Metric Compatibility

A connection and a metric are, at birth, strangers that happen to live on the same manifold. The metric $g_{\alpha\beta}$ is a ruler-and-protractor: it reports how long a vector is and what angle two vectors make. The connection $\Gamma^\lambda_{\mu\nu}$ — assembled in aq-gr-affine-connection and made to act through the covariant derivative of aq-gr-covariant-derivative — is a pure transport rule: it says how to carry a vector from one tangent space to the neighbouring one, with no reference to length whatsoever. Nothing in the definition of either object forces them to agree. You can build a perfectly good connection that stretches every vector it moves, and a perfectly good metric that the connection completely ignores. Metric compatibility is the single postulate that welds the two together. It demands that the covariant derivative of the metric vanish, $\nabla_\mu g_{\alpha\beta}=0$ — the metric is covariantly constant. Read geometrically, this is the promise that when a vector is parallel transported (the transport of aq-gr-parallel-transport) along any curve at all, its length never changes and the angle it makes with any other transported vector never drifts. The ruler carried by the connection stays calibrated against the ruler defined by the metric. The payoff is the fundamental theorem of (pseudo-)Riemannian geometry: out of the enormous space of possible connections, the twin demands "preserve lengths" and "no torsion" select exactly one, whose coefficients are the Christoffel symbols of aq-gr-christoffel-symbols. Physically that theorem is the equivalence principle wearing geometric clothing — and it is the quiet reason that index-raising commutes with differentiation, that geodesics keep a fixed causal character, and that $\nabla_\mu T^{\mu\nu}=0$ makes sense at all.

Key takeaways

Metric compatibility is the postulate $\nabla_\mu g_{\alpha\beta}=0$ : the metric tensor is covariantly constant, so the transport rule respects the measuring rule.
Geometrically it means parallel transport preserves every inner product — vector lengths and the angles between vectors are invariant along any curve.
Together with vanishing torsion it uniquely fixes the connection to be the Levi-Civita connection, whose coefficients are the Christoffel symbols $\tfrac12 g^{\lambda\sigma}(\partial_\mu g_{\sigma\nu}+\partial_\nu g_{\sigma\mu}-\partial_\sigma g_{\mu\nu})$ .
It makes the covariant derivative commute with raising and lowering indices, and forces $\nabla_\mu g^{\alpha\beta}=0$ for the inverse metric too.
It is a physical assumption — the equivalence principle in geometric form — not a mathematical necessity; non-metricity theories such as Weyl geometry deliberately relax it, measuring the failure by the tensor $Q_{\mu\alpha\beta}\equiv\nabla_\mu g_{\alpha\beta}$ .
Key idea: demanding that the metric be covariantly constant, $\nabla_\mu g_{\alpha\beta}=0$ , makes parallel transport preserve all inner products; combined with zero torsion it singles out the unique Levi-Civita connection given by the Christoffel symbols.

Where this lives today

The postulate that once looked like a mathematician's tidiness is now an experimental target. Because a non-metric connection would let a transported clock or ruler drift its calibration, metric compatibility is tested every time we check that atomic transition frequencies are the same here and in distant galaxies. Bounds on any spacetime variation of fundamental constants — from quasar absorption spectra, from the Oklo natural fission reactor, and from laboratory optical-clock comparisons now reaching parts in $10^{18}$ — all constrain the non-metricity that Weyl first contemplated in 1918. At the theoretical frontier the same freedom is being reopened deliberately: symmetric-teleparallel gravity recasts gravitation with zero curvature and zero torsion but nonzero non-metricity $Q$ , so that the entire gravitational field lives in $Q_{\mu\alpha\beta}=\nabla_\mu g_{\alpha\beta}$ — the very object GR sets to zero. Whether such " $f(Q)$ " theories can match cosmological data as well as standard GR, and whether the cosmic-acceleration story needs anything beyond a metric-compatible connection, is an active research question. Metric compatibility, in other words, is no longer a definition to memorize but a hypothesis the universe keeps being asked to confirm.

The physical picture

Picture carrying a ruler and a protractor across a curved surface, always keeping them "as parallel as possible" in the sense fixed by the connection. Two things could in principle happen. The ruler could quietly stretch, so that a vector which measured one metre at the start reads a metre and a half after a long detour; and the protractor could drift, so that two vectors initially at right angles end up oblique. Metric compatibility is the flat refusal of both. It is the guarantee that the geometry's own notion of distance — encoded in $g_{\alpha\beta}$ — is exactly the notion of distance the connection preserves as it slides vectors around. The rule for moving vectors is forced to honour the rule for measuring them. That is what lets us differentiate inside a dot product just as in flat space, $\frac{\mathrm{d}}{\mathrm{d}\lambda}(V\cdot W)=(\nabla_u V)\cdot W+V\cdot(\nabla_u W)$ : the metric can be slid past the derivative because it does not change under it. The most vivid demonstration is a closed loop on a sphere. Transport a vector all the way around a spherical triangle and it comes back rotated — that rotation is curvature, the connection at work — but it comes back exactly the same length. Length preservation is compatibility; the rotation is a separate fact about torsion-free curvature. The first figure separates these two effects cleanly.

Figure

Parallel transport around one octant of a unit sphere: the vector returns rotated by the enclosed solid angle, here exactly π/2 = 90°, yet its length is preserved at every station because ∇ g = 0 keeps the inner product g_αβ Vᵃ Vᵇ constant

The index changes with nothing here — length is fixed — but the direction rotates, and that is the crucial split: compatibility is about magnitude, curvature is about direction. It is the geometric form of the equivalence principle. In a freely falling laboratory the metric looks like Minkowski's $\eta_{\alpha\beta}$ with vanishing first derivatives, so special relativity's constancy of intervals holds instantaneously; asserting "locally inertial everywhere" is precisely $\nabla g=0$ . Without compatibility, freely falling rods and clocks would silently recalibrate as they moved, and no experiment has ever seen that happen. The fundamental theorem is the punchline the second figure makes concrete. An arbitrary connection in four dimensions carries sixty-four independent numbers. Metric compatibility clamps forty of them; demanding zero torsion clamps the remaining twenty-four; sixty-four constraints on sixty-four unknowns leave exactly one connection standing — the Christoffel symbols. Nature, to every precision yet measured, uses that one.

Figure

The fundamental theorem as bookkeeping in n = 4: an affine connection has n³ = 64 independent components; metric compatibility ∇ g = 0 imposes ½n²(n+1) = 40 equations and zero torsion imposes the remaining n²(n−1)/2 = 24, leaving a single connection — the Christoffel symbols

The core equations

The whole subject is one equation and its consequences. The metric-compatibility condition says the covariant derivative of the metric tensor vanishes identically — every component, at every point, for every direction of differentiation. It is a postulate about the connection, imposed by hand.

\nabla_\mu g_{\alpha\beta}=0 \qquad (1)

where: $\nabla_\mu$ — covariant derivative associated with the connection; $g_{\alpha\beta}$ — components of the metric tensor. Physical reading: the metric is covariantly constant, so parallel transport moves vectors without ever touching their lengths or mutual angles. Written out through the connection coefficients, equation (1) becomes an algebraic relation between the coordinate derivatives of the metric and the $\Gamma$ 's. The covariant derivative of any $(0,2)$ tensor subtracts one connection term for each lower index, so two negative pieces appear.

\nabla_\mu g_{\alpha\beta}=\partial_\mu g_{\alpha\beta}-\Gamma^\lambda_{\mu\alpha}g_{\lambda\beta}-\Gamma^\lambda_{\mu\beta}g_{\alpha\lambda}=0 \qquad (2)

where: $\partial_\mu$ — ordinary partial (coordinate) derivative; $\Gamma^\lambda_{\mu\alpha}$ — connection coefficients. Physical reading: any gradient of the metric must be exactly matched by the $\Gamma$ terms, so where the metric varies from point to point the connection cannot be zero. Solving (2) — with the extra input that the connection is symmetric in its lower indices — gives the unique closed form for the coefficients: the Christoffel symbols, built purely from first derivatives of the metric and the inverse metric.

\Gamma^\lambda_{\mu\nu}=\tfrac12 g^{\lambda\sigma}\left(\partial_\mu g_{\sigma\nu}+\partial_\nu g_{\sigma\mu}-\partial_\sigma g_{\mu\nu}\right) \qquad (3)

where: $g^{\lambda\sigma}$ — inverse metric, defined by $g^{\lambda\sigma}g_{\sigma\rho}=\delta^\lambda_\rho$ ; $\Gamma^\lambda_{\mu\nu}$ — symmetric, torsion-free connection coefficients. Physical reading: gravity's connection is completely determined by the metric — there is no independent "force field" beyond the geometry itself. The geometric meaning is cleanest along a curve. Compatibility lets the metric pass through the derivative of an inner product, so the rate of change of $V\cdot W$ is carried entirely by the covariant derivatives of $V$ and $W$ .

\frac{\mathrm{d}}{\mathrm{d}\lambda}\left(g_{\alpha\beta}V^\alpha W^\beta\right)=g_{\alpha\beta}\left(\frac{DV^\alpha}{\mathrm{d}\lambda}W^\beta+V^\alpha\frac{DW^\beta}{\mathrm{d}\lambda}\right) \qquad (4)

where: $V^\alpha,W^\beta$ — two vector fields along the curve; $D/\mathrm{d}\lambda=u^\mu\nabla_\mu$ — covariant derivative along the curve; $\lambda$ — parameter along the curve. Physical reading: no $\nabla g$ term survives, precisely because it is zero — this is the whole content of compatibility in action. Impose that both vectors are parallel transported, $DV/\mathrm{d}\lambda=DW/\mathrm{d}\lambda=0$ , and the inner product is frozen: lengths and angles ride along unchanged.

\frac{D}{\mathrm{d}\lambda}\left(g_{\alpha\beta}V^\alpha W^\beta\right)=0\quad\text{when}\quad \frac{DV^\alpha}{\mathrm{d}\lambda}=\frac{DW^\beta}{\mathrm{d}\lambda}=0 \qquad (5)

where: $DV/\mathrm{d}\lambda=0$ — the statement that $V$ is parallel transported. Physical reading: the geometry ships a whole orthonormal frame from point to point without distortion — the basis of local inertial frames. Because $g$ is covariantly constant, so is its inverse, and index-raising or lowering therefore commutes with covariant differentiation — one of the workhorse manipulations of general relativity.

\nabla_\mu g^{\alpha\beta}=0,\qquad \nabla_\mu V_\alpha=g_{\alpha\beta}\,\nabla_\mu V^\beta \qquad (6)

where: $g^{\alpha\beta}$ — inverse metric; $V_\alpha=g_{\alpha\beta}V^\beta$ — lowered-index component. Physical reading: you may slide indices up and down freely inside a $\nabla$ , which is what makes covariant tensor calculus internally consistent. Finally, when compatibility is dropped the failure is itself a tensor. The non-metricity tensor measures exactly how much a vector's norm changes per unit transport, and it vanishes identically in any metric-compatible theory.

Q_{\mu\alpha\beta}\equiv\nabla_\mu g_{\alpha\beta},\qquad u^\mu\nabla_\mu\!\left(g_{\alpha\beta}u^\alpha u^\beta\right)=0 \ \text{for a geodesic} \qquad (7)

where: $Q_{\mu\alpha\beta}$ — non-metricity, zero in GR; $u^\alpha$ — tangent of an affinely parametrized geodesic, $u^\mu\nabla_\mu u^\alpha=0$ . Physical reading: in GR $Q=0$ , so a geodesic's tangent keeps a constant norm and a particle's worldline cannot change between timelike, null, and spacelike.

Derivation AThe fundamental theorem — deriving the Christoffel symbols

The goal is to prove that metric compatibility together with the torsion-free (symmetric) condition determines the connection uniquely and to produce its closed form. The strategy is a classic index trick: write the compatibility condition $\nabla_\mu g_{\nu\rho}=0$ three separate times with the free indices cyclically permuted, then add the first two and subtract the third. Symmetry of the connection makes six of the resulting connection terms cancel in pairs, leaving a single $\Gamma$ that can be freed by contracting with the inverse metric. The result is equation (3), the Christoffel symbols, and it establishes the fundamental theorem of Riemannian geometry.

Step 1. Expand compatibility and move the partial derivative to one side: $\partial_\mu g_{\nu\rho}=\Gamma^\lambda_{\mu\nu}g_{\lambda\rho}+\Gamma^\lambda_{\mu\rho}g_{\nu\lambda}$ . Why: this converts the geometric statement $\nabla g=0$ into an algebraic relation in which the unknown $\Gamma$ 's appear linearly, ready to be solved.

Step 2. Cyclically permute $(\mu\nu\rho)\to(\nu\rho\mu)$ to get a second copy, $\partial_\nu g_{\rho\mu}=\Gamma^\lambda_{\nu\rho}g_{\lambda\mu}+\Gamma^\lambda_{\nu\mu}g_{\rho\lambda}$ . Why: permuting the same identity generates independent equations that share the same $\Gamma$ 's, so combining them can isolate one particular coefficient.

Step 3. Permute once more $\to(\rho\mu\nu)$ for the third copy, $\partial_\rho g_{\mu\nu}=\Gamma^\lambda_{\rho\mu}g_{\lambda\nu}+\Gamma^\lambda_{\rho\nu}g_{\mu\lambda}$ . Why: three cyclic versions are exactly what is needed so that the unwanted terms line up to cancel in the next step.

Step 4. Form (Step 1) + (Step 2) − (Step 3). Using the symmetry $\Gamma^\lambda_{\mu\nu}=\Gamma^\lambda_{\nu\mu}$ (zero torsion), four of the six $\Gamma$ terms cancel pairwise and the survivors combine to give $\partial_\mu g_{\nu\rho}+\partial_\nu g_{\rho\mu}-\partial_\rho g_{\mu\nu}=2\,\Gamma^\lambda_{\mu\nu}g_{\lambda\rho}$ . Why: this is where torsion-freeness earns its keep — without symmetry the antisymmetric parts would not cancel and no closed form would exist.

Step 5. Contract both sides with $\tfrac12 g^{\rho\sigma}$ to strip the metric off the surviving $\Gamma$ : $\Gamma^\sigma_{\mu\nu}=\tfrac12 g^{\sigma\rho}(\partial_\mu g_{\nu\rho}+\partial_\nu g_{\rho\mu}-\partial_\rho g_{\mu\nu})$ . Why: invertibility of the metric turns " $2\Gamma g$ " into " $\Gamma$ ", proving the solution both exists and is unique — the fundamental theorem. *As a check: a symbolic computation on the unit 2-sphere reproduces $\Gamma^\theta_{\phi\phi}=-\sin\theta\cos\theta$ and $\Gamma^\phi_{\theta\phi}=\cot\theta$ , and the formula correctly reduces to $\Gamma=0$ for a constant metric.*

Go deeper · Derivation B · 1 minParallel transport preserves lengths and angles

Derivation BParallel transport preserves lengths and angles

The goal is to show that metric compatibility is equivalent to the physical statement that inner products are conserved under parallel transport — the length of a vector and the angle between two vectors do not change as they are carried along a curve. The method is to differentiate the scalar $g_{\alpha\beta}V^\alpha W^\beta$ along the curve, apply the Leibniz rule for the covariant derivative, and watch the metric term vanish by compatibility. Whatever remains is proportional to the transport equations, so if the vectors are transported the whole thing is zero. This is the result that makes local orthonormal frames globally meaningful.

Step 1. For a scalar the ordinary and covariant derivatives coincide, so $\frac{\mathrm{d}}{\mathrm{d}\lambda}(g_{\alpha\beta}V^\alpha W^\beta)=u^\mu\nabla_\mu(g_{\alpha\beta}V^\alpha W^\beta)$ with $u^\mu=\mathrm{d}x^\mu/\mathrm{d}\lambda$ . Why: promoting the plain derivative to a covariant one lets us use the tensorial Leibniz rule and the compatibility identity, which live at the level of $\nabla$ .

Step 2. Apply the product rule: $u^\mu\big[(\nabla_\mu g_{\alpha\beta})V^\alpha W^\beta+g_{\alpha\beta}(\nabla_\mu V^\alpha)W^\beta+g_{\alpha\beta}V^\alpha(\nabla_\mu W^\beta)\big]$ . Why: the covariant derivative distributes across products exactly like an ordinary derivative, so the three factors each contribute one term.

Step 3. Invoke compatibility, $\nabla_\mu g_{\alpha\beta}=0$ , killing the first term to leave $g_{\alpha\beta}\big(\tfrac{DV^\alpha}{\mathrm{d}\lambda}W^\beta+V^\alpha\tfrac{DW^\beta}{\mathrm{d}\lambda}\big)$ . Why: this is the single place compatibility is used — remove it and a stray $Q_{\mu\alpha\beta}V^\alpha W^\beta u^\mu$ term survives and lengths drift.

Step 4. Impose parallel transport, $DV^\alpha/\mathrm{d}\lambda=DW^\beta/\mathrm{d}\lambda=0$ , so $\frac{\mathrm{d}}{\mathrm{d}\lambda}(V\cdot W)=0$ . Setting $W=V$ gives $|V|^2=$ const; keeping them distinct gives constant $\cos\theta=V\cdot W/(|V|\,|W|)$ . Why: both the norms and the angle between the vectors are frozen, which is precisely "the ruler and protractor never change" — the geometric meaning of compatibility. *The derivation is reversible: preservation of all inner products under transport implies $\nabla g=0$ , so the two statements are equivalent.*

Go deeper · Derivation C · 1 minThe residual freedom — connection equals Christoffel plus contortion

Derivation CThe residual freedom — connection equals Christoffel plus contortion

The goal is to make precise a subtle and often-missed point: metric compatibility by itself does not fix the connection. It leaves a residual freedom parametrized entirely by the torsion. The method is to recall that the difference of any two connections is a genuine tensor, then ask what compatibility permits that tensor to be. The answer is a contortion tensor built from torsion, so the general metric-compatible connection is the Christoffel symbol plus contortion. Setting torsion to zero recovers Levi-Civita, and this is exactly why the fundamental theorem needs two postulates, not one.

Step 1. Define the torsion as the antisymmetric part of the connection in its lower indices, $T^\lambda{}_{\mu\nu}=\Gamma^\lambda_{\mu\nu}-\Gamma^\lambda_{\nu\mu}$ ; unlike $\Gamma$ itself this transforms as a tensor. Why: the non-tensorial, symmetric part of $\Gamma$ is what compatibility pins down, so the leftover freedom must live in the tensorial, antisymmetric torsion.

Step 2. Build the contortion tensor from torsion, $K^\lambda{}_{\mu\nu}=\tfrac12(T^\lambda{}_{\mu\nu}+T_\mu{}^\lambda{}_\nu+T_\nu{}^\lambda{}_\mu)$ , which when its first and third indices are lowered is antisymmetric in that pair. Why: that antisymmetry is exactly the condition for adding $K$ to a compatible connection without spoiling $\nabla g=0$ — it contributes nothing to the symmetric metric equation.

Step 3. Write the general solution $\Gamma^\lambda_{\mu\nu}=\bar\Gamma^\lambda_{\mu\nu}+K^\lambda{}_{\mu\nu}$ , where $\bar\Gamma$ is the Christoffel symbol of equation (3). Setting $T^\lambda{}_{\mu\nu}=0$ makes $K=0$ and recovers the unique Levi-Civita connection. Why: this displays compatibility's residual freedom explicitly — the connection is fixed only once torsion is also chosen (zero, in GR; nonzero in Einstein–Cartan gravity). *Setting $T^\lambda{}_{\mu\nu}=0$ recovers the Christoffel symbols, as expected. In $n=4$ the torsion carries $n^2(n-1)/2=24$ components — exactly the freedom left after the 40 compatibility equations constrain the 64 connection components.*

Limiting cases

How the condition specializes or is deliberately relaxed in the important settings:

Case	Result	Meaning
Flat / inertial frame	$\partial_\mu\eta_{\alpha\beta}=0,\ \Gamma^\lambda_{\mu\nu}=0$	In Cartesian coordinates on Minkowski space the metric is constant, the Christoffel symbols vanish, compatibility is trivial, and $\nabla\to\partial$ .
Levi-Civita (GR)	$\nabla_\mu g_{\alpha\beta}=0,\ T^\lambda{}_{\mu\nu}=0$	Both compatibility and zero torsion hold; the connection is the unique Christoffel connection used throughout general relativity.
Einstein–Cartan / torsion	$\nabla_\mu g_{\alpha\beta}=0,\ T^\lambda{}_{\mu\nu}\neq0$	Compatibility retained but torsion allowed; the connection gains a contortion piece. Lengths are still preserved, but the uniqueness theorem no longer applies.
Non-metricity (metric-affine)	$Q_{\mu\alpha\beta}\equiv\nabla_\mu g_{\alpha\beta}\neq0$	Compatibility deliberately dropped; $Q$ measures how vector norms change under transport, as in Weyl geometry and symmetric-teleparallel gravity.

Common trapMetric compatibility alone determines the connection

It does not. Compatibility supplies only $\tfrac12 n^2(n+1)$ equations — forty in four dimensions — while an affine connection has $n^3=64$ components. The gap of twenty-four is precisely the torsion, and it stays free until you separately impose the torsion-free (symmetric) condition. Only both postulates together give the Christoffel symbols; either one alone is not enough. The complementary error is to treat $\nabla_\mu g_{\alpha\beta}=0$ as automatic or purely mathematical. It is a physical postulate — the geometric equivalence principle — and entire families of theories (Weyl geometry, metric-affine and symmetric-teleparallel gravity) keep the metric while abandoning compatibility, encoding the departure in a non-metricity tensor $Q$ that is measured, and found consistent with zero, in the real universe.

Worked examples

1 · Metric compatibility on the sphere, checked explicitly (confidence builder)

Take the round 2-sphere of radius $a$ with $\mathrm{d}s^2=a^2(\mathrm{d}\theta^2+\sin^2\theta\,\mathrm{d}\phi^2)$ , so $g_{\theta\theta}=a^2$ , $g_{\phi\phi}=a^2\sin^2\theta$ . Compute its Christoffel symbols from equation (3) and verify $\nabla_\theta g_{\phi\phi}=0$ by hand.

Step 1. The inverse metric is diagonal, $g^{\theta\theta}=1/a^2$ and $g^{\phi\phi}=1/(a^2\sin^2\theta)$ . Only $g_{\phi\phi}$ depends on a coordinate, and $\partial_\theta g_{\phi\phi}=2a^2\sin\theta\cos\theta$ . Every other first derivative of the metric vanishes, so at most a handful of $\Gamma$ 's can be nonzero.

Step 2. Apply equation (3). For $\Gamma^\theta_{\phi\phi}=\tfrac12 g^{\theta\theta}(2\partial_\phi g_{\theta\phi}-\partial_\theta g_{\phi\phi})=\tfrac12(1/a^2)(-2a^2\sin\theta\cos\theta)=\mathbf{-\sin\theta\cos\theta}$ . For $\Gamma^\phi_{\theta\phi}=\tfrac12 g^{\phi\phi}\,\partial_\theta g_{\phi\phi}=\tfrac12\,\dfrac{1}{a^2\sin^2\theta}\,(2a^2\sin\theta\cos\theta)=\mathbf{\cot\theta}$ . All other components are zero. Notice these are independent of $a$ — the sphere's shape fixes the connection, its size does not.

Step 3. Now test compatibility directly from equation (2): $\nabla_\theta g_{\phi\phi}=\partial_\theta g_{\phi\phi}-2\Gamma^\phi_{\theta\phi}g_{\phi\phi}=2a^2\sin\theta\cos\theta-2(\cot\theta)(a^2\sin^2\theta)=2a^2\sin\theta\cos\theta-2a^2\sin\theta\cos\theta=\mathbf{0}$ . The two terms cancel exactly. A symbolic pass over all indices confirms $\nabla_\mu g_{\alpha\beta}=0$ for every $(\mu,\alpha,\beta)$ — the Christoffel connection is metric-compatible on the sphere, as the fundamental theorem guarantees.

2 · Counting the freedom in four dimensions

In $n=4$ spacetime dimensions, count the independent components of a general affine connection, the equations imposed by metric compatibility, and the components of torsion, and confirm that compatibility plus zero torsion leaves a unique connection.

Step 1. A connection $\Gamma^\lambda_{\mu\nu}$ has one upper and two lower indices with no symmetry assumed, so $n^3=4^3=\mathbf{64}$ independent components. This is the full pool of freedom before any postulate.

Step 2. Metric compatibility $\nabla_\mu g_{\alpha\beta}=0$ carries one free index $\mu$ ( $n$ choices) and a symmetric pair $(\alpha\beta)$ ( $\tfrac12 n(n+1)$ choices), for $\tfrac12 n^2(n+1)=\tfrac12(16)(5)=\mathbf{40}$ independent equations. That leaves $64-40=\mathbf{24}$ components undetermined.

Step 3. Torsion $T^\lambda{}_{\mu\nu}=\Gamma^\lambda_{\mu\nu}-\Gamma^\lambda_{\nu\mu}$ is antisymmetric in $\mu\nu$ , so it has $n\cdot\tfrac12 n(n-1)=4\cdot 6=\mathbf{24}$ components — exactly the residual freedom. Setting $T=0$ therefore imposes the missing 24 equations and $40+24=64$ constraints fix all 64 components. The connection is unique: it is the Levi-Civita connection given by the Christoffel symbols.

Physics of the portrait

Metric compatibility is where two centuries of pure geometry became physics. The condition was implicit in Riemann's manifolds and Christoffel's symbols, but its meaning — that transport preserves length — only became visible once someone gave the connection an intrinsic, coordinate-free interpretation. That someone was Levi-Civita.

Origins · 1 minTullio Levi-Civita and how this was discovered

Tullio Levi-Civita

1873–1941 · the notion of parallel transport and the metric connection that bears his name

Tullio Levi-Civita, an Italian mathematician trained at Padua under Gregorio Ricci-Curbastro, gave the covariant derivative its geometric soul. In 1900 he and Ricci published the absolute differential calculus — tensor calculus — the formalism Einstein would borrow to write general relativity in 1915. But Levi-Civita's decisive personal contribution came in 1917, when he introduced the notion of parallelism on an arbitrary manifold: a coordinate-free rule for saying that a vector at one point is "the same" as a vector at a neighbouring point. This is parallel transport, and it revealed that the Christoffel symbols, until then a computational bookkeeping device, actually encode how vectors are dragged through a curved space. His construction singled out the connection that leaves lengths and angles invariant — the metric-compatible, torsion-free connection now universally called the Levi-Civita connection. It is the unique connection GR uses, and the fundamental theorem of Riemannian geometry is the statement that Levi-Civita's two demands, compatibility and symmetry, determine it completely. Einstein corresponded warmly with him and once said the Italian's work was among the most elegant he knew; when the Fascist racial laws of 1938 stripped Levi-Civita of his chair and his memberships, the loss was felt across all of mathematical physics. The concept he named turned the metric and the connection from two independent structures into one.

History

1854

— Riemann's habilitation lecture founds the geometry of curved manifolds

1869

— Christoffel introduces the symbols Γ and covariant differentiation of quadratic forms

1900

— Ricci-Curbastro and Levi-Civita systematize the absolute differential calculus

1917

— Levi-Civita defines parallel transport, giving the connection its geometric meaning

1918

— Weyl relaxes metric compatibility, introducing non-metricity in his gauge geometry

Going further

Texts. Misner, Thorne & Wheeler, Gravitation, §8.5 and Ch. 10 (connections and the metric) · Wald, General Relativity, §3.1 (the derivative operator and metric compatibility, eqs. 3.1.21–3.1.30) · Carroll, Spacetime and Geometry, §3.2 (covariant derivative, compatibility eq. 3.23, Christoffel eq. 3.27) · Schutz, A First Course in General Relativity, §6 · Hartle, Gravity, §8. The Christoffel formula (3) follows from the three-permutation derivation; $\nabla_\mu g^{\alpha\beta}=0$ follows from $\nabla_\mu g_{\alpha\beta}=0$ ; and the contortion decomposition reduces to Levi-Civita at zero torsion. A symbolic computation on the round 2-sphere reproduces $\Gamma^\theta_{\phi\phi}=-\sin\theta\cos\theta$ , $\Gamma^\phi_{\theta\phi}=\cot\theta$ and confirms $\nabla_\mu g_{\alpha\beta}=0$ for every index; the component count in $n=4$ gives $64=40+24$ exactly. No empirical inputs are needed beyond the signature convention $(-,+,+,+)$ and units $c=1$ .

Prerequisites. The Covariant Derivative · The Affine Connection · Parallel Transport · Christoffel Symbols. Leads to. The Levi-Civita Connection · The Geodesic Equation · The Torsion Tensor · Killing Vectors.

Ask the tutor: Ask me to parallel-transport a vector around a loop on any surface and watch its length stay fixed while its direction rotates — the clearest way to feel why $\nabla_\mu g_{\alpha\beta}=0$ is the equivalence principle in disguise.

Is metric compatibility a theorem or an assumption?

It is an assumption — a physical postulate about which connection nature uses. Mathematically a manifold can carry any number of connections, and most of them are not compatible with a given metric: they stretch vectors as they transport them. General relativity postulates that the physical connection is the one for which $\nabla_\mu g_{\alpha\beta}=0$ , because that is the geometric content of the equivalence principle — freely falling rulers and clocks keep their calibration. Once you also postulate zero torsion, the fundamental theorem elevates this to uniqueness: there is exactly one such connection, the Levi-Civita connection. So compatibility is an assumption; the uniqueness of its consequence is the theorem.

Why exactly are there two minus signs in ∇μ g_αβ?

Because the covariant derivative of a tensor subtracts one connection term for every lower index and adds one for every upper index. The metric $g_{\alpha\beta}$ has two lower indices and no upper ones, so it collects two $-\Gamma$ corrections: $\nabla_\mu g_{\alpha\beta}=\partial_\mu g_{\alpha\beta}-\Gamma^\lambda_{\mu\alpha}g_{\lambda\beta}-\Gamma^\lambda_{\mu\beta}g_{\alpha\lambda}$ . Each term "corrects" one index for the fact that the coordinate basis itself twists from point to point. Had you differentiated the inverse metric $g^{\alpha\beta}$ instead, the two indices would be upstairs and you would get two $+\Gamma$ terms — and, reassuringly, that expression also vanishes in a compatible theory.

Does compatibility mean spacetime is flat?

No, and this is a common confusion. Compatibility is about lengths under transport, not about curvature. A vector parallel-transported around a closed loop on a curved manifold comes back rotated — that rotation is curvature and it is entirely consistent with compatibility — but it comes back the same length. The sphere is the standard example: it is intensely curved, yet its Levi-Civita connection is perfectly metric-compatible, as the worked example shows. Curvature lives in the second derivatives of the metric (the Riemann tensor); compatibility is a condition on how the first derivatives feed the connection. The two ideas are independent.

What actually goes wrong in a theory without metric compatibility?

Vector norms stop being invariant under transport. In Weyl's 1918 geometry, for instance, a vector carried around a closed loop can return rescaled by a factor depending on the path — so a clock's ticking rate would depend on its history, and two identical atoms brought together from different journeys could disagree on their transition frequencies. Einstein's immediate objection to Weyl's theory was exactly this: we do not observe such "second clock effects", so the non-metricity must be extremely small. The failure is quantified by the tensor $Q_{\mu\alpha\beta}=\nabla_\mu g_{\alpha\beta}$ ; every high-precision test of the constancy of atomic spectra across space and time is, in effect, a bound on $Q$ .

How does compatibility let me raise and lower indices inside a derivative?

Because a covariantly constant object can be slid freely past $\nabla$ . Since $\nabla_\mu g_{\alpha\beta}=0$ and (by the same token) $\nabla_\mu g^{\alpha\beta}=0$ , the metric behaves like a constant with respect to covariant differentiation: $\nabla_\mu(g_{\alpha\beta}V^\beta)=g_{\alpha\beta}\nabla_\mu V^\beta$ , i.e. $\nabla_\mu V_\alpha=g_{\alpha\beta}\nabla_\mu V^\beta$ . This is the property that keeps tensor calculus consistent — you can lower an index and then differentiate, or differentiate and then lower, and get the same answer. Without compatibility these two operations would disagree, and the whole apparatus of covariant conservation laws such as $\nabla_\mu T^{\mu\nu}=0$ would lose its meaning.

Why does a geodesic keep the same causal character?

Because compatibility freezes the norm of its tangent vector. An affinely parametrized geodesic satisfies $u^\mu\nabla_\mu u^\alpha=0$ , and then $u^\mu\nabla_\mu(g_{\alpha\beta}u^\alpha u^\beta)=g_{\alpha\beta}\,2u^\alpha(u^\mu\nabla_\mu u^\beta)=0$ , where the metric passed through the derivative only because $\nabla g=0$ . So $u\cdot u$ is constant along the whole geodesic. If it starts negative (timelike), it stays negative; if zero (null), it stays null; if positive (spacelike), spacelike. This is why a massive particle's worldline can be normalized to $g_{\alpha\beta}u^\alpha u^\beta=-1$ once and for all, and why light rays remain light rays — the causal structure is preserved by transport.