On approximation theory and functional equations (Q2549431): Difference between revisions

Let \(X\) and \(Y\) be compact spaces and let \(S = X\times Y\); \(C[S]\) is the space of continuous real-valued functions on \(S\), with the usual uniform norm. \(H\) is the subspace of \(C[S]\) of functions of the form \(f(x,y) = A(x) + B(y)\). For any compact set \(K\subset S\), \(H_K\) is the class of functions by restricting those in \(H\) to the set \(K\); \(H_K\) is a subspace of \(C[K]\). The main results in this paper provide conditions under which \(H_K\) coincides with or is dense in \(C[K]\), relating this to the existence of exact or approximate solutions of functional equations. In the simplest case, let \(\alpha\) and \(\beta\) be continuous mappings from \(X\) into \(Y\), and let \(K\) be the union of their graphs. Let \(\Gamma\) be the set of \(x\in X\) such that \(\alpha(x)=\beta(x)\). Theorem 2. \(H_K=C[K]\) if and only if the functional equation \[ \varphi(\alpha(x))-\varphi(\beta(x))=u(x)\tag{*} \] has a continuous solution \(\varphi\) for every choice of \(u(x)\in C[K]\) obeying the condition (N): \(u(x)=0\) for all \(x\in\Gamma\). Furthermore, \(H_K\) is dense in \(C[K]\) if and only if (*) admits arbitrarily good approximate solutions for each such choice of \(u\). This is generalized to systems. Given mappings \(\beta_0, \beta_1,\ldots, \beta_n\) from \(X\) into \(Y\), consider the system \[ \varphi(\beta_{i-1}(x))- \varphi(\beta_i(x))-u_i(x)=0,\quad i=1,2,\ldots,n.\tag{**} \] Given choices of the \(u_i\in C[K]\), one seeks either a function \(\varphi\in C[Y]\) such that all then equations (**) hold for all \(x\), or one seeks, for each \(\delta>0\), a function \(\varphi\) such that the left side of (**) is uniformly smaller than \(\delta\) for all \(x\in X\) and all \(i\). For each \(i<j\), let \(\Gamma_{ij}\) be the set of \(x\) for which \(\beta_i(x)=\beta_j(x)\). Theorem 1. The system (**) cannot have solutions in either sense unless the functions \(u_i\) obey conditions (N): \(u_k+u_{k+1}+\ldots+u_n\) vanishes on the set \(\Gamma_{k-1,m}\) for all \(k\) and \(m\). Furthermore, if conditions (N) hold, then (**) admits arbitrarily good approximate solutions \(\varphi\) if and only if \(H_K\) is dense in \(C[K]\) where \(K\) is the union of the graphs of the mappings \(\beta_i\), \(i=1,2,\ldots,n\). Previous work with the functional equation (*) has dealt with the special case \(\alpha(x)\equiv x\) [\textit{M. Kuczma}, Functional equations in a single variable. Warszawa: PWN - Polish Scientific Publishers (1968; Zbl 0196.16403)]. In this classical case, the author shows (Theorem 3) that if \(X=Y=[0,1]\) and \(\beta\) is monotonic increasing with \(\beta(0)=0\) and \(\beta(1)=1\) (this corrects a misprint in the present article), and the corresponding set \(\Gamma\) is finite, then (*) admits arbitrarily good approximate solutions for any choice of \(u\) which vanishes on \(\Gamma\). As an immediate corollary, any sequence \(\{\mu_k\}\) of moments on \([0,1]\) which obeys the relation \(\mu_n=\mu_{pn}\) for some integer \(p>1\) and all \(n\), must be constant.