Port-Hamiltonian Systems Theory An Introductory Overview-Van Der Schaft, A., Jeltsema, D PDF
Port-Hamiltonian Systems Theory An Introductory Overview-Van Der Schaft, A., Jeltsema, D PDF
Port-Hamiltonian Systems Theory An Introductory Overview-Van Der Schaft, A., Jeltsema, D PDF
Theory: An Introductory
Overview
Port-Hamiltonian Systems
Theory: An Introductory
Overview
Dimitri Jeltsema
Delft Institute of Applied Mathematics
Delft University of Technology, the Netherlands
[email protected]
Boston — Delft
Foundations and Trends
R
in Systems and Control
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system,
or transmitted in any form or by any means, mechanical, photocopying, recording or otherwise,
without prior written permission of the publishers.
Photocopying. In the USA: This journal is registered at the Copyright Clearance Center, Inc., 222
Rosewood Drive, Danvers, MA 01923. Authorization to photocopy items for internal or personal
use, or the internal or personal use of specific clients, is granted by now Publishers Inc for users
registered with the Copyright Clearance Center (CCC). The ‘services’ for users can be found on
the internet at: www.copyright.com
For those organizations that have been granted a photocopy license, a separate system of pay-
ment has been arranged. Authorization does not extend to other kinds of copying, such as that
for general distribution, for advertising or promotional purposes, for creating new collective
works, or for resale. In the rest of the world: Permission to photocopy must be obtained from
the copyright owner. Please apply to now Publishers Inc., PO Box 1024, Hanover, MA 02339,
USA; Tel. +1 781 871 0245; www.nowpublishers.com; [email protected]
now Publishers Inc. has an exclusive license to publish this material worldwide. Permission
to use this content must be obtained from the copyright license holder. Please apply to now
Publishers, PO Box 179, 2600 AD Delft, The Netherlands, www.nowpublishers.com; e-mail:
[email protected]
Foundations and Trends
R
in
Systems and Control
Volume 1, Issue 2-3, 2014
Editorial Board
Editors-in-Chief
Editors
Topics
• Control applications
1 Introduction 3
1.1 Origins of port-Hamiltonian systems theory . . . . . . . 3
1.2 Summary of contents . . . . . . . . . . . . . . . . . . . 6
ix
x
Appendices 199
A Proofs 201
A.1 Proof of Proposition 2.1 . . . . . . . . . . . . . . . . . . 201
A.2 Proof of Proposition 2.2 . . . . . . . . . . . . . . . . . . 202
A.3 Extension of Proposition 2.1 . . . . . . . . . . . . . . . . 202
References 207
Abstract
3
4 Introduction
‘modeling for control’. Since the addition of control will anyway modify the dynami-
cal properties of the system the emphasis is on relatively simple models reflecting the
main dynamical characteristics of the system.
4
For specific physical domains (e.g., mechanical, electrical, chemical, hydraulic, ..)
there are many network modeling and simulation software packages available.
1.2. Summary of contents 7
2.1 Introduction
11
12 From modeling to port-Hamiltonian systems
fk
k m
em
fm
ek
ergy; see Figure 2.1. For the spring system the potential energy is ex-
pressed in terms of the elongation q of the spring. In case of a linear
spring, satisfying Hooke’s law, this potential energy is 12 kq 2 . This leads
to the system equations
q̇ = −fk ,
spring:
d 1 2
ek =
kq ,
dq 2
where1 −fk denotes the velocity of the endpoint of the spring (where
it is attached to the mass), and ek = kq denotes the spring force at this
endpoint.
For the mass system we obtain similar equations using the kinetic
1 2
energy 2m p expressed in terms of the momentum p of the mass
ṗ = −fm ,
mass: d
1 2
em =
p ,
dp 2m
p
where −fm denotes the force exerted on the mass, and em = m is the
velocity of the mass.
Finally, we couple the spring and the mass subsystems to each other
through the interconnection element using Newton’s third law (action
= −reaction) (
−fk = em ,
interconnection:
fm = ek ,
1
The reason for the minus sign in front of fk is that we want the product fk ek to
be incoming power with respect to the interconnection. This sign convention will be
adopted throughout.
2.1. Introduction 13
eS eR
storage D dissipation
fS fR
eP fP
where < e | f > denotes the duality product, that is, the linear func-
5
Usually one can take F = Rk . However, there are interesting cases where the
coordinate-free viewpoint is really rewarding, e.g., in rigid body dynamics the space
of flows is given as the space of twists F = se(3), the Lie algebra of the matrix group
SE(3), while the space of efforts is given by the space of wrenches E = se∗ (3), the
dual Lie algebra. We refer to Chapter 3 for some developments in this direction.
6
The definition E = F ∗ for the effort space is in some sense the minimal required
structure. All definitions and results directly extend to the case that F has an inner-
product structure. In this case we may take E = F with the duality product < e | f >
replaced by the inner product he, f i.
16 From modeling to port-Hamiltonian systems
2. dim D = dim F.
Property (1) corresponds to power-conservation, and expresses the
fact that the total power entering (or leaving) a Dirac structure is zero.
It can be shown that the maximal dimension of any subspace D ⊂ F × E
satisfying Property (1) is equal to dim F. Instead of proving this di-
rectly, we will give an equivalent definition of a Dirac structure from
which this claim immediately follows. Furthermore, this equivalent
definition of a Dirac structure has the advantage that it generalizes to
the case of an infinite-dimensional linear space F, leading to the defini-
tion of an infinite-dimensional Dirac structure. This will be instrumen-
tal in the definition of a distributed-parameter port-Hamiltonian system
in Chapter 14.
In order to give this equivalent characterization of a Dirac struc-
ture, we look more closely at the geometric structure of the total space
of flow and effort variables F × E. Related to the definition of power,
there exists a canonically defined bilinear form ≪, ≫ on the space F ×E,
defined as
≪ (f a , ea ), (f b , eb ) ≫:=< ea | f b > + < eb | f a >, (2.4)
with (f a , ea ), (f b , eb ) ∈ F × E. Note that this bilinear form is indefi-
nite, that is, ≪ (f, e), (f, e) ≫ may be positive or negative. It is non-
degenerate in the sense that ≪ (f a , ea ), (f b , eb ) ≫= 0 for all (f b , eb )
implies that (f a , ea ) = 0.
Proposition 2.1 (Courant (1990); Dorfman (1993)). A Dirac structure
on F × E is a subspace D ⊂ F × E such that
D = D ⊥⊥ , (2.5)
2.2. Port-based modeling and Dirac structures 17
is a Dirac structure.
is a Dirac structure.
f1 = βe2 ,
(2.11)
βe1 = −f2 ,
is defining a Dirac structure D, and the same holds for the ideal flow
constraint n o
D := (f, e) | f = 0 .
Finally, the equations of a so-called k-dimensional 0-junction (termi-
nology from bond graph theory, cf. Paynter (1960); Breedveld (1984))
e1 = e2 = · · · = ek , f1 + f2 + · · · + fk = 0,
f1 = f2 = · · · = fk , e1 + e2 + · · · + ek = 0,
BI = 0, (2.13)
V = B T λ, (2.14)
(V a )T I b + (V b )T I a = 0,
AT V = 0, F = Aλ, (2.16)
where the columns of A form a basis for the space of allowed reaction
forces, and λ is the vector of reaction force Lagrange multipliers.
where Fi are the impressed forces, and δqi denotes the virtual displace-
ments that are compatible with the kinematic constraints of the sys-
P
tem. The expression ni=1 Fi δqi equals the infinitesimal work due to
the impressed forces and an infinitesimal displacement. If the kine-
matic constraints of the system are given as AT δq = 0, with δq =
8
However, for 3D mechanical systems the matrix A will often depend on the con-
figuration coordinates; thus defining a Dirac structure on a manifold, see Chapter
3.
22 From modeling to port-Hamiltonian systems
(δq1 , · · · , δqn )T , then it follows that the impressed forces are given as
F = Aλ, with F = (F1 , · · · , Fn )T , as in the previous subsection; see
Chapter 3 for more details. We conclude that, like in the case of Kirch-
hoff’s laws in the electrical domain, the principle of virtual work can
be formulated as defining a separable Dirac structure on the product
of the space of virtual displacements and impressed forces.
Originally, the principle of virtual work (2.17) is formulated as an
equilibrium condition. Indeed, a system with configuration coordinates
q = (q1 , q2 , . . . , qn )T , which is subject to forces F (q), is at equilibrium
P
q̄ if the virtual work ni=1 Fi (q̄)δqi corresponding to any admissible
virtual displacement δq from q̄ is equal to zero. In d’Alembert’s prin-
ciple this was extended by adding the inertial forces ṗ to the impressed
forces. This can be interpreted as linking the Dirac structure to energy-
storage (in this case, kinetic energy).
d ∂H ∂T H
H =< (x) | ẋ >= (x)ẋ. (2.18)
dt ∂x ∂x
The interconnection of the energy-storing elements to the storage port
(fS , eS ) of the Dirac structure is accomplished by setting
∂H
fS = −ẋ and eS = (x). (2.19)
∂x
Hence, the power-balance (2.18) can be also written as
d ∂T H
H= (x)ẋ = −eTS fS . (2.20)
dt ∂x
Remark 2.2. The minus sign in (2.19) is inserted in order to have a
T
consistent power flow convention: ∂∂xH (x)ẋ is the power flowing into
the energy-storing elements, whereas eTS fS is the power flowing into
the Dirac structure.
The external port (fP , eP ) models the interaction of the system with
its environment. This comprises different situations. One are the port
variables which are accessible for controller action. Another type of
external port variables corresponds to an interaction port. Typical ex-
ample of the latter is a controlled robotic system interacting with its
physical environment. Still another type of external port variables are
variables corresponding to sources. For example, in an electrical circuit
with voltage source the input is the voltage of the source, while the
current through the source is the (resulting) output variable.
Taking the external port into account the power-balance (2.22) ex-
tends to
eTS fS + eTR fR + eTP fP = 0, (2.26)
whereby (2.23) extends to
d
H = eTR fR + eTP fP ≤ eTP fP , (2.27)
dt
26 From modeling to port-Hamiltonian systems
since eTR fR ≤ 0. This inequality expresses the basic fact that the in-
crease of the internally stored energy (the Hamiltonian) is always less
than or equal to the externally supplied power.
H : X → R,
D ⊂ Tx X × Tx∗ X × FR × ER × FP × EP ,
φ̇i = −Vi ,
dHi
Ii = (ϕi ),
dϕi
for i = 1, 2, where Ii are the currents through the inductors with
flux-linkages φi , magnetic energy Hi (φ1 ), and −Vi the voltages across
2.6. Port-Hamiltonian dynamics 27
L1 L2
− ϕ1 + + ϕ2 −
R
R R
ϕ̇1 = − ϕ1 − ϕ2 ,
L1 L2
R R
ϕ̇2 = − ϕ1 − ϕ2 .
L1 L2
In Chapter 12, we will elaborate on port-Hamiltonian models for gen-
eral RLC circuits.
p2 ϕ2
H(q, p, ϕ) = mgq + + ,
2m 2L(q)
where the inductance L(q) depends on the height q. In fact, the mag-
ϕ2
netic energy 2L(q) depends both on the flux ϕ and the mechanical vari-
able q. As a result the right-hand side of the second equation (describ-
ing the evolution of the mechanical momentum variable p) depends
on the magnetic variable ϕ, and conversely the right-hand side of the
2.6. Port-Hamiltonian dynamics 29
R L
+
ω
V K J
_ I τ
b
given as
VR = −RI, τb = −bω,
with R, b > 0, where τb is a damping torque. Furthermore, the equa-
tions of the gyrator (converting magnetic power into mechanical, or
conversely) are
VK = −Kω, τ = KI.
for a certain positive constant K (the gyrator constant). Finally, the
subsystems are interconnected by the equations
VL + VR + VK + V = 0, τJ + τb + τ = 0.
# ϕ
" # " " #
ϕ̇ −R −K
L + 1 V,
=
ṗ K −b p 0
ϕ J (2.30)
h i L
I= 1 0 p.
J
While in the previous example the coupling between the mechani-
cal and magnetic domain was provided by the Hamiltonian (depend-
ing in a non-separable way on the mechanical state variables and the
magnetic state variable), in this example the inter-domain coupling is
given by the Dirac structure (through the gyrator constant K).
C1 C2
Q2
Q1 +
ϕ L
−
thermore, Qi are the charges stored in the capacitors and are regarded
as basic state variables.11 Similarly, the linear inductor is described by
the dynamical equations
ϕ̇ = −VL ,
ϕ
IL = ,
L
where IL is the current through the inductor, and VL is the voltage
across the inductor. Here, the (magnetic) flux-linkage ϕ is taken as the
state variable of the inductor, and L denotes its inductance.
Parallel interconnection of these three subsystems by Kirchhoff’s
laws amounts to the interconnection equations
V1 = V2 = VL , I1 + I2 + IL = 0,
Q1 Q2
= , (2.31)
C1 C2
11
In the port-Hamiltonian formulation there is a clear preference for taking the
charges Qi to be the state variables instead of the voltages Vi . This is due to the fact
that the charges satisfy a conservation law, while the voltages do not. Furthermore,
although the introduction of charge variables comes at the expense of extra variables,
it will turn out to be very advantageous from a geometric point of view as well: the
charge variables live in the ’right’ state space.
2.7. Port-Hamiltonian differential-algebraic equations 33
ẋ = Sv(x),
h iT
with S the stoichiometric matrix, and v(x) = v1 (x) · · · vr (x) ∈
Rr the vector of reaction rates. We assume that v(x) is given by mass ac-
tion kinetics; the most basic way of modeling reaction rates. Following
van der Schaft et al. (2013) we will show how, under the assumption
of existence of a thermodynamic equilibrium, the dynamics of the re-
action network can be naturally modeled as a port-Hamiltonian sys-
tem, with Hamiltonian given by the free Gibbs’ energy.
In order to do so we first need to introduce some concepts and ter-
minology. The collection of all the different left- and right-hand sides
of the reactions are called the chemical complexes of the reaction net-
work, or briefly, the complexes. Denoting the number of complexes by
2.8. Detailed-balanced chemical reaction networks 35
Eq. (2.35) can be rewritten in the following way. Let ZSj and ZPj
denote the columns of Z corresponding to the substrate complex Sj
and the product complex Sj of the j-th reaction. Defining the mapping
Ln : Rc+ → Rc as the component-wise natural logarithm, (2.35) takes
the form
vj (x) = kj+ exp ZSTj Ln(x) − kj− exp ZPTj Ln(x) . (2.36)
kj+
Kjeq := (2.38)
kj−
+ −
x∗ ∈ Rm
+ if and only if kj > 0, kj > 0, for all j = 1, . . . , r, and further-
more
Ln (K eq ) ∈ im S T (2.41)
It also follows that once a thermodynamic equilibrium x∗ is given, the
set of all thermodynamic equilibria is given by
E := {x∗∗ ∈ Rm T ∗∗ T ∗
+ | S Ln (x ) = S Ln (x )} (2.42)
for j = 1, · · · , r. Then the mass action reaction rate (2.36) of the j-th
reaction can be rewritten as
∗ x x
vj (x) = κj (x ) exp ZSTj Ln − exp ZPTj Ln ,
x∗ x∗
where for any vectors x, z ∈ Rm the quotient vector xz ∈ Rm is defined
element-wise. Defining the r × r diagonal matrix of conductances as
K := diag κ1 (x∗ ), · · · , κr (x∗ ) , (2.44)
it follows that the mass action reaction rate vector v(x) of a balanced
reaction network equals
T x T
v(x) = −KB Exp Z Ln ,
x∗
and thus the dynamics of a balanced reaction network takes the form
x
ẋ = −ZBKB T Exp Z T Ln , K > 0. (2.45)
x∗
The matrix L := BKB T in (2.45) defines a weighted Laplacian ma-
trix for the complex graph, with weights given by the conductances
κ1 (x∗ ), · · · , κr (x∗ ). Note that K, and therefore the Laplacian matrix
38 From modeling to port-Hamiltonian systems
Indeed, by using the properties of the Laplacian matrix BKB T and the
fact that the exponential function is strictly increasing, it can be shown
that van der Schaft et al. (2013)
ria, and a Lyapunov analysis using the Gibbs’ free energy (the Hamil-
tonian) shows that that starting from any initial state in the positive
orthant the system will converge to a unique thermodynamic equilib-
rium (at least under the assumption of persistence of the reaction net-
work: the vector of concentrations does not approach the boundary of
the positive orthant Rm 12
+ ), cf. van der Schaft et al. (2013) for details.
12
For an extension of these results to complex-balanced mass action kinetics reaction
networks we refer to Rao et al. (2013).
3
Port-Hamiltonian systems on manifolds
For quite a few system classes, in particular those with 3-D mechan-
ical components, the Dirac structure is modulated by the state vari-
ables. Furthermore, the state space X is not necessarily anymore a lin-
ear space but instead a (differentiable1 ) manifold. As before, the flows
fS = −ẋ corresponding to energy-storage are elements of the tangent
space Tx X at the state x ∈ X , while the efforts eS = ∂H
∂x (x) are elements
∗
of the co-tangent space Tx X . The modulation of the Dirac structure is
usually intimately related to the underlying geometry of the system.
1
’Manifold’ will always mean ’differentiable manifold’.
41
42 Port-Hamiltonian systems on manifolds
where B(q)u are the external forces (controls) applied to the system,
for some n × m matrix B(q), while A(q)λ are the constraint forces. The
Lagrange multipliers λ(t) are uniquely determined by the require-
ment that the constraints AT (q(t))q̇(t) = 0 have to be satisfied for all
times t.
Defining the generalized momenta
∂L
p= = M (q)q̇, (3.7)
∂ q̇
the constrained Euler-Lagrange equations (3.6) transform into con-
strained Hamiltonian equations
∂H
q̇ = (q, p)
∂p
∂H
ṗ = − (q, p) + A(q)λ + B(q)u
∂q
(3.8)
∂H
y = B T (q) (q, p)
∂p
∂H
0 = AT (q) (q, p)
∂p
44 Port-Hamiltonian systems on manifolds
l1
φ1 y1
g
x1
m1 y2
l2
φ2
x2
m2
Figure 3.1: Double pendulum.
θ
(x, y)
ϕ
x
The total energy is H = 12 p2x + 21 p2y + 12 p2θ + 12 p2ϕ . The rolling con-
straints are ẋ = θ̇ cos ϕ and ẏ = θ̇ sin ϕ, i.e., rolling without slipping,
which can be written in the form (3.3) by defining
" #
T 1 0 − cos φ 0
A (x, y, θ, φ) = . (3.11)
0 1 − sin φ 0
D(x) ⊂ Tx X × Tx∗ X
3.2 Integrability
< LX1 α2 | X3 > + < LX2 α3 | X1 > + < LX3 α1 | X2 >= 0, (3.22)
We call x0 a regular point for the Dirac structure if both the distribution
GD and the co-distribution PD have constant dimension around x0 .
If the Dirac structure is integrable and x0 is a regular point, then,
again by a version of Darboux’s theorem, we can choose local coor-
dinates x = (q, p, r, s) for X (with dim q = dim p), such that, in the
resulting bases for (fq , fp , fr , fs ) for Tx X and (eq , ep , er , es ) for Tx∗ X ,
the Dirac structure on this coordinate neighborhood is given as
(
fq = −ep , fp = eq
(3.23)
fr = 0, es = 0.
(1994); Dalsmo & van der Schaft (1999). Here we only state a slightly
simplified version of this result, detailed in Dalsmo & van der Schaft
(1999). We assume that the actuation matrix B(q) has the special form
(often encountered in examples) that its j-th column is given as
0
∂Cj .
(q)
∂q
for some function Cj (q) only depending on the configuration variables
q, j = 1, . . . , m. In this case, the Dirac structure D is integrable if and
only if the kinematic constraints are holonomic. Thus the Dirac structure
corresponding to the Double pendulum example is integrable, while
the Dirac structure corresponding to the Rolling euro example is not
integrable.
4
Input-state-output port-Hamiltonian systems
53
54 Input-state-output port-Hamiltonian systems
ables x ∈ X )
That is, any pair (fR , eR ) satisfying (4.1) also satisfies (4.4) for some λ,
and conversely, every (fR , eR ) satisfying (4.4) for some λ also satisfies
(4.1). Hence by (4.2) for all (fR , eR ) satisfying (4.1)
T
eTR fR = − RfT (x)λ ReT (x)λ = −λT Rf (x)ReT (x)λ ≤ 0,
Then,
(D(x) ◦ R(x))⊥⊥ = D(x) ◦ (−R) (x).
Then, by factorizing K(x) into its skew-symmetric part Kss (x) and its
symmetric part Ks (x) we have
T
K(x) = Kss (x) + Ks (x), Kss (x) = −Kss (x), KsT (x) = Ks (x) ≥ 0.
(4.6)
implying Ks (x) ≥ 0.
∂H
Hence, together with the energy-storage relations eS = ∂x (x), fS =
−ẋ, we obtain the port-Hamiltonian system
∂H
ẋ = [J(x) − R(x)] (x) + [g(x) − P (x)] u,
∂x
(4.9)
h i ∂H
T
y = g (x) + P (x) (x) + [M (x) + S(x)] u,
∂x
called an input-state-output port-Hamiltonian system with feedthrough
term. Along the trajectories of the system we recover the fundamen-
tal power-balance
" #" #
d h i R(x) P (x) eS
H(x) = −eTS fS = uT y − eTS uT ≤ uT y.
dt P T (x) S(x) u
I + ϕ −
+ Q C
V R1 R2
R3
−
Hamiltonian form
R2 R3 R3 ∂H
− −
" # " #
ϕ̇ R + R R + R ∂ϕ 1
2 3 2 3
= + V
Q̇ R3 1 ∂H 0
−
R2 + R3 R2 + R3 ∂Q
∂H
h i ∂ϕ V
I= 1 0 + ,
∂H R1
∂Q
1 2 1
with H(ϕ, Q) = 2L ϕ + 2C Q2 . This defines a port-Hamiltonian input-
state-output system with feedthrough specified by
R3 R2 R3
0 − 0
R2 + R3
R + R
2 3
J = , R = ,
R3 1
0 0
R2 + R3 R2 + R3
M = 0, P = 0, and S = 1/R1 .
(2009).
ated to the memristive port, and let xfM ∈ XM and xeM ∈ XM ∗ the
∂ΦM
eM = −RM (xfM )fM , RM (xfM ) = (xfM ),
∂xfM
1
From a mathematical perspective, the behavior of a resistor, inductor, and a ca-
pacitor, whether linear or nonlinear, is described by relationships between two of the
four basic electrical variables: voltage, current, charge, and flux linkage. Indeed, a
resistor is described by the relationship of current and voltage; an inductor by that
of current and flux linkage, and a capacitor by that of voltage and charge. But what
about the relationship between charge and flux linkage? This missing relationship
defines the memristor.
4.4. Relation with classical Hamiltonian systems 59
Note that the memory effect stems from the fact that the memristor
‘remembers’ the amount of flow that has passed through it via ẋfM =
fM .
Now, locally around xfM ∈ XM , the memristive structure M de-
fines a port-Hamiltonian system with a direct feedthrough term. In-
deed, let HM : XM → R be the zero function, then the dynamics on M
locally take the from
ẋfM = fM ,
∂HM (4.11)
eM = (xfM ) − RM (xfM )fM .
∂xfM
The fact that HM (xfM ) = 0, for all xfM ∈ XM , together with the fact
that eM ≡ 0 whenever fM ≡ 0 regardless of the internal state xfM ,
clearly underscores the ‘no energy discharge property’ as discussed in
Chua (1971). A dual representation can be obtained starting from the
xeM –controlled relationship xfM = −Φ∗M (xeM ).
The concept of the memristor and its generalizations can be use-
ful in modelling a wide variety of phenomena, including thermistors,
Josephson junctions, discharge tubes, and even ionic systems like the
Hodgkin-Huxley model of a neuron; see Jeltsema & van der Schaft
(2010) and Jeltsema & Doria (2012) for a further discussion, some il-
lustrative examples, and the inclusion of so-called meminductors and
memcapacitors, the memory equivalents of inductors and capacitors,
in the port-Hamiltonian framework.
state space of (4.16) with local coordinates (q, p) is usually called the
phase space.
The following power-balance immediately follows from (4.15):
d ∂T H ∂T H ∂T H
H= (q, p)q̇ + (q, p)ṗ = (q, p)τ = q̇ T τ, (4.17)
dt ∂q ∂p ∂p
expressing that the increase in energy of the system is equal to the sup-
plied work (conservation of energy). Hence by defining the input to be
d
u = τ and the output to be y := q̇ we obtain dt H = y T u. In particular, if
the Hamiltonian H(q, p) is assumed to be the sum of a positive kinetic
energy and a potential energy which is nonnegative, then it follows
that the system (4.15) with inputs u = τ and outputs y := q̇ is passive
(in fact, lossless) with storage function H(q, p).
System (4.15) with inputs u = τ and outputs y := q̇ is an exam-
ple of a Hamiltonian system with collocated inputs and outputs, which
more generally is given in the following form
∂H
q̇ = (q, p),
∂p
∂H
ṗ = − (q, p) + B(q)u, u ∈ Rm , (4.18)
∂q
∂H
y = B T (q) (q, p) (= B T (q)q̇), y ∈ Rm ,
∂p
where q = (q1 , . . . , qk )T and p = (p1 , . . . , pk )T , and B(q) is the in-
put force matrix, with B(q)u denoting the generalized forces resulting
from the control inputs u ∈ Rm . (In case m < k we speak of an un-
deractuated system. If m = k and the matrix B(q) is invertible for all q,
then the system is fully actuated.) Again we obtain the energy balance
dH
(q(t), p(t)) = uT (t)y(t). (4.19)
dt
A further generalization of the class of Hamiltonian systems (4.18)
with collocated inputs and outputs consists in considering systems
which are described in local coordinates as
∂H
ẋ = J(x) (x) + g(x)u, x ∈ X , u ∈ Rm
∂x (4.20)
∂H
y = gT (x) (x), y ∈ Rm ,
∂x
62 Input-state-output port-Hamiltonian systems
63
64 Representations of Dirac structures
Then, cf. Bloch & Crouch (1999), the matrix [F1 | E2 ] is invertible, and
(" # " # " # " #)
f1 e1 f
1 e1
D= ∈ F, ∈E =J ,
f2 e2 e2 f2
69
70 Interconnection of port-Hamiltonian systems
f1 fA fB f3
DA DB
e1 eA eB e3
equal the outgoing power from DB . Thus we cannot simply equate the
flows fA and fB and the efforts eA and eB , but instead we define the
interconnection constraints as
fA = −fB ∈ F2 , eA = eB ∈ E2 . (6.1)
Then,
F = LA F1 | LB F3 , E = LA E1 | LB E3 , (6.4)
is a relaxed matrix kernel/image representation of DA ◦ DB .
Separable Dirac structures turn out to have the following special
compositional property (van der Schaft & Maschke (2013)):
Proposition 6.1. Let DA ⊂ F2 ×E1 ×F2 ×E2 and DB ⊂ F2 ×E2 ×F3 ×E3
be two separable Dirac structures given as
⊥ ⊥
DA = KA × KA , DB = KB × KB
where KA ⊂ F1 × F2 and KB ⊂ F2 × F3 Define the composition
n
KA ◦ KB = (f1 , f3 ) ∈ F1 × F3 | ∃f2 ∈ F2
o
s.t. (f1 , f2 ) ∈ KA , (−f2 , f3 ) ∈ KB
X1 , F1 , D1 , H1
f1
e1
f1 f
X2 , F2 , D2 , H2 DI
e2 e
fk
ek
Xk , Fk , Dk , Hk
75
76 Port-Hamiltonian systems and passivity
∂T S
(x)f (x, u) ≤ uT h(x, u), (7.2)
∂x
for all x, u. For affine nonlinear systems ẋ = f (x) + g(x)u, y = h(x),
with g(x) an n × m matrix, this is easily seen to reduce to
∂T S ∂S
(x)f (x) ≤ 0, h(x) = gT (x) (x). (7.3)
∂x ∂x
An integral form of the differential dissipation inequality (7.1) is pro-
vided by
Z t2
S x(t1 ) − S x(t0 ) ≤ uT (t)y(t)dt, (7.4)
t1
for all time instants t1 ≤ t2 , all states x(t0 ), and all input functions
u : [t0 , t1 ] → Rm , where x(t1 ) denotes the state at time t1 resulting from
initial state x(t0 ) at time t0 . This integral form allows one to relax the
requirement of differentiability of S. Clearly (7.1) implies (7.4), while
if S is differentiable then conversely (7.4) can be seen to imply (7.1).
For port-Hamiltonian systems, passivity can be directly inferred
from the power-balance (2.27), and thus is a direct consequence of the
properties of the Dirac structure and the energy-dissipating relation.
Indeed, since by definition of the energy-dissipating relation, the term
eTR fR is always less than or equal to zero, the power-balance (2.27) can
be written as
d
H = eTR fR + eTP fP ≤ eTP fP . (7.5)
dt
7.1. Linear port-Hamiltonian systems 77
for x(0) = x. It can be shown Willems (1972a); van der Schaft (2000),
that the system is passive if and only if the function Sa is well-defined
for all x ∈ X , that is, the righthand side of (7.10) is finite for all x ∈ X .
It is easy to see, cf. van der Schaft (2000), that, whenever the system
is reachable from a certain state x∗ , the function Sa is well-defined (and
thus the system is passive) if and only if Sa (x∗ ) < ∞.
The quantity Sa (x) represents the maximal amount of energy that
can be extracted from the system starting from the initial state x(0) =
x, and is therefore called the available storage. Moreover, it is the smallest
of all possible storage functions: Sa (x) ≤ S(x), x ∈ X , for all other
storage functions S.
Furthermore, if we assume that the system is reachable from a cer-
tain state x∗ then there exists a largest storage function in the following
sense. Define the expression
Z 0
Sr (x) = inf − uT (t)y(t)dt, (7.11)
u(·) −τ
τ ≥0
∂H ∂ H̃
[J(x) − R(x)] (x) = [J˜(x) − R̃(x)] (x).
∂x ∂x
Hence, the system is port-Hamiltonian with respect to a different
Hamiltonian and a different Dirac structure and energy-dissipating re-
lation specified by J˜ and R̃.
D ⊂ Tx X × Tx∗ X × FR × ER × FP × EP ,
3
In thermodynamics this is called the availability function, cf. Kennan (1951). In
convex analysis it is also known as the Bregman function, cf. Bürger et al. (2013).
8
Conserved quantities and algebraic
constraints
d
H = eTR fR ≤ 0, (8.2)
dt
which implies that if H(x∗ ) = 0 and H(x) > 0, for every x 6= x∗ , then
x∗ is a stable equilibrium.1 However, the point where the Hamiltonian
is minimal (which typically coincides with the zero state) is often not
the one of practical interest for set-point regulation, in which case the
Hamiltonian alone can not be employed as a Lyapunov function.
1
The equilibrium x∗ is asymptotically stable if the dissipation term eTR fR < 0
for all x 6= x∗ , or alternatively if a detectability condition is satisfied, guaranteeing
asymptotic stability by the use of LaSalle’s Invariance principle.
83
84 Conserved quantities and algebraic constraints
has full rank. Indeed, in this case the differentiated constraint equation
d ∂H ∂2H
0= GT (x) (x) = ∗ + GT (x) 2 (x)G(x)λ
dt ∂x ∂x
(with ∗ denoting unspecified terms) can be uniquely solved for λ, lead-
ing to a uniquely defined dynamics on the constrained state space Xc .
Hence the set of consistent states for the port-Hamiltonian differential-
algebraic system (the set of initial conditions for which the system has
a unique ordinary solution) is equal to the constrained state space Xc .
Using terminology from the theory of DAEs, the condition that the
matrix in (8.6) has full rank ensures that the index of the DAEs speci-
fied by the port-Hamiltonian system is equal to one. This can be sum-
marized as
If the matrix in (8.6) does not have full rank, then the index of
the port-Hamiltonian differential-algebraic system will be larger than
one, and it will be necessary to further constrain the space Xc by
considering apart from the ’primary’ algebraic constraints (8.5), also
their (repeated) time-derivatives (called secondary constraints). We re-
fer to van der Schaft (1987); Nijmeijer & van der Schaft (1990) for a
detailed treatment and conditions for reducing the port-Hamiltonian
DAE system to a system without algebraic constraints in case J(x) cor-
responds to a symplectic structure. For the linear case, and the relation
with the theory of matrix pencils see van der Schaft (2013).
A particular elegant representation of the algebraic constraints
arises from the canonical coordinate representation. We will only consider
the case of a system without energy-dissipation and external ports. If
the Dirac structure D on the state space manifold is integrable, cf. Chap-
ter 3, then there exist local coordinates x = (q, p, r, s) for X in which
the system (without energy-dissipation and external ports) takes the
8.4. Elimination of algebraic constraints 89
form
∂H
q̇ = (q, p, r, s)
∂p
∂H
ṗ = − (q, p, r, s)
∂q (8.7)
ṙ = 0
∂H
0= (q, p, r, s)
∂s
Hence the Casimirs are all state functions only depending on r, while
the algebraic constraints take the simple form ∂H∂s = 0.
The condition that the matrix in (8.6) has full rank is in the canoni-
cal coordinate representation equivalent to the partial Hessian matrix
∂2H
∂s2 being invertible. Solving, by the Implicit Function theorem, the
algebraic constraints ∂H
∂s = 0 for s as a function s(q, p, r) reduces the
DAEs (8.7) to the ODEs
∂ H̄
q̇ = (q, p.r)
∂p
∂ H̄ (8.8)
ṗ = − (q, p, r)
∂q
ṙ = 0
It follows that the composition of the Dirac structure D with the resis-
tive structure R, defined as
D ◦ R := {(fS , eS , fP , eP ) ∈ F)S × ES × FP × EP |
91
92 Incrementally port-Hamiltonian systems
N := {(fS , eS , fP , eP ) ∈ FS × ES × FP × EP }
holds.
Furthermore, consider a maximal monotone relation
M ⊂ FS × ES × FP × EP ,
I = Φ(V − V0 ) + I0 ,
C := {(fP , eP ) | fP = c} (9.9)
d 1
(x1 (t) − x2 (t))T Q(x1 (t) − x2 (t))
dt 2
≤ (e1P (t) − e2P (t))T (fP1 (t) − fP2 (t)).
Recall furthermore from Forni & Sepulchre (2013); van der Schaft
(2013) the following definition of differential passivity.
∂Hj
δyj (t) = (x(t))δx(t) , j = 1, . . . , m ,
∂x
with state δx ∈ Rn , where δu = (δu1 , . . . , δum ) denote the inputs of
the variational system and δy = (δy1 , . . . , δym ) the outputs. Then the
system (9.11) is called differentially passive if the system together with
all its variational systems is passive with respect to the supply rate
98 Incrementally port-Hamiltonian systems
Proof. Consider the infinitesimal version of (9.6). In fact, let (fP1 , e1P , x1 )
and (fP2 , e2P , x2 ) be two triples of system trajectories arbitrarily near
each other. Taking the limit we deduce from (9.6)
∂2H
(∂x)T (x)∂ ẋ ≤ (∂eP )T ∂fP , (9.14)
∂x2
where ∂x denotes the variational state, and ∂fP , ∂eP the variational
inputs and outputs). If the Hamiltonian H is a quadratic function
H(x) = 21 xT Qx then the left-hand side of the inequality (9.14) is equal
d 1 T
to dt 2 (∂x) Q∂x, and hence amounts to the differential dissipativity
inequality
d1
(∂x)T Q∂x ≤ (∂eP )T ∂fP , (9.15)
dt 2
implying that the incrementally port-Hamiltonian system is differen-
tially passive, with differential storage function 12 (∂x)T Q∂x.
now based on the fact that the composition of maximal monotone re-
lations is again maximal monotone.
Consider two maximal monotonous relations Ma ⊂ Fa × Fa∗ ×
Va × Va∗ , with typical element denoted by (fa , ea , va , wa ) and Mb ⊂
Fb × Fb∗ × Vb × Vb∗ , with typical element denoted by (fb , eb , vb , wb ),
where Va = Vb = V, and thus Va∗ = Vb∗ = V ∗ (shared flow and effort
variables). Define as before the composition of Ma and Mb , denoted as
Ma ◦ Mb , by
n
Ma ◦ Mb := (fa , ea , fb , eb ) ∈ Fa × Fa∗ × Fb × Fb∗ |
∃v ∈ V, w ∈ V ∗ s.t. (fa , ea , v, w) ∈ Fa × Fa∗ × V × V ∗ ,
o
(fb , eb , −v, w) ∈ Fb × Fb∗ × V × V ∗ .
Thus the composition of Ma and Mb is obtained by imposing on the
vectors (fa , ea , va , wa ) ∈ Ma and (fb , eb , vb , wb ) ∈ Mb the interconnec-
tion constraints
va = −vb , wa = wb , (9.16)
and looking at the resulting vectors (fa , ea , fb , eb ) ∈ Fa × Fa∗ × Fb × Fb∗ .
The main result of this section is that, whenever Ma ◦Mb satisfies a
technical condition, then the composition Ma ◦Mb is again a maximal
monotone relation. The key ingredient in the proof is the following
theorem from Rockafellar & Wets (1998) [Ex. 12.46].
Theorem 9.1. Let M ⊂ Fα × Fα∗ × Fβ × Fβ∗ be maximal monotone.
Assume that the reduced relation (with e¯β a constant vector)
Mr := {(fα , eα ) | ∃fβ s.t. (fα , eα , fβ , eβ = e¯β )} (9.17)
is such that there exists e¯α for which (e¯α , e¯β ) is in the relative interior
of the projection of M on the space of efforts {(eα , eβ )}. Then Mr is
maximal monotone.
This theorem can be applied to the situation at hand after applying
the following transformation. Define
va + vb va − vb
yv := √ , zv := √ , yv , zv ∈ V
2 2
wa + wb wa − wb
yw := √ , zw := √ , yw , zw ∈ V ∗ .
2 2
100 Incrementally port-Hamiltonian systems
∂H
ẋ = [J(x) − R(x)] (x) + g(x)u, x ∈ X , u ∈ Rm
∂x (10.1)
∂H
y = gT (x) (x), y ∈ Rm ,
∂x
101
102 Input-output Hamiltonian systems
Now suppose that the input matrix g(x) satisfies an integrability condi-
tion1 , in the sense that there exists a mapping C : X → Rm such that2
∂C T
g(x) = −[J(x) − R(x)] (x). (10.3)
∂x
Then the system equations can be rewritten as
!
∂H ∂C T
ẋ = [J(x) − R(x)] (x) − (x)u ,
∂x ∂x
!T
∂C T ∂H
y= (x) [J(x) + R(x)] (x).
∂x ∂x
Noting that
T
R −R ∂C
∂x In h i
∂C T ≥0
T T
T T
=
∂C T
T R I −
n ∂x
− ∂C
R ∂C
R ∂C
T
− ∂x
∂x ∂x ∂x
k1 k2
F m1 m2 v
Proposition 10.1. The affine IOHD system (10.4) with output equa-
tion
!T !
∂C T ∂H ∂C T
ỹ := ż = (x) [J(x) − R(x)] (x) − (x)u
∂x ∂x ∂x
defines an input-state-output port-Hamiltonian system with
feedthrough term.
Then the system with differentiated outputs ỹ1 = ż1 , ỹ2 = ż2 is a port-
Hamiltonian input-state-output system (with feedthrough term), but
having outputs ỹ1 , ỹ2 differing from the original outputs y1 , y2 . In fact,
p1
ỹ1 = ż1 = k2 q2 + F, ỹ2 = ż2 = + v.
m1
Example 10.2. Consider a linear system
ẋ = Ax + Bu, x ∈ Rn , u ∈ Rm ,
(10.7)
y = Cx + Du, y ∈ Rm ,
with transfer matrix G(s) = C(Is − A)−1 B + D. In Lanzon & Petersen
(2008, 2010) G(s) is called negative imaginary4 if the transfer matrix
H(s) := s(G(s) − D) is positive real and D = DT . In Angeli (2006) the
same notion (mostly for the case D = 0) was coined as counterclockwise
input-output dynamics.
In van der Schaft (2011) it has been shown that the system (10.7)
has negative imaginary transfer matrix if and only if it can be written
as
ẋ = (J − R)(Qx − C T u),
(10.8)
y = Cx + Du, D = DT ,
for some matrices Q, J, R of appropriate dimensions satisfying
Q = QT , J = −J T , R = RT ≥ 0, (10.9)
with Q > 0. We conclude that a linear system (10.7) has negative imag-
inary transfer matrix if and only it is a linear input-output Hamiltonian
system with dissipation (10.8) satisfying Q > 0.
A typical instance of a linear IOHD system is a linear mechani-
cal system with co-located position sensors and force actuators, repre-
sented in Hamiltonian state space form (with q denoting the position
vector and p the momentum vector) as
" # " #" #" # " #
q̇ 0n In K N q 0
= + T u,
ṗ −In 0n NT M −1 p L (10.10)
y = Lq.
4
The terminology ’negative imaginary’, stems, similarly to ’positive real’, from the
Nyquist plot interpretation for single-input single-output systems. For the precise
definition in the frequency domain we refer to Lanzon & Petersen (2008); Angeli
(2006).
106 Input-output Hamiltonian systems
Remark 10.2. In case of the linear IOHD system (10.8) the dc-gain
amounts to the symmetric linear map ȳ = CQ−1 C T + D ū.
We have seen before, cf. Chapter 6, that the basic interconnection prop-
erty of port-Hamiltonian systems is the fact that the power-conserving
interconnection of port-Hamiltonian systems again defines a port-
Hamiltonian system, with Dirac structure being the composition of
the Dirac structures of the composing port-Hamiltonian systems, and
Hamiltonian function and resistive structure being the ’sum’ of the re-
spective Hamiltonian functions and resistive structures. A particular
instance of a power-conserving interconnection is the standard nega-
tive feedback interconnection of two input-state-output systems given
as
u1 = −y2 + e1 , u2 = y1 + e2 ,
systems
!
∂Hi ∂CiT
ẋi = [Ji (xi ) − Ri (xi )] (xi ) − (xi )ui , ui ∈ Rm ,
∂xi ∂xi (10.16)
m
yi = Ci (xi ), y ∈ R , i = 1, 2,
interconnected by the positive feedback interconnection
u1 = y2 + e1 , u2 = y1 + e2 . (10.17)
∗ ∗
Then (x1 , x2 ) is a stable equilibrium of the interconnected affine IOHD
system (10.18) if the interconnected Hamiltonian Hint given by (10.19)
has a strict minimum at the origin (x∗1 , x∗2 ). A sufficient condition
2 2
for this is that the Hessian matrices ∂∂xH21 (x∗1 ), ∂∂xH22 (x∗2 ) are positive-
1 2
definite, and furthermore the following coupling condition holds on
110 Input-output Hamiltonian systems
Example 10.3. The dc-gain of the linear IOHD system (10.10) is given
as LK −1 LT , and thus only depends on the compliance matrix K (e.g.,
the spring constants) and the collocated sensor/actuator locations.
Note that in this case positive feedback amounts to positive position
feedback, while negative feedback of z = ẏ = LM −1 p = Lq̇ cor-
responds to negative velocity feedback; see also Lanzon & Petersen
(2010).
111
112 Pseudo-gradient representations
∂2H ∗ ∂2H ∗
∂z 2 (z) ∂z1 ∂z2
(z)
1
Q(z) =
∂2H ∗ ∂2H ∗
− (z) − (z)
∂z2 ∂z1 ∂z22
∂2K
Qij (z) = (z),
∂zi ∂zj
for i, j = 1, . . . , n.
structure
DBM = (fS , eS , fP , eP ) ∈ FS × FS∗ × FP × FP∗
" #
g1 h i
Q(z)fS = −eS + fP , eP = g1T 0 fP ,
0
defined with respect to the bilinear form
C1
L2
q1 q2 p2
p1
R2
2
Note that if we would start from V (z, u) = P (z) − z1T g1 u, (constant) non-zero
inputs can naturally be taken into account in the Lyapunov analysis by generating an
admissible pair (Q̃, Ṽ ), with Ṽ satisfying Ṽ ≥ 0 for all z.
12
Port-Hamiltonian systems on graphs
119
120 Port-Hamiltonian systems on graphs
the identity map and ⊗ denotes the Kronecker product. B will be called
the incidence operator. For R = R the incidence operator reduces to the
linear map given by the matrix B itself, in which case we will through-
out use B both for the incidence matrix and for the incidence operator. The
adjoint map of B̂ is denoted as
B̂ ∗ : Λ0 → Λ1 ,
and is called the co-incidence operator. For R = R3 the co-incidence
operator is given by B T ⊗I3 , while for R = R the co-incidence operator
is simply given by the transposed matrix B T , and we will throughout
use B T both for the co-incidence matrix and for the co-incidence operator.
An open graph G is obtained from an ordinary graph with set of
vertices V by identifying a subset Vb ⊂ V of Nb boundary vertices.
The interpretation of Vb is that these are the vertices that are open to
interconnection (i.e., with other open graphs). The remaining subset
Vi := V − Vb are the Ni internal vertices of the open graph.
The splitting of the vertices into internal and boundary vertices
induces a splitting of the vertex space and its dual, given as
Λ0 = Λ0i ⊕ Λ0b , Λ0 = Λ0i ⊕ Λ0b ,
where Λ0i is the vertex space corresponding to the internal vertices
and Λ0b the vertex space corresponding to the boundary vertices. Con-
sequently, the incidence operator B̂ : Λ1 → Λ0 splits as
B̂ = B̂i ⊕ B̂b ,
with B̂i : Λ1 → Λ0i and B̂b : Λ1 → Λ0b . For R = R we will simply
write " #
Bi
B= .
Bb
Furthermore, we will define the boundary space Λb as the linear space
of all functions from the set of boundary vertices Vb to the linear space
R. Note that the boundary space Λb is equal to the linear space Λ0b ,
and that the linear mapping B̂b can be also regarded as a mapping
B̂b : Λ1 → Λb , called the boundary incidence operator. The dual space of
Λb will be denoted as Λb . The elements fb ∈ Λb are called the boundary
flows and the elements eb ∈ Λb the boundary efforts.
122 Port-Hamiltonian systems on graphs
damper 1
spring 2
(a)
(b)
defined with respect to the Poisson structure on the state space Λ1 ×Λ0
given by the skew-symmetric matrix
" #
0 BT
J := . (12.2)
−B 0
∂H
q̇ = B T (q, p),
∂p
∂H
ṗ = −B (q, p) + Efb , (12.3)
∂q
∂H
eb = E T (q, p).
∂p
nal vertices)
∂H
q̇ = BiT (q, pi ) + BbT eb ,
∂pi
∂H
ṗi = −Bi (q, pi ), (12.4)
∂q
∂H
f b = Bb (q, pi ),
∂q
1
One can also consider the hybrid case where some of the boundary vertices are
associated to masses while the remaining ones are massless.
2
Note that these equation follow from (12.4) by replacing −q̇ by e1 and ∂H
∂q
(q, p)
by f 1 .
12.2. Mass-spring-damper systems 125
tions
Bi f1 = −ṗi ,
Bb f 1 = f b , (12.5)
∂H
e1 = −BiT (pi ) − BbT eb ,
∂pi
where f1 , e1 are the flows and efforts corresponding to the dampers
(damping forces, respectively, velocities). For linear dampers f1 =
−Re1 , where R is the positive diagonal matrix with the damping con-
stants on its diagonal. Substitution into (12.5) then yields the port-
Hamiltonian system
∂H
ṗi = −Bi RBiT (pi ) − Bi RBbT eb ,
∂pi
(12.6)
∂H
fb = Bb RBiT (pi ) + BbT RBbT eb ,
∂pi
where, as before, the inputs eb are the boundary velocities and fb are
the forces as experienced at the massless boundary vertices. Note that
the matrix " #
Bi h i
L := R BiT BbT = BRB T
Bb
is the weighted Laplacian matrix of the graph G (with weights given by
the diagonal elements of R). It is well-known Bollobas (1998) that for
a connected graph the matrix L has exactly one eigenvalue 0, with
eigenvector 1, while all other eigenvalues are positive.
ẋ = Bν, (12.8)
ẋ = −BKB T M −1 x, (12.10)
δ̇i = ωib − ω r , i = 1, · · · , n,
X
Mi ω̇i = −ai (ωib − ω r ) − Vi Vj Sij [sin(δi − δj ) + ui ,
j6=i
where the summation in the last line is over all buses j which are ad-
jacent to bus i; that is, all buses j that are directly linked to bus i by a
transmission line (defining an edge of the graph). Here δi denotes the
voltage angle, vi the voltage amplitude, ωib the frequency, ωi := ωib −ω r
the frequency deviation, and ui the power generation/consumption;
all at bus i. Furthermore, ω r is the nominal (reference) frequency for
the network, Mi and ai are inertia and damping constants at bus i, and
Sij is the transfer susceptance of the line between bus i and j.
Define zk := δi − δj and ck := Ei Ej Sij , if the k-th edge is point-
ing from vertex i to vertex j. Furthermore, define the momenta pi =
Mi ωi , i = 1, · · · , n. Then the equations can be written in the vector
form
ż = B T M −1 p,
ṗ = −AM −1 p − BC Sin z + u,
where z is the m-vector with components zk , M is the diagonal matrix
with diagonal elements Mi , A is the diagonal matrix with diagonal
elements ai , and C is the diagonal matrix with elements ck . Further-
more, Sin : Rm → Rm denotes the elementwise sin function, and z ∗ is
the m-vector with k-th component δij∗.
12.4. Available storage 129
ẋ = Bu, u ∈ Rm , x ∈ Rn ,
∂H (12.13)
y = BT (x), y ∈ Rm ,
∂x
where B is the incidence matrix of the graph, and H(x) = 21 kxk2 is
the Hamiltonian function. Clearly, since H is non-negative it defines a
storage function, and the system is passive. On the other hand it will
turn out that the minimal storage function for the system, called the
available storage (see Chapter 7), is different from H.
Throughout this section we will assume3 that the graph is con-
nected, or equivalently Bollobas (1998) ker B T = span 1. Based on
3
Without loss of generality, since otherwise the analysis can be repeated for every
connected component of the graph.
130 Port-Hamiltonian systems on graphs
where we consider the supremum over all τ ≥ 0 and all input func-
tions u : [0, τ ] → Rm , and where y : [0, τ ] → Rm is the output resulting
from the input function u : [0, τ ] → Rm and initial condition x(0) = x.
Noting that
Z τ Z τ
T
u (t)y(t)dt = uT (t)B T x(t)dt
0 0
Z τ 1 1
= ẋT (t)x(t)dt = kx(τ )k2 − kx(0)k2 ,
0 2 2
we see that the available storage is equivalently given as
1 1
Sa (x) = sup kxk2 − kx(τ )k2 , (12.15)
2 2
where we take the supremum over all τ ≥ 0 and all possible states
x(τ ) resulting from input functions u : [0, τ ] → Rm . By connectedness
of the graph, we know that from x(0) = x we can reach, by choosing
the input function suitably, any state x(τ ) satisfying
1T x(τ ) = 1T x. (12.16)
h i
dampers of the system, with incidence matrix B = Bs Bd , where
the columns of Bs reflect the spring edges and the columns of Bd
the damper edges. Without boundary vertices the dynamics takes the
form (see equation (12.7) in Section 12.2.3)
" # " # ∂H (q, p)
q̇ 0 BsT ∂q
= T
. (12.20)
ṗ −Bs −B RB
d d
∂H
(q, p)
∂p
Throughout this section we make the following simplifying assump-
tion5 . The graph G(V, Es ∪ Ed ) is connected, or equivalently ker BsT ∩
ker BdT = span 1.
∂T H ∂H
(q, p)Bd RBdT (q, p) = 0,
∂p ∂p
or equivalently BdT ∂H ∂H
∂p (q, p) = 0, which implies Bs ∂q (q, p) = 0.
Proposition 12.2. The Casimir functions are all functions C(q, p) sat-
isfying
∂C ∂C
(q, p) ∈ span 1, (q, p) ∈ ker Bs . (12.21)
∂p ∂q
Proof. The function C(q, p) is a Casimir if
" #
∂C ∂C 0 BsT
(q, p) (q, p) = 0,
∂q ∂p −Bs −Bd RBdT
or equivalently
∂T C ∂T C ∂T C
(q, p)Bs = 0, (q, p)BsT + (q, p)Bd RBdT = 0.
∂p ∂q ∂p
∂C
Postmultiplication of the second equation by ∂p (q, p), making use of
the first equation, gives the result.
d ∂T H ∂H
H(q, p) = − (q, p)Bd RBdT (q, p)
dt ∂p ∂p (12.27)
= −pT Bd GRBdT Gp ≤ 0,
d ∂T H ∂H
H̄(q, p) = − (q, p)Bd RBdT (q, p) ≤ 0. (12.30)
dt ∂p ∂p
Specializing to H(q, p) = 21 q T Kq + 12 pT Gp, in which case H̄(q, p) =
1 T 1 T
2 (q − q̄) K(q − q̄)+ 2 p Gp, we obtain the following analog of Theorem
12.1.
From standard reduction theory, see e.g. Marsden & Ratiu (1999);
Libermann & Marle (1987) and the references quoted therein, it fol-
lows that we may factor out the configuration space Qc := Λ0 to the
reduced configuration space
Q := Λ0 /G (12.34)
Q := Λ0 /G ≃ B T Λ0 ⊂ Λ1 . (12.35)
im B T × Λ0 ⊂ Λ1 × Λ0 obtained before:
∂Hc ∂H
q̇ = B T q̇c = B T (qc , p) = B T (q, p),
∂p ∂p
∂Hc ∂H
ṗ = − (qc , p) + Efb = −B (q, p) + Efb , (12.36)
∂qc ∂q
∂H
eb = E T (q, p).
∂p
In case the graph is not connected, then the above symmetry reduction
can be performed for each component of the graph (i.e., the symmetry
group is RcG , with cG denoting the number of components of the graph
G), yielding again the reduced state space8 im B T × Λ0 .
For a mass-spring-damper system, although in the standard sym-
metry reduction framework not considered as a Hamiltonian system,
the same reduction procedure can still be applied. A mass-spring-
damper system in coordinates (qc , p) takes the form
∂Hc
q̇c = (qc , p),
∂p
∂Hc ∂Hc
ṗ = − (qc , p) − Bd RBdT (qc , p) + Efb , (12.37)
∂qc ∂p
∂Hc
eb = E T (qc , p),
∂p
where Hc (qc , p) = H(BsT q, p) with q = BsT qc the spring elonga-
tions. Here Bs and Bd denote, as before, the incidence matrices of the
spring, respectively, damper graph. Under the same symmetry action
as above this reduces to the equations (12.20) on the reduced state
space im BsT × Λ0 .
Furthermore we obtain the following corollary to Theorem 12.1
regarding to ’second-order consensus’ (see also Goldin et al. (2010);
Camlibel & Zhang (2012)):
Then for all initial conditions qc (t) → span 1, p(t) → span 1 if and
only the largest GLs -invariant subspace contained in ker BdT is equal
to span 1, and moreover ker Bs = 0.
By Proposition 2.3 both Df (G) and De (G) are separable Dirac struc-
tures. Note that Df (G) and De (G) only differ in the role of the bound-
ary flows and efforts, and that Df (G) = De (G) if there are no boundary
vertices.
Interconnection of two open graphs G α and G β is done by identify-
ing some of their boundary vertices, and equating (up to a minus sign)
the boundary efforts and flows corresponding to these boundary ver-
tices, resulting in a new graph. For simplicity of exposition consider
the case that the open graphs have all their boundary vertices in com-
mon, resulting in a (closed) graph with set of vertices Viα ∪ Viβ ∪ V,
where V := Vbα = Vbβ denotes the set of boundary vertices of both
graphs. The incidence operator of the interconnected (closed) graph is
12.8. The Kirchhoff-Dirac structure 141
BbC Q̇ = IbC ,
BiC Q̇ = 0,
BbCT ψbC = C −1 Q − BiCT ψiC .
with Q the vector of charges of the capacitors and C the diagonal ma-
trix with diagonal elements given by the capacitances of the capaci-
144 Port-Hamiltonian systems on graphs
1Tb fb = 0 ,
This difference stems from the fact that the energy variable Q of a
capacitor, as well as the current I, takes values in the linear space Λ1 ,
while the state variable Φ of an inductor, as well as the voltage V ,
146 Port-Hamiltonian systems on graphs
takes values in the dual space Λ1 . Recalling from Section 12.2.1 the
description of a spring system
" #
vα
q̇ = bT ,
vβ
" # (12.48)
Fα dH
=b (q),
Fβ dq
with q the elongation of the spring, and H(q) its potential energy, we
conclude that there is a strict analogy between a spring and an induc-
tor12 . On the other hand, a moving mass is not a strict analog of a ca-
pacitor. Instead, it can be considered to be the analog of a grounded
capacitor, while the strict analog of a capacitor (12.47) is the so-called
inerter Smith (2002)
" # " #
F dH v
bṗ = α , (p) = bT α ,
Fβ dp vβ
where p is the momentum of the inerter and H(p) its kinetic energy,
while Fα , Fβ and v1 , v2 denote the forces, respectively, velocities, at the
two terminals of the inerter. For a further discussion on analogies, see
Appendix B.
12
Thus we favor the so-called force-current analogy instead of the force-voltage anal-
ogy.
13
Switching port-Hamiltonian systems
147
148 Switching port-Hamiltonian systems
D ⊂ Fx × Ex × FR × ER × FP × EP × FS × ES . (13.1)
We will say that in switch configuration π, for all i ∈ π the i-th switch
is closed, while for j ∈
/ π the j-th switch is open.
For each fixed switch configuration π this leads to the following
subspace Dπ of the restricted space of flows and efforts Fx × Ex × FR ×
ER × FP × EP :
n
Dπ = (fx , ex , fR , eR , fP , eP ) | ∃fS ∈ FS , eS ∈ ES
s.t. eiS = 0, i ∈ π, fSj = 0, j 6∈ π, and (13.3)
o
(fx , ex , fR , eR , fP , eP , fS , eS ) ∈ D .
For every π the subspace Dπ defines a Dirac structure. Indeed, every
switch configuration π given by (13.2) defines a Dirac structure on the
space of flow and effort variables fS , eS of the switches, and Dπ equals
the composition of this Dirac structure with the overall Dirac structure
D. Since the composition of any two Dirac structures is again a Dirac
structure (cf. Chapter 6) it thus follows that Dπ is a Dirac structure for
every switch configuration π.
The dynamics of the switching port-Hamiltonian system is defined
by specifying as before, next to its Dirac structure D, the constitutive
relations of the energy-storing and energy-dissipating (resistive) ele-
ments. We will restrict to linear resistive structures given by the graph
of a resistive mapping (see Chapter 4).
Definition 13.1. Consider a linear state space X = Fx , a Dirac struc-
ture D given by (13.1), a Hamiltonian H : X → R, and a linear resis-
tive structure fR = −ReR with R = RT ≥ 0. Then the dynamics of the
corresponding switching port-Hamiltonian system is given as
∂H
(−ẋ(t), (x(t)), −ReR (t), eR (t), fP (t), eP (t)) ∈ Dπ (13.4)
∂x
at all time instants t during which the system is in switch configura-
tion π.
It follows from the power-conservation property of Dirac struc-
tures that during the time-interval in which the system is in a fixed
switch configuration
d
H = −eTR ReR + eTP fP ≤ eTP fP , (13.5)
dt
150 Switching port-Hamiltonian systems
thus showing passivity for each fixed switch configuration if the Hamilto-
nian H is non-negative.
g m
Figure 13.1: Bouncing pogo-stick: definition of the variables (left), flying phase (mid-
dle), contact phase (right).
where fR , eR are the flow and effort variables associated to the damper
(energy dissipation), and d is the damping constant. Furthermore, the
overall Dirac structure of the system is described by the linear equa-
13.1. Switching port-Hamiltonian systems 151
tions
fy = fx − fS , fR = fx , fp = ex + ey + eR ,
eS + ex + eR = 0, ep = −fy .
Here the third equation fp = ex + ey + eR represents the total force
balance on the mass m. In the switch configuration eS = 0 (no external
force on the foot) the pogo-stick is in its flying mode, while for fS = 0
the foot is in contact with the horizontal plate. Hence the equation
eS +ex +eR = 0 expresses that for eS = 0 (flying mode) the spring force
on the mass-less foot balances the damping force, while for fS = 0
(contact mode) fy = fx and eS represents the constraint force exerted
by the ground.
pM (momentum of the foot) with kinetic energy 21 p2M , and the equa-
tion eS + ex + eR = 0 changes into fM = eS + ex + eR with fM = −ṗM ,
while furthermore, an extra equation epM == fy − fx with eM = pm M
The following crucial relation between the jump space Jπ and the
effort constraint subspace Cπ holds true. Recall that Jπ ⊂ Fx while
Cπ ⊂ Ex , where Ex = Fx∗ .
Theorem 13.1.
Jπ = Cπ⊥ , (13.9)
where ⊥ denotes the orthogonal complement with respect to the dual-
ity product between the dual spaces Fx and Ex .
In this subsection we will show how the jump rule for switching
port-Hamiltonian systems, as formulated above, includes the classi-
cal charge and flux conservation principles for RLC-circuits with switches
as a special case, and in fact formalizes these principles in an insightful
way.
Consider an RLC-circuit with switches with an arbitrary topology.
It can be described as a switched port-Hamiltonian system as follows
(see also Escobar et al. (1999)). First consider the directed graph asso-
ciated with the circuit. Identify every capacitor, every inductor, every
resistor and every switch with an edge. Furthermore, associate with
every external port an edge (between the terminals of the port). De-
note the incidence matrix of this directed graph by B; cf. Section 12.1.
The incidence matrix has as many columns as there are edges, and as
many rows as there are vertices in the graph. h By reordering the edgesi
we partition the incidence matrix as B = BC BL BR BS BP ,
where the sub-matrices BC , BL , BR , BS correspond, respectively, to
the capacitor, inductor, resistor, and switch edges, and BP corresponds
to the external ports. Then Kirchhoff’s current laws are given as
BC IC + BL IL + BR IR + BS IS + BP IP = 0, (13.16)
VC = BCT ψ,
VL = BLT ψ,
T ψ,
V R = BR (13.17)
VS = BST ψ,
VP = BPT ψ,
fx = (IC , VL ),
ex = (VC , IL ),
fR = VR ,
eR = IR ,
(13.18)
fS = VS ,
eS = IS ,
fP = VP ,
eP = IP .
1 1
H(Q, Φ) = QT C −1 Q + ΦT L−1 Φ, (13.20)
2 2
where the diagonal elements of the diagonal matrices C and L are the
capacitances, respectively, inductances, of the capacitors and induc-
tors.
Similarly, the constitutive relations for the linear (Ohmic) resistors
are given as
VR = −RIR , (13.21)
0 = BC IC + BL IL + BR IR + BS IS + BP IP ,
VC = BCT ψ,
VL = BLT ψ,
T ψ,
V R = BR (13.22)
VS = BST ψ,
VP = BPT ψ,
VSi = 0, i ∈ π, ISj = 0, j 6∈ π.
Thus, all switches corresponding to the subset π are closed, while the
remaining ones are open.
The constraint subspace Cπ for each switch configuration π is given
as
Cπ = {(VC , IL ) | ∃IC , VL , VR , IR , VS , IS , VP , IP
(13.23)
such that (13.21) and (13.22) is satisfied }.
Furthermore, the jump space Jπ is given as the set of all (IC , VL ) sat-
isfying for some ψ the equations
0 = BC IC + BS IS ,
0 = BCT ψ,
VL = BLT ψ,
T ψ,
0 = BR (13.24)
VS = BST ψ,
0= BPT ψ,
VSi = 0, i ∈ π, ISj = 0, j 6∈ π.
Hence the jump space can be written as the product of the space
It follows that for RLC circuits with switches the jump (state trans-
fer) splits into a charge transfer Q+ − Q− = Qtransfer and a flux transfer
Φ+ − Φ− = Φtransfer . The charge transfer Qtransfer corresponding to
the switch configuration π is specified as follows. The direction of the
charge transfer Qtransfer is determined by
0 = BCT ψ,
Φtransfer = BLT ψ,
T ψ,
0 = BR (13.26)
VS = BST ψ, VSi = 0, i ∈ π,
0= BPT ψ,
dH ∂H T ∂H
(q, p) = − (q, p)R̄(q, p) (q, p) + v T F ≤ v T F.
dt ∂p ∂p
We throughout assume that the matrix R̄(q, p) admits a factorization
where the matrix Aπ (q) is obtained from the matrix A(q) by leaving
out every j-th column with j 6∈ π.
Thus jump rule in this case amounts to a jump in the momentum
variables p given as
∂H
ptransfer = p+ − p− ∈ Aπ (q), ATπ (q) (q, p+ ) = 0
∂p
If H can be written as the sum of a kinetic and a potential energy
H(q, p) = 12 pT M −1 (q)p + V (q), with M (q) > 0 denoting the general-
ized mass matrix, then a variational characterization of the jump rule
is given by defining p+ to be the unique minimum of
1
min (p − p− )T M −1 (q)(p − p− ) (13.31)
p,AT (q)M −1 (q)p=0 2
Π
163
164 Distributed-parameter systems
with α ∈ Ωk (∂Z), β ∈ Ωn−1−k (∂Z). Now, let us define the linear space
p + q = n + 1, (14.4)
where for i = 1, 2,
fpi ∈ Ωp (Z), fqi ∈ Ωq (Z), eip ∈ Ωn−p (Z), eip ∈ Ωn−q (Z),
The spaces of differential forms Ωp (Z) and Ωq (Z) represent the en-
ergy variables of two different physical energy domains interacting
with each other, while Ωn−p (∂Z) and Ωn−q (∂Z) will denote the bound-
ary variables whose (wedge) product represents the boundary energy
flow. The following theorem is proved in van der Schaft & Maschke
(2002), based on ’integration by parts’ and Stokes’ theorem. Recall that
d denotes the exterior derivative, mapping k-forms into k + 1-forms
(and generalizing in R3 the vector calculus operations grad, curl, div).
Theorem 14.1. Consider Fp,q and Ep,q given in (14.3) and (14.5), re-
spectively, with p, q satisfying (14.4), and bilinear form ≪, ≫ given by
(14.6). Let (·)|∂Z denote the restriction to the boundary ∂Z, then the
166 Distributed-parameter systems
linear subspace
D = (fp , fq , fb , ep , eq , eb ) ∈ Fp,q × Ep,q
" # " #" #
fp 0 (−1)pq+1 d ep
= , (14.7)
fq d 0 eq
" # " #" #
fb 1 0 ep|∂Z
= ,
eb 0 −(−1)n−q eq|∂Z
is a Dirac structure.
The subspace (14.7) is called a Stokes-Dirac structure.
with r = pq + 1.
with
" # Q(t, z) " #
δQ H C(z) V (t, z)
ϕ(t, z) = I(t, z) ,
=
δϕ H
L(z)
14.2. Distributed-parameter port-Hamiltonian systems 169
where C(z) and L(z) are the distributed capacitance and distributed
inductance of the line, respectively, whereas V (t, z) and I(t, z) rep-
resent the corresponding voltage and current. The resulting port-
Hamiltonian system is given by
" # ∂ " #
∂ Q(t, z) 0 ∂z
V (t, z)
− = ∂ ,
∂t ϕ(t, z) 0 I(t, z)
∂z
which represents the well-known telegrapher’s equations, together
with the boundary variables
" # " # " # " #
fb0 (t) V (t, 0) e0b (t) I(t, 0)
= , =− .
fb1 (t) V (t, 1) e1b (t) I(t, 1)
yielding
1 2
eh = δh H = v + gh (Bernoulli function),
2
ev = δv H = hv (mass flow).
which expresses that the power flow through the boundary of the
channel equals velocity × pressure + energy flux through the bound-
ary.
Energy exchange through the boundary is not the only possible way a
distributed-parameter system may interact with its environment. An
example of this is provided by Maxwell’s equations (Example 14.3),
where interaction may also take place via the current density J, which
directly affects the electric charge distribution in the domain Z. In or-
der to cope with this situation, we augment the spaces Fp,q and Ep,q as
defined in (14.3) and (14.5), respectively, to
s
Fq,p := Fp,q × Ωs (S),
s
(14.14)
Eq,p := Ep,q × Ωn−s (S),
with dual map (again we consider Ωn−p (Z) and Ωn−q (Z) as dual
spaces to Ωp (Z) and Ωn−q (Z), respectively)
satisfying
Z Z h i
[ep ∧ Gp (fs ) + eq ∧ Gq (fs )] = G∗p (ep ) + G∗q (eq ) ∧ fs ,
Z S
D s ⊂ Fp,q
s s
× Ep,q ,
fs = −R(es )
⋆Js = σE,
be a function satisfying
d(δp C) = 0,
(14.26)
d(δq C) = 0.
14.4. Conservation laws 175
= eb ∧ fbC + eC
b ∧ fb ,
∂Z ∂Z
fbC := δp C|∂Z , eC
b := −(−1)
n−q
δq C|∂Z .
δp C|∂Z = 0,
(14.27)
δq C|∂Z = 0,
then dC
dt = 0 along the system trajectories of (14.11) for any Hamilto-
nian H. Therefore, a function C satisfying (14.26) and (14.27) is called
a Casimir function. If C satisfies (14.26), but not (14.27), then the time-
derivative of C is determined by the boundary conditions of (14.11),
and therefore is called a conservation law for (14.11).
γp := αp + (−1)r δq H ∧ dt,
γq := αq + δp H ∧ dt,
respectively. Then the first part of the equations (14.11) can be equiva-
lently stated as
¯ p = 0,
L ∂ dγ
∂t
(14.28)
¯ q = 0,
L ∂ dγ
∂t
¯ p = βp ,
dγ
(14.29)
¯ q = βq ,
dγ
∂αp
dαp + ∧ dt + (−1)r d(δq H) ∧ dt = βp ,
∂t (14.30)
∂αq
dαq + ∧ dt + d(δp H) ∧ dt = βq ,
∂t
14.5. Covariant formulation of port-Hamiltonian systems 177
with d denoting the exterior derivative with respect to the spatial vari-
ables z, resulting in the equations of a port-Hamiltonian system (14.11)
∂αp
− = (−1)r d(δq H),
∂t (14.31)
∂αq
− = d(δp H),
∂t
together with the conserved quantities (cf. Chapter 8) dαp = βp
and dαq = βq . Furthermore, the boundary variables of the port-
Hamiltonian system (14.11) can be re-formulated as
i ∂ γq = fb ,
∂t ∂Z
= (−1)q eb .
i∂ γp
∂t ∂Z
15
Control of port-Hamiltonian systems
179
180 Control of port-Hamiltonian systems
any interconnection between the plant and the controller through their
respective external ports satisfying the power-preserving property
which actually is p = 0 if, and only if, gk 6= 0 for all k ∈ {x, y, z}.
15.2. Energy transfer control 181
Stabilizing a system at the origin, which often coincides with the open-
loop minimum energy, is generally not an enticing control problem.
Of wider practical interest is to stabilize the system at a non-zero set-
point. Indeed, suppose that we want to stabilize a plant pH system
around a set-point x∗p . We know that for any controller pH system, the
closed-loop power-balance satisfies
u y
plant
−
ȳ ū
controller
∂Hp
ẋp = Jp (xp ) − R(xp ) (xp ) + gp (xp )u,
∂xp
(15.6)
∂Hp
y = gpT (xp ) (xp ),
∂xp
∂Hc
ẋc = Jc (xc ) − Rc (xc ) (xc ) + gc (xc )ū,
∂xc
(15.7)
∂Hc
ȳ = gcT (xc ) (xc ),
∂xc
closed-loop system
∂Hp
" # "
T
#(xp )
ẋp Jp (xp ) − Rp (xp ) −gp (xp )gc (xc ) ∂xp
= ,
ẋc gc (xc )gpT (xp ) Jc (xc ) − Rc (xc ) ∂Hc
(xc )
∂xc
(15.8)
∂H p
∂x (xp )
" # " #
y gpT (xp ) 0 p
= ,
ȳ 0 gcT (xc ) ∂Hc
(xc )
∂xc
which is again a pH system with Hamiltonian Hp (xp ) + Hc (xc ).
The main idea then is to design the controller system such that the
closed-loop system (15.8) has useful Casimirs. If both the plant and
the controller are lossless, i.e., Rp (xp ) = 0 and Rc (xc ) = 0, we thus
look for functions C(xd , xc ) satisfying
" #
∂T C ∂T C
(xp , xc ) (xp , xc )
∂xp ∂xc
" # (15.9)
Jp (xp ) −gp (xp )gcT (xc )
× = 0,
gc (xc )gpT (xp ) Jc (xc )
such that the candidate Lyapunov function
V (xp , xc ) = Hp (xp ) + Hc (xc ) + C(xp , xc ) (15.10)
has a minimum at (x∗p , x∗c ) for a certain x∗c . Subsequently, one may add
extra damping, directly or in the dynamics of the controller, to achieve
asymptotic stability.
Example 15.2. Consider a pendulum with normalized Hamiltonian
1
H(q, p) = p2 + 1 − cos q
2
actuated by a torque u, with output y = p (angular velocity). Suppose
we wish to stabilize the pendulum at q ∗ 6= 0 and p∗ = 0. Apply the
nonlinear integral control
ẋc = ū
∂Hc
ȳ = (xc ),
∂xc
15.3. Stabilization by Casimir generation 185
ẋc = ū
∂Hc
ȳ = (xc ),
∂xc
with Hamiltonian Hc (xc ) = gh∗ xc , via the feedback interconnection
1
ū = y = h(b)v(b), ȳ = −u = v 2 (b) + gh(b).
2
By mass-balance, we find that
Z b
h(z, t)dz + xc + κ
a
∂C ∂C
Rp (xp ) (xp , xc ) = 0, Rc (xc ) (xp , xc ) = 0.
∂xp ∂xc
Example 15.4. Consider the levitated ball system of Example 2.4 and
assume that the inductance of the coil is given by L(q) = k/(1 − q),
with k some real parameter depending on the geometry of the coil.
Then, the desired equilibrium is given by
p
(q ∗ , p∗ , ϕ∗ ) = (q ∗ , 0, ± 2kmg),
Using combinations of the stated options, there are three major ap-
proaches to solve the PDE’s of (15.23):
197
Appendices
A
Proofs
201
202 Proofs
203
204 Physical meaning of efforts and flows
f R x dH e
(x)
dx
207
208 References
J. Cortes, A.J. van der Schaft, P.E. Crouch, "Characterization of gradient con-
trol systems", SIAM J. Control Optim., 44(4): 1192–1214, 2005.
T.J. Courant, “Dirac manifolds”, Trans. Amer. Math. Soc., 319, pp. 631–661,
1990.
P.E. Crouch, "Geometric structures in systems theory, Proc. IEE-D, 128: 242–
252, 1981.
P.E. Crouch, "Spacecraft attitude control and stabilization: applications of ge-
ometric control theory to rigid body models", IEEE Trans. Automatic Con-
trol, AC-29(4), pp. 321-331, 1984.
P.E. Crouch, A.J. van der Schaft, Variational and Hamiltonian control systems,
Lect. Notes in Control and Information Sciences, Vol. 101, Springer-Verlag,
Berlin, 1987, p. 121.
M. Dalsmo, A.J. van der Schaft, “On representations and integrability of
mathematical structures in energy-conserving physical systems”, SIAM J.
Control and Optimization, vol.37, pp. 54–91, 1999.
C. DePersis, C.S. Kallesoe. Pressure regulation in nonlinear hydraulic net-
works by positive and quantized control. IEEE Transactions on Control Sys-
tems Technology, 19(6), pp. 1371–1383, 2011.
C.A. Desoer, M. Vidyasagar, Feedback systems: input-output properties, Siam
Classics in Applied Mathematics, SIAM, 1975.
D.A. Dirksz, J.M.A. Scherpen. Power-based control: Canonical coordinate
transformations, integral and adaptive control. Automatica, Vol. 48, pp.
1046–1056, 2012.
P.A.M. Dirac, Lectures in Quantum Mechanics, Belfer Graduate School of Sci-
ence, Yeshiva University Press, New York, 1964; Can. J. Math., 2, pp. 129–
148, 1950.
P.A.M. Dirac, "Generalized Hamiltonian Dynamics", Proc. R. Soc. A, 246, pp.
333–348, 1958.
I. Dorfman. Dirac Structures and Integrability of Nonlinear Evolution Equations.
John Wiley, Chichester, 1993.
V. Duindam, G. Blankenstein and S. Stramigioli, "Port-Based Modeling and
Analysis of Snake-board Locomotion”, Proc. 16th Int. Symp. on Mathemati-
cal Theory of Networks and Systems (MTNS2004), Leuven, 2004.
Modeling and Control of Complex Physical Systems; The Port-Hamiltonian Ap-
proach, eds. V. Duindam, A. Macchelli, S. Stramigioli, H. Bruyninckx,
Springer, 2009.
210 References
G. Golo, A. van der Schaft, P.C. Breedveld, B.M. Maschke, “Hamiltonian for-
mulation of bond graphs”, Nonlinear and Hybrid Systems in Automotive Con-
trol, Eds. R. Johansson, A. Rantzer, pp. 351–372, Springer London, 2003.
K. Gerritsen, A.J. van der Schaft, W.P.M.H. Heemels, “On switched Hamil-
tonian systems”, Proceedings 15th International Symposium on Mathematical
Theory of Networks and Systems (MTNS2002), Eds. D.S. Gilliam, J. Rosen-
thal, South Bend, August 12–16, 2002.
V. Guillemin, S. Sternberg, Some problems in integral geometry and some
related problems in micro-local analysis. American Journal of Mathematics,
101(4), 915–955, 1979.
M. Hernandez-Gomez, R. Ortega, F. Lamnabhi-Lagarrigue, and G. Esco-
bar. Adaptive PI Stabilization of Switched Power Converters, IEEE Trans,
Contr. Sys. Tech., 18(3):688–698, pp. May 2010.
D. Hill and P. Moylan. "The stability of nonlinear dissipative systems”, IEEE
Trans. Aut. Contr., 21(5):708–711, October 1976.
L. Hörmander, Fourier integral operators, I. Acta Math., 127, 79–183, 1971.
F.J.M. Horn, R. Jackson, “General mass action kinetics", Arch. Rational Mech.
Anal., 47, pp. 81–116, 1972.
R.S. Ingarden, A. Jamiolkowski, Classical Electrodynamics, PWN-Polish Sc.
Publ., Warszawa, Elsevier, 1985.
B. Jacob, H.J. Zwart, Linear port-Hamiltonian systems on infinite-dimensional
spaces, Series: Operator Theory: Advances and Applications, Vol. 223, Sub-
series: Linear Operators and Linear Systems, Birkhäuser, Basel, 2012.
B. Jayawardhana, R. Ortega, E. Garcia-Canseco, and F. Castanos, "Passivity of
nonlinear incremental systems: application to PI stabilization of nonlinear
RLC circuits”, Systems & Control Letters, Vol. 56, No. 9-10, pp. 618-622, 2007.
D. Jeltsema, R. Ortega, J.M.A. Scherpen, "On passivity and power-balance
inequalities of nonlinear RLC circuits", IEEE Transactions on Circuits and
Systems I: Fundamental Theory and Applications, 50(9): 1174–1179, 2003.
D. Jeltsema and J.M.A. Scherpen, "On Brayton and Moser’s Missing Stability
Theorem”, it IEEE Trans. Circ. and Syst-II, Vol. 52, No. 9, 2005, pp. 550–552.
D. Jeltsema and J.M.A. Scherpen, "Multi-domain modeling of nonlinear net-
works and systems: energy- and power-based perspectives”, in August edi-
tion of IEEE Control Systems Magazine, pp. 28-59, 2009.
D. Jeltsema and A.J. van der Schaft, "Memristive Port-Hamiltonian Systems",
Mathematical and Computer Modelling of Dynamical Systems, Vol. 16, Issue 2,
pp. 75–93, April 2010.
212 References
R. Polyuga, A.J. van der Schaft, "Structure preserving moment matching for
port-Hamiltonian systems: Arnoldi and Lanczos", IEEE Transactions on Au-
tomatic Control, 56(6), pp. 1458–1462, 2011.
S. Rao, A.J. van der Schaft, B. Jayawardhana, "A graph-theoretical approach
for the analysis and model reduction of complex-balanced chemical reac-
tion networks", J. Math. Chem., 51:2401–2422, 2013.
J.A. Roberson, C.T. Crowe. Engineering fluid mechanics. Houghton Mifflin
Company, 1993.
R.T. Rockafellar and J.-B. Wets, Variational Analysis, Series of Comprehensive
Studies in Mathematics 317. Springer, 1998.
M. Seslija, A. J. van der Schaft, and J. M.A. Scherpen, "Discrete Exterior Ge-
ometry Approach to Structure-Preserving Discretization of Distributed-
Parameter Port-Hamiltonian Systems", Journal of Geometry and Physics, Vol-
ume 62, 1509–1531, 2012.
A.J. van der Schaft, “Hamiltonian dynamics with external forces and obser-
vations”, Mathematical Systems Theory, 15, pp. 145–168, 1982.
A.J. van der Schaft, “Observability and controllability for smooth nonlinear
systems”, SIAM J. Control & Opt., 20, pp. 338–354, 1982.
A.J. van der Schaft, "Linearization of Hamiltonian and gradient systems”,
IMA J. Mathematical Control & Inf., 1:185–198, 1984.
A.J. van der Schaft, System theoretic descriptions of physical systems, CWI Tract
No. 3, CWI, Amsterdam, 1984, p. 256.
A.J. van der Schaft, “Equations of motion for Hamiltonian systems with con-
straints”, J. Physics A: Math. Gen., 20, pp. 3271–3277, 1987.
A.J. van der Schaft, “Implicit Hamiltonian systems with symmetry”, Rep.
Math. Phys., 41, pp.203–221, 1998.
A.J. van der Schaft, “Interconnection and geometry”, in The Mathematics of
Systems and Control: from Intelligent Control to Behavioral Systems, Editors
J.W. Polderman, H.L. Trentelman, Groningen, pp. 203–218, 1999.
A.J. van der Schaft, L2 -Gain and Passivity Techniques in Nonlinear Control,
Lect. Notes in Control and Information Sciences, Vol. 218, Springer-Verlag,
Berlin, 1996, p. 168, 2nd revised and enlarged edition, Springer-Verlag,
London, 2000 (Springer Communications and Control Engineering series),
p.xvi+249.
A.J. van der Schaft, "Port-Hamiltonian Systems", Chapter 2 in Modeling and
Control of Complex Physical Systems; The Port-Hamiltonian Approach, eds. V.
Duindam, A. Macchelli, S. Stramigioli, H. Bruyninckx, Springer, 2009.
References 215
A.J. van der Schaft, "Characterization and partial synthesis of the behavior
of resistive circuits at their terminals", Systems & Control Letters, vol. 59, pp.
423–428, 2010.
A.J. van der Schaft, "On the relation between port-Hamiltonian and gradient
systems", Preprints of the 18th IFAC World Congress, Milano (Italy) August
28 - September 2, pp. 3321–3326, 2011.
A.J. van der Schaft, "Positive feedback interconnection of Hamiltonian sys-
tems", 50th IEEE Conference on Decision and Control and European Control
Conference (CDC-ECC) Orlando, FL, USA, December 12-15, pp. 6510–6515,
2011.
A.J. van der Schaft, "Port-Hamiltonian differential-algebraic systems", pp.
173–226 in Surveys in Differential-Algebraic Equations I (eds. A. Ilchmann,
T. Reis), Differential-Algebraic Equations Forum, Springer, 2013.
A.J. van der Schaft, "On differential passivity", pp. 21–25 in Proc. 9th IFAC
Symposium on Nonlinear Control Systems (NOLCOS), Toulouse, France,
September 4–6, 2013.
A.J. van der Schaft, M.K. Camlibel, "A state transfer principle for switching
port-Hamiltonian systems", pp. 45–50 in Proc. 48th IEEE Conf. on Decision
and Control, Shanghai, China, December 16–18, 2009.
A.J. van der Schaft, B.M. Maschke, “On the Hamiltonian formulation of non-
holonomic mechanical systems”, Rep. Mathematical Physics, 34, pp. 225–
233, 1994
A.J. van der Schaft, B.M. Maschke, “The Hamiltonian formulation of energy
conserving physical systems with external ports”, Archiv für Elektronik und
Übertragungstechnik, 49: 362–371, 1995. .
A.J. van der Schaft, B.M. Maschke, “Hamiltonian formulation of distributed-
parameter systems with boundary energy flow”, Journal of Geometry and
Physics, vol. 42, pp.166–194, 2002.
A.J. van der Schaft, B.M. Maschke, "Conservation Laws and Lumped System
Dynamics", in Model-Based Control; Bridging Rigorous Theory and Advanced
Technology, P.M.J. Van den Hof, C. Scherer, P.S.C. Heuberger, eds., Springer,
ISBN 978-1-4419-0894-0, pp. 31–48, 2009.
A.J. van der Schaft, B.M. Maschke, "Port-Hamiltonian systems on graphs",
SIAM J. Control Optim., 51(2), 906–937, 2013.
A.J. van der Schaft, S. Rao, B. Jayawardhana, "On the Mathematical Struc-
ture of Balanced Chemical Reaction Networks Governed by Mass Action
Kinetics", SIAM J. Appl. Math., 73(2), 953Ð-973, 2013.
S. Seshu and N. Balabanian, Linear Network Analysis, John Wiley & Sons, 1964.
216 References