Paperlane

no. 15 · Monday 16 November 2026

The elephant random walk with proportionally growing recent memory

Zhishui Hu and Shuwen Xue

math.PR

Abstract

In this work, we study an elephant random walk with recent restricted memory, where at step $n$ the elephant remembers only the last $m_n=\max\{\lfloor cn\rfloor,1\}$ steps for $c\in(0,1)$. We show that the random walk undergoes a phase transition at the critical point $p_c=(3+\sqrt{1-c})/4$, which depends on the memory proportion $c$. For $0<p<p_c$, $S_n/\sqrt{n}$ is asymptotically normal; for $p=p_c$, $S_n/\sqrt{n\log n}$ is asymptotically normal; and for $p_c<p<1$, $S_n/n^ρ$ converges almost surely and in $L^2$ to a random variable for some constant $ρ\in (1/2, 1)$. This phase transition differs from that of the classical elephant random walk and its remote-memory counterpart, both of which have the same fixed critical point $3/4$.

fields
Probability
pdf ↗ page arXiv:2611.17664 ↗