Chapter 10

Qualitative Theory of Differential

Equations

by

Dwijendra Narain Pandey

Department of Mathematics
Indian Institute of Technology Roorkee
Roorkee 247667, Uttarakhand.
E-mail: [email protected]
Module 3: Stability of linear systems.

1 Paths of linear system

A general linear system may be given as
dx
= ax + by
dt
dy
= cx + dy (1)
dt
where a, b, c, and d are real constants. Clearly, the origin (0, 0) is a critical
point of (1). Further the condition
a b
6= 0
c d
ensures that (0, 0) is the only critical point of (1). The characteristic equation
of (1) is given by
λ2 − (a + d)λ + (ad − bc) = 0, (2)
Consider the following Lemma which provide the proof of (i) and (ii) part of
the Theorem given in previous module.
Lemma 1. The critical point (0, 0) of the linear system (1) is a node, if the
roots λ1 and λ2 of the characteristic equation (2) are real, unequal, and of the
same sign.
Proof. Let us assume that λ1 and λ2 are both negative with λ1 < λ2 < 0.
Then, the general solution of (1) is given by
x = c 1 u 1 e λ 1 t + c 2 u2 e λ2 t
y = c 1 v 1 e λ1 t + c 2 v 2 e λ2 t (3)
where u1 , v1 , u2 , and v2 are definite constants, u1 v2 6= u2 v1 and c1 and c2 are
arbitrary constants.
• Let us first consider that c2 = 0. Then the solution (3) reduces to
x = c 1 u1 e λ1 t
y = c 1 v 1 e λ1 t . (4)
We can find the orbit of the solution by eliminating λ1 from (4) and we
get the straight line v1 x = u1 y as the orbit of the solution. For any c1 > 0,
the solution (4) represents a path consisting of half of the line and for any
c1 < 0, the solution (4) represents a path consisting of other half of the
line. Since λ1 < 0, both of these half-line paths approach (0, 0) as t → +∞
and enter (0, 0) with slope v1 /u1 .

1
 • Similarly, by choosing c1 = 0, we get the solution
x = c 2 u 2 e λ2 t
y = c 2 v 2 e λ2 t . (5)
eliminating λ2 from (5), we get the straight line v2 x = u2 y. For any
c2 > 0, the solution (5) represents a path consisting of half of the line and
for any c2 < 0, the solution (5) represents a path consisting of other half
of the line. Since λ2 < 0, both of these half-line paths approach (0, 0) as
t → +∞ and enter (0, 0) with slope v2 /u2 .
Therefore, the solutions (4) and (5) gives four half-line paths which ap-
proach and enter (0, 0) as t → +∞.

• Now consider the case when c1 6= 0 and c2 6= 0. In this case the general
solution (3) represents non-rectilinear paths. Since λ1 < λ2 < 0, all of
these paths approach (0, 0) as t → +∞. Also,
y c 1 v 1 e λ 1 t + c 2 v 2 e λ2 t
lim = lim
t→∞ x t→∞ c1 u1 eλ1 t + c2 u2 eλ2 t

(c1 v1 /c2 )e(λ1 −λ2 )t + v2 v2

= lim (λ )t
=
t→∞ (c1 u1 /c2 )e 1 −λ 2 + u2 u2
⇒ all of these paths enter(0, 0)with limiting slope v2 /u2 .

Hence, all the paths (both rectilinear and non-rectilinear) enter (0, 0) as t →
+∞ and by definition, the critical point (0, 0) is a node and it is asymptotically
stable.
Let λ1 > λ2 > 0, then the general solution of (1) is still of the form (3)
and particular solution be of the forms (4) and (5); and they all exist. The
situation is the same as before, except now all the paths approach and enter
(0, 0) as t → −∞. The critical point (0, 0) is still a node, but in this case it is
an unstable one.
Consider the following Lemma which provide the proof of (iii) part of the
Theorem given in previous module.
Lemma 2. The critical point (0, 0) of the linear system (1) is a saddle point,
if the roots λ1 and λ2 of the characteristic equation (2) are real, unequal, and
of the opposite sign.
Proof. Let λ1 < 0 and λ2 > 0. The general solution of (1) is (same as (3))
x = c 1 u 1 e λ1 t + c 2 u2 e λ2 t
y = c 1 v 1 eλ1 t + c 2 v 2 eλ2 t
and particular solutions are (same as (4) and (5))
x = c1 u1 eλ1 t , y = c1 v1 eλ1 t (c2 = 0)
x = c2 u2 eλ2 t , y = c2 v2 eλ2 t (c1 = 0).

2
 Figure 1: Stable node

Figure 2: Unstable node

3
 Figure 3: Saddle point

Eliminating λ1 from (4), we get the straight line v1 x = u1 y. For any c1 > 0,
the solution (4) represents a path consisting of half of the line and for any c1 < 0,
the solution (4) represents a path consisting of other half of the line. Both of
these half-line paths approach and enter (0, 0) as t → +∞, since λ1 < 0.
In the similar manner, eliminating λ2 from (5), we get the straight line
v2 x = u2 y. For any c2 > 0, the solution (5) represents a path consisting of half
of the line and for any c2 < 0, the solution (5) represents a path consisting of
other half of the line. However, in this case, both of these half-line paths now
approach and enter (0, 0) as t → −∞, since λ2 > 0.

If c1 6= 0 and c2 6= 0, the general solution (3) represents non-rectilinear

paths. Since λ1 < 0 < λ2 , none of these paths can approach (0, 0) as t → +∞
or as t → −∞. Further, none of them pass through (0, 0) for any t0 such that
−∞ < t0 < ∞. As t → +∞, we observe from (3) that each of these non-
rectilinear paths becomes asymptotic to one of the half-line paths defined by
(5). As t → −∞, each of them becomes asymptotic to one of the paths defined
by (4).

Therefore, there are two half-line paths which approach and enter (0, 0)
as t → +∞ and two other half-line paths which approach and enter (0, 0) as
t → −∞. All other paths are non-rectilinear paths which do not approach (0, 0)
as t → +∞ or as as t → −∞, but which become asymptotic to one or another
of the four half-line paths as t → +∞ and t → −∞. By definition, the critical
point (0, 0) is a saddle point and it is unstable.

2 Stability of linear systems

In this section we consider the stability question for solutions of autonomous
differential equations. Let x = φ(t) be a solution of autonomous differential

4
equation
ẋ = f (x). (6)
We are interested in determining weather φ(t) is a stable or unstable solution.
That is whether every solution ψ(t) of (6) which starts sufficiently close to φ(t)
at t = t0 must remain close to φ(t) for all future time t ≥ t0 .

Definition 1. The solution x = φ(t) of (6) is stable if every solution ψ(t) of

(6) which starts sufficiently close to φ(t) at t = 0 must remain close to φ(t) for
all future time t.
In mathematical terms, the solution x = φ(t) of (6) is stable if for every
ǫ > 0 there exists δ = δ(ǫ) such that
kψj (t) − φj (t)k < ǫ if kψj (t0 ) − φj (t0 )k < δ(ǫ), j = 1, . . . , n.
for every solution ψ(t) of (6). The stability issue can be completely resolved for
linear autonomous differential system. The following theorem gives a complete
information.
Theorem 1. (a) Every solution x = φ(t) of ẋ = Ax is stable if all the eigen-
values of A have negative real part.
(b) Every solution x = φ(t) of ẋ = Ax is unstable if at least one eigenvalue of
A has positive real part.
(c) Suppose that all the eigenvalues of A have real part ≤ 0 and λ1 = iβ1 , . . . , λl =
iβl have non zero real part. Let λj = iβj have multiplicity kj i.e. the char-
acteristic polynomial of A can be factored into the form
p(λ) = (λ − iβ1 )k1 . . . (λ − iβl )kl q(λ)
Where all roots of q(λ) have negative real part. Then, every solution x =
φ(t) of (6) is stable if A has kj linearly independent eigenvectors for each
eigenvalue λj = iβj . Otherwise, every solution φ(t) is unstable.
Proof. (a) We know that every solution of ẋ = Ax can be written in the form
ψ(t) = eAt ψ(0). Let φij (t) be the ij element of the matrix eAt , and let ψ10 , . . . , ψn0
be the components of ψ(0). Then, the ith component of ψ(t) is
n
X
ψi (t) = φi1 (t)ψ10 + · · · + φin (t)ψn0 ≡ φij (t)ψj0 . (7)
j=1

Now let all the eigenvalues of A have negative real part and let −α1 be the
largest of the real part of the eigenvalues of A. It is easy to show that for
every number −α, with −α1 < −α < 0, we can find a number K such that
|φij (t)| 6 Ke−αt , t > 0. Consequently,
n
X n
X
|ψi (t)| 6 Ke−αt |ψj0 | = Ke−αt |ψj0 |
j=1 j=1

5
for some positive constants K and α. Now, |ψj0 | 6 kψ(0)k.
Hence,

kψi (t)k = max{|ψ1 (t)|, . . . , |ψn (t)|}

6 nKe−αt kψ(0)k.

Let ǫ > 0 be given. Choose δ(ǫ) = ǫ/nK. Then, kψi (t)k < ǫ if kψ(0)k < δ(ǫ)
and t > 0, since

kψi (t)k 6 nKe−αt kψ(0)k < nKǫ/nK = ǫ. (8)

Thus, the equilibrium solution x(t) ≡ 0 is stable.

(b) Let λ be an eigenvalue of A with positive real part and let v be an eigenvector
of A with eigenvalue λ. Then, ψ(t) = ceλt v is a solution of ẋ = Ax for any
constant c. If λ is real then v is also real and kψ(t)k = |c|eλt kvk.

Clearly, kψ(t)k → ∞ as t → ∞, for any arbitrary c 6= 0. Therefore, x(t) ≡ 0

is unstable. If λ = α + iβ is complex, then v = v 1 + iv 2 is also complex. In this
case
1
+iv 2 )
e(α+iβ)t(v = eαt (cos βt + i sin βt)(v 1 + iv 2 )
= eαt [(v 1 cos βt − v 2 sin βt)
+i(v 1 sin βt + v 2 cos βt)]

is a complex valued solution of ẋ = Ax.

Therefore

ψ 1 (t) = ceαt (v 1 cos βt − v 2 sin βt)

is a real valued solution of ẋ = Ax. Clearly if kψ 1 (t)k is unbounded as t → ∞

if c and either v 1 or v 2 is nonzero. Thus x(t) ≡ 0 is unstable.

(c) If A has kj linearly independent eigenvectors for each eigenvalue λj = iβj

of multiplicity kj , then we can find a constant K such that |(eAt )ij | 6 K.
Therefore kψ(t)k 6 nKkψ(0)k for every solution ψ(t) of ẋ = Ax. Thus from
proof of (a) we have x(t) ≡ 0 is stable. On the other hand, if A has fewer than
kj linearly independent eigenvectors for each eigenvalue λj = iβj , then ẋ = Ax
has solution ψ(t) of the form

ψ(t) = ceiβj t [v + t(A − iβj I)v]

where (A−iβj I)v 6= 0. If βj = 0, then ψ(t) = c(v+tAv) is real valued. Moreover,

kψ(t)k is unbounded as t → ∞ for any arbitrary c 6= 0. Similarly both the real
and imaginary parts of ψ(t)are unbounded in magnitude for arbitrary small
ψ(0) 6= 0, if βj 6= 0, Therefore, the equilibrium solution x(t) ≡ 0 is unstable.
Definition 2. A solution x = φ(t) of (6) is asymptotically stable if it is stable,
and if every solution ψ(t) which starts sufficiently close to φ(t) must approach
φ(t) as t approaches infinity.

6
Example 1. Determine whether each solution x(t) of differential equation
 
−1 0 0
ẋ =  0 −1 1  x
0 −1 −1

is stable, asymptotically stable, or unstable.

Solution The eigenvalues of the coefficient matrix are λ = −1 and λ = −1 ± i.

Since all the three eigenvalues have negative real part. Therefore every solution
of differential equation ẋ = Ax is asymptotically stable.