What's in a Differential Form? – Part 2

Sep 7, 2026 14 min read Mathematics

In Part 1, we arrived at differential one-forms by reconsidering the derivative. If $f$ is a smooth real-valued function on a manifold $M$, then its differential at $p$ is a linear functional $$ df_p:T_pM\longrightarrow \mathbb{R}. $$ More generally, a one-form $\omega$ assigns to each point $p$ a covector $\omega_p\in T_p^\ast M$, and it does so smoothly. In local coordinates $(x^1,\ldots,x^n)$, every one-form has an expression $$ \omega=\sum_{i=1}^n a_i dx^i. $$ This description explains the individual symbols $dx^i$, but it does not yet explain familiar expressions such as $dx\wedge dy$, nor does it tell us what sort of object can be integrated over a surface or a higher-dimensional manifold. To answer those questions, we must pass from linear algebra to multilinear algebra.

From Linear Maps to Tensors

Let $V_1,\ldots,V_k$, and $W$ be real vector spaces. A map $$ T:V_1\times\cdots\times V_k\longrightarrow W $$ is called multilinear if it is linear in each argument when all the other arguments are held fixed. Thus, in its $i$th argument, it satisfies $$ T(v_1,\ldots,av_i+bv_i’,\ldots,v_k) =aT(v_1,\ldots,v_i,\ldots,v_k) +bT(v_1,\ldots,v_i’,\ldots,v_k). $$ A covariant $k$-tensor on $V$ is a multilinear map $V^k\to\mathbb{R}$. A one-form at a point is therefore already a covariant one-tensor. The essential new feature for $k>1$ is that a tensor can compare several independent directions at once.

Given covectors $\alpha^1,\ldots,\alpha^k\in V^\ast$, their tensor product is the covariant $k$-tensor defined by $$ (\alpha^1\otimes\cdots\otimes\alpha^k)(v_1,\ldots,v_k) =\alpha^1(v_1)\cdots\alpha^k(v_k). $$ If $(e_1,\ldots,e_n)$ is a basis of $V$ and $(\varepsilon^1,\ldots,\varepsilon^n)$ is its dual basis, then the tensors $$ \varepsilon^{i_1}\otimes\cdots\otimes\varepsilon^{i_k} $$ form a basis for the space of covariant $k$-tensors. Consequently, any such tensor is determined by the collection of numbers obtained by evaluating it on ordered $k$-tuples of basis vectors.

Not every covariant tensor is a differential form. Differential forms are built from tensors satisfying an additional condition that encodes orientation.

Alternation and Oriented Volume

A covariant $k$-tensor $\omega$ is alternating if interchanging any two arguments changes its sign. If $\sigma$ is a permutation of $\lbrace 1,\ldots,k\rbrace$, this condition can be written $$ \omega(v_{\sigma(1)},\ldots,v_{\sigma(k)}) =\operatorname{sgn}(\sigma)\omega(v_1,\ldots,v_k). $$ In particular, an alternating tensor vanishes whenever two of its arguments agree. More generally, it vanishes whenever its arguments are linearly dependent. Indeed, if two arguments are equal, exchanging them leaves the input unchanged while alternation changes the sign of the output, so the output must be zero. Multilinearity then extends this observation to every linearly dependent tuple.

The space of alternating covariant $k$-tensors on $V$ is denoted $\Lambda^k(V^\ast)$. An element of this space is also called a $k$-covector. A differential $k$-form on a manifold $M$ is a smooth assignment $$ p\longmapsto \omega_p\in\Lambda^k(T_p^\ast M). $$ The vector space of smooth differential $k$-forms on $M$ is denoted $\Omega^k(M)$. Thus $\Omega^0(M)$ is the space $C^\infty(M)$ of smooth functions, while $\Omega^1(M)$ is the space of one-forms discussed in Part 1.

The alternating condition gives a $k$-form its geometric meaning. A $k$-form consumes $k$ tangent vectors and returns the signed $k$-dimensional volume of the parallelepiped that those vectors determine, measured according to the form. Reversing the orientation reverses the sign, and collapsing the parallelepiped into a lower-dimensional subspace makes the result zero. This interpretation requires no ambient Euclidean coordinates. It is intrinsic to the tangent space.

Consider the two-form $dx\wedge dy$ on $\mathbb{R}^2$. If $$ u=u^1\frac{\partial}{\partial x}+u^2\frac{\partial}{\partial y} \qquad\text{and}\qquad v=v^1\frac{\partial}{\partial x}+v^2\frac{\partial}{\partial y}, $$ then $$ (dx\wedge dy)(u,v) =u^1v^2-u^2v^1 =\det\begin{pmatrix}u^1&v^1\u^2&v^2\end{pmatrix}. $$ The determinant is precisely the signed area-scaling factor associated with the ordered pair $(u,v)$. The appearance of the determinant is not an incidental computational trick; it is the algebraic expression of alternation.

This also clarifies a subtle point about the discussion of duality in Part 1. On Euclidean space, a one-form $$ \omega=f dx+g dy+h dz $$ is commonly associated with the vector field $(f,g,h)$. That association uses the Euclidean inner product. On a general manifold, there is no canonical identification between tangent vectors and cotangent vectors. A Riemannian metric $g$ supplies one by assigning to a vector field $X$ the one-form $X^\flat=g(X,\mathord{\cdot})$, but without such a choice, vectors and covectors are genuinely different kinds of objects. Differential forms themselves require neither a metric nor a notion of angle or length.

The Wedge Product

The multiplication operation appropriate to differential forms is the wedge product. If $\alpha$ is a one-form and $\beta$ is a one-form, their wedge product is the two-form defined by $$ (\alpha\wedge\beta)(u,v) =\alpha(u)\beta(v)-\alpha(v)\beta(u). $$ Some authors include a normalizing factor in the abstract alternation operator; the convention above is the standard one for wedge products of one-forms and fixes the convention used throughout this article.

The two terms ensure that $\alpha\wedge\beta$ is alternating. In particular, $$ \alpha\wedge\beta=-\beta\wedge\alpha $$ and therefore $$ \alpha\wedge\alpha=0. $$ The equation $dx\wedge dx=0$ should not be understood as a mysterious rule for manipulating infinitesimals. It says that an alternating area measurement vanishes when it is given the same direction twice.

More generally, the wedge product combines a $k$-form and an $\ell$-form to produce a $(k+\ell)$-form, $$ \wedge:\Omega^k(M)\times\Omega^\ell(M)\longrightarrow\Omega^{k+\ell}(M). $$ It is bilinear and associative, and it satisfies the graded-commutativity law $$ \alpha\wedge\beta=(-1)^{k\ell}\beta\wedge\alpha $$ when $\alpha$ has degree $k$ and $\beta$ has degree $\ell$. Ordinary scalar functions are zero-forms, so if $f\in C^\infty(M)$, then $f\wedge\alpha$ is simply written $f\alpha$. Two two-forms commute under the wedge product, whereas a one-form and a three-form anticommute. The sign depends on the degrees of the forms rather than merely on the fact that two factors have been exchanged.

In a coordinate chart $(x^1,\ldots,x^n)$, a general $k$-form can be written uniquely as $$ \omega=\sum_{1\leq i_1<\cdots<i_k\leq n} a_{i_1\cdots i_k} dx^{i_1}\wedge\cdots\wedge dx^{i_k}. $$ Only strictly increasing index sequences are needed. Any repeated index makes a term zero, and any other ordering can be rearranged into increasing order at the cost of the sign of the corresponding permutation. It follows that an $n$-dimensional vector space has $\binom{n}{k}$ independent $k$-covectors. If $k>n$, every alternating $k$-tensor vanishes, so $\Omega^k(M)=\lbrace 0\rbrace$ on an $n$-dimensional manifold.

For example, every two-form on $\mathbb{R}^3$ has the form $$ \omega=P dy\wedge dz+Q dz\wedge dx+R dx\wedge dy. $$ The ordering in this expression is chosen to match the usual cyclic orientation. With the Euclidean metric and the standard orientation, the coefficient triple $(P,Q,R)$ may be identified with a vector field. As before, this identification is useful but depends on additional geometric structure. The two-form is intrinsically an oriented area-measuring object, not a vector in disguise.

Pulling Forms Back

Part 1 introduced the pullback of a one-form. The same construction applies in every degree. If $F:M\to N$ is smooth and $\omega\in\Omega^k(N)$, then $F^\ast\omega\in\Omega^k(M)$ is defined at $p\in M$ by $$ (F^\ast\omega)(p)\bigl(v^{(1)},\ldots,v^{(k)}\bigr) =\omega(F(p))\bigl(dF(p)v^{(1)},\ldots,dF(p)v^{(k)}\bigr). $$ The direction of this operation deserves emphasis. The differential $dF_p$ sends tangent vectors forward from $M$ to $N$, but the pullback sends covectors and differential forms backward from $N$ to $M$. It does so because a covector on the target can act on a source vector only after that vector has been pushed forward by $dF_p$.

Pullbacks respect the algebra of forms. For forms $\alpha$ and $\beta$ on $N$, $$ F^\ast(\alpha\wedge\beta)=F^\ast\alpha\wedge F^\ast\beta. $$ They also behave contravariantly under composition. If $G:N\to P$ is another smooth map, then $$ (G\circ F)^\ast=F^\ast\circ G^\ast. $$ These identities make pullback the correct language for both coordinate changes and parameterized integration.

Suppose, for example, that $F:U\subset\mathbb{R}^2\to\mathbb{R}^2$ is given in coordinates by $$ F(u,v)=\bigl(x(u,v),y(u,v)\bigr). $$ Since pulling back a coordinate one-form is the same as differentiating the corresponding coordinate function, $$ F^\ast(dx)=d(x\circ F)=x_u du+x_v dv $$ and similarly $$ F^\ast(dy)=y_u du+y_v dv. $$ Consequently, $$ \begin{aligned} F^\ast(dx\wedge dy) &=(x_u du+x_v dv)\wedge(y_u du+y_v dv)\ &=(x_uy_v-x_vy_u)du\wedge dv\ &=\det(DF)du\wedge dv. \end{aligned} $$ The Jacobian determinant in the change-of-variables formula therefore arises automatically from the alternating algebra of forms. Its sign records whether $F$ preserves or reverses orientation. When one instead integrates with respect to an unsigned measure, the absolute value $|\det(DF)|$ appears because orientation is deliberately discarded.

The Exterior Derivative

The ordinary differential sends a smooth function $f$, regarded as a zero-form, to the one-form $$ df=\sum_i\frac{\partial f}{\partial x^i}dx^i. $$ The exterior derivative extends this operation to forms of every degree. It is a linear map $$ d:\Omega^k(M)\longrightarrow\Omega^{k+1}(M). $$ In coordinates, if $$ \omega=\sum_{i_1<\cdots<i_k}a_{i_1\cdots i_k} dx^{i_1}\wedge\cdots\wedge dx^{i_k}, $$ then $$ d\omega= \sum_{i_1<\cdots<i_k} da_{i_1\cdots i_k}\wedge dx^{i_1}\wedge\cdots\wedge dx^{i_k}. $$ Since $$ da_{i_1\cdots i_k} =\sum_j\frac{\partial a_{i_1\cdots i_k}}{\partial x^j}dx^j, $$ this definition differentiates the coefficient functions and then uses the wedge product to incorporate the new direction of differentiation.

Although the formula is written using coordinates, the resulting form is independent of the chosen coordinate chart. One way to see why is to characterize $d$ by properties that make no reference to coordinates. It agrees with the ordinary differential on functions, it satisfies $d^2=0$, and it obeys the graded Leibniz rule $$ d(\alpha\wedge\beta) =d\alpha\wedge\beta+(-1)^k\alpha\wedge d\beta $$ for $\alpha\in\Omega^k(M)$. These requirements determine the exterior derivative uniquely. Moreover, exterior differentiation commutes with pullback: $$ d(F^\ast\omega)=F^\ast(d\omega). $$ This naturality is one reason the exterior derivative behaves so cleanly under changes of coordinates.

The identity $d^2=0$ generalizes the familiar fact that sufficiently smooth mixed partial derivatives commute. For a function $f$ on $\mathbb{R}^2$, $$ \begin{aligned} d(df) &=d(f_x dx+f_y dy)\ &=f_{yx}dy\wedge dx+f_{xy}dx\wedge dy\ &=(f_{xy}-f_{yx})dx\wedge dy=0. \end{aligned} $$ The cancellation combines two facts: mixed partials are equal, while exchanging $dx$ and $dy$ changes the sign.

A form $\omega$ is called closed if $d\omega=0$, and it is called exact if $\omega=d\eta$ for some form $\eta$ of one lower degree. Since $d^2=0$, every exact form is closed. The converse is true locally on a sufficiently small ball, by the Poincaré lemma, but it need not be true globally. On the punctured plane $\mathbb{R}^2\setminus\lbrace 0\rbrace$, the one-form $$ \omega=\frac{-y dx+x dy}{x^2+y^2} $$ is closed. It is not exact on the entire punctured plane, because its integral around the unit circle is $2\pi$, whereas the integral of an exact one-form around any closed curve must vanish. This failure is not caused by a defect in differentiation. It records the hole in the domain. The quotient of closed $k$-forms by exact $k$-forms is the $k$th de Rham cohomology group, which turns this analytic distinction into an invariant of the topology of $M$.

Exterior Differentiation in Three Dimensions

The vector-calculus operators gradient, curl, and divergence are all manifestations of the same exterior derivative. Their familiar vector-valued notation obscures this unity because it repeatedly uses the Euclidean metric to identify forms with vector fields.

For a function $f$ on $\mathbb{R}^3$, $$ df=f_x dx+f_y dy+f_z dz. $$ After identifying one-forms and vectors using the Euclidean inner product, the coefficients of $df$ form the gradient $\nabla f$.

For a one-form $$ \alpha=P dx+Q dy+R dz, $$ direct calculation gives $$ d\alpha =(R_y-Q_z)dy\wedge dz +(P_z-R_x)dz\wedge dx +(Q_x-P_y)dx\wedge dy. $$ Under the metric- and orientation-dependent identification of two-forms with vectors, these coefficients form $\nabla\times(P,Q,R)$. Thus the curl is the exterior derivative of a one-form, expressed as a vector field.

For a two-form $$ \beta=P dy\wedge dz+Q dz\wedge dx+R dx\wedge dy, $$ we obtain $$ d\beta=(P_x+Q_y+R_z)dx\wedge dy\wedge dz. $$ Relative to the standard volume form, the coefficient is the divergence $\nabla\cdot(P,Q,R)$. The identities $\nabla\times\nabla f=0$ and $\nabla\cdot(\nabla\times X)=0$ are consequently both instances of $d^2=0$.

This interpretation separates the intrinsic operation from the conveniences of Euclidean geometry. The exterior derivative exists on every smooth manifold without a metric. Gradient, curl, and divergence in their standard vector-valued forms require extra structure to convert between vectors, covectors, and volume forms.

Integrating Differential Forms

A $k$-form is the appropriate object to integrate over an oriented $k$-dimensional manifold. The matching of degrees is essential: a one-form is integrated along a curve, a two-form over a surface, and an $n$-form over an $n$-dimensional manifold. The form accepts an oriented tangent $k$-frame and returns the infinitesimal signed volume assigned to that frame.

Let $S$ be an oriented $k$-dimensional manifold parameterized by a smooth map $$ \varphi:U\subset\mathbb{R}^k\longrightarrow S. $$ For a compactly supported $k$-form $\omega$ on $S$, integration is defined locally by pulling the form back to the parameter domain: $$ \int_S\omega=\int_U\varphi^\ast\omega. $$ Every $k$-form on $U$ has the form $$ \varphi^\ast\omega=f(u^1,\ldots,u^k)du^1\wedge\cdots\wedge du^k, $$ so the right-hand side is computed using an ordinary multiple integral of the coefficient $f$. For a general manifold, one covers the support of the form by oriented coordinate charts and combines the local integrals with a partition of unity. The transformation law established by pullback guarantees that the final value does not depend on the chosen parameterizations.

Orientation is what makes the sign of this integral coherent. Reversing the orientation of $S$ changes the sign of $\int_S\omega$. If the desired quantity should not depend on orientation, as with mass or probability, then differential forms are not quite the intrinsic object one needs. One instead integrates a density, whose coordinate transformation law contains an absolute Jacobian determinant. This is the precise version of the signed-versus-unsigned distinction introduced at the beginning of Part 1.

As a concrete example, let $\gamma:[a,b]\to M$ be a smooth oriented curve and let $\omega$ be a one-form on $M$. Then $$ \int_\gamma\omega =\int_{[a,b]}\gamma^\ast\omega =\int_a^b\omega_{\gamma(t)}\bigl(\gamma’(t)\bigr)dt. $$ If the parameterization is reversed, then $\gamma’$ changes direction and the integral changes sign. When $M=\mathbb{R}^3$ and $\omega=P dx+Q dy+R dz$, this formula is the usual work integral $$ \int_a^b (P,Q,R)\bigl(\gamma(t)\bigr)\mathbin{\cdot}\gamma’(t)dt, $$ where the dot product enters only through the Euclidean identification of the one-form with a vector field.

Stokes’ Theorem

The fundamental theorem of calculus says that integrating the derivative of a function over an interval depends only on the values of the function at the boundary: $$ \int_a^b f’(x)dx=f(b)-f(a). $$ The generalized Stokes theorem states the same principle in every dimension. If $M$ is an oriented smooth $n$-manifold with boundary and $\omega$ is a compactly supported $(n-1)$-form, then $$ \boxed{\int_M d\omega=\int_{\partial M}\omega.} $$ The boundary $\partial M$ carries the orientation induced from $M$. Informally, an ordered basis of $T_p(\partial M)$ is positive when placing an outward-pointing normal vector before it produces a positive basis of $T_pM$. This convention is responsible for the signs at the endpoints of an interval and for the counterclockwise orientation of the boundary of a positively oriented planar region.

Many theorems ordinarily taught as separate results are instances of this single formula. If $M=[a,b]$ and $\omega=f$, then Stokes’ theorem is the fundamental theorem of calculus. If $M$ is an oriented surface and $\omega$ is a one-form, it becomes the classical Stokes theorem relating circulation around the boundary to the surface integral of curl. If $M$ is a region in the plane, the same case yields Green’s theorem. If $M$ is a three-dimensional region and $\omega$ is a two-form, it becomes the divergence theorem relating outward flux through the boundary to the integral of divergence over the interior.

The theorem also explains why exact one-forms have path-independent integrals. If $\omega=df$ and $\gamma$ is a curve from $p$ to $q$, then $$ \int_\gamma df=f(q)-f(p). $$ If $\gamma$ is closed, its boundary is empty and the integral is zero. Conversely, a nonzero integral around a closed curve certifies that the form cannot be exact on a region containing that curve. The punctured-plane example above uses precisely this obstruction.

Stokes’ theorem is the organizing principle behind the entire calculus of differential forms. The wedge product combines oriented measurements, the exterior derivative describes their infinitesimal variation, pullback transports them between spaces, and integration accumulates them over oriented domains. Stokes’ theorem then asserts that the accumulated variation in the interior is exactly the accumulated original quantity on the boundary.

What Is in a Differential Form?

It is tempting to regard symbols such as $dx$, $dy$, and $dx\wedge dy$ as infinitesimal lengths and areas. That intuition can be useful, provided it is not mistaken for a definition. At a point $p$, a differential $k$-form is an alternating multilinear functional on $k$ tangent vectors. Across a manifold, it is a smooth field of such functionals. Its alternation records orientation, its multilinearity records how signed volume scales in each direction, and its degree specifies the dimension of the objects over which it may be integrated.

This perspective resolves several apparent coincidences from elementary calculus. Jacobian determinants appear because top-degree alternating forms transform by determinants. Curl and divergence resemble one another because they are consecutive instances of the same operator $d$. The fundamental theorem, Green’s theorem, the classical Stokes theorem, and the divergence theorem have the same shape because they are all the generalized Stokes theorem in different degrees and dimensions.

Differential forms are therefore not merely a compact notation for vector calculus. They isolate the parts of calculus that depend only on smooth structure and orientation, leaving metrics to enter only when lengths, angles, orthogonality, or identifications with vector fields are genuinely required. That separation is what allows the same theory to operate naturally on curves, surfaces, higher-dimensional manifolds, and spaces whose geometry is far removed from ordinary Euclidean coordinates.