Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Statement

Setting (p. 137). The walk is the symmetric nearest-neighbor walk on Z2\mathbb Z^2 started at the origin.

Theorem 7 (p. 151). Let ff be a monotonic function increasing to +∞+\infty, and let EnE_n be the event that the planar walk does not return to the origin between times nn and nf(n)n^{f(n)}. Then P{En i.o.}\mathbf P\{E_n\text{ i.o.}\} is 00 or 11 according as

∑k=1∞1f(22k)\sum_{k=1}^{\infty}\frac1{f\bigl(2^{2^k}\bigr)}

converges or diverges. The series runs over the doubly exponential sequence 22k2^{2^k}, the sequence nkn_k of the proof. The print writes the growth hypothesis as "increases to +∞+\infty as x→∞x\to\infty" [sic], with the variable xx for nn.

The paper motivates the theorem (p. 151) as the question of how long the gaps between returns can be: for which monotone gg the interval (n,n+g(n))(n,n+g(n)) contains a return for all but finitely many nn.

Theorem 7A (p. 153). The paper states without proof the line analogue, saying it can be proved by similar methods: for the walk on Z\mathbb Z, with ff as above and EnE_n the event of no return to the origin between nn and n{f(n)}2n\{f(n)\}^2, P(En i.o.)\mathbf P(E_n\text{ i.o.}) is 00 or 11 according as ∑1/f(2k)\sum 1/f(2^k) converges or diverges.

Source. P. Erdős and S. J. Taylor, Some problems concerning the structure of random walk paths, Acta Math. Acad. Sci. Hungar. 11 (1960), 137--162: the walk on p. 137, (2.16) on p. 141, Theorem 7 and (4.8)--(4.9) on p. 151, the proof on pp. 151--153, Theorem 7A on p. 153. The edition read is identified on the source card.

Read depth. Claims checked: Theorems 7 and 7A were read clause by clause on the printed pages. The proof on pp. 151--153 was read for the pointer below and not checked step by step. Nothing here is independently reviewed.

Proof pointer

Pages 151--153. The paper's estimate (2.16) (p. 141) for the probability that a planar walk started at distance ϱ\varrho avoids the origin for nn steps gives, from the position at time nn, two-sided bounds 13⋅1f(n)<P(En)<3f(n)\frac13\cdot\frac1{f(n)}<\mathbf P(E_n)<\frac3{f(n)} for large nn ((4.8)--(4.9), p. 151). Convergence of the series gives the zero case by Borel--Cantelli along nk=22kn_k=2^{2^k}, with ff replaced by f/2f/2 to cover the intermediate nn. For divergence the paper bounds the overlap P(En5k∩En5r)\mathbf P(E_{n_{5k}}\cap E_{n_{5r}}) in two cases ((4.14)--(4.15), p. 153), so that disjointified events carry at least half the probability, and concludes with the zero-one law.

Dependencies

The planar avoidance estimate (2.16) of the same paper (p. 141), which rests on (2.5) and the local estimates (2.9)--(2.10).

Bears on

No problem page of this corpus.