1: \begin{abstract}

2: 			%% Text of abstract

3: 			We analyze a stochastic process resulting from the normalization of states in the zeroth-order optimization method CMA-ES.

4: 			On a specific class of minimization problems \del{known as}\nnew{where the objective function is} scaling-invariant, this process defines a time-homogeneous Markov chain\del{, and we can study its convergence.} whose convergence at a geometric rate can \nnew{imply the}\del{yield\del{ to} guarantees}\niko{is there more than one guaranty of linear convergence?}\del{ of} linear convergence of CMA-ES.

5: 			However, the analysis of the \new{intricate}\del{complex} updates for this process \del{presents}\nnew{constitute} a great mathematical challenge.

6:  We establish that this Markov chain is an irreducible and aperiodic T-chain.

7: 			These contributions represent a first major step for the convergence analysis towards a stationary distribution.

8: 	We rely for this analysis on conditions for the irreducibility of nonsmooth state-space models on manifolds.

9: 			\new{To obtain our results,}\del{Besides,} we extend these \new{conditions to address}\del{results to conclude to\niko{my dictionaries don't contain "conclude" with the word "to" after}} the irreducibility in different hyperparameter settings that define different Markov chains, 

10: 				and to include nonsmooth state spaces.

11: 

12: 			

13: 		\end{abstract}

14: