A Cheatsheet on Prametric Optimization
Table of Contents
Continuity Results
Let \( X:T\rightrightarrows \mathbb{R}^n \) be lower semicontinuous and let
\( f:\text{graph}(X)\rightarrow\mathbb{R} \) be lower semicontinuous. Then,
the extended real optimal-value function
$$
\varphi(t) := \sup_{ x\in X(t) } \ f(t,x)
$$
is lower semicontinuous.
\textcite[Lemma~17.29]{aliprantis2006}.
Let \( X:T\rightrightarrows \mathbb{R}^n \) be lower semicontinuous and let
\( f:\text{graph}(X)\rightarrow\mathbb{R} \) be upper semicontinuous. Then,
the extended real optimal-value function $$ \varphi(t) := \inf_{x\in X(t)} \
f(t,x) $$ is upper semicontinuous.
\textcite[Lemma~17.29]{aliprantis2006}.
Let \( X:T\rightrightarrows \mathbb{R}^n \) be upper semicontinuous with
nonempty compact values and let \( f:\text{graph}(X)\rightarrow\mathbb{R} \)
be upper semicontinuous. Then, the extended real optimal-value function $$
\varphi(t) := \max_{ x\in X(t) } \ f(t,x) $$ is upper semicontinuous.
\textcite[Lemma~17.30]{aliprantis2006}.
Let \( X:T\rightrightarrows \mathbb{R}^n \) be upper semicontinuous with
nonempty compact values and let \( f:\text{graph}(X)\rightarrow\mathbb{R} \) be
lower semicontinuous. Then, the extended real optimal-value function $$
\varphi(t) := \min_{x\in X(t)} \ f(t,x) $$ is lower semicontinuous.
\textcite[Lemma~17.30]{aliprantis2006}.
(Berge Maximum Theorem)
Let \( X:T\rightrightarrows \mathbb{R}^n \) be continuous with nonempty
compact values and let \( f:\text{graph}(X)\rightarrow\mathbb{R} \) be
continuous. Then, the optimal-value function $$ \varphi(t) := \max_{ x\in
X(t) } \ f(t,x) $$ is continuous. Moreover, the solution set mapping is
upper semicontinuous.
\textcite[Theorem~17.31]{aliprantis2006}.
(Convex Optimization - Lower Semicontinuity)
Let \( T\subseteq\mathbb{R}^m \) be a given set and let \( S(t) \) denote
the solution set of the convex problem
$$
\begin{align*}
\min_{x} \quad & f(x) \\
\text{s.t.} \quad & g_i(x) \le t_i, \quad \text{for all }i = 1,\dotsc,m.
\end{align*}
$$
Assume that \( f \) and \( g_i \) are convex and weakly analytic for all \(
i=1,\dotsc,m \) and that \( S(t) \neq \emptyset \) holds for all \( t\in T
\). Then, the solution set mapping \( S \) is lower semicontinuous on \( T
\).
\textcite[Theorem~4.3.5]{Bank_1982}.
References
@book{Bank_1982,
title = {Non-Linear Parametric Optimization},
isbn = {9783034863285},
doi = {10.1007/978-3-0348-6328-5},
publisher = {Birkhäuser Basel},
author = {Bank, B. and Guddat, J. and Klatte, D. and Kummer, B. and Tammer, K.},
year = {1982}
}
@book{aliprantis2006,
title={Infinite dimensional analysis: a hitchhiker's guide},
author={Aliprantis, Charalambos D and Border, Kim C},
year={2006},
publisher={Springer},
doi={10.1007/3-540-29587-9}
}