥ) — which, being a ratio of finite positive numbers, can never let |u′| reach c

Limiting cases

Case	Result	Meaning
$u, v \ll c$	$u' \approx u - v$	Galileo, recovered. First correction: $u' \approx (u-v)(1 + uv/c^2)$
$u = c$	$u' = \dfrac{c-v}{1 - v/c} = c$	light comes out at $c$ for every $v$ — the denominator exists to make this exact
$u = v$	$u' = 0$	you are at rest in your own frame. A sanity check the formula must pass, and does
$u = v = c$	$u' = 0/0$	indeterminate, and rightly so — there is no rest frame of a photon to ask about
$u, v \to c$	$u \oplus v \to c^-$	$0.99999 \oplus 0.99999 = 0.9999999999$ . Approached, never reached
$\gamma$	absent from eq (1)	it cancelled in the ratio — and reappears, in eq (3), as the denominator itself

Common trapFour ways to get this wrong

1 · The forward rule is not the inverse rule. $u' = (u-v)/(1 - uv/c^2)$ gives the velocity in $S'$ of something moving at $u$ in $S$ . If you are adding — a ship at $v$ , a probe at $u'$ relative to the ship — you want the inverse, $u = (u' + v)/(1 + u'v/c^2)$ . The two differ by $v \to -v$ , and the chapter's neighbours are not all written in the same one. Read the frames before you read the formula.

2 · This rule is for the parallel component only. A velocity component perpendicular to the boost does not transform this way: it picks up an extra $1/\gamma$ from time dilation, giving $u'_{\perp} = u_{\perp}/[\gamma(1 - uv/c^2)]$ even though $y' = y$ . That rule, and the reason the $\gamma$ appears without any transverse contraction, belongs to The Transverse Velocity Transformation.

3 · Non-collinear boosts do not compose into a boost. Compose two boosts in different directions and the result is a boost times a rotation — the Thomas–Wigner rotation. The collinear rule is the special case in which that rotation is the identity, and it hides the fact completely. See The Thomas–Wigner Rotation.

4 · Closing speed is not relative velocity. Two bunches approaching each other at $0.99999c$ each have a lab closing rate of $1.99998c$ — and that is fine, because it is a coordinate rate, not the speed of anything. The velocity of one bunch in the rest frame of the other is $0.9999999999c$ . The first number is a bookkeeping artefact of one frame; the second is physics, and it is what sets the collision energy.

Worked examples

1 · The ship and the probe, and the γ that multiplies

A ship moves at $v = 0.6c$ . It fires a probe forward at $u' = 0.6c$ relative to itself. What does the launch frame measure — and what is the probe's $\gamma$ ?

Step 1 — use the inverse form, because we are adding: u = \frac{u' + v}{1 + u'v/c^2} = \frac{0.6c + 0.6c}{1 + 0.36} = \frac{1.2c}{1.36} = \mathbf{0.8824\,c}. Not $1.2c$ . The Galilean answer overshoots by $36\%$ , which is exactly the size of the denominator.

Step 2 — now the $\gamma$ , the easy way. Eq (3) in its inverse form reads $\gamma_u = \gamma_{u'}\gamma_v(1 + u'v/c^2)$ . With $\gamma_{0.6} = 1/\sqrt{1 - 0.36} = 1.25$ for both: \gamma_u = 1.25 \times 1.25 \times 1.36 = \mathbf{2.125}\quad\text{exactly.}

Step 3 — check it the hard way. Directly, $\gamma_u = 1/\sqrt{1 - 0.8824^2}$ . With the exact value $u/c = 1.2/1.36 = 15/17$ : $1 - (15/17)^2 = (289 - 225)/289 = 64/289$ , so $\gamma_u = 17/8 = \mathbf{2.125}$ ✓. The two routes agree to the last digit, which is what eq (3) promised.

Step 4 — the moral. The $\gamma$ 's multiply, up to the denominator. That is the whole content of the composition law, and it is why the answer can never run away: $2.125$ is finite because $1.25$ , $1.25$ and $1.36$ are finite.

2 · Fizeau's water tube — the drag coefficient falls out

Light travels at $c/n$ in still water ( $n = 1.333$ ). The water flows at $v = 7$ m/s. How fast does the light go in the lab?

Step 1 — compose, don't add. The lab speed is not $c/n + v$ but w = \frac{c/n + v}{1 + \dfrac{v}{nc}}.

Step 2 — expand to first order in $v/c$ . w \approx \left(\frac{c}{n} + v\right)\left(1 - \frac{v}{nc}\right) \approx \frac{c}{n} + v\left(1 - \frac{1}{n^2}\right). The water does not drag the light along at its own speed $v$ ; it drags it at $v$ times the Fresnel drag coefficient 1 - \frac{1}{n^2} = 1 - \frac{1}{1.777} = \mathbf{0.437}\quad (n = 1.333).

Step 3 — the number. The drag is $0.437 \times 7 = \mathbf{3.06\ m\,s^{-1}}$ , not $7\ \mathrm{m\,s^{-1}}$ and not zero. Evaluating the exact composition with $c = 299\,792\,458\ \mathrm{m\,s^{-1}}$ gives $w - c/n = \mathbf{3.06\ m\,s^{-1}}$ — the first-order formula is already exact to the digits quoted, because $v/c \sim 10^{-8}$ .

Step 4 — why this mattered. Fresnel invented that coefficient in 1818 to save the stationary ether, and it worked embarrassingly well: Fizeau measured it in 1851 and Michelson and Morley confirmed it in 1886. It was one of the ugliest facts in nineteenth-century physics — an ether that a moving medium drags partially, by a factor conjured to fit. In 1907, von Laue showed it is nothing but the velocity-addition rule, expanded to first order. A generation of ether mechanics evaporated into one line of algebra.

Physics of the portrait

Origins · 2 minRelativistic velocity addition and how this was discovered

Relativistic velocity addition

1818–1907 · Fresnel's drag coefficient · Fizeau's water tube · Einstein's §5 · Laue's one-line explanation

The rule has a strange history: its most famous experimental confirmation was performed fifty-four years before it was written down, to test a completely different theory. In 1818, in a letter to Arago, Augustin Fresnel faced an awkward problem. If light is a wave in a stationary ether, then a moving transparent body ought to drag the ether — but a full drag contradicts stellar aberration, and no drag contradicts other observations. Fresnel's escape was a compromise so arbitrary it embarrassed him: the medium drags the light partially, by the factor $1 - 1/n^2$ . For water this is $0.437$ — a number with no obvious right to exist. In 1851 Hippolyte Fizeau built the experiment that should have killed it (Comptes Rendus 33, 349): light split into two beams, sent up- and down-stream through flowing water, and recombined. The fringes shifted, and the shift matched Fresnel's coefficient. Michelson and Morley repeated it far more carefully in 1886 (Am. J. Sci. 31, 377) and confirmed it again; Zeeman, in 1914–1915, measured it precisely enough to see the dispersion correction. The coefficient was real. Nobody could say why. Then Einstein, in §5 of the 1905 paper (Annalen der Physik 17, 891), derived velocity composition from the kinematics — and in 1907 Max von Laue (Ann. Phys. 23, 989) pointed out the obvious: compose $c/n$ with $v$ relativistically, expand to first order, and Fresnel's coefficient falls out on its own. There is no drag, no ether, no partial anything. There is a denominator. Einstein himself said the Fizeau result was one of the experiments that mattered most to him — a strange, arbitrary number that his kinematics simply explained.

History

1818

— Fresnel, in a letter to Arago, introduces the partial-drag coefficient 1 − 1/n² to reconcile a stationary ether with stellar aberration. For water it is 0.437, and nobody can say why.

1851

— Fizeau measures the drag of flowing water on light (Comptes Rendus 33, 349) and finds agreement with Fresnel's coefficient.

1886

— Michelson and Morley repeat the water-tube experiment far more precisely (American Journal of Science 31, 377) and confirm the coefficient again.

1905

— Einstein derives the composition of velocities in §5 of Zur Elektrodynamik bewegter Körper (Annalen der Physik 17, 891), directly from the Lorentz transformation of coordinates.

1907

— Max von Laue shows that Fresnel's drag coefficient is simply the relativistic velocity-addition rule expanded to first order (Annalen der Physik 23, 989). The ether's most successful number becomes a corollary.

1915

— Zeeman's precision repetitions of the Fizeau experiment resolve the dispersion correction to the drag coefficient.

1926

— Thomas shows that non-collinear boosts do not compose into a boost: the missing piece is a rotation, which the collinear rule hides completely.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §13 · Griffiths, Introduction to Electrodynamics, 4th ed., §12.1.2 · Taylor & Wheeler, Spacetime Physics, 2nd ed., ch. 3. Primary: Einstein, Annalen der Physik 17, 891 (1905), §5; von Laue, Annalen der Physik 23, 989 (1907); Fizeau, Comptes Rendus 33, 349 (1851).

Provenance. The base for this article is signature-neutral (it writes velocity algebra only) and is written in the chapter's forward boost convention, so no sign had to be converted — but its third derivation card (rapidity additivity and the boost group) belongs to Rapidity as an Additive Velocity Parameter, and is stated here and handed over, not derived. Its equation card $\gamma_u = \gamma_{u'}\gamma_v(1 + u'v/c^2)$ is asserted and never derived; deriving it in the forward form — as the time component of the linearly-transforming four-velocity — is this article's owned result, and it is what turns subluminality from a numerical fact into a factorisation. All numbers are recomputed in verify_aq-sr-relativistic-velocity-addition.py. Signature: $(+,-,-,-)$ , timelike-positive; $U\cdot U = +c^2$ .

Builds on. The Lorentz Transformation Equations · Four-Vectors · Time Dilation · Relativity of Simultaneity. Leads to. The Transverse Velocity Transformation · Velocity Composition and the Invariance of Light · Rapidity as an Additive Velocity Parameter · The Thomas–Wigner Rotation · Relativistic Aberration.

[//]: # (faq2)

Why is there no γ in the velocity-addition formula, when γ is everywhere else?

Because it cancels — and finding out where it went is the point of this article. In $u' = \mathrm{d}x'/\mathrm{d}t'$ both differentials carry the same factor $\gamma$ , so it divides straight out of the ratio, leaving a formula that looks suspiciously un-relativistic. But the $\gamma$ has not disappeared from the physics: boost the four-velocity instead of dividing immediately, and it turns up in the time component as $\gamma_{u'} = \gamma_u\gamma_v(1 - uv/c^2)$ . The rule contains no $\gamma$ precisely because velocity is a ratio; the $\gamma$ 's live in the numerator and denominator separately, and they cancel only when you insist on collapsing them into one number.

Is the denominator really "the boosted gamma"? That sounds like numerology.

It is an identity, provable in two lines. The four-velocity's time component is $U^0 = \gamma_u c$ . Boost it: $U'^0 = \gamma_v(U^0 - \beta_v U^1) = \gamma_v(\gamma_u c - \tfrac{v}{c}\gamma_u u) = \gamma_u\gamma_v c\,(1 - uv/c^2)$ . But $U'^0 = \gamma_{u'}c$ by definition. Equate and divide by $c$ : $\gamma_{u'} = \gamma_u\gamma_v(1 - uv/c^2)$ , so $1 - uv/c^2 = \gamma_{u'}/(\gamma_u\gamma_v)$ exactly. Check it numerically with $u = v = 0.6c$ : $\gamma_{u'} = 1$ (the object is at rest in $S'$ ), and indeed $1.25 \times 1.25 \times (1 - 0.36) = 1.5625 \times 0.64 = 1$ ✓.

Two beams approach each other at 0.99999c each. Isn't their relative speed 1.99998c?

Their lab closing rate is $1.99998c$ , and that number is perfectly legitimate — it is the rate at which the gap between them shrinks in the lab's coordinates, and nothing is travelling at it. The relative velocity — the speed of one beam measured in the rest frame of the other — is what the composition rule gives: $(0.99999 + 0.99999)/(1 + 0.99999^2) = \mathbf{0.9999999999c}$ . Only the second is a physical velocity, and only the second determines the collision energy. Coordinate rates in a single frame are allowed to exceed $c$ ; the speed of anything, measured in any inertial frame, is not.

Does the rule work for a photon? What is u ⊕ v when u = c?

It works, and it returns $c$ : $(c - v)/(1 - vc/c^2) = (c-v)/[(c-v)/c] = c$ , for every $v$ with $|v| < c$ . That is the invariance of the speed of light, sitting inside the arithmetic as a fixed point of the composition. The one case that breaks is $u = v = c$ , which gives $0/0$ — and that is not a defect but a warning: there is no rest frame of a photon, so "the velocity of light relative to light" is not a question the theory answers. That $c$ is the only speed invariant under every composition is the theorem of Velocity Composition and the Invariance of Light.

Why does Fresnel's ether drag coefficient come out of a theory with no ether?

Because it was never a statement about ether — it was a first-order expansion of the composition law, discovered eighty-seven years early by someone with the wrong picture. Compose $c/n$ (light in still water) with $v$ (the water's flow): $w = (c/n + v)/(1 + v/nc)$ . Expand in $v/c$ and you get $w \approx c/n + v(1 - 1/n^2)$ . The bracket is Fresnel's coefficient exactly — $0.437$ for water — and von Laue pointed this out in 1907. Fresnel had guessed the right number by imagining a partially-dragged ether; the number survives, and the ether does not. It is one of the cleanest examples in physics of a correct formula rescued from a demolished theory.

If velocity is a "projective coordinate", what is rapidity?

The natural additive coordinate on the same line. Because the boost acts linearly on $(U^0, U^1)$ , its action on the ratio $u = c\,U^1/U^0$ is fractional-linear, and the eigenvectors of the boost matrix are the null directions — which is a fancy way of saying that $c$ and $-c$ are its fixed points. The coordinate that turns that fractional-linear action into a shift is $\varphi = \operatorname{artanh}(u/c)$ , and in it the composition law becomes plain addition. This article deliberately stops at the doorstep: the hyperbolic geometry is owned by Hyperbolic Geometry of Boosts and Rapidity, and the kinematic payoff — that rapidity differences are boost-invariant, which is why collider physicists use them — by Rapidity as an Additive Velocity Parameter.