1

§9.6 Liapunov’s Second Method

In section 9.4 we defined the stability of the equilibrium of an autonomous system.
We presented some examples on stability of an equilibrium. These examples were
based on the explicit solution of the system, or the stability of the linearized system.
Is there any other ways or convenient method to determine the stability? The answer
is positive. We will discuss other approach in this section.
Liapunov, a Russian mathematician, put forward a very good idea to determine
the stability of an equilibrium. The method is referred to as a direct method because
no knowledge of the solution of the system of differential equations is required by
constructing a suitable auxiliary function. The technique is a very powerful one that
provides a more global type of information.
Basically Liapunov’s second method is a generalization of the physical principles
that for a conservative system: (1) a rest position is stable if the potential energy is
local minimum, otherwise it is unstable, and (2) the total energy is a constant during
any motion.

9.6.1 Typical examples

To illustrate the idea of the Liapunov’s second method we consider the following
two example: one is from mechanical view, the other is from geometrical view.
Example 9.6.1 Consider the undamped pendulum, which is governed by the
equation
d2 θ r
+ sinθ = 0. (9.6.1)
dt2 l
By introducing new variables we obtain the first order equations corresponding to
(9.6.1)
dx dy g
= y, = − sinx, (9.6.2)
dt dt l
where x = θ is the angular displacement, and y = dθ dt
is the angular velocity.
Omitting an arbitrary constant, the potential energy U is the work done in lifting
the pendulum above its lowest position, namely U = mgh(1 − cosθ) = mgh(1 − cosx).
The kinetic energy is E = m2 l2 ( dθ
dt
)2 = m2 l2 y 2 . (0, 0) and (π, 0) are critical point of
the system (9.6.2). Physically, we expect that (0,0) will be stable since the potential
energy U is minimum, and (π, 0) be unstable since U is a maximum.
Next consider the total energy V, which is the sum of the potential energy U and
the kinetic energy E:
1
V = V (x, y) = mgh(1 − cosx) + ml2 y 2 . (9.6.3)
2
 2

On a trajectory corresponding to a solution x = φ(t), y = ψ(t) of Eqs. (9.6.2), V can

be considered as a function of t:
1
V = V (t) = mgh(1 − cosφ(t)) + ml2 ψ(t)2 . (9.6.4)
2
The derivative of V (t) is called the rate of change of V following the trajectory. By
the chain rule
dV ∂V (x, y) dx ∂V (x, y) dy
= + , (9.6.5)
dt ∂x dt ∂y dt
where it is understood that x = φ(t), y = ψ(t). But dx dt
and dydt
can be obtained in
terms of x and y from Eqs. (9.6.2). Substituting in Eqs. (9.6.4) for dx dt
and dy
dt
, we find
dV
that dt = 0. hence V is a constant along any trajectory of the system (9.6.2), which
is in agreement with the fact that the total energy is constant during any motion of
a conservative system.
It is important to note that at any point (x, y) the rate of change of V along the
trajectory was computed without actually solving the system (9.6.2). It is precisely
that fact that allows us to use Liapunov’s second method for systems whose solutions
we do not know, and hence makes it such an important technique.
At the stable point (0,0), the energy V is zero, which is the minimum of V (x, y) in
the neighborhood of (0,0). If the initial state, say (x1 , y1 ), of the pendulum is sufficient
near the origin (0,0), then the energy V (x1 , y1 ) is small, and the motion(trajectory)
associated with this energy stays close to the critical point. It can be shown that if
V (x1 , y1 ) is sufficiently small, then the trajectory is closed. In fact, we can solve the
equation V (x, y) = V (x1 , y1 ) to obtain
2
y2 = (V (x1 , y1 ) − mgl(1 − cosx)).
ml2
V is a constant along any trajectory and and V (x, y) = V (x1 , y1 ) is closed curve
imply that the origin (0,0) is stable.
Consider the system governing the damped pendulum
dx dy g c
= y, = − sinx − y.
dt dt l ml
The derivative of V (t) following the trajectory is

dV ∂V (x, y) dx ∂V (x, y) dy
= + = −lcψ(t) ≤ 0.
dt ∂x dt ∂y dt
The above inequality implies that

lim φ(t) = 0, lim ψ(t) = 0,

t→+∞ t→+∞
 3

and the (0,0) is asymptotically stable.

Geometrically, Liapunov’s second method is a generalization of the geometrical
technical for the following system
½ dx
dt
= −y − x3 ,
dy (9.6.6)
dt
= x − y3,

The explicit solution of (9.6.6) is not easy to find. Hence we can not use the expression
of the solution and the definition of stability to discuss the stability of the critical
point (0,0). We will investigate the stability of the critical point (0,0) geometrically.
Let x = x(t, x0 , y0 ), y = y(t, x0 , y0 ) be the solution of (9.6.6) with the initial value
x(0) = x0 , y(0) = y0 . The critical point (0,0) is stable if, given any ε > 0, it is
possible to find a δ such that
p
x2 (t, x0 , y0 ) + y 2 (t, x0 , y0 ) < ε
p
for all t > 0 and x20 + y02 < δ. The critical point (0,0) is asymptoticallly stable if
limt→+∞ x(t, x0 , y0 ) = 0, limt→+∞ y(t, x0 , y0 ) = 0.
Consider the function V = V (x, y) = x2 + y 2 . The square of the distance be-
tween (0,0) and (x(t, x0 , y0 ), y(t, x0 , y0 )) is V = V (t) = V (x(t, x0 , y0 ), y(t, x0 , y0 )) =
x2 (t, x0 , y0 ) + y 2 (t, x0 , y0 ). Computing the derivative of V (t) gives

dV (t) dx dy
= 2x + 2y = −2x4 − 2y 4 ≤ 0. (9.6.7)
dt dt dt
(9.6.7) implies that the function V (t) decreases monotonically, that is, the distance
between the (0,0) and (x(t, x0 , y0 ), y(t, x0 , y0 )) is less than the original distance, and
the distance tends to zero. In fact, if V (t) > δ > 0 then from the derivative of V (t)
we see that
dV (t)
= −2V 2 (t)(cos4 θ + sin4 θ) ≤ −αδ,
dt
where α is a positive constant. Integrating the above equation we obtain V (t) ≤
−αδ(t − t0 ) + V0 → −∞, which is a contradiction. Therefore, the critical point (0,0)
is aymptotically stable. The geometrical interpretation of the above process is that
V (x, y) = c is a family of a closed curves, and the point (x(t, x0 , y0 ), y(t, x0 , y0 )) on
the trajectory crosses any closed curve from outside to inside.

9.6.2 The main idea of the method

Consider the system
dx dy
= F (x, y), = G(x, y), ((9.6.8))
dt dt
 4

where F (x, y) and G(x, y) have continuous first partial derivatives, and F (0, 0) = 0,
G(0, 0) = 0,
∂F (0, 0) ∂G(0, 0) ∂F (0, 0) ∂G(0, 0)
− 6= 0.
∂x ∂y ∂y ∂x
Let V (x, y) = C be a family of monotonic increasing closed curves. x = x(t), y = y(t)
be a solution of (9.6.8). If V (t) = V (x(t), y(t)) is a decreasing function of t, then
(x(t), y(t)) must locate inside the curve V (x, y) = V (x0 , y0 ) = V (t0 ) for all t ≥ t0 .
Example 9.6.2 The critical point (0,0) of the system
½ dx
dt
= −y + x(x2 + y 2 − 1)
dy
dt
= x + y(x2 + y 2 − 1)

is asymptotically stable. In fact, we can choose V = x2 + y 2 and calculate the

derivative V̇ = (x2 + y 2 )(x2 + y 2 − 1). V = x2 + y 2 = c are closed curve and
V (t) = x2 (t) + y 2 (t) decreases monotonically.

9.6.3 Definition
Let V = V (x, y) be defined and have continuous first partial derivatives on some
domain D containing the origin. Then V is said to be positive(negative) definite on
D if V (0, 0) = 0 and V (x, y) > (<)0 for all other points in D. If the inequalities >
and < are replaced by ≥ and ≤ then V is said to be positive semidefinite and negative
semidefinite, respectively.
Examples

V = x2 + y 2 , V = x2 + y 4 , V = sin(x2 + y 2 )

are positive definite functions.

2
V = (x − y)2 , V = sin(x − 2y)2 , V = ex − 1

are positive semipositive definite.

Proposition If V (x, y) = ax2 + bxy + cy 2 and a > 0, b2 − 4ac < 0. Then V (x, y)
is a positive definite function, and V (x, y) = h(h > 0) is a closed curve.
Proof Direct calculation implies that V is positive definite. Let x = αs, y = βs
be the line starting from the origin, and V (s) = V (αs, βs) = s2 (aα2 + bαβ + cβ 2 ).
For any h > 0, define f (s) = V (s) − h. It is easy to see that f (s) is continuous,
increasing, and f (0) = −h < 0, f (+∞) = +∞. Therefore, there exists one and only
one s such that f (s) = 0. This means that V (x, y) = h intersects with any line from
origin once. Consequently, V (x, y) = h is a closed curve.
If V (x, y) is positive definite, then it can be shown that V (x, y) = c is a closed
curve for 0 < h << 1.
 5

Along any solution x = x(t) = x(t, x0 , y0 ), y = y(t) = y(t, x0 , y0 ) V (t) =

V (x(t), y(t)) is a function of t. The derivative respect to t is
dV (t) ∂V (x, y) ∂V (x, y)
V̇ = = F (x, y) + G(x, y).
dt ∂x ∂y

V̇ is usually refereed to as the derivative of V with respect to the system (9.6.8).

9.6.4 Liapunov’s Stability Theorems

Theorem 9.3 pp294(stable)
Theorem 9.4 pp294(unstable)
The main ideal of proof.
Examples 9.6.2
dx dy 1
= −x + x3 , = −y + y 5 , V = (x2 + y 2 ).
dt dt 2
dx dy 1
= y + x3 , = x − y3, V = (x2 − y 2 ).
dt dt 2
dx dy 1
= y, = −sinx − y, V = y 2 + 1 − cosx.
dt dt 2
dx dy
= x(1 − y), = y(−1 + x),
dt dt
critical point (0,0), V = x2 , unstable,
critical point (1,1), x = x1 + 1, y = y1 + 1,
dx1 dy1
= −y1 − x1 y1 , = x1 + x1 y1 , V = x1 + y1 − ln(1 + x1 ) − ln(1 + y1 ).
dt dt
1 1 1
V = x − (x − x2 + ...) + y − (y − y 2 ...) = (x2 + y 2 ) + ....
2 2 2
1 1
V̇ = (1 − )(−y1 − x1 y1 ) + (1 − )(x1 + x1 y1 ) = 0.
1 + x1 1 + y1
The Liapunov’s theorem implies that (1,1) is stable.

Example
Study the stability of the positive equilibrium point E ∗ ( 32 , 23 ) for the system

 dx = x(2 − 2x − y),

dt ,
 dy
 = y(2 − x − 2y).,
dt
 6

where x and y are the populations of the two species at time t.

The local stability can be obtained from the linearization of the system at E ∗ . To
study the global stability we define a Liapunov function
2 2 3x 2 2 3y
L = x − − ln + y − − ln .
3 3 2 3 3 2
The derivative of L respect to the system is
dL (x − 23 ) (y − 23 )
= x(2 − 2x − y) + y(2 − x − 2y)
dt x y
2 2 2 2 2 2
= (x − )[−2(x − ) − (y − )] + (y − )[−(x − ) − 2(y − )]
3 3 3 3 3 3
2 2 2 2 2 2
= −2(x − ) − 2(x − )(y − ) − 2(y − ) ≤ 0.
3 3 3 3
∗
Hence the positive equilibrium E is globally asymptotically in the domain x > 0 first
y > 0.

Example
Show that the equilibrium point E 0 (1, 0) of the SIR epidemic model

 dS = µ(1 − S) − βSI,

dt
 dI
 = βSI − (µ + γ)I
dt
β
is globally asymptotically stable if R0 = < 1, where S and I are the proportions
µ+γ
of the susceptibles and infectives at time t, respectively.
To study the global stability we define a Liapunov function
V = I + S − 1 − lnS.
The derivative of V respect to the system is
dV (S − 1)
= [µ(1 − S) − βSI] + βSI − (µ + γ)I
dt S
µ(S − 1)2
=− + βI − (µ + γ)I
S
µ(S − 1)2
=− + (µ + γ)I(R0 − 1) ≤ 0.
S
Hence the disease free equilibrium E 0 is globally asymptotically.

Life is a series of ².