In this paper, a crossover operator for genetic algorithms is introduced to solve partially separable global optimization problems involving many variables. The fitness function must be an addition of positive sub-functions involving only a subset of the variables. A ''local fitness'' is associated to each variable and a parameter $\Delta$ controlling the operator's determinism is introduced. Combined with sharing and simulated annealing, this operator improves GAs efficiency to optimize combinational problems involving many variables. A polynomial function is given as an example and the operator is then used to solve a $200$ cities' TSP. The operator becomes necessary for problems such as conflict resolution involving many aircraft for air traffic control.