The aim of this thesis is to understand the structure of colliding wind binaries composed of a massive star and a young pulsar, called gamma-ray binaries. They are expected to display a similar structure to colliding wind binaries composed of massive stars, with some particularities due to the relativistic nature of the pulsar wind. The interaction of the supersonic winds from massive stars creates a shocked structure with observational signatures from the radio domain to the X-rays. The structure is affected by various instabilities and by the orbital motion of the stars. To understand their impact, I carried out high resolution simulations of colliding wind binaries with the hydrodynamical code RAMSES. They are computationally demanding, especially when one of the winds strongly dominates the other one. Small scale simulations highlight the importance of the Non-linear Thin Shell Instability in isothermal collisions while the Kelvin-Helmholtz instability may strongly impact the dynamics of adiabatic collisions. I found that, at larger scales, this instability can destroy the expected large scale spiral structure when there is an important velocity gradient between the winds. WR 104 is a system that displays a spiral structure with important dust emission. The simulation of this system shows a good agreement with the observed structure and indicates cooling processes are necessary to enable dust formation. To model the pulsar wind in gamma-ray binaries, an extension of RAMSES has been developed, that incorporates relativistic hydrodynamics. I used this new relativistic code to perform preliminary simulations of gamma-ray binaries. They display a similar structure to colliding wind binaries with small relativistic corrections. We expect to use this code to perform large scale simulations of gamma-ray binaries. It will be part of the next public release of RAMSES and is suited for the study of many astrophysical problems such as relativistic jets, pulsar wind nebulae or gamma-ray bursts.
