We investigate the run-up of a shock wave from inside to the surface of a perfect fluid star in equilibrium and bounded by vacuum. Near the surface we approximate the fluid motion as plane-symmetric and the gravitational field as constant. We consider the "hot" equation of state $P=(\Gamma-1)\rho e$ and its "cold" (fixed entropy, barotropic) form $P=K_0\rho^\Gamma$ (the latter does not allow for shock heating). We find numerically that the evolution of generic initial data approaches universal similarity solutions sufficiently near the surface, and we construct these similarity solutions explicitly. The two equations of state show very different behaviour, because shock heating becomes the dominant effect when it is allowed. In the barotropic case, the fluid velocity behind the shock approaches a constant value, while the density behind the shock approaches a power law in space, as the shock approaches the surface. In the hot case with shock heating,the density jumps by a constant factor through the shock, while the sound speed and fluid velocity behind the shock diverge in a whiplash effect. We tabulate the similarity exponents as a function of the equation of state parameter $\Gamma$and the stratification index $n_*$.

We have developed a novel computer code designed to follow the evolution of cosmic-ray modified shocks, including the full momentum dependence of the particles for a realistic diffusion coefficient model. In this form the problem is technically very difficult, because one needs to cover a wide range of diffusive scales, beginning with those slightly larger than the physical shock thickness. With most finite difference schemes for Euler's equations the numerical shock thickness is at least one zone across, so this provides a lower bound on the physical scale for diffusive transport computation. Our code uses sub-zone shock tracking and multi-level adaptive mesh refinement to provide enhanced spatial resolution around shocks at modest cost compared to the coarse grid and vastly improved cost effectiveness compared to a uniform, highly refined grid. We present and discuss the implications from our initial results.