3

How do I find the points on the surface: $$x^3+y^3+z^3=1$$ such that the distance to the origin is minimized?

My Thoughts:

Perhaps we can minimize the distance squared? Not sure.

gabby
  • 33
  • 3
  • One important point is to formulate the problem in a good way. You want to minimize $\sqrt{x^2+y^2+z^2}$ subject to $x^3+y^3+z^3=1.$ This is a good start to generate more ideas – epsilon Jul 16 '15 at 17:40
  • 1
    Hint: method of Lagrange multiplier (https://en.wikipedia.org/wiki/Lagrange_multiplier) – Leo Jul 16 '15 at 17:41
  • @Leo Okay. So I have set up f(x,y,z) and g(x,y,z). Now can you show me what I do next? – gabby Jul 16 '15 at 17:46

3 Answers3

1

You want to minimise the function: $f(x,y,z) = x^2+y^2+z^2$ with the condition that $g(x,y,z) = 1- x^3-y^3-z^3 = 0$

Note that if $x^2+y^2+z^2$ is minimal so is $\sqrt{x^2+y^2+z^2}$

This is best done using Lagrange multipliers. You define $ F(x,y,z,\lambda ): = f(x,y,z) + \lambda\cdot g(x,y,z)$

Then the set of equations

$$ \frac{\partial F}{\partial x} = 0,\ \frac{\partial F}{\partial y} = 0,\ \frac{\partial F}{\partial z} = 0,\ \frac{\partial F}{\partial \lambda}= 0 $$

will give us the answer. We get $x=\lambda 3x^2$ same for $y,z$ plus $g(x,y,z) = 0$

(i) $x,y,z \ne 0$

We get: $2x=\lambda 3x^2 \Rightarrow \lambda = \frac{2}{3x}$

Inserting into the second and third equation: $x = y = z$

With the last equation: $3x^3 = 1 \Rightarrow x =y=z = \frac{1}{\sqrt[3]{3}}$

(ii) $x,y\ne 0$ and $z = 0$

$ 2x^3 = 1 \Rightarrow x =y = \frac{1}{\sqrt[3]{2}}, z = 0$

(iii) $x\ne 0, y,z = 0$

$ x^3 = 1 \Rightarrow x = 1, y = z = 0$ and $\lambda = \frac23$.

Because of symmetry we will get the last two cases three times for the permutations of $x,y,z$

Now we calculate the distances:

(i) $d = 3^{\frac{1}{6}} = 1,2...$

(ii) $d = 2^{\frac{1}{6}} = 1,1...$

(iii) $d = 1$

So the points with minimal distance are case (1) $ x = 1, y=z=0$ (2) $ y = 1, x=z=0$ (3) $ z = 1, x =y=0$.

To check if we have a local minimum we could also calculate the Hessian matrix on the tangent space of $\nabla g(x,y,z) = (-3, 0, 0)^T$ for say case (1):

\begin{align} \frac{\partial^2F}{\partial y^2} = \frac{\partial^2F}{\partial z^2} &= 2 - 6\lambda y = 2 \\ \frac{\partial^2F}{\partial x \partial y} &= 0 \end{align} That is $$ \nabla^2_{(y,z)} f(1,0,0) = \begin{pmatrix} 2 & 0 \\ 0 & 2 \end{pmatrix}.$$

This is obviously positive define, thus $x=1,y=z=0$ is a strict local minimum.

Here is an alternative solution for the minima argument, not sure if this is correct though:

If I vary $y,z$ to be slightly different from zero: $y = \epsilon$, $z = \delta$ then x is given by $x = (1-\epsilon^3 -\delta^3)^{\frac{1}{3}}$. Now checking the distance: $ d= (1-\epsilon^3 -\delta^3)^{\frac{2}{3}} + \epsilon^2 +\delta^2$ tayor expanding the first term $d= 1 - \frac{2}{3}\epsilon^3 - \frac{2}{3}\delta^3 + \epsilon^2 +\delta^2 = 1 + \epsilon^2(1- \frac{2\epsilon}{3}) +\delta^2(1- \frac{2\delta}{3}) > 1$

similarly for just varying one component. Thus the points are indeed local minima.

user251257
  • 9,229
john
  • 1,470
  • I got $2^{(1/6)}$. Is that right? – gabby Jul 16 '15 at 18:45
  • Well done John, it might be nice to conclude your answer with the answer, thus calculating $\sqrt{x^2 + y^2 +z^2}$ for these three cases. – Jan Jul 16 '15 at 18:46
  • @Jan I got $2^{(1/6)}$. Is that right? – gabby Jul 16 '15 at 18:48
  • Yes gabby, taking $x =y=z = \frac{1}{\sqrt[3]{3}}$ we see that $\sqrt{x^2 + y^2 +z^2} = \sqrt{(3^{-\frac{1}{3}})^2 +(3^{-\frac{1}{3}})^2 + (3^{-\frac{1}{3}})^2} = \sqrt{3 \cdot 3^{-\frac{2}{3}}} = \sqrt{3^{\frac{1}{3}}} = (3^{\frac{1}{3}})^{\frac{1}{2}} =3^{\frac{1}{3} \cdot \frac{1}{2}}= 3^{\frac{1}{6}}$ – Jan Jul 16 '15 at 18:51
  • thanks, and yes I can do that, though I think the soltion is not yet right – john Jul 16 '15 at 18:52
  • @Jan Wouldn't the second case give you the minimum and not the first? – gabby Jul 16 '15 at 18:56
  • @John what do you mean by "Because of symmetry we will get the last two cases for three times for the permutations of x,y,z" – gabby Jul 16 '15 at 18:59
  • No, since $\sqrt[3]{3} > \sqrt[3]{2}$ we see that $\frac{1}{\sqrt[3]{3}} < \frac{1}{\sqrt[3]{2}}$. We want to take the smallest values for $x,y,z$ since we will square these numbers, add them and than take the root. Squaring a smaller number yields a smaller number, the same goes for taking roots (if the original number is bigger than $1$), so we take $x=y=z= \frac{1}{\sqrt[3]{3}}$. – Jan Jul 16 '15 at 19:02
  • Okay and what does john mean by "Because of symmetry we will get the last two cases for three times for the permutations of x,y,z" – gabby Jul 16 '15 at 19:03
  • He means that if you take $y\ne 0, x,z = 0$ instead of $x\ne 0, y,z = 0$ we will find that we switched the values for $x$ and $y$. This does not influence the length. – Jan Jul 16 '15 at 19:05
  • @Jan Okay. When I plugged in 3^(1/6) and 2^(1/6) into my calculator, I got that 2^(1/6)<3^(1/6), but you are saying the opposite. – gabby Jul 16 '15 at 19:06
  • @John Where are you getting the distances from? – gabby Jul 16 '15 at 19:12
  • Correct but if you take $x=y=z=\frac{1}{\sqrt[3]{2}}$ we get for $\sqrt{x^2 + y^2 +z^2} = \sqrt{ (\frac{1}{\sqrt[3]{2}})^2 +(\frac{1}{\sqrt[3]{2}})^2 +(\frac{1}{\sqrt[3]{2}})^2} = \sqrt{ 3 \cdot 2^{\frac{2}{3}}} = \sqrt{3} \cdot (2^{\frac{2}{3}})^{\frac{1}{2}}= \frac{\sqrt{3}}{\sqrt[3]{2}}$. And $3^{\frac{1}{6}} < \frac{\sqrt{3}}{\sqrt[3]{2}}$. – Jan Jul 16 '15 at 19:12
  • @Jan Can you fix it, Its hard to see – gabby Jul 16 '15 at 19:12
  • @user251257 Why, is this wrong? – gabby Jul 16 '15 at 19:19
  • oh, my bad. had inequality in mind. – user251257 Jul 16 '15 at 19:20
  • @user251257 So is this right? – gabby Jul 16 '15 at 19:21
  • just ignore my comment. my bad :) – user251257 Jul 16 '15 at 19:22
  • @user251257 Or do you maen they are still wrong, because I believe Jan has it wrong. for $x = y = z = 2^{-\frac{1}{3}}$ there is no extremum, in case (ii) $ z = 0$ so $d = \sqrt{(2^{-\frac{1}{3}})^2 + (2^{-\frac{1}{3}})^2+0^2} = \sqrt{2\cdot (2^{-\frac{1}{3}})^2} = 2^{\frac{1}{2}}\cdot 2^{-\frac{1}{3}} = 2^{\frac{3}{6}}\cdot 2^{-\frac{2}{6}} =2^{\frac{1}{6}} $ – john Jul 16 '15 at 21:33
  • @john The numbers in the post are right. I got the same numbers. But I made a mistake about the Hessian. It need to be positive definite. – user251257 Jul 16 '15 at 21:35
  • Yes thanks. I will add the Hessian to make it complete – john Jul 16 '15 at 21:56
  • @user251257 somehow the argument with the hessian matrix doesn't work. Any ideas why? – john Jul 18 '15 at 14:45
  • @john: I made an edit to fix your second order condition argument. It is going through review right now. – user251257 Jul 18 '15 at 16:23
0

For any point $ \mathbf{P} = (x, y, z) $, the distance to the origin is given by $ d(x, y, z) = \sqrt{x^2+y^2+z^2}$. This is what you wish to minimize, given the constraint $ g(x, y, z) = x^3+y^3+z^3 = 1$. Luckily, a man named Lagrange found a clever way to do this problem. You instead solve the equation $ \nabla d(x, y, z) = \lambda \nabla g(x, y, z) $, where $ \nabla g(x) = \langle g_x, g_y, g_z \rangle $. The $(x, y, z) $ solutions will give you your min/max values.

Triangle
  • 393
  • This is where I'm having trouble. I got to this point. I also wrote down the three equations, but now I don't know how to proceed. – gabby Jul 16 '15 at 17:48
  • Try solving for $ \lambda $ in each separate equation, then setting them all equal, you should then be able to solve for all three variables in terms of one – Triangle Jul 16 '15 at 17:50
  • Can you also work this problem with me to see if I get it correct? – gabby Jul 16 '15 at 17:51
  • I got $2^{(1/6)}$. Is that right? – gabby Jul 16 '15 at 18:45
0

If we take $x=\rho\sin\theta,y=\rho\cos\theta\sin\phi,z=\rho\cos\theta\cos\phi$, then we have to find the minimum of $\rho$ under the constraint:

$$ \rho^3(\sin^3\theta+(\sin^3\phi+\cos^3\phi)\cos^3\theta)=1 $$ or, by switching to the dual problem, the maximum of: $$ f(\theta,\rho) = \sin^3\theta+(\sin^3\phi+\cos^3\phi)\cos^3\theta.$$ The stationary points of $f(\theta,\rho)$ are given by: $$ \left\{\begin{array}{rcl}\frac{\partial}{\partial \theta}\,f(\theta,\phi)=-3\cos\theta\sin\theta\left(-\sin\theta+\cos^3\phi\cos\theta+\cos\phi\cos^3\theta\right)&=&0\\\frac{\partial}{\partial \phi}\,f(\theta,\phi)=-3\cos\phi\sin\phi\left(-\sin\phi+\cos^3\phi\cos\theta+\cos\phi\cos^3\theta\right)&=&0\end{array}\right.$$ so, at least in principle, we can locate them. Anyway, since: $$ 1-\left(\cos^3 x+\sin^3 x\right) = 4\sin^2\left(\frac{\pi}{4}-\frac{x}{2}\right)\sin^2\left(\frac{x}{2}\right)\left(2+\sqrt{2}\sin\left(\frac{\pi}{4}+x\right)\right)\geq 0$$

$$ 1+\left(\cos^3 x+\sin^3 x\right) = 4\sin^2\left(\frac{\pi}{4}+\frac{x}{2}\right)\cos^2\left(\frac{x}{2}\right)\left(2-\sqrt{2}\sin\left(\frac{\pi}{4}+x\right)\right)\geq 0$$ the range of $g(x)=\sin^3 x+\cos^3 x$ is $[-1,1]$, so the maximum of $f(\theta,\phi)$ is one and the minimum of $\rho$ is $\color{red}{1}$, too. Obviously, it is attained at $(1,0,0),(0,1,0),(0,0,1)$.

For non-believers, here it is a picture of the unit sphere with our surface being a nice hat:

enter image description here

Jack D'Aurizio
  • 353,855