MATH 219

Fall 2024
Lecture 4
Lecture notes by Özgür Kişisel

Content: Differences between linear and nonlinear equations (section 2.4).

Suggested Problems: (Boyce, Di Prima, 10th edition)
§2.4: 1, 3, 5, 6, 8, 9, 12, 14, 16, 23, 27, 30
Recall that a first order ODE is called linear if it can be written in the form
dy
+ p(t)y = q(t)
dt
for some functions p(t) and q(t). This definition does not give us much insight about
the adjective “linear”; it doesn’t tell us much about distinguishing features of linear
equations. In this lecture we will study linear equations and non-linear equations
from a more theoretical point of view. The characteristic feature of linear equations
is the principle of superposition, which we will discuss first. Afterwards, we will in-
vestigate the existence and uniqueness of solutions for initial value problems. There
are some subtle differences between linear and non-linear equations in this respect.
The possibility of finite time blow-up for non-linear equations is the most notable of
these differences. We conclude the lecture by stating some other differences between
linear and nonlinear equations.

1 Principle of superposition
The characteristic feature of linear equations is that one can combine some given
solutions in order to produce new solutions.

Definition 1.1 Suppose that f1 , · · · , fn are functions of t. A function of the form

f = c1 f1 + c2 f2 + · · · + cn fn , where c1 , · · · , cn ∈ R, is called a linear combination
of f1 , · · · , fn . If furthermore c1 + · · · + cn = 1, then f is called an affine linear
combination of f1 , · · · , fn .

1
Remark 1.1 It is useful to make a more general definition by allowing c1 , . . . , cn ∈ C,
the set of complex numbers. But it is important to note that c1 , . . . , cn must be
constants.

Proposition 1.1 Suppose that f1 , f2 , · · · , fn are solutions of the first order linear
equation y ′ + p(t)y = q(t). Then any affine linear combination of f1 , · · · , fn is also
a solution. If furthermore q(t) is the zero function, then any linear combination of
f1 , . . . , fn is a solution.

Proof: Since f1 , · · · , fn are solutions, we know that the following equations hold:

f1′ + p(t)f1 = q(t)

f2′ + p(t)f2 = q(t)
···
fn′ + p(t)fn = q(t)

Multiply the first equation by c1 , second equation by c2 , ... , the nth equation by
cn and add them up. We get

(c1 f1 + · · · + cn fn )′ + p(t)(c1 f1 + · · · + cn fn ) = (c1 + · · · + cn )q(t).

Therefore, if c1 +· · ·+cn = 1 then the right hand side becomes q(t) and consequently
f = c1 f1 + · · · + cn fn is a solution of the equation. If q(t) = 0, then the value of
c1 + · · · + cn doesn’t matter and any linear combination of f1 , · · · , fn is a solution.
□
The theorem above describes precisely in which sense combinations of solutions of
linear equations are again solutions. This is called the principle of superposition.
Example: The principle of superposition fails in general for non-linear equations.
For instance, consider the nonlinear ODE y ′ = y 2 . It is separable and we can solve
it as follows:
Z Z
dy
= dt
y2
1
− = t+c
y
1
y(t) = −
t+c

2
But a linear combination of two solutions of this form is almost never in the same
form, therefore it will almost never be a solution. For example, both −1/t and
−1/(t − 1) are solutions. But
   
1 1 (c1 + c2 )t − c1
−c1 − c2 =−
t t−1 t(t − 1)
is not a solution unless c1 = 0, c2 = 1 or c1 = 1, c2 = 0.

2 Existence and uniqueness theorems

A first order ODE typically has infinitely many solutions because of the constant
c appearing in the solution. Our experience so far is that if an additional initial
condition y(t0 ) = y0 is given, then the constant and therefore the solution is uniquely
determined. One can ask whether this is always the case. An answer to this question
is given by the existence-uniqueness theorem. Let us begin investigating this problem
starting from the case of linear equations.

Theorem 2.1 Consider the initial value problem y ′ + p(t)y = q(t) and y(t0 ) = y0
where the functions p(t), q(t) are continuous on an open interval (a, b) and t0 ∈ (a, b).
Then this initial value problem has a unique solution y(t) which is valid over (at
least) the open interval (a, b).

Remark 2.1 It is possible that a = −∞, b = +∞ or both.

Proof: We have already derived a formula for the general solution of a first order
linear equation in section 2: R
µ(t)q(t)dt
y=
µ(t)
R
where µ(t) = exp( p(t)dt). If p(t) and q(t) are both continuous on (a, b) then
they are integrable. Same holds for µ(t). Also, µ(t) ̸= 0 since it is an exponential
function; so division by µ(t) does not cause a problem. Therefore y(t) is defined
for all t ∈ (a, b). What about the constant? Suppose that F (t) is one of the
antiderivatives of µ(t)q(t). Then we can rewrite the solution as
F (t) + c
y=
µ(t)

3
If y(t0 ) = y0 then we see that c = µ(t0 )y0 −F (t0 ), therefore it is uniquely determined
by the initial condition. Hence the solution is unique. □
Now, let us look at the case of non-linear equations. An arbitrary first order equation
(linear or non-linear) can be written in the form

dy
= f (t, y)
dt
provided that we can isolate the term dy/dt in the equation by using some algebraic
operations. We will assume that this can indeed be done and work with equations
in the form above (even though the required algebraic equations may be very com-
plicated, let us not consider this problem now and concentrate on the ODE side of
the story). Notice that if the equation is linear then f (t, y) = q(t) − p(t)y for some
functions p(t) and q(t).
The first thing to think about regarding the existence-uniqueness theorem for the
non-linear case is its statement. In general, it will be impossible to decompose f (t, y)
in the form q(t) − p(t)y or in any reasonable simple form. In the linear case, the
continuity of p(t) and q(t) implies the continuity of the two variable function f (t, y).
However, it implies more than just that: Notice that ∂f /∂y = −p(t). Therefore the
continuity of p(t) in the linear case is equivalent to the continuity of ∂f /∂y. This
gives a good idea about a possible formulation in the non-linear case, which turns
out to be a correct theorem:

Theorem 2.2 Consider the initial value problem y ′ = f (t, y), y(t0 ) = y0 where
f (t, y) and ∂f /∂y are both continuous on an open rectangle (a, b) × (c, d) containing
(t0 , y0 ). Then there exists a unique solution of this initial value problem, defined
possibly for t belonging to a smaller open subinterval of (a, b) containing t0 .

Since f and ∂f /∂y are functions of two variables, it is necessary to check continuity
on an open rectangle containing the initial point rather than on an interval. An
actual rectangle is not absolutely necessary, any open set containing the initial point
would do the same job. The proof of this theorem requires a knowledge of some
important facts from mathematical analysis, and it is beyond the scope of these
lectures. The phrase “possibly smaller subinterval” at the end of the statement is
related to the notion of “finite time blow-up”. This will be explained in the next
section.

4
Example: We want to see that the condition of continuity of ∂f /∂y really plays an
important role in the theorem and cannot be entirely discarded. For this purpose,
consider the ODE
y ′ = y 1/3
together with the initial condition y(0) = 0. Then f (t, y) = y 1/3 is continuous
everywhere, however
∂f y −2/3
=
∂y 3
is not continuous on any open set containing a point from the line y = 0. This
line contains the initial point (0, 0), so the conditions of the existence-uniqueness
theorem are not satisfied here. Let us indeed see that there are multiple solutions
to this initial value problem. The equation is separable:
Z Z
dy
= dt
y 1/3
3y 2/3
= t+c
2
 3/2
2
y = ± (t + c)
3

The condition y(0) = 0 gives c = 0. But there are still at least two solutions of the
initial value problem, namely
( (
(2t/3)3/2 , t ≥ 0 −(2t/3)3/2 , t ≥ 0
y1 (t) = , y2 (t) =
0, t<0 0, t<0

Also, in the derivation we assumed that y ̸= 0. The function y3 = 0 is yet a third

solution. (In fact, literally speaking, one can construct infinitely many solutions to
this initial value problem. Can you see how?)

3 Finite time blow-up

In the previous section it was mentioned that the domain of the solution of a non-
linear initial value problem may not extend to the largest interval which is the
projection to the t-axis where the functions f and ∂f /∂y are continuous. This
phenomenon is called finite time blow-up, and it is especially annoying in any

5
theoretical or practical consideration of differential equations. Let us first see that
this can indeed happen, by looking at an example.
Example: Consider the non-linear ODE y ′ = y 2 together with the initial condition
y(0) = y0 > 0. We obtained the general solution to this ODE to be
1
y(t) = − .
t+c
Since y(0) = y0 , we get y0 = −1/c, so that c = −1/y0 . Then
1
y(t) = − .
t − 1/y0
Let us now note that the conditions of the existence-uniqueness theorem are satisfied
for this ODE, where the rectangle (a, b) × (c, d) is all of R2 . Indeed, the functions
f (t, y) = y 2 and ∂f /∂y = 2y are continuous on all of R2 . The projection of R2 to
the t-axis is all of the t-axis, namely (−∞, +∞). However, the solution is defined
only on the interval (−∞, 1/y0 ) and not on all of (−∞, ∞) (We say the solution
blows up at 1/y0 ). Notice that the place of the blow-up depends on the initial
condition, hence it can not be quickly predicted from the ODE only. One actually
has to solve the ODE in order to understand what happens.
Let us also emphasize that in this example, the appropriate domain for the solution
y(t) above is (−∞, 1/y0 ), and not R \ {1/y0 }. Even though the function y(t) is
algebraically defined on all of R \ {1/y0 }, in a real life problem there is no way for
the solution on the left of the point 1/y0 to “talk to” the part on the right. Hence,
we take the largest connected open interval containing the initial point t0 = 0 where
y(t) is defined.

4 Some other differences

There are some further differences between linear and non-linear equations, two of
which we remark below:

ˆ The general solution of a linear first order ODE can be written as a formula in
terms of the functions p(t) and q(t). This is not the case for non-linear ODE’s.
Even if one can find all solutions, not all solutions need to correspond to special
values of a constant in some general formula. It might be quite difficult to

6
 understand how the solution depends on the parameters and functions in the
original ODE.

ˆ When we solve a non-linear ODE, it is often needed to leave the final solution
in an implicit form. This is never necessary for a first order linear ODE since
the formula expresses y explicitly in terms of t.