1: \begin{abstract}

2: We study nondeterministic and probabilistic versions of a discrete

3: dynamical system (due to T. Antal, P. L. Krapivsky, and S. Redner

4: \cite{redner-balance}) inspired by Heider's social balance theory.

5: We investigate the convergence time of this dynamics on several

6: classes of graphs.  Our contributions include:

7: \begin{enumerate}

8: \item We point out the connection between the triad

9: dynamics and a generalization of {\em annihilating walks}

10: to hypergraphs. In particular, this connection allows us

11: to completely characterize the recurrent states in graphs

12: where each edge belongs to at most two triangles.

13: \item We also solve the case of hypergraphs that do not contain edges consisting of one or two vertices.

14:  \item We show that on the so-called ``triadic

15: cycle''graph, the convergence time is linear.

16: \item We obtain a cubic upper bound on the convergence

17: time on 2-regular triadic simplexes $G$. This bound can be further

18: improved to a quantity that depends on the {\em Cheeger constant} of

19: $G$. In particular this provides some rigorous counterparts to

20: experimental observations in \cite{triad-lattice}.

21: %\item In contrast, for graphs $G$ for which an associated

22: %``triad graph'' is an expander, the dynamics displays a

23: %phase transition in the convergence time similar to the

24: %one displayed in \cite{redner-balance} for the complete

25: %graph $K_{n}$.

26: \end{enumerate}

27: 

28: We also point out an application to the analysis of the random walk algorithm on certain instances of the 3-XOR-SAT problem.

29: \end{abstract}

30: