I am trying to deduce the next identity (Green-Gauss theorem) $$\int_\Omega \dfrac{\partial u}{\partial x_i} dx = \int_{\partial \Omega} uv_i dS$$ from the generalized Stoke's theorem for manifolds. Here $u$ is a $C^\infty$ function on an open bounded set $\Omega$ with $C^1$ boundary and $v_i$ denotes the i-th component of the unit normal vector field.

I've been trying to define $n-1$ forms on $\Omega$ whose derivatives are $\dfrac{\partial u}{\partial x_i} dx$ to apply the theorem but I cannot prove that those forms have the same integral as $uv_i$ on $\partial \Omega$.

Any ideas?

Also, I would really appreciate references of the proofs of classical calculus integration results using the generalized Stoke's theorem.

Thanks in advance.

  • $\begingroup$ As stated, this does not make sense, since the integrand on the LHS is a $1$-form, which you can only integrate over a curve. Actually, if your coordinates are $x_i$, I do not know what $dx$ all by itself means. Did you mean $dV$ here? $\endgroup$ – Steven Gubkin Oct 30 '14 at 18:09
  • $\begingroup$ Is just a matter of notation, $dx$ means the volume form on $\mathbb{R} ^{n}$, $dx = dx_1 dx_2 ... dx_n$ if you want to see it that way. $\endgroup$ – Frank Trujillo Oct 30 '14 at 19:30
  • $\begingroup$ Ah, then I can answer your question. $\endgroup$ – Steven Gubkin Oct 30 '14 at 19:48

First of all, you need to understand what $dS$ is.

$dS$ is the form $dV$ contracted by the vector field $\nu$ on $\partial \Omega$. In other words, to find $dS(v_2,v_3,...,v_{n})$ for tangent vectors $v_2,v_3,...,v_{n}$ to $\partial \Omega$, you just evaluate $dV(\nu, v_2,v_3,...,v_{n})$. The motivation for this definition is that the area of a parallelogram $P$ is the same as the volume of the parallelepiped with base $P$ and height $1$.

So now note that,

$ \frac{\partial f}{\partial x_1} dV = d(f dx_2 \wedge dx_3 \wedge ...\wedge dx_n) $

Letting $ \widehat{dx_1} = dx_2 \wedge dx_3 \wedge ...\wedge dx_n$

Let $p \in \partial \Omega$ and $v_2,v_3,...,v_n \in T_p( \partial \Omega)$ then

$ \begin{align*} f(p) \widehat{dx_1}(v_2,v_3,...,v_n) &= f(p) dV(e_1,v_2,...,v_n)\\ &=f(p) dV(\textrm{proj}_{\nu(p)}(e_1),v_2,...,v_n)\\ &=f(p) dV(\nu_1(p)\nu(p),v_2,...,v_n)\\ &=f(p)\nu_1(p) dV(\nu(p),v_2,...,v_n)\\ &=f(p)\nu_1(p) dS(v_2,...,v_n) \end{align*} $

so as forms on $\partial S$, we have $f\widehat{dx_1} = f\nu_1dS$. The big step is the second line in the equality above, which is justified by writing $e_1$ as a linear combination of $\nu$ and $v_i$'s. All of the $v_i$ terms die because of the alternating property of forms.

Now we conclude that

$\begin{align*} \int_\Omega \frac{\partial f}{\partial x_1} dV &= \int_\Omega d(f \widehat{dx_i})\\ &=\int_{\partial \Omega} f \widehat{dx_i} \text{by Stokes' theorem}\\ &=\int_{\partial \Omega} f \nu_1 dS \end{align*} $

If any of this does not make sense, please ask for follow up in the comments. Notation in differential geometry is horrible, and everyone makes it up for themselves I think. So I understand if some part of this is hard to parse. I am also no differential geometer, but I do a lot of calculus in high dimensional spaces (I work in several complex variables). This kind of thing used to frustrate me a lot.

  • $\begingroup$ There is some care needed with signs if you look at $x_i$ instead of just $x_1$ as I did here. But I think we get a $(-1)^i$ once when rewritting the original form as the exterior differential of something else, and we get another $(-1)^i$ when we contract by the vector field $e_i$ later on, so the conclusion you seek is still true without any sign worries. Sticking to $x_1$ let me glide over this issue, which I thought would be easier as a first pass. $\endgroup$ – Steven Gubkin Oct 30 '14 at 20:22
  • $\begingroup$ Also I realized that you may not have the same perspective on forms as I do. I know in some presentations they are "just symbols of the form...", with various definitions of integration. My definition of a form on a vector space is an alternating multilinear map to the base field. A form on a manifold is then a smooth section of a form bundle. $\endgroup$ – Steven Gubkin Oct 30 '14 at 20:53
  • $\begingroup$ Don't worry, I was aware of the sign and I also understand the notation. That was exactly the form that I was trying, thank you. You were right, I was not seeing the second equality. $\endgroup$ – Frank Trujillo Oct 30 '14 at 21:21

Steven's answer is good, let me reformulate it in more formal flavor. By Stokes, it suffices to prove $$l^*(dx^2 \wedge...\wedge dx^n) = dS := l^*(i_v(dx^1\wedge...\wedge dx^n))$$ Where $l:\partial U \to U$ the inclusion map, $i_v$ is the contraction operator with respect to $v$. Let $i_{\frac{\partial}{\partial x^1}}$ be the contrition operator with respect to $\frac{\partial}{\partial x^1}$, then we have $l^*(dx^2 \wedge...\wedge dx^n) = l^*i_{\frac{\partial}{\partial x^1}}(dx^1,...,dx^n)$. Then we just need to prove $l^*i_{\frac{\partial}{\partial x^1}}(dx^1,...,dx^n) = \langle \frac{\partial}{\partial x^1}, v \rangle l^*(i_v(dx^1\wedge...\wedge dx^n))$. To see this, decompose $\frac{\partial}{\partial x^1} = \langle \frac{\partial}{\partial x^1}, v \rangle v + \alpha$, where $\alpha$ is tangent to $\partial U$.(essentially we used Lemma 16.30 in Lee's introduction to smooth manifold.), then we have $$l^*i_{\frac{\partial}{\partial x^1}}(dx^1,...,dx^n) = \langle \frac{\partial}{\partial x^1}, v \rangle l^*(i_v(dx^1\wedge...\wedge dx^n))+l^*i_\alpha (dx^1\wedge...\wedge dx^n)$$ Now it remains to prove $l^*i_\alpha (dx^1\wedge...\wedge dx^n) = 0$, which is because choosing $v_2,...,v_n$ tangent to $\partial U$, then $\alpha$ and $v_2,...,v_n$ must be linear dependent thus have zero determinant, so $l^*i_\alpha (dx^1\wedge...\wedge dx^n)(v_2,...,v_n)=(dx^1 \wedge...\wedge dx^n)(\alpha, v_2,...,v_n)=0$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.