(NC) 3 Stability of Equilibrium Points / 3.

1 Basic Concepts

1. (Stability) is defined for the specific points, not for the system itself.
cf) the system is stable (X) ) the system has any stable equilibrium points (O)

• An equilibrium point is stable if all solutions starting at nearby points stay nearby; oth-
erwise it is unstable
• It is asymptotically stable if all solutions starting at nearby points not only stay nearby,
but also tend to the equilibrium point as time approaches infinity

2. (Equilibrium Point) Let us denote an equilibrium point of ẋ = f (x) as x̄ 2 D; namely f (x̄) = 0.

• The equilibrium point can be always shifted to the origin via a change of variables. For
example, when x̄ 6= 0, if the change of variables y = x x̄ is utilized, since the derivative
of y is given by

ẏ = ẋ = f (x) = f (y + x̄) , g(y) where g(0) = 0,

then the system has an equilibrium point at the origin in the new variable y.
• Without loss of generality, we assume that f (x) satisfies f (0) = 0 and thus we can check
the stability of the origin x = 0.
3. (Definition 3.1) (✏ Requirement for Stability) Let f (x) be a locally Lipschitz function defined
over a domain D 2 <n , which contains the origin, and f (0) = 0. The equilibrium point x = 0 of
ẋ = f (x) is

• stable, for each ✏ > 0, if 9 > 0 (dependent on ✏) 3

kx(0)k < (✏) ! kx(t)k < ✏, 8t 0

• unstable if it is not stable.

• asymptotically stable if it is stable and (region of attraction) can be chosen 3

kx(0)k < ! lim x(t) = 0

t!1

a) For any ✏, we must produce (dependent on ✏) such that a trajectory starting in a

neighborhood of the origin will never leave the ✏ neighborhood.
b) Trying to apply ✏ theory becomes actually finding all solutions of the state equation,
but it may be difficult or even impossible.
c) As an alternative, Lyapunov’s method provides us with a tool to investigate stability of
equilibrium points w/o solving the state equation.
4. (Scalar System) For given one-dimensional system
• the ✏ requirement for stability is
violated if xf (x) > 0. (see Figure 3.1)

• a necessary condition for the origin to

be stable is to have xf (x)  0 in some
neighborhood of the origin. (see Fig-
ure 3.2/3.3)

• the origin will be asymptotically sta-

ble if and only if xf (x) < 0 in some
neighborhood of the origin. (see Fig-
ure 3.3)

• Let us guess any criterion about the

stability using ẋ = f (x)

V̇ (x) , xf (x) = xẋ

+
1
V (x) = x2 > 0 except the origin
2
– if V̇ > 0, unstable
– if V̇  0, stable
– if V̇ < 0, asymptotically stable
5. (Region of Attraction) In Figure 3.3(a), if x0 = x(0) exists within a < x < b, then x(t) ! 0 as
t ! 1, otherwise it is unstable.
6. (Linear Systems) For the diagonlization, let us obtain the eigenvalue decomposition of A in
either real or complex number domain

0
A = M 1 ⇤M $ ⇤ = M AM 1 = 1
0 2

where 1, 2 2 < or 1,2 = ↵ ± j 2 C. For linear time-invariant system, the solution is given

ẋ = Ax ! x(t) = eAt x(0) = M 1 ⇤t

e M x(0) (14)

7. (Theorem 3.1) The equilibrium point x = 0 of ẋ = Ax is stable if and only if

• all eigenvalues of A 2 <n⇥n satisfy Re[ i ]  0 and

• for every eigenvalue with Re[ i ] = 0 and algebraic multiplicity qi 2, rank(A i I) = n qi

The equilibrium point x = 0 is globally asymptotically stable if and only if all eigenvalues of A
satisfy Re[ i ] < 0.
8. (Example 3.1) Assume two same systems having following form are connected by series or by
parallel. Check the stability of series- or parallel-connected system?

ẋ1 = Ax1 + Bu1 ẋ2 = Ax2 + Bu2

y1 = Cx1 y2 = Cx2

where  
0 1 0 ⇥ ⇤
A= B= C= 1 0
1 0 1
• Since the series-connected (u2 = y1 ) or parallel-connected (u2 = u1 ) system has

ẋ1 = Ax1 + Bu1 ẋ1 = Ax1 + Bu1

ẋ2 = Ax2 + Bu2 = Ax2 + B(Cx1 ) = BCx1 + Ax2 ẋ2 = Ax2 + Bu2

the system matrix will be either

2 3 2 3
 0 1 0 0  0 1 0 0
A 0 6 1 0 0 07 A 0 6 1 0 0 07
As = =6 40 0 0
7 Ap = =6 7
BC A 15 0 A 40 0 0 15
1 0 1 0 0 0 1 0

• The matrices Ap and As have the same eigenvalues on ±j with multiplicity qi = 2, for
i = 1, 2. For 1 = j,

rank(As jI) = 3 6= 2(= n q1 ) rank(Ap jI) = 2 = 2(= n q1 )

• By Theorem 3.1, the origin of series-connected system is unstable, but the origin of
parallel-connected system is stable.
 9. (Hurwitz) When Re[ i ] < 0 for i = 1, · · · , n, A is called a Hurwitz matrix. The origin of ẋ = Ax
is asymptotically stable if and only if A is Hurwitz. In this case, its solution satisfies the
inequality

kx(t)k  kkx(0)ke t , 8 t 0 with k > 0 and <0 (15)

where can be chosen as a real part of the maximal eigenvalue among the negative eigenval-
ues i for i = 1, 2, · · · , n

10. (Definition 3.3) (Exponential Stability) Let f (x) be a locally Lipschitz function defined over
a domain D ⇢ <n , which contains the origin, and f (0) = 0. The equilibrium point x = 0 of
ẋ = f (x) is

• exponentially stable if 9 positive c, k and 3 inequality Eq. (15) is satisfied 8 kx(0)k < c.
• globally exponentially stable if the inequality is satisfied for every initial state x(0).
11. (Example 3.2) Show that the origin of ẋ = x3 is asymptotically stable, but not exponentially
stable ?

(1) f (x) = x3 is locally Lipschitz b/c f 0 (x) = 3x2 is continuous and locally bounded for all x
in a domain D ⇢ <.
(2) Since xf (x) < 0, the origin is asymptotically stable. For given initial condition x(0) = a,
the solution cannot leave the compact set {|x|  |a|}. Thus we conclude that it has a
unique solution 8 t 0
Z x(t) Z t x(t)
dx 3 dx 3 1 x(0)
= x ! = dt ! x dx = dt ! = t ! x(t) = p
dt x3 x(0) 0 2x2 x(0) 1 + 2tx2 (0)

(3) Since the solution does not satisfy inequality of the exponential form (15), we know that
it is asymptotically stable, not exponentially stable.
(NC) 3.2 Linearization
1. Stability can be easily checked by seeing the local behavior (convergence or divergence) near
the specific point.

2. (Theorem 3.2) (Lyapunov’s Indirect Theorem) Let x = 0 be an equilibrium point for the nonlin-
ear system ẋ = f (x), where f (x) is continuously differentiable in a neighborhood of the origin.

@f
A= (x)
@x x=0

• the origin is exponentially stable if and only if Re[ i ] < 0 for all eigenvalues of A
• the origin is unstable if Re[ i ] > 0 for one or more of the eigenvalues.
• but, theorem does not say anything about the case when Re[ i ]  0 for all i. In this case,
linearization fails to determine the stability.
3. (Example 3.4) The pendulum system has infinite number of equilibrium points, but consider
only two equilibrium points at (0, 0) and (⇡, 0), with b > 0.

ẋ1 = x2 ẋ2 = sin x1 bx2


0 1
• Jacobian matrix is A = @f@x = cos x1 b

0 1 p
• At (0, 0), A = and 1,2 = 0.5b ± 0.5 b2 4.
1 b
Since all eigenvalues are with Re[ i ] < 0, the origin is exponentially stable.

0 1 p
• At (⇡, 0), A = and 1,2 = 0.5b ± 0.5 b2 + 4.
1 b
Since one eigenvalue of the both is with Re[ i ] > 0, the (⇡, 0) is unstable.
(NC) 3.3 Lyapunov’s Method
1. Reconsider the pendulum equation

ẋ1 = x2 ẋ2 = sin x1 bx2

In fact,

• the origin is stable focus when b > 0 by drawing phase portraits (Example 2.3)
• the origin is exponential stable when b > 0 by linearization (Example 3.4).

As an another approach, the energy concept can be used to determine the stability.
(e.g., The origin is stable when b = 0 and asymptotically stable when b > 0)
2. (Energy Function) Consider the kinetic energy plus potential energy
1 1 ˙ 2 + mg(l
E(x) = x22 + (1 cos x1 ) ( m(l✓) l cos ✓)
2 2
By examining the derivative of E along the trajectories of the system, it is possible to deter-
mine the stability of the equilibrium point.

• When b = 0, the origin x = 0 is a stable equilibrium point b/c

dE
= x2 ẋ2 + sin x1 ẋ1 = x2 sin x1 + sin x1 x2 = 0
dt
• When b > 0, it is a stable equilibrium point b/c
dE
= x2 ẋ2 + sin x1 ẋ1 = x2 sin x1 bx22 + sin x1 x2 = bx22  0
dt
actually, when b > 0, the origin is asymptotically stable, although we cannot show it by
using energy function. (LaSalle’s Theorem)

3. In 1892, Lyapunov showed that certain energy-like functions could be used instead of energy
to determine stability of an equilibrium point.
4. (Theorem 3.3) (Lyapunov’s Direct Theorem) Let f (x) be a locally Lipschitz function defined
over a domain D ⇢ <n , which contains the origin, and f (0) = 0.
Let V (x) be a continuously differentiable function defined over D 3

V (0) = 0 and V (x) > 0 8x 2 D with x 6= 0 (16)

V̇ (x)  0 8x 2 D (17)

Then the origin is a stable equilibrium point of ẋ = f (x).

Moreover, if Eq. (16) holds and

V̇ (x) < 0 8x 2 D with x 6= 0 (18)

then the origin is asymptotically stable

Furthermore, if D = <n , Eqs. (16) and (18) hold 8x 6= 0, and

kxk ! 1 ) V (x) ! 1 (19)

then the origin is globally asymptotically stable

• V (0) = 0 and V (x) > 0 (V (x) < 0) for x 6= 0 : is said to be positive (negative) definite
• V (0) = 0 and V (x) 0 (V (x)  0) for x 6= 0 : is said to be positive (negative) semidefinite
• If V (x) does not have a definite sign, it is said to be indefinite
• V (x) ! 1 as kxk ! 1 : is said to be radially unbounded.
5. (Rephrasing Lyapunov’s Theorem)

• The origin is stable if 9 a continuously differentiable V (x) > 0 3 V̇  0.

• It is asymptotically stable if 9 continuously differentiable V (x) > 0 3 V̇ < 0.
• It is globally asymptotically stable if the conditions for asymptotic stability hold globally
and V (x) is radially unbounded

6. (How to Check Positive Definiteness) For the quadratic form

V (x) = xT P x

where P is a real symmetric matrix, V (x) > 0 if and only if all the eigenvalues of P are positive,
which is also true if and only if all the leading principal minors of P are positive.

7. (Example 3.5) Find the conditions for V (x) > 0 and V (x) < 0, respectively?
2 32 3
⇥ ⇤ a 0 1 x1
V (x) = x1 x2 x3 40 a 25 4x2 5
1 2 a x3

The leading principal minors of P are

2 3
 a 0 1
a 0
det[a] = a > 0, det = a2 > 0, and det 40 a 25 = a(a2 5) > 0
0 a
1 2 a

The leading principal minors of P are a > 0, a2 > 0, and a(a2 5) > 0
p
• a > 5, for V (x) > 0
p
• a< 5, for V (x) < 0
8. (Advantage and Disadvantage of Lyapunov Theorem) Lyapunov’s Theorem can be applied
w/o solving the differential equation ẋ = f (x), but there is no systematic method for finding
Lyapunov functions

• Total Energy
• Variable Gradient Method
(NC) 3.4 The Invariance Principle

1. (Revisited Example 3.6)

ẋ1 = x2 ẋ2 = sin x1 bx2

• The energy Lyapunov function proves just that it is not asymptotically stable but stable
1 dV
V (x) = x22 + (1 cos x1 ) ! = bx22  0
2 dt

• When V̇ = 0, it means x2 = 0. From the dynamics, we have

x2 = 0 ) ẋ2 = 0 ) sin x1 = 0 over D = {|x1 | < ⇡} ) x1 = 0

• The system can maintain the V̇ (x) = 0 condition only at the origin x = 0.

2. (LaSalle’s Invariance Principle) If we can find V (x) > 0 3 V̇ (x)  0 and if we can establish
that no trajectory can stay identically at points where V̇ (x) = 0 except x = 0, then the origin
is asymptotically stable.
(NC) 3.5 Linear Systems and Region of Attraction

1. (For Linear Systems) Consider V (x) = xT P x, where any matrix P = P T > 0, the derivative of
V along ẋ = Ax is given by

V̇ (x) = ẋT P x + xT P ẋ = xT (AT P + P A)x , xT Qx

where Q = QT > 0 defined by

AT P + P A = Q (20)

If 9 Q = QT > 0, we can say by Theorem 3.3 that the origin is globally asymptotically (or
exponentially) stable; that is, A is Hurwitz. Equation (20) is called the Lyapunov equation.

2. (Theorem 3.7) A matrix A is Hurwitz if and only if, for every Q = QT > 0, 9 P = P T > 0 that
satisfies AT P + P A + Q = 0. Moreover, if A is Hurwitz, then P is unique.
3. If the linearization is applied to the nonlinear system ẋ = f (x) and f (0) = 0, we have

@f
ẋ = f (x) = [A + G(x)]x where A = and G(x) ! 0 as x ! 0
@x x=0

• When A is Hurwitz, we can solve Lyapunov equation AT P + P A + Q = 0 for Q > 0, and

use V (x) = xT P x as a Lyapunov function candidate for the nonlinear system. Then

V̇ (x) = ẋT P x + xT P ẋ
= xT [A + G]T P x + xT P [A + G]x
= xT Qx + 2xT P Gx

• Since G(x) ! 0 as x ! 0, for given any 0 < k < 1, we can find r > 0 3 2kP Gk < k min (Q)
in domain D = {kxk < r}.
The origin is exponentially stable in D (region of attraction) b/c
2
V̇ (x)  (1 k) min (Q)kxk