Identifying biomarkers- objective, quantifiable biologically-based measures to complement traditional clinical assessments- is critical for studying the links between brain and disorders. Recent advances in neuroimaging have shifted biomarker discovery from traditional univariate brain mapping techniques, which analyze individual brain regions separately, to multivariate predictive models that consider complex patterns across multiple regions, with dynamic functional network connectivity (dFNC) emerging as a key approach offering a dynamic view of the temporal coupling between different brain networks. Here, we introduce an innovative approach to estimate dynamic double functional independent primitives (ddFIP) by first applying a spatially constrained independent component analysis (ICA) to derive intrinsic connectivity networks (ICNs), followed by a second ICA applied to dFNC matrices. This procedure provides a set of states that reflect dynamic connectivity patterns. To characterize these states, we propose several dynamic measures: (1) amplitude convergence, which quantifies the extent to which multiple states contribute similarly to the connectivity profile at a given time (indicating more uniform state contributions); (2) amplitude divergence, quantifying the tendency for states to contribute at varying levels which does not assume dominance but rather reflects a spread of amplitudes across states; as well as (3) dynamic state density which shows the number of strongly occupied states, reflecting the brains preference for spending time in a smaller or larger set of dominant states. We apply this approach to uncover ddFIP-based biomarkers from seven resting-state functional magnetic resonance imaging (rs-fMRI) clinical datasets, which include four neuropsychiatric disorders- schizophrenia (SCZ), autism spectrum disorder (ASD), major depressive disorder (MDD), and bipolar disorder (BPD)- comprising a total of 5,805 participants. Our results revealed disorder-specific patterns in dynamic connectivity measures. SCZ exhibited widespread disruptions with high variability and increased divergence, suggesting a tendency for states to contribute at varying levels rather than uniformly. ASD, in contrast, showed significantly reduced divergence and increased convergence, indicating more uniform contributions across states and atypical stability in dynamic connectivity. BPD demonstrated heightened variability, particularly in mood regulation networks, while MDD displayed moderate disruptions, especially in self-referential processing networks. Notably, ASDs increased state convergence reflects a pattern where state weights are more similar, was sharply distinct from SCZs increased divergence, as indicated by state occupancy measures. In sum, our findings highlight the potential of continuous dFNC as a FNC-based biomarker to capture disorder-specific connectivity signatures. Moreover, by analyzing both the convergence and divergence of dynamic states, we capture a detailed view of connectivity, reflecting the brains adaptability and resilience within each disorder.