MPI-I-2004-1-006

The Hadwiger number m(G) of a graph G is the largest integer h such that the complete graph on h nodes is a minor of G. Equivalently, it is the largest integer such that any graph on at most m(G) nodes is a minor of G. The Hadwiger's conjecture states that for any graph G, m(G) >= chi(G), where chi(G) is the chromatic number of G. It is well-known that for any connected undirected graph G, there exists a unique prime factorization with respect to Cartesian graph products. If the unique prime factorization of G is given as G1 X G2 X ... X Gk, where each Gi is prime, then we say that the product dimension of G is k. Such a factorization can be computed efficiently. In this paper, we study the Hadwiger's conjecture for graphs in terms of their prime factorization. We show that the Hadwiger's conjecture is true for a graph G if the product dimension of G is at least 2log(chi(G)) + 3. In fact, it is enough for G to have a connected graph M as a minor whose product dimension is at least 2log(chi(G)) + 3, for G to satisfy the Hadwiger's conjecture. We show also that if a graph G is isomorphic to F^d for some F, then mr(G) >= chi(G)^{\lfloor \frac{d-1}{2} \rfloor}, and thus G satisfies the Hadwiger's conjecture when d >= 3. For sufficiently large d, our lower bound is exponentially higher than what is implied by the Hadwiger's conjecture. Our approach also yields (almost) sharp lower bounds for the Hadwiger number of well-known graph products like d--dimensional hypercubes, Hamming graphs and the d--dimensional grids. In particular, we show that for a d--dimensional hypercube Hd, $2^{\lfloor\frac{d-1}{2}\rfloor} <= m(Hd) <= 2^{\frac{d}{2}}\sqrt{d} +1$. We also derive similar bounds for G^d for almost all G with n nodes and at least nlog(n) edges.