KON 509E: Design of Discrete Control Systems

Discrete-Time State Equations and

Internal Stability

Assoc. Prof. Dr. İlker Üstoğlu

 Discrete-time State Equations

The basic representation for a discrete-time linear system is the linear state equation:
x[k] x[k+1] x[j-1] x[j]
𝑥 𝑘+1 =𝐴 𝑘 𝑥 𝑘 +𝐵 𝑘 𝑢 𝑘
(1)
…
𝑦 𝑘 = 𝐶 𝑘 𝑥 𝑘 + 𝐷 𝑘 𝑢(𝑘) 𝑘 𝑘+1 𝑗−1 𝑗

input signal 𝑚 × 1

output signal 𝑝 × 1 state vector 𝑛 × 1

state variables

State summarizes effect of past 𝑥1 𝑘 The linear state equation is called time-varying
inputs on future output – like 𝑥2 𝑘 if any entry in any coefficient matrix sequence
𝑥 𝑘 = changes with k.
the memory of the system ⋮
𝑥𝑛 𝑘
Slide 2
Discrete-time State Equations

The standard terminology is that (1) is time invariant if all coefficient-matrix sequences are constant.

𝑥 𝑘+1 =𝑨𝑥 𝑘 +𝑩𝑢 𝑘

𝑦 𝑘 = 𝑪 𝑥 𝑘 + 𝑫 𝑢(𝑘)

{C, A, B, D}

Slide 3
RSolve (Mathematica) / R recurrence! 𝑥 𝑘 +1 = 𝟐 𝑥 𝑘 +𝟏
𝑥 0 =1

k x[k] x[k+1]
0 x[0]= 1 x[1]= 3
1 x[1]= 3 x[2]= 7
2 x[2]= 7 x[3]= 15
3 x[3]= 15 x[4]= 31
4 x[4]= 31 x[5]= 63
5 x[5]= 63 x[6]= 127
6 x[6]= 127 x[7]= 255
pattern: multiply by two, add one! 7 x[7]= 255 x[8]= 511
8 x[8]= 511 x[9]= 1023
9 x[9]= 1023 x[10]= 2047
10 x[10]= 2047 x[11]= 4095
x[k] x[k+1] x[j-1] x[j]

…
𝑘 𝑘+1 𝑗−1 𝑗 Slide 4
State Equation Implementation
A discrete-time linear state equation can be implemented in software on a digital computer.

A state equation also can be implemented directly in electronic hardware using devices that perform
the three underlying operations involved in the state equation:

(signed) sum of scalar multiplication of a scalar signal by unit delay

sequences a (time-varying) coefficient,
𝑥1 𝑘0

𝑥1 𝑘 + 𝑥1 𝑘 − 𝑥2 𝑘
𝑥1 𝑘 + 1 𝑥1 𝑘
_ D
𝑥1 𝑘 α 𝑘 𝑥1 𝑘
α𝑘
𝑥2 𝑘

Slide 5
State Equation Implementation

These basic building blocks can be connected together as prescribed by a given linear state equation
to obtain a state variable diagram:

From a theoretical perspective such a diagram sometimes reveals structural features of the linear
state equation that are not apparent from the coefficient matrices. From an implementation
perspective, a state variable diagram provides a blueprint for hardware realization of the state
equation.
Slide 6
State Equation Solution
Given an initial time 𝑘0, initial state x(𝑘0) = 𝑥0, and input signal u(𝑘) defined for all 𝑘, we can
generate a solution of the linear state equation for 𝑘 > 𝑘0.

existence of a solution uniqueness of the solution Slide 7

State Equation Solution
The situation can be quite different when solution of (1) backward in the time index is attempted.

As a first step, given x0 and u(𝑘0-1), we would want to compute x(𝑘0 − 1) such that, writing (1)
at 𝑘 = 𝑘0 − 1

𝑥 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 +𝐵 𝑘 𝑢 𝑘
𝑘 =𝑘0-1

x(𝑘0) = 𝑥0 = 𝐴 𝑘0 −1 𝑥 𝑘0−1 + 𝐵 𝑘0 −1 𝑢 𝑘0 −1 If 𝐴 𝑘0 −1 is invertible, then

𝑥 𝑘0 −1 = 𝐴−1 𝑘0−1 x0 − 𝐴−1 𝑘0−1 𝐵 𝑘0−1 𝑢 𝑘0−1

If 𝐴 𝑘0 −1 is not invertible, this may yield an infinite

number of solutions for 𝑥 𝑘0−1 , or none at all. Therefore
neither existence nor uniqueness of solutions for k < 𝑘0 can
be claimed in general for (1).
Slide 8
Discrete-time (State) Transition Matrix
If 𝐴 𝑘 is invertible for all 𝑘, then given 𝑘0, x(𝑘0) and u(𝑘) defined for all 𝑘, there exists a unique solution
x(𝑘) of (1) defined for all 𝑘 both backward and forward from 𝑘0.

We typically work only with the forward solution, viewing the backward solution as an uninteresting
artifact.

𝐴 𝑘 − 1 𝐴 𝑘 − 2 … 𝐴(𝑗), 𝑘 ≥𝑗+1
Φ 𝑘, 𝑗 = 𝑥 𝑘 + 1 = 𝐴(𝑘) 𝑥 𝑘
𝐼, 𝑘=𝑗

x(2)=A(1)x(1)
x(3)=A(2)x(2)=A(2)A(1)x(1)
x(4)=A(3)x(3)=A(3)A(2)x(2)=A(3)A(2)A(1)x(1)
…

𝑗 𝑗+1 … 𝑘−1 𝑘 Slide 9

Discrete-time (State) Transition Matrix
The default is to leave Φ 𝑘, 𝑗 undefined for 𝑘 ≤ 𝑗 −1. However under the hypothesis that A 𝑘 is
invertable for every 𝑘 we set

Φ 𝑘, 𝑗 = 𝐴−1(𝑘) 𝐴−1(𝑘 + 1) … 𝐴−1 (𝑗 − 1) , 𝑘 ≤ 𝑗 −1

…
𝑘 𝑘+1 𝑗−1 𝑗

Slide 10
Discrete-time (State) Transition Matrix
In terms of transition-matrix notation, the unique solution of (1) provided by the forward iteration
can be written as

𝑘−1
𝑥 𝑘 = 𝛷 𝑘, 𝑘0 𝑥0+ 𝑗=𝑘0 𝛷 𝑘, 𝑗 + 1 𝐵 𝑗 𝑢 𝑗 , 𝑘 ≥ 𝑘0 + 1 𝑥 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 +𝐵 𝑘 𝑢 𝑘

𝑥 1 =𝐴 0 𝑥 0 +𝐵 0 𝑢 0
𝑥 2 = 𝐴 1 𝑥 1 + 𝐵 1 𝑢 1 = 𝐴 1 (𝐴 0 𝑥 0 + 𝐵 0 𝑢 0 ) + 𝐵 1 𝑢 1
𝑥 3 = 𝐴 2 𝑥 2 + 𝐵 2 𝑢 2 = 𝐴 2 (𝐴 1 (𝐴 0 𝑥 0 + 𝐵 0 𝑢 0 ) + 𝐵 1 𝑢 1 )+𝐵 2 𝑢 2
= (𝐴 2 𝐴 1 𝐴 0 𝑥 0 + 𝐴 2 𝐴 1 𝐵 0 𝑢 0 ) + 𝐴 2 𝐵 1 𝑢 1 ) + 𝐵 2 𝑢 2

Slide 11
The Complete Solution / Zero-Input Response

𝐶 𝑘0 𝑥0 + 𝐷 𝑘0 𝑢0 𝑘 = 𝑘0
𝑦 𝑘 =
𝐶(𝑘)𝛷 𝑘, 𝑘0 𝑥0+ 𝑘−1
𝑗=𝑘0 𝐶 𝑘 𝛷 𝑘, 𝑗 + 1 𝐵 𝑗 𝑢 𝑗 + 𝐷 𝑘 𝑢(𝑘) 𝑘 ≥ 𝑘0 + 1

The 𝑖 𝑡ℎ column of 𝛷 𝑘, 𝑘0 represents the zero-input response to the initial state 𝑥(𝑘0 ) = 𝑒𝑖 , the
𝑖 𝑡ℎ column of 𝐼𝑛 .
0
0
⋮
0
𝐼𝑛 = [𝑒1 𝑒2 … 𝑒𝑖 … 𝑒𝑛 ] 𝑒𝑖 =
1 𝑖 𝑡ℎ 𝑟𝑜𝑤
0
⋮
0
𝑖𝑡ℎ 𝑐𝑜𝑙𝑢𝑚𝑛
Slide 12
The Complete Solution / Zero-Input Response

Thus a transition matrix can be computed for fixed 𝑘0, by computing the zero input response to n
initial states at 𝑘0. In general if 𝑘0changes, then the whole computation must be repeated at the new
initial time.

Slide 13
The Complete Solution / Zero-State Response

The zero-state response can be investigated in terms of a simple class of input signals. Define the
scalar unit pulse signal by

1, 𝑘=0
𝛿 𝑘 =
0, otherwise k0 𝑘

Consider the complete solution for fixed k 0, x(k 0 ) = 0, and the input signal that has all zero entries
except for a unit pulse as the ith entry. That is, u k = ei δ k − k 0 , where ei now is the ith column
of Im . This gives

𝐷 𝑘0 𝑒𝑖 𝑘 = 𝑘0
𝑦 𝑘 =
𝐶 𝑘 𝛷 𝑘, 𝑘0 + 1 𝐵 𝑘0 𝑒𝑖 𝑘 ≥ 𝑘0 + 1

Slide 14
Transition Matrix Properties 1

𝛷 𝑘 + 1, 𝑗 = 𝐴 𝑘 𝛷 𝑘, 𝑗 𝑘≥𝑗

𝑗−1 𝑗 … 𝑘 𝑘+1
𝛷 𝑘, 𝑗 − 1 = 𝛷 𝑘, 𝑗 𝐴 𝑗 − 1 𝑘≥𝑗

𝑥 𝑘 = 𝛷 𝑘, 𝑗 𝑥 𝑗
𝑥 𝑘 + 1 = 𝐴 𝑘 𝑥 𝑘 = 𝐴 𝑘 𝛷 𝑘, 𝑗 𝑥 𝑗 = 𝛷 𝑘 + 1, 𝑗 𝑥 𝑗

𝑥 𝑘 = 𝛷 𝑘, 𝑗 𝑥 𝑗 = 𝛷 𝑘, 𝑗 𝐴 𝑗 − 1 𝑥 𝑗 − 1 = 𝛷 𝑘, 𝑗 − 1 𝑥 𝑗 − 1

Slide 15
 Transition Matrix Properties 2
The transition matrix for an n × n matrix sequence 𝐴 𝑘 satisfies

𝛷 𝑘, 𝑖 = 𝛷 𝑘, 𝑗 · 𝛷 𝑗, 𝑖 𝑖≤𝑗 ≤𝑘

If 𝐴 𝑘 is invertible for every k, then 𝛷 𝑘, 𝑖 = 𝛷 𝑘, 𝑗 𝛷 𝑗, 𝑖 holds without restriction on the indices 𝑖, 𝑗, 𝑘.

𝛷 𝑗, 𝑖 𝛷 𝑘, 𝑗

… 𝑗 …
𝑖 𝑘
Slide 16
 Transition Matrix Properties 3

If the n × n matrix sequence 𝐴 𝑘 is invertible for every 𝑘, then the transition matrix 𝛷 𝑘, 𝑗 is
invertible for every 𝑘 and 𝑗, and

𝛷 −1 𝑘, 𝑗 = 𝛷 𝑗, 𝑘

Slide 17
 Transition Matrix Properties 4

Suppose 𝑃 𝑘 is an n × n matrix sequence that is invertible at each 𝑘.

If the transition matrix for the n × n matris sequence 𝐴 𝑘 𝑖𝑠 𝛷𝐴 𝑘, 𝑗 , 𝑘 ≥ 𝑗, then the transition matrix for
F 𝑘 = 𝑃 −1 𝑘 + 1 𝐴 𝑘 𝑃(𝑘) is 𝛷𝐹 𝑘, 𝑗 = 𝑃 −1 𝑘 𝛷𝐴 𝑘, 𝑗 𝑃 𝑗 , 𝑘 ≥ 𝑗.

state variable change!

𝑥 𝑘 =𝑃 𝑘 𝑧 𝑘
𝑥 𝑘+1 = 𝑃 𝑘+1 𝑧 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 +𝐵 𝑘 𝑢 𝑘
𝑧 𝑘 + 1 = 𝑃 −1 𝑘 + 1 𝐴 𝑘 𝑥 𝑘 + 𝑃 −1 𝑘 + 1 𝐵 𝑘 𝑢 𝑘
= 𝑃 −1 𝑘 + 1 𝐴 𝑘 𝑃 𝑘 𝑧 𝑘 + 𝑃 −1 𝑘 + 1 𝐵 𝑘 𝑢 𝑘

𝑧(𝑘0 )=𝑃 −1 𝑘0 𝑥 𝑘0
𝑦 𝑘 = 𝐶 𝑘 𝑃 𝑘 𝑧 𝑘 + 𝐷 𝑘 𝑢(𝑘) Slide 18
 Transition Matrix Properties 4
𝑥 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 +𝐵 𝑘 𝑢 𝑘

𝑦 𝑘 = 𝐶 𝑘 𝑥 𝑘 + 𝐷 𝑘 𝑢(𝑘)

𝑥 𝑘0

𝑧 𝑘 + 1 = 𝑷−𝟏 𝒌 + 𝟏 𝑨 𝒌 𝑷 𝒌 𝑧 𝑘 + 𝑷−𝟏 𝒌 + 𝟏 𝑩 𝒌 𝑢 𝑘

𝑦 𝑘 = 𝑪 𝒌 𝑷 𝒌 𝑧 𝑘 + 𝑫 𝒌 𝑢(𝑘)

𝑧(𝑘0 )=𝑃 −1 𝑘0 𝑥 𝑘0

Slide 19
 Transition Matrix Properties 4
𝑥 𝑘 =𝑃 𝑘 𝑧 𝑘

(𝐶 𝑘 ,𝐴 𝑘 ,𝐵 𝑘 ,𝐷 𝑘 )

( 𝐶 𝑘 𝑃 𝑘 , 𝑃 −1 𝑘 + 1 𝐴 𝑘 𝑃 𝑘 , 𝑃 −1 𝑘 + 1 𝐵 𝑘 , 𝐷 𝑘 )

k=0 𝑧 1 = 𝑷−𝟏 𝟏 𝑨 𝟎 𝑷 𝟎 𝑧 0 + 𝑷−𝟏 𝟏 𝑩 𝟎 𝑢 0

= 𝑷−𝟏 𝟏 𝑨 𝟎 𝑷 𝟎 𝑃−1 0 𝑥(0) + 𝑷−𝟏 𝟏 𝑩 𝟎 𝑢 0
= 𝑷−𝟏 𝟏 (𝑨 𝟎 𝑥(0) + 𝑩 𝟎 𝑢 0 ) = 𝑷−𝟏 𝟏 𝑥(1) 𝑧(𝑘0 )=𝑃 −1 𝑘0 𝑥 𝑘0

Slide 20
 Transition Matrix Properties 5

The linear n × n matrix difference equation

𝑋 𝑘+1 = 𝐴 𝑘 𝑋 𝑘 , 𝑋 𝑘0 = 𝐼

has the unique solution

𝑋 𝑘 = 𝛷𝐴 𝑘, 𝑘0 , 𝑘 ≥ 𝑘0

If the initial condition is an arbitrary n × n matrix 𝑋 𝑘0 =𝑋0, in place of identity, then the unique
solution for 𝑘 ≥ 𝑘0 is 𝑋 𝑘 = 𝛷𝐴 𝑘, 𝑘0 𝑋0.

Slide 21
 Transition Matrix Properties 6

The linear n × n matrix difference equation

𝑍 𝑘 − 1 = 𝐴𝑇 𝑘 − 1 𝑍 𝑘 , 𝑍 𝑘0 = 𝐼

has the unique solution

𝑍 𝑘 = 𝛷𝐴 𝑇 𝑘0, 𝑘 , 𝑘 ≤ 𝑘0

𝑧 𝑘 = 𝛷𝐴 𝑇 𝑘0 ,𝑘 𝑧0 𝑘 ≤ 𝑘0
𝑥 𝑘 = 𝛷𝐴 𝑘, 𝑘0 𝑥0 𝑘 ≥ 𝑘0

Slide 22
 Special Case 1: Time-Invariant Case

If all coefficient matrices are constant, then standard notation for the discrete-time linear
state equation is

𝑥 𝑘+1 = 𝐴𝑥 𝑘 +𝐵𝑢 𝑘 ,

𝑦 𝑘 =𝐶 𝑥 𝑘 +𝐷𝑢 𝑘

The transition matrix for the matrix A follows directly from the general formula in the time-varying
case as

𝛷𝐴 𝑘, 𝑗 = 𝐴𝑘−𝑗 , 𝑘 ≥ 𝑗.

If A is invertible, then this definition extends to k < j without writing a separate formula.

Slide 23
 Time-Invariant Case

𝐶 𝑥0 + 𝐷 𝑢 𝑘0 𝑘 = 𝑘0
𝑦 𝑘 =
𝐶 𝐴𝑘−𝑘0 𝑥0 + 𝑘−1
𝑗=𝑘0 𝐶 𝐴
𝑘−𝑗−1 𝐵 𝑢 𝑗 + 𝐷 𝑢(𝑘) 𝑘 ≥ 𝑘0 + 1

𝑘−1

𝑘0 = 0 𝑦 𝑘 = 𝐶 𝐴𝑘 𝑥0+ 𝐶 𝐴𝑘−𝑗−1 𝐵 𝑢 𝑗 + 𝐷 𝑢(𝑘) 𝑘≥1

𝑗=0

If the matrix A is invertible then:

−1

𝑘0 = 0 𝑦 𝑘 = 𝐶 𝐴𝑘 𝑥0 − 𝐶 𝐴𝑘−𝑗−1 𝐵 𝑢 𝑗 + 𝐷 𝑢 𝑘 𝑘<0
𝑗=𝑘

We do not consider solutions for k < 0 unless special mention is made. Slide 24
 Time-Invariant Case
By defining the p × m matrix sequence

𝐷 𝑘=0 (unit-pulse response)

𝐺 𝑘 =
𝐶 𝐴𝑘−1𝐵 𝑘≥1

we can write the (forward) solution as

𝑘

𝑦 𝑘 = 𝐶 𝐴𝑘 𝑥0 + 𝐺(𝑘 − 𝑗) 𝑢 𝑗 𝑘≥0
𝑗=0

(convolution)

Slide 25
 Time-Invariant Case
Recall that 𝛷𝐴 𝑘, 0 = 𝐴𝑘 , 𝑘 ≥ 0 is the unique solution of the n × n matrix difference equation
𝑋 𝑘+1 = 𝐴𝑋 𝑘 , 𝑋 0 = 𝐼

The z-transform, can be used to develop a representation for 𝐴𝑘 as follows. Applying the z-transform
to both sides yields an algebraic equation in X(z) = Z{X(k)} that solves to

X(z) = 𝑧 (𝑧𝐼 − 𝐴)−1

𝑧 𝑎𝑑𝑗(𝑧𝐼−𝐴)
Z{𝐴𝑘} = 𝑧 (𝑧𝐼 − 𝐴)−1 = Z{𝑋 𝑘 + 1 } = z X(z) – z 𝑋 0 = 𝐴 X(z)
det(𝑧𝐼−𝐴)

det (𝑧𝐼 − 𝐴) is a degree-n polynomial in z, so (𝑧𝐼 − 𝐴)−1exists for all but at most n values of z.

Slide 26
Time-Invariant Case

Slide 27
Time-Invariant Case

𝐴𝑘

Slide 28
Transfer Function

𝟏
0 1
𝟐 1
𝑥 𝑘+1 = 𝑥 𝑘 + 𝑢 𝑘 𝑥1 𝑘 + 1 = 𝑥1 𝑘 + 𝑢 𝑘
𝟏 0 2
0
𝟑 𝐵
1
𝐴 𝑥2 𝑘 + 1 = 𝑥2 𝑘
3
𝐶 𝐷
𝑦 𝑘 = 1 0 𝑥 𝑘 +1𝑢 𝑘 𝑦 𝑘 = 𝑥1 𝑘 + 𝑢 𝑘

Slide 29
 Transfer Function
𝟏
uncontrollable, unobservable
G(z) = 𝐶 (𝑧𝐼 − 𝐴)−1𝐵 + 𝐷 𝟑
eigenvalues
det(z I -A) = 0 𝟏
𝟏 𝟐
0
𝟐 1
𝑥 𝑘+1 = 𝑥 𝑘 + 𝑢 𝑘
𝟏 0
0 G(z) 1 + 2𝑧
𝟑 𝐺 𝑧 =
−1 + 2𝑧
1 1
𝑦 𝑘 = 1 0 𝑥 𝑘 +𝑢 𝑘 −
2 2
zero pole

Slide 30
 Special Case 2: Periodic Case

The second special case we consider involves linear state equations with coefficients that are
repetitive matrix sequences.

A matrix sequence F(k) is called K-periodic if K is a positive integer such that for all k, F(k + K) = F(k).

It is convenient to call the least such integer K the period of F(k).

If K = 1, then F(k) is constant.

Slide 31
 Special Case 2: Periodic Case

Example:

(−1)𝑘 0
𝑥 𝑘+1 = 𝑥 𝑘
0 1

(−1) 𝑘 0
𝐴 𝑘 =
0 1

(−1) 𝑘+𝐾 0 ? (−1)𝑘 0

𝐴 𝑘+𝐾 = = =𝐴 𝑘 ⇔ 𝐾 = 2.
0 1 0 1

𝐴 𝑘 is a "2 −periodic" matrix sequence.

Slide 32
 Special Case 2: Periodic Case

Property

Suppose the 𝑛 × 𝑛 matrix sequence 𝐴 𝑘 is invertible for every 𝑘 and 𝐾 −periodic. Then the transition
matrix for 𝐴 𝑘 can be written in the form

𝛷 𝑘, 𝑗 = 𝑃 𝑘 𝑅𝑘−𝑗 𝑃 −1(𝑗)

for all 𝑘, 𝑗, where 𝑅 is a constant (possibly complex), invertible, 𝑛 × 𝑛 matrix , and 𝑃(𝑘) is a
𝐾 −periodic, 𝑛 × 𝑛 matrix sequence that is invertible for every 𝑘.

Slide 33
Special Case 2: Periodic Case
Proof of Property:

Define an 𝑛 × 𝑛 matrix 𝑅 by setting

𝑅 𝐾 = 𝛷 𝐾, 0 .

Also define 𝑃 𝑘 via

𝑃 𝑘 = 𝛷 𝑘, 0 𝑅−𝑘 .

Obviously 𝑃 𝑘 is invertible for every 𝑘.

Slide 34
Special Case 2: Periodic Case
Proof of Property:

𝑃 𝑘 + 𝐾 = 𝛷 𝑘 + 𝐾, 0 𝑅 −(𝑘+𝐾) = 𝛷 𝑘 + 𝐾, 𝐾 𝛷 𝐾, 0 𝑅−𝐾 𝑅−𝑘

Since 𝛷 𝐾, 0 𝑅−𝐾 = 𝐼 (identity matrix),

𝑃 𝑘 + 𝐾 = 𝛷 𝑘 + 𝐾, 𝐾 𝑅−𝑘.

From the definition of the state transition matrix and the periodicity property of A(k):

𝛷 𝑘 + 𝐾, 𝐾 = 𝛷 𝑘, 0 ∀𝑘.

Thus we obtain 𝑃 𝑘 + 𝐾 = 𝑃(𝑘) ∀𝑘.

Slide 35
Special Case 2: Periodic Case
Proof of Property:

𝑃 𝑘 = 𝛷 𝑘, 0 𝑅−𝑘 𝑃 𝑘 𝑅 𝑘 = 𝛷 𝑘, 0

𝑃 −1 𝑘 = 𝑅 𝑘 𝛷 𝑘, 0 −1= 𝑅 𝑘 𝛷 0,𝑘

𝑃 −1 𝑗 = 𝑅 𝑗 𝛷 0,𝑗

𝛷 0,𝑗 = 𝑅 −𝑗 𝑃−1 𝑗

𝛷 𝑘, 𝑗 = 𝛷 𝑘, 0 𝛷 0, 𝑗 = 𝑃 𝑘 𝑅𝑘 𝑅−𝑗 𝑃−1 𝑗 = 𝑃 𝑘 𝑅𝑘−𝑗 𝑃−1 𝑗

Slide 36
Special Case 2: Periodic Case - Example
𝑥 𝑘 = 𝛷 𝑘, 𝑗 𝑥 𝑗 = 𝑃 𝑘 𝑅𝑘−𝑗 𝑃−1 𝑗 𝑥 𝑗

𝑃 𝑘 = 𝛷 𝑘, 0 𝑅−𝑘

𝑅 𝐾 = 𝛷 𝐾, 0 "2 −periodic" (−1)𝑘 0

𝑥 𝑘+1 = 𝑥 𝑘
0 1
−1 0 𝑥1 (𝑘)
𝑅 2 = 𝛷 2,0 = A 1 A 0 = 𝑥 𝑘 =
0 1 𝑥2 (𝑘)

𝑖 0 3
⇔𝑅 = , 𝑖 = −1 𝑥 0 =
0 1 4

𝑃 0 = 𝛷 0,0 𝑅 −0 = 𝐼2 (identity matrix)

−𝑖 0
𝑃 1 = 𝛷 1,0 𝑅 −1 = 𝐴 0 𝑅 −1= = {𝑃 3 , 𝑃 5 , … }
0 1
𝑃 2 = 𝛷 2,0 𝑅 −2 = 𝐴 1 𝐴 0 𝑅−2 = 𝐼2 = {𝑃 4 , 𝑃 6 , … } Slide 37
Special Case 2: Periodic Case - Example
𝛷 𝑘, 𝑗 = 𝑃 𝑘 𝑅𝑘−𝑗 𝑃 −1 𝑗 𝛷 𝑘, 𝑗 = 𝑃 𝑘 𝑅𝑘−𝑗 𝑃 −1 𝑗

1 0
𝑘 𝑒𝑣𝑒𝑛
𝑃 𝑘 = 0 1 1 0 𝑖 0 𝑘−𝑗 1 0
−𝑖 0 𝑘, 𝑗 𝑒𝑣𝑒𝑛
𝑘 𝑜𝑑𝑑 0 1 0 1 0 1
0 1 −𝑖 0 𝑖 0 𝑘−𝑗
𝑖 0
𝑘, 𝑗 𝑜𝑑𝑑
𝛷 𝑘, 𝑗 = 0 1 0 1 0 1
1 0 𝑖 0 𝑘−𝑗 𝑖 0
1 0 𝑘 𝑒𝑣𝑒𝑛, 𝑗 𝑜𝑑𝑑
𝑘 𝑒𝑣𝑒𝑛 0 1 0 1 0 1
𝑃 −1 𝑗 = 0 1
𝑖 0 −𝑖 0 𝑖 0 𝑘−𝑗 1 0
𝑘 𝑜𝑑𝑑 𝑘 𝑜𝑑𝑑, 𝑗 𝑒𝑣𝑒𝑛
0 1 0 1 0 1 0 1

𝑖 0
𝑅=
0 1

Slide 38
Special Case 2: Periodic Case - Example
𝑘−𝑗
𝑖 0
𝑘, 𝑗 𝑒𝑣𝑒𝑛
0 1
−𝑖 0 𝑖 0 𝑘−𝑗 𝑖 0
𝑘, 𝑗 𝑜𝑑𝑑
𝛷 𝑘, 𝑗 = 0 1 0 1 0 1
k>j
𝑖 0 𝑘−𝑗 𝑖 0
𝑘 𝑒𝑣𝑒𝑛, 𝑗 𝑜𝑑𝑑
0 1 0 1
−𝑖 0 𝑖 0 𝑘−𝑗
𝑘 𝑜𝑑𝑑, 𝑗 𝑒𝑣𝑒𝑛
0 1 0 1

−1 𝑘 𝑖 𝑘+𝑘2 0 𝑥(0)
𝑥 𝑘 =
0 1

Slide 39
Special Case 2: Periodic Case - Example

(−1)𝑘 0
𝑥 𝑘+1 = 𝑥 𝑘
0 1

𝑥1 (𝑘)
𝑥 𝑘 =
𝑥2 (𝑘)

3
𝑥 0 =
4

Slide 40
Special Case 2: Periodic Case
𝑥 𝑘 + 𝑗𝐾 = 𝛷 𝑘 + 𝑗𝐾, 𝑘0 𝑥0 = 𝛷 𝑘 + 𝑗𝐾, 𝑘 + (𝑗 − 1 𝐾) 𝛷 𝑘 + (𝑗 − 1)𝐾, 𝑘 + (𝑗 − 2 𝐾) … 𝛷 𝑘 + 𝐾, 𝑘 𝛷 𝑘, 𝑘0 𝑥0

= 𝑃 𝑘 + 𝑗𝐾 𝑅 𝐾 𝑃−1(𝑘 + 𝑗 − 1 𝐾) 𝑃 𝑘 + (𝑗 − 1)𝐾 𝑅 𝐾 𝑃−1(𝑘 + 𝑗 − 2 𝐾)… 𝑃 𝑘 + 𝐾 𝑅 𝐾 𝑃 −1 𝑘 𝑥(𝑘)

𝑥 𝑘 + 𝑗𝐾 = 𝑃 𝑘 (𝑅 𝐾 )𝑗 𝑃−1 𝑘 𝑥(𝑘)

If all eigenvalues of 𝑅 𝐾 have magnitude less than unity, then the zero-input solution goes to zero.

If 𝑅 𝐾has at least one eigenvalue with magnitude greater than unity, there are initial states (formed from
corresponding eigenvectors) for which the solution grows without bound.

The case where R has at least one unity eigenvalue relates to existence of K-periodic solutions.

Slide 41
Special Case 2: Periodic Case

Theorem:

Suppose 𝐴 𝑘 is invertible for every 𝑘 and 𝐾 −periodic.

Given any 𝑘0 there exists a nonzero 𝑥0 such that the solution of 𝑥 𝑘 + 1 = 𝐴 𝑘 𝑥 𝑘 , 𝑥 𝑘0 = 𝑥0 is

𝐾 −periodic if and only if (iff) at least one eigenvalue of 𝑅 𝐾 = 𝛷 𝐾, 0 is unity.

Theorem:

For the time-invariant linear state equation 𝑥 𝑘 + 1 = 𝐴 𝑥 𝑘 + 𝐵 𝑢(𝑘), 𝑥 0 = 𝑥0, suppose 𝐴 is invertible.
If 𝐴𝐾 has no unity eigenvalue, then for every 𝐾 −periodic input signal 𝑢(𝑘) there exists an 𝑥0 such that the
corresponding solution is 𝐾 −periodic.
Slide 42
 Uniform Stability

Internal stability deals with boundedness properties and asymptotic behavior (as 𝑘 → ∞ ) of
solutions of the zero-input linear state equation

(1) 𝑥 𝑘 + 1 = 𝐴 𝑘 𝑥 𝑘 , 𝑥 𝑘0 = 𝑥0

The first notion involves boundedness of solutions of (1):

Definition:
The discrete-time linear state equation (1) is called uniformly stable if there exists a finite positive
constant 𝜸 such that for any 𝑘0 and 𝑥0 the corresponding solution satisfies

𝑥(𝑘) ≤ 𝛾 𝑥0 , 𝑘 ≥ 𝑘0. (vector norm)

𝑘 = 𝑘0 shows that the constant 𝜸 must satisfy 𝜸 ≥ 1.

Slide 43
 Uniform Stability

𝑥1 𝑘0
𝑥3 𝑘
𝒙0 = 𝒙 𝑘0 = 𝑥2 𝑘0
𝑥3 𝑘0
𝒙 𝑘0 + 2
𝑥2 𝑘

𝑥1 𝑘 𝒙 𝑘0 + 1

The adjective uniform in the definition refers precisely to the fact that 𝜸 must not depend on the
choice of initial time.

Slide 44
Uniform Stability

The linear state equation (1) is uniformly stable iff there exists a positive constant 𝜸 such that

‖Φ(𝑘, 𝑗)‖ ≤ 𝜸 (induced norm)

for all 𝑘, 𝑗 such that 𝑘 ≥ 𝑗.

Slide 45
 Uniform Exponential Stability

Next we consider a stability property for (1) that addresses both boundedness of solutions and
asymptotic behavior of solutions.

It implies uniform stability, and imposes an additional requirement that all solutions approach zero
exponentially as 𝑘 → ∞.

(1) 𝑥 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 , 𝑥 𝑘0 = 𝑥0

Definition:
The linear state equation (1) is called uniformly exponentially stable if there exists a finite positive
constant 𝜸 and a constant 0 ≤ 𝝀 < 1 such that for any 𝑘0 and 𝑥0 the corresponding solution satisfies

𝑥(𝑘) ≤ 𝜸 𝝀𝑘−𝑘0 𝑥0 , 𝑘 ≥ 𝑘0.

Slide 46
Uniform Exponential Stability

𝜸 is no less than unity, and the adjective uniform refers to the fact that 𝜸 and 𝝀 are independent of 𝑘0.

The property of uniform exponential stability can be expressed in terms of an exponential bound on
the transition matrix norm:
Slide 47
 Uniform Exponential Stability

(1) 𝑥 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 , 𝑥 𝑘0 = 𝑥0

Theorem:
The linear state equation (1) is uniformly exponentially stable iff (if and only if) there exists a finite
positive constant 𝜸 and a constant 0 ≤ 𝝀 < 1 such that

‖Φ(𝑘, 𝑗)‖ ≤ 𝜸 𝝀𝑘−𝑗

for all 𝑘, 𝑗 such that 𝑘 ≥ 𝑗.

Slide 48
 Uniform Exponential Stability

(1) 𝑥 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 , 𝑥 𝑘0 = 𝑥0

Proof:
First suppose 𝜸 > 0 and 0 ≤ 𝝀 < 1 such that ‖Φ(𝑘, 𝑗)‖ ≤ 𝜸 𝝀𝑘−𝑗 holds for all 𝑘, 𝑗 such that 𝑘 ≥ 𝑗.
Then for any 𝑘0 and 𝑥0 the solution of (1) satisfies

𝑥(𝑘) = ‖Φ 𝑘, 𝑘0 𝑥0‖ ≤ ‖Φ 𝑘, 𝑘0 ‖ ‖𝑥0‖ ≤ 𝜸 𝝀𝑘−𝑘0 𝑥0 , 𝑘 ≥ 𝑘0

and uniform exponential stability is established. For the reverse implication suppose that the state
equation (1) is uniformly exponentially stable. Then there is a finite 𝜸 > 0 and 0 ≤ 𝝀 < 1 such that
for any 𝑘0 and 𝑥0 the corresponding solution satisfies 𝑥(𝑘) ≤ 𝜸 𝝀𝑘−𝑘0 𝑥0 , 𝑘 ≥ 𝑘0.

Slide 49
 Uniform Exponential Stability

(1) 𝑥 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 , 𝑥 𝑘0 = 𝑥0

Proof:
For the reverse implication suppose that the state equation (1) is uniformly exponentially stable.
Then there is a finite 𝜸 > 0 and 0 ≤ 𝝀 < 1 such that for any 𝑘0 and 𝑥0 the corresponding solution
satisfies 𝑥(𝑘) ≤ 𝜸 𝝀𝑘−𝑘0 𝑥0 , 𝑘 ≥ 𝑘0. Given any 𝑘0 and 𝑘𝑎 ≥ 𝑘0, let 𝑥𝑎 be such that ‖𝑥𝑎 ‖ = 1,
‖Φ 𝑘𝑎 ,𝑘0 𝑥𝑎 ‖ =‖Φ 𝑘𝑎 , 𝑘0 ‖. (Such an 𝑥𝑎 exists by definition of the induced norm.)

Then the initial state 𝑥 𝑘0 = 𝑥𝑎 yields a solution of (1) that at time 𝑘𝑎 satisfies

𝑥(𝑘𝑎 ) = ‖Φ 𝑘𝑎, 𝑘0 𝑥𝑎‖ ≤ ‖Φ 𝑘𝑎, 𝑘0 ‖ ≤ 𝜸 𝝀𝑘𝑎−𝑘0 𝑥𝑎

Since ‖𝑥𝑎 ‖ = 1, this shows that ‖Φ 𝑘𝑎 ,𝑘0 ‖ ≤ 𝜸 𝝀𝑘𝑎−𝑘0. Because such an 𝑥𝑎 can be selected for any
𝑘0 and 𝑘𝑎 ≥ 𝑘0. The proof is complete.
Slide 50
 Uniform Exponential Stability

(1) 𝑥 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 , 𝑥 𝑘0 = 𝑥0

Theorem:
The linear state equation (1) is uniformly exponentially stable iff there exists a finite positive
constant 𝜷 such that
𝑘
‖Φ(𝑘, 𝑖)‖ ≤ 𝜷
𝑖=𝑗+1

for all 𝑘, 𝑗 such that 𝑘 ≥ 𝑗 + 1.

Slide 51
 Uniform Exponential Stability

(1) 𝑥 𝑘+1 = 𝐴 𝑘 𝑥 𝑘 , 𝑥 𝑘0 = 𝑥0

Proof:
If the state equation (1) is uniformly exponentially stable then there exist finite 𝜸 > 0 and 0 ≤ 𝝀 < 1
such that ‖Φ(𝑘, 𝑖)‖ ≤ 𝜸 𝝀𝑘−𝑖 for all 𝑘, 𝑖 such that 𝑘 ≥ 𝑖.

𝑘 𝑘 𝑘−𝑗−1 ∞
1
‖Φ(𝑘, 𝑖)‖ ≤ 𝜸 𝝀𝑘−𝑖 = 𝜸 𝝀𝑞 ≤ 𝜸 𝝀𝑞 ≤ 𝜸 =𝜷
1−𝝀
𝑖=𝑗+1 𝑖=𝑗+1 𝑞=0 𝑞=0

for all 𝑘, 𝑗 such that 𝑘 ≥ 𝑗 + 1.

𝛾
𝛽= Slide 52
1−𝜆
Uniform Exponential Stability

Proof:

𝑘
Conversely suppose 𝑖=𝑗+1 ‖Φ(𝑘, 𝑖)‖ ≤ 𝛽 holds for all 𝑘, 𝑗 such that 𝑘 ≥ 𝑗 + 1.

𝑘 𝑘

Φ 𝑘, 𝑗 = 𝐼 + {Φ 𝑘, 𝑖 − 1 − Φ 𝑘, 𝑖 } = 𝐼 + {Φ 𝑘, 𝑖 𝐴(𝑖 − 1) − Φ 𝑘, 𝑖 }
𝑖=𝑗+1 𝑖=𝑗+1

{telescoping sum}

{see Rugh, Linear System Theory, 2nd Ed. pg. 426, 22.8 Theorem}

‖Φ(𝑘, 𝑗)‖ ≤ 𝜸 𝝀𝑘−𝑗

Slide 53
Uniform Exponential Stability

For time-invariant linear state equations, where 𝐴 𝑘 = 𝐴 and Φ 𝑘, 𝑗 = 𝐴𝑘−𝑗 , a summation-variable

change in 𝑘𝑖=𝑗+1 ‖Φ(𝑘, 𝑖)‖ ≤ 𝛽 shows that uniform exponential stability is equavalent to existence
of a finite 𝛽 such that ∞
𝑘=0 ‖𝐴 ‖ ≤ 𝛽.
𝑘

The adjective 'uniform' is useless in the time-invariant case, and we drop it in clear contexts. Though
exponential stability usually is called asymptotic stability when discussing time-invariant linear
state equations, we keep the term exponential stability.

A linear state equation (1) with constant 𝐴 𝑘 = 𝐴 is exponentially stable if and only if all
eigenvalues of A have magnitude strictly less than unity.

Slide 54
Uniform Exponential Stability

lim 𝐴𝑘 = 0 is a necessary and sufficient condition for uniform exponential stability in the time-
𝑘→∞
invariant case.

The analogous statement for time-varying linear state equations is not true.

Slide 55
Example

𝑓 𝑘+1 For any 𝑘0, lim Φ 𝑘, 𝑘0 = 0.

𝑥 𝑘+1 = 𝑥 𝑘 , 𝑥 𝑘0 = 𝑥0 𝑘→∞
𝑓 𝑘
However with 𝑘0 = 1 suppose there exist 𝜸 > 0
𝑓 𝑘 and 0 ≤ 𝝀 < 1 such that
Φ 𝑘, 𝑗 =
𝑓 𝑗
1
≤𝜸 𝝀𝑘−1, 𝑘 ≥ 1
𝑘

1 𝑘≤0 1
𝑓 𝑘 = 1 This implies ≤ 𝑘 𝝀𝑘−1, 𝑘 ≥ 1
𝜸
𝑘>0
𝑘
Which is a contradiction since 0 ≤ 𝝀 < 1.
𝑘0
𝑘 ≥ 𝑘0 > 0 Thus the state equation is not uniformly
𝑘
exponentially stable.
Φ 𝑘, 𝑘0 = 1
𝑘 ≥ 0 > 𝑘0
𝑘
1 0 > 𝑘 ≥ 𝑘0 Slide 56
Finite-time Asymptotic Stability

Case: 𝐴(𝑘) is a constant, Nilpotent matrix …

The response to every initial state is zero after a finite number of time steps!

Finite-time asymptotic stability does not occur in continuous-time linear state equations.

Slide 57
Uniform Asymptotic Stability

Definition
The linear state equation (1) is called uniformly asymptotically stable if it is uniformly stable, and if
given any positive constant 𝜹 there exists a positive integer 𝐾 such that for any 𝑘0 and 𝑥0, the
corresponding solution satisfies ‖𝑥(𝑘)‖ ≤ 𝜹 ‖𝑥0‖ , 𝑘 ≥ 𝑘0 + 𝐾.

Note that the elapsed time 𝐾 until the solution satisfies the bound {‖𝑥(𝑘)‖ ≤ 𝜹 ‖𝑥0‖ } must be
independent of the initial time

Slide 58
Uniform Asymptotic Stability

Theorem

The linear state equation (1) is uniformly asymptotically stable iff it is uniformly exponentially stable.

Proof

Suppose that the state equation is uniformly exponentially stable, that is, there exist finite positive 𝜸
and 0 ≤ 𝝀 < 1 such that ‖Φ(𝑘, 𝑗)‖ ≤ 𝜸 𝝀𝑘−𝑗 for all 𝑘, 𝑗 whenever 𝑘 ≥ 𝑗. Then the state equation clearly
is uniformly stable. To show it is uniformly asymptotically stable, for a given 𝜹 > 0 select a positive
integer 𝐾 such that 𝝀𝐾 ≤ 𝜹/𝜸. Then for any 𝑘0 and 𝑥0 and 𝑘 ≥ 𝑘0 + 𝐾.

‖𝑥(𝑘)‖ = ‖Φ(𝑘, 𝑘0 )𝑥0‖ ≤ ‖Φ(𝑘, 𝑘0 )‖ ‖𝑥0‖ ≤ 𝜸 𝝀𝑘−𝑘0 ‖𝑥0 ‖ ≤ 𝜸 𝝀𝐾 ‖𝑥0 ‖ ≤ 𝜹‖𝑥0 ‖, 𝑘 ≥ 𝑘0 + 𝐾.

This demonstrates uniform asymptotic stability.

Slide 59
Uniform Asymptotic Stability

Proof

Conversely suppose the state equation is uniformly asymptotically stable.

{see Rugh, Linear System Theory, 2nd Ed. pg. 431, 22.14 Theorem}

Slide 60