Stability of Linear Differential Equation Systems

This document discusses the stability behavior of 2x2 linear differential equation systems. It explains that the stability at the origin can be determined by plotting the trace and determinant on the (τ,Δ) plane. Different regions of this plane correspond to whether the system exhibits a sink, source, center, spiral, or saddle point. The eigenvalues also provide insight into the stability and can be plotted on the complex plane. Key thresholds like the parabola Δ=τ^2/4 divide the stable and unstable regions.

Uploaded by

Dan Drake

Available Formats

Download as PDF, TXT or read online on Scribd

100% found this document useful (3 votes)

11K views

Stability of Linear Differential Equation Systems

Uploaded by

Dan Drake

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 3

Stability behavior of linear differential equation systems

This is a note about first-order 2 × 2 linear differential equation systems:

d
x = Ax,
dt
where A is a 2 × 2 real-valued matrix. I’m assuming you know a bit about these, from, say Zill
and Cullen or another basic differential equations book.
Let τ be the trace of A and ∆ its determinant; the eigenvalues of A can be written in terms of
τ and ∆:
1 p
λ= τ ± τ 2 − 4∆ .
2
These linear systems all have a single critical point at the origin, and the eigenvalues determine the
behavior of the system there. Since τ and ∆ determine the eigenvalues, we can plot the behavior
of these systems on a (τ, ∆) plane.

τ2
∆ ∆= 4

stable spiral unstable spiral

deg. sink deg. source

center

sink source

saddle point

Green background means the origin is stable ; areas with a red or yellow background are
unstable . The orange axes and plots are a representative picture of what solutions near the
origin will look like for such a trace and determinant; the actual appearance will depend on the
eigenvectors. The blue axes represent the complex plane and we draw dots where the eigenvalues

1
will be. Note that, for repeated roots, there is one big dot, which is supposed to stand for two
roots stacked on top of one another.1 For example, for the stable spiral, we draw

In orange, we have plotted actual solutions x(t) to such a system, and in blue we’ve drawn
roots with a negative real part and nonzero complex part, because such a system will have those
kind of roots.
Each pair of plots is positioned in the (τ, ∆) plane at the origin of the orange axes—so, for
example, the orange origin of the degenerate sink plot lies precisely on the parabola ∆ = τ 2 /4,
and the orange origin of the center plot is exactly on the positive ∆ axis.
For saddle points, we only draw one orange plot, because they look roughly the same regardless
of the trace. We do, however, draw three different blue plots, corresponding to negative, zero, and
positive trace.
Notice that there are three different kinds of two-dimensional plots in the figure! Given a matrix
A, we we can plot solutions to the DE system in an (x, y) plane (orange plots), the eigenvalues can
be placed in the complex plane (blue plots), and we classify stability behavior in the (τ, ∆) plane.
Quite a large number of concepts all come together in the main figure.

Stability on the axes and the parabola

The τ and ∆ axes, along with the parabola ∆ = τ 2 /4, are colored black in the main figure, but we
can figure out the stability behavior for those places too. Here’s a picture of those axes and the
parabola, made very fat since that’s what we are interested in:

2
1 3

4 5 6

In region 1—that is, the part of the parabola ∆ = τ 2 /4 with τ < 0—our system has a single
negative eigenvalue. We may or may not have two linearly independent eigenvectors, but in any
case, our solutions will be linear combinations of terms of the form exp(λt)v and t exp(λt)w with λ
negative. As t goes to infinity, such terms will always go to 0, so everything in region 1 is stable .
By a similar argument, everything in region 3 is unstable .
In region 2, the systems have purely imaginary eigenvalues and all the solutions are periodic,
and hence region 2 is stable . On the other side of the ∆ axis in region 7, we always have one
positive and one negative eigenvalue, so all such systems are in region 7 are unstable at the origin.
1 The
√
radius of those larger dots is 2 times that of the smaller dots, so the area of the big dots is exactly the
same as the sum of the areas of two small dots!

2
In region 4, the trace is negative and the determinant is zero, so one eigenvalue is zero and
the other is negative. Any such system will have two linearly independent eigenvectors and all
solutions are described by
C1 exp(λt)v1 + C2 v2 ,
which for any choice of C1 and C2 goes to C2 v2 as t goes to infinity; region 4 is stable . Everything
works the same in region 6 but (most of) the solutions go off to infinity. (If C1 is zero, those solutions
will be constant and not go to infinity.) All such systems in region 6 are unstable .
We are left with region 5: the origin. If the trace and determinant are zero, then the only
eigenvalue is zero. The matrix A could be the zero matrix, in which case we clearly have a stable
critical point at the origin. On the other hand, we might have a defective matrix: say that v is the
eigenvector for such a matrix and w the generalized eigenvector; then all solutions to the system
are given by C1 v + C2 (v + tw), which is clearly unstable because of the tw term. Therefore,
systems in region 5 could be stable or unstable. Let’s color it half green and half red.
Putting all these things together and coloring the regions according to their stability or insta-
bility, we can re-draw our diagram:

2
1 3

4 5 6

This means you can classify the stability behavior of any 2 × 2 linear DE system simply by
finding its trace and determinant, and as long as one of those is nonzero, you know how the system
will behave—and if both are zero, the system will be unstable unless it’s the zero matrix.

This document may be distributed and modified under the terms of the Creative Commons
Attribution-Share Alike 3.0 license. See http://creativecommons.org/licenses/by-sa/3.0/.
Dan Drake (http://mathsci.kaist.ac.kr/~drake)
16 September 2010