8
$\begingroup$

I know the problem is traditionally solved via the isoperimetric inequality, but I was hoping to solve it by minimizing a surface of revolution subject to a volume constraint.

The surface area of a surface of revolution is:

$$A=2\pi\int_{-1}^{1}y\sqrt{1+\dot{y}^2}~dx$$

and the volume will be:

$$V=\pi\int_{-1}^{1}y^2~dx$$

I'd like to show that the sphere (or in this case the function $x^2+y^2=1$) minimizes the surface area functional for any fixed volume.

The combined action will then be:

$$S=\int_{-1}^{1}2\pi y\sqrt{1+\dot{y}^2}+\lambda\pi y^2 dx$$

The Euler Lagrange Equation simplifies to:

$$-\frac{2 \pi y'(x)^2}{\sqrt{y'(x)^2+1}}+2 \pi \sqrt{y'(x)^2+1}+\frac{2 \pi y(x) y'(x)^2 y''(x)}{\left(y'(x)^2+1\right)^{3/2}}-\frac{2 \pi y(x) y''(x)}{\sqrt{y'(x)^2+1}}+2 \pi \lambda y(x)=0$$

I have no clear idea how to proceed from here. I can solve for $\lambda$ using Mathematica, and get

$$\lambda=\frac{y(x) y''(x)-y'(x)^2-1}{y(x) \left(y'(x)^2+1\right)^{3/2}}$$

The ODE solution is extremely convoluted, involving about half a dozen elliptic integrals, and doesn't seem to be going anywhere. Is there something fundamentally wrong with my approach, or is there an obvious next step in the problem?

EDIT: Using the Beltrami identity instead, and reversing the problem so the volume is varied and a constant surface area is the constraint, the EL equation can be simplified to:

$$y(x)^2+\frac{2\lambda y(x)}{\sqrt{y'(x)^2+1}}=k$$

Which, while much simpler, doesn't get me closer to an answer.

If $k$ is equal to $0$ and $y(0)$ is likewise, then the ODE reduces nicely, and can be solved. One solutions is:

$$y(x)=\sqrt{4 \lambda ^2-x^2}$$

Which is the equation of a circle (which when rotated around the axis becomes a sphere). However, some assumptions are made there which are not necessarily true.

$\endgroup$
  • $\begingroup$ $S$ denotes both surface area and action. Kind of confusing. $\endgroup$ – Rodrigo de Azevedo May 25 '16 at 0:09
  • $\begingroup$ Fixed. Thanks. Accustomed to using S for almost everything. $\endgroup$ – JAustin May 25 '16 at 0:11
  • 1
    $\begingroup$ $k$ should be zero since you should assume that $y(-1) = y(1) = 0$. $\endgroup$ – user99914 May 25 '16 at 2:14
7
$\begingroup$

I spent a few more hours on the problem, and eventually found the solution. No major breakthroughs, but a lot of algebra.

First, I reversed the problem, instead maximizing the volume subject to a surface area constraint. Not strictly necessary, but it simplifies the computation. I also gave the surface area a magnitude $k$, per user7530's comment. The new action is:

$$S=\int_{-\lambda/2}^{\lambda/2}\pi y^2 + 2\lambda\pi y\sqrt{1+\dot{y}^2}-\lambda k~dx$$

The integration boundaries were also changed for convenience.

Then, because the Lagrangian does not depend explicitly on the variable $x$, I was able to apply the Beltrami identity instead of the traditional EL equation.

$$\frac{d}{dx}\left(\pi y(x)^2 + 2\lambda\pi y(x) \sqrt{1+y'(x)^2}-\lambda k - \frac{2\pi\lambda y'(x)^2y(x)}{\sqrt{1+y'(x)^2}}\right) = 0$$

Integrating both sides, we find that the functional inside the derivative is equal to a constant. Because $y$ is equal to $0$ at $-\lambda/2$ and $\lambda/2$, and each term in the derivative depends on $y(x)$ except for the $\lambda k$ term, the arbitrary additive constant must be $-\lambda k$, which cancels with the $-\lambda k$ on the left side.

The DE becomes:

$$\pi y(x)^2 + 2\lambda\pi y(x) \sqrt{1+y'(x)^2} - \frac{2\pi\lambda y'(x)^2y(x)}{\sqrt{1+y'(x)^2}} = 0$$

Multiplying the second term by $\frac{\sqrt{1+y'(x)^2}}{\sqrt{1+y'(x)^2}}$, that in turn simplifies to:

$$y(x)^2 + \frac{2\lambda y(x)}{\sqrt{1+y'(x)^2}}=0$$

This can be solved for $y(x)$, and with the constant chosen to equal zero, the solution is the equation of a circle of radius $2\lambda$.

$$y(x)=\pm\sqrt{4 \lambda ^2-x^2}$$

$\endgroup$
  • 1
    $\begingroup$ Radius $2 \lambda$ right? $\endgroup$ – ITA Sep 13 '16 at 15:29
  • $\begingroup$ Fixed. Thank you. For the record, I am also accepting my own answer, since the problem is solved. $\endgroup$ – JAustin Sep 14 '16 at 19:29
5
$\begingroup$

There are a few problems with the formulation:

  1. The volume constraint: right now you are constraining the total volume to be zero (check by differentiating $S$ with respect to $\lambda$). If you want the total volume to be equal to some fixed value $C$, you need instead the action

$$S = \int_{-1}^1 \left[2\pi y \sqrt{1+y'^2} + \lambda(\pi y^2-C/2)\right]\,dx.$$

  1. The boundary conditions: your problem restricts your surface of revolution to touch the bounding points $(\pm 1,0,0)$. Of course, depending on the value of $C$ the sphere of volume $C$ will not usually touch these points. You can still minimize surface area, given these boundary conditions, but realize you are solving a harder version of the isoperimetric problem. (Which I also cannot solve at the moment).
$\endgroup$
  • $\begingroup$ Where is the $\mathrm{d}x$? $\endgroup$ – Rodrigo de Azevedo May 25 '16 at 1:26
  • $\begingroup$ @RodrigodeAzevedo at the end, of course. added. $\endgroup$ – user7530 May 25 '16 at 1:42
  • $\begingroup$ I had the same thoughts about boundary conditions when I began the problem, and couldn't see how to actually constrain the problem with the boundary value. I also half-deliberately ignored the magnitude of the volume because it makes the integral even more unpleasant. But hey, who doesn't love page-long elliptic integrals?. $\endgroup$ – JAustin May 25 '16 at 1:57
  • $\begingroup$ I eventually decided to reverse the problem and vary the volume subject to a constant surface area, which makes the problem marginally easier, but I'm still not getting anywhere. As I mentioned in my edit, I can get an answer if I assume $k$, $C$, and $y(0)$ are equal to zero, but then the problem isn't very meaningful. $\endgroup$ – JAustin May 25 '16 at 1:57
2
$\begingroup$

I have to solve a similar problem, here are my thoughts.

We could express the problem in spherical coordinates with the domain $\Omega \in \mathbb{R}^3$ delimited by $\partial \Omega$, the Surface $A = \vert \partial \Omega \vert$ and the Volume $V = \vert \Omega \vert = constant$:

$$min\{A[R]\} = \int_{\partial\Omega} R^2(\varphi,\theta) {\rm sin}(\theta){\rm d}\varphi {\rm d}\theta$$ $$constant = V = \int_{\Omega} r^2 {\rm sin}(\theta) {\rm d} r {\rm d} \varphi {\rm d} \theta = \int_{\partial\Omega} \left( \int \limits_{0}^{R(\varphi,\theta)} r^2 {\rm d} r \right) {\rm sin}(\theta) {\rm d}\varphi {\rm d}\theta = \int_{\partial\Omega} \frac{R^3(\varphi,\theta)}{3} {\rm sin}(\theta) {\rm d}\varphi {\rm d}\theta $$

Now we have two integrals over the same domain $\partial\Omega$ and we can use formulas for problems under isoperimetric conditions.

With $F := R^2(\varphi,\theta){\rm sin}(\theta)$ and $G := \frac{R^3(\varphi,\theta)}{3} {\rm sin}(\theta)$ and $F^* = F + \lambda G$, we can use the formula $$ 0 = \underbrace{ \frac{{\rm d}}{{\rm d} \theta} \frac{ \partial F^* }{ \partial \frac{ \partial R }{ \partial \theta}} }_{=0} + \underbrace{ \frac{{\rm d}}{{\rm d} \varphi} \frac{ \partial F^* }{ \partial \frac{ \partial R }{ \partial \varphi}} }_{=0} - \frac{ \partial F^* }{ \partial R} $$

Therein is $$ F^* = \underbrace{ R^2(\varphi,\theta) {\rm sin}(\theta) }_{=F} + \lambda \cdot \underbrace{ \frac{R^3(\varphi,\theta)}{3} {\rm sin}(\theta) }_{=G} $$

So it follows $$ 0 = \frac{ \partial F^* }{ \partial R} = 2 R (\varphi,\theta){\rm sin}(\theta) + \lambda \cdot R^2(\varphi,\theta) {\rm sin}(\theta) $$

Solving the equation for $R$: $$ R(\varphi,\theta) = \frac{2}{\lambda} $$

So we have shown, that $R(\varphi,\theta) = constant$, because $\lambda$ is constant in $\varphi$ and $\theta$.

$\endgroup$

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.