7

The Exact Form and General

Integrating Factors

In the previous chapters, we’ve seen how separable and linear differential equations can be solved
using methods for converting them to forms that can be easily integrated. In this chapter, we will
develop a more general approach to converting a differential equation to a form (the “exact form”)
that can be integrated through a relatively straightforward procedure. We will see just what it
means for a differential equation to be in exact form and how to solve differential equations in
this form. Because it is not always obvious when a given equation is in exact form, a practical
“test for exactness” will also be developed. Finally, we will generalize the notion of integrating
factors to help us find exact forms for a variety of differential equations.
The theory and methods we will develop here are more general than those developed earlier
for separable and linear equations. In fact, the procedures developed here can be used to solve
any separable or linear differential equation (though you’ll probably prefer using the methods
developed earlier). More importantly, the methods developed in this chapter can, in theory at
least, be used to solve a great number of other first-order differential equations. As we will see
though, practical issues will reduce the applicability of these methods to a somewhat smaller (but
still significant) number of differential equations.
By the way, the theory, the computational procedures, and even the notation that we will
develop for equations in exact form are all very similar to that often developed in the later part of
many calculus courses for two-dimensional conservative vector fields. If you’ve seen that theory,
look for the parallels between it and what follows.

7.1 The Chain Rule

The exact form for a differential equation comes from one of the chain rules for differentiating
a composite function of two variables. Because of this, it may be wise to briefly review these
differentiation rules.
First, suppose φ is a differentiable function of a single variable y (so φ = φ(y) ), and that
y , itself, is a differentiable function of another variable t (so y = y(t) ). Then the composite
function φ(y(t)) is a differentiable function of t whose derivative is given by the (elementary)
chain rule
d
[φ(y(t))] = φ ′ (y(t)) y ′ (t) .
dt

129
130 The Exact Form and General Integrating Factors

A less precise (but more suggestive) description of this chain rule is

d dφ d y
[φ(y(t))] = .
dt d y dt

!◮Example 7.1: Let

y(t) = t 2 and φ(y) = sin(y) .
Then

φ(y(t)) = sin t 2 ,
and
d
d dφ d y
sin t 2 = [φ(y(t))] =
dt dt d y dt
d d  2

= [sin(y)] · t = cos(y) · 2t = cos t 2 2t .
dy dt
(In practice, of course, you probably do not explicitly write out all the steps listed above.)

Now suppose φ is a differentiable function of two variables x and y (so φ = φ(x, y) ),

while both x and y are differentiable functions of a single variable t (so x = x(t) and
y = y(t) ). Then the composite function φ(x(t), y(t)) is a differentiable function of t , and its
derivative can be computed using a chain rule typically encountered later in the study of calculus;
namely,
d ∂φ dx ∂φ d y
[φ(x(t), y(t))] = + . (7.1)
dt ∂ x dt ∂ y dt
In practice, it is usually easier to compute this derivative by simply replacing the x and y in
the formula for φ(x, y) with the corresponding formulas x(t) and y(t) , and then computing
that formula of t to compute the above derivative. Still, this chain rule (and other chain rules
involving functions of several variables) can be quite useful in more advanced applications. Our
particular interest is in the corresponding chain rule for computing
d
[φ(x, y(x))] ,
dx
which we can obtain from equation (7.1) by simply letting x = t . Then
dx dy dy
= 1 , y = y(t) = y(x) , = ,
dt dt dx
and equation (7.1) reduces to
d ∂φ ∂φ d y
[φ(x, y(x))] = + . (7.2)
dx ∂x ∂ y dx
For brevity, we will henceforth refer to this formula as chain rule (7.2) (not very original, but
better that constantly repeating “the chain rule described in equation (7.2)”).
Don’t forget the difference between d/dx and ∂/∂ x . If φ = φ(x, y) , then
dφ
= the derivative of φ(x, y) assuming x is the variable and y is a function of x .
dx

while
∂φ
= the derivative of φ(x, y) assuming x is the variable and y is a constant.
∂x
The Exact Form, Defined 131

!◮Example 7.2: Assume y is some function of x (i.e., y = y(x) ) and

φ(x, y) = y 2 + x 2 y .
Then
∂φ ∂  2  ∂φ ∂  2 
= y + x 2 y = 2x y , = y + x 2 y = 2y + x 2 ,
∂x ∂x ∂y ∂y
and, by chain rule (7.2),
d ∂φ ∂φ d y   dy
[φ(x, y(x))] = + = 2x y + 2y + x 2 .
dx ∂x ∂ y dx dx
If, for example, y = sin(x) , then the above becomes
d  
[φ(x, y(x))] = 2x sin(x) + 2 sin(x) + x 2 cos(x) .
dx
On the other hand, if y = y(x) is some unknown function, then, after replacing φ with its
formula, we simply have
d  2    dy
y + x 2 y = 2x y + 2y + x 2 . (7.3)
dx dx

In our use of chain rule (7.2), y will be an unknown function of x , and the right side of
equation (7.2) will correspond to one side of whatever differential equation is being considered.

7.2 The Exact Form, Defined

Let R be some region in the XY –plane. We will say that a first-order differential equation is in
exact form (on R ) if and only if both of the following hold:
1. The differential equation is written in the form
dy
M(x, y) + N (x, y) = 0 (7.4a)
dx
where M(x, y) and N (x, y) are known functions of x and y .
and
2. There is a differentiable function φ = φ(x, y) on R such that
∂φ ∂φ
= M(x, y) and = N (x, y) (7.4b)
∂x ∂y
at every point in R .
We will refer to the above φ(x, y) as a potential function for the differential equation.1
In practice, the region R is often either the entire XY –plane or a significant portion of it.
There are a few technical issues regarding this region, but we can deal with these issues later. In
the meantime, little harm will be done by not explicitly stating the region. Just keep in mind that
if we say an certain equation is in exact form, then it is in exact form on some region R , and
that the graph of any solution y = y(x) derived will be restricted being a curve in that region.
1 Referring to φ as a “potential function” comes from the theory of conservative vector fields. In fact, it is not
common terminology in other differential equation texts. Most other texts just refer to this function as “ φ ”.
132 The Exact Form and General Integrating Factors

!◮Example 7.3: Consider the differential equation

  dy
2x y + 2y + x 2 = 0 . (7.5)
dx

This equation is in the form

dy
M(x, y) + N (x, y) = 0
dx

with
M(x, y) = 2x y and N (x, y) = 2y + x 2 .

Moreover, if we glance back at example 7.2, we immediately see that, letting

φ(x, y) = y 2 + x 2 y ,

we have
∂φ ∂  2 
= y + x 2 y = 2x y = M(x, y)
∂x ∂x
and
∂φ ∂  2 
= y + x 2 y = 2y + x 2 = N (x, y)
∂y ∂y

everywhere in the XY –plane. So equation (7.5) is in exact form (and we can take R to be
the entire XY –plane).

In the next several sections, we will discuss how to determine when a differential equation
is in exact form, how to convert one not in exact form into exact form, how to find a potential
function, and how to solve the differential equation using a potential function. However, we
will develop the material backwards, starting with solving a differential equation given a known
potential function, and working our way towards dealing with equations not in exact form. This
ends up being the natural (and least confusing) way to develop the material.
But first, a few general comments regarding exact forms and potential functions:

1. Just being written as

dy
M(x, y) + N (x, y) = 0
dx
does not guarantee that a differential equation is in exact form; there still might not be a
φ(x, y) with
∂φ ∂φ
= M(x, y) and = N (x, y) .
∂x ∂y
For example, we will discover that there is no φ(x, y) satisfying

∂φ ∂φ
= 3y + 3y 3 and = x y2 − x .
∂x ∂y

So
  dy
3y + 3y 3 + x y 2 − x = 0
dx
is not in exact form.
The Exact Form, Defined 133

2. A single differential equation will have several potential functions. In particular, adding
any constant to a potential function yields another potential function. After all, if
∂φ0 ∂φ0
= M(x, y) and = N (x, y)
∂x ∂y

and
φ1 (x, y) = φ0 (x, y) + c
for some constant c , then, since any derivative of any constant is zero,
∂φ0 ∂φ1 ∂φ0 ∂φ1
= = M(x, y) and = = N (x, y) .
∂x ∂x ∂y ∂y

Moreover, as you will verify in exercises 7.2 and 7.3 (starting on page 154), any
differential equation that can be written in one exact form will actually have several exact
forms, all corresponding to different (but related) potential functions.
Because of the uniqueness results concerning solutions to first-order differential equa-
tions, it does not matter which of these potential functions is used to solve the equation.
Thus, in the following, we will only worry about finding a potential function for any
given differential equation.

3. Many authors refer to an equation in exact form as an exact equation. We are avoiding this
terminology because, unlike linear and separable differential equations, the “exactness”
of an equation can be destroyed by legitimate rearrangements of that equation. For
example,
dy
+ x 2 y = e2x
dx
is a linear differential equation whether it is written as above or written as
dy
= e2x − x 2 y .
dx
Consider, on the other hand,
  dy
2x y + 2y + x 2 = 0 .
dx
As we saw in example 7.3, this differential equation is in exact form (i.e., is an “exact
equation”). However, if we rewrite it as
dy 2x y
= − ,
dx 2y + x 2

we have an equation that is not in exact form (i.e., is not an “exact equation”).
134 The Exact Form and General Integrating Factors

7.3 Solving Equations in Exact Form

Using a Known Potential Function
Observe that, if φ = φ(x, y) is any sufficiently differentiable function of x and y , and
∂φ ∂φ
= M(x, y) and = N (x, y) ,
∂x ∂y

then the differential equation

dy
M(x, y) + N (x, y) = 0
dx
can be rewritten as
∂φ ∂φ d y
+ = 0 .
∂x ∂ y dx
Chain rule (7.2) then tells us that the left side of this equation is just the derivative of φ(x, y(x)) .
So this last equation can be written more concisely as
d
[φ(x, y)] = 0
dx
(with y = y(x) ). Not only is this concise, it is easily integrated:
Z Z
d
[φ(x, y)] d x = 0 dx
dx

H⇒ φ(x, y) = c .

Think about this last equation for a moment. It describes the relation between x and y = y(x)
assuming y satisfies the differential equation
dy
M(x, y) + N (x, y) = 0 .
dx
In other words, the equation φ(x, y) = c is an implicit solution to the differential equation for
which it is a potential function.
All this is important enough to be restated as a theorem.

Theorem 7.1 (importance of a potential function)

Let φ(x, y) be a potential function for a given first-order differential equation. Then that
differential equation can be written as
∂φ ∂φ d y
+ = 0 .
∂x ∂ y dx

This, in turn, can be rewritten as

d
[φ(x, y)] = 0 with y = y(x) ,
dx
which can be integrated, Z Z
d
[φ(x, y)] d x = 0 dx ,
dx
Solving Equations in Exact Form 135

to obtain the implicit solution

φ(x, y) = c
where c is an arbitrary constant.

The above theorem contains all the steps for finding an implicit solution to a differential
equation that can be put in exact form, provided you have a corresponding potential function. Of
course, you have probably already noticed that this theorem can be shortened to the following:

Corollary 7.2
Let φ(x, y) be a potential function for a given first-order differential equation. Then
φ(x, y) = c
is an implicit solution to that differential equation.

In solving at least the first few differential equations in the exercises at the end of this chapter,
you should use all the steps described in theorem 7.1 simply to reinforce your understanding of
why φ(x, y) = c . Then feel free to cut out the intermediate steps (i.e., use the corollary). And,
of course, don’t forget to see if the implicit solution can be solved for y in terms of x , yielding
an explicit solution.

!◮Example 7.4: Consider the differential equation

  dy
2x y + 2y + x 2 = 0 .
dx
From example 7.3, we know this is in exact form and has corresponding potential function
φ(x, y) = y 2 + x 2 y .
Since
∂  2  ∂  2 
y + x 2 y = 2x y and y + x 2 y = 2y + x 2
∂x ∂y
Our differential equation can be rewritten as
 
∂  2  ∂  2  dy
y + x2y + y + x2y = 0 .
∂x ∂y dx
By chain rule (7.2), this reduces to
d  2 
y + x2y = 0 with y = y(x) .
dx
Integrating this, Z Z
d  2 
y + x2y dx = 0 dx ,
dx
yields the implicit solution
y2 + x 2 y = c .
Rewriting this as
y2 + x 2 y − c = 0
and then solving for y provides the explicit solution
√
−x 2 ± x 4 + 4c
y = .
2
136 The Exact Form and General Integrating Factors

Finding the Potential Function

Let use now consider a more difficult problem: Finding a potential function, φ(x, y) , for a given
differential equation that has been written in the form
dy
M(x, y) + N (x, y) = 0 .
dx
This means finding a function φ(x, y) satisfying both
∂φ ∂φ
= M(x, y) and = N (x, y) .
∂x ∂y
A relatively straightforward procedure for finding this φ will be outlined in a moment. But first,
let us make a few observations regarding this pair of partial differential:
1. We are not assuming the given differential equation is in exact form, so there might not
be a solution to this above pair of partial differential equations. Our method will find φ
if it exists (i.e., if the equation is in exact form), and will lead to an obviously impossible
equation otherwise.
2. Because the above pair of equations are partial differential equations, we must treat x
and y as independent variables. Do not view y as a function of x in solving for φ .
3. If you are acquainted with methods for “recovering the potential for a conservative vector
field”, then you will recognize the following as one of those methods.
Now, here is the procedure:

Basic Procedure for Finding a Potential Function

Assume we have a differential equation in the form
dy
M(x, y) + N (x, y) = 0 ,
dx
For purposes of illustration, we will use the differential equation
  dy
2x y + 2 + x 2 + 4 = 0 .
dx
To find a potential function φ(x, y) for this differential equation (if it exists), do the following:

1. Identify the formulas for M(x, y) and N (x, y) , and, using these formulas, write out the
pair of partial differential equations to be solved,
∂φ ∂φ
= M(x, y) and = N (x, y) .
∂x ∂y
(In doing so, we are naively assuming the given differential equation is in exact form.)
For our example,
  dy
2x y + 2 + |x 2{z
+ 4} = 0 .
| {z } dx
M(x,y) N (x,y)

So the pair of partial differential equations to be solved is

∂φ ∂φ
= 2x y + 2 and = x2 + 4 .
∂x ∂y
Solving Equations in Exact Form 137

2. Integrate the first equation in the pair with respect to x , treating y as a constant.
Z Z
∂φ
dx = M(x, y) d x .
∂x
In computing the integral of M with respect to x , you get a “constant of integration”.
Keep in mind that this “constant” is based on y being a constant, and may thus depend
on the value of y . Consequently, this “constant of integration” is actually some yet
unknown function of y — call it h(y) .
For our example,
Z Z
∂φ
dx = [2x y + 2] d x
∂x

H⇒ φ(x, y) = x 2 y + 2x + h(y) .

(Observe that this step yields a formula for φ(x, y) involving one yet unknown function
of y only. We now just have to determine h(y) to know the formula for φ(x, y) .)
3. Replace φ in the other partial differential equation,
∂φ
= N (x, y) ,
∂y
with the formula just derived for φ(x, y) , and compute the partial derivative. Keep in
mind that, because h(y) is a function of y only,
∂
[h(y)] = h ′ (y) .
∂y
Then algebraically solve the resulting equation for h ′ (y) .
For our example:
∂φ
= N (x, y)
∂y
∂  2 
H⇒ x y + 2x + h(y) = x 2 + 4
∂y

H⇒ x 2 + h ′ (y) = x 2 + 4

H⇒ h ′ (y) = 2 .

4. Look at the formula just obtained for h ′ (y) . Because h ′ (y) is a function of y only, its
formula must involve only the variable y , no x may appear.
If the x’s do not cancel out, then we have an impossible equation. This means the
naive assumption made in the first step was wrong; the given differential equation was
not in exact form, and there is no φ(x, y) satisfying the desired pair of partial differential
equations. In this case, Stop! Go no further in this procedure!
On the other hand, if the previous step yields
h ′ (y) = a formula of y only (no x’s ) ,
then integrate both sides of this equation with respect to y to obtain h(y) . (Because
h(y) does not depend on x , the constant of integration here will truly be a constant, not
a function of the other variable.)
138 The Exact Form and General Integrating Factors

For our example, the last step yielded

h ′ (y) = 2 .

The right side does not contain x , so we can continue and integrate to obtain
Z Z
h(y) = h(y) dy = 2 dy = 2y + c1 .

(We will later see that the constant c1 is not that important.)

5. Combine the formula just obtained for h with the formula obtained for φ in step 2.
In our example, combining the results from step 2 and the last step above
yields
φ(x, y) = x 2 y + 2x + h(y)
= x 2 y + 2x + 2y + c1

where c1 is an arbitrary constant.

(Because of the arbitrary constant from the integration of h ′ (y) , the formula obtained
actually describes all possible φ’s satisfying the desired pair of partial differential equa-
tions. If you look at our discussion above, it should be clear that this formula will always
be of the form
φ(x, y) = φ0 (x, y) + c1
where φ0 (x, y) is a particular formula and c1 is an arbitrary constant. But, as noted
earlier, we only need to find one potential function φ(x, y) . So you can set c1 equal to
your favorite constant, 0 , or keep it arbitrary and see what happens.)

If the above procedure does yield a formula φ(x, y) , then it immediately follows that the
given equation is in exact form and φ(x, y) is a potential function for the given differential
equation. But don’t forget that the goal is usually to solve the given differential equation,

dy
M(x, y) + N (x, y) = 0 .
dx

The function φ = φ(x, y) is not that solution. It is a function such that the differential equation
can be rewritten as
d
[φ(x, y)] = 0 .
dx
Integrating this yields the implicit solution

φ(x, y) = c

from which, if the implicit solution is not too complicated, we can obtain an explicit solution
y = y(x) .

!◮Example 7.5: Consider solving

  dy
2x y + 2 + x 2 + 4 = 0 .
dx
Solving Equations in Exact Form 139

As just illustrated, this equation is in exact form, and has a potential function

φ(x, y) = x 2 y + 2x + 2y + c1

where c1 can be any constant. So the differential equation can be rewritten as

d  2 
x y + 2x + 2y + c1 = 0 ,
dx
which integrates to
x 2 y + 2x + 2y + c1 = c2 .
This is an implicit solution for the differential equation. Solving this for y is easy, and (after
letting c = c2 − c1 ) gives us the explicit solution
c − 2x
y = .
x2 + 2

Note that, in the last example, the constants arising from the integration of h ′ (y) and dφ/
dx
were combined at the end. It is easy to see that this will always be possible.

Other Ways to Find a Potential Function

There are other ways to find a function φ(x, y) satisfying both
∂φ ∂φ
= M(x, y) and = N (x, y) .
∂x ∂y

Two, in particular, are worth mentioning:

1. The first is the obvious modification of the one already given in which the roles of
∂φ/ ∂φ/
∂ x = M and ∂ y = N are interchanged. That is, instead of
∂φ/
integrating ∂x = M with respect to x , and then plugging the result into
∂φ/
∂y = N ,

you
∂φ/
integrate ∂y = N with respect to y , and then plug the result into ∂φ/∂ x = M .
The integration of ∂φ/∂ y = N will yield some formula involving x , y and an unknown
function of x , g(x) . Plugging this formula into ∂φ/∂ x = M should yield a ordinary
differential equation for g(x) which does not contain y . If the y’s do not cancel out,
the desired φ does not exist. Otherwise, g(x) can be obtained by integration, and then
combined with the formula just obtained for φ(x, R y) .
This
R is usually the preferred method when N (x, y) dy is much easier to compute
than M(x, y) d x . Indeed, it usually is a good idea to scan these two integrals and, if
one looks much easier to compute, compute that one, and plug the result into the partial
differential equation corresponding to the other integral. Don’t forget to check to see if
equation resulting from that just involves the appropriate variable.
2. The other method is one often attempted by beginners who do not understand why it
should not be used: First independently integrate both ∂φ/∂ x = M and ∂φ/∂ y = N .
Then stare at the two different formulas obtained for φ(x, y) (each involving a different
unknown function) and try to guess what single formula for φ(x, y) (without unknown
functions) matches the results from the two integrations.
140 The Exact Form and General Integrating Factors

Fight any temptation to take this approach. Yes, with a little luck and skill, you can
get φ this way. But it is usually more work, it doesn’t easily warn you when φ(x, y)
does not exist, and is more likely to result in errors. Why use a method involving two
two integrations and two unknown functions when you can use a method involving just
one one integration and one unknown function with a straightforward way to determine
that one function?

7.4 Testing for Exactness — Part I

The procedure just discussed for finding a potential function φ for

dy
M(x, y) + N (x, y) = 0 (7.6)
dx

does not tell us whether such a φ even exists until step 4, after a possibly tricky integration.
Fortunately, there is a simple test that that can often tell us when seeking that φ would be futile.
This test is based on the fact that, for any sufficiently differentiable φ(x, y) ,

∂ 2φ ∂ 2φ
= .
∂y ∂x ∂x ∂y

Now let R be any region in the XY –plane on which φ is sufficiently differentiable and satisfies

∂φ ∂φ
= M(x, y) and = N (x, y) .
∂x ∂y

Then, at every point in R ,

   
∂M ∂ ∂φ ∂ 2φ ∂ 2φ ∂ ∂φ ∂N
= = = = = .
∂y ∂y ∂x ∂y ∂x ∂x ∂y ∂x ∂y ∂x

So, for equation (7.6) to be in exact form over R , we must have

∂M ∂N
= (7.7)
∂y ∂x

at every point in R . If this equation does not hold, differential equation (7.6) is not in exact
form over that region — no corresponding potential function φ exists.
What we have not shown is that equality (7.7) necessarily implies that differential equation
(7.6) is in exact form. In fact, equality (7.7) does imply that, given any point in R , the equation
is in exact form over some subregion of R containing that point. Unfortunately, showing that
and describing those regions takes more development than is appropriate here. For now, let us
just say that, in practice, the equality (7.7) implies that differential equation (7.6) is “probably” in
exact form over the given region, and it is worthwhile to seek a corresponding potential function
φ via the method outlined earlier.
Let us summarize what has just been derived, accepting the term “suitably differentiable”
as simply meaning that the necessary partial derivatives can be computed:
Testing for Exactness — Part I 141

Theorem 7.3 (test for probable exactness)

Let M(x, y) and N (x, y) be two suitably differentiable functions of two variables over a region
R in the XY –plane, and consider the differential equation
dy
M(x, y) + N (x, y) = 0 .
dx

1. If
∂M ∂N
6= on R ,
∂y ∂x
then the above differential equation is not in exact form on R .
2. If
∂M ∂N
= on R ,
∂y ∂x
then the above differential equation might be in exact form on R . It is worth seeking a
corresponding potential function.

!◮Example 7.6: Consider the differential equation

  dy
3y + 3y 3 + x y 2 − x = 0 .
dx
Here
∂M ∂  
= 3y + 3y 3 = 3 + 9y 2
∂y ∂y

and
∂N ∂  2 
= x y − x = y2 − 1 .
∂x ∂x
So
∂M ∂N
6= ,
∂y ∂x
telling us that the given differential equation is not in exact form over any region.

!◮Example 7.7: Consider the differential equation

y x dy
− + 2 = 0 .
x2 +y 2 2
x + y dx
Here
y x
M(x, y) = − and N (x, y) =
x2 + y2 x2 + y2
are well defined and differentiable everywhere on the XY –plane except (x, y) = (0, 0) .
So let’s take R to be the entire XY –plane with the origin removed. Computing the partial
derivatives, we get
 
∂M ∂ y 1(x 2 + y 2 ) − y(2y) y2 − x 2
= − 2 2
= − 2 2 2
= 2 2 2
∂y ∂y x +y (x + y ) (x + y )

and  
∂N ∂ x 1(x 2 + y 2 ) − x(2x) y2 − x 2
= = = .
∂x ∂x x 2 + y2 2
(x + y )2 2 (x 2 + y 2 )2
142 The Exact Form and General Integrating Factors

So these two partial derivatives are equal throughout R , and our test for probable exactness
tells us that the given differential equation might be in exact form on R — it is worthwhile
to try to find a corresponding potential function.

For many, the test described above in theorem 7.3 will suffice. Those who wish a more
complete test should jump to section 7.7 starting on page 150 (where we will also finish solving
the differential equation in example7.7).

7.5 Exact Equations: A Summary

To review:

If you suspect that a given differential equation,

dy
M(x, y) + N (x, y) = 0 ,
dx
is in exact form, then you can quickly check for at least probable exactness by
computing ∂ M/∂ y and ∂ N/∂ x and seeing if
∂M ∂N
= .
∂y ∂x

If the two partial derivatives are equal, then follow the procedure for finding a
potential function φ(x, y) outlined on pages 136 to 138.
If that procedure is successful and yields a φ(x, y) , then finish solving the given
differential equation using the fact that the differential equation can be rewritten as
d
[φ(x, y)] = 0 ,
dx
the integration of which yields the implicit solution

φ(x, y) = 0 .

If this equation can be solved for y in terms of x , do so.

If the given differential equation is not in exact form, then there is a possibility that it can
be put into an exact form using appropriate “integrating factors”. We will discuss these next.
By the way, don’t forget that these equations may be solvable by other means. For example,
the equation used to illustrate the procedure for finding φ was a linear differential equation, and
could have been solved a bit more quickly using the methods from chapter 5.
Converting Equations to Exact Form 143

7.6 Converting Equations to Exact Form

Basic Notions
Obviously, the first step to converting a given first-order differential equation to exact form is to
get it into the form
dy
M(x, y) + N (x, y) = 0 .
dx
Then apply the test for (probable) exactness. With luck, the test result will be positive. More
likely, it will not.
To see how we might further convert our equation to exact form, it may help to recall why
we want the exact form. It is so that the left side of the differential equation can be identified as
an ordinary derivative of some formula of x and y(x) ,
d
φ(x, y(x)) .
dx
We had a similar situation with linear equations. Given a linear equation
dy
+ p(x)y = f (x) ,
dx
we found that, after multiplying it by an integrating factor µ to get
dy
µ + µpy = µ f ,
dx
we could identify the equation’s left side as a complete derivative of a formula of x and y(x) ,
namely,
d
[µ(x)y(x)] .
dx
The same idea can be applied to convert an equation not in exact form to one that is in exact
form.

!◮Example 7.8: Consider the differential equation

  dy
3y + 3y 3 + x y 2 − x = 0 .
dx
In example 7.6, we saw that this equation is not in exact form. But look what happens after
we multiply through by µ = x 2 y −2 ,
 
2 −2 3
 2  dy
x y 3y + 3y + x y − x = 0 .
dx

We get
  dy
3x 2 y −1 + 3x 2 y + x 3 − x 3 y −2 = 0
| {z } | {z } dx
Mnew (x,y) Nnew (x,y)

with
∂ Mnew ∂  2 −1 
= 3x y + 3x 2 y = −3x 2 y −2 + 3x 2 = 3x 2 − 3x 2 y −2
∂y ∂y
144 The Exact Form and General Integrating Factors

and
∂ Nnew ∂  3 
= x − x 3 y −2 = 3x 2 − 3x 2 y −2 .
∂x ∂x
So
∂ Mnew ∂ Nnew
= ,
∂y ∂x
telling us that the equation is now (probably) in exact form (over any region where y never
equals 0 ).

We will refer to any nonzero function µ = µ(x, y) as an integrating factor for a first-order
differential equation
dy
M + N = 0
dx
if and only if multiplying that equation through by µ ,
dy
µM + µN = 0 ,
dx
yields a differential equation in exact form.2 This integrating factor may be a function of x or
of y or of both x and y . Notice that, because
dy
µM + µN = 0
|{z} |{z} dx
“new” M “new” N

is exact, our test for exactness tells us that

∂ ∂
[µM] = [µN ] .
∂y ∂x

Finding Integrating Factors

General Approach
Finding an integrating factor for an equation
dy
M(x, y) + N (x, y) = 0
dx
starts with the requirement just derived,
∂ ∂
[µM] = [µN ] . (7.8)
∂y ∂x
Remember, M and N will be known formulas of x and y . So equation (7.8) is a differential
equation in which the unknown function is our integrating factor, µ . Unfortunately, it is a rather
nontrivial partial differential equation (unlike the partial differential equations in section 7.2),
and a complete discussion of how to solve it for µ is beyond the scope of any introductory
differential equations course. Fortunately, there are some common cases where this partial
differential equation reduces to an equation we can handle. The cases we will consider are where
there is an integrating factor µ that is either as function of x only, or a function of y only, or a
‘simple’ formula of x and y . In all cases, the approach is basically the same:
2 You can show that the integrating factors found for linear equations are just special cases of the integrating factors
considered here. If there is any danger of confusion, we’ll refer to the integrating factors now being discussed as
the “more general” integrating factors.
Converting Equations to Exact Form 145

1. Decide on which case you think is appropriate, and make the corresponding assumptions
on µ .

2. Expand equation (7.8) by computing out the derivatives as far as possible, taking into
account the assumptions made on µ .

3. See if the resulting equation can be solved for a µ satisfying the assumptions made. If
so, do so. If not, start over using different assumptions on µ (unless you’ve run out of
reasonable options).

We’ll illustrate the above approach in a moment. Before that, however, a few more comments
should be made:

1. Only one integrating factor is needed. So go ahead and assign convenient values to the
arbitrary constants that arise in solving for µ (just as in finding an integrating factors for
linear equations).

2. Once you have found an integrating factor µ , remember why you wanted it. Use it to
rewrite your differential equation in exact form, and then solve the differential equation
as described earlier in this chapter.

3. There are tests to determine if there are integrating factors that are functions of just x
or just y . Moreover, there are formulas for µ that can be used if any of the tests are
satisfied (see exercise 7.6 on page 156). DO NOT WASTE YOUR TIME TRYING TO
USE THESE FORMULAS! They are hard to memorize correctly and are worth learning
only if you expect to compute many, many integrating factors over a relatively short
time frame. Chances are, you won’t have that need, just a need to understand the basic
concepts and to compute the occasional integrating factor.

Now, let’s look at the common cases:

First Case: µ Being a Function of x Only

If you think the integrating factor µ could be function of x only, then assume it, and see what
happens when you compute out equation (7.8). Under this assumption,
∂µ dµ ∂µ
µ = µ(x) , = and = 0 .
∂x dx ∂y

Consequently, equation (7.8) should immediately reduce to an ordinary differential equation for
µ . Moreover, since µ is supposed to be just a function of x here, this differential equation for
µ should not contain any y’s . Thus, if the y’s do not cancel out, the assumption that µ could
be just a function of x is wrong — go to the next case. But if the y’s do cancel out, then solve
the differential equation just obtained for a µ = µ(x) .

!◮Example 7.9: Consider the differential equation

dy
1 + y3 + x y2 = 0 .
dx
Here
∂M ∂   ∂N ∂  2
= 1 + y 3 = 3y 2 and = x y = y2 .
∂y ∂y ∂x ∂x
146 The Exact Form and General Integrating Factors

∂ M/
So ∂y 6= ∂ N/∂ x . The equation is not exact, but, with luck, we can find a function µ so that
    dy
µ 1 + y3 + µ x y2 = 0
dx

is exact. This integrating factor µ must satisfy

∂  
∂  

µ 1 + y3 = µ x y2 .
∂y ∂x

Let’s see if µ could be a function of just x . Assume µ = µ(x) . Then

∂µ dµ ∂µ
= and = 0 .
∂x dx ∂y

Using this, we have

∂  
∂  

µ 1 + y3 = µ x y2
∂y ∂x
∂µ   ∂   ∂µ  2  ∂  2
H⇒ 1 + y3 + µ 1 + y3 = xy + µ xy
∂y ∂y ∂x ∂x
    dµ  2   
H⇒ 0 · 1 + y 3 + µ 0 + 3y 2 = x y + µ y2
dx
dµ
H⇒ 3y 2 µ = x y 2 + y2µ .
dx

The y ’s do cancel out, leaving us with

dµ
3µ = x + µ ,
dx

which simplies to
dµ 2µ
= .
dx x
So there is an integrating factor µ that is a function of x alone. Moreover, to find such an
integrating factor, we need only find one (nonzero) solution to the above simple, separable
ordinary differential equation. Proceding to do so, we get
Z Z
1 dµ 2
dx = dx
µ dx x

H⇒ ln |µ| = 2 ln |x| + c = ln |x|2 + c

2
H⇒ µ = ±eln x +c
= Ax 2 .

Since only one nonzero integrating factor is needed, we can take A = 1 , giving us

µ(x) = x 2

as our integrating factor.

Unfortunately, there is always the possibility that the y’s will not cancel out.
Converting Equations to Exact Form 147

!◮Example 7.10: It is easily verified that

  dy
6x y + 5 x 2 + y = 0
dx
is not in exact form. To be a corresponding integrating factor, µ must satisfy
∂
∂  

µ[6x y] = µ5 x 2 + y .
∂y ∂x
Assume µ is a function of x only. Then
∂µ dµ ∂µ
= and = 0 ,
∂x dx ∂y
and, thus,
∂
∂  

µ[6x y] = µ5 x 2 + y
∂y ∂x
∂µ ∂ ∂µ  2  ∂  

H⇒ [6x y] + µ [6x y] = 5 x +y + µ 5 x2 + y
∂y ∂y ∂x ∂x
dµ  2 
H⇒ 0 · 6x y + µ[6x] = 5 x + 5 + µ[10x]
dx
dµ 10xµ − 6xµ 4xµ
H⇒ = 2
= .
dx 4x + 5y 4x 2 + 5y
Here, the y ’s do not cancel out, as they should if our assumption that µ depended only on
x were true. Hence, that assumption was wrong. This equation does not have an integrating
factor that is a function of x only. We will have to try something else.

Second Case: µ Being a Function of y Only

This is just like the first case, but with the roles of x and y switched. If you think the integrating
factor µ could be function of y only, then assume it, and see what happens when you compute
out equation (7.8). Under this assumption, µ = µ(y) ,
∂µ ∂µ dµ
= 0 and = .
∂x ∂y dy
Again, equation (7.8) should immediately reduce to an ordinary differential equation for µ , only
this time the differential equation for µ should not contain any x’s . If the x’s do not cancel
out, our assumption that µ could be just a function of y is wrong and we can go no further with
this hope. But if the x’s do cancel out, then solving the differential equation just obtained for a
µ = µ(y) yields the desired integrating factor.

!◮Example 7.11: As just seen in example 7.10,

  dy
6x y + 5 x 2 + y = 0
dx
does not have an integrating factor depending only on x . So instead, let’s try to find one that
depends on just y . Assuming this,
∂µ ∂µ dµ
µ = µ(y) , = 0 and = .
∂x ∂y dy
148 The Exact Form and General Integrating Factors

Combining this with equation (7.8):

∂
∂  

µ[6x y] = µ 5x 2 + 5y
∂y ∂x
∂µ ∂   ∂µ  2  ∂  2 
H⇒ [6x y] + µ 6x y = 5x + 5y + µ 5x + 5y
∂y ∂y ∂x ∂x
dµ  
H⇒ [6x y] + µ[6x] = 0 · 5x 2 + 5y + µ[10x]
dy
dµ 10xµ − 6xµ 2µ
H⇒ = = .
dx 6x y 3y

The x ’s cancel out, as hoped, and we have a simple differential equation for µ = µ(y) .
Solving it:
Z Z
1 dµ 2
dy = dy
µ dy 3y
2 2
H⇒ ln |µ| = ln |y| + c = ln |y| /3 + c
3
2/
H⇒ µ = Ay 3 .

Taking A = 1 gives the integrating factor

2/
µ(y) = y 3 .

Third Case: µ Being a ‘Simple’ Function of Both Variables

Of course, it is quite possible that no function of just x or just y will be an integrating factor
for a given equation
dy
M(x, y) + N (x, y) = 0 .
dx
In this case, the best we can usually hope for is that there is a relatively simple function of x
and y that will work as an integrating factor. This means making a ‘good guess’ at µ(x, y) ,
and verifying that it satisfies equation (7.8). One ‘guess’ sometimes worth trying is

µ(x, y) = x α y β

where the exponents, α and β , are constants to be determined. To determine the values of
the constants so that the guess works (if such constants exist), just plug this formula for µ
into equation (7.8) and see if it reduces to an equation that can be solved for α and β . If so,
find those values and use µ(x, y) = x α y β (with the values just found for α and β ) as your
integrating factor. If not, keep searching (or consider dealing with the differential equation using
the graphical and numerical methods we’ll discuss in chapter 8).

!◮Example 7.12: Consider the differential equation from example 7.6,

  dy
3y + 3y 3 + x y 2 − x = 0 .
dx
Converting Equations to Exact Form 149

Plugging
µ(x, y) = x α y β
into equation (7.8) for our differential equation yields:
∂ ∂
(µM) = (µN )
∂y ∂x
∂  
∂  

H⇒ x α y β 3y + 3y 3 = x α yβ x y2 − x
∂y ∂x
∂
∂

H⇒ 3x α y β+1 + 3x α y β+3 = x α+1 y β+2 − x α+1 y β
∂y ∂x

H⇒ 3(β + 1)x α y β + 3(β + 3)x α y β+2 = (α + 1)x α y β+2 − (α + 1)x α y β .

Combining like terms then gives

[3β + α + 4]x α y β + [3β − α + 8]x α y β+2 = 0 ,

which, in turn, holds if and only if

3β + α + 4 = 0 and 3β − α + 8 = 0 .

This last pair of equations constitute a simple system of linear equations,

3β + α + 4 = 0
3β − α + 8 = 0

which can be easily solved by any of a number of ways, yielding

α=2 and β = −2 .

Thus, the differential equation we started with,

  dy
3y + 3y 3 + x y 2 − x = 0 ,
dx

does have an integrating factor of the form µ(x, y) = x α y β , and it is

µ(x, y) = x 2 y −2

(just as was used in example 7.8 on page 143).

150 The Exact Form and General Integrating Factors

∗
7.7 Testing for Exactness — Part II
Simple Connectivity and the Complete Test for Exactness
A more complete test for exactness than given in theorem 7.3 can be described if we are more
careful about describing our situation. So suppose we have an equation
dy
M(x, y) + N (x, y) = 0 ,
dx
and that, on some open region R of the XY –plane, all of the following hold:
∂ M/ ∂ N/
1. The functions M(x, y) and N (x, y) , along with the derivatives ∂y and ∂x , are
continuous everywhere in R .

2. At each point (x, y) in R ,

∂M ∂N
= .
∂y ∂x
That region R will said to be simply connected if each and every simple closed curve (i.e.,
loop) in R encloses only points in R . If any simple closed curve in R encloses any point not
in R , then we will say that R is not simply connected. If you think about it, you will realize
that saying a region is simply connected is just a precise way of saying that the region has no
“holes”. And if you think a little more about the situation, you will realize that, if our open region
R has “holes” (i.e., is not simply connected), then it is probably because these are points where
M(x, y) or N (x, y) or their partial derivatives fail to exist.

!◮Example 7.13: Again, consider the differential equation

y x dy
− + 2 = 0 .
x2 +y 2 2
x + y dx

As noted in example 7.7,

y x
M(x, y) = − and N (x, y) =
x 2 + y2 x 2 + y2

are well defined and differentiable and satisfy

∂M ∂N
=
∂y ∂x

everywhere in the region R consisting of the XY –plane with the origin removed. Removing
this point (the origin) creates a “hole” in R . This point is also a point not in R but which is
enclosed by any loop in R around the origin.

Now we can state the full test for exactness. (Its proof will be briefly discussed at the end
of this section.)

∗ For the interested reader.

Testing for Exactness — Part II 151

Theorem 7.4 (complete test for exactness)

Let R be a simply-connected open region in the XY –plane, and let M(x, y) and N (x, y) be
two continuous functions on R whose partial derivatives are also continuous on R . Then
dy
M(x, y) + N (x, y) = 0
dx
is in exact form on R if and only if
∂M ∂N
=
∂y ∂x
at every point in R .

This theorem assures us that, if our region R is simply connected, then we can (in theory at
least) use the procedure outlined on pages 136 to 138 to find the corresponding potential function
φ(x, y) on R , and from that, derive an implicit solution φ(x, y) = c to our differential equation.
Theorem 7.4 does not definitely say the differential equation is not in exact form if R is
not simply connected. Whether the equation is or is not be in exact form over all of R is still
uncertain. What is certain, however, is the following immediate consequence of theorem 7.4.

Corollary 7.5
Assume M(x, y) and N (x, y) are two continuous functions on some open region R of the
XY –plane. Assume further that, on R , the partial derivatives of M and N are continuous and
satisfy
∂M ∂N
= .
∂y ∂x
Then
dy
M(x, y) + N (x, y) = 0
dx
is in exact form on each simply-connected open subregion of R .

Thus, even if our original region is not simply connected, we can at least pick any open,
simply connected subregion R1 , and (in theory at least) use the procedure outlined in section
7.3 to find a corresponding potential function φ1 (x, y) on R1 , and from that, derive the implicit
solution φ1 (x, y) = c to our differential equation, valid on subregion R1 .
But then, why might there not be a potential function φ(x, y) valid on the entire region R ?
Let’s go back to an example started earlier to see.

!◮Example 7.14: Let us continue our consideration of the differential equation

y x dy
− + 2 = 0
x2 +y 2 2
x + y dx

from example 7.7. As just noted in the previous example, the region R consisting of all the
XY –plane except for the origin (0, 0) is not simply connected. But we can partition it into
the left and right half-planes

R+ = {(x, y) : x > 0} and R− = {(x, y) : x < 0} ,

which are simply connected. Theorem 7.4 assures us that our differential equation is in exact
form over each of these half-planes, and, indeed, you can easy show that all the corresponding
152 The Exact Form and General Integrating Factors

potential functions on these regions for our differential equation are given by
y
φ+ (x, y) = Arctan + c+ on R+
x
and y
φ− (x, y) = Arctan + c− on R−
x
where c+ and c− are arbitrary constants.
But could there be a potential function φ on all of R corresponding to our differential
equation? If so, then φ would also be a potential function over the left and right half-planes
R+ and R− , and, as just noted, there would be constants c+ and c− such that
y
φ(x, y) = Arctan + c+ for x > 0
x
and y
φ(x, y) = Arctan + c− for x <0 .
x
Since φ must be continuous everywhere except at the origin, it must, in particular, be con-
tinuous at any point on the positive Y –axis, (0, y) with y > 0 . So, letting x → 0 from the
positive side, we have
y
φ(0, y) = lim+ φ(x, y) = lim+ Arctan + c+ .
x→0 x→0 x
Using the substitution t = y/x and recalling the limiting values of the Arctangent function,
we can rewrite the above as
π
φ(0, y) = lim Arctan(t) + c+ = + c+ .
t→+∞ 2
Likewise, letting x → 0 from the negative side, we have
φ(0, y) = lim− φ(x, y)
x→0
y
= lim− Arctan + c−
x→0 x
π
= lim Arctan(t) + c− = − + c− .
t→−∞ 2
Together, the above tells us that
π π
− + c− = φ(0, 1) = + c+ ,
2 2
which, of course, means that c− = π + c+ . Because of the arbitrariness of the constants
added to potential functions, we may, for simplicity, let c+ = 0 . Then
  
y


 Arctan if x > 0

  x
y
φ(x, y) = Arctan + π if x < 0 .

 x

 π
 if x = 0 and y > 0
2
But look at what must now happen at a point on the negative Y –axis, say, at (0, −1) .
 
−1 π
lim+ φ(x, −1) = lim+ Arctan = lim Arctan(t) = −
x→0 x→0 x t→−∞ 2
Testing for Exactness — Part II 153

and  
−1 3π
lim− φ(x, −1) = lim− Arctan + π = lim Arctan(t) + π = .
x→0 x→0 x t→+∞ 2

So
lim φ(x, −1) 6= lim− φ(x, −1) .
x→0+ x→0

Thus, there are points in R at which φ(x, y) is not continuous, contrary to the fact that a
potential function on R must be continuous everywhere on R . And thus, the answer to our
question “Could there be a potential function φ on all of R corresponding to our differential
equation?” is “No”.

The above example illustrates that, while we can partition a non-simply connected region
into simply-connected subregions and then find all possible potential functions for our differential
equation over each subregion, it may still be impossible to “paste together” these regions and
potential functions to obtain a potential function that is well defined across all the boundaries
between the partitions.
Is this truly a problem for us, whose main interest is in solving the differential equation?
Not really. We can still solve the given differential equation. All we need to do is to choose our
simply-connected partitions reasonably intelligently.

!◮Example 7.15: Let’s solve the initial-value problem

y x dy
− + 2 = 0 with y(1) = 3
x2 +y 2 2
x + y dx

using the potential functions from the last example.

Since (x, y) = (1, 3) is in the right half plane, it makes sense to use φ+ from the last
example. Letting c+ = 0 , this potential function is
y
φ+ (x, y) = Arctan for x > 0 .
x
So our differential equation has an implicit solution
y
Arctan = C+ for x >0 .
x
Taking the tangent of both sides, letting A = tan(C+ ) , and then solving for y yields the
general solution
y = Ax for x > 0 .
Combined with the initial condition, this is

3 = y(1) = A · 1 .

So A = 3 and the solution to our initial-value problem is

y = 3x for x >0 .

(We will leave the issue of whether we truly need to restrict x to being positive as an exercise
for the interested.)
154 The Exact Form and General Integrating Factors

Proving Theorem 7.4

This is one theorem we will not attempt to prove. A good proof would require a review of
“integrals over curves in the plane” and “Green’s theorem”, both of which are subjects you may
recall seeing near the end of your calculus course. Moreover, if you replace the equation
dy
M(x, y) + N (x, y) = 0
dx
with
F(x, y) = M(x.y)i + N (x, y)j
and replace the phrase “in exact form” with “a conservative vector field”, then theorem 7.4
becomes the theorem describing when a two-dimensional vector field is conservative. The
proofs of these two theorems are virtually the same, and since the theorem about conservative
vector fields is in any good calculus text, this author will save space and writing by directing the
interested student to reviewing the relevant chapters in his/her old calculus book.

Additional Exercises

7.1. For each choice of φ(x, y) given below, find a differential equation for which the
given φ is a potential function, and then solve the differential equation using the given
potential function.
a. φ(x, y) = 3x y b. φ(x, y) = y 2 − 2x 3 y
c. φ(x, y) = x 2 y − x y 3 d. φ(x, y) = x Arctan(y)
7.2. The following concern the differential equation
dy y 1
= − . (7.9)
dx 2x y
a. Verify that the above differential equation can be rewritten as
  dy
2x − y 2 + 2x y = 0 ,
dx
and then verify that this is an exact form for equation (7.9) by showing that
φ(x, y) = x 2 − x y 2
is a corresponding potential function.
b. Solve equation (7.9) using the above potential function.
c. Note that we can also rewrite equation (7.9) as
2 2   2 2 dy
e x −x y 2x − y 2 + e x −x y 2x y = 0 .
dx
Show that this is also an exact form by showing that
2 −x y 2
ψ(x, y) = e x
is a corresponding potential function.
Additional Exercises 155

7.3. Assume φ(x, y) is a potential function corresponding to

dy
M(x, y) + N (x, y) = 0 .
dx
Show that

ψ1 (x, y) = eφ(x,y) and ψ2 (x, y) = sin(φ(x, y))

are also potential functions for this differential equation, though corresponding to dif-
ferent exact forms.

7.4. Each of the following differential equations is in exact form. Find a corresponding
potential function for each, and then find a general solution to the differential equation
using that potential function (even if it can be solved by simpler means).
  dy
a. 2x y + y 2 + 2x y + x 2 = 0
dx
dy
b. 2x y 3 + 4x 3 + 3x 2 y 2 = 0
dx
dy
c. 2 − 2x + 3y 2 = 0
dx
  dy
d. 1 + 3x 2 y 2 + 2x 3 y + 6y = 0
dx
4 dy
 
e. 4x 3 y + x 4 − y = 0
dx
x dy
f. 1 + ln |x y| + = 0
y dx
dy
g. 1 + e y + xe y = 0
dx
  dy
h. e y + xe y + 1 = 0
dx
7.5. For each of the following differential equations,

i. verify that the equation is not in exact form,

ii. find an integrating factor, and

iii. solve the given differential equation (using the integrating factor just found).
  dy
a. 2y 3 + 4x 3 y 3 − 3x y 2 = 0
dx
  dy
b. y + y 4 − 3x = 0
dx
  dy
c. 2x −1 y + 4x 2 y − 3 = 0
dx
dy
d. 1 + [1 − x tan(y)] = 0
dx
  dy
e. 3y + 3y 2 + 2x + 4x y = 0
dx
156 The Exact Form and General Integrating Factors

dy
f. 2x(y + 1) − = 0
dx
dy
g. 1 + y 4 + x y 3 = 0
dx
  dy
h. 4x y + 3x 2 + 5y = 0 for y > 0
dx
 x  dy
i. 6 + 12x 2 y 2 + 7x 3 y + = 0
y dx

7.6. The following problems concern the differential equation

dy
M(x, y) + N (x, y) = 0 . (7.10)
dx
Assume M and N are continuous and have continuous partial derivatives, and let P
and Q be the functions given by
∂M ∂N ∂N ∂M
− −
∂y ∂x ∂x ∂y
P = and Q = .
N M

a. Show that, if P is a function of x only (so all the y ’s cancel out), then
R
P(x) dx
µ(x) = e

is an integrating factor for equation (7.10).

b. Show that, if Q is a function of y only (so all the x ’s cancel out), then
R
Q(x) dx
µ(y) = e

is an integrating factor for equation (7.10).