arXiv++ Combinatorics

Higher dimensional graphs can be used to colour two-dimensional geometric graphs. If G the boundary of a three dimensional graph H for example, we can refine the interior until it is colourable with 4 colours. The later goal is achieved if all interior edge degrees are even. Using a refinement process which cuts the interior along surfaces we can adapt the degrees along the boundary of that surface. More efficient is a self-cobordism of G with itself with a host graph discretizing the product of G with an interval. It follows from the fact that Euler curvature is zero everywhere for three dimensional geometric graphs, that the odd degree edge set O is a cycle and so a boundary if H is simply connected. A reduction to minimal colouring would imply the four colour theorem. The method is expected to give a reason "why 4 colours suffice" and suggests that every two dimensional geometric graph of arbitrary degree and orientation can be coloured by 5 colours: since the projective plane can not be a boundary of a 3-dimensional graph and because for higher genus surfaces, the interior H is not simply connected, we need in general to embed a surface into a 4-dimensional simply connected graph in order to colour it. This explains the appearance of the chromatic number 5 for higher degree or non-orientable situations, a number we believe to be the upper limit. For every surface type, we construct examples with chromatic number 3,4 or 5, where the construction of surfaces with chromatic number 5 is based on a method of Fisk. We have implemented and illustrated all the topological aspects described in this paper on a computer. So far we still need human guidance or simulated annealing to do the refinements in the higher dimensional host graph.

Given a finite simple graph G=(V,E) with chromatic number c and chromatic polynomial C(x). Every vertex graph coloring f of G defines an index i_f(x) satisfying the Poincare-Hopf theorem sum_x i_f(x)=chi(G). As a variant to the index expectation result we prove that E[i_f(x)] is equal to curvature K(x) satisfying Gauss-Bonnet sum_x K(x) = χ(G), where the expectation is the average over the finite probability space containing the C(c) possible colorings with c colors, for which each coloring has the same probability.

We discuss some linear algebra related to the Dirac matrix D of a finite simple graph G=(V,E).

For any finite simple graph G=(V,E), the discrete Dirac operator D=d+d* and the Laplace-Beltrami operator L=d d* + d* d on the exterior algebra bundle Omega are finite v times v matrices, where dim(Omega) = v is the sum of the cardinalities v(k) of the set G(k) of complete subgraphs K(k) of G. We prove the McKean-Singer formula chi(G) = str(exp(-t L)) which holds for any complex time t, where chi(G) = str(1)= sum (-1)k v(k) is the Euler characteristic of G. The super trace of the heat kernel interpolates so the Euler-Poincare formula for t=0 with the Hodge theorem in the real limit t going to infinity. More generally, for any continuous complex valued function f satisfying f(0)=0, one has the formula chi(G) = str(exp(f(D))). This includes for example the Schroedinger evolutions chi(G) = str(cos(t D)) on the graph. After stating some general facts about the spectrum of D which includes statements about the complexity, the product of the non-zero eigenvalues as well as a perturbation result estimating the spectral difference of two graphs, we mention as a combinatorial consequence that the spectrum of D encodes the number of closed paths in the simplex space of a graph. McKean-Singer implies that the number of closed paths of length n starting at an even dimensional simplex is the same than the number of closed paths of length n starting at an odd dimensional simplex. We give a couple of worked out examples and see that McKean-Singer allows to find explicit pairs of non-isometric graphs which have isospectral Dirac operators.

The inductive dimension dim(G) of a finite undirected graph G=(V,E) is a rational number defined inductively as 1 plus the arithmetic mean of the dimensions of the unit spheres dim(S(x)) at vertices x primed by the requirement that the empty graph has dimension -1. We look at the distribution of the random variable "dim" on the Erdos-Renyi probability space G(n,p), where each of the n(n-1)/2 edges appears independently with probability p. We show here that the average dimension E[dim] is a computable polynomial of degree n(n-1)/2 in p. The explicit formulas allow experimentally to explore limiting laws for the dimension of large graphs. We also study the expectation E[X] of the Euler characteristic X, considered as a random variable on G(n,p). We look experimentally at the statistics of curvature K(v) and local dimension dim(v) = 1+dim(S(v)) which satisfy the Gauss-Bonnet formula X(G) = sum K(v) and by definition dim(G) = sum dim(v)/|V|. We also look at the signature functions f(p)=E[dim], g(p)=E[X] and matrix values functions A(p) = Cov[{dim(v),dim(w)], B(p) = Cov[K(v),K(w)] on the probability space G(p) of all subgraphs of a host graph G=(V,E) with the same vertex set V, where each edge is turned on with probability p. If G is the complete graph or a union of cyclic graphs with have explicit formulas for the signature polynomials f and g.