Dynamic Macroeconomic Theory
Prof. Thomas Lux
Bifurcation Theory
Bifurcation: qualitative change in the nature of the solution occurs if a parameter passes through
a critical point (bifurcation or branch value).
Local Bifurcation Theory: Continuous Time Systems
Consider the system:
y′ = ϕ ( y, α ) , (1)
where α is a parameter. The equilibrium point ye of this system is given by solving
ϕ ( y e ,α ) = 0 . (2)
Note: the equilibrium point ye , a stationary point, depends on the value of α . At certain values
of α the characteristics of the system changes, sometimes quite dramatically.
By the implicit function theorem the equilibria are continuously differentiable functions of α :
ye = y e (α ) . (3)
Some definitions:
(a) If, at an equilibrium point ( y e ,0 ,α 0 ) the Jacobian is zero and several branches of
equilibria come together, one says that ( ye,0 ,α 0 ) is a point of bifurcation.
(b) Alternative definition: let N α denote the number of equilibrium values of the system
when the parameter is equal to α , then, if for any interval, (α 0 − ε ,α 0 + ε ) , N α is not
constant, α 0 is called a bifurcation value, and the system is said to undergo a bifurcation
as α passes through α 0 .
(c) Another definition: a value α 0 of equation (1) for which the solution of (1) is not
structurally stable is a bifurcation value of α .
(d) An equilibrium point at which no bifurcation occurs is called a hyperbolic fixed point.
(e) The bifurcation diagram is a diagram in which the branches of equilibria are shown in
( ye ,α ) space.
1
Note: in the case of codimension one bifurcations1 the conditions on the Jacobian and its roots
can be replaced with single partial derivatives.
1. Saddle-node (fold) bifurcation (prototype function: y ′ = α − y 2 )
Consider the one parameter first-order differential equation
y ′ = f ( y,α ) (4)
and assume that when α = α 0 there is an equilibrium ye, 0 for which the following assumptions
are satisfied:
∂f ( y e , 0 , α 0 )
= 0 , [ f ( y,α ) has a stationary point with respect to y at ( ye,0 ,α 0 ) ] (5.a)
∂y
∂ 2 f ( y e , 0 ,α 0 )
≠ 0 , [ ( y e,0 ,α 0 ) is an extremum] (5.b)
∂y 2
∂f ( ye, 0 ,α 0 )
≠ 0 . [ f ( y,α ) is not stationary with respect to α at ( ye,0 ,α 0 ) ] (5.c)
∂α
Then, depending on the signs of the expression (5.b) and (5.c), there are
(i) no equilibria near ( ye, 0 ,α 0 ) when
α < α 0 ( α > α 0 );
(ii) two equilibria near ( ye, 0 ,α 0 ) for
each parameter value α > α 0
( α < α 0 ). These equilibria are
hyperbolic; one of them is stable
and the other unstable.
Example: y ′ = α − y 2
2. Transcritical bifurcation (prototype function: y ′ = αy − y 2 )
Consider the one parameter first-order differential equation
y ′ = f ( y,α ) (6)
and assume that when α = α 0 there is an equilibrium ye, 0 for which the following assumptions
1
One-dimensional systems.
2
are satisfied:
∂f ( ye, 0 ,α 0 )
= 0 , [ f ( y,α ) has a stationary point with respect to y at ( ye,0 ,α 0 ) ] (7.a)
∂y
∂ 2 f ( y e , 0 ,α 0 )
≠ 0 , [ ( ye,0 ,α 0 ) is an extremum] (7.b)
∂y 2
∂ 2 f ( y e , 0 ,α 0 )
≠ 0 . [Change in α shifts the phase curve] (7.c)
∂α∂y
Then, depending on the signs of the expression (7.b) and (7.c)
(i) the equilibrium y e,0 is stable (unstable) when α < α 0
( α > α 0 );
(ii) the equilibrium y e,0 becomes unstable (stable) for each
parameter value α > α 0 ( α < α 0 ), and a branch of
additional stable (unstable) equilibria ye (α ) emerges.
Example: y ′ = αy − y 2
3. Pitchfork bifurcation (prototype function: y ′ = αy − y 3 )
Consider the one parameter first-order differential equation
y ′ = f ( y,α ) (8)
and assume that when α = α 0 there is an equilibrium y e, 0 for which the following assumptions
are satisfied:
∂f ( ye, 0 ,α 0 )
= 0 , [ f ( y,α ) has a stationary point with respect to y at ( ye,0 ,α 0 ) ] (9.a)
∂y
∂ 3 f ( y e , 0 ,α 0 )
≠ 0 , [Excluding the presence of a horizontal inflection at ye,0 ] (9.b)
∂y 3
∂ 2 f ( y e , 0 ,α 0 )
≠ 0 . [Shift of the phase curve] (9.c)
∂α∂y
Then, depending on the signs of the expression (9.b) and (9.c)
(iii) the equilibrium y e,0 is stable (unstable) when α < α 0 ( α > α 0 );
3
(iv) the equilibrium y e,0 becomes unstable (stable) for each parameter value α > α 0
( α < α 0 ), and two branches of additional stable (unstable) equilibria ye (α ) emerge2.
Example: y ′ = αy − y 3
The Hopf bifurcations in continuous time
Note: the Hopf bifurcation requires at least a 2× 2 system to appear.
Consider the 2× 2 system of first-order difference equations
y1′ = ϕ 1 ( y1 , y 2 ,α ),
(10)
y 2′ = ϕ 2 ( y1 , y 2 ,α ),
and assume that for each α in the relevant range this system has an isolated equilibrium point
ye = ( y1e , y 2e ) obtained by solving the system
ϕ1 ( y1 , y 2 ,α ) = 0,
(11)
ϕ 2 ( y1 , y 2 ,α ) = 0.
The solution to (11) will give y1e , y 2e as continuously differentiable functions of the parameter,
namely (3), if the following Jacobian matrix of (11) is non singular at the equilibrium point
∂ϕ1 ( y1 , y 2 ,α ) ∂ϕ 1 ( y1 , y 2 ,α )
∂y1 ∂y 2
J (α ) = , (12)
2 1 , y 2 ,α )
∂ϕ ( y ∂ϕ 2 ( y1 , y 2 ,α )
∂y1 ∂y 2
2
The case in which the two additional equilibria were stable is called supercritical pitchfork. The case in which the
two additional equilibria were unstable is called subcritical pitchfork.
4
Hopf bifurcation theorem
Assume that Jacobian matrix (12) evaluated at ( ye, 0 (α 0 ),α 0 ) has the following properties:
(i) it possesses a pair of simple complex conjugate eigenvalues θ (α ) ± iω (α ) that
become pure imaginary at the critical value α 0 of the parameter – i.e., θ (α 0 ) = 0 ,
while ω (α 0 ) ≠ 0 ;
dθ (α )
(ii) ≠ 0;
dα α =α 0
THEN system (3) has a family of periodic solutions.
Note: the critical value α 0 is called Hopf bifurcation
point of system (10).
When a stable cycle emerges, we have supercritical
Hopf bifurcation (see fig. i), otherwise it is
subcritical (see fig. ii).
Note:
(1) Conditions for determining the type of Hopf bifurcation (supercritical or subcritical) do
exist, but involve the coefficients of third-order approximation to the nonlinear terms,
which are typically undetermined in economic models (Perko L.: Differential Equation
and Dynamical Systems. Springer-Verlag, Berlin, 1991. Chapter 4).
(2) Conditions (i) and (ii) can also be applied to more general n × n systems of differential
equations. In the case n ≥ 3 , the Hopf bifurcation theorem requires all remaining roots
(except for the pair of complex conjugate roots under investigation) to have a negative
real part as otherwise the system would be unstable anyway.
5
Local Bifurcation Theory: Discrete Time Systems
Note: in discrete time systems the root with unit modulus takes the place of the zero-real-part
root of the Jacobian matrix.
1. Saddle-node (fold) bifurcation (prototype functions: yt +1 = α − yt2 , yt +1 = yt + α − yt2 ).
Consider the one parameter first-order difference equation
yt +1 = f ( yt ,α ) (13)
and assume that when α = α 0 there is an equilibrium ye, 0 . We have fold bifurcation if the
following hypotheses are satisfied:
∂f ( ye ,0 ,α 0 )
=1, (14.a)
∂y
∂ 2 f ( ye , 0 , α 0 )
≠ 0, (14.b)
∂y 2
∂f ( ye, 0 ,α 0 )
≠ 0. (14.c)
∂α
2. Transcritical bifurcation (prototype functions: yt +1 = αyt − yt2 , yt +1 = yt + αyt − yt2 ).
Consider the one parameter first-order difference equation
yt +1 = f ( yt ,α ) (15)
and assume that when α = α 0 there is an equilibrium ye, 0 . We have a transcritical bifurcation if
the following hypotheses are satisfied:
∂f ( ye ,0 ,α 0 )
=1
∂y (16.a)
∂ 2 f ( ye , 0 , α 0 )
≠ 0, (16.b)
∂y 2
∂ 2 f ( ye , 0 , α 0 )
≠ 0. (16.c)
∂α∂y
6
3. Pitchfork bifurcation (prototype functions: yt +1 = αyt − yt3 , yt +1 = yt + αyt − yt3 ).
Consider the one parameter first-order difference equation
yt +1 = f ( yt ,α ) (17)
and assume that when α = α 0 there is an equilibrium ye, 0 . We have pitchfork bifurcation if the
following hypotheses are satisfied:
∂f ( ye ,0 ,α 0 )
=1
∂y (18.a)
∂ 3 f ( ye , 0 , α 0 )
≠ 0, (18.b)
∂y 3
∂ 2 f ( ye , 0 ,α 0 )
≠ 0. (18.c)
∂αy
4. Flip (period doubling) bifurcation (prototype functions: yt +1 = αyt − αyt2 ,
yt +1 = − yt − αyt − αyt3 ).
Note: flip bifurcation can only arise in discrete dynamical systems.
Consider the one parameter first-order difference equation
yt +1 = f ( yt ,α ) (19)
and assume that when α = α 0 there is an equilibrium ye, 0 for which the following hypotheses
are satisfied:
∂f ( ye,0 ,α 0 )
= −1 , (20.a)
∂y
∂f ( ye,0 ,α 0 ) ∂ 2 f ( ye,0 ,α 0 ) ∂ 2 f ( ye , 0 , α 0 )
+2 ≠ 0, (20.b)
∂ α ∂ y 2
∂ α y
∂ 2 f ( y ,α ) 2 ∂ 3 f ( ye ,0 ,α 0 )
a ≡ 3 ≠ 0.
+2
e,0 0
(20.c)
∂y 2 ∂ y 3
Then, depending of the signs of the expressions (20.b) and (20.c),
(i) the equilibrium y e, 0 is stable (unstable) when α < α 0 ( α > α 0 );
(ii) the equilibrium ye,0 becomes unstable (stable) for each parameter value α > α 0
7
( α < α 0 ), and a branch of additional stable (unstable) equilibria ye (α ) of order 2
emerges (two-cycle).
Note: an equilibrium point (or fixed point) of order 2 is an equilibrium point of the
following difference equitation3:
yt +2 = f ( yt +1 ,α ) = f ( f ( yt ,α ),α ) ≡ f ( 2) ( yt ,α ) . (21)
When a in (20.c) is positive (negative), the emerging equilibrium points of order 2 are stable
(unstable), and the flip bifurcation is said to be supercritical (subcritical) respectively.
Hopf bifurcation theorem for discrete-time systems
Note: unlike with the continuous time case the Hopf bifurcation theorem exists only for 2× 2
discrete time systems.
Consider a 2× 2 non-linear difference system with one parameter
y t +1 = ϕ (y t ,α ) , (22)
and suppose that for each α it has a smooth family of equilibrium points y e = y e (α ) at which
the eigenvalues are complex conjugate, λ1, 2 = θ (α ) ± iω (α ) . If there is a critical value α 0 of the
parameter such that
(i) λ1, 2 (α 0 ) = + θ 2 + ω 2 = 1 , λ1j, 2 (α 0 ) ≠ 1 for j = 1,2,3,4 ;
d λ1, 2 (α 0 )
(ii) ≠ 0;
dα
α =α 0
THEN there is an invariant closed curve bifurcating from α 0 .
3
An equilibrium point of order 2 is a point that repeats itself every two periods, i.e. a constant-amplitude alternation.