2
$\begingroup$

Let $X$ denote the non-negative "orthant" of the Banach space $L^2$ (or whatever you call the set of functions in $L^2$ that are non-negative), and let $C$ be a closed, convex subset of $X$. Let $f$ be a strictly convex functional defined on $X$ and suppose that the problem $$\text{minimize}~ f(x)~~~s.t.~~~x\in C$$admits a unique solution $x_0\in C$. Further, suppose that there is also a unique "tangent plane" to $C$ at $x_0$, i.e., a unique functional $y_0 \in L^2$ such that $\left\langle x_{0},y_{0}\right\rangle = q$ and $\left\langle x,y_{0}\right\rangle\leq q$ for all other $x\in C$. Is it also true that $x_0$ minimizes the relaxed problem $$\text{minimize}~ f(x)~~~s.t.~~~\left\langle x,y_{0}\right\rangle\leq q~?$$ Could I make my problem more tractable by considering a different space of functions instead of $L^2$, e.g. continuous functions or $L^\infty$, for example?

$\endgroup$
1
$\begingroup$

Let us assume that your $f$ is at least directionally differentiable at $x_0$ and that the directional derivative depends continuously on the direction (this may be satisfied under rather general assumptions, e.g. if $f$ is locally Lipschitz).

Then, the first order (necessary and - due to convexity - sufficient) optimality conditions of your first problem read $$f'(x_0;d) \ge 0 \quad\forall d \in T_C(x_0),$$ where $T_C(x_0)$ is the tangent cone of $C$ in $x_0$. Since your functional $y_0$ is unique, we find $T_C(x_0) = y_0^\circ$, where $y_0^\circ = \{d : \langle d, y_0\rangle \le 0\}$ is the polar cone of $y_0$. This yields the optimality conditions $$f'(x_0;d) \ge 0 \quad\forall d \in y_0^\circ,$$ but these are precisely the first order (nec. and suff.) conditions of the second problem.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.