DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE

NOTES

1 Functions, Derivatives, and Notation

A function, informally, is an input-output correspondence between values in a domain or input
space, and values in a codomain or output space. When studying a particular function in this course,
our domain will always be either a subset of the set of all real numbers, always denoted R, or a subset
of n-dimensional Euclidean space, always denoted Rn . Formally, Rn is the set of all ordered n-tuples
(x1 , x2 , ..., xn ) where each of x1 , x2 , ..., xn is a real number. Our codomain will always be R, i.e. our
functions under consideration will be real-valued.
We will typically functions in two ways. If we wish to highlight the variable or input parameter,
we will write a function using first a letter which represents the function, followed by one or more letters
in parentheses which represent input variables. For example, we write

f (x) or y(t) or F (x, y).

In the first two examples above, we implicitly assume that the variables x and t vary over all possible
inputs in the domain R, while in the third example we assume (x, y)√varies over all possible inputs in the
domain R2 . For some specific examples of functions, e.g. f (x) = x − 5, we assume that the domain
includes√only real numbers x for which it makes sense to plug x into the formula, i.e., the domain of
f (x) = x − 5 is just the interval [5, ∞).
Whenever it does not lead to confusion, we will refer to functions just by their names f , y, F , etc.,
and omit reference to the input variable. This shorthand can occasionally be cause for confusion, and
so the student will be asked to parse which letters are input variables and which are functions from the
context. For example, in the differential equation

x2
y0 = y2 cos y,

the use of the single-variable derivative notation y 0 implies that y = y(x) is a function of the single input
variable x. On the other hand in the differential equation

fxx + fyy = 0,

we note the use of partial derivative notation, and assume that f = f (x, y) is a function of the two
variables x and y. (More comments on derivative notation below.)

2 Remarks on Leibniz Notation (For the Students’ Reading Pleasure - Not

To Be Covered in Lecture)
The student will have seen at least two common families of notation for derivatives and partial deriva-
tives. One kind looks like:

y 0 , f 00 (x), y (4) , fx , fxy (x, y), ..., etc.,

while the other looks like

dy d2 d4 y δf δ 2
, 2 f (x), 4 , , f (x, y), etc.
dx dx dt δx δyδx
1
2 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

The former group is often referred to as the Lagrange notation, while the latter group is the
Leibniz notation. Both are acceptable and commonly used in practice, and have distinct advantages
and disadvantages over one another. The major advantage of the Leibniz notation is that it is completely
explicit about which variable the given derivative respects. However, the Leibniz notation is notoriously
a source of great confusion for calculus students (this fact may be mostly blamed on obsolete textbook
authorship), and so I ask the student to thoroughly understand its meaning before we proceed to the
main content of the course.
dy
The major drawback of the notation dx is that it is misleading: it looks like a fraction. The derivative
dy
dx is in no way a fraction: if y = y(x) is a real-valued function of a single-variable, then the derivative
dy dy
dx = dx (x) is a real-valued function of a single variable.
Moreover, the components of the notation, dy and dx, are not rigorously defined mathematical objects
of their own. Many textbooks refer to dy, dx, and similar symbols as “infinitesimals,” “differentials”
or “elements” (Isaac Newton called them “fluxions”!). This is an outdated misuse of terminology that
dy
should be abolished from the student’s brain and vocabulary. The notation dx is to be regarded as a
single unified notation which cannot be separated into constituent parts. For instance, we may write
dy
dx = 5x, which is a meaningful mathematical statement, but we do not consider it meaningful to
“multiply on both sides by dx” to obtain the statement “dy = 5xdx.” Unfortunately many authors
choose to use expressions like dy = 5xdx on a regular basis. Whenever we see a statement of the latter
form, we will regard it mentally as just a shorthand for a statement of the former form.
Why has the “fraction notation” become so entrenched when we are not really working with fractions
at all? One popular feature of the Leibniz notation is that it synergizes well with the chain rule for
derivatives and the substitution rule for integrals, in the sense that it makes them look nice visually.
Recall that if y(x) and u(x) are real-valued functions of a single variable x, then the composition y(u(x))
is a real-valued function of x, and the chain rule says that its derivative is equal to

d
dx y(u(x)) = y 0 (u(x)) · u0 (x).

People often switch to the Leibniz notation to write the chain rule in the following condensed form:

dy dy du
= · .
dx du dx

This form is aesthetically pleasing because it looks like cancellation of fractions (again, a source of
confusion). But there is clearly no cancellation happening; the above statement is of the form “function
dy
equals function times function.” Moreover, the statement above is ambiguous because the notation du ,
where y and u are both functions, is rarely if ever defined carefully at the introductory calculus level,
and the input variables, which are of crucial importance, are omitted. The useful content of the chain
dy
rule is that the derivative of the composition y ◦ u may be found by computing dx and du
dx separately,
and then adjusting the input variables. In other words, a correct version of the chain rule, written in
Leibniz notation, should look like:

dy dy du
(x) = (u(x)) · (x),
dx dx dx

which helps shatter any illusion of fraction cancellation.

A similar problem is encountered when we work with the substitution rule for integrals:

Rb R g(b)
a
y 0 (g(x)) · g 0 (x)dx = g(a)
y 0 (u)du.

Rb
(Crucially recall here that the string of symbols a dx means “take the integral of over the
interval [a, b] with respect to x”; the dx is again an inseparable part of the notation and should not be
regarded as a mathematical object all on its own.)
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 3

The substitution rule is usually applied in practice by means of the following mnemonic device: “Set
0
u = g(x). Then du
dx = g (x), and u runs over [g(a), g(b)] as x runs over [a, b], so

Rb Rb R g(b)
a
y 0 (g(x)) · g 0 (x)dx = a
y 0 (u) du
dx dx = g(a)
y 0 (u)du.”

Again we obtain the illusion of fraction cancellation, caused by the judicious omission of input vari-
ables, together with a highly suspect duplication of the symbol u as both a function and a dummy
variable of integration.
The moral of the story here is that the Leibniz notation can be useful (particularly as a mnemonic
device for certain calculus techniques) but it comes with many misleading characteristics, and so I will
ask the student to be very careful mentally when reading and working with these symbols. Especially
be aware, please, when an author is using “differentials” (dx, dy, dt etc.) as a mnemonic device or as a
non-rigorous shorthand notation.
As a last remark, the rigorousness of infinitesimals and infinitesimal notation was a major debate for
many decades and has become an interesting and important part of mathematical history. Our modern
approach to calculus avoids reference to infinitesimals by instead relying on the notion of a limit; this
approach is essentially due to Karl Weierstrass. It is actually possible to make things like differentials
and infinitesimals dy, dx, etc. rigorous notions (for example, google: “differential geometry” or “non-
standard analysis”) but these tools are well beyond the scope of our course and don’t really shed any
light on our present purposes. So we will regard dx, dy, etc. purely as notational conveniences, at their
best, and confusing encumbrances at their worst.

3 Terminology of Differential Equations

Definition 3.1. A differential equation is an equation relating some function to its derivatives. An
ordinary differential equation (ODE) is a differential equation involving a function of a single-
variable. A partial differential equation (PDE) is a differential equation involving a function of
multiple variables. A solution to a differential equation is any function which satisfies the equation.
Example 3.2 (Examples of Differential Equations). .

ODE’s PDE’s
dv
dt= −32 uxx + uyy = 0
0
y = 3x2 − 4x fx + f fyy = 0
y 00 + y = 0 δu δu
δx δy =u
d2 x
dt2+ 2 dx
+ 3x = sin t
dt Fx = 5x2 y − 3y
00 0
y + y − 2y = x3
Example 3.3. (1) Find all solutions to the differential equation y 0 = 3x2 − 4x.

(2) Find at least one solution to the differential equation y 00 + y = 0.

dv
Example 3.4. Find the unique solution to the differential equation dt = −32 which satisfies the initial
condition v(0) = −64.
Definition 3.5 (Informal). An initial value problem (IVP) is a differential equation together with
one or more initial conditions.
Example 3.6. Find the solution to the IVP y 0 = 3x2 − 4x, y(1) = 4.
Definition 3.7. The order of a differential equation is the highest degree of derivative which appears
in the equation.
Example 3.8. Find the order of each of the following differential equations.
dy x2
(1) dx = y2 cos y (degree 1)
4 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

(2) uxx + uyy = 0 (degree 2)

(3) Fxx − Fxyx = xy (degree 3)

dy 4
(4) ( dx ) =y+x (degree 1)

d7 x
(5) dt7 + 2 dx
dt + 3x = sin t (degree 7)
Example 3.9. (1) Show that y = c1 sin t + c2 cos t satisfies the ODE y 00 + y = 0 for all real numbers
c1 , c2 .
(2) Find a solution to the IVP y 00 + y = 0, y(0) = 0, y 0 (0) = 1.
Definition 3.10. An ODE (involving the function y of the independent variable x) of order n is called
linear if it may be written in the form

an (x)y (n) + an−1 (x)y (n−1) + ... + a2 (x)y 00 + a1 (x)y 0 + a0 (x)y = f (x),

where a0 , a1 , ..., an and f are functions of x and an is not the zero function. If f is identically the 0
function, then the ODE is called homogeneous. Note that the function y(x) = 0 is a solution to any
homogeneous ODE, called the trivial solution.
Example 3.11. Determine if the following ODE’s are (a) linear or nonlinear, and (b) homogeneous or
nonhomogeneous.
dy
(1) dx + x2 y = x (linear nonhomogeneous)
dy
(2) dx = x3 (linear nonhomogeneous)

(3) y − 1 + (x cos y)y 0 = 0 (nonlinear)

(4) u00 + u − ex = 0 (linear nonhomogeneous)

d2 f
(5) dx2 = −f (homogeneous)

(6) y (3) + yy 0 = x (nonlinear)

d2 x
(7) dt2 + sin x = 0 (nonlinear)
Example 3.12. Solve the following differential equations.
(1) y 0 = cos x (easy)
x
(2) y 0 = √ (substitution rule)
x2 +1
1
(3) y 0 = (trig substitution)
x2 + 16
(4) y 0 = xex (integration by parts)
1
(5) y 0 = (partial fractions)
4 − x2
(6) y 0 = y (remembering what we already know + cleverness)
Definition 3.13 (Informal). Given an ODE (involving a function y of an independent variable x), an
explicit solution is a function y = y(x) which satisfies the equation. An implicit solution is an equation
in y and x, such that if a given differentiable function satisfies the equation, then the function also
satisfies the given ODE.
Example 3.14. Show that an implicit solution to the ODE
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 5

2y − 4x − 5
y0 =
2y − 2x

is given by the equation 2x2 + y 2 − 2xy + 5x = 0. (Hint: implicit differentiation on the solution.)
Example 3.15 (Visualizing Solutions to ODEs). Sketch the slope field associated to the differential
2 2
equation y 0 = e−x . Sketch a possible solution. (For context, recall that e−x does not have a nice
explicit antiderivative.)

4 Existence of Solutions to First-Order Differential Equations

Theorem 4.1 (Picard-Lindelöf Existence and Uniqueness Theorem for First-Order ODE’s). Let f be a
function in two variables and consider the IVP

y 0 = f (t, y), y(t0 ) = y0 .

If f has first-order partial derivatives which are continuous on an open set containing (t0 , y0 ), then
there exists an open interval (t0 − h, t0 + h) on which there exists one and only one function which is a
solution to the IVP.
Example 4.2. Verify that the IVP y 0 = y, y(0) = 1 has a unique solution.

5 Separable Equations
Definition 5.1. A first-order ODE is called separable if it may be written in the form

g(y)y 0 = f (t),

for some functions g, f in one variable. (Note: some authors will use the shorthand g(y)dy = f (t)dt; see
Section 2 of these notes.)
Example 5.2. We wish to model the population P of a group of apes in the forest, as a function of
time t measured in years. Suppose that the birth rate of the apes is proportional to the population, so
that the birth rate may be written bP (t) for some constant b. Likewise suppose the mortality rate is
proportional to the population, say mP (t) for some constant m. Setting k = b − m, then the rate of
change of the ape population is modeled by

dP
=(Birth Rate)−(Mortality Rate)= bP − mP = kP .
dt
Assuming that the initial population of apes is P0 , find a model P (t) for the population by solving
the above first-order ODE.
Solution. First note that the equation is separable, as the equation P 0 = kP may be rewritten:

1
· P 0 = k.
P

Integrate both sides with respect to t:

1
P 0 (t)dt = kdt.
R R
P (t)

Note that the left-hand side is a substitution rule problem (mentally set u = P (t)). Integration gives us

ln |P (t)| + C1 = kt + C2
6 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

where C1 and C2 are arbitrary constants. Before finishing the problem, let us make a simple observation
that will serve us the remainder of the course: there is no reason to write two arbitrary constants when
one will do. Set C3 = C2 − C1 (so C3 is still arbitrary) and we get

ln |P (t)| = kt + C3 .

We finish the problem by exponentiating both sides:

|P (t)| = eln |P (t)| = ekt+C3 = ekt · eC3 .

Setting C = eC3 (so C is still an arbitrary constant) we find the implicit (general) solution |P (t)| =
Cekt . From the context we know P must be positive, so we write P (t) = Cekt . Lastly since P (0) = y0 ,
we get that C = Cek(0) = y0 , so the solution to our IVP is

P (t) = y0 ekt .

The student should verify that this is indeed a solution to the IVP. 

Technique 5.3 (General Method for Solving Separable First-Order ODE’s). (1) First separate all
the y-terms onto the one side of the equation and all of the t-terms onto the other side, so the
equation is in the form

g(y)y 0 = f (t).

dy
(If you wish, you may write y 0 = dt and use the mnemonic device g(y)dy = f (t)dt.)

(2) Integrate both sides of the equation with respect to t. The left-hand side is a substitution-rule
integration, so the y 0 -term ends up disappearing.

(3) Solve the equation for y if possible to obtain a general explicit solution. (You may be stuck with
just an implicit solution.) Replace arbitrary constants as necessary to keep things simple.

Example 5.4. Solve y 0 = − yt .

Example 5.5. Solve the IVP y 0 = e2t+y , y(0) = 0.

et
Example 5.6. Solve the IVP y 0 = , y(0) = −2.
y+1
Example 5.7. Solve the ODE y 0 = sin(t − y) + sin(t + y).

Definition 5.8. An autonomous first-order ODE is one of the form y 0 = f (y), i.e., an an autonomous
first-order ODE is one where the independent variable t does not explicitly appear. An equilibrium
solution to an autonomous ODE is a constant solution, i.e. a function of the form y(t) = C which
satisfies the equation.

Example 5.9. Consider y 0 = 2y − y 2 .

(1) Observe that the ODE is autonomous.
(2) Find a general solution to the ODE.
(3) Find any equilibrium solutions to the ODE.
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 7

6 First-Order Linear Equations

Definition 6.1. Recall that a first-order linear ODE is one of the form

y 0 + p(t)y = q(t)

for some functions p and q. The corresponding homogeneous equation is the ODE:

y 0 + p(t)y = 0,

i.e. we obtain the corresponding homogeneous equation by just dropping the q(t) term from the original
equation.
Recall from a previous course that if F1 and F2 are two functions for which F10 = y and F20 = y, then
F2 = F1 + C for some constant C. In other words, any two antiderivatives of a given function y differ
only by a constant. By analogy, the next theorem says that any two solutions to a first-order linear
ODE differ only by a solution to the corresponding homogeneous equation.
Theorem 6.2. Let y 0 + p(t)y = q(t) be a first-order linear ODE, and suppose y1 and y2 are two partic-
ular solutions. Then

y2 = y1 + yh ,

where yh is some solution to the corresponding homogeneous equation.

Proof. Define the function yh by setting yh = y2 − y1 . Then clearly, y2 = y1 + yh , so we need only check
that yh is a solution to the corresponding homogeneous equation, given by

y 0 + p(t)y = 0.

This is immediate because y1 and y2 are solutions to the original ODE; check that:

yh0 + p(t)y = (y20 − y10 ) + p(t)(y2 − y1 ) = (y20 + p(t)y2 ) − (y10 + y1 p(t)) = q(t) − q(t) = 0.

Corollary 6.3 (General Solution for First-Order Linear ODEs, Part I). Let yp be any particular solution
to the first-order linear ODE y 0 + p(t)y = q(t). Then every solution to the ODE is of the form

yp + yh ,

where yh is some solution to the corresponding homogeneous equation.

Example 6.4. Consider the ODE

dy 3t t
+ 2
y= .
dt 1+t (1 + t2 )2
1
(1) Show that yp = 1+t 2 is a particular solution.

(2) Find all solutions to the ODE.

Solution. Part (a) is a straightforward computation, so we take it as a fact and move on to part (b). By
the previous corollary, to find a general solution to the ODE it suffices to find a general solution to the
corresponding homogeneous equation

3t
y0 + y = 0.
1 + t2
8 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

Thankfully this latter equation is separable:

 
1 3t
y0 = − .
y 1 + t2

Integrating both sides (substitution rule on both sides, with u = 1 + t2 on the right-hand side) yields

ln |y| = − 32 ln |1 + t2 | + C.

Exponentiate both sides (and replace the arbitrary constant):

2
|y| = eC e−(3/2) ln |1+t | = C|1 + t2 |−3/2 .

Noting that (1 + t2 ) is always positive, and that the positivity or negativity of y may be absorbed
into the arbitrary constant C, we drop absolute value signs and arrive at the general solution

y = C(1 + t2 )−3/2 .

1
Since a particular solution to the original ODE is 1+t2 , we conclude that a general solution to the
original ODE looks like

1
y= + C(1 + t2 )−3/2 .
1 + t2

(The student should verify that this solution works.) 

We would like to develop a complete technique for solving first-order linear ODEs. To that end, we
introduce the following definition, which is otherwise impossible to motivate.
Definition 6.5. Let y 0 + p(t)y = q(t) be a first-order linear ODE, where p(t) is continuous. An inte-
grating factor µ(t) associated to the ODE is any function of the form
Rt
p(u)du
µ(t) = e c ,

where c is an arbitrary constant in the domain of p. In other words, µ is the exponentiation of any
antiderivative of p.
Theorem 6.6. Let y 0 + p(t)y = q(t) be a first-order linear ODE, where p(t) is continuous. Then the
general solution to the corresponding homogeneous equation y 0 + p(t)y = 0 is

y = Cµ(t)−1 ,

where µ is any integrating factor and C is any constant. Using the indefinite integral notation, we may
write the general solution as
R
y = e− p(t)dt
.

Proof. Use the separability of the corresponding homogeneous equation:

 
1
y 0 = −p(t),
y
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 9

and integrate both sides to obtain

R
ln |y| = − p(t)dt.

Let g(t) be any antiderivative of p(t), and set µ(t) = eg(t) , so µ is an integrating factor. Since any
two antiderivatives of p(t) differ only by a constant, we may write

ln |y| = −g(t) + C.

Exponentiating both sides, and consolidating arbitrary constants, yields the desired result:

y = Ce−g(t) = Cµ(t)−1 .

Technique 6.7 (General Method for Solving First-Order Linear ODEs). (1) Given a first-order lin-
0
ear ODE ofR
the form y + p(t)y = q(t), choose a representative µ(t) from the class of integrating
factors e− p(t)dt .

(2) Multiply both sides of the differential equation by µ(t) to obtain y 0 µ(t) + µ(t)p(t)y = µ(t)q(t).
Note that since µ0 (t) = µ(t)p(t), this equation may be rewritten as

y 0 (t)µ(t) + µ0 (t)y(t) = µ(t)q(t).

(3) Integrate both sides with respect to t. By the product rule, the left-hand side has y(t)µ(t) for
an antiderivative.

(4) Solve for y if possible to obtain a general explicit solution.

Corollary 6.8 (General Solution for First-Order Linear ODEs, Part II). Let y 0 + p(t)y = q(t) be a
first-order linear ODE where p and q are continuous, and let µ(t) be an integrating factor. Then the
general solution to the ODE is given by

y = µ(t)−1
R
q(t)µ(t)dt.
Example 6.9. Solve the IVP y 0 + 5t4 y = t4 , y(0) = 7.
dr π
Example 6.10. Find the general solution to the ODE dt = sin t − r tan t, 0 < t < 2.

Example 6.11. Show that the IVP ty 0 − y = t2 cos t, y(0) = 0 has infinitely many solutions. Why does
this not contradict the Picard-Lindelöf Theorem 4.1?
3t
Example 6.12. Solve y 0 − 2 y = t.
(t − 4)

7 Non-Linear Special Case: Exact Equations

Definition 7.1. A first-order ODE of the form

M (t, y) + N (t, y)y 0 = 0

is called exact if there exists a differentiable function f (a, b) for which

M (t, y) = fa (t, y) and N (t, y) = fb (t, y).

10 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

Such a function f is called potential function.

(To make a connection with multivariable calculus, note that an ODE is exact if and only if the vector
field (M, N ) is conservative, i.e. (M, N ) = ∇f for some potential function f .)
Theorem 7.2 (Test for Exactness). Suppose M and N have continuous partial derivatives on an open
domain D. The first-order ODE M (t, y) + N (t, y)y 0 = 0 is exact if and only if

δ δ
δy M = δt N .

Example 7.3. Determine whether the following equations are exact.

(1) 2ty 3 + (1 + 3t2 y 2 )y 0 = 0

(2) t2 y + 5ty 2 y 0 = 0
Theorem 7.4 (Multivariable Chain Rule). Let f (a, b) be a differentiable function of two variables and
let g(t) and h(t) be differentiable functions of one variable. Then

d
dt f (g(t), h(t)) = fa (g(t), h(t))g 0 (t) + fb (g(t), h(t))h0 (t).
Technique 7.5 (General Method for Solving First-Order Exact ODEs). (1) Given an exact equa-
tion M (t, y) + N (t, y)y 0 = 0, find a potential function f (a, b) for which M (t, y) = fa (t, y) and
N (t, y) = fb (t, y), and rewrite the equation as

fa (t, y) · 1 + fb (t, y) · y 0 = 0.
d
(2) Note, using the multivariable chain rule, that the above is of the form dt f (t, y(t)) = 0. Therefore
a general solution to the ODE is f (t, y) = C for an arbitrary constant C.

(3) Solve for y if possible to obtain a general explicit solution.

Example 7.6. Find a general solution of sin y + y cos t + (sin t + t cos y)y 0 = 0.
Example 7.7. Solve the IVP 2t sin y + (t2 cos y − 1)y 0 = 0, y(0) = 21 .

Example 7.8. Solve ey/t − yt ey/t + 1

1+t2 + ey/t y 0 = 0.

8 Non-Linear Special Case: Bernoulli Equations

Definition 8.1. A Bernoulli equation is a first-order ODE of the form

y 0 + p(t)y = q(t)y n

for some integer n.

Technique 8.2 (General Method for Solving Bernoulli Equations). (1) Given an equation of the
form y 0 + p(t)y = q(t)y n , multiply throughout by (1 − n)y −n to obtain

(1 − n)y −n y 0 + (1 − n)p(t)y 1−n = (1 − n)q(t).

(2) Make the substitution w = y 1−n , so w0 = (1 − n)y −n y 0 . This transforms the equation into

w0 + (1 − n)p(t)w = (1 − n)q(t).

(3) The new equation as above is linear; so solve for w using Technique 6.7.
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 11

(4) Once you have found w, un-substitute and solve for y by taking y = wn−1 . (Warning: if n is
odd, and hence w is an even power of y, then you will probably be stuck with just an implicit
solution.)
y
Example 8.3. Solve y 0 + x = 1
xy 2 , x > 0.

Example 8.4. Solve y 0 + y = ty 2 .

9 Non-Linear Special Case: Homogeneous Differential Equations

Definition 9.1. Let n be a positive integer. A function f (a, b) of two variables is called homogeneous
of degree n if

f (xa, xb) = xn f (a, b)

for every real number x.

Example 9.2. Show that the function f (a, b) = 2a2 − 7b2 + 4ab is homogeneous of degree 2.
Definition 9.3. A first-order ODE is called homogeneous of degree n if it can be written in the form

M (t, y) + N (t, y)y 0 = 0

where each of M and N is homogeneous of degree n.

Our use of the term homogeneous here bears absolutely no relation to its meaning in Definition 3.10.
With sympathy for the reader, we remark that homogeneous is just one of those words which over time
has accumulated an excess of different meanings in mathematical writing (see also for instance regular
or normal ), and this is sadly beyond the author’s power to change.
Technique 9.4 (General Method for Solving Homogeneous Equations). (1) The strategy here is to
make a clever substitution and thereby reduce to the separable case. Given an ODE of the form

M (t, y) + N (t, y)y 0 = 0

where both M and N are homogeneous of degree n, choose one of the following substitutions:

y
u= t (hence y = ut)

or

t
v= y (hence t = vy).

As a general guideline, the y = ut substitution tends to be better if N (t, y) is less complicated

than M (t, y), whereas the t = vy substitution tends to be better if it’s the other way around. If
you try one substitution and get stuck, try the other choice.

(2) If you chose y = ut:

(a) Observe that y 0 = u0 t + u by the product rule, and make the substitution to get:

M (t, ut) + N (t, ut)(u0 t + u) = 0.

By the homogeneity of M and N , we can rewrite this as:

tn M (1, u) + tn N (1, u)(u0 t + u) = 0.

12 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

(b) Divide out the tn from all terms to obtain

M (1, u) + N (1, u)(u0 t + u) = 0.

The resulting equation is separable in t and u:

 
N (1, u) 1
− u0 = ,
M (1, u) + uN (1, u) t
so proceed as in Technique 5.3 to integrate both sides and obtain an implicit solution in u
and t.

(c) Un-substitute u = yt , and solve for y if possible.

(3) If you chose t = vy:

(a) Observe that 1 = v 0 y + vy 0 by the product rule, and make the substitution to get:

M (vy, y)(v 0 y + vy 0 ) + N (vy, y)y 0 = 0.

By the homogeneity of M and N , we can rewrite this as:

y n M (v, 1)(v 0 y + vy 0 ) + y n N (v, 1)y 0 = 0.

(b) Divide out the y n from all terms to obtain

M (v, 1)(v 0 y + vy 0 ) + N (v, 1)y 0 = 0.

The equation above is separable in v and y:

   
1 M (v, 1)
y0 = − v0 ,
y vM (v, 1) + N (v, 1)
so proceed as in Technique 5.3 to integrate both sides and obtain an implicit solution in y
and v.

(c) Un-substitute v = yt , and solve for y if possible.

Example 9.5. Solve t3 + y 3 − ty 2 y 0 = 0 twice, using each of the substitutions y = ut and t = vy.
Assume that t > 0 and y > 0.

Example 9.6. Solve ty − y 2 + t(t − 3y)y 0 .

Technique 9.7 (Summary of Techniques for Solving First-Order ODEs).

(1) Separable Equations - Of the form g(y)y 0 = f (t). To solve: Separate and integrate.

(2) Linear Equations - Of the form y + p(t)y 0 = q(t). To solve: Multiply by an integrating
factor.

(3) Exact Equations - Of the form M (t, y) + N (t, y)y 0 = 0 where M , N are the partial derivatives
of some potential function. To solve: Find a potential function.

(4) Bernoulli Equations - Of the form y 0 + p(t)y = q(t)y n . To solve: Substitute and reduce
to the linear case.

(5) Homogeneous Equations - Of the form M (t, y)+N (t, y)y 0 = 0 where M , N are homogeneous.
To solve: Substitute and reduce to the separable case.
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 13

10 Applications: Population Growth

Definition 10.1. The Malthus model for population growth is the IVP discussed in Example 5.2, i.e.
y 0 = ky, y(0) = y0 ,

where y is population at time t, and y0 is some initial population. Recall that the solution to the above
IVP is

y = y0 ekt .
Despite the simplicity and effectiveness of the Malthus model, it has drawbacks. Primarily, the
Malthus model assumes that population growth is always directly proportional to population size, i.e.
y 0 = ky for constant k, and thus all population growth must be exponential. This is a situation that is
usually not encountered in the long term in reality, as typically the growth of a population depends on
the natural resources available in a given system. One way to try to reflect this reality is by changing
our fundamental assumption: we still assume that the growth y 0 is proportional to the total population
y, but now as the population y grows larger, the population growth will be inhibited due to scarcity of
resources. In other words, instead of assuming that y 0 is a constant k times population size at any time,
we assume that y 0 is a function (k − ay) times population size, where k and a are constant; as y increases
the proportion of y 0 to y decreases. This new assumption leads to consideration of the following famous
differential equation.
Definition 10.2. The logistic equation is the ODE

y 0 = (k − ay)y = ky − ay 2 ,

where k and a are constant.

Example 10.3. Solve the IVP given by the logistic equation and the initial condition y(0) = y0 .
Solution. The logistic equation is autonomous, hence separable. The usual methods yield the following
general solution:
ky0
y= .
ay0 + (k − ay0 )e−kt

Example 10.4. Show that if y is a solution to the logistic equation, then y is increasing but lim y is
t→∞
finite.
Definition 10.5. Let y be a solution to the logistic equation with given constants k and a. Then the
number ka = lim y is the maximum sustainable population of the model.
t→∞

11 Applications: Newton’s Law of Cooling

Theorem 11.1 (Newton’s Law of Cooling). The rate at which the temperature T changes with respect
to time t in a cooling body is proportional to the difference T − Ts between the temperature of the body
T and the constant temperature Ts of the surrounding medium. In other words,

T 0 = k(T − Ts ) for some constant k < 0.

Example 11.2. (1) Find a solution to the IVP T 0 = k(T − Ts ), T (0) = T0 .
(2) Compute lim T .
t→∞

Example 11.3. A pie is removed from a 350◦ Fahrenheit oven and placed to cool in a room with
temperature 75◦ Fahrenheit. In 15 minutes, the pie has a temperature of 150◦ Fahrenheit. Determine
the time required to cool the pie to a temperature of 80◦ .
14 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

12 Applications: Mixing Problems

Example 12.1. Suppose a tank holds V0 gallons of a brine solution. Another brine solution, with
concentration S1 pounds of salt per gallon, is allowed to flow into the tank at a rate R1 gallons per
minute. A drain at the bottom of the tank allows the mixture in the tank to flow out at a rate of R2
gallons per minute. We assume the brine flowing out is well-mixed. If y represents the number of pounds
of salt in the tank after t minutes, then y is modeled by the ODE:
y(t)
y 0 =(Rate salt enters the tank) − (Rate salt leaves the tank) = S1 R1 − V (t) · R2 ,

where V (t) is the volume of liquid in the tank at time t. This function V is modelled by the ODE

V 0 = R1 − R2 .
Example 12.2. Suppose S1 = 1, R1 = 4, and R2 = 3. In addition, suppose the tank initially holds 500
gallons of liquid and 250 pounds of salt.
(1) Find the amount of salt contained in the tank at any time.
(2) Determine how much time is required for 1000 pounds of salt to accumulate.
(3) If the tank holds a maximum of 800 gallons, can 1000 pounds of salt be present in the tank at
any time before the tank reaches its maximum capacity?

13 Complex Numbers and Euler’s Formula

Definition 13.1.√A complex number is a number of the form z = a + bi, where a and b are real
numbers and i = −1. The real part of z is the real number Re z = a, and the imaginary part of z
is the real number Im z = b. We denote the set of all complex numbers by C.

We visualize the complex numbers as a set by identifying each complex number a + bi with the point
(a, b) in R2 ; this gives a one-to-one correspondence between C and R2 and leads us to commonly refer
to C as the complex plane.

Utilizing this visual representation, the modulus of a complex number z = a + bi, denoted |z|, is the
Cartesian distance from (a, b) to the origin, i.e.
√
|z| = a2 + b2 .

In addition, by identifying C with R2 , we can formally talk about limits, continuity, etc. in C by
declaring that a limit exists in C if and only if the limit of the associated points exists in R2 (limits in
R2 were studied in the student’s multivariable calculus course).
Fact 13.2. Given any non-zero complex number z = a+bi, there exists a unique angle θ, with 0 ≤ θ < 2π,
called the argument of z, for which

z = |z|(cos θ + i sin θ).

Definition 13.3. Given two complex numbers z = a + bi and w = c + di, one can define their sum z + w
and their product zw by extending the addition and multiplication of the reals in the natural way. The
sum z + w is defined to be:

z + w = (a + bi) + (c + di) = a + c + bi + di = (a + c) + (b + d)i.

By associating a + bi and c + di in C with the points (a, b) and (c, d) in R2 in the usual way, it
is easy to see that complex addition corresponds exactly to vector addition in R2 . However, complex
multiplication (defined below) clearly distinguishes C from R2 :

zw = (a + bi)(c + di) = ac + adi + cbi + bdi2 = (ac − bd) + (ad + bc)i.

 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 15

It is straightforward to verify that the usual commutative properties of addition and multiplication,
associative properties of addition and multiplication, and distributive property of multiplication are true
in C just like they are in R.
Example 13.4. Let z = 1 + i and w = 2 + 3i. Compute z + w and zw.
Definition 13.5. The complex conjugate of a complex number z = a + bi, denoted z, is the number
z = a − bi.
Example 13.6. Show that for any complex number z, we have zz = |z|2 .
Example 13.7. Show that for any complex number z = a + bi and any complex number w = c + di 6= 0,
the quotient wz exists and is a unique complex number. Therefore division makes sense in C.
Initially the operations of multiplication and division in C may seem fairly mysterious. We will
understand them better after we observe Euler’s formula below, but in order to get Euler’s formula we
need to recall the following facts, developed in a previous calculus course.
Fact 13.8 (Maclaurin Series for ex , sin x, and cos x). The functions ex , sin x, and cos x have the fol-
lowing Maclaurin series representations:
∞
x
X xk x2 x3 x4
(1) e = =1+x+ + + + ...
k! 2! 3! 4!
k=0
∞
X x2k+1 x3 x5 x7
(2) sin x = (−1)k =x− + − + ...
(2k + 1)! 3! 5! 7!
k=0
∞
X x2k x2 x4 x6
(3) cos x = (−1)k =1− + − + ...
(2k)! 2! 4! 6!
k=0
Moreover each of the series above have interval of convergence (−∞, ∞) = R. In other words, the
equalities above hold for all real numbers x.
Until now we have always regarded the functions ex , sin x, and cos x as real-valued functions of a real
variable. The following fact says that there is a meaningful way to extend the functions ex , sin x, and
cos x to be complex-valued functions of a complex variable, that is, to have domain C and codomain
C. This fact is incredibly useful and non-obvious, and the proof is not very difficult to understand,
but is a little beyond the scope of our course and so we omit it. (One typically sees this proof in an
undergraduate complex analysis course.)
Fact 13.9. Each of the Maclaurin series in Fact 13.8 above converges not just for all real numbers x in
R, but in fact for all complex numbers x in C. Of course if x is real then ex , sin x, and cos x are real,
but if x is complex non-real then ex , sin x, and cos x are complex non-real.

Moreover, the function ex , now regarded as a complex function of a complex variable, still obeys the
exponentiation rule: for any two complex numbers z and w, ez+w = ez ew .
Theorem 13.10 (Euler’s Formula). For all real numbers x,

eix = cos x + i sin x.

Proof. The proof is by computation using the previous two facts:

(ix)2 (ix)3 (ix)4

eix = 1 + (ix) + + + + ...
2! 3! 4!
i2 x2 i 4 x4 i6 x6 i 2 x3 i4 x5 i6 x7
  
= 1+ + + + ... + i x + + + + ...
2! 4! 6! 3! 5! 7!
x2 x4 x6 x3 x5 x7
   
= 1− + − + ... + x − + − + ...
2! 4! 6! 3! 5! 7!
= cos x + i sin x,
16 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES


Now it is time to harvest the corollaries. The first is a cute fact relating some of our favorite mathe-
matical constants.
Corollary 13.11. eπi = −1.
Proof. Take x = π in Euler’s formula. 
The next two shed a great deal of light on the connection between the planar representation of C and
the multiplication in C.
Corollary 13.12. Suppose z is a complex number with argument θ. Then z = |z|eiθ .
Corollary 13.13. Let z be a complex number with argument θ1 , so z = |z|eiθ1 , and let w be a complex
number with argument θ2 , so w = |w|eiθ2 . Then the product zw is uniquely determined by the following
facts:
(1) zw has modulus |zw| = |z||w|; and

(2) zw has argument θ1 + θ2 .

Proof. Compute: zw = |z|eiθ1 |w|eiθ2 = |z||w|eiθ1 +iθ2 = |z||w|ei(θ1 +θ2 ) . 

14 Second-Order Linear Homogeneous Equations with Constant Coefficients

Definition 14.1. Recall that a second-order linear ODE is of the form

y 00 + p(t)y 0 + q(t)y = f (t).

Recall also that a second-order linear ODE in the above form is homogeneous if f (t) = 0. A
second-order linear homogeneous equation has constant coefficients if p(t) = p and q(t) = q are just
constants.
Example 14.2. Consider the ODE y 00 − 2y 0 − 15y = 0.
(1) Show that the equation has at least two solutions of the form y = eλ1 t , y = eλ2 t , where λ1 and
λ2 are real numbers.
(2) Show that any function of the form y = C1 eλ1 t +C2 eλ2 t , where C1 and C2 are arbitrary constants,
are also solutions to the ODE.
Solution - Characteristic Polynomial with Distinct Real Roots. (a) Let us drop subscripts for a moment
and consider functions of the form y = eλt . Differentiating once and twice, we get

y 0 = λeλt ;
y 00 = λ2 eλt .

Plugging these into the ODE, we get

λ2 eλt − 2λeλt − 15eλt = (λ2 − 2λ − 15)eλt = 0.

Judging from the above, y = eλt will be a solution if either λ2 − 2λ − 15 = 0 or eλt = 0. Since the
latter equality never happens (eλt is always positive), we need λ to be a root of the quadratic polynomial
λ2 − 2λ − 15 = (λ + 3)(λ − 5). So it suffices to take λ1 = −3 and λ2 = 5, i.e. y = e−3t and y = e5t are
solutions to the ODE (this is easy to verify directly).

(b) Suppose y1 = e−3t and y2 = e5t , so y1 and y2 are solutions to the ODE, and C1 and C2 are
arbitrary constants. Then taking y = C1 y1 + C2 y2 , we get
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 17

y 00 − 2y 0 − 15y = 0 = (C1 y1 + C2 y2 )00 − 2(C1 y1 + C2 y2 )y 0 − 15(C1 y1 + C2 y2 )

= C1 (y100 − 2y10 − 15y1 ) + C2 (y200 − 2y20 − 15y2 )
= C1 · 0 + C2 · 0 = 0.

So y = C1 y1 + C2 y2 is a solution as well. 
In the above example, given a second-order linear homogeneous ODE with constant coefficients, we
found infinitely many solutions of the form y = C1 eλ1 t + C2 eλ2 t . We will see in the next section that in
fact all of its solutions have this form.
The example and remarks above suggest that given a second-order linear homogeneous ODE with
constant coefficients ay 00 + by 0 + cy = 0, the task of finding solutions y = eλt may reduce in general to
the task of finding solutions to the associated polynomial equation aλ2 + bλ + c = 0.
Definition 14.3. Given a second-order linear homogeneous ODE with constant coefficients ay 00 + by 0 +
cy = 0, the associated characteristic polynomial is

aλ2 + bλ + c.

The characteristic equation is

aλ2 + bλ + c = 0.
Theorem 14.4. Let ay 00 + by 0 + cy = 0 be a second-order linear homogeneous ODE with constant
coefficients. If r is a solution to the characteristic equation, then y = ert is a solution to the ODE.
In our previous example, the characteristic polynomial happened to have two distinct real roots; what
happens if the polynomial has complex roots, or a single repeated real root? The next two examples
(and their preceding observations) address these situations.
Theorem 14.5. Let ay 00 + by 0 + cy = 0 be a second-order linear homogeneous ODE with constant
coefficients. A complex-valued function y is a solution to the ODE if and only if both Re y and Im y are
solutions.
Proof. Write f = Re y and g = Im y, so y = f + gi. Plugging in and using the linearity of the derivative,
we get

ay 00 + by 0 + cy = a(f 00 + g 00 i) + b(f 0 + g 0 i) + c(f + gi) = (af 00 + bf 0 + cf ) + (ag 00 + bg 0 + cg)i = 0.

The left-hand side will be 0 if and only if both the real part and the complex part of the right-hand
side are 0. So y is a solution if and only if both f and g are. 
Example 14.6. Solve y 00 + 4y 0 + 20y = 0.
Solution - Characteristic Polynomial with Complex Conjugate Roots. We proceed as in the previous ex-
ample by assuming that y has solutions of the form y = eλt . Differentiating to find y 0 = λeλt and
y 00 = λ2 eλt , we can plug into the differential equation and solve to obtain (as in the previous example):

(λ2 + 4λ + 20)eλt = 0.

Since eλt above is never zero, we have λ2 + 4λ + 20 = 0 and we wish to solve for λ. The quadratic
formula gives us:

λ = −2 + 4i or λ = −2 − 4i.
18 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

We may safely conclude that the complex-valued functions of a complex variable y = e(−2+4i)t and
y = e(−2−4i)t are solutions to the given differential equation. But we are interested in finding only
real-valued functions of a real variable which satisfy the ODE. Now given a complex function y, it is
easy to check that y satisfies a given differential equation if and only if both its real part Re y and its
imaginary part Im y satisfy the same equation. Thus to find some real-valued solutions, it suffices to
take the real and complex parts of our complex solutions using Euler’s formula. Recall that

e(−2+4i)t = e−2t e(4t)i = e−2t (cos(4t) + i sin(4t))

and likewise

e(−2−4i)t = e−2t e(−4t)i = e−2t (cos(−4t) + i sin(−4t)) = e−2 (cos(4t) − i sin(−4t))

Thus both of the functions y = e−2t cos(4t) and y = e−2t sin(4t) are solutions of the equation. (The
student should verify this manually.) And as in the previous example, it will turn out that all solutions
to the ODE are of the form

y = C1 e−2t cos(4t) + C2 e−2t sin(4t)

where C1 and C2 are arbitrary constants. (We will observe this rigorously in the next section.) 
Theorem 14.7. Let ay 00 + by 0 + cy = 0 be a second-order linear homogeneous ODE with constant
coefficients. Suppose the characteristic polynomial has a single repeated real root, i.e. aλ2 + bλ + c) =
a(λ − r)2 for some real number r. Then y = ert is a solution, and y = tert is also a solution.
Proof. The fact that y = ert is a solution follows from what we’ve already done. The fact that y = tert
is a clever non-obvious observation. Let us see that it’s true. Compute by the product rule:

y 0 = rtert + ert = (rt + 1)ert

and

y 00 = r(rt + 1)ert + rert = r(rt + 1 + 1)ert = (r2 t + 2r)ert .

Now plugging into the ODE we get

ay 00 + by 0 + cy = [a(r2 t + 2r) + b(rt + 1) + ct]ert

= [(ar2 + br + c)t + (2ra + b)]ert
b
= [(ar2 + br + c)t + a(2r + )]er t.
a
Now ar2 + br + c = 0 since r is a root of the characteristic polynomial. Even better, −2r = ab since
r is a repeated root (to see this, foil out the right-hand side of the equality λ2 + ab λ + ac = (λ − r)2 ), so
a(2r + ab ) = 0. Thus the right-hand side above equals 0, showing that y = tert is a solution. 
Example 14.8. Solve y 00 + 2y 0 + 1 = 0.
Solution - Characteristic Polynomial with Real Repeated Root. Once again assume there are solutions of
the form y = eλt . Differentiate, plug in and solve (see previous examples) to reduce to the following
equation:
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 19

λ2 + 2λ + 1 = 0.

The above factors into (λ + 1)2 = 1, and thus has only one real solution λ = −1. So y = e−t and
y = te−t are solutions. We conclude that y = C1 e−t + C2 tet is a general solution to the ODE. 
Technique 14.9 (General Method for Solving Second-Order Linear Homogeneous ODE’s with Constant Coefficients).
Given an ODE of the form ay 00 + by 0 + c = 0, solve the associated characteristic equation
aλ2 + bλ + c = 0.

(2) If the characteristic equation has two distinct real solutions λ1 and λ2 , then a general solution
to the ODE is given by y = C1 eλ1 t + C2 eλ2 t .

(3) If the characteristic equation has one repeated real solution λ, then a general solution to the
ODE is given by y = C1 eλt + C2 teλt .

(4) If the characteristic equation has two complex conjugate solutions z = a + bi and z = a − bi,
then a general solution to the ODE is given by y = C1 eat cos(bt) + C2 eat sin(bt).

15 The Wronskian and the Structure of Solutions to Second-Order Linear

Homogeneous ODEs
The main purpose of this section will be to justify the fact that the solution sets given in Technique
14.9 in fact encompass all possible solutions to a given second-order linear homogeneous ODE. First
we will state without proof an analogue of the Picard-Lindelöf theorem for second-order differential
equations.
Theorem 15.1 (Existence and Uniqueness of Solutions for Second-Order Linear Equations). Suppose
that p(t), q(t), and f (t) are continuous functions on an open interval I containing t0 , and consider the
IVP

y 00 + p(t)y 0 + q(t)y = f (t), y(t0 ) = y0 , y 0 (t0 ) = y00 .

Then there exists one and only one solution to the IVP on the interval I.
As we proceed, we will need to recall some terminology from linear algebra.
Definition 15.2. Let f1 , f2 , ..., fn be a finite collection of functions of an independent variable t. If
c1 , c2 , ..., cn are any constants, then the function

c1 f1 + c2 f2 + ... + cn fn

is called a linear combination of the functions f1 , f2 , ..., fn .

The functions f1 , ..., fn are called linearly dependent if there exist constants c1 , c2 , ..., cn , at least
one of which is non-zero, for which

c1 f1 + c2 f2 + ... + cn fn = 0.

If f1 , ..., fn are not linearly dependent, they are called linearly independent.
Example 15.3. Determine if the following pairs of functions are linearly dependent or linearly inde-
pendent.
(1) f1 (t) = 2t; f2 (t) = 4t

(2) f1 (t) = et ; f2 (t) = e−2t

20 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

(3) f1 (t) = sin 2t, f2 (t) = 5 sin t cos t

(4) f1 (t) = t, f2 (t) = 0

Theorem 15.4. Two functions f1 and f2 are linearly dependent if and only if one is a constant multiple
of the other.
Theorem 15.5 (Principle of Superposition). Suppose f1 and f2 are two solutions to the linear homo-
geneous ODE y 00 + p(t)y 0 + q(t)y = 0 on the interval I, and let c1 and c2 be any constants. Then the
linear combination c1 f1 (t) + c2 f2 (t) is also a solution on the interval I.
Proof. Plug it in and see that it works! 
Definition 15.6. Let f1 (t) and f2 (t) be two differentiable functions of an independent variable t. The
Wronskian of f1 and f2 is the function
 
f1 (t) f2 (t)
W (f1 , f2 )(t) = det .
f10 (t) f20 (t)
Example 15.7. Compute the Wronskian of the following functions.
(1) cos t and sin t.
(2) et and tet .
(3) 2t and 4t.
(4) sin(2t) and 5 sin t cos t.
Theorem 15.8. Let y1 and y2 be two solutions to the ODE y 00 + p(t)y 0 + q(t)y = 0, where p(t) and q(t)
are continuous. Then y1 and y2 are linearly dependent if and only if W (y1 , y2 ) = 0, i.e. W (y1 , y2 ) is
identically the zero function.
Proof. If y1 and y2 are linearly dependent, then there is a constant C for which y2 = Cy1 . In this case
W (y1 , y2 ) = y1 y20 − y10 y2 = y1 (Cy10 ) − y10 (Cy1 ) = 0.
Conversely, suppose W (y1 , y2 ) = 0. Choose any point t0 in the domain of y1 and y2 . Since the
Wronskian is 0, we have y1 (t0 )y20 (t0 ) − y10 (t0 )y2 (t0 ) = 0 and hence

y1 (t0 ) y 0 (t0 )
= 10 .
y2 (t0 ) y2 (t0 )

y 0 (t )
Let C = yy12 (t0) 1 0
(t0 ) = y20 (t0 ) . Now consider the function y = y1 − Cy2 . By the Principle of Superposition
Theorem 15.5, y is a solution to the given ODE. In addition, notice that

y(t0 ) = y1 (t0 ) − Cy2 (t0 ) = 0 and y 0 (t0 ) = y10 (t0 ) − Cy20 (t0 ) = 0.

In other words, y = y1 −Cy2 is a solution to the IVP y 00 +p(t)y 0 +q(t)y = 0, y(t0 ) = 0, y 0 (t0 ) = 0. But
the constant zero function 0 is also a solution to this IVP! Therefore by the Existence and Uniqueness
Theorem 15.1, we must have y1 − Cy2 = 0. It follows that y1 = Cy2 , whence y1 and y2 are a linearly
dependent pair, as claimed. 
Lemma 15.9. Let y1 and y2 be functions. Then d
dt W (y1 , y2 ) = y1 y200 − y100 y2 .
Proof.
d d
W (y1 , y2 ) = (y1 y20 − y10 y2 )
dt dt
= (y1 y200 + y10 y20 ) − (y10 y20 + y100 y2 )
= y1 y200 − y100 y2 .

 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 21

Lemma 15.10 (Abel’s Theorem). Let y1 and y2 be two solutions of the second-order linear homogeneous
ODE y 00 + p(t)y 0 + q(t)y = 0, where p and q are continuous functions of t. Let P (t) be an antiderivative
of p(t). If y1 and y2 are linearly independent, then

W (y1 , y2 )(t) = Ce−P (t) ,

where C is a non-zero constant.

Proof. Suppose y1 and y2 are two independent solutions to y 00 + p(t)y 0 + q(t)y = 0. Since y1 and y2 are
solutions to the ODE, we have (omitting arguments):

qy1 = −y100 − py10 and qy2 = −y200 − py2 .

Therefore,

0 = y2 · qy1 − y1 · qy2
= y2 (−y100 − py10 ) − y1 (−y200 − py20 )
= (y1 y200 − y100 y2 ) + p(y1 y20 − y10 y2 )
d
= W (y1 , y2 ) + pW (y1 , y2 ).
dt

The last line above follows from the definition of the Wronskian plus Lemma 15.9. So the above
computation yields a separable ODE in t and W (t) = W (y1 , y2 )(t):

W 0 + p(t)W = 0.

Let P (t) be some antiderivative of p(t), so P 0 = p. By Theorem 6.6, all solutions to the ODE above
have the form W (t) = Ce−P (t) , where C is a constant. Since y1 and y2 are independent, by Theorem
15.8, we must have C 6= 0. 
Corollary 15.11. Let y1 and y2 be two solutions of the second-order linear homogeneous ODE y 00 +
p(t)y 0 + q(t)y = 0, where p and q are continuous functions of t. If y1 and y2 are linearly independent,
then W (y1 , y2 )(t) 6= 0 for all t, i.e. W (y1 , y2 ) is nowhere zero.
Theorem 15.12 (General Solutions to Second-Order Linear Homogeneous ODE’s). Let y1 and y2 be two
linearly independent solutions to the second-order linear homogeneous ODE y 00 + p(t)y 0 + q(t)y = 0 where
p and q are continuous functions of t. Then all solutions to the ODE are of the form y = C1 y1 + C2 y2
for some constants C1 and C2 .
Proof. Let Y be an arbitrary solution to the ODE; we will in fact compute the constants C1 and C2
for which Y = C1 y1 + C2 y2 ! Fix any real number t0 in the domain of Y . By Corollary 15.11, we have
W (y1 , y2 )(t0 ) 6= 0. Therefore it makes sense to set:

W (Y, y2 )(t0 )
C1 =
W (y1 , y2 )(t0 )

and

W (y1 , Y )(t0 )
C2 = .
W (y1 , y2 )(t0 )
22 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

Set Z = C1 y1 +C2 y2 . Note that Z is a solution to the ODE by the Principle of Superposition Theorem
15.5. If we can show Y = Z, then we will have shown that Y is a linear combination of y1 and y2 , and
hence the proof will be complete. Let us compute Z(t0 ) and Z 0 (t0 ):

Z(t0 ) = C1 y1 (t0 ) + C2 y2 (t0 )

W (Y, y2 )(t0 ) W (y1 , Y )(t0 )
= · y1 (t0 ) + y2 (t0 )
W (y1 , y2 )(t0 ) W (y1 , y2 )(t0 )
(Y (t0 )y20 (t0 ) − Y 0 (t0 )y2 (t0 ))y1 (t0 ) − (y1 (t0 )Y 0 (t0 ) − y10 (t0 )Y (t0 ))y2 (t0 )
=
W (y1 , y2 )(t0 )
Y (t0 )(y1 (t0 )y20 (t0 ) − y10 (t0 )y2 (t0 ))
=
W (y1 , y2 )(t0 )
= Y (t0 ).

Z 0 (t0 ) = C1 y10 (t0 ) + C2 y20 (t0 )

W (Y, y2 )(t0 ) W (y1 , Y )(t0 ) 0
= · y10 (t0 ) + y (t0 )
W (y1 , y2 )(t0 ) W (y1 , y2 )(t0 ) 2
(Y (t0 )y20 (t0 ) − Y 0 (t0 )y2 (t0 ))y10 (t0 ) − (y1 (t0 )Y 0 (t0 ) − y10 (t0 )Y (t0 ))y20 (t0 )
=
W (y1 , y2 )(t0 )
Y 0 (t0 )(y1 (t0 )y20 (t0 ) − y10 (t0 )y2 (t0 ))
=
W (y1 , y2 )(t0 )
0
= Y (t0 ).

Thus we have shown that Y and Z are two solutions to the ODE which agree on the initial conditions
y(t0 ) = Y (t0 ) = Z(t0 ) and y 0 (t0 ) = Y 0 (t0 ) = Z 0 (t0 ). But by the Existence and Uniqueness Theorem
15.1, there is only one such function! Therefore Y = Z, and this completes the proof. 
The theorem above justifies the general solutions we gave in Technique 14.9, and motivates the
following definition.
Definition 15.13. A pair of functions {y1 , y2 } is called a fundamental set of solutions to a linear
homogeneous ODE y 00 + p(t)y 0 + q(t)y = 0 if y1 and y2 are both solutions and if they are linearly
independent. If {y1 , y2 } is a fundamental set of solutions, then all solutions to the ODE are linear
combinations of y1 and y2 .

16 Second Order Linear Non-Homogeneous ODE’s: The Method of Unde-

termined Coefficients
Definition 16.1. Given a second-order linear ODE y 00 + p(t)y 0 + q(t)y = f (t), the corresponding
homogeneous equation is y 00 + p(t)y 0 + q(t)y = 0.
The next theorem reduces the task of finding a general solution to a second-order linear ODE to the
task of finding a single solution, together with solving the corresponding homogeneous equation.
Theorem 16.2. If y1 and y2 are solutions to the second-order linear ODE y 00 + p(t)y 0 + q(t)y = f (t),
then
y2 = y1 + yh ,

where yh is some solution to the corresponding homogeneous equation. In other words, two solutions
differ only by a solution to the corresponding homogeneous equation.
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 23

Proof. This proof mimics exactly the proof we gave long ago in the first-order case. With y1 and y2
given, set yh = y2 − y1 . Then plug yh into the corresponding homogeneous equation and see that it is a
solution. 
Example 16.3. Consider the ODE y 00 − 4y = −3 sin t.
(1) Verify that yp (t) = 53 sin t is a solution.
(2) Find all solutions.
Solution. Checking (a) is easy, so we proceed with (b). Since we know a particular solution (from part
(a)), by the previous theorem we can find all possible solutions simply by finding all solutions to the
corresponding homogeneous equation:

y 00 − 4y = 0.

For this we consider the characteristic equation λ2 − 4 = 0. This has two distinct roots 2 and −2, so
a general solution is given by

y = C1 e2t + C2 e−2t .

Therefore all possible solutions to the original ODE have the form

y= 3
5 sin t + C1 e2t + C2 e−2t .

The next technique is much more easily understood via practice than by reading the description, but
we will attempt to describe the process in words anyway. See the examples following the technique.
Technique 16.4 (Guessing a Particular Solution to a Second-Order Linear Non-Homogeneous ODE
with Constant Coefficients). Given an ODE of the form ay 00 + by 0 + cy = f (t), we look at the function
f (t) on the right-hand side. This method is only effective if f (t) is a linear combination of functions
of the forms tm , tm eαt , tm eαt cos(βt), and tm eαt sin(βt), where m is a positive integer and α, β are real
numbers.
(1) If f is indeed a linear combination of functions of this form, then we consider the following sets
of functions:

(a) F = {1, t, t2 , ..., tm } if tm appears in f ;

(b) F = {eαt , teαt , t2 eαt , ..., tm eαt } if tm eαt appears in f ; and

(c) F = {eαt cos(βt), teαt cos(βt), ..., tm eαt cos(βt), eαt sin(βt), teαt sin(βt), ..., tm eαt sin(βt)} if
tm eαt cos(βt) or tm eαt sin(βt) appears in f .

(2) If none of the functions listed above appear as solutions to the corresponding homogeneous equa-
tion, then we guess that a particular solution yp to the equation will be a linear combination of
the functions as above.

(3) On the other hand, if some functions in F appear as solutions to the corresponding homogeneous
equation, then we multiply F by tr to obtain a new set tr F , where r is the smallest integer for
which no function in tr F is a solution to the corresponding homogeneous equation. In this case
we guess that yp is a linear combination of the functions above, but with F replaced by tr F .
Technique 16.5 (Method of Undetermined Coefficients). Given an ODE of the form ay 00 + by 0 + cy =
f (t):
(1) Solve the corresponding homogeneous equation, to obtain a general solution yh .
24 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

(2) Guess the form of a particular solution yp to the ODE, as in the previous technique.
(3) With yp written as a general linear combination (with undetermined coefficients) of functions,
compute yp0 and yp00 and plug into the ODE. Equate the coefficients of like terms, and solve for
the coefficients. This gives a particular solution yp .
(4) A general solution to the original ODE is yp + yh (where yp is a particular function, and yh
denotes infinitely many functions).
Example 16.6. Solve y 00 + 5y 0 + 6y = 2et + 4 cos t.
Solution. First we go ahead and solve the corresponding homogeneous equation:
y 00 + 5y 0 + 6y = 0.

By the usual method of considering the characteristic equation, it is easy to see that a fundamental
set of solutions to the corresponding homogeneous equation is {e2t , e3t }. So a general solution is of the
form yh = C1 e2t + C2 e3t .
We note the 2et and the 4 cos t on the right-hand side, and we guess that a particular solution yp
will be a linear combination of the functions et , cos t, and sin t. (Because none of these functions is a
solution to the corresponding homogeneous equation, there are no wrinkles here- for remarks on what
to do if one of our guesses appears as a solution to the corresponding homogeneous equation, see the
next example). In other words, we guess that

yp = Aet + B cos t + C sin t,

where A, B, and C are constants to be determined. We compute derivatives:

yp0 = Aet − B sin t + C cos t

yp00 = Aet − B cos t − C sin t

and plug into the ODE:

Aet − B cos t − C sin t + 5(Aet − B sin t + C cos t) + 6(Aet + B cos t + C sin t) = 2et + 4 cos t.

Regrouping terms, we get:

3Aet + C cos t − B sin t = 2et + 4 cos t.

We conclude that A = 32 , C = 4, and B = 0. So yp (t) = 23 et + 4 sin t is a particular solution. We

conclude that a general solution to our given ODE is any function of the form

y = 23 et + 4 sin t + C1 e2t + C2 e3t .


Example 16.7. Solve the IVP y 00 − 3y 0 = −e3t − 2t, y(0) = 0, y 0 (0) = 8
9.

Solution. First we solve the corresponding homogeneous equation y 00 − 3y 0 = 0 by considering the char-
acteristic equation λ2 − 3λ = λ(λ − 3) = 0. This has roots 0 and 3, so a fundamental set of solutions is
{e0t , e3t } = {1, e3t }, and a general solution has the form

yh = C1 + C2 e3t .
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 25

Now looking at the right-hand side of our original ODE, we are inclined to guess that a particular so-
lution yp will look like some linear combination of the functions e3t , 1, and t. However, e3t and 1 already
appear as solutions to the corresponding homogeneous equation; therefore, we multiply everything by t
(in other words by t1 , where 1 is the least power which will prevent all functions from being solutions to
the corresponding homogeneous equation). So we guess that yp is a linear combination of the functions
te3t , t, and t2 :

yp (t) = Ate3t + Bt + Ct2

for some undetermined coefficients A, B, C. Now compute derivatives:

yp0 (t) = Ae3t + 3Ate3t + B + 2Ct

yp00 (t) = 3Ae3t + 3Ae3t + 9Ate3t + 2C

= 6Ae3t + 9Ate3t + 2C.

Plugging these into the ODE we get

6Ae3t + 9Ate3t + 2C − 3(Ae3t + 3Ate3t + B + 2Ct) = −e3t − 2t.

Regrouping, we have

3Ae3t + (2C − 3B) − 6Ct = −e−3t − 2t.

It follows from the above that 3A = −1 and −6C = −2, whence A = − 31 and C = 13 . Since
2C − 3B = 0, we can also deduce that B = 29 . So a particular solution to the ODE is given by

yp = − 13 te3t + 29 t + 13 t2 .

We conclude that any solution to the ODE must have the form

y = − 13 te3t + 29 t + 13 t2 + C1 + C2 e3t .

Now it remains only to solve for the initial conditions. We first pin down C2 by considering y 0 :

y 0 (t) = −te3t + (3C 2 − 31 )e3t + 23 t + 92 .

Our initial condition gives us

8
9 = y 0 (0) = 3C2 − 1
3 + 2
9

whence C2 = 13 . Now we can use our other initial condition y(0) = 0 to find C1 :
1
0 = y(0) = C1 + C2 = C1 + 3
26 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

The above gives C1 = − 13 . So the unique solution to our IVP is

y(t) = − 31 te3t + 29 t + 31 t2 − 1
3 + 13 e3t .


17 Second-Order Linear Non-Homogeneous ODE’s: The Method of Varia-

tion of Parameters
We have seen that solving a second-order linear ODE reduces to finding a general solution to the
corresponding homogeneous equation, which is usually easy, together with finding a single particular
solution. The method of undetermined coefficients in the previous section is excellent for its simplicity,
but it only works in very special cases. We will now give a more general strategy for solving ODE’s,
which involves recalling the following fact about 2 × 2 matrices.
 
a b
Fact 17.1. Let A = be a 2 × 2 matrix. Then A is invertible if and only if det A 6= 0, and
c d
 
−1 1 d −b
A = det A .
−c a
Theorem 17.2 (Variation of Parameters for Second-Order Linear ODE’s). Consider the ODE y 00 +
p(t)y 0 + q(t)y = f (t), where p, q, and f are continuous. Let {y1 , y2 } be a fundamental set of solutions
to the corresponding homogeneous equation. Then there exist functions u1 (t), u2 (t) such that

yp (t) = u1 (t)y1 (t) + u2 (t)y2 (t)

is a particular solution to the ODE.

Proof. The strategy of this proof is to approach the equation as a 2-dimensional linear algebra prob-
lem. One beautiful feature of this proof is that the Wronskian falls out naturally from the necessary
computations.
We work backwards: Suppose there is such a solution y = u1 y1 + u2 y2 . Compute the first derivative:

y 0 = u1 y10 + u01 y1 + u2 y20 + u02 y2 .

At this moment let us make an additional assumption, which we enshrine in a box:

u01 y1 + u02 y2 = 0

If we make the above assumption, the derivative simplifies to

y 0 = u1 y10 + u2 y20 .

Now compute the second derivative:

y 00 = u1 y100 + u01 y10 + u2 y200 + u02 y20 .

Plug everything into the ODE:

u1 y100 + u01 y10 + u2 y200 + u02 y20 + p(u1 y10 + u2 y20 ) + q(u1 y1 + u2 y2 ) = f .
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 27

Regrouping terms yields:

u1 [y100 + py10 + qy1 ] + u2 [y200 + py20 + qy2 ] + u01 y10 + u02 y20 = f .

Since y1 and y2 are solutions to the corresponding homogeneous equation, the above reduces to the
following boxed equation:

u01 y10 + u02 y20 = f

With these computations completed, the question “Is the theorem true?” reduces to the question
“Do there exist differentiable functions u1 and u2 which satisfy the two boxed equations above?”, in the
sense that if the answer to the latter question is “Yes” then so is the answer to the former question.
Regard u01 and u02 as unknowns, and y1 and y2 as given- then the two boxed equations give a linear
system which is equivalent to the following matrix equation:

u01
    
y1 y2 0
= .
y10 y20 u02 f

(In the above we have omitted the argument t from each of the functions, but we are implicitly
considering the
 equation  above for all possible values of t.)
y1 y2
Let A = . Since y1 and y2 are linearly independent, the determinant det A = W (y1 , y2 )
y10 y20
is non-zero for all t, and hence the matrix A(t) is invertible for all t. Moreover, we can compute the
inverse using the familiar formula for 2 × 2-matrix inverses:

y20
 
−1 1 −y2
A = .
W (y1 , y2 ) −y10 y1

Multiplying on the left of our matrix equation by A−1 on both sides, we get

u01 y20
      
1 −y2 0 1 −y2 f
= = .
u02 W (y1 , y2 ) −y10 y1 f W (y1 , y2 ) y1 f

y2 f y1 f
In other words the equation has a solution and we get u01 = − and u02 = . The
W (y1 , y2 ) W (y1 , y2 )
functions u1 and u2 may now be obtained through integration:

−y2 f
Z Z
y1 f
u1 = dt and u2 = dt.
W (y1 , y2 ) W (y1 , y2 )

Technique 17.3 (General Method for Solving Second-Order Linear ODE’s). Given an ODE of the form
y 00 + p(t)y 0 + q(t)y = f (t):
(1) Find a fundamental set of solutions {y1 , y2 } to the corresponding homogeneous equation y 00 +
p(t)y 0 + q(t)y = 0.

(2) Compute the Wronskian W (y1 , y2 ).

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

(3) Set u01 (t) = − and u02 (t) = , and integrate to find u1 and u2 .
W (y1 , y2 )(t) W (y1 , y2 )(t)
28 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

(4) A particular solution to the ODE is given by yp = u1 y1 + u2 y2 , where u1 and u2 are chosen as
above.

(5) The general solution the the ODE is y = yp + C1 y1 + C2 y2 .

Example 17.4. Solve y 00 − 2y 0 + y = et ln t, t > 0.
Example 17.5. Solve y 00 + 4y = sec(2t), y(0) = 1, y 0 (0) = 1, − π4 < t < π
4.

18 Remarks on Higher-Order Linear ODEs

The student has probably surmised that many of the linear algebra techniques we use to solve second-
order ODE’s are likely to generalize to higher-order problems. The purpose of this section is to briefly
confirm the aforementioned thought. We omit the vast majority of the details due to a lack of sufficient
lecture time.
Theorem 18.1 (Existence and Uniqueness). Consider the n-th order linear ODE y (n) +an−1 (t)y (n−1) +
... + a2 (t)y 00 + a1 (t)y 0 (t) + a0 (t)y = f (t), where a0 , a1 , ..., an−1 , and f are all continuous functions of t.
Then there exists one and only one solution to the ODE satisfying the given initial conditions y 0 (t0 ) = y0 ,
(n−1)
y 0 (t0 ) = y00 ,..., y(n−1) (t0 ) = y0 .
Theorem 18.2 (Principle of Superposition). Let y1 , ..., yn all be solutions to the n-th order linear
homogeneous ODE y (n) + an−1 (t)y (n−1) + ... + a2 (t)y 00 + a1 (t)y 0 (t) + a0 (t)y = 0. Then any function of
the form y = C1 y1 + C2 y1 + ... + Cn yn , where C1 , ..., Cn are constants, is also a solution to the ODE.
Theorem 18.3 (General Solutions to n-th Order Linear Homogeneous ODE’s). Let y1 , ..., yn all be
solutions to the n-th order linear homogeneous ODE y (n) + an−1 (t)y (n−1) + ... + a2 (t)y 00 + a1 (t)y 0 (t) +
a0 (t)y = 0, where a0 , a1 , ..., an−1 are all continuous functions of t. Then all solutions to the ODE are of
the form y = C1 y1 + C2 y2 + ... + Cn yn for some constants C1 , ..., Cn .
Solving n-th order linear homogeneous ODE’s with constant coefficients is always theoretically possi-
ble, thanks to the following well-known theorem (which requires some complex analysis to prove).
Theorem 18.4 (Fundamental Theorem of Algebra). Every polynomial of degree n has exactly n complex
roots (counting multiplicities).
Example 18.5. Given that (λ − 1)3 (λ − 2)(λ + i)(λ − i) = λ6 − 6λ5 + 13λ4 − 16λ3 + 15λ2 − 10λ + 3,
solve y (6) − 6y (5) + 16y (4) − 16y (3) + 15y 00 − 10y 0 + 3y = 0.
Solution. The characteristic polynomial has roots 2, i, −i, and 1 with a multiplicity of 3. The 2 gives a
solution of e2t , while the complex conjugate roots ±i give solutions of cos t and sin t. The root 1 gives
the solution et , while the multiplicity of 3 for this root implies that tet and t2 et are also solutions (a
root of multiplicity 3 should give 3 distinct solutions). So a fundamental set of solutions is

{e2t , cos t, sin t, et , tet , t2 et }

and a general solution to the ODE is given by

y = C1 e2t + C2 cos t + C3 sin t + C4 et + C5 tet + C6 t2 et .

Note that the solution set to this 6-th order linear homogeneous ODE is a 6-dimensional vector space.
This nice correspondence always happens. 
Example 18.6. Solve 2y (6) − 7y (5) − 4y (4) = 0.
Theorem 18.7 (DeMoivre’s Formula). For any real numbers x and and n, we have

(cos x + i sin x)n = cos(nx) + i sin(nx).

 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 29

Proof. This follows from Euler’s formula:

(cos x + i sin x)n = (eix )n

= ei(nx)
= cos(nx) + i sin(nx).

Corollary 18.8 (Roots of Unity). For every integer n, there are exactly n distinct complex numbers
(called the n-th roots of unity) which satisfy the equation z n = 1. These roots, denoted ωk for
1 ≤ k ≤ n, are given by
   
2πk 2πk
ωk = cos + i sin .
n n

Note that the ωk ’s form the vertices of a regular n-sided polygon inscribed within the unit circle of the
complex plane. Also note that ωk = (ω1 )k for each k and that ωn = 1.
Proof. This is an easy consequence of DeMoivre’s formula (or Euler’s formula), if you note that e2πki = 1
for any integer value of k. 
Example 18.9. Solve y (8) − y = 0.
Solution. The characteristic equation is λ8 − 1 = 0, or λ8 = 1. Thus the solutions are the 8-th roots of
unity:

λ = eπi/4 , eπi/4 , e3πi/4 , eπi/2 , e5πi/4 , e3πi/2 , e7πi/4 , e2πi .

Rewriting these solutions as real numbers and complex conjugate pairs:

√ √ √ √
2 2 2 2
λ = 1, −1, ±i, 2 ± 2 i, − 2 ± 2 i.

The above roots yield the following fundamental set of solutions to the ODE:

√ √ √ √ √ √ √ √
{et , e−t , cos t, sin t, e 2t/2
cos( 2
2 t), e
2t/2
sin( 2
2 t), e
− 2t/2
cos( 2
2 t), e
− 2t/2
sin( 2
2 t)}.

A general solution is any linear combination of the eight functions above. 

(5) 000 −t
Example 18.10. Solve y + 4y = 48t − 6 − 10e .
Solution. We will solve this non-homogeneous equation using the method of undetermined coefficients,
i.e., we will guess the form of a solution and then use computations to refine our guess.
We begin by solving the corresponding homogeneous equation y (5) + 4y 000 = 0. This has characteristic
equation λ5 + 4λ3 = λ3 (λ2 + 4) = λ3 (λ + 2i)(λ − 2i) = 0, whose roots are 0 (with multiplicity 3), 2i,
and −2i. So a fundamental set of solutions to the corresponding homogeneous equation is

{1, t, t2 , cos(2t), sin(2t)}.

Now we will make an educated guess as to the form of a particular solution. Since we see an e−t on
the right-hand side of the ODE, and e−t is not a solution to the corresponding homogeneous equation,
we guess that some multiple of e−t will appear in our solution. Since we see a linear function 48t − 6 on
the right-hand side, we are inclined to guess that some linear combination of the functions t and 1 might
yield a solution; however, both of these are solutions to the corresponding homogeneous equation. So
30 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

we scale up by t3 (where the exponent 3 is the smallest choice that will prevent our guesses from being
solutions to the corresponding homogeneous equation) and guess a linear combination of t4 = t3 · t and
t3 = t4 · 1 instead. Thus we assume that we will be able to find a solution of the form

y = Ae−t + Bt4 + Ct3 .

Now compute derivatives:

y 0 = −Ae−t + 4Bt3 + 3Ct2

y 00 = Ae−t + 12Bt2 + 6Ct
y 000 = −Ae−t + 24Bt + 6C
y (4) = Ae−t + 24B
y (5) = −Ae−t

Plugging the above into the ODE, we get

−Ae−t + 4(−Ae−t + 24Bt + 6C) = −5Ae−t + 96Bt + 24C = 48t − 6 − 10e−t .

The above implies −5A = −10, 96B = 48, and 24C = −6, which in turn imply that A = 2, B = 21 ,
and C = − 41 . This gives us a particular solution to the ODE. And since we already know a fundamental
set of solutions to the corresponding homogeneous equation, we obtain a full general solution to the ODE:

y = 2e−t + 12 t4 − 14 t3 + C1 + C2 t + C3 t2 + C4 cos(2t) + C5 sin(2t).


Definition 18.11. Let y1 , y2 , ..., yn be (n − 1)-times differentiable functions, where n is some positive
integer. The Wronskian of this collection of functions is
 
y1 y2 y3 ... yn
 y10 y20 y30 ... yn0 
y100 y200 y300 yn00
 
W (y1 , y2 , ..., yn ) = det 
 ... .

 ... 
(n−1) (n−1) (n−1) (n−1)
y1 y2 y3 ... yn

Theorem 18.12. Let y1 , ..., yn all be solutions to the n-th order linear homogeneous ODE y (n) +
an−1 (t)y (n−1) + ... + a2 (t)y 00 + a1 (t)y 0 (t) + a0 (t)y = 0, where a0 , a1 , ..., an−1 are all continuous func-
tions of t. Then y1 , ..., yn are linearly dependent if and only if W (y1 , ..., yn ) = 0.
Theorem 18.13 (Variation of Parameters). Consider an n-th order linear ODE y (n) + an−1 (t)y (n−1) +
... + a2 (t)y 00 + a1 (t)y 0 + a0 (t)y = f (t). Let {y1 , y2 , ..., yn } be a fundamental set of solutions to the corre-
sponding homogeneous equation. Let A denote the n×n matrix defined in the definition of the Wronskian
above, i.e., let A be the matrix for which W (y1 , ..., yn ) = det A. For each integer k with 1 ≤ k ≤ n, let
Ak be the (n − 1) × (n − 1) matrix obtained by omitting the k-th column and the bottom row from A. Let
Wk (y1 , ..., yn ) = det Ak . Then there exist functions u1 (t), u2 (t), ..., un (t) for which

yp = u1 y1 + u2 y2 + ... + un yn

is a particular solution to the ODE. Moreover, each uk (1 ≤ k ≤ n) satisfies the equation

Wk (y1 , ..., yn ) · f
u0k = (−1)n .
W (y1 , ..., yn )
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 31

Example 18.14. Solve ty 000 +3y 00 = −t−3 , given that a fundamental set of solutions to the corresponding
homogeneous equation is {1, t, t−1 }.

19 The Laplace Transform: Definition and Basics

The central idea of the Laplace transform is this: to each function f we permanently associate a
new function F = Lf (called its Laplace transform). For “reasonable” functions f , this induces a one-
to-one correspondence, i.e. each reasonable f has a Laplace transform and that transform is uniquely
determined. The reason for making this association is that calculus operations (taking derivatives,
antiderivatives) on a function f end up corresponding to algebraic operations (adding, subtracting,
multiplying) on that function’s Laplace transform F . In other words, a “difficult” calculus problem, such
as solving a differential equation, may be reduced to an “easy” algebra problem by applying Laplace
transforms. This principle is most strongly embodied in the upcoming Theorems 20.5 and 22.1.
Definition 19.1. Let f (t) be a function defined on the interval (0, ∞). The Laplace transform of f ,
denoted Lf , is the function (of an independent variable s)
R∞
Lf (s) = 0
e−st f (t)dt,

provided that the improper integral exists.

As a shorthand notation, we will frequently refer to the Laplace transform of a function with a lower-
case letter name, by the capital letter, i.e. for f (t) we will often denote F (s) = Lf (s). We will also
sometimes use the longer notation L{f (t)}(s), when we wish to call attention to the respective arguments
t and s.
Remark 19.2. The student should recognize that the Laplace transform operator L is a “function of
functions,” i.e., L takes a function f for input and returns a function F = Lf as output. In this way, it
d
is similar to the operation of taking a derivative, since the derivative operator dt takes a function f for
0 d
input and returns a function f = dt f as output.
d
However, conceptually there is a big difference between the operators L and dt that we wish to
d
emphasize, which is this: The derivative operator dt can be regarded as depending “locally” on f , in
the sense that if you fix any particular real input t0 , the value of the derivative f 0 (t0 ) at t0 depends only
on the values of f (t) for inputs t in a very small neighborhood around t0 . (Think about the definition
and meaning of the derivative.) In other words, if you know what is happening with f “very close” to
t0 then you know the value of f 0 (t0 ).
The Laplace transform Lf is quite different- it depends “globally” on f . If you fix a particular input
s0 and wish to know the value of Lf (s0 ), then you are computing the integral of e−s0 t f (t) over the entire
interval [0, ∞), and therefore Lf (s0 ) depends not just on the value of f at inputs close to s0 , but in fact
it depends on the value of f at every possible input on the right-half line [0, ∞).
This is the reason why, when we go from a function f (t) to its Laplace transform F (s), we choose to
change the variable name from t to s- we are emphasizing the fact that any given any real number s,
the value of the Laplace transform F (s) depends on the values f (t) for all t in [0, ∞).
Remark 19.3. For most contexts in which the Laplace transform is used, one actually thinks of the
map F = Lf as a complex-valued function of a complex variable F : C → C. However, for our purposes
in this introductory course, it will suffice to regard F as a real-valued function of a real variable, i.e.
F : R → R, and we do so until further notice.
Example 19.4. Compute Lf (s), if possible, for the following functions f (t).
(1) f (t) = 1

(2) f (t) = C for constant C

(3) f (t) = eat for constant a

32 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

(4) f (t) = t

(5) f (t) = sin kt

(6) f (t) = tt

Solution. (a) If f (t) = 1, we have

Z ∞
Lf (s) = e−st · 1dt
0
 t=b
1 −st
= lim − e
b→∞ s
 t=0
1 −sb 1
= lim − e +
b→∞ s s
1
= .
s

Note that the above computation is only true for s > 0; if s ≤ 0, then the integral in the definition of
the Laplace transform diverges.
R∞ R∞
(b) If f (t) = C, then Lf (s) = 0
e−st · Cdt = C 0
e−st dt = CL{1}(s) = C
s, by our computation in
part (a).

(c) For f (t) = eat , compute:

Z ∞
Lf (s) = e−st eat dt
Z0 ∞
= e−(s−a)t dt
0
 t=b
1 −(s−a)t
= lim − e
b→∞ s−a t=0
1
= .
s−a

As in part (a), we note that the computation above is only valid for s > a.

(d) For f (t) = t, we use integration by parts:

 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 33

Z ∞
Lf (s) = e−st tdt
0
 t=b Z b
t −st 1 −st
= lim − e + e dt
b→∞ s t=0 0 s
 t=b  t=b
t −st 1 −st
= lim − e + − 2e
b→∞ s t=0 s t=0
1
=0+ 2
s
1
= 2.
s

The above holds only for s > 0.

(e) For f (t) = sin(kt), we need to use integration by parts twice. (Students may remember integrating
functions of the form eat sin(bt) as a challenging integration by parts exercise in Cal II.)

Z ∞
Lf (s) = e−st sin(kt)dt
0
t=a
k b −st
 Z
1
= lim − e−st sin(kt) + e cos(kt)dt
b→∞ s t=0 s 0
k ∞ −st
Z
=0+ e cos(kt)dt
s 0
" t=a #
k a −st
 Z
k 1 −st
= lim − e cos(kt) − e sin(kt)dt
s a→∞ s t=0 s 0
k 1 k ∞ −st
 Z 
= − e sin(kt)dt
s s s 0
2
k k
= 2 − 2 Lf (s).
s s

Taking the first and last lines above together, we write:

k2
 
k
1+ 2 Lf (s) =
s s2

and therefore

k
Lf (s) = .
s2 + k 2

(f ) If f (t) = tt , then the Laplace transform Lf (s) does

R ∞ not exist for any value of s. This is because
the definition of the Laplace transform gives Lf (s) = 0 e−st tt dt; but tt grows faster in the long term
R∞
than est does for any value of s; therefore e−st tt → ∞. It follows that the indefinite integral 0 e−st tt dt
cannot possibly converge for any value of s. 
Example 19.5. Compute L{sinh t}, where sinh t = 21 (ex − e−x ). (This computation strongly hints that
the Laplace transform is linear.)
34 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

Fact 19.6 (Laplace Transforms of Common Functions).

f (t) F (s) = Lf (s)
1
1 ,s>0
s
1
eat s−a , s > a

k
sin(kt)
s2 + k2
k
sinh(kt) = 12 (ekt − e−kt ) ,s>k
s2 − k 2
1
t ,s>0
s2

Definition 19.7. A function f of an independent variable t has exponential order b, where b is a real
number, if there exist constants C > 0 and T > 0 for which

|f (t)| ≤ Cebt for all t > T .

Theorem 19.8. If f (t) is a piecewise continuous function of exponential order b, then the Laplace
transform Lf (s) exists for s > b.

We will state the next lemma without proof, but drawing an appropriate picture lends it a lot of
credibility.
Rb Rb
Lemma 19.9. For any integrable function g(t) on any interval [a, b], | a g(t)dt| ≤ a |g(t)|dt.

Theorem 19.10. If f (t) is a piecewise continuous function of exponential order b and F = Lf , then
lim F (s) = 0.
s→∞

Proof. It suffices (by the squeeze theorem) to prove that lim |F (s)| = 0. Since f is of exponential order
s→∞
b, there are constants C and T such that |f (t)| ≤ Cebt for all t > T . Then by the previous lemma, for
all t > T we have

Z ∞
|F (s)| = e−st f (t)dt
0
Z ∞
≤ |e−st f (t)|dt
0
Z ∞
= e−st |f (t)|dt
0
Z ∞
≤ e−st Cebt dt
0
Z ∞
=C e−st ebt dt
0
= CL{ebt }(s)
1
=C· .
s−b

C
It follows then that lim |F (s)| ≤ lim = 0. This completes the proof. 
s→∞ s→∞ s − b
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 35

20 Properties of the Laplace Transform

Theorem 20.1 (Linearity of the Laplace Transform). The Laplace transform operator L is linear. That
is,

L(af + bg) = aLf + bLg

for any functions f and g and any constants a and b.

R∞
Proof. This is just because the integral operator 0 is linear as well; plug into the definition and
check. 
Example 20.2. Compute L{5 − 2e−t }.
Solution. By the linearity of L,

L{5 − 2e−t }(s) = 5L{1}(s) − 2L{e−t }(s)

1 1
=5· −2·
s s+1
3s + 5
= ,
s(s + 1)

for s > 0. 
Theorem 20.3 (Shifting Property of the Laplace Transform). Let f (t) be a function and suppose
F (s) = Lf (s) exists for s > b. Then

L{eat f (t)}(s) = F (s − a)

for s > b + a.
Proof. This is just the definition of the Laplace transform again- plug it in and see. 
Example 20.4. Compute Lf where
(1) f (t) = e−2t sin t.
(2) f (t) = 4te3t .
Solution. (a) By the shifting property,

L{e−2t sin t}(s) = L{sin t}(s + 2)

1
= .
(s + 2)2 + 1

(b) By the shifting property as well as the linearity property,

L{4te3t }(s) = 4L{e3t t}(s)

= 4L{t}(s − 3)
4
= .
(s − 3)2

36 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

Theorem 20.5 (Laplace Transform of a Derivative). Let f (t) be a differentiable function of exponential
order b and let F = Lf . Then

L(f 0 )(s) = sF (s) − f (0)

for s > b.

Proof. The proof is integration by parts (taking u = e−st and dv = f 0 (t)dt):

Z ∞
0
L{f (t)}(s) = e−st f 0 (t)dt
0
t=a
Z a
= lim e−st f (t) t=0 + se−st f (t)dt

a→∞ 0
Z a
−sa
= lim e f (a) − f (0) + s e−st f (t)dt
a→∞ 0
= 0 − f (0) + Lf (s)
= sF (s) − f (0).

The fact that the third line is equal to the fourth line, i.e. that e−sa f (a) approaches 0 as a → ∞, is
true when s > b because f is of exponential order b. 

Example 20.6. Compute L{cos(kt)}.

Solution. Note that if f (t) = sin(kt) then f 0 (t) = k cos(kt). So we apply the previous theorem (and also
use the linearity of L):

1
L{cos(kt)}(s) = L{k cos(kt)}(s)
k
1
= [sL{sin(kt)}(s) − sin(k · 0)]
k  
1 k
= ·s 2
k s + k2
s
= 2 .
s + k2

Theorem 20.7 (Derivatives of the Laplace Transform). If f (t) is a piecewise continuous function of
exponential order b, and F = Lf is its Laplace transform, then for s > b,

L{tn f (t)}(s) = (−1)n F (n) (s).

n
−st
Proof Sketch. First note that, by an easy computation, ds d
ne = (−1)n tn e−st . Now the proof consists
of the following string of equalities which turn out to be true:
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 37

dn
F (n) (s) = L(f (t))(s)
dsn Z
n ∞
d
= n e−st f (t)dt
ds 0
Z ∞ n 
d −st
= e f (t) dt
dsn
Z0 ∞
= [(−1)n tn e−st ]f (t)dt
0
Z ∞
= (−1) n
e−st tn f (t)dt
0
= (−1)n L{tn f (t)}(s).

All of the equalities between the lines above should be clear to a differential equations student, except
the equality between lines 2 and 3. The fact that the n-th derivative (with respect to s) can be pulled
inside the integral (with respect to t) without changing the value of the expression is not obvious, and
is in fact not true for all functions of two variables s and t- however, it is true for a for a special class of
functions which includes the function e−st f (t), because f (t) is of exponential order. The student may
see a proof of this fact sometime later in a good real analysis course. 

Example 20.8. Compute the Laplace transforms of

(1) t cos(2t)

(2) t2 e−3t

(3) tn
Solution. (a) Using the previous theorem,

 
1 d
L{t cos(2t)}(s) = (−1) L{cos(2t)}(s)
ds
 
d s
=−
ds s2 + 4
s + 4 − 2s2
2
=−
(s2 + 4)2
2
s −4
= 2 .
(s + 4)2

(b) Again using the previous theorem:

d2
 
L{t2 e−3t }(s) = (−1)2 L{e−3t }(s)
ds2
d2 1
=
ds2 s + 3
2
= .
(s + 3)3
38 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

dn 1 n!
(c) Note that dsn s = (−1)n sn+1 . Then:

dn
 
n n
L{t }(s) = (−1) L{1}(s)
dsn
 n 
d 1
= (−1)n
dsn s
n!
= (−1)n · (−1)n n+1
s
n!
= n+1 .
s

Example 20.9. Compute the Laplace transforms of f , f 0 , and f 00 for f (t) = (3t − 1)3 .
Fact 20.10 (More Laplace Transforms of Common Functions).
f (t) F (s) = Lf (s)
n!
tn ,s>0
sn+1
s
cos(kt)
s + k2
2
s
cosh(kt) = 21 (ekt − e−kt ) ,s>k
s2 − k 2
k
eat sin(kt)
(s − a)2 + k 2
s−a
eat cos(kt) ,s>0
(s − a)2 + k 2
k
eat sinh(kt)
(s − a)2 − k 2
s−a
eat cosh(kt) ,s>0
(s − a)2 − k 2

21 The Inverse Laplace Transform

Definition 21.1. The inverse Laplace transform of a function F (s) is the function f (t) for which
Lf = F , if such a function exists. If such a function exists, we denote it by f = L−1 F .
Example 21.2. Observe that the functions F (s) = 1, F (s) = s, F (s) = s2 , for example, have no inverse
Laplace transform. (Use Theorem 19.10.)
Example 21.3. Find the inverse Laplace transform of

1
(1) F (s) = s−6

2
(2) F (s) = s2 +4

6
(3) F (s) = s4

6
(4) F (s) = (s+2)4

Theorem 21.4 (Linearity of the Inverse Laplace Transform). The inverse Laplace transform operator
L is linear. That is,

L−1 (aF + bG) = aL−1 F + bL−1 G

 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 39

for any functions F and G and any constants a and b.

Example 21.5. Find the inverse Laplace transform of
(1) s13

7
(2) − s2 +16

1 7
(3) s3 − s2 +16

5 2
(4) s − s−10
2s−9
Example 21.6. Find the inverse Laplace transform of F (s) = s2 +25 .

Solution. We use the linearity of L−1 .

   
−1 −1 s −1 1
L F (t) = 2L (t) − 9L (t)
s2 + 25 s2 + 25
   
s 9 5
= 2L−1 (t) − L−1 (t)
s2 + 25 5 s2 + 25
9
= 2 cos(5t) − sin(5t).
5

Example 21.7. Compute L−1 { s2 +2s+5
s
}. (Hint: Rewrite the denominator in the form (s − a)2 + k 2 .)

22 Solving IVPs with the Laplace Transform

Theorem 22.1 (Laplace Transform of a Higher Derivative). Let f be an n-times differentiable function
of exponential order b, and let F = Lf . Then

L(f (n) )(s) = sn F (s) − sn−1 f (0) − ... − sf (n−2) (0) − f (n−1) (0).

Example 22.2. Compute L{sin2 (kt)}.

Solution. If f (t) = sin2 (kt), then f 0 (t) = 2k sin(kt) cos(kt) = k sin(2kt). Then it follows from Theorem
22.1 that

L{k sin(2kt)}(s) = sL{sin2 (kt)}(s) − sin(2k(0)).

So:

1
L{sin2 (kt)}(s) = L{k sin(2kt)}(s)
s
k
= L{sin(2kt)}(s)
s
k 2k
= · 2
s s + 4k 2
2k 2
= .
s(s2 + 4k 2 )

0 4t
Example 22.3. Solve the IVP y − 4y = e , y(0) = 0.
40 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

Solution. Let Y = Ly. Take the Laplace transform of both sides of the ODE:

L(y 0 − 4y) = L{e4t }

By the linearity of L and by Theorem 22.1, the above translates into the following equation:

1
(sY − y(0)) − 4Y = s−4

After simplifying and grouping like terms this gives:

1
(s − 4)Y = s−4

1
It follows that Ly = Y = (s−4)2 . Then

n o
y = L−1 Y = L−1 1
(s−4)2 = te4t

and the problem is finished. (Check that this answer is actually a solution to the IVP.) 

Example 22.4. Solve y 00 − 4y 0 = 0, y(0) = 3, y 0 (0) = 8.

Solution. Again let Y = Ly, and take the Laplace transform of both sides of the ODE:

(s2 Y − sy(0) − y 0 (0)) − 4(sY − y(0)) = 0

Plugging in initial conditions and grouping like terms, we get:

(s2 − 4s)Y − 3s + 4 = 0

and hence Y = s3s−4 3s−4

2 −4s = s(s−4) . Now one can use method of partial fractions to show that Y =
1
s + 2
s−4 .
To find y, we compute the inverse Laplace transform of Y :

y = L−1 Y
 
−1 1 2
=L +
s s−4
   
−1 1 −1 1
=L + 2L
s s−4
= 1 + 2e4t .

This completes the problem. 

Example 22.5. Solve y 00 + 2y 0 + y = 6, y(0) = 5, y 0 (0) = 10.

Example 22.6. Solve y 000 + 4y 0 = −10et , y(0) = 2, y 0 (0) = 2, y 00 (0) = −10.
Example 22.7. Solve y 00 + 3ty 0 − 6y = 3, y(0) = y 0 (0) = 0.
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 41

Solution. Note that this example is interesting because it is a second-order ODE where the coefficients
are not constant. So the methods we developed earlier in the course for second-order ODE’s are not
obviously useful. But the Laplace transform turns out to be adequate to the task.
As usual, we let Y = Ly and take Laplace transforms of both sides of the ODE. The tricky thing here
is computing L{3ty 0 }. By Theorem 20.7, we have

d
L{3ty 0 } = − L{3y 0 }
ds
d
= −3 · (sY − y(0))
ds
d
= −3 · (sY )
ds
= −3(Y + sY 0 ).

Then, applying L to both sides of the ODE, we get:

s2 Y − sy(0) − y 0 (0) − 3(Y + sY 0 ) − 6Y = 3s .

Setting y(0) = y 0 (0) = 0 and rearranging terms, we get

−3sY 0 + (s2 − 9)Y = 3s .

Thus we have reduced the problem to a first-order linear ODE, which we can solve using an integrating
factor. Rewrite the ODE in standard form as follows:

Y 0 + (− 31 s + 3s−1 )Y = − s12 .

(−(1/3)s+3s−1 )ds 2
R
Set µ(s) = e = e−(1/6)s +3 ln s
, and multiply on both sides of the equation by µ. This
gives us

d −(1/6)s2 +3 ln s 1 2
(e Y ) = − 2 e−(1/6)s +3 ln s
ds s
1 2
= − 2 e−(1/6)s · s3
s
2
= −se−(1/6)s .

It follows that

2 2 2
e−(1/6)s +3 ln s
Y = (−se(1/6)s )ds = 3e−(1/6)s + C
R

and hence

2 2
C (1/6)s2
Y = (3e−(1/6)s + C) · e(1/6)s −3 ln s
= 3
s3 + s3 e .
42 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

Now consider the expression above; we have determined that the function Y = Ly is one of an infinite
class of functions,
 which depend
 on the choice of constant C. But notice that if C is not equal to zero,
3 C (1/6)s2
then lim + 3e is either ∞ or −∞ (depending on the sign of C); this is impossible if Y is
s→∞ s3 s
the Laplace transform of y, since we must have lim Y = 0. So we assume C = 0 and hence Y = s33 . It
s→∞
follows then that

y = L−1 Y = L−1 { s33 } = 32 t2 .

This completes the problem. The reader should verify that y is actually a solution to the IVP. 

23 Laplace Transforms of Step Functions

Definition 23.1. The Heaviside function, or unit step function, denoted U(t), is the piecewise
continuous function defined as follows:

(
0 if t < 0;
U(t) =
1 if t ≥ 0.

We will frequently work with horizontal shifts of the Heaviside function, i.e. functions of the form
U(t − a) where a is a real number. Note that

(
0 if t < a;
U(t − a) =
1 if t ≥ a.

We will also find it convenient to work with sums of the form U(t−a)−U(t−b), where a < b. Note that


0 if t < a;

U(t − a) − U(t − b) = 1 if a ≤ t < b;

0 if t ≥ b.


Example 23.2. Sketch graphs of the following functions.

(1) U(t − 5)

(2) U(t − 3π) sin t

(3) [U(t − π) − U(t − 3π)] cos t

(4) [U(t − π) − U(t − 3π)] cos t + U(t − 3π)et−3π

Theorem 23.3. Suppose f (t) is a function and F (s) = Lf (s) exists for s > b. If a > 0, then

L{f (t − a)U(t − a)}(s) = e−as F (s)

and

L−1 {e−as F (s)} = f (t − a)U(t − a).

 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 43

Proof. To prove the first equality, we compute:

Z ∞
L{f (t − a)U(t − a)} = e−st f (t − a)U(t − a)dt
0
Z a Z ∞
= e−st f (t − a)U(t − a)dt + e−st f (t − a)U(t − a)dt
Z0 a Z ∞ a
−st
= e f (t − a) · 0dt + e−st f (t − a) · 1dt
0 a
Z ∞
= e−st f (t − a)dt.
a

Now use substitution rule with u = t − a:

Z ∞
L{f (t − a)U(t − a)} = e−st f (t − a)dt
Za∞
= e−s(u+a) f (u)du
0
Z ∞
= e−as e−su f (u)du
0
= e−as F (s).

The second equality in the theorem is just a re-writing of the first equality. 
Example 23.4. Find
(1) L{U(t − a)}

(2) L{(t − 3)5 U(t − 3)}

(3) L{sin(t − π6 )U(t − π6 )}

(4) L{t2 U(t − 1)}

(5) L{U(t − π) sin t}

Example 23.5. Find
−4s
(1) L−1 { e s3 }
−πs/2
(2) L−1 { es2 +16 }
Example 23.6. Solve
(
00 1 if 0 < t < π;
y + 9y =
0 if t ≥ π;

where y(0) = y 0 (0) = 0.

Solution. Let f (t) denote the piecewise continuous function on the right-hand side of the ODE. Rewrite
f as follows:
44 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

f (t) = 1 · [U(t) − U(t − π)] = 1 − U(t − π).

Now we compute the Laplace transform of f according to Theorem 23.3:

1
Lf (s) = L{1 − U(t − π)} = L{1} − L{(U )(t − π)} = s − eπs · 1s .

Now we apply L to both sides of the ODE. As usual we denote Y = Ly:

s2 Y − sy(0) − y 0 (0) + 9Y = 1
s − eπs · 1s .

By applying the initial conditions y(0) = y 0 (0) = 0 and collecting like terms, the above reduces to:

(s2 + 9)Y = 1s (1 − e−πs )

or

e−πs
Y = 1
s(s2 +9) (1 − e−πs ) = 1
s(s2 +9) − s(s2 +9) .

Now to finish the problem it suffices to compute L−1 of the above, since y = L−1 Y . Let us do the
two terms on the right-hand separately. (This is ok because L−1 is linear.)

1 1 s
The method of partial fractions reveals that s(s2 +9) = 9s − 9(s2 +9) , and hence

     
−1 1 1 −1 1 1 −1 s
L 2
= L − L
s(s + 9) 9 s 9 s2 + 9
1 1
= − cos(3t).
9 9

Applying Theorem 23.3 again, we have

e−πs
   
−1 −1 1
L =L (t − π) · U(t − π)
s(s2 + 9) s(s2 + 9)
 
1 1
= − cos(3(t − π)) U(t − π).
9 9

Combining all of the above computations, we get

y = L−1 Y = 1
9 − 1
9 cos(3t) − ( 91 − 1
9 cos(3(t − π)))U(t − π).

The above solution to the IVP may be written as a piecewise function as follows:

(
1
− 91 cos(3t) if 0 < t < π;
9
y=
− 92 cos(3t) if t ≥ π.

 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 45

24 The Convolution Theorem

Definition 24.1. Let f (t) and g(t) be functions with domain [0, ∞). The convolution of f and g,
denoted f ∗ g, is the function (of independent variable t) defined as follows:
Rt
(f ∗ g)(t) = 0
f (t − u)g(u)du,

whenever the integral exists.

Example 24.2. Let f (t) = t and g(t) = sin t for t ≥ 0. Compute f ∗ g.

Recall the following immensely important theorem from Calculus III:

Lemma 24.3 (Fubini’s Theorem). Let F (t, y) be an integrable function on the rectangular region R =
[a, b] × [c, d]. Then f is integrable, and
RR RdRb RbRd
R
F (t, y)d(t, y) = c a F (t, y)dtdy = a c F (t, y)dydt.

The student should recall that Fubini’s theorem is actually stronger than what is stated above, since
it is true not just for rectangular regions of integration but actually for much more general regions. We
will not benefit much at the moment from attempting to formulate a precise version of this statement,
but we will use the stronger version of Fubini’s theorem to prove the next fact.

Theorem 24.4 (Convolution Theorem). Suppose f (t) and g(t) are both piecewise continuous functions
of exponential order b. Let F = Lf and G = Lg. Then

L−1 {F (s)G(s)} = f ∗ g.

Proof. Write the product F (s)G(s) using the definition of the Laplace transform:

R∞ R∞
F (s)G(s) = 0
e−sx f (x)dx · 0
e−sy g(y)dy.

The above may be rewritten as an iterated integral:

R∞R∞
F (s)G(s) = 0 0
e−s(x+y) f (x)g(y)dxdy.

R∞
Now briefly consider the inner integral 0 e−s(x+y) f (x)g(y)dx in the equation above (imagining y
held constant). Applying the substitution rule with t = x + y, we get:

R∞ R∞
0
e−s(x+y) f (x)g(y)dx = y
e−st f (t − y)g(y)dx

and therefore

R∞R∞
F (s)G(s) = 0 y
e−st f (t − y)g(y)dtdy.

Note that the region of integration in the above is exactly the set {(t, y) : 0 ≤ t ≤ y, 0 ≤ y ≤ ∞},
which corresponds to the “infinite triangle” bounded below by the line y = 0 and on the left by the line
y = t in the ty-plane. We can rewrite this region of integration as {(t, y) : 0 ≤ t ≤ ∞, 0 ≤ y ≤ t}, and
use Fubini’s theorem to exchange the order of the iterated integral:
46 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

Z ∞ Z t
F (s)G(s) = e−st f (t − y)g(y)dydt
0 0
Z ∞ Z t 
−st
= e f (t − y)g(y)dy dt
Z0 ∞ 0

= e−st (f ∗ g)(t)dt
0
= L{(f ∗ g)(t)}(s).

This proves the theorem. 

Example 24.5. Verify the convolution theorem for f (t) = t and g(t) = sin t.
Corollary 24.6. The convolution operation ∗ is commutative. That is,

f ∗g =g∗f

for any two functions f and g.

Proof. Although the theorem is true in general (and can be shown via elementary means, e.g. substi-
tution rule), we will just prove the case where both the Laplace transforms F = Lf and G = Lg exist,
by using the convolution theorem. Since multiplication is commutative, in this case by the convolution
theorem we have:

f ∗ g = L−1 (F G)
= L−1 (GF )
= g ∗ f.

The next corollary is of great theoretical significance. It gives a direct method for producing a solution
to any linear ODE with constant coefficients, provided the function f (t) on the right-hand side is Laplace
transformable. By careful analysis of characteristic polynomials, one may use Duhamel’s principle to
deduce the method of variation of parameters.
Corollary 24.7 (Duhamel’s Principle). Let a0 , a1 , ..., an be constants, and let f (t) be a function of ex-
ponential order b for some b. Consider the linear IVP:

an y (n) + ... + a2 y 00 + a1 y 0 + a0 y = f (t);

y(0) = y 0 (0) = ... = y (n−1) (0) = 0.

Let p(s) be the characteristic polynomial of the above ODE, i.e.

p(s) = an sn + ... + a2 s2 + a1 s + a0 .

Then the unique solution to the IVP is given by

 
1
y = L−1 ∗ f.
p(s)
Proof. Let y be the unique solution to the IVP and let Y = Ly. Taking Laplace transforms of both
sides of the ODE and plugging in initial conditions yields:
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 47

(an sn + ... + a2 s2 + a1 s + a0 )Y = p(s)Y = Lf .

1
Therefore Y = p(s) · Lf , and hence by the convolution theorem,

y = L−1 Y
 
−1 1
=L · Lf
p(s)
 
1
= L−1 ∗ L(L−1 f )
p(s)
 
−1 1
=L ∗ f.
p(s)


25 The Gamma Function

Definition 25.1. The Gamma function, denoted Γ(x), is a complex-valued function of a complex
number x with Re(x) > 0, defined as follows:
R∞
Γ(x) = 0
e−u ux−1 du.
Example 25.2. Verify mentally that the improper integral ought to converge if x is a real number
strictly bigger than 0.
Theorem 25.3. For every complex number x with Re(x) > 0,

Γ(x + 1) = xΓ(x).

Proof. We use integration by parts on the definition of the Gamma function at x + 1:

Z ∞
Γ(x + 1) = e−u ux du
0
Z ∞
 −u x ∞
= −e u u=0 + e−u (xux−1 )du
0
Z ∞
=0+x e−u ux−1 du
0
= xΓ(x).


Corollary 25.4. The function Γ(x + 1) is an extension of the factorial function to complex inputs x
with Re(x) > −1, i.e. for every nonnegative integer n,

Γ(n + 1) = n!.

Moreover, the Gamma function is continuous, in fact infinitely differentiable, in fact infinitely complex-
differentiable.
Proof. The claim in the last sentence of the corollary is a bit beyond our scope to verify at the moment.
However, we have done enough to verify that Γ extends the factorial function. First check it for n = 0
by using the definition of the Gamma function:
48 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

Z ∞
Γ(0 + 1) = e−u u0 du
Z0 ∞
= e−u du
0
= [−e−u ]∞
u=0
= 0 − (−1)
=1
= 0!

Now use the functional equation given by Theorem 25.3 to check it for the remaining integers n:

Γ(1 + 1) = 1 · Γ(1) = 1 · 1 = 1 = 1!
Γ(2 + 1) = 2 · Γ(2) = 2 · 1 = 2 = 2!
Γ(3 + 1) = 3 · Γ(3) = 3 · 2 = 6 = 3!
Γ(4 + 1) = 4 · Γ(4) = 4 · 6 = 24 = 4!
Γ(5 + 1) = 5 · Γ(5) = 5 · 24 = 120 = 5!
...

In general if Γ(n + 1) = n!, then Γ((n + 1) + 1) = (n + 1)Γ(n + 1) = (n + 1) · n! = (n + 1)!, so the

pattern continues forever. 
It turns out that the function Γ can actually be extended even further in a natural way to all com-
plex inputs x which are not equal to a nonnegative integer 0, −1, −2, −3, ..., and in such a way that
it still satisfies the functional equation Γ(x + 1) = xΓ(x). One may check that the integral in the
definition of the Gamma function does not converge for x = 0, i.e. Γ(0) is not defined, and therefore
there is no way to extend Γ to any negative integer in such a way that Γ(x+1) = xΓ(x) for all x. (Why?)

Since the Gamma function extends the factorial function, the next theorem is an extension of the rule
n!
L{tn } = n+1 which we computed earlier.
s
Theorem 25.5. Let a be a positive real number. Then

Γ(a + 1)
L{ta } = .
sa+1
Proof. The trick here is simply to look at the definition of the Laplace transform of ta , and use substi-
tution rule with u = st, which gives t = s−1 u and dt = s−1 du. We proceed:

Z ∞
L{ta }(s) = e−st ta dt
0
Z ∞
= e−u (s−1 u)a · s−1 du
0
Z ∞
= s−a · s−1 · e−u ua du
0
Γ(a + 1)
= .
sa+1
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 49


The next theorem is a very useful computational tool about the Gamma function (and is also surpris-
ing, beautiful, and strange). Its usual proof relies on a famous theorem called Weierstrass’ factorization
theorem, which is beyond our scope.
Theorem 25.6 (Euler’s Reflection Formula). For complex numbers x which are not integers,
π
Γ(1 − x)Γ(x) = .
sin(πx)
Example 25.7. Use Euler’s reflection formula to compute:
(1) Γ( 21 )
(2) Γ(− 21 )
Solution. (1) For x = 12 , we have

 2    
1 1 1
Γ =Γ 1− Γ
2 2 2
π
=
sin(π/2)
=π

√
and therefore since Γ( 21 ) must be positive, we have Γ( 21 ) = π.

(2) As an intermediate step, note that by Theorem 25.3 we have

√
π
Γ( 23 ) = 12 Γ( 12 ) = 2 .

Now using the reflection formula,

       
1 3 3 3
Γ − Γ =Γ 1− Γ
2 2 2 2
π
=
sin(3π/2)
= −π

√
and therefore Γ(− 12 ) = −π/Γ( 32 ) = −π · √2
π
= −2 π. 
Example 25.8. Compute:
√
(1) L{ t}
(2) L{ √1t }
Solution. Using the results of the previous example, we get:

√
Γ( 23 ) π
(1) L{t1/2 } = 3/2
= 3/2
, and
s 2s
Γ( 21 )
r
π
(2) L{t−1/2 } = 1/2 = .
s s

50 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

26 Systems of First-Order Linear Homogeneous Equations

Definition 26.1. A system of n first-order linear homogeneous ODE’s in n functions is a set
of equations of the form

y10 = a11 (t)y1 + a12 (t)y2 + ... + a1n (t)yn ;

y20 = a21 (t)y1 + a22 (t)y2 + ... + a2n (t)yn ;
y30 = a31 (t)y1 + a32 (t)y2 + ... + a3n (t)yn ;
...
yn0 = an1 (t)y1 + an2 (t)y2 + ... + ann (t)yn ;

where each aij (1 ≤ i, j ≤ n) is a function of an independent variable t, and each variable y1 , ..., yn
represents a differentiable function of t.
Example 26.2. Rewrite the following ODE’s as systems of first-order ODE’s.
(1) y 00 − 5y 0 + 6y = 0

(2) y (4) + 6y (3) − 2y 00 − y 0 + 17y = 0

Definition 26.3. A vector-valued function, or Rn -valued function, is a function ~y which takes a real
number t for input and returns a vector ~y (t) in Rn for output. We will typically denote vector-valued
functions using column vector notation, so we will write
 
y1 (t)
 y2 (t) 
~y (t) = 
 ... 


yn (t)

where each y1 , ..., yn is a real-valued function.

 
y1
 y2 
If ~y = 
 ...  is a vector valued function and each y1 , ..., yn is differentiable, then we define the


yn
derivative ~y 0 of ~y to be

y10
 
 y20 
~y 0 = 
 ... .


yn0

A linear homogeneous ODE of a vector-valued function is an equation of the form

~y 0 = A~y ,

where
 
a11 (t) a12 (t) ... a1n (t)
 a21 (t) a22 (t) ... a2n (t) 
A= 
 ... 
an1 (t) an2 (t) ... ann (t)
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 51

is an n × n matrix of functions aij (t), 1 ≤ i, j ≤ n.

Fact 26.4. Every system of n linear homogeneous ODE’s in y1 , ..., yn (with constant coefficients)
 cor-
y1
 y2 
responds uniquely to a linear homogeneous ODE ~y 0 = A~y of the vector-valued function ~y =  
 ... 
yn
(where A is a matrix with real entries).

Example 26.5. Solve the system

y10 = −y1
y20 = 2y2

Solution. It is easy to see that the general solution to the system above is y1 = C1 e−t , y2 = C2 e2t , where
C1 , C2 are arbitrary constants.

Note that in this problem, y10 depends only on y1 and y20 depends only on y2 . This system corresponds
to the matrix equation

 
−1 0
~y 0 = A~y , where A =
0 2

where A is a diagonal matrix. 

Definition 26.6. A systen of n linear homogeneous ODE’s is called decoupled if each equation in the
system is of the form yi0 = aii (t)yi . Decoupled systems correspond to matrix equations ~y 0 = D~y with
diagonal matrices D.

We can solve many decoupled systems in one swift stroke.

Theorem 26.7. Let ~y ‘ = D~y be the matrix representation of a decoupled system of first-order lin-
ear ODE’s with constant coefficients. (In other words D is a diagonal matrix with real entries.) Let
λ1 , λ2 , ..., λn denote the entries along the diagonal of D, and assume that the entries are distinct real
numbers. For each 1 ≤ i ≤ n, let ~ei denote the i-th standard basis vector of Rn , i.e. ~ei has a 1 for the
i-th entry any all 0’s in the other entries. Then any vector-valued function of the form

~y = C1~e1 eλ1 t + C2 e~2 eλ2 t + ... + C3~en eλn t

is a solution to the ODE.

Proof. For each 1 ≤ i ≤ n, the i-th equation in the system is yi0 = λi yi , which has general solution
y = C1 eλi yi . So a general solution to the vector-valued ODE is

C1 eλ1 t
 
 C2 eλ2 t 
~y =  .
 ... 
λn t
Cn e

The above is just the statement of the theorem written differently. 

52 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

27 The Structure of Solutions to Systems of Linear Homogeneous Equations

From this point on we will permanently identify a given system of linear first-order homogeneous
ODE’s with its associated vector-valued ODE ~y 0 = A~y , as in the previous section.
Theorem 27.1 (Existence and Uniqueness of Solutions to Systems of Linear First-Order ODE’s). Let
A = A(t) be an n × n matrix whose entries aij (t) are functions of an independent variable t, and are
continuous on an interval I containing t0 , for 1 ≤ i, j ≤ n. let ~y0 be any vector in Rn . Consider the
IVP

~y 0 = A~y , ~y (t0 ) = ~y .

Then there is one and only one solution to the IVP on I.

Theorem 27.2. Let ~y 0 = A~y be a system of n linear first-order homogeneous ODE’s, where the entries
of A are continuous functions of an independent variable t, and let S be the set of solutions to the ODE.
Then S is an n-dimensional vector space.
Proof. Let V denote the set of all differentiable real-valued functions of t. The student may easily verify
that V together with its usual addition and scalar multiplication is a vector space, i.e., V satisfies all
the axioms given in the usual definition of a vector space.
We wish to show S is a vector space. Since S is a subset of V , it suffices to check that it is closed
under linear combinations. So let ~y1 and ~y2 be in S, i.e. let ~y1 and ~y2 be two solutions to the given
ODE, and let C1 and C2 be constants. Let ~x = C1 ~y1 + C2 ~y2 ; we wish to check that ~x is still in S. This
is immediate, since

A~x = A(C1 ~y1 + C2 ~y2 ) = C1 A~y1 + C2 A~y2 = C1 ~y1 0 + C2 ~y2 0 = ~x 0 ,

thus ~x is a solution to the ODE. So S is indeed a vector space.

We still need to verify that S is n-dimensional. To show this, we will check that S is vector-space
isomorphic to Rn . Define a map T from S to Rn by the rule

T (~y ) = ~y (0).

Note that if C1 and C2 are constants and ~y1 and ~y2 are in S, then

T (C1 ~y1 + C2 ~y2 ) = C1 ~y1 (0) + C2 ~y2 (0) = C1 T (~y1 ) + C2 T (~y2 ),

so T is a linear transformation. T is onto by the “existence” part of the Existence and Uniqueness
Theorem 27.1: if ~y0 is any vector in Rn , then there is a solution ~y in S with ~y (0) = T (~y ) = ~y0 . T is also
one-to-one by the “uniqueness” part of Theorem 27.1: if ~y0 is a vector in Rn , there is at most one ~y in
S with ~y (0) = T (~y ) = ~y0 .
We have shown T is a vector space isomorphism (in a natural way!) of the space of solutions S with
the space of initial conditions Rn . So S is n-dimensional as claimed. 
Corollary 27.3. Let ~y 0 = A~y be a system of n linear first-order homogeneous ODE’s, where the entries
of A are continuous functions of an independent variable t. Let {y1 , y2 , ..., yn } be any set of linearly
independent solutions to the ODE. Then all solutions to the ODE are of the form ~y = C1 ~y1 + ... + Cn ~yn
for some constants C1 , ..., Cn .
Proof. This is because any set of n-linearly independent solutions must be a basis for the n-dimensional
vector space of all solutions. 
Definition 27.4. Let ~y 0 = A~y be a system of n linear first-order homogeneous ODE’s, where the entries
of A are continuous functions of an independent variable t. A fundamental set of solutions to the
ODE is any set {y1 , ..., yn } of n linearly independent solutions.
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 53

28 Solving Systems With Distinct Real Eigenvalues

We need to recall the following definitions and theorems from linear algebra.
Definition 28.1. Let A be an n × n matrix. A number λ is called an eigenvalue of A if there exists a
non-zero vector ~v in Rn for which

A~v = λ~v .

Any vector ~v for which A~v = λ~v is called an eigenvector of A corresponding to the eigenvalue λ.
Fact 28.2. A number λ0 is an eigenvalue of an n × n matrix A if and only if λ0 is a root of the
characteristic polynomial det(A − λI).
Fact 28.3. A vector ~v is an eigenvector of A corresponding to the eigenvalue λ if and only if ~v is in the
null space of the matrix A − λI.
 
4 −6
Example 28.4. Find the eigenvalues and eigenvectors of A = .
3 −7
 
4−λ −6
Solution. The characteristic polynomial is det(A − λI) = det = λ2 + 3λ − 10 =
3 −7 − λ
(λ + 5)(λ − 2), so the eigenvalues of A are the roots −5 and 2.
To find an eigenvector
 corresponding
 to the eigenvalue −5, we are looking for a vector ~v in the null
9 −6 3 −2
space of A + 5I = . This matrix row-reduces to , which has the same null space.
3 −2 0 0
A vector ~v gets sent to 0 by the latter
 matrix
 ifand only if 3v1 − 2v2 = 0, if and only if v1 = 32 v2 , if and
2/3 2/3
only if ~v is in the span of . So is an eigenvector corresponding to −5 (and in fact all
1 1
other eigenvectors are a scalar multiple of it).  
3
A similar computation yields for instance (or any scalar multiple of it) as an eigenvector
1
corresponding to 2. 

Theorem 28.5 (Diagonalizing Matrices with Distinct Eigenvalues). Let A be an n × n matrix with
distinct real eigenvalues λ1 , ..., λn and corresponding non-zero eigenvectors ~v1 , ..., ~vn respectively. Let C
be the matrix
 

C =  ~v1 ... ~vn 

whose columns are the eigenvectors of A. Then C is invertible, and A = CDC −1 and D = C −1 AC,
where D is a diagonal matrix whose entries along the diagonal are λ1 , ..., λn in order.
Theorem 28.6. Let A be an n × n matrix with real entries, and suppose that A has n distinct real
eigenvalues λ1 , ..., λn . Let ~v1 , ..., ~vn be eigenvectors corresponding to λ1 , ..., λn respectively. Then a fun-
damental set of solutions to the ODE ~y 0 = A~y is

{~v1 eλ1 t , ..., ~vn eλn t }.

Proof. By Theorem 28.5, A = CDC −1 where D is a diagonal matrix whose entries along the diagonal
are λ1 , ..., λn respectively and C is a matrix whose columns are ~v1 , ..., ~vn respectively. Then we can
rewrite the ODE ~y 0 = A~y = CDC −1 ~y as

C −1 ~y 0 = DC −1 ~y .
54 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

We introduce a change of variable: let ~x = C −1 ~y . Then ~x 0 = C −1 ~y 0 , since the derivative operator is

linear. Plugging this change into the above, we get:

~x 0 = D~x.

The above represents a decoupled system, and hence by Theorem 26.7, a general solution to it is

~x = C1~e1 eλ1 t + ... + Cn~en eλn t .

Each of the terms in the linear combination on the right-hand side above are obviously linearly inde-
pendent with respect to one another. So a fundamental set of solutions to ~x 0 = D~x is

{e~1 eλ1 t , ..., e~n eλn t }.

It follows from our change of variables that for each 1 ≤ i ≤ n, the function

~y = C~x
= C e~i eλi t
 

=  ~v1 ... ~vn  e~i eλi t

= ~vi eλi t

is a solution to ~y 0 = A~y . Thus the set

{~v1 eλ1 t , ..., ~vn eλn t }

is a set of solutions. Moreover, the functions above are linearly independent since they are just the
images of the invertible matrix C of the fundamental set of solutions to ~x 0 = D~x, already a linearly
independent set. So they comprise a fundamental set of solutions and the theorem is proved. 
Example 28.7. Solve the system y10 = 4y1 − 6y2 , y20 = 3y1 − 7y2 .
 
0 4 −6
Solution. This corresponds to the matrix system ~y = A~y , where ~y = . In a previous exam-
3 −7
ple, we computed that the eigenvalues of A are

λ1 = −5, λ2 = 2,

with corresponding eigenvectors

   
2/3 3
~v1 = , ~v2 = ,
1 1

respectively. So by our previous theorem, a fundamental set of solutions to the system is exactly
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 55

(2/3)e−5t
        2t 
(2/3) 3 3e
e−5t , e2t = , .
1 1 e−5t e2t

Thus all possible solutions to the ODE are linear combinations of the above, i.e. a general solution is
   
2/3 −5t 3
~y = C1 e + C2 e2t .
1 1


29 Solving Systems with Complex Eigenvalues

Theorem 29.1. Let A be an n × n matrix with real entries, and suppose a + bi is a complex eigenvalue
of A with corresponding non-zero eigenvector ~v1 + ~v2 i, where ~v1 and ~v2 are real vectors. Then the
complex conjugate a − bi is also an eigenvalue of A, and the complex conjugate ~v1 − ~v2 i is an eigenvector
corresponding to a − bi.
Proof. Note that since ~v1 + ~v2 i is an eigenvector corresponding to a + bi, we have

A(~v1 + ~v2 i) = (a + bi)(~v1 + ~v2 i) = (a~v1 − b~v2 ) + (b~v1 + a~v2 )i.

Since A has real entries and ~v1 , ~v2 are real vectors, it follows that

A~v1 = a~v1 − b~v2 and A~v2 = b~v1 + a~v2 .

In that case,

A(~v1 − ~v2 i) = (a~v1 − b~v2 ) − (b~v1 + a~v2 )i = (a − bi)(~v1 − ~v2 i)

and therefore ~v1 − ~v2 i is an eigenvector corresponding to a − bi as claimed. 

Theorem 29.2. Let A be an n × n matrix with real entries, and suppose a + bi, a − bi are distinct
complex conjugate eigenvalues of A with corresponding complex conjugate eigenvectors ~v1 + ~v2 i, ~v1 − ~v2 i
respectively. Then two linearly independent real vector-valued solutions of the ODE ~y 0 = A~y are given by

eat (~v1 cos(bt) − ~v2 sin(bt))

and

eat (~v2 cos(bt) + ~v1 sin(bt)).

Proof. First note that the complex vector-valued function ~y1 = (~v1 + ~v2 i)e(a+bi)t is a solution to the
ODE: since ~v is an eigenvector corresponding to a + bi, we have A~y = A~v e(a+bi) = (a + bi)~v ea+bi = ~y 0 .
Now expand this solution ~y1 using Euler’s formula:

~y1 = (~v1 + ~v2 i)e(a+bi)t

= (~v1 + ~v2 i)eat (cos bt + i sin bt)
= eat (~v1 cos bt − ~v2 sin bt) + ieat (~v2 cos bt + ~v1 sin bt).

Since the ODE is linear, both the real and imaginary parts of the above have to be solutions to the
ODE as well as ~y1 . (We will not repeat the usual argument here- see Theorem 14.5 to remember how it
56 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES

goes.) This proves the theorem. (One may check that the same pair of solutions falls out when repeating
the argument with a − bi and ~v1 − ~v2 i instead of a + bi and ~v1 + ~v2 i, as we have done.) 
 
0 3 −2
Example 29.3. Find a general solution to ~y = A~y , where A = .
4 −1
Solution. First we compute eigenvalues and eigenvectors of A. The characteristic polynomial of A is

det(A − λI) = (3 − λ)(−1 − λ) − 4(−2) = λ2 − 2λ + 5.

The quadratic formula yields λ = 1 ± 2i as roots to the above, so 1 + 2i and 1 − 2i are the complex
conjugate eigenvalues of A.
To apply our previous theorem and solve the ODE, we actually only need to find an eigenvector
corresponding to 1 + 2i. So we consider the matrix

 
2 − 2i −2
A − (1 + 2i)I = .
4 −2 − 2i

   
8 −4 + 4i 2 −1 + i
The above row-reduces to , and then to . The null-space of this matrix
  4 −2 − 2i 0 0
v1
is all vectors ~v = for which 2v1 + (−1 − i)v2 = 0, i.e. for which v1 = ( 21 + 12 i)v2 . Thus an
v2
eigenvector corresponding to 1 + 2i is

     
1/2 + (1/2)i 1/2 1/2
= + i.
1 1 0

Lastly, we apply our previous theorem to determine that a pair of linearly independent solutions to
~y 0 = A~y is given by

          
t 1/2 1/2 t 1/2 1/2
e cos 2t − sin 2t , e cos 2t + sin 2t .
1 0 0 1

Since A is 2 × 2, the above is actually a complete basis for the solution space. So a general solution
has the form

         
t 1/2 1/2 t 1/2 1/2
~y = C1 e cos 2t − sin 2t + C2 e cos 2t + sin 2t .
1 0 0 1


 
1 0 0
Example 29.4. Solve ~y 0 =  0 1 −1 .
0 1 1
 
1 0 0
Solution. The characteristic polynomial of A =  0 1 −1  is det(A−λI) = (1−λ)[(1−λ)2 −1(−1)] =
0 1 1
−(λ3 − 3λ2 + 4λ − 2) = −(λ − 1)(λ − 2λ + 2). This polynomial has a real root λ1 = 1, and two complex
conjugate roots λ2 = 1 + i, λ3 = 1 − i. These are the eigenvalues of A.
 DIFFERENTIAL EQUATIONS MATH 266 SPRING 2014 (COHEN) LECTURE NOTES 57
 
0 0 0
For the real eigenvalue λ1 = i, compute an eigenvector: we have A − I =  0 0 −1 , which row-
0 1 0
   
0 1 0 v1
reduces to  0 0 1 . This matrix will send a vector ~v =  v2  to ~0 if and only if v2 = v3 = 0. So
0 0 0 v3
   
1 1
a good choice of non-zero eigenvector is, for instance, ~v1 =  0 . This yields a solution ~y1 =  0  et
0 0
to the ODE.
For the complex
 conjugate roots  1 ± i, we compute aneigenvector corresponding  to 1+ i. We have
−i 0 0 1 0 0 v1
A − (1 + i)I =  0 −i −1 , which row-reduces to  0 1 −i . For ~v =  v2  to be in the
0 1 −i 0 0 0 v3
null
 space
  of the latter
  matrix,
 we need v 1 = 0 and v 2 − iv 3 = 0, i.e. v 2 = iv 3 . Choosing v3 = 1, we get
0 0 0
 i  =  0  +  1  i as such a vector.
1 1 0          
0 0 0 0
This yields the two functions ~y2 = et  0  cos t −  1  sin t and ~y3 = et  1  cos t +  0  sin t
1 0 0 1
as solutions to the ODE. The colletion {~y1 , ~y2 , ~y3 } are linearly independent, so we have found a funda-
mental set of solutions. A general solution looks like:
           
1 0 0 0 0
~y = C1  0  et + C2 et  0  cos t −  1  sin t + C3 et  1  cos t +  0  sin t.
0 1 0 0 1

Technique 29.5 (General Strategy for Solving Linear Homogeneous Systems with Constant Coefficients). (1)
Given a system ~y 0 = A~y , find the eigenvalues λ1 , ..., λn of A, which are precisely the roots of
the characteristic polynomial det(A − λI). Assume all of the eigenvalues are distinct.
(2) For each real eigenvector λi , find a corresponding non-zero eigenvector ~vi . This may be done by
row reducing the matrix A−λi I and picking a vector ~vi from its null space. Then a corresponding
solution to the ODE is ~yi = ~vi eλi t .
(3) For each pair of complex conjugate eigenvectors λi = a + bi, λi+1 = a − bi, find a non-zero
eigenvector ~vi corresponding to a + bi. Again, this may be done by row reducing A − λi I and
picking a vector ~vi from its null space. This ~vi will be complex, so write ~vi = ~v1 + ~v2 i. Then two
solutions to the ODE are ~yi = eat (~v1 cos bt − ~v2 sin bt) and ~yi+1 = eat (~v2 cos bt + ~v1 sin bt).
(4) The solutions {~y1 , ..., ~yn } obtained in parts (2) and (3) above comprise a linearly independent
set, and thus are a fundamental set of solutions to the system. Therefore a general solution is
~y = C1 ~y1 + ... + Cn ~yn .
Remark 29.6. Although the technique above goes a long way, we have not completely solved the
problem of finding solutions to ODE’s of the form ~y 0 = A~y where A is a real matrix, for the following
reason: it is possible for A to have repeated eigenvalues λi , i.e. roots of the characteristic polynomial
det(A − λI) with multiplicity k > 1. When this happens, the situation is a bit complicated because it
breaks into two cases. In one case, the repeated eigenvalue λi actually has k many linearly independent
eigenvectors (matching its multiplicity of k). This is the easy case and we proceed to give solutions as
in the technique above. In the other case, however, it is possible for an eigenvalue λi with multiplicity k
to have strictly fewer than k linearly independent eigenvectors. In this case, a bit more work remains to
be done to find k many solutions to the ODE. Unfortunately we have run out of time for the semester!
Feel free to ask me about this if you are curious, and the details will probably be added to a future
incarnation of the notes.