No more than three favorite sites for simple random walk (Q1872196)

Consider a simple symmetric random walk on the integers. The site \(x\) is a favorite site of the random walk at time \(n\) if the number of visits to \(x\) before time \(n\) is larger than or equal to the number of visits to any other site \(y\). It is obvious that for infinitely many times \(n\), there is exactly one favorite site of the random walk. It is also easy to verify that for infinitely many times \(n\), there are exactly two favorite sites of the random walk. A famous question of Erdős and Revesz is the following: Is the number of favorite sites almost surely for infinitely many times, larger than or equal to 3, 4, 5, \dots ? The paper gives a partial answer to this question, by showing that with probability 1, there are at most finitely many times when there are 4 or more favorite sites of the random walk. Let \(f(r)\) be the (possibly infinite) number of steps, when the currently occupied site is one of the \(r\) actual favorites. The author shows that \(f(4)\) has finite expectation, which implies of course that \(f(4)\) is almost surely finite. The proof uses the Ray-Knight representation of the local time process of the random walk, stopped at inverse local times, to relate \(f(4)\) to a critical Galton-Watson process with geometric offspring distribution. It can be deduced from the proof that in contrast to \(f(4)\), \(f(3)\) has infinite expectation. The (open) conjecture is that \(f(3)\) is finite, almost surely, too.