JOVANA
Explore Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

Orientation, Integration on Manifolds & the Generalized Stokes' Theorem

Top-degree forms are exactly the things you can integrate over a manifold — once you choose an orientation. We assemble the integral chart by chart, then prove the one theorem that swallows the fundamental theorem of calculus, Green's, the divergence theorem, and classical Stokes whole.

Why a form is the natural thing to integrate

From the first two guides in this rung you already know a differential k-form eats k tangent vectors and spits out a number, antisymmetrically. The payoff of that antisymmetry is exactly here: a top-degree form, an n-form on an n-dimensional manifold M, assigns a signed number to each little parallelepiped spanned by a tangent frame. That signed number IS an oriented volume element. So when you write "integral over M of omega" for an n-form omega, you are summing oriented volumes — and the antisymmetry is what makes the change-of-variables Jacobian appear automatically, with its sign.

Contrast this with how undergraduate calculus integrates a function: that needs a measure, hence a metric, to even know what "dx dy" weighs. A form needs no metric at all. This is the secret reason forms, not functions, are the right currency of integration on a bare smooth manifold — there is no canonical volume until you add geometry, but there is always a canonical pairing between an n-form and an oriented n-region. Keep this clean: integrating a function needs extra structure; integrating an n-form needs only an orientation.

Orientation: choosing a consistent sign

An orientation of M is a coherent choice of "which way is positive" in every tangent space T_p M, varying continuously with p. Concretely: at each point declare one equivalence class of ordered frames to be positively oriented (two frames agree when the linear map between them has positive determinant), and require neighboring points to agree. On a connected M there are at most two such choices — like the two faces of a sheet of paper. A manifold that admits one is called orientable; the Klein bottle and the Möbius band are the famous manifolds that admit none.

The slickest test uses forms directly: M is orientable exactly when it carries a nowhere-vanishing n-form. Such a form never changes sign as you move around, so it picks out the positive frames globally. On the Möbius band, any top-form you write down is forced to vanish somewhere — try to extend a sign all the way around the loop and it flips on you. This is why orientability is properly a statement about the top exterior power of the cotangent bundle T*M, not about pictures of arrows.

Building the integral chart by chart

Now define "integral over M of omega" for a compactly supported n-form. Inside a single oriented chart with coordinates (x^1, ..., x^n), omega looks like f dx^1 ^ ... ^ dx^n, and we just declare its integral to be the ordinary Riemann (or Lebesgue) integral of f over that coordinate patch. The miracle is consistency: if you redo it in overlapping coordinates, the pullback of omega multiplies f by the Jacobian determinant of the transition map — and on an oriented atlas that determinant is positive, so the ordinary change-of-variables formula makes the two answers agree. Orientation is precisely what kills the absolute value bars that change-of-variables usually carries.

To patch the local pieces into one global number we use a partition of unity subordinate to the oriented atlas — a family of bump functions rho_i summing to 1, each supported in one chart. Write omega = sum_i rho_i omega, integrate each piece in its own chart, and add. A short argument shows the total does not depend on which partition you chose. This is the standard "globalize a local definition" move from the manifolds rung, here doing real analytic work rather than bookkeeping.

single chart:   integral over U of (f dx^1 ^ ... ^ dx^n) := Riemann integral of f over phi(U) in R^n

overlap, transition F = psi o phi^(-1):   pullback gives  f_psi = (f_phi o F^(-1)) * det(DF)
   det(DF) > 0 on an oriented atlas  =>  no absolute value  =>  the two chart-integrals agree

global:   integral over M of omega := sum_i  integral over chart_i of (rho_i * omega),   sum_i rho_i = 1
The integral is local-in-charts, then glued by a partition of unity; orientation removes the |det| bars.

Manifolds with boundary and induced orientation

Stokes' theorem relates a region to its edge, so we need an edge. A manifold with boundary is modeled not on R^n but on the closed half-space {x^n >= 0}; its boundary dM is the set where x^n = 0, and that boundary is itself a smooth manifold one dimension down (a metric is nowhere in sight yet). Think of the closed disk (boundary a circle), the solid ball (boundary a sphere), or a cylinder (boundary two circles). The interior behaves like an ordinary manifold; only the rim is special.

Here is the subtle, sign-critical part. An orientation on M induces a specific orientation on dM by the outward-normal-first convention: at a boundary point, take an outward-pointing vector, place it first in a frame, and call the remaining (n-1) vectors positively oriented for dM when the whole list is positive for M. This rule is a convention — Lee, Spivak, and Bott-Tu all use it, but a few sources put the outward normal last, which flips the boundary orientation and therefore the sign of every boundary integral. Pick one and announce it; we use outward-first here.

The generalized Stokes' theorem

Everything has been built for one clean statement. For a compact oriented n-manifold M with boundary, and any smooth (n-1)-form omega, the generalized Stokes' theorem says: integral over M of d omega equals integral over dM of (the restriction of) omega. In one line, the exterior derivative inside is traded for the boundary outside. The proof is honestly short once the machinery is in place: use a partition of unity to reduce to a single half-space chart, where the claim collapses to the ordinary fundamental theorem of calculus applied in the x^n direction, with all other directions integrating to zero.

Let it sink in how much this unifies. With n = 1, M an interval, omega a function, d omega = f' dx and dM two endpoints with opposite signs — that is literally the fundamental theorem of calculus. With n = 2 it is Green's theorem; with n = 3 and the metric dictionary translating d into grad/curl/div, it becomes both the classical Stokes theorem and the divergence theorem. The reason this guide can say "the one operator that unifies grad, curl, and div" is that all those vector-calculus identities are this single equation read through a metric.

  1. Cover M by oriented charts and pick a partition of unity rho_i subordinate to them; write omega = sum_i rho_i omega, so it suffices to prove the theorem for a single compactly-supported piece in one chart.
  2. In an interior chart (no boundary), expand d omega in coordinates; each term integrates to zero by the fundamental theorem because the form has compact support, so the inside integral vanishes and there is no boundary to match — consistent.
  3. In a boundary chart modeled on {x^n >= 0}, all coordinate directions except x^n still integrate to zero; the x^n term gives, via the fundamental theorem, exactly the value of omega on the slice x^n = 0.
  4. Check that the outward-normal-first convention makes that slice carry the induced boundary orientation; summing over i reassembles the global identity integral over M of d omega = integral over dM of omega.

Why this is the gateway to cohomology

Stokes immediately pays a topological dividend. On a closed manifold (compact, no boundary), the right side is an integral over the empty set, hence zero: so integral over M of d omega = 0 for every omega. Therefore the integral of a closed form (recall closed and exact from guide 2) over a closed manifold cannot tell apart two forms that differ by an exact one — the difference integrates to zero. Integration descends to a pairing on equivalence classes, and those classes are exactly de Rham cohomology. The next guide makes this precise: H^k(M) measures forms that are closed modulo those that are exact.

A tiny example to anchor it: on the circle, the angle form d theta is closed but not exact (theta is not a single-valued function), and its integral around the loop is 2 pi, not zero. Stokes does not apply to make that vanish because there is no 0-form on the circle whose derivative is d theta globally — that nonzero integral is detecting the hole, and it is the seed of H^1 of the circle being one-dimensional.

Be honest about scope. We proved Stokes for smooth forms on smooth manifolds with smooth boundary, and stated it for the compact case. Real life often needs corners (a square has them), manifolds-with-corners, or weaker regularity, and each extension is a genuine theorem with its own hypotheses — not an automatic corollary. The slogan "d inside equals boundary outside" is reliable; the fine print on corners, completeness, and decay is where careful sources differ, so when you reach for Stokes on a non-compact or cornered space, check the version you are actually using.