2023 European Control Conference (ECC)

June 13-16, 2023. Bucharest, Romania

Relationship between Systems with Counterclockwise Input-Output

Dynamics and Negative Imaginary Systems
Suleiman Kurawa
Rolls-Royce PLC

Abstract— In this paper, we provide some extension results to the simple stability condition that only depends on the DC
for systems with counterclockwise input-output dynamics and loop gain. NI theory finds potential applications in vibration
give their relationship with negative imaginary systems. We control of lightly damped flexible structures [1], cantilever
give a method of characterizing systems with counterclockwise
input-output dynamics in terms of the state vectors of the beams [4], large space structures [9] and robotic manipulators
system. We do this by using the dissipative property of the [9], in control of nano-positioning systems [10], in control
systems. We provide conditions under which local counter- of large vehicle platoons [11], etc.
clockwise input-output dynamics of a nonlinear system can The nonlinear equivalence of the Kalman-Yakubovich-
be inferred from the negative imaginary property of the Popov lemma for passive systems proposed by [16] has
linearized system. Furthermore, we show that if a system has
the counterclockwise input-output dynamics, then the linearized been applied to various control problems such as nonlinear
system will be a negative imaginary system. Finally, we provide H∞ control [17], [18] and nonlinear feedback stabilization
a numerical example where we use some of the developed result problems [?], [20], [21]. This is due to the fact that syn-
to show that there are nonlinear systems which do not possess thesis and analysis of control system is more intuitive using
the counterclockwise input-output property, but the linearized the algebraic structure of the system than the input-output
system is negative imaginary.
representation.
I. INTRODUCTION It is also common knowledge that the control of nonlinear
The study of counterclockwise (CCW) systems was mo- systems is achieved in most cases by applying linear control
tivated by the study of multistability in positive feedback techniques to the linearized model of the nonlinear systems.
interconnection [5]. CCW dynamics provides a tool for the Local stability of nonlinear systems can be inferred from
convergence analysis of positive feedback in a similar way the linearized model about an equilibrium point. It therefore
that passivity is used for analysing negative feedback. For becomes important to be able to determine how good the
periodic input and output, the counterclockwise condition is linear model captures the properties of the nonlinear systems.
equivalent to having the signed area encircled by a curve in The main contributions of this paper are therefore
the input-output plane to be non-negative [5]. 1) To provide a characterization of CCW systems that
Since its inception, CCW dynamics theory has been used relates the state equations of the system to its input-
in a number of applications. It was used in the stability output behaviour. This will provide a better tool for the
analysis of systems with Duhem hysteresis [12], [13] and analysis and synthesis of the CCW systems. Our result
the study of hysteretic models [14]. CCW systems theory is different from that of [5], where the CCW property
was also used for analysis of stability and convergence of was treated via an input-output approach.
neural networks [15]. 2) To show that the CCW dynamics of a nonlinear system
Negative Imaginary (NI) systems was introduced in [1]. is equivalent to the NI property of the first order approx-
The theory was motivated by the study of inertial system with imation of the nonlinear system. This will enable the
colocated position sensors and force actuators [1], [2]. NI well documented literature of NI control and synthesis
theory has unified many existing vibration control techniques to be applied to systems with CCW dynamics.
such as positive position feedback [3] and integral resonant 3) To show the conditions under which the NI property of
control [4], used for the control of lightly damped flexible a system is equivalent to the CCW dynamics of the non-
structures. linear system. This is important because there are some
Negative imaginary systems is closely related to coun- systems, such as neural networks where CCW dynamics
terclockwise input-output dynamics in a nonlinear setting facilitate convergence analysis. However, there is no
[5], to input-output Hamiltonian systems in both linear and easy method of defining a storage function for this sys-
nonlinear setting [6], [7] and to a class of dissipative systems tem to determine the property. It is therefore important
involving the input to the system and the time-derivative of that we can be able to infer the CCW dynamics property
the system’s output [8]. from the linearized system, since the NI property can
NI systems theory has been used for a number of practical easily be checked via linear matrix inequalities.
applications for controller design and stability analysis due The notations are standard throughout. R≥0 and R>0 de-
note respectively the sets of all non-negative and all positive
The author is with Rolls-Royce Plc, Derby-A Site-
Sinfin Complex. Any correspondence should be directed to: real numbers. C denotes the set of all complex numbers.
[email protected] Rm×n and Cm×n denote respectively the sets of real and

978-3-907144-09-1 ©2023 EUCA 258

complex matrices of dimensions (m × n). ℜ(·) and ℑ(·)
express the real and the imaginary parts respectively. A⊤ , A∗
(
ẋ = f (x) + g(x)u,
and Ā denote the transpose, the complex conjugate transpose Σ: (4)
y = h(x),
and the complex conjugate of
a matrix A.
A−∗ and A−⊤
∗ ⊤
represent shorthand for A−1 and A−1 respectively. where f : Rn → Rn is Lipschitz continuous function,
For a transfer function matrix M (s), M (jω) = M (−jω)⊤
∗
h : Rn → R is of class C 1 , x ∈ Rn , u ∈ Rm and y ∈ Rm .
and M (s)∗ = M (s̄)⊤ where s̄ denotes the complex conju- Let x = 0 be an equilibrium point of (4) such that f (0) = 0
gate of s. R m×n denotes the set of all proper, real, rational and h(0) = 0. The input u(t), state trajectory x(t) and output
transfer function matrices and RH m×n∞ denotes the set of y(t) are all real valued locally square integrable functions.
all proper, real, rational and asymptotically stable transfer The state transition function Φ(t1 , 0, x(0), u) is the state
function matrices, both of dimensions (m × n). x(t1 ) of the system Σ at time t1 , with initial condition x(0),
when input u(t) is applied to the system for time t ∈ [0, t1 ].
II. T ECHNICAL PRELIMINARIES
Definition 2: (Reachability) [16] System (4) is com-
In this section, we present essential technical preliminar- pletely reachable if for a given x1 and t1 , there exist a t0 ≤ t1
ies, definitions and lemmas which underpin the proofs of the and a locally square integrable u(·) such that the state can
main results of the paper. be driven from x(t0 ) = 0 to x(t1 ) = x1 .
Definition 3: (Dissipative Systems) [28] The system (4)
A. Briefs of NI theory
is said to be dissipative with respect to a supply rate w(u, y)
Definition 1: (NI System) [9], [22] Let G(s) be the real, if there exist a storage function V : Rn → R≥0 such that
rational and proper transfer function matrix of a finite-
dimensional, square, LTI system. Then, G(s) is said to be
Z T

NI if V (x(0)) + w(u, y)dt ≥ V (x(T )) (5)

0
i) G(s) has no poles in {s ∈ C : ℜ[s] > 0};
ii) j[G(jω) − G(jω)∗ ] ≥ 0 for all ω ∈ (0, ∞) except the with T ∈ [ 0, ∞) , x(0) ∈ Rn , u(t) ∈ R for all t ∈ R≥0 and
values of ω where s = jω is a pole of G(s); x(T ) and w(u, y) evaluated along the trajectories of (4).
iii) If s = jω0 with ω0 ∈ (0, ∞) is a pole of G(s), then it is Remark 2: If the storage function V (x) is atleast of class
at most a simple pole and the residue matrix lim (s − C 1 , then the dissipative inequality (5) can be written in the
s→jω0 differential form as V̇ (x) ≤ w(u, y). The supply rate w(u, y)
jω0 )jG(s) is Hermitian and positive semidefinite. can be thought of as the input power in physical systems.
The following lemma, referred to as the NI Lemma, provides Equation (5) simply means that the amount of energy in a
a state-space characterisation for NI systems without poles system at any given time is at most equal to the initial energy
at the origin. of the system and the energy supplied to the system.
Lemma 1: (NI Lemma) [26] Let G(s) be the real, ra-
Definition 4: (CCW System) [5] The system (4) is said
tional and proper transfer function matrix of a square
to have the CCW input-output dynamics if
and causal system having a minimal state-space realisation
(A, B, C, D), where A ∈ Rn×n , B ∈ Rn×m , C ∈ Rm×n Z T
and D ∈ Rm×m . Then, G(s) is NI without poles at the lim inf ẏ(t)T u(t)dt > −∞ (6)
T →+∞ 0
origin if and only if det(A) ̸= 0, D = D⊤ and there exist a Lemma 3: (CCW Lemma) [5] Σ is said to have CCW
real matrices P = P ⊤ > 0, L ∈ Rm×n , W ∈ Rm×m such dynamics if there exists a non negative function V (x) :
that Rn → R≥0 of class C 1 , with V (0) = 0 such that for all
x ∈ R n , u ∈ Rm
P A + AT P P B − AT C T
 
 
= ∂V (x) ∂h(x)
B T P − CA −(CB + B T C T ) ẋ ≤ ẋ u, (7)
∂t ∂t
−LT L −LT W
 
≤ 0. (1)
−W T L −W T W holds.
Remark 1: The LMI (1) can be reformulated into [1] Remark 3: The system is said to have R T strictly
T
CCW
AY + Y A⊤ ≤ 0 and B + AY C ⊤ = 0, (2) (SCCW) dynamics if lim  T →+∞inf 0 ẏ(t) u(t) −
Ψ(t)dt > −∞ or ∂V∂t(s) ẋ ≤ ∂h(x) ∂t ẋ u − Ψ(x) for some
where Y = Y T = P −1 .
positive definite function Ψ(·).
Lemma 2: (NI Lemma) [27] Let (A, B, C, D) be the
minimal realization of a real, rational proper transfer function
matrix G(s). Then G(s) is NI if and only if III. M AIN RESULTS

V̇ (x(t)) ≤ ẏ(t)u(t) (3) In this section, we will present the main contributions of
this paper. Here, we consider the system given by (4) with
for all t ≥ 0, where V = 12 xT P x, and P satisfies LMI (1). the subsequent assumptions to ensure that a solution exist
Consider the class of nonlinear analytic system given by forward in time.

259
A. Counterclockwise Characterization Comparing coefficients, we get
Theorem 1: Let Σ be a nonlinear system given by (4).
Suppose Σ is reachable from the origin and is dissipative
∂ T V (x)
with respect to a supply rate w(u, ẏ) = ẏu and a storage f (x) = −L(x)T L(x) (13)
function V (x) ≥ 0, where V (0) = 0 exists. Then Σ is CCW ∂x
1 T ∂V (x) ∂h(x)
if and only if there exist real functions L(x) : Rn → Rz and g (x) = f (x) − W (x)T L(x) (14)
W (x) : Rn → Rz×m satisfying 2 ∂x ∂x
∂ T h(x) ∂h(x)
g T (x) + g(x) = W (x)T W (x). (15)
∂ T V (x) ∂x ∂x
f (x) = −LT (x)L(x) (8)
∂x We obtain the desired result with V (x) = V (x).
1 T ∂V (x) ∂h(x) Sufficiency: Using (8)-(10), we have
g (x) = f (x) − W T (x)L(x) (9)
2 ∂x ∂x
T T
∂h(x) ∂ h(x) 
∂h(x) ∂h(x)
g(x) + g T (x) = W T (x)W (x) (10) w(u, ẏ) = 2ẏ T u = 2 f (x) + g(x)u u=
∂x ∂x ∂x ∂x
Proof
Necessity: Since Σ is reachable from the origin, there exist ∂ T V (x)
2L(x)T W (x)u + g(x)u + uT W (x)T W (x)u =
u(·) defined on [t0 , 0] such that x(t0 ) = 0 and x(0) = x0 . ∂x
Since Σ is dissipative with respect to the supply rate w(u, ẏ), ∂ T V (x)
2L(x)T W (x)u + (f (x) + g(x))u−
there exist V (x) ≥ 0, with V (0) = 0 such that ∂x
T
∂ V (x)
Z T Z 0 Z T g(x) + uT W (x)T W (x)u = 2L(x)T W (x)u+
∂x
w(u, ẏ)dt = w(u, ẏ)dt + w(u, ẏ)dt,
∂ T V (x)
t0 t0 0 ẋ + LT (x)L(x) + uT W (x)T W (x)u.
Z T Z T ∂x
w(u, ẏ)dt ≥ − w(u, ẏ)dt,
0 t0 The above equality condition can easily be factored as
R0
where t0 w(u, ẏ)dt only depends on x0 . Therefore, there T
exist a function C : Rn → R such that w(u, ẏ) = (L(x) + W (x)u) (L(x) + W (x)u) +
Z T ∂ T V (x) ∂ T V (x)
w(u, ẏ)dt ≥ C(x0 ) > −∞. ẋ ≥ ẋ = V̇ (x).
∂x ∂x
0
Hence, the available storage is bounded by Hence, using Lemma 3, the system Σ is CCW since V̇ ≤
Z T
2ẏu.
0 ≤ V (x) = − sup w(u, ẏ)dt < ∞.
u(·),T ≥0 0 This completes the proof.
Dissipativeness of Σ implies that V (0) = 0 and V (x) is the ■
storage function such that Corollary 1: Let Σ be a nonlinear system satisfying the
conditions of Theorem 1 and ∂h(x)
∂x g(x) = 0. Then Σ is
Z t CCW if and only if
V (x(t)) − V (x(0)) ≥ w(u, ẏ)dt ∀t ≥ 0,
0
Z t 
dV (x(τ )) ∂V (x)
w(u, ẏ) − dτ ≥ 0, ∀t ≥ 0. f (x) ≤ 0 (16)
0 dt ∂x
Taking the derivative of the above inequality gives ∂V (x) ∂h(x)
dV (x)
g(x) = f (x) (17)
dt ≤ w(u, ẏ) along the trajectories of Σ. Let E(x, u) : ∂x ∂x
Rn × Rm → R be such that
B. Relationship Between NI and CCW Systems
dV (x)
E=− + w(u, ẏ) ≥ 0 (11) In this section, we will use a first order Taylor series
dt
approximation to establish the relationship between CCW
 
∂V (x) ∂h(x) ∂h(x)
E=− ẋ + w u, f (x) + g(x)u . and NI systems. We will first present some preliminary
∂x ∂x ∂x
(12) results before stating the main contributions of this section.
E(x, u) is a quadratic function in u and can be factored Consider the nonlinear Σ system given by (4)
as E(x, u) = (L(x) + W (x)u)T (L(x) + W (x)u) ≥ 0. Then (
ẋ = f (x) + g(x)u,
(12) can be written as Σ:
y = h(x).
∂ T V (x) ∂ T h(x)
− (f (x) + g(x)u) + 2f T (x) u+
∂x ∂x
∂ T h(x) Then ẏ = ∂h(x) ∂h(x)
∂x f (x) + ∂x g(x)u = h̃(x) + J(x)u. Let
2uT g T (x) u = L(x) + W (x)u)T (L(x) + W (x)u .


∂x
∂f (x)
A = ∂x |x=0 , B = g(0) and C = ∂h(x)
∂x |x=0 . Using Taylor

260
series expansion, The above inequality holds since the poles of ΣL have a
∂h(x) ∂h(x) minimum decay rate of 21 ϵ.
h̃(x) = f (x) = f (x)|x=0 + Let V = 12 xT P x also be a local storage function of Σ.
∂x  ∂x
 
∂ ∂h(x)
 Then
f (x) |x=0 x + H(x) = CAx + H(x)
∂x ∂x uT ẏ − V̇ = uT (CAx + H(x) + (CB + M (x)) u)
(18a) ∂V
∂h(x) ∂h(x) − (Ax + F (x) + (B + G(x))u) ,
J(x) = g(x) = g(x)|x=0 + M (x) ∂x
∂x ∂x
= CB + M (x) (18b)
   uT ẏ − V̇ = uT (CAx + CBu) + uT (H(x) + M (x)u)
∂f (x)
f (x) = f (x)|x=0 + |x=0 x + F (x) 1
∂x − xT P (Ax + Bu) − xT P (F (x) + G(x)u) ≥ ϵxT P x
2
= Ax + F (x) (18c) (24)
g(x) = g(x)|x=0 + G(x) = B + G(x) (18d) + uT (H(x) + M (x)u) − xT P (F (x) + G(x)u) = Θ(x, u).
  
∂h(x)
h(x) = h(x)|x=0 + |x=0 x + H̄(x) We would like Θ(x, u) ≥ 0 around the neighbourhood of
∂x the equilibrium point x = 0, u = 0.
= Cx + H̄(x), (18e)
where H(x), M (x), F (x), G(x) and H̄(x) are functions 1 T
Θ(x, u) ≥ ϵx P x + uT (H(x) + M (x)u)
that contain higher order terms. Using (18a)-(18e), the non- 2
linear system Σ can be written as − xT P (F (x) + G(x)u)
The rest of the proof is similar to [29, Theorem 2] and is
(
ẋ = Ax + F (x) + (B + G(x)) u,
Σ: (19) thus omitted for brevity. ■
y = Cx + H̄(x),
Theorem 2 gives conditions for which we can infer the
with ẏ = CAx + H(x) + (CB + M (x)) u. The linearized CCW dynamics property of a system from the linearized
system ΣL , obtained from the nonlinear system Σ is given dynamics. Next, we show that linearising a CCW system
by results in an NI system.
( Theorem 3: Consider the nonlinear system Σ given by (4).
ẋ = Ax + Bu, Suppose Σ has CCW dynamics, then ΣL is NI and there exist
ΣL : (20)
y = Cx, matrices P ≥ 0, L and W of appropriate dimensions such
that
where A = ∂f∂x (x)
|x=0 , B = g(0) and C = ∂h(x)
∂x |x=0 .
Theorem 2: Let Σ be nonlinear system given by (4) and AT P + P A = −LT L (25)
T T
ΣL be the linearized system given by (20), with minimal state P B = (CA) − L W (26)
space realization. Suppose ΣL is stable NI having poles in T
CB + (CB) = W W, T
(27)
ℜ[s] ≤ − 21 ϵ and there exists γ ≥ 0 such that
∂f (x) ∂h(x)
∥J(x) − CB∥
where A = ∂x |x=0 , B = g(0) and C = ∂x |x=0 .
lim ≤ γ, (21) Proof
∥x∥ 2 →0 ∥x∥2 Since Σ is CCW, then there exist a non-negative storage
then Σ is locally CCW. function V (x) : Rn → R, with V (0) = 0 such that (8)-(10)
Proof hold. V (x) can be expanded using Taylor series as
Since by supposition ΣL is stable NI, the Lemmas 2 and
1 ensure that ẏ(t)u(t) − V̇ (x(t)), with V (x(t)) = 12 xT P x V (x) = xT P x + Vh (x), (28)
and where Vh (x) : Rn → R contains the higher order terms of
T T
P A + A P = −L L − ϵP V (x). Using (28), (18a)-(18e) and the fact that
T
P B = (CA) − LT W (22) L(x) = Lx + Lh (x) (29)
T W (x) = W + Wh (x)
W T W = CB + (CB) , (30)
are satisfied. where Lh (x) and Wh (x) contain higher order terms of L(x)
∂V and W (x) respectively, we can re-write (8)-(10) via the
uT ẏ − V̇ = uT (CAx + CBu) − ẋ ≥ 0, Taylor series expansion. Equation (8) becomes
∂x
 
T T T ∂Vh (x)
u ẏ − V̇ = u (CAx + CBu) − x P (Ax + Bu) 2P x + (Ax + F (x)) +
∂x
1
≥ ϵxT P x. (23) T
(Lx + Lh (x)) (Lx + Lh (x)) = 0
2

261
Simplifying the above equation gives Σ can be linearized as follows
" #
∂x2 ∂x2
∂f (x) ∂x1 ∂x2
A= = −∂ sin(x1 ) −∂ sin(x1 ) =
xT P A + AT P + LT L x + Ω(x) = 0,


(31) ∂x x=0 ∂x1 ∂x2
  x=0
T T
where Ω(x) = 2xT P F (x) + ∂ V∂xh (x) Ax + ∂ V∂xh (x) F (x) + 0 1
2xT LT Lh (x) + LTh (x)Lh (x). Similarly, (9) becomes −1 0
 
0
B = g(0) =
1

∂Vh (x)
 1
T
2x P + (B + G(x)) − (CAx + H(x)) ∂h(x) h i
2 ∂x C= = ∂ sin(x 1) ∂ sin(x1 )
∂x x=0 ∂x1 ∂x2 x=0
T
= (Lx + Lh (x)) (W + Wh (x)) ,  
= 1 0 .
which can be simplified as
 Using  Remark 1 of Lemma 1 and choosing Y =
1 0
, (2) becomes
xT P B − (CA)T + LT W + Π(x) = 0,


(32) 0 1

where Π(x) = xT P G(x) + 12 ∂V∂x h (x)

B + 21 ∂V∂x
h (x)
G(x) − AY + Y AT =
T T T T T
H (x) + x L Wh (x) + Lh (x)W + Lh (x)Wh (x).      
0 1 1 0 1 0 0 1
Taking (31) and (32) as the Taylor series expansion of (8) + =0
−1 0 0 1 0 1 −1 0
and (9) respectively and noting that      
0 0 1 1 0 1
∥Ω(x)∥ ∥Π(x)∥ B + AY C T = + = 0.
lim = 0, lim = 0, 1 −1 0 0 1 0
∥x∥→0 ∥x∥2 ∥x∥→0 ∥x∥
Hence, the linearized system is lossless NI. Next, we show
it follows that P satisfies (25) and (26). Now, (27) follows that the nonlinear system Σ is not  CCW. First, notice
by setting x = 0 in (10).

∂h(x)   0
This completes the proof. ■ that ∂x g(x) = cos(x1 ) 0 = 0. Hence, Σ
1
will have CCW dynamics if conditions of Corollary 1 are
IV. N UMERICAL E XAMPLES satisfied. From (16), we have
In this section, we will use some numerical example to ∂V (x)
 
 ∂V ∂V
 x2
elucidate the efficacy of the results presented in this paper. f (x) = ∂x ∂x2 ≤0
∂x 1 − sin(x1 )
∂V ∂V
x2 − sin(x1 ) ≤ 0. (33)
∂x1 ∂x2
𝑢 From (17), we obtain
l= 1 m
 
 ∂V ∂V
 0
∂x1 ∂x2 =
1
𝜑 𝑚 = 1 kg i 
h
∂ sin(x1 ) ∂ sin(x1 ) x 2
∂x1 ∂x2 − sin(x1 )
Fig. 1: A simple pendulum ∂V  

x2

= cos(x1 ) 0 = x2 cos(x1 ). (34)
∂x2 − sin(x1 )
Example 1: Consider the dynamics of a simple pendulum
shown in Fig.1. The pendulum is assumed to be of mass Integrating (34) with respect to x2 , we get
m = 1kg and length l = 1m. The input to the system is 1 2
V = x cos(x1 ) + F (x1 ). (35)
u and the angular position is φ. The desired output of the 2 2
system is the horizontal position of the pendulum. If we Substituting (35) into (33) yields
take x1 = φ, then the dynamics of the pendulum can be  
1 dF (x1 )
represented as − x22 sin(x1 ) + x2 − x2 cos(x1 ) sin(x1 ) ≤ 0,
 2 dx1
ẋ1 = x2
 
 1 dF (x1 )
Σ : ẋ2 = − sin(x1 ) + u − x32 sin(x1 ) + x2 − cos(x1 ) sin(x1 ) ≤ 0.
 2 dx1

y = sin(x1 ), (36)

x2
  
0 For a fixed x1 , we can write the left hand side of the
where f (x) = and g(x) = and h(x) = above inequality as ax32 + bx2 . For a very small x2 ,
− sin(x1 ) 1
sin(x1 ). We will use Theorem 1 to show that the nonlinear the
h linear term dominates i and (36) means that b =
dF (x1 )
system Σ is not CCW even though the linearized system is dx1 − cos(x 1 ) sin(x 1 ) = 0 and − 12 x32 sin(x1 ) ≤ 0 for
NI. all x1 , x2 close to zero. This is a contradiction, which implies

262
that there exist no storage function and hence, the system is [17] A. van der Schaft, L2-Gain and Passivity Techniques in Nonlinear
not CCW. Control, 3rd ed. Springer Publishing Company, Incorporated, 2016.
[18] M. Aliyu, Nonlinear H-infinity control, Hamiltonian systems and
V. CONCLUSIONS Hamilton-Jacobi equations. Hoboken, NJ: CRC Press, 2011.
[19] “A positive real condition for global stabilization of nonlinear sys-
In this paper, we presented a method of characterising tems,” Systems & Control Letters, vol. 13, no. 2, pp. 125–133, 1989.
CCW systems using the state vectors of the system via a [20] R. A. Freeman and P. V. Kototovic, Robust Nonlinear Control Design:
State-Space and Lyapunov Techniques. USA: Birkhauser Boston Inc.,
nonlinear-like KYP lemma. We provided conditions under 1996.
which the NI property of a linearized system infers the CCW [21] R. Sepulchre, M. Jankovic, and P. Kokotovic, “Constructive nonlinear
property of the underlying nonlinear system. Moreover, we control,” 1997.
[22] A. Lanzon and H.-J. Chen, “Feedback stability of negative imaginary
also showed that linearising a CCW system will result in an systems,” IEEE Transactions on Automatic Control, vol. 62, no. 11,
NI system. Finally, we used some of the result presented in pp. 5620–5633, Nov 2017.
this paper to show that a linearized system having the NI [23] A. Ferrante and L. Ntogramatzidis, “Some new results in the theory of
negative imaginary systems with symmetric transfer matrix function,”
system property does not necessarily imply CCW dynamics Automatica, vol. 49, no. 7, pp. 2138–2144, July 2013.
of the original nonlinear system. [24] A. Ferrante, A. Lanzon, and L. Ntogramatzidis, “Foundations of
not necessarily rational negative imaginary systems theory: Relations
R EFERENCES between classes of negative imaginary and positive real systems,” IEEE
Transactions on Automatic Control, vol. 61, no. 10, pp. 3052–3057,
[1] A. Lanzon and I. R. Petersen, “Stability robustness of a feedback inter- Oct 2016.
connection of systems with negative imaginary frequency response,” [25] M. Liu and J. Xiong, “On non-proper negative imaginary systems,”
IEEE Transactions on Automatic Control, vol. 53, no. 4, pp. 1042– Systems & Control Letters, vol. 88, pp. 47–53, Feb 2016.
1046, May 2008. [26] M. A. Mabrok, A. G. Kallapur, I. R. Petersen, and A. Lanzon,
[2] I. R. Petersen and A. Lanzon, “Feedback control of negative-imaginary “Stability analysis for a class of negative imaginary feedback sys-
systems,” IEEE Control Systems Magazine, vol. 30, no. 5, pp. 54–72, tems including an integrator,” in 2011 8th Asian Control Conference
Oct 2010. (ASCC), 2011, pp. 1481–1486.
[3] J. L. Fanson and T. K. Caughey, “Positive position feedback control [27] A. G. Ghallab, M. A. Mabrok, and I. R. Petersen, “Extending neg-
for large space structures,” AIAA Journal, vol. 28, no. 4, pp. 717–724, ative imaginary systems theory to nonlinear systems,” in 2018 IEEE
1990. Conference on Decision and Control (CDC), 2018, pp. 2348–2353.
[4] B. Bhikkaji, S. O. Reza Moheimani, and I. R. Petersen, “A negative [28] J. C. Willems, “Dissipative dynamical systems part I: General theory,”
imaginary approach to modeling and control of a collocated structure,” Archive for Rational Mechanics and Analysis, vol. 45, no. 5, pp. 321–
IEEE/ASME Transactions on Mechatronics, vol. 17, no. 4, pp. 717– 351, Jan 1972.
727, Aug 2012. [29] M. Xia, P. Antsaklis, V. Gupta, and M. McCourt, “Passivity and
[5] D. Angeli, “Systems with counterclockwise input-output dynamics,” dissipativity of a nonlinear system and its linearization technical report
IEEE Transactions on Automatic Control, vol. 51, no. 7, pp. 1130– of the isis group at the university of notre dame isis-12-008 september
1143, July 2006. 2012,” 2012.
[6] A. J. van der Schaft, “Positive feedback interconnection of Hamilto-
nian systems,” in Proceedings of 50th IEEE Conference on Decision
and Control and European Control Conference, Dec 2011, pp. 6510–
6515.
[7] ——, “Interconnections of input-output Hamiltonian systems with
dissipation,” in Proceedings of 55th IEEE Conference on Decision
and Control, Dec 2016, pp. 4686–4691.
[8] P. Bhowmick and A. Lanzon, “Output strictly negative imaginary
systems and its connections to dissipativity theory,” in Proceedings
of 58th IEEE Conference on Decision and Control, Dec 2019, pp.
6754–6759.
[9] M. A. Mabrok, A. G. Kallapur, I. R. Petersen, and A. Lanzon,
“Generalizing negative imaginary systems theory to include free
body dynamics: Control of highly resonant structures with free body
motion,” IEEE Transactions on Automatic Control, vol. 59, no. 10,
pp. 2692–2707, Oct 2014.
[10] ——, “Spectral conditions for negative imaginary systems with appli-
cations to nanopositioning,” IEEE/ASME Transactions on Mechatron-
ics, vol. 19, no. 3, pp. 895–903, June 2014.
[11] C. Cai and G. Hagen, “Stability analysis for a string of coupled
stable subsystems with negative imaginary frequency response,” IEEE
Transactions on Automatic Control, vol. 55, no. 8, pp. 1958–1963,
Aug 2010.
[12] R. Ouyang and B. Jayawardhana, “Absolute stability analysis of linear
systems with duhem hysteresis operator,” Automatica, vol. 50, no. 7,
pp. 1860–1866, 2014.
[13] B. Jayawardhana, R. Ouyang, and V. Andrieu, “Stability of systems
with the duhem hysteresis operator: The dissipativity approach,”
Automatica, vol. 48, no. 10, pp. 2657–2662, 2012.
[14] A. K. Padthe, JinHyoung Oh, and D. S. Bernstein, “Counterclockwise
dynamics of a rate-independent semilinear duhem model,” in Proceed-
ings of the 44th IEEE Conference on Decision and Control, 2005, pp.
8000–8005.
[15] D. Angeli, “Convergence in networks with counterclockwise neural
dynamics,” IEEE Transactions on Neural Networks, vol. 20, no. 5,
pp. 794–804, 2009.
[16] D. J. Hill and P. J. Moylan, “Stability results for nonlinear feedback
systems,” Automatica, vol. 13, no. 4, pp. 377–382, July 1977.

263