A geometrical interpretation of the without-exception feasibility of PL homotopy methods (Q750942)

This is a well-written exposition of a talk on PL-homotopy methods which shows a nice geometric flavour: Let f: \({\mathbb{R}}^ n\to {\mathbb{R}}^ n\) be continuous. A PL-method for finding the zeros of f consists of a homotopy H: \({\mathbb{R}}^ n\times [0,1]\to {\mathbb{R}}^ n\) connecting \(f=H(\cdot,0)\) to an auxiliary mapping \(g:=H(\cdot,1)\), a triangulation G of \({\mathbb{R}}^ n\times [0,1]\), and a PL-approximation L: \({\mathbb{R}}^ n\times [0,1]\to {\mathbb{R}}^ n\) to H. Let \(J(\epsilon):=\{(s,...,s^ n)|\) \(0<s<\epsilon \}\). An n-simplex \(<v_ 0,...,v_ n>\) in \({\mathbb{R}}^ n\times [0,1]\) is said to be complete if J(\(\epsilon\))\(\subset <H(v_ 0),...,H(v_ n)>\) for some \(\epsilon >0\). An \((n+1)\)-simplex is complete if one of its facets is a complete n-simplex. It turns out that this implies that there are precisely two facets which are complete n- simplices. Let \(\tau^ k=<v_ 0,...,v_ n>\) be a complete n-simplex which is a facet of the complete \((n+1)\)-simplex \(\sigma^{k+1}\). Let v be the vertex of \(\sigma^{k+1}\) opposite to \(\tau^ k\). Then there is a unique \(i\in \{1,..,n\}\) such that \(\tau^{k+1}=<v_ 0,...,v_{i- 1},v,v_{i+1},...,v_ n>\) is complete. This is complementary pivoting. The author proves that in this situation there is an \(\epsilon >0\) such that \(L^{-1}(J(\epsilon))\cap [\sigma^{k+1}]\) is a nondegenerate piece of a ruled surface, the ruled surface intersects the boundary of \(\sigma^{k+1}\) in exactly two facets (namely \(\tau^ k\) and \(\tau^{k+1})\) and these intersections are nonsingular affine images of some J(\(\epsilon\)). This gives a nice geometric interpretation of why one does not have to worry about nondegeneracy conditions when using vector labelling and complementary pivoting.