Chapter 13

Phase Plane Analysis

We have seen that the linear approximation of a system gives

information about the stability of an equilibrium. Insight into other
properties can also be expected. We will discuss this matter in detail
for the case of two state variables only. An advantage of this special
case is that the results can easily be presented graphically. A diagram
where one state variable is plotted as a function of the other is called
a phase plane. The curves in the phase plane are called trajectories or
orbits. We will begin with a discussion of linear systems.

13.1 Phase Planes for Linear Systems

Consider the system
x = Ax (13.1)
We know that the eigenvalues of A play an important role for the
dynamic properties of (13.1). We will see that also the eigenvectors are
important.
Lemma 13.1 Consider the system (13.1) and let A be an eigenvalue
of A with corresponding eigenvector v. Then
x(t) = aextv (13.2)
(where a is an arbitrary scalar) is a solution of (13.1).

Proof: Substitution of (13.2) into (13.1) gives

a\extv

339
340 Chapter 13 Phase Plane Analysis

for the left hand side and

aextAv = a\extv

for the right hand side. D

The lemma is illustrated in Figure 13.1 for the case of a real eigenvalue
and a two-dimensional state space. It follows that rays from the

Figure 13.1: Eigenvector and solutions of a linear differential equation.

origin which are tangent to the eigenvector are also trajectories. If the
eigenvalue is negative, the solution tends towards the origin otherwise
away from it.
It is possible to classify singular points depending on whether the
eigenvalues are complex or real and whether the real parts are positive
or negative. We will concentrate on the case where A is a two by two
matrix.

Two Real, Different, Negative Eigenvalues

Assume that the system (13.1) has two real eigenvalues, AI and A2,
satisfying

AI < A2 < 0

Let v\ and v2 be corresponding eigenvectors. According to Lemma 13.1

there are then trajectories approaching the origin along straight lines
tangent to v\ and v2. Since solutions along v\ approach the origin
faster than solutions along v2, it is customary to call v\ and v2 the fast
13.1 Phase Planes for Linear Systems 341

and slow eigenvector, respectively. Solutions which do not start on an

eigenvector will be linear combinations
x(t) = cieXltvi + c2eX2tv2 (13.3)
where c\ and c2 are constants. When t goes to infinity the second
term of (13.3) will dominate, so trajectories will approach the origin
along the slow eigenvector. For t tending to minus infinity the first
term will dominate, so initially trajectories will be parallel to the
fast eigenvectors. The phase plane will look like Figure 13.2. An
equilibrium of this type is called a stable node.

Figure 13.2: Stable node.

Two Real Distinct Positive Eigenvalues

When both eigenvalues are real, distinct and positive the trajectories
look the same as in the previous case. The difference is that the
exponential functions in the right hand side of (13.3) grow with time.
This gives the same figure as in 13.2 but with the trajectories traveling
in the opposite direction. Such an equilibrium is called an unstable
node, see Figure 13.3.

Two Real Eigenvalues of Opposite Sign

Let the two eigenvalues have opposite signs, AI < 0 < A2 and call the
corresponding eigenvectors v\ and v2. Since AI is negative, trajectories
342 Chapter 13 Phase Plane Analysis

Figure 13.3: Unstable node.

along ^i will approach the origin and it is natural to call v\ the "stable
eigenvector". Analogously, v^ is called the "unstable eigenvector" since
trajectories are directed away from the origin. A general trajectory
is a linear combination as in (13.3), where one component decreases
towards zero and the other one increases towards infinity. Trajectories
will therefore tend to the stable eigenvector when moving backwards
in time and towards the unstable one when moving forwards in time.
The phase plane will then have the general appearance of Figure 13.4.
This type of equilibrium is called a saddle point.

Multiple Eigenvalues

In the situation where the two eigenvalues of a second order system

coincide, AI = A2 = A, there are two possibilities, depending on the
number of linearly independent eigenvectors. When there is just one
independent eigenvector we have again a node, but with the slow and
fast eigenvector coinciding. The node can be either stable or unstable
depending on the sign of A. Figure 13.5 shows such a node. If there are
two linearly independent eigenvectors, then all linear combinations of
them are also eigenvectors. This means that all straight lines through
the origin correspond to trajectories, see Figure 13.6. This type of
equilibrium is called a star node.
13.1 Phase Planes for Linear Systems 343

Figure 13.4: Saddle point.

Figure 13.5: Stable node.

344 Chapter 13 Phase Plane Analysis

Figure 13.6: Stable star node.

Complex Eigenvalues
An example of a system with complex eigenvalues is

* = [ ' -"I* (13.4)

[w a \

which has the eigenvalues a ± iu. With polar coordinates

xi = rcos<^, X2 = rsin<^

the system can be written

r = or (13.5)
j> = u (13.6)

The trajectories become spirals since <£ is monotone. If a < 0 then

the distance to the origin decreases all the time and the phase plane of
Figure 13.7 is obtained. An equilibrium of this type is called a stable
focus. If a > 0, r increases and the trajectories spiral outwards from
the origin. This is known as an unstable focus. Finally, if a = 0 then
the distance to the origin remains constant and the trajectories are
circles as shown in Figure 13.8. The phase plane is called a center.
Now consider an arbitrary second order system
13.1 Phase Planes for Linear Systems 345

Figure 13.7: Stable focus.

Figure 13.8: Center.

346 Chapter 13 Phase Plane Analysis

x — Ax

By making the transformation x = Tx, where T is a nonsingular

matrix, one obtains the system

5 = T~lATx (13.7)
If A has the complex eigenvalues a ± zu; with complex eigenvectors
v ± ztu, then the choice

T=[v -w]
gives, after some calculations

Any second order system with complex eigenvalues can thus be

transformed into the form (13.4). Complex eigenvalues with negative
real parts give a stable focus, while positive real parts give an unstable
focus, and purely imaginary eigenvalues give rise to a center. The
phase planes will usually be deformed versions of Figures 13.7 and
13.8 due to the influence of the transformation T. It is often possible
to sketch approximately how the figure is deformed by checking the
angles at which the trajectories cross the coordinate axes. One simply
puts x\ — 0 or X2 = 0 into the system equations and forms the ratio
X2/xi. See also Section 13.2 below.

13.2 Phase Planes for Nonlinear Systems

The Phase Plane Close to an Equilibrium
Let us now return to the nonlinear system. We assume that the right
hand side is continuously differentiable. Disregarding the input and
assuming the equilibrium to be at the origin for simplicity, we can
write
x = Ax + <?(*), \g(x)\/\x\ -> 0, \x\ -» 0 (13.8)
according to Theorem 11.1. It is natural to apply the linear analysis
of the preceding section to the matrix A. We can expect the nonlinear
system to have almost the same phase plane as the linear part close to
the equilibrium, since g is very small compared to Ax there.
13.2 Phase Planes for Nonlinear Systems 347

Let us consider the case of a system with two states having a

linearization with complex eigenvalues a ± iu where u ^ 0. The same
variable transformations as in Section 13.1, will replace equations (13.5)
and (13.6) by

r = <rr + pi(r,<t>), |pi(r,$|/r -)• 0, r -» 0

^ = w + p2(r,^), |p2(r,^)|-»0, r-K)

where p\ and p^ are functions coming from the nonlinear part g of

(11.12). If a is nonzero then the first equation will be dominated by
the linear part close to the equilibrium. The second equation shows that
<f> will be monotonically increasing or decreasing if r is small enough so
that \p2\ < |w|. If the linear part is a focus, it follows that the nonlinear
system has the same qualitative behavior close to the equilibrium. If
the linear part is a center, the situation is different. The r-equation is
then

r = p\(r,<t>)

Depending on the nonlinearity r, can either increase, decrease or remain

constant. The nonlinear behavior can thus be similar to a stable or
unstable focus or a center (or possibly something entirely different).
This fact was illustrated in Examples 12.1 and 12.2.
A similar analysis can be made for other types of equilibria and
leads to the following conclusions.

• If the linearized system has a node, focus or saddle point, then

the nonlinear system has a phase plane with the same qualitative
properties close to the equilibrium. If the linearized system has
a star node, then the qualitative behavior remains the same for
the nonlinear system if in (13.8) we have \g(x)\ < C\x\l+s close
to the origin, for some positive constants C and 6.

t If the linearized system has a center, then the nonlinear system

can behave qualitatively like a center or focus (stable or unstable)
close to the origin.
348 Chapter 13 Phase Plane Analysis

The Phase Plane Far from Equilibria

It is possible to form an idea of phase plane properties of a second order
system

*' = >f"**> (,3.9)

0:2 = h\xl,x<2)

by forming the derivative

dX2 _ X2
dx\ x\
which gives

dx2 _ /2(#1,0:2)
dx\ fi (0:1,0:2)

This is a differential equation where the time is eliminated. The right

hand side of (13.10) gives the slope of the trajectories as each point.
In particular the slope is horizontal at points where

72(0:1,0:2) = 0

and vertical where

/i (#1,0:2) = 0

By studying the slopes at interesting points, e.g., the coordinate axes,

it is often possible to get a good qualitative feeling for the system
behavior. Prom the limits

,. /20El,0:2) ,. /20El,0:2)
lim -— -. lim -—; -
Xi-»±00 /l(o:i,0:2) X2->±00 /l(o:i,0:2)

it is possible to get some idea of system properties for large amplitudes.

In some special cases it is also possible to solve (13.10) explicitly.

Some Examples
We finish our discussion of phase planes by showing the complete phase
planes of our two examples.
13.2 Phase Planes for Nonlinear Systems 349

Example 13.1: Phase Plane of the Generator

Consider the generator of Example 11.2, with a = 1, b = 2, and u = 0.
The linearizations around the equilibria were computed in Example 11.5. For
x2 = 0, xi = i>27r, we get

l
A-\°
A
-[-2 -ij]

with the characteristic equation

A2 + A + 2 = 0

The eigenvalues are complex with negative real parts, so these equilibria are
stable foci. For x2 = 0, x\ = I/TT, we get instead

AA-P
-[2 -ijM
with the eigenvalues 1 and —2. This is a saddle point with the unstable
eigenvector [1 l]T (X = 1) and the stable eigenvector [1 - 2]T (A = -2).
The phase plane will thus be periodic in the x\ direction with a focus at
every even multiple of TT and a saddle point at every odd multiple. To get
more information on the phase plane we could check the slope of trajectories.
Equation (13.10) gives
dx2 _ 1 2sinxi
dxi ~ x2
We see that trajectories have vertical slope when they pass the x\ axis and that
the slope tends to -1 as \x%\ —» oo. The phase plane is shown in Figure 13.9.

Example 13.2: Phase Plane of the System with Friction

Let us consider the mechanical system with friction given by (11.6). We use
the parameter values K = 1, u0 = 1> ^b = 1-5, and F\ = 1. As we noted
when discussing the equations (11.10), the system is piecewise linear. With
the parameter values above we have
Xi = Z2
- - ifx2>l (13.11a)
X2 = -Xi-/X 2 -l

il=I2
, ^ i f ^< x
X2 = —Xi — /X2 + 1
(13.11b)
350 Chapter 13 Phase Plane Analysis

Figure 13.9: Phase plane for the generator. The directions of the stable and
unstable eigenvectors are shown with dotted lines at x*z = 0, xi = ±TT.

xi = 1
if X2 = 1 and |xi + f x ^ \ < 1.5 (13.11c)
X2 = 0

(The set described by x% = 1, xi + fx<2 > 1.5 is not important. In this set
we have ±2 < 0, and trajectories starting there immediately enter (13.11b).
Analogously the set x<z = 1, xi + fx<2 < 1.5 is unimportant.)
For (13.1 la) there is an equilibrium at x\ = — 1, x<z = 0, which is outside
the region where the equations are valid. For (13.lib) there is an equilibrium
at xi = 1, X2 = 0. The equations (13.11c) have no equilibrium. Since xi is
increasing, solutions starting there will eventually leave the region and enter
(13.11b). Both for (13.11a) and (13.11b) the linear dynamics is described by
the matrix
r o 11
l-i -/]
with the characteristic equation

A2 + /A + 1 = 0

For / = 0 the eigenvalues are imaginary and the phase plane has a center.
For 0 < / < 4 the eigenvalues are complex with negative real parts and the
dynamics is described by a stable focus. In particular, we consider the case
of / being a small positive number. The trajectories are then almost circular
with a slow spiraling motion towards the equilibrium. For 0 < / < 4 we get
the following piecewise linear system:
13.2 Phase Planes for Nonlinear Systems 351

A
2 o

-0.5

-1

(a) Phase plane of a mechanical (b) Phase plane of a mechanical

system with friction. The damping system with friction. The damping
is / = 0.1. is / = 0.02.

Figure 13.10: Phase plane of a mechanical system with friction for different
values of the damping.

1. x2 > 1
The upper part of a stable focus with equilibrium (xi = — 1, x% =
0) outside the region. For small values of /, the trajectories are
approximately circles centered at x\ = —1, X2 = 0.
2. x2 < 1
A stable focus around an equilibrium at x\ = 1, x2 = 0.
3. ar2 = l, |zi+/| < 1.5.
The region consists of one single trajectory between xi = —1.5 — / and
xi = 1.5 - /. At the latter point it joins a trajectory of region 2.

For / = 0.1 we get the phase plane of Figure 13.10a. Region 3 is the solid
line between points A and B in the figure.
If the damping term is reduced to / = 0.02 we get the interesting phenomenon
of Figure 13.10b. We see that trajectories that leave region 3 at B in the phase
plane return to region 3 at C. There is thus a closed curve in the phase plane
corresponding to a periodic solution. It is clear that all trajectories arriving
at the segment AB will end in this periodic motion. A solution which is a
limit of trajectories in this way is called a limit cycle. From the geometry of
the phase plane we see that the limit cycle is only formed if B is to the right
of xi = 1. This situation occurs when the static friction is greater than the
sliding friction.
352 Chapter 13 Phase Plane Analysis

13.3 Comments
Main Points of the Chapter
• It is possible to get a good idea of the properties of a nonlinear
system close to an equilibrium by studying the eigenvalues and
eigenvectors of the linear approximation. This is especially true
in two dimensions where the result can be plotted (the phase
plane).

• Two real distinct eigenvalues of the same sign give a tangent

node, stable for negative real values, unstable for positive ones.

• Complex eigenvalues with nonzero real parts give a focus, stable

for negative real parts, unstable for positive ones.

• Real eigenvalues of opposite signs give a saddle point.

• In all these cases a nonlinear system described by

x = Ax + g(x), g(x)/\x\ -> 0, x -> 0

has a phase plane which essentially agrees with that of the linear
part close to the origin.

• In two dimensions a good qualitative picture of the phase plane

is obtained by plotting the slope of the trajectories at different
points.

Literature
The classification of equilibria according to their linearizations was
done by 19th century mathematicians, in particular Poincare. An
overview with many examples is given in Verhulst (1990). Solutions
which do not approach any equilibrium or limit cycle but remain
bounded can show so called chaotic behavior, described in e.g.,
Thompson & Stewart (1986). The use of mathematical results of phase
plane analysis in control theory became common at the middle of the
20th century. It is treated in e.g., Slotine & Li (1991), Leigh (1983)
and Cook (1994).
13.3 Comments 353

Software
In MATLAB trajectories can be simulated using SIMULINK. There are
also the commands
trim Determination of equilibria.
linmod Linearization.
eig Determination of eigenvalues and eigenvectors,
to analyze the character of the equilibria.