1: \begin{abstract}

2: We study scaling limits of non-increasing Markov chains with values in

3: the set of non-negative integers, under the assumption that the large jump

4: events are rare and happen at rates that behave like a negative power

5: of the current state. We show that the chain starting from $n$ and

6: appropriately rescaled, converges in distribution, as $n \rightarrow

7: \infty$, to a non-increasing self-similar Markov process. This

8: convergence holds jointly with that of the rescaled absorption time to

9: the time at which the self-similar Markov process reaches first 0.

10: 

11: 

12: We discuss various applications to the study of random walks with a

13: barrier, of the number of collisions in $\Lambda$-coalescents that do

14: not descend from infinity and of non-consistent regenerative

15: compositions.  Further applications to the scaling limits of Markov

16: branching trees are developed in the forthcoming paper

17: \cite{HMMarkovBranching}.

18: \end{abstract}

19: