1: \begin{definition}

2: Let $H=(V,E)$ be a connected hypergraph. Define an {\it annihilating random walk on $H$} (Figure~\ref{annrw}) by the following:

3: \begin{itemize} 

4: \item[(a).]

5:  {\bf Initial state:} Initially: $A_{i}\in  \{0,1\}$. We identify this configuration with  $\mathcal{B}=\{i\in V:A_{i}=1\}$, and call such a vertex $i$ {\it live}. 

6:  \item[(b).]

7:  {\bf Moves:} Choose pair $i,e$ consisting of a {\bf random live node} $i$ and a random hyperedge $e=(i,j_{1},\ldots, j_{k})$ containing $i$. Simultaneously set $A_{v}=A_{v}\oplus A_{i}$ for all $v\in e$ (including $v=i$, which will result in $A_{i}=0$).  

8:  \item[(c).] {\bf Annihilation:} The event we want to time is {\it annihilation}, defined by condition $A_{v}=0$ for all  $v$. 

9:  \item[(c).] {\bf Stabilizing configurations:} To be able to talk about the expected time to annihilation we will limit ourselves to \textit{stabilizing configurations}, i.e. configurations $C$ of the annihilating random walk such that the annihilating configuration \textbf{0} is reachable from $C$ and all its descendants. We will denote the set of stabilizing configurations of $G$ by $\mathcal{S}(G)$. As implicitly shown in Theorem~\ref{rec-xor}, configurations $\mathcal{B}$ obtained by duality from assignments to satisfiable instances of XOR-SAT are indeed stabilizing. 

10:   \end{itemize} 

11:   

12:  \end{definition}