Skip to content

Normal Coordinates and Local Metric Expansions

Scrollable formula. Use Left and Right arrow keys to reveal hidden content.

Scrollable table. Use Left and Right arrow keys to reveal additional columns.

Scrollable figure. Use Left and Right arrow keys to inspect hidden content.

Normal coordinates use the exponential map of the Levi-Civita connection to make geodesics through a chosen point appear as straight radial lines. At that point the connection coefficients and first metric derivatives vanish, but curvature remains in the quadratic metric term. With the site’s curvature convention, gμν=ημν−13Rμανβxαxβ+O(∣x∣3)g_{\mu\nu}=\eta_{\mu\nu}-\frac13R_{\mu\alpha\nu\beta}x^\alpha x^\beta+O(\lvert x\rvert^3).

Required background. Tensor Fields and Natural Operations supplies charts, tangent vectors, and Taylor expansion of tensor components; Pseudo-Riemannian Metrics and Signature supplies an orthonormal basis at a point.

Helpful background. Calculus and Linear Algebra for Relativity supplies multivariable Taylor expansions and inverse-function arguments.

Fix pp in a smooth pseudo-Riemannian manifold and choose an orthonormal basis eμe_\mu of TpMT_pM. For v=vμeμv=v^\mu e_\mu sufficiently small, let exp⁡p(v)\exp_p(v) be the point reached at affine parameter one by the Levi-Civita geodesic with initial tangent vv. Since d(exp⁡p)0d(\exp_p)_0 is the identity, the inverse-function theorem makes exp⁡p\exp_p a diffeomorphism from a neighborhood of 0∈TpM0\in T_pM to a neighborhood of pp. Define the coordinates of exp⁡p(v)\exp_p(v) to be xμ=vμx^\mu=v^\mu.

Radial lines xμ(λ)=λvμx^\mu(\lambda)=\lambda v^\mu are then affinely parametrized geodesics. Substitution into the geodesic equation gives

Γρμν(λv)vμvν=0.\Gamma^\rho{}_{\mu\nu}(\lambda v)v^\mu v^\nu=0.

At λ=0\lambda=0, this holds for every vv. Because the Levi-Civita coefficients are symmetric in μ,ν\mu,\nu, polarization implies

Γρμν(p)=0.\Gamma^\rho{}_{\mu\nu}(p)=0.

Metric compatibility then yields

∂αgμν(p)=0,gμν(p)=ημν.\partial_\alpha g_{\mu\nu}(p)=0, \qquad g_{\mu\nu}(p)=\eta_{\mu\nu}.

Tong constructs these coordinates and states their immediate consequences in Tong 2019, §3.3.2, pp. 119–123, PDF.

Vanishing Γ(p)\Gamma(p) removes no tensorial curvature. Adopt the site convention

Rρσμν=∂μΓρνσ−∂νΓρμσ+ΓρμλΓλνσ−ΓρνλΓλμσ.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}.

Write a comma for a coordinate derivative evaluated at pp. Since Γρμν(p)=0\Gamma^\rho{}_{\mu\nu}(p)=0, expanding the radial-geodesic identity one order farther gives

0=Γρμν,α(p)xμxνxα+O(∣x∣4).0 =\Gamma^\rho{}_{\mu\nu,\alpha}(p) x^\mu x^\nu x^\alpha+O(\lvert x\rvert^4).

Replace xx by txtx, divide by t3t^3, and let t→0t\to0. The leading cubic polynomial must vanish for every xx, so polarization says explicitly that

Γρμν,α(p)+Γρνα,μ(p)+Γραμ,ν(p)=0.\Gamma^\rho{}_{\mu\nu,\alpha}(p) +\Gamma^\rho{}_{\nu\alpha,\mu}(p) +\Gamma^\rho{}_{\alpha\mu,\nu}(p)=0.

At the same point, the quadratic ΓΓ\Gamma\Gamma terms in the curvature vanish, leaving

Rρσμν(p)=Γρνσ,μ(p)−Γρμσ,ν(p).R^\rho{}_{\sigma\mu\nu}(p) =\Gamma^\rho{}_{\nu\sigma,\mu}(p) -\Gamma^\rho{}_{\mu\sigma,\nu}(p).

Solving these two linear identities, using the symmetry of the Levi-Civita coefficients in their lower indices, yields

Γρμν,α(p)=−13(Rρμνα(p)+Rρνμα(p)).\Gamma^\rho{}_{\mu\nu,\alpha}(p) =-\frac13\left( R^\rho{}_{\mu\nu\alpha}(p) +R^\rho{}_{\nu\mu\alpha}(p) \right).

Now lower the output index by writing Γρμν=gρλΓλμν\Gamma_{\rho\mu\nu}=g_{\rho\lambda}\Gamma^\lambda{}_{\mu\nu}. Differentiating metric compatibility once and evaluating at pp gives

gμν,αβ(p)=Γναμ,β(p)+Γμαν,β(p)=−13(Rμανβ(p)+Rμβνα(p)).g_{\mu\nu,\alpha\beta}(p) =\Gamma_{\nu\alpha\mu,\beta}(p) +\Gamma_{\mu\alpha\nu,\beta}(p) =-\frac13\left( R_{\mu\alpha\nu\beta}(p) +R_{\mu\beta\nu\alpha}(p) \right).

The ordinary Taylor formula therefore makes the factor of two explicit:

gμν(x)=ημν+12gμν,αβ(p)xαxβ+O(∣x∣3)=ημν−13Rμανβ(p)xαxβ+O(∣x∣3),\begin{aligned} g_{\mu\nu}(x) &=\eta_{\mu\nu} +\frac12g_{\mu\nu,\alpha\beta}(p)x^\alpha x^\beta +O(\lvert x\rvert^3)\\ &=\eta_{\mu\nu} -\frac13R_{\mu\alpha\nu\beta}(p)x^\alpha x^\beta +O(\lvert x\rvert^3), \end{aligned}

where the last step exchanges the dummy indices α\alpha and β\beta in one term. This calculation is also a useful diagnostic: the curvature definition, the connection derivative, and the metric expansion must carry compatible signs.

The minus sign is tied to the displayed curvature convention. Importing the formula from a source with the opposite Riemann-tensor sign without translation would reverse it. The local proof needs nondegeneracy rather than positive definiteness. For the normal-coordinate construction with an arbitrary nondegenerate signature, see Wendl 2021–2022, §23.1, Propositions 23.1–23.2, pp. 192–193, PDF.

The notation O(∣x∣3)O(\lvert x\rvert^3) is coordinate-local: it means that the remaining components are bounded by a constant times a Euclidean coordinate norm cubed in a sufficiently small chart. It is not a global distance estimate.

On a unit two-sphere, geodesic polar coordinates about the north pole give

ds2=dr2+sin⁡2r dθ2.ds^2=dr^2+\sin^2r\,d\theta^2.

Normal Cartesian coordinates are x=rcos⁡θx=r\cos\theta, y=rsin⁡θy=r\sin\theta. The radial direction retains unit coefficient, while the tangential direction is scaled by

(sin⁡rr)2=1−r23+O(r4).\left(\frac{\sin r}{r}\right)^2 =1-\frac{r^2}{3}+O(r^4).

For constant sectional curvature K=+1K=+1,

Riαjβ=δijδαβ−δiβδαj.R_{i\alpha j\beta} =\delta_{ij}\delta_{\alpha\beta} -\delta_{i\beta}\delta_{\alpha j}.

The general expansion therefore becomes

gij=δij−13(r2δij−xixj)+O(r3),g_{ij} =\delta_{ij} -\frac13\left(r^2\delta_{ij}-x_ix_j\right) +O(r^3),

which changes only the tangential coefficient by −r2/3-r^2/3. Agreement with (sin⁡r/r)2(\sin r/r)^2 independently verifies the quadratic sign.

At pp, normal coordinates remove the Levi-Civita coefficients and the first metric derivatives. They do not remove curvature, which depends on derivatives of the coefficients and appears at second order. They also do not make the metric constant on a neighborhood. A freely falling-coordinate interpretation belongs to gravitational physics; geometrically, the statement is simply a local coordinate normalization.

The chart exists only where exp⁡p\exp_p is one-to-one and nonsingular. Beyond a normal neighborhood, distinct radial geodesics can reach the same point or conjugate behavior can spoil the coordinate map. For a torsionful connection, coordinate transformations cannot remove the tensorial antisymmetric coefficient part. For a degenerate metric, there is no Levi-Civita connection from which to construct these coordinates.

Finally, Γ(p)=0\Gamma(p)=0 does not imply R(p)=0R(p)=0. If curvature happens to vanish at pp, the quadratic term vanishes in exponential-map normal coordinates, but higher derivatives may still make the neighborhood nonflat. Flatness on an open simply connected region is a much stronger statement.

The volume density supplies another contraction check. Taking the determinant of the quadratic metric expansion gives, in the same normal coordinates and curvature convention,

∣g(x)∣=1−16Rαβ(p)xαxβ+O(∣x∣3).\sqrt{\lvert g(x)\rvert} =1-\frac16R_{\alpha\beta}(p)x^\alpha x^\beta +O(\lvert x\rvert^3).

This follows from the first determinant variation δ∣g∣=12∣g∣ gμνδgμν\delta\sqrt{\lvert g\rvert} =\tfrac12\sqrt{\lvert g\rvert}\,g^{\mu\nu}\delta g_{\mu\nu} and the contraction gμνRμανβ=Rαβg^{\mu\nu}R_{\mu\alpha\nu\beta}=R_{\alpha\beta}. It verifies both the factor 1/31/3 in the metric and the Ricci contraction sign without calculating every connection derivative separately.

Exponential-map normal coordinates are not unique because the orthonormal basis at pp is not unique; changing it applies a constant pseudo-orthogonal transformation (a Lorentz transformation in Lorentzian signature), under which the displayed curvature coefficient transforms as a tensor. This restricted statement must not be extended to every chart that merely satisfies g(p)=ηg(p)=\eta and ∂g(p)=0\partial g(p)=0. A coordinate change that starts at cubic order preserves those two pointwise conditions but can alter the quadratic metric coefficients and destroy the radial-geodesic property.

Flat one-dimensional space gives a decisive check. Starting from ds2=dx2ds^2=dx^2, set y=x+ax3y=x+a x^3. Near the origin, x=y−ay3+O(y5)x=y-a y^3+O(y^5), and hence

gyy=(dxdy)2=1−6ay2+O(y4).g_{yy}=\left(\frac{dx}{dy}\right)^2 =1-6a y^2+O(y^4).

Thus gyy(0)=1g_{yy}(0)=1 and ∂ygyy(0)=0\partial_y g_{yy}(0)=0 although the quadratic coefficient is nonzero while curvature vanishes identically. Curvature is the invariant combination of second derivatives; the simple −Rμανβ/3-R_{\mu\alpha\nu\beta}/3 coefficient belongs specifically to exponential-map normal coordinates. Cubic and higher coefficients there require derivatives of curvature and further geometric data.

For a sphere of radius aa, write the normal-coordinate expansion through quadratic order and check it against ds2=dr2+a2sin⁡2(r/a)dθ2ds^2=dr^2+a^2\sin^2(r/a)d\theta^2.

Solution

The sectional curvature is K=a−2K=a^{-2}, so

gij=δij−13a2(r2δij−xixj)+O(r3).g_{ij} =\delta_{ij} -\frac{1}{3a^2} \left(r^2\delta_{ij}-x_ix_j\right) +O(r^3).

The exact tangential scale is

(asin⁡(r/a)r)2=1−r23a2+O(r4),\left(\frac{a\sin(r/a)}{r}\right)^2 =1-\frac{r^2}{3a^2}+O(r^4),

while the radial coefficient remains one. This agrees with the tensor expansion and reduces to the flat metric as a→∞a\to\infty.

  • Tong, David. General Relativity. University of Cambridge lecture notes, 2019, §3.3.2, pp. 119–123. Official course PDF.
  • Wendl, Chris. Differential Geometry I and II. Humboldt-Universität zu Berlin lecture notes, 2021–2022, §23.1, pp. 192–193. Course PDF.
  • Lee, John M. Introduction to Riemannian Manifolds. 2nd ed., Graduate Texts in Mathematics 176, Springer, 2018, Chapter 5, pp. 115–150. DOI.