1: \begin{abstract}

2: The class of algorithms called \emph{Hessian Estimation Evolution

3: Strategies (HE-ESs)} update the covariance matrix of their sampling

4: distribution by directly estimating the curvature of the objective

5: function. The approach is practically efficient, as attested by

6: respectable performance on the BBOB testbed, even on rather irregular

7: functions.

8: 

9: In this paper we formally prove two strong guarantees for the

10: (1+4)-HE-ES, a minimal elitist member of the family: stability of the

11: covariance matrix update, and as a consequence, linear convergence on

12: all convex quadratic problems at a rate that is independent of the problem instance.

13: \end{abstract}

14: