We propose here a new approach to study co-evolution and we apply it to the well-known iterated prisoner's dilemma. The originality of our work is that it uses a simplified version of the game, and thus, restrict the search space of evolutionary dynamics. This allows to have a look at the totality of the search space in permanence, and so, a complete understanding of the phenomenon of co-evolution in process. The paper includes a little game-theoretic introduction to iterated prisoner's dilemma, a survey of previous works on evolution in this game and the exposition of the questions that were still asked to us. We describe then our special approach to the problem, using populations larger than the search space, or even infinite. The experimental results that we present complete the actual knowledge of iterated prisoner's dilemma.