Definition of Stability

Stability is an important concept in linear systems —we all want to fly in airplanes with
stable control systems! Although many of us have an intuitive feel for the idea of
“stability,” we need a working definition that will allow us to classify systems as either
stable or unstable.
Working with a partner, draft a definition of stability.
The definition should be specific enough that you can test whether a system with impulse
response g(t) (or equivalently, transfer function G(s)) is stable or not.
Note: There are dozens, and maybe hundreds, of definitions of stability. There is no
“wrong” answer!
When you are finished, press “1” on your PRS remote.
Definition of Stability
Solution
Gu(t) y(t)
The LTI system G is Bounded Input / Bounded Output
(BIBO) stable if every bounded input u(t) produces a bounded output y(t).
Basically, this definition says that every “nice” input produces a “nice” output.
BIBO Stability I
Consider the systems F, G, and H, with impulse responses given by
f(t) = _(t)e−2t g(t) = _(t) h(t) = _(t)et
Which of the systems are BIBO stable?
1. F, G, and H
2. F and G
3. F only
4. G and H
5. H only
6. none of the above
BIBO Stability I
Consider the systems F, G, and H, with impulse responses given by f(t) = _(t)e−2t
g(t) = _(t) h(t) = _(t)et
Which of the systems are BIBO stable?
1. F, G, and H
2. F and G
3. ~ F only
4. G and H
5. H only
6. none of the above
The correct answer is:
BIBO Stability II
Consider the system G with impulse response given by g(t) = 1 1 + t _(t)
Is the system G BIBO stable?
1. Yes
2. No
3. Don’t know
BIBO Stability II
Consider the system G with impulse response given by g(t) =1 1 + t_(t)
Is the system G BIBO stable?
1. Yes
2. ~ No
3. Don’t know
The correct answer is:
BIBO Stability III
Consider the system G with impulse response given by g(t) = 1 p t _(t)
Is the system G BIBO stable?
1. Yes
2. No
3. Don’t know
BIBO Stability III
Consider the system G with impulse response given by g(t) = 1 pt _(t)
Is the system G BIBO stable?
1. Yes
2. ~ No
3. Don’t know
The correct answer is:

Stability of Linear Systems

We consider systems described by the state equations for continuous time systems, and
(1)
(2)
In the Circuits and Systems course, these equations were solved in the frequency domain by using
Laplace transforms for continuous time systems and z-transforms for discrete time systems. This
approach is limited to linear systems. Since we are eventually interested in nonlinear systems, we will
perform the analysis in the time domain.
The free solution (u = 0) can be represented as
(3)
where
(4)
for continuous time systems and
(5) for discrete time systems. Note that in general, x is a vector and A, B, C and D are matrices.

Definition 1:
a) System is asymptotically stable if for all x0 we have

(6)
 b) System is (simply) stable if for all x0 there exists C such that for all t

(7)
c) System is weakly unstable if it is not stable and if for all x0 there exist C and n such that for all t

(8)
d) System is strongly unstable if it is neither stable nor weakly unstable.

Example 1: Discharging a capacitor

Fig.1
State equation:
(9)
Solutions:
(10)
• System is asymptotically stable if R > 0
• System is stable if R = ∞ (open circuit)
• System is strongly unstable if R <0. <in this case the solutions diverge exponentially fast.
Note that the system is never weakly unstable.

Example 2: Free frictionless motion

Fig.2
State equations:
(11)
Solutions:
(12)
• System is weakly unstable, with n = 0.

Example 3: Population growth model

State equation:
(13)
Solution:
(14)
• Asymptotically stable if 0 < a < 1
• Stable for a = 1
• Strongly unstable for a > 1

Note that the system is never weakly unstable.

Example 4: Counter with adjustable increment
The system uses the first state variable to model the increment. State equations:
(15)
Solutions:
(16)
• System is weakly unstable

Conditions for stability, continuous time systems

Form of the free solutions:
(17)
Try to diagonalize the matrix A by a coordinate change:
(18)
This is most of the time possible, but not always. Suppose temporarily that it is possible, i.e. that A is
diagonalizable. Then λi are the eigenvalues of A and the i-th column of S the eigenvector for λi.
We can represent the matrix exponential as
(19)
and thus
 (20)
Hence, the solutions become
(21)
and in transformed coordinates
(22)
This leads directly to the following proposition.

Proposition 1: • The system is stable if all eigenvalues λi satisfy Reλi ≤ 0.

If A is diagonalizable • The system is strongly unstable if Reλi > 0 for at least one
• The system is eigenvalue λi.
asymptotically
stable if all Note that if the system has a diagonalizable matrix A it cannot be
eigenvalues λi weakly unstable. If we take example 2, the matrix A is
satisfy Reλi < 0 (23)
This matrix cannot be diagonalized. Otherwise, it would have two
linearly independent eigenvectors with eigenvalue 0 (since it is a
triangular matrix, all eigenvalues are diagonal elements).But if x is an
eigenvector with eigenvalue 0, we must have
(24)
and thus all eigenvectors lie in the 1-dimensional subspace x2 = 0, i.e. there are no two linearly
independent eigenvectors.

Jordan normal form:

Any matrix A can be reduced to the Jordan normal form:
(25)
where
(26)
and the Jordan blocks Ji
have the form
(27)
Now
(28)
and
(29)
After some calculations, one
obtains
 (30) The solution x(t) is a linear combination of the terms in this matrix.
(31) From this the following theorem follows:
(32)

Theorem 1:
• The system is asymptotically stable if all eigenvalues λi satisfy Reλi < 0
• The system is stable, if all eigenvalues λi satisfy
- either Reλi < 0
- or Reλi = 0 and the corresponding Jordan block is of dimension 1
• The system is weakly unstable if all eigenvalues λi satisfy Reλi ≤ 0 and if there is at least one
eigenvalue λi with Reλi = 0 and Jordan block with dimension higher than 1.
• The system is strongly unstable if there is at least one eigenvalue λi with Reλi > 0.

Conditions for stability, discrete time systems

Everything is similar as in the case of continuous time systems. Form of the free solutions:
 (33)
Jordan normal form
(34)
(35)
(36)
(37)
The solution x(t) is a linear combination of the terms in this matrix. From this the following theorem
follows:

Theorem 2:
• The system is asymptotically stable if all eigenvalues λi satisfy |λi| < 1
• The system is stable, if all eigenvalues λi satisfy
- either |λi| < 1
- or |λi| = 1 and the corresponding Jordan block is of dimension 1
• The system is weakly unstable if all eigenvalues λi satisfy |λi| ≤ 0 and if there is at least one
eigenvalue λi with |λi| = 1 and Jordan block with dimension higher than 1.
• The system is strongly unstable if there is at least one eigenvalue λi with |λi| > 1

Remarks:
1. If the initial conditions x0 coincide with an eigenvector of A, the trajectory x(t) lies on the
straight line through x0 and the origin.
2. A system that has no eigenvalue with Reλi = 0 (continuous time) resp. |λi| = 1 (discrete time) is
called hyperbolic. Hyperbolic systems are robust against parameter changes, and non-
hyperbolic systems are fragile.
 3. There are two kinds of multiplicities of an eigenvalue λi:
- The algebraic multiplicity is the number of times it appears on the diagonal of J. This is
equal to the sum of the dimensions of all Jordan blocks with λi.
- The geometric multiplicity is the number of linearly independent eigenvectors with
eigenvalue λi. This is equal to the number of Jordan blocks with eigenvalue λi.

Connection with frequency domain analysis

Continuous time systems
Example of dimension 1:
(38)
Laplace transform
(39)
The constant a is at the same time eigenvalue of the (1x1) state matrix and pole of the Laplace
transformed free solution, i.e. natural frequency. We now show that this is the case in general.
Dimension n:
(40)
Laplace transform
(41)
This is a system of n linear equations for the components of X(p).
Reminder: Cramers rule for the solution of linear equations
 (42)
Applied to our case:
(43)
Both the numerator and the denominator are polynomials in p. The zeros of the denominator are the
natural frequencies of the system. At the same time they are the eigenvalues of the matrix A.

Discrete time systems:

Everything is similar to the case of continuous time systems:
(44)
Again, the numerator and the denominator are polynomials in z. The zeros of the denominator are the
natural frequencies and the eigenvalues of A
Summary:
Poles of the Laplace/z-transform of the free solution = Natural frequencies
= Eigenvalues of the state matrix A