[proofplan]
The point $b$ lies between $a$ and $c$, so it can be written as a convex combination of $a$ and $c$. Applying convexity of $f$ to this representation gives a single inequality relating $f(a)$, $f(b)$, and $f(c)$. Rearranging that inequality in two different ways yields the two desired comparisons of secant slopes.
[/proofplan]
custom_env
admin
[step:Express $b$ as a convex combination of $a$ and $c$]
Define
\begin{align*}
t := \frac{b-a}{c-a}.
\end{align*}
Since $a<b<c$, we have $c-a>0$ and $0<t<1$. Moreover,
\begin{align*}
(1-t)a+tc=b.
\end{align*}
Because $I$ is an interval and $a,c\in I$, every point between $a$ and $c$ lies in $I$; in particular, this convex combination is an element of $I$.
[/step]
custom_env
admin
[step:Apply convexity at the point $b$]Since $f:I\to\mathbb{R}$ is convex, $a,c\in I$, and $t\in[0,1]$, convexity gives
\begin{align*}
f((1-t)a+tc)\le (1-t)f(a)+t f(c).
\end{align*}
Using $(1-t)a+tc=b$ and substituting the value of $t$, this becomes
\begin{align*}
f(b)\le \frac{c-b}{c-a}f(a)+\frac{b-a}{c-a}f(c).
\end{align*}[/step]
custom_env
admin
[guided]The key point is that convexity controls the value of $f$ at an interior point by the corresponding linear interpolation of the endpoint values. We have already written $b$ in the form
\begin{align*}
b=(1-t)a+tc
\end{align*}
with
\begin{align*}
t=\frac{b-a}{c-a}.
\end{align*}
Because $a<b<c$, the denominator $c-a$ is positive and the numerator $b-a$ is strictly between $0$ and $c-a$, so $0<t<1$. Thus the convexity inequality for the map $f:I\to\mathbb{R}$ applies to the points $a,c\in I$ and the parameter $t\in[0,1]$.
Convexity gives
\begin{align*}
f((1-t)a+tc)\le (1-t)f(a)+t f(c).
\end{align*}
Since $(1-t)a+tc=b$, the left-hand side is $f(b)$. Also,
\begin{align*}
1-t=1-\frac{b-a}{c-a}=\frac{c-b}{c-a}.
\end{align*}
Substituting these identities into the convexity inequality gives
\begin{align*}
f(b)\le \frac{c-b}{c-a}f(a)+\frac{b-a}{c-a}f(c).
\end{align*}
This is the single inequality from which both secant slope comparisons will be extracted.[/guided]
custom_env
admin
[step:Rearrange the convexity inequality to compare $S_f(a,b)$ and $S_f(a,c)$]
From
\begin{align*}
f(b)\le \frac{c-b}{c-a}f(a)+\frac{b-a}{c-a}f(c),
\end{align*}
subtract $f(a)$ from both sides:
\begin{align*}
f(b)-f(a)\le \frac{b-a}{c-a}(f(c)-f(a)).
\end{align*}
Since $b-a>0$, division by $b-a$ preserves the inequality and gives
\begin{align*}
\frac{f(b)-f(a)}{b-a}\le \frac{f(c)-f(a)}{c-a}.
\end{align*}
By the definition of the secant slope, this is
\begin{align*}
S_f(a,b)\le S_f(a,c).
\end{align*}
[/step]
custom_env
admin
[step:Rearrange the same inequality to compare $S_f(a,c)$ and $S_f(b,c)$]
Again starting from
\begin{align*}
f(b)\le \frac{c-b}{c-a}f(a)+\frac{b-a}{c-a}f(c),
\end{align*}
subtract $f(c)$ from both sides:
\begin{align*}
f(b)-f(c)\le \frac{c-b}{c-a}(f(a)-f(c)).
\end{align*}
Multiplying by $-1$ reverses the inequality:
\begin{align*}
f(c)-f(b)\ge \frac{c-b}{c-a}(f(c)-f(a)).
\end{align*}
Since $c-b>0$, division by $c-b$ preserves the inequality and gives
\begin{align*}
\frac{f(c)-f(b)}{c-b}\ge \frac{f(c)-f(a)}{c-a}.
\end{align*}
Equivalently,
\begin{align*}
S_f(a,c)\le S_f(b,c).
\end{align*}
Combining this inequality with $S_f(a,b)\le S_f(a,c)$ proves
\begin{align*}
S_f(a,b)\le S_f(a,c)\le S_f(b,c).
\end{align*}
[/step]