Autonomic computing systems are self-aware and self-healing. Microstates of components of a system are organized into macrostates, and the macrostates are organized into meta-dynamic states (MDS's). When the system is perturbed, it is determined whether the system enters a given undesirable MDS. Where the system will enter an undesirable MDS, a microstate/MDS graph (MMG) is constructed depicting which meta-dynamic states the system enters when its microstates are changed. A self-heal graph (SHG) is constructed from the MMG. The undesirable MDS is at the center of, and a desired MDS is at the periphery of, the SHG. The SHG has paths from the undesirable MDS to the desired MDS, each path having an associated cost incurred if the system moves along the path. An optimal path is selected from these paths that have the lowest associated costs. The system moved on the selected optimal path to reach the desired MDS.