Linear Dynamical Systems

Week 1- State-space solutions and

realizations

1
Linear Dynamical
Outline of
Week 1

1 Introduction
2 State-space solution of linear
systems Linear Time Varying
(LTV) systems Linear Time
3 Invariant (LTI) systems
4 Equivalent representation of linear state-space
systems Realization problem and its solution

2
Linear Dynamical
Dynamical
System

Inpu Dynamic Outpu

t al
t
system

Continuous-Time (CT): accepts CT signals and

generates CT signals
Discrete-Time (DT): accepts DT signals and
generates DT signals
3
Linear Dynamical
Dynamical
System
Causality:
1 If the current output depends on past and current
input(s) but not on future input(s)
2 a necessary condition for a system to be built in the
real world

4
Linear Dynamical
Dynamical
System
Causality:
1 If the current output depends on past and current
input(s) but not on future input(s)
2 a necessary condition for a system to be built in the
real world
3 “Current Output of a causal system is affected by

past input”
Question
How far back in time will the past input affects the
current output?

4
Linear Dynamical
Dynamical
System
Causality:
1 If the current output depends on past and current
input(s) but not on future input(s)
2 a necessary condition for a system to be built in the
real world
3 “Current Output of a causal system is affected by

past input”
Question
How far back in time will the past input affects the
current output?

Answer S
u(t), −∞ < t −−→ y(t)
However, tracking u(t) from t = −∞ is very
inconvenient.
4
Linear Dynamical
Dynamical
System
Causality:
1 If the current output depends on past and current
input(s) but not on future input(s)
2 a necessary condition for a system to be built in the
real world
3 “Current Output of a causal system is affected by

past input”
Question
How far back in time will the past input affects the
current output?

Answer S
u(t), −∞ < t −−→ y(t)
However, tracking u(t) from t = −∞ is very
inconvenient.
the concept of state deals with this
problem! 4
Linear Dynamical
Dynamical
System

State:
1 The state x(t 0 ) of a system at time t0 is the
information at t0 that, together with the input u(t), for
t ≥ t0 determines uniquely the output y(t) ∀t ≥ t0

5
Linear Dynamical
Dynamical
System

State:
1 The state x(t 0 ) of a system at time t0 is the
information at t0 that, together with the input u(t), for
t ≥ t0 determines uniquely the output y(t) ∀t ≥ t0

2
no need to know the input u(t) applied
before t0 in determining the output y(t)
after t0.
3 the state summarizes the effect of past input on
future output

5
Linear Dynamical
Exampl
e

Consider the electrical

circuit

If we know the voltages x 1 (t 0 ) and x 2 (t 0 ) across the two

capacitors and the current x 3 (t 0 ) passing through the
inductor...
6
Linear Dynamical
Example:
Continued...

...then for any input applied on and after t0 you can

determine uniquely the output for t ≥ t0

7
Linear Dynamical
Example:
Continued...

...then for any input applied on and after t0 you can

determine uniquely the output for t ≥ t0

 
x1
1 State Variables x = 2  x 3

x∈3 R

2 S : using the state

at t0 )
x(t 0 ) S
−−→ y(t), t ≥ t0
u(t), t0 ≤
t
7
Linear Dynamical
Dynamical Systems:
Linearity
S is linear if (Superposition
Property)
α 1x 1(t0) + α 2x 2(t0) ) S
— −−→1α1 y (t) +2 α y (t), t 0
α 1u1(t) + α 2u2(t), t ≥ t0 ≥ t 2

for any real constants

α 1, α 2

8
Linear Dynamical
Dynamical Systems:
Linearity
S is linear if (Superposition
Property)
α 1x 1(t0) + α 2x 2(t0) ) S
— −−→1α1 y (t) +2 α y (t), t 0
α 1u1(t) + α 2u2(t), t ≥ t0 ≥ t 2

for any real constants α 1, α 2

Based on the input-state-output variables, two types of

responses can now be defined

1 Zero Input
Response:
x(t0)
) 0
u(t) = 0, t ≥ 0 − −S−→ y , t
zi

t
2 ≥t
Zero State Response: )
x(t 0) = S
0 —− (t), t ≥ t0
u(t), t ≥ t0 −→ y zs

8
Linear Dynamical
Response of linear
systems

The additivity property implies

that:
)
output due to )
( u(t), t ≥ u(t) = 0, t ≥
= output due to
x (t 0 ) 0 0
(
x (t 0 ) (
u(t),
x (tt0≥
)=0
+ output due to
)
0

i.e.
,
y(t) = yz i (t) +
yz s (t)

9
Linear Dynamical
Zero-state response of linear
systems
Consider the SISO system.
Let δ ∆ (t − t1 ) be the pulse as shown in the figure, then
every input can be approximated by a sequence of the
pulses
64(t—ti)Δ u(t
u(t i )i )64 (t — t i )4

ti
ti ti + 4
(a) Pulse (b) Time-shifted (c) Step approximation
pulse

1
Linear Dynamical 0
Zero-state response of linear
systems
Consider the SISO system.
Let δ ∆ (t − t1 ) be the pulse as shown in the figure, then
every input can be approximated by a sequence of the
pulses
64(t—ti)Δ u(t
u(t i )i )64 (t — t i )4

ti
ti ti + 4
(a) Pulse (b) Time-shifted (c) Step approximation
pulse

The input can be expressed

symbolically as :
u(t) ≈ Σ u(ti )δ ∆ (t − i
t )∆ i

1
Linear Dynamical 0
Zero-state response of linear
systems
Let g∆ (t, ti ) be the output at time t excited by
the pulse
u(t) = δ∆ (t − ti ) applied at time ti then:
δ∆ (t − ti ) −−→ g ∆ (t, t i )
S

1
Linear Dynamical 1
Zero-state response of linear
systems
Let g∆ (t, ti ) be the output at time t excited by
the pulse
u(t) = δ∆ (t − ti ) applied at time ti then:
S
δ∆ (t − ti ) −−→ g ∆ (t, t i )
S
u(ti )δ ∆ (t − ti ) ∆ −−→ g∆ (t, i i (homogeneit
t )u(t ) ∆ y)

1
Linear Dynamical 1
Zero-state response of linear
systems
Let g∆ (t, ti ) be the output at time t excited by
the pulse
u(t) = δ∆ (t − ti ) applied at time ti then:
S
δ∆ (t − ti ) −−→ g ∆ (t, t i )
S
u(ti )δ ∆ (t − ti ) ∆ −−→ g∆ (t, i i (homogeneit
Σ t )u(t ) ∆ S
Σ y)
u(ti )δ (t − ti ) ∆ −−→ g∆ (t, i i (additivit
i t )u(t
∆ )∆ i y)
Thus
,
Σ
y(t) ≈ g∆ (t, i i
t )u(t ) ∆i

1
Linear Dynamical 1
Zero-state response of linear
systems
y(t) ≈ g∆ (t, i i
t )u(t )∆i Σ

1
Linear Dynamical 2
Zero-state response of linear
systems Σ
y(t) ≈ g∆ (t, i i
t )u(t )∆i

If ∆ approaches zero, then δ ∆ ( t − t i ) becomes an impulse at

t = t i i.e.

δ(t − t i ) and the corresponding output will be denoted by

g(t, t i )

1
Linear Dynamical 2
Zero-state response of linear
systems Σ
y(t) ≈ g∆ (t, i i
t )u(t )∆i

If ∆ approaches zero, then δ ∆ ( t − t i ) becomes an impulse at

t = t i i.e.

δ(t − t i ) and the corresponding output will be denoted by

g(t, t i )

As ∆ approaches zero,
∆ can be written as dτ
discrete t i becomes a continuous and can
be replaced by τ
summation becomes an
integration approximation
becomes an equality

1
Linear Dynamical 2
Zero-state response of linear
systems Σ
y(t) ≈ g∆ (t, i i
t )u(t )∆i

If ∆ approaches zero, then δ ∆ ( t − t i ) becomes an impulse at

t = t i i.e.

δ(t − t i ) and the corresponding output will be denoted by

g(t, t i )

As ∆ approaches zero,
∆ can be written as dτ
discrete tΣi becomes a continuous ∫ ∞and can
y(t)be= replaced
lim by g∆ τ(t, ti )u(ti ) ∆ g(t,
∆ →0
summation becomes an −∞
= i τ)u(τ)dτ
integration approximation
where t is the time at which the output is observed; τ is the time at which
becomes an equality
the impulse input is applied; and g(t, τ ) is the impulse response

1
The last equation is a consequence of the definition of the Riemann integral, i.e.
, −∞∞ Σ 1
f ( τ ) d τ =∆ →l 0i m i f ( k ∆ ) ∆ . It implicitly assumes that the limit and the integral 2
Linear Dynamical
Zero-state response of linear
systems
If a system is causal, the output will not appear before the input
is applied.
Thu
∫ t
s
Causal ⇐⇒ g(t, τ) = 0 for t < τ =⇒ y(t) = g(t,
τ)u(τ)dτ t0

1
Linear Dynamical 3
Zero-state response of linear
systems
If a system is causal, the output will not appear before the input
is applied.
Thu
∫ t
s valid for
Causal ⇐⇒ g(t, τ) = 0 for t < τ =⇒ y(t) = g(t, τ)u(τ)dτ
LTI and LTV
⇐= t0

1
Linear Dynamical 3
Zero-state response of linear
systems
If a system is causal, the output will not appear before the input
is applied.
Thu
∫ t
s valid for
Causal ⇐⇒ g(t, τ) = 0 for t < τ =⇒ y(t) = g(t, τ)u(τ)dτ
LTI and LTV
⇐= t0

Theorem (Impulse Response)

Consider a continuous-time linear system with m inputs and p
outputs. There exists a matrix-valued signal G(t, τ) ∈ R p × m
such that for every input u a corresponding output is given
by ∫ t
y(t) = G (t, τ)u(τ)dτ ∀t ≥
t0 t0

1
Linear Dynamical 3
Zero-state response of linear
systems
If a system is causal, the output will not appear before the input
is applied.
Thu
∫ t
s valid for
Causal ⇐⇒ g(t, τ) = 0 for t < τ =⇒ y(t) = g(t, τ)u(τ)dτ
LTI and LTV
⇐= t0

Theorem (Impulse Response)

Consider a continuous-time linear system with m inputs and p
outputs. There exists a matrix-valued signal G(t, τ) ∈ R p × m
such that for every input u a corresponding output is given
by ∫ t
y(t) = G (t, τ)u(τ)dτ ∀t ≥
t0 t0

If the system is time-invariant as well, then

G (t, τ) = G (t + T, τ + T ) = G (t − τ, 0) = G (t − τ) for
any T
∫ t
and assuming t 0 = 0
y(t) = G (t − τ)u(τ)dτ , (G * u)(t) ∀t ≥
0 0
where s denotes the convolution 13
operator.
Linear Dynamical
Zero-state response of linear
systems
For Discrete-time systems:
Theorem (Impulse Response)
Consider a discrete-time linear system with m inputs
and p outputs. There exists a matrix valued function
G(t, τ) ∈ R p×m such that for every input u a
corresponding output is given by
Σt
y(t) = G (t, τ)u(τ); ∀t ≥0t , t, τ
∈ N t0
If the system is time-invariant as well, then the
time-shifting property holds and assuming t0 = 0

Σt
y(t) = G (t − τ)u(τ)dτ , (G * u)(t) ∀t ∈ N
≥0
0
where s denotes the convolution
operator. 14
Linear Dynamical
Zero-state response of linear systems: Transfer
Function
Particularly, for computing the zero-state response of LTI systems,
frequency domain tools offers a great flexibility.

1
Linear Dynamical 5
Zero-state response of linear systems: Transfer
Function
Particularly, for computing the zero-state response of LTI systems,
frequency domain tools offers a great flexibility.
The continuous-time linear system has an output
∫ ∞
y(t) = G (t − τ)u(τ)dτ; ∀t ≥
0 0
Taking its Laplace transform, one
obtains ∫ ∞ ∫ ∞
yˆ(s) = e− s t G (t −
τ)u(τ)dτdt
0 0

1
Linear Dynamical 5
Zero-state response of linear systems: Transfer
Function
Particularly, for computing the zero-state response of LTI systems,
frequency domain tools offers a great flexibility.
The continuous-time linear system has an output
∫ ∞
y(t) = G (t − τ)u(τ)dτ; ∀t ≥
0 0
Taking its Laplace transform, one
obtains ∫ ∞ ∫ ∞
yˆ(s) = e− s t G (t −
τ)u(τ)dτdt
0 0
Changing the order of integration and rearranging
integrals, one gets∫ ∞ ∫ ∫ ∞ ,
−s ( t−τ )
yˆ(s) = G (t − τ)dt e− s τ (1
e 0 0 u(τ)dτ )

1
Linear Dynamical 5
Zero-state response of linear systems: Transfer
Function
Particularly, for computing the zero-state response of LTI systems,
frequency domain tools offers a great flexibility.
The continuous-time linear system has an output
∫ ∞
y(t) = G (t − τ)u(τ)dτ; ∀t ≥
0 0
Taking its Laplace transform, one
obtains ∫ ∞ ∫ ∞
yˆ(s) = e− s t G (t −
τ)u(τ)dτdt
0 0
Changing the order of integration and rearranging
integrals, one gets∫ ∞ ∫ ∫ ∞ ,
−s ( t−τ )
yˆ(s) = G (t − τ)dt e− s τ (1
e 0 0 u(τ)dτ )
But because of
∫ ∞
causality, ∫ ∞ ∫ ∞
¯ ¯
e− s ( t − τ ) G (t − τ)dt = e−st G (t)dt
¯ ¯ = e−st G (t)dt
¯ ¯= G
ˆ (s)
0 (2) − τ 0

1
Linear Dynamical 5
Zero-state response of linear systems: Transfer
Function
Particularly, for computing the zero-state response of LTI systems,
frequency domain tools offers a great flexibility.
The continuous-time linear system has an output
∫ ∞
y(t) = G (t − τ)u(τ)dτ; ∀t ≥
0 0
Taking its Laplace transform, one
obtains ∫ ∞ ∫ ∞
yˆ(s) = e− s t G (t −
τ)u(τ)dτdt
0 0
Changing the order of integration and rearranging
integrals, one gets∫ ∞ ∫ ∫ ∞ ,
−s ( t−τ )
yˆ(s) = G (t − τ)dt e− s τ (1
e 0 0 u(τ)dτ )
But because of
∫ ∞
causality, ∫ ∞ ∫ ∞
¯ ¯
e− s ( t − τ ) G (t − τ)dt = e−st G (t)dt
¯ ¯ = e−st G (t)dt
¯ ¯= G
ˆ (s)
0 (2) − τ 0

Substituting (2) into (1) and removing Gˆ(s) from the integral,
we conclude that
∫ ∫
0 ∞ 0 ∞
yˆ(s) = Gˆ(s)e−sτ u(τ)dτ = Gˆ(s) e − s τ u(τ)dτ = 1
5
Linear Dynamical
Zero-state response of linear systems: Transfer
Function
Definition (Transfer function)
The transfer function of a CT causal LTI system is the
Laplace transform
∫ ∞
Gˆ (s) = L [G (t)] = e−st G (t)dt, s ∈
C 0

of an impulse response G (t2 , t1 ) = G (t2 − t1 ), ∀t2 ≥ t1 ≥ 0.

Definition (Transfer function)

The transfer function of a DT causal LTI system is the Z-
transform
Σ∞
ˆ
G (z) = Z[G (t)] = z−t G (t), z ∈
C t=0

of an impulse response G (t2 , t1 ) = G (t2 − t1 ), ∀t2 ≥ t1

≥ 0. 16
Linear Dynamical
State-space
systems
State-space representation of linear systems
Using the state variable, as introduced earlier, a continuous-
time
state-space linear system is represented by the following two
equations: x˙(t) = A (t)x(t) +
B (t)u(t) , LTV
y(t) = C (t)x(t) + )
(3)
D (t)u(t)
where

u : [0, ∞) → Rm , x : [0, ∞) → Rn , y : [0, ∞) → Rp

are called the input, state, and output signals of the

system and the time-varying matrices (A, B , C , D)(t) are of
appropriate dimensions.

1
Linear Dynamical 7
State-space
systems
State-space representation of linear systems
Using the state variable, as introduced earlier, a continuous-
time
state-space linear system is represented by the following two
equations: x˙(t) = A (t)x(t) +
B (t)u(t) , LTV
y(t) = C (t)x(t) + )
(3)
D (t)u(t)
where

u : [0, ∞) → Rm , x : [0, ∞) → Rn , y : [0, ∞) → Rp

are called the input, state, and output signals of the

system and the time-varying matrices (A, B , C , D)(t) are of
appropriate dimensions.
Note:
The first equation of (3) is called the state equation and the
second equation of (3) is called the output equation.
when all the matrices (A, B , C , D)(t) are constant ∀t ≥
0, the system is LTI 1
7
Linear Dynamical
Interconnectio
ns Interconnections of block diagrams are especially
useful to highlight special structures in state-
space equations.

Figure: Negative feedback

interconnection

Given P 1 : z ›→ y x˙ = A 1 x + B 1 z, y1 = C 1 x +
D 1z
Compute S : u ›→ y

1
Linear Dynamical 8
Interconnectio
ns Interconnections of block diagrams are especially
useful to highlight special structures in state-
space equations.

Figure: Negative feedback

interconnection
Given P 1 : z ›→ y x˙ = A 1 x + y1 = C 1 x + D 1 z
B 1 z,
Compute S : u ›→ y + B 1 (I − (I +
x˙ = (A 1 − B 1 (I + D 1 )−1 D 1 )u
D 1 )−1 C 1 )x y———-
= +
Show By Yourself! (I + D 1 )−1 D 1 u
1
———-
(I + D )−1 C x Linear Dynamical 8
Impulse Response and Transfer function for LTI
system
Consider the continuous-time LTI system
x˙ = Ax + B u,

y = C x + D u,

1
Linear Dynamical 9
Impulse Response and Transfer function for LTI
system
Consider the continuous-time LTI system
x˙ = Ax + B u, y = Cx +
D u,
Taking the Laplace transform of both sides, we obtain
sxˆ(s) − x(0) = A xˆ(s) + B uˆ(s), yˆ(s) = C xˆ(s) +
D uˆ(s)

1
Linear Dynamical 9
Impulse Response and Transfer function for LTI
system
Consider the continuous-time LTI system
x˙ = Ax + B u, y = Cx +
D u,
Taking the Laplace transform of both sides, we obtain
sxˆ(s) − x(0) = A xˆ(s) + B uˆ(s), yˆ(s) = C xˆ(s) +
D uˆ(s)
Solving for xˆ(s), we obtain
xˆ(s) = (sI − A )−1 B uˆ(s) + (sI − A )−1 x(0)

1
Linear Dynamical 9
Impulse Response and Transfer function for LTI
system
Consider the continuous-time LTI system
x˙ = Ax + B u, y = Cx +
D u,
Taking the Laplace transform of both sides, we obtain
sxˆ(s) − x(0) = A xˆ(s) + B uˆ(s), yˆ(s) = C xˆ(s)
+ D uˆ(s)
Solving for xˆ(s), we obtain
xˆ(s) = (sI − A )−1 B uˆ(s) + (sI − A )−1 x(0)
Ψˆ (s) = C (sI − A )
from =
yˆ(s) Ψ(ˆ we
which ˆ(s)uˆ(s)
conclude
s)x(0)+G that
Gˆ(s) = C (sI − A )−1 B +
−1

D
where

1
Linear Dynamical 9
Impulse Response and Transfer function for LTI
system
Consider the continuous-time LTI system
x˙ = Ax + B u, y = Cx +
D u,
Taking the Laplace transform of both sides, we obtain
sxˆ(s) − x(0) = A xˆ(s) + B uˆ(s), yˆ(s) = C xˆ(s)
+ D uˆ(s)
Solving for xˆ(s), we obtain
xˆ(s) = (sI − A )−1 B uˆ(s) + (sI − A )−1 x(0)
Ψˆ (s) = C (sI − A )
from =
yˆ(s) Ψ(ˆ we
which ˆ(s)uˆ(s)
conclude
s)x(0)+G that
Gˆ(s) = C (sI − A )−1 B +
−1

Coming back to the time domain by D applying

where
inverse Laplace transforms, we obtain

y(t) = Ψ(t)x(0) + (G * u)(t) G(t) = L−1[Gˆ(s)],

where
Ψ(t) = L −1 [Ψˆ 1
Linear Dynamical 9
Impulse Response and Transfer function for LTI
system
Theorem (In continuous-time domain)
The impulse response and transfer function of the CLTI system are given
by: −1 −1
G (t) = L C (sI − A ) B + D and Gˆ (s) = C (sI − A )− 1 B +
D
respectively Moreover, the response y(t) = (G * u)(t) corresponds to the zero
initial condition x(0) = 0.

2
Linear Dynamical 0
Impulse Response and Transfer function for LTI
system
Theorem (In continuous-time domain)
The impulse response and transfer function of the CLTI system are given
by: −1 −1
G (t) = L C (sI − A ) B + D and Gˆ (s) = C (sI − A )− 1 B +
D
respectively Moreover, the response y(t) = (G * u)(t) corresponds to the zero
initial condition x(0) = 0.

Consider the discrete-time LTI system

x+ = A x + y = Cx +
B u, Du
Theorem (In discrete-time domain)
The impulse response and transfer function of the DLTI system are given
by: −1 −1
G (t) = Z C (zI − A ) B + D and Gˆ (z) = C (zI − A )− 1 B +
D
respectively Moreover, the response y(t) = (G * u)(t) corresponds to the zero
initial condition x(0) = 0.
2
Linear Dynamical 0
Impulse Response and Transfer
function

Laplace transforms can be used for solving the LTI state-

space systems, however for time-varying linear systems,
this tool cannot be used

2
Linear Dynamical 1
Impulse Response and Transfer
function

Laplace transforms can be used for solving the LTI state-

space systems, however for time-varying linear systems,
this tool cannot be used
1 The Laplace transform of G(t, τ) is a function of two
2 variables
L [A(t)x(t)] =/ L[A(t)]L[x(t)]

2
Linear Dynamical 1
Impulse Response and Transfer
function

Laplace transforms can be used for solving the LTI state-

space systems, however for time-varying linear systems,
this tool cannot be used
1 The Laplace transform of G(t, τ) is a function of two
2 variables
L [A(t)x(t)] =/ L[A(t)]L[x(t)]
First we will see the solution of LTV systems and then
tailor it for LTI systems

2
Linear Dynamical 1
Week 1 -
Lecture 2

In the last lecture, we discussed

Key properties of dynamical systems and the
physical significance of the state
Zero-state response of linear (TI and TV) systems in
(CT and DT ) -domain
Zero-state response of LTI system in frequency
domain and its relation with the state-space
representation

22
Linear Dynamical
Solution to homogeneous LTV
systems
We start by considering the solution to a CTLTV
system with a given initial condition but zero input
x˙(t) = A(t)x(t); x(t 0 ) = x 0 ∈ Rn; t
≥0 (4)

2
Linear Dynamical 3
Solution to homogeneous LTV
systems
We start by considering the solution to a CTLTV
system with a given initial condition but zero input
x˙(t) = A(t)x(t); x(t 0 ) = x 0 ∈ Rn;
t≥0 (4)
A key property of homogeneous systems is that the map from
the initial condition x(t0) = x0 ∈ R n to the solution x(t) ∈ R n
at a given time
t ≥ 0 is always linear.

2
Linear Dynamical 3
Solution to homogeneous LTV
systems
We start by considering the solution to a CTLTV
system with a given initial condition but zero input
x˙(t) = A(t)x(t); x(t 0 ) = x 0 ∈ Rn; t
≥0 (4)
Acondition x(t0) =
key property of xhomogeneous
0 ∈ R
n to the solution x(t) ∈ R n at a
systems is that the map
tfrom
≥ 0 is
giventhe always
time
initial linear.
Theorem (Peano-Baker
The unique solution to (4) is given by x(t) = φ(t, t 0 )x0 , x0 ∈ R n , t ≥
Series)
0
where
∫ t ∫ t ∫ τ1
φ(t, t0) = I + A(τ1)dτ1 + A(τ1) A(τ2)dτ2
dτ1+ ∫ t t0
∫ τ1 ∫ τ2
t0 t0

A (τ1) A (τ2)
A (τ3)dτ3dτ2dτ1 + . . .
t0 t0 t0
2
The n × n matrix φ(t, t0) is called theLinear
stateDynamical
transition matrix. 3
Solution to homogeneous LTV
systems
Properties of the state-transition
matrix
1 For every t ≥ 0, φ(t, t ) is the unique
0 0
solution to
d φ(t, t0) = A (t)φ(t, t0) φ(t0, t0) t≥
= I,
dt 0.

2
Linear Dynamical 4
Solution to homogeneous LTV
systems
Properties of the state-transition
matrix
1 For every t ≥ 0, φ(t, t ) is the unique
0 0
solution to
d φ(t, t0) = A (t)φ(t, t0) φ(t0, t0) t≥
= I,
dt 0.

2 For every t, s, τ ≥
0,
φ(t, s)φ(s, τ) = φ(t,
τ)

x 1 = $(s, ⌧)x 0
x 2 = $(t, ⌧)x 0 = $(t,
x0
s)x 1
t

2
Linear Dynamical 4
Solution to homogeneous LTV
systems
Properties of the state-transition
matrix
1 For every t ≥ 0, φ(t, t ) is the unique
0 0
solution to
d φ(t, t0) = A (t)φ(t, t0) φ(t0, t0) t≥
= I,
dt 0.

2 For every t, s, τ ≥
0,
φ(t, s)φ(s, τ) = φ(t,
τ)

x 1 = $(s, ⌧)x 0
x 2 = $(t, ⌧)x 0 = $(t,
x0
s)x 1
t

3 For every t, τ ≥ 0, φ(t, τ) is non-singular and φ(t, τ) −1 =

φ(τ, t) 2
Linear Dynamical 4
Computation of φ(t,
t0) Consid
er x˙ = (5
A (t)x )
where A ∈ R n × n is a continuous function, then for every
initial state x i (t 0 ) ∈ R n , there exists a unique solution
x i (t) ∈ R n for i = 1, 2, 3, . . . , n

2
Linear Dynamical 5
Computation of φ(t,
t0) Consid
er x˙ = (5
A (t)x )
where A ∈ R n × n is a continuous function, then for every
initial state x i (t 0 ) ∈ R n , there exists a unique solution
x i (t) ∈ R n for i = 1, 2, 3, . . . , n
1 Arrange these n solutions as X 1 x 2 . . . xn a
= x
matrix square(5),
of order n. Because every x i satisfies
we have

X˙ (t) = A (t)X (t)

2
Linear Dynamical 5
Computation of φ(t,
t0) Consid
er x˙ = (5
A (t)x )
where A ∈ R n × n is a continuous function, then for every
initial state x i (t 0 ) ∈ R n , there exists a unique solution
x i (t) ∈ R n for i = 1, 2, 3, . . . , n
1 Arrange these n solutions as X 1 x 2 . . . xn a
= x
matrix square(5),
of order n. Because every x i satisfies
we have

X˙ (t) = A (t)X (t)

2 If X(t 0 ) is non-singular or the n initial states are
linearly independent, then X(t) is called a
fundamental matrix of (5)

2
Linear Dynamical 5
Computation of φ(t,
t0) Consid
er x˙ = (5
A (t)x )
where A ∈ R n × n is a continuous function, then for every
initial state x i (t 0 ) ∈ R n , there exists a unique solution
x i (t) ∈ R n for i = 1, 2, 3, . . . , n
1 Arrange these n solutions as X 1 x 2 . . . xn a
= x
matrix square(5),
of order n. Because every x i satisfies
we have

X˙ (t) = A (t)X (t)

2 If X(t 0 ) is non-singular or the n initial states are
linearly independent, then X(t) is called a
fundamental matrix of (5)
Question
1 Is X (t) unique?

2 Is X(t) non-singular for all 2

t? Linear Dynamical 5
Computation of φ(t,
t0) Example: Consider the homogeneous
equation
0
x˙1 x(t)
0
=
o t
r x˙1(t) = 0, 0 x˙2 = tx 1 (t)

2
Linear Dynamical 6
Computation of φ(t,
t0) Example: Consider the homogeneous
equation
0
x˙1 x(t)
0
=
o t
r x˙1(t) = 0, 0

x˙2 = tx 1 (t)

the solution of x˙1(t) = 0 for t 0 = 0 is x 1 (t) = x 1 (0);

2
Linear Dynamical 6
Computation of φ(t,
t0) Example: Consider the homogeneous
equation
0
x˙1 x(t)
0
=
o t
r x˙1(t) = 0, 0

∫ x˙2 = tx 1 (t)
t
2
x 2 (t)of=x˙1(t)
the solution τ x (0)dτ + x (0) = 0. 5t x 1 (0) + x 2 (0)
0 = 01 for t 0 = 0 2is x 1 (t) = x 1 (0);

the solution of x˙2 = t x 1 = tx 1 (0) is

2
Linear Dynamical 6
Computation of φ(t,
t0) Example: Consider the homogeneous
equation
0
x˙1 x(t)
0
=
o t
r x˙1(t) = 0, 0

∫ x˙2 = tx 1 (t)
t
2
x 2 (t)of=x˙1(t)
the solution τ x (0)dτ + x (0) = 0. 5t x 1 (0) + x 2 (0)
0 = 01 for t 0 = 0 2is x 1 (t) = x 1 (0);
Choos
e the solution of x˙2 = t x 1 = tx 1 (0) is
x 1 (0)
x 1 (0) = =
1 0
x 2 (0)
an
d x 1 (0) 1
x 2 (0) = =
x 2 (0)
The two initial states are linearly
2
independent.
2
Linear Dynamical 6
Computation of φ(t,
t0) Example: Consider the homogeneous
equation
0
x˙1 x(t)
0
=
o t
r x˙1(t) = 0, 0

∫ x˙2 = tx 1 (t)
t
2
x 2 (t)of=x˙1(t)
the solution τ x (0)dτ + x (0) = 0. 5t x 1 (0) + x 2 (0)
0 = 01 for t 0 = 0 2is x 1 (t) = x 1 (0);
Thu
s the solution of x˙2 = t x 1 = tx 1 (0) is
x 1 (0) 1
x 1 (0) = =
1 0 0.5t2 = x 1 (t)
x 2 (0) =⇒
an
d x 1 (0) 1 1
x 2 (0) = =
0.5t2 + = x 2 (t)
x 2 (0) =⇒
The two initial states are linearly 2
2
independent.
2
Linear Dynamical 6
Computation of φ(t,
t0) Example: Consider the homogeneous
equation
0
x˙1 x(t)
0
=
o t
r x˙1(t) = 0, 0

∫ x˙2 = tx 1 (t)
t
2
x 2 (t)of=x˙1(t)
the solution τ x (0)dτ + x (0) = 0. 5t x 1 (0) + x 2 (0)
0 = 01 for t 0 = 0 2is x 1 (t) = x 1 (0);
Thu
s the solution of x˙2 = t x 1 = tx 1 (0) is
x 1 (0) 1
x 1 (0) = =
1 0 0.5t2 = x 1 (t)
x 2 (0) =⇒
an
d x 1 (0) 1 1
x 2 (0) = =
0.5t2 + = x 2 (t)
x 2 (0) =⇒
The two initial states are linearly 2
2
independent. Thus 1
X (t) =
1 2
0.5t 0.5t2 + 2
2 Linear Dynamical 6
Computation of φ(t,
t0)
Theorem
Let X (t) be a fundamental matrix of x˙ = A (t)x. Then

φ(t, t0 ) = X (t)X −1
(t0 ).

Because X(t) is non-singular for all t, its inverse is well

defined

2
Linear Dynamical 7
Computation of φ(t,
t0)
Theorem
Let X (t) be a fundamental matrix of x˙ = A (t)x. Then

φ(t, t0 ) = X (t)X −1
(t0 ).

Because X(t) is non-singular for all t, its inverse is well

defined Revisit the last example:

1 1
X (t)
= 0.5t2 0.5t2 +
2
The state-transition matrix is
given by
1
20 − 2t )
0.5(t
φ(t, t0 ) = 0
1
Verify the three earlier listed properties of 2
φ(t, t0 ). Linear Dynamical 7
Solution of non-homogeneous LTV
systems
We now go back to the original non-homogeneous LTV
system
x˙ = A (t)x +
B (t)u x(t0 ) = x0 ∈ Rn , t ≥ (6
, 0 )
y = C (t)x +
D (t)u

2
Linear Dynamical 8
Solution of non-homogeneous LTV
systems
We now go back to the original non-homogeneous LTV
system
x˙ = A (t)x +
B (t)u x(t0 ) = x0 ∈ Rn , t ≥ (6
, 0 )
y = C (t)x +
D (t)u
Theorem (Variation of constants)
The unique solution to (6) is given
by ∫ t
(7
x (t) = φ(t , t 0 )x 0 + φ(t , τ )B (τ )u(τ )dτ
t0
∫ )
t
(8
y(t) = C (t)φ(t , t 0 )x 0 + C (t) φ(t , τ )B (τ )u(τ )dτ + D (t)u(t)
t0 )

where φ(t, t 0 ) is the state-transition matrix.

2
Linear Dynamical 8
Solution of non-homogeneous LTV
systems
We now go back to the original non-homogeneous LTV
system
x˙ = A (t)x +
B (t)u x(t0 ) = x0 ∈ Rn , t ≥ (6
, 0 )
y = C (t)x +
D (t)u
Theorem (Variation of constants)
The unique solution to (6) is given
by ∫ t
(7
x (t) = φ(t , t 0 )x 0 + φ(t , τ )B (τ )u(τ )dτ
t0
∫ )
t
(8
y(t) = C (t)φ(t , t 0 )x 0 + C (t) φ(t , τ )B (τ )u(τ )dτ + D (t)u(t)
t0 )

where φ(t, t 0 ) is the state-transition matrix.

Equation (7) is known as the variation of constants

formula. Homogeneous response: y z i¸(t) ≡ y h (t) =
t
Forced response:z s y (t) ≡f y (t) = C (t) φ(t, τ)B (τ)u(τ)dτ +
C(t)φ(t, t 0 )x 0 t0
D (t)u(t)
2
Linear Dynamical 8
Solution of non-homogeneous LTV
x˙ = A (t)x + B (t)u
systems
n
, 0 0 ≥0
y = C (t)x + x(t ) = x 2 R , t (6)
D (t)u
Zt
x ( t ) = $ ( t , t 0) x 0 + $ ( t , ⌧)B(⌧)u(⌧)d⌧
t0
Z t
(7)
y ( t ) = C ( t ) $ ( t , t 0) x 0 + C ( t ) $ ( t , ⌧)B(⌧)u(⌧)d⌧ + D( t ) u ( t ) (8)
t0

2
Linear Dynamical 9
Solution of non-homogeneous LTV
x˙ = A (t)x + B (t)u
systems
n
, 0 0 ≥0
y = C (t)x + x(t ) = x 2 R , t (6)
D (t)u
Zt
x ( t ) = $ ( t , t 0) x 0 + $ ( t , ⌧)B(⌧)u(⌧)d⌧
t0
Z t
(7)
y ( t ) = C ( t ) $ ( t , t 0) x 0 + C ( t ) $ ( t , ⌧)B(⌧)u(⌧)d⌧ + D( t ) u ( t ) (8)
t0

Proof
To verify (7) is a solution to (6), note that at t = t0, the
integral in (7) disappears and we get x(t0) = x0.

2
Linear Dynamical 9
Solution of non-homogeneous LTV
x˙ = A (t)x + B (t)u
systems
n
, 0 0 ≥0
y = C (t)x + x(t ) = x 2 R , t (6)
D (t)u
Zt
x ( t ) = $ ( t , t 0) x 0 + $ ( t , ⌧)B(⌧)u(⌧)d⌧
t0
Z t
(7)
y ( t ) = C ( t ) $ ( t , t 0) x 0 + C ( t ) $ ( t , ⌧)B(⌧)u(⌧)d⌧ + D( t ) u ( t ) (8)
t0

Proof
To verify (7) is a solution to (6), note that at t = t0, the
integral in (7) disappears and we get x(t0) = x0.
Taking the derivative of (7), we
obtain ∫ t
dφ(t, t0) dφ(t, τ)
x˙ x 0 + φ(t, t)B (t)u B(τ)u(τ)d
dt dt
= + t0
∫ t τ
= A (t)φ(t, t0)x 0 + B (t)u(t) + A (t) φ(t,
τ)B (τ)u(τ)dτ t0
= A (t)x(t) + B (t)u(t)

which shows that (7) is indeed a solution

to (6).
2
Linear Dynamical 9
Solution of non-homogeneous LTV
x˙ = A (t)x + B (t)u
systems
n
, 0 0 ≥0
y = C (t)x + x(t ) = x 2 R , t (6)
D (t)u
Zt
x ( t ) = $ ( t , t 0) x 0 + $ ( t , ⌧)B(⌧)u(⌧)d⌧
t0
Z t
(7)
y ( t ) = C ( t ) $ ( t , t 0) x 0 + C ( t ) $ ( t , ⌧)B(⌧)u(⌧)d⌧ + D( t ) u ( t ) (8)
t0

Proof
To verify (7) is a solution to (6), note that at t = t0, the
integral in (7) disappears and we get x(t0) = x0.
Taking the derivative of (7), we
obtain ∫ t
dφ(t, t0) dφ(t, τ)
x˙ x 0 + φ(t, t)B (t)u B(τ)u(τ)d
dt dt
= + t0
∫ t τ
= A (t)φ(t, t0)x 0 + B (t)u(t) + A (t) φ(t,
τ)B (τ)u(τ)dτ t0
= A (t)x(t) + B (t)u(t)

which shows that (7) is indeed a solution to (6).

(8) is obtained by direct substitution of x(t) in y(t) = C(t)x + 2

D(t)u. Linear Dynamical 9
Solution of non-homogeneous LTV systems:
Facts
Relation between input-output and state-space
descriptions: The zero-state response is given as
∫ t
yz s (t) = C (t) φ(t, τ)B (τ)u(τ)dτ +
D (t)u(t) t 0

3
Linear Dynamical 0
Solution of non-homogeneous LTV systems:
Facts
Relation between input-output and state-space
descriptions: The zero-state response is given as
∫ t
yz s (t) = C (t) φ(t, τ)B (τ)u(τ)dτ +
D (t)u(t) t 0

which can also be

written as ∫ t
yz s (t) = [C (t)φ(t, τ)B (τ) + D (t)δ(t − τ)]
u(τ)dτ t 0

3
Linear Dynamical 0
Solution of non-homogeneous LTV systems:
Facts
Relation between input-output and state-space
descriptions: The zero-state response is given as
∫ t
yz s (t) = C (t) φ(t, τ)B (τ)u(τ)dτ +
D (t)u(t) t 0

which can also be

written as ∫ t
yz s (t) = [C (t)φ(t, τ)B (τ) + D (t)δ(t − τ)]
u(τ)dτ t 0
and is equivalent
to ∫ t
yz s (t) = G (t,
τ)u(τ)dτ t 0
implying
G(t, τ) , C(t)φ(t, τ)B(τ) + D(t)δ(t −
τ) 3
Linear Dynamical 0
Solution of non-homogeneous LTV systems:
Facts
Computing φ(t, t0 ) is generally difficult

3
Linear Dynamical 1
Solution of non-homogeneous LTV systems:
Facts
Computing φ(t, t0 ) is generally
difficult
Recall that the solution ∫ t
x(t) = φ(t, t0)x 0 + φ(t,
τ)B (τ)u(τ)dτ t0

hinge on
solving
d φ(t, t0) = A (t)φ(t,
t0)
dt

3
Linear Dynamical 1
Solution of non-homogeneous LTV systems:
Facts
Computing φ(t, t0 ) is generally
difficult
Recall that the solution ∫ t
x(t) = φ(t, t0)x 0 + φ(t,
τ)B (τ)u(τ)dτ t0

hinge on
solving
d φ(t, t0) = A (t)φ(t, t0)
dt
If A(t) is triangular such as
x˙1 a (t) 0 x (t)
= 11
(t)2(t
x˙ a21
1 (t) a22(t) x2(t)
)

3
Linear Dynamical 1
Solution of non-homogeneous LTV systems:
Facts
Computing φ(t, t0 ) is generally
difficult
Recall that the solution ∫ t
x(t) = φ(t, t0)x 0 + φ(t,
τ)B (τ)u(τ)dτ t0

hinge on
solving
d φ(t, t0) = A (t)φ(t, t0)
dt
If A(t) is triangular such as
x˙1 a (t) 0 x (t)
= 11
(t)2(t)
x˙ a121(t) a22(t) x2(t)

we can solve the scalar equation x˙1(t) = a11(t) and then

substitute it into

x˙2(t) = a 22(t)x 2(t) + a 21(t)x 1(t)

3
Linear Dynamical 1
Solution of non-homogeneous LTV systems:
Facts
Computing φ(t, t0 ) is generally
difficult
Recall that the solution ∫ t
x(t) = φ(t, t0)x 0 + φ(t,
τ)B (τ)u(τ)dτ t0

hinge on
solving
d φ(t, t0) = A (t)φ(t, t0)
dt
If A(t) is triangular such as
x˙1 a (t) 0 x (t)
= 11
(t)2(t)
x˙ a121(t) a22(t) x2(t)

we can solve the scalar equation x˙1(t) = a11(t) and then

substitute it into

x˙2(t) = a 22(t)x 2(t) + a 21(t)x 1(t)

Because x1(t) has been solved, the preceding scalar 3

equation can be solved for x2(t). This is what we did in the
Linear Dynamical 1
Solution of homogeneous DTLTV
systems

x(t + 1) = A (t)x(t), x(t0 ) = x 0 ∈ t∈

Rn , N

3
Linear Dynamical 2
Solution of homogeneous DTLTV
systems
The (unique) solution to the homogeneous discrete-
time linear time-varying system

x(t + 1) = A (t)x(t), x(t0 ) = x 0 ∈ Rn , t∈

N

is given by

x(t) = φ(t, t0 )x 0 , x(t0 ) = x 0 ∈ Rn , t ≥ t0

3
Linear Dynamical 2
Solution of homogeneous DTLTV
systems
The (unique) solution to the homogeneous discrete-
time linear time-varying system

x(t + 1) = A (t)x(t), x(t0 ) = x 0 ∈ Rn , t∈N

is given by

x(t) = φ(t, t0 )x 0 , x(t0 ) = x 0 ∈ Rn , t ≥ t0

where
I, for t = 0
φ(t, t0 )
t (t−1)A (t−2) . . . A (t0 + 1)A (t0 ), for t >
A
=
t0
is called the (discrete-time) state transition
matrix
3
Linear Dynamical 2
Solution of homogeneous DTLTV
systems
Note
Since the state equation is algebraic, it can be computed
recursively for a given initial state.

3
Linear Dynamical 3
Solution of homogeneous DTLTV
systems
Note
Since the state equation is algebraic, it can be computed
recursively for a given initial state.
Because the fundamental matrix in the CT case is non-
singular for all t, φ(t, t0) is defined for t ≥ t0 and t < t0.

3
Linear Dynamical 3
Solution of homogeneous DTLTV
systems
Note
Since the state equation is algebraic, it can be computed
recursively for a given initial state.
Because the fundamental matrix in the CT case is non-
singular for all t, φ(t, t0) is defined for t ≥ t0 and t < t0.
In the DT case, the A−matrix can be singular. Thus the
inverse of φ(t, t0) may not be defined. Consequently, φ(t,
t0) is defined only for t ≥ t0.

3
Linear Dynamical 3
Solution of homogeneous DTLTV
systems
Note
Since the state equation is algebraic, it can be computed
recursively for a given initial state.
Because the fundamental matrix in the CT case is non-
singular for all t, φ(t, t0) is defined for t ≥ t0 and t < t0.
In the DT case, the A−matrix can be singular. Thus the
inverse of φ(t, t0) may not be defined. Consequently, φ(t,
t0) is defined only for t ≥ t0.
Properties of φ(t, t0 )
1 For every t0 ≥ 0, φ(t, t0) is the unique
solution to t ≥ t0
2
φ(t + 1, t0) = A (t)φ(t, t0), φ(t0,
t0) = I ,
For every t ≥ s ≥ τ ≥ 0,
3 φ(t, t0) may be
φ(t, s)φ(s, τ) = φ(t, τ) 3
singular Linear Dynamical 3
Solution of non-homogeneous DTLTV
systems
Theorem (Variation of constants)
The unique solution to

x(t + 1) = A (t)x(t) + B (t)u(t)

y(t) = C (t)x(t) + D (t)u(t)

with x(t0) = x 0 ∈ Rn , t ∈ N is given by

t− 1
Σ
x(t) = φ(t, t0 )x + φ(t, τ + ∀t ≥ t0
1)B (τ)u(τ),0 τ =t0
t− 1
Σ
y(t) = C (t)φ(t, t0 )x + C (t) φ(t, τ + 1)B (τ)u(τ) + D (t)u(t), ∀t 0
≥t 0
τ =t0

where φ(t, t0) is the discrete-time state transition matrix.

———-Show by yourself
———- 34
Linear Dynamical
Solution of non-homogeneous LTV systems:
Facts
Relation between input-output and state-space
descriptions: The zero-state response is given as
t−1
Σ
yz s (t) = C (t) φ(t, τ + 1)B (τ)u(τ) + D (t)u(t), ∀t 0
≥t τ = t0

3
Linear Dynamical 5
Solution of non-homogeneous LTV systems:
Facts
Relation between input-output and state-space
descriptions: The zero-state response is given as
t−1
Σ
yz s (t) = C (t) φ(t, τ + 1)B (τ)u(τ) + D (t)u(t), ∀t 0
≥t τ = t0

If we define φ(t, τ) = 0 for t < τ ,

then
Σt
yz s (t) = [C (t)φ(t, τ + 1)B (τ) + D (τ)δ(t − τ)]
u(τ) τ = t0

where the impulse sequence δ(t − τ) equals 1 if t = τ

and 0 if
t /= τ .

3
Linear Dynamical 5
Solution of non-homogeneous LTV systems:
Facts
Relation between input-output and state-space
descriptions: The zero-state response is given as
t−1
Σ
yz s (t) = C (t) φ(t, τ + 1)B (τ)u(τ) + D (t)u(t), ∀t 0
≥t τ = t0

If we define φ(t, τ) = 0 for t < τ ,

then
Σt
yz s (t) = [C (t)φ(t, τ + 1)B (τ) + D (τ)δ(t − τ)]
u(τ) τ = t0

where the impulse sequence δ(t − τ) equals 1 if t = τ

and 0 if
t /= τ . Therefore,

G (t, τ) , C (t)φ(t, τ + 1)B (τ) + D (τ)δ(t − τ)

for t ≥ 3
τ. Linear Dynamical 5
Week 1 -
Lecture 3

In the last lecture, we discussed

solution of LTV state-space system in
CT solution of LTV state-space
system in DT properties and
implications of these solution

36
Linear Dynamical
Solution to LTI systems:
Homogeneous case

x˙ = Ax, x(t0) = x 0 ∈ t≥ (9)

Rn , 0

3
Linear Dynamical 7
Solution to LTI systems:
Homogeneous case
By applying the earlier results to the homogeneous time-
invariant
systems x˙ = Ax, t≥ (9)
x(t0) = x 0 ∈ Rn , 0
we have the following result.
Theorem (Peano-Baker Series)
The unique solution to (9) is given by x ( t ) = φ ( t , t 0 ) x 0 , x 0 ∈ R n , t ≥ 0
where
∫ t ∫ t ∫ τ ! ∫ t ∫ τ ∫ τ
1 1 2
φ ( t , t 0) = I + Adτ1 + A A d τ2 d τ 1 + A A A d τ3 d τ 2d τ 1 + . . .
t0 t0 t0 t0 t0 t0

The n × n matrix φ( t , t 0 ) is called the state transition

matrix.

3
Linear Dynamical 7
Solution to LTI systems:
Homogeneous case
By applying the earlier results to the homogeneous time-
invariant
systems x˙ = Ax, t≥ (9)
x(t0) = x 0 ∈ Rn , 0
we have the following result.
Theorem (Peano-Baker Series)
The unique solution to (9) is given by x ( t ) = φ ( t , t 0 ) x 0 , x 0 ∈ R n , t ≥ 0
where
∫ t ∫ t ∫ τ ! ∫ t ∫ τ ∫ τ
1 1 2
φ ( t , t 0) = I + Adτ1 + A A d τ2 d τ 1 + A A A d τ3 d τ 2d τ 1 + . . .
t0 t0 t0 t0 t0 t0

The n × n matrix φ( t , t 0 ) is called the state transition

matrix.

∫ t∫
Since τ1 ∫ τk− ∫ τ
2
(t − t0)k k
·· k− 1
A k dτk k− 1· · · dτ
2 dτ A
k
t0 t0 · t0 t0 dτ 1 =
we conclude !
that Σ∞ (t − t0 )k
φ(t, t0 ) A k.
k
= k =0 3
!
Linear Dynamical 7
Solution to LTI systems:
Homogeneous case

Σ∞ (t − t0 )k
φ(t, t0 ) Ak (10
k
= k= 0 )
!

3
Linear Dynamical 8
Solution to LTI systems:
Homogeneous case

Σ∞ (t − t0 )k
φ(t, t0 ) Ak (10
k
= k= 0 )
!

Define the matrix exponential of a given n × n

matrix M by
Σ∞ 1
M
e Mk
k
= k= 0
!

3
Linear Dynamical 8
Solution to LTI systems:
Homogeneous case

Σ∞ (t − t0 )k
φ(t, t0 ) Ak (10
k
= k= 0 )
!

Define the matrix exponential of a given n × n

matrix M by
Σ∞ 1
M
e Mk
k
= k= 0
!
which allows us to rewrite (10) as

φ(t, t0 ) = eA ( t−t 0 )

3
Linear Dynamical 8
Solution to LTI systems: Non-
homogeneous case

x˙ = Ax +
Bu x(t0 ) = x 0 ∈ Rn , t ≥ t0
,
y = Cx +
Du

3
Linear Dynamical 9
Solution to LTI systems: Non-
homogeneous case

From the variation of constants formula, the

solution to
x˙ = Ax +
Bu x(t0 ) = x 0 ∈ Rn , t ≥0
,
y = Cx + t
is given Du
by ∫t
A(t−t 0) A ( t−τ )
x(t) = e x0 + e B(τ)u(τ)dτ
t0
∫ t ,
A(t−t 0) A ( t−τ )
y(t) = Ce x0 + Ce B(τ)u(τ)d
t0 τ

3
Linear Dynamical 9
Properties of the matrix
exponential
1 The function e At is the unique
solution tod
e At = Ae A t , eA . 0 = t≥
dt I ,
0
2 For every t, τ ∈
R,
eA t eA τ = eA ( t+ τ )

In general, =/ e ( A + B ) t
3 eFor
At eB t
every t ∈ R, eA t is nonsingular
and
−1
e At = e −At
4 For every n × n
matrix A ,
Ae A t = e A t A, ∀t ∈
R 40
Linear Dynamical
Computation of matrix
exponential

e At is uniquely
d defined by
e At = Ae A t , eA . 0 = I , t≥
dt
0

4
Linear Dynamical 1
Computation of matrix
exponential
e At is uniquely defined
by d
eA t = Ae A t , eA . 0 = I , t
dt ≥ 0
Taking the Laplace Transform, we
conclude that
d At At
L e =L
dt
Ae
se^At − e At . t= 0 = Ae^At

(sI − A )e^A t = I

e^A t = (sI − A )
−1

4
Linear Dynamical 1
Computation of matrix
exponential
e At is uniquely defined
by d
eA t = Ae A t , eA . 0 = I , t
dt ≥ 0
Taking the Laplace Transform, we
conclude that
d At At
L e =L
dt
Ae
se^At − e At . t= 0 = Ae^At

(sI − A )e^A t = I

e^A t = (sI − A )
−1

Therefore, 4
Linear Dynamical 1
At −1 −1
Importance of the Characteristic
polynomial
1
(sI − A )−1 = [adj(sI − A )]J
det(sI −
wher A)
e m1
m2
mk
1
det(sI − A ) = (s − λ 1 ) (s − λ 2 ) · · · (s − λ k )
is the characteristic polynomial of A, whose roots λ i
2 adj(sI A ) is the adjoint
are the−eigenvalues matrix of sI − A whose
of A and,
entries are polynomials in s of degree (n − 1) or
lower
To compute L − 1 [(sI − A)−1], we need to perform the
partial fraction expansion.

42
Linear Dynamical
Importance of the Characteristic
polynomial
These are of the forms

α 1 sn − 1 + α 2 sn − 2 + · · · + α n − 1 s + α n
(s − λ 1 )m 1 (s −aλ112 )m 2 · · · a(s12− a 1m1
λ k )m k = + +···+ +··
· s − λ1 (s − λ 1 )2 (sa−k 1λ 1 )m 1 ak m k
+ (s − λ k ) + · · · +(s −
λ k )m k

4
Linear Dynamical 3
Importance of the Characteristic
polynomial
These are of the forms

α 1 sn − 1 + α 2 sn − 2 + · · · + α n − 1 s + α n
(s − λ 1 )m 1 (s −aλ112 )m 2 · · · a(s12− a 1m1
λ k )m k = + +···+ +··
· s − λ1 (s − λ 1 )2 (sa−k 1λ 1 )m 1 ak m k
+ (s − λ k ) + · · · +(s −
λ k )m k
The inverse Laplace transform is then
given by
α 1 sn − 1 + · · · +n − 1 s + n
L −1
α(s − λ 1 )m 1 · · · (s −α
= a11 eλλ1 tk +
)m k· · · + a tm 1 − 1 eλ 1t + · · · + a λ ket + · · ·
1m 1 k1 km k t
m k − 1 λ kt

+a e

4
Linear Dynamical 3
Importance of the Characteristic
polynomial
These are of the forms

α 1 sn − 1 + α 2 sn − 2 + · · · + α n − 1 s + α n
(s − λ 1 )m 1 (s −aλ112 )m 2 · · · a(s12− a 1m1
λ k )m k = + +···+ +··
· s − λ1 (s − λ 1 )2 (sa−k 1λ 1 )m 1 ak m k
+ (s − λ k ) + · · · +(s −
λ k )m k
The inverse Laplace transform is then
given by
α 1 sn − 1 + · · · +n − 1 s + n
L −1
α(s − λ 1 )m 1 · · · (s −α
= a11 eλλ1 tk +
)m k· · · + a tm 1 − 1 eλ 1t + · · · + a λ ket + · · ·
1m k1 km tm k − 1 λ kt
1 k
+a e
Thus when all the eigenvalues λ i of A have strictly negative real parts, all
entries of e A t converge to zero as t → ∞, i.e., y(t) converges to
the forced
respons ∫ t
e
yf (t) = C eA ( t − τ ) B u(τ)dτ +
D u(t) t 0
4
Linear Dynamical 3
Discretizati
on Consider the continuous-time state
equation
x˙(t) = Ax (t) + B u(t) (11
)
y(t) = C x(t) +
D u(t)

4
Linear Dynamical 4
Discretizati
on Consider the continuous-time state
equation
x˙(t) = Ax (t) + B u(t) (11
)
y(t) = C x(t) +
D u(t)
x(t + T ) −
For discretization, since x(t)
x˙(t) =T → 0
lim T
we can approximate (11)
as
x(t + T ) = x(t) + Ax (t)T +
B u(t)T

4
Linear Dynamical 4
Discretizati
on Consider the continuous-time state
equation
x˙(t) = Ax (t) + B u(t) (11
)
y(t) = C x(t) +
D u(t)
x(t + T ) −
For discretization, since x(t)
x˙(t) =T → 0
lim T
we can approximate (11)
as
x(t + T ) = x(t) + Ax (t)T + B u(t)T
If we compute x(t) and y(t) only at t = kT for k = 1,
2, . . ., then
x((k + 1)T ) = (I + T A )x(kT ) + T B u(kT )
y(kT ) = C x(kT ) + D u(kT )
results for the This discretization is easy to carry out 4
same T . but yieldsLinear
theDynamical
least accurate 4
Discretization: Another
method
Let
u(t) = u(kT ) , u[k], for kT ≤ t ≤ (k +
1)T (12) for k = 0, 1, 2, . . .. This input changes values
only at discrete-time instants.

4
Linear Dynamical 5
Discretization: Another
method
Let
u(t) = u(kT ) , u[k], for kT ≤ t ≤ (k +
1)T (12) for k = 0, 1, 2, . . .. This input changes values
only at discrete-time instants.
∫ kT
Compute the solution
AkT of CT system at t = kT and t =
(k + 1)T x[k] , x(kT ) = e x(0) + eA ( k T − τ ) B u(τ)dτ
(13) 0
an
d ∫ (k+ 1) T
x[k + 1] , x((k + 1)T ) A ( k + 1 ) T x(0) eA ( ( k + 1 ) T − τ ) Bu(τ)dτ
=e + 0

4
Linear Dynamical 5
Discretization: Another
method
Let
u(t) = u(kT ) , u[k], for kT ≤ t ≤ (k +
1)T (12) for k = 0, 1, 2, . . .. This input changes values
only at discrete-time instants.
∫ kT
Compute the solution
AkT of CT system at t = kT and t =
(k + 1)T x[k] , x(kT ) = e x(0) + eA ( k T − τ ) B u(τ)dτ
(13) 0
an
d ∫ (k+ 1) T
x[k + 1] , x((k + 1)T ) =A (ek + 1 ) T x(0) eA ( ( k + 1 ) T − τ ) Bu(τ)dτ
0
+
kT
= eA T ∫ eA k T x(0) + A (k T −τ )
Bu(τ)dτ
e ∫
0
(k+ 1) T
A (k T + T −τ )
+ Bu(τ)dτ
e kT

4
Linear Dynamical 5
Discretization: Another
method
Let
u(t) = u(kT ) , u[k], for kT ≤ t ≤ (k +
1)T (12) for k = 0, 1, 2, . . .. This input changes values
only at discrete-time instants.
∫ kT
Compute the solution
AkT of CT system at t = kT and t =
(k + 1)T x[k] , x(kT ) = e x(0) + eA ( k T − τ ) B u(τ)dτ
(13) 0
an
d ∫ (k+ 1) T
x[k + 1] , x((k + 1)T ) =A (ek + 1 ) T x(0) eA ( ( k + 1 ) T − τ ) Bu(τ)dτ
0
+
kT
= eA T ∫ eA k T x(0) + A (k T −τ )
Bu(τ)dτ
e ∫
0
(k+ 1) T
+ eA ( k T + T − τ )
Bu(τ)dτ
kT

Substituting (12) and (13) and introducing the new variable α =

kT + T − τ , we get
∫∫ T ,
x[k + 1] = AeT x[k] eA α d α Bu[k]
+ 0
4
Linear Dynamical 5
Discretization: Another
method AT
∫ T
x [k + 1] = e x [k ] + eA α dα Bu[ k]
0

4
Linear Dynamical 6
Discretization: Another
method AT
∫ T
x [k + 1] = e x [k ] + eA α dα Bu[ k]
0

Thus
, x[k + 1] = A d x[k] + y[k] = C d x[k] +
wit B d u[k], D d u[k]
,
h ∫∫ T
A d = eA t Bd = eA α d α B Cd = C Dd = D
0

4
Linear Dynamical 6
Discretization: Another
method AT
∫ T
x [k + 1] = e x [k ] + eA α dα Bu[ k]
0

Thus
, x[k + 1] = A d x[k] + y[k] = C d x[k] +
wit B d u[k], D d u[k]
,
h ∫∫ T
A d = eA t Bd = eA α d α B Cd = C Dd = D
0
Computation of
B d : Note that
,
∫ T ∫ 2 2
I + Aτ + A2 τ T 3 T2I +
+ · ·T· dτ =
0 A+
A
+ · · · 2! 2!
2!

4
Linear Dynamical 6
Discretization: Another
method AT
∫ T
x [k + 1] = e x [k ] + eA α dα Bu[ k]
0

Thus
, x[k + 1] = A d x[k] + y[k] = C d x[k] +
wit B d u[k], D d u[k]
,
h ∫∫ T
A d = eA t Bd = eA α d α B Cd = C Dd = D
0
Computation of
B d : Note that
,
∫ T ∫ 2 2
I + Aτ + A2 τ + · ·T· dτ =T 3 T2I +
0 A+
A
+ · · · 2!
If A is nonsingular, then the series can be 2!
, 2!
written as ∫
A − 1 T A + T 2 A 2 + T 3A 3 + · · · + I − I = A− 1 eA T −
I
2! 2!
Thus, we
have B = A − 1 (A − I )B
d d

MATLAB code: [ad, bd] = c2d(a, b, T) 4

Linear Dynamical 6
Solution of Discrete-time
Equations
Consider the discrete-time state-space
equations
x[k + 1] = Ax[k] +
B u[k] x(t 0) = x ∈ Rn , k≥
,
y[k] = C x[k] + D u[k] 0 0

4
Linear Dynamical 7
Solution of Discrete-time
Equations
Consider the discrete-time state-space
equations
x[k + 1] = Ax[k] +
B u[k] x(t 0) = x ∈ Rn , k ≥
,
y[k] = C x[k] + D u[k] 0 0
Once x[0] and u[k], k = 0, 1, . . ., are given, the response
can be computed recursively from the equations.
x[1] = Ax[0] + B u[0]
x[2] = Ax[1] + B u[1] = A 2x[0] + A B u[0] +
B u[1]

4
Linear Dynamical 7
Solution of Discrete-time
Equations
Consider the discrete-time state-space
equations
x[k + 1] = Ax[k] +
B u[k] x(t 0) = x ∈ Rn , k ≥
,
y[k] = C x[k] + D u[k] 0 0
Once x[0] and u[k], k = 0, 1, . . ., are given, the response
can be computed recursively from the equations.
x[1] = Ax[0] + B u[0]
x[2] = Ax[1] + B u[1] = A 2x[0] + A B u[0] +
B u[1]
Proceeding forward, we cankΣ− 1readily obtain, for k > 0,
k
x[k] = A x[0] + A k − 1− m
B u[m] m =0
kΣ− 1
y[k] = C Ak x[0] + C A k − 1− m B u[m] +
D u[k] m =0

State transition matrix, φ[k, k0] = A k − k 0 , ∀k

4
≥ k0 Linear Dynamical 7
Computation of φ[k,
k0]The matrix power can be computed using Z-
transforms. ∞ !
h i Σ Σ∞ Σ∞
Z Ak + 1 , z−k
k+ 1
A =z z−( k + 1 ) A k + 1 = z −k
k
A −
k= 0 k= 0 z k= 0 I

4
Linear Dynamical 8
Computation of φ[k,
k0]The matrix power can be computed using Z-
transforms. ∞ !
h i Σ Σ∞ Σ∞
Z Ak + 1 , z −k A k + 1 = z z−( k + 1 ) A k + 1 = z −k
k
A −
k= 0 k= 0 z k= 0 I
h i
= z Z Ak − I

Also, Z A k + 1 = AZ
Ak .

4
Linear Dynamical 8
Computation of φ[k,
k0]The matrix power can be computed using Z-
transforms. ∞ !
h i Σ Σ∞ Σ∞
Z Ak + 1 , z −k A k + 1 = z z−( k + 1 ) A k + 1 = z −k
k
A −
k= 0 k= 0 z k= 0 I
h i
= z Z Ak − I

Also, Z A k + 1 = AZ A k .
Therefore

AA^k = z A^k − I ⇔ (zI −

A)A^k = zI ⇔ A^k = z(zI − A) − 1

4
Linear Dynamical 8
Computation of φ[k,
k0]The matrix power can be computed using Z-
transforms. ∞ !
h i Σ Σ∞ Σ∞
Z Ak + 1 , z −k A k + 1 = z z−( k + 1 ) A k + 1 = z −k
k
A −
k= 0 k= 0 z k= 0 I
h i
= z Z Ak − I

Also, Z A k + 1 = AZ A k .
AA^k = z ^Ak − I ⇔ (zI − A )A^k = zI ⇔ ^
Ak = z(zI − −1
Therefore
A)
Taking inverse Z-transform, we obtain

A k = Z −1 z(zI − A )−1

4
Linear Dynamical 8
Computation of φ[k,
k0]The matrix power can be computed using Z-
transforms. ∞ !
h i Σ Σ∞ Σ∞
Z Ak + 1 , z −k A k + 1 = z z−( k + 1 ) A k + 1 = z −k
k
A −
k= 0 k= 0 z k= 0 I
h i
= z Z Ak − I

Also, Z A k + 1 = AZ A k .
^k = z ^Ak − I ⇔ (zI − A )A
AA ^k = zI ⇔ ^ Ak = z(zI − −1
Therefore
A)
Taking inverse Z-transform, we
obtain
A k = Z−1 z(zI −
−1
A)
Now, when all eigenvalues of A have magnitude smaller
than 1, all entries of A k will converge to zero as t → ∞,
which means that the output will converge to the forced 4
response. Linear Dynamical 8
Week 1 -
Lecture 4

In the last lecture, we discussed

Solution of LTI system both in CT and DT
domain Two methods of discretization
Importance of characteristic polynomial

49
Linear Dynamical
Equivalent state equations
Consider the network shown
below:

5
Linear Dynamical 0
Equivalent state
equations
Consider the network shown
below:

State variable, x(t) State variable,

x 1 : Inductor current; x 2 : Capacitor x¯(t)
voltage x¯ 1 , x¯ 2 : Loop currents

5
Linear Dynamical 0
Equivalent state
equations
Consider the network shown
below:

State variable, x(t) State variable,

x 1 : Inductor current; x 2 : Capacitor x¯(t)
voltage x¯ 1 , x¯ 2 : Loop currents
x˙1 0 −1 x 1 1 ẋ 1 −1 1 x¯1 1
= + u = + u
x˙2 1 −1 x2 0 ¯x¯˙ 0 x¯2 1
2
−1

5
Linear Dynamical 0
Equivalent state
equations
Consider the network shown
below:

State variable, x(t) State variable,

x 1 : Inductor current; x 2 : Capacitor x¯(t)
voltage x¯ 1 , x¯ 2 : Loop currents
x˙1 0 −1 x 1 1 ẋ 1 −1 1 x¯1 1
= + u = + u
x˙2 1 −1 x2 0 ¯x¯˙ 0 x¯2 1
2
−1

Problem
Given two or more state-space equations, when can we
say that
these equations are equivalent or describe the 5
same system? Linear Dynamical 0
Equivalent LTI state
equations
Consider the n-dimensional continuous-time LTI
system
x˙ = Ax + B u, y
= Cx + Du

5
Linear Dynamical 1
Equivalent LTI state
equations
Consider the n-dimensional continuous-time LTI
system
x˙ = Ax + B u, y = Cx
+ Du
Given a nonsingular matrix T , suppose that we
define
x¯ , T x

5
Linear Dynamical 1
Equivalent LTI state
equations
Consider the n-dimensional continuous-time LTI
system
x˙ = Ax + B u, y = Cx
+ Du
Given a nonsingular matrix T , suppose that we
define
x¯ , T x
The same system can be defined using x¯ as the
state,

x¯˙ = T x˙ = T Ax + T B u = T AT −1 x¯
+ T B u y = C x + D u = C T −1 x¯ + D u

5
Linear Dynamical 1
Equivalent LTI state
equations
Consider the n-dimensional continuous-time LTI
system
x˙ = Ax + B u, y = Cx +
Du
Given a nonsingular matrix T , suppose that we
define
x¯ , T x
The same system can be defined using x¯ as the
state,

x¯˙ = T x˙ = T Ax + T B u = T AT −1 x¯
+ T B u y = C x + D u = C T −1 x¯ + D u
which can
A¯ ,beT written
AT Bas
¯, T C¯ , C T −1 , D¯ , D
5
−1
, ¯
B, ¯ Linear Dynamical 1
Equivalent LTI state
equations
Consider the n-dimensional continuous-time LTI system
x˙ = A x + B u , (14
)
y = Cx + Du

5
Linear Dynamical 2
Equivalent LTI state
equations
Consider the n-dimensional continuous-time LTI system
x˙ = A x + B u , (14
)
y = Cx + Du
Let T be an(Algebraically
Definition n × n real nonsingular matrix and let x¯ = T x, then
Equivalent)
the state equation
x¯˙ = A¯x¯ + B¯ u, y=
C¯ x¯ + D¯ u

where A¯ = T AT − 1 , B¯ = T B , C¯ = C T − 1 , D¯ = D is said to be
(algebraically) equivalent to (14) and x¯ = T x is called an equivalence
transformation.

5
Linear Dynamical 2
Equivalent LTI state
equations
Consider the n-dimensional continuous-time LTI system
x˙ = A x + B u , (14
)
y = Cx + Du
Let T be an(Algebraically
Definition n × n real nonsingular matrix and let x¯ = T x, then
Equivalent)
the state equation
x¯˙ = A¯x¯ + B¯ u, y=
C¯ x¯ + D¯ u

where A¯ = T AT − 1 , B¯ = T B , C¯ = C T − 1 , D¯ = D is said to be
(algebraically) equivalent to (14) and x¯ = T x is called an equivalence
transformation.

Property
The equivalent transformations have the same
set of eigenvalues

transfer
functions

5
Linear Dynamical 2
Equivalent LTI state
equations
Consider the n-dimensional continuous-time LTI system
x˙ = A x + B u , (14
)
y = Cx + Du
Let T be an(Algebraically
Definition n × n real nonsingular matrix and let x¯ = T x, then
Equivalent)
the state equation
x¯˙ = A¯x¯ + B¯ u, y=
C¯ x¯ + D¯ u

where A¯ = T AT − 1 , B¯ = T B , C¯ = C T − 1 , D¯ = D is said to be
(algebraically) equivalent to (14) and x¯ = T x is called an equivalence
transformation.

Property
The equivalent transformations have the same
set of eigenvalues

∆¯ ( λ ) = det( λ I − A¯ ) = det( λ T T −1
− T AT −1
)

−1
= det T ( λ I − A ) T = det(λI − A )
= ∆(λ)
5
transfer functions Linear Dynamical 2
Equivalent LTI state
equations
Consider the n-dimensional continuous-time LTI system
x˙ = A x + B u , (14
)
y = Cx + Du
Let T be an(Algebraically
Definition n × n real nonsingular matrix and let x¯ = T x, then
Equivalent)
the state equation
x¯˙ = A¯x¯ + B¯ u, y=
C¯ x¯ + D¯ u

where A¯ = T AT − 1 , B¯ = T B , C¯ = C T − 1 , D¯ = D is said to be
(algebraically) equivalent to (14) and x¯ = T x is called an equivalence
transformation.

Property
The equivalent transformations have the same
set of eigenvalues

∆¯ ( λ ) = det( λ I − A¯ ) = det( λ T T −1
− T AT −1
)

−1
= det T ( λ I − A ) T = det(λI − A )
= ∆(λ)
5
transfer functions Linear Dynamical 2
Equivalent LTI state
equations

Definition (Zero-state equivalent)

Two state equations are said to be zero state equivalent
whenever they have the same transfer function matrix

5
Linear Dynamical 3
Equivalent LTI state
equations

Definition (Zero-state equivalent)

Two state equations are said to be zero state equivalent
whenever they have the same transfer function matrix

Zero- Algebra
state ic
equivalen equivalen
ce ce

5
Linear Dynamical 3
Equivalent LTI state
equations

Definition (Zero-state equivalent)

Two state equations are said to be zero state equivalent
whenever they have the same transfer function matrix

Zero-state Algebra
⇐= ic
equivalence equivalen
ce

5
Linear Dynamical 3
Equivalent LTI state
equations

Definition (Zero-state equivalent)

Two state equations are said to be zero state equivalent
whenever they have the same transfer function matrix

Zero- ⇐=
state Algebraic
equivalen =/ ⇒
ce equivalence

5
Linear Dynamical 3
Equivalent LTI state
equations

Definition (Zero-state equivalent)

Two state equations are said to be zero state equivalent
whenever they have the same transfer function matrix

Zero- ⇐=
state Algebraic
equivalen =/ ⇒
ce equivalence

Under what conditions we can

ensure the zero-state
equivalence?

5
Linear Dynamical 3
Markov
parameters

5
Linear Dynamical 4
Markov
parameters
We know
that
"
Σ∞ t i
#
( s I − A ) − 1 = L [eA t ] = L Ai
i = 0 i!

5
Linear Dynamical 4
Markov
parameters
We know
that
"
Σ∞ t i
#
( s I − A ) − 1 = L [eA t ] = L Ai
i = 0 i!
Sinc
ti
e L
i! = s −(i+1),

we conclude
Σ∞
that ( s I − A )− 1
= s− A .
(i+ 1) i
i= 0

5
Linear Dynamical 4
Markov
parameters
We know
that
"
Σ∞ t i
#
( s I − A ) − 1 = L [eA t ] = L Ai
i = 0 i!
Sinc
ti
e L
i! = s −(i+1),

we conclude
Σ∞
that ( s I − A )− 1
= s− A .
(i+ 1) i
i= 0

Therefor Σ∞
e, Gˆ ( s ) = C ( s I − A ) − 1 B + D = D + s−(i + 1 ) C A i B
i=0

The matrices D , C A i B , i ≥ 0 are called the

Markov parameters

5
Linear Dynamical 4
Markov
parameters
We know
that
"
Σ∞ t i
#
( s I − A ) − 1 = L [eA t ] = L Ai
i = 0 i!
Sinc
ti
e L
i! = s −(i+1),

we conclude
Σ∞
that ( s I − A )− 1
= s− A .
(i+ 1) i
i= 0

Therefor Σ∞
e, Gˆ ( s ) = C ( s I − A ) − 1 B + D = D + s−(i + 1 ) C A i B
i= 0

The matrices D , C A i B , i ≥ 0 are called the Markov parameters which

related
are alsoto the system’s impulse response i.e.

G ( t ) = L − 1 [G ˆ (s)] = L − 1 [ C ( s I − A ) − 1 B + D ] = C e A t B + Dδt

5
Linear Dynamical 4
Markov
parameters
We know
that
"
Σ∞ t i
#
( s I − A ) − 1 = L [eA t ] = L Ai
i = 0 i!
Sinc
ti
e L
i! = s −(i+1),

we conclude
Σ∞
that ( s I − A )− 1
= s− A .
(i+ 1) i
i= 0

Therefor Σ∞
e, Gˆ ( s ) = C ( s I − A ) − 1 B + D = D + s−(i + 1 ) C A i B
i= 0

The matrices D , C A i B , i ≥ 0 are called the Markov parameters which

related
are alsoto the system’s impulse response i.e.

G ( t ) = L − 1 [G ˆ (s)] = L − 1 [ C ( s I − A ) − 1 B + D ] = C e A t B + Dδt

Taking derivative of the RHS, we get

di G(t)
= C A i eA t B , ∀i ≥ 1, t >
dt i 0
from which we obtain the relationship:limt → 0 di G(t)
= CAiB, ∀i ≥
dti 5
1 4
Linear Dynamical
Equivalent LTI state
equations
Theorem
Two state-space representations

x˙ = Ax + y = Cx +
B u, Du
and

x¯˙ = A¯x¯ + B¯u, y = C¯x¯ + D¯ u

are zero-state equivalent or have the same transfer function

matrix if and only if they have the same Markov parameters i.e.,

D = D¯ , C A i B = C¯A¯i B¯, ∀i ≥ 0.

5
Linear Dynamical 5
Equivalent LTI state
equations
Theorem
Two state-space representations

x˙ = Ax + y = Cx +
B u, Du
and

x¯˙ = A¯x¯ + B¯u, y = C¯x¯ + D¯ u

are zero-state equivalent or have the same transfer function

matrix if and only if they have the same Markov parameters i.e.,

D = D¯ , C A i B = C¯A¯i B¯, ∀i ≥ 0.

Prove it by yourself!!
5
Linear Dynamical 5
Equivalent LTV state
equations
Consider the n-dimensional continuous-time LTV system

x˙ = A (t)x + B (t)u, y= (15

C (t)x + D (t)u )

5
Linear Dynamical 6
Equivalent LTV state
equations
Consider the n-dimensional continuous-time LTV system

x˙ = A (t)x + B (t)u, y= (15

C (t)x + D (t)u )

Definition (Algebraically Equivalent)

Let P(t) ∈ R n × n be a non-singular matrix and both P(t) and
P˙ (t) are for all t. Let x¯ , P(t)x, then the state
continuous
equation
(16
x¯˙ = A¯(t)x¯ + B¯ (t)u, y )
wher
e = C¯ (t)x¯ + D¯ (t)u
h i
A¯(t) = P (t)A (t) + P˙ (t) C¯ (t) =
−1
P¯ (t)(t), C (t)P − 1 (t)
B =
P (t)Balgebraically
is said to be (t), equivalent to (15) and P(t)
D¯ (t) = is called an
D (t)
algebraic equivalent transformation.

5
Linear Dynamical 6
Equivalent LTV state
equations
Theorem (Equivalence of fundamental matrix)

Let X (t) be a fundamental matrix of (15), then X¯ (t) = P(t)X(t) is

a fundamental matrix of (16).

57
Linear Dynamical
Equivalent LTV state
equations
Theorem

Let A 0 be an arbitrary constant matrix. Then there exists an equivalent

transformation that transforms (15) into (16) with A¯(t) = A 0

5
Linear Dynamical 8
Equivalent LTV state
equations
Theorem

Let A 0 be an arbitrary constant matrix. Then there exists an equivalent

transformation that transforms (15) into (16) with A¯(t) = A 0

Proof.
Let X ( t ) be a fundamental matrix of x˙ = A ( t ) x .

5
Linear Dynamical 8
Equivalent LTV state
equations
Theorem

Let A 0 be an arbitrary constant matrix. Then there exists an equivalent

transformation that transforms (15) into (16) with A¯(t) = A 0

Proof.
Let X ( t ) be a fundamental matrix of x˙ = A ( t ) x . The
differentiation of
X − 1 ( t ) X ( t ) = I yields X˙ − 1 ( t ) X ( t ) + X − 1 ( t ) X ˙ (t) = 0 which
implies (17
)
X˙ − 1 (t) = − X − 1 ( t ) A ( t ) X ( t ) X − 1 ( t ) = − X − 1 ( t ) A ( t )

5
Linear Dynamical 8
Equivalent LTV state
equations
Theorem

Let A 0 be an arbitrary constant matrix. Then there exists an equivalent

transformation that transforms (15) into (16) with A¯(t) = A 0

Proof.
Let X ( t ) be a fundamental matrix of x˙ = A ( t ) x . The differentiation of
X − 1 ( t ) X ( t ) = I yields X˙ − 1 ( t ) X ( t ) + X − 1 ( t ) X ˙ (t) = 0 which implies

(17
X˙ − 1 (t) = − X − 1 ( t ) A ( t ) X ( t ) X − 1 ( t ) = − X − 1 ( t ) A ( t )
)
Because A¯(t) = A 0 is a constant matrix, X¯ (t) = e A 0 t is a fundamental
matrix of
x¯˙ = A¯(t)x¯ = A0x¯.

5
Linear Dynamical 8
Equivalent LTV state
equations
Theorem

Let A 0 be an arbitrary constant matrix. Then there exists an equivalent

transformation that transforms (15) into (16) with A¯(t) = A 0

Proof.
Let X ( t ) be a fundamental matrix of x˙ = A ( t ) x . The differentiation of
X − 1 ( t ) X ( t ) = I yields X˙ − 1 ( t ) X ( t ) + X − 1 ( t ) X ˙ (t) = 0 which implies

(17
X˙ − 1 (t) = − X − 1 ( t ) A ( t ) X ( t ) X − 1 ( t ) = − X − 1 ( t ) A ( t )
)
Because A¯(t) = A 0 is a constant matrix, X¯ (t) = e A 0 t is a fundamental
matrix of
x¯˙ = A¯(t)x¯ = A0x¯.

Define X¯ (t) = P ( t ) X ( t ) = ⇒ P (t) = X¯ ( t ) X − 1 ( t ) = e A 0 t X − 1 ( t )

5
Linear Dynamical 8
Equivalent LTV state
equations
Theorem

Let A 0 be an arbitrary constant matrix. Then there exists an equivalent

transformation that transforms (15) into (16) with A¯(t) = A 0

Proof.
Let X ( t ) be a fundamental matrix of x˙ = A ( t ) x . The differentiation of
X − 1 ( t ) X ( t ) = I yields X˙ − 1 ( t ) X ( t ) + X − 1 ( t ) X ˙ (t) = 0 which implies

(17
X˙ − 1 (t) = − X − 1 ( t ) A ( t ) X ( t ) X − 1 ( t ) = − X − 1 ( t ) A ( t )
)
Because A¯(t) = A 0 is a constant matrix, X¯ (t) = e A 0 t is a fundamental
matrix of
x¯˙ = A¯(t)x¯
Define X¯ (t)= =A0Px¯.( t ) X ( t ) = ⇒ P (t) = X¯ ( t ) X − 1 ( t ) = e A 0 t X − 1 ( t )
and compute

h i
A¯(t) = P (t)A (t) + P˙ ( t ) P −1
(t)

h i
= eA 0 t X − 1 ( t ) A ( t ) + A 0 e A 0 t X − 1 ( t ) + e A 0 tX˙ − 1 ( t ) X ( t ) e − A 0 t

5
Linear Dynamical 8
Equivalent LTV state
equations
Theorem

Let A 0 be an arbitrary constant matrix. Then there exists an equivalent

transformation that transforms (15) into (16) with A¯(t) = A 0

Proof.
Let X ( t ) be a fundamental matrix of x˙ = A ( t ) x . The differentiation of
X − 1 ( t ) X ( t ) = I yields X˙ − 1 ( t ) X ( t ) + X − 1 ( t ) X ˙ (t) = 0 which implies

(17
X˙ − 1 (t) = − X − 1 ( t ) A ( t ) X ( t ) X − 1 ( t ) = − X − 1 ( t ) A ( t )
)
Because A¯(t) = A 0 is a constant matrix, X¯ (t) = e A 0 t is a fundamental
matrix of
x¯˙ = A¯(t)x¯
Define X¯ (t)= =A0Px¯.( t ) X ( t ) = ⇒ P (t) = X¯ ( t ) X − 1 ( t ) = e A 0 t X − 1 ( t )
and compute

h i
A¯(t) = P (t)A (t) + P˙ ( t ) P −1
(t)

h i
= eA 0 t X − 1 ( t ) A ( t ) + A 0 e A 0 t X − 1 ( t ) + e A 0 tX˙ − 1 ( t ) X ( t ) e − A 0 t

which becomes after substituting (17)

5
Linear Dynamical 8
¯ A 0t −1 −A 0t
Equivalent LTV state equations: Additional
points
If A 0 is chosen as zero matrix, then P(t) = X − 1 (t), thus
A¯(t) = 0, B¯(t) = X − 1 (t)B(t), C¯(t) = C(t)X(t), D¯ (t) =
D(t)

5
Linear Dynamical 9
Equivalent LTV state equations: Additional
points
If A 0 is chosen as zero matrix, then P(t) = X − 1 (t), thus
A¯(t) = 0, B¯(t) = X − 1 (t)B(t), C¯(t) = C(t)X(t), D¯ (t) =
D(t)

1 Every time-varying state equation can be transformed

into such a block diagram

5
Linear Dynamical 9
Equivalent LTV state equations: Additional
points
If A 0 is chosen as zero matrix, then P(t) = X − 1 (t), thus
A¯(t) = 0, B¯(t) = X − 1 (t)B(t), C¯(t) = C(t)X(t), D¯ (t) =
D(t)

1 Every time-varying state equation can be transformed

into such a block diagram
2 However, the challenge is to determine its fundamental
matrix. 5
9
Linear Dynamical
Equivalent LTV state equations: Additional
points
Invariance of Impulse Response matrix

G(t, τ) = C(t)φ(t, τ)B(τ) + D(t)δ(t − τ)

= C (t)X (t)X −1
(τ)B (τ) + D (t)δ(t − τ)

Using the above substitutions, we get

G¯(t, τ) = C¯(t)X¯ (t)X¯ −1(τ)B¯(τ ) + D¯ (t)δ(t − τ)

= C P −1 P X X −1
(τ)P −1 (τ)P (τ)B (τ) + D (t)δ(t −
τ)
= CX X −1
(τ)B (τ) + D (t)δ(t − τ) = G (t, τ)

Thus, the impulse response is invariant under any

equivalence transformation.
60
Linear Dynamical
Concluding remarks on
equivalence
Definition (Lyapunov transformation)
A matrix P(t) is called a Lyapunov transformation
whenever
1 P(t) is non singular

2
P (t) and P˙ (t) are continuous
3
P(t) and P − 1 (t) are bounded for
all t.

6
Linear Dynamical 1
Concluding remarks on
equivalence
Definition (Lyapunov transformation)
A matrix P(t) is called a Lyapunov transformation
whenever
1 P(t) is non singular

2
P (t) and P˙ (t) are continuous
3
P(t) and P − 1 (t) are bounded for
Recaall t.
ll
x˙ = A (t)x + y = C (t)x + (18
D (t)u y = )
B (t)u, x¯˙ =
(19
C¯(t)x¯ + D¯ (t)u )
A¯(t)x¯
Definition + B¯(t)u,equivalent)
(Lyapunov
Equations (18) and (19) are said to be Lyapunov
equivalent whenever P(t) is a Lyapunov
transformation 6
Linear Dynamical 1
Week 1 -
Lecture 5

In the last lecture, we discussed

What is the equivalent representation
problem? Algebraic equivalence of LTI
and LTV systems Zero-state equivalence
of LTI and LTV systems
Relationship between algebraic and zero-
state equivalence.

62
Linear Dynamical
Realization: LTI
systems
Every LTI system can be described by the input-output
description

yˆ(s) = Gˆ(s)uˆ(s)

and if the system is lumped as well, by input-

system-output description

x˙ = Ax + B u, y = Cx + Du

6
Linear Dynamical 3
Realization: LTI
systems
Every LTI system can be described by the input-output
description

yˆ(s) = Gˆ(s)uˆ(s)

and if the system is lumped as well, by input-

system-output description

x˙ = Ax + B u, y = Cx + Du

If the state equation is known, the transfer function

matrix can be computed as Gˆ(s) = C (sI − A )−1 B + D .

6
Linear Dynamical 3
Realization: LTI
systems
Every LTI system can be described by the input-output
description

yˆ(s) = Gˆ(s)uˆ(s)

and if the system is lumped as well, by input-

system-output description

x˙ = Ax + B u, y = Cx + Du

If the state equation is known, the transfer function

Thebe
matrix can computed transfer
computed function
as Gˆ(s) = C (sImatrix
− A )−1isB + D .
unique

6
Linear Dynamical 3
Realization: LTI
systems
Every LTI system can be described by the input-output
description

yˆ(s) = Gˆ(s)uˆ(s)

and if the system is lumped as well, by input-

system-output description

x˙ = Ax + B u, y = Cx + Du

If the state equation is known, the transfer function

Thebe
matrix can computed transfer
computed function
as Gˆ(s) = C (sImatrix
− A )−1isB + D .
unique
Realization problem
Find a state-space equation from a given transfer
matrix. 6
Linear Dynamical 3
Realization: LTI
systems
Definition (Realization)

A transfer function matrix Gˆ(ξ), ξ ∈ {s, z } is said to be

realizable
whenever there exists a finite-dimensional state equation
or simply
{ A, B , C , D } such that

Gˆ(ξ) = C (ξI − A )−1 B + D , ξ ∈ { s,

z}

and { A , B , C , D } is called a realization of Gˆ(ξ)

6
Linear Dynamical 4
Realization: LTI
systems
Definition (Realization)

A transfer function matrix Gˆ(ξ), ξ ∈ {s, z } is said to be

realizable
whenever there exists a finite-dimensional state equation
or simply
{ A, B , C , D } such that

Gˆ(ξ) = C (ξI − A )−1 B + D , ξ ∈ { s,

z}
1
Gˆ, (ξ)
and If{ A B , Cis, D
realizable then
} is called it has “infinitely”
a realization of many
realizations, not necessarily of the same dimension
Gˆ(ξ) Note the realization problem is fairly
complex

6
Linear Dynamical 4
Realization: LTI
systems
Definition (Realization)

A transfer function matrix Gˆ(ξ), ξ ∈ {s, z } is said to be

realizable
whenever there exists a finite-dimensional state equation
or simply
{ A, B , C , D } such that

Gˆ(ξ) = C (ξI − A )−1 B + D , ξ ∈ { s,

z}
1
Gˆ, (ξ)
and If{ A B , Cis, D
realizable then
} is called it has “infinitely”
a realization of many
realizations, not necessarily of the same dimension
Gˆ(ξ) Note the realization problem is fairly
2 Here we shall complex
study the “realizability condition” and
compute one realization
6
Linear Dynamical 4
Realization: LTI
systems

Theorem

A transfer function matrix Gˆ(s) is realizable if and only if Gˆ(s) is

a proper rational matrix.

6
Linear Dynamical 5
Realization: LTI
systems

Theorem

A transfer function matrix Gˆ(s) is realizable if and only if Gˆ(s) is

a proper rational matrix.

The proof shall be done in two

parts. Theorem (Necessary

part)

If Gˆ(s) is realizable then Gˆ(s) is a

proper rational matrix.

Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix 6

Linear Dynamical 5
Realization: LTI
systems
Theorem (Necessary part)

If Gˆ(s) is realizable then Gˆ(s) is a proper rational

matrix.

6
Linear Dynamical 6
Realization: LTI
systems
Theorem (Necessary part)

If Gˆ(s) is realizable then Gˆ(s) is a proper rational

matrix.
Proof.

If Gˆ is realizable, then we can

write 1
(s) = C (sI − A )− 1 B =
Gˆs p ′
C [A dj (sI − A )]
B det(sI − A )

6
Linear Dynamical 6
Realization: LTI
systems
Theorem (Necessary part)

If Gˆ(s) is realizable then Gˆ(s) is a proper rational

matrix.
Proof.

If Gˆ is realizable, then we can

write 1
(s) = C (sI − A )− 1 B =
Gˆs p ′
C [A dj (sI − A )]
B det(sI − A )

If A is n × n, then det(sI − A ) has degree n

Every entry of A dj (sI − A ) has at most degree
(n − 1)

6
Linear Dynamical 6
Realization: LTI
systems
Theorem (Necessary part)

If Gˆ(s) is realizable then Gˆ(s) is a proper rational

matrix.
Proof.

If Gˆ is realizable, then we can

write 1
(s) = C (sI − A )− 1 B =
Gˆs p ′
C [A dj (sI − A )]
B det(sI − A )

If A is n × n, then det(sI − A ) has degree n

Every entry of A dj (sI − A ) has at most degree
(n − 1)
Thus C(sI − A ) − 1 B is a strictly proper rational
matrix.

If D is non-zero, then C(sI − A ) − 1 B + D , Gˆ(s) is

proper.

6
Linear Dynamical 6
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is

realizable.

6
Linear Dynamical 7
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is realizable.

We show the converse; i.e., if Gˆ(s) is a q × p proper rational matrix,

then there exists a realization.

6
Linear Dynamical 7
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is realizable.

We show the converse; i.e., if Gˆ(s) is a q × p proper rational matrix,

then there exists a realization.
Proof

Decompose Gˆ
as

Gˆ (s) = Gˆ (∞ ) +

Gˆ s p (s).

6
Linear Dynamical 7
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is realizable.

We show the converse; i.e., if Gˆ(s) is a q × p proper rational matrix,

then there exists a realization.
Proof

Decompose Gˆ
as

Gˆ (s) = Gˆ (∞ ) +

Gˆ s p (s).

Let

d(s) = s r + α 1 s r −1

+ · · · + αr −1s + 6
Linear Dynamical 7
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is realizable.

We show the converse; i.e., if Gˆ(s) is a q × p proper rational matrix,

then there exists a realization.
Proof

Decompose Gˆ
as

Gˆ (s) = Gˆ (∞ ) +

Gˆ s p (s).

Let 1 1
(s) =
Gˆs p
[N (s)] = N1 sr − 1 + N2 sr − 2 + ... + r −1s
+ Nr
d(s)
N (20)
d(s) = s r + α 1 sd(s)
r −1

where N i are q × p constant matrix.

+ · · · + αr −1s + 6
Linear Dynamical 7
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is realizable.

Proof (Cont.)
We claim that the set of equations
   
... Ip
− α 1I p − α 2I p −α r − 1 I p −α r 0

Ip 0 ... 0 0   
x˙ = . . . Ip 0 0
 0 Ip x+ 0u
    (21
.. .. .. .. ..   
 )
0 0 .. 0
Ip
.
...
y = N1 N2 N r−1
... 0
where I p ∈ R p × p , 0 ∈ R p × p , A ∈ R r p × r p , B ∈ R r p × p , C ∈ R q × r p is a
ˆ
realization of G (s) with dimensionN rp.x + Gˆ (∞ )u
r

6
Linear Dynamical 8
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is realizable.

Proof (Cont.)
We claim that the set of equations
   
... Ip
− α 1I p − α 2I p −α r − 1 I p −α r 0

Ip 0 ... 0 0   
x˙ = . . . Ip 0 0
 0 Ip x+ 0u
    (21
.. .. .. .. ..   
 )
0 0 .. 0
Ip
.
...
y = N1 N2 N r−1
... 0
where I p ∈ R p × p , 0 ∈ R p × p , A ∈ R r p × r p , B ∈ R r p × p , C ∈ R q × r p is a
ˆ
realization of G (s) with dimension rp.We shall show that (21) is a
N r x + Gˆ (∞ )u
realization.

6
Linear Dynamical 8
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is

realizable.
Let us(Cont.)
Proof
define  
Z1
Z =  Z2 
 , (sI − A )− 1 (22
 .. 
 B )
Zr

where Z i is p × p and Z is r p × p.

6
Linear Dynamical 9
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is

realizable.
Let us(Cont.)
Proof
define  
Z1
Z =  Z2 
 , (sI − A )− 1 (22
 .. 
 B )
Zr

where Z i is p × p and Z is rp × p. Then the transfer matrix of (21)

(23
C (s I − A )− 1 B + Gˆ (∞ ) = N 1 Z 1 + N 2 Z 2 + · · · + N r Z r + Gˆ (∞ )
equals )
Write (22) as (sI − A )Z = B or

sZ = A Z + B

6
Linear Dynamical 9
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is realizable.

Proof (Cont.)
Using the shifting property of the matrix A, from the second to the
last block, we can readily obtain,
       
Z1 − α1 p I −α 2 I p −αr I p Z1 Ip
−α r− 1 I p
... Ip 0 0
Z 2   0  Z2   0
Z 3  0 ... 0 0  Z 3   0
sZ = A Z + B ≡ s = Ip +
. . ... . .
 .  . . . .   ..   
 .  . . .   .   
. . .
0 0 ... 0. 0
Z r Ip Z r

sZ2 = Z 1, sZ3 = Z 2, ..., s Z r = Z r−1

7
Linear Dynamical 0
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is realizable.

Proof (Cont.)
Using the shifting property of the matrix A, from the second to the
last block, we can readily obtain,
       
Z1 − α1 p I −α 2 I p −αr I p Z1 Ip
−α r− 1 I p
... Ip 0 0
Z 2   0  Z2   0
Z 3  0 ... 0 0  Z 3   0
sZ = A Z + B ≡ s = Ip +
. . ... . .
 .  . . . .   ..   
 .  . . . . .   .   
. .
0 0 ... 0. 0
Z r Ip Z r

...,
sZ2 = Z 1, sZ3 = Z 2, s Z r = Z r−1
which
implies Z1 Z1 Z1
Z2 = , Z3 = , . . . , Zr =
s s2 s r−1
Substituting these into the first block of
A yields
s Z 1 = −α 1 Z 1 − α 2 Z 2 − · · · − α r Z r +
α2 αr
Ip = − α1 + +··· Z1 + p
s 7
+ s r−1 I 0
Linear Dynamical
Realization: LTI
systems
Theorem (Sufficient part)

If Gˆ(s) is a proper rational matrix then Gˆ(s) is realizable.

Proof
(Cont.)
Using d(s) α2 αr d(s)
s + α1 + +··· Z1 = Z1 = Ip
s s r−1 s r−1
+
Thus
, s r−1 s r−2 1
··
Z1 = Ip, Zr = Ip
d(s) ·, d(s)
Z2 = Ip,
d(s)
C (s I − these
Substituting + Gˆ (∞ ) = N 1 Z 1 + N 2 Z 2 + · · · + N r Z r + Gˆ (∞ )
A )− 1 B into

yields
1
C (sI − A )− 1 B + ˆG (∞) N 1 sr − 1 + N2 sr − 2 + · · · + rN + ˆ
d(s)
= G (∞)

71
Linear Dynamical
Realization: LTV systems
The Laplace Transform cannot be
used

7
Linear Dynamical 2
Realization: LTV
systems
The Laplace Transform cannot be
used input-output description
∫ t
y(t) = G (t,
τ)u(τ)dτt 0

input-state-output description

x˙ = A (t)x +
B (t)u y = C (t)x
+ D (t)u

7
Linear Dynamical 2
Realization: LTV
systems
The Laplace Transform cannot be
used input-output description
∫ t
y(t) = G (t,
τ)u(τ)dτt 0

input-state-output description

x˙ = A (t)x +
B (t)u y = C (t)x
+ D (t)u

If the state equation is available, the impulse

response can be computed from

G (t, τ) = C (t)X (t)X −1

(τ)B (τ) + D (t)δ(t − τ),
∀t ≥ τ 7
Linear Dynamical 2
Realization: LTV
systems
Theorem
A q × p impulse response matrix G(t, τ) is realizable if and only if G(t,
τ)
can be decomposed as

G (t, τ) = M (t)N (τ) + D (t)δ(t − τ), ∀t ≥ τ

where M , N and D are respectively q × n, n × p and q × p matrices

for some integer n.

7
Linear Dynamical 3
Realization: LTV
systems
Theorem
A q × p impulse response matrix G(t, τ) is realizable if and only if G(t,
τ)
can be decomposed as

G (t, τ) = M (t)N (τ) + D (t)δ(t − τ), ∀t ≥ τ

where M , N and D are respectively q × n, n × p and q × p matrices
some integer n.
for
Proof shall be done in two
parts.
Theorem
If G(t, τ) is(Necessary part)there exists a realization that satisfies
realizable then
−1
G (t, τ) = C (t)X (t)X (τ)B (τ) + D (t)δ(t − τ), ∀t ≥ τ

where X (t) is the fundamental matrix.

Theorem (Sufficient part)

If G(t, τ) can be decomposed as mentioned above then G(t, τ) is 7
realizable. Linear Dynamical 3
Realization: LTV
systems
Theorem (Necessary part)
If G(t, τ) is realizable then there exists a realization that satisfies
−1
G (t, τ) = C (t)X (t)X (τ)B (τ) + D (t)δ(t − τ),

∀t ≥ τ

where X (t) is the fundamental matrix.

7
Linear Dynamical 4
Realization: LTV
systems
Theorem (Necessary part)
If G(t, τ) is realizable then there exists a realization that
satisfies
G (t, τ) = C (t)X (t)X − 1 (τ)B (τ) + D (t)δ(t − ∀t ≥
τ), τ

where X (t) is the fundamental matrix.

Proof.
Identifying M(t) = C(t)X(t) and N(τ) = X − 1 (τ)B (τ)
establishes the necessary part of the theorem

7
Linear Dynamical 4
Realization: LTV
systems
Theorem (Necessary part)
If G(t, τ) is realizable then there exists a realization that
satisfies
G (t, τ) = C (t)X (t)X − 1 (τ)B (τ) + D (t)δ(t − ∀t ≥
τ), τ

where X (t) is the fundamental matrix.

Proof.
Identifying M(t) = C(t)X(t) and N(τ) = X − 1 (τ)B (τ)
establishes the necessary part of the theorem

Theorem (Sufficient part)

If G(t, τ) can be decomposed as mentioned above then G(t, τ) is
realizable.

7
Linear Dynamical 4
Realization: LTV
systems
Theorem (Necessary part)
If G(t, τ) is realizable then there exists a realization that
satisfies
G (t, τ) = C (t)X (t)X − 1 (τ)B (τ) + D (t)δ(t − ∀t ≥
τ), τ

where X (t) is the fundamental matrix.

Proof.
Identifying M(t) = C(t)X(t) and N(τ) = X − 1 (τ)B (τ)
establishes the necessary part of the theorem

Theorem (Sufficient part)

If G(t, τ) can be decomposed as mentioned above then G(t, τ) is
realizable.
Proof.
If G(t, τ) can be decomposed as above, then the
n−dimensional state equation

x˙ = N (t)u, y = M (t)x + D (t)u

is a realization. Indeed, a fundamental matrix of x˙ = 0.x is X (t)

7
= I. Thus the impulse response is 4
Linear Dynamical