4
$\begingroup$

Which of the following functions, $f: [0,\infty) \rightarrow [0,\infty)$, can be composed with a metric $d$ to get a new metric $f \circ d$:

a)$\;f(x) = \begin{cases}0 & \text{if $x=0$} \\x+1 & \text{if $x >0$}\end{cases}$

b)$f(x) = x^2, x \ge 0$

c)$f(x) = \arctan(x), x \ge 0$

In order for $f \circ d$ to be a metric we need $f$ to be monotonic and subadditive (and of course evaluate to 0 only when the input is 0) so that we can have the triangle inequality property satisfied. This is what I have so far:

a) $f$ is monotonic (looking at the graph, it is always increasing over $[0,\infty)$)
$\;\;\;f$ is subadditive since $1+(x+y) \le (1+x) + (1+y)$, for all $x,y$

b)$f$ is monotonic(looking at the graph, it is always increasing over $[0,\infty)$)
$\;\;\;f$ is not subadditive since $(x+y)^2 = x^2 + 2xy + y^2 \not\le x^2 + y^2$

c)$f$ is monotonic(looking at the graph, it is always increasing over $[0,\infty)$)
$\;\;\;f$ is or is not subadditive ??

I'm thinking that $\arctan(x)$ is subadditive since I can't think of any case where $\arctan(x+y) \le \arctan(x) + \arctan(y)$ doesn't hold over $[0,\infty)$.

Feedback on what I have so far is appreciated.

$\endgroup$
7
  • 1
    $\begingroup$ You have a mistake in a), $1 + (x+y) < (1+x) + (1+y) = 2+(x+y)$ for all $x,y$. $\endgroup$ – Daniel Fischer Oct 23 '14 at 15:20
  • 1
    $\begingroup$ For c), it may be a good idea to apply $\tan$ to both sides. $\endgroup$ – Daniel Fischer Oct 23 '14 at 15:22
  • 1
    $\begingroup$ For c), use that $\tan$ is increasing and the double angle formula $\tan(a + b) = \frac{\tan a + \tan b}{1 - \tan a \tan b}$ $\endgroup$ – Christopher Oct 23 '14 at 15:23
  • 1
    $\begingroup$ @user92638 I fixed the cases for a), you had $x=0$ and $x>1$ leaving $f$ undefined in $(0,1]$. $\endgroup$ – AlexR Oct 23 '14 at 15:29
  • 1
    $\begingroup$ Of course there are certain metrics $d$ for which b) does give a valid metric. The problem statement might be formulated in a better way to stress that $f\circ d$ should be a metric for all metrics $d$. $\endgroup$ – Hagen von Eitzen Oct 23 '14 at 15:31
9
$\begingroup$

In general, if $d$ is a metric, and $f$ is strictly increasing, $f(0)=0$, and concave on $[0,\infty)$, then $f\circ d$ is again an metric.

Here is a short proof for triangle inequality. Let $d_1=d(x,y),d_2=d(y,z),d_3=d(x,z)$, we have $d_1\leq d_2+d_3$. By increasing, we have $f(d_1)\leq f(d_2+d_3)$, and we want $f(d_2+f_3)\leq f(d_2)+f(d_3)$, which is equivalent to $\frac{f(d_2+d_3)-f(d_2)}{(d_2+d_3)-d_2}\leq \frac{f(d_3)-f(0)}{d_3-0}$. But this follows directly from concavity of $f$ since $0\leq d_2\leq d_2+d_3$.

Note, if $f$ is 2nd order differentiable, $f$ is concave iff $f''(x)\leq 0$ for all $x\in(0,\infty)$.

So by this criterion, a),c) will give you a metric.

$\endgroup$
1
  • $\begingroup$ Actually, it suffices that $f$ is increasing, $f(x) \geq 0 \wedge f(x) = 0 \Leftrightarrow x = 0$ (positive definite) and $f(x+y) \leq f(x) + f(y)$ (subadditive). Then we get the triangle inequality via: $f(d(x, y)) \leq f(d(x, y) + d(y, z)) \leq f(d(x, y)) + f(d(y, z))$$. $\endgroup$ – ComFreek Apr 30 '19 at 7:31

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.