I was trying to find out if every positive (do we need this?) monotone increasing function $f :\mathbb{R}_{+}\to\mathbb{R}_{+}$ was necessarily the derivative of a convex function. I have found this answer that seems related but I am not so sure if this is the good approach. Do I have to verify that for $n$ points $x_1,\ldots, x_n$ I have $\langle x_0-x_1,f(x_0)\rangle+\ldots+\langle x_{n-1}-x_n,f(x_{n-1})\rangle+\langle x_{n}-x_0,f(x_n)\rangle\ge 0$? (for every $f$ as above)

  • $\begingroup$ Any function $f :\mathbb{R}_{+}\to\mathbb{R}_{+}$ is necessarily positive $\endgroup$ – Henry Apr 13 at 14:43
  • $\begingroup$ Hint: A function with increasing derivative is convex. $\endgroup$ – Giorgos Giapitzakis Apr 13 at 14:45
  • $\begingroup$ It slightly depends on your definition of derivative. $f(x) =x$ for $0 < x < 1$; $f(1)=3$; $f(x)=x+10$ for $x>1$ is a monotone increasing function but not obviously the derivative of its integral $\endgroup$ – Henry Apr 13 at 14:47
  • $\begingroup$ you don't need positive. Check - $x^2$ has negative derivative for $x<0$, but you may want to add continuous to the requirement? $\endgroup$ – Rahul Madhavan Apr 13 at 14:48

If $F: I \to \Bbb R$ is convex and differentiable on an interval $I \subset \Bbb R$ then $F'$ is increasing and continuous, see for example Continuity of derivative of convex function.

Conversely, if $f: I \to \Bbb R$ is increasing and continuous then $$ \tag{*} F(x) = \int_a^x f(t) \, dt $$ (for some $a \in I$) is convex with $F'=f$.

If $f$ is only increasing but not necessarily continuous then we can still define $F$ via $(*)$ because monotone functions are Lebesgue integrable. Then $F' = f$ a.e. in $I$ and $F$ is convex: For $x < y < z$ is $$ \frac{F(y)-F(x)}{y-x} = \frac{1}{y-x} \int_x^y f(t) \, dt = \int_0^1 f(x + s(y-x)) \, ds \\ \le f(y) \le \int_0^1 f(y + s(z-y)) \, ds = \frac{F(z)-F(y)}{z-y} \, . $$

  • $\begingroup$ If $f$ is continuous I understand that $F$ is convex, since $F'=f$ and $f$ is increasing. But what is the argument in the case $F'=f$ a.e.? $\endgroup$ – roi_saumon Apr 13 at 20:04
  • $\begingroup$ @roi_saumon: I have added a sketch of the argument. $\endgroup$ – Martin R Apr 13 at 20:11
  • $\begingroup$ Thank you. Is there a typo $f'\to f$? $\endgroup$ – roi_saumon Apr 13 at 23:22
  • $\begingroup$ @roi_saumon: Yes, it should be $f$ (or $F'$). Hopefully correct now. $\endgroup$ – Martin R Apr 14 at 4:52

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.