Particle-In-Cell codes are widely used for plasma physics simulations. It is often the case that particles within a computational cell need to be split to improve the statistics or, in the case of non-uniform meshes, to avoid the development of fictitious self-forces. Existing particle splitting methods are largely empirical and their accuracy in preserving the distribution function has not been evaluated in a quantitative way. Here we present a new method

<p>Collisionless magentized plasmas a priori need to be evolved using Vlasov-Maxwell kinetic formalism.<br>However the tremendous number of spatial and temporal scales involved in phenomena of interest makes it prohibitive, from a computational standpoint.<br>Fully kinetic particle in cell and single fluid MHD codes are commonly used at very small or very large scales.<br>The hybrid formalism, treating ions kinetically and electrons as a fluid, is in principle advantageous to fill the gap between these two extremities.<br>However, a correct treatment of critical regions such as reconnection X-lines require a good resolution of sub-ion dissipative scales, which<br>still constitute a major challenge if aiming at simulating meso/macro scale systems.<br>This work presents a new code, named PHARE, which successfully implements the adaptive mesh refinement mechanism in a hybrid particle-in-cell code.<br>Such a code is able to dynamically focus the resolution in critical regions while others not only have a coarser spatial resolution, but are also<br>evolved much less often thanks to a recursive time stepping procedure.<br>Adopting a patch based AMR mechanism, the code architecture is made so that the specific solver/physical model that is solved at a given refined level<br>is abstracted, thus giving the opportunity to handling multi-formalisms AMR patch hierarchies, where, for instance, coarsest levels are solved in MHD while<br>dynamicall created refined levels are solved within the Hybrid framework.</p>