When reading about the optimization problem for Support Vector Machine in Bishop's book (Pattern Recognition and Machine Learning) he wrote that:

The optimization problem then simply requires that we maximize $\| \mathbf w\|^{-1}$ which is equivalent to minimizing $\| \mathbf w\|^{2}$

Here $\| \mathbf w\|$ is the euclidian norm of the vector $\mathbf w$ so it is always a positive number.

I am wondering why do we not minimize $\| \mathbf w\|$ instead of $\| \mathbf w\|^2$ ?

Does using $\| \mathbf w\|^{2}$ simplifies the optimization problem?

At first I thought that taking the square made the problem convex, but the function $$f(x,y)=\sqrt{x^2+y^2}$$ looks convex to me too.

  • 3
    $\begingroup$ Minimizing $\lVert\cdot\rVert$ and minimizing $\lVert\cdot\rVert^2$ are the same thing, aren't they? $\endgroup$
    – Clement C.
    Jun 6, 2016 at 13:04

1 Answer 1


Minimizing $|w|$ is the same as minimizing $|w|^2$. As I understand it, the reason we choose the latter is purely for convenience. Would you rather deal with partials of $\sqrt{x^2+y^2}$ or $x^2+y^2$? The latter, obviously, because it's easier.

  • 1
    $\begingroup$ And its derivative is well defined. $\endgroup$
    – Royi
    Jun 9, 2016 at 15:12

You must log in to answer this question.

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