Figure 2
From: Topological network features determine convergence rate of distributed average algorithms
(a) In a SW graph with size \(n = 1000\) and average degree \(deg_{avg} = 20\), after a faster transient phase, the logarithm of the normalised error decreases linearly but at a slower rate than the complete graph of the same size. The convergence rate can be estimated as the angular coefficient of the regression line of the normalised error with respect to time. (b) In graphs where nodes are assigned values from a Gaussian random distribution with \(\mu = 0\) and \(\sigma \) varying in [0, 1000], the angular coefficient of the regression line is constant, i.e. the convergence rate is independent of the initial values. (c) The convergence rate in complete graphs is independent of the graph size n.
