Counterexample to a problem on tensor product approximation (Q1270280)

Let \(K\) be a compact set on the plane. A sequence of distinct points \(P_1,\dots,P_{2k}\) is called a closed twisting path when consecutive points can be connected with alternatively horizontal and vertical segments. When \(K\) contains a closed twisting path then there are functions \(F(x,y)\) on \(K\) which cannot be approximated by sums \(g(x)+h(y)\) of functions of a single variable only. On the other hand when \(K\) does not contain a closed twisting path one can easily represent each function on \(K\) in the form \(F(x,y) = g(x)+h(y)\). M. v. Golitschek conjectured that every \textit{continuous} function \(F\) on such a set \(K\) can be uniformly approximated by sums \(g(x)+h(y)\) of \textit{continuous} functions \(g\) and \(h\). In the present paper a set \(K\) is constructed for which this conjecture is not valid.