Lecture Notes for Math 524

Dr. Michael Y. Li December 7, 2004

These notes are based on the lecture notes of Professor James S. Muldowney, the books of Hale, Copple, Coddington and Levinson, and Perko. They are for the use of students in my graduate ODE class.

Chapter 5. Elementary Bifurcation Theory

5.1. Introduction
Consider a family of dierential equations x = f (x, ), (1)

where f : Rn+1 Rn is analytic for R, x Rn . Let x = x0 () be a family of equilibria of (1), namely, f (x0 (), ) = 0. Set z = x x0 (). f (x0 (), ). x Let 1 (), , n () be the eigenvalues of A(). If, for some i, Rei () changes sign at = 0 , we say that 0 is a bifurcation point of (1). Sometimes, we also call (x0 (0 ), 0 ) a bifurcation point. z = A() z + O(|z |2 ), A() = Remarks: (1) f is analytic in x, implies that x0 () is analytic in , provided det A() = det f (x0 (), ) = 0. x Then,

Analyticity may fail at a bifurcation point since det A() = 1 () n (). (2) Being the roots of det( I A()) = 0, = i () are also analytic in except possibly at bifurcation points.

Simplifying Assumption: i () = j () if i = j. We will look at some examples when n = 2. Example 1. One-dimensional bifurcation (only one-dimensional eigenspace changes with ):

Example 2. Two-dimensional bifurcation (two-dimensional eigenspace changes with ):

5.2. One-dimensional bifurcations

Since only one-dimensional eigenspace changes with , we may simply assume n = 1. Therefore, f : R2 R1 , and x0 () is a real-valued analytic function of provided 1 () = fx (x0 (), ) = A() = 0. Therefore, the equilibrium x0 () is u.a.s. if 1 () < 0, and unstable if 1 () > 0. This implies that 0 is a bifurcation point if 1 (0 ) = 0. Therefore, bifurcation points (x0 (0 ), 0 ) are solutions of f (x, ) = 0, and (2) fx (x, ) = 0. The bifurcation diagram describes the general shape of x0 () for near the bifurcation point 0 . At = 0 , 1 () = fx (x, ) = 0. By the Implicit Function Theorem, x = x0 () is the unique solution of f (x, ) = 0, and f dx f + = 0. x d 2

Since 1 (0 ) = fx (x0 (0 ), 0 ) = 0, if f (x0 (0 ), 0 ) = 0, then | dx | , d as 0 .

Therefore the curve x = x0 () has a vertical tangent line when = 0 if f (x0 (0 ), 0 ) = 0. In the subsequent discussions, w.l.o.g., we will assume that the bifurcation point is at (0, 0) , namely, 0 = 0 and x0 (0 ) = 0. The most common bifurcation types are illustrated by the following examples. Example 3. Saddle-Node Bifurcation. Consider x = x2 . In this case, the bifurcation equations (2) becomes x2 = , 2x = 0. There are two branches of equilibria: x0 () = , and the bifurcation point is at (x, ) = (0, 0). Since f (0, 0) = 1 = 0, we should expect a vertical tangent line at (0, 0) for x0 (). The bifurcation diagram is shown in the following gure.

Therefore, the saddle-node bifurcation can be described as a single branch of equilibria undergoes a change in stability, from being stable to unstable. Example 4. Transcritical Bifurcation. Consider x = x x2 . The bifurcation equation (2) becomes x x2 = 0, 2x = 0. This gives two branches of equilibria x0 = 0, and x0 = . For the branch x0 = 0, we have 1 = and thus the stability changes from stable to unstable as increases cross 0, and 0 = 0 is the bifurcation point. For the second branch, x0 = , we have 1 = . Therefore, this branches changes stability in the opposite direction to the rst branch, and the bifurcation point is also 0 = 0. See the bifurcation diagram below.

The transcritical bifurcation can be described as two branches of equilibria intersect and exchange stability type at the bifurcation point. Example 5. Pitchfork Bifurcation. Consider x = x x3 . From the bifurcation equations, we nd that there are three branches of equilibria: x0 = 0, x0 = , and x0 = . The corresponding 1 for the three branches are 1 = , 2, and 2, respectively. It is easy to see that (0, 0 is the bifurcation point for all three branches, and the bifurcation diagram is as shown below. Case 1: Subcritical pitchfork.

Case 2: Supercritical pitchfork.

The subcritical (or supercritical) pitchfork bifurcation can be described as a branch of equilibria which changes stability type at the bifurcation point is intersected there by a stable (or unstable) branch. The following theorem establishes that the three type of bifurcations observed in Examples 3-5 are indeed the only generic ones there are. Theorem 5.1 Suppose (i) f (x, u) is an analytic function of (x, ) near (0, 0). 4

(ii) (x, ) = (0, 0) is a bifurcation point (namely 0 = f (0, 0) = fx (0, 0) ). Then (a) If f (0, 0) = 0 and fxx (0, 0) = 0, then there exists, in a neighborhood of (0, 0), a single branch of critical points which has a saddle node bifurcation at (0, 0). (b) If f (0, 0) = 0, let D = f fx fx fxx
2 , then = f fxx fx (0,0)

(b1) if D > 0, then (0, 0) is an isolated critical point. (b2) if D < 0, then there are two branches of critical points which intersect at (0, 0). The bifurcation is either transcritical or pitchfork. Proof. From the analyticity of f we have 1 f (x, ) = + (a2 + 2bx + cx2 ) + O(x3 , x2 , x2 , 3 ), 2 where = f (0, 0), a = f (0, 0), b = fx (0, 0), c = fxx (0, 0). Then D = ac b2 . Therefore fx (x, ) = b + cx + O(x2 , x, 2 ) f (x, u) = + a + bx + O(x2 , x, 2 ). (4) (3)

In case (a), we have 0 = = f (0, 0). Then the IFT implies f (x, ) = 0 can be solved uniquely in the form = (x) near (0, 0), and (0) = 0. Now (3) implies that (x) = O(x2 ), near (0, 0), and thus c 2 x + O(x3 ). (x) = 2 From (4) we know 1 = fx (x, (x)) = b( c 2 )x + cx + O(x2 ) = cx + O(x2 ). 2

Since c = fxx (0, 0) = 0, 1 (x) changes sign at x = 0, and thus (0, 0) is the bifurcation point. It is easy to see du dx = 0 at the bifurcation point. In summary, = (x) gives a single branch of equilibria which changes stability type at the bifurcation point; we have a saddle-node bifurcation in the case. In case (b), we have = f (0, 0) = 0. Then 1 f (x, ) = (a2 + 2bx + cx2 ) + O(x3 , x2 , x2 , 3 ). 2 (5)

2 2 If D > 0, then 1 2 (a + 2bx + cx ) is a positive denite quadratic form, and (x, ) = (0, 0) is the only critical point in a neighborhood of (0, 0).

Now we assume D < 0. First suppose c = fxx (0, 0) = 0. Then (x, ) = (x, 0) is not a solution of f (x, ) = 0 if x = 0. Set x = z, and g (z, ) = 5 f ( z, ) . 2

Then f (x, ) = 0, = 0 if and only if g (z, ) = 0. From (5) we have 1 g (z, ) = (a + 2b z + c z 2 ) + O(). 2 Therefore, g (z, 0) = 0 has two distinct solutions b D = 1 , 2 z= c and gz (1,2 , 0) = D = 0. The IFT implies that g (z, ) = 0 has two distinct branches of analytic solutions z = z1 (), z = z2 (), zi (0) = i , i = 1, 2. Therefore, f (x, ) = 0 has two distinct branches of solutions x = z1 () and x = z2 () which intersect at the bifurcation point, and 1 , 2 are slopes of these two curves at (0, 0). To see that these branches change stability types, observe 1,2 () = fx (x1,2 (), ) = b + c z1,2 () + O(2 ),
2

from (5)

= b + c 1,2 + O( ) since z1,2 () = 1,2 + a1,2 + O(2 ) = D + O(2 ) Therefore, these two branches have opposite stability types and exchange stability at the bifurcation point; namely, this is a transcritical bifurcation. Next consider the case c = fxx (0, 0) = 0 (together with = f (0, 0) = 0 and D < 0 ). In this case 1 g (z, ) = (a + 2b z ) + O(). 2
a 2 Therefore g (z, 0) = 0 has exactly one solution z = 2 b ( b = 0 since D = ac b < 0 and c = 0 ), a and gz ( 2b , 0) = b = 0. The IFT implies that there exists a unique analytic solution z = z () of a g (z, ) = 0, z (0) = 2 b . Hence, there is a branch of critical points of the form x = x() = z (), a with 2b as its slope at = 0. From (4), we have () = b + O(2 ). So this branch changes stability type at the bifurcation point. To obtain the second branch, let

= x, Then

and h(x, ) =

f (x, x ) . x2

1 h(x, ) = (a 2 + 2b ) + O(x), (6) 2 and h(0, 0) = 0 and h (0, 0) = b = 0. The IFT implies that h(x, ) = 0 has a unique analytic solution = (x) with (0) = 0. Therefore, there is a second branch of critical points = (x) = x (x) = x2 + O(x3 ). For the stability of this branch, observe that, from (4) (with c = 0 ), 1 = bx2 + O(x3 ). Therefore, this branch is stable if b < 0 (supercritical pitchfork) and unstable if b > 0 (subcritical pitchfork). 6

If = 0, then the sign of the rst nonzero term in (x) will determine the sign of 1 . Therefore, in case (b), the bifurcation is always transcritical or pitchfork. Example. Consider x = x( x2 )(x + 2). (7) The critical points are x = 0, x2 = , and x = 2. Bifurcation points are (i) (0, 0), (ii) (2, 0), (iii) (1, 1), and (iv) (4, 2). Also, fx (x, ) = ( x2 )(x 2) 2x2 (x 2) + x( x2 ), and thus (2 ), 2 1 = 2x (2 + x x2 ), x(2 + x x2 ), if x = 0, if x2 = , if x = 2,

Therefore, a supercritical pitchfork at (i), and transcritical at (ii), (iii), and (iv).

5.3. One-dimensional bifurcations ( n = 2 )

Consider a two-dimensional system x = f (x, y, ) y = g (x, y, ), (8)

where f, g are analytic in (x, y, ). Suppose that (x, y, ) = (0, 0, 0) is a bifurcation point. Rewrite system (8) in the following form x y = a() x F (x, y, ) + A0 () + , b() y G(x, y, ) (9)

where F, G = O(r2 ), as r , r2 = x2 + y 2 . Let 1 () and 2 () be the eigenvalues of A0 (). Then, the assumption that (0, 0, 0) is a bifurcation point implies that a(0) = b(0) = 0, 1 (0) = 0, and 2 (0) < 0 (or > 0 ). Therefore, near = 0, ( ) and 2 () are analytic, 2 () < 0 (or > 0 ). W.o.l.g., we assume () 0 A0 () = 1 , 0 2 () and thus f (x, y, ) = a() + 1 ()x + F (x, y, ), g (x, y, ) = b() + 2 ()y + G(x, y, ), 7

with g (0, 0, 0) = 0, gy (0, 0, 0) = 2 (0) = 0. Therefore g (x, y, ) = 0 has a unique branch of solutions y = Y (x, ) analytic in (x, ). Furthermore Y (x, ) = b (0) + O(x2 , x, 2 ). 2 (0)

(Note: Yx (0, 0) = 0. ) Substitute Y (x, ) into f (x, y, ) and let (x, ) = f (x, Y (x, ), ) = a() + 1 ()x + F (x, Y (x, ), ), f we have f x = fx + fy Yx = 1 () + Fx + Fy Yx f = f + fy Y = a () + 1 () x + F + Fy Y .

(0, 0) = f Therefore, f x (0, 0) = 0. This is similar to the case of n = 1. Let f fx D= . fx f xx Now, the equilibria of the system lie on the surface x = X (), y = Y (x, ), where X () is any (x, ) = 0. The determination of the number of branches of x = X () is exactly solution of f (x, ). The stability of an equilibrium is the same as in the Theorem 5.1, replacing f (x, ) by f determined from the eigenvalues 1 (), 2 () of A(), the linearization about the equilibrium. (Note: A() = A0 () if the equilibrium is (0, 0, 0), but not necessarily otherwise.) Since 2 (0) = 2 (0) < 0, we have 2 () < 0 for near 0 : the stability character is therefore determined by 1 (). Observe 1 2 = fx gy fy gx (x = X (), y = Y (x, )),

and that g (x, Y, ) = 0 implies gx + gy Yx = 0, we have 1 2 = gy (fx + fy Yx ) = gy f x. But gy (0, 0, 0) = 2 (0) = 2 (0) < 0, we thus have 1 = f x [1 + O (, x)]. Therefore, near (0, 0) the sign of 1 is the same as that of f x . We thus have the following result. Theorem 5.2 Suppose that a(0) = b(0) = 0, 1 (0) = 0, and 2 (0) < 0. Then (1) If a (0) = 0, f xx (0, 0) = 0, there exists near (0, 0, 0) a single branch of critical points which has a saddle-node bifurcation at (0, 0, 0). (2) If a (0) = 0 and D > 0, the equilibrium at (0, 0, 0) is isolated. (3) If a (0) = 0 and D < 0, there are two branches of critical points which intersect at (0, 0, 0). The bifurcation is either transcritical and pitchfork. 8

Example. Consider a Lotka-Volterra food-chain model x = x( x) (x + 1) y 2 y = y (x 1). The equilibria are (1) x = 0, y = 0 (2) x = , y = 0 (3) x = 1, 2y 2 = 1. We have A() = At an equilibrium in (1), A() = 0 , 0 1 2x y 2 2(x + 1)y fx fy = . gx gy y x1 (10)

and thus it is stable if < 0 and unstable for > 0. At an equilibrium in (2), A() = 0 , 0 1

and thus it is stable for 0 < < 1 and unstable for < 0 and > 1. At an equilibrium in (3), A() = Its eigenvalues are 1 2 [y 1 2 1 (y 2 9)2 80] = [( 3) 4 ( 19)2 320]. y 2 1 4y , y 0 2y 2 = 1.

This branch in the plane x = 1 exists only in 1 and it is stable if 1 < < 3, and unstable for > 3. In summary, we see that branches (1) and (2) undergo a transcritical bifurcation at (0, 0, 0). branches (2) and (3) undergo a supercritical pitchfork bifurcation at (1, 0, 1).

5.4. Hopf Bifurcation (n 2)

We rst examine an example. Example. Consider x = y + x( x2 y 2 ) y = x + y ( x2 y 2 ). 9

(11)

The only equilibrium lies at (0, 0, ), at which A() = 1 . 1

Thus, it is stable if < 0 and unstable if > 0. In polar coordinates, the system can be written as r = r( r2 ) = 1. We see that r2 = gives rise a one-parameter family of stable limit cycles.

As increases through 0, the equilibrium changes its stability character and a periodic limit cycle bifurcates from the equilibrium. This phenomenon is called a Hopf bifurcation. In general, we consider a 2 -dimensional analytic system x = f (x, y, ) y = g (x, y, ). The equilibria x = x0 (), y = y0 () satisfy f (x0 (), y0 ()) = g (x0 (), y0 ()) = 0. with Jacobian matrix A() = fx fy gx gy ,
(x0 (),y0 (),)

(12)

whose eigenvalues are 1 (), 2 () such that 1 () = 2 () = () + i (). We assume that () changes sign at = 0 so that = 0 is a bifurcation value. det(A()) = 2 () + 2 () = 0 near = 0. Change variables from x to x + x0 () and y to y + y0 () , we obtain x y = A() x F (x, y, ) + , y G(x, y, ) (13)

the equilibrium remains at (0, 0, ), and F, G = O(r2 ), as r = A() = () () . () ()

x2 + y 2 0. W.l.o.g., we assume

Therefore, (0, 0, ) is a focus which changes stability character as passes through 0. 10

Theorem 5.3 (Poincar e-Andronov-Hopf Bifurcation Theorem) Suppose (i) The system (13) has an equilibrium at (0, 0) if is near 0. (ii) The eigenvalues 1,2 () = () i () of A() satises (0) = 0, (0) = 0. (iii) (0) > 0. (iv) The equilibrium at (0, 0) is asymptotically stable when = 0. Then, as increases through 0, the stable equilibrium at (0, 0) bifurcates to a unstable equilibrium at (0, 0) surrounded by a stable limit cycle. Furthermore, the period of the limit cycle is 2 + O(), (0) 0.

Remark. The bifurcation stated in Theorem 5.3 is called a supercritical Hopf bifurcation. If we change the assumption (iv) of the theorem to (iv) The equilibrium at (0, 0) is negatively asymptotically stable (as t ). then the conclusion will be: as decreases through 0, the unstable equilibrium at (0, 0) bifurcates into a stable equilibrium at (0, 0) surrounded by an unstable limit cycle, of period |2 (0)| + O (). In this case, the bifurcation is called a subcritial Hopf bifurcation. [See, for example, D. K. Arrowsmith and C. M. Mace, An Introduction to Dynamical Systems, Cambridge University Press, 1990.] Proof. We will rewrite (13) in complex form z = (() i ()) z + N (z, z, ), namely z = z + N (z, z, ), where z = x + i y, z = x i y, () = () i (), and N (z, z, ) = F (x, y, ) + i G(x, y, ). Therefore, x = (z + z )/2, y = (z z )/(2i), and N (z, z, ) = O(|z |2 ). Proposition 5.4 There exists a near-identity analytic transformation S = z + S (z, z, ), with S = O(|z |2 ), as |z | 0, so that the system takes the normal form S = (() i ()) S + ( () + i ())|S|2 S + O(|S|4 ), where (), () are analytic in . 11 (14)

Proof of Proposition 5.4. 1 1 N (z, z, ) = n1 z 2 + n2 z z + n3 z 2 + O(|z |3 ). 2 2 Let w = z + Q(z, z, ), where 1 1 Q(z, z, ) = q1 z 2 + q2 z z + q3 z 2 , 2 2 and q1 , q2 , and q3 are to be determined so that all quadratic terms in the transformed equation are 0. The inverse transformation is z = w Q(w, w, ) + O(|w|3 ). We thus have w = z + (q1 z + q2 z )z + (q2 z + q3 z )z 1 1 = ( i )[w q1 w2 q2 ww q3 w2 + O(|w|3 )] 2 2 1 1 2 2 + n1 w + n2 ww + n3 w + O(|w|3 ) 2 2 + (q1 w + q2 w)( i )w + (q2 w + q3 w)( + i )w + O(|w|3 ). (from (15)) Therefore where n 1 = n1 + ( i )q1 , n 2 = n2 + ( + i )q2 , n 3 = n3 + ( + 3i )q3 . Since (0) = 0 but (0) = 0, near = 0, analytic q1 , q2 , and q3 may be chosen such that n 1 = n 2 = n 3 = 0. Therefore w = ( i )w + M (w, w, ), and M (w, w, ) = O(|w|3 ) near = 0. Next, we will consider another near-identity transformation to get rid of most of the cubic terms. Let 1 1 M (w, w) = m1 w3 + m2 w2 w + m3 ww2 + m4 w3 , 3 3 and set S = w + R(w, w, ), where Then S = w + (r1 w2 + 2r2 ww + r3 w2 )w + (r2 w2 + 2r3 ww + r4 w2 )w , 1 1 R(w, w, ) = r1 w3 + r2 w2 w + r3 ww2 + r4 w3 . 3 3 1 1 w = ( i )w + n 1 w2 + n 2 ww + n 3 w2 + O(|w|3 ), 2 2 (15)

12

and thus

1 1 2 3 S = ( i )[S r2 S 3 r2 S 2 S r3 SS r4 S + O(|S|4 )] 3 3 1 1 2 3 + m1 S 3 + m2 S 2 S + m3 SS + m4 S + O(|S|4 ) 3 3 2 2 + (r1 S + 2r2 SS + r3 S )( i )S + (r2 S 2 + 2r3 SS + r4 S )( + i )S + O(|S|4 ).

2

Therefore where

1 1 2 3 S = ( i )S + m 1S 3 + m 2S 2S + m 3 SS + m 4 S + O(|S|4 ), 3 3 m 1 = m1 + 2( i )r1 , m 3 = m3 + 2( + i )r3 , m 2 = m2 + 2 r2 , m 4 = m4 + 2( + 2i )r4 .

Again, since (0) = 0, (0) = 0 we can choose r1 , r2 , and r3 such that m 1 = m 3 = m 4 = 0, and that S = ( i )S + m2 |S|2 S + O(|S|4 ), (16) with m2 = + i . Remarks. (1) Linearized equation at S = 0, = 0 is z = ((0) i (0))z = i (0)z, whose solution is z (t) = z (0)ei (0)t . (2) m2 is given by the formula m2 = n5 (3) 16 (0) = [(fxxx + fxyy + gxxy + gyyy ) + n1 n2 n1 n2 |n2 |2 |n3 |2 = + i . i 2( + i ) + i 2( + 3i ) 1 (fxy (fxx + fyy ))
(x,y )=(0,0)

gxy [gxx + gyy ] fxx gxx + fyy gyy ]

(4) Write z = rei , and S = Rei . Then S = z + O(|z |2 ) can be written as R = r + O(r2 ) = + O(r). Cubic approximation of system (16) is R = ()R + ()R3 , = () + ()R2 . 13

If (0) = 0, then there exist two branches of critical points for the R equation: R = 0, R2 = () (0) = + O(2 ). () (0)

(0) = 0, (0) = 0 means stability character of 0 equailibrium changes as passes through 0. There is a pitchfork bifurcation at R = 0, = 0. (5) If (0) = 0, we can continue the procedure to obtain the approximation R = ()R + ()R2m+1 , = () + ()R2 . The sign of 1 () = () determines the stability of R = 0. The sign of 1 () = (R + rR3 ) R
R2 =()/ ()

= 2()

determines the stability of the other equilibrium. (6) (t) = (0) ( () + () where w0 = (0). () )t = 0 + (w0 + (1))t, ()

Proof of P-A-H Theorem ( (0) = 0) . From (16), we have R = ()R + ()R3 + O(R4 ) = () + ()R2 + O(R3 ). Let R = , =
2 ,

with

being a small parameter, and w0 = (0). Then d = d

2

H (, , ),

where w0 H (, , ) = (0) + (0)3 + O( ). Now () = 0 +

2 0

H ((s), s, ) ds,

0 = (0),

14

is periodic if and only if (2, 0 , ) = 0 if and only if P (0 , ) := Observe the following

0+ 0 2

H ((s), s, ) ds = 0.

lim P (0 , ) = P0 (0 ) :=

2 [ (0)0 + (0)3 0 ], w0

(0) and P0 (0 ) = 0 if and only if 0 = 0 or 2 0 = (0) , and

2 P (0 , 0) = [2 (0) ] = 0 0 w0
if = 0. Applying the IFT, we obtain that there exists a unique 0 = 0 ( ), 0 (0)2 = (0) (0) such that P (0 ( ), ) = 0, and thus 0 = 0 ( ) gives rise to a family of periodic solutions, whose period is obtained from the equation for .

Example. Consider x = x + y y = x + y x2 y The complex form is 1 z = ( i)z (z 3 + z 2 z zz 2 z 3 ). 8 Since the quadratic terms are already zero, there is a transformation S = z + R(z, z, ) which changes the equation to normal form 1 S = ( i)S |S|2 S + O(|S|4 ), 8 or in polar form 1 R = R R3 + O(R4 ), 8 = 1 + O(R2 ). R2 = 8 + O(2 ).

Therefore, there exists a unique family of periodic solutions (through Hopf bifurcation)

Remark. A somewhat technical further development of the proof of the P-A-H Theorem leads to the result that (when (0) > 0 ) R () > 0 implies the periodic orbit is stable and R () < 0 15

implies it is unstable [see J. K. Hale and H. Ko cak, Dynamics and Bifurcations, Springer-Verlag, 1991, page 353]. The periodic solution branch may thus undego further bifurcation. Example. Consider x = y + x( + cr2 r4 ) y = x + y ( + cr2 r4 ), where r2 = x2 + y 2 . In polar coordinates, we have r = r( + cr2 r4 ) = 1.
c When c > 0, the branch = cr2 + r4 undergoes a saddle-node bifurcation when = 3 16 .
2

16