Math 94 Professor: Padraic Bartlett

Lecture 7: Differential Equations

Week 7 UCSB 2015

This is the seventh week of the Mathematics Subject Test GRE prep course; here, we
review various techniques used to solve differential equations!

1 Definitions and Concepts

A differential equation is any mathematical equation that relates some collection of
functions to their derivatives; the mathematical fields of ordinary and partial differential
equations study various methods used to find all of the functions that satisfy such equations.
More so than most other fields of mathematics, the study of ordinary and partial dif-
ferential equations is far more focused on its processes than its theorems; in other words,
mathematicians in these areas are less concerned with recognizing and applying various
large theorems, and are more concerned with the practical nitty-gritty of solving and work-
ing with specific classes of equations. To give an example, think of the difference in feel
between studying limits and series in calculus, versus integrating various functions in cal-
culus. In the first subject, most of your proofs involve sticking together various results (i.e.
the mean value theorem, the squeeze theorem, etc) to solve problems. The second, however,
feels much more like an art — when integrating functions, you’re mostly looking for the
right clever use of substitution/parts/trig identities/change of variables/etc to solve your
specific equation!
Accordingly, our review section here is a bit different than other review sections. Here,
we sketch several basic techniques for solving various differential equations; in the second
half of this talk, we use these techniques to solve several GRE-styled problems from the field.
Our listing of techniques is (of course) partial, as mathematics itself only knows techniques
for solving some families of differential equations! However, almost all of the problems you
encounter on the GRE will be approachable with one of these techniques.
Finally, one last note about the specific structure of the GRE itself: due to the nature
of the GRE as a multiple-choice test, it will sometimes be possible to simply “plug in” your
five possible answers into your differential equation, and eliminate answers based on those
that do not solve your question! This may not always be the most time-efficient approach,
as it will require you to take derivatives of between one to five functions; but derivation is
usually pretty “easy” as far as mathematical operations go. Do not be afraid to “game”
the GRE in this manner; it is a multiple-choice test, and you should exploit this structure
to your advantage!

1. Separable first-order differential equations. Suppose that you have a differential

equation of the form
dy
M (x)N (y) = .
dx

1
 for two functions M (x), N (y). We can solve this equation by “separating” M (x) from
N (y): that is, by dividing both sides by N (x) and “multiplying” by dx to get1 the
following:
1
M (x)dx = dy.
N (y)
Integrating both sides yields
Z Z
1
M (x)dx = dy,
N (y)
which gives us a relation that can be used to solve for y with algebra/other techniques.
Be aware that this equation above only gives us solutions for which g(y) 6= 0. In the
event that g(y) is identically 0 — i.e. y is a constant — you would need to check this
manually by seeing if a constant value of y can solve our equation.
We calculate an example here:

Example. Solve the differential equation

3x2 + 1 dy
= ,
2y − 3 dx
with the boundary condition that when x = 0, y = 0 as well.

Proof. We simply separate variables and solve:

3x2 + 1 dy
=
2y − 3 dx
Z Z
2
⇒ (3x + 1)dx = (2y − 3)dy

⇒x3 + x = y 2 − 3y + c

Because (0, 0) is a point that should be a solution to our equation, we can see that
c = 0, and that our equation is (solving for y)
x3 + x = y 2 − 3y
⇒y 2 − 3y − (x3 + x) = 0
p
3 ± 9 + 4(x3 + x)
⇒y= .
2
At x = 0, this expression is 3±3
2 , which we know should be 0; this tells us that we
want the positive branch of this expression, i.e.
p
3 + 9 + 4(x3 + x)
y= .
2

1
Formally speaking, we are doing something more subtle than multiplying through by dx, because what
would that even mean? What is a dx, outside of an integral? For rigorous answers to this, take courses on
analysis and differential equations! For now, however, just roll with it.

2
 2. Homogeneous first-order differential equations. Suppose that you have a dif-
ferential equation of the form
dy
M (x, y) + N (x, y) = 0,
dx
where M (x, y), N (x, y) are a pair of degree-n homogeneous equations2 . We can solve
this differential equation by defining v = y/x, which lets us make the substitution
y = xv, and yields the equation
d
M (x, xv) + N (x, xv)
(xv) = 0.
dx
If M, N are homogeneous of degree n, this yields
 
n n dv
x · M (1, v) + x · N (1, v) · v + x = 0,
dx
dv
which we can solve for dx to get
 
dv N (1, v) 1
= − −v · .
dx M (1, v) x
This is a separable differential equation, and therefore is solvable by our earlier meth-
ods! Use them to solve this differential equation, and then finally substitute v = y/x
back to get a solution for our original problem.
We calculate an example here:

Example. Solve the differential equation

dy
x2 y + y 2 x
= 0,
dx
given the boundary condition that at x = 0 we want y = 1.

Proof. The two functions M (x, y) = x2 y, N (x, y) = xy 2 are both homogeneous of

degree 3, so we can attempt to proceed as directed above. We start by substituting
in y = xv, and performing various algebraic manipulations:
 
3 3 2 dv
x v+x v v+x =0
dx
dv −x3 v
⇒v + x = 3 2
dx x v
dv 1
⇒x =− −v
dx v
1 1
⇒− 1 dv = dx.
v +v
x

2
A function f (x, y) of two variables is called homogeneous of degree n if f (tx, ty) = tn f (x, y) for all
t, x, y. For example, f (x, y) = x2 + xy + y 2 is homogeneous of degree 2, as f (tx, ty) = t2 x2 + txty + t2 y 2 =
t2 f (x, y).

3
 Integrating both sides of this separable equation gives us
Z Z Z
1 v 1 2 1
− 1 dv = − 2
dv = − ln(1 + v ) = dx = ln(x) + c.
v +v
1 + v 2 x
Plugging in v = y/x yields
y2
 
1
− ln 1 + 2 = ln(x) + c
2 x
y 2  −2 c
⇒1 + 2 = e−2 ln(x)+c = c · eln(x) = 2
x x
⇒y 2 = c − x2
p
⇒y = ± c − x2 .

At x = 0 we wanted√ y = 1; so this tells us that c = 1 and our sign is positive, and

therefore that y = 1 − x2 is our answer!
Notice how we repeatedly ignore the constants, signs, etc. that c is multiplied by, as
it’s just a constant (and therefore we don’t really care if it’s two times some other
constant.)

3. Linear first-order differential equations. Suppose that you have a differential

equation of the form
dy
+ P (x)y = Q(x),
dx
and we want to solve for y as a function of x. We can solve this equation by multiplying
both sides by the “integrating factor3 ”
R
P (x)dx
µ(x) = e .
If we do this, then we get
dy
µ(x) + P (x)yµ(x) = Q(x)µ(x).
dx
d
However, the product rule tells us that the LHS above is just dx (µ(x) · y). Therefore,
our equation is now of the form
d
(µ(x) · y) = Q(x)µ(x).
dx
Integrating both sides with respect to x yields
Z
1
y= Q(x)µ(x)dx.
µ(x)
Success!
We calculate an example here:
3
An integrating factor is some cleverly-chosen function that we multiply both sides of a differential
equation by to make it simpler in some appropriate sense.

4
 Example. Solve the differential equation
dy
+ x2 y = x5 ,
dx
given the boundary condition that at x = 0 we want y = 0.

Proof. This is a linear

R 2 differential equation; therefore if we multiply both sides by the
3
integrating factor e x dx = ex /3 , and go through all of the steps above we get
dy x3 /3 3 3
e + x2 yex /3 = x5 ex /3
dx
∂  x3 /3  3
⇒ ye = x5 ex /3
∂x
 3  Z 3
⇒ yex /3 = x5 ex /3 dx.

To integrate the RHS, we use the substitution u = x3 /3, motivated by the fact that
eanything not just a single variable is a total pain to calculate: as du = x2 dx and 3u = x3 ,
we get
 3  Z 3
yex /3 = x5 ex /3 dx
Z
3
⇒ = x3 ex /3 x2 dx
Z
= 3ueu du

= 3ueu − 3eu + C
3 /3
= ex (x3 − 3) + C
C
⇒ y = x3 − 3 + .
ex3 /3
Our boundary conditions tell us that we want (0, 0) to be a solution to our equation:
in other words, that 3 = C.

4. Exact first-order differential equations. Suppose that you have a differential

equation of the form
dy
M (x, y) + N (x, y) = 0,
dx
∂ ∂
where ∂y M (x, y) = ∂x N (x, y). Now, take any function F (x, y) that is the antideriva-
tive of M (x, y) with respect to x, and also the antiderivative of N (x, y) with respect

5
to y: i.e. some function F (x, y) such that
Z
F (x, y) = M (x, y)dx + Cy ,
Z
F (x, y) = N (x, y)dy + Cx ,

Note that I have written Cy , Cx instead of the normal constants C; this is because when
we integrate with respect to x, y is held constant (and similarly for y, x.) Therefore,
terms involving the coefficient we are not integrating by are “constants” that can show
up in our solution! (This is why we need to consider integrating both M (x, y) and
N (x, y), and not just one of the two.)
Fun fact we’re not proving here: such a function always exists for exact differential
equations, and you can always find it!
∂ ∂
When you do, you’ll get that ∂x F (x, y) = M (x, y), ∂y F (x, y) = N (x, y). Therefore,
we can write our differential equation in the form
∂ ∂ dx
F (x, y) + F (x, y) = 0.
∂x ∂y dy

But this is simply the total derivative of the function F (x, y)! Therefore, if we are
asking that this total derivative is 0, we are looking for the set of all points (x, y) on
which F (x, y) is constant; that is, the set of all level curves of F (x, y), i.e.

F (x, y) = c.

We calculate an example here:

Example. Solve the differential equation

dy
y cos(xy) + x cos(xy) = 0,
dx
given the boundary condition that at x = 1 we want y = 0.

Proof. We first notice that because

∂ ∂
(y cos(xy)) = cos(xy) − xy sin(xy) = (x cos(xy)),
∂y ∂x
this differential equation is indeed exact. Therefore, we are seeking some function
F (x, y) such that
Z Z
F (x, y) = M (x, y)dx = y cos(xy)dx = sin(xy) + Cy ,
Z Z
F (x, y) = N (x, y)dy = x cos(xy)dy = sin(xy) + Cx ;

6
 i.e. F (x, y) = sin(xy) + C. Our solutions are simply the level curves of this function;
i.e. the set of all points (x, y) satisfying sin(xy) = c. If we want (1, 0) on this curve,
we want sin(0) = C; i.e. C = 0, and therefore that our solutions are the set of all
points satisfying sin(xy) = 0.

5. Nonexact first-order differential equations. Sometimes you have a differential

equation
dy
M (x, y) + N (x, y) = 0,
dx
that is not exact, but is not too far from it. In these situations, we can sometimes be
“clever” and multiply both the LHS and RHS by a cleverly-chosen integrating factor
µ(x) or µ(y), so that the resulting equation is exact! In practice, it can be very easy
or very difficult to find an integrating factor. We list a few special cases where such
factors have predictable forms here:
∂ ∂
∂y M (x, y) − ∂x N (x, y)
• Suppose that is a function ξ(x) that only depends on
N (x, y) R
the variable x. Then µ(x) = e ξ(x)dx is an integrating factor for our differential
equation.
∂ ∂
∂y M (x, y) − ∂x N (x, y)
• Suppose that is a function ψ(y) that only depends on
M (x, y) R
the variable y. Then µ(y) = e ψ(y)dy is an integrating factor for our differential
equation.

We calculate an example here:

Example. Solve the differential equation

dy
(3x2 y + y 3 + 2yx) + (x2 + y 2 ) = 0,
dx
given the boundary condition that at x = 0 we want y = 0.

Proof. We first notice that because

∂
(3x2 y + y 3 + 2yx) = 3x2 + 3y 2 + 2x,
∂y
∂ 2
(x + y 2 ) = 2x,
∂x

we are sadly not exact. However, we do have that

∂ ∂
∂y M (x, y) − ∂x N (x, y) 3x2 + 3y 2 + 2x − 2x
= =3
N (x, y) x2 + y 2

7
 is indeed a function ξ(x) that only depends4 on the variable x! RTherefore, as suggested
above, we can multiply both sides by the integrating factor e ξ(x)dx = e3x , to get
dy
(3x2 y + y 3 + 2yx)e3x + (x2 + y 2 )e3x = 0.
dx
We can see that this equation now is exact, as
∂
(3x2 y + y 3 + 2yx)e3x = (3x2 + 3y 2 + 2x)e3x ,
∂y
∂ 2
(x + y 2 )e3x = 2xe3x + (x2 + y 2 )3e3x
∂x

are both equal. Therefore, we can find a solution by integrating (3x2 y+y 3 +2yx)e3x , (x2 +
y 2 )e3x appropriately:
Z
F (x, y) = (3x2 y + y 3 + 2yx)e3x dx

y 3 3x
   
2 3x 2 3x 2 3x 2 3x 2 3x
y x e − xe + e + e +y xe − e + Cy
3 9 3 3 9
y3
=yx2 e3x + e3x + Cy .
Z 3
F (x, y) = (x2 + y 2 )e3x dy

y 3 3x
=yx2 e3x + e + Cx .
3

So we have
 
3x 2 3x 2 3x
F (x, y) = e +y xe − e + C,
3 9

Solutions to our differential equation are level curves of this function; i.e. all x, y such
that
y 3 3x
yx2 e3x + e = C.
3
Asking that (0, 0) is on such a curve is simply the restriction that C = 0; that is, we
have
y 3 3x
yx2 e3x + e = 0.
3

4
Well, really, it doesn’t depend on anything. But that’s OK: constant functions are functions!

8
2 Example GRE Problems
We work four example problems here, taken from the three GRE exams you’ve completed
thus far in this class:

Problem. Let y = f (x) be a solution of the differential equation

xdy + (y − xex )dx = 0,

chosen such that y = 0 at x = 1. What is the value of f (2)?

1 1 e2
(a) 2e (b) e (c) 2 (d) 2e (e) 2e2

Answer. Notice that this equation is exact, as

∂
(y − xex ) = 1,
∂y
∂
(x) = 1.
∂x

Therefore, we can solve this by simply integrating these two functions appropriately:
Z
F (x, y) = (y − xex )dx = xy − xex + ex + Cy ,
Z
F (x, y) = x dy = xy + Cx .

Combining these results gives us F (x, y) = xy − xex + ex + C, which we want to find level
curves for; i.e. our solutions look like xy − xex + ex = C. If we plug in the point (1, 0), we
get C = 0. Finally, if we want to find out what happens when we have x = 2, note that

2y − 2e2 + e2 = 0

e2
implies that y = f (2) is just 2. In other words, our answer is (c).

Problem. Which of the following five pictures gives the graphs of two functions satisfying
the differential equation
 2
dy dy
+ 2y + y 2 = 0?
dx dx

9
Answer. Factoring our equation yields
 2
dy
+y = 0,
dx

so we can simply take a square root to get the simpler problem

dy
+ y = 0.
dx
This is a very simple separable differential equation; we can separate it accordingly to get
1
dy = −dx ⇒ ln(y) = −x + C ⇒ y = Ce−x .
y

The only answer whose curves have Ce−x -like behavior is (a), so we have answered our
question.

Problem. Suppose that we have a tank of water. This tank is a cube with vertical sides,
no top, and side length 10 feet. Let h(t) denote the height of the water level, in feet, above
the floor of the tank at time t.
Suppose that at time t = 0 water begins to pour into the tank at a constant rate of 1
cubic foot per second, and also begins to pour out of the tank at a rate of h(t) 4 cubic feet
per second. As t approaches infinity, what is the limit of the volume of the water in the
tank?

(a) 400 ft3 (b) 600 ft3 (c) 1000 ft3 (d) The limit DNE.
(e) We do not have enough information to solve this problem.

10
Answer. We note that on one hand, if we let V denote the volume of our tank, we have
V = 100h; consequently, we have that dV dh dV
dt = 100 dt . Conversely, we are given dt directly as
1 − h/4; therefore, by combining, we have the differential equation

dh 1 1
+ h= .
dt 400 100
R
This is linear; therefore, if we multiply both sides by the integrating factor e (1/400)dt =
et/400 , we get
dh t/400 1 t/400 1
e + e h = et/400
dt 400 100
d  t/400  1
⇒ he = et/400
dt Z 100
1
⇒het/400 = et/400 dt
100
= 4et/400 + C
C
⇒ h = 4 + t/400 .
e

As t goes to infinity, this expression converges to 4; therefore the volume, which is 100h,
goes to 400. So our answer is (a).

11