Let $f:[a,b]\rightarrow \mathbb R$ be a continuous function such that

$$\frac{f(x) - f(a)}{x-a}$$

is an increasing function of $x\in [a,b]$. Is $f$ necessarily convex? What if we also assume that

$$\frac{f(b) - f(x)}{b-x}$$

is increasing in $x$?

  • $\begingroup$ I am adding the condition that $f$ be continuous. $\endgroup$ – user15464 Jul 18 '12 at 2:21

A counterexample (to both) can be constructed from the function $f(x)=x^2$ on the interval $[0,1]$ by adding a little bump to the graph, say, near the point $(1/2,1/4)$.

  • $\begingroup$ Or, more crudely, try $f(x) = x^2$ when $x \in [0,\frac{1}{2}]$, and $f(x) = x^2+1$, when $x \in (\frac{1}{2},1]$. $f$ is not continuous at $x=\frac{1}{2}$, hence it is not convex. $\endgroup$ – copper.hat Jul 18 '12 at 2:11
  • $\begingroup$ I don't think one can make a small enough bump for non-convexity to hold, while still satisfying both given assumptions. Perhaps I'm not interpreting "bump" correctly. Could you elaborate, perhaps? $\endgroup$ – Cameron Buie Jul 18 '12 at 2:12
  • $\begingroup$ Or for a continuous counterexample, $f(x) = x^2$ on $[0,\frac{1}{2}]$, $f(x) = (x-\frac{1}{2})^2+\frac{3}{4}(x-\frac{1}{2})+\frac{1}{4}$ on $[\frac{1}{2},1]$. A little harder to verify, but it is non-convex, and $\frac{f(x)}{x}$ is increasing. $\endgroup$ – copper.hat Jul 18 '12 at 2:27
  • $\begingroup$ @Cameron What I had in mind: if you replaced a part of parabola with its chord, both quotients remain strictly increasing, but the function is no longer strictly convex. Now replace the chord with a slightly concave arc. $\endgroup$ – user31373 Jul 18 '12 at 2:44
  • $\begingroup$ Ah! I thought you meant a vertical shift to give a jump discontinuity. That makes more sense. $\endgroup$ – Cameron Buie Jul 18 '12 at 2:51

Yes $f$ is necessarily convex (with the added assumption continuous), if $ \displaystyle x\mapsto \frac{f(x)-f(a)}{x-a}$ is increasing over $(a,b]$. Let $a<x<y$ and define $t\in (0,1)$ by $x=ta+(1-t)y$. We have $\displaystyle \frac{f(x)-f(a)}{(1-t)(y-a)} = \frac{f(x)-f(a)}{x-a}\leqslant \frac{f(y)-f(a)}{y-a}$, hence $f(x)\leqslant tf(a) +(1-t)f(y)$.


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.