7 16 Notes
7 16 Notes
Spirals / Centers
Supposed we have a 2D-system represented by the matrix A such that A has complex
eigenvalues. Then the eigenvalues are conjugates:
λ = a + bi, λ̄ = a − bi
Eigenvectors for each eigenvalue also come in conjugate pairs v, v̄. Solutions to the system in
C are then of the form:
λt λ̄t
c1 e v + c2 e v̄
We consider the real parts of such solutions to get solutions in R.
Euler’s identity states that:
xi
e = cos(x) + i sin(x)
Therefore:
(a+bi)t at bit at
e =e e = e (cos(bt) + i sin(bt))
The graph of this equation in C, which can be visualized as a plane, is a spiral whose radius
at bit
is controlled by e and whose angle and speed of rotation is controlled by e .
2 at
The same thing is true for the phase portrait and trajectories in R . e describes how
trajectories move away or toward (0, 0) and the complex part as well as the eigenvectors
describe how the spiral is shaped.
Therefore we have three cases:
at
1. Unstable Spiral: Re(λ) = a > 0, then e → ∞ as t → ∞ implying that solutions move
away from the origin.
at
2. Center: Re(λ) = a = 0, then e = 1 and trajectories are closed circles or ellipses around
the origin.
at
3. Stable Spiral: Re(λ) < 0, then e → 0 as t → ∞, implying that solutions end up at
the origin.
Example 1. Draw the phase portrait and describe stability for the following system:
ẋ = x − 3y, ẏ = 3x + y
x(0) = 2, y(0) = 3
1 −3
A=[ ]
3 1
1
This has characteristic polynomial:
2
χ(t) = t − 2t + 10
3
2
1
0
−1
−2
−3
−3 −2 −1 0 1 2 3
We now solve the system for the given initial conditions. We can restrict to just one eigenvalue.
Let λ = 1 + 3i. Then:
−3i −3 i
ker A − λI = ker [ ] = span {[ ]}
3 −3i 1
Since:
2
−3i(i) = −3i = −3(−1) = 3
Then for the initial condition [i, 1], the system restricts to the equation ẋ = λx which has
solution:
λt i
e [ ]
1
To find real solutions, we use Euler’s identity and take the real and imaginary parts separately:
λt i i
e [ ] = e e = e (cos(3t) + i sin(3t)) [ ]
t 3it t
1 1
− sin(3t) cos(3t)
= e ([ ] + i[ ])
t
cos(3t) sin(3t)
t − sin(3t) cos(3t)
e [ ], e [ ]
t
cos(3t) sin(3t)
2
Because the other eigenvalue / eigenvector pair is just a conjugate of this one, the real and
imaginary parts we get from them are up to a scalar the same as these. Therefore the real
solutions to the system are spanned by the ones above.
Therefore the general solution is:
t − sin(3t) t cos(3t)
c1 e [ ] + c2 e [ ]
cos(3t) sin(3t)
For the initial condition above, we have:
2 x(0) − sin(0) cos(0)
[ ]=[ ] = c1 [ ] + c2 [ ]
3 y(0) cos(0) sin(0)
Since cos(0) = 1, sin(0) = 0 we get that c1 = 3 and c2 = 2. Therefore the solution is:
x(t) t − sin(3t) t cos(3t)
[ ] = 3e [ ] + 2e [ ]
y(t) cos(3t) sin(3t)
Example 2. Draw the phase portrait for the system:
0 1
A=[ ]
−4 0
2
This has characteristic polynomial χ(t) = t + 4, so λ = ±2i. Since the eigenvalues have 0
real part, the origin is a Center.
Eigenvectors for 2i:
−2i 1 −i
ker A − 2i = ker [ ] = span {[ ]}
−4 −2i 2
The general solution is then:
sin(2t) − cos(2t)
c1 [ ] + c2 [ ]
2 cos(2t) 2 sin(2t)
Notice that, because of the components of the eigenvector, the scaling in the first component
is different from the scaling in the second component. In particular, for any solution curve,
its intersection along the y-axis will be 2× the distance from the origin as its intersection
along the x-axis. Therefore trajectories are ellipses stretched along the y-axis:
3
2
1
0
−1
−2
−3
−3 −2 −1 0 1 2 3
3
Degenerate Nodes
If the 2D system given by A has only one eigenvalue λ, then it is either a star node or a
degenerate node.
It is a star node if the eigenspace for λ is 2-dimensional. This means every vector is an
eigenvector and all trajectories travel along lines towards or away from the origin.
It is a degenerate node if the eigenspace is only 1-dimensional. In this case the origin is a
stable or unstable degenerate node depending on the sign of the eigenvalue. To find solutions,
instead of diagonalizing, you put the system into Jordan normal form.
Example 3. Draw the phase portrait, classify the origin, and describe solutions for the
system:
ẋ = x + y, ẏ = −x + 3y
3
2
1
0
−1
−2
−3
−3 −2 −1 0 1 2 3
You can see from the picture that solutions spiral out of the origin but along the eigenvector
[1, 1] trajectories follow the same line.
To describe the general solution to the system, we calculate the generalized eigenspace.
In 2 dimensions, this is done by finding a vector [x, y] such that:
x 1
(A − 2I) [ ] = [ ]
y 1
4
This gives us the equation:
−x + y = 1
and we can choose the vector [−1, 0]. Then the general solution has the form:
2t 1 1 −1
c1 e [ ] + c2 e (t ⋅ [ ] + [ ])
2t
1 1 0
2t
Notice the extra term te v in the solution corresponding to the generalized eigenvector. This
is because restricting the system along the generalized eigenvector [−1, 0], the system is of
the form:
ẋ = 2x + c, c ∈ R
Nonlinear spirals
Example 4 (Strogatz 6.3.13). Consider the system:
3
ẋ = −y − x , ẏ = x
3
The x-nullcline is in blue and is given by y = −x . The y-nullcline is in red and is given by
x = 0, the y-axis:
y
Then the only fixed point is (0, 0). The Jacobian for the system is:
2
−3x −1
[ ]
1 0
Therefore:
3 4
ṙ = −r cos (θ)
5
This vanishes when cos(θ) = 0. For r to not change along a trajectory, this implies that
cos(θ) = 0 along the entire trajectory, so that θ is constant. Then θ̇ = 0 along the trajectory.
But the origin is the only fixed point, implying that r changes along all other trajectories.
We can also explicitly check:
xẏ − y ẋ
θ̇ =
r2
2 2 3
x + y + yx
=
r2
3
= 1 + rr cos (θ) sin(θ)
−1
−1 0 1
Example 5. Consider the Lotka-Volterra equations where we do not assume the prey
population grows exponentially:
6
At the origin we linearize to get:
a 0
[ ]
0 −k
Then the origin is a saddle point , with the x-axis as the unstable manifold and y as the
stable manifold.
At the second fixed point we get:
−bk/λ −ck/λ
[ λa−bk ]
c
0
−a −ca/b
[ ]
0 (−k + λa/b)
7
x-axis. The eigenvector for 1 is (−2, 3). The phase portrait is:
0
0 1 2