Why a Single Theorem Governs the Boundary
The fundamental theorem of calculus says that integrating a derivative over an interval returns the original
function evaluated at the two endpoints. The interior integral of \(f'\) is determined entirely by data on the
boundary \(\{a, b\}\). This is a striking economy. To recover the total change of \(f\) across \([a, b]\) one need
not examine every point inside. The endpoints already encode it.
Stokes's theorem is the assertion that this economy is not special to intervals. On an oriented manifold of any
dimension, the integral of a derivative over the interior equals an integral over the boundary alone.
We are now in a position to make that statement precise, because the two halves of it are already in hand. On an
oriented smooth manifold we know how to form the
integral of a top-degree form,
and we know how to differentiate a form of any degree by the
exterior derivative
\(d\). The remaining ingredient is the boundary. A manifold with boundary carries a canonical orientation on its
boundary, the
Stokes orientation,
fixed precisely so that the signs come out right. Three pieces are now in place: an integral, a derivative, and an
oriented boundary. With them the relationship can be stated in one line, and the bulk of this page is spent proving
it and reading off its consequences.
What that one line earns is unification. The fundamental theorem itself is its lowest-dimensional instance, with
the endpoints \(\{a, b\}\) playing the role of the boundary. Specialized to the plane and to three-dimensional
space it reproduces the classical integral theorems of vector calculus: Green's theorem, the divergence theorem,
and the classical Stokes theorem for the curl. Each of the three was historically a separate result with a separate
proof, and here they become a single statement under different choices of form and domain. Those specializations
are taken up once the theorem is proved. The point now is that the unification is genuine, not a slogan.
Why the Result Is Worth the Machinery
It is fair to ask whether the apparatus of forms, orientations, and exterior derivatives earns its keep, when
the classical theorems can each be proved by hand. The answer is that the apparatus is what makes the theorems
one theorem. A computation tailored to surfaces in space says nothing about a region in seven dimensions, or
about a manifold that sits in no Euclidean space at all. The form-theoretic statement holds in every dimension
and on every oriented manifold, with a single proof. This generality is not idle. The theorem is needed in
exactly this form wherever integration runs over abstract manifolds. Those manifolds include orbit spaces of
Lie groups, configuration spaces, and the curved domains on which geometric models of data are built.
The boundary relationship is also the hinge of de Rham cohomology, the structure that detects, through
integration alone, whether a closed form fails to be exact or a cycle fails to bound. Several of the
corollaries below are the first visible signs of that structure.
The Theorem and Its Proof
We state the theorem in full generality and then prove it. The strategy of the proof is the one that has served the
development so far. We verify the claim first in a single coordinate model, where it reduces to the fundamental
theorem of calculus applied one variable at a time. The general case is then assembled from coordinate pieces using
a partition of unity. The model is the upper half-space, and the boundary signs that emerged when its boundary
orientation was computed are exactly the signs that will make the two sides agree.
Theorem: Stokes's Theorem
Let \(M\) be an oriented smooth \(n\)-manifold with boundary, and let \(\omega\) be a compactly supported smooth
\((n-1)\)-form on \(M\). Then
\[
\int_M d\omega = \int_{\partial M} \omega,
\]
where \(\partial M\) carries the Stokes orientation and the right-hand side is read as the integral of the
pullback \(\iota_{\partial M}^*\omega\) along the inclusion \(\iota_{\partial M} : \partial M \hookrightarrow M\).
If \(\partial M = \varnothing\), the right-hand side is interpreted as zero.
The degrees match by design. The derivative \(d\omega\) is an \(n\)-form, of exactly the degree integrable over the
\(n\)-manifold \(M\), while \(\omega\) has degree \(n-1\), exactly the degree integrable over the
\((n-1)\)-dimensional boundary. Differentiation raises degree by one and passing to the boundary lowers dimension
by one, so the two operations meet at a single equality. Compact support keeps both integrals finite even when
\(M\) is noncompact. When \(\partial M = \varnothing\) the right-hand side is zero, so the theorem then asserts
that an exact top form integrates to zero over a boundaryless manifold. That degenerate case will turn out to carry
real content.
Proof.
The proof proceeds in three steps of increasing generality: the half-space \(\mathbb{H}^n\) itself, then a form
supported in a single chart, then an arbitrary form patched together from charts. The first step carries all
the analytic content. The later steps are bookkeeping.
Step 1: The upper half-space.
Suppose first that \(M\) is the upper half-space \(\mathbb{H}^n = \{x^n \geq 0\}\) itself, with its standard
orientation. Because \(\omega\) is compactly supported, there is a number \(R \gt 0\) for which the support of
\(\omega\) lies in the box \(A = [-R, R] \times \cdots \times [-R, R] \times [0, R]\). Write \(\omega\) in
standard coordinates, using the hat to mark the omitted factor:
\[
\omega = \sum_{i=1}^{n} \omega_i\, dx^1 \wedge \cdots \wedge \widehat{dx^i} \wedge \cdots \wedge dx^n.
\]
Each \(\omega_i\) is a smooth function vanishing outside \(A\). We compute \(d\omega\) term by term. The
exterior derivative of \(\omega_i\) is \(\sum_j (\partial \omega_i / \partial x^j)\, dx^j\), so
\[
d\omega = \sum_{i, j} \frac{\partial \omega_i}{\partial x^j}\, dx^j \wedge dx^1 \wedge \cdots \wedge \widehat{dx^i}
\wedge \cdots \wedge dx^n.
\]
In the wedge \(dx^j \wedge dx^1 \wedge \cdots \wedge \widehat{dx^i} \wedge \cdots \wedge dx^n\), every
coordinate factor except \(dx^i\) already appears, so the term vanishes unless \(j = i\). When \(j = i\), the
factor \(dx^i\) sits at the front and must be moved into its natural position, passing the \(i - 1\) factors
\(dx^1, \dots, dx^{i-1}\) ahead of it. Each transposition contributes a sign, for a total of \((-1)^{i-1}\).
Hence
\[
d\omega = \sum_{i=1}^{n} (-1)^{i-1}\, \frac{\partial \omega_i}{\partial x^i}\, dx^1 \wedge \cdots \wedge dx^n.
\]
Integrating over \(\mathbb{H}^n\) means erasing the wedges and computing the iterated integral over the box
\(A\). For each term we may carry out the \(x^i\)-integration first. The fundamental theorem of calculus turns
the inner integral of \(\partial \omega_i / \partial x^i\) into a difference of boundary values of
\(\omega_i\):
\[
\int_{\mathbb{H}^n} d\omega = \sum_{i=1}^{n} (-1)^{i-1} \int \cdots \int
\Bigl[\, \omega_i\, \Bigr]_{x^i = (\text{lower})}^{x^i = (\text{upper})} \, dx^1 \cdots \widehat{dx^i} \cdots dx^n.
\]
Now the support condition does the pruning. For an index \(i \neq n\), the variable \(x^i\) ranges over the
full interval \([-R, R]\), and \(\omega_i\) vanishes at both endpoints \(x^i = \pm R\). The bracketed
difference is zero, and the entire term drops out. The only surviving term is \(i = n\), where \(x^n\) ranges
over \([0, R]\). At the upper limit \(x^n = R\) the coefficient \(\omega_n\) vanishes, but at the lower limit
\(x^n = 0\) it need not, so the difference of boundary values reduces to
\(-\,\omega_n(x^1, \dots, x^{n-1}, 0)\). Therefore
\[
\int_{\mathbb{H}^n} d\omega = (-1)^{n-1} \cdot (-1) \int_{\mathbb{R}^{n-1}} \omega_n(x^1, \dots, x^{n-1}, 0)\,
dx^1 \cdots dx^{n-1} = (-1)^{n} \int_{\mathbb{R}^{n-1}} \omega_n(\cdot, 0)\, dx^1 \cdots dx^{n-1},
\]
the factor \((-1)\) coming from the lower limit of integration.
It remains to evaluate the other side, \(\int_{\partial \mathbb{H}^n} \omega\). The boundary is
\(\{x^n = 0\}\), identified with \(\mathbb{R}^{n-1}\) through \((x^1, \dots, x^{n-1})\). Pulling \(\omega\)
back to the boundary kills every term but one. For \(i \neq n\), the factor \(dx^n\) survives in the wedge, and
the pullback of \(dx^n\) to the slice \(x^n = 0\) is zero, since \(x^n\) is constant there. Only the \(i = n\)
term survives, since its wedge \(dx^1 \wedge \cdots \wedge dx^{n-1}\) contains no \(dx^n\). So the pullback is
\(\omega_n(x^1, \dots, x^{n-1}, 0)\, dx^1 \wedge \cdots \wedge dx^{n-1}\), and the boundary integral is the
integral of \(\omega_n(\cdot, 0)\) over \(\mathbb{R}^{n-1}\). That integral is taken with the
Stokes orientation of the half-space boundary,
which differs from the standard orientation of \(\mathbb{R}^{n-1}\) by precisely the factor \((-1)^n\).
Inserting it, we obtain
\[
\int_{\partial \mathbb{H}^n} \omega = (-1)^{n} \int_{\mathbb{R}^{n-1}} \omega_n(\cdot, 0)\, dx^1 \cdots dx^{n-1}.
\]
The two computations agree, and the theorem holds on \(\mathbb{H}^n\).
Replacing \(\mathbb{H}^n\) by \(\mathbb{R}^n\) repeats the computation with one change. The variable \(x^n\)
now also ranges over the full interval \([-R, R]\), so the \(i = n\) term vanishes along with the rest. Both
sides are then zero. The left side vanishes because every boundary difference does, and the right side because
\(\mathbb{R}^n\) has no boundary, so the equality holds trivially.
Step 2: A form supported in one chart.
Let \(M\) now be an arbitrary oriented smooth manifold with boundary, and suppose \(\omega\) is compactly
supported in the domain of a single smooth chart \((U, \varphi)\), which we may take to be positively
oriented. A negatively oriented chart introduces a matching minus sign on both sides and changes nothing.
The chart carries \(U\) to an open subset of \(\mathbb{R}^n\) or of \(\mathbb{H}^n\), according to whether
\(U\) meets the boundary. Pushing the form to coordinates, the definition of the integral over a chart gives
\[
\int_M d\omega = \int_{\varphi(U)} \bigl(\varphi^{-1}\bigr)^* d\omega
= \int_{\varphi(U)} d\Bigl( \bigl(\varphi^{-1}\bigr)^*\omega \Bigr),
\]
where the second equality is the
commutation of
pullback with the exterior derivative. The coordinate form \((\varphi^{-1})^*\omega\) is a
compactly supported \((n-1)\)-form on an open subset of \(\mathbb{H}^n\) (or \(\mathbb{R}^n\)), so Step 1
applies and equates its integral over the model with the integral of \((\varphi^{-1})^*\omega\) over
\(\partial \mathbb{H}^n\). Because \(\varphi\) restricts to an orientation-preserving diffeomorphism of
\(\partial M \cap U\) onto its image in \(\partial \mathbb{H}^n\), the boundary integral in coordinates
equals \(\int_{\partial M}\omega\). That compatibility is exactly what the Stokes orientation was defined to
secure. Tracing the equalities back,
\[
\int_M d\omega = \int_{\partial M} \omega
\]
for any form supported in a single chart. When \(U\) is an interior chart, not meeting the boundary, the
\(\mathbb{R}^n\) version of Step 1 applies and both sides are zero.
Step 3: Patching with a partition of unity.
Finally let \(\omega\) be an arbitrary compactly supported \((n-1)\)-form. Cover its support by finitely many
chart domains \(U_1, \dots, U_m\), each positively or negatively oriented, and choose a smooth
partition of unity
\(\{\psi_k\}\) subordinate to this cover. Each \(\psi_k \omega\) is then a compactly supported \((n-1)\)-form
lying inside a single chart, so Step 2 applies to it. Summing over \(k\) and using that the \(\psi_k\) sum to
\(1\) on the support of \(\omega\),
\[
\begin{align*}
\int_M d\omega
&= \int_M d\!\left( \sum_k \psi_k\, \omega \right)
= \sum_k \int_M d(\psi_k\, \omega)\\\\
&= \sum_k \int_{\partial M} \psi_k\, \omega
= \int_{\partial M} \left( \sum_k \psi_k \right) \omega
= \int_{\partial M} \omega.
\end{align*}
\]
The interchange of \(d\) with the finite sum is the linearity of the exterior derivative, and the interchange
of each integral with the sum is the linearity of the integral. Since the sums are finite, no convergence
question arises. This proves the theorem in general.
Consequences and Classical Theorems
With the theorem in hand, a sequence of consequences follows almost mechanically. Several of them are nevertheless
far from trivial in content, and one is the first place we meet the idea that integration can detect the global
shape of a manifold. We take them in order of increasing depth, then close with the classical theorem of Green as a
direct specialization.
The Fundamental Theorem for Line Integrals
The first consequence is a sanity check that also recovers a classical fact. Let \(\gamma : [a, b] \to M\) be a
smooth embedding, so that its image \(S = \gamma([a,b])\) is an
embedded \(1\)-dimensional submanifold with boundary,
oriented so that \(\gamma\) is orientation-preserving. Its boundary consists of the two endpoints
\(\gamma(a)\) and \(\gamma(b)\), and the Stokes orientation assigns \(\gamma(b)\) the sign \(+\) and \(\gamma(a)\) the sign \(-\), as
the outward direction at each end dictates. A smooth function \(f \in C^\infty(M)\) is a \(0\)-form, so the theorem
applies to \(\omega = f\) on the manifold \(S\) and gives
\[
\int_S df = \int_{\partial S} f = f(\gamma(b)) - f(\gamma(a)),
\]
since integrating a \(0\)-form over an oriented finite set of points is, by definition, the signed sum of its
values.
The integral on the left is the integral of the \(1\)-form \(df\) over the curve, so this is exactly the statement
that the integral of a gradient along a path depends only on the endpoints. Reducing one dimension further, taking
\(\gamma\) to be the inclusion of \([a,b]\) into \(\mathbb{R}\), it is the ordinary fundamental theorem of
calculus. No machinery beyond the theorem itself is used. The entire content is that the boundary of a segment is
its two signed endpoints.
Two Vanishing Theorems
Two special cases occur so often that they are worth recording as named corollaries. Both are immediate from the
theorem once one tracks which side vanishes and why. Recall that a form is
closed if its exterior
derivative is zero, and exact
if it is itself the exterior derivative of another form.
Proof.
Apply the theorem to the \((n-1)\)-form \(\omega\). Compactness of \(M\) makes \(\omega\) automatically
compactly supported, so the theorem applies and \(\int_M d\omega = \int_{\partial M}\omega\). The boundary is
empty, so by the convention fixed in the statement the right-hand side is zero. Hence \(\int_M d\omega = 0\).
Proof.
Again \(\omega\) is compactly supported because \(M\) is compact, so the theorem gives
\(\int_{\partial M}\omega = \int_M d\omega\). But \(\omega\) is closed, meaning \(d\omega = 0\) identically, so
the integrand on the right vanishes and the integral is zero.
The two corollaries are mirror images. The first sends the vanishing from the empty boundary back into the
interior, and the second sends the vanishing of \(d\omega\) out to the boundary. Each isolates one side of the
equality and exploits that the other side is forced to be zero.
Integration Detects When a Form Is Not Exact
The corollaries so far have used the theorem to prove that certain integrals vanish. Read in reverse, the same
facts become a tool of an entirely different character. A single nonzero integral can certify that a form is not
exact and that a submanifold is not a boundary, both of them global conclusions drawn from one number. This is the
first appearance of the idea underlying de Rham cohomology, and it deserves to be stated carefully.
Corollary: A Nonzero Integral Obstructs Exactness and Bounding
Let \(M\) be a smooth manifold, let \(S \subseteq M\) be an oriented compact \(k\)-dimensional submanifold
without boundary, and let \(\omega\) be a closed \(k\)-form on \(M\). If
\[
\int_S \omega \neq 0,
\]
then both of the following hold:
(a) \(\omega\) is not exact on \(M\).
(b) \(S\) is not the boundary of any oriented compact \((k+1)\)-dimensional
embedded submanifold with boundary
in \(M\).
Proof.
Each part is proved by assuming the opposite and deriving that the integral would vanish, contradicting the
hypothesis. The two arguments use the two vanishing corollaries above, one each.
Part (a). Suppose, for contradiction, that \(\omega\) were exact, say \(\omega = d\eta\)
for some \((k-1)\)-form \(\eta\) on \(M\). Restrict attention to \(S\), which is compact, oriented, and has
no boundary. The restriction \(\iota_S^*\omega\) is the integrand defining \(\int_S\omega\), and
pullback commutes
with the exterior derivative, so \(\iota_S^*\omega = \iota_S^*(d\eta) = d(\iota_S^*\eta)\) is
exact on \(S\). By the corollary on integrals of exact forms over closed manifolds, applied to the closed
manifold \(S\),
\[
\int_S \omega = \int_S d(\iota_S^*\eta) = 0,
\]
contradicting \(\int_S\omega \neq 0\). Therefore \(\omega\) is not exact.
Part (b). Suppose, for contradiction, that \(S = \partial N\) for some oriented compact
\((k+1)\)-dimensional embedded submanifold with boundary \(N \subseteq M\), with \(S\) carrying the Stokes orientation
induced by \(N\). The form \(\omega\) is closed, so its restriction to \(N\) is a closed \(k\)-form on a
compact manifold with boundary. By the corollary on integrals of closed forms over boundaries, applied to
\(N\),
\[
\int_S \omega = \int_{\partial N}\omega = 0,
\]
again contradicting the hypothesis. Therefore \(S\) bounds no such \(N\).
The orientation in part (b) matters. The conclusion is that \(S\) with its given orientation is not a boundary.
Reversing the orientation of \(S\) flips the sign of \(\int_S\omega\) but not its vanishing, so the
orientation-free statement "\(S\) is not a boundary up to orientation" holds equally.
Being non-exact and failing to bound are global properties, hard to verify directly. The force of this corollary is
that it converts either one into the evaluation of a single integral, which is often a routine computation. The
next example is the canonical instance, on the punctured plane.
Example.
On the punctured plane \(\mathbb{R}^2 \setminus \{0\}\), consider the \(1\)-form
\[
\omega = \frac{x\, dy - y\, dx}{x^2 + y^2}.
\]
A direct computation of the exterior derivative shows \(d\omega = 0\), so \(\omega\) is closed. Integrating it
over the unit circle \(S^1\), parametrized counterclockwise, gives \(\int_{S^1}\omega = 2\pi \neq 0\). In polar
form \(\omega\) is exactly the angle differential \(d\theta\), whose integral around the circle is the total
angle swept.
By the corollary, the nonzero value forces two conclusions at once. First, \(\omega\) is not exact on the
punctured plane. There is no globally defined smooth function whose differential is \(\omega\), even though
\(\omega\) looks locally like \(d\theta\). The angle \(\theta\) cannot be defined consistently all the way
around the puncture. Second, \(S^1\) is not the boundary of any compact surface lying within
\(\mathbb{R}^2 \setminus \{0\}\). Any disk it would bound must enclose the missing origin and so cannot fit in
the punctured plane. A single integral has detected the hole.
Green's Theorem
The classical theorems of vector calculus are specializations of the theorem to low dimensions, obtained by
choosing \(\omega\) suitably. The simplest is Green's theorem, which is nothing more than the two-dimensional case
written out in coordinates. We state it for a compact regular domain, that is, for a compact embedded
\(2\)-dimensional submanifold with boundary sitting inside \(\mathbb{R}^2\). A closed disk and a closed
annulus are typical, and compactness makes the properness demanded by that definition automatic, since such a
subset is
properly embedded exactly when it is closed.
Theorem: Green's Theorem
Let \(D \subseteq \mathbb{R}^2\) be a compact
regular domain,
and let \(P\) and \(Q\) be smooth real-valued functions on \(D\). Then
\[
\int_D \left( \frac{\partial Q}{\partial x} - \frac{\partial P}{\partial y} \right) dx\, dy
= \int_{\partial D} P\, dx + Q\, dy,
\]
where \(\partial D\) carries the Stokes orientation. That orientation traverses each boundary curve so as to
keep \(D\) on the left, counterclockwise around an outer boundary and clockwise around the boundary of a hole.
Proof.
Apply the theorem to the \(1\)-form \(\omega = P\, dx + Q\, dy\) on \(D\). Its exterior derivative is computed
term by term. Differentiating \(P\) gives
\(dP = \frac{\partial P}{\partial x}\,dx + \frac{\partial P}{\partial y}\,dy\), and wedging with \(dx\)
annihilates the \(dx\) term while \(dy \wedge dx = -\,dx \wedge dy\), so
\[
d(P\, dx) = \frac{\partial P}{\partial y}\, dy \wedge dx = -\frac{\partial P}{\partial y}\, dx \wedge dy.
\]
Similarly \(d(Q\, dy) = \frac{\partial Q}{\partial x}\, dx \wedge dy\), the \(dy \wedge dy\) term being zero.
Adding,
\[
d\omega = \left( \frac{\partial Q}{\partial x} - \frac{\partial P}{\partial y} \right) dx \wedge dy.
\]
The theorem equates \(\int_D d\omega\) with \(\int_{\partial D}\omega\). Erasing the wedge on the left turns
\(\int_D d\omega\) into the ordinary double integral of the coefficient over \(D\), and the right-hand side is
the line integral of \(P\, dx + Q\, dy\) around the boundary. This is precisely the stated identity.
The mechanism is entirely contained in the sign \(dy \wedge dx = -\,dx \wedge dy\). The antisymmetry of the wedge
product is what produces the difference \(\partial_x Q - \partial_y P\) rather than a sum. The divergence theorem
and the classical Stokes theorem for the curl arise the same way, in three dimensions, once the correspondence
between forms and the vector-calculus operators is set up. That correspondence requires a metric and is developed
in the treatment of integration on Riemannian manifolds.