1
$\begingroup$

Find the vector with $||x||^2=x^Tx=1$ that maximizes the following function.

$f(x) = 2x_1^2+2x_2^2-x_3^3+2x_1x_2$

I have rewritten the quadratic form as

$f(x) = \frac{1}{2}x^T \begin{bmatrix} 4&2&0\\2&4&0\\0&0&-2\end{bmatrix} x$

in which the gradient can be found as $\nabla{f}=Qx+b$ where $b = 0$ But this is exactly the point where I'm stuck.

I want to use the First-order necessary condition to find the maximizer but how can I find the vector that maximizes the function if I can't use the FONC because $b=0$?

EDIT: Due to the comment section I found out that this function is NOT quadratic. So if I just take the gradient I get

$\nabla f(x) = \begin{bmatrix} 4x_1+2x_2\\4x_2+2x_1\\-3x_3^2\end{bmatrix} $ but still, can someone show me how to do it from here?

EDIT2: Using Lagrange multiplier I get the Lagrange function:

$L(x,y) = 2x_1^2+2x_2^2-x_3^3+2x_1x_2-\lambda(x_1^2+x_2^2+x_3^2-1)$

Taking partial derivatives and setting it equal to zero I get

$\begin{bmatrix} 4x_1+2x_2-2\lambda x_1\\4x_2+2x_1-2\lambda x_2\\-3x_3^2-2\lambda x_3\\-x_1^2-x_2^2-x_3^2+1 \end{bmatrix} = \begin{bmatrix} 0\\0\\0\\0\end{bmatrix}$

But I get from the third equation that lambda equals -1 and then filling it in the other equations I get $x_1=0, x_2=0$ and $x_3 = 1$ or $x_3 = -1$ But as the person below kindly showed the answer should be 3.

$\endgroup$
  • 1
    $\begingroup$ Your function $f$ is NOT quadratic. $\endgroup$ – Batominovski Jul 26 '15 at 14:49
  • 1
    $\begingroup$ Is there a specific technique that you are trying to use to solve this problem? Because otherwise it is a simple matter of a lagrange multiplier in 3 variables. $\endgroup$ – DanielV Jul 26 '15 at 16:25
  • $\begingroup$ @DanielV I have edited and used the lagrange multiplier as you suggested but my answer differs from what was said below. Can you see what I am doing wrong? $\endgroup$ – Peter Jul 26 '15 at 17:14
  • $\begingroup$ The equations actually have several solutions~, you've only found 1 $\endgroup$ – DanielV Jul 26 '15 at 17:26
3
$\begingroup$

We have:

$$ 2x_1^2+2 x_1 x_2+2x_2^2-x_3^3 = 3\left(\frac{x_1+x_2}{\sqrt{2}}\right)^2+\left(\frac{x_1-x_2}{\sqrt{2}}\right)^2-x_3^3$$ hence the problem is equivalent to finding the maximum of: $$ 3a^2+b^2-c^3 $$ under the constraint $a^2+b^2+c^2=1$. Given that $a^2+b^2=1-c^2$, the maximum of $3a^2+b^2$ is achieved when $b=0$, so the last problem is equivalent to finding the maximum of: $$ f(c)=3(1-c^2)-c^3 $$ under the constrant $c\in[-1,1]$. Such a maximum is attained at $c=0$ and it equals $\color{red}{3}$.

$\endgroup$
  • $\begingroup$ Thank you very much sir, can you please tell me why I get another value if I use the Lagrange multiplier? $\endgroup$ – Peter Jul 26 '15 at 17:22
  • $\begingroup$ @Peter: the stationary point you computed is the minimum. The maximum is given by $\lambda=3$. $\endgroup$ – Jack D'Aurizio Jul 26 '15 at 17:32

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.