We propose an approach to define numerical methods for exascale computational systems.
We propose an adapted programming model for these computing systems.
We point out the characteristics of such numerical methods.
We show that the methods defined according to the unite and conquer approach could achieve our objective.