arXiv subjects
Publications and source records attributed to Joel Friedman.
We show that the abelian girth of a graph is at least three times its girth. We prove an analogue of the Moore bound for the abelian girth of regular graphs, where the degree of the graph is fixed and the number of vertices is large. We conclude that one could try to improve the Moore bound for graphs of fixed degree and many vertices by trying to improve its analogue concerning the abelian girth.
We show that logarithmic derivative of the Zeta function of any regular graph is given by a power series about infinity whose coefficients are given in terms of the traces of powers of the graph's Hashimoto matrix. We then consider the expected value of this power series over random, $d$-regular graph on $n$ vertices, with $d$ fixed and $n$ tending to infinity. Under rather speculative assumptions, we make a formal calculation that suggests that for fixed $d$ and $n$ large, this expected value should have simple poles of residue $-1/2$ at $\pm (d-1)^{-1/2}$. We shall explain that calculation suggests that for fixed $d$ there is an $f(d)>1/2$ such that a $d$-regular graph on $n$ vertices is Ramanujan with probability at least $f(d)$ for $n$ sufficiently large. Our formal computation has a natural analogue when we consider random covering graphs of degree $n$ over a fixed, regular "base graph." This again suggests that for $n$ large, a strict majority of random covering graphs are relatively Ramanujan. We do not regard our formal calculations as providing overwhelming evidence regarding the frequency of Ramanujan graphs. However, these calculations are quite simple, and yield intiguing suggestions which we feel merit further study.
We prove a relativization of the Alon Second Eigenvalue Conjecture for all $d$-regular base graphs, $B$, with $d\ge 3$: for any $ε>0$, we show that a random covering map of degree $n$ to $B$ has a new eigenvalue greater than $2\sqrt{d-1}+ε$ in absolute value with probability $O(1/n)$. Furthermore, if $B$ is a Ramanujan graph, we show that this probability is proportional to $n^{-{η_{\rm \,fund}}(B)}$, where ${η_{\rm \,fund}}(B)$ is an integer depending on $B$, which can be computed by a finite algorithm for any fixed $B$. For any $d$-regular graph, $B$, ${η_{\rm \,fund}}(B)$ is greater than $\sqrt{d-1}$. Our proof introduces a number of ideas that simplify and strengthen the methods of Friedman's proof of the original conjecture of Alon. The most significant new idea is that of a ``certified trace,'' which is not only greatly simplifies our trace methods, but is the reason we can obtain the $n^{-{η_{\rm \,fund}}(B)}$ estimate above. This estimate represents an improvement over Friedman's results of the original Alon conjecture for random $d$-regular graphs, for certain values of $d$.
We introduce a notion of a sheaf of vector spaces on a graph, and develop the foundations of homology theories for such sheaves. One sheaf invariant, its "maximum excess," has a number of remarkable properties. It has a simple definition, with no reference to homology theory, that resembles graph expansion. Yet it is a "limit" of Betti numbers, and hence has a short/long exact sequence theory and resembles the $L^2$ Betti numbers of Atiyah. Also, the maximum excess is defined via a supermodular function, which gives the maximum excess much stronger properties than one has of a typical Betti number. The maximum excess gives a simple interpretation of an important graph invariant, which will be used to study the Hanna Neumann Conjecture in a future paper. Our sheaf theory can be viewed as a vast generalization of algebraic graph theory: each sheaf has invariants associated to it---such as Betti numbers and Laplacian matrices---that generalize those in classical graph theory.
In this paper we establish some foundations regarding sheaves of vector spaces on graphs and their invariants, such as homology groups and their limits. We then use these ideas to prove the Hanna Neumann Conjecture of the 1950's; in fact, we prove a strengthened form of the conjecture. We introduce a notion of a sheaf of vector spaces on a graph, and develop the foundations of homology theories for such sheaves. One sheaf invariant, its "maximum excess," has a number of remarkable properties. It has a simple definition, with no reference to homology theory, that resembles graph expansion. Yet it is a "limit" of Betti numbers, and hence has a short/long exact sequence theory and resembles the $L^2$ Betti numbers of Atiyah. Also, the maximum excess is defined via a supermodular function, which gives the maximum excess much stronger properties than one has of a typical Betti number. Our sheaf theory can be viewed as a vast generalization of algebraic graph theory: each sheaf has invariants associated to it---such as Betti numbers and Laplacian matrices---that generalize those in classical graph theory. We shall use "Galois graph theory" to reduce the Strengthened Hanna Neumann Conjecture to showing that certain sheaves, that we call $ρ$-kernels, have zero maximum excess. We use the symmetry in Galois theory to argue that if the Strengthened Hanna Neumann Conjecture is false, then the maximum excess of "most of" these $ρ$-kernels must be large. We then give an inductive argument to show that this is impossible.
We prove the Strengthened Hanna Neumann Conjecture, in its common graph theoretic formulation. Our original approach to this conjecture used cohomology of sheaves on graphs, although here we give a short combinatorial proof that we found in a succession of simplifications of the cohomological approach. Our proof is inductive, and requires Tardos' previous result settling the rank two case of the conjecture.
We prove the Strengthened Hanna Neumann Conjecture. We give a more direct cohomological interpretation of the conjecture in terms of "typical" covering maps, and use graph Galois theory to "symmetrize" the conjecture. The conjecture is then related to certain kernel of a morphism of sheaves, and is implied provided these kernels are co-acyclic in the covering cohomology theory. This allows us to prove a slightly generalized Strengthened Hanna Neumann Conjecture; this conjecture is false if generalized to all sheaves. The kernels we use do not exist in the theory of graphs, so our use of sheaf theory seems essential to this approach.
A non-backtracking walk on a graph, $H$, is a directed path of directed edges of $H$ such that no edge is the inverse of its preceding edge. Non-backtracking walks of a given length can be counted using the non-backtracking adjacency matrix, $B$, indexed by $H$'s directed edges and related to Ihara's Zeta function. We show how to determine $B$'s spectrum in the case where $H$ is a tree covering a finite graph. We show that when $H$ is not regular, this spectrum can have positive measure in the complex plane, unlike the regular case. We show that outside of $B$'s spectrum, the corresponding Green function has ``periodic decay ratios.'' The existence of such a ``ratio system'' can be effectively checked, and is equivalent to being outside the spectrum. We also prove that the spectral radius of the non-backtracking walk operator on the tree covering a finite graph is exactly $\sqrt\gr$, where $\gr$ is the growth rate of the tree. This further motivates the definition of the graph theoretical Riemann hypothesis proposed by Stark and Terras \cite{ST}. Finally, we give experimental evidence that for a fixed, finite graph, $H$, a random lift of large degree has non-backtracking new spectrum near that of $H$'s universal cover. This suggests a new generalization of Alon's second eigenvalue conjecture.
In a previous paper we have suggested a number of ideas to attack circuit size complexity with cohomology. As a simple example, we take circuits that can only compute the AND of two inputs, which essentially reduces to SET COVER. We show a very special case of the cohomological approach (one particular free category, using injective and superskyscraper sheaves) gives the linear programming bound coming from the relaxation of the standard integer programming reformulation of SET COVER.
This paper is motivated by questions such as P vs. NP and other questions in Boolean complexity theory. We describe an approach to attacking such questions with cohomology, and we show that using Grothendieck topologies and other ideas from the Grothendieck school gives new hope for such an attack. We focus on circuit depth complexity, and consider only finite topological spaces or Grothendieck topologies based on finite categories; as such, we do not use algebraic geometry or manifolds. Given two sheaves on a Grothendieck topology, their "cohomological complexity" is the sum of the dimensions of their Ext groups. We seek to model the depth complexity of Boolean functions by the cohomological complexity of sheaves on a Grothendieck topology. We propose that the logical AND of two Boolean functions will have its corresponding cohomological complexity bounded in terms of those of the two functions using ``virtual zero extensions.'' We propose that the logical negation of a function will have its corresponding cohomological complexity equal to that of the original function using duality theory. We explain these approaches and show that they are stable under pullbacks and base change. It is the subject of ongoing work to achieve AND and negation bounds simultaneously in a way that yields an interesting depth lower bound.
The purpose of this paper is to develop a "calculus" on graphs that allows graph theory to have new connections to analysis. For example, our framework gives rise to many new partial differential equations on graphs, most notably a new (Laplacian based) wave equation; this wave equation gives rise to a partial improvement on the Chung-Faber-Manteuffel diameter/eigenvalue bound in graph theory, and the Chung-Grigoryan-Yau and (in a certain case) Bobkov-Ledoux distance/eigenvalue bounds in analysis. Our framework also allows most techniques for the non-linear p-Laplacian in analysis to be easily carried over to graph theory.
In this paper we show the following conjecture of Noga Alon. Fix a positive integer d>2 and real epsilon > 0; consider the probability that a random d-regular graph on n vertices has the second eigenvalue of its adjacency matrix greater than 2 sqrt(d-1) + epsilon; then this probability goes to zero as n tends to infinity. We prove the conjecture for a number of notions of random d-regular graph, including models for d odd. We also estimate the aforementioned probability more precisely, showing in many cases and models (but not all) that it decays like a polynomial in 1/n.