The Failure of Global Inertial Frames and the Road to Curvature

Special relativity is a theory about inertial frames. Every postulate names one; every result compares two of them. Take the global inertial frame away and there is nothing left to say. Gravity takes it away. But the interesting question is not whether — `aq-sr-equivalence-principle` and `aq-sr-tidal-forces-equivalence-limits` settled that. It is how much. Free fall buys back an enormous amount: at any single event you can make the metric exactly Minkowski and its first derivatives exactly zero, and special relativity is then exactly true at that point. So what is left over?

Twenty numbers. Not "some residual curvature". Not "higher-order corrections". Twenty — and you can count them with nothing but the chain rule:

\underbrace{16 - 10 = 6}_{\text{the Lorentz group}} \qquad \underbrace{40 - 40 = 0}_{\text{free fall works — exactly}} \qquad \underbrace{100 - 80 = \mathbf{20}}_{\text{and here it stops}}

Those twenty survivors are the Riemann curvature tensor, and the theorem that closes this chapter is an iff: a global inertial frame exists if and only if all twenty vanish.

Key takeaways

Signature $(+,-,-,-)$ : $\eta_{\mu\nu} = \operatorname{diag}(+1,-1,-1,-1)$ , timelike positive. The base prints $\operatorname{diag}(-1,+1,+1,+1)$ and a mostly-plus Rindler metric; both are re-derived here.
The theorem. $\boxed{\text{a global inertial frame exists} \iff R^{\rho}{}_{\sigma\mu\nu} = 0}$ — an iff. Curvature is not evidence that the frame is gone; it is the obstruction, and its vanishing is sufficient.
Owned derivation — THE COUNTING. Expand a coordinate change to third order and count parameters against conditions. $16 - 10 = 6$ (surplus: the Lorentz group). $40 - 40 = 0$ (exactly enough: free fall works, nothing to spare). $100 - 80 = 20$ (deficit: irremovable). The equivalence principle is a counting coincidence, and it runs out at third order.
The residue IS the Riemann tensor — and a second, independent count agrees: Riemann's symmetries cut its $4^4 = 256$ components to $21$ , the first Bianchi identity kills one more, $\mathbf{21 - 1 = 20}$ . Two unrelated roads, one number.
Owned: the ⅓, and a base defect nobody caught. In Riemann normal coordinates $\boxed{g_{\mu\nu}(x) = \eta_{\mu\nu} - \tfrac{1}{3}R_{\mu\alpha\nu\beta}\,x^{\alpha}x^{\beta} + O(x^3)}$ . The coefficient is $1/3$ ; the base prints $1/2$ — confusing the Taylor coefficient $\tfrac12\partial\partial g$ with the curvature coefficient. Machine-checked on the sphere.
Owned: what the twenty are made of. $\mathbf{20 = 10 + 10}$ — ten Ricci (fixed by matter) and ten Weyl (the free field: vacuum tides, gravitational waves). In three dimensions it is $6 = 6 + 0$ : Weyl vanishes identically, and with it go vacuum curvature, gravitational waves and every local degree of freedom. Gravity propagates because spacetime is four-dimensional.
⚠ A hard sign error in this article's base. Its eq[2] prints $\Delta a_r \approx -\tfrac{2GM}{r^3}\Delta r$ , $\Delta a_\perp \approx +\tfrac{GM}{r^3}\Delta x_\perp$ and calls them relative accelerations. Both signs are backwards — those are the eigenvalues of the tidal tensor, and the accelerations are $\ddot\xi = -E\xi$ , their negatives. The base contradicts its own derivation card, which correctly says $\ddot\xi_r > 0$ (stretch), $\ddot\xi_\perp < 0$ (squeeze).

The core insight

Everyone knows gravity ruins the global inertial frame. Almost nobody can tell you by how much — and the answer is not a matter of taste or approximation. Write down the most general coordinate change there is, Taylor-expand it, and count what you can cancel against what you must. The books balance at first order with six to spare, balance exactly at second order, and fail at third order by twenty. Those twenty cannot be argued away, transformed away, or fallen away from, in this or any other four-dimensional universe. They are not a sign of curvature. They are curvature. And when they are all zero the global inertial frame comes back — which is why special relativity is not a good approximation to general relativity but an exact special case of it.

Where this lives today

The counting argument is how you find the degrees of freedom of any field theory. Count the components, count the gauge freedom, subtract. Gravity in four dimensions has two propagating polarisations — and that is why a gravitational wave has a plus mode and a cross mode and no others.

The Weyl/Ricci split is the working vocabulary of relativity. Ricci is where the matter is; Weyl is where it isn't. Every vacuum solution — Schwarzschild, Kerr, a passing wave — is pure Weyl. The Petrov classification, the peeling theorem, the definition of an isolated system: all are statements about the ten Weyl components, and none exists in three dimensions.

And three-dimensional gravity is a real research subject because it is trivial. $2+1$ Einstein gravity has no local degrees of freedom, which makes it exactly solvable — so it is where black-hole entropy, holography and quantum gravity get tested before anyone tries them in four dimensions. The BTZ black hole exists not because $2+1$ gravity has vacuum curvature (it does not) but because it can carry a cosmological constant, and topology does the rest.

The physical picture

Stand at an event $P$ and ask what a change of coordinates can do for you. It is a bargaining problem in three rounds.

Round one you win easily. Ten numbers to fix, sixteen knobs. You set $g_{\mu\nu}(P) = \eta_{\mu\nu}$ and have six knobs left over — and those six are not slack. They are exactly the maps that preserve $\eta_{\mu\nu}$ : three rotations, three boosts. The surplus of the first round is the Lorentz group, which is a startling thing to fall out of a subtraction.

Round two you win exactly. Forty first derivatives of the metric; forty second-order coordinate parameters. You kill them all — the Christoffel symbols vanish at $P$ , and you are in free fall — but with nothing whatever to spare. The equivalence principle is not a deep symmetry dressed up in coordinates. It is an accounting coincidence, and coincidences do not extend.

Round three you lose, a hundred to eighty. Twenty numbers survive every coordinate transformation that will ever be written — and twenty is exactly the number of independent components of the Riemann tensor. Not a coincidence but a theorem, and the subject of this article.

Figure

THE COUNTING THEOREM — the whole of this chapter in three subtractions, and every number is obtained with nothing but the chain rule. Take an event P and ask what the most general change of coordinates can do for you, order by order in its Taylor expansion. AT FIRST ORDER: the metric at P has ten independent components (a symmetric 4×4 matrix) and a linear coordinate change has sixteen free parameters (a general 4×4 matrix). Sixteen knobs, ten conditions — you win with SIX TO SPARE, and those six leftover parameters are precisely the transformations that preserve the Minkowski metric: three rotations and three boosts. THE SURPLUS OF THE FIRST COUNT IS THE LORENTZ GROUP. AT SECOND ORDER: the forty first derivatives of the metric must be killed, and the second-order coordinate freedom is a set of coefficients symmetric in two lower indices, which is also forty. FORTY AGAINST FORTY — you win, but EXACTLY, with nothing whatever to spare. That is why free fall works, and the equivalence principle is revealed as an accounting coincidence rather than a deep symmetry. AT THIRD ORDER: there are one hundred second derivatives of the metric and only eighty third-order coordinate parameters. YOU LOSE BY TWENTY. Twenty second derivatives survive every coordinate transformation that will ever be written, and twenty is exactly n²(n²−1)/12 at n = 4, the number of independent components of the Riemann curvature tensor. The residue is not evidence of curvature. It IS curvature

The core equations

\boxed{\;\text{a global inertial frame exists on a region} \quad\Longleftrightarrow\quad R^{\rho}{}_{\sigma\mu\nu} = 0 \ \text{ there}\;}

where: $R^{\rho}{}_{\sigma\mu\nu} = \partial_\mu\Gamma^{\rho}_{\nu\sigma} - \partial_\nu\Gamma^{\rho}_{\mu\sigma} + \Gamma^{\rho}_{\mu\lambda}\Gamma^{\lambda}_{\nu\sigma} - \Gamma^{\rho}_{\nu\lambda}\Gamma^{\lambda}_{\mu\sigma}$ . Physical reading: the theorem this chapter has been building toward, and it is an iff. Necessity is easy; sufficiency is the substantive half — vanishing curvature does not merely permit a global inertial frame, it produces one — and that is why special relativity is an exact special case of general relativity rather than an approximation.

\underbrace{16 - 10 = 6}_{\text{Lorentz}} \qquad \underbrace{40 - 40 = 0}_{\text{free fall}} \qquad \underbrace{100 - 80 = 20}_{\text{Riemann}} \qquad\qquad N = \frac{n^{2}(n^{2}-1)}{12}\ \xrightarrow{\;n=4\;}\ \mathbf{20}

where: coordinate freedom counted against conditions, at first, second and third order. Physical reading: the whole article — a surplus, an exact balance, and a deficit, and the deficit is the Riemann tensor.

\boxed{\;g_{\mu\nu}(x) \;=\; \eta_{\mu\nu} \;-\; \tfrac{1}{3}\,R_{\mu\alpha\nu\beta}\,x^{\alpha}x^{\beta} \;+\; O(x^{3})\;}

where: these are Riemann normal coordinates at $P$ . Physical reading: the sharpest statement of "how flat is a falling lab". The linear term is gone — that is $\Gamma(P) = 0$ , that is free fall — and curvature is the first thing that survives. Note the coefficient: $\mathbf{1/3}$ . The base prints $1/2$ , confusing the Taylor coefficient $\tfrac12\partial_\alpha\partial_\beta g_{\mu\nu}$ with the curvature coefficient; the two differ because $\partial_\alpha\partial_\beta g_{\mu\nu}|_P = -\tfrac13(R_{\mu\alpha\nu\beta} + R_{\mu\beta\nu\alpha})$ . Machine-checked on the sphere.

\boxed{\;\ddot{\xi}^{\,i} = -\,E_{ij}\,\xi^{j}\;} \qquad\qquad \mathbf{20} \;=\; \underbrace{\mathbf{10}}_{\text{Ricci}} \;+\; \underbrace{\mathbf{10}}_{\text{Weyl}} \qquad\qquad\text{but in } n=3: \quad \mathbf{6} = \mathbf{6} + \mathbf{0}

where: $E_{ij} = \partial_i\partial_j\Phi$ is `aq-sr-equivalence-principle`'s tidal tensor, eigenvalues $(-2GM/r^3, +GM/r^3, +GM/r^3)$ ; Ricci is what Einstein's equations tie to the matter, and Weyl is the trace-free remainder that survives in vacuum. Physical reading, twice over. First, the minus sign — and the base does not have it: the tensor's eigenvalues and the accelerations they produce are negatives, so the true values are $+\tfrac{2GM}{r^3}\xi_r$ (stretch) and $-\tfrac{GM}{r^3}\xi_\perp$ (squeeze). Second, why gravity propagates: the ten Weyl components are the free field — vacuum tides, gravitational waves — and in three dimensions there are none. Ricci exhausts Riemann, vacuum is flat, and $2+1$ Einstein gravity has no waves and no local degrees of freedom at all.

Counting what cannot be removed

Derivation AThe counting theorem — how much of gravity survives every coordinate change

The equivalence principle is usually presented as a physical insight, and it is one; but it is also an arithmetic fact, and the arithmetic tells you something the insight cannot: exactly where it stops. This card does the counting properly — the most general coordinate transformation there is, expanded about an event $P$ , with the free parameters counted against the conditions they must satisfy at each order. Three subtractions, and the size of the third is the answer to the chapter.

Step 1. Write the transformation and expand about $P$ :

x'^{\mu} = A^{\mu}{}_{\alpha}x^{\alpha} + \tfrac12 B^{\mu}{}_{\alpha\beta}\,x^{\alpha}x^{\beta} + \tfrac16 C^{\mu}{}_{\alpha\beta\gamma}\,x^{\alpha}x^{\beta}x^{\gamma} + \cdots

$B$ is symmetric in $(\alpha\beta)$ and $C$ in $(\alpha\beta\gamma)$ — because partial derivatives commute, and for no other reason. Why: the symmetry of $B$ and $C$ is the entire source of the deficit. If mixed partials did not commute you would have more knobs and there would be no curvature to speak of. Everything below follows from $\partial_\alpha\partial_\beta = \partial_\beta\partial_\alpha$ .

Step 2. First order: $16 - 10 = 6$ . The matrix $A^{\mu}{}_{\alpha}$ has $\mathbf{16}$ entries; imposing $g'_{\mu\nu}(P) = \eta_{\mu\nu}$ is $\mathbf{10}$ conditions. So $16 - 10 = \mathbf{6}$ parameters are left over. Why: the six are not slack — they are the Lorentz group. Exactly the linear maps that leave $\eta_{\mu\nu}$ alone: three rotations, three boosts, $\dim SO(1,3) = 6$ .

Step 3. Second order: $40 - 40 = 0$ . Killing $\partial_\gamma g'_{\mu\nu}(P)$ is $4 \times 10 = \mathbf{40}$ conditions. The knobs are $B^{\mu}{}_{\alpha\beta}$ : an upper index (4) times a symmetric pair of lower ones ( $\tfrac12\cdot4\cdot5 = 10$ ), so $\mathbf{40}$ .

40 - 40 = \mathbf{0}\ \text{ — exactly enough, with nothing to spare.}

Equivalently $\Gamma^{\lambda}_{\mu\nu}(P) = 0$ can always be arranged. This is free fall, and it is the whole content of the equivalence principle. Why: stare at this. Forty and forty. Not forty-one and forty. The equivalence principle is true because two integers happen to coincide — and an exact coincidence has no margin. It cannot be pushed one order further, and the failure at the next order is what general relativity is.

Step 4. Third order: $100 - 80 = 20$ . Killing $\partial_\gamma\partial_\delta g'_{\mu\nu}(P)$ is a symmetric derivative pair (10) times a symmetric metric (10) $= \mathbf{100}$ conditions. The knobs are $C^{\mu}{}_{\alpha\beta\gamma}$ : an upper index (4) times a totally symmetric triple of lower ones, and symmetric triples from four values number $\binom{4+3-1}{3} = \binom{6}{3} = 20$ . So $4\times 20 = \mathbf{80}$ .

\boxed{\;100 - 80 = \mathbf{20}\;}

Why: you have run out of coordinates. Twenty combinations of second derivatives cannot be set to zero at $P$ by any transformation whatsoever — not a clever one, not a singular one, not one nobody has thought of yet. They are invariant under the whole diffeomorphism group, and such an object is a tensor.

Step 5. Identify the residue. The Riemann tensor has $N = \tfrac{1}{12}n^{2}(n^{2}-1)$ independent components, and at $n=4$ that is $\tfrac{16\cdot15}{12} = \mathbf{20}$ . Why: the same twenty. The irremovable second derivatives of the metric are the Riemann tensor — not "measured by", not "related to". They are it. And the theorem runs both ways: if $R = 0$ the residue vanishes, the third-order transformation does exist, the metric is $\eta$ everywhere, and the global inertial frame comes back. Special relativity is general relativity with twenty numbers set to zero. *Verified in verify_aq-sr-failure-global-inertial-frames.py: all three counts are recomputed from first principles (the multiset coefficients $\binom{n+k-1}{k}$ are built by brute-force enumeration, not asserted); $100-80$ is checked equal to $n^2(n^2-1)/12$ at $n=4$ ; the six leftover parameters are checked equal to $\dim SO(1,3)$ by explicitly counting the generators of $\Lambda^T\eta\Lambda = \eta$ ; and artefact guards vary $n$ and confirm every count moves.*

Go deeper · Derivation B · 2 minThe other road to twenty — and the ⅓ that the base gets wrong

Derivation BThe other road to twenty — and the ⅓ that the base gets wrong

A number obtained one way is a result; the same number obtained two unrelated ways is a fact. The counting card reached twenty by asking what a coordinate change cannot do. This one reaches twenty by asking what a tensor with Riemann's symmetries can be — a question with no coordinates in it at all — and then uses the answer to write the metric of a falling laboratory, where a factor the base gets wrong is waiting.

Step 1. Start from $4^4 = 256$ and impose the symmetries. $R_{\mu\nu\alpha\beta}$ is antisymmetric in $(\mu\nu)$ , antisymmetric in $(\alpha\beta)$ , and symmetric under exchange of the pairs. An antisymmetric pair from four indices has $\binom{4}{2} = 6$ values, so $R$ is a symmetric $6\times6$ matrix on pair-space: $\tfrac{6\cdot 7}{2} = \mathbf{21}$ . Why: $256 \to 21$ with three symmetries and no effort. The pair-index trick is the whole reason curvature is tractable.

Step 2. Impose the first Bianchi identity, $R_{\mu[\nu\alpha\beta]} = 0$ . The totally antisymmetric part of $R_{\mu\nu\alpha\beta}$ has $\binom{4}{4} = \mathbf{1}$ component in four dimensions, and the identity kills it:

\boxed{\;21 - 1 = \mathbf{20}\;}

Why: twenty again — and nothing here knew anything about coordinate transformations. One count asked "what survives a chart change?"; the other, "what can an algebraic object with these symmetries be?". Different questions, same answer — which is how you know the identification is not a coincidence.

Step 3. Build the falling laboratory. Choose Riemann normal coordinates at $P$ : run geodesics out from $P$ and label each point by the tangent vector that reaches it. By construction $g_{\mu\nu}(P) = \eta_{\mu\nu}$ and $\Gamma^\lambda_{\mu\nu}(P) = 0$ — the previous card's second-order count, cashed in. Why: the freely-falling frame, made concrete. Every statement about "a small enough lab" becomes a statement about how many terms of this expansion you keep.

Step 4. Expand, and mind the factor. Taylor gives $g_{\mu\nu}(x) = \eta_{\mu\nu} + \tfrac12\,\partial_\alpha\partial_\beta g_{\mu\nu}\big|_P\,x^{\alpha}x^{\beta} + O(x^3)$ , and the normal-coordinate condition forces $\partial_\alpha\partial_\beta g_{\mu\nu}\big|_P = -\tfrac13(R_{\mu\alpha\nu\beta} + R_{\mu\beta\nu\alpha})$ . Substituting — and $x^{\alpha}x^{\beta}$ is symmetric, so the two terms contribute equally:

g_{\mu\nu}(x) = \eta_{\mu\nu} + \tfrac12\cdot\left(-\tfrac13\right)\cdot 2\,R_{\mu\alpha\nu\beta}\,x^{\alpha}x^{\beta} = \boxed{\;\eta_{\mu\nu} - \tfrac{1}{3}R_{\mu\alpha\nu\beta}\,x^{\alpha}x^{\beta} + O(x^3)\;}

Why: the coefficient is $1/3$ , and the base prints $1/2$ . The error is exactly what the algebra above is built to catch: $\tfrac12$ is the Taylor coefficient and multiplies $\partial\partial g$ , not $R$ . Converting one to the other costs a further $-\tfrac13\times 2 = -\tfrac23$ , and $\tfrac12 \times (-\tfrac23) = -\tfrac13$ . The base stopped one substitution early. The verify script builds normal coordinates on a sphere of radius $a$ and reads the coefficient off: it comes out $-1/(3a^2)$ , and $R_{xyxy} = 1/a^2$ — so $1/3$ , not $1/2$ .

Step 5. Read the physics off. There is no linear term — that is $\Gamma(P) = 0$ , that is the equivalence principle — and the first surviving term is quadratic, with curvature as its coefficient. The fractional deviation from flatness across a lab of size $L$ is $\sim \tfrac13 |R| L^2$ , and that is what "a small enough laboratory" means: small compared with the curvature radius $|R|^{-1/2}$ . Why: it is the quantitative form of `aq-sr-equivalence-principle`'s expiry inequality $LT^2 < \delta r^3/GM$ . The equivalence principle does not fail suddenly. It fails quadratically, and the coefficient of the failure is the Riemann tensor. *Verified in verify_aq-sr-failure-global-inertial-frames.py: the $256 \to 21 \to 20$ count is done by brute-force enumeration over all $4^4$ index quadruples, canonicalising under the symmetries and then computing the rank of the first-Bianchi constraint system — no formula is assumed; and the $1/3$ is extracted from an explicit normal-coordinate expansion of the round sphere, with the base's $1/2$ asserted NOT to reproduce it.*

Go deeper · Derivation C · 2 minWhat the twenty are made of — and why gravity would not propagate in a flatter universe

Derivation CWhat the twenty are made of — and why gravity would not propagate in a flatter universe

Twenty answers "how much survives", but not "what is it". The twenty split, cleanly and invariantly, into two halves of ten — and the split is the difference between the gravity matter causes and the gravity that simply is. This card makes the split, runs the same arithmetic in three dimensions where it collapses, and notices what the collapse costs: gravitational waves, vacuum tides, and black holes.

Step 1. Contract. From $R_{\rho\sigma\mu\nu}$ (20 components) form the Ricci tensor $R_{\mu\nu} = R^{\lambda}{}_{\mu\lambda\nu}$ — symmetric, $\tfrac12 n(n+1) = \mathbf{10}$ components in four dimensions. Why: Ricci is what the Einstein field equations speak to: $R_{\mu\nu} - \tfrac12 R\,g_{\mu\nu} = \tfrac{8\pi G}{c^4}T_{\mu\nu}$ ties all ten, algebraically and locally, to the matter that is right there. Where there is no matter, Ricci is zero.

Step 2. Subtract. The trace-free remainder is the Weyl tensor $C_{\rho\sigma\mu\nu}$ : $\ \mathbf{20} = \underbrace{\mathbf{10}}_{\text{Ricci}} + \underbrace{\mathbf{10}}_{\text{Weyl}}$ . Why: the ten Weyl components are the free field. Not fixed by the local matter — they are what curvature is doing where there is no matter at all. In vacuum $R_{\mu\nu} = 0$ and yet $C_{\rho\sigma\mu\nu} \neq 0$ : that is the tide outside a star, and that is a gravitational wave.

Step 3. Run the same count in three dimensions. Riemann: $\tfrac{1}{12}n^2(n^2-1)$ at $n=3$ gives $\tfrac{9\cdot 8}{12} = \mathbf{6}$ . Ricci: $\tfrac12 n(n+1)$ at $n=3$ gives $\tfrac{3\cdot 4}{2} = \mathbf{6}$ .

\mathbf{6} = \underbrace{\mathbf{6}}_{\text{Ricci}} + \underbrace{\mathbf{0}}_{\text{Weyl}} \qquad\Longrightarrow\qquad \boxed{\;C_{\rho\sigma\mu\nu} \equiv 0 \ \text{ in } n = 3\;}

Why: Ricci exhausts Riemann. Curvature is then determined pointwise and algebraically by the matter at that point — so vacuum is flat, exactly, and there is nothing for gravity to do outside a source.

Step 4. Count the cost. In $2+1$ -dimensional Einstein gravity there are therefore no gravitational waves (nothing to propagate), no tidal forces in vacuum (nothing to feel), and no Schwarzschild solution (a point mass makes a conical defect, not a field). Gravity has zero local degrees of freedom. Why: now look back at four dimensions with fresh eyes. The reason gravity propagates — the reason LIGO detects anything, the reason there are black holes — is that $20 > 10$ : ten components more than the matter can dictate. In three dimensions that surplus is zero.

Step 5. Close the chapter. Every result here is now one sentence: a global inertial frame exists on a region if and only if all twenty components of $R^{\rho}{}_{\sigma\mu\nu}$ vanish there. Necessity is the counting theorem; sufficiency is the converse — if $R = 0$ , parallel transport is path-independent, so one frame can be carried consistently over the whole region, and it is inertial everywhere. Why: that is why special relativity is exact rather than approximate. Not the leading term in an expansion — the theory you get by setting twenty numbers to zero, and if they are zero it is right, everywhere, forever. `aq-sr-rindler-metric` is the illustration: the Rindler chart looks curved, has non-zero Christoffels, has a horizon — and its Riemann tensor is identically zero. Flat spacetime in disguise, with a global inertial frame sitting there the whole time. *Verified in verify_aq-sr-failure-global-inertial-frames.py: the Riemann, Ricci and Weyl counts are recomputed for $n = 2,3,4,5$ ; the $n=3$ Weyl count is asserted to be exactly zero (and the $n=4$ count exactly ten); the Weyl tensor of an explicit 3-metric is built with sympy and shown to vanish identically; and the Rindler chart is shown to have non-zero Christoffels and identically zero Riemann.*

Limiting cases

Case	Result	Meaning
single event $P$	$g = \eta$ , $\Gamma = 0$	SR is exactly true at a point. $40 - 40 = 0$
lab of size $L$	deviation $\sim \tfrac13\lvert R\rvert L^{2}$	quadratic, not sudden. The coefficient is the curvature
$R^{\rho}{}_{\sigma\mu\nu} = 0$	global inertial frame exists	the sufficiency half. SR is exact, not approximate
Rindler chart	$\Gamma \neq 0$ but $R = 0$	non-trivial chart, flat spacetime. `aq-sr-rindler-metric`
vacuum, $n = 4$	$R_{\mu\nu} = 0$ , $C \neq 0$	ten Weyl components survive: tides and waves
vacuum, $n = 3$	$C \equiv 0 \Rightarrow R = 0$	no waves, no vacuum tides, no black holes

Figure

WHAT THE TWENTY ARE MADE OF — and why gravity would not propagate in a flatter universe. The twenty irremovable components of the Riemann tensor split, cleanly and invariantly, into two halves of ten. TEN ARE RICCI. Einstein's field equations tie all ten of them algebraically and locally to the stress-energy tensor: where there is matter, Ricci is fixed by it, and where there is no matter, Ricci is exactly zero. TEN ARE WEYL — the trace-free remainder, which the field equations say nothing about. Those ten are the FREE FIELD: the curvature that exists where there is no matter at all. They are the tide you feel outside a star, they are what propagates as a gravitational wave, and they are what remains of a star that has collapsed into a black hole. Gravity is interesting because 20 is greater than 10 — because there are ten more components than the local matter can dictate. NOW RUN THE SAME ARITHMETIC IN THREE DIMENSIONS. Riemann has n²(n²−1)/12 = 6 components; Ricci has n(n+1)/2 = 6. Six equals six, so the remainder is ZERO: the Weyl tensor vanishes identically in three dimensions, Ricci exhausts Riemann, and curvature is completely determined pointwise by the matter sitting at that point. Vacuum is exactly flat. The consequences are total — no gravitational waves, because there is nothing to propagate; no tidal forces in vacuum, because there is nothing to feel; no Schwarzschild solution, because a point mass produces a conical defect rather than a field; and no local degrees of freedom whatsoever. We live one dimension above the threshold, and everything interesting about gravity lives in that margin of ten

Common trapFour things in the base of this article are wrong, and two of them invert the physics

1 · The sign of the tidal accelerations is backwards — and the base contradicts itself. Its eq[2], captioned "Newtonian tidal accelerations (irremovable)", prints

\Delta a_r \approx -\frac{2GM}{r^{3}}\,\Delta r, \qquad \Delta a_\perp \approx +\frac{GM}{r^{3}}\,\Delta x_\perp .

Those are the eigenvalues of the tidal tensor $E_{ij}$ , not accelerations. The relative acceleration is $\ddot\xi^i = -E_{ij}\xi^j$ — the negative — so the true values are $+\tfrac{2GM}{r^3}\xi_r$ (stretch) and $-\tfrac{GM}{r^3}\xi_\perp$ (squeeze). As printed, the base has the oceans bulging at the poles. And it knows better: its own derivation card, two panels later, correctly states $\ddot\xi_r > 0$ (stretch), $\ddot\xi_\perp < 0$ (squeeze). The two cards contradict each other and only one can be right. The check takes two seconds: a tide pulls you apart along the field line.

2 · The Riemann-normal-coordinate coefficient is $1/2$ ; it should be $1/3$ . The base's limits card prints $g_{\mu\nu}(x) \approx \eta_{\mu\nu} + \tfrac12 R_{\mu\alpha\nu\beta}x^{\alpha}x^{\beta}$ . The $\tfrac12$ is the Taylor coefficient — it belongs in front of $\partial_\alpha\partial_\beta g_{\mu\nu}$ , not $R$ . Converting one to the other costs a further $-\tfrac23$ , and the correct expansion is $\eta_{\mu\nu} - \tfrac13 R_{\mu\alpha\nu\beta}x^{\alpha}x^{\beta}$ . The base is wrong by a factor of $-3/2$ , and the verify script settles it by building normal coordinates on a round sphere.

3 · $\eta_{\mu\nu} = \operatorname{diag}(-1,+1,+1,+1)$ , and 4 · the Rindler metric, are both mostly-plus. In this lane's signature $\eta_{\mu\nu} = \operatorname{diag}(+1,-1,-1,-1)$ and the Rindler chart reads $\mathrm{d}s^2 = +(1 + a\xi/c^2)^2c^2\mathrm{d}\tau^2 - \mathrm{d}\xi^2 - \dots$ ; `aq-sr-rindler-metric` owns it and states it correctly.

Not re-derived here. The tidal tensor, its eigenvalues and the expiry inequality are `aq-sr-equivalence-principle`'s; geodesic deviation is `aq-sr-tidal-forces-equivalence-limits`'s — used, not repeated. This article's job is the complementary one: not that a tide is irremovable, but how much is, and what object the residue is.

Worked examples

How big can a laboratory be before SR breaks?

Step 1 — identify. The deviation from flatness across a lab of size $L$ is $\sim \tfrac13|R|L^2$ , from the normal-coordinate expansion. Demand it stay below a fractional tolerance $\epsilon$ .

Step 2 — the curvature scale. Near a mass $M$ at radius $r$ the dominant component is $R^{r}{}_{0r0} \approx \tfrac{1}{c^2}\cdot\tfrac{2GM}{r^3}$ . At the Earth's surface $\tfrac{2GM}{r^3} = 3.083\times10^{-6}\ \mathrm{s^{-2}}$ (`aq-sr-equivalence-principle`'s number), so $|R| \approx 3.4\times10^{-23}\ \mathrm{m^{-2}}$ .

Step 3 — the radius of curvature. $|R|^{-1/2} \approx \mathbf{1.7\times10^{11}\ m}$ — 1.14 astronomical units.

Step 4 — interpret. That is the length over which spacetime near the Earth is appreciably curved. A metre-sized lab is smaller by eleven orders of magnitude, so the fractional deviation from flatness is $\sim\tfrac13(1/1.7\times10^{11})^2 \approx 10^{-23}$ .

Step 5 — the boundary number. A curvature radius of 1.14 AU is why special relativity works on a bench. Curvature is fantastically weak and fantastically irremovable — and those are independent facts. The counting theorem speaks only to the second.

Origins · 2 minRiemann curvature tensor and how this was discovered

Riemann curvature tensor

1854–1915 · Riemann builds the instrument · Einstein spends sixty years' worth of mathematics in three

The instrument. On 10 June 1854 Bernhard Riemann delivered his habilitation lecture at Göttingen — Über die Hypothesen, welche der Geometrie zu Grunde liegen — with Gauss, near the end of his life, in the audience. Riemann proposed that space need not be flat, that its geometry is a physical question rather than an a-priori one, and that the deviation from flatness at a point is measured by an object built from second derivatives of the metric. He did not know what it was for. He built the tensor and put it on the shelf, where it sat for sixty years. The application. In 1912, back in Zurich and stuck, Einstein turned to Marcel Grossmann for the mathematics he needed and did not have. Grossmann pointed him at Ricci and Levi-Civita's absolute differential calculus — and at Riemann's tensor. The Entwurf paper of 1913 was the first attempt, and was wrong; three more years produced, on 25 November 1915, the field equations. What Einstein had recognised is that Riemann's obstruction to flatness and gravity's obstruction to a global inertial frame are the same obstruction — that the twenty numbers you cannot transform away are the twenty numbers you feel as a tide. What survives. Everything. The tensor is still the object and the counting is still the reason, and the statement that closes this chapter — global inertial frames exist iff $R = 0$ — is the precise sense in which special relativity is not overthrown by general relativity but contained in it: the case where all twenty happen to vanish.

History

1854

— Riemann's habilitation lecture at Göttingen proposes that the geometry of space is a physical question, and that curvature is measured by second derivatives of the metric. Gauss is in the audience. The lecture is published posthumously in 1868.

1861

— Riemann introduces the curvature tensor explicitly, in a prize essay for the Paris Academy; it too is published only after his death.

1869

— Christoffel introduces the symbols that carry his name, giving the connection its calculational form.

1900

— Ricci-Curbastro and Levi-Civita publish the absolute differential calculus, the machinery that makes tensors usable.

1912

— Einstein, back in Zurich and unable to proceed, turns to Marcel Grossmann, who points him at Ricci, Levi-Civita and Riemann.

1913

— Einstein and Grossmann publish the Entwurf paper: the right variables, the wrong field equations.

1915

— On 25 November, Einstein presents the field equations to the Prussian Academy. The obstruction to a global inertial frame and the curvature of spacetime are identified as one object.

Going further

Texts. Misner, Thorne & Wheeler, Gravitation, §11.6 — Riemann normal coordinates, and the counting argument · Brewin, arXiv:0903.2087, eq. (11) — the normal-coordinate expansion with its coefficients, if you want the $1/3$ in print · Wald, General Relativity, §3.4 (curvature and geodesic deviation) · Weinberg, Gravitation and Cosmology, §6.2 · Carroll, Spacetime and Geometry, §3.10 · Carlip, Quantum Gravity in 2+1 Dimensions (Living Reviews) — for what the vanishing of Weyl costs · Riemann, Über die Hypothesen, welche der Geometrie zu Grunde liegen (1854; published 1868).

Provenance. Signature $(+,-,-,-)$ . Every count is recomputed in verify_aq-sr-failure-global-inertial-frames.py, which reads this file, regexes the printed integers out, and derives each by brute-force enumeration rather than formula — the $256 \to 21 \to 20$ reduction canonicalises all $4^4$ index quadruples and takes the rank of the Bianchi system. Four base defects repaired: the inverted sign of the tidal accelerations (which contradicts the base's own derivation card); the normal-coordinate coefficient ( $1/2 \to 1/3$ , machine-checked on the sphere); and two mostly-plus metrics.

Scope. The tidal tensor, its eigenvalues and the expiry inequality are `aq-sr-equivalence-principle`'s; geodesic deviation is `aq-sr-tidal-forces-equivalence-limits`'s, used here and not repeated; the Rindler chart's flatness is `aq-sr-rindler-metric`'s. The Einstein field equations, the Schwarzschild solution and gravitational-wave physics belong to the general-relativity track.

Builds on. The Equivalence Principle · Tidal Forces and the Limits of the Equivalence Principle · The Rindler Metric · Gravitational Time Dilation. Leads to. From Special to General Relativity.

Why is the equivalence principle described here as a "coincidence"?

Because the count comes out $40 - 40$ , and there is no deeper reason for it to. The first derivatives of the metric number $4 \times 10 = 40$ ; the second-order coordinate parameters number $4 \times 10 = 40$ . They match exactly — so the derivatives can always be killed, free fall always works, and the equivalence principle is true. But it is true because two integers happen to be equal, and an exact match has no reserve. Push one order further and the arithmetic gives $100$ against $80$ : you are twenty short. The equivalence principle is not a symmetry that cleverness could extend; it is a numerical accident that runs out. That account explains what the "physical insight" version cannot: why the obstruction has exactly the size it does.

Why 80 third-order parameters and not 4 × 4 × 4 × 4 = 256?

Because $\partial_\alpha\partial_\beta\partial_\gamma x'^{\mu}$ is totally symmetric in its three lower indices — mixed partials commute — so orderings may not be counted separately. Unordered triples drawn with repetition from four values number $\binom{4+3-1}{3} = \binom{6}{3} = 20$ , and with the free upper index that is $4\times20 = 80$ . The same logic gives the $40$ at second order: symmetric pairs from four values number $\binom{5}{2} = 10$ , and $4\times10 = 40$ . The symmetry of mixed partials is the entire source of the deficit — it stops the coordinate freedom growing as fast as the derivatives it must cancel. If derivatives did not commute there would be no curvature to speak of.

Is "the Riemann tensor is the obstruction" a definition or a theorem?

A theorem, and it matters which half. Necessity — global inertial frame $\Rightarrow R = 0$ — is nearly a definition: in that frame $g = \eta$ everywhere, so all derivatives vanish, so $R$ (built from them) vanishes; and being a tensor, it then vanishes in every frame. Sufficiency is the substantive half: if $R = 0$ on a region, parallel transport there is path-independent, so a frame chosen at one point can be carried consistently to every other, and the result is a genuine global inertial frame. That direction is what makes special relativity exact rather than approximate — it is not the leading term of an expansion in curvature, it is the theory you get when twenty numbers are zero. The two counts establish the size of the obstruction; the iff establishes that the obstruction is the only thing in the way.

The Rindler chart has non-zero Christoffel symbols. Isn't it therefore curved?

No, and this is the cleanest test of whether one has understood the counting. Christoffel symbols are not tensors: they measure how your coordinates twist, and the $40 - 40$ argument makes them vanish at any point you like. The Rindler chart has non-zero $\Gamma$ , a position-dependent clock rate, and even a horizon — and `aq-sr-rindler-metric` computes its Riemann tensor and gets identically zero, every component. It is flat spacetime, described awkwardly. A non-zero $\Gamma$ tells you about your chart. A non-zero $R$ tells you about the world — which is why the criterion for a global inertial frame is stated in terms of $R$ and never $\Gamma$ .

What exactly do the ten Weyl components do?

They are the gravitational field where there is no matter. Einstein's equations tie the ten Ricci components algebraically to the local stress-energy — so in vacuum Ricci is zero and the equations are satisfied — and yet curvature is not zero, because the ten Weyl components are unconstrained by them. Those ten are the tide outside a star (`aq-sr-tidal-forces-equivalence-limits` computes it), what propagates as a gravitational wave, and what remains of a collapsed star. They are the free field: not caused by matter here, only influenced by matter elsewhere. And they are the reason gravity is interesting. Remove them — go to three dimensions — and gravity has nothing to say outside a source.

If curvature is so tiny near the Earth, why does any of this matter?

Because tiny and removable are different properties, and only one is relevant. Near the Earth's surface the curvature radius $|R|^{-1/2}$ is about $1.7\times10^{11}$ m — roughly one astronomical unit — so across a metre-sized laboratory the fractional deviation from flatness is around $10^{-23}$ . Utterly negligible, and that is exactly why special relativity works on a bench. But it is not zero, and it cannot be made zero, by any coordinate change, ever. The counting theorem does not say the residue is large; it says the residue exists, and that no chart in the space of all charts disposes of it. Special relativity is exactly right only in the limit, and structurally incomplete otherwise — and repairing that incompleteness is what general relativity is.