Lower bounds for graph embeddings and combinatorial preconditioners

• Authors: Gary L. Miller;Peter C. Richter
• Venue: Proceedings of the sixteenth annual ACM symposium on Parallelism in algorithms and architectures
• Year: 2004

Given a general graph G, a fundamental problem is to find a spanning tree H that best approximates G by some measure. Often this measure is some combination of the congestion and dilation of an embedding of G into H. One example is the routing time p(G, H) ≤ O (congestion + dilation), the number of steps necessary to route pairwise demands G on network links H in the store-and-forward packet routing model. Another is the condition number kf(G, H) ≤ O (congestion · dilation), the square root of which bounds the number of iterations necessary to solve a linear system with coefficient matrix G preconditioned by H using the classical conjugate gradient method. The algorithmic applications of being able to find (efficiently) a good tree approximation H for a graph G are numerous; but what if no good tree exists.In this paper, we seek to identify the class of graphs G which are intrinsically difficult to approximate by a particular measure. It is easily seen that with respect to routing time, G is hardest to approximate by a tree H precisely when it contains either long cycles (which yield high dilation)or large separators (which yield high congestion). We show that with respect to condition number, the existence of long cycles or large separators in G is sufficient but not necessary or it to be hardest to approximate, by demonstrating a nearly-linear lower bound or the case in which G is a square mesh. The proof uses concepts from circuit theory, linear algebra, and geometry, and it generalizes to the case in which H is a spanning subgraph of G of Euler characteristic k. The result has consequences or the design of preconditioners or symmetric M-matrices and perhaps also of communication networks.