In the distributed operations of route discovery and maintenance, strong interaction occurs across mobile ad hoc network (MANET) protocol layers in the transport of service packets between source and destination nodes. Quality of service (QoS) requirements are specified for multimedia service classes by the application layer. Due to node mobility, knowledge of their locations is inherently incomplete. Network topology is thus only partially known in time and in space. Cross-layer interactions in the protocol are modeled in terms of a set of concatenated design parameters, updates on node positions and associated energy costs. Functional dependencies of the QoS metrics on the concatenated parameters are described. An analytical framework is established for development of new cross-layer designs that optimize layer interdependencies to achieve the “best” QoS available in the MANET given a partially known, time-varying topology. The designs, based on a reactive MANET protocol, adapt provisioned QoS to dynamic network conditions and residual energy capacities. Cross-layer optimization, given partially known topology, is based on stochastic dynamic programming and predictive estimation conditions derived from time-dependent models of MANET behavior. Models of real-time behavior are based on the control of conditional rates of multivariate point processes (MVPPs). These rates depend on the concatenated protocol and resource parameters. The analytical framework supports predictive models to estimate node mobility in addition to measurement-based estimates of probability distributions for voice, video, data, and other Internet traffic. Estimates of QoS metrics are given in terms of recursive stochastic filters of the network state, based on complete or partial observations of topology and events.