CDS140a

Nonlinear Systems: Local Theory

02/03/2011

4 Center Manifold Theory

Theorem (Local Center Manifold Theorem) Let f ∈ C r (E), where E is an open subset of Rn containing
the origin and r ≥ 1. Suppose that f (0) = 0 and that Df (0) has c eigenvalues with zero real part, and
s=n−c eigenvalues with negative real part. The system (1) then can be written in diagonal form

ẋ = Cx + F (x, y)
(4)
ẏ = P y + G(x, y)

where (x, y) ∈ Rc × Rs , C c eigenvalues with zero real parts, P is a square matrix

is a square matrix with
with s F (0) = G(0) = 0, DF (0) = DG(0) = 0; furthermore, there
eigenvalues with negative real parts, and
r
exists a δ > 0 and a function h ∈ C (Nδ (0)), h(0) = 0, Dh(0) = 0 the denes the local center manifold
c c s
W (0) := {(x, y) ∈ R × R | y = h(x) for |x| < δ} and satises

Dh(x) [Cx + F (x, h(x))] = P h(x) + G(x, h(x)) (5)

for |x| < δ ; and the ow on the center manifold W c (0) is dened by the system of dierential equations

ẋ = Cx + F (x, h(x)) (6)

for all x ∈ Rc with |x| < δ .

This theorem can be used to determine the ow near nonhyperbolic equilibrium points. The strategy is:

1. Convert (1) in diagonal form (4)

2. Use a series expansion for the components of h(x) (up to the degree of accuracy we need, provided
that r is suciently large)

3. Determine the components of the expansion of h(x) using (5)

4. Substitute this approximate expression of h(x) into (6) to determine the ow.

Example: Perko 2.12.3 Use the above theorem to determine the qualitative behavior of the origin for the
system

ẋ = xy
ẏ = −y − x2

Solution: The system is already in the desired form, with C = 0, P = −1, F (x, y) = xy and G(x, y) =
−x2 . Let

h(x) = ax2 + bx3 + · · ·

Dh(x) = 2ax + 3bx2 + · · ·
2ax + 3bx2 + · · · x ax2 + bx3 + · · ·



Dh(x) [Cx + F (x, h(x))] =
= − ax2 + bx3 + · · · − x2


P h(x) + G(x, h(x))
⇓ Collecting terms (Using (5))

O(x2 ) : −a − 1 = 0
3
O(x ) : b=0
.
.
.

15
So h(x) = −x2 + O(x4 ). The ow on the center manifold is given by (6)

ẋ = F (x, h(x)) = −x3 + O(x5 )

i.e., the system is stable.

Example: Perko 2.12.5(a) Use the above theorem to determine the qualitative behavior of the origin for
the system

ẋ1 = −x2 + x1 y
ẋ2 = x1 + x2 y
ẏ = −y − x21 − x22 + y 2
   
0 −1 x1 y
Solution: The system is already in the desired form, with C = , P = −1, F (x, y) =
1 0 x2 y
and G(x, y) = −x21 − x22 + y 2 . Let

h(x1 , x2 ) = ax21 + bx1 x2 + cx22 + · · ·

[2ax1 + bx2 + · · · , bx1 + 2cx2 · · · ]
Dh(x) =


 −x2 + x1 ax21 + bx1 x2 + cx22 + · · ·


Dh(x) [Cx + F (x, h(x))] = 2ax1 + bx2 + · · · , bx1 + 2cx2 · · ·

x1 + x2 ax21 + bx1 x2 + cx22 + · · ·

2
P h(x) + G(x, h(x)) = − ax21 + bx1 x2 + cx22 + · · · − x21 − x22 + ax21 + bx1 x2 + cx22 + · · ·

⇓ Collecting terms (Using (5))

x21 : b = −a − 1
x22 : −b = −c − 1
x1 x2 : −2a + 2c = −b
.
.
.

3
we get a = −1, b = 0, c = −1, and so h(x1 , x2 ) = −x21 − x22 + O(|x| ). The ow on the center manifold is
given by (6)
 
3 4
ẋ1 = −x2 + x1 −x21 − x22 + O(|x| ) = −x2 − x31 − x1 x22 + O(|x| )
 
3 4
ẋ2 = x1 + x2 −x21 − x22 + O(|x| ) = x1 − x32 − x21 x2 + O(|x| )

Note: The center manifold theorem allows us to determine the local behavior of the system by looking at
the ow on a lower dimensional manifold, i.e., instead of working with an n-dimensional system, we can just
deal with a c-dimensional one (i.e., we reduce the full n-dimensional c-dimensional system).
system into a
In this example, instead of trying to determine the behavior of the full 3-d system, we reduced the problem
to determine the behavior of the 2-d system above.
To determine the stability of the origin, let's try changing to polar coordinates

5
= x1 ẋ1 + x2 ẋ2 = x1 −x2 − x31 − x1 x22 + x2 x1 − x32 − x21 x2 + O(|r| )



rṙ
5
2 5 5
= −x41 − 2x21 x22 − x42 + O(|r| ) = − x21 + x22 + O(|r| ) = −r4 + O(|r| )
4
ṙ = −r3 + O(|r| )
i.e., the system is stable.

Example: Perko 2.12.6 Use the above theorem to determine the qualitative behavior of the origin for the
system

ẋ = ax2 + bxy + cy 2
ẏ = −y + dx2 + exy + f y 2

16
 Solution: The system is already in the desired form, with C = 0, P = −1, F (x, y) = ax2 + bxy + cy 2 and
2 2
G(x, y) = dx + exy + f y . Let

h(x) = k2 x2 + k3 x3 + · · ·
Dh(x) = 2k2 x + 3k3 x2 + · · ·


2 
2k2 x + 3k3 x2 + · · ·
ax2 + bx k2 x2 + k3 x3 + · · · + c k2 x2 + k3 x3 + · · ·


Dh(x) [Cx + F (x, h(x))] =

2
= − k2 x2 + k3 x3 + · · · + dx2 + ex k2 x2 + k3 x3 + · · · + f k2 x2 + k3 x3 + · · ·



P h(x) + G(x, h(x))
⇓ Collecting terms (Using (5))

O(x2 ) : −k2 + d = 0
O(x3 ) : 2k2 a = −k3 + ek2 ⇒ k3 = (e − 2a)k2
.
.
.

So h(x) = dx2 + (e − 2a)dx3 + O(x4 ). The ow on the center manifold is given by (6)

2
= ax2 + bx dx2 + (e − 2a)dx3 + O(x4 ) + c dx2 + (e − 2a)dx3 + O(x4 )


ẋ = F (x, h(x))
= ax2 + bdx3 + bd(e − 2a) + cd2 x4 + O(x5 )

So unstable for a 6= 0. For a=0 and bd > 0 unstable, bd < 0 stable. For a = 0, bd = 0, c 6= 0 it is unstable.

Example: Perko 2.12.4 Use the above theorem to determine the qualitative behavior of the origin for the
system

ẋ = −x3
ẏ = −y + x2

Solution: First, it is clear that the system is stable since the ow on the center manifold is ẋ = −x3 .
Let's calculate the center manifold anyway. The system is already in the desired form, with C = 0, P = −1,
F (x, y) = −x3 and G(x, y) = x2 . Let

h(x) = a2 x2 + a3 x3 + · · ·
Dh(x) = 2a2 x + 3a3 x2 + · · ·
= − 2a2 x + 3a3 x2 + · · · x3


Dh(x) [Cx + F (x, h(x))]
= − a2 x2 + a3 x3 + · · · + x2


P h(x) + G(x, h(x))
⇓ Collecting terms (Using (5))
2
O(x ) : −a2 + 1 = 0
O(x3 ) : a3 = 0
4
O(x ) : −2a2 = −a4 ⇒ a4 = 2

In general, a2k+1 = 0 and an+2 = nan for n even. So the Taylor series x2 + 2x4 + 8x6 + · · · which diverges
for x 6= 0. What happened? Let's try to solve for h(x) without using the power expansion

dh(x)
(−x3 ) = −h(x) + x2
dx
⇓
0 1 1
h = h−
x3 x
The existence and uniqueness theorem does not apply (since the vector eld is not continous in x at 0) .
Indeed using mathematica to solve, it appears to give a continuoum of solutions (all satisfying h(0) = 0,

17
h0 (0) = 0),      
1 1 1 1
h(x) = exp − 2 a + exp − 2 ExpIntegralEi
2x 2 2x 2x2
, where a is some constant and
ˆ ∞
e−t
ExpIntegralEi(z) = Ei(z) = − dt.
−z t

Remarks:

• There may be many functions h(x) that determine diferent center manifolds. However, the ows on
the dierent center manifolds are determined by 6 and are all topologically equivalent near the origin.

• For an analytical system, if the series expansion of h converges, then there exists a unique analytical
center manifold.

• For a analytical system (even polynomial), if the series does not converge, then an analytical center
manifold need not exist.

Theorem f ∈ C 1 (E), where E is an open subset of Rn containing the origin. Suppose that f (0) = 0
Let
and that Df (0) = diag [C, P, Q], where the square matric C has c eigenvalues with zero real part, the square
matric P has s eigenvalues with negative real part, and the square matric Q has u = n − c − s eigenvalues
1
with positive real part. Then there exist C functions h1 (x) , h2 (x) satisfying

Dh1 (x) [Cx + F (x, h1 (x), h2 (x))] = P h1 (x) + G(x, h1 (x), h2 (x))
Dh2 (x) [Cx + F (x, h1 (x), h2 (x))] = Qh2 (x) + H(x, h1 (x), h2 (x))

in a neighborhood of the origin such that the nonlinear system (1) which can be writen as

ẋ = Cx + F (x, y, z)
ẏ = P y + G(x, y, z)
ż = Qz + H(x, y, z)

is topologically conjugate to the C 1 system

ẋ = Cx + F (x, h1 (x), h2 (x))

ẏ = Py
ż = Qz

for (x, y, z) ∈ Rc × Rs × Ru in a neighbrhood of the origin.

References

[1] H. K. Khalil. Nonlinear Systems. Prentice Hall, 3rd edition, 2002.

[2] L. Perko. Dierential Equations and Dynamical Systems. Springer, 3rd edition, 2001.

[3] S. H. Strogatz. Nonlinear Dynamics And Chaos: With Applications To Physics, Biology, Chemistry,
And Engineering. Westview Press, 2001.

18