Is my proof for $f$ is convex iff $f'$ is monotonically increasing correct?

Question

I am following up on my previous question. My previous attempt for the proof was wildly incorrect (my question was how that proof was exactly my old proof was incorrect) and I have now come up with a new proof.

I have to prove:

Let $f:(a, b) \to R^1$ be differentiable. Prove that $f$ is convex iff $f'$ is monotonically increasing.

What I have for the proof:

($\Rightarrow$) Assume $f$ is convex in $(a, b)$. Let $a<s<t<u<b$. By Exercise 23 in Chapter 4, \begin{align}\tag{14.1} \frac{f(t)-f(s)}{t-s} \le \frac{f(u)-f(s)}{u-s} \le \frac{f(u)-f(t)}{u-t} \end{align} Since $f$ is differentiable on $(a,b)$, both $ f'(s) = \lim_{t \to s} \frac{f(t)-f(s)}{t-s}$ and $f'(t)=\lim_{u \to t} \frac{f(u)-f(t)}{u-t}$ exist. However, applying the Order Limit Theorem on (14.1) gives \begin{align*} \lim_{t \to s} \frac{f(t)-f(s)}{t-s} \le \lim_{u \to t} \frac{f(u)-f(t)}{u-t} \implies f'(s) \le f'(t) \end{align*} which shows that $f'$ is monotonically increasing in $(a, b)$.

($\Leftarrow$) Assume $f'$ is monotonically increasing in $(a, b)$ and $a<x<y<b$. Fix $0 < \lambda< 1$. By Exercise 23 in Chapter 4, we must show that \begin{equation}\tag{14.0} f(\lambda x + (1- \lambda)y) \le \lambda f(x) + (1-\lambda) f(y) \end{equation} Denote $z=\lambda x+ (1-\lambda)y$.Then, $z=\lambda(x-y)+y$ which implies that $\lambda=\frac{z-y}{x-y}$. Since $\lambda>0, z-y>x-y \implies z>x$. Also, $1-\lambda=\frac{x-y-z+y}{x-y} = \frac{x-z}{x-y}$. Since $\lambda<1, x-z>x-y \implies z < y$. Thus, $x<z<y$. Then, (14.0) can be simplified as: \begin{align*} f(z) &\le f(y) + \lambda f(x) - \lambda f(y) \\ \lambda f(z) - \lambda f(x) &\le f(y) - f(z) - \lambda f(y) + \lambda f(z) \\ \lambda[f(z)-f(x)] &\le (1-\lambda)[f(y)-f(z)] \end{align*} Thus, since $\lambda = \frac{y-z}{y-x}$ and $1-\lambda = \frac{z-x}{y-x}$, it suffices to show that \begin{equation}\tag{14.2} \frac{f(z)-f(x)}{z-x} \le \frac{f(y)-f(z)}{y-z} \end{equation} Now, as we take $z\to x$ on the left of (14.2) and $y\to z$ on the right of (14.2), then we have $f'(x)\le f'(z)$, which holds since $x<z$ and $f'$ is monotonically increasing.

Exercise 23 in Chapter 4 in Rudin:

A real-valued function $f$ defined in $(a, b)$ is said to be convex if $$ f \left( \lambda x + (1- \lambda) y \right) \leq \lambda f(x) + (1-\lambda) f(y)$$ whenever $a < x < b$, $a < y < b$, $0 < \lambda < 1$. Prove that every convex function is continuous.

Hint: If $f$ is convex in $(a, b)$ and if $a < s < t < u < b$, show that $$ \frac{ f(t)-f(s)}{t-s} \leq \frac{ f(u)-f(s)}{u-s} \leq \frac{ f(u)-f(t)}{u-t}.$$

Can someone please read over my proof and see if there is something that I did incorrectly? Also, specifically, is my usage of the Order Limit Theorem correct and is the argument right below (14.2) correct?

Answer 1

Hint: (reverse implication)

If $s<t<u$, then by the mean value theorem there exists $\xi_1 \in (s,t)$ and $\xi_2 \in (t,u)$ such that (since $f'$ is monotonically increasing) $$\frac{f(t)-f(s)}{t-s} = f'(\xi_1) \leqslant f'(\xi_2) = \frac{f(u)-f(t)}{u-t}$$

---

Forward Implication

By convexity, for $s < t < u$, we have

$$\frac{f(t)-f(s)}{t-s} \leqslant \frac{f(u)-f(s)}{u-s} \leqslant \frac{f(u)-f(t)}{u-t}$$

Thus,

$$f'(s) = \lim_{t \to s+}\frac{f(t)-f(s)}{t-s} \leqslant\lim_{t \to s+}\frac{f(u)-f(t)}{u-t} = \frac{f(u)-f(s)}{u-s}, $$

and

$$\frac{f(u)-f(s)}{u-s} = \lim_{t \to u-}\frac{f(t)-f(s)}{t-s} \leqslant \lim_{t \to u-} \frac{f(u)-f(t)}{u-t} = f'(u)$$

Therfore, $f'(u) \geqslant f'(s)$ when $u > s$ and $f$ is monotonically increasing.

Answer 2

Monotonic increasing means that the function $f(x)$ cannot decrease with increasing $x$, i.e. $f''(x)\geq 0$.

Define $h=y-x$.To show a continuous convex function is monotonically increasing:

$$0\leq\lim_{y\to x}\lambda f(x)+(1-\lambda)f(y)-f(\lambda x+(1-\lambda)y)=\lim_{h\to 0}\lambda f(x)+(1-\lambda)f(x+h)-f(x+(1-\lambda)h)=\lim_{h\to 0}f(x)+(1-\lambda)f'(x)h+\frac{1}{2}(1-\lambda)f''(x)h^2+o(h^3)-f(x)-f'(x)(1-\lambda)h-\frac{1}{2}f''(x)(1-\lambda)^2h^2+o(h^3)=\lim_{h\to 0}\frac{1}{2}f''(x)h^2\lambda(1-\lambda)+o(h^3).$$

To prove the reverse direction, just read the equations in the opposite direction.