Koopman operator theory lifts nonlinear dynamics into a linear framework. Instead of tracking where a state goes (nonlinear), it tracks how functions of the state evolve (linear, but infinite-dimensional). The tradeoff: the state space becomes simple (linear), but the operator space becomes infinite. In practice, you truncate — keep a finite number of Koopman modes and hope they capture the essential dynamics.
Port-Hamiltonian systems encode energy structure. Every state variable has a conjugate, and the dynamics preserve a Hamiltonian — typically the total energy. The structure guarantees passivity: the system can only dissipate energy, never generate it from nothing. This structure is load-bearing for control design, because passivity guarantees stability under interconnection.
The Koopman generator decomposition for port-Hamiltonian systems connects these two frameworks. The Koopman modes — the spectral building blocks of the linearized dynamics — are shown to respect the port-Hamiltonian structure. The energy-preserving and energy-dissipating parts of the dynamics decompose cleanly in the Koopman basis. The Hamiltonian structure isn't just preserved as an abstract property; it's visible in the spectral decomposition.
This matters because Koopman-based methods are increasingly used for data-driven modeling — learning dynamics from observed trajectories. Without structural guarantees, a data-driven Koopman model might predict energy creation or destruction that the physical system can't produce. The port-Hamiltonian decomposition constrains the Koopman model to be physically consistent, even when learned from noisy data.