1: \begin{abstract}

2: Derivative-free optimization (DFO) consists in finding the best value of an 

3: objective function without relying on derivatives. To tackle such problems, 

4: one may build approximate derivatives, using for instance finite-difference 

5: estimates. One may also design algorithmic strategies that perform space 

6: exploration and seek improvement over the current point. The first type of 

7: strategy often provides good performance on smooth problems but at the expense 

8: of more function evaluations. The second type is cheaper and 

9: typically handles non-smoothness or noise in the objective better. Recently, 

10: full-low evaluation methods have been proposed as a hybrid class of DFO 

11: algorithms that combine both strategies, respectively denoted as Full-Eval and 

12: Low-Eval. In the unconstrained case, these methods showed promising numerical 

13: performance.

14: 

15: In this paper, we extend the full-low evaluation framework to bound and 

16: linearly constrained derivative-free optimization. We derive convergence 

17: results for an instance of this framework, that combines finite-difference

18: quasi-Newton steps with probabilistic direct-search steps. The former are 

19: projected onto the feasible set, while the latter are defined within tangent cones 

20: identified by nearby active constraints. We illustrate the practical 

21: performance of our instance on standard linearly constrained problems, that 

22: we adapt to introduce noisy evaluations as well as non-smoothness. In all cases, 

23: our method performs favorably compared to algorithms that rely solely on 

24: Full-eval or Low-eval iterations.

25: \end{abstract}

26: