Midpoint-Convex Function is Rational Convex
Theorem
Let $I$ be a non-empty real interval.
Let $f: I \to \R$ be a real function.
If $f$ is midpoint-convex, then $f$ is rational-convex.
Proof
It suffices to show that for each $n \in \N$ and for any choice of $n$ elements $x_1, \dots, x_n \in I$:
\(\ds \map f {\frac {x_1 + \dots + x_n} n}\) | \(\leq\) | \(\ds \frac {\map f {x_1} + \dots + \map f {x_n} } n\) |
via Forward-Backward Induction.
Basis for the Induction
The statement holds:
- for $n = 0$ vacuously
- for $n = 1$ as $\map f {\dfrac x 1} = \dfrac {\map f x} 1$ for each $x \in I$.
Let $x_1, x_2$ be two points in $I$.
Then as $f$ is midpoint-convex:
\(\ds \map f {\frac {x_1 + x_2} {2^1} }\) | \(\le\) | \(\ds \frac {\map f {x_1} + \map f {x_2} } {2^1}\) |
This is the basis for the induction.
Induction Hypothesis
Suppose that if $n = 2^k$, then for any choice $x_1, \dots, x_n$ of $n$ elements in $I$, we have:
\(\ds \map f {\frac {x_1 + \dots + x_n} {2^k} }\) | \(\le\) | \(\ds \frac {\map f {x_1} + \dots + \map f {x_n} } {2^k}\) |
This is the induction hypothesis.
Induction Step
This is the induction step:
Let $n = 2^{k + 1}$.
Let $x_1, \dots, x_n \in I$.
Then:
\(\ds \map f {\frac {x_1 + \dots + x_{2^{k + 1} } } {2^{k + 1} } }\) | \(=\) | \(\ds \map f {\frac {\frac {x_1 + \dots + x_{2^k} } {2^k} + \frac {x_{2^k + 1} + \dots + x_{2^{k + 1} } } {2^k} } 2}\) | ||||||||||||
\(\ds \) | \(\le\) | \(\ds \frac {\map f {\frac {x_1 + \dots + x_{2^k} } {2^k} } + \map f {\frac {x_{2^k + 1} + \dots + x_{2^{k + 1} } } {2^k} } } 2\) | as $f$ is midpoint-convex | |||||||||||
\(\ds \) | \(\le\) | \(\ds \frac {\frac {\map f {x_1} + \dots + \map f {x_{2^k} } } {2^k} + \frac {\map f {x_{2^k + 1} } + \dots + \map f {x_{2^{k + 1} } } } {2^k} } 2\) | Induction Hypothesis | |||||||||||
\(\ds \) | \(=\) | \(\ds \frac {\map f {x_1} + \dots + \map f {x_{2^{k + 1} } } } {2^{k + 1} }\) |
which completes the Proof by Mathematical Induction for integers of the form $2^k$.
Backward Step
Let $2^k \le n \le 2^{k + 1}$ for some integer $k$.
Let $x_1, \dots, x_n \in I$.
Defined $\overline x := \dfrac {x_1 + \dots + x_n} n$.
Then:
\(\ds \map f {\overline x}\) | \(=\) | \(\ds \map f {\frac {x_1 + \dots + x_n} n}\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \map f {\frac {n \overline x + \paren {2^k - n} \overline x} {2^k} }\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \map f {\frac {x_1 + \dots + x_n + \paren {2^k - n} \overline x} {2^k} }\) | ||||||||||||
\(\ds \) | \(\le\) | \(\ds \frac {\map f {x_1} + \dots + \map f {x_n} + \paren {2^k - n} \map f {\overline x} } {2^k}\) | from above |
from which we obtain:
\(\ds \map f {\overline x}\) | \(=\) | \(\ds \map f {\frac {x_1 + \dots + x_n} n}\) | ||||||||||||
\(\ds \) | \(\le\) | \(\ds \frac {\map f {x_1} + \dots + \map f {x_n} } n\) |
completing the proof.
$\blacksquare$