aDifferential forms provide a modern, coordinate-free framework for calculus on manifolds, unifying classical vector analysis with the language of algebraic topology. Rather than working with gradient, curl, and divergence as separate operations on vector fields, this course develops a single geometric object—the differential form—whose behavior under integration, differentiation, and pullback captures all of multivariable calculus in a unified system. The course explores how forms encode information about smooth spaces and reveals deep connections between the local analytic properties of a manifold (captured by the [exterior derivative](/theorems/1525)) and its global topological structure. The chapters progress systematically from foundations to applications. We begin with exterior algebra, the algebraic machinery underlying forms, then introduce differential forms concretely on open subsets of ℝⁿ where calculus intuition is strongest. The [exterior derivative](/theorems/1525) emerges as the natural generalization of classical differentiation, and the pullback operation shows how forms behave under smooth maps. Once these tools are mastered, we extend the theory to smooth manifolds—abstract curved spaces where these ideas become essential—and develop integration via orientation and the generalised Stokes theorem. The Poincaré lemma establishes when closed forms are exact, leading naturally to de Rham cohomology: a way of extracting topological invariants from the algebra of forms. De Rham's theorem, the course's central result, reveals that the cohomology computed from differential forms coincides exactly with singular cohomology, a fundamental topological invariant. This identification shows that smooth and purely topological information are intimately related, and the final chapters explore how this perspective illuminates problems across differential geometry, physics, and topology. Throughout, the framework develops the conceptual tools that professional mathematicians and physicists use daily to understand smooth manifolds. # Introduction This opening chapter sets the perspective for the course. Differential forms give a single language for line integrals, surface flux, volume integration, change of variables, and the boundary terms that appear in Stokes-type theorems. The first aim is to replace several coordinate-dependent vector calculus constructions by one functorial object. The second aim is to understand why the equation $d\omega = 0$ carries topological information about the space on which $\omega$ lives. The course starts from linear algebra on finite-dimensional real vector spaces and then moves to smooth manifolds. Smooth manifolds will be used as spaces that locally look like open subsets of $\mathbb R^n$, so the early chapters develop every construction first on open sets. Later chapters globalise these constructions by checking that they behave correctly under change of coordinates. ## Why Forms Replace Vector Calculus Notation What goes wrong if line integrals, surface integrals, and volume integrals are treated as unrelated operations? The formulas of vector calculus depend heavily on the coordinates chosen to write them down. A line integral such as $P\,dx + Q\,dy$ does not transform like a vector field; it transforms by pullback along parametrisations. Differential forms are designed so that the integrand already contains the transformation rule needed for integration. [definition: Differential Form On An Open Set] Let $U \subset \mathbb R^n$ be open. A differential $k$-form on $U$ is a smooth assignment $\omega$ sending each point $x \in U$ to an alternating $k$-[linear map](/page/Linear%20Map) \begin{align*} \omega_x : (\mathbb R^n)^k \to \mathbb R. \end{align*} The [vector space](/page/Vector%20Space) of smooth $k$-forms on $U$ is denoted $\Omega^k(U)$. For $k=0$, set $\Omega^0(U)=C^\infty(U)$. [/definition] The word alternating means that the value changes sign when two input vectors are exchanged and is zero when two input vectors are equal. Thus a $1$-form eats one tangent vector, a $2$-form eats an ordered pair of tangent vectors, and a top-degree $n$-form eats an ordered basis of $\mathbb R^n$. [example: Line Integral As A One Form] Let $U\subset \mathbb R^2$ be open, and let \begin{align*} \omega=P\,dx+Q\,dy\in\Omega^1(U), \end{align*} where $P,Q\in C^\infty(U)$. Let $\gamma:[a,b]\to U$ be a smooth curve, and write \begin{align*} \gamma(t)=(\gamma_1(t),\gamma_2(t)),\qquad \dot\gamma(t)=(\dot\gamma_1(t),\dot\gamma_2(t)). \end{align*} For $p\in U$ and $v=(v_1,v_2)\in\mathbb R^2$, the coordinate covectors satisfy \begin{align*} dx_p(v)=v_1,\qquad dy_p(v)=v_2. \end{align*} The integral of $\omega$ along $\gamma$ is \begin{align*} \int_\gamma\omega =\int_a^b\left(P(\gamma(t))\dot\gamma_1(t)+Q(\gamma(t))\dot\gamma_2(t)\right)\,dt. \end{align*} By the definition of integrating a $1$-form over a parametrised curve, \begin{align*} \int_\gamma\omega =\int_a^b \omega_{\gamma(t)}(\dot\gamma(t))\,dt. \end{align*} For each $t\in[a,b]$, evaluate the covector $\omega_{\gamma(t)}$ on the velocity vector: \begin{align*} \omega_{\gamma(t)}(\dot\gamma(t)) &=\left(P(\gamma(t))\,dx_{\gamma(t)}+Q(\gamma(t))\,dy_{\gamma(t)}\right)(\dot\gamma(t)) \\ &=P(\gamma(t))\,dx_{\gamma(t)}(\dot\gamma(t)) +Q(\gamma(t))\,dy_{\gamma(t)}(\dot\gamma(t)) \\ &=P(\gamma(t))\,dx_{\gamma(t)}(\dot\gamma_1(t),\dot\gamma_2(t)) +Q(\gamma(t))\,dy_{\gamma(t)}(\dot\gamma_1(t),\dot\gamma_2(t)) \\ &=P(\gamma(t))\dot\gamma_1(t)+Q(\gamma(t))\dot\gamma_2(t). \end{align*} Substituting this pointwise identity into the defining integral gives \begin{align*} \int_\gamma\omega &=\int_a^b \omega_{\gamma(t)}(\dot\gamma(t))\,dt \\ &=\int_a^b\left(P(\gamma(t))\dot\gamma_1(t)+Q(\gamma(t))\dot\gamma_2(t)\right)\,dt. \end{align*} The parametrised curve contributes the velocity vector $\dot\gamma(t)$, and the $1$-form contributes the covector $P(\gamma(t))\,dx_{\gamma(t)}+Q(\gamma(t))\,dy_{\gamma(t)}$. Feeding the velocity into this covector produces exactly the coordinate integrand $P(\gamma(t))\dot\gamma_1(t)+Q(\gamma(t))\dot\gamma_2(t)$, so the ordinary planar line integral is naturally the integral of a $1$-form. [/example] Coordinate notation is useful because it gives a basis for all forms on an open subset of Euclidean space. If $x=(x_1,\dots,x_n)$ are the standard coordinates, then $dx_i$ denotes the $i$-th coordinate covector. [example: Coordinate Basis For Forms] Let $U\subset \mathbb R^n$ be open, and let \begin{align*} \omega\in\Omega^k(U). \end{align*} Write $e_1,\dots,e_n$ for the standard basis of $\mathbb R^n$, and write $dx_1,\dots,dx_n$ for the dual coordinate covectors, so \begin{align*} dx_i(e_j)= \begin{cases} 1,& i=j,\\ 0,& i\ne j. \end{cases} \end{align*} For a strictly increasing multi-index $I=(i_1,\dots,i_k)$, set \begin{align*} dx_I:=dx_{i_1}\wedge\cdots\wedge dx_{i_k}. \end{align*} There are unique smooth functions $a_I\in C^\infty(U)$ such that \begin{align*} \omega=\sum_{1\le i_1<\cdots<i_k\le n} a_{i_1\cdots i_k}\,dx_{i_1}\wedge\cdots\wedge dx_{i_k}. \end{align*} For $n=3$ and $k=2$, this becomes \begin{align*} \omega=A\,dy\wedge dz+B\,dz\wedge dx+C\,dx\wedge dy. \end{align*} Fix $x\in U$. Since $\omega_x$ is an alternating $k$-[linear map](/page/Linear%20Map), its values on arbitrary vectors are determined by its values on ordered $k$-tuples of basis vectors. Define \begin{align*} a_{i_1\cdots i_k}(x):=\omega_x(e_{i_1},\dots,e_{i_k}) \end{align*} for every $1\le i_1<\cdots<i_k\le n$. Now evaluate the coordinate wedge $dx_I$ on an ordered basis tuple $(e_{j_1},\dots,e_{j_k})$. By the defining determinant formula for the wedge of coordinate covectors, \begin{align*} dx_I(e_{j_1},\dots,e_{j_k}) &=(dx_{i_1}\wedge\cdots\wedge dx_{i_k})(e_{j_1},\dots,e_{j_k})\\ &=\det\left(dx_{i_p}(e_{j_q})\right)_{1\le p,q\le k}. \end{align*} If $(j_1,\dots,j_k)=(i_1,\dots,i_k)$, the matrix is the identity matrix, so \begin{align*} dx_I(e_{i_1},\dots,e_{i_k})=\det(I_k)=1. \end{align*} If some $j_q$ is not one of $i_1,\dots,i_k$, then the $q$-th column of the matrix has all entries $0$, so \begin{align*} dx_I(e_{j_1},\dots,e_{j_k})=0. \end{align*} If $(j_1,\dots,j_k)$ is a permutation of $(i_1,\dots,i_k)$, then the matrix is the corresponding permutation matrix, so \begin{align*} dx_I(e_{j_1},\dots,e_{j_k})=\operatorname{sgn}(j_1,\dots,j_k). \end{align*} Set \begin{align*} \widetilde\omega_x =\sum_{1\le i_1<\cdots<i_k\le n} a_{i_1\cdots i_k}(x)\,dx_{i_1}\wedge\cdots\wedge dx_{i_k}. \end{align*} For every strictly increasing tuple $J=(j_1,\dots,j_k)$, \begin{align*} \widetilde\omega_x(e_{j_1},\dots,e_{j_k}) &=\sum_I a_I(x)\,dx_I(e_{j_1},\dots,e_{j_k})\\ &=a_J(x)\\ &=\omega_x(e_{j_1},\dots,e_{j_k}). \end{align*} Both $\widetilde\omega_x$ and $\omega_x$ are alternating and $k$-linear, so equality on the ordered basis tuples gives equality on all $k$ input vectors. Hence \begin{align*} \omega_x=\sum_{1\le i_1<\cdots<i_k\le n} a_{i_1\cdots i_k}(x)\,dx_{i_1}\wedge\cdots\wedge dx_{i_k}. \end{align*} Since this holds for every $x\in U$, \begin{align*} \omega=\sum_{1\le i_1<\cdots<i_k\le n} a_{i_1\cdots i_k}\,dx_{i_1}\wedge\cdots\wedge dx_{i_k}. \end{align*} The functions $a_I$ are smooth because they are the coordinate coefficient functions of the smooth form $\omega$. For uniqueness, suppose also that \begin{align*} \omega=\sum_I b_I\,dx_I. \end{align*} Evaluating both sides at $x$ on $(e_{i_1},\dots,e_{i_k})$ gives \begin{align*} \omega_x(e_{i_1},\dots,e_{i_k}) &=\sum_J b_J(x)\,dx_J(e_{i_1},\dots,e_{i_k})\\ &=b_{i_1\cdots i_k}(x). \end{align*} But by the definition of $a_{i_1\cdots i_k}$, \begin{align*} \omega_x(e_{i_1},\dots,e_{i_k})=a_{i_1\cdots i_k}(x). \end{align*} Therefore \begin{align*} a_{i_1\cdots i_k}(x)=b_{i_1\cdots i_k}(x) \end{align*} for every $x\in U$ and every strictly increasing multi-index, so the coefficients are unique. For $n=3$ and $k=2$, the strictly increasing pairs are \begin{align*} (1,2),\qquad (1,3),\qquad (2,3). \end{align*} Thus every $2$-form on $U\subset\mathbb R^3$ has the form \begin{align*} \omega=a_{12}\,dx\wedge dy+a_{13}\,dx\wedge dz+a_{23}\,dy\wedge dz. \end{align*} Using $dx\wedge dz=-dz\wedge dx$, this can be rewritten as \begin{align*} \omega &=a_{23}\,dy\wedge dz-a_{13}\,dz\wedge dx+a_{12}\,dx\wedge dy. \end{align*} If we set \begin{align*} A=a_{23},\qquad B=-a_{13},\qquad C=a_{12}, \end{align*} then \begin{align*} \omega=A\,dy\wedge dz+B\,dz\wedge dx+C\,dx\wedge dy. \end{align*} Coordinate wedge products form the natural coordinate basis for differential forms. The coefficients are not arbitrary decorations: each coefficient is recovered by feeding $\omega_x$ the corresponding ordered coordinate vectors. In dimension $3$, the three basic $2$-forms $dy\wedge dz$, $dz\wedge dx$, and $dx\wedge dy$ record the three oriented coordinate planes. [/example] ## The Operations This Course Builds Which operations on integrands survive change of coordinates and still remember orientation, dimension, and boundary? The course develops three fundamental constructions: the wedge product, the [exterior derivative](/theorems/1525), and pullback. Together they replace dot products, cross products, gradients, curls, divergences, and Jacobian determinants by coordinate-free operations. [definition: Wedge Product] Let $U \subset \mathbb R^n$ be open. The wedge product is a bilinear operation \begin{align*} \wedge : \Omega^p(U)\times \Omega^q(U)\to \Omega^{p+q}(U) \end{align*} whose value at each point is the exterior product of alternating covectors. On coordinate $1$-forms it is determined by multilinearity and the relations \begin{align*} dx_i\wedge dx_j=-dx_j\wedge dx_i,\qquad dx_i\wedge dx_i=0. \end{align*} [/definition] The wedge product encodes signed area, signed volume, and higher-dimensional oriented content. The relation $dx_i\wedge dx_i=0$ is the algebraic expression of the fact that a parallelogram with two equal directions has zero area. [definition: Exterior Derivative] Let $U \subset \mathbb R^n$ be open. The [exterior derivative](/theorems/1525) is the family of linear maps \begin{align*} d: \Omega^k(U)\to \Omega^{k+1}(U) \end{align*} defined in coordinates by \begin{align*} d\left(\sum_I a_I\,dx_{i_1}\wedge\cdots\wedge dx_{i_k}\right) =\sum_I\sum_{j=1}^n \frac{\partial a_I}{\partial x_j}\,dx_j\wedge dx_{i_1}\wedge\cdots\wedge dx_{i_k}, \end{align*} where $I=(i_1,\dots,i_k)$ ranges over strictly increasing multi-indices. [/definition] For a function $f\in C^\infty(U)$, this gives $df=\sum_i (\partial f/\partial x_i)\,dx_i$. For a $1$-form on $\mathbb R^3$, it packages the curl; for a $2$-form on $\mathbb R^3$, it packages the divergence. [quotetheorem:1525] The identity $d^2=0$ is the algebraic seed of de Rham cohomology. It says that every form produced by an [exterior derivative](/theorems/1525) automatically satisfies a compatibility equation. [example: Curl And Divergence Inside Exterior Derivative] Work on $\mathbb R^3$ with standard coordinates $(x,y,z)$ and coordinate $1$-forms $dx,dy,dz$. Let $P,Q,R,A,B,C\in C^\infty(\mathbb R^3)$, and set \begin{align*} \alpha=P\,dx+Q\,dy+R\,dz \end{align*} and \begin{align*} \beta=A\,dy\wedge dz+B\,dz\wedge dx+C\,dx\wedge dy. \end{align*} We use the wedge relations \begin{align*} dx\wedge dx=dy\wedge dy=dz\wedge dz=0 \end{align*} and \begin{align*} dy\wedge dx=-dx\wedge dy,\qquad dz\wedge dy=-dy\wedge dz,\qquad dx\wedge dz=-dz\wedge dx. \end{align*} The [exterior derivative](/theorems/1525) of the $1$-form $\alpha$ records the curl components: \begin{align*} d\alpha &=\left(\frac{\partial R}{\partial y}-\frac{\partial Q}{\partial z}\right)dy\wedge dz +\left(\frac{\partial P}{\partial z}-\frac{\partial R}{\partial x}\right)dz\wedge dx +\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy. \end{align*} The [exterior derivative](/theorems/1525) of the $2$-form $\beta$ records the divergence component: \begin{align*} d\beta=\left(\frac{\partial A}{\partial x}+\frac{\partial B}{\partial y}+\frac{\partial C}{\partial z}\right)dx\wedge dy\wedge dz. \end{align*} By linearity of $d$ and the coordinate definition of the [exterior derivative](/theorems/1525), \begin{align*} d\alpha &=d(P\,dx)+d(Q\,dy)+d(R\,dz)\\ &=\left(\frac{\partial P}{\partial x}dx+\frac{\partial P}{\partial y}dy+\frac{\partial P}{\partial z}dz\right)\wedge dx\\ &\quad+\left(\frac{\partial Q}{\partial x}dx+\frac{\partial Q}{\partial y}dy+\frac{\partial Q}{\partial z}dz\right)\wedge dy\\ &\quad+\left(\frac{\partial R}{\partial x}dx+\frac{\partial R}{\partial y}dy+\frac{\partial R}{\partial z}dz\right)\wedge dz. \end{align*} Expand each wedge product term by term: \begin{align*} d\alpha &=\frac{\partial P}{\partial x}dx\wedge dx +\frac{\partial P}{\partial y}dy\wedge dx +\frac{\partial P}{\partial z}dz\wedge dx\\ &\quad+\frac{\partial Q}{\partial x}dx\wedge dy +\frac{\partial Q}{\partial y}dy\wedge dy +\frac{\partial Q}{\partial z}dz\wedge dy\\ &\quad+\frac{\partial R}{\partial x}dx\wedge dz +\frac{\partial R}{\partial y}dy\wedge dz +\frac{\partial R}{\partial z}dz\wedge dz. \end{align*} The repeated-coordinate terms vanish, so \begin{align*} d\alpha &=\frac{\partial P}{\partial y}dy\wedge dx +\frac{\partial P}{\partial z}dz\wedge dx +\frac{\partial Q}{\partial x}dx\wedge dy +\frac{\partial Q}{\partial z}dz\wedge dy\\ &\quad+\frac{\partial R}{\partial x}dx\wedge dz +\frac{\partial R}{\partial y}dy\wedge dz. \end{align*} Rewrite every term in the ordered $2$-form basis $dy\wedge dz$, $dz\wedge dx$, $dx\wedge dy$: \begin{align*} d\alpha &=-\frac{\partial P}{\partial y}dx\wedge dy +\frac{\partial P}{\partial z}dz\wedge dx +\frac{\partial Q}{\partial x}dx\wedge dy -\frac{\partial Q}{\partial z}dy\wedge dz\\ &\quad-\frac{\partial R}{\partial x}dz\wedge dx +\frac{\partial R}{\partial y}dy\wedge dz. \end{align*} Collect coefficients of each basis wedge: \begin{align*} d\alpha &=\left(\frac{\partial R}{\partial y}-\frac{\partial Q}{\partial z}\right)dy\wedge dz +\left(\frac{\partial P}{\partial z}-\frac{\partial R}{\partial x}\right)dz\wedge dx +\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy. \end{align*} Now compute $d\beta$. By linearity and the coordinate definition of $d$, \begin{align*} d\beta &=d(A\,dy\wedge dz)+d(B\,dz\wedge dx)+d(C\,dx\wedge dy)\\ &=\left(\frac{\partial A}{\partial x}dx+\frac{\partial A}{\partial y}dy+\frac{\partial A}{\partial z}dz\right)\wedge dy\wedge dz\\ &\quad+\left(\frac{\partial B}{\partial x}dx+\frac{\partial B}{\partial y}dy+\frac{\partial B}{\partial z}dz\right)\wedge dz\wedge dx\\ &\quad+\left(\frac{\partial C}{\partial x}dx+\frac{\partial C}{\partial y}dy+\frac{\partial C}{\partial z}dz\right)\wedge dx\wedge dy. \end{align*} Expand term by term: \begin{align*} d\beta &=\frac{\partial A}{\partial x}dx\wedge dy\wedge dz +\frac{\partial A}{\partial y}dy\wedge dy\wedge dz +\frac{\partial A}{\partial z}dz\wedge dy\wedge dz\\ &\quad+\frac{\partial B}{\partial x}dx\wedge dz\wedge dx +\frac{\partial B}{\partial y}dy\wedge dz\wedge dx +\frac{\partial B}{\partial z}dz\wedge dz\wedge dx\\ &\quad+\frac{\partial C}{\partial x}dx\wedge dx\wedge dy +\frac{\partial C}{\partial y}dy\wedge dx\wedge dy +\frac{\partial C}{\partial z}dz\wedge dx\wedge dy. \end{align*} Every term with a repeated coordinate factor vanishes: \begin{align*} d\beta &=\frac{\partial A}{\partial x}dx\wedge dy\wedge dz +\frac{\partial B}{\partial y}dy\wedge dz\wedge dx +\frac{\partial C}{\partial z}dz\wedge dx\wedge dy. \end{align*} The remaining two cyclic permutations have positive sign: \begin{align*} dy\wedge dz\wedge dx &=-dy\wedge dx\wedge dz\\ &=dx\wedge dy\wedge dz, \end{align*} and \begin{align*} dz\wedge dx\wedge dy &=-dx\wedge dz\wedge dy\\ &=dx\wedge dy\wedge dz. \end{align*} Therefore \begin{align*} d\beta &=\frac{\partial A}{\partial x}dx\wedge dy\wedge dz +\frac{\partial B}{\partial y}dx\wedge dy\wedge dz +\frac{\partial C}{\partial z}dx\wedge dy\wedge dz\\ &=\left(\frac{\partial A}{\partial x}+\frac{\partial B}{\partial y}+\frac{\partial C}{\partial z}\right)dx\wedge dy\wedge dz. \end{align*} Applying $d$ to a $1$-form on $\mathbb R^3$ produces the three signed curl coefficients in the $2$-form basis $dy\wedge dz$, $dz\wedge dx$, $dx\wedge dy$. Applying $d$ to a $2$-form produces the single divergence coefficient multiplying the volume form $dx\wedge dy\wedge dz$. Thus curl and divergence are the same operation, the [exterior derivative](/theorems/1525), applied in adjacent degrees. [/example] ## Integration And Orientation Why does a change of parametrisation sometimes preserve an integral and sometimes change its sign? Integration of forms depends on orientation. A top-degree form measures signed volume, so reversing the order of a basis reverses the sign of the measured volume. [definition: Orientation Of A Vector Space] Let $V$ be an $n$-dimensional real [vector space](/page/Vector%20Space). An orientation of $V$ is a choice of one of the two classes of ordered bases of $V$, where two ordered bases are in the same class when the change-of-basis matrix between them has positive determinant. [/definition] On a manifold, an orientation is a smoothly varying choice of orientation on every tangent space. This condition is exactly what lets local integrals in coordinate charts combine into a global integral. [definition: Integral Of A Top Degree Form On A Coordinate Domain] Let $U\subset\mathbb R^n$ be open and let $\omega=f\,dx_1\wedge\cdots\wedge dx_n\in\Omega^n(U)$, where $f\in C_c^\infty(U)$. The integral of $\omega$ over $U$ with the standard orientation is \begin{align*} \int_U \omega := \int_U f\,d\mathcal L^n. \end{align*} [/definition] The determinant appears because it is the top-degree alternating form applied to the columns of a Jacobian matrix. This is why differential forms absorb the Jacobian factor in the change of variables formula. [quotetheorem:3554] This theorem is the bridge from multivariable calculus to integration on manifolds. Once the integral is independent of the oriented coordinate chart, it can be defined by choosing charts, integrating locally, and summing with a [partition of unity](/page/Partition%20of%20Unity). [example: Polar Coordinates And The Area Form] Let \begin{align*} F:(0,\infty)\times(0,2\pi)\to\mathbb R^2\setminus\{(x,0):x\ge0\} \end{align*} be the polar coordinate map \begin{align*} F(r,\theta)=(r\cos\theta,r\sin\theta). \end{align*} Write $x,y$ for the standard coordinates on $\mathbb R^2$, and write $r,\theta$ for the standard coordinates on $(0,\infty)\times(0,2\pi)$. The standard area form on $\mathbb R^2$ is \begin{align*} dx\wedge dy. \end{align*} The pullback of the standard area form is \begin{align*} F^*(dx\wedge dy)=r\,dr\wedge d\theta. \end{align*} Thus the polar Jacobian factor $r$ is already contained in the pullback of the area form. By definition of pullback on coordinate functions, \begin{align*} x\circ F(r,\theta)&=r\cos\theta,\\ y\circ F(r,\theta)&=r\sin\theta. \end{align*} Therefore \begin{align*} F^*(dx)&=d(x\circ F)=d(r\cos\theta),\\ F^*(dy)&=d(y\circ F)=d(r\sin\theta). \end{align*} Using the product rule for differentials, \begin{align*} d(r\cos\theta) &=\cos\theta\,dr+r\,d(\cos\theta)\\ &=\cos\theta\,dr-r\sin\theta\,d\theta, \end{align*} and \begin{align*} d(r\sin\theta) &=\sin\theta\,dr+r\,d(\sin\theta)\\ &=\sin\theta\,dr+r\cos\theta\,d\theta. \end{align*} By compatibility of pullback with wedge products, \begin{align*} F^*(dx\wedge dy) &=F^*(dx)\wedge F^*(dy)\\ &=d(r\cos\theta)\wedge d(r\sin\theta)\\ &=(\cos\theta\,dr-r\sin\theta\,d\theta) \wedge(\sin\theta\,dr+r\cos\theta\,d\theta). \end{align*} Expand by bilinearity of the wedge product: \begin{align*} F^*(dx\wedge dy) &=\cos\theta\sin\theta\,dr\wedge dr +r\cos^2\theta\,dr\wedge d\theta\\ &\quad-r\sin^2\theta\,d\theta\wedge dr -r^2\sin\theta\cos\theta\,d\theta\wedge d\theta. \end{align*} The repeated wedge products vanish: \begin{align*} dr\wedge dr=0,\qquad d\theta\wedge d\theta=0. \end{align*} Also \begin{align*} d\theta\wedge dr=-dr\wedge d\theta. \end{align*} Substituting these wedge relations gives \begin{align*} F^*(dx\wedge dy) &=r\cos^2\theta\,dr\wedge d\theta -r\sin^2\theta(-dr\wedge d\theta)\\ &=r\cos^2\theta\,dr\wedge d\theta +r\sin^2\theta\,dr\wedge d\theta\\ &=r(\cos^2\theta+\sin^2\theta)\,dr\wedge d\theta\\ &=r\,dr\wedge d\theta. \end{align*} Pulling back the area form means pulling back both coordinate covectors and then wedging them. In polar coordinates, this produces the factor $r$ from the expansion of $d(r\cos\theta)\wedge d(r\sin\theta)$. The usual polar area element $r\,dr\,d\theta$ is therefore not an added correction term; it is the coordinate expression of the pulled-back form $F^*(dx\wedge dy)$. [/example] ## Stokes As The Unifying Principle What single theorem contains the [fundamental theorem of calculus](/theorems/632), Green's theorem, the [divergence theorem](/theorems/2754), and the Kelvin-Stokes theorem? The answer is the generalised Stokes theorem. Its strength is that the same formula holds in every dimension and for every degree of form. [quotetheorem:3555] The theorem says that integration of a derivative over a region is the same as integration over the oriented boundary. The many classical vector calculus theorems differ only in how forms are translated into vector-field notation. [example: Fundamental Theorem As Stokes] Let $M=[a,b]$ with its standard orientation, so the positive coordinate vector on the interior is $\partial/\partial x$. Let $f\in C^\infty([a,b])$, and regard $f$ as a $0$-form on $M$. For an oriented point $+p$, integration of a $0$-form is evaluation: \begin{align*} \int_{+p} f=f(p). \end{align*} For the same point with the opposite orientation, denoted $-p$, integration changes sign: \begin{align*} \int_{-p} f=-f(p). \end{align*} The generalised Stokes formula on the oriented interval $[a,b]$ is exactly \begin{align*} \int_a^b f'(x)\,dx=f(b)-f(a). \end{align*} The signs come from the induced boundary orientation \begin{align*} \partial[a,b]=\{b\}-\{a\}. \end{align*} Since $f$ is a $0$-form, the coordinate definition of the [exterior derivative](/theorems/1525) gives \begin{align*} df=\frac{\partial f}{\partial x}\,dx=f'(x)\,dx. \end{align*} Therefore, by the definition of integration of a top-degree form on the coordinate interval, \begin{align*} \int_M df &=\int_{[a,b]} f'(x)\,dx\\ &=\int_a^b f'(x)\,dx. \end{align*} Now determine the boundary orientation. At the right endpoint $b$, the outward-pointing vector is $+\partial/\partial x$, which agrees with the standard orientation of $[a,b]$. Thus $b$ receives positive orientation: \begin{align*} b=+b. \end{align*} At the left endpoint $a$, the outward-pointing vector is $-\partial/\partial x$, which is the negative of the standard positive vector. Thus $a$ receives negative orientation: \begin{align*} a=-a. \end{align*} Hence the oriented boundary is \begin{align*} \partial M=+b+(-a)=\{b\}-\{a\}. \end{align*} Using the definition of integration over oriented points, \begin{align*} \int_{\partial M} f &=\int_{\{b\}-\{a\}} f\\ &=\int_{+b} f+\int_{-a} f\\ &=f(b)-f(a). \end{align*} By *Generalised Stokes Theorem*, applied to the $0$-form $f$ on the oriented $1$-manifold $M=[a,b]$, \begin{align*} \int_M df=\int_{\partial M} f. \end{align*} Substituting the two computed sides gives \begin{align*} \int_a^b f'(x)\,dx=f(b)-f(a). \end{align*} The ordinary [fundamental theorem of calculus](/theorems/632) is the $1$-dimensional case of Stokes theorem. The right endpoint appears with a plus sign because its outward direction agrees with the orientation of the interval, and the left endpoint appears with a minus sign because its outward direction is opposite to the orientation. [/example] The next case raises the dimension by one. Instead of endpoints, the boundary is now an oriented curve, and the same sign convention becomes the positive orientation around a planar region. [example: Green Theorem As Stokes] Let $D\subset\mathbb R^2$ be a compact oriented region with positively oriented boundary $\partial D$. Let $P,Q$ be smooth functions on an [open set](/page/Open%20Set) containing $D$, and set \begin{align*} \omega=P\,dx+Q\,dy. \end{align*} Orient $D$ by the standard area form $dx\wedge dy$. The generalised Stokes theorem applied to $\omega$ gives Green's theorem: \begin{align*} \int_{\partial D} P\,dx+Q\,dy =\int_D\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)d\mathcal L^2. \end{align*} By linearity of the [exterior derivative](/theorems/1525) and the coordinate definition of $d$, \begin{align*} d\omega &=d(P\,dx+Q\,dy)\\ &=d(P\,dx)+d(Q\,dy)\\ &=\left(\frac{\partial P}{\partial x}dx+\frac{\partial P}{\partial y}dy\right)\wedge dx +\left(\frac{\partial Q}{\partial x}dx+\frac{\partial Q}{\partial y}dy\right)\wedge dy. \end{align*} Expand the wedge products term by term: \begin{align*} d\omega &=\frac{\partial P}{\partial x}dx\wedge dx +\frac{\partial P}{\partial y}dy\wedge dx +\frac{\partial Q}{\partial x}dx\wedge dy +\frac{\partial Q}{\partial y}dy\wedge dy. \end{align*} Using \begin{align*} dx\wedge dx=0,\qquad dy\wedge dy=0,\qquad dy\wedge dx=-dx\wedge dy, \end{align*} this becomes \begin{align*} d\omega &=0-\frac{\partial P}{\partial y}dx\wedge dy +\frac{\partial Q}{\partial x}dx\wedge dy+0\\ &=\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy. \end{align*} By *Generalised Stokes Theorem*, applied to the oriented $2$-manifold with boundary $D$ and the $1$-form $\omega$, \begin{align*} \int_D d\omega=\int_{\partial D}\omega. \end{align*} Substitute the expression for $d\omega$: \begin{align*} \int_{\partial D}\omega &=\int_D d\omega\\ &=\int_D\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy. \end{align*} By the definition of integration of a top-degree form on a coordinate domain with the standard orientation, \begin{align*} \int_D\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy =\int_D\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)d\mathcal L^2. \end{align*} Since $\omega=P\,dx+Q\,dy$, the boundary integral is \begin{align*} \int_{\partial D}\omega=\int_{\partial D}P\,dx+Q\,dy. \end{align*} Combining these equalities gives \begin{align*} \int_{\partial D}P\,dx+Q\,dy =\int_D\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)d\mathcal L^2. \end{align*} Green's theorem is the case of Stokes theorem with $M=D\subset\mathbb R^2$ and $\omega=P\,dx+Q\,dy$. The [exterior derivative](/theorems/1525) turns the line-integrand coefficients $P$ and $Q$ into the signed area coefficient $\partial Q/\partial x-\partial P/\partial y$, while the induced positive boundary orientation supplies the orientation used in the boundary line integral. [/example] ## Cohomology From Closed And Exact Forms If $d^2=0$, what information is lost when a closed form is not itself a derivative? This question leads from analysis to topology. The failure of a closed form to be exact can detect holes in the underlying space. [definition: Closed And Exact Forms] Let $M$ be a smooth manifold. A form $\omega\in\Omega^k(M)$ is closed if $d\omega=0$. It is exact if there exists $\eta\in\Omega^{k-1}(M)$ such that $\omega=d\eta$. [/definition] Since $d^2=0$, every exact form is closed. The quotient by exact forms measures the closed forms that remain after derivatives have been ignored. [definition: De Rham Cohomology] Let $M$ be a smooth manifold. The $k$-th de Rham cohomology group of $M$ is \begin{align*} H_{\mathrm{dR}}^k(M):=\frac{\ker(d:\Omega^k(M)\to\Omega^{k+1}(M))}{\operatorname{im}(d:\Omega^{k-1}(M)\to\Omega^k(M))}. \end{align*} [/definition] A class in $H_{\mathrm{dR}}^k(M)$ is represented by a closed $k$-form, and changing the representative by an exact form does not change the class. Stokes theorem explains this quotient: exact corrections integrate to zero over closed cycles. [quotetheorem:832] The Poincare lemma says that local cohomology in positive degree vanishes on sufficiently simple coordinate domains. De Rham cohomology is therefore a global invariant: it appears when local primitives cannot be chosen consistently across the whole space. [example: A Closed One Form On The Punctured Plane] Let $M=\mathbb R^2\setminus\{(0,0)\}$, with standard coordinates $x,y$. Define \begin{align*} \omega=\frac{-y}{x^2+y^2}\,dx+\frac{x}{x^2+y^2}\,dy. \end{align*} Since $x^2+y^2>0$ on $M$, the coefficient functions are smooth on $M$. Write \begin{align*} P(x,y)=\frac{-y}{x^2+y^2},\qquad Q(x,y)=\frac{x}{x^2+y^2}, \end{align*} so that $\omega=P\,dx+Q\,dy$. Let \begin{align*} \gamma:[0,2\pi]\to M,\qquad \gamma(t)=(\cos t,\sin t) \end{align*} be the counterclockwise parametrisation of the unit circle. The $1$-form $\omega$ is closed but not exact on $M$. More precisely, \begin{align*} d\omega=0 \end{align*} and \begin{align*} \int_\gamma\omega=2\pi. \end{align*} The nonzero integral around the closed curve $\gamma$ rules out the existence of a global smooth function $f$ on $M$ with $\omega=df$. By the [coordinate formula for the exterior derivative](/theorems/3564) of a $1$-form on an open subset of $\mathbb R^2$, \begin{align*} d\omega &=d(P\,dx+Q\,dy)\\ &=d(P\,dx)+d(Q\,dy)\\ &=\left(\frac{\partial P}{\partial x}dx+\frac{\partial P}{\partial y}dy\right)\wedge dx +\left(\frac{\partial Q}{\partial x}dx+\frac{\partial Q}{\partial y}dy\right)\wedge dy\\ &=\frac{\partial P}{\partial x}dx\wedge dx +\frac{\partial P}{\partial y}dy\wedge dx +\frac{\partial Q}{\partial x}dx\wedge dy +\frac{\partial Q}{\partial y}dy\wedge dy. \end{align*} Using \begin{align*} dx\wedge dx=0,\qquad dy\wedge dy=0,\qquad dy\wedge dx=-dx\wedge dy, \end{align*} this becomes \begin{align*} d\omega &=-\frac{\partial P}{\partial y}dx\wedge dy +\frac{\partial Q}{\partial x}dx\wedge dy\\ &=\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy. \end{align*} Now compute the two partial derivatives: \begin{align*} \frac{\partial Q}{\partial x} &=\frac{\partial}{\partial x}\left(x(x^2+y^2)^{-1}\right)\\ &=(x^2+y^2)^{-1}+x\left(-1\right)(x^2+y^2)^{-2}(2x)\\ &=\frac{1}{x^2+y^2}-\frac{2x^2}{(x^2+y^2)^2}\\ &=\frac{x^2+y^2-2x^2}{(x^2+y^2)^2}\\ &=\frac{y^2-x^2}{(x^2+y^2)^2}, \end{align*} and \begin{align*} \frac{\partial P}{\partial y} &=\frac{\partial}{\partial y}\left(-y(x^2+y^2)^{-1}\right)\\ &=-(x^2+y^2)^{-1}+(-y)\left(-1\right)(x^2+y^2)^{-2}(2y)\\ &=-\frac{1}{x^2+y^2}+\frac{2y^2}{(x^2+y^2)^2}\\ &=\frac{-(x^2+y^2)+2y^2}{(x^2+y^2)^2}\\ &=\frac{y^2-x^2}{(x^2+y^2)^2}. \end{align*} Therefore \begin{align*} \frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y} &=\frac{y^2-x^2}{(x^2+y^2)^2} -\frac{y^2-x^2}{(x^2+y^2)^2}\\ &=0, \end{align*} so \begin{align*} d\omega=0. \end{align*} Thus $\omega$ is closed. Next compute the pullback along $\gamma$. Since \begin{align*} x\circ\gamma(t)=\cos t,\qquad y\circ\gamma(t)=\sin t, \end{align*} we have \begin{align*} \gamma^*(dx)=d(\cos t)=-\sin t\,dt,\qquad \gamma^*(dy)=d(\sin t)=\cos t\,dt. \end{align*} Also \begin{align*} (P\circ\gamma)(t) &=\frac{-\sin t}{\cos^2 t+\sin^2 t}\\ &=-\sin t, \end{align*} and \begin{align*} (Q\circ\gamma)(t) &=\frac{\cos t}{\cos^2 t+\sin^2 t}\\ &=\cos t. \end{align*} By the definition of pullback of a $1$-form, \begin{align*} \gamma^*\omega &=(P\circ\gamma)\,\gamma^*(dx)+(Q\circ\gamma)\,\gamma^*(dy)\\ &=(-\sin t)(-\sin t\,dt)+(\cos t)(\cos t\,dt)\\ &=\sin^2 t\,dt+\cos^2 t\,dt\\ &=(\sin^2 t+\cos^2 t)\,dt\\ &=dt. \end{align*} Therefore \begin{align*} \int_\gamma\omega &=\int_0^{2\pi}\gamma^*\omega\\ &=\int_0^{2\pi}dt\\ &=\int_0^{2\pi}1\,dt\\ &=2\pi-0\\ &=2\pi. \end{align*} It remains to show that $\omega$ is not exact. Suppose, for contradiction, that there exists $f\in C^\infty(M)$ such that \begin{align*} \omega=df. \end{align*} Let $C=S^1\subset M$ be the unit circle with the orientation induced by $\gamma$. Since $C$ is a closed oriented $1$-manifold, \begin{align*} \partial C=\varnothing. \end{align*} By *Generalised Stokes Theorem*, applied to the $0$-form $f|_C$ on $C$, \begin{align*} \int_C df=\int_{\partial C} f. \end{align*} Because $\partial C=\varnothing$, \begin{align*} \int_{\partial C} f=0. \end{align*} Thus \begin{align*} \int_C df=0. \end{align*} But $\omega=df$, and $\gamma$ parametrises $C$ once with its chosen orientation, so \begin{align*} 0 &=\int_C df\\ &=\int_C \omega\\ &=\int_\gamma\omega\\ &=2\pi. \end{align*} This contradiction shows that no such global smooth function $f$ exists. Hence $\omega$ is not exact on $M$. The form \begin{align*} \omega=\frac{-y}{x^2+y^2}\,dx+\frac{x}{x^2+y^2}\,dy \end{align*} is closed because the two mixed coefficient derivatives agree exactly. It is not exact because its integral around the unit circle is $2\pi$, while an exact $1$-form has integral $0$ over every closed oriented $1$-manifold by Stokes theorem. Therefore $\omega$ represents a nonzero de Rham cohomology class in $H_{\mathrm{dR}}^1(\mathbb R^2\setminus\{0\})$, detecting the hole at the origin. [/example] ## How The Course Progresses What background should a reader keep active while moving through the course? The main prerequisites are multivariable calculus, linear algebra of dual spaces and determinants, and basic topology of open sets, compactness, and continuity. Smooth manifolds enter gradually: first as spaces covered by coordinate charts, then as objects on which forms, pullbacks, orientations, and integrals are intrinsic. The first part of the course is algebraic. It constructs alternating tensors, wedge products, and bases for $\Lambda^k(V^*)$. This explains why $k$-forms have the coordinate expressions used throughout the notes. The second part is analytic and geometric. It defines differential forms on open sets and manifolds, proves the coordinate invariance of [exterior derivative](/theorems/1525) and pullback, develops oriented integration, and proves the generalised Stokes theorem. The final part is topological. It introduces closed and exact forms, proves the Poincare lemma, computes de Rham cohomology for standard examples, and states de Rham's comparison theorem relating differential forms to singular cohomology. The guiding principle is that a form is an integrand with its transformation law built in. Once this viewpoint is in place, the major theorems of vector calculus become shadows of one theorem, and the obstruction to finding primitives becomes a computable invariant of the underlying space. The exterior algebra provides the necessary algebraic foundation for differential forms. We begin by studying alternating tensors on a single [vector space](/page/Vector%20Space), which will become, at each point of a manifold, the domain on which differential k-forms act. # 1. Exterior Algebra This opening chapter builds the algebraic language that later becomes the language of differential forms on manifolds. At a single point of a manifold, a differential $k$-form is an alternating $k$-linear function on tangent vectors, so the first task is to understand alternating tensors on an ordinary finite-dimensional real [vector space](/page/Vector%20Space). The chapter moves from multilinearity, to antisymmetrisation, to the wedge product, and ends with the coordinate basis and the determinant as the model top-degree form. ## Multilinear Maps and Alternating Tensors What algebraic object should receive $k$ vectors and produce a number in a way compatible with linear algebra in each input? For integration and orientation, the answer must also change sign when two directions are swapped, because reversing an oriented parallelepiped should reverse its signed volume. Let $V$ be a finite-dimensional real [vector space](/page/Vector%20Space). A map of $k$ vector inputs records how $k$ directions interact, but multilinearity is the condition that lets the map be determined by its values on a basis. [definition: Multilinear Map] Let $k \in \mathbb N$. A $k$-[linear map](/page/Linear%20Map) on $V$ is a function $T: V^k \to \mathbb R$ such that for each $j \in \{1, \dots, k\}$, each fixed choice of the other $k-1$ inputs, and all $a,b \in \mathbb R$ and $v,w \in V$, \begin{align*} T(v_1, \dots, av + bw, \dots, v_k) &= aT(v_1, \dots, v, \dots, v_k) + bT(v_1, \dots, w, \dots, v_k). \end{align*} The [vector space](/page/Vector%20Space) of all $k$-linear maps $V^k \to \mathbb R$ is denoted by $\operatorname{Mult}^k(V)$. [/definition] For $k=1$, this is just the [dual space](/page/Dual%20Space) $V^*$. For larger $k$, the new feature is the ability to compare what happens under a rearrangement of inputs. [definition: Alternating Multilinear Map] Let $k \in \mathbb N$. A $k$-[linear map](/page/Linear%20Map) $\alpha: V^k \to \mathbb R$ is alternating if \begin{align*} \alpha(v_1, \dots, v_k) = 0 \end{align*} whenever $v_i = v_j$ for some distinct $i,j \in \{1, \dots, k\}$. The [vector space](/page/Vector%20Space) of alternating $k$-linear maps $V^k \to \mathbb R$ is denoted by $\Lambda^k(V^*)$. We set $\Lambda^0(V^*) = \mathbb R$. [/definition] The notation $\Lambda^k(V^*)$ anticipates the construction of exterior powers. Its elements are called alternating $k$-tensors, exterior $k$-forms on $V$, or simply $k$-forms when the [vector space](/page/Vector%20Space) is fixed. [quotetheorem:3556] This theorem is the algebraic source of orientation signs. Alternating tensors do not merely vanish on degenerate $k$-tuples; they remember the order of independent directions up to the sign of the corresponding permutation. [example: A Two Form on the Plane] Let $V=\mathbb R^2$ with coordinate covectors $dx,dy\in V^*$, so $dx(x,y)=x$ and $dy(x,y)=y$. Define \begin{align*} \omega: V^2 &\to \mathbb R \\ (v,w) &\mapsto dx(v)dy(w)-dx(w)dy(v). \end{align*} We compute $\omega$ on the pair $v=(a,b)$, $w=(c,d)$. The coordinate covectors return the components directly: $dx(v)=a$, $dy(v)=b$, $dx(w)=c$, $dy(w)=d$. Substituting, \begin{align*} \omega(v,w) =dx(v)dy(w)-dx(w)dy(v) =ad-cb =ad-bc. \end{align*} This is exactly the determinant of the matrix whose columns are $v$ and $w$: \begin{align*} \det\begin{pmatrix} a & c\\ b & d \end{pmatrix} =ad-bc=\omega(v,w). \end{align*} Geometrically, $\omega(v,w)$ is the signed area of the parallelogram spanned by $v$ and $w$ — the same quantity that the determinant computes. The sign is the orientation of the ordered pair: swapping the inputs flips it, \begin{align*} \omega(w,v)=dx(w)dy(v)-dx(v)dy(w)=cb-ad=-\omega(v,w), \end{align*} which is the alternating property that makes $\omega$ a two-form rather than a generic bilinear map. [/example] The example already contains the pattern of the wedge product: take products of covectors and antisymmetrise them so that repeated directions disappear and interchanges of directions introduce signs. ## Antisymmetrisation and the Wedge Product How can an arbitrary multilinear expression be turned into an alternating one without changing the alternating part it already contains? The operation is to average over all permutations with the signs attached, producing a projection onto $\Lambda^k(V^*)$. [definition: Antisymmetrisation] Let $T \in \operatorname{Mult}^k(V)$. The antisymmetrisation of $T$ is the $k$-[linear map](/page/Linear%20Map) $\operatorname{Alt}(T): V^k \to \mathbb R$ defined by \begin{align*} \operatorname{Alt}(T)(v_1, \dots, v_k) = \frac{1}{k!}\sum_{\sigma \in S_k} \operatorname{sgn}(\sigma) T(v_{\sigma(1)}, \dots, v_{\sigma(k)}). \end{align*} [/definition] Antisymmetrisation is the formal version of keeping only the oriented part of a multilinear expression. The factor $1/k!$ makes it a projection rather than only a signed sum. [quotetheorem:3557] The [projection theorem](/theorems/1985) explains why antisymmetrisation is the right operation rather than just a convenient formula. Linearity means it can be applied term-by-term after expanding a tensor, the identity property means already alternating tensors are left unchanged, and idempotence means applying it a second time does not alter the result. The factor $1/k!$ is responsible for that idempotence; without it, alternating tensors would be multiplied by $k!$ rather than fixed. This theorem does not say that $\operatorname{Alt}$ respects tensor products. Even if two factors are alternating, their ordinary [tensor product](/page/Tensor%20Product) need not be alternating in all variables, so the [tensor product](/page/Tensor%20Product) must be antisymmetrised before it becomes a form of higher degree. [example: Tensor Product Before Antisymmetrisation] On $V=\mathbb R^2$ with coordinate covectors $dx,dy\in V^*$ and standard basis $e_1=(1,0)$, $e_2=(0,1)$, consider the tensor product \begin{align*} T=dx\otimes dy, \qquad T(v,w)=dx(v)dy(w). \end{align*} Evaluating on the basis pair $(e_1,e_2)$ gives $T(e_1,e_2)=dx(e_1)dy(e_2)=1$, while $T(e_2,e_1)=dx(e_2)dy(e_1)=0\cdot 0=0$. If $T$ were alternating it would need to satisfy $T(e_2,e_1)=-T(e_1,e_2)=-1$, so the asymmetry $0\neq -1$ shows $dx\otimes dy$ is not alternating. Antisymmetrising removes the symmetric part. For two inputs, the only permutations are the identity (sign $+1$) and the swap (sign $-1$), so \begin{align*} \operatorname{Alt}(T)(v,w) =\frac{1}{2}\bigl(T(v,w)-T(w,v)\bigr) =\frac{1}{2}\bigl(dx(v)dy(w)-dx(w)dy(v)\bigr). \end{align*} The same expression appears on the right of $dx\otimes dy-dy\otimes dx$, using commutativity of multiplication in $\mathbb R$: \begin{align*} (dx\otimes dy-dy\otimes dx)(v,w) =dx(v)dy(w)-dy(v)dx(w) =dx(v)dy(w)-dx(w)dy(v). \end{align*} Hence \begin{align*} \operatorname{Alt}(dx\otimes dy) =\tfrac{1}{2}(dx\otimes dy-dy\otimes dx). \end{align*} The wedge product is the alternating part rescaled by the combinatorial factor $\frac{(p+q)!}{p!q!}$. For two one-forms ($p=q=1$) this factor is $2$, so \begin{align*} dx\wedge dy =2\operatorname{Alt}(dx\otimes dy) =dx\otimes dy-dy\otimes dx. \end{align*} The rescaling is chosen precisely so that $(dx\wedge dy)(e_1,e_2)=1$: the wedge of dual covectors evaluates to $1$ on the corresponding basis pair, recovering a clean duality with the basis of $\Lambda^2(V^*)$. The unscaled $\operatorname{Alt}(dx\otimes dy)$ would give $\tfrac{1}{2}$ instead, and chains of wedges would carry inconvenient factorial factors. [/example] The wedge product is therefore defined by multiplying two alternating tensors and then applying antisymmetrisation. The normalising constant is chosen so that wedge products of basis covectors have the expected determinant formula without extra factorials. [definition: Wedge Product] Let $\alpha \in \Lambda^p(V^*)$ and $\beta \in \Lambda^q(V^*)$. Their wedge product is the element $\alpha \wedge \beta \in \Lambda^{p+q}(V^*)$ defined by \begin{align*} \alpha \wedge \beta = \frac{(p+q)!}{p!q!}\operatorname{Alt}(\alpha \otimes \beta), \end{align*} where \begin{align*} (\alpha \otimes \beta)(v_1, \dots, v_{p+q}) = \alpha(v_1, \dots, v_p)\beta(v_{p+1}, \dots, v_{p+q}). \end{align*} [/definition] The formula says that $\alpha$ is evaluated on $p$ of the vectors, $\beta$ on the remaining $q$, and then every way of choosing and ordering those inputs is combined with the correct orientation sign. [quotetheorem:3558] The shuffle formula is often the most efficient way to compute wedges by hand. In small degrees it recovers familiar determinant expressions. [example: Wedge of Two One Forms in Three Dimensions] Let $V=\mathbb R^3$ with coordinate covectors $dx,dy,dz\in V^*$, and let \begin{align*} \alpha = a_1dx+a_2dy+a_3dz, \qquad \beta = b_1dx+b_2dy+b_3dz \end{align*} be one-forms with scalar coefficients $a_i,b_i\in\mathbb R$. [claim] \begin{align*} \alpha \wedge \beta = (a_1b_2-a_2b_1)\,dx\wedge dy + (a_1b_3-a_3b_1)\,dx\wedge dz + (a_2b_3-a_3b_2)\,dy\wedge dz. \end{align*} [/claim] [proof] Two identities for one-form wedges drive the calculation. For $\eta,\theta\in V^*$ and $v,w\in V$, the definition $\eta\wedge\theta=2\operatorname{Alt}(\eta\otimes\theta)$ unpacks to \begin{align*} (\eta\wedge\theta)(v,w) =\eta(v)\theta(w)-\eta(w)\theta(v). \end{align*} Setting $\theta=\eta$ gives $(\eta\wedge\eta)(v,w)=\eta(v)\eta(w)-\eta(w)\eta(v)=0$, so any one-form wedged with itself vanishes. Swapping the order of $\eta,\theta$ negates the result, so $\theta\wedge\eta=-\eta\wedge\theta$. Applied to the coordinate covectors, \begin{align*} dx\wedge dx=dy\wedge dy=dz\wedge dz=0, \quad dy\wedge dx=-dx\wedge dy, \quad dz\wedge dx=-dx\wedge dz, \quad dz\wedge dy=-dy\wedge dz. \end{align*} The wedge product is bilinear (it factors through the [tensor product](/page/Tensor%20Product) followed by the [linear map](/page/Linear%20Map) $\operatorname{Alt}$), so expanding $\alpha\wedge\beta$ produces nine terms $a_ib_j\,dx^i\wedge dx^j$ where $dx^1=dx$, $dx^2=dy$, $dx^3=dz$. The three diagonal terms vanish, and the off-diagonal pairs combine using $dx^j\wedge dx^i=-dx^i\wedge dx^j$: \begin{align*} \alpha\wedge\beta &=a_1b_2\,dx\wedge dy+a_2b_1\,dy\wedge dx +a_1b_3\,dx\wedge dz+a_3b_1\,dz\wedge dx +a_2b_3\,dy\wedge dz+a_3b_2\,dz\wedge dy\\ &=(a_1b_2-a_2b_1)\,dx\wedge dy +(a_1b_3-a_3b_1)\,dx\wedge dz +(a_2b_3-a_3b_2)\,dy\wedge dz. \end{align*} [/proof] The three coefficients $a_ib_j-a_jb_i$ are the $2\times 2$ minors of the matrix $\begin{pmatrix}a_1&a_2&a_3\\b_1&b_2&b_3\end{pmatrix}$. The wedge $\alpha\wedge\beta$ therefore records the oriented coordinate-plane areas associated to the two one-forms — the same data that the classical cross product packages as a vector in $\mathbb R^3$. [/example] The wedge product is the multiplication law of exterior algebra. Its two structural features are associativity and a sign rule recording the degrees of the factors. [quotetheorem:3559] Associativity is the point at which the normalisation in the wedge product pays for itself. The two parenthesisations both reduce to the same signed sum over $(p,q,r)$-shuffles, with no extra binomial coefficients left over from choosing intermediate blocks. Thus exterior forms assemble into an algebra $\Lambda^\ast(V^*)=\bigoplus_{k=0}^n \Lambda^k(V^*)$ whose multiplication is the wedge product. Because of associativity, expressions such as $dx\wedge dy\wedge dz$ need no parentheses. The next theorem explains how much commutativity remains. [quotetheorem:3560] For one-forms this gives anticommutativity, while for a two-form and another two-form it gives commutativity. In particular, if $\alpha$ has odd degree, then $\alpha\wedge\alpha=-\alpha\wedge\alpha$, so over $\mathbb R$ we get $\alpha\wedge\alpha=0$. This conclusion uses the sign and the ability to divide by $2$; in characteristic $2$ the sign disappears, so alternating behaviour has to be imposed separately rather than recovered from graded commutativity. The theorem also belongs to exterior forms, not to arbitrary multilinear tensors: without alternation of both factors, the shuffle-sign argument has no reason to apply. ## Coordinate Bases and Dimension Once a basis of $V$ is chosen, which alternating tensors are needed to describe every $k$-form? Since alternating tensors vanish on repeated vectors, only ordered collections of distinct coordinate covectors can survive. Let $\dim V=n$, and let $e_1,\dots,e_n$ be a basis of $V$ with [dual basis](/theorems/414) $dx_1,\dots,dx_n \in V^*$. The notation $dx_i$ is chosen to match the later differential-geometric setting, where these covectors are differentials of coordinate functions. [definition: Coordinate Wedge] For indices $1 \le i_1 < \dots < i_k \le n$, the coordinate $k$-form \begin{align*} dx_{i_1}\wedge \dots \wedge dx_{i_k} \in \Lambda^k(V^*) \end{align*} is the wedge product of the corresponding [dual basis](/theorems/414) covectors. [/definition] The strict inequality in the indices is not cosmetic. If an index is repeated, the wedge is zero; if the same indices are written in a different order, the result changes by the sign of the ordering permutation. [quotetheorem:3561] This theorem is the computational backbone of exterior algebra. It turns the abstract [vector space](/page/Vector%20Space) $\Lambda^k(V^*)$ into a concrete coordinate space whose coordinates are indexed by increasing $k$-element subsets of $\{1,\dots,n\}$. [example: Two Forms in Three Dimensions] Let $V=\mathbb R^3$ with standard basis $e_1,e_2,e_3$ and [dual basis](/theorems/414) $dx,dy,dz\in V^*$, so $dx(e_i)=\delta_{1i}$, $dy(e_i)=\delta_{2i}$, $dz(e_i)=\delta_{3i}$. By the basis theorem for exterior powers, every $\omega\in\Lambda^2(V^*)$ has a unique expansion \begin{align*} \omega = A\,dx\wedge dy + B\,dx\wedge dz + C\,dy\wedge dz, \qquad A,B,C\in\mathbb R. \end{align*} [claim] The coefficients are recovered by evaluating $\omega$ on the increasing pairs of basis vectors: \begin{align*} A = \omega(e_1,e_2), \qquad B = \omega(e_1,e_3), \qquad C = \omega(e_2,e_3). \end{align*} [/claim] [proof] The wedge of two coordinate covectors satisfies $(\eta\wedge\theta)(r,s)=\eta(r)\theta(s)-\eta(s)\theta(r)$, so on the basis pair $(e_1,e_2)$, \begin{align*} (dx\wedge dy)(e_1,e_2) &=dx(e_1)dy(e_2)-dx(e_2)dy(e_1) =1\cdot 1-0\cdot 0=1,\\ (dx\wedge dz)(e_1,e_2) &=dx(e_1)dz(e_2)-dx(e_2)dz(e_1) =1\cdot 0-0\cdot 0=0,\\ (dy\wedge dz)(e_1,e_2) &=dy(e_1)dz(e_2)-dy(e_2)dz(e_1) =0\cdot 0-1\cdot 0=0. \end{align*} By linearity of evaluation, $\omega(e_1,e_2)=A\cdot 1+B\cdot 0+C\cdot 0=A$. The same Kronecker-delta calculation on the pairs $(e_1,e_3)$ and $(e_2,e_3)$ singles out $B$ and $C$ respectively: each coordinate wedge $dx^i\wedge dx^j$ returns $1$ on its own increasing basis pair and $0$ on the other two. [/proof] These three values determine $\omega$ on every pair of vectors. Alternation forces $\omega(e_i,e_i)=0$, and the polarisation identity $\omega(e_i+e_j,e_i+e_j)=0$ combined with bilinearity gives $\omega(e_j,e_i)=-\omega(e_i,e_j)$. Expanding arbitrary $r=\sum x_i e_i$ and $s=\sum y_j e_j$ by bilinearity then yields \begin{align*} \omega(r,s) =A(x_1y_2-x_2y_1)+B(x_1y_3-x_3y_1)+C(x_2y_3-x_3y_2), \end{align*} a sum of three signed $2\times 2$ minors of the matrix $\begin{pmatrix}x_1&x_2&x_3\\y_1&y_2&y_3\end{pmatrix}$. A two-form on $\mathbb R^3$ is therefore three numbers — one per oriented coordinate plane — exactly mirroring the three components of a vector in $\mathbb R^3$, which is why two-forms in three dimensions are classically identified with vector fields via the cross product. [/example] The binomial coefficient reflects the fact that a $k$-form chooses $k$ independent coordinate directions at a time. This is why exterior algebra is finite in dimension even though wedge products can be iterated formally. ## Top-Degree Forms and the Determinant What remains in degree $n$ when $V$ itself has dimension $n$? There is only one increasing choice of all coordinate directions, so the space of top-degree forms is one-dimensional. [quotetheorem:3561] The finite-dimensional hypothesis is essential here. In an infinite-dimensional [vector space](/page/Vector%20Space) there is no largest finite degree, and exterior powers do not collapse to a single top-dimensional line. For finite-dimensional $V$, an orientation can be described as a choice of one of the two possible positive rays in the one-dimensional space $\Lambda^n(V^*)$; choosing a nonzero top form fixes which ordered bases are positively oriented. Dually, $\Lambda^n(V)$ is also one-dimensional and pairs naturally with $\Lambda^n(V^*)$ by evaluation. The choice of basis matters: replacing the basis rescales the coordinate top form by the determinant of the change-of-basis matrix. This is the first appearance of the determinant as the transformation factor for oriented volume. [quotetheorem:393] This theorem connects exterior algebra with the Jacobian determinants from multivariable calculus. Later, when forms are pulled back along smooth maps, this determinant factor will appear automatically from the algebra of top-degree forms. [example: The Standard Volume Form in Three Dimensions] Let $V=\mathbb R^3$ with coordinate covectors $dx,dy,dz\in V^*$, and set \begin{align*} \Omega = dx\wedge dy\wedge dz. \end{align*} For vectors $u,v,w\in\mathbb R^3$ with components $u_i,v_i,w_i$, let $A$ denote the matrix with columns $u,v,w$. [claim] \begin{align*} \Omega(u,v,w)=\det A. \end{align*} [/claim] [proof] Write $\Omega=\varphi\wedge dz$ where $\varphi=dx\wedge dy$, and let $T=\varphi\otimes dz$, so $T(r,s,t)=\varphi(r,s)dz(t)$. The wedge of a $2$-form and a $1$-form has combinatorial factor $\frac{(2+1)!}{2!1!}=3$, giving $\Omega=3\operatorname{Alt}(T)$. For three inputs, $\operatorname{Alt}$ averages $T$ over the six signed permutations: \begin{align*} \Omega(u,v,w) =\frac{1}{2}\bigl( T(u,v,w)-T(v,u,w)-T(u,w,v)+T(w,u,v)+T(v,w,u)-T(w,v,u) \bigr). \end{align*} Because $\varphi$ is itself alternating, three of the six terms collapse onto the others: $T(v,u,w)=-T(u,v,w)$, $T(w,u,v)=-T(u,w,v)$, and $T(w,v,u)=-T(v,w,u)$. The expression reduces to \begin{align*} \Omega(u,v,w) =T(u,v,w)-T(u,w,v)+T(v,w,u). \end{align*} Each term unpacks via $\varphi(r,s)=dx(r)dy(s)-dx(s)dy(r)$: \begin{align*} T(u,v,w)&=(u_1v_2-v_1u_2)w_3,\\ T(u,w,v)&=(u_1w_2-w_1u_2)v_3,\\ T(v,w,u)&=(v_1w_2-w_1v_2)u_3. \end{align*} Summing with the correct signs gives \begin{align*} \Omega(u,v,w) =u_1v_2w_3-v_1u_2w_3-u_1w_2v_3+w_1u_2v_3+v_1w_2u_3-w_1v_2u_3. \end{align*} Cofactor expansion of $\det A$ along the first row produces the same six signed monomials: \begin{align*} \det A =u_1(v_2w_3-w_2v_3)-v_1(u_2w_3-w_2u_3)+w_1(u_2v_3-v_2u_3). \end{align*} Distributing matches term-by-term, so $\Omega(u,v,w)=\det A$. [/proof] This is the structural reason the determinant deserves the name "signed volume." The three-form $dx\wedge dy\wedge dz$ is, up to scalar, the unique alternating trilinear map on $\mathbb R^3$, and the determinant is the unique normalisation that gives the unit cube volume $1$. Linear dependence of $u,v,w$ collapses the parallelepiped — and the determinant — to zero, while orientation-reversing permutations flip the sign. [/example] The determinant interpretation also explains why volume is sensitive to linear dependence rather than merely to the lengths of the input vectors. The next example refines the repeated-direction case by perturbing one vector and watching only the independent component survive. [example: A Nearly Repeated Direction] On $V=\mathbb R^3$, the standard volume form $\Omega=dx\wedge dy\wedge dz\in\Lambda^3(V^*)$ is a trilinear alternating map $V^3\to\mathbb R$. Fix $u,v,w\in\mathbb R^3$ and a scalar $\varepsilon\in\mathbb R$, and consider the slightly perturbed triple $(u,u+\varepsilon v,w)$ — the second slot starts at $u$ and drifts toward $v$ by a small amount. Trilinearity in the second input splits $\Omega(u,u+\varepsilon v,w)$ along the sum: \begin{align*} \Omega(u,u+\varepsilon v,w) =\Omega(u,u,w)+\varepsilon\Omega(u,v,w). \end{align*} The first term vanishes because $\Omega$ is alternating and the first two slots are equal. What remains is \begin{align*} (dx\wedge dy\wedge dz)(u,u+\varepsilon v,w) =\varepsilon(dx\wedge dy\wedge dz)(u,v,w). \end{align*} The geometric content is that the repeated direction $u$ contributes no volume — the parallelepiped collapses along that axis — so only the transverse perturbation $\varepsilon v$ generates new signed volume, and it does so linearly. This linear sensitivity is the reason alternating top-forms detect first-order deformations of degenerate configurations, and it foreshadows how the [exterior derivative](/theorems/1525) will measure infinitesimal twisting of forms by exploiting the same kind of cancellation. [/example] Exterior algebra therefore packages three facts into a single structure: multilinearity, antisymmetry, and determinant-like volume. The next chapter globalises this pointwise algebra by assigning to every point of an [open set](/page/Open%20Set) or manifold an alternating tensor on the tangent space, producing differential forms as fields of exterior-algebra elements. With the algebraic structure of exterior algebra in place at a single point, we now globalize it by letting these constructions vary smoothly across an [open set](/page/Open%20Set) of ℝⁿ. This produces differential forms as smooth fields of exterior algebra elements. # 2. Differential k-Forms on Open Subsets of Rⁿ The first chapter built exterior algebra at a single [vector space](/page/Vector%20Space): alternating covectors, their wedge product, and the determinant as the top-degree case. We now let those algebraic objects vary smoothly from point to point on an [open set](/page/Open%20Set) $U\subset \mathbb R^n$. The resulting objects are differential forms, and they are the basic quantities that will later be differentiated, pulled back, and integrated. The guiding shift is from a single alternating map to a field of alternating maps. A $k$-form assigns to each point $x\in U$ an element of $\Lambda^k(T_x^*U)$, with coordinate coefficients depending smoothly on $x$. On open subsets of Euclidean space, this can be written without bundle machinery because each tangent space $T_xU$ is naturally identified with $\mathbb R^n$. ## Forms as Smooth Covector Fields How can the alternating covectors from exterior algebra become objects on an [open set](/page/Open%20Set) rather than at a single point? The coordinate functions $x_1,\dots,x_n$ on $U\subset\mathbb R^n$ provide a standard covector basis at every point, so a form is described by smooth coefficient functions multiplying the exterior algebra basis. [definition: Coordinate One-Forms] Let $U\subset \mathbb R^n$ be open. For $1\le i\le n$, the coordinate one-form $dx_i$ assigns to $x\in U$ the covector $dx_i|_x:T_xU\to\mathbb R$ defined by \begin{align*} dx_i|_x(v)=v_i \end{align*} for $v=(v_1,\dots,v_n)\in T_xU$. [/definition] These coordinate one-forms are the moving version of the [dual basis](/theorems/414) from exterior algebra. Since $U$ is open in $\mathbb R^n$, the notation suppresses the identification $T_xU\cong \mathbb R^n$. [definition: Smooth k-Form On An Open Set] Let $U\subset\mathbb R^n$ be open and let $0\le k\le n$. A smooth $k$-form $\omega$ on $U$ is an assignment \begin{align*} x\mapsto \omega_x\in \Lambda^k(T_x^*U) \end{align*} for which there exist functions $a_I\in C^\infty(U)$, indexed by increasing $k$-tuples $I=(i_1<\dots<i_k)$, satisfying \begin{align*} \omega_x=\sum_I a_I(x)\,dx_{i_1}|_x\wedge\dots\wedge dx_{i_k}|_x \end{align*} for every $x\in U$. The set of all smooth $k$-forms on $U$ is denoted $\Omega^k(U)$. [/definition] The standard shorthand omits the point $x$ and writes \begin{align*} \omega=\sum_I a_I\,dx_{i_1}\wedge\dots\wedge dx_{i_k}. \end{align*} The smoothness requirement is not cosmetic. If the coefficients were merely assigned pointwise with no regularity, the exterior algebra operations would still make sense at individual tangent spaces, but the next operations in the course would fail to be stable: differentiating the coefficients would not produce smooth forms, and integration over smooth parametrised pieces would lose the regularity needed for change-of-variables arguments. Thus $\Omega^k(U)$ is the class where algebra, differentiation, and integration can be developed together. When $k=0$, the space $\Lambda^0(T_x^*U)$ is $\mathbb R$, so a smooth $0$-form is a smooth function $f\in C^\infty(U)$. When $k=n$, there is only one basis form, $dx_1\wedge\dots\wedge dx_n$, so every $n$-form is $a\,dx_1\wedge\dots\wedge dx_n$ for a unique $a\in C^\infty(U)$. [quotetheorem:3562] The theorem says that differential forms on an open subset of $\mathbb R^n$ can be manipulated by their coefficient functions, provided the exterior basis is kept in increasing order. The increasing-index convention prevents multiple names for the same term: for instance $dx_j\wedge dx_i=-dx_i\wedge dx_j$ and $dx_i\wedge dx_i=0$, so an unrestricted sum would contain redundant or vanishing pieces. Writing every term in increasing order makes the coefficient of each basis element well-defined and gives the uniqueness statement real content. The free $C^\infty(U)$-module structure is the computational reason for using coordinates at this stage. It lets later operators be defined by formulas on basis elements and smooth coefficient functions, then extended by linearity and product rules. Without smooth coefficients, this coordinate method would produce expressions that no longer lie in the same class of objects. [example: Polynomial One-Form On The Plane] On $U=\mathbb R^2$ with coordinate functions $x,y$, consider \begin{align*} \omega=x\,dy-y\,dx=(-y)\,dx+x\,dy. \end{align*} Its coefficients $-y$ and $x$ are polynomial — hence smooth — functions on $\mathbb R^2$, so by the coordinate definition of a smooth $1$-form, $\omega\in\Omega^1(\mathbb R^2)$. Evaluating at $p=(x,y)$ on a tangent vector $v=(v_1,v_2)\in T_p\mathbb R^2$ uses $dx|_p(v)=v_1$ and $dy|_p(v)=v_2$: \begin{align*} \omega_p(v) =x\,dy|_p(v)-y\,dx|_p(v) =xv_2-yv_1. \end{align*} This is the formula that will reappear, multiplied by $1/(x^2+y^2)$, as the angle form on the punctured plane. Geometrically, $\omega_p$ measures the component of $v$ orthogonal to the radial direction at $p$, weighted by $|p|$; the renormalisation needed to make it the differential of an angle is the subject of the next example. [/example] ## The Wedge Product and the Graded Algebra of Forms How do forms of different degrees multiply while remembering orientation and sign? The answer is to apply the exterior algebra wedge product pointwise, then track the degree of each form. [definition: Total Space Of Differential Forms] Let $U\subset\mathbb R^n$ be open. The graded space of smooth differential forms on $U$ is \begin{align*} \Omega^*(U)=\bigoplus_{k=0}^n\Omega^k(U). \end{align*} An element of $\Omega^k(U)$ has degree $k$. [/definition] The direct sum keeps the degree visible. Functions live in degree $0$, ordinary work-type integrands live in degree $1$, and top-degree forms live in degree $n$. [definition: Wedge Product Of Differential Forms] Let $\alpha\in\Omega^p(U)$ and $\beta\in\Omega^q(U)$. The wedge product $\alpha\wedge\beta\in\Omega^{p+q}(U)$ is defined by \begin{align*} (\alpha\wedge\beta)_x=\alpha_x\wedge\beta_x \end{align*} for every $x\in U$, with $\Omega^m(U)=0$ for $m>n$. [/definition] In coordinates, if $\alpha=\sum_I a_I dx_I$ and $\beta=\sum_J b_J dx_J$, then \begin{align*} \alpha\wedge\beta=\sum_{I,J}a_Ib_J\,dx_I\wedge dx_J, \end{align*} with repeated coordinate covectors giving zero and the remaining terms reordered into increasing order with the sign of the required permutation. A reliable computation has three steps: expand bilinearly, delete any term containing a repeated coordinate covector, and reorder each surviving wedge into increasing index order while multiplying by the sign of the permutation. In particular, a wedge of more than $n$ one-forms on $U\subset\mathbb R^n$ vanishes, and any local wedge expression vanishes whenever every expanded term has a repeated coordinate covector or cancelling coefficients after reordering. [quotetheorem:3563] This result is the global form of the exterior algebra identities from the previous chapter. The new point is that coefficient functions multiply in $C^\infty(U)$ while the signs are still controlled by degrees. Ordinary commutativity would forget the orientation reversal caused by swapping the inputs of alternating forms; graded-commutativity records exactly that sign. When $\alpha$ has odd degree, the identity gives $\alpha\wedge\alpha=-\alpha\wedge\alpha$, hence $\alpha\wedge\alpha=0$ over $\mathbb R$. This is why $1$-forms behave like oriented infinitesimal directions rather than scalar factors. The same sign rule is what later makes the [exterior derivative](/theorems/1525) satisfy a signed product rule: \begin{align*} d(\alpha\wedge\beta)=d\alpha\wedge\beta+(-1)^p\alpha\wedge d\beta \end{align*} for $\alpha\in\Omega^p(U)$. The algebraic signs in this theorem are therefore not bookkeeping only; they are the mechanism that makes calculus of forms consistent across degrees. [example: Wedge Product Computation In Three Variables] On $\mathbb R^3$ with coordinates $x,y,z$, take the two one-forms \begin{align*} \alpha=x\,dx+y\,dy,\qquad \beta=z\,dy+dx, \end{align*} and expand $\alpha\wedge\beta$ by bilinearity over smooth functions: \begin{align*} \alpha\wedge\beta =xz\,(dx\wedge dy)+x\,(dx\wedge dx)+yz\,(dy\wedge dy)+y\,(dy\wedge dx). \end{align*} Alternation forces $dx\wedge dx=dy\wedge dy=0$ and $dy\wedge dx=-dx\wedge dy$, so \begin{align*} \alpha\wedge\beta =xz\,dx\wedge dy-y\,dx\wedge dy =(xz-y)\,dx\wedge dy. \end{align*} The point is mechanical: in coordinates, taking the wedge of two one-forms collapses to keeping the $dx\wedge dy$ basis term and treating the coefficient as a determinant of the $(dx,dy)$-components of $\alpha$ and $\beta$ at each point. The same routine — expand, kill diagonals, reorder — drives every coordinate wedge computation in the chapters that follow. [/example] ## The Vector Calculus Dictionary in Three Dimensions How does the notation of gradients, line integrals, fluxes, and volume integrals fit into the language of forms? In $\mathbb R^3$, the usual vector calculus integrands appear as forms of degree $0,1,2,3$, but the translation uses the Euclidean metric and the chosen orientation. [explanation: Vector Calculus Dictionary In R Three] Let $U\subset\mathbb R^3$ be open with coordinates $x,y,z$. The basic dictionary is \begin{align*} \text{scalar function } f&\longleftrightarrow f\in\Omega^0(U),\\ \text{work integrand for }F=(P,Q,R)&\longleftrightarrow \alpha_F=P\,dx+Q\,dy+R\,dz\in\Omega^1(U),\\ \text{flux integrand for }F=(P,Q,R)&\longleftrightarrow \beta_F=P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy\in\Omega^2(U),\\ \text{volume integrand } g&\longleftrightarrow \gamma_g=g\,dx\wedge dy\wedge dz\in\Omega^3(U). \end{align*} A $1$-form evaluates on one tangent vector, matching the way a line integral samples a velocity vector along a curve. A $2$-form evaluates on an ordered pair of tangent vectors, matching the oriented parallelogram element used in flux. A $3$-form evaluates on an ordered triple of tangent vectors, matching signed volume. [/explanation] This dictionary explains why two different forms can be associated to the same vector field $F$: the $1$-form $\alpha_F$ is the work integrand, while the $2$-form $\beta_F$ is the flux integrand. They are related by Euclidean metric and orientation data, not by exterior algebra alone. Later, the [exterior derivative](/theorems/1525) will turn this dictionary into the familiar operations grad, curl, and div. [remark: Metric Dependence Of The Dictionary] The spaces $\Omega^k(U)$ and their wedge product are defined without an inner product. The identifications between vector fields and $1$-forms or $2$-forms in $\mathbb R^3$ use the dot product and the standard orientation. On a general manifold, these identifications require extra geometric structure. [/remark] ## Guiding Examples What examples should be kept in mind before exterior differentiation and integration are introduced? Two models are especially useful: the angular form on the punctured plane, which detects rotation around a missing point, and coordinate area forms in $\mathbb R^3$, which encode oriented projected area. [illustration:forms-angular-one-form] [example: Angle One-Form On The Punctured Plane] On the punctured plane $U=\mathbb R^2\setminus\{0\}$, define \begin{align*} d\theta=\frac{-y}{x^2+y^2}\,dx+\frac{x}{x^2+y^2}\,dy. \end{align*} The denominator $x^2+y^2$ is smooth and nonvanishing on $U$, so both coefficients are smooth and $d\theta\in\Omega^1(U)$. At a point $p=(a,b)\in U$, the formula $dx|_p(v)=v_1$, $dy|_p(v)=v_2$ on $v=(v_1,v_2)\in T_pU$ gives \begin{align*} d\theta_p(v)=\frac{-bv_1+av_2}{a^2+b^2}. \end{align*} [claim] On the radial vector $r_p=(a,b)$ and the counterclockwise angular vector $t_p=(-b,a)$, \begin{align*} d\theta_p(r_p)=0,\qquad d\theta_p(t_p)=1. \end{align*} [/claim] [proof] Substituting $(v_1,v_2)=(a,b)$ into the evaluation formula gives a numerator $-ba+ab=0$, hence $d\theta_p(r_p)=0$. Substituting $(v_1,v_2)=(-b,a)$ gives numerator $-b(-b)+a\cdot a=a^2+b^2$, which cancels the denominator to leave $d\theta_p(t_p)=1$. [/proof] The form $d\theta$ records infinitesimal angular motion: it kills radial displacement (no angle change) and reads off unit angular speed in the standard counterclockwise direction. Despite the notation, $d\theta$ is *not* globally the differential of any smooth function on $U$ — the polar angle $\theta$ has a branch cut, and the failure of $d\theta$ to be exact is exactly what de Rham cohomology will detect in $H^1(U)$. [/example] The notation $d\theta$ is local notation from polar coordinates. There is no globally defined smooth angle function $\theta:U\to\mathbb R$ on the whole punctured plane, a fact that will later reappear as the first basic example of de Rham cohomology. This is a useful warning about forms: a formula can make perfect sense as a smooth $1$-form even when it is not the differential of any globally defined smooth function. The missing origin creates a topological obstruction to choosing a single-valued angle, so differential forms can record global failure modes that scalar potentials cannot see. [illustration:forms-area-two-form-projection] [example: Coordinate Area Two-Form In R Three] On $U=\mathbb R^3$ with coordinates $x,y,z$, fix a point $p$ and tangent vectors $v=(v_1,v_2,v_3)$, $w=(w_1,w_2,w_3)$ in $T_p\mathbb R^3$. The one-form evaluations $dx|_p(v)=v_1$, $dy|_p(v)=v_2$ (and likewise on $w$) combine via $(\alpha\wedge\beta)(v,w)=\alpha(v)\beta(w)-\alpha(w)\beta(v)$ to give \begin{align*} (dx\wedge dy)|_p(v,w) =v_1w_2-v_2w_1 =\det\begin{pmatrix}v_1 & w_1\\ v_2 & w_2\end{pmatrix}. \end{align*} This is the signed area of the parallelogram spanned by the projections $\pi_{xy}(v)=(v_1,v_2)$ and $\pi_{xy}(w)=(w_1,w_2)$ onto the $xy$-plane: the $z$-components of $v$ and $w$ never enter the calculation. The same template gives the other two coordinate area forms, \begin{align*} (dy\wedge dz)|_p(v,w)=v_2w_3-v_3w_2, \qquad (dz\wedge dx)|_p(v,w)=v_3w_1-v_1w_3, \end{align*} measuring signed projected area in the $yz$- and $zx$-planes respectively. The three coordinate two-forms $dx\wedge dy$, $dy\wedge dz$, $dz\wedge dx$ together package the three signed projected areas of a parallelogram in $\mathbb R^3$ — the same data classical vector calculus stores in the cross product. This identification, taken pointwise, is the bridge that turns "flux of a vector field" into "integral of a $2$-form" in the Stokes-theorem chapter. [/example] The chapter has converted the fixed exterior algebra of a [vector space](/page/Vector%20Space) into a smooth graded algebra over every [open set](/page/Open%20Set) $U\subset\mathbb R^n$. The next operation to add is the [exterior derivative](/theorems/1525), which differentiates coefficient functions and raises degree by one while respecting the wedge product through a signed product rule. With differential forms defined on open sets of ℝⁿ, we now add the derivative operator that respects their algebraic structure. The [exterior derivative](/theorems/1525) raises degree by one, respects the wedge product through a signed rule, and squares to zero. # 3. The Exterior Derivative The [exterior derivative](/theorems/1525) is the operation that turns the algebra of differential forms into a differential complex. In the previous chapters, forms were built from alternating covectors and the wedge product; now we add the derivative compatible with that alternating algebra. The guiding point is that antisymmetry removes the symmetric second-derivative terms, so the same operator explains gradients, curls, divergences, and the obstruction theory behind de Rham cohomology. ## Axioms for Differentiating Forms What should a derivative of forms do that ordinary partial differentiation does not already do? It must raise degree by one, obey a signed product rule for the wedge product, commute with change of variables, and reduce to the usual differential on functions. If we merely differentiated coefficients without the wedge signs, the result would depend on the chosen coordinate expression and the cancellation behind $d^2=0$ would no longer be forced by antisymmetry. [definition: Degree Plus One Antiderivation] Let $U \subset \mathbb R^n$ be open. A family of $\mathbb R$-linear maps $(D_k:\Omega^k(U) \to \Omega^{k+1}(U))_{k\geq 0}$ is a degree plus one antiderivation if, for every $\omega \in \Omega^k(U)$ and every $\eta \in \Omega^\ell(U)$, \begin{align*} D(\omega \wedge \eta)=D\omega \wedge \eta+(-1)^k\omega \wedge D\eta. \end{align*} [/definition] The sign records the degree of the form past which the derivative moves. It is the same sign convention that makes the wedge product graded-commutative. [quotetheorem:1525] This theorem is the reason the notation $d$ is used without choosing coordinates. The coordinate formula is a way to compute the operator, while the axioms explain why the result is invariant under smooth changes of variables. Naturality is essential on manifolds: without compatibility with pullback, formulas written in two overlapping charts might define different forms. Uniqueness matters because it says these requirements leave no room for a second [exterior derivative](/theorems/1525) with different signs or extra terms. The condition $d^2=0$ is what turns the spaces of forms into a complex, so dropping it would lose the later connection with closed forms, exact forms, and cohomology. ## The Coordinate Formula How do we compute $d\omega$ once a form is written in coordinates? The answer is to differentiate only the coefficient functions and wedge the resulting $1$-forms onto the existing basis wedge. For an increasing multi-index $I=(i_1<\cdots<i_k)$, write \begin{align*} dx_I=dx_{i_1}\wedge\cdots\wedge dx_{i_k}. \end{align*} Every $k$-form on $U\subset\mathbb R^n$ has a unique expression \begin{align*} \omega=\sum_{|I|=k} a_I dx_I, \end{align*} with $a_I\in C^\infty(U)$. [quotetheorem:3564] The formula is local, but it is compatible on overlaps because of naturality. On a manifold, this lets us compute $d$ in any chart $(U,\varphi)$ and obtain a globally defined form. The assumption that the coefficients are smooth ensures that all derivatives appearing in the formula exist and that the mixed second derivatives needed for $d^2=0$ agree. The increasing multi-index convention gives each basis wedge a preferred order, so signs are tracked by reordering rather than by duplicating terms. The same ordering convention reappears when top-degree forms are integrated with an orientation. [example: Exterior Derivative Of A One Form] On $\mathbb R^3$ with coordinates $(x,y,z)$, let $\omega=P\,dx+Q\,dy+R\,dz$ with $P,Q,R\in C^\infty(\mathbb R^3)$. We compute $d\omega$ using $\mathbb R$-linearity, the signed product rule $d(f\,\eta)=df\wedge\eta+f\,d\eta$, and the identity $d(dx^i)=0$ for each coordinate. Because $d^2=0$ on coordinate functions, every term $P\,d(dx)$, $Q\,d(dy)$, $R\,d(dz)$ vanishes, leaving \begin{align*} d\omega=dP\wedge dx+dQ\wedge dy+dR\wedge dz. \end{align*} Expand $dP=\partial_xP\,dx+\partial_yP\,dy+\partial_zP\,dz$ and likewise for $dQ$, $dR$. Each wedge with a single coordinate collapses three of the nine terms via $dx^i\wedge dx^i=0$ and rewrites the remaining two using $dx^j\wedge dx^i=-dx^i\wedge dx^j$: \begin{align*} dP\wedge dx &=\partial_zP\,dz\wedge dx-\partial_yP\,dx\wedge dy,\\ dQ\wedge dy &=-\partial_zQ\,dy\wedge dz+\partial_xQ\,dx\wedge dy,\\ dR\wedge dz &=\partial_yR\,dy\wedge dz-\partial_xR\,dz\wedge dx. \end{align*} Summing and collecting by basis $2$-form gives \begin{align*} d\omega =\left(\frac{\partial R}{\partial y}-\frac{\partial Q}{\partial z}\right)dy\wedge dz +\left(\frac{\partial P}{\partial z}-\frac{\partial R}{\partial x}\right)dz\wedge dx +\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy. \end{align*} In the ordered basis $dy\wedge dz$, $dz\wedge dx$, $dx\wedge dy$ — the standard $2$-form basis on $\mathbb R^3$ — these three coefficients are exactly the components of $\nabla\times(P,Q,R)$. The cross-partial differences are not chosen; they are forced by the antisymmetry $dx^j\wedge dx^i=-dx^i\wedge dx^j$. This is the first sign that classical vector calculus has the exterior derivative hiding behind it. [/example] This example is the first sign that the [exterior derivative](/theorems/1525) contains curl. The signs are not an added convention; they are forced by alternating multiplication. [example: Exterior Derivative Of A Two Form] On $\mathbb R^3$ with coordinates $(x,y,z)$, let \begin{align*} \beta=P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy, \qquad P,Q,R\in C^\infty(\mathbb R^3). \end{align*} By $\mathbb R$-linearity and the signed product rule, $d\beta=dP\wedge dy\wedge dz+dQ\wedge dz\wedge dx+dR\wedge dx\wedge dy$ — the trailing $d(dy\wedge dz)$-type terms vanish because $d$ kills each coordinate one-form. Expand each $dP=\partial_xP\,dx+\partial_yP\,dy+\partial_zP\,dz$ (and likewise $dQ$, $dR$) and wedge against the corresponding $2$-form. In $\mathbb R^3$ there is only one independent $3$-form, $dx\wedge dy\wedge dz$, and any wedge of three coordinate one-forms with a repeat is zero. Only the partial derivative *transverse* to the trailing two-form survives: \begin{align*} dP\wedge dy\wedge dz &=\partial_xP\,dx\wedge dy\wedge dz,\\ dQ\wedge dz\wedge dx &=\partial_yQ\,dy\wedge dz\wedge dx=\partial_yQ\,dx\wedge dy\wedge dz,\\ dR\wedge dx\wedge dy &=\partial_zR\,dz\wedge dx\wedge dy=\partial_zR\,dx\wedge dy\wedge dz, \end{align*} using two transpositions in each of the last two rows to put the basis $3$-form in standard order. Summing, \begin{align*} d\beta =\left(\frac{\partial P}{\partial x}+\frac{\partial Q}{\partial y}+\frac{\partial R}{\partial z}\right)dx\wedge dy\wedge dz. \end{align*} Under the identification of a vector field $(P,Q,R)$ with the flux $2$-form $P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy$, the scalar coefficient is exactly $\nabla\cdot(P,Q,R)$. Together with the curl-formula in the previous example, this gives both halves of the de Rham translation table on $\mathbb R^3$: $d$ on $1$-forms is curl, $d$ on $2$-forms is divergence. The composite $d^2=0$ then becomes $\nabla\cdot(\nabla\times F)=0$ for free. [/example] The second example shows the divergence pattern. The coordinate formula therefore packages several vector calculus operations into one construction. ## Closed And Exact Forms When does the differential equation $d\eta=\omega$ have a solution, and what condition must $\omega$ satisfy before such a solution can exist? The identity $d^2=0$ supplies the universal compatibility condition. [definition: Closed Differential Form] Let $U\subset\mathbb R^n$ be open, and let $\omega\in\Omega^k(U)$. The form $\omega$ is closed if \begin{align*} d\omega=0. \end{align*} [/definition] Closure is a differential condition on the coefficients of $\omega$. For a $1$-form it says that the mixed first derivatives line up in the pattern detected by curl. [definition: Exact Differential Form] Let $U\subset\mathbb R^n$ be open, and let $\omega\in\Omega^k(U)$. The form $\omega$ is exact if there exists $\eta\in\Omega^{k-1}(U)$ such that \begin{align*} \omega=d\eta. \end{align*} [/definition] Exactness is a solvability condition: $\eta$ is a potential for $\omega$. The central question of de Rham cohomology is how far closed forms are from being exact. [quotetheorem:3565] The converse is both a local question and a global question. Closedness is checked locally from the coefficients of $d\omega$, while exactness asks for a single potential defined on the whole domain. Locally, the Poincare lemma later proves that closed forms have potentials on star-shaped regions; globally, holes in the domain can obstruct the existence of a potential. De Rham cohomology records exactly this obstruction by measuring closed forms modulo exact forms. [example: Closed One Form On The Punctured Plane] On $U=\mathbb R^2\setminus\{0\}$, define \begin{align*} \omega=\frac{-y}{x^2+y^2}\,dx+\frac{x}{x^2+y^2}\,dy=P\,dx+Q\,dy. \end{align*} Both coefficients are smooth on $U$ since $x^2+y^2$ is positive there. [claim] $\omega$ is closed on $U$ but not exact on $U$. [/claim] [proof] **Closedness.** The general formula $d(P\,dx+Q\,dy)=(\partial_xQ-\partial_yP)\,dx\wedge dy$ requires the two cross-partials. Differentiating $Q=x(x^2+y^2)^{-1}$ in $x$, \begin{align*} \frac{\partial Q}{\partial x} =\frac{1}{x^2+y^2}-\frac{2x^2}{(x^2+y^2)^2} =\frac{y^2-x^2}{(x^2+y^2)^2}. \end{align*} Differentiating $P=-y(x^2+y^2)^{-1}$ in $y$, \begin{align*} \frac{\partial P}{\partial y} =-\frac{1}{x^2+y^2}+\frac{2y^2}{(x^2+y^2)^2} =\frac{y^2-x^2}{(x^2+y^2)^2}. \end{align*} The two are equal, so $d\omega=0$. **Non-exactness.** Consider the unit circle $\gamma:[0,2\pi]\to U$, $\gamma(t)=(\cos t,\sin t)$. Along $\gamma$ the denominator $x^2+y^2=1$, and $\gamma^*\omega=\sin^2 t\,dt+\cos^2 t\,dt=dt$, so \begin{align*} \int_\gamma \omega=\int_0^{2\pi}dt=2\pi. \end{align*} If $\omega=df$ for some $f\in C^\infty(U)$, the fundamental theorem of calculus along $\gamma$ would force $\int_\gamma\omega=f(\gamma(2\pi))-f(\gamma(0))=f(1,0)-f(1,0)=0$. Since $2\pi\neq 0$, no such global potential exists. [/proof] The equality $d\omega=0$ is purely local — the cross-partial test is satisfied at every point of $U$. Exactness is a global question, and it fails because the unit circle is a closed loop that *cannot* be contracted to a point inside $U$: the hole at the origin is in the way. This is the prototypical generator of $H^1(\mathbb R^2\setminus\{0\})\cong\mathbb R$, and the difference between $d\omega=0$ and $\omega=df$ is exactly what de Rham cohomology measures. [/example] This example motivates the quotient of closed forms by exact forms. That quotient is the de Rham cohomology group, introduced after integration and homotopy invariance are available. ## Gradient Curl And Divergence Why do the identities $\nabla\times\nabla f=0$ and $\nabla\cdot(\nabla\times F)=0$ have the same proof? After choosing the Euclidean metric and the standard orientation on $\mathbb R^3$, they are both the single identity $d^2=0$ written in different degrees. A smooth function $f\in C^\infty(\mathbb R^3)$ is a $0$-form. A vector field $F=(P,Q,R)$ may be encoded either as the $1$-form \begin{align*} \alpha_F=P\,dx+Q\,dy+R\,dz \end{align*} or as the $2$-form \begin{align*} \beta_F=P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy. \end{align*} The first encoding is used for curl, and the second encoding is used for divergence. [quotetheorem:3566] The theorem also explains why the classical operators occur in the order gradient, curl, divergence. They are the degree-by-degree pieces of the de Rham complex on $\mathbb R^3$: \begin{align*} 0\longrightarrow \Omega^0(\mathbb R^3)\xrightarrow{d}\Omega^1(\mathbb R^3)\xrightarrow{d}\Omega^2(\mathbb R^3)\xrightarrow{d}\Omega^3(\mathbb R^3)\longrightarrow 0. \end{align*} [example: A Curl Computation] On $\mathbb R^3$, take $F=(yz,xz,xy)$ and encode it as the $1$-form \begin{align*} \alpha_F=yz\,dx+xz\,dy+xy\,dz. \end{align*} Applying the coordinate formula $d\alpha_F=(\partial_yR-\partial_zQ)\,dy\wedge dz+(\partial_zP-\partial_xR)\,dz\wedge dx+(\partial_xQ-\partial_yP)\,dx\wedge dy$ with $P=yz$, $Q=xz$, $R=xy$, each coefficient is a difference of two equal cross-partials: \begin{align*} \partial_yR-\partial_zQ=x-x=0, \quad \partial_zP-\partial_xR=y-y=0, \quad \partial_xQ-\partial_yP=z-z=0. \end{align*} So $d\alpha_F=0$, and under the encoding, $\nabla\times F=0$. The cancellations are not coincidence: $F$ is a gradient. Setting $f=xyz$, \begin{align*} \nabla f =(\partial_xf,\partial_yf,\partial_zf) =(yz,xz,xy) =F. \end{align*} Since $\mathbb R^3$ is contractible, the Poincaré lemma in the next chapter will guarantee that *every* closed $1$-form on $\mathbb R^3$ is exact, so curl-free vector fields and gradient fields coincide globally on $\mathbb R^3$ — a clean fact that fails on the punctured plane, as the previous example showed. [/example] The curl example illustrates the degree-$1$ part of the complex: a $1$-form is differentiated into a $2$-form, and vanishing means that a local potential may exist. The divergence example moves one step further, from $2$-forms to top-degree $3$-forms, where the output is a scalar multiple of the oriented volume form. Reading the two examples together shows how $d$ links the vector calculus operators in sequence rather than treating them as unrelated formulas. [example: A Divergence Computation] On $\mathbb R^3$, take $F=(x^2,y^2,z^2)$ and encode it as the flux $2$-form \begin{align*} \beta_F=x^2\,dy\wedge dz+y^2\,dz\wedge dx+z^2\,dx\wedge dy. \end{align*} The general formula $d(P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy)=(\partial_xP+\partial_yQ+\partial_zR)\,dx\wedge dy\wedge dz$ applies with $P=x^2$, $Q=y^2$, $R=z^2$, giving partials $2x$, $2y$, $2z$ and therefore \begin{align*} d\beta_F=2(x+y+z)\,dx\wedge dy\wedge dz. \end{align*} Under the flux encoding, $\nabla\cdot F=2x+2y+2z$. The divergence is not constant: it grows linearly outward in every direction, which is what one expects from a vector field whose components grow quadratically along their own axis. The same form will reappear in the Stokes-theorem chapter, where the integral of $d\beta_F$ over a region equals the flux $\int_{\partial\Omega}\beta_F$ through its boundary. [/example] The [exterior derivative](/theorems/1525) is therefore not a new vector calculus operator beside grad, curl, and div. It is the coordinate-free operator whose shadows in dimension three are precisely those familiar operations, and whose square-zero property is the algebraic source of the compatibility conditions used throughout the rest of the course. Beyond the [exterior derivative](/theorems/1525), we must understand how forms behave under smooth maps between spaces. Pullback is the operation that transports a differential form from a target space back to a source space, and it will be compatible with the [exterior derivative](/theorems/1525) in a way that underpins [Stokes' theorem](/theorems/1530). # 4. Pullback of Forms Pullback answers the question: how does a differential form on a target space become a form on a source space when we have a smooth map into that target? The construction is forced by the interpretation of a $k$-form as something that takes $k$ tangent vectors as input. A map pushes tangent vectors forward by its total derivative, so a form pulls back by first pushing the input vectors forward and then evaluating the original form. This chapter develops that operation, proves its compatibility with wedge products and exterior derivatives, and explains why the determinant in multivariable change of variables is a top-degree pullback coefficient. ## 1. Pulling Forms Back Along Smooth Maps What data are needed to measure a $k$-dimensional infinitesimal parallelepiped in the source using a form defined on the target? Suppose $U$ and $V$ are open subsets of Euclidean spaces and $f: U \to V$ is smooth. A tangent vector at $x \in U$ is sent to a tangent vector at $f(x)$ by $Df_x$, so the natural way to evaluate a target form on source vectors is to feed their images into the target form. [definition: Pullback Of A Differential Form] Let $U \subseteq \mathbb{R}^m$ and $V \subseteq \mathbb{R}^n$ be open sets, let $f: U \to V$ be smooth, and let $\omega \in \Omega^k(V)$. The pullback of $\omega$ by $f$ is the $k$-form $f^*\omega \in \Omega^k(U)$ defined by \begin{align*} (f^*\omega)_x(v_1,\dots,v_k) = \omega_{f(x)}(Df_x(v_1),\dots,Df_x(v_k)) \end{align*} for $x \in U$ and $v_1,\dots,v_k \in \mathbb{R}^m$. [/definition] The definition is pointwise, but smoothness of $f$ ensures that the resulting coefficients vary smoothly in $x$. For $0$-forms, which are smooth functions, this recovers composition: if $g \in C^\infty(V)$, then $f^*g = g \circ f$. The same formula works for smooth maps $F:M\to N$ between manifolds, replacing $Df_x$ by the differential $dF_p:T_pM\to T_{F(p)}N$. [example: Pullback Of A One-Form On The Plane] Let $f:\mathbb R^2\to\mathbb R^2$, $f(u,v)=(u^2-v,uv)$, with source coordinates $(u,v)$ and target coordinates $(x,y)$, and take $\omega=y\,dx+x\,dy$ on the target. Pullback acts on coefficient functions by composition and on coordinate differentials by $f^*(dx^i)=d(f^*x^i)$, so $f^*x=u^2-v$, $f^*y=uv$, $f^*(dx)=d(u^2-v)=2u\,du-dv$, and $f^*(dy)=d(uv)=v\,du+u\,dv$. Substituting, \begin{align*} f^*\omega &=f^*y\,f^*(dx)+f^*x\,f^*(dy)\\ &=(uv)(2u\,du-dv)+(u^2-v)(v\,du+u\,dv)\\ &=(2u^2v+u^2v-v^2)\,du+(-uv+u^3-uv)\,dv\\ &=(3u^2v-v^2)\,du+(u^3-2uv)\,dv. \end{align*} Two things happened simultaneously: the coefficient functions $x,y$ on the target were replaced by their compositions with $f$, and the target differentials $dx,dy$ were replaced by the differentials of those composed functions. This is the universal recipe — pullback of a coordinate form is naturally a $C^\infty$-linear combination of source differentials, with coefficients computed once one knows the Jacobian of $f$. [/example] This example already suggests the two parts of every coordinate calculation: substitute through $f$, then apply the differential to the coordinate functions of $f$. Before using this as a general operation, however, we must check that the pointwise formula really produces another smooth differential form. The issue is not alternating multilinearity, which is inherited from the original form, but smooth dependence on the base point after the vectors have been pushed forward by $Df_x$. [quotetheorem:3567] This result lets us use $f^*$ as an operator $\Omega^k(V) \to \Omega^k(U)$, not merely as a pointwise formula. Smoothness of $f$ is essential here: a merely continuous map has no derivative with which to push tangent vectors forward, and even a differentiable map with poorly behaved derivative need not produce smooth coefficients. Degenerate points of $Df_x$ cause no problem; they simply make some pulled-back forms vanish on directions that are collapsed by the map. The next issue is whether pullback respects the algebraic structure built in the first chapters. ## 2. Algebraic Behaviour Of Pullback If differential forms are multiplied by wedge product, what kind of map should pullback be on the exterior algebra? Since pullback applies the same derivative $Df_x$ to every input vector before evaluation, it should preserve all alternating algebraic operations. It also reverses the direction of maps: a map $f: U \to V$ gives a map on forms from $V$ back to $U$. [quotetheorem:3568] Functoriality is the formal reason pullback is the right operation for changing coordinates. It says that changing from one coordinate system to another and then to a third gives the same result as changing directly. The direction reversal is important: maps of spaces compose as $U \to V \to W$, while the induced maps on forms go $\Omega^k(W) \to \Omega^k(V) \to \Omega^k(U)$. This contravariance is also the pattern that later reappears when smooth maps induce maps on de Rham cohomology. [quotetheorem:3569] This theorem says that every algebraic identity among forms survives pullback. For instance, if $\omega\wedge\omega=0$ because $\omega$ has odd degree, then $f^*\omega\wedge f^*\omega=0$ as well. [example: Pullback Under The Circle Inclusion] Let $\iota:S^1\hookrightarrow\mathbb R^2\setminus\{0\}$ be the inclusion, parametrised on the angular chart by $\iota(t)=(\cos t,\sin t)$, and consider the angle form $d\theta=(-y\,dx+x\,dy)/(x^2+y^2)$ on $\mathbb R^2\setminus\{0\}$. Pulling back through $\iota$ gives $\iota^*x=\cos t$, $\iota^*y=\sin t$, $\iota^*(dx)=-\sin t\,dt$, $\iota^*(dy)=\cos t\,dt$. The numerator becomes \begin{align*} \iota^*(-y\,dx+x\,dy) =-\sin t(-\sin t\,dt)+\cos t(\cos t\,dt) =(\sin^2 t+\cos^2 t)\,dt =dt, \end{align*} and the denominator $\iota^*(x^2+y^2)=\cos^2 t+\sin^2 t=1$ is constant, so $\iota^*(d\theta)=dt$. The seemingly mysterious form $d\theta$ on the punctured plane is, when restricted to the unit circle, just the angular differential $dt$ — exactly what its name promises. The point of $d\theta$ is precisely that it *agrees* with $dt$ on every circle around the origin but cannot be patched to a single function $\theta$ on $\mathbb R^2\setminus\{0\}$, because $t$ has a branch cut. [/example] The form $d\theta$ is locally the differential of an angle function, but it is not globally the differential of a single-valued smooth function on the punctured plane. Pullback to the circle records the winding behaviour that later becomes visible in de Rham cohomology. ## 3. Pullback And The Exterior Derivative Can differentiation of forms commute with changing variables? It must do so if the [exterior derivative](/theorems/1525) is to be independent of the coordinates used to compute it. The central compatibility result is naturality of $d$: pull back first and then differentiate, or differentiate first and then pull back, and the result is the same. [quotetheorem:1525] This theorem is the reason closed and exact forms behave well under smooth maps. If $d\omega=0$, then $d(f^*\omega)=0$; if $\omega=d\eta$, then $f^*\omega=d(f^*\eta)$. [remark: Pullback Preserves Closed And Exact Forms] For a smooth map $f: U \to V$, pullback sends closed forms on $V$ to closed forms on $U$ and exact forms on $V$ to exact forms on $U$. Therefore $f^*$ descends to a map on de Rham cohomology groups, once those groups have been defined. [/remark] The remark previews the role of pullback in cohomology. Smooth maps between spaces will induce linear maps between cohomology groups in the opposite direction. ## 4. Coordinate Formulae How do we compute pullbacks without returning to the multilinear definition each time? The answer is to express a form in coordinate differentials and replace each target coordinate function by its component under $f$. This produces a compact formula that is the practical workhorse for examples. [quotetheorem:3570] This formula is often the shortest path through a computation. It also records dimension effects automatically: if $k>m$, then every pulled-back $k$-form on $U\subseteq\mathbb{R}^m$ is zero because there are not enough independent coordinate differentials on the source. The map $f$ need not be injective for pullback to make sense; non-injectivity affects global integration, but the pulled-back form is still defined pointwise from $Df_x$. The same coordinate formula also shows why the Jacobian appears when pulling back top-degree forms. [quotetheorem:3571] The determinant is therefore not an additional ingredient imposed on integration. It is the coefficient by which a smooth map rescales the oriented top-degree form. [illustration:forms-polar-coordinate-sector] [example: Polar Coordinates And Area] Let $f:(0,\infty)\times(0,2\pi)\to\mathbb R^2\setminus\{(x,0):x\ge 0\}$, $f(r,\theta)=(r\cos\theta,r\sin\theta)$. The pullbacks of the coordinate differentials follow from the product rule: \begin{align*} f^*(dx)=d(r\cos\theta)=\cos\theta\,dr-r\sin\theta\,d\theta, \qquad f^*(dy)=d(r\sin\theta)=\sin\theta\,dr+r\cos\theta\,d\theta. \end{align*} Since pullback commutes with $\wedge$, \begin{align*} f^*(dx\wedge dy) &=(\cos\theta\,dr-r\sin\theta\,d\theta)\wedge(\sin\theta\,dr+r\cos\theta\,d\theta). \end{align*} Expanding and killing the diagonal $dr\wedge dr$ and $d\theta\wedge d\theta$ terms, the $dr\wedge d\theta$ coefficient assembles as $r\cos^2\theta+r\sin^2\theta=r$. Hence \begin{align*} f^*(dx\wedge dy)=r\,dr\wedge d\theta. \end{align*} The factor $r$ is no coincidence: it is exactly the determinant of the Jacobian $\det Jf=r\cos^2\theta+r\sin^2\theta=r$. The general pattern that pullback of a top-degree form picks up the Jacobian determinant is the engine of the change-of-variables theorem for integration on manifolds — and it is the reason the elementary "polar area element $r\,dr\,d\theta$" appears in every multivariable calculus course without explanation. Here the explanation is structural: it falls out of the wedge product axioms. [/example] Coordinate formulae also make sense for maps that are not between spaces of the same dimension. Pulling a top-degree form from a higher-dimensional target to a lower-dimensional source produces the form that measures the target quantity along the parametrised source. [illustration:forms-inverse-stereographic-area] [example: Stereographic Projection And The Area Form] Let $\sigma:\mathbb R^2\to S^2\setminus\{N\}$ be inverse stereographic projection from the north pole $N=(0,0,1)$, \begin{align*} \sigma(u,v)=\frac{1}{s}\bigl(2u,2v,u^2+v^2-1\bigr),\qquad s=1+u^2+v^2. \end{align*} Let $\omega_{S^2}$ denote the Riemannian area form on $S^2$, oriented so that $(u,v)$ is positively oriented. [claim] \begin{align*} \sigma^*\omega_{S^2}=\frac{4}{(1+u^2+v^2)^2}\,du\wedge dv. \end{align*} [/claim] [proof] The coordinate tangent vectors $\sigma_u,\sigma_v\in T_{\sigma(u,v)}S^2$ are obtained by component-wise differentiation of $\sigma$. Quotient-rule calculation (using $\partial_us=2u$, $\partial_vs=2v$) yields \begin{align*} \sigma_u=\tfrac{1}{s^2}\bigl(2(1-u^2+v^2),-4uv,4u\bigr), \qquad \sigma_v=\tfrac{1}{s^2}\bigl(-4uv,2(1+u^2-v^2),4v\bigr). \end{align*} For a positively oriented pair on an oriented Riemannian surface, $\omega(A,B)=\sqrt{(A\cdot A)(B\cdot B)-(A\cdot B)^2}$, so the task reduces to three inner products. Expanding $\sigma_u\cdot\sigma_u$ and grouping powers of $u,v$, \begin{align*} \sigma_u\cdot\sigma_u &=\frac{4(1-u^2+v^2)^2+16u^2v^2+16u^2}{s^4} =\frac{4(1+u^2+v^2)^2}{s^4} =\frac{4}{s^2}. \end{align*} The bracketed identity $(1-u^2+v^2)^2+4u^2v^2+4u^2=(1+u^2+v^2)^2$ verifies on inspection of the constant, quadratic, and quartic terms. By the same algebra (swap roles of $u$ and $v$), $\sigma_v\cdot\sigma_v=4/s^2$. The mixed product simplifies similarly: \begin{align*} \sigma_u\cdot\sigma_v =\frac{-8uv\bigl((1-u^2+v^2)+(1+u^2-v^2)\bigr)+16uv}{s^4} =\frac{-16uv+16uv}{s^4}=0, \end{align*} so the coordinate frame is orthogonal and conformal — both partials have the same length $2/s$. Plugging into the area-form formula, \begin{align*} \omega_{S^2}(\sigma_u,\sigma_v) =\sqrt{\tfrac{4}{s^2}\cdot\tfrac{4}{s^2}-0} =\frac{4}{s^2}. \end{align*} By the definition of pullback, $(\sigma^*\omega_{S^2})(\partial_u,\partial_v)=\omega_{S^2}(\sigma_u,\sigma_v)=4/s^2$, and since $du\wedge dv(\partial_u,\partial_v)=1$, the coefficient of $du\wedge dv$ is $4/s^2=4/(1+u^2+v^2)^2$. [/proof] The pulled-back area form makes the conformal nature of stereographic projection visible: $\sigma_u$ and $\sigma_v$ are orthogonal and have equal length $2/s$ at every point, so $\sigma$ is angle-preserving but stretches area by the factor $4/s^2$. The total spherical area $\int_{\mathbb R^2}4/(1+u^2+v^2)^2\,du\,dv=4\pi$ falls out as the integral of this density — a clean computation of the area of $S^2$ that motivates why stereographic coordinates are the right tool for many problems on the round sphere. [/example] The stereographic formula is a model for later manifold calculations: even when the form lives geometrically on a curved space, pullback to a coordinate chart turns it into an ordinary differential form on an open subset of Euclidean space. ## 5. Change Of Variables From Pullback Why does the change-of-variables theorem place an absolute value around the Jacobian determinant, while pullback of the volume form gives an oriented determinant? Differential forms integrate over oriented domains, so orientation-preserving maps contribute $\det(Jf_x)$ and orientation-reversing maps contribute a negative sign. The classical theorem for non-oriented volume uses the positive density $|\det(Jf_x)|$. [quotetheorem:3554] This theorem recovers the familiar formula for ordinary multiple integrals by replacing the oriented form with its associated density. [example: Orientation Reversal On The Line] Let $f:(-1,1)\to(-1,1)$, $f(x)=-x$, with source coordinate $x$ and target coordinate $y$, and take $\omega=a(y)\,dy$ with $a$ compactly supported on $(-1,1)$. Then $f^*y=-x$ and $f^*(dy)=d(-x)=-dx$, so \begin{align*} f^*\omega=a(-x)\cdot(-dx)=-a(-x)\,dx. \end{align*} The minus sign is the *signature* of $f$ as an orientation-reversing diffeomorphism — it does not appear because of any choice of integration convention, only because of how $d$ acts on $-x$. Integrating with the standard orientation on both intervals, \begin{align*} \int_{-1}^{1}f^*\omega =-\int_{-1}^{1}a(-x)\,dx =-\int_{-1}^{1}a(y)\,dy =-\int_{-1}^{1}\omega, \end{align*} where the middle step substitutes $y=-x$ (one sign from $dx=-dy$, one from swapping the limits). The minus sign here is the oriented change-of-variables theorem at work: an orientation-reversing diffeomorphism contributes a global minus sign to the integral of a top-degree form. Compare with the *unsigned* Lebesgue change of variables, which uses $|f'(x)|=1$: \begin{align*} \int_{-1}^{1}a(y)\,d\mathcal L^1(y) =\int_{-1}^{1}a(-x)\,d\mathcal L^1(x). \end{align*} No sign. The two formulas measure different things: pullback-and-integrate respects orientation, Lebesgue integration measures unsigned mass. This distinction is the reason differential forms — not just functions weighted by Jacobians — are the right objects to integrate on manifolds when orientation matters, as it does for Stokes' theorem and the de Rham comparison theorem later in the course. [/example] The line example is the simplest case of the general phenomenon. Pullback keeps track of whether a parametrisation preserves or reverses the chosen orientation, whereas densities record only unsigned size. [quotetheorem:22] The absolute value is therefore the price of forgetting orientation. Differential forms keep the sign because they are designed to integrate over oriented parametrised objects. [remark: What Pullback Prepares] Pullback is the operation that makes integration on manifolds local. To integrate a form over a parametrised curve, surface, or coordinate chart, we pull the form back to a [Euclidean domain](/page/Euclidean%20Domain) and integrate there. The compatibility $f^*(d\omega)=d(f^*\omega)$ will later be the mechanism behind [Stokes' theorem](/theorems/1530) and the induced maps on de Rham cohomology. [/remark] So far, all constructions—exterior algebra, differential forms, pullback, and the [exterior derivative](/theorems/1525)—have lived in coordinates on open subsets of ℝⁿ. To fully unlock the geometric meaning of these tools, we must reformulate them on abstract manifolds, where forms become genuinely coordinate-free geometric objects. # 5. Smooth Manifolds and Forms on Manifolds Smooth manifolds are the setting in which differential forms stop being coordinate-dependent formulae and become geometric objects. The preceding chapters built exterior algebra, forms on open subsets of Euclidean space, pullback, wedge product, and [exterior derivative](/theorems/1525) in coordinates. This chapter explains how those constructions survive a change of chart, how they assemble on an abstract manifold, and why the same algebraic and differential operations are available globally. The guiding question is: if a manifold is only locally an open subset of $\mathbb R^n$, what data must be required so that a $k$-form written in one coordinate system is the same geometric object as the corresponding expression in another? The answer is compatibility under pullback by smooth transition maps. Once that compatibility is in place, $\Omega^*(M)$ becomes a graded algebra and $d$ becomes a global operator, giving the language needed for integration and [Stokes' theorem](/theorems/1530) in the next part of the course. ## Smooth Atlases and the Cotangent Bundle How much structure is needed on a topological space before calculus makes sense on it? A chart identifies a small part of the space with an open subset of $\mathbb R^n$, but calculus also requires that overlapping charts use compatible coordinate changes. [definition: Smooth Chart] Let $M$ be a topological space. An $n$-dimensional chart on $M$ is a pair $(U,\varphi)$ such that $U \subset M$ is open, $\varphi: U \to \varphi(U) \subseteq \mathbb R^n$ is a homeomorphism, and $\varphi(U)$ is open in $\mathbb R^n$. [/definition] A chart gives local coordinates $x_1,\dots,x_n$ on $U$ by writing $\varphi(p)=(x_1(p),\dots,x_n(p))$. The point is not that $M$ is a subset of Euclidean space, but that near each point it admits a coordinate description in which limits, derivatives, and forms can be computed. The danger is that two coordinate systems can describe the same points while giving incompatible calculus; the next condition rules out such fake coordinate changes. [definition: Smoothly Compatible Charts] Two $n$-dimensional charts $(U,\varphi)$ and $(V,\psi)$ on $M$ are smoothly compatible if either $U \cap V=\varnothing$, or the transition map \begin{align*} \psi \circ \varphi^{-1}:\varphi(U\cap V)\longrightarrow \psi(U\cap V) \end{align*} is a diffeomorphism between open subsets of $\mathbb R^n$. [/definition] The transition map is the change of coordinates from the $\varphi$-coordinates to the $\psi$-coordinates. Smooth compatibility is what permits a derivative computed in one chart to be transported to another chart without changing the underlying geometric statement. [definition: Smooth Manifold] A smooth $n$-manifold is a second-countable Hausdorff topological space $M$ equipped with a maximal atlas $\mathcal A$ of smoothly compatible $n$-dimensional charts whose domains cover $M$. [/definition] The maximality condition means that every chart smoothly compatible with the atlas is already included. In practice, one usually specifies a smaller atlas and understands the smooth structure to be the maximal atlas it generates. [example: Standard Smooth Structure on the Sphere] Let $S^n\subset\mathbb R^{n+1}$ be the unit sphere with north pole $N=(0,\dots,0,1)$ and south pole $S=(0,\dots,0,-1)$. Define the open covers $U_N=S^n\setminus\{N\}$, $U_S=S^n\setminus\{S\}$, and stereographic projections \begin{align*} \sigma_N(x',x_{n+1})=\frac{x'}{1-x_{n+1}}, \qquad \sigma_S(x',x_{n+1})=\frac{x'}{1+x_{n+1}}, \end{align*} where $x'=(x_1,\dots,x_n)$. The denominators are nonzero on the indicated domains, so the formulas are well defined. [claim] $(U_N,\sigma_N)$ and $(U_S,\sigma_S)$ form a smooth atlas on $S^n$. The maximal atlas it generates is the standard smooth structure. [/claim] [proof] **Charts.** The inverse of $\sigma_N$ is the map $\tau_N(u)=(2u/(|u|^2+1),(|u|^2-1)/(|u|^2+1))$: substituting and using $4|u|^2+(|u|^2-1)^2=(|u|^2+1)^2$ shows $|\tau_N(u)|=1$, so the image lies in $S^n$, and the last coordinate equals $1$ only if $-1=1$, so $\tau_N(u)\in U_N$. The compositions $\sigma_N\circ\tau_N(u)=u$ and $\tau_N\circ\sigma_N(x)=x$ verify by direct algebraic computation, using the identity $|u|^2+1=2/(1-x_{n+1})$ when $u=\sigma_N(x)$. The analogous formula $\tau_S(v)=(2v/(|v|^2+1),(1-|v|^2)/(|v|^2+1))$ inverts $\sigma_S$. **Transition map.** On the overlap $U_N\cap U_S=S^n\setminus\{N,S\}$, the image $\sigma_N(U_N\cap U_S)=\mathbb R^n\setminus\{0\}$ (the removed south pole maps to $u=0$). For $u\neq 0$, \begin{align*} (\sigma_S\circ\tau_N)(u) =\frac{2u/(|u|^2+1)}{1+(|u|^2-1)/(|u|^2+1)} =\frac{2u/(|u|^2+1)}{2|u|^2/(|u|^2+1)} =\frac{u}{|u|^2}. \end{align*} The transition $u\mapsto u/|u|^2$ is its own inverse — substituting into itself returns $u$ — and each component $u_i/(u_1^2+\cdots+u_n^2)$ is a quotient of polynomials with nonvanishing denominator on $\mathbb R^n\setminus\{0\}$, hence smooth with smooth inverse. The two charts therefore form a smooth atlas. [/proof] This is the universal construction of a smooth structure by overlapping charts: two coordinate maps whose transition function is a diffeomorphism of an open subset of $\mathbb R^n$. The transition $u\mapsto u/|u|^2$ is geometrically the inversion in the unit sphere in $\mathbb R^n$ — a fact that reflects the conformal nature of stereographic projection seen earlier with $\mathbb R^2\to S^2$. [/example] The cotangent bundle packages all covectors on all tangent spaces into a single geometric object. Differential forms will be sections of exterior powers of this bundle. [definition: Cotangent Bundle] Let $M$ be a smooth $n$-manifold. The cotangent bundle of $M$ is \begin{align*} T^*M=\bigsqcup_{p\in M} T_p^*M, \end{align*} where $T_p^*M$ is the dual [vector space](/page/Vector%20Space) of the tangent space $T_pM$. [/definition] In a chart $(U,\varphi)$ with coordinates $x_1,\dots,x_n$, the coordinate covectors $dx_1|_p,\dots,dx_n|_p$ form a basis of $T_p^*M$ for each $p\in U$. Thus a local covector field can be written as $\sum_i a_i\,dx_i$, with coefficient functions $a_i:U\to\mathbb R$. [example: Cotangent Coordinates on the Circle] Let $\theta:U\to\mathbb R$ and $\tilde\theta:V\to\mathbb R$ be two angular charts on $S^1$, and let $h=\tilde\theta\circ\theta^{-1}$ be the coordinate change on the overlap. A covector field locally has two coefficient expressions, $\omega_\theta=f(\theta)\,d\theta$ in the first chart and $\omega_{\tilde\theta}=\tilde f(\tilde\theta)\,d\tilde\theta$ in the second. These represent the same global form exactly when $h^*\omega_{\tilde\theta}=\omega_\theta$. Pullback gives $h^*\omega_{\tilde\theta}=\tilde f(h(\theta))\,d(h(\theta))=\tilde f(h(\theta))h'(\theta)\,d\theta$, so the equality of one-forms forces \begin{align*} f(\theta)=\tilde f(h(\theta))\,h'(\theta), \qquad \tilde f(\tilde\theta)=\frac{f(\theta)}{h'(\theta)}. \end{align*} A covector field is not a "function plus a label $d\theta$" — the label transforms too. When the coordinate scales by $h'$, the basis covector scales by $h'$, so the coefficient must scale by $1/h'$ to compensate. This is the one-dimensional case of the universal pullback compatibility rule for differential forms, which generalises directly to $k$-forms on $n$-manifolds with the Jacobian determinant in place of $h'$. [/example] ## Differential Forms Defined by Charts A formula such as $f(x,y)\,dx\wedge dy$ is meaningful on a coordinate patch, but it should not depend on the chosen coordinates. The next definition answers the question of when a family of local coordinate formulae represents one global $k$-form. [definition: Differential Form on a Manifold] Let $M$ be a smooth $n$-manifold with smooth atlas $\mathcal A$. A smooth $k$-form on $M$ is an assignment which, for each chart $(U,\varphi)\in\mathcal A$, gives a smooth $k$-form $\omega_\varphi\in\Omega^k(\varphi(U))$ such that for any two charts $(U,\varphi)$ and $(V,\psi)$, the compatibility condition \begin{align*} (\psi\circ\varphi^{-1})^*(\omega_\psi|_{\psi(U\cap V)})=\omega_\varphi|_{\varphi(U\cap V)} \end{align*} holds on $\varphi(U\cap V)$. [/definition] The compatibility condition says that if a form is written in $\psi$-coordinates and then pulled back through the coordinate change to $\varphi$-coordinates, the result is the form already written in $\varphi$-coordinates. This is the same rule that appeared for pullbacks on open subsets of Euclidean space, now used as the glue holding the local pieces together. Without it, one could assign $dx$ in one coordinate and $2\,dx$ in another coordinate on the same interval; the two formulae would not describe a single covector field on the overlap. [definition: Space of Smooth Forms] For a smooth manifold $M$, the [vector space](/page/Vector%20Space) of smooth $k$-forms on $M$ is denoted by $\Omega^k(M)$. The full space of smooth differential forms is \begin{align*} \Omega^*(M)=\bigoplus_{k=0}^{n}\Omega^k(M). \end{align*} [/definition] Here $\Omega^0(M)$ is the algebra $C^\infty(M)$ of smooth functions on $M$. For $k>n$, the space $\Omega^k(M)$ is zero because each tangent space has dimension $n$. [example: Forms on the Sphere by Restriction] Let $i:S^2\hookrightarrow\mathbb R^3$ be the inclusion, and consider the $2$-form \begin{align*} \sigma=x_1\,dx_2\wedge dx_3+x_2\,dx_3\wedge dx_1+x_3\,dx_1\wedge dx_2 \end{align*} on $\mathbb R^3$. More generally, any $\alpha\in\Omega^k(\mathbb R^{n+1})$ pulls back to a $k$-form $i^*\alpha\in\Omega^k(S^n)$ via the inclusion. [claim] $i^*\sigma$ is the standard oriented area form on $S^2$. [/claim] [proof] Since $i$ is the inclusion, $di_x$ is the identity on tangent vectors, so $(i^*\sigma)_x(v,w)=\sigma_x(v,w)$ for $v,w\in T_xS^2$. Each wedge term evaluates via $(dx_a\wedge dx_b)(v,w)=v_aw_b-v_bw_a$, giving \begin{align*} \sigma_x(v,w) =x_1(v_2w_3-v_3w_2)+x_2(v_3w_1-v_1w_3)+x_3(v_1w_2-v_2w_1) =x\cdot(v\times w). \end{align*} To identify this with the standard area form, note that $x\cdot x=1$ on $S^2$ and differentiating along any tangent vector $v$ yields $2x\cdot v=0$, so $x$ is the outward unit normal at the point $x$. For any positively oriented orthonormal basis $(e_1,e_2)$ of $T_xS^2$ the right-hand rule gives $e_1\times e_2=x$, hence \begin{align*} (i^*\sigma)_x(e_1,e_2)=x\cdot(e_1\times e_2)=x\cdot x=1. \end{align*} A $2$-form that evaluates to $1$ on every positively oriented orthonormal basis is the standard area form. [/proof] The form $\sigma$ on $\mathbb R^3$ looks more elaborate than necessary, but the asymmetric pattern $x_1\,dx_2\wedge dx_3+\dots$ is exactly the *interior product* of the radial vector field with the volume form $dx_1\wedge dx_2\wedge dx_3$. Restricting to $S^2$ via $i^*$ amputates the radial direction and leaves the area form — a clean instance of the general fact that the volume form on a level set is the contraction of the ambient volume form with the unit normal. [/example] This example illustrates an important source of forms on submanifolds: forms from the ambient Euclidean space restrict by pullback. Not every form on a manifold has to be presented in this way, but many geometric examples arise from ambient formulae. [quotetheorem:3572] This theorem justifies switching between two viewpoints. In calculations, forms are local coordinate expressions; conceptually, they are coordinate-independent sections of exterior powers of the cotangent bundle. The Hausdorff condition keeps the local pieces of the space separated enough for tangent and cotangent spaces to assemble into bundles, while second-countability supplies the countability hypotheses used later for partitions of unity. The section viewpoint is what makes restriction, pullback, support, and integration statements independent of a chosen atlas; the chart viewpoint remains the practical way to compute coefficients. ## Wedge Product as a Global Operation The wedge product was defined algebraically on each cotangent space and analytically on open subsets of $\mathbb R^n$. The issue on a manifold is whether multiplying two compatible families of local forms gives another compatible family. [definition: Wedge Product on a Manifold] Let $\alpha\in\Omega^p(M)$ and $\beta\in\Omega^q(M)$. Their wedge product $\alpha\wedge\beta\in\Omega^{p+q}(M)$ is the form whose expression in every chart is \begin{align*} (\alpha\wedge\beta)_\varphi=\alpha_\varphi\wedge\beta_\varphi. \end{align*} [/definition] The definition is local, but it is not arbitrary: pullback commutes with the wedge product, so the local expressions continue to agree under transition maps. The next result records that no algebraic law is lost when we pass from Euclidean coordinate patches to an abstract manifold. This matters because integration and [Stokes' theorem](/theorems/1530) will multiply forms of different degrees, so the signs and degrees must be globally consistent rather than chartwise conventions. [quotetheorem:3573] The theorem is the first place where the coordinate-free formalism pays off. A global algebraic identity between forms is proved by reducing to the Euclidean coordinate calculation and using compatibility to glue the results. The sign $(-1)^{pq}$ is not optional: if $\alpha$ has odd degree, then $\alpha\wedge\alpha=-\alpha\wedge\alpha$, so over $\mathbb R$ this forces $\alpha\wedge\alpha=0$. On a space with non-smooth transition functions, the coefficient transformations needed for this algebra may fail to be smooth; on badly non-Hausdorff spaces, the local pieces need not assemble into well-behaved global sections. [example: Wedge Products on a Surface] On a chart $(U,\varphi)$ of a smooth surface with coordinates $x_1,x_2$, let $\alpha=a_1\,dx_1+a_2\,dx_2$ and $\beta=b_1\,dx_1+b_2\,dx_2$. Bilinearity expands $\alpha\wedge\beta$ into four terms; alternation kills the two diagonals and combines the off-diagonals through $dx_2\wedge dx_1=-dx_1\wedge dx_2$: \begin{align*} \alpha\wedge\beta =(a_1b_2-a_2b_1)\,dx_1\wedge dx_2 =\det\begin{pmatrix}a_1 & a_2\\ b_1 & b_2\end{pmatrix}\,dx_1\wedge dx_2. \end{align*} The coefficient is no accident. The basis $2$-form $dx_1\wedge dx_2$ transforms under a chart change $(x_1,x_2)\to(y_1,y_2)$ by the same determinant rule: expanding $dx_i=\sum_j(\partial x_i/\partial y_j)\,dy_j$ and applying the same bilinearity-plus-alternation gives \begin{align*} dx_1\wedge dx_2=\det\!\left(\frac{\partial(x_1,x_2)}{\partial(y_1,y_2)}\right)\,dy_1\wedge dy_2. \end{align*} So the local coefficient of a $2$-form picks up the Jacobian on a chart change — precisely the compensating factor that makes $\alpha\wedge\beta$ a chart-independent geometric object. This is the local mechanism behind orientability and the change-of-variables formula on surfaces. [/example] ## Exterior Derivative on Manifolds The [exterior derivative](/theorems/1525) should differentiate a form without requiring a preferred coordinate system. Since the Euclidean [exterior derivative](/theorems/1525) commutes with pullback by smooth maps, it can be applied chart by chart and then glued. [definition: Exterior Derivative on a Manifold] Let $\omega\in\Omega^k(M)$. The [exterior derivative](/theorems/1525) $d\omega\in\Omega^{k+1}(M)$ is the form whose expression in every chart is \begin{align*} (d\omega)_\varphi=d(\omega_\varphi). \end{align*} [/definition] This gives the same operation as ordinary differentiation on functions: if $f\in\Omega^0(M)=C^\infty(M)$, then in local coordinates \begin{align*} df=\sum_{i=1}^n \frac{\partial f}{\partial x_i}\,dx_i. \end{align*} For higher-degree forms, $d$ differentiates the coefficient functions and wedges the corresponding coordinate covectors in front. [quotetheorem:1525] This is the mechanism behind much of the theory of forms on manifolds: prove a natural identity in Euclidean coordinates, check that it commutes with pullback, and then regard it as a statement on every smooth manifold. The identity $d^2=0$ is the algebraic reason closed forms and exact forms form a cochain complex, which is the starting point of de Rham cohomology. Locality is just as important: if the chart representatives did not satisfy the overlap compatibility condition, differentiating them would produce unrelated local $(k+1)$-forms rather than a single global object. [example: Differentiating a One-Form in Local Coordinates] Let $(U,x,y)$ be a coordinate patch on a smooth surface and $\omega=P\,dx+Q\,dy\in\Omega^1(U)$. The exterior derivative $d$ is $\mathbb R$-linear and obeys the signed product rule $d(f\,\eta)=df\wedge\eta+f\,d\eta$ with $d(dx^i)=0$ for coordinate one-forms, so \begin{align*} d\omega=dP\wedge dx+dQ\wedge dy. \end{align*} Expanding $dP=\partial_xP\,dx+\partial_yP\,dy$ (and similarly $dQ$), the wedge against the single coordinate kills two of the four resulting terms via $dx\wedge dx=dy\wedge dy=0$ and reorders the survivors by $dy\wedge dx=-dx\wedge dy$: \begin{align*} d\omega =\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy. \end{align*} This is the scalar curl of vector calculus, now living naturally on a $2$-manifold. A different chart on the overlap would produce the corresponding expression in those coordinates, related to this one by the Jacobian rule of the previous example — confirming that $d\omega$ is a globally defined $2$-form, not a coordinate artifact. [/example] This computation also shows why $d$ is more rigid than ordinary partial differentiation of coefficients. The alternating combination is exactly what survives a change of coordinates. [remark: Locality of the Exterior Derivative] If two forms agree on an open subset $U\subset M$, then their exterior derivatives agree on $U$. This locality is inherited from the coordinate definition and will be used when partitions of unity and integration are introduced. [/remark] ## Pullback of Forms Between Manifolds How should a form on one manifold be transported to another when the two manifolds need not sit inside a common Euclidean space? A smooth map $F:M\to N$ compares tangent vectors by its differential, so a form on $N$ can be pulled back to a form on $M$ by applying $dF_p$ to tangent vectors before evaluating the original form. [definition: Pullback on Manifolds] Let $F:M\to N$ be a smooth map between smooth manifolds, and let $\omega\in\Omega^k(N)$. The pullback $F^*\omega\in\Omega^k(M)$ is defined pointwise by \begin{align*} (F^*\omega)_p(v_1,\dots,v_k)=\omega_{F(p)}(dF_p(v_1),\dots,dF_p(v_k)) \end{align*} for $p\in M$ and $v_1,\dots,v_k\in T_pM$. [/definition] In local coordinates, this is the same substitution rule already used for forms on open subsets of Euclidean space. The coordinate-free pointwise formula explains why the construction is independent of charts. [quotetheorem:3574] Pullback is contravariant: forms move from the target of a map back to its source. Smoothness of $F$ is essential, because the formula uses $dF_p$ and the identity $d(F^*\alpha)=F^*(d\alpha)$ differentiates the pulled-back coefficients. If $F$ is only continuous, there is no derivative with which to pull back positive-degree forms; if $F$ has only finite differentiability, pulling back and then applying $d$ consumes derivatives. This theorem is the naturality statement that later makes de Rham cohomology functorial. ## Manifolds with Boundary and Collars Integration on manifolds requires a precise meaning of the boundary. The local model changes from all of $\mathbb R^n$ to a half-space, because near a boundary point only one side of the manifold is present. [definition: Half-Space] The closed upper half-space in $\mathbb R^n$ is \begin{align*} \mathbb H^n=\{(x_1,\dots,x_n)\in\mathbb R^n:x_n\ge 0\}. \end{align*} [/definition] A boundary chart maps an open neighbourhood in the manifold to an open subset of $\mathbb H^n$, with transition maps smooth in the sense that they extend smoothly to open subsets of $\mathbb R^n$. [definition: Smooth Manifold with Boundary] A smooth $n$-manifold with boundary is a second-countable Hausdorff topological space $M$ equipped with a maximal atlas of charts $(U,\varphi)$ where $\varphi:U\to\varphi(U)\subseteq\mathbb H^n$ is a homeomorphism onto a relatively open subset of $\mathbb H^n$, and the transition maps are smooth up to the boundary. [/definition] The boundary consists of the points that are sent to the hyperplane $x_n=0$ in some boundary chart. This does not depend on the chosen chart, because transition maps between half-space charts preserve the distinction between interior points and boundary points. [definition: Boundary of a Smooth Manifold with Boundary] Let $M$ be a smooth $n$-manifold with boundary. Its boundary is \begin{align*} \partial M=\{p\in M: \varphi_n(p)=0 \text{ in some boundary chart }(U,\varphi)\text{ around }p\}. \end{align*} [/definition] The set $M\setminus\partial M$ is the interior of $M$, and $\partial M$ itself carries a natural smooth structure of dimension $n-1$. Boundary orientation will be addressed when integration is introduced. [definition: Compact Manifold with Boundary] A compact smooth manifold with boundary is a smooth manifold with boundary whose underlying topological space is compact. [/definition] Compactness is the hypothesis under which global integration behaves especially well: finite subcovers, partitions of unity, and finite sums over coordinate charts become available. Near the boundary, however, compactness alone does not tell us that the manifold separates into a tangential boundary direction and a normal inward direction. The collar theorem supplies exactly that product model, so boundary integrals can be compared cleanly with integrals over the interior. [quotetheorem:3575] This theorem is quoted without proof in this course. It says that near its boundary, a compact smooth manifold with boundary looks like a product of the boundary with a half-open interval; this product structure is essential for stating and proving [Stokes' theorem](/theorems/1530) cleanly. [illustration:forms-collar-neighbourhood] [example: The Closed Ball] Let $M=\overline B(0,1)\subset\mathbb R^n$ with boundary $\partial M=S^{n-1}$. For $0<\varepsilon<1$, define \begin{align*} \Phi:S^{n-1}\times[0,\varepsilon)\to M, \qquad \Phi(p,t)=(1-t)p. \end{align*} $\Phi$ is a *collar* of $\partial M$: a diffeomorphism from $S^{n-1}\times[0,\varepsilon)$ onto the open shell $U=\{x\in M:1-\varepsilon<|x|\le 1\}$ that fixes $\partial M$ at $t=0$. The image identification is immediate: $|\Phi(p,t)|=(1-t)|p|=1-t\in(1-\varepsilon,1]$, so $\Phi(p,t)\in U$, and the inverse on $U$ is $x\mapsto(x/|x|,1-|x|)$ — well defined since $|x|>1-\varepsilon>0$. Both $\Phi$ and $\Phi^{-1}$ are smooth in their arguments, and $\Phi(p,0)=p$. Injectivity of $\Phi$ falls out of taking norms: $\Phi(p,t)=\Phi(q,s)$ forces $1-t=1-s$, hence $t=s$, and then $p=q$ on dividing by $1-t>0$. The parameter $t$ is exactly inward radial distance from the boundary: $t=0$ is $\partial M$, and increasing $t$ pushes inward along the radius through $p$. Collars exist for every smooth manifold with boundary — the closed ball is the cleanest possible example — and they are the technical device that makes Stokes' theorem work near the boundary, because they reduce integration near $\partial M$ to integration on a product. [/example] ## Orientability and the Möbius Band Top-degree forms are local volume elements. The question for integration is whether these local volume elements can be chosen consistently across the whole manifold. [definition: Orientation by an Atlas] An orientation on a smooth $n$-manifold $M$ is a smooth atlas such that every transition map has positive Jacobian determinant, considered up to refinement by atlases with the same positivity property. [/definition] This definition matches the intuition that a change of coordinates should preserve the chosen sign of a local volume form. If a global nowhere-vanishing $n$-form exists, it determines such an orientation by declaring its local coefficient relative to oriented coordinates to be positive. [quotetheorem:3576] The course will use orientability to define integration of top-degree forms. Non-orientable manifolds show why the hypothesis is not cosmetic: there may be no globally consistent choice of sign for volume. Connectedness is not required for the theorem: each connected component may be oriented independently, and a nowhere-vanishing top-degree form records all those choices at once. Passing to an orientable double cover can restore a consistent local sign upstairs, but the sign may reverse under the covering symmetry; the Möbius band below is the basic example where this obstruction appears. [illustration:forms-mobius-orientation-reversal] [example: Möbius Band] Let $M$ be the Möbius band, the quotient of $\mathbb R\times[-1,1]$ by the action of $\rho(s,t)=(s+1,-t)$, with quotient map $\pi$. Boundary charts near $t=\pm 1$ use the boundary coordinate $r=1-|t|$, so $M$ is a smooth surface with boundary modelled on $\mathbb H^2$ where appropriate. The gluing $\rho$ has Jacobian $\det D\rho=-1$, so it reverses the sign of the local area form: \begin{align*} \rho^*(ds\wedge dt)=d(s+1)\wedge d(-t)=-\,ds\wedge dt. \end{align*} [claim] $M$ admits no smooth nowhere-vanishing global $2$-form. Equivalently (by the orientability-via-top-form theorem), $M$ is not orientable. [/claim] [proof] Suppose $\omega\in\Omega^2(M)$ is nowhere zero. Pulling back along $\pi$ gives $\pi^*\omega=f(s,t)\,ds\wedge dt$ for a smooth $f$, nowhere zero because $\pi$ is a local diffeomorphism. The relation $\pi\circ\rho=\pi$ implies $\rho^*\pi^*\omega=\pi^*\omega$, so \begin{align*} -f(s+1,-t)\,ds\wedge dt=f(s,t)\,ds\wedge dt, \end{align*} giving the cocycle $f(s,t)=-f(s+1,-t)$. Restricting to $t=0$ and writing $g(s)=f(s,0)$, the relation becomes $g(s)=-g(s+1)$, so $g(0)$ and $g(1)$ have opposite signs. By the intermediate value theorem, $g$ must vanish somewhere in $[0,1]$ — contradicting nowhere-vanishing. [/proof] The Möbius band is the archetypal non-orientable surface: a single trip around its central circle returns the local area element with the opposite sign. Algebraically, the orientation obstruction lives in the sign of $\det D\rho$ of the gluing map. Geometrically, it is the topological fact that the band has only one side. This obstruction is exactly what $H^2(M;\mathbb Z/2)$ detects, and it is the prototype for the orientation double cover used later when comparing de Rham cohomology with $\mathbb Z$-coefficient singular cohomology. [/example] The Möbius band is the standard warning that local data are not enough. Differential forms can always be defined locally, but global statements about integration and orientation require compatibility around loops as well as across individual overlaps. With differential forms defined on manifolds, the next task is to integrate them over domains. This requires equipping manifolds with an orientation, a consistent choice of sign for the top-degree form that lets us assign [real numbers](/page/Real%20Numbers) to integrals. # 6. Orientation and Integration of Forms This chapter turns differential forms from algebraic objects into quantities that can be integrated. The central difficulty is that integration is signed: a coordinate system tells us whether $dx_1 \wedge \cdots \wedge dx_n$ is positive or negative, and different coordinate systems need to make compatible choices. Once this compatibility is encoded by an orientation, compactly supported top-degree forms have a well-defined integral over a manifold, and the familiar line, surface, and volume integrals become instances of the same construction. The chapter also prepares the boundary convention needed for Stokes theorem. The orientation on $\partial M$ is not an extra arbitrary choice once $M$ is oriented; it is fixed by the outward normal first convention. This is the sign convention that makes the general Stokes theorem match Green theorem, the [divergence theorem](/theorems/2754), and the classical orientation rules for curves and surfaces. ## Orientability and Top-Degree Forms What data lets a manifold decide whether a local coordinate volume element is positive? On an $n$-manifold, the space of top-degree covectors at each point is one-dimensional, so choosing a positive side of that line at every point is the geometric content of orientation. [definition: Orientation of a Tangent Space] Let $V$ be a real [vector space](/page/Vector%20Space) of dimension $n$. An orientation of $V$ is a choice of one of the two equivalence classes of ordered bases of $V$, where two ordered bases $(v_1,\dots,v_n)$ and $(w_1,\dots,w_n)$ are equivalent when the change-of-basis matrix from the first basis to the second has positive determinant. [/definition] Thus an ordered basis is either positively oriented or negatively oriented relative to the chosen class. For a one-dimensional [vector space](/page/Vector%20Space), this amounts to choosing which nonzero vectors point in the positive direction. [definition: Orientation of a Manifold] Let $M$ be a smooth $n$-manifold. An orientation of $M$ is a choice of orientation of each tangent space $T_pM$, for $p \in M$, such that around every point there is a coordinate chart $(U,\varphi)$ whose coordinate basis \begin{align*} \left(\frac{\partial}{\partial x_1}\bigg|_p,\dots,\frac{\partial}{\partial x_n}\bigg|_p\right) \end{align*} is positively oriented for every $p \in U$. [/definition] The local compatibility condition prevents the orientation from changing sign discontinuously from point to point. A manifold admitting such a choice is called orientable. [definition: Volume Form] Let $M$ be a smooth $n$-manifold. A volume form on $M$ is a smooth differential form $\omega \in \Omega^n(M)$ such that $\omega_p \ne 0$ in $\Lambda^n(T_p^*M)$ for every $p \in M$. [/definition] A volume form chooses the positive bases by evaluation: an ordered basis $(v_1,\dots,v_n)$ of $T_pM$ is positive when $\omega_p(v_1,\dots,v_n)>0$. The next result is the first major bridge between orientations and forms. [quotetheorem:3576] This theorem is useful because a nowhere-vanishing top-degree form is often easier to construct than an orientation atlas. On a submanifold of Euclidean space, a normal vector field can also encode an orientation, as surface examples below show. [example: Standard Orientation of Euclidean Space] On $\mathbb R^n$ with global coordinates $(x_1,\dots,x_n)$, consider the top-degree form $\omega=dx_1\wedge\cdots\wedge dx_n$. At any point $p\in\mathbb R^n$, the evaluation rule $(dx_1\wedge\cdots\wedge dx_n)(v_1,\dots,v_n)=\det(dx_i(v_j))$ specialises on the coordinate basis $e_i=\partial/\partial x_i|_p$ to $\det(\delta_{ij})=1$, so $\omega$ is nowhere vanishing. For an arbitrary ordered basis $(v_1,\dots,v_n)$ of $T_p\mathbb R^n$ with change-of-basis matrix $A=(a_{ij})$, $v_j=\sum_i a_{ij}e_i$, linearity gives $dx_i(v_j)=a_{ij}$, hence \begin{align*} \omega_p(v_1,\dots,v_n)=\det A. \end{align*} The basis is positive (i.e., $\omega_p>0$) precisely when $\det A>0$ — the classical orientation criterion for $\mathbb R^n$. The point is that an "orientation" — a notion that looks like extra data — is already encoded in a single nowhere-vanishing top-degree form. The standard orientation of $\mathbb R^n$ is just the orientation determined by $dx_1\wedge\cdots\wedge dx_n$, and on a general $n$-manifold any choice of nowhere-vanishing top-form plays the same role. [/example] The Euclidean example is the reference model: orientation is detected by a nowhere-zero top-degree form whose sign is globally consistent. The next example shows exactly what fails when that local sign cannot be transported consistently around the manifold. [example: Non-Orientability of the Mobius Strip] Let $M$ be the open Möbius strip, the quotient of $\widetilde M=\mathbb R\times(-1,1)$ by the deck transformation $T(s,t)=(s+1,-t)$, with quotient $\pi:\widetilde M\to M$. (This refines the Möbius band of the previous chapter; the boundary is removed here so we are on an honest manifold without boundary.) [claim] $M$ admits no volume form, hence is not orientable. [/claim] [proof] Suppose $\Omega\in\Omega^2(M)$ is nowhere zero. Pullback gives $\pi^*\Omega=f(s,t)\,ds\wedge dt$ with $f$ smooth and nowhere zero. The identity $\pi\circ T=\pi$ forces $T^*\pi^*\Omega=\pi^*\Omega$, and since $T^*(ds\wedge dt)=ds\wedge(-dt)=-ds\wedge dt$, \begin{align*} -f(s+1,-t)\,ds\wedge dt=f(s,t)\,ds\wedge dt, \end{align*} i.e., $f(s,t)=-f(s+1,-t)$. Restricting to $t=0$ yields $h(s)=f(s,0)$ satisfying $h(s+1)=-h(s)$, so $h(0)$ and $h(1)$ have opposite signs. By the intermediate value theorem, $h$ vanishes somewhere on $[0,1]$, contradicting nowhere-vanishing. [/proof] The obstruction is visible in $\det DT=-1$: the gluing reverses the transverse coordinate, so one circuit around the central circle returns the local area element with the opposite sign. Non-orientability is precisely the impossibility of patching local area forms across a cover when the transition cocycle has a sign-reversing loop, and the Möbius strip is the simplest manifold where this happens. This is the prototype for the orientation double cover and the orientability obstruction $w_1\in H^1(M;\mathbb Z/2)$. [/example] ## Oriented Atlases and Boundary Orientation How can an orientation be recorded in coordinates, and what sign should a boundary inherit? Coordinate changes compare local choices by their Jacobian determinants, while the boundary convention is fixed by placing the outward normal before the boundary basis. [definition: Oriented Atlas] Let $M$ be a smooth $n$-manifold. An oriented atlas on $M$ is an atlas $\{(U_i,\varphi_i)\}_{i\in I}$ such that for every overlap $U_i\cap U_j$, the transition map \begin{align*} \varphi_j\circ \varphi_i^{-1}:\varphi_i(U_i\cap U_j)\to \varphi_j(U_i\cap U_j) \end{align*} has positive Jacobian determinant at every point of its domain. [/definition] An oriented atlas is a coordinate-level expression of the same orientation from the previous section. Its advantage is computational: it tells us which coordinate formula for an $n$-form should be integrated with a positive sign. [quotetheorem:3577] The boundary of an oriented manifold must be oriented in a way compatible with outward-pointing directions. This convention is the source of many signs in Stokes theorem, so it is fixed now rather than postponed. [definition: Boundary Orientation] Let $M$ be an oriented smooth $n$-manifold with boundary. The induced orientation on $\partial M$ is the orientation for which an ordered basis $(v_1,\dots,v_{n-1})$ of $T_p(\partial M)$ is positive exactly when $(\nu,v_1,\dots,v_{n-1})$ is a positive ordered basis of $T_pM$, where $\nu\in T_pM$ is any outward-pointing vector transverse to $T_p(\partial M)$. [/definition] The outward vector is placed first. For a region in the plane with the standard orientation, this produces counterclockwise orientation on the outer boundary and clockwise orientation on holes. [example: Boundary of the Unit Interval] Give $[0,1]$ the standard orientation determined by $dt$: at every interior point, $(\partial_t)$ is positive because $dt(\partial_t)=1>0$. The boundary $\partial[0,1]=\{0,1\}$ is a $0$-manifold, where each tangent space is trivial and the only ordered basis is the empty one $()$. The induced boundary convention: $()$ is positive at a boundary point $p$ iff $(\nu)$ is positive in $T_p[0,1]$, where $\nu$ is outward-pointing. At $p=1$ the outward direction is $\nu_1=\partial_t|_1$, which is positive ($dt(\nu_1)=1>0$), so the endpoint $1$ inherits sign $+1$. At $p=0$ the outward direction is $\nu_0=-\partial_t|_0$, which is negative ($dt(\nu_0)=-1<0$), so $0$ inherits sign $-1$. Symbolically, $\partial[0,1]=\{1\}-\{0\}$ as an oriented $0$-manifold — the convention that makes Stokes' theorem for a $1$-form $f\,dt$ read $\int_0^1 df=f(1)-f(0)$. The minus sign at the left endpoint is not arbitrary; it is forced by the outward-normal-first orientation convention combined with the standard orientation of $[0,1]$. [/example] The interval is the one-dimensional version of the same outward-normal-first rule. In dimension two, the rule becomes the familiar direction of travel around a boundary curve. [example: Boundary Orientation of the Unit Square] Give $Q=[0,1]^2$ the standard orientation determined by $dx\wedge dy$. The induced boundary convention asks, at each edge, for the tangent direction $v$ such that $(\nu,v)$ is positive in $T_pQ$, where $\nu$ is outward-pointing. Using $(dx\wedge dy)(u,v)=dx(u)dy(v)-dx(v)dy(u)$: - **Bottom edge** ($y=0$, $\nu=-\partial_y$): the test $v=\partial_x$ gives $(dx\wedge dy)(-\partial_y,\partial_x)=0\cdot 0-1\cdot(-1)=1>0$, so the positive direction is $+\partial_x$ (rightward). - **Right edge** ($x=1$, $\nu=\partial_x$): the test $v=\partial_y$ gives $(dx\wedge dy)(\partial_x,\partial_y)=1$, so the positive direction is $+\partial_y$ (upward). - **Top edge** ($y=1$, $\nu=\partial_y$): the test $v=-\partial_x$ gives $(dx\wedge dy)(\partial_y,-\partial_x)=0\cdot 0-(-1)\cdot 1=1$, so the positive direction is $-\partial_x$ (leftward). - **Left edge** ($x=0$, $\nu=-\partial_x$): the test $v=-\partial_y$ gives $(dx\wedge dy)(-\partial_x,-\partial_y)=(-1)(-1)=1$, so the positive direction is $-\partial_y$ (downward). Tracing these directions around the square — right, up, left, down — recovers the familiar counterclockwise orientation of $\partial Q$. The outward-normal-first convention is precisely what makes Stokes' theorem on the square read $\int_Q d\omega=\int_{\partial Q}\omega$ with the conventional ccw orientation on $\partial Q$, giving Green's theorem its usual sign. [/example] ## Integrating Compactly Supported Top-Degree Forms How can we integrate an $n$-form on a manifold that may not have one global coordinate system? The construction is local: break the form using a [partition of unity](/page/Partition%20of%20Unity), integrate each piece in a positive coordinate chart, and prove that the sum is independent of every auxiliary choice. [definition: Local Integral of a Top-Degree Form] Let $(U,\varphi)$ be a positive coordinate chart on an oriented smooth $n$-manifold $M$, with coordinates $(x_1,\dots,x_n)$. If $\omega\in\Omega_c^n(U)$ has coordinate expression \begin{align*} \omega = f\,dx_1\wedge\cdots\wedge dx_n, \end{align*} then its local integral over $U$ is \begin{align*} \int_U \omega := \int_{\varphi(U)} f\circ\varphi^{-1}\,d\mathcal L^n. \end{align*} [/definition] This definition uses a positive chart. If a negative chart were used instead, the sign would change, since the determinant in the change-of-variables formula would have the opposite sign. [quotetheorem:3578] Now the global integral can be assembled. Compact support ensures that only finitely many partition-of-unity terms contribute after choosing a locally finite cover. [definition: Integral of a Compactly Supported Top-Degree Form] Let $M$ be an oriented smooth $n$-manifold and let $\omega\in\Omega_c^n(M)$. Choose a locally finite cover by positive coordinate charts $(U_i,\varphi_i)$ and a smooth [partition of unity](/page/Partition%20of%20Unity) $(\rho_i)_{i\in I}$ subordinate to this cover. The integral of $\omega$ over $M$ is \begin{align*} \int_M \omega := \sum_{i\in I}\int_{U_i} \rho_i\omega. \end{align*} [/definition] The sum is finite because $\operatorname{supp}\omega$ is compact and the chart cover is locally finite. The theorem below is the main well-definedness result for integration on oriented manifolds. [quotetheorem:3579] This result is the analytic foundation for the rest of the course. From now on, $\int_M\omega$ is meaningful whenever $M$ is oriented and $\omega$ is compactly supported of degree $\dim M$. [quotetheorem:3580] This theorem shows that integration of forms depends on the orientation, not just on the underlying smooth manifold. This differs from measure-theoretic integration, where changing an orientation has no effect on the integral of a nonnegative density. [example: Integrating the Standard Area Form on the Unit Square] On $Q=[0,1]^2$ with the standard orientation determined by $dx\wedge dy$, the global chart $\varphi(x,y)=(x,y)$ is positive (the coordinate basis evaluates to $1$ on $dx\wedge dy$), so the manifold integral reduces to the ordinary Lebesgue integral of the coefficient: \begin{align*} \int_Q dx\wedge dy=\int_{[0,1]^2}1\,d\mathcal L^2=1. \end{align*} Reversing the orientation flips the sign. Substituting $u=-x$, $v=y$ gives a chart positive for the opposite orientation in which $dx\wedge dy=-du\wedge dv$, and the image is $[-1,0]\times[0,1]$, so \begin{align*} \int_{-Q}dx\wedge dy=\int_{[-1,0]\times[0,1]}(-1)\,d\mathcal L^2=-1. \end{align*} Two facts coexist here: the geometric area of the square is $1$ regardless of orientation, while the *signed* integral of $dx\wedge dy$ depends on orientation. Lebesgue measure measures size; differential-form integration measures size *with sign*, and the sign tracks orientation. This is what allows $\int_M d\omega=\int_{\partial M}\omega$ to be a true identity rather than an equality up to sign. [/example] The square illustrates orientation on a compact coordinate domain. The next example records the noncompact case where compact support, rather than compactness of the manifold, makes the integral finite and well-defined. [example: A Compactly Supported Form on the Plane] Let $\eta\in C_c^\infty(\mathbb R^2)$ and $\omega=\eta(x,y)\,dx\wedge dy\in\Omega_c^2(\mathbb R^2)$. The single global chart $(x,y)$ is positive for the standard orientation, and a partition of unity subordinate to the cover $\{\mathbb R^2\}$ consists of the constant function $1$. The local integral formula in this chart gives \begin{align*} \int_{\mathbb R^2}\omega=\int_{\mathbb R^2}\eta(x,y)\,d\mathcal L^2(x,y), \end{align*} the ordinary Lebesgue double integral of $\eta$. This is the canonical sanity check that differential-form integration on $\mathbb R^n$ specialises correctly: a compactly supported top-degree form is just a function written in a coordinate disguise. The manifold integral is well-defined because compact support makes both the form-side and the Lebesgue-side integrals finite. On a general $n$-manifold the same local formula assembles via a partition of unity, and Stokes' theorem will eventually let us extend integration to forms that are not compactly supported but vanish suitably at infinity. [/example] ## Line and Surface Integrals as Form Integrals Why do line integrals, flux integrals, and surface integrals have different formulas in vector calculus? They look different because they are written using coordinates and Euclidean vector identifications, but differential forms express them by the same rule: integrate a top-degree form over an oriented manifold of the matching dimension. For a curve, the top degree is $1$. If $\gamma:[a,b]\to M$ is a smooth parametrised curve compatible with the chosen orientation, integrating a $1$-form $\alpha$ over the oriented curve means integrating the pullback $\gamma^*\alpha$ over $[a,b]$. [definition: Integral of a One-Form Along a Parametrised Curve] Let $M$ be a smooth manifold, let $\alpha\in\Omega^1(M)$, and let $\gamma:[a,b]\to M$ be a smooth curve. The line integral of $\alpha$ along $\gamma$ is \begin{align*} \int_\gamma \alpha := \int_{[a,b]}\gamma^*\alpha. \end{align*} [/definition] If $\alpha=P\,dx+Q\,dy+R\,dz$ on an open subset of $\mathbb R^3$ and $\gamma(t)=(x(t),y(t),z(t))$, then \begin{align*} \gamma^*\alpha=(P(\gamma(t))x^{\prime}(t)+Q(\gamma(t))y^{\prime}(t)+R(\gamma(t))z^{\prime}(t))\,dt. \end{align*} Thus the familiar dot-product line integral is the pullback formula for $1$-forms. [example: Work Integral as a One-Form Integral] Let $\alpha=-y\,dx+x\,dy$ and let $\gamma:[0,2\pi]\to\mathbb R^2$, $\gamma(t)=(\cos t,\sin t)$, parametrise the unit circle positively. The line integral of a $1$-form along a curve is defined as $\int_\gamma\alpha=\int_{[0,2\pi]}\gamma^*\alpha$. Pulling back via $\gamma^*dx=-\sin t\,dt$, $\gamma^*dy=\cos t\,dt$, \begin{align*} \gamma^*\alpha=-\sin t(-\sin t\,dt)+\cos t(\cos t\,dt)=(\sin^2t+\cos^2t)\,dt=dt, \end{align*} so $\int_\gamma\alpha=\int_0^{2\pi}dt=2\pi$. The same form $\alpha=-y\,dx+x\,dy$ is twice the angle form $d\theta$ on the punctured plane (without the $1/(x^2+y^2)$ normalisation), and its integral $2\pi$ around the unit circle is the same period that obstructs $d\theta$ from being exact in the de Rham chapter. This circle integral is the canonical generator of $H^1_{\rm dR}(\mathbb R^2\setminus\{0\})$, and it is the value Stokes' theorem produces when one tries — and fails — to apply it across the missing origin. [/example] For an oriented surface in $\mathbb R^3$, the top degree is $2$. A vector field $F=(F_1,F_2,F_3)$ can be encoded as the flux $2$-form \begin{align*} \omega_F = F_1\,dy\wedge dz+F_2\,dz\wedge dx+F_3\,dx\wedge dy. \end{align*} Integrating $\omega_F$ over an oriented surface gives the flux of $F$ through that surface with the chosen orientation. [definition: Flux Form] Let $U\subset\mathbb R^3$ be open and let $F:U\to\mathbb R^3$ be a smooth vector field, written $F=(F_1,F_2,F_3)$. The flux form associated to $F$ is the $2$-form \begin{align*} \omega_F := F_1\,dy\wedge dz+F_2\,dz\wedge dx+F_3\,dx\wedge dy. \end{align*} [/definition] For a parametrised oriented surface $r:D\subset\mathbb R^2\to\mathbb R^3$, the pullback $r^*\omega_F$ is a multiple of $du\wedge dv$. The coefficient is the classical scalar triple product $F(r(u,v))\cdot(r_u\times r_v)$ when $(u,v)$ is a positive parameter orientation. [quotetheorem:3581] The formula identifies the abstract integral of a $2$-form with the vector-calculus flux integral whenever the parametrisation and orientation agree. The following example uses the simplest closed oriented surface to check the sign convention. [example: Outward Flux Through the Unit Sphere] On the unit sphere $S^2$ with outward orientation, let $F(x,y,z)=(x,y,z)$ — the radial vector field — and let \begin{align*} \omega_F=x\,dy\wedge dz+y\,dz\wedge dx+z\,dx\wedge dy \end{align*} be its flux $2$-form. Parametrise by $r(\theta,\phi)=(\sin\theta\cos\phi,\sin\theta\sin\phi,\cos\theta)$, $(\theta,\phi)\in[0,\pi]\times[0,2\pi]$. [claim] $\int_{S^2}\omega_F=4\pi.$ [/claim] [proof] **Orientation compatibility.** The cross product $r_\theta\times r_\phi$ computes to $\sin\theta\,r(\theta,\phi)$, which on the interior $0<\theta<\pi$ points outward radially. So $(\theta,\phi)$ is a positively oriented chart away from the parameter-rectangle boundary, which has measure zero. **Pullback.** Using $r^*dx=\cos\theta\cos\phi\,d\theta-\sin\theta\sin\phi\,d\phi$, $r^*dy=\cos\theta\sin\phi\,d\theta+\sin\theta\cos\phi\,d\phi$, $r^*dz=-\sin\theta\,d\theta$, the three coordinate $2$-forms pull back to \begin{align*} r^*(dy\wedge dz) &=\sin^2\theta\cos\phi\,d\theta\wedge d\phi,\\ r^*(dz\wedge dx) &=\sin^2\theta\sin\phi\,d\theta\wedge d\phi,\\ r^*(dx\wedge dy) &=\sin\theta\cos\theta\,d\theta\wedge d\phi. \end{align*} Each is a routine wedge expansion in which the diagonal $d\theta\wedge d\theta$ and $d\phi\wedge d\phi$ terms vanish. **Coefficient simplification.** Substituting the coefficients $x,y,z$ evaluated along $r$, \begin{align*} r^*\omega_F &=(\sin\theta\cos\phi)(\sin^2\theta\cos\phi)+(\sin\theta\sin\phi)(\sin^2\theta\sin\phi)+(\cos\theta)(\sin\theta\cos\theta)\\ &\qquad\text{(times $d\theta\wedge d\phi$)}\\ &=\bigl(\sin^3\theta(\cos^2\phi+\sin^2\phi)+\sin\theta\cos^2\theta\bigr)\,d\theta\wedge d\phi\\ &=\sin\theta(\sin^2\theta+\cos^2\theta)\,d\theta\wedge d\phi =\sin\theta\,d\theta\wedge d\phi. \end{align*} **Integration.** $\int_0^{2\pi}\int_0^\pi\sin\theta\,d\theta\,d\phi=\int_0^{2\pi}2\,d\phi=4\pi$. [/proof] The radial flux form $\omega_F=x\,dy\wedge dz+y\,dz\wedge dx+z\,dx\wedge dy$ restricts to $S^2$ as the area form (the previous chapter's example showed it equals $x\cdot(v\times w)$ on tangent vectors), so this integral computes the surface area of the unit sphere. The two derivations of $4\pi$ — once via $\sin\theta\,d\theta\,d\phi$ here, and once via the conformal factor $4/(1+u^2+v^2)^2$ in the stereographic example — converge on the same number from completely different coordinate vantages, which is the geometric content of "the integral of a $2$-form is chart-independent." [/example] Flux forms also recover unsigned surface area once a unit normal has been chosen to convert area into an oriented $2$-form. This is the bridge between the metric notion of area and the orientation-sensitive integration of forms. [example: Surface Area Form from a Parametrisation] Let $S$ be a smooth oriented surface in $\mathbb R^3$ with chosen unit normal field $n=(n_1,n_2,n_3)$, parametrised by $r:D\to S$ in an orientation-compatible way (so $r_u\times r_v$ points along $n$). Encode the normal as the flux form \begin{align*} \omega_n=n_1\,dy\wedge dz+n_2\,dz\wedge dx+n_3\,dx\wedge dy. \end{align*} [claim] \begin{align*} \int_S\omega_n=\int_D|r_u\times r_v|\,d\mathcal L^2. \end{align*} [/claim] [proof] Pulling back each coordinate $2$-form along $r$ produces a $2\times 2$ minor of the Jacobian as the coefficient of $du\wedge dv$: \begin{align*} r^*(dy\wedge dz)&=(y_uz_v-y_vz_u)\,du\wedge dv,\\ r^*(dz\wedge dx)&=(z_ux_v-z_vx_u)\,du\wedge dv,\\ r^*(dx\wedge dy)&=(x_uy_v-x_vy_u)\,du\wedge dv. \end{align*} These three minors are the components of $r_u\times r_v$, so combining, \begin{align*} r^*\omega_n=\bigl(n(r)\cdot(r_u\times r_v)\bigr)\,du\wedge dv. \end{align*} Orientation compatibility says $n(r)=(r_u\times r_v)/|r_u\times r_v|$, hence $n(r)\cdot(r_u\times r_v)=|r_u\times r_v|$ and \begin{align*} r^*\omega_n=|r_u\times r_v|\,du\wedge dv. \end{align*} Integration over $D$ via the definition $\int_S\omega_n=\int_D r^*\omega_n$ delivers the claim. [/proof] This is the bridge between the form-theoretic surface integral and the classical scalar surface-area integral $\iint|r_u\times r_v|\,dA$ from multivariable calculus. The orientation hypothesis is exactly what makes $n\cdot(r_u\times r_v)=+|r_u\times r_v|$ rather than $-|r_u\times r_v|$; reversing the orientation flips the sign of the flux integral. The same calculation will appear in disguise as the integral side of Stokes' theorem, where divergence theorem and surface integrals fall out as special cases. [/example] The point of these examples is not that vector calculus disappears, but that its sign conventions become part of the orientation of the domain. In the next chapter, exterior differentiation and boundary orientation combine in Stokes theorem, which turns the identity between an integral over a boundary and an integral over the interior into a single formula for all degrees. With orientations and integration in place, we can now state the theorem unifying all of vector calculus. The generalized Stokes theorem relates the integral of an [exterior derivative](/theorems/1525) over an oriented manifold to an integral of the original form over its boundary. # 7. The Generalised Stokes Theorem The previous chapters built the machinery needed to integrate differential forms: exterior products, the [exterior derivative](/theorems/1525), pullbacks, orientations, and integration over oriented manifolds. This chapter brings those constructions together in a single theorem. The generalised Stokes theorem says that integrating an [exterior derivative](/theorems/1525) over a manifold is the same as integrating the original form over the oriented boundary. This is the point at which differential forms repay the setup. The [fundamental theorem of calculus](/theorems/632), the theorem of Green, the classical Stokes theorem, and the [divergence theorem](/theorems/2754) become the same statement in different dimensions, with different choices of form. ## Why Compact Support and Boundary Orientation Are Needed What hypotheses make the two integrals in Stokes formula finite and give them compatible signs? The compact-support condition prevents contributions from escaping to infinity on a non-compact manifold. On a compact manifold it is automatic, but on an open manifold it is the condition that makes [integration by parts](/theorems/2098) have only the geometric boundary term. [definition: Compactly Supported Differential Form] Let $M$ be a smooth manifold. A differential form $\omega \in \Omega^k(M)$ is compactly supported if \begin{align*} \operatorname{supp}\omega := \overline{\{p \in M : \omega_p \ne 0\}} \end{align*} is compact in $M$. The subspace of compactly supported $k$-forms is \begin{align*} \Omega_c^k(M) := \{\omega \in \Omega^k(M) : \operatorname{supp}\omega \text{ is compact in } M\}\subset \Omega^k(M). \end{align*} [/definition] The sign on the boundary integral is not an extra convention attached to the theorem; it is forced by the orientation convention for manifolds with boundary. This convention is chosen so that the half-space calculation matches the [fundamental theorem of calculus](/theorems/632) in the normal coordinate. [definition: Boundary Orientation] Let $M$ be an oriented smooth $n$-manifold with boundary. The boundary orientation on $\partial M$ is the orientation for which a basis $(v_1,\dots,v_{n-1})$ of $T_p\partial M$ is positive precisely when $(\nu,v_1,\dots,v_{n-1})$ is a positive basis of $T_pM$, where $\nu \in T_pM$ is any outward-pointing vector transverse to $T_p\partial M$. [/definition] For the model half-space $H^n=\{x\in \mathbb R^n:x_n\ge 0\}$ with its standard orientation, the outward normal is $-\partial_{x_n}$ along $\partial H^n$. Hence the induced boundary orientation is represented by $(-1)^n dx_1\wedge\cdots\wedge dx_{n-1}$. [illustration:forms-halfspace-boundary-orientation] ## The Generalised Stokes Theorem How can the integral of a derivative over all of $M$ be detected only from values on $\partial M$? [quotetheorem:3555] The theorem is local in nature: away from the boundary, compact support makes all coordinate-direction boundary terms cancel; near the boundary, the outward normal direction is the only source of a remaining term. The theorem should be read as an integration-by-parts formula without coordinates. The [exterior derivative](/theorems/1525) supplies the differentiated quantity, the orientation supplies the sign, and the pullback $\iota^*\omega$ is the operation that restricts the form to the boundary. If the orientation of $M$ is reversed, both sides of the formula change sign because the boundary orientation is induced from the orientation of $M$. If the orientation of $M$ is fixed but $\partial M$ is given the opposite orientation, only the boundary integral changes sign, so the stated equality no longer has the correct sign. The compact-support hypothesis is also part of the theorem rather than a technical decoration. It removes boundary terms at infinity on non-compact manifolds. [example: Failure Without Compact Support On The Line] Take $f(x)=\arctan x$ as a smooth $0$-form on $\mathbb R$, which has no boundary. Its exterior derivative is $df=(1+x^2)^{-1}\,dx$, and \begin{align*} \int_{\mathbb R}df=\int_{-\infty}^\infty\frac{dx}{1+x^2}=\arctan(\infty)-\arctan(-\infty)=\pi, \end{align*} while the empty boundary contributes $\int_{\partial\mathbb R}f=0$. Stokes would say $\pi=0$, which it manifestly does not. The missing $\pi$ is the value $\lim_{x\to\infty}f(x)-\lim_{x\to-\infty}f(x)$ — a *boundary at infinity* that the manifold $\mathbb R$ does not see. Compact support kills this contribution because a compactly supported function on $\mathbb R$ vanishes outside a finite interval, so both limits are zero. This is the entire purpose of the compact-support hypothesis in Stokes' theorem on a non-compact manifold: without it, mass can leak to infinity and the formula breaks. [/example] This is the analytic reason compact support appears in the global statement. ## The Half-Space Calculation Which coordinate computation is hidden inside the proof of the theorem? The local model is $H^n=\{x\in\mathbb R^n:x_n\ge 0\}$ with standard coordinates. For a compactly supported $(n-1)$-form on $H^n$, write \begin{align*} \omega = \sum_{i=1}^n f_i\, dx_1\wedge\cdots\wedge dx_{i-1}\wedge dx_{i+1}\wedge\cdots\wedge dx_n. \end{align*} Then \begin{align*} d\omega = \sum_{i=1}^n (-1)^{i-1}\frac{\partial f_i}{\partial x_i}\, dx_1\wedge\cdots\wedge dx_n. \end{align*} For $i<n$, integration in the $x_i$ direction gives zero because $f_i$ has compact support. The normal term gives \begin{align*} \int_{H^n} (-1)^{n-1}\frac{\partial f_n}{\partial x_n}\,d\mathcal L^n = (-1)^n\int_{\mathbb R^{n-1}} f_n(x_1,\dots,x_{n-1},0)\,d\mathcal L^{n-1}. \end{align*} This is precisely the integral of $\omega$ over $\partial H^n$ with the boundary orientation described above. This computation is the [fundamental theorem of calculus](/theorems/632) applied in the coordinate normal to the boundary. [example: A Compactly Supported Form On The Half-Plane] On the upper half-plane $H^2=\{y\ge 0\}$ with standard orientation $dx\wedge dy$, let $\omega=f(x,y)\,dx$ for $f\in C_c^\infty(H^2)$, supported in $(-R,R)\times[0,R)$. [claim] $\int_{H^2}d\omega=\int_{\partial H^2}\iota^*\omega=\int_{\mathbb R}f(x,0)\,dx$. [/claim] [proof] The exterior derivative is $d\omega=df\wedge dx=(\partial_yf\,dy)\wedge dx=-\partial_yf\,dx\wedge dy$. Fubini together with the fundamental theorem of calculus in $y$ — using $f(x,R)=0$ from compact support — gives \begin{align*} \int_{H^2}d\omega =\int_{-R}^R\int_0^R-\partial_yf\,dy\,dx =\int_{-R}^R\bigl(f(x,0)-f(x,R)\bigr)\,dx =\int_{\mathbb R}f(x,0)\,dx. \end{align*} For the boundary side, the outward normal on $\partial H^2=\{y=0\}$ is $-\partial_y$, so the induced orientation makes $(\partial_x)$ positive: $\det\!\begin{pmatrix}0&1\\-1&0\end{pmatrix}=1>0$. The parametrisation $\tau(t)=(t,0)$ is therefore positive, and $\tau^*\iota^*\omega=f(t,0)\,dt$, giving $\int_{\partial H^2}\iota^*\omega=\int_{\mathbb R}f(x,0)\,dx$. [/proof] This is the local model that Stokes' theorem on a manifold-with-boundary reduces to: a chart compatible with $\partial M\cong\mathbb R^{n-1}\times\{0\}\subset H^n$ converts the global statement into exactly this $y$-direction integration-by-parts in coordinates. The minus sign from $dy\wedge dx=-dx\wedge dy$ is precisely what makes the boundary orientation come out correct — change either convention and the whole formula picks up a sign error. [/example] The example shows why the outward-normal-first convention is the right convention for this course. It makes the boundary contribution agree with the sign produced by integration in the inward coordinate direction. ## Boundaries of Boundaries What remains if the form being integrated is already an [exterior derivative](/theorems/1525)? The algebraic identity $d^2=0$ becomes a geometric statement after applying Stokes. It says that the oriented boundary of an oriented boundary contributes nothing to integration. [quotetheorem:3582] This corollary is often the first hint of cohomology, but it still depends on the same support and integrability hypotheses as Stokes theorem. Without compact support, an exact form on a non-compact boundary can retain a contribution from infinity: for $M=H^2$ and $\eta(x,y)=\arctan x$, the restriction to $\partial H^2\cong\mathbb R$ satisfies \begin{align*} \int_{\partial H^2} d(\iota^*\eta)=\int_{\mathbb R}\frac{1}{1+x^2}\,dx=\pi. \end{align*} The result also does not say that every closed form is exact. It says that exact forms integrate to zero over oriented boundaries under the stated hypotheses; the converse question is local in the Poincare lemma and global in de Rham cohomology. ## Recovering The Classical Integral Theorems What familiar formulas appear when the dimension and the form are chosen in the standard Euclidean ways? Each classical theorem is obtained by choosing $M$ and $\omega$ so that $d\omega$ is the usual integrand in the interior and $\omega$ is the usual line, surface, or endpoint integrand on the boundary. [example: Fundamental Theorem Of Calculus] On $M=[a,b]$ with the standard orientation $dx$, take a $0$-form $f\in C^\infty([a,b])$. Then $df=f'(x)\,dx$, so $\int_Mdf=\int_a^b f'(x)\,dx$, and the boundary computation of the previous chapter gives $\partial[a,b]=\{b\}-\{a\}$, hence $\int_{\partial M}f=f(b)-f(a)$. Applying the generalised Stokes theorem, \begin{align*} \int_a^b f'(x)\,dx=f(b)-f(a), \end{align*} which is exactly the [fundamental theorem of calculus](/theorems/632). Stokes' theorem is the only known generalisation of FTC that scales correctly to higher dimensions, and every classical integral theorem in this chapter — Green, Kelvin–Stokes, divergence — is the same identity in a different dimension and codegree. [/example] This is the one-dimensional case from which the theorem gets its sign convention. The higher-dimensional statements are the same principle applied to forms of higher degree. [example: Planar Green Formula From Differential Forms] Let $D\subset\mathbb R^2$ be a compact oriented region with smooth boundary, oriented by $dx\wedge dy$, and let $\omega=P\,dx+Q\,dy$ for $P,Q\in C^\infty(D)$. [claim] \begin{align*} \int_{\partial D}P\,dx+Q\,dy=\int_D\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)d\mathcal L^2. \end{align*} [/claim] [proof] The coordinate formula for $d$ on a $1$-form (Chapter 3) gives \begin{align*} d\omega=\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy, \end{align*} so $\int_D d\omega=\int_D(\partial_xQ-\partial_yP)\,d\mathcal L^2$ since $dx\wedge dy$ is the positive area form on $D$. The boundary orientation matches the usual planar "$D$ on the left" convention: if $\gamma(t)=(x(t),y(t))$ is a boundary parametrisation with $D$ on the left, the outward normal is $\nu_{\rm out}=(y',-x')/|\gamma'|$, and \begin{align*} \det(\nu_{\rm out}\mid T)=\frac{(y')^2+(x')^2}{|\gamma'|}=|\gamma'|>0, \end{align*} so $(\nu_{\rm out},T)$ is positive in $\mathbb R^2$. Hence the outer boundary is ccw, inner boundaries are cw, and $\int_{\partial D}\iota^*\omega=\int_{\partial D}P\,dx+Q\,dy$. Stokes' theorem $\int_Dd\omega=\int_{\partial D}\iota^*\omega$ delivers the claim. [/proof] Green's theorem is the $n=2$, $k=1$ case of the generalised Stokes theorem. The "scalar curl" $\partial_xQ-\partial_yP$ that classical multivariable calculus introduces ad hoc is the coefficient of $d\omega$ in coordinates — a derived quantity, not a definition. This is the structural point of the de Rham machinery: the apparently disparate classical theorems (Green, Stokes, divergence) are one statement in three dimensional disguises. [/example] In this example the two-form $d\omega$ is the signed area-density form whose coefficient is the scalar curl of the planar vector field $(P,Q)$. [example: Classical Stokes Theorem For Curl] Let $\Sigma\subset\mathbb R^3$ be a compact oriented smooth surface with smooth boundary $\partial\Sigma$ (boundary-oriented), with unit normal $\nu$ compatible with the orientation. For $F=(P,Q,R)\in C^\infty$ on a neighbourhood of $\Sigma$, set $\omega=P\,dx+Q\,dy+R\,dz$. [claim] \begin{align*} \int_{\partial\Sigma}F\cdot dr=\int_\Sigma(\operatorname{curl}F)\cdot\nu\,dS. \end{align*} [/claim] [proof] **$d\omega$ is the curl flux form.** The Chapter 3 coordinate formula gives \begin{align*} d\omega=(R_y-Q_z)\,dy\wedge dz+(P_z-R_x)\,dz\wedge dx+(Q_x-P_y)\,dx\wedge dy, \end{align*} whose coefficients are the three components of $\operatorname{curl}F$. **Surface side.** For a positively oriented parametrisation $\phi(u,v)$ with $\phi_u\times\phi_v=|N|\nu$, the pullbacks $\phi^*(dy\wedge dz)$, $\phi^*(dz\wedge dx)$, $\phi^*(dx\wedge dy)$ are the components of $\phi_u\times\phi_v$ times $du\wedge dv$ (the parametrised-surface example of the previous chapter). Hence \begin{align*} \phi^*(d\omega)=((\operatorname{curl}F)\circ\phi)\cdot(\phi_u\times\phi_v)\,du\wedge dv=(\operatorname{curl}F\cdot\nu)|N|\,du\wedge dv, \end{align*} and integrating gives $\int_\Sigma d\omega=\int_\Sigma(\operatorname{curl}F)\cdot\nu\,dS$. **Boundary side.** A boundary parametrisation $\gamma(t)$ gives $\gamma^*\omega=F(\gamma)\cdot\gamma'(t)\,dt$, so $\int_{\partial\Sigma}\omega=\int_{\partial\Sigma}F\cdot dr$. **Apply Stokes.** $\int_\Sigma d\omega=\int_{\partial\Sigma}\omega$ then delivers the identity. [/proof] The classical "curl theorem" (sometimes called Kelvin–Stokes) is the $n=2$, $k=1$ form of Stokes' theorem on a $2$-manifold embedded in $\mathbb R^3$ — the dimension didn't go up, but the embedding lets us re-express the wedge structure of $d\omega$ as a flux integral involving the surface's external geometry. Without the differential-forms framework, the classical proof requires triangulating the surface and applying Green's theorem patch by patch; here it falls out of one identity. [/example] This is the version traditionally called Stokes theorem in vector calculus. The differential-form statement removes the need to remember a separate curl theorem. [example: Divergence Theorem From Differential Forms] Let $V\subset\mathbb R^3$ be a compact oriented solid region with smooth boundary, oriented by $dx\wedge dy\wedge dz$, with outward unit normal $\nu$ on $\partial V$. For $F=(P,Q,R)\in C^\infty$ on a neighbourhood of $V$, set \begin{align*} \omega=P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy. \end{align*} [claim] \begin{align*} \int_{\partial V}F\cdot\nu\,dS=\int_V\operatorname{div}F\,d\mathcal L^3. \end{align*} [/claim] [proof] The Chapter 3 coordinate formula gives $d\omega=(P_x+Q_y+R_z)\,dx\wedge dy\wedge dz=\operatorname{div}F\,dx\wedge dy\wedge dz$, so $\int_Vd\omega=\int_V\operatorname{div}F\,d\mathcal L^3$. For the boundary, the outward-normal-first orientation makes a positively oriented parametrisation $\psi(u,v)$ of $\partial V$ satisfy $\psi_u\times\psi_v=|N|\nu$, and the parametrised-surface pullback identity from Chapter 6 gives \begin{align*} \iota^*\omega=F\cdot\nu\,dS \qquad \text{on each oriented boundary patch.} \end{align*} Hence $\int_{\partial V}\iota^*\omega=\int_{\partial V}F\cdot\nu\,dS$, and Stokes' theorem $\int_Vd\omega=\int_{\partial V}\iota^*\omega$ delivers the identity. [/proof] The [divergence theorem](/theorems/2754) is the $n=3$, $k=2$ case of Stokes. Together with the $k=1$ (curl) and $k=0$ (FTC) instances, the classical trilogy of vector-calculus integral theorems are recovered as a single identity $\int_Md\omega=\int_{\partial M}\omega$ specialised to the three possible codegrees in three dimensions. From here, "Stokes on a manifold" is just the same machinery on arbitrary $(M,\partial M)$ — no new content beyond what compactness, orientation, and the local model on $H^n$ already provide. [/example] The [divergence theorem](/theorems/2754) is therefore Stokes theorem for a two-form on a three-dimensional manifold. The boundary orientation is what enforces the outward normal in the usual statement. ## Applying The Theorem Reliably When using Stokes theorem in a calculation, which choices must be fixed before computing? First choose the oriented manifold $M$ and record whether it has boundary. Next choose the $(n-1)$-form $\omega$ whose [exterior derivative](/theorems/1525) is the desired interior integrand. Then check compact support, or compactness of $M$, so that the integrals are defined without an additional term at infinity. The last step is the orientation check. The boundary orientation determines whether the boundary is traversed counterclockwise, clockwise, outward-normal-first, or with another sign convention inherited from $M$. Most sign errors in applications of Stokes come from replacing this induced orientation with an informal geometric guess. This chapter completes the integration theory of differential forms. The next part of the course studies what happens when $d\omega=0$ and asks whether such a closed form must be of the form $d\eta$; the answer is de Rham cohomology. Having developed the full integration machinery, the fundamental question becomes: when can we solve dη = ω for a given closed form ω? The Poincaré lemma answers this locally on contractible domains, but the failure of exactness globally will measure the topology of the space. # 8. The Poincaré Lemma The previous chapters developed differential forms, pullback, exterior differentiation, and [Stokes' theorem](/theorems/1530) as a unified language for integration. This chapter asks when the equation $d\eta=\omega$ can be solved, given the necessary condition $d\omega=0$. The answer is local: on domains that can be contracted to a point in a controlled way, closed forms of positive degree are exact. The same statement fails on spaces with holes, and the failure becomes the first visible link between differential forms and topology. The central tool is a homotopy operator. It turns a geometric contraction into an algebraic identity $dK+Kd=\operatorname{id}$ on positive-degree forms, so the proof is not a search for primitives by guesswork. The angle form on the punctured plane then shows why the hypothesis matters: a closed form can have nonzero integral around a loop, and no exact form can do that. ## Closed Forms and the Local Exactness Problem When does the differential equation $d\eta=\omega$ have a solution for a given differential form $\omega$? Since $d^2=0$, every exact form is closed, so closedness is the first compatibility condition. The Poincare lemma says that, on a star-shaped region of Euclidean space, this compatibility condition is also sufficient in positive degrees. [definition: Closed and Exact Forms] Let $U \subseteq \mathbb R^n$ be open, and let $\omega \in \Omega^k(U)$. The form $\omega$ is closed if $d\omega=0$. For $k \ge 1$, the form $\omega$ is exact if there exists $\eta \in \Omega^{k-1}(U)$ such that $\omega=d\eta$. [/definition] This terminology packages the cohomological question into a quotient: closed forms are potential cohomology classes, and exact forms are the classes counted as zero. Thus \begin{align*} H^k_{\mathrm{dR}}(U) = \frac{\ker(d:\Omega^k(U)\to \Omega^{k+1}(U))}{\operatorname{im}(d:\Omega^{k-1}(U)\to \Omega^k(U))}. \end{align*} The lemma will prove that this quotient vanishes in positive degrees for star-shaped open sets. The geometric hypothesis says that the whole domain can be shrunk linearly to one point without leaving the domain. [definition: Star-Shaped Open Set] Let $U \subseteq \mathbb R^n$ be open and let $a \in U$. The set $U$ is star-shaped with centre $a$ if, for every $x \in U$ and every $t \in [0,1]$, the point $a+t(x-a)$ lies in $U$. [/definition] Star-shaped domains include balls, convex open sets, and all of $\mathbb R^n$. The hypothesis is stronger than contractibility, but it is exactly the condition needed for the explicit radial homotopy used in the proof. [illustration:forms-radial-contraction] [quotetheorem:832] The statement says that positive-degree de Rham cohomology has no local content on such domains. Any obstruction to exactness must therefore come from the global shape of the domain rather than from the differential equation alone. The proof reduces the lemma to the construction and verification of $K$. The next section gives the formula, since the same pattern will later prove homotopy invariance of cohomology. [example: Exact One-Forms on the Plane] Let $\omega=P\,dx+Q\,dy\in\Omega^1(\mathbb R^2)$ with $P,Q\in C^\infty(\mathbb R^2)$. The Chapter 3 formula $d\omega=(\partial_xQ-\partial_yP)\,dx\wedge dy$ makes the closedness condition $d\omega=0$ equivalent to $\partial_xQ=\partial_yP$. [claim] If $d\omega=0$, the function \begin{align*} f(x,y)=\int_0^1\bigl(xP(tx,ty)+yQ(tx,ty)\bigr)dt \end{align*} satisfies $df=\omega$. [/claim] [proof] Differentiation under the integral (justified by smoothness of the integrand on the compact interval) gives \begin{align*} \partial_xf =\int_0^1\bigl(P(tx,ty)+xt\,\partial_xP+yt\,\partial_xQ\bigr)dt. \end{align*} Substituting $\partial_xQ=\partial_yP$ (the closedness hypothesis) makes the integrand recognisable as the total $t$-derivative of $tP(tx,ty)$: \begin{align*} \frac{d}{dt}\bigl(tP(tx,ty)\bigr)=P(tx,ty)+t(x\,\partial_xP+y\,\partial_yP). \end{align*} Hence $\partial_xf=tP(tx,ty)\big|_0^1=P(x,y)$. The symmetric calculation, using the same identity $\partial_xQ=\partial_yP$ now to rewrite $\partial_yP$ inside the integrand, gives $\partial_yf=Q(x,y)$. So $df=P\,dx+Q\,dy=\omega$. [/proof] The function $f$ is constructed by *integrating $\omega$ along the radial segment from $0$ to $(x,y)$* — a one-dimensional path. The closedness condition is exactly the cross-partials test that makes this radial integral *path-independent* among all paths in $\mathbb R^2$, even though we only used radial paths in the construction. This is the prototype of the homotopy operator for the Poincaré lemma: replace $\mathbb R^2$ by any star-shaped domain in $\mathbb R^n$ and the same recipe (with $n$ coordinate functions instead of $2$) builds a global potential. [/example] This example is the differential-form version of finding a potential function for a conservative vector field. The point is that the formula uses the whole straight segment from $0$ to $(x,y)$, so it depends on the star-shaped geometry of the domain. ## The Homotopy Operator How can a contraction of a domain produce a primitive for every closed form at once? The answer is to integrate the form along the contraction direction while keeping the remaining variables as tangent inputs. This produces an operator of degree $-1$, and its commutator with $d$ measures the difference between the two ends of the homotopy. [definition: Homotopy Operator for a Radial Contraction] Let $U \subseteq \mathbb R^n$ be star-shaped with centre $a$. For $t\in[0,1]$, set \begin{align*} H_t(x)=a+t(x-a). \end{align*} For each $k\ge 1$, the homotopy operator is the [linear map](/page/Linear%20Map) \begin{align*} K &: \Omega^k(U)\to\Omega^{k-1}(U) \end{align*} defined by \begin{align*} (K\omega)_x(v_1,\dots,v_{k-1}) = \int_0^1 t^{k-1}\,\omega_{a+t(x-a)}(x-a,v_1,\dots,v_{k-1})\,dt, \end{align*} where $\omega\in\Omega^k(U)$, $x\in U$, and $v_1,\dots,v_{k-1}\in\mathbb R^n$. For $k=0$, set $Kf=0$ for $f\in\Omega^0(U)$. [/definition] The factor $t^{k-1}$ records how the spatial tangent vectors are scaled by the radial map, while the inserted vector $x-a$ is the contraction direction. The star-shaped hypothesis ensures that every point $a+t(x-a)$ at which the integrand is evaluated remains inside $U$. [quotetheorem:3583] This is the chain-homotopy identity behind the Poincare lemma. It says that the identity pullback and the constant pullback induce the same map on cohomology, with $K$ recording the explicit coboundary between them. The hypotheses are doing real work: the homotopy must be smooth so that pullback, exterior differentiation, and integration in $t$ interact as stated, and the star-shaped condition keeps the entire segment $a+t(x-a)$ inside $U$. If the segment leaves $U$, then $H_t$ is not a homotopy through maps $U\to U$, the formula for $K$ is not defined on all of $U$, and global periods such as those on the punctured plane can survive. The formula also explains why functions behave differently. For $k=0$, closedness means $df=0$, and the identity becomes $Kd f=f-f(a)$, which is the usual [fundamental theorem of calculus](/theorems/632) along the line from $a$ to $x$. [example: Recovering the Fundamental Theorem Along Rays] Let $U\subseteq\mathbb R^n$ be star-shaped with centre $a$, take $f\in C^\infty(U)$, and consider the radial path $\gamma_x(t)=a+t(x-a)$. The homotopy operator $K$ for a $1$-form on a star-shaped domain has weight factor $t^{k-1}=t^0=1$ in degree $k=1$, so \begin{align*} Kdf(x)=\int_0^1(df)_{\gamma_x(t)}(x-a)\,dt. \end{align*} The integrand is the chain-rule derivative of $f\circ\gamma_x$: $\gamma_x'(t)=x-a$, so $(df)_{\gamma_x(t)}(x-a)=\frac{d}{dt}f(\gamma_x(t))$. The one-variable [fundamental theorem of calculus](/theorems/632) gives \begin{align*} Kdf(x)=\int_0^1\frac{d}{dt}f(\gamma_x(t))\,dt=f(x)-f(a). \end{align*} The general homotopy identity $dK+Kd=\mathrm{id}-c_a^*$ in degree zero (where $K=0$ on $0$-forms and $c_a^*f=f(a)$) reduces to $Kdf=f-f(a)$ — exactly what we computed. So the FTC is *not* an auxiliary tool used to prove the Poincaré lemma; it *is* the degree-zero case of the Poincaré lemma. The radial weight factor $t^{k-1}$ in higher degrees is what makes the chain-rule trick continue to work for $k$-forms. [/example] Thus $K$ is not an artificial device. It is the higher-degree analogue of integrating a derivative along a path, with the contraction direction inserted into the form. ## The Angle Form on the Punctured Plane What breaks when the domain has a hole? The punctured plane $\mathbb R^2\setminus\{0\}$ admits loops that cannot be contracted through the domain, and a closed form can record how many times such a loop winds around the missing point. On $\mathbb R^2\setminus\{0\}$, define the angle form \begin{align*} \omega_\theta=\frac{-y\,dx+x\,dy}{x^2+y^2}. \end{align*} On any angular coordinate chart, this form is the differential of the local angle coordinate $\theta$. There is no globally defined smooth function $\theta:\mathbb R^2\setminus\{0\}\to\mathbb R$ whose differential is $\omega_\theta$, and the integral around the unit circle detects this failure. [illustration:forms-angle-winding-puncture] [quotetheorem:3584] This gives a practical test for non-exactness. To disprove exactness of a closed $1$-form, it is enough to find one closed curve on which the integral is nonzero. The converse requires global information: on a simply connected domain, closed $1$-forms have path-independent integrals and are exact, but on a general domain a closed form may be locally exact while nonzero periods obstruct a global primitive. In higher degree, the analogous obstruction is the integral of a closed $k$-form over a closed $k$-cycle rather than over a loop. The angle form passes the closedness test but fails the period test. That is exactly the phenomenon excluded by the Poincare lemma on star-shaped domains. [example: The Angle Form Is Closed But Not Exact] On $M=\mathbb R^2\setminus\{0\}$, take the angle form \begin{align*} \omega_\theta=\frac{-y\,dx+x\,dy}{x^2+y^2}=P\,dx+Q\,dy. \end{align*} [claim] $\omega_\theta$ is closed but not exact on $M$. Consequently $H^1_{\rm dR}(\mathbb R^2\setminus\{0\})\neq 0$. [/claim] [proof] **Closed.** The closedness criterion $\partial_xQ-\partial_yP=0$ holds: with $s=x^2+y^2$, the quotient rule gives \begin{align*} \partial_yP=-\frac{1}{s}+\frac{2y^2}{s^2}=\frac{y^2-x^2}{s^2}, \qquad \partial_xQ=\frac{1}{s}-\frac{2x^2}{s^2}=\frac{y^2-x^2}{s^2}, \end{align*} which are equal. **Not exact.** Along $\gamma(t)=(\cos t,\sin t)$ on $[0,2\pi]$, the denominator pulls back to $1$, and $\gamma^*(-y\,dx+x\,dy)=\sin^2t\,dt+\cos^2t\,dt=dt$, so $\int_\gamma\omega_\theta=\int_0^{2\pi}dt=2\pi$. If $\omega_\theta=df$ for $f\in C^\infty(M)$, the [fundamental theorem of calculus](/theorems/632) along the closed loop $\gamma$ would force $\int_\gamma\omega_\theta=f(\gamma(2\pi))-f(\gamma(0))=0\neq 2\pi$, a contradiction. **Cohomology.** $[\omega_\theta]\in\ker d/\operatorname{im}d=H^1_{\rm dR}(M)$ is nonzero because $\omega_\theta$ is closed but not in the image of $d$. [/proof] The number $2\pi$ is the *period* of $\omega_\theta$ around the unit circle, and it is a topological invariant of $M$: any loop encircling the origin once gives the same period, by closedness and Stokes' theorem. The non-vanishing period detects the hole at the origin, and de Rham cohomology turns this analytic obstruction into a vector space. The same construction in higher dimensions — $\mathbb R^n\setminus\{0\}$, the analogous angular form, the unit sphere $S^{n-1}$ as the integration cycle — gives the canonical generator of $H^{n-1}_{\rm dR}(\mathbb R^n\setminus\{0\})\cong\mathbb R$. [/example] The notation $d\theta$ is therefore local notation on the punctured plane, not the differential of a global real-valued angle function. Its nonzero period measures the obstruction to choosing a continuous single-valued angle around the origin. [remark: Topological Obstruction] The contrapositive of the Poincare lemma is useful: if an [open set](/page/Open%20Set) admits a closed positive-degree form that is not exact, then the [open set](/page/Open%20Set) is not star-shaped. In the punctured plane, the nonzero period of $\omega_\theta$ detects the missing origin. Later computations refine this observation by identifying $H^1_{\mathrm{dR}}(\mathbb R^2\setminus\{0\})$ with a one-dimensional [vector space](/page/Vector%20Space) generated by $[\omega_\theta]$. [/remark] The contrast with the plane is sharp. Removing a single point changes which loops can be collapsed, and the angle form records that change through its period. The next example returns to $\mathbb R^2$ itself to isolate the role of the missing point: once the radial contraction is available again, every closed one-form has a global potential. [example: Contrast with the Whole Plane] Let $\alpha\in\Omega^1(\mathbb R^2)$ be closed. Since $\mathbb R^2$ is star-shaped with centre $0$ (every $t\in[0,1]$ keeps $t(x,y)\in\mathbb R^2$), the *Poincaré lemma for star-shaped domains* delivers $f\in C^\infty(\mathbb R^2)$ with $df=\alpha$. Hence $[\alpha]=0$ in $H^1_{\rm dR}(\mathbb R^2)$, and the first de Rham cohomology of $\mathbb R^2$ vanishes. For any closed curve $\gamma:[0,1]\to\mathbb R^2$ with $\gamma(0)=\gamma(1)$, exactness forces \begin{align*} \int_\gamma\alpha=\int_0^1d(f\circ\gamma)=f(\gamma(1))-f(\gamma(0))=0. \end{align*} Every closed $1$-form on $\mathbb R^2$ has zero period around every closed loop. The contrast with the punctured plane is the entire content of de Rham cohomology in low dimension: removing a single point changes $H^1$ from $0$ to $\mathbb R$. The mechanism is the same star-shaped-or-not dichotomy — $\mathbb R^2$ is star-shaped about every interior point, $\mathbb R^2\setminus\{0\}$ is not star-shaped about *any* point because the deleted origin always blocks the radial contraction. The Mayer–Vietoris sequence in the next chapter will systematise this picture for arbitrary open sets in $\mathbb R^n$. [/example] ## Homotopy Invariance of de Rham Cohomology Why should a deformation of a map preserve the induced map on cohomology? The homotopy operator proves that two pullbacks along homotopic maps differ by an exact term on closed forms. This turns de Rham cohomology into a topological invariant rather than a construction depending only on coordinates. [definition: Smooth Homotopy] Let $M$ and $N$ be smooth manifolds, and let $f,g:M\to N$ be smooth maps. A smooth homotopy from $f$ to $g$ is a smooth map $F:M\times[0,1]\to N$ such that $F(p,0)=f(p)$ and $F(p,1)=g(p)$ for every $p\in M$. [/definition] The interval direction in $M\times[0,1]$ plays the same role as the radial direction in the proof of the Poincare lemma. Integrating the contraction of $F^*\omega$ against $\partial_t$ gives a degree $-1$ operator between forms on $N$ and forms on $M$. [quotetheorem:3585] This theorem is the global version of the homotopy operator identity. It says that cohomology cannot distinguish maps that are connected by a smooth deformation. The proof uses no special feature of Euclidean domains beyond the existence of a smooth homotopy. This is why the Poincare lemma extends from star-shaped subsets of $\mathbb R^n$ to smoothly contractible manifolds. [quotetheorem:3586] This form of the lemma says that positive-degree de Rham cohomology is invariant under smooth contraction. It also explains why nonzero cohomology classes are obstructions to contractibility. The angle form on $\mathbb R^2\setminus\{0\}$ is therefore not just a counterexample to a naive exactness statement. It proves that the punctured plane is not smoothly contractible, and it anticipates the general principle that de Rham cohomology detects holes through integrals of closed forms over cycles. The Poincaré lemma showed that closed forms are not always exact. De Rham cohomology makes this obstruction precise by quotient—it measures the space of closed forms modulo exact forms—and the resulting invariant captures the global topology of the manifold. # 9. de Rham Cohomology This chapter turns the calculus of differential forms into a cohomology theory. The guiding question is: when does a closed form arise as an [exterior derivative](/theorems/1525)? The quotient by exact forms measures the obstruction, and the answer depends only on the global shape of the manifold rather than on a particular coordinate system. We use the Poincare lemma from the previous chapter as the local input and Mayer-Vietoris as the mechanism for assembling local information into global computations. ## Closed Forms Modulo Exact Forms The equation $d\alpha=0$ is a local compatibility condition, but solving $\alpha=d\beta$ asks for a single primitive defined on all of $M$. The obstruction appears when local primitives disagree as one moves around loops or across overlaps of coordinate charts. De Rham cohomology is designed to keep exactly this residual obstruction: two closed forms represent the same class when their difference is globally exact. [definition: Closed Differential Form] Let $M$ be a smooth manifold. A $k$-form $\alpha\in \Omega^k(M)$ is closed if $d\alpha = 0$. [/definition] Closedness is the equation imposed by the [exterior derivative](/theorems/1525). Exactness is stronger because it asks the form to have a global primitive. [definition: Exact Differential Form] Let $M$ be a smooth manifold. A $k$-form $\alpha\in \Omega^{k}(M)$ is exact if there exists $\beta\in \Omega^{k-1}(M)$ such that $\alpha = d\beta$. [/definition] Since $d^2=0$, every exact form is closed. The failure of the converse is the central object of this chapter. [definition: De Rham Cohomology] Let $M$ be a smooth manifold. The $k$-th de Rham cohomology group of $M$ is \begin{align*} H^k_{\mathrm{dR}}(M) := \frac{\ker(d:\Omega^k(M)\to \Omega^{k+1}(M))}{\operatorname{im}(d:\Omega^{k-1}(M)\to \Omega^k(M))}. \end{align*} An element of $H^k_{\mathrm{dR}}(M)$ is called a de Rham cohomology class. [/definition] We write $[\alpha]$ for the class of a closed form $\alpha$. Thus $[\alpha]=[\alpha']$ exactly when $\alpha-\alpha'=d\beta$ for some $\beta\in\Omega^{k-1}(M)$. [example: Closed Not Exact On The Circle] On $S^1$, take the angle form $\alpha=d\theta=(-y\,dx+x\,dy)|_{S^1}$ — well defined globally as the restriction of a $1$-form on $\mathbb R^2$. The circle is $1$-dimensional, so $\Omega^2(S^1)=0$ (any alternating bilinear form on a $1$-dimensional tangent space vanishes), which makes the closedness $d\alpha\in\Omega^2(S^1)=0$ automatic. [claim] $\alpha$ is not exact on $S^1$; consequently $[\alpha]\neq 0$ in $H^1_{\rm dR}(S^1)$. [/claim] [proof] The positive parametrisation $\gamma(t)=(\cos t,\sin t)$ pulls $\alpha$ back to $dt$ (Chapter 4), so \begin{align*} \oint_{S^1}d\theta=\int_0^{2\pi}dt=2\pi. \end{align*} If $\alpha=df$ for some $f\in C^\infty(S^1)$, then $F=f\circ\gamma$ satisfies $F'(t)\,dt=\gamma^*\alpha=dt$, so $F'\equiv 1$ and $F(2\pi)-F(0)=2\pi$. But $\gamma(0)=\gamma(2\pi)$, hence $F(2\pi)=F(0)$. The two conclusions contradict, so no such $f$ exists. [/proof] This is the same non-exactness obstruction as the angle form on $\mathbb R^2\setminus\{0\}$ from the previous chapter, transplanted to the circle itself. The number $2\pi$ is the integral of the generator of $H^1_{\rm dR}(S^1)$ over the fundamental cycle, and once we normalise $\alpha$ by dividing by $2\pi$, we obtain a generator whose period is $1$ — the canonical bridge to the integer-coefficient comparison theorem. [/example] This example is the prototype: the equation $d\alpha=0$ is local, while being $d\beta$ for a globally defined $\beta$ is a global condition. ## Low-Degree Cohomology Before computing new examples, it is useful to ask what the definition says in the lowest degrees. Degree zero is controlled by functions with zero derivative, while positive-degree cohomology of a point vanishes because there are no positive-degree forms on a point. [quotetheorem:3587] The zeroth group therefore detects connected components, but the hypotheses are doing real work. The proof uses the fact that connected components of a smooth manifold are path connected, so a zero derivative forces constancy along paths. In a general topological space this implication has no differential meaning, and even in pathological smooth settings without the usual manifold assumptions the relation between components and paths can fail. If $M$ has infinitely many connected components, the notation $\mathbb R^{\pi_0(M)}$ means the [vector space](/page/Vector%20Space) of all real-valued functions on the set of components, not necessarily a finite-dimensional space. Higher groups detect higher-dimensional holes, as the computations below will show. [quotetheorem:3588] This follows from the description of $H^0$ and the fact that $\Omega^k(\{p\})=0$ for $k\ge 1$: the tangent space of a zero-dimensional manifold is zero, so there are no non-zero alternating $k$-linear forms once $k\ge 1$. The dimension is essential here; even a one-dimensional manifold can carry non-zero $1$-forms, and a circle carries a closed $1$-form which is not exact. Contractibility removes that global obstruction in the local Euclidean situation: the Poincare lemma says that every closed positive-degree form on a contractible open subset of $\mathbb R^n$ is exact. Without contractibility, local exactness need not assemble into a global primitive, which is precisely the failure measured by de Rham cohomology. [example: Disconnected Manifold] Let $M=S^1\sqcup S^2$. [claim] $H^0_{\rm dR}(M)\cong\mathbb R^2$, with one coordinate per connected component. [/claim] [proof] Since $\Omega^{-1}=0$, the zeroth cohomology is just $\ker(d:\Omega^0\to\Omega^1)$ — the closed $0$-forms, equivalently locally constant smooth functions. A smooth function on $M$ is a pair $(f_1,f_2)$ with $f_1\in C^\infty(S^1)$ and $f_2\in C^\infty(S^2)$. Each $S^1$ and $S^2$ is path connected, so $df_i=0$ together with the fundamental theorem of calculus along a piecewise smooth path between any two points forces $f_i\equiv a_i$ for some constant $a_i\in\mathbb R$ on that component. Conversely, any function constant on each component has $df=0$. Hence $\ker d\cong\mathbb R\oplus\mathbb R=\mathbb R^2$ via $(f_1,f_2)\mapsto(a_1,a_2)$. [/proof] This is the prototype for $H^0_{\rm dR}(M)\cong\mathbb R^{\#(\pi_0(M))}$: the dimension of zeroth de Rham cohomology counts connected components. Combined with the universal coefficient theorem in the comparison chapter, this also gives $H_0(M;\mathbb R)\cong\mathbb R^{\#(\pi_0)}$, confirming that "number of connected components" is the same topological invariant counted by either the singular or the de Rham machinery. [/example] ## Mayer-Vietoris For De Rham Cohomology Local computations are useful only if they can be glued. Suppose $M=U\cup V$ with $U$ and $V$ open. A form on $M$ restricts to forms on $U$ and $V$ which agree on $U\cap V$, and the Mayer-Vietoris sequence packages this gluing condition into a long exact sequence in cohomology. [quotetheorem:3589] The openness of $U$ and $V$ is not a cosmetic assumption. Differential forms restrict naturally to open submanifolds, and smooth partitions of unity subordinate to open covers are what make the short exact sequence of complexes exact on the right. For arbitrary closed covers, extension by zero can fail to be smooth at the boundary, so the same proof does not apply without extra collar or relative hypotheses. Mayer-Vietoris also does not assert that $H^k_{\mathrm{dR}}(M)$ splits as a direct sum of the neighbouring groups; exactness controls kernels and images, while the connecting homomorphism records the obstruction to such naive splitting. The connecting homomorphism is the part of the sequence that carries the geometry. It turns a class on the overlap into the obstruction to choosing compatible primitives on the two pieces. [explanation: Connecting Homomorphism] Given a closed $k$-form $\eta\in\Omega^k(U\cap V)$, choose forms $\beta\in\Omega^k(U)$ and $\gamma\in\Omega^k(V)$ such that $\beta|_{U\cap V}-\gamma|_{U\cap V}=\eta$. The forms $d\beta$ and $d\gamma$ agree on $U\cap V$, so they glue to a closed $(k+1)$-form on $M$. The cohomology class of this glued form is $\delta[\eta]\in H^{k+1}_{\mathrm{dR}}(M)$. Different choices of $\beta$ and $\gamma$ change the glued form by an exact form, so the class is well-defined. [/explanation] Exactness means that the image of each map is the kernel of the next. In computations, this lets us recover unknown groups by comparing the known cohomology of $U$, $V$, and $U\cap V$. ## Spheres From Mayer-Vietoris The sphere is the first major test case: it is locally contractible, but its top-dimensional cohomology should remember the fundamental enclosed volume. Mayer-Vietoris proves this by covering the sphere by two contractible caps whose overlap deformation retracts onto an equatorial sphere. [quotetheorem:3590] The restriction $n\ge 1$ avoids the exceptional zero-sphere: $S^0$ is two points, so its cohomology is concentrated in degree zero with $H^0_{\mathrm{dR}}(S^0)\cong\mathbb R^2$. For $n\ge 1$, the overlap in the two-cap cover must deformation retract onto $S^{n-1}$ because Mayer-Vietoris reads the cohomology of the whole sphere from the cohomology carried by this equatorial overlap. If the overlap were contractible, the sequence would lose the class that shifts into the top degree. The resulting top class is represented by a volume form, and its non-vanishing can also be detected by integration: an exact top-degree form has integral zero over a compact boundaryless oriented manifold, while a positive volume form has non-zero integral. This computation is also the template for later examples: choose a cover whose pieces are simple, identify the overlap, and use exactness to see which overlap classes become global classes one degree higher. [example: Full Mayer-Vietoris Computation For The Two-Sphere] Cover $S^2$ by $U=S^2\setminus\{S\}$ and $V=S^2\setminus\{N\}$, the two open neighbourhoods of the hemispheres. Stereographic projection identifies each with $\mathbb R^2$, which is star-shaped, so the Poincaré lemma gives \begin{align*} H^0(U)=H^0(V)=\mathbb R, \qquad H^k(U)=H^k(V)=0 \text{ for } k\ge 1. \end{align*} The intersection $U\cap V=S^2\setminus\{N,S\}$ is diffeomorphic to $S^1\times(-1,1)$ via $(x,y,z)\mapsto((x,y)/\sqrt{x^2+y^2},z)$, which deformation retracts onto $S^1\times\{0\}\cong S^1$. By homotopy invariance and the de Rham cohomology of $S^1$, \begin{align*} H^0(U\cap V)=\mathbb R, \qquad H^1(U\cap V)=\mathbb R, \qquad H^k(U\cap V)=0 \text{ for } k\ge 2. \end{align*} The Mayer–Vietoris long exact sequence pieces these together. The non-zero portion reads \begin{align*} 0\to H^0(S^2)\xrightarrow{r_0}\mathbb R^2\xrightarrow{s_0}\mathbb R\xrightarrow{\delta_0}H^1(S^2)\to 0 \to \mathbb R\xrightarrow{\delta_1}H^2(S^2)\to 0. \end{align*} *Degree $0$.* The difference map $s_0(a,b)=a-b$ has $\ker s_0=\{(a,a)\}\cong\mathbb R$ and $\operatorname{im}s_0=\mathbb R$. Exactness identifies $H^0(S^2)\cong\ker s_0\cong\mathbb R$, confirming $S^2$ is connected. *Degree $1$.* Exactness at $H^0(U\cap V)$ forces $\ker\delta_0=\operatorname{im}s_0=\mathbb R$, so $\delta_0=0$ and $\operatorname{im}\delta_0=0$. The next term in the sequence is $H^1(U)\oplus H^1(V)=0$, so $H^1(S^2)=\ker(\to 0)=\operatorname{im}\delta_0=0$. *Degree $2$.* The map $s_1:0\to\mathbb R$ has image $0$, so $\delta_1$ has trivial kernel by exactness — hence injective. Exactness at $H^2(S^2)$ then equates $\operatorname{im}\delta_1$ with $\ker(H^2(S^2)\to 0)=H^2(S^2)$, so $\delta_1$ is also surjective. Thus $H^2(S^2)\cong\mathbb R$. The pattern is mechanical and revealing: the two contractible pieces $U$, $V$ contribute no positive-degree cohomology of their own, but the topology of the overlap — the equatorial circle — gets "transferred up by one degree" by the connecting homomorphism $\delta$. This is the general Mayer–Vietoris philosophy: cohomology of the whole is assembled from cohomology of the pieces, with the overlap contributing degree shifts that detect the gluing topology. [/example] This computation is the model for higher spheres. The overlap carries the cohomology one dimension lower, and the connecting map shifts it into the top degree of the whole sphere. ## The Torus And Product Behaviour The torus has two independent circular directions, so we expect two independent degree-one classes and one degree-two class obtained by wedging them. This is formalised by the Kunneth theorem, which describes the cohomology of a product from the cohomology of its factors. [quotetheorem:3591] The proof uses the exterior product of forms and homological algebra; in this course we use it as a computational tool. The real coefficients matter: over a field, tensor products behave cleanly and there are no torsion correction terms. By contrast, integral singular cohomology has a Kunneth formula with possible $\operatorname{Tor}$ terms, so the displayed formula should not be copied unchanged into integer-coefficient topology. The finite-dimensional smooth-manifold setting also keeps the product differential forms under ordinary algebraic control; outside this setting, topological completions and infinite-dimensional phenomena require additional care. Applying it to $T^2=S^1\times S^1$ gives the cohomology of the torus. [quotetheorem:3592] If $\theta_1$ and $\theta_2$ are angular coordinates on the two circle factors, then $d\theta_1$ and $d\theta_2$ represent the two degree-one classes, while $d\theta_1\wedge d\theta_2$ represents the degree-two class. [example: Betti Numbers Of The Torus] For $T^2=S^1\times S^1$, the de Rham cohomology theorem for the two-torus gives $H^0(T^2)\cong\mathbb R$, $H^1(T^2)\cong\mathbb R^2$, $H^2(T^2)\cong\mathbb R$, and $H^k(T^2)=0$ for $k\geq 3$. Taking dimensions yields the Betti numbers \begin{align*} b_0(T^2)=1, \qquad b_1(T^2)=2, \qquad b_2(T^2)=1. \end{align*} The numbers $1,2,1$ are not coincidence — they are the dimensions of $\Lambda^kE$ for a two-dimensional vector space $E$. Take $E=\mathbb Re_1\oplus\mathbb Re_2$ with $e_1,e_2$ representing the two circle factors: $\Lambda^0E$ has basis $\{1\}$, $\Lambda^1E$ has basis $\{e_1,e_2\}$, and $\Lambda^2E$ has basis $\{e_1\wedge e_2\}$ (all higher powers vanish since $e_i\wedge e_i=0$). This matches because the de Rham cohomology of a product is the graded tensor product of the factors (Künneth in cohomology), and $H^*(S^1)=\mathbb R[e]/e^2$ with $|e|=1$, so $H^*(T^2)=H^*(S^1)\otimes H^*(S^1)$ is the free exterior algebra on two degree-one generators. The "two independent loops" intuition for the torus is exactly the two-dimensional space $H^1(T^2)$, and their wedge generates the area class in $H^2$. [/example] ## Betti Numbers And Compact Manifolds The examples above produced finite-dimensional cohomology groups, but this need not be taken for granted from the definition: the spaces of differential forms are infinite-dimensional. Compactness is the analytic hypothesis that brings finite-dimensionality into the theory. [quotetheorem:3593] In a full Hodge theory course, this theorem is proved by choosing a Riemannian metric, identifying each cohomology class with a unique harmonic representative, and using elliptic regularity to show that the space of harmonic $k$-forms is finite-dimensional. Here we use the result as a structural fact about compact manifolds. [definition: Betti Number] Let $M$ be a compact smooth manifold. The $k$-th Betti number of $M$ is \begin{align*} b_k(M):=\dim H^k_{\mathrm{dR}}(M). \end{align*} [/definition] Betti numbers turn cohomology groups into numerical invariants. They forget the preferred representatives of cohomology classes, but retain the number of independent classes in each degree. [example: Betti Numbers Of Spheres] For $n\geq 1$, the de Rham theorem for spheres gives $H^k(S^n)\cong\mathbb R$ for $k\in\{0,n\}$ and $H^k(S^n)=0$ otherwise. Hence \begin{align*} b_k(S^n)=\begin{cases}1, & k=0,\\ 1, & k=n,\\ 0, & \text{otherwise}.\end{cases} \end{align*} The two non-zero Betti numbers have distinct geometric meanings: $b_0=1$ records that $S^n$ is connected, and $b_n=1$ records the existence of a normalised top-degree volume form whose integral over $S^n$ equals the volume — equivalently, that $S^n$ is closed (compact without boundary) and orientable. All intermediate degrees vanish because $S^n$ has no non-trivial topology "in between": there are no embedded codimension-$k$ submanifolds whose duals could carry independent cohomology classes for $0<k<n$. This is a special structural feature of spheres; tori, projective spaces, and Lie groups all have richer cohomology profiles, and de Rham cohomology will distinguish them in the next chapter. [/example] De Rham cohomology is therefore a bridge between analysis and topology. It starts with smooth forms and the differential operator $d$, but its answers are stable under global deformation and reveal connected components, loops, enclosed voids, and their higher-dimensional analogues. De Rham cohomology encodes topological information through the [exterior derivative](/theorems/1525), but to make this precise, we must connect it to topology constructed from singular simplices. De Rham's theorem shows that these two cohomology theories coincide, revealing that differential forms compute topological invariants. # 10. de Rham's Theorem and Comparison with Singular Cohomology The earlier chapters built differential forms, exterior differentiation, pullback, integration on oriented manifolds, and the generalized Stokes theorem. This chapter compares that analytic complex with a topological complex built from singular simplices. The main point is that integrating closed forms over cycles loses no information: de Rham cohomology is naturally the same as singular cohomology with real coefficients. This comparison turns computations with forms into topological invariants and explains why periods, degrees, and Euler characteristics appear in the same theory. ## Singular Chains and Real Cochains How can a topological space be converted into algebra that remembers its holes? Singular homology answers this by mapping standard simplices into the space, allowing many parameterized pieces with formal real coefficients. Cohomology then takes linear functionals on those chains and studies which functionals vanish on boundaries. [definition: Standard Simplex] For $k \ge 0$, the standard $k$-simplex is \begin{align*} \Delta^k = \left\{(t_0,\dots,t_k) \in \mathbb R^{k+1} : t_i \ge 0, \sum_{i=0}^k t_i = 1\right\}. \end{align*} [/definition] The vertices of $\Delta^k$ are the coordinate vectors $e_0,\dots,e_k$. Faces are obtained by omitting one of these vertices. [definition: Singular Simplex] A singular $k$-simplex in a topological space $M$ is a continuous map $\sigma:\Delta^k \to M$. [/definition] For a smooth manifold, the underlying topological space is being used here. In the integration construction below, smooth singular simplices are used first; the smoothing theorem identifies their cohomology with ordinary singular cohomology with real coefficients. [definition: Singular Chain Group] For $k \ge 0$, the real singular chain group $C_k(M;\mathbb R)$ is the real [vector space](/page/Vector%20Space) of finite formal sums \begin{align*} c = \sum_{j=1}^N a_j \sigma_j, \end{align*} where $a_j \in \mathbb R$ and each $\sigma_j:\Delta^k \to M$ is a singular $k$-simplex. [/definition] A chain is a finite linear combination of parameterized simplices. The boundary operator records the oriented faces of each simplex. [definition: Boundary Operator] Let $\iota_i:\Delta^{k-1}\to \Delta^k$ be the affine map whose image is the face opposite $e_i$, determined on vertices by \begin{align*} \iota_i(e_j)= \begin{cases} e_j, & j<i,\\ e_{j+1}, & j\ge i. \end{cases} \end{align*} The boundary of a singular $k$-simplex $\sigma:\Delta^k \to M$ is \begin{align*} \partial \sigma = \sum_{i=0}^k (-1)^i \sigma \circ \iota_i, \end{align*} and $\partial:C_k(M;\mathbb R)\to C_{k-1}(M;\mathbb R)$ is extended linearly. [/definition] The alternating signs encode the orientation of the faces. They are chosen so that taking the boundary twice cancels every codimension-two face. [illustration:forms-simplex-boundary-signs] [quotetheorem:2232] The identity $\partial^2=0$ is what makes homology possible: boundaries automatically become cycles, so quotienting cycles by boundaries is consistent. Without this cancellation, a boundary could have a nonzero boundary of its own, and the phrase "cycles modulo boundaries" would not define the intended invariant. This is the chain-level analogue of $d^2=0$ for differential forms. Cochains reverse direction: they assign numbers to chains. This dual viewpoint is the one naturally reached by integrating differential forms. [definition: Singular Cochain Complex] The degree $k$ singular cochain group with real coefficients is \begin{align*} C^k_{\mathrm{sing}}(M;\mathbb R)=\operatorname{Hom}_{\mathbb R}(C_k(M;\mathbb R),\mathbb R). \end{align*} The coboundary map $\delta:C^k_{\mathrm{sing}}(M;\mathbb R)\to C^{k+1}_{\mathrm{sing}}(M;\mathbb R)$ is defined by \begin{align*} (\delta \varphi)(c)=\varphi(\partial c) \end{align*} for every $c \in C_{k+1}(M;\mathbb R)$. [/definition] Since $\partial^2=0$, the coboundary also squares to zero. Thus singular cochains form a cochain complex. [definition: Singular Cohomology] The singular cocycles, coboundaries, and cohomology groups are \begin{align*} Z^k_{\mathrm{sing}}(M;\mathbb R) &= \ker\left(\delta:C^k_{\mathrm{sing}}(M;\mathbb R)\to C^{k+1}_{\mathrm{sing}}(M;\mathbb R)\right),\\ B^k_{\mathrm{sing}}(M;\mathbb R) &= \operatorname{im}\left(\delta:C^{k-1}_{\mathrm{sing}}(M;\mathbb R)\to C^k_{\mathrm{sing}}(M;\mathbb R)\right),\\ H^k_{\mathrm{sing}}(M;\mathbb R) &= Z^k_{\mathrm{sing}}(M;\mathbb R)/B^k_{\mathrm{sing}}(M;\mathbb R). \end{align*} [/definition] The analogy with de Rham cohomology is now visible. In de Rham theory, closed forms are killed by $d$, exact forms are images under $d$, and $d^2=0$. In singular cohomology, cocycles are killed by $\delta$, coboundaries are images under $\delta$, and $\delta^2=0$. Real coefficients are essential for this comparison because integration produces [real numbers](/page/Real%20Numbers). Integer cohomology contains torsion information that real-valued periods cannot see; for example, torsion classes vanish after tensoring with $\mathbb R$ and therefore have no de Rham representative. [example: Zero-Dimensional Cohomology] Let $M$ be a nonempty path-connected smooth manifold. A singular $0$-cochain $\varphi$ assigns a real number to each point. The coboundary $\delta\varphi$ acts on a singular $1$-simplex $\sigma$ from $p=\sigma(e_0)$ to $q=\sigma(e_1)$ as \begin{align*} (\delta\varphi)(\sigma)=\varphi(\partial\sigma)=\varphi(q)-\varphi(p), \end{align*} so $\varphi$ is a $0$-cocycle iff $\varphi(q)=\varphi(p)$ for every singular path. Path-connectedness lifts this to *all* pairs $p,q\in M$, forcing $\varphi$ to be globally constant; conversely any constant cochain is a cocycle. There are no $(-1)$-cochains, hence no coboundaries, and $H^0_{\rm sing}(M;\mathbb R)\cong\mathbb R$ via $\varphi\mapsto$ (its common value). The de Rham side runs in parallel. A $0$-form is a smooth function $f$, and $df=0$ along every smooth path $\gamma$ gives $(f\circ\gamma)'\equiv 0$, so $f$ is constant on each path component — hence globally constant by path-connectedness of $M$. No $-1$-forms exist, so $H^0_{\rm dR}(M)\cong\mathbb R$ as well. Both invariants count connected components, by completely different mechanisms (one combinatorial, one analytic), and they agree. This is the first non-trivial agreement in the de Rham comparison theorem — a case where the matching is essentially tautological because both sides reduce to "locally constant real-valued functions modulo nothing". The general comparison theorem will extend this agreement to every degree. [/example] ## Integration as a Cochain Map What must be checked before integration over simplices can define a map on cohomology? A $k$-form can be pulled back to a smooth $k$-simplex and integrated over the standard simplex, giving a real number. The decisive point is compatibility with the two differentials: the [exterior derivative](/theorems/1525) $d$ on forms and the coboundary $\delta$ on singular cochains. [definition: Integration Cochain] Let $M$ be a smooth manifold and let $\omega \in \Omega^k(M)$. For a smooth singular $k$-simplex $\sigma:\Delta^k\to M$, define \begin{align*} I_k(\omega)(\sigma)=\int_{\Delta^k} \sigma^*\omega. \end{align*} Extend $I_k(\omega)$ linearly to smooth singular $k$-chains. [/definition] Smoothness of $\sigma$ is needed because $\sigma^*\omega$ uses derivatives of $\sigma$. For a merely continuous singular simplex, the pullback of a differential form is not defined in the usual smooth sense. The smoothing theorem is what permits this smooth-chain construction to recover ordinary singular cohomology. The orientation on $\Delta^k$ is the standard one induced by the ordered vertices $(e_0,\dots,e_k)$. With this convention, the boundary signs in singular homology match the boundary orientation in Stokes theorem. [quotetheorem:3594] This identity is the whole mechanism behind the comparison map. Closed forms give singular cocycles, and exact forms give singular coboundaries. [quotetheorem:3595] The formula says that a de Rham class is detected by its periods over cycles. In practice, to identify a class one chooses cycles representing a basis of homology, integrates the closed form over those cycles, and compares the resulting numbers with a known [dual basis](/theorems/414) in cohomology. On a compact connected oriented $n$-manifold, a normalized top-degree form represents the generator dual to the fundamental class when its integral over $M$ is $1$. The next example is the basic model for all period computations. [illustration:forms-s1-period-pairing] [example: The Period Pairing on the Circle] On $S^1$ with positive orientation, take the normalised angle form \begin{align*} \alpha=\frac{1}{2\pi}(-y\,dx+x\,dy)\big|_{S^1}\in\Omega^1(S^1), \end{align*} and parametrise the fundamental cycle by $\gamma:[0,1]\to S^1$, $\gamma(t)=(\cos 2\pi t,\sin 2\pi t)$. Pulling back coordinate differentials, $dx=-2\pi\sin 2\pi t\,dt$ and $dy=2\pi\cos 2\pi t\,dt$, so \begin{align*} \gamma^*(-y\,dx+x\,dy)=2\pi(\sin^2 2\pi t+\cos^2 2\pi t)\,dt=2\pi\,dt, \end{align*} and $\gamma^*\alpha=dt$. Integrating, \begin{align*} I([\alpha])([\gamma])=\int_{\Delta^1}\gamma^*\alpha=\int_0^1dt=1. \end{align*} The normalised angle form has period $1$ over the fundamental cycle by construction — the factor of $1/(2\pi)$ in $\alpha$ is *chosen* to make the pairing unit. Under the comparison isomorphism, $[\alpha]\in H^1_{\rm dR}(S^1)$ is therefore the dual of $[\gamma]\in H_1(S^1;\mathbb R)$. This duality is the prototype: for any closed oriented $n$-manifold, normalising the volume form to integrate to $1$ produces the de Rham generator dual to the fundamental class. [/example] ## The De Rham Comparison Theorem Why should integration over cycles capture every singular cohomology class and identify only exact ambiguity? This is the central comparison question. Local contractibility makes both theories look the same on coordinate balls, while partitions of unity allow the local comparison to be assembled over the manifold. [quotetheorem:3596] The theorem is stronger than equality of dimensions. It gives a canonical way to replace a differential form computation by a singular cochain computation, and conversely to represent real singular cohomology classes by closed differential forms. [remark: Ring Structure] The comparison is compatible with products: wedge product of forms corresponds to the cup product in singular cohomology. Thus [de Rham theorem](/theorems/3596) identifies $H^*_{\mathrm{dR}}(M)$ and $H^*_{\mathrm{sing}}(M;\mathbb R)$ as graded real algebras, after using the standard singular cup product conventions. [/remark] Naturality is especially useful when a map acts on top-degree cohomology. It turns the analytic operation of pulling back a volume form into the topological degree of a map. [example: Degree of a Smooth Self-Map of the Sphere] Let $f:S^n\to S^n$ be smooth, $n\geq 1$, with $S^n$ positively oriented. Fix a normalised orientation form $\omega\in\Omega^n(S^n)$ with $\int_{S^n}\omega=1$. By definition, the topological degree satisfies $f_*[S^n]=\deg(f)\,[S^n]$ in singular homology. [claim] \begin{align*} \int_{S^n}f^*\omega=\deg(f). \end{align*} [/claim] [proof] Since $\Omega^{n+1}(S^n)=0$, $d\omega=0$, so $[\omega]\in H^n_{\rm dR}(S^n)$. The class is nonzero: if $\omega=d\eta$, Stokes on the boundaryless $S^n$ gives $\int_{S^n}\omega=\int_\varnothing\eta=0$, contradicting the normalisation. By the de Rham theorem $H^n_{\rm dR}(S^n)\cong\mathbb R$, so $[\omega]$ is a basis vector, and there is a unique $\lambda\in\mathbb R$ with $f^*[\omega]=\lambda[\omega]$ — i.e., $f^*\omega-\lambda\omega=d\eta$ for some $\eta$. Integrating over $S^n$ and using Stokes, \begin{align*} \int_{S^n}f^*\omega-\lambda\int_{S^n}\omega=0 \quad\Longrightarrow\quad \int_{S^n}f^*\omega=\lambda. \end{align*} Naturality of the comparison map $I$ (which is a chain map) gives $I(f^*[\omega])=f^*_{\rm sing}I([\omega])$. Evaluating on $[S^n]$: \begin{align*} \lambda=I(\lambda[\omega])([S^n])=I(f^*[\omega])([S^n])=I([\omega])(f_*[S^n])=I([\omega])(\deg(f)\,[S^n])=\deg(f). \end{align*} [/proof] This is the most concrete instance of the de Rham theorem at work: a topological invariant (the degree, defined by lifting and counting preimages) is computed by an integral of a smooth top-degree form. The same formula generalises to closed oriented $n$-manifolds: $\int_M f^*\omega=\deg(f)\int_M\omega$ for any normalised volume form. It is the analytical engine behind the Gauss–Bonnet theorem in the next chapter, where the degree of the Gauss map equals the Euler characteristic. [/example] ## Poincare Duality, Betti Numbers, and Euler Characteristic What topological information becomes available once de Rham cohomology has been identified with singular cohomology? The comparison theorem imports the major structural results of algebraic topology into differential forms. It also gives analytic formulas for invariants that were originally defined by triangulations or singular chains. [definition: De Rham Betti Number] If $H^k_{\mathrm{dR}}(M)$ is finite-dimensional, the degree $k$ Betti number of $M$ is \begin{align*} b_k(M)=\dim_{\mathbb R} H^k_{\mathrm{dR}}(M). \end{align*} [/definition] By [de Rham theorem](/theorems/3596), this agrees with $\dim_{\mathbb R}H^k_{\mathrm{sing}}(M;\mathbb R)$. Therefore the Betti numbers are topological invariants of the underlying space. [quotetheorem:3597] This consequence is conceptually important: the de Rham groups are defined using smooth forms, but their isomorphism type depends only on the topology. Smoothness is still needed to define the complex $\Omega^*(M)$ and to produce explicit pullback maps on forms; a general homeomorphism need not pull back smooth forms to smooth forms. [De Rham theorem](/theorems/3596) supplies the missing bridge by passing through singular cohomology, so homeomorphic smooth manifolds have isomorphic de Rham groups even when the homeomorphism is not a diffeomorphism. [quotetheorem:3598] The proof is not part of this chapter. It is usually obtained either from singular Poincare duality together with [de Rham theorem](/theorems/3596), or analytically from Hodge theory after choosing a Riemannian metric. The hypotheses matter: non-orientability removes the global fundamental class needed for this real pairing, and noncompactness can make the pairing degenerate unless compactly supported cohomology is used. For instance, $\mathbb R$ has $H^0_{\mathrm{dR}}(\mathbb R)\cong \mathbb R$ and $H^1_{\mathrm{dR}}(\mathbb R)=0$, so the symmetry $b_0=b_1$ fails without compactness. [definition: Euler Characteristic] For a smooth manifold $M$ with finite-dimensional de Rham cohomology and only finitely many nonzero Betti numbers, its Euler characteristic is \begin{align*} \chi(M)=\sum_{k\ge 0}(-1)^k b_k(M). \end{align*} [/definition] This formula is often the most efficient way to compute $\chi(M)$ once de Rham cohomology is known. The equality with the topological Euler characteristic follows from [de Rham theorem](/theorems/3596) and the corresponding singular cohomology formula. [example: Euler Characteristic of the Circle] The de Rham cohomology of $S^1$ is $H^0\cong\mathbb R$, $H^1\cong\mathbb R$, and $H^k=0$ for $k\geq 2$, so the Betti numbers are $b_0=b_1=1$ and $b_k=0$ otherwise. The Euler characteristic, $\chi=\sum_k(-1)^kb_k$, is \begin{align*} \chi(S^1)=1-1=0. \end{align*} The two non-zero Betti numbers cancel because $b_0$ counts components (one) and $b_1$ counts independent loops (one), and they enter the sum with opposite signs. The cancellation is not a coincidence: every closed odd-dimensional manifold has $\chi=0$, by Poincaré duality which pairs $H^k$ with $H^{n-k}$ of equal dimension, making each $b_k(-1)^k$ cancel against $b_{n-k}(-1)^{n-k}$ when $n$ is odd. The previous example also exhibited the concrete generator of $H^1(S^1)$: the normalised form $\alpha=\frac{1}{2\pi}(-y\,dx+x\,dy)|_{S^1}$ has period $1$ around the fundamental cycle, hence $[\alpha]\neq 0$ in the $1$-dimensional space $H^1_{\rm dR}(S^1)$, so $[\alpha]$ spans it. The negative term in $\chi(S^1)=b_0-b_1$ is geometrically the class of this loop — Euler characteristic counts the alternating sum of cells of each dimension, and the loop in $S^1$ is the single $1$-cell that cancels the single $0$-cell. [/example] The chapter closes the circle begun with exterior algebra. Alternating multilinear forms led to differential forms; the [exterior derivative](/theorems/1525) produced a cochain complex; Stokes theorem made integration compatible with boundaries; [de Rham theorem](/theorems/3596) identifies the resulting cohomology with singular cohomology. From this point onward, computations with differential forms can be read as computations of topological invariants. With de Rham's theorem establishing the link between differential forms and topological invariants, we now apply this machinery to concrete geometric problems. The final chapter demonstrates how questions about global manifold structure reduce to explicit calculations with differential forms. # 11. Selected Applications The earlier chapters built the basic machine of differential forms: exterior differentiation, pullback, integration on oriented manifolds, [Stokes' theorem](/theorems/1530), the Poincare lemma, Mayer-Vietoris, and de Rham cohomology. This final chapter shows how that machine turns global geometric questions into integrals of differential forms. The applications here are selective rather than exhaustive: degree detects how a sphere wraps around itself, Gauss-Bonnet turns curvature into topology, and Hodge theory explains why Riemannian geometry can choose canonical representatives of cohomology classes. The connecting theme is that integration over top-dimensional cycles pairs differential forms with topology. A closed form may carry more information than a local formula suggests, because its integral over a cycle can be unchanged under deformation. The degree, the Euler characteristic, and [harmonic representatives](/theorems/2747) are three ways this principle appears in geometry. ## Degree of Smooth Maps The first problem is to measure the global winding of a smooth map $f:S^n \to S^n$. Pointwise data such as the derivative can change under a homotopy, but the total signed number of times the domain covers the target should remain fixed. Differential forms give a compact definition: compare the integral of a volume form on the target with the integral of its pullback to the domain. [definition: Degree of a Smooth Map Between Spheres] Let $S^n$ carry its standard orientation. For a smooth map $f:S^n \to S^n$ and a volume form $\omega \in \Omega^n(S^n)$ with $\int_{S^n}\omega \ne 0$, the degree of $f$ is \begin{align*} \deg(f) = \frac{\int_{S^n} f^*\omega}{\int_{S^n} \omega}. \end{align*} [/definition] The definition uses a choice of $\omega$, but the answer is forced by the one-dimensionality of $H^n_{\mathrm{dR}}(S^n)$. The pullback $f^*$ acts on top cohomology by multiplication by a scalar, and the displayed ratio computes that scalar. [quotetheorem:3599] The regular-value formula is useful because it turns a global integral into local orientation data. The condition that $y$ is regular is essential: at a critical value the preimage need not be a finite collection of points, and the derivative may not decide an orientation sign. The sphere hypothesis keeps the target oriented and gives $H^n_{\mathrm{dR}}(S^n)\cong \mathbb R$, so pullback on top cohomology is multiplication by a single scalar. For a proper smooth map between compact connected oriented $n$-manifolds the same construction defines a degree; without compactness or properness, preimages can escape to infinity and the signed count need not be stable. [example: Antipodal Map on the Sphere] Let $A:S^n\to S^n$ be the antipodal map $A(x)=-x$, with $S^n\subset\mathbb R^{n+1}$ oriented by the outward-normal convention: $(v_1,\dots,v_n)$ is positive in $T_pS^n$ iff $(p,v_1,\dots,v_n)$ is positive in $\mathbb R^{n+1}$. The ambient derivative is $-I_{n+1}$, with determinant $(-1)^{n+1}$. [claim] $\deg(A)=(-1)^{n+1}$. [/claim] [proof] Every $y\in S^n$ is a regular value: $A^{-1}(y)=\{-y\}$ and $DA_{-y}:v\mapsto-v$ is an isomorphism. To compute the local degree sign, take a positive basis $(v_1,\dots,v_n)$ of $T_{-y}S^n$, so $(-y,v_1,\dots,v_n)$ is positive in $\mathbb R^{n+1}$. The image basis at $y$ is $(-v_1,\dots,-v_n)$, and prepending the outward normal $y$ gives \begin{align*} (y,-v_1,\dots,-v_n)=(-1)^{n+1}(-y,v_1,\dots,v_n), \end{align*} so the sign is $(-1)^{n+1}$. Summing over the unique preimage, $\deg(A)=(-1)^{n+1}$. [/proof] This is the cleanest example of degree theory: a single isolated preimage at a regular value carries one sign that determines the whole degree. The geometric meaning is that the antipodal map preserves orientation on odd-dimensional spheres ($S^1$, $S^3$, $\dots$) and reverses it on even-dimensional ones ($S^2$, $S^4$, $\dots$) — the sign is just $\det(-I_{n+1})$. Two consequences worth flagging: $A$ is homotopic to the identity iff $n$ is odd, and the hairy ball theorem (no continuous tangent vector field on $S^{2k}$) ultimately rests on this sign computation. [/example] The antipodal computation has a sharp topological consequence: in even dimensions $n=2k$, $\deg(A)=-1\ne 1=\deg(\operatorname{id})$, so the antipodal map is not homotopic to the identity on $S^{2k}$. This is the precise obstruction that powers the [hairy ball theorem](/theorems/2248) and the Borsuk-Ulam theorem in even dimensions. In odd dimensions, by contrast, $\deg(A)=+1$, consistent with the fact that $S^{2k-1}$ admits nowhere-vanishing tangent vector fields whose flows interpolate between the identity and the antipodal map. The degree is also a faithful invariant in lower dimensions. For $n=1$, the degree recovers the classical [winding number](/page/Winding%20Number) of a map $S^1\to S^1$: the map $z\mapsto z^m$ wraps the circle $m$ times around itself, so its degree is $m$ for $m\ge 1$ and $-|m|$ for $m\le -1$. Two circle maps are homotopic if and only if they have the same degree, so $[S^1,S^1]\cong\mathbb Z$ via the degree. The Hopf degree theorem extends this to $[S^n,S^n]\cong\mathbb Z$ for every $n\ge 1$. With this language in place, the [fundamental theorem of algebra](/theorems/347) becomes a statement about the degree of a polynomial's radial boundary map. [quotetheorem:347] The proof is a model application of de Rham cohomology: the obstruction to extending a circle map over the disk is detected by integrating a closed $1$-form around the boundary. The argument depends on three structural inputs that are worth isolating. First, the closed disk is compact with boundary $S^1$, so [Stokes' theorem](/theorems/1530) applies and forces the boundary integral of an exact form to vanish; on a non-compact replacement of the disk this conclusion would fail. Second, the disk is contractible, so the Poincare lemma converts every closed form into an exact form there; on a domain with a hole, the closed pullback could carry a nonzero period. Third, the boundary map $F_R$ would have to be smooth on the entire disk for the pullback to be defined, which is why the hypothetical absence of a zero matters: a zero of $p$ on the disk is precisely the singularity that would prevent the extension. The same circle of ideas generalises far beyond polynomials. Replacing $S^1\to S^1$ with $S^n\to S^n$ and the disk with $\overline{B}(0,1)\subset\mathbb R^{n+1}$ gives Brouwer's fixed-point theorem and the no-retraction principle, since a continuous retraction of the ball onto its boundary would contradict the degree of the identity map. The same degree machinery underlies the Hopf degree theorem identifying $[S^n,S^n]$ with $\mathbb Z$, the topological proof of the [hairy ball theorem](/theorems/2248) on $S^{2k}$, and the more general Poincare-Hopf theorem relating vector field indices to the Euler characteristic. The [fundamental theorem of algebra](/theorems/347) is the one-dimensional, polynomial-flavoured instance of this pattern. [illustration:forms-fta-winding-obstruction] [example: The Form Detecting Winding Number] On $U=\mathbb R^2\setminus\{0\}$, take $\alpha=(x\,dy-y\,dx)/(x^2+y^2)=P\,dx+Q\,dy$ (note: opposite sign convention from the angle form of earlier chapters; some authors call this $d\theta$ as well). The closedness condition $\partial_xQ=\partial_yP$ holds: both equal $(y^2-x^2)/(x^2+y^2)^2$ by direct quotient-rule calculation. Along the unit circle $\gamma(\theta)=(\cos\theta,\sin\theta)$, $\theta\in[0,2\pi]$, pulling back gives $\gamma^*(x\,dy-y\,dx)=(\cos^2\theta+\sin^2\theta)\,d\theta=d\theta$ and $\gamma^*(x^2+y^2)=1$, so $\gamma^*\alpha=d\theta$ and \begin{align*} \int_\gamma\alpha=\int_0^{2\pi}d\theta=2\pi. \end{align*} A non-zero period around a closed loop is impossible for an exact $1$-form (the [fundamental theorem of calculus](/theorems/632) would give zero), so $\alpha$ is not exact. Integration of $\alpha$ over a smooth closed loop computes $2\pi$ times the *winding number* of the loop around the origin — this is essentially the definition of winding number, packaged as a topological invariant by Stokes' theorem. The class $[\alpha]/(2\pi)\in H^1_{\rm dR}(U)$ is the canonical integral generator: an element whose periods on all integer-coefficient cycles are integers. This integrality is the de Rham side of the comparison theorem $H^1_{\rm dR}(U;\mathbb R)\cong H^1(U;\mathbb R)$. [/example] ## Gauss-Bonnet and Curvature as a Cohomology Class The next problem is more geometric: can the total curvature of a surface remember its topology? Gaussian curvature is defined from a Riemannian metric and may vary from point to point, while the Euler characteristic is a homotopy invariant. Gauss-Bonnet says that the integral of curvature is exactly the bridge between them. [definition: Euler Characteristic of a Closed Surface] Let $M$ be a compact connected oriented smooth surface. Its Euler characteristic is \begin{align*} \chi(M)=b_0(M)-b_1(M)+b_2(M), \qquad b_k(M)=\dim H^k_{\mathrm{dR}}(M). \end{align*} [/definition] For a connected closed oriented surface, $b_0=b_2=1$, so the topology is controlled by $b_1$. If $M$ has genus $g$, then $b_1=2g$ and $\chi(M)=2-2g$. [definition: Curvature Two-Form of an Oriented Surface] Let $(M,g)$ be an oriented Riemannian surface. Let $K:M\to\mathbb R$ be its Gaussian curvature and let $dA$ be its Riemannian area form. The curvature two-form is \begin{align*} \Omega_K = K\,dA \in \Omega^2(M). \end{align*} [/definition] Since every $2$-form on a surface is closed, $\Omega_K$ defines a de Rham cohomology class. The remarkable fact is that this class is not arbitrary: after division by $2\pi$, it is the Euler class of the tangent bundle. [quotetheorem:3600] This proof places Gauss-Bonnet inside the same framework as the earlier chapters: local differential forms are integrated, [Stokes' theorem](/theorems/1530) moves the information to boundaries, and cancellation leaves a topological invariant. Each hypothesis pulls its weight. Orientability gives a global area form $dA$ and a consistent sign on each triangle, so the boundary integrals cancel between adjacent faces rather than doubling up; on a non-orientable surface such as $\mathbb RP^2$ or the Klein bottle, the global area form fails to exist and the statement must be reformulated through the Euler class of the orthonormal frame bundle, with $K\,dA$ promoted to a twisted-density. Closedness of $M$ is equally essential: any boundary $\partial M$ would contribute an extra geodesic-curvature term $\int_{\partial M}\kappa_g\,ds$, recovering the boundary version $\int_M K\,dA + \int_{\partial M}\kappa_g\,ds = 2\pi\chi(M)$. Smoothness lets the connection $1$-forms be differentiated, while finiteness of the triangulation lets the vertex contributions add to a finite combinatorial number. On a non-compact surface, the integral $\int_M K\,dA$ need not converge, and even when it does, no triangulation identity links it to a Betti-number sum. [illustration:forms-triangulated-surface-curvature] [explanation: Cohomological Form of Gauss-Bonnet] The curvature form gives a class \begin{align*} \left[\frac{\Omega_K}{2\pi}\right]\in H^2_{\mathrm{dR}}(M). \end{align*} Gauss-Bonnet says that pairing this class with the fundamental class $[M]$ gives the Euler characteristic: \begin{align*} \int_M \frac{\Omega_K}{2\pi}=\chi(M). \end{align*} In this form, the metric dependence has disappeared from the final number. Different Riemannian metrics change the representative $\Omega_K$, but not its cohomology class when paired with $[M]$. [/explanation] Two examples make the metric-independence concrete. The first inspects a single closed surface — the round two-sphere — under metrics of different sizes and verifies that the total curvature is unchanged. The second reads Gauss-Bonnet in the opposite direction: fixing the topology of a surface forces the total curvature to take a specific value, regardless of how the metric is deformed. [example: Round Two-Sphere] Let $S^2_R\subset\mathbb R^3$ be the sphere of radius $R$ with the outward orientation and induced round metric. The outward unit normal is $N(p)=p/R$, so $dN_p(v)=v/R$ for tangent $v$. In any orthonormal tangent frame, $dN_p$ has matrix $(1/R)I$, hence Gaussian curvature \begin{align*} K=\det(dN_p)=1/R^2. \end{align*} The area is computed in the standard spherical parametrisation $\Phi(\varphi,\theta)=(R\sin\varphi\cos\theta,R\sin\varphi\sin\theta,R\cos\varphi)$. The cross product $\Phi_\varphi\times\Phi_\theta$ has magnitude $R^2\sin\varphi$ (and points outward, confirming the orientation), so \begin{align*} \operatorname{Area}(S^2_R)=\int_0^{2\pi}\int_0^\pi R^2\sin\varphi\,d\varphi\,d\theta=4\pi R^2. \end{align*} Combining, \begin{align*} \int_{S^2_R}K\,dA=\frac{1}{R^2}\cdot 4\pi R^2=4\pi. \end{align*} By the [Gauss–Bonnet theorem](/theorems/3640) for closed oriented surfaces, $\int K\,dA=2\pi\chi$, so $\chi(S^2)=2$. The radius cancels: $K\sim 1/R^2$ shrinks as the sphere grows, but the area grows like $R^2$, and the product stays $4\pi$ — a topological invariant. This radius-independence is the entire content of Gauss–Bonnet: total curvature is a topological functional of the surface, not a metric one. Stretch the sphere, dent it, give it a different metric; the total curvature is still $4\pi$ so long as the topology stays $S^2$. [/example] For surfaces of higher genus, the same equation runs in the opposite direction: the topology is fixed first, and Gauss-Bonnet then constrains the total curvature. This is the form in which the theorem becomes a rigidity statement, since no smooth metric on a genus-$g$ surface can violate the prescribed total. The torus case is especially striking, because it forces any embedded torus in $\mathbb R^3$ to balance positive and negative curvature exactly. [example: Surfaces of Genus g] Let $M_g$ be a compact connected oriented smooth surface of genus $g$ with any Riemannian metric. The classical Euler-characteristic formula gives $\chi(M_g)=2-2g$, and the [Gauss–Bonnet theorem](/theorems/3640) for closed oriented surfaces gives \begin{align*} \int_{M_g}K\,dA=2\pi\chi(M_g)=2\pi(2-2g)=4\pi(1-g). \end{align*} The total curvature is determined entirely by genus: $4\pi$ for the sphere ($g=0$), $0$ for the torus ($g=1$), negative for every $g\geq 2$. The torus case is especially striking: every embedded torus in $\mathbb R^3$ has regions of positive curvature (the outer ring) and negative curvature (the inner ring), and they must exactly cancel — a metric constraint forced by topology. The flat quotient torus realises $\int K\,dA=0$ trivially, with $K\equiv 0$ pointwise. For $g\geq 2$, the total curvature is negative, so the metric must be "mostly hyperbolic" on average. The uniformisation theorem makes this precise: every surface of genus $g\geq 2$ admits a metric of constant negative curvature $K=-1$, and the area is then exactly $4\pi(g-1)$ by Gauss–Bonnet. This identifies surfaces of high genus as quotients of the hyperbolic plane. [/example] ## Harmonic Representatives and the Hodge Decomposition The last problem asks for a preferred representative of a de Rham cohomology class. A class is an equivalence class of closed forms modulo exact forms, so it usually has many representatives. Once a Riemannian metric is chosen on a compact manifold, Hodge theory selects the representative satisfying an elliptic differential equation. [example: Oscillating Representatives on the Circle] On $S^1=\mathbb R/2\pi\mathbb Z$ with metric $g=d\theta^2$ and orientation given by $d\theta$, take $\alpha_0=d\theta$ and, for $N\in\mathbb N$, $\alpha_N=d\theta+d(\sin N\theta)$. The function $\sin N\theta$ is well defined on $S^1$ because $\sin N(\theta+2\pi)=\sin N\theta$. Expanding $\alpha_N=(1+N\cos N\theta)\,d\theta$, the difference $\alpha_N-\alpha_0=d(\sin N\theta)$ is exact, so $[\alpha_N]=[\alpha_0]$ in $H^1_{\rm dR}(S^1)$. Both are closed: $d\alpha_0=d^2\theta=0$, and $d\alpha_N=-N^2\sin N\theta\,d\theta\wedge d\theta=0$. But the pointwise norm $|\alpha_N|_g=|1+N\cos N\theta|$ oscillates between $|1-N|$ and $1+N$. Taking $N$ large makes the pointwise norm arbitrarily large somewhere, while remaining in the same cohomology class. A cohomology class does not select a preferred representative. This is exactly the problem that Hodge theory solves: among all closed representatives of a given class, exactly one minimises the $L^2$ norm — the *harmonic* representative. On $S^1$ with the standard metric, that minimiser is $\alpha_0=d\theta$ itself, and all the oscillating $\alpha_N$ are non-harmonic perturbations of it. This selection principle becomes especially powerful on higher-dimensional manifolds where the harmonic representative often has additional symmetry or invariance properties the generic representative does not. [/example] This example shows why a quotient description does not by itself choose a canonical form. At each point $p\in M$, a Riemannian metric induces a pointwise inner product $(\cdot,\cdot)_g$ on $\Lambda^kT_p^*M$; the Hodge star is the operator built from this pointwise inner product and the volume form. [definition: Hodge Star] Let $(M,g)$ be an oriented Riemannian $n$-manifold with Riemannian volume form $dV_g$, and let $(\cdot,\cdot)_g$ denote the fiberwise inner product induced by $g$ on each exterior power $\Lambda^kT_p^*M$. The Hodge star is the bundle map $*:\Lambda^kT^*M\to \Lambda^{n-k}T^*M$, extended to global sections as $*:\Omega^k(M)\to \Omega^{n-k}(M)$, determined pointwise by \begin{align*} \alpha_p\wedge (*\beta)_p = (\alpha_p,\beta_p)_g\,(dV_g)_p \end{align*} for all $\alpha_p,\beta_p\in \Lambda^kT_p^*M$ and all $p\in M$. [/definition] The Hodge star turns the metric into an $L^2$ inner product on forms. For compact $M$, define \begin{align*} (\alpha,\beta)_{L^2}=\int_M \alpha\wedge *\beta. \end{align*} [definition: Codifferential] Let $(M,g)$ be a compact oriented Riemannian manifold. The codifferential $d^*:\Omega^k(M)\to\Omega^{k-1}(M)$ is the formal adjoint of $d$ with respect to the $L^2$ inner product: \begin{align*} (d\alpha,\beta)_{L^2}=(\alpha,d^*\beta)_{L^2} \end{align*} for all compatible degrees. [/definition] Closed forms solve $d\alpha=0$. Hodge theory adds the adjoint condition $d^*\alpha=0$, producing a form that is both closed and co-closed. [definition: Hodge Laplacian] Let $(M,g)$ be a compact oriented Riemannian manifold. The Hodge Laplacian on $k$-forms is \begin{align*} \Delta = dd^*+d^*d:\Omega^k(M)\to\Omega^k(M). \end{align*} [/definition] The Hodge Laplacian is a second-order elliptic operator that reduces, on functions in Euclidean space, to minus the ordinary Laplacian. Its symmetry under the $L^2$ inner product follows from the adjoint relation between $d$ and $d^*$, and this symmetry is the reason its kernel singles out a distinguished subspace of forms. The forms in this kernel are the geometric objects that combine both halves of the de Rham complex into a single equation. [definition: Harmonic Form] Let $(M,g)$ be a compact oriented Riemannian manifold. A $k$-form $\alpha\in\Omega^k(M)$ is harmonic if \begin{align*} \Delta\alpha=0. \end{align*} The [vector space](/page/Vector%20Space) of harmonic $k$-forms is denoted $\mathcal H^k(M)$. [/definition] On a compact manifold without boundary, the identity \begin{align*} (\Delta\alpha,\alpha)_{L^2}=\|d\alpha\|_{L^2}^2+\|d^*\alpha\|_{L^2}^2 \end{align*} shows that a harmonic form is exactly a form satisfying $d\alpha=0$ and $d^*\alpha=0$. [quotetheorem:2745] The proof uses elliptic regularity and Fredholm theory for the Hodge Laplacian, which go beyond the differential-form tools developed in this course. The hypotheses are doing real work. Compactness gives Fredholm behaviour and finite-dimensional harmonic spaces; on non-compact manifolds, harmonic forms can fail to represent de Rham cohomology in this clean way. The metric is also essential: without it there is no Hodge star, no codifferential, and no Laplacian selecting a preferred representative. Orientability is part of the ordinary-form presentation used here, since it supplies a global volume form for the Hodge star and the integral pairing; non-orientable manifolds require density or twisted-coefficient versions. In these notes, the theorem is used as a structural statement explaining how analysis refines de Rham cohomology. [example: Harmonic Forms on the Flat Torus] On $T^2=\mathbb R^2/(2\pi\mathbb Z)^2$ with flat metric $g=d\theta^2+d\phi^2$, oriented by $dA=d\theta\wedge d\phi$, the coframe $(d\theta,d\phi)$ is orthonormal. [claim] \begin{align*} \mathcal H^0(T^2)=\operatorname{span}\{1\}, \qquad \mathcal H^1(T^2)=\operatorname{span}\{d\theta,d\phi\}, \qquad \mathcal H^2(T^2)=\operatorname{span}\{d\theta\wedge d\phi\}. \end{align*} [/claim] [proof] The Hodge star on this oriented Riemannian $2$-manifold acts on the orthonormal basis as $*1=dA$, $*d\theta=d\phi$, $*d\phi=-d\theta$, $*dA=1$. The codifferential on a $k$-form satisfies $d^*=-*d*$ in dimension $2$. **Listed forms are harmonic.** Each of $1$, $d\theta$, $d\phi$, $dA$ satisfies $d(\cdot)=0$ (the first two by $d^2=0$, the third similarly, the last because $\Omega^3(T^2)=0$). Each satisfies $d^*(\cdot)=0$: \begin{align*} d^*d\theta=-*d*d\theta=-*d(d\phi)=0, \quad d^*d\phi=-*d*d\phi=-*d(-d\theta)=0, \quad d^*dA=-*d*dA=-*d1=0. \end{align*} Hence $\Delta=dd^*+d^*d$ kills each. **No other harmonic forms.** By the [Hodge decomposition theorem](/theorems/2745), $\mathcal H^k(T^2)\cong H^k_{\rm dR}(T^2)$, and the Mayer–Vietoris computation in Chapter 9 gave Betti numbers $(b_0,b_1,b_2)=(1,2,1)$. The listed spans have matching dimensions $1, 2, 1$ (linear independence is immediate by evaluating on $\partial_\theta,\partial_\phi$), so the inclusions $\operatorname{span}\subseteq\mathcal H^k$ are equalities. [/proof] The flat torus has the perfect harmonic theory: every cohomology class has a *constant-coefficient* representative in this metric, and the harmonic space is built from the translation-invariant forms descending from $\mathbb R^2$. The two independent harmonic $1$-forms $d\theta$, $d\phi$ are the de Rham incarnations of the two circle factors in $S^1\times S^1$, and their wedge generates $H^2$. This neat picture breaks on curved surfaces, as the next example shows. [/example] The flat torus is the cleanest case because translation-invariance picks out the [harmonic representatives](/theorems/2747) without any analysis: the constant-coefficient forms automatically satisfy both $d\alpha=0$ and $d^*\alpha=0$. Once the geometry is no longer flat, however, the harmonic equation becomes a genuine elliptic problem and curvature enters the picture. The sphere is the simplest such case: its constant positive curvature changes the answer in degree $1$ and illustrates how Hodge theory interacts with Riemannian geometry rather than merely with topology. [example: Harmonic Representatives on the Sphere] Equip $S^2$ with the round metric ($K\equiv 1$) and the induced orientation. The cohomology is $H^0\cong\mathbb R$, $H^1=0$, $H^2\cong\mathbb R$. [claim] \begin{align*} \mathcal H^0(S^2)=\operatorname{span}\{1\}, \qquad \mathcal H^1(S^2)=\{0\}, \qquad \mathcal H^2(S^2)=\operatorname{span}\{dA\}. \end{align*} [/claim] [proof] **Degree $0$.** A harmonic $f$ on a closed manifold satisfies $0=(\Delta f,f)_{L^2}=\|df\|^2+\|d^*f\|^2$, with $d^*f=0$ automatic since $\Omega^{-1}=0$. So $df=0$, hence $f$ is constant on the connected $S^2$. Constants are evidently harmonic. **Degree $1$.** This is where the geometry matters. The Bochner formula on a closed Riemannian surface reads \begin{align*} \int_{S^2}|\nabla\alpha|^2\,dA+\int_{S^2}K|\alpha|^2\,dA=\int_{S^2}(|d\alpha|^2+|d^*\alpha|^2)\,dA. \end{align*} A harmonic $\alpha$ kills the right side. Substituting $K=1$ gives $\int|\nabla\alpha|^2+\int|\alpha|^2=0$, and since both integrands are nonnegative, $|\alpha|^2\equiv 0$, hence $\alpha=0$. **Degree $2$.** A $2$-form is $h\,dA$ for a smooth function $h$ (since $\Lambda^2T^*S^2$ is $1$-dimensional). Closedness is automatic ($d(h\,dA)\in\Omega^3=0$), and the codifferential condition reads $d^*(h\,dA)=-*d*(h\,dA)=-*dh=0$, forcing $dh=0$, hence $h$ is constant. The constants times $dA$ are visibly harmonic. [/proof] The contrast with the flat torus is striking: positive curvature ($K=1$) eliminates harmonic $1$-forms through the Bochner identity, while flatness ($K=0$) preserves them. This is the prototype of the *vanishing theorem* phenomenon — geometric positivity kills cohomology in certain degrees. The same idea, in many disguises, drives Kodaira vanishing for complex manifolds, Bochner–Kodaira–Nakano for hermitian vector bundles, and Witten's analytic proof of the Morse inequalities. Positive curvature is a topological constraint, mediated by analysis. [/example] Hodge theory closes the circle of the course. De Rham cohomology began as a quotient of closed forms by exact forms; with a metric, the same cohomology can be represented by solutions of the elliptic equation $\Delta\alpha=0$. This is the point where topology, geometry, and analysis meet. ## References