Chapter 2. Linear Systems: Lecture Notes For MA2327

Download as pdf or txt
Download as pdf or txt
You are on page 1of 50

Chapter 2.

Linear systems
Lecture notes for MA2327

P. Karageorgis

[email protected]

1 / 50
Linear homogeneous systems

Definition 2.1 – Linear homogeneous systems


A linear homogeneous system is a system that has the form

y ′ (t) = A(t)y(t), (LHS)

where y(t) is the vector of unknowns and A(t) is a square matrix.

Theorem 2.2 – Superposition principle


The set of solutions of a linear homogeneous system is closed under
addition and scalar multiplication. In other words, the sum of any two
solutions is a solution and scalar multiples of solutions are solutions.

The superposition principle asserts that the solutions of (LHS) form a


vector space. If one can find some solutions that form a basis for this
vector space, then every solution will be a linear combination of them.
2 / 50
Linear independence of functions

Definition 2.3 – Linear independence of functions


The functions y1 (t), y2 (t), . . . , yn (t) are called linearly dependent, if
there exist constants c1 , c2 , . . . , cn which are not all zero such that

c1 y1 (t) + c2 y2 (t) + . . . + cn yn (t) = 0 for all t.

Otherwise, we say that the functions are linearly independent.

Linear independence of vector-valued functions is a bit more subtle


than linear independence of constant vectors. This is because the
coefficients ck are not allowed to depend on t. For instance,
   
1 t
y1 (t) = , y2 (t) =
0 0

are linearly independent functions, even though one has y2 = ty1 .


3 / 50
Linear independence of functions: Example

We check that y1 (t), y2 (t), y3 (t) are linearly independent when


 t   
e t 1
y1 (t) = , y2 (t) = , y3 (t) = .
t t 1

Suppose that c1 y1 (t) + c2 y2 (t) + c3 y3 (t) = 0, in which case

c1 et + c2 t + c3 = c1 t + c2 t + c3 = 0 for all t.

One may analyse this relation by considering special values of t or by


differentiating, for instance. Differentiating twice, we get c1 et = 0 for
all t, hence also c1 = 0. The given relation may thus be reduced to

c2 t + c3 = 0 for all t.

Letting t = 0 and t = 1, we now get c3 = 0 = c2 + c3 . This implies


that ck = 0 for all k, so the given functions are linearly independent.
4 / 50
Linear independence of solutions

Theorem 2.4 – Linear independence of solutions


Suppose that y1 (t), y2 (t), . . . , yn (t) are solutions of the n × n system

y ′ (t) = A(t)y(t). (LHS)

Then y1 (t), y2 (t), . . . , yn (t) are linearly independent functions if and


only if y1 (0), y2 (0), . . . , yn (0) are linearly independent vectors.

The solutions of an n × n linear homogeneous system form a vector


space of dimension n. In fact, let v1 , v2 , . . . , vn be any basis of Rn
and let yk (t) be the unique solution of the initial value problem

yk′ (t) = A(t)yk (t), yk (0) = vk .

Then y1 (t), y2 (t), . . . , yn (t) are easily seen to form a basis for the
space of solutions. However, such a basis is not usually explicit.
5 / 50
Basis of solutions: Example, page 1
We obtain a basis of solutions for the linear homogeneous system
 
′ 1 0
y (t) = A(t)y(t), A(t) = t .
e 2

In this case, A(t) is lower triangular, so it is easier to look at the


corresponding equations one by one. Let us start by writing

x′ (t) = x(t), y ′ (t) = et x(t) + 2y(t).

When it comes to the leftmost equation, one clearly has

x′ (t) = x(t) =⇒ x(t) = c1 et .

We now insert this fact in the rightmost equation to find that

y ′ (t) − 2y(t) = et x(t) = c1 e2t .

This is a first-order linear equation with integrating factor µ = e−2t .


6 / 50
Basis of solutions: Example, page 2

Multiplying by the integrating factor, we conclude that

(e−2t y)′ = c1 =⇒ e−2t y(t) = c1 t + c2


=⇒ y(t) = (c1 t + c2 )e2t .

This shows that every solution of the system has the form

c 1 et
   t   
e 0
y(t) = = c1 + c2 2t .
(c1 t + c2 )e2t te2t e

In other words, every solution is a linear combination of


 t   
e 0
y1 (t) = 2t , y2 (t) = 2t
te e

and these functions form a basis for the space of solutions.

7 / 50
Systems with constant coefficients

When A is constant, the linear system y ′ (t) = Ay(t) can always be


solved explicitly by relating A to its Jordan form, say J = B −1 AB.
More precisely, the change of variables z(t) = B −1 y(t) gives

z ′ (t) = B −1 y ′ (t) = B −1 Ay(t) = B −1 ABz(t) = Jz(t).

This is a linear system that involves a lower triangular matrix, while


each of the corresponding equations has the form

zk′ = λk zk or zk′ = zk−1 + λk zk .

In particular, each of these equations is first-order linear and one may


determine the variables zk inductively using integrating factors.
The corresponding formula for the solution y(t) = Bz(t) turns out to
be simple when A is diagonalisable but a bit technical, otherwise. We
shall thus use another approach to deal with the general case later.
8 / 50
Eigenvector method

Theorem 2.5 – Eigenvector method


Consider the n × n linear system y ′ (t) = Ay(t) in the case that A is
constant and diagonalisable. Let v1 , v2 , . . . , vn be linearly independent
eigenvectors and let λ1 , λ2 , . . . , λn be the corresponding eigenvalues.
Then every solution of the system has the form

y(t) = c1 eλ1 t v1 + c2 eλ2 t v2 + . . . + cn eλn t vn .

The coefficients ck may be taken to be real, if the eigenvalues of A


are all real. Otherwise, the coefficients ck will generally be complex.
To prove this theorem, we note that each yk (t) = eλk t vk satisfies

yk′ (t) = λk eλk t vk = eλk t Avk = Ayk (t).

This gives n solutions which are linearly independent when t = 0, so


every other solution must be a linear combination of them.
9 / 50
Eigenvector method: Example 1

We use the eigenvector method to solve the linear system


 
′ 3 2
y (t) = Ay(t), A= .
4 5

The eigenvalues of A are the roots of the characteristic polynomial

f (λ) = λ2 − (tr A)λ + det A = λ2 − 8λ + 7 = (λ − 7)(λ − 1),

namely λ1 = 7 and λ2 = 1. These correspond to the eigenvectors


   
1 1
v1 = , v2 = .
2 −1

According to the previous theorem, the solution of the system is

c1 e7t + c2 et
 
y(t) = c1 e7t v1 + c2 et v2 = .
2c1 e7t − c2 et

10 / 50
Eigenvector method: Example 2
We use the eigenvector method to solve the linear system
 
1 0 0
y ′ (t) = Ay(t), A = 2 3 0 .
2 1 4
Since A is lower triangular, its eigenvalues λ = 1, 3, 4 are merely the
diagonal entries of A. These are distinct, so A is diagonalisable and
one may easily check that the corresponding eigenvectors are
     
−3 0 0
v1 =  3 , v2 = −1 , v3 = 0 .
1 1 1
In view of the previous theorem, the solution of the system is thus
−3c1 et
 

y(t) = c1 et v1 + c2 e3t v2 + c3 e4t v3 =  3c1 et − c2 e3t  .


c1 et + c2 e3t + c3 e4t
11 / 50
Matrix exponential: Definition
Definition 2.6 – Matrix exponential

Given a square matrix A, we define its exponential eA as the series



1 2 X 1
eA = I + A + A + ... = Ak .
2! k!
k=0

It can be shown that this series converges for every square matrix A.

To compute the powers of a square matrix, one relates them to the


powers of its Jordan form J = B −1 AB using the computation
Ak = (BJB −1 )k = BJ k B −1 .
A similar approach can be used for the exponential of A since
∞ ∞
A
X 1 k X 1
e = A = BJ k B −1 = BeJ B −1 .
k! k!
k=0 k=0
12 / 50
Matrix exponential: Properties

Theorem 2.7 – Properties of the matrix exponential

Suppose A, B are n × n matrices and let Φ(t) = etA for all t ∈ R.


1 The exponential property eA+B = eA eB holds when AB = BA,
but this property is generally false for arbitrary matrices.
2 The exponential function Φ(t) = etA is such that Φ′ (t) = AΦ(t).
In particular, it is a matrix solution of the system y ′ (t) = Ay(t).
3 The columns of Φ(t) = etA are vector solutions of y ′ (t) = Ay(t)
and they also form a basis for the space of all solutions.

The second property may be stated simply as (etA )′ = AetA . This


resembles the chain rule for the standard exponential function.
The product rule (AB)′ = A′ B + AB ′ also holds for matrix-valued
functions, but the chain rule (A2 )′ = 2AA′ is generally false.

13 / 50
Matrix exponential: Jordan forms
Theorem 2.8 – Matrix exponential of a Jordan form
Suppose that J is a k × k Jordan block with eigenvalue λ. Then the
exponential etJ is a lower triangular matrix and the entries that lie i
j
steps below the diagonal are equal to tj! eλt for each 0 ≤ j < k.

For instance, the exponential of a 3 × 3 Jordan block is given by


 λt
e
  
λ
J =  1 λ  =⇒ etJ =  teλt eλt .
t 2
1 λ λt λt λt
2e te e
The exponential of a Jordan form is obtained by exponentiating each
Jordan block separately. As a typical example, one has
   2t 
2 e
J = 3  =⇒ etJ =  e3t .
1 3 te 3t e 3t

14 / 50
Matrix exponential: Example 1, page 1
We compute the matrix exponential of the diagonalisable matrix
 
4 1
A= .
2 3
The characteristic polynomial of this matrix is given by
f (λ) = λ2 − (tr A)λ + det A = λ2 − 7λ + 10 = (λ − 2)(λ − 5),
so the eigenvalues are real and distinct, namely λ1 = 2 and λ2 = 5.
The corresponding eigenvectors are easily found to be
   
1 1
v1 = , v2 = .
−2 1
Once we now merge the eigenvectors to form a matrix B, we get
   
1 1 −1 2
B= =⇒ J = B AB = .
−2 1 5
15 / 50
Matrix exponential: Example 1, page 2

Since the Jordan form J is diagonal, the same is true for etJ and
   2t 
−1 2 tJ e
J = B AB = =⇒ e = .
5 e5t

As for the exponential of the original matrix A, this is given by

J = B −1 AB =⇒ A = BJB −1 =⇒ etA = BetJ B −1 .

In view of our computations above, we must thus have

1 1 e2t
   
tA 1/3 −1/3
e =
−2 1 e5t 2/3 1/3
2t 5t 2t 5t
 
1 e + 2e −e + e
= 2t 5t .
3 −2e + 2e 2e2t + e5t

The exact same approach applies for any diagonalisable matrix A.


16 / 50
Matrix exponential: Example 2, page 1
We compute the matrix exponential of the non-diagonalisable matrix
 
9 −4
A= .
9 −3
In this case, the characteristic polynomial of A is given by
f (λ) = λ2 − (tr A)λ + det A = λ2 − 6λ + 9 = (λ − 3)2 ,
so the only eigenvalue is λ = 3. The only eigenvector turns out to be
 
2
v= .
3
This implies that A is not diagonalisable and that the Jordan form is
   3t 
−1 3 tJ e
J = B AB = =⇒ e = .
1 3 te3t e3t

Let us now find a matrix B such that J = B −1 AB is in Jordan form.


17 / 50
Matrix exponential: Example 2, page 2

Pick any nonzero vector v1 which is not an eigenvector and let


 
v2 = (A − λI)v1 , B = v1 v2 .

There are obviously infinitely many choices and one possibility is


     
1 6 1 6
v1 = , v2 = (A − 3I)v1 = , B= .
0 9 0 9

In view of our computations above, we must thus have

1 6 e3t
   
tA tJ −1 1 −2/3
e = Be B =
0 9 te3t e3t 0 1/9
 
1 + 6t −4t
= e3t .
9t 1 − 6t

This approach applies for any non-diagonalisable 2 × 2 matrix A.


18 / 50
Matrix exponential: Example 3, page 1
Finally, we consider a real matrix A with complex eigenvalues, say
 
1 −1
A= .
1 1
The characteristic polynomial of this matrix is easily found to be
f (λ) = λ2 − (tr A)λ + det A = λ2 − 2λ + 2 = (λ − 1)2 + 1.
The eigenvalues λ = 1 ± i are complex conjugates of one another and
the same is true for the corresponding eigenvectors which are given by
   
1 1
v1 = , v2 = .
−i i
This implies that A is diagonalisable and that we also have
   
1 1 1+i
B= =⇒ J = B −1 AB = .
−i i 1−i
In particular, one may proceed as before to compute etJ and then etA .
19 / 50
Matrix exponential: Example 3, page 2
Since the Jordan form J is diagonal, the same is true for etJ and
   t it 
−1 1+i tJ ee
J = B AB = =⇒ e = .
1−i et e−it
In view of our computations above, we must thus have
1 1 et eit
   
tA tJ −1 1/2 i/2
e = Be B =
−i i et e−it 1/2 −i/2
et eit + e−it i(eit − e−it )
 
= .
2 i(e−it − eit ) eit + e−it

On the other hand, one has e±it = cos t ± i sin t, so this implies that
et 2 cos t −2 sin t
  t
e cos t −et sin t
 
tA
e = = t .
2 2 sin t 2 cos t e sin t et cos t

Needless to say, etA will always turn out to be real when A is real.
20 / 50
Fundamental matrix

Definition 2.9 – Fundamental matrix


We say that Φ(t) is a fundamental matrix for a linear homogeneous
system, if the columns of Φ(t) form a basis for the space of solutions.

The most common example is the matrix exponential Φ(t) = etA . It is


a fundamental matrix for the system y ′ (t) = Ay(t), if A is constant.

Theorem 2.10 – Properties of fundamental matrices


Let Φ(t) be a fundamental matrix for the system y ′ (t) = A(t)y(t).
1 Every solution is a linear combination of the columns of Φ(t).
2 Every solution has the form y(t) = Φ(t)c for some vector c.
3 The fundamental matrix itself is a matrix solution of the system.
In other words, one has the matrix identity Φ′ (t) = A(t)Φ(t).
21 / 50
Fundamental matrix: Special cases

It is only in a few special cases that one may explicitly determine a


fundamental matrix for the linear system y ′ (t) = A(t)y(t).
When A(t) is either upper or lower triangular, the system can be
solved explicitly by solving the corresponding equations one by one.
Let yk (t) be the unique solution of the initial value problem

yk′ (t) = A(t)yk (t), yk (0) = ek .

Then y1 (t), y2 (t), . . . , yn (t) form a basis for the space of solutions.
When A(t) is a matrix that commutes with its antiderivative B(t), a
fundamental matrix for the system is given by
Z t
B(t)
Φ(t) = e , B(t) = A(s) ds.
0

This is the case, in particular, when A(t) = A is a constant matrix.


22 / 50
Variation of parameters: Intuition

Let us now turn our attention to the linear inhomogeneous system

y ′ (t) = A(t)y(t) + b(t). (LIS)

When it comes to the special case b(t) = 0, there is a fundamental


matrix Φ(t) which satisfies the identity Φ′ (t) = A(t)Φ(t) and every
solution has the form y(t) = Φ(t)c for some constant vector c.
To deal with the general case, we look for solutions that have the
form y(t) = Φ(t)c(t), where c(t) is not necessarily constant. Since

y ′ (t) = Φ′ (t)c(t) + Φ(t)c′ (t)


= A(t)Φ(t)c(t) + Φ(t)c′ (t)
= A(t)y(t) + Φ(t)c′ (t),

we do obtain a solution of (LIS), provided that Φ(t)c′ (t) = b(t).


Thus, one may use Φ(t) to solve the inhomogeneous system as well.
23 / 50
Variation of parameters: Main result

Theorem 2.11 – Variation of parameters


Consider the linear inhomogeneous system

y ′ (t) = A(t)y(t) + b(t). (LIS)

If A(t) and b(t) are continuous, then every solution has the form
Z
y(t) = Φ(t)c + Φ(t) Φ(t)−1 b(t) dt,

where c is a constant vector and Φ(t) is a fundamental matrix for the


associated linear homogeneous system y ′ (t) = A(t)y(t).

The integral term in the equation above is itself a particular solution


of the system. According to the theorem, every solution is thus the
sum of the homogeneous solution Φ(t)c and a particular solution.
24 / 50
Variation of parameters: Example
We use variation of parameters to solve the inhomogeneous system
   
′ 1 0 1
y (t) = Ay(t) + b(t), A= , b(t) = .
1 1 t
Since A is constant, a fundamental matrix is Φ(t) = etA and thus
Z
tA tA
y(t) = e c + e e−tA b(t) dt
 t    t  Z  −t  
e 0 c1 e 0 e 0 1
= + dt
tet et c2 tet et −te−t e−t t
c 1 et
   t  Z  −t 
e 0 e
= t t + t t dt
c1 te + c2 e te e 0
c 1 et
  t
0 −e−t
  
e
= +
c1 tet + c2 et tet et 0
c 1 et − 1
 
= .
c1 te + c2 et + tet − t
t

25 / 50
Higher-order scalar equations

Suppose that we need to solve a scalar linear equation such as

y ′′′ (t) − 5y ′′ (t) + 7y ′ (t) − 3y(t) = 0.

This is a 3rd-order equation, so one may express it as a 3 × 3 system.


More precisely, let y be the vector with entries y, y ′ , y ′′ and note that
   ′   
y y 0 1 0 y
y =  y ′  =⇒ y ′ =  y ′′  = 0 0 1  y ′  = Ay.
y ′′ y ′′′ 3 −7 5 y ′′

Since the scalar equation is linear, the same is true for the system, so
one may determine y using methods we have already developed.
This kind of approach is certainly valid, but it is not very efficient, as
we are only interested in the first entry of y. It is thus worth having
some related results that deal with scalar equations directly.
26 / 50
Linear homogeneous equations

Theorem 2.12 – Linear homogeneous equations


Consider the scalar linear homogeneous equation

an y (n) (t) + . . . + a2 y ′′ (t) + a1 y ′ (t) + a0 y(t) = 0. (LHE)

If the coefficients ak are all constant, then one may obtain a basis of
solutions by solving the corresponding characteristic equation

a n λn + . . . + a 2 λ2 + a 1 λ + a 0 = 0

and by associating each root λ with solutions of (LHE) as follows.


1 If a real root λ has multiplicity k, it gets associated with the k
k−1
functions {tj eλt }j=0 , namely with eλt , teλt , . . . , tk−1 eλt .
2 If a pair of complex roots λ = a ± bi has multiplicity k, it gets
k−1
associated with the 2k functions {tj eat sin(bt), tj eat cos(bt)}j=0 .
27 / 50
Linear homogeneous equations: Example 1

We use the previous theorem to solve the homogeneous equation

y ′′′ (t) − 5y ′′ (t) + 7y ′ (t) − 3y(t) = 0.

In this case, the associated characteristic equation is given by

λ3 − 5λ2 + 7λ − 3 = 0.

Noting that λ = 1 is a root, one easily finds that

λ3 − 5λ2 + 7λ − 3 = (λ − 1)(λ2 − 4λ + 3) = (λ − 1)2 (λ − 3).

This means that λ = 1 is a double root, while λ = 3 is a simple root.


Thus, a basis of solutions is formed by the functions et , tet , e3t and
every solution of the given equation has the form

y(t) = c1 et + c2 tet + c3 e3t .

28 / 50
Linear homogeneous equations: Example 2
Let us now solve an initial value problem such as

y ′′ (t) − y(t) = 0, y(0) = 1, y ′ (0) = 3.

In this case, the associated characteristic equation gives

λ2 − 1 = 0 =⇒ (λ + 1)(λ − 1) = 0 =⇒ λ = −1, 1.

Since the roots are both simple, every solution has the form

y(t) = c1 et + c2 e−t .

Next, we turn to the initial conditions and we note that

y(t) = c1 et + c2 e−t =⇒ 1 = y(0) = c1 + c2 ,


′ t −t
y (t) = c1 e − c2 e =⇒ 3 = y ′ (0) = c1 − c2 .

Solving this system of equations, we find that c1 = 2 and c2 = −1.


Thus, the unique solution is given by y(t) = 2et − e−t .
29 / 50
Linear homogeneous equations: Example 3

The equation that describes a simple harmonic oscillator is

my ′′ (t) = −ky(t).

Here, the constants k, m are both positive, so one may also write

y ′′ (t) + ω 2 y(t) = 0,
p
ω = k/m.

Solving the associated characteristic equation, we now get

λ2 + ω 2 = 0 =⇒ λ2 = −ω 2 =⇒ λ = ±iω.

This is a pair of complex roots, so every solution has the form

y(t) = c1 sin(ωt) + c2 cos(ωt).

In particular, every solution of the given equation is periodic.


30 / 50
Linear homogeneous equations: Example 4
As our last example on homogeneous equations, we now solve
y ′′′ (t) + 7y ′′ (t) + 19y ′ (t) + 13y(t) = 0.
In this case, the associated characteristic equation is given by
λ3 + 7λ2 + 19λ + 13 = 0.
Noting that λ = −1 is a root, one may factor the cubic as
λ3 + 7λ2 + 19λ + 13 = (λ + 1)(λ2 + 6λ + 13).
The roots of the quadratic factor are easily found to be

−6 ± 62 − 4 · 13 −6 ± 4i
λ= = = −3 ± 2i.
2 2
We may thus conclude that every solution has the form
y(t) = c1 e−t + c2 e−3t sin(2t) + c3 e−3t cos(2t).
31 / 50
Linear inhomogeneous equations

Suppose that we need to solve an inhomogeneous equation such as

y ′′ (t) − 3y ′ (t) + 2y(t) = 2t + 5.

The solution of such an equation can be expressed as the sum of the


homogeneous solution yh and a particular solution yp . More precisely,
the difference z = y − yp between any two solutions satisfies

z ′′ (t) − 3z ′ (t) + 2z(t) = 0,

so it is a solution of the corresponding homogeneous equation.


This proves the useful identity y = yh + yp . We already know how to
find the homogeneous solution yh , so we need only worry about yp .
There are two methods for finding a particular solution: the method
of undetermined coefficients and variation of parameters. The former
is generally simpler, but it only applies in a few special cases.
32 / 50
Undetermined coefficients: Main result

Theorem 2.13 – Undetermined coefficients


Consider the scalar linear inhomogeneous equation

an y (n) (t) + . . . + a2 y ′′ (t) + a1 y ′ (t) + a0 y(t) = f (t). (LIE)

Suppose that the coefficients ak are all constant and that the right
hand side f (t) is a linear combination of terms that have the form

tj eλt , tj eat sin(bt), tj eat cos(bt).

Then the solution y(t) satisfies a higher-order homogeneous equation,


so it can itself be expressed as a linear combination of such terms.

One typically uses this theorem to write down an explicit formula for
a particular solution yp . It is easy to predict the terms that appear in
the formula, but their exact coefficients need to be determined.
33 / 50
Undetermined coefficients: General rules
The general rules for finding a particular solution yp are the following.
1 If f (t) contains the term tk eλt , then yp contains the expression
k
X
Aj tj eλt = Ak tk eλt + . . . + A1 teλt + A0 eλt .
j=0

2 If f (t) contains either the term tk eat sin(bt) or the term tk eat cos(bt),
but not necessarily both, then yp contains the expression
k
X k
X
Aj tj eat sin(bt) + Bj tj eat cos(bt).
j=0 j=0

3 If either of the expressions above repeats part of the homogeneous


solution, then it needs to be multiplied by t repeatedly until it no
longer contains terms which appear in the homogeneous solution.
34 / 50
Undetermined coefficients: Some comments

Let us explain the overall approach by looking at the special case

y ′′ (t) − y(t) = f (t).

Our initial guess for a particular solution yp is dictated by the right


hand side f (t). Some typical choices appear in the table below.
f (t) yp
t2 e2t At2 e2t + Bte2t + Ce2t
te2t − e3t Ate2t + Be2t + Ce3t
t3 + 1 At3 + Bt2 + Ct + D
t + cos t At + B + C sin t + D cos t
These choices are dictated by rules 1 and 2 . According to the last
rule, we also need to adjust our initial choice whenever it repeats part
of the homogeneous solution. In this case, we have yh = c1 et + c2 e−t ,
so there is no overlap with yp and thus no need for adjustments.
35 / 50
Undetermined coefficients: Example 1

We use undetermined coefficients in order to solve the equation

y ′′ (t) − 3y ′ (t) + 2y(t) = 2t + 5.

We have y = yh + yp and the homogeneous solution is given by

λ2 − 3λ + 2 = 0 =⇒ (λ − 1)(λ − 2) = 0
=⇒ yh = c1 et + c2 e2t .

To find a particular solution yp , we let yp = At + B. This gives

yp′′ − 3yp′ + 2yp = −3A + 2At + 2B,

so we need to have 2A = 2 and 2B − 3A = 5. It easily follows that

A=1 =⇒ B=4 =⇒ y = c1 et + c2 e2t + t + 4.

36 / 50
Undetermined coefficients: Example 2

We use undetermined coefficients in order to solve the equation

y ′′ (t) + 5y ′ (t) + 6y(t) = 8e2t .

Once again, y = yh + yp and the homogeneous solution is given by

λ2 + 5λ + 6 = 0 =⇒ (λ + 2)(λ + 3) = 0
=⇒ yh = c1 e−2t + c2 e−3t .

To find a particular solution yp , we let yp = Ae2t . This gives

yp′′ + 5yp′ + 6yp = 4Ae2t + 5(2Ae2t ) + 6Ae2t = 20Ae2t ,

so we need to have 20A = 8. In other words, A = 2/5 and thus


2
y = yh + yp = c1 e−2t + c2 e−3t + e2t .
5
37 / 50
Undetermined coefficients: Example 3

We use undetermined coefficients in order to solve the equation

y ′′ (t) + 5y ′ (t) + 6y(t) = sin t.

As in the previous example, the homogeneous solution is given by

yh = c1 e−2t + c2 e−3t .

To find a particular solution, we let yp = A sin t + B cos t and we note


that yp′ = A cos t − B sin t, while yp′′ = −A sin t − B cos t. This gives

yp′′ + 5yp′ + 6yp = 5(A − B) sin t + 5(A + B) cos t,

so we need to have A − B = 1/5 and A + B = 0.


Solving these two equations, we get A = 1/10 and B = −1/10, so
1 1
y = yh + yp = c1 e−2t + c2 e−3t + sin t − cos t.
10 10
38 / 50
Undetermined coefficients: Example 4, page 1

We use undetermined coefficients in order to solve the equation

y ′′ (t) + y(t) = 2 sin t + 4et .

The homogeneous solution yh can be found by noting that

λ2 + 1 = 0 =⇒ λ = ±i =⇒ yh = c1 sin t + c2 cos t.

Let us now worry about the particular solution yp . Based on the right
hand side of the given equation, a natural guess for yp would be

yp = A sin t + B cos t + Cet .

However, this function repeats terms that are already present in yh ,


so we need to adjust these terms and seek a solution of the form

yp = At sin t + Bt cos t + Cet .

39 / 50
Undetermined coefficients: Example 4, page 2

Differentiating the last equation twice, one finds that

yp = At sin t + Bt cos t + Cet ,


yp′ = A sin t + At cos t + B cos t − Bt sin t + Cet ,
yp′′ = 2A cos t − At sin t − 2B sin t − Bt cos t + Cet .

We need to ensure that yp′′ + yp = 2 sin t + 4et and we also have

yp′′ + yp = 2A cos t − 2B sin t + 2Cet

by above. Comparing these two equations, we arrive at the system

2A = 0, −2B = 2, 2C = 4.

This determines the coefficients A, B and C, so the solution is

y = yh + yp = c1 sin t + c2 cos t − t cos t + 2et .


40 / 50
Undetermined coefficients: Example 5, page 1

We use undetermined coefficients in order to solve the equation

y ′′ (t) − 2y ′ (t) + y(t) = 2et + 3t + 4.

The homogeneous solution yh can be found by noting that

λ2 − 2λ + 1 = 0 =⇒ (λ − 1)2 = 0
=⇒ yh = c1 et + c2 tet .

Next, we turn to the particular solution yp . Our initial guess

yp = Aet + Bt + C

repeats part of the homogeneous solution, so this part needs to be


adjusted. Since tet is also repeating part of yh , one needs to take

yp = At2 et + Bt + C.

41 / 50
Undetermined coefficients: Example 5, page 2
Differentiating the last equation twice, one easily finds that
yp = At2 et + Bt + C,
yp′ = 2Atet + At2 et + B,
yp′′ = 2Aet + 4Atet + At2 et ,
yp′′ − 2yp′ + yp = 2Aet + Bt + C − 2B.
On the other hand, we need to ensure that the solution yp satisfies
yp′′ − 2yp′ + yp = 2et + 3t + 4.
Comparing these two expressions, we arrive at the system
2A = 2, B = 3, C − 2B = 4.
This determines the coefficients A, B and C, so the solution is
y = yh + yp = c1 et + c2 tet + t2 et + 3t + 10.
42 / 50
Linear independence and Wronskian
Definition 2.14 – Wronskian
The Wronskian of the functions y1 (t), y2 (t), . . . , yn (t) is defined as

y1 (t) y2 (t) ... yn (t)


 
 y1′ (t) y2′ (t) ... yn′ (t) 
W (t) = det  .. .. .. .. .
 
 . . . . 
(n−1) (n−1) (n−1)
y1 (t) y2 (t) . . . yn (t)

Theorem 2.15 – Linear independence and Wronskian


Suppose that the Wronskian of some scalar functions is not identically
zero. Then these scalar functions are linearly independent.

The converse of this theorem is not true in general. For instance, the
Wronskian of the functions y1 (t) = t2 and y2 (t) = t|t| is identically
zero, but these functions are linearly independent.
43 / 50
Variation of parameters: General case

Theorem 2.16 – Variation of parameters (General case)


Consider the general scalar linear inhomogeneous equation

an (t)y (n) (t) + . . . + a1 (t)y ′ (t) + a0 (t)y(t) = f (t). (LIE)

Suppose that y1 (t), y2 (t), . . . , yn (t) are linearly independent solutions


of the associated homogeneous equation. A particular solution of (LIE)
is then yp (t) = c1 (t)y1 (t) + c2 (t)y2 (t) + . . . + cn (t)yn (t), where the
coefficients ck (t) are determined using the equation
−1 
c′1 (t) y1 (t) y2 (t) ... yn (t)
   
0
 c′ (t)   y1′ (t) y2′ (t) ... yn′ (t)   .. 
 2   .
 ..  =  .. .. .. .. .
  
.
 
 .   . . .   0 
′ (n−1) (n−1) (n−1)
cn (t) y1 (t) y2 (t) . . . yn (t) f (t)/an (t)

44 / 50
Variation of parameters: Second-order case

Theorem 2.17 – Variation of parameters (Second-order case)


Suppose that y1 (t), y2 (t) are linearly independent solutions of

a(t)y ′′ (t) + b(t)y ′ (t) + c(t)y(t) = 0 (LHE)

and consider the corresponding inhomogeneous equation

a(t)y ′′ (t) + b(t)y ′ (t) + c(t)y(t) = f (t). (LIE)

A particular solution of (LIE) is then provided by the formula

y2 (t)f (t) y1 (t)f (t)


Z Z
yp (t) = −y1 (t) dt + y2 (t) dt,
a(t)W (t) a(t)W (t)

where W (t) = y1 (t)y2′ (t) − y1′ (t)y2 (t) is the Wronskian of y1 and y2 .

45 / 50
Variation of parameters: Example
We use variation of parameters to find a particular solution of
y ′′ (t) + y(t) = sec t.
The solution of the associated homogeneous equation is given by
λ2 + 1 = 0 =⇒ λ = ±i =⇒ yh = c1 sin t + c2 cos t.
Letting y1 (t) = sin t and y2 (t) = cos t, we now find that
 
sin t cos t
W (t) = det = − sin2 t − cos2 t = −1.
cos t − sin t
According to the previous theorem, a particular solution is thus
Z Z
yp (t) = sin t cos t · sec t dt − cos t sin t · sec t dt
cos t sin t
Z Z
= sin t dt − cos t dt
cos t cos t
= t sin t + (cos t) log(cos t).
46 / 50
Reduction of order

Suppose that we know one solution y1 of the homogeneous equation

an (t)y (n) (t) + . . . + a1 (t)y ′ (t) + a0 (t)y(t) = 0 (LHE)

and that we need to solve the associated inhomogeneous equation

an (t)z (n) (t) + . . . + a1 (t)z ′ (t) + a0 (t)z(t) = f (t). (LIE)

Then the substitution z = y1 v gives rise to an equation for v which


involves the derivatives of v but not v itself. Such an equation is a
lower-order equation for v ′ , so it is generally easier to solve.
This approach can be used for any linear inhomogeneous equation. In
particular, we are not assuming that the coefficients ak are constant.
When it comes to second-order equations, one may use this approach
to find all solutions of (LIE), if just one solution of (LHE) is known.

47 / 50
Reduction of order: Example, page 1

It is easy to check that y1 (t) = t2 satisfies the homogeneous equation

t2 y ′′ (t) − 2ty ′ (t) + 2y(t) = 0.

We now use this fact to solve the inhomogeneous equation



t2 z ′′ (t) − 2tz ′ (t) + 2z(t) = t t, t > 0.

First of all, we change variables by letting z = y1 v. This gives

z = t2 v, z ′ = 2tv + t2 v ′ , z ′′ = 2v + 4tv ′ + t2 v ′′

and the inhomogeneous equation that needs to be solved becomes



t t = t2 z ′′ − 2tz ′ + 2z
2✟
2t✟
=✟ v + 4t3 v ′ + t4 v ′′ − ✟ 2✟
4t✟v − 2t3 v ′ + ✟ 2✟
2t✟v
= t4 v ′′ + 2t3 v ′ .

48 / 50
Reduction of order: Example, page 2

Setting w = v ′ for convenience, we now arrive at the equation



t4 w′ + 2t3 w = t t =⇒ w′ + 2t−1 w = t−5/2 .

This is a first-order linear equation with integrating factor


Z 
µ = exp 2t−1 dt = e2 log t+C = Kt2 .

Letting K = 1 for simplicity, we may finally conclude that

(µw)′ = t−1/2 =⇒ µw = 2t1/2 + c1


=⇒ w = 2t−3/2 + c1 t−2 .

Since v ′ = w and z = t2 v by above, this also implies that



v = −4t−1/2 − c1 t−1 + c2 =⇒ z = −4t t − c1 t + c2 t2 .
49 / 50
Summary of available methods

Homogeneous systems: y ′ (t) = A(t)y(t).


−→ Eigenvector method: if A(t) is constant and diagonalisable.
−→ Matrix exponential: if A(t) is constant.
−→ Solvable equations: if A(t) is lower/upper triangular.
Inhomogeneous systems: y ′ (t) = A(t)y(t) + b(t).
−→ Variation of parameters: this method applies in all cases.
Pn (k) (t) = 0.
Homogeneous scalar equations: k=0 ak (t)y
−→ Characteristic equation: if the coefficients ak are constant.
−→ Reduction of order: if one solution is already known.
Pn (k) (t) = f (t).
Inhomogeneous scalar equations: k=0 ak (t)y
−→ Undetermined coefficients: if ak are constant and f is simple.
−→ Variation of parameters: this method applies in all cases.
−→ Reduction of order: if one homogeneous solution is known.
50 / 50

You might also like