Title:Unlocking the walk matrix of a graph

Abstract:Let $G$ be a graph on the vertex set $V=\{v_{1},\dots,v_{n}\}$ with adjacency matrix $A.$ For a subset $S$ of $V$ let ${\rm e}=(x_{1},\,\dots,\,x_{n})^{\rm T}$ be the characteristic vector of $S,$ that is, $x_{\ell}=1$ if $v_{\ell}\in S$ and $x_{\ell}=0$ otherwise. Then the $\,walk \,\,matrix$ of $G$ for $S$ is $$W^{S}:=\big[{\rm e},\,A{\rm e},\,A^{2}{\rm e},\dots,A^{n-1}{\rm e}\big].$$ In this matrix the $k^{\rm th}$ entry in the row corresponding to $v_{\ell}$ is the number of walks of length $k-1$ from $v_{\ell}$ to some vertex in $S.$
We investigate two related problems: To what degree does a walk matrix determine the adjacency matrix of the graph, and how can walk matrices be computed from the eigenvalues and eigenvectors of the graph? The first question depends essentially on the rank of the walk matrix. For given walk matrix $W=W^{S}$ we define its $\,core$ $A_{W}.$ This matrix has ${\rm rank}(W)$ or ${\rm rank}(W)-1.$ We show that $A=A_{W}+N$ where $N$ is a symmetric matrix with $N\cdot A_{W}=0.$ This theorem has several applications. For instance, we show that if $W$ has rank $\geq n-1$ then the adjacency matrix can be computed from $W,$ and this result is best possible. Similar conclusions are available in probabilistic models or if suitable assumptions on the characteristic polynomial of the graph are made. We mention applications to graph automorphisms and to isomorphism testing problems when some walk matrix has large rank.
For the second question we show that $W$ is determined by an eigenvector matrix $E$ and a matrix $M$ formed from certain eigenvalues of $A$ that are related to the set $S.$ In the literature these are known as the $\,main \,\,eigenvectors$ and $\, main \,\,eigenvalues$ of $G$ for the given set. We show that $W=M\cdot E.$ In the opposite direction we show that $W$ determines both $M$ and $E.$