Here we prove the ODE result that the initial-value formulation used as an input. We work with regular embedded surfaces in , using elementary multivariable calculus and completeness of the real numbers. Length-minimizing normal neighborhoods and Hopf–Rinow are not assumed or proved here.
Motivation
Deriving a differential equation does not show that it has solutions. Even a local solution does not explain what happens when its coordinates leave a chart. A further argument is needed to rule out finite-time escape from the whole surface. These are three different questions: local construction, compatibility, and continuation.
The geodesic equation is particularly suited to continuation because it preserves physical speed. On a compact surface, both position and velocity then remain in a compact part of the tangent bundle. A local existence interval can be chosen uniformly near any limiting phase point, allowing the solution to continue past a proposed finite endpoint. The proof does not turn a geodesic into a shortest path: existence for all time and minimization between endpoints are different properties.
Definitions
Let be a regular patch, open, with . Repeated indices range from one to two. The metric is and its Christoffel functions
are .
Definition 4.3.1
(Affinely parametrized geodesic)
An affinely parametrized geodesic is a curve whose tangential acceleration vanishes. In the patch, with , this is
The initial data are a point and a tangent vector , not a pair of endpoints.
Definition 4.3.2
(Phase points, maximal solutions and completeness)
The phase point of the data is . The tangent bundle is here the concrete set ; a patch gives coordinates via . A solution is maximal if no solution on a larger interval extends it. A surface is geodesically complete if the maximal affine geodesic for every initial datum is defined on all of .
Constant curves, with zero initial velocity, are included. The absence of boundary will matter in the compactness theorem.
Theorem/proposition
Theorem 4.3.3
(Local geodesic existence and uniqueness)
For every initial phase point on a regular surface without boundary, there is a unique geodesic for sufficiently small positive and negative times. Near any fixed phase point, one common positive existence time works for all sufficiently nearby initial data. Every solution has constant speed and joins uniquely across overlapping charts.
Theorem 4.3.4
(Compact phase sets prevent finite-time escape)
If the phase points of a geodesic remain in a compact subset of as time approaches a finite endpoint of its domain, the solution extends across that endpoint. Therefore every compact regular embedded surface without boundary is geodesically complete.
Proof sketch or proof idea
Proof
(Picard iteration in phase space)
Write and . On the open phase-coordinate set , the equation becomes
The vector field is . Around an initial point , choose a closed Euclidean ball contained in this open set. Continuity bounds there. Bounded first derivatives give a Lipschitz constant on the ball: integrate the derivative along the straight segment between two points, which stays in the convex ball.
Choose with and . A zero value of either bound simply removes the corresponding restriction. Consider continuous paths , with the supremum norm. They form a complete metric space: a uniformly Cauchy sequence converges pointwise in , its limit is uniform and continuous, and the closed-ball constraint passes to the limit. Define
The bound keeps the image in the ball. Also . Starting with the constant path , iterate . Successive differences are bounded by . Their geometric sum makes the iterates uniformly Cauchy. The limit is fixed by , since the Lipschitz estimate permits passage to the limit. Two fixed points satisfy , hence coincide.
The integral equation makes the fixed point and gives . Its first components satisfy , so is and solves the geodesic equation. Any other solution with the same initial values stays in the ball on a sufficiently small interval by continuity; the same contraction argument there proves local uniqueness. Thus uniqueness is not restricted to the paths originally chosen for the construction.
The uniform-time assertion follows from the same proof. Fix a closed phase ball of radius around . For starting points within radius , use paths staying within radius of their starting point. All stay in the fixed ball. Its same constants work, and the conditions , give one time for every such start. Uniform existence time, rather than differentiability with respect to initial data, is what continuation will require.
Proof
(Chart compatibility, constant speed and maximal solutions)
To verify compatibility, differentiate twice. The metric computation from the preceding chapter gives
Since is invertible, the coordinate equation is equivalent to the coordinate-independent statement that is perpendicular to the tangent plane. Hence changing charts preserves the equation. The vector determines unique coordinate velocity components because maps isomorphically onto . Local uniqueness therefore identifies solutions written in different charts.
For speed, differentiation gives
The cancellation uses and the differential equation. Thus is constant. Zero initial speed gives the constant solution by uniqueness; nonzero initial speed never vanishes.
All local extensions with the same initial data agree wherever their time intervals overlap. One way to justify propagation is to consider the set of times where their phase points agree: it is closed by continuity and open by local uniqueness, and contains the initial time. The overlap is an interval, so agreement holds throughout. Taking their union gives a well-defined solution on one maximal open interval containing zero.
Proof
(Continuation through a compact phase set)
Now suppose its right endpoint is finite and its phase stays in a compact subset near . Choose . Compactness in the ambient Euclidean space gives a subsequence of phase points converging to . In a phase chart around , these points eventually lie in the neighborhood with the common existence time proved above. Choose one with . The solution starting there exists past and agrees with the old one on their overlap by uniqueness, contradicting maximality. The argument at a left endpoint is identical with time reversed. A convergent subsequence suffices because its existence time is uniform.
Proof X-ray
Three ingredients that defeat a finite endpoint
The continuation step combines three separately proved facts. Compactness of inside , not merely boundedness in , gives an accumulation point that is a genuine phase point, so a phase chart exists around it; in the punctured plane the bounded phase points accumulate outside . The key move is the uniform existence time near : a time that depended on the starting point could shrink to zero along , and then no start would reach past . Uniqueness glues the new solution to the old. For compact surfaces, conserved speed is what places every phase point in the compact set , and the absence of boundary is what makes the phase charts open, as the Picard ball requires.
Proof
(Compact surfaces without boundary are complete)
Finally assume compact without boundary and let be the conserved speed. The set
is compact. To check closedness, the orthogonal projection onto is locally , continuous in and independent of the patch. The constraint is closed. Thus is a closed subset of the compact product of with the radius- sphere in ; when , that second factor is . Every phase point lies in . Neither endpoint of the maximal interval can be finite, proving completeness. Without boundary, phase coordinates have open domains, exactly what the Picard ball construction required.
Worked examples
Worked example 4.3.5
(A great circle continues through the poles)
On the sphere of radius , take , , and . Then
has initial data , stays on the sphere, and satisfies , which is normal. It is the unique geodesic for those data and exists for all real time. If it crosses a pole, a latitude-longitude formula may fail there, but this curve does not. For , take the constant curve instead.
Worked example 4.3.6
(A missing point gives genuine incompleteness)
On the punctured plane , the geodesic has speed one and maximal interval . Any continuous extension through would take the value , which is absent. Position and velocity remain bounded near the endpoint, but their limiting phase is outside . A straight line on the other side of the hole does not repair continuity across it.
Worked example 4.3.7
(A noncompact complete cylinder)
On the radius- cylinder,
has radial normal acceleration and constant speed . Every initial point and tangent velocity has this form for suitable constants, so uniqueness proves completeness although the cylinder is noncompact. Taking and gives a closed geodesic. A full circuit has positive length while its identical endpoints admit the zero-length constant path. Completeness does not imply global minimizing behavior.
Common mistakes
Counterexample mode
Compactness with boundary is not enough
Claim: every compact surface, with or without boundary, is geodesically complete. The closed unit disc in a plane is compact as a surface with boundary. Starting at its center with velocity produces the straight line in the interior. At time one it reaches the boundary; any continued solution of the same flat equation leaves the disc, so the maximal interval in the disc does not extend past . The claim fails. The theorem assumes no boundary, because the Picard construction needs open phase charts around every limiting phase point.
Common mistake
(Iterating position alone)
The contraction acts on paths of position and velocity together. Applying it only to position while ignoring the second-order initial velocity does not solve the stated initial-value problem. The constants bound and its derivative on a phase ball, not just the Christoffel functions on a position chart.
Common mistake
(Mistaking a chart endpoint for a geometric endpoint)
Coordinate velocity can become large while physical speed remains constant, because the coordinate basis can shrink. A chart endpoint is therefore not automatically a geometric endpoint. Conversely, switching charts cannot fill an actual missing point of the surface.
Common mistake
(Reading more into completeness than it gives)
Unique initial-value solutions do not mean unique geodesics between endpoints. Nor has a proof of completeness supplied normal neighborhoods, a distance-realizing geodesic, or Hopf–Rinow. Those require additional arguments concerning the exponential map and length.
The continuation proof shows that a maximal solution with a finite endpoint leaves every compact set for good.
A path to generalization
The escape argument for any differential equation
The continuation proof shows that a maximal solution with a finite endpoint leaves every compact set for good.
Nothing in the continuation proof used the geodesic form of . For any vector field on an open set, if a maximal solution has a finite right endpoint , then for every compact there is with for , because the proof needed only a sequence of times in . Without conserved speed, escape can happen through the velocity: , has the solution , which blows up at .
Summary
Picard iteration constructs and uniquely determines a local phase-space solution with a uniform nearby existence time. Tangential acceleration makes the equation independent of charts, and the metric identity preserves speed. Compact phase confinement rules out finite-time escape. A compact surface without boundary provides precisely such confinement for each speed, while holes and boundaries explain the hypotheses.
Exercises
Quick check
(If and , bound the error after Picard iterations. What guarantees error below using this bound?)
Solution
Summing successive differences from iteration onward gives . For , this is ; gives only , which does not certify the requested error. The estimate concerns both position and velocity in the chosen phase norm, not a global error guarantee after leaving the chart.
Quick check
(Let be a nonconstant affine geodesic and , where is with nonzero derivative. Prove that is affine geodesic exactly when . What changes for a constant curve?)
Solution
Differentiation gives . The first term is normal to the surface and the second tangent. Since never vanishes, tangential acceleration is zero exactly when . These are affine time changes on the connected parameter interval. For a constant curve , any such reparametrization still gives the same constant geodesic, so the necessity fails.
Quick check
(Explain exactly why bounded ambient position and velocity do not imply the compact-phase continuation hypothesis. Use the punctured-plane example and identify the missing closedness.)
Solution
A bounded phase sequence has a convergent subsequence in ambient , but its limit need not belong to . Here , whose base point is missing. The closure in is compact but not contained in ; the phase set is not closed there. The theorem requires a compact subset of , ensuring an actual phase chart at the limit.
Quick check
(The open half-plane has global patch . Compute the geodesic equation for and verify that , has physical speed one for . Does the smooth metric on all coordinate imply completeness?)
Solution
The metric is , with and all other symbols zero. Thus , . The proposed function has and , so it solves the equation. Its speed squared is . Yet reaches the omitted boundary line at time one. The coefficients are smooth and positive definite at every finite coordinate point, but escapes every compact coordinate region. Smoothness alone gives local existence, not a global time interval.
Quick check
(On a sphere, show that infinitely many distinct unit-speed geodesics go from to in time . Explain why this is consistent with initial-value uniqueness.)
Solution
For each unit tangent vector at , the sphere formula with gives and . The unit tangent circle supplies infinitely many choices, and their initial derivatives are different, so these are different parametrized geodesics. Each individual initial datum still determines exactly one of them. Endpoint data have not specified the initial direction.