Branch-and-reduce algorithms aim at finding all the solutions of constraint satisfaction problems. They combine branching and reduction steps using e.g. interval methods. In this framework, we introduce expansion steps that calculate approximate solutions with a local optimization solver and find enclosing regions of the exact solutions in their neighborhoods. It may result in a better separation of the solutions early in the solving process, hence accelerating the convergence of the algorithm. We expect that such a local/global approach can benefit continuous global optimization.