This course develops the calculus of differential forms as a unified language for multivariable analysis, geometry, and integration. It begins by building the algebraic framework needed to talk about alternating multilinear maps and exterior powers, then turns to differential forms on open subsets of $\mathbb{R}^n$. From there, the central operators and ideas of the subject appear: the [exterior derivative](/theorems/1525), pullbacks, orientation, and integration of forms. Together, these tools provide a coordinate-flexible way to express differentiation and integration in several variables. The later chapters show why the subject is so powerful. Stokes’ theorem ties the [exterior derivative](/theorems/1525) to boundary integration and serves as the organizing principle for the entire theory. Closed and exact forms, along with Poincaré’s lemma, explain when local differential information comes from potentials and when topological obstructions appear. The course concludes with applications that reconnect the abstract formalism to classical vector calculus, showing how familiar results about gradients, curls, divergence, and line and surface integrals are natural consequences of the language of differential forms. # Introduction This course takes the line integrals, surface integrals, change-of-variables formulas, and integral theorems of multivariable calculus and rewrites them in one language. The central objects are differential forms: algebraic devices that know how many tangent directions they should eat, how orientation affects sign, and how integration changes under parametrisation. We begin on open subsets of $\mathbb R^n$, where coordinates are available for computation, and build toward a coordinate-free formulation that later extends to smooth manifolds. The guiding theme is that calculus has two compatible operations: differentiating inside a region and restricting to the boundary of that region. Exterior calculus is the formalism in which these operations become the [exterior derivative](/theorems/1525) $d$, the boundary operator $\partial$, and the identity \begin{align*} \int_M d\omega = \int_{\partial M} \omega. \end{align*} The point of the course is not to discard familiar vector calculus, but to explain why gradient, curl, divergence, [Green's theorem](/theorems/3612), Gauss' theorem, and [Stokes' theorem](/theorems/1530) are shadows of the same construction. ## Why Vector Calculus Needs A Common Language The same integral theorem appears in several disguises, so what is the single object being integrated in each case? In $\mathbb R^3$, a vector field may be used in a line integral, a flux integral, or a divergence integral, but these are not the same operation. A line integral measures a vector field along one tangent direction, while a flux integral measures oriented area, and a volume integral measures oriented volume. Differential forms separate these roles by degree. A $0$-form is a function, a $1$-form is integrated along oriented curves, a $2$-form is integrated over oriented surfaces, and a $3$-form is integrated over oriented regions in $\mathbb R^3$. The degree tells us the dimension of the object of integration. [example: Vector Calculus Dictionary In R3] Let $U\subset \mathbb R^3$ be open, write $x=(x_1,x_2,x_3)$, and let $F=(P,Q,R):U\to\mathbb R^3$ be smooth. The same coordinate functions give different forms according to the dimension of the object being integrated: \begin{align*} \omega_1 &= P\,dx_1 + Q\,dx_2 + R\,dx_3,\\ \omega_2 &= P\,dx_2\wedge dx_3 + Q\,dx_3\wedge dx_1 + R\,dx_1\wedge dx_2,\\ \omega_3 &= P\,dx_1\wedge dx_2\wedge dx_3. \end{align*} For an oriented curve $\gamma:[a,b]\to U$, write $\gamma(t)=(\gamma_1(t),\gamma_2(t),\gamma_3(t))$. Since $\gamma^*dx_i=d(\gamma_i)=\dot\gamma_i(t)\,dt$, we get \begin{align*} \gamma^*\omega_1 &= P(\gamma(t))\,\gamma^*dx_1 + Q(\gamma(t))\,\gamma^*dx_2 + R(\gamma(t))\,\gamma^*dx_3\\ &= P(\gamma(t))\dot\gamma_1(t)\,dt + Q(\gamma(t))\dot\gamma_2(t)\,dt + R(\gamma(t))\dot\gamma_3(t)\,dt\\ &= \bigl(P(\gamma(t))\dot\gamma_1(t) + Q(\gamma(t))\dot\gamma_2(t) + R(\gamma(t))\dot\gamma_3(t)\bigr)\,dt\\ &= F(\gamma(t))\cdot \dot\gamma(t)\,dt. \end{align*} Thus \begin{align*} \int_\gamma \omega_1 = \int_a^b F(\gamma(t))\cdot \dot\gamma(t)\,dt, \end{align*} so $\omega_1$ is the form corresponding to the usual work integral of $F$ along an oriented curve. For an oriented parametrised surface $\Phi:W\subset\mathbb R^2\to U$, write $\Phi(u,v)=(X(u,v),Y(u,v),Z(u,v))$. Then \begin{align*} \Phi^*dx_1 &= X_u\,du+X_v\,dv,\\ \Phi^*dx_2 &= Y_u\,du+Y_v\,dv,\\ \Phi^*dx_3 &= Z_u\,du+Z_v\,dv. \end{align*} Using $du\wedge du=0$, $dv\wedge dv=0$, and $dv\wedge du=-du\wedge dv$, \begin{align*} \Phi^*(dx_2\wedge dx_3) &= (Y_u\,du+Y_v\,dv)\wedge (Z_u\,du+Z_v\,dv)\\ &= Y_uZ_u\,du\wedge du+Y_uZ_v\,du\wedge dv +Y_vZ_u\,dv\wedge du+Y_vZ_v\,dv\wedge dv\\ &= (Y_uZ_v-Y_vZ_u)\,du\wedge dv,\\ \Phi^*(dx_3\wedge dx_1) &= (Z_u\,du+Z_v\,dv)\wedge (X_u\,du+X_v\,dv)\\ &= (Z_uX_v-Z_vX_u)\,du\wedge dv,\\ \Phi^*(dx_1\wedge dx_2) &= (X_u\,du+X_v\,dv)\wedge (Y_u\,du+Y_v\,dv)\\ &= (X_uY_v-X_vY_u)\,du\wedge dv. \end{align*} Therefore \begin{align*} \Phi^*\omega_2 &= \Bigl[ P(\Phi)(Y_uZ_v-Y_vZ_u) +Q(\Phi)(Z_uX_v-Z_vX_u) +R(\Phi)(X_uY_v-X_vY_u) \Bigr]\,du\wedge dv\\ &= F(\Phi(u,v))\cdot \bigl(Y_uZ_v-Y_vZ_u,\; Z_uX_v-Z_vX_u,\; X_uY_v-X_vY_u\bigr)\,du\wedge dv\\ &= F(\Phi(u,v))\cdot(\partial_u\Phi\times \partial_v\Phi)\,du\wedge dv. \end{align*} Thus $\omega_2$ is the form corresponding to the flux of $F$ through the oriented tangent parallelogram spanned in order by $\partial_u\Phi$ and $\partial_v\Phi$. Finally, if a volume parametrisation $\Psi:A\subset\mathbb R^3\to U$ is written $\Psi(s,t,r)=(X,Y,Z)$, then \begin{align*} \Psi^*\omega_3 &= P(\Psi)\,\Psi^*(dx_1\wedge dx_2\wedge dx_3)\\ &= P(\Psi)\,dX\wedge dY\wedge dZ\\ &= P(\Psi)\,\det D\Psi\,ds\wedge dt\wedge dr. \end{align*} So $\omega_3$ is the top-degree form corresponding to an oriented volume integral with scalar density $P$. [/example] This dictionary is not merely a notation change. It tells us which algebraic object is being pulled back to a parameter domain before integration, and it keeps track of the sign changes caused by reversing orientation. A concrete failure of naive notation already appears for curves. If $\gamma:[0,1]\to U$ is traversed backwards by $\tilde\gamma(t)=\gamma(1-t)$, then the pullback calculation gives \begin{align*} \int_{\tilde\gamma}\omega = -\int_\gamma \omega. \end{align*} Writing only "$\int_C F\cdot dx$" hides the fact that the sign is attached to the oriented parametrisation, not just to the underlying set of points. Differential forms force this bookkeeping: the $1$-form is pulled back to the parameter interval, where the derivative of $1-t$ supplies the minus sign. ## The Algebra Of Oriented Measurement What algebraic object can record length, oriented area, and oriented volume without choosing a preferred coordinate system? The answer is an alternating multilinear form. Alternation is the sign rule that makes area change sign when two ordered tangent directions are swapped, and it is the reason repeated tangent directions contribute zero oriented area. [definition: Alternating k-linear Form] Let $V$ be a real [vector space](/page/Vector%20Space) and let $k\in\mathbb N$. An alternating $k$-linear form on $V$ is a map \begin{align*} \alpha: V^k \to \mathbb R \end{align*} that is linear in each argument and satisfies \begin{align*} \alpha(v_1,\dots,v_i,\dots,v_j,\dots,v_k) = -\alpha(v_1,\dots,v_j,\dots,v_i,\dots,v_k) \end{align*} for all $v_1,\dots,v_k\in V$ and all $i\neq j$. [/definition] The [vector space](/page/Vector%20Space) of alternating $k$-linear forms on $V$ is denoted $\Lambda^k(V^*)$. We use the convention $\Lambda^0(V^*)=\mathbb R$. For $k=1$ this is the ordinary [dual space](/page/Dual%20Space) $V^*$; for $k=2$ it consists of signed area-measuring functions; for $k=\dim V$ it consists of signed volume forms. A fixed alternating form measures tangent data in one vector space, but calculus needs measurements attached to every point of a region. Without a smooth dependence on the point, there is no way to differentiate coefficients, pull the measurement through a parametrisation, or integrate it consistently over a moving surface. The definition therefore treats a differential form as a pointwise choice of alternating covector that varies smoothly across the open set. [definition: Differential k-form] Let $U\subset\mathbb R^n$ be open and let $0\le k\le n$. A smooth differential $k$-form on $U$ is a smooth map \begin{align*} \omega: U \to \Lambda^k((\mathbb R^n)^*). \end{align*} The space of smooth differential $k$-forms on $U$ is denoted $\Omega^k(U)$. [/definition] A differential form is therefore a smoothly varying rule for measuring oriented tangent $k$-parallelepipeds. The course first develops the finite-dimensional algebra of $\Lambda^k(V^*)$, then allows the coefficients of those algebraic expressions to vary with $x\in U$. [quotetheorem:3561] The number of basis elements is $\binom{n}{k}$, because choosing an alternating coordinate measurement is the same as choosing which $k$ coordinate directions are measured, with the order then fixed up to sign. Thus $\Lambda^k(V^*)=\{0\}$ for $k>n$ is not just a computational shortcut: there are no nonzero oriented $k$-dimensional volume measurements inside an $n$-dimensional [vector space](/page/Vector%20Space). The theorem does depend on the prior choice of a basis of $V$ and its [dual basis](/theorems/414); it does not supply a canonical inner product on forms or a preferred basis of $\Lambda^k(V^*)$ for an abstract [vector space](/page/Vector%20Space). Later, once an inner product and orientation are chosen, the Hodge star will add extra structure by relating $k$-forms to $(n-k)$-forms. [citeproof:3561] The theorem is the computational entry point for the course. On an [open set](/page/Open%20Set) $U\subset\mathbb R^n$, every $k$-form has a unique coordinate expansion in the basis $dx_{i_1}\wedge\cdots\wedge dx_{i_k}$ with smooth coefficient functions. [example: Coordinate Expansion Of A Two-form] [claim]Let $U\subset\mathbb R^3$ be open and let $\omega\in\Omega^2(U)$. There are unique smooth functions $A,B,C:U\to\mathbb R$ such that \begin{align*} \omega = A\,dx_2\wedge dx_3 + B\,dx_3\wedge dx_1 + C\,dx_1\wedge dx_2. \end{align*}[/claim] [proof]Let $e_1,e_2,e_3$ be the standard basis of $\mathbb R^3$. For $x\in U$, define \begin{align*} A(x)&=\omega_x(e_2,e_3),& B(x)&=\omega_x(e_3,e_1),& C(x)&=\omega_x(e_1,e_2). \end{align*} These functions are smooth because $\omega:U\to\Lambda^2((\mathbb R^3)^*)$ is smooth and evaluation on fixed vectors is linear. Now take arbitrary tangent vectors \begin{align*} u&=u_1e_1+u_2e_2+u_3e_3,& v&=v_1e_1+v_2e_2+v_3e_3. \end{align*} By bilinearity and alternation, \begin{align*} \omega_x(u,v) &=\sum_{i=1}^3\sum_{j=1}^3 u_i v_j\,\omega_x(e_i,e_j)\\ &=u_1v_2\,\omega_x(e_1,e_2)+u_2v_1\,\omega_x(e_2,e_1) +u_1v_3\,\omega_x(e_1,e_3)+u_3v_1\,\omega_x(e_3,e_1)\\ &\quad +u_2v_3\,\omega_x(e_2,e_3)+u_3v_2\,\omega_x(e_3,e_2)\\ &=C(x)(u_1v_2-u_2v_1)+B(x)(u_3v_1-u_1v_3)+A(x)(u_2v_3-u_3v_2). \end{align*} On the other hand, \begin{align*} (dx_2\wedge dx_3)(u,v)&=u_2v_3-u_3v_2,\\ (dx_3\wedge dx_1)(u,v)&=u_3v_1-u_1v_3,\\ (dx_1\wedge dx_2)(u,v)&=u_1v_2-u_2v_1. \end{align*} Therefore \begin{align*} \omega_x(u,v) &=\bigl(A(x)\,dx_2\wedge dx_3 +B(x)\,dx_3\wedge dx_1 +C(x)\,dx_1\wedge dx_2\bigr)(u,v) \end{align*} for all $x\in U$ and all $u,v\in\mathbb R^3$, which proves the expansion. For uniqueness, suppose also \begin{align*} \omega=A'\,dx_2\wedge dx_3+B'\,dx_3\wedge dx_1+C'\,dx_1\wedge dx_2. \end{align*} Evaluating at $(e_2,e_3)$ gives \begin{align*} \omega_x(e_2,e_3) &=A'(x)(dx_2\wedge dx_3)(e_2,e_3) +B'(x)(dx_3\wedge dx_1)(e_2,e_3) +C'(x)(dx_1\wedge dx_2)(e_2,e_3)\\ &=A'(x)\cdot 1+B'(x)\cdot 0+C'(x)\cdot 0\\ &=A'(x), \end{align*} so $A'=A$. The same calculation on $(e_3,e_1)$ and $(e_1,e_2)$ gives $B'=B$ and $C'=C$.[/proof] Thus a $2$-form in $\mathbb R^3$ has exactly three coordinate coefficients, each measuring oriented area in one coordinate plane. [/example] ## Differentiation That Knows The Degree How should a derivative turn integrands over $k$-dimensional objects into integrands over $(k+1)$-dimensional objects? The [exterior derivative](/theorems/1525) $d$ is designed to do exactly this. It sends $k$-forms to $(k+1)$-forms and packages the gradient, curl, and divergence into a single operator. [definition: Exterior Derivative In Coordinates] Let $U\subset\mathbb R^n$ be open. The [exterior derivative](/theorems/1525) in degree $k$ is the map \begin{align*} d: \Omega^k(U) \to \Omega^{k+1}(U) \end{align*} defined as follows. For a smooth $k$-form \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*} set \begin{align*} d\omega = \sum_{1\le i_1<\cdots<i_k\le n}\sum_{j=1}^{n} \partial_j a_{i_1\cdots i_k}\,dx_j\wedge dx_{i_1}\wedge\cdots\wedge dx_{i_k}. \end{align*} [/definition] The formula differentiates only the coefficient functions and then wedges the new differential in front. The signs that appear when the wedge factors are reordered are the mechanism by which curl and divergence acquire their standard coordinate expressions. Once this operator is defined, the essential structural question is whether it behaves like a differential in a chain complex and how it interacts with wedge multiplication. In Euclidean coordinates the answer comes from the formula itself: mixed second partial derivatives cancel in pairs, and the sign in the wedge product forces the graded Leibniz rule. These properties are what make exterior calculus coordinate computations fit together into a coherent algebra. The equation $d(d\omega)=0$ is the analytic source of familiar identities such as $\operatorname{curl}(\nabla f)=0$ and $\operatorname{div}(\operatorname{curl}F)=0$. Its proof uses equality of mixed second partial derivatives, so the smoothness hypothesis is not cosmetic: for merely $C^1$ coefficients, the second derivatives needed in $d(d\omega)$ need not exist classically, and the cancellation argument is no longer a classical coordinate computation. The Leibniz rule is also a graded rule, not the ordinary ungraded product rule. If $\omega\in\Omega^1(U)$, then \begin{align*} d(\omega\wedge\eta)=d\omega\wedge\eta-\omega\wedge d\eta, \end{align*} not $d\omega\wedge\eta+\omega\wedge d\eta$. The minus sign is forced by moving the new differential past one existing wedge factor, and forgetting it is one of the most common ways to get the wrong sign in exterior calculus. In later courses, $d^2=0$ becomes the starting point for de Rham cohomology. [example: Gradient Curl And Divergence] Let $U\subset\mathbb R^3$ be open, let $f:U\to\mathbb R$ be smooth, and let $F=(P,Q,R):U\to\mathbb R^3$ be smooth. We compute the [exterior derivative](/theorems/1525) in degrees $0$, $1$, and $2$ and compare the resulting coefficients with the usual gradient, curl, and divergence. For the $0$-form $f$, the coordinate definition of $d$ gives \begin{align*} df &=\partial_1f\,dx_1+\partial_2f\,dx_2+\partial_3f\,dx_3. \end{align*} Thus the coefficients of $df$ are the three components of $\nabla f$. Now set \begin{align*} \alpha=P\,dx_1+Q\,dx_2+R\,dx_3. \end{align*} Using linearity of $d$ and the coordinate definition, \begin{align*} d\alpha &=d(P\,dx_1)+d(Q\,dx_2)+d(R\,dx_3)\\ &=(\partial_1P\,dx_1+\partial_2P\,dx_2+\partial_3P\,dx_3)\wedge dx_1\\ &\quad+(\partial_1Q\,dx_1+\partial_2Q\,dx_2+\partial_3Q\,dx_3)\wedge dx_2\\ &\quad+(\partial_1R\,dx_1+\partial_2R\,dx_2+\partial_3R\,dx_3)\wedge dx_3\\ &=\partial_1P\,dx_1\wedge dx_1+\partial_2P\,dx_2\wedge dx_1+\partial_3P\,dx_3\wedge dx_1\\ &\quad+\partial_1Q\,dx_1\wedge dx_2+\partial_2Q\,dx_2\wedge dx_2+\partial_3Q\,dx_3\wedge dx_2\\ &\quad+\partial_1R\,dx_1\wedge dx_3+\partial_2R\,dx_2\wedge dx_3+\partial_3R\,dx_3\wedge dx_3. \end{align*} Since $dx_i\wedge dx_i=0$ and $dx_i\wedge dx_j=-dx_j\wedge dx_i$ for $i\neq j$, \begin{align*} d\alpha &=\partial_2P(-dx_1\wedge dx_2)+\partial_3P\,dx_3\wedge dx_1\\ &\quad+\partial_1Q\,dx_1\wedge dx_2+\partial_3Q(-dx_2\wedge dx_3)\\ &\quad+\partial_1R(-dx_3\wedge dx_1)+\partial_2R\,dx_2\wedge dx_3\\ &=(\partial_2R-\partial_3Q)\,dx_2\wedge dx_3\\ &\quad+(\partial_3P-\partial_1R)\,dx_3\wedge dx_1\\ &\quad+(\partial_1Q-\partial_2P)\,dx_1\wedge dx_2. \end{align*} These three coefficients are the components of $\nabla\times F$ in the standard cyclic basis of $2$-forms. Finally set \begin{align*} \beta=P\,dx_2\wedge dx_3+Q\,dx_3\wedge dx_1+R\,dx_1\wedge dx_2. \end{align*} Again using the coordinate definition, \begin{align*} d\beta &=(\partial_1P\,dx_1+\partial_2P\,dx_2+\partial_3P\,dx_3)\wedge dx_2\wedge dx_3\\ &\quad+(\partial_1Q\,dx_1+\partial_2Q\,dx_2+\partial_3Q\,dx_3)\wedge dx_3\wedge dx_1\\ &\quad+(\partial_1R\,dx_1+\partial_2R\,dx_2+\partial_3R\,dx_3)\wedge dx_1\wedge dx_2\\ &=\partial_1P\,dx_1\wedge dx_2\wedge dx_3\\ &\quad+\partial_2Q\,dx_2\wedge dx_3\wedge dx_1\\ &\quad+\partial_3R\,dx_3\wedge dx_1\wedge dx_2, \end{align*} because every other term contains a repeated factor and is zero. The two remaining cyclic reorderings have positive sign: \begin{align*} dx_2\wedge dx_3\wedge dx_1 &=-dx_2\wedge dx_1\wedge dx_3 =dx_1\wedge dx_2\wedge dx_3,\\ dx_3\wedge dx_1\wedge dx_2 &=-dx_1\wedge dx_3\wedge dx_2 =dx_1\wedge dx_2\wedge dx_3. \end{align*} Therefore \begin{align*} d\beta &=(\partial_1P+\partial_2Q+\partial_3R)\,dx_1\wedge dx_2\wedge dx_3. \end{align*} The coefficient of the standard volume form is exactly $\nabla\cdot F$, so $d$ packages gradient, curl, and divergence as the same degree-raising operation on forms. [/example] ## Pullback As The Right Composition Rule If a curve, surface, or coordinate chart parametrises a region, how does a differential form move back to the parameter domain where the integral is computed? The answer is the pullback. It is the operation that makes forms compatible with substitution and change of variables. [definition: Pullback Of A Differential Form] Let $F:U\to V$ be a smooth map between open sets $U\subset\mathbb R^m$ and $V\subset\mathbb R^n$. For $\omega\in\Omega^k(V)$, the pullback $F^*\omega\in\Omega^k(U)$ is 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 pullback uses the total derivative $DF_x$ to transport tangent vectors forward, then lets $\omega$ measure the transported parallelepiped. This direction is important: maps push tangent vectors forward, while forms pull back. After defining pullback pointwise, we need the algebraic rules that let it pass through sums, wedge products, and exterior derivatives. These rules are the formal reason substitution can be carried out before or after the exterior-calculus operations. [quotetheorem:3601] [citeproof:3601] These identities explain why differential forms are stable under reparametrisation. They allow calculations to move between physical coordinates and parameter coordinates without changing the geometric content of the integral. The smoothness of $F$ is essential because the definition uses $DF_x$ and the commutation rule $F^*d=dF^*$ is a chain-rule statement. A merely continuous map may pull points around, but it does not provide a [linear map](/page/Linear%20Map) on tangent vectors, so there is no classical pullback of differential forms of positive degree. The definition must pull the form back rather than pull tangent vectors back: for example, if $F:\mathbb R\to\mathbb R^2$ parametrises a curve, then $DF_t$ sends the single tangent direction of the parameter line into the plane, while there is no canonical way to pull an arbitrary tangent vector in $\mathbb R^2$ back to $\mathbb R$. The theorem also does not say that $F^*$ is invertible; if $F$ collapses dimensions or is not a diffeomorphism, information can be lost. [example: Polar Pullback Of The Area Form] Let $F(r,\theta)=(r\cos\theta,r\sin\theta)$. The coordinate functions pulled back along $F$ are \begin{align*} x_1\circ F(r,\theta)&=r\cos\theta,\\ x_2\circ F(r,\theta)&=r\sin\theta. \end{align*} Hence \begin{align*} F^*dx_1 &=d(x_1\circ F) =d(r\cos\theta) =\cos\theta\,dr-r\sin\theta\,d\theta,\\ F^*dx_2 &=d(x_2\circ F) =d(r\sin\theta) =\sin\theta\,dr+r\cos\theta\,d\theta. \end{align*} Using compatibility of pullback with wedge products, \begin{align*} F^*(dx_1\wedge dx_2) &=F^*dx_1\wedge F^*dx_2\\ &=(\cos\theta\,dr-r\sin\theta\,d\theta) \wedge(\sin\theta\,dr+r\cos\theta\,d\theta)\\ &=\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*} Since $dr\wedge dr=0$, $d\theta\wedge d\theta=0$, and $d\theta\wedge dr=-dr\wedge d\theta$, \begin{align*} F^*(dx_1\wedge dx_2) &=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*} Thus the polar parametrisation contributes the Jacobian factor $r$, with the positive sign coming from the ordered coordinates $(r,\theta)$. [/example] [illustration:oriented-parametrised-surface-patch] ## Integration Requires Orientation What extra structure is needed before a $k$-form can be integrated over a $k$-dimensional surface? Besides a parametrisation and a form of the matching degree, we need an orientation. Orientation records which ordered bases of tangent vectors count as positive, and reversing it changes the sign of the integral. [definition: Orientation Of A Vector Space] Let $V$ be an $n$-dimensional real [vector space](/page/Vector%20Space). An orientation of $V$ is one of the two equivalence classes of ordered bases under the relation \begin{align*} (v_1,\dots,v_n)\sim(w_1,\dots,w_n) \end{align*} when the change-of-basis matrix from $(v_1,\dots,v_n)$ to $(w_1,\dots,w_n)$ has positive determinant. [/definition] For parametrised domains, the standard orientation of $\mathbb R^k$ supplies the ordered tangent directions. Pulling a form back to the parameter domain converts the geometric integral into an ordinary multiple integral. There is still a choice to make: the surface itself may live in $\mathbb R^n$, while the ordinary integral must be performed over a subset of $\mathbb R^k$. The pullback is the device that transports both the coefficients and the ordered tangent vectors back to the parameter space, so that stretching and orientation are recorded before integration occurs. This gives a practical definition before any global theory of manifolds is needed. [definition: Integral Over A Parametrised Domain] Let $W\subset\mathbb R^k$ be an oriented parameter domain, let $U\subset\mathbb R^n$ be open, let $\Phi:W\to U$ be a smooth parametrisation, and let $\omega\in\Omega^k(U)$ be smooth with $\Phi^*\omega$ compactly supported in $W$. The integral of $\omega$ over the parametrised domain $\Phi$ is \begin{align*} \int_{\Phi}\omega := \int_W \Phi^*\omega. \end{align*} [/definition] This definition is local and computational. The course then proves that the value is independent of orientation-preserving changes of parameter, so it descends from parametrisations to oriented geometric domains. [quotetheorem:3602] [citeproof:3602] The theorem is the bridge from calculus on parameter domains to geometry on curves, surfaces, and higher-dimensional domains. It is also the first place where determinants enter as orientation data rather than as isolated Jacobian factors. The differentiability hypothesis on $\psi$ cannot be replaced by a purely topological one. If $\psi$ is only a homeomorphism, the Jacobian determinant in the change-of-variables formula is not defined, so there is no classical mechanism for comparing the pulled-back forms. In dimension one, the sign rule is completely visible: for $\psi:[-1,1]\to[-1,1]$ given by $\psi(t)=-t$, one has $\psi^*dt=-dt$, and hence reversing the parametrisation of a curve changes the integral of every pulled-back $1$-form by a minus sign. [example: Work Integral As A Pullback] Let $\gamma:[a,b]\to U\subset\mathbb R^n$ be a smooth oriented curve, write $\gamma(t)=(\gamma_1(t),\dots,\gamma_n(t))$, and let $\omega=\sum_{i=1}^n P_i\,dx_i\in\Omega^1(U)$. For each coordinate function $x_i$, we have $x_i\circ\gamma=\gamma_i$, so \begin{align*} \gamma^*dx_i &=d(x_i\circ\gamma)\\ &=d\gamma_i\\ &=\dot{\gamma}_i(t)\,dt. \end{align*} Using linearity of pullback, \begin{align*} \gamma^*\omega &=\gamma^*\left(\sum_{i=1}^n P_i\,dx_i\right)\\ &=\sum_{i=1}^n \gamma^*(P_i\,dx_i)\\ &=\sum_{i=1}^n (P_i\circ\gamma)\,\gamma^*dx_i\\ &=\sum_{i=1}^n P_i(\gamma(t))\,\dot{\gamma}_i(t)\,dt\\ &=\left(\sum_{i=1}^n P_i(\gamma(t))\,\dot{\gamma}_i(t)\right)dt. \end{align*} Therefore, by the definition of integration over a parametrised domain, \begin{align*} \int_{\gamma}\omega &=\int_{[a,b]}\gamma^*\omega\\ &=\int_a^b \sum_{i=1}^n P_i(\gamma(t))\,\dot{\gamma}_i(t)\,dt. \end{align*} Thus the pullback definition of integrating a $1$-form over an oriented curve recovers the classical work integral. [/example] ## The Boundary Formula That Ends The Course Which theorem makes the whole formalism worth building? The answer is the general Stokes theorem. It says that exterior differentiation inside an oriented domain is adjoint to restricting the original form to the oriented boundary. In preview form, the theorem says that once oriented manifolds with boundary and integration of forms have been built, the identity \begin{align*} \int_M d\omega=\int_{\partial M}\omega \end{align*} will unify the fundamental theorem of calculus, Green's theorem, the classical Stokes theorem, and the divergence theorem. This preview should be read as a map of the course rather than as a finished proof. Each earlier topic supplies one ingredient: exterior algebra gives forms, the [exterior derivative](/theorems/1525) gives $d\omega$, pullback handles charts, orientation fixes signs, and integration turns local coordinate formulas into a global theorem. [illustration:induced-boundary-orientation] [example: Green's Theorem As Stokes] [claim]Let $D\subset\mathbb R^2$ be a positively oriented compact planar domain with smooth boundary, and let $P,Q$ be smooth on an open neighbourhood of $D$. For \begin{align*} \omega=P\,dx_1+Q\,dx_2, \end{align*} the identity \begin{align*} \int_{\partial D}(P\,dx_1+Q\,dx_2) = \int_D (\partial_1Q-\partial_2P)\,dx_1\wedge dx_2 \end{align*} is the two-dimensional case of [Stokes' theorem](/theorems/1530).[/claim] [proof]Using the coordinate definition of the [exterior derivative](/theorems/1525) and linearity, \begin{align*} d\omega &=d(P\,dx_1)+d(Q\,dx_2)\\ &=(\partial_1P\,dx_1+\partial_2P\,dx_2)\wedge dx_1 +(\partial_1Q\,dx_1+\partial_2Q\,dx_2)\wedge dx_2\\ &=\partial_1P\,dx_1\wedge dx_1 +\partial_2P\,dx_2\wedge dx_1 +\partial_1Q\,dx_1\wedge dx_2 +\partial_2Q\,dx_2\wedge dx_2. \end{align*} Since $dx_1\wedge dx_1=0$, $dx_2\wedge dx_2=0$, and $dx_2\wedge dx_1=-dx_1\wedge dx_2$, \begin{align*} d\omega &=0+\partial_2P(-dx_1\wedge dx_2) +\partial_1Q\,dx_1\wedge dx_2+0\\ &=(\partial_1Q-\partial_2P)\,dx_1\wedge dx_2. \end{align*} By the generalized Stokes theorem applied to the oriented surface $D$ and the $1$-form $\omega$, \begin{align*} \int_{\partial D}(P\,dx_1+Q\,dx_2) &=\int_{\partial D}\omega\\ &=\int_D d\omega\\ &=\int_D(\partial_1Q-\partial_2P)\,dx_1\wedge dx_2. \end{align*} [/proof] For the standard positive orientation on $D$, the induced boundary orientation is the usual counterclockwise orientation, so this is exactly [Green's theorem](/theorems/3612) in differential-form notation. [/example] ## How To Read The Rest Of The Course How should the later chapters be organised so that the final theorem feels inevitable rather than mysterious? The course proceeds from algebra to analysis to geometry. Each step introduces one operation and then checks that it behaves well under coordinates. The first part studies multilinear algebra: tensor products, alternating maps, wedge products, bases of exterior powers, and the determinant as a top-degree form. This part explains why the signs in exterior calculus are forced by orientation. The second part studies differential forms on open subsets of $\mathbb R^n$: coordinate expressions, the [exterior derivative](/theorems/1525), closed and exact forms, pullbacks, and the identities $d^2=0$ and $F^*\circ d=d\circ F^*$. This part is where vector calculus is translated into forms. The third part studies integration: oriented parametrised domains, changes of variables, manifolds-with-boundary in Euclidean space, and the [induced orientation on the boundary](/theorems/1528). This part prepares the precise statement and proof of [Stokes' theorem](/theorems/1530). The final part applies [Stokes' theorem](/theorems/1530) back to classical calculus. [Green's theorem](/theorems/3612), the [divergence theorem](/theorems/3614), and the Kelvin-Stokes theorem become special cases obtained by choosing the degree of the form and the dimension of the domain. A reader should keep three questions active throughout the course. What degree is the form? What is the orientation of the domain? What happens to the form under pullback to coordinates? Most computations in exterior calculus are disciplined answers to these three questions. The opening chapter raises the three questions that guide the whole course: degree, orientation, and behavior under pullback. To answer them precisely, we first need the exterior algebra of alternating multilinear objects at a single point, where those questions can be stated and manipulated cleanly. # 1. Multilinear Algebra: The Exterior Algebra This opening chapter supplies the algebra that differential forms use at every point of an [open set](/page/Open%20Set) in $\mathbb R^n$. The main issue is how to encode objects that accept several tangent vectors, change sign when two inputs are interchanged, and multiply in a way compatible with oriented volume. Later chapters will let the [vector space](/page/Vector%20Space) vary from point to point; here we work at a single finite-dimensional real [vector space](/page/Vector%20Space) $V$. The guiding theme is that the exterior algebra is the part of multilinear algebra designed for integration. Alternating forms measure directed $k$-dimensional content, the wedge product builds higher-dimensional measurements from lower-dimensional ones, and the determinant appears as the top-degree alternating form on $\mathbb R^n$. ## Multilinear Maps and Tensor Products How should we formalise an expression that is linear in several vector inputs at once? Ordinary linear functionals measure one vector, but line integrals, surface integrals, and determinants require functions of several vectors with a separate linearity condition in each slot. [definition: Multilinear Form] Let $V$ be a real [vector space](/page/Vector%20Space) and let $k \in \mathbb N$. A $k$-linear form on $V$ is a map \begin{align*} A: V^k \to \mathbb R \end{align*} such that, for each $j \in \{1,\dots,k\}$, the map \begin{align*} v_j \mapsto A(v_1,\dots,v_j,\dots,v_k) \end{align*} is linear when all other entries are fixed. [/definition] For $k=1$ this is an element of the [dual space](/page/Dual%20Space) $V^*$. For $k=2$ it is a [bilinear form](/page/Bilinear%20Form), such as an inner product or a matrix pairing. The point of separating the entries is that the same object can respond independently to several geometric directions. [example: Bilinear Form From A Matrix] Let $V=\mathbb R^n$ and let $A\in\mathbb R^{n\times n}$. Define $B:V^2\to\mathbb R$ by $B(v,w)=v^\top Aw$. We show that $B$ is linear in each input separately. For $r,s\in\mathbb R$ and $v_1,v_2,w\in V$, transpose linearity and distributivity of matrix multiplication give \begin{align*} B(rv_1+sv_2,w) &=(rv_1+sv_2)^\top Aw\\ &=(rv_1^\top+sv_2^\top)Aw\\ &=r\,v_1^\top Aw+s\,v_2^\top Aw\\ &=rB(v_1,w)+sB(v_2,w). \end{align*} Thus $v\mapsto B(v,w)$ is linear for each fixed $w$. For the second input, take $r,s\in\mathbb R$ and $v,w_1,w_2\in V$. Linearity of $A:\mathbb R^n\to\mathbb R^n$ and distributivity of row-vector multiplication give \begin{align*} B(v,rw_1+sw_2) &=v^\top A(rw_1+sw_2)\\ &=v^\top(rAw_1+sAw_2)\\ &=r\,v^\top Aw_1+s\,v^\top Aw_2\\ &=rB(v,w_1)+sB(v,w_2). \end{align*} Therefore $B$ is linear in each slot when the other slot is fixed, so $B$ is a [bilinear form](/page/Bilinear%20Form) on $\mathbb R^n$. [/example] Tensor products provide a bookkeeping device for all multilinear forms. They allow us to multiply linear functionals before imposing any alternating symmetry. Before defining the exterior product, we need a non-antisymmetric product that simply remembers which covector acts on which input slot. The tensor product supplies that ordered raw product. [definition: Tensor Product Of Covectors] Let $\alpha_1,\dots,\alpha_k \in V^*$. Their [tensor product](/page/Tensor%20Product) is the $k$-linear form \begin{align*} \alpha_1 \otimes \cdots \otimes \alpha_k : V^k \to \mathbb R \end{align*} defined by \begin{align*} (\alpha_1 \otimes \cdots \otimes \alpha_k)(v_1,\dots,v_k) = \alpha_1(v_1)\cdots \alpha_k(v_k). \end{align*} [/definition] The [tensor product](/page/Tensor%20Product) is not meant to be alternating. It remembers the ordered slots. This ordered algebra is the raw material from which the exterior algebra will be obtained by antisymmetrising. Before imposing sign relations, we need a single ambient algebra in which tensors of different degrees can be multiplied and compared. Keeping all covariant tensor powers together makes concatenation the basic product, and only after that can alternation be introduced as a structured restriction of this larger object. [definition: Covariant Tensor Algebra] Let $V$ be a finite-dimensional real [vector space](/page/Vector%20Space). For $k\ge 0$, write $(V^*)^{\otimes k}$ for the [vector space](/page/Vector%20Space) spanned by tensor products of $k$ covectors, with $(V^*)^{\otimes 0}=\mathbb R$. The covariant tensor algebra of $V$ is \begin{align*} T^*(V)=\bigoplus_{k=0}^\infty (V^*)^{\otimes k}. \end{align*} [/definition] Multiplication in $T^*(V)$ is concatenation of tensor factors. Differential forms will live in a quotient-like substructure of this algebra where repeated directions vanish and permutations contribute signs. ## Alternating Forms and the Wedge Product Which multilinear forms measure oriented area, volume, and their higher-dimensional analogues? Such measurements should vanish when two input directions coincide, because two equal directions span no two-dimensional area and no higher-dimensional content. [definition: Alternating Form] Let $V$ be a real [vector space](/page/Vector%20Space) and let $k\ge 1$. A $k$-linear form $\omega:V^k\to\mathbb R$ is alternating if \begin{align*} \omega(v_1,\dots,v_k)=0 \end{align*} whenever $v_i=v_j$ for some $i\ne j$. [/definition] Over $\mathbb R$, alternation is equivalent to changing sign whenever two adjacent arguments are interchanged. This sign rule is the algebraic source of orientation. Individual alternating forms are not enough for an algebra: we also need to add them, scale them, and keep track of their degree. The zero-degree case must be included as well, because ordinary scalar functions will later multiply forms and act as degree-zero elements. The following notation packages all alternating measurements of one fixed degree into a vector space. [definition: Space Of Alternating Forms] Let $V$ be a real [vector space](/page/Vector%20Space). For $k\ge 1$, $\Lambda^k(V^*)$ is the [vector space](/page/Vector%20Space) of alternating $k$-linear forms on $V$. Set $\Lambda^0(V^*)=\mathbb R$. [/definition] A $0$-form in this algebra is a scalar, a $1$-form is an ordinary covector, and a $k$-form is an alternating measurement of $k$ ordered vectors. The next construction explains how to multiply such measurements. [definition: Antisymmetrisation] Let $V$ be a real [vector space](/page/Vector%20Space) and let $k\ge 1$. The antisymmetrisation operator sends a $k$-linear form $A:V^k\to\mathbb R$ to the alternating $k$-linear form $\operatorname{Alt}(A):V^k\to\mathbb R$ given by \begin{align*} \operatorname{Alt}(A)(v_1,\dots,v_k) =\frac{1}{k!}\sum_{\sigma\in S_k}\operatorname{sgn}(\sigma) A(v_{\sigma(1)},\dots,v_{\sigma(k)}). \end{align*} [/definition] The factor $1/k!$ makes $\operatorname{Alt}$ a projection onto alternating forms. The sign of the permutation records how the orientation of an ordered list changes under reordering. The remaining task is to multiply two alternating forms and stay inside the alternating world. The wedge product does this by tensoring first and then antisymmetrising with the normalisation used in coordinates. [definition: Wedge Product] Let $V$ be a real [vector space](/page/Vector%20Space), let $k,l\ge 0$, and let $\alpha\in \Lambda^k(V^*)$ and $\beta\in \Lambda^l(V^*)$. The wedge product operation sends $(\alpha,\beta)$ to the element $\alpha\wedge\beta\in \Lambda^{k+l}(V^*)$ defined by \begin{align*} \alpha\wedge\beta =\frac{(k+l)!}{k!l!}\operatorname{Alt}(\alpha\otimes\beta). \end{align*} [/definition] This normalisation is chosen so that coordinate basis forms multiply without extra binomial constants. It also makes the wedge product associative with the standard basis conventions. [example: Basic Wedge Product In Three Variables] Let $V=\mathbb R^3$ with standard basis $e_1,e_2,e_3$ and coordinate covectors $dx,dy,dz$, so $dx(e_1)=1$, $dy(e_2)=1$, $dz(e_3)=1$, and each of these covectors is $0$ on the other two standard basis vectors. For $u,v\in V$, the definition of the wedge product of two $1$-forms gives \begin{align*} (dx\wedge dy)(u,v) &=2\,\operatorname{Alt}(dx\otimes dy)(u,v)\\ &=2\cdot \frac{1}{2}\bigl(dx(u)dy(v)-dx(v)dy(u)\bigr)\\ &=dx(u)dy(v)-dx(v)dy(u). \end{align*} Hence \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. \end{align*} Now apply the wedge product definition to the $2$-form $dx\wedge dy$ and the $1$-form $dz$. Since $k=2$ and $l=1$, \begin{align*} ((dx\wedge dy)\wedge dz)(e_1,e_2,e_3) &=3\,\operatorname{Alt}\bigl((dx\wedge dy)\otimes dz\bigr)(e_1,e_2,e_3)\\ &=3\cdot \frac{1}{6}\sum_{\sigma\in S_3}\operatorname{sgn}(\sigma) (dx\wedge dy)(e_{\sigma(1)},e_{\sigma(2)})dz(e_{\sigma(3)}). \end{align*} The six summands are \begin{align*} &\phantom{{}+{}}(dx\wedge dy)(e_1,e_2)dz(e_3)\\ &-(dx\wedge dy)(e_2,e_1)dz(e_3)\\ &-(dx\wedge dy)(e_1,e_3)dz(e_2)\\ &+(dx\wedge dy)(e_3,e_1)dz(e_2)\\ &+(dx\wedge dy)(e_2,e_3)dz(e_1)\\ &-(dx\wedge dy)(e_3,e_2)dz(e_1). \end{align*} Using $dz(e_3)=1$ and $dz(e_1)=dz(e_2)=0$, this becomes \begin{align*} & (dx\wedge dy)(e_1,e_2)\cdot 1 -(dx\wedge dy)(e_2,e_1)\cdot 1\\ &\qquad -(dx\wedge dy)(e_1,e_3)\cdot 0 +(dx\wedge dy)(e_3,e_1)\cdot 0 +(dx\wedge dy)(e_2,e_3)\cdot 0 -(dx\wedge dy)(e_3,e_2)\cdot 0. \end{align*} Also \begin{align*} (dx\wedge dy)(e_2,e_1) &=dx(e_2)dy(e_1)-dx(e_1)dy(e_2)\\ &=0\cdot 0-1\cdot 1\\ &=-1. \end{align*} Therefore \begin{align*} ((dx\wedge dy)\wedge dz)(e_1,e_2,e_3) &=3\cdot \frac{1}{6}\bigl(1-(-1)\bigr)\\ &=3\cdot \frac{2}{6}\\ &=1. \end{align*} The form $(dx\wedge dy)\wedge dz$ is alternating by the definition of the wedge product, so its value on any reordered standard triple is the sign of the corresponding permutation, and its value is $0$ when a standard basis vector is repeated. Thus $(dx\wedge dy)\wedge dz$ is exactly the standard oriented volume form $dx\wedge dy\wedge dz$. [/example] The example used two sign descriptions of alternation at once: vanishing on repeated directions and changing sign under swaps. Those conditions look different: one is a degeneracy condition, while the other controls how signs change under reordering. To use whichever version is convenient in later wedge-product computations, we need a result showing that multilinearity over $\mathbb R$ forces the two descriptions to agree. For real multilinear forms, these two descriptions agree. If an alternating multilinear form vanishes whenever two inputs are equal, then swapping two inputs changes the sign: \begin{align*} \omega(\dots,v_i,\dots,v_j,\dots)=-\omega(\dots,v_j,\dots,v_i,\dots). \end{align*} The multilinearity hypothesis is essential: expanding the value on $v_i+v_j$ is what forces the sign rule. The result does not classify all multilinear forms, only the alternating ones, and it prepares the computational rule used in wedge products: reordering inputs changes only the sign. For the wedge product used in these notes, two structural rules will be used constantly. First, wedge multiplication is associative, so $\alpha\wedge\beta\wedge\gamma$ may be written without parentheses. Second, homogeneous forms satisfy the graded sign rule \begin{align*} \alpha\wedge\beta=(-1)^{k\ell}\beta\wedge\alpha \end{align*} when $\alpha$ has degree $k$ and $\beta$ has degree $\ell$. For $1$-forms this gives $\alpha\wedge\beta=-\beta\wedge\alpha$, so $\alpha\wedge\alpha=0$. For two $2$-forms the sign is positive because $(-1)^{2\cdot 2}=1$. These rules are the computational heart of the exterior algebra. They do not make wedge multiplication ordinary commutative; they describe the controlled failure of commutativity that governs signs in coordinate calculations. The word homogeneous is important because a sum of forms of different degrees has no single degree-dependent sign. ## Coordinate Bases and Dimension Once alternating forms have been defined abstractly, how many independent $k$-forms are there on $\mathbb R^n$? The answer should match the number of choices of $k$ coordinate directions, since an alternating form ignores repeated directions and changes sign under reordering. Let $e_1,\dots,e_n$ be the standard basis of $\mathbb R^n$, and let $dx_1,\dots,dx_n$ be the [dual basis](/theorems/414) of $(\mathbb R^n)^*$, so $dx_i(e_j)=\delta_{ij}$. [quotetheorem:3561] [citeproof:3561] This theorem identifies a $k$-form with its coordinate coefficients. The strict ordering $i_1<\cdots<i_k$ is not cosmetic: without it, the list would contain duplicates such as $dx_2\wedge dx_1=-dx_1\wedge dx_2$ and zero terms such as $dx_1\wedge dx_1$. The theorem is basis-dependent in its displayed form, so it does not choose canonical coordinates on an arbitrary [vector space](/page/Vector%20Space); it says that once a basis is chosen, alternating forms are counted by subsets of basis directions. This prepares the top-degree case, where there is only one basis element and the determinant appears. [example: Two Forms On Three Space] Let $V=\mathbb R^3$ with standard basis $e_1,e_2,e_3$ and coordinate covectors $dx,dy,dz$. For \begin{align*} \omega=a\,dx\wedge dy+b\,dx\wedge dz+c\,dy\wedge dz, \end{align*} we compute its values on the three positively ordered coordinate planes. For two $1$-forms, the wedge product definition gives \begin{align*} (\alpha\wedge\beta)(u,v)=\alpha(u)\beta(v)-\alpha(v)\beta(u). \end{align*} Thus \begin{align*} \omega(e_1,e_2) &=a(dx\wedge dy)(e_1,e_2)+b(dx\wedge dz)(e_1,e_2)+c(dy\wedge dz)(e_1,e_2)\\ &=a\bigl(dx(e_1)dy(e_2)-dx(e_2)dy(e_1)\bigr)\\ &\qquad +b\bigl(dx(e_1)dz(e_2)-dx(e_2)dz(e_1)\bigr)\\ &\qquad +c\bigl(dy(e_1)dz(e_2)-dy(e_2)dz(e_1)\bigr)\\ &=a(1\cdot 1-0\cdot 0)+b(1\cdot 0-0\cdot 0)+c(0\cdot 0-1\cdot 0)\\ &=a. \end{align*} Similarly, \begin{align*} \omega(e_1,e_3) &=a(dx\wedge dy)(e_1,e_3)+b(dx\wedge dz)(e_1,e_3)+c(dy\wedge dz)(e_1,e_3)\\ &=a(1\cdot 0-0\cdot 0)+b(1\cdot 1-0\cdot 0)+c(0\cdot 1-0\cdot 0)\\ &=b, \end{align*} and \begin{align*} \omega(e_2,e_3) &=a(dx\wedge dy)(e_2,e_3)+b(dx\wedge dz)(e_2,e_3)+c(dy\wedge dz)(e_2,e_3)\\ &=a(0\cdot 0-0\cdot 1)+b(0\cdot 1-0\cdot 0)+c(1\cdot 1-0\cdot 0)\\ &=c. \end{align*} So $a$, $b$, and $c$ are exactly the oriented measurements of $\omega$ on the coordinate planes spanned by $(e_1,e_2)$, $(e_1,e_3)$, and $(e_2,e_3)$, respectively. [/example] When computing, the basis theorem also gives a reduction rule: reorder wedge factors into increasing order and attach the sign of the permutation. Any repeated factor makes the term vanish. [example: Reordering A Wedge] Let $U\subset\mathbb R^3$, let $v=P\partial_x+Q\partial_y+R\partial_z$, and put \begin{align*} \operatorname{vol}=dx\wedge dy\wedge dz. \end{align*} This example previews the contraction operation $\iota_v$, which is defined formally in the next chapter. We compute the resulting $2$-form by evaluating it on the coordinate pairs. At a point $p\in U$, \begin{align*} (\iota_v\operatorname{vol})_p(\partial_y|_p,\partial_z|_p) &=\operatorname{vol}_p(v_p,\partial_y|_p,\partial_z|_p)\\ &=\operatorname{vol}_p(P(p)\partial_x|_p+Q(p)\partial_y|_p+R(p)\partial_z|_p,\partial_y|_p,\partial_z|_p)\\ &=P(p)\operatorname{vol}_p(\partial_x|_p,\partial_y|_p,\partial_z|_p) +Q(p)\operatorname{vol}_p(\partial_y|_p,\partial_y|_p,\partial_z|_p)\\ &\qquad +R(p)\operatorname{vol}_p(\partial_z|_p,\partial_y|_p,\partial_z|_p)\\ &=P(p)\cdot 1+Q(p)\cdot 0+R(p)\cdot 0\\ &=P(p). \end{align*} The two zero terms vanish because an alternating form is zero when two arguments are equal. Similarly, \begin{align*} (\iota_v\operatorname{vol})_p(\partial_x|_p,\partial_z|_p) &=\operatorname{vol}_p(v_p,\partial_x|_p,\partial_z|_p)\\ &=P(p)\operatorname{vol}_p(\partial_x|_p,\partial_x|_p,\partial_z|_p) +Q(p)\operatorname{vol}_p(\partial_y|_p,\partial_x|_p,\partial_z|_p)\\ &\qquad +R(p)\operatorname{vol}_p(\partial_z|_p,\partial_x|_p,\partial_z|_p)\\ &=P(p)\cdot 0+Q(p)(-1)+R(p)\cdot 0\\ &=-Q(p), \end{align*} because $(\partial_y,\partial_x,\partial_z)$ is obtained from $(\partial_x,\partial_y,\partial_z)$ by one transposition. Finally, \begin{align*} (\iota_v\operatorname{vol})_p(\partial_x|_p,\partial_y|_p) &=\operatorname{vol}_p(v_p,\partial_x|_p,\partial_y|_p)\\ &=P(p)\operatorname{vol}_p(\partial_x|_p,\partial_x|_p,\partial_y|_p) +Q(p)\operatorname{vol}_p(\partial_y|_p,\partial_x|_p,\partial_y|_p)\\ &\qquad +R(p)\operatorname{vol}_p(\partial_z|_p,\partial_x|_p,\partial_y|_p)\\ &=P(p)\cdot 0+Q(p)\cdot 0+R(p)\cdot 1\\ &=R(p), \end{align*} because $(\partial_z,\partial_x,\partial_y)$ is obtained from $(\partial_x,\partial_y,\partial_z)$ by two transpositions. These three values are exactly the coefficients of the coordinate $2$-form expansion, since $dy\wedge dz$ evaluates to $1$ on $(\partial_y,\partial_z)$, $dx\wedge dz$ evaluates to $1$ on $(\partial_x,\partial_z)$, and $dx\wedge dy$ evaluates to $1$ on $(\partial_x,\partial_y)$. Therefore \begin{align*} \iota_v(dx\wedge dy\wedge dz) &=P\,dy\wedge dz-Q\,dx\wedge dz+R\,dx\wedge dy. \end{align*} Using graded commutativity for $1$-forms, \begin{align*} dz\wedge dx&=-dx\wedge dz, \end{align*} so the same identity can be written as \begin{align*} \iota_v(dx\wedge dy\wedge dz) &=P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy. \end{align*} Thus contracting the standard volume form by $v$ produces the usual flux $2$-form whose coefficients are the coordinate components of $v$. [/example] ## The Determinant As A Top Form Why does the determinant keep appearing whenever oriented volume is involved? The exterior algebra answer is that the determinant is the fundamental alternating $n$-linear form on $\mathbb R^n$, and every other top-degree form is a scalar multiple of it. [definition: Standard Volume Form] On $\mathbb R^n$, the standard volume form is \begin{align*} \operatorname{vol}=dx_1\wedge\cdots\wedge dx_n\in\Lambda^n((\mathbb R^n)^*). \end{align*} [/definition] This form evaluates an ordered list of $n$ vectors by extracting its signed coordinate volume. In coordinates, that signed volume is the determinant of the matrix with those vectors as columns. The point still needs to be made algebraically: the coordinate top form should agree exactly with the determinant function on column vectors. This identification ties exterior algebra back to the familiar determinant from linear algebra. Concretely, if $v_1,\dots,v_n$ are the columns of a matrix $A$, then \begin{align*} (dx_1\wedge\cdots\wedge dx_n)(v_1,\dots,v_n)=\det A. \end{align*} This gives the geometric meaning of the wedge product: wedging $n$ coordinate covectors produces the signed volume functional. It does not define Euclidean volume for arbitrary sets, only the alternating volume of a parallelepiped spanned by $n$ vectors. It closes the pointwise algebraic story by showing that top-degree forms transform by determinants, which is the algebraic origin of orientation. [example: Recovering A Three By Three Determinant] Let $V=\mathbb R^3$ with coordinate covectors $dx,dy,dz$. Write \begin{align*} v_j=x_j e_1+y_j e_2+z_j e_3 \end{align*} for $j=1,2,3$, so the matrix with columns $v_1,v_2,v_3$ is \begin{align*} A= \begin{pmatrix} x_1&x_2&x_3\\ y_1&y_2&y_3\\ z_1&z_2&z_3 \end{pmatrix}. \end{align*} Since $dx(v_j)=x_j$, $dy(v_j)=y_j$, and $dz(v_j)=z_j$, expanding the wedge product as the signed sum over $S_3$ gives \begin{align*} (dx\wedge dy\wedge dz)(v_1,v_2,v_3) &=dx(v_1)dy(v_2)dz(v_3)-dx(v_1)dy(v_3)dz(v_2)\\ &\qquad -dx(v_2)dy(v_1)dz(v_3)+dx(v_2)dy(v_3)dz(v_1)\\ &\qquad +dx(v_3)dy(v_1)dz(v_2)-dx(v_3)dy(v_2)dz(v_1)\\ &=x_1y_2z_3-x_1y_3z_2-x_2y_1z_3+x_2y_3z_1+x_3y_1z_2-x_3y_2z_1. \end{align*} On the other hand, expanding $\det A$ along the first row gives \begin{align*} \det A &=x_1 \begin{vmatrix} y_2&y_3\\ z_2&z_3 \end{vmatrix} -x_2 \begin{vmatrix} y_1&y_3\\ z_1&z_3 \end{vmatrix} +x_3 \begin{vmatrix} y_1&y_2\\ z_1&z_2 \end{vmatrix}\\ &=x_1(y_2z_3-y_3z_2)-x_2(y_1z_3-y_3z_1)+x_3(y_1z_2-y_2z_1)\\ &=x_1y_2z_3-x_1y_3z_2-x_2y_1z_3+x_2y_3z_1+x_3y_1z_2-x_3y_2z_1. \end{align*} The two displayed expansions are identical term by term, so \begin{align*} (dx\wedge dy\wedge dz)(v_1,v_2,v_3)=\det A. \end{align*} Thus the coordinate top form evaluates three vectors by taking the signed volume of the parallelepiped whose columns are those vectors. [/example] This example is the computational version of the top-form theorem. It also fixes the sign convention used below: reversing two of the input vectors reverses the sign of the determinant and hence the sign of the top form. [remark: Signed Volume] The sign of $\operatorname{vol}(v_1,\dots,v_n)$ records whether the ordered basis $(v_1,\dots,v_n)$ has the same orientation as the standard basis. Its absolute value is the Euclidean $n$-dimensional volume of the parallelepiped spanned by the vectors. [/remark] The algebra developed in this chapter is pointwise. In the next chapter, the coefficients of these alternating forms are allowed to vary smoothly with $x\in U\subset\mathbb R^n$, producing differential $k$-forms such as $f(x)\,dx_i\wedge dx_j$ on open subsets of Euclidean space. Once the exterior algebra is in place at a single [vector space](/page/Vector%20Space), the next step is to let its coefficients vary smoothly from point to point. That turns static alternating forms into differential forms on open sets of $\mathbb R^n$, where wedge products and coordinate expressions become the basic language of calculation. # 2. Differential Forms on Open Sets of $\mathbb{R}^n$ The previous chapter built the exterior algebra at a single [vector space](/page/Vector%20Space). This chapter turns that pointwise algebra into a calculus on an [open set](/page/Open%20Set) $U \subset \mathbb R^n$. The guiding question is how to let alternating covectors vary smoothly with the base point while retaining the wedge product and the coordinate bases from multilinear algebra. The result is the space $\Omega^k(U)$ of smooth $k$-forms, which is the basic object of exterior calculus on open subsets of Euclidean space. The central shift is from algebra at one tangent space to algebra in a smoothly varying family of tangent spaces. Because $U$ is open in $\mathbb R^n$, translation in $\mathbb R^n$ identifies each tangent space $T_p\mathbb R^n$ with $\mathbb R^n$, and the coordinate covectors $dx_1,\dots,dx_n$ give a common notation across all points. This Euclidean setting keeps the geometry visible while preparing the notation used later on manifolds. ## Smooth Alternating Covectors Varying With A Point The first problem is to decide what it means for an alternating multilinear object to depend smoothly on a point. At a fixed point $p$, a $k$-form should be an element of $\Lambda^k(T_p^*\mathbb R^n)$. The naive approach would allow an arbitrary choice of alternating covector at each point, but that would leave no coefficient functions to differentiate later and hence no [exterior derivative](/theorems/1525). Smoothness is therefore not decoration: it is the condition that turns pointwise exterior algebra into calculus. Smoothness is tested by writing the form in the coordinate exterior basis and asking its coefficient functions to be smooth. For a strictly increasing $k$-tuple $I=(i_1,\dots,i_k)$ with $1 \le i_1 < \cdots < i_k \le n$, write \begin{align*} dx_I = dx_{i_1}\wedge\cdots\wedge dx_{i_k}. \end{align*} For $k=0$, the empty index corresponds to the constant basis element $1$. [definition: Smooth Differential Form] Let $U \subset \mathbb R^n$ be open, and let $0 \le k \le n$. A smooth $k$-form on $U$ is an assignment $p \mapsto \omega_p$ with \begin{align*} \omega_p \in \Lambda^k(T_p^*\mathbb R^n) \end{align*} for each $p \in U$, such that in coordinates \begin{align*} \omega_p = \sum_I f_I(p)\,(dx_I)_p \end{align*} each coefficient function $f_I:U\to\mathbb R$ is smooth. [/definition] The space of smooth $k$-forms on $U$ is denoted $\Omega^k(U)$. The summation ranges over all strictly increasing $k$-tuples $I$, so there are $\binom{n}{k}$ coordinate basis forms when $0\le k\le n$ and no such forms when $k>n$. The case $k=0$ deserves to be stated separately because there are no tangent-vector inputs to measure. In that degree, a form is just a smooth scalar field on the domain. [definition: Zero Form] A $0$-form on $U$ is a smooth function $f:U\to\mathbb R$. [/definition] This convention makes functions the degree-zero part of the theory. Multiplication of functions will become wedge multiplication in degree zero, and this is why the algebra of forms contains ordinary calculus as its first layer. For computations, one also needs to know that every smooth form has a unique coordinate expansion with smooth coefficient functions. That basis statement is what turns abstract forms into the sums used in exterior differentiation and integration. Every smooth $k$-form on an open set $U\subset\mathbb R^n$ can be written uniquely as \begin{align*} \omega=\sum_I f_I\,dx_I \end{align*} with smooth coefficient functions $f_I$. This coordinate expansion is the backbone of computations with forms. Openness of $U$ lets the coordinate vector fields and covectors restrict smoothly to every point of $U$; without an open domain in $\mathbb R^n$, ordinary partial derivatives and the later formula for $d$ would not be available in this form. Later arguments use the expansion by checking an identity on the basis forms $dx_I$ and then extending by smooth coefficients. [example: One Forms On The Plane] Let $U\subset\mathbb R^2$ have coordinates $(x,y)$, and let \begin{align*} \omega=P\,dx+Q\,dy \end{align*} with $P,Q\in C^\infty(U)$. Fix $p\in U$ and a tangent vector \begin{align*} X_p=a\,\partial_x|_p+b\,\partial_y|_p. \end{align*} The coordinate covectors are dual to the coordinate vectors, so \begin{align*} dx_p(X_p) &=dx_p(a\,\partial_x|_p+b\,\partial_y|_p)\\ &=a\,dx_p(\partial_x|_p)+b\,dx_p(\partial_y|_p)\\ &=a\cdot 1+b\cdot 0\\ &=a, \end{align*} and similarly \begin{align*} dy_p(X_p) &=dy_p(a\,\partial_x|_p+b\,\partial_y|_p)\\ &=a\,dy_p(\partial_x|_p)+b\,dy_p(\partial_y|_p)\\ &=a\cdot 0+b\cdot 1\\ &=b. \end{align*} Therefore \begin{align*} \omega_p(X_p) &=(P(p)\,dx_p+Q(p)\,dy_p)(X_p)\\ &=P(p)\,dx_p(X_p)+Q(p)\,dy_p(X_p)\\ &=P(p)a+Q(p)b. \end{align*} Thus a one-form on the plane evaluates a tangent vector by pairing its coefficient functions at $p$ with the coordinate components of that vector. [/example] In this example, the coefficients look like the components of a vector field, but their transformation law is covariant rather than vectorial. On open subsets of $\mathbb R^n$ the coordinate basis hides this distinction, while the notation $dx_i$ keeps track of the fact that forms eat tangent vectors. ## The Graded Algebra Of Forms Once forms can vary over $U$, the next question is how much of the exterior algebra survives point by point. The wedge product should combine a $k$-form and an $\ell$-form into a $(k+\ell)$-form, multiply coefficients as functions, and preserve the sign rules from alternating algebra. [definition: Differential Form Algebra] For an [open set](/page/Open%20Set) $U\subset\mathbb R^n$, the graded [vector space](/page/Vector%20Space) of smooth differential forms is \begin{align*} \Omega^*(U)=\bigoplus_{k=0}^{n}\Omega^k(U), \end{align*} equipped with the product defined by \begin{align*} (\alpha\wedge\beta)_p=\alpha_p\wedge\beta_p \end{align*} for homogeneous forms $\alpha$ and $\beta$. [/definition] The word homogeneous means that a form has a single degree. General elements of $\Omega^*(U)$ are finite sums of homogeneous pieces, and the wedge product is extended across such sums by bilinearity. Once forms of different degrees share one algebra, ordinary commutativity is no longer the right expectation. Swapping two blocks of alternating arguments introduces a sign determined by how many pairwise interchanges are required, so the algebra needs a precise graded commutativity law rather than a usual commutativity law. The algebra of smooth forms is graded-commutative: if $\alpha$ has degree $k$ and $\beta$ has degree $\ell$, then \begin{align*} \alpha\wedge\beta=(-1)^{k\ell}\beta\wedge\alpha. \end{align*} The sign is forced by alternation: interchanging a block of $k$ inputs with a block of $\ell$ inputs requires $k\ell$ pairwise swaps. A genuinely commutative product would identify $dx_i\wedge dx_j$ with $dx_j\wedge dx_i$, while alternation requires these two forms to be negatives. The sign has immediate consequences: if $\alpha$ has odd degree, then $\alpha\wedge\alpha=-\alpha\wedge\alpha$, so $\alpha\wedge\alpha=0$, whereas even-degree forms need not square to zero. [example: Wedge Product In Coordinates] Let $U\subset\mathbb R^3$, let $\alpha=f\,dx+g\,dy$, and let $\beta=h\,dy+r\,dz$ with $f,g,h,r\in C^\infty(U)$. We compute their wedge product by expanding the two sums and moving smooth coefficients outside the wedge product: \begin{align*} \alpha\wedge\beta &=(f\,dx+g\,dy)\wedge(h\,dy+r\,dz)\\ &=(f\,dx)\wedge(h\,dy)+(f\,dx)\wedge(r\,dz)+(g\,dy)\wedge(h\,dy)+(g\,dy)\wedge(r\,dz)\\ &=fh\,dx\wedge dy+fr\,dx\wedge dz+gh\,dy\wedge dy+gr\,dy\wedge dz. \end{align*} The repeated factor vanishes because a $1$-form wedges with itself to zero: \begin{align*} dy\wedge dy&=-(dy\wedge dy),\\ 2\,dy\wedge dy&=0,\\ dy\wedge dy&=0. \end{align*} Substituting this into the expansion gives \begin{align*} \alpha\wedge\beta &=fh\,dx\wedge dy+fr\,dx\wedge dz+gh\cdot 0+gr\,dy\wedge dz\\ &=fh\,dx\wedge dy+fr\,dx\wedge dz+gr\,dy\wedge dz. \end{align*} Thus the wedge product keeps exactly the coordinate $2$-form terms with distinct coordinate covectors, with signs determined by their order. [/example] This example illustrates why alternating signs are not decoration. The term $g h\,dy\wedge dy$ vanishes because a $2$-form changes sign when its two inputs are exchanged, forcing repeated coordinate covectors to cancel. ## Top Forms And The Standard Volume Element Which differential forms represent the objects integrated over $n$-dimensional regions? In ordinary multivariable calculus, volume integrals use a density such as $f(x)\,dx_1\cdots dx_n$. In exterior calculus on an oriented [open set](/page/Open%20Set) of $\mathbb R^n$, the corresponding object is an $n$-form. [definition: Standard Volume Form] Let $U\subset\mathbb R^n$ be open with standard coordinates $(x_1,\dots,x_n)$. The standard volume form on $U$ is \begin{align*} \operatorname{vol}=dx_1\wedge\cdots\wedge dx_n\in\Omega^n(U). \end{align*} [/definition] The standard volume form is tied to the usual orientation of $\mathbb R^n$. Reversing orientation changes the sign of the preferred volume form, a convention that will matter when integration over oriented domains is introduced. Top forms are therefore the exterior-calculus replacement for signed volume densities. The form $dx_1\wedge\cdots\wedge dx_n$ is tied to the chosen coordinates and the standard orientation of $\mathbb R^n$; a general manifold needs an orientation before a comparable signed volume form can be chosen. This is exactly the feature that makes top forms suitable for integration: under a change of variables, the coefficient of a pulled-back top form picks up the determinant of the Jacobian matrix, including its sign. [example: Area Forms In The Plane] Let $U\subset\mathbb R^2$ and let $\omega=f\,dx\wedge dy$. Fix $p\in U$, and write two tangent vectors in the coordinate frame as \begin{align*} a&=a_1\,\partial_x|_p+a_2\,\partial_y|_p,\\ b&=b_1\,\partial_x|_p+b_2\,\partial_y|_p. \end{align*} We compute $\omega_p(a,b)$ from the defining formula for the wedge of two $1$-forms: \begin{align*} (dx\wedge dy)_p(a,b) &=dx_p(a)\,dy_p(b)-dx_p(b)\,dy_p(a). \end{align*} The coordinate covectors are dual to the coordinate vectors, so \begin{align*} dx_p(a)&=dx_p(a_1\partial_x|_p+a_2\partial_y|_p) =a_1dx_p(\partial_x|_p)+a_2dx_p(\partial_y|_p) =a_1,\\ dy_p(a)&=dy_p(a_1\partial_x|_p+a_2\partial_y|_p) =a_1dy_p(\partial_x|_p)+a_2dy_p(\partial_y|_p) =a_2, \end{align*} and similarly $dx_p(b)=b_1$ and $dy_p(b)=b_2$. Substituting these four values gives \begin{align*} (dx\wedge dy)_p(a,b) &=a_1b_2-b_1a_2\\ &=a_1b_2-a_2b_1\\ &=\det\begin{pmatrix} a_1 & b_1\\ a_2 & b_2 \end{pmatrix}. \end{align*} Therefore \begin{align*} \omega_p(a,b) &=\bigl(f(p)(dx\wedge dy)_p\bigr)(a,b)\\ &=f(p)(dx\wedge dy)_p(a,b)\\ &=f(p)\det\begin{pmatrix} a_1 & b_1\\ a_2 & b_2 \end{pmatrix}. \end{align*} Thus $dx\wedge dy$ measures the signed area of the parallelogram spanned by the two coordinate vectors, and the coefficient $f(p)$ scales that oriented area at the point $p$. [/example] [illustration:oriented-area-form-plane] The determinant interpretation explains why top forms measure oriented volume. The coefficient $f$ scales the standard oriented area or volume element point by point. [example: Two Forms And Vector Fields In Three Dimensions] Let $p\in U$ and write \begin{align*} a&=a_1\partial_x|_p+a_2\partial_y|_p+a_3\partial_z|_p,\\ b&=b_1\partial_x|_p+b_2\partial_y|_p+b_3\partial_z|_p. \end{align*} The coordinate covectors are dual to the coordinate vectors, so by linearity \begin{align*} dx_p(a)=a_1,\quad dy_p(a)=a_2,\quad dz_p(a)=a_3, \end{align*} and similarly \begin{align*} dx_p(b)=b_1,\quad dy_p(b)=b_2,\quad dz_p(b)=b_3. \end{align*} Using $(\eta\wedge\theta)(a,b)=\eta(a)\theta(b)-\eta(b)\theta(a)$ for $1$-forms $\eta,\theta$, we get \begin{align*} (dy\wedge dz)_p(a,b) &=dy_p(a)dz_p(b)-dy_p(b)dz_p(a)\\ &=a_2b_3-b_2a_3\\ &=a_2b_3-a_3b_2,\\ (dz\wedge dx)_p(a,b) &=dz_p(a)dx_p(b)-dz_p(b)dx_p(a)\\ &=a_3b_1-b_3a_1\\ &=a_3b_1-a_1b_3,\\ (dx\wedge dy)_p(a,b) &=dx_p(a)dy_p(b)-dx_p(b)dy_p(a)\\ &=a_1b_2-b_1a_2\\ &=a_1b_2-a_2b_1. \end{align*} Therefore \begin{align*} \omega_p(a,b) &=\bigl(P(p)\,dy\wedge dz+Q(p)\,dz\wedge dx+R(p)\,dx\wedge dy\bigr)_p(a,b)\\ &=P(p)(a_2b_3-a_3b_2)+Q(p)(a_3b_1-a_1b_3)+R(p)(a_1b_2-a_2b_1). \end{align*} On the other hand, for \begin{align*} v_p=P(p)\partial_x|_p+Q(p)\partial_y|_p+R(p)\partial_z|_p, \end{align*} the standard scalar triple product with columns $v_p,a,b$ is \begin{align*} \det\begin{pmatrix} P(p) & a_1 & b_1\\ Q(p) & a_2 & b_2\\ R(p) & a_3 & b_3 \end{pmatrix} &=P(p)\det\begin{pmatrix}a_2&b_2\\a_3&b_3\end{pmatrix} -Q(p)\det\begin{pmatrix}a_1&b_1\\a_3&b_3\end{pmatrix} +R(p)\det\begin{pmatrix}a_1&b_1\\a_2&b_2\end{pmatrix}\\ &=P(p)(a_2b_3-b_2a_3)-Q(p)(a_1b_3-b_1a_3)+R(p)(a_1b_2-b_1a_2)\\ &=P(p)(a_2b_3-a_3b_2)+Q(p)(a_3b_1-a_1b_3)+R(p)(a_1b_2-a_2b_1). \end{align*} Thus $\omega_p(a,b)$ is exactly the scalar triple product of $v_p$, $a$, and $b$, so the coefficients of the $2$-form are the components of the vector field associated to $\omega$ by the standard orientation of $\mathbb R^3$. [/example] This is the exterior-calculus form of flux. The coefficients $(P,Q,R)$ are the components that classical vector calculus places in a vector field, while the form itself is the object naturally integrated over oriented surfaces. ## Contraction By A Vector Field How does a vector field interact with a differential form before any differentiation is introduced? A vector field supplies one tangent vector at each point, so it can be inserted into the first slot of a $k$-form. The resulting operation lowers degree by one and is called contraction or interior product. Let $v$ be a smooth vector field on $U$, written in coordinates as \begin{align*} v=\sum_{i=1}^n v_i\partial_{x_i}. \end{align*} [definition: Interior Product] Let $v$ be a smooth vector field on an [open set](/page/Open%20Set) $U\subset\mathbb R^n$. For $\omega\in\Omega^k(U)$ with $k\ge 1$, the interior product $\iota_v\omega\in\Omega^{k-1}(U)$ is given by \begin{align*} (\iota_v\omega)_p(w_1,\dots,w_{k-1})=\omega_p(v_p,w_1,\dots,w_{k-1}). \end{align*} For $f\in\Omega^0(U)$, define $\iota_v f=0$. [/definition] Contraction is algebraic rather than differential: it uses the values of $v$ and $\omega$ at the same point, with no derivatives of either object. Smoothness of $\iota_v\omega$ comes from the smooth coefficient functions of $v$ and $\omega$. [example: Contracting The Standard Volume Form] Let $U\subset\mathbb R^3$, let $v=P\partial_x+Q\partial_y+R\partial_z$, and put \begin{align*} \operatorname{vol}=dx\wedge dy\wedge dz. \end{align*} We compute the $2$-form $\iota_v\operatorname{vol}$ by evaluating it on the coordinate pairs. At a point $p\in U$, \begin{align*} (\iota_v\operatorname{vol})_p(\partial_y|_p,\partial_z|_p) &=\operatorname{vol}_p(v_p,\partial_y|_p,\partial_z|_p)\\ &=\operatorname{vol}_p(P(p)\partial_x|_p+Q(p)\partial_y|_p+R(p)\partial_z|_p,\partial_y|_p,\partial_z|_p)\\ &=P(p)\operatorname{vol}_p(\partial_x|_p,\partial_y|_p,\partial_z|_p) +Q(p)\operatorname{vol}_p(\partial_y|_p,\partial_y|_p,\partial_z|_p)\\ &\qquad +R(p)\operatorname{vol}_p(\partial_z|_p,\partial_y|_p,\partial_z|_p)\\ &=P(p)\cdot 1+Q(p)\cdot 0+R(p)\cdot 0\\ &=P(p). \end{align*} The two zero terms vanish because an alternating form is zero when two arguments are equal. Similarly, \begin{align*} (\iota_v\operatorname{vol})_p(\partial_x|_p,\partial_z|_p) &=\operatorname{vol}_p(v_p,\partial_x|_p,\partial_z|_p)\\ &=P(p)\operatorname{vol}_p(\partial_x|_p,\partial_x|_p,\partial_z|_p) +Q(p)\operatorname{vol}_p(\partial_y|_p,\partial_x|_p,\partial_z|_p)\\ &\qquad +R(p)\operatorname{vol}_p(\partial_z|_p,\partial_x|_p,\partial_z|_p)\\ &=P(p)\cdot 0+Q(p)(-1)+R(p)\cdot 0\\ &=-Q(p), \end{align*} because $(\partial_y,\partial_x,\partial_z)$ is obtained from $(\partial_x,\partial_y,\partial_z)$ by one transposition. Finally, \begin{align*} (\iota_v\operatorname{vol})_p(\partial_x|_p,\partial_y|_p) &=\operatorname{vol}_p(v_p,\partial_x|_p,\partial_y|_p)\\ &=P(p)\operatorname{vol}_p(\partial_x|_p,\partial_x|_p,\partial_y|_p) +Q(p)\operatorname{vol}_p(\partial_y|_p,\partial_x|_p,\partial_y|_p)\\ &\qquad +R(p)\operatorname{vol}_p(\partial_z|_p,\partial_x|_p,\partial_y|_p)\\ &=P(p)\cdot 0+Q(p)\cdot 0+R(p)\cdot 1\\ &=R(p), \end{align*} because $(\partial_z,\partial_x,\partial_y)$ is obtained from $(\partial_x,\partial_y,\partial_z)$ by two transpositions. These three values are exactly the coefficients of the coordinate $2$-form expansion, since $dy\wedge dz$ evaluates to $1$ on $(\partial_y,\partial_z)$, $dx\wedge dz$ evaluates to $1$ on $(\partial_x,\partial_z)$, and $dx\wedge dy$ evaluates to $1$ on $(\partial_x,\partial_y)$. Therefore \begin{align*} \iota_v(dx\wedge dy\wedge dz) &=P\,dy\wedge dz-Q\,dx\wedge dz+R\,dx\wedge dy. \end{align*} Using graded commutativity for $1$-forms, \begin{align*} dz\wedge dx&=-dx\wedge dz, \end{align*} so the same identity can be written as \begin{align*} \iota_v(dx\wedge dy\wedge dz) &=P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy. \end{align*} Thus contracting the standard volume form by $v$ produces the usual flux $2$-form whose coefficients are the coordinate components of $v$. [/example] This example connects the vector-field notation from multivariable calculus to the $2$-form notation used for flux. A vector field in three dimensions corresponds to the contraction of the standard volume form by that field. For contraction to be useful beyond single coordinate volume forms, it must interact predictably with wedge products. The obstruction is that contraction lowers degree by one, so moving it past a factor of degree $k$ should create a sign. We write $\mathfrak X(U)$ for the space of smooth vector fields on $U$. The next structural rule records exactly how contraction by an element of $\mathfrak X(U)$ distributes over a wedge product. [quotetheorem:3603] [citeproof:3603] This theorem says that $\iota_v$ is a graded derivation of degree $-1$ on the wedge algebra. The sign is $(-1)^k$ because the inserted vector must pass the whole degree-$k$ factor $\alpha$ before it can be inserted into $\beta$; the degree of $\beta$ is not what controls that move. The convention $\iota_v f=0$ for $0$-forms is also forced by this derivation viewpoint, since evaluation of a vector field on a function would not lower degree and would make the Leibniz rule for $\iota_v(fg)$ incompatible with ordinary multiplication. A further structural consequence is $\iota_v^2=0$, because inserting the same vector twice into an alternating form gives zero. It is the first example of a degree-changing operator on forms, and its sign rule is the model for the later Leibniz rule for the [exterior derivative](/theorems/1525). [example: Contracting A Product Of One Forms] Let $U\subset\mathbb R^2$, let $v=a\partial_x+b\partial_y$, and let $\alpha=P\,dx+Q\,dy$ and $\beta=R\,dx+S\,dy$, with all coefficient functions smooth on $U$. We show that \begin{align*} \iota_v(\alpha\wedge\beta)=\alpha(v)\,\beta-\beta(v)\,\alpha. \end{align*} For a $1$-form, contraction by $v$ is evaluation on $v$. Since the coordinate covectors are dual to the coordinate vector fields, \begin{align*} \iota_v\alpha &=\alpha(v)\\ &=(P\,dx+Q\,dy)(a\partial_x+b\partial_y)\\ &=P\,dx(a\partial_x+b\partial_y)+Q\,dy(a\partial_x+b\partial_y)\\ &=P\bigl(a\,dx(\partial_x)+b\,dx(\partial_y)\bigr)+Q\bigl(a\,dy(\partial_x)+b\,dy(\partial_y)\bigr)\\ &=P(a\cdot 1+b\cdot 0)+Q(a\cdot 0+b\cdot 1)\\ &=aP+bQ, \end{align*} and \begin{align*} \iota_v\beta &=\beta(v)\\ &=(R\,dx+S\,dy)(a\partial_x+b\partial_y)\\ &=R\,dx(a\partial_x+b\partial_y)+S\,dy(a\partial_x+b\partial_y)\\ &=R\bigl(a\,dx(\partial_x)+b\,dx(\partial_y)\bigr)+S\bigl(a\,dy(\partial_x)+b\,dy(\partial_y)\bigr)\\ &=R(a\cdot 1+b\cdot 0)+S(a\cdot 0+b\cdot 1)\\ &=aR+bS. \end{align*} By the *Interior Product Graded Leibniz Rule* with $\deg \alpha=1$, \begin{align*} \iota_v(\alpha\wedge\beta) &=(\iota_v\alpha)\wedge\beta+(-1)^1\alpha\wedge(\iota_v\beta)\\ &=\alpha(v)\wedge\beta-\alpha\wedge\beta(v). \end{align*} The forms $\alpha(v)$ and $\beta(v)$ are $0$-forms, so wedging with them is ordinary multiplication by functions, and a $0$-form commutes with a $1$-form. Hence \begin{align*} \iota_v(\alpha\wedge\beta) &=\alpha(v)\,\beta-\beta(v)\,\alpha. \end{align*} The minus sign is the degree-one sign: inserting $v$ into the second factor requires passing the degree-$1$ form $\alpha$. [/example] The negative sign in the last formula is the first sign a reader should associate with contraction. Inserting a vector into the second factor of a wedge product requires moving it past a degree-one form, and that movement contributes the sign. The chapter has now extended the exterior algebra from a single [vector space](/page/Vector%20Space) to open subsets of $\mathbb R^n$. The objects $\Omega^k(U)$ carry coordinate descriptions, a graded-commutative wedge product, top-degree volume forms, and the contraction operation by vector fields. The next step in exterior calculus is to add differentiation in a way that respects this graded algebra structure. After forms are allowed to vary smoothly, the natural next operation is differentiation. The [exterior derivative](/theorems/1525) extends calculus in a way that respects the graded algebra structure, and the familiar vector calculus identities will emerge as special cases of that framework. # 3. The Exterior Derivative Until now a differential form has been an algebraic object varying smoothly over an [open set](/page/Open%20Set) $U \subset \mathbb R^n$: at each point it is an alternating multilinear functional on tangent vectors. The next problem is to differentiate such objects in a way that respects both the smooth coefficients and the alternating wedge product. The [exterior derivative](/theorems/1525) is the operator that does this, and its defining feature is that it turns the separate operations of gradient, curl, and divergence into one graded differential. This chapter introduces $d$, proves its coordinate formula, and explains why applying it twice gives zero. That single identity produces the de Rham complex and the distinction between closed and exact forms. The final section translates the formalism back into the vector calculus operators on $\mathbb R^3$. ## Definition and Uniqueness What properties should a derivative of forms have if it is to extend the usual differential of a function and remain compatible with wedge products? The answer is surprisingly rigid: once we demand linearity, the graded product rule, nilpotence, and the usual formula on functions, there is no freedom left. [definition: Exterior Derivative Axioms] Let $U \subset \mathbb R^n$ be open. An [exterior derivative](/theorems/1525) on $U$ is a family of maps \begin{align*} d : \Omega^k(U) \to \Omega^{k+1}(U), \qquad k \ge 0, \end{align*} satisfying the following conditions: - $d(a\alpha + b\beta) = a\,d\alpha + b\,d\beta$ for all $a,b \in \mathbb R$ and all $\alpha,\beta \in \Omega^k(U)$. - If $f \in \Omega^0(U) = C^\infty(U)$, then \begin{align*} df = \sum_{i=1}^n \frac{\partial f}{\partial x_i}\,dx_i. \end{align*} - If $\alpha \in \Omega^k(U)$ and $\beta \in \Omega^\ell(U)$, then \begin{align*} d(\alpha \wedge \beta) = d\alpha \wedge \beta + (-1)^k \alpha \wedge d\beta. \end{align*} - $d(d\omega)=0$ for every $\omega \in \Omega^k(U)$. [/definition] The sign in the product rule is the same sign that appears whenever a degree-one operation is moved past a $k$-form. This is the first place where the grading of the exterior algebra controls an analytic formula. The definition leaves an existence-and-uniqueness question. If the exterior derivative is to be a genuine calculus operator rather than a list of separate formulas in each degree, there should be exactly one construction extending the ordinary differential of functions, compatible with wedge products, and satisfying $d^2=0$. In Euclidean coordinates this construction is unique: the coordinate one-forms $dx_i=d x_i$ are closed, and every form is built from smooth coefficients and wedges of these basis forms. Thus the [exterior derivative](/theorems/1525) is not a new choice made in each degree. It is the single extension of the differential of functions that respects the alternating algebra. The nilpotence axiom is being included here as part of the characterisation; with only linearity, the formula on functions, and the graded Leibniz rule, the values on the coordinate one-forms would not yet be forced in the same way. In the uniqueness argument, the step $d(dx_i)=d(d x_i)=0$ is exactly where this axiom enters. This is a coordinate theorem on open subsets of $\mathbb R^n$, where the global one-forms $dx_1,\dots,dx_n$ provide a fixed basis and every form has a unique expansion in that basis. On a manifold the same operator is defined by applying this formula in charts and then checking that the result is independent of the chosen chart; that independence is the naturality property of $d$ under smooth pullback. Thus the coordinate formula gives the local construction, while naturality explains why it is geometrically meaningful rather than an artefact of coordinates. ## Coordinate Formula How do we actually compute $d\omega$ when a form is written in coordinates? The axioms reduce every calculation to differentiating coefficient functions and then wedging the result with the basis form already present. Let $I=(i_1<\cdots<i_k)$ be an increasing multi-index and 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 f_I\,dx_I, \end{align*} with $f_I \in C^\infty(U)$. [quotetheorem:3564] The coordinate formula is exactly what the product rule predicts when the basis one-forms are the differentials of coordinate functions. [citeproof:3564] The practical rule is therefore simple: differentiate only the scalar coefficients, wedge the resulting one-form on the left, and then simplify signs. The reason no derivatives of the basis forms appear is that $dx_i=d x_i$ and $d^2x_i=0$, so the coordinate coframe is closed. This is a special feature of coordinate one-forms: in a general moving coframe, extra structure terms can appear because the basis one-forms themselves may have nonzero [exterior derivative](/theorems/1525). The formula should therefore be read as a coordinate calculation, not as permission to ignore the variation of every possible frame. [example: Polynomial One Form] Let $U=\mathbb R^2$ with coordinates $(x,y)$, and consider \begin{align*} \omega=x^2y\,dx+xy^2\,dy. \end{align*} Put $P=x^2y$ and $Q=xy^2$, so $\omega=P\,dx+Q\,dy$. Since $dx=d x$ and $dy=d y$, nilpotence gives $d(dx)=d^2x=0$ and $d(dy)=d^2y=0$, hence \begin{align*} d\omega &=d(P\,dx)+d(Q\,dy) \\ &=dP\wedge dx+P\,d(dx)+dQ\wedge dy+Q\,d(dy) \\ &=dP\wedge dx+dQ\wedge dy. \end{align*} The coefficient differentials are \begin{align*} dP &=\frac{\partial}{\partial x}(x^2y)\,dx+\frac{\partial}{\partial y}(x^2y)\,dy \\ &=2xy\,dx+x^2\,dy, \\ dQ &=\frac{\partial}{\partial x}(xy^2)\,dx+\frac{\partial}{\partial y}(xy^2)\,dy \\ &=y^2\,dx+2xy\,dy. \end{align*} Substituting and distributing the wedge product gives \begin{align*} d\omega &=(2xy\,dx+x^2\,dy)\wedge dx+(y^2\,dx+2xy\,dy)\wedge dy \\ &=2xy\,dx\wedge dx+x^2\,dy\wedge dx +y^2\,dx\wedge dy+2xy\,dy\wedge dy \\ &=0+x^2(-dx\wedge dy)+y^2\,dx\wedge dy+0 \\ &=(y^2-x^2)\,dx\wedge dy. \end{align*} Thus for this polynomial one-form, $d$ keeps exactly the antisymmetric first-order part $\partial_xQ-\partial_yP=y^2-x^2$. [/example] The preceding computation is mechanical, but it exposes the cancellation built into the wedge product: symmetric first-order data is discarded and antisymmetric first-order data remains. The next example is a boundary case in top degree. It shows that a coefficient can vary without contributing to $d\omega$ when its differential points in a direction already present in the wedge. [example: Repeated Differentials Kill Tangential Variation] Let $U=\mathbb R^3$ with coordinates $(x,y,z)$, and let \begin{align*} \omega=f(x,y,z)\,dy\wedge dz. \end{align*} We compute $d\omega$ and show that only the derivative of $f$ in the missing $x$-direction survives. Since $dy=d y$ and $dz=d z$, nilpotence gives $d(dy)=0$ and $d(dz)=0$. Hence the graded product rule gives \begin{align*} d\omega &=d(f\,dy\wedge dz) \\ &=df\wedge dy\wedge dz+f\,d(dy\wedge dz) \\ &=df\wedge dy\wedge dz+f\bigl(d(dy)\wedge dz-dy\wedge d(dz)\bigr) \\ &=df\wedge dy\wedge dz. \end{align*} Now expand the coefficient differential: \begin{align*} df &=\frac{\partial f}{\partial x}\,dx+\frac{\partial f}{\partial y}\,dy+\frac{\partial f}{\partial z}\,dz. \end{align*} Substituting and distributing the wedge product gives \begin{align*} d\omega &=\left(\frac{\partial f}{\partial x}\,dx+\frac{\partial f}{\partial y}\,dy+\frac{\partial f}{\partial z}\,dz\right)\wedge dy\wedge dz \\ &=\frac{\partial f}{\partial x}\,dx\wedge dy\wedge dz +\frac{\partial f}{\partial y}\,dy\wedge dy\wedge dz +\frac{\partial f}{\partial z}\,dz\wedge dy\wedge dz \\ &=\frac{\partial f}{\partial x}\,dx\wedge dy\wedge dz+0+0 \\ &=\frac{\partial f}{\partial x}\,dx\wedge dy\wedge dz. \end{align*} For example, if $f(y,z)=y^2+z^2$, then $\partial f/\partial x=0$, so \begin{align*} d\bigl((y^2+z^2)\,dy\wedge dz\bigr)=0. \end{align*} Thus a coefficient can vary in the $y$- and $z$-directions without contributing to $d\omega$, because those directions are already present in the wedge. [/example] ## The De Rham Complex What new structure appears once applying $d$ twice gives zero? The spaces of forms become a cochain complex: each image lies inside the next kernel, so we can compare forms killed by $d$ with forms obtained by applying $d$. For an [open set](/page/Open%20Set) $U \subset \mathbb R^n$, the de Rham complex is \begin{align*} 0 \longrightarrow \Omega^0(U) \xrightarrow{d} \Omega^1(U) \xrightarrow{d} \cdots \xrightarrow{d} \Omega^n(U) \longrightarrow 0. \end{align*} The complex stops at $\Omega^n(U)$ because there are no nonzero alternating $(n+1)$-forms on $\mathbb R^n$. The cancellation $d^{k+1}\circ d^k=0$ is the analytic heart of the [exterior derivative](/theorems/1525). This cancellation is the reason closed and exact forms are compatible notions, but it should not be read as saying that every closed form has a potential. It says only that the image of one [exterior derivative](/theorems/1525) lies in the kernel of the next, so exactness implies closedness; whether the reverse implication holds is a topological question. The proof also shows why the smoothness hypothesis matters: the cancellation uses equality of mixed partial derivatives, and for merely $C^1$ coefficients those second derivatives need not exist as ordinary continuous functions. The cohomological consequence is that the quotient \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*} measures the gap between closed and exact $k$-forms. If this quotient is zero, closed $k$-forms have global potentials; if it is nonzero, the obstruction comes from the global shape of $U$. The next two definitions name the kernel and image at a fixed degree. [definition: Closed Form] Let $U \subset \mathbb R^n$ be open. A form $\omega \in \Omega^k(U)$ is closed if \begin{align*} d\omega=0. \end{align*} [/definition] Closedness is a differential condition: it says the form has no [exterior derivative](/theorems/1525). It is therefore a local equation on the coefficients of the form. Exactness is stronger because it asks for a global primitive whose [exterior derivative](/theorems/1525) is the given form. [definition: Exact Form] Let $U \subset \mathbb R^n$ be open. A form $\omega \in \Omega^k(U)$ is exact if there exists $\eta \in \Omega^{k-1}(U)$ such that \begin{align*} \omega=d\eta. \end{align*} [/definition] Exactness is a potential condition: the form is the [exterior derivative](/theorems/1525) of a form one degree lower. Nilpotence immediately turns every potential into a closed form, so the implication from exact to closed is formal. The difficult direction, when it holds, is constructing the potential from the differential equation $d\omega=0$. [quotetheorem:3565] This is the basic structural consequence of $d^2=0$. [citeproof:3565] The converse is a deeper question and depends on the topology of the domain. On a contractible [open set](/page/Open%20Set), closed forms are locally expected to have potentials; on domains with holes, a closed form can fail to be exact. [illustration:angle-form-punctured-plane] [example: Angle Form] Let $U=\mathbb R^2\setminus\{0\}$ and define \begin{align*} \omega=\frac{-y\,dx+x\,dy}{x^2+y^2}. \end{align*} We show that $\omega$ is closed but not exact on $U$. Write $\omega=P\,dx+Q\,dy$, where \begin{align*} P=-\frac{y}{x^2+y^2}, \qquad Q=\frac{x}{x^2+y^2}. \end{align*} Using the quotient rule, \begin{align*} \frac{\partial Q}{\partial x} &=\frac{\partial}{\partial x}\left(\frac{x}{x^2+y^2}\right) \\ &=\frac{(x^2+y^2)\cdot 1-x\cdot 2x}{(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 similarly \begin{align*} \frac{\partial P}{\partial y} &=\frac{\partial}{\partial y}\left(-\frac{y}{x^2+y^2}\right) \\ &=-\frac{(x^2+y^2)\cdot 1-y\cdot 2y}{(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*} d\omega &=d(P\,dx+Q\,dy) \\ &=dP\wedge dx+dQ\wedge 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 \\ &=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 \\ &=0. \end{align*} Thus $\omega$ is closed. To prove that $\omega$ is not exact, suppose for contradiction that $\omega=dF$ for some smooth function $F:U\to\mathbb R$. Then \begin{align*} dF=\frac{\partial F}{\partial x}\,dx+\frac{\partial F}{\partial y}\,dy=P\,dx+Q\,dy, \end{align*} so \begin{align*} \frac{\partial F}{\partial x}=P=-\frac{y}{x^2+y^2}, \qquad \frac{\partial F}{\partial y}=Q=\frac{x}{x^2+y^2}. \end{align*} Let $\gamma(t)=(\cos t,\sin t)$ for $0\le t\le 2\pi$. Since $\cos^2t+\sin^2t=1$, along this circle we have \begin{align*} \frac{\partial F}{\partial x}(\gamma(t))=-\sin t, \qquad \frac{\partial F}{\partial y}(\gamma(t))=\cos t. \end{align*} By the chain rule, \begin{align*} \frac{d}{dt}F(\gamma(t)) &=\frac{\partial F}{\partial x}(\gamma(t))\frac{d}{dt}(\cos t) +\frac{\partial F}{\partial y}(\gamma(t))\frac{d}{dt}(\sin t) \\ &=(-\sin t)(-\sin t)+(\cos t)(\cos t) \\ &=\sin^2t+\cos^2t \\ &=1. \end{align*} Integrating from $0$ to $2\pi$ gives \begin{align*} F(\gamma(2\pi))-F(\gamma(0)) &=\int_0^{2\pi}\frac{d}{dt}F(\gamma(t))\,dt \\ &=\int_0^{2\pi}1\,dt \\ &=2\pi. \end{align*} But $\gamma(0)=(1,0)=\gamma(2\pi)$, so the left-hand side is \begin{align*} F(\gamma(2\pi))-F(\gamma(0))=F(1,0)-F(1,0)=0, \end{align*} a contradiction. Hence $\omega$ is closed but not exact, showing that exactness can fail for global topological reasons even when $d\omega=0$. [/example] ## Classical Operators Recovered Where did the familiar gradient, curl, and divergence go? They are all instances of the same operator $d$, after making the usual Euclidean identifications between scalar functions, vector fields, one-forms, two-forms, and three-forms. [remark: Metric Identification] The [exterior derivative](/theorems/1525) produces forms, not vector fields. On Euclidean space, the inner product identifies a vector field $V=(V_1,V_2,V_3)$ with the one-form $V_1dx+V_2dy+V_3dz$, and the orientation identifies a vector field $V$ with the two-form \begin{align*} V_1\,dy\wedge dz+V_2\,dz\wedge dx+V_3\,dx\wedge dy. \end{align*} These identifications are what turn $d$ into the classical vector calculus operators. [/remark] The following formulas are written on an [open set](/page/Open%20Set) $U\subset\mathbb R^3$ with standard coordinates $(x,y,z)$ and standard orientation $dx\wedge dy\wedge dz$. The choice of orientation fixes the signs in the identification between vector fields and two-forms. With a different orientation, the displayed divergence identification would acquire the corresponding sign change. [quotetheorem:3566] The theorem is mostly bookkeeping, but it explains why the signs in curl have exactly their familiar pattern. [citeproof:3566] This recovers the vector calculus chain \begin{align*} \text{functions} \xrightarrow{\operatorname{grad}} \text{vector fields} \xrightarrow{\operatorname{curl}} \text{vector fields} \xrightarrow{\operatorname{div}} \text{functions} \end{align*} from the de Rham complex in dimension three. The identities $\operatorname{curl}\nabla f=0$ and $\operatorname{div}(\operatorname{curl} V)=0$ are both the single statement $d^2=0$ written through these identifications. With differentiation available, the next structural question is how forms change under smooth maps between spaces. Pullbacks answer that question by transporting forms through coordinate changes, preserving wedge products and encoding the chain rule in geometric form. # 4. Pullbacks The previous chapters built differential forms from alternating multilinear algebra and introduced the [exterior derivative](/theorems/1525) as the operation that differentiates forms. Pullbacks answer the next structural question: if a smooth map sends one space into another, how do forms on the target become forms on the source? This chapter develops that operation, proves its compatibility with wedge products and exterior derivatives, and explains why Jacobian determinants appear in change-of-variables formulae. ## Pulling Forms Back Along Smooth Maps What should it mean to transport a covector or a $k$-form backward along a smooth map $F: U \to V$? A tangent vector at a point $p \in U$ is pushed forward by the differential $dF_p$, so a form on $V$ can be evaluated after all tangent vectors have been pushed forward to $T_{F(p)}V$. This reversal of direction is the defining feature of pullback. [definition: Pullback Of A Differential Form] Let $U \subset \mathbb R^m$ and $V \subset \mathbb R^n$ be open sets, let $F:U\to V$ be smooth, and let $\omega \in \Omega^k(V)$. The pullback operator is the map \begin{align*} F^*: \Omega^k(V) &\longrightarrow \Omega^k(U), & \omega &\longmapsto F^*\omega, \end{align*} where $F^*\omega \in \Omega^k(U)$ is defined 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 $p\in U$ and $v_1,\dots,v_k\in T_pU$. [/definition] The formula is forced by a concrete obstruction: a $k$-form on $V$ cannot be evaluated directly on tangent vectors based at $p\in U$, because its inputs must lie in $T_{F(p)}V$. The differential $dF_p:T_pU\to T_{F(p)}V$ is precisely the device that moves those inputs to the correct tangent space, so pullback is evaluation after pushing all vectors forward. For $0$-forms, which are smooth functions, the same convention gives $F^*f=f\circ F$; the definition is therefore the higher-degree version of ordinary substitution. The reversal of direction is essential: $F$ sends points forward from $U$ to $V$, while $F^*$ sends forms backward from $V$ to $U$. [example: Pullback Along A Projection] Let $F:\mathbb R^2\to\mathbb R$ be given by $F(s,t)=s^2+t$, let $y$ be the coordinate on $\mathbb R$, and let $\omega=y\,dy$. We compute the pullback by evaluating it on an arbitrary tangent vector \begin{align*} X&=A\,\partial_s+B\,\partial_t\in T_{(s,t)}\mathbb R^2. \end{align*} Since \begin{align*} dF_{(s,t)}(X) &=\frac{\partial F}{\partial s}(s,t)A\,\partial_y+\frac{\partial F}{\partial t}(s,t)B\,\partial_y \\ &=(2sA+B)\,\partial_y, \end{align*} the definition of pullback gives \begin{align*} (F^*\omega)_{(s,t)}(X) &=\omega_{F(s,t)}\bigl(dF_{(s,t)}(X)\bigr) \\ &=(y\,dy)_{s^2+t}\bigl((2sA+B)\partial_y\bigr) \\ &=(s^2+t)\,dy\bigl((2sA+B)\partial_y\bigr) \\ &=(s^2+t)(2sA+B). \end{align*} On the other hand, \begin{align*} d(s^2+t) &=\frac{\partial}{\partial s}(s^2+t)\,ds+\frac{\partial}{\partial t}(s^2+t)\,dt \\ &=2s\,ds+dt, \end{align*} so \begin{align*} \bigl((s^2+t)(2s\,ds+dt)\bigr)_{(s,t)}(X) &=(s^2+t)\bigl(2s\,ds(X)+dt(X)\bigr) \\ &=(s^2+t)(2sA+B). \end{align*} Because the two $1$-forms agree on every tangent vector $X$, we have \begin{align*} F^*\omega &=(s^2+t)\,d(s^2+t) \\ &=(s^2+t)(2s\,ds+dt). \end{align*} Thus the coefficient $y$ is replaced by $y\circ F=s^2+t$, while the coordinate covector $dy$ is replaced by $d(y\circ F)=d(s^2+t)$. [/example] This example is the template for computations: pull back the coefficient functions, then pull back the coordinate differentials. ## Coordinate Formula For Pullbacks How can the definition be turned into a calculation in coordinates? Since every form is a sum of coefficient functions times wedge products of coordinate differentials, it suffices to know how pullback acts on those two ingredients. [quotetheorem:3570] This is the computational form of the definition: the coefficient is substituted by $F$, while each target coordinate differential is replaced by the differential of the corresponding component function of $F$. [citeproof:3570] The theorem depends on smoothness of $F$: if some component $F_j$ is not differentiable, then $d(y_j\circ F)$ may not exist and the displayed coordinate expression has no meaning as a smooth differential form. It also does not say that every form on $U$ is a pullback from $V$; for instance, a projection $\mathbb R^2\to\mathbb R$ cannot produce a form whose coefficient genuinely depends on the forgotten coordinate. What the theorem does give is a reliable computational bridge from the abstract tangent-vector definition to partial derivatives of the component functions $F_j$. That bridge is the basis for the parametrised-surface and Jacobian computations that follow. [example: Spherical Parametrisation And A Two-Form] Let \begin{align*} S(\theta,\phi) &=(\sin\phi\cos\theta,\sin\phi\sin\theta,\cos\phi) \end{align*} for $(\theta,\phi)\in(0,2\pi)\times(0,\pi)$, with target coordinates $(x,y,z)$. We compute the pullback of $dy\wedge dz$. The relevant coordinate functions are \begin{align*} y\circ S&=\sin\phi\sin\theta, & z\circ S&=\cos\phi. \end{align*} Hence \begin{align*} d(y\circ S) &=\frac{\partial}{\partial\theta}(\sin\phi\sin\theta)\,d\theta +\frac{\partial}{\partial\phi}(\sin\phi\sin\theta)\,d\phi \\ &=\sin\phi\cos\theta\,d\theta+\cos\phi\sin\theta\,d\phi, \end{align*} and \begin{align*} d(z\circ S) &=\frac{\partial}{\partial\theta}(\cos\phi)\,d\theta +\frac{\partial}{\partial\phi}(\cos\phi)\,d\phi \\ &=0\,d\theta-\sin\phi\,d\phi \\ &=-\sin\phi\,d\phi. \end{align*} Therefore \begin{align*} S^*(dy\wedge dz) &=d(y\circ S)\wedge d(z\circ S) \\ &=(\sin\phi\cos\theta\,d\theta+\cos\phi\sin\theta\,d\phi) \wedge(-\sin\phi\,d\phi) \\ &=(\sin\phi\cos\theta\,d\theta)\wedge(-\sin\phi\,d\phi) +(\cos\phi\sin\theta\,d\phi)\wedge(-\sin\phi\,d\phi) \\ &=-\sin^2\phi\cos\theta\,d\theta\wedge d\phi -\sin\phi\cos\phi\sin\theta\,d\phi\wedge d\phi \\ &=-\sin^2\phi\cos\theta\,d\theta\wedge d\phi, \end{align*} because $d\phi\wedge d\phi=0$ by alternation. This target-space form detects only the oriented $yz$-coordinate component of the parametrised tangent parallelogram. Indeed, \begin{align*} S_\theta &=(-\sin\phi\sin\theta,\sin\phi\cos\theta,0),\\ S_\phi &=(\cos\phi\cos\theta,\cos\phi\sin\theta,-\sin\phi), \end{align*} so \begin{align*} S_\theta\times S_\phi &=(-\sin^2\phi\cos\theta,-\sin^2\phi\sin\theta,-\sin\phi\cos\phi) \end{align*} and \begin{align*} |S_\theta\times S_\phi| &=\sqrt{\sin^4\phi\cos^2\theta+\sin^4\phi\sin^2\theta+\sin^2\phi\cos^2\phi} \\ &=\sqrt{\sin^2\phi(\sin^2\phi+\cos^2\phi)} \\ &=\sin\phi>0. \end{align*} Thus $S$ is regular on the parameter domain, but $S^*(dy\wedge dz)$ vanishes when $\cos\theta=0$. The form $dy\wedge dz$ measures the oriented projection of the surface element onto the $yz$-plane, not the full surface area element. [/example] The signs in this example come from the order of the wedge product, not from a separate orientation convention imposed after the calculation. This matters because changing the order of the target coordinates or of the parameters would change the sign of the pulled-back $2$-form. The computation also hints at why orientations must be tracked before one can integrate forms consistently over parametrised surfaces. ## Functoriality And Algebraic Compatibility If forms can be pulled back along maps, what happens when two maps are composed or two forms are wedged? A naive construction might choose coordinates after each change of variables and hope that the resulting formula is independent of the order of calculation. The danger is that repeated reparametrisation could introduce incompatible Jacobian factors or sign conventions. The point of the next results is that pullback behaves like substitution: it reverses composition of maps and preserves the algebraic operations on forms. [quotetheorem:3568] The smoothness hypotheses are needed because the proof uses the differentials $dF_p$, $dG_{F(p)}$, and the chain rule; without differentiability, there is no pullback of positive-degree smooth forms in this sense. The statement does not make $F^*$ invertible: if $F$ collapses directions, nonzero forms on $V$ can pull back to zero on $U$. The order on the right-hand side is reversed because the form is first pulled from $W$ to $V$, then from $V$ to $U$. This naturality is what later allows [Stokes' theorem](/theorems/1530) and de Rham cohomology to be expressed independently of a particular parametrisation. [citeproof:3568] Functoriality is why pullback is the correct operation for reparametrising forms: changing variables twice gives the same form as changing variables by the composite map. It also prevents parametrisation-dependent bookkeeping from entering later integral formulae. The next compatibility result addresses a different obstruction: a form is often built as a wedge product of simpler forms, and a useful pullback operation must not depend on whether we multiply first or substitute first. [quotetheorem:3569] This says that pullback is an algebra homomorphism from the exterior algebra of forms on $V$ to the exterior algebra of forms on $U$. The hypothesis that both forms live on the same target $V$ is part of the statement: wedge products are defined pointwise only when the factors can be evaluated on tangent vectors at the same point. The theorem does not assert that pullback preserves division, contraction, or any operation not built functorially from multilinear evaluation. Its importance is practical as well as conceptual, because coordinate forms and volume forms are wedges of $1$-forms. [citeproof:3569] Together with linearity, this theorem lets us pull back a complicated form by pulling back its factors. Without this compatibility, computations with determinants and surface elements would have to be redone from the full alternating definition each time. The final compatibility in this group asks whether pullback also respects differentiation, which is the analytic operation needed for closed forms, exact forms, and Stokes-type arguments. On open subsets of Euclidean space, the chain rule gives the key identity \begin{align*} d(F^*\omega)=F^*(d\omega) \end{align*} for smooth maps $F:U\to V$ and smooth forms $\omega$ on $V$. Smoothness is essential, since the [exterior derivative](/theorems/1525) of the pulled-back coefficients differentiates the component functions of $F$. The identity does not say that a primitive can be pulled back to every desired primitive on $U$; exactness is preserved in the forward pullback direction, but information may be lost when $F$ is not injective or has rank drop. Its forward role is substantial: it is the coordinate version of the naturality statement behind closed forms, exact forms, de Rham cohomology, and the pullback form of [Stokes' theorem](/theorems/1530). This compatibility is the mechanism behind the invariance of closed and exact forms under smooth changes of variables: if $d\omega=0$, then $d(F^*\omega)=0$, and if $\omega=d\eta$, then $F^*\omega=d(F^*\eta)$. The point is not merely computational; it says that pullback respects the differential complex of forms. That is the reason it descends to maps on de Rham cohomology in later courses. [example: Pulling A One-Form Back To A Curve] Let $\gamma:(a,b)\to\mathbb R^n$ be a regular smooth curve, and restrict to a subinterval $I\subset(a,b)$ on which $\gamma$ is embedded. Along the image arc, define \begin{align*} \alpha_{\gamma(t)}(v) &=v\cdot \frac{\dot\gamma(t)}{|\dot\gamma(t)|}. \end{align*} Regularity gives $|\dot\gamma(t)|\neq 0$, so the displayed unit tangent vector is defined. The embeddedness condition makes the assignment unambiguous: if $\gamma(t_1)=\gamma(t_2)$ inside this arc, then $t_1=t_2$, so the same image point is not assigned covectors coming from two different parameter values. Assume $\alpha$ has been extended to a smooth $1$-form on a neighbourhood of the arc. We show that \begin{align*} \gamma^*\alpha &=|\dot\gamma(t)|\,dt \end{align*} on $I$. Every tangent vector in $T_tI$ has the form $c\,\partial_t$ for some $c\in\mathbb R$. Since \begin{align*} d\gamma_t(c\,\partial_t) &=c\,\dot\gamma(t), \end{align*} the definition of pullback gives \begin{align*} (\gamma^*\alpha)_t(c\,\partial_t) &=\alpha_{\gamma(t)}\bigl(d\gamma_t(c\,\partial_t)\bigr) \\ &=\alpha_{\gamma(t)}\bigl(c\,\dot\gamma(t)\bigr) \\ &=\bigl(c\,\dot\gamma(t)\bigr)\cdot \frac{\dot\gamma(t)}{|\dot\gamma(t)|} \\ &=c\,\frac{\dot\gamma(t)\cdot\dot\gamma(t)}{|\dot\gamma(t)|} \\ &=c\,\frac{|\dot\gamma(t)|^2}{|\dot\gamma(t)|} \\ &=c\,|\dot\gamma(t)|. \end{align*} On the other hand, \begin{align*} \bigl(|\dot\gamma(t)|\,dt\bigr)_t(c\,\partial_t) &=|\dot\gamma(t)|\,dt_t(c\,\partial_t) \\ &=|\dot\gamma(t)|\,c. \end{align*} The two $1$-forms agree on every tangent vector $c\,\partial_t$, so $\gamma^*\alpha=|\dot\gamma(t)|\,dt$. Thus the usual arc-length integrand is the pullback of the unit tangent covector along the parametrised curve. Regularity supplies the nonzero velocity, while embeddedness is what makes this particular ambient covector field well-defined along the image without conflicting tangent choices. [/example] The example also explains why line integrals of $1$-forms are naturally written as integrals over the parameter interval after pullback. It separates two issues that are often blurred: pullback along $\gamma$ only needs the parametrised curve, but defining an ambient form by referring to the image may require the image to have no self-overlap with different tangent data. This distinction becomes important when curves are reparametrised or decomposed into embedded arcs before integration. ## Volume Forms And The Jacobian Determinant Where does the determinant in multivariable substitution come from? A naive guess might be that each coordinate differential contributes an independent scaling factor, but this misses the mixing of directions caused by off-diagonal partial derivatives. Alternation is exactly what removes repeated directions and records the signed volume contribution of the linearly independent ones. The determinant is therefore not an extra convention; it is the coefficient produced when the standard volume form is pulled back by a map between spaces of the same dimension. [quotetheorem:3571] The equal-dimension hypothesis is essential: if the source and target dimensions differ, the pullback of a top-degree form on the target is either not a top-degree form on the source or may vanish for degree reasons when too many covectors are pulled back to a smaller tangent space. The formula is also oriented, so the sign of $\det(JF_x)$ matters rather than only its absolute value. What it does not assert is that $F$ is locally invertible; a zero determinant simply means the pulled-back volume form vanishes at that point. This result prepares the measure-theoretic change-of-variables formula, where orientation determines whether the determinant or its absolute value appears. [citeproof:3571] This result is the differential-form version of the familiar statement that linear maps scale oriented volume by their determinant. It also explains why maps with the same coordinatewise diagonal behaviour can have different volume effects when they shear or rotate directions into one another. The following example is chosen because it has off-diagonal mixing but no change in oriented area, so it tests the misconception that nonzero off-diagonal Jacobian entries must change volume. [example: A Planar Shear Has Unit Volume Pullback] Let $F:\mathbb R^2\to\mathbb R^2$ be $F(x_1,x_2)=(x_1+ax_2,x_2)$, where $a\in\mathbb R$. The coordinate functions of $F$ are \begin{align*} y_1\circ F&=x_1+ax_2, & y_2\circ F&=x_2. \end{align*} Their first derivatives give \begin{align*} JF_x &=\begin{pmatrix} \frac{\partial}{\partial x_1}(x_1+ax_2) & \frac{\partial}{\partial x_2}(x_1+ax_2)\\ \frac{\partial}{\partial x_1}(x_2) & \frac{\partial}{\partial x_2}(x_2) \end{pmatrix} \\ &=\begin{pmatrix} 1 & a\\ 0 & 1 \end{pmatrix}, \end{align*} so \begin{align*} \det(JF_x) &=1\cdot 1-a\cdot 0 \\ &=1. \end{align*} We compute the pullback of the standard oriented area form. Since \begin{align*} d(y_1\circ F) &=d(x_1+ax_2) \\ &=\frac{\partial}{\partial x_1}(x_1+ax_2)\,dx_1 +\frac{\partial}{\partial x_2}(x_1+ax_2)\,dx_2 \\ &=dx_1+a\,dx_2 \end{align*} and \begin{align*} d(y_2\circ F) &=d(x_2) \\ &=\frac{\partial x_2}{\partial x_1}\,dx_1 +\frac{\partial x_2}{\partial x_2}\,dx_2 \\ &=0\,dx_1+1\,dx_2 \\ &=dx_2, \end{align*} we have \begin{align*} F^*(dy_1\wedge dy_2) &=d(y_1\circ F)\wedge d(y_2\circ F) \\ &=(dx_1+a\,dx_2)\wedge dx_2 \\ &=dx_1\wedge dx_2+a\,dx_2\wedge dx_2 \\ &=dx_1\wedge dx_2+a\cdot 0 \\ &=dx_1\wedge dx_2, \end{align*} where $dx_2\wedge dx_2=0$ by alternation. Thus the shear changes the shape of tangent parallelograms, but its pulled-back oriented area form is unchanged. [/example] The same calculation, with a negative determinant, records orientation reversal rather than mere change of size. This distinction is invisible if one only measures unsigned area, but it is central for [integration of differential forms](/theorems/1529). At the level of integrals, this raises a precise comparison problem. The usual change-of-variables formula uses an absolute value to count unsigned volume, while integration of top forms keeps the orientation sign. The final result identifies the determinant factor that appears when an orientation-preserving diffeomorphism is used for substitution. [quotetheorem:3554] This corollary previews the integration theory developed in Chapter 6: once compactly supported top forms are integrated over oriented domains, the identification of $dy_1\wedge\cdots\wedge dy_n$ with Lebesgue measure in the chosen positive orientation gives the displayed formula. The diffeomorphism hypothesis prevents multiplicity: if $F$ is not one-to-one, the same point of $V$ may be counted more than once by the right-hand side unless an area-formula correction is introduced. Compact support keeps the integrals finite and avoids boundary convergence issues, while orientation preservation is exactly what permits $\det(JF_x)$ rather than $|\det(JF_x)|$. The theorem does not replace the full measure-theoretic change-of-variables theorem; it explains its determinant factor from the geometry of pulled-back volume forms. [citeproof:3554] The chapter's main message is that pullback is the universal form of substitution. It acts on coefficients by composition, acts on coordinate differentials by the chain rule, respects wedge products, commutes with the [exterior derivative](/theorems/1525), and turns the Jacobian determinant into the coefficient of a pulled-back volume form. Pullbacks explain how forms transform, but integration forces the issue of sign. That is why the next chapter must address orientation, since the same geometric domain can contribute positively or negatively depending on the chosen convention. # 5. Orientation The previous chapters built differential forms as alternating objects and explained how their algebra changes under pullback. Integration of top-degree forms now forces a sign choice: the same parallelepiped can be read with either of two orientations, and changing that choice changes the sign of an integral. This chapter isolates that sign convention first for vector spaces, then for open sets and manifolds with boundary, so that the statement of Stokes theorem has a well-defined meaning. The central point is that orientation is not extra metric data. It is a coherent choice of which ordered bases count as positive. Once that choice is made on a manifold, the boundary inherits its own orientation by the outward-normal-first convention. ## Orientations of Vector Spaces The first question is how to record the difference between an ordered basis and the same basis with two vectors swapped. Determinants measure this difference, because the sign of a change-of-basis determinant is the part of the determinant that survives after ignoring volume scale. [definition: Orientation Of A Vector Space] Let $V$ be a real [vector space](/page/Vector%20Space) of dimension $n \ge 1$. Two ordered bases $e=(e_1,\dots,e_n)$ and $f=(f_1,\dots,f_n)$ of $V$ are orientation-equivalent if the matrix $A=(a_{ij})$ defined by \begin{align*} f_j=\sum_{i=1}^{n} a_{ij}e_i \end{align*} has $\det A>0$. An orientation of $V$ is an orientation-equivalence class of ordered bases. An ordered basis in the chosen class is called positively oriented. [/definition] The determinant condition is stable under composition of changes of basis, so it defines two possible classes of bases. Passing from $(e_1,e_2)$ to $(e_2,e_1)$ in a two-dimensional [vector space](/page/Vector%20Space) reverses orientation, while multiplying one basis vector by a positive scalar preserves it. [example: The Plane Has Two Orientations] Let $V=\mathbb R^2$ and let $e=(e_1,e_2)$ be the reference ordered basis. For $f=(e_2,e_1)$, the change-of-basis coefficients relative to $e$ are \begin{align*} f_1=e_2=0e_1+1e_2,\qquad f_2=e_1=1e_1+0e_2, \end{align*} so the matrix whose $j$th column gives the coordinates of $f_j$ in the basis $e$ is \begin{align*} A=\begin{pmatrix}0&1\\1&0\end{pmatrix}. \end{align*} Its determinant is \begin{align*} \det A=(0)(0)-(1)(1)=-1<0. \end{align*} Thus $(e_2,e_1)$ is not orientation-equivalent to $(e_1,e_2)$, so it represents the opposite orientation. For $g=(2e_1,3e_2)$, the coordinate equations are \begin{align*} g_1=2e_1+0e_2,\qquad g_2=0e_1+3e_2, \end{align*} hence \begin{align*} B=\begin{pmatrix}2&0\\0&3\end{pmatrix}. \end{align*} Now \begin{align*} \det B=(2)(3)-(0)(0)=6>0. \end{align*} Therefore $(2e_1,3e_2)$ is orientation-equivalent to $e$, so multiplying basis vectors by positive scalars preserves the orientation in this example. [/example] Top-degree alternating forms give another way to see the same dichotomy. A non-zero element of $\Lambda^n(V^*)$ evaluates on every ordered basis, and the sign of that number distinguishes the two orientation classes. To connect this sign test with integration, we need language for top forms that agree with the chosen orientation. A top form may be nonzero but still assign negative volume to the bases declared positive, so positivity must be defined relative to an orientation rather than as an intrinsic property of the form alone. [definition: Positive Top Form] Let $V$ be an oriented real [vector space](/page/Vector%20Space) of dimension $n$. A non-zero form $\omega\in\Lambda^n(V^*)$ is positive if \begin{align*} \omega(e_1,\dots,e_n)>0 \end{align*} for every positively oriented ordered basis $(e_1,\dots,e_n)$ of $V$. [/definition] This definition matches the geometric idea that a top form is a signed volume measurement. Reversing orientation changes which non-zero top forms are positive. [quotetheorem:3604] [citeproof:3604] This result explains why orientation can be encoded either by positive bases or by positive top-degree forms. The form viewpoint is the one that interacts directly with integration: choosing the opposite orientation multiplies every positive top form by a negative sign. Later, when forms are integrated over manifolds, this is the mechanism behind the sign change under orientation reversal. ## Orientations on Open Sets of Euclidean Space The next problem is to choose orientations at every point of a region without allowing the sign to jump from point to point. On an [open set](/page/Open%20Set) $U\subset\mathbb R^n$, each tangent space $T_xU$ is naturally identified with $\mathbb R^n$, so the standard coordinate basis gives a reference orientation everywhere. [definition: Orientation Of An Open Set] Let $U\subset\mathbb R^n$ be open. An orientation of $U$ is a choice of orientation of $T_xU$ for each $x\in U$ such that every $x_0\in U$ has a neighbourhood $W\subset U$ and a continuous $n$-form $\omega$ on $W$ satisfying $\omega_x\ne 0$ for all $x\in W$ and positive on the chosen orientation of $T_xU$ for all $x\in W$. [/definition] The continuity requirement is the analytic ingredient: it prevents a region from being assigned the standard sign on one side of a point and the opposite sign arbitrarily close by. To compare arbitrary orientations with coordinate calculations, we need a fixed reference orientation on Euclidean open sets. The coordinate vector fields provide the same ordered basis at every tangent space, so the next definition records this canonical choice before any Jacobian sign can be interpreted as preserving or reversing orientation. [definition: Standard Orientation On Euclidean Space] Let $U\subset\mathbb R^n$ be open. The standard orientation on $U$ is the orientation for which \begin{align*} \left(\frac{\partial}{\partial x_1}\Big|_x,\dots,\frac{\partial}{\partial x_n}\Big|_x\right) \end{align*} is positively oriented for every $x\in U$. [/definition] With this convention the coordinate top form $dx_1\wedge\cdots\wedge dx_n$ is positive. Reversing the order of two coordinate directions changes the sign of the form and hence reverses orientation. Once a reference orientation is fixed, the next question is whether a connected Euclidean domain can carry any more exotic orientation. The local continuity condition suggests that the comparison sign with the standard orientation is locally constant, and connectedness should then force it to be globally constant. The theorem makes this classification precise. [quotetheorem:3605] [citeproof:3605] The theorem shows that on a connected [Euclidean domain](/page/Euclidean%20Domain) there is no hidden third choice of orientation: every orientation agrees everywhere with the coordinate orientation or disagrees everywhere. This is why the sign of a Jacobian determinant controls orientation under smooth changes of variables. The next example is the basic local model for an orientation-reversing map. [example: A Change Of Variables Reversing Orientation] Let $F:\mathbb R^2\to\mathbb R^2$ be given by $F(x_1,x_2)=(x_1,-x_2)$. We show that $F$ reverses the standard orientation of $\mathbb R^2$. Writing $F_1(x_1,x_2)=x_1$ and $F_2(x_1,x_2)=-x_2$, its Jacobian matrix at any point $x$ is \begin{align*} JF_x &= \begin{pmatrix} \frac{\partial F_1}{\partial x_1} & \frac{\partial F_1}{\partial x_2}\\ \frac{\partial F_2}{\partial x_1} & \frac{\partial F_2}{\partial x_2} \end{pmatrix}\\ &= \begin{pmatrix} 1&0\\ 0&-1 \end{pmatrix}. \end{align*} Hence \begin{align*} \det JF_x=(1)(-1)-(0)(0)=-1<0. \end{align*} The same sign appears from the coordinate area form. Since pullback commutes with exterior products and is linear, \begin{align*} F^*(dx_1)&=d(x_1\circ F)=d(x_1)=dx_1,\\ F^*(dx_2)&=d(x_2\circ F)=d(-x_2)=-dx_2, \end{align*} and therefore \begin{align*} F^*(dx_1\wedge dx_2) &=F^*(dx_1)\wedge F^*(dx_2)\\ &=dx_1\wedge(-dx_2)\\ &=-\,dx_1\wedge dx_2. \end{align*} Thus $F$ sends the positive coordinate area form to its negative, so it reverses the standard orientation. [/example] ## Manifolds With Boundary in Euclidean Space Integration over curves, surfaces, and higher-dimensional domains needs spaces that locally look like open sets, except that some points are allowed to lie on an edge. The local model is a half-space, because near an ordinary boundary point there is exactly one missing side. For $k\ge 1$, set \begin{align*} \mathbb H^k=\{y\in\mathbb R^k:y_k\ge 0\}, \qquad \partial\mathbb H^k=\{y\in\mathbb R^k:y_k=0\}. \end{align*} [definition: Manifold With Boundary In Euclidean Space] Let $1\le k\le n$. A $k$-dimensional smooth manifold with boundary in $\mathbb R^n$ is a subset $M\subset\mathbb R^n$ such that for every $p\in M$ there are an open neighbourhood $W\subset\mathbb R^n$ of $p$, an [open set](/page/Open%20Set) $O\subset\mathbb R^k$, and a diffeomorphism of local models \begin{align*} \psi:W\cap M\to O\cap\mathbb H^k \end{align*} whose inverse is a smooth parametrisation of rank $k$. [/definition] The points mapping to $y_k>0$ are interior points of the manifold, while points mapping to $y_k=0$ are boundary points. To use this distinction globally, we need a chart-independent subset of $M$ that collects exactly the points lying on the half-space face. The next definition names that subset and separates it from the interior. [definition: Boundary Of A Manifold With Boundary] Let $M$ be a $k$-dimensional smooth manifold with boundary. The boundary $\partial M$ is the set of points $p\in M$ whose image in a boundary chart lies in $\partial\mathbb H^k$. The interior $M^\circ$ is $M\setminus\partial M$. [/definition] This definition packages the familiar examples: an interval has two boundary points, a disk has a boundary circle, and a hemisphere has a boundary equator. The structural question is whether this boundary is itself a smooth manifold of the expected dimension, rather than merely a distinguished subset. The theorem supplies that geometric regularity, which is needed before forms can be restricted to the boundary and integrated there. [quotetheorem:3606] [citeproof:3606] This result justifies treating $\partial M$ as a legitimate smooth space in its own right, with dimension one less than $M$. The last phrase is important for Stokes theorem: a boundary has no further boundary in the smooth manifold-with-boundary category. Corners are handled separately as a chain-level extension, where the same sign rule makes vertex contributions cancel. The next issue is the algebraic shadow of this geometry. Since integration over boundaries will be paired with exterior differentiation, the boundary operation itself should satisfy the same kind of nilpotence as $d^2=0$. The following theorem records this as the smooth principle that a boundary has no boundary. [quotetheorem:3607] [citeproof:3607] This theorem is the smooth version of the principle that a boundary has no boundary. It is essential for Stokes theorem to be compatible with applying the boundary operation twice: the formal identity behind the theory is $\partial^2=0$. For domains with corners the same cancellation is still present, but it is not expressed by taking the boundary of a smooth manifold with boundary. ## The Boundary Orientation The sign in Stokes theorem depends on how the orientation of a manifold tells us to orient its boundary. The convention used here is outward-normal-first: first point out of the manifold, then read the remaining ordered basis along the boundary. [definition: Orientation Of A Manifold With Boundary] Let $M$ be a $k$-dimensional smooth manifold with boundary. An orientation of $M$ is a choice of orientation of $T_pM$ for each $p\in M$ such that every $p\in M$ has a boundary chart $\phi:O\cap\mathbb H^k\to M$ whose derivative carries the standard orientation of $\mathbb R^k$ to the chosen orientation of $T_{\phi(y)}M$ for all $y\in O\cap\mathbb H^k$. [/definition] In a half-space chart with $y_k\ge 0$, an outward-pointing tangent vector is one whose $y_k$-component is negative. This agrees with the usual outward normal along the boundary of a planar region or a surface in $\mathbb R^3$. To integrate over $\partial M$, however, the boundary needs its own orientation, not just an ambient orientation on $M$. The convention must say how an outward transverse direction and a boundary basis combine to recover the positive basis of $T_pM$. The next definition makes that transfer of orientation explicit. [definition: Induced Boundary Orientation] Let $M$ be an oriented $k$-dimensional smooth manifold with boundary, and let $p\in\partial M$. An ordered basis $(v_1,\dots,v_{k-1})$ of $T_p\partial M$ is positive for the induced boundary orientation if \begin{align*} (\nu_p,v_1,\dots,v_{k-1}) \end{align*} is a positive ordered basis of $T_pM$ for an outward-pointing vector $\nu_p\in T_pM$ transverse to $T_p\partial M$. [/definition] The convention would be unusable if different outward transverse vectors or boundary charts produced different orientations on $\partial M$. The obstruction is choice-dependence: Stokes theorem needs one boundary orientation attached to $M$, not a family of orientations depending on auxiliary normals. [quotetheorem:3608] [citeproof:3608] Well-definedness is what makes the boundary orientation intrinsic rather than a feature of a particular normal vector field or chart. The convention is asymmetric: using inward-normal-first would reverse every boundary orientation and would change the sign in Stokes theorem. The following planar example fixes the sign convention in the most familiar case. [illustration:unit-square-boundary-orientation] [example: Orientation Of The Unit Square] Let $Q=[0,1]^2$ carry the standard orientation, so at every smooth boundary point the ordered basis \begin{align*} \left(\frac{\partial}{\partial x_1},\frac{\partial}{\partial x_2}\right) \end{align*} is positive. By the outward-normal-first convention, a boundary tangent vector $v$ is positive exactly when $(\nu,v)$ is a positive ordered basis of $TQ$, where $\nu$ points outward from $Q$. On the bottom edge $\{x_2=0\}$, the outward vector is \begin{align*} \nu=-\frac{\partial}{\partial x_2}. \end{align*} Testing $v=\frac{\partial}{\partial x_1}$, the coordinate matrix of $(\nu,v)$ relative to $\left(\frac{\partial}{\partial x_1},\frac{\partial}{\partial x_2}\right)$ is \begin{align*} \begin{pmatrix} 0&1\\ -1&0 \end{pmatrix}, \end{align*} and its determinant is \begin{align*} (0)(0)-(1)(-1)=1>0. \end{align*} Thus the bottom edge is oriented in the $\frac{\partial}{\partial x_1}$ direction, from left to right. On the right edge $\{x_1=1\}$, the outward vector is $\nu=\frac{\partial}{\partial x_1}$. With $v=\frac{\partial}{\partial x_2}$, the coordinate matrix is \begin{align*} \begin{pmatrix} 1&0\\ 0&1 \end{pmatrix}, \end{align*} so \begin{align*} (1)(1)-(0)(0)=1>0. \end{align*} The right edge is therefore oriented upward, from bottom to top. On the top edge $\{x_2=1\}$, the outward vector is $\nu=\frac{\partial}{\partial x_2}$. With $v=-\frac{\partial}{\partial x_1}$, the coordinate matrix is \begin{align*} \begin{pmatrix} 0&-1\\ 1&0 \end{pmatrix}, \end{align*} and \begin{align*} (0)(0)-(-1)(1)=1>0. \end{align*} Hence the top edge is oriented in the $-\frac{\partial}{\partial x_1}$ direction, from right to left. On the left edge $\{x_1=0\}$, the outward vector is $\nu=-\frac{\partial}{\partial x_1}$. With $v=-\frac{\partial}{\partial x_2}$, the coordinate matrix is \begin{align*} \begin{pmatrix} -1&0\\ 0&-1 \end{pmatrix}, \end{align*} so \begin{align*} (-1)(-1)-(0)(0)=1>0. \end{align*} Thus the left edge is oriented downward, from top to bottom. Combining the four edges gives the familiar counterclockwise boundary orientation of the positively oriented square. [/example] This is the familiar counterclockwise orientation of the boundary of a positively oriented planar region. The corner points do not form a second boundary in the manifold-with-boundary sense; in chain notation their contributions cancel in pairs. [example: Upper Hemisphere And The Equator] Let \begin{align*} S^2_+=\{x\in\mathbb R^3:|x|=1,\ x_3\ge 0\} \end{align*} be oriented by the outward unit normal $n(x)=x$. Its boundary is the equator \begin{align*} C=\{x\in\mathbb R^3:|x|=1,\ x_3=0\}. \end{align*} We show that the induced boundary orientation on $C$ is represented by \begin{align*} \gamma(t)=(\cos t,\sin t,0),\qquad 0\le t<2\pi. \end{align*} At the point $\gamma(t)$, the outward unit normal to the hemisphere is \begin{align*} n(\gamma(t))=\gamma(t)=(\cos t,\sin t,0). \end{align*} The tangent vector to the equator is \begin{align*} \dot\gamma(t)=(-\sin t,\cos t,0). \end{align*} Along the boundary of the upper hemisphere, the outward-pointing tangent vector inside $S^2_+$ is \begin{align*} \nu=(0,0,-1), \end{align*} because moving in the negative $x_3$-direction leaves the half-sphere condition $x_3\ge 0$. To check the sign, write the ordered triple \begin{align*} \bigl(n(\gamma(t)),\nu,\dot\gamma(t)\bigr) \end{align*} as columns in the standard basis of $\mathbb R^3$: \begin{align*} \begin{pmatrix} \cos t&0&-\sin t\\ \sin t&0&\cos t\\ 0&-1&0 \end{pmatrix}. \end{align*} Its determinant is \begin{align*} &\cos t\bigl((0)(0)-(\cos t)(-1)\bigr) -0\bigl((\sin t)(0)-(\cos t)(0)\bigr) +(-\sin t)\bigl((\sin t)(-1)-(0)(0)\bigr)\\ &=\cos t(\cos t)+(-\sin t)(-\sin t)\\ &=\cos^2 t+\sin^2 t\\ &=1>0. \end{align*} Thus $\bigl(n(\gamma(t)),\nu,\dot\gamma(t)\bigr)$ is positively oriented in $\mathbb R^3$, so $(\nu,\dot\gamma(t))$ is a positive ordered basis of $T_{\gamma(t)}S^2_+$ for the surface orientation. By the outward-normal-first convention, $\dot\gamma(t)$ is therefore the positive boundary direction on $C$. Hence the induced boundary orientation is the counterclockwise parametrisation $\gamma(t)=(\cos t,\sin t,0)$ as viewed from above. [/example] ## Orientability and the Mobius Band The final question is whether every manifold admits a continuous orientation in the first place. Local orientations always exist in a chart, but when charts are glued around a loop the sign may return reversed. [definition: Orientable Manifold] A smooth manifold with boundary $M$ is orientable if it admits an orientation. If an orientation has been chosen, $M$ is called oriented. [/definition] Orientability is therefore a global compatibility condition, not a pointwise condition. The failure is detected by transporting an ordered basis around a closed path and comparing the sign when it returns. [example: Mobius Band Is Not Orientable] [claim]The Mobius band \begin{align*} M=([0,1]\times[-1,1])/((0,t)\sim(1,-t)) \end{align*} is not orientable.[/claim] [proof]Let $q:[0,1]\times[-1,1]\to M$ be the quotient map, and let $c(s)=q(s,0)$ be the core circle. Suppose that $M$ had an orientation. Along the path $0\le s\le 1$, read tangent vectors using the rectangle coordinates before the endpoint identification. Let $\epsilon(s)=1$ if $(\partial_s,\partial_t)$ is positive at $c(s)$, and let $\epsilon(s)=-1$ if it is negative. In a coordinate neighbourhood, a local positive area form has the form \begin{align*} \omega=a(s,t)\,ds\wedge dt \end{align*} with $a$ continuous and nonzero. Since \begin{align*} \omega(\partial_s,\partial_t) &=a(s,t)(ds\wedge dt)(\partial_s,\partial_t)\\ &=a(s,t)\bigl(ds(\partial_s)dt(\partial_t)-ds(\partial_t)dt(\partial_s)\bigr)\\ &=a(s,t)(1\cdot 1-0\cdot 0)\\ &=a(s,t), \end{align*} the sign of $(\partial_s,\partial_t)$ is $\operatorname{sgn}(a(s,0))$. This sign is locally constant because a continuous nonzero function cannot change sign. Therefore transporting the orientation once along the connected core gives \begin{align*} \epsilon(0)=\epsilon(1). \end{align*} The endpoint identification is \begin{align*} (0,t)\sim(1,-t), \end{align*} so the transition map from the left endpoint coordinates to the right endpoint coordinates is \begin{align*} G(s,t)=(s+1,-t). \end{align*} Its derivative at $(0,0)$ has matrix \begin{align*} DG_{(0,0)} &= \begin{pmatrix} \frac{\partial(s+1)}{\partial s} & \frac{\partial(s+1)}{\partial t}\\ \frac{\partial(-t)}{\partial s} & \frac{\partial(-t)}{\partial t} \end{pmatrix}\\ &= \begin{pmatrix} 1&0\\ 0&-1 \end{pmatrix}. \end{align*} Thus \begin{align*} DG_{(0,0)}(\partial_s)=\partial_s, \qquad DG_{(0,0)}(\partial_t)=-\partial_t. \end{align*} The coordinate matrix of the image basis $(\partial_s,-\partial_t)$ relative to $(\partial_s,\partial_t)$ at the right endpoint is \begin{align*} \begin{pmatrix} 1&0\\ 0&-1 \end{pmatrix}, \end{align*} and its determinant is \begin{align*} (1)(-1)-(0)(0)=-1<0. \end{align*} Hence the gluing reverses the orientation class, so the same endpoint identification forces \begin{align*} \epsilon(0)=-\epsilon(1). \end{align*} This contradicts $\epsilon(0)=\epsilon(1)$. Therefore no continuous global orientation exists on $M$.[/proof] The obstruction is that one full trip around the core circle returns to the same tangent plane with the orientation reversed. [/example] This example explains why orientation must be stated as a hypothesis in integration theorems. Manifolds such as intervals, disks, spheres, and orientable surfaces support a global sign convention, while the Mobius band does not. Once orientation is fixed, top-degree forms can be integrated over oriented domains in a way that generalizes ordinary line, surface, and volume integrals. The next chapter uses that setup to relate integrals over a region to integrals over its boundary. # 6. Integration of Differential Forms Integration is where differential forms stop being only algebraic objects and become quantities attached to oriented geometric domains. The guiding question in this chapter is: what should it mean to add up a $k$-form over a $k$-dimensional oriented object? The answer is built from pullback: reduce every integral to an ordinary integral on an open subset of Euclidean space, and let the alternating determinant factor keep track of orientation. The construction also explains why the older vector-calculus integrals fit into one framework. Line integrals, surface integrals, flux integrals, and the change-of-variables formula all arise by writing the relevant object as a differential form and pulling it back to parameters. ## Top Forms on Oriented Open Sets How should an $n$-form on an [open set](/page/Open%20Set) $U\subseteq \mathbb R^n$ be integrated when it already has the same degree as the ambient dimension? The coordinate volume form $dx_1\wedge\cdots\wedge dx_n$ gives a reference against which every top form can be measured, while the orientation decides which sign is positive. [definition: Integral of a Top Form on an Oriented Open Set] Let $U\subseteq \mathbb R^n$ be open with the standard orientation. Let $\omega\in \Omega^n(U)$ have coordinate expression \begin{align*} \omega = f\,dx_1\wedge\cdots\wedge dx_n, \end{align*} where $f:U\to\mathbb R$ is Lebesgue integrable. The integral of $\omega$ over $U$ is \begin{align*} \int_U \omega := \int_U f\,d\mathcal L^n. \end{align*} For $U$ with the opposite orientation, the integral is \begin{align*} \int_{-U}\omega := -\int_U\omega. \end{align*} [/definition] This definition says that the coefficient of the oriented volume form is the density being integrated. The alternating product is not decorative: changing the orientation changes the sign of the basis top form, and the integral records that change. [example: Integral of a Planar Top Form] Let $U=(0,1)^2$ carry the standard orientation, and let $\omega=(x+y)\,dx_1\wedge dx_2$. Write a point of $U$ as $(x,y)=(x_1,x_2)$. Since the coefficient of the positive volume form $dx_1\wedge dx_2$ is $f(x,y)=x+y$, the definition of the integral of a top form gives \begin{align*} \int_U\omega &=\int_0^1\int_0^1(x+y)\,dy\,dx\\ &=\int_0^1\left(\int_0^1x\,dy+\int_0^1y\,dy\right)\,dx\\ &=\int_0^1\left(\left[xy\right]_{y=0}^{y=1}+\left[\frac{y^2}{2}\right]_{y=0}^{y=1}\right)\,dx\\ &=\int_0^1\left(x+\frac{1}{2}\right)\,dx\\ &=\left[\frac{x^2}{2}+\frac{x}{2}\right]_{x=0}^{x=1}\\ &=\frac{1}{2}+\frac{1}{2}\\ &=1. \end{align*} For the same square with the opposite orientation, the definition reads the same coefficient with the opposite sign, so \begin{align*} \int_{-U}\omega=-\int_U\omega=-1. \end{align*} Thus this form assigns signed area-weighted value $1$ to the standard orientation and $-1$ to the reversed orientation. [/example] The next result isolates the sign behaviour because it will reappear for every oriented manifold. Without recording orientation, the same region $U$ and the oppositely oriented region $-U$ would give identical values, so form integration would lose the sign information carried by determinants. This is why integration of top forms is not merely integration of nonnegative volume. [quotetheorem:3609] [citeproof:3609] The hypothesis that $\omega$ has top degree is essential: a $1$-form on an open subset of $\mathbb R^2$ is not integrated over the whole [open set](/page/Open%20Set) by this definition, because there is no scalar coefficient of $dx_1\wedge dx_2$ to integrate. The result also does not say that ordinary volume changes sign; for example $\mathcal L^n(U)$ is unchanged when the orientation is reversed. The statement is a convention at the level of open subsets, but it becomes a theorem for manifolds once integrals are defined using oriented charts. Top-degree integration on open sets is the local model for every later construction. The only extra ingredient for submanifolds is a way to compare a $k$-form on a curved object with a $k$-form on a parameter domain. ## Integrals Over Oriented Parameterized Manifolds If a curve or surface is described by parameters, what should be integrated on the parameter domain? The only natural candidate is the pullback of the form, because it evaluates the form on the tangent vectors supplied by the parametrization. [definition: Integral over a Parameterized Patch] Let $A\subseteq\mathbb R^k$ be open with the standard orientation. Let $\phi:A\to\mathbb R^n$ be a smooth regular embedding onto the $k$-dimensional patch $P=\phi(A)$, and let $\omega$ be a $k$-form defined on an open neighbourhood of $P$ such that $\phi^*\omega$ is integrable on $A$. The integral of $\omega$ over the parametrized patch $(P,\phi)$ is \begin{align*} \int_{(P,\phi)}\omega := \int_A \phi^*\omega. \end{align*} [/definition] The induced orientation means that an ordered basis of tangent vectors is positive on the patch when it is the image under $d\phi$ of a positive ordered basis in $\mathbb R^k$. Thus the same geometric patch can acquire the opposite orientation if the parameters are reversed. [example: Reversing a Parameter on a Circle] Let $C\subseteq\mathbb R^2$ be the unit circle and let $\omega=x\,dy-y\,dx$. For the counterclockwise parametrization $\phi(t)=(\cos t,\sin t)$, $0\le t\le 2\pi$, we have $x\circ\phi=\cos t$ and $y\circ\phi=\sin t$, so \begin{align*} \phi^*\omega &=(x\circ\phi)\,d(y\circ\phi)-(y\circ\phi)\,d(x\circ\phi)\\ &=\cos t\,d(\sin t)-\sin t\,d(\cos t)\\ &=\cos t(\cos t\,dt)-\sin t(-\sin t\,dt)\\ &=(\cos^2 t+\sin^2 t)\,dt\\ &=dt. \end{align*} Therefore \begin{align*} \int_C\omega &=\int_0^{2\pi}dt\\ &=\left[t\right]_{0}^{2\pi}\\ &=2\pi \end{align*} for the counterclockwise orientation. For the clockwise parametrization $\psi(t)=(\cos t,-\sin t)$, we have $x\circ\psi=\cos t$ and $y\circ\psi=-\sin t$, hence \begin{align*} \psi^*\omega &=(x\circ\psi)\,d(y\circ\psi)-(y\circ\psi)\,d(x\circ\psi)\\ &=\cos t\,d(-\sin t)-(-\sin t)\,d(\cos t)\\ &=\cos t(-\cos t\,dt)+\sin t(-\sin t\,dt)\\ &=-(\cos^2 t+\sin^2 t)\,dt\\ &=-dt. \end{align*} Thus \begin{align*} \int_C\omega &=\int_0^{2\pi}(-dt)\\ &=-\left[t\right]_{0}^{2\pi}\\ &=-2\pi \end{align*} for the clockwise orientation. Reversing the parameter direction reverses the induced orientation of the circle, and the same $1$-form records that reversal by changing the sign of the integral. [/example] The definition by parameters would be unusable unless it gave the same value for different positive parametrizations. This is not just a technical detail: a patch of an oriented manifold should have one integral, not one value for each coordinate description. The next theorem is the invariance check that makes integration of forms a geometric operation rather than a coordinate convention. [quotetheorem:3610] This is the local well-definedness theorem for integration of forms. Orientation-preserving changes of parameters do not alter the integral, while orientation-reversing changes of parameters reverse the sign. [citeproof:3610] The regular embedding hypothesis is needed because otherwise the transition map may not exist; for instance, the map $t\mapsto (t^2,t^3)$ has a singular tangent at $0$ and is not a valid oriented curve chart there. The sign condition is also essential: the parameter changes $t\mapsto t$ and $t\mapsto -t$ describe the same interval with opposite orientations and give opposite integrals for $dt$. The theorem does not identify unoriented patches; it says only that once an orientation is fixed, all positive parametrizations produce the same signed value. For manifolds covered by more than one chart, the same idea is assembled with a [partition of unity](/page/Partition%20of%20Unity). Compact support ensures that only finitely many chart contributions meet the part of the manifold where the form is nonzero after passing to a finite subcover of the support. [definition: Integral over an Oriented Manifold] Let $M\subseteq\mathbb R^n$ be an oriented smooth $k$-manifold, and let $\omega\in\Omega^k(M)$ have compact support. Choose an oriented smooth atlas $(\phi_a:A_a\to M)_{a\in I}$ and a smooth [partition of unity](/page/Partition%20of%20Unity) $(\rho_a)_{a\in I}$ subordinate to this atlas, where each $\rho_a:M\to\mathbb R$ is smooth, $\operatorname{supp}\rho_a\subseteq \phi_a(A_a)$, the family $(\operatorname{supp}\rho_a)_{a\in I}$ is locally finite, and $\sum_{a\in I}\rho_a=1$ on a neighbourhood of $\operatorname{supp}\omega$. The integral of $\omega$ over $M$ is \begin{align*} \int_M\omega:=\sum_{a\in I}\int_{A_a}\phi_a^*(\rho_a\omega). \end{align*} [/definition] The formula is local, but the value is global. Each term integrates the part of the form assigned to one oriented chart, and overlap terms agree because positive transition maps preserve the oriented integral. Because the definition involves choices of atlas and partition of unity, it still needs a well-definedness result. The next theorem confirms that changing those auxiliary choices does not change the integral, so the notation $\int_M\omega$ depends only on the oriented manifold and the compactly supported form. [quotetheorem:3579] The theorem is the manifold version of independence of parametrization. It says that orientation is the only chart-level datum remembered by the integral. [citeproof:3579] Compact support is part of the statement because, without it or another integrability hypothesis, the sum of local integrals can fail to converge; for example the constant top form $dx$ on $\mathbb R$ has no finite integral over all of $\mathbb R$. The orientation hypothesis is also necessary: if a chart transition reverses orientation, the local contributions differ by a sign rather than agreeing. This theorem permits the notation $\int_M\omega$ without mentioning charts, and it prepares the orientation-reversal formula by showing exactly which chart data have disappeared from the final value. The orientation remains part of the input, so replacing $M$ by the same manifold with the opposite orientation should change the result in a predictable way. Since every local chart has its orientation reversed at once, each top-degree coordinate integral should acquire the same minus sign. The theorem records this global sign rule. [quotetheorem:3580] The sign is not a new analytic phenomenon; it is the determinant sign from every local chart. [citeproof:3580] The compact support assumption again prevents analytic divergence from obscuring the orientation statement; on a noncompact manifold a nonintegrable form may have neither side defined. The conclusion also does not say that the underlying set or its [Hausdorff measure](/page/Hausdorff%20Measure) changes, since $M$ and $-M$ have the same points and the same unoriented size. Orientation reversal is the first place where differential forms differ sharply from unoriented volume integration, and this distinction is what later makes boundary orientation in [Stokes' theorem](/theorems/1530) meaningful. A form integral measures signed geometry, not only size. ## Classical Line, Surface, and Flux Integrals How do the integrals from multivariable calculus appear inside this definition? They are recovered by writing the classical integrand as a differential form and pulling it back along the familiar parametrization. [quotetheorem:3611] This is precisely the usual line integral of the vector field $P=(P_1,\dots,P_n)$ along the oriented curve. [citeproof:3611] Smoothness of $\gamma$ is used so that the pullbacks $\gamma^*dx_i=d(\gamma_i)$ have the displayed coefficient functions; for a curve with corners the same formula must be applied piecewise. The theorem does not assert parametrization by arclength, and it does not compute the unsigned work against speed $|\gamma'(t)|$ unless the $1$-form has been chosen for that purpose. The sign of the line integral is therefore controlled by the orientation of the parameter interval, and reversing the path replaces $dt$ by $-dt$ in exactly the same way that reversing an oriented manifold changes the sign. This line-integral formula is the first classical recovery result; the surface and flux formulas use the same pullback mechanism in degree $2$. [example: Signed Area from a One-Form] Let $C$ be a positively oriented piecewise smooth closed plane curve bounding a region $D\subseteq\mathbb R^2$, and consider \begin{align*} \alpha=\frac{1}{2}(x\,dy-y\,dx). \end{align*} Writing $\alpha=P\,dx+Q\,dy$ with $P=-\frac{y}{2}$ and $Q=\frac{x}{2}$, the Green theorem proved in the next chapter gives \begin{align*} \int_C\alpha &=\int_C P\,dx+Q\,dy\\ &=\int_D\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)\,d\mathcal L^2\\ &=\int_D\left(\frac{\partial}{\partial x}\frac{x}{2}-\frac{\partial}{\partial y}\left(-\frac{y}{2}\right)\right)\,d\mathcal L^2\\ &=\int_D\left(\frac{1}{2}-\left(-\frac{1}{2}\right)\right)\,d\mathcal L^2\\ &=\int_D 1\,d\mathcal L^2\\ &=\mathcal L^2(D). \end{align*} Thus $\alpha$ integrates to the signed area enclosed by $C$ for the positive orientation. If $r(t)=(x(t),y(t))$ parametrizes $C$ on $[a,b]$, then \begin{align*} r^*\alpha=\frac{1}{2}\bigl(x(t)y'(t)-y(t)x'(t)\bigr)\,dt. \end{align*} For the reversed parametrization $\widetilde r(t)=r(a+b-t)$, we have \begin{align*} \widetilde r^*\alpha &=\frac{1}{2}\bigl(x(a+b-t)(-y'(a+b-t))-y(a+b-t)(-x'(a+b-t))\bigr)\,dt\\ &=-\frac{1}{2}\bigl(x(a+b-t)y'(a+b-t)-y(a+b-t)x'(a+b-t)\bigr)\,dt, \end{align*} so the reversed orientation changes the integral by a factor of $-1$. For the circle $r(t)=(R\cos t,R\sin t)$, $0\le t\le 2\pi$, we have \begin{align*} dx&=-R\sin t\,dt,\\ dy&=R\cos t\,dt. \end{align*} Therefore \begin{align*} r^*\alpha &=\frac{1}{2}\left((R\cos t)(R\cos t\,dt)-(R\sin t)(-R\sin t\,dt)\right)\\ &=\frac{1}{2}\left(R^2\cos^2 t+R^2\sin^2 t\right)\,dt\\ &=\frac{R^2}{2}(\cos^2 t+\sin^2 t)\,dt\\ &=\frac{R^2}{2}\,dt. \end{align*} Hence \begin{align*} \int_C\alpha &=\int_0^{2\pi}\frac{R^2}{2}\,dt\\ &=\frac{R^2}{2}\left[t\right]_0^{2\pi}\\ &=\frac{R^2}{2}(2\pi)\\ &=\pi R^2. \end{align*} The one-form $\frac{1}{2}(x\,dy-y\,dx)$ therefore packages the planar signed-[area formula](/theorems/3075) as an ordinary integral of a differential form over the boundary curve. [/example] Surface integrals are similar, but $2$-forms in $\mathbb R^3$ encode oriented area through pairs of tangent vectors. The key problem is to connect the form integral with the vector-calculus flux integral: the orientation of the parameter domain must determine the normal vector and therefore the sign of the flux. [illustration:oriented-surface-normal] [quotetheorem:3581] Thus the flux integral of $F$ through an oriented surface is the integral of the associated $2$-form $P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy$. [citeproof:3581] The orientation hypothesis is needed because replacing the parametrization order $(u,v)$ by $(v,u)$ changes $r_u\times r_v$ to its negative and reverses the flux. The theorem also does not define an unoriented surface integral of a scalar function such as surface area; that requires the positive density $|r_u\times r_v|\,d\mathcal L^2(u,v)$ rather than a signed $2$-form. The normal vector has not been added by hand: it is encoded by the ordered pair of tangent vectors and the alternating nature of the $2$-form. This prepares the change-of-variables viewpoint, where the same determinant mechanism appears in top degree. [example: Flux Through the Unit Sphere] Let $S^2\subseteq\mathbb R^3$ have outward orientation and let $F(x,y,z)=(x,y,z)$. The associated flux form is \begin{align*} \omega=x\,dy\wedge dz+y\,dz\wedge dx+z\,dx\wedge dy. \end{align*} Use \begin{align*} r(\varphi,\theta)=(\sin\varphi\cos\theta,\sin\varphi\sin\theta,\cos\varphi), \end{align*} with $0<\varphi<\pi$ and $0<\theta<2\pi$, ordered as $(\varphi,\theta)$. The coordinate functions are \begin{align*} x(\varphi,\theta)&=\sin\varphi\cos\theta,\\ y(\varphi,\theta)&=\sin\varphi\sin\theta,\\ z(\varphi,\theta)&=\cos\varphi. \end{align*} Their differentials are \begin{align*} dx&=\cos\varphi\cos\theta\,d\varphi-\sin\varphi\sin\theta\,d\theta,\\ dy&=\cos\varphi\sin\theta\,d\varphi+\sin\varphi\cos\theta\,d\theta,\\ dz&=-\sin\varphi\,d\varphi. \end{align*} Therefore \begin{align*} dy\wedge dz &=(\cos\varphi\sin\theta\,d\varphi+\sin\varphi\cos\theta\,d\theta)\wedge(-\sin\varphi\,d\varphi)\\ &=-\cos\varphi\sin\varphi\sin\theta\,d\varphi\wedge d\varphi-\sin^2\varphi\cos\theta\,d\theta\wedge d\varphi\\ &=\sin^2\varphi\cos\theta\,d\varphi\wedge d\theta,\\ dz\wedge dx &=(-\sin\varphi\,d\varphi)\wedge(\cos\varphi\cos\theta\,d\varphi-\sin\varphi\sin\theta\,d\theta)\\ &=-\sin\varphi\cos\varphi\cos\theta\,d\varphi\wedge d\varphi+\sin^2\varphi\sin\theta\,d\varphi\wedge d\theta\\ &=\sin^2\varphi\sin\theta\,d\varphi\wedge d\theta,\\ dx\wedge dy &=(\cos\varphi\cos\theta\,d\varphi-\sin\varphi\sin\theta\,d\theta)\wedge(\cos\varphi\sin\theta\,d\varphi+\sin\varphi\cos\theta\,d\theta)\\ &=\sin\varphi\cos\varphi\cos^2\theta\,d\varphi\wedge d\theta+\sin\varphi\cos\varphi\sin^2\theta\,d\varphi\wedge d\theta\\ &=\sin\varphi\cos\varphi\,d\varphi\wedge d\theta. \end{align*} Substituting these into $r^*\omega$ gives \begin{align*} r^*\omega &=(\sin\varphi\cos\theta)(\sin^2\varphi\cos\theta\,d\varphi\wedge d\theta)\\ &\quad+(\sin\varphi\sin\theta)(\sin^2\varphi\sin\theta\,d\varphi\wedge d\theta)\\ &\quad+(\cos\varphi)(\sin\varphi\cos\varphi\,d\varphi\wedge d\theta)\\ &=\left(\sin^3\varphi\cos^2\theta+\sin^3\varphi\sin^2\theta+\sin\varphi\cos^2\varphi\right)d\varphi\wedge d\theta\\ &=\sin\varphi\left(\sin^2\varphi(\cos^2\theta+\sin^2\theta)+\cos^2\varphi\right)d\varphi\wedge d\theta\\ &=\sin\varphi\,d\varphi\wedge d\theta. \end{align*} Hence \begin{align*} \int_{S^2}\omega &=\int_0^{2\pi}\int_0^\pi \sin\varphi\,d\varphi\,d\theta\\ &=\int_0^{2\pi}\left[-\cos\varphi\right]_{\varphi=0}^{\varphi=\pi}\,d\theta\\ &=\int_0^{2\pi}\bigl(-\cos\pi+\cos 0\bigr)\,d\theta\\ &=\int_0^{2\pi}2\,d\theta\\ &=2[\theta]_{\theta=0}^{\theta=2\pi}\\ &=4\pi. \end{align*} The outward radial field has unit dot product with the outward unit normal on the unit sphere, so the same value is the total outward flux through $S^2$. [/example] These recoveries are more than translations of notation. They show that forms automatically combine the integrand, tangent directions, and orientation into a single object. ## Change of Variables as Pullback Why does the Jacobian determinant appear in the ordinary change-of-variables theorem? For differential forms, the determinant appears because a top form evaluates on $n$ transformed tangent vectors, and multilinearity plus alternation produce the determinant. [definition: Orientation Sign of a Diffeomorphism] Let $U,V\subseteq\mathbb R^n$ be connected open sets, and let $F:U\to V$ be a diffeomorphism. The orientation sign of $F$ is $+1$ if $\det JF_x>0$ for all $x\in U$, and $-1$ if $\det JF_x<0$ for all $x\in U$. [/definition] The determinant cannot change sign on a connected domain without vanishing somewhere, and a diffeomorphism has invertible derivative at every point. The remaining question is how this sign enters integration: top forms should transform by the determinant itself, so orientation-reversing maps must change the sign of the integral. [quotetheorem:3554] This is the clean form-valued change-of-variables law. The sign is absent from ordinary positive-density integration because scalar densities use $|\det JF_x|$ instead of $\det JF_x$. [citeproof:3554] The diffeomorphism hypothesis is essential: if $F(x)=x^2$ on $\mathbb R$, then $JF_0=0$ and the map is not locally invertible at $0$, so there is no globally signed determinant rule of this kind. Connectedness is used only to make the orientation sign a single number; on disconnected domains the sign may differ from component to component. The theorem does not say that scalar volume integrals use signed determinants, and the standard change-of-variables formula is recovered by applying this theorem to a coordinate volume form and then replacing the signed determinant by its absolute value for ordinary measure integration. [quotetheorem:22] The form version remembers orientation; the measure version remembers unsigned volume scaling. [citeproof:22] [example: Polar Coordinates and the Area Form] Let $A=(a,b)\times(0,2\pi)$ with $0<a<b$, and define \begin{align*} F(r,\theta)=(x(r,\theta),y(r,\theta))=(r\cos\theta,r\sin\theta). \end{align*} Then \begin{align*} dx&=d(r\cos\theta)=\cos\theta\,dr-r\sin\theta\,d\theta,\\ dy&=d(r\sin\theta)=\sin\theta\,dr+r\cos\theta\,d\theta. \end{align*} Therefore \begin{align*} F^*(dx\wedge dy) &=(\cos\theta\,dr-r\sin\theta\,d\theta)\wedge(\sin\theta\,dr+r\cos\theta\,d\theta)\\ &=\cos\theta\sin\theta\,dr\wedge dr+r\cos^2\theta\,dr\wedge d\theta-r\sin^2\theta\,d\theta\wedge dr-r^2\sin\theta\cos\theta\,d\theta\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*} Since $r>0$ on $A$, the coefficient $r$ is positive. The map $F$ covers the annulus $a<|x|<b$ except for the positive $x$-axis segment with $a<x<b$ and $y=0$, which has planar measure $0$, so the area is computed from the coefficient of the pulled-back area form: \begin{align*} \int_A r\,d\mathcal L^2(r,\theta) &=\int_0^{2\pi}\int_a^b r\,dr\,d\theta\\ &=\int_0^{2\pi}\left[\frac{r^2}{2}\right]_{r=a}^{r=b}\,d\theta\\ &=\int_0^{2\pi}\frac{b^2-a^2}{2}\,d\theta\\ &=\frac{b^2-a^2}{2}[\theta]_{\theta=0}^{\theta=2\pi}\\ &=\frac{b^2-a^2}{2}(2\pi)\\ &=\pi(b^2-a^2). \end{align*} Thus the familiar polar-coordinate Jacobian factor is exactly the coefficient produced by pulling back the oriented area form $dx\wedge dy$. [/example] Integration of forms is therefore built from three linked ideas: top forms are signed densities, pullback transports forms to parameter spaces, and orientation decides which parameter changes preserve sign. These ideas prepare the next step of the course, where exterior differentiation and oriented boundary integration combine into [Stokes' theorem](/theorems/1530). [Stokes' theorem](/theorems/1530) is where the algebra of forms, the [exterior derivative](/theorems/1525), pullbacks, orientation, and integration come together in one statement. After seeing that boundary integration encodes differentiation, the next question is when a closed form actually comes from differentiating something smaller. # 7. Stokes' Theorem This chapter is the point at which the algebra of forms, the [exterior derivative](/theorems/1525), pullbacks, orientations, and integration become one theorem. Stokes theorem says that integrating $d\omega$ over an oriented manifold is the same operation as integrating $\omega$ over its oriented boundary. The familiar integral theorems of vector calculus are not separate phenomena; they are different coordinate presentations of this single statement. The previous chapter defined integration of $k$-forms over oriented $k$-manifolds, and Chapter 5 explained how the boundary orientation is chosen. We now use that orientation convention to make cancellation exact: contributions from artificial interior faces disappear, while the true boundary remains. ## Why Boundaries Control Exterior Derivatives The central question is how an integral over a region can be determined by data on its boundary. In one variable, the [fundamental theorem of calculus](/theorems/632) says that the integral of a derivative over an interval records the endpoint values. Stokes theorem is the same principle after replacing intervals by oriented manifolds and ordinary derivatives by the [exterior derivative](/theorems/1525). [definition: Boundary Orientation] Let $M$ be an oriented $k$-manifold-with-boundary. The boundary orientation on $\partial M$ is the orientation for which, at each $p \in \partial M$, an ordered basis $(v_1,\dots,v_{k-1})$ of $T_p\partial M$ is positive exactly when $(\nu,v_1,\dots,v_{k-1})$ is a positive basis of $T_pM$, where $\nu$ is an outward-pointing transverse vector in $T_pM$. [/definition] The outward normal is placed first in this convention. This choice is what makes the signs in the one-dimensional fundamental theorem match the signs in all higher-dimensional boundary integrals. [example: Interval Orientation] Let $M=[a,b]$ carry the standard orientation determined by $dx$, and let $f$ be a smooth $0$-form on an open interval containing $[a,b]$. At $b$ the outward normal is $+\partial_x$, so $b$ has positive boundary sign; at $a$ the outward normal is $-\partial_x$, so $a$ has negative boundary sign. Thus \begin{align*} \partial [a,b]=\{b\}-\{a\}. \end{align*} For the $0$-form $f$, the [exterior derivative](/theorems/1525) is \begin{align*} df=f'(x)\,dx. \end{align*} Therefore the left side of Stokes theorem is \begin{align*} \int_{[a,b]}df &=\int_{[a,b]} f'(x)\,dx\\ &=\int_a^b f'(x)\,dx, \end{align*} because $dx$ is the positive coordinate volume form on the oriented interval. The right side is the integral of $f$ over the oriented $0$-manifold $\{b\}-\{a\}$, namely \begin{align*} \int_{\partial [a,b]} f &=\int_{\{b\}-\{a\}}f\\ &=f(b)-f(a). \end{align*} Hence Stokes theorem on the oriented interval reads \begin{align*} \int_a^b f'(x)\,dx=f(b)-f(a), \end{align*} which is exactly the one-variable [fundamental theorem of calculus](/theorems/632) with the endpoint signs supplied by the boundary orientation. [/example] The interval case is the local model. Higher-dimensional proofs reduce to coordinate boxes and half-boxes, apply this one-dimensional calculation in one coordinate at a time, and then account for how faces are oriented. ## The General Stokes Theorem The theorem has to compare a $k$-form integrated over $M$ with a $(k-1)$-form integrated over $\partial M$. The [exterior derivative](/theorems/1525) is the operation that changes degree by one, so it is exactly the correct analytic object for this comparison. There are three obstructions the hypotheses remove. Without an orientation, the two sides have no consistent sign: reversing the orientation of an interval swaps $f(b)-f(a)$ with $f(a)-f(b)$. Without compactness or a compact support condition, the integrals need not exist; for instance, $d(x)=dx$ on $\mathbb R$ has infinite integral over the non-compact line. Without a smooth boundary, the pullback of a form to $\partial M$ and the boundary orientation may require extra measure-theoretic interpretation rather than the manifold-level construction used here. [quotetheorem:3555] [citeproof:3555] Compactness is used to reduce the argument to finitely many coordinate pieces; a non-compact version needs compact support or an integrability hypothesis. The orientation hypotheses are not decoration: reversing the orientation of $M$ reverses both sides, while choosing the wrong boundary orientation reverses only the boundary side and breaks the identity. The theorem does not say that every closed form is exact, nor that a zero boundary integral forces $d\omega=0$ pointwise; it relates one global integral to another. This distinction is the bridge to de Rham cohomology, where closed forms modulo exact forms measure the failure of local primitives to fit together globally. [remark: Closed Manifolds] If $M$ is compact, oriented, and has empty boundary, then Stokes theorem gives \begin{align*} \int_M d\omega = 0 \end{align*} for every smooth $(n-1)$-form $\omega$ on the $n$-dimensional manifold $M$. [/remark] This is the first indication that exact forms have no total integral on closed oriented manifolds. In later courses this observation becomes one of the entry points to de Rham cohomology. [example: Exact One-Form Around A Circle] Let $S^1$ have its usual counterclockwise orientation, and let $f:S^1\to\mathbb R$ be smooth. Since $\partial S^1=\varnothing$, *General Stokes Theorem* applied to the compact oriented $1$-manifold $S^1$ and the $0$-form $f$ gives \begin{align*} \int_{S^1}df &=\int_{\partial S^1}f\\ &=\int_{\varnothing}f\\ &=0, \end{align*} where the last equality is the definition of integration over the empty oriented $0$-manifold. In the standard angular parameter $\gamma(t)=e^{it}$, $0\le t\le 2\pi$, define $g(t)=f(\gamma(t))=f(e^{it})$. Since $e^{0i}=1$ and $e^{2\pi i}=1$, \begin{align*} g(0)&=f(1),\\ g(2\pi)&=f(1), \end{align*} so $g(2\pi)=g(0)$. Pulling back $df$ to the parameter interval gives \begin{align*} \gamma^*(df) &=d(f\circ\gamma)\\ &=dg\\ &=g'(t)\,dt\\ &=\frac{d}{dt}f(e^{it})\,dt. \end{align*} Hence the coordinate computation is \begin{align*} \int_{S^1}df &=\int_0^{2\pi}g'(t)\,dt\\ &=g(2\pi)-g(0)\\ &=f(1)-f(1)\\ &=0. \end{align*} The vanishing is the boundary-free form of Stokes theorem; in angular coordinates it is exactly the cancellation of the endpoint values after one full period. [/example] ## Green Theorem In The Plane The next question is how the usual circulation theorem in the plane appears from the general formula. A $1$-form on a plane region records the infinitesimal work done along tangent directions, and its [exterior derivative](/theorems/1525) records the signed infinitesimal circulation density. The obstruction in the plane is sign, not computation. The same geometric curve bounds the same region with two possible orientations, and reversing the boundary orientation changes the line integral while the area integral is unchanged unless the orientation of the region is also reversed. Smoothness of the region and of $P,Q$ ensures that the boundary integral is the ordinary line integral and that the mixed coordinate derivatives appearing in $d(P\,dx+Q\,dy)$ are legitimate. [quotetheorem:3612] [citeproof:3612] Green theorem is therefore not a new integration principle. It is the degree-one, dimension-two case of Stokes theorem written in the coordinate basis $dx,dy$. The positive boundary orientation is necessary: the unit disk with clockwise boundary orientation would make the boundary integral the negative of the displayed area integral. The theorem also does not assert that zero circulation forces the integrand $\partial Q/\partial x-\partial P/\partial y$ to vanish; cancellations over the region can still produce total integral zero. This planar case is the model for translating geometric sign conventions into algebraic wedge-product signs. [example: Green Theorem On The Unit Disk] Let $M=\{(x,y)\in\mathbb R^2:x^2+y^2\le 1\}$ carry the standard orientation, and let $\omega=x^2\,dy$. Writing $\omega=P\,dx+Q\,dy$ gives $P=0$ and $Q=x^2$, so \begin{align*} d\omega &=d(x^2)\wedge dy\\ &=(2x\,dx)\wedge dy\\ &=2x\,dx\wedge dy. \end{align*} Thus the area side is \begin{align*} \iint_M 2x\,dA &=\int_0^{2\pi}\int_0^1 2(r\cos\theta)\,r\,dr\,d\theta\\ &=\int_0^{2\pi}\int_0^1 2r^2\cos\theta\,dr\,d\theta\\ &=\left(\int_0^1 2r^2\,dr\right)\left(\int_0^{2\pi}\cos\theta\,d\theta\right)\\ &=\left(\frac{2}{3}\right)\left(\sin(2\pi)-\sin(0)\right)\\ &=0. \end{align*} The boundary orientation induced by the standard orientation of the disk is counterclockwise, so we use $\gamma(t)=(\cos t,\sin t)$ for $0\le t\le 2\pi$. Along this curve, \begin{align*} \gamma^*(x^2\,dy) &=(\cos^2 t)\,d(\sin t)\\ &=(\cos^2 t)(\cos t\,dt)\\ &=\cos^3 t\,dt. \end{align*} Therefore the boundary side is \begin{align*} \oint_{\partial M}x^2\,dy &=\int_0^{2\pi}\cos^3 t\,dt\\ &=\int_0^{2\pi}\cos t(1-\sin^2 t)\,dt\\ &=\left[\sin t-\frac{\sin^3 t}{3}\right]_0^{2\pi}\\ &=\left(\sin(2\pi)-\frac{\sin^3(2\pi)}{3}\right)-\left(\sin(0)-\frac{\sin^3(0)}{3}\right)\\ &=0. \end{align*} Both sides vanish: the signed $x$-contributions cancel over the disk, and the boundary integral records the same cancellation over one full counterclockwise circuit. [/example] The example also shows why orientation matters. Reversing the boundary orientation changes the sign of the line integral, while the chosen orientation of the disk fixes the sign of the area integral. ## Classical Stokes Theorem For Surfaces The vector calculus version of Stokes theorem asks how circulation around a curve is related to curl through a spanning surface. Differential forms translate the vector field into a $1$-form and translate curl into the [exterior derivative](/theorems/1525) of that form. Without this translation, the vector formula depends on Euclidean structure that is not part of an abstract surface. Curl uses a metric, an orientation, and the three-dimensional cross-product convention, while Stokes theorem itself only needs forms, pullback, and orientation. The boundary orientation is again a genuine hypothesis: reversing the chosen normal reverses the induced orientation of the boundary curve and changes the sign of the circulation. [definition: One-Form Associated To A Vector Field] Let $U\subset\mathbb R^3$ be open. The association from smooth vector fields to $1$-forms is the map \begin{align*} C^\infty(U;\mathbb R^3)&\longrightarrow \Omega^1(U),\\ F=(F_1,F_2,F_3)&\longmapsto \omega_F=F_1\,dx+F_2\,dy+F_3\,dz. \end{align*} [/definition] With the Euclidean metric in the background, this definition is the usual identification between vectors and covectors. The [exterior derivative](/theorems/1525) of $\omega_F$ stores the same components as $\nabla\times F$, but as a $2$-form rather than as another vector field. [quotetheorem:3613] [citeproof:3613] The theorem says that a circulation integral only depends on the oriented surface through the curl flux. If two surfaces share the same oriented boundary, the difference between their curl fluxes is the integral of an exact form over a closed surface, hence vanishes when the hypotheses allow the surfaces to be joined. Smoothness and orientability are doing work here: a Mobius strip has no global choice of normal compatible with an orientation, and a vector field with singularities on the spanning surface cannot be inserted into the theorem without removing the singular set or changing the hypotheses. The result also does not say that a curve bounds a surface uniquely; it says every admissible spanning surface gives the same circulation calculation once the orientations match. [example: Curl Integral Over The Upper Hemisphere] Let $S=\{(x,y,z):x^2+y^2+z^2=1,\ z\ge 0\}$, oriented by the outward unit normal $N(x,y,z)=(x,y,z)$, and let $F=(-y/2,x/2,0)$. We show that the curl flux through $S$ is $\pi$. First compute the curl: \begin{align*} \nabla\times F &=\left( \frac{\partial 0}{\partial y}-\frac{\partial (x/2)}{\partial z}, \frac{\partial (-y/2)}{\partial z}-\frac{\partial 0}{\partial x}, \frac{\partial (x/2)}{\partial x}-\frac{\partial (-y/2)}{\partial y} \right)\\ &=\left(0-0,\ 0-0,\ \frac12-\left(-\frac12\right)\right)\\ &=(0,0,1). \end{align*} The boundary $\partial S$ is the unit circle in the plane $z=0$. Parametrize it by \begin{align*} \gamma(t)=(\cos t,\sin t,0),\qquad 0\le t\le 2\pi. \end{align*} At $\gamma(t)$, the outward-pointing normal to the boundary inside the surface is $\nu(t)=(0,0,-1)$, and \begin{align*} \nu(t)\times \gamma'(t) &=(0,0,-1)\times(-\sin t,\cos t,0)\\ &=(\cos t,\sin t,0)\\ &=N(\gamma(t)). \end{align*} Thus $\gamma$ has the boundary orientation induced by the outward orientation of the hemisphere. The line-integral $1$-form associated to $F$ is \begin{align*} F\cdot dr &=-\frac{y}{2}\,dx+\frac{x}{2}\,dy. \end{align*} Pulling it back along $\gamma$ gives \begin{align*} \gamma^*(F\cdot dr) &=-\frac{\sin t}{2}\,d(\cos t)+\frac{\cos t}{2}\,d(\sin t)\\ &=-\frac{\sin t}{2}(-\sin t\,dt)+\frac{\cos t}{2}(\cos t\,dt)\\ &=\frac{\sin^2 t}{2}\,dt+\frac{\cos^2 t}{2}\,dt\\ &=\frac{\sin^2 t+\cos^2 t}{2}\,dt\\ &=\frac12\,dt. \end{align*} Therefore \begin{align*} \oint_{\partial S}F\cdot dr &=\int_0^{2\pi}\frac12\,dt\\ &=\left[\frac{t}{2}\right]_0^{2\pi}\\ &=\frac{2\pi}{2}-0\\ &=\pi. \end{align*} By *Classical Stokes Theorem*, \begin{align*} \iint_S(\nabla\times F)\cdot n\,dS &=\oint_{\partial S}F\cdot dr\\ &=\pi. \end{align*} The hemisphere flux is therefore obtained entirely from the simpler circulation around its equatorial boundary. [/example] This computation avoids parametrising the hemisphere. Stokes theorem permits replacing a surface integral by a boundary integral when the boundary is simpler than the surface. ## Gauss Divergence Theorem As Stokes Theorem The [divergence theorem](/theorems/2754) has a different surface-level appearance: it compares volume production inside a solid with flux through its boundary. In forms language, a vector field is converted to a $2$-form whose [exterior derivative](/theorems/1525) is divergence times the volume form. The obstruction is that flux through a surface needs both a normal direction and an area density. In Euclidean $\mathbb R^3$ these are packaged by the standard orientation and metric, but on the form side they become the pullback of a $2$-form to the oriented boundary. If the boundary orientation were inward instead of outward, the surface flux would acquire the opposite sign while the volume integral would not. [definition: Flux Two-Form Associated To A Vector Field] Let $U\subset\mathbb R^3$ be open. The association from smooth vector fields to flux $2$-forms is the map \begin{align*} C^\infty(U;\mathbb R^3)&\longrightarrow \Omega^2(U),\\ F=(F_1,F_2,F_3)&\longmapsto \eta_F=F_1\,dy\wedge dz+F_2\,dz\wedge dx+F_3\,dx\wedge dy. \end{align*} [/definition] Flux through a closed boundary should measure the total source strength inside the region. The obstruction is to show that the exterior derivative of the flux form produces exactly the divergence density, so the general Stokes theorem becomes the classical divergence theorem. [quotetheorem:3614] [citeproof:3614] The [divergence theorem](/theorems/3614) is the top-degree case in dimension three. It is often the most efficient form when a volume integral is easier than a surface integral, or conversely when boundary flux can be read from symmetry. The outward orientation is essential: using the inward normal multiplies the boundary integral by $-1$. Compactness and smooth boundary rule out escape of flux through infinity or singular boundary pieces; versions for unbounded domains or rough boundaries require additional decay or weak trace hypotheses. The theorem also does not imply that zero total flux means zero divergence everywhere, since positive and negative divergence can cancel in the integral. [example: Flux Through A Sphere] Let $R>0$, let $B=\{x\in\mathbb R^3:|x|\le R\}$ have the standard orientation, and give $\partial B$ the outward boundary orientation. For $F(x,y,z)=(x,y,z)$, we compute the outward flux through $\partial B$. The divergence is \begin{align*} \nabla\cdot F &=\frac{\partial x}{\partial x}+\frac{\partial y}{\partial y}+\frac{\partial z}{\partial z}\\ &=1+1+1\\ &=3. \end{align*} By *Gauss [Divergence Theorem](/theorems/2754)*, \begin{align*} \iint_{\partial B}F\cdot n\,dS &=\iiint_B \nabla\cdot F\,dV\\ &=\iiint_B 3\,dV. \end{align*} Using spherical coordinates \begin{align*} x=\rho\sin\phi\cos\theta,\qquad y=\rho\sin\phi\sin\theta,\qquad z=\rho\cos\phi, \end{align*} with $0\le \rho\le R$, $0\le\phi\le\pi$, $0\le\theta\le 2\pi$, and $dV=\rho^2\sin\phi\,d\rho\,d\phi\,d\theta$, this becomes \begin{align*} \iiint_B 3\,dV &=\int_0^{2\pi}\int_0^\pi\int_0^R 3\rho^2\sin\phi\,d\rho\,d\phi\,d\theta\\ &=\left(\int_0^{2\pi}d\theta\right)\left(\int_0^\pi \sin\phi\,d\phi\right)\left(\int_0^R 3\rho^2\,d\rho\right)\\ &=(2\pi)\left[-\cos\phi\right]_0^\pi\left[\rho^3\right]_0^R\\ &=(2\pi)\bigl(-\cos\pi+\cos0\bigr)(R^3-0)\\ &=(2\pi)(1+1)R^3\\ &=4\pi R^3. \end{align*} The same value is obtained from the surface integral itself. At a boundary point $p\in\partial B$, the outward unit normal is $n(p)=p/R$, so \begin{align*} F(p)\cdot n(p) &=p\cdot\frac{p}{R}\\ &=\frac{|p|^2}{R}\\ &=\frac{R^2}{R}\\ &=R. \end{align*} Parametrize the sphere by \begin{align*} \sigma(\phi,\theta) =\bigl(R\sin\phi\cos\theta,\ R\sin\phi\sin\theta,\ R\cos\phi\bigr). \end{align*} Then \begin{align*} |\sigma_\phi\times\sigma_\theta| &=R^2\sin\phi, \end{align*} so \begin{align*} \iint_{\partial B}F\cdot n\,dS &=\int_0^{2\pi}\int_0^\pi R\cdot R^2\sin\phi\,d\phi\,d\theta\\ &=R^3\left(\int_0^{2\pi}d\theta\right)\left(\int_0^\pi\sin\phi\,d\phi\right)\\ &=R^3(2\pi)(2)\\ &=4\pi R^3. \end{align*} Thus the volume calculation and the outward surface-flux calculation agree, and the outward flux through the sphere is $4\pi R^3$. [/example] This example illustrates the role of the chosen normal. Inward orientation would multiply the boundary integral by $-1$, while the standard orientation of the ball corresponds to the outward normal in the theorem. ## How The Classical Theorems Fit Together The final question is what the three classical theorems have in common. They differ mainly in the degree of the form and the dimension of the manifold, not in the underlying mechanism. [explanation: Degree And Dimension] Green theorem integrates a $1$-form over the boundary of a $2$-dimensional region and integrates its [exterior derivative](/theorems/1525) over the region. Classical Stokes theorem does the same for a $1$-form on a $2$-dimensional surface in $\mathbb R^3$. Gauss theorem integrates a $2$-form over the boundary of a $3$-dimensional region and integrates its [exterior derivative](/theorems/1525) over the solid. In all cases, the pattern is \begin{align*} \text{integral of derivative over region} =\text{integral of original form over boundary}. \end{align*} The apparent vector operations $\operatorname{curl}$ and $\operatorname{div}$ arise only after using the Euclidean metric and the standard volume form to identify forms with vector fields. [/explanation] The exterior calculus formulation separates the theorem from these Euclidean identifications. That is why the same statement works on abstract oriented manifolds, where there may be no preferred dot product, no cross product, and no global coordinate system. [remark: Orientation Summary] For an oriented interval, the boundary is endpoint minus starting point. For an oriented plane region, the positive boundary orientation is counterclockwise with respect to the standard area orientation. For an oriented surface in $\mathbb R^3$, the induced boundary orientation is given by the right-hand rule. For an oriented solid in $\mathbb R^3$, the induced boundary orientation is the outward orientation. [/remark] Stokes theorem is therefore the unifying endpoint of the first arc of the course. Exterior algebra supplies the correct objects, the [exterior derivative](/theorems/1525) supplies the correct differential operation, pullback and orientation make integration invariant, and the boundary orientation makes the signs work. Closed forms are the differential-form analogue of curl-free fields, while exact forms are those that arise as derivatives of lower-degree forms. With that distinction in hand, the next chapter turns to the global obstructions captured by Poincaré's lemma and de Rham cohomology. # 8. Closed and Exact Forms; Poincaré's Lemma Closed forms are the differential-form version of curl-free vector fields: their [exterior derivative](/theorems/1525) vanishes, so the first-order compatibility conditions have been met. Exact forms are stronger: they actually arise as exterior derivatives of lower-degree forms. This chapter asks when these two notions coincide, why topology can prevent coincidence, and how the homotopy operator proves that star-shaped open sets have no positive-degree de Rham cohomology. The central result is Poincaré's Lemma. It says that on a domain that contracts linearly to a point, every closed form of positive degree has a primitive. The proof gives more than existence: it writes down a concrete operator $K$ which constructs the primitive and satisfies the identity $dK+Kd=\operatorname{id}$. ## The Exactness Problem for Differential Forms When does the equation $d\eta=\omega$ have a solution, and what condition can be checked before trying to solve it? The condition $d\omega=0$ is necessary because applying $d$ twice always gives zero. The main question of this chapter is whether that necessary condition is also sufficient, and the answer depends on the shape of the domain. [definition: Closed and Exact Form] Let $U\subset \mathbb R^n$ be open, and let $d^k:\Omega^k(U)\to \Omega^{k+1}(U)$ denote the [exterior derivative](/theorems/1525) on $k$-forms. A form $\omega\in \Omega^k(U)$ is closed if $d\omega=0$. For $k\geq 1$, a form $\omega\in \Omega^k(U)$ is exact if there exists $\eta\in \Omega^{k-1}(U)$ such that $\omega=d\eta$. [/definition] The closed $k$-forms form $\ker d^k$, and the exact $k$-forms form $\operatorname{im} d^{k-1}$. The first universal test is one-way: if a form is exact, applying $d$ again must kill it. The next theorem records this necessary compatibility condition before we ask when the condition is also sufficient. [quotetheorem:3565] [citeproof:3565] This theorem is the source of the compatibility conditions in vector calculus. Its hypothesis is genuinely stronger than closedness on an arbitrary domain: the angle form on $\mathbb R^2\setminus\{0\}$ studied below is closed but not exact. The result also does not say how to find a primitive, or that a primitive exists; it only says that any proposed primitive must pass the test $d\omega=0$. For $1$-forms in the plane, this test is the equality of mixed partial derivatives which appears in the usual test for conservative vector fields, and it motivates asking when the compatibility condition is sufficient. [example: Recovering the Plane Curl Condition] Let $U\subset \mathbb R^2$ be open with coordinates $(x,y)$, and let $\omega=P\,dx+Q\,dy$ where $P,Q\in C^1(U)$. Suppose $\omega$ is exact, so $\omega=df$ for some smooth function $f:U\to\mathbb R$. Since \begin{align*} df=\frac{\partial f}{\partial x}\,dx+\frac{\partial f}{\partial y}\,dy, \end{align*} comparison of the $dx$ and $dy$ coefficients in $df=P\,dx+Q\,dy$ gives $P=\partial f/\partial x$ and $Q=\partial f/\partial y$. Now compute $d\omega$ term by term. Using $d(dx)=d(dy)=0$, $dx\wedge dx=0$, $dy\wedge dy=0$, and $dy\wedge dx=-dx\wedge dy$, we have \begin{align*} d\omega &=d(P\,dx)+d(Q\,dy)\\ &=dP\wedge dx+dQ\wedge 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\\ &=-\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*} Because $\omega=df$, we also have $d\omega=d(df)=0$. Hence the coefficient of $dx\wedge dy$ must vanish: \begin{align*} \frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}=0, \end{align*} so \begin{align*} \frac{\partial P}{\partial y}=\frac{\partial Q}{\partial x}. \end{align*} Thus exactness of $P\,dx+Q\,dy$ forces precisely the classical two-dimensional curl-free condition. [/example] The converse direction is not purely a matter of partial derivatives. A closed form can fail to be exact because the domain contains a hole, so integrating the form around a closed loop can detect global information that local differentiation does not see. [example: A Local Compatibility Condition With a Global Obstruction] On $U=\mathbb R^2\setminus\{0\}$ consider \begin{align*} \alpha=\frac{-y\,dx+x\,dy}{x^2+y^2}. \end{align*} Write $\alpha=P\,dx+Q\,dy$, where \begin{align*} P(x,y)=\frac{-y}{x^2+y^2}, \qquad Q(x,y)=\frac{x}{x^2+y^2}. \end{align*} Then \begin{align*} \frac{\partial P}{\partial y} &=\frac{(-1)(x^2+y^2)-(-y)(2y)}{(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*} and \begin{align*} \frac{\partial Q}{\partial x} &=\frac{(1)(x^2+y^2)-x(2x)}{(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*} Using $d(P\,dx+Q\,dy)=(\partial Q/\partial x-\partial P/\partial y)\,dx\wedge dy$, we get \begin{align*} d\alpha &=\left(\frac{y^2-x^2}{(x^2+y^2)^2} -\frac{y^2-x^2}{(x^2+y^2)^2}\right)dx\wedge dy\\ &=0. \end{align*} Thus the local closedness condition is satisfied. Now let $\gamma:[0,2\pi]\to U$ be the unit circle \begin{align*} \gamma(t)=(\cos t,\sin t). \end{align*} Along $\gamma$, \begin{align*} \gamma^*x=\cos t,\qquad \gamma^*y=\sin t,\qquad \gamma^*dx=-\sin t\,dt,\qquad \gamma^*dy=\cos t\,dt, \end{align*} so \begin{align*} \gamma^*\alpha &=\frac{-(\sin t)(-\sin t\,dt)+(\cos t)(\cos t\,dt)} {\cos^2 t+\sin^2 t}\\ &=\frac{\sin^2 t+\cos^2 t}{1}\,dt\\ &=dt. \end{align*} Therefore \begin{align*} \int_\gamma \alpha =\int_0^{2\pi}\gamma^*\alpha =\int_0^{2\pi}dt =2\pi. \end{align*} If $\alpha$ had a global primitive $f$ on $U$, then $\alpha=df$, and by the chain rule \begin{align*} \gamma^*(df)=d(f\circ\gamma)=\frac{d}{dt}(f(\gamma(t)))\,dt. \end{align*} Hence \begin{align*} \int_\gamma df &=\int_0^{2\pi}\frac{d}{dt}(f(\gamma(t)))\,dt\\ &=f(\gamma(2\pi))-f(\gamma(0))\\ &=f(1,0)-f(1,0)\\ &=0. \end{align*} This contradicts $\int_\gamma\alpha=2\pi$, so $\alpha$ is closed but not exact on $\mathbb R^2\setminus\{0\}$; the obstruction is global, detected by one trip around the missing origin. [/example] ## Star-Shaped Domains and Poincaré's Lemma Which domains remove the global obstruction and make closedness sufficient for exactness? A star-shaped [open set](/page/Open%20Set) has a preferred centre from which every point can be reached by a straight line segment remaining inside the domain. This linear contraction is strong enough to build primitives by integrating along the radial direction. [definition: Star-Shaped Open Set] Let $U\subset \mathbb R^n$ be open. The set $U$ is star-shaped with centre $a\in U$ if for every $x\in U$ and every $t\in [0,1]$, the point $a+t(x-a)$ lies in $U$. [/definition] Star-shaped sets include balls, convex open sets, and many domains that are not convex but still contract to one point by straight-line paths. This geometry is strong enough to remove the global obstruction to primitives: the straight-line contraction gives a concrete way to build a primitive from a closed form. [quotetheorem:3615] [citeproof:3615] The star-shaped hypothesis is doing real work: on $\mathbb R^2\setminus\{0\}$ the angle form is closed but has no global primitive. The theorem does not say primitives are unique, since adding a closed $(k-1)$-form to a primitive gives another primitive; in degree $1$, this includes adding constants to potentials on connected domains. What it gives is a constructive sufficient condition for existence. The point of the lemma is not that primitives can be guessed, but that the geometry of the domain supplies a canonical construction. The next section records the operator used in the proof and the identity it satisfies. [example: Constructing a Potential by Radial Integration] Let $U=\mathbb R^2$, which is star-shaped with centre $0$, and let \begin{align*} \omega=2xy\,dx+x^2\,dy. \end{align*} Writing $\omega=P\,dx+Q\,dy$ with $P(x,y)=2xy$ and $Q(x,y)=x^2$, we first check the closedness condition: \begin{align*} \frac{\partial Q}{\partial x}=2x, \qquad \frac{\partial P}{\partial y}=2x, \end{align*} so \begin{align*} d\omega =\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)dx\wedge dy =(2x-2x)\,dx\wedge dy =0. \end{align*} For a $1$-form $P\,dx+Q\,dy$ on a star-shaped domain centred at $0$, the radial homotopy operator is \begin{align*} K(P\,dx+Q\,dy)(x,y)=\int_0^1 \bigl(xP(tx,ty)+yQ(tx,ty)\bigr)\,dt. \end{align*} Here \begin{align*} P(tx,ty)=2(tx)(ty)=2t^2xy, \qquad Q(tx,ty)=(tx)^2=t^2x^2. \end{align*} Therefore \begin{align*} K\omega(x,y) &=\int_0^1 \bigl(x(2t^2xy)+y(t^2x^2)\bigr)\,dt\\ &=\int_0^1 \bigl(2t^2x^2y+t^2x^2y\bigr)\,dt\\ &=\int_0^1 3t^2x^2y\,dt\\ &=3x^2y\int_0^1 t^2\,dt\\ &=3x^2y\left[\frac{t^3}{3}\right]_0^1\\ &=x^2y. \end{align*} Finally, \begin{align*} d(x^2y) &=\frac{\partial}{\partial x}(x^2y)\,dx+\frac{\partial}{\partial y}(x^2y)\,dy\\ &=2xy\,dx+x^2\,dy\\ &=\omega. \end{align*} Thus radial integration has constructed the potential $x^2y$ for $\omega$. [/example] ## The Homotopy Operator and the Identity $dK+Kd=\operatorname{id}$ How does the radial contraction produce a primitive rather than merely suggest one? The operator $K$ inserts the radial vector $x-a$ into the form and averages the result along the segment from $a$ to $x$. The power of $t$ accounts for the scaling of tangent vectors under the map $x\mapsto a+t(x-a)$. [definition: Radial Homotopy Operator] Let $U\subset \mathbb R^n$ be star-shaped with centre $a\in U$, and let $k\geq 1$. Define \begin{align*} K:\Omega^k(U)\to\Omega^{k-1}(U) \end{align*} by the following formula: for $\omega\in \Omega^k(U)$, \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*} for $x\in U$ and $v_1,\dots,v_{k-1}\in \mathbb R^n$. [/definition] The formula is best read as contraction with the radial vector field, pulled back along the contraction. The obstruction is to prove that this candidate really interacts with $d$ like a homotopy inverse, rather than merely having the right degree. [quotetheorem:3583] [citeproof:3583] This is a chain homotopy identity. The assumption that the straight-line homotopy stays inside $U$ is essential for this operator; on the punctured plane the same radial collapse would pass through the missing origin, and the angle form shows that no global positive-degree identity of this kind can hold there. The formula is also a statement about positive-degree forms, since the pullback by the constant map is not zero on $0$-forms. It says that the identity map on positive-degree forms is homologous, in the de Rham complex, to the zero map induced by collapsing $U$ to the centre. This algebraic identity is what turns the analytic construction of $K$ into the cohomological vanishing statement in the next section. [example: The Operator on Two-Forms in Three Variables] Let $U\subset \mathbb R^3$ be star-shaped with centre $0$, and let \begin{align*} \omega=R(x,y,z)\,dy\wedge dz+S(x,y,z)\,dz\wedge dx+T(x,y,z)\,dx\wedge dy. \end{align*} Since $\omega$ is a $2$-form, the radial homotopy operator produces a $1$-form. At the point $(x,y,z)$, the definition gives \begin{align*} (K\omega)_{(x,y,z)}(v) = \int_0^1 t\,\omega_{(tx,ty,tz)}\bigl((x,y,z),v\bigr)\,dt. \end{align*} Write $v=a\,\partial_x+b\,\partial_y+c\,\partial_z$. We evaluate each wedge term on the pair $((x,y,z),v)$: \begin{align*} (dy\wedge dz)((x,y,z),v) &=dy(x,y,z)\,dz(v)-dy(v)\,dz(x,y,z)\\ &=y c-b z,\\ (dz\wedge dx)((x,y,z),v) &=dz(x,y,z)\,dx(v)-dz(v)\,dx(x,y,z)\\ &=z a-c x,\\ (dx\wedge dy)((x,y,z),v) &=dx(x,y,z)\,dy(v)-dx(v)\,dy(x,y,z)\\ &=x b-a y. \end{align*} Therefore \begin{align*} \omega_{(tx,ty,tz)}((x,y,z),v) ={}&R(tx,ty,tz)(yc-bz)\\ &+S(tx,ty,tz)(za-cx)\\ &+T(tx,ty,tz)(xb-ay). \end{align*} Collecting the coefficients of $a$, $b$, and $c$ gives \begin{align*} \omega_{(tx,ty,tz)}((x,y,z),v) ={}&a\bigl(zS(tx,ty,tz)-yT(tx,ty,tz)\bigr)\\ &+b\bigl(xT(tx,ty,tz)-zR(tx,ty,tz)\bigr)\\ &+c\bigl(yR(tx,ty,tz)-xS(tx,ty,tz)\bigr). \end{align*} Thus the $dx$, $dy$, and $dz$ coefficients of $K\omega$ are the corresponding coefficients of $a$, $b$, and $c$ after integrating in $t$: \begin{align*} K\omega={}&\left(\int_0^1 t\bigl(zS(tx,ty,tz)-yT(tx,ty,tz)\bigr)\,dt\right)dx\\ &+\left(\int_0^1 t\bigl(xT(tx,ty,tz)-zR(tx,ty,tz)\bigr)\,dt\right)dy\\ &+\left(\int_0^1 t\bigl(yR(tx,ty,tz)-xS(tx,ty,tz)\bigr)\,dt\right)dz. \end{align*} In particular, the $dy$ coefficient comes exactly from the $-zR(tx,ty,tz)$ contribution of $R\,dy\wedge dz$ and the $xT(tx,ty,tz)$ contribution of $T\,dx\wedge dy$. If $d\omega=0$, then the homotopy identity gives \begin{align*} d(K\omega)+K(d\omega)=\omega, \end{align*} so \begin{align*} d(K\omega)=\omega. \end{align*} Thus the displayed $1$-form is the primitive constructed by radial contraction whenever $\omega$ is closed. [/example] ## De Rham Cohomology and the Angle Form What does the failure of exactness measure when a closed form has no primitive? De Rham cohomology records closed forms modulo exact forms. In this language, Poincaré's Lemma says that star-shaped domains have no positive-degree de Rham cohomology, while punctured domains can carry nonzero classes. [definition: De Rham Cohomology of an Open Set] Let $U\subset \mathbb R^n$ be open. For $k\geq 1$, the $k$th de Rham cohomology [vector space](/page/Vector%20Space) of $U$ is \begin{align*} H^k_{\mathrm{dR}}(U)=\ker(d^k:\Omega^k(U)\to \Omega^{k+1}(U))\,/\,\operatorname{im}(d^{k-1}:\Omega^{k-1}(U)\to \Omega^k(U)). \end{align*} [/definition] The quotient is defined because [exact forms are closed](/theorems/3565). A closed form represents the zero class precisely when it is exact, so the natural question is what this quotient becomes on domains where Poincaré's lemma supplies primitives for all positive-degree closed forms. [quotetheorem:832] [citeproof:832] The star-shaped hypothesis cannot simply be omitted: the punctured plane has a closed $1$-form with nonzero integral around the unit circle. In degree $0$, the theorem records the separate fact that smooth functions with zero differential are constant on the connected star-shaped domain. In positive degree, its role here is to translate Poincaré's Lemma into the language of quotient vector spaces: every closed form represents the zero cohomology class. The first cohomology group has a concrete interpretation: it detects closed $1$-forms whose line integrals around closed loops may be nonzero. The punctured plane is the basic example, and the angle form is the standard detector. [quotetheorem:3616] [citeproof:3616] This example explains why the hypotheses of Poincaré's Lemma include a contraction assumption on the domain. The theorem does not compute the whole space $H^1_{\mathrm{dR}}(\mathbb R^2\setminus\{0\})$; it only shows that it contains a nonzero class. The failure is global rather than local: every sufficiently small ball in the punctured plane is star-shaped, so Poincaré's Lemma gives a primitive there. What fails is compatibility of these local primitives after going once around the hole, which is why the next example studies a branch cut. [example: Local Angle Primitives] Let $U_+=\mathbb R^2\setminus\{(x,0):x\leq 0\}$, and restrict the angle form \begin{align*} \alpha=\frac{-y\,dx+x\,dy}{x^2+y^2} \end{align*} to $U_+$. On this cut plane the principal argument $\theta=\operatorname{Arg}(x+iy)\in(-\pi,\pi)$ is a smooth real-valued function, and we show that $d\theta=\alpha$. Write \begin{align*} r=\sqrt{x^2+y^2}. \end{align*} Then $r>0$ on $U_+$, and the polar identities are \begin{align*} x=r\cos\theta,\qquad y=r\sin\theta. \end{align*} Differentiating these two identities gives \begin{align*} dx&=\cos\theta\,dr-r\sin\theta\,d\theta,\\ dy&=\sin\theta\,dr+r\cos\theta\,d\theta. \end{align*} Substitute these expressions into the numerator of $\alpha$: \begin{align*} -y\,dx+x\,dy &=-(r\sin\theta)(\cos\theta\,dr-r\sin\theta\,d\theta) +(r\cos\theta)(\sin\theta\,dr+r\cos\theta\,d\theta)\\ &=-r\sin\theta\cos\theta\,dr+r^2\sin^2\theta\,d\theta +r\sin\theta\cos\theta\,dr+r^2\cos^2\theta\,d\theta\\ &=r^2(\sin^2\theta+\cos^2\theta)\,d\theta\\ &=r^2\,d\theta. \end{align*} Since $r^2=x^2+y^2$, this becomes \begin{align*} -y\,dx+x\,dy=(x^2+y^2)d\theta. \end{align*} Dividing by $x^2+y^2$, which is nonzero on $U_+$, gives \begin{align*} d\theta=\frac{-y\,dx+x\,dy}{x^2+y^2}=\alpha. \end{align*} Thus $\alpha=d\theta$ on $U_+$, so the angle form is exact after the branch cut is made. Changing the branch cut changes the chosen angle function by integer multiples of $2\pi$ on overlaps, and the jump of $2\pi$ after one full turn is exactly the obstruction detected by the integral around the origin. [/example] The chapter therefore separates three levels of information. Closedness is the differential condition $d\omega=0$; exactness is the existence of a primitive; de Rham cohomology measures the global obstruction to passing from the first condition to the second. On star-shaped sets the homotopy formula kills the obstruction, while on $\mathbb R^2\setminus\{0\}$ the angle form records it. The abstract theory now returns to familiar vector calculus, but with a clearer geometric meaning. Having translated line, surface, and flux integrals into the language of forms, the final chapter uses [Stokes' theorem](/theorems/1530) to organize those classical operations into a single framework. # 9. Applications and Classical Vector Calculus Revisited The preceding chapters showed that line integrals, surface integrals, and flux integrals are all instances of [integration of differential forms](/theorems/1529), with [Stokes' theorem](/theorems/1530) as the organising result. This final chapter returns to the classical vector calculus language and translates its familiar operations into exterior calculus. The aim is not to replace the old notation in every computation, but to understand why the identities of vector calculus have the form they do and where topology enters the discussion. The chapter has four themes. First, conservative vector fields are exact $1$-forms, and path independence is the integral manifestation of exactness. Second, the sequence of maps behind gradient, curl, and divergence is the de Rham complex in low degrees. Third, the Hodge star uses a Euclidean metric and an orientation to identify complementary degrees of forms. Finally, Maxwell's equations illustrate how much structure is compressed by the notation $dF=0$ and $d*F=*J$. ## Conservative Vector Fields and Exact One-Forms When does a vector field have a scalar potential, and how can the answer be detected without guessing the potential? Exterior calculus separates the local differential test from the global topological issue. The local test is closedness; the global issue is whether closed $1$-forms are exact on the domain. [definition: Closed and Exact Forms] Let $U \subset \mathbb R^n$ be open. A differential form $\omega \in \Omega^k(U)$ is closed if $d\omega=0$. A differential form $\omega \in \Omega^k(U)$ with $k\geq 1$ is exact if there exists $\eta \in \Omega^{k-1}(U)$ such that $d\eta=\omega$. [/definition] Since $d^2=0$, every exact form is closed. The reverse implication is the first place in this course where the topology of the domain matters. [definition: Conservative Vector Field] Let $U \subset \mathbb R^n$ be open. A smooth vector field $F\in C^\infty(U;\mathbb R^n)$ is conservative if there exists $f\in C^\infty(U)$ such that $F=\nabla f$. The associated $1$-form map is \begin{align*} \Phi: C^\infty(U;\mathbb R^n)&\longrightarrow \Omega^1(U),\\ F=(F_1,\dots,F_n)&\longmapsto \omega_F=\sum_{i=1}^n F_i\,dx_i. \end{align*} [/definition] With the Euclidean metric, $F=\nabla f$ is the same statement as $\omega_F=df$. The vector-calculus question is therefore the exactness question for $1$-forms: when can a line integral be recovered from a potential evaluated only at endpoints? [quotetheorem:3617] [citeproof:3617] This theorem explains why potentials are useful: once $\omega=df$, line integrals no longer require a parametrisation of the whole path. The path-connectedness assumption is a bookkeeping condition for constructing one potential from a single base point; if $U=(0,1)\cup(2,3)\subset \mathbb R$, the construction must be made separately on the two components and additive constants cannot be compared across components. The theorem does not say that every closed form is exact: on $\mathbb R^2\setminus\{0\}$ the angular form below is closed but has nonzero integral around the unit circle. The next result adds exactly the missing global hypothesis by requiring simple connectedness, which turns the local condition $d\omega=0$ into path independence. [quotetheorem:3618] [citeproof:3618] The hypothesis on the domain is not cosmetic. On domains with holes, a closed $1$-form may record circulation around a missing region and fail to be exact; the form $\frac{-y\,dx+x\,dy}{x^2+y^2}$ on $\mathbb R^2\setminus\{0\}$ is the standard counterexample. The theorem does not assert that every [open set](/page/Open%20Set) has no such obstruction, nor does it give a practical method for computing a potential when simple connectedness fails. The following examples separate the two tasks: first checking exactness on a contractible domain, then seeing how a nonzero period detects a missing point. [example: Conservative Field In Three Space] For $F(x,y,z)=(yz,xz,xy)$, the associated $1$-form is \begin{align*} \omega_F=yz\,dx+xz\,dy+xy\,dz. \end{align*} We first compute its [exterior derivative](/theorems/1525), using the product rule and the identities $dx\wedge dx=dy\wedge dy=dz\wedge dz=0$: \begin{align*} d\omega_F &=d(yz)\wedge dx+d(xz)\wedge dy+d(xy)\wedge dz\\ &=(z\,dy+y\,dz)\wedge dx+(z\,dx+x\,dz)\wedge dy+(y\,dx+x\,dy)\wedge dz\\ &=z\,dy\wedge dx+y\,dz\wedge dx+z\,dx\wedge dy+x\,dz\wedge dy+y\,dx\wedge dz+x\,dy\wedge dz\\ &=-z\,dx\wedge dy+y\,dz\wedge dx+z\,dx\wedge dy-x\,dy\wedge dz-y\,dz\wedge dx+x\,dy\wedge dz\\ &=(-z+z)\,dx\wedge dy+(y-y)\,dz\wedge dx+(-x+x)\,dy\wedge dz\\ &=0. \end{align*} Thus $\omega_F$ is closed; since $\mathbb R^3$ is simply connected, *Closed One Forms on Simply Connected Domains* gives exactness. The proposed potential is checked by differentiating $f(x,y,z)=xyz$: \begin{align*} df &=\frac{\partial}{\partial x}(xyz)\,dx+\frac{\partial}{\partial y}(xyz)\,dy+\frac{\partial}{\partial z}(xyz)\,dz\\ &=yz\,dx+xz\,dy+xy\,dz\\ &=\omega_F. \end{align*} Equivalently, $\nabla f=(yz,xz,xy)=F$, so $F$ is conservative, and on the connected domain $\mathbb R^3$ this potential is determined up to an additive constant. [/example] The computation above works because $\mathbb R^3$ has no topological obstruction, so the local equation $d\omega_F=0$ already forces a global potential. That is not a feature of closedness itself: closedness only compares mixed partial derivatives in small coordinate neighbourhoods. The next example is the standard warning that exactness can fail when loops cannot be contracted through the domain, and it gives the counterexample promised after the simply connected theorem. [example: Punctured Plane Circulation] On $U=\mathbb R^2\setminus\{0\}$, consider \begin{align*} \omega=\frac{-y\,dx+x\,dy}{x^2+y^2}. \end{align*} We show that $\omega$ is closed on $U$ but not exact. Write \begin{align*} P(x,y)&=\frac{-y}{x^2+y^2},& Q(x,y)&=\frac{x}{x^2+y^2}, \end{align*} so that $\omega=P\,dx+Q\,dy$. Since $x^2+y^2\neq 0$ on $U$, \begin{align*} P_x&=\frac{2xy}{(x^2+y^2)^2},& P_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},\\ Q_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},& Q_y&=-\frac{2xy}{(x^2+y^2)^2}. \end{align*} Therefore \begin{align*} d\omega &=dP\wedge dx+dQ\wedge dy\\ &=(P_x\,dx+P_y\,dy)\wedge dx+(Q_x\,dx+Q_y\,dy)\wedge dy\\ &=P_y\,dy\wedge dx+Q_x\,dx\wedge dy\\ &=(-P_y+Q_x)\,dx\wedge dy\\ &=\left(-\frac{y^2-x^2}{(x^2+y^2)^2}+\frac{y^2-x^2}{(x^2+y^2)^2}\right)dx\wedge dy\\ &=0. \end{align*} Thus $\omega$ is closed. Now let $\gamma(t)=(\cos t,\sin t)$ for $0\leq t\leq 2\pi$. Along $\gamma$, \begin{align*} \gamma^*x&=\cos t,& \gamma^*y&=\sin t,& \gamma^*dx&=-\sin t\,dt,& \gamma^*dy&=\cos t\,dt. \end{align*} Hence \begin{align*} \gamma^*\omega &=\frac{-(\sin t)(-\sin t\,dt)+(\cos t)(\cos t\,dt)}{\cos^2t+\sin^2t}\\ &=\frac{\sin^2t+\cos^2t}{1}\,dt\\ &=dt, \end{align*} and therefore \begin{align*} \int_\gamma\omega =\int_0^{2\pi}dt =2\pi. \end{align*} If $\omega=df$ for some smooth function $f$ on $U$, then \begin{align*} \int_\gamma\omega &=\int_\gamma df\\ &=\int_0^{2\pi}(f\circ\gamma)'(t)\,dt\\ &=f(\gamma(2\pi))-f(\gamma(0))\\ &=f(1,0)-f(1,0)\\ &=0, \end{align*} contradicting $\int_\gamma\omega=2\pi$. Therefore $\omega$ is not exact. The nonzero integral records one positive winding of the unit circle around the missing origin. [/example] For a general closed $1$-form on the punctured plane, the obstruction to exactness is detected by its period around a generator of the fundamental group. In the language of this course, topology enters through the possible nonzero integrals of closed forms over closed cycles. ## The De Rham Complex Behind Gradient, Curl, and Divergence Why do the formulas $\operatorname{curl}(\nabla f)=0$ and $\operatorname{div}(\operatorname{curl}F)=0$ always appear together? Exterior calculus gives a single answer: both identities are instances of $d^2=0$. The familiar operators arise after identifying forms in $\mathbb R^3$ with scalar fields and vector fields. For an [open set](/page/Open%20Set) $U\subset \mathbb R^3$, use the following degree-by-degree identifications: \begin{align*} \Omega^0(U)&\ni f,\\ \Omega^1(U)&\ni A\,dx+B\,dy+C\,dz \longleftrightarrow (A,B,C),\\ \Omega^2(U)&\ni P\,dy\wedge dz+Q\,dz\wedge dx+R\,dx\wedge dy \longleftrightarrow (P,Q,R),\\ \Omega^3(U)&\ni h\,dx\wedge dy\wedge dz \longleftrightarrow h. \end{align*} The ordering of the $2$-form basis is chosen so that the signs match the usual orientation convention for curl and divergence. [quotetheorem:3619] [citeproof:3619] This is the conceptual gain of forms: the two vector identities no longer need separate verification. They are shadows of the same algebraic fact. The statement depends on the chosen orientation and on the three-dimensional identification of $2$-forms with vector fields; for example, replacing the basis term $dz\wedge dx$ by $dx\wedge dz$ reverses the sign of the middle curl component. The theorem does not say that every curl-free field is a gradient or every divergence-free field is a curl; on $U=(\mathbb R^2\setminus\{0\})\times \mathbb R$, the field $F(x,y,z)=(-y/(x^2+y^2),x/(x^2+y^2),0)$ has zero curl but is not a gradient because its line integral around the unit circle at $z=0$ is $2\pi$. The next theorem supplies a domain condition, star-shapedness, under which the low-degree de Rham complex has the stronger exactness property. [quotetheorem:832] [citeproof:832] The previous theorem is the Poincare lemma in the setting most useful for calculations in this course. It says that holes, not coordinate complexity, are what obstruct exactness; on $U=(\mathbb R^2\setminus\{0\})\times \mathbb R$, the pullback of the angular $1$-form is closed but not exact. The theorem does not say star-shapedness is necessary, only that it is a convenient geometric hypothesis that gives an explicit homotopy operator. The next example uses the theorem in degree $2$: instead of stopping at a closedness computation, it constructs a potential by solving $d\alpha=\omega$. [example: Closed Two Form With A Potential] Let \begin{align*} \omega=xy\,dx\wedge dy+yz\,dy\wedge dz \end{align*} on $\mathbb R^3$. We compute $d\omega$ using $d(f\eta)=df\wedge\eta+f\,d\eta$, $d(dx)=d(dy)=d(dz)=0$, and the fact that any wedge product with a repeated $1$-form is zero: \begin{align*} d\omega &=d(xy\,dx\wedge dy)+d(yz\,dy\wedge dz)\\ &=d(xy)\wedge dx\wedge dy+d(yz)\wedge dy\wedge dz\\ &=(y\,dx+x\,dy)\wedge dx\wedge dy+(z\,dy+y\,dz)\wedge dy\wedge dz\\ &=y\,dx\wedge dx\wedge dy+x\,dy\wedge dx\wedge dy +z\,dy\wedge dy\wedge dz+y\,dz\wedge dy\wedge dz\\ &=0+x(-dx\wedge dy\wedge dy)+0+y(-dy\wedge dz\wedge dz)\\ &=0. \end{align*} Thus $\omega$ is closed. To find a $1$-form potential of the form $\alpha=A\,dy$, compute \begin{align*} d(A\,dy) &=dA\wedge dy\\ &=(A_x\,dx+A_y\,dy+A_z\,dz)\wedge dy\\ &=A_x\,dx\wedge dy+A_y\,dy\wedge dy+A_z\,dz\wedge dy\\ &=A_x\,dx\wedge dy-A_z\,dy\wedge dz. \end{align*} Therefore $d(A\,dy)=xy\,dx\wedge dy+yz\,dy\wedge dz$ is achieved if \begin{align*} A_x=xy,\qquad -A_z=yz. \end{align*} For \begin{align*} A(x,y,z)=\frac{1}{2}y(x^2-z^2), \end{align*} the needed partial derivatives are \begin{align*} A_x&=\frac{1}{2}y\cdot 2x=xy,\\ A_y&=\frac{1}{2}(x^2-z^2),\\ A_z&=\frac{1}{2}y(-2z)=-yz. \end{align*} Hence, with \begin{align*} \alpha=\frac{1}{2}y(x^2-z^2)\,dy, \end{align*} we get \begin{align*} d\alpha &=d\left(\frac{1}{2}y(x^2-z^2)\right)\wedge dy\\ &=\left(xy\,dx+\frac{1}{2}(x^2-z^2)\,dy-yz\,dz\right)\wedge dy\\ &=xy\,dx\wedge dy+\frac{1}{2}(x^2-z^2)\,dy\wedge dy-yz\,dz\wedge dy\\ &=xy\,dx\wedge dy+0+yz\,dy\wedge dz\\ &=\omega. \end{align*} Thus $\omega$ is exact as well as closed; the displayed $\alpha$ is an explicit primitive for the $2$-form. [/example] ## The Hodge Star and Metric Duality How can a $1$-form be compared with a $2$-form, or a line integral operation with a flux operation? The [exterior derivative](/theorems/1525) only uses the smooth structure, while the Hodge star also uses the Euclidean metric and an orientation. This is why the Hodge star is the bridge from exterior calculus back to metric-dependent vector operations such as the cross product. [definition: Euclidean Inner Product on Forms] Let $V=\mathbb R^n$ have its standard Euclidean inner product, and let $dx_1,\dots,dx_n$ be the oriented orthonormal coframe. For $0\leq k\leq n$, the Euclidean inner product on $k$-forms is the map \begin{align*} \langle \cdot,\cdot\rangle:\Lambda^k(V^*)\times\Lambda^k(V^*)&\longrightarrow \mathbb R. \end{align*} For $I=(i_1<\cdots<i_k)$, write $dx_I=dx_{i_1}\wedge\cdots\wedge dx_{i_k}$. If \begin{align*} \alpha=\sum_I a_I\,dx_I, \qquad \beta=\sum_I b_I\,dx_I, \end{align*} then \begin{align*} \langle \alpha,\beta\rangle=\sum_I a_Ib_I. \end{align*} [/definition] This inner product on $k$-forms makes the standard wedge basis orthonormal. To compare a $k$-form with an $(n-k)$-form, we need an operator that converts metric information into oriented volume. The Hodge star is defined by requiring the wedge product with the original form to reproduce that volume pairing. [definition: Hodge Star] Let $V=\mathbb R^n$ have its standard Euclidean inner product and orientation, with volume form $\operatorname{vol}=dx_1\wedge\cdots\wedge dx_n$. The Hodge star is the [linear map](/page/Linear%20Map) \begin{align*} *: \Lambda^k(V^*)\to \Lambda^{n-k}(V^*) \end{align*} determined by \begin{align*} \alpha\wedge (*\beta)=\langle \alpha,\beta\rangle\operatorname{vol} \end{align*} for all $\alpha,\beta\in \Lambda^k(V^*)$. [/definition] Changing the orientation changes signs, and changing the metric changes the inner product used in the defining equation. In this chapter we only use the standard Euclidean metric and the standard orientation. [example: Hodge Star In Three Space] Work in $\mathbb R^3$ with the oriented orthonormal coframe $dx,dy,dz$ and volume form \begin{align*} \operatorname{vol}=dx\wedge dy\wedge dz. \end{align*} We compute $*(dx\wedge dy)$ from the defining equation \begin{align*} \alpha\wedge *\beta=\langle \alpha,\beta\rangle \operatorname{vol}. \end{align*} Write \begin{align*} *(dx\wedge dy)=a\,dx+b\,dy+c\,dz. \end{align*} Using the orthonormal basis $dy\wedge dz,\ dz\wedge dx,\ dx\wedge dy$ of $2$-forms, we get \begin{align*} (dy\wedge dz)\wedge *(dx\wedge dy) &=(dy\wedge dz)\wedge(a\,dx+b\,dy+c\,dz)\\ &=a\,dy\wedge dz\wedge dx\\ &=a\,dx\wedge dy\wedge dz, \end{align*} while \begin{align*} \langle dy\wedge dz,dx\wedge dy\rangle \operatorname{vol}=0. \end{align*} Thus $a=0$. Similarly, \begin{align*} (dz\wedge dx)\wedge *(dx\wedge dy) &=(dz\wedge dx)\wedge(a\,dx+b\,dy+c\,dz)\\ &=b\,dz\wedge dx\wedge dy\\ &=b\,dx\wedge dy\wedge dz, \end{align*} and \begin{align*} \langle dz\wedge dx,dx\wedge dy\rangle \operatorname{vol}=0, \end{align*} so $b=0$. Finally, \begin{align*} (dx\wedge dy)\wedge *(dx\wedge dy) &=(dx\wedge dy)\wedge(a\,dx+b\,dy+c\,dz)\\ &=c\,dx\wedge dy\wedge dz, \end{align*} and \begin{align*} \langle dx\wedge dy,dx\wedge dy\rangle \operatorname{vol} =dx\wedge dy\wedge dz, \end{align*} so $c=1$. Therefore \begin{align*} *(dx\wedge dy)=dz. \end{align*} The same coefficient test gives \begin{align*} *(dy\wedge dz)&=dx,\\ *(dz\wedge dx)&=dy,\\ *(dx\wedge dy)&=dz. \end{align*} These signs come from the chosen orientation: each of \begin{align*} dy\wedge dz\wedge dx,\qquad dz\wedge dx\wedge dy,\qquad dx\wedge dy\wedge dz \end{align*} equals the positive volume form $dx\wedge dy\wedge dz$. [/example] The signs in the Hodge star encode the orientation. The obstruction is that applying the star twice swaps complementary degrees twice, and the commutation of a $k$-form with an $(n-k)$-form can introduce a sign. The exact sign must be known before the star can be used reliably in vector-calculus identities. For the Euclidean Hodge star on $\mathbb R^n$, applying the star twice to a $k$-form gives \begin{align*} **\alpha=(-1)^{k(n-k)}\alpha. \end{align*} The sign formula is a metric-and-orientation statement, not a property of the wedge product alone. If the Euclidean metric is replaced by the Lorentzian metric used in spacetime, the corresponding $**$ formula changes sign in some degrees. The formula also does not by itself turn forms into vector fields, because that requires the metric duality between $V$ and $V^*$. That extra metric identification is the next ingredient and is what allows the cross product to be recovered from wedge product and Hodge star. [definition: Musical Isomorphisms] The flat map is \begin{align*} \flat:\mathbb R^n&\longrightarrow (\mathbb R^n)^*,\\ v&\longmapsto v^\flat, \end{align*} where $v^\flat(w)=v\cdot w$ for all $w\in \mathbb R^n$. The sharp map is \begin{align*} \sharp:(\mathbb R^n)^*&\longrightarrow \mathbb R^n,\\ \alpha&\longmapsto \alpha^\sharp, \end{align*} where $\alpha(w)=\alpha^\sharp\cdot w$ for all $w\in \mathbb R^n$. [/definition] These maps use the Euclidean inner product to pass between vectors and covectors. The final comparison with vector calculus needs all three ingredients at once: wedge product for alternation, Hodge star for degree conversion, and musical maps for translating covectors back into vectors. [quotetheorem:3620] [citeproof:3620] This identity is a useful memory aid: wedge product is the alternating multiplication, the Hodge star changes degree, and the musical maps translate between covectors and vectors. The dimension hypothesis is essential: in $\mathbb R^4$, $*(u^\flat\wedge v^\flat)$ is again a $2$-form, not a $1$-form, so the formula cannot define a vector-valued binary product with the usual three-dimensional type. The theorem also does not make the cross product metric-free; changing the metric or orientation changes the flat map or the Hodge star and therefore changes the resulting vector. Maxwell's equations use the same separation of roles: $d$ carries the topological differential identities, while $*$ supplies the metric-dependent duality. ## Maxwell's Equations as a Form Equation What does exterior calculus buy beyond cleaner notation for old vector identities? Maxwell's equations provide the standard answer: four coupled vector equations can be written as two differential-form equations. This course does not develop Lorentzian geometry, so the discussion here is only an illustration of the language. On spacetime with coordinates $(t,x,y,z)$, package the electric field $E=(E_1,E_2,E_3)$ and magnetic field $B=(B_1,B_2,B_3)$ into the electromagnetic $2$-form \begin{align*} F=B_1\,dy\wedge dz+B_2\,dz\wedge dx+B_3\,dx\wedge dy-E_1\,dt\wedge dx-E_2\,dt\wedge dy-E_3\,dt\wedge dz. \end{align*} Let $\rho$ denote charge density and let $j=(j_1,j_2,j_3)$ denote current density. These are packaged into a spacetime current $1$-form $J$, so $*J$ is a $3$-form. With a compatible Lorentzian Hodge star $*:\Omega^k\to \Omega^{4-k}$, Maxwell's equations take the compact form \begin{align*} dF&=0, & d*F&=*J. \end{align*} The first equation contains the absence of magnetic monopoles and Faraday's law, \begin{align*} \operatorname{div}B&=0, & \frac{\partial B}{\partial t}+\operatorname{curl}E&=0. \end{align*} The second equation contains Gauss' law and the Ampere-Maxwell law, \begin{align*} \operatorname{div}E&=\rho, & \operatorname{curl}B-\frac{\partial E}{\partial t}&=j, \end{align*} up to the sign convention chosen for the Lorentzian star and the current $J$. This compact form shows the two roles played throughout the course. The [exterior derivative](/theorems/1525) expresses the structure forced by [Stokes' theorem](/theorems/1530) and $d^2=0$; the Hodge star inserts metric information and turns geometric conservation laws into equations relating complementary degrees. Classical vector calculus is therefore not separate from exterior calculus, but a low-dimensional metric presentation of it. ## References - Spivak, M., *Calculus on Manifolds*, W. A. Benjamin. - Lee, J. M., *Introduction to Smooth Manifolds*, Springer. - Tu, L. W., *An Introduction to Manifolds*, Springer. - Bott, R. and Tu, L. W., *Differential Forms in Algebraic Topology*, Springer. - Flanders, H., *Differential Forms with Applications to the Physical Sciences*, Dover.