4 Fermi normal coordinates: the Manasse–Misner construction
Where we are
Part I produced observables of a single kind. Every one of them mixes source physics, propagation, and observer terms that acquire meaning only in combination, which is why establishing gauge invariance took the machinery of the last two chapters. Part II turns to measurements that need none of it.
The preceding observables compare a source with an observer across a finite null geodesic and are therefore intrinsically nonlocal. We now ask a different question: what gravitational information can be measured inside a small freely falling laboratory? Fermi normal coordinates answer this by constructing coordinates around an entire timelike geodesic. The construction removes the connection on the central worldline while retaining the curvature that produces relative acceleration.
Two debts from Part I are settled here. Chapter 1 asserted that the local gravitational operators available to a formation process are the density and the trace-free tide, on the grounds that a freely falling observer can discard a constant potential and a uniform acceleration. That was a Newtonian argument. This chapter replaces it with a theorem: there exist coordinates in which the connection vanishes along an entire geodesic, the metric is Minkowski to first order in the spatial distance from it, and the leading correction is quadratic with coefficients fixed by the Riemann tensor. Whatever a local experiment can measure must therefore be built from curvature. Separately, the chapter supplies the frame in which Chapter 5 defines an intrinsic galaxy shape, which must be a tensor in a local orthonormal rest frame rather than in a cosmological chart.
The chapter opens by fixing two conventions, the curvature sign and the meaning of a hatted index, and then, in Section 4.2, by asking what such coordinates could possibly achieve. Counting the freedom in a coordinate change against the number of metric coefficients at each order settles both how far the gravitational field can be removed and what is guaranteed to survive. The construction then proceeds in four steps: fix a tetrad on a central geodesic and parallel transport it; use the exponential map along orthogonal spacelike geodesics to assign coordinates to nearby events; show that the resulting chart is regular near the worldline; and Taylor expand the metric, using the geodesic identities to convert connection derivatives into curvature. No field equation is used at any point, so the result holds in any smooth Lorentzian spacetime.
Figure placeholder. Construction of Fermi normal coordinates. The central timelike geodesic \(\mathcal G\) with the parallel-transported tetrad drawn at two values of proper time; from one of them, a fan of spacelike geodesics orthogonal to \(e_{\hat 0}\) sweeping out the constant-\(X^{\hat 0}\) surface; a nearby event \(P\) labelled by the proper time \(\tau\) of its originating point and the proper distance \(s\) along the geodesic reaching it.
4.1 Conventions
Two conventions are fixed here and then used without further comment: the sign and index order of the Riemann tensor, and what a hat on an index means.
4.1.1 Curvature
We use
\[ \begin{aligned} R^\rho{}_{\sigma\mu\nu} ={}&\partial_\mu\Gamma^\rho_{\nu\sigma} -\partial_\nu\Gamma^\rho_{\mu\sigma} +\Gamma^\rho_{\mu\lambda}\Gamma^\lambda_{\nu\sigma} -\Gamma^\rho_{\nu\lambda}\Gamma^\lambda_{\mu\sigma},\\ [\nabla_\mu,\nabla_\nu]V^\rho ={}&R^\rho{}_{\sigma\mu\nu}V^\sigma. \end{aligned} \tag{4.1}\]
Fixing the curvature convention is essential because the sign of the quadratic Fermi metric and of geodesic deviation changes when the Riemann tensor is defined with the opposite commutator order. The formulas below are internally tied to Eq. 4.1; translating a result from another reference requires translating both its index order and its curvature sign.
Manasse and Misner write the first covariant index before the contravariant index, \(R_\mu{}^\nu{}_{\alpha\beta}\). Their mixed-index differential definition agrees with Eq. 4.1 after translating the index order. However, their all-lowered object \(R^{\mathrm{MM}}_{\mu\nu\alpha\beta}\) has the first pair reversed relative to ours, so \(R^{\mathrm{MM}}_{\mu\nu\alpha\beta}=-R_{\mu\nu\alpha\beta}\). This accounts for the opposite signs between their printed Eq. (66) and Eq. 4.48, Eq. 4.49, and Eq. 4.50 below.
4.1.2 Frame indices
Two kinds of index appear from here on, and the distinction matters enough to state before it is used.
An unhatted Greek index \(\mu,\nu,\rho,\sigma\), as in Eq. 4.1, labels a component in an arbitrary chart on spacetime. Nothing is assumed about that chart. A hatted index labels a component in the Fermi coordinate chart \(X^{\hat a}\).
On the central geodesic the Fermi coordinate basis is constructed to coincide with the parallel-transported orthonormal tetrad \(e_{\hat a}\), which is Eq. 4.24. Hence hatted components of a tensor evaluated on \(\mathcal G\) also equal its tetrad projections. Away from the worldline they are Fermi-coordinate components and should not be read as local Lorentz components. This distinction matters for the metric expansion itself, whose hatted components are functions of \(X^i\) away from \(\mathcal G\).
Within this chapter the notation is:
| Index | Range | Meaning |
|---|---|---|
| \(\mu,\nu,\rho,\sigma\), and \(\mu'\) for a second chart | 0 to 3 | arbitrary chart on spacetime; never a frame index |
| \(\hat a,\hat b,\hat c,\hat d\) | 0 to 3 | Fermi-coordinate components; on \(\mathcal G\), also observer-tetrad components |
| \(\hat 0\) | Fermi time direction; on \(\mathcal G\), the observer-tetrad time direction | |
| \(i,j,k,l,m\) | 1 to 3 | spatial Fermi-coordinate directions within this chapter; hats are suppressed |
Within this chapter, Latin letters denote spatial Fermi-coordinate indices, so the hat is suppressed. Every four-dimensional Fermi index carries a hat, including the time direction, which is \(\hat 0\). In Chapter 5, named weak-field and TT charts also use Latin letters for their Cartesian spatial coordinates; there a bare time index \(0\) identifies a chart component, while \(\hat 0\) identifies the Fermi-frame quantity measured by the observer.
Two consequences are worth spelling out. The coordinates are \(X^{\hat a}=(X^{\hat 0},X^i)\) and the tetrad is \(e_{\hat a}=(e_{\hat 0},e_i)\), so the exponential map Eq. 4.17 pairs \(X^i\) with \(e_i\) and the notation matches the structure. And \(\Gamma^{\hat a}_{ij}\), to take the case that most often provokes the question, is a Christoffel symbol of the Fermi chart whose two lower indices have been restricted to spatial directions while the upper one still runs over all four; the hat on \(\hat a\) and its absence on \(i,j\) say exactly that.
Chapter 5 writes \(R_{0i0j}\) with a bare time index in two places, and this is deliberate rather than a lapse. Both occur inside a named chart rather than in the observer frame: the weak-field metric of Eq. 5.1, written in local physical coordinates, and the transverse-traceless chart of Eq. 5.15. The frame quantities in that chapter, \(\delta R_{\hat 0i\hat 0j}\) and \(\mathcal E^{\mathrm{GW}}_{ij}=R^{\mathrm{GW}}_{\hat 0i\hat 0j}\), carry the hat as they should. Reading the hat is therefore enough to tell a chart component from a measured one.
4.2 The goal, and a degree-of-freedom count
Before constructing anything it is worth stating precisely what is wanted and asking whether it can be had.
We want coordinates adapted to one freely falling observer. Erect an orthonormal tetrad at a point of the observer’s worldline, thread coordinates outward from it, and then expand the metric in the resulting chart. The aim is to remove as much of the gravitational field as the coordinates permit, and to find out what is left when no more can be removed. The leftover is the physics, because it is what no observer can transform away.
The question is how far the removal can go. Write the map from an arbitrary chart \(x^\mu\) to the coordinates \(X^{\hat a}\) we are building as a Taylor series about the origin,
\[ x^\mu =A^\mu{}_{\hat a}X^{\hat a} +\frac{1}{2!}B^\mu{}_{\hat a\hat b}X^{\hat a}X^{\hat b} +\frac{1}{3!}C^\mu{}_{\hat a\hat b\hat c} X^{\hat a}X^{\hat b}X^{\hat c} +\mathcal O(X^4), \tag{4.2}\]
with \(B\) symmetric in its two lower indices and \(C\) totally symmetric in its three. The metric in the new chart is likewise a series,
\[ g_{\hat a\hat b}(X) =g_{\hat a\hat b}(0) +g_{\hat a\hat b,\hat c}(0)X^{\hat c} +\frac{1}{2}g_{\hat a\hat b,\hat c\hat d}(0) X^{\hat c}X^{\hat d} +\mathcal O(X^3). \tag{4.3}\]
At each order the coefficients of Eq. 4.2 are the freedom, and the coefficients of Eq. 4.3 are what we are trying to set to zero. Counting both in \(n=4\) spacetime dimensions settles the matter.
| Order | Metric coefficients | Available freedom | Balance |
|---|---|---|---|
| \(g_{\hat a\hat b}\) | \(n(n+1)/2=10\) | \(A^\mu{}_{\hat a}\): \(n^2=16\) | 6 left over |
| \(g_{\hat a\hat b,\hat c}\) | \(10n=40\) | \(B^\mu{}_{\hat a\hat b}\): \(40\) | exactly enough |
| \(g_{\hat a\hat b,\hat c\hat d}\) | \(10\times10=100\) | \(C^\mu{}_{\hat a\hat b\hat c}\): \(80\) | 20 cannot be removed |
A counting argument is only as good as the structure that licenses it, so it is worth saying why the rows can be read independently. The system is triangular. The value \(g_{\hat a\hat b}(0)\) involves \(A\) alone; the first derivatives involve \(A\) and \(B\), with \(B\) entering linearly; the second derivatives involve \(A\), \(B\), and \(C\), with \(C\) entering linearly. One may therefore solve order by order, each order fixing the next coefficient while the earlier ones are already determined and merely contribute inhomogeneous terms. What matters at each order is not a vague supply of adjustable parameters but whether one specific linear map is onto.
Read the rows in turn.
At zeroth order there is more freedom than there are conditions. Ten conditions fix the metric to \(\eta_{\hat a\hat b}\), and the six surviving parameters are the Lorentz group: having chosen one orthonormal frame, a boost or a rotation gives another.
At first order the count is tight, and the reason is worth making explicit rather than leaving to the arithmetic. Define the Jacobian and its inverse by
\[ A^\mu{}_{\hat a} \equiv \left.\frac{\partial x^\mu}{\partial X^{\hat a}}\right|_0, \qquad A^{\hat a}{}_{\mu}A^\mu{}_{\hat b} =\delta^{\hat a}{}_{\hat b}, \qquad A^\mu{}_{\hat a}A^{\hat a}{}_{\nu} =\delta^\mu{}_{\nu}. \tag{4.4}\]
The hatted Christoffel symbol is a coordinate connection coefficient of the Fermi chart,
\[ \nabla_{\partial_{\hat b}}\partial_{\hat c} =\Gamma^{\hat a}{}_{\hat b\hat c}\partial_{\hat a}, \qquad \Gamma^{\hat a}{}_{\hat b\hat c} =\frac12g^{\hat a\hat d} (\partial_{\hat b}g_{\hat c\hat d} +\partial_{\hat c}g_{\hat b\hat d} -\partial_{\hat d}g_{\hat b\hat c}). \tag{4.5}\]
It is not the tetrad projection of the old Christoffel symbols, because a connection is not a tensor. At the origin its transformation law is
\[ \Gamma^{\hat a}{}_{\hat b\hat c}(0) =A^{\hat a}{}_{\rho} \left[ B^\rho{}_{\hat b\hat c} +\Gamma^\rho{}_{\mu\nu}(0) A^\mu{}_{\hat b}A^\nu{}_{\hat c} \right]. \tag{4.6}\]
The Hessian \(B\) is the inhomogeneous part. Therefore demanding \(\Gamma^{\hat a}{}_{\hat b\hat c}(0)=0\) determines it uniquely:
\[ \boxed{ B^{\rho}{}_{\hat a\hat b} =-A^{\mu}{}_{\hat a}A^{\nu}{}_{\hat b}\, \Gamma^{\rho}_{\mu\nu}(0) }. \tag{4.7}\]
Here \(\Gamma^{\rho}_{\mu\nu}\) carries Greek indices throughout: it is the ordinary spacetime connection of the original chart \(x^\mu\), evaluated at the origin, and not a tetrad-frame quantity. The two factors of \(A\) convert its lower spacetime indices into the hatted indices carried by \(B\); the inverse Jacobian in Eq. 4.6 converts the upper index of the new connection.
The same result follows directly from the metric. Differentiating \(g_{\hat a\hat b}=g_{\mu\nu}(\partial_{\hat a}x^\mu)(\partial_{\hat b}x^\nu)\) and taking the cyclic combination \(g_{\hat a\hat b,\hat c}+g_{\hat a\hat c,\hat b}-g_{\hat b\hat c,\hat a}\) gives
\[ 0=2g_{\mu\nu}A^\mu{}_{\hat a} \left[ B^\nu{}_{\hat b\hat c} +\Gamma^\nu{}_{\rho\sigma} A^\rho{}_{\hat b}A^\sigma{}_{\hat c} \right], \tag{4.8}\]
so nondegeneracy of \(g_{\mu\nu}A^\mu{}_{\hat a}\) reproduces Eq. 4.7. The local-inertial map through quadratic order is consequently
\[ x^\mu=z^\mu+A^\mu{}_{\hat a}X^{\hat a} -\frac12\Gamma^\mu{}_{\nu\rho}(z) A^\nu{}_{\hat a}A^\rho{}_{\hat b} X^{\hat a}X^{\hat b} +\mathcal O(X^3). \tag{4.9}\]
\(B\) is therefore not an independent supply of forty knobs that happens to match forty conditions; it is solved for, uniquely, in terms of \(A\) and the spacetime connection at the origin. The equality of the two counts is the shadow of a sharper fact: a connection and the inhomogeneous term \(B^\rho{}_{\hat a\hat b}\) carry the same index symmetry, symmetric in the lower pair, so the map from \(B\) to the connection it induces is a bijection. That is why the first derivatives can always be removed, and why nothing is left over once they are. This is the equivalence principle as a statement about linear algebra: a freely falling observer detects no gravitational field through the connection because the connection is pure inhomogeneous term.
At second order the map is no longer onto. The coefficient \(C^\mu{}_{\hat a\hat b\hat c}\) is totally symmetric in its three lower indices, while \(g_{\hat a\hat b,\hat c\hat d}\) is symmetric within the pair \(\hat a\hat b\) and within the pair \(\hat c\hat d\) but has no symmetry relating the two pairs. The two objects sit in different symmetry classes, and the shift that \(C\) can generate spans only an eighty-dimensional subspace of the hundred-dimensional target. The twenty directions outside that image are unreachable by any choice of coordinates, and they are precisely the combinations carrying the algebraic symmetries of the Riemann tensor. Twenty is not a coincidence: it is \(n^2(n^2-1)/12\) evaluated at \(n=4\).
Two ways of trying to evade this fail, and seeing why confirms that the twenty are genuine. The six Lorentz parameters left unspent at zeroth order act on the surviving components as a linear representation; a group action moves a nonzero tensor around its orbit but never to the origin, so the residual frame freedom rotates the twenty rather than reducing them. And one might try spending \(B\) differently, declining to remove all forty first derivatives in the hope of buying something at second order. That trades away forty quantities a local experiment would detect for at most twenty it cannot remove anyway, which is never worth doing.
The three rows answer both questions at once. They say that \(g_{\hat a\hat b,\hat c}=0\) is the right target, because it is the last condition that can be met completely. And they say that the expansion must therefore be carried to \(X^2\), because that is the first order at which anything survives, and that what survives there is exactly the Riemann tensor. The quadratic term of the Fermi metric is not one contribution among several; at that order it is the whole of the local gravitational field.
There remains the question the counting above does not answer. It was carried out at a single point, which gives Riemann normal coordinates. We want the conditions to hold not at one event but at every proper time along the observer’s worldline, which is infinitely many copies of the same demand. Is that still possible?
It is, and the reason is that two independent geometric prescriptions divide the work exactly between them. Along the central geodesic the connection has the same forty components, which split according to how many of the two lower indices are spatial:
| Components | Count | Removed by |
|---|---|---|
| \(\Gamma^{\hat a}_{ij}\) | \(4\times6=24\) | the coordinate lines being spatial geodesics |
| \(\Gamma^{\hat b}_{\hat 0\hat a}\) | \(4\times4=16\) | parallel transport of the tetrad |
The first block is killed by threading the coordinates outward along geodesics that leave the worldline orthogonally, so that each radial coordinate line has vanishing acceleration. The second is killed by carrying the tetrad along the worldline without rotating it, which also covers \(\Gamma^{\hat a}_{\hat 0\hat 0}=0\), the statement that the central curve is itself a geodesic. Together they account for all forty, with none left over and none demanded twice.
So the construction is available, and the remainder of the chapter carries it out. Section 4.4 builds the map by the exponential prescription, Section 4.6 verifies the two blocks in the table, and Section 4.9 extracts the twenty surviving components at order \(X^2\).
Two caveats belong with the count. It is local: nothing here guarantees that the coordinates remain single valued far from the worldline, and they do not. And it assumes the central worldline is a geodesic. Allowing acceleration removes the geodesic condition that supplied \(\Gamma^{\hat a}_{\hat 0\hat 0}=0\), and a term linear in \(X^i\) reappears in \(g_{\hat 0\hat 0}\).
4.3 Geometric data on the central geodesic
The central geodesic represents the ideal freely falling observer. A coordinate system around this observer also requires a prescription for spatial axes. Parallel transport supplies nonaccelerating, nonrotating axes along a geodesic and allows tensor components measured at different proper times to be compared in one tetrad frame.
The idea of erecting coordinates along a worldline so that the metric is Minkowski and its first derivatives vanish all along it, rather than at a single event as in Riemann normal coordinates, goes back to1. Fermi’s construction addressed the general question of what happens in the neighbourhood of a timelike curve; the extension to accelerated and rotating observers, now standard in relativistic metrology, uses Fermi–Walker rather than parallel transport. The systematic quadratic expansion used in this chapter, with the metric coefficients written in terms of the Riemann tensor on the central geodesic, is due to2, who also introduced the affine-rescaling argument reproduced in Eq. 4.18. Their index conventions differ from the ones used here, which is the source of an apparent sign discrepancy; Appendix D carries out the translation.
Choose a point \(P_0\) and a future-directed unit timelike vector \(e_{\hat 0}(0)\) at \(P_0\). Let
\[ z:\tau\longmapsto z(\tau) \tag{4.10}\]
be the timelike geodesic with
\[ z(0)=P_0, \qquad \dot z^\mu=e_{\hat 0}^\mu, \qquad \nabla_{e_{\hat 0}}e_{\hat 0}=0, \qquad e_{\hat 0}\cdot e_{\hat 0}=-1. \tag{4.11}\]
Choose three spacelike vectors \(e_i(0)\) at \(P_0\) so that
\[ e_{\hat a}(0)\cdot e_{\hat b}(0)=\eta_{\hat a\hat b}. \tag{4.12}\]
Parallel transport the full tetrad along \(z(\tau)\):
\[ \nabla_{e_{\hat 0}}e_{\hat a}=0. \tag{4.13}\]
Metric compatibility preserves orthonormality for all \(\tau\).
Because the tetrad is orthonormal, \(e_{\hat 0}\) is the observer’s four-velocity and the three \(e_i\) span its instantaneous rest space. Because it is parallel transported, no inertial acceleration or rotation is introduced by the frame. If the central worldline were accelerated, Fermi–Walker rather than parallel transport would be required and linear acceleration terms would remain in the local metric.
The construction depends on the choice of the initial spatial triad only through a constant spatial rotation. Choosing a different central geodesic defines a different local observer.
4.4 Orthogonal geodesics and the coordinate map
At each proper time on the central worldline, shoot spacelike geodesics in all directions orthogonal to \(e_{\hat 0}\). The initial tangent specifies a spatial displacement in the local tetrad. The exponential map sends that tangent vector to the endpoint of the corresponding geodesic. This construction defines both simultaneity and spatial position relative to the observer.
Following Secs. II and IV of2, let
\[ h(\tau;\alpha^i;\lambda) \tag{4.14}\]
denote the spacelike geodesic with affine parameter \(\lambda\) satisfying
\[ \begin{aligned} h(\tau;\alpha;0)&=z(\tau),\\ \left.\frac{\partial h}{\partial\lambda}\right|_{\lambda=0} &=\alpha^i e_i(\tau), \qquad \delta_{ij}\alpha^i\alpha^j=1. \end{aligned} \tag{4.15}\]
The initial tangent is orthogonal to the central four-velocity. For a point with proposed Fermi coordinates \((X^{\hat 0},X^i)\), define
\[ s\equiv(\delta_{ij}X^iX^j)^{1/2}, \qquad \alpha^i\equiv\frac{X^i}{s} \tag{4.16}\]
for \(s\neq0\). The physical point is
\[ P(X)=h(X^{\hat 0};\alpha^i;s) =\exp_{z(X^{\hat 0})}\!\left(X^i e_i(X^{\hat 0})\right). \tag{4.17}\]
Thus \(X^{\hat 0}=\tau\) is proper time on the central geodesic, while \(s\) is proper distance along the orthogonal spacelike geodesic at its initial point.
The statement about \(s\) is local to the initial point: affine parametrization preserves the norm of the tangent along the spacelike geodesic, so its parameter length equals proper length. Surfaces of constant \(X^{\hat 0}\) are generated by these orthogonal geodesics; away from the central worldline they need not be orthogonal to a global timelike congruence.
Manasse and Misner use the affine-rescaling identity
\[ h(\tau;s\alpha^i;\lambda) =h(\tau;\alpha^i;s\lambda). \tag{4.18}\]
Both sides solve the same geodesic equation with the same initial point and tangent after the parameter rescaling, so uniqueness proves the identity. Eq. 4.17 can therefore be written without separating length and direction:
\[ P(X)=h(X^{\hat 0};X^i;1). \tag{4.19}\]
This is the form used to establish differentiability and regularity at \(X^i=0\).
Writing the map directly in terms of \(X^i\) avoids the apparent singularity of the direction \(\alpha^i=X^i/s\) at the origin. Smooth dependence of geodesic solutions on their initial data makes \(h(\tau;X^i;1)\) a smooth function of the Cartesian tangent-space components.
4.5 Regularity near the central geodesic
A geometric prescription does not automatically guarantee a valid coordinate chart. The Jacobian must be nonsingular, and each nearby event must be reached by a unique orthogonal geodesic from the central worldline. The first property follows locally from the tetrad; the second restricts the construction to a normal tubular neighborhood.
Let \(y^{\mu'}\) be any regular coordinates in a neighborhood of the central geodesic. The coordinate transformation is
\[ y^{\mu'}(X)=h^{\mu'}(X^{\hat 0};X^i;1). \tag{4.20}\]
From the initial tangent of the orthogonal geodesics,
\[ \left. \frac{\partial y^{\mu'}}{\partial X^i} \right|_{X^j=0} =e_i^{\mu'}(X^{\hat 0}). \tag{4.21}\]
Along \(X^i=0\), \(P(X)=z(X^{\hat 0})\), so
\[ \left. \frac{\partial y^{\mu'}}{\partial X^{\hat 0}} \right|_{X^j=0} =e_{\hat 0}^{\mu'}(X^{\hat 0}). \tag{4.22}\]
Therefore
\[ \left. \det\left(\frac{\partial y^{\mu'}}{\partial X^{\hat a}}\right) \right|_{\mathcal G} =\det(e_{\hat a}^{\mu'})\neq0, \tag{4.23}\]
where \(\mathcal G\) denotes the central geodesic. The inverse-function theorem guarantees a nonsingular coordinate system in a sufficiently small tubular neighborhood. Globally the construction fails when orthogonal geodesics intersect or reach conjugate points; FNC are intrinsically local.
The failure at a caustic is geometric rather than perturbative: two different spatial tangent vectors would be assigned to the same event, so the Fermi coordinates cease to be one-to-one. For cosmological applications the laboratory size is chosen well below both the curvature radius and the scale at which such intersections occur.
4.6 The Fermi conditions
Riemann normal coordinates make the connection vanish at one event. Fermi normal coordinates extend the same local-inertial property along the full central geodesic. The extension is possible because the temporal basis is the geodesic tangent and the spatial basis is parallel transported, while the off-worldline coordinate curves are themselves geodesics at fixed central proper time.
On \(\mathcal G\), the Fermi coordinate basis is the transported tetrad:
\[ \left.\frac{\partial}{\partial X^{\hat a}}\right|_{\mathcal G} =e_{\hat a}. \tag{4.24}\]
Hence
\[ \boxed{ g_{\hat a\hat b}|_{\mathcal G}=\eta_{\hat a\hat b} }. \tag{4.25}\]
For fixed \(X^{\hat 0}=\tau\) and fixed direction \(\alpha^i\), the coordinate curve
\[ X^{\hat 0}=\tau, \qquad X^i=\lambda\alpha^i \tag{4.26}\]
is precisely one of the orthogonal geodesics. Its coordinate acceleration vanishes, so the geodesic equation gives
\[ \Gamma^{\hat a}_{ij}(\tau,\lambda\alpha) \alpha^i\alpha^j=0. \tag{4.27}\]
At \(\lambda=0\), the direction \(\alpha^i\) is arbitrary and the connection is symmetric in \(i,j\). Therefore
\[ \Gamma^{\hat a}_{ij}|_{\mathcal G}=0. \tag{4.28}\]
The coordinate basis vectors on \(\mathcal G\) are parallel transported by construction. Thus
\[ 0=\nabla_{e_{\hat 0}}e_{\hat a} =\Gamma^{\hat b}_{\hat 0\hat a}|_{\mathcal G} \frac{\partial}{\partial X^{\hat b}}, \tag{4.29}\]
which implies
\[ \Gamma^{\hat b}_{\hat 0\hat a}|_{\mathcal G}=0. \tag{4.30}\]
Together with torsion freedom and Eq. 4.28,
\[ \boxed{ \Gamma^{\hat c}_{\hat a\hat b}|_{\mathcal G}=0 }, \qquad \boxed{ \partial_{\hat c}g_{\hat a\hat b}|_{\mathcal G}=0 }. \tag{4.31}\]
The second statement follows from metric compatibility. These are the Fermi conditions: the metric is rectangular and has no linear spatial dependence along the chosen geodesic.
The vanishing of first derivatives means that a pointlike experiment on the central worldline cannot detect a gravitational field through the connection. It does not imply that the metric is constant in a finite neighborhood. Second spatial derivatives remain and are fixed by curvature, so an extended experiment can measure differential acceleration.
Eq. 4.31 is the precise content of the equivalence principle for an extended experiment. What can be transformed away is \(\Gamma^{\hat c}_{\hat a\hat b}\), and it can be removed not merely at one event, as in Riemann normal coordinates, but along the entire worldline. What cannot be removed is the second derivative of the metric, because it contains the Riemann tensor, and a tensor that vanishes in one chart vanishes in all of them. This is why the gravitational field available to a local experiment begins at tidal order, and why the size of the apparatus enters: the dimensionless effect scales as \(RL^2\), so a pointlike detector measures nothing at all. Every local operator used later in these notes, from the tidal bias of Chapter 1 to the intrinsic alignment and gravitational-wave response of Chapter 5, follows from this single statement.
4.7 Geodesic deviation in the convention of these notes
Geodesic deviation gives the operational meaning of the curvature components that enter the Fermi metric. The deviation vector \(\xi^\mu\) joins neighboring geodesics at equal values of their affine parameter. Its second covariant derivative measures relative rather than absolute acceleration, so it is invariant under the removal of a uniform gravitational field.
Consider a two-parameter family of geodesics \(P(\lambda,\sigma)\). Define
\[ u\equiv\frac{\partial}{\partial\lambda}, \qquad \xi\equiv\frac{\partial}{\partial\sigma}. \tag{4.32}\]
Because \((\lambda,\sigma)\) are coordinates on the two-surface swept out by the family,
\[ [u,\xi]=0, \qquad \nabla_u\xi=\nabla_\xi u. \tag{4.33}\]
The second equality follows from the first because the connection is torsion free, so the antisymmetric part of \(\nabla u\) acting on \(\xi\) reduces to the Lie bracket.
Each \(\lambda\) curve is geodesic, \(\nabla_u u=0\). Therefore
\[ \begin{aligned} \nabla_u\nabla_u\xi &=\nabla_u\nabla_\xi u\\ &=\nabla_\xi\nabla_u u+R(u,\xi)u\\ &=R(u,\xi)u. \end{aligned} \tag{4.34}\]
In components, using Eq. 4.1,
\[ \boxed{ \frac{D^2\xi^\mu}{D\lambda^2} =-R^\mu{}_{\nu\alpha\beta} u^\nu\xi^\alpha u^\beta }. \tag{4.35}\]
The minus-sign form follows by using antisymmetry in the last two curvature indices. For neighboring freely falling particles near the central timelike geodesic,
\[ \frac{\mathrm d^2X^i}{\mathrm dT^2} =-R^i{}_{\hat 0j\hat 0}(T)X^j +\mathcal O(X^2,\dot X X). \tag{4.36}\]
The matrix \(R^i{}_{\hat 0j\hat 0}\) is the tidal acceleration per unit separation measured by the central observer. Its trace changes the volume of a small cloud of test particles, while its trace-free part shears the cloud. This same tensor will be used to describe scalar large-scale tides and gravitational waves.
4.8 Connection derivatives on the central geodesic
Since the connection itself vanishes on the central geodesic, the leading spatial variation of the connection is its first derivative. Curvature is precisely the antisymmetrized derivative of the connection at that location. The additional identities supplied by the radial geodesics determine the symmetric combinations needed for the Taylor expansion of the metric.
The quadratic metric requires first spatial derivatives of the connection. The Fermi conditions determine them algebraically from curvature, as in Sec. VII of2.
Since \(\Gamma^{\hat a}_{\hat b\hat 0}=0\) on \(\mathcal G\) for every \(X^{\hat 0}\),
\[ \Gamma^{\hat a}_{\hat b\hat 0,\hat 0}|_{\mathcal G}=0. \tag{4.37}\]
At \(\mathcal G\), where quadratic connection terms vanish, the curvature definition gives
\[ R^{\hat a}{}_{\hat b i\hat 0} =\Gamma^{\hat a}_{\hat b\hat 0,i} -\Gamma^{\hat a}_{\hat b i,\hat 0}. \tag{4.38}\]
The second term vanishes because \(\Gamma^{\hat a}_{\hat b i}=0\) along the full central geodesic. Hence
\[ \boxed{ \Gamma^{\hat a}_{\hat b\hat 0,i}|_{\mathcal G} =R^{\hat a}{}_{\hat b i\hat 0}|_{\mathcal G} }. \tag{4.39}\]
Differentiate the radial identity Eq. 4.27 with respect to \(\lambda\) at the origin:
\[ \Gamma^{\hat a}_{(ij,k)}|_{\mathcal G}=0. \tag{4.40}\]
At \(\mathcal G\),
\[ R^{\hat a}{}_{i jk} =\Gamma^{\hat a}_{ik,j} -\Gamma^{\hat a}_{ij,k}. \tag{4.41}\]
Writing the analogous equation with \(i\) and \(j\) exchanged and using Eq. 4.40 gives
\[ \boxed{ \Gamma^{\hat a}_{ij,k}|_{\mathcal G} =-\frac13 \left( R^{\hat a}{}_{i jk} +R^{\hat a}{}_{j ik} \right)_{\mathcal G} }. \tag{4.42}\]
The cyclic identity of the Riemann tensor is equivalent to the triple symmetry required by the radial geodesics.
Eq. 4.39 and Eq. 4.42 are the technical core of the Manasse–Misner construction. They convert a coordinate statement about the geodesic spray into covariant curvature coefficients. No field equation is used; the result holds in any smooth Lorentzian spacetime.
4.9 Quadratic expansion of the Fermi metric
The Fermi conditions remove the constant correction and all terms linear in the spatial coordinates. The first nontrivial metric coefficients are therefore quadratic in \(X^i\). Metric compatibility relates their coefficients to the connection derivatives just obtained, and hence to the Riemann tensor on the central geodesic.
Metric compatibility gives
\[ g_{\hat a\hat b,\hat c} =g_{\hat a\hat d} \Gamma^{\hat d}_{\hat b\hat c} +g_{\hat b\hat d} \Gamma^{\hat d}_{\hat a\hat c}. \tag{4.43}\]
Differentiating with respect to \(X^i\) and evaluating on \(\mathcal G\) yields
\[ g_{\hat a\hat b,ij}|_{\mathcal G} =\eta_{\hat a\hat d} \Gamma^{\hat d}_{\hat b i,j}|_{\mathcal G} +\eta_{\hat b\hat d} \Gamma^{\hat d}_{\hat a i,j}|_{\mathcal G}. \tag{4.44}\]
For the time–time component, Eq. 4.39 gives
\[ \begin{aligned} g_{\hat 0\hat 0,ij}|_{\mathcal G} &=-2\Gamma^{\hat 0}_{\hat 0i,j}|_{\mathcal G}\\ &=-2R_{\hat 0i\hat 0j}|_{\mathcal G}. \end{aligned} \tag{4.45}\]
For the mixed and spatial components, using Eq. 4.39 and Eq. 4.42 and the Riemann symmetries gives
\[ \begin{aligned} g_{\hat 0k,ij}|_{\mathcal G} &=-\frac23 \left(R_{\hat 0ikj}+R_{\hat 0jki}\right)_{\mathcal G},\\ g_{\ell m,ij}|_{\mathcal G} &=-\frac13 \left(R_{\ell i m j}+R_{\ell j m i}\right)_{\mathcal G}. \end{aligned} \tag{4.46}\]
The Taylor series
\[ g_{\hat a\hat b}(T,\boldsymbol{X}) =\eta_{\hat a\hat b} +\frac12g_{\hat a\hat b,ij}(T,\boldsymbol{0})X^iX^j +\mathcal O(X^3) \tag{4.47}\]
then gives the Manasse–Misner metric in our all-lowered curvature convention:
\[ \boxed{ g_{\hat 0\hat 0} =-1-R_{\hat 0i\hat 0j}(T)X^iX^j+\mathcal O(X^3) }. \tag{4.48}\]
\[ \boxed{ g_{\hat 0i} =-\frac23R_{\hat 0jik}(T)X^jX^k+\mathcal O(X^3) }. \tag{4.49}\]
\[ \boxed{ g_{ij} =\delta_{ij}-\frac13R_{ikjl}(T)X^kX^l+\mathcal O(X^3) }. \tag{4.50}\]
Curvature is evaluated on the central geodesic at the same proper time \(T=X^{\hat 0}\). Cubic terms contain covariant derivatives of curvature; their coefficients retain arbitrary proper-time dependence along the central geodesic.
The three equations encode different parts of local gravity. \(g_{\hat 0\hat 0}\) controls the leading tidal acceleration of slow particles. \(g_{\hat 0i}\) contains gravitomagnetic-type curvature and affects moving particles and clock transport. \(g_{ij}\) describes the spatial curvature of the local coordinate grid. All are fixed by the same spacetime Riemann tensor rather than by independent potentials.
The mixed and spatial coefficients require a short manipulation using pair exchange and the first Bianchi identity, which is carried out in Appendix D. That appendix also translates the index order of2, whose printed metric carries the opposite sign for the reason given in the caveat above.
As a check, the slow-particle geodesic equation from Eq. 4.48 gives
\[ \Gamma^i_{\hat 0\hat 0}=R^i{}_{\hat 0j\hat 0}X^j+\mathcal O(X^2), \qquad \ddot X^i=-R^i{}_{\hat 0j\hat 0}X^j, \tag{4.51}\]
which agrees with geodesic deviation.
4.10 What is locally measurable?
The equivalence principle removes gravitational effects that can be represented by one value of the connection along a freely falling worldline. A finite-size experiment compares neighboring worldlines and is sensitive to derivatives of that connection, namely curvature. The operator content of a local effective description must therefore be expressible in terms of curvature, locally measured matter fields, and covariant derivatives or histories of these quantities.
The connection can be set to zero on \(\mathcal G\), but curvature cannot. The electric part of the Riemann tensor seen by the observer is
\[ \mathcal R_{ij}\equiv R_{\hat 0 i\hat 0 j}. \tag{4.52}\]
Its trace and trace-free parts are
\[ \mathcal R_{ij} =\frac13\delta_{ij}\mathcal R^k{}_k +\mathcal R_{\langle ij\rangle}, \qquad \mathcal R_{\langle ij\rangle} \equiv\left(\delta_i{}^k\delta_j{}^l -\frac13\delta_{ij}\delta^{kl}\right)\mathcal R_{kl}. \tag{4.53}\]
The trace contributes isotropic geodesic focusing; the trace-free part is the shear tide. In vacuum it equals the electric Weyl tensor. In matter, the Ricci part carries local density and pressure information.
This is the relativistic completion of the Newtonian operator statement in Chapter 1. A local formation process cannot depend on a coordinate potential or a uniform acceleration. It can depend on curvature, its covariant derivatives, matter fields measured in the local tetrad, and their history along the worldline.
For nonrelativistic scalar perturbations the trace is related to the local density through the field equations, and the trace-free part reduces to the Newtonian tidal tensor. The Fermi construction explains why these are the leading gravitational operators in the bias expansion without relying on a particular cosmological gauge.
The value of the construction is not that it produces a convenient chart. It is that it settles which quantities a local process is permitted to depend on. A galaxy forming in a long-wavelength environment cannot know the value of \(\Phi\), because that value can be changed by a coordinate transformation that no local measurement can detect; it cannot know \(\partial_i\Phi\) either, since a uniform acceleration is indistinguishable from free fall. It can respond to \(\mathcal R_{ij}\), to matter fields measured in its own tetrad, and to the histories of both along its worldline. That is the entire list, and it is fixed by geometry rather than by modelling taste. The bias expansion of Chapter 1 was written down on the strength of a Newtonian version of this argument; here it is derived, and the derivation is what allows the same operator basis to be used in a relativistic setting where no preferred gauge is available.
4.11 Domain of validity
For a laboratory of characteristic size \(L\), the quadratic approximation requires
\[ |R_{\hat a\hat b\hat c\hat d}|L^2\ll1, \qquad |\nabla_{\hat e}R_{\hat a\hat b\hat c\hat d}|L^3\ll1. \tag{4.54}\]
The coordinate construction additionally requires a normal tubular neighborhood without intersections of orthogonal geodesics. FNC describe a local experiment around one timelike observer; they do not replace the past-light-cone coordinates required for cosmological observations over horizon-scale distances.
The conditions in Eq. 4.54 also state the derivative expansion explicitly. The dimensionless effect of curvature across the apparatus is \(R L^2\); variation of the tide across the apparatus begins at \(\nabla R L^3\). A long-wavelength perturbation can therefore be treated as a nearly homogeneous local tide even when its metric amplitude is not itself a local observable.
Summary
Key ideas. Fermi normal coordinates extend the local-inertial property of Riemann normal coordinates from a single event to an entire timelike geodesic. Their construction uses only parallel transport and the exponential map, so it is available in any smooth Lorentzian spacetime and invokes no field equation. Along the central worldline the metric is \(\eta_{\hat a\hat b}\) and the connection vanishes, which is the equivalence principle stated exactly; the first surviving structure is quadratic in the spatial coordinates, with coefficients built from the Riemann tensor. The chart is intrinsically local, failing where orthogonal geodesics intersect.
Counting settles in advance both where the construction must stop and what it will find. The forty first derivatives of the metric are matched by exactly forty coefficients in a coordinate change, so they can always be removed and nothing is left over; the one hundred second derivatives face only eighty, so twenty survive, and twenty is the number of independent Riemann components in four dimensions. Along a worldline rather than at a point the same forty connection components split into twenty-four killed by the radial geodesics and sixteen killed by parallel transport, again with nothing left over.
Main results.
- The degree-of-freedom count of Section 4.2, based on the chart expansion Eq. 4.2 and the metric expansion Eq. 4.3.
- The exponential-map definition Eq. 4.17 and its regular form Eq. 4.19, with regularity established by Eq. 4.23.
- The Fermi conditions Eq. 4.31: vanishing connection and vanishing first metric derivatives along \(\mathcal G\).
- Geodesic deviation Eq. 4.35, and its nonrelativistic form Eq. 4.36 in which \(R^i{}_{\hat 0j\hat 0}\) is the tidal acceleration per unit separation.
- The connection derivatives Eq. 4.39 and Eq. 4.42, which are the technical core of the construction.
- The quadratic Fermi metric Eq. 4.48, Eq. 4.49, and Eq. 4.50, and the consistency check Eq. 4.51.
- The electric Riemann tensor Eq. 4.52 and its trace split Eq. 4.53, giving isotropic focusing and shear tide.
- The domain of validity Eq. 4.54.
What the next chapter builds on. The chapter has identified the object a local experiment can measure: \(\mathcal R_{ij}=R_{\hat 0i\hat 0j}\), evaluated in the observer’s tetrad. Chapter 5 shows that three effects usually presented separately are the same object in different regimes. Its subhorizon scalar limit is the tidal operator \(K_{ij}\) assumed in Chapter 1; its trace-free part drives the intrinsic shape response that produces intrinsic alignment; and a gravitational wave contributes to it through \(-\tfrac12\ddot h_{ij}^{\mathrm{TT}}\). The chapter closes by contrasting this local response with the propagation distortion of Chapter 3, which can produce the same spin-two signal on the sky by an entirely different mechanism.