8

Let $f:(a,b)\rightarrow \mathbb{R}$.

$f$ satisfied the following property:

If $\forall x_{1},x_{0},x_{2}\in(a,b)$ and $x_{1}<x_{0}<x_{2};$then$\frac{f(x_{0})-f(x_{1})}{x_{0}-x_{1}}\geq \frac{f(x_{2})-f(x_{0})}{x_{2}-x_{0}}.$

My question : whether the function $f\in C((a,b))?$


We can get $\displaystyle\frac{f(x_{1})-f(x_{0})}{x_{1}-x_{0}}\geq \frac{f(x_{2})-f(x_{0})}{x_{2}-x_{0}}. $Let $\displaystyle g(x)=\frac{f(x)-f(x_{0})}{x-x_{0}},$ if we can prove $g(x)$ is bound and decreasing (not strictly ) in $(x_{0}-\delta ,x_{0});$ then we have $f^{'}_{-}(x_{0}) $ exists .In a similar way ,$f^{'}_{+}(x_{0}) $ exists. $f^{'}_{-}(x_{0}) $ exists $\Rightarrow\displaystyle\lim_{x\rightarrow x_{0}-}f(x)=f(x_{0});$ $f^{'}_{+}(x_{0}) $ exists $\Rightarrow\displaystyle\lim_{x\rightarrow x_{0}+}f(x)=f(x_{0}).$ Obviously, $f$ is coutinuous at $x=x_{0}.$Further, $f\in C((a,b))!$


But I failed to prove $g(x)$ is bound and decreasing (not strictly ) in $(x_{0}-\delta ,x_{0}).$Sometimes I doubt the conculsion that $f\in C((a,b)).$ Either make a counterexample to deny it ,or prove the conculsion is correct?


Elliot
  • 496
  • I believe if a function is measurable and convex on an open interval it is continuous. There are discontinuous convex functions, but they have very wild behavior. – RRL Sep 22 '14 at 03:47
  • The inequality for the slopes goes the other way for convex functions, you stated the one for concave functions. In either case, the conclusion is Yes, it is continuous (provided the domain is convex and $\it{open}$ ) – orangeskid Sep 22 '14 at 08:45

2 Answers2

5

Let $x < y < z$. The inequalities: \begin{eqnarray*} \frac{f(y)-f(x)}{y-x} \le \frac{f(z)- f(y)}{z-y} \\ \frac{f(y)-f(x)}{y-x} \le \frac{f(z)- f(x)}{z-x}\\ \frac{f(z)-f(x)}{z-x} \le \frac{f(z)- f(y)}{z-y} \end{eqnarray*} are equivalent.

A function $f$ is convex if for any $x<y<z$ in the domain any of the equivalent inequalities from above hold.

It is easy to see now that if $f$ is convex and $x<y$, $x'<y'$ and $x\le x'$, $y\le y'$ then \begin{eqnarray*} \frac{f(y)-f(x)}{y-x} \le \frac{f(y')- f(x')}{y'-x'} \end{eqnarray*} (the slopes are increasing). Indeed we have \begin{eqnarray*} \frac{f(y)-f(x)}{y-x} \le \frac{f(y')- f(x)}{y'-x}\le \frac{f(y')- f(x')}{y'-x'} \end{eqnarray*} Let $[c,d]$ a closed interval contained in $(a,b)$. Let $c'< c$ and $d'>d$ so that $[c,d]\subset[c',d']\subset (a,b)$. For any $x<y$ in $[a,b]$ we have \begin{eqnarray*} \frac{f(c)- f(c')}{c - c'} \le \frac{f(y)- f(x)}{y-x}\le \frac{f(d')- f(d)}{d'-d} \end{eqnarray*} Let $M = \max \{ | \frac{f(c)- f(c')}{c - c'} |, | \frac{f(d')- f(d)}{d'-d}|\}$. It follows that \begin{eqnarray*} |\frac{f(y)- f(x)}{y-x}|\le M \end{eqnarray*} for all $x<y$ in the interval $[c,d]$ so $f$ is Lipschitz on that interval.

Note that $M$ can improved to $\max\{ |f'_{-}(c)|, |f'_{+}(d)|\}$.

In any case, $f$ is Lipschitz on any closed interval and hence continuous.

$\bf{Added:}$ We can prove that a convex function defined on an open subset of $\mathbb{R}^n$ is Lipschitz on any compact subset, and so continuous. Let's sketch the argument for an open subset of $\mathbb{R}^2$. It is enough to show that is is Lipschitz on any square $S$ contained in the domain. Consider a slightly larger $S'\supset S$ also in the domain. On the boundary of $S'$ and $S$ the function is continuous, so bounded. Take two points $x$, $y$ in $S$. With the argument above we have $$\frac{|f(x)-f(y)|}{\|x-y\|} \le \frac{2A}{\delta}$$ where $A= \sup_{\partial S \cup \partial S'} |f|$ and $\delta = \operatorname{dist}(\partial S, \partial S')$.

A similar argument with induction on the dimension would show it for convex functions on open subsets of $\mathbb{R}^n$.

orangeskid
  • 56,630
  • Do the inequalities hold true because of the second-order condition on convex functions (i.e. "slopes are increasing")? If so, is this proof not restricted to functions with a differentiable first-order derivative? – Nurmister Sep 29 '18 at 09:58
  • @Nurmister: the slopes are of the secants. Now, a convex function on an open interval will have left and right derivatives, with the argument above. – orangeskid Feb 15 '22 at 22:36
3

You can prove $g(x)$ is decreasing over all of $(a,x_0)$. To do this, I'll use the lemma that if $\frac{a}b\ge \frac{c}d$, then $\frac{c+a}{d+b}\ge \frac{c}d$, whenever $a,b,c,d\ge0$ (this is easy to prove). Then, for any $x<y<x_0$, we have

$$ g(x)=\frac{f(x_0)-f(x)}{x_0-x}=\frac{f(x_0)-f(y)+f(y)-f(x)}{x_0-y+y-x}\ge\frac{f(x_0)-f(y)}{x_0-y}=g(y) $$ where the $\ge$ part follows from the lemma since $\frac{f(y)-f(x)}{y-x}\ge \frac{f(x_0)-f(y)}{x_0-y}$ by the convexity property. This proves $g$ is decreasing; you can also show $g$ is bounded below in this region, namely, for $x<x_0<x_2$, we always have $g(x)\ge \frac{f(x_2)-f(x_0)}{x_2-x_0}$, and then the rest of the proof goes jsut like you said.

Mike Earnest
  • 84,902
  • For completeness sake.... Could it not be the case that left and right limits exist but are not equal ? Also proving that f is continuous for an arbitrary $x_0$ in this open, does it necessarily imply that it is true for any $x$ in this open ? – zebullon Dec 12 '16 at 14:41
  • 1
    @zebullon I think you are confusing the limit of the difference quotient with that of the original function. The right and left limits of the difference quotient need not to be the equal as you say. But since they exists $f$ is right sided continuous and left sided continuous and therefore continuous. – mdcq Jan 22 '18 at 11:49