A Sard theorem for graph theory

Abstract

The zero locus of a function f on a graph G is defined as the graph with vertex set consisting of all complete subgraphs of G, on which f changes sign and where x,y are connected if one is contained in the other. For d-graphs, finite simple graphs for which every unit sphere is a d-sphere, the zero locus of (f-c) is a (d-1)-graph for all c different from the range of f. If this Sard lemma is inductively applied to an ordered list functions f_1,...,f_k in which the functions are extended on the level surfaces, the set of critical values (c_1,...,c_k) for which F-c=0 is not a (d-k)-graph is a finite set. This discrete Sard result allows to construct explicit graphs triangulating a given algebraic set. We also look at a second setup: for a function F from the vertex set to R^k, we give conditions for which the simultaneous discrete algebraic set { F=c } defined as the set of simplices of dimension in {k, k+1,...,n} on which all f_i change sign, is a (d-k)-graph in the barycentric refinement of G. This maximal rank condition is adapted from the continuum and the graph { F=c } is a (n-k)-graph. While now, the critical values can have positive measure, we are closer to calculus: for k=2 for example, extrema of functions f under a constraint {g=c} happen at points, where the gradients of f and g are parallel D f = L D g, the Lagrange equations on the discrete network. As for an application, we illustrate eigenfunctions of geometric graphs and especially the second eigenvector of 3-spheres, which by Courant-Fiedler has exactly two nodal regions. The separating nodal surface of the second eigenfunction f_2 belonging to the smallest nonzero eigenvalue always appears to be a 2-sphere in experiments if G is a 3-sphere.