betaFeedback

WorksMathematics

Descriptor representations of geometric port-Hamiltonian systems on Hilbert spaces

Abstract

We establish conversions in both directions between geometric and descriptor representations of linear port-Hamiltonian systems on infinite-dimensional Hilbert spaces. For the geometric-to-descriptor conversion, we represent the possibly multivalued inverse Dirac relation by a skew-adjoint operator on enlarged Hilbert spaces, retaining the original operator-domain restrictions. The construction accommodates internal and external ports. We show that the conversions preserve trajectories and the power balance. An incompressible Kelvin–Voigt solid illustrates the results.

Introduction

Port-Hamiltonian systems separate energy storage, power-conserving interconnection, and dissipation. This makes power exchange in coupled systems transparent, particularly for distributed-parameter models with boundary interactions [1, 2, 3, 4]. System-node formulations and boundary-triplet methods offer operator-theoretic approaches to these infinite-dimensional systems [5, 6].

Linear port-Hamiltonian systems have two common formulations. The geometric formulation describes admissible trajectories behaviorally: at each tt in an interval II, they satisfy the Dirac, Lagrangian, and resistive inclusions

((x˙(t)−fR(t)−y(t)),(eS(t)eR(t)u(t)))∈D,(x(t),eS(t))∈L,(fR(t),eR(t))∈R.(1)\left(\begin{pmatrix} \dot{x}(t) \\ -f_R(t) \\ -y(t) \end{pmatrix},\begin{pmatrix} e_S(t) \\ e_R(t) \\ u(t) \end{pmatrix}\right) \in\mathcal{D},\qquad(x(t),e_S(t)) \in\mathcal{L},\qquad(f_R(t),e_R(t)) \in\mathcal{R}. \tag*{(1)}

Here xx is the state, its derivative x˙\dot{x} is the storage flow, (fR,eR)(f_R,e_R) denotes the internal resistive port, and uu and yy are the external input and output. The relations specify which trajectories are admissible without first solving for the state derivative or the output.

In the descriptor formulation, the same dynamics are expressed in an operator-based input–output form,

\begin{pmatrix} \frac{\mathrm{d}}{\mathrm{d}t}(E\zeta(t)) \\ -y(t) \end{pmatrix} = M\begin{pmatrix} Q\zeta(t) \\ u(t) \end{pmatrix}. \tag*{(2)}\end{pmatrix}

The variable ζ\zeta collects storage, resistive, and auxiliary variables; EE and QQ are bounded operators, while MM is closed and may be unbounded. Its domain is part of the equation and can impose additional restrictions. In this form the external input uu and output yy are singled out, although existence or uniqueness of a trajectory is not presupposed.

To relate this formulation to the established finite-dimensional theory, consider the time-invariant port-Hamiltonian descriptor systems of [7 Def. 4],

ddt(Ez(t))=(J−R)Qz(t)+(B−P)u(t),\frac{\mathrm{d}}{\mathrm{d}t}(Ez(t))=(J-R)Qz(t)+(B-P)u(t),
y(t)=(B+P)∗Qz(t)+(S+N)u(t).y(t)=(B+P)^{*}Qz(t)+(S+N)u(t).

These equations can be written in the form (2) by setting

M=[J−RB−P−(B+P)∗−(S+N)].M=\begin{bmatrix} J-R & B-P \\ -(B+P)^{*} & -(S+N) \end{bmatrix}.

Here J=−J∗J=-J^{*}, N=−N∗N=-N^{*}, and the dissipation matrix [RPP∗S]\begin{bmatrix} R & P \\ P^{*} & S \end{bmatrix} is positive semidefinite. The energy matrices satisfy E∗Q=Q∗E≥0E^{*}Q=Q^{*}E\ge0. In finite dimensions, MM can be split into a skew-Hermitian and a negative semidefinite Hermitian part. For unbounded operators AA, this procedure may fail: dom⁡A∩dom⁡(A∗)\operatorname{dom} A \cap\operatorname{dom}(A^{*}) can be trivial, even if AA is maximal dissipative [8]. We therefore derive the power structure directly from the geometric relations. In infinite dimensions, MM may be unbounded or non-densely defined; its domain is essential and, together with auxiliary variables, can encode constraints.

For infinite-dimensional descriptor equations, see also the operator-theoretic treatment in [9]. Our aim is to convert the geometric and descriptor formulations into one another without changing the admissible trajectories or their power balance.

The relation between geometric and descriptor formulations has been studied extensively in finite dimensions [10, 11, 12, 13]; see also the recent results on nonlinear systems in [14]. In particular, [15 Thm. 15] established a geometric-to-descriptor conversion with internal and external ports. More recently, [16] developed equivalent finite-dimensional representations of complete port-Hamiltonian systems with Dirac and Lagrange constraints, including resistive and external ports, and related them to constrained input–output formulations; see also [17].

Algebraic constraints arising from Dirac structures are already familiar in finite dimensions. The additional difficulty in infinite dimensions is that the operator part of a Dirac relation may be unbounded. Its domain can encode spatial regularity and boundary conditions that cannot be recovered merely from the algebraic constraints. A conversion to descriptor form must preserve these domain restrictions as well. Moreover, the spaces associated with algebraic constraints may themselves be infinite-dimensional, as in models of incompressible continua.

Every closed linear relation admits bounded image and kernel representations. The challenge here is therefore not simply to write the geometric inclusions as operator equations. Rather, we seek a structured descriptor representation in which the Dirac interconnection is realized by a skew-adjoint operator on enlarged Hilbert spaces. The transition from a structured descriptor equation to geometric relations is comparatively direct: the inverse graph of the interconnection operator gives a Dirac structure. For the converse direction, we use the operator–multivalued decomposition of the inverse Dirac relation to obtain a skew-adjoint operator representation. An auxiliary variable accounts for the multivalued part, while the original operator domain is retained. The construction draws on the theory of linear relations [18, 19, 20, 21]; related infinite-dimensional Dirac representations are studied in [22]. Combined with parametrizations of the storage and resistive relations, this gives a closed descriptor operator with internal and external ports. We prove an exact correspondence of classical trajectories, including preservation of the power balance.

An incompressible Kelvin–Voigt solid illustrates how the descriptor representation retains unbounded differential operators, infinite-dimensional incompressibility constraints, and operator-domain restrictions.

Structure of skew-adjoint relations

We introduce notation for Hilbert spaces and linear relations, recall the operator–multivalued decomposition of skew-adjoint relations, and construct a skew-adjoint operator representation on enlarged Hilbert spaces without additional domain regularity.

Hilbert-space notation and duality

All Hilbert spaces are complex, with inner products linear in the first argument. The anti-dual U∗\mathcal{U}^{*} of a Hilbert space U\mathcal{U} consists of its bounded antilinear functionals. We denote the canonical duality pairing between U∗\mathcal{U}^{*} and U\mathcal{U} by ⟨u′,u⟩U∗,U\langle u',u\rangle_{\mathcal{U}^{*},\mathcal{U}} and use

⟨u,u′⟩U,U∗:=⟨u′,u⟩U∗,U‾\langle u,u'\rangle_{\mathcal{U},\mathcal{U}^{*}} := \overline{\langle u',u\rangle_{\mathcal{U}^{*},\mathcal{U}}}

for the reversed order. Like the inner products, both pairings are linear in the first and antilinear in the second argument. The Riesz isomorphism and the canonical biduality map are characterized by

RU:U→U∗,⟨RUu,v⟩U∗,U=⟨u,v⟩U,R_{\mathcal U}:\mathcal U\to\mathcal U^*,\qquad\langle R_{\mathcal U}u,v\rangle_{\mathcal U^*,\mathcal U}=\langle u,v\rangle_{\mathcal U},
ιU:U→U∗∗,⟨ιUu,u′⟩U∗∗,U∗=⟨u,u′⟩U,U∗.\iota_{\mathcal U}:\mathcal U\to\mathcal U^{**},\qquad\langle\iota_{\mathcal U}u,u'\rangle_{\mathcal U^{**},\mathcal U^*}=\langle u,u'\rangle_{\mathcal U,\mathcal U^*}.

Both maps are linear and unitary. For a bounded linear operator A:U→WA:\mathcal U\to\mathcal W, its anti-dual adjoint A∗:W∗→U∗A^*:\mathcal W^*\to\mathcal U^* is characterized by

⟨A∗w′,u⟩U∗,U=⟨w′,Au⟩W∗,W.\langle A^*w',u\rangle_{\mathcal U^*,\mathcal U}=\langle w',Au\rangle_{\mathcal W^*,\mathcal W}.

The same notation is used for densely defined operators, with the anti-dual adjoint defined on its natural domain.

We use U×V\mathcal U\times\mathcal V for Cartesian products and S1⊥S2\mathcal S_1\perp\mathcal S_2 for orthogonal direct sums of subspaces of Hilbert spaces. Hilbert-space products carry the product inner product. Let U,V\mathcal U,\mathcal V be Hilbert spaces and let Γ:V→U∗\Gamma:\mathcal V\to\mathcal U^* be a bounded isomorphism. We use the duality pairing ⟨v,u⟩Γ:=⟨Γv,u⟩U∗,U\langle v,u\rangle_\Gamma:=\langle\Gamma v,u\rangle_{\mathcal U^*,\mathcal U} and denote the annihilator of a subspace S⊆U\mathcal S\subseteq\mathcal U by S∘:={v∈V∣⟨v,s⟩Γ=0 for all s∈S}\mathcal S^\circ:=\{v\in\mathcal V\mid\langle v,s\rangle_\Gamma=0\text{ for all }s\in\mathcal S\}. By contrast, S⊥\mathcal S^\perp denotes the Hilbert-space orthogonal complement. Orthogonal complements and projections refer to the given inner products.

Linear relations and their adjoints

A linear relation from U\mathcal U to V\mathcal V is a subspace T⊆U×V\mathcal T\subseteq\mathcal U\times\mathcal V. We write dom⁡T\operatorname{dom}\mathcal T and ran⁡T\operatorname{ran}\mathcal T for its domain and range. Its multivalued part is {v∈V∣(0,v)∈T}\{v\in\mathcal V\mid(0,v)\in\mathcal T\}. As usual, T−1:={(v,u)∣(u,v)∈T}\mathcal T^{-1}:=\{(v,u)\mid(u,v)\in\mathcal T\} and −T:={(u,−v)∣(u,v)∈T}-\mathcal T:=\{(u,-v)\mid(u,v)\in\mathcal T\}.

For an operator A:dom⁡A⊆U→VA:\operatorname{dom}A\subseteq\mathcal U\to\mathcal V, we write graph⁡A:={(u,Au)∣u∈dom⁡A}\operatorname{graph}A:=\{(u,Au)\mid u\in\operatorname{dom}A\}. A relation is an operator graph exactly when its multivalued part vanishes. For background on linear relations, see [18, 19, 20, 21, 23].

With respect to the duality map Γ\Gamma, the adjoint relation of T\mathcal T is

T∗Γ:={(a,c)∈U×V∣⟨c,u⟩Γ=⟨v,a⟩Γ for all (u,v)∈T}.(3)\mathcal T^{*_\Gamma}:=\left\{(a,c)\in\mathcal U\times\mathcal V\mathrel{\big|}\langle c,u\rangle_\Gamma=\langle v,a\rangle_\Gamma\text{ for all }(u,v)\in\mathcal T\right\}. \tag*{(3)}

We call T\mathcal T self-adjoint or skew-adjoint if T∗Γ=T\mathcal T^{*_\Gamma}=\mathcal T or T∗Γ=−T\mathcal T^{*_\Gamma}=-\mathcal T, respectively. For a skew-adjoint relation, the multivalued part equals (dom⁡T)∘=K∘(\operatorname{dom}\mathcal T)^\circ=\mathcal K^\circ, where K:=dom⁡T‾\mathcal K:=\overline{\operatorname{dom}\mathcal T}. Thus T\mathcal T is single-valued if and only if its domain is dense in U\mathcal U.

Operator representations of skew-adjoint relations

The possible multivaluedness of a skew-adjoint relation is thus determined by the closure of its domain. In infinite dimensions, dom⁡T\operatorname{dom}\mathcal T may moreover be a proper dense subspace of its closure K\mathcal K. The following theorem treats these two features separately. Its first part is the standard operator–multivalued decomposition of a self-adjoint relation, transferred to the skew-adjoint setting [20, 21]. The second part realizes the entire relation by a skew-adjoint operator without replacing its domain by its closure.

Theorem 2.1 (Representation of skew-adjoint relations) Let T⊆U×VT \subseteq\mathcal{U} \times\mathcal{V} be skew-adjoint with respect to a unitary isomorphism Γ:V→U∗\Gamma:\mathcal{V}\to\mathcal{U}^{*}. Set K:=dom⁡T\mathcal{K}:=\operatorname{dom}T and M:=K⊥\mathcal{M}:=\mathcal{K}^{\perp}. Then the following statements hold.

(a) Operator–multivalued decomposition. There exists a unique closed operator JK:dom⁡T⊆K→M∘J_{\mathcal{K}}:\operatorname{dom}T\subseteq\mathcal{K}\to\mathcal{M}^{\circ} such that

T=graph⁡JK⊥({0}×K∘).T=\operatorname{graph}J_{\mathcal{K}}\perp\left(\{0\}\times\mathcal{K}^{\circ}\right).

Its zero extension J0J_{0}, with dom⁡J0=dom⁡T⊥M\operatorname{dom}J_{0}=\operatorname{dom}T\perp\mathcal{M} and J0(uK+u⊥)=JKuKJ_{0}(u_{\mathcal{K}}+u_{\perp})=J_{\mathcal{K}}u_{\mathcal{K}} for uK∈dom⁡Tu_{\mathcal{K}}\in\operatorname{dom}T and u⊥∈Mu_{\perp}\in\mathcal{M}, is closed, densely defined, and skew-adjoint with respect to Γ\Gamma.

(b) Skew-adjoint operator representation. Equip Y:=U/K\mathcal{Y}:=\mathcal{U}/\mathcal{K} with its quotient Hilbert norm, let q:U→Yq:\mathcal{U}\to\mathcal{Y} be the quotient map, and set j:=Γ−1q∗:Y∗→Vj:=\Gamma^{-1}q^{*}:\mathcal{Y}^{*}\to\mathcal{V}. The block operator

J^:=[J0j−q0],dom⁡J^=dom⁡J0×Y∗,(4)\widehat{J}:=\begin{bmatrix}J_{0}&j\\-q&0\end{bmatrix},\qquad\operatorname{dom}\widehat{J}=\operatorname{dom}J_{0}\times\mathcal{Y}^{*}, \tag*{(4)}

from U×Y∗\mathcal{U}\times\mathcal{Y}^{*} to V×Y\mathcal{V}\times\mathcal{Y} is closed, densely defined, and skew-adjoint with respect to the unitary duality map diag⁡(Γ,ιY)\operatorname{diag}(\Gamma,\iota_{\mathcal{Y}}) from V×Y\mathcal{V}\times\mathcal{Y} to U∗×Y∗∗\mathcal{U}^{*}\times\mathcal{Y}^{**}. For every (u,v)∈U×V(u,v)\in\mathcal{U}\times\mathcal{V},

(u,v)∈T⟺∃!λ∈Y∗:J^(uλ)=(v0).(u,v)\in T\quad\Longleftrightarrow\quad\exists!\lambda\in\mathcal{Y}^{*}:\quad\widehat{J}\begin{pmatrix}u\\\lambda\end{pmatrix}=\begin{pmatrix}v\\0\end{pmatrix}.

Moreover, jj is an isometric isomorphism onto K∘\mathcal{K}^{\circ}. The unique auxiliary variable is λ=j−1PK∘v\lambda=j^{-1}P_{\mathcal{K}^{\circ}}v and satisfies ∥λ∥Y∗≤∥v∥V\|\lambda\|_{\mathcal{Y}^{*}}\leq\|v\|_{\mathcal{V}}, where PK∘P_{\mathcal{K}^{\circ}} is the orthogonal projection in V\mathcal{V}.

Proof. (a) Set W:=RΓ−1:V→UW:=R_{\Gamma}^{-1}:\mathcal{V}\to\mathcal{U}. The relation A:={(u,iWv)∣(u,v)∈T}A:=\{(u,iWv)\mid(u,v)\in T\} is self-adjoint in U\mathcal{U} and has multivalued part M\mathcal{M}. By [21 Thm. 1.5.1], it decomposes as A=graph⁡LK⊥({0}×M)A=\operatorname{graph}L_{\mathcal{K}}\perp(\{0\}\times\mathcal{M}), where LKL_{\mathcal{K}} is self-adjoint in K\mathcal{K} and dom⁡LK=dom⁡T\operatorname{dom}L_{\mathcal{K}}=\operatorname{dom}T. Since W−1K=M∘W^{-1}\mathcal{K}=\mathcal{M}^{\circ} and W−1M=K∘W^{-1}\mathcal{M}=\mathcal{K}^{\circ}, the operator JK:=−iW−1LKJ_{\mathcal{K}}:=-iW^{-1}L_{\mathcal{K}} yields the asserted unique decomposition. Moreover, WJ0WJ_{0} acts as −iLK-iL_{\mathcal{K}} on dom⁡LK\operatorname{dom}L_{\mathcal{K}} and vanishes on M\mathcal{M}. Since U=K⊥M\mathcal{U}=\mathcal{K}\perp\mathcal{M}, this is a closed, densely defined, skew-adjoint operator, and hence so is J0J_{0} with respect to Γ\Gamma.

(b) The anti-dual adjoint q∗q^{*} maps Y∗\mathcal{Y}^{*} isometrically onto the annihilator of K\mathcal{K} in U∗\mathcal{U}^{*}. Thus j=Γ−1q∗j=\Gamma^{-1}q^{*} maps Y∗\mathcal{Y}^{*} isometrically onto K∘\mathcal{K}^{\circ}. The off-diagonal operator N:=[0j−q0]N:=\begin{bmatrix}0&j\\-q&0\end{bmatrix}, defined on U×Y∗\mathcal{U}\times\mathcal{Y}^{*}, is bounded and skew-adjoint with respect to diag⁡(Γ,ιY)\operatorname{diag}(\Gamma,\iota_{\mathcal{Y}}), since Γj=q∗\Gamma j=q^{*}. The block-diagonal operator diag⁡(J0,0)\operatorname{diag}(J_{0},0) is skew-adjoint as well. By the bounded-perturbation adjoint formula, J^=diag⁡(J0,0)+N\widehat{J}=\operatorname{diag}(J_{0},0)+N is therefore skew-adjoint on dom⁡J0×Y∗\operatorname{dom}J_{0}\times\mathcal{Y}^{*}, hence closed and densely defined.

The second component of the lifted equation gives qu=0qu=0. Together with u∈dom⁡J0u\in\operatorname{dom}J_{0}, this is equivalent to u∈dom⁡Tu\in\operatorname{dom}T; its first component is then v=JKu+jλv=J_{\mathcal{K}}u+j\lambda. By (a), this holds for a unique λ\lambda exactly when (u,v)∈T(u,v)\in T. Since JKu∈M∘J_{\mathcal{K}}u\in\mathcal{M}^{\circ} and jλ∈K∘j\lambda\in\mathcal{K}^{\circ}, orthogonal projection gives the multiplier formula and its norm estimate. □

The equation qu=0qu=0 enforces the closed subspace constraint u∈Ku\in\mathcal{K}, whereas the domain of J0J_{0} retains the possibly nonclosed domain of TT. The construction depends on the chosen Hilbert-space inner products through U=K⊥K⊥\mathcal{U}=\mathcal{K}\perp\mathcal{K}^{\perp}.

Remark 2.2 (Converse of the decomposition) Let K⊂U\mathcal{K}\subset\mathcal{U} be closed, and suppose that JK:dom⁡JK⊆K→(K∘)J_{\mathcal{K}}:\operatorname{dom}J_{\mathcal{K}}\subseteq\mathcal{K}\to(\mathcal{K}^{\circ}) has a zero extension to dom⁡JK⊥K⊥\operatorname{dom}J_{\mathcal{K}}\perp\mathcal{K}^{\perp} that is skew-adjoint with respect to Γ\Gamma. Then graph⁡JK⊥({0}×K∘)\operatorname{graph}J_{\mathcal{K}}\perp(\{0\}\times\mathcal{K}^{\circ}) is a skew-adjoint relation. This follows by reversing the construction in (a) of Theorem 2.1.

Remark 2.3 (Finite-dimensional case). In finite dimensions, K=dom⁡T\mathcal{K}=\operatorname{dom}T, and J0J_0 is everywhere defined. Identify V\mathcal{V} with U\mathcal{U} via W=RV−1ΓW=R_{\mathcal{V}}^{-1}\Gamma and identify Y\mathcal{Y} and Y∗\mathcal{Y}^{*} with M=K⊥\mathcal{M}=\mathcal{K}^{\perp} by the Hilbert metrics. In these coordinates, J^\widehat{J} becomes the skew-Hermitian block matrix

[WJ0incl⁡M−PM0].\begin{bmatrix} WJ_0 & \operatorname{incl}_{\mathcal{M}}\\ -P_{\mathcal{M}} & 0 \end{bmatrix}.

where incl⁡M:M→U\operatorname{incl}_{\mathcal{M}}:\mathcal{M}\to\mathcal{U} is the inclusion and PM:U→MP_{\mathcal{M}}:\mathcal{U}\to\mathcal{M} the orthogonal projection. In infinite dimensions, dom⁡T\operatorname{dom}T can be strictly smaller than K\mathcal{K}, so it cannot in general be replaced by its closure. ◇

Geometric port-Hamiltonian systems

State and port spaces

Let X\mathcal{X}, FR\mathcal{F}_{R}, and U\mathcal{U} be Hilbert spaces. The storage flow space is FS:=X\mathcal{F}_{S}:=\mathcal{X}, with ES:=X∗\mathcal{E}_{S}:=\mathcal{X}^{*}, so that the storage flow is x˙\dot{x}. We set ER:=FR∗\mathcal{E}_{R}:=\mathcal{F}_{R}^{*} for the internal resistive port, while u∈Uu\in\mathcal{U} and y∈U∗y\in\mathcal{U}^{*} are the external input and output. The storage and resistive efforts act by canonical duality; the external port uses the reversed pairing ⟨u,y⟩U,U∗\langle u,y\rangle_{\mathcal{U},\mathcal{U}^{*}} introduced above. The total flow and effort spaces are the Cartesian products

Ftot:=X×FR×U∗,Etot:=X∗×ER×U.\mathcal{F}_{\mathrm{tot}}:=\mathcal{X}\times\mathcal{F}_{R}\times\mathcal{U}^{*},\qquad\mathcal{E}_{\mathrm{tot}}:=\mathcal{X}^{*}\times\mathcal{E}_{R}\times\mathcal{U}.

With the canonical identification Ftot∗=X∗×ER×U∗∗\mathcal{F}_{\mathrm{tot}}^{*}=\mathcal{X}^{*}\times\mathcal{E}_{R}\times\mathcal{U}^{**}, the unitary duality map Γtot:Etot→Ftot∗\Gamma_{\mathrm{tot}}:\mathcal{E}_{\mathrm{tot}}\to\mathcal{F}_{\mathrm{tot}}^{*} is diag⁡(I,I,ιU)\operatorname{diag}(I,I,\iota_{\mathcal{U}}). For total efforts ee and flows ff, ordered by storage, resistive, and external components, the corresponding pairing is

⟨e,f⟩Γtot=⟨eS,fS⟩X∗,X+⟨eR,fR⟩ER,FR+⟨u,y⟩U,U∗.\langle e,f\rangle_{\Gamma_{\mathrm{tot}}}=\langle e_{S},f_{S}\rangle_{\mathcal{X}^{*},\mathcal{X}}+\langle e_{R},f_{R}\rangle_{\mathcal{E}_{R},\mathcal{F}_{R}}+\langle u,y\rangle_{\mathcal{U},\mathcal{U}^{*}}.

For the reverse direction, we use the unitary map Γtot∨:=Γtot∗ιFtot\Gamma_{\mathrm{tot}}^{\vee}:=\Gamma_{\mathrm{tot}}^{*}\iota_{\mathcal{F}_{\mathrm{tot}}} from Ftot\mathcal{F}_{\mathrm{tot}} to Etot∗\mathcal{E}_{\mathrm{tot}}^{*}. Its pairing satisfies ⟨f,e⟩Γtot∨=⟨e,f⟩Γtot\langle f,e\rangle_{\Gamma_{\mathrm{tot}}^{\vee}}=\langle e,f\rangle_{\Gamma_{\mathrm{tot}}}.

Dirac, Lagrangian, and resistive structures

The Dirac, Lagrangian, and resistive structures describe power-conserving interconnection, state–effort relations, and admissible internal port variables, respectively.

On Ftot×Etot\mathcal{F}_{\mathrm{tot}}\times\mathcal{E}_{\mathrm{tot}} define the Hermitian form

⟨⟨(f1,e1),(f2,e2)⟩⟩+:=⟨e1,f2⟩Γtot+⟨e2,f1⟩Γtot‾.\left\langle\left\langle(f_{1},e_{1}),(f_{2},e_{2})\right\rangle\right\rangle_{+}:=\langle e_{1},f_{2}\rangle_{\Gamma_{\mathrm{tot}}}+\overline{\langle e_{2},f_{1}\rangle_{\Gamma_{\mathrm{tot}}}}.

For a subspace S\mathcal{S} of this product space, let S⊥+\mathcal{S}^{\perp+} denote its orthogonal complement with respect to this form. A Dirac structure is a closed subspace D⊆Ftot×Etot\mathcal{D}\subseteq\mathcal{F}_{\mathrm{tot}}\times\mathcal{E}_{\mathrm{tot}} satisfying

D=D⊥+.\mathcal{D}=\mathcal{D}^{\perp+}.

By the definition of the relation adjoint in (3), this is equivalent to D∗Γtot=−D\mathcal{D}^{*_{\Gamma_{\mathrm{tot}}}}=-\mathcal{D} when D\mathcal{D} is viewed as a relation from flows to efforts. Thus the Dirac condition expresses maximality, beyond the power identity Re⁡⟨e,f⟩Γtot=0\operatorname{Re}\langle e,f\rangle_{\Gamma_{\mathrm{tot}}}=0 for (f,e)∈D(f,e)\in\mathcal{D}.

On X×X∗\mathcal{X} \times\mathcal{X}^{*} define the skew-Hermitian form

⟨ ⁣⟨(x1,e1),(x2,e2)⟩ ⁣⟩−:=⟨x1,e2⟩X,X∗−⟨e1,x2⟩X∗,X.\left\langle\!\left\langle(x_1,e_1),(x_2,e_2)\right\rangle\!\right\rangle_{-} := \langle x_1,e_2\rangle_{\mathcal{X},\mathcal{X}^{*}}-\langle e_1,x_2\rangle_{\mathcal{X}^{*},\mathcal{X}}.

For subspaces S⊆X×X∗\mathcal{S} \subseteq\mathcal{X} \times\mathcal{X}^{*}, write S⊥−\mathcal{S}^{\perp-} for the corresponding skew-orthogonal complement. A Lagrangian structure is a closed subspace L⊆X×X∗\mathcal{L} \subseteq\mathcal{X} \times\mathcal{X}^{*} satisfying L=L⊥−\mathcal{L} = \mathcal{L}^{\perp-}, equivalently a self-adjoint relation with respect to the canonical state–effort pairing.

The Lagrangian structure determines the stored energy H(x,eS):=12⟨eS,x⟩X∗,X\mathcal{H}(x,e_{\mathrm{S}}) := \frac{1}{2}\langle e_{\mathrm{S}},x\rangle_{\mathcal{X}^{*},\mathcal{X}} for (x,eS)∈L(x,e_{\mathrm{S}}) \in\mathcal{L}. By the Lagrangian property, this value is real and depends only on xx. Indeed, if (x,eS),(x,e~S)∈L(x,e_{\mathrm{S}}),(x,\tilde{e}_{\mathrm{S}}) \in\mathcal{L}, then (0,eS−e~S)∈L(0,e_{\mathrm{S}}-\tilde{e}_{\mathrm{S}}) \in\mathcal{L}; self-adjointness gives ⟨eS−e~S,x⟩X∗,X=0\langle e_{\mathrm{S}}-\tilde{e}_{\mathrm{S}},x\rangle_{\mathcal{X}^{*},\mathcal{X}} = 0. We do not assume that this energy is positive.

The resistive relation is a closed linear relation R⊆FR×ER\mathcal{R} \subseteq\mathcal{F}_{\mathrm{R}} \times\mathcal{E}_{\mathrm{R}}. We call it dissipative if

\operatorname{Re}\langle e_{\mathrm{R},f_{\mathrm{R}}\rangle}_{\mathcal{E}_{\mathrm{R}},\mathcal{F}_{\mathrm{R}}} \le0 \quad\text{for all } (f_{\mathrm{R}},e_{\mathrm{R}}) \in\mathcal{R}. \tag*{(5)}

The conversion results below require only that R\mathcal{R} be closed; dissipativity is relevant when interpreting the sign of its contribution to the energy balance.

Geometric systems and classical trajectories

The Dirac structure couples the storage flow x˙\dot{x} to the resistive port (fR,eR)(f_{\mathrm{R}},e_{\mathrm{R}}) and the external input uu and output yy without contributing net power. Resistive and external flows are oriented opposite to the storage flow, yielding the usual signs in the energy balance.

Definition 3.1 (Geometric port-Hamiltonian system) Let D⊆Ftot×Etot\mathcal{D} \subseteq\mathcal{F}_{\mathrm{tot}} \times\mathcal{E}_{\mathrm{tot}} be a Dirac structure, L⊆X×X∗\mathcal{L} \subseteq\mathcal{X} \times\mathcal{X}^{*} a Lagrangian structure, and R⊆FR×ER\mathcal{R} \subseteq\mathcal{F}_{\mathrm{R}} \times\mathcal{E}_{\mathrm{R}} a closed linear relation. The geometric port-Hamiltonian system associated with these relations is the differential-algebraic inclusion (1), with variables in the spaces specified above. A classical trajectory on an interval I⊆RI \subseteq\mathbb{R} consists of functions

x∈C1(I;X),fR∈C(I;FR),u∈C(I;U),eS∈C(I;X∗),eR∈C(I;ER),y∈C(I;U∗).\begin{aligned} x &\in C^{1}(I;\mathcal{X}),\quad f_{\mathrm{R}} \in C(I;\mathcal{F}_{\mathrm{R}}),\quad u \in C(I;\mathcal{U}),\\ e_{\mathrm{S}} &\in C(I;\mathcal{X}^{*}),\quad e_{\mathrm{R}} \in C(I;\mathcal{E}_{\mathrm{R}}),\quad y \in C(I;\mathcal{U}^{*}). \end{aligned}

satisfying (1) for every t∈It \in I.

The formulation is behavioral: the inclusion specifies the admissible trajectories without requiring the state derivative or the output to be uniquely determined by the other variables. To explain the signs, Dirac isotropy implies that the storage power equals the sum of resistive and external port powers. The Lagrangian property gives the chain rule ddtH(x(t),eS(t))=Re⁡⟨eS(t),x˙(t)⟩X∗,X\frac{\mathrm{d}}{\mathrm{d}t}\mathcal{H}(x(t),e_{\mathrm{S}}(t)) = \operatorname{Re}\langle e_{\mathrm{S}}(t),\dot{x}(t)\rangle_{\mathcal{X}^{*},\mathcal{X}}. Hence every classical trajectory satisfies

ddtH(x(t),eS(t))=Re⁡⟨eR(t),fR(t)⟩ER,FR+Re⁡⟨u(t),y(t)⟩U,U∗.\frac{\mathrm{d}}{\mathrm{d}t}\mathcal{H}(x(t),e_{\mathrm{S}}(t)) = \operatorname{Re}\langle e_{\mathrm{R}}(t),f_{\mathrm{R}}(t)\rangle_{\mathcal{E}_{\mathrm{R}},\mathcal{F}_{\mathrm{R}}} + \operatorname{Re}\langle u(t),y(t)\rangle_{\mathcal{U},\mathcal{U}^{*}}.

The resistive contribution is nonpositive when R\mathcal{R} is dissipative. The storage chain rule is justified under the stated regularity assumptions in § 4.5.

Between geometric systems and descriptor equations

To convert between (1) and (2), we first parametrize the storage and resistive relations, then treat the Dirac interconnection.

Parametrizing storage and resistive relations

Since L\mathcal{L} and R\mathcal{R} are closed linear relations, there exist Hilbert spaces ZL\mathcal{Z}_{\mathcal{L}} and ZR\mathcal{Z}_{\mathcal{R}} and bounded isomorphisms (EL,QL):ZL→L(E_{\mathcal{L}},Q_{\mathcal{L}}):\mathcal{Z}_{\mathcal{L}}\to\mathcal{L} and (FR,ER):ZR→R(F_{\mathcal{R}},E_{\mathcal{R}}):\mathcal{Z}_{\mathcal{R}}\to\mathcal{R} [21 Thm. 1.10.1 and Prop. 1.10.3]. Thus (x,eS)=(ELz,QLz)(x,e_{\mathcal{S}})=(E_{\mathcal{L}}z,Q_{\mathcal{L}}z) and (fR,eR)=(FRr,ERr)(f_{\mathcal{R}},e_{\mathcal{R}})=(F_{\mathcal{R}}r,E_{\mathcal{R}}r). For L=graph⁡A\mathcal{L}=\operatorname{graph}A with AA closed and possibly unbounded, one may take ZL=dom⁡A\mathcal{Z}_{\mathcal{L}}=\operatorname{dom}A with its graph norm and (EL,QL)=(I,A)(E_{\mathcal{L}},Q_{\mathcal{L}})=(I,A). Only ELzE_{\mathcal{L}}z, not necessarily zz in the graph norm, is differentiated in the descriptor equation.

The Dirac structure D\mathcal{D} also admits bounded image parametrizations. Here we seek a skew-adjoint operator representation of the interconnection on enlarged effort and flow spaces, using Theorem 2.1.

From operator-based descriptors to geometric systems

Proposition 4.1 (Geometric interpretation of a structured descriptor equation) Let J:dom⁡J⊆Etot→FtotJ:\operatorname{dom}J\subseteq\mathcal{E}_{\mathrm{tot}}\to\mathcal{F}_{\mathrm{tot}} be densely defined and skew-adjoint with respect to Γtot\Gamma_{\mathrm{tot}}. Let (EL,QL)(E_{\mathcal{L}},Q_{\mathcal{L}}) and (FR,ER)(F_{\mathcal{R}},E_{\mathcal{R}}) be bounded isomorphisms onto a Lagrangian structure L\mathcal{L} and a closed resistive relation R\mathcal{R}, respectively. Consider

(ddt(ELz(t))−FRr(t)−y(t))=Je(t),e(t)=(QLz(t)ERr(t)u(t))∈dom⁡J.\begin{pmatrix} \frac{\mathrm{d}}{\mathrm{d}t}(E_{\mathcal{L}}z(t))\\ -F_{\mathcal{R}}r(t)\\ -y(t) \end{pmatrix} =Je(t),\qquad e(t)= \begin{pmatrix} Q_{\mathcal{L}}z(t)\\ E_{\mathcal{R}}r(t)\\ u(t) \end{pmatrix} \in\operatorname{dom}J.

Then D:={(Je,e)∣e∈dom⁡J}\mathcal{D}:=\{(Je,e)\mid e\in\operatorname{dom}J\} is a Dirac structure. Under the regularity assumptions ELz∈C1(I;X)E_{\mathcal{L}}z\in C^1(I;\mathcal{X}) and continuity of zz, rr, uu, yy in their respective spaces, solutions of the displayed equation correspond bijectively to classical trajectories of (1), with x=ELzx=E_{\mathcal{L}}z, eS=QLze_{\mathcal{S}}=Q_{\mathcal{L}}z, fR=FRrf_{\mathcal{R}}=F_{\mathcal{R}}r, and eR=ERre_{\mathcal{R}}=E_{\mathcal{R}}r.

Proof The inverse of the skew-adjoint graph of JJ is a Dirac structure. Under the two bijective parametrizations, the displayed equation is precisely the Dirac inclusion in (1); the other inclusions hold by construction. □

From geometric Dirac structures to operators

For a general Dirac structure D\mathcal{D}, the inverse relation TD:=D−1⊆Etot×Ftot\mathcal{T}_{\mathcal{D}}:=\mathcal{D}^{-1}\subseteq\mathcal{E}_{\mathrm{tot}}\times\mathcal{F}_{\mathrm{tot}} need not be single-valued. With the reversed duality Γtot∨\Gamma_{\mathrm{tot}}^{\vee} from § 3, we have (D−1)tot∗Γtot∨=(D∗Γtot)−1=−D−1(\mathcal{D}^{-1})_{\mathrm{tot}}^{*_{\Gamma_{\mathrm{tot}}^{\vee}}}=(\mathcal{D}^{*_{\Gamma_{\mathrm{tot}}}})^{-1}=-\mathcal{D}^{-1}; see also [22 Prop. 2.8(i)]. Thus TD\mathcal{T}_{\mathcal{D}} is skew-adjoint and Theorem 2.1 applies. Put KD:=dom⁡TD=ran⁡D\mathcal{K}_{\mathcal{D}}:=\operatorname{dom}\mathcal{T}_{\mathcal{D}}=\operatorname{ran}\mathcal{D}. Its multivalued part is KD∘\mathcal{K}_{\mathcal{D}}^{\circ} by Theorem 2.1.

Corollary 4.2 (Operator representation of the Dirac structure) Let YD:=Etot/KD\mathcal{Y}_{\mathcal{D}}:=\mathcal{E}_{\mathrm{tot}}/\mathcal{K}_{\mathcal{D}} with quotient map qD:Etot→YDq_{\mathcal{D}}:\mathcal{E}_{\mathrm{tot}}\to\mathcal{Y}_{\mathcal{D}}, and set jD:=(Γtot∨)−1qD∗j_{\mathcal{D}}:=(\Gamma_{\mathrm{tot}}^{\vee})^{-1}q_{\mathcal{D}}^{*}. Applying (4) to TD\mathcal{T}_{\mathcal{D}} with Γtot∨\Gamma_{\mathrm{tot}}^{\vee} gives

a closed, densely defined operator J^D\widehat{J}_D with domain dom⁡J^D=(dom⁡TD⊥KD⊥)×YD∗\operatorname{dom}\widehat{J}_D = (\operatorname{dom}T_D \perp K_D^\perp) \times Y_D^*, skew-adjoint with respect to diag⁡(Γtot∨,ιYD)\operatorname{diag}(\Gamma_{\mathrm{tot}}^\vee,\iota_{Y_D}). Moreover, (f,e)∈D(f,e) \in D if and only if there is a unique λ∈YD∗\lambda\in Y_D^* satisfying the first equation below; the second gives its recovery:

J^D(eλ)=(f0)(6)\widehat{J}_D \begin{pmatrix} e\\ \lambda \end{pmatrix} = \begin{pmatrix} f\\ 0 \end{pmatrix} \tag*{(6)}
λ=ΛD(f,e):=J^D−1PKD∘f(7)\lambda= \Lambda_D(f,e) := \widehat{J}_D^{-1}P_{K_D^\circ}f \tag*{(7)}

where PKD∘P_{K_D^\circ} is the orthogonal projection in FtotF_{\mathrm{tot}}. The recovery map is bounded and satisfies ∥ΛD(f,e)∥YD∗≤∥f∥Ftot\lVert\Lambda_D(f,e)\rVert_{Y_D^*} \le\lVert f\rVert_{F_{\mathrm{tot}}}.

Proof. Apply Theorem 2.1 to T=D−1⊆Etot×FtotT=D^{-1}\subseteq E_{\mathrm{tot}}\times F_{\mathrm{tot}} with Γ=Γtot∨\Gamma=\Gamma_{\mathrm{tot}}^\vee.

Remark 4.3 (Effort constraints and operator domains) The quotient equation qDe=0q_D e=0 in (6) enforces e∈KDe\in K_D. Together with the domain of J^D\widehat{J}_D, it also enforces e∈dom⁡TDe\in\operatorname{dom}T_D, since (dom⁡TD⊥KD⊥)∩KD=dom⁡TD(\operatorname{dom}T_D\perp K_D^\perp)\cap K_D=\operatorname{dom}T_D. Thus the operator domain retains spatial regularity and boundary conditions independently of algebraic constraints. If KD=EtotK_D=E_{\mathrm{tot}}, no multiplier is needed (YD={0}Y_D=\{0\}), even when dom⁡TD\operatorname{dom}T_D is a proper dense subspace. Moreover, J^D\widehat{J}_D is bounded precisely when dom⁡TD=KD\operatorname{dom}T_D=K_D.

Construction and exact correspondence

Combining the parametrizations from Section 4.1 with (6), define the bounded map S:ES×ZR×YD∗×U→Etot×YD∗S:E_S\times Z_R\times Y_D^*\times U\to E_{\mathrm{tot}}\times Y_D^* by

S(erλu)=(eERruλ).S \begin{pmatrix} e\\ r\\ \lambda\\ u \end{pmatrix} = \begin{pmatrix} e\\ E_Rr\\ u\\ \lambda \end{pmatrix}.

Let Π\Pi exchange the last two components of Ftot×YDF_{\mathrm{tot}}\times Y_D. For ww with components e,r,λ,ue,r,\lambda,u, define

dom⁡M:=S−1(dom⁡J^D)(8)\operatorname{dom}M:=S^{-1}(\operatorname{dom}\widehat{J}_D) \tag*{(8)}
Mw:=ΠJ^DSw+(0FRr00)(w∈dom⁡M).(9)Mw:=\Pi\widehat{J}_DSw+ \begin{pmatrix} 0\\ F_Rr\\ 0\\ 0 \end{pmatrix} \qquad(w\in\operatorname{dom}M). \tag*{(9)}

The zero rows impose the resistive and Dirac constraints. The finite-dimensional construction with internal and external ports is given in [15 Thm. 15].

Theorem 4.4 (Equivalence of the geometric and descriptor formulations) Let DD be a Dirac structure, LL a Lagrangian structure, and RR a closed resistive relation, not necessarily dissipative. Choose bounded isomorphisms (EL,QL)(E_L,Q_L) and (FR,ER)(F_R,E_R) as in Section 4.1 and construct J^D\widehat{J}_D and MM as above. Then MM is closed.

Set Z:=ZC×ZR×YD∗\mathcal{Z} := \mathcal{Z}_{C} \times\mathcal{Z}_{R} \times\mathcal{Y}_{D}^{*}. Define bounded operators E:Z→FS×FR×YDE : \mathcal{Z} \to\mathcal{F}_{S} \times\mathcal{F}_{R} \times\mathcal{Y}_{D} and Q:Z→ES×ZR×YD∗Q : \mathcal{Z} \to\mathcal{E}_{S} \times\mathcal{Z}_{R} \times\mathcal{Y}_{D}^{*} by

Eζ=(ECz00),Qζ=(QLzrλ),where ζ=(zrλ).E\zeta= \begin{pmatrix} E_{C}z\\ 0\\ 0 \end{pmatrix}, \qquad Q\zeta= \begin{pmatrix} Q_{L}z\\ r\\ \lambda \end{pmatrix}, \qquad \text{where }\zeta= \begin{pmatrix} z\\ r\\ \lambda \end{pmatrix}.

Then x=ECzx = E_{C}z, eS=QLze_{S} = Q_{L}z, fR=FRrf_{R} = F_{R}r, and eR=ERre_{R} = E_{R}r define a bijection between classical geometric trajectories on II and triples ζ∈C(I;Z)\zeta\in C(I;\mathcal{Z}), u∈C(I;U)u \in C(I;U), y∈C(I;U∗)y \in C(I;U^{*}) such that ECz∈C1(I;X)E_{C}z \in C^{1}(I;\mathcal{X}), (QLz(t)u(t))∈dom⁡M\begin{pmatrix} Q_{L}z(t) \\ u(t) \end{pmatrix} \in\operatorname{dom} M, and (2) holds at every t∈It \in I. We call such triples classical descriptor trajectories. The external input and output are preserved.

Proof Closedness. Suppose wk∈dom⁡Mw_{k} \in\operatorname{dom} M, wk→ww_{k} \to w, and Mwk→gMw_{k} \to g. Since SS and FRF_{R} are bounded, Swk→SwSw_{k} \to Sw, and (9) implies that J^DSwk\widehat{J}_{D}Sw_{k} converges. Closedness of J^D\widehat{J}_{D} gives w∈dom⁡Mw \in\operatorname{dom} M, and (9) yields Mw=gMw = g.

Correspondence. For a classical descriptor trajectory, the parametrizations yield the regularity and membership in L\mathcal{L} and R\mathcal{R}. By (8)–(9) and (6), the descriptor equation reproduces the Dirac inclusion in (1). Conversely, a classical geometric trajectory determines unique continuous zz and rr by the bounded inverse parametrizations. The bounded map (7) supplies a unique continuous λ=ΛD(f,e)\lambda= \Lambda_{D}(f,e). These variables satisfy the descriptor equation and its domain condition. The uniqueness of z,r,λz,r,\lambda proves injectivity. □

Remark 4.5 (Non-dense descriptor domains) Although J^D\widehat{J}_{D} is densely defined, dom⁡M=S−1(dom⁡J^D)\operatorname{dom} M = S^{-1}(\operatorname{dom}\widehat{J}_{D}) need not be dense. ♢

Remark 4.6 (Port-free form) Without internal and external ports, the construction reduces to ddt(Eζ(t))=J^DQζ(t)\frac{\mathrm{d}}{\mathrm{d}t}(E\zeta(t)) = \widehat{J}_{D}Q\zeta(t), where E(zλ)=(ECz0)E\begin{pmatrix} z \\ \lambda\end{pmatrix} = \begin{pmatrix} E_{C}z \\ 0 \end{pmatrix} and Q(zλ)=(QLzλ)Q\begin{pmatrix} z \\ \lambda\end{pmatrix} = \begin{pmatrix} Q_{L}z \\ \lambda\end{pmatrix}. The interconnection operator remains closed and skew-adjoint. ♢

Power structure and balance

The descriptor representation retains the geometric power structure. Let ww have components e,r,λ,ue,r,\lambda,u, and let g1,…,g4g_{1},\ldots,g_{4} denote the storage, resistive, constraint, and external components of a descriptor flow gg. The bounded sesquilinear effort–flow pairing is defined by

⟨w,g⟩Θ:=⟨e,g1⟩X∗,X+⟨ERr,g2⟩ER,FR+⟨λ,g3⟩YD∗,YD+⟨u,g4⟩U,U∗.(10)\langle w,g\rangle_{\Theta} := \langle e,g_{1}\rangle_{\mathcal{X}^{*},\mathcal{X}} + \langle E_{R}r,g_{2}\rangle_{\mathcal{E}_{R},\mathcal{F}_{R}} + \langle\lambda,g_{3}\rangle_{\mathcal{Y}_{D}^{*},\mathcal{Y}_{D}} + \langle u,g_{4}\rangle_{U,U^{*}}. \tag*{(10)}

This pairing may be degenerate.

Proposition 4.7 (Port-Hamiltonian structure) Under the assumptions of Theorem 4.4:

(a) Dissipation identity. For every w∈dom⁡Mw \in\operatorname{dom} M,

Re⁡⟨w,Mw⟩Θ=Re⁡⟨ERr,FRr⟩ER,FR.\operatorname{Re}\langle w,Mw\rangle_{\Theta} = \operatorname{Re}\langle E_{R}r,F_{R}r\rangle_{\mathcal{E}_{R},\mathcal{F}_{R}}.

The resistive contribution is nonpositive for all $r if and only if RR is dissipative.

(b) Power balance. Every classical descriptor trajectory satisfies

ddtH(ELz(t),QLz(t))=Re⁡⟨ERr(t),FRr(t)⟩ER,FR+Re⁡⟨u(t),y(t)⟩U,U∗(11)\frac{\mathrm{d}}{\mathrm{d}t}\mathcal{H}(E_{\mathcal{L}}z(t),Q_{\mathcal{L}}z(t))=\operatorname{Re}\langle E_{\mathcal{R}}r(t),F_{\mathcal{R}}r(t)\rangle_{\mathcal{E}_{\mathcal{R}},\mathcal{F}_{\mathcal{R}}}+\operatorname{Re}\langle u(t),y(t)\rangle_{U,U^*} \tag*{(11)}

where t↦H(ELz(t),QLz(t))t\mapsto\mathcal{H}(E_{\mathcal{L}}z(t),Q_{\mathcal{L}}z(t)) is continuously differentiable.

Proof For (a), the pairing of ww with ΠJ^DSw\Pi\widehat{J}_{D}Sw is the conjugate of the pairing of SwSw with J^DSw\widehat{J}_{D}Sw, whose real part vanishes by skew-adjointness. The second term in (9) contributes ⟨ERr,FRr⟩ER,FR\langle E_{\mathcal{R}}r,F_{\mathcal{R}}r\rangle_{\mathcal{E}_{\mathcal{R}},\mathcal{F}_{\mathcal{R}}} by (10). The identity follows, and the sign characterization comes from the bijective parametrization of R\mathcal{R} and (5).

For (b), write x=ELzx=E_{\mathcal{L}}z and eS=QLze_{\mathcal{S}}=Q_{\mathcal{L}}z. The self-adjointness of L\mathcal{L} gives ddtH(x,eS)=Re⁡⟨eS,x˙⟩X∗,X\frac{\mathrm{d}}{\mathrm{d}t}\mathcal{H}(x,e_{\mathcal{S}})=\operatorname{Re}\langle e_{\mathcal{S}},\dot{x}\rangle_{X^*,X} without differentiating the effort. Indeed, for s,t∈Is,t\in I,

2(H(x(t),eS(t))−H(x(s),eS(s)))=Re⁡⟨eS(t)+eS(s),x(t)−x(s)⟩X∗,X.2\bigl(\mathcal{H}(x(t),e_{\mathcal{S}}(t))-\mathcal{H}(x(s),e_{\mathcal{S}}(s))\bigr)=\operatorname{Re}\langle e_{\mathcal{S}}(t)+e_{\mathcal{S}}(s),x(t)-x(s)\rangle_{X^*,X}.

Difference quotients justify the chain rule under the stated regularity. Finally, the descriptor equation and (a) yield (11). □

Example: An incompressible Kelvin–Voigt solid with a boundary port

We consider an incompressible Kelvin–Voigt solid with a traction boundary port, illustrating unbounded Dirac operators, infinite-dimensional constraints, and dissipation; see also [4 Sec. 3.2].

The mechanical model

Let Ω⊂Rd\Omega\subset\mathbb{R}^{d}, d∈{2,3}d\in\{2,3\}, be a bounded, connected Lipschitz domain. Its boundary is decomposed, up to sets of surface measure zero, into disjoint relatively open parts ΓD\Gamma_{D} and ΓP\Gamma_{P}, each of positive surface measure.

Let ρ,μ,η∈L∞(Ω)\rho,\mu,\eta\in L^{\infty}(\Omega) denote the mass density, shear modulus, and shear viscosity, respectively. These coefficients are real-valued and satisfy ρ≥ρ−>0\rho\geq\rho_{-}>0, μ≥μ−>0\mu\geq\mu_{-}>0, and η≥0\eta\geq0 almost everywhere. We denote the displacement by qq, the velocity by v=∂tqv=\partial_{t}q, and the momentum density by p=ρvp=\rho v. The linearized strain tensor is defined by ε(w):=12(∇w+∇w⊤)\varepsilon(w):=\frac{1}{2}(\nabla w+\nabla w^{\top}). The linearized incompressible Kelvin–Voigt equations are

∂tq(t)=v(t),∂tp(t)=div⁡T(t),p(t)=ρv(t),div⁡v(t)=0,T(t)=2με(q(t))+2ηε(v(t))−π(t)I.(12)\begin{aligned} \partial_{t}q(t)&=v(t), & \partial_{t}p(t)&=\operatorname{div}T(t),\\ p(t)&=\rho v(t), & \operatorname{div}v(t)&=0,\\ T(t)&=2\mu\varepsilon(q(t))+2\eta\varepsilon(v(t))-\pi(t)I. \tag*{(12)} \end{aligned}

Here π\pi is the pressure multiplier enforcing incompressibility, and TT is the total stress tensor, comprising elastic, viscous, and pressure contributions.

The boundary conditions and external output are

q(t)∣ΓD=0,T(t)n∣ΓP=u(t),y(t)=γPv(t).(13)q(t)|_{\Gamma_{D}}=0,\qquad T(t)n|_{\Gamma_{P}}=u(t),\qquad y(t)=\gamma_{P}v(t). \tag*{(13)}

Here nn is the outward unit normal, γP\gamma_{P} the trace on ΓP\Gamma_{P}, uu the prescribed traction, and yy the boundary velocity. The traction condition is understood weakly; without body forces, all external power enters through ΓP\Gamma_{P}.

Velocity, trace, and constraint spaces

Let γD\gamma_D denote the trace on ΓD\Gamma_D. Put

H:=L2(Ω;Cd),Hsym:=L2(Ω;Csymd×d),H := L^2(\Omega;\mathbb{C}^d), \qquad H_{\mathrm{sym}} := L^2(\Omega;\mathbb{C}_{\mathrm{sym}}^{d \times d}),
W:={w∈H1(Ω;Cd)∣γDw=0},V:={w∈W∣div⁡w=0}.W := \{w \in H^1(\Omega;\mathbb{C}^d) \mid\gamma_D w = 0\}, \qquad V := \{w \in W \mid\operatorname{div} w = 0\}.

Let G:=γPWG := \gamma_P W, equipped with the quotient norm

∥g∥G:=inf⁡w∈WγPw=g∥w∥H1(Ω;Cd).\lVert g \rVert_G := \inf_{\substack{w \in W \\ \gamma_P w = g}} \lVert w \rVert_{H^1(\Omega;\mathbb{C}^d)}.

Equivalently, GG consists of traces on ΓP\Gamma_P whose extensions by zero to ΓD\Gamma_D belong to H1/2(∂Ω;Cd)H^{1/2}(\partial\Omega;\mathbb{C}^d). Define the subspace of traces with zero total flux by

Gσ:={g∈G | ∫ΓPg⋅n dS=0}.G_\sigma:= \left\{g \in G \,\middle|\, \int_{\Gamma_P} g \cdot n\,\mathrm{d}S = 0\right\}.

The divergence theorem gives γPV⊆Gσ\gamma_P V \subseteq G_\sigma. Conversely, any g∈Gσg \in G_\sigma can be extended to w∈Ww \in W, whose divergence has zero mean. Using Bogovskii’s operator [24 Thm. III.3.1], this divergence can be corrected without changing the boundary trace, yielding a field in VV with trace gg. Hence γPV=Gσ\gamma_P V = G_\sigma. Since the total-flux functional g↦∫ΓPg⋅n dSg \mapsto\int_{\Gamma_P} g \cdot n\,\mathrm{d}S is nonzero on GG, its kernel GσG_\sigma has codimension one in GG.

We deliberately choose the full space GG, rather than GσG_\sigma, for the external effort and flow pairing U:=G∗U := G^*, U∗≅GU^* \cong G. Under the canonical identification G≅U∗G \cong U^*, the two pairings agree via ⟨u,g⟩U,G=⟨u,ιGg⟩U,U∗\langle u,g\rangle_{U,G} = \langle u,\iota_G g\rangle_{U,U^*} for all u∈Uu \in U, g∈Gg \in G. Here u∈Uu \in U is the boundary traction, whereas y=γPv∈Gσy = \gamma_P v \in G_\sigma is identified with an element of U∗U^*. Using GG rather than GσG_\sigma retains the constant-normal traction component and thus fixes the pressure level.

Define the solenoidal subspace and its orthogonal complement by

Hσ:=V⊥,H∇,P:={∇φ∣φ∈H1(Ω), γPφ=0}.H_\sigma:= V^\perp, \qquad H_{\nabla,P} := \{\nabla\varphi\mid\varphi\in H^1(\Omega),\ \gamma_P \varphi= 0\}.

The mixed-boundary Helmholtz decomposition reads

H=Hσ⊥H∇,P.(14)H = H_\sigma\perp H_{\nabla,P}. \tag*{(14)}

The divergence map div⁡:W→L2(Ω)\operatorname{div}: W \to L^2(\Omega) is surjective: Bogovskii’s operator provides all mean-zero divergences, while a field in WW with nonzero flux through ΓP\Gamma_P supplies the constant component. Since V=ker⁡(div⁡∣W)V = \ker(\operatorname{div}|_W), every h∈V⊥h \in V^\perp induces a well-defined functional on L2(Ω)L^2(\Omega) through div⁡w↦⟨h,w⟩H\operatorname{div} w \mapsto\langle h,w\rangle_H. By the open mapping and Riesz representation theorems, this functional is represented by some φ∈L2(Ω)\varphi\in L^2(\Omega). Integration by parts yields h=−∇φh = -\nabla\varphi, with φ∈H1(Ω)\varphi\in H^1(\Omega) and γPφ=0\gamma_P\varphi= 0. The converse follows from the Green identity, proving (14). The space H∇,PH_{\nabla,P} is infinite-dimensional.

By Korn’s first inequality for fields with vanishing trace on a nonempty Dirichlet boundary portion [25 Cor. 8.8], the symmetric-gradient operator

A:V⊂Hσ⟶Hsym,Aw:=ε(w),A : V \subset H_{\sigma} \longrightarrow H_{\mathrm{sym}}, \qquad Aw := \varepsilon(w),

is closed and densely defined. The trace γP:V→Gσ\gamma_P : V \to G_{\sigma} is unbounded for the HH-norm: otherwise the compact embedding V↪HσV \hookrightarrow H_{\sigma} would make this surjective trace map compact, contradicting dim⁡Gσ=∞\dim G_{\sigma} = \infty.

A boundary-coupled Dirac operator

To include the traction and both elastic and resistive stresses in a single adjoint, introduce

C:V⊂Hσ⟶Hsym×Hsym×G,Cw:=(AwAw−γPw).C : V \subset H_{\sigma} \longrightarrow H_{\mathrm{sym}} \times H_{\mathrm{sym}} \times G, \qquad Cw := \begin{pmatrix} Aw \\ Aw \\ -\gamma_P w \end{pmatrix}.

This operator is densely defined and closed: convergence in its graph norm implies convergence in WW by Korn’s inequality. We identify the L2L^2-based spaces HH and HsymH_{\mathrm{sym}} with their anti-duals through the Riesz isomorphisms. We also identify GG with U∗U^* by canonical biduality. With these identifications, the anti-dual adjoint C∗C^* is characterized by

⟨C∗(σsσRu),w⟩H=⟨σs+σR,Aw⟩Hsym−⟨u,γPw⟩U,G(w∈V).(15)\left\langle C^* \begin{pmatrix} \sigma_s \\ \sigma_R \\ u \end{pmatrix}, w \right\rangle_H = \langle\sigma_s + \sigma_R, Aw \rangle_{H_{\mathrm{sym}}} - \langle u, \gamma_P w \rangle_{U,G} \qquad(w \in V). \tag*{(15)}

The displayed column belongs to dom⁡C∗\operatorname{dom} C^* precisely when the right-hand side is bounded in the HH-norm on VV; (15) then determines its image uniquely. In particular, the domain condition couples the sum of the elastic and viscous stresses with the boundary traction. Neither stress need have a normal trace by itself.

For the geometric system choose the storage and resistive spaces and relations

X:=Hsym×H,L:=graph⁡diag⁡(2μ,ρ−1),\mathcal{X} := H_{\mathrm{sym}} \times H, \qquad\mathcal{L} := \operatorname{graph}\operatorname{diag}(2\mu,\rho^{-1}),
FR=ER:=Hsym,R:={(−r,2ηr)∣r∈Hsym}.\mathcal{F}_R = \mathcal{E}_R := H_{\mathrm{sym}}, \qquad\mathcal{R} := \{(-r,2\eta r) \mid r \in H_{\mathrm{sym}}\}.

The storage state is x=(sp)x = \begin{pmatrix} s \\ p \end{pmatrix}, with effort eS=(2μsρ−1p)e_S = \begin{pmatrix} 2\mu s \\ \rho^{-1}p \end{pmatrix}, where ss denotes the strain variable and pp the momentum density. The energy is

H(s,p)=12⟨2μs,s⟩Hsym+12⟨ρ−1p,p⟩H.\mathcal{H}(s,p) = \tfrac{1}{2}\langle2\mu s,s\rangle_{H_{\mathrm{sym}}} + \tfrac{1}{2}\langle\rho^{-1}p,p\rangle_H.

Both multipliers are bounded, and R\mathcal{R} is a closed dissipative relation because η≥0\eta\ge0.

The interconnection will restrict the velocity effort ep=ρ−1pe_p = \rho^{-1}p to VV. Since V‾H=Hσ\overline{V}^{H} = H_{\sigma} and dom⁡C∗\operatorname{dom} C^* is dense in Hsym×Hsym×UH_{\mathrm{sym}} \times H_{\mathrm{sym}} \times U, the closed effort subspace and its annihilator are

K:=Hsym×Hσ×Hsym×U,K∘:={0}×H∇,P×{0}×{0}.\mathcal{K} := H_{\mathrm{sym}} \times H_{\sigma} \times H_{\mathrm{sym}} \times U, \qquad\mathcal{K}^{\circ} := \{0\} \times H_{\nabla,P} \times\{0\} \times\{0\}.

The expression for K∘K^\circ follows from (14). On K\mathcal{K} define

dom⁡JK:={e∈K | ep∈V, (eseRu)∈dom⁡C∗},JK(esepeRu):=(Aep−C∗(eseRu)Aep−γPep).\operatorname{dom} J_{\mathcal{K}} := \left\{ e \in\mathcal{K} \,\middle|\, e_p \in V,\ \begin{pmatrix} e_s \\ e_R \\ u \end{pmatrix} \in\operatorname{dom} C^* \right\},\qquad J_{\mathcal{K}}\begin{pmatrix} e_s \\ e_p \\ e_R \\ u \end{pmatrix} := \begin{pmatrix} A e_p \\ -C^*\begin{pmatrix} e_s \\ e_R \\ u \end{pmatrix} \\ A e_p \\ -\gamma_P e_p \end{pmatrix}.

The last component is understood in U∗\mathcal{U}^*. After placing the velocity effort first and grouping the two stress efforts with the boundary traction, and reordering the flow coordinates accordingly, JKJ_{\mathcal{K}} takes the block form [0−C∗C0]\begin{bmatrix} 0 & -C^* \\ C & 0 \end{bmatrix} on dom⁡C×dom⁡C∗\operatorname{dom} C \times\operatorname{dom} C^*. Since CC is closed and densely defined, its bidual adjoint coincides with CC under the canonical identifications. Thus the block operator is skew-adjoint with respect to the stated duality. Its zero extension to dom⁡JK⊥K⊥\operatorname{dom} J_{\mathcal{K}} \perp\mathcal{K}^\perp is skew-adjoint as well. Consequently,

D:={(JKe+g,e)∣e∈dom⁡JK, g∈K∘}.\mathcal{D} := \left\{ (J_{\mathcal{K}}e+g,e) \mid e \in\operatorname{dom} J_{\mathcal{K}},\ g \in\mathcal{K}^\circ\right\}.

is a Dirac structure by Remark 2.2. The two restrictions K⊆Etot\mathcal{K} \subseteq\mathcal{E}_{\mathrm{tot}} and dom⁡JK⊆K\operatorname{dom} J_{\mathcal{K}} \subseteq\mathcal{K} exhibit the algebraic incompressibility constraint and the unbounded spatial and boundary-port operator domain, respectively.

Descriptor representation

By the Helmholtz decomposition (14), the quotient space YD=Etot/K\mathcal{Y}_{D}=\mathcal{E}_{\mathrm{tot}}/\mathcal{K} can be identified with H∇,PH_{\nabla,P}. We identify YD∗\mathcal{Y}_{D}^{*} with the same space by Riesz and write λ∈H∇,P\lambda\in H_{\nabla,P}. Let Pσ∈Lb(H)P_{\sigma} \in\mathcal{L}_{b}(H) denote the orthogonal projection onto HσH_{\sigma} from (14). The maps qDq_{\mathcal{D}} and jDj_{\mathcal{D}} are

qDe=(I−Pσ)ep,jDλ=(0λ00).q_{\mathcal{D}}e=(I-P_{\sigma})e_p,\qquad j_{\mathcal{D}}\lambda=\begin{pmatrix}0\\ \lambda\\0\\0\end{pmatrix}.

Using the storage parametrization EL=IE_{\mathcal{L}}=I, QL=diag⁡(2μ,ρ−1)Q_{\mathcal{L}}=\operatorname{diag}(2\mu,\rho^{-1}) and the resistive parametrization FRr=−rF_{Rr}=-r, ERr=2ηrE_{Rr}=2\eta r, the general construction (8)–(9) gives the descriptor system

\begin{pmatrix} \partial_t s(t)\\ \partial_t p(t)\\ 0\\ 0\\ -y(t) \end{pmatrix} = M\begin{pmatrix} 2\mu s(t)\\ \rho^{-1}p(t)\\ r(t)\\ \lambda(t)\\ u(t) \end{pmatrix}, \qquad M\begin{pmatrix} e_s\\ e_p\\ r\\ \lambda\\ u \end{pmatrix} = \begin{pmatrix} A P_{\sigma}e_p\\ -C^*\begin{pmatrix}e_s\\2\eta r\\u\end{pmatrix}+\lambda\\ A P_{\sigma}e_p-r\\ -(I-P_{\sigma})e_p\\ -\gamma_P P_{\sigma}e_p \end{pmatrix}. \tag*{(16)}\end{pmatrix}

With ww denoting the effort column in (16), the domain of MM on W:=Hsym×H×Hsym×H∇,P×U\mathcal{W}:=H_{\mathrm{sym}}\times H\times H_{\mathrm{sym}}\times H_{\nabla,P}\times\mathcal{U} is

dom⁡M:={w∈W | Pσep∈V, (es2ηru)∈dom⁡C∗}.\operatorname{dom} M := \left\{w\in\mathcal{W}\,\middle|\,P_{\sigma}e_p\in V,\ \begin{pmatrix}e_s\\2\eta r\\u\end{pmatrix}\in\operatorname{dom} C^*\right\}.

The third and fourth rows of (16) enforce r=Avr = Av and v=Pσv∈Vv = P_{\sigma}v \in V, respectively, with v=ρ−1pv = \rho^{-1}p. The fifth row yields y=γPvy = \gamma_{P}v. Thus the input is a boundary traction and the output a velocity trace, not a distributed body-force/velocity pair.

Equivalence and power balance

We first show that classical descriptor trajectories of (16) with compatible initial strain recover (12) and its boundary conditions. Set v=ρ−1pv = \rho^{-1}p and σ:=2μs+2ηr\sigma:= 2\mu s + 2\eta r. Since λ∈H∇,P=V⊥\lambda\in H_{\nabla,P} = V^{\perp}, the second row of (16) and (15) yield

⟨∂tp(t),w⟩H+⟨σ(t),ε(w)⟩Hsym−⟨u(t),γPw⟩U,G=0(w∈V).\langle\partial_t p(t), w\rangle_H + \langle\sigma(t), \varepsilon(w)\rangle_{H_{\mathrm{sym}}} - \langle u(t), \gamma_P w\rangle_{U,G} = 0 \qquad(w \in V).

To recover the pressure, observe that the left-hand side defines a bounded functional on WW that vanishes on V=ker⁡(div⁡∣W)V = \ker(\operatorname{div}|_W). Since div⁡:W→L2(Ω)\operatorname{div}: W \to L^2(\Omega) is bounded and surjective, the open mapping and Riesz representation theorems yield, for each t∈It \in I, a unique π(t)∈L2(Ω)\pi(t) \in L^2(\Omega) such that

⟨∂tp(t),w⟩H+⟨σ(t),ε(w)⟩Hsym−⟨u(t),γPw⟩U,G=⟨π(t),div⁡w⟩L2(Ω)(w∈W).(17)\langle\partial_t p(t), w\rangle_H + \langle\sigma(t), \varepsilon(w)\rangle_{H_{\mathrm{sym}}} - \langle u(t), \gamma_P w\rangle_{U,G} = \langle\pi(t), \operatorname{div} w\rangle_{L^2(\Omega)} \qquad(w \in W). \tag*{(17)}

Hence T(t):=σ(t)−π(t)IT(t) := \sigma(t) - \pi(t)I satisfies div⁡T(t)=∂tp(t)∈H\operatorname{div} T(t) = \partial_t p(t) \in H distributionally, so that T(t)∈H(div⁡;Ω)T(t) \in H(\operatorname{div};\Omega). The Green identity [26 Ch. I, Sec. 2.1, Thm. 2.2 and Cor. 2.1] defines its weak boundary traction tT(t)∈Ut_T(t) \in U by

⟨tT(t),γPw⟩U,G=⟨∂tp(t),w⟩H+⟨T(t),∇w⟩L2(Ω;Cd×d)(w∈W).\langle t_T(t), \gamma_P w\rangle_{U,G} = \langle\partial_t p(t), w\rangle_H + \langle T(t), \nabla w\rangle_{L^2(\Omega;\mathbb{C}^{d \times d})} \qquad(w \in W).

By (17), tT(t)=u(t)t_T(t) = u(t) for the complete stress, including pressure.

It remains to recover the displacement. Since v(t)∈Vv(t) \in V, v∈C(I;H)v \in C(I;H), and Av=r∈C(I;Hsym)Av = r \in C(I;H_{\mathrm{sym}}), Korn’s inequality gives v∈C(I;V)v \in C(I;V). The descriptor equations also imply ∂ts=Av=ε(v)\partial_t s = Av = \varepsilon(v). Hence, if s(t0)=ε(q0)s(t_0) = \varepsilon(q_0) for some q0∈Wq_0 \in W, then q(t):=q0+∫t0tv(τ) dτq(t) := q_0 + \int_{t_0}^{t} v(\tau)\,\mathrm{d}\tau satisfies ∂tq=v\partial_t q = v and s=ε(q)s = \varepsilon(q) for all t∈It \in I.

Conversely, suppose (12)–(13) hold pointwise in time with q∈C1(I;W)q \in C^1(I;W), p∈C1(I;H)p \in C^1(I;H), u∈C(I;U)u \in C(I;U), and π(t)∈L2(Ω)\pi(t) \in L^2(\Omega) for every t∈It \in I. Then v=∂tq=ρ−1p∈Vv = \partial_t q = \rho^{-1}p \in V, and the Green identity yields the domain condition for C∗C^* with elastic stress 2με(q)2\mu\varepsilon(q), viscous stress 2ηAv2\eta Av, and boundary traction uu. Thus a classical descriptor trajectory is obtained with s=ε(q)s = \varepsilon(q), r=Avr = Av, y=γPvy = \gamma_Pv, and λ=(I−Pσ)∂tp\lambda= (I - P_{\sigma})\partial_t p. The multiplier λ∈H∇,P\lambda\in H_{\nabla,P} has a unique H1H^1-potential vanishing on ΓP\Gamma_P, which need not coincide with the physical pressure π∈L2(Ω)\pi\in L^2(\Omega). The additive constant of the pressure is fixed by the prescribed traction. By Korn’s inequality, ε:W→Hsym\varepsilon: W \to H_{\mathrm{sym}} is injective with closed range. The constructions above are inverse to each other, establishing a bijection between classical solutions of (12)–(13) with the stated regularity and classical descriptor trajectories satisfying s(t0)∈ε(W)s(t_0) \in\varepsilon(W).

The power balance (11) takes the form

ddtH(s(t),p(t))=−∫Ω2η∣ε(v(t))∣2 dx+Re⁡⟨u(t),y(t)⟩U,U∗.\frac{\mathrm{d}}{\mathrm{d}t}\mathcal{H}(s(t),p(t)) = -\int_{\Omega} 2\eta|\varepsilon(v(t))|^2\,\mathrm{d}x + \operatorname{Re}\langle u(t),y(t)\rangle_{U,U^*}.

Statements and Declarations

Funding

This work was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project-ID 531152215 – CRC 1701.

Competing interests

The authors declare that they have no competing interests.

Not applicable.

Not applicable.

Data availability

No datasets were generated or analysed during this study.

Materials availability

Not applicable.

Code availability

Not applicable.

Authors’ contributions

Both authors contributed to the mathematical analysis and the preparation of the manuscript.

References

  1. [1]van der Schaft, A.J., Maschke, B.M.: Hamiltonian formulation of distributed-parameter systems with boundary energy flow. J. Geom. Phys. 42(1–2), 166–194 (2002) https://doi.org/10.1016/S0393-0440(01)00083-3
  2. [2]Le Gorrec, Y., Zwart, H., Maschke, B.: Dirac structures and boundary control systems associated with skew-symmetric differential operators. SIAM J. Control Optim. 44(5), 1864–1892 (2005) https://doi.org/10.1137/040611677
  3. [3]Jacob, B., Zwart, H.J.: Linear Port-Hamiltonian Systems on Infinite-dimensional Spaces. Operator Theory: Advances and Applications, vol. 223. Birkhäuser, Basel (2012). https://doi.org/10.1007/978-3-0348-0399-1
  4. [4]Brugnoli, A., Haine, G., Matignon, D.: Stokes-Dirac structures for distributed parameter port-Hamiltonian systems: An analytical viewpoint. Commun. Anal. Mech. 15(3), 362–387 (2023) https://doi.org/10.3934/cam.2023018
  5. [5]Philipp, F.M., Reis, T., Schaller, M.: Infinite-dimensional port-Hamiltonian systems: A system node approach. Math. Control Signals Syst. 37(3), 573–620 (2025) https://doi.org/10.1007/s00498-025-00412-0
  6. [6]Gernandt, H., Philipp, F.M., Preuster, T., Schaller, M.: Extension theory via boundary triplets for infinite-dimensional implicit port-Hamiltonian systems. Integr. Equ. Oper. Theory 98, 17 (2026) https://doi.org/10.1007/s00020-026-02841-1
  7. [7]Beattie, C., Mehrmann, V., Xu, H., Zwart, H.: Linear port-Hamiltonian descriptor systems. Math. Control Signals Syst. 30(4), 17 (2018) https://doi.org/10.1007/s00498-018-0223-3
  8. [8]Arlinskii, Y., Tretter, C.: Everything is possible for the domain intersection dom T ∩ dom T*. Adv. Math. 374, 107383 (2020) https://doi.org/10.1016/j.aim.2020.107383
  9. [9]Erbay, M., Jacob, B., Morris, K.: On the Weierstraß form of infinite-dimensional differential algebraic equations. J. Evol. Equ. 24, 73 (2024) https://doi.org/10.1007/s00028-024-01003-3
  10. [10]van der Schaft, A., Maschke, B.: Generalized port-Hamiltonian DAE systems. Syst. Control Lett. 121, 31–37 (2018) https://doi.org/10.1016/j.sysconle.2018.09.008
  11. [11]Gernandt, H., Haller, F.E., Reis, T.: A linear relation approach to port-Hamiltonian differential-algebraic equations. SIAM J. Matrix Anal. Appl. 42(2), 1011–1044 (2021) https://doi.org/10.1137/20M1371166
  12. [12]Mehrmann, V., van der Schaft, A.: Differential–algebraic systems with dissipative Hamiltonian structure. Math. Control Signals Syst. 35(3), 541–584 (2023) https://doi.org/10.1007/s00498-023-00349-2
  13. [13]van der Schaft, A., Mehrmann, V.: Linear port-Hamiltonian DAE systems revisited. Syst. Control Lett. 177, 105564 (2023) https://doi.org/10.1016/j.sysconle.2023.105564
  14. [14]Mehrmann, V., van der Schaft, A.: Nonlinear differential-algebraic systems with dissipative port-Hamiltonian structure. arXiv preprint arXiv:2610.11455 (2026). https://doi.org/10.48550/arXiv.2610.11455
  15. [15]Gernandt, H., Philipp, F.M., Preuster, T., Schaller, M.: On the equivalence of geometric and descriptor representations of linear port-Hamiltonian systems. In: Schwenninger, F.L., Waurick, M. (eds.) Systems Theory and PDEs. Trends in Mathematics, pp. 149–165. Birkhäuser, Cham (2024). https://doi.org/10.1007/978-3-031-64991-2_6
  16. [16]Caballeria, J., Ramirez, H., Lefèvre, L., Le Gorrec, Y.: On representations of linear port-Hamiltonian systems defined by Dirac and Lagrange structures. IFAC-PapersOnLine 60(1), 115–120 (2026) https://doi.org/10.1016/j.ifacol.2026.09.020
  17. [17]van der Schaft, A., Maschke, B.: Dirac and Lagrange algebraic constraints in nonlinear port-Hamiltonian systems. Vietnam J. Math. 48(4), 929–939 (2020) https://doi.org/10.1007/s10013-020-00419-x
  18. [18]Arens, R.: Operational calculus of linear relations. Pacific J. Math. 11(1), 9–23 (1961) https://doi.org/10.2140/pjm.1961.11.9
  19. [19]Cross, R.: Multivalued Linear Operators. Monographs and Textbooks in Pure and Applied Mathematics, vol. 213. Marcel Dekker, New York (1998)
  20. [20]Hassi, S., de Snoo, H.S.V., Szafraniec, F.H.: Componentwise and Cartesian decompositions of linear relations. Dissertationes Math. 465, 1–59 (2009) https://doi.org/10.4064/dm465-0-1
  21. [21]Behrndt, J., Hassi, S., de Snoo, H.: Boundary Value Problems, Weyl Functions, and Differential Operators. Monographs in Mathematics, vol. 108. Birkhäuser, Cham (2020). https://doi.org/10.1007/978-3-030-36714-5
  22. [22]Iftime, O.V., Roman, M., Sandovici, A.: A kernel representation of Dirac structures for infinite-dimensional systems. Math. Model. Nat. Phenom. 9(5), 295–308 (2014) https://doi.org/10.1051/mmnp/20149520
  23. [23]Hassi, S., Sebestyén, Z., de Snoo, H.S.V., Szafraniec, F.H.: A canonical decomposition for linear operators and linear relations. Acta Math. Hungar. 115(4), 281–307 (2007) https://doi.org/10.1007/s10474-007-5247-y
  24. [24]Galdi, G.P.: An Introduction to the Mathematical Theory of the Navier–Stokes Equations: Steady-State Problems, 2nd edn. Springer Monographs in Mathematics. Springer, New York, NY (2011). https://doi.org/10.1007/978-0-387-09620-9
  25. [25]Neff, P., Pauly, D., Witsch, K.-J.: On an extension of the first Korn inequality to incompatible tensor fields on domains of arbitrary dimensions. In: Modeling, Simulation and Optimization for Science and Technology. Computational Methods in Applied Sciences, vol. 34, pp. 139–159. Springer, Dordrecht (2014). https://doi.org/10.1007/978-94-017-9054-3_8
  26. [26]Girault, V., Raviart, P.-A.: Finite Element Approximation of the Navier–Stokes Equations. Lecture Notes in Mathematics, vol. 749. Springer, Berlin, Heidelberg (1979). https://doi.org/10.1007/BFb0063447

Paper details

vlnc:W_bs9svrye
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents