Page j splits one vote equally among the cj pages it links to, and a page is important if important pages vote for it: pi = (1 − d) + d ∑j (Lij/cj) pj. Equivalently, a surfer follows a random link with probability d and jumps to a random page otherwise; the PageRanks are N times the fraction of time the surfer spends on each page. Run the power method step by step, change the damping d, release a surfer, and see what goes wrong when d = 1.
Disc areas show the PageRanks after k iterations of p ← Ap, p ← Np/eTp (14.110), with A = (1 − d)eeT/N + dLDc−1 (14.109). A page with no outgoing links (dangling) is treated as linking to every page, the usual fix; the book's examples have none. The surfer starts on page 1 and uses a fixed random seed.
code/ch14_unsupervised_learning_b.py (pagerank, fig_14_46, fig_14_47); this page runs the same power method.