Dimension, superposition of functions and separation of points, in compact metric spaces (Q1076367): Difference between revisions

The work of \textit{A. N. Kolmogorov} [Dokl. Akad. Nauk SSSR 114, 953-956 (1957; Zbl 0090.271); Am. Math. Soc., Transl. II. Ser. 28, 55-61 (1963)], \textit{P. A. Ostrand} [Bull. Am. Math. Soc. 71, 619-622 (1965; Zbl 0134.418)] and \textit{M. Katětov} [Čas. Pěst. Mat. Fys. 75, 1-16 (1950; Zbl 0037.354)] are thoroughly and profoundly generalized in a unique method. Spaces are compact metric. Let \(F=\{\phi_ i:\) \(i=1,...,k\}\) be a family of \(\phi_ i\in C(X,Y_ i)\). F is said to be basic if each \(f\in C(X)\) admits a representation \(f(x)=\sum^{k}_{i=1}g_ i(\phi_ i(x))\), \(x\in X\), with \(g_ i\in C(Y_ i)\). When X and \(Y_ i\) are merely sets, F is said to be a uniformly separating family if there exists a constant \(0<\lambda \leq 1\) such that for each \(\mu \in \ell_ 1(X)\), \(\| \mu \circ \phi \| \geq \lambda \| \mu \|\) holds for some \(\phi\in F\). These two concepts are important in the argument. The following three theorems are equivalent. (1) Let \(n\geq 1\). Then dim \(X\leq n\) if and only if there exists a basic family \(F\subset C(X)\) with \(| F| \leq 2n+1\). (2) Let \(n\geq 1\). Then dim \(X\leq n\) if and only if there exists an embedding \(\phi\) : \(X\to R^{2n+1}\) satisfying the condition; any \(f\in C(\phi (X))\) can be represented as \(f(t_ 1,...,t_{2n+1})=\sum^{2n+1}_{i=1}g_ i(t_ i)\), \((t_ 1,...,t_{2n+1})\in \phi (X)\), \(g_ i\in C(R)\). (3) Let \(n\geq 1\). Then dim \(X\leq n\) if and only if C(X) is the (algebraic) sum of \(2n+1\) subalgebras, each of which contains the constants and is generated by one element. A stronger version of (3) is given as follows. (4) Let \(n\geq 1\). Then dim \(X\leq n\) if and only if there exist subalgebras \(A_ i\), \(1\leq i\leq 2n+1\), and \(B_ j\), \(1\leq j\leq n\), of C(X), which contain the constants, such that the \(A_ i's\) are generated by one element, and \(B_ j's\) are analytically generated by one element, so that for each \(0\leq k\leq n\), the algebraic sum of any k of the \(B_ j's\) and any \(2(n- k)+1\) of the \(A_ i's\) is C(X). A stronger version of (1) is also given as follows. (5) If dim X\(=n\geq 2\), and \(F\subset C(X)\) is a uniformly separating family, then \(| F| \geq 2n+1\).