19.1 Convexity

Definition 19.1.1 (Epigraph).label Let $E$ be a vector space over $\real$ and $f: E \to (-\infty, \infty]$, then the epigraph of $f$ is the set

\[\text{epi}(f) = \bracs{(x, y) \in E \times \real| y \ge f(x)}\]

Definition 19.1.2 (Convex Function).label Let $E$ be a vector space over $\real$ and $f: E \to (-\infty, \infty]$, then the following are equivalent:

  1. (1)

    For every $x, y \in E$ and $t \in [0, 1]$,

    \[f((1 - t)x + ty) \le (1 - t)f(x) + tf(y)\]

  2. (2)

    $\text{epi}(f)$ is convex.

If the above holds, then $f$ is convex.

Proof. (1) $\Rightarrow$ (2): Let $(x, \alpha), (y, \beta) \in \text{epi}(f)$ and $t \in [0, 1]$, then

\begin{align*}(1 - t)\alpha + t\beta&\ge (1 - t)f(x) + tf(y) \\&\ge f((1 - t)x + ty)\end{align*}

so $((1 - t)x + ty, (1 - t)\alpha + t\beta) \in \text{epi}(f)$.

(2) $\Rightarrow$ (1): Let $x, y \in E$ and $t \in [0, 1]$, then

\[((1 - t)x + ty, (1 - t)f(x) + tf(y)) \in \text{epi}(f)\]

so $f((1 - t)x + ty) \le (1 - t)f(x) + tf(y)$.$\square$

Lemma 19.1.3.label Let $E$ be a vector space over $\real$ and $f: E \to (-\infty, \infty]$, then $f$ is convex if and only if $\bracs{f < \infty}$ is convex and $f|_{\bracs{f < \infty}}$ is a convex function.

Proof. If $f$ is convex, then for any $x, y \in \bracs{f < \infty}$ and $t \in [0, 1]$,

\[f((1 - t)x + ty) \le (1 - t)f(x) + tf(y) < \infty\]

so $\bracs{f < \infty}$ is convex, and the restriction $f|_{\bracs{f < \infty}}$ is a convex function.

On the other hand, for any $x, y \in E$ and $t \in [0, 1]$, if $f(x) = \infty$ or $f(y) = \infty$, then

\[f((1 - t)x + ty) \le = \infty (1 - t)f(x) + tf(y)\]

Otherwise, $x, y \in \bracs{f < \infty}$, and

\[f((1 - t)x + ty) \le (1 - t)f(x) + tf(y)\]

by convexity of $f|_{\bracs{f < \infty}}$.$\square$

Lemma 19.1.4.label Let $E$ be a vector space over $\real$ and $f: E \to (-\infty, \infty]$ be convex, then for any $x, y \in E$ and $t \in \real \setminus [0, 1]$,

\[f((1 - t)x + ty) \ge (1 - t)f(x) + tf(y)\]

Proof. Via an affine transformation, assume without loss of generality that $x = 0$ and $f(x) = 0$. By exchanging $x$ and $y$, assume without loss of generality that $t > 1$. In which case, since $f$ is convex and $y = t^{-1}ty$, $f(y) \le t^{-1}f(ty)$ and $tf(y) \le f(ty)$.$\square$

Proposition 19.1.5.label Let $E$ be a vector space over $\real$, $f: E \to (-\infty, \infty]$ be convex, and $x \in \bracs{f < \infty}$, then for any $h \in E$,

\[\lim_{t \downto 0}\frac{f(x + th) - f(x)}{t}= \inf_{t > 0}\frac{f(x + th) - f(x)}{t}\]

exists in $[-\infty, \infty]$.

Proof. Let $0 < s \le t$, then since $f$ is convex,

\begin{align*}f(x + sh)&\le \paren{1 - \frac{s}{t}}f(x) + \frac{s}{t}f(x + th) \\ f(x + sh) - f(x)&\le \frac{s}{t}[f(x + th) - f(x)] \\ \frac{f(x + sh) - f(x)}{s}&\le \frac{f(x + th) - f(x)}{t}\end{align*}

$\square$

Proposition 19.1.6.label Let $E$ be a vector space over $\real$, then:

  1. (1)

    For any $f, g: E \to (-\infty, \infty]$ convex and $\lambda \ge 0$, $\lambda f + g$ is convex.

  2. (2)

    For any convex functions $\cf \subset (-\infty, \infty]^{E}$, $\sup_{f \in \cf}f$ is convex.

Post a Comment

Name:Email:
Please enter the tag of the current page (11V) to post the comment.
Tag: