Some characterization of best approximants in normed linear spaces (Q5954629)

The author considers the classical problem of characterization for the best approximation elements from a linear subspace~\(G\) of a normed linear space \((X,\|{\cdot}\|)\). For \(x,y\in X\) let \((x,y)_{i(s)}=\lim_{t\to 0^{-(+)}} (\|y+txt\|^2-\|y\|)/2t\) be the inferior (superior) semi-inner product. Let \(x_0\notin G\), \(g_0\in G\), \(G_{x_0}:=\operatorname{lin}(G,x_0)\), \(s_0=(x_0-g_0)/\|x_0-g_0\|\). Then the following statements are equivalent: (i) \(g_0\) is the best approximation from~\(G\) to~\(x_0\); (ii) \(\|f\|_{G_{x_0}}(x,\operatorname{sign}f(x_0)s_0)_i \leqslant f(x)\leqslant \|f\|_{G_{x_0}} (x,\operatorname{sign}f(x_0)s_0)_s\) for every \(f\in G_{x_0}^*\) with \(\text{Ker}f=G\), and every \(x\in G_{x_0}\); (iii) for every \(f\in G_{x_0}^*\) with \(\text{Ker}f=G\), the element \(u_0=s_0 \operatorname{sign}f(x_0)\) minimises the quadratic functional \(F_f(x)=\|x\|^2-2f(x)\), \(x\in G_{x_0}\).