In this Wikipedia article (see "Higher dimensions") there seems to be a connection between the Reynolds transport theorem (here) and the Lie derivative:

$$\frac{d}{dt}\int_{\Omega(t)}\omega=\int_{\Omega(t)} i_{\vec{\textbf v}}(d\omega)+\int_{\partial \Omega(t)} i_{\vec{\textbf v}} \omega+\int_{\Omega(t)}\dot{\omega} \qquad(1)$$

were $\omega$ is a p-form and the domain is time-varying. The symbol $i_X$ denotes the contraction with the vector field $X$.

By using the Cartan magic formula $L_{X} \omega = d (i_X \omega) + i_X (d \omega)$, it seems that

$$\frac{d}{dt}\int_{\Omega(t)}\omega=\int_{\Omega(t)} i_{\vec{\textbf v}}(d\omega)+\int_{\Omega(t)} d(i_{\vec{\textbf v}} \omega)+\int_{\Omega(t)}\dot{\omega} \, ,$$


$$\frac{d}{dt}\int_{\Omega(t)}\omega = \int_{\Omega(t)} L_{\vec{\textbf v}}\omega+\int_{\Omega(t)}\dot{\omega} \, .$$

It is clear that the Reynolds transport theorem is equivalent to the conservation of mass in hydrodynamics:

$$\frac{d}{dt}\int_{\Omega(t)}\rho = \int_{\Omega(t)} div(\rho \vec{\bf{v}}) +\int_{\Omega(t)}\dot{\rho} \, .$$

What is the relation between the two formulations? Clearly $\omega =\rho$ is not the correct thing to do: we can only integrate 3-forms and $\rho$ is a 0-form (the domain $\Omega(t)$ is three dimensional). How to derive the mass conservation from equation (1)?


There is indeed a connection. Recall the definition of the Lie derivative for a $k$-form $\omega$ along a vector field $X$. If $\chi$ is the one-parameter family of diffeomorphisms defined (at least locally) by $X$, then $$L_X\omega = \left.\frac{d}{d\tau}\right|_0(\chi_{\tau}^{*}\omega)$$ Integrating over some $k$-submanifold $\Omega$, we get $$ \int_\Omega L_X\omega = \int_{\Omega}\left.\frac{d}{d\tau}\right|_0(\chi_{\tau}^{*}\omega) = \left.\frac{d}{d\tau}\right|_0\int_{\Omega}\chi_{\tau}^{*}\omega = \left.\frac{d}{d\tau}\right|_0\int_{\chi_\tau[\Omega]}\omega $$ Note that we first used the Leibniz integral rule in one variable, and then the fact that diffeomorphisms preserve the integral. I claim Reynold's transport theorem is a special case of the formula $$\left.\frac{d}{d\tau}\right|_0\int_{\chi_\tau[\Omega]}\omega = \int_\Omega L_X\omega$$ we just obtained.

Say your space is an $n$-dimensional manifold $M$, and consider an interval $I\subseteq \mathbb R$ (say with $0\in I$) which for us will represent time. In the context of Reynold's transport theorem you have a submanifold $\Omega$ of $M$ (say of dimension $k$) which "varies (smoothly) over time". We can interpret this as having a fixed manifold $\Omega$ and a family of embeddings $i_t:\Omega\to M$. Using these embeddings, we can trace the "worldsheet" of $\Omega$ inside the "spacetime" $I\times M$. Define another family of embeddings $j_t:\Omega\to I\times M$ given by $$j_t(x)=(t,i_t(x)),$$ and define $\Omega_t:=j_t[\Omega]\subseteq I\times M$. Now we want to get the "spacetime velocity" of $\Omega$, which will be a vector field defined on $\Theta:=\bigcup_{t\in I}\Omega_t$. To do this, we observe that the embeddings $j_t$ are smooth, so they define a family of smooth curves $\gamma_x:I\to I\times M$ given for every $x\in\Omega$ by $\gamma_x(t)=(t,i_t(x))$. The vectors tangent to this family of curves give us a vector field $Y$ defined by $$Y_{(t,i_t(x))}=\gamma'_x(t)$$ for every $(t,i_t(x))\in\Theta=\bigcup_{t\in I}\Omega_t$. This is the spacetime velocity we wanted. Using the extension lemma for vector fields on submanifolds, we get a vector field $X$ on $I\times M$ which agrees with $M$ on $\Theta$ (we actually need to show that $\Theta$ is a submanifold, but I think this follows from the smoothness of all the maps).

Similarly, if you had a $k$-form $\bar{\omega}$ on $M$ which depends on time $t\in I$, then it can be regarded as a $k$-form $\omega$ on $M\times I$ by defining $$\omega_{(t,m)}(v_{(t,m)}) := \bar{\omega(t)_m}(\text{proj}_{TM}(v_{(t,m)}))$$ for every tangent vector $v_{(t,m)}\in T_{(t,m)}(I\times M)$.

Now applying the formula we first derived to the submanifold $\Omega_0$, we have $$\left.\frac{d}{dt}\right|_0\int_{\chi_t[\Omega_0]}\omega = \int_{\Omega_0} L_X\omega$$ where $\chi$ is the one-parameter family of diffeomorphisms induced by $X$ (we might need to show that $\chi$ is actually defined for all $t\in I$, but I think this is true at least in a neighbourhood of $\Omega_0$). Since $X$ extends $Y$, I think one can show that $\chi_t[\Omega_0]=\Omega_t$. Also, by Cartan's magic formula, we have $L_X\omega = d(\iota_X\omega) + \iota_X(d\omega)$. Hence

$$\left.\frac{d}{dt}\right|_0\int_{\Omega_t}\omega = \int_{\Omega_0} d(\iota_X\omega) + \int_{\Omega_0}\iota_X(d\omega)$$

When integrating the right hand side we can take charts consisting of a chart on $I$ times a chart on $M$, so $\omega$ has no $dt$ component because we defined it using $\bar\omega$. This way, letting $V=\text{proj}_{TM}Y$ we have $\iota_Y\omega=\iota_V\omega$ and even $\iota_X\omega=\iota_Y\omega=\iota_V\omega$ on $\Omega_0$. On the other hand, since $\Omega_0$ is contained in $\{0\}\times M\subseteq I\times M$ we have $dt|_{\Omega_0}=0$. These observations give the first two equalities in $$\int_{\Omega_0} d(\iota_X\omega)=\int_{\Omega_0} d(\iota_V\omega)=\int_{\Omega_0} d_M(\iota_V\omega)=\int_{\partial\Omega_0}\iota_V\omega$$ and the third one comes from Stokes' theorem.

Similarly, over $\Omega_0$ we have $X=Y=\frac{\partial}{\partial t}+V$, and since $\omega=\omega_Idx^I$ (where $I$ is a multiindex where $t$ does not appear) we have $$\begin{aligned} d\omega &=\partial_t\omega_Idt\wedge dx^I+\sum_i \partial_i\omega_Idx^i\wedge dx^I \\ &=\partial_t\omega_Idt\wedge dx^I+d_M\omega \end{aligned}$$ and the second integral on the right hand side becomes $$\begin{aligned} \int_{\Omega_0}\iota_X(d\omega) &= \int_{\Omega_0}\iota_{\frac{\partial}{\partial t}+V}(\partial_t\omega_Idt\wedge dx^I+d_M\omega) \\ &= \int_{\Omega_0}\left[\iota_{\frac{\partial}{\partial t}}(\partial_t\omega_Idt\wedge dx^I)+\iota_{\frac{\partial}{\partial t}}(d_M\omega)+\iota_V(\partial_t\omega_Idt\wedge dx^I)+\iota_V(d_M\omega)\right] \\ &= \int_{\Omega_0}\left[\partial_t\omega_Idx^I+\iota_V(d_M\omega)\right] \\ &= \int_{\Omega_0}\dot\omega+\int_{\Omega_0}\iota_V(d_M\omega) \end{aligned}$$ (note the second and third terms vanish because they pair "spacelike" basis forms with "timelike" basis vectors and viceversa).

Putting al this together, we have $$\left.\frac{d}{dt}\right|_0\int_{\Omega_t}\omega = \int_{\partial\Omega_0}\iota_V\omega+\int_{\Omega_0}\dot\omega+\int_{\Omega_0}\iota_V(d_M\omega)$$ as desired.


Instead of $\rho$, think about the three form $\rho ~\mathrm{d}x$. (Like I tell my calculus students, you should always write the integration element in an integral.) Here $\mathrm{d}x$ is the standard volume form.

By definition of the divergence of a vector field you have that $$ \mathrm{div}(\rho \vec{v}) ~\mathrm{d}x = \mathrm{d}( \iota_{\rho\vec{v}} ~\mathrm{d}x ) = \mathrm{d} (\iota_{\vec{v}} \rho ~\mathrm{d}x ) = L_{\vec{v}} \left(\rho~\mathrm{d}x\right) $$ which is what you expect.


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.