DIFFERENTIATING UNDER THE INTEGRAL SIGN

KEITH CONRAD
I had learned to do integrals by various methods shown in a book that my high
school physics teacher Mr. Bader had given me. [It] showed how to dierentiate
parameters under the integral sign its a certain operation. It turns out thats
not taught very much in the universities; they dont emphasize it. But I caught on
how to use that method, and I used that one damn tool again and again. [If] guys
at MIT or Princeton had trouble doing a certain integral, [then] I come along and
try dierentiating under the integral sign, and often it worked. So I got a great
reputation for doing integrals, only because my box of tools was dierent from
everybody elses, and they had tried all their tools on it before giving the problem
to me. Richard Feynman [2, pp. 7172]
1. Introduction
The method of dierentiation under the integral sign, due originally to Leibniz, concerns integrals
depending on a parameter, such as
_
1
0
x
2
e
tx
dx. Here t is the extra parameter. (Since x is the
variable of integration, x is not a parameter.) In general, we might write such an integral as
(1.1)
_
b
a
f(x, t) dx,
where f(x, t) is a function of two variables like f(x, t) = x
2
e
tx
.
Example 1.1. Let f(x, t) = (2x + t
3
)
2
. Then
_
1
0
f(x, t) dx =
_
1
0
(2x + t
3
)
2
dx.
An anti-derivative of (2x + t
3
)
2
with respect to x is
1
6
(2x + t
3
)
3
, so
_
1
0
(2x + t
3
)
2
dx =
(2x + t
3
)
3
6

x=1
x=0
=
(2 + t
3
)
3
t
9
6
=
4
3
+ 2t
3
+ t
6
.
This answer is a function of t, which makes sense since the integrand depends on t. We integrate
over x and are left with something that depends only on t, not x.
An integral like
_
b
a
f(x, t) dx is a function of t, so we can ask about its t-derivative, assuming
that f(x, t) is nicely behaved. The rule is: the t-derivative of the integral of f(x, t) is the integral
of the t-derivative of f(x, t):
(1.2)
d
dt
_
b
a
f(x, t) dx =
_
b
a

t
f(x, t) dx.
1
2 KEITH CONRAD
This is called dierentiation under the integral sign. If you are used to thinking mostly about
functions with one variable, not two, keep in mind that (1.2) involves integrals and derivatives with
respect to separate variables: integration with respect to x and dierentiation with respect to t.
Example 1.2. We saw in Example 1.1 that
_
1
0
(2x +t
3
)
2
dx = 4/3 +2t
3
+t
6
, whose t-derivative is
6t
2
+ 6t
5
. According to (1.2), we can also compute the t-derivative of the integral like this:
d
dt
_
1
0
(2x + t
3
)
2
dx =
_
1
0

t
(2x + t
3
)
2
dx
=
_
1
0
2(2x + t
3
)(3t
2
) dx
=
_
1
0
(12t
2
x + 6t
5
) dx
= 6t
2
x
2
+ 6t
5
x

x=1
x=0
= 6t
2
+ 6t
5
.
The answers agree.
2. Eulers factorial integral in a new light
For integers n 0, Eulers integral formula for n! is
(2.1)
_

0
x
n
e
x
dx = n!,
which can be obtained by repeated integration by parts starting from the formula
(2.2)
_

0
e
x
dx = 1
when n = 0. Now we are going to derive Eulers formula in another way, by repeated dierentiation
after introducing a parameter t into (2.2).
For any t > 0, let x = tu. Then dx = t du and (2.2) becomes
_

0
te
tu
du = 1.
Dividing by t and writing u as x (why is this not a problem?), we get
(2.3)
_

0
e
tx
dx =
1
t
.
This is a parametric form of (2.2), where both sides are now functions of t. We need t > 0 in order
that e
tx
is integrable over the region x 0.
Now we bring in dierentiation under the integral sign. Dierentiate both sides of (2.3) with
respect to t, using (1.2) to treat the left side. We obtain
_

0
xe
tx
dx =
1
t
2
,
so
(2.4)
_

0
xe
tx
dx =
1
t
2
.
DIFFERENTIATING UNDER THE INTEGRAL SIGN 3
Dierentiate both sides of (2.4) with respect to t, again using (1.2) to handle the left side. We get
_

0
x
2
e
tx
dx =
2
t
3
.
Taking out the sign on both sides,
(2.5)
_

0
x
2
e
tx
dx =
2
t
3
.
If we continue to dierentiate each new equation with respect to t a few more times, we obtain
_

0
x
3
e
tx
dx =
6
t
4
,
_

0
x
4
e
tx
dx =
24
t
5
,
and
_

0
x
5
e
tx
dx =
120
t
6
.
Do you see the pattern? It is
(2.6)
_

0
x
n
e
tx
dx =
n!
t
n+1
.
We have used the presence of the extra variable t to get these equations by repeatedly applying
d/dt. Now specialize t to 1 in (2.6). We obtain
_

0
x
n
e
x
dx = n!,
which is our old friend (2.1). Voila!
The idea that made this work is introducing a parameter t, using calculus on t, and then setting
t to a particular value so it disappears from the nal formula. In other words, sometimes to solve
a problem it is useful to solve a more general problem. Compare (2.1) to (2.6).
3. A damped sine integral
We are going to use dierentiation under the integral sign to prove
_

0
e
tx
sin x
x
dx =

2
arctan t
for t > 0.
Call this integral F(t) and set f(x, t) = e
tx
(sin x)/x, so (/t)f(x, t) = e
tx
sin x. Then
F

(t) =
_

0
e
tx
(sin x) dx.
The integrand e
tx
sin x, as a function of x, can be integrated by parts:
_
e
ax
sin xdx =
(a sin x cos x)
1 + a
2
e
ax
.
Applying this with a = t and turning the indenite integral into a denite integral,
F

(t) =
_

0
e
tx
(sin x) dx =
(t sin x + cos x)
1 + t
2
e
tx

x=
x=0
.
4 KEITH CONRAD
As x , t sin x + cos x oscillates a lot, but in a bounded way (since sin x and cos x are bounded
functions), while the term e
tx
decays exponentially to 0 since t > 0. So the value at x = is 0.
Therefore
F

(t) =
_

0
e
tx
(sin x) dx =
1
1 + t
2
.
We know an explicit antiderivative of 1/(1 + t
2
), namely arctan t. Since F(t) has the same
t-derivative as arctan t, they dier by a constant: for some number C,
(3.1)
_

0
e
tx
sin x
x
dx = arctan t + C for t > 0.
Weve computed the integral, up to an additive constant, without nding an antiderivative of
e
tx
(sin x)/x.
To compute C in (3.1), let t on both sides. Since |(sin x)/x| 1, the absolute value of
the integral on the left is bounded from above by
_

0
e
tx
dx = 1/t, so the integral on the left in
(3.1) tends to 0 as t . Since arctan t /2 as t , equation (3.1) as t becomes
0 =

2
+ C, so C = /2. Feeding this back into (3.1),
(3.2)
_

0
e
tx
sin x
x
dx =

2
arctan t for t > 0.
If we let t 0
+
in (3.2), this equation suggests that
(3.3)
_

0
sin x
x
dx =

2
,
which is true and it is important in signal processing and Fourier analysis. It is a delicate matter to
derive (3.3) from (3.2) since the integral in (3.3) is not absolutely convergent. Details are provided
in an appendix.
4. The Gaussian integral
The improper integral formula
(4.1)
_

e
x
2
/2
dx =

2
is fundamental to probability theory and Fourier analysis. The function
1

2
e
x
2
/2
is called a
Gaussian, and (4.1) says the integral of the Gaussian over the whole real line is 1.
The physicist Lord Kelvin (after whom the Kelvin temperature scale is named) once wrote (4.1)
on the board in a class and said A mathematician is one to whom that [pointing at the formula] is
as obvious as twice two makes four is to you. We will prove (4.1) using dierentiation under the
integral sign. The method will not make (4.1) as obvious as 2 2 = 4. If you take further courses
you may learn more natural derivations of (4.1) so that the result really does become obvious. For
now, just try to follow the argument here step-by-step.
We are going to aim not at (4.1), but at an equivalent formula over the range x 0:
(4.2)
_

0
e
x
2
/2
dx =

2
2
=
_

2
.
For t > 0, set
A(t) =
__
t
0
e
x
2
/2
dx
_
2
.
We want to calculate A() and then take a square root.
DIFFERENTIATING UNDER THE INTEGRAL SIGN 5
Dierentiating with respect to t,
A

(t) = 2
_
t
0
e
x
2
/2
dx e
t
2
/2
= 2e
t
2
/2
_
t
0
e
x
2
/2
dx.
Let x = ty, so
A

(t) = 2e
t
2
/2
_
1
0
te
t
2
y
2
/2
dy =
_
1
0
2te
(1+y
2
)t
2
/2
dy.
The function under the integral sign is easily antidierentiated with respect to t:
A

(t) =
_
1
0

t
2e
(1+y
2
)t
2
/2
1 + y
2
dy = 2
d
dt
_
1
0
e
(1+y
2
)t
2
/2
1 + y
2
dy.
Letting
B(t) =
_
1
0
e
(1+x
2
)t
2
/2
1 + x
2
dx,
we have A

(t) = 2B

(t) for all t > 0, so there is a constant C such that

(4.3) A(t) = 2B(t) + C
for all t > 0. To nd C, we let t 0
+
in (4.3). The left side tends to (
_
0
0
e
x
2
dx)
2
= 0 while the
right side tends to 2
_
1
0
dx/(1 + x
2
) + C = /2 + C. Thus C = /2, so (4.3) becomes
__
t
0
e
x
2
/2
dx
_
2
=

2
2
_
1
0
e
(1+x
2
)t
2
/2
1 + x
2
dx.
Letting t here, we get (
_

0
e
x
2
/2
dx)
2
= /2, so
_

0
e
x
2
/2
dx =
_
/2. That is (4.2).
5. Higher moments of the Gaussian
For every integer n 0 we want to compute a formula for
(5.1)
_

x
n
e
x
2
/2
dx.
(Integrals of the type
_
x
n
f(x) dx for n = 0, 1, 2, . . . are called the moments of f(x), so (5.1) is the
n-th moment of the Gaussian.) When n is odd, (5.1) vanishes since x
n
e
x
2
/2
is an odd function.
What if n = 0, 2, 4, . . . is even?
The rst case, n = 0, is the Gaussian integral (4.1):
(5.2)
_

e
x
2
/2
dx =

2.
To get formulas for (5.1) when n = 0, we follow the same strategy as our treatment of the factorial
integral in Section 2: stick a t into the exponent of e
x
2
/2
and then dierentiate repeatedly with
respect to t.
For t > 0, replacing x with

tx in (5.2) gives
(5.3)
_

e
tx
2
/2
dx =

t
.
Dierentiate both sides of (5.3) with respect to t, using dierentiation under the integral sign on
the left:
_

x
2
2
e
tx
2
/2
dx =

2
2t
3/2
,
6 KEITH CONRAD
so
(5.4)
_

x
2
e
tx
2
/2
dx =

2
t
3/2
.
Dierentiate both sides of (5.4) with respect to t. After removing a common factor of 1/2 on
both sides, we get
(5.5)
_

x
4
e
tx
2
/2
dx =
3

2
t
5/2
.
Dierentiating both sides of (5.5) with respect to t a few more times, we get
_

x
6
e
tx
2
/2
dx =
3 5

2
t
7/2
,
_

x
8
e
tx
2
/2
dx =
3 5 7

2
t
9/2
,
and
_

x
10
e
tx
2
/2
dx =
3 5 7 9

2
t
11/2
.
Quite generally, when n is even
_

x
n
e
tx
2
/2
dx =
1 3 5 (n 1)
t
n/2
_
2
t
,
where the numerator is the product of the positive odd integers from 1 to n 1 (understood to be
the empty product 1 when n = 0).
In particular, taking t = 1 we have computed (5.1):
_

x
n
e
x
2
/2
dx = 1 3 5 (n 1)

2.
As an application of (5.4), we now compute (
1
2
)! :=
_

0
x
1/2
e
x
dx, where the notation (
1
2
)! and
its denition are inspired by Eulers integral formula (2.1) for n! when n is a nonnegative integer.
Using the substitution u = x
1/2
in
_

0
x
1/2
e
x
dx, we have
_
1
2
_
! =
_

0
x
1/2
e
x
dx
=
_

0
ue
u
2
(2u) du
= 2
_

0
u
2
e
u
2
du
=
_

u
2
e
u
2
du
=

2
2
3/2
by (5.4) at t = 2
=

2
.
DIFFERENTIATING UNDER THE INTEGRAL SIGN 7
6. A cosine transform of the Gaussian
We are going to compute
F(t) =
_

0
cos(tx)e
x
2
/2
dx
by looking at its t-derivative:
(6.1) F

(t) =
_

0
xsin(tx)e
x
2
/2
dx.
This is good from the viewpoint of integration by parts since xe
x
2
/2
is the derivative of e
x
2
/2
.
So we apply integration by parts to (6.1):
u = sin(tx), dv = xe
x
2
dx
and
du = t cos(tx) dx, v = e
x
2
/2
.
Then
F

(t) =
_

0
udv
= uv

_

0
v du
=
sin(tx)
e
x
2
/2

x=
x=0
t
_

0
cos(tx)e
x
2
/2
dx
=
sin(tx)
e
x
2
/2

x=
x=0
tF(t).
As x , e
x
2
/2
blows up while sin(tx) stays bounded, so sin(tx)/e
x
2
/2
goes to 0. Therefore
F

(t) = tF(t).
We know the solutions to this dierential equation: constant multiples of e
t
2
/2
. So
_

0
cos(tx)e
x
2
/2
dx = Ce
t
2
/2
for some constant C. To nd C, set t = 0. The left side is
_

0
e
x
2
/2
dx, which is
_
/2 by (4.2).
The right side is C. Thus C =
_
/2, so we are done: for all real t,
_

0
cos(tx)e
x
2
/2
dx =
_

2
e
t
2
/2
.
Remark 6.1. If we want to compute G(t) =
_

0
sin(tx)e
x
2
/2
dx, with sin(tx) in place of cos(tx),
then in place of F

(t) = tF(t) we have G

(t) = 1tG(t), and G(0) = 0. From the dierential equa-

tion, (e
t
2
/2
G(t))

= e
t
2
/2
, so G(t) = e
t
2
/2
_
t
0
e
x
2
/2
dx. So while
_

0
cos(tx)e
x
2
/2
dx =
_

2
e
t
2
/2
,
the integral
_

0
sin(tx)e
x
2
/2
dx is impossible to express in terms of elementary functions.
8 KEITH CONRAD
7. Logs in the denominator, part I
Consider the following integral over [0, 1], where t > 0:
_
1
0
x
t
1
log x
dx.
Since 1/ log x 0 as x 0
+
, the integrand vanishes at x = 0. As x 1

, (x
t
1)/ log x 0.
Therefore when t is xed the integrand is a continuous function of x on [0, 1], so the integral is not
an improper integral.
The t-derivative of this integral is
_
1
0
x
t
log x
log x
dx =
_
1
0
x
t
dx =
1
1 + t
,
which we recognize as the t-derivative of log(1 + t). Therefore
_
1
0
x
t
1
log x
dx = log(1 + t) + C.
To nd C, let t 0
+
. On the right side the term log(1 + t) tends to 0. On the left side, the
integrand tends to 0: |(x
t
1)/ log x| = |(e
t log x
1)/ log x| t because |e
a
1| |a| when a 0.
Therefore the integral on the left tends to 0. So C = 0, which implies
_
1
0
x
t
1
log x
dx = log(1 + t)
for all t > 0. For example,
_
1
0
x 1
log x
dx = log 2.
We computed this denite integral without computing an anti-derivative of (x 1)/ log x.
8. Logs in the denominator, part II
We now consider the integral
F(t) =
_

2
dx
x
t
log x
for t > 1. The integral converges by comparison with
_

2
dx/x
t
. We know that at t = 1 the
integral diverges to :
_

2
dx
xlog x
= lim
b
_
b
2
dx
xlog x
= lim
b
log log x

b
2
= lim
b
log log b log log 2
= .
So we expect that as t 1
+
, F(t) should blow up. But how does it blow up? By analyzing F

(t)
and then integrating back, we are going to show F(t) behaves essentially like log(t1) as t 1
+
.
DIFFERENTIATING UNDER THE INTEGRAL SIGN 9
Using dierentiation under the integral sign, for t > 1
F

(t) =
_

2

t
_
1
x
t
log x
_
dx
=
_

2
x
t
(log x)
log x
dx
=
_

2
dx
x
t
=
x
t+1
t + 1

x=
x=2
=
2
1t
1 t
.
We want to bound this derivative from above and below when t > 1. Then we will integrate to get
bounds on the size of F(t).
For t > 1, the dierence 1 t is negative, so 2
1t
< 1. Dividing both sides of this by 1 t, which
is negative, reverses the sense of the inequality and gives
2
1t
1 t
>
1
1 t
.
This is a lower bound on F

(t). To get an upper bound on F

(t), we want to use a lower bound

on 2
1t
. Since e
a
a + 1 for all a (the graph of y = e
x
lies on or above its tangent line at x = 0,
which is y = x + 1),
2
x
= e
xlog 2
(log 2)x + 1
for all x. Taking x = 1 t,
(8.1) 2
1t
(log 2)(1 t) + 1.
When t > 1, 1 t is negative, so dividing (8.1) by 1 t reverses the sense of the inequality:
2
1t
t 1
log 2 +
1
1 t
.
This is an upper bound on F

(t). Putting the upper and lower bounds on F

(t) together,
(8.2)
1
1 t
< F

(t) log 2 +
1
1 t
for all t > 1.
We are concerned with the behavior of F(t) as t 1
+
. Lets integrate (8.2) from a to 2, where
1 < a < 2:
_
2
a
dt
1 t
<
_
2
a
F

(t) dt
_
2
a
_
log 2 +
1
1 t
_
dt.
Using the Fundamental Theorem of Calculus,
log(t 1)

2
a
< F(t)

2
a
((log 2)t log(t 1))

2
a
,
so
log(a 1) < F(2) F(a) (log 2)(2 a) + log(a 1).
Manipulating to get inequalities on F(a), we have
(log 2)(a 2) log(a 1) + F(2) F(a) < log(a 1) + F(2)
10 KEITH CONRAD
Since a 2 > 1 for 1 < a < 2, (log 2)(a 2) is greater than log 2. This gives the bounds
log(a 1) + F(2) log 2 F(a) < log(a 1) + F(2)
Writing a as t, we get
log(t 1) + F(2) log 2 F(t) < log(t 1) + F(2),
so F(t) is a bounded distance from log(t1) when 1 < t < 2. In particular, F(t) as t 1
+
.
9. Smoothly dividing by t
Let h(t) be an innitely dierentiable function for all real t such that h(0) = 0. The ratio h(t)/t
makes sense for t = 0, and it also can be given a reasonable meaning at t = 0: from the very
denition of the derivative, when t 0 we have
h(t)
t
=
h(t) h(0)
t 0
h

(0).
Therefore the function
r(t) =
_
h(t)/t, if t = 0,
h

(0), if t = 0
is continuous for all t. We can see immediately from the denition of r(t) that it is better than
continuous when t = 0: it is innitely dierentiable when t = 0. The question we want to address
is this: is r(t) innitely dierentiable at t = 0 too?
If h(t) has a power series representation around t = 0, then it is easy to show that r(t) is innitely
dierentiable at t = 0 by working with the series for h(t). Indeed, write
h(t) = c
1
t + c
2
t
2
+ c
3
t
3
+
for all small t. Here c
1
= h

(0), c
2
= h

(0)/2! and so on. For small t = 0, we divide by t and get

(9.1) r(t) = c
1
+ c
2
t + c
3
t
3
+ ,
which is a power series representation for r(t) for all small t = 0. The value of the right side of
(9.1) at t = 0 is c
1
= h

(0), which is also the dened value of r(0), so (9.1) is valid for all small x
(including t = 0). Therefore r(t) has a power series representation around 0 (its just the power
series for h(t) at 0 divided by t). Since functions with power series representations around a point
are innitely dierentiable at the point, r(t) is innitely dierentiable at t = 0.
However, this is an incomplete answer to our question about the innite dierentiability of r(t)
at t = 0 because we know by the key example of e
1/t
2
(at t = 0) that a function can be innitely
dierentiable at a point without having a power series representation at the point. How are we
going to show r(t) = h(t)/t is innitely dierentiable at t = 0 if we dont have a power series to
help us out? Might there actually be a counterexample?
The solution is to write h(t) in a very clever way using dierentiation under the integral sign.
Start with
h(t) =
_
t
0
h

(u) du.
(This is correct since h(0) = 0.) For t = 0, introduce the change of variables u = tx, so du = t dx.
At the boundary, if u = 0 then x = 0. If u = t then x = 1 (we can divide the equation t = tx by t
because t = 0). Therefore
h(t) =
_
1
0
h

(tx)t dx = t
_
1
0
h

(tx) dx.
DIFFERENTIATING UNDER THE INTEGRAL SIGN 11
Dividing by t when t = 0, we get
r(t) =
h(t)
t
=
_
1
0
h

(tx) dx.
The left and right sides dont have any t in the denominator. Are they equal at t = 0 too? The
left side at t = 0 is r(0) = h

(0). The right side is

_
1
0
h

(0) dx = h

(0) too, so
(9.2) r(t) =
_
1
0
h

(tx) dx
for all t, including t = 0. This is a formula for h(t)/t where there is no longer a t being divided!
Now were set to use dierentiation under the integral sign. The way we have set things up
here, we want to dierentiate with respect to t; the integration variable on the right is x. We can
use dierentiation under the integral sign on (9.2) when the integrand is dierentiable. Since the
integrand is innitely dierentiable, r(t) is innitely dierentiable!
Explicitly,
r

(t) =
_
1
0
vh

(tx) dx
and
r

(t) =
_
1
0
vh

(tx) dx
and more generally
r
(k)
(t) =
_
1
0
x
k
h
(k+1)
(tx) dx.
In particular, r
(k)
(0) =
_
1
0
x
k
h
(k+1)
(0) dx =
h
(k+1)
(0)
k+1
.
10. Counterexamples
We have seen many examples where dierentiation under the integral sign can be carried out
with interesting results, but we have not actually stated conditions under which (1.2) is valid.
Something does need to be checked. In [6], an incorrect use of dierentiation under the integral
sign due to Cauchy is discussed, where a divergent integral is evaluated as a nite expression. Here
are two other examples where dierentiation under the integral sign does not work.
Example 10.1. It is pointed out in [3] that the formula
_

0
sin x
x
dx =

2
,
which we discussed at the end of Section 3, can be rewritten as
(10.1)
_

0
sin(ty)
y
dy =

2
,
for any t > 0, by the change of variables x = ty. Then dierentiation under the integral sign implies
_

0
cos(ty) dy = 0,
which doesnt make sense.
The next example shows that even if both sides of (1.2) make sense, they need not be equal.
12 KEITH CONRAD
Example 10.2. For any real numbers x and t, let
f(x, t) =
_
_
_
xt
3
(x
2
+ t
2
)
2
, if x = 0 or t = 0,
0, if x = 0 and t = 0.
Let
F(t) =
_
1
0
f(x, t) dx.
For instance, F(0) =
_
1
0
f(x, 0) dx =
_
1
0
0 dx = 0. When t = 0,
F(t) =
_
1
0
xt
3
(x
2
+ t
2
)
2
dx
=
_
1+t
2
t
2
t
3
2u
2
du (where u = x
2
+ t
2
)
=
t
3
2u

u=1+t
2
u=t
2
=
t
3
2(1 + t
2
)
+
t
3
2t
2
=
t
2(1 + t
2
)
.
This formula also works at t = 0, so F(t) = t/(2(1 + t
2
)) for all t. Therefore F(t) is dierentiable
and
F

(t) =
1 t
2
2(1 + t
2
)
2
for all t. In particular, F

(0) =
1
2
.
Now we compute

t
f(x, t) and then
_
1
0

t
f(x, t) dx. Since f(0, t) = 0 for all t, f(0, t) is dieren-
tiable in t and

t
f(0, t) = 0. For x = 0, f(x, t) is dierentiable in t and

t
f(x, t) =
(x
2
+ t
2
)
2
(3xt
2
) xt
3
2(x
2
+ t
2
)2t
(x
2
+ t
2
)
4
=
xt
2
(x
2
+ t
2
)(3(x
2
+ t
2
) 4t
2
)
(x
2
+ t
2
)
4
=
xt
2
(3x
2
t
2
)
(x
2
+ t
2
)
3
.
Combining both cases (x = 0 and x = 0),
(10.2)

t
f(x, t) =
_
xt
2
(3x
2
t
2
)
(x
2
+t
2
)
3
, if x = 0,
0, if x = 0.
In particular

t

t=0
f(x, t) = 0. Therefore at t = 0 the left side of the formula
d
dt
_
1
0
f(x, t) dx =
_
1
0

t
f(x, t) dx.
is F

(0) = 1/2 and the right side is

_
1
0

t=0
f(x, t) dx = 0. The two sides are unequal!
DIFFERENTIATING UNDER THE INTEGRAL SIGN 13
The problem in this example is that

t
f(x, t) is not a continuous function of (x, t). Indeed, the
denominator in the formula in (10.2) is (x
2
+ t
2
)
3
, which has a problem near (0, 0). Specically,
while this derivative vanishes at (0, 0), it we let (x, t) (0, 0) along the line x = t, then on this
line

t
f(x, t) has the value 1/(4x), which does not tend to 0 as (x, t) (0, 0).
Theorem 10.3. The two sides of (1.2) both exist and are equal at a point t = t
0
provided the
following two conditions hold:
f(x, t) and

t
f(x, t) are continuous functions of two variables when x is in the range of
integration and t is in some interval around t
0
,
there are upper bounds |f(x, t)| A(x) and |

t
f(x, t)| B(x), both being independent of t,
such that
_
b
a
A(x) dx and
_
b
a
B(x) dx converge.
Proof. See [4, pp. 337339].
In Table 1 we include choices for A(x) and B(x) for each of the functions we have treated. Since
the calculation of a derivative at a point only depends on an interval around the point, we have
replaced a t-range such as t > 0 with t c > 0 in some cases to obtain choices for A(x) and B(x).
Section f(x, t) x range t range t we want A(x) B(x)
2 x
n
e
tx
[0, ) t c > 0 1 x
n
e
cx
x
n+1
e
cx
3 e
tx sin x
x
(0, ) t c > 0 0 e
cx
e
cx
4
e
t
2
(1+x
2
)
1+x
2
[0, 1] 0 t c t
1
1+x
2
2c
5 x
n
e
tx
2
R t c > 0 1 x
n
e
cx
2
x
n+2
e
cx
2
6 cos(tx)e
x
2
/2
[0, ) R all t e
x
2
/2
|x|e
x
2
/2
7
x
t
1
log x
(0, 1] 0 < t < c 1
1x
c
log x
1
8
1
x
t
log x
[2, ) t c > 1 t > 1
1
x
2
log x
1
x
c
9 x
k
h
(k+1)
(tx) [0, 1] |t| < c 0 max
|y|c
|h
(k+1)
(y)| max
|y|c
|h
(k+2)
(y)|
Table 1. Summary
A version of dierentiation under the integral sign for t a complex variable is in [5, pp. 392393].
11. An example needing a change of variables
Our next example is taken from [1, pp. 78,84]. For all t R, we will show by dierentiation
under the integral sign that
(11.1)
_
R
cos(xt)
1 + x
2
dx = e
|t|
.
Here f(x, t) = cos(xt)/(1 + x
2
). Since f(x, t) is continuous and |f(x, t)| 1/(1 + x
2
), the integral
exists for all t. The function e
|t|
is not dierentiable at t = 0, so we shouldnt expect to be able
to prove (11.1) at t = 0 using dierentiation under the integral sign; this special case can be treated
with elementary calculus:
_
R
dx
1 + x
2
= arctan x

= .
14 KEITH CONRAD
The integral in (11.1) is an even function of t, so to compute it for t = 0 it suces to treat the case
t > 0.
1
Let
F(t) =
_
R
cos(xt)
1 + x
2
dx.
If we try to compute F

(t) for t > 0 using dierentiation under the integral sign, we get
(11.2) F

(t)
?
=
_
R

t
_
cos(xt)
1 + x
2
_
dx =
_
R
xsin(xt)
1 + x
2
dx.
Unfortunately, there is no upper bound |

t
f(x, t)| B(x) that justies dierentiating F(t) under
the integral sign (or even justies that F(t) is dierentiable). Indeed, when x is near a large odd
multiple of (/2)/t, the integrand in (11.2) has values that are approximately x/(1 + x
2
) 1/x,
which is not integrable for large x. That does not mean (11.2) is actually false, although if we
werent already told the answer on the right side of (11.1) then we might be suspicious about
whether the integral is dierentiable for all t > 0; after all, you cant easily tell from the integral
that it is not dierentiable at t = 0.
Having already raised suspicions about (11.2), we can get something really crazy if we dierentiate
under the integral sign a second time:
F

(t)
?
=
_
R
x
2
cos(xt)
1 + x
2
dx.
If this made sense then
(11.3) F

(t) F(t) =
_
R
(x
2
+ 1) cos(xt)
1 + x
2
dx =
_
R
cos(xt) dx =???.
All is not lost! Lets make a change of variables. Fixing t > 0, set y = xt, so dy = t dx and
F(t) =
_
R
cos y
1 + y
2
/t
2
dy
t
=
_
R
t cos y
t
2
+ y
2
dy.
This new integral will be accessible to dierentiation under the integral sign. (Although the new
integral is an odd function of t while F(t) is an even function of t, there is no contradiction because
this new integral was derived only for t > 0.)
Fix c

> c > 0. For t (c, c

), the integrand in
_
R
t cos y
t
2
+ y
2
dy
is bounded above in absolute value by t/(t
2
+ y
2
) c

/(c
2
+ y
2
), which is independent of t and
integrable over R. The t-partial derivative of the integrand is (y
2
t
2
)(cos y)/(t
2
+ y
2
)
2
, which
is bounded above in absolute value by (y
2
+ t
2
)/(t
2
+ y
2
)
2
= 1/(t
2
+ y
2
) 1/(c
2
+ y
2
), which is
independent of t and integrable over R. This justies the use dierentiation under the integral sign
according to Theorem 10.3: for c < t < c

, and hence for all t > 0 since we never specied c or c

,
F

(t) =
_
R

t
_
t cos y
t
2
+ y
2
_
dy =
_
R
y
2
t
2
(t
2
+ y
2
)
2
cos y dy.
1
A reader who knows complex analysis can derive (11.1) for t > 0 by the residue theorem, viewing cos(xt) as the
real part of e
ixt
.
DIFFERENTIATING UNDER THE INTEGRAL SIGN 15
We want to compute F

(t) using dierentiation under the integral sign. For 0 < c < t < c

, the
t-partial derivative of the integrand for F

(t) is bounded above in absolute value by a function of

y that is independent of t and integrable over R (exercise), so for all t > 0 we have
F

(t) =
_
R

2
t
2
_
t cos y
t
2
+ y
2
_
dy =
_
R

2
t
2
_
t
t
2
+ y
2
_
cos y dy.
It turns out that (
2
/t
2
)(t/(t
2
+ y
2
)) = (
2
/y
2
)(t/(t
2
+ y
2
)), so
F

(t) =
_
R

2
y
2
_
t
t
2
+ y
2
_
cos y dy.
Using integration by parts on this formula for F

(t) twice (starting with u = cos y and dv =

(
2
/y
2
)(t/(t
2
+ y
2
)), we obtain
F

(t) =
_
R

y
_
t
t
2
+ y
2
_
sin y dy =
_
R
_
t
t
2
+ y
2
_
cos y dy = F(t).
The equation F

(t) = F(t) is a second order linear ODE whose general solution is ae

t
+ be
t
, so
(11.4)
_
R
cos(xt)
1 + x
2
dx = ae
t
+ be
t
for all t > 0 and some real constants a and b. To determine a and b we look at the behavior of the
integral in (11.4) as t 0
+
and as t .
As t 0
+
, the integrand in (11.4) tends pointwise to 1/(1 +x
2
), so we expect the integral tends
to
_
R
dx/(1+x
2
) = as t 0
+
. To justify this, we will bound the absolute value of the dierence

_
R
cos(xt)
1 + x
2
dx
_
R
dx
1 + x
2

_
R
| cos(xt) 1|
1 + x
2
dx
by an expression that is arbitrarily small as t 0
+
. For any N > 0, break up the integral over R
into the regions |x| N and |x| N. We have
_
R
| cos(xt) 1|
1 + x
2
dx
_
|x|N
| cos(xt) 1|
1 + x
2
dx +
_
|x|N
2
1 + x
2
dx

_
|x|N
t|x|
1 + x
2
dx +
_
|x|N
2
1 + x
2
dx
= t
_
|x|N
|x|
1 + x
2
dx + 4
_

2
arctan N
_
.
Taking N suciently large, we can make /2 arctan N as small as we wish, and after doing that
we can make the rst term as small as we wish by taking t suciently small. Returning to (11.4),
letting t 0
+
we obtain = a + b, so
(11.5)
_
R
cos(xt)
1 + x
2
dx = ae
t
+ ( a)e
t
for all t > 0.
Now let t in (11.5). The integral tends to 0 by the RiemannLebesgue lemma from Fourier
analysis, although we can explain this concretely in our special case: using integration by parts
with u = 1/(1 + x
2
) and dv = cos(xt) dx, we get
_
R
cos(xt)
1 + x
2
dx =
1
t
_
R
2xsin(xt)
(1 + x
2
)
2
dx.
16 KEITH CONRAD
The absolute value of the term on the right is bounded above by a constant divided by t, which
tends to 0 as t . Therefore ae
t
+( a)e
t
0 as t . This forces a = 0, which completes
the proof that F(t) = e
t
for t > 0.
12. Exercises
1. From the formula
_

0
e
tx
sin x
x
dx =

2
arctan t for t > 0, in Section 3, use a change
of variables to obtain a formula for
_

0
e
ax
sin(bx)
x
dx when a and b are positive. Then use
dierentiation under the integral sign with respect to b to nd a formula for
_

0
e
ax
cos(bx) dx
when a and b are positive. (Dierentiation under the integral sign with respect to a will produce
a formula for
_

0
e
ax
sin(bx) dx, but that would be circular in our approach since we used that
integral in our derivation of the formula for
_

0
e
tx
sin x
x
dx.)
2. From the formula
_

0
e
tx
sin x
x
dx =

2
arctan t for t > 0, the change of variables x = ay
with a > 0 implies
_

0
e
tay
sin(ay)
y
dy =

2
arctan t,
so the integral on the left is independent of a and thus has a-derivative 0. Dierentiation under
the integral sign, with respect to a, implies
_

0
e
tay
(cos(ay) t sin(ay)) dy = 0.
Verify that this application of dierentiation under the integral sign is valid when a > 0 and t > 0.
What happens if t = 0?
3. Prove for t > 0 that
_

0
e
tx
cos x 1
x
dx = log
_
t

1 + t
2
_
. What happens to the integral
as t 0
+
?
4. In calculus textbooks, formulas for the indenite integrals
_
x
n
sin xdx and
_
x
n
cos xdx
are derived recursively using integration by parts. Find formulas for these integrals when n =
1, 2, 3, 4 using dierentiation under the integral sign starting with the formulas
_
cos(tx) dx =
sin(tx)
t
,
_
sin(tx) dx =
cos(tx)
t
for t > 0.
5. If you are familiar with integration of complex-valued functions, show
_

e
(x+iy)
2
dx =

2
DIFFERENTIATING UNDER THE INTEGRAL SIGN 17
for all y R. In other words, show the integral on the left side is independent of y. (Hint: Use
dierentiation under the integral sign to compute the y-derivative of the left side.)
Appendix A. Justifying passage to the limit in a sine integral
In Section 3 we derived the equation
(A.1)
_

0
e
tx
sin x
x
dx =

2
arctan t for t > 0,
which by naive passage to the limit as t 0
+
suggests that
(A.2)
_

0
sin x
x
dx =

2
.
To prove (A.2) is correct, we will show
_

0
sin x
x
dx exists and then show the dierence
(A.3)
_

0
sin x
x
dx
_

0
e
tx
sin x
x
dx =
_

0
(1 e
tx
)
sin x
x
dx
tends to 0 as t 0
+
. The key in both cases is alternating series.
On the interval [k, (k + 1)], where k is an integer, we can write sin x = (1)
k
| sin x|, so
convergence of
_

0
sin x
x
dx = lim
b
_
b
0
sin x
x
dx is equivalent to convergence of the series

k0
_
(k+1)
k
sin x
x
dx =

k0
(1)
k
_
(k+1)
k
| sin x|
x
dx.
This is an alternating series in which the terms a
k
=
_
(k+1)
k
| sin x|
x
dx are monotonically decreasing:
a
k+1
=
_
(k+2)
(k+1)
| sin x|
x
dx =
_
(k+1)
k
| sin(x + )|
x +
dx =
_
(k+1)
k
| sin x|
x +
dx < a
k
.
By a simple estimate a
k

1
k
for k 1, so a
k
0. Thus
_

0
sin x
x
dx =

k0
(1)
k
a
k
converges.
To show the right side of (A.3) tends to 0 as t 0
+
, we write it as an alternating series. Breaking
up the interval of integration [0, ) into a union of intervals [k, (k + 1)] for k 0,
(A.4)
_

0
(1 e
tx
)
sin x
x
dx =

k0
(1)
k
I
k
(t), where I
k
(t) =
_
(k+1)
k
(1 e
tx
)
| sin x|
x
dx.
Since 1 e
tx
> 0 for t > 0 and x > 0, the series

k0
(1)
k
I
k
(t) is alternating. The upper bound
1 e
tx
< 1 tells us I
k
(t)
1
k
for k 1, so I
k
(t) 0 as k . To show the terms I
k
(t) are
monotonically decreasing with k, set this up as the inequality
(A.5) I
k
(t) I
k+1
(t) > 0 for t > 0.
Each I
k
(t) is a function of t for all t, not just t > 0 (note I
k
(t) only involves integration on a
bounded interval). The dierence I
k
(t) I
k+1
(t) vanishes when t = 0 (in fact both terms are then
0), and I

k
(t) =
_
(k+1)
k
e
tx
| sin x| dx for all t by dierentiation under the integral sign, so (A.5)
would follow from the derivative inequality I

k
(t) I

k+1
(t) > 0 for t > 0. By a change of variables
y = x in the integral for I

k+1
(t),
I

k+1
(t) =
_
(k+1)
k
e
t(y+)
| sin(y + )| dy = e
t
_
(k+1)
k
e
ty
| sin y| dy < I

k
(t).
This completes the proof that the series in (A.4) for t > 0 satises the alternating series test.
18 KEITH CONRAD
If we truncate the series

k0
(1)
k
I
k
(t) after the Nth term, the magnitude of the error is no
greater than the absolute value of the next term:

k0
(1)
k
I
k
(t) =
N

k=0
(1)
k
I
k
(t) + r
N
, |r
N
| |I
N+1
(t)|
1
(N + 1)
.
Since |1 e
tx
| tx,

k=0
(1)
k
I
k
(t)

_
(N+1)
0
(1 e
tx
)
| sin x|
x
dx =
_
(N+1)
0
t dx = t(N + 1).
Thus

_

0
(1 e
tx
)
sin x
x
dx

t(N + 1) +
1
(N + 1)
.
For any > 0 we can make the second term at most /2 by a suitable choice of N. Then the rst
term is at most /2 for all small enough t (depending on N), and that shows (A.3) tends to 0 as
t 0
+
.
References
[1] W. Appel, Mathematics for Physics and Physicists, Princeton Univ. Press, Princeton, 2007.
[2] R. P. Feynman, Surely Youre Joking, Mr. Feynman!, Bantam, New York, 1985.
[3] S. K. Goel and A. J. Zajta, Parametric Integration Techniques, Math. Mag. 62 (1989), 318322.
[4] S. Lang, Undergraduate Analysis, 2nd ed., Springer-Verlag, New York, 1997.
[5] S. Lang, Complex Analysis, 3rd ed., Springer-Verlag, New York, 1993.
[6] E. Talvila, Some Divergent Trigonometric Integrals, Amer. Math. Monthly 108 (2001), 432436.