Motivation
The condition that acceleration be normal to a surface describes geodesic motion in the surrounding space. There is also an intrinsic formulation: the metric determines an energy for paths, and stationary paths satisfy differential equations involving only its coefficients. The distinction between stationary and minimizing is essential. A critical point of a functional need not be a minimum.
Definitions
Energy and first variation
Definition 4.1.1
(Energy of a coordinate path)
Let be a regular surface patch, with metric coefficients . For a coordinate path , write . The energy of is
A fixed-endpoint variation of is a smooth family of such paths in , defined for small , with , and . Its variation field is , so . The path is stationary for energy when for every fixed-endpoint variation.
Repeated indices in the displays below are summed from one to two.
Theorem 4.1.2
(Energy and first variation)
Let be a coordinate path in a regular patch. Then is stationary for energy if and only if
These Euler–Lagrange equations are equivalent to the geodesic equations
Proof
(Integrate by parts, then invert the metric)
Differentiating under the integral and using symmetry of gives
Integrating the second term by parts has no boundary contribution, because . Hence
If the bracket vanishes, every first variation is zero. Conversely, for any smooth vanishing near and , the family stays in the open set for small , because is compact; it is therefore an admissible variation. The bracket is continuous, so the fundamental lemma of the calculus of variations forces it to vanish.
Expanding the derivative and using symmetry in yields
The inverse matrix exists because the patch is regular. Multiplying by it, which can be undone by multiplying by , produces the geodesic equations with the stated .
Proof X-ray
Where the endpoint condition and regularity enter
The proof uses three separate ingredients. The fixed endpoints are used exactly once, to remove the boundary term evaluated at and ; this does not follow merely because the original path is a geodesic. The key move is the converse direction: arbitrary variation fields vanishing near the ends must be admissible, which is where openness of and compactness of the image enter, and the fundamental lemma needs the bracket to be continuous, which is why the path is and the patch . Regularity of the patch is used only at the end, to invert the metric matrix; since that step is reversible, the equivalence survives. The argument closes with stationarity, not with minimum length, and the zero-speed constant path satisfies the same equations.
Stationarity is a local variational conclusion for variations inside this patch, with fixed endpoints and fixed parameter interval. For a regular geodesic one additionally requires nonzero initial velocity. The equations preserve speed: differentiating and substituting the equations cancels the terms. Reparametrizing an affine solution by , , gives another affine solution, but a general nonlinear reparametrization need not satisfy these equations.
Concept lensGeometric
No tangential acceleration in space
The same equations describe the space curve . By the chain rule, . Differentiating and combining the three derivative expressions gives . Therefore
which is exactly the left side of the expanded Euler–Lagrange equation. The energy equations say : the acceleration in space is normal to the surface. The extrinsic description shows how a geodesic sits in ; the intrinsic one shows that only the enter, so a local isometry carries geodesics to geodesics.
Worked examples
Example: sphere coordinates
Worked example 4.1.3
(Example: sphere coordinates)
On the radius- sphere use , in a local interval, and
Here . The energy equations reduce to
For the first equation, ; the second follows by differentiating . A meridian has constant and affine , so it is a geodesic within the chart. A nonconstant parallel has constant and nonzero constant . The first equation then requires . In the chart , leaving : only the equator is such a geodesic. The coordinates fail at the poles; that is a chart singularity, not a singularity of the sphere.
Energy, length and constant speed
The length of a piecewise path is , with . Cauchy–Schwarz applied to and the constant function one gives
For a regular continuous-speed path, equality holds exactly when speed is constant. A geometric path can be traversed at different speeds, so its energy depends on the parametrization even though its length does not under an orientation-preserving reparametrization with positive derivative. On a fixed parameter interval, constant-speed traversal minimizes energy among parametrizations of that same regular path. This comparison alone does not show that the path is shorter than other paths.
A smooth length minimizer, locally expressible in a coordinate patch and parametrized at constant speed, is stationary for energy under fixed-endpoint variations; this is a necessary condition. Conversely, an energy stationary path need not be a minimizer. A long great-circle arc is an explicit example. Comparing paths of varying speed without accounting for the preceding inequality can lead to the false conclusion that energy and length are interchangeable for arbitrary parametrizations.
Counterexample mode
An energy-stationary arc that is not shortest
Claim: a path that is stationary for energy between two points has the least length among paths joining them. On the radius- sphere, take and for , inside the chart . It runs at unit speed from to and satisfies both spherical equations, as checked in the equatorial example below, so it is stationary. Its length is . The other equatorial arc between the same endpoints passes through and has length , which is smaller. The claim fails. The correct statement is one-directional: stationarity is necessary for a constant-speed minimizer, not sufficient. On the sphere, an arc of a great circle is length-minimizing exactly when its length is at most .
Theorem/proposition
Initial-value formulation
Theorem 4.1.4
(Initial-value formulation)
Let be a regular patch. For every point and every pair of initial tangent coordinates there are and a solution of the geodesic equations with and . Two solutions with the same initial data agree on the intersection of their intervals. If , the solution has constant positive speed and is therefore regular.
Proof
(Reduction to a first-order system)
The metric coefficients are , so the are in the chart. With the geodesic equation becomes the first-order system , , whose right side is and hence locally Lipschitz. The local existence-and-uniqueness theorem for ordinary differential equations, used here without proof, gives the local solution and its uniqueness. The speed is constant along solutions, as computed above, and equals when because the metric is positive definite.
The theorem is local: a solution may leave the chosen chart, so its coordinates cease to describe it even when the geometric curve continues. A meridian reaches a pole in finite time, where the spherical coordinates above fail. On the sphere the curve continues smoothly through the pole in another chart. Neither this example nor local existence alone proves that all geodesics on every surface exist for all time.
A conserved quantity on the sphere
Let a unit-speed curve make signed angle from the unit meridian direction toward the unit parallel direction . Its parallel velocity component is , hence
The second geodesic equation therefore makes constant. This is a form of Clairaut's relation for the sphere: the radius of the parallel times the signed sine of the angle with a meridian stays fixed. For a meridian the constant is zero; for a unit-speed equator it is . The relation is necessary for geodesics but a single conserved scalar, by itself, does not replace the full system of two differential equations.
A flat example in curved coordinates
In the Euclidean plane, use polar coordinates on a patch with and a restricted angle interval. The metric is
Substituting these coefficients into the energy equations gives
The first equation contains a term involving the angular speed, and the second contains a mixed term. Yet the geometric solutions are straight lines traversed at constant speed. The coordinate values of a straight line need not be linear functions of time, because the coordinate grid itself is curved.
Common mistake
(Nonzero Christoffel symbols do not mean curvature)
From these equations one reads off and , and it is tempting to conclude that the plane is curved in these coordinates. Christoffel symbols depend on the chart: for the same plane in Cartesian coordinates every symbol is zero. A nonzero symbol may record only the way the chosen coordinate basis changes from point to point. Curvature is a separate, chart-independent quantity, and here it is zero.
A line that misses the origin
Worked example 4.1.5
(A line that misses the origin)
Fix and take the Cartesian line . Locally choose its continuous polar angle. Its polar radius is and its angular derivative is . Thus and , while . The radial equation holds. Differentiating gives , which cancels . Both equations hold although neither polar coordinate is generally affine in .
If the line passes through the origin, polar coordinates break down there. The straight geodesic does not disappear; only the chart fails. One can use Cartesian coordinates through the origin and polar coordinates elsewhere. This example separates three questions: whether a curve is regular, whether a given parametrization of the surface is regular, and whether a particular chart covers the whole curve.
The second polar equation can be written . The conserved quantity is a consequence of rotational symmetry of the metric, and it equals the signed planar angular momentum per unit mass for this parameter. When it is zero and , the angle is locally constant, giving a radial straight line. Nonzero constant value does not mean the path is a circle: the line above has .
Existence, uniqueness and the choice of speed
The local initial-value theorem says that a starting point and a tangent vector determine a geodesic for a short time. It does not say that two prescribed endpoints have a unique joining geodesic. For example, antipodal points on a sphere lie on infinitely many great circles; each great circle supplies geodesic arcs joining them. The endpoints alone therefore do not select one initial direction. Even distinct non-antipodal points can be joined by the two different arcs of their common great circle, with different lengths and initial directions.
If the initial tangent vector is multiplied by a positive constant, the same local geometric path is traversed at a rescaled constant speed. This follows by substituting an affine time change into the differential equations and invoking uniqueness for the new initial velocity. Multiplying it by a negative constant also reverses the traversal direction. In contrast, replacing the parameter by an arbitrary increasing nonlinear function generally adds an acceleration component in the tangent direction. The image can still be a geodesic path, but the new parameter need not be affine and need not satisfy the displayed equations.
Common mistake
(A geodesic image with a non-affine parameter)
The equator traversed by , for has the image of a geodesic, but it is not a solution: the second spherical equation gives . The geodesic equations constrain the parametrized curve, not only its image. Only affine changes of parameter preserve them.
The energy functional makes this parameter choice visible. A time interval is part of its definition, and two traversals of the same path at different speeds can have different energies. The length integral is unchanged by regular orientation-preserving reparametrization. Constant speed is the bridge between their variational treatments, not a property that may be assumed for every parametrization. These distinctions prevent a common circular argument: one cannot first assert every stationary path is shortest and then use that false assertion to justify the geodesic equations.
Testing a proposed spherical solution
Worked example 4.1.6
(Testing a proposed spherical solution)
Set and in the spherical equations, where is constant. The first equation holds because ; the second holds because both and vanish. The speed is , so gives a regular constant-speed equatorial geodesic. Choosing makes the parameter arc length.
A different constant latitude does not pass the first equation when the angular speed is nonzero. Checking only the second equation would overlook this obstruction: conservation of the longitude momentum is necessary but does not enforce the radial balance represented by the first equation. Both components of a vector equation must be checked. The sign of changes the travel direction, not the geometric equator or the fact that it is a geodesic.
When solving an initial-value problem in practice, first record the chart domain and the starting velocity. Solve the coordinate system only while the solution remains in that domain, and use a different chart if the geometric path crosses a coordinate boundary. An apparent singularity in the coordinate formula may reflect the chart rather than a failure of the curve.
Exercises
Quick check
(For the Euclidean metric , solve the geodesic equations.)
Every metric derivative vanishes.
Solution · Answer
Every metric derivative is zero, so every vanishes and . Thus ; nonzero gives a straight line traversed at constant speed.
Quick check
(For the sphere metric above, show that is constant along every geodesic segment in the chart. Does this by itself show the segment minimizes length?)
Use the second Euler–Lagrange equation.
Solution · Answer
The second Euler–Lagrange equation is , giving the conserved quantity after division by . It does not prove minimization: in the chart , the equatorial arc from to has length and satisfies the equations. Its complementary arc on the sphere has length , so it is shorter.
Further practice
Quick check
(A regular path has length and is traversed on an interval of length . What lower bound does the energy–length inequality give, and when is it attained?)
Rearrange .
Solution · Answer
, with equality exactly for constant speed. This compares parametrizations of a fixed path and does not assert it minimizes length among paths.
Quick check
(For a unit-speed equatorial geodesic on a sphere of radius , calculate the constant in each direction of travel.)
At the equator the tangent is parallel to a parallel circle.
Solution · Answer
At the equator and the tangent points along the parallel, so and the constants are . Reversing travel reverses the sign.