Differential Equations 2019/20

MA 209

Solutions to homework for week 8

Extra Notes 4: Questions 1 – 5.

Extra Notes 4

1 It is easy to see that the function V (x1 , x2 ) = x21 + x22 is continuous differentiable on R2 and
is a positive definite function (i.e. V (0) = 0 and V (x) > 0 for all x 6= 0). So we only have to
check, for each of the given systems, if the final condition for strong Lyapunov functions is
∂V (x) ∂V (x)
satisfied: is the directional derivative [V (x)]0 = f1 (x) + f2 (x) negative definite?
∂x1 ∂x2
∂V (x) ∂V (x)
Note that for V (x1 , x2 ) = x21 + x22 we have = 2x1 and = 2x2 .
∂x1 ∂x2
 0
x1 = −x31 + x2 sin(x1 ),
(a) For the system we find:
x02 = −x2 − x21 x2 − x1 sin(x1 ),

[V (x)]0 = 2x1 · (−x31 + x2 sin(x1 )) + 2x2 · (−x2 − x21 x2 − x1 sin(x1 ))

= −2x41 − 2x22 − 2x21 x22 .

So for all x we have [V (x)]0 ≤ 0, so [V (x)]0 is definitely negative semi-definite.

And in fact we can only have [V (x)]0 = 0 if x41 = 0, x22 = 0 and x21 x22 = 0, hence if both
x1 = 0 and x2 = 0. It follows that for all x 6= 0 we have [V (x)]0 < 0. So [V (x)]0 is
negative definite. Combined with the remarks at the beginning of this answer, we can
conclude that V (x) is a strong Lyapunov function on the whole of R2 .
 0
x1 = −x1 − 2x22 ,
(b) For the system we obtain:
x02 = 2x1 x2 − x32 ,

[V (x)]0 = 2x1 · (−x1 − 2x22 ) + 2x2 · (2x1 x2 − x32 ) = −2x21 − 2x42 .

So for all x we have [V (x)]0 ≤ 0, so [V (x)]0 is definitely negative semi-definite.

And in fact we can only have [V (x)]0 = 0 if both x21 = 0 and x42 = 0, hence if both x1 = 0
and x2 = 0. It follows that for all x 6= 0 we have [V (x)]0 < 0. So [V (x)]0 is negative
definite. Combined with the remarks at the beginning of this answer, we can conclude
that V (x) is a strong Lyapunov function on the whole of R2 .

c London School of Economics, 2019

 MA 209 Differential Equations Solutions to homework for week 8 — Page 2

x01 = −x1 sin2 (x1 ),


(c) For the system we determine:
x02 = −x2 − x52 ,

[V (x)]0 = 2x1 · (−x1 sin2 (x1 )) + 2x2 · (−x2 − x52 ) = −2x21 sin2 (x1 ) − 2x22 − 2x62 .

So for all x we have [V (x)]0 ≤ 0, so [V (x)]0 is definitely negative semi-definite.

Additionally, we have that if x2 6= 0, then [V (x)]0 < 0. But there exist x1 6= 0 so that
−2x21 sin2 (x1 ) = 0, for instance if x1 = π. Nevertheless, for −π < x1 < π we know that
sin2 (x1 ) = 0 only if x1 = 0. So if we take N = { (x1 , x2 ) ∈ R2 | −π < x1 < π }, then N
is a neighbourhood of 0 so that for all x ∈ N with x 6= 0 we have [V (x)]0 < 0. So [V (x)]0
is negative definite on this neighbourhood N .
It follows that V (x) is a strong Lyapunov function.

x1 = −(1 − x2 )x1 ,
(d) For the system we calculate:
x02 = −(1 − x1 )x2 ,

[V (x)]0 = 2x1 · (−(1 − x2 )x1 ) + 2x2 · (−(1 − x1 )x2 ) = −2x21 (1 − x2 ) − 2x22 (1 − x1 ).

It’s easy to see that provided we take x2 < 1 and x1 < 1, then [V (x)]0 < 0 for all x 6= 0.
So if we take N = { (x1 , x2 ) | x1 < 1, x2 < 1 }, then N is a neighbourhood of 0 on which
[V (x)]0 is negative definite.
It follows that V (x) is a strong Lyapunov function.

2 (a) In part (a) above we found that V (x1 , x2 ) = x21 + x22 is a strong Lyapunov function on
the whole R2 . So for every constant C > 0, the set { x ∈ R2 | V (x) < C } = { (x1 , x2 ) |
x21 + x22 < C } is a domain of stability.
(b) In part (b) above we found that V (x1 , x2 ) = x21 + x22 is a strong Lyapunov function on
the whole R2 . So for every constant C > 0, the set { x ∈ R2 | V (x) < C } = { (x1 , x2 ) |
x21 + x22 < C } is a domain of stability.
(c) In part (c) above we found that V (x1 , x2 ) = x21 + x22 is a strong Lyapunov function on
the set N = { (x1 , x2 ) ∈ R2 | −π < x1 < π }. So in order to get the largest domain of
stability, we must try to find the largest C > 0 so that the set { x ∈ R2 | V (x) < C }
is completely contained within the set N . Expanding what this all means, it appears
that we must find the largest C so that for all (x1 , x2 ) with x21 + x22 < C we also have
−π < x1 < π. Using that x22 ≥ 0 for all x2 , it is obvious that we can choose C = π 2 . So
the largest domain of stability we can guarantee with our current knowledge is the set
D = { (x1 , x2 ) ∈ R2 | x21 + x22 < π 2 }.
(d) In part (d) above we found that V (x1 , x2 ) = x21 + x22 is a strong Lyapunov function on
the set N = { (x1 , x2 ) ∈ R2 | x1 < 1, x2 < 1 }. Arguing as in part (c), we find that
by taking C = 1 the set { x ∈ R2 | V (x) < C } is completely contained in N . But for
larger C this is no longer the case. Hence the largest domain of stability we can guarantee
is D = { (x1 , x2 ) ∈ R2 | x21 + x22 < 1 }.
 MA 209 Differential Equations Solutions to homework for week 8 — Page 3

3 It is easy to see that the function V (x1 , x2 ) = x21 +x22 is continuous differentiable on R2 and is a
positive definite function (i.e. V (0) = 0 and V (x) > 0 for all x 6= 0). So we only have to check,
for each of the given systems, if the final condition for weak Lyapunov functions is satisfied: is
∂V (x) ∂V (x)
the directional derivative [V (x)]0 = f1 (x) + f2 (x) negative semi-definite? Note
∂x1 ∂x2
∂V (x) ∂V (x)
that for V (x1 , x2 ) = x21 + x22 we have = 2x1 and = 2x2 .
∂x1 ∂x2
 0
x1 = x2 ,
(a) For the system we find:
x02 = −x1 − x32 (1 − x21 )2 ,

[V (x)]0 = 2x1 · x2 + 2x2 · (−x1 − x32 (1 − x21 )2 ) = −2x42 (1 − x21 )2 .

So for all x we have [V (x)]0 ≤ 0, so [V (x)]0 is negative semi-definite. And hence V is a

weak Lyapunov function.
The function V is not a strong Lyapunov function, since for any (a, 0) 6= (0, 0) we have
[V (a, 0)]0 = 0. So in that way we cannot conclude that the system is asymptotically
stable.
But we can try to use Extra Theorem 4.1 of Extra Notes 4. The function [V (x)]0 is only
zero if x2 = 0 or if x1 = ±1. If we limit ourselves to the neighbourhood N = { (x1 , x2 ) |
|x1 | < 1 }, then we don’t have to worry about the points with x1 = ±1. Moreover, if
(x1 , x2 ) ∈ N with x2 = 0 and x1 6= 0, then x02 = −x1 6= 0. So a trajectory starting at
a point with x2 = 0, but not the origin, won’t stay on the line x2 = 0. So according
to Extra Theorem 4.1 we can guarantee that the origin is an asymptotically stable fixed
point.
 0
x1 = −x1 + x22 ,
(b) For the system we obtain:
x02 = −x1 x2 − x21 ,

[V (x)]0 = 2x1 · (−x1 + x22 ) + 2x2 · (−x1 x2 − x21 ) = −2x21 − 2x21 x2 = −2x21 (1 + x2 ).

So for all x = (x1 , x2 ) with x2 ≥ −1 we have [V (x)]0 ≤ 0. Hence [V (x)]0 is negative

semi-definite for x2 ≥ −1. So V is a weak Lyapunov function on the neighbourhood
N = { (x1 , x2 ) | x2 ≥ −1 }.
The function V is not a strong Lyapunov function, since for any (0, a) 6= (0, 0) we have
[V (0, a)]0 = 0. So in that way we cannot conclude that the system is asymptotically
stable.
For x2 > −1 we have that [V (x)]0 = 0 only if x1 = 0. If x1 = 0 but x2 6= 0, then
x01 = x22 6= 0. So a the trajectory starting in a point with x1 = 0 not the origin, won’t
stay on the line x1 = 0. And so Extra Theorem 4.1 can be used to guarantee that the
origin is an asymptotically stable fixed point.
 MA 209 Differential Equations Solutions to homework for week 8 — Page 4

x01 = −x31 ,

(c) For the system we determine:
x02 = −x21 x2 ,

[V (x)]0 = 2x1 · (−x31 ) + 2x2 · (−x21 x2 ) = −2x41 − 2x21 x22 .

This expression is clearly non-positive for any x. So [V (x)]0 is negative semi-definite and
hence V is a weak Lyapunov function. But since [V (x)]0 = 0 for any (x1 , x2 ) = (0, a), it
is not a strong Lyapunov function.
We can try to use Extra Theorem 4.1 from Extra Notes 4 again. The function [V (x)]0 is
only zero if x1 = 0. But if x1 = 0, then x01 = 0 and x02 = 0. So all points on the line x1 = 0
are fixed points. And hence trajectories starting on that line won’t approach the origin.
(They won’t go anywhere.) So no matter how small a neighbourhood of the origin we
take (remember, the origin must be an interior point of any such neighbourhood), there
always will be point from the line x1 = 0 in that neighbourhood. So the origin is not an
asymptotically stable fixed point of the system, but only a neutrally stable fixed point.

4 The intuition behind the proof that the origin is unstable for the system x0 = f (x) is as
follows: If x0 = −f (x) has a strong Lyapunov function, then the origin is an asymptotically
stable fixed point of that system. That means that all solutions of x0 = −f (x) move towards
the origin. (Or at least, the solutions starting close to the origin.) But if we look at a solution
of x0 = −f (x), but now go back in time, we see a solution of x0 = f (x). So if this solution
moves towards the origin, then going back in time we move away from the origin.

O.k., now we need to compose a formal proof out of the intuition from above. The main thing
is to do this in small steps, carefully working out what we know and what we need to prove.
Here we go . . .
Suppose the system x0 = −f (x) has a strong Lyapunov function in a neighbourhood N of
the origin. In particularly this means that the origin is an asymptotically stable fixed point
of x0 = −f (x). So there exists an r > 0 so that for any solution x(t) of the system with
x(0) ∈ B(0, r) we have x(t) −→ 0 if t −→ ∞. Since N is a neighbourhood of the origin, we
can take r small enough to make sure that B(0, r) ⊆ N .
Now take x0 ∈ B(0, r) with x0 6= 0, and let y(t) be the solution of x0 = −f (x) with y(0) = x0 .
Note that this means that y(−t) is a solution of x0 = f (x).
Next we use the solution y(t) of x0 = −f (x) to show that the origin is not a stable fixed
point of the system x0 = f (x). By the definition of stability, we must show that there is an
R > 0 so that for all r0 > 0 we have that there is a solution z(t) of the system x0 = f (x) with
z(0) ∈ B(0, r0 ), but so that it is not the case that for all t ≥ 0 we have z(t) ∈ B(0, R).
O.k., take R > 0 small enough so that x0 ∈ / B(0, R) (possible since x0 6= 0). And take any
r > 0. Since y(t) −→ 0 if t −→ ∞, here must be a T > 0 so that y(T ) ∈ B(0, r0 ). Define
0

the function z(t) by z(t) = y(T − t). Then z(t) is a solution of x0 = f (x). We also have
z(0) = y(T ) ∈ B(0, r0 ), but z(T ) = y(0) = x0 ∈
/ B(0, R).
This shows that the system x0 = f (x) is not stable at the origin.
 MA 209 Differential Equations Solutions to homework for week 8 — Page 5

If there is only a weak Lyapunov function for x0 = −f (x), we can’t say much. It is still
possible that the system x0 = −f (x) has an asymptotically stable fixed point at the origin.
And then it follows from the argumentation above that in the system x0 = f (x) is an unstable
fixed point.
But if the origin is not an asymptotically stable fixed point for x0 = −f (x), but only a
neutrally stable fixed point, then the proof above is no longer valid. That doesn’t mean the
conclusion is false, it’s just that we have to look at the situation again.
In fact, we can find examples so that, even though x0 = −f (x) has a weak Lyapunov function,
the system x0 = f (x) is also stable. Take for instance the 2-dimensional system x = f (x)
 0
x1 = x2 ,
given by This is the standard linear system whose solutions are concentric
x02 = −x1 .
circles around the origin. In particular,  0 the origin is a neutrally stable fixed point. The
x1 = −x2 ,
negative version x = −f (x) of it is And this system has a weak Lyapunov
x02 = x1 .
function: V (x1 , x2 ) = x21 + x22 .
So it is possible for x = −f (x) to have a weak Lyapunov function, while the original system
x = f (x) still has a stable fixed point at the origin.

x01 = x2 ,

5 For the system we again first try V (x) = x21 + x22 as a possible
x02 = −x2 + x32 − x51 ,
Lyapunov function. Then for the directional derivative we get:

[V (x)]0 = 2x1 · x2 + 2x2 · (−x2 + x32 − x51 ) = 2x1 x2 − 2x22 + 2x42 − 2x51 x2 .

This doesn’t seem to be negative semi-definite or definite, mainly because of the terms 2x1 x2
and 2x51 x2 . In order to have these cancel one another, we try a second function of the form
V1 (x) = ax61 + bx22 , where we must take a, b positive to guarantee V1 to be positive definite.
We have

[V1 (x)]0 = 6ax51 · x2 + 2bx2 · (−x2 + x32 − x51 )

= 6ax51 x2 − 2bx22 + 2bx42 − 2bx51 x2 = (6a − 2b)x51 x2 − 2bx22 (1 − x22 ).

By making sure that 6a − 2b = 0, for instance by taking a = 1, b = 3, the terms with odd
powers will be eliminated. In that case we have

[V1 (x)]0 = −6x22 (1 − x22 ).

This function is not negative definite, since [V1 (x1 , 0)]0 = 0 for all x1 ∈ R, but it will be
negative semi-definite provided 1 − x22 ≥ 0, hence for |x2 | ≤ 1. So V1 (x) is a weak Lyapunov
function on the set N = { (x1 , x2 ) ∈ R2 | |x2 | ≤ 1 }, which is indeed a neighbourhood of the
origin.
To find a domain of stability, we must choose a C so that V1 (x) < C guarantees x22 ≤ 1.
Since V1 (x) = x61 + 3x22 , this can be achieved by taking C = 3. So the domain given by
D = { (x1 , x2 ) | x61 + 3x22 < 3 } is a domain such that for every neighbourhood within that
domain the origin is a stable fixed point.
MA 209 Differential Equations Solutions to homework for week 8 — Page 6

But in fact we can say more. Note that if x ∈ D, then we have [V1 (x)]0 = 0 only if x2 = 0.
(There are no points with x22 = 1 in D.) If x2 = 0, but x1 6= 0, then x02 = −x51 6= 0. So a
trajectory starting on a point with x2 = 0 not the origin, won’t stay on the line x2 = 0.
All in all, it follows that the origin is an asymptotically stable fixed point, with a domain of
stability D = { (x1 , x2 ) | x61 + 3x22 < 3 }