MA-417-ODE

Lecture 27&28&29

Debanjana Mitra

Department of Mathematics
Indian Institute of Technology Bombay
Powai, Mumbai - 76

October 22&25&26, 2021

Stability theory: Nonlinear system

Linearization method

Liapunov stability
Consider an autonomous system :
Y 0 (t) = f (Y (t)),
where f : Ω → Rn is a given C 1 -function and Ω is an open subset of Rn .

Special solutions: Equilibrium point of the flow.

Definition (Equilibrium point): A point X̂ ∈ Ω is said to be an
equilibrium point of the given autonomous system Y 0 (t) = f (Y (t)), if
f (X̂ ) = 0.
i.e., the zeros of function f . The equilibrium point is also called as a
critical point or singular point or zero or fixed point.
Observe Autonomous system with the initial condition at X̂ :
Y 0 (t) = f (Y (t)), Y (0) = X̂ . Then the unique solution of this IVP is
given by the constant map for all t ∈ R
φ(t; X̂ ) = X̂ , ∀ t ∈ R.
The trajectory starting from X̂ ∈ Ω, an equilibrium point remains at that
point for all t ∈ R.
Our goal is to study the behavior of any trajectory of the autonomous
system around the equilibrium point.
Throughout, we assume the equilibrium point is an ‘isolated equilibrium’
point of the system.
Example:

Find the equilibrium points of the following system. Also determine when
an equilibrium point is isolated.
1. y 0 (t) = y (t)(1 − y (t)).
2. Y 0 (t) = AY (t), where A is a n × n matrix with real entities and
det(A) 6= 0.
3. Y 0 (t) = AY (t), where A is a n × n matrix with real entities and
det(A) = 0.
Definitions of stability
Let X̂ be an isolated point of the ODE-system: Y 0 (t) = f (Y (t)), where
f : Ω → Rn .
Stability: The equilibrium point X̂ is called stable if for any given  > 0,
there exists a δ > 0 (depending on ) such that for any X ∈ Ω satisfying
kX − X̂ k < δ, the IVP: Y 0 (t) = f (Y (t)), Y (0) = X has a unique
solution denoted by φ(t; X ) for all t ≥ 0 and the solution obeys

kφ(t; X ) − X̂ k < , ∀ t > 0.

Asymptotic stability: The equilibrium point X̂ is called asymptotically

stable if for any given  > 0, there exists a δ > 0 (depending on ) such
that for any X ∈ Ω satisfying kX − X̂ k < δ, the IVP:
Y 0 (t) = f (Y (t)), Y (0) = X has a unique solution denoted by φ(t; X )
for all t ≥ 0 and the solution obeys

kφ(t; X ) − X̂ k < , ∀ t > 0, and limt→∞ kφ(t; X ) − X̂ k = 0.

Unstability: The equilibrium point X̂ is called unstable, if it is not stable.

Stability theory for linear system
Consider
Y 0 (t) = AY (t), Y (0) = Y0 ,
where A is a n × n matrix with real entities.

Assume det(A) 6= 0. i.e., 0 is not an eigenvalue.

Note the system has an isolated equilibrium point X̂ = 0 in Rn (at origin).

Recall that the solution of the system is given by φ(t; Y0 ) = e tA Y0 for all
t ∈ R.

Asymptotic stability: We call that the linear system is ‘asymptotically

stable’ (global) if for all Y0 ∈ Rn ,

lim e tA Y0 = 0 in Rn .
t→∞

Recall the characterization of the (global) asymptotic stability property of

a linear system using the eigenvalues of A has been studied. See Lecture
slides 25&26.pdf.
Let Ω be an open domain in Rn and X ∈ Ω, X = (x1 , · · · , xn ).
Let f : Ω → Rn be a differentiable function, where
f (X ) = (f1 (X ), · · · , fn (X )).

The the ‘total derivative’ is an operator Df : Ω ∈ L(Rn ; Rn ), i.e., for all

X ∈ Ω, Df (X ) is a linear operator defined from Rn to Rn . The operator
Df (X ) is represented by
 ∂f1 ∂f1 ∂f1
∂x1 (X ) ∂x2 (X ) ··· ∂xn (X )

 ∂f2 (X ) ∂f1
··· ∂f2
∂x2 (X ) ∂xn (X ) .

Df (X ) =   ∂x1
··· 
∂fn ∂fn ∂fn
∂x1 (X ) ∂x2 (X ) ··· ∂xn (X )

Taylor’s expansion till 1st order: Let f : Ω → Rn be a C 2 function, where

Ω is an open domain in Rn . Then for any X̂ ∈ Ω, there exists a
neighborhood Bδ (X̂ ) of X̂ such that ,

f (X ) = f (X̂ ) + Df (X̂ )(X − X̂ ) + R(X − X̂ ), ∀ X ∈ Bδ (X̂ ),

kR(X −X̂ )k
where kX −X̂ k
→ 0 as kX − X̂ k → 0.
Linearization method

Let us consider the autonomous ODE: for f ∈ C 2 (Ω; Rn ),

Y 0 (t) = f (Y (t)).

Let X̂ ∈ Ω be an isolated equilibrium point of the system.

The linearized system corresponding to the above ODE around the

equilibrium point X̂ : denote Ye (t) = Y (t) − X̂ ;

Ye 0 (t) = Df (X̂ )Ye (t),

noting that f (X̂ ) = 0.

Qn. Can we determine the stability behavior of X̂ for the Nonlinear

ODE-system by the stability behavior of X̂ for the linearized system, i.e.,
by the spectrum of Df (X̂ )?
Ans. Yes, provided all eigenvalues of Df (X̂ ) have non-zero real part.
Main result

Hyperbolic flow: An equilibrium point X̂ of the system Y 0 (t) = f (Y (t))

is called hyperbolic if all eigenvalues of Df (X̂ ) have non-zero real part.
Theorem
Let f ∈ C 2 (Ω; Rn ), where Ω be an open domain in Rn . Let X̂ be an
isolated, hyperbolic equilibrium point of the nonlinear system.
I Then X̂ is a stable equilibrium point of the non-linear system if and
only if X̂ is the stable equilibrium of the corresponding linearized
system.
I Similarly, X̂ is an unstable equilibrium point of the non-linear system
if and only if X̂ is the unstable equilibrium of the corresponding
linearized system.

Warning! If Df (X̂ ) has an eigenvalue with zero real part i.e., the
equilibrium is non-hyperbolic, then No conclusion can be derived for the
nonlinear system from the linearized system.
Examples.

I y10 (t) = sin(y1 (t)) − 4y2 (t), y20 (t) = sin(y2 (t)) − 5y2 (t). Find
equilibrium points. Determine the behavior of the equilibrium point
(0, 0). Is the linearized method applicable here?
p
y10 (t) = y2 (t) + y1 (t) y12 (t) + y22 (t),
I p .
y20 (t) = −y1 (t) + y2 (t) y12 (t) + y22 (t)
Find equilibrium points. Is the linearized method applicable here?
Definitions and notations
In the sequel, we use following notations: for any r > 0, and X̂ ∈ Rn

Br (X̂ ) = {X ∈ Rn | kX − X̂ kRn < r },

the open ball in Rn with center X̂ and radius r .

The boundary of the ball Br (X̂ ) is denoted by

∂Br (X̂ ) = {X ∈ Rn | kX − X̂ kR n = r }.

The closed ball B̄r (X̂ ) = Br (X̂ ) ∪ ∂Br (X̂ ).

Let Ω be an open domain in Rn . Let V : Ω → R be a C 1 -scalar function

and φ(t; X ) be the solution map of the IVP:

Y 0 (t) = f (Y (t)), Y (0) = X ∈ Ω.

Then the derivative of V along the trajectory φ(t; X )) for any X ∈ Ω is

d
defined by dt V (φ(t; X )). Note that

d
V (φ(t; X )) = ∇V (φ(t; x)).f (φ(t; X )), ∀ t ∈ R.
dt
Liapunov stability
Let us consider the autonomous ODE: for f ∈ C 1 (Ω; Rn ),
Y 0 (t) = f (Y (t)).

Let X̂ ∈ Ω be an isolated equilibrium point of the system.

Let Φ(t; X ) denote the solution of the IVP:

Y 0 (t) = f (Y (t)), Y (0) = X ∈ Ω.

Liapunov function: A scalar valued C 1 -function V : Ω → R is called a

Liapunov function for the system corresponding to the equilibrium point
X̂ if V satisfies
1. V (X ) > 0, ∀ X ∈ Ω, X 6= X̂ ,
2. V (X ) = 0 if and only if X = X̂ .
d
3. For all X ∈ Ω, dt V (φ(t; X )) ≤ 0, for all t ≥ 0.

d
Note the condition dt V (φ(t; X )) ≤ 0, for all t ≥ 0, for X ∈ Ω means
that V (·) is non-increasing along the trajectory, i.e., for any t0 < t1 ,
V (φ(t1 ; X )) ≤ V (φ(t0 ; X )).
Main result
Theorem (Liapunov stability)
Let X̂ be an isolated equilibrium point of the autonomous system
Y 0 (t) = f (Y (t)), where f is in C 1 (Ω; Rn ) and Ω is an open domain in
Rn . Let φ(t; X ) be the solution of Y 0 (t) = f (Y (t)), with
Y (0) = X ∈ Ω.
1. If there exists a Liapunov function V for the system and X̂ , then the
equilibrium point X̂ is stable.
2. If there exists a Liapunov function V for the system and X̂ satisfying
d
V (φ(t; X )) < 0, ∀ X ∈ Ω\{X̂ }, ∀t ≥ 0
dt

then the equilibrium point X̂ is asymptotically stable.

Proof
Part 1. Claim: for any  > 0, there exists a δ > 0 such that for all
ξ ∈ Bδ (X̂ ), φ(t; ξ) ∈ B (X̂ ) for all t ≥ 0.
For any given  > 0, set m = min{V (X ) | X ∈ ∂B (X̂ )}.
Proof contd...

Note that m > 0, since V (X ) > 0 for all X ∈ ∂B (X̂ ).

From the continuity of V at X̂ and the fact that V (X̂ ) = 0, it follows
that there exists a δ > 0 such that for all X ∈ Bδ (X̂ ), V (X ) < m .
d
Note that since for all X ∈ Ω, dt V (φ(t; X )) ≤ 0 for all t ≥ 0, we have
V (φ(t; X )) ≤ V (X ) for all X ∈ Ω and for all t ≥ 0. Using this, we get
that for all ξ ∈ Bδ (X̂ ), V (φ(t; ξ)) ≤ V (ξ) < m for all t ≥ 0.

Now our claim is that for all ξ ∈ Bδ (X̂ ), φ(t; ξ) ∈ B (X̂ ) for all t ≥ 0. If
not, then there exists a τ > 0 such that φ(τ ; ξ) ∈ ∂B (X̂ ) and it
contradicts the fact that V (φ(τ ; ξ)) < m .
Thus, X̂ is a stable equilibrium.
Proof contd....
Part 2. From Part 1, it follows that for any  > 0, there exists a δ > 0
such that for all ξ ∈ Bδ (X̂ ), φ(t; ξ) ∈ B (X̂ ) for all t ≥ 0.
Our aim is to show that kφ(t; ξ) − X̂ k → 0 as t → ∞, for all ξ ∈ Bδ (X̂ ).
If ξ = X̂ , then it is obvious. So, let us consider any ξ ∈ Bδ (X̂ )\{X̂ }.

Choose any sequence {tn }n∈N of positive real numbers such that
tn → ∞. Since B̄ (X̂ ) is a compact set in Rn , the sequence
{φ(tn ; ξ) | n ∈ N} ⊂ B̄ (X̂ ) has a convergent subsequence that converges
to a point in B̄ (X̂ ).
Our proof will be complete if we show that any convergent subsequence
of {φ(tn ; ξ) | n ∈ N} converges to the same limit X̂ .

Let {t` }`∈N ⊂ {tn }n∈N be a subsequence such that

{φ(t` ; ξ) | ` ∈ N} ⊂ {φ(tn ; ξ) | n ∈ N} and the subsequence converges to
a limit ξ0 ∈ B̄ (X̂ ), i.e.,

kφ(t` ; ξ) − ξ0 k → 0, as t` → ∞.
 d
Note that, since ξ ∈ Ω\{X̂ }, dt V (φ(t; ξ)) < 0, ∀ t ≥ 0, we have

0 ≤ V (ξ0 ) < V (φ(t` ; ξ)), ∀ t` ,

and thus 0 ≤ V (ξ0 ) < V (φ(t; ξ)) for all t ≥ 0.

To prove our claim, we need to show that V (ξ0 ) = 0. Then by definition

of V , it follows that ξ0 = X̂ .

The proof is via a contradiction argument. If V (ξ0 ) > 0, then by

continuity of V and V (X̂ ) = 0, there exists a η > 0 such that for all
X ∈ Bη (X̂ ), V (X ) < V (ξ0 ). Thus, from the above relation, it follows
that φ(t; ξ) ∈ B̄ (X̂ )\Bη (X̂ ) for all t ≥ 0.
Set m = {−∇V (X ).f (X ) | X ∈ B̄ (X̂ )\Bη (X̂ )}. Note that m > 0 since
d
∇V (X ).f (X ) = dt V (φ(t; X ))|t=0 < 0.

Since we have φ(t; ξ) ∈ B̄ (X̂ )\Bη (X̂ ) for all t ≥ 0, we get
d
dt V (φ(t; ξ)) < −m for all t ≥ 0. Using the fundamental theorem of
calculus, we get
Z t
d
V (φ(t; ξ)) − V (ξ) = V (φ(s; ξ)) ds ≤ −mt.
0 ds
Proof contd..

Consequently, taking t → ∞, V (φ(t; ξ) → −∞ which contradicts the

fact that V (X ) ≥ 0 for all X ∈ Ω.
Thus, it follows that V (ξ0 ) = 0 and hence ξ0 = X̂ .

Finally, since for ξ ∈ Bδ (X̂ ), {φ(t; ξ) | t ≥ 0} ⊂ B̄ (X̂ ) and any

convergent subsequence of {φ(t; ξ) | t ≥ 0} converges to the same limit
X̂ , we conclude kΦ(t; ξ) − X̂ k → 0 as t → ∞.

Thus, X̂ is asymptotically stable.

Theorem (Liapunov unstability)
Let X̂ be an isolated equilibrium point of the autonomous system
Y 0 (t) = f (Y (t)), where f is in C 1 (Ω; Rn ) and Ω is an open domain in
Rn . Let φ(t; X ) be the solution of Y 0 (t) = f (Y (t)), with Y (0) = X ∈ Ω.
If there exists a scalar valued C 1 -function V : Ω → R satisfying
I V (X ) > 0, ∀ X ∈ Ω, X 6= X̂ ,
I V (X ) = 0 if and only if X = X̂ .
d
I For all X ∈ Ω with X 6= X̂ , dt V (φ(t; X )) > 0, for all t ≥ 0.
then the equilibrium point X̂ is unstable.
Proof. For any given  > 0, consider any δ such that 0 < δ < .
Claim: For any ξ ∈ Bδ (X̂ ) with ξ 6= X̂ , there exists a tξ > 0 such that
φ(tξ ; ξ) ∈
/ B̄ (X̂ ).
Since ξ 6= X̂ , V (ξ) > 0. Using the continuity of V at X̂ , since V (X̂ ) = 0,
there exists a η > 0 such that for all X ∈ Bη (X̂ ), V (X ) < V (ξ).
d
Since for X ∈ Ω with X 6= X̂ , dt V (φ(t; X )) > 0, for all t ≥ 0, we get
V (φ(t; ξ)) > V (ξ) for all t > 0, and thus φ(t; ξ) ∈/ Bη (X̂ ) for all t ≥ 0.
Proof contd..

Let m = min{∇ V (X ).f (X ) | X ∈ B̄ (X̂ )\Bη (X̂ )}.

Then we have that m > 0 because
d
∇ V (X ).f (X ) = dt V (φ(t; X ))|t=0 > 0 for all X ∈ B̄ (X̂ )\Bη (X̂ ).
Therefore, if τ > 0 is such that for all t ∈ [0, τ ], φ(t; ξ) ∈ B̄ (X̂ )\Bη (X̂ ),
then using Fundamental theorem of calculus, we obtain
Z τ
d
V (φ(τ ; ξ)) − V (ξ) = V (φ(s; ξ)) ds ≥ mτ,
0 ds

and hence V (φ(τ ; ξ)) ≥ mτ . Further, since V (φ(τ ; ξ)) ∈ B̄ (X̂ ), we have
V (φ(τ ; ξ)) ≤ max{V (X ) | X ∈ B̄ (X̂ )} = M > 0. In particular, we get

M ≥ V (φ(τ ; ξ)) ≥ mτ.

M
The above relation yields that τ has to satisfy the upper bound τ ≤ m.
So, it implies that for some large tξ > M
m , φ(tξ ; ξ) ∈
/ B̄ (X̂ ).
Thus the equilibrium point X̂ is unstable.
Strategy to find Liapunov function
In general, there is no standard rule to find the Liapunov function.
Often the following strategy is used:
1. Depending on the right hand side of the ODE-system, first choice of
the Liapunov function could be
n
X
V (X ) = aj1 xj2 , where X = (x1 , · · · , xn ) ∈ Rn ,
j=1

and determine the coefficients (non-negative) {aj1 | j = 1, · · · , n}

such that the conditions in the definition of Liapunov functions are
satisfied.
2. If the choice in Step 1 does not work then, add more positive terms.
For example, look for the function in the form:
p X
X n
V (X ) = ajk xj2k , where X = (x1 , · · · , xn ) ∈ Rn ,
k=1 j=1

and assign the value of p progressively until the required conditions

for Liapunov function are satisfied.
Examples.

Construct the Liapunov functions for the following system and study the
stability:
y10 (t) = −y1 (t) − 2y2 (t) + y12 (t)y22 (t)
1. y 3 (t)y (t) .
y20 (t) = y1 (t) − 12 y2 (t) − 1 2 2
Ans. Check: V (x1 , x2 ) = α( 12 x12 + x22 ) for any α > 0.

2. y10 (t) = y2 (t), y20 (t) = −y1 (t) − y2 (t) − y13 (t).
Ans. Check: V (x1 , x2 ) = 2(x12 + x22 ) + x14 .
Special case

Ex. Find Liapunov function of the following system: Let f be a

continuous real valued function defined on R with f ≥ 0 and f (x) = 0 iff
x = 0;
y10 (t) = y2 (t), y20 (t) = −f (y1 (t)).
Discuss the stability of the system.