Port-Hamiltonian systems on graphs

In this paper we present a unifying geometric and compositional framework for modeling complex physical network dynamics as port-Hamiltonian systems on open graphs. Basic idea is to associate with the incidence matrix of the graph a Dirac structure relating the flow and effort variables associated to the edges, internal vertices, as well as boundary vertices of the graph, and to formulate energy-storing or energy-dissipating relations between the flow and effort variables of the edges and internal vertices. This allows for state variables associated to the edges, and formalizes the interconnection of networks. Examples from different origins such as consensus algorithms are shown to share the same structure. It is shown how the identified Hamiltonian structure offers systematic tools for the analysis of the resulting dynamics.

💡 Research Summary

The paper presents a unified geometric and compositional framework that models a wide variety of physical network dynamics as port‑Hamiltonian (PH) systems defined on open graphs. The central construction starts from the incidence matrix of a directed graph, which induces a linear incidence operator B mapping edge‑space variables to vertex‑space variables, and its adjoint B* (the co‑incidence operator). Flow variables (currents, velocities, etc.) live in the vertex and edge spaces, while effort variables (voltages, forces, etc.) reside in the corresponding dual spaces.

Two canonical Dirac structures are introduced: a flow‑continuous Dirac structure D_f(G) and an effort‑continuous Dirac structure D_e(G). In D_f(G) the boundary flows are directly linked to edge flows, whereas in D_e(G) the boundary efforts are determined by a part of the internal vertex efforts. Both structures satisfy the defining properties of a Dirac structure—power‑conservation (⟨e|f⟩=0) and maximal dimensionality—so they encode the power‑preserving interconnection laws inherent in the graph topology.

The authors show that Dirac structures are closed under composition: given two Dirac structures sharing a common port space, their interconnection (defined by matching flows and opposite efforts on the shared ports) yields another Dirac structure. This compositionality is crucial for PH theory because it guarantees that interconnecting PH subsystems results in a larger PH system whose Dirac structure is the composition of the subsystems’ Dirac structures and whose Hamiltonian is the sum of the subsystems’ Hamiltonians.

Applying this machinery, the paper models a mass‑spring‑damper network. Vertices represent masses (state variables are momenta), edges represent springs (energy‑storing) and dampers (energy‑dissipating). The incidence matrix encodes Kirchhoff‑type balance laws, while the Hamiltonian consists of kinetic energy of the masses and potential energy of the springs. Depending on whether the boundary vertices have mass, either D_f(G) (massless boundary) or D_e(G) (massful boundary) is used.

Beyond mechanical examples, the framework is shown to encompass electrical circuits, hydraulic and chemical reaction networks, and even algorithmic processes such as consensus and clustering. In the consensus case, the graph Laplacian appears as the incidence‑derived operator, and the algorithm’s dynamics can be written as a PH system with a quadratic Hamiltonian and a dissipative term. This unifies disparate fields under a single energy‑based description, enabling the use of PH tools—passivity analysis, energy shaping, and structure‑preserving discretization—for stability and control.

A special “Kirchhoff‑Dirac” structure is identified for the situation where no storage or dissipation is attached to vertices, reproducing the classical Kirchhoff current and voltage laws. The authors also hint at extensions to higher‑order complexes (k‑complexes), which would allow structure‑preserving spatial discretization of distributed‑parameter systems governed by partial differential equations.

In summary, the paper establishes that any dynamics defined on a directed graph can be captured by a Dirac structure derived from the graph’s incidence matrix together with a Hamiltonian that accounts for stored energy and a set of static relations that model dissipation. This yields a powerful, modular, and geometrically transparent framework for modeling, analysis, and control of complex, multi‑scale physical networks.