A generative power-law search tree model

It is now a well-established fact that search algorithms can exhibit heavy-tailed behavior. However, the reasons behind this fact are not well understood. We provide a generative search tree model whose distribution of the number of nodes visited during search is formally heavy-tailed. Our model allows us to generate search trees with any degree of heavy-tailedness. We also show how the different regimes observed for the runtime distributions of backtrack search methods across different constrainedness regions of random CSP models can be captured by a mixture of the so-called stable distributions.