Consider a time-dependent family of curves given in polar coordinates, so described by $r(\theta,t)$, which is assumed to be a positive $2\pi$-periodic function of $\theta$ for each fixed $t$, with the property that the periodic extension is smooth. (In other words the induced function on $S^1$ is smooth.)

Suppose that the point on the curve at each $\theta$ moves with velocity vector $v(\theta,t) \nu(\theta,t)$. Here $v$ is a real-valued function (with the same hypotheses as $r$ except that it can have either sign), and $\nu(\theta,t)$ is the outward unit normal to the curve at $\theta$. (If it helps later, you may assume $v>0$.) In general the outward unit normal has both a radial and an angular component; the angular component vanishes if and only if $\partial_\theta r = 0$ at that point.

I am trying to come up with a PDE for the evolution of $r(\theta,t)$, of the form $\frac{\partial r}{\partial t}=\dots$.

It seems like there is a transport term in this PDE. To see that, consider that in a time interval of length $h$ the point $r \mathbf{e}_r(\theta)$ is approximately displaced to $(r+avh) \mathbf{e}_r(\theta) + bvh \mathbf{e}_\theta(\theta)$, where $\mathbf{e}_r$ and $\mathbf{e}_\theta$ are unit radial and angular vectors. Here $a$ and $b$ are the coefficients in the expansion of $\nu(\theta,t)$ into components. Explicitly, we can write

$$\nu(\theta,t)=\frac{1}{\sqrt{r^2+(\partial_\theta r)^2}} (r\cos(\theta)+\partial_\theta r \sin(\theta),r\sin(\theta) - \partial_\theta r \cos(\theta))^T$$

with $\mathbf{e}_r=(\cos(\theta),\sin(\theta))^T$ and $\mathbf{e}_\theta=(-\sin(\theta),\cos(\theta))^T$. Therefore $a=\frac{r}{\sqrt{r^2+(\partial_\theta r)^2}}$ and $b=-\frac{\partial_\theta r}{\sqrt{r^2+(\partial_\theta r)^2}}$.

If I try to push this a little further, it seems like the angle that is displaced to end up at $\theta$ is $\theta' \approx \theta - \frac{bvh}{r}$. The initial $r$ at $\theta'$ (before accounting for the motion in the radial direction) is approximately $r-\frac{bvh}{r} \partial_\theta r$. So it seems that we recover

$$\frac{\partial r}{\partial t} = v \cdot \left ( a - b \frac{\partial_\theta r}{r} \right )$$

which I guess can be simplified further by plugging in $a$ and $b$, to get

$$\frac{\partial r}{\partial t}=\frac{v}{\sqrt{r^2+(\partial_\theta r)^2}} \left ( r + \frac{(\partial_\theta r)^2}{r} \right ) \\ = \frac{v}{r\sqrt{r^2+(\partial_\theta r)^2}} \left ( r^2+(\partial_\theta r)^2 \right ) \\ = \frac{v \sqrt{r^2+(\partial_\theta r)^2}}{r}.$$

Is this correct? I can tell it is correct in the case of the circle but it seems simpler than I expected in the general situation.

  • $\begingroup$ Since to first order every curve is a line, just check on a generic line $r=p\sec(\theta-\theta_0)$ with $v$ constant. That gives $r(\theta,t)=(p+tv)\sec(\theta-\theta_0)$ so you should end up with $\frac{\partial}{\partial t}r=v\sec(\theta-\theta_0)$, which it does. $\endgroup$ Oct 8 at 16:20
  • $\begingroup$ @user10354138 With $r(\theta,0)=\sec(\theta)$, say, you have $r^2+(\partial_\theta r)^2=\sec(\theta)^2 + \sec(\theta)^2 \tan(\theta)^2 = \sec(\theta)^2+\sec(\theta)^4-\sec(\theta)^2=\sec(\theta)^4$, then the square root gives $\sec(\theta)^2$, then the division by $r$ gives $\sec(\theta)$ back again. So the vertical line with constant velocity case is handled correctly also (a decent sign...) $\endgroup$
    – Ian
    Oct 8 at 16:24

To do it more rigorously (read analysis-y), we have $S^1=\mathbb{R}/2\pi$, a smooth $$ (\rho,\psi)\colon\mathbb{R}\times S^1\to\mathbb{R}^+\times S^1 $$ such that:

  • $\psi(t,-)$ is a diffeomorphism $S^1\to S^1$ for all $t$, say orientation preserving, with inverse $\phi(t,-)$;
  • normal deformation flow (*) \begin{align*} (\partial_1\rho,\partial_1\psi) &=v(\psi)\frac{(\rho\partial_2\psi,-\frac1\rho\partial_2\rho)}{\sqrt{(\rho\partial_2\psi)^2+\rho^2(\frac1\rho\partial_2\rho)^2}}\\ &=\left( \frac{v(\psi)\rho\partial_2\psi}{\sqrt{(\rho\partial_2\psi)^2+(\partial_2\rho)^2}} , -\frac{v(\psi)\frac1\rho\partial_2\rho}{\sqrt{(\rho\partial_2\psi)^2+(\partial_2\rho)^2}}\right) \end{align*}

Then $$ r(\theta,t):=\rho(t,\phi(t,\theta)) $$ so we calculate \begin{align*} \psi(t,\phi(t,\theta))=\theta &\Rightarrow \frac{\partial}{\partial t}\psi(t,\phi(t,\theta))=0,\quad \frac{\partial}{\partial \theta}\psi(t,\phi(t,\theta))=1 \\ &\Rightarrow \partial_1\psi(t,\phi(t,\theta))+\partial_2\psi(t,\phi(t,\theta))\partial_1\phi(t,\theta)=0,\quad \partial_2\psi(t,\phi(t,\theta))\partial_2\phi(t,\theta)=1\\ &\Rightarrow \partial_1\phi(t,\theta)=-\frac{\partial_1\psi(t,\phi(t,\theta))}{\partial_2\psi(t,\phi(t,\theta))},\quad \partial_2\phi(t,\theta)=\frac1{\partial_2\psi(t,\phi(t,\theta))} \end{align*} hence $$ \frac{\partial r}{\partial\theta}(\theta,t)=\frac{\partial}{\partial\theta}\rho(t,\phi(t,\theta))=\partial_2\rho(t,\phi(t,\theta))\cdot\partial_2\phi(t,\theta)=\left.\frac{\partial_2\rho}{\partial_2\psi}\right\rvert_{t,\phi(t,\theta)} $$ and \begin{align*} \frac{\partial}{\partial t}r(\theta,t) &=\partial_1\rho(t,\phi(t,\theta))+\partial_2\rho(t,\phi(t,\theta))\partial_1\phi(t,\theta)\\ &=\partial_1\rho(t,\phi(t,\theta))-\partial_2\rho(t,\phi(t,\theta))\frac{\partial_1\psi(t,\phi(t,\theta))}{\partial_2\psi(t,\phi(t,\theta))}\\ &=\left.\left(\partial_1\rho-\partial_1\psi\frac{\partial_2\rho}{\partial_2\psi}\right)\right\rvert_{(t,\phi(t,\theta))}\\ &=\left.\left(\frac{v(\psi)\rho\partial_2\psi}{\sqrt{(\rho\partial_2\psi)^2+(\partial_2\rho)^2}}+\frac{v(\psi)\frac1\rho\partial_2\rho}{\sqrt{(\rho\partial_2\psi)^2+(\partial_2\rho)^2}} \frac{\partial_2\rho}{\partial_2\psi}\right)\right\rvert_{(t,\phi(t,\theta))}\\ &=v(\theta)\left.\left( \frac{(\rho\partial_2\psi)^2+(\partial_2\rho)^2}{\rho\partial_2\psi\sqrt{(\rho\partial_2\psi)^2+(\partial_2\rho)^2}} \right)\right\rvert_{(t,\phi(t,\theta))}\\ &=v(\theta)\left.\left( \frac1\rho\sqrt{\rho^2+\left(\frac{\partial_2\rho}{\partial_2\psi}\right)^2} \right)\right\rvert_{(t,\phi(t,\theta))}\\ &=\frac{v(\theta)}{r(\theta,t)}\,\sqrt{r(\theta,t)^2+\left(\frac{\partial r(\theta,t)}{\partial\theta}\right)^2}. \end{align*}

(*) The tangent (of the $S^1\to\mathbb{R}^+\times S^1$ at fixed $t$) is $(\partial_2\rho,\partial_2\psi)\in\mathbb{R}^2=T_\rho\mathbb{R}^+\times T_\psi S^1$, taking orthogonal with respect to metric $dr^2+r^2d\theta^2$ on $\mathbb{R}^+\times S^1$ at $(\rho,\psi)$ gives $\pm(\rho\partial_2\psi,-\frac1\rho\partial_2\rho)$. Since we assume $\psi(t,-)$ is orientation-preserving we choose the + sign to have the first component positive.


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.