Lecture slides for UCLA LS 30B, Spring 2020
License: GPL3
Image: ubuntu2004
Learning goals:
Know that linear stability analysis is inconclusive in some cases.
Be able to tell when it is valid to apply linear stability analysis and when the method is inconclusive. (This is an important theorem called the Hartman–Grobman Theorem.)
Recall the linear stability analysis method, in one variable:
Suppose you have a differential equation , with an equilibrium point at .
Step 1: Compute the derivative:
Step 2: Plug in the equilibrium point: .
This number (slope) represents the linear approximation of at the equilibrium point.
Step 3: That number now tells us the stability:
If , then the equilibrium point is stable.
If , then the equilibrium point is unstable.
If , then this method is inconclusive.
Compare this with our recent conclusions, for two or more variables:
Suppose you have a system of differential equations
with an equilibrium point at .
Step 1: Compute the Jacobian:
Step 2: Plug in the equilibrium point:
This matrix represents the linear approximation of the differential equation at the equilibrium point.
Step 3: The eigenvalues of that matrix now tell us the stability:
If all eigenvalues are , the equilibrium point is stable.
If any eigenvalues are , the equilibrium point is unstable.
More specifically, if some eigenvalues are and some are , then it's a saddle point (still unstable).
If the eigenvalues are complex with real part , the equilibrium point is a stable spiral.
If the eigenvalues are complex with real part , the equilibrium point is an unstable spiral.
If ... (?) then this method is inconclusive.
What's wrong with having an eigenvalue that's ?
Example: One eigenvalue is exactly , the other is negative
Example: One eigenvalue is exactly , the other is negative
Example: One eigenvalue is exactly , the other is positive
Example: One eigenvalue is exactly , the other is positive
Example: Eigenvalues are complex, with real part
Example: Eigenvalues are complex, with real part
A generalization of both of the above examples: (purely imaginary eigenvalues)
What's wrong with having an eigenvalue that is pure imaginary?
All of these “edge cases”: (at least in the 2-variable setting)
Remember that the characteristic polynomial of is .
That is, we set ,
where (the trace of the matrix)
and (the determinant) of the matrix.
Then from the quadratic formula, the eigenvalues are
Conclusion: “linear stability analysis” in two or more dimensions
Definition: A matrix is called degenerate (or non-generic, or you can just call it “an edge case”) if any of its eigenvalues is exactly , or is complex with real part equal to , or has any repeated eigenvalues (double roots of the characteristic polynomial).
Theorem: (The Hartman–Grobman Theorem)
As long as the Jacobian matrix at the equilibrium point is not degenerate (so not an edge case), then it can be used (via its eigenvalues) to classify the type of the equilbrium point.
But if the Jacobian matrix has a eigenvalue, or an eigenvalue that is pure imaginary (complex with real part ), or repeated eigenvalues, then the linear stability method is inconclusive.