Matthew Toseland wrote: > Anyone feel like running some > simulations to confirm the theory and the fix?
OK, but we need to come up with a precise model of what's happening before we can simulate it. There seem to be two possible causes: non-uniform clustering and non-uniform lifetime (maybe both). Lifetime is relatively easy: assign each new node a lifetime drawn from some distribution (fixed, exponential, power law seem like plausible candidates). Clustering is harder: here's an initial suggestion adapted from [1]. Each new node connects to m neighbours. The first neighbour is chosen at random. Then for each of the m-1 remaining neighbours, with probability p you choose a random neighbour of your existing neighbours (this creates clustering); with probability 1-p you choose a random node. The value of p can be different for each node, so we can test what happens when clustering is correlated with lifetime (ie long-lived nodes form clusters and short-lived nodes are peripheral). Cheers, Michael [1] http://nlsc.ustc.edu.cn/BJKim/PAPER/PRE_CLUS.PDF (the paper uses preferential attachment and therefore creates scale-free networks, but my suggestion above doesn't use preferential attachment)
