Hafsteinn a Computer Scientist with a background in Comp. Neuro and ML

Percolation with inhibition - part 2

This part 2 in the series on percolation with inhbition. If you have not yet read part 1 you can find it here.

In this post we will try to get an intuition for how many vertices eventually turn active in the process. Let us assume that we have vertices in total, of them which send positive messages and of them which send negative messages.

For a fixed vertex in the graph we can assume that it does not know if its neighbours are positive or negative. Since it only receives at most one message from each neighbour we can think of these messages being independently positive with probability and negative with probability . Formally these events are not independent but the dependence is weak so you can assume they are to get some intuition behind the process.

This allows us to model the process as a Markov chain. If you are unfamiliar with Markov chains there is a great pictorial explanation here. Our Markov chain consists of infinitely many states which represent the potential of our single vertex under consideration. The states are given by the set and correspond to the value of the potential of the vertex. Like all vertices which are not active at the start it starts with its potential in state 0 and it can only stop if it ever reaches state . The transition probabilities are given as the probability of increasing or decreasing the potential as explained in the previous paragraph.

Below you see an interactive version of the process. The fact that state is an absorbing state is denoted by a self-loop. Since we do not have space for infinitely many states the left-most state represents all states smaller or equal to . You can imagine the chain extending infinitely to the left. Additionally we reset the process if it has not finished after 10 steps (you can change this variable below). Note that the process should always reach if with enough many steps. However, if we have a drift to the left. This means that if the potential becomes too small there is hardly any chance that it will be able to recover. Try it out by choosing and for example.

Below you can see some statistics of the simulations of the Markov chain. The histogram shows after how many steps the potential reached . Observe that for the most likely outcome is that the process goes directly to .

The plot below shows which fraction of all the simulations ever reach within the given number of steps. You can use it to estimate which fraction of the vertices turn active in the bootstrap percolation process. Just set the maximum number of steps in the Markov chain to the average degree of the graph.

By pressing the update parameters button you start a new simulation and draw a new curve. The old one is not lost however so you can compare different parameters.