On page 198 of Arnold's Mathematical Methods of Classical Mechanics, he asks the reader to prove Cartan's formula $$\tag{1}L_X=\mathrm{d}i_X+i_X\mathrm{d}$$ where $L_X$ is the Lie derivative wrt. $X$, $\mathrm{d}$ is the exterior derivative, and $i_X$ is the interior derivative (interior product).

I am aware of the "usual" proof, i.e. to show that the action of $L_X$ on functions and differentials agrees with that of the action of the rhs. of (1). From that one may show this equality holds on all $p$-forms.

However, Arnold offers a hint that completely changes the nature of the problem:

We denote by $H$ the "homotopy operator" associating to a $k$-chain $\gamma: \sigma\to M$ the $(k + 1)$-chain $H\gamma: (I \times \sigma) \to M$ according to the formula $(H\gamma)(t, x) = g^t\gamma(x)$ (where $I = [0, 1]$). Then $g^1\gamma - \gamma = \partial(H\gamma) + H(\partial\gamma).$

($g^t$ is the flow of $X$. I understand how to obtain the equation for the boundary of the homotopy of $\gamma$.)

Since we are given the equation for the boundary of $H\gamma$, I thought it would be a good idea to integrate the differential of a $k$-form $\omega$ over $H\gamma$ and use Stokes' theorem, followed by the equation for the boundary: $$\int_{H\gamma}\mathrm{d}\omega=\int_{\partial(H\gamma)}\omega=\int_{g^1\gamma-\gamma-H(\partial\gamma)}\omega$$ I don't see how this helps. Alternatively, on the right hand side, we can use the equation $\int_{f(D)}\omega=\int_Df^*\omega$ ($f^*$ is the pullback of $f$) to write $$\int_{H\gamma}\mathrm{d}\omega=\int_{I\times\sigma}(H\gamma)^*\mathrm{d}\omega$$ Unfortunately, I don't know how to calculate $(H\gamma)^*\mathrm{d}\omega$. How does one continue here?


First of all, I think the formula for the homotopy should be $$ g^1 \gamma -\gamma = \partial(H\gamma) - H(\partial \gamma).$$ As the left hand side is the top minus the bottom, the right hand side then should be the boundary of $I\times \sigma$ minus the cylindrical part of the boundary ($I\times \partial \gamma$).

Using that, we proceed to the proof. By fundamental theorem of calculus and $g^{s+t} = g^s g^t$, for any $k$-form $\omega$ we have

\begin{align} (g^1)^*\omega - \omega &= \int_0^1\frac{d}{dt} (g^t)^*\omega|_{t=s} \ ds \\ &=\int_0^1 (g^s)^* \frac{d}{dt} (g^t)^* \omega \bigg|_{t=0} ds \\ &= \int_0^1 (g^s)^* \mathcal L_X \omega \ ds. \end{align}

If we integrate this on $\gamma$, by Fubini's theorem, $$\int_\gamma \int_0^1 (g^s)^* \mathcal L_X \omega \ ds = (-1)^k\int_0^1 \int_\gamma H_s^* \mathcal L_X \omega \ ds,$$ where $H_s : \sigma \to M$ is given by $H_s(x) = H(s, x) = g^s \gamma(x)$. Note that $H_0 = \gamma$.

On the other hand, $$\begin{split} \int_\gamma (g^1)^* \omega - \omega &= \int_{g^1 \gamma - \gamma} \omega \\ &= \int_{\partial (H\gamma) - H(\partial \gamma)} \omega \\ &= \int_{H\gamma} d\omega - \int_{H(\partial \gamma)} \omega. \end{split}$$

Now we use that (prove later) for any $r$-form $\alpha$ and for $r-1$ chain $\gamma$ we have $$\tag{1} \int_{H\gamma}\alpha = (-1)^{r-1}\int_0^1 \int_\gamma H_s^* i_X \alpha\ ds,$$

then $$\begin{split} \int_\gamma (g^1)^* \omega - \omega &= (-1)^k\int_0^1 \int_\gamma H_s^* i_X d\omega \ ds - (-1)^{k-1} \int_0^1 \int_{\partial\gamma}H_s^* i_X \omega \ ds\\ &= (-1)^k \int_0^1 \int_\gamma H_s^* (i_X d\omega + di_X \omega) \ ds . \end{split}$$ Hence $$\int_0^1 \int_\gamma H_s^* \mathcal L_X \omega \ ds = \int_0^1 \int_\gamma H_s^* (i_X d\omega + di_X \omega) \ ds .$$ Note that if we consider the homotopy $G_t$, which is the restriction of $H\gamma$ to $[0, t]\times \sigma$, then similarly we have $$\int_0^t \int_\gamma H_s^* \mathcal L_X \omega \ ds = \int_0^t \int_\gamma H_s^* (i_X d\omega + di_X \omega) \ ds $$ for all $t >0$. Thus (using $H_0 = \gamma$) $$ \int_\gamma \mathcal L_X \omega= \int_\gamma (i_X d\omega + di_X \omega).$$ As the above holds for all $\gamma$, we have $$ \mathcal L_X \omega = i_X d\omega + di_X \omega.$$

It remains to prove $(1)$. Let $\alpha$ be an $r$ form. By definition, (call $H\gamma = F$)

$$\int_{H\gamma} \alpha = \int_{I \times \sigma} F^*\alpha$$

if we write in local coordinates $(t = x^0, x_1, \cdots, x_{r-1})$ in $I\times \sigma$ and $(y^1, \cdots, y^m)$ of $M$.

$$\begin{split} F^* \alpha (t, x)&= F^* \left(\sum_{|I| = r} \alpha_{i_1\cdots i_{r}} dy^{i_1} \wedge \cdots \wedge dy^{i_{r}}\right) \\ &= \sum_{|I| =r} \alpha_{i_1\cdots i_{r}} (F(t, x)) dF^{i_1} \wedge \cdots \wedge dF^{i_{r}} \\ \end{split}$$

Note that $$dF^{i_j} = \frac{\partial F^{i_j}}{\partial t} (t, x) dt + \sum_{p=1}^{r-1} \frac{\partial F^{i_j}}{\partial x^p} (t, x) dx^p = X^{i_j}(F(t,x)) dt + H_t^* dy^{i_j} $$

as $X = \frac{\partial F}{\partial t}$ by definition of $F = H\gamma$. Thus

$$\begin{split} F^* \alpha&= \sum_{|I| =r} \alpha_{i_1\cdots i_{r}} \left( X^{i_1} dt + H_t^* dy^{i_1}\right) \wedge \cdots \wedge \left( X^{i_{k+1}} dt + H_t^* dy^{i_{r}}\right) \\ &= \sum_{|I| =r} \alpha_{i_1\cdots i_{r}} \ dt \wedge H_t^* \left( \sum_{j=1}^{r}(-1)^{j-1} X^{i_j} dy^{i_1} \wedge \cdots \wedge \hat{dy^{i_j}} \wedge \cdots \wedge dy^{i_{r}}\right) \\ &= \sum_{|I| =k+1} \alpha_{i_1\cdots i_{r}} \ dt \wedge H_t^*\left( i_X (dy^{i_1} \wedge \cdots \wedge dy^{i_{r}})\right) \\ &= dt \wedge H_t^*(i_X \alpha) = (-1)^{r-1} H_t^* (i_X\alpha) \wedge dt. \end{split}$$ Thus $$ \int_{H\gamma} \alpha = (-1)^{r-1} \int_{I\times \sigma} H_t^*(i_X \alpha) \wedge dt = (-1)^{r-1} \int_0^1 \int_\gamma H_t^*(i_X \alpha) dt.$$

  • $\begingroup$ Thanks for your answer. I have some questions. Why does Fubini's theorem produce a factor $(-1)^k$? Also, how did you obtain $\mathrm{d}F^{i_j}$? I don't see how you got either term in the result. $\endgroup$
    – Ryan Unger
    Oct 15 '15 at 14:46
  • 1
    $\begingroup$ In the "usual" Fubini theorem, you basically ignore the order of intergration, $dx dy$ is the same as $dydx$. But if you treat that as a differential form, they are different. In our situation I am using $dt\wedge dx^1\wedge \cdots \wedge dx^k = (-1)^k dx^1 \wedge \cdots \wedge dx^k \wedge dt$. $\endgroup$
    – user99914
    Oct 15 '15 at 19:36
  • $\begingroup$ For your second question, note that the pullback is calculated by (if $F = (F^1, \cdots, F^m)$) $F^* dy^{i_j} = dF^{i_j}$. @0celo7 $\endgroup$
    – user99914
    Oct 15 '15 at 19:37
  • 1
    $\begingroup$ The first one comes from the definition as $F (t, x)= g^t \gamma(x)$, so $\frac{\partial F}{\partial t} (t, x) = \frac {d}{dt} g^t \gamma(x) = X(F(t, x))$. $\endgroup$
    – user99914
    Oct 15 '15 at 20:09
  • 1
    $\begingroup$ For the second one, the definition of $H_t$ is $H_t(x) = F(t, x)$. So $H_t^* dy^{i_j} = dH_t^{i_j} = \frac{\partial H_t^{i_j}}{\partial x^p} dx^p$. But $\frac{\partial H_t^{i_j}}{\partial x^p} = \frac{\partial F^{i_j}}{\partial x^p} $ @0celo7 $\endgroup$
    – user99914
    Oct 15 '15 at 20:12

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.