Motivation
A straight line in the plane has no acceleration. On a surface, a path may have acceleration forced by the surface's bending while having no acceleration along the surface. This motivates geodesics: their acceleration is normal to the surface, so a traveler feels no intrinsic turning.
Definitions
Geodesics and local equations
Definition 4.2.1
(Geodesic)
A curve on a regular surface, parametrized with constant nonzero speed, is a geodesic when the tangential projection of is zero.
To compute with this condition, write the second derivatives of a patch in the tangent frame and the unit normal : , with repeated indices summed.
Theorem 4.2.2
(Geodesics and local equations)
Let be a regular patch with . A curve with constant nonzero speed is a geodesic if and only if
Proof
(Split the acceleration along the frame)
The chain rule gives . Substituting the decomposition of ,
The tangential projection is the first sum. Because are linearly independent, it vanishes exactly when both coefficients vanish.
The coordinate formula depends only on the metric and its first derivatives, although this derivation does not establish that further fact.
On a sphere centered at the origin, a great circle has unit speed and , which is normal to the sphere. It is a geodesic. A great-circle segment longer than is not globally length-minimizing between its endpoints: a shorter route exists. Geodesic is a local differential condition, not a guarantee that every segment is shortest.
Theorem/proposition
Curvature, boundary and topology
Theorem 4.2.3
(Curvature, boundary and topology)
For a compact oriented smooth surface with piecewise smooth positively oriented boundary, the Gauss–Bonnet theorem states, under the usual regularity assumptions,
Here is signed geodesic curvature relative to the induced boundary orientation, are the signed exterior turning angles at corners, and is the Euler characteristic. A smooth boundary has no corner terms.
We state this global theorem without proof: a proof requires a careful treatment of parallel transport or triangulation and boundary orientations. Its local-to-global conclusion should not be confused with a definition of .
Worked examples
Example: a spherical triangle
Worked example 4.2.4
(Example: a spherical triangle)
On a sphere of radius , choose the north pole and two points on the equator separated by longitude , where . Join them by the two meridians and the shorter equatorial arc. These are great-circle segments, so along each side. The interior angles are . Since the region is a disk, , and each exterior angle is minus its interior angle. Gauss–Bonnet gives
Thus its area is . For this is one eighth of the sphere. The Euclidean angle-sum rule fails here precisely by the integrated positive curvature.
Counterexample mode
A geodesic need not be globally shortest
The equatorial arc from longitude to on a sphere of radius is a geodesic of length . The complementary arc joins the same endpoints with length . The repaired claim is local: the geodesic equation says tangential acceleration vanishes, and it does not by itself compare every competing path.
Orientations and boundary terms
The boundary orientation is induced by the chosen orientation of the surface: in the usual oriented plane the interior lies to the left as the boundary is traversed. At a smooth unit-speed boundary curve, signed geodesic curvature records the tangential turning relative to this orientation. It is not the full space curvature of the boundary. Corners contribute the signed turn from the incoming tangent to the outgoing tangent, chosen consistently with the oriented boundary.
For a positively oriented convex plane polygon, and each straight side has . Its region is a disk, so the theorem reduces to the fact that the exterior angles add to . For a nonconvex simple polygon, signed exterior angles must be used; treating every angle as a positive turn would give the wrong result at a reflex vertex. A boundary with several components also needs care: an inner boundary receives the opposite traversal direction from an outer boundary.
Common mistake
(Counting a reflex corner as a positive turn)
Take the L-shaped hexagon with vertices , traversed counterclockwise. At five vertices the tangent turns left by ; at the reflex vertex it turns right by . The signed exterior angles sum to , as the theorem requires for a flat disk with straight sides. Counting every turn as positive gives , which is wrong. The exterior angle at a corner is the signed turn from the incoming to the outgoing tangent.
Example: a planar annulus
Worked example 4.2.5
(Example: a planar annulus)
Let with and the standard orientation. The outer circle is traversed counterclockwise and the inner clockwise. The plane has ; the signed boundary integrals are and , respectively. There are no corners, and their sum is zero. This agrees with : the annulus deformation retracts onto a circle. Counting both boundary circles counterclockwise would incorrectly produce and would not use the induced orientation.
Example: a geodesic triangle in a sphere
Take a simple spherical triangle whose sides are geodesic arcs and whose interior angles are . With the disk orientation and no side curvature, the exterior angles are , , . Since , the theorem yields
This formula assumes the sides bound the chosen disk and the angles correspond to that region. Three arbitrary great circles also bound other regions, possibly with different angles; the formula does not select a region on its own. The earlier right triangle is the case , .
Why local straightness is not global minimality
On a surface, geodesic curvature zero is the intrinsic straightness condition for a unit-speed curve. Gauss–Bonnet shows that several such straight sides can enclose nonzero curvature, so even geodesic triangles need not obey Euclidean angle sums. A geodesic segment can also cease to minimize after traveling far enough, as the great-circle example demonstrates. These two statements have different logical roles: the first concerns the curvature enclosed by a region; the second compares lengths among paths with the same endpoints. Neither follows merely from drawing a curve that looks straight in a chosen coordinate chart.
The signed boundary curvature in a calculation
Definition 4.2.6
(Signed geodesic curvature)
Let be a unit-speed boundary arc on an oriented surface with unit normal , and write . The vector lies in the tangent plane, has unit length, and points to the left of the oriented tangent. With this convention the signed geodesic curvature is
The ordinary acceleration may also have a component along ; that component is not part of . On a plane, the normal component vanishes for a curve staying in the plane, and becomes the familiar signed plane curvature. On a sphere, a great circle has acceleration along the radius and therefore even though its ordinary curvature is nonzero.
Changing orientation needs care. Reversing the surface normal reverses the chosen left vector for a fixed directed curve. For a boundary of an oriented region, however, the boundary direction also changes when the surface orientation is reversed. A calculation must use the induced boundary direction and the stated sign convention consistently. It is unsafe to assign an unsigned ordinary curvature to every boundary component and expect the global formula to hold.
Example: a spherical cap
Worked example 4.2.7
(Example: a spherical cap)
On a radius- sphere, consider the cap , where , in the usual spherical coordinates and outward orientation. Traverse its boundary at increasing longitude . Its speed with respect to is , and its unit tangent is . Differentiating with respect to arc length yields
The left tangent normal is . Their dot product is . Consequently
The cap area, computed by integrating the spherical area element, is . Its Gaussian-curvature integral is therefore . The two terms sum to , as Gauss–Bonnet predicts for a disk. For a hemisphere , the boundary term vanishes because the equator is a great circle. For a cap larger than a hemisphere, is negative; the signed boundary term is allowed to be negative even though its length is positive.
Closed surfaces and what the integral records
For a closed oriented surface, the boundary terms disappear. For a sphere of any radius the integral of Gaussian curvature is : increasing the radius decreases the curvature by the same factor by which it increases the area. For a standard torus the integral is zero, since the positive outer curvature and negative inner curvature cancel. These two numbers agree with Euler characteristics two and zero respectively. The equality is global: it does not say curvature is constant on the torus, nor that the sign of its integral determines the sign at each point.
Euler characteristic can be computed from a finite triangulation as vertices minus edges plus faces. Its invariance under suitable subdivisions and the existence of triangulations in this setting are part of the topological background and are not proved here. Using a different triangulation changes the three counts but not their alternating sum. The theorem's power is that a local geometric quantity integrates to this global topological number. A proof requires more than checking a few convenient shapes, because it must control all patches and their oriented boundary contributions.
Concept lensCombinatorial
The closed-surface formula as bookkeeping of angle excesses
Granting the triangle case, take a triangulation of a closed oriented surface by small geodesic triangles with vertices, edges and faces. Each triangle has equal to its angle sum minus . All angles together total , since the angles at each vertex fill a full turn. Each face has three edges and each edge borders two faces, so . Hence
The formula stated here assumes an oriented surface and an induced boundary orientation. Versions for nonorientable surfaces require additional formulation; one should not insert a globally chosen unit normal on a surface where none exists. The regularity and corner conditions also matter: an arbitrary rough boundary need not have the curvature integral used in the statement. These qualifications specify where the equation applies, rather than weakening its conclusion within its hypotheses.
No: its total curvature is zero, so positive curvature somewhere is balanced by negative curvature elsewhere.
A natural next question
Can a torus be positively curved everywhere?
No: its total curvature is zero, so positive curvature somewhere is balanced by negative curvature elsewhere.
A torus has , so . Hence everywhere is impossible, and everywhere forces because is continuous. Conversely, everywhere on a closed oriented surface gives ; since for genus , the surface is a topological sphere.
The sign of spherical excess
For a geodesic triangle bounding a disk on a surface, the theorem gives its angle excess as the integral of Gaussian curvature over that disk. Thus positive curvature throughout the region forces the angle sum to exceed , while negative curvature throughout forces it to be less than . This comparison depends on the whole region and the chosen simple boundary, not just on the curvature at one vertex. If curvature changes sign, the integral can have either sign or vanish through cancellation.
Common mistake
(Dropping a boundary term that does not vanish)
The spherical cap calculation illustrates why one must keep boundary terms until they are actually known to vanish. Its boundary is a latitude, which is not usually a geodesic. Omitting its geodesic curvature would incorrectly make the cap's curvature integral equal to for every latitude. Only at the equator is the latitude a great circle and the boundary integral zero. Conversely, a polygon made from geodesic arcs has zero smooth-side contributions but retains its corner turns. Identify each term before setting it to zero.
The theorem accounts for the whole boundary, including every connected component and each corner where the tangent changes abruptly. For a region with a hole, treating the inner curve as though it were an outer boundary reverses its contribution. Writing the orientations before evaluating integrals is often the simplest way to prevent this sign error. The resulting signed contributions can cancel even when each individual boundary component has positive length.
Exercises
Quick check
(Why is a great circle geodesic even though its acceleration is nonzero?)
Look at the direction of .
Solution · Answer
Its acceleration points along the radius, perpendicular to the tangent plane. Its tangential component is zero.
Quick check
(What does Gauss–Bonnet say for a closed oriented compact surface with no boundary?)
Remove the terms that need a boundary.
Solution · Answer
The boundary integral and corner sum vanish, leaving . For the radius- sphere this is .
Further practice
Quick check
(Use Gauss–Bonnet to find the area of a hemisphere of radius , treating its equator as a smooth geodesic boundary.)
A hemisphere is a disk with constant curvature .
Solution · Answer
The hemisphere is a disk, so ; its smooth equator is a geodesic and contributes zero boundary curvature. Hence , giving .
Quick check
(Why do the two boundary contributions of an oriented planar annulus have opposite signs?)
Keep the region on the left of each boundary circle.
Solution · Answer
The induced orientation keeps the region on the left: the outer boundary is counterclockwise and the inner boundary clockwise. Their signed turning integrals are and .