ordinary differential equations - Department of Mathematics ...
ordinary differential equations - Department of Mathematics ...
ordinary differential equations - Department of Mathematics ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
ORDINARY DIFFERENTIAL EQUATIONS<br />
GABRIEL NAGY<br />
<strong>Mathematics</strong> <strong>Department</strong>,<br />
Michigan State University,<br />
East Lansing, MI, 48824.<br />
JULY 2, 2013<br />
Summary. These notes try to be an introduction to <strong>ordinary</strong> <strong>differential</strong> <strong>equations</strong>. We<br />
describe a collection <strong>of</strong> methods and techniques used to find solutions to several types<br />
<strong>of</strong> <strong>differential</strong> <strong>equations</strong>. These <strong>equations</strong> include first order scalar <strong>equations</strong>, second<br />
order linear <strong>equations</strong>, and systems <strong>of</strong> linear <strong>equations</strong>. There is one chapter dedicated<br />
to power series methods to find solutions to variable coefficients second order linear<br />
<strong>equations</strong>. There is another chapter dedicated to Laplace transform methods to find<br />
solutions to constant coefficients <strong>equations</strong> with generalized source functions. The last<br />
chapter provides a brief introduction to boundary value problems and the Fourier transform.<br />
We end the chapter showing how to use these ideas to solve an initial boundary<br />
value problem for a particular partial <strong>differential</strong> equation, the heat equation.<br />
Date: July 2, 2013, gnagy@math.msu.edu.<br />
1
G. NAGY – ODE July 2, 2013 I<br />
Table <strong>of</strong> Contents<br />
Chapter 1. First order <strong>equations</strong> 1<br />
1.1. Linear constant coefficients <strong>equations</strong> 1<br />
1.1.1. Overview <strong>of</strong> <strong>differential</strong> <strong>equations</strong> 1<br />
1.1.2. Linear <strong>equations</strong> 2<br />
1.1.3. Linear <strong>equations</strong> with constant coefficients 3<br />
1.1.4. The Initial Value Problem 6<br />
1.1.5. Exercises 7<br />
1.2. Linear variable coefficients <strong>equations</strong> 8<br />
1.2.1. Linear <strong>equations</strong> with variable coefficients 8<br />
1.2.2. The Initial Value Problem 10<br />
1.2.3. The Bernoulli equation 13<br />
1.2.4. Exercises 16<br />
1.3. Separable <strong>equations</strong> 17<br />
1.3.1. Separable <strong>equations</strong> 17<br />
1.3.2. Euler Homogeneous <strong>equations</strong> 22<br />
1.3.3. Exercises 25<br />
1.4. Exact <strong>equations</strong> 26<br />
1.4.1. Exact <strong>differential</strong> <strong>equations</strong> 26<br />
1.4.2. The Poincaré Lemma 27<br />
1.4.3. The integrating factor method 30<br />
1.4.4. Exercises 34<br />
1.5. Applications 35<br />
1.5.1. Radioactive decay 35<br />
1.5.2. Newton’s cooling law 36<br />
1.5.3. Salt in a water tank 36<br />
1.5.4. The Main <strong>equations</strong> 37<br />
1.5.5. Predictions in particular cases 38<br />
1.5.6. Exercises 42<br />
1.6. Non-linear <strong>equations</strong> 43<br />
1.6.1. On solutions <strong>of</strong> linear <strong>differential</strong> <strong>equations</strong> 43<br />
1.6.2. On solutions <strong>of</strong> non-linear <strong>differential</strong> <strong>equations</strong> 44<br />
1.6.3. Direction Fields 47<br />
1.6.4. Exercises 50<br />
Chapter 2. Second order linear <strong>equations</strong> 51<br />
2.1. Variable coefficients 51<br />
2.1.1. Main definitions and properties 51<br />
2.1.2. Wronskian and linear dependence 52<br />
2.1.3. Fundamental and general solutions 54<br />
2.1.4. The Abel Theorem 56<br />
2.1.5. Special Second order nonlinear <strong>equations</strong> 57<br />
2.1.6. Exercises 60<br />
2.2. Constant coefficients 61<br />
2.2.1. Solution formulas for constant coefficients 61<br />
2.2.2. Different real-valued roots 63<br />
2.2.3. Exercises 65<br />
2.3. Complex roots 66
II G. NAGY – ODE july 2, 2013<br />
2.3.1. Real-valued fundamental solutions 66<br />
2.3.2. The RLC electric circuit 68<br />
2.3.3. Exercises 70<br />
2.4. Repeated roots 71<br />
2.4.1. Fundamental solutions for repeated roots 71<br />
2.4.2. Reduction <strong>of</strong> order method 73<br />
2.4.3. Exercises 75<br />
2.5. Undetermined coefficients 76<br />
2.5.1. Operator notation 76<br />
2.5.2. The undetermined coefficients method 77<br />
2.5.3. Exercises 80<br />
2.6. Variation <strong>of</strong> parameters 81<br />
2.6.1. The variation <strong>of</strong> parameters method 81<br />
2.6.2. Exercises 85<br />
Chapter 3. Power series solutions 86<br />
3.1. Regular points 86<br />
3.1.1. Review <strong>of</strong> power series 86<br />
3.1.2. Equations with regular points 90<br />
3.1.3. Exercises 96<br />
3.2. The Euler equation 97<br />
3.2.1. The main result 97<br />
3.2.2. The complex-valued roots 99<br />
3.2.3. Exercises 101<br />
3.3. Regular-singular points 102<br />
3.3.1. Finding regular-singular points 102<br />
3.3.2. Solving <strong>equations</strong> with regular-singular points 104<br />
3.3.3. Exercises 108<br />
Chapter 4. The Laplace Transform 109<br />
4.1. Definition and properties 109<br />
4.1.1. Review <strong>of</strong> improper integrals 109<br />
4.1.2. Definition <strong>of</strong> the Laplace Transform 110<br />
4.1.3. Properties <strong>of</strong> the Laplace Transform 113<br />
4.1.4. Exercises 115<br />
4.2. The initial value problem 116<br />
4.2.1. Summary <strong>of</strong> the Laplace Transform method 116<br />
4.2.2. Exercises 121<br />
4.3. Discontinuous sources 122<br />
4.3.1. The Step function 122<br />
4.3.2. The Laplace Transform <strong>of</strong> Step functions 123<br />
4.3.3. The Laplace Transform <strong>of</strong> translations 124<br />
4.3.4. Differential <strong>equations</strong> with discontinuous sources 127<br />
4.3.5. Exercises 132<br />
4.4. Generalized sources 133<br />
4.4.1. The Dirac delta generalized function 133<br />
4.4.2. The Laplace Transform <strong>of</strong> the Dirac delta 135<br />
4.4.3. The fundamental solution 136<br />
4.4.4. Exercises 139
G. NAGY – ODE July 2, 2013 III<br />
4.5. Convolution solutions 140<br />
4.5.1. The convolution <strong>of</strong> two functions 140<br />
4.5.2. The Laplace Transform <strong>of</strong> a convolution 141<br />
4.5.3. Impulse response solutions 143<br />
4.5.4. The solution decomposition Theorem 144<br />
4.5.5. Exercises 146<br />
Chapter 5. Systems <strong>of</strong> <strong>differential</strong> <strong>equations</strong> 147<br />
5.1. Introduction 147<br />
5.1.1. Systems <strong>of</strong> <strong>differential</strong> <strong>equations</strong> 147<br />
5.1.2. Reduction to first order systems 150<br />
5.1.3. The Picard-Lindelöf Theorem 151<br />
5.1.4. Exercises 153<br />
5.2. Systems <strong>of</strong> algebraic <strong>equations</strong> 154<br />
5.2.1. Linear algebraic systems 154<br />
5.2.2. Gauss elimination operations 158<br />
5.2.3. Linearly dependence 161<br />
5.2.4. Exercises 162<br />
5.3. Matrix algebra 163<br />
5.3.1. A matrix is a function 163<br />
5.3.2. Matrix operations 164<br />
5.3.3. The inverse matrix 168<br />
5.3.4. Determinants 170<br />
5.3.5. Exercises 172<br />
5.4. Linear <strong>differential</strong> Systems 173<br />
5.4.1. Matrix-vector notation 173<br />
5.4.2. Homogeneous systems 175<br />
5.4.3. Abel’s Theorem for systems 177<br />
5.4.4. Existence <strong>of</strong> fundamental solutions 179<br />
5.4.5. Exercises 182<br />
5.5. Diagonalizable matrices 183<br />
5.5.1. Eigenvalues and eigenvectors 183<br />
5.5.2. Diagonalizable matrices 189<br />
5.5.3. The exponential <strong>of</strong> a matrix 193<br />
5.5.4. Exercises 196<br />
5.6. Constant coefficient <strong>differential</strong> systems 197<br />
5.6.1. Diagonalizable homogeneous systems 197<br />
5.6.2. The Fundamental Solutions 200<br />
5.6.3. Diagonalizable systems with continuous sources 203<br />
5.6.4. Exercises 206<br />
5.7. 2 × 2 Real distinct eigenvalues 207<br />
5.7.1. Phase diagrams 207<br />
5.7.2. Case 0 < λ2 < λ1 208<br />
5.7.3. Case λ 2 < 0 < λ 1 209<br />
5.7.4. Case λ 2 < λ 1 < 0 210<br />
5.7.5. Exercises 212<br />
5.8. 2 × 2 Complex eigenvalues 213<br />
5.8.1. Complex conjugate pairs 213<br />
5.8.2. Qualitative analysis 215<br />
5.8.3. Exercises 216
IV G. NAGY – ODE july 2, 2013<br />
5.9. 2 × 2 Repeated eigenvalues 217<br />
5.9.1. Diagonalizable systems 217<br />
5.9.2. Non-diagonalizable systems 217<br />
5.9.3. Exercises 221<br />
Chapter 6. Boundary value problems 222<br />
6.1. Eigenvalue-eigenfunction problems 222<br />
6.1.1. Comparison: IVP and BVP 222<br />
6.1.2. Eigenvalue-eigenfunction problems 226<br />
6.1.3. Exercises 231<br />
6.2. Overview <strong>of</strong> Fourier series 232<br />
6.2.1. Origins <strong>of</strong> Fourier series 232<br />
6.2.2. Fourier series 234<br />
6.2.3. Even and odd functions 238<br />
6.2.4. Exercises 240<br />
6.3. Application: The heat equation 241<br />
6.3.1. The initial-boundary value problem 241<br />
6.3.2. Solution <strong>of</strong> the initial-boundary value problem 242<br />
6.3.3. Exercises 245<br />
Chapter 7. Appendix 246<br />
7.1. Review exercises 246<br />
7.2. Practice Exams 252<br />
7.3. Answers to exercises 254<br />
References 263
G. NAGY – ODE July 2, 2013 1<br />
Chapter 1. First order <strong>equations</strong><br />
A first order <strong>differential</strong> <strong>equations</strong> is an equation where the unknown function and only its<br />
first derivative appear in the equation. This Chapter is a collection <strong>of</strong> methods or recipes to<br />
get solutions to certain types <strong>of</strong> <strong>differential</strong> <strong>equations</strong>, including linear <strong>equations</strong>, separable<br />
<strong>equations</strong>, Euler homogeneous <strong>equations</strong>, and exact <strong>equations</strong>. The methods we present<br />
here were the first attempts done in the eighteen and nineteen centuries to obtain solutions<br />
to these <strong>differential</strong> <strong>equations</strong>.<br />
1.1. Linear constant coefficients <strong>equations</strong><br />
1.1.1. Overview <strong>of</strong> <strong>differential</strong> <strong>equations</strong>. A <strong>differential</strong> equation is an equation where<br />
the unknown is a function and both the function and its derivatives may appear in the<br />
equation. Differential <strong>equations</strong> are an essential part in the description <strong>of</strong> nature done by<br />
physics. The central part <strong>of</strong> many physical theories is a <strong>differential</strong> equation: Newton’s and<br />
Lagrange <strong>equations</strong> for classical mechanics, Maxwell’s <strong>equations</strong> for classical electromagnetism,<br />
Schrödinger’s equation for quantum mechanics, Einstein’s equation for the general<br />
theory <strong>of</strong> gravitation. We highlight the following examples <strong>of</strong> <strong>differential</strong> <strong>equations</strong> in a<br />
more precise way:<br />
(a) Newton’s second law <strong>of</strong> motion for a single particle. The unknown is a vector-valued<br />
function x : R → R 3 , where the vector x(t) represents the position <strong>of</strong> a particle at the<br />
time t. The <strong>differential</strong> equation is<br />
m d2x (t) = f (t, x(t)),<br />
dt2 where m is a positive constant representing the mass <strong>of</strong> the particle and f : R×R3 → R3 is the force acting on the particle, which depends on the time t and the position in space<br />
x. This is the well-known “mass times acceleration equals force” law <strong>of</strong> motion.<br />
(b) The time decay <strong>of</strong> a radioactive substance. The unknown is a scalar-valued function<br />
u : R → R, where u(t) represents the concentration <strong>of</strong> the radioactive substance at the<br />
time t. The <strong>differential</strong> equation is<br />
du<br />
(t) = −k u(t),<br />
dt<br />
where k is a positive constant. The equation says that the higher the material concentration<br />
the faster it decays.<br />
(c) The wave equation, which describes waves propagating in a media. An example is sound,<br />
where pressure waves propagate in the air. The unknown is a scalar-valued function <strong>of</strong><br />
two variables u : R × R3 → R, where u(t, x) represents a perturbation in the air density<br />
at the time t and point x = (x, y, z) in space. (We used the same notation for vectors<br />
and points, although they are different type <strong>of</strong> objects.) The equation is<br />
∂ttu(t, x) = v 2 ∂xxu(t, x) + ∂yyu(t, x) + ∂zzu(t, x) ,<br />
where v is a positive constant describing the wave speed, and we have used the notation<br />
∂ to mean partial derivative.<br />
(d) The heat conduction equation, which describes the variation <strong>of</strong> temperature in a solid<br />
material. The unknown is a scalar-valued function u : R × R 3 → R, where u(t, x)<br />
represents the temperature at time t and the point x = (x, y, z) in the solid. The<br />
equation is<br />
∂tu(t, x) = k ∂xxu(t, x) + ∂yyu(t, x) + ∂zzu(t, x) ,<br />
where k is a positive constant representing thermal properties <strong>of</strong> the material.
2 G. NAGY – ODE july 2, 2013<br />
The <strong>equations</strong> in examples (a) and (b) are called <strong>ordinary</strong> <strong>differential</strong> <strong>equations</strong> (ODE),<br />
since the unknown function depends on a single independent variable, t in these examples.<br />
The <strong>equations</strong> in examples (c) and (d) are called partial <strong>differential</strong> <strong>equations</strong> (PDE), since<br />
the unknown function depends on two or more independent variables and their partial<br />
derivatives appear in the <strong>equations</strong>.<br />
The order <strong>of</strong> a <strong>differential</strong> equation is the highest derivative order that appears in the<br />
equation. Newton’s equation in Example (a) is second order, the time decay equation in<br />
Example (b) is first order, the wave equation in Example (c) and the heat equation in<br />
Example (d) are second order. These last two cases are PDE, and in these cases one usually<br />
gives extra information, saying that the wave equation is second order both in space and<br />
time derivatives but the heat equation is second order in space derivatives and first order<br />
in time derivatives.<br />
1.1.2. Linear <strong>equations</strong>. We start with a more precise definition <strong>of</strong> the type <strong>of</strong> <strong>differential</strong><br />
equation we are about to study. When there is no risk <strong>of</strong> confusion we usually denote with<br />
a prime a derivative <strong>of</strong> a single variable function. So, the derivative <strong>of</strong> a function y : R → R<br />
is denoted as<br />
dy<br />
dt (t) = y′ (t).<br />
Definition 1.1.1. Given a function f : R 2 → R, a first order ODE in the unknown<br />
function y : R → R is the equation<br />
y ′ (t) = f(t, y(t)). (1.1.1)<br />
The ODE in Eq. (1.1.1) is called linear iff the function with values f(t, y) is linear on its<br />
second argument y, that is, there exist functions a, b : R → R such that<br />
y ′ (t) = a(t) y(t) + b(t), f(t, y) = a(t) y + b(t). (1.1.2)<br />
Sometimes in the literature there is a different sign convention for Eq. (1.1.2). For<br />
example, Boyce-DiPrima [1] writes it as y ′ = −a y + b. The sign choice in front <strong>of</strong> function a<br />
is just a convention. Some people like the negative sign, because later on, when they write<br />
the equation as y ′ + a y = b, they get a plus sign on the left-hand side. In any case, we stick<br />
here to the convention y ′ = ay + b.<br />
A linear ODE is called <strong>of</strong> constant coefficients iff the functions a and b in Eq. (1.1.2) are<br />
both constants. Otherwise, the linear ODE is called <strong>of</strong> variable coefficients.<br />
Example 1.1.1:<br />
(a) An example <strong>of</strong> a first order linear ODE is the equation<br />
y ′ (t) = 2 y(t) + 3.<br />
In this case, the right-hand side is given by the function f(t, y) = 2y + 3, where we can<br />
see that a(t) = 2 and b(t) = 3. Since these coefficients do not depend on t, this is a<br />
constant coefficients equation.<br />
(b) Another example <strong>of</strong> a first order linear ODE is the equation<br />
y ′ (t) = − 2<br />
y(t) + 4t.<br />
t<br />
In this case, the right-hand side is given by the function f(t, y) = −2y/t + 4t, where<br />
a(t) = −2/t and b(t) = 4t. Since the coefficients are non-constant functions <strong>of</strong> t, this is<br />
a variable coefficients equation.<br />
⊳
G. NAGY – ODE July 2, 2013 3<br />
A function y : D ⊂ R → R is solution <strong>of</strong> the <strong>differential</strong> equation in (1.1.1) iff the equation<br />
is satisfied for all values <strong>of</strong> the independent variable t in the domain D <strong>of</strong> the function y.<br />
Example 1.1.2: Show that y(t) = e 2t − 3<br />
2 is solution <strong>of</strong> the ODE y′ (t) = 2 y(t) + 3.<br />
Solution: On the one hand we compute y ′ (t) = 2e2t . On the other hand we compute<br />
<br />
2 y(t) + 3 = 2 e 2t − 3<br />
<br />
+ 3 = 2e<br />
2<br />
2t .<br />
We conclude that y ′ (t) = 2 y(t) + 3 for all t ∈ R. ⊳<br />
1.1.3. Linear <strong>equations</strong> with constant coefficients. It is not difficult to obtain all<br />
solutions to first order linear ODE in the case that the equation has constant coefficients.<br />
One multiplies the <strong>differential</strong> equation by a particularly chosen non-zero function, called the<br />
integrating factor. The integrating factor has the following property: The whole equation<br />
is transformed into a total derivative <strong>of</strong> a function, called potential function. Integrating<br />
the <strong>differential</strong> equation is now trivial. The function solution <strong>of</strong> the <strong>differential</strong> equation is<br />
simple to obtain inverting the potential function. This whole idea is called the integrating<br />
factor method. In the next Section we generalize this idea to find solutions to variable<br />
coefficients linear ODE, and in Section 1.4 we generalize this idea to certain non-linear<br />
<strong>differential</strong> <strong>equations</strong>.<br />
Theorem 1.1.2 (Constant coefficients). If the constants a, b ∈ R satisfy that a = 0,<br />
then linear <strong>differential</strong> equation<br />
y ′ (t) = a y(t) + b (1.1.3)<br />
has infinitely many solutions and every solution can be labeled by c ∈ R as follows,<br />
y(t) = c e at − b<br />
. (1.1.4)<br />
a<br />
Remark: Theorem 1.1.2 says that Eq. (1.1.3) has infinitely many solutions, one solution for<br />
each value <strong>of</strong> the constant c, which is not determined by the equation. This is reasonable.<br />
Since the <strong>differential</strong> equation contains one derivative <strong>of</strong> the unknown function y, finding<br />
a solution <strong>of</strong> the <strong>differential</strong> equation requires to compute an integral. Every indefinite<br />
integration introduces an integration constant. This is the origin <strong>of</strong> the constant c above.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 1.1.2: The integrating factor method is the key idea in the pro<strong>of</strong> <strong>of</strong><br />
Theorem 1.1.2. We first write down the <strong>differential</strong> equation as<br />
y ′ (t) − a y(t) = b,<br />
and then multiply the <strong>differential</strong> equation by the exponential e −at , where the exponent<br />
is the constant coefficient a in the <strong>differential</strong> equation multiplied by t. This exponential<br />
function is an integrating factor for the <strong>differential</strong> equation. The reason to choose this<br />
particular function, e −at , is explained in Lemma 1.1.3, below. The result is<br />
y ′ (t) − a y(t) e −at = b e −at ⇔ e −at y ′ (t) − a e −at y(t) = b e −at .<br />
This exponential is chosen because <strong>of</strong> the following property,<br />
−a e −at = e −ay ′ .<br />
Introducing this property into the <strong>differential</strong> equation we get<br />
e −at y ′ (t) + e −at ′ y(t) = b e −at .<br />
Now the product rule for derivatives implies that the left-hand side above is a total derivative,<br />
e −at y(t) ′ = b e −at .
4 G. NAGY – ODE july 2, 2013<br />
At this point we can go one step further, expresing the right-hand side in the <strong>differential</strong><br />
equation as b e −at <br />
= − b<br />
a e−at ′<br />
. We obtain<br />
<br />
e −at y(t) + b<br />
a e−at ′<br />
<br />
= 0 ⇔ y(t) + b<br />
<br />
e<br />
a<br />
−at ′<br />
= 0.<br />
Hence, the whole <strong>differential</strong> equation is a total derivative. The whole <strong>differential</strong> equation<br />
is the total derivative <strong>of</strong> the function,<br />
<br />
ψ(t, y(t)) = y(t) + b<br />
<br />
e<br />
a<br />
−at ,<br />
called potential function. The equation now has the form<br />
dψ<br />
(t, y(t)) = 0.<br />
dt<br />
It is trivial to integrate the <strong>differential</strong> equation written in terms <strong>of</strong> the potential function,<br />
<br />
ψ(t, y(t)) = c ⇔ y(t) + b<br />
<br />
e<br />
a<br />
−at = c.<br />
Simple algebraic manipulations imply that<br />
y(t) = c e at − b<br />
a .<br />
This establishes the Theorem. <br />
Remarks:<br />
(a) The Pro<strong>of</strong> <strong>of</strong> Theorem 1.1.2 starts multiplying the <strong>differential</strong> equation by the exponential<br />
function e−at . People could argue that it is not clear where this idea comes from.<br />
In Lemma 1.1.3 we show that only functions proportional to the exponential e−at have<br />
the property needed to be an integrating factor for the <strong>differential</strong> equation. The pro<strong>of</strong><br />
<strong>of</strong> Lemma 1.1.3 starts multiplying the <strong>differential</strong> equation by a function µ, and then<br />
one finds that µ(t) = c e−at .<br />
(b) Since the function µ is used to multiply the original <strong>differential</strong> equation, we can freely<br />
choose the function µ with c = 1. This is the choice made in the pro<strong>of</strong> <strong>of</strong> Theorem 1.1.2.<br />
(c) It is important we understand the origin <strong>of</strong> the integrating factor e−at in order to extend<br />
results from constant coefficients <strong>equations</strong> to variable coefficients <strong>equations</strong>.<br />
The following Lemma states that those functions proportional to e−at are integrating factors<br />
for the <strong>differential</strong> equation in (1.1.3).<br />
Lemma 1.1.3 (Integrating Factor). Given any differentiable function y and constant a,<br />
every function µ satisfying<br />
(y ′ − ay)µ = (yµ) ′ ,<br />
must be given by the expression below, for any c ∈ R,<br />
µ(t) = c e −at .<br />
Pro<strong>of</strong> <strong>of</strong> Lemma 1.1.3: Multiply Eq. (1.1.3) by a non-vanishing otherwise arbitrary<br />
function with values µ(t), and order the terms in the equation as follows<br />
µ(y ′ − a y) = b µ. (1.1.5)<br />
The key idea <strong>of</strong> the pro<strong>of</strong> is to choose the function µ such that the following equation holds<br />
µ(y ′ − a y) = µy ′ . (1.1.6)<br />
Eq. (1.1.6), this is an equation for µ, since it can be rewritten as follows,<br />
µ y ′ − a µy = µ ′ y + µ y ′ ⇔ −a µy = µ ′ y ⇔ −a µ = µ ′ ,
G. NAGY – ODE July 2, 2013 5<br />
that is, the function y does not appear in Eq. (1.1.6). The equation on the far right above<br />
can be solved for µ as follows:<br />
Integrating in the equation above we obtain<br />
µ ′<br />
µ = −a ⇔ ln(µ) ′ = −a.<br />
ln(µ) = −at + c0 ⇔ e ln(µ) = e −at+c0 = e −at e c0 ,<br />
where c0 is an arbitrary constant. Denoting c = e c0, the integrating factor is the nonvanishing<br />
function<br />
µ(t) = c e −at .<br />
This establishes the Lemma. <br />
We now solve the problem in Example 1.1.3 following the pro<strong>of</strong> <strong>of</strong> Theorem 1.1.2.<br />
Example 1.1.3: Find all solutions to the constant coefficient equation<br />
Solution: Write down the equation in (1.1.7) as follows,<br />
y ′ = 2y + 3 (1.1.7)<br />
y ′ − 2y = 3.<br />
Multiply this equation by the exponential e −2t , that is,<br />
e −2t y ′ − 2 e −2t y = 3 e −2t ⇔ e −2t y ′ + e −2t ′ y = 3 e −2t .<br />
The equation on the far right above is<br />
(e −2t y) ′ = 3 e −2t .<br />
Integrating on both sides get<br />
<br />
(e −2t y) ′ <br />
dt = 3 e −2t dt + c,<br />
e −2t y = − 3<br />
2 e−2t + c.<br />
Multiplying the whole equation by e 2t ,<br />
y(t) = c e 2t − 3<br />
, c ∈ R.<br />
2<br />
⊳<br />
y<br />
0<br />
−3/2<br />
c > 0<br />
t<br />
c = 0<br />
c < 0<br />
Figure 1. The graph <strong>of</strong> a<br />
few solutions to Eq. (1.1.7)<br />
for different constants c.<br />
We now solve same problem above but using the formula in Theorem 1.1.2.<br />
Example 1.1.4: Find all solutions to the constant coefficient equation<br />
y ′ = 2y + 3 (1.1.8)<br />
Solution: The equation above is the particular case <strong>of</strong> Eq. (1.1.3) given by a = 2 and<br />
b = 3. Therefore, using these values in the expression for the solution given in Eq. (1.1.4)<br />
we obtain<br />
y(t) = ce 2t − 3<br />
2 .<br />
⊳
6 G. NAGY – ODE july 2, 2013<br />
1.1.4. The Initial Value Problem. Sometimes in physics one is not interested in all<br />
solutions to a <strong>differential</strong> equation, but only in those solutions satisfying an extra condition.<br />
For example, in the case <strong>of</strong> Newton’s second law <strong>of</strong> motion for a point particle, one could<br />
be interested only in those solutions satisfying an extra condition: At an initial time the<br />
particle must be at a specified initial position. Such condition is called an initial condition,<br />
and it selects a subset <strong>of</strong> solutions <strong>of</strong> the <strong>differential</strong> equation. An initial value problem<br />
means to find a solution to both a <strong>differential</strong> equation and an initial condition.<br />
Definition 1.1.4. The initial value problem (IVP) for a constant coefficients first order<br />
linear ODE is the following: Given a, b, t 0, y 0 ∈ R, find a solution y : R → R <strong>of</strong> the problem<br />
y ′ = a y + b, y(t0) = y0. (1.1.9)<br />
The second equation in (1.1.9) is called the initial condition <strong>of</strong> the problem. Although<br />
the <strong>differential</strong> equation in (1.1.9) has infinitely many solutions, the associated initial value<br />
problem has a unique solution.<br />
Theorem 1.1.5 (Constant coefficients IVP). Given constants a, b, t0, y0 ∈ R, with a =<br />
0, the initial value problem in (1.1.9) has the unique solution<br />
<br />
y(t) = y0 + b<br />
<br />
e<br />
a<br />
a(t−t0) b<br />
− . (1.1.10)<br />
a<br />
To prove this result we use Theorem 1.1.2 to write down the general solution <strong>of</strong> the<br />
<strong>differential</strong> equation; the initial condition to fixes the value <strong>of</strong> the integration constant c.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 1.1.5: The general solution <strong>of</strong> the <strong>differential</strong> equation in (1.1.9) is<br />
given in Eq. (1.1.4), that is,<br />
y(t) = c e at − b<br />
a .<br />
The initial condition determines the value <strong>of</strong> the constant c, as follows<br />
y0 = y(t0) = c e at0<br />
b<br />
−<br />
a<br />
⇔<br />
<br />
c = y0 + b<br />
<br />
e<br />
a<br />
−at0 .<br />
Introduce this expression for the constant c into the <strong>differential</strong> equation in Eq. (1.1.9),<br />
<br />
y(t) = y0 + b<br />
<br />
e<br />
a<br />
a(t−t0) b<br />
−<br />
a .<br />
This establishes the Theorem. <br />
Example 1.1.5: Find the unique solution <strong>of</strong> the initial value problem<br />
y ′ = 2y + 3, y(0) = 1. (1.1.11)<br />
Solution: Every solution <strong>of</strong> the <strong>differential</strong> equation is given by y(t) = ce2t − (3/2), where<br />
c is an arbitrary constant. The initial condition in Eq. (1.1.11) determines the value <strong>of</strong> c,<br />
1 = y(0) = c − 3<br />
2<br />
⇔<br />
5<br />
c =<br />
2 .<br />
Then, the unique solution to the IVP above is<br />
y(t) = 5<br />
2 e2t − 3<br />
2 .<br />
⊳
1.1.5. Exercises.<br />
1.1.1.- Verify that y(t) = (t + 2) e 2t is solution<br />
<strong>of</strong> the IVP<br />
y ′ = 2y + e 2t , y(0) = 2.<br />
1.1.2.- Find the general solution <strong>of</strong><br />
y ′ = −4y + 2<br />
1.1.3.- Find the solution <strong>of</strong> the IVP<br />
y ′ = −4y + 2, y(0) = 5.<br />
1.1.4.- Find the solution <strong>of</strong> the IVP<br />
dy<br />
(t) = 3 y(t) − 2, y(1) = 1.<br />
dt<br />
G. NAGY – ODE July 2, 2013 7<br />
1.1.5.- Express the <strong>differential</strong> equation<br />
y ′ (t) = 6 y(t) + 1 (1.1.12)<br />
as a total derivative <strong>of</strong> a potential function<br />
ψ(t, y(t)), that is, find ψ satisfying<br />
dψ<br />
= 0<br />
dt<br />
for every y solution <strong>of</strong> Eq. (1.1.12).<br />
Then, find the general solution y <strong>of</strong> the<br />
equation above.<br />
1.1.6.- Find the solution <strong>of</strong> the IVP<br />
y ′ = 6 y + 1, y(0) = 1.
8 G. NAGY – ODE july 2, 2013<br />
1.2. Linear variable coefficients <strong>equations</strong><br />
In Section 1.1 we found an expression for the general solution <strong>of</strong> a first order, linear,<br />
constant coefficient <strong>differential</strong> equation, Theorem 1.1.2. We also learned that the initial<br />
value problem for such equation has a unique solution, Theorem 1.1.5. In the first part <strong>of</strong><br />
this Section we generalize these results to variable coefficients <strong>equations</strong>, that is, <strong>equations</strong><br />
<strong>of</strong> the form<br />
y ′ (t) = a(t) y(t) + b(t),<br />
where a, b : (t 1, t 2) → R are continuous functions. We will show that the integrating<br />
factor method studied in the last Section can be generalized to variable coefficients linear<br />
<strong>equations</strong>. In the second part <strong>of</strong> this Section we introduce the Bernoulli equation, which is a<br />
non-linear <strong>differential</strong> equation. This non-linear equation has a particular property: It can<br />
be transformed into a linear equation by and appropriate change in the unknown function.<br />
One then solves the linear equation for the changed function using the integrating factor<br />
method. Finally one transforms back the changed function into the original function.<br />
1.2.1. Linear <strong>equations</strong> with variable coefficients. We start generalizing the result in<br />
Theorem 1.1.2 to variable coefficinents <strong>equations</strong>.<br />
Theorem 1.2.1 (Variable coefficients). If the functions a, b : (t 1, t 2) → R are continuous,<br />
where t1 < t2, then the <strong>differential</strong> equation<br />
y ′ = a(t) y + b(t), (1.2.1)<br />
has infinitely many solutions and every solution can be labeled by c ∈ R as follows<br />
y(t) = c e A(t) + e A(t)<br />
<br />
e −A(t) <br />
b(t) dt, (1.2.2)<br />
where we introduced the function A(t) = a(t) dt, any primitive <strong>of</strong> the function a.<br />
Remark: In the particular case <strong>of</strong> constant coefficients we see that a primitive for the<br />
constant function a ∈ R is A(t) = at, while<br />
e A(t)<br />
<br />
e −A(t) b(t) dt = e at<br />
<br />
b e −at dt = e at<br />
− b<br />
a e−at = − b<br />
a ,<br />
hence we recover the expression y(t) = c eat − b<br />
given in Eq. (1.1.3).<br />
a<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 1.2.1: We generalize the integrating factor method to variable coefficients<br />
<strong>equations</strong>. We first write down the <strong>differential</strong> equation as<br />
y ′ − a(t) y = b(t).<br />
<br />
If we denote any primitive <strong>of</strong> coefficient a as follows, A(t) = a(t) dt, then multiply the<br />
equation above by the function e −A(t) , called the integrating factor,<br />
y ′ − a(t) y e A(t) = b(t) e −A(t) ⇔ e −A(t) y ′ − a(t) e −A(t) y = b(t) e −A(t) .<br />
This exponential was chosen because <strong>of</strong> the property<br />
−a(t) e −A(t) = e −A(t) ′ ,<br />
since A ′ (t) = a(t). Introducing this property into the <strong>differential</strong> equation,<br />
e −A(t) y ′ + e −A(t) ′ y = b(t) e −A(t) .<br />
The product rule for derivatives says that the left-hand side above is a total derivative,<br />
e −A(t) y ′ = b(t) e −A(t) .
G. NAGY – ODE July 2, 2013 9<br />
One way to proceed at this point is to rewrite the right-hand side above in terms <strong>of</strong> its<br />
primitive function, B(t) = e −A(t) b(t) dt, that is,<br />
e −A(t) y ′ = B ′ (t) ⇔ e −A(t) y − B(t) ′ = 0.<br />
As in the constant coefficient case, the whole <strong>differential</strong> equation has been rewritten as the<br />
total derivative <strong>of</strong> a function, in this case,<br />
ψ(t, y(t)) = e −A(t) <br />
y(t) − e −A(t) b(t) dt,<br />
called potential function. The <strong>differential</strong> equation now has the form<br />
dψ<br />
(t, y(t)) = 0.<br />
dt<br />
It is trivial to integrate the equation in terms <strong>of</strong> the potential function,<br />
ψ(t, y(t)) = c ⇔ e −A(t) <br />
y(t) − e −A(t) b(t) dt = c.<br />
Therefore, for each value <strong>of</strong> the constant c ∈ R we have the solution<br />
y(t) = c e A(t) + e A(t)<br />
<br />
e −A(t) b(t) dt.<br />
This establishes the Theorem. <br />
Lemma 1.1.3 can be generalized to the variable coefficient case, henceforth stating that<br />
the integrating factor µ(t) = e −A(t) describes all possible integrating factors for Eq. (1.2.1).<br />
Lemma 1.2.2 (Integrating Factor). Given any differentiable function y and integrable<br />
function a, every function µ satisfying<br />
(y ′ − ay)µ = (yµ) ′ ,<br />
<br />
must be given in terms <strong>of</strong> any primitive <strong>of</strong> function a, A(t) = a(t) dt, as follows,<br />
µ(t) = e −A(t) .<br />
Pro<strong>of</strong> <strong>of</strong> Lemma 1.2.2: Multiply Eq. (1.2.1) by a non-vanishing otherwise arbitrary<br />
function with values µ(t), and order the terms in the equation as follows<br />
µ(y ′ − a(t) y) = b(t) µ. (1.2.3)<br />
The key idea <strong>of</strong> the pro<strong>of</strong> is to choose the function µ such that the following equation holds<br />
µ(y ′ − a(t) y) = µy ′ . (1.2.4)<br />
Eq. (1.2.4), this is an equation for µ, since it can be rewritten as follows,<br />
µ y ′ − a(t) µy = µ ′ y + µ y ′ ⇔ −a(t) µy = µ ′ y ⇔ −a(t) µ = µ ′ ,<br />
that is, the function y does not appear in Eq. (1.2.4). The equation on the far right above<br />
can be solved for µ as follows:<br />
µ ′<br />
µ = −a(t) ⇔ ln(µ) ′ = −a(t).<br />
<br />
Integrating in the equation above, and introducing a primitive A(t) = a(t) dt, we obtain<br />
ln(µ) = −A(t) ⇔ e ln(µ) = e −A(t) ⇔ µ(t) = e −A(t) .<br />
This establishes the Lemma.
10 G. NAGY – ODE july 2, 2013<br />
Example 1.2.1: Find all solutions y to the <strong>differential</strong> equation<br />
y ′ = 3<br />
t y + t5 .<br />
Solution: Rewrite the equation as<br />
y ′ − 3<br />
t y = t5 .<br />
Introduce a primitive <strong>of</strong> the coefficient function a(t) = 3/t,<br />
<br />
3<br />
A(t) =<br />
t dt = 3 ln(t) = ln(t3 ),<br />
so we have A(t) = ln(t 3 ). The integrating factor µ is then<br />
µ(t) = e −A(t) = e − ln(t3 ) = e ln(t −3 ) = t −3 ,<br />
hence µ(t) = t −3 . We multiply the <strong>differential</strong> equation by the integrating factor<br />
t −3 y ′ − 3<br />
t y − t5 = 0 ⇔ t −3 y ′ − 3 t −4 y − t 2 Since −3 t<br />
= 0.<br />
−4 = (t−3 ) ′ , we get<br />
t −3 y ′ + (t −3 ) ′ y − t 2 = 0 ⇔<br />
<br />
t −3 y − t3<br />
3<br />
′<br />
= 0.<br />
The potential function in this case is ψ(t, y) = t −3 y − t3<br />
. Integrating the total derivative<br />
3<br />
we obtain<br />
t −3 y − t3<br />
3 = c ⇒ t−3 y = c + t3<br />
3 ,<br />
so all solutions to the <strong>differential</strong> equation are y(t) = c t 3 + t6<br />
, with c ∈ R. ⊳<br />
3<br />
1.2.2. The Initial Value Problem. We now generalize to variable coefficients <strong>equations</strong><br />
the existence and uniqueness <strong>of</strong> solutions to initial value problems for constant coefficients<br />
linear <strong>differential</strong> <strong>equations</strong> presented in Theorems 1.1.5. We first introduce the initial value<br />
problem for a variable coefficients <strong>equations</strong>, a simple generalization <strong>of</strong> Def. 1.1.4.<br />
Definition 1.2.3. The initial value problem (IVP) for a first order linear ODE is the<br />
following: Given functions a, b : R → R and constants t0, y0 ∈ R, find y : R → R solution <strong>of</strong><br />
y ′ = a(t) y + b(t), y(t 0) = y 0. (1.2.5)<br />
As we did with the constant coefficients IVP, the second equation in (1.2.5) is called the<br />
initial condition <strong>of</strong> the problem. We saw in Theorem 1.2.1 that the <strong>differential</strong> equation<br />
in (1.2.5) has infinitely many solutions, parametrized by a real constant c. The associated<br />
initial value problem has a unique solution though, because the initial condition fixes the<br />
arbitrary ccnstant c.<br />
Theorem 1.2.4 (Variable coefficients IVP). Given continuous functions a, b : (t 1, t 2) →<br />
R and constants t0 ∈ (t1, t2) and y0 ∈ R, the initial value problem<br />
has the unique solution y : (t 1, t 2) → R given by<br />
y ′ = a(t) y + b(t), y(t 0) = y 0, (1.2.6)<br />
y(t) = e A(t)<br />
y0 +<br />
t<br />
t0<br />
e −A(s) <br />
b(s) ds , (1.2.7)
where we introduced the function A(t) =<br />
G. NAGY – ODE July 2, 2013 11<br />
t<br />
t0<br />
a(s) ds, a particular primitive <strong>of</strong> function a.<br />
Remark: In the particular case <strong>of</strong> a constant coefficients equation, that is, a, b ∈ R, the<br />
solution given in Eq. (1.2.7) reduces to the one given in Eq. (1.1.10). Indeed,<br />
A(t) = −<br />
t<br />
t0<br />
a ds = −a(t − t 0),<br />
t<br />
t0<br />
e −a(s−t0) b ds = − b<br />
a e−a(t−t0) + b<br />
a .<br />
Therefore, the solution y can be written as<br />
y(t) = y0 e a(t−t0)<br />
<br />
+ − b<br />
a e−a(t−t0) b<br />
<br />
+ e<br />
a<br />
a(t−t0)<br />
<br />
= y0 + b<br />
<br />
e<br />
a<br />
a(t−t0) b<br />
−<br />
a .<br />
Pro<strong>of</strong> Theorem 1.2.4: We follow closely the pro<strong>of</strong> <strong>of</strong> Theorem 1.1.2. From Theorem 1.2.1<br />
we know that all solutions to the <strong>differential</strong> equation in (1.2.6) are given by<br />
y(t) = c e A(t) + e A(t)<br />
<br />
e −A(t) b(t) dt,<br />
<br />
for every c ∈ R. Let us use again the notation B(t) = e −A(t) b(t) dt, and then introduce<br />
the initial condition in (1.2.6), which fixes the constant c,<br />
so we get the constant c,<br />
y0 = y(t0) = c e A(t0) + e A(t0) B(t0),<br />
c = y0 e −A(t0) − B(t0).<br />
Using this expression in the general solution above,<br />
<br />
y(t) = y0 e −A(t0)<br />
<br />
− B(t0) e A(t) + e A(t) B(t) = y0 e A(t)−A(t0) A(t)<br />
+ e B(t) − B(t0) .<br />
Let us introduce the particular primitives Â(t) = A(t) − A(t0) and ˆ B(t) = B(t) − B(t0),<br />
which vanish at t0, that is,<br />
Â(t) =<br />
t<br />
t0<br />
a(s) ds,<br />
Then the solution y <strong>of</strong> the IVP has the form<br />
which is equivalent to<br />
so we conclude that<br />
ˆ B(t) =<br />
y(t) = y0 e Â(t) + e A(t)<br />
t<br />
y(t) = y0 e Â(t) + e A(t)−A(t0)<br />
t<br />
y(t) = e Â(t)<br />
y 0 +<br />
t<br />
t0<br />
t0<br />
t0<br />
t<br />
t0<br />
e −A(s) b(s) ds.<br />
e −A(s) b(s) ds<br />
e −[A(s)−A(t0)] b(s) ds,<br />
e −Â(s) <br />
b(s) ds .<br />
Renaming the particular primitive  simply by A, we then establish the Theorem. <br />
We solve the next Example following the main steps in the pro<strong>of</strong> <strong>of</strong> Theorem 1.2.4 above.<br />
Example 1.2.2: Find the function y solution <strong>of</strong> the initial value problem<br />
ty ′ + 2y = 4t 2 , y(1) = 2.
12 G. NAGY – ODE july 2, 2013<br />
Solution: We first write the equation above in a way it is simple to see the functions a<br />
and b in Theorem 1.2.4. In this case we obtain<br />
y ′ = − 2<br />
t<br />
Now rewrite the equation as<br />
and multiply it by the function µ = e A(t) ,<br />
y + 4t ⇔ a(t) = −2 , b(t) = 4t. (1.2.8)<br />
t<br />
y ′ = 2<br />
y = 4t,<br />
t<br />
e A(t) y ′ + 2<br />
t eA(t) y = 4t e A(t) .<br />
The function A in the integrating factor e A(t) must be the one satisfying<br />
A ′ (t) = 2<br />
t<br />
⇔ A(t) =<br />
In this case the <strong>differential</strong> equation can be written as<br />
2<br />
t dt.<br />
e A(t) y ′ + A ′ (t) e A(t) y = 4t e A(t) ⇔ e A(t) y ′ = 4t e A(t) .<br />
We now compute the function A,<br />
<br />
dt<br />
A(t) = 2<br />
t = 2 ln(|t|) ⇒ A(t) = ln(t2 ).<br />
This implies that<br />
e A(t) = t 2 .<br />
Therefore, the <strong>differential</strong> equation can be written as<br />
t 2 y ′ = 4t 3 .<br />
Integrating on both sides we obtain that<br />
The initial condition implies that<br />
t 2 y = t 4 + c ⇒ y(t) = t 2 + c<br />
.<br />
t2 2 = y(1) = c + 1 ⇒ c = 1 ⇒ y(t) = 1<br />
t 2 + t2 .<br />
Remark: It is not needed to compute the potential function to find the solution in the<br />
Example above. However, it could be useful to see this function for the <strong>differential</strong> equation<br />
in the Example. When the equation is written as<br />
t 2 y ′ = 4t 3 ⇔ (t 2 y) ′ = (t 4 ) ′ ⇔ (t 2 y − t 4 ) ′ = 0,<br />
it is simple to see that the potential function is<br />
ψ(t, y(t)) = t 2 y(t) − t 4 .<br />
The <strong>differential</strong> equation is then trivial to integrate,<br />
t 2 y − t 4 = c ⇔ t 2 y = c + t 4 ⇔ y(t) = c<br />
t 2 + t2 .<br />
⊳
G. NAGY – ODE July 2, 2013 13<br />
Example 1.2.3: Find the solution <strong>of</strong> the problem given in Example 1.2.2 using the results<br />
<strong>of</strong> Theorem 1.2.4.<br />
Solution: We find the solution simply by using Eq. (1.2.7). First, find the integrating<br />
factor function µ as follows:<br />
A(t) = −<br />
t<br />
1<br />
2<br />
s ds = −2 ln(t) − ln(1) = −2 ln(t) ⇒ A(t) = ln(t −2 ).<br />
The integrating factor is µ(t) = e −A(t) , that is,<br />
Then, we compute the solution as follows:<br />
µ(t) = e − ln(t−2 ) = e ln(t 2 ) ⇒ µ(t) = t 2 .<br />
y(t) = 1<br />
t2 <br />
2 +<br />
= 2 1<br />
+<br />
t2 t2 2<br />
1<br />
t<br />
1<br />
s 2 <br />
4s ds<br />
4s 3 ds<br />
= 2 1<br />
+<br />
t2 t2 (t4 − 1)<br />
= 2<br />
t2 + t2 − 1<br />
t2 ⇒<br />
1<br />
y(t) =<br />
t2 + t2 .<br />
1.2.3. The Bernoulli equation. In 1696 Jacob Bernoulli struggled for months trying to<br />
solve a particular <strong>differential</strong> equation, now known as Bernoulli’s <strong>differential</strong> equation, see<br />
Def. 1.2.5. He could not solve it, so he organized a contest among his peers to solve the<br />
equation. In short time his brother Johann Bernoulli solved it. Later on the equation was<br />
solved by Leibniz, transforming the original non-linear equation into a linear equation. In<br />
what follows we explain Leibniz’s idea in more detail.<br />
Definition 1.2.5. A Bernoulli equation in the unknown function y, determined by the<br />
functions p, q : (t1, t2) → R and a number n ∈ R, is the <strong>differential</strong> equation<br />
y ′ = p(t) y + q(t) y n . (1.2.9)<br />
In the case that n = 0 or n = 1 the Bernoulli equation reduces to a linear equation. The<br />
interesting cases are when the Bernoulli equation is non-linear. We now show in an Example<br />
the main idea to solve a Bernoulli equation: To transform the original non-linear equation<br />
into a linear equation.<br />
Example 1.2.4: Find every solution <strong>of</strong> the <strong>differential</strong> equation<br />
y ′ = y + 2y 5 .<br />
Solution: This is a Bernoulli equation for n = 5. Divide the equation by y 5 ,<br />
y ′ 1<br />
= + 2.<br />
y5 y4 Introduce the function v = 1/y 4 and its derivative v ′ = −4(y ′ /y 5 ), into the <strong>differential</strong><br />
equation above,<br />
− v′<br />
4 = v + 2 ⇒ v′ = −4 v − 8 ⇒ v ′ + 4 v = −8.<br />
⊳
14 G. NAGY – ODE july 2, 2013<br />
The last equation is a linear <strong>differential</strong> equation for v. This equation can be solved using<br />
the integrating factor method. Multiply the equation by µ(t) = e 4t , then<br />
e 4t v ′ = −8 e 4t ⇒ e 4t v = − 8<br />
4 e4t + c<br />
We obtain that v = c e −4t − 2. Since v = 1/y 4 ,<br />
1<br />
y4 = c e−4t − 2 ⇒ y(t) = ±<br />
1<br />
c e −4t − 2 1/4 .<br />
The following result summarizes the main ideas used in the previous example to find<br />
solutions to the non-linear Bernoulli equation.<br />
Theorem 1.2.6 (Bernoulli). The function y is a solution <strong>of</strong> the Bernoulli equation<br />
y ′ = p(t) y + q(t) y n , n = 1,<br />
iff the function v = 1/y (n−1) is solution <strong>of</strong> the linear <strong>differential</strong> equation<br />
v ′ = −(n − 1)p(t) v − (n − 1)q(t).<br />
This result says, solve the linear equation for function v, a the compute the solution y <strong>of</strong><br />
the Bernoulli equation as y = (1/v) 1/(n−1) .<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 1.2.6: Divide the Bernoulli equation by y n ,<br />
y ′ p(t)<br />
= + q(t).<br />
yn yn−1 Introduce the new unknown v = y −(n−1) and compute its derivative,<br />
v ′ = y −(n−1) ′ = −(n − 1)y −n y ′ ⇒ − v ′ (t)<br />
(n − 1) = y′ (t)<br />
y n (t) .<br />
If we substitute v and this last equation into the Bernoulli equation we get<br />
− v′<br />
(n − 1) = p(t) v + q(t) ⇒ v′ = −(n − 1)p(t) v − (n − 1)q(t).<br />
This establishes the Theorem. <br />
Example 1.2.5: Given any constants a0, b0, find every solution <strong>of</strong> the <strong>differential</strong> equation<br />
y ′ = a0y + b0y 3 .<br />
Solution: This is a Bernoulli equation. Divide the equation by y 3 ,<br />
y ′<br />
y<br />
3 = a0<br />
+ b0.<br />
y2 Introduce the function v = 1/y 2 and its derivative v ′ = −2(y ′ /y 3 ), into the <strong>differential</strong><br />
equation above,<br />
− v′<br />
2 = a0v + b0 ⇒ v ′ = −2a0v − 2b0 ⇒ v ′ + 2a0v = −2b0.<br />
The last equation is a linear <strong>differential</strong> equation for v. This equation can be solved using<br />
the integrating factor method. Multiply the equation by µ(t) = e2a0t ,<br />
2a0t ′<br />
e v = −2b0 e 2a0t ⇒ e 2a0t v = − b0<br />
a0<br />
e 2a0t + c<br />
⊳
We obtain that v = c e −2a0t − b0<br />
. Since v = 1/y2 ,<br />
a0<br />
1<br />
y 2 = c e−2a0t − b0<br />
a0<br />
G. NAGY – ODE July 2, 2013 15<br />
⇒ y(t) = ±<br />
1<br />
.<br />
c e−2a0t b0<br />
1/2<br />
− a0<br />
Example 1.2.6: Find every solution <strong>of</strong> the equation t y ′ = 3y + t 5 y 1/3 .<br />
Solution: Rewrite the <strong>differential</strong> equation as<br />
y ′ = 3<br />
t y + t4 y 1/3 .<br />
This is a Bernoulli equation for n = 1/3. Divide the equation by y 1/3 ,<br />
y ′ 3<br />
=<br />
y1/3 t y2/3 + t 4 .<br />
Define the new unknown function v = 1/y (n−1) , that is, v = y 2/3 , compute is derivative,<br />
v ′ = 2<br />
3<br />
y ′<br />
, and introduce them in the <strong>differential</strong> equation,<br />
y1/3 3<br />
2 v′ = 3<br />
t v + t4 ⇒ v ′ − 2 2<br />
v =<br />
t 3 t4 .<br />
Integrate the linear equation <strong>of</strong> v using the integrating factor method.<br />
factor is computed as follows:<br />
<br />
2<br />
A(t) =<br />
t<br />
The integrating<br />
dt = 2 ln(t) = ln(t2 ),<br />
and recalling that µ(t) = e −A(t) , we obtain<br />
µ(t) = e − ln(t2 ) ln(t<br />
= e −2 ) 1<br />
⇒ µ(t) = . 2<br />
t<br />
Therefore, the equation for v is<br />
1<br />
t2 ′ 2<br />
v −<br />
t v = 2<br />
3 t2 <br />
v 2<br />
⇒ −<br />
t2 9 t3 ′<br />
= 0.<br />
The potential function is ψ(t, v) = v/t2−(2/9)t3 and the solution <strong>of</strong> the <strong>differential</strong> equation<br />
is ψ(t, v(t)) = c, that is,<br />
v 2<br />
−<br />
t2 9 t3 <br />
2<br />
= c ⇒ v(t) = t c + 2<br />
9 t3 ⇒ v(t) = c t 2 + 2<br />
9 t5 .<br />
Once v is known we compute the original unknown y = ±v3/2 , where the double sign is<br />
related to taking the square root. So we conclude,<br />
<br />
y(t) = ± c t 2 + 2<br />
9 t5 3/2<br />
⊳<br />
⊳
16 G. NAGY – ODE july 2, 2013<br />
1.2.4. Exercises.<br />
1.2.1.- Find the solution y to the IVP<br />
y ′ = −y + e −2t , y(0) = 3.<br />
1.2.2.- Find the solution y to the IVP<br />
y ′ = y + 2te 2t , y(0) = 0.<br />
1.2.3.- Find the solution y to the IVP<br />
t y ′ + 2 y = sin(t)<br />
<br />
π<br />
<br />
, y =<br />
t 2<br />
2<br />
π ,<br />
for t > 0.<br />
1.2.4.- Find all solutions y to the ODE<br />
y ′<br />
(t2 = 4t.<br />
+ 1)y<br />
1.2.5.- Find all solutions y to the ODE<br />
ty ′ + n y = t 2 ,<br />
with n a positive integer.<br />
1.2.6.- Find all solutions to the ODE<br />
2ty − y ′ = 0.<br />
Show that given two solutions y1 and<br />
y2 <strong>of</strong> the equation above, the addition<br />
y1 + y2 is also a solution.<br />
1.2.7.- Find every solution <strong>of</strong> the equation<br />
y ′ + t y = t y 2 .<br />
1.2.8.- Find every solution <strong>of</strong> the equation<br />
y ′ = −x y = 6x √ y.
G. NAGY – ODE July 2, 2013 17<br />
1.3. Separable <strong>equations</strong><br />
1.3.1. Separable <strong>equations</strong>. Non-linear <strong>differential</strong> <strong>equations</strong> in general are more complicated<br />
to solve than the linear ones. However, one particular type <strong>of</strong> non-linear <strong>differential</strong><br />
<strong>equations</strong> are among the simplest <strong>equations</strong> to solve, even simpler than linear <strong>differential</strong><br />
<strong>equations</strong>. This type <strong>of</strong> non-linear <strong>equations</strong> is called separable <strong>equations</strong>, and they are<br />
solved simply by integrating on both sides <strong>of</strong> the <strong>differential</strong> equation.<br />
Definition 1.3.1. Given functions h, g : R → R, a first order ODE on the unknown function<br />
y : R → R is called separable iff the ODE has the form<br />
h(y) y ′ (t) = g(t).<br />
It is not difficult to see that a <strong>differential</strong> equation y ′ (t) = f(t, y(t)) is separable iff<br />
Example 1.3.1:<br />
(a) The <strong>differential</strong> equation<br />
y ′ = g(t)<br />
h(y)<br />
is separable, since it is equivalent to<br />
(b) The <strong>differential</strong> equation<br />
y ′ (t) =<br />
1 − y 2 y ′ (t) = t 2 ⇒<br />
is separable, since it is equivalent to<br />
⇔ f(t, y) = g(t)<br />
h(y) .<br />
t 2<br />
1 − y 2 (t)<br />
g(t) = t 2 ,<br />
y ′ (t) + y 2 (t) cos(2t) = 0<br />
1<br />
y 2 y′ (t) = − cos(2t) ⇒<br />
⎧<br />
⎨<br />
⎩<br />
h(y) = 1 − y 2 .<br />
g(t) = − cos(2t),<br />
h(y) = 1<br />
.<br />
y2 The functions g and h are not uniquely defined; another choice in this example is:<br />
g(t) = cos(2t), h(y) = − 1<br />
.<br />
y2 (c) The linear <strong>differential</strong> equation y ′ (t) = −a(t) y(t) is separable, since it is equivalent to<br />
1<br />
y y′ (t) = −a(t) ⇒<br />
⎧<br />
⎨<br />
⎩<br />
g(t) = −a(t),<br />
h(y) = 1<br />
y .<br />
(d) The constant coefficients linear <strong>differential</strong> equation y ′ (t) = −a0 y(t) + b0 is separable,<br />
since it is equivalent to<br />
1<br />
(−a0 y + b0) y′ (t) = 1 ⇒<br />
⎧<br />
⎨ g(t) = 1,<br />
1<br />
⎩ h(y) =<br />
(−a0 y + b0) .<br />
(e) The <strong>differential</strong> equation y ′ (t) = e y(t) + cos(t) is not separable.<br />
(f) The linear <strong>differential</strong> equation y ′ (t) = −a(t) y(t) + b(t), with b(t) non-constant, is not<br />
separable.
18 G. NAGY – ODE july 2, 2013<br />
We see in the examples above that a linear ODE with function b = 0 is separable, so the<br />
solutions <strong>of</strong> such equation can be computed using both the result we show below and the<br />
integrating factor method studied in Sect. 1.2. We now describe a way to find solutions to<br />
every first order separable <strong>differential</strong> equation.<br />
Theorem 1.3.2 (Separable <strong>equations</strong>). If the functions h, g : R → R are continuous,<br />
with h = 0 and with primitives H and G, respectively; that is,<br />
then, the separable ODE<br />
H ′ (u) = h(u), G ′ (t) = g(t),<br />
h(y) y ′ = g(t) (1.3.1)<br />
has infinitely many solutions y : R → R satisfying the algebraic equation<br />
where c ∈ R is arbitrary.<br />
H(y(t)) = G(t) + c, (1.3.2)<br />
Before presenting the pro<strong>of</strong> <strong>of</strong> the Theorem above we use it in the example below and we<br />
explicitly show how to find the primitives G and H <strong>of</strong> the given functions g and h.<br />
Example 1.3.2: Find all solutions y : R → R to the ODE<br />
y ′ (t) =<br />
t2 1 − y2 . (1.3.3)<br />
(t)<br />
Solution: Theorem 1.3.2 tell us how to obtain the solution y. Writing Eq. (1.3.3) as<br />
1 − y 2 y ′ (t) = t 2 ,<br />
we see that the functions h, g are given by<br />
h(u) = 1 − u 2 , g(t) = t 2 .<br />
Their primitive functions, H and G, respectively, are simple to compute,<br />
h(u) = 1 − u 2 ⇒ H(u) = u − u3<br />
3 ,<br />
g(t) = t 2 ⇒ G(t) = t3<br />
3 .<br />
Then, Theorem (1.3.2) implies that the solution y satisfies the algebraic equation<br />
y(t) − y3 (t) t3<br />
= + c,<br />
3 3<br />
(1.3.4)<br />
where c ∈ R is arbitrary. ⊳<br />
Notice that we have not obtained a definition <strong>of</strong> a solution function y in Example 1.3.2<br />
above. We have obtained an algebraic equation satisfied by the solution, Eq. (1.3.4). The<br />
equation is algebraic, since it does not contain any derivatives <strong>of</strong> function y. This is precisely<br />
the result <strong>of</strong> Theorem 1.3.2 given in Eq. (1.3.2), which is an algebraic equation satisfied by<br />
the solution y. The following notions are useful in this Section.<br />
Definition 1.3.3. Assume the notation in Theorem 1.3.2. The solution y <strong>of</strong> a separable<br />
ODE is given in implicit form iff function y is specified by<br />
H y(t) = G(t) + c,<br />
⊳
G. NAGY – ODE July 2, 2013 19<br />
The solution y <strong>of</strong> a separable ODE is given in explicit form iff function H is invertible<br />
and y is specified by<br />
y(t) = H −1 G(t) + c .<br />
Sometimes is difficult to find the inverse <strong>of</strong> function H. This is the case in Example 1.3.2.<br />
In such cases we leave the solution y written in implicit form. If H−1 is simple to compute,<br />
we express the solution y in explicit form. We now show the pro<strong>of</strong> <strong>of</strong> Theorem 1.3.2, which<br />
is based in an integration by substitution.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 1.3.2: Integrate on both sides in (1.3.1),<br />
h(y(t)) y ′ <br />
(t) = g(t) ⇒ h(y(t)) y ′ <br />
(t) dt = g(t) dt + c,<br />
where c is an arbitrary constant. Introduce on the left-hand side <strong>of</strong> the second equation<br />
above the substitution<br />
u = y(t), du = y ′ (t) dt,<br />
which implies,<br />
<br />
h(y(t)) y ′ <br />
<br />
<br />
(t) dt = h(u)du ⇒ h(u) du = g(t) dt + c.<br />
Recalling that H and G are the primitives <strong>of</strong> h and g, respectively,<br />
<br />
<br />
H(u) = h(u) du, G(t) = g(t) dt,<br />
then we obtain that<br />
H(u) = G(t) + c.<br />
Substitute y(t) back in the place <strong>of</strong> u. We conclude that the solution function y satisfies<br />
the algebraic equation<br />
H y(t) = G(t) + c.<br />
This establishes the Theorem. <br />
In the following Example we show that an initial value problem can be solved even in the<br />
case where the solutions <strong>of</strong> the ODE are given only in implicit form.<br />
Example 1.3.3: Find the solution <strong>of</strong> the initial value problem<br />
y ′ (t) =<br />
t2 1 − y2 , y(0) = 1. (1.3.5)<br />
(t)<br />
Solution: From Example 1.3.2 we know that all solutions to the <strong>differential</strong> equation<br />
in (1.3.5) are given by<br />
y(t) − y3 (t) t3<br />
= + c,<br />
3 3<br />
where c ∈ R is arbitrary. This constant c is now fixed with the initial condition in Eq. (1.3.5)<br />
y(0) − y3 (0)<br />
1<br />
2<br />
= c ⇒ 1 − = c ⇔ c =<br />
3 3 3 ⇒ y(t) − y3 (t) t3 2<br />
= +<br />
3 3 3 .<br />
So we can rewrite the algebraic equation defining the solution functions y as the roots <strong>of</strong> a<br />
cubic polynomial,<br />
y 3 (t) − 3y(t) + t 3 + 2 = 0.<br />
⊳<br />
In Examples 1.3.2 and 1.3.3 above we have used Theorem 1.3.2 to find the implicit form<br />
<strong>of</strong> the solution y. In the Example below we solve the same problem than in Example 1.3.2<br />
but now we follow step by step the pro<strong>of</strong> <strong>of</strong> Theorem 1.3.2.
20 G. NAGY – ODE july 2, 2013<br />
Example 1.3.4: Follow the pro<strong>of</strong> <strong>of</strong> Theorem 1.3.2 to find all solutions y to the equation<br />
y ′ (t) =<br />
Solution: We write the <strong>differential</strong> equation in (1.3.6) as follows<br />
1 − y 2 (t) y ′ (t) = t 2 ;<br />
t2 1 − y2 . (1.3.6)<br />
(t)<br />
We now integrate in the variable t on both sides <strong>of</strong> the equation above,<br />
<br />
1 2 ′<br />
− y (t) y (t) dt = t 2 dt + c;<br />
where c is any constant, since the integrals are indefinite. Introduce the substitution<br />
then we obtain<br />
<br />
(1 − u 2 <br />
) du =<br />
u = y(t), du = y ′ (t) dt,<br />
t 2 dt + c ⇔ u − u3<br />
3<br />
= t3<br />
3<br />
Substitute back the original unknown y into the last expression above and we obtain<br />
y(t) − y3 (t) t3<br />
= + c,<br />
3 3<br />
which is the implicit form <strong>of</strong> the solution y. ⊳<br />
We present now few more Examples.<br />
Example 1.3.5: Find the solution <strong>of</strong> the initial value problem<br />
+ c.<br />
y ′ (t) + y 2 (t) cos(2t) = 0, y(0) = 1. (1.3.7)<br />
Solution: The <strong>differential</strong> equation above is separable, with<br />
g(t) = − cos(2t), h(y) = 1<br />
,<br />
y2 therefore, it can be integrated as follows:<br />
Again the substitution<br />
y ′ (t)<br />
y2 = − cos(2t) ⇔<br />
(t)<br />
′ y (t)<br />
y2 <br />
dt = −<br />
(t)<br />
u = y(t), du = y ′ (t) dt<br />
cos(2t) dt + c.<br />
implies that <br />
du<br />
= − cos(2t) dt + c ⇔ −<br />
u2 1<br />
= −1 sin(2t) + c.<br />
u 2<br />
Substitute the unknown function y back in the equation above,<br />
− 1<br />
= −1 sin(2t) + c.<br />
y(t) 2<br />
The solution is given in implicit form. However, in this case is simple to solve this algebraic<br />
equation for y and we obtain the following explicit form for the solutions,<br />
2<br />
y(t) =<br />
sin(2t) − 2c .<br />
The initial condition implies that<br />
1 = y(0) = 2<br />
0 − 2c<br />
⇔ c = −1.
So, the solution to the IVP is given in explicit form by<br />
G. NAGY – ODE July 2, 2013 21<br />
y(t) =<br />
2<br />
sin(2t) + 2 .<br />
Example 1.3.6: Follow the pro<strong>of</strong> in Theorem 1.3.2 to find all solutions y <strong>of</strong> the ODE<br />
y ′ (t) =<br />
4t − t3<br />
4 + y 3 (t) .<br />
Solution: The <strong>differential</strong> equation above is separable, with<br />
therefore, it can be integrated as follows:<br />
4 + y 3 (t) y ′ (t) = 4t − t 3 ⇔<br />
Again the substitution<br />
g(t) = 4t − t 3 , h(y) = 4 + y 3 ,<br />
4 + y 3 (t) y ′ (t) dt =<br />
u = y(t), du = y ′ (t) dt<br />
<br />
(4t − t 3 ) dt + c.<br />
implies that<br />
<br />
(4 + u 3 <br />
) du = (4t − t 3 ) dt + c0. ⇔ 4u + u4<br />
4 = 2t2 − t4<br />
4 + c0. Substitute the unknown function y back in the equation above and calling c1 = 4c0 we<br />
obtain the following implicit form for the solution,<br />
y 4 (t) + 16y(t) − 8t 2 + t 4 = c 1.<br />
Example 1.3.7: Find the explicit form <strong>of</strong> the solution to the initial value problem<br />
y ′ (t) =<br />
2 − t<br />
1 + y(t)<br />
Solution: The <strong>differential</strong> equation above is separable with<br />
Their primitives are respectively given by,<br />
g(t) = 2 − t, h(u) = 1 + u.<br />
y(0) = 1. (1.3.8)<br />
g(t) = 2 − t ⇒ G(t) = 2t − t2<br />
, h(u) = 1 + u<br />
2<br />
⇒<br />
u2<br />
H(u) = u +<br />
2 .<br />
Therefore, the implicit form <strong>of</strong> all solutions y to the ODE above are given by<br />
y(t) + y2 (t)<br />
2<br />
= 2t − t2<br />
2<br />
with c ∈ R. The initial condition in Eq. (1.3.8) fixes the value <strong>of</strong> constant c, as follows,<br />
+ c,<br />
y(o) + y2 (0)<br />
= 0 + c<br />
2<br />
⇒<br />
1<br />
1 + = c<br />
2<br />
⇒<br />
3<br />
c =<br />
2 .<br />
We conclude that the implicit form <strong>of</strong> the solution y is given by<br />
y(t) + y2 (t)<br />
2<br />
= 2t − t2<br />
2<br />
+ 3<br />
2 , ⇔ y2 (t) + 2y(t) + (t 2 − 4t − 3) = 0.<br />
⊳<br />
⊳
22 G. NAGY – ODE july 2, 2013<br />
The explicit form <strong>of</strong> the solution can be obtained realizing that y(t) is a root in the quadratic<br />
polynomial above. The two roots <strong>of</strong> that polynomial are given by<br />
y±(t) = 1<br />
<br />
−2 ± 4 − 4(t2 − 4t − 3) ⇔ y±(t) = −1 ±<br />
2<br />
−t2 + 4t + 4.<br />
We have obtained two functions y+ and Y−. However, we know that there is only one<br />
solution to the IVP. We can decide which one is the solution by evaluating them at the<br />
value t = 0 given in the initial condition. We obtain<br />
y+(0) = −1 + √ 4 = 1,<br />
y−(0) = −1 − √ 4 = −3.<br />
Therefore, the solution is y+, that is, the explicit form <strong>of</strong> the solution is<br />
y(t) = −1 + −t 2 + 4t + 4.<br />
1.3.2. Euler Homogeneous <strong>equations</strong>. Sometimes a <strong>differential</strong> equation is not separable<br />
but a change <strong>of</strong> unknown in the equation can transform a non-separable equation into<br />
a separable equation. This is the case for a type <strong>of</strong> <strong>differential</strong> <strong>equations</strong> called Euler<br />
homogeneous <strong>equations</strong>.<br />
Definition 1.3.4. The first order <strong>differential</strong> equation <strong>of</strong> the form y ′ (t) = f t, y(t) is called<br />
Euler homogeneous iff for every real numbers t, u and every c = 0 the function f satisfies<br />
f(ct, cu) = f(t, u).<br />
A <strong>differential</strong> equation is Euler homogeneous when the function f is invariant under the<br />
change <strong>of</strong> scale <strong>of</strong> its arguments. It is not difficult to prove that a function with values<br />
f(t, u) having the property above must depend on (t, u) only through their quotient. That<br />
is, there exists a function F : R → R such that<br />
<br />
u<br />
<br />
f(t, u) = F .<br />
t<br />
Therefore, a first order <strong>differential</strong> equation is Euler homogeneous iff it has the form<br />
y ′ (t) = F<br />
y(t)<br />
t<br />
Example 1.3.8: Show that the equation below is Euler homogeneous,<br />
<br />
.<br />
(t − y) y ′ − 2y + 3t + y2<br />
t<br />
Solution: Rewrite the equation in the standard form<br />
(t − y) y ′ = 2y − 3t − y2<br />
t ⇒ y′ =<br />
= 0.<br />
<br />
2y − 3t − y2<br />
<br />
t .<br />
(t − y)<br />
Divide numerator and denominator by t. We get,<br />
y ′ <br />
2y − 3t −<br />
=<br />
y2<br />
<br />
1<br />
<br />
t<br />
<br />
t<br />
(t − y) 1<br />
<br />
t<br />
⇒ y ′ <br />
y<br />
<br />
y<br />
2 2 − 3 −<br />
= t<br />
t<br />
y<br />
.<br />
1 −<br />
t<br />
⊳
G. NAGY – ODE July 2, 2013 23<br />
We conclude that the <strong>differential</strong> equation is Euler homogeneous, because the right-hand side<br />
<strong>of</strong> the equation above depends only on y/t. Indeed, we have shown that f(t, y) = F (y/t),<br />
where<br />
f(t, y) = 2y − 3t − (y2 /t)<br />
, F (x) =<br />
t − y<br />
2x − 3 − x2<br />
,<br />
1 − x<br />
Example 1.3.9: Determine whether the equation below is Euler homogeneous,<br />
y ′ = t2<br />
.<br />
1 − y3 Solution: Dividing numerator and denominator by t3 we obtain<br />
y ′ t<br />
=<br />
2<br />
(1 − y3 <br />
1<br />
t<br />
)<br />
3<br />
<br />
<br />
1<br />
t3 ⇒ y ′ <br />
1<br />
<br />
= t<br />
<br />
1<br />
t3 <br />
y<br />
3 .<br />
−<br />
t<br />
We conclude that the <strong>differential</strong> equation is not Euler homogeneous. ⊳<br />
Theorem 1.3.5 (Euler Homogeneous). If the <strong>differential</strong> equation y ′ (t) = f t, y(t) is<br />
Euler homogeneous, then the <strong>differential</strong> equation for the unknown v(t) = y(t)<br />
is separable.<br />
t<br />
The Theorem says that Euler homogeneous <strong>equations</strong> can be transformed into separable<br />
<strong>equations</strong>. This is analogous to the idea used to solve a Bernoulli equation, where we<br />
transformed the non-linear equation into a linear one. In the case <strong>of</strong> an Euler homogeneous<br />
equation for the unknown function y, we transform it into a separable equation for the<br />
unknown function v = y/t. We then solve for v in implicit or explicit form. Finally we<br />
replace back y = t v.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 1.3.5: If y ′ = f(t, y) is homogeneous, then it can be written as<br />
y ′ = F (y/t) for some function F . Introduce v = y/t. This means,<br />
y(t) = t v(t) ⇒ y ′ (t) = v(t) + t v ′ (t).<br />
Introducing these expressions into the <strong>differential</strong> equation for y we get<br />
<br />
F (v) − v<br />
v + t v ′ = F (v) ⇒ v ′ =<br />
This last equation is separable. This establishes the Theorem. <br />
Example 1.3.10: Find all solutions y <strong>of</strong> the ODE y ′ = t2 + 3y2 .<br />
2ty<br />
Solution: The equation is homogeneous, since<br />
y ′ = (t2 + 3y2 <br />
1<br />
) t<br />
2ty<br />
2<br />
<br />
<br />
1<br />
t2 ⇒ y ′ <br />
y<br />
2 1 + 3<br />
= t<br />
y<br />
.<br />
2<br />
t<br />
Therefore, we introduce the change <strong>of</strong> unknown v = y/t, so y = t v and y ′ = v + t v ′ . Hence<br />
v + t v ′ =<br />
1 + 3v2<br />
2v<br />
⇒ t v ′ =<br />
t<br />
1 + 3v2<br />
2v − v = 1 + 3v2 − 2v2 2v<br />
.<br />
⊳
24 G. NAGY – ODE july 2, 2013<br />
We obtain the separable equation v ′ = 1<br />
2 1 + v<br />
<br />
. We rewrite and integrate it,<br />
t 2v<br />
2v<br />
1 + v2 v′ = 1<br />
t ⇒<br />
<br />
2v<br />
1 + v2 v′ <br />
1<br />
dt = dt + c0.<br />
t<br />
The substitution u = 1 + v 2 (t) implies du = 2v(t) v ′ (t) dt, so<br />
du<br />
u =<br />
dt<br />
t + c0 ⇒ ln(u) = ln(t) + c0 ⇒ u = e ln(t)+c0 .<br />
But u = e ln(t) e c0 , so denoting c1 = e c0 , then u = c1t. Hence, the explicit form <strong>of</strong> the solution<br />
can be computed as follows,<br />
1 + v 2 = c1t ⇒ 1 +<br />
<br />
y<br />
2 = c1t ⇒ y(t) = ±t<br />
t<br />
√ c1t − 1.<br />
Example 1.3.11: Find all solutions y <strong>of</strong> the ODE y ′ t(y + 1) + (y + 1)2<br />
= .<br />
Solution: This equation is not homogeneous in the unknown y and variable t, however,<br />
it becomes homogeneous in the unknown u(t) = y(t) + 1 and the same variable t. Indeed,<br />
u ′ = y ′ , thus we obtain<br />
y ′ t(y + 1) + (y + 1)2<br />
=<br />
t2 ⇔ u ′ tu + u2<br />
=<br />
t2 ⇔ u ′ = u<br />
t +<br />
t<br />
Therefore, we introduce the new variable v = u/t, which satisfies u = t v and u ′ = v + t v ′ .<br />
The <strong>differential</strong> equation for v is<br />
v + t v ′ = v + v 2 ⇔ t v ′ = v 2 ⇔<br />
v ′<br />
t 2<br />
dt =<br />
v2 u<br />
<br />
1<br />
dt + c,<br />
t<br />
with c ∈ R. The substitution w = v(t) implies dw = v ′ dt, so<br />
<br />
w −2 <br />
1<br />
dw =<br />
t dt + c ⇔ −w−1 1<br />
= ln(|t|) + c ⇔ w = −<br />
ln(|t|) + c .<br />
Substituting back v, u and y, we obtain w = v(t) = u(t)/t = [y(t) + 1]/t, so<br />
y + 1 1<br />
= −<br />
t ln(|t|) + c<br />
⇔<br />
t<br />
y(t) = − − 1.<br />
ln(|t|) + c<br />
2<br />
.<br />
⊳<br />
⊳
1.3.3. Exercises.<br />
1.3.1.- Find all solutions y to the ODE<br />
y ′ = t2<br />
y .<br />
Express the solutions in explicit form.<br />
1.3.2.- Find every solution y <strong>of</strong> the ODE<br />
3t 2 + 4y 3 y ′ − 1 + y ′ = 0.<br />
Leave the solution in implicit form.<br />
1.3.3.- Find the solution y to the IVP<br />
y ′ = t 2 y 2 , y(0) = 1.<br />
1.3.4.- Find every solution y <strong>of</strong> the ODE<br />
ty + 1 + t 2 y ′ = 0.<br />
G. NAGY – ODE July 2, 2013 25<br />
1.3.5.- Find every solution y <strong>of</strong> the Euler<br />
homogeneous equation<br />
y ′ y + t<br />
= .<br />
t<br />
1.3.6.- Find all solutions y to the ODE<br />
y ′ = t2 + y 2<br />
.<br />
ty<br />
1.3.7.- Find the explicit solution to the IVP<br />
(t 2 + 2ty) y ′ = y 2 , y(1) = 1.<br />
1.3.8.- Prove that if y ′ = f(t, y) is an Euler<br />
homogeneous equation and y1(t) is a solution,<br />
then y(t) = (1/k) y1(kt) is also a<br />
solution for every non-zero k ∈ R.
26 G. NAGY – ODE july 2, 2013<br />
1.4. Exact <strong>equations</strong><br />
We introduce a particular type <strong>of</strong> <strong>differential</strong> <strong>equations</strong> called exact. They are called exact<br />
<strong>differential</strong> <strong>equations</strong> because they can be rewritten as the vanishing <strong>of</strong> a total derivative<br />
<strong>of</strong> an appropriate function, called potential function. This property makes it simple to find<br />
implicit solutions to an exact equation. We just need to find this potential function. In<br />
the second part <strong>of</strong> this Section we study a type <strong>of</strong> <strong>differential</strong> <strong>equations</strong> that are not exact<br />
but they can be converted into exact <strong>equations</strong> when they are multiplied by an appropriate<br />
function. This function is called an integrating factor. The integrating factor converts a<br />
non-exact equation into an exact equation. Linear <strong>differential</strong> <strong>equations</strong> are a particular<br />
case <strong>of</strong> this type <strong>of</strong> <strong>equations</strong>, and we have studied them in Sections 1.1 and 1.2. For linear<br />
<strong>equations</strong> we computed integrating factors that transformed the equation into a derivative<br />
<strong>of</strong> a potential function. We now generalize this idea to a class <strong>of</strong> non-linear <strong>equations</strong>.<br />
1.4.1. Exact <strong>differential</strong> <strong>equations</strong>. We start with a definition <strong>of</strong> an exact equation.<br />
Definition 1.4.1. The <strong>differential</strong> equation in the unknown function y : (t1, t2) → R<br />
N(t, y(t)) y ′ (t) + M(t, y(t)) = 0<br />
is called exact in an open rectangle R = (t1, t2) × (u1, u2) ⊂ R 2 iff for every point (t, u) ∈ R<br />
the functions M, N : R → R are continuously differentiable and satisfy the equation<br />
∂tN(t, u) = ∂uM(t, u)<br />
We use the notation for partial derivatives ∂tN = ∂N<br />
∂t and ∂uM = ∂M<br />
. Let us see<br />
∂u<br />
whether the following <strong>equations</strong> are exact or not.<br />
Example 1.4.1: Show whether the <strong>differential</strong> equation below is exact or not,<br />
2ty(t) y ′ (t) + 2t + y 2 (t) = 0.<br />
Solution: We first identify the functions N and M. This is simple in this case, since<br />
2ty(t) y ′ (t) + 2t + y 2 (t) = 0 ⇒ N(t, u) = 2tu, M(t, u) = 2t + u 2 .<br />
The equation is indeed exact, since<br />
<br />
N(t, u) = 2tu ⇒ ∂tN(t, u) = 2u,<br />
M(t, u) = 2t + u 2 ⇒ ∂uM(t, u) = 2u,<br />
⇒ ∂tN(t, u) = ∂uM(t, u).<br />
Example 1.4.2: Show whether the <strong>differential</strong> equation below is exact or not,<br />
sin(t)y ′ (t) + t 2 e y(t) y ′ (t) − y ′ (t) = −y(t) cos(t) − 2te y(t) .<br />
Solution: We first identify the functions N and M by rewriting the equation as follows,<br />
sin(t) + t 2 e y(t) − 1 y ′ (t) + y(t) cos(t) + 2te y(t) = 0<br />
we can see that<br />
N(t, u) = sin(t) + t 2 e u − 1 ⇒ ∂tN(t, u) = cos(t) + 2te u ,<br />
M(t, u) = u cos(t) + 2te u<br />
Therefore, the equation is exact, since<br />
⇒ ∂uM(t, u) = cos(t) + 2te u .<br />
∂tN(t, u) = ∂uM(t, u).<br />
⊳<br />
⊳
G. NAGY – ODE July 2, 2013 27<br />
The last example shows whether the linear ODE we studied in Section 1.2 is exact.<br />
Example 1.4.3: Show whether the linear <strong>differential</strong> equation below is exact or not,<br />
y ′ (t) = −a(t) y(t) + b(t), a(t) = 0.<br />
Solution: We first find the functions N and M rewriting the equation as follows,<br />
y ′ + a(t)y − b(t) = 0 ⇒ N(t, u) = 1, M(t, u) = a(t) u − b(t).<br />
We can see that the <strong>differential</strong> equation is not exact, since<br />
<br />
N(t, u) = 1 ⇒ ∂tN(t, u) = 0,<br />
⇒ ∂tN(t, u) = ∂uM(t, u).<br />
M(t, u) = a(t)u − b(t) ⇒ ∂uM(t, u) = a(t),<br />
1.4.2. The Poincaré Lemma. It is well-known how to find implicit solutions to exact<br />
<strong>differential</strong> <strong>equations</strong>. The main idea is to find the potential function mentioned at the<br />
begining <strong>of</strong> this Section. We first show that this potential function always exists for an<br />
exact equation. This result is due to Henri Poincaré around 1880 and we show it without<br />
pro<strong>of</strong>. Later on we show a method to find the potential function whose existence is asserted<br />
by Poincaré’s result.<br />
Lemma 1.4.2 (Poincaré). Continuously differentiable functions M, N : R → R, on an<br />
open rectangle R = (t1, t2) × (u1, u2), satisfy the equation<br />
∂tN(t, u) = ∂uM(t, u) (1.4.1)<br />
iff there exists a twice continuously differentiable function ψ : R → R, called potential<br />
function, such that for all (t, u) ∈ R holds<br />
∂uψ(t, u) = N(t, u), ∂tψ(t, u) = M(t, u). (1.4.2)<br />
Pro<strong>of</strong> <strong>of</strong> Lemma 1.4.2:<br />
(⇐) We assume that the potential function ψ is given ad satisfies Eq. (1.4.2). Since ψ is<br />
twice continuously differentiable, its cross derivatives are the same, that is, ∂t∂uψ = ∂u∂tψ.<br />
We then conclude that<br />
∂tN = ∂t∂uψ = ∂u∂tψ = ∂uM.<br />
(⇒) It is not given. See [2]. <br />
We now show the potential function for a particular exact <strong>differential</strong> equation. Later on<br />
we show how to actually find such potential function.<br />
Example 1.4.4: Show that the function ψ(t, u) = t 2 + tu 2 is the potential function for the<br />
exact <strong>differential</strong> equation<br />
2ty(t) y ′ (t) + 2t + y 2 (t) = 0.<br />
Solution: We saw in Example 1.4.1 that the <strong>differential</strong> equation above is exact, since the<br />
functions M and N,<br />
N(t, u) = 2tu, M(t, u) = 2t + u 2 ,<br />
satisfy that ∂tN = 2u = ∂uM. It is simple to see that ψ(t, u) = t 2 + tu 2 is the potential<br />
function, since<br />
∂tψ = 2t + u 2 = M, ∂uψ = 2tu = N.<br />
We remark that the Poincaré Lemma does not tell us how to find the potential function, it<br />
only states necessary and sufficient conditions on N and M for the existence <strong>of</strong> ψ. ⊳<br />
⊳
28 G. NAGY – ODE july 2, 2013<br />
The potential function ψ is crucial to find implicit solutions to exact ODE, as it can be<br />
seen in the following result.<br />
Theorem 1.4.3 (Exact equation). If the <strong>differential</strong> equation<br />
N(t, y(t)) y ′ (t) + M(t, y(t)) = 0 (1.4.3)<br />
is exact on R = (t1, t2) × (u1, u2), then every solution y must satisfy the algebraic equation<br />
ψ(t, y(t)) = c, (1.4.4)<br />
where c ∈ R and ψ : R → R is a potential function for Eq. (1.4.3).<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 1.4.3: Since the <strong>differential</strong> equation in (1.4.3) is exact, Lemma 1.4.2<br />
implies that there exists a potential function ψ satisfying Eqs. (1.4.2). Hence, we can replace<br />
in the <strong>differential</strong> equation the functions N and M by ∂yψ and ∂tψ, respectively. That is,<br />
0 = N(t, y(t)) y ′ (t) + M(t, y(t))<br />
= ∂yψ(t, y(t)) d<br />
dt y(t) + ∂tψ(t, y(t))<br />
= d<br />
ψ(t, y(t))<br />
dt<br />
⇔ ψ(t, y(t)) = c,<br />
where c is an arbitrary constant. This establishes the Theorem. <br />
Example 1.4.5: Find all solutions y to the <strong>differential</strong> equation<br />
2ty(t) y ′ (t) + 2t + y 2 (t) = 0.<br />
Solution: The first step is to verify whether the <strong>differential</strong> equation is exact. We know<br />
the answer, the equation is exact, we did this calculation before in Example 1.4.1, but we<br />
reproduce it here anyway.<br />
N(t, u) = 2tu ⇒ ∂tN(t, u) = 2u,<br />
M(t, u) = 2t + u 2 <br />
⇒ ∂tN(t, u) = ∂uM(t, u).<br />
⇒ ∂uM(t, u) = 2u.<br />
Since the equation is exact, Lemma 1.4.2 implies that there exists a potential function ψ<br />
satisfying the <strong>equations</strong><br />
∂uψ(t, u) = N(t, u), (1.4.5)<br />
∂tψ(t, u) = M(t, u). (1.4.6)<br />
We now proceed to compute the function ψ. Integrate Eq. (1.4.5) in the variable u keeping<br />
the variable t constant,<br />
<br />
∂uψ(t, u) = 2tu ⇒ ψ(t, u) = 2tu du + g(t),<br />
where g is a constant <strong>of</strong> integration on the variable u, so g can only depend on t. We obtain<br />
ψ(t, u) = tu 2 + g(t). (1.4.7)<br />
Introduce into Eq. (1.4.6) the expression for the function ψ in Eq. (1.4.7) above, that is,<br />
u 2 + g ′ (t) = ∂tψ(t, u) = M(t, u) = 2t + u 2 ,<br />
g ′ (t) = 2t<br />
Integrate in t the last equation above, and choose the integration constant to be zero,<br />
g(t) = t 2 .<br />
We have found that a potential function is given by<br />
ψ(t, u) = tu 2 + t 2 .
G. NAGY – ODE July 2, 2013 29<br />
Therefore, Theorem 1.4.3 implies that all solutions y satisfy the implicit equation<br />
ty 2 (t) + t 2 = c0,<br />
where c0 ∈ R is an arbitrary constant. Notice that choosing g(t) = t 2 + c has the only effect<br />
<strong>of</strong> modifying the constant c 0 above.<br />
⊳<br />
Example 1.4.6: Find all solutions y to the equation<br />
sin(t) + t 2 e y(t) − 1 y ′ (t) + y(t) cos(t) + 2te y(t) = 0.<br />
Solution: The first step is to verify whether the <strong>differential</strong> equation is exact. We know<br />
the answer, the equation is exact, we did this calculation before in Example 1.4.2, but we<br />
reproduce it here anyway.<br />
N(t, u) = sin(t) + t 2 e u − 1 ⇒ ∂tN(t, u) = cos(t) + 2te u ,<br />
M(t, u) = u cos(t) + 2te u<br />
⇒ ∂uM(t, u) = cos(t) + 2te u .<br />
Therefore, the <strong>differential</strong> equation is exact. Then, Lemma 1.4.2 implies that there exists a<br />
potential function ψ satisfying the <strong>equations</strong><br />
∂uψ(t, u) = N(t, u), (1.4.8)<br />
∂tψ(t, u) = M(t, u). (1.4.9)<br />
We know proceed to compute the function ψ. We first integrate in the variable u the<br />
equation ∂uψ = N keeping the variable t constant,<br />
∂uψ(t, u) = sin(t) + t 2 e u − 1 ⇒<br />
<br />
sin(t) 2 u<br />
ψ(t, u) = + t e − 1 du + g(t)<br />
where g is a constant <strong>of</strong> integration on the variable u, so g can only depend on t. We obtain<br />
ψ(t, u) = u sin(t) + t 2 e u − u + g(t).<br />
Now introduce the expression above for the potential function ψ in Eq. (1.4.9), that is,<br />
u cos(t) + 2te u + g ′ (t) = ∂tψ(t, u) = M(t, u) = u cos(t) + 2te u ,<br />
g ′ (t) = 0.<br />
The solution is g(t) = c, with c a constant, but we can always choose that constant to be<br />
zero, so we conclude that<br />
We have obtained the potential function<br />
g(t) = 0.<br />
ψ(t, u) = u sin(t) + t 2 e u − u.<br />
Therefore, Theorem 1.4.3 implies that the solution y satisfies the implicit equation<br />
y(t) sin(t) + t 2 e y(t) − y(t) = c0.<br />
The solution y above cannot be written in explicit form. Notice that choosing g(t) = c has<br />
the only effect <strong>of</strong> modifying the constant c 0 above. ⊳
30 G. NAGY – ODE july 2, 2013<br />
1.4.3. The integrating factor method. There exist non-exact <strong>differential</strong> <strong>equations</strong> that<br />
can be converted into an exact <strong>differential</strong> equation by multiplying the original equation<br />
by an appropriate function, called integrating factor. One example is the case <strong>of</strong> linear<br />
<strong>differential</strong> <strong>equations</strong>. We have seen in Example 1.4.3 that linear <strong>equations</strong> with coefficient<br />
a = 0 are not exact. Nevertheless, in Section 1.2 we have obtained solutions to linear<br />
<strong>equations</strong> multiplying the equation by an appropriate function, called the integrating factor.<br />
This functions converts the original <strong>differential</strong> equation into a total derivative. Using the<br />
terminology <strong>of</strong> this Section, the integrating factor transforms a linear equation into an exact<br />
equation. Now we generalize this idea to non-linear <strong>differential</strong> <strong>equations</strong>.<br />
Theorem 1.4.4 (Integrating factor). Assume that the <strong>differential</strong> equation<br />
N(t, y(t)) y ′ (t) + M(t, y(t)) = 0 (1.4.10)<br />
is not exact in the sense that the continuously differentiable functions M, N : R → R satisfy<br />
∂tN(t, u) = ∂uM(t, u) on R = (t1, t2) × (u1, u2). If N = 0 on R and the function<br />
1 <br />
∂uM(t, u) − ∂tN(t, u)<br />
N(t, u)<br />
<br />
does not depend on the variable u, then the equation<br />
(1.4.11)<br />
µ(t)N(t, y(t)) y ′ (t) + µ(t)M(t, y(t)) = 0 (1.4.12)<br />
is exact, where the function µ : (t1, t2) → R is the solution <strong>of</strong> the equation<br />
µ ′ (t)<br />
µ(t) =<br />
1 <br />
∂uM(t, u) − ∂tN(t, u)<br />
N(t, u)<br />
.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 1.4.4: Since the original <strong>differential</strong> equation in (1.4.10) is not exact,<br />
we multiply this equation by a non-zero function µ,<br />
µ(t)N(t, y) y ′ + µ(t)M(t, y) = 0. (1.4.13)<br />
Introducing the functions Ñ(t, u) = µ(t)N(t, u) and ˜ M(t, u) = µ(t)M(t, u), we see that this<br />
new equation is exact iff<br />
The calculation<br />
∂t Ñ(t, u) = ∂u ˜ M(t, u).<br />
∂t Ñ(t, u) = µ′ (t)N(t, u) + µ(t)∂tN(t, u), ∂u ˜ M(t, u) = µ(t)∂uM(t, u),<br />
says that Eq.(1.4.13) is exact iff<br />
which is equivalent to<br />
and in turns it is equivalent to<br />
µ ′ (t)N(t, u) + µ(t)∂tN(t, u) = µ(t)∂uM(t, u),<br />
µ ′ (t)N(t, u) = µ(t) ∂uM(t, u) − ∂tN(t, u) ,<br />
µ ′ (t)<br />
µ(t) =<br />
1 <br />
∂uM(t, u) − ∂tN(t, u)<br />
N(t, u)<br />
.<br />
Since the left-hand side does not depend on the variable u, the equation above is solvable<br />
iff the right-hand side does not depend on this variable u. This observation establishes the<br />
Theorem.
G. NAGY – ODE July 2, 2013 31<br />
Example 1.4.7: Find all solutions y to the <strong>differential</strong> equation<br />
t 2 + t y(t) y ′ (t) + 3t y(t) + y 2 (t) = 0. (1.4.14)<br />
Solution: We first verify whether this equation is exact:<br />
N(t, u) = t 2 + tu ⇒ ∂tN(t, u) = 2t + u,<br />
M(t, u) = 3tu + u 2<br />
⇒ ∂uM(t, u) = 3t + 2u,<br />
therefore, the <strong>differential</strong> equation is not exact. We now verify whether the extra condition<br />
in Theorem 1.4.4 holds, that is, whether the function in (1.4.11) is u-independent:<br />
1 <br />
∂uM(t, u) − ∂tN(t, u)<br />
N(t, u)<br />
1<br />
=<br />
(t2 <br />
(3t + 2u) − (2t + u)<br />
+ tu)<br />
1<br />
= (t + u)<br />
t(t + u)<br />
= 1<br />
t .<br />
We conclude that the function (∂uM −∂tN)/N does not depend on the variable u. Therefore,<br />
Theorem 1.4.4 implies that the <strong>differential</strong> equation in (1.4.14) can be transformed into an<br />
exact equation multiplying it by the function µ solution <strong>of</strong> the equation<br />
µ ′ (t) 1 <br />
= ∂uM − ∂tN<br />
µ(t) N<br />
= 1<br />
⇒ ln(µ(t)) = ln(t) ⇒ µ(t) = t,<br />
t<br />
where we have chosen in second equation the integration constant to be zero. Then, multiplying<br />
the original <strong>differential</strong> equation in (1.4.14) by the integrating factor µ we obtain<br />
3t 2 y(t) + t y 2 (t) + t 3 + t 2 y(t) y ′ (t) = 0. (1.4.15)<br />
This latter equation is exact, since<br />
Ñ(t, u) = t 3 + t 2 u ⇒ ∂t Ñ(t, u) = 3t2 + 2tu,<br />
˜M(t, u) = 3t 2 u + tu 2<br />
⇒ ∂u ˜ M(t, u) = 3t 2 + 2tu,<br />
that is, ∂t Ñ = ∂u ˜ M. This establishes that the new equation in (1.4.15) is exact. Therefore,<br />
the solution can be found as we did in the previous examples in this Section. That is, we<br />
find the potential function ψ solution <strong>of</strong> the <strong>equations</strong><br />
∂uψ(t, u) = Ñ(t, u), (1.4.16)<br />
∂tψ(t, u) = ˜ M(t, u). (1.4.17)<br />
From the first equation above we obtain<br />
∂uψ = t 3 + t 2 <br />
t 3 2<br />
u ⇒ ψ(t, u) = + t u du + g(t).<br />
Integrating on the right hand side above we arrive to<br />
ψ(t, u) = t 3 u + 1<br />
2 t2 u 2 + g(t).<br />
Introduce the expression above for ψ in Eq. (1.4.17), that is,<br />
3t 2 u + tu 2 + g ′ (t) = ∂tψ(t, u) = ˜ M(t, u) = 3t 2 u + tu 2 ,<br />
g ′ (t) = 0.<br />
A solution to this last equation is g(t) = 0. We conclude that a potential function is<br />
ψ(t, u) = t 3 u + 1<br />
2 t2 u 2 .
32 G. NAGY – ODE july 2, 2013<br />
All solutions y to the <strong>differential</strong> equation in (1.4.14) satisfy the equation<br />
t 3 y(t) + 1<br />
2 t2 y(t) 2 = c0,<br />
where c0 ∈ R is arbitrary. ⊳<br />
We have seen in Example 1.4.3 that linear <strong>differential</strong> <strong>equations</strong> with a = 0 are not<br />
exact. We have seen in Section 1.2 that the integrating factor method transforms the linear<br />
<strong>differential</strong> equation into a total derivative. Such calculation is now a particular case <strong>of</strong><br />
Theorem 1.4.4, as it can be seen in our last Example.<br />
Example 1.4.8: Use Theorem 1.4.4 to find all solutions to the linear <strong>differential</strong> equation<br />
Solution: We first write the <strong>differential</strong> equation as<br />
y ′ (t) = a(t) y(t) + b(t), a(t) = 0. (1.4.18)<br />
y ′ + −a(t) y − b(t) = 0.<br />
Now it is simple to verify whether the equation is exact or not. We have seen in Example 1.4.3<br />
that this equation is not exact, since<br />
N(t, u) = 1 ⇒ ∂tN(t, u) = 0,<br />
M(t, u) = −a(t)u − b(t) ⇒ ∂uM(t, u) = −a(t).<br />
However, the function<br />
1 <br />
∂uM(t, u) − ∂tN(t, u)<br />
N(t, u)<br />
= −a(t)<br />
is independent <strong>of</strong> the variable u. Therefore, we can compute the integrating factor µ solving<br />
the equation<br />
µ ′ (t)<br />
µ(t) = −a(t) ⇒ µ(t) = e−A(t) <br />
, A(t) = a(t)dt.<br />
Therefore, the equation<br />
is exact, since<br />
Ñ(t, u) = e −A(t)<br />
˜M(t, u) = −a(t) e −A(t) u − b(t) e −A(t)<br />
e −A(t) y ′ − a(t) e −A(t) y − b(t) e −A(t) = 0 (1.4.19)<br />
⇒ ∂t Ñ(t, u) = −a(t) e−A(t) ,<br />
⇒ ∂u ˜ M(t, u) = −a(t) e −A(t) .<br />
This establishes that the new equation in (1.4.19) is exact. Therefore, the solution can be<br />
found as we did in the previous examples in this Section. That is, we find the potential<br />
function ψ solution <strong>of</strong> the <strong>equations</strong><br />
∂uψ(t, u) = Ñ(t, u), (1.4.20)<br />
∂tψ(t, u) = ˜ M(t, u). (1.4.21)<br />
From the first equation above we obtain<br />
∂uψ = e −A(t) <br />
⇒ ψ(t, u) = e −A(t) du + g(t).<br />
The integrating is trivial, since e −A(t) is u-independent, so we arrive to<br />
ψ(t, u) = e −A(t) u + g(t).
Introduce the expression above for ψ in Eq. (1.4.17), that is,<br />
G. NAGY – ODE July 2, 2013 33<br />
−a(t) e −A(t) u + g ′ (t) = ∂tψ(t, u) = ˜ M(t, u) = −a(t) e −A(t) u − b(t) e −A(t) ,<br />
A function g is then given by<br />
g ′ (t) = −b(t) e −A(t) .<br />
<br />
g(t) = −<br />
b(t) e −A(t) dt.<br />
We conclude that a potential function is<br />
ψ(t, u) = e −A(t) <br />
u − b(t) e −A(t) dt.<br />
All solutions y to the <strong>differential</strong> equation in (1.4.18) satisfy the equation<br />
e −A(t) <br />
y(t) − b(t) e −A(t) dt = c0, where c0 ∈ R is arbitrary. This is the implicit form <strong>of</strong> the solution. The explicit form is<br />
simple to find, and is given by<br />
y(t) = e A(t)<br />
<br />
c0 + b(t) e −A(t) <br />
dt .<br />
This expression agrees with the one in Theorem 1.2.4, when we studied linear <strong>equations</strong>. ⊳
34 G. NAGY – ODE july 2, 2013<br />
1.4.4. Exercises.<br />
1.4.1.- Consider the equation<br />
(1 + t 2 ) y ′ = −2t y.<br />
(a) Determine whether the <strong>differential</strong><br />
equation is exact.<br />
(b) Find every solution <strong>of</strong> the equation<br />
above.<br />
1.4.2.- Consider the equation<br />
t cos(y) y ′ − 2y y ′ = −t − sin(y).<br />
(a) Determine whether the <strong>differential</strong><br />
equation is exact.<br />
(b) Find every solution <strong>of</strong> the equation<br />
above.<br />
1.4.3.- Consider the equation<br />
y ′ =<br />
−2 − y ety<br />
.<br />
−2y + t ety (a) Determine whether the <strong>differential</strong><br />
equation is exact.<br />
(b) Find every solution <strong>of</strong> the equation<br />
above.<br />
1.4.4.- Consider the equation<br />
(6x 5 − xy) + (−x 2 + xy 2 )y ′ = 0,<br />
with initial condition y(0) = 1.<br />
(a) Find the integrating factor µ that<br />
converts the equation above into an<br />
exact equation.<br />
(b) Find an implicit expression for the<br />
solution y <strong>of</strong> the IVP.<br />
1.4.5.- Consider the equation<br />
<br />
2x 2 y + y<br />
x 2<br />
<br />
y ′ + 4xy 2 = 0,<br />
with initial condition y(0) = −2.<br />
(a) Find the integrating factor µ that<br />
converts the equation above into an<br />
exact equation.<br />
(b) Find an implicit expression for the<br />
solution y <strong>of</strong> the IVP.
G. NAGY – ODE July 2, 2013 35<br />
1.5. Applications<br />
Ordinary <strong>differential</strong> <strong>equations</strong> can be used to describe situations that change in time.<br />
For example the radioactive decay <strong>of</strong> a radioactive substance. The amount <strong>of</strong> the radioactive<br />
substance present in a sample decays exponentially with time. This function is the solution<br />
<strong>of</strong> a particularly simple <strong>differential</strong> equation. The same type <strong>of</strong> <strong>differential</strong> <strong>equations</strong> also<br />
describes how materials cool down. This is the Newton cooling law. The first half <strong>of</strong> this<br />
Section is dedicated to study these situations. In the second half we study the salt concentration<br />
inside a water tank in the case that salty water is allowed in and out the tank. We<br />
will see that the <strong>differential</strong> equation that describes the salt in the tank is more complicated<br />
than the equation that appears in the exponential decay <strong>of</strong> a radioactive substance.<br />
1.5.1. Radioactive decay. A radioactive decay is a process that happens in the nucleus<br />
<strong>of</strong> certain substances, like Uranium-234, Radium-226, Radon-222, Polonium-218, Lead-214,<br />
Cobalt-60, Carbon-14, etc. The nucleus loses energy by emitting different types <strong>of</strong> charged<br />
particles. Therefore, there are many types <strong>of</strong> exponential decays. A nucleus can emit an<br />
alpha particle (Helium nucleus), can emit a proton (Hydrogen nucleus), can emit an electron,<br />
gamma-rays, neutrinos, etc. The radioactive decay <strong>of</strong> a single atom is impossible to predict,<br />
but the decay <strong>of</strong> a large number <strong>of</strong> atoms is predictable. It can be seen in the laboratory that<br />
the rate <strong>of</strong> change in the amount <strong>of</strong> a radioactive substance is proportional to the negative<br />
<strong>of</strong> the amount <strong>of</strong> substance in the sample. If we denote by N(t) the amount <strong>of</strong> a radioactive<br />
substance in a sample as function <strong>of</strong> time, then<br />
N ′ (t) = −a N(t),<br />
where a > 0 is called the decay constant and characterizes the radioactive material. The<br />
<strong>differential</strong> equation above is simple to solve using the integration factor method. Indeed,<br />
N ′ (t) + a N(t) e at = 0 ⇔ e at N(t) ′ = 0 ⇔ N(t) = N0 e −at ,<br />
where N0 = N(0) is the initial amount <strong>of</strong> the substance. Radioactive substances are <strong>of</strong>ten<br />
characterized not by their decay constant a but by their half-life. The half-life is a time τ<br />
needed for half <strong>of</strong> the radioactive substance to decay, that is,<br />
N(τ) = N0<br />
2 .<br />
The equation above provides a simple relation between the decay constant a and the half-life<br />
τ, namely,<br />
N0<br />
2 = N0 e −aτ <br />
1<br />
<br />
⇒ −aτ = ln ⇒ a =<br />
2<br />
ln(2)<br />
.<br />
τ<br />
Therefore, the solution N can be expressed in terms <strong>of</strong> the half-life as<br />
N(t) = N0 e (−t/τ) ln(2) ⇒ N(t) = N0 e ln[2(−t/τ) ] ⇒ N(t) = N0 2 −t/τ .<br />
From this last expression is clear that for t = τ we get N(τ) = N0/2.<br />
Example 1.5.1: The Carbon-14 is a radioactive isotope <strong>of</strong> Carbon that is constantly renovated<br />
in the atmosphere and is accumulated by living organisms. While the organism<br />
lives, the amount <strong>of</strong> Carbon-14 is held constant, because the decay is compensated with<br />
new amounts being incorporated into the organism. When the organism dies, the amount<br />
<strong>of</strong> Carbon-14 in its remains decays. The half-life <strong>of</strong> Carbon-14 can be measured in the<br />
laboratory and is τ = 5730 years. If certain remains are found containing an amount <strong>of</strong> 14%<br />
<strong>of</strong> the original amount <strong>of</strong> Carbon-14, find the date <strong>of</strong> the remains.
36 G. NAGY – ODE july 2, 2013<br />
Solution: Suppose that t = 0 is set at the time that the organism dies. If the remains<br />
contain 14% <strong>of</strong> the original amount at the present time t, that means<br />
N(t)<br />
N0<br />
= 14<br />
100 .<br />
Since the amount <strong>of</strong> Carbon-14 only decays after the organism dies, the amount <strong>of</strong> Carbon-14<br />
as function <strong>of</strong> time is given by<br />
N(t) = N0 2 −t/τ ,<br />
where τ = 5730 years is the Carbon-14 half-life. Therefore,<br />
2 −t/τ = 14<br />
100<br />
⇒<br />
t<br />
−<br />
τ = log2(14/100) ⇒ t = τ log2(100/14). We obtain that t = 16, 253 years. The organism died more that 16, 000 years ago. ⊳<br />
1.5.2. Newton’s cooling law. The Newton’s cooling law is an empirical law that describes<br />
how objects cool down. The temperature difference ∆T = T − T0 between the temperature<br />
<strong>of</strong> an object, T , and the constant temperature <strong>of</strong> the surrounding medium where it is placed,<br />
Ts, evolves in time t following the equation<br />
(∆T ) ′ = −k(∆T ), T (0) = T0, k > 0.<br />
This is the exponential decay equation for the function ∆T . We know that the solution is<br />
(∆T )(t) = (∆T )0 e −kt , that is,<br />
(T − Ts)(t) = (T0 − Ts) e −kt ⇒ T (t) = (T0 − Ts) e −kt + Ts.<br />
where the constant k depends on the material and the surroundings.<br />
Example 1.5.2: A cup with water at 45 C is placed in the cooler held at 5 C. If after 2<br />
minutes the water temperature is 25 C, when will the water temperature be 15 C? while<br />
Solution: We know that T (t) = (T0 − Ts) e −kt + Ts, and also<br />
T0 = 45, Ts = 5, T (2) = 25.<br />
Find t1 such that T (t1) = 15. First we find k,<br />
Now recall that T (2) = 25 C, that is,<br />
T (t) = (45 − 5) e −kt + 5 ⇒ T (t) = 40 e −kt + 5.<br />
20 = T (2) = 40 e −2k ⇒ ln(1/2) = −2k ⇒ k = 1<br />
2 ln(2).<br />
Having the constant k it is simple to find the time t1 such that T (t1) = 15 C. Indeed,<br />
T (t) = 40 e −t ln(√ 2) + 5 ⇒ 10 = 40 e −t1 ln( √ 2) ⇒ t1 = 4.<br />
1.5.3. Salt in a water tank. Consider the situation described in Fig. 2. The tank contains<br />
a volume V (t) <strong>of</strong> water at a time t with a salt mass Q(t) dissolved in it. Water is pouring<br />
into the tank at a rate ri(t) having salt concentration qi(t). Water is also leaving the tank<br />
at a rate ro(t) with salt concentration qo(t). We recall that a rate r represents water volume<br />
per unit time, while a concentration q represent salt mass per unit volume. We finally<br />
assume that there is a mixing mechanism in the tank such that the salt entering the tank<br />
is instantaneously mixed. This mixing mechanism then implies that the salt concentration<br />
in the tank is homogeneous at every time.<br />
⊳
water<br />
salt<br />
pipe<br />
V (t)<br />
Q (t)<br />
G. NAGY – ODE July 2, 2013 37<br />
r i<br />
q (t)<br />
i<br />
instantaneously mixed<br />
tank<br />
Figure 2. Description <strong>of</strong> the water tank problem.<br />
r<br />
o<br />
q (t)<br />
Before stating the problems we are interested in solving, we review the physical units <strong>of</strong><br />
the main fields involved in the problem. Denoting by [ri] the units <strong>of</strong> the quantity ri, then<br />
we have<br />
[ri] = [ro] = Volume<br />
Time , [qi] = [qo] = Mass<br />
, [V ] = Volume, [Q] = Mass.<br />
Volume<br />
1.5.4. The Main <strong>equations</strong>. The mass conservation law provides the main <strong>equations</strong> <strong>of</strong><br />
the mathematical description <strong>of</strong> the experimental situation above. The main <strong>equations</strong> are<br />
the following,<br />
V ′ (t) = ri(t) − ro(t), (1.5.1)<br />
Q ′ (t) = ri(t) qi(t) − ro(t), qo(t), (1.5.2)<br />
qo(t) = Q(t)<br />
, (1.5.3)<br />
V (t)<br />
r ′ i(t) = r ′ o(t) = 0. (1.5.4)<br />
The first and second <strong>equations</strong> above are just the mass conservation <strong>of</strong> water and salt,<br />
respectively. Water volume and mass are proportional, so both are conserved, and we<br />
chose the volume to write down this conservation in Eq. (1.5.1). This equation is indeed<br />
a conservation because it says that the water volume variation in time is equal to the<br />
difference <strong>of</strong> volume time rates coming in and going out <strong>of</strong> the tank. Eq. (1.5.2) is the salt<br />
mass conservation, since the salt mass variation in time is equal to the difference <strong>of</strong> the<br />
salt mass time rates coming in and going out <strong>of</strong> the tank. The product <strong>of</strong> a rate r times<br />
a concentration q has units <strong>of</strong> mass per time and represents the amount <strong>of</strong> salt entering or<br />
leaving the tank per unit time. Eq.(1.5.3) is implied by the instantaneous mixing mechanism<br />
in the tank. Since the salt is mixed instantaneously in the tank, the salt concentration in<br />
the tank is homogeneous with value Q(t)/V (t). Finally the <strong>equations</strong> in (1.5.4) say that<br />
both rates in and out are time independent, that is, constants.<br />
We now study the solutions <strong>of</strong> Eqs. (1.5.1)-(1.5.4). Since Eq. (1.5.4) says that the water<br />
rates are constant, we denote them as ri and ro. This information in Eq. (1.5.1) implies<br />
that V ′ (t) is constant, so we can easily integrate this equation to obtain<br />
V (t) = (ri − ro) t + V0, (1.5.5)<br />
o
38 G. NAGY – ODE july 2, 2013<br />
where V0 = V (0) is the water volume in the tank at the initial time t = 0. On the other<br />
hand, Eqs.(1.5.2) and (1.5.3) imply that<br />
Q ′ (t) = ri qi(t) − ro<br />
V (t) Q(t).<br />
Since V (t) is known from Eq. (1.5.5), we conclude that the function Q must be solution <strong>of</strong><br />
the <strong>differential</strong> equation<br />
Q ′ (t) = ri qi(t) −<br />
Q(t).<br />
(ri − ro) t + V0<br />
This is a linear ODE for the function Q. Indeed, introducing the functions<br />
the <strong>differential</strong> equation for Q has the form<br />
ro<br />
ro<br />
a(t) =<br />
, b(t) = ri qi(t), (1.5.6)<br />
(ri − ro) t + V0<br />
Q ′ (t) = −a(t) Q(t) + b(t). (1.5.7)<br />
Solutions <strong>of</strong> this equation can be obtained with the integrating factor method.<br />
1.5.5. Predictions in particular cases. Instead <strong>of</strong> finding the solution Q <strong>of</strong> Eq. (1.5.7)<br />
for the most general functions a and b given in Eq. (1.5.6), we find solutions in particular<br />
cases given by specific choices <strong>of</strong> the rate constants ri, ro, the concentration function qi, and<br />
the initial data constants V0 and Q0 = Q(0). The study <strong>of</strong> solutions to Eq. (1.5.7) in several<br />
particular cases might provide a deeper understanding <strong>of</strong> the physical situation under study<br />
than the expression <strong>of</strong> the solution Q in the most general case.<br />
Example 1.5.3: Consider the particular case given by the rates ri = ro = r, the incoming<br />
concentration function qi is constant, and the initial water volume in the tank V0 is given.<br />
Then, find the solution to the IVP<br />
Q ′ (t) = −a(t) Q(t) + b(t), Q(0) = Q0,<br />
where function a and b are given in Eq. (1.5.6). Graph the solution function Q for different<br />
values <strong>of</strong> the initial condition Q0.<br />
Solution: The assumption ri = ro = r implies that the function a is constant, while the<br />
assumption that qi is constant implies that the function b is also constant,<br />
ro<br />
a(t) =<br />
(ri − ro) t + V0<br />
⇒ a(t) = r<br />
V0<br />
= a0,<br />
b(t) = ri qi(t) ⇒ b(t) = ri qi = b0.<br />
We then must solve the IVP for the following constant coefficients linear ODE,<br />
Q ′ (t) = −a0 Q(t) + b0, Q(0) = Q0,<br />
The solution is found using the integrating factor method, the result was found in Theorem<br />
1.1.5 and is given by<br />
<br />
Q(t) = Q0 − b0<br />
<br />
e −a0t + b0<br />
.<br />
a0<br />
In our case the we can evaluate the constant b0/a0, and the result is<br />
b0<br />
a0<br />
<br />
V0<br />
= (rqi)<br />
r<br />
⇒ b0<br />
a0<br />
= qiV0.<br />
Then, the solution Q has the form,<br />
Q(t) = −rt/V0<br />
Q0 − qiV0 e + qiV0. (1.5.8)<br />
a0
G. NAGY – ODE July 2, 2013 39<br />
The initial amount <strong>of</strong> salt Q0 in the tank can be any non-negative real number. The solution<br />
behaves differently for different values <strong>of</strong> Q0. We classify these values in three classes:<br />
(a) The initial amount <strong>of</strong> salt in the tank has the critical value Q0 = qiV0. In this case the<br />
solution Q remains constant equal to this critical value, that is, Q(t) = qiV0.<br />
(b) The initial amount <strong>of</strong> salt in the tank is bigger than the critical value, that is, Q0 > qiV0.<br />
In this case the salt in the tank Q(t) decreases exponentially towards the critical value.<br />
(c) The initial amount <strong>of</strong> salt in the tank is smaller than the critical value, that is, Q0 < qiV0.<br />
In this case the salt in the tank Q(t) increases exponentially towards the critical value.<br />
The graphs <strong>of</strong> few solutions in these three classes are plotted in Fig. 3. ⊳<br />
Q<br />
Q<br />
Q<br />
Q<br />
Q<br />
0<br />
0<br />
Q = q v<br />
0 i 0<br />
0<br />
0<br />
Figure 3. The graph <strong>of</strong> the function q given in Eq. (1.5.8) for different<br />
values <strong>of</strong> the initial condition Q0.<br />
Example 1.5.4: Consider the particular case where the rates ri = ro = r, fresh water is<br />
coming into the tank, that is, qi = 0. Then, find the time t1 such that the salt concentration<br />
in the tank Q(t)/V (t) is 1% the initial value.<br />
Solution: The first step to find the time t1 is to find the solution Q(t) to the IVP<br />
Q ′ (t) = −a(t) Q(t) + b(t), Q(0) = Q0,<br />
where function a and b are given in Eq. (1.5.6). In this case we have<br />
ro<br />
a(t) =<br />
(ri − ro) t + V0<br />
⇒ a(t) = r<br />
,<br />
V0<br />
b(t) = ri qi(t) ⇒ b(t) = 0.<br />
The IVP we need to solve is<br />
The solution is given by<br />
Q ′ (t) = − r<br />
V0<br />
Q(t), Q(0) = Q0.<br />
Q(t) = Q0 e −rt/V0 .<br />
We can now proceed to find the time t1. We first need to find the concentration Q(t)/V (t).<br />
This is simple to do, since we already have Q(t), and V (t) = V0, since ri = ro. Therefore,<br />
The condition that defines t1 is<br />
Q(t) Q(t)<br />
= =<br />
V (t) V0<br />
Q0<br />
e<br />
V0<br />
−rt/V0 .<br />
Q(t1) 1<br />
=<br />
V (t1) 100<br />
Q0<br />
V0<br />
.<br />
t
40 G. NAGY – ODE july 2, 2013<br />
From these two <strong>equations</strong> above we conclude that<br />
1 Q0<br />
=<br />
100 V0<br />
Q(t1) Q0<br />
= e<br />
V (t1) V0<br />
−rt1/V0 .<br />
The time t1 is simple to obtain, since<br />
1<br />
100 = e−rt1/V0 ⇔<br />
<br />
1<br />
<br />
ln<br />
100<br />
The final result is given by<br />
= − rt1<br />
V0<br />
t1 = V0<br />
r ln(100).<br />
⇔ ln(100) = rt1<br />
.<br />
Example 1.5.5: Consider the particular case where the rates ri = ro = r; there is initially<br />
fresh water in the tank, that is, Q0 = 0; the initial water volume in the tank V0 is given;<br />
and the incoming salt concentration is the function<br />
Then, find the salt in the tank Q(t).<br />
qi(t) = 2 + sin(2t).<br />
Solution: We need to find the solution Q(t) to the IVP<br />
Q ′ (t) = −a(t) Q(t) + b(t), Q(0) = 0,<br />
where function a and b are given in Eq. (1.5.6). In this case we have<br />
ro<br />
a(t) =<br />
(ri − ro) t + V0<br />
⇒ a(t) = r<br />
b(t) = ri qi(t) ⇒<br />
= a0,<br />
V0<br />
b(t) = r 2 + sin(2t) .<br />
The IVP we need to solve is<br />
Q ′ (t) = −a0 Q(t) + b(t), Q(0) = 0.<br />
The solution is computed using the integrating factor method and the result is<br />
Q(t) = e −a0t<br />
t<br />
0<br />
e a0s b(s) ds,<br />
where we already introduced the initial condition Q0 = 0. Introducing the form <strong>of</strong> the<br />
function b we obtain the expression<br />
Q(t) = e −a0t<br />
t<br />
0<br />
e a0s 2 + sin(2s) ds.<br />
This is the solution <strong>of</strong> the problem. We only need to compute the integral given in the<br />
equation above. This is not a difficult calculation, although it is not straightforward. We<br />
start with the following integral found in an integration table,<br />
0<br />
<br />
e ks sin(ls) ds = eks<br />
k2 + l2 <br />
k sin(ls) − l cos(ls) ,<br />
where k and l are constants. Therefore,<br />
t<br />
e a0s 2 + sin(2s) <br />
2<br />
ds = e a0s t a0s<br />
e<br />
+<br />
a0<br />
0<br />
a2 0 + 22<br />
= 2 a0t e<br />
e − 1 +<br />
a0<br />
a0t<br />
a2 0 + 22<br />
V0<br />
<br />
a0 sin(2s) − 2 cos(2s) t <br />
,<br />
a0 sin(2t) − 2 cos(2t) +<br />
0<br />
a 2 0<br />
2<br />
.<br />
+ 22<br />
⊳
G. NAGY – ODE July 2, 2013 41<br />
With the integral above we can compute the solution Q as follows,<br />
Q(t) = e −a0t 2 a0t<br />
e − 1 +<br />
ea0t <br />
a0 sin(2t) − 2 cos(2t) +<br />
a0<br />
a2 0 + 22<br />
recalling that a0 = r/V0. We rewrite expression above as follows,<br />
Q(t) = 2 1 <br />
+ a0 sin(2t) − 2 cos(2t)<br />
+ 22<br />
<br />
2 2<br />
+ −<br />
+ 22<br />
a0<br />
a 2 0<br />
y<br />
1<br />
Q ( t )<br />
− 2 t<br />
1 − ( 3 / 4 ) e<br />
a 2 0<br />
t<br />
a0<br />
2<br />
a2 0 + 22<br />
<br />
,<br />
<br />
e −a0t . (1.5.9)<br />
Figure 4. The graph <strong>of</strong> the function Q given in Eq. (1.5.9) for a0 = 2.<br />
⊳
42 G. NAGY – ODE july 2, 2013<br />
1.5.6. Exercises.<br />
1.5.1.- A radioactive material decays at<br />
a rate proportional to the amount<br />
present. Initially there is 50 milligrams<br />
<strong>of</strong> the material present and after one<br />
hour the material has lost 80% <strong>of</strong> its<br />
original mass.<br />
(a) Find the mass <strong>of</strong> the material as<br />
function <strong>of</strong> time.<br />
(b) Find the mass <strong>of</strong> the material after<br />
four hours.<br />
(c) Find the half-life <strong>of</strong> the material.<br />
1.5.2.- A tank initially contains V0 = 100<br />
liters <strong>of</strong> water with Q0 = 25 grams <strong>of</strong><br />
salt. The tank is rinsed with fresh water<br />
flowing in at a rate <strong>of</strong> ri = 5 liters<br />
per minute and leaving the tank at the<br />
same rate. The water in the tank is wellstirred.<br />
Find the time such that the<br />
amount the salt in the tank is Q1 = 5<br />
grams.<br />
1.5.3.- A tank initially contains V0 = 100<br />
liters <strong>of</strong> pure water. Water enters the<br />
tank at a rate <strong>of</strong> ri = 2 liters per minute<br />
with a salt concentration <strong>of</strong> q1 = 3<br />
grams per liter. The instantaneously<br />
mixed mixture leaves the tank at the<br />
same rate it enters the tank. Find the<br />
salt concentration in the tank at any<br />
time t 0. Also find the limiting<br />
amount <strong>of</strong> salt in the tank in the limit<br />
t → ∞.<br />
1.5.4.- A tank with a capacity <strong>of</strong> Vm = 500<br />
liters originally contains V0 = 200 liters<br />
<strong>of</strong> water with Q0 = 100 grams <strong>of</strong> salt<br />
in solution. Water containing salt with<br />
concentration <strong>of</strong> qi = 1 gram per liter<br />
is poured in at a rate <strong>of</strong> ri = 3 liters<br />
per minute. The well-stirred water is<br />
allowed to pour out the tank at a rate<br />
<strong>of</strong> ro = 2 liter per minute. Find the<br />
salt concentration in the tank at the<br />
time when the tank is about to overflow.<br />
Compare this concentration with the<br />
limiting concentration at infinity time<br />
if the tank had infinity capacity.
G. NAGY – ODE July 2, 2013 43<br />
1.6. Non-linear <strong>equations</strong><br />
This short Section intends to highlight the main differences between solutions to linear<br />
<strong>differential</strong> <strong>equations</strong> and solutions to non-linear <strong>differential</strong> <strong>equations</strong>. The former <strong>equations</strong><br />
were studied in Section 1.2 and several examples <strong>of</strong> non-linear <strong>equations</strong> were studied<br />
in Sections 1.3 and 1.4. The plan here is first, to review the main result on solutions to<br />
an initial value problem involving a linear ODE, already given in Theorem 1.2.4. From<br />
this result we highlight the main properties shared by solutions <strong>of</strong> linear ODE, listed in the<br />
items (a)-(c) below. Then we define what is a non-linear ODE and we state that none <strong>of</strong> the<br />
properties in (a)-(c) below hold for solutions <strong>of</strong> non-linear ODE. We prove this statement<br />
showing three examples <strong>of</strong> non-linear ODE with solutions having the properties listed in<br />
items (i)-(iii) below, respectively. We finally state in Theorem 1.6.2 below, without pro<strong>of</strong>,<br />
the main result about existence and uniqueness <strong>of</strong> solutions to an initial value problem<br />
involving a non-linear ODE.<br />
1.6.1. On solutions <strong>of</strong> linear <strong>differential</strong> <strong>equations</strong>. The initial value problem for a<br />
linear <strong>differential</strong> equation is the following: Given continuous functions a, b : (t1, t2) → R,<br />
with t2 > t1, and constants t0 ∈ (t1, t2), y0 ∈ R, find the function y : (t1, t2) → R solution<br />
<strong>of</strong> the <strong>equations</strong><br />
y ′ = −a(t) y + b(t), y(t0) = y0. (1.6.1)<br />
The main result regarding solutions to this problem is summarized in Theorem 1.2.4, which<br />
we reproduce it below.<br />
Theorem 1.2.4 (Variable coefficients). Given continuous functions a, b : (t1, t2) → R<br />
and constants t 0 ∈ (t 1, t 2) and y 0 ∈ R, the initial value problem<br />
has the unique solution y : (t1, t2) → R given by<br />
y ′ = a(t) y + b(t), y(t0) = y0, (1.2.6)<br />
y(t) = e A(t)<br />
y0 +<br />
where we introduced the function A(t) =<br />
t<br />
t0<br />
t<br />
t0<br />
a(s) ds.<br />
e −A(s) <br />
b(s) ds , (1.2.7)<br />
We highlight now three main properties that solutions to a linear initial value problem<br />
have according to Theorem 1.2.4.<br />
Properties <strong>of</strong> solutions to linear <strong>differential</strong> <strong>equations</strong>:<br />
(a) There is an explicit expression for the solutions <strong>of</strong> a linear IVP, given in Eq. (1.2.7).<br />
(b) For every initial condition y 0 ∈ R there exists a unique solution to a linear IVP.<br />
(c) For every initial condition y 0 ∈ R the corresponding solution y(t) <strong>of</strong> a linear IVP is<br />
defined for all t ∈ (t1, t2).<br />
None <strong>of</strong> these properties hold for solutions <strong>of</strong> non-linear <strong>differential</strong> <strong>equations</strong>. We will<br />
see that there exist non-linear <strong>differential</strong> <strong>equations</strong> having solutions that cannot be written<br />
in explicit form. We will also see that an initial value problem for a non-linear <strong>differential</strong><br />
equation has two solutions instead <strong>of</strong> a unique solution. We will finally see that the solution<br />
to an initial value problem for a non-linear equation is defined in a domain that depends on<br />
the initial data.
44 G. NAGY – ODE july 2, 2013<br />
1.6.2. On solutions <strong>of</strong> non-linear <strong>differential</strong> <strong>equations</strong>. We now state more precisely<br />
what is a non-linear <strong>ordinary</strong> <strong>differential</strong> equation and we then present the main result about<br />
the solutions <strong>of</strong> an IVP for a non-linear ODE.<br />
Definition 1.6.1. An <strong>ordinary</strong> <strong>differential</strong> equation y ′ (t) = f(t, y(t)) is called non-linear<br />
iff the function (t, u) ↦→ f(t, u) is non-linear in the second argument.<br />
Example 1.6.1:<br />
(a) The <strong>differential</strong> equation<br />
y ′ (t) = t2<br />
y 3 (t)<br />
is non-linear, since the function f(t, u) = t 2 /u 3 is non-linear in the second argument.<br />
(b) The <strong>differential</strong> equation<br />
y ′ (t) = 2ty(t) + ln y(t) <br />
is non-linear, since the function f(t, u) = 2tu + ln(u) is non-linear in the second argument,<br />
due to the term ln(u).<br />
(c) The <strong>differential</strong> equation<br />
y ′ (t)<br />
= 2t2<br />
y(t)<br />
is linear, since the function f(t, u) = 2t 2 u is linear in the second argument.<br />
The initial value problem involving a non-linear ODE has a unique solution in the case<br />
that the ODE meets certain conditions. The precise statement is summarized below.<br />
Theorem 1.6.2 (Non-linear ODE). Fix a non-empty rectangle R = (t1, t2)×(u1, u2) ⊂ R 2<br />
and fix a function f : R → R denoted as (t, u) ↦→ f(t, u). If the function f and its derivative<br />
∂uf are continuous on R, and the point (t0, y0) ∈ R, then there exists a smaller open<br />
rectangle ˆ R ⊂ R with (t0, y0) ∈ ˆ R such that the IVP<br />
has a unique solution y on the set ˆ R ⊂ R 2 .<br />
y ′ (t) = f(t, y(t)), y(t0) = y0 (1.6.2)<br />
We do not provide the pro<strong>of</strong> <strong>of</strong> this Theorem, since it requires ideas beyond the level <strong>of</strong><br />
these notes. This statement is the generalization <strong>of</strong> Theorem 1.2.4 from linear to non-linear<br />
ODE. We now highlight three main properties <strong>of</strong> the solutions to non-linear ODE, properties<br />
that are allowed by the statement in Theorem 1.6.2.<br />
Properties <strong>of</strong> solutions to non-linear <strong>differential</strong> <strong>equations</strong>:<br />
(i) There is no general explicit expression for the solution y(t) to a non-linear <strong>differential</strong><br />
equation.<br />
(ii) Non-uniqueness <strong>of</strong> solution to the IVP in Eq. (1.6.2) may happen at points (t, u) ∈ R 2<br />
where ∂uf is not continuous.<br />
(iii) Changing the initial data y 0 in Eq. (1.6.2) may change the domain on the variable t<br />
where the solution y(t) is defined.<br />
The next three examples (1.6.2)-(1.6.4) below address the statements in (i)-(iii), respectively.<br />
In the first example we have an equation whose solution cannot be written in explicit<br />
form. The reason is not lack <strong>of</strong> ingenuity, it is actually proven that such explicit expression<br />
does not exist.<br />
⊳
G. NAGY – ODE July 2, 2013 45<br />
Example 1.6.2: Given non-zero constants a1, a2, a3, a4, find every solution y to the ODE<br />
y ′ (t) =<br />
t2 <br />
y4 (t) + a4 y3 (t) + a3 y2 . (1.6.3)<br />
(t) + a2 y(t) + a1<br />
Solution: We see that the non-linear <strong>differential</strong> equation above is separable, so we can<br />
find solutions following the ideas given in Sect. 1.3. That is, we first rewrite the equation as<br />
4<br />
y (t) + a4 y 3 (t) + a3 y 2 ′ 2<br />
(t) + a2 y(t) + a1 y (t) = t ,<br />
then we integrate on both sides <strong>of</strong> the equation,<br />
<br />
y4 (t) + a4 y 3 (t) + a3 y 2 <br />
′<br />
(t) + a2 y(t) + a1 y (t) dt =<br />
Introduce the substitution u = y(t), so du = y ′ (t) dt,<br />
<br />
(u 4 + a4 u 3 + a3 u 2 <br />
<br />
+ a2 u + a1 du =<br />
t 2 dt + c.<br />
t 2 dt + c.<br />
Integrate and in the result substitute back the function y, hence we obtain<br />
1<br />
5 y5 (t) + a4<br />
4 y4 (t) + a3<br />
3 y3 (t) + a2<br />
2 y(t) + a1 y(t) = t3<br />
+ c.<br />
3<br />
This is an implicit form for the solution y <strong>of</strong> the problem, given by the root <strong>of</strong> a polynomial<br />
degree five. Since there is no formula for the roots <strong>of</strong> a general polynomial degree bigger or<br />
equal five, we see that there is no explicit expression for solutions y <strong>of</strong> Eq. (1.6.3). ⊳<br />
We now give an example <strong>of</strong> the statement in (ii). We consider a <strong>differential</strong> equation<br />
defined by a function f that does not satisfy one hypothesis in Theorem 1.6.2. The function<br />
values f(t, u) have a discontinuity at a line in the (t, u) plane where the initial condition for<br />
the initial value problem is given. We then show that such initial value problem has two<br />
solutions instead <strong>of</strong> a unique solution.<br />
Example 1.6.3: Find every solution y <strong>of</strong> the initial value problem<br />
y ′ (t) = y 1/3 (t), y(0) = 0. (1.6.4)<br />
Remark: The equation above is non-linear and separable, and the function f(t, u) = u1/3 has derivative<br />
∂uf = 1 1<br />
,<br />
3 u2/3 so the function ∂uf is not continuous at u = 0.<br />
Solution: The solution to the initial value problem in Eq. (1.6.4) exists but it is not unique,<br />
since we now show that it has two solutions. The first solution is<br />
y 1(t) = 0.<br />
The second solution can be computed as using the ideas from separable <strong>equations</strong>, that is,<br />
<br />
y(t) −1/3 ′<br />
y (t) dt = dt + c0.<br />
Then, the substitution u = y(t), with du = y ′ (t) dt, implies that<br />
<br />
u −1/3 <br />
du = dt + c0. Integrate and substitute back the function y. The result is<br />
3<br />
2/3 y(t) = t + c0,<br />
2
46 G. NAGY – ODE july 2, 2013<br />
so the solution is<br />
The initial condition above implies<br />
so the second solution is:<br />
y(t) =<br />
0 = y(0) =<br />
<br />
2<br />
3 (t + c 3/2 0) .<br />
<br />
2<br />
3 c 3/2 0<br />
y2(t) =<br />
<br />
2<br />
3 t<br />
3/2 .<br />
⇒ c 0 = 0,<br />
Finally, an example <strong>of</strong> the statement in (iii). We show that the solution <strong>of</strong> an initial value<br />
problem for a non-linear equation is defined in a domain that depends on the initial data.<br />
Example 1.6.4: Find the solution y to the initial value problem<br />
y ′ (t) = y 2 (t), y(0) = y 0.<br />
Solution: This is a non-linear separable equation, so we can again apply the ideas in<br />
Sect. 1.3. Performing all the calculations in one line, we summarize them as follows,<br />
′ y (t) dt<br />
y2 (t) =<br />
<br />
dt + c0 ⇒ − 1<br />
y(t) = t + c0 ⇒ y(t) = − 1<br />
c0 + t .<br />
Using the initial condition in equation above we obtain<br />
y0 = y(0) = − 1<br />
⇒ c0 = − 1<br />
.<br />
c0<br />
So the solution <strong>of</strong> the initial value problem above is:<br />
1<br />
y(t) = <br />
1<br />
.<br />
− t<br />
y0 This solution diverges at t = 1/y0, so the domain <strong>of</strong> the solution y is not the whole real line<br />
R. Instead, the domain is R − {y0}, so it depends on the values <strong>of</strong> the initial data y0. ⊳<br />
The last example tries to help understand the continuity hypothesis given in Theorem<br />
1.6.2 on the nonlinear function f that defines the <strong>differential</strong> equation.<br />
Example 1.6.5: Consider the non-linear IVP:<br />
y ′ 1<br />
(t) =<br />
(t − 1)(t + 1)(y(t) − 2)(y(t) + 3) , y(t0) = y0. (1.6.5)<br />
Find the regions on the plane R 2 where Theorem 1.6.2 does not hold.<br />
Solution: In this case the function f is given by:<br />
1<br />
f(t, u) =<br />
, (1.6.6)<br />
(t − 1)(t + 1)(u − 2)(u + 3)<br />
therefore, f is not defined on the lines<br />
t = 1, t = −1, u = 2, u = −3.<br />
where we denote (t, u) ∈ R 2 . See Fig. 5. For example, in the case that the initial data is<br />
t 0 = 0, y 0 = 1, then Theorem 1.6.2 implies that there exists a unique solution on any region<br />
ˆR contained in the rectangle R = (−1, 1) × (−3, 2). On the other hand, if the initial data<br />
for the IVP in Eq. (1.6.5) is t = 0, y0 = 2, then the hypothesis <strong>of</strong> Theorem 1.6.2 are not<br />
satisfied. ⊳<br />
y0<br />
⊳
G. NAGY – ODE July 2, 2013 47<br />
t=−1<br />
−1<br />
u<br />
1<br />
−1<br />
(0,2)<br />
(0,1)<br />
R<br />
1<br />
t=1<br />
u=2<br />
u=−3<br />
Figure 5. Regions in the plane where the function f defined in Eq. (1.6.6)<br />
is not defined.<br />
Summary: Both Theorems 1.2.4 and 1.6.2 state that there exist solutions to linear and<br />
non-linear <strong>differential</strong> <strong>equations</strong>, respectively. However, Theorem 1.2.4 provides more information<br />
about the solutions to a reduced type <strong>of</strong> <strong>equations</strong>, linear problems; while Theorem<br />
1.6.2 provides less information about solutions to wider type <strong>of</strong> <strong>equations</strong>, linear and<br />
non-linear. Once again:<br />
• Linear <strong>differential</strong> <strong>equations</strong>:<br />
(a) There is an explicit expression for the solution <strong>of</strong> a linear IVP.<br />
(b) For every initial condition y0 ∈ R there exists a unique solution to a linear IVP.<br />
(c) The domain <strong>of</strong> the solution <strong>of</strong> a linear IVP is defined for every initial condition<br />
y0 ∈ R.<br />
• Non-linear <strong>differential</strong> <strong>equations</strong>:<br />
(i) There is no general explicit expression for the solution y(t) to a non-linear<br />
<strong>differential</strong> equation.<br />
(ii) Non-uniqueness <strong>of</strong> solution to a non-linear IVP may happen at points (t, u) ∈<br />
R 2 where ∂uf is not continuous.<br />
(iii) Changing the initial data y0 may change the domain on the variable t where<br />
the solution y(t) is defined.<br />
1.6.3. Direction Fields. One conclusion <strong>of</strong> the discusion in this section is that non-linear<br />
<strong>differential</strong> <strong>equations</strong> are more difficult to solve that the linear ones. It is important to<br />
develop methods to obtain any type <strong>of</strong> information from the solution <strong>of</strong> a <strong>differential</strong> equation<br />
without having to actually solve the equation. One <strong>of</strong> such method is based on the idea <strong>of</strong><br />
direction fields. We focus this discussion on <strong>differential</strong> <strong>equations</strong> <strong>of</strong> the form<br />
y ′ (t) = f(t, y(t)).<br />
One does not need to solve the <strong>differential</strong> equation above to have a qualitative idea <strong>of</strong> the<br />
solution. We only need to recall that y ′ (t) represents the slope <strong>of</strong> the tangent line to the<br />
graph <strong>of</strong> function y at the point (t, y(t)) in the yt-plane. Therefore, what the <strong>differential</strong><br />
equation above provides is all these slopes, f(t, y(t)), for every point (t, y(t)) in the yt-plane.<br />
The Key idea is this: Graph the function f(t, y) on the yt-plane, not as points, but as slopes<br />
<strong>of</strong> small segments.<br />
t
48 G. NAGY – ODE july 2, 2013<br />
Definition 1.6.3. A direction field for the <strong>differential</strong> equation y ′ (t) = f(t, y(t)) is the<br />
graph on the yt-pane <strong>of</strong> the values f(t, y) as slopes <strong>of</strong> a small segments.<br />
Example 1.6.6: We know that the solution <strong>of</strong> y ′ = y are given by the exponentials y(t) =<br />
y0 e t , for ever constant y0 ∈ R. The graph <strong>of</strong> these solution is simple. So is the direction<br />
field shown in Fig. 6. ⊳<br />
Figure 6. Direction field for the equation y ′ = y.<br />
Example 1.6.7: The solution <strong>of</strong> y ′ = sin(y) is simple to compute. The equation is separable.<br />
After some calculations the implicit solution are<br />
<br />
<br />
ln csc(y0) + cot(y)<br />
<br />
<br />
= t.<br />
csc(y) + cot(y)<br />
for y0 ∈ R. The graph <strong>of</strong> these solution is not simple to do. But the direction field is simple<br />
to plot and can be seen in Fig. 7. ⊳<br />
Example 1.6.8: The solution <strong>of</strong> y ′ = (1 + y3 )<br />
(1 + t2 could be hard to compute. But the direction<br />
)<br />
field is simple to plot and can be seen in Fig. 8. ⊳
G. NAGY – ODE July 2, 2013 49<br />
Figure 7. Direction field for the equation y ′ = sin(y).<br />
Figure 8. Direction field for the equation y ′ = (1 + y 3 )/(1 + t 2 ).
50 G. NAGY – ODE july 2, 2013<br />
1.6.4. Exercises.<br />
1.6.1.- By looking at the equation coefficients,<br />
find a domain where the solution<br />
<strong>of</strong> the initial value problem below exists,<br />
(a) (t 2 −4) y ′ +2 ln(t) y = 3t, and initial<br />
condition y(1) = −2.<br />
(b) y ′ y<br />
= , and initial condition<br />
t(t − 3)<br />
y(−1) = 2.<br />
1.6.2.- State where in the plane with points<br />
(t, y) the hypothesis <strong>of</strong> Theorem 1.6.2<br />
are not satisfied.<br />
(a) y ′ =<br />
y2<br />
2t − 3y .<br />
(b) y ′ = 1 − t2 − y2 .<br />
1.6.3.- Find the domain where the solution<br />
<strong>of</strong> the initial value problems below is<br />
well-defined.<br />
(a)<br />
y ′ = −4t<br />
, y(0) = y0 > 0.<br />
y<br />
(b)<br />
y ′ = 2ty 2 , y(0) = y0 > 0.
G. NAGY – ODE July 2, 2013 51<br />
Chapter 2. Second order linear <strong>equations</strong><br />
In Chapter 1 we studied methods to find solutions to particular types <strong>of</strong> first order<br />
<strong>differential</strong> <strong>equations</strong>. A first oder <strong>differential</strong> equation is an equation where besides the<br />
function only its first derivative appear in the equation. In this Chapter we study second<br />
order <strong>differential</strong> <strong>equations</strong>, that is, <strong>differential</strong> <strong>equations</strong> where the second derivative <strong>of</strong> the<br />
unknown function also appears in the equation. The whole Chapter is dedicated to study<br />
linear second order <strong>equations</strong> only. This observation suggests that second order <strong>equations</strong><br />
are more complicated to solve than first order <strong>equations</strong>, since we studied and solved linear<br />
first order <strong>equations</strong> in just the first Section <strong>of</strong> Chapter 1.<br />
2.1. Variable coefficients<br />
2.1.1. Main definitions and properties. In this Section we present the precise definition<br />
<strong>of</strong> linear second order <strong>differential</strong> <strong>equations</strong> and few properties <strong>of</strong> their solutions. The central<br />
result, Theorem 2.1.6, asserts that there exist solutions to this type <strong>of</strong> <strong>equations</strong> in the case<br />
that the equation coefficients are continuous functions. We present the statement without<br />
pro<strong>of</strong>, since the pro<strong>of</strong> requires functional analysis techniques beyond the scope <strong>of</strong> this class.<br />
Definition 2.1.1. Given functions a1, a0, b : (t1, t2) → R, the <strong>differential</strong> equation in the<br />
unknown function y : (t1, t2) → R given by<br />
y ′′ + a 1(t) y ′ + a 0(t) y = b(t) (2.1.1)<br />
is called a second order linear <strong>differential</strong> equation. The equation in (2.2.1) is called<br />
homogeneous iff for all t ∈ R holds<br />
b(t) = 0.<br />
The equation in (2.2.1) is called <strong>of</strong> constant coefficients iff a1, a0, and b are constants,<br />
otherwise the equation is called <strong>of</strong> variable coefficients.<br />
We remark that the notion <strong>of</strong> an homogeneous equation presented here is completely<br />
different from the homogeneous <strong>equations</strong> we studied in Section 1.3.<br />
Example 2.1.1:<br />
(a) A second order, linear, homogeneous, constant coefficients equation is<br />
y ′′ + 5y ′ + 6 = 0.<br />
(b) A second order order, linear, constant coefficients, non-homogeneous equation is<br />
y ′′ − 3y ′ + y = 1.<br />
(c) A second order, linear, non-homogeneous, variable coefficients equation is<br />
y ′′ + 2t y ′ − ln(t) y = e 3t .<br />
(d) Newton’s second law <strong>of</strong> motion for point particles <strong>of</strong> mass m moving in one space<br />
dimension under a force f : R → R is given by<br />
m y ′′ (t) = f(t).<br />
This is the equation that says mass times acceleration equals force. The acceleration is<br />
the second time derivative <strong>of</strong> the position function y. ⊳<br />
Solutions to linear homogeneous <strong>equations</strong> have the property that any linear combination<br />
<strong>of</strong> solutions is also a solution. Here is a more precise statement.
52 G. NAGY – ODE july 2, 2013<br />
Theorem 2.1.2 (Superposition). If the functions y 1 and y 2 are solutions to the linear<br />
homogeneous equation<br />
y ′′ + a1(t) y ′ + a0(t) y = 0, (2.1.2)<br />
then the linear combination c1y1(t) + c2y2(t) is also a solution for any constants c1, c2 ∈ R.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 2.1.2: Verify that the function y = c 1y 1 + c 2y 2 satisfies Eq. (2.1.2) for<br />
every constants c 1, c 2, that is,<br />
(c1y1 + c2y2) ′′ + a1(t)(c1y1 + c2y2) ′ + a0(t)(c1y1 + c2y2)<br />
= (c 1y ′′<br />
1 + c 2y ′′<br />
2 ) + a 1(t)(c 1y ′ 1 + c 2y ′ 2) + a 0(t)(c 1y 1 + c 2y 2)<br />
′′<br />
= c1 y 1 + a1(t)y ′ <br />
1 + a0(t)y1 + c2<br />
y ′′<br />
2 + a 1(t)y ′ 2 + a 0(t)y 2<br />
= 0.<br />
2.1.2. Wronskian and linear dependence. We now introduce the notion <strong>of</strong> linearly dependent<br />
and linearly independent functions.<br />
Definition 2.1.3. Two continuous functions y1, y2 : (t1, t2) ⊂ R → R are called linearly<br />
dependent on the interval (t1, t2) iff there exists a constant c such that for all t ∈ I holds<br />
y1(t) = c y2(t).<br />
The two functions are called linearly independent on the interval (t1, t2) iff they are not<br />
linearly dependent.<br />
We see that two functions are linearly dependent on an interval iff the functions are<br />
proportional to each other on that interval, and they are linearly independent if they are<br />
not proportional to each other in that interval. Notice that the function zero is proportional<br />
to every function, so any set containing the zero function is linearly dependent.<br />
The definitions <strong>of</strong> linearly dependent or independent functions found in the literature is<br />
equivalent to the definition given here, but it is worded in a slight different way. One usually<br />
finds in the literature that two functions are called linearly dependent on the interval (t1, t2)<br />
iff there exist constants c1, c2, not both zero, such that for all t ∈ (t1, t2) holds<br />
c 1y 1(t) + c 2y 2(t) = 0.<br />
Two functions are called linearly independent on the interval (t1, t2) iff they are not linearly<br />
dependent, that is, the only constants c1 and c2 that for all t ∈ (t1, t2) satisfy the equation<br />
c 1y 1(t) + c 2y 2(t) = 0<br />
are the constants c1 = c2 = 0. This latter wording makes it simple to generalize these<br />
definitions to an arbitrary number <strong>of</strong> functions.<br />
Example 2.1.2:<br />
(a) Show that y 1(t) = sin(t), y 2(t) = 2 sin(t) are linearly dependent.<br />
(b) Show that y1(t) = sin(t), y2(t) = t sin(t) are linearly independent.<br />
Solution:<br />
Part (a): This is trivial, since 2y 1(t) − y 2(t) = 0.<br />
Part (b): Find constants c1, c2 such that for all t ∈ R holds<br />
c 1 sin(t) + c 2t sin(t) = 0.<br />
Evaluating at t = π/2 and t = 3π/2 we obtain<br />
c1 + π<br />
2 c2 = 0, c1 + 3π<br />
2 c2 = 0 ⇒ c1 = 0, c2 = 0.<br />
We conclude: The functions y1 and y2 are linearly independent. ⊳
G. NAGY – ODE July 2, 2013 53<br />
We now introduce a function W that determines whether two functions y 1, y 2 are linearly<br />
independent or not. This function W was first introduced by a polish scientist Josef Wronski<br />
in 1821 while studying a different problem, and it is now called the Wronskian.<br />
Definition 2.1.4. The Wronskian <strong>of</strong> functions y 1, y 2 : (t 1, t 2) → R is the function<br />
Wy1y2 (t) = y1(t)y ′ 2(t) − y ′ 1(t)y2(t).<br />
Remarks:<br />
<br />
y1(t) y2(t)<br />
(a) If we introduce the matrix-valued function A(t) =<br />
, then the Wronskian<br />
y ′ 1(t) y ′ 2(t)<br />
can be written using the determinant <strong>of</strong> a 2 × 2 matrix, that is, Wy1y2 (t) = detA(t) .<br />
An alternative notation is: Wy1y2 =<br />
<br />
<br />
<br />
y1 y2 y ′ 1 y ′ <br />
<br />
<br />
<br />
2<br />
.<br />
(b) We will see that the Wronskian determines whether two functions are linearly dependent<br />
or linearly independent.<br />
Example 2.1.3: Find the Wronskian <strong>of</strong> the functions:<br />
(a) y 1(t) = sin(t) and y 2(t) = 2 sin(t). (ld)<br />
(b) y1(t) = sin(t) and y2(t) = t sin(t). (li)<br />
Solution:<br />
Part (a): By the definition <strong>of</strong> the Wronskian:<br />
<br />
<br />
Wy1y2 (t) = <br />
y1(t)<br />
<br />
y2(t) <br />
<br />
=<br />
<br />
<br />
<br />
sin(t)<br />
2 sin(t) <br />
<br />
cos(t) 2 cos(t) = sin(t)2 cos(t) − cos(t)2 sin(t)<br />
y ′ 1(t) y ′ 2(t)<br />
We conclude that Wy1y2 (t) = 0. Notice that y1 and y2 are linearly dependent.<br />
Part (b): Again, by the definition <strong>of</strong> the Wronskian:<br />
<br />
<br />
<br />
Wy1y2 (t) = sin(t)<br />
t sin(t) <br />
<br />
cos(t) sin(t) + t cos(t) = sin(t)sin(t) + t cos(t) − cos(t)t sin(t).<br />
We conclude that Wy1y2 (t) = sin2 (t). Notice that y1 and y2 are linearly independent. ⊳<br />
We can see that the Wronskian detects whether the two functions are linearly dependent<br />
or independent. The Wronskian vanishes identically iff the functions are linearly dependent.<br />
This property is the main result <strong>of</strong> the following statement.<br />
Lemma 2.1.5 (Wronskian and linearly dependence). The continuously differentiable<br />
functions y 1, y 2 : (t 1, t 2) → R are linearly dependent iff Wy1y2 (t) = 0 holds for all t ∈ (t 1, t 2).<br />
Pro<strong>of</strong> <strong>of</strong> Lemma 2.1.5:<br />
(⇒) Since the functions y1, y2 form a linearly dependent set, then there exists a nonzero<br />
constant c such that y 1 = cy 2. Therefore,<br />
Wy1y2 (t) = c y 2 y ′ 2 − c y ′ 2 y 2 = 0.<br />
(⇐) Since the Wronskian <strong>of</strong> the non-zero functions y1, y2 vanishes, we obtain that<br />
0 = Wy1y2 = y1 y ′ 2 − y ′ 1<br />
y2 ⇔ y′ 1<br />
⇔<br />
<br />
y1(t)<br />
<br />
y2(t)<br />
<br />
ln = ln<br />
y1(t0) y2(t0)<br />
y1<br />
= y′ 2<br />
y2<br />
therefore, calling c = y1(t0)/y2(t0), we obtain that<br />
y1(t) = c y2(t).<br />
This establishes the Lemma.
54 G. NAGY – ODE july 2, 2013<br />
Example 2.1.4: Show whether the following two functions form a l.d. or l.i. set:<br />
y 1(t) = cos(2t) − 2 cos 2 (t), y 2(t) = cos(2t) + 2 sin 2 (t).<br />
Solution: In this case it is not clear whether these two functions are proportional to each<br />
other or not. They could be proportional due to a trigonometric identity. Actually this is<br />
the case, as we can see by computing their Wronskian:<br />
Wy1y2 (t) = cos(2t) − 2 cos 2 (t) −2 sin(2t) + 4 sin(t) cos(t) <br />
− − 2 sin(2t) + 4 sin(t) cos(t) cos(2t) + 2 sin 2 (t) ,<br />
but sin(2t) = 2 sin(t) cos(t), which implies that Wy1y2 (t) = 0. We conclude that the functions<br />
y1 and y2 are linearly dependent. ⊳<br />
2.1.3. Fundamental and general solutions. Second order linear <strong>differential</strong> <strong>equations</strong><br />
have solutions in the case that the equation coefficients are continuous functions. We present<br />
the statement without pro<strong>of</strong>.<br />
Theorem 2.1.6 (Variable coefficients). If a 1, a 0, b : (t 1, t 2) → R are continuous, then<br />
there exist two linearly independent solutions y 1, y 2 : (t 1, t 2) → R to the equation<br />
y ′′ + a1(t) y ′ + a0(t) y = b(t). (2.1.3)<br />
Furthermore, for every constant t0 ∈ (t1, t2) and y0, y1 ∈ R, there exists a unique solution<br />
y : (t1, t2) → R to the initial value problem given by Eq. (2.1.3) with the initial conditions<br />
y(t 0) = y 0 and y ′ (t 0) = y 1.<br />
Remarks:<br />
(a) Unlike the first order linear <strong>differential</strong> <strong>equations</strong> where we have an explicit expression<br />
for the solution, Theorem 1.2.4, there is no explicit expression for the solution <strong>of</strong> second<br />
order linear equation in the Theorem above. This suggests that second order <strong>equations</strong><br />
are more difficult to solve than first order <strong>equations</strong>.<br />
(b) We will provide a formula for solutions to second order linear <strong>equations</strong> though, in<br />
the particular case that the equation has constant coefficients. We study this case in<br />
Section 2.2 and the result is Theorem 2.2.2.<br />
(c) Two integrations must be done to find solutions to second order linear <strong>differential</strong> <strong>equations</strong>.<br />
Therefore, initial value problems with two initial conditions can have a unique<br />
solution.<br />
Example 2.1.5: Find the longest interval I ∈ R such that there exists a unique solution to<br />
the initial value problem<br />
(t − 1)y ′′ − 3ty ′ + 4y = t(t − 1), y(−2) = 2, y ′ (−2) = 1.<br />
Solution: We first write the equation above in the form given in Theorem (2.1.6,<br />
y ′′ − 3t<br />
t − 1 y′ + 4<br />
y = t.<br />
t − 1<br />
The intervals where the hypotheses in Theorem 2.1.6 are satisfied, that is, where the equation<br />
coefficients are continuous, are I1 = (−∞, 1) and I2 = (1, ∞). Since the initial condition<br />
belongs to I1, the solution domain is<br />
I 1 = (−∞, 1). ⊳
G. NAGY – ODE July 2, 2013 55<br />
The Theorem above and the Wronskian <strong>of</strong> two functions are the key ideas needed to<br />
find properties <strong>of</strong> the solutions to <strong>differential</strong> <strong>equations</strong>, in the case where the solutions are<br />
not explicitly known. For example, the only information needed from the solutions in the<br />
statement below is that they are continuous functions.<br />
Theorem 2.1.7 (Linear independent solutions). If the functions a 1, a 0 : (t 1, t 2) → R<br />
are continuous, then each initial value problem<br />
y ′′<br />
1 + a1(t) y ′ 1 + a0(t) y1 = 0, y1(0) = 1, y ′ 1(0) = 0,<br />
y ′′<br />
2 + a1(t) y ′ 2 + a0(t) y2 = 0, y2(0) = 0, y ′ 2(0) = 1,<br />
has a unique solution y1, y2 : (t1, t2) → R for any t0 ∈ (t1, t2), respectively, and these<br />
solutions are linearly independent.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 2.1.7: We know that the solutions actually exist because <strong>of</strong> Theorem<br />
2.1.6. We now only compute the Wronskian at t0, that is,<br />
Wy1y2 (t 0) = y 1(t 0) y ′ 2(t 0) − y ′ 1(t 0) y 2(t 0) = 1.<br />
Since the Wronskian is a continuous function, it will be non-zero in a neighborhood <strong>of</strong> t0.<br />
Then, Lemma 2.1.5 establishes the Theorem. <br />
The superposition property says that every linear combination y(t) = c1 y1(t) + c2 y2(t),<br />
is also a solution <strong>of</strong> the <strong>differential</strong> equation<br />
y ′′ + a1(t) y ′ + a0(t) y = 0,<br />
Conversely, one can prove that every solution y <strong>of</strong> the equation above can be written as a<br />
linear combination <strong>of</strong> the solutions y 1, y 2. These comments suggest the following definitions.<br />
Definition 2.1.8. Two linearly independent solutions y1, y2 <strong>of</strong> the homogeneous equation<br />
y ′′ + a 1(t)y ′ + a 0(t)y = 0,<br />
are called fundamental solutions <strong>of</strong> the <strong>differential</strong> equation. Given any two fundamental<br />
solutions y1, y2, and arbitrary constants c1, c2, the function<br />
y(t) = c 1 y 1(t) + c 2 y 2(t)<br />
is called the general solution <strong>of</strong> the homogeneous <strong>differential</strong> equation.<br />
Example 2.1.6: Find two linearly independent solutions to the equation<br />
y ′′ + y ′ − 2y = 0.<br />
Solution: One way is to find any two solutions, and later on verify that they are linearly<br />
independent. For example, we look for solutions <strong>of</strong> the form y(t) = ert . Then, we obtain<br />
the characteristic equation<br />
r 2 + r − 2 = 0 ⇒ r = 1<br />
<br />
√ r1 = 1<br />
−1 ± 1 + 8 ⇒<br />
2<br />
r2 = −2.<br />
Therefore, y1(t) = e t , y2(t) = e −2t are linearly independent solutions to the equation above.<br />
Another way to solve this problem is to use Theorem 2.1.7. This Theorem says that two<br />
linearly independent solutions <strong>of</strong> the equation above cane be obtained by solving the two<br />
initial value problems mentioned in that Theorem. The respective solutions are<br />
y1(t) = 2<br />
3 et + 1<br />
3 e−2t , y2(t) = 1<br />
t −2t<br />
e − e<br />
3<br />
.<br />
⊳
56 G. NAGY – ODE july 2, 2013<br />
2.1.4. The Abel Theorem. Theorem 2.1.7 says that the functions y 1 and y 2 that solve<br />
the initial values problems in that Theorem are linearly independent in a neighborhood <strong>of</strong><br />
t0. We extend now this result from a neighborhood <strong>of</strong> t0 to the whole domain (t1, t2) where<br />
the solution is defined. The extension is based in Abel’s Theorem, which also answers the<br />
second question asked earlier in this Section.<br />
Theorem 2.1.9 (Abel). Let a 1, a 0 : (t 1, t 2) → R be continuous functions, and y 1, y 2 be<br />
twice continuously differentiable solutions <strong>of</strong> the equation<br />
Then, the Wronskian Wy1y2<br />
y ′′ + a1(t) y ′ + a0(t) y = 0. (2.1.4)<br />
is a solution <strong>of</strong> the equation<br />
W ′ y1y2 (t) + a1(t) Wy1y2 (t) = 0.<br />
Therefore, for any t0 ∈ (t1, t2), the Wronskian Wy1y2<br />
where W0 = Wy1y2 (t0) and A1(t) =<br />
Wy1y2 (t) = W0 e A1(t) ,<br />
t<br />
t0<br />
a1(s) ds.<br />
is given by the expression<br />
This Theorem says that the Wronskian <strong>of</strong> two solutions <strong>of</strong> Eq. (2.1.4) is non-zero at a<br />
single point t0 ∈ (t1, t2) iff it is non-zero on the whole interval (t1, t2) where Wy1y2 is defined.<br />
In other words, if the two solutions y1, y2 are linearly independent in a neighborhood <strong>of</strong> a<br />
point t0, then they are linearly independent on the whole interval (t1, t2) where they are<br />
defined. This statement answers the second question above.<br />
Example 2.1.7: Find the Wronskian <strong>of</strong> two solutions <strong>of</strong> the equation<br />
t 2 y ′′ − t(t + 2) y ′ + (t + 2) y = 0, t > 0.<br />
Solution: Notice that we do not known the explicit expression for the solutions. Nevertheless,<br />
Theorem 2.1.9 says that we can compute their Wronskian. First, we have to rewrite<br />
the <strong>differential</strong> equation in the form given in that Theorem, namely,<br />
y ′′ <br />
2<br />
<br />
− + 1 y<br />
t ′ <br />
2 1<br />
<br />
+ + y = 0.<br />
t2 t<br />
Then, Theorem 2.1.9 says that the Wronskian satisfies the <strong>differential</strong> equation<br />
W ′ <br />
2<br />
<br />
y1y2 (t) − + 1 Wy1y2 (t) = 0.<br />
t<br />
This is a first order, linear equation for Wy1y2 , so its solution can be computed using the<br />
method <strong>of</strong> integrating factors. That is, first compute the integral<br />
t<br />
2<br />
<br />
t<br />
<br />
− + 1 ds = −2 ln − (t − t0) t0 s t0 2 t<br />
<br />
0<br />
= ln − (t − t0).<br />
Then, the integrating factor µ is given by<br />
t 2<br />
µ(t) = t2 0<br />
t 2 e−(t−t0) ,<br />
which satisfies the condition µ(t0) = 1. So the solution, Wy1y2 is given by<br />
<br />
µ(t)Wy1y2 (t)<br />
′<br />
= 0 ⇒ µ(t)Wy1y2 (t) − µ(t0)Wy1y2 (t0) = 0
so, the solution is<br />
G. NAGY – ODE July 2, 2013 57<br />
Wy1y2<br />
t2<br />
(t) = Wy1y2 (t0) e (t−t0)<br />
.<br />
If we call the constant c = Wy1y2 (t0)/[t 2 0e t0], then the Wronskian has the simpler form<br />
t 2 0<br />
Wy1y2 (t) = c t2 e t .<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 2.1.9: Recall that both y1 and y2 are solutions to Eq. (2.1.4), that is<br />
y ′′<br />
1 + a1(t) y ′ 1 + a0(t) y1 = 0<br />
y ′′<br />
2 + a1(t) y ′ 2 + a0(t) y2 = 0.<br />
Now, multiply the first equation above by −y2, the second equation by y1, and then add the<br />
resulting <strong>equations</strong> together. One obtains<br />
Since<br />
(y1 y ′′<br />
2 − y ′′<br />
1 y2) + a1(t) (y1 y ′ 2 − y ′ 1 y2) = 0.<br />
Wy1y2 = y1 y ′ 2 − y ′ 1 y2 ⇒ W ′ y1y2 = y1 y ′′<br />
2 − y ′′<br />
1 y2,<br />
we obtain the equation<br />
W ′ y1y2 + a1(t) Wy1y2 = 0.<br />
This is a first order linear equation, and its solution can be found using the method <strong>of</strong><br />
integrating factors. The result is the expression given in the Theorem. This establishes the<br />
Theorem. <br />
2.1.5. Special Second order nonlinear <strong>equations</strong>. Just for completeness, we now take<br />
the time to introduce definiton <strong>of</strong> nonlinear second order <strong>differential</strong> <strong>equations</strong>.<br />
Definition 2.1.10. Given a functions f : R 3 → R, a second order <strong>differential</strong> equation in<br />
the unknown function y : R → R is given by<br />
y ′′ = f(t, y, y ′ ).<br />
The equation is linear iff function f is linear in the arguments y and y ′ .<br />
One could say that nonlinear second order <strong>differential</strong> equation are usually more difficult<br />
to solve than first order nonlinear <strong>equations</strong>. However, there are two particular cases where<br />
second order <strong>equations</strong> can be transformed into first order <strong>equations</strong>.<br />
(a) y ′′ = f(t, y ′ ). The function y is missing.<br />
(b) y ′′ = f(y, y ′ ). The independent variable t is missing.<br />
It is simple to solve diffenrential <strong>equations</strong> in case (a), where function y is missing. Indeed,<br />
if second order <strong>differential</strong> equation has the form y ′′ = f(t, y ′ ), then the equation for v = y ′<br />
is the first order equation v ′ = f(t, v).<br />
Example 2.1.8: Find the y solution <strong>of</strong> the second order nonlinear equation y ′′ = −2t (y ′ ) 2<br />
with initial conditions y(0) = 2, y ′ (0) = 1.<br />
Solution: Introduce v = y ′ . Then v ′ = y ′′ , and<br />
v ′ = −2t v 2 ⇒ v′<br />
= −2t<br />
v2 ⇒<br />
1<br />
−<br />
v = −t2 + c.<br />
So, 1<br />
y ′ = t2 − c, that is, y ′ = 1<br />
t2 . The initial condition implies<br />
− c<br />
1 = y ′ (0) = − 1<br />
c ⇒ c = −1 ⇒ y′ = 1<br />
t2 − 1 .<br />
⊳
58 G. NAGY – ODE july 2, 2013<br />
<br />
Then, y =<br />
dt<br />
t2 + c. We integrate using the method <strong>of</strong> partial fractions,<br />
− 1<br />
1<br />
t 2 − 1 =<br />
1<br />
(t − 1)(t + 1) =<br />
a<br />
(t − 1) +<br />
b<br />
(t + 1) .<br />
Hence, 1 = a(t + 1) + b(t − 1). Evaluating at t = 1 and t = −1 we get a = 1<br />
, b = −1 . So<br />
2 2<br />
1<br />
t2 1<br />
<br />
1<br />
=<br />
− 1 2 (t − 1) −<br />
1<br />
<br />
.<br />
(t + 1)<br />
Therefore, the integral is simple to do,<br />
y = 1<br />
<br />
ln |t − 1| − ln |t + 1| + c.<br />
2<br />
1<br />
2 = y(0) = (0 − 0) + c.<br />
2<br />
We conclude y = 1<br />
<br />
ln |t − 1| − ln |t + 1| + 2.<br />
2<br />
⊳<br />
We now consider the case (b), that is, a <strong>differential</strong> equation y ′′ = f(y, y ′ ), where the<br />
independent variable t is missing. The method to find the solution in this case is more<br />
involved.<br />
Theorem 2.1.11. Consider a second order <strong>differential</strong> equation y ′′ = f(y, y ′ ), and introduce<br />
the function v(t) = y ′ (t). If the function y is invertible, then the new function ˆv(y) = v(t(y))<br />
satisfies the first order <strong>differential</strong> equation<br />
dˆv 1<br />
= f(y, ˆv(y)).<br />
dy ˆv<br />
Pro<strong>of</strong>: Notice that v ′ (t) = f(y, v(t)). Now, by chain rule<br />
dˆv<br />
<br />
<br />
=<br />
dy y<br />
dv<br />
<br />
dt<br />
<br />
<br />
=<br />
dt t(y) dy t(y)<br />
v′<br />
y ′<br />
<br />
<br />
=<br />
t(y)<br />
v′<br />
<br />
<br />
v<br />
Therefore, dˆv<br />
dy<br />
t(y)<br />
= f y, v)<br />
v<br />
<br />
<br />
t(y)<br />
1<br />
= f(y, ˆv(y)). <br />
ˆv<br />
Example 2.1.9: Find a solution y to the second order equation y ′′ = 2y y ′ .<br />
Solution: The variable t does not appear in the equation. Hence, v(t) = y ′ (t). The<br />
equation is v ′ (t) = 2y(t) v(t). Now introduce ˆv(y) = v(t(y)). Then<br />
dˆv<br />
dy =<br />
<br />
dv dt<br />
<br />
t(y)<br />
=<br />
dt dy<br />
v′<br />
y ′<br />
<br />
<br />
=<br />
t(y)<br />
v′<br />
<br />
<br />
.<br />
v t(y)<br />
Using the <strong>differential</strong> equation,<br />
dˆv 2yv<br />
<br />
<br />
=<br />
dy v<br />
t(y)<br />
⇒ dˆv<br />
dy = 2y ⇒ ˆv(y) = y2 + c.<br />
Since v(t) = ˆv(y(t)), we get v(t) = y 2 (t) + c. This is a separable equation,<br />
y ′ (t)<br />
y2 = 1.<br />
(t) + c<br />
Since we only need to find a solution <strong>of</strong> the equation, and the integral depends on whether<br />
c > 0, c = 0, c < 0, we choose (for no special reason) only one case, c = 1.<br />
<br />
dy<br />
= dt + c0 ⇒ arctan(y) = t + c0y(t) = tan(t + c0).<br />
1 + y2 .
G. NAGY – ODE July 2, 2013 59<br />
Again, for no reason, we choose c 0 = 0, and we conclude that one possible solution to our<br />
problem is y(t) = tan(t). ⊳
60 G. NAGY – ODE july 2, 2013<br />
2.1.6. Exercises.<br />
2.1.1.- Compute the Wronskian <strong>of</strong> the following<br />
functions:<br />
(a) f(t) = sin(t), g(t) = cos(t).<br />
(b) f(x) = x, g(x) = x e x .<br />
(c) f(θ) = cos 2 (θ), g(θ) = 1 + cos(2θ).<br />
2.1.2.- Find the longest interval where the<br />
solution y <strong>of</strong> the initial value problems<br />
below is defined. (Do not try to solve<br />
the <strong>differential</strong> <strong>equations</strong>.)<br />
(a) t 2 y ′′ + 6y = 2t, y(1) = 2, y ′ (1) = 3.<br />
(b) (t − 6)y ′ + 3ty ′ − y = 1, y(3) =<br />
−1, y ′ (3) = 2.<br />
2.1.3.- (a) Verify that y1(t) = t 2 and<br />
y2(t) = 1/t are solutions to the <strong>differential</strong><br />
equation<br />
t 2 y ′′ − 2y = 0, t > 0.<br />
(b) Show that y(t) = a t 2 + b<br />
is so-<br />
t<br />
lution <strong>of</strong> the same equation for all<br />
constants a, b ∈ R.<br />
2.1.4.- Can the function y(t) = sin(t 2 ) be<br />
solution on an open interval containing<br />
t = 0 <strong>of</strong> a <strong>differential</strong> equation<br />
y ′′ + a(t) y ′ + b(t)y = 0,<br />
with continuous coefficients a and b?<br />
Explain your answer.<br />
2.1.5.- Verify whether the functions y1, y2<br />
below are a fundamental set for the <strong>differential</strong><br />
<strong>equations</strong> given below:<br />
(a) y ′′ +4y = 0, y1(t) = cos(2t), y2(t) =<br />
sin(2t).<br />
(b) y ′′ − 2y ′ + y = 0, y1(t) = e t , y2(t) =<br />
t e t .<br />
(c) x 2 y ′′ − 2x(x + 2) y ′ + (x + 2) y =<br />
0, y1(x) = x, y2(t) = x e x .<br />
2.1.6.- If the Wronskian <strong>of</strong> any two solutions<br />
<strong>of</strong> the <strong>differential</strong> equation<br />
y ′′ + p(t) y ′ + q(t) y = 0<br />
is constant, what does this imply about<br />
the coefficients p and q?<br />
2.1.7.- Find the solution y to the second order,<br />
nonlinear equation<br />
t 2 y ′′ + 6t y ′ = 1, t > 0.
G. NAGY – ODE July 2, 2013 61<br />
2.2. Constant coefficients<br />
2.2.1. Solution formulas for constant coefficients. In the previous Section we described<br />
general properties <strong>of</strong> solutions to second order, linear, homogeneous, <strong>equations</strong> with variable<br />
coefficients. Among them we mentioned the superposition property, how to obtain fundamental<br />
solutions by solving appropriate initial value problems, and the evolution equation<br />
for the Wronskian <strong>of</strong> two solutions. In this Section we focus on the particular case <strong>of</strong> constant<br />
coefficients <strong>equations</strong>. We find every solution to <strong>differential</strong> <strong>equations</strong> in this reduced<br />
class. We start with a useful definition before we present the main result <strong>of</strong> this Section.<br />
Definition 2.2.1. Consider the homogeneous, constant coefficients <strong>differential</strong> equation<br />
y ′′ + a 1y ′ + a 0 = 0.<br />
The characteristic polynomial <strong>of</strong> the equation is<br />
p(r) = r 2 + a1r + a0.<br />
The characteristic equation is the equation p(r) = 0.<br />
The following result says that the roots <strong>of</strong> the characteristic polynomial are crucial to<br />
express the solutions <strong>of</strong> the homogeneous, constant coefficients, <strong>differential</strong> equation above.<br />
The characteristic polynomial is a real coefficients, second degree polynomial and the general<br />
expression for its roots is<br />
r± = 1<br />
<br />
<br />
.<br />
2<br />
−a1 ± a 2 1 − 4a0<br />
If the discriminant (a 2 1 − 4a0) is positive, zero, or negative, then the roots <strong>of</strong> p are different<br />
real numbers, only one real number, or a complex-conjugate pair <strong>of</strong> complex numbers. For<br />
each case the solution <strong>of</strong> the <strong>differential</strong> equation can be expressed in different forms.<br />
Theorem 2.2.2 (Constant coefficients). Given real constants a1, a0 consider the homogeneous,<br />
linear <strong>differential</strong> equation on the unknown y : R → R given by<br />
y ′′ + a1y ′ + y0 = 0. (2.2.1)<br />
Let r± be the roots <strong>of</strong> the characteristic polynomial and let c0, c1 be arbitrary constants.<br />
(a) If r+ = r−, real or complex, then the general solution <strong>of</strong> Eq. (2.2.1) is given by<br />
y(t) = c0 e r-t + c1 e r-t .<br />
(b) If r+ = r− = ˆr ∈ R, then the general solution <strong>of</strong> Eq. (2.2.1) is given by<br />
y(t) = c 0 e ˆrt + c 1 te ˆrt .<br />
Furthermore, given real constants t 0, y 0 and y 1, there is a unique solution to the initial value<br />
problem given by Eq. (2.2.1) and the initial conditions y(t 0) = y 0 and y ′ (t 0) = y 1.<br />
In the case that the characteristic polynomial has different roots, these roots can be either<br />
real-valued <strong>of</strong> complex-valued. We present examples <strong>of</strong> solutions with real-valued roots<br />
in this Section. In the case that the roots are complex-valued, one root is the complex<br />
conjugate <strong>of</strong> the other, say r± = α ± βi. This is so because the characteristic polynomial<br />
has real coefficients. In this case fundamental solutions <strong>of</strong> the <strong>differential</strong> equation are still<br />
given by the expression <strong>of</strong> the Theorem, that is,<br />
y1(t) = e (α+iβ)t , y2(t) = e (α−iβ)t .<br />
We will see in the next Section that simple linear combinations <strong>of</strong> these solutions define<br />
real-valued fundamental solutions. More <strong>of</strong>ten than not such solutions are useful when one<br />
solves problems in physics or engineering.
62 G. NAGY – ODE july 2, 2013<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 2.2.2: The pro<strong>of</strong> has two main parts: First, we transform the original<br />
equation into an equation simpler to solve for a new unknown; second, we solve this simpler<br />
problem.<br />
In order to transform the problem into a simpler one, we express the solution y as a<br />
product <strong>of</strong> two functions, that is, y(t) = u(t)v(t). Choosing v in an appropriate way the<br />
equation for u will be simpler to solve than the equation for y. Hence,<br />
y = uv ⇒ y ′ = u ′ v + v ′ u ⇒ y ′′ = u ′′ v + 2u ′ v ′ + v ′′ u.<br />
Therefore, Eq. (2.2.1) implies that<br />
that is,<br />
(u ′′ v + 2u ′ v ′ + v ′′ u) + a 1 (u ′ v + v ′ u) + a 0 uv = 0,<br />
<br />
u ′′ <br />
+ a1 + 2 v′<br />
v<br />
We now choose the function v such that<br />
<br />
u ′ <br />
+ a0 u v + (v ′′ + a1 v ′ ) u = 0. (2.2.2)<br />
a1 + 2 v′<br />
= 0<br />
v<br />
⇔<br />
v′<br />
= −a1 .<br />
v 2<br />
(2.2.3)<br />
We choose a simple solution <strong>of</strong> this equation, given by<br />
v(t) = e −a1t/2 .<br />
Having this expression for v one can compute v ′ and v ′′ , and it is simple to check that<br />
v ′′ + a 1v ′ = − a2 1<br />
v. (2.2.4)<br />
4<br />
Introducing the first equation in (2.2.3) and Eq. (2.2.4) into Eq. (2.2.2), and recalling that<br />
v is non-zero, we obtain the simplified equation for the function u, given by<br />
u ′′ − k u = 0, k = a2 1<br />
4 − a 0. (2.2.5)<br />
Eq. (2.2.5) for u is simpler than the original equation (2.2.1) for y since in the former there<br />
is no term with the first derivative <strong>of</strong> the unknown function.<br />
In order to solve Eq. (2.2.5) we repeat the idea followed to obtain this equation, that<br />
is, express function u as a product <strong>of</strong> two functions, and solve a simple problem <strong>of</strong> one <strong>of</strong><br />
the functions. We first consider the harder case, which is when k = 0. In this case, let us<br />
express u(t) = e √ kt w(t). Hence,<br />
u ′ = √ √ √<br />
kt kt ′<br />
ke w + e w ⇒<br />
√ √ √ √<br />
′′ kt kt ′ kt ′′<br />
u = ke w + 2 ke w + e w .<br />
Therefore, Eq. (2.2.5) for function u implies the following equation for function w<br />
0 = u ′′ − ku = e<br />
√ kt (2 √ k w ′ + w ′′ ) ⇒ w ′′ + 2 √ kw ′ = 0.<br />
Only derivatives <strong>of</strong> w appear in the latter equation, so denoting x(t) = w ′ (t) we have to<br />
solve a simple equation<br />
Integrating we obtain w as follows,<br />
renaming c1 = −x0/(2 √ k), we obtain<br />
x ′ = −2 √ k x ⇒ x(t) = x0e −2√ kt , x0 ∈ R.<br />
w ′ = x0e −2√ kt ⇒ w(t) = − x0<br />
2 √ k e−2√ kt + c0.<br />
w(t) = c1e −2√ √<br />
kt kt<br />
+ c0 ⇒ u(t) = c0e + c1e −√kt .
G. NAGY – ODE July 2, 2013 63<br />
We then obtain the expression for the solution y = uv, given by<br />
y(t) = c0e (− a 1 2 + √ k)t + c1e (− a 1 2 − √ k)t .<br />
Since k = (a2 1/4 − a0), the numbers<br />
r± = − a1<br />
2 ± √ k ⇔ r± = 1<br />
<br />
−a1 ±<br />
2<br />
a2 <br />
1 − 4a0 are the roots <strong>of</strong> the characteristic polynomial<br />
r 2 + a1 r + a0 = 0,<br />
we can express all solutions <strong>of</strong> the Eq. (2.2.1) as follows<br />
y(t) = c 0e r-t + c 1e r-t , k = 0.<br />
Finally, consider the case k = 0. Then, Eq. (2.2.5) is simply given by<br />
u ′′ = 0 ⇒ u(t) = (c 0 + c 1t) c 0, c 1 ∈ R.<br />
Then, the solution y to Eq. (2.2.1) in this case is given by<br />
y(t) = (c0 + c1t) e −a1t/2 .<br />
Since k = 0, the characteristic equation r 2 +a1 r+a0 = 0 has only one root r+ = r− = −a1/2,<br />
so the solution y above can be expressed as<br />
y(t) = (c0 + c1t) e r-t , k = 0.<br />
The furthermore part follows from solving a 2 × 2 linear system for the unknowns c0 and c1.<br />
The initial conditions for the case r- = r- are the following,<br />
y 0 = c 0 e r-t0 + c1 e r-t0 , y1 = r+c 0 e r-t0 + r−c 1 e r-t0 .<br />
It is not difficult to verify that this system is always solvable and the solutions are<br />
c0 = − (r−y0 − y1) (r+ − r−) er-t0 , c1 = (r+y0 − y1) .<br />
(r+ − r−) er-t0 The initial conditions for the case r+ = r− are the following,<br />
y0 = (c0 + c1t0) e r-t0 , y1 = c1 e r-t0 + r+(c0 + c1t0) e r-t0 .<br />
It is also not difficult to verify that this system is always solvable and the solutions are<br />
c0 = y0 + t0(r+y0 − y1)<br />
er-t0 , c1 = − (r+y0 − y0) er-t0 .<br />
This establishes the Theorem. <br />
2.2.2. Different real-valued roots. We now present few examples in the case that the<br />
characteristic polynomial has two different real-valued roots.<br />
Example 2.2.1: Find solutions to the equation<br />
y ′′ + 5y ′ + 6y = 0. (2.2.6)<br />
Solution: We see in Theorem 2.2.2 that at least one solution <strong>of</strong> the <strong>differential</strong> equation<br />
above has the form y(t) = e rt . Introduce this function and its derivatives, y ′ (t) = re rt and<br />
y ′′ (t) = r 2 e rt in the equation. The result is the characteristic equation for r, that is,<br />
(r 2 + 5r + 6)e rt = 0 ⇔ r 2 + 5r + 6 = 0. (2.2.7)<br />
The solutions <strong>of</strong> the characteristic equation are<br />
r± = 1 √ 1<br />
−5 ± 25 − 24 = (−5 ± 1) ⇒<br />
2<br />
2<br />
r+ = −2,<br />
r− = −3.
64 G. NAGY – ODE july 2, 2013<br />
Therefore, a set <strong>of</strong> fundamental solutions to the <strong>differential</strong> equation in (2.2.6) is given by<br />
y 1(t) = e −2t , y 2(t) = e −3t .<br />
The superposition property in Proposition 2.1.2 implies that we have found infinitely many<br />
solutions, called the general solution, given by<br />
y(t) = c 1e −2t + c 2e −3t , c 1, c 2 ∈ R. (2.2.8)<br />
Example 2.2.2: Find the solution y <strong>of</strong> the initial value problem<br />
y ′′ + 5y ′ + 6 = 0, y(0) = 1, y ′ (0) = −1.<br />
Solution: We know that a solution <strong>of</strong> the <strong>differential</strong> equation above is<br />
y(t) = c 1e −2t + c 2e −3t .<br />
We now find the constants c1 and c2 that satisfy the initial conditions above:<br />
1 = y(0) = c 1 + c 2<br />
−1 = y ′ (0) = −2c1 − 3c2<br />
Therefore, the unique solution to the initial value problem is<br />
<br />
⇒<br />
y(t) = 2e −2t − e −3t .<br />
c1 = 2,<br />
c 2 = −1.<br />
Example 2.2.3: Find the general solution y <strong>of</strong> the <strong>differential</strong> equation<br />
2y ′′ − 3y ′ + y = 0.<br />
Solution: We look for every solutions <strong>of</strong> the form y(t) = ert , where r is solution <strong>of</strong> the<br />
characteristic equation<br />
2r 2 − 3r + 1 = 0 ⇒ r = 1<br />
⎧<br />
√ ⎨ r1 = 1,<br />
3 ± 9 − 8 ⇒<br />
4<br />
⎩ r2 = 1<br />
2 .<br />
Therefore, the general solution <strong>of</strong> the equation above is<br />
y(t) = c1e t + c2e t/2 .<br />
⊳<br />
⊳<br />
⊳
2.2.3. Exercises.<br />
2.2.1.- . 2.2.2.- .<br />
G. NAGY – ODE July 2, 2013 65
66 G. NAGY – ODE july 2, 2013<br />
2.3. Complex roots<br />
We have seen in the previous Section that Theorem 2.2.2 provides fundamental solutions<br />
to the constant coefficient homogeneous equation<br />
y ′′ + a1 y ′ + a0 y = 0, a1, a0 ∈ R. (2.3.1)<br />
In the case (a1 − 4a0) < 0 the roots <strong>of</strong> the characteristic polynomial are different and<br />
complex-valued, say r± = α±iβ, with α = −a1/2 and β = (a1 − a 2 1/4). The fundamental<br />
solutions shown in Theorem 2.2.2 for this case are the complex-valued functions<br />
˜y 1 = e (α+iβ)t , ˜y 2 = e (α−iβ)t ,<br />
hence, the general solutions <strong>of</strong> the <strong>differential</strong> equation is<br />
y1(t) = ˜c1e (α+iβ)t + ˜c2 e (α−iβ)t , ˜c1, ˜c2 ∈ C.<br />
This expression for the general solution has the unfortunate property that it is not straightforward<br />
to separate real-valued from complex-valued solutions. Such separation could be<br />
important in physical applications. If a physical system is described by a <strong>differential</strong> equation<br />
with real coefficients, then more <strong>of</strong>ten than not one is interested in finding real-valued<br />
solutions to that equation. In the first part <strong>of</strong> this Section we provide a new set <strong>of</strong> fundamental<br />
solutions that are real-valued. These solutions makes it trivial to separate real-valued<br />
from complex-valued solutions. In the second part <strong>of</strong> this Section we describe the electric<br />
current in a circuit formed by a resistance, an inductance, and a capacitor, called the RLC<br />
circuit. This system is described by a second order, linear, constant coefficients, homogeneous<br />
<strong>differential</strong> equation with a characteristic polynomial having complex-valued roots.<br />
The RLC circuit is a simple example <strong>of</strong> a physical situation involving oscillations and dissipation.<br />
Such systems are usually described by <strong>differential</strong> <strong>equations</strong> with complex-valued<br />
roots <strong>of</strong> the characteristic polynomial.<br />
2.3.1. Real-valued fundamental solutions. We mentioned that Theorem 2.2.2 provides<br />
complex-valued fundamental solutions to the equation in (2.3.1) in the case that the associated<br />
characteristic polynomial has complex roots. The following result provides real-valued<br />
fundamental solutions in this same case.<br />
Corollary 2.3.1 (Real-valued fundamental solutions). Assume that a1, a0 ∈ R satisfy<br />
that a 2 1 − 4a 0 < 0, so that Theorem 2.2.2 implies that the <strong>differential</strong> equation<br />
y ′′ + a 1 y ′ + a 0 y = 0 (2.3.2)<br />
has the characteristic polynomial p(r) = r2 +a1r +a0 with complex-valued roots r± = α±iβ,<br />
α = − a1 1<br />
, β = 4a0 − a<br />
2 2<br />
2 1.<br />
and the <strong>differential</strong> equation in (2.3.2) has complex-valued fundamental solutions<br />
˜y1(t) = e (α+iβt) , ˜y2(t) = e (α−iβt) .<br />
Then, real-valued fundamental solutions to the <strong>differential</strong> equation in (2.3.2) are given by<br />
y1(t) = e αt cos(βt), y2(t) = e αt sin(βt).<br />
We know from Theorem 2.2.2 that ˜y 1, ˜y 2 are fundamental set <strong>of</strong> solutions to Eq. (2.3.2).<br />
These fundamental solutions are complex-valued. The corollary above says that there exists<br />
real-valued fundamental solutions given by y1 and y2.<br />
Pro<strong>of</strong> <strong>of</strong> Corollary 2.3.1: Since the functions<br />
˜y 1(t) = e (α+iβ)t , ˜y 2(t) = e (α−iβ)t ,
G. NAGY – ODE July 2, 2013 67<br />
are solutions to the <strong>differential</strong> equation in the Corollary, then so are the functions<br />
y1(t) = 1<br />
˜y1(t) + ˜y2(t)<br />
2<br />
, y2(t) = 1 <br />
˜y1(t) − ˜y2(t)<br />
2i<br />
.<br />
It is simple to check that<br />
y1(t) = 1<br />
2 eαte iβt + e −iβt<br />
y2(t) = 1<br />
2i eαte iβt − e −iβt ,<br />
where e (α±iβ)t = eαte ±iβt . Recalling the Euler equation and its complex-conjugate,<br />
e iβt = cos(βt) + i sin(βt), e −iβt = cos(βt) − i sin(βt),<br />
we obtain that<br />
y1(t) = e αt cos(βt), y2(t) = e αt sin(βt).<br />
This establishes the Corollary. <br />
Example 2.3.1: Find the real-valued general solution <strong>of</strong> the equation<br />
y ′′ − 2y ′ + 6y = 0.<br />
Solution: We first find the roots <strong>of</strong> the characteristic polynomial,<br />
r 2 − 2r + 6 = 0 ⇒ r± = 1<br />
√ <br />
2 ± 4 − 24 ⇒ r± = 1 ± i<br />
2<br />
√ 5.<br />
We know from Theorem 2.2.2 that the complex-valued general solution <strong>of</strong> the <strong>differential</strong><br />
equation above can be written as follows,<br />
y(t) = ˜c1e (1+i√ 5) t + ˜c2e (1−i√ 5) t , ˜c1, ˜c2 ∈ C.<br />
The main information given in the Corollary above is that this general solution can also be<br />
written the equivalent way,<br />
y(t) = c1 cos( √ 5 t) + c2 sin( √ 5 t) e t , c1, c2 ∈ R. (2.3.3)<br />
In order to translate from the former expression to the latter one, we only need to remember<br />
two main relations: First, that the following equation holds<br />
e (1±i√ 5) t = e t e ±i √ 5 t .<br />
Second, that adding and subtracting Euler’s formula and its complex-conjugate formula<br />
gives the relations<br />
e i√ 5 t = cos( √ 5 t) + i sin( √ 5 t),<br />
e −i√ 5 t = cos( √ 5 t) − i sin( √ 5 t),<br />
i<br />
e √ 5 t −i<br />
+ e √ 5 t , sin( √ 5 t) = 1 i<br />
e √ 5 t −i<br />
− e √ 5 t ,<br />
cos( √ 5 t) = 1<br />
2<br />
2i<br />
Therefore, it is simple to show that the following relations hold,<br />
e t cos( √ 5 t) = 1<br />
(1+i<br />
e<br />
2<br />
√ 5)t (1−i<br />
+ e √ 5)t , e t sin( √ 5 t) = 1<br />
2i<br />
and so, the functions<br />
y1(t) = e t cos( √ 5 t), y2(t) = e t sin( √ 5 t),<br />
e (1+i √ 5)t − e (1−i √ 5)t .<br />
are a fundamental set <strong>of</strong> solutions to the <strong>differential</strong> equation above. The real-valued general<br />
solution is given by<br />
y(t) = c1 cos( √ 5 t) + c2 sin( √ 5 t) e t , c1, c2 ∈ R.<br />
⊳
68 G. NAGY – ODE july 2, 2013<br />
Example 2.3.2: Find real-valued fundamental solutions to the equation<br />
y ′′ + 2 y ′ + 6 y = 0.<br />
Solution: The roots <strong>of</strong> the characteristic polynomial p(r) = r2 + 2r + 6 are<br />
r± = 1<br />
√ 1<br />
√ <br />
−2 ± 4 − 24 = −2 ± −20<br />
2<br />
2<br />
⇒ r± = −1 ± i √ These are complex-valued roots, with<br />
5.<br />
Real-valued fundamental solutions are<br />
y<br />
1<br />
α = −1, β = √ 5.<br />
y 1(t) = e −t cos( √ 5 t), y 2(t) = e −t sin( √ 5 t).<br />
− t<br />
e<br />
Figure 9. The graph <strong>of</strong> the solution<br />
y1 found in Example 2.3.1.<br />
Example 2.3.3: Find the real-valued general solution <strong>of</strong> y ′′ + 5 y = 0.<br />
t<br />
Second order <strong>differential</strong> <strong>equations</strong> with<br />
characteristic polynomials having complex<br />
roots, like the one in this example,<br />
describe physical processes related<br />
to damped oscillations. An example<br />
from physics is a pendulums with friction.<br />
⊳<br />
Solution: The characteristic polynomial is p(r) = r 2 + 5. Its roots are r± = ± √ 5 i. This<br />
is the case α = 0, and β = √ 5. Real-valued fundamental solutions are<br />
The real-valued general solution is<br />
y1(t) = cos( √ 5 t), y2(t) = sin( √ 5 t).<br />
y(t) = c1 cos( √ 5 t) + c2 sin( √ 5 t), c1, c2 ∈ R.<br />
Remark: Physical processes that oscillate in time without dissipation could be described<br />
by <strong>differential</strong> <strong>equations</strong> like the one in this example. ⊳<br />
2.3.2. The RLC electric circuit. We describe the electric current flowing through an<br />
electric circuit consisting <strong>of</strong> a resistance, a non-zero capacitor, and non-zero inductance as<br />
shown in Fig. 10. A capacitor and an inductance alone can be used to produce and keep<br />
flowing an electric current I through the circuit, from one plate <strong>of</strong> the capacitor to the other<br />
and vice-versa, endlessly. The presence <strong>of</strong> a resistance transforms the current energy into<br />
heat, damping the current oscillation. This system is described by an integro-<strong>differential</strong><br />
equation found by Kirchh<strong>of</strong>f, now called Kirchh<strong>of</strong>f’s voltage law, relating the resistance R,<br />
capacitance C, inductance L, and the current I in a circuit as follows,<br />
L I ′ (t) + R I(t) + 1<br />
C<br />
t<br />
t0<br />
I(s) ds = 0.
Derivate both sides above,<br />
and divide by L,<br />
G. NAGY – ODE July 2, 2013 69<br />
R C L<br />
I (t) : electric current.<br />
Figure 10. The RLC electric circuit.<br />
L I ′′ (t) + R I ′ (t) + 1<br />
I(t) = 0,<br />
C<br />
I ′′ <br />
R<br />
<br />
(t) + 2 I<br />
2L<br />
′ (t) + 1<br />
I(t) = 0.<br />
LC<br />
If we introduce α = R<br />
1<br />
and ω = √ , then Kirchh<strong>of</strong>f’s law can be expressed as the second<br />
2L LC<br />
order, homogeneous, constant coefficients, <strong>differential</strong> equation<br />
I ′′ + 2α I ′ + ω 2 I = 0.<br />
This equation is what we need to solve our final example.<br />
Example 2.3.4: Find real-valued fundamental solutions to I ′′ + 2α I ′ + ω 2 I = 0, where<br />
α = R/(2L), ω 2 = 1/(LC), in the cases (a), (b) below.<br />
Solution: The roots <strong>of</strong> the characteristic polynomial is p(r) = r2 + 2αr + ω2 , are given by<br />
r± = 1<br />
<br />
−2α ± 4α2 − 4ω2 2<br />
⇒ r± = −α ± α2 − ω2 .<br />
Case (a): R = 0. This implies α = 0, so r± = ±iω. Therefore,<br />
I1(t) = cos(ωt), I2(t) = sin(ωt).<br />
Remark: When the circuit has no resistance, the current oscillates without dissipation.<br />
Case (b): R < 4L/C. This implies<br />
R 2 < 4L<br />
C<br />
R2 1<br />
⇔ <<br />
4L2 LC ⇔ α2 < ω 2 .<br />
Therefore, the characteristic polynomial has complex roots r± = −α ± i √ ω 2 − α 2 , hence<br />
the fundamental solutions are<br />
I1(t) = e −αt cos ω 2 − α 2 t ,<br />
I 2(t) = e −αt sin ω 2 − α 2 t .<br />
Therefore, the resistance R damps the current<br />
oscillations produced by the capacitor<br />
and the inductance. ⊳<br />
I<br />
1<br />
− t<br />
e<br />
Figure 11. The electric current<br />
I as function <strong>of</strong> time.<br />
t
70 G. NAGY – ODE july 2, 2013<br />
2.3.3. Exercises.<br />
2.3.1.- . 2.3.2.- .
G. NAGY – ODE July 2, 2013 71<br />
2.4. Repeated roots<br />
In Section 2.2 we have proven Theorem 2.2.2, which states every solution to the equation<br />
y ′′ + a 1 y ′ + a 0 y = 0<br />
for every possible value <strong>of</strong> the coefficients a1, a0 ∈ R. In particular, in the case that a 2 1 = 4a0,<br />
that is, the characteristic polynomial has the repeated root r+ = r− = −a 1/2, the Theorem<br />
states that a fundamental set <strong>of</strong> solutions is given by<br />
y 1(t) = e r-t , y 2(t) = te r-t . (2.4.1)<br />
In the first part <strong>of</strong> this Section we present two new pro<strong>of</strong>s for the repeated roots part <strong>of</strong><br />
Theorem 2.2.2. The first pro<strong>of</strong> contains a nice argument showing that the fundamental<br />
solutions for the case <strong>of</strong> repeated roots can be obtained from a limit procedure on solutions<br />
to <strong>equations</strong> with complex roots. More precisely, we show that solutions <strong>of</strong> <strong>equations</strong><br />
with complex roots having small imaginary part are close to solutions <strong>of</strong> <strong>equations</strong> with<br />
appropriate repeated roots. We then present a third pro<strong>of</strong> <strong>of</strong> the repeated root part in<br />
Theorem 2.2.2. Although this is essentially the original pro<strong>of</strong> given in Section 2.2, here we<br />
emphasize its main idea. The reason for highlighting this idea becomes clear in the second<br />
part <strong>of</strong> this Section. There we generalize this idea, called the reduction <strong>of</strong> order method, to<br />
<strong>differential</strong> <strong>equations</strong> with variable coefficients.<br />
2.4.1. Fundamental solutions for repeated roots. We start with a simple example<br />
showing how we use the repeated root part <strong>of</strong> Theorem 2.2.2.<br />
Example 2.4.1: Find the solution to the initial value problem<br />
9y ′′ + 6y ′ + y = 0, y(0) = 1, y ′ (0) = 5<br />
3 .<br />
Solution: The characteristic polynomial is p(r) = 9r2 + 6r + 1, with root<br />
r± = 1 √ <br />
−6 ± 36 − 36 ⇒ r+ = r− = −<br />
18<br />
1<br />
3 .<br />
Therefore, Theorem 2.2.2 says that the general solution has the form<br />
y(t) = c1e −t/3 + c2te −t/3 .<br />
Since its derivative is given by<br />
y ′ (t) = − c1<br />
3 e−t/3 <br />
+ c2 1 − t<br />
<br />
e<br />
3<br />
−t/3 ,<br />
then, the initial conditions imply that<br />
1 = y(0) = c1, 5<br />
3 = y′ (0) = − c1 3 + c ⎫<br />
⎬<br />
⎭ 2<br />
⇒ c1 = 1, c2 = 2.<br />
So, the solution to the initial value problem above is: y(t) = (1 + 2t) e −t/3 . ⊳<br />
Second pro<strong>of</strong> <strong>of</strong> repeated root part <strong>of</strong> Theorem 2.2.2: We now show that the solution<br />
y 2 in Eq. (2.4.1) is close to solutions <strong>of</strong> an appropriate <strong>differential</strong> equation having complex<br />
roots in the case that the imaginary part is small enough. Indeed, consider the <strong>differential</strong><br />
equation<br />
˜y ′′ + ã 1 ˜y ′ + ã 0 ˜y = 0,<br />
where ã1 − 4ã0 < 0. We know that real-valued fundamental solutions are given by<br />
˜y1 = e α t cos(βt), ˜y2 = e α t sin(βt),
72 G. NAGY – ODE july 2, 2013<br />
where α = −ã 1/2 and β = ã 0 − ã 2 1/4, as usual. We now study what happen to these<br />
solutions ˜y1 and ˜y2 in the following limit: The variable t is held constant, α is held constant,<br />
and β → 0. The last two conditions are conditions on the equation coefficients, ã1, ã0.<br />
Since cos(βt) → 1 as β → 0 with t fixed, then keeping α fixed too, we obtain<br />
Since sin(βt)<br />
βt<br />
˜y1β(t) = e α t cos(βt) −→ e α t = y1(t).<br />
→ 1 as β → 0 with t constant, that is, sin(βt) → βt, we conclude that<br />
˜y2(t)<br />
β<br />
= sin(βt)<br />
β<br />
e α t = sin(βt)<br />
βt<br />
t e α t −→ t e α t = y2(t).<br />
The calculation above says that the function ˜y2(t) α t<br />
/β is close to the function y2(t) = t e<br />
in the limit mentioned above. This calculation only suggests that y2(t) = t y1(t) could be<br />
solutions <strong>of</strong> the equation having repeated roots in its characteristic polynomial. It gives a<br />
hint on what could be the solution. One has to introducing this function y2 in the <strong>differential</strong><br />
equation to prove that this function y2 is indeed a solution. Since y2 is not proportional to<br />
y1, one then concludes the functions y1, y2 are a fundamental set for the <strong>differential</strong> equation<br />
in with repeated roots. This establishes part (b) in Theorem 2.2.2. <br />
We now repeat one more time the last part <strong>of</strong> the pro<strong>of</strong> <strong>of</strong> Theorem 2.2.2 in order to<br />
highlight its main idea. It turns out that this idea can be generalized from <strong>equations</strong> with<br />
constant coefficients to <strong>equations</strong> with variable coefficients, and it is called the reduction <strong>of</strong><br />
order method. This generalization occupies the last part <strong>of</strong> this Section.<br />
Third pro<strong>of</strong> <strong>of</strong> repeated root part <strong>of</strong> Theorem 2.2.2: Since a2 1 = 4a0, there is only<br />
one root to the characteristic polynomial p(r) = r2 + a1r + a0 given by r1 = −a1/2. This<br />
implies that one solution <strong>of</strong> the <strong>differential</strong> equation y ′′ + a1 y ′ + a0 y = 0 is given by<br />
y 1(t) = e r1t .<br />
The main idea in the reduction <strong>of</strong> order method is to look for a second solution to the<br />
<strong>differential</strong> equation as a product <strong>of</strong> a function times the known solution, that is,<br />
y2(t) = v(t) y1(t) ⇒ y2(t) = v(t) e r1t .<br />
Introducing this function y2 into the <strong>differential</strong> equation one obtains a <strong>differential</strong> equation<br />
for the new function v. First compute<br />
y ′ 2 = (v ′ + r1 v) e r1t , y ′′<br />
2 = (v ′′ + 2r1 v ′ + r 2 1 v) e r1t .<br />
As we said, introduce y2 into the <strong>differential</strong> equation and get the equation for v,<br />
0 = (v ′′ + 2r1 v ′ + r 2 1 v) + a1(v ′ + r1 v) + a0 v,<br />
= v ′′ + (2r 1 + a 1) v ′ + (r 2 1 + a 1r + a 0) v.<br />
However, the equation for v above is simpler than the original equation for y2, since y1 = e r1t<br />
is a solution <strong>of</strong> the original <strong>differential</strong> equation, which in this case means both that<br />
r 2 1 + a1r1 + a0 = 0 and 2r1 + a1 = 0.<br />
So, the equation for the function v is simple and given by<br />
v ′′ = 0.<br />
The solutions are given in terms <strong>of</strong> two arbitrary constants c 1 and c 2 as follows<br />
v(t) = c 1 + c 2t ⇒ y 2(t) = c 1e r1t + c2te r1t .
G. NAGY – ODE July 2, 2013 73<br />
Since the constants c 1 and c 2 are arbitrary, choosing c 1 = 0 and c 2 = 1 one obtains the<br />
following set <strong>of</strong> solutions to the original <strong>differential</strong> equation:<br />
y1(t) = e r1t , y2(t) = te r1t .<br />
This establishes part (b) in Theorem 2.2.2. <br />
2.4.2. Reduction <strong>of</strong> order method. This same idea used in the pro<strong>of</strong> above, the reduction<br />
<strong>of</strong> order method, can be generalized to variable coefficient <strong>equations</strong>.<br />
Theorem 2.4.1 (Reduction <strong>of</strong> order). Given continuous functions p, q : (t 1, t 2) → R,<br />
let y1 : (t1, t2) → R be a solution <strong>of</strong> the <strong>differential</strong> equation<br />
If a function v : (t 1, t 2) → R is solution <strong>of</strong><br />
y ′′ + p(t) y ′ + q(t) y = 0. (2.4.2)<br />
y 1(t) v ′′ + 2y ′ (t) + p(t)y 1(t) v ′ = 0, (2.4.3)<br />
then the functions y 1 and y 2 = v y 1 are fundamental solutions to Eq. (2.4.2).<br />
The reason for the name reduction <strong>of</strong> order method is that the equation for the function<br />
v involves v ′′ and v ′ but does not involve v itself. So this equation is a first order equation<br />
for the unknown w = v ′ .<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 2.4.1: If we express y2 = vy1, then the <strong>differential</strong> equation for y<br />
implies an equation for v, as the following calculation shows. First, compute y ′ 2 and y ′′<br />
2 ,<br />
y ′ 2 = v ′ y1 + v y ′ 1, y ′′<br />
2 = v ′′ y1 + 2v ′ y ′ 1 + v y ′′<br />
1 .<br />
Introduce this information introduced the <strong>differential</strong> equation,<br />
0 = (v ′′ y1 + 2v ′ y ′ 1 + v y ′′<br />
1 ) + p (v ′ y1 + v y ′ 1) + qv y1<br />
= y1 v ′′ + (2y ′ 1 + p y1) v ′ + (y ′′<br />
1 + p y ′ 1 + q y1) v.<br />
The function y1 is solution to the <strong>differential</strong> original <strong>differential</strong> equation,<br />
y ′′<br />
1 + p y ′ 1 + q y1 = 0,<br />
then, the equation for v is given by Eq. (2.4.3), that is,<br />
y1 v ′′ + (2y ′ 1 + p y1) v ′ = 0.<br />
We now need to show that y1 and y2 = vy1 are linearly independent.<br />
Wy1y2 =<br />
<br />
<br />
<br />
y1 vy1<br />
y ′ 1 (v ′ y1 + vy ′ <br />
<br />
<br />
1)<br />
We need to find v ′ . Denote w = v ′ , so<br />
= y1(v ′ y1 + vy ′ 1) − vy1y ′ 1 ⇒ Wy1y2 = v′ y 2 1.<br />
y1 w ′ + (2y ′ 1 + p y1) w = 0 ⇒ w′<br />
w = −2y′ 1<br />
Let P be a primitive <strong>of</strong> p, that is, P ′ (t) = p(t), then<br />
ln(w) = −2 ln(y1) − P ⇒ w = e [ln(y−2<br />
y1<br />
− p.<br />
1 )−P ] ⇒ w = y −2<br />
1<br />
e −P .<br />
We obtain v ′ y 2 1 = e −P , hence Wy1y2 = e−P , which is non-zero. We conclude that y1 and<br />
y 2 = vy 1 are linearly independent. This establishes the Theorem.
74 G. NAGY – ODE july 2, 2013<br />
Example 2.4.2: Find a second solution y 2 linearly independent to the solution y 1(t) = t <strong>of</strong><br />
the <strong>differential</strong> equation<br />
t 2 y ′′ + 2ty ′ − 2y = 0.<br />
Solution: We look for a solution <strong>of</strong> the form y 2(t) = t v(t). This implies that<br />
So, the equation for v is given by<br />
y ′ 2 = t v ′ + v, y ′′<br />
2 = t v ′′ + 2v ′ .<br />
0 = t 2 t v ′′ + 2v ′ + 2t t v ′ + v − 2t v<br />
= t 3 v ′′ + (2t 2 + 2t 2 ) v ′ + (2t − 2t) v<br />
= t 3 v ′′ + (4t 2 ) v ′ ⇒ v ′′ + 4<br />
t v′ = 0.<br />
Notice that this last equation is precisely Eq. (2.4.3), since in our case we have<br />
y1 = t, p(t) = 2<br />
t ⇒ t v′′ + 2 + 2<br />
t t v ′ = 0.<br />
The equation for v is a first order equation for w = v ′ , given by<br />
w ′<br />
= −4<br />
w t ⇒ w(t) = c1t −4 , c1 ∈ R.<br />
Therefore, integrating once again we obtain that<br />
v = c2t −3 + c3, c2, c3 ∈ R,<br />
and recalling that y 2 = t v we then conclude that<br />
y2 = c2t −2 + c3t.<br />
Choosing c2 = 1 and c3 = 0 we obtain that y2(t) = t−2 . Therefore, a fundamental solution<br />
set to the original <strong>differential</strong> equation is given by<br />
y1(t) = t, y2(t) = 1<br />
.<br />
t2 ⊳
2.4.3. Exercises.<br />
2.4.1.- . 2.4.2.- .<br />
G. NAGY – ODE July 2, 2013 75
76 G. NAGY – ODE july 2, 2013<br />
2.5. Undetermined coefficients<br />
The method <strong>of</strong> undetermined coefficients is a trial-and-error procedure to find solutions<br />
<strong>of</strong> non-homogeneous <strong>differential</strong> <strong>equations</strong>. Recall that a <strong>differential</strong> equation <strong>of</strong> the form<br />
y ′′ + p(t) y ′ + q(t) y = f(t) (2.5.1)<br />
is called non-homogeneous iff the function f is not identically zero; see Definition 2.1.1.<br />
The method <strong>of</strong> undetermined coefficients is then, given a particular form <strong>of</strong> the function<br />
f, propose a clever guess for the solution y, guess that depends on some free parameters,<br />
or yet undetermined coefficients. Introduce this guess y into the <strong>differential</strong> equation and<br />
determine the free parameters.<br />
2.5.1. Operator notation. Given functions p, q : (t1, t2) → R, we denote by L(y) the<br />
left-hand side in Eq. (2.5.1), that is<br />
L(y) = y ′′ + p(t) y ′ + q(t) y. (2.5.2)<br />
The function L acting on the function y is called an operator. Therefore, Eq. (2.5.1) can be<br />
written as L(y) = f, while the homogeneous equation is simply L(y) = 0.<br />
It is also convenient to emphasize that the superposition property proven in Proposition<br />
2.1.2 is a consequence <strong>of</strong> the fact that the operator L is a linear function <strong>of</strong> y. This is<br />
stated in the following result.<br />
Theorem 2.5.1 (Linearity <strong>of</strong> L). For every continuously differentiable pair <strong>of</strong> functions<br />
y 1, y 2 : (t 1, t 2) → R and every c 1, c 2 ∈ R the operator in (2.5.2) satisfies that<br />
L(c1y1 + c2y2) = c1L(y1) + c2L(y2). (2.5.3)<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 2.5.1: This is a straightforward calculation:<br />
L(c 1y 1 + c 2y 2) = (c 1y 1 + c 2y 2) ′′ + p(t) (c 1y 1 + c 2y 2) ′ + q(t) (c 1y 1 + c 2y 2).<br />
Re-order terms in the following way,<br />
L(c1y1 + c2y2) = c1y ′′<br />
1 + p(t) c1y ′ 1 + q(t) c1y1<br />
<br />
+ c2y ′′<br />
2 + p(t) c2y ′ <br />
2 + q(t) c2y2 .<br />
Introduce the definition <strong>of</strong> L back on the right-hand side. We then conclude that<br />
L(c1y1 + c2y2) = c1L(y1) + c2L(y2).<br />
This establishes the Theorem. <br />
It is now simple to see that the superposition property follows from the linearity <strong>of</strong> the<br />
operator L. Indeed, for every functions y 1, y 2, solutions <strong>of</strong> the homogeneous <strong>equations</strong><br />
L(y 1) = 0 and L(y 2) = 0 holds that their linear combination (c 1y 1 + c 2y 2) is also solution <strong>of</strong><br />
the homogeneous equation, since<br />
L(c 1y 1 + c 2y 2) = c 1L(y 1) + c 2L(y 2) = 0.<br />
Finally, we state two simple results we need later on. Their pro<strong>of</strong> is a straightforward<br />
consequence <strong>of</strong> Eq. (2.5.3), and they are left as exercise for the reader.<br />
Theorem 2.5.2 (General solution). Given functions p, q, f, let L(y) = y ′′ + p y ′ + q y.<br />
If the functions y 1 and y 2 are fundamental solutions <strong>of</strong> the homogeneous <strong>equations</strong><br />
L(y 1) = 0, L(y 2) = 0,<br />
and yp is any solution <strong>of</strong> the non-homogeneous equation<br />
L(yp) = f, (2.5.4)<br />
then any other solution y <strong>of</strong> the non-homogeneous equation above is given by<br />
y(t) = c1y1(t) + c2y2(t) + yp(t), c1, c2 ∈ R. (2.5.5)
G. NAGY – ODE July 2, 2013 77<br />
We just mention that the expression for y in Eq. (2.5.5) is called the general solution <strong>of</strong><br />
the non-homogeneous Eq. (2.5.4). Finally, we present a second result that uses the linearity<br />
property <strong>of</strong> the operator L. This result is useful to simplify solving certain <strong>differential</strong><br />
<strong>equations</strong>.<br />
Theorem 2.5.3 (Source simplification). Given functions p, q, let L(y) = y ′′ + p y ′ + q y.<br />
If the function f can be written as f(t) = f1(t) + · · · + fn(t), with n 1, and if there exists<br />
functions yp1 , · · · , ypn such that<br />
L(ypi ) = fi, i = 1, · · · , n,<br />
then the function yp = yp1 + · · · + ypn satisfies the non-homogeneous equation<br />
L(yp) = f.<br />
2.5.2. The undetermined coefficients method. The method <strong>of</strong> undetermined coefficients<br />
is used to solve the following type <strong>of</strong> problems: Given a constant coefficients linear<br />
operator L(y) = y ′′ +a1y ′ +a0y, with a1, a2 ∈ R, find every solution <strong>of</strong> the non-homogeneous<br />
<strong>differential</strong> equation<br />
L(y) = f.<br />
A summary <strong>of</strong> the undetermined coefficients method is the following:<br />
(1) Find the general solution to the homogeneous equation L(yh) = 0.<br />
(2) If the source function f has the form f = f1+· · ·+fn, with n 1, then look for solutions<br />
yp1 , · · · , ypn , to the <strong>equations</strong> L(ypi ) = fi, with i = 1, · · · , n. Once the functions ypi are<br />
found, then construct yp = yp1 + · · · + ypn .<br />
(3) Given the source functions fi, guess the solutions ypi following the Table 1 below.<br />
(4) If any guessed function ypi satisfies the homogeneous equation L(ypi) = 0, then change<br />
the guess to the function tsypi , with s 1 sufficiently large such that L(tsypi ) = 0.<br />
(5) Impose the equation L(ypi) = fi to find the undetermined constants k1, · · · , kn given in<br />
the table below. Then, compute yp = yp1 + · · · + ypn .<br />
(6) The solution to the original problem is then given by y(t) = yh(t) + yp(t).<br />
In the rest <strong>of</strong> the Section we use this method in several examples.<br />
Example 2.5.1: Find all solutions to the inhomogeneous equation<br />
y ′′ − 3y ′ − 4y = 3e 2t .<br />
Solution: Form the problem we get that L(y) = y ′′ − 3y ′ − 4y and f(t) = 3e 2t .<br />
(1): Find all solutions yh to the homogeneous equation L(yh) = 0. The characteristic<br />
equation is<br />
r 2 − 3r − 4 = 0 ⇒ r 1 = 4, r 2 = −1, ⇒ yh(t) = c 1 e 4t + c 2 e −t .<br />
(2): Trivial in our case. The source function f(t) = 3e 2t cannot be simplified into a sum <strong>of</strong><br />
simpler functions.<br />
(3): Table says: For f(t) = 3e 2t guess yp(t) = k e 2t . The constant k is the undetermined<br />
coefficient we must find.<br />
(4): Trivial here, since L(yp) = 0. Therefore, we do not modify our guess. (Recall: L(yh) = 0<br />
iff yh(t) = c 1 e 4t + c 2 e −t .)
78 G. NAGY – ODE july 2, 2013<br />
(5): Introduce yp into L(yp) = f and find k.<br />
(2 2 − 6 − 4)ke 2t = 3e 2t ⇒ −6k = 3 ⇒ k = − 1<br />
2 .<br />
Notice that the guessed solution must be proportional to the exponential e 2t in order to<br />
cancel out the exponentials in the equation above. We have obtained that<br />
yp(t) = − 1<br />
2 e2t .<br />
(6): The general solution to the inhomogeneous equation is then given by<br />
y(t) = c1e 4t + c2e −t − 1<br />
2 e2t .<br />
fi(t) (K, m, a, b, given.) ypi (t) (Guess) (k not given.)<br />
Ke at ke at<br />
Kt m kmt m + km−1t m−1 + · · · + k0<br />
K cos(bt) k1 cos(bt) + k2 sin(bt)<br />
K sin(bt) k 1 cos(bt) + k 2 sin(bt)<br />
Kt m e at e at (kmt m + · · · + k 0)<br />
Ke at cos(bt) e at k 1 cos(bt) + k 2 sin(bt) <br />
KKeat sin(bt) eatk1 cos(bt) + k2 sin(bt) <br />
Ktm <br />
cos(bt)<br />
kmtm <br />
+ · · · + k0 a1 cos(bt) + a2 sin(bt) <br />
Ktm <br />
sin(bt)<br />
kmtm <br />
+ · · · + k0 a1 cos(bt) + a2 sin(bt) <br />
Table 1. List <strong>of</strong> sources fi and solutions ypi to the equation L(ypi ) = fi.<br />
Example 2.5.2: Find all solutions to the inhomogeneous equation<br />
y ′′ − 3y ′ − 4y = 3e 4t .<br />
Solution: We have the same left hand side as in the example above, so the first step,<br />
finding the solutions to the homogeneous equation, is the same as in the example above:<br />
yh(t) = c1e 4t + c2e −t .<br />
The second step is again trivial, since the source function f(t) = 3e 4y cannot be simplified<br />
into a sum <strong>of</strong> simpler functions. The third step says for the source f(t) = 3 e 4t guess<br />
yp(t) = k e 4t . However, step four tells us that we have to change our guess, since yp = k yh1<br />
is a solution <strong>of</strong> the homogeneous equation. Step four says we have to change our guess to<br />
yp(t) = kt e 4t .<br />
Step five is to introduce the guess into L(yp) = f. We need to compute<br />
y ′ p = (1 + 4t) ke 4t , y ′′<br />
p = (8 + 16t) ke 4t ⇒ 8 − 3 + (16 − 12 − 4)t ke 4t = 3e 4t ,<br />
⊳
therefore, we get that<br />
G. NAGY – ODE July 2, 2013 79<br />
5k = 3 ⇒ k = 3<br />
5 ⇒ yp(t) = 3<br />
5 te4t .<br />
The general solution <strong>of</strong> the non-homogeneous equation is<br />
y(t) = c 1e 4t + c 2e −t + 3<br />
5 te4t .<br />
Example 2.5.3: Find all the solutions to the inhomogeneous equation<br />
y ′′ − 3y ′ − 4y = 2 sin(t).<br />
Solution: We again have the same left hand side as in the example above, so the first step,<br />
finding the solutions to the homogeneous equation, is the same as in the example above:<br />
yh(t) = c1e 4t + c2e −t .<br />
The second step is still not needed, the third step says that for the source f(t) = 2 sin(t) the<br />
guess must be yp(t) = k1 sin(t) + k2 cos(t). Since our guess is not a solution to the homogeneous<br />
equation, we proceed to step five and we introduce our guess into the <strong>differential</strong><br />
equation. We first compute<br />
y ′ p = k 1 cos(t) − k 2 sin(t), y ′′<br />
p = −k 1 sin(t) − k 2 cos(t),<br />
and then we obtain<br />
[−k 1 sin(t) − k 2 cos(t)] − 3[k 1 cos(t) − k 2 sin(t)] − 4[k 1 sin(t) + k 2 cos(t)] = 2 sin(t).<br />
Re-ordering terms in the expression above we get<br />
(−5k 1 + 3k 2) sin(t) + (−3k 1 − 5k 2) cos(t) = 2 sin(t).<br />
The last equation must hold for all t ∈ R. In particular, it must hold for t = π/2 and for<br />
t = 0. At these two points we obtain, respectively,<br />
⎧<br />
⎪⎨<br />
−5k1 + 3k2 = 2,<br />
k1 = −<br />
⇒<br />
−3k1 − 5k2 = 0, ⎪⎩<br />
5<br />
17 ,<br />
k2 = 3<br />
17 .<br />
So the particular solution to the inhomogeneous equation is given by<br />
yp(t) = 1 <br />
−5 sin(t) + 3 cos(t) ,<br />
17<br />
and the general solution is<br />
<br />
−5 sin(t) + 3 cos(t) .<br />
y(t) = c1e 4t + c2e −t + 1<br />
17<br />
Example 2.5.4: We provide few more examples <strong>of</strong> non-homogeneous <strong>equations</strong> and the<br />
appropriate guesses for the particular solutions.<br />
(a) For y ′′ − 3y ′ − 4y = 3e 2t sin(t), guess, yp(t) = k1 sin(t) + k2 cos(t) e 2t .<br />
(b) For y ′′ − 3y ′ − 4y = 2t 2 e 3t , guess, yp(t) = k2t 2 + k1t + k0<br />
(c) For y ′′ − 3y ′ − 4y = 2t 2 e 4t , guess, yp(t) = t k2t 2 + k1t + k0<br />
e 3t .<br />
e 4t .<br />
(d) For y ′′ − 3y ′ − 4y = 3t sin(t), guess, yp(t) = (1 + k1t) k2 sin(t) + k3 cos(t) .<br />
⊳<br />
⊳
80 G. NAGY – ODE july 2, 2013<br />
2.5.3. Exercises.<br />
2.5.1.- . 2.5.2.- .
G. NAGY – ODE July 2, 2013 81<br />
2.6. Variation <strong>of</strong> parameters<br />
We describe a method to find a solution to a non-homogeneous <strong>differential</strong> equation with<br />
variable coefficients with a continuous but otherwise arbitrary source function,<br />
y ′′ + p(t) y ′ + q(t) y = f(t),<br />
in the case that one happens to know a two linearly independent solutions to the associated<br />
homogeneous equation. This way to find solutions to non-homogeneous <strong>differential</strong> <strong>equations</strong><br />
is called the variation <strong>of</strong> parameters method and is summarized in Theorem 2.6.1 below.<br />
The variation <strong>of</strong> parameter method can be applied to more general <strong>equations</strong> than the<br />
undetermined coefficients method studied in the previous Section. However, the former<br />
usually takes more time to implement than the latter simpler method.<br />
2.6.1. The variation <strong>of</strong> parameters method. The whole method can be summarized in<br />
the following statement.<br />
Theorem 2.6.1 (Variation <strong>of</strong> parameters). Let p, q, f : (t1, t2) → R be continuous<br />
functions, let y1, y2 : (t1, t2) → R be fundamental solutions to the homogeneous equation<br />
y ′′ + p(t) y ′ + q(t) y = 0,<br />
and let Wy1y2 be the Wronskian <strong>of</strong> y1 and y2. If the functions u1 and u2 are defined by<br />
<br />
u1(t) = − y2(t)f(t) Wy1y2 (t) dt, u2(t)<br />
<br />
y1(t)f(t)<br />
=<br />
dt, (2.6.1)<br />
Wy1y2 (t)<br />
then a particular solution to the non-homogeneous equation y ′′ + p(t) y ′ + q(t) y = f(t) is<br />
yp = u 1y 1 + u 2y 2.<br />
The result above is more general than the trial-and-error method studied earlier since<br />
Theorem 2.6.1 applies to any continuous source function f, not only to those functions that<br />
appear in Table 1. As in the undetermined coefficients method, the variation <strong>of</strong> parameters<br />
method also requires to know two solutions to the homogeneous equation in order to solve<br />
the non-homogeneous one. One can say that the pro<strong>of</strong> <strong>of</strong> Theorem 2.6.1 is based in two new<br />
ideas. While the first idea is a good one, the second idea is much better.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 2.6.1: The problem is to find a solution <strong>of</strong> an inhomogeneous equation<br />
y ′′ + p(t) y ′ + q(t) y = f(t) (2.6.2)<br />
knowing that there exist fundamental solutions y1 and y2 to the homogeneous equation<br />
y ′′ + p(t) y ′ + q(t) y = 0.<br />
One can say that the first idea to solve this inhomogeneous equation comes from the reduction<br />
<strong>of</strong> order method. In that case, one finds a second solution y 2 to an homogeneous<br />
equation if one already knows another solution y 1, by proposing y 2 = uy 1. One then hopes<br />
that the equation for u will be simpler than the original equation for y2, since y1 is solution<br />
<strong>of</strong> the homogeneous equation. We now adapt this idea to our situation as follows: we<br />
propose a form <strong>of</strong> the solution yp <strong>of</strong> the form<br />
yp = u1y1 + u2y2,<br />
and we hope that the equation for u 1 and u 2 will be simpler than the original equation,<br />
since y 1 and y 2 are solutions to the homogeneous equation. We introduce this expression<br />
for yp into the original <strong>differential</strong> equation. We must first compute the derivatives<br />
y ′ p = u ′ 1y 1 + u 1y ′ 1 + u ′ 2y 2 + u 2y ′ 2, y ′′<br />
p = u ′′<br />
1 y 1 + 2u ′ 1y ′ 1 + u 1y ′′<br />
1 + u ′′<br />
2 y 2 + 2u ′ 2y ′ 2 + u 2y ′′<br />
2 .
82 G. NAGY – ODE july 2, 2013<br />
After some reordering <strong>of</strong> terms, one obtains that Eq. (2.6.2) has the form<br />
f = u ′′<br />
1 y 1 + u ′′<br />
2 y 2 + 2(u ′ 1y ′ 1 + u 2y ′ 2) + p (u ′ 1y 1 + u ′ 2y 2)<br />
+ u1 (y ′′<br />
1 + p y ′ 1 + q y1) + u2 (y ′′<br />
2 + p y ′ 2 + q y2)<br />
The functions y1 and y2 are solutions to the homogeneous <strong>equations</strong><br />
y ′′<br />
1 + p y ′ 1 + q y1 = 0, y ′′<br />
2 + p y ′ 2 + q y2 = 0,<br />
so we obtain that u1 and u2 must be solution <strong>of</strong> a simpler equation that the one above,<br />
given by<br />
f = u ′′<br />
1 y1 + u ′′<br />
2 y2 + 2(u ′ 1y ′ 1 + u ′ 2y ′ 2) + p (u ′ 1y1 + u ′ 2y2). (2.6.3)<br />
The second idea is the following: Look for functions u1 and u2 that satisfy the extra equation<br />
u ′ 1y1 + u ′ 2y2 = 0. (2.6.4)<br />
Any<br />
<br />
solution <strong>of</strong> this extra equation must satisfy that its derivative is also zero, that is,<br />
′ u 1y1 + u ′ ′<br />
2y2 = 0, and this last equation is explicitly given by<br />
u ′′<br />
1 y1 + u ′′<br />
2 y2 + (u ′ 1y ′ 1 + u ′ 2y ′ 2) = 0. (2.6.5)<br />
Using Eqs. (2.6.4) and Eq. (2.6.5) into the original Eq. (2.6.3) we obtain that<br />
u ′ 1y ′ 1 + u ′ 2y ′ 2 = f. (2.6.6)<br />
Therefore, instead <strong>of</strong> solving the original equation in (2.6.3) we find solutions u1 and u2 for<br />
the system given by Eqs. (2.6.4) and (2.6.6). This system is simple to solve since it is an<br />
algebraic system <strong>of</strong> two <strong>equations</strong> for the two unknowns u ′ 1 and u ′ 2. From Eq. (2.6.4) we<br />
compute u ′ 2 and we introduce it into Eq. (2.6.6), as follows,<br />
u ′ 2 = − y1<br />
u<br />
y2 ′ 1 ⇒ u ′ 1y ′ 1 − y1y ′ 2<br />
u<br />
y2 ′ 1 = f ⇒ u ′ ′ y 1y2 − y1y<br />
1<br />
′ 2<br />
y2 Since Wy1y2 = y1y ′ 2 − y ′ 1y2, we obtain that<br />
u ′ 1 = − y2f<br />
Wy1y2<br />
⇒ u ′ 2 = y1f<br />
Wy1y2<br />
.<br />
<br />
= f.<br />
Integrating in the variable t we obtain the expressions in Eq. (2.6.1), if we choose the<br />
integration constants to be zero. Then yp = u 1y 1 + u 2y 2 is a solution to the inhomogeneous<br />
<strong>differential</strong> equation in Theorem (2.6.1). Notice that the constant <strong>of</strong> integration can always<br />
be chosen to be zero, because <strong>of</strong> the following argument: Denote by u1 and u2 the functions<br />
in Eq. (2.6.1), and given any real numbers c1 and c2 define<br />
ũ1 = u1 + c1, ũ2 = u2 + c2.<br />
Then the corresponding solution ˜yp is given by<br />
˜yp = ũ 1y 1 + ũ 2y 2 = u 1y 1 + u 2y 2 + c 1y 1 + c 2y 2 ⇒ ˜yp = yp + c 1y 1 + c 2y 2.<br />
That is, the two solutions ˜yp and yp differ by a solution to the homogeneous <strong>differential</strong><br />
equation, so both are solutions to the inhomogeneous equation. One is then free to chose<br />
the constants c 1 and c 2 in any way. We choose them to be zero. This establishes the<br />
Theorem. <br />
Example 2.6.1: Find the general solution <strong>of</strong> the non-homogeneous equation<br />
y ′′ − 5y ′ + 6y = 2e t .
G. NAGY – ODE July 2, 2013 83<br />
Since the construction <strong>of</strong> yp by Theorem 2.6.1 requires to know fundamental solutions to<br />
the homogeneous problem, let us find these solutions first. Since the equation has constant<br />
coefficients, we compute the characteristic equation,<br />
r 2 − 5r + 6 = 0 ⇒ r = 1<br />
√ <br />
5 ± 25 − 24 ⇒<br />
2<br />
So, the functions y1 and y2 in Theorem 2.6.1 are in our case given by<br />
y 1(t) = e 3t , y 2(t) = e 2t .<br />
The Wronskian <strong>of</strong> these two functions is given by<br />
r1 = 3,<br />
r 2 = 2.<br />
Wy1y2 (t) = (e3t )(2e 2t ) − (3e 3t )(e 2t ) ⇒ Wy1y2 (t) = −e5t .<br />
We are now ready to compute the functions u1 and u2. Notice that Eq. (2.6.1) the following<br />
<strong>differential</strong> <strong>equations</strong><br />
u ′ 1 = − y2f<br />
, u ′ 2 = y1f<br />
.<br />
So, the equation for u1 is the following,<br />
Wy1y2<br />
Wy1y2<br />
u ′ 1 = −e 2t (2e t )(−e −5t ) ⇒ u ′ 1 = 2e −2t ⇒ u1 = −e −2t ,<br />
u ′ 2 = e 3t (2e t )(−e −5t ) ⇒ u ′ 2 = −2e −t ⇒ u2 = 2e −t ,<br />
where we have chosen the constant <strong>of</strong> integration to be zero. Therefore, the particular<br />
solution we are looking for is given by<br />
yp = (−e −2t )(e 3t ) + (2e −t )(e 2t ) ⇒ yp = e t .<br />
Then, the general solution to the inhomogeneous equation above is given by<br />
y(t) = c1e 3t + c2e 2t + e t<br />
c1, c2 ∈ R.<br />
Example 2.6.2: Find a particular solution to the <strong>differential</strong> equation<br />
t 2 y ′′ − 2y = 3t 2 − 1,<br />
knowing that y1 = t 2 and y2 = 1/t are solutions to the homogeneous equation t 2 y ′′ −2y = 0.<br />
Solution: We first rewrite the inhomogeneous equation above in the form given in Theorem<br />
2.6.1, that is, we must divide the whole equation by t 2 ,<br />
y ′′ − 2 1<br />
1<br />
y = 3 − . ⇒ f(t) = 3 −<br />
t2 t2 t<br />
We now proceed to compute the Wronskian<br />
Wy1y2 (t) = (t2 <br />
−1<br />
)<br />
t2 <br />
1<br />
<br />
− (2t)<br />
t<br />
2 .<br />
⇒ Wy1y2 (t) = −3.<br />
We now use the equation in (2.6.1) to obtain the functions u1 and u2, that is,<br />
u ′ 1 = − 1<br />
t<br />
<br />
3 − 1<br />
t 2<br />
1<br />
−3<br />
= 1 1<br />
−<br />
t 3 t−3 ⇒ u1 = ln(t) + 1<br />
6 t−2 ,<br />
u ′ 2 = (t 2 )<br />
<br />
3 − 1<br />
t2 <br />
1<br />
−3<br />
= −t 2 + 1<br />
3 ⇒ u 2 = − 1<br />
3 t3 + 1<br />
3 t.<br />
⊳
84 G. NAGY – ODE july 2, 2013<br />
Therefore, a particular solution to the inhomogeneous equation above is ˜yp = u1y1 + u2y2, that is,<br />
˜yp =<br />
<br />
ln(t) + 1<br />
6 t−2(t<br />
2 ) + 1<br />
3 (−t3 + t)(t −1 )<br />
= t 2 ln(t) + 1 1<br />
−<br />
6 3 t2 + 1<br />
3<br />
= t 2 ln(t) + 1 1<br />
−<br />
2 3 t2<br />
= t 2 ln(t) + 1 1<br />
−<br />
2 3 y1(t).<br />
However, a simpler expression for a solution <strong>of</strong> the inhomogeneous equation above is<br />
yp = t 2 ln(t) + 1<br />
2 .<br />
Remark: Sometimes it could be difficult to remember the formulas for functions u1 and u2<br />
in (2.6.1). In such case one can always go back to the place in the pro<strong>of</strong> <strong>of</strong> Theorem 2.6.1<br />
where these formulas come from, that is, the system<br />
u ′ 1y1 + u ′ 2y2 = 0<br />
u ′ 1y ′ 1 + u ′ 2y ′ 2 = f.<br />
The system above could be simpler to remember than the <strong>equations</strong> in (2.6.1). We end this<br />
Section using the <strong>equations</strong> above to solve the problem in Example 2.6.2. Recall that the<br />
solutions to the homogeneous equation in Example 2.6.2 are y1(t) = t 2 , and y2(t) = 1/t,<br />
while the source function is f(t) = 3 − 1/t 2 . Then, we need to solve the system<br />
t 2 u ′ 1 + u ′ 1<br />
2<br />
2t u ′ 1 + u ′ 2<br />
= 0,<br />
t<br />
(−1)<br />
t2 1<br />
= 3 −<br />
t<br />
This is an algebraic linear system for u ′ 1 and u ′ 2. Those are simple to solve. From the equation<br />
on top we get u ′ 2 in terms <strong>of</strong> u ′ 1, and we use that expression on the bottom equation,<br />
u ′ 2 = −t 3 u ′ 1 ⇒ 2t u ′ 1 + t u ′ 1 = 3 − 1<br />
t2 ⇒ u′ 1 = 1 1<br />
− .<br />
t 3t3 Substitue back the expression for u ′ 1 in the first equation above and we get u ′ 2. We get,<br />
u ′ 1 = 1 1<br />
−<br />
t 3t3 u ′ 2 = −t 2 + 1<br />
3 .<br />
We should now integrate these functions to get u1 and u2 and then get the particular solution<br />
˜yp = u1y1 + u2y2. We do not repeat these calculations, since they are done Example 2.6.2.<br />
2 .<br />
⊳
2.6.2. Exercises.<br />
2.6.1.- . 2.6.2.- .<br />
G. NAGY – ODE July 2, 2013 85
86 G. NAGY – ODE july 2, 2013<br />
Chapter 3. Power series solutions<br />
This Chapter is focused to find solutions to second order linear <strong>differential</strong> <strong>equations</strong><br />
having variable coefficients. We use power series representations to find the solution. This<br />
solutions is usually defined only in a neighborhood <strong>of</strong> a point where the power series is<br />
centered at. In Section 3.1 we study the simplest case, when the power series is centered<br />
at a regular point <strong>of</strong> the equation. A point is regular iff the coefficient that multiplies<br />
the second derivative <strong>of</strong> the unknown does not vanishes there. In Section 3.2 we study a<br />
particular type <strong>of</strong> variable coefficients <strong>equations</strong> with singular points, called Euler <strong>differential</strong><br />
<strong>equations</strong>. We find solutions to Euler <strong>equations</strong> in a neighborhood <strong>of</strong> their singular points.<br />
In Section 3.3 we combine the results in the previous two Sections to find solutions to a<br />
large class <strong>of</strong> <strong>equations</strong>, called <strong>equations</strong> with regular-singular points, which include both<br />
Euler <strong>differential</strong> <strong>equations</strong> and regular point <strong>equations</strong>.<br />
3.1. Regular points<br />
In this Section we describe an idea to find solutions <strong>of</strong> variable coefficients <strong>equations</strong><br />
P (x) y ′′ + Q(x) y ′ + R(x) y = 0, (3.1.1)<br />
where the functions P , Q, and R are continuous. The main idea is to assume that the<br />
solution admits a power series representation around a point x0 ∈ R <strong>of</strong> the form<br />
∞<br />
y(x) = an(x − x0) n . (3.1.2)<br />
n=0<br />
In the case that this power series assumption is true, we can obtain the solution by simply<br />
introducing the expansion (3.1.2) into the <strong>differential</strong> equation in (3.1.1) and then obtaining<br />
the coefficients an that solve this equation. The power series representation <strong>of</strong> the solution<br />
defined in this way will be defined in an interval around x 0, where the radius <strong>of</strong> convergence<br />
can be determined after the solution is obtained. Notice that the assumption we have done<br />
in Eq. (3.1.2) implies that the solution y is well-defined at x0, since y(x0) = a0.<br />
In the case that the power series assumption in Eq. (3.1.2) is not true, when we solve for<br />
the coefficients an we obtain a contradiction. There exist <strong>equations</strong> where this power series<br />
assumption for their solutions is in fact not true, for example the Euler <strong>differential</strong> equation<br />
(x − 2) 2 y ′′ + 6(x − 2) y ′ + 6 y = 0. (3.1.3)<br />
We will see in the next Section that this equation has two linearly independent solutions<br />
1<br />
y1(x) =<br />
(x − 2) 2 , y 1<br />
2(x) = .<br />
(x − 2) 3<br />
Both solutions above diverge at x0 = 2, so if one choose to find solutions <strong>of</strong> the equation<br />
above around x0 = 2 using a power series expansion for the solution <strong>of</strong> the form given<br />
in (3.1.2), one is going to arrive to a contradiction. Equations similar to (3.1.3) will be<br />
studied in later sections.<br />
3.1.1. Review <strong>of</strong> power series. We present basic concepts about the power series representation<br />
<strong>of</strong> functions that we will need to find solutions to <strong>differential</strong> <strong>equations</strong>. We review<br />
only the main results without giving their pro<strong>of</strong>s, which can be found in standard Calculus<br />
textbooks. We start with the definition <strong>of</strong> a power series representation <strong>of</strong> a function.<br />
Definition 3.1.1. The power series <strong>of</strong> a function y : R → R centered at x0 ∈ R is<br />
∞<br />
y(x) = an (x − x0) n .<br />
n=0
G. NAGY – ODE July 2, 2013 87<br />
Example 3.1.1: We show few examples <strong>of</strong> power series representation <strong>of</strong> functions:<br />
1<br />
(a)<br />
1 − x =<br />
∞<br />
x n = 1 + x + x 2 + · · · . Here x0 = 0 and |x| < 1.<br />
(b) e x =<br />
n=0<br />
∞<br />
n=0<br />
n=0<br />
x n<br />
n!<br />
= 1 + x + x2<br />
2! + · · · . Here x 0 = 0 and x ∈ R.<br />
(c) The Taylor series <strong>of</strong> y : R → R centered at x0 ∈ R is<br />
∞ y<br />
y(x) =<br />
(n) (x0)<br />
n!<br />
(x − x 0) n ⇒ y(x) = y(x 0) + y ′ (x 0) (x − x 0) + y′′ (x0)<br />
2!<br />
(x − x 0) 2 + · · · .<br />
The Taylor series, reviewed in the Example above, can be very useful to find the power<br />
series representation <strong>of</strong> a function having infinitely many continuous derivatives.<br />
Example 3.1.2: Find the Taylor series <strong>of</strong> y(x) = sin(x) centered at x0 = 0.<br />
Solution: We need to compute the derivatives <strong>of</strong> the function y and evaluate these derivatives<br />
at the point we center the expansion, in this case x 0 = 0.<br />
y(x) = sin(x) ⇒ y(0) = 0, y ′ (x) = cos(x) ⇒ y ′ (0) = 1,<br />
y ′′ (x) = − sin(x) ⇒ y ′′ (0) = 0, y ′′′ (x) = − cos(x) ⇒ y ′′′ (0) = −1.<br />
Introduce the numbers y (n) (0) into Taylor’s formula,<br />
sin(x) = x − x3<br />
3!<br />
+ x5<br />
5!<br />
− · · · ⇒ sin(x) =<br />
∞<br />
n=0<br />
(−1) n<br />
(2n + 1)! x(2n+1) .<br />
A similar calculation leads us to the Taylor series <strong>of</strong> y(x) = cos(x) centered at x0 = 0,<br />
∞ (−1)<br />
cos(x) =<br />
n<br />
(2n)! x(2n) .<br />
n=0<br />
It is important to recall that the power series <strong>of</strong> a function may not be defined on the<br />
whole domain <strong>of</strong> the function. The following example shows a function with this property.<br />
Example 3.1.3: Find the Taylor series for y(x) = 1<br />
1 − x centered at x0 = 0.<br />
The power series expansion for the function y = 1<br />
(1−x)<br />
⊳<br />
⊳<br />
does not converge for |x| > 1. For<br />
example, the power series y(x) = 1<br />
1−x = ∞<br />
n=0 xn evaluated at x = 1 is the sum<br />
∞<br />
1 n = 1 + 1 + 1 + · · · ,<br />
n=0<br />
1<br />
which diverges. We conclude that the power series expansion for the function y(x) =<br />
(1 − x)<br />
diverges for x ∈ (−∞, −1) ∪ [1, ∞). We will show later on that this power series expansion<br />
converges for x ∈ (−1, 1). ⊳<br />
Later on we will need the notion <strong>of</strong> convergence <strong>of</strong> an infinite series in absolute value.
88 G. NAGY – ODE july 2, 2013<br />
Solution: Notice that this function is well<br />
defined for every x ∈ R − {1}. The function<br />
graph can be seen in Fig. 12. To find<br />
the Taylor series we need to compute the nderivative,<br />
y (n) (0). It simple to check that,<br />
y (n) (x) =<br />
n!<br />
(1 − x) n+1 , so y(n) (0) = n!.<br />
We conclude: y(x) = 1<br />
1 − x =<br />
∞<br />
x n .<br />
n=0<br />
Definition 3.1.2. The power series y(x) =<br />
series<br />
∞<br />
|an| |x − x0| n converges.<br />
n=0<br />
(<br />
−1<br />
y<br />
1<br />
)<br />
1<br />
y ( x ) = 1 / ( 1 − x )<br />
Figure 12. The graph <strong>of</strong><br />
1<br />
y =<br />
(1 − x) .<br />
∞<br />
an (x − x0) n converges absolutely iff the<br />
n=0<br />
∞ (−1)<br />
Example 3.1.4: One can show that the series s =<br />
n=1<br />
n<br />
converges, but this series does<br />
n<br />
∞ 1<br />
not converge absolutely, since diverges. ⊳<br />
n<br />
n=1<br />
Since power series representations <strong>of</strong> functions might not converge on the same domain<br />
where the function is defined, it is useful to introduce the region where the power series<br />
converges.<br />
∞<br />
Definition 3.1.3. The radius <strong>of</strong> convergence <strong>of</strong> a power series y(x) = an (x − x0) n<br />
is the number ρ 0 that satisfies both the series converges absolutely for |x − x0| < ρ and<br />
the series diverges for |x − x0| > ρ.<br />
The radius <strong>of</strong> convergence defines the size <strong>of</strong> the biggest open interval where the power series<br />
converges. This interval is symmetric around the series center point x0.<br />
diverges<br />
converges absolutely<br />
( )<br />
x<br />
0<br />
rho<br />
diverges<br />
Figure 13. Example <strong>of</strong> the radius <strong>of</strong> convergence.<br />
Example 3.1.5: We state the radius <strong>of</strong> convergence <strong>of</strong> few power series.<br />
1<br />
(1) The series<br />
1 − x =<br />
∞<br />
x n has radius <strong>of</strong> convergence ρ = 1.<br />
n=0<br />
x<br />
n=0<br />
x
(2) The series e x =<br />
∞<br />
n=0<br />
(3) The series sin(x) =<br />
(4) The series cos(x) =<br />
x n<br />
n!<br />
∞<br />
n=0<br />
∞<br />
n=0<br />
G. NAGY – ODE July 2, 2013 89<br />
has radius <strong>of</strong> convergence ρ = ∞.<br />
(−1) n<br />
(2n + 1)! x(2n+1) has radius <strong>of</strong> convergence ρ = ∞.<br />
(−1) n<br />
(2n)! x(2n) has radius <strong>of</strong> convergence ρ = ∞.<br />
The next result provides a way to compute the radius <strong>of</strong> convergence <strong>of</strong> a power series.<br />
∞<br />
Theorem 3.1.4 (Ratio test). Given the power series y(x) = an (x − x0) n , introduce<br />
the number L = lim<br />
n→∞<br />
n=0<br />
|an+1|<br />
. Then, the following statements hold:<br />
|an|<br />
(1) The power series converges in the domain |x − x0|L < 1.<br />
(2) The power series diverges in the domain |x − x0|L > 1.<br />
(3) The power series may or may not converge at |x − x0|L = 1.<br />
Therefore, if L = 0, then ρ = 1<br />
is the series radius <strong>of</strong> convergence; if L = 0, then the radius<br />
L<br />
<strong>of</strong> convergence is ρ = ∞.<br />
We finish this review subsection with a comment on summation index manipulations. It<br />
is well-known that the series<br />
y(x) = a0 + a1(x − x0) + a2(x − x0) 2 + · · ·<br />
is usually written using the summation notation<br />
∞<br />
y(x) = an (x − x0) n .<br />
n=0<br />
Now, the label n has nothing in particular, any other label defines the same series, for<br />
example the labels k and m below,<br />
∞<br />
y(x) = ak (x − x0) k ∞<br />
= am+3 (x − x0) m+3 .<br />
k=0<br />
m=−3<br />
In the first sum we just changed the label name from n to k, that is, k = n. In the second<br />
sum above we re-label the sum as follows, n = m + 3. Since the initial value for n is<br />
n = 0, then the initial value <strong>of</strong> m is m = −3. Derivatives <strong>of</strong> power series can be computed<br />
derivating every term in the power series, that is,<br />
y ′ ∞<br />
(x) = n an (x − x0) n−1 ∞<br />
= n an (x − x0) n−1 = a1 + 2a2(x − x0) + · · · .<br />
n=0<br />
n=1<br />
Notice that the power series for the y ′ can start either at n = 0 or n = 1, since the coefficients<br />
have a multiplicative factor n. We will usually re-label derivatives <strong>of</strong> power series as follows,<br />
y ′ ∞<br />
(x) = n an (x − x0) n−1 ∞<br />
= (m + 1) am+1 (x − x0) m<br />
n=1<br />
where m = n − 1, that is, n = m + 1.<br />
m=0
90 G. NAGY – ODE july 2, 2013<br />
3.1.2. Equations with regular points. In this subsection we concentrate in solving the<br />
following problem: Find a power series representation <strong>of</strong> the solutions y <strong>of</strong> the variable<br />
coefficients equation<br />
P (x) y ′′ + Q(x) y ′ + R(x) y = 0<br />
centered at any point x0 ∈ R where P (x0) = 0, The condition on the point x0, where we<br />
center the power series, is so important that we give it a particular name.<br />
Definition 3.1.5. Given continuous functions P , Q, R : (x1, x2) → R and the equation<br />
P (x) y ′′ + Q(x) y ′ + R(x) y = 0,<br />
a point x0 ∈ (x1, x2) is called a regular point <strong>of</strong> the equation above iff holds that P (x0) = 0.<br />
The point x 0 is called a singular point iff holds that P (x 0) = 0.<br />
The <strong>differential</strong> equation changes order at singular points. Since the coefficient P vanishes<br />
at a singular point, in a neighborhood <strong>of</strong> such a point the equation is second order and first<br />
order. This property makes finding solutions <strong>of</strong> the <strong>differential</strong> equation more difficult near<br />
singular points than near regular points.<br />
Example 3.1.6: The <strong>differential</strong> equation given in (3.1.3) has a singular point at x0 = 2,<br />
since the function P (x) = (x − 2) 2 vanishes at that point. Any other point is a regular<br />
point. ⊳<br />
Here is a summary <strong>of</strong> the power series method to find solutions to <strong>differential</strong> <strong>equations</strong><br />
<strong>of</strong> the form given in Eq. (3.1.1) near regular points x0.<br />
(a) Verify that the power series expansion is centered at a regular point, x 0;<br />
(b) Propose a power series representation <strong>of</strong> the solution centered at x0, given by<br />
∞<br />
y(x) = an (x − x0) n ;<br />
n=0<br />
(c) Introduce the power series expansion above into the <strong>differential</strong> equation and find a<br />
recurrence relation among the coefficients an;<br />
(d) Solve the recurrence relation in terms <strong>of</strong> free coefficients;<br />
(e) If possible, add up the resulting power series for the solution function y.<br />
We follow these steps in the examples below to find solutions to several <strong>differential</strong> <strong>equations</strong>.<br />
We start with a first order constant coefficient equation, and then we continue with<br />
a second order constant coefficient equation. The last two examples consider variable coefficient<br />
<strong>equations</strong>.<br />
Example 3.1.7: Find a power series solution y around the point x 0 = 0 <strong>of</strong> the equation<br />
y ′ + c y = 0, c ∈ R.<br />
Solution: We already know every solution to this equation. This is a first order, linear,<br />
<strong>differential</strong> equation, so using the method <strong>of</strong> integrating factor we find that the solution is<br />
y(x) = a 0 e −c x , a 0 ∈ R.<br />
We are now interested in obtaining such solution with the power series method. Although<br />
this is not a second order equation, the power series method still works in this example.<br />
Propose a solution <strong>of</strong> the form<br />
∞<br />
y = an x n ⇒ y ′ ∞<br />
= nan x (n−1) .<br />
n=0<br />
n=1
G. NAGY – ODE July 2, 2013 91<br />
We can start the sum in y ′ at n = 0 or n = 1. We choose n = 1, since it is more convenient<br />
later on. Introduce the expressions above into the <strong>differential</strong> equation,<br />
∞<br />
nan x n−1 ∞<br />
+ c an x n = 0.<br />
n=1<br />
Re-label the first sum above so that the functions x n−1 and x n in the first and second sum<br />
have the same label. One way is the following,<br />
∞<br />
n=0<br />
n=0<br />
(n + 1)a (n+1) x n +<br />
∞<br />
c an x n = 0<br />
n=0<br />
We can now write down both sums into one single sum,<br />
∞ n<br />
(n + 1)a(n+1) + c an x = 0.<br />
n=0<br />
Since the function on the left-hand side must be zero for every x ∈ R, we conclude that<br />
every coefficient that multiplies x n must vanish, that is,<br />
(n + 1)a (n+1) + c an = 0, n 0.<br />
The last equation is called a recurrence relation among the coefficients an. The solution <strong>of</strong><br />
this relation can be found by writing down the first few cases and then guessing the general<br />
expression for the solution, that is,<br />
n = 0, a1 = −c a0 ⇒ a1 = −c a0,<br />
n = 1, 2a 2 = −c a 1 ⇒ a 2 = c2<br />
2! a 0,<br />
n = 2, 3a3 = −c a2 ⇒ a3 = − c3<br />
3! a0,<br />
n = 3, 4a4 = −c a3 ⇒ a4 = c4<br />
4! a0.<br />
One can check that the coefficient an can be written as<br />
n cn<br />
an = (−1)<br />
n! a0,<br />
which implies that the solution <strong>of</strong> the <strong>differential</strong> equation is given by<br />
∞<br />
n cn<br />
y(x) = a0 (−1)<br />
n! xn ∞ (−c x)<br />
⇒ y(x) = a0 n<br />
n!<br />
n=0<br />
n=0<br />
⇒ y(x) = a 0 e −c x .<br />
Example 3.1.8: Find a power series solution y(x) around the point x 0 = 0 <strong>of</strong> the equation<br />
y ′′ + y = 0.<br />
Solution: We know that the solution can be found computing the roots <strong>of</strong> the characteristic<br />
polynomial r 2 + 1 = 0, which gives us the solutions<br />
y(x) = a 0 cos(x) + a 1 sin(x).<br />
We now recover this solution using the power series,<br />
∞<br />
∞<br />
y =<br />
n=0<br />
an x n ⇒ y ′ =<br />
n=1<br />
nan x (n−1) , ⇒ y ′′ =<br />
∞<br />
n(n − 1)an x (n−2) .<br />
n=2<br />
⊳
92 G. NAGY – ODE july 2, 2013<br />
Introduce the expressions above into the <strong>differential</strong> equation, which involves only the function<br />
and its second derivative,<br />
∞<br />
n(n − 1)an x n−2 ∞<br />
+ an x n = 0.<br />
n=2<br />
Re-label the first sum above, so that both sums have the same factor xn . One way is,<br />
∞<br />
(n + 2)(n + 1)a (n+2) x n ∞<br />
+ an x n = 0.<br />
n=0<br />
Now we can write both sums using one single sum as follows,<br />
∞<br />
n<br />
x = 0 ⇒ (n + 2)(n + 1)a(n+2) + an = 0. n 0.<br />
n=0<br />
(n + 2)(n + 1)a(n+2) + an<br />
The last equation is the recurrence relation. The solution <strong>of</strong> this relation can again be found<br />
by writing down the first few cases, and we start with even values <strong>of</strong> n, that is,<br />
n = 0, (2)(1)a2 = −a0 ⇒ a2 = − 1<br />
2! a0,<br />
n=0<br />
n=0<br />
n = 2, (4)(3)a4 = −a2 ⇒ a4 = 1<br />
4! a0,<br />
n = 4, (6)(5)a6 = −a4 ⇒ a6 = − 1<br />
6! a0.<br />
One can check that the even coefficients a2k can be written as<br />
a2k = (−1)k<br />
(2k)! a0.<br />
The coefficients an for the odd values <strong>of</strong> n can be found in the same way, that is,<br />
n = 1, (3)(2)a3 = −a1 ⇒ a3 = − 1<br />
3! a1,<br />
n = 3, (5)(4)a5 = −a3 ⇒ a5 = 1<br />
5! a1,<br />
n = 5, (7)(6)a7 = −a5 ⇒ a7 = − 1<br />
7! a1.<br />
One can check that the odd coefficients a2k+1 can be written as<br />
a2k+1 = (−1)k<br />
(2k + 1)! a1.<br />
Split the sum in the expression for y into even and odd sums. We have the expression for<br />
the even and odd coefficients. Therefore, the solution <strong>of</strong> the <strong>differential</strong> equation is given by<br />
y(x) = a 0<br />
∞<br />
k=0<br />
(−1) k<br />
(2k)! x2k + a 1<br />
∞<br />
k=0<br />
(−1) k<br />
(2k + 1)! x2k+1 .<br />
One can check that these are precisely the power series representations <strong>of</strong> the cosine and<br />
sine functions, respectively,<br />
y(x) = a0 cos(x) + a1 sin(x).<br />
⊳
G. NAGY – ODE July 2, 2013 93<br />
Example 3.1.9: Find the first four terms <strong>of</strong> the power series expansion around the point<br />
x0 = 1 <strong>of</strong> each fundamental solution to the <strong>differential</strong> equation<br />
y ′′ − x y ′ − y = 0.<br />
Solution: This is a <strong>differential</strong> equation we cannot solve with the methods <strong>of</strong> previous<br />
sections. This is a second order, variable coefficients equation. We use the power series<br />
method, so we look for solutions <strong>of</strong> the form<br />
∞<br />
y = an(x − 1) n ⇒ y ′ ∞<br />
= nan(x − 1) n−1 ⇒ y ′′ ∞<br />
= n(n − 1)an(x − 1) n−2 .<br />
n=0<br />
n=1<br />
We start working in the middle term in the <strong>differential</strong> equation. Since the power series is<br />
centered at x 0 = 1, it is convenient to re-write this term as x y ′ = [(x − 1) + 1] y ′ , that is,<br />
x y ′ =<br />
=<br />
∞<br />
nanx(x − 1) n−1<br />
n=1<br />
∞<br />
n=1<br />
n=1<br />
n−1<br />
nan (x − 1) + 1 (x − 1)<br />
n=2<br />
∞<br />
= nan(x − 1) n ∞<br />
+ nan(x − 1) n−1 . (3.1.4)<br />
As usual by now, the first sum on the right-hand side <strong>of</strong> Eq. (3.1.4) can start at n = 0, since<br />
we are only adding a zero term to the sum, that is,<br />
∞<br />
nan(x − 1) n ∞<br />
= nan(x − 1) n ;<br />
n=1<br />
while it is convenient to re-label the second sum in Eq. (3.1.4) follows,<br />
∞<br />
nan(x − 1) n−1 ∞<br />
= (n + 1)a (n+1)(x − 1) n ;<br />
n=1<br />
so both sums in Eq. (3.1.4) have the same factors (x − 1) n . We obtain the expression<br />
x y ′ ∞<br />
= nan(x − 1) n ∞<br />
+ (n + 1)a (n+1)(x − 1) n<br />
=<br />
n=0<br />
n=0<br />
n=0<br />
n=0<br />
n=1<br />
∞ n<br />
nan + (n + 1)a (n+1) (x − 1) . (3.1.5)<br />
n=0<br />
In a similar way re-label the index in the expression for y ′′ , so we obtain<br />
y ′′ ∞<br />
=<br />
n=0<br />
(n + 2)(n + 1)a (n+2)(x − 1) n . (3.1.6)<br />
If we use Eqs. (3.1.5)-(3.1.6) in the <strong>differential</strong> equation, together with the expression for y,<br />
the <strong>differential</strong> equation can be written as follows<br />
∞<br />
(n + 2)(n + 1)a (n+2)(x − 1) n ∞ <br />
∞<br />
n<br />
− nan + (n + 1)a (n+1) (x − 1) − an(x − 1) n = 0.<br />
n=0<br />
n=0<br />
We can now put all the terms above into a single sum,<br />
∞ <br />
n=0<br />
(n + 2)(n + 1)a (n+2) − (n + 1)a (n+1) − nan − an<br />
n=0<br />
<br />
(x − 1) n = 0.
94 G. NAGY – ODE july 2, 2013<br />
This expression provides the recurrence relation for the coefficients an with n 0, that is,<br />
which can be rewritten as follows,<br />
(n + 2)(n + 1)a (n+2) − (n + 1)a (n+1) − (n + 1)an = 0<br />
<br />
<br />
(n + 1)<br />
= 0,<br />
(n + 2)a (n+2) − a (n+1) − an<br />
(n + 2)a (n+2) − a (n+1) − an = 0. (3.1.7)<br />
We can solve this recurrence relation for the first four coefficients,<br />
n = 0 2a 2 − a 1 − a 0 = 0 ⇒ a 2 = a 1<br />
2 + a 0<br />
2 ,<br />
n = 1 3a 3 − a 2 − a 1 = 0 ⇒ a 3 = a1<br />
2<br />
+ a0<br />
6 ,<br />
n = 2 4a4 − a3 − a2 = 0 ⇒ a4 = a1 a0<br />
+<br />
4 6 .<br />
Therefore, the first terms in the power series expression for the solution y <strong>of</strong> the <strong>differential</strong><br />
equation are given by<br />
<br />
a0<br />
y = a0 + a1(x − 1) +<br />
2 + a <br />
1<br />
(x − 1)<br />
2<br />
2 <br />
a0<br />
+<br />
6 + a <br />
1<br />
(x − 1)<br />
2<br />
3 <br />
a0<br />
+<br />
6 + a <br />
1<br />
(x − 1)<br />
4<br />
4 + · · ·<br />
which can be rewritten as<br />
<br />
y = a0 1 + 1<br />
2 (x − 1)2 + 1<br />
6 (x − 1)3 + 1<br />
6 (x − 1)4 <br />
+ · · ·<br />
<br />
+ a1 (x − 1) + 1<br />
2 (x − 1)2 + 1<br />
2 (x − 1)3 + 1<br />
4 (x − 1)4 <br />
+ · · ·<br />
So the first four terms on each fundamental solution are given by<br />
y 1 = 1 + 1<br />
2 (x − 1)2 + 1<br />
6 (x − 1)3 + 1<br />
6 (x − 1)4 ,<br />
y 2 = (x − 1) + 1<br />
2 (x − 1)2 + 1<br />
2 (x − 1)3 + 1<br />
4 (x − 1)4 .<br />
Example 3.1.10: Find the first three terms <strong>of</strong> the power series expansion around the point<br />
x0 = 2 <strong>of</strong> each fundamental solution to the <strong>differential</strong> equation<br />
y ′′ − x y = 0.<br />
Solution: We then look for solutions <strong>of</strong> the form<br />
∞<br />
y = an(x − 2) n .<br />
n=0<br />
It is convenient to rewrite the function x y = [(x − 2) + 2] y, that is,<br />
∞<br />
xy = anx(x − 2) n<br />
=<br />
n=0<br />
∞<br />
n=0<br />
n=0<br />
n<br />
an (x − 2) + 2 (x − 2)<br />
∞<br />
= an(x − 2) n+1 ∞<br />
+ 2an(x − 2) n . (3.1.8)<br />
n=0<br />
⊳
G. NAGY – ODE July 2, 2013 95<br />
We now relabel the first sum on the right-hand side <strong>of</strong> Eq. (3.1.8) in the following way,<br />
∞<br />
an(x − 2) n+1 ∞<br />
=<br />
n=0<br />
n=1<br />
We do the same type <strong>of</strong> re-labeling on the expression for y ′′ ,<br />
y ′′ ∞<br />
= n(n − 1)an(x − 2) n−2<br />
=<br />
n=2<br />
∞<br />
n=0<br />
(n + 2)(n + 1)a (n+2)(x − 2) n .<br />
Then, the <strong>differential</strong> equation above can be written as follows<br />
∞<br />
(n + 2)(n + 1)a (n+2)(x − 2) n ∞<br />
− 2an(x − 2) n ∞<br />
−<br />
n=0<br />
(2)(1)a2 − 2a0 +<br />
n=0<br />
a (n−1)(x − 2) n . (3.1.9)<br />
n=1<br />
a (n−1)(x − 2) n = 0<br />
∞ <br />
<br />
(n + 2)(n + 1)a (n+2) − 2an − a (n−1) (x − 2) n = 0.<br />
n=1<br />
So the recurrence relation for the coefficients an is given by<br />
a2 − a0 = 0, (n + 2)(n + 1)a (n+2) − 2an − a (n−1) = 0, n 1.<br />
We can solve this recurrence relation for the first four coefficients,<br />
n = 0 a2 − a0 = 0 ⇒ a2 = a0,<br />
n = 1 (3)(2)a3 − 2a1 − a0 = 0 ⇒ a3 = a0<br />
6<br />
+ a1<br />
3 ,<br />
n = 2 (4)(3)a4 − 2a2 − a1 = 0 ⇒ a4 = a0 a1<br />
+<br />
6 12 .<br />
Therefore, the first terms in the power series expression for the solution y <strong>of</strong> the <strong>differential</strong><br />
equation are given by<br />
y = a0 + a1(x − 2) + a0(x − 2) 2 <br />
a0 a1<br />
+ + (x − 2)<br />
6 3<br />
3 <br />
a0 a1<br />
+ + (x − 2)<br />
6 12<br />
4 + · · ·<br />
which can be rewritten as<br />
<br />
y = a0 1 + (x − 2) 2 + 1<br />
6 (x − 2)3 + 1<br />
6 (x − 2)4 <br />
+ · · ·<br />
<br />
+ a1 (x − 2) + 1<br />
3 (x − 2)3 + 1<br />
12 (x − 2)4 <br />
+ · · ·<br />
So the first three terms on each fundamental solution are given by<br />
y 1 = 1 + (x − 2) 2 + 1<br />
6 (x − 2)3 ,<br />
y 2 = (x − 2) + 1<br />
3 (x − 2)3 + 1<br />
12 (x − 2)4 .<br />
⊳
96 G. NAGY – ODE july 2, 2013<br />
3.1.3. Exercises.<br />
3.1.1.- . 3.1.2.- .
G. NAGY – ODE July 2, 2013 97<br />
3.2. The Euler equation<br />
3.2.1. The main result. In this Section we study how to find solutions <strong>of</strong> certain <strong>differential</strong><br />
<strong>equations</strong> with singular points, for example an equation like this one:<br />
(x − 2) 2 y ′′ + 6(x − 2) y ′ + 6 y = 0.<br />
We are interested in finding the solution values y(x) for x arbitrary close to the singular<br />
point x0 = 2. Therefore, we are not interested in a power series expansion <strong>of</strong> the solution<br />
around a regular point, which in the example above is any point different from x0 = 2. In<br />
the previous Section we have stated without pro<strong>of</strong> that the solutions defined arbitrary close<br />
to x0 = 2 are given by<br />
1<br />
y1(x) =<br />
(x − 2) 2 , y 1<br />
2(x) = .<br />
(x − 2) 3<br />
In this Section we show how to obtain these solutions. We start by introducing the type <strong>of</strong><br />
<strong>equations</strong> we are interested in.<br />
Definition 3.2.1. Given real constants p0, q0, the Euler <strong>differential</strong> equation for the<br />
unknown y with singular point at x0 ∈ R is given by<br />
(x − x0) 2 y ′′ + p0 (x − x0) y ′ + q0 y = 0.<br />
We take the time to raise few comments <strong>of</strong> this Euler <strong>differential</strong> equation. First, this is<br />
a variable coefficients equation. So, it is clear that the exponential functions y(x) = e rx are<br />
not solutions <strong>of</strong> the Euler equation. Just introduce such a function into the equation, and<br />
it is simple to show that there is no r such that the exponential is solution. Second, notice<br />
that the point x0 ∈ R is a singular point <strong>of</strong> the equation. The particular case x0 = 0 is<br />
x 2 y ′′ + p0 x y ′ + q0 y = 0.<br />
The following result summarizes what is known about solutions <strong>of</strong> an Euler equation.<br />
Theorem 3.2.2 (Euler equation). Given p0, q0, and x0 ∈ R consider the Euler equation<br />
(x − x0) 2 y ′′ + p0 (x − x0) y ′ + q0 y = 0. (3.2.1)<br />
Let r± be the roots <strong>of</strong> the Euler characteristic polynomial p(r) = r(r − 1) + p 0r + q 0.<br />
(a) If r+ = r−, then the general solution <strong>of</strong> Eq. (3.2.1) is given by<br />
y(t) = c0|x − x0| r- + c1|x − x0| r- , x = x0, c0, c1 ∈ R.<br />
(b) If r+ = r− ∈ R, then the general solution <strong>of</strong> Eq. (3.2.1) is given by<br />
<br />
y(t) = c0 + c1 ln |x − x0| <br />
|x − x0| r- , x = x0, c0, c1 ∈ R.<br />
Furthermore, given real constants x 1 = x 0, y 0 and y 1, there is a unique solution to the initial<br />
value problem given by Eq. (3.2.1) and the initial conditions<br />
y(x 1) = y 0, y ′ (x 1) = y 1.<br />
Remark: Although we call the solutions found in (a)-(b) the general solution, they do not<br />
include all possible solutions to the <strong>differential</strong> equation. For example, consider the case that<br />
the roots <strong>of</strong> the Euler characteristic polynomial are different and one root is odd, say r+ is<br />
odd and r+ = r−. Then, the formula provided in (a) is given by y = c1|x−x0| r- +c2|x−x0| r- ,<br />
but this expression does not include the solution y = (x − x0) r- . So, to simplify the writing<br />
<strong>of</strong> the Theorem above we do not include all possible solutions to the equation, although we<br />
keep the name general solution for the solutions we present in that Theorem.
98 G. NAGY – ODE july 2, 2013<br />
We do not provide a pro<strong>of</strong> <strong>of</strong> this Theorem. We just mention the main idea behind this<br />
pro<strong>of</strong>. To understand this main idea it is useful to recall what we did to find solutions to<br />
constant coefficient <strong>equations</strong> <strong>of</strong> the form<br />
y ′′ + a1 y ′ + a0 y = 0.<br />
Since the derivatives <strong>of</strong> the exponential function are proportional to the same exponential,<br />
that is, e rx ′ = re rx , one can see that introducing this exponential y = e rx into the equation<br />
above converts the <strong>differential</strong> equation for y into an algebraic equation for r as follows,<br />
(r 2 + a1r + a0) e rx = 0 ⇔ r 2 + a1r + a0 = 0.<br />
This is the central idea to find solutions to a constant coefficient equation, that the exponential<br />
function disappears from the equation, leaving only an algebraic equation for the<br />
coefficient r. Later on one can show that the solutions obtained in this way are in fact all<br />
possible solution, as it was shown in the pro<strong>of</strong> <strong>of</strong> Theorem 2.2.2.<br />
We have seen that exponential functions are not appropriate to find solutions <strong>of</strong> the Euler<br />
equation. But the idea above can be carried out on the Euler equation<br />
(x − x0) 2 y ′′ + p0 (x − x0) y ′ + q0 y = 0,<br />
we just need a different function. Now, recall that power functions y = (x − x0) r satisfy<br />
y ′ = r(x − x0) r−1 ⇒ (x − x0)y ′ = r(x − x0) r .<br />
A similar property holds for the second derivative,<br />
y ′′ = r(r − 1)(x − x0) r−2 ⇒ (x − x0) 2 y ′ = r(r − 1)(x − x0) r .<br />
Hence, introducing this power function into the Euler equation above transform this <strong>differential</strong><br />
equation for y into an algebraic equation for r as follows<br />
<br />
r(r − 1) + p0r + q0 (x − x0) r = 0 ⇔ r(r − 1) + p0r + q0 = 0.<br />
The constant r must be a root <strong>of</strong> the Euler characteristic polynomial p(r) = r(r−1)+p0r+q0.<br />
And we again have different cases depending whether this polynomial has two different roots<br />
or only one. These cases are described in Theorem 3.2.2. Notice that this Theorem asserts<br />
more than that. The Theorem asserts that these are the only solutions <strong>of</strong> the Euler equation,<br />
and we are not proving this here.<br />
Example 3.2.1: Find the general solution <strong>of</strong> the Euler equation<br />
x 2 y ′′ + 4x y ′ + 2 y = 0.<br />
Solution: We look for solutions <strong>of</strong> the form y(x) = x r , which implies that<br />
x y ′ (x) = rx r , x 2 y ′′ (x) = r(r − 1) x r ,<br />
therefore, introducing this function y into the <strong>differential</strong> equation we obtain<br />
r(r − 1) + 4r + 2 x r = 0 ⇔ r(r − 1) + 4r + 2 = 0.<br />
The solutions are computed in the usual way,<br />
r 2 + 3r + 2 = 0 ⇒ r = 1<br />
√ <br />
−3 ± 9 − 8 ⇒<br />
2<br />
So the general solution <strong>of</strong> the <strong>differential</strong> equation above is given by<br />
y(x) = c1 x −1 + c2 x −2 .<br />
r1 = −1<br />
r2 = −2.<br />
⊳
G. NAGY – ODE July 2, 2013 99<br />
Example 3.2.2: Find the general solution <strong>of</strong> the Euler equation<br />
x 2 y ′′ − 3x y ′ + 13 y = 0.<br />
Solution: We look for solutions <strong>of</strong> the form y(x) = x r , which implies that<br />
x y ′ (x) = rx r , x 2 y ′′ (x) = r(r − 1) x r ,<br />
therefore, introducing this function y into the <strong>differential</strong> equation we obtain<br />
r(r − 1) − 3r + 13 x r = 0 ⇔ r(r − 1) − 3r + 13 = 0.<br />
The solutions are computed in the usual way,<br />
r 2 − 4r + 13 = 0 ⇒ r = 1<br />
√ <br />
4 ± −36 ⇒<br />
2<br />
So the general solution <strong>of</strong> the <strong>differential</strong> equation above is given by<br />
r1 = 2 + 3i<br />
r2 = 2 − 3i.<br />
y(x) = c1 x (2+3i) + c2 x (2−3i) . (3.2.2)<br />
⊳<br />
3.2.2. The complex-valued roots. In the case that the roots <strong>of</strong> the Euler characteristic<br />
polynomial are complex-valued, there always is linear combination <strong>of</strong> complex-valued fundamental<br />
solutions that results in real-valued fundamental solutions. The reason is that the<br />
Euler <strong>differential</strong> equation coefficients P 0, q 0 are real-valued. For example, the solution <strong>of</strong><br />
the example above, given in Eq. (3.2.2) can also be expressed as follows,<br />
y(t) = ˜c1 x 2 cos ln(|x| 3 ) + ˜c2 x 2 sin ln(|x| 3 ) , (3.2.3)<br />
where the constants ˜c 1, ˜c 2 are different form the constants c1, c2 present in (3.2.2). The<br />
steps needed to obtain Eq. (3.2.3) are summarized in the pro<strong>of</strong> <strong>of</strong> the Corollary below.<br />
Corollary 3.2.3 (Complex roots). If p0, q0 ∈ R satisfy that (p0 − 1) 2 <br />
− 4q0 < 0, then<br />
the indicial polynomial p(r) = r(r − 1) + p0r + q0 <strong>of</strong> the Euler equation<br />
has complex roots r± = α ± iβ, where<br />
(x − x0) 2 y ′′ + p0(x − x0) y ′ + q0 y = 0 (3.2.4)<br />
4q0 − (p 0 − 1) 2 .<br />
α = − (p0 − 1)<br />
, β =<br />
2<br />
1<br />
2<br />
Furthermore, a fundamental set <strong>of</strong> solution to Eq. (3.2.4) is<br />
˜y 1(x) = |x − x 0| (α+iβ) , ˜y 2(x) = |x − x 0| (α−iβ) ,<br />
while another fundamental set <strong>of</strong> solutions to Eq. (3.2.4) is<br />
y1(x) = |x − x0| α cos β ln |x − x0| , y2(x) = |x − x0| α sin β ln |x − x0| .<br />
Pro<strong>of</strong> <strong>of</strong> Corollary 3.2.3: For simplicity consider the case x0 = 0. Since the functions<br />
˜y1(x) = x (α+iβ) , ˜y2(x) = x (α−iβ) ,<br />
are solutions to the <strong>differential</strong> equation in the Corollary, then so are the functions<br />
y1(x) = 1<br />
˜y1(x) + ˜y2(x)<br />
2<br />
, y2(x) = 1 <br />
˜y1(x) − ˜y2(x)<br />
2i<br />
.<br />
Now, we need to recall several relations. First, recall the main property <strong>of</strong> exponentials,<br />
e (α±iβ)t = e αt e ±iβt . Second, recall the definition <strong>of</strong> the power function for complex exponents,<br />
that is,<br />
x iβ = e ln(|x|iβ ) = e iβ ln(|x|) .
100 G. NAGY – ODE july 2, 2013<br />
Third, recall the Euler equation for complex exponentials,<br />
e iβt = cos(βt) + i sin(βt),<br />
e −iβt = cos(βt) − i sin(βt).<br />
All these properties above are used in the following calculation:<br />
y1(x) = 1<br />
2 xα x iβ + x −iβ<br />
= 1<br />
2 xαe<br />
ln(|x|iβ ) ln(|x|<br />
+ e −iβ ) <br />
e iβ ln(|x|) −iβ<br />
+ e<br />
ln(|x|)<br />
= 1<br />
2 xα<br />
= x α cos β ln(|x|) <br />
= x α cos ln(|x| β ) ,<br />
We have then concluded that<br />
y2(x) = 1<br />
2i xα x iβ − x −iβ<br />
= 1<br />
2i xαe<br />
ln(|x|iβ ) ln(|x|<br />
− e −iβ ) <br />
e iβ ln(|x|) −iβ<br />
− e<br />
ln(|x|)<br />
= 1<br />
2i xα<br />
= x α sin β ln(|x|) <br />
= x α sin ln(|x| β ) ,<br />
y1(x) = x α cos ln(|x| β ) , y2(x) = x α sin ln(|x| β ) .<br />
This establishes the Corollary. <br />
Example 3.2.3: Find a real-valued general solution <strong>of</strong> the Euler equation<br />
x 2 y ′′ − 3x y ′ + 13 y = 0.<br />
Solution: The indicial equation is r(r − 1) − 3r + 13 = 0, with solutions<br />
A complex-valued general solution is<br />
A real-valued general solution is<br />
r 2 − 4r + 13 = 0 ⇒ r+ = 2 + 3i, r− = 2 − 3i.<br />
y(x) = ˜c1 |x| (2+3i) + ˜c2 |x| (2−3i)<br />
˜c1, ˜c2 ∈ C.<br />
y(x) = c1 |x| 2 cos 3 ln |x| + c2 |x| 2 sin 3 ln |x| , c1, c2 ∈ R.<br />
Our last example considers the case where the Euler characteristic polynomial has a<br />
repeated root.<br />
Example 3.2.4: Find the general solution <strong>of</strong> the Euler equation<br />
x 2 y ′′ − 3x y ′ + 4 y = 0.<br />
Solution: We look for solutions <strong>of</strong> the form y(x) = x r , then the constant r must be solution<br />
<strong>of</strong> the Euler characteristic polynomial<br />
r(r − 1) − 3r + 4 = 0 ⇔ r 2 − 4r + 4 = 0 ⇒ r+ = r− = 2.<br />
Therefore, the general solution <strong>of</strong> the Euler equation in this case is given by<br />
y(x) = c1x 2 + c2x 2 ln(|x|).<br />
⊳<br />
⊳
3.2.3. Exercises.<br />
3.2.1.- . 3.2.2.- .<br />
G. NAGY – ODE July 2, 2013 101
102 G. NAGY – ODE july 2, 2013<br />
3.3. Regular-singular points<br />
In Sect. 3.2 we have introduced the Euler equation and we have seen that contains one<br />
singular point. In the same Section we have seen how to find solutions to this Euler equation<br />
that are well-defined arbitrary close to the singular point. In this Section we introduce a<br />
large set <strong>of</strong> <strong>differential</strong> <strong>equations</strong> containing singular points similar to the singular points<br />
<strong>of</strong> the Euler equation. We will call them <strong>differential</strong> <strong>equations</strong> with regular-singular points,<br />
where the name tries to suggest that they are singular points which happen to be not<br />
so singular. This type <strong>of</strong> <strong>equations</strong> include the <strong>equations</strong> with regular points studied in<br />
Section 3.1 and the Euler equation studied in Section 3.2. We show how to find at least<br />
one solution well-defined arbitrary close to the regular-singular points to this broad type <strong>of</strong><br />
<strong>equations</strong>.<br />
3.3.1. Finding regular-singular points. Recall that that a point x0 ∈ R is a singular<br />
point <strong>of</strong> the second order, linear, <strong>differential</strong> equation<br />
P (x) y ′′ + Q(x) y ′ + R(x) y = 0<br />
iff holds that P (x0) = 0. Regular-singular points are singular points which are not so<br />
singular. Here is a precise definition.<br />
Definition 3.3.1. A singular point x0 ∈ R <strong>of</strong> the equation<br />
P (x) y ′′ + Q(x) y ′ + R(x) y = 0<br />
is called a regular-singular point iff the following limits are finite,<br />
and both functions<br />
lim<br />
x→x0<br />
(x − x0) Q(x)<br />
, lim<br />
P (x)<br />
x→x0<br />
(x − x0) Q(x)<br />
,<br />
P (x)<br />
admit convergent Taylor series expansions around x0.<br />
(x − x0) 2 R(x)<br />
,<br />
P (x)<br />
(x − x0) 2 R(x)<br />
,<br />
P (x)<br />
Example 3.3.1: Show that the singular point <strong>of</strong> every Euler equation is a regular-singular.<br />
Solution: Given real constants p0, q0, x0, consider the general Euler equation<br />
(x − x0) 2 y ′′ + p0(x − x0) y ′ + q0 y = 0.<br />
In this case the functions P , Q and R introduced in the Definition 3.3.1 are given by<br />
Therefore, we obtain,<br />
P (x) = (x − x0) 2 , Q(x) = p0(x − x0), R(x) = q0.<br />
(x − x0) Q(x)<br />
P (x)<br />
= p0,<br />
(x − x0) 2 R(x)<br />
P (x)<br />
= q0.<br />
From these <strong>equations</strong> is trivial to see that both conditions given in the Definition 3.3.1 are<br />
satisfied by the coefficients <strong>of</strong> every Euler equation. Therefore, the singular point <strong>of</strong> every<br />
Euler equation is regular-singular. ⊳<br />
Example 3.3.2: Find the regular-singular points <strong>of</strong> the <strong>differential</strong> equation<br />
where α is a real constant.<br />
(1 − x 2 ) y ′′ − 2x y ′ + α(α + 1) y = 0,
G. NAGY – ODE July 2, 2013 103<br />
Solution: We first find the singular points <strong>of</strong> this equation, which are given by the zeros<br />
<strong>of</strong> the coefficient function that multiplies y ′′ , that is,<br />
P (x) = (1 − x 2 <br />
x0 = 1,<br />
) = (1 − x)(1 + x) = 0 ⇒<br />
x1 = −1.<br />
Let us start with the singular point x0 = 1. We then have<br />
(x − 1) Q(x)<br />
P (x)<br />
= (x − 1)(−2x)<br />
(1 − x)(1 + x)<br />
= 2x<br />
1 + x ,<br />
(x − 1) 2 R(x)<br />
P (x)<br />
= (x − 1)2α(α + 1) <br />
(1 − x)(1 + x)<br />
= (x − 1)α(α + 1) <br />
;<br />
1 + x<br />
which implies that both functions above have power series expansions around x0 = 1.<br />
Furthermore, the following limits are finite, namely,<br />
(x − 1) Q(x)<br />
(x − 1)<br />
lim<br />
= 1; lim<br />
x→1 P (x)<br />
x→1<br />
2 R(x)<br />
= 0.<br />
P (x)<br />
We conclude that x 0 = 1 is a regular-singular point <strong>of</strong> the <strong>differential</strong> equation. We now<br />
perform the same calculation with the point x 1 = −1, that means,<br />
(x + 1) Q(x)<br />
P (x)<br />
= (x + 1)(−2x)<br />
(1 − x)(1 + x)<br />
= − 2x<br />
1 − x ,<br />
(x + 1) 2 R(x)<br />
P (x)<br />
= (x + 1)2 α(α + 1) <br />
(1 − x)(1 + x)<br />
= (x + 1)α(α + 1) <br />
;<br />
1 − x<br />
which implies that both functions above have power series expansions around x1 = −1.<br />
Furthermore, the following limits are finite, namely,<br />
(x + 1) Q(x)<br />
(x + 1)<br />
lim<br />
= 1; lim<br />
x→−1 P (x)<br />
x→−1<br />
2 R(x)<br />
= 0.<br />
P (x)<br />
Therefore, the point x1 = −1 is also a regular-singular point <strong>of</strong> the <strong>differential</strong> equation. ⊳<br />
Example 3.3.3: Find the regular-singular points <strong>of</strong> the <strong>differential</strong> equation<br />
(x + 2) 2 (x − 1) y ′′ + 3(x − 1) y ′ + 2 y = 0.<br />
Solution: The singular point <strong>of</strong> the <strong>differential</strong> equation above are given by x0 = −2 and<br />
x 1 = 1. In the first case we have that<br />
(x + 2)Q(x) (x + 2)3(x − 1)<br />
lim<br />
= lim<br />
x→−2 P (x) x→−2 (x + 2) 2 = lim<br />
(x − 1) x→−2<br />
3<br />
= ∞,<br />
(x + 2)<br />
so x0 = −2 is not a regular-singular point. In the second case, x1 = 1, we obtain<br />
(x − 1) Q(x)<br />
P (x)<br />
= (x − 1)2(x − 1) <br />
(x + 2)(x − 1)<br />
2(x − 1)<br />
= − ,<br />
(x + 2) 2<br />
(x − 1) 2 R(x)<br />
P (x)<br />
2(x − 1)<br />
=<br />
2<br />
(x + 2) 2 (x − 1)<br />
= 2(x − 1)<br />
;<br />
(x + 2) 2<br />
which implies that both functions above have power series expansions around x1 = 1.<br />
Furthermore, the following limits are finite, namely,<br />
(x − 1) Q(x)<br />
(x − 1)<br />
lim<br />
= 0; lim<br />
x→1 P (x)<br />
x→1<br />
2 R(x)<br />
= 0.<br />
P (x)<br />
Therefore, the point x1 = −1 is a regular-singular point <strong>of</strong> the <strong>differential</strong> equation. ⊳
104 G. NAGY – ODE july 2, 2013<br />
Example 3.3.4: Find the regular-singular points <strong>of</strong> the <strong>differential</strong> equation<br />
x y ′′ − x ln(|x|) y ′ + 3x y = 0.<br />
Solution: The singular point is x 0 = 0. Although the following limits are finite,<br />
xQ(x) x<br />
lim = lim<br />
x→0 P (x) x→0<br />
x ln(|x|) <br />
ln(|x|)<br />
= lim<br />
x<br />
x→0 1 = lim<br />
x→0<br />
x<br />
x<br />
lim<br />
x→0<br />
2R(x) P (x)<br />
x<br />
= lim<br />
x→0<br />
2 (3x)<br />
x<br />
1<br />
x<br />
− 1<br />
x 2<br />
= lim<br />
x→0 3x2 = 0,<br />
= lim<br />
x→0 −x = 0,<br />
the point x0 = 0 is not a regular-singular point, since the function xQ/P does not have a<br />
power series expansion around zero, since<br />
xQ(x)<br />
P (x)<br />
= x ln(|x|),<br />
and the log function does not have a Taylor series at x0 = 0. ⊳<br />
3.3.2. Solving <strong>equations</strong> with regular-singular points. In the last part <strong>of</strong> this Section<br />
we study how to find solutions to <strong>differential</strong> <strong>equations</strong> having regular-singular points. We<br />
are interested in solutions well-defined arbitrary close to the regular-singular point. The<br />
main idea is this: Since near a regular-singular point the coefficients <strong>of</strong> the <strong>differential</strong><br />
equation are close to the coefficients <strong>of</strong> a particular Euler equation, then it could happen<br />
that near that regular-singular point their solutions could be also close to each other. More<br />
precisely, if x0 is a regular-singular point <strong>of</strong><br />
with limits<br />
P (x) y ′′ + Q(x) y ′ + R(x) y = 0,<br />
(x − x0)Q(x) (x − x0) lim<br />
= p0 and lim<br />
x→x0 P (x)<br />
x→x0<br />
2R(x) = q0, P (x)<br />
then the coefficients <strong>of</strong> the <strong>differential</strong> equation above near x0 are close to the coefficients<br />
<strong>of</strong> the Euler equation<br />
(x − x0) 2 y ′′ + p0(x − x0) y ′ + q0 y = 0.<br />
Since the <strong>differential</strong> equation is close to an Euler equation, then it could happen that the<br />
solutions <strong>of</strong> the <strong>differential</strong> equation might be close to the solutions <strong>of</strong> an Euler equation.<br />
And we know that at least one solution <strong>of</strong> an Euler equation is<br />
ye(x) = (x − x 0) r , (3.3.1)<br />
where the constant r is the root <strong>of</strong> the Euler characteristic polynomial, and x0 is the regularsingular<br />
point <strong>of</strong> the original equation. Therefore, the idea is to look for solutions <strong>of</strong> the<br />
original <strong>differential</strong> equation having the form<br />
y(x) = (x − x0) r<br />
∞<br />
an (x − x0) n ∞<br />
⇔ y(x) = an (x − x0) (n+r) . (3.3.2)<br />
n=0<br />
It is simple to see that this expression <strong>of</strong> the solution contains the idea that the values y(x)<br />
are close to ye(x) for values <strong>of</strong> x close to x0, since in this latter case we have that<br />
n=0<br />
y(x) = ye(x) a0 + a1(x − x0) + · · · a0 ye(x) for x x0,<br />
where the symbol a b, with a, b ∈ R means that |a − b| is close to zero.<br />
We remark that it could be possible that not every solution <strong>of</strong> a <strong>differential</strong> equation<br />
containing regular-singular points will have the form given in Eq. (3.3.2). However, it can<br />
be shown that at least one solution will have this form. A similar situation occurs with the
G. NAGY – ODE July 2, 2013 105<br />
solutions to the Euler equation, where there are cases with solutions containing logarithmic<br />
terms, which means that they do not have the form given in Eq. (3.3.1). For an explicit<br />
example <strong>of</strong> this situation, see the comment at the end <strong>of</strong> Example 3.3.5.<br />
A summary <strong>of</strong> the method to find solutions to <strong>differential</strong> <strong>equations</strong> having regularsingular<br />
points is the following:<br />
∞<br />
(1) Look for a solution y <strong>of</strong> the form y(x) =<br />
n=0<br />
an (x − x0) (n+r) ;<br />
(2) Introduce this power series expansion into the <strong>differential</strong> equation and find both the<br />
exponent r and a recurrence relation for the coefficients an;<br />
(3) First find the solutions for the constant r. Then, introduce this result for r into the<br />
recurrence relation for the coefficients an. Only then, solve this latter recurrence relation<br />
for the coefficients an.<br />
Example 3.3.5: Find the solution y near the regular-singular point x0 = 0 <strong>of</strong> the equation<br />
x 2 y ′′ − x(x + 3) y ′ + (x + 3) y = 0.<br />
Solution: We propose a solution <strong>of</strong> the form<br />
∞<br />
y(x) = an x (n+r) .<br />
n=0<br />
The first and second derivatives are given by<br />
y ′ ∞<br />
(x) =<br />
n=0<br />
(n + r)an x (n+r−1) , y ′′ (x) =<br />
∞<br />
(n + r)(n + r − 1)an x (n+r−2) .<br />
In the case r = 0 we had the relation ∞ n=0 nan x (n−1) = ∞ n=1 nan x (n−1) . But in our<br />
case r = 0, so we do not have the freedom to change in this way the starting value <strong>of</strong> the<br />
summation index n. If we want to change the initial value for n, we have to re-label the<br />
summation index. We now introduce these expressions into the <strong>differential</strong> equation. It is<br />
convenient to do this step by step. We start with the term (x + 3)y, which has the form,<br />
∞<br />
(x + 3) y = (x + 3) an x (n+r)<br />
=<br />
∞<br />
n=0<br />
n=1<br />
n=0<br />
an x (n+r+1) +<br />
n=0<br />
∞<br />
3an x (n+r)<br />
n=0<br />
∞<br />
= a (n−1) x (n+r) ∞<br />
+ 3an x (n+r) . (3.3.3)<br />
n=0<br />
We continue with the term containing y ′ , that is,<br />
−x(x + 3) y ′ = −(x 2 ∞<br />
+ 3x) (n + r)an x (n+r−1)<br />
n=0<br />
n=0<br />
∞<br />
= − (n + r)an x (n+r+1) ∞<br />
− 3(n + r)an x (n+r)<br />
n=1<br />
n=0<br />
∞<br />
= − (n + r − 1)a (n−1) x (n+r) ∞<br />
− 3(n + r)an x (n+r) . (3.3.4)<br />
n=0
106 G. NAGY – ODE july 2, 2013<br />
Then, we compute the term containing y ′′ as follows,<br />
x 2 y ′′ = x 2<br />
=<br />
∞<br />
(n + r)(n + r − 1)an x (n+r−2)<br />
n=0<br />
∞<br />
(n + r)(n + r − 1)an x (n+r) . (3.3.5)<br />
n=0<br />
As one can see from Eqs.(3.3.3)-(3.3.5), the guiding principle to rewrite each term is to<br />
have the power function x (n+r) labeled in the same way on every term. For example, in<br />
Eqs.(3.3.3)-(3.3.5) we do not have a sum involving terms with factors x (n+r−1) or factors<br />
x (n+r+1) . Then, the <strong>differential</strong> equation can be written as follows,<br />
∞<br />
(n + r)(n + r − 1)an x (n+r) ∞<br />
− (n + r − 1)a (n−1) x (n+r)<br />
n=0<br />
n=0<br />
n=1<br />
n=1<br />
∞<br />
− 3(n + r)an x (n+r) ∞<br />
+ a (n−1) x (n+r) ∞<br />
+ 3an x (n+r) = 0.<br />
In the equation above we need to split the sums containing terms with n 0 into the term<br />
n = 0 and a sum containing the terms with n 1, that is,<br />
<br />
r(r − 1) − 3r + 3 a0x r +<br />
∞ <br />
<br />
(n + r)(n + r − 1)an − (n + r − 1)a (n−1) − 3(n + r)an + a (n−1) + 3an x (n+r) = 0,<br />
n=1<br />
and this expression can be rewritten as follows,<br />
<br />
r(r − 1) − 3r + 3 a0x r +<br />
∞ (n <br />
+ r)(n + r − 1) − 3(n + r) + 3 an − (n + r − 1 − 1)a (n−1) x (n+r) = 0<br />
n=1<br />
and then,<br />
<br />
r(r − 1) − 3r + 3 a0x r +<br />
∞ (n <br />
+ r)(n + r − 1) − 3(n + r − 1) an − (n + r − 2)a (n−1) x (n+r) = 0<br />
hence,<br />
n=1<br />
r(r − 1) − 3r + 3 a0x r +<br />
n=0<br />
∞ <br />
<br />
(n + r − 1)(n + r − 3)an − (n + r − 2)a (n−1) x (n+r) = 0.<br />
n=1<br />
The Euler characteristic equation and the recurrence relation are given by the <strong>equations</strong><br />
r(r − 1) − 3r + 3 = 0, (3.3.6)<br />
(n + r − 1)(n + r − 3)an − (n + r − 2)a (n−1) = 0. (3.3.7)<br />
The way to solve these <strong>equations</strong> in (3.3.6)-(3.3.7) is the following: First, solve Eq. (3.3.6) for<br />
the exponent r, which in this case has two solutions r±; second, introduce the first solution<br />
r+ into the recurrence relation in Eq. (3.3.7) and solve for the coefficients an; the result is a<br />
solution y+ <strong>of</strong> the original <strong>differential</strong> equation; then introduce the second solution r− into
G. NAGY – ODE July 2, 2013 107<br />
Eq. (3.3.7) and solve again for the coefficients an; the new result is a second solution y−.<br />
Let us follow this procedure in the case <strong>of</strong> the <strong>equations</strong> above:<br />
r 2 − 4r + 3 = 0 ⇒ r± = 1<br />
<br />
√ r+ = 3,<br />
4 ± 16 − 12 ⇒<br />
2<br />
r− = 1.<br />
Introducing the value r+ = 3 into Eq. (3.3.7) we obtain<br />
(n + 2)n an − (n + 1)an−1 = 0.<br />
One can check that the solution y+ obtained form this recurrence relation is given by<br />
y+(x) = a0 x 3<br />
1 + 2 1<br />
x +<br />
3 4 x2 + 1<br />
15 x3 <br />
+ · · · .<br />
Introducing the value r− = 1 into Eq. (3.3.7) we obtain<br />
n(n − 2)an − (n − 1)an−1 = 0.<br />
One can also check that the solution y− obtained form this recurrence relation is given by<br />
<br />
y−(x) = a2 x x 2 + 2<br />
3 x3 + 1<br />
4 x4 + 1<br />
15 x5 <br />
+ · · · ,<br />
= a2 x 3<br />
1 + 2 1<br />
x +<br />
3 4 x2 + 1<br />
15 x3 <br />
+ · · · ⇒ y− = a2<br />
y+.<br />
a1<br />
Remark: We have found that the solution y− is not linearly independent <strong>of</strong> the solution<br />
y+. This Example shows an important property <strong>of</strong> the method described in this Section<br />
to find solutions <strong>of</strong> <strong>equations</strong> near a regular-singular point; a property that we already<br />
mentioned earlier in this Section: The method does not provide all solutions <strong>of</strong> a <strong>differential</strong><br />
equation near a regular-singular point, it only provides at least one solution near a regularsingular<br />
point. It can be shown the following result: If the roots <strong>of</strong> the Euler characteristic<br />
polynomial r+, r− differ by an integer, then the corresponding solutions y+, y− are linearly<br />
dependent. It can be proven that second solution linearly independent to y+ exists, but it<br />
has the solution a more complicated expression than the one given in Eq. (3.3.2). This new<br />
solution involves logarithmic terms. We do not study such solutions in these notes. ⊳
108 G. NAGY – ODE july 2, 2013<br />
3.3.3. Exercises.<br />
3.3.1.- . 3.3.2.- .
G. NAGY – ODE July 2, 2013 109<br />
Chapter 4. The Laplace Transform<br />
In this Chapter we introduce and develop an idea to find solutions to <strong>differential</strong> <strong>equations</strong><br />
that is radically different from all we have studied so far. The idea is centered around the<br />
Laplace Transform <strong>of</strong> a function. The Laplace Transform is a functional, that is, is a map<br />
that transforms a function into a new function. The Laplace Transform converts derivatives<br />
into multiplications, hence it converts <strong>differential</strong> <strong>equations</strong> into algebraic <strong>equations</strong>.<br />
For certain types <strong>of</strong> functions the Laplace Transform is invertible. So, Laplace transforms<br />
can be used to solve <strong>differential</strong> <strong>equations</strong>: First transform the whole <strong>differential</strong> equation<br />
into an algebraic equation; second solve the algebraic equation; third transform back the<br />
result, which is the solution to the <strong>differential</strong> equation. Laplace Transforms can be used<br />
to find solutions to <strong>differential</strong> <strong>equations</strong> with arbitrary source functions but with constant<br />
coefficients.<br />
4.1. Definition and properties<br />
The Laplace Transform is defined by a particular improper integral, so we start this<br />
Section with a brief review on improper integral and few examples. Then we introduce the<br />
definition <strong>of</strong> the Laplace Transform and then present several examples. In the following<br />
Sections we explain how to use such transform to find solutions to <strong>differential</strong> <strong>equations</strong>.<br />
The Laplace transform method is useful to solve non-homogeneous <strong>differential</strong> <strong>equations</strong>,<br />
but only with constant coefficients.<br />
4.1.1. Review <strong>of</strong> improper integrals. Improper integrals are integrals on unbounded<br />
domains. They are defined as a limit <strong>of</strong> definite integrals. More precisely,<br />
∞<br />
N<br />
g(t) dt = lim g(t) dt.<br />
N→∞<br />
t0<br />
We say that the integral above converges iff the limit exists, and diverges iff the limit does<br />
not exist. In the following Example we compute an improper integral that is very useful to<br />
compute Laplace Transforms.<br />
∞<br />
Example 4.1.1: Compute the improper integral e −at dt, with a ∈ R.<br />
Solution: Following the definition above we need to first compute a definite integral and<br />
then take a limit. So, from the definition,<br />
∞<br />
N<br />
e −at dt.<br />
We first compute the definite integral,<br />
∞<br />
0<br />
0<br />
e −at dt = lim<br />
N→∞<br />
e −at dt = lim<br />
N→∞<br />
Since lim<br />
N→∞ e−aN = 0, for a > 0, we conclude<br />
∞<br />
0<br />
t0<br />
0<br />
0<br />
1<br />
<br />
− e<br />
a<br />
−aN <br />
− 1 .<br />
e −at dt = 1<br />
, a > 0. (4.1.1)<br />
a<br />
In the case that a = 0, the improper integral diverges, since we are computing the area <strong>of</strong><br />
a rectangle with sides equal to one and infinity. In the case a < 0 holds lim<br />
N→∞ e−aN = ∞.<br />
Therefore, the improper integral converges only for a > 0 and it is given by the expression<br />
in Eq. (4.1.1). ⊳
110 G. NAGY – ODE july 2, 2013<br />
4.1.2. Definition <strong>of</strong> the Laplace Transform. The Laplace Transform <strong>of</strong> a function is<br />
defined as an improper integral <strong>of</strong> the function times an exponential.<br />
Definition 4.1.1. The function ˆ f : Dfˆ → R is the Laplace Transform <strong>of</strong> the function<br />
f : (0, ∞) → R iff holds<br />
∞<br />
ˆf(s) = e −st f(t) dt (4.1.2)<br />
where D ˆ f ⊂ R is the open set where the integral above converges.<br />
We will <strong>of</strong>ten use an alternative notation for ˆ f, the Laplace transform <strong>of</strong> f, namely,<br />
0<br />
ˆf = L[f].<br />
This notation emphasizes that the Laplace transform defines a map from the set <strong>of</strong> piecewise<br />
continuous functions into the same set. Maps between function spaces are <strong>of</strong>ten called<br />
functionals. Recalling that a scalar-valued function <strong>of</strong> a single variable is a map denoted<br />
by t ↦→ f(t), the same notation can be used with the Laplace transform, that is, f ↦→ L[f].<br />
In this Section we are interested to study the properties <strong>of</strong> this map L. In particular we<br />
will show in Theorem 4.1.3 that this map is linear, and in Theorem 4.1.5 that this map is<br />
one-to-one and so invertible on the appropriate sets.<br />
It can be seen from this definition that the Laplace transform <strong>of</strong> a function f is an improper<br />
integral. We will see in the Examples below that this improper integral in Eq. (4.1.2)<br />
does not converge for every s ∈ (0, ∞). The interval where the Laplace transform <strong>of</strong> a function<br />
f is well-defined depends on the particular function f. So, we will see below that L[e at ]<br />
with a > 0 is well-defined for s > a, while L[sin(at)] is well-defined for s > 0. We now<br />
compute the Laplace transforms <strong>of</strong> simple functions.<br />
Example 4.1.2: Compute L[1].<br />
Solution: The function f(t) = 1 is the simplest function we can use to find its Laplace<br />
transform. Following the definition,<br />
∞<br />
L[1] = e −st dt<br />
0<br />
We have computed this improper integral in Example 4.1.1. Just replace a = s in that<br />
example. The result is,<br />
L[1] = 1<br />
s<br />
for s > 0,<br />
and the Laplace Transform is not defined for s 0. ⊳<br />
Example 4.1.3: Compute L[e at ], where a ∈ R.<br />
Solution: Following the definition <strong>of</strong> a Laplace Transform we obtain,<br />
L[e at ∞<br />
] = e −st e at ∞<br />
dt = e −(s−a)t dt.<br />
0<br />
Here we can again use the result in Example 4.1.1, just replace a in that Example by (s−a).<br />
We obtain,<br />
L[e at 1<br />
] =<br />
for s > a,<br />
(s − a)<br />
and the Laplace Transform is not defined for s a. ⊳<br />
0
Example 4.1.4: Compute L[te at ], where a ∈ R.<br />
G. NAGY – ODE July 2, 2013 111<br />
Solution: In this case the calculation is more complicated than above, since we need to<br />
integrate by parts. We start with the definition <strong>of</strong> the Laplace Transform,<br />
L[te at ∞<br />
n<br />
] =<br />
te −(s−a)t dt.<br />
We now integrate by parts,<br />
L[te at ] = lim<br />
n→∞<br />
0<br />
e −st te at dt = lim<br />
n→∞<br />
<br />
− 1<br />
(s − a) te−(s−a)t<br />
<br />
<br />
n<br />
0<br />
0<br />
+ 1<br />
s − a<br />
n<br />
0<br />
e −(s−a)t <br />
dt ,<br />
that is,<br />
L[te at <br />
] = lim −<br />
n→∞<br />
1<br />
(s − a) te−(s−a)t<br />
<br />
<br />
n 1<br />
<br />
<br />
− e−(s−a)t<br />
0 (s − a) 2 n<br />
0<br />
Notice that the first term on the right-hand side <strong>of</strong> the last line above vanishes, since<br />
Regarding the other term,<br />
1<br />
lim −<br />
n→∞ (s − a) ne−(s−a)n = 0,<br />
lim<br />
n→∞ −<br />
Therefore, we conclude that<br />
1<br />
(s − a) 2 e−(s−a)n = 0,<br />
L[te at ] =<br />
1<br />
(s − a) te−(s−a)t t=0 = 0.<br />
1<br />
(s − a) 2 e−(s−a)t t=0 =<br />
1<br />
, s > a.<br />
(s − a) 2<br />
Example 4.1.5: Compute L[sin(at)], where a ∈ R.<br />
Solution: In this case we need to compute<br />
L[sin(at)] = lim<br />
n→∞<br />
n<br />
0<br />
e −st sin(at) dt.<br />
1<br />
.<br />
(s − a) 2<br />
The definite integral in the last line above can be computed as follows: Integrating by parts<br />
twice it is not difficult to check that the following formula holds,<br />
n<br />
0<br />
e −st sin(at) dt = − 1<br />
−st<br />
e sin(at)<br />
s<br />
n <br />
−<br />
0<br />
a<br />
s2 −st<br />
e cos(at) n <br />
−<br />
0<br />
a<br />
s2 which implies that<br />
<br />
1 + a<br />
s2 n<br />
Hence, it is not difficult to see that<br />
2 2 s + a<br />
∞<br />
which is equivalent to<br />
0<br />
e −st sin(at) dt = − 1<br />
−st<br />
e sin(at)<br />
s<br />
n <br />
−<br />
0<br />
a<br />
s2 s 2<br />
0<br />
L[sin(at)] =<br />
e −st sin(at) dt = a<br />
, s > 0,<br />
s2 a<br />
s2 , s > 0.<br />
+ a2 n<br />
0<br />
e −st sin(at) dt,<br />
−st<br />
e cos(at) n <br />
.<br />
We here present a short list <strong>of</strong> Laplace transforms. They can be computed in the same<br />
way we computed the examples above.<br />
0<br />
⊳<br />
⊳
112 G. NAGY – ODE july 2, 2013<br />
f(t)<br />
f(t) = 1<br />
ˆ f(s) = L[f(t)]<br />
ˆ f(s) = 1<br />
s<br />
f(t) = e at ˆ f(s) =<br />
1<br />
(s − a)<br />
f(t) = t n ˆ f(s) = n!<br />
s (n+1)<br />
f(t) = sin(at)<br />
f(t) = cos(at)<br />
f(t) = sinh(at)<br />
f(t) = cosh(at)<br />
ˆ f(s) =<br />
ˆ f(s) =<br />
ˆ f(s) =<br />
ˆ f(s) =<br />
f(t) = t n e at ˆ f(s) =<br />
f(t) = e at sin(bt)<br />
f(t) = e at cos(bt)<br />
ˆ f(s) =<br />
ˆ f(s) =<br />
a<br />
s 2 + a 2<br />
s<br />
s 2 + a 2<br />
a<br />
s 2 − a 2<br />
s<br />
s 2 − a 2<br />
n!<br />
(s − a) (n+1)<br />
b<br />
(s − a) 2 + b 2<br />
(s − a)<br />
(s − a) 2 + b 2<br />
s > 0<br />
s > max{a, 0}<br />
s > 0<br />
s > 0<br />
s > 0<br />
s > |a|<br />
s > |a|<br />
Table 2. List <strong>of</strong> particular Laplace Transforms.<br />
s > max{a, 0}<br />
s > max{a, 0}<br />
s > max{a, 0}<br />
We have seen how to compute the Laplace transform <strong>of</strong> several functions. We now answer<br />
the following question: How big is the set <strong>of</strong> functions that have a well-defined Laplace<br />
transform? It turns out that this is a big set <strong>of</strong> functions, containing all functions that can<br />
be bounded by an exponential. This is the main concept is the following result.<br />
Theorem 4.1.2 (Sufficient conditions). If the function f : R+ → R is piecewise continuous<br />
and there exist positive constants k and a such that<br />
|f(t)| k e at ,<br />
then the Laplace transform <strong>of</strong> f exists for all s > a.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 4.1.2: We show that ˆ f is well-defined, that is, finite, by bounding its<br />
absolute value by a finite number. One can check that the following inequalities hold,<br />
| ˆ <br />
∞<br />
<br />
f(s)| = e −st ∞<br />
<br />
f(t) dt<br />
e −st ∞<br />
|f(t)| dt e −st ke at dt = k L[e at ].<br />
0<br />
Therefore, we can bound | ˆ f(s)| by the function<br />
0<br />
| ˆ f(s)| k L[e at ] = k<br />
s − a<br />
0<br />
s > a.<br />
Since | ˆ f(s)| is bounded by a finite number for s > a, then ˆ f(s) is well-defined. This bound<br />
establishes the Theorem.
G. NAGY – ODE July 2, 2013 113<br />
4.1.3. Properties <strong>of</strong> the Laplace Transform. We conclude this Section presenting three<br />
important properties <strong>of</strong> the Laplace transform, which are summarized in the following three<br />
Theorems. The first one says that the Laplace transform <strong>of</strong> a linear combination <strong>of</strong> functions<br />
is the linear combination <strong>of</strong> their Laplace transforms.<br />
Theorem 4.1.3 (Linear combination). If the Laplace transforms L[f] and L[g] <strong>of</strong> the<br />
functions f and g are well-defined and a, b are constants, then the following equation holds<br />
L[af + bg] = a L[f] + b L[g].<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 4.1.3: Since integration is a linear operation, so is the Laplace transform,<br />
as this calculation shows,<br />
∞<br />
L[af + bg] = e<br />
0<br />
−st af(t) + bg(t) dt<br />
∞<br />
= a e −st ∞<br />
f(t) dt + b e −st g(t) dt<br />
0<br />
= a L[f] + b cL[g].<br />
This establishes the Theorem. <br />
Example 4.1.6: Compute L[3t 2 + 5 cos(4t)].<br />
Solution: From the Theorem above we know that<br />
Therefore,<br />
L[3t 2 + 5 cos(4t)] = 3 L[t 2 ] + 5 L[cos(4t)]<br />
<br />
2<br />
= 3<br />
s3 <br />
4<br />
+ 5<br />
s2 + 42 <br />
= 6 20<br />
+<br />
s3 s2 .<br />
+ 42 L[3t 2 + 5 cos(4t)] = 20s3 + 6s2 + 96<br />
s3 (s2 .<br />
+ 16)<br />
The second Theorem relates L[f ′ ] with L[f]. This result is crucial to use the Laplace<br />
transform to solve <strong>differential</strong> <strong>equations</strong>.<br />
Theorem 4.1.4 (Derivatives). If the L[f] and L[f ′ ] are well-defined, then holds<br />
L[f ′ ] = s L[f] + f(0).<br />
Furthermore, if L[f ′′ ] is well-defined, then it also holds<br />
L[f ′′ ] = s 2 L[f] − s f(0) − f ′ (0).<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 4.1.4: The first formula in the Theorem is proved by an integration<br />
by parts, namely,<br />
L[f ′ ] = lim<br />
n→∞<br />
n<br />
e<br />
0<br />
−st f ′ (t) dt<br />
n<br />
<br />
= lim e<br />
n→∞<br />
−st n <br />
f(t) − (−s)e<br />
0 0<br />
−st <br />
f(t) dt<br />
<br />
= lim e<br />
n→∞<br />
−sn ∞<br />
f(n) − f(0) + s e −st f(t) dt<br />
= −f(0) + s L[f].<br />
0<br />
0<br />
⊳
114 G. NAGY – ODE july 2, 2013<br />
Since L[f] is well-defined, we have used the equation<br />
lim<br />
n→∞ e−snf(n) = 0.<br />
we then conclude that<br />
L[f ′ ] = s L[f] − f(0).<br />
This proves the first part <strong>of</strong> the Theorem. The furthermore part is obtained using the<br />
relation we have just proved, namely<br />
L[f ′′ ] = s L[f ′ ] − f ′ (0)<br />
= s s L[f] − f(0) − f ′ (0)<br />
= s 2 L[f] − s f(0) − f ′ (0).<br />
This establishes the Theorem. <br />
The third and last Theorem <strong>of</strong> this Section is also crucial to use this transform to solve<br />
<strong>differential</strong> <strong>equations</strong>. It says that the map L acting on the set <strong>of</strong> piecewise continuous<br />
functions is one-to-one. This property is essential to define the inverse <strong>of</strong> the Laplace<br />
transform, L −1 . We present this result without the pro<strong>of</strong>.<br />
Theorem 4.1.5 (One-to-one). Given any piecewise continuous functions f and g, it holds<br />
L[f] = L[g] ⇒ f = g.<br />
As we mentioned before, this last result allows to define the inverse Laplace transform:<br />
Given a continuous function with values ˆ f(s), we define L −1 as follows,<br />
L −1 [ ˆ f(s)] = f(t) ⇔ L[f(t)] = ˆ f(s).
4.1.4. Exercises.<br />
4.1.1.- . 4.1.2.- .<br />
G. NAGY – ODE July 2, 2013 115
116 G. NAGY – ODE july 2, 2013<br />
4.2. The initial value problem<br />
4.2.1. Summary <strong>of</strong> the Laplace Transform method. We use the Laplace Transform<br />
to find solutions to second order <strong>differential</strong> <strong>equations</strong>. On the one hand this method works<br />
only with constant coefficients <strong>equations</strong>, which is a restriction. On the other hand, this<br />
method works with very general source functions, including discontinuous step functions<br />
and Dirac’s delta generalized functions. The main idea <strong>of</strong> the method can be summarized<br />
in the following scheme:<br />
L<br />
<br />
<strong>differential</strong> eq.<br />
for y(t).<br />
(1)<br />
−→<br />
Algebraic eq.<br />
for L[y(t)].<br />
(2)<br />
−→<br />
Solve the<br />
algebraic eq.<br />
for L[y(t)].<br />
(a) Compute the Laplace transform <strong>of</strong> the whole <strong>differential</strong> equation;<br />
(b) Use Theorems 4.1.3 and 4.1.4, that is,<br />
L[a f(t) + b g(t)] = a L[f(t)] + b L[g(t)],<br />
(3)<br />
−→<br />
L y (n) = s n L[y] − s (n−1) y(0) − s (n−2) y ′ (0) − · · · − y (n−1) (0),<br />
Transform back<br />
to obtain y(t).<br />
(Use the table.)<br />
to transform the original <strong>differential</strong> equation for the unknown function y into an algebraic<br />
equation for the unknown function L[y];<br />
(c) Solve the algebraic equation for L[y];<br />
(d) Transform back to obtain the solution function y.<br />
We provides several examples to help understand how this method works.<br />
Example 4.2.1: Use the Laplace transform to find the solution y to the initial value problem<br />
y ′ + 2y = 0, y(0) = 3.<br />
Solution: We have seen in Section 1.1 that the solution to this problem is y(t) = 3e −2t .<br />
We now recover this result using Laplace Transforms. First, compute the Laplace Transform<br />
<strong>of</strong> the <strong>differential</strong> equation,<br />
L[y ′ + 2y] = L[0] = 0.<br />
Theorem 4.1.3 says the Laplace Transform is a linear operation, that is,<br />
L[y ′ ] + 2 L[y] = 0.<br />
Theorem 4.1.4 relates derivatives and multiplications, as follows,<br />
<br />
<br />
s L[y] − y(0) + 2 L[y] = 0 ⇒ (s + 2)L[y] = y(0).<br />
In the last equation we have been able to transform the original <strong>differential</strong> equation for y<br />
into an algebraic equation for L[y]. We can solve for the unknown L[y] as follows,<br />
L[y] = y(0)<br />
s + 2<br />
⇒<br />
3<br />
L[y] =<br />
s + 2 ,<br />
where in the last step we introduced the initial condition y(0) = 3. From the list <strong>of</strong> Laplace<br />
transforms given in Sect. 4.1 we know that<br />
L[e at ] = 1<br />
s − a ⇒<br />
3<br />
s + 2 = 3L[e−2t ] ⇒<br />
3<br />
s + 2 = L[3e−2t ].<br />
So we arrive at L[y] = L[3e −2t ]. The Laplace Transform is invertible in the set <strong>of</strong> piecewise<br />
continuous functions, then<br />
y(t) = 3e −2t .
G. NAGY – ODE July 2, 2013 117<br />
Example 4.2.2: Use the Laplace transform to find the solution y to the initial value problem<br />
y ′′ − y ′ − 2y = 0, y(0) = 1, y ′ (0) = 0.<br />
Solution: First, compute the Laplace transform <strong>of</strong> the <strong>differential</strong> equation,<br />
L[y ′′ − y ′ − 2y] = L[0] = 0.<br />
Theorem 4.1.3 says that the Laplace Transform is a linear operation,<br />
L[y ′′ ] − L[y ′ ] − 2 L[y] = 0.<br />
Then, Theorem 4.1.4 relates derivatives and multiplications,<br />
<br />
s 2 L[y] − s y(0) − y ′ <br />
<br />
(0) − s L[y] − y(0) − 2 L[y] = 0,<br />
which is equivalent to the equation<br />
(s 2 − s − 2) L[y] = (s − 1) y(0) + y ′ (0).<br />
Once again we have transformed the original <strong>differential</strong> equation for y into an algebraic<br />
equation for L[y]. Introduce the initial condition into the last equation above, that is,<br />
Solve for the unknown L[y] as follows,<br />
(s 2 − s − 2) L[y] = (s − 1).<br />
L[y] =<br />
(s − 1)<br />
(s 2 − s − 2) .<br />
We now use partial fraction methods to find a function whose Laplace transform is the righthand<br />
side <strong>of</strong> the equation above. First find the roots <strong>of</strong> the polynomial in the denominator,<br />
s 2 − s − 2 = 0 ⇒ s± = 1<br />
<br />
√ s+ = 2,<br />
1 ± 1 + 8 ⇒<br />
2<br />
s− = −1,<br />
that is, the polynomial has two real roots. In this case we factorize the denominator,<br />
L[y] =<br />
(s − 1)<br />
(s − 2)(s + 1) .<br />
The partial fraction decomposition <strong>of</strong> the right-hand side in the equation above is the following:<br />
Find constants a and b such that<br />
A simple calculation shows<br />
(s − 1) a b<br />
= +<br />
(s − 2)(s + 1) s − 2 s + 1 .<br />
(s − 1) a b a(s + 1) + b(s − 2)<br />
= + =<br />
(s − 2)(s + 1) s − 2 s + 1 (s − 2)(s + 1)<br />
Hence constants a and b must be solutions <strong>of</strong> the <strong>equations</strong><br />
(s − 1) = s(a + b) + (a − 2b) ⇒<br />
<br />
a + b = 1,<br />
a − 2b = −1.<br />
The solution is a = 1<br />
3<br />
2<br />
and b = . Hence,<br />
3<br />
L[y] = 1<br />
3<br />
1 2<br />
+<br />
(s − 2) 3<br />
1<br />
(s + 1) .<br />
s(a + b) + (a − 2b)<br />
= .<br />
(s − 2)(s + 1)<br />
⊳
118 G. NAGY – ODE july 2, 2013<br />
From the list <strong>of</strong> Laplace transforms given in Sect. 4.1 we know that<br />
L[e at ] = 1<br />
s − a ⇒<br />
1<br />
s − 2 = L[e2t ],<br />
1<br />
s + 1 = L[e−t ].<br />
So we arrive at the equation<br />
L[y] = 1<br />
3 L[e2t ] + 2<br />
3 L[e−t <br />
1<br />
2t −t<br />
] = L e + 2e<br />
3<br />
<br />
We conclude that<br />
y(t) = 1<br />
2t −t<br />
e + 2e<br />
3<br />
.<br />
Example 4.2.3: Use the Laplace transform to find the solution y to the initial value problem<br />
y ′′ − 4y ′ + 4y = 0, y(0) = 1, y ′ (0) = 1.<br />
Solution: First, compute the Laplace transform <strong>of</strong> the <strong>differential</strong> equation,<br />
L[y ′′ − 4y ′ + 4y] = L[0] = 0.<br />
Theorem 4.1.3 says that the Laplace Transform is a linear operation,<br />
L[y ′′ ] − 4 L[y ′ ] + 4 L[y] = 0.<br />
Theorem 4.1.4 relates derivatives with multiplication,<br />
<br />
s 2 L[y] − s y(0) − y ′ <br />
<br />
(0) − 4 s L[y] − y(0) + 4 L[y] = 0,<br />
which is equivalent to the equation<br />
(s 2 − 4s + 4) L[y] = (s − 4) y(0) + y ′ (0).<br />
Introduce the initial conditions y(0) = 1 and y ′ (0) = 1 into the equation above,<br />
Solve the algebraic equation for L[y],<br />
(s 2 − 4s + 4) L[y] = s − 3.<br />
L[y] =<br />
(s − 3)<br />
(s 2 − 4s + 4) .<br />
We now want to find a function y whose Laplace transform is the right-hand side in the<br />
equation above. In order to see if partial fraction methods will be needed, we now find the<br />
roots <strong>of</strong> the polynomial in the denominator,<br />
s 2 − 4s + 4 = 0 ⇒ s± = 1<br />
√ <br />
4 ± 16 − 16 ⇒ s+ = s− = 2.<br />
2<br />
that is, the polynomial has a single real root, so L[y] can be written as<br />
L[y] =<br />
(s − 3)<br />
.<br />
(s − 2) 2<br />
This expression is already in the partial fraction decomposition. We now rewrite the righthand<br />
side <strong>of</strong> the equation above in a way it is simple to use the Laplace Transform table in<br />
Section 4.1,<br />
L[y] =<br />
(s − 2) + 2 − 3<br />
(s − 2) 2<br />
= (s − 2)<br />
−<br />
(s − 2) 2<br />
1<br />
1<br />
⇒ L[y] =<br />
(s − 2) 2 s − 2 −<br />
From the list <strong>of</strong> Laplace Transforms given in Sect. 4.1 we know that<br />
L[e at ] = 1<br />
s − a ⇒<br />
1<br />
s − 2 = L[e2t ],<br />
1<br />
.<br />
(s − 2) 2<br />
⊳
L[te at ] =<br />
So we arrive at the equation<br />
G. NAGY – ODE July 2, 2013 119<br />
1<br />
(s − a) 2<br />
⇒<br />
1<br />
(s − 2) 2 = L[te2t ].<br />
L[y] = L[e 2t ] − L[te 2t ] = L e 2t − te 2t ⇒ y(t) = e 2t − te 2t .<br />
Example 4.2.4: Use the Laplace transform to find the solution y to the initial value problem<br />
y ′′ − 4y ′ + 4y = 3 sin(2t), y(0) = 1, y ′ (0) = 1.<br />
Solution: First, compute the Laplace transform <strong>of</strong> the <strong>differential</strong> equation,<br />
L[y ′′ − 4y ′ + 4y] = L[3 sin(2t)].<br />
The right-hand side above can be expressed as follows,<br />
L[3 sin(2t)] = 3 L[sin(2t)] = 3<br />
2<br />
s2 6<br />
=<br />
+ 22 s2 + 4 .<br />
Theorem 4.1.3 says that the Laplace Transform is a linear operation,<br />
L[y ′′ ] − 4 L[y ′ ] + 4 L[y] = 6<br />
s 2 + 4 ,<br />
and Theorem 4.1.4 relates derivatives with multiplications,<br />
<br />
s 2 L[y] − s y(0) − y ′ <br />
<br />
(0) − 4 s L[y] − y(0) + 4 L[y] = 6<br />
s2 + 4 .<br />
Re-order terms,<br />
(s 2 − 4s + 4) L[y] = (s − 4) y(0) + y ′ (0) + 6<br />
s 2 + 4 .<br />
Introduce the initial conditions y(0) = 1 and y ′ (0) = 1,<br />
Solve this algebraic equation for L[y], that is,<br />
L[y] =<br />
(s 2 − 4s + 4) L[y] = s − 3 + 6<br />
s 2 + 4 .<br />
(s − 3)<br />
(s2 − 4s + 4) +<br />
6<br />
(s2 − 4 + 4)(s2 + 4) .<br />
From the Example above we know that s 2 − 4s + 4 = (s − 2) 2 , so we obtain<br />
L[y] = 1<br />
s − 2 −<br />
From the previous example we know that<br />
1<br />
+<br />
(s − 2) 2<br />
6<br />
(s − 2) 2 (s2 . (4.2.1)<br />
+ 4)<br />
L[e 2t − te 2t ] = 1<br />
s − 2 −<br />
1<br />
. (4.2.2)<br />
(s − 2) 2<br />
We know use partial fractions to simplify the third term on the right-hand side <strong>of</strong> Eq. (4.2.1).<br />
The appropriate partial fraction decomposition for this term is the following: Find constants<br />
a, b, c, d, such that<br />
6<br />
(s − 2) 2 (s2 as + b<br />
=<br />
+ 4) s2 + 4 +<br />
c<br />
(s − 2) +<br />
d<br />
(s − 2) 2<br />
⊳
120 G. NAGY – ODE july 2, 2013<br />
Take common denominator on the right-hand side above, and one obtains the system<br />
a + c = 0,<br />
−4a + b − 2c + d = 0,<br />
4a − 4b + 4c = 0,<br />
4b − 8c + 4d = 6.<br />
The solution for this linear system <strong>of</strong> <strong>equations</strong> is the following:<br />
a = 3<br />
,<br />
8<br />
b = 0, c = −3 ,<br />
8<br />
3<br />
d =<br />
4 .<br />
Therefore,<br />
6<br />
(s − 2) 2 (s2 3 s<br />
=<br />
+ 4) 8 s2 3 1 3 1<br />
− +<br />
+ 4 8 (s − 2) 4 (s − 2) 2<br />
We can rewrite this expression above in terms <strong>of</strong> the Laplace transforms given on the list in<br />
Sect. 4.1, as follows,<br />
6<br />
(s − 2) 2 (s2 3<br />
3<br />
= L[cos(2t)] −<br />
+ 4) 8 8 L[e2t ] + 3<br />
4 L[te2t ],<br />
and using the linearity <strong>of</strong> the Laplace Transform,<br />
6<br />
(s − 2) 2 (s2 <br />
3 3<br />
= L cos(2t) −<br />
+ 4) 8 8 e2t + 3<br />
4 te2t.<br />
(4.2.3)<br />
Finally, introducing Eqs. (4.2.2) and (4.2.3) into Eq. (4.2.1) we obtain<br />
L[y(t)] = L (1 − t) e 2t + 3<br />
8 (−1 + 2t) e2t + 3<br />
8 cos(2t)<br />
<br />
.<br />
Since the Laplace Transform is an invertible transformation, we conclude that<br />
y(t) = (1 − t) e 2t + 3<br />
8 (2t − 1) e2t + 3<br />
8 cos(2t).<br />
⊳
4.2.2. Exercises.<br />
4.2.1.- . 4.2.2.- .<br />
G. NAGY – ODE July 2, 2013 121
122 G. NAGY – ODE july 2, 2013<br />
4.3. Discontinuous sources<br />
We have mentioned that the Laplace Transform method is useful to solve <strong>differential</strong> <strong>equations</strong><br />
with discontinuous source functions. In this Section review the simplest discontinuous<br />
function, the step function, and we use it to construct more general piecewise continuous<br />
functions. We then compute the Laplace Transform <strong>of</strong> these discontinuous functions. We<br />
also establish useful formulas for the Laplace Transform <strong>of</strong> certain function translations.<br />
These formulas and the Laplace Transform table in Section 4.1 are the key to solve a large<br />
set <strong>of</strong> <strong>differential</strong> <strong>equations</strong> with discontinuous sources.<br />
4.3.1. The Step function. We start with a definition <strong>of</strong> a step function.<br />
Definition 4.3.1. A function u is called a step function at t = 0 iff holds<br />
<br />
0 t < 0,<br />
u(t) =<br />
1 t 0.<br />
The step function u at t = 0 and its right and left translations are plotted in Fig. 14.<br />
u(t)<br />
1<br />
0<br />
t<br />
u(t − c)<br />
1<br />
0 c<br />
t<br />
− c<br />
u(t + c)<br />
Figure 14. The graph <strong>of</strong> the step function given in Eq. (4.3.1), a right an<br />
a left translation by a constant c > 0, respectively, <strong>of</strong> this step function.<br />
1<br />
0<br />
(4.3.1)<br />
Recall that given a function with values f(t) and a positive constant c, then f(t − c) and<br />
f(t + c) are the function values <strong>of</strong> the right translation and the left translation, respectively,<br />
<strong>of</strong> the original function f. In Fig. 15 we plot the graph <strong>of</strong> functions f(t) = e at , g(t) = u(t) e at<br />
and their respective right translations by c ∈ R.<br />
a t<br />
a ( t − c )<br />
f ( t ) f ( t ) = e<br />
f ( t ) f ( t ) = e<br />
1<br />
0 t<br />
1<br />
0 c<br />
t<br />
a t<br />
a ( t − c )<br />
f ( t ) f ( t ) = u ( t ) e f ( t ) f ( t ) = u ( t − c ) e<br />
1<br />
0 t<br />
1<br />
0 c<br />
t<br />
Figure 15. An exponential function, its right translation, the function<br />
f(t) = u(t) e at and its right translation.<br />
Right and left translations <strong>of</strong> step functions are useful to construct bump functions. A<br />
bump function is a function with nonzero values only on a finite interval [a, b]. The simplest<br />
bump function is a box function, b, defined as follows,<br />
⎧<br />
⎪⎨<br />
0 t < a,<br />
b(t) = u(t − a) − u(t − b) ⇔ b(t) = 1<br />
⎪⎩<br />
0<br />
a t < b<br />
t b.<br />
(4.3.2)<br />
In our first Example we plot the graph <strong>of</strong> a step function using the first expression above,<br />
that is, as a combination <strong>of</strong> step functions.<br />
t
u(t −a)<br />
1<br />
0 a b<br />
t<br />
G. NAGY – ODE July 2, 2013 123<br />
u(t −b)<br />
1<br />
0 a b<br />
b(t)<br />
1<br />
t 0 a b t<br />
Figure 16. A bump function b constructed with translated step functions.<br />
Example 4.3.1: Graph <strong>of</strong> the function b(t) = u(t − a) − u(t − b), with 0 < a < b.<br />
Solution: The bump function b can be graphed as follows:<br />
Step and bump functions are useful to construct more general piecewise continuous functions.<br />
Example 4.3.2: Graph <strong>of</strong> the function f(t) = e at u(t − 1) − u(t − 2) .<br />
Solution:<br />
y<br />
1<br />
a t<br />
f ( t ) = e [ u ( t −1 ) − u ( t −2 ) ]<br />
1 2<br />
e<br />
a t<br />
[ u ( t −1 ) − u ( t −2 ) ]<br />
4.3.2. The Laplace Transform <strong>of</strong> Step functions. We are interested to find the Laplace<br />
transform <strong>of</strong> the step functions. The following statement says they are not difficult to<br />
compute.<br />
Theorem 4.3.2. For every real number c and s > 0 holds L[u(t − c)] = e−cs<br />
s .<br />
Pro<strong>of</strong> <strong>of</strong> Lemma 4.3.2: By the definition <strong>of</strong> the Laplace Transform,<br />
∞<br />
L[u(t − c)] = e −st ∞<br />
u(t − c) dt = e −st dt,<br />
0<br />
where we used that the step function vanishes for t < c. Now compute the improper integral,<br />
e −Ns − e −cs = e−cs<br />
L[u(t − c)] = lim<br />
N→∞ −1<br />
s<br />
s<br />
⇒ L[u(t − c)] = e−cs<br />
s<br />
This establishes the Theorem. <br />
Example 4.3.3: Compute L[3u(t − 2)].<br />
Solution: The Laplace Transform is a linear operation, so L[3u(t − 2)] = 3 L[u(t − 2)], and<br />
the Theorem above implies that L[3u(t − 2)] = 3e−2s<br />
. ⊳<br />
s<br />
Example 4.3.4: Compute L −1 e3s <br />
.<br />
s<br />
Solution: Since L −1 e3s <br />
= L<br />
s<br />
−1 e−(−3)s <br />
= u(t − (−3)), then, L<br />
s<br />
−1 e3s <br />
= u(t + 3). ⊳<br />
s<br />
c<br />
t<br />
⊳
124 G. NAGY – ODE july 2, 2013<br />
4.3.3. The Laplace Transform <strong>of</strong> translations. We now introduce two important properties<br />
<strong>of</strong> the Laplace transform.<br />
Theorem 4.3.3 (Translations). If ˆ f(s) = L[f(t)] exists for s > a 0 and c 0, then<br />
both <strong>equations</strong> hold,<br />
L[u(t − c)f(t − c)] = e −cs L[f(t)], s > a, (4.3.3)<br />
L[e ct f(t)] = L[f(t)](s − c), s > a + c. (4.3.4)<br />
We can summarize the relations in the Theorem above as follows,<br />
L translation (uf) = (exp) L[f] ,<br />
L (exp) (f) = translation L[f] .<br />
Denoting ˆ f(s) = L[f(t)] we have the following equivalent expression for Eqs. (4.3.3)-(4.3.4),<br />
L[u(t − c)f(t − c)] = e −cs ˆ f(s),<br />
L[e ct f(t)] = ˆ f(s − c).<br />
The inverse form <strong>of</strong> Eqs. (4.3.3)-(4.3.4) is given by,<br />
L −1 [e −cs ˆ f(s)] = u(t − c)f(t − c), (4.3.5)<br />
L −1 [ ˆ f(s − c)] = e ct f(t). (4.3.6)<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 4.3.3: The pro<strong>of</strong> is again based in a change <strong>of</strong> the integration variable.<br />
We start with Eq. (4.3.3), as follows,<br />
∞<br />
L[u(t − c)f(t − c)] = e<br />
0<br />
−st u(t − c)f(t − c) dt<br />
∞<br />
= e<br />
c<br />
−st f(t − c) dt, τ = t − c, dτ = dt<br />
∞<br />
= e<br />
0<br />
−s(τ+c) f(τ) dτ<br />
= e −cs<br />
∞<br />
e −sτ f(τ) dτ<br />
0<br />
= e −cs L[f(t)], s > a.<br />
The pro<strong>of</strong> <strong>of</strong> Eq. (4.3.4) is even simpler,<br />
L e ct f(t) ∞<br />
= e<br />
0<br />
−st e ct f(t) dt<br />
∞<br />
= e −(s−c)t f(t) dt<br />
0<br />
= L[f(t)](s − c), s − c > a.<br />
This establishes the Theorem. <br />
Example 4.3.5: Compute L u(t − 2) sin(a(t − 2)) .<br />
a<br />
Solution: Both L[sin(at)] =<br />
s2 + a2 and L[u(t − c)f(t − c)] = e−cs L[f(t)] imply<br />
L u(t − 2) sin(a(t − 2)) = e −2s L[sin(at)] = e −2s<br />
We conclude: L u(t − 2) sin(a(t − 2)) =<br />
a<br />
s2 .<br />
+ a2 a e−2s<br />
s2 . ⊳<br />
+ a2
Example 4.3.6: Compute L e 3t sin(at) .<br />
G. NAGY – ODE July 2, 2013 125<br />
Solution: Since L[ectf(t)] = L[f](s − c), then we get<br />
L e 3t sin(at) a<br />
=<br />
(s − 3) 2 , s > 3.<br />
+ a2 Example 4.3.7: Compute both L u(t − 2) cos a(t − 2) and L e3t cos(at) .<br />
Solution: Since L cos(at) s<br />
=<br />
s2 , then<br />
+ a2 L u(t − 2) cos a(t − 2) = e −2s<br />
Example 4.3.8: Find the Laplace transform <strong>of</strong> the function<br />
<br />
0 t < 1,<br />
f(t) =<br />
(t2 − 2t + 2) t 1.<br />
s<br />
(s2 + a2 ) , Le 3t cos(at) (s − 3)<br />
=<br />
(s − 3) 2 .<br />
+ a2 ⊳<br />
⊳<br />
(4.3.7)<br />
Solution: The idea is to re-write function f so we can use the Laplace Transform table in<br />
Section 4.1 to compute its Laplace Transform. Since the function f vanishes for all t < 1,<br />
we use step functions to write f as<br />
f(t) = u(t − 1)(t 2 − 2t + 2).<br />
Now, notice that completing the square we obtain,<br />
t 2 − 2t + 2 = (t 2 − 2t + 1) − 1 + 2 = (t − 1) 2 + 1.<br />
That is, the polynomial is a parabola t 2 translated to the right by 1 and up by one. This is<br />
a discontinuous functions, as it can be seen in Fig. 17.<br />
f ( t )<br />
1<br />
0<br />
1<br />
Figure 17. The graph <strong>of</strong> the function f given in Eq. (4.3.7).<br />
So the function f can be written as follows,<br />
f(t) = u(t − 1) (t − 1) 2 + u(t − 1).<br />
Since we know that L[t 2 ] = 2<br />
, then Eq. (4.3.3) implies<br />
s3 L[f(t)] = L[u(t − 1) (t − 1) 2 ] + L[u(t − 1)]<br />
−s 2<br />
= e<br />
s<br />
s<br />
3 + e−s 1<br />
t<br />
e−s<br />
⇒ L[f(t)] =<br />
s3 2<br />
2 + s .<br />
⊳
126 G. NAGY – ODE july 2, 2013<br />
Example 4.3.9: Find the function f such that L[f(t)] = e−4s<br />
s 2 + 5 .<br />
Solution: Notice that<br />
L[f(t)] = e −4s<br />
Recall that L[sin(at)] =<br />
1<br />
s 2 + 5<br />
⇒ L[f(t)] = 1<br />
√ 5 e −4s<br />
√ 5<br />
s 2 + √ 5 2 .<br />
a<br />
(s2 + a2 , then Eq. (4.3.3), or its inverse form Eq. (4.3.5) imply<br />
)<br />
L[f(t)] = 1<br />
√ 5 L u(t − 4) sin √ 5 (t − 4) ⇒ f(t) = 1<br />
√ 5 u(t − 4) sin √ 5 (t − 4) .<br />
Example 4.3.10: Find the function f(t) such that L[f(t)] =<br />
Solution: We first rewrite the right-hand side above as follows,<br />
L[f(t)] =<br />
= (s − 2)<br />
(s − 1 − 1 + 1)<br />
(s − 2) 2 + 3<br />
(s − 2) 2 + 3 +<br />
(s − 2)<br />
=<br />
(s − 2) 2 + √ 3 2 1<br />
(s − 2) 2 + 3<br />
+ 1<br />
√ 3<br />
(s − 1)<br />
(s − 2) 2 + 3 .<br />
√ 3<br />
(s − 2) 2 + √ 3 2 .<br />
We now recall Eq. (4.3.4) or its inverse form Eq. (4.3.6), which imply<br />
So, we conclude that<br />
Example 4.3.11: Find L −1 2e−3s s2 <br />
.<br />
− 4<br />
Solution: Since L −1 a<br />
s2 − a2 <br />
L −1 2e−3s s2 <br />
= L<br />
− 4<br />
−1<br />
e −3s 2<br />
s2 <br />
− 4<br />
L[f(t)] = L e 2t cos √ 3 t + 1<br />
√ 3 L e 2t sin √ 3 t .<br />
f(t) = e2t<br />
√ √ √ <br />
√ 3 cos 3 t + sin 3 t<br />
3<br />
<br />
.<br />
Example 4.3.12: Find a function f such that L[f(t)] =<br />
= sinh(at) and L −1 e −cs ˆ f(s) = u(t − c) f(t − c), then<br />
⇒ L −1 2e−3s s2 <br />
= u(t − 3) sinh<br />
− 4<br />
2(t − 3) .<br />
e −2s<br />
s 2 + s − 2 .<br />
Solution: Since the right-hand side above does not appear in the Laplace Transform table<br />
in Section 4.1, we need to simplify it in the appropriate way. The plan is to rewrite the<br />
denominator <strong>of</strong> the rational function 1/(s 2 +s−2), so we can use the partial fraction method<br />
to simplify this rational function. We first find out whether this denominator has real or<br />
complex roots:<br />
s± = 1<br />
√ <br />
−1 ± 1 + 8 ⇒<br />
2<br />
s+ = 1,<br />
s− = −2.<br />
⊳<br />
⊳<br />
⊳
We are in the case <strong>of</strong> real roots, so we can rewrite<br />
G. NAGY – ODE July 2, 2013 127<br />
s 2 + s − 2 = (s − 1) (s + 2).<br />
The partial fraction decomposition in this case is given by<br />
1<br />
(s − 1) (s + 2) =<br />
a<br />
(s − 1) +<br />
b (a + b) s + (2a − b)<br />
=<br />
(s + 2) (s − 1) (s + 2)<br />
The solution is a = 1/3 and b = −1/3, so we arrive to the expression<br />
Recalling that<br />
L[f(t)] = 1<br />
3 e−2s 1 1<br />
−<br />
s − 1 3 e−2s 1<br />
s + 2 .<br />
L[e at ] = 1<br />
s − a ,<br />
and Eq. (4.3.3) we obtain the equation<br />
L[f(t)] = 1<br />
3 L u(t − 2) e (t−2) − 1<br />
3 L u(t − 2) e −2(t−2)<br />
which leads to the conclusion:<br />
f(t) = 1<br />
<br />
u(t − 2)<br />
3<br />
e (t−2) − e −2(t−2)<br />
.<br />
⇒<br />
a + b = 0,<br />
2a − b = 1.<br />
4.3.4. Differential <strong>equations</strong> with discontinuous sources. The last three examples in<br />
this Section show how to use the methods presented above to solve <strong>differential</strong> <strong>equations</strong><br />
with discontinuous source functions.<br />
Example 4.3.13: Use the Laplace Transform to find the solution <strong>of</strong> the initial value problem<br />
y ′ + 2y = u(t − 4), y(0) = 3.<br />
Solution: We compute the Laplace transform <strong>of</strong> the whole equation,<br />
L[y ′ ] + 2 L[y] = L[u(t − 4)] = e−4s<br />
s .<br />
From the previous Section we know that<br />
s L[y] − y(0) + 2 L[y] = e −4s<br />
s<br />
⇒ (s + 2) L[y] = y(0) + e−4s<br />
s .<br />
We introduce the initial condition y(0) = 3 into equation above,<br />
L[y] =<br />
3<br />
1<br />
+ e−4s<br />
(s + 2)<br />
s(s + 2) ⇒ L[y] = 3 Le −2t + e −4s 1<br />
s(s + 2) .<br />
We need to invert the Laplace transform on the last term on the right-hand side in equation<br />
above. We use the partial fraction decomposition on the rational function above, as follows<br />
1 a<br />
=<br />
s(s + 2) s +<br />
<br />
b a(s + 2) + bs (a + b) s + (2a) a + b = 0,<br />
= = ⇒<br />
(s + 2) s(s + 2) s(s + 2)<br />
2a = 1.<br />
We conclude that a = 1/2 and b = −1/2, so<br />
1 1<br />
=<br />
s(s + 2) 2<br />
1<br />
s −<br />
1<br />
<br />
.<br />
(s + 2)<br />
⊳
128 G. NAGY – ODE july 2, 2013<br />
We then obtain<br />
L[y] = 3 L e −2t + 1<br />
<br />
−4s 1<br />
e<br />
2 s<br />
= 3 L e −2t + 1<br />
2<br />
Hence, we conclude that<br />
− e−4s<br />
1<br />
<br />
(s + 2)<br />
<br />
L[u(t − 4)] − L u(t − 4) e −2(t−4)<br />
.<br />
y(t) = 3e −2t + 1<br />
<br />
u(t − 4) 1 − e<br />
2 −2(t−4)<br />
.<br />
Example 4.3.14: Use the Laplace transform to find the solution to the initial value problem<br />
y ′′ + y ′ + 5<br />
4 y = b(t), y(0) = 0, y′ (0) = 0, b(t) =<br />
1 0 t < π<br />
0 t π.<br />
Solution: From Fig. 18, the source function g can be written as the following product,<br />
u ( t )<br />
1<br />
0 pi<br />
t<br />
b(t) = u(t) − u(t − π) .<br />
u ( t − pi )<br />
1<br />
0 pi t<br />
b ( t )<br />
1<br />
0 pi t<br />
Figure 18. The graph <strong>of</strong> the u, its translation and b as given in Eq. (4.3.8).<br />
The last expression for b is particularly useful to find its Laplace transform,<br />
L[b(t)] = L[u(t)] − L[u(t − π)] = 1 1<br />
+ e−πs<br />
s s ⇒ L[b(t)] = (1 − e−πs ) 1<br />
s .<br />
Now Laplace transform the whole equation,<br />
L[y] = L[b].<br />
4<br />
Since the initial condition are y(0) = 0 and y ′ (0) = 0, we obtain<br />
<br />
s 2 + s + 5<br />
<br />
L[y] =<br />
4<br />
1 − e −πs 1<br />
s ⇒ L[y] = 1 − e −πs 1<br />
<br />
s s2 + s + 5<br />
.<br />
4<br />
Introduce the function<br />
1<br />
H(s) = <br />
s s2 + s + 5<br />
4<br />
L[y ′′ ] + L[y ′ ] + 5<br />
⇒ y(t) = L −1 [H(s)] − L −1 [e −πs H(s)].<br />
⊳<br />
(4.3.8)<br />
That is, we only need to find the inverse Laplace Transform <strong>of</strong> H. We use the partial fraction<br />
method to simplify the expression <strong>of</strong> H. We first find out whether the denominator has real<br />
or complex roots:<br />
s 2 + s + 5<br />
4 = 0 ⇒ s± = 1<br />
√ <br />
−1 ± 1 − 5 ,<br />
2<br />
so the roots are complex-valued. The appropriate partial fraction decomposition is<br />
H(s) =<br />
1<br />
s s2 + s + 5<br />
=<br />
4<br />
a (bs + c)<br />
+ <br />
s s2 5 + s + 4
G. NAGY – ODE July 2, 2013 129<br />
Therefore, we get<br />
<br />
1 = a s 2 + s + 5<br />
<br />
+ s (bs + c) = (a + b) s<br />
4<br />
2 + (a + c) s + 5<br />
4 a.<br />
This equation implies that a, b, and c, satisfy the <strong>equations</strong><br />
The solution is, a = 4<br />
5<br />
, b = −4<br />
5<br />
a + b = 0, a + c = 0,<br />
5<br />
a = 1.<br />
4<br />
, c = −4 . Hence, we have found that,<br />
5<br />
1<br />
H(s) = <br />
s2 + s + 5<br />
=<br />
4 s<br />
4<br />
<br />
1<br />
5 s −<br />
(s + 1)<br />
<br />
<br />
s2 5 + s + 4<br />
Complete the square in the denominator,<br />
s 2 + s + 5<br />
4 =<br />
<br />
s 2 <br />
1<br />
<br />
+ 2 s +<br />
2<br />
1<br />
<br />
−<br />
4<br />
1 5<br />
+<br />
4 4 =<br />
<br />
s + 1<br />
2<br />
Replace this expression in the definition <strong>of</strong> H, that is,<br />
H(s) = 4<br />
<br />
1<br />
5 s −<br />
(s + 1)<br />
<br />
<br />
1 2 <br />
s + 2 + 1<br />
Rewrite the polynomial in the numerator,<br />
(s + 1) =<br />
<br />
s + 1<br />
2<br />
1<br />
<br />
+ =<br />
2<br />
s + 1<br />
2<br />
<br />
+ 1<br />
2 ,<br />
hence we get<br />
H(s) = 4<br />
<br />
1<br />
5 s −<br />
<br />
s + 1<br />
<br />
2<br />
<br />
1 2 −<br />
s + 2 + 1 1 1<br />
<br />
<br />
2 1 2 .<br />
s + 2 + 1<br />
Use the Laplace Transform table to get H(s) equal to<br />
H(s) = 4<br />
<br />
L[1] − L<br />
5<br />
e −t/2 cos(t) − 1<br />
2 L[e−t/2 <br />
sin(t)] ,<br />
equivalently<br />
Denote:<br />
<br />
4<br />
<br />
H(s) = L 1 − e<br />
5<br />
−t/2 cos(t) − 1<br />
2 e−t/2 <br />
sin(t) .<br />
2<br />
+ 1.<br />
h(t) = 4<br />
<br />
1 − e<br />
5<br />
−t/2 cos(t) − 1<br />
2 e−t/2 <br />
sin(t) . ⇒ H(s) = L[h(t)].<br />
Recalling L[y(t)] = H(s) + e −πs H(s), we obtain L[y(t)] = L[h(t)] + e −πs L[h(t)], that is,<br />
y(t) = h(t) + u(t − π)h(t − π).<br />
Example 4.3.15: Use the Laplace transform to find the solution to the initial value problem<br />
y ′′ + y ′ + 5<br />
4 y = g(t), y(0) = 0, y′ (0) = 0, g(t) =<br />
sin(t) 0 t < π<br />
0 t π.<br />
Solution: From Fig. 19, the source function g can be written as the following product,<br />
g(t) = u(t) − u(t − π) sin(t),<br />
⊳<br />
(4.3.9)
130 G. NAGY – ODE july 2, 2013<br />
since u(t) − u(t − π) is a box function, taking value one in the interval [0, π] and zero on<br />
the complement. Finally, notice that the equation sin(t) = − sin(t − π) implies that the<br />
function g can be expressed as follows,<br />
g(t) = u(t) sin(t) − u(t − π) sin(t) ⇒ g(t) = u(t) sin(t) + u(t − π) sin(t − π).<br />
The last expression for g is particularly useful to find its Laplace transform,<br />
y<br />
1<br />
sin ( t )<br />
0 pi t<br />
y<br />
1<br />
0<br />
u ( t ) − u ( t − pi )<br />
pi t<br />
y<br />
1<br />
g ( t )<br />
0 pi<br />
t<br />
Figure 19. The graph <strong>of</strong> the sine function, a square function u(t)−u(t−π)<br />
and the source function g given in Eq. (4.3.9).<br />
L[g(t)] =<br />
1<br />
(s2 1<br />
+ e−πs<br />
+ 1)<br />
(s 2 + 1) .<br />
Having this last transform is then not difficult to solve the <strong>differential</strong> equation. As usual,<br />
Laplace transform the whole equation,<br />
L[y] = L[g].<br />
4<br />
Since the initial condition are y(0) = 0 and y ′ (0) = 0, we obtain<br />
<br />
s 2 + s + 5<br />
<br />
L[y] =<br />
4<br />
1 + e −πs 1<br />
Introduce the function<br />
1<br />
H(s) = <br />
s2 + s + 5<br />
<br />
4 (s2 + 1)<br />
L[y ′′ ] + L[y ′ ] + 5<br />
(s2 + 1) ⇒ L[y] = 1 + e −πs 1<br />
<br />
s 2 + s + 5<br />
⇒ y(t) = L −1 [H(s)] + L −1 [e −πs H(s)].<br />
4<br />
<br />
(s2 .<br />
+ 1)<br />
That is, we only need to find the inverse Laplace Transform <strong>of</strong> H. We use the partial fraction<br />
method to simplify the expression <strong>of</strong> H. We first find out whether the denominator has real<br />
or complex roots:<br />
s 2 + s + 5<br />
4 = 0 ⇒ s± = 1<br />
√ <br />
−1 ± 1 − 5 ,<br />
2<br />
so the roots are complex-valued. The appropriate partial fraction decomposition is<br />
1<br />
(as + b) (cs + d)<br />
H(s) = <br />
s2 5 = <br />
+ s + 4 (s2 + 1) s2 5 +<br />
+ s + (s 4<br />
2 + 1) .<br />
Therefore, we get<br />
1 = (as + b)(s 2 <br />
+ 1) + (cs + d) s 2 + s + 5<br />
<br />
,<br />
4<br />
equivalently,<br />
1 = (a + c) s 3 + (b + c + d) s 2 <br />
+ a + 5<br />
<br />
c + d s + b +<br />
4 5<br />
4 d<br />
<br />
.<br />
This equation implies that a, b, c, and d, are solutions <strong>of</strong><br />
a + c = 0, b + c + d = 0, a + 5<br />
c + d = 0,<br />
4<br />
5<br />
b + d = 1.<br />
4
Here is the solution to this system:<br />
a = 16<br />
17<br />
G. NAGY – ODE July 2, 2013 131<br />
, b = 12<br />
17<br />
4<br />
, c = −16,<br />
d =<br />
17 17 .<br />
We have found that,<br />
H(s) = 4<br />
<br />
(4s + 3) (−4s + 1)<br />
<br />
17 s2 5 +<br />
+ s + (s 4<br />
2 <br />
.<br />
+ 1)<br />
Complete the square in the denominator,<br />
s 2 + s + 5<br />
4 =<br />
<br />
s 2 <br />
1<br />
<br />
+ 2 s +<br />
2<br />
1<br />
<br />
−<br />
4<br />
1 5<br />
+<br />
4 4 =<br />
<br />
s + 1<br />
2<br />
H(s) = 4<br />
<br />
(4s + 3) (−4s + 1)<br />
<br />
17<br />
1 2 +<br />
s + 2 + 1 (s2 <br />
.<br />
+ 1)<br />
Rewrite the polynomial in the numerator,<br />
<br />
(4s + 3) = 4 s + 1 1<br />
<br />
− + 3 = 4 s +<br />
2 2<br />
1<br />
<br />
+ 1,<br />
2<br />
hence we get<br />
H(s) = 4<br />
<br />
<br />
1 s + 2<br />
1<br />
s<br />
4 <br />
17<br />
1 2 + <br />
1 2 − 4<br />
s + 2 + 1 s + 2 + 1 (s2 + 1) +<br />
1<br />
(s2 <br />
.<br />
+ 1)<br />
Use the Laplace Transform table to get H(s) equal to<br />
H(s) = 4<br />
<br />
4 L<br />
17<br />
e −t/2 cos(t) + L e −t/2 sin(t) <br />
− 4 L[cos(t)] + L[sin(t)] ,<br />
equivalently<br />
Denote:<br />
h(t) = 4<br />
17<br />
2<br />
+ 1.<br />
<br />
4<br />
<br />
H(s) = L 4e<br />
17<br />
−t/2 cos(t) + e −t/2 <br />
sin(t) − 4 cos(t) + sin(t) .<br />
<br />
4e −t/2 cos(t) + e −t/2 <br />
sin(t) − 4 cos(t) + sin(t)<br />
⇒ H(s) = L[h(t)].<br />
Recalling L[y(t)] = H(s) + e −πs H(s), we obtain L[y(t)] = L[h(t)] + e −πs L[h(t)], that is,<br />
y(t) = h(t) + u(t − π)h(t − π).<br />
⊳
132 G. NAGY – ODE july 2, 2013<br />
4.3.5. Exercises.<br />
4.3.1.- . 4.3.2.- .
G. NAGY – ODE July 2, 2013 133<br />
4.4. Generalized sources<br />
This Section provides an elementary presentation <strong>of</strong> a particular generalized function, the<br />
Dirac delta generalized function. We introduce Dirac’s delta using an appropriate sequence<br />
<strong>of</strong> functions, {δn : R → R}, that is, the sequence domain is the whole set <strong>of</strong> real numbers.<br />
The limit <strong>of</strong> the sequence as n → ∞ is not a function R → R though, since the limit diverges<br />
at {0}. Although this limit is a function on the domain R − {0}, it is important to study<br />
the limit <strong>of</strong> sequences in the same domain where the sequence is defined. And in this same<br />
domain, the limit is not a function. For that reason we call this limit a generalized function,<br />
the Dirac delta generalized function. Moreover, we show that this sequence {δn} has a<br />
well-defined sequence <strong>of</strong> Laplace Transforms {L[δn]}. What is much more important, the<br />
sequence <strong>of</strong> Laplace Transforms has a well-defined limit as n → ∞. This is all we need to<br />
solve certain <strong>differential</strong> <strong>equations</strong> having the Dirac delta generalized function as a source.<br />
4.4.1. The Dirac delta generalized function. Since this is an elementary presentation<br />
<strong>of</strong> few generalized functions, we do not mention the theory <strong>of</strong> distributions, which should<br />
be the most appropriate way to present a complete theory <strong>of</strong> generalized functions. Since<br />
we restrict ourselves mainly to the Dirac delta generalized function only, we have chosen to<br />
present it as a limit <strong>of</strong> a function sequence.<br />
A continuous function sequence is a sequence<br />
whose elements are continuous<br />
functions. An example is the continuous<br />
function sequence {un}, for n 1,<br />
is given by<br />
⎧<br />
0, t < 0<br />
⎪⎨<br />
un(t) = nt, 0 t <br />
⎪⎩<br />
1<br />
n<br />
1, t > 1<br />
n .<br />
u<br />
n<br />
1<br />
u u u<br />
3 2 1<br />
0 1/3 1/2 1<br />
Figure 20. The graph <strong>of</strong> functions<br />
un for n = 1, 2, and 3.<br />
It is not difficult to see that every continuous function can be thought as the limit <strong>of</strong> a<br />
continuous function sequence. So the set <strong>of</strong> limits to continuous function sequences contains<br />
the set <strong>of</strong> all continuous functions. However, the set <strong>of</strong> limits is much bigger than the set<br />
<strong>of</strong> continuous functions. For example, it contains discontinuous functions too. The function<br />
sequence {un} is an example <strong>of</strong> a continuous function sequence whose limit is a discontinuous<br />
function. From the few graphs in Fig. 20 we can infer that limn→∞ un(t) = u(t), the step<br />
function, for t = 0, and the limit does not exist at t = 0. So the limit is a discontinuous<br />
function, although every function in the sequence is continuous. The set <strong>of</strong> limits include<br />
even stranger objects. May be the most famous <strong>of</strong> these, which is very important to describe<br />
physical processes, is the Dirac delta generalized function. This object is the limit <strong>of</strong> certain<br />
continuous function sequences. Since there are many different sequences that have the same<br />
limit and calculations are simple with a discontinuous sequence, we introduce here the Dirac<br />
delta generalized function as the limit <strong>of</strong> a sequence <strong>of</strong> bump functions, that is, a sequence<br />
<strong>of</strong> discontinuous functions.<br />
Definition 4.4.1. Given the sequence for n 1 <strong>of</strong> bump functions<br />
<br />
δn(t) = n u(t) − u t − 1 <br />
n<br />
<br />
, (4.4.1)<br />
we call the Dirac delta generalized function to the pointwise limit, for every t ∈ R,<br />
lim<br />
n→∞ δn(t) = δ(t).<br />
t
134 G. NAGY – ODE july 2, 2013<br />
It is simple to see that the sequence<br />
<strong>of</strong> bump functions introduced above<br />
can be re-written as follows,<br />
⎧<br />
0, t < 0<br />
⎪⎨<br />
δn(t) = n, 0 t <br />
⎪⎩<br />
1<br />
n<br />
0, t > 1<br />
n .<br />
We remark that there exist infinitely<br />
many sequences, different for δn<br />
above, that nevertheless define the<br />
same generalized function δ in the<br />
limit n → ∞.<br />
d<br />
n<br />
3<br />
2<br />
1<br />
0<br />
d<br />
3<br />
d<br />
2<br />
1<br />
d<br />
1<br />
Figure 21. The graph <strong>of</strong> the functions<br />
δn for n = 1, 2, and 3.<br />
We see that the Dirac delta generalized function is a function on the domain R−{0}, indeed<br />
it is the zero function, and it is not defined at t = 0, since the limit diverges at that point.<br />
From the graphs given in Fig. 21 is simple to see that<br />
1<br />
0<br />
t<br />
δn(t) dt = 1 for every n 1.<br />
Therefore, the sequence <strong>of</strong> integrals is the constant sequence, {1, 1, · · · }, which has a trivial<br />
limit equal to one as n → ∞. This says that the divergence at t = 0 <strong>of</strong> the sequence<br />
{δn} is <strong>of</strong> a very particular type. Using this limit procedure, one can generalize several<br />
operations from functions to limits <strong>of</strong> function sequences. For example, translations, linear<br />
combinations, and multiplications <strong>of</strong> a function by a generalized function, integration and<br />
Laplace Transforms, as follows:<br />
Definition 4.4.2. We introduce the following operations on the Dirac delta:<br />
<br />
f(t) δ(t − c) + g(t) δ(t − c) = lim f(t) δn(t − c) + g(t) δn(t − c) ,<br />
b<br />
a<br />
δ(t − c) dt = lim<br />
n→∞<br />
b<br />
a<br />
n→∞<br />
δn(t − c) dt, L[δ(t − c)] = lim<br />
n→∞ L[δn(t − c)].<br />
Remark: The notation in the definitions above could be misleading. In the left hand<br />
sides above we use the same notation as we use on functions, although Dirac’s delta is not<br />
a function on R. Take the integral, for example. When we integrate a function f, the<br />
integration symbol means “take a limit <strong>of</strong> Riemann sums”, that is,<br />
b<br />
a<br />
f(t) dt = lim<br />
n<br />
n→∞<br />
i=0<br />
f(xi) ∆x, xi = a + i<br />
b − a<br />
, ∆x =<br />
n n .<br />
However, when f is a generalized function in the sense <strong>of</strong> a limit <strong>of</strong> a sequence <strong>of</strong> functions<br />
{fn}, then by the integration symbol we mean to compute a different limit,<br />
b<br />
a<br />
f(t) dt = lim<br />
n→∞<br />
b<br />
a<br />
fn(t) dt.<br />
We use the same symbol, the integration, to mean two different things, depending whether<br />
we integrate a function or a generalized function. This remark also holds for all the operations<br />
we introduce on generalized functions, specially the Laplace Transform, that will be<br />
<strong>of</strong>ten used in the rest <strong>of</strong> the Section.
G. NAGY – ODE July 2, 2013 135<br />
4.4.2. The Laplace Transform <strong>of</strong> the Dirac delta. Once we have the definitions <strong>of</strong><br />
operations involving the Dirac delta, we can actually compute these limits. The following<br />
statement summarizes few interesting results.<br />
Theorem 4.4.3 (Dirac’s delta). If f : (a, b) → R is continuous and c ∈ (a, b), then<br />
b<br />
a<br />
δ(t − c) dt = 1,<br />
b<br />
a<br />
f(t) δ(t − c) dt = f(c), L[δ(t − c)] = e −cs . (4.4.2)<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 4.4.3: The first equation in (4.4.2) is simple to see from Fig. 21 and<br />
it is left as an exercise. The second equation can be proven as follows.<br />
b<br />
a<br />
δ(t − c) f(t) dt = lim<br />
n→∞<br />
= lim<br />
n→∞<br />
= lim<br />
n→∞<br />
b<br />
a<br />
b<br />
a<br />
δn(t − c) f(t) dt<br />
<br />
n<br />
c+ 1<br />
n<br />
c<br />
u(t − c) − u t − c − 1<br />
n<br />
n f(t) dt,<br />
<br />
f(t), dt<br />
where in the last line we used that c ∈ [a, b]. If we denote by F any primitive <strong>of</strong> f, that is,<br />
F ′ = f, then we can write,<br />
b<br />
a<br />
δ(t − c) f(t) dt = lim<br />
n→∞ nF c + 1 <br />
− F (c)<br />
n<br />
= lim<br />
n→∞<br />
1<br />
F c + 1 <br />
− F (c)<br />
n<br />
1<br />
n<br />
= F ′ (c) ⇒<br />
b<br />
The third equation in (4.4.2) can be proven as follows.<br />
a<br />
δ(t − c) f(t) dt = f(c).<br />
L[δ(t − c)] = lim<br />
n→∞ L[δn(t − c)]<br />
= lim<br />
n→∞ n<br />
<br />
<br />
L[u(t − c)] − L u t − c − 1 <br />
n<br />
<br />
= lim<br />
n→∞ n<br />
1<br />
−cs e e−(c+ n<br />
−<br />
s )s <br />
s<br />
= e −cs s − (1 − e n )<br />
lim <br />
n→∞ s<br />
.<br />
n<br />
The limit on the last line above is a singular limit <strong>of</strong> the form 0<br />
0 , so we can use the l’Hôpital<br />
rule to compute it, that is,<br />
<br />
s − (1 − e n ) −<br />
lim <br />
n→∞ s<br />
= lim<br />
n→∞<br />
n<br />
s<br />
<br />
s<br />
e− n<br />
n2 <br />
− s<br />
n2 s<br />
= lim e− n = 1.<br />
n→∞<br />
We therefore conclude that<br />
L[δ(t − c)] = e −cs .<br />
This establishes the Theorem.
136 G. NAGY – ODE july 2, 2013<br />
4.4.3. The fundamental solution. In this Subsection we solve <strong>differential</strong> <strong>equations</strong> with<br />
the Dirac delta as a source. Since Dirac’s delta is not a function, we need to clarify a bit<br />
what do we mean by a solution to a <strong>differential</strong> equation with such a source.<br />
Definition 4.4.4. Given constants a1, a0, y0, y1, c ∈ R, consider the sequence <strong>of</strong> functions<br />
{yn}, with n 1, solutions <strong>of</strong> the initial value problems<br />
y ′′<br />
n + a1 y ′ n + a0 yn = δn(t − c), yn(0) = y0, y ′ n(0) = y1. (4.4.3)<br />
We define y(t) = L −1<br />
lim<br />
n→∞ L[yn(t)]<br />
<br />
to be the solution <strong>of</strong> the IVP with Dirac’s delta source<br />
y ′′ + a1 y ′ + a0 y = δ(t − c), y(0) = y0, y ′ (0) = y1, (4.4.4)<br />
The definition above seems difficult to work with. However, the following result says that it<br />
is fairly simple to find the function y solution <strong>of</strong> <strong>differential</strong> <strong>equations</strong> with Dirac’s deltas.<br />
Theorem 4.4.5. The function y is solution <strong>of</strong> the IVP with Dirac’s delta source<br />
y ′′ + a1 y ′ + a0 y = δ(t − c), y(0) = y0, y ′ (0) = y1,<br />
iff its Laplace Transform satisfies the equation<br />
<br />
+ a1 s L[y] − y0 − a0 L[y] = e −cs .<br />
s 2 L[y] − sy0 − y 1<br />
This Theorem tells us that to find the solution y to an IVP when the source is a Dirac’s delta<br />
we have to apply the Laplace Transform to the equation and perform the same calculations<br />
as if the Dirac’s delta were a function. So, the final calculation is remarkably simple.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 4.4.5: Compute the Laplace Transform on Eq. (4.4.3),<br />
L[y ′′<br />
n] + a 1 L[y ′ n] + a 0 L[yn] = L[δn(t − c)].<br />
Recall the relations between the Laplace Transform and derivatives and use the initial<br />
conditions,<br />
L[y ′′<br />
n] = s 2 L[yn] − sy0 − y1, L[y ′ ] = s L[yn] − y0,<br />
and use these relation in the <strong>differential</strong> equation,<br />
(s 2 + a1s + a0) L[yn] − sy0 − y1 − a1y0 = L[δn(t − c)],<br />
Since we have shown that lim<br />
n→∞ L[δn(t − c)] = e −cs , then<br />
(s 2 + a1s + a0) lim<br />
n→∞ L[yn] − sy0 − y1 − a1y0 = e −cs .<br />
From the definition <strong>of</strong> y we see that we can re-write the equation above as<br />
(s 2 + a1s + a0) L[y] − sy0 − y1 − a1y0 = e −cs ,<br />
which is equivalent to<br />
2<br />
s L[y] − sy0 − y1 + a1 s L[y] − y0 − a0 L[y] = e −cs .<br />
This establishes the Theorem. <br />
We now introduce a name for the solution <strong>of</strong> a <strong>differential</strong> equation with a Dirac’s delta<br />
generalized function as a source and zero initial conditions. Such solution will play an<br />
important role later on, that is why we give that function a particular name.<br />
Definition 4.4.6. The function yδ is a fundamental solution <strong>of</strong> the linear operator<br />
L(y) = y ′′ + p y ′ + q y, where p, q are continuous functions, iff the function yδ is solution <strong>of</strong><br />
the initial value problem<br />
L(yδ) = δ(t − c), yδ(0) = 0, y ′ δ(0) = 0, c ∈ R.
G. NAGY – ODE July 2, 2013 137<br />
Example 4.4.1: Find the function yδ, the fundamental solution <strong>of</strong> the initial value problem<br />
y ′′ + 2y ′ + 2y = δ(t), y(0) = 0, y ′ (0) = 0.<br />
Solution: We compute the Laplace transform on both sides <strong>of</strong> the equation above,<br />
L[y ′′ ] + 2 L[y ′ ] + 2 L[y] = L[δ(t)] = 1 ⇒ (s 2 + 2s + 2) L[y] = 1,<br />
where we have introduced the initial conditions on the last equation above. So we obtain<br />
1<br />
L[y] =<br />
(s2 + 2s + 2) .<br />
The denomination in the equation above has complex valued roots, since<br />
s± = 1<br />
√ <br />
−2 ± 4 − 8 ,<br />
2<br />
therefore, we complete squares s2 + 2s + 2 = (s + 1) 2 + 1. We need to solve the equation<br />
L[y] =<br />
1<br />
(s + 1) 2 + 1 = L[e −t sin(t)] ⇒ yδ(t) = e −t sin(t).<br />
Example 4.4.2: Find the solution y to the initial value problem<br />
y ′′ − y = −20 δ(t − 3), y(0) = 1, y ′ (0) = 0.<br />
Solution: The Laplace transform <strong>of</strong> the <strong>differential</strong> equation is given by<br />
L[y ′′ ] − L[y] = −20 L[δ(t − 3)] ⇒ (s 2 − 1) L[y] − s = −20 e −3s ,<br />
where in the second equation we have already introduced the initial conditions. We arrive<br />
to the equation<br />
s<br />
L[y] =<br />
(s2 1<br />
− 20 e−3s<br />
− 1) (s2 = L[cosh(t)] − 20 L[u(t − 3) sinh(t − 3)],<br />
− 1)<br />
which leads to the solution<br />
y(t) = cosh(t) − 20 u(t − 3) sinh(t − 3).<br />
Example 4.4.3: Find the solution to the initial value problem<br />
y ′′ + 4y = δ(t − π) − δ(t − 2π), y(0) = 0, y ′ (0) = 0.<br />
Solution: We again Laplace transform both sides <strong>of</strong> the <strong>differential</strong> equation,<br />
L[y ′′ ] + 4 L[y] = L[δ(t − π)] − L[δ(t − 2π)] ⇒ (s 2 + 4) L[y] = e −πs − e −2πs ,<br />
where in the second equation above we have introduced the initial conditions. Then,<br />
L[y] = e−πs<br />
(s2 e−2πs<br />
−<br />
+ 4) (s2 + 4)<br />
= e−πs 2<br />
2 (s2 e−2πs 2<br />
−<br />
+ 4) 2 (s2 + 4)<br />
= 1<br />
2 L<br />
<br />
u(t − π) sin 2(t − π) <br />
− 1<br />
2 L<br />
<br />
u(t − 2π) sin 2(t − 2π) <br />
.<br />
The last equation can be rewritten as follows,<br />
y(t) = 1<br />
2 u(t − π) sin2(t − π) − 1<br />
2 u(t − 2π) sin2(t − 2π) ,<br />
⊳<br />
⊳
138 G. NAGY – ODE july 2, 2013<br />
which leads to the conclusion that<br />
y(t) = 1 <br />
u(t − π) − u(t − 2π) sin(2t).<br />
2<br />
Remark: Dirac’s delta generalized function is very useful to describe certain situation in<br />
physical systems. One example is to describe impulsive forces in mechanical systems. An<br />
impulsive force on a particle is a force that transmits a finite momentum to the particle in<br />
an infinitely short time. For example, the momentum transmitted to a pendulum when hit<br />
by a hammer. The force acting on the pendulum due to the hammer can be described with<br />
a Dirac’s delta. Recall Newton’s law <strong>of</strong> motion,<br />
m v ′ (t) = F (t), with F (t) = F0 δ(t − t0).<br />
Then, the linear momentum transfer from the hammer to the pendulum is,<br />
∆I = lim<br />
∆t→0 mv(t)<br />
<br />
<br />
t0+∆t<br />
t0+∆t<br />
= lim F (t) dt = F0.<br />
∆t→0<br />
That is, ∆I = F 0.<br />
t0−∆t<br />
t0−∆t<br />
⊳
4.4.4. Exercises.<br />
4.4.1.- . 4.4.2.- .<br />
G. NAGY – ODE July 2, 2013 139
140 G. NAGY – ODE july 2, 2013<br />
4.5. Convolution solutions<br />
This Section describes a way to express a solution to an inhomogeneous linear <strong>differential</strong><br />
equation in terms <strong>of</strong> a convolution between the fundamental solution and the source<br />
function <strong>of</strong> the <strong>differential</strong> equation. This particular form to express the solution is useful<br />
in mathematical physics to study different properties <strong>of</strong> solution. It is also a widely used<br />
in physics and engineering. The plan for this Section is to first present the definition <strong>of</strong> a<br />
convolution <strong>of</strong> two functions, and then discuss the main properties <strong>of</strong> this operation, with<br />
Theorems 4.5.3 and 4.5.5 containing perhaps the most important results <strong>of</strong> this Section. The<br />
latter result uses convolutions to write solutions to inhomogeneous <strong>differential</strong> <strong>equations</strong>.<br />
4.5.1. The convolution <strong>of</strong> two functions. We start with a definition <strong>of</strong> the convolution<br />
<strong>of</strong> two functions. One can say that the convolution is a generalization <strong>of</strong> the product <strong>of</strong> two<br />
functions. In a convolution one multiplies the two functions evaluated at different points,<br />
and then integrates the result. Here is a precise definition.<br />
Definition 4.5.1. Assume that f and g are piecewise continuous functions, or one <strong>of</strong> them<br />
is a Dirac’s delta generalized function. The convolution <strong>of</strong> f and g is a function denoted<br />
by f ∗ g and given by the following expression,<br />
(f ∗ g)(t) =<br />
t<br />
0<br />
f(τ)g(t − τ) dτ.<br />
In the Example we explicitly compute the integral defining the convolution <strong>of</strong> two functions.<br />
Example 4.5.1: Find the convolution <strong>of</strong> the functions f(t) = e −t and g(t) = sin(t).<br />
Solution: The definition <strong>of</strong> convolution is,<br />
(f ∗ g)(t) =<br />
t<br />
0<br />
e −τ sin(t − τ) dτ.<br />
This integral is not difficult to compute. Integrate by parts twice,<br />
t<br />
e −τ <br />
sin(t − τ) dτ = e −τ t <br />
<br />
cos(t − τ) − e −τ t <br />
sin(t − τ) −<br />
0<br />
that is,<br />
2<br />
t<br />
0<br />
e −τ sin(t − τ) dτ =<br />
We then conclude that<br />
0<br />
0<br />
t<br />
0<br />
e −τ sin(t − τ) dτ,<br />
<br />
e −τ t <br />
<br />
cos(t − τ) − e<br />
0<br />
−τ t <br />
sin(t − τ) = e<br />
0<br />
−t − cos(t) − 0 + sin(t).<br />
(f ∗ g)(t) = 1<br />
−t<br />
e + sin(t) − cos(t) . (4.5.1)<br />
2<br />
⊳<br />
The main properties <strong>of</strong> the convolution are summarized in the following result:<br />
Lemma 4.5.2 (Properties). For every piecewise continuous functions f, g, and h, hold:<br />
(i) Commutativity: f ∗ g = g ∗ f;<br />
(ii) Associativity: f ∗ (g ∗ h) = (f ∗ g) ∗ h;<br />
(iii) Distributivity: f ∗ (g + h) = f ∗ g + f ∗ h;<br />
(iv) Neutral element: f ∗ 0 = 0;<br />
(v) Identity element: f ∗ δ = f.
G. NAGY – ODE July 2, 2013 141<br />
Pro<strong>of</strong> <strong>of</strong> Lemma 4.5.2: We only prove properties (i) and (v), the rest are simple to obtain<br />
from the definition <strong>of</strong> convolution. The first property can be obtained by a change <strong>of</strong> the<br />
integration variable as follows,<br />
(f ∗ g)(t) =<br />
=<br />
=<br />
t<br />
0<br />
0<br />
t<br />
t<br />
0<br />
f(τ) g(t − τ) dτ, ˆτ = t − τ, dˆτ = −dτ,<br />
f(t − ˆτ) g(ˆτ)(−1) dˆτ<br />
g(ˆτ) f(t − ˆτ) dˆτ ⇒ (f ∗ g)(t) = (g ∗ f)(t).<br />
Finally, property (v) is straightforward form the definition <strong>of</strong> the delta generalized function,<br />
(f ∗ δ)(t) =<br />
t<br />
0<br />
f(τ) δ(t − τ) dτ = f(t).<br />
This establishes the Lemma. <br />
4.5.2. The Laplace Transform <strong>of</strong> a convolution. We now show the relation between<br />
the Laplace transform and the convolution operations. This is one <strong>of</strong> the most important<br />
results we need to solve <strong>differential</strong> <strong>equations</strong>. The Laplace transform <strong>of</strong> a convolution is<br />
the usual product <strong>of</strong> the individual Laplace transforms.<br />
Theorem 4.5.3 (Laplace Transform). If the functions f and g have well-defined Laplace<br />
Transforms L[f] and L[g], including the case that one <strong>of</strong> them is a Dirac’s delta, then<br />
L[f ∗ g] = L[f] L[g].<br />
Example 4.5.2: Compute the Laplace transform <strong>of</strong> the function u(t) =<br />
Solution: The function u above is the convolution <strong>of</strong> the functions<br />
f(t) = e −t , g(t) = sin(t),<br />
that is, u = f ∗ g. Therefore, Theorem 4.5.3 says that<br />
Since,<br />
L[u] = L[f ∗ g] = L[f] L[g].<br />
L[f] = L[e −t ] = 1<br />
1<br />
, L[g] = L[sin(t)] =<br />
s + 1 s2 + 1 ,<br />
we then conclude that L[u] = L[f ∗ g] is given by<br />
L[f ∗ g] =<br />
1<br />
(s + 1)(s 2 + 1) .<br />
t<br />
0<br />
e −τ sin(t−τ) dτ.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 4.5.3: The key step is to interchange two integrals. We start we the<br />
product <strong>of</strong> the Laplace Transforms,<br />
<br />
L[f] L[g] =<br />
∞<br />
e<br />
0<br />
−st <br />
f(t) dt<br />
∞<br />
e<br />
0<br />
−s˜t<br />
<br />
g(˜t) d˜t<br />
∞<br />
= e<br />
0<br />
−s˜t<br />
<br />
g(˜t)<br />
∞<br />
e<br />
0<br />
−st <br />
f(t) dt d˜t<br />
∞ <br />
= g(˜t)<br />
∞<br />
e −s(t+˜t)<br />
<br />
f(t) dt d˜t,<br />
0<br />
0<br />
⊳
142 G. NAGY – ODE july 2, 2013<br />
where we only introduced the integral in t as a constant inside the integral in ˜t. Introduce<br />
the Change <strong>of</strong> variables in the inside integral τ = t + ˜t, hence dτ = dt. Then, we get<br />
∞ <br />
L[f] L[g] = g(˜t)<br />
0<br />
∞<br />
e<br />
˜t<br />
−sτ <br />
f(τ − ˜t) dτ d˜t (4.5.2)<br />
∞ ∞<br />
= e −sτ g(˜t) f(τ − ˜t) dτ d˜t. (4.5.3)<br />
Here is the key step, we must switch the order <strong>of</strong><br />
integration. From Fig. 22 we see that changing the<br />
order <strong>of</strong> integration gives the following expression,<br />
∞ τ<br />
L[f] L[g] = e −sτ g(˜t) f(τ − ˜t) d˜t dτ.<br />
0<br />
0<br />
Then, is straightforward to check that<br />
∞<br />
L[f] L[g] = e<br />
0<br />
−sτ τ<br />
<br />
g(˜t) f(τ − ˜t) d˜t dτ<br />
0<br />
∞<br />
= e −sτ (g ∗ f)(τ) dt<br />
0<br />
= L[g ∗ f] ⇒ L[f] L[g] = L[f ∗ g].<br />
This establishes the Theorem. <br />
Example 4.5.3: Use Laplace transforms to compute u(t) =<br />
0<br />
˜t<br />
t<br />
0<br />
t = tau<br />
Figure 22. Domain <strong>of</strong><br />
integration in (4.5.3).<br />
t<br />
0<br />
e −τ sin(t − τ) dτ.<br />
Solution: We know that u = f ∗ g, with f(t) = e−t and g(t) = sin(t), and we have seen in<br />
Example 4.5.2 that<br />
1<br />
L[u] = L[f ∗ g] =<br />
(s + 1)(s2 + 1) .<br />
A partial fraction decomposition <strong>of</strong> the right-hand side above implies that<br />
This says that<br />
We then conclude that<br />
L[u] = 1<br />
<br />
1 (1 − s)<br />
+<br />
2 (s + 1) (s2 <br />
+ 1)<br />
= 1<br />
<br />
1<br />
2 (s + 1) +<br />
1<br />
(s2 + 1) −<br />
s<br />
(s2 <br />
+ 1)<br />
= 1<br />
<br />
L[e<br />
2<br />
−t <br />
] + L[sin(t)] − L[cos(t)] .<br />
u(t) = 1<br />
−t<br />
e + sin(t) − cos(t) .<br />
2<br />
(f ∗ g)(t) = 1<br />
−t<br />
e + sin(t) − cos(t) ,<br />
2<br />
which agrees with Eq. (4.5.1) in the first example above. ⊳<br />
tau
G. NAGY – ODE July 2, 2013 143<br />
4.5.3. Impulse response solutions. We now introduce the notion <strong>of</strong> the impulse response<br />
solution to a <strong>differential</strong> equation. This is the solution <strong>of</strong> the the initial value problem for<br />
the <strong>differential</strong> equation with zero initial conditions and a Dirac’s delta as a source function.<br />
Definition 4.5.4. The impulse response solution is the function yδ solution <strong>of</strong> the IVP<br />
y ′′<br />
δ + a1 y ′ δ + a0 yδ = δ(t − c), yδ(0) = 0, y ′ δ(0) = 0, c ∈ R.<br />
Suppose the <strong>differential</strong> equation describes the motion <strong>of</strong> a particle and the the source is the<br />
force applied to the particle. The impulse response solution describes the movement <strong>of</strong> the<br />
particle that starts at the initial position with zero velocity and moves because an impulsive<br />
force kicked the particle in an infinitesimal short time.<br />
Example 4.5.4: Find the impulse response solution <strong>of</strong> the initial value problem<br />
y ′′<br />
δ + 2 y ′ δ + 2 yδ = δ(t − c), yδ(0) = 0, y ′ δ(0) = 0, .<br />
Solution: We have to use the Laplace Transform to solve this problem because the source<br />
is a Dirac’s delta generalized function. So, compute the Laplace Transform <strong>of</strong> the <strong>differential</strong><br />
equation,<br />
L[y ′′<br />
δ ] + 2 L[y ′ δ] + 2 L[yδ] = L[δ(t − c)].<br />
Since the initial conditions are all zero, we get<br />
Find the roots <strong>of</strong> the denominator,<br />
(s 2 + 2s + 2) L[yδ] = e −cs ⇒ L[yδ] =<br />
e −cs<br />
(s 2 + 2s + 2) .<br />
s 2 + 2s + 2 = 0 ⇒ s± = 1 √ <br />
−2 ± 4 − 8<br />
2<br />
The denominator has complex roots. Then, it is convenient to complete the square in the<br />
denominator,<br />
s 2 <br />
+ 2s + 2 = s 2 <br />
2<br />
<br />
+ 2 s + 1 − 1 + 2 = (s + 1)<br />
2<br />
2 + 1.<br />
Therefore, we obtain the expression,<br />
L[yδ] =<br />
e −cs<br />
(s + 1) 2 + 1 .<br />
Recall that L[sin(t)] = 1<br />
s 2 + 1 , and L[f](s − c) = L[ect f(t)]. Then,<br />
1<br />
(s + 1) 2 + 1 = L[e−t sin(t)] ⇒ L[yδ] = e −cs L[e −t sin(t)].<br />
Since e −cs L[f](s) = L[u(t − c) f(t − c)], we conclude that<br />
yδ(t) = u(t − c) e −(t−c) sin(t − c).<br />
⊳
144 G. NAGY – ODE july 2, 2013<br />
4.5.4. The solution decomposition Theorem. One could say Theorem 4.5.5 is the main<br />
result <strong>of</strong> this Section. It presents a decomposition <strong>of</strong> the solution to a general second order,<br />
linear, constant coefficients initial value problem. The statement says that the solution to<br />
such problem can always be split in two terms, the first containing information only about<br />
the initial data, while the second contains information only about the source function.<br />
Theorem 4.5.5 (Solution decomposition). Given constants a 0, a 1, y 0, y 1 and a piecewise<br />
continuous function g, the solution y to the initial value problem<br />
can be decomposed as<br />
y ′′ + a 1 y ′ + a 0 y = g(t), y(0) = y 0, y ′ (0) = y 1, (4.5.4)<br />
y(t) = yh(t) + (yδ ∗ g)(t), (4.5.5)<br />
where yh is the solution <strong>of</strong> the homogeneous initial value problem<br />
y ′′<br />
h + a 1 y ′ h + a 0 yh = 0, yh(0) = y 0, y ′ h(0) = y 1, (4.5.6)<br />
and yδ is the impulse response solution, that is,<br />
y ′′<br />
δ + a1 y ′ δ + a0 yδ = δ(t), yδ(0) = 0, y ′ δ(0) = 0.<br />
The solution decomposition in Eq. (4.5.5) can be expressed in the equivalent way<br />
y(t) = yh(t) +<br />
t<br />
0<br />
yδ(τ)g(t − τ) dτ.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem4.5.5: Compute the Laplace Transform <strong>of</strong> the <strong>differential</strong> equation,<br />
L[y ′′ ] + a1 L[y ′ ] + a0 L[y] = L[g(t)].<br />
Recalling the relations between Laplace Transforms and derivatives,<br />
L[y ′′ ] = s 2 L[y] − sy0 − y1, L[y ′ ] = s L[y] − y0.<br />
we re-write the <strong>differential</strong> equation for y as an algebraic equation for cL[y],<br />
(s 2 + a1s + a0) L[y] − sy0 − y1 − a1y0 = L[g(t)].<br />
As usual, it is simple to solve the algebraic equation for cL[y],<br />
L[y] = (s + a1)y0 + y1<br />
(s2 + a1s + a0) +<br />
1<br />
(s2 + a1s + a0) L[g(t)].<br />
Now, the function yh is the solution <strong>of</strong> Eq. (4.5.6), that is,<br />
L[yh] = (s + a 1)y 0 + y 1<br />
(s 2 + a1s + a0) .<br />
And by the definition <strong>of</strong> the impulse response solution yδ we have that<br />
1<br />
L[yδ] =<br />
(s2 + a1s + a0) .<br />
These last three equation imply,<br />
L[y] = L[yh] + L[yδ] L[g(t)].<br />
This is the Laplace Transform version <strong>of</strong> Eq. (4.5.5). Inverting the Laplace Transform above,<br />
y(t) = yh(t) + L −1 L[yδ] L[g(t)] .<br />
Using the result in Theorem 4.5.3 in the last term above we conclude that<br />
y(t) = yh(t) + (yδ ∗ g)(t).
G. NAGY – ODE July 2, 2013 145<br />
Example 4.5.5: Use the Solution Decomposition Theorem to express the solution <strong>of</strong><br />
y ′′ + 2 y ′ + 2 y = sin(at), y(0) = 1, y ′ (0) = −1.<br />
Solution: Compute the Laplace Transform <strong>of</strong> the <strong>differential</strong> equation above,<br />
L[y ′′ ] + 2 L[y ′ ] + 2 L[y] = L[sin(at)],<br />
and recall the relations between the Laplace Transform and derivatives,<br />
L[y ′′ ] = s 2 L[y] − sy(0) − y ′ (0), L[y ′ ] = s L[y] − y(0).<br />
Introduce the initial conditions in the equation above,<br />
L[y ′′ ] = s 2 L[y] − s (1) − (−1), L[y ′ ] = s L[y] − 1,<br />
and these two equation into the <strong>differential</strong> equation,<br />
Re-order terms to get<br />
(s 2 + 2s + 2) L[y] − s + 1 − 2 = L[sin(at)].<br />
L[y] =<br />
(s + 1)<br />
(s2 + 2s + 2) +<br />
1<br />
(s2 + 2s + 2) L[sin(at)].<br />
Now, the function yh is the solution <strong>of</strong> the homogeneous initial value problem with the same<br />
initial conditions as y, that is,<br />
L[yh] =<br />
(s + 1)<br />
(s2 (s + 1)<br />
=<br />
+ 2s + 2) (s + 1) 2 + 1 = L[e−t cos(t)].<br />
Now, the function yδ is the impulse response solution for the <strong>differential</strong> equation in this<br />
Example, that is,<br />
1<br />
cL[yδ] =<br />
(s2 + 2s + 2) =<br />
1<br />
(s + 1) 2 + 1 = L[e−t sin(t)].<br />
Put all this information together and denote g(t) = sin(at) to get<br />
More explicitly, we get<br />
L[y] = L[yh] + L[yδ] L[g(t)] ⇒ y(t) = yh(t) + (yδ ∗ g)(t),<br />
y(t) = e −t cos(t) +<br />
t<br />
0<br />
e −τ sin(τ) sin[a(t − τ)] dτ.<br />
⊳
146 G. NAGY – ODE july 2, 2013<br />
4.5.5. Exercises.<br />
4.5.1.- . 4.5.2.- .
G. NAGY – ODE July 2, 2013 147<br />
Chapter 5. Systems <strong>of</strong> <strong>differential</strong> <strong>equations</strong><br />
The main interest <strong>of</strong> this Chapter resides in its last Section, where we study in some detail<br />
the solutions <strong>of</strong> 2 × 2 linear first order <strong>differential</strong> <strong>equations</strong> with constant coefficients. A<br />
system <strong>of</strong> <strong>differential</strong> <strong>equations</strong> is harder to solve than a single <strong>differential</strong> equation because<br />
the <strong>equations</strong> are coupled. The derivative <strong>of</strong> the i-th unknown depends on all the other<br />
unknowns, not just the i-th one. In the last Section <strong>of</strong> this Chapter we find solutions to<br />
linear <strong>differential</strong> systems in the case that the constant coefficients matrix is diagonalizable.<br />
The latter means that when the system is written in terms <strong>of</strong> new unknowns based on<br />
the coefficient matrix eigenvectors, the system decouples. One then solves the individual<br />
decoupled <strong>equations</strong>, and then transforms back the result to the original unknowns. In order<br />
to present this method we first need to review well-known material form linear algebra. This<br />
review will focus on on matrix algebra, determinants, linearly dependent and independent<br />
vector sets, and the calculation <strong>of</strong> eigenvalues and eigenvectors <strong>of</strong> a square matrix.<br />
5.1. Introduction<br />
5.1.1. Systems <strong>of</strong> <strong>differential</strong> <strong>equations</strong>. A system <strong>of</strong> <strong>differential</strong> <strong>equations</strong> is a set<br />
<strong>of</strong> more than one <strong>differential</strong> equation on more than one unknown function. Systems <strong>of</strong><br />
<strong>differential</strong> <strong>equations</strong> <strong>of</strong>ten appear in physical descriptions <strong>of</strong> nature. We start with may be<br />
the most common example, Newton’s law <strong>of</strong> motion for point particles.<br />
Example 5.1.1: Newton’s second law <strong>of</strong> motion for a point particle <strong>of</strong> mass m moving<br />
in space is a system <strong>of</strong> <strong>differential</strong> <strong>equations</strong>. The unknown functions are the particle<br />
coordinates in space, x 1, x 2, and x 3. The source functions are the three components <strong>of</strong> the<br />
force, F 1, F 2, and F 3, which depend on time and space. Newton’s second law <strong>of</strong> motion is<br />
a system <strong>of</strong> three <strong>differential</strong> <strong>equations</strong> given by<br />
m d2x1 dt2 = F <br />
1 t, x1(t), x2(t), x3(t) ,<br />
m d2x2 <br />
= F2 t, x1(t), x2(t), x3(t)<br />
dt2 ,<br />
m d2x3 <br />
= F3 t, x1(t), x2(t), x3(t)<br />
dt2 .<br />
This is an example <strong>of</strong> a system <strong>of</strong> three <strong>differential</strong> <strong>equations</strong> in the three unknown functions<br />
x1, x2, and x3. The <strong>equations</strong> are <strong>of</strong> second order, since they involve second derivatives <strong>of</strong><br />
the unknown functions. We will see in later Sections that it is convenient to group both the<br />
unknown functions and the source functions in vector valued functions,<br />
⎡ ⎤<br />
⎡ ⎤<br />
x1(t) F1(t, x)<br />
x(t) = ⎣x2(t)<br />
⎦ , F(t) = ⎣F2(t,<br />
x) ⎦ .<br />
x3(t)<br />
F3(t, x)<br />
We will then use vector notation to express systems <strong>of</strong> <strong>differential</strong> <strong>equations</strong>. In the case <strong>of</strong><br />
Newton’s law <strong>of</strong> motion we get<br />
m d2 x<br />
dt 2 (t) = F t, x(t) .<br />
Newton’s law <strong>of</strong> motion mentioned above is a particular case <strong>of</strong> what is called a second<br />
order <strong>differential</strong> system. This is our first example <strong>of</strong> a system <strong>of</strong> <strong>differential</strong> <strong>equations</strong><br />
because Newton’s law are so widely used in physics. However, in these notes we start<br />
introducing a first order <strong>differential</strong> system.<br />
⊳
148 G. NAGY – ODE july 2, 2013<br />
Definition 5.1.1. An n × n first order system <strong>of</strong> <strong>differential</strong> <strong>equations</strong> is the following:<br />
Given functions Fi : [t1, t2] × Rn → R, where i = 1, · · · , n, find n functions<br />
xj : [t1, t2] → R solutions <strong>of</strong> the n linear <strong>differential</strong> <strong>equations</strong><br />
x ′ <br />
1(t) = F1 t, x1(t), · · · , xn(t) , (5.1.1)<br />
.<br />
x ′ <br />
n(t) = Fn t, x1(t), · · · , xn(t) . (5.1.2)<br />
Example 5.1.2: n = 1: This is a single <strong>differential</strong> equation: Find x1(t) solution <strong>of</strong><br />
x ′ (x1) 1 = −<br />
3<br />
<br />
2x1 + t(x1) 2<br />
Actually, this equation can be transformed into an exact equation when multiplied by an<br />
appropriate integrating factor. We have seen how to find solutions solutions to <strong>equations</strong> <strong>of</strong><br />
this type in Section 1.4. ⊳<br />
Example 5.1.3: n = 2: This is a 2 × 2 <strong>differential</strong> system: Find x1(t) and x2(t) solutions <strong>of</strong><br />
x ′ 1 = e t (x1) 2 + t 2 (x2) 2 ,<br />
x ′ 2 = e x1x2 . ⊳<br />
An important particular case <strong>of</strong> an n × n <strong>differential</strong> system is when the <strong>differential</strong><br />
<strong>equations</strong> are linear in the unknown functions.<br />
Definition 5.1.2. An n × n first order system <strong>of</strong> linear <strong>differential</strong> <strong>equations</strong> is<br />
the following: Given functions aij, gi : [t1, t2] → R, where i, j = 1, · · · , n, find n functions<br />
xj : [t1, t2] → R solutions <strong>of</strong> the n linear <strong>differential</strong> <strong>equations</strong><br />
x ′ 1 = a11(t) x1 + · · · + a1n(t) xn + g1(t),<br />
.<br />
x ′ n = an1(t) x 1 + · · · + ann(t) xn + gn(t).<br />
The system is called homogeneous iff the source functions satisfy that g 1 = · · · = gn = 0.<br />
The system is called <strong>of</strong> constant coefficients iff all the functions aij are constants.<br />
We see that the linear <strong>differential</strong> system is a particular case <strong>of</strong> a <strong>differential</strong> system where<br />
functions Fi, for i = 1, · · · , n have the form<br />
Fi(t, x1, · · · , xn) = ai1(t) x1 + · · · + ain(t) xn + gi(t),<br />
that is, the Fi are linear functions <strong>of</strong> the xi.<br />
Example 5.1.4: n = 1: This is a single <strong>differential</strong> equation: Find x1(t) solution <strong>of</strong><br />
x ′ 1 = a11(t) x 1 + g 1(t).<br />
This is a linear first order equation, and solutions can be found with the integrating factor<br />
method described in Section 1.2. ⊳<br />
Example 5.1.5: n = 2: This is a 2 × 2 linear system: Find x 1(t) and x 2(t) solutions <strong>of</strong><br />
x ′ 1 = a11(t) x1 + a12(t) x2 + g1(t),<br />
x ′ 2 = a21(t) x1 + a22(t) x2 + g2(t). ⊳
G. NAGY – ODE July 2, 2013 149<br />
Example 5.1.6: n = 2: A 2 × 2 homogeneous linear system: Find x 1(t) and x 2(t) such that<br />
x ′ 1 = a11(t) x 1 + a12(t) x 2<br />
x ′ 2 = a21(t) x1 + a22(t) x2. ⊳<br />
A solution <strong>of</strong> an n × n <strong>differential</strong> system is a set <strong>of</strong> n functions {x 1, · · · , xn} that satisfy<br />
every equation in the <strong>differential</strong> system. Usually it is more complicated to solve a system<br />
<strong>of</strong> <strong>differential</strong> <strong>equations</strong> than a single <strong>differential</strong> equation. The reason is that in a system<br />
the <strong>equations</strong> are usually coupled. For example, consider the 2 × 2 system in Example 5.1.6<br />
above. One must know the function x2 in order to integrate the first equation to obtain the<br />
function x1. Similarly, one has to know function x1 to integrate the second equation to get<br />
function x2. This is what we mean by a coupled system. One cannot integrate one equation<br />
at a time; one must integrate the whole system together. We now show how this can be<br />
done in a simple example. Although the example is simple, the idea to find solutions is<br />
powerful. Generalizations <strong>of</strong> this idea is what one does to solve much more general systems<br />
<strong>of</strong> <strong>differential</strong> <strong>equations</strong>.<br />
Example 5.1.7: Find functions x1, x2 solutions <strong>of</strong> the first order, 2×2, constant coefficients,<br />
homogeneous <strong>differential</strong> system<br />
x ′ 1 = x1 − x2,<br />
x ′ 2 = −x1 + x2.<br />
Solution: The main idea to solve this system comes from the following observation. If we<br />
add up the two <strong>equations</strong> <strong>equations</strong>, and if we subtract the second equation from the first,<br />
we obtain, respectively,<br />
(x1 + x2) ′ = 0, (x1 − x2) ′ = 2(x1 − x2).<br />
To understand the consequences <strong>of</strong> what we have done, let us introduce the new unknowns<br />
v = x1 + x2, and w = x1 − x2, and re-write the <strong>equations</strong> above with these new unknowns,<br />
v ′ = 0, w ′ = 2w.<br />
We have decoupled the original system. The <strong>equations</strong> for x1 and x2 are coupled, but we<br />
have found a linear combination <strong>of</strong> the <strong>equations</strong> such that the <strong>equations</strong> for v and w are<br />
not coupled. We now solve each equation independently <strong>of</strong> the other.<br />
v ′ = 0 ⇒ v = c 1,<br />
w ′ = 2w ⇒ w = c2e 2t ,<br />
with c1, c2 ∈ R. Having obtained the solutions for the decoupled system, we now transform<br />
back the solutions to the original unknown functions. From the definitions <strong>of</strong> v and w we<br />
see that<br />
x1 = 1<br />
(v − w).<br />
2 (v + w), x2 = 1<br />
2<br />
We conclude that for all c1, c2 ∈ R the functions x1, x2 below are solutions <strong>of</strong> the 2 × 2<br />
<strong>differential</strong> system in the example, namely,<br />
x 1(t) = 1<br />
2 (c 1 + c 2e 2t ), x 2(t) = 1<br />
2 (c 1 − c 2e 2t ).<br />
This example contains the main idea to solve systems <strong>of</strong> linear <strong>differential</strong> <strong>equations</strong>. If<br />
the <strong>differential</strong> <strong>equations</strong> in a linear system are coupled, then find appropriate linear combinations<br />
<strong>of</strong> the <strong>equations</strong> and the unknowns such that the new system for new unknowns<br />
⊳
150 G. NAGY – ODE july 2, 2013<br />
is not coupled. Solve each equation in the uncoupled system using the methods from Section<br />
1.2. Once the solution for the new unknowns are obtained, transform them back to the<br />
original unknowns. The resulting functions will be solution to the original linear <strong>differential</strong><br />
system.<br />
We key part in this method is to find the appropriate linear combinations <strong>of</strong> the <strong>equations</strong><br />
and unknowns that decouples the system. In the example above that transformation was<br />
simple to find. This is not the case in a general linear system, and this is the reason we will<br />
need to review concepts from Linear Algebra. We will find out that the transformation that<br />
decouples a linear <strong>differential</strong> system is deeply related with the eigenvalues and eigenvectors<br />
<strong>of</strong> the system coefficient matrix.<br />
5.1.2. Reduction to first order systems. A second order linear <strong>differential</strong> equation for<br />
a single unknown function can be transformed into a 2 × 2 linear system for the unknown<br />
and its first derivative. More precisely, we have the following statement.<br />
Theorem 5.1.3 (First order reduction). Every solution y to the second order equation<br />
y ′′ + p(t) y ′ + q(t) y = g(t), (5.1.3)<br />
defines a solution x1 = y and x2 = y ′ <strong>of</strong> the 2 × 2 first order linear <strong>differential</strong> system<br />
x ′ 1 = x 2, (5.1.4)<br />
x ′ 2 = −q(t) x1 − p(t) x2 + g(t). (5.1.5)<br />
Conversely, every solution x 1, x 2 <strong>of</strong> the 2×2 first order linear system in Eqs. (5.1.4)-(5.1.5)<br />
defines a solution y = x 1 <strong>of</strong> the second order <strong>differential</strong> equation in (5.1.3).<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.1.3:<br />
(⇒) Given a solution y <strong>of</strong> Eq. (5.1.3), introduce the functions x1 = y and x2 = y ′ . Therefore<br />
Eq. (5.1.4) holds, due to the relation<br />
x ′ 1 = y ′ = x 2,<br />
Also Eq. (5.1.5) holds, because <strong>of</strong> the equation<br />
x ′ 2 = y ′′ = −q(t) y − p(t) y ′ + g(t) ⇒ x ′′<br />
2 = −q(t) x1 − p(t) x2 + g(t).<br />
(⇐) Differentiate Eq. (5.1.4) and introduce the result into Eq. (5.1.5), that is,<br />
Denoting y = x1, we obtain,<br />
x ′′<br />
1 = x ′ 2 ⇒ x ′′<br />
1 = −q(t) x1 − p(t) x ′ 1 + g(t).<br />
y ′′ + p(t) y ′ + q(t) y = g(t).<br />
This establishes the Theorem. <br />
Example 5.1.8: Express as a first order system the second order equation<br />
Solution: Introduce the new unknowns<br />
y ′′ + 2y ′ + 2y = sin(at).<br />
x1 = y, x2 = y ′ ⇒ x ′ 1 = x2.<br />
Then, the <strong>differential</strong> equation can be written as<br />
We conclude that<br />
x ′ 2 + 2x2 + 2x1 = sin(at).<br />
x ′ 1 = x2, x ′ 2 = −2x1 − 2x2 + sin(at).<br />
⊳
G. NAGY – ODE July 2, 2013 151<br />
Example 5.1.9: Express as a single second order equation the 2 × 2 system and solve it,<br />
x ′ 1 = −x1 + 3x2,<br />
x ′ 2 = x1 − x2.<br />
Solution: Compute x1 from the second equation: x1 = x ′ 2 + x2. Introduce this expression<br />
into the first equation,<br />
(x ′ 2 + x2) ′ = −(x ′ 2 + x2) + 3x2 ⇒ x ′′<br />
2 + x ′ 2 = −x ′ 2 − x2 + 3x2,<br />
so we obtain the second order equation<br />
x ′′<br />
2 + 2x ′ 2 − 2x2 = 0.<br />
We solve this equation with the methods studied in Chapter 2, that is, we look for solutions<br />
<strong>of</strong> the form x2(t) = e rt , with r solution <strong>of</strong> the characteristic equation<br />
r 2 + 2r − 2 = 0 ⇒ r± = 1 √ <br />
−2 ± 4 + 8 ⇒ r± = −1 ±<br />
2<br />
√ 3.<br />
Therefore, the general solution to the second order equation above is x2 = c1 e r+ t + c2 e r− t ,<br />
with c1, c2 ∈ R. Since x1 = x ′ 2 + x2, we obtain<br />
x 1 = c 1r+ e r+ t + c 2r− e r− t + c 1 e r+ t + c 2 e r− t ,<br />
Re-ordering terms we get the expression for the function x1, that is,<br />
x1 = c1(1 + r+) e r+ t + c2(1 + r−) e r− t .<br />
5.1.3. The Picard-Lindelöf Theorem. The Picard-Lindelöf Theorem says that every<br />
initial value problem for a certain class <strong>of</strong> <strong>differential</strong> systems always has a solution and<br />
the solution is unique. We start introducing the notion <strong>of</strong> an initial value problem for<br />
<strong>differential</strong> systems. This is a straightforward generalization <strong>of</strong> an initial value problem for<br />
a single equation.<br />
Definition 5.1.4. An initial value problem for the n × n <strong>differential</strong> system given in<br />
Definition 5.1.1 is the following: Given the set <strong>of</strong> n + 1 real numbers {x 01, · · · , x 0 n} and t 0,<br />
find a set <strong>of</strong> n functions {x1, · · · , xn} solutions <strong>of</strong> the Eqs. (5.1.1)-(5.1.2) together with the<br />
initial conditions<br />
x 1(t 0) = x 01, · · · xn(t 0) = x 0 n.<br />
Since an n × n <strong>differential</strong> system has n unknown functions, we define the initial value<br />
problem imposing n additional conditions, one for each unknown function.<br />
We now present a version <strong>of</strong> the Picard-Lindelöf Theorem, which states the following:<br />
If the functions Fi defining the <strong>differential</strong> system are differentiable both in t and in the<br />
arguments where we place the unknown functions, then the initial value problem in Definition<br />
5.1.4 has a unique solution for every real numbers {x01, · · · , x0 n} and t0. This is not the<br />
strongest version <strong>of</strong> the Theorem but it is simpler to describe in these notes. The strongest<br />
version requires that Fi be only continuous in t and a condition between continuity and<br />
differentiability, called Lipschitz continuous, in the arguments where we place the unknown<br />
functions. Charles Émile Picard presented in 1890 a successive approximation method to<br />
find solutions to <strong>differential</strong> <strong>equations</strong> and Ernst Lindelöf applied this method in 1894 to<br />
find solutions to Eqs. (5.1.1)-(5.1.2).<br />
⊳
152 G. NAGY – ODE july 2, 2013<br />
Theorem 5.1.5 (Picard-Lindelöf). If D ⊂ R×R n is a non-empty open set, and the functions<br />
Fi : D → R are differentiable, i = 1, · · · , n, then for every point (t0, x01, · · · , x0 n) ∈ D<br />
there exists an interval (t1, t2) ⊂ R containing t0 such that there exists a unique solution set<br />
{x1, · · · , xn}, with xi : (t1, t2) → R, <strong>of</strong> the initial value problem<br />
x ′ <br />
1(t) = F1 t, x1(t), · · · , xn(t) , · · · x ′ <br />
n(t) = Fn t, x1(t), · · · , xn(t) ,<br />
x 1(t 0) = x 01, · · · xn(t 0) = x 0 n.<br />
Although the pro<strong>of</strong> <strong>of</strong> this result involves techniques beyond the scope <strong>of</strong> these notes, the<br />
main ideas <strong>of</strong> the pro<strong>of</strong> are not difficult to understand, specially using vector notation. That<br />
is why we show these ideas after we finished our review <strong>of</strong> Linear Algebra.<br />
The Picard-Lindelöf Theorem applies to linear <strong>differential</strong> systems introduced in Definition<br />
5.1.2, since these systems are a particular case <strong>of</strong> the <strong>differential</strong> systems in Definition<br />
5.1.1. Nevertheless, in Section 5.6 we present a simpler pro<strong>of</strong> <strong>of</strong> Picard-Lindelöf<br />
Theorem for linear, constant coefficients, <strong>differential</strong> systems in the case that the equation<br />
coefficients satisfy an extra condition, called diagonalizability. This is a rather particular<br />
type <strong>of</strong> systems but they are fairly common in applications. This simple pro<strong>of</strong> follows the<br />
main idea we used to solve Example 5.1.7. To understand the diagonalizability condition is<br />
why we need to review several notions from Linear Algebra. We dedicate few Sections to<br />
such review.
5.1.4. Exercises.<br />
5.1.1.- . 5.1.2.- .<br />
G. NAGY – ODE July 2, 2013 153
154 G. NAGY – ODE july 2, 2013<br />
5.2. Systems <strong>of</strong> algebraic <strong>equations</strong><br />
We said in the previous Section that one way to solve a linear <strong>differential</strong> system is to<br />
transform the original system and unknowns into a decoupled system, solve the decoupled<br />
system for the new unknowns, and then transform back the solution to the original unknowns.<br />
These transformations, to and from, the decoupled system are the key steps to<br />
solve a linear <strong>differential</strong> system. One way to understand these transformations is using the<br />
ideas and notation from Linear Algebra. This Section is a review <strong>of</strong> concepts from Linear<br />
Algebra needed to introduce the transformations mentioned above. We start introducing<br />
an algebraic linear system, we then introduce matrices and matrix operations, column vectors<br />
and matrix vector products. The transformations on a system <strong>of</strong> <strong>differential</strong> <strong>equations</strong><br />
mentioned above will be introduced in a later Section, when we study the eigenvalues and<br />
eigenvectors <strong>of</strong> a square matrix.<br />
5.2.1. Linear algebraic systems. One could say that the set <strong>of</strong> results we call Linear<br />
Algebra originated with the study <strong>of</strong> linear systems <strong>of</strong> algebraic <strong>equations</strong>. Our review <strong>of</strong><br />
elementary concepts from Linear Algebra starts with a study <strong>of</strong> these type <strong>of</strong> <strong>equations</strong>.<br />
Definition 5.2.1. An n × n system <strong>of</strong> linear algebraic <strong>equations</strong> is the following:<br />
Given constants aij and bi, where i, j = 1 · · · , n 1, find the constants xj solutions <strong>of</strong><br />
a11x1 + · · · + a1nxn = b1, (5.2.1)<br />
.<br />
an1x1 + · · · + annxn = bn. (5.2.2)<br />
The system is called homogeneous iff all sources vanish, that is, b1 = · · · = bn = 0.<br />
Example 5.2.1:<br />
(a) A 2 × 2 linear system on the unknowns x 1 and x 2 is the following:<br />
2x1 − x2 = 0,<br />
−x1 + 2x2 = 3.<br />
(b) A 3 × 3 linear system on the unknowns x1, x2 and x3 is the following:<br />
x 1 + 2x 2 + x 3 = 1,<br />
−3x 1 + x 2 + 3x 3 = 24,<br />
x 2 − 4x 3 = −1. ⊳<br />
One way to find a solution to an n × n linear system is by substitution. Compute x1<br />
from the first equation and introduce it into all the other <strong>equations</strong>. Then compute x2 from<br />
this new second equation and introduce it into all the remaining <strong>equations</strong>. Repeat this<br />
procedure till the last equation, where one finally obtains xn. Then substitute back and<br />
find all the xi, for i = 1, · · · , n − 1. A computational more efficient way to find a solution<br />
is to perform Gauss elimination operations on the augmented matrix <strong>of</strong> the system. Since<br />
matrix notation will simplify calculations, it is convenient we spend some time on this. We<br />
start with the basic definitions.<br />
Definition 5.2.2. An m × n matrix, A, is an array <strong>of</strong> numbers<br />
⎡<br />
⎤<br />
a11 · · · a1n<br />
⎢<br />
⎥ m rows,<br />
A = ⎣ . . ⎦ ,<br />
aij ∈ C,<br />
n columns,<br />
am1 · · · amn<br />
where i = 1, · · · , m, and j = 1, · · · , n. An n × n matrix is called a square matrix.
G. NAGY – ODE July 2, 2013 155<br />
Example 5.2.2:<br />
(a) Examples <strong>of</strong> 2 × 2, 2 × 3, 3 × 2 real-valued matrices, and a 2 × 2 complex-valued matrix:<br />
⎡ ⎤<br />
<br />
1 2<br />
<br />
1 2<br />
1 2 3<br />
A = , B = , C = ⎣3<br />
4⎦<br />
1 + i 2 − i<br />
, D =<br />
.<br />
3 4<br />
4 5 6<br />
3 4i<br />
5 6<br />
(b) The coefficients <strong>of</strong> the algebraic linear systems in Example 5.2.1 can be grouped in<br />
matrices, as follows,<br />
<br />
2x1 − x2 = 0,<br />
2<br />
⇒ A =<br />
−x1 + 2x2 = 3,<br />
−1<br />
<br />
−1<br />
.<br />
2<br />
⎫<br />
x1 + 2x2 + x3 = 1,<br />
⎡<br />
⎪⎬<br />
1<br />
−3x1 + x2 + 3x3 = 24, ⇒ A = ⎣−3<br />
⎪⎭<br />
x2 − 4x3 = −1.<br />
0<br />
2<br />
1<br />
1<br />
⎤<br />
1<br />
3 ⎦ .<br />
−4<br />
The particular case <strong>of</strong> an m × 1 matrix is called an m-vector.<br />
Definition 5.2.3. An m-vector, v, is the array <strong>of</strong> numbers v =<br />
components vi ∈ C, with i = 1, · · · , m.<br />
⎡<br />
⎢<br />
⎣<br />
v1<br />
.<br />
vm<br />
⎤<br />
⊳<br />
⎥<br />
⎦, where the vector<br />
Example 5.2.3: The unknowns <strong>of</strong> the algebraic linear systems in Example 5.2.1 can be<br />
grouped in vectors, as follows,<br />
⎫<br />
x1 + 2x2 + x3 = 1,<br />
⎡<br />
⎪⎬<br />
2x1 − x2 = 0,<br />
x1<br />
⇒ x = . −3x1 + x2 + 3x3 = 24, ⇒ x = ⎣<br />
−x1 + 2x2 = 3,<br />
x2<br />
⎪⎭<br />
x2 − 4x3 = −1.<br />
x1<br />
⎤<br />
x2⎦<br />
.<br />
x3<br />
Definition 5.2.4. The matrix-vector product <strong>of</strong> an n × n matrix A and an n-vector x<br />
is an n-vector given by<br />
⎡<br />
⎤ ⎡ ⎤ ⎡<br />
⎤<br />
Ax =<br />
⎢<br />
⎣<br />
a11 · · · a1n<br />
.<br />
an1 · · · ann<br />
.<br />
⎥ ⎢<br />
⎦ ⎣<br />
x1<br />
.<br />
xn<br />
⎥ ⎢<br />
⎦ = ⎣<br />
a11x1 + · · · + a1nxn<br />
.<br />
an1x1 + · · · + a1nxn<br />
The matrix-vector product <strong>of</strong> an n × n matrix with an n-vector is another n-vector. This<br />
product is useful to express linear systems <strong>of</strong> algebraic <strong>equations</strong> in terms <strong>of</strong> matrices and<br />
vectors.<br />
Example 5.2.4: Find the matrix-vector products for the matrices A and vectors x in Examples<br />
5.2.2(b) and Example 5.2.3, respectively.<br />
Solution: In the 2 × 2 case we get<br />
<br />
2 −1 x1 2x1 − x2<br />
Ax =<br />
=<br />
.<br />
−1 2 x2 −x1 + 2x2<br />
In the 3 × 3 case we get,<br />
⎡<br />
1<br />
Ax = ⎣−3 2<br />
1<br />
⎤ ⎡<br />
1<br />
3 ⎦<br />
0 1 −4<br />
⎣ x1<br />
x2<br />
x3 ⎤ ⎡<br />
⎦ = ⎣ x1<br />
⎤<br />
+ 2x2 + x3<br />
−3x1 + x2 + 3x3⎦<br />
.<br />
x 2 − 4x 3<br />
⎥<br />
⎦<br />
⊳<br />
⊳
156 G. NAGY – ODE july 2, 2013<br />
Example 5.2.5: Use the matrix-vector product to express the algebraic linear system below,<br />
2x1 − x2 = 0,<br />
−x1 + 2x2 = 3.<br />
Solution: Introduce the coefficient matrix A, the unknown vector x, and the source vector<br />
b as follows,<br />
<br />
2<br />
A =<br />
−1<br />
<br />
−1<br />
,<br />
2<br />
<br />
x1<br />
x = ,<br />
<br />
0<br />
b = .<br />
3<br />
Since the matrix-vector product Ax is given by<br />
<br />
2 −1 x1 2x1 − x2<br />
Ax =<br />
=<br />
,<br />
−1 2 x2 −x1 + 2x2<br />
then we conclude that<br />
<br />
2x1 − x2 = 0,<br />
−x1 + 2x2 = 3,<br />
⇔<br />
x2<br />
<br />
2x1 − x2<br />
=<br />
−x1 + 2x2<br />
<br />
0<br />
3<br />
⇔ Ax = b.<br />
It is simple to see that the result found in the Example above can be generalized to every<br />
n × n algebraic linear system.<br />
Theorem 5.2.5. Given the algebraic linear system in Eqs. (5.2.1)-(5.2.2), introduce the<br />
coefficient matrix A, the unknown vector x, and the source vector b, as follows,<br />
⎡<br />
⎤ ⎡ ⎤ ⎡ ⎤<br />
A =<br />
⎢<br />
⎣<br />
a11 · · · a1n<br />
.<br />
an1 · · · ann<br />
.<br />
⎥ ⎢<br />
⎦ , x = ⎣<br />
Then, the algebraic linear system can be written as<br />
Ax = b.<br />
x1<br />
.<br />
xn<br />
⎥ ⎢<br />
⎦ , b = ⎣<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.2.5: From the definition <strong>of</strong> the matrix-vector product we have that<br />
⎡<br />
a11<br />
⎢<br />
Ax = ⎣ .<br />
· · ·<br />
⎤ ⎡ ⎤ ⎡<br />
⎤<br />
a1n x1 a11x1 + · · · + a1nxn<br />
⎥ ⎢ ⎥ ⎢<br />
⎥<br />
. ⎦ ⎣ . ⎦ = ⎣ . ⎦ .<br />
an1x1 + · · · + a1nxn<br />
an1 · · · ann<br />
Then, we conclude that<br />
⎫<br />
a11x1 + · · · + a1nxn = b1,<br />
⎪⎬<br />
.<br />
⎪⎭<br />
an1x1 + · · · + annxn = bn,<br />
⇔<br />
⎡<br />
⎢<br />
⎣<br />
xn<br />
a11x1 + · · · + a1nxn<br />
.<br />
an1x1 + · · · + a1nxn<br />
⎤<br />
⎥ ⎢<br />
⎦ = ⎣<br />
We introduce one last definition, which will be helpful in the next subsection.<br />
⎡<br />
b1<br />
.<br />
bn<br />
b1<br />
.<br />
bn<br />
⎥<br />
⎦ .<br />
⎤<br />
⎥<br />
⎦ ⇔ Ax = b.<br />
Definition 5.2.6. The augmented matrix <strong>of</strong> Ax = b is the n × (n + 1) matrix [A|b].<br />
The augmented matrix <strong>of</strong> an algebraic linear system contains the equation coefficients and<br />
the sources. Therefore, the augmented matrix <strong>of</strong> a linear system contains the complete<br />
information about the system.<br />
⊳
G. NAGY – ODE July 2, 2013 157<br />
Example 5.2.6: Find the augmented matrix <strong>of</strong> both the linear systems in Example 5.2.1.<br />
Solution: The coefficient matrix and source vector <strong>of</strong> the first system imply that<br />
<br />
2<br />
A =<br />
−1<br />
<br />
−1<br />
,<br />
2<br />
<br />
0<br />
b =<br />
3<br />
⇒<br />
<br />
2<br />
[A|b] =<br />
−1<br />
−1<br />
2<br />
<br />
<br />
<br />
<br />
<br />
0<br />
.<br />
3<br />
The coefficient matrix and source vector <strong>of</strong> the second system imply that<br />
⎡<br />
1<br />
A = ⎣−3<br />
0<br />
2<br />
1<br />
1<br />
⎤<br />
1<br />
3 ⎦ ,<br />
−4<br />
⎡ ⎤<br />
1<br />
b = ⎣ 24 ⎦<br />
−1<br />
⇒<br />
⎡<br />
1<br />
[A|b] = ⎣−3<br />
0<br />
2<br />
1<br />
1<br />
1<br />
3<br />
−4<br />
<br />
<br />
<br />
<br />
<br />
<br />
⎤<br />
1<br />
24 ⎦ .<br />
−1<br />
Recall that the linear combination <strong>of</strong> two vectors is defined component-wise, that is, given<br />
any numbers a, b ∈ R and any vectors x, y, their linear combination is the vector given by<br />
⎡ ⎤<br />
⎡ ⎤ ⎡ ⎤<br />
ax1 + by1 x1 y1 ⎢ ⎥<br />
⎢ ⎥ ⎢ ⎥<br />
ax + by = ⎣ . ⎦ , where x = ⎣ . ⎦ , y = ⎣ . ⎦ .<br />
axn + byn<br />
With this definition <strong>of</strong> linear combination <strong>of</strong> vectors it is simple to see that the matrix-vector<br />
product is a linear operation.<br />
Theorem 5.2.7. The matrix-vector product is a linear operation, that is, given an n × n<br />
matrix A, then for all n-vectors x, y and all numbers a, b ∈ R holds<br />
xn<br />
A(ax + by) = a Ax + b Ay. (5.2.3)<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.2.7: Just write down the matrix-vector product in components,<br />
⎡<br />
a11<br />
⎢<br />
A(ax+by) = ⎣ .<br />
· · ·<br />
⎤ ⎡ ⎤ ⎡<br />
⎤<br />
a1n ax1 + by1 a11(ax1 + by1) + · · · + a1n(axn + byn)<br />
⎥ ⎢ ⎥ ⎢<br />
⎥<br />
. ⎦ ⎣ . ⎦ = ⎣<br />
.<br />
⎦ .<br />
am1 · · · amn axn + byn an1(ax1 + by1) + · · · + ann(axn + byn)<br />
Expand the linear combinations on each component on the far right-hand side above and<br />
re-order terms as follows,<br />
⎡<br />
⎤<br />
A(ax + by) =<br />
⎢<br />
⎣<br />
Separate the right-hand side above,<br />
⎡<br />
A(ax + by) = a<br />
We then conclude that<br />
⎢<br />
⎣<br />
a (a11x 1 + · · · + a1nxn) + b (a11y 1 + · · · + a1nyn)<br />
.<br />
a (an1x 1 + · · · + annxn) + b (an1y 1 + · · · + annyn)<br />
(a11x1 + · · · + a1nxn)<br />
.<br />
(an1x1 + · · · + annxn)<br />
⎤<br />
⎡<br />
⎥ ⎢<br />
⎦ + b ⎣<br />
A(ax + by) = a Ax + b Ay.<br />
yn<br />
⎥<br />
⎦ .<br />
(a11y1 + · · · + a1nyn)<br />
.<br />
(an1y1 + · · · + annyn)<br />
This establishes the Theorem. <br />
⎤<br />
⎥<br />
⎦ .<br />
⊳
158 G. NAGY – ODE july 2, 2013<br />
5.2.2. Gauss elimination operations. We now review three operations that can be performed<br />
on an augmented matrix <strong>of</strong> a linear system. These operations change the augmented<br />
matrix <strong>of</strong> the system but they do not change the solutions <strong>of</strong> the system. The Gauss elimination<br />
operations were already known in China around 200 BC. We call them after Carl<br />
Friedrich Gauss, since he made them very popular around 1810, when he used them to study<br />
the orbit <strong>of</strong> the asteroid Pallas, giving a systematic method to solve a 6 × 6 algebraic linear<br />
system.<br />
Definition 5.2.8. The Gauss elimination operations are three operations on a matrix:<br />
(i) Adding to one row a multiple <strong>of</strong> the another;<br />
(ii) Interchanging two rows;<br />
(iii) Multiplying a row by a non-zero number.<br />
These operations are respectively represented by the symbols given in Fig. 23.<br />
a<br />
a = 0<br />
Figure 23. A representation <strong>of</strong> the Gauss elimination operations.<br />
As we said above, the Gauss elimination operations change the coefficients <strong>of</strong> the augmented<br />
matrix <strong>of</strong> a system but do not change its solution. Two systems <strong>of</strong> linear <strong>equations</strong> having<br />
the same solutions are called equivalent. It can be shown that there is an algorithm using<br />
these operations that transforms any n × n linear system into an equivalent system where<br />
the solutions are explicitly given.<br />
Example 5.2.7: Find the solution to the 2 × 2 linear system given in Example 5.2.1 using<br />
the Gauss elimination operations.<br />
Solution: Consider the augmented matrix <strong>of</strong> the 2 × 2 linear system in Example (5.2.1),<br />
and perform the following Gauss elimination operations,<br />
<br />
2 −1 <br />
0 2 −1 0<br />
−1 2 →<br />
2 −1 <br />
3 −2 4 → <br />
6 0 3 <br />
<br />
0 2<br />
→<br />
6 0<br />
−1<br />
1<br />
<br />
<br />
<br />
<br />
<br />
0<br />
→<br />
2<br />
<br />
2<br />
0<br />
0<br />
1<br />
<br />
<br />
<br />
<br />
<br />
2 1<br />
→<br />
2 0<br />
0<br />
1<br />
<br />
<br />
<br />
<br />
<br />
1<br />
2<br />
⇔<br />
<br />
x1 + 0 = 1<br />
0 + x2 = 2<br />
⇔<br />
<br />
x1 = 1<br />
x2 = 2<br />
⊳<br />
Example 5.2.8: Find the solution to the 3 × 3 linear system given in Example 5.2.1 using<br />
the Gauss elimination operations<br />
Solution: Consider the augmented matrix <strong>of</strong> the 3 × 3 linear system in Example 5.2.1 and<br />
perform the following Gauss elimination operations,<br />
⎡<br />
⎤ ⎡<br />
⎤ ⎡<br />
1 2 1 <br />
1 1 2 1 <br />
1 1<br />
⎣−3<br />
1 3 <br />
24 ⎦ → ⎣0<br />
7 6 27 ⎦<br />
→ ⎣0<br />
0 1 −4 −1 0 1 −4 −1 0<br />
2<br />
1<br />
7<br />
1<br />
−4<br />
6<br />
<br />
<br />
<br />
<br />
<br />
<br />
⎤<br />
1<br />
−1⎦<br />
,<br />
27<br />
⎡<br />
1<br />
⎣0<br />
0<br />
0<br />
1<br />
0<br />
9<br />
−4<br />
34<br />
<br />
<br />
<br />
<br />
<br />
<br />
⎤ ⎡<br />
3 1<br />
−1⎦<br />
→ ⎣0<br />
34 0<br />
0<br />
1<br />
0<br />
9<br />
−4<br />
1<br />
<br />
<br />
<br />
<br />
<br />
<br />
⎤ ⎡<br />
3 1<br />
−1⎦<br />
→ ⎣0<br />
1 0<br />
0<br />
1<br />
0<br />
0<br />
0<br />
1<br />
<br />
<br />
<br />
<br />
<br />
<br />
⎤<br />
−6<br />
3 ⎦<br />
1<br />
⇒<br />
⎧<br />
⎪⎨<br />
x1 = −6,<br />
x2 = 3,<br />
⎪⎩<br />
x3 = 1.<br />
⊳
G. NAGY – ODE July 2, 2013 159<br />
In the last augmented matrix on both Examples 5.2.7 and 5.2.8 the solution is given explicitly.<br />
This is not always the case with every augmented matrix. A precise way to define<br />
the final augmented matrix in the Gauss elimination method is captured in the notion <strong>of</strong><br />
echelon form and reduced echelon form <strong>of</strong> a matrix.<br />
Definition 5.2.9. An m × n matrix is in echelon form iff the following conditions hold:<br />
(i) The zero rows are located at the bottom rows <strong>of</strong> the matrix;<br />
(ii) The first non-zero coefficient on a row is always to the right <strong>of</strong> the first non-zero<br />
coefficient <strong>of</strong> the row above it.<br />
The pivot coefficient is the first non-zero coefficient on every non-zero row in a matrix in<br />
echelon form.<br />
Example 5.2.9: The 6 × 8, 3 × 5 and 3 × 3 matrices given below are in echelon form, where<br />
the ∗ means any non-zero number and pivots are highlighted.<br />
⎡<br />
⎤<br />
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗<br />
⎢<br />
⎢0<br />
0 ∗ ∗ ∗ ∗ ∗ ∗⎥<br />
⎡<br />
⎤<br />
⎥<br />
⎢<br />
⎢0<br />
0 0 ∗ ∗ ∗ ∗ ∗⎥<br />
∗ ∗ ∗ ∗ ∗<br />
⎥<br />
⎢<br />
⎢0<br />
0 0 0 0 0 ∗ ∗⎥<br />
, ⎣0 0 ∗ ∗ ∗⎦<br />
,<br />
⎥<br />
⎣0<br />
0 0 0 0 0 0 0⎦<br />
0 0 0 0 0<br />
⎡<br />
∗<br />
⎣0 0<br />
∗<br />
∗<br />
0<br />
⎤<br />
∗<br />
∗⎦<br />
.<br />
∗<br />
0 0 0 0 0 0 0 0<br />
⊳<br />
Example 5.2.10: The following matrices are in echelon form, with pivot highlighted.<br />
⎡ ⎤<br />
2 1 1<br />
1 3 2 3 2<br />
,<br />
, ⎣0 3 4⎦<br />
.<br />
0 1 0 4 −2<br />
0 0 0<br />
Definition 5.2.10. An m × n matrix is in reduced echelon form iff the matrix is in<br />
echelon form and the following two conditions hold:<br />
(i) The pivot coefficient is equal to 1;<br />
(ii) The pivot coefficient is the only non-zero coefficient in that column.<br />
We denote by EA a reduced echelon form <strong>of</strong> a matrix A.<br />
Example 5.2.11: The 6×8, 3×5 and 3×3 matrices given below are in echelon form, where<br />
the ∗ means any non-zero number and pivots are highlighted.<br />
⎡<br />
⎤<br />
1 ∗ 0 0 ∗ ∗ 0 ∗<br />
⎢<br />
⎢0<br />
0 1 0 ∗ ∗ 0 ∗⎥<br />
⎡<br />
⎤<br />
⎥<br />
⎢<br />
⎢0<br />
0 0 1 ∗ ∗ 0 ∗⎥<br />
1 ∗ 0 ∗ ∗<br />
⎥<br />
⎢<br />
⎢0<br />
0 0 0 0 0 1 ∗⎥<br />
, ⎣0<br />
0 1 ∗ ∗⎦<br />
,<br />
⎥<br />
⎣0<br />
0 0 0 0 0 0 0⎦<br />
0 0 0 0 0<br />
⎡<br />
1<br />
⎣0<br />
0<br />
0<br />
1<br />
0<br />
⎤<br />
0<br />
0⎦<br />
.<br />
1<br />
0 0 0 0 0 0 0 0<br />
⊳<br />
Example 5.2.12: And the following matrices are not only in echelon form but also in<br />
reduced echelon form; again, pivot coefficients are highlighted.<br />
⎡ ⎤<br />
1 0 0<br />
1 0 1 0 4<br />
,<br />
, ⎣0<br />
1 0⎦<br />
.<br />
0 1 0 1 5<br />
0 0 0<br />
⊳<br />
Summarizing, the Gauss elimination operations can transform any matrix into reduced<br />
echelon form. Once the augmented matrix <strong>of</strong> a linear system is written in reduced echelon<br />
form, it is not difficult to decide whether the system has solutions or not.<br />
⊳
160 G. NAGY – ODE july 2, 2013<br />
Example 5.2.13: Use Gauss operations to find the solution <strong>of</strong> the linear system<br />
2x1 − x2 = 0,<br />
− 1<br />
2 x1 + 1<br />
4 x2 = − 1<br />
4 .<br />
Solution: We find the system augmented matrix and perform appropriate Gauss elimination<br />
operations,<br />
<br />
2 −1 0<br />
− 1<br />
<br />
<br />
2 −1 0<br />
1 1 →<br />
2 −1 0<br />
2 4 − 4 −2 1 → <br />
−1 0 0 1<br />
From the last augmented matrix above we see that the original linear system has the same<br />
solutions as the linear system given by<br />
2x1 − x2 = 0,<br />
0 = 1.<br />
Since the latter system has no solutions, the original system has no solutions. ⊳<br />
The situation shown in Example 5.2.13 is true in general. If the augmented matrix [A|b] <strong>of</strong><br />
an algebraic linear system is transformed by Gauss operations into the augmented matrix<br />
[ Ã|˜ b] having a row <strong>of</strong> the form [0, · · · , 0|1], then the original algebraic linear system Ax = b<br />
has no solution.<br />
Example 5.2.14: Find all vectors b such that the system Ax = b has solutions, where<br />
⎡<br />
1<br />
A = ⎣−1 −2<br />
1<br />
⎤<br />
3<br />
−2⎦<br />
,<br />
⎡<br />
b = ⎣<br />
2 −1 3<br />
b1<br />
⎤<br />
b2⎦<br />
.<br />
Solution: We do not need to write down the algebraic linear system, we only need its<br />
augmented matrix,<br />
⎡<br />
1<br />
[A|b] = ⎣−1<br />
2<br />
−2<br />
1<br />
−1<br />
3<br />
−2<br />
3<br />
<br />
<br />
<br />
<br />
<br />
<br />
⎤ ⎡<br />
b1 1<br />
b2⎦<br />
→ ⎣0<br />
b3 2<br />
−2<br />
−1<br />
−1<br />
3<br />
1<br />
3<br />
<br />
<br />
<br />
<br />
<br />
<br />
⎤<br />
b1<br />
b1 + b2⎦<br />
→<br />
b3<br />
⎡<br />
1<br />
⎣0<br />
0<br />
−2<br />
1<br />
3<br />
3<br />
−1<br />
−3<br />
<br />
<br />
<br />
<br />
<br />
<br />
⎤ ⎡<br />
b1 1<br />
−b1 − b2⎦<br />
→ ⎣0<br />
b3 − 2b1 0<br />
−2<br />
1<br />
0<br />
3<br />
−1<br />
0<br />
<br />
<br />
<br />
<br />
<br />
<br />
⎤<br />
b1<br />
−b1 − b2 ⎦ .<br />
b3 + b1 + 3b2<br />
Therefore, the linear system Ax = b has solutions<br />
⇔ the source vector satisfies the equation<br />
holds b1 + 3b2 + b3 = 0.<br />
That is, every source vector b that lie on the<br />
plane normal to the vector n is a source vector<br />
such that the linear system Ax = b has<br />
solution, where<br />
⎡ ⎤<br />
1<br />
n = ⎣3⎦<br />
.<br />
1<br />
⊳<br />
b3<br />
b 1<br />
b<br />
3<br />
b<br />
n<br />
b<br />
2
G. NAGY – ODE July 2, 2013 161<br />
5.2.3. Linearly dependence. We generalize the idea <strong>of</strong> two vectors lying on the same line,<br />
and three vectors lying on the same plane, to an arbitrary number <strong>of</strong> vectors.<br />
Definition 5.2.11. A set <strong>of</strong> vectors {v 1, · · · , vk}, with k 1 is called linearly dependent<br />
iff there exists constants c1, · · · , ck, with at least one <strong>of</strong> them non-zero, such that<br />
c1 v1 + · · · + ck vk = 0. (5.2.4)<br />
The set <strong>of</strong> vectors is called linearly independent iff it is not linearly dependent, that is,<br />
the only constants c1, · · · , ck that satisfy Eq. (5.2.4) are given by c1 = · · · = ck = 0.<br />
In other words, a set <strong>of</strong> vectors is linearly dependent iff one <strong>of</strong> the vectors is a linear combination<br />
<strong>of</strong> the other vectors. When this is not possible, the set is called linearly independent.<br />
Example 5.2.15: Show that the following set <strong>of</strong> vectors is linearly dependent,<br />
⎡<br />
⎣ 1<br />
⎤ ⎡<br />
2⎦<br />
, ⎣<br />
3<br />
3<br />
⎤ ⎡<br />
2⎦<br />
, ⎣<br />
1<br />
−1<br />
⎤<br />
2 ⎦<br />
5<br />
<br />
,<br />
and express one <strong>of</strong> the vectors as a linear combination <strong>of</strong> the other two.<br />
Solution: We need to find constants c1, c2, and c3 solutions <strong>of</strong> the equation<br />
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤<br />
1 3 −1 0 1 3 −1 c1 0<br />
⎣2⎦<br />
c1 + ⎣2⎦<br />
c2 + ⎣ 2 ⎦ c3 = ⎣0⎦<br />
⇔ ⎣2<br />
2 2 ⎦ ⎣c<br />
⎦ 2 + ⎣0⎦<br />
.<br />
3 1 5 0 3 1 5 c3 0<br />
The solution to this linear system can be obtained with Gauss elimination operations,<br />
⎡ ⎤ ⎡<br />
⎤ ⎡ ⎤ ⎡ ⎤ ⎧<br />
1 3 −1 1 3 −1 1 3 −1 1 0 2 ⎪⎨<br />
c1 = −2c3, ⎣2 2 2 ⎦ → ⎣0<br />
−4 4 ⎦ → ⎣0<br />
1 −1⎦<br />
→ ⎣0<br />
1 −1⎦<br />
⇒ c2 = c3, ⎪⎩<br />
3 1 5 0 −8 8 0 1 −1 0 0 0<br />
c3 = free.<br />
Since there are non-zero constants c1, c2, c3 solutions to the linear system above, the vectors<br />
are linearly dependent. Choosing c3 = −1 we obtain the third vector as a linear combination<br />
<strong>of</strong> the other two vectors,<br />
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤<br />
1 3 −1 0<br />
2 ⎣2⎦<br />
− ⎣2⎦<br />
− ⎣ 2 ⎦ = ⎣0⎦<br />
⇔<br />
3 1 5 0<br />
−1 1 3<br />
⎣ 2 ⎦ = 2 ⎣2⎦<br />
− ⎣2⎦<br />
.<br />
5 3 1<br />
⊳
162 G. NAGY – ODE july 2, 2013<br />
5.2.4. Exercises.<br />
5.2.1.- . 5.2.2.- .
G. NAGY – ODE July 2, 2013 163<br />
5.3. Matrix algebra<br />
The matrix-vector product introduced in Section 5.2 implies that an n × n matrix A is a<br />
function A : R n → R n . This idea leads to introduce matrix operations, like the operations<br />
introduced for functions f : R → R. These operations include linear combinations <strong>of</strong><br />
matrices; the composition <strong>of</strong> matrices, also called matrix multiplication; the inverse <strong>of</strong> a<br />
square matrix; and the determinant <strong>of</strong> a square matrix. These operations will be needed in<br />
later Sections when we study systems <strong>of</strong> <strong>differential</strong> <strong>equations</strong>.<br />
5.3.1. A matrix is a function. The matrix-vector product leads to the interpretation<br />
that an n × n matrix A is a function. If we denote by R n the space <strong>of</strong> all n-vectors, we see<br />
that the matrix-vector product associates to the n-vector x the unique n-vector y = Ax.<br />
Therefore the matrix A determines a function A : R n → R n .<br />
Example 5.3.1: Describe the action on R2 <strong>of</strong> the function given by the 2 × 2 matrix<br />
<br />
0 1<br />
A = . (5.3.1)<br />
1 0<br />
Solution: The action <strong>of</strong> this matrix on an arbitrary element x ∈ R2 is given below,<br />
<br />
0 1 x1 x2<br />
Ax =<br />
= .<br />
1 0<br />
Therefore, this matrix interchanges the components x 1 and x 2 <strong>of</strong> the vector x. It can be<br />
seen in the first picture in Fig. 24 that this action can be interpreted as a reflection on the<br />
plane along the line x1 = x2. ⊳<br />
x<br />
2<br />
Ax<br />
x = x<br />
1<br />
2<br />
x<br />
x<br />
1<br />
Figure 24. Geometrical meaning <strong>of</strong> the function determined by the matrix<br />
in Eq. (5.3.1) and the matrix in Eq. (5.3.2), respectively.<br />
Example 5.3.2: Describe the action on R2 <strong>of</strong> the function given by the 2 × 2 matrix<br />
<br />
0 −1<br />
A = . (5.3.2)<br />
1 0<br />
Solution: The action <strong>of</strong> this matrix on an arbitrary element x ∈ R2 is given below,<br />
<br />
0 −1 x1 −x2<br />
Ax =<br />
= .<br />
1 0<br />
In order to understand the action <strong>of</strong> this matrix, we give the following particular cases:<br />
<br />
0 −1 1 0 0 −1 1 −1 0 −1 −1 0<br />
= ,<br />
= ,<br />
= .<br />
1 0 0 1 1 0 1 1 1 0 0 −1<br />
x2<br />
x2<br />
Ay<br />
z<br />
x1<br />
x1<br />
x<br />
2<br />
Ax<br />
Az<br />
x<br />
y<br />
x<br />
1
164 G. NAGY – ODE july 2, 2013<br />
These cases are plotted in the second figure on Fig. 24, and the vectors are called x, y and<br />
z, respectively. We therefore conclude that this matrix produces a ninety degree counterclockwise<br />
rotation <strong>of</strong> the plane. ⊳<br />
An example <strong>of</strong> a scalar-valued function is f : R → R. We have seen here that an n×n matrix<br />
A is a function A : R n → R n . Therefore, one can think an n × n matrix as a generalization<br />
<strong>of</strong> the concept <strong>of</strong> function from R to R n , for any positive integer n. It is well-known how to<br />
define several operations on scalar-valued functions, like linear combinations, compositions,<br />
and the inverse function. Therefore, it is reasonable to ask if these operation on scalarvalued<br />
functions can be generalized as well to matrices. The answer is yes, and the study<br />
<strong>of</strong> these and other operations is the subject <strong>of</strong> the rest <strong>of</strong> this Section.<br />
5.3.2. Matrix operations. The linear combination <strong>of</strong> matrices refers to the addition <strong>of</strong> two<br />
matrices and the multiplication <strong>of</strong> a matrix by scalar. Linear combinations <strong>of</strong> matrices are<br />
defined component by component. For this reason we introduce the component notation<br />
for matrices and vectors. We denote an m × n matrix by A = [Aij], where Aij are the<br />
components <strong>of</strong> matrix A, with i = 1, · · · , m and j = 1, · · · , n. Analogously, an n-vector is<br />
denoted by v = [vj], where vj are the components <strong>of</strong> the vector v. We also introduce the<br />
notation F = {R, C}, that is, the set F can be the real numbers or the complex numbers.<br />
Definition 5.3.1. Let A = [Aij] and B = [Bij] be m × n matrices in F m,n and a, b be<br />
numbers in F. The linear combination <strong>of</strong> A and B is also and m × n matrix in F m,n ,<br />
denoted as a A + b B, and given by<br />
a A + b B = [a Aij + b Bij].<br />
The particular case where a = b = 1 corresponds to the addition <strong>of</strong> two matrices, and the<br />
particular case <strong>of</strong> b = 0 corresponds to the multiplication <strong>of</strong> a matrix by a number, that is,<br />
A + B = [Aij + Bij], a A = [a Aij].<br />
<br />
1 2 2 3<br />
Example 5.3.3: Find the A + B, where A = , B = .<br />
3 4 5 1<br />
Solution: The addition <strong>of</strong> two equal size matrices is performed component-wise:<br />
<br />
1 2 2 3 (1 + 2) (2 + 3) 3 5<br />
A + B = + =<br />
= .<br />
3 4 5 1 (3 + 5) (4 + 1) 8 5<br />
<br />
1 2 1 2 3<br />
Example 5.3.4: Find the A + B, where A = , B = .<br />
3 4 4 5 6<br />
Solution: The matrices have different sizes, so their addition is not defined. ⊳<br />
<br />
1<br />
Example 5.3.5: Find 2A and A/3, where A =<br />
2<br />
3<br />
4<br />
<br />
5<br />
.<br />
6<br />
Solution: The multiplication <strong>of</strong> a matrix by a number is done component-wise, therefore<br />
<br />
1<br />
2A = 2<br />
2<br />
3<br />
4<br />
<br />
5 2<br />
=<br />
6 4<br />
6<br />
8<br />
<br />
10<br />
,<br />
12<br />
<br />
A 1 1<br />
=<br />
3 3 2<br />
3<br />
4<br />
⎡1<br />
<br />
5 ⎢3<br />
= ⎢<br />
6 ⎣<br />
2<br />
3<br />
1<br />
4<br />
3<br />
5⎤<br />
3 ⎥<br />
⎦<br />
2<br />
.<br />
⊳<br />
Since matrices are generalizations <strong>of</strong> scalar-valued functions, one can define operations<br />
on matrices that, unlike linear combinations, have no analogs on scalar-valued functions.<br />
One <strong>of</strong> such operations is the transpose <strong>of</strong> a matrix, which is a new matrix with the rows<br />
and columns interchanged.<br />
⊳
G. NAGY – ODE July 2, 2013 165<br />
Definition 5.3.2. The transpose <strong>of</strong> a matrix A = [Aij] ∈ Fm,n is the matrix denoted as<br />
AT = (AT <br />
n,m T<br />
)kl ∈ F , with its components given by A <br />
= Alk.<br />
kl<br />
<br />
1 3 5<br />
Example 5.3.6: Find the transpose <strong>of</strong> the 2 × 3 matrix A = .<br />
2 4 6<br />
Solution: Matrix A has components Aij with i = 1, 2 and j = 1, 2, 3. Therefore, its<br />
transpose has components (A T )ji = Aij, that is, A T has three rows and two columns,<br />
A T ⎡<br />
1<br />
⎤<br />
2<br />
= ⎣3<br />
4⎦<br />
.<br />
5 6<br />
If a matrix has complex-valued coefficients, then the conjugate <strong>of</strong> a matrix can be defined<br />
as the conjugate <strong>of</strong> each component.<br />
Definition 5.3.3. The complex conjugate <strong>of</strong> a matrix A = [Aij] ∈ Fm,n is the matrix<br />
A = <br />
m,n<br />
Aij ∈ F .<br />
Example 5.3.7: A matrix A and its conjugate is given below,<br />
<br />
1<br />
A =<br />
−i<br />
<br />
2 + i<br />
,<br />
3 − 4i<br />
⇔<br />
<br />
1<br />
A =<br />
i<br />
<br />
2 − i<br />
.<br />
3 + 4i<br />
Example 5.3.8: A matrix A has real coefficients iff A = A; It has purely imaginary coefficients<br />
iff A = −A. Here are examples <strong>of</strong> these two situations:<br />
<br />
<br />
1 2<br />
1 2<br />
A =<br />
⇒ A = = A;<br />
3 4<br />
3 4<br />
<br />
<br />
i 2i<br />
−i −2i<br />
A =<br />
⇒ A =<br />
= −A.<br />
3i 4i<br />
−3i −4i<br />
⊳<br />
Definition 5.3.4. The adjoint <strong>of</strong> a matrix A ∈ Fm,n is the matrix A∗ = A T<br />
∈ Fn,m .<br />
Since A T = (AT ), the order <strong>of</strong> the operations does not change the result, that is why there<br />
is no parenthesis in the definition <strong>of</strong> A∗ .<br />
Example 5.3.9: A matrix A and its adjoint is given below,<br />
<br />
1<br />
A =<br />
−i<br />
<br />
2 + i<br />
,<br />
3 − 4i<br />
⇔ A ∗ <br />
1<br />
=<br />
2 − i<br />
<br />
i<br />
.<br />
3 + 4i<br />
The transpose, conjugate and adjoint operations are useful to specify certain classes <strong>of</strong><br />
matrices with particular symmetries. Here we introduce few <strong>of</strong> these classes.<br />
Definition 5.3.5. An n × n matrix A is called:<br />
(a) symmetric iff holds A = A T ;<br />
(b) skew-symmetric iff holds A = −A T ;<br />
(c) Hermitian iff holds A = A ∗ ;<br />
(d) skew-Hermitian iff holds A = −A ∗ .<br />
⊳<br />
⊳<br />
⊳
166 G. NAGY – ODE july 2, 2013<br />
Example 5.3.10: We present examples <strong>of</strong> each <strong>of</strong> the classes introduced in Def. 5.3.5.<br />
Part (a): Matrices A and B are symmetric. Notice that A is also Hermitian, while B is<br />
not Hermitian,<br />
⎡<br />
1 2<br />
⎤<br />
3<br />
A = ⎣2<br />
7 4⎦<br />
= A<br />
3 4 8<br />
T ⎡<br />
1 2 + 3i<br />
⎤<br />
3<br />
, B = ⎣2<br />
+ 3i 7 4i⎦<br />
= B<br />
3 4i 8<br />
T .<br />
Part (b): Matrix C is skew-symmetric,<br />
⎡<br />
0<br />
C = ⎣ 2<br />
−2<br />
0<br />
⎤<br />
3<br />
−4⎦<br />
⇒ C<br />
−3 4 0<br />
T ⎡<br />
0<br />
= ⎣−2 2<br />
0<br />
⎤<br />
−3<br />
4 ⎦ = −C.<br />
3 −4 0<br />
Notice that the diagonal elements in a skew-symmetric matrix must vanish, since Cij = −Cji<br />
in the case i = j means Cii = −Cii, that is, Cii = 0.<br />
Part (c): Matrix D is Hermitian but is not symmetric:<br />
⎡<br />
1 2 + i 3<br />
⎤<br />
D = ⎣2<br />
− i 7 4 + i⎦<br />
⇒ D<br />
3 4 − i 8<br />
T ⎡<br />
1 2 − i 3<br />
⎤<br />
= ⎣2<br />
+ i 7 4 − i⎦<br />
= D,<br />
3 4 + i 8<br />
however,<br />
D ∗ = D T<br />
⎡<br />
1<br />
= ⎣2 − i<br />
2 + i<br />
7<br />
⎤<br />
3<br />
4 + i⎦<br />
= D.<br />
3 4 − i 8<br />
Notice that the diagonal elements in a Hermitian matrix must be real numbers, since the<br />
condition Aij = Aji in the case i = j implies Aii = Aii, that is, 2iIm(Aii) = Aii − Aii = 0.<br />
We can also verify what we said in part (a), matrix A is Hermitian since A∗ = A T<br />
= AT Part (d): The following matrix E is skew-Hermitian:<br />
= A.<br />
⎡<br />
i<br />
E = ⎣−2 + i<br />
2 + i<br />
7i<br />
⎤<br />
−3<br />
4 + i⎦<br />
⇒ E<br />
3 −4 + i 8i<br />
T ⎡<br />
i<br />
= ⎣2 + i<br />
−2 + i<br />
7i<br />
⎤<br />
3<br />
−4 + i⎦<br />
−3 4 + i 8i<br />
therefore,<br />
E ∗ = E T<br />
⎡<br />
−i<br />
⎣2<br />
− i<br />
−2 − i<br />
−7i<br />
⎤<br />
3<br />
−4 − i⎦<br />
= −E.<br />
−3 4 − i −8i<br />
A skew-Hermitian matrix has purely imaginary elements in its diagonal, and the <strong>of</strong>f diagonal<br />
elements have skew-symmetric real parts with symmetric imaginary parts. ⊳<br />
The trace <strong>of</strong> a square matrix is a number, the sum <strong>of</strong> all the diagonal elements <strong>of</strong> the matrix.<br />
Definition 5.3.6. The trace <strong>of</strong> a square matrix A = <br />
n,n<br />
Aij ∈ F , denoted as tr (A) ∈ F,<br />
is the sum <strong>of</strong> its diagonal elements, that is, the scalar given by tr (A) = A11 + · · · + Ann.<br />
⎡<br />
1<br />
Example 5.3.11: Find the trace <strong>of</strong> the matrix A = ⎣4 2<br />
5<br />
⎤<br />
3<br />
6⎦.<br />
7 8 9<br />
Solution: We only have to add up the diagonal elements:<br />
tr (A) = 1 + 5 + 9 ⇒ tr (A) = 15.<br />
⊳
G. NAGY – ODE July 2, 2013 167<br />
The operation <strong>of</strong> matrix multiplication originates in the composition <strong>of</strong> functions. We<br />
call it matrix multiplication instead <strong>of</strong> matrix composition because it reduces to the multiplication<br />
<strong>of</strong> real numbers in the case <strong>of</strong> 1 × 1 real matrices. Unlike the multiplication <strong>of</strong><br />
real numbers, the product <strong>of</strong> general matrices is not commutative, that is, AB = BA in<br />
the general case. This property reflects the fact that the composition <strong>of</strong> two functions is a<br />
non-commutative operation.<br />
Definition 5.3.7. The matrix multiplication <strong>of</strong> the m × n matrix A = [Aij] and the<br />
n × ℓ matrix B = [Bjk], where i = 1, · · · , m, j = 1, · · · , n and k = 1, · · · , ℓ, is the m × ℓ<br />
matrix AB given by<br />
n<br />
(AB)ik = AijBjk. (5.3.3)<br />
j=1<br />
The product is not defined for two arbitrary matrices, since the size <strong>of</strong> the matrices is<br />
important: The numbers <strong>of</strong> columns in the first matrix must match the numbers <strong>of</strong> rows in<br />
the second matrix.<br />
A times B defines AB<br />
m × n n × ℓ m × ℓ<br />
<br />
2<br />
Example 5.3.12: Compute AB, where A =<br />
−1<br />
<br />
−1<br />
and B =<br />
2<br />
<br />
3 0<br />
.<br />
2 −1<br />
Solution: The component (AB)11 = 4 is obtained from the first row in matrix A and the<br />
first column in matrix B as follows:<br />
<br />
2 −1 3 0 4<br />
=<br />
−1 2 2 −1 1<br />
<br />
1<br />
,<br />
−2<br />
(2)(3) + (−1)(2) = 4;<br />
The component (AB)12 = −1 is obtained as follows:<br />
<br />
2 −1 3 0 4 1<br />
= , (2)(0) + (−1)(1) = −1;<br />
−1 2 2 −1 1 −2<br />
The component (AB)21 = 1 is obtained as follows:<br />
<br />
2 −1 3 0 4 1<br />
= , (−1)(3) + (2)(2) = 1;<br />
−1 2 2 −1 1 −2<br />
And finally the component (AB)22 = −2 is obtained as follows:<br />
<br />
2 −1 3 0 4 1<br />
= , (−1)(0) + (2)(−1) = −2.<br />
−1 2 2 −1 1 −2<br />
Example 5.3.13: Compute BA, where A =<br />
Solution: We find that BA =<br />
Example 5.3.14: Compute AB and BA, where A =<br />
Solution: The product AB is<br />
<br />
4 3 1 2 3<br />
AB =<br />
2 1 4 5 6<br />
<br />
2 −1<br />
and B =<br />
−1 2<br />
<br />
3 0<br />
.<br />
2 −1<br />
<br />
6 −3<br />
. Notice that in this case AB = BA. ⊳<br />
5 −4<br />
⇒ AB =<br />
<br />
4 3<br />
and B =<br />
2 1<br />
<br />
16 23 30<br />
.<br />
6 9 12<br />
<br />
1 2 3<br />
.<br />
4 5 6<br />
The product BA is not possible. ⊳<br />
⊳
168 G. NAGY – ODE july 2, 2013<br />
5.3.3. The inverse matrix. We now introduce the concept <strong>of</strong> the inverse <strong>of</strong> a square<br />
matrix. Not every square matrix is invertible. The inverse <strong>of</strong> a matrix is useful to compute<br />
solutions to linear systems <strong>of</strong> algebraic <strong>equations</strong>.<br />
Definition 5.3.8. The matrix In ∈ F n,n is the identity matrix iff Inx = x for all x ∈ F n .<br />
It is simple to see that the components <strong>of</strong> the identity matrix are given by<br />
<br />
Iii = 1<br />
The cases n = 2, 3 are given by<br />
In = [Iij] with<br />
I2 =<br />
Iij = 0 i = j.<br />
⎡ ⎤<br />
1 0 0<br />
1 0<br />
, I3 = ⎣0 1 0⎦<br />
.<br />
0 1<br />
0 0 1<br />
Definition 5.3.9. A matrix A ∈ F n,n is called invertible iff there exists a matrix, denoted<br />
as A −1 , such that A −1 A = In, and A A −1 = In.<br />
Example 5.3.15: Verify that the matrix and its inverse are given by<br />
<br />
2 2<br />
A = , A<br />
1 3<br />
−1 = 1<br />
<br />
3 −2<br />
.<br />
4 −1 2<br />
Solution: We have to compute the products,<br />
A A −1 <br />
2 2 1 3 −2<br />
=<br />
=<br />
1 3 4 −1 2<br />
1<br />
<br />
4 0<br />
4 0 4<br />
⇒ A A −1 = I2.<br />
It is simple to check that the equation A −1 A = I2 also holds. ⊳<br />
Theorem 5.3.10. Given a 2 × 2 matrix A introduce the number ∆ as follows,<br />
<br />
a b<br />
A = , ∆ = ad − bc.<br />
c d<br />
The matrix A is invertible iff ∆ = 0. Furthermore, if A is invertible, its inverse is given by<br />
A −1 = 1<br />
∆<br />
<br />
d −b<br />
. (5.3.4)<br />
−c a<br />
The number ∆ is called the determinant <strong>of</strong> A, since it is the number that determines whether<br />
A is invertible or not.<br />
<br />
2 2<br />
Example 5.3.16: Compute the inverse <strong>of</strong> matrix A = , given in Example 5.3.15.<br />
1 3<br />
Solution: Following Theorem 5.3.10 we first compute ∆ = 6 − 4 = 4. Since ∆ = 0, then<br />
A−1 exists and it is given by<br />
A −1 = 1<br />
<br />
3<br />
4 −1<br />
<br />
−2<br />
.<br />
2<br />
⊳<br />
<br />
1<br />
Example 5.3.17: Compute the inverse <strong>of</strong> matrix A =<br />
3<br />
<br />
2<br />
.<br />
6<br />
Solution: Following Theorem 5.3.10 we first compute ∆ = 6 − 6 = 0. Since ∆ = 0, then<br />
matrix A is not invertible. ⊳
G. NAGY – ODE July 2, 2013 169<br />
Gauss operations can be used to compute the inverse <strong>of</strong> a matrix. The reason for this is<br />
simple to understand in the case <strong>of</strong> 2 × 2 matrices, as can be seen in the following Example.<br />
Example 5.3.18: Given any 2 × 2 matrix A, find its inverse matrix, A −1 , or show that the<br />
inverse does not exist.<br />
Solution: If the inverse matrix, A −1 exists, then denote it as A −1 = [x1, x2]. The equation<br />
A(A −1 ) = I2 is then equivalent to A [x1, x2] =<br />
two algebraic linear systems,<br />
A x1 =<br />
<br />
1 0<br />
. This equation is equivalent to solving<br />
0 1<br />
<br />
1<br />
, A x2 =<br />
0<br />
<br />
0<br />
.<br />
1<br />
Here is where we can use Gauss elimination operations. We use them on both systems<br />
<br />
<br />
1<br />
<br />
0<br />
A , A .<br />
0<br />
1<br />
However, we can solve both systems at the same time if we do Gauss operations on the<br />
bigger augmented matrix <br />
<br />
1<br />
A <br />
0<br />
<br />
0<br />
.<br />
1<br />
Now, perform Gauss operations until we obtain the reduced echelon form for [A|I2]. Then<br />
we can have two different types <strong>of</strong> results:<br />
• If there is no line <strong>of</strong> the form [0, 0|∗, ∗] with any <strong>of</strong> the star coefficients non-zero,<br />
then matrix A is invertible and the solution vectors x1, x2 form the columns <strong>of</strong> the<br />
inverse matrix, that is, A−1 = [x1, x2].<br />
• If there is a line <strong>of</strong> the form [0, 0|∗, ∗] with any <strong>of</strong> the star coefficients non-zero, then<br />
matrix A is not invertible. ⊳<br />
<br />
2 2<br />
Example 5.3.19: Use Gauss operations to find the inverse <strong>of</strong> A = .<br />
1 3<br />
Solution: As we said in the Example above, perform Gauss operation on the augmented<br />
matrix [A|I2] until the reduced echelon form is obtained, that is,<br />
<br />
2 2 <br />
1 0 1 3 0 1<br />
1 3 → <br />
1 3 <br />
0 1 2 2 → 0<br />
1 0 0 −4 1<br />
<br />
1 3 <br />
0 1 1 0 3 1<br />
0 1 1 1 → 4 − 2<br />
− 4 2 0 1 1 1 − 4 2<br />
<br />
1<br />
→<br />
−2<br />
That is, matrix A is invertible and the inverse is<br />
A −1 <br />
3 1<br />
= 4 − 2<br />
− 1<br />
<br />
1 ⇔ A<br />
4 2<br />
−1 = 1<br />
<br />
3<br />
4 −1<br />
<br />
−2<br />
.<br />
2<br />
⎡<br />
1 2<br />
⎤<br />
3<br />
⊳<br />
Example 5.3.20: Use Gauss operations to find the inverse <strong>of</strong> A = ⎣2<br />
5 7⎦.<br />
3 7 9<br />
Solution: We perform Gauss operations on the augmented matrix [A|I3] until we obtain<br />
its reduced echelon form, that is,<br />
⎡ <br />
1 2 3 <br />
1<br />
⎣2<br />
5 7 <br />
0<br />
3 7 9 0<br />
0<br />
1<br />
0<br />
⎤ ⎡<br />
0 1<br />
0⎦<br />
→ ⎣0<br />
1 0<br />
2<br />
1<br />
1<br />
3<br />
1<br />
0<br />
<br />
<br />
<br />
<br />
<br />
<br />
1<br />
−2<br />
−3<br />
0<br />
1<br />
0<br />
⎤<br />
0<br />
0⎦<br />
→<br />
1
170 G. NAGY – ODE july 2, 2013<br />
⎡<br />
1<br />
⎣0<br />
0<br />
0<br />
1<br />
0<br />
<br />
⎤ ⎡<br />
1 <br />
5 −2 0 1<br />
1 <br />
−2 1 0⎦<br />
→ ⎣<br />
−1 −1 −1 1<br />
0 1 0 1 1<br />
5 −2 0<br />
0 0 1<br />
−2 1 0<br />
⎡<br />
1<br />
⎣0<br />
0<br />
0<br />
1<br />
0<br />
<br />
⎤ ⎡<br />
1 <br />
5 −2 0 1<br />
1 −2 1 0 ⎦<br />
<br />
→ ⎣0<br />
1 1 1 −1 0<br />
0<br />
1<br />
0<br />
0<br />
0<br />
1<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
1<br />
4<br />
−3<br />
1<br />
1<br />
−3<br />
0<br />
1<br />
⎤<br />
⎦<br />
−1<br />
⎤<br />
1<br />
1 ⎦<br />
−1<br />
We conclude that matrix A is invertible and<br />
A −1 ⎡<br />
4<br />
= ⎣−3 −3<br />
0<br />
⎤<br />
1<br />
1 ⎦ .<br />
1 1 −1<br />
5.3.4. Determinants. A determinant is a scalar computed form a square matrix that gives<br />
important information about the matrix, for example if the matrix is invertible or not. We<br />
now review the definition and properties <strong>of</strong> the determinant <strong>of</strong> 2 × 2 and 3 × 3 matrices.<br />
<br />
a11 a12<br />
Definition 5.3.11. The determinant <strong>of</strong> a 2 × 2 matrix A =<br />
is given by<br />
<br />
<br />
det(A) = <br />
a11 a12<br />
a21 a22<br />
The determinant <strong>of</strong> a 3 × 3 matrix A = ⎣<br />
<br />
<br />
<br />
det(A) = <br />
<br />
<br />
a11 a12 a13<br />
a21 a22 a23<br />
a31 a32 a33<br />
<br />
<br />
<br />
<br />
= a11 <br />
<br />
<br />
<br />
<br />
= a11a22 − a12a21.<br />
⎡<br />
⎤<br />
a22 a23<br />
a32 a33<br />
a11 a12 a13<br />
a21 a22 a23<br />
a31 a32 a33<br />
<br />
<br />
<br />
− a12 <br />
a21 a22<br />
⎦ is given by<br />
a21 a23<br />
a31 a33<br />
<br />
<br />
<br />
+ a13 <br />
a21 a22<br />
a31 a32<br />
Example 5.3.21: The following three examples show that the determinant can be a negative,<br />
zero or positive number.<br />
<br />
<br />
1<br />
2<br />
<br />
3 4<br />
= 4 − 6 = −2,<br />
<br />
<br />
2<br />
3 <br />
1<br />
<br />
4<br />
= 8 − 3 = 5,<br />
<br />
<br />
1<br />
2 <br />
2<br />
<br />
4<br />
= 4 − 4 = 0.<br />
The following is an example shows how to compute the determinant <strong>of</strong> a 3 × 3 matrix,<br />
<br />
<br />
1<br />
3 −1<br />
<br />
<br />
<br />
2<br />
1 1 <br />
= (1) 1<br />
1<br />
<br />
<br />
<br />
3<br />
2 1 2 1<br />
− 3 2<br />
1<br />
<br />
<br />
3 1<br />
+ (−1) 2<br />
1<br />
<br />
3 2<br />
= (1 − 2) − 3 (2 − 3) − (4 − 3)<br />
= −1 + 3 − 1<br />
= 1. ⊳<br />
The absolute value <strong>of</strong> the determinant <strong>of</strong> a 2 × 2 matrix A = [a1, a2] has a geometrical<br />
meaning: It is the area <strong>of</strong> the parallelogram whose sides are given by a1 and bia2, that is, by<br />
the columns <strong>of</strong> the matrix A; see Fig. 25. Analogously, the absolute value <strong>of</strong> the determinant<br />
<strong>of</strong> a 3 × 3 matrix A = [a 1, a 2, a 3] also has a geometrical meaning: It is the volume <strong>of</strong> the<br />
parallelepiped whose sides are given by a 1, a 2 and a 3, that is, by the columns <strong>of</strong> the matrix<br />
A; see Fig. 25.<br />
The determinant <strong>of</strong> an n × n matrix A can be defined generalizing the properties that<br />
areas <strong>of</strong> parallelogram have in two dimensions and volumes <strong>of</strong> parallelepipeds have in three<br />
dimensions. One <strong>of</strong> these properties is the following: if one <strong>of</strong> the column vectors <strong>of</strong> the<br />
<br />
<br />
<br />
.<br />
⊳
0<br />
a 1<br />
R<br />
2<br />
G. NAGY – ODE July 2, 2013 171<br />
a 2<br />
Figure 25. Geometrical meaning <strong>of</strong> the determinant.<br />
matrix A is a linear combination <strong>of</strong> the others, then the figure determined by these column<br />
vectors is not n-dimensional but (n−1)-dimensional, so its volume must vanish. We highlight<br />
this property <strong>of</strong> the determinant <strong>of</strong> n × n matrices in the following result.<br />
Theorem 5.3.12. The set <strong>of</strong> n-vectors {v 1, · · · , vn}, with n 1, is linearly dependent iff<br />
a 1<br />
det[v1, · · · , vn] = 0.<br />
Example 5.3.22: Show whether the set <strong>of</strong> vectors below linearly independent,<br />
⎡<br />
⎣ 1<br />
⎤ ⎡<br />
2⎦<br />
, ⎣<br />
3<br />
3<br />
⎤ ⎡<br />
2⎦<br />
, ⎣<br />
1<br />
−3<br />
⎤<br />
2 ⎦<br />
7<br />
<br />
.<br />
The determinant <strong>of</strong> the matrix whose column vectors are the vectors above is given by<br />
<br />
<br />
1<br />
3 −3<br />
<br />
<br />
2<br />
2 2 <br />
= (1) (14 − 2) − 3 (14 − 6) + (−3) (2 − 6) = 12 − 24 + 12 = 0.<br />
3<br />
1 7 <br />
Therefore, the set <strong>of</strong> vectors above is linearly dependent. ⊳<br />
The determinant <strong>of</strong> a square matrix also determines whether the matrix is invertible or not.<br />
Theorem 5.3.13. An n × n matrix A is invertible iff holds det(A) = 0.<br />
⎡<br />
1 2<br />
⎤<br />
3<br />
Example 5.3.23: Is matrix A = ⎣2<br />
5 7⎦<br />
invertible?<br />
3 7 9<br />
Solution: We only need to compute the determinant <strong>of</strong> A.<br />
<br />
<br />
1<br />
2 3<br />
<br />
<br />
det(A) = <br />
2<br />
5 7<br />
= (1) 5<br />
7<br />
<br />
<br />
<br />
3<br />
7 9<br />
7 9<br />
− (2) 2<br />
7<br />
<br />
<br />
3 9<br />
+ (3) 2<br />
5<br />
<br />
3 7<br />
.<br />
Using the definition <strong>of</strong> determinant for 2 × 2 matrices we obtain<br />
det(A) = (45 − 49) − 2(18 − 21) + 3(14 − 15) = −4 + 6 − 3.<br />
Since det(A) = −1, that is, non-zero, matrix A is invertible. ⊳<br />
R 3<br />
a 2<br />
a<br />
3
172 G. NAGY – ODE july 2, 2013<br />
5.3.5. Exercises.<br />
5.3.1.- . 5.3.2.- .
G. NAGY – ODE July 2, 2013 173<br />
5.4. Linear <strong>differential</strong> Systems<br />
In Section 5.1 we introduced <strong>differential</strong> systems and the particular case <strong>of</strong> linear <strong>differential</strong><br />
systems. We also defined the initial value problem for <strong>differential</strong> systems and stated<br />
a version <strong>of</strong> the Picard-Lindelöf Theorem. This result says that the initial value problem for<br />
certain <strong>differential</strong> systems has a unique solution. In this Section we focus on linear <strong>differential</strong><br />
systems. We use matrix-vector notation to simplify the presentation. We define several<br />
useful concepts like fundamental solutions, general solutions, fundamental matrix, and the<br />
Wronskian. We also describe few properties shared by solutions to linear <strong>differential</strong> systems.<br />
The superposition property <strong>of</strong> solutions to homogeneous linear <strong>differential</strong> systems is<br />
one example. The generalization <strong>of</strong> Abel’s Theorem from a single linear <strong>differential</strong> equation<br />
to a linear <strong>differential</strong> system is another example.<br />
5.4.1. Matrix-vector notation. In Section 5.1, Definition 5.1.2 we introduced an n × n<br />
first order system <strong>of</strong> linear <strong>differential</strong> <strong>equations</strong>. We said that one has to fix both the<br />
coefficient and source functions aij, gi : [t1, t2] → R, where i, j = 1, · · · , n, and then one has<br />
to find n functions xj : [t1, t2] → R solutions <strong>of</strong> the n <strong>equations</strong><br />
x ′ 1 = a11(t) x1 + · · · + a1n(t) xn + g1(t),<br />
.<br />
x ′ n = an1(t) x 1 + · · · + ann(t) xn + gn(t).<br />
We also said that the system is called homogeneous iff the source functions satisfy that<br />
g1 = · · · = gn = 0, and the system is called <strong>of</strong> constant coefficients iff all functions aij are<br />
constants. Matrix-vector notation is very helpful to express linear <strong>differential</strong> systems in a<br />
compact way. Since the coefficients aij, the sources gi, and the unknowns xj are functions,<br />
we need to introduce a matrix-valued function and two vector-valued functions,<br />
⎡<br />
⎤<br />
⎡ ⎤<br />
⎡ ⎤<br />
a11(t) · · · a1n(t)<br />
g1(t)<br />
x1(t)<br />
⎢<br />
⎥<br />
⎢ ⎥<br />
⎢ ⎥<br />
A(t) = ⎣ .<br />
. ⎦ , g(t) = ⎣ . ⎦ , x(t) = ⎣ . ⎦ .<br />
an1(t) · · · ann(t)<br />
gn(t)<br />
xn(t)<br />
Define the derivative <strong>of</strong> a vector-valued function x : [t1, t2] → Rn component-wise, that is,<br />
x ′ ⎡<br />
x<br />
⎢<br />
(t) = ⎣<br />
′ 1(t)<br />
.<br />
x ′ ⎤<br />
⎥<br />
⎦ . (5.4.1)<br />
n(t)<br />
Then, the n × n system <strong>of</strong> linear <strong>differential</strong> <strong>equations</strong> given above can be written in a<br />
compact was as follows,<br />
x ′ = A(t) x + g(t).<br />
Example 5.4.1: Use matrix notation to write down the 2 × 2 system given by<br />
x ′ 1 = x1 − x2,<br />
x ′ 2 = −x 1 + x 2.<br />
Solution: In this case, the matrix <strong>of</strong> coefficients and the unknown vector have the form<br />
<br />
1<br />
A =<br />
−1<br />
<br />
−1<br />
,<br />
1<br />
<br />
x1(t)<br />
x(t) = .<br />
x2(t)
174 G. NAGY – ODE july 2, 2013<br />
This is an homogeneous system, so the source vector g = 0. The <strong>differential</strong> equation can<br />
be written as follows,<br />
x ′ 1 = x1 − x2<br />
x ′ <br />
′ x 1 ⇔<br />
2 = −x1 + x x 2<br />
′ <br />
1 −1 x1<br />
=<br />
⇔ x<br />
2 −1 1 x2<br />
′ = Ax.<br />
Example 5.4.2: Find the explicit expression for the linear system x ′ = Ax + b, where<br />
<br />
1<br />
A =<br />
3<br />
<br />
3<br />
,<br />
1<br />
<br />
t e<br />
g(t) = ,<br />
<br />
x1<br />
x = .<br />
Solution: The 2 × 2 linear system is given by<br />
<br />
′ x 1<br />
x ′ <br />
t<br />
1 3 x1 e<br />
=<br />
+<br />
2 3 1 x2 2e3t <br />
2e 3t<br />
x2<br />
, ⇔ x′ 1 = x1 + 3x2 + e t ,<br />
x ′ 2 = 3x1 + x2 + 2e 3t .<br />
Example 5.4.3: Find and express in matrix form the first order reduction to the system<br />
y ′′ + 2y ′ + 3y = sin(5t).<br />
Solution: We then introduce the functions x 1 = y and x 2 = y ′ . Therefore, x 1 and x 2<br />
satisfy the 2 × 2 <strong>differential</strong> linear system,<br />
x ′ 1 = x 2,<br />
x ′ 2 = −3x1 − 2x2 + sin(5t).<br />
Introduce the matrix, source vector and unknown vector,<br />
<br />
0<br />
A =<br />
−3<br />
<br />
1<br />
,<br />
−2<br />
<br />
0<br />
g(t) = ,<br />
sin(5t)<br />
<br />
x1(t)<br />
x(t) = ,<br />
x2(t)<br />
then the system above can be written as follows,<br />
<br />
′ x 1<br />
x ′ <br />
0 1 x1 0<br />
=<br />
+<br />
⇔ x<br />
2 −3 −2 x2 sin(5t)<br />
′ = Ax + g(t).<br />
⊳<br />
Example 5.4.4: Show that the vector-valued functions x (1) <br />
2<br />
= e<br />
1<br />
2t and x (2) <br />
1<br />
= e<br />
2<br />
−t<br />
are solutions to the 2 × 2 linear system x ′ <br />
3 −2<br />
= Ax, where A = .<br />
2 −2<br />
Solution: We start computing the right-hand side in the <strong>differential</strong> equation above for<br />
the function x (1) , that is,<br />
Ax (1) <br />
3 −2 2<br />
=<br />
e<br />
2 −2 1<br />
2t <br />
4<br />
= e<br />
2<br />
2t <br />
2<br />
= 2 e<br />
1<br />
2t <br />
2<br />
= 2e<br />
1<br />
2t<br />
=<br />
so we conclude that x (1) ′ = Ax (1) . Analogously,<br />
Ax (2) =<br />
<br />
3<br />
<br />
−2 1<br />
2 −2 2<br />
e −t =<br />
<br />
−1<br />
−2<br />
e −t = −<br />
<br />
1<br />
2<br />
e −t =<br />
<br />
1<br />
−e<br />
2<br />
−t<br />
=<br />
<br />
2<br />
e<br />
1<br />
2t ′<br />
= x (1)′ ,<br />
⊳<br />
⊳<br />
<br />
1<br />
e<br />
2<br />
−t ′<br />
= x (2) ′ ,<br />
so we conclude that x (2) ′ = Ax (2) . ⊳
G. NAGY – ODE July 2, 2013 175<br />
5.4.2. Homogeneous systems. May be the most important property satisfied by solutions<br />
to linear homogeneous <strong>differential</strong> system is the superposition property. Given two solutions<br />
<strong>of</strong> the homogeneous system, their linear combination is also a solution to that system. This<br />
property will play a crucial role to compute solutions to linear systems.<br />
Theorem 5.4.1 (Superposition). If each vector-valued function x (1) , x (2) : [t 1, t 2] → R n<br />
is solution <strong>of</strong> the homogeneous linear system x ′ = A(t) x, then so is the linear combination<br />
ax (1) + bx (2) , for all a, b ∈ R.<br />
This Theorem contains the particular case a = b = 1. So, if x (1) and x (2) are solutions <strong>of</strong><br />
an homogeneous linear system, so is x (1) + x (2) . Another particular case is b = 0 and a<br />
arbitrary. So, if x (1) is a solution <strong>of</strong> an homogeneous linear system, so is ax (1) .<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.4.1: The vector-valued functions x (1) and x (2) are solutions to the<br />
homogeneous linear system, that is,<br />
x (1) ′ = Ax (1) , x (2) ′ = Ax (2) . (5.4.2)<br />
It is simple to see that the derivative <strong>of</strong> a vector-valued function, as defined in Eq. (5.4.1),<br />
is a linear operation, meaning that for all a, b ∈ R holds,<br />
ax (1) + bx (2) ′ = a x (1) ′ + b x (2) ′ .<br />
Replacing Eq. (5.4.2) above we get<br />
ax (1) + bx (2) ′ = a Ax (1) + b Ax (2) .<br />
It was shown in Theorem 5.2.7 that the matrix-vector product is a linear operation, that is,<br />
it satisfies Eq. (5.2.3), therefore, a Ax (1) + b Ax (2) = A ax (1) + bx (2) . We then obtain,<br />
ax (1) + bx (2) ′ = A ax (1) + bx (2) ,<br />
showing that the vector ax (1) + bx (2) is solution <strong>of</strong> the homogeneous equation too. This<br />
establishes the Theorem. <br />
Example 5.4.5: Verify that the functions x (1) <br />
1<br />
= e<br />
1<br />
−2t and x (2) <br />
−1<br />
= e<br />
1<br />
4t are solutions<br />
to the homogeneous linear system<br />
x ′ = Ax, A =<br />
Verify that x (1) + x (2) is also a solution.<br />
<br />
1 −3<br />
.<br />
−3 1<br />
Solution: The function x (1) is solution to the <strong>differential</strong> equation, since<br />
x (1) ′ <br />
1<br />
= −2 e<br />
1<br />
−2t , Ax (1) <br />
1 −3 1<br />
=<br />
e<br />
−3 1 1<br />
−2t <br />
−2<br />
= e<br />
−2<br />
−2t <br />
1<br />
= −2 e<br />
1<br />
−2t .<br />
We then conclude that x (1) ′ = Ax (1) . Analogously, the function x (2) is solution to the<br />
<strong>differential</strong> equation, since<br />
x (2) ′ <br />
−1<br />
= 4 e<br />
1<br />
4t , Ax (2) =<br />
1 −3<br />
−3 1<br />
−1<br />
1<br />
<br />
e 4t =<br />
<br />
−4<br />
e<br />
4<br />
4t <br />
−1<br />
= 4 e<br />
1<br />
4t .<br />
We then conclude that x (2) ′ = Ax (2) . To show that x (1) + x (2) is also a solution we could<br />
use the linearity <strong>of</strong> the matrix-vector product, as we did in the pro<strong>of</strong> <strong>of</strong> the Theorem 5.4.1.<br />
Here we choose the straightforward, although more obscure, calculation: On the one hand,<br />
x (1) + x (2) <br />
−2t 4t e − e<br />
=<br />
⇒ x (1) + x (2) <br />
−2t 4t<br />
′ −2e − 4e<br />
=<br />
.<br />
e −2t + e 4t<br />
−2e −2t + 4e 4t
176 G. NAGY – ODE july 2, 2013<br />
On the other hand,<br />
that is,<br />
A x (1) + x (2) =<br />
1 −3<br />
−3 1<br />
e −2t − e 4t<br />
e −2t + e 4t<br />
A x (1) + x (2) =<br />
<br />
=<br />
e −2t − e 4t − 3e −2t − 3e 4t<br />
−2e −2t − 4e 4t<br />
−3e −2t + 3e 4t + e −2t + e 4t<br />
−2e −2t + 4e 4t<br />
We conclude that x (1) + x (2) ′ = A x (1) + x (2) . ⊳<br />
Given n solutions x (1) , · · · , x (n) to an n × n linear homogeneous <strong>differential</strong> system, one<br />
can always construct a matrix having these solutions as columns, that is, x (1) , · · · , x (n) .<br />
This solution matrix is an n×n matrix for every value <strong>of</strong> the parameter t, so we can compute<br />
several things with such a matrix. For example we can compute its determinant. Since this<br />
determinant has important applications, we give it a name.<br />
Definition 5.4.2. Given n vector-valued functions x (1) , · · · , x (n) solutions to an n×n linear<br />
homogeneous system x ′ = A(t) x, we call the Wronskian <strong>of</strong> these solutions to the function<br />
w = det x (1) , · · · , x (n) .<br />
Suppose that the linear system in Definition 5.4.2 is a first order reduction <strong>of</strong> a second order<br />
linear homogeneous equation, y ′′ + a1y ′ + a0y = 0, studied in Sect. 2.1. That is, suppose<br />
that the linear system is<br />
x ′ 1 = x 2, x ′ 2 = −a 0x 1 − a 1x 2.<br />
In this case, the Wronskian defined here coincides with the definition given in Sect. 2.1. The<br />
pro<strong>of</strong> is simple. Suppose that y1, y2 are fundamental solutions <strong>of</strong> the second order equation,<br />
then the vector-valued functions<br />
x (1) <br />
y1<br />
=<br />
y ′ <br />
, x<br />
1<br />
(2) <br />
y2<br />
=<br />
y ′ <br />
,<br />
2<br />
are solutions <strong>of</strong> the first order reduction system. Then holds,<br />
w = det x (1) , x (2) <br />
<br />
= <br />
y1 y2<br />
y ′ 1 y ′ <br />
<br />
<br />
= Wy1y2 .<br />
2<br />
Example 5.4.6: Compute the Wronskian <strong>of</strong> the vector-valued functions given in Example<br />
5.4.5, that is, x (1) <br />
1<br />
= e<br />
1<br />
−2t and x (2) <br />
−1<br />
= e<br />
1<br />
4t .<br />
Solution: The Wronskian is the determinant <strong>of</strong> the solution matrix, with the vectors<br />
placed in any order. For example, we can choose the order x (1) , x (2) . If we choose the<br />
order x (2) , x (1) , this second Wronskian is the negative <strong>of</strong> the first one. Choosing the first<br />
order we get,<br />
w(t) = det x (1) , x (2) <br />
<br />
= <br />
<br />
<br />
<br />
<br />
.<br />
e−2t −e4t e−2t e4t = e−2t e 4t + e −2t e 4t .<br />
We conclude that w(t) = 2e 2t . ⊳<br />
We saw in Section 5.3, Theorem 5.3.12, that the determinant <strong>of</strong> an n × n matrix is non-zero<br />
iff the matrix column vectors form a linearly independent set. Since in Example 5.4.6 the<br />
Wronskian is non-zero for all t ∈ R, this implies that the solution set {x (1) , x (2) } is linearly<br />
independent for every t ∈ R.<br />
<br />
,
G. NAGY – ODE July 2, 2013 177<br />
Example 5.4.7: Show that the vector-valued functions x (1) <br />
3t e<br />
=<br />
2e3t <br />
and x (2) <br />
−t e<br />
=<br />
−2e−t <br />
form a linearly independent set for all t ∈ R.<br />
<br />
3t e e<br />
Solution: We compute the determinant <strong>of</strong> the matrix X(t) =<br />
−t<br />
<br />
, that is,<br />
<br />
<br />
w(t) = <br />
e3t e−t 2e3t −2e−t 2e 3t −2e −t<br />
<br />
<br />
<br />
= −2e2t − 2e 2t ⇒ w(t) = −4e 2t = 0 t ∈ R.<br />
Wronskians are used to find linearly independent set <strong>of</strong> n solutions to n × n linear homogeneous<br />
<strong>differential</strong> systems. These linearly independent solution sets contain very important<br />
information about the system. We will see that they actually contain all the information<br />
needed to construct all possible solutions to that system. Because <strong>of</strong> that we give such<br />
solutions a name.<br />
Definition 5.4.3. Consider the n × n linear homogeneous <strong>differential</strong> system x ′ = A(t) x,<br />
where A is an n × n matrix-valued function. The set {x (1) , · · · , x (n) } <strong>of</strong> n solutions to this<br />
<strong>differential</strong> system is called a fundamental set in the interval I ⊂ R iff the set is linearly<br />
independent for all t ∈ I. The n × n matrix-valued function<br />
X = x (1) , · · · , x (n)<br />
is called a fundamental matrix <strong>of</strong> the system.<br />
Using the notation above, the Wronskian <strong>of</strong> a fundamental set is given by w(t) = det X(t) .<br />
Since a fundamental set is linearly independent, that means the Wronskian <strong>of</strong> the fundamental<br />
matrix is non-zero for every t where the solutions are defined.<br />
Example 5.4.8: Find two fundamental matrices for the linear homogeneous system in Example<br />
5.4.5.<br />
Solution: One fundamental matrix is simple to find, is it constructed with the solutions<br />
given in Example 5.4.5, that is,<br />
X = x (1) , x (2) ⇒<br />
<br />
−2t e<br />
X(t) =<br />
<br />
4t −e<br />
.<br />
e −2t e 4t<br />
A second fundamental matrix can be obtained multiplying by any non-zero constant each<br />
solution above. For example, another fundamental matrix is<br />
˜X = 2x (1) , 3x (2) ⇒ ˜ <br />
−2t 2e<br />
X(t) =<br />
2e<br />
4t −3e −2t 3e4t <br />
.<br />
⊳<br />
5.4.3. Abel’s Theorem for systems. The following result is a generalization <strong>of</strong> Abel’s<br />
Theorem 2.1.9 in Section 2.1 to n solutions <strong>of</strong> an n×n linear homogeneous <strong>differential</strong> system.<br />
This result shows an important property <strong>of</strong> the Wronskian: The solution Wronskian vanishes<br />
at a single point iff it vanishes identically for all t ∈ R.<br />
Theorem 5.4.4 (Abel for systems). Let A be an n×n continuous matrix-valued function,<br />
let tr A be the trace <strong>of</strong> A, let {x (1) , · · · , x (n) } be a set <strong>of</strong> n solutions to the <strong>differential</strong><br />
system x ′ = A(t)x, and let t0 ∈ R be an arbitrary number. Then, the Wronskian function<br />
w = det x (1) , · · · , x (n) is given by the expression<br />
w(t) = w(t0) e α(t) , α(t) =<br />
t<br />
t0<br />
tr A(τ) dτ.<br />
⊳
178 G. NAGY – ODE july 2, 2013<br />
In the case <strong>of</strong> a constant matrix A, the equation above for the Wronskian reduces to<br />
w(t) = w(t0) e (tr A) (t−t0) .<br />
Once again, this result implies that the solution Wronskian vanishes at a single point iff<br />
it vanishes identically for all t ∈ R. One consequence is this: If n solutions to a linear<br />
homogeneous <strong>differential</strong> system are linearly independent at the initial time t0, then they<br />
remain linearly independent for every time t ∈ R.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.4.4: The pro<strong>of</strong> is based in an identity satisfied by the determinant<br />
<strong>of</strong> certain matrix-valued functions. The pro<strong>of</strong> <strong>of</strong> this identity is quite involved, so we do<br />
not provide it here. The identity is the following: Every n × n, differentiable, invertible,<br />
matrix-valued function Z, with values Z(t) for t ∈ R, satisfies the identity:<br />
d<br />
det(Z) = det(Z) tr<br />
dt<br />
<br />
−1 d<br />
Z<br />
dt Z<br />
We use this identity with any fundamental matrix X = x (1) , · · · , x (n) <strong>of</strong> the linear homogeneous<br />
<strong>differential</strong> system x ′ = Ax. Recalling that the Wronskian w(t) = det X(t) , the<br />
identity above says,<br />
w ′ (t) = w(t) tr X −1 (t) X ′ (t) .<br />
We now compute the derivative <strong>of</strong> the fundamental matrix,<br />
X ′ = x (1) ′ , · · · , x (n) ′ = Ax (1) , · · · , Ax (n) = AX,<br />
where the equation on the far right comes from the definition <strong>of</strong> matrix multiplication.<br />
Replacing this equation in the Wronskian equation we get<br />
w ′ (t) = w(t) tr X −1 AX = w(t) tr X X −1 A = w(t) tr (A),<br />
where in the second equation above we used a property <strong>of</strong> the trace <strong>of</strong> three matrices:<br />
tr (ABC) = tr (CAB) = tr (BCA). Therefore, we have seen that the Wronskian satisfies<br />
the equation<br />
w ′ (t) = tr A(t) w(t). (5.4.3)<br />
This is a linear <strong>differential</strong> equation <strong>of</strong> a single function w : R → R. We integrate it using<br />
the integrating factor method from Section 1.2. The result is<br />
w(t) = w(t0) e α(t) , α(t) =<br />
t<br />
t0<br />
<br />
.<br />
tr A(τ) dτ.<br />
This establishes the Theorem. <br />
Example 5.4.9: Show that the Wronskian <strong>of</strong> the fundamental matrix constructed with the<br />
solutions given in Example 5.4.4 satisfies Eq. (5.4.3) above.<br />
Solution: In Example 5.4.4 we have shown that the vector-valued functions x (1) <br />
2<br />
= e<br />
1<br />
2t<br />
and x (2) <br />
1<br />
= e<br />
2<br />
−t are solutions to the system x ′ <br />
3 −2<br />
= Ax, where A = . The matrix<br />
2 −2<br />
<br />
2t 2e e<br />
X(t) =<br />
−t<br />
<br />
e 2t 2e −t<br />
is a fundamental matrix <strong>of</strong> the system, since its Wronskian is non-zero,<br />
<br />
<br />
w(t) = <br />
2e2t e−t <br />
<br />
<br />
= 4et − e t ⇒ w(t) = 3e t .<br />
e 2t 2e −t
G. NAGY – ODE July 2, 2013 179<br />
We need to compute the right-hand side and the left-hand side <strong>of</strong> Eq. (5.4.3) and verify that<br />
they coincide. We start with the left-hand side,<br />
w ′ (t) = 3e t = w(t).<br />
The right-hand side is<br />
tr (A) w(t) = (3 − 2) w(t) = w(t).<br />
Therefore, we have shown that w ′ (t) = tr (A) w(t). ⊳<br />
5.4.4. Existence <strong>of</strong> fundamental solutions. We now show that every n × n linear homogeneous<br />
<strong>differential</strong> system has a set <strong>of</strong> n fundamental solutions. Abel’s Theorem above<br />
together with Theorem 5.1.5 in Section 5.1 are the crucial statements needed to show this<br />
existence result. We also need to recall the Definition 5.1.4 <strong>of</strong> an initial value problem,<br />
which in the case <strong>of</strong> a linear non-homogeneous <strong>differential</strong> system is the following: Given an<br />
n × n matrix-valued function A, a source vector-valued function g, an initial n-vector x0,<br />
and an initial time t0, find the vector-valued function x solution <strong>of</strong><br />
x ′ = Ax + g, x(t0) = x0.<br />
The following statement says that a fundamental set <strong>of</strong> solutions to a linear <strong>differential</strong><br />
system can be found solving n initial value problems. We use the notation e (i) to denote<br />
the i-th column vector <strong>of</strong> the identity matrix I, that is, I = e (1) , · · · , e (n) .<br />
Theorem 5.4.5 (Fundamental solutions). If the n × n matrix-valued function A is<br />
continuous on an open interval I ⊂ R, then the <strong>differential</strong> equation x ′ = A(t) x has a<br />
fundamental set given by {x (1) , · · · , x (n) }, where the vector-valued function x (i) : I → R n ,<br />
for i = 1, · · · , n, is the solution <strong>of</strong> the initial value problem<br />
x (i) ′ = A(t) x (i) , x (i) (t0) = e (i) . (5.4.4)<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.4.5: Since linear homogeneous <strong>differential</strong> systems are particular<br />
cases <strong>of</strong> the <strong>differential</strong> systems in Picard-Lindelöf Theorem 5.1.5, each initial value problem<br />
in Eq. (5.4.4) has a unique solution x (i) , for i = 1, · · · , n. These solutions form a<br />
fundamental set iff the set is linearly independent. We can check whether these solutions<br />
form an independent set or not computing their Wronskian,<br />
w(t) = det x (1) , · · · , x (n) .<br />
Abel’s Theorem 5.4.4 says that the Wronskian satisfies<br />
w(t) = e α(t) w(t0), α(t) =<br />
t<br />
t0<br />
tr A(τ) dτ.<br />
No matter what function α actually is, the function values e α(t) = 0. Hence, w(t) = 0 iff<br />
w(t 0) = 0. However, the Wronskian at the initial value t 0 is simple to compute using the<br />
initial conditions in Eq. (5.4.4),<br />
w(t 0) = det e (1) , · · · , e (n) ⇒ w(t 0) = 1.<br />
Since w(t0) = 0, so is w(t) for all t, showing that the set {x (1) , · · · , x (n) } is a fundamental<br />
set. This establishes the Theorem. <br />
We now know that the Definition 5.4.3 <strong>of</strong> fundamental set is indeed a useful one, since<br />
every linear homogeneous <strong>differential</strong> system has at least one fundamental set. The next<br />
result explains why this solution set is called fundamental. We now show that knowing a<br />
fundamental set is equivalent to knowing all solutions to a system. The reason is that every<br />
solution <strong>of</strong> a linear homogeneous <strong>differential</strong> system can be expressed as a particular linear<br />
combination <strong>of</strong> the solutions in a given fundamental set.
180 G. NAGY – ODE july 2, 2013<br />
Theorem 5.4.6. If A is a continuous n×n matrix-valued function, and the set <strong>of</strong> n-vectorvalued<br />
functions {x (1) , · · · , x (n) } is a fundamental set <strong>of</strong> the n×n linear system x ′ = A(t) x,<br />
then for every solution, x, to this <strong>differential</strong> system there exists a unique set <strong>of</strong> constants<br />
{c1, · · · , cn} such that the following expression holds,<br />
x(t) = c 1x (1) (t) + · · · + cnx (n) (t).<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.4.6: We know that each function x, x (1) , · · · , x (n) is solution <strong>of</strong> the<br />
<strong>differential</strong> equation x ′ = A(t) x. This means that for every t where the solutions are defined<br />
we can construct the set {x, x (1) , · · · , x (n) } containing n+1 vectors in R n . From basic results<br />
in Linear Algebra we know that this set must be linearly dependent. An example is a set <strong>of</strong><br />
three vectors on a plane; that set must be linearly dependent. Therefore, there exist n + 1<br />
numbers, c 0(t), ˜c 1(t), · · · , ˜cn(t), which may depend on t, not all <strong>of</strong> them zero, such that,<br />
c0(t) x(t) + ˜c1(t) x (1) (t) · · · + ˜cn(t) x (n) (t) = 0.<br />
Since the set {x (1) (t), · · · , x (n) (t)} is linearly independent for all t, then the number c0(t)<br />
must be non-zero for all t. Otherwise, all functions c 0, · · · , cn must vanish, which we just<br />
said it is not the case. So, re-ordering terms and dividing by c 0(t) we get the equation<br />
x(t) = c1(t) x (1) (t) · · · + cn(t) x (n) (t), (5.4.5)<br />
where ci(t) = −˜ci(t)/c0(t), for i = 1, · · · , n. It can be shown that every solution <strong>of</strong><br />
x ′ = A(t) x must be differentiable, since A is continuous. Knowing that all functions<br />
x, x (1) , · · · , x (n) are differentiable implies that the functions c1, · · · , cn must be differentiable<br />
too. Now compute the derivative with respect to t <strong>of</strong> the equation above. We get,<br />
x ′ = c1 x (1) ′ + · · · + cn x (n) ′ + c ′ 1 x (1) · · · + c ′ n x (n) .<br />
Recalling that all the vector-valued functions above are solution to the <strong>differential</strong> equation,<br />
we can replace their derivatives by multiplication by matrix A, that is,<br />
A(t) x = c1 A(t) x (1) + · · · + cn A(t) x (n) + c ′ 1 x (1) · · · + c ′ n x (n) .<br />
Using that the matrix-vector product is a linear operation, we get,<br />
Recalling Eq. (5.4.5), we get<br />
A(t) x = A(t) c1 x (1) + · · · + cn x (n) + c ′ 1 x (1) · · · + c ′ n x (n) .<br />
A(t) x = A(t) x + c ′ 1 x (1) · · · + c ′ n x (n) ⇔ c ′ 1 x (1) · · · + c ′ n x (n) = 0.<br />
Use again that the set {x (1) (t), · · · , x (n) (t)} is linearly independent for all t, we get that<br />
c ′ 1 = 0, · · · c ′ n = 0.<br />
So every function ci must be a constant. Then Eq. (5.4.5) now has the form<br />
x(t) = c1 x (1) (t) · · · + cn x (n) (t),<br />
with constants c1, · · · , cn. We only need to show that this set <strong>of</strong> constants is unique. Suppose<br />
that we have a second set <strong>of</strong> constants ĉ1, · · · , ĉn such that<br />
x(t) = ĉ1 x (1) (t) · · · + ĉn x (n) (t).<br />
Subtract the latter expression from the former,<br />
(c1 − ĉ1) x (1) (t) · · · + (cn − ĉn) x (n) (t) = 0.<br />
One last time we use that the set <strong>of</strong> vectors {x (1) (t), · · · , x (n) (t)} is linearly independent for<br />
all t, then we get that ci = ĉi. The constants are unique. This establishes the Theorem. <br />
The Theorem above is the reason for the following definition.
G. NAGY – ODE July 2, 2013 181<br />
Definition 5.4.7. The general solution <strong>of</strong> an n×n linear homogeneous <strong>differential</strong> system<br />
x ′ = Ax is the set <strong>of</strong> n-vector-valued functions x(t) = c1 x (1) (t) · · · + cn x (n) (t), for every<br />
constants c1, · · · , cn.<br />
Example 5.4.10: Find the general solution to <strong>differential</strong> equation in Example 5.4.4 and<br />
then use this general solution to find the solution <strong>of</strong> the initial value problem<br />
x ′ <br />
1<br />
3 −2<br />
= Ax, x(0) = , A = .<br />
5<br />
2 −2<br />
Solution: From Example 5.4.4 we know that the general solution <strong>of</strong> the <strong>differential</strong> equation<br />
above can be written as<br />
<br />
2<br />
x(t) = c1 e<br />
1<br />
2t <br />
1<br />
+ c2 e<br />
2<br />
−t .<br />
Before imposing the initial condition on this general solution, it is convenient to write this<br />
general solution in terms <strong>of</strong> the fundamental matrix <strong>of</strong> the equation above, that is,<br />
<br />
2t 2e e<br />
x(t) =<br />
−t<br />
<br />
c1<br />
⇔ x(t) = X(t)c,<br />
e 2t 2e −t<br />
where we introduced the fundamental and the constant vector<br />
<br />
2t 2e e<br />
X(t) =<br />
−t<br />
<br />
c1<br />
, c = .<br />
c2<br />
e 2t 2e −t<br />
The initial condition fixes the vector c, that is, its components c1, c2, as follows,<br />
x(0) = X(0) c ⇒ c = X(0) −1 x(0).<br />
Since the fundamental matrix at t = 0 has the form,<br />
<br />
2 1<br />
X(0) = ⇒<br />
1 2<br />
X(0) −1 1<br />
=<br />
3<br />
introducing X(0) −1 in the equation for c above we get<br />
c = 1<br />
<br />
2<br />
3 −1<br />
<br />
−1 1 −1<br />
=<br />
2 5 3<br />
⇒<br />
c2<br />
<br />
2 −1<br />
,<br />
−1 2<br />
c1 = −1,<br />
c 2 = 3.<br />
We conclude that the solution to the initial value problem above is given by<br />
<br />
2<br />
x(t) = − e<br />
1<br />
2t <br />
1<br />
+ 3 e<br />
2<br />
−t .<br />
⊳
182 G. NAGY – ODE july 2, 2013<br />
5.4.5. Exercises.<br />
5.4.1.- . 5.4.2.- .
G. NAGY – ODE July 2, 2013 183<br />
5.5. Diagonalizable matrices<br />
We continue with the review on Linear Algebra we started in Sections 5.2 and 5.3. We<br />
have seen that a square matrix is a function on the space <strong>of</strong> vectors. The matrix acts on a<br />
vector and the result is another vector. In this section we see that, given an n × n matrix,<br />
there may exist lines in R n that are left invariant under the action <strong>of</strong> the matrix. The<br />
matrix acting on a non-zero vectors in such line is proportional to that vector. The vector<br />
is called an eigenvector <strong>of</strong> the matrix, and the proportionality factor is called an eigenvalue.<br />
If an n × n matrix has a linearly independent set containing n eigenvectors, then we call<br />
that matrix diagonalizable. In the next Section we will study linear <strong>differential</strong> systems<br />
with constant coefficients having a diagonalizable coefficients matrix. We will see that it is<br />
fairly simple to find solutions to such <strong>differential</strong> systems. The solutions can be expressed in<br />
terms <strong>of</strong> the exponential <strong>of</strong> the coefficient matrix. For that reason we study in this Section<br />
how to compute exponentials <strong>of</strong> square diagonalizable matrices.<br />
5.5.1. Eigenvalues and eigenvectors. When a square matrix acts on a vector the result<br />
is another vector that, more <strong>of</strong>ten than not, points in a different direction from the original<br />
vector. However, there may exist vectors whose direction is not changed by the matrix. We<br />
give these vectors a name.<br />
Definition 5.5.1. A non-zero n-vector v and a number λ are respectively called an eigenvector<br />
and eigenvalue <strong>of</strong> an n × n matrix A iff the following equation holds,<br />
Av = λv.<br />
We see that an eigenvector v determines a particular direction in the space R n , given by<br />
(av) for a ∈ R, that remains invariant under the action <strong>of</strong> the function given by matrix A.<br />
That is, the result <strong>of</strong> matrix A acting on any vector (av) on the line determined by v is<br />
again a vector on the same line, as the following calculation shows it,<br />
A(av) = a Av = aλv = λ(av).<br />
Example 5.5.1: Verify that the pair λ1, v1 and the pair λ2, v2 are eigenvalue and eigenvector<br />
pairs <strong>of</strong> matrix A given below,<br />
A =<br />
<br />
1 3<br />
,<br />
3 1<br />
⎧<br />
<br />
1<br />
⎪⎨ λ1 = 4 v1 = ,<br />
1<br />
<br />
⎪⎩<br />
−1<br />
λ2 = −2 v2 = .<br />
1<br />
Solution: We just must verify the definition <strong>of</strong> eigenvalue and eigenvector given above.<br />
We start with the first pair,<br />
<br />
1 3 1 4 1<br />
Av1 =<br />
= = 4 = λ1v1<br />
3 1 1 4 1<br />
⇒ Av1 = λ1v1.<br />
A similar calculation for the second pair implies,<br />
<br />
1 3 −1 2 −1<br />
Av2 =<br />
= = −2 = λ<br />
3 1 1 −2 1<br />
2v2 ⇒ Av2 = λ2v2. Example 5.5.2: Find the eigenvalues and eigenvectors <strong>of</strong> the matrix A =<br />
<br />
0 1<br />
.<br />
1 0<br />
Solution: This is the matrix given in Example 5.3.1. The action <strong>of</strong> this matrix on the<br />
plane is a reflection along the line x 1 = x 2, as it was shown in Fig. 24. Therefore, this line<br />
⊳
184 G. NAGY – ODE july 2, 2013<br />
x 1 = x 2 is left invariant under the action <strong>of</strong> this matrix. This property suggests that an<br />
eigenvector is any vector on that line, for example<br />
<br />
1 0 1 1 1<br />
v1 = ⇒<br />
= ⇒ λ1 = 1.<br />
1 1 0 1 1<br />
<br />
1<br />
So, we have found one eigenvalue-eigenvector pair: λ1 = 1, with v1 = . We remark that<br />
1<br />
any non-zero vector proportional to v1 is also an eigenvector. Another choice fo eigenvalue-<br />
3<br />
eigenvector pair is λ1 = 1, with v1 = . It is not so easy to find a second eigenvector<br />
3<br />
which does not belong to the line determined by v1. One way to find such eigenvector is<br />
noticing that the line perpendicular to the line x1 = x2 is also left invariant by matrix A.<br />
Therefore, any non-zero vector on that line must be an eigenvector. For example the vector<br />
v2 below, since<br />
<br />
−1 0 1 −1 1 −1<br />
v2 = ⇒<br />
= = (−1) ⇒ λ2 = −1.<br />
1 1 0 1 −1<br />
1<br />
<br />
−1<br />
So, we have found a second eigenvalue-eigenvector pair: λ2 = −1, with v2 = . These<br />
1<br />
two eigenvectors are displayed on Fig. 26. ⊳<br />
v<br />
2<br />
x<br />
2<br />
Ax<br />
x = x<br />
Av = v<br />
1<br />
1 2<br />
2<br />
1<br />
Av = −v<br />
2 2<br />
x<br />
x 1<br />
Figure 26. The first picture shows the eigenvalues and eigenvectors <strong>of</strong><br />
the matrix in Example 5.5.2. The second picture shows that the matrix<br />
in Example 5.5.3 makes a counterclockwise rotation by an angle θ, which<br />
proves that this matrix does not have eigenvalues or eigenvectors.<br />
There exist matrices that do not have eigenvalues and eigenvectors, as it is show in the<br />
example below.<br />
<br />
cos(θ) − sin(θ)<br />
Example 5.5.3: Fix any number θ ∈ (0, 2π) and define the matrix A =<br />
.<br />
sin(θ) cos(θ)<br />
Show that A has no real eigenvalues.<br />
Solution: One can compute the action <strong>of</strong> matrix A on several vectors and verify that the<br />
action <strong>of</strong> this matrix on the plane is a rotation counterclockwise by and angle θ, as shown<br />
in Fig. 26. A particular case <strong>of</strong> this matrix was shown in Example 5.3.2, where θ = π/2.<br />
Since eigenvectors <strong>of</strong> a matrix determine directions which are left invariant by the action <strong>of</strong><br />
the matrix, and a rotation does not have such directions, we conclude that the matrix A<br />
above does not have eigenvectors and so it does not have eigenvalues either. ⊳<br />
x<br />
Ax<br />
0<br />
x<br />
x 1
G. NAGY – ODE July 2, 2013 185<br />
Remark: We will show that matrix A in Example 5.5.3 has complex-valued eigenvalues.<br />
We now describe a method to find eigenvalue-eigenvector pairs <strong>of</strong> a matrix, if they exit.<br />
In other words, we are going to solve the eigenvalue-eigenvector problem: Given an n × n<br />
matrix A find, if possible, all its eigenvalues and eigenvectors, that is, all pairs λ and v = 0<br />
solutions <strong>of</strong> the equation<br />
Av = λv.<br />
This problem is more complicated than finding the solution x to a linear system Ax = b,<br />
where A and b are known. In the eigenvalue-eigenvector problem above neither λ nor v are<br />
known. To solve the eigenvalue-eigenvector problem for a matrix A we proceed as follows:<br />
(a) First, find the eigenvalues λ;<br />
(b) Second, for each eigenvalue λ find the corresponding eigenvectors v.<br />
The following result summarizes a way to solve the steps above.<br />
Theorem 5.5.2 (Eigenvalues-eigenvectors).<br />
(a) The number λ is an eigenvalue <strong>of</strong> an n × n matrix A iff holds,<br />
det(A − λI) = 0. (5.5.1)<br />
(b) Given an eigenvalue λ <strong>of</strong> an n × n matrix A, the corresponding eigenvectors v are the<br />
non-zero solutions to the homogeneous linear system<br />
(A − λI)v = 0. (5.5.2)<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.5.2: The number λ and the non-zero vector v are an eigenvalueeigenvector<br />
pair <strong>of</strong> matrix A iff holds<br />
Av = λv ⇔ (A − λI)v = 0,<br />
where I is the n × n identity matrix. Since v = 0, the last equation above says that the<br />
columns <strong>of</strong> the matrix (A − λI) are linearly dependent. This last property is equivalent, by<br />
Theorem 5.3.12, to the equation<br />
det(A − λI) = 0,<br />
which is the equation that determines the eigenvalues λ. Once this equation is solved,<br />
substitute each solution λ back into the original eigenvalue-eigenvector equation<br />
(A − λI)v = 0.<br />
Since λ is known, this is a linear homogeneous system for the eigenvector components. It<br />
always has non-zero solutions, since λ is precisely the number that makes the coefficient<br />
matrix (A − λI) not invertible. This establishes the Theorem. <br />
<br />
1 3<br />
Example 5.5.4: Find the eigenvalues λ and eigenvectors v <strong>of</strong> the matrix A = .<br />
3 1<br />
Solution: We first find the eigenvalues as the solutions <strong>of</strong> the Eq. (5.5.1). Compute<br />
<br />
1 3 1 0 1 3 λ 0 (1 − λ) 3<br />
A − λI = − λ = − =<br />
.<br />
3 1 0 1 3 1 0 λ 3 (1 − λ)<br />
Then we compute its determinant,<br />
<br />
<br />
<br />
0 = det(A − λI) = (1<br />
− λ) 3 <br />
<br />
3 (1 − λ) = (λ − 1)2 − 9 ⇒<br />
λ+ = 4,<br />
λ- = −2.<br />
We have obtained two eigenvalues, so now we introduce λ + = 4 into Eq. (5.5.2), that is,<br />
<br />
1 − 4 3 −3 3<br />
A − 4I =<br />
=<br />
.<br />
3 1 − 4 3 −3
186 G. NAGY – ODE july 2, 2013<br />
Then we solve for v + the equation<br />
(A − 4I)v + = 0 ⇔<br />
−3 3<br />
3 −3<br />
v +<br />
1<br />
v +<br />
2<br />
<br />
=<br />
<br />
0<br />
.<br />
0<br />
The solution can be found using Gauss elimination operations, as follows,<br />
<br />
+<br />
−3 3 1 −1 1 −1<br />
v 1 = v<br />
→ →<br />
⇒<br />
3 −3 3 −3 0 0<br />
+<br />
2,<br />
free.<br />
Al solutions to the equation above are then given by<br />
v + <br />
1<br />
= =<br />
1<br />
v +<br />
2<br />
v +<br />
2<br />
v +<br />
2 ⇒ v + =<br />
<br />
1<br />
,<br />
1<br />
where we have chosen v +<br />
2 = 1. A similar calculation provides the eigenvector v - associated<br />
with the eigenvalue λ- = −2, that is, first compute the matrix<br />
<br />
3 3<br />
A + 2I =<br />
3 3<br />
then we solve for v - the equation<br />
(A + 2I)v - = 0 ⇔<br />
3 3<br />
3 3<br />
v -<br />
1<br />
v -<br />
2<br />
<br />
=<br />
v +<br />
2<br />
<br />
0<br />
.<br />
0<br />
The solution can be found using Gauss elimination operations, as follows,<br />
<br />
-<br />
3 3 1 1 1 1<br />
v1 = −v<br />
→ → ⇒<br />
3 3 3 3 0 0<br />
-<br />
2,<br />
free.<br />
Al solutions to the equation above are then given by<br />
v - <br />
−1<br />
= =<br />
1<br />
−v -<br />
2<br />
v -<br />
2<br />
v -<br />
2 ⇒ v - =<br />
v -<br />
2<br />
<br />
−1<br />
,<br />
1<br />
where we have chosen v -<br />
2 = 1. We therefore conclude that the eigenvalues and eigenvectors<br />
<strong>of</strong> the matrix A above are given by<br />
λ+ = 4, v + =<br />
<br />
1<br />
, λ- = −2, v<br />
1<br />
- =<br />
<br />
−1<br />
.<br />
1<br />
It is useful to introduce few more concepts, that are common in the literature.<br />
Definition 5.5.3. The characteristic polynomial <strong>of</strong> an n × n matrix A is the function<br />
p(λ) = det(A − λI).<br />
Example 5.5.5: Find the characteristic polynomial <strong>of</strong> matrix A =<br />
<br />
1 3<br />
.<br />
3 1<br />
Solution: We need to compute the determinant<br />
<br />
<br />
<br />
p(λ) = det(A − λI) = (1<br />
− λ) 3 <br />
<br />
3 (1 − λ) = (1 − λ)2 − 9 = λ 2 − 2λ + 1 − 9.<br />
We conclude that the characteristic polynomial is p(λ) = λ 2 − 2λ − 8. ⊳<br />
Since the matrix A in this example is 2 × 2, its characteristic polynomial has degree two.<br />
One can show that the characteristic polynomial <strong>of</strong> an n × n matrix has degree n. The<br />
eigenvalues <strong>of</strong> the matrix are the roots <strong>of</strong> the characteristic polynomial. Different matrices<br />
may have different types <strong>of</strong> roots, so we try to classify these roots in the following definition.<br />
⊳
G. NAGY – ODE July 2, 2013 187<br />
Definition 5.5.4. Given an n × n matrix A with eigenvalues λi, with i = 1, · · · , k n, it<br />
is always possible to express the matrix characteristic polynomial as<br />
p(λ) = (λ − λ 1) r1 · · · (λ − λk) rk .<br />
The number ri is called the algebraic multiplicity <strong>of</strong> the eigenvalue λi. Furthermore, the<br />
geometric multiplicity <strong>of</strong> the eigenvalue λi, denoted as si, is the maximum number <strong>of</strong><br />
eigenvectors vectors corresponding to that eigenvalue λi forming a linearly independent set.<br />
Example 5.5.6: Find the eigenvalues algebraic and geometric multiplicities <strong>of</strong> the matrix<br />
<br />
1 3<br />
A =<br />
3 1<br />
Solution: In order to find the algebraic multiplicity <strong>of</strong> the eigenvalues we need first to find<br />
the eigenvalues. We now that the characteristic polynomial <strong>of</strong> this matrix is given by<br />
<br />
<br />
<br />
p(λ) = (1<br />
− λ) 3 <br />
<br />
3 (1 − λ) = (λ − 1)2 − 9.<br />
The roots <strong>of</strong> this polynomial are λ1 = 4 and λ2 = −2, so we know that p(λ) can be rewritten<br />
in the following way,<br />
p(λ) = (λ − 4)(λ + 2).<br />
We conclude that the algebraic multiplicity <strong>of</strong> the eigenvalues are both one, that is,<br />
λ1 = 4, r1 = 1, and λ2 = −2, r2 = 1.<br />
In order to find the geometric multiplicities <strong>of</strong> matrix eigenvalues we need first to find the<br />
matrix eigenvectors. This part <strong>of</strong> the work was already done in the Example 5.5.4 above<br />
and the result is<br />
λ1 = 4, v (1) =<br />
<br />
1<br />
, λ2 = −2, v<br />
1<br />
(2) =<br />
<br />
−1<br />
.<br />
1<br />
From this expression we conclude that the geometric multiplicities for each eigenvalue are<br />
just one, that is,<br />
λ1 = 4, s 1 = 1, and λ2 = −2, s 2 = 1.<br />
⊳<br />
The following example shows that two matrices can have the same eigenvalues, and so the<br />
same algebraic multiplicities, but different eigenvectors with different geometric multiplicities.<br />
⎡<br />
3 0<br />
⎤<br />
1<br />
Example 5.5.7: Find the eigenvalues and eigenvectors <strong>of</strong> the matrix A = ⎣0<br />
3 2⎦.<br />
0 0 1<br />
Solution: We start finding the eigenvalues, the roots <strong>of</strong> the characteristic polynomial<br />
<br />
<br />
<br />
(3<br />
− λ) 0 1 <br />
<br />
p(λ) = <br />
0 (3 − λ) 2 <br />
<br />
0 0 (1 − λ) = −(λ − 1)(λ − 3)2 <br />
λ1 = 1, r1 = 1,<br />
⇒<br />
λ2 = 3, r2 = 2.<br />
We now compute the eigenvector associated with the eigenvalue λ1 = 1, which is the solution<br />
<strong>of</strong> the linear system<br />
(A − I)v (1) = 0 ⇔<br />
⎡<br />
2<br />
⎣0<br />
0<br />
2<br />
⎤ ⎡<br />
1 v<br />
2⎦<br />
⎢<br />
⎣<br />
0 0 0<br />
(1)<br />
1<br />
v (1)<br />
⎤ ⎡ ⎤<br />
0<br />
⎥<br />
⎦ = ⎣0⎦<br />
.<br />
0<br />
2<br />
v (1)<br />
3
188 G. NAGY – ODE july 2, 2013<br />
After the few Gauss elimination operation we obtain the following,<br />
⎡<br />
2<br />
⎣0 0<br />
0<br />
2<br />
0<br />
⎤ ⎡<br />
1<br />
2⎦<br />
→ ⎣<br />
0<br />
1 0 0<br />
0<br />
1<br />
0<br />
⎤<br />
1<br />
2<br />
1⎦<br />
0<br />
⇒<br />
⎧<br />
⎪⎨<br />
v<br />
⎪⎩<br />
(1)<br />
1 = − v(1) 3<br />
2 ,<br />
v (1)<br />
2 = −v (1)<br />
3 ,<br />
free.<br />
Therefore, choosing v (1)<br />
3<br />
v (1)<br />
3<br />
= 2 we obtain that<br />
v (1) ⎡ ⎤<br />
−1<br />
= ⎣−2⎦<br />
,<br />
2<br />
λ1 = 1, r1 = 1, s1 = 1.<br />
In a similar way we now compute the eigenvectors for the eigenvalue λ2 = 3, which are all<br />
solutions <strong>of</strong> the linear system<br />
(A − 3I)v (2) = 0 ⇔<br />
⎡<br />
0<br />
⎣0<br />
0<br />
0<br />
⎤ ⎡<br />
1 v<br />
2 ⎦ ⎢<br />
⎣<br />
0 0 −2<br />
(2)<br />
1<br />
v (2)<br />
⎤ ⎡ ⎤<br />
0<br />
⎥<br />
⎦ = ⎣0⎦<br />
.<br />
0<br />
2<br />
v (2)<br />
3<br />
After the few Gauss elimination operation we obtain the following,<br />
⎡<br />
0<br />
⎣0<br />
0<br />
0<br />
0<br />
0<br />
⎤ ⎡<br />
1 0<br />
2 ⎦ → ⎣0<br />
−2 0<br />
0<br />
0<br />
0<br />
⎤<br />
1<br />
0⎦<br />
0<br />
⇒<br />
⎧<br />
⎪⎨<br />
v<br />
⎪⎩<br />
(2)<br />
1 free,<br />
v (2)<br />
2 free,<br />
= 0.<br />
Therefore, we obtain two linearly independent solutions, the first one v (2) with the choice<br />
v (2)<br />
1 = 1, v (2)<br />
2 = 0, and the second one w (2) with the choice v (2)<br />
1 = 0, v (2)<br />
2 = 1, that is<br />
v (2) ⎡<br />
= ⎣ 1<br />
⎤<br />
0⎦<br />
, w<br />
0<br />
(2) ⎡<br />
= ⎣ 0<br />
⎤<br />
1⎦<br />
, λ2 = 3, r2 = 2, s2 = 2.<br />
0<br />
Summarizing, the matrix in this example has three linearly independent eigenvectors. ⊳<br />
⎡<br />
3 1<br />
⎤<br />
1<br />
Example 5.5.8: Find the eigenvalues and eigenvectors <strong>of</strong> the matrix A = ⎣0<br />
3 2⎦.<br />
0 0 1<br />
Solution: Notice that this matrix has only the coefficient a12 different from the previous<br />
example. Again, we start finding the eigenvalues, which are the roots <strong>of</strong> the characteristic<br />
polynomial<br />
<br />
<br />
<br />
(3<br />
− λ) 1 1 <br />
<br />
p(λ) = <br />
0 (3 − λ) 2 <br />
<br />
0 0 (1 − λ) = −(λ − 1)(λ − 3)2 <br />
λ1 = 1, r1 = 1,<br />
⇒<br />
λ2 = 3, r2 = 2.<br />
So this matrix has the same eigenvalues and algebraic multiplicities as the matrix in the<br />
previous example. We now compute the eigenvector associated with the eigenvalue λ1 = 1,<br />
which is the solution <strong>of</strong> the linear system<br />
⎡ ⎤ ⎡<br />
2 1 1 v<br />
⎢<br />
(1)<br />
⎤ ⎡ ⎤<br />
1 0<br />
⎥<br />
(A − I)v (1) = 0 ⇔<br />
⎣0<br />
2 2⎦<br />
⎣<br />
0 0 0<br />
v (1)<br />
v (2)<br />
3<br />
2<br />
v (1)<br />
3<br />
⎦ = ⎣0⎦<br />
.<br />
0
G. NAGY – ODE July 2, 2013 189<br />
After the few Gauss elimination operation we obtain the following,<br />
⎧<br />
⎡ ⎤ ⎡ ⎤ ⎡ ⎤<br />
2 1 1 1 1 1 1 0 0 ⎪⎨<br />
⎣0<br />
2 2⎦<br />
→ ⎣0<br />
1 1⎦<br />
→ ⎣0<br />
1 1⎦<br />
⇒<br />
0 0 0 0 0 0 0 0 0 ⎪⎩<br />
Therefore, choosing v (1)<br />
3<br />
v (1)<br />
1<br />
v (1)<br />
2<br />
v (1)<br />
3<br />
= 1 we obtain that<br />
v (1) ⎡ ⎤<br />
0<br />
= ⎣−1⎦<br />
,<br />
1<br />
λ1 = 1, r1 = 1, s1 = 1.<br />
= 0,<br />
= −v (1)<br />
3 ,<br />
In a similar way we now compute the eigenvectors for the eigenvalue λ2 = 3. However, in<br />
this case we obtain only one solution, as this calculation shows,<br />
(A − 3I)v (2) ⎡ ⎤ ⎡<br />
0 1 1 v<br />
= 0 ⇔ ⎣0<br />
0 2 ⎦ ⎢<br />
⎣<br />
0 0 −2<br />
(2)<br />
1<br />
v (2)<br />
⎤ ⎡ ⎤<br />
0<br />
⎥<br />
⎦ = ⎣0⎦<br />
.<br />
0<br />
2<br />
v (2)<br />
3<br />
After the few Gauss elimination operation we obtain the following,<br />
⎧<br />
⎡ ⎤ ⎡ ⎤<br />
0 1 1 0 1 0 ⎪⎨<br />
v<br />
⎣0<br />
0 2 ⎦ → ⎣0<br />
0 1⎦<br />
⇒<br />
0 0 −2 0 0 0 ⎪⎩<br />
(2)<br />
1 free,<br />
v (2)<br />
2 = 0,<br />
= 0.<br />
Therefore, we obtain only one linearly independent solution, which corresponds to the choice<br />
v (2)<br />
1 = 1, that is,<br />
v (2) ⎡<br />
= ⎣ 1<br />
⎤<br />
0⎦<br />
,<br />
0<br />
λ2 = 3, r2 = 2, s2 = 1.<br />
Summarizing, the matrix in this example has only two linearly independent eigenvectors,<br />
and in the case <strong>of</strong> the eigenvalue λ2 = 3 we have the strict inequality<br />
1 = s2 < r2 = 2.<br />
5.5.2. Diagonalizable matrices. We first introduce the notion <strong>of</strong> a diagonal matrix. Later<br />
on we define the idea <strong>of</strong> a diagonalizable matrix as a matrix that can be transformed into a<br />
diagonal matrix by a simple transformation.<br />
⎡<br />
⎤<br />
Definition 5.5.5. An n × n matrix A is called diagonal iff holds A =<br />
v (2)<br />
3<br />
free.<br />
⎢<br />
⎣<br />
⊳<br />
a11<br />
.<br />
· · ·<br />
. ..<br />
0<br />
.<br />
⎥<br />
⎦.<br />
0 · · · ann<br />
That is, a matrix is diagonal iff every non-diagonal coefficient vanishes. From now on we<br />
use the following notation for a diagonal matrix A:<br />
A = diag ⎡<br />
a11<br />
⎢<br />
a11, · · · , ann = ⎣ .<br />
· · ·<br />
. ..<br />
0<br />
.<br />
⎤<br />
⎥<br />
⎦ .<br />
0 · · · ann<br />
This notation says that the matrix is diagonal and shows only the diagonal coefficients,<br />
since any other coefficient vanishes. Diagonal matrices are simple to manipulate since they<br />
share many properties with numbers. For example the product <strong>of</strong> two diagonal matrices is
190 G. NAGY – ODE july 2, 2013<br />
commutative. It is simple to compute power functions <strong>of</strong> a diagonal matrix. It is also simple<br />
to compute more involved functions <strong>of</strong> a diagonal matrix, like the exponential function.<br />
Example 5.5.9: For every positive integer n find An <br />
2 0<br />
, where A = .<br />
0 3<br />
Solution: We start computing A2 as follows,<br />
A 2 <br />
2 0 2 0<br />
= AA =<br />
=<br />
0 3 0 3<br />
<br />
2 2 0<br />
0 32 <br />
.<br />
We now compute A3 ,<br />
A 3 = A 2 <br />
2 2 0<br />
A =<br />
0 32 <br />
3 2 0 2 0<br />
=<br />
0 3 0 33 <br />
.<br />
Using induction, it is simple to see that An <br />
n 2 0<br />
=<br />
0 3n <br />
. ⊳<br />
Many properties <strong>of</strong> diagonal matrices are shared by diagonalizable matrices. These are<br />
matrices that can be transformed into a diagonal matrix by a simple transformation.<br />
Definition 5.5.6. An n × n matrix A is called diagonalizable iff there exists an invertible<br />
matrix P and a diagonal matrix D such that<br />
A = P DP −1 .<br />
Systems <strong>of</strong> linear <strong>differential</strong> <strong>equations</strong> are simple to solve in the case that the coefficient<br />
matrix is diagonalizable. In such case, it is simple to decouple the <strong>differential</strong> <strong>equations</strong>.<br />
One solves the decoupled <strong>equations</strong>, and then transforms back to the original unknowns.<br />
Example 5.5.10: We will see later on that matrix A is diagonalizable while B is not, where<br />
<br />
1 3<br />
A = , B =<br />
3 1<br />
1<br />
<br />
3 1<br />
.<br />
2 −1 5<br />
⊳<br />
There is a deep relation between the eigenvalues and eigenvectors <strong>of</strong> a matrix and the<br />
property <strong>of</strong> diagonalizability.<br />
Theorem 5.5.7 (Diagonalizable matrices). An n × n matrix A is diagonalizable iff<br />
matrix A has a linearly independent set <strong>of</strong> n eigenvectors. Furthermore,<br />
A = P DP −1 , P = [v1, · · · , vn], D = diag <br />
λ1, · · · , λn ,<br />
where λi, vi, for i = 1, · · · , n, are eigenvalue-eigenvector pairs <strong>of</strong> matrix A.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.5.7:<br />
(⇒) Since matrix A is diagonalizable, there exist an invertible matrix P and a diagonal<br />
matrix D such that A = P DP −1 . Multiply this equation by P −1 on the left and by P on<br />
the right, we get<br />
D = P −1 AP. (5.5.3)<br />
Since n × n matrix D is diagonal, it has a linearly independent set <strong>of</strong> n eigenvectors, given<br />
by the column vectors <strong>of</strong> the identity matrix, that is,<br />
De (i) = diie (i) , D = diag <br />
d11, · · · , dnn , I = (1) (n) e , · · · , e .<br />
So, the pair dii, e (i) is an eigenvalue-eigenvector pair <strong>of</strong> D, for i = 1 · · · , n. Using this<br />
information in Eq. (5.5.3) we get<br />
(i)<br />
P e ,<br />
diie (i) = De (i) = P −1 AP e (i) ⇒ A P e (i) = dii
G. NAGY – ODE July 2, 2013 191<br />
where the last equation comes from multiplying the former equation by P on the left. This<br />
last equation says that the vectors v (i) = P e (i) are eigenvectors <strong>of</strong> A with eigenvalue dii.<br />
By definition, v (i) is the i-th column <strong>of</strong> matrix P , that is,<br />
P = v (1) , · · · , v (n) .<br />
Since matrix P is invertible, the eigenvectors set {v (1) , · · · , v (n) } is linearly independent.<br />
This establishes this part <strong>of</strong> the Theorem.<br />
(⇐) Let λi, v (i) be eigenvalue-eigenvector pairs <strong>of</strong> matrix A, for i = 1, · · · , n. Now use the<br />
eigenvectors to construct matrix P = v (1) , · · · , v (n) . This matrix is invertible, since the<br />
eigenvector set {v (1) , · · · , v (n) } is linearly independent. We now show that matrix P −1 AP<br />
is diagonal. We start computing the product<br />
that is,<br />
AP = A v (1) , · · · , v (n) = Av (1) , · · · , Av (n) , = λ 1v (1) · · · , λnv (n) .<br />
P −1 AP = P −1 λ 1v (1) , · · · , λnv (n) = λ 1P −1 v (1) , · · · , λnP −1 v (n) .<br />
At this point it is useful to recall that P −1 is the inverse <strong>of</strong> P ,<br />
I = P −1 P ⇔ e (1) , · · · , e (n) = P −1 v (1) , · · · , v (n) = P −1 v (1) , · · · , P −1 v (n) .<br />
We conclude that e (i) = P −1 v (i) , for i = 1 · · · , n. Using these <strong>equations</strong> in the equation for<br />
P −1 AP we get<br />
P −1 AP = λ1e (1) , · · · , λne (n) = diag <br />
λ1, · · · , λn .<br />
Denoting D = diag <br />
−1 λ1, · · · , λn we conlude that P AP = D, or equivalently<br />
<br />
.<br />
A = P DP −1 , P = v (1) , · · · , v (n) , D = diag λ 1, · · · , λn<br />
This means that A is diagonalizable. This establishes the Theorem. <br />
<br />
1<br />
Example 5.5.11: Show that matrix A =<br />
3<br />
<br />
3<br />
is diagonalizable.<br />
1<br />
Solution: We known that the eigenvalue eigenvector pairs are<br />
λ1 = 4,<br />
<br />
1<br />
v1 =<br />
1<br />
and λ2 = −2, v2 =<br />
Introduce P and D as follows,<br />
<br />
1 −1<br />
P =<br />
1 1<br />
⇒ P −1 = 1<br />
2<br />
<br />
1 1<br />
, D =<br />
−1 1<br />
We must show that A = P DP −1 . This is indeed the case, since<br />
P DP −1 <br />
1<br />
=<br />
1<br />
<br />
−1 4<br />
1 0<br />
<br />
0 1 1<br />
−2 2 −1<br />
<br />
1<br />
.<br />
1<br />
<br />
−1<br />
.<br />
1<br />
<br />
4 0<br />
.<br />
0 −2<br />
P DP −1 <br />
4 2 1 1 1 2 1 1 1<br />
=<br />
=<br />
4 −2 2 −1 1 2 −1 −1 1<br />
We conclude, P DP −1 <br />
1 3<br />
= ⇒ P DP<br />
3 1<br />
−1 = A, that is, A is diagonalizable. ⊳
192 G. NAGY – ODE july 2, 2013<br />
Example 5.5.12: Show that matrix B = 1<br />
2<br />
<br />
3 1<br />
is not diagonalizable.<br />
−1 5<br />
Solution: We first compute the matrix eigenvalues. The characteristic polynomial is<br />
<br />
<br />
<br />
3<br />
<br />
1<br />
− λ<br />
p(λ) = 2 2<br />
<br />
− 1<br />
<br />
<br />
<br />
3<br />
<br />
5<br />
<br />
<br />
5<br />
<br />
= − λ − λ +<br />
− λ<br />
2 2<br />
<br />
2 2 1<br />
4 = λ2 − 4λ + 4.<br />
The roots <strong>of</strong> the characteristic polynomial are computed in the usual way,<br />
λ = 1<br />
√ <br />
4 ± 16 − 16<br />
2<br />
⇒ λ = 2, r = 2.<br />
We have obtained a single eigenvalue with algebraic multiplicity r = 2. The associated<br />
eigenvectors are computed as the solutions to the equation (A − 2I)v = 0. Then,<br />
⎡<br />
3<br />
<br />
⎢ − 2<br />
(A − 2I) = ⎣ 2<br />
−<br />
1<br />
2<br />
1<br />
2<br />
⎡<br />
⎤<br />
⎢−<br />
⎥ ⎢<br />
<br />
5<br />
⎦<br />
= ⎢<br />
− 2 ⎣<br />
2 1<br />
2<br />
−<br />
1<br />
2<br />
1<br />
2<br />
⎤<br />
⎥<br />
1<br />
⎥<br />
⎦<br />
2<br />
→<br />
<br />
1<br />
0<br />
<br />
−1<br />
0<br />
⇒<br />
<br />
1<br />
v = ,<br />
1<br />
s = 1.<br />
We conclude that the biggest linearly independent set <strong>of</strong> eigenvalues for the 2 × 2 matrix B<br />
contains only one vector, insted <strong>of</strong> two. Therefore, matrix B is not diagonalizable. ⊳<br />
Because <strong>of</strong> Theorem 5.5.7 it is important in applications to know whether an n×n matrix<br />
has a linearly independent set <strong>of</strong> n eigenvectors. More <strong>of</strong>ten than not there is no simple way<br />
to check this property other than to compute all the matrix eigenvectors. However, there is<br />
a simple particular case: When the n × n matrix has n different eigenvalues. In such case<br />
we do not need to compute the eigenvectors. The following result says that such matrix<br />
always have a linearly independent set <strong>of</strong> n eigenvectors, and so, by Theorem 5.5.7, will be<br />
diagonalizable.<br />
Theorem 5.5.8 (Different eigenvalues). If an n × n matrix has n different eigenvalues,<br />
then this matrix has a linearly independent set <strong>of</strong> n eigenvectors.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.5.8: Let λ 1, · · · , λn be the eigenvalues <strong>of</strong> an n × n matrix A,<br />
all different from each other. Let v (1) , · · · , v (n) the corresponding eigenvectors, that is,<br />
Av (i) = λiv (i) , with i = 1, · · · , n. We have to show that the set {v (1) , · · · , v (n) } is linearly<br />
independent. We assume that the opposite is true and we obtain a contradiction. Let us<br />
assume that the set above is linearly dependent, that is, there exists constants c1, · · · , cn,<br />
not all zero, such that,<br />
c1v (1) + · · · + cnv (n) = 0. (5.5.4)<br />
Let us name the eigenvalues and eigenvectors such that c1 = 0. Now, multiply the equation<br />
above by the matrix A, the result is,<br />
c 1λ 1v (1) + · · · + cnλnv (n) = 0.<br />
Multiply Eq. (5.5.4) by the eigenvalue λn, the result is,<br />
c1λnv (1) + · · · + cnλnv (n) = 0.<br />
Subtract the second from the first <strong>of</strong> the <strong>equations</strong> above, then the last term on the righthand<br />
sides cancels out, and we obtain,<br />
c1(λ1 − λn)v (1) + · · · + cn−1(λn−1 − λn)v (n−1) = 0. (5.5.5)
G. NAGY – ODE July 2, 2013 193<br />
Repeat the whole procedure starting with Eq. (5.5.5), that is, multiply this later equation<br />
by matrix A and also by λn−1, then subtract the second from the first, the result is,<br />
c1(λ1 − λn)(λ1 − λn−1)v (1) + · · · + cn−2(λn−2 − λn)(λn−2 − λn−1)v (n−2) = 0.<br />
Repeat the whole procedure a total <strong>of</strong> n − 1 times, in the last step we obtain the equation<br />
c1(λ1 − λn)(λ1 − λn−1) · · · (λ1 − λ2)v (1) = 0.<br />
Since all the eigenvalues are different, we conclude that c 1 = 0, however this contradicts our<br />
assumption that c1 = 0. Therefore, the set <strong>of</strong> n eigenvectors must be linearly independent.<br />
Example 5.5.13: Is matrix A =<br />
<br />
1 1<br />
diagonalizable?<br />
1 1<br />
Solution: We compute the matrix eigenvalues, starting with the characteristic polynomial,<br />
<br />
<br />
<br />
p(λ) = (1<br />
− λ) 1 <br />
<br />
1 (1 − λ) = (1 − λ)2 − 1 = λ 2 − 2λ ⇒ p(λ) = λ(λ − 2).<br />
The roots <strong>of</strong> the characteristic polynomial are the matrix eigenvalues,<br />
λ 1 = 0, λ 2 = 2.<br />
The eigenvalues are different, so by Theorem 5.5.8, matrix A is diagonalizable. ⊳<br />
5.5.3. The exponential <strong>of</strong> a matrix. Functions <strong>of</strong> a diagonalizable matrix are simple<br />
to compute in the case that the function admits a Taylor series expansion. One <strong>of</strong> such<br />
functions is the exponential function. In this Section we compute the exponential <strong>of</strong> a<br />
diagonalizable matrix. However, we emphasize that this method can be trivially extended<br />
to any function admitting a Taylor series expansion. A key result needed to compute the<br />
Taylor series expansion <strong>of</strong> a matrix is the n-power <strong>of</strong> the matrix.<br />
Theorem 5.5.9. If the n × n matrix A is diagonalizable, with invertible matrix P and<br />
diagonal matrix D satisfying A = P DP −1 , then for every integer n 1 holds<br />
A n = P D n P −1 . (5.5.6)<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.5.9: It is not difficult to generalize the calculation done in Example<br />
5.5.9 to obtain the n-th power <strong>of</strong> a diagonal matrix D = diag <br />
d1, · · · , dn , and the result<br />
is another diagonal matrix given by<br />
<br />
.<br />
D n = diag d n 1 , · · · , d n n<br />
We use this result and induction in n to prove Eq.(5.5.6). Since the case n = 1 is trivially<br />
true, we start computing the case n = 2. We get<br />
A 2 = P DP −1 2 = P DP −1 P DP −1 = P DDP −1 ⇒ A 2 = P D 2 P −1 ,<br />
that is, Eq. (5.5.6) holds for n = 2. Now, assuming that this Equation holds for n, we show<br />
that it also holds for n + 1. Indeed,<br />
A (n+1) = A n A = P D n P −1 P DP −1 = P D n P −1 P DP −1 = P D n DP −1 .<br />
We conclude that A (n+1) = P D (n+1) P −1 . This establishes the Theorem. <br />
The exponential function f(x) = e x is usually defined as the inverse <strong>of</strong> the natural<br />
logarithm function g(x) = ln(x), which in turns is defined as the area under the graph <strong>of</strong><br />
the function h(x) = 1/x from 1 to x, that is,<br />
ln(x) =<br />
x<br />
1<br />
1<br />
dy, x ∈ (0, ∞).<br />
y
194 G. NAGY – ODE july 2, 2013<br />
It is not clear how to extend to matrices this way <strong>of</strong> defining the exponential function on<br />
real numbers. However, the exponential function on real numbers satisfies several identities<br />
that can be used as definition for the exponential on matrices. One <strong>of</strong> these identities is the<br />
Taylor series expansion<br />
e x =<br />
∞<br />
k=0<br />
x k<br />
k!<br />
= 1 + x + x2<br />
2!<br />
+ x3<br />
3!<br />
+ · · · .<br />
This identity is the key to generalize the exponential function to diagonalizable matrices.<br />
Definition 5.5.10. The exponential <strong>of</strong> a square matrix A is defined as the infinite sum<br />
e A ∞ A<br />
=<br />
n<br />
. (5.5.7)<br />
n!<br />
n=0<br />
One must show that the definition makes sense, that is, that the infinite sum in Eq. (5.5.7)<br />
converges. We show in these notes that this is the case when matrix A is diagonal and when<br />
matrix A is diagonalizable. The case <strong>of</strong> non-diagonalizable matrix is more difficult to prove,<br />
and we do not do it in these notes.<br />
Theorem 5.5.11. If D = diag <br />
D d1 dn<br />
d1, · · · , dn , then holds e = diag e , · · · , e .<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.5.11: It is simple to see that the infinite sum in Eq (5.5.7) converges<br />
for diagonal matrices. Start computing<br />
e D ∞ 1 <br />
∞ k 1<br />
= diag d1, · · · , dn =<br />
k!<br />
k! diag (d1) k , · · · , (dn) k .<br />
k=0<br />
Since each matrix in the sum on the far right above is diagonal, then holds<br />
e D ∞ (d1)<br />
= diag<br />
k ∞ (dn)<br />
, · · · ,<br />
k!<br />
k <br />
.<br />
k!<br />
k=0<br />
∞ (di)<br />
Now, each sum in the diagonal <strong>of</strong> matrix above satisfies<br />
k=0<br />
k<br />
k! = edi . Therefore, we<br />
arrive to the equation eD = diag ed1 , · · · , edn <br />
. This establishes the Theorem. <br />
Example 5.5.14: Compute eA <br />
2<br />
, where A =<br />
0<br />
<br />
0<br />
.<br />
7<br />
Solution: Theorem 5.5.11 implies that eA <br />
2 e<br />
=<br />
0<br />
0<br />
e7 <br />
. ⊳<br />
The case <strong>of</strong> diagonalizable matrices is more involved and is summarized below.<br />
Theorem 5.5.12. If the n × n matrix A is diagonalizable, with invertible matrix P and<br />
diagonal matrix D satisfying A = P DP −1 , then the exponential <strong>of</strong> matrix A is given by<br />
k=0<br />
k=0<br />
e A = P e D P −1 . (5.5.8)<br />
The formula above says that to find the exponential <strong>of</strong> a diagonalizable matrix there is no<br />
need to compute the infinite sum in Definition 5.5.10. To compute the exponential <strong>of</strong> a<br />
diagonalizable matrix it is only needed to compute the product <strong>of</strong> three matrices. It also<br />
says that to compute the exponential <strong>of</strong> a diagonalizable matrix we need to compute the<br />
eigenvalues and eigenvectors <strong>of</strong> the matrix.
k=0<br />
G. NAGY – ODE July 2, 2013 195<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.5.12: We again start with Eq. (5.5.7),<br />
e A ∞ 1<br />
=<br />
k! An ∞ 1<br />
=<br />
k! (P DP −1 ) n ∞ 1<br />
=<br />
k! (P DnP −1 ),<br />
k=0<br />
where the last step comes from Theorem 5.5.9. Now, in the expression on the far right we<br />
can take common factor P on the left and P −1 on the right, that is,<br />
e A ∞ 1<br />
= P<br />
k! Dn P −1 .<br />
k=0<br />
The terms between brackets sum up to the exponential <strong>of</strong> the diagonal matrix D, that is,<br />
e A = P e D P −1 .<br />
Since the exponential <strong>of</strong> a diagonal matrix is computed in Theorem 5.5.11, this establishes<br />
the Theorem.<br />
Example 5.5.15: Compute e<br />
<br />
A <br />
1<br />
, where A =<br />
3<br />
<br />
3<br />
.<br />
1<br />
Solution: To compute the exponential <strong>of</strong> matrix A above we need the decomposition<br />
A = P DP −1 . Matrices P and D are constructed with the eigenvectors and eigenvalues <strong>of</strong><br />
k=0<br />
matrix A, respectively. From Example 5.5.4 we know that<br />
<br />
1<br />
λ1 = 4, v1 =<br />
1<br />
Introduce P and D as follows,<br />
<br />
1 −1<br />
P =<br />
1 1<br />
⇒ P −1 = 1<br />
2<br />
Then, the exponential function is given by<br />
e At = P e Dt P −1 =<br />
and λ2 = −2, v2 =<br />
1 −1<br />
1 1<br />
<br />
1 1<br />
, D =<br />
−1 1<br />
e 4t 0<br />
0 e −2t<br />
<br />
1<br />
2<br />
<br />
−1<br />
.<br />
1<br />
<br />
4 0<br />
.<br />
0 −2<br />
<br />
1 1<br />
.<br />
−1 1<br />
Usually one leaves the function in this form. If we multiply the three matrices out we get<br />
e At = 1<br />
<br />
4t −2t 4t −2t (e + e ) (e − e )<br />
2 (e4t − e−2t ) (e4t + e−2t <br />
.<br />
)<br />
⊳<br />
The exponential <strong>of</strong> an n × n matrix A can be used to define the matrix-valued function<br />
with values e At . In the case that the matrix A is diagonalizable we obtain<br />
e At = P e Dt P −1 ,<br />
where eDt = diag ed1t , · · · , ednt and D = diag <br />
d1, · · · , dn . It is not difficult to show that<br />
de At<br />
dt = A eAt = e At A.<br />
This exponential function will be useful to express solutions <strong>of</strong> a linear homogeneous <strong>differential</strong><br />
system x ′ = Ax. This is the subject <strong>of</strong> the next Section.
196 G. NAGY – ODE july 2, 2013<br />
5.5.4. Exercises.<br />
5.5.1.- . 5.5.2.- .
G. NAGY – ODE July 2, 2013 197<br />
5.6. Constant coefficient <strong>differential</strong> systems<br />
In the rest <strong>of</strong> this Chapter we study linear <strong>differential</strong> systems with constant coefficients.<br />
More precisely, given an n × n constant matrix A, an n-vector valued function g, and a<br />
constant n-vector x0, find an n-vector valued function x solution to the initial value problem<br />
x ′ (t) = A x(t) + g(t), x(t 0) = x 0. (5.6.1)<br />
We have mentioned in Section 5.1 that the Picard-Lindelöf Theorem 5.1.5 applies to linear<br />
<strong>differential</strong> systems, hence the initial value problem above always has a unique solution.<br />
However, we have not said in a precise way how to obtain such solution. In this Section we<br />
present a formula for the solution <strong>of</strong> the initial value problem in Eq. (5.6.1) in the case that<br />
the coefficient matrix A is diagonalizable. We start studying the particular case <strong>of</strong> homogeneous<br />
<strong>equations</strong>, that is, g = 0. We then study non-homogeneous <strong>equations</strong> with constant<br />
source vector g. We finally study non-homogeneous <strong>equations</strong> with continuous source function<br />
g. The expression for the solutions in these cases will be a generalization <strong>of</strong> the solution<br />
for linear, constant coefficients, scalar <strong>equations</strong> given in Section 1.1, Eq. (1.1.10).<br />
5.6.1. Diagonalizable homogeneous systems. We start studying the particular case <strong>of</strong><br />
homogeneous linear <strong>differential</strong> systems. We restrict our study to systems with diagonalizable<br />
coefficient matrices. In this case it is easy to show that the solutions to <strong>differential</strong><br />
systems are given by a simple generalization <strong>of</strong> the solution formula for scalar <strong>differential</strong><br />
<strong>equations</strong>, Eq. (1.1.10) in the case b = 0. The solution formula for <strong>differential</strong> systems involves<br />
the exponential <strong>of</strong> the equation coefficients, that is, the exponential <strong>of</strong> the coefficient<br />
matrix.<br />
Theorem 5.6.1 (Exponential expression). If the n × n constant matrix A is diagonalizable,<br />
then the homogeneous initial value problem<br />
x ′ (t) = A x(t), x(t0) = x0, (5.6.2)<br />
has a unique solution for every initial condition x0 ∈ R n given by<br />
x(t) = e A(t−t0) x0. (5.6.3)<br />
Because the coefficient matrix A is diagonalizable, we can transform the coupled system<br />
for the unknown vector components xi into a decoupled system for appropriately chosen<br />
unknonwns. We solve each <strong>of</strong> the n decoupled <strong>equations</strong> using Eq. (1.1.10), and then we<br />
transform this solution back to the original unknown vector components xi.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.6.1: Since the coefficient matrix A is diagonalizable, there exist an<br />
n×n invertible matrix P and an n×n diagonal matrix D such that A = P DP −1 . Introducing<br />
this information into the <strong>differential</strong> equation in Eq. (1.1.10), and then multiplying the whole<br />
equation by P −1 , we obtain,<br />
P −1 x ′ (t) = P −1 P DP −1 x(t).<br />
Since matrix A is constant, so is P and D. In particular P −1 x ′ = P −1 x ′ . Therefore,<br />
(P −1 x) ′ = D P −1 x .<br />
Introduce the new unknown function y = P −1 x , then the equation above has the form<br />
y ′ (t) = D y(t).<br />
Now transform the initial condition too, that is, P −1 x(t 0) = P −1 x 0. Introducing the notation<br />
y 0 = P −1 x0, we get the initial condition<br />
y(t 0) = y 0.
198 G. NAGY – ODE july 2, 2013<br />
Since the coefficient matrix D is diagonal, the initial value problem above for the unknown<br />
y is decoupled. That is, expressing D = diag[λ1, · · · , λn], y 0 = [y0i], and y(t) = [yi(t)], then<br />
for each i = 1, · · · , n holds that the initial value problem above has the form<br />
y ′ i(t) = λi yi(t), yi(t 0) = y 0i.<br />
Since the i-th equation involves only the unknown yi, we only need to solve n decoupled<br />
scalar <strong>equations</strong>, each one for a single unknown function. The solution <strong>of</strong> each initial value<br />
problem above can be found using Eq. (1.1.10), that is,<br />
yi(t) = e λi(t−t0) y0i.<br />
Once the <strong>equations</strong> are solved, we rewrite the solutions above back in vector notation,<br />
y(t) = e D(t−t0) y0,<br />
where we have used the expression e D(t−t0) = diag e λ1(t−t0) , · · · , e λn(t−t0) . We now multiply<br />
the whole equation by matrix P and we recall that P −1 P = In, then<br />
P y(t) = P e D(t−t0) (P −1 P )y0.<br />
Recalling that P y = x, P y 0 = x0, and e At = P e Dt P −1 , we obtain<br />
x(t) = e A(t−t0) x0.<br />
This establishes the Theorem. <br />
The expression for the solution given in Eq. (5.6.3) is beautiful, simple. However, the<br />
exponential <strong>of</strong> a matrix requires some work to be computed, as the following example shows.<br />
Example 5.6.1: Find the vector-valued function x solution to the <strong>differential</strong> system<br />
x ′ <br />
3 1 2<br />
= A x, x(0) = , A = .<br />
2 2 1<br />
Solution: First we need to find out whether the coefficient matrix A is diagonalizable or<br />
not. Theorem 5.5.7 says that a 2 × 2 matrix is diagonalizable iff there exists a linearly<br />
independent set <strong>of</strong> two eigenvectors. So we start computing the matrix eigenvalues, which<br />
are the roots <strong>of</strong> the characteristic polynomial<br />
<br />
<br />
<br />
p(λ) = det(A − λI2) = (1<br />
− λ) 2 <br />
<br />
2 (1 − λ) = (1 − λ)2 − 4.<br />
The roots <strong>of</strong> the characteristic polynomial are<br />
(λ − 1) 2 = 4 ⇔ λ± = 1 ± 2 ⇔ λ+ = 3, λ− = −1.<br />
The eigenvectors corresponding to the eigenvalue λ+ = 3 are the solutions v <strong>of</strong> the linear<br />
system (A − 3I2)v (+) = 0. To find them, we perform Gauss operations on the matrix<br />
<br />
−2 2 1 −1<br />
A − 3I2 =<br />
→<br />
⇒ v1 = v2 ⇒ v<br />
2 −2 0 0<br />
(+) <br />
1<br />
= .<br />
1<br />
The eigenvectors corresponding to the eigenvalue λ− = −1 are the solutions v <strong>of</strong> the linear<br />
system (A + I2)v (-) = 0. To find them, we perform Gauss operations on the matrix<br />
<br />
2 2 1 1<br />
A + I2 = → ⇒ v1 = −v2 ⇒ v<br />
2 2 0 0<br />
(-) <br />
−1<br />
= .<br />
1<br />
With this information and Theorem 5.5.7 we obtain that<br />
A = P DP −1 , P =<br />
1 −1<br />
1 1<br />
<br />
, D =<br />
<br />
3 0<br />
,<br />
0 −1
and also that<br />
so we conclude that<br />
e At = P e Dt P −1 =<br />
e At = 1<br />
2<br />
G. NAGY – ODE July 2, 2013 199<br />
<br />
1<br />
<br />
3t −1 e 0<br />
1 1 0 e−t <br />
1<br />
2<br />
<br />
3t −t 3t −t (e + e ) (e − e )<br />
(e3t − e−t ) (e3t + e−t <br />
.<br />
)<br />
The solution to the initial value problem above is,<br />
x(t) = e At x0 = 1<br />
<br />
3t −t 3t −t (e + e ) (e − e )<br />
2 (e3t − e−t ) (e3t + e−t <br />
3<br />
) 2<br />
<br />
1 1<br />
,<br />
−1 1<br />
⇒ x(t) = 1<br />
2<br />
5e 3t + e −t<br />
5e 3t − e −t<br />
We now present an equivalent way to write down the solution function x given in Theorem<br />
5.6.1, Eq. (5.6.3). This new expression for x looks simpler because it does not contain<br />
the exponential <strong>of</strong> the system coefficient matrix. Only the matrix eigenvalues and a simple<br />
linear combination <strong>of</strong> the matrix eigenvectors appear in this new expression.<br />
Theorem 5.6.2 (Eigenpairs expression). If the n×n constant matrix A is diagonalizable,<br />
with eigenpairs λi, v (i) for i = 1, · · · , n, then, the homogeneous initial value problem<br />
has a unique for every initial condition x0 ∈ R n given by<br />
<br />
.<br />
x ′ (t) = A x(t) x(t 0) = x 0. (5.6.4)<br />
x(t) = c1v (1) e λ1(t−t0) + · · · + cnv (n) e λn(t−t0) , (5.6.5)<br />
where the constants c 1, · · · , cn are solution <strong>of</strong> the algebraic linear system<br />
x0 = c1v (1) + · · · + cnv (n) . (5.6.6)<br />
Just by looking at Eqs. (5.6.5)-(5.6.6) one can understand the reason for the simplicity <strong>of</strong><br />
this new expression for x; it is simple because it is not complete. The coefficients c1, · · · , cn<br />
are said to be solutions <strong>of</strong> Eq. (5.6.6), but such solution is actually not computed. If one<br />
solves the <strong>equations</strong> for c 1, · · · , cn, the resulting expression for x is the one in Eq. (5.6.3).<br />
We think that Eq. (5.6.5) deserves to be highlighted in a Theorem, because is very common<br />
in the literature, and simple to work with.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.6.2: Recall Theorem 5.5.7, where we showed that matrix A is<br />
diagonalizable iff it can be written as<br />
A = P DP −1 , P = v (1) , · · · , v (n) , D = diag[λ1, · · · , λn],<br />
where λi, v (i) are eigenpairs <strong>of</strong> the matrix A. Let us use this information and the exponential<br />
formula e At = P e Dt P −1 to rewrite the solution found in Eq. (5.6.3), that is,<br />
x(t) = e At x0 = P e Dt P −1 x0.<br />
Introduce the constant vector c = P −1 x0 and recall that<br />
P e Dt = v (1) , · · · , v (n) diag e λ1t , · · · , e λnt = e λ1t v (1) , · · · , e λnt v (n) .<br />
We then arrive at the expression<br />
x(t) = e λ1t v (1) , · · · , e λnt v (n) c ⇔ x(t) = c 1e λ1t v (1) + · · · + cne λnt v (n) ,<br />
where the constant vector c satisfies the equation P c = x0, that is,<br />
c1v (1) + · · · + cnv (n) = x0.<br />
This establishes the Theorem. <br />
⊳
200 G. NAGY – ODE july 2, 2013<br />
Remark: An alternative pro<strong>of</strong> for Theorem 5.6.2 is simply to verify that x in Eq. (5.6.5) is<br />
solution <strong>of</strong> the initial value problem in Eq. (5.6.4). Indeed, x is solution <strong>of</strong> the <strong>differential</strong><br />
equation, since for all constants c1, · · · , cn holds<br />
Ax(t) = c 1Av (1) e λ1t + · · · + cnAv (n) e λnt<br />
= c 1λ 1v (1) e λ1t + · · · + cnλnv (n) e λnt<br />
= c1v (1) e λ1t ′ + · · · + cnv (n) e λnt ′<br />
= x ′ (t).<br />
The function x in Eq. (5.6.5) satisfies the initial condition because<br />
x0 = x(0) = c1v (1) + · · · + cnv (n) ,<br />
which is Eq. (5.6.6) for the constants c 1, · · · , cn.<br />
Example 5.6.2: Find the vector-valued function x solution to the <strong>differential</strong> system<br />
x ′ <br />
3 1 2<br />
= A x, x(0) = , A = .<br />
2 2 1<br />
Solution: In Example 5.6.1 we have found the eigenvalues and eigenvectors <strong>of</strong> the coefficient<br />
matrix, and the result is<br />
λ1 = 3, v (1) <br />
1<br />
= ,<br />
1<br />
and λ2 = −1, v (2) <br />
−1<br />
= .<br />
1<br />
So, Eq. (5.6.5) implies that<br />
x(t) = c 1<br />
<br />
1<br />
e<br />
1<br />
3t + c2 <br />
−1<br />
e<br />
1<br />
−t ,<br />
where the coefficients c1 and c2 are solutions <strong>of</strong> the initial condition equation<br />
<br />
<br />
c1<br />
c1<br />
c1<br />
1<br />
1<br />
+ c2<br />
−1<br />
1<br />
=<br />
3<br />
2<br />
⇒<br />
1 −1<br />
1 1<br />
c2<br />
=<br />
3<br />
2<br />
⇒<br />
c2<br />
= 1<br />
2<br />
1 1<br />
−1 1<br />
<br />
3<br />
.<br />
2<br />
We conclude that c1 = 5/2 and c2 = −1/2, so we reobtain the result in Example 5.6.1,<br />
<br />
<br />
3t −t 5e + e<br />
x(t) = 5<br />
2<br />
1<br />
1<br />
e 3t − 1<br />
2<br />
−1<br />
1<br />
<br />
e −t ⇔ x(t) = 1<br />
2<br />
5e 3t − e −t<br />
5.6.2. The Fundamental Solutions. The expression in Eq. (5.6.5) for the solution <strong>of</strong> a<br />
linear, homogeneous <strong>differential</strong> system says that there exist n different solutions to the<br />
linear system x ′ = Ax. Just choose one <strong>of</strong> the coefficients ci non-zero and all the rest zero.<br />
It is then useful to introduce the following definitions.<br />
Definition 5.6.3. The set <strong>of</strong> functions x (1) , · · · , x (n) is called a fundamental set <strong>of</strong><br />
solutions to the homogeneous system x ′ = Ax iff each function x (i) for i = 1, · · · , n is solution<br />
<strong>of</strong> the system above and the set is linealry independent for all t ∈ R. Each function x (i)<br />
in a fundamental set is called a fundamental solution to the system above. Furthermore,<br />
the matrix-valued function X = x (1) , · · · , x (n) is called the fundamental matrix <strong>of</strong> the<br />
system above. Finally, the general solution <strong>of</strong> the <strong>differential</strong> equation above is the set <strong>of</strong><br />
all functions <strong>of</strong> the form<br />
⎡ ⎤<br />
x(t) = c1x (1) (t) + · · · + cnx (n) (t) ⇔ x(t) = X(t) c, c =<br />
.<br />
⎢<br />
⎣<br />
c1<br />
.<br />
cn<br />
⎥<br />
⎦ .<br />
⊳
G. NAGY – ODE July 2, 2013 201<br />
Any linearly independent set <strong>of</strong> n solutions to an n × n linear homogeneous <strong>differential</strong><br />
system defines a fundamental set. Hence, fundamental sets <strong>of</strong> solutions are not unique at all.<br />
If the system coefficient matrix is diagonalizable, then Theorem 5.6.5 provides an explicit<br />
expression for a fundamental set, a fundamental matrix and a general solution. They are<br />
constructed with the eigenvalues and eigenvectors <strong>of</strong> the coefficient matrix.<br />
Theorem 5.6.4. If the n × n constant matrix A is diagonalizable, with eigenpairs λi, v (i)<br />
for i = 1, · · · , n, then, the linear homogeneous system x ′ = Ax has a fundamental set <strong>of</strong><br />
solutions given by<br />
(1) (1) λ1t (n) (n) λnt<br />
x (t) = v e , · · · , x (t) = v e <br />
(5.6.7)<br />
Then, the associated fundamental matrix is<br />
X(t) = v (1) e λ1t , · · · , v (n) e λnt ,<br />
and the general solution is given by every linear combination<br />
x(t) = c1v (1) e λ1t + · · · + cnv (n) e λnt .<br />
Remark: It is simple to see the relation between the fundamental solution matrix X <strong>of</strong> an<br />
homogeneous system x ′ = Ax and the exponential matrix eAt . Indeed, if A = P DP −1 , then<br />
X(t) = P eDt ; equivalently, eAt = X(t)P −1 . Also notice that X(0) = P .<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.6.4: We have already seen in the Remark after Theorem 5.6.2 that<br />
every function in the set given in (5.6.7) is solution to the system x ′ = Ax. We only need<br />
to show that this set is linearly independent. For that reason we compute the determinant<br />
<strong>of</strong> the matrix X. Since det P −1 = 0, we can do the following,<br />
det X(t) = det X(t) detP −1<br />
det P −1 = detX(t)P −1<br />
det P −1<br />
.<br />
Recalling that X(t)P −1 = P e Dt P −1 we obtain that<br />
det X(t) = det P e Dt P −1<br />
det P −1<br />
= det P det e Dt det P −1<br />
det P −1<br />
= det e Dt<br />
det P −1.<br />
Therefore, det X(t) = 0 iff det e Dt = 0. But this latter equatio is true, since<br />
det e Dt = e λ1t · · · e λnt ⇔ det e Dt = e (λ1+···λn)t = 0.<br />
Therefore det X(t) = 0 for all t ∈ R. This establishes the Theorem. <br />
Remark: The function W (t) = det X(t) is called the Wronskian <strong>of</strong> a set <strong>of</strong> n solutions<br />
to a <strong>differential</strong> system.<br />
Example 5.6.3: Find the general solution to the 2 × 2 <strong>differential</strong> system<br />
x ′ <br />
1 3<br />
= Ax, A = .<br />
3 1<br />
Solution: Following Theorem 5.6.5, in order to find the solution to the equation above we<br />
must start finding the eigenvalues and eigenvectors <strong>of</strong> the coefficient matrix A. This part<br />
<strong>of</strong> the work was already done in Example 5.5.4. We have found that A has two linearly<br />
independent eigenvectors, more precisely,<br />
λ1 = 4, v (1) <br />
1<br />
=<br />
⇒ x (1) <br />
1<br />
= e 4t ,<br />
λ 2 = −2, v (2) =<br />
1<br />
−1<br />
1<br />
<br />
1<br />
⇒ x (2) <br />
−1<br />
= e<br />
1<br />
−2t .
202 G. NAGY – ODE july 2, 2013<br />
Therefore, the general solution <strong>of</strong> the <strong>differential</strong> equation is<br />
<br />
1<br />
x(t) = c1 e<br />
1<br />
4t <br />
−1<br />
+ c2 e<br />
1<br />
−2t , c1, c2 ∈ R.<br />
Remark: We can write the general solution using the fundamental matrix X(t) and the<br />
constant vector c,<br />
<br />
4t e<br />
X(t) =<br />
<br />
−2t −e<br />
,<br />
<br />
c1<br />
c = ,<br />
e 4t e −2t<br />
and the general solution is x(t) = X(t)c, that is,<br />
<br />
x1(t)<br />
=<br />
x2(t) e 4t −e −2t<br />
e 4t e −2t<br />
c2<br />
<br />
c1<br />
.<br />
c2 Example 5.6.4: Use the fundamental solution matrix found in Example 5.6.3 to find the<br />
solution to the initial value problem<br />
x ′ <br />
2<br />
1 3<br />
= Ax, x(0) = , A = .<br />
4<br />
3 1<br />
Solution: We known from Example 5.6.3 that the general solution to the <strong>differential</strong><br />
equation above is given by<br />
<br />
1<br />
x(t) = c1 e<br />
1<br />
4t <br />
−1<br />
+ c2 e<br />
1<br />
−2t , c1, c2 ∈ R.<br />
Equivalently, introducing the fundamental matrix X and the vector c as in Example 5.6.3,<br />
<br />
4t e<br />
X(t) =<br />
e<br />
−2t −e 4t e−2t <br />
,<br />
<br />
c1<br />
c =<br />
c2 ⇒ x(t) = X(t)c.<br />
The initial condition is an equation for the constant vector c,<br />
X(0)c = x(0) ⇔<br />
<br />
1<br />
1<br />
<br />
−1 c1<br />
=<br />
1 c2 <br />
2<br />
.<br />
4<br />
The solution <strong>of</strong> the linear system above can be expressed in terms <strong>of</strong> the inverse matrix<br />
<br />
−1 1 1 1<br />
X(0) = ,<br />
2 −1 1<br />
as follows,<br />
c = X(0) −1 x(0) ⇔<br />
c1<br />
c 2<br />
<br />
= 1<br />
2<br />
1 1<br />
−1 1<br />
<br />
2<br />
4<br />
So, the solution to the initial value problem is given by<br />
<br />
1<br />
x(t) = 3 e<br />
1<br />
4t <br />
−1<br />
+ e<br />
1<br />
−2t .<br />
⇒<br />
c1<br />
<br />
=<br />
c2 <br />
3<br />
.<br />
1<br />
Remark: In the case that the coefficient matrix A <strong>of</strong> a linear homogeneous <strong>differential</strong><br />
system has n distinct eigenvalues, then Theorem 5.5.8 implies that A is diagonalizable, and<br />
every the result in this Section applies to this case.<br />
⊳<br />
⊳
G. NAGY – ODE July 2, 2013 203<br />
5.6.3. Diagonalizable systems with continuous sources. We continue our study <strong>of</strong><br />
diagonalizable linear <strong>differential</strong> systems with the case <strong>of</strong> non-homogeneous, continuous<br />
sources. The solution <strong>of</strong> an initial value problem in this case involves again the exponential<br />
<strong>of</strong> the coefficient matrix. The solution for these systems is a generalization <strong>of</strong> the solution<br />
formula for single scalar <strong>equations</strong>, which is given by Eq. (1.2.7), in the case that the<br />
coefficient function a is constant.<br />
Theorem 5.6.5 (Exponential expression). If the n × n constant matrix A is diagonalizable<br />
and the n-vector-valued function g is constant, then the initial value problem<br />
x ′ (t) = A x(t) + g, x(t0) = x0.<br />
has a unique solution for every initial condition x0 ∈ R n given by<br />
<br />
A(t−t0)<br />
x(t) = e x0 +<br />
t<br />
t0<br />
e −A(τ−t0) g(τ) dτ<br />
<br />
. (5.6.8)<br />
Remark: In the case <strong>of</strong> an homogeneous system, g = 0, and we reobtain Eq. (5.6.3). In<br />
the case that the coefficient matrix A is invertible and the source function g is constant, the<br />
integral in Eq. (5.6.8) can be computed explicitly and the result is<br />
x(t) = e A(t−t0) x0 − A −1 g − A −1 g.<br />
The expression above is the generalizations for systems <strong>of</strong> Eq. (1.1.10) for scalar <strong>equations</strong>.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.6.5: Since the coefficient matrix A is diagonalizable, there exist an<br />
n×n invertible matrix P and an n×n diagonal matrix D such that A = P DP −1 . Introducing<br />
this information into the <strong>differential</strong> equation in Eq. (1.1.10), and then multiplying the whole<br />
equation by P −1 , we obtain,<br />
P −1 x ′ (t) = P −1 P DP −1 x(t) + P −1 g(t).<br />
Since matrix A is constant, so is P and D. In particular P −1 x ′ = P −1 x ′ . Therefore,<br />
(P −1 x) ′ = D P −1 x + P −1 g .<br />
Introduce the new unknown function y = P −1 x and the new source function h = P −1 g ,<br />
then the equation above has the form<br />
y ′ (t) = D y(t) + h(t).<br />
Now transform the initial condition too, that is, P −1 x(t 0) = P −1 x 0. Introducing the notation<br />
y 0 = P −1 x 0, we get the initial condition<br />
y(t0) = y 0.<br />
Since the coefficient matrix D is diagonal, the initial value problem above for the unknown<br />
y is decoupled. That is, expressing D = diag[λ1, · · · , λn], h = [hi], y 0 = [y0i], and y = [yi],<br />
then for i = 1, · · · , n holds that the initial value problem above has the form<br />
y ′ i(t) = λi yi(t) + hi(t), yi(t0) = y0i.<br />
The solution <strong>of</strong> each initial value problem above can be found using Eq. (1.2.7), that is,<br />
<br />
λi(t−t0)<br />
yi(t) = e y0i +<br />
t<br />
t0<br />
t0<br />
e −λi(τ−t0) hi(τ) dτ<br />
We rewrite the <strong>equations</strong> above in vector notation as follows,<br />
t<br />
D(t−t0)<br />
y(t) = e y0 + e −D(τ−t0)<br />
<br />
h(τ) dτ ,<br />
<br />
.
204 G. NAGY – ODE july 2, 2013<br />
where we recall that eD(t−t0) <br />
λ1(t−t0) λn(t−t0) = diag e , · · · , e . We now multiply the whole<br />
equation by the constant matrix P and we recall that P −1P = In, then<br />
e −D(τ−t0)<br />
<br />
−1<br />
(P P )h(τ) dτ .<br />
<br />
D(t−t0)<br />
P y(t) = P e (P −1 P )y0 + (P −1 t<br />
P )<br />
t0<br />
Recalling that P y = x, P y0 = x0, P h = g, and eAt = P eDtP −1 , we obtain<br />
t<br />
<br />
A(t−t0)<br />
x(t) = e x0 +<br />
t0<br />
e −A(τ−t0) g(τ) dτ<br />
This establishes the Theorem. <br />
We said it in the homogenoeus equation and we now say it again. Although the expression<br />
for the solution given in Eq. (5.6.8) looks simple, the exponentials <strong>of</strong> the coefficient matrix<br />
are actually not that simple to compute. We show this in the following example.<br />
Example 5.6.5: Find the vector-valued solution x to the <strong>differential</strong> system<br />
x ′ <br />
3 1 2 1<br />
= A x + g, x(0) = , A = , g = .<br />
2 2 1 2<br />
Solution: In Example 5.6.1 we have found the eigenvalues and eigenvectors <strong>of</strong> the coefficient<br />
matrix, and the result is<br />
λ1 = 3, v (1) <br />
1<br />
= ,<br />
1<br />
and λ2 = −1, v (2) <br />
−1<br />
= .<br />
1<br />
With this information and Theorem 5.5.7 we obtain that<br />
and also that<br />
so we conclude that<br />
e At = 1<br />
2<br />
A = P DP −1 , P =<br />
e At = P e Dt P −1 =<br />
(e 3t + e −t ) (e 3t − e −t )<br />
(e 3t − e −t ) (e 3t + e −t )<br />
1 −1<br />
1 1<br />
<br />
, D =<br />
<br />
1<br />
<br />
3t −1 e 0<br />
1 1 0 e−t <br />
1<br />
2<br />
The solution to the initial value problem above is,<br />
Since<br />
e At x0 = 1<br />
2<br />
in a similar way<br />
e −Aτ g = 1<br />
2<br />
<br />
⇒ e −At = 1<br />
2<br />
x(t) = e At x0 + e At<br />
t<br />
(e 3t + e −t ) (e 3t − e −t )<br />
(e 3t − e −t ) (e 3t + e −t )<br />
(e −3τ + e τ ) (e −3τ − e τ )<br />
(e −3τ − e τ ) (e −3τ + e τ )<br />
0<br />
e −Aτ g dτ.<br />
<br />
3<br />
2<br />
<br />
1<br />
2<br />
Integrating the last expresion above, we get<br />
t<br />
e −Aτ g dτ = 1<br />
<br />
−3t t −e − e<br />
2 −e−3t + et <br />
+<br />
Therefore, we get<br />
x(t) = 1<br />
2<br />
5e 3t + e −t<br />
5e 3t − e −t<br />
0<br />
<br />
+ 1<br />
2<br />
<br />
.<br />
<br />
3 0<br />
,<br />
0 −1<br />
<br />
1 1<br />
,<br />
−1 1<br />
<br />
−3t t −3t t (e + e ) (e − e )<br />
(e−3t − et ) (e−3t + et <br />
.<br />
)<br />
= 1<br />
2<br />
= 1<br />
2<br />
5e 3t + e −t<br />
5e 3t − e −t<br />
3e −3τ − e τ<br />
<br />
1<br />
.<br />
0<br />
<br />
3t −t (e + e ) 3t −t (e − e )<br />
(e3t − e−t ) (e3t + e−t 1<br />
) 2<br />
3e −3τ + e τ<br />
<br />
,<br />
<br />
.<br />
<br />
−3t t −e − e<br />
+<br />
−e −3t + e t<br />
<br />
1<br />
<br />
.<br />
0
G. NAGY – ODE July 2, 2013 205<br />
Multiplying the matrix-vector product on the second term <strong>of</strong> the left-hand side above,<br />
x(t) = 1<br />
<br />
3t −t 5e + e<br />
2 5e3t − e−t <br />
1<br />
− +<br />
0<br />
1<br />
<br />
3t −t (e + e )<br />
2 (e3t − e−t <br />
.<br />
)<br />
We conclude that the solution to the initial value problem above is<br />
x(t) =<br />
3e 3t + e −t − 1<br />
3e 3t − e −t<br />
<br />
.<br />
⊳
206 G. NAGY – ODE july 2, 2013<br />
5.6.4. Exercises.<br />
5.6.1.- . 5.6.2.- .
G. NAGY – ODE July 2, 2013 207<br />
5.7. 2 × 2 Real distinct eigenvalues<br />
This and the next two Sections are dedicated to study 2×2 linear, homogeneous <strong>differential</strong><br />
systems. In this Section we consider the case that the coefficient matrix is diagonalizable<br />
and has two different, real, eigenvalues. We study in some depth the solutions <strong>of</strong> this particular<br />
type <strong>of</strong> systems<br />
5.7.1. Phase diagrams. The main tool will be the sketch <strong>of</strong> phase diagrams, also called<br />
phase portraits. The solution value at a particular t is given by a 2-vector<br />
<br />
x1(t)<br />
x(t) = ,<br />
x2(t)<br />
so it can be represented by a point on a plane, while the solution function for all t ∈ R<br />
corresponds to a curve on that plane. In the case that the solution vector x(t) is interpreted<br />
as the position function <strong>of</strong> a particle moving in a plane at the time t, the curve given in<br />
the phase diagram is the trajectory <strong>of</strong> the particle. Arrows are added to this trajectory to<br />
indicate the motion <strong>of</strong> the particle as time increases.<br />
In this Section we concentrate in constructing phase diagrams <strong>of</strong> solutions <strong>of</strong> 2×2 systems<br />
x ′ (t) = Ax(t),<br />
where matrix A has two real, distinct, eigenvalues. Theorem 5.5.8 implies that A is diagonalizable,<br />
and we can use the results in Section 5.6 to describe the general solution <strong>of</strong> the<br />
system, given by Eq. (5.6.5), that is,<br />
x(t) = c1 v (1) e λ1t + c2 v (2) e λ2t . (5.7.1)<br />
where λ1, v (1) and λ2, v (2) are eigenpairs <strong>of</strong> matrix A, with the eigenvectors forming a<br />
linearly independent set.<br />
In the case that one <strong>of</strong> the eigenvalues vanishes the study is simple and is left as an<br />
exercise. In the case that the eigenvalues are non-zero, all possible phase diagrams can<br />
be classified into three main classes, defined by the relative signs <strong>of</strong> the coefficient matrix<br />
eigenvalues λ1 = λ2, as follows:<br />
(i) 0 < λ 2 < λ 1, that is, both eigenvalues positive;<br />
(ii) λ2 < 0 < λ1, that is, one eigenvalue negative and the other positive;<br />
(iii) λ2 < λ1 < 0, that is, both eigenvalues negative.<br />
A phase diagram contains several curves, each one corresponding to a solution given in<br />
Eq. (5.7.1) for a particular choice <strong>of</strong> constants c 1 and c 2. A phase diagram can be sketched<br />
following these simple steps:<br />
(a) Plot the eigenvectors v (2) and v (1) corresponding to the eigenvalues λ 2 and λ 1.<br />
(b) Draw the whole lines parallel to these vectors and passing through the origin. These<br />
straight lines correspond to solutions with one <strong>of</strong> the coefficients c1 or c2 vanishing.<br />
(c) Draw arrows on these lines indicate how the solution changes as the variable t grows.<br />
If t is interpreted as time, the arrows indicate how the solution changes into the future.<br />
The arrows point towards the origin if the corresponding eigenvalue λ is negative, and<br />
they point away form the origin if the eigenvalue is positive.<br />
(d) Find the non-straight curves correspond to solutions with both coefficient c 1 and c 2<br />
non-zero. Again, arrows on these curves indicate the how the solution moves into the<br />
future.<br />
We now find the phase diagrams for three examples, one for every eigenvalue class above.
208 G. NAGY – ODE july 2, 2013<br />
5.7.2. Case 0 < λ 2 < λ 1.<br />
Example 5.7.1: Sketch the phase diagram <strong>of</strong> the solutions to the <strong>differential</strong> equation<br />
x ′ = Ax, A = 1<br />
<br />
11<br />
4 1<br />
<br />
3<br />
.<br />
9<br />
(5.7.2)<br />
Solution: The characteristic equation for this matrix A is given by<br />
det(A − λI) = λ 2 − 5λ + 6 = 0 ⇒<br />
<br />
λ1 = 3,<br />
λ2 = 2.<br />
One can show that the corresponding eigenvectors are given by<br />
v (1) <br />
3<br />
= , v<br />
1<br />
(2) <br />
−2<br />
= .<br />
2<br />
So the general solution to the <strong>differential</strong> equation above is given by<br />
x(t) = c1 v (1) e λ1t<br />
+ c2 v (2) e λ2t<br />
⇔<br />
<br />
3<br />
x(t) = c1<br />
1<br />
e 3t + c 2<br />
<br />
−2<br />
e<br />
2<br />
2t .<br />
In Fig. 27 we have sketched four curves, each representing a solution x corresponding to a<br />
particular choice <strong>of</strong> the constants c1 and c2. These curves actually represent eight different<br />
solutions, for eight different choices <strong>of</strong> the constants c 1 and c 2, as is described below. The<br />
arrows on these curves represent the change in the solution as the variable t grows. Since<br />
both eigenvalues are positive, the length <strong>of</strong> the solution vector always increases as t grows.<br />
The straight lines correspond to the following four solutions:<br />
c1 = 1, c2 = 0, Line on the first quadrant, starting at the origin, parallel to v (1) ;<br />
c 1 = 0, c 2 = 1, Line on the second quadrant, starting at the origin, parallel to v (2) ;<br />
c1 = −1, c2 = 0, Line on the third quadrant, starting at the origin, parallel to −v (1) ;<br />
c 1 = 0, c 2 = −1, Line on the fourth quadrant, starting at the origin, parallel to −v (2) .<br />
2<br />
v<br />
x 2<br />
Figure 27. The graph <strong>of</strong> several solutions to Eq. (5.7.2) corresponding to<br />
the case 0 < λ2 < λ1, for different values <strong>of</strong> the constants c1 and c2. The<br />
trivial solution x = 0 is called an unstable point.<br />
v<br />
1<br />
x 1
G. NAGY – ODE July 2, 2013 209<br />
Finally, the curved lines on each quadrant start at the origin, and they correspond to the<br />
following choices <strong>of</strong> the constants:<br />
c 1 > 0, c 2 > 0, Line starting on the second to the first quadrant;<br />
c1 < 0, c2 > 0, Line starting on the second to the third quadrant;<br />
c1 < 0, c2 < 0, Line starting on the fourth to the third quadrant,<br />
c 1 > 0, c 2 < 0, Line starting on the fourth to the first quadrant.<br />
5.7.3. Case λ 2 < 0 < λ 1.<br />
Example 5.7.2: Sketch the phase diagram <strong>of</strong> the solutions to the <strong>differential</strong> equation<br />
x ′ = Ax,<br />
<br />
1<br />
A =<br />
3<br />
<br />
3<br />
.<br />
1<br />
(5.7.3)<br />
Solution: In Example 5.6.3 we computed the eigenvalues and eigenvectors <strong>of</strong> the coefficient<br />
matrix, and the result was<br />
λ1 = 4, v (1) <br />
1<br />
= and λ2 = −2, v<br />
1<br />
(2) <br />
−1<br />
= .<br />
1<br />
In that Example we also computed the general solution to the <strong>differential</strong> equation above,<br />
x(t) = c1 v (1) e λ1t<br />
+ c2 v (2) e λ2t<br />
⇔<br />
<br />
1<br />
x(t) = c1 e<br />
1<br />
4t <br />
−1<br />
+ c2 e<br />
1<br />
−2t ,<br />
In Fig. 28 we have sketched four curves, each representing a solution x corresponding to a<br />
particular choice <strong>of</strong> the constants c 1 and c 2. These curves actually represent eight different<br />
solutions, for eight different choices <strong>of</strong> the constants c 1 and c 2, as is described below. The<br />
arrows on these curves represent the change in the solution as the variable t grows. The<br />
part <strong>of</strong> the solution with positive eigenvalue increases exponentially when t grows, while the<br />
part <strong>of</strong> the solution with negative eigenvalue decreases exponentially when t grows. The<br />
straight lines correspond to the following four solutions:<br />
c1 = 1, c2 = 0, Line on the first quadrant, starting at the origin, parallel to v (1) ;<br />
c 1 = 0, c 2 = 1, Line on the second quadrant, ending at the origin, parallel to v (2) ;<br />
c1 = −1, c2 = 0, Line on the third quadrant, starting at the origin, parallel to −v (1) ;<br />
c 1 = 0, c 2 = −1, Line on the fourth quadrant, ending at the origin, parallel to −v (2) .<br />
Finally, the curved lines on each quadrant correspond to the following choices <strong>of</strong> the<br />
constants:<br />
c 1 > 0, c 2 > 0, Line from the second to the first quadrant,<br />
c1 < 0, c2 > 0, Line from the second to the third quadrant,<br />
c1 < 0, c2 < 0, Line from the fourth to the third quadrant,<br />
c 1 > 0, c 2 < 0, Line from the fourth to the first quadrant.<br />
⊳<br />
⊳
210 G. NAGY – ODE july 2, 2013<br />
v<br />
1<br />
2 v<br />
−1<br />
x<br />
2<br />
−1<br />
Figure 28. The graph <strong>of</strong> several solutions to Eq. (5.7.3) corresponding to<br />
the case λ2 < 0 < λ1, for different values <strong>of</strong> the constants c1 and c2. The<br />
trivial solution x = 0 is called a saddle point.<br />
5.7.4. Case λ2 < λ1 < 0.<br />
Example 5.7.3: Sketch the phase diagram <strong>of</strong> the solutions to the <strong>differential</strong> equation<br />
x ′ = Ax, A = 1<br />
<br />
−9 3<br />
. (5.7.4)<br />
4 1 −11<br />
Solution: The characteristic equation for this matrix A is given by<br />
det(A − λI) = λ 2 + 5λ + 6 = 0 ⇒<br />
<br />
λ1 = −2,<br />
λ2 = −3.<br />
One can show that the corresponding eigenvectors are given by<br />
v (1) <br />
3<br />
= , v<br />
1<br />
(2) <br />
−2<br />
= .<br />
2<br />
So the general solution to the <strong>differential</strong> equation above is given by<br />
x(t) = c1 v (1) e λ1t<br />
+ c2 v (2) e λ2t<br />
<br />
3<br />
⇔ x(t) = c1 e<br />
1<br />
−2t + c2<br />
1<br />
1<br />
x<br />
1<br />
<br />
−2<br />
e<br />
2<br />
−3t .<br />
In Fig. 29 we have sketched four curves, each representing a solution x corresponding to a<br />
particular choice <strong>of</strong> the constants c1 and c2. These curves actually represent eight different<br />
solutions, for eight different choices <strong>of</strong> the constants c 1 and c 2, as is described below.<br />
The arrows on these curves represent the change in the solution as the variable t grows.<br />
Since both eigenvalues are negative, the length <strong>of</strong> the solution vector always decreases as t<br />
grows and the solution vector always approaches zero. The straight lines correspond to the<br />
following four solutions:<br />
c1 = 1, c2 = 0, Line on the first quadrant, ending at the origin, parallel to v (1) ;<br />
c 1 = 0, c 2 = 1, Line on the second quadrant, ending at the origin, parallel to v (2) ;<br />
c1 = −1, c2 = 0, Line on the third quadrant, ending at the origin, parallel to −v (1) ;<br />
c 1 = 0, c 2 = −1, Line on the fourth quadrant, ending at the origin, parallel to −v (2) .
G. NAGY – ODE July 2, 2013 211<br />
v<br />
2<br />
x 2<br />
Figure 29. The graph <strong>of</strong> several solutions to Eq. (5.7.4) corresponding to<br />
the case λ2 < λ1 < 0, for different values <strong>of</strong> the constants c1 and c2. The<br />
trivial solution x = 0 is called a stable point.<br />
Finally, the curved lines on each quadrant start at the origin, and they correspond to the<br />
following choices <strong>of</strong> the constants:<br />
c 1 > 0, c 2 > 0, Line entering the first from the second quadrant;<br />
c1 < 0, c2 > 0, Line entering the third from the second quadrant;<br />
c1 < 0, c2 < 0, Line entering the third from the fourth quadrant,<br />
c 1 > 0, c 2 < 0, Line entering the first from the fourth quadrant.<br />
v 1<br />
x 1<br />
⊳
212 G. NAGY – ODE july 2, 2013<br />
5.7.5. Exercises.<br />
5.7.1.- . 5.7.2.- .
G. NAGY – ODE July 2, 2013 213<br />
5.8. 2 × 2 Complex eigenvalues<br />
5.8.1. Complex conjugate pairs. In this Section we study the case that the coefficient<br />
matrix is real-valued, diagonalizable, and has two complex-valued eigenvalues. Since the<br />
coefficient matrix is real-valued, the complex-valued eigenvalues are always a complex conjugate<br />
pair. The followng result holds for n × n matrices, not just 2 × 2.<br />
Theorem 5.8.1 (Conjugate pairs). If an n × n real-valued matrix A has a complex<br />
eigenvalue eigenvector pair λ, v, then the complex conjugate pair λ, v is an eigenvalue<br />
eigenvector pair <strong>of</strong> matrix A.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.8.1: Complex conjugate the eigenvalue eigenvector equation for λ<br />
and v, and recall that matrix A is real-valued, meaning that A = A, then we obtain,<br />
Av = λv ⇔ Av = λ v,<br />
This establishes the Theorem. <br />
Since the complex eigenvalues <strong>of</strong> a matrix with real coefficients are always complex conjugate<br />
pairs, there are an even number <strong>of</strong> complex eigenvalues and it holds that λ+ = λ−,<br />
with a similar equation for their respective eigenvectors, that is, v (+) = v (−) . Therefore, it<br />
is not difficult to see that an eigenvalue and eigenvector pair has the form<br />
λ± = α ± iβ, v (±) = a ± ib, (5.8.1)<br />
where α, β ∈ R and a, b ∈ R n .<br />
It is simple to obtain a linearly independent set <strong>of</strong> two solutions to x ′ = Ax in the case<br />
that matrix A has a complex conjugate pair <strong>of</strong> eigenvalues and eigenvectors. These solutions<br />
are complex valued, but a simple linear combinantio <strong>of</strong> them is real-valued, as the following<br />
result shows.<br />
Theorem 5.8.2 (Complex and real solutions). If λ± = α ± iβ are two eigenvalues <strong>of</strong><br />
an n × n constant matrix A with respective eigenvectors v (±) = a ± ib, where α, β ∈ R and<br />
a, b ∈ R n , and n 2, then the complex-valued functions<br />
x (+) (t) = v (+) e λ+t , x (−) (t) = v (−) e λ-t , (5.8.2)<br />
are solutions to the system x ′ = Ax and form a linearly independent set. Furthermore, the<br />
real-valued functions<br />
x (1) (t) = a cos(βt) − b sin(βt) e αt , x (2) (t) = a sin(βt) + b cos(βt) e αt . (5.8.3)<br />
are also solutions to the system above and form a linearly independent set.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.8.2: We know from Theorem 5.6.2 that the two solutions in (5.8.2)<br />
form a linearly independent set. The new information in Theorem 5.8.2 above is the realvalued<br />
solutions in Eq. (5.8.3). They can be obtained from Eq. (5.8.2) as follows:<br />
x (±) = (a ± ib) e (α±iβ)t<br />
= e αt (a ± ib) e ±iβt<br />
= e αt (a ± ib) cos(βt) ± i sin(βt) <br />
= e αt a cos(βt) − b sin(βt) ± ie αt a sin(βt) + b cos(βt) .<br />
Since the <strong>differential</strong> equation x ′ = Ax is linear, the functions below are also solutions,<br />
x (1) = 1<br />
(+) (−)<br />
x + x<br />
2<br />
= e αt a cos(βt) − b sin(βt) ,<br />
x (2) = 1 (+) (−)<br />
x − x<br />
2i<br />
= e αt a sin(βt) + b cos(βt) .<br />
This establishes the Theorem.
214 G. NAGY – ODE july 2, 2013<br />
We now go back to 2 × 2 coefficient matrices.<br />
Example 5.8.1: Find a real-valued set <strong>of</strong> fundamental solutions to the <strong>differential</strong> equation<br />
x ′ <br />
2 3<br />
= Ax, A = , (5.8.4)<br />
−3 2<br />
and then sketch a phase diagram for the solutions <strong>of</strong> this equation.<br />
Solution: Fist find the eigenvalues <strong>of</strong> matrix A above,<br />
<br />
<br />
<br />
0 = (2 − λ) 3 <br />
<br />
−3 (2 − λ) = (λ − 2)2 + 9 ⇒ λ± = 2 ± 3i.<br />
Then find the respective eigenvectors. The one corresponding to λ+ is the solution <strong>of</strong> the<br />
homogeneous linear system with coefficients given by<br />
<br />
2 − (2 + 3i) 3 −3i 3 −i<br />
=<br />
→<br />
−3 2 − (2 + 3i) −3 −3i −1<br />
Therefore the eigenvector v<br />
<br />
1 1<br />
→<br />
−i −1<br />
<br />
i 1<br />
→<br />
−i 0<br />
<br />
i<br />
.<br />
0<br />
(+) <br />
v1<br />
= is given by<br />
v1 = −iv2 ⇒ v2 = 1, v1 = −i, ⇒ v (+) =<br />
v2<br />
<br />
−i<br />
, λ+ = 2 + 3i.<br />
1<br />
The second eigenvector is the complex conjugate <strong>of</strong> the eigenvector found above, that is,<br />
v (−) <br />
i<br />
= , λ− = 2 − 3i.<br />
1<br />
Notice that<br />
v (±) =<br />
<br />
0<br />
±<br />
1<br />
<br />
−1<br />
i.<br />
0<br />
Then, the real and imaginary parts <strong>of</strong> the eigenvalues and <strong>of</strong> the eigenvectors are given by<br />
<br />
0<br />
−1<br />
α = 2, β = 3, a = , b = .<br />
1<br />
0<br />
So a real-valued expression for a fundamental set <strong>of</strong> solutions is given by<br />
x (1) <br />
=<br />
<br />
0<br />
−1<br />
<br />
cos(3t) − sin(3t) e<br />
1<br />
0<br />
2t ⇒ x (tione) <br />
sin(3t)<br />
= e<br />
cos(3t)<br />
2t ,<br />
x (2) <br />
=<br />
<br />
0<br />
−1<br />
<br />
sin(3t) + cos(3t) e<br />
1<br />
0<br />
2t ⇒ x (2) <br />
− cos(3t)<br />
=<br />
e<br />
sin(3t)<br />
2t .<br />
The phase diagram <strong>of</strong> these two fundamental solutions is given in Fig. 30 below. There<br />
is also a circle given in that diagram, corresponding to the trajectory <strong>of</strong> the vectors<br />
˜x (1) <br />
sin(3t)<br />
=<br />
˜x<br />
cos(3t)<br />
(2) <br />
− cos(3t)<br />
=<br />
.<br />
sin(3t)<br />
The trajectory <strong>of</strong> these vectors is a circle since their length is constant equal to one. ⊳<br />
Remark: In the particular case that the coefficient matrix A in x ′ = Ax is 2 ×2, then there<br />
are no other solutions linearly independent to the solutions given in Eq. (5.8.3). That is,<br />
the general solution <strong>of</strong> given by<br />
where<br />
x(t) = c 1 x (1) (t) + c 2 x (2) (t),<br />
x (1) = a cos(βt) − b sin(βt) e αt , x (2) = a sin(βt) + b cos(βt) e αt .
G. NAGY – ODE July 2, 2013 215<br />
x (1)<br />
b<br />
x 2<br />
a<br />
Figure 30. The graph <strong>of</strong> the fundamental solutions x (1) and x (2) <strong>of</strong> the Eq. (5.8.4).<br />
5.8.2. Qualitative analysis. We end this Section with a qualitative study <strong>of</strong> the phase<br />
diagrams <strong>of</strong> the solutions for this case. We first fix the vectors a and b, and the plot phase<br />
diagrams for solutions having α > 0, α = 0, and α < 0. These diagrams are given in Fig. 31.<br />
We see that for α > 0 the solutions spiral outward as t increases, and for α < 0 the solutions<br />
spiral inwards to the origin as t increases.<br />
b<br />
x 2<br />
(2)<br />
x<br />
x (1)<br />
a<br />
x 1<br />
b<br />
(2)<br />
x<br />
x 2<br />
Figure 31. The graph <strong>of</strong> the fundamental solutions x (1) and x (2) (dashed<br />
line) <strong>of</strong> the Eq. (5.8.3) in the case <strong>of</strong> α > 0, α = 0, and α < 0, respectively.<br />
Finally, let us study the following cases: We fix α > 0 and the vector a, and we plot the<br />
phase diagrams for solutions with two choices <strong>of</strong> the vector b, as shown in Fig. 32. It can<br />
then bee seen that the relative directions <strong>of</strong> the vectors a and b determines the rotation<br />
direction <strong>of</strong> the solutions as t increases.<br />
b<br />
x 2<br />
x<br />
x (1)<br />
(2)<br />
a<br />
x 1<br />
Figure 32. The graph <strong>of</strong> the fundamental solutions x (1) and x (2) <strong>of</strong> the<br />
Eq. (5.8.3) for α > 0, a given vector a, and for two choices <strong>of</strong> the vector b.<br />
The relative positions <strong>of</strong> a and b determines the rotation direction.<br />
x (1)<br />
(1)<br />
x<br />
x<br />
a<br />
(2)<br />
x 1<br />
x1<br />
x 2<br />
x (1)<br />
b<br />
(2)<br />
x<br />
b<br />
x 2<br />
x<br />
(2)<br />
a<br />
x 1<br />
a<br />
x 1
216 G. NAGY – ODE july 2, 2013<br />
5.8.3. Exercises.<br />
5.8.1.- . 5.8.2.- .
G. NAGY – ODE July 2, 2013 217<br />
5.9. 2 × 2 Repeated eigenvalues<br />
In Sections 5.7 and 5.8 we considered <strong>differential</strong> systems with coefficient matrix having<br />
distinct eigenvalues. Theorem 5.5.8 implies that these matrices are diagonalizable. Hence,<br />
the results in Section 5.6 can be applied to find the solutions to linear homogeneous systems<br />
with such coefficient matrices. In this Section we study the remaining case, when the<br />
coefficient matrix has repeated eigenvalues. We concentrate in the particular case where the<br />
coefficient matrix has an eigenvalue with algebraic multiplicity r = 2, that is, the simplest<br />
case. We have seen examples <strong>of</strong> matrices with r = 2 that are diagonalizable, Example 5.5.7,<br />
and matrices that are not diagonalizable, Example 5.5.8. Both cases are possible, but only<br />
the latter is new for us. Therefore we present a brief section where we summarize the main<br />
result for diagonalizable systems with repeated eigenvalues, and then we concentrate in the<br />
case <strong>of</strong> non-diagonalizable systems.<br />
5.9.1. Diagonalizable systems. There is no new concepts in this case. If the n × n<br />
coefficient matrix A <strong>of</strong> the linear system x ′ = Ax is diagonalizable, then the results in<br />
Theorem 5.6.1 or Theorem 5.6.2 apply to this system. It does not matter whether matrix A<br />
has repeated eigenvalues or not, as long as A has a linear independent set <strong>of</strong> n eigenvectors.<br />
The following example illustrates this. The coefficient matrix has a repeated eigenvalue, but<br />
since this matrix is diagonalizable, we use Theorem 5.6.2 to write down the general solution<br />
<strong>of</strong> the <strong>differential</strong> system.<br />
Example 5.9.1: Find a fundamental set <strong>of</strong> solutions to<br />
x ′ ⎡<br />
3 0<br />
⎤<br />
1<br />
(t) = A x(t), A = ⎣0<br />
3 2⎦<br />
,<br />
0 0 1<br />
Solution: It is simple to see, and it was already computed in Example 5.5.7, that the<br />
characteristic polynomial p(λ) = det(A − λI) is given by<br />
p(λ) = −(λ − 3) 2 (λ − 1),<br />
hence the eigenvalue λ1 = 3 has algebraic multiplicity r1 = 2, while the eigenvalue λ2 = 1<br />
has algebraic multiplicity r2 = 1. We have also seen in Example 5.5.7 that the matrix A is<br />
diagonalizable, with eigenvalues and eigenvectors<br />
<br />
λ1 = 3,<br />
⎡<br />
⎣ 1<br />
⎤ ⎡<br />
0⎦<br />
, ⎣<br />
0<br />
0<br />
⎤<br />
1⎦<br />
0<br />
<br />
and λ2 = 1,<br />
⎡<br />
⎣ −1<br />
⎤<br />
−2⎦<br />
2<br />
<br />
.<br />
Since A is diagonalizable, we use Theorem 5.6.2 to write down a fundamental set <strong>of</strong> solution<br />
to x ′ = Ax, and the result is,<br />
⎡<br />
<br />
x1(t) = ⎣ 1<br />
⎤<br />
0⎦<br />
e<br />
0<br />
3t ⎡<br />
, x2(t) = ⎣ 0<br />
⎤<br />
1⎦<br />
e<br />
0<br />
3t ⎡<br />
, x3(t) = ⎣ −1<br />
⎤<br />
−2⎦<br />
e<br />
2<br />
t<br />
.<br />
5.9.2. Non-diagonalizable systems. Consider an n × n matrix A having an eigenvalue<br />
λ with algebraic multiplicity r = 2, so λ is repeated, and geometric multiplicity s = 1, so<br />
the eigenspace associated with λ has dimension one. A concrete example <strong>of</strong> this situation<br />
was given in Example 5.5.8. If v is an eigenvector associated with λ, then a solution to the<br />
<strong>differential</strong> system x ′ = Ax in this case is given by<br />
x (1) (t) = v e λt .<br />
⊳
218 G. NAGY – ODE july 2, 2013<br />
However, every other eigenvector associated with λ is proportional to the v above. In other<br />
words, every set <strong>of</strong> two eigenvectors associated with λ is linearly dependent. Then both<br />
Theorem 5.6.1 and Theorem 5.6.2 do not hold for this system. The following result, stated<br />
without a pro<strong>of</strong>, provides a linearly independent set <strong>of</strong> two solutions to the system x ′ = Ax<br />
associated with the repeated eigenvalue λ.<br />
Theorem 5.9.1 (Repeated eigenvalue). If an n × n matrix A has an eigenvalue λ with<br />
algebraic multiplicity r = 2 and only one associated eigen-direction, given by the eigenvector<br />
v, then the <strong>differential</strong> system x ′ (t) = A x(t) has a linearly independent set <strong>of</strong> solutions<br />
x (1) (t) = v e λt , x (2) (t) = v t + w e λt ,<br />
where the vector w is one <strong>of</strong> infinitely many solutions <strong>of</strong> the algebraic linear system<br />
(A − λI)w = v. (5.9.1)<br />
Remark: We know that matrix (A − λI) is not invertible, since λ is an eigenvalue <strong>of</strong> A,<br />
that is, det(A − λI) = 0. So an algebraic linear system with coefficient matrix (A − λI)<br />
is not consistent for every source. Nevertheless, the Theorem above says that Eq. (5.9.1)<br />
has solutions, and the property <strong>of</strong> v being an eigenvector <strong>of</strong> A is crucial to show that this<br />
system is consistent. This pro<strong>of</strong> is not reproduced here.<br />
Example 5.9.2: Find the fundamental solutions <strong>of</strong> the <strong>differential</strong> equation<br />
x ′ = Ax, A = 1<br />
<br />
−6 4<br />
.<br />
4 −1 −2<br />
Solution: As usual, we start finding the eigenvalues and eigenvectors <strong>of</strong> matrix A. The<br />
former are the solutions <strong>of</strong> the characteristic equation<br />
<br />
3<br />
0 = − 2 − λ 1<br />
− 1<br />
<br />
<br />
<br />
1<br />
4 − 2 − λ<br />
=<br />
<br />
λ + 3<br />
<br />
λ +<br />
2<br />
1<br />
<br />
+<br />
2<br />
1<br />
4 = λ2 + 2λ + 1 = (λ + 1) 2 .<br />
Therefore, there solution is<br />
λ = −1, r = 2.<br />
There is only one linearly independent eigenvector associated with this eigenvalue, and it is<br />
computed as the solution <strong>of</strong> the linear system (A + I)v = 0, that is,<br />
<br />
3 − 2 + 1 1<br />
− 1<br />
<br />
1<br />
−<br />
1 = 2 1<br />
4 − 2 + 1 − 1<br />
<br />
1 −2 1 −2<br />
1 → →<br />
⇒ v1 = 2v<br />
4 2 1 −2 0 0<br />
2.<br />
Choosing v2 = 1, then v1 = 2, and we obtain<br />
<br />
2<br />
λ = −1, v = , r = 2, s = 1.<br />
1<br />
Following Theorem 5.9.1 we now must solve for the vector w the linear system (A+I)w = v.<br />
The augmented matrix for this system is given by,<br />
<br />
1 − 2 1 <br />
2 1 −2 −4<br />
→ 1 −2<br />
1 1 −2 →<br />
−4 0 0<br />
<br />
<br />
<br />
<br />
<br />
−4<br />
0<br />
⇒ w1 = 2w2 − 4.<br />
− 1<br />
4<br />
1<br />
2<br />
We have obtained infinitely many solutions given by<br />
<br />
2 −4<br />
w = w2 + .<br />
1 0<br />
As one could have imagined, given any solution w, the cv + w is also a solution for any<br />
c ∈ R. We choose the simplest solution given by<br />
<br />
−4<br />
w = .<br />
0
G. NAGY – ODE July 2, 2013 219<br />
Therefore, a fundamental set <strong>of</strong> solutions to the <strong>differential</strong> equation above is formed by<br />
x (1) <br />
2<br />
(t) = e<br />
1<br />
−t , x (2) <br />
(t) =<br />
<br />
2 −4<br />
<br />
t + e<br />
1 0<br />
−t . (5.9.2)<br />
−x<br />
(1)<br />
x<br />
(2)<br />
w<br />
x 2<br />
Figure 33. The graph <strong>of</strong> the fundamental solutions x (1) , x (2) computed<br />
in Example 5.9.2 and given in Eq. (5.9.2), together with their negatives.<br />
The phase diagram <strong>of</strong> the fundamental solutions in Example 5.9.2 above are given in<br />
Fig. 33. This diagram can be obtained as follows. First draw the vectors v and w. The<br />
solution x (1) (t) is easy to draw, since it is represented by a straight line in the first quadrant<br />
parallel to v ending in the origin. The solution approaches the origin as the variable t grows<br />
since the eigenvalue λ = −1 is negative. The solution −x (1) is the the straight line in the<br />
third quadrant ending in the origin and parallel to −v.<br />
x (2)<br />
−x<br />
(1)<br />
w<br />
x 2<br />
v<br />
x<br />
(1)<br />
v<br />
−x<br />
(2)<br />
x 1 x 1<br />
−x (2)<br />
(1)<br />
−x<br />
w<br />
x<br />
(2)<br />
−x<br />
(1)<br />
x<br />
1<br />
x x 2<br />
(2)<br />
Figure 34. The trajectories <strong>of</strong> the functions x (1) , x (2) given in Eq. (5.9.3)<br />
for the cases λ < 0 and λ > 0, respectively.<br />
v<br />
x (1)<br />
⊳
220 G. NAGY – ODE july 2, 2013<br />
The solution x (2) (t) is more difficult to draw. One way is to first draw the trajectory <strong>of</strong><br />
the time-dependent vector<br />
˜x (2) <br />
2 −4<br />
= t + .<br />
1 0<br />
This is a straight line parallel to v passing through w, and is represented by a dashed line<br />
in Fig. 33. As t grows the function ˜x (2) moves from the third quadrant to the second and<br />
then to the first. The solution x (2) differs from ˜x (2) by the multiplicative factor e −t , so for<br />
t < 0 we have that x (2) (t) has bigger length than ˜x (2) (t), while the opposite happens for<br />
t > 0. In the limit t → ∞ the solution x (2) (t) approaches the origin, since the exponential<br />
factor e −t decreases faster than the linear factor t grows. The result is then the upper half<br />
<strong>of</strong> the s-shaped trajectory in Fig. 33. The lower half corresponds to −x (2) .<br />
Given any vectors v and w, and any constant λ, the phase diagrams <strong>of</strong> functions<br />
x (1) (t) = v e λt , x (2) (t) = v t + w e λt , (5.9.3)<br />
can be obtained in a similar way as the one in Fig. 33. The results are displayed in Fig. 34<br />
for the case λ < 0 and λ > 0, respectively.
5.9.3. Exercises.<br />
5.9.1.- . 5.9.2.- .<br />
G. NAGY – ODE July 2, 2013 221
222 G. NAGY – ODE july 2, 2013<br />
Chapter 6. Boundary value problems<br />
6.1. Eigenvalue-eigenfunction problems<br />
In this Section we consider second order, linear, <strong>ordinary</strong> <strong>differential</strong> <strong>equations</strong>. In the<br />
first half <strong>of</strong> the Section we study boundary value problems for these <strong>equations</strong> and in the<br />
second half we focus on a particular type <strong>of</strong> boundary value problems, called the eigenvalueeigenfunction<br />
problem for these <strong>equations</strong>.<br />
6.1.1. Comparison: IVP and BVP. Given real constants a1 and a0, consider the second<br />
order, linear, homogeneous, constant coefficients, <strong>ordinary</strong> <strong>differential</strong> equation<br />
y ′′ + a 1 y ′ + a 0 y = 0. (6.1.1)<br />
We now review the initial boundary value problem for the equation above, which was discussed<br />
in Sect. 2.2, where we showed in Theorem 2.2.2 that this initial value problem always<br />
has a unique solution.<br />
Definition 6.1.1 (IVP). Given the constants t0, y0 and y1, find a solution y <strong>of</strong> Eq. (6.1.1)<br />
satisfying the initial conditions<br />
y(t 0) = y 0, y ′ (t 0) = y 1. (6.1.2)<br />
There are other problems associated to the <strong>differential</strong> equation above. The following<br />
one is called a boundary value problem.<br />
Definition 6.1.2 (BVP). Given the constants t0 = t1, y0 and y1, find a solution y <strong>of</strong><br />
Eq. (6.1.1) satisfying the boundary conditions<br />
y(t0) = y0, y(t1) = y1. (6.1.3)<br />
One could say that the origins <strong>of</strong> the names “initial value problem” and “boundary value<br />
problem” originates in physics. Newton’s second law <strong>of</strong> motion for a point particle was<br />
the <strong>differential</strong> equation to solve in an initial value problem; the unknown function y was<br />
interpreted as the position <strong>of</strong> the point particle; the independent variable t was interpreted<br />
as time; and the additional conditions in Eq. (6.1.2) were interpreted as specifying the<br />
position and velocity <strong>of</strong> the particle at an initial time. In a boundary value problem, the<br />
<strong>differential</strong> equation was any equation describing a physical property <strong>of</strong> the system under<br />
study, for example the temperature <strong>of</strong> a solid bar; the unknown function y represented any<br />
physical property <strong>of</strong> the system, for example the temperature; the independent variable t<br />
represented position in space, and it is usually denoted by x; and the additional conditions<br />
given in Eq. (6.1.3) represent conditions on the physical quantity y at two different positions<br />
in space given by t 0 and t 1, which are usually the boundaries <strong>of</strong> the system under study, for<br />
example the temperature at the boundaries <strong>of</strong> the bar. This originates the name “boundary<br />
value problem”.<br />
We mentioned above that the initial value problem for Eq. (6.1.1) always has a unique<br />
solution for every constants y0 and y1, result presented in Theorem 2.2.2. The case <strong>of</strong> the<br />
boundary value problem for Eq. (6.1.1) is more complicated. A boundary value problem<br />
may have a unique solution, or may have infinitely many solutions, or may have no solution,<br />
depending on the boundary conditions. This result is stated in a precise way below.<br />
Theorem 6.1.3 (BVP). Fix real constants a1, a0, and let r± be the roots <strong>of</strong> the characteristic<br />
polynomial p(r) = r 2 + a 1r + a 0.<br />
(i) If the roots r± ∈ R, then the boundary value problem given by Eqs. (6.1.1) and (6.1.3)<br />
has a unique solution for all y 0, y 1 ∈ R.
G. NAGY – ODE July 2, 2013 223<br />
(ii) If the roots r± form a complex conjugate pair, that is, r± = α±βi, with α, β ∈ R, then<br />
the solution <strong>of</strong> the boundary value problem given by Eqs. (6.1.1) and (6.1.3) belongs<br />
to only one <strong>of</strong> the following three possibilities:<br />
(a) There exists a unique solution;<br />
(b) There exists infinitely many solutions;<br />
(c) There exists no solution.<br />
Before presenting the pro<strong>of</strong> <strong>of</strong> Theorem 6.1.3 concerning boundary value problem, let us<br />
review part <strong>of</strong> the pro<strong>of</strong> <strong>of</strong> Theorem 2.2.2 concerning initial value problems using matrix<br />
notation to highlight the crucial part <strong>of</strong> the calculations. For the simplicity <strong>of</strong> this review,<br />
we only consider the case r+ = r-. In this case the general solution <strong>of</strong> Eq. (6.1.1) can be<br />
expressed as follows,<br />
y(t) = c1 e r- t + c2 e r+ t , c1, c2 ∈ R.<br />
The initial conditions in Eq. (6.1.2) determine the values <strong>of</strong> the constants c 1 and c 2 as<br />
follows:<br />
y0 = y(t0) = c1 e r- t0 r+ t0 + c2 e<br />
y1 = y ′ (t0) = c1r- e r- t0 r+ t0 + c2r + e<br />
<br />
⇒<br />
<br />
r- t0 e r+ t0 e<br />
r- er-t0 r+ t0 r + e<br />
<br />
c1<br />
=<br />
c2 y0<br />
<br />
.<br />
y1 The linear system above has a unique solution c 1 and c 2 for every constants y 0 and y 1 iff<br />
the determinant <strong>of</strong> the coefficient matrix Z is non-zero, where<br />
A simple calculation shows<br />
Z =<br />
det(Z) = r+ − r-<br />
<br />
r- t0 e r+ t0 e<br />
r- er- t0 r+ t0 r+ e<br />
<br />
.<br />
e (r++r-) t0 = 0 ⇔ r+ = r-.<br />
Since r+ = r-, the matrix Z is invertible and so the initial value problem above a unique<br />
solution for every choice <strong>of</strong> the constants y 0 and y 1. The pro<strong>of</strong> <strong>of</strong> Theorem 6.1.3 for the<br />
boundary value problem follows the same steps we described above: First find the general<br />
solution to the <strong>differential</strong> equation, second find whether the matrix Z above is invertible<br />
or not.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 6.1.3:<br />
Part (i): Assume that r± are real numbers. We have two cases, r+ = r- and r+ = r-. In<br />
the former case the general solution to Eq. (6.1.1) is given by<br />
y(t) = c1 e r- t + c2 e r+ t , c1, c2 ∈ R. (6.1.4)<br />
The boundary conditions in Eq. (6.1.3) determine the values <strong>of</strong> the constants c1, c2, since<br />
y0 = y(t0) = c1e r- t0<br />
<br />
r+ t0 <br />
+ c2e r- t0 r+ t0<br />
e e<br />
⇒<br />
er- t1<br />
<br />
c1 y0<br />
r+ t1 = . (6.1.5)<br />
e c2 y1 y1 = y(t1) = c1e r- t1 r+ t1 + c2e<br />
The linear system above has a unique solution c 1 and c 2 for every constants y 0 and y 1 iff<br />
the determinant <strong>of</strong> the coefficient matrix Z is non-zero, where<br />
<br />
r- t0 r+ t0<br />
e e<br />
Z =<br />
. (6.1.6)<br />
A straightforward calculation shows<br />
er- t1 r+ t1 e<br />
det(Z) = e r+ t1 e r- t0 − e r+ t0 e r- t1 = e r+ t0 e r- t0 e r+ (t1−t0) − e r- (t1−t0) . (6.1.7)<br />
So it is simple to verify that<br />
det(Z) = 0 ⇔ e r+ (t1−t0) = e r- (t1−t0) ⇔ r+ = r -. (6.1.8)
224 G. NAGY – ODE july 2, 2013<br />
Therefore, in the case r + = r - the matrix Z is invertible and so the boundary value problem<br />
above has a unique solution for every choice <strong>of</strong> the constants y0 and y1. In the case that<br />
r+ = r- = r0, then we have to start over, since the general solution <strong>of</strong> Eq. (6.1.1) is not given<br />
by Eq. (6.1.4) but by the following expression<br />
y(t) = (c1 + c2 t) e r0t , c1, c2 ∈ R.<br />
Again, the boundary conditions in Eq. (6.1.3) determine the values <strong>of</strong> the constants c1 and<br />
c2 as follows:<br />
y0 = y(t0) = c1e r0t0 + c2t0e r0t0<br />
<br />
⇒<br />
<br />
r0t0 e t0er0t0 er0t1 t1er0t1 <br />
c1 y0<br />
= .<br />
c2 y1 y 1 = y(t 1) = c 1e r0t1 + c2t 1e r0t1<br />
The linear system above has a unique solution c1 and c2 for every constants y0 and y1 iff<br />
the determinant <strong>of</strong> the coefficient matrix Z is non-zero, where<br />
<br />
r0t0 e t0e<br />
Z =<br />
r0t0<br />
<br />
A simple calculation shows<br />
e r0t1 t1e r0t1<br />
det(Z) = t1e r0(t1+t0) − t0e r0(t1+t0) = (t1 − t0)e r0(t1+t0) = 0 ⇔ t1 = t0.<br />
Therefore, in the case r+ = r- = r0 the matrix Z is again invertible and so the boundary<br />
value problem above has a unique solution for every choice <strong>of</strong> the constants y0 and y1. This<br />
establishes part (i) <strong>of</strong> the Theorem.<br />
Part (ii): Assume that the roots <strong>of</strong> the characteristic polynomial have the form r± = α±βi<br />
with β = 0. In this case the general solution to Eq. (6.1.1) is still given by Eq. (6.1.4), and<br />
the boundary condition <strong>of</strong> the problem are still given by Eq. (6.1.5). The determinant <strong>of</strong><br />
matrix Z introduced in Eq. (6.1.6) is still given by Eq. (6.1.7). However, the claim given<br />
in Eq. (6.1.8) is true only in the case that r± are real numbers, and it does not hold in the<br />
case that r± are complex numbers. The reason is the following calculation: Let us start<br />
with Eq. (6.1.7) and then introduce that r± = α ± βi;<br />
det(Z) = e (r++r-) t0 e r+ (t1−t0) − e r- (t1−t0) <br />
= e 2αt0 e α(t1−t0) e iβ(t1−t0) − e −iβ(t1−t0) <br />
= 2i e α(t1+t0) sin β(t1 − t 0) .<br />
We then conclude that<br />
det(Z) = 0 ⇔ sin β(t1 − t0) nπ<br />
= 0 ⇔ β = , (6.1.9)<br />
(t1 − t0)<br />
where n = 1, 2, · · · and we are using that t1 = t0. This last equation in (6.1.9) is the key<br />
to obtain all three cases in part (ii) <strong>of</strong> this Theorem, as can be seen from the following<br />
argument:<br />
Part (iia): If the coefficients a1, a0 in Eq. (6.1.1) and the numbers t1, t0 in the boundary<br />
conditions in (6.1.3) are such that Eq. (6.1.9) does not hold, that is,<br />
nπ<br />
β =<br />
(t1 − t0) ,<br />
then det(Z) = 0 and so the boundary value problem for Eq. (6.1.1) has a unique solution<br />
for all constants y 0 and y 1. This establishes part (iia).<br />
Part (iib) and Part (iic): If the coefficients a1, a0 in Eq. (6.1.1) and the numbers t1, t0<br />
in the boundary conditions in (6.1.3) are such that Eq. (6.1.9) holds, then det(Z) = 0, and<br />
so the system <strong>of</strong> linear equation given in (6.1.5) may or may not have solutions, depending<br />
on the values <strong>of</strong> the constants y0 and y1. In the case that there exists a solution, then there
G. NAGY – ODE July 2, 2013 225<br />
are infinitely many solutions, since det(Z) = 0. This establishes part (iib). The remaining<br />
case, when y0 and y1 are such that Eq. (6.1.5) has no solution is the case in part (iic). This<br />
establishes the Theorem. <br />
Our first example is a boundary value problem with a unique solution. This corresponds<br />
to case (iia) in Theorem 6.1.3. The matrix Z defined in the pro<strong>of</strong> <strong>of</strong> that Theorem is<br />
invertible and the boundary value problem has a unique solution for every y 0 and y 1.<br />
Example 6.1.1: Find the solution y(x) to the boundary value problem<br />
y ′′ + 4y = 0, y(0) = 1, y(π/4) = −1.<br />
Solution: We first find the general solution to the <strong>differential</strong> equation above. We know<br />
that we have to look for solutions <strong>of</strong> the form y(x) = e rx , with the constant r being solutions<br />
<strong>of</strong> the characteristic equation<br />
r 2 + 4 = 0 ⇔ r± = ±2i.<br />
We know that in this case we can express the general solution to the <strong>differential</strong> equation<br />
above as follows,<br />
y(x) = c1 cos(2x) + c2 sin(2x).<br />
The boundary conditions imply the following system <strong>of</strong> linear equation for c1 and c2,<br />
<br />
1 = y(0) = c1 1 0 c1 1<br />
⇒<br />
= .<br />
−1 = y(π/4) = c2 0 1 c2 −1<br />
The linear system above has the unique solution c1 = 1 and c2 = −1. Hence, the boundary<br />
value problem above has the unique solution<br />
y(x) = cos(2x) − sin(2x).<br />
The following example is a small variation <strong>of</strong> the previous one, we change the value<br />
<strong>of</strong> the constant t1, which is the place where we impose the second boundary condition,<br />
from π/4 to π/2. This is enough to have a boundary value problem with infinitely many<br />
solutions, corresponding to case (iib) in Theorem 6.1.3. The matrix Z in the pro<strong>of</strong> <strong>of</strong> this<br />
Theorem 6.1.3 is not invertible in this case, and the values <strong>of</strong> the constants t0, t1, y0, y1, a1<br />
and a0 are such that there are infinitely many solutions.<br />
Example 6.1.2: Find the solution y(x) to the boundary value problem<br />
y ′′ + 4y = 0, y(0) = 1, y(π/2) = −1.<br />
Solution: The general solution is the same as in Example 6.1.1 above, that is,<br />
y(x) = c 1 cos(2x) + c 2 sin(2x).<br />
The boundary conditions imply the following system <strong>of</strong> linear equation for c1 and c2,<br />
<br />
1 = y(0) = c1 1 0 c1 1<br />
⇒<br />
= .<br />
−1 = y(π/2) = −c1 −1 0 c2 −1<br />
The linear system above has infinitely many solution, as can be seen from the following:<br />
<br />
1 0 | 1 1 0 | 1 c1 = 1,<br />
→<br />
⇒<br />
−1 0 | −1 0 0 | 0 c2 free.<br />
Hence, the boundary value problem above has infinitely many solutions given by<br />
y(x) = cos(2x) + c2 sin(2x), c2 ∈ R.<br />
⊳<br />
⊳
226 G. NAGY – ODE july 2, 2013<br />
The following example again is a small variation <strong>of</strong> the previous one, this time we change<br />
the value <strong>of</strong> the constant y1 from −1 to 1. This is enough to have a boundary value problem<br />
with no solutions, corresponding to case (iic) in Theorem 6.1.3. The matrix Z in the pro<strong>of</strong><br />
<strong>of</strong> this Theorem 6.1.3 is still not invertible, and the values <strong>of</strong> the constants t0, t1, y0, y1, a1<br />
and a 0 are such that there is not solution.<br />
Example 6.1.3: Find the solution y to the boundary value problem<br />
y ′′ (x) + 4y(x) = 0, y(0) = 1, y(π/2) = 1.<br />
Solution: The general solution is the same as in Examples 6.1.1 and 6.1.2 above, that is,<br />
y(x) = c1 cos(2x) + c2 sin(2x).<br />
The boundary conditions imply the following system <strong>of</strong> linear equation for c 1 and c 2,<br />
1 = y(0) = c1<br />
1 = y(π/2) = −c 1<br />
From the <strong>equations</strong> above we see that there is no solution for c1, hence there is no solution<br />
for the boundary value problem above.<br />
Remark: We now use matrix notation, in order to follow the same steps we did in the<br />
pro<strong>of</strong> <strong>of</strong> Theorem 6.1.3:<br />
<br />
1 0 c1<br />
=<br />
−1 0<br />
c2<br />
<br />
1<br />
.<br />
1<br />
The linear system above has infinitely many solutions, as can be seen from the following<br />
Gauss elimination operations<br />
<br />
1 0 <br />
1 1 0 1<br />
−1 0 → 1 0 1<br />
1 0 0 → <br />
2 0 0 1<br />
Hence, there are no solutions to the linear system above. ⊳<br />
6.1.2. Eigenvalue-eigenfunction problems. A particular type <strong>of</strong> boundary value problems<br />
are called eigenvalue-eigenfunction problems. The main example we study in this<br />
Section is the following: Find all the numbers λ and the non-zero functions with values y(x)<br />
solutions <strong>of</strong> the homogeneous boundary value problem<br />
y ′′ = λy, y(0) = 0, y(ℓ) = 0, ℓ > 0. (6.1.10)<br />
This problem is analogous to the eigenvalue-eigenvector problem studied in Sect. 5.5, that is,<br />
given an n × n matrix A find all numbers λ and non-zero vectors v solution <strong>of</strong> the algebraic<br />
linear system Av = λv. The role <strong>of</strong> matrix A is played by d 2 /dx 2 , the role <strong>of</strong> the vector<br />
space R n is played by the vector space <strong>of</strong> all infinitely differentiable functions f with domain<br />
[0, ℓ] ⊂ R satisfying f(0) = f(ℓ) = 0. We mentioned in Sect. 5.5 that given any n × n matrix<br />
A there exist at most n eigenvalues and eigenvectors. In the case <strong>of</strong> the boundary value<br />
problem in Eq. (6.1.10) there exist infinitely many solutions λ and y(x), as can be seen in<br />
the following result.<br />
Theorem 6.1.4 (Eigenvalues-eigenfunctions). The homogeneous boundary value problem<br />
in Eq. (6.1.10) has the infinitely many solutions, labeled by a subindex n ∈ N,<br />
λn = − n2 π 2<br />
ℓ2 , yn(x) = sin nπx <br />
.<br />
ℓ
G. NAGY – ODE July 2, 2013 227<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 6.1.4: We first look for solutions having eigenvalue λ = 0. In such a<br />
case the general solution to the <strong>differential</strong> equation in (6.1.10) is given by<br />
y(x) = c1 + c2x, c1, c2 ∈ R.<br />
The boundary conditions imply the following conditions on c1 and c2,<br />
<br />
0 = y(0) = c1, 0 = y(ℓ) = c1 + c2ℓ<br />
⇒ c1 = c2 = 0.<br />
Since the only solution in this case is y = 0, there are no non-zero solutions.<br />
We now look for solutions having eigenvalue λ > 0. In this case we redefine the eigenvalue<br />
as λ = µ 2 , with µ > 0. The general solution to the <strong>differential</strong> equation in (6.1.10) is given<br />
by<br />
y(x) = c 1e −µx + c 2e µx ,<br />
where we used that the roots <strong>of</strong> the characteristic polynomial r2 − µ 2 = 0 are given by<br />
r± = ±µ. The boundary conditions imply the following conditions on c1 and c2,<br />
<br />
0 = y(0) = c1 + c2,<br />
<br />
1 1<br />
⇒<br />
e−µℓ e µℓ<br />
<br />
c1 0<br />
= .<br />
0<br />
Denoting by<br />
we see that<br />
0 = y(ℓ) = c1e −µℓ + c2e µℓ<br />
Z =<br />
<br />
1 1<br />
e−µℓ e µℓ<br />
<br />
det(Z) = e µℓ − e −µℓ = 0 ⇔ µ = 0.<br />
Hence the matrix Z is invertible, and then we conclude that the linear system above for c 1,<br />
c2 has a unique solution given by c1 = c2 = 0, and so y = 0. Therefore there are no non-zero<br />
solutions y in the case that λ > 0.<br />
We now study the last case, when the eigenvalue λ < 0. In this case we redefine the<br />
eigenvalue as λ = −µ 2 , with µ > 0, and the general solution to the <strong>differential</strong> equation<br />
in (6.1.10) is given by<br />
y(x) = ˜c 1e −iµx + ˜c 2e iµx ,<br />
where we used that the roots <strong>of</strong> the characteristic polynomial r 2 + µ 2 = 0 are given by<br />
r± = ±iµ. In a case like this one, when the roots <strong>of</strong> the characteristic polynomial are<br />
complex, it is convenient to express the general solution above as a linear combination <strong>of</strong><br />
real-valued functions,<br />
y(x) = c 1 cos(µx) + c 2 sin(µx).<br />
The boundary conditions imply the following conditions on c1 and c2,<br />
<br />
0 = y(0) = c1, ⇒ c2 sin(µℓ) = 0.<br />
0 = y(ℓ) = c1 cos(µℓ) + c2 sin(µℓ)<br />
Since we are interested in non-zero solutions y, we look for solutions with c2 = 0. This<br />
implies that µ cannot be arbitrary but must satisfy the equation<br />
sin(µℓ) = 0 ⇔ µnℓ = nπ, n ∈ N.<br />
We therefore conclude that the eigenvalues and eigenfunctions are given by<br />
λn = − n2π2 ℓ2 , yn(x) = cn sin nπx <br />
.<br />
ℓ<br />
Choosing the free constants cn = 1 we establish the Theorem. <br />
c2
228 G. NAGY – ODE july 2, 2013<br />
Example 6.1.4: Find the numbers λ and the non-zero functions with values y(x) solutions<br />
<strong>of</strong> the following homogeneous boundary value problem<br />
y ′′ = λy, y(0) = 0, y ′ (π) = 0.<br />
Solution: This is also an eigenvalue-eigenfunction problem, the only difference with the<br />
case studied in Theorem 6.1.4 is that the second boundary condition here involves the<br />
derivative <strong>of</strong> the unknown function y. The solution is obtained following exactly the same<br />
steps performed in the pro<strong>of</strong> <strong>of</strong> Theorem 6.1.4.<br />
We first look for solutions having eigenvalue λ = 0. In such a case the general solution<br />
to the <strong>differential</strong> equation is given by<br />
y(x) = c1 + c2x, c1, c2 ∈ R.<br />
The boundary conditions imply the following conditions on c1 and c2,<br />
0 = y(0) = c 1, 0 = y ′ (π) = c 2.<br />
Since the only solution in this case is y = 0, there are no non-zero solutions with λ = 0.<br />
We now look for solutions having eigenvalue λ > 0. In this case we redefine the eigenvalue<br />
as λ = µ 2 , with µ > 0. The general solution to the <strong>differential</strong> equation is given by<br />
y(x) = c1e −µx + c2e µx ,<br />
where we used that the roots <strong>of</strong> the characteristic polynomial r2 − µ 2 = 0 are given by<br />
r± = ±µ. The boundary conditions imply the following conditions on c1 and c2, <br />
0 = y(0) = c1 + c2,<br />
1 1<br />
⇒<br />
−µe−µπ µe µπ<br />
<br />
c1 0<br />
= .<br />
c2 0<br />
Denoting by<br />
we see that<br />
0 = y ′ (π) = −µc 1e −µπ + µc 2e µπ<br />
<br />
Z =<br />
1 1<br />
−µe −µπ µe µπ<br />
det(Z) = µ e µπ + e −µπ = 0.<br />
Hence the matrix Z is invertible, and then we conclude that the linear system above for c 1,<br />
c 2 has a unique solution given by c 1 = c 2 = 0, and so y = 0. Therefore there are no non-zero<br />
solutions y in the case that λ > 0.<br />
We now study the last case, when the eigenvalue λ < 0. In this case we redefine the<br />
eigenvalue as λ = −µ 2 , with µ > 0, and the general solution to the <strong>differential</strong> equation<br />
in (6.1.10) is given by<br />
y(x) = ˜c1e −iµx + ˜c2e iµx ,<br />
where we used that the roots <strong>of</strong> the characteristic polynomial r 2 + µ 2 = 0 are given by<br />
r± = ±iµ. As we did in the pro<strong>of</strong> <strong>of</strong> Theorem 6.1.4, it is convenient to express the general<br />
solution above as a linear combination <strong>of</strong> real-valued functions,<br />
y(x) = c1 cos(µx) + c2 sin(µx).<br />
The boundary conditions imply the following conditions on c1 and c2,<br />
0 = y(0) = c1, 0 = y ′ <br />
⇒ c2 cos(µπ) = 0.<br />
(π) = −µc1 sin(µπ) + µc2 cos(µπ)<br />
Since we are interested in non-zero solutions y, we look for solutions with c2 = 0. This<br />
implies that µ cannot be arbitrary but must satisfy the equation<br />
cos(µπ) = 0 ⇔ µnπ = (2n + 1) π<br />
, n ∈ N.<br />
2
G. NAGY – ODE July 2, 2013 229<br />
We therefore conclude that the eigenvalues and eigenfunctions are given by<br />
(2n + 1)2<br />
λn = − ,<br />
4<br />
<br />
(2n + 1)x<br />
<br />
yn(x) = cn sin<br />
,<br />
2<br />
n ∈ N.<br />
Example 6.1.5: Find the numbers λ and the non-zero functions with values y(x) solutions<br />
<strong>of</strong> the homogeneous boundary value problem<br />
x 2 y ′′ − x y ′ = λ y, y(1) = 0, y(ℓ) = 0, ℓ > 1.<br />
Solution: This is also an eigenvalue-eigenfunction problem, the only difference with the<br />
case studied in Theorem 6.1.4 is that the <strong>differential</strong> equation is now the Euler equation,<br />
studied in Sect. 3.2, instead <strong>of</strong> a constant coefficient equation. Nevertheless, the solution is<br />
obtained following exactly the same steps performed in the pro<strong>of</strong> <strong>of</strong> Theorem 6.1.4.<br />
Writing the <strong>differential</strong> equation above in the standard form <strong>of</strong> an Euler equation,<br />
x 2 y ′′ − x y ′ − λy = 0,<br />
we know that the general solution to the Euler equation is given by<br />
y(x) = c 1 + c 2 ln(x) x r0<br />
in the case that the constants r+ = r- = r0, where r± are the solutions <strong>of</strong> the Euler characteristic<br />
equation<br />
r(r − 1) − r − λ = 0 ⇒ r± = 1 ± √ 1 + λ.<br />
In the case that r+ = r-, then the general solution to the Euler equation has the form<br />
y(x) = c 1x r- + c 2x r+ .<br />
Let us start with the first case, when the roots <strong>of</strong> the Euler characteristic polynomial are<br />
repeated r+ = r- = r0. In our case this happens if 1 + λ = 0. In such a case r0 = 1, and the<br />
general solution to the Euler equation is<br />
y(x) = c1 + c2 ln(x) x.<br />
The boundary conditions imply the following conditions on c1 and c2, 0 = y(1) = c1,<br />
0 = y(ℓ) = c1 + c2 ln(ℓ) <br />
⇒ c2ℓ ln(ℓ) = 0,<br />
ℓ<br />
hence c2 = 0. We conclude that the linear system above for c1, c2 has a unique solution<br />
given by c1 = c2 = 0, and so y = 0. Therefore there are no non-zero solutions y in the case<br />
that 1 + λ = 0.<br />
We now look for solutions having eigenvalue λ satisfying the condition 1 + λ > 0. In this<br />
case we redefine the eigenvalue as 1 + λ = µ 2 , with µ > 0. Then, r± = 1 ± µ, and so the<br />
general solution to the <strong>differential</strong> equation is given by<br />
y(x) = c 1x (1−µ) + c 2x (1+µ) ,<br />
The boundary conditions imply the following conditions on c1 and c2,<br />
0 = y(1) = c1 + c2,<br />
0 = y(ℓ) = c1ℓ (1−µ) + c2ℓ (1+µ)<br />
<br />
1 1<br />
⇒<br />
ℓ (1−µ) ℓ (1+µ)<br />
<br />
c1<br />
=<br />
c2 Denoting by<br />
Z =<br />
<br />
1 1<br />
ℓ (1−µ) ℓ (1+µ)<br />
<br />
<br />
0<br />
.<br />
0<br />
⊳
230 G. NAGY – ODE july 2, 2013<br />
we see that<br />
det(Z) = ℓ ℓ µ − ℓ −µ = 0 ⇔ ℓ = ±1.<br />
Hence the matrix Z is invertible, and then we conclude that the linear system above for c1,<br />
c 2 has a unique solution given by c 1 = c 2 = 0, and so y = 0. Therefore there are no non-zero<br />
solutions y in the case that 1 + λ > 0.<br />
We now study the second case, when the eigenvalue satisfies that 1 + λ < 0. In this case<br />
we redefine the eigenvalue as 1 + λ = −µ 2 , with µ > 0. Then r± = 1 ± iµ, and the general<br />
solution to the <strong>differential</strong> equation is given by<br />
y(x) = ˜c 1x (1−iµ) + ˜c 2x (1+iµ) ,<br />
As we did in the pro<strong>of</strong> <strong>of</strong> Theorem 6.1.4, it is convenient to express the general solution<br />
above as a linear combination <strong>of</strong> real-valued functions,<br />
y(x) = x c1 cos µ ln(x) + c2 sin µ ln(x) .<br />
The boundary conditions imply the following conditions on c1 and c2,<br />
0 = y(1) = c1, 0 = y(ℓ) = c1ℓ cos µ ln(ℓ) + c2ℓ sin (µ ln(ℓ) <br />
<br />
⇒ c2ℓ sin µ ln(ℓ) = 0.<br />
Since we are interested in non-zero solutions y, we look for solutions with c 2 = 0. This<br />
implies that µ cannot be arbitrary but must satisfy the equation<br />
sin µ ln(ℓ) = 0 ⇔ µn ln(ℓ) = nπ, n ∈ N.<br />
We therefore conclude that the eigenvalues and eigenfunctions are given by<br />
λn = −1 − n2π2 ln 2 (ℓ) , yn(x)<br />
<br />
nπ ln(x)<br />
<br />
= cnx sin ,<br />
ln(ℓ)<br />
n ∈ N.<br />
⊳
6.1.3. Exercises.<br />
6.1.1.- . 6.1.2.- .<br />
G. NAGY – ODE July 2, 2013 231
232 G. NAGY – ODE july 2, 2013<br />
6.2. Overview <strong>of</strong> Fourier series<br />
This Section is a brief introduction to the Fourier series expansions <strong>of</strong> periodic functions.<br />
We first recall the origins <strong>of</strong> this type <strong>of</strong> series expansions. We then review basic few notions<br />
<strong>of</strong> linear algebra that are satisfied by the set <strong>of</strong> infinitely differentiable functions. One crucial<br />
concept is the orthogonality <strong>of</strong> the sine and cosine functions. We then introduce the Fourier<br />
series <strong>of</strong> periodic functions, and we end this Section with the study two particular cases:<br />
The Fourier series <strong>of</strong> odd and <strong>of</strong> even functions, which are called sine and cosine series,<br />
respectively.<br />
6.2.1. Origins <strong>of</strong> Fourier series. The study <strong>of</strong> solutions to the wave equation in one space<br />
dimension by Daniel Bernoulli in the 1750s is a possible starting point to describe the origins<br />
<strong>of</strong> the Fourier series. The physical system is a vibrating elastic string with fixed ends, the<br />
unknown function with values u(t, x) represents the vertical displacement <strong>of</strong> a point in the<br />
string at the time t and position x, as can be seen in the sketch given in Fig. 35. A constant<br />
c > 0 characterizes the material that form the string.<br />
y<br />
0<br />
x<br />
u(t,x)<br />
Figure 35. A sketch <strong>of</strong> a vibrating string in one space dimension that<br />
moves only on the vertical direction and has fixed ends.<br />
The mathematical problem to solve is the following initial-boundary value problem: Given<br />
a function with values f(x) defined in the interval [0, ℓ] ⊂ R satisfying f(0) = f(ℓ) = 0, find<br />
a function with values u(t, x) solution <strong>of</strong> the wave equation<br />
∂ 2 t u(t, x) = c 2 ∂ 2 xu(t, x),<br />
u(t, 0) = 0, u(t, ℓ) = 0,<br />
u(0, x) = f(x), ∂tu(0, x) = 0.<br />
The <strong>equations</strong> on the second line are called boundary conditions, since they are conditions<br />
at the boundary <strong>of</strong> the vibrating string for all times. The <strong>equations</strong> on the third line are<br />
called initial conditions, since they are equation that hold at the initial time only. The first<br />
equation says that the initial position <strong>of</strong> the string is given by the function f, while the<br />
second equation says that the initial velocity <strong>of</strong> the string vanishes. Bernoulli found that<br />
the functions<br />
<br />
cnπt<br />
<br />
nπx<br />
<br />
un(t, x) = cos sin<br />
ℓ ℓ<br />
are particular solutions to the problem above in the case that the initial position function<br />
is given by<br />
He also found that the function<br />
u(t, x) =<br />
<br />
nπx<br />
<br />
fn(x) = sin .<br />
ℓ<br />
∞ <br />
cnπt<br />
<br />
nπx<br />
<br />
cn cos sin<br />
ℓ ℓ<br />
n=1<br />
x
G. NAGY – ODE July 2, 2013 233<br />
is also a solution to the problem above with initial condition<br />
∞ <br />
nπx<br />
<br />
f(x) = cn sin . (6.2.1)<br />
ℓ<br />
n=1<br />
Is the set <strong>of</strong> initial functions f given in Eq. (6.2.1) big enough to include all continuous<br />
functions satisfying the compatibility conditions f(0) = f(ℓ) = 0? Bernoulli said the answer<br />
was yes, and his argument was that with infinitely many coefficients cn one can compute<br />
every function f satisfying the compatibility conditions. Unfortunately this argument does<br />
not prove Bernoulli’s claim. A pro<strong>of</strong> would be a formula to compute the coefficients cn in<br />
terms <strong>of</strong> the function f. However, Bernoulli could not find such a formula.<br />
A formula was obtained by Joseph Fourier in the 1800s while studying a different problem.<br />
He was looking for solutions to the following initial-boundary value problem: Given a<br />
function with values f(x) defined in the interval [0, ℓ] ⊂ R satisfying f(0) = f(ℓ) = 0, and<br />
given a positive constant k, find a function with values u(t, x) solution <strong>of</strong> the <strong>differential</strong><br />
equation<br />
∂tu(t, x) = k ∂ 2 xu(t, x),<br />
u(t, 0) = 0, u(t, ℓ) = 0,<br />
u(0, x) = f(x).<br />
The values <strong>of</strong> the unknown function u(t, x) are interpreted as the temperature <strong>of</strong> a solid<br />
body at the time t and position x. The temperature in this problem does not depend on the<br />
y and z coordinates. The partial <strong>differential</strong> equation on the first line above is called the<br />
heat equation, and describes the variation <strong>of</strong> the body temperature. The thermal properties<br />
<strong>of</strong> the body material are specified by the positive constant k, called the thermal diffusivity.<br />
The main difference with the wave equation above is that only first time derivatives appear<br />
in the equation. The boundary conditions on the second line say that both borders <strong>of</strong> the<br />
body are held at constant temperature. The initial condition on the third line provides the<br />
initial temperature <strong>of</strong> the body. Fourier found that the functions<br />
nπ −(<br />
un(t, x) = e ℓ )2kt sin<br />
n=1<br />
<br />
nπx<br />
<br />
are particular solutions to the problem above in the case that the initial position function<br />
is given by<br />
<br />
nπx<br />
<br />
fn(x) = sin .<br />
ℓ<br />
Fourier also found that the function<br />
∞<br />
nπ −(<br />
u(t, x) = cn e ℓ )2 <br />
kt nπx<br />
<br />
sin<br />
ℓ<br />
is also a solution to the problem above with initial condition<br />
∞ <br />
nπx<br />
<br />
f(x) = cn sin . (6.2.2)<br />
ℓ<br />
n=1<br />
Fourier was able to show that any continuous function f defined on the domain [0, ℓ] ⊂ R<br />
satisfying the conditions f(0) = f(ℓ) = 0 can be written as the series given in Eq. (6.2.2),<br />
where the coefficients cn can be computed with the formula<br />
cn = 2<br />
ℓ<br />
ℓ<br />
0<br />
f(x) sin<br />
ℓ<br />
<br />
nπx<br />
<br />
dx.<br />
ℓ
234 G. NAGY – ODE july 2, 2013<br />
This formula for the coefficients cn, together with few other formulas that we will study<br />
later on in this Section, was an important reason to name after Fourier instead <strong>of</strong> Bernoulli<br />
the series containing those given in Eq. (6.2.2).<br />
6.2.2. Fourier series. Every continuous τ-periodic function f can be expressed as an infinite<br />
linear combination <strong>of</strong> sine and cosine functions. Before we present this result in a precise<br />
form we need to introduce few definitions and to recall few concepts from linear algebra.<br />
We start defining a peridic function saying that it is invariant under certain translations.<br />
Definition 6.2.1. A function f : R → R is called τ-periodic iff for all x ∈ R holds<br />
f(x − τ) = f(x), τ > 0.<br />
The number τ is called the period <strong>of</strong> f, and the definition says that a function τ-periodic iff<br />
it is invariant under translations by τ and so, under translations by any multiple <strong>of</strong> τ.<br />
Example 6.2.1: The following functions are periodic, with period τ,<br />
f(x) = sin(x), τ = 2π,<br />
f(x) = cos(x), τ = 2π,<br />
f(x) = tan(x), τ = π,<br />
f(x) = sin(ax), τ = 2π<br />
a .<br />
The following function is also periodic and its graph is given in Fig. 36,<br />
f(x) = e x , x ∈ [0, 2), f(x − 2) = f(x). (6.2.3)<br />
y<br />
y = f(x)<br />
0 1 x<br />
Figure 36. The graph <strong>of</strong> the function given in Eq. (6.2.3).<br />
Example 6.2.2: Show that the following functions are τ-periodic for all n ∈ N,<br />
<br />
2πnx<br />
<br />
<br />
2πnx<br />
<br />
fn(x) = cos , gn(x) = sin .<br />
τ<br />
τ<br />
Solution: The following calculation shows that fn is τ-periodic,<br />
<br />
2πn(x + τ)<br />
<br />
fn(x + τ) = cos<br />
,<br />
τ<br />
<br />
2πnx<br />
<br />
= cos + 2πn ,<br />
τ<br />
<br />
2πnx<br />
<br />
= cos ,<br />
τ<br />
= fn(x) ⇒ fn(xτ ) = fn(x).<br />
A similar calculation shows that gn is τ-periodic. ⊳<br />
⊳
G. NAGY – ODE July 2, 2013 235<br />
It is simple to see that a linear combination <strong>of</strong> τ-periodic functions is also τ-periodic.<br />
Indeed, let f and g be τ-periodic function, that is, f(x − τ) = f(x) and g(x − τ) = g(x).<br />
Then, given any constants a, b holds<br />
a f(x − τ) + b g(x − τ) = a f(x) + b g(x).<br />
A simple application <strong>of</strong> this result is given in the following example.<br />
Example 6.2.3: Show that the following function is τ-periodic,<br />
f(x) = a0<br />
2 +<br />
∞<br />
n=1<br />
<br />
an cos<br />
2πnx<br />
τ<br />
<br />
, +bn sin<br />
2πnx<br />
Solution: f is τ-periodic, since it is a linear combination <strong>of</strong> τ-periodic functions. ⊳<br />
The set <strong>of</strong> infinitely many differentiable functions f : [−ℓ, ℓ] ⊂ R → R, with ℓ > 0,<br />
together with the operation <strong>of</strong> linear combination <strong>of</strong> functions define a vector space. The<br />
main difference between this vector space and Rn , for any n 1, is that the space <strong>of</strong><br />
functions is infinite dimensional. An inner product in a vector space is an operation that<br />
associates to every pair <strong>of</strong> vectors a real number, and this operation is positive definite,<br />
symmetric and bilinear. An inner product in the vector space <strong>of</strong> functions is defined as<br />
follows: Given any two functions f and g, define its inner product, denoted by (f, g), as the<br />
following number,<br />
(f, g) =<br />
ℓ<br />
−ℓ<br />
f(x) g(x) dx.<br />
This is an inner product since it is positive definite,<br />
it is symmetric,<br />
(f, f) =<br />
ℓ<br />
−ℓ<br />
(f, g) =<br />
f 2 (x) dx 0, with<br />
ℓ<br />
−ℓ<br />
f(x) g(x) dx =<br />
and bilinear since it is symmetric and linear,<br />
f, [ag + bh] =<br />
= a<br />
ℓ<br />
ℓ<br />
−ℓ<br />
τ<br />
<br />
.<br />
<br />
<br />
(f, f) = 0 ⇔ f = 0 ;<br />
g(x) f(x) dx = (g, f);<br />
f(x) a g(x) + b h(x) dx<br />
−ℓ<br />
ℓ<br />
f(x) g(x) dx + b ∈<br />
−ℓ<br />
ℓ −ℓ f(x) h(x) dx<br />
= a(f, g) + b(f, h).<br />
An inner product provides the notion <strong>of</strong> angle in a vector space, and so the notion <strong>of</strong><br />
orthogonality <strong>of</strong> vectors. The idea <strong>of</strong> perpendicular vectors in the three dimensional space<br />
is indeed a notion that belongs to a vector space with an inner product, hence it can be<br />
generalized to the space <strong>of</strong> functions. The following result states that certain sine and cosine<br />
functions can be perpendicular to each oder.
236 G. NAGY – ODE july 2, 2013<br />
Theorem 6.2.2 (Orthogonality). The following relations hold for all n, m ∈ N,<br />
⎧<br />
ℓ <br />
nπx<br />
<br />
mπx<br />
⎪⎨<br />
0 n = m,<br />
cos cos dx = ℓ n = m = 0,<br />
−ℓ ℓ ℓ ⎪⎩<br />
2ℓ n = m = 0,<br />
ℓ <br />
nπx<br />
<br />
mπx<br />
<br />
0 n = m,<br />
sin sin dx =<br />
−ℓ ℓ ℓ<br />
ℓ n = m,<br />
ℓ <br />
nπx<br />
<br />
mπx<br />
<br />
cos sin dx = 0.<br />
ℓ ℓ<br />
−ℓ<br />
Remark: The pro<strong>of</strong> <strong>of</strong> this result is based in the following trigonometric identities:<br />
cos(θ) cos(φ) = 1<br />
<br />
cos(θ + φ) + cos(θ − φ) ,<br />
2<br />
sin(θ) sin(φ) = 1<br />
<br />
cos(θ − φ) − cos(θ + φ) ,<br />
2<br />
sin(θ) cos(φ) = 1<br />
<br />
sin(θ + φ) + sin(θ − φ) .<br />
2<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 6.2.2: We show the pro<strong>of</strong> <strong>of</strong> the first equation in Theorem 6.2.2, the<br />
pro<strong>of</strong> <strong>of</strong> the other two <strong>equations</strong> is similar. So, From the trigonometric identities above we<br />
obtain<br />
ℓ <br />
nπx<br />
<br />
mπx<br />
<br />
cos cos dx =<br />
−ℓ ℓ ℓ<br />
1<br />
ℓ <br />
(n + m)πx<br />
<br />
cos<br />
dx +<br />
2 −ℓ ℓ<br />
1<br />
ℓ <br />
(n − m)πx<br />
<br />
cos<br />
dx.<br />
2 −ℓ ℓ<br />
Now, consider the case that at least one <strong>of</strong> n or m is strictly greater than zero. In this case<br />
it holds the first term always vanishes, since<br />
ℓ<br />
1<br />
<br />
(n + m)πx<br />
<br />
ℓ<br />
cos<br />
dx =<br />
2<br />
ℓ<br />
2(n + m)π sin<br />
<br />
(n + m)πx<br />
ℓ <br />
= 0;<br />
ℓ −ℓ<br />
−ℓ<br />
while the remaining term is zero in the sub-case n = m, due to the same argument as above,<br />
ℓ<br />
1<br />
<br />
(n − m)πx<br />
<br />
ℓ<br />
cos<br />
dx =<br />
2<br />
ℓ<br />
2(n − m)π sin<br />
<br />
(n − m)πx<br />
ℓ <br />
= 0;<br />
ℓ −ℓ<br />
−ℓ<br />
while in the sub-case that n = m = 0 we have that<br />
ℓ<br />
1<br />
<br />
(n − m)πx<br />
<br />
cos<br />
dx =<br />
2<br />
ℓ<br />
1<br />
2<br />
−ℓ<br />
ℓ<br />
−ℓ<br />
dx = ℓ.<br />
Finally, in the case that both n = m = 0 is simple to see that<br />
ℓ <br />
nπx<br />
<br />
mπx<br />
ℓ<br />
cos cos dx = dx = 2ℓ.<br />
ℓ ℓ<br />
−ℓ<br />
This establishes the first equation in Theorem 6.2.2. The remaining <strong>equations</strong> are proven<br />
in a similar way. <br />
The main result <strong>of</strong> this Section is that any twice continuously differentiable function<br />
defined on an interval [−ℓ, ℓ] ⊂ R, with ℓ > 0, admits a Fourier series expansion.<br />
Theorem 6.2.3 (Fourier Series). Given ℓ > 0, assume that the functions f, f ′ f<br />
and<br />
′′ : [−ℓ, ℓ] ⊂ R → R are continuous. Then, f can be expressed as an infinite series<br />
f(x) = a0<br />
2 +<br />
∞ <br />
nπx<br />
<br />
nπx<br />
<br />
an cos + bn sin<br />
ℓ<br />
ℓ<br />
(6.2.4)<br />
n=1<br />
−ℓ
G. NAGY – ODE July 2, 2013 237<br />
with the constants an and bn given by<br />
an = 1<br />
ℓ <br />
nπx<br />
<br />
f(x) cos dx, n 0, (6.2.5)<br />
ℓ −ℓ<br />
ℓ<br />
bn = 1<br />
ℓ <br />
nπx<br />
<br />
f(x) sin dx, n 1. (6.2.6)<br />
ℓ<br />
ℓ<br />
−ℓ<br />
Furthermore, the Fourier series in Eq. (6.2.4) provides a 2ℓ-periodic extension <strong>of</strong> f from<br />
the domain [−ℓ, ℓ] ⊂ R to R.<br />
Remark: The Fourier series given in Eq. (6.2.4) also exists for functions f which are only<br />
piecewise continuous, although the pro<strong>of</strong> <strong>of</strong> the Theorem for such functions is more involved<br />
than in the case where f and its first two derivatives are continuous. In this notes we present<br />
the simpler pro<strong>of</strong>.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 6.2.3: We split the pro<strong>of</strong> in two parts: First, we show that if f admits<br />
an infinite series <strong>of</strong> the form given in Eq. (6.2.4), then the coefficients an and bn must<br />
be given by Eqs. (6.2.5)-(6.2.6); second, we show that the functions fN approaches f as<br />
N → ∞, where<br />
fN (x) = a0<br />
2 +<br />
N<br />
n=1<br />
<br />
an cos<br />
nπx<br />
ℓ<br />
<br />
+ bn sin<br />
nπx<br />
Regarding the first part, assume that f can be expressed by a series as given in Eq. (6.2.4).<br />
Then, multiply both sides <strong>of</strong> this equation by a cosine function and integrate as follows,<br />
ℓ <br />
mπx<br />
<br />
f(x) cos dx =<br />
−ℓ<br />
ℓ<br />
a0<br />
ℓ <br />
mπx<br />
<br />
cos dx<br />
2 −ℓ ℓ<br />
∞ ℓ <br />
mπx<br />
<br />
nπx<br />
<br />
+ an cos cos dx<br />
ℓ ℓ<br />
n=1<br />
ℓ<br />
+ bn<br />
−ℓ<br />
−ℓ<br />
ℓ<br />
<br />
.<br />
<br />
mπx<br />
<br />
nπx<br />
<br />
cos sin dx .<br />
ℓ ℓ<br />
In the case that m = 0 only the first term on the right-hand side above is nonzero, and<br />
Theorem 6.2.2 implies that<br />
ℓ<br />
−ℓ<br />
f(x) dx = a0<br />
2 2ℓ ⇒ a0 = 1<br />
ℓ<br />
ℓ<br />
−ℓ<br />
f(x) dx,<br />
which agrees with Eq. (6.2.5) for n = 0. In the case m 1 Theorem 6.2.2 implies that<br />
ℓ <br />
mπx<br />
<br />
f(x) cos dx = anℓ,<br />
ℓ<br />
−ℓ<br />
which again agrees with Eq. (6.2.5). Instead <strong>of</strong> multiplying by a cosine one multiplies the<br />
equation above by sin <br />
nπx<br />
ℓ , then one obtains Eq. (6.2.6). The second part <strong>of</strong> the pro<strong>of</strong> is<br />
similar and it is left as an exercise. This establishes the Theorem. <br />
Example 6.2.4: Find the Fourier series expansion <strong>of</strong> the function<br />
<br />
1 + x x ∈ [−1, 0),<br />
f(x) =<br />
1 − x x ∈ [0, 1].<br />
Solution: The Fourier series expansion is given by<br />
f(x) = a0<br />
2 +<br />
∞ <br />
an cos(nπx) + bn sin(nπx) ,<br />
n=1
238 G. NAGY – ODE july 2, 2013<br />
where the coefficients an, bn are given in Theorem 6.2.3. We start computing a0, that is,<br />
Similarly,<br />
an =<br />
=<br />
Recalling the integrals <br />
<br />
1<br />
−1<br />
0<br />
a0 =<br />
−1<br />
=<br />
1<br />
−1<br />
0<br />
−1<br />
f(x) dx<br />
(1 + x) dx +<br />
1<br />
0<br />
(1 − x) dx<br />
<br />
= x + x2<br />
0 <br />
<br />
+ x −<br />
2 −1<br />
x2<br />
1 <br />
2 0<br />
<br />
= 1 − 1<br />
<br />
+<br />
2<br />
1 − 1<br />
<br />
⇒ a0 = 1.<br />
2<br />
f(x) cos(nπx) dx<br />
(1 + x) cos(nπx) dx +<br />
1<br />
cos(nπx) dx = 1<br />
nπ sin(nπx),<br />
x cos(nπx) dx = x<br />
nπ<br />
0<br />
(1 − x) cos(nπx) dx.<br />
1<br />
sin(nπx) +<br />
n2 cos(nπx),<br />
π2 it is not difficult to see that<br />
an = 1<br />
nπ sin(nπx)<br />
<br />
<br />
0 <br />
x<br />
1<br />
+ sin(nπx) +<br />
−1 nπ n2 0 <br />
cos(nπx)<br />
π2 −1<br />
+ 1<br />
nπ sin(nπx)<br />
<br />
<br />
1 <br />
x<br />
1<br />
− sin(nπx) +<br />
0 nπ n2 1 <br />
cos(nπx)<br />
π2 0<br />
<br />
1<br />
=<br />
n2 1<br />
−<br />
π2 n2 <br />
1<br />
cos(−nπ) −<br />
π2 n2 1<br />
cos(−nπ) −<br />
π2 n2π2 <br />
,<br />
we then conclude that<br />
an = 2<br />
n2π2 <br />
1 − cos(−nπ) .<br />
Finally, we must find the coefficients bn. The calculation is similar to the one done above<br />
for an, and it is left as an exercise: Show that bn = 0. Then, the Fourier series <strong>of</strong> f is<br />
f(x) = 1<br />
2 +<br />
∞ 2<br />
n2π2 <br />
1 − cos(−nπ) cos(nπx).<br />
n=1<br />
6.2.3. Even and odd functions. There exist particular classes <strong>of</strong> functions with simple<br />
Fourier series expansions. Simple means that either the bn or the an coefficients vanish.<br />
These functions are called even or odd, respectively.<br />
Definition 6.2.4. Given any ℓ > 0, a function f : [−ℓ, ℓ ] → R is called even iff holds<br />
f(−x) = f(x), for all x ∈ [−ℓ, ℓ ].<br />
A function f : [−ℓ, ℓ ] → R is called odd iff holds<br />
f(−x) = −f(x), for all x ∈ [−ℓ, ℓ ].<br />
⊳
Example 6.2.5: Even functions are the following:<br />
G. NAGY – ODE July 2, 2013 239<br />
f(x) = cos(x), f(x) = x 2 , f(x) = 3x 4 − 7x 2 + cos(3x).<br />
Odd functions are the following:<br />
f(x) = sin(x), f(x) = x 3 , f(x) = 3x 3 − 7x + sin(3x).<br />
There exist functions that are neither even nor odd:<br />
f(x) = e x , f(x) = x 2 + 2x − 1.<br />
Notice that the product <strong>of</strong> two odd functions is even: For example f(x) = x sin(x) is even,<br />
while both x and sin(x) are odd functions. Also notice that the product <strong>of</strong> an odd function<br />
with an even function is again an odd function. ⊳<br />
The Fourier series <strong>of</strong> a function which is either even or odd is simple to find.<br />
Theorem 6.2.5 (Cosine and sine series). Consider the Fourier series <strong>of</strong> the function<br />
f : [−ℓ, ℓ] → R, that is,<br />
f(x) = a0<br />
2 +<br />
∞ <br />
nπx<br />
<br />
nπx<br />
<br />
an cos + bn sin .<br />
ℓ<br />
ℓ<br />
n=1<br />
(a) The function f is even iff the coefficients bn = 0 for all n 1. In this case the Fourier<br />
series is called a cosine series.<br />
(b) The function f is odd iff the coefficients an = 0 for all n 0. In this case the Fourier<br />
series is called a sine series.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 6.2.5:<br />
Part (a): Suppose that f is even, that is, f(−x) = f(x), and compute the coefficients bn,<br />
ℓ <br />
nπx<br />
<br />
bn = f(x) sin dx<br />
−ℓ<br />
ℓ<br />
−ℓ <br />
nπ(−y)<br />
<br />
= f(−y) sin (−dy), y = −x, dy = −dx,<br />
ℓ<br />
ℓ<br />
ℓ <br />
nπ(−y)<br />
<br />
= f(−y) sin dy,<br />
−ℓ<br />
ℓ<br />
ℓ <br />
nπy<br />
<br />
= − f(y) sin dy,<br />
−ℓ<br />
ℓ<br />
= −bn ⇒ bn = 0.<br />
Part (b): The pro<strong>of</strong> is similar. This establishes the Theorem.
240 G. NAGY – ODE july 2, 2013<br />
6.2.4. Exercises.<br />
6.2.1.- . 6.2.2.- .
G. NAGY – ODE July 2, 2013 241<br />
6.3. Application: The heat equation<br />
In this Section we show how to find the solution to an initial-boundary value problem<br />
for the heat equation. This equation is a partial <strong>differential</strong> equation, since the unknown<br />
function u depends on two independent variables t and x, and the partial derivatives <strong>of</strong><br />
both variables appear into the equation. We find the solution to this problem solving<br />
infinitely many initial value problems and boundary value problems for appropriate <strong>ordinary</strong><br />
<strong>differential</strong> <strong>equations</strong>.<br />
6.3.1. The initial-boundary value problem. Consider the solid bar sketched in Fig. 37.<br />
Assume that the bar is thermally insulated from the surrounding media except at the surfaces<br />
x = 0 and x = ℓ, where the temperature is held constant to T0 and Tℓ, respectively.<br />
This setting implies that the temperature in the bar does not depend on the coordinates<br />
y and z, it depends only on the space coordinate x and time t. The thermal properties <strong>of</strong><br />
the bar material are characterized by its thermal diffusivity k > 0, which we assume to be<br />
constant. Denoting by u(t, x) the bar temperature at the time t and the plane given by x<br />
constant, it is known that the temperature must satisfy the heat equation<br />
∂tu(t, x) = k ∂ 2 xu(t, x),<br />
where ∂ denotes partial derivative. Notice that this is a partial <strong>differential</strong> equation for the<br />
temperature function u.<br />
T = 0<br />
0<br />
z<br />
y<br />
0 x L<br />
insulation<br />
insulation<br />
T = 0<br />
Figure 37. A solid bar thermally insulated on all surfaces except the x = 0<br />
and x = L surfaces, which are held at temperatures T0 and TL, respectively.<br />
The temperature <strong>of</strong> the bar is a function <strong>of</strong> the coordinate x only.<br />
In this Section we are interested in solving an initial-boundary value problem for the<br />
bar temperature: Fix positive constant ℓ and k, fix the constants T0 = Tℓ = 0, and fix an<br />
infinitely differentiable function f : [0, ℓ] → R satisfying the conditions f(0) = f(ℓ) = 0.<br />
Then, find a function u : [0, ∞) × [0, ℓ] → R, with values u(t, x), solution <strong>of</strong> the initialboundary<br />
value problem,<br />
∂tu(t, x) = k ∂ 2 xu(t, x), (6.3.1)<br />
u(0, x) = f(x), (6.3.2)<br />
u(t, 0) = 0, u(t, ℓ) = 0. (6.3.3)<br />
The second line in the equation above is called the initial condition, while the <strong>equations</strong> on<br />
the third line above are called the boundary conditions <strong>of</strong> the problem. A sketch on the tx<br />
plane is useful to understand this type <strong>of</strong> problems, as can be seen in Fig. 38.<br />
x<br />
L
242 G. NAGY – ODE july 2, 2013<br />
u ( t, 0 ) = 0<br />
t<br />
t<br />
2<br />
x<br />
d u = k d u<br />
0 u ( 0, x ) = f ( x ) L<br />
u ( t, L ) = 0<br />
Figure 38. A sketch on the tx plane <strong>of</strong> an initial-boundary value problem<br />
for the heat equation.<br />
6.3.2. Solution <strong>of</strong> the initial-boundary value problem. We use the method <strong>of</strong> separation<br />
<strong>of</strong> variables to look for solutions to the Eqs. (6.3.1)-(6.3.3). The method <strong>of</strong> separation<br />
<strong>of</strong> variables can be summarized in three main steps. First, look for a solution u as an infinite<br />
series <strong>of</strong> the form u(t, x) = ∞ n=0 an un(t, x), where each function un is a solution <strong>of</strong> the<br />
heat equation having a particularly simple form, in the sense that<br />
un(t, x) = vn(t) wn(x).<br />
The name <strong>of</strong> the method, separation <strong>of</strong> variables, originates in the equation above. The heat<br />
equation for un implies particular <strong>ordinary</strong> <strong>differential</strong> <strong>equations</strong> for vn(t) and wn(x). So,<br />
second, solve an appropriate initial value problem for each function vn and an appropriate<br />
boundary value problem for each function wn, where the boundary conditions for wn are the<br />
same as the boundary conditions for the function u. There exist infinitely many solutions<br />
functions wn, which are mutually orthogonal. So, third, use the orthogonality relations<br />
among the functions wn to find the coefficients an in terms <strong>of</strong> the initial data function f.<br />
We then start looking for solutions <strong>of</strong> the form<br />
∞<br />
u(t, x) = an un(t, x), un(t, x) = vn(t) wn(x),<br />
n=0<br />
where each term is itself a solution <strong>of</strong> the heat equation, that is,<br />
∂tun(t, x) − k ∂ 2 xun(t, x) = 0. (6.3.4)<br />
Then it is clear that u will be solution <strong>of</strong> the heat equation, since<br />
∂tu − k∂ 2 ∞ <br />
xu = an ∂tun − k∂ 2 <br />
xun = 0.<br />
n=0<br />
returning to Eq. (6.3.4) we see that this equation implies<br />
wn(x) ∂tvn(t) = k vn(t)∂ 2 xwn(t) ⇒ 1 ∂tvn(t)<br />
k vn(t) = ∂2 xwn(x)<br />
wn(x) ,<br />
where the last equation is obtained dividing the previous one by un. Since the left-hand<br />
side on the last equation above depends only on t and the right-hand side depends only on<br />
x, then any solution vn and wn must be such that each side is equal to a constant, that is,<br />
1 ∂tvn(t)<br />
k vn(t) = ∂2 xwn(x)<br />
= λn,<br />
wn(x)<br />
where λn is a constant. The equation above contains two separate <strong>equations</strong>, given by<br />
1 ∂tvn(t)<br />
= λn,<br />
k vn(t)<br />
∂ 2 xwn(x)<br />
wn(x)<br />
x<br />
= λn.
G. NAGY – ODE July 2, 2013 243<br />
These <strong>equations</strong> are the motivation to propose the following problems for vn(t) and for<br />
wn(x), respectively:<br />
(i) First, find the function vn solution <strong>of</strong> initial value problem<br />
v ′ n(t) = kλn vn(t) vn(0) = 1. (6.3.5)<br />
(ii) Second, find all eigenvalues λn and all non-zero functions wn solutions <strong>of</strong> the boundary<br />
value problem<br />
w ′′<br />
n(x) = λn wn(x), wn(0) = 0, wn(ℓ) = 0. (6.3.6)<br />
The solution <strong>of</strong> Eq. (6.3.5) is straight forward to find, since it is a linear, first order, constant<br />
coefficient equation. We have seen in Theorem 1.1.5 that the solution is given by<br />
vn(t) = e kλnt .<br />
The eigenvalue-eigenfunction problem in Eq. (6.3.6) was already studied in Sect. 6.1 and<br />
the solution was found in Theorem 6.1.4. There is an infinite sequence <strong>of</strong> eigenvalues and<br />
eigenfunctions given by<br />
λn = − n2π2 ℓ2 , wn(x)<br />
<br />
nπx<br />
<br />
= sin , n ∈ N.<br />
ℓ<br />
With the functions vn and wn we can construct the function<br />
u(t, x) =<br />
∞<br />
n=1<br />
an e − n2π 2<br />
ℓ2 kt <br />
nπx<br />
<br />
sin .<br />
ℓ<br />
We verify that this function satisfies the boundary conditions, since<br />
∞<br />
u(t, 0) = an e<br />
n=1<br />
− n2π 2<br />
ℓ2 kt <br />
nπ0<br />
<br />
sin = 0,<br />
ℓ<br />
∞<br />
u(t, ℓ) = an e − n2π 2<br />
ℓ2 kt sin(nπ) = 0,<br />
n=1<br />
Therefore, the function u above is solution <strong>of</strong> Eqs. (6.3.1)-(6.3.3) iff it satisfies the initial<br />
condition u(0, x) = f(x), that is,<br />
∞ <br />
nπx<br />
<br />
f(x) = an sin . (6.3.7)<br />
ℓ<br />
n=1<br />
We have seen in Theorem 6.2.2 that the functions<br />
<br />
nπx<br />
∞ sin<br />
ℓ<br />
are mutually orthogonal on the interval [−ℓ, ℓ]. It is also not difficult to show that<br />
⎧<br />
ℓ <br />
nπx<br />
<br />
mπx<br />
⎨ 0 n = m,<br />
sin sin dx =<br />
0 ℓ ℓ ⎩<br />
ℓ<br />
n = m,<br />
2<br />
Therefore the coefficients am can be obtained as follows: Multiply the Eq. (6.3.7) by the<br />
function sin(mπx/ℓ) and integrate on the interval [0, ℓ]. Use the equation above to conclude<br />
that<br />
an = 2<br />
ℓ<br />
ℓ<br />
0<br />
n=1<br />
<br />
nπx<br />
<br />
f(x) sin dx.<br />
ℓ
244 G. NAGY – ODE july 2, 2013<br />
We therefore conclude that the solution to the initial-boundary value problem in Eqs. (6.3.1)-<br />
(6.3.3) is given by<br />
∞<br />
u(t, x) = an e − n2π 2<br />
ℓ2 kt <br />
nπx<br />
<br />
sin , an =<br />
ℓ<br />
2<br />
ℓ <br />
nπx<br />
<br />
f(x) sin dx.<br />
ℓ<br />
ℓ<br />
n=1<br />
0
6.3.3. Exercises.<br />
6.3.1.- . 6.3.2.- .<br />
G. NAGY – ODE July 2, 2013 245
246 G. NAGY – ODE july 2, 2013<br />
Chapter 1: First order <strong>equations</strong><br />
Chapter 7. Appendix<br />
7.1. Review exercises<br />
1.- . 2.- .
Chapter 2: Second order linear <strong>equations</strong><br />
1.- . 2.- .<br />
G. NAGY – ODE July 2, 2013 247
248 G. NAGY – ODE july 2, 2013<br />
Chapter 3: Power series solutions<br />
1.- . 2.- .
Chapter 4: The Laplace Transform<br />
1.- . 2.- .<br />
G. NAGY – ODE July 2, 2013 249
250 G. NAGY – ODE july 2, 2013<br />
Chapter 5: Systems <strong>of</strong> <strong>differential</strong> <strong>equations</strong><br />
1.- . 2.- .
Chapter 6: Boundary value problems<br />
1.- . 2.- .<br />
G. NAGY – ODE July 2, 2013 251
252 G. NAGY – ODE july 2, 2013<br />
7.2. Practice Exams<br />
Instructions to use the Practice Exams. The idea <strong>of</strong> these lecture notes is to help anyone<br />
interested in learning linear algebra. More <strong>of</strong>ten than not such person is a college student.<br />
An unfortunate aspect <strong>of</strong> our education system is that students must pass an exam to prove<br />
they understood the ideas <strong>of</strong> a given subject. To prepare the student for such exam is the<br />
purpose <strong>of</strong> the practice exams in this Section.<br />
These practice exams once were actual exams taken by student in previous courses. These<br />
exams can be useful to other students if they are used properly; one way is the following:<br />
Study the course material first, do all the exercises at the end <strong>of</strong> every Section; do the review<br />
problems in Section 7.1; then and only then take the first practice exam and do it. Think<br />
<strong>of</strong> it as an actual exam. Do not look at the notes or any other literature. Do the whole<br />
exam. Watch your time. You have only two hours to do it. After you finish, you can grade<br />
yourself. You have the solutions to the exam at the end <strong>of</strong> the Chapter. Never, ever look at<br />
the solutions before you finish the exam. If you do it, the practice exam is worthless. Really,<br />
worthless; you will not have the solutions when you do the actual exam. The story does not<br />
finish here. Pay close attention at the exercises you did wrong, if any. Go back to the class<br />
material and do extra problems on those subjects. Review all subjects you think you are<br />
not well prepared. Then take the second practice exam. Follow a similar procedure. Review<br />
the subject related to the practice exam exercises you had difficulties to solve. Then, do the<br />
next practice exam. You have three <strong>of</strong> them. After doing the last one, you should have a<br />
good idea <strong>of</strong> what your actual grade in the class should be.
Practice Exam 1. (Two hours.)<br />
1. .<br />
G. NAGY – ODE July 2, 2013 253
254 G. NAGY – ODE july 2, 2013<br />
Chapter 1: First order <strong>equations</strong><br />
7.3. Answers to exercises<br />
Section 1.1: Linear constant coefficients <strong>equations</strong><br />
1.1.1.- It is simple to see that:<br />
and that<br />
y ′ = 2(t + 2) e 2t + e 2t ,<br />
2y + e 2t = 2(t + 2) e 2t + e 2t .<br />
Hence y ′ = 2y + e 2t . Also holds that<br />
y(0) = (0 + 2) e 0 = 2.<br />
Section 1.2: Linear variable coefficients <strong>equations</strong><br />
1.2.1.- y(t) = 4e −t − e −2t .<br />
1.2.2.- y(t) = 2e t + 2(t − 1) e 2t .<br />
1.2.3.- y(t) = π cos(t)<br />
−<br />
2t2 t2 .<br />
1.2.4.- y(t) = c e (1+t2 ) 2<br />
, with c ∈ R.<br />
1.2.5.- y(t) = t2<br />
n + 2<br />
c<br />
+ , with c ∈ R.<br />
tn Section 1.3: Separable <strong>equations</strong><br />
1.3.1.- Implicit form: y2 t3<br />
= + c.<br />
2 3<br />
<br />
2t3 Explicit form: y = ± + 2c.<br />
3<br />
1.3.2.- y 4 + y = t − t 3 + c, with c ∈ R.<br />
1.3.3.- y(t) = 3<br />
.<br />
3 − t3 1.1.2.- y(t) = c e −4t + 1<br />
, with c ∈ R.<br />
2<br />
1.1.3.- y(t) = 9<br />
2 e−4t + 1<br />
2 .<br />
1.1.4.- y(x) = c e 6t − 1<br />
, with c ∈ R.<br />
6<br />
1.1.5.- y(x) = 7<br />
6 e6t − 1<br />
6 .<br />
1.1.6.- y(t) = 1<br />
3 e3(t−1) + 2<br />
3 .<br />
1.2.6.- y(t) = c e t2<br />
, with c ∈ R.<br />
Let y1 and y2 be solutions, that is,<br />
2ty1 − y ′ 1 = 0 and 2ty2 − y ′ 2 = 0. Then<br />
2t(y1 + y2) − (y1 + y2) ′ =<br />
(2ty1 − y ′ 1) + (2ty2 − y ′ 2) = 0.<br />
1.2.7.- Define v(t) = 1/y(t). The equation<br />
for v is v ′ = t v − t. Its solution is<br />
v(t) = c e t2 /2<br />
+ 1. Therefore,<br />
1<br />
y(t) =<br />
t et2 .<br />
/2 + 1<br />
1.2.8.- y(x) = 6 + c e −x2 /4 2<br />
√<br />
1+t2 1.3.4.- y(t) = c e .<br />
1.3.5.- y(t) = t ln(|t|) + c .<br />
1.3.6.- y 2 (t) = 2t 2 ln(|t|) + c .<br />
1.3.7.- Implicit: y 2 + ty − 2t = 0.<br />
Explicit: y(t) = 1 <br />
−t + t2 + 8t .<br />
2<br />
1.3.8.- Hint: Recall the Definition<br />
1.3.4 and use that<br />
y ′ 1(x) = f x, y1(x) ,<br />
for any independent variable x, for example<br />
for x = kt.
Section 1.4: Exact <strong>equations</strong><br />
1.4.1.-<br />
(a) The equation is exact. N = (1+t 2 ),<br />
M = 2t y, so ∂tN = 2t = ∂yM.<br />
(b) Since a potential function is given<br />
by ψ(t, y) = t 2 y + y, the solution is<br />
y(t) = c<br />
t2 , c ∈ R.<br />
+ 1<br />
1.4.2.-<br />
(a) The equation is exact. We have<br />
N = t cos(y) − 2y, M = t + sin(y),<br />
∂tN = cos(y) = ∂yM.<br />
(b) Since a potential function is given<br />
by ψ(t, y) = t2<br />
2 + t sin(y) − y2 , the<br />
solution is<br />
t 2<br />
2 + t sin(y(t)) − y2 (t) = c,<br />
for c ∈ R.<br />
G. NAGY – ODE July 2, 2013 255<br />
1.4.3.-<br />
(a) The equation is exact. We have<br />
N = −2y + t e ty , M = 2 + y e ty ,<br />
∂tN = (1 + t y) e ty = ∂yM.<br />
(b) Since the potential function is given<br />
by ψ(t, y) = 2t + e ty − y 2 , the solution<br />
is<br />
2t + e t y(t) − y 2 (t) = c,<br />
for c ∈ R.<br />
1.4.4.-<br />
(a) µ(x) = 1/x.<br />
(b) y 3 − 3xy + 18<br />
5 x5 = 1.<br />
1.4.5.-<br />
(a) µ(x) = x 2 .<br />
2<br />
(b) y = − √ . The negative<br />
1 + 2x4 square root is selected because the<br />
the initial condition is y(0) < 0.
256 G. NAGY – ODE july 2, 2013<br />
Section 1.5: Applications<br />
1.5.1.-<br />
(a) Denote m(t) the material mass as<br />
function <strong>of</strong> time. Use m in mgr and<br />
t in hours. Then<br />
m(t) = m0 e −kt ,<br />
where m0 = 50 mgr and k = ln(5)<br />
hours.<br />
(b) m(4) = 2<br />
25 mgr.<br />
(c) τ = ln(2)<br />
hours, so τ 0.43 hours.<br />
ln(5)<br />
1.5.2.- Since<br />
the condition<br />
implies that<br />
Q(t) = Q0e −(ro/V0)t ,<br />
Q1 = Q0e −(ro/V0)t1<br />
t1 = V0<br />
ro<br />
ln<br />
Q0<br />
Q1<br />
<br />
.<br />
Therefore, t = 20 ln(5) minutes.<br />
1.5.3.- Since<br />
−(ro/V0)t<br />
Q(t) = V0qi 1 − e <br />
and<br />
lim Q(t) = V0qi,<br />
t→∞<br />
the result in this problem is<br />
Q(t) = 300 1 − e −t/50<br />
and<br />
lim Q(t) = 300 grams.<br />
t→∞<br />
1.5.4.- Denoting ∆r = ri − ro and<br />
V (t) = ∆r t + V0, we obtain<br />
Q(t) =<br />
+ qi<br />
V0<br />
V (t)<br />
ro<br />
∆r<br />
Q0<br />
<br />
V (t) − V0<br />
V0<br />
V (t)<br />
ro<br />
∆r <br />
.<br />
A reordering <strong>of</strong> terms gives<br />
ro<br />
V0 ∆r<br />
Q(t) = qiV (t) − (qiV0 − Q0)<br />
V (t)<br />
and replacing the problem values yields<br />
Q(t) = t + 200 − 100<br />
(200) 2<br />
.<br />
(t + 200) 2<br />
The concentration q(t) = Q(t)/V (t) is<br />
q(t) = qi −<br />
V0<br />
V (t)<br />
ro<br />
∆r +1<br />
qi − Q0<br />
V0<br />
The concentration at V (t) = Vm is<br />
qm = qi −<br />
V0<br />
Vm<br />
ro<br />
∆r +1<br />
qi − Q0<br />
V0<br />
<br />
.<br />
<br />
,<br />
which gives the value<br />
qm = 121<br />
125 grams/liter.<br />
In the case <strong>of</strong> an unlimited capacity,<br />
limt→∞ V (t) = ∞, thus the equation for<br />
q(t) above says<br />
lim q(t) = qi.<br />
t→∞
Section 1.6: Non-linear <strong>equations</strong><br />
1.6.1.-<br />
(a) Write the equation as<br />
y ′ 2 ln(t)<br />
= −<br />
(t2 − 4) y.<br />
The equation is not defined for<br />
t = 0 t = ±2.<br />
This provides the intervals<br />
(−∞, −2), (−2, 2), (2, ∞).<br />
Since the initial condition is at t =<br />
1, the interval where the solution is<br />
defined is<br />
D = (0, 2).<br />
(b) The equation is not defined for<br />
t = 0, t = 3.<br />
This provides the intervals<br />
(−∞, 0), (0, 3), (3, ∞).<br />
Since the initial condition is at t =<br />
−1, the interval where the solution<br />
is defined is<br />
D = (−∞, 0).<br />
G. NAGY – ODE July 2, 2013 257<br />
1.6.2.-<br />
(a) y = 2<br />
3 t.<br />
(b) Outside the disk t 2 + y 2 1.<br />
1.6.3.-<br />
<br />
(a) Since y = y2 0 − 4t2 , and the initial<br />
condition is at t = 0, the solution<br />
domain is<br />
<br />
D = − y0<br />
<br />
y0<br />
, .<br />
2 2<br />
y0<br />
(b) Since y =<br />
1 − t2 and the initial<br />
y0<br />
condition is at t = 0, the solution<br />
domain is<br />
<br />
D = − 1<br />
√ ,<br />
y0<br />
1<br />
<br />
√ .<br />
y0
258 G. NAGY – ODE july 2, 2013<br />
Chapter 2: Second order linear <strong>equations</strong><br />
Section 2.1: Variable coefficients<br />
2.1.1.- . 2.1.2.- .<br />
Section 2.2: Constant coefficients<br />
2.2.1.- . 2.2.2.- .<br />
Section 2.3: Complex roots<br />
2.3.1.- . 2.3.2.- .<br />
Section 2.4: Repeated roots<br />
2.4.1.- . 2.4.2.- .<br />
Section 2.5: Undetermined coefficients<br />
2.5.1.- . 2.5.2.- .<br />
Section 2.6: Variation <strong>of</strong> parameters<br />
2.6.1.- . 2.6.2.- .
Chapter 3: Power series solutions<br />
Section 3.1: Regular points<br />
3.1.1.- . 3.1.2.- .<br />
Section 3.2: The Euler equation<br />
3.2.1.- . 3.2.2.- .<br />
Section 3.3: Regular-singular points<br />
3.3.1.- . 3.3.2.- .<br />
G. NAGY – ODE July 2, 2013 259
260 G. NAGY – ODE july 2, 2013<br />
Chapter 4: The Laplace Transform<br />
Section 4.1: Regular points<br />
4.1.1.- . 4.1.2.- .<br />
Section 4.2: The initial value problem<br />
4.2.1.- . 4.2.2.- .<br />
Section 4.3: Discontinuous sources<br />
4.3.1.- . 4.3.2.- .<br />
Section 4.4: Generalized sources<br />
4.4.1.- . 4.4.2.- .<br />
Section 4.5: Convolution solutions<br />
4.5.1.- . 4.5.2.- .
Chapter 5: Systems <strong>of</strong> <strong>differential</strong> <strong>equations</strong><br />
Section 5.1: Introduction<br />
5.1.1.- . 5.1.2.- .<br />
Section 5.2: Systems <strong>of</strong> algebraic <strong>equations</strong><br />
5.2.1.- . 5.2.2.- .<br />
Section 5.3: Matrix algebra<br />
5.3.1.- . 5.3.2.- .<br />
Section 5.4: Linear <strong>differential</strong> systems<br />
5.4.1.- . 5.4.2.- .<br />
Section 5.5: Diagonalizable matrices<br />
5.5.1.- . 5.5.2.- .<br />
Section 5.6: Constant coefficients systems<br />
5.6.1.- . 5.6.2.- .<br />
G. NAGY – ODE July 2, 2013 261
262 G. NAGY – ODE july 2, 2013<br />
Chapter 6: Boundary value problems<br />
Section 6.1: Eigenvalue-eigenfunction problems<br />
6.1.1.- . 6.1.2.- .<br />
Section 6.2: Overview <strong>of</strong> Fourier series<br />
6.2.1.- . 6.2.2.- .<br />
Section 6.3: Applications: The heat equation<br />
6.3.1.- . 6.3.2.- .
G. NAGY – ODE July 2, 2013 263<br />
References<br />
[1] W. Boyce and R. DiPrima. Elementary <strong>differential</strong> <strong>equations</strong> and boundary value problems. Wiley, New<br />
Jersey, 2009. 9th edition.<br />
[2] W. Rudin. Principles <strong>of</strong> Mathematical Analysis. McGraw-Hill, New York, NY, 1953.