3 Geometric interpretation of the first order ODE.

Existence and
uniqueness theorem
It was stated that our main goal for the first half of the course is to learn analytical methods of solving
ODE. Especially, of solving the first order ODE in the form

y ′ = f (x, y). (1)

Sometimes we will need the initial condition

y(x0 ) = y0 , x0 , y0 ∈ R. (2)

In this lecture, however, we will discuss the geometric interpretation of the problem (1)–(2). This
geometric interpretation allows us to use the graphical method to obtain approximate solutions to
(1)–(2). First, recall that φ is a solution to (1)–(2) on the interval I = (a, b) if φ(x0 ) = y0 , φ ∈
C (1) (I; R) (this means that φ has a continuous first derivative everywhere on I) and satisfies (1) (i.e.,
φ′ (x) ≡ f (x, φ(x)) for any x ∈ I).

Definition 1. The graph of the solution y = φ(x) to (1)–(2) is called an integral curve of equation (1).

If we change the initial conditions, we obtain different integral curves. Can we figure out an
approximate form of the integral curves of (1) without actually solving (1)? The answer is “yes,” and
it is based on the notion of the direction (or slope) field for the equation (1).
Recall from Calculus that the geometric meaning of the derivative at the point x0 is the value of
function tan of the angle at which the tangent line at x0 intersects the x axis (this is what we usually
call “slope”). Therefore, from (1) it follows that for any point (x̃, ỹ), for which the right hand side
of (1) is defined, we can calculate the slope of the graph of y(x̃) at this point, which is given simply
by f (x̃, ỹ). Now change the coordinates (x̃, ỹ) and we will find the slope of y(x̃) at another point.
These slopes can be represented graphically as small line segments, all together, these line segments
constitute the direction (slope) field of a given ODE. Now consider a curve which is tangent at each
point to the given direction field (it means this curve is tangent to the line segments at each point).
The key fact is that this curve is an integral curve of the ODE. Indeed, at each point the slope y ′ (x)
of this curve is given by construction as f (x, y), which exactly means that this function solves ODE.
Therefore, if we are able to plot the direction field, we can sketch the integral curves of our equation.
There are two main ways to plot the direction field. The first one, which is suitable for a computer,
is to divide the area in (x, y) plane into rectangular lattice, and calculate the slopes at all the points
in the vertices of this lattice.

Example 2. Consider as an example the equation

x
y′ = − .
y
The direction field, built by computer, is presented in the figure below.
MATH266: Intro to ODE by Artem Novozhilov, e-mail: artem.novozhilov@ndsu.edu. Spring 2024

1
y

y
x x

Figure 1: The direction field of the equation y ′ = −x/y. On the right two integral curves are also
shown. Note that they are tangent to the direction field at each point.

An easier way for a human being to get an idea about the direction field is to consider the isoclines
for the equation (1). The isoclines (“iso-cline” means lines of the “equal slope”) are defined by the
equations
f (x, y) = C, C = constant. (3)
If we fix C, we find an implicitly defined curve f (x, y) = C, on every point of which the direction field
is the same and has the slope C. We can plot this curve and put the line segments with the slope
C on it. For example, if we take C = 1 and get a curve f (x, y) = 1 this means that at each point
on this curve the direction field has the slope 1 (angle π/4). Now change C and find another curve
f (x, y) = C, at every point of which the slopes are again the same, but different from the previous.
Changing the constant C we can find enough isoclines so that the general pattern of the direction field
becomes obvious.
Consider the same example
x
y′ = − .
y
The family of isoclines is given by
x
− = C,
y
or, more convenient,
x
y=− ,
C
which defines the family of straight lines passing through the origin. Take C = 1, we obtain y = −x.
This means that on the line y = −x, at every point, the direction field has the slope 1 (see the left panel
on the figure below). Now consider some other values of C, for instance, C → 0, C = −1, C → ∞.
The corresponding isoclines will be x = 0, y = x, y = 0 (see the figure).

2
 C=1

C = 0.05 C = −1

C=1
C = 200
y

y
x x

Figure 2: The direction field of the equation y ′ = −x/y sketched by the isocline method. Left panel:
The isocline corresponding to the slope C = 1 is shown in the dashed red. Four isoclines and two
integral curves in the right panel are shown.

Example 3. Here is one more example:

y ′ = x2 − y.

The family of isoclines is

y = x2 − C.
In the figure below we have the direction field plotted by the computer method (on the left) and by
the isocline method (on the right).
Now we can state the geometric interpretation of the problem (1)–(2): Geometrically, to solve the
IVP (1)–(2) means to find an integral curve belonging to the direction field of the equation and passing
through the point with the coordinates (x0 , y0 ).
Now try to think about the following questions: Can two isoclines intersect? Can they be tangent
to each other? The first question is simple: if we assume that the right-hand side of the equation
(1) is continuous both in x and y, then the isoclines cannot intersect. Indeed, assume the opposite,
let two isoclines intersect with non-zero angle. This means that at the point of intersection we must
have two different slopes, which is impossible for the continuous f (x, y), hence the contradiction. The
second question is much more subtle and requires the theory which is beyond our course. Therefore,
we just state the most important facts without going into the details of justifying them.
First, let us assume that we do have two isoclines tangent to each other at some point (x0 , y0 ).
What are some not very nice consequences of this fact? Assume that our differential equation models
some real world process, and we are given the initial condition (x0 , y0 ). Since there are at least two
integral curves passing through this point, by solving the equation we will have more than one solution.
Which means, from the practical point of view, that we do not have any solutions, since we cannot
use our mathematical model as a predictive tool. Therefore, we need to identify conditions, which

3
y

y
x x

Figure 3: The direction field of the equation y ′ = x2 −y sketched by the computer method (on the left)
and by the isocline method (on the right). The green lines are numerically obtained integral curves.

would allow us to avoid such situations. These conditions are provided by the existence and uniqueness
theorem for the IVP (1)–(2), which is of paramount importance for the theory of differential equations.
Theorem 4 (The existence and uniqueness theorem). Consider problem (1)–(2) and assume that
function f (x, y) is continuous in x and continuously differentiable in y is a neighborhood of (x0 , y0 ),
including this point. Then there exists an ε > 0 such that for (x0 − ε, x0 + ε) there exists one and only
one integral curve of the equation (1) passing through the point (x0 , y0 ).
This theorem shows that, for example, the integral curves in our last example do not intersect
and are not tangent. Now consider an example, which is not covered by the existence and uniqueness
theorem.
Example 5. Consider the equation
√
y′ = y.
The direction field together with a few integral curves are shown in the figure below.
√
Here the integral curves are tangent at certain points, and the reason for this is that y does not
√ 1
have a continuous first derivative with respect to y for y = 0 (indeed, ( y)′y = 2√ y , which is undefined
for y = 0).
To prove rigorously that the integral curves do indeed tangent to one another in certain cases (the
figure cannot serve as a proof!), we solve the equation. This equation is separable, and we can solve
it as
dy
Z Z
√ = dx =⇒
y
√
2 y = x + C =⇒
x 2 x
y= +C , + C ≥ 0.
2 2

4
 y

x
√
Figure 4: The direction field of the equation y ′ = y and three integral curves.

Note that y = 0 is also a solution.

Now consider the function 
0, x<0
y(x) = x2
 , x ≥ 0.
4
This function is continuous and has a continuous first derivative (check this). It passes through the
origin, and satisfies the equation, hence it is a solution. On the other hand the function y = 0 is also
a solution passing through the origin. Hence, we have two solutions whose integral curves are tangent
at the point x = 0 (see the figure). Q: Can you construct one more function passing through the
origin and satisfying the ODE, and different from the two we discussed? Can you construct infinitely
many such functions?
The last thing to discuss here is the meaning of the constant ε in the statement of the existence
and uniqueness theorem. Its presence actually implies that the theorem is essentially local, and the
existence and uniqueness is guaranteed only for some small interval of values of x. This is indeed
important as the following example shows.
Example 6. Consider the equation
y′ = 1 + y2.
Its solution (you should verify it) is given by
y(x) = tan(x + C).
For the initial condition y(0) = 0 we find C = 0 and y(x) = tan x. Recall that if x → π/2 then
tan x → ±∞, therefore the solutions approach infinity on the finite interval of x. Again, in the
equation itself the right hand side is not only continuous, but infinitely many times differentiable for
any points (x, y), and still the actual solutions cannot be continued infinitely. This is why we need ε
in the statement of the existence and uniqueness theorem.

5
 ∗
3.1 For a mathematically inclined student
I am not planning to prove the existence and uniqueness theorem in our course (if you are interested in
the proof you can consult many of the standard ODE textbooks, for example, I especially like for the
clarity of exposition, Hirsch, M. W., & Smale, S. (1974). Differential equations, dynamical systems,
and linear algebra), since it would take too much time. Here I would like to give two very particular
examples of the existence and uniqueness proof.
First, consider again the equation
y ′ = ay,
where a some constant. We know now that the general solution to this equation is y(x) = Ceax , where
C is an arbitrary constant.

Proposition 7. There are no other solutions to y ′ = ay except for Ceax .

Proof. Indeed, assume that u(x) is also a solution and consider function

u(x)e−ax .

Its derivative is
d
(u(x)e−ax ) = u′ (x)e−ax − au(x)e−ax = au(x)e−ax − au(x)e−ax = 0,
dx
which implies that
u(x)e−ax = C,
or
u(x) = Ceax .


This proposition proves that the only solution to the IVP

y ′ = ay, y(x0 ) = y0

is
y(x) = y0 ea(x−x0 ) .
Actually, we can consider even more general case.

Proposition 8. Consider a separable ODE

y ′ = f1 (x)f2 (y)

in the rectangular Q defined by a < x < b, c < y < d. Assume that f1 and f2 are continuous in Q,
and additionally f2 (y) 6= 0 for any c < y < d. Then through every point (x0 , y0 ) ∈ Q passes one and
only one solution.

6
Proof. Assume first that there is solution φ(x) to our equation such that φ(x0 ) = y0 . It means that

dφ
(x) = f1 (x)f2 (φ(x)),
dx
or, since f2 (y) 6= 0,
dφ(x)
= f1 (x) dx.
f2 (φ(x))
Integrate the last equality from x0 to x and find
φ(x) x
dφ(ξ)
Z Z
= f1 (ξ) dξ.
φ(x0 )=y0 f2 (φ(ξ)) x0

Let F2 (y) be any antiderivative of 1/f2 (y) and F1 (x) be any antiderivative of f1 (x). Then the last
equality takes the form
F2 (φ(x)) − F2 (y0 ) = F1 (x) − F1 (x2 ).
Since F2 (y) is a monotonous function (since its derivative does not change sign), the last equality can
be solved for φ(x):

φ(x) = F2−1 F2 (y0 ) + F1 (x) − F1 (x0 ) .
Therefore, by assuming the existence of solution φ(x) we were able to present it in a concrete form,
which implies that such solution is unique: all functions in the solution are determined through the
equation itself and the initial conditions.
On the other hand it is straightforward to show that φ(x) in this form actually is a solution. For
this we need to check the initial conditions and show that that the expression for φ(x) satisfies the
equation. 

The last proposition may seem too abstract and not very informative. However, it actually shows
how non uniqueness can appear in such equations. For this it is necessary and sufficient to have some
ŷ such that f2 (ŷ) = 0 and the improper integral
Z y
dη
y0 f2 (η)

converges when y → ŷ. In this case there are infinitely many integral curves passing through some
points of Q, they all tangent to the integral curve y = ŷ. If, however, this integral diverges, then the
uniqueness is not broken.