Measuring uncertainty without a norm (Q798906)

Let F and G be given sets and \(2^ G\) be the class of all subsets of G. By a solution operator one means an operator \(S:F\times R^+\to 2^ G\) such that, for all f in F S(f,0)\(\neq\emptyset \) and for all f in F and all \(\epsilon_ 1\epsilon_ 2\) in \(R^+\) with \(\epsilon_ 1\leq\epsilon_ 2 S(f,\epsilon_ 1)\subseteq S(f,\epsilon_ 2)\). The set S(f,\(\epsilon)\) consists of the \(\epsilon\) -approximations to the exact solution S(f,0). The author shows that if \(S:F\times R^+\to 2^ G\) is a solution operator then for any \(\epsilon_ 0>0\) then exist an operator \(\bar S:F\to G\) and a family of pseudometrics \(D=\{d_ f| f\in F\}\) on G such that S(f,\(\epsilon)\subseteq B(\bar Sf,d_ f,\epsilon)\subseteq S(f,\epsilon ')\) for all \(f\in F\) and any \(\epsilon\in [0,\epsilon_ 0)\) and \(\epsilon '\in (\epsilon_ 1,\epsilon_ 0]\) where \(B(\bar Sf,d_ f,\epsilon)=\{x\in G| d_ f(x,\bar Sf)\leq\epsilon \}.\) Moreover, he shows that if \(\bar S:F\to R\) is defined by letting \(\bar Sf\) be the zero of f that is smallest in magnitude and a family of pseudometrics \(d_ f:R\times R\to R^+\) (\(f\in F)\) defined by \(d_ f(x,y)=| f(x)- f(y)|,\) then \(S(f,\epsilon)=B(\bar Sf,d_ f,\epsilon)\) for all \(f\in F\) and \(\epsilon \geq 0\). Further, the author shows that there do not exist an operator \(\bar S:F\to R\) and a single pseudometric d on R such that \(S(f,\epsilon)=B(\bar Sf,d,\epsilon)\) for all \(f\in F\) and \(\epsilon \geq 0\).