2 Without loss of generality, we consider Boolean domains. We have three
3 elements taking their value in the set $\{0,1\}$. Thus, a configuration is an
4 element of $\{0,1\}^3$, \textit{i.e.}, a binary number between 0 and 7. In
5 \Fig{fig:map} is given the function defining the dynamic and
6 \Fig{fig:xplgraph} depicts its associated connection graph. Operators
7 $\overline{a}$, sum and product are the classical Boolean operators. In the
8 incidence graph, or connection graph, the vertices are the elements of the
9 system and an edge from $i$ to $j$ indicates that $f_j$ depends on $i$. Notice
10 that the connection graph contains five cycles.
21 % f_1(x_1,x_2,x_3) & = & x_1.(x_2 + x_3) \\
22 % f_2(x_1,x_2,x_3) & = & \overline{x_1}.\overline{x_2}.x_3 \\
23 % f_3(x_1,x_2,x_3) & = & x_1.x_2.x_3
26 f_1(x_1,x_2,x_3) & = & x_1.\overline{x_2} + x_3 \\
27 f_2(x_1,x_2,x_3) & = & x_1 + \overline{x_3} \\
28 f_3(x_1,x_2,x_3) & = & x_2.x_3
37 \subfloat[Connexion Graph]{
41 \includegraphics[width=4cm]{xplCnx.pdf}
47 \caption{Dynamic of the running example.}
53 % %\includegraphics[width=3cm]{xplgraph.ps}
54 % \includegraphics[width=3cm]{xplgraph.pdf}
56 % \caption{Incidence graph of running example\label{fig:xplgraph}}
62 % \includegraphics[width=3cm]{xplgraph.ps}
63 % %\includegraphics[width=3cm]{xplgraph.pdf}
65 % \caption{Incidence graph of running example\label{fig:xplgraph}}
69 % The prototype developed for~\cite{BCVC10:ir} has shown
70 % that pure parallel iterations are convergent.
71 % It has helped us to find pseudo-periodic strategies making chaotic
72 % iterations (and then asynchronous iterations) diverging.
73 % This example shows then a case when convergence is established
74 % under some conditions whereas universal convergence is not.
78 % \includegraphics[width=12cm]{implementation/para_iterate_dec.ps}
80 % \caption{Parallel iterations of running example\label{fig:xplparaFig}}
83 % In what follows and for brevity reasons, configurations are represented as
84 % decimal numbers instead of binary numbers. The graph of parallel iterations is
85 % given in Fig.~\ref{fig:xplparaFig}.
86 % Starting from any configuration, the network
87 % converges to the fixed point corresponding to the decimal number 19.
91 % \includegraphics[width=4cm]{chao_iterate_excerpt.ps}
93 % \caption{Excerpt of chaotic iterations of running example
94 % \label{fig:xplchaoFig}}
97 % An extract of the graph of chaotic iterations is given in Fig.~\ref{fig:xplchaoFig}.
98 % Arc label is the characteristic function of activated elements
99 % (i.e. the matrix $J^t$) expressed as a
100 % decimal number. It allows us to shortly represent the strategy.
102 % in this graph we only consider the pseudo-periodic strategy that alternately activates both the
103 % first two elements and the last four elements. This is formalized as
106 % \begin{array}{lllll}
118 % \begin{array}{lllll}
126 % $, $p=0,1,2,\ldots$.
127 % The former matrix is represented as 24 and the later as 15.
129 % Iterations do not converge for this strategy: for instance,
130 % starting from the configuration 3 (resp. configuration 7), the network goes to the configuration 11 (resp. configuration 15) and
131 % goes back to the configuration 3 (resp. configuration 7) by applying the strategy depicted above.
134 % Since chaotic iterations do not converge for some strategies, it is obvious that
135 % convergence cannot be observed for some asynchronous iterations with the
136 % corresponding strategies. However, even if all the components
137 % are activated, that is the asynchronous counterpart of pure parallel iterations,
138 % delays may introduce divergence.
140 % For instance, let $J^t$ be constantly equal to the identity matrix and
141 % consider the matrix
144 % \begin{array}{lllll}
145 % t & t' & t & t & t \\
146 % t & t & t & t & t \\
147 % t & t & t & t & t \\
148 % t & t & t & t & t\\
149 % t & t & t & t & t \\
155 % t & \textrm{ if t is even;} \\
156 % t-1 & \textrm{ otherwise.}
163 % X^{2k+1} & =& F(X^{2k}) \\
166 % F_1(X_1^{2k-1},X_2^{2k-2},X_3^{2k-1},X_4^{2k-1},X_5^{2k-1}) \\
175 % Starting from $X^0 = (0,0,0,1,1)$, the system reaches
176 % $X^1 = (0,1,0,1,1)$ and enters in a cycle between these
177 % two configurations (as in some chaotic iterations).
180 % the next section, a particular execution mode is described which enables
181 % asynchronism in iterations while guaranteeing the convergence.
184 In what follows and for brevity reasons, configurations are represented as
185 decimal numbers instead of binary numbers. It is not hard to establish that in
186 parallel synchronous iterations, starting from any configuration different from
187 7 (111) leads to the fixed point 2 (010), and starting from 7 leads to 7.
188 With other strategies,
189 the behavior is quite different although the system always
190 reaches a stabilization. Indeed, for the initial states 1, 3 and 5, the system
191 converges towards anyone of the two fixed points 2 and 7. However, there is no
192 infinite cycle and the convergence property defined in \Equ{eq:ltl:conv} is
194 Finally, due to the presence of
195 cycle in the connexion graph,
196 known sufficient conditions~\cite{Bah00} establishing convergence
197 results for asynchronous modes cannot be applied here.
198 We address this question in what follows.
205 %%% ispell-dictionary: "american"
207 %%% TeX-master: "main"