1
$\begingroup$

Just trying to solve this question: $f(x,y,z) = x^2 + y^2 +3z^2 -xy +2xz+ yz$. Found the only critical point of the function and explain why she is an absolute minimum. We learn at class how to found that a point is a critical point so i found her and its $(0,0,0)$. I know how to say that its a Local minimum, but i don't know how i am supposed to explain why its an absolute minimum and i want to know how to approach it.

$\endgroup$
5
  • 1
    $\begingroup$ After differentiating with respect to $x, y, z$, setting their derivatives to $0$ yields a linear system of equations with a unique solution. Does that help? $\endgroup$
    – Badr B
    Commented May 25, 2021 at 10:57
  • $\begingroup$ @BadrB, I see why that implies there is a unique critical point, but why does it imply that the local minimum is a global minimum? $\endgroup$
    – Joe
    Commented May 25, 2021 at 11:48
  • $\begingroup$ @Joe If we assume that there is a local minimum that is not the global minimum, then that would mean there are at least two critical points (the local minimum and global minimum). If we know there is exactly one critical point, then it follows that the local minimum must in fact be the global minimum as well. $\endgroup$
    – Badr B
    Commented May 25, 2021 at 12:13
  • $\begingroup$ @BadrB, not true, see here: math.stackexchange.com/questions/121326/… $\endgroup$
    – Joe
    Commented May 25, 2021 at 12:15
  • 1
    $\begingroup$ @Joe Ah, I was assuming that the global minimum must also be a critical point. I suppose we can't make that assumption in this case. $\endgroup$
    – Badr B
    Commented May 25, 2021 at 12:36

1 Answer 1

2
$\begingroup$

I don't know if there's some immediate way to know that the local minimum is a global minimum, and I would like to see such answers if there are.

I would do some algebra to rewrite $f$ as:

$f(x,y,z) = 0.5(x-y)^2 + 0.5(y+z)^2 +(x+z)^2-0.5x^2+1.5z^2$

Then, the only potentially problematic term is $-0.5x^2$.

But minimizing $(x+z)^2-0.5x^2+1.5z^2$ over $z$, which occurs when $2(x+z)+3z=0$, i.e. $z=-0.4x$, we see that:

$(x+z)^2-0.5x^2+1.5z^2 \ge 0.36x^2 -0.5x^2 +0.24x^2 = 0.1x^2 \ge 0$

Therefore $f(x,y,z) \ge 0$

Note that, when I minimized $(x+z)^2-0.5x^2+1.5z^2$ over $z$ to show that $(x+z)^2-0.5x^2+1.5z^2 \ge 0.1x^2$, that was only to derive that inequality. I was not trying to minimize $f$ over only $z$, while ignoring $y$.

Another way to derive the same inequality would be to let $t = z + 0.4x$, with the choice inspired by the fact that we derived above that $z = -0.4x$ minimizes that expression.

Then we have $z = t - 0.4x$ and $z^2 = t^2 - 0.8tx + 0.16x^2$, and:

\begin{align} (x+z)^2-0.5x^2+1.5z^2 &= x^2 + 2xz + z^2 - 0.5x^2+1.5z^2 \\ & = 0.5x^2 +2xz+2.5z^2 \\ & = 0.5x^2 +2x(t - 0.4x)+2.5(t^2 - 0.8tx + 0.16x^2) \\ & = 0.1x^2 +2.5t^2 \\ &\ge 0.1x^2 \end{align}

Using this inequality in $f$ gives: \begin{align} f(x,y,z) &= 0.5(x-y)^2 + 0.5(y+z)^2 +(x+z)^2-0.5x^2+1.5z^2 \\ &\ge 0.5(x-y)^2 + 0.5(y+z)^2 + 0.1x^2 \\ &\ge 0 \qquad \text{ since it is the sum of squares}\\ \end{align}

$\endgroup$
4
  • $\begingroup$ I am not sure about the minimizing part. lets say h(x) = $(x+z)^2 - 0.5x^2 +1.5z^2$, if you try to minimize it as a 3 variable function than you get that the hessian is 0 always (deriatives of y deriatives are 0) and therefore you can't use the theoreme that when gradiant is 0 and hessian is > 0 there is a local minimum. If you will look at the function as two variables function you will get that x and z are both zero at the local minimums and that every point is a local minimum which is really weird and does not make sense. Also, i am not sure that you can look at it as 2 variables function. $\endgroup$
    – yuval
    Commented May 25, 2021 at 13:27
  • 1
    $\begingroup$ @yuval, I have edited my answer to show another derivation of the inequality that I used to show that $f(x,y,z)\ge 0$. Please let me know if you are still not sure about that part. $\endgroup$
    – Joe
    Commented May 25, 2021 at 15:12
  • $\begingroup$ Hi, thank you, i found another way to solve this question so maybe i will post this later but i think you solution is also really nice $\endgroup$
    – yuval
    Commented May 26, 2021 at 9:29
  • $\begingroup$ @yuval, thanks! I hope you will. I'd like to see it. $\endgroup$
    – Joe
    Commented May 26, 2021 at 11:02

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .