Hard Integrals

STEM Horizons
Hrishabh Ayush and Ong Cheng Yiin
March 2021

Problem 1
Z ∞
sin x
dx
0 x
iz
Proof. (Complex Analysis) Let us define f (z) = ez . We’re going to use indented semicir-
cular contour since the function has a pole at the origin.

Im(z)

Re(z)
−R −ε ε R


2t
 
We will be using the fact that sin(t) > π valid for all values of t ∈ 0, π2 .
I Z R Z Z −ε Z
f (z)dz = f (z)dz + f (z)dz + f (z)dz + f (z)dt
C ε Γ −R Y

By our definition, we know that,

Z −ε Z −ε iz
e
f (z)dz = dz
−R −R z

1
STEM Horizons
Hard Integrals
Hrishabh Ayush and Ong Cheng Yiin (March 2021)

Remark. Substitute z = − x then dz = −dx for replacing the negative sign.

e−ix ε e−ix
Z −ε iz
e
Z ε Z
dz = dx = −
dx
−R z R −x R x
By changing the orientation of the integral we will get
Z ε −ix Z R −ix
e e
dx = − dx
R x ε x
Now,
Z R −ix
e − e−ix
Z R ix Z R ix
e e
dx − dx = dx
ε x ε x ε x
Recall the property of sinx in terms of exponential functions

eix − e−ix
sin( x ) =
2i
2i sin( x ) eix − e−ix
=
x x

Therefore, the whole integral becomes

Z R
sin x
2i dx
ε x
Also, we need to calculate the other integrals
Z  Z iz 
   e 
 f ( x ) dz =
 τ z dz
 
  τ

Remark. (Supporting substitution) z = Reit . So, dz = iReit dt, where 0 ≤ t ≤ π.

Z it

eiRe
 π  Z π 
it iReit
 
iRe dt = i  e dt
  

Re it
  0

Hence, Z π  Z π 
  2 iReit Z π
 iRe it iRe it 
 0 e
 dt =  e dt + π e dt
0 2

Theorem
Euler’s Identity eit = cos t + i sin t

Now we’ll calculate the integrals individually and use a bit of triangle inequality
Z π
 2 iReit  Z π2 iR cos t i2 R sin t 
  
 e dt = 
  e ·e dt
0 0

Z π    
2  iR cos t   − R sin t 
≤  e ·
  e  dt
0

2
STEM Horizons
Hard Integrals
Hrishabh Ayush and Ong Cheng Yiin (March 2021)

We know that for some t ∈ R, we have

|eit | = 1

Hence, |eiR cos t | = 1 makes a unit circle on the complex plane.

Z π
2
= e− R sin t dt
0

2t
 
Using, the initial condition of sin t ≥ π , ∀t ∈ 0, π2

π
−π h −2Rt i π2
Z
2 −2Rt
≤ e e π
π dt =
0 2R 0
−π − R π −
πe R
= [ e − 1] = −
2R 2R 2R

π πe− R
Remark. When R goes to infity, so the value 2R also goes to 0, so does 2R . Therefore, the
whole thing goes to 0.

Hence,
Z π
2 it dt
eiRe = 0.
0
Similarly, Z π 
 2 iReit  Z π − R sin t
 0 e
 dt ≤ π e dt
2

Remark. (Supporting subtitution) t = π − u, So u = π − t =⇒ dt = −du.

Therefore, from the previous substitution we get,

Z 0 Z π
2

π
e− R sin(π −u) du = e− R sin(π −u) du
2 0
Z π
2
= e− R sin u du (∵ sin(π − u) = sin u)
0

Lastly, we have
eiz
Z Z
f (z)dz = dz
γ γ z

3
STEM Horizons
Hard Integrals
Hrishabh Ayush and Ong Cheng Yiin (March 2021)

Theorem
Taylor Series:

x2 x3
ex = 1 + x + + +···
2! 3!
eiz 1 i 2 z i 3 z2
= +i+ + +···
z z 2! 3!

So, we get a series of the type 1z + P(z) where P(z) is the power series in z. We can now
find a strong upper bound for the series P(z). Suppose that,

max | P(z)| = M > 0

z∈γ

By the ML inequality, we get

Z
P(z)dz ≤ πε max | P(z)| = πεM.
γ z∈γ

Since, ε goes to 0, therefore the whole quantity, πεM goes to 0. For the other part we
have by the line integral,
1
Z
= −iπ
γ z

Hence, we finally have

Z ∞
sin x
0 = 2i dx − iπ
0 x
Z ∞
sin x π
∴ dx = .
0 x 2

Remark. This formula is usually one of the high points of a course in complex analysis, and
is a consequence of the so-called residue theorem. But there are are several, real-variable
proofs of this, including using doubles integrals or using differentiation under the integral
sign (popularized by Feynman, as an alternative to residue calculus).

Proof. (Elementary Calculus) We’ll use Feynman Method of Integration to compute this
integral, let Z ∞
sin x −ax
I ( a) = e dx,
0 x
notice that I (0) is our desired function. By differentiating I ( a) with respect to a, we have
Z ∞   Z ∞
dI ( a) ∂ sin x −ax
= e dx = − e−ax sin x dx.
da 0 ∂a x 0

We now compute the RHS integral utilising a standard trick:

Z ∞ Z ∞ Z ∞  Z ∞

− ax − ax − ax −( a−i ) x
e sin x dx = = e cos x dx + i e sin x dx = = e dx
0 0 0 0

4
STEM Horizons
Hard Integrals
Hrishabh Ayush and Ong Cheng Yiin (March 2021)
√
where i = −1 is the imaginary unit and =( x ) is the imaginary component of the
complex number x. Thus,
Z ∞ Z ∞ 
e−ax sin x dx = = e−(a−i)x dx
0 0
∞ !
e−(a−i)x 
== −
a−i 

0
 
1
==
a−i
 
a+i
== 2
a +1
1
= 2
a +1
which means
dI ( a) 1 1
Z
=− 2 =⇒ I ( a) = − 2
da = − arctan( a) + C.
da a +1 a +1
By letting a → ∞, we see that because e−ax → 0, I ( a) vanishes. So,
π π
0 = lim I ( a) = lim (− arctan( a) + C ) = − + C =⇒ C = .
a→∞ a→∞ 2 2
Back to our desired integral,
Z ∞
sin x π π
dx = I (0) = − arctan(0) + = .
0 x 2 2


Problem 2
Z π
ln(sin x ) dx
0

Proof. We know that

x x
Z π Z π 
I= ln sin x dx = ln 2 · sin · cos dx
0 0 2 2
x x
Z π 
= π ln 2 + ln sin cos dx
0 2 2
Z π Z π 
y= 2x 2 2
= π ln 2 + 2 ln sin y dy + ln cos y dy .
0 0

But by using symmetry between sin and cos, as

Z π Z π Z 0 Z π
2 2
 π  z= π2 −y 2
ln cos y dy = ln sin −y dy = − π ln sin z dz = ln sin z dz,
0 0 2 2 0

we have
Z π Z π 
2 2
I = π ln 2 + 2 ln sin y dy + ln cos y dy
0 0
Z π Z π 
2 2
= π ln 2 + 2 ln sin y dy + ln sin y dy
0 0
Z π
2
= π ln 2 + 4 sin y dy.
0

5
STEM Horizons
Hard Integrals
Hrishabh Ayush and Ong Cheng Yiin (March 2021)

Now, with the same technique we can find the integral

Z π
2
0
I = sin x dx,
0

which we know that

π π
x x
Z Z 
2 2
0
I = ln sin x dx = ln 2 · sin · cos dx
0 0 2 2
Z π
π 2
 x x
= ln 2 + ln sin cos dx
2 0 2 2
Z π Z π 
y= 2x π 4 4
= ln 2 + 2 ln sin y dy + ln cos y dy .
2 0 0

But by using symmetry between sin and cos, as

π π Z π
z= π2 −y
Z Z  π 
4 4 2
ln cos y dy = ln sin −y dy = ln sin z dz,
0 0 2 π
4

we have
Z π Z π 
0 π 4 4
I = ln 2 + 2 ln sin y dy + ln cos y dy
2 0 0
Z π Z π 
π 4 2
= ln 2 + 2 ln sin y dy + π ln sin y dy
2 0 4
π 0
= + 2I
2

so, I 0 = − π2 ln 2 and finally, I = π ln 2 + 4I 0 = −π ln 2 . 

Remark. There are many problems in Putnam, MIT Integration bee based on this with the
limits changed. In particular Putnam 1953 A4, MIT Integration Bee 2019, etc.

Problem 3
Z ∞
sin( x2 )dx
0

Proof. We define f (z) = e − z2

6
STEM Horizons
Hard Integrals
Hrishabh Ayush and Ong Cheng Yiin (March 2021)

Im(z) Rei 4
π

Γ3 Γ2

π
4
Γ1 Re(z)

Remark. The path of our contour integral begins from the origin then goes to point via path
Γ1 , then follows a curve Γ2 and then comes back to the origin through the path Γ3 .
Z Z Z Z
f (z) dz = f (z) dz + f (z) dz + f (z) dz
C T1 T2 T3
Let, γ(t)=t, where, 0 ≥ t ≥ R

Z Z R
0
f (z) dz = f (t) · γ (t) dt
T1 0
Z R
z
= e−t dt
√0
π
=
2
π
Now, we have γ(t) = Rei 4 t dt where 0 ≤ t ≤ 1.
Z  Z 
(−γ(t)2 ) 0
   
 Γ f (z) dz =  Γ e
   · γ (t) dt
2
Z 12 π

−( Rei 4 t)2 π i π4 t
 
= e
 · R(i ) Re dt
0 4
Z 1  
πt    
i
≤ e− R2 e 2   Ri π  ei π4 t  dt
0
  4

Remark. For the previous 2 parts, we have used the Gaussian integral result and the second
conclusion with ML inequality. We’re trying to find the answer in all the cases where R
tends to infinity.

Recall
π
2 ei 2 t 2
= e− R (cos( 2 t)+i sin( 2 t))
π π
e− R
 2
  2

= e− R cos( 2 t)  · e−iR sin( 2 t) 
 π   π 
 2

= e− R cos( 2 t) 
 π 

7
STEM Horizons
Hard Integrals
Hrishabh Ayush and Ong Cheng Yiin (March 2021)

Hence,
Z 1  
− R2 cos( π2 t) πR
≤ e · dt
0 4
Z 1
π 2 cos( π2 t)
= e− R dt
4 0

We know,
d2   π  −π 2 π 
cos t = cos t ≤0
dt2 2 4 2
π


on (0, 1), So, cos 2t is concave on (0, 1). So, tangent line at t = 1, makes the equation:
π  −π π 
y − cos t = sin t ( t − 1)
2 2 2
−π
y−0 = ( t − 1)
2
π π
y = t+
π  2 2
π π
So , cos t ≤ − t+
2 2 2

Hence, the integral becomes,

Z  Z 1
 ≤ πR 2
e− R (− 2 t+ 2 )
π π
 
 f ( z ) dz
 T
2
 4 0

Remark. (Supporting substitution)Let, u = − π2 t + π2 , du = − π2 dt. Hence, we can now do

the substitution in terms of u instead of t.

Z 0
−2
 
π − R2 u
= R e dt
4 π
2
π
 Z π
π 2 2 2
= R e− R u du
4 π 0
" 2
#
R θ −R u
=
2 − R2
−1 −R22 π
= (e − 1) = 0 (As R goes to infinity)
2R
iπ
Finally, γ(t) = e 4 ( R − t), 0 ≤ t ≤ R
Z Z R
2 0
f (z) dz = e−γ(t) · γ (t) dt
T3 0
Z R 2
iπ
2 ( R−t)
e−e e−i 4 dt
π
=
0
Z R 2
iπ
2 ( R−t)
e−e
π
= ei 4 dt
0

8
STEM Horizons
Hard Integrals
Hrishabh Ayush and Ong Cheng Yiin (March 2021)

Substitute l = R − t then dl = −dt So, we get

Z 0 2
Z R 2
iπ
2t iπ
2t
− e −i 4 e−e (−dt) = e−i 4 e−e
π π
dt
R 0
Z R
2
= −e i π4
e−it dt
Z0 ∞
2
= −e i π4
e−it dt (as R tends to infinity)
0

Now, by Cauchy’s Integral Theorem, we get:

√ Z ∞
π 2
e−it dt
π
0= + 0 + (−ei 4 )
2 0

So, now we only need to find the value of the left integral
Z ∞ √ √
−it2 π −i π t π i
e dt = e 4 = √ −√
0 2 2 2

Theorem
We will use Euler’s Identity e−iθ = cos θ − i sin θ.

Z ∞ Z ∞
2
cos(t ) dt − i sin(t2 ) dt
0 0

√ √ r
π π π
= √ −i √ = (1 − i )
2 2 2 2 8
Hence, comparing the real and and imaginary parts we get
Z ∞ r
2 π
sin( x ) dx =
0 8


Remark. In particular it is called the Fresnel Integral.

Proof. Since we’re dealing with a non-elementary integral (in particular, it’s called Fres-
nel integral), we can’t directly evaluate the desired integral. However, notice that sin
and cos are usually related to each other! Well, let’s try the same trick again by combining
it with cos as in Problem 1. It is equivalent to
Z ∞ Z ∞  Z ∞ 
2
sin( x2 ) dx = = cos( x2 ) + i sin( x2 ) dx = = eix dx
0 0 0

2
but the integral of eix from 0 to infinity doesn’t do any good. But hey! If we can somehow
2
change it to be the integral of e− x from 0 to infinity, wouldn’t it be great? It’s the Gaussian

9
STEM Horizons
Hard Integrals
Hrishabh Ayush and Ong Cheng Yiin (March 2021)

Integral! So, let’s take

Z ∞ Z ∞

sin( x2 ) dx = −= cos( x2 ) − i sin( x2 ) dx
0 0
∞
Z 
2 2
= −= cos(− x ) + i sin(− x ) dx
0
Z ∞ 
2
= −= e−ix dx .
0
√
We’re close, let’s make a substitution, u = ix, and it becomes
Z ∞  Z ∞  √ Z ∞  √ √ 
2 1 − u2 − u2 π
sin( x ) dx = −= √ e du = = i i e du = < i·
0 i 0 0 2
√ √
and we know i = 22 (1 + i ) by Euler’s Formula: ei 4 , so
π

Z ∞ √ √  r  r
2 π π π
sin( x ) dx = < i· =< (1 + i ) = .
0 2 8 8

10