Work with smooth embedded surfaces, possibly with smooth boundary. Ordinary change of variables, Fubini and the fundamental theorem of calculus are explicit multivariable-calculus inputs. We prove the partition construction and surface Stokes, rather than import Green's or Stokes's theorem. Global triangulation existence and the planar turning theorem remain separate inputs not proved here.
Motivation
A local area element must agree when charts overlap. Unsigned area survives any coordinate reversal, while an oriented two-form changes sign. This difference explains both why area can be integrated on a nonorientable surface and why a global Stokes formula needs an orientation.
Local coordinate calculations alone also do not define an integral over an entire surface: charts can overlap. A finite smooth partition assigns weights whose sum is one, allowing local integrals to be combined without double counting. The same device reduces Stokes to a compactly supported calculation in a plane or half-plane.
Definitions
Definition 2.3.1
(Positive charts and orientation)
A chart parametrization is positive for a unit normal when points along . An oriented atlas has positive Jacobian determinants between its coordinate systems. An orientation is a choice of such an atlas.
Below we prove that for an embedded surface an orientation is equivalent to a continuous unit normal.
Definition 2.3.2
(One-forms, two-forms and the exterior derivative)
A one-form assigns a linear functional to each tangent plane, with coordinate coefficients. Its pullback is . An oriented two-form has local expression , where a coordinate change multiplies the coefficient by its signed Jacobian. Define
The coordinate invariance of this definition will be checked. The oriented area form is in positive charts. Unsigned area instead uses the density , with absolute Jacobians.
Definition 2.3.3
(Positive boundary direction)
For an oriented surface with boundary, the positive boundary tangent is chosen so that (outward tangent conormal, ) is a positive tangent basis.
In the positive half-plane chart , the outward conormal points in the negative direction, so the boundary direction is increasing . This convention fixes the signs in the theorem.
Theorem/proposition
Theorem 2.3.4
(Normals, charts and area)
A smooth embedded surface is orientable exactly when it admits a continuous unit normal. On an oriented surface the local area forms agree. Unsigned area densities agree even without an orientation.
Theorem 2.3.5
(Finite partitions and well-defined integration)
Every finite open chart cover of a compact smooth surface with boundary admits a finite smooth nonnegative partition with sum one and each support compactly contained in an assigned chart. Integrating a two-form using these weights and positive charts is independent of the chosen cover and partition.
Theorem 2.3.6
(Surface Stokes)
For a compact oriented smooth surface with smooth boundary and a one-form ,
The boundary carries the outward-first convention above. The same proof applies when finitely many corner charts are smooth coordinate quadrants; it does not assert an arbitrary global triangulation theorem.
Concept lensGeometric
The same theorem in vector-field language
For a surface with positive unit normal , restrict the one-form of a field . Then restricted to equals , and along is , so the theorem reads
In the hemisphere example has , and the left side is twice the area of the shadow disk, .
Proof sketch or proof idea
Proof
(Jacobians: normals, area and the exterior derivative)
Under with Jacobian , the chain rule gives
Thus positive transitions glue the ordered unit normals, and a chosen continuous normal lets each connected chart be oriented by reversing one coordinate if necessary (the tangential coordinate in a boundary chart). The dot product of a local ordered normal with the chosen normal takes values plus or minus one and is continuous, hence locally constant; shrinking charts makes the choice consistent. It is then a smooth normal locally. Also . With positive this is exactly the two-form transformation law; with either sign it is the density law. This proves both orientation equivalence and the area assertions.
To check , write in coordinates by the chain rule. Differentiate its new coefficient in and subtract the derivative of the coefficient. The terms involving mixed second derivatives of and cancel. The remaining terms are
precisely the two-form transformation. Thus is not dependent on a preferred chart. The same product calculation gives .
Proof
(Bumps, compactness and independence of the integral)
We construct the partition explicitly. At each point choose a chart from the given cover and a small coordinate ball, or its intersection with the boundary half-plane, whose doubled closed ball stays inside that chart domain. Restrict a smooth Euclidean nonnegative bump supported in the doubled ball and positive on the smaller one. Such a bump follows, for example, by translating and scaling the function for , extended by zero outside. All derivatives vanish at its support boundary. Pull it to the surface and extend by zero outside its chart. Compact containment makes that extension smooth, including at chart edges; a boundary half-ball simply restricts the same smooth function.
The smaller neighborhoods cover . Compactness selects finitely many bumps whose positive regions cover . Then and are smooth, nonnegative, sum to one, and have the required supports. Multiple weights may be assigned to the same original chart. No disjoint chart decomposition is being assumed.
Define as the finite sum of coordinate integrals of . Each coefficient has compact support within its chart. For another positive-chart partition , insert into each integral. On overlaps, ordinary change of variables and the two-form transformation law identify the integral of computed in either chart. Compact support permits extending its coefficient by zero outside the overlap. Summing the finite double sum proves independence of both partition and charts. The corresponding one-dimensional substitution proves that boundary one-form integration is invariant under positive boundary parametrizations. These are the only change-of-variables inputs.
Proof
(Reduce to one chart and integrate in the plane)
Now reduce Stokes by
It is enough to prove the theorem for one chart-supported form. In an interior chart extend its coefficient functions by zero to the plane. Fubini and the one-variable fundamental theorem give , since every integrated derivative has zero end values. There is no boundary contribution from this chart.
In a boundary chart extend by zero to the closed half-plane . Compact support and the same calculation give
The term integrates to zero in ; the term gives the value of at . The right side is exactly the integral along the positively directed boundary, increasing . Summing all chart-supported forms gives .
For a coordinate quadrant , the same direct computation instead gives
These are its bottom edge directed right and left edge directed down. The vertex contributes no extra line term. Bumps and the independence argument work after restriction to a quadrant, proving the stated finite-corner version when such smooth corner charts are part of the hypotheses. A reentrant corner requires a separate suitable decomposition, not an unmentioned quadrant chart.
Proof X-ray
Why only the boundary survives
The identity is used twice. First it makes well defined without overcounting overlaps. Then, the key move, removes every derivative of the weights, so splits into the chart-supported pieces . Compact support inside a chart turns each piece into a planar integral of derivatives with zero end values: in an interior chart every term vanishes, and in a half-plane chart only the lower end of the term survives. The outward-first convention is exactly the rule that makes this surviving term , the integral in the positive boundary direction. Orientation is needed so that the chartwise integrals of add up to one number; compactness is needed so that the partition is finite and every piece has compact support. Curvature, a chosen triangulation and a planar Green theorem are not inputs.
Worked examples
Worked example 2.3.7
(A Möbius strip has no global unit normal)
For use the band
with , , identifying with . Its tangent has nonzero angular component and its tangent lies in the radial/vertical plane, so they are independent. Polar angle determines modulo the indicated identification and then , making the quotient parametrization injective; compact-to-Hausdorff gives an embedding of the band. Along its center , the ordered normal is . It reverses between and , which represent the same center point. A continuous chosen normal would be either this normal or its negative, with a continuous sign taking only plus or minus one. That sign is constant along the interval, so it cannot repair the reversal. The strip is nonorientable, though its unsigned area density remains valid.
Worked example 2.3.8
(The upper hemisphere fixes the boundary sign)
On the radius- upper hemisphere with outward normal take the restriction of . Then . With colatitude and longitude , the positive area orientation gives
At the equator the positive direction is counterclockwise as viewed from above; becomes , with the same integral. The polar-coordinate pole and longitude seam have area zero, justified in regular charts by ordinary coordinate integration; alternatively omit small caps/seam strips and pass to the limit. Reversing the normal reverses both form integration and the boundary direction, retaining equality.
Worked example 2.3.9
(The sphere's area form has no global one-form potential)
A sphere of radius has area , obtained by integrating in the same spherical coordinates, with the pole/seam limit just explained. If a global one-form had , Stokes on this closed surface would make that positive integral equal to the integral over its empty boundary, namely zero. Thus no such global exists. Local potentials may exist in individual charts; the obstruction is gluing them into one global form, not differentiating a local coefficient.
Common mistakes
Common mistake
(A cross product is a local orientation choice)
A parametrization's cross product is an orientation choice, not a globally fixed normal on every surface. On a nonorientable surface these local choices cannot all agree, although unsigned area is still meaningful.
Common mistake
(Overlapping chart integrals cannot be added)
Chart integrals cannot simply be added over overlapping patches. Nor is an arbitrary open cover already a smooth partition. Both the weights and independence of the weighted integral were proved.
Common mistake
(Boundary direction is not always counterclockwise)
Boundary direction follows from the surface orientation and outward conormal. It is not assigned counterclockwise in every planar-looking picture; inner boundaries of an annulus have the opposite direction.
The compactness hypothesis of Stokes's theorem is equally essential.
Counterexample mode
Stokes fails without compactness
Claim to test: for every oriented surface. Let be the open unit disk in the plane, an oriented smooth surface with empty boundary, and . Then and , while the integral over the empty boundary is . The proof breaks at the partition step: does not have compact support in , so no finite partition reduces it to chart-supported pieces. The repaired statement requires compact, or more generally with compact support; the closed disk with its boundary circle gives on both sides.
Summary
Coordinate Jacobians distinguish oriented forms from unsigned densities. Smooth bumps and compactness give a finite partition, making integration independent of overlapping charts. The product rule cancels partition derivatives, and the fundamental theorem on interior, half-plane and specified quadrant charts proves Stokes with its boundary signs.
Exercises
Quick check
(Derive the cross-product coordinate rule and explain why a negative Jacobian reverses an ordered normal while leaving unsigned area unchanged.)
Solution
The new tangents are and . Bilinearity and antisymmetry of the cross product leave . Its sign reverses the normal when the determinant is negative. Taking its magnitude gives the absolute determinant, precisely the factor in unsigned area transformation.
Quick check
(In the partition construction explain why extending a chart bump by zero is smooth, and why dividing by is globally legitimate.)
Solution
Each bump's support is a compact subset of its chart domain relative to the surface, so every point outside that support has a neighborhood where the extension is zero or agrees with the original smooth function. The Euclidean bump and all derivatives vanish at its support boundary. At the physical surface boundary it is the restriction of a smooth ambient coordinate function. Finitely many positive regions cover , so never vanishes; its reciprocal and the normalized weights are smooth.
Quick check
(On the oriented planar annulus , compute both boundary integrals of , verify , and explain why is not the differential of one globally defined function.)
Solution
On a circle in an angular parametrization, so the outer positive integral is and the inner clockwise one is . In coordinates , , differentiation gives , hence . Their sum is zero, agreeing with Stokes. If globally, integrating along any closed parametrized circle would give zero by the one-variable chain rule, contradicting its integral . Local angular functions do not provide one global function.
Quick check
(Show that if on a compact oriented surface without boundary, then its integral vanishes. Explain why the converse has not been proved here.)
Solution
Stokes gives . A converse would require constructing a global potential from a zero integral and additional topology/regularity analysis; no such construction or general cohomology theorem has been supplied. The necessary condition alone does not prove sufficiency.
Quick check
(Verify the two signs in the quadrant formula by applying the outward-first rule separately on and . Does the corner create an additional point integral?)
Solution
At the outward direction is ; pairing it with gives the positive orientation, hence . At the outward direction is ; the positive tangent is , hence . The shared vertex has no length, and the fundamental-theorem computation produces no extra point contribution. This addresses Stokes for forms, not the separate turning-angle terms of Gauss–Bonnet.