Four-Acceleration

Strap an accelerometer to a rocket. It reads a number. That number is not "the rate of change of the rocket's velocity" — because velocity is frame-dependent, and different observers watching the same rocket disagree about $d\vec{v}/dt$ by factors of $\gamma^2$ , $\gamma^3$ , even $\gamma^4$ . Yet the accelerometer reads one number, and everyone agrees on what it reads, because it is a physical instrument with a needle on it. Something invariant is being measured. What? The answer is the four-acceleration $A^\mu = dU^\mu/d\tau$ , and the striking fact about it is that it is always spacelike. It has to be. The four-velocity has constant length, $U\cdot U = +c^2$ , and differentiating a constant gives $A\cdot U = 0$ : the four-acceleration is Minkowski-orthogonal to the four-velocity. But $U$ is timelike, and anything orthogonal to a timelike vector is spacelike. So $A\cdot A = -\alpha^2 \le 0$ , and $\alpha$ — the positive number whose square is $-A\cdot A$ — is an invariant. Go to the particle's instantaneous rest frame and $A^\mu = (0,\ \vec{\alpha}\,)$ : purely spatial. That is what the accelerometer reads. Proper acceleration is not a coordinate artefact to be transformed away. It is a scalar, defined without reference to any frame, and it is the honest measure of how hard something is being pushed. Two consequences follow, and they are why engineers care. In the lab, a longitudinal push obeys $\alpha = \gamma^3 a_\parallel$ — so a particle at $\gamma = 1000$ needs a billion times the lab-frame acceleration to feel the same shove. That $\gamma^3$ is why a linac shows such brutal diminishing returns, and why accelerator physicists quote energy and never speed. And if $\alpha$ is held constant, the resulting worldline is not a parabola but a hyperbola — the curve on which Rindler coordinates, Born rigidity, the equivalence principle and the Unruh effect are all built.

Key takeaways

Signature $(+,-,-,-)$ : $\eta_{\mu\nu} = \mathrm{diag}(+1,-1,-1,-1)$ . Recalled by name from Four-Vectors (`aq-sr-four-vectors`), which derives $U\cdot U = +c^2$ and $A\cdot U = 0$ . This article takes those as given.
$A^\mu \equiv dU^\mu/d\tau = d^2x^\mu/d\tau^2$ — differentiate with respect to the invariant proper time.
Owned result: orthogonality plus a timelike $U$ forces $A$ to be SPACELIKE, hence $\boxed{A\cdot A = -\alpha^2 \le 0}$ . It is negative. In the instantaneous rest frame $A^\mu = (0,\ \vec{\alpha}\,)$ , so $\alpha$ is literally what an accelerometer reads — proper acceleration is thereby defined as an invariant, not computed in some frame.
Owned result: the lab reduction $\alpha = \gamma^3 a_\parallel$ and $\alpha = \gamma^2 a_\perp$ . Different powers of $\gamma$ , and the $\gamma^3$ is why pushing a fast particle along its motion is so much harder than deflecting it.
Owned result: the general formula $\alpha^2 = \gamma^6\big[a^2 - (\vec{v}\times\vec{a}\,)^2/c^2\big]$ , containing both special cases.
Owned result: constant $\alpha$ integrates to HYPERBOLIC motion, $x = (c^2/\alpha)\cosh(\alpha\tau/c)$ , $ct = (c^2/\alpha)\sinh(\alpha\tau/c)$ — not the Newtonian parabola. The worldline is $x^2 - c^2t^2 = (c^2/\alpha)^2$ , and it reproduces $A\cdot A = -\alpha^2$ exactly.
Key idea: the four-acceleration is always orthogonal to the four-velocity, and therefore always spacelike. A particle cannot accelerate "in the time direction" — which is precisely why $|\vec{v}|$ can approach $c$ forever and never arrive. The speed limit is orthogonality, not a force running out.

The core insight

The three-acceleration $\vec{a} = d\vec{v}/dt$ is bookkeeping: frame-dependent, and not even the spatial part of $A^\mu$ . The invariant is $\alpha$ , and the cleanest way to see it is that $\alpha/c^2$ is the curvature of the worldline in Minkowski space. A straight worldline has zero curvature and zero proper acceleration; a curved one has both, and how tightly it curves is what the accelerometer reports.

Where this lives today

Proper acceleration is the quantity synchrotron radiation actually depends on. The radiated power is $P = \dfrac{q^2\gamma^6}{6\pi\epsilon_0c^3}\big[a^2 - (\vec v\times\vec a)^2/c^2\big] = \dfrac{q^2\alpha^2}{6\pi\epsilon_0c^3}$ — the relativistic Larmor formula, whose bracket is precisely equation (4). The loss is therefore set by the invariant $\alpha$ and nothing else, which is why electrons in a storage ring bleed energy catastrophically (huge transverse $\alpha$ ) while the same electrons in a linac barely radiate at all. It is the reason LEP was replaced by the LHC in the same tunnel: protons, $\sim\!1836$ times heavier, reach a given energy at a far smaller $\gamma$ , and $\alpha$ — hence the radiated power — collapses with it. The hyperbolic worldline of equation (5) is the flat-spacetime laboratory for horizon physics. A uniformly accelerated observer asymptotes to the light cone, which becomes a causal horizon: signals from beyond it never arrive. That is the Rindler horizon, the arena in which Born rigidity is defined (`aq-sr-born-rigidity`) and in which the Unruh effect assigns a temperature proportional to the invariant $\alpha$ — as it must be, since a temperature cannot depend on a coordinate choice.

Origins · 1 minFour-acceleration and how this was discovered

Four-acceleration

1909–1966 · Minkowski's Eigenzeit · Born's hyperbolic worldline · Rindler's horizon · the invariant an accelerometer reads

The four-acceleration has no single author; it is what you get the moment you have a four-velocity and take another derivative with respect to proper time — a step implicit in Minkowski's 1908 Grundgleichungen. What had to be discovered was the physics on the far side of it. In 1909 Max Born, working out what "rigid" could mean for an accelerating electron, wrote down the worldline of constant proper acceleration and found it is a hyperbola, asymptotic to the light cone: the motion now universally called hyperbolic. That single curve became one of the most productive objects in physics — the trajectory of the twin who turns around, the congruence that defines Born-rigid motion, and, once Wolfgang Rindler had drawn out its coordinate system in the 1960s, the flat-spacetime model of a black-hole horizon. The invariant $\alpha$ later acquired a stranger role still: in 1976 William Unruh showed that an observer with proper acceleration $\alpha$ perceives the Minkowski vacuum as a thermal bath at $T = \hbar\alpha/(2\pi ck_B)$ . Reaching one kelvin needs $\alpha \approx 2.5\times10^{20}\ \mathrm{m/s^2}$ — comparable to what an LHC proton feels on the bend — and it has never been observed. It remains the sharpest known link between acceleration, horizons and quantum theory.

History

1908

— Minkowski's Grundgleichungen introduces proper time and the four-dimensional derivative dx/dτ, from which the four-acceleration follows immediately.

1909

— Max Born, in "Die Theorie des starren Elektrons in der Kinematik des Relativitätsprinzips" (Ann. Phys. 335, 1–56), finds that constant proper acceleration gives a hyperbolic worldline, and uses it to define rigid motion.

1909

— Sommerfeld and von Laue take up hyperbolic motion; it becomes the standard example of relativistic accelerated motion.

1960

s — Wolfgang Rindler develops the coordinate system of the uniformly accelerated observer and identifies the causal horizon of hyperbolic motion.

1976

— William Unruh shows that an observer of proper acceleration α sees the vacuum as a thermal bath at T = ħα/(2πck_B), tying the invariant directly to quantum field theory.

The physical picture

Everything here descends from one line, established at `aq-sr-four-vectors`: since $U\cdot U = c^2$ is constant, differentiating along the worldline gives $\frac{d}{d\tau}(U\cdot U) = 2\,A\cdot U = 0$ , hence $A\cdot U = 0$ . We recall that result rather than re-derive it. What we do here is take it seriously, because it has a consequence that is easy to state and hard to over-state: $A^\mu$ is always spacelike. Why? A timelike vector — and $U$ is timelike, $U\cdot U = +c^2 > 0$ — has a rest frame, one in which it points purely along the time axis, $U^\mu = (c, \vec{0}\,)$ . There, $A\cdot U = 0$ reads $A^0c = 0$ , so $A^0 = 0$ . The four-acceleration has no time component in the particle's own frame: it is purely spatial, and a purely spatial four-vector is spacelike, with a negative norm. Hence $A\cdot A = -|\vec{A}|^2 \equiv -\alpha^2 \le 0$ — and since $A\cdot A$ is an invariant, that classification holds in every frame, whatever the components look like. This is not a formal curiosity. It says a particle cannot accelerate in the time direction, and that is the real reason $v$ approaches $c$ and never reaches it. It is not that the engines give out. It is that the four-velocity is confined to a hyperbola by its own constant length, and all an acceleration can do is slide it along.

Figure

PUNCHLINE — the four-acceleration is ALWAYS orthogonal to the four-velocity, and therefore ALWAYS spacelike. LEFT: U rides the hyperbola U·U = +c², so it has fixed length; A = dU/dτ is tangent to that hyperbola, which is exactly the Minkowski-orthogonal direction. In the particle's own rest frame U = (c, 0) and orthogonality forces A⁰ = 0, so A = (0, α) is purely spatial — and α is what the accelerometer reads. RIGHT: because A is spacelike, A·A = −α² is NEGATIVE. In the (−,+,+,+) signature it would be +α². Get this sign wrong and every threshold in the dynamics chapter is corrupted, with no numeric check to catch it

Now the lab. The relation between the invariant $\alpha$ and the coordinate acceleration $\vec{a} = d\vec{v}/dt$ that an observer measures is not a single factor of $\gamma$ , and it is not even a single factor at all: it depends on the direction of the push relative to the motion. Push along the velocity and you pay $\gamma^3$ . Push across it and you pay only $\gamma^2$ . That asymmetry is real, it is measurable, and it is the reason linear accelerators are long.

Figure

PUNCHLINE — the γ³ is why a linac has brutal diminishing returns. An electron in a 20 MV/m field feels a CONSTANT proper acceleration α = eE/m = 3.52×10¹⁸ m/s² (about 3.6×10¹⁷ g) — the accelerometer reading never changes. But the LAB-frame acceleration a = α/γ³ collapses: by γ = 1000 it has fallen by a factor of a BILLION, to 3.52×10⁹ m/s². The particle is being pushed exactly as hard as ever and is barely speeding up at all. This is why accelerator physicists quote ENERGY and never speed. Transverse pushes cost only γ², which is why bending a beam is comparatively cheap — an LHC proton at γ = 7247 on the ring feels α = γ²c²/R = 1.11×10²¹ m/s²

The core equations

A^{\mu} \equiv \frac{dU^{\mu}}{d\tau} = \frac{d^{2}x^{\mu}}{d\tau^{2}}, \qquad\qquad \boxed{\;A\cdot U = 0\;} \qquad (1)

where: $\tau$ — proper time; the orthogonality $A\cdot U = 0$ is derived at `aq-sr-four-vectors` and recalled here. Physical reading: $A^\mu$ is a four-vector because $d\tau$ is a scalar. The orthogonality is not an assumption but a theorem, and it follows from nothing more than $U\cdot U$ being constant.

\boxed{\;A\cdot A = \eta_{\mu\nu}A^{\mu}A^{\nu} = -\,\alpha^{2} \;\le\; 0\;}, \qquad\qquad A^{\mu}\big|_{\text{rest frame}} = \big(0,\ \vec{\alpha}\,\big) \qquad (2)

where: $\alpha \equiv \sqrt{-A\cdot A}$ — the proper acceleration, a Lorentz scalar. Physical reading: the sign is negative and it matters. $A$ is spacelike because it is orthogonal to a timelike vector. In $(-,+,+,+)$ this reads $+\alpha^2$ , and a silent flip between conventions corrupts every downstream result without changing a digit of any numeric check. The lab-frame components, obtained by pushing $d/d\tau = \gamma\,d/dt$ through $U^\mu = \gamma(c,\vec{v}\,)$ :

A^{\mu} = \left(\ \gamma^{4}\frac{\vec{v}\cdot\vec{a}}{c}\ ,\ \ \gamma^{2}\vec{a} + \gamma^{4}\frac{(\vec{v}\cdot\vec{a}\,)}{c^{2}}\,\vec{v}\ \right) \qquad (3)

where: $\vec{a} = d\vec{v}/dt$ — the ordinary three-acceleration. Physical reading: the spatial part of $A^\mu$ is not $\vec{a}$ , and not $\gamma^2\vec{a}$ either — there is an extra term along $\vec{v}$ , and $\vec{A}$ is not even parallel to $\vec{a}$ in general. This is the single most common algebra slip in the subject. Contracting (3) with itself gives the general lab formula, and its two famous special cases:

\boxed{\;\alpha^{2} = \gamma^{6}\!\left[\,a^{2} - \frac{(\vec{v}\times\vec{a}\,)^{2}}{c^{2}}\,\right]\;} \qquad\Longrightarrow\qquad \boxed{\;\alpha = \gamma^{3}a_{\parallel}\;}\quad\text{and}\quad \boxed{\;\alpha = \gamma^{2}a_{\perp}\;} \qquad (4)

where: $a_\parallel$ , $a_\perp$ — the components of $\vec{a}$ along and across $\vec{v}$ . Physical reading: for a purely longitudinal push $\vec{v}\times\vec{a} = 0$ and the bracket is $a^2$ , giving $\alpha = \gamma^3 a$ . For a purely transverse push $|\vec{v}\times\vec{a}| = va$ , the bracket is $a^2(1-\beta^2) = a^2/\gamma^2$ , and $\alpha = \gamma^2 a$ . The two differ by a factor of $\gamma$ , and that factor is the whole economics of accelerator design. Finally, the integral of $\alpha = \text{const}$ — Newton's parabola, relativistically corrected into a hyperbola:

x(\tau) = \frac{c^{2}}{\alpha}\cosh\!\frac{\alpha\tau}{c}, \qquad ct(\tau) = \frac{c^{2}}{\alpha}\sinh\!\frac{\alpha\tau}{c} \qquad\Longrightarrow\qquad x^{2} - c^{2}t^{2} = \left(\frac{c^{2}}{\alpha}\right)^{2} \qquad (5)

where: the worldline is the invariant hyperbola at spacelike interval $c^2/\alpha$ from the origin. Physical reading: hyperbolic motion. As $\tau \to \infty$ the worldline hugs the light cone asymptotically but never touches it. This is the curve on which Rindler coordinates, Born rigidity (`aq-sr-born-rigidity`) and the Unruh effect are all built.

Deriving the four-acceleration

Derivation AOrthogonality forces a spacelike four-acceleration — so alpha is an invariant

The goal is to establish the article's foundational claim: that $A\cdot A \le 0$ always, with no exceptions, and that the resulting invariant $\alpha$ is exactly the accelerometer reading. The argument is short, but the order of the logic matters and is often garbled. We do not first define $\alpha$ in the rest frame and then show it is invariant — that would be circular. We show the vector is spacelike by a frame-free argument, deduce that a rest frame exists in which it is purely spatial, and only then read off what the instrument measures.

Step 1. Recall the orthogonality. From `aq-sr-four-vectors`: $U\cdot U = +c^2$ is constant along every massive worldline, so $\dfrac{d}{d\tau}(U\cdot U) = 2\,A\cdot U = 0$ , hence $A\cdot U = 0$ . Why: this is recalled, not re-derived — it is 1890's owned result. The metric $\eta$ is constant, so it passes through the derivative, and the product rule supplies the factor 2. No components were used, so it holds for arbitrary motion.

Step 2. Deduce spacelikeness, without choosing a frame. $U$ is timelike ( $U\cdot U = +c^2 > 0$ ), and a general theorem of Minkowski geometry says any vector orthogonal to a timelike vector is spacelike (or zero). Hence $A\cdot A \le 0$ , and we define $\alpha \ge 0$ by $\boxed{A\cdot A = -\alpha^2}$ . Why: $A\cdot A$ is a full contraction, hence a Lorentz scalar — the same number in every frame. "Spacelike" is therefore an absolute classification, and $\alpha$ is well-defined before any frame has been named. That is what makes the next step a measurement rather than a convention.

Step 3. Now go to the instantaneous rest frame and read the instrument. Because $U$ is timelike there is a frame in which $U^\mu = (c,\ \vec{0}\,)$ . In it, $A\cdot U = \eta_{\mu\nu}A^\mu U^\nu = A^0 c = 0$ , forcing $A^0 = 0$ . So $A^\mu = (0,\ \vec{\alpha}\,)$ — purely spatial — and $A\cdot A = 0 - |\vec\alpha|^2 = -|\vec\alpha|^2$ , confirming $|\vec{\alpha}| = \alpha$ . Why: an accelerometer is a device that is momentarily at rest with respect to the thing it is bolted to. It therefore measures precisely $|\vec{\alpha}|$ in this frame — and by Step 2 that number is the invariant. Proper acceleration is not "the acceleration in a particular frame"; it is a scalar that happens to be easiest to read in that frame.

Step 4. Check the boundary case. If $A\cdot A = 0$ with $A \ne 0$ , then $A$ would be null and orthogonal to timelike $U$ — impossible, since the only such vector is zero. So $A\cdot A = 0 \iff A^\mu = 0 \iff$ inertial motion. Why: the classification is clean. Zero proper acceleration means a straight worldline, means free fall, means the accelerometer reads zero. There is no degenerate middle case. *Verified symbolically in verify_aq-sr-four-acceleration.py: $A\cdot U = 0$ for an arbitrary $\vec{v}(t)$ in 3D; $A\cdot A \le 0$ checked over 300 random $(\vec v, \vec a)$ with $|\vec v| < c$ ; and $A\cdot A = -\alpha^2$ asserted negative (the base's summary calls it "positive with $(+,-,-,-)$ ", which is wrong twice over).*

Go deeper · Derivation B · 2 minThe lab-frame reduction — where the gamma-cubed comes from

Derivation BThe lab-frame reduction — where the gamma-cubed comes from

An accelerometer reads $\alpha$ , but an experimenter with a ruler and a clock measures $\vec{a} = d\vec{v}/dt$ . The relation between them is the practical heart of relativistic dynamics, and it is not a single factor of $\gamma$ — it depends on the angle between the push and the motion. This card derives the general formula and extracts the two cases that matter, because the difference between $\gamma^3$ and $\gamma^2$ is the difference between a three-kilometre linac and a bending magnet.

Step 1. Convert the derivative. $\dfrac{d}{d\tau} = \dfrac{dt}{d\tau}\dfrac{d}{dt} = \gamma\dfrac{d}{dt}$ , so $A^\mu = \gamma\dfrac{d}{dt}\Big[\gamma\big(c,\ \vec{v}\,\big)\Big]$ . We will need $\dfrac{d\gamma}{dt} = \gamma^3\dfrac{\vec{v}\cdot\vec{a}}{c^2}$ , which follows from differentiating $\gamma = (1-v^2/c^2)^{-1/2}$ . Why: everything below is the product rule applied to this one expression. The $\gamma^3$ that will dominate the story is already visible here — it is born in $d\gamma/dt$ , not in the boost.

Step 2. Assemble the components. Time slot: $A^0 = \gamma\,\dfrac{d(\gamma c)}{dt} = \gamma\cdot c\cdot\gamma^3\dfrac{\vec{v}\cdot\vec{a}}{c^2} = \gamma^4\dfrac{\vec{v}\cdot\vec{a}}{c}$ . Space slot: $\vec{A} = \gamma\dfrac{d(\gamma\vec{v}\,)}{dt} = \gamma\Big[\gamma\vec{a} + \gamma^3\dfrac{(\vec{v}\cdot\vec{a}\,)}{c^2}\vec{v}\Big] = \gamma^2\vec{a} + \gamma^4\dfrac{(\vec{v}\cdot\vec{a}\,)}{c^2}\vec{v}$ . Why: note the second term. The spatial part of $A^\mu$ is not simply $\gamma^2\vec{a}$ , and $\vec{A}$ is generally not even parallel to $\vec{a}$ . Forgetting this term is the classic slip.

Step 3. Contract, and simplify. $A\cdot A = (A^0)^2 - |\vec{A}|^2$ ; grinding through (done symbolically in the verify script) it collapses to $A\cdot A = -\gamma^{6}\big[a^{2} - (\vec{v}\times\vec{a}\,)^{2}/c^{2}\big]$ , hence $\alpha^{2} = \gamma^{6}\big[a^{2} - (\vec{v}\times\vec{a}\,)^{2}/c^{2}\big]$ . Why: the cross-product is doing real work. It measures how much of $\vec{a}$ is transverse to $\vec{v}$ , and transverse acceleration is discounted — which is exactly why it will cost one power of $\gamma$ less.

Step 4. Read off the two cases. Longitudinal ( $\vec{a} \parallel \vec{v}$ ): the cross product vanishes, the bracket is $a^2$ , and $\boxed{\alpha = \gamma^3 a_\parallel}$ . Transverse ( $\vec{a} \perp \vec{v}$ ): $|\vec{v}\times\vec{a}| = va$ , so the bracket is $a^2(1 - \beta^2) = a^2/\gamma^2$ , and $\alpha^2 = \gamma^6 a^2/\gamma^2 = \gamma^4a^2$ , giving $\boxed{\alpha = \gamma^2 a_\perp}$ . Why: a factor of $\gamma$ separates them. To give a fast particle the same felt push, you must supply $\gamma^3$ times the lab acceleration if you push it forwards, but only $\gamma^2$ if you push it sideways. Equivalently: at fixed lab force, a fast particle is far more reluctant to speed up than to turn.

Step 5. Convert to money. An electron in a $20$ MV/m linac field feels $\alpha = eE/m_e = \mathbf{3.52\times10^{18}\ m/s^{2}}$ — and it is constant, because for a longitudinal field $\alpha = \gamma^3 a = F/m$ exactly. But the lab acceleration is $a = \alpha/\gamma^3$ , so by $\gamma = 1000$ it has fallen to $\mathbf{3.52\times10^{9}\ m/s^{2}}$ : a billion-fold collapse. Why: the machine pushes exactly as hard as it ever did and the particle barely speeds up. This is why accelerator physicists never quote speed. Speed stopped being informative around $\gamma \approx 3$ ; energy never stops. *Verified symbolically in verify_aq-sr-four-acceleration.py: equation (3) derived from $\gamma\,d/dt[\gamma(c,\vec v)]$ ; $\alpha^2 = \gamma^6[a^2 - (\vec v\times\vec a)^2/c^2]$ proved for arbitrary symbolic $\vec{v}(t)$ in 3D; both special cases $\gamma^3a_\parallel$ and $\gamma^2a_\perp$ recovered from it; and $\alpha = eE/m_e = 3.52\times10^{18}$ m/s², with $a(\gamma{=}1000) = 3.52\times10^{9}$ m/s².*

Go deeper · Derivation C · 2 minConstant proper acceleration gives a hyperbola, not a parabola

Derivation CConstant proper acceleration gives a hyperbola, not a parabola

Newton says: push with constant acceleration and the worldline is a parabola, $x = \tfrac12 at^2$ , and the speed grows without limit. Relativity cannot allow that, so something must give — and what gives is the shape of the curve. This card integrates $A\cdot A = -\alpha^2 = \text{const}$ and shows the worldline is a hyperbola, then verifies the solution by feeding it back through the definition. This is 1917's owned result; `aq-sr-born-rigidity` uses it as given.

Step 1. Set up in rapidity, the variable that linearises the problem. Write $U^\mu = c\,(\cosh\varphi,\ \sinh\varphi)$ , which satisfies $U\cdot U = c^2(\cosh^2\varphi - \sinh^2\varphi) = c^2$ identically, so $\varphi$ automatically parametrises the hyperbola; then $\beta = \tanh\varphi$ , $\gamma = \cosh\varphi$ . Why: any timelike unit vector can be written this way — nothing has been assumed, only good coordinates chosen. The payoff is immediate.

Step 2. Compute $A^\mu$ and impose the constraint. $A^\mu = dU^\mu/d\tau = c\,\dot\varphi\,(\sinh\varphi,\ \cosh\varphi)$ , so $A\cdot A = c^2\dot\varphi^2(\sinh^2\varphi - \cosh^2\varphi) = -c^2\dot\varphi^2$ . Setting this to $-\alpha^2$ gives $\boxed{\dot\varphi = \alpha/c}$ , hence $\varphi(\tau) = \alpha\tau/c$ for a particle starting from rest. Why: rapidity grows linearly in proper time. That is the relativistic replacement for "velocity grows linearly in time" — and rapidity, unlike velocity, has no ceiling to bump against, which is exactly how the motion continues forever without $v$ reaching $c$ .

Step 3. Integrate to get the worldline. $\dfrac{d(ct)}{d\tau} = U^0 = c\cosh(\alpha\tau/c)$ and $\dfrac{dx}{d\tau} = U^1 = c\sinh(\alpha\tau/c)$ . Integrating, with $x(0) = c^2/\alpha$ and $t(0) = 0$ :

ct = \frac{c^{2}}{\alpha}\sinh\frac{\alpha\tau}{c}, \qquad x = \frac{c^{2}}{\alpha}\cosh\frac{\alpha\tau}{c} \qquad\Longrightarrow\qquad x^{2} - c^{2}t^{2} = \left(\frac{c^{2}}{\alpha}\right)^{2}.

Why: a hyperbola, of constant spacelike interval $c^2/\alpha$ from the origin — hence the name "hyperbolic motion". The offset $x(0) = c^2/\alpha$ is a choice of origin that makes the algebra clean; it puts the vertex of the hyperbola on the $x$ -axis.

Step 4. Verify by substitution, which is the only honest check. Differentiate twice: $A^0 = d^2(ct)/d\tau^2 = \alpha\sinh(\alpha\tau/c)$ and $A^1 = d^2x/d\tau^2 = \alpha\cosh(\alpha\tau/c)$ . Then $A\cdot A = \alpha^2\sinh^2 - \alpha^2\cosh^2 = \boxed{-\alpha^2}$ ✓, and $A\cdot U = \alpha\sinh\cdot c\cosh - \alpha\cosh\cdot c\sinh = 0$ ✓. Why: the solution reproduces both structural facts exactly — the constant negative norm and the orthogonality. That closes the loop.

Step 5. Recover Newton, and see where he fails. For $\alpha\tau/c \ll 1$ : $\cosh u \approx 1 + u^2/2$ , so $x \approx c^2/\alpha + \tfrac12\alpha\tau^2$ — the parabola. But as $\tau \to \infty$ the hyperbola's asymptote is the light line $x = ct$ : the worldline approaches the light cone and never crosses it. Why: the parabola is the small- $\tau$ tangent to the hyperbola. Newton was not wrong; he was looking at a short arc of a curve whose far end he could not see. At $1\,g$ , "short" means much less than $c/g \approx 1$ year. *Verified symbolically in verify_aq-sr-four-acceleration.py: the hyperbolic worldline substituted back into $A^\mu = d^2x^\mu/d\tau^2$ reproduces $A\cdot A = -\alpha^2$ and $A\cdot U = 0$ exactly; $U\cdot U = +c^2$ identically; $\dot\varphi = \alpha/c$ ; $x^2 - c^2t^2 = (c^2/\alpha)^2$ ; and the small- $\tau$ expansion recovers $x \approx c^2/\alpha + \tfrac12\alpha\tau^2$ .*

Limiting cases

Case	Result	Meaning
$v \to 0$ , $\gamma \to 1$	$A^\mu \to (0,\ \vec{a}\,)$ , $\ \alpha \to a$	proper and coordinate acceleration coincide — the Newtonian regime
instantaneous rest frame	$A^\mu = (0,\ \vec{\alpha}\,)$	purely spatial; $\alpha$ is the accelerometer reading
longitudinal push	$\alpha = \gamma^3 a_\parallel$	brutal: at $\gamma = 1000$ , a billion times harder to speed up
transverse push	$\alpha = \gamma^2 a_\perp$	one power of $\gamma$ cheaper — bending beats boosting
inertial motion	$A^\mu = 0$ , $\ \alpha = 0$	straight worldline; the accelerometer reads zero
$\alpha = \text{const}$	hyperbola, $x^2 - c^2t^2 = (c^2/\alpha)^2$	rapidity grows linearly in $\tau$ ; $v \to c$ asymptotically, never arriving

Common trapA dot A is NEGATIVE. The sign is the whole convention, and no number will catch it

In $(+,-,-,-)$ the four-acceleration is spacelike, so $A\cdot A = -\alpha^2 \le 0$ . It cannot be otherwise: $A$ is orthogonal to the timelike $U$ , and orthogonal-to-timelike means spacelike. In $(-,+,+,+)$ the same physics is written $+\alpha^2$ . Mixing them is silent, catastrophic and undetectable by any numeric check — $\alpha$ itself comes out the same positive number either way, so a unit test on $\alpha$ passes while every sign downstream is wrong. The AllQFT base for this article contains exactly this defect: its equation card correctly gives $a^\mu a_\mu = -\alpha^2$ , while its summary says $a^\mu a_\mu = \alpha^2$ and calls it "spacelike, positive with $(+,-,-,-)$ " — wrong twice, since spacelike is negative here. Read the metric before you read the formula.

Worked examples

1 · An electron in a linac: what it feels, and what the lab sees

An electron sits in a linear accelerator with a longitudinal field $E = 20$ MV/m. What proper acceleration does it feel? What lab-frame acceleration does an observer measure when the electron has reached $\gamma = 1000$ ?

Step 1 — identify. Longitudinal push, so $\alpha = \gamma^3 a_\parallel$ . For a longitudinal field the force in the lab is $F = eE$ and $dp/dt = F$ with $p = \gamma m v$ , which gives $F = \gamma^3 m a$ — so $\alpha = \gamma^3 a = F/m$ exactly, and independently of $\gamma$ . The proper acceleration is constant.

Step 2 — what the electron feels. $\alpha = \dfrac{eE}{m_e} = \dfrac{(1.602176634\times10^{-19}\ \mathrm{C})(2\times10^{7}\ \mathrm{V/m})}{9.1093837\times10^{-31}\ \mathrm{kg}} = \mathbf{3.52\times10^{18}\ m/s^{2}}$ .

Step 3 — in human units. Dividing by $g = 9.81$ : $\alpha = \mathbf{3.6\times10^{17}\,g}$ — and an accelerometer bolted to the electron reads this the whole way down the machine, at injection and at full energy alike.

Step 4 — what the lab sees at $\gamma = 1000$ . $a = \alpha/\gamma^3 = \dfrac{3.52\times10^{18}}{10^{9}} = \mathbf{3.52\times10^{9}\ m/s^{2}}$ .

Step 5 — the boundary number. The lab acceleration has fallen by a factor of $\mathbf{10^{9}}$ while the force, and the felt acceleration, have not changed at all. The machine pushes exactly as hard as it did on day one, and the electron speeds up a billion times more slowly. At $\gamma = 1000$ the electron is at $\beta = 0.9999995$ ; another kilometre of accelerator will barely move that digit, but it will double the energy. That is why "how fast is the beam?" gets a shrug at any modern accelerator, and "how energetic?" gets a number.

2 · The LHC ring: the most violently accelerated matter we make

A proton circulates in the LHC at $6.8$ TeV. Treat the ring as a circle of radius $R = C/2\pi$ with circumference $C = 26.66$ km. What proper acceleration does the proton feel on the bend?

Step 1 — identify. Circular motion is a purely transverse push ( $\vec{v}\cdot\vec{a} = 0$ ), so $\alpha = \gamma^2 a_\perp$ with $a_\perp = v^2/R \approx c^2/R$ .

Step 2 — the Lorentz factor. $\gamma = \dfrac{E}{m_pc^2} = \dfrac{6.8\times10^{6}\ \mathrm{MeV}}{938.272\ \mathrm{MeV}} = \mathbf{7247}$ .

Step 3 — the lab acceleration. $R = 26659\ \mathrm{m}/2\pi = \mathbf{4243\ m}$ , so $a_\perp \approx \dfrac{c^2}{R} = \dfrac{8.98755\times10^{16}}{4243} = \mathbf{2.12\times10^{13}\ m/s^{2}}$ .

Step 4 — the proper acceleration. $\alpha = \gamma^2 a_\perp = (7247)^2 \times 2.12\times10^{13} = 5.25\times10^{7} \times 2.12\times10^{13} = \mathbf{1.11\times10^{21}\ m/s^{2}}$ .

Step 5 — the boundary number. That is $\mathbf{1.13\times10^{20}\,g}$ — about as violent an acceleration as humanity produces anywhere. And note it cost only $\gamma^2$ , not $\gamma^3$ : had this been a longitudinal push, the same felt acceleration would have demanded another factor of $7247$ . Bending is cheap; boosting is not. That asymmetry is exactly what makes a storage ring a sane way to hold a beam and an insane way to accelerate one.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §§ on hyperbolic motion · Griffiths, Introduction to Electrodynamics, 4th ed., §12.2 · Jackson, Classical Electrodynamics, 3rd ed., §14.1 (the relativistic Larmor formula) · Born, M., Ann. Phys. 335, 1–56 (1909) · Unruh, W. G., "Notes on black-hole evaporation", Phys. Rev. D 14, 870 (1976).

Provenance. Signature $(+,-,-,-)$ , so $U\cdot U = +c^2$ and $A\cdot A = -\alpha^2$ (spacelike, NEGATIVE). Everything is recomputed in verify_aq-sr-four-acceleration.py, which also greps this article for each printed value: $A\cdot U = 0$ and $A\cdot A \le 0$ for arbitrary symbolic $\vec{v}(t)$ in 3D; $\alpha^2 = \gamma^6[a^2 - (\vec v\times\vec a\,)^2/c^2]$ proved symbolically, with $\alpha = \gamma^3a_\parallel$ and $\alpha = \gamma^2a_\perp$ recovered as its two special cases; the hyperbolic worldline substituted back to reproduce $A\cdot A = -\alpha^2$ and $A\cdot U = 0$ exactly; $\alpha_{\text{linac}} = eE/m_e = 3.52\times10^{18}\ \mathrm{m/s^2} = 3.6\times10^{17}g$ , with $a(\gamma{=}1000) = 3.52\times10^{9}\ \mathrm{m/s^2}$ ; and the LHC proton at $\gamma = 7247$ , $R = 4243$ m, $\alpha = 1.11\times10^{21}\ \mathrm{m/s^2} = 1.13\times10^{20}g$ . Base defect repaired: the base's eqCard correctly gives $a^\mu a_\mu = -\alpha^2$ , but its summary states $a^\mu a_\mu = \alpha^2$ and calls it "spacelike, positive with $(+,-,-,-)$ " — wrong twice. The sign is re-derived from $\eta$ here.

Scope. $U\cdot U = +c^2$ and $A\cdot U = 0$ are derived at `aq-sr-four-vectors` and recalled by name. The Minkowski four-force $K^\mu = mA^\mu$ is named, not developed — that is `aq-sr-four-force`, and the force law `aq-sr-relativistic-force`. Born-rigid motion, which uses the hyperbolic worldline derived here, is `aq-sr-born-rigidity`. The Unruh temperature is quoted, not derived.

Builds on. Four-Velocity · Four-Vectors · Proper Time · The Lorentz Factor. Leads to. Four-Force · Relativistic Force · Born Rigidity · Relativistic Kinetic Energy.

If A is orthogonal to U, does that mean they look perpendicular on a spacetime diagram?

No, and expecting them to is the commonest confusion here. Minkowski orthogonality is $A\cdot U = \eta_{\mu\nu}A^\mu U^\nu = 0$ , with the metric's minus signs, and it does not correspond to a right angle on Euclidean paper — Minkowski-orthogonal vectors appear to make equal angles with the light line, which for a fast particle looks alarmingly like the two arrows nearly coincide. The physical content is unambiguous regardless: in the rest frame $U = (c,\vec 0\,)$ and orthogonality forces $A^0 = 0$ . Trust the contraction, not the picture.

Why is the longitudinal factor γ³ and the transverse one γ²?

Because a longitudinal push must change the particle's speed, and a transverse push need not. Changing speed changes $\gamma$ , and that variation contributes the extra term in $\vec{A} = \gamma^2\vec{a} + \gamma^4(\vec{v}\cdot\vec{a}\,)\vec{v}/c^2$ — nonzero only when $\vec{v}\cdot\vec{a} \ne 0$ . For a transverse push that term dies and you are left with $\gamma^2$ ; for a longitudinal one it survives and conspires with the first term to give $\gamma^3$ . Physically: turning a particle is cheaper than speeding it up, because turning does not have to fight the divergence of $\gamma$ as $v \to c$ .

Is proper acceleration bounded? Can α be arbitrarily large?

There is no relativistic bound on $\alpha$ at all. Unlike velocity, it is an invariant with no ceiling — you may make it as large as you can afford. What relativity forbids is not large $\alpha$ but reaching $c$ , and a particle at enormous constant $\alpha$ still only approaches the light cone asymptotically along its hyperbola. The real limits come from other physics: radiation reaction (power $\propto \alpha^2$ ), field breakdown in the accelerating structure, and — around $\alpha \sim 10^{29}\ \mathrm{m/s^2}$ for an electron — the Schwinger limit, where the vacuum itself starts to break down.

Does a photon have a four-acceleration?

No, for the same reason it has no four-velocity: $d\tau = 0$ along a null worldline, so neither $dx^\mu/d\tau$ nor $d^2x^\mu/d\tau^2$ exists. The whole construction here — differentiate with respect to proper time, use the constancy of $U\cdot U$ — presupposes a timelike worldline with a proper time to differentiate against, and light has neither. Nor is this a gap: a photon in vacuum travels on a straight null geodesic and is never accelerated.

What is the physical meaning of A·A being an invariant?

It means "how hard is this thing being pushed" has a single, frame-independent answer — which is not true of $\vec{a} = d\vec{v}/dt$ . Different observers watching the same rocket disagree wildly about its coordinate acceleration, by factors of $\gamma^2$ to $\gamma^4$ depending on geometry. But they all agree on $\alpha = \sqrt{-A\cdot A}$ , and they must, because it is the reading of a physical dial. This is the four-vector formalism doing the job it was built for: separating what is real (the invariant) from what is bookkeeping (the components).

Why is the hyperbola the right curve, when Newton gave a parabola?

Because a parabola implies unbounded speed, and the hyperbola is what you get when $U\cdot U = c^2$ is enforced honestly. In rapidity variables it is transparent: constant proper acceleration means rapidity grows linearly with proper time, $\varphi = \alpha\tau/c$ . Rapidity has no ceiling, so the motion continues forever — but $\beta = \tanh\varphi$ , and $\tanh$ saturates at 1. Rapidity is the quantity that behaves the way Newton expected velocity to behave. The parabola is the small- $\tau$ approximation, $\cosh u \approx 1 + u^2/2$ , excellent right up until $\alpha\tau \sim c$ — which at $1\,g$ is about a year.