Chapter 4 › 4.2 Cartan's Structural Equations › 4.2.2 Exterior derivative
Subsection

4.2.2  Exterior derivative

One way to “produce” a differential 1-form is to take the “gradient” of a scalar function \(f\). For a \(C^1\)-function \(f:\Sigma\to\mathbb{R}\) locally parametrized by \(F(u^1,u^2)\), we define \(df\) to be the following differential 1-form: \[df:=\frac{\partial f}{\partial u^1}\,du^1+\frac{\partial f}{\partial u^2}\,du^2.\] It is analogous to the gradient vector \(\nabla f\) in multivariable calculus. For example, if \(f(u^1,u^2)=u_1^2+u_2^2\), then \(df=2u_1\,du^1+2u_2\,du^2\).

Given a differential 1-form \(\alpha=\sum_i\alpha_i(u^1,u^2)\,du^i\), one can also define its exterior derivative \(d\alpha\), which is a differential 2-form: \[d\alpha:=\sum_i d\alpha_i\wedge du^i=\sum_i\sum_j\frac{\partial\alpha_i}{\partial u^j}\,du^j\wedge du^i.\] On regular surfaces (\(i\) runs from 1 to 2 and \(du^i\wedge du^i=0\)): \[d\alpha=\left(\frac{\partial\alpha_2}{\partial u^1}-\frac{\partial\alpha_1}{\partial u^2}\right)du^1\wedge du^2.\] The Green's and Stokes' Theorems can be written elegantly as: for any differential 1-form on \(\Sigma^2\), \[\int_{\partial\Sigma^2}\alpha=\int_{\Sigma^2}d\alpha.\]

Exercise 4.6

Show that for a \(C^2\) scalar function \(f(u^1,u^2)\), we have \(d^2f:=d(df)=0\).

Need help?

It is in fact possible to express \(d\alpha\), where \(\alpha\) is a differential 1-form, without using local coordinates. For any vector fields \(X,Y\), we have:

Lemma 4.9

\[(4.2)\qquad d\alpha(X,Y)=D_X(\alpha(Y))-D_Y(\alpha(X))-\alpha(\nabla_X Y-\nabla_Y X).\] Note that \(\alpha(Y)\) is a scalar function, so \(D_X(\alpha(Y))\) is simply the directional derivative of \(\alpha(Y)\) along the vector field \(X\).

Checkpoint
Lemma 4.9 gives a formula for the exterior derivative of a 1-form \(\alpha\):
\[d\alpha(X,Y)=D_X(\alpha(Y))-D_Y(\alpha(X))-\alpha(\nabla_X Y-\nabla_Y X).\]What role does the term \(\alpha(\nabla_X Y - \nabla_Y X)\) play?
WhyOn a hypersurface, \(\nabla_X Y\) is the covariant (projected) derivative, while \(D_X Y\) is the full ambient derivative. The correction term \(\alpha(\nabla_X Y-\nabla_Y X)\) accounts for how the 1-form \(\alpha\) sees the difference between applying \(\alpha\) to the covariant commutator versus the Lie bracket — it is needed to express \(d\alpha\) purely in terms of intrinsic data.
Proof — revealed step by step

Denote the RHS by \(T(X,Y):=D_X(\alpha(Y))-D_Y(\alpha(X))-\alpha(\nabla_X Y-\nabla_Y X)\). We first check that \(T\) is a 2-tensor by verifying \(T(\varphi X,\psi Y)=\varphi\psi T(X,Y)\) for scalar functions \(\varphi,\psi\). We compute \(D_{\varphi X}(\alpha(\psi Y))=\varphi\alpha(Y)D_X\psi+\varphi\psi D_X(\alpha(Y))\) and \(D_{\psi Y}(\alpha(\varphi X))=\psi\alpha(X)D_Y\varphi+\psi\varphi D_Y(\alpha(X))\).

Checking the connection term: \(\alpha(\nabla_{\varphi X}(\psi Y)-\nabla_{\psi Y}(\varphi X))=\varphi D_X\psi\cdot\alpha(Y)-\psi D_Y\varphi\cdot\alpha(X)+\varphi\psi\alpha(\nabla_X Y-\nabla_Y X)\). Subtracting, the extra terms cancel and we get \(T(\varphi X,\psi Y)=\varphi\psi T(X,Y)\). Thus \(T\) is a 2-tensor (alternating, since \(T(X,Y)=-T(Y,X)\)), hence a differential 2-form.

As both \(d\alpha\) and \(T\) are 2-forms on surfaces, it suffices to verify they agree on \(\left(\frac{\partial F}{\partial u^1},\frac{\partial F}{\partial u^2}\right)\). Using \(\alpha\!\left(\frac{\partial F}{\partial u^j}\right)=\alpha_j\) and \(\Gamma^j_{12}=\Gamma^j_{21}\), we get \(T\!\left(\frac{\partial F}{\partial u^1},\frac{\partial F}{\partial u^2}\right)=\frac{\partial\alpha_2}{\partial u^1}-\frac{\partial\alpha_1}{\partial u^2}=d\alpha\!\left(\frac{\partial F}{\partial u^1},\frac{\partial F}{\partial u^2}\right)\). Hence \(T=d\alpha\). \(\square\)

Exercise 4.7

Some textbooks write (4.2) as \(d\alpha(X,Y)=D_X(\alpha(Y))-D_Y(\alpha(X))-\alpha(D_XY-D_YX)\). Does it make any difference? Why?

Need help?