The Exterior Derivative on Euclidean Space
The differential of a smooth function turns a \(0\)-form into a \(1\)-form, recording how the function changes in every direction.
We now extend this operation to forms of every degree: a single differential operator that raises the degree of a form by one, agrees
with the ordinary differential on functions, and — most importantly — squares to zero. That last property, innocuous as it looks, is
the algebraic engine behind the classical integral theorems of vector calculus and the obstruction theory that distinguishes one manifold
from another.
We begin on an open subset of Euclidean space, where the definition is explicit, and afterward show that it transfers unambiguously to manifolds.
Recall that a differential \(k\)-form
on an open set \(U \subseteq \mathbb{R}^n\) can be written in coordinates as a primed sum over increasing multi-indices,
\[
\omega = {\sum_J}'\, \omega_J\, dx^J,
\]
with smooth coefficient functions \(\omega_J\). The differential \(d\omega_J\) of each coefficient is a \(1\)-form;
wedging it onto \(dx^J\) produces a \((k+1)\)-form, and summing gives the exterior derivative.
Definition: Exterior Derivative on Euclidean Space
Let \(U \subseteq \mathbb{R}^n\) be open and let \(\omega = {\sum_J}'\, \omega_J\, dx^J\) be a smooth \(k\)-form on \(U\).
Its exterior derivative is the \((k+1)\)-form
\[
d\omega = {\sum_J}'\, d\omega_J \wedge dx^J,
\]
where \(d\omega_J = \sum_i (\partial \omega_J / \partial x^i)\, dx^i\) is the ordinary differential of the coefficient function.
Spelled out over the coordinate basis,
\[
d\Bigl({\sum_J}'\, \omega_J\, dx^{j_1} \wedge \cdots \wedge dx^{j_k}\Bigr) = {\sum_J}'\, \sum_i \frac{\partial \omega_J}{\partial x^i}\, dx^i \wedge dx^{j_1} \wedge \cdots \wedge dx^{j_k}.
\]
Two extreme cases anchor the definition. For a \(0\)-form, that is a smooth function \(f\), there are no \(dx^J\) factors
and the formula collapses to
\[
df = \sum_i \frac{\partial f}{\partial x^i}\, dx^i,
\]
the ordinary differential. The exterior derivative therefore extends the differential of functions, not merely imitates it.
For a \(1\)-form \(\omega = \sum_j \omega_j\, dx^j\), the definition gives
\[
d\omega = \sum_{i,j} (\partial \omega_j / \partial x^i)\, dx^i \wedge dx^j.
\]
Splitting the double sum into \(i \lt j\) and \(i \gt j\), interchanging the names \(i\) and \(j\) in the second part, and
using \(dx^j \wedge dx^i = -\,dx^i \wedge dx^j\) collects the two halves onto each basis \(2\)-form:
\[
d\omega = \sum_{i \lt j} \left( \frac{\partial \omega_j}{\partial x^i} - \frac{\partial \omega_i}{\partial x^j} \right) dx^i \wedge dx^j.
\]
The coefficient is antisymmetric in \(i\) and \(j\), exactly as a \(2\)-form requires. This formula already carries a familiar shadow:
its vanishing is the symmetry-of-mixed-partials condition that a \(1\)-form must satisfy to be the differential of a function.
The Integrability Condition
A \(1\)-form is called closed when its exterior derivative is zero. By the formula above, \(\omega = \sum_j \omega_j\, dx^j\) is
closed precisely when \(\partial \omega_j / \partial x^i = \partial \omega_i / \partial x^j\) for all \(i, j\).
When \(\omega = df\), these equations
hold automatically because mixed second partials of a smooth function commute. The exterior derivative thus packages the classical
integrability condition — the test for whether a field of "infinitesimal increments" can be assembled into a single potential function — into
one coordinate-free equation, \(d\omega = 0\). Whether a closed form is actually the differential of a function turns out to depend
on the global shape of the domain, a theme we return to once the operator is in place.
Properties
The exterior derivative on Euclidean space obeys four properties that together make it a genuinely algebraic operation rather than
a mere bookkeeping device. It is linear; it satisfies a graded product rule across the wedge product; it squares to zero; and
it is compatible with pullback. These are the properties we will eventually elevate into a coordinate-free characterization on manifolds,
so we establish them carefully here, where everything is explicit.
Theorem: Properties of the Exterior Derivative on Euclidean Space
Let \(U \subseteq \mathbb{R}^n\) be open. The exterior derivative satisfies:
- Linearity. \(d\) is linear over \(\mathbb{R}\).
- Graded product rule (antiderivation). If \(\omega\) is a \(k\)-form and \(\eta\) is any form, then
\[
d(\omega \wedge \eta) = d\omega \wedge \eta + (-1)^k\, \omega \wedge d\eta.
\]
- Nilpotence. \(d \circ d = 0\); that is, \(d(d\omega) = 0\) for every form \(\omega\).
- Naturality. If \(F : V \to U\) is a smooth map from an open subset \(V\) of \(\mathbb{R}^m\),
then \(F^*(d\omega) = d(F^*\omega)\) for every form \(\omega\) on \(U\).
Linearity is immediate from the definition, since both the differential of coefficients and the wedge are linear.
The remaining three properties require argument. A preliminary observation makes the computations clean.
Lemma: Differential of a Coefficient Times an Arbitrary Wedge
For any multi-index \(I\), increasing or not, and any smooth function \(u\),
\[
d(u\, dx^I) = du \wedge dx^I.
\]
Proof:
The definition uses increasing multi-indices, so we must check the general case. If \(I\) has a repeated index,
then \(dx^I = 0\), and both sides vanish. If the entries of \(I\) are distinct, let \(\sigma\) be the permutation rearranging \(I\)
into an increasing multi-index \(J\), so that \(dx^I = (\operatorname{sgn}\sigma)\, dx^J\). Then
\[
\begin{align*}
d(u\, dx^I)
&= (\operatorname{sgn}\sigma)\, d(u\, dx^J) \\\\
&= (\operatorname{sgn}\sigma)\, du \wedge dx^J = du \wedge dx^I,
\end{align*}
\]
using the definition on the increasing index \(J\). We use this freely below.
Proof:
Part (2): the product rule.
By linearity it suffices to treat decomposable terms \(\omega = u\, dx^I\) (a \(k\)-form) and \(\eta = v\, dx^J\).
Their wedge is \(\omega \wedge \eta = uv\, dx^I \wedge dx^J\), and the ordinary product rule for the differential
of the function \(uv\) gives \(d(uv) = v\, du + u\, dv\). Hence
\[
\begin{align*}
d(\omega \wedge \eta)
&= d(uv) \wedge dx^I \wedge dx^J \\\\
&= (v\, du + u\, dv) \wedge dx^I \wedge dx^J.
\end{align*}
\]
The first term is \((du \wedge dx^I) \wedge (v\, dx^J) = d\omega \wedge \eta\). In the second term, the \(1\)-form \(dv\) must
move past the \(k\)-form \(dx^I\) to reach \(dx^J\); each of the \(k\) transpositions contributes a sign, so
\(dv \wedge dx^I = (-1)^k\, dx^I \wedge dv\), giving
\[
\begin{align*}
u\, dv \wedge dx^I \wedge dx^J
= (-1)^k\, (u\, dx^I) \wedge (dv \wedge dx^J) \\\\
= (-1)^k\, \omega \wedge d\eta.
\end{align*}
\]
Adding the two terms yields the stated formula.
Part (3): nilpotence.
We first verify \(d \circ d = 0\) on a \(0\)-form, that is a smooth function \(f\).
Here \(df = \sum_j (\partial f / \partial x^j)\, dx^j\), and applying \(d\) again,
\[
\begin{align*}
d(df) &= \sum_{i,j} \frac{\partial^2 f}{\partial x^i \partial x^j}\, dx^i \wedge dx^j \\\\
&= \sum_{i \lt j} \left( \frac{\partial^2 f}{\partial x^i \partial x^j} - \frac{\partial^2 f}{\partial x^j \partial x^i} \right) dx^i \wedge dx^j = 0,
\end{align*}
\]
where the antisymmetrization comes from collecting \(i \lt j\) and \(i \gt j\) using \(dx^j \wedge dx^i = -\,dx^i \wedge dx^j\), and the bracket vanishes because mixed second partials of a smooth function are equal.
For a general \(k\)-form, linearity reduces the claim to a decomposable term \(\omega = u\, dx^I\) with \(I = (i_1, \dots, i_k)\).
Then \(d\omega = du \wedge dx^I\), and since each \(dx^{i_r}\) is itself the differential of a coordinate function, the lemma
and the product rule give
\[
d(d\omega) = d(du \wedge dx^I) = d(du) \wedge dx^I - du \wedge d(dx^I) = 0,
\]
because \(d(du) = 0\) by the \(0\)-form case applied to each component of \(du\), and \(d(dx^I) = 0\)
because \(dx^I\) is a wedge of differentials \(dx^{i_r} = d(x^{i_r})\), each of which is closed by the same \(0\)-form case.
Thus \(d \circ d = 0\) in every degree.
Part (4): naturality.
By linearity it again suffices to consider \(\omega = u\, dx^{i_1} \wedge \cdots \wedge dx^{i_k}\).
The pullback of a coordinate differential is the differential of the composite, \(F^*(dx^i) = d(x^i \circ F)\), the coordinate
identity established for pullbacks of forms. Applying \(d\) to \(F^*\omega\) and using that pullback respects wedge products,
\[
\begin{align*}
d(F^*\omega)
&= d\bigl((u \circ F)\, d(x^{i_1} \circ F) \wedge \cdots \wedge d(x^{i_k} \circ F)\bigr) \\\\
&= d(u \circ F) \wedge d(x^{i_1} \circ F) \wedge \cdots \wedge d(x^{i_k} \circ F),
\end{align*}
\]
where every later factor \(d(x^{i_r} \circ F)\) is closed, so the product rule leaves only
the term differentiating the leading coefficient. On the other hand,
\[
\begin{align*}
F^*(d\omega)
= F^*\bigl(du \wedge dx^{i_1} \wedge \cdots \wedge dx^{i_k}\bigr) \\\\
= d(u \circ F) \wedge d(x^{i_1} \circ F) \wedge \cdots \wedge d(x^{i_k} \circ F),
\end{align*}
\]
using \(F^*(du) = d(u \circ F)\) and \(F^*(dx^{i_r}) = d(x^{i_r} \circ F)\). The two expressions coincide.
The third property is the one with the longest reach. It says that the image of \(d\) is annihilated by \(d\):
every form of the type \(d\eta\) automatically satisfies \(d(d\eta) = 0\). This single equation is the algebraic source of
the identities relating gradient, curl, and divergence, and the reason certain vector fields cannot be gradients or curls
— a point we make precise once the operator is available on manifolds and expressed in the language of vector calculus.
Existence and Uniqueness on Manifolds
The Euclidean definition depends on a choice of coordinates. To make the exterior derivative an operator on a manifold,
we must check that the coordinate formula gives the same answer in overlapping charts. It does
— and in fact the four properties of the previous section single the operator out completely: there is exactly one operator
on forms satisfying them. This is the cleanest possible state of affairs, since it means the exterior derivative can be
characterized without reference to coordinates at all, by its algebraic behavior alone.
Theorem: Existence and Uniqueness of Exterior Differentiation
Let \(M\) be a smooth manifold with or without boundary.
There is a unique collection of operators \(d : \Omega^k(M) \to \Omega^{k+1}(M)\), one for each \(k\), called
exterior differentiation, satisfying:
- \(d\) is linear over \(\mathbb{R}\);
- \(d(\omega \wedge \eta) = d\omega \wedge \eta + (-1)^k\, \omega \wedge d\eta\) for \(\omega \in \Omega^k(M)\) and \(\eta\) of any degree;
- \(d \circ d = 0\);
- for a \(0\)-form \(f\), \(df\) is the ordinary differential, \(df(X) = Xf\).
In every smooth chart, this operator is given by the coordinate formula of the Euclidean definition.
Proof:
Existence.
Define \(d\) in each smooth chart by the Euclidean coordinate formula. To know this is meaningful on \(M\),
we must check that two overlapping charts produce the same \((k+1)\)-form on the overlap.
The coordinate formula is built entirely from the differential of functions and the wedge product, both of which
the chart-to-chart transition respects; concretely, applying the naturality property of the Euclidean operator to
the transition diffeomorphism shows that the form defined in one chart pulls back to the form defined in the other.
Since the two definitions agree wherever the charts overlap, they patch together into a single globally defined \((k+1)\)-form \(d\omega\).
The operator so defined satisfies (1)-(4), because each property was verified chart-by-chart and each is preserved under the patching.
Uniqueness.
Suppose \(d\) is any operator satisfying (1)-(4). We first show that \(d\) is local: if two forms
\(\omega_1\) and \(\omega_2\) agree on an open set \(U\), then \(d\omega_1\) and \(d\omega_2\) agree on \(U\).
By linearity it is enough to show that if a form \(\eta\) vanishes on \(U\), then \(d\eta\) vanishes on \(U\).
Fix a point \(p \in U\) and choose a
smooth bump function
\(\psi\) that equals \(1\) near \(p\) and is supported inside \(U\).
Then \(\psi\eta \equiv 0\) on all of \(M\), because \(\eta\) is already zero on \(U\) and \(\psi\) is zero outside \(U\).
Applying \(d\) to the identically zero form \(\psi\eta\) and using the product rule (2) with the \(0\)-form \(\psi\),
\[
0 = d(\psi\eta) = d\psi \wedge \eta + \psi\, d\eta.
\]
Evaluating at \(p\): there \(\psi = 1\), so \(d\psi = 0\) at \(p\), and \(\eta = 0\) at \(p\), leaving \(d\eta|_p = 0\).
Since \(p \in U\) was arbitrary, \(d\eta\) vanishes on \(U\), proving locality.
Now let \(\omega\) be any form and \((U, (x^i))\) a smooth chart. On \(U\) we may write \(\omega = {\sum_J}'\, \omega_J\, dx^J\).
Locality lets us compute \(d\omega\) at any point \(p \in U\) using forms that agree with the \(\omega_J\) and \(x^i\) only near \(p\);
extending these by a bump function to globally defined smooth objects and applying properties (1)-(4) forces
\[
d\omega = {\sum_J}'\, d\omega_J \wedge dx^J
\]
at \(p\), since (4) fixes \(d\omega_J\) and \(d x^i\) as ordinary differentials, (3) gives \(d(dx^i) = 0\),
and (2) distributes \(d\) across the wedge. This is exactly the coordinate formula. As \(p\) and the chart were arbitrary,
any operator satisfying (1)-(4) must equal the one constructed in the existence part.
The properties characterizing \(d\) have a clean algebraic name. If \(A = \bigoplus_k A^k\) is a graded algebra, a linear map raising degree
by \(1\) and satisfying the graded product rule \(T(\alpha\beta) = (T\alpha)\beta + (-1)^k \alpha\,(T\beta)\) for \(\alpha \in A^k\) is called
an antiderivation of degree \(+1\). The theorem says that the ordinary differential on functions extends uniquely
to an antiderivation of degree \(+1\) on the algebra of differential forms whose square is zero. The
interior multiplication operator
is, by contrast, an antiderivation of degree \(-1\) whose square is also zero; the two will meet in a single identity when the Lie derivative
of forms is taken up.
One consequence is worth isolating, because it travels to every later use of forms: the exterior derivative commutes with pullback on manifolds,
not merely on Euclidean space.
Theorem: Naturality of the Exterior Derivative
Let \(F : M \to N\) be a smooth map. For every \(\omega \in \Omega^k(N)\),
\[
F^*(d\omega) = d(F^*\omega).
\]
That is, the
pullback map
intertwines the exterior derivatives on \(N\) and \(M\).
Proof:
The statement is local, so we may work in smooth charts \((U, \varphi)\) for \(M\) and \((V, \psi)\) for \(N\) with \(F(U) \subseteq V\).
In these charts the exterior derivative is given by the Euclidean coordinate formula, and \(F\) is represented
by its coordinate expression between open subsets of Euclidean (or half-) space. The naturality already established
for the Euclidean operator applies verbatim to this coordinate representation, giving \(F^*(d\omega) = d(F^*\omega)\) on \(U\).
Since the charts cover \(M\), the identity holds throughout.
Naturality is what allows differential forms and their derivatives to be transported across maps without distortion: restricting
to a submanifold, changing coordinates, or comparing forms on different spaces all commute with \(d\). It is the property that makes
the exterior derivative the right derivative for objects meant to be integrated, where compatibility with change of variables
is everything.
Closed and Exact Forms
The nilpotence \(d \circ d = 0\) splits the differential forms on a manifold into two nested families.
A form is closed if its exterior derivative vanishes, and exact if it is itself the exterior
derivative of another form. Nilpotence says immediately that every exact form is closed; the converse
— whether every closed form is exact — is a question that the algebra alone cannot answer, and
its failure is precisely what records the global shape of the manifold.
The two notions are linked in one direction by the structure of \(d\).
Every exact form is closed
If \(\omega = d\eta\) is exact, then \(d\omega = d(d\eta) = 0\) by nilpotence.
Hence \(\omega\) is closed. Equivalently, the image of \(d\) is contained in its kernel: exactness implies closedness,
for every degree, on every manifold.
For a \(1\)-form this recovers a classical statement. A \(1\)-form is exact when it is the differential of a function,
\(\omega = df\); the function \(f\) is a potential in the sense of physics, and \(\omega\) records its infinitesimal increments.
The closedness condition \(d\omega = 0\) is then the integrability condition met earlier
— the equality of the mixed partials of the would-be potential. So a necessary condition for a \(1\)-form to admit a potential
is that it be closed.
Whether this necessary condition is also sufficient is a genuinely different question, and the answer depends on where the form lives.
On a star-shaped or otherwise topologically simple region, every closed form is exact: a potential can always be built. But on a domain
with a hole, closed forms can fail to be exact. The standard witness is the angular \(1\)-form on the punctured plane,
\[
\omega = \frac{-y\, dx + x\, dy}{x^2 + y^2},
\]
which is closed everywhere on \(\mathbb{R}^2 \setminus \{0\}\) yet is not the differential of any globally
defined smooth function there: integrating it once around the origin returns \(2\pi\) rather than \(0\),
so no single-valued potential can exist. Locally, near any point away from the origin, \(\omega\) is the differential
of an angle function; the obstruction is entirely global, created by the missing point.
Closedness Is Local, Exactness Is Global
The gap between closed and exact is not a defect of the forms but a measurement of the space.
Closedness is a local, differential condition: it can be checked in any chart, one point at a time.
Exactness is a global, integral condition: it asks whether local potentials can be reconciled
into a single one across the whole manifold. The size of the discrepancy
— how many closed forms fail to be exact in each degree — is an invariant of the manifold, computed
by assembling these forms into a quotient. Built from smooth forms, this quotient is at first an invariant
of the manifold's smooth structure; that it depends only on the underlying topology is a deeper theorem we do not prove here.
Either way, it is the entry point to a cohomology theory built entirely from the exterior derivative, where
the algebra of \(d\) becomes a tool for detecting holes. What matters now is the recognition that \(d \circ d = 0\)
is exactly the relation that makes such a quotient possible, since it guarantees the exact forms sit inside the closed ones.
Exterior Derivatives and Vector Calculus
The classical differential operators of vector calculus on \(\mathbb{R}^3\) — gradient, curl, and divergence —
are not three unrelated constructions but three appearances of the single exterior derivative, read off in degrees zero, one, and two.
Making this precise both unifies the operators and explains, in one stroke, the two identities every student of vector calculus memorizes:
that the curl of a gradient vanishes and the divergence of a curl vanishes. Both are instances of \(d \circ d = 0\).
Work on \(\mathbb{R}^3\) with coordinates \((x, y, z)\). Recall the classical operators: the gradient of a function \(f\)
is the vector field \(\operatorname{grad} f = (\partial f/\partial x,\, \partial f/\partial y,\, \partial f/\partial z)\);
the curl of a vector field \(X = (P, Q, R)\) is
\[
\operatorname{curl} X = \left( \frac{\partial R}{\partial y} - \frac{\partial Q}{\partial z},\ \frac{\partial P}{\partial z} - \frac{\partial R}{\partial x},\ \frac{\partial Q}{\partial x} - \frac{\partial P}{\partial y} \right);
\]
and the divergence of \(X\) is the function \(\operatorname{div} X = \partial P/\partial x + \partial Q/\partial y + \partial R/\partial z\).
Now compute the exterior derivative in each degree. For a \(0\)-form \(f\),
\[
df = \frac{\partial f}{\partial x}\, dx + \frac{\partial f}{\partial y}\, dy + \frac{\partial f}{\partial z}\, dz,
\]
whose coefficients are exactly the components of \(\operatorname{grad} f\). For a \(1\)-form \(\omega = P\, dx + Q\, dy + R\, dz\),
the computation of \(d\omega\) collects the mixed terms as
\[
\begin{align*}
d\omega = {}&\left( \frac{\partial Q}{\partial x} - \frac{\partial P}{\partial y} \right) dx \wedge dy + \left( \frac{\partial R}{\partial x} - \frac{\partial P}{\partial z} \right) dx \wedge dz\\\\
&+ \left( \frac{\partial R}{\partial y} - \frac{\partial Q}{\partial z} \right) dy \wedge dz,
\end{align*}
\]
whose three coefficients are exactly the components of \(\operatorname{curl} X\), up to
the ordering and signs of the basis \(2\)-forms.
Finally, for a \(2\)-form \(\eta = u\, dy \wedge dz + v\, dz \wedge dx + w\, dx \wedge dy\),
\[
d\eta = \left( \frac{\partial u}{\partial x} + \frac{\partial v}{\partial y} + \frac{\partial w}{\partial z} \right) dx \wedge dy \wedge dz,
\]
whose single coefficient is the divergence of the field \((u, v, w)\).
So the same operator \(d\), applied to forms of degree \(0\), \(1\), and \(2\), reproduces gradient, curl,
and divergence in turn. The translation between vector fields or functions and forms of the appropriate degree is supplied
by the Euclidean metric, which identifies a vector field with a \(1\)-form by
lowering an index,
and identifies functions and vector fields with top- and middle-degree forms through the volume form. Under those identifications,
the three classical operators line up along a single sequence in which each step is an exterior derivative:
The grad-curl-div sequence
Reading left to right in increasing degree, the operators form the chain
\[
C^\infty \xrightarrow{\ \operatorname{grad}\ } \mathfrak{X} \xrightarrow{\ \operatorname{curl}\ } \mathfrak{X} \xrightarrow{\ \operatorname{div}\ } C^\infty,
\]
which corresponds, degree by degree under the metric identifications, to
\[
\Omega^0 \xrightarrow{\ d\ } \Omega^1 \xrightarrow{\ d\ } \Omega^2 \xrightarrow{\ d\ } \Omega^3.
\]
Each vertical correspondence commutes with the horizontal operators, so a statement about the bottom row is automatically
a statement about the top row.
The payoff is immediate. Since \(d \circ d = 0\), composing two consecutive arrows in the bottom row gives zero, and
the same must hold for the top row. Composing gradient then curl corresponds to applying \(d\) twice starting from a \(0\)-form:
\[
\operatorname{curl}(\operatorname{grad} f) \ \longleftrightarrow\ d(df) = 0.
\]
Composing curl then divergence corresponds to applying \(d\) twice starting from a \(1\)-form:
\[
\operatorname{div}(\operatorname{curl} X) \ \longleftrightarrow\ d(d\omega) = 0.
\]
The two identities of vector calculus — that gradients are curl-free and curls are divergence-free
— are thus not separate facts to be verified by grinding through partial derivatives, but
two readings of the single relation \(d \circ d = 0\).
Why Forms, and Why All Dimensions
The classical operators are tied to three dimensions: the curl, in particular, exists as an operation on vector fields only
on \(\mathbb{R}^3\), where a \(2\)-form happens to have the same number of components as a vector field.
The exterior derivative carries no such restriction. It is defined on forms of every degree on a manifold of any dimension,
and it expresses the same information — how a field of infinitesimal data changes — without depending on the coincidences
that make the vector-field formulation possible in dimension three. The desire to extend gradient, curl, and divergence
beyond \(\mathbb{R}^3\) was one of the motivations for the theory of differential forms; the exterior derivative is the operator
that achieves it, with \(d \circ d = 0\) as the structural identity that survives into every dimension.
An Invariant Formula
The coordinate formula for \(d\) is the one to compute with, but it hides a coordinate-free description that is often more illuminating.
The exterior derivative can be expressed entirely in terms of how forms act on vector fields and how those vector fields fail to commute
— that is, in terms of the Lie bracket. For a \(1\)-form the formula is clean and worth committing to memory; it shows that \(d\) is,
in a precise sense, dual to the bracket.
Proof:
Both sides are linear in \(\omega\), and every smooth \(1\)-form is locally a sum of terms \(u\, dv\)
for smooth functions \(u, v\), so it suffices to verify the identity for \(\omega = u\, dv\). On the left,
\(d\omega = du \wedge dv\), and evaluating a wedge of two \(1\)-forms on a pair of vectors gives the determinant
\[
d\omega(X, Y) = (du \wedge dv)(X, Y) = du(X)\, dv(Y) - du(Y)\, dv(X) = (Xu)(Yv) - (Yu)(Xv),
\]
using \(du(X) = Xu\) for the action of a differential on a vector field.
On the right, \(\omega(Y) = u\, dv(Y) = u\,(Yv)\) and \(\omega(X) = u\,(Xv)\).
Differentiating with the product rule and using that \([X, Y]\) acts as \(X \circ Y - Y \circ X\) on functions,
\[
\begin{align*}
X\bigl(\omega(Y)\bigr) - Y\bigl(\omega(X)\bigr) - \omega\bigl([X, Y]\bigr) &= X\bigl(u\,(Yv)\bigr) - Y\bigl(u\,(Xv)\bigr) - u\,\bigl([X,Y]v\bigr)\\\\
&= (Xu)(Yv) + u\,(XYv) - (Yu)(Xv) - u\,(YXv) - u\,(XYv - YXv)\\\\
&= (Xu)(Yv) - (Yu)(Xv).
\end{align*}
\]
The second-order terms \(u\,(XYv)\) and \(u\,(YXv)\) produced by the product rule are exactly cancelled by the expansion \(\omega([X,Y]v) = u\,(XYv - YXv)\), leaving exactly the left-hand side.
The formula deserves emphasis because it inverts a familiar dependency. The Lie bracket measures the failure of two flows to commute;
the formula above shows that knowing all the brackets of a frame of vector fields determines the exterior derivatives of the dual coframe,
and conversely. Exterior differentiation and the Lie bracket carry the same information, packaged dually.
The same identity generalizes to forms of arbitrary degree, at the cost of a longer expression.
Proof sketch
Denote the right-hand side by \(D\omega(X_1, \dots, X_{k+1})\). One first checks that \(D\omega\) is multilinear
over smooth functions in each slot: replacing some \(X_p\) by \(fX_p\) produces, from the derivative terms,
extra factors \((X_i f)\) that are exactly matched and cancelled by the corresponding terms generated
in the bracket sum through the identity \([fX_p, X_j] = f[X_p, X_j] - (X_j f)X_p\). Function-multilinearity means \(D\omega\)
is a tensor, so it equals \(d\omega\) as soon as the two agree on a single convenient frame at each point.
Choosing a coordinate frame, where all the Lie brackets vanish, kills the second sum, and a direct computation
with \(\omega = u\, dx^I\) shows that the remaining first sum reproduces the coordinate formula for \(d\omega\).
Hence \(D\omega = d\omega\) everywhere.
A Coordinate-Free Definition
The arbitrary-degree formula could serve as the definition of \(d\), bypassing coordinates entirely: its right-hand side
is manifestly independent of any chart, being built only from the action of vector fields, the form, and the bracket.
From it the existence, uniqueness, and properties of \(d\) can all be recovered.
In practice the coordinate formula remains the computational tool, and the invariant formula past degree one is unwieldy,
with one serious inconvenience — to evaluate \(d\omega\) on tangent vectors at a point, one must first extend them
to vector fields nearby. Its value is theoretical: it makes the relationship between \(d\) and the bracket transparent,
and that relationship is what surfaces whenever the geometry of a frame is encoded in the derivatives of its coframe.
Structure Constants
The duality between the exterior derivative and the Lie bracket becomes a concrete numerical relation when both are read
against a local frame. A frame of vector fields fails to commute in a way recorded by a collection of functions, the structure
constants of the frame; the dual coframe has exterior derivatives recorded by another such collection. The invariant formula
for a \(1\)-form turns one collection into the other, up to a sign.
Let \((E_i)\) be a smooth local frame for a manifold and let \((\varepsilon^i)\) be its dual
coframe,
so that \(\varepsilon^i(E_j) = \delta^i_j\). The brackets of the frame fields are themselves combinations of the frame,
and their coefficients are the structure constants \(c^i_{jk}\):
\[
[E_j, E_k] = \sum_i c^i_{jk}\, E_i.
\]
Likewise, each \(d\varepsilon^i\) is a \(2\)-form, expressible over the coframe with coefficient functions \(b^i_{jk}\):
\[
d\varepsilon^i = \sum_{j \lt k} b^i_{jk}\, \varepsilon^j \wedge \varepsilon^k.
\]
These two families of functions are not independent.
Theorem: Structure Constants and the Coframe
With the notation above, the coframe coefficients and the frame structure constants are related by
\[
b^i_{jk} = -\,c^i_{jk}.
\]
Proof:
Apply the invariant formula for a \(1\)-form to \(\omega = \varepsilon^i\) and the frame fields \(X = E_j\), \(Y = E_k\):
\[
d\varepsilon^i(E_j, E_k) = E_j\bigl(\varepsilon^i(E_k)\bigr) - E_k\bigl(\varepsilon^i(E_j)\bigr) - \varepsilon^i\bigl([E_j, E_k]\bigr).
\]
The first two terms vanish: \(\varepsilon^i(E_k) = \delta^i_k\) and \(\varepsilon^i(E_j) = \delta^i_j\) are constant functions,
so any vector field annihilates them. The third term evaluates the coframe on the bracket:
\[
\begin{align*}
\varepsilon^i\bigl([E_j, E_k]\bigr)
&= \varepsilon^i\Bigl(\sum_m c^m_{jk}\, E_m\Bigr) \\\\
&= \sum_m c^m_{jk}\, \varepsilon^i(E_m) \\\\
&= c^i_{jk}.
\end{align*}
\]
Hence \(d\varepsilon^i(E_j, E_k) = -\,c^i_{jk}\). On the other hand, the expansion
\(d\varepsilon^i = \sum_{j \lt k} b^i_{jk}\,\varepsilon^j \wedge \varepsilon^k\), evaluated
on the same pair with \(j \lt k\), returns \(b^i_{jk}\), because \((\varepsilon^p \wedge \varepsilon^q)(E_j, E_k)\) equals
\(1\) when \((p, q) = (j, k)\) and \(0\) for any other increasing pair. Comparing the two evaluations gives \(b^i_{jk} = -\,c^i_{jk}\).
The relation makes the equivalence between bracket data and exterior-derivative data exact and computable:
a single sign converts the structure constants of a frame into the coefficients describing how its coframe twists under \(d\),
and back. The numbers \(c^i_{jk}\) themselves are antisymmetric in \(j\) and \(k\) and constrained by the Jacobi identity
of the bracket; the equation above transports those constraints onto the coframe side without further work.
The Coframe Equation as a Structural Fingerprint
On a Lie group equipped with a frame of left-invariant vector fields, the structure constants are genuinely constant
— the same at every point — and they encode the entire infinitesimal structure of the group.
The relation \(b^i_{jk} = -c^i_{jk}\) then says that the exterior derivatives of the left-invariant coframe carry
precisely the same information, written as a system of equations among \(2\)-forms. That system, expressing \(d\varepsilon^i\)
through wedge products of the \(\varepsilon^j\), is the equation governing how a group's invariant forms differentiate,
and it reappears as a central object whenever the geometry of a group is studied through its invariant forms.
The exterior derivative, the Lie bracket, and the algebraic skeleton of a group are here three views of one structure.
Lie Derivatives of Differential Forms
The Lie derivative measures how a tensor field changes as it is dragged along the flow of a vector field.
For differential forms there is a far more efficient way to compute it than the general tensor formula:
the exterior derivative and interior multiplication combine into a single identity that expresses the Lie derivative
without any reference to a flow at all. This identity is one
of the most useful in differential geometry. We first record a product rule, then state and prove the formula.
Since differential forms are
covariant tensor fields,
the
product rule for Lie derivatives of tensors
applies. Specialized to the wedge product, it takes the following form.
The sign is the unmixed one of an ordinary product rule, with no \((-1)^k\) factor, because
the Lie derivative preserves the degree of a form rather than raising it. This is the structural difference
between \(\mathcal{L}_V\) and \(d\): the exterior derivative is an antiderivation of degree \(+1\), so
it carries a sign when commuted past a form; the Lie derivative is a derivation of degree \(0\), so it does not.
We now reach the identity promised when the exterior derivative was first characterized, where \(d\) appeared
as an antiderivation of degree \(+1\) and interior multiplication \(\iota_V\) as an antiderivation of degree \(-1\).
Their anticommutator is the Lie derivative.
Proof:
We argue by induction on the degree \(k\) of \(\omega\). For a \(0\)-form, that is a function \(f\),
interior multiplication \(\iota_V f = 0\) on functions, so the right-hand side is \(\iota_V(df) + 0 = df(V) = Vf\),
which is exactly \(\mathcal{L}_V f\). This establishes the base case.
Assume the formula holds for forms of degree less than \(k\), and let \(\omega\) be a \(k\)-form.
Working locally, \(\omega\) is a sum of terms \(u\, dx^{i_1} \wedge \cdots \wedge dx^{i_k}\), each
of which can be written \(du \wedge \beta\) with \(u = x^{i_1}\) a coordinate function
and \(\beta\) a \((k-1)\)-form, or handled by linearity; it therefore suffices
to treat a product \(\mathcal{L}_V(du \wedge \beta)\) for a function \(u\) and a \((k-1)\)-form \(\beta\).
The wedge derivation rule above and the
Lie derivative of a differential,
\(\mathcal{L}_V(du) = d(\mathcal{L}_V u) = d(Vu)\), give
\[
\mathcal{L}_V(du \wedge \beta) = d(Vu) \wedge \beta + du \wedge (\mathcal{L}_V \beta).
\]
Now compute the right-hand side of Cartan's formula on the same term. Both \(d\) and \(\iota_V\)
are antiderivations, \(d\) of degree \(+1\) and \(\iota_V\) of degree \(-1\), and \(\iota_V(du) = du(V) = Vu\).
Expanding \(\iota_V\, d(du \wedge \beta) + d\,\iota_V(du \wedge \beta)\) using these antiderivation rules,
together with \(d(du) = 0\), and applying the induction hypothesis to \(\beta\) in the form
\(\iota_V\, d\beta + d\,\iota_V\beta = \mathcal{L}_V\beta\), the terms involving \((Vu)\,d\beta\) cancel in pairs,
and what remains is \(d(Vu) \wedge \beta + du \wedge (\mathcal{L}_V\beta)\). This matches the previous display,
completing the induction.
The formula's power is that it computes a flow-defined quantity, the Lie derivative, purely algebraically,
from the exterior derivative and the contraction with \(V\) — neither of which requires solving for the flow.
It also makes the two antiderivations of the form algebra, \(d\) and \(\iota_V\), partners in a single relation:
their anticommutator on forms is the degree-preserving operator \(\mathcal{L}_V\). One consequence is recorded at once.
Proof:
Apply Cartan's formula to \(d\omega\) and use nilpotence \(d \circ d = 0\):
\[
\begin{align*}
\mathcal{L}_V(d\omega)
&= \iota_V\bigl(d(d\omega)\bigr) + d\bigl(\iota_V(d\omega)\bigr) \\\\
&= d\bigl(\iota_V(d\omega)\bigr).
\end{align*}
\]
On the other hand, applying \(d\) to Cartan's formula for \(\omega\) and again using \(d \circ d = 0\),
\[
\begin{align*}
d(\mathcal{L}_V \omega)
&= d\bigl(\iota_V(d\omega)\bigr) + d\bigl(d(\iota_V \omega)\bigr) \\\\
&= d\bigl(\iota_V(d\omega)\bigr).
\end{align*}
\]
The two right-hand sides are identical, so \(\mathcal{L}_V(d\omega) = d(\mathcal{L}_V \omega)\).
One Identity, Three Operators
Cartan's formula ties together the three derivatives that act on differential forms — the exterior derivative \(d\),
interior multiplication \(\iota_V\), and the Lie derivative \(\mathcal{L}_V\) — into a single relation.
Two of them, \(d\) and \(\iota_V\), are antiderivations whose squares vanish; the third is their anticommutator,
and the relation \(\mathcal{L}_V(d\omega) = d(\mathcal{L}_V\omega)\)
follows formally, with no computation, from \(d^2 = 0\). This kind of bookkeeping among graded operators, where
identities among derivations are forced by their degrees and their squares, is the algebraic style
that the calculus of forms makes available throughout geometry: a flow-theoretic statement is reduced to a sign count.