Consider $\mathbb{R}_t \times \mathbb{R}_x ^n$ , let $b_1(t,x)$ and $b_2 (t,x)$ be two velocity fields with all the regularity you want and consider the flow of the point $(0,x_0)$ for a time $T$. Now, I can view the trajectories of $(0,x_0)$ as curves in $\mathbb{R}^{n+1}$ like $ \gamma_1 =(t, \Phi_1 (t,x_0))$ and $\gamma _2= (t, \Phi_2 (t,x_0))$ with $\Phi_i$ the flux relative to $b_i$. The question is: if I consider $\tilde{\gamma} =\gamma_1 \cup \gamma_2 \cup I_T$ where $I_T$ is a segment in $\mathbb{R}^{n+1}$ joining $(T,\Phi_1 (T, x_0) )$ and $(T, \Phi_2 (T, x_0))$ , I can consider the Plateau problem associated to the rectifiable set $\tilde{\gamma}$, and by the general theory I'll have a minimal surface whose boundary is $\tilde{\gamma}$. The question is: given how my $\gamma$ is constructed, is there a way to estimate the area of such surface in terms of $b_1$ and $b_2$?

EDIT: Actually, what I want is to integrate such an estimate, starting from a certain mass distribution at time 0, so I would like something whixh takes that into account.


1 Answer 1


Almgren (https://mathscinet.ams.org/mathscinet-getitem?mr=855173) proves that the Plateau solution will satisfy the 2-dimensional Euclidean isoperimetric inequality. So you can bound $$ \textrm{Area(Plateau Solution)} \leq (4\pi)^{-1}(L(\gamma_1)+L(\gamma_2)+L(I_T))^2. $$ If you just want a coarse estimate you should be able to bound each term by the $L^\infty$ norm of $b_1,b_2$.

Of course, this inequality is unlikely to be close to optimal. (For example, if $b_1=b_2$ then the area of the Plateau solution is $=0$ while the RHS above is nonzero). If you want an estimate that works better in this case you might construct the explicit competitor $$ \Gamma : = \bigcup_{t=0}^T I_{\Phi_1(t,x_0),\Phi_2(t,x_1)} \times \{t\} $$ namely you union up all the lines between the points in the flow. You should be able to compute the area of this in terms of $b_1,b_2$ using the co-area formula.

EDIT: As requested here is a formula for the area of $\Gamma$. I did it via a parametrization instead of co-area but I think co-area would also work.

I parametrize $\Gamma$ by $F: [0,1]\times [0,T]\to\mathbb{R}^{n+1}$ $$ F(s,t) : = (s\Phi_1(t)+(1-s)\Phi_2(t),t) $$ (dropping the $x_0$ notation). Then we have $$ \partial_s F = \Phi_1-\Phi_2, \qquad \partial_t F = s\partial_t\Phi_1 +(1-s)\partial_t \Phi_2 + \mathbf{e}_{n+1}. $$ In particular, the induced metric is \begin{align*} g_{ss} & = |\Phi_1-\Phi_2|^2 \\ g_{st} & = \langle \Phi_1-\Phi_2,s\partial_t\Phi_1 +(1-s)\partial_t \Phi_2\rangle\\ g_{tt} & = 1 + |s\partial_t\Phi_1 +(1-s)\partial_t \Phi_2|^2. \end{align*} Hence $$ \det g = (1 + |s\partial_t\Phi_1 +(1-s)\partial_t \Phi_2|^2)|\Phi_1-\Phi_2|^2 - \langle \Phi_1-\Phi_2,s\partial_t\Phi_1 +(1-s)\partial_t \Phi_2\rangle^2. $$ This gives the explicit formula $$ \textrm{area}(\Gamma) = \int_{[0,1]\times [0,T]} \sqrt{(1 + |s\partial_t\Phi_1 +(1-s)\partial_t \Phi_2|^2)|\Phi_1-\Phi_2|^2 - \langle \Phi_1-\Phi_2,s\partial_t\Phi_1 +(1-s)\partial_t \Phi_2\rangle^2}. $$ If you want to write this in terms of $b_1$ and $b_2$ you can do the following (perhaps one can be more precise here to get a better estimate). First of all $$ |\partial_t \Phi_i| = |b_i| \leq \Vert b_i\Vert_{C^0} $$ Secondly, by the argument here we can bound $$ |\Phi_1-\Phi_2| \leq L^{-1}(e^{Lt}-1)|b_1-b_2|_{L^\infty}. $$ for $L$ the Lipschitz constant of one of the vector fields.

Using this above we should get $$ \textrm{area}(\Gamma) \leq (1 + (|b_1|_{C^0}+|b_2|_{C^0}) |b_1-b_2|_{C^0}^2 L^{-2}(e^{LT}-LT-1). $$

Note that I discarded the $g_{st}$ term, which could improve this estimate if you had some way of knowing it was big. But it does have the nice feature that it vanishes if $b_1=b_2$ and it's $O(T^2)$ for $T$ small.

I'm sure you can improve the estimate used above in the case of large $T$ (e.g. if $b_i$ are bounded, then the flows should diverge at most linearly).

  • $\begingroup$ Yeah, I kept it simple in the question but I'll edit it. In fact I don't have a single point, but rather a mass distribution at time 0 which flows, so I should, in a sense, integrate my estimate over all points $x_0$. Could you please give an idea on how to use the coarea formula on your example? I thought about that competitor you are mentioning, but how do you explicitly compute its area? $\endgroup$
    – tommy1996q
    Jun 8 at 16:56
  • 1
    $\begingroup$ Marvelous! Thanks a lot! $\endgroup$
    – tommy1996q
    Jun 9 at 17:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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