I am attempting problem 3.9 in Brezis' book:
Let $E$ be a Banach space; let $M\subseteq E$ be a linear subspace, and let $f_0\in E^\star$. Prove that there exists some $g_0\in M^\perp$ such that $$ \inf_{g\in M^\perp}\{\Vert f_0-g\Vert \}=\Vert f_0-g_0\Vert $$ I have attempted the following solution:
Let $\varphi : E^\star\to \mathbb{R}:g\mapsto \Vert f_0 - g\Vert$. Note that $\varphi$ is weak$^*$ly l.s.c. In addition, $M^\perp$ is always closed. Let $\varepsilon > 0$ and $$ r_0\colon=\inf_{\overline{B_{E^\star}(f_0, r)}\cap M^\perp\neq \varnothing}\{r>0\} $$ $$ B\colon= \overline{B_{E^\star}(f_0, r_0+\varepsilon)} $$ Then $B\cap M^\perp$ is a closed subspace of $B$ and therefore compact in the weak$^*$ topology. Thus, by lower semicontinuity, $\varphi$ attains its infimum on $B\cap M^\perp$.
I am stuck on trying to prove that $\varphi$ attaining its infimum on $B$ for every $\varepsilon$ implies that the global infimum is attained. Specifically, is it true that the global infimum of $\varphi$ lies in $B$ for every $\varepsilon$? Any hints or suggestions? Thanks!