Find the minimum e max distance (in $R^2$) ) between the point $Q = ( 3/ 2 , − 3/ 2 )$ and the set $$B = \{(x, y) ∈ R^2 : yx = 1, x ≥ 0, y ≥ 0\}$$

In other words I have to find max /min points of the function

$${(x-3/2)^2 + (y+3/2)^2}$$

The set $B$ is clearly neither bounded nor convex. If I use Lagrange I can only find local min/max, but how do I show they are global? In a previous question :Does the following function admit a maximum? you suggested me to maximize/minimize the $x$ component and the $y$ component independently, but it is not clear to me if I can do it in this exercise as well. It seens to me that choosing $x=3/2$, which clearly minimizes the first component of the sum, would restrict the choice of $y$ as $yx = 1 $.

Is there any way to show that the point found using lagrange is a global min? Possibly without using the bordered hessian method?

  • $\begingroup$ Note that $C=\{ (x,y) | yx \ge 1, x,y \ge 0 \}$ is convex (and closed) and $Q \notin C$, so there is a unique point of minimal norm. Since the minimiser must be on the boundary and $B$ is the boundary, it is the minimiser on $B$ and hence is unique. $\endgroup$ – copper.hat Jan 27 at 17:17
  • $\begingroup$ Why the fact that the set C is convex and closed implies that there is a unique min point ? $\endgroup$ – LearningProb Jan 27 at 17:28
  • $\begingroup$ If the point is $(3/2,-3/2)$, then the distance is $\dots+(y+3/2)^2$. The problem can be solved directly by writing it as a function of $x$ and taking 1st derivative. I'm not sure if that's what you want. $\endgroup$ – bjorn93 Jan 27 at 17:31
  • $\begingroup$ Well, since $A=\overline{B}(Q, 100)$ (say) is closed and bounded and must contain any closest point we can consider the minimum of a convex function (distance) over a convex compact set and a minimiser must exist. (It is a more general result that a closed convex set has a unique nearest point in the Euclidean norm.) $\endgroup$ – copper.hat Jan 27 at 17:32
  • $\begingroup$ Now it clear thank you. $\endgroup$ – LearningProb Jan 27 at 17:57

(Note that you have a typo in your function, if the point is $(3/2,-3/2)$ then the second minus in your function should be a plus)

Using Lagrange multipliers is overkill here. In any case, Lagrange multipliers will give you critical points, which then you can check to see if they are the actual max/min.

Here, $x,y>0$ (otherwise $xy\ne1$), and you have that $y=1/x$. So your function becomes $$ \left(x-\tfrac32\right)^2+\left(\tfrac1x+\tfrac32\right)^2,\ \ \ \ x>0. $$ Now you can do this using one-variable calculus.

| cite | improve this answer | |
  • $\begingroup$ Is the equation $f'(x,y)=0$ more easy to solve than the Lagrange system? $\endgroup$ – Emilio Novati Jan 27 at 18:10
  • $\begingroup$ As far as I can tell, you end up getting the exact same quartic, but it has two easy roots. $\endgroup$ – Martin Argerami Jan 27 at 18:11
  • $\begingroup$ The equation $f'=0$ is of degree 3, but the system is of degree 2 (I think). Anyway, yes, there is the simple solution $x=2$. $\endgroup$ – Emilio Novati Jan 27 at 18:13
  • $\begingroup$ Yes, you are right. What happened is that I solved super quickly just to check and I used a non-efficient idea that increased the degree. $\endgroup$ – Martin Argerami Jan 27 at 18:32
  • $\begingroup$ Using both lagrange and $f'=0$ I still get $x^4$ but collecting terms it can then be simplified a little bit. $\endgroup$ – LearningProb Jan 27 at 18:40


The square of the distance function that you want minimize is $f(x,y)=(x-\frac{3}{2})^2+(y+\frac{3}{2})^2$ ( it seems that you have a wrong sign) with the condition $g(xy)=xy=1$ so, using Lagrange multipliers, you have to solve; $$ \begin{cases} \nabla f=\lambda \nabla g\\ xy=1 \end{cases} $$

can you do this?

| cite | improve this answer | |
  • $\begingroup$ Hi Emilio, yes I can do this but then how do I prove that the critical points are actually minima? $\endgroup$ – LearningProb Jan 27 at 18:35
  • $\begingroup$ I got that $(2,1/2)$ is a critical but how do I prove that it is a global min? $\endgroup$ – LearningProb Jan 27 at 21:49
  • $\begingroup$ Use the Hessian test (very simple in this case): en.wikipedia.org/wiki/Hessian_matrix $\endgroup$ – Emilio Novati Jan 27 at 22:47
  • $\begingroup$ Do you mean the bordered hessian method? Actually I do not know that method, that it is why I was looking for an alternative to it. $\endgroup$ – LearningProb Jan 28 at 1:09
  • $\begingroup$ Hi, I can not find it $\endgroup$ – LearningProb Jan 28 at 12:21

Lets say y->0 and x->$\infty$ then value of above expression -> $\infty$ (so yeah no maximum).

$(x-\frac{3}{2})^2 +(y+\frac{3}{2})^2$(You have made a mistake here or given the point wrong) = $x^2 +y^2 +\frac{9}{2} + 3(y-x) = z^2 +3z +\frac{13}{2}$.

(as $x=\frac{1}{y}$, $(y-\frac{1}{y})$=z, z can be any real number , the quadratic equation have its minimum at $\frac{-3}{2}$ which is the required answer)

So minimum value is $\sqrt{\frac{17}{4}}$

| cite | improve this answer | |

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.