The Equivalence Principle and the Boundary of Special Relativity

There is a piece of folklore that has to die before this chapter can end: "you need general relativity to handle accelerated frames." You do not. The four articles before this one handled accelerated frames completely, in flat spacetime, with nothing but special relativity — hyperbolic worldlines, a position-dependent proper acceleration, a chart with a clock gradient, a causal horizon, a rigid body whose ends never agree about time. `aq-sr-rindler-metric` computes the full Riemann tensor of the Rindler chart and gets $R^{\lambda}{}_{\mu\nu\kappa} = 0$ , every component, exactly. Curved-looking coordinates. Flat spacetime. No gravity anywhere. So what does need general relativity?

Tides. Not acceleration — gradients of acceleration. A uniform field can be transformed away by falling; that is Einstein's insight, and it is exact. A real field, made by a real mass, is never uniform: it points inward, so a freely-falling ball of dust is squeezed sideways and stretched vertically, and no change of coordinates removes that. The object that measures it is the tidal tensor $\partial_i\partial_j\Phi$ , whose relativistic form is the Riemann tensor. That is the boundary. Not acceleration. Curvature. This article puts a number on it. The principle is local, everyone says so, and almost nobody says how local. It has a size and an expiry time, related by one inequality:

L\,T^{2} \;<\; \frac{\delta\, r^{3}}{GM} .

A lab of size $L$ , falling freely for a time $T$ , measuring positions to $\delta$ . At Earth's surface, to a micron: a one-metre lab is an inertial frame for 0.81 seconds. After that the tides show up and special relativity has run out of road.

Key takeaways

Signature $(+,-,-,-)$ : $\mathrm{d}s^2 = c^2\mathrm{d}t^2 - \mathrm{d}x^2$ , timelike positive. Everything in this article except the tidal tensor lives in flat spacetime.
The folklore is false. "You need GR for accelerated frames" is wrong. SR handles acceleration completely: `aq-sr-relativistic-uniform-acceleration`, `aq-sr-rindler-coordinates`, `aq-sr-rindler-metric` (which computes $R^{\lambda}{}_{\mu\nu\kappa} = 0$ explicitly), `aq-sr-rindler-horizon`, `aq-sr-born-rigidity-s6`. Horizons, redshift, clock gradients, rigid bodies — all flat.
What needs GR is TIDES. A uniform field is exactly removable by falling; a real field has a gradient, and gradients are not removable. The tidal tensor $\boxed{E_{ij} = \partial_i\partial_j\Phi}$ is the obstruction, and its covariant form is the Riemann tensor: $\boxed{R^{i}{}_{0j0} = \dfrac{1}{c^2}\,\partial_i\partial_j\Phi}$ .
Owned result: the tidal tensor is TRACELESS in vacuum. $\nabla^2\Phi = 0 \Rightarrow \operatorname{tr}E = 0$ , so its eigenvalues at radius $r$ are $\left(-\dfrac{2GM}{r^3},\; +\dfrac{GM}{r^3},\; +\dfrac{GM}{r^3}\right)$ : stretch vertically, squeeze twice horizontally, conserve volume. Free fall removes gravity's monopole and leaves a volume-preserving distortion behind.
Owned result: the equivalence principle has an expiry time. $\boxed{L\,T^2 < \delta\,r^3/GM}$ . At Earth's surface with $\delta = 1\ \mu$ m: a 1 m lab is inertial for 0.81 s, a 1 cm lab for 8.1 s. That is what "local" means.
Owned result: the numbers. At Earth's surface $2GM/r^3 = \mathbf{3.083\times10^{-6}\ s^{-2}}$ . A $2$ m vertical box: relative tidal acceleration $\mathbf{6.17\times10^{-6}\ m/s^2}$ — $6.29\times10^{-7}$ of $g$ — drifting $\mathbf{3.08\ \mu m}$ after one second of free fall. A human-sized falling lab is an inertial frame to six parts in ten million, and no better.
Owned result: why MICROSCOPE's test masses are concentric. To beat its own tidal gradient by brute force, a lab at Earth's surface testing the WEP to $3\times10^{-15}$ would have to be 9.5 nanometres tall. Impossible — so the experiment nulls the gradient instead, making the two test masses concentric cylinders sharing a centre of mass. The tidal tensor dictated the hardware.
Acceleration does not affect clock rates — only velocity does. The CERN muon ring (Bailey et al., 1977) held muons at a transverse proper acceleration of $\sim\!10^{18}\,g$ and their dilated lifetime followed $\gamma$ exactly: the clock postulate, tested — which is why SR needs no repair to handle acceleration.
Key idea: the boundary between SR and GR is not acceleration. It is curvature.

The core insight

Einstein's happiest thought was that a falling man feels no weight. The precise version: a uniform gravitational field is not a field at all — it is a choice of coordinates, and you undo it by letting go. But no real mass makes a uniform field. Real fields converge on their source, and the convergence survives every coordinate change you can throw at it. So the equivalence principle is exactly true for the part of gravity that is fake and says nothing about the part that is real. The real part is the tidal tensor, its covariant name is Riemann, and general relativity is the theory of what special relativity cannot transform away.

Where this lives today

Every GPS receiver is running this article. A satellite clock at $r \approx 2.66\times10^7$ m gains about $+45.7\ \mu$ s/day from the potential difference and loses $-7.2\ \mu$ s/day from its orbital speed: net $+38.5\ \mu$ s/day. (The commonly quoted $+45.9$ and $+38.6$ use the rotating geoid rather than a spherical $-GM/R$ ; the difference is a few tenths of a microsecond.) Uncorrected, that is a positioning error of about $11$ km per day — and the gravitational term is six times the special-relativistic one, with the opposite sign. A system that assumed "you only need SR" would fail catastrophically, in the wrong direction.

Microgravity is tidal residue, and it is quoted as such. The ISS is in free fall, so it is weightless — but it is $109$ m across, and the tidal tensor does not care that the station is comfortable. Its advertised residual acceleration, of order $10^{-6}\,g$ at the extremities, is $2GMr^{-3}L$ and nothing else. Drop towers trade $T$ against $L$ on exactly the curve above. An experiment that specifies "microgravity" is specifying its place on the equivalence principle's expiry curve.

And the frontier is still the same equality. Torsion balances, lunar laser ranging, atom interferometers dropping two isotopes side by side, MICROSCOPE in orbit — all ask whether $m_{\mathrm{i}} = m_{\mathrm{g}}$ exactly. Nearly every proposed extension of the Standard Model — new scalars, dilatons, fifth forces — predicts a composition-dependent violation at some level. The equivalence principle is not a museum piece: it is the most sensitive null experiment in physics, and what it tests is the assumption general relativity was built on.

The physical picture

Einstein, 1907: "If a person falls freely, he will not feel his own weight." Drop the laboratory and everything inside it is weightless, and every experiment performed inside returns the answer it would return in deep space. Free fall is not "cancelling gravity with acceleration". Free fall is inertial motion, and the ground is the thing that is accelerating. Run it the other way for the same statement. Stand in a windowless box far from all matter and fire the engine at $9.8\ \mathrm{m/s^2}$ : you have weight, dropped objects fall, light crossing the box bends. No local experiment distinguishes it from a room on Earth. That is the equivalence principle, and every one of those effects is pure special relativity in a Rindler chart — which is why the previous four articles did all of it with $R^{\lambda}{}_{\mu\nu\kappa} = 0$ .

Now make the box bigger. Release two balls side by side and fall. On Earth both fall toward the centre, so their paths converge and they drift together. Release two balls one above the other and they drift apart, the lower one being in a stronger field. Nothing you can do to the box — no boost, no rotation, no acceleration, no coordinate change of any kind — makes those drifts go away. They are the fingerprints of a real mass, and they are what an accelerating rocket cannot fake.

Figure

THE BOUNDARY IS NOT ACCELERATION — IT IS CURVATURE. Left: a small freely-falling box. The balls hang motionless, the experimenter is weightless, and every SR experiment works. The equivalence principle is EXACT here, and everything about it — redshift, clock gradients, horizons — is done in FLAT spacetime by the four preceding articles, with the Riemann tensor computed and found to be exactly zero. Right: the SAME box, made bigger. Now the field's convergence shows: the vertical pair pulls APART (the lower ball is in a stronger field), the horizontal pair squeezes TOGETHER (both fall toward the same centre). Volume is conserved — the tidal tensor is traceless in vacuum — but the SHAPE is not, and no coordinate change in existence removes it. That residue is the Riemann tensor. It is the one thing special relativity cannot transform away, and it is where general relativity begins

The right-hand panel is the whole of general relativity in one picture. Notice what it is not: it is not "the box is accelerating". Both boxes are freely falling; neither feels a thing. The difference is entirely that the field on the right has a gradient, and gradients are not coordinate artefacts. Notice too that the ball of dust does not change volume — one stretch of $-2GM/r^3$ , two squeezes of $+GM/r^3$ , sum zero. That is Laplace's equation, and it is why gravity in vacuum is shear without compression.

The core equations

\boxed{\;m_{\mathrm{i}} = m_{\mathrm{g}}\;}\qquad\Longrightarrow\qquad \mathbf{a} = \frac{m_{\mathrm{g}}}{m_{\mathrm{i}}}\,\mathbf{g} = \mathbf{g}\quad\text{for every body}

where: $m_{\mathrm{i}}$ is inertial mass (resistance to force), $m_{\mathrm{g}}$ gravitational mass (coupling to the field). Physical reading: logically independent quantities that experiment finds equal — and that equality is the entire empirical content of the weak equivalence principle. It is why free fall can be a property of spacetime rather than of the falling object, an assumption no other force permits.

E_{ij} \;\equiv\; \partial_i\partial_j\Phi \;=\; \frac{GM}{r^3}\left(\delta_{ij} - 3n_in_j\right), \qquad n_i = x_i/r

where: $\Phi = -GM/r$ is the Newtonian potential, $\mathbf{n}$ the radial unit vector. Physical reading: the tidal tensor. Eigenvalues $(-2GM/r^3,\ +GM/r^3,\ +GM/r^3)$ — radial stretch, transverse squeeze, twice. Its trace is $\nabla^2\Phi = 0$ in vacuum, so it is traceless: free fall preserves a dust ball's volume and distorts its shape.

\boxed{\;R^{i}{}_{0j0} \;=\; \frac{1}{c^{2}}\,\partial_i\partial_j\Phi\;}

where: $R^{\lambda}{}_{\mu\nu\kappa}$ is the Riemann tensor; $i,j$ run over space. Physical reading: the relativistic tidal tensor, and the obstruction. Free indices $i$ and $j$ on both sides — the base's card, which prints $R^{i}{}_{0j0} \sim \partial_j\partial_k\Phi$ , is not a tensor equation at all. Being a tensor, $R = 0$ in one frame means $R = 0$ in every frame: a curved spacetime cannot be flattened by any coordinate change, ever. That, and nothing else, separates GR from SR.

\frac{\mathrm{D}^2\xi^i}{\mathrm{d}\tau^2} \;=\; -\,c^2\,R^{i}{}_{0j0}\,\xi^j \;\longrightarrow\; \frac{\mathrm{d}^2\xi^i}{\mathrm{d}t^2} \;=\; -\,\partial_i\partial_j\Phi\;\xi^j

where: $\xi^i$ is the separation between two neighbouring freely-falling particles. Physical reading: geodesic deviation — the equation that turns curvature into an observable. The left side is GR, the right its Newtonian limit, and their agreement fixes the coefficient in the card above.

\boxed{\;L\,T^{2} \;<\; \frac{\delta\,r^{3}}{GM}\;}

where: $L$ is the lab's size, $T$ the experiment's duration, $\delta$ the position resolution, $r$ the distance to the mass. Physical reading: the equivalence principle's domain of validity, as an inequality — and what the word "local" has been standing in for all along.

Derivation AThe tidal tensor is traceless — free fall removes gravity's volume, not its shape

The equivalence principle is usually stated and then hedged with the word "local". The goal here is to compute what the hedge is hiding: the exact residue left after free fall has removed everything it can, and to show that this residue has a specific algebraic structure — trace-free — which is why the principle's failure looks like a squeeze and never like a push.

Step 1. Fall, and expand. Let the lab's origin fall freely at $\mathbf{r}_0$ . A particle at $\boldsymbol{\xi}$ relative to it has, in the lab frame, the acceleration

\ddot\xi_i = -\partial_i\Phi(\mathbf{r}_0 + \boldsymbol{\xi}) + \partial_i\Phi(\mathbf{r}_0) = -\,\partial_i\partial_j\Phi\big|_{\mathbf{r}_0}\,\xi^j + O(\xi^2).

Why: the uniform part $\partial_i\Phi(\mathbf{r}_0)$ cancels exactly. That cancellation is the equivalence principle, and it is why the leading term is a gradient. Gravity's monopole is a coordinate artefact; its gradient is not.

Step 2. Compute the gradient for a point mass, $\Phi = -GM/r$ :

E_{ij} \;=\; \partial_i\partial_j\!\left(-\frac{GM}{r}\right) \;=\; \frac{GM}{r^3}\bigl(\delta_{ij} - 3n_in_j\bigr), \qquad n_i = \frac{x_i}{r}.

Why: symmetric, so three real eigenvalues and three orthogonal axes. The tides have preferred directions: radial and transverse.

Step 3. Diagonalise. Along the radial direction $\mathbf{n}$ : $E\mathbf{n} = (GM/r^3)(1 - 3)\mathbf{n} = -\,(2GM/r^3)\,\mathbf{n}$ . Perpendicular to it, $n_in_j$ contributes nothing, so the eigenvalue is $+GM/r^3$ , twice. The spectrum is

\left(-\frac{2GM}{r^3},\;\; +\frac{GM}{r^3},\;\; +\frac{GM}{r^3}\right).

Why: the vertical pair pulls apart; the horizontal pair squeezes together at exactly half the rate, in both transverse directions. The factor of two is not decoration — it is why there are two tides a day and not one.

Step 4. Take the trace.

\operatorname{tr}E = \nabla^2\Phi = 4\pi G\rho \;=\; 0 \quad\text{in vacuum.}

$-2 + 1 + 1 = 0$ . ✓ Why: the tidal tensor is traceless wherever there is no matter. A freely-falling ball of dust changes shape but not volume — gravity in empty space is pure shear. That is the classical shadow of the vacuum Einstein equation, which sets the Ricci tensor (Riemann's trace) to zero and leaves the Weyl tensor (its trace-free part) free to propagate. Tides are Weyl curvature, seen by a Newtonian.

Step 5. Promote to a tensor. Matching geodesic deviation to Step 1 in the Newtonian limit gives

R^{i}{}_{0j0} = \frac{1}{c^2}\,\partial_i\partial_j\Phi \;\neq\; 0 .

Why: the point of no return. $E_{ij}$ is a Newtonian object one might hope to argue away. $R^{\lambda}{}_{\mu\nu\kappa}$ is a tensor: non-zero in one frame means non-zero in all. No coordinate system, no acceleration, no boost, no clever chart makes a real gravitational field absent. That is the end of special relativity's jurisdiction, in one line. *Verified in verify_aq-sr-equivalence-principle.py: sympy differentiates $\Phi = -GM/r$ twice, confirms $E_{ij} = (GM/r^3)(\delta_{ij} - 3n_in_j)$ , extracts the eigenvalues $(-2GM/r^3, +GM/r^3, +GM/r^3)$ , and confirms $\operatorname{tr}E = \nabla^2\Phi = 0$ identically away from the origin.*

Go deeper · Derivation B · 2 minThe equivalence principle has an expiry time

Derivation BThe equivalence principle has an expiry time

Everyone says the principle is "local". The goal here is to replace the word with an inequality: how big may a laboratory be, and for how long may it fall, before its own tidal drift becomes visible at a stated precision? The answer is a single closed formula, and the numbers it returns are much less comfortable than the word "local" suggests.

Step 1. Take the worst case. The largest eigenvalue of $E_{ij}$ is the radial one, $2GM/r^3$ , so two particles separated vertically by $L$ have relative acceleration

a_{\text{rel}} = \frac{2GM}{r^3}\,L .

Why: the largest eigenvalue gives a bound that holds for every orientation of the apparatus. A design rule has to be conservative.

Step 2. Integrate for a time $T$ of free fall, starting from rest relative to one another:

\Delta = \tfrac12 a_{\text{rel}} T^2 = \frac{GM}{r^3}\,L\,T^2 .

Why: the drift is quadratic in time. Doubling the fall time quadruples the failure. The principle degrades faster than intuition expects.

Step 3. Demand invisibility. If the apparatus resolves positions to $\delta$ , the equivalence principle survives as an experimental statement precisely while $\Delta < \delta$ :

\boxed{\;L\,T^{2} \;<\; \frac{\delta\, r^{3}}{GM}\;}

Why: three knobs — size, time, precision — and one budget. Make the box smaller, fall for less time, or measure worse. There is no fourth option, and no experiment makes the tides go away.

Step 4. Put Earth in it. $GM/r^3 = 1.5414\times10^{-6}\ \mathrm{s^{-2}}$ at the surface, so $LT^2 < \delta / (1.5414\times10^{-6}\ \mathrm{s^{-2}})$ . At an interferometric $\delta = 1\ \mu\mathrm{m}$ :

L\,T^2 \;<\; 0.649\ \mathrm{m\cdot s^2} \qquad\Longrightarrow\qquad \begin{cases} L = 1\ \mathrm{m}: & T < \mathbf{0.81\ s} \\[2pt] L = 1\ \mathrm{cm}: & T < \mathbf{8.1\ s} \end{cases}

Why: a metre-scale laboratory in free fall at Earth's surface is an inertial frame for eight-tenths of a second. That is the honest content of "local". Not a philosophical caveat — a stopwatch.

Step 5. Read the rule backwards. Everything on the right is fixed by where you are, so the only freedom is to trade $L$ against $T^2$ . Drop towers and orbital microgravity facilities live on this curve, and their advertised "residual $g$ " is $2GM r^{-3}L$ and nothing else. Why: the equivalence principle used as engineering rather than philosophy — the form in which it is actually applied. *Verified in verify_aq-sr-equivalence-principle.py: the bound is re-derived symbolically from the eigenvalue, and $GM/r^3 = 1.5414\times10^{-6}$ s⁻², $LT^2 < 0.649$ m·s², $T < 0.8055$ s at $L = 1$ m and $T < 8.055$ s at $L = 1$ cm are recomputed at fifty digits and grepped out of this file.*

Go deeper · Derivation C · 2 minWhat SR can do, what it cannot, and where exactly the line falls

Derivation CWhat SR can do, what it cannot, and where exactly the line falls

The purpose of this derivation is to kill a piece of folklore with evidence. "Special relativity cannot handle accelerated frames — that is what general relativity is for" is repeated in lecture halls constantly, and it is false. The goal is to say precisely what SR handles (all of acceleration) and precisely what it cannot (any non-zero Riemann tensor), and so to locate the boundary between the two theories at curvature rather than at acceleration.

Step 1. Inventory what SR already did in this chapter — flat, exact, no approximation:

phenomenon	article	flat?
accelerated worldlines, arbitrary $\alpha$	`aq-sr-relativistic-uniform-acceleration`	yes
a chart for an accelerated observer	`aq-sr-rindler-coordinates`	yes
position-dependent clock rates, gravitational-type redshift	`aq-sr-rindler-metric`	yes — $R^{\lambda}{}_{\mu\nu\kappa} = 0$ computed
a causal horizon, an observer frozen and reddened on it	`aq-sr-rindler-horizon`	yes
a rigid body whose ends disagree about time	`aq-sr-born-rigidity-s6`	yes

Why: every effect the folklore attributes to general relativity is in this list, and every entry is flat. The Rindler metric has non-zero Christoffel symbols, a non-trivial lapse and a horizon — and zero curvature. Non-zero Christoffels do not mean curvature. They mean you chose curvilinear coordinates.

Step 2. Add the clock postulate, and test it. SR asserts — it does not derive — that an ideal clock's rate depends on its instantaneous velocity and not at all on its acceleration. Bailey et al. (1977) stored muons in the CERN ring at $\gamma \approx 29.3$ , where the transverse proper acceleration is

\alpha = \frac{\gamma^2 v^2}{R} \approx 10^{19}\ \mathrm{m/s^2} \;\approx\; \mathbf{10^{18}\,g},

and the muons' dilated lifetime tracked $\gamma$ to $2\times10^{-3}$ (95% CL), with no acceleration-dependent term. Why: $10^{18}\,g$ , and the clock does not notice. If acceleration broke special relativity, this is where it would have broken. It did not — so SR needs no repair to handle acceleration, and the four preceding articles are not approximations but exact.

Step 3. Now find something SR genuinely cannot do. $R^{\lambda}{}_{\mu\nu\kappa}$ is a tensor. Special relativity's spacetime is Minkowski, whose Riemann tensor is identically zero, and a tensor that vanishes in one frame vanishes in all of them. Therefore:

\text{SR}\;\Longrightarrow\; R^{\lambda}{}_{\mu\nu\kappa} = 0 \;\Longrightarrow\; \partial_i\partial_j\Phi = 0 \;\Longrightarrow\; \textbf{no tides, ever.}

Why: a theorem, not a limitation of technique. No clever chart, no exotic congruence, no amount of care produces geodesic deviation in flat spacetime. If you have measured a tide, you have measured curvature and left special relativity behind for good.

Step 4. Locate the boundary.

\textbf{Not acceleration.}\qquad \textbf{Curvature.}

Uniform field $\Leftrightarrow$ Rindler chart $\Leftrightarrow$ flat $\Leftrightarrow$ SR. A field with a gradient $\Leftrightarrow$ $R \neq 0$ $\Leftrightarrow$ GR, necessarily. Why: which also explains why the principle had to be local from the very beginning. It equates gravity with acceleration; acceleration lives in flat spacetime; flat spacetime has no tides — so the equivalence can hold only in a region small enough that the tides are invisible. "Local" is not a caveat bolted on afterwards. It is the exact size of the region in which $R$ can be neglected, and the previous derivation measures it. *Verified in verify_aq-sr-equivalence-principle.py: the Riemann tensor of the Rindler chart is recomputed from the metric (all components zero, confirming the flatness claim independently of `aq-sr-rindler-metric`); the CERN transverse proper acceleration $\gamma^2v^2/R$ is recomputed at $\gamma = 29.3$ , $R = 7$ m and returned in units of $g$ ; and the tidal tensor of $\Phi = -GM/r$ is confirmed non-zero, establishing the two sides of the boundary.*

Figure

THE EQUIVALENCE PRINCIPLE HAS AN EXPIRY TIME. "Local" is not a philosophical hedge — it is a stopwatch, and this is its face. The tidal drift between two points of a freely-falling lab of size L, after a fall of duration T, is (GM/r³)·L·T². Demand that it stay below your position resolution δ and you get one closed inequality: L·T² < δ·r³/GM. Everything on the right is fixed by where you are; the only freedom an experimenter has is to trade SIZE against TIME-SQUARED. The curve is the boundary at Earth's surface with a micron of resolution. BELOW it, special relativity is exact and you are in a local inertial frame. ABOVE it, the tides are visible, the Riemann tensor is doing observable work, and you have left SR behind. A one-metre laboratory is an inertial frame for eight-tenths of a second. A one-centimetre laboratory gets eight seconds. That is what "local" means

Limiting cases

Case	Result	Meaning
$L \to 0$ , $T \to 0$	$R\,L\,T^2 \to 0$	the principle becomes exact — the definition of a local inertial frame, and why GR reduces to SR pointwise
uniform field, any $L$ , any $T$	equivalence exact	a genuinely uniform field is a Rindler chart. Flat. SR handles it completely — this chapter is the proof
real field, $L\,T^2 \geq \delta r^3/GM$	equivalence fails	tidal drift becomes visible; nothing done to the frame prevents it
vacuum	$\operatorname{tr}E_{ij} = 0$	volume-preserving. Gravity outside matter is pure shear: stretch once, squeeze twice
inside matter	$\operatorname{tr}E_{ij} = 4\pi G\rho > 0$	the dust ball does shrink — the Ricci part, which the Einstein equation constrains
$\alpha \to 10^{18}\,g$	clock rate still $= \gamma$	the clock postulate, measured (Bailey 1977). Acceleration does not affect clocks. Only velocity does.

Common trapThe equivalence principle is not "gravity = acceleration". It is "gravity's MONOPOLE = acceleration"

The slogan drops the most important word. What free fall removes is the uniform part of the field — the first derivative of the potential, at one point. What it leaves behind is every higher derivative, starting with $\partial_i\partial_j\Phi$ , and no choice of frame touches those. The correct statement is not "a gravitational field is indistinguishable from an acceleration" but "the value of a gravitational field at a point is indistinguishable from an acceleration, and its gradient is indistinguishable from nothing at all." Two consequences, both routinely mangled. First, the principle cannot be used to argue that gravity is fictitious — the trace-free part is an invariant, as real as anything in physics. Second, it cannot be extended to a finite region by taking more care: the tidal residue is not an error term you beat down with technique, it is a tensor, and it is still there when you have finished being careful. The principle is exactly true at a point and approximately true near one, and "near" means $L\,T^2 < \delta\,r^3/GM$ .

Common trapWhat this article does not own — and three defects in its own base

The Rindler chart, the clock-rate law and the flatness proof $R^{\lambda}{}_{\mu\nu\kappa} = 0$ are owned by `aq-sr-rindler-metric`; so is the Pound–Rebka analysis (predicted $gh/c^2 = 2.46\times10^{-15}$ ; the published two-way figures $-4.92\times10^{-15}$ predicted against $-(5.13 \pm 0.51)\times10^{-15}$ measured, ratio $1.05 \pm 0.10$ ). Both are recalled here, never re-derived. The horizon belongs to `aq-sr-rindler-horizon`, the rigid congruence to `aq-sr-born-rigidity-s6`. The Unruh effect (`aq-sr-unruh-effect`) is gestured at and nothing more: it is not canon here, and no claim whatever is made about its contents.

Base defects repaired. (1) The base's tidal card prints $R^{i}{}_{0j0} \sim \partial_j\partial_k\Phi$ — $k$ is free on the right and absent on the left, $i$ free on the left and absent on the right. It is not a tensor equation. The correct relation, derived above, is $R^{i}{}_{0j0} = (1/c^2)\,\partial_i\partial_j\Phi$ . (2) The base's Rindler metric card is mostly-plus, against this chapter's $(+,-,-,-)$ lock. (3) The base's Pound–Rebka card says the experiment agreed "to about 1% (later refined to 1% by Pound and Snider)" — self-contradictory, and wrong twice: Pound and Rebka (1960) reached about 10% (published ratio $1.05 \pm 0.10$ ); Pound and Snider (1965) reached about 1%.

Worked examples

1 · How good an inertial frame is a falling lift?

Step 1 — the tidal scale at Earth's surface. With $GM = 3.9860\times10^{14}\ \mathrm{m^3/s^2}$ and $r = 6.371\times10^6$ m,

\frac{2GM}{r^3} = \mathbf{3.083\times10^{-6}\ s^{-2}} .

Step 2 — a two-metre box. Two balls, one at the ceiling and one at the floor, $L = 2$ m apart vertically. Their relative acceleration is

a_{\text{rel}} = \frac{2GM}{r^3}\,L = \mathbf{6.17\times10^{-6}\ m/s^2} \;=\; \mathbf{6.29\times10^{-7}\,g} .

Step 3 — the drift. After one second of free fall: $\Delta = \tfrac12 a_{\text{rel}} (1\ \mathrm{s})^2 = \mathbf{3.08\ \mu m}$ .

Step 4 — the verdict. A human-sized freely-falling laboratory is an inertial frame to about six parts in ten million, and its ends drift apart by three microns a second. If your apparatus cannot see three microns, you are in special relativity. If it can, you are in general relativity — and you were the whole time.

Step 5 — the boundary number. $6.29\times10^{-7}$ is some $10^{9}$ times the frequency stability of a good optical clock. The equivalence principle is an excellent approximation for a person and a terrible one for an instrument.

2 · Why MICROSCOPE's test masses are concentric cylinders

Step 1 — the target. MICROSCOPE (launched 2016, final results 2022) tests the universality of free fall by comparing the accelerations of a titanium and a platinum test mass, constraining the Eötvös parameter to $|\eta| \lesssim 3\times10^{-15}$ .

Step 2 — the obstacle. Two test masses whose centres are separated by $L$ feel a differential acceleration from the tidal tensor alone — no equivalence-principle violation required. To hide that beneath $3\times10^{-15}$ of the local $g$ by making the apparatus small, you would need

L \;<\; \frac{3\times10^{-15}\,g}{2GM/r^3} \;=\; \frac{3\times10^{-15} \times 9.80665}{3.083\times10^{-6}} \;=\; \mathbf{9.5\ nanometres} .

(For a $10^{-15}$ target: $3.2$ nm.)

Step 3 — the impossibility. You cannot build a torsion balance from two masses whose centres are nine nanometres apart. Brute force is not available.

Step 4 — the solution, which the tidal tensor dictated. Do not make $L$ small. Make it zero. MICROSCOPE's test masses are concentric hollow cylinders sharing a common centre of mass, so $L = 0$ to the precision of the machining and the leading tidal term — linear in $L$ — cancels identically. What survives is a far smaller residual from the difference in the masses' second moments, which is calibrated and subtracted.

Step 5 — the boundary number. The most precise test of $m_{\mathrm{i}} = m_{\mathrm{g}}$ ever performed has a geometry chosen by the tidal tensor — concentric, because $E_{ij}\xi^j$ is linear in $\xi$ and the only way to beat a linear term you cannot outrun is to set its argument to zero. The tides do not merely limit the equivalence principle. They design the experiments that test it.

Origins · 2 minEquivalence principle and how this was discovered

Equivalence principle

1592–2022 · Galileo drops them together · Eötvös weighs the difference · Einstein has his happiest thought · MICROSCOPE closes to 3 parts in 10¹⁵

The observation. That all bodies fall alike is the oldest surviving anomaly in physics. Galileo argued it around 1590 by pure logic — tie a light body to a heavy one and ask whether the composite falls faster (it is heavier) or slower (the light one drags); the contradiction settles it without dropping anything. Newton wrote it into the Principia as the exact proportionality of weight to mass, tested with pendulums to one part in $10^3$ . The measurement. Loránd Eötvös spent decades on the torsion balance and by 1909 had $m_{\mathrm i} = m_{\mathrm g}$ to $5\times10^{-9}$ — stupefying for the era, and the number Einstein leaned on. Modern torsion balances (Eöt-Wash) reach $10^{-13}$ ; MICROSCOPE, in orbit, reaches $|\eta| \lesssim 3\times10^{-15}$ . In four centuries the bound has improved by twelve orders of magnitude and the answer has never changed. The idea. Einstein's step, in 1907, was to stop treating $m_{\mathrm{i}} = m_{\mathrm{g}}$ as a coincidence and start treating it as a principle. If every body falls the same way, falling is not a property of bodies — it is a property of spacetime, and the trajectories are geometry. Nothing else in physics permits this move: two charges of different $q/m$ do not follow the same path in an electric field, so there is no "equivalence principle" for electromagnetism and no geometric theory of it. Gravity is geometrisable because, and only because, Eötvös kept getting zero.

History

1592

— Galileo argues from a thought experiment that bodies of different weight must fall alike; the Pisa drop is probably apocryphal, the argument is not.

1687

— Newton's Principia makes weight exactly proportional to mass, tested with pendulums to

\sim10^{-3}

1889–1909

— Eötvös's torsion-balance campaign establishes

m_{\mathrm i} = m_{\mathrm g}

5\times10^{-9}

1907

— Einstein, Jahrbuch der Radioaktivität und Elektronik 4, 411, elevates the equality to a principle and derives gravitational time dilation and redshift from it. (The "happiest thought" phrasing comes from a 1920 recollection, not this paper; light bending arrives in 1911.)

1915

— General relativity: free-falling frames become local Lorentz frames, and gravity becomes curvature.

1960

— Pound and Rebka measure the redshift in a 22.5 m tower, agreeing to about 10% (`aq-sr-rindler-metric` owns the analysis); Pound and Snider reach 1% in 1965.

1977

— Bailey et al. store muons at

\sim10^{18}\,g

transverse proper acceleration at CERN; the dilated lifetime follows

\gamma

with no acceleration term, confirming the clock postulate.

2016–2022

— MICROSCOPE flies and reports

|\eta| \lesssim 3\times10^{-15}

: still zero, to fifteen digits.

Going further

Texts. Einstein, Jahrbuch der Radioaktivität und Elektronik 4, 411 (1907) · Misner, Thorne & Wheeler, Gravitation, §§1.6, 16 · Wald, General Relativity, §4.3 · Rindler, Relativity: Special, General and Cosmological, 2nd ed., ch. 1 · Will, Theory and Experiment in Gravitational Physics, 2nd ed. · Bailey et al., Nature 268, 301 (1977) · Touboul et al., Phys. Rev. Lett. 129, 121102 (2022), the MICROSCOPE final result.

Provenance. Signature $(+,-,-,-)$ . Every equation and number is recomputed in verify_aq-sr-equivalence-principle.py, which reads this file, regexes the printed values out, and asserts each against what it computed at fifty digits. Base defects repaired: the tidal card's free-index mismatch; the mostly-plus Rindler metric card; the self-contradictory Pound–Rebka precision claim ("agreed to 1%… later refined to 1%"), corrected to $\sim$ 10% for Pound–Rebka 1960 and $\sim$ 1% for Pound–Snider 1965; and the MICROSCOPE card's date and bound, corrected to launched 2016, final results 2022, $|\eta| \lesssim 3\times10^{-15}$ .

Scope. The Rindler chart, the exact clock-rate law and the Pound–Rebka analysis are owned by `aq-sr-rindler-metric`, recalled and never re-derived. The Unruh effect is gestured at only.

So is it true or false that you need general relativity for accelerated frames?

False, and worth being blunt about. Every phenomenon the folklore attributes to GR — position-dependent clock rates, redshift between the ends of an accelerating ship, a causal horizon, a body whose ends cannot stay synchronised — is derived in this chapter in flat spacetime, and `aq-sr-rindler-metric` computes the Riemann tensor of the accelerated chart explicitly and gets zero, every component. What confuses people is that the Rindler metric has non-constant coefficients and non-zero Christoffel symbols, which look like general relativity. They are not: they are curvilinear coordinates, and polar coordinates on flat paper have non-zero Christoffels too. Curvature is $R^{\lambda}{}_{\mu\nu\kappa}$ , not $\Gamma^{\lambda}{}_{\mu\nu}$ , and only one of those is a tensor.

If the equivalence principle is only local, how can it be the foundation of GR?

Because "local" is exactly the right amount. GR's content is that spacetime is a manifold with a metric, and the defining property of a manifold is that it looks flat if you zoom in far enough. The equivalence principle is the physical statement of that mathematical fact: at every event there exists a frame — the freely-falling one — in which the laws of special relativity hold exactly, at that event. GR is then built by demanding the laws of physics be written so this is automatically true everywhere. The principle is not a global claim that failed and got downgraded. It was always a statement about a point, and the theory built on it is the theory of how to glue those points together — and the glue is curvature.

Why is the tidal tensor traceless, and why should I care?

Traceless because $\operatorname{tr}(\partial_i\partial_j\Phi) = \nabla^2\Phi = 4\pi G\rho$ , zero wherever there is no matter — Laplace's equation, straight out of Newton. Care because it means gravity outside matter is pure shear: it distorts a ball of dust without changing its volume, one stretch balanced by two squeezes. That is the classical shadow of a deep fact in GR, where the Riemann tensor splits into a trace part (Ricci), tied by the Einstein equation to the local matter, and a trace-free part (Weyl), free to propagate through vacuum. Tides in empty space are Weyl curvature. So are gravitational waves. When you feel the tide, you are feeling the part of the field that can travel.

Can I get around the tidal limit by making the box freely falling AND rotating, or by some cleverer frame?

No, and the reason is a theorem rather than a difficulty. $R^{\lambda}{}_{\mu\nu\kappa}$ is a tensor: non-zero in one coordinate system means non-zero in every one, because tensors transform linearly and a linear map cannot send a non-zero object to zero. Boosting, rotating, accelerating, exotic non-orthogonal charts — all change the components of Riemann; none can change whether it vanishes. Geodesic deviation is therefore an invariant, observable fact about a region of spacetime, and there is no frame in which two freely-falling particles near a real mass fail to drift. Your only knob is the size of the region, which is what $LT^2 < \delta r^3/GM$ quantifies.

Does light bend because of the equivalence principle, or because of curvature?

Both, and the split is instructive. In an accelerating box a horizontally-crossing light beam does bend — pure SR, visible in the Rindler chart, and it is what Einstein used in 1911 to predict a deflection of $0.87''$ for starlight grazing the Sun. That is half the right answer. The other half comes from the spatial curvature of the Schwarzschild metric, which the equivalence principle knows nothing about, being a statement about a single point. Einstein got the full $1.75''$ only in 1915, with the field equations. The factor of two is the cleanest demonstration there is that the equivalence principle is the beginning of general relativity and not the whole of it.

Where does the Unruh effect fit in?

It is the quantum sequel to this chapter, and this article deliberately does not derive it. The short version, claimed no further: the accelerated observer of `aq-sr-rindler-coordinates` and `aq-sr-rindler-horizon` has a horizon, and quantum field theory says an observer with a horizon does not agree with an inertial one about what counts as the vacuum. `aq-sr-unruh-effect` is where that is done. It is not part of this chapter's canon and nothing here depends on it; it is noted only because the flat-spacetime accelerated frames built in these five articles are the arena in which it is computed, and because the same logic — horizon, thermality, information — reappears for black holes, where the spacetime is genuinely curved and none of this chapter's flatness survives.