1: \begin{abstract}

2: At each iteration of a Block Coordinate Descent method one minimizes

3: an approximation of the objective function with respect to a generally

4: small set of variables subject to constraints in which these variables

5: are involved. The unconstrained case and the case in which the

6: constraints are simple were analyzed in the recent literature. In this

7: paper we address the problem in which block constraints are not simple

8: and, moreover, the case in which they are not defined by global sets

9: of equations and inequations. A general algorithm that minimizes

10: quadratic models with quadratric regularization over blocks of

11: variables is defined and convergence and complexity are proved. In

12: particular, given tolerances $\delta>0$ and $\varepsilon>0$ for

13: feasibility/complementarity and optimality, respectively, it is shown

14: that a measure of $(\delta,0)$-criticality tends to zero; and the the

15: number of iterations and functional evaluations required to achieve

16: $(\delta,\varepsilon)$-criticality is $O(\varepsilon^2)$. Numerical

17: experiments in which the proposed method is used to solve a continuous

18: version of the traveling salesman problem are presented.\\

19:   

20: \noindent

21: \textbf{Key words:} Coordinate descent methods, convergence, complexity.\\

22: 

23: \noindent

24: \textbf{AMS subject classifications:} 90C30, 65K05, 49M37, 90C60,

25: 68Q25.

26: \end{abstract}

27: