Changing the way we change the variables.

Refresher notes in real analysis

Valentin Fadeev
To my Mother
 Introduction

Dealing with problems of calculus often requires the ability to manipulate ex-
pressions casting them in certain adequate forms. Some of these transformations
take a lot of skill and ability to think several steps ahead. Frequently used tricks
are classified and offered in the text books in the form of ready made solutions.
Learning these techniques often gives the impression that they are artificially in-
vented and would be nearly impossible to come up with without prior knowledge
of the answer. Most of the arguments in modern text books are conducted in such
synthetical way, with the result in mind, the details polished so that they nicely fit
together in the end.
We shall concentrate on a specific subject of analysis namely, calculating inte-
grals through change of variable. Our main objective will be to demonstrate the
way of looking at these problems which allows to arrive at the necessary transfor-
mations in a natural way. Instead of trying to plug in the the standard catalogue
solutions we shall try to extract them from the given problem. The role of patterns
in mathematics is enormous, but it is recognizing, not enforcing them that delivers
the best results.
We shall also examine the value of looking at the same problem in many dif-
ferent ways. The general principle here would be the following: it is better to be
able to solve the one problem using many methods, than to solve many problems
using the same method.
This work does not offer an exhaustive account of the integration methods.
Routine calculation procedures are intentionally avoided. The aim is to give the
reader through a series of worked examples and supporting exercises a notion of
flexibility when working with analytic expressions. There are no new methods
offered which is hardly possible in such well-researched subject, but those selected
are arranged to maintain the overall style of solutions. Applying the same approach
can make learning and applying new methods easier.

3
 CHAPTER 1

Indefinite integrals

The need to cast expressions into forms fitting some recognizable patterns arises
very early in the problems of analysis. A simple rearrangement often allows us to
say something about the behavior of a seemingly inscrutable expression. A typical
example here is the task of calculating limits of certain sequences. There are some
standard limits known:

sin ↵
(1.0.1) lim =1
↵!0 ↵

✓ ◆n
1
(1.0.2) lim 1+ = e ⇡ 2.71828
n!1 n

ln (1 + x)
(1.0.3) lim =1
x!0 x
These can be occur as building blocks of more complicated expressions.
Example 1. ◆x ✓
x
lim
x!1 x + 1
This is an indeterminate expression of the type 11 , so additional work is re-
quired to find its value. Since we have the variable both in the base and the
exponent, (1.0.2) seems an appropriate option to apply. Divide both numerator
and denominator by x:

✓ ◆x ✓ ◆x ✓ ◆x 1
x 1 1
lim = lim = lim 1+ =
x!1 x+1 x!1 1 + 1 x!1 x
x
 ✓ ◆x 1
1
= lim 1 + = lim e x = 0
x!1 x x!1

The third transition is legitimate since f (t) = 1t is continuous as t ! 1. After

the second transition we arrived at a familiar expression which allowed us to invoke
the standard result.
Exercise 1. Calculate the limits
x
a) limx!1 1 + x12
b) limx!e x e
ln x 1

We shall pursue the same approach when dealing with integrals, not making a
substitute until it becomes obvious. Effectively, a new variable should be under-
stood as a placeholder or a “wrapper” for a frequently occurring part of expression.
4
 1. INDEFINITE INTEGRALS 5

There is an analogue to this procedure in the theory of programming which forms

one of the foundation of object-oriented approach. Fragments of code frequently
referred to in the project are given specific pointers by which they are called when
necessary without rewriting them again.
Now we shall revisit the basic concepts of the changing of variable under the
integral sign, before working through several examples.
Given a function g (t) for which a primitive is known:
ˆ
g (t) dt = G (t) + C

integral of the function f (t) can be calculated provided that a function is

found such that:

0
f [ (w)] (w) = g (w)
Making the change of variable: t = (w) we obtain:
ˆ ˆ ˆ
0
f (t) dt = f [ (w)] (w) dw = g (w) dw = G (w) + C
.
The difficulty of this method is contained in finding the appropriate function
which depends on the structure of the integrand. Various classes of integrands
have been identified for which certain standard substitutes almost always simplify
the expressions involved. However, our aim is not to browse the catalogues looking
for suitable expressions, but to work it out in a natural way.
In order to achieve better understanding and agility in manipulations it is useful
to revisit the basics and pay special attention to the d sign under the integral. It is
often seen as a formal thing, like a closing bracket for the expression. However, in
fact it is a linear operator which can significantly help to transform the integrand.
In the theory of the Lebesgue-Stieltjes integral the following notation is adopted:
ˆ b
I= f (x) dF (x)
a
where F (x) is a non-negative monotonic function inducing the measure µ on
the interval [a, b]. If F (x) is differentiable on [a, b] then the expression reduces to
an ordinary Riemann integral:
ˆ b
I= f (x) F 0 (x) dx
a
Taken out of context of the Lebesgue-Stieltjes integration, the inverse transition
(1.0.4) F 0 (x) dx = d (F (x))
(adding brackets to empathize the role of d as an operator) forms the basis of a
technique referred to as “inserting expressions under d-sign”. It is sometimes taught
as a stand-alone method of calculating integrals together with changing of variable,
integration by parts and several others. We shall use the notation like d (F (x))
extensively in further examples, without referring to the concepts of measure theory.
Finally, the obvious, but nonetheless useful property of d as a linear operator
implies that for constant c 6= 0 the following identity holds:
 1. INDEFINITE INTEGRALS 6

d (cf ) = cdf
For all its simplicity it happens to be very useful in the arguments.
Example 2.
du
ˆ
I=
a(1 + u2 ) + b(1 u2 )
Before making any changes of the variable we perform some rearrangements of
the integrand:

du du
ˆ ˆ
I= 2 2
= =
a(1 + u ) + b(1 u ) (a + b) + (a b)u2
1 du 1 du
ˆ ˆ
= a b 2
= ⇣q ⌘2
a+b 1 + a+b u a + b a b
1+ a+b u
Now that we have isolated some expression squared in the denominator we will
push the same under the differential, canceling out the artificially inserted factors:
⇣q ⌘
p d a b
1 a b 1 du 1 a+b u
ˆ ˆ
I= p p p ⇣q ⌘2 = p ⇣q ⌘2
a+b a+b a b a b a2 b2 a b
1+ a+b u 1+ a+b u
q
From the last expression it is clear that wrapping aa+bb u under a new variable,
say v, we can cast the integrand to a standard form found in the tables:
r !
1 dv 1 1 a b
ˆ
I=p 2
=p arctan v + C = p arctan u +C
a2 b2 1+v a 2 b 2 a 2 b2 a+b
Example 3. Evaluate
dt
ˆ
I=
a2 cos2 t + b2 sin2 t
First factor out the cosine:

dt 1 dt
ˆ ˆ
I= =
a2 cos2 t + b2 sin2 t a2 + b2 tan2 t cos2 t
0
Now noticing that (tan t) = cos12 t we can apply (1.0.4) to push the function
under d. At the same time we can add the required constant factor balancing it
out outside the integral:
b ✓ ◆
1 a d a tan t 1 b
ˆ
I= 2 = arctan tan t + C
a2 b
1+ b ab a
a tan t
In both of the above examples we actually did not need to introduce the new
variable, because we could effectively operate with expression blocks as with vari-
ables. In other cases calculations can get bulky, especially if the expression is not
immediately cast into the standard form. In such instances assigning a placeholder
to the expression is convenient.
 1. INDEFINITE INTEGRALS 7

Example 4. ˆ p ⇥ ⇤
x2 + 1 ln x2 + 1 2 ln x
I= dx
x4
A first glimpse at the integrand indicates similar expressions under the radical
and logarithm, as well as the possibility to replace the difference of the logarithms
by a logarithm of the ratio:

ˆ p ⇥ ⇤ ˆ r ✓ ◆
x2 + 1 ln x2 + 1 2 ln x 1 1 dx
I= dx = 1 + ln 1 + =
x4 x2 x2 x3
ˆ r ✓ ◆ ✓ ◆
1 1 1 1
= 1 + 2 ln 1 + 2 d 1 + 2
2 x x x
1 0
In the last transition we used the fact that x2 = 1
x3 and the following
identity:

1
df = d (af + b)
a
following from the linearity of d. Now that there are clearly distinguishable
blocks 1 + x12 it makes sense to let:

1
1+ =t
x2
to obtain:

1 p
ˆ
I= t ln tdt
2
This integral can be evaluated using integration by parts:
ˆ ˆ
udv = uv vdu
p
We can push t under d:
p 2 ⇣p 3 ⌘ 2 ⇣ p ⌘
tdt = d t = d t t
3 3

1 p
ˆ
1
ˆ ⇣p⌘ 1 p 1
ˆ p
I= t ln tdt = ln td t t = t t ln t + t td (ln t) =
2 3 3 3
1 p
ˆ p
1 dt 1 p 1 p
ˆ
= t t ln t + t t = t t ln t + tdt =
3 3 t 3 3
1 p 1 p 1 p
= t t ln t + t t + C = t t (1 3 ln t) + C
3 9 9
Now we can return to variable x if necessary.
In the previous examples we used to add the missing coefficients and cancel
them out before the integral. In many cases it is necessary to add and subtract
additional terms in order to split the integral in several parts. When the integrand
is a rational function one can apply a general method of elementary fractions using
indeterminate coefficients. However, there are cases when a simple algebraic trick
can help.
 1. INDEFINITE INTEGRALS 8

Example 5.
⌧ d⌧
ˆ
p I=
t ⌧
Clearly a term is missing to get t ⌧ in the numerator. Therefore, we add and
subtract t and also insert it under d taking care of the sign d (t ⌧ ) = d⌧ :

⌧ d⌧ t ⌧ t p d (t ⌧)
ˆ ˆ ˆ ˆ
I= p = p d (t ⌧) = t ⌧ d (t ⌧) t p =
t ⌧ t ⌧ t ⌧
2 3 1
= (t ⌧ ) 2 2t (t ⌧ ) 2 + C
3
After the second transition we got the integrand decomposed into two terms
each containing distinct components t ⌧ which allows to evaluate them again
without changing the variable.
Example 6.
(x + 3) dx
ˆ
I= p
4x2 + 4x 3
One of the most helpful tricks for decomposing expressions into manageable
blocks is separating a squared binomial. Looking at the expression under the radical
2
we see that the missing term to get (2x + 1) is 1. So we can add and subtract it:

(x + 3) dx (x + 3) dx (x + 3) dx
ˆ ˆ ˆ
I= p = p = q
4x2 + 4x 3 4x2 + 4x + 1 4 (2x + 1)
2
4
Now
h i
2
1 (8x + 4 + 20) dx
ˆ
1
ˆ d (2x + 1) 5
ˆ
dx
I= q = q + q =
8 2 4 2 2 2
(2x + 1) 4 2 (2x + 1) 4 (2x + 1) 4
h i
ˆ d (2x + 1)2 4
1 5 d (2x + 1)
ˆ
= q + q
4 2 4 2
2 (2x + 1) 4 (2x + 1) 4
The first integral is evaluated easily as the integrand is a power function f (t) =
p 0
1
p
2 t
= t . The second integral is evaluated using the Euler’s substitute which
we shall touch upon further. It is also included in the tables as a standard result.
Finally the answer is:

1p 2 5 p
4x + 4x 3 + ln 2x + 1 + 4x2 + 4x
I= 3 +C
4 4
Exercise 2. I = px2 +2x+5 dx
5x+4
´

Several techniques exist that are aimed at eliminating the radical in the inte-
grand. The general direction is to try to convert it in the form of a rational function.
Integrals of rational functions can be easily calculated using the routine method of
elementary fractions with indeterminate coefficients. It is also useful to eliminate
the linear term of the quadratic polynomial under the radical. In this form it is
 1. INDEFINITE INTEGRALS 9

often easily converted to one of the standard integrals. We shall consider some of
the tricks attributed to Euler and Abel that serve for these purposes.
Example 7.
dx
ˆ
I= p
x2 + a
This is a standard integral, however it is useful to work through its calculation.
Substitute

p
(1.0.5) z =x+ x2 + a

p
xdx x + x2 + a z
dz = dx + p = p dx = p dx
2
x +a 2
x +a 2
x +a
dx dz
p =
2
x +a z
ˆ
dz p
I= = ln |z| + C = ln x + x2 + a + C
z
The suggested change of variable leads to an elegant solution, however it is
not clear how it would follow from the structure of the given integrand. In fact
rewriting (1.0.5) in the form:
p
x2 + a = z x
We find express it as a particular case of on of the Euler’s substitutes applicable
to a class of integrals of the form:
ˆ ⇣ p ⌘
I= R x, ax2 + bx + c dx

where R is some rational function. These substitutes are the following ones:

p p
(1.0.6) ax2 + bx + c = t ± x a, a>0

p p
(1.0.7) ax2 + bx + c = tx ± c, c>0

p
(1.0.8) ax2 + bx + c = (x ↵) t,
where

ax2 + bx + c = a (x ↵) (x ), ↵, 2R
The purpose of the Euler’s substitutes is to express the radical in terms of the
new variable in such way that squaring and canceling out on both sides gives a
linear equation for the old variable in terms of the new variable.s
 1. INDEFINITE INTEGRALS 10

Example 8.
dx
I= p
(1 + x) 1 + x x2
Since c = 1 > 0 we can use (1.0.7):
p
1 + x x2 = tx + 1

x x2 = t2 x2 + 2tx

1 2t
x=
1 + t2
t2 t 1
dx = 2 dt
(1 + t2 )
p t2 t 1
1+x x2 = tx + 1 =
1 + t2
t2 2t
1+x=
1 + t2

1 + t2 1 + t 2 t 2 t 1 dt
ˆ ˆ
I= 2 2 2 2
dt = 2 2
=
t 2t t t 1 (1 + t ) t 2t
(t 2 t) dt dt d (t 2) t
ˆ ˆ ˆ
= 2
= = ln +C
t 2t t t 2 t 2
Returning to variable x:
p
1 + x x2 1
I = ln p +C
1 + x x2 2x 1

Exercise 3. I = p dx
´
1 x2 1

Another way to get rid of the radical in the integrand is to use the Abel’s
substitute. It applies to the integrals of the following forms:

dx dx
ˆ ˆ
m p , mp
(ax2 + bx + c) ax2 + bx + c (x2 + p) kx2 + l
The trick is to denote the derivative of the radical as a new variable. The
technique relies on the way the derivative of the radical is expressed in terms of
itself. Thus let, respectively:
⇣p ⌘0 ⇣p ⌘0
t= ax2 + bx + c , t= kx2 + l

Example 9.
dx
ˆ
I= p
(x2
+ 2) x2 + 1
⇣p ⌘0 x
t= x2 + 1 = p
x2 + 1
p
t x2 + 1 = x
 1. INDEFINITE INTEGRALS 11

Differentiating both sides by x we obtain:

p
dt p 2 d x2 + 1
x +1+t =1
dx dx
Using the definition of t:

dt p 2
x + 1 + t2 = 1
dx
dx dt
p = 2
x2 + 1 t 1
t2
x2 =
1 t2
2 t2
x2 + 2 =
1 t2

1 t2 dt dt
1 dt
ˆ ˆ ˆ
I= = = 2 =
2 t2 1 t2 2 t 2 2 1 t2
! p 0 ˆ ⇣ ⌘ ⇣ ⌘1
1
ˆ
dt
ˆ
dt 2@ d 1 pt2 ˆ d 1 + pt
2
= + = + A=
4 1 pt2 1 + pt2 4 1 pt2 1 + pt2
p p
1 2+t 1 2x2 + 2 + x
= p ln p + C = p ln p +C
2 2 2 t 2 2 2x2 + 2 x

Exercise 4. I = dx
´
7
2 (2x x+2) 2

Often it is possible to express the integrand as a rational function of a linear

ratio ax+b
cx+d . In these cases it is convenient to let t = cx+d raising it to the appropriate
ax+b

power so as to eliminate the radicals.

Example 10.
dx
ˆ
I= p
(1 x) 1 x2
First we need to cast the integrand in the appropriate form:

dx dx
ˆ ˆ
I= q = q
1+x 2 2 1+x
(1 x) 1 x (1 x) (1 x) 1 x

Now let t2 = 1+x

1 x to obtain:

1 t2
x=
1 + t2
2t 1 + t2 1 t2 2t 4tdt
dx = 2 dt = 2
(1 + t2 ) (1 + t2 )
2
1 x=
(1 + t2 )
 1. INDEFINITE INTEGRALS 12

2
1 + t2 4t dt
ˆ ˆ
I= 2 t = dt = t + C =
t 2
(1 + t )
r
1+x
= +C
1 x
dx dx
ˆ ˆ
I= q = q
2
(1 x) 1 x
1+x (1 + x) (1 x2 ) 11+xx
1 x
t2 = 1+x

1 t2
x=
1 + t2
2t 1 + t2 2t 1 t2 4tdt
dx = 2 dt = 2
(1 + t2 ) (1 + t2 )
2 2
1 + t2 1 t2 4t2
1 x2 = 2 = 2
(1 + t2 ) (1 + t2 )
2 r
1 + t2 4t dt dt 1 1+x
ˆ ˆ
I= 2 t = = +C = +C
4t2 2
(1 + t ) t2 t 1 x

Exercise 5. I = p dx
´
3 2 4
(x+1) (x 1)

We have already mentioned the use of casting the quadratic polynomials in the
form of squared binomials (here assuming a > 0):
✓ ◆
b b2 c b2
ax2 + bx + c = a x2 + 2 x + 2 + =
2a 4a a 4a2
✓ ◆2
b b2 4ac
=a x+
2a 2a
In p
general it is possible to transform the integrand being a rational function of
x and ax2 + bx + c to one of the following, after appropriate scaling:
ˆ ⇣ p ⌘
(1.0.9) R t, t2 + 1 dt

ˆ ⇣ p ⌘
(1.0.10) R t, t2 1 dt

ˆ ⇣ p ⌘
(1.0.11) R t, 1 t2 dt

The first type of these is often rationalized by a hyperbolic substitute t = sinh z

which takes advantage of the identity:

cosh2 z sinh2 z = 1
 1. INDEFINITE INTEGRALS 13

thereby removing the radical. A general note here is that any integral that
can be cast to a simple form and calculated using a hyperbolic substitute can be
with equal success handled using ordinary trigonometric (circle) functions. For the
above three cases it is appropriate to let respectively:

(1.0.12) t = tan z or t = sinh z

1
(1.0.13) t= or t=coshz
cos z

(1.0.14) t = sin z or t=tanhz

Example 11.
dt
ˆ
I= p
t2 t2 + 1

t = sinh z
dt = cosh zdz

cosh dz dz
ˆ ˆ
I= 2 = = coth z + C =
sinh z cosh z sinh z
p p
1 + sinh2 z 1 + t2
= +C = +C
sinh z t
Alternatively,
t = tan z
dz
dt =
cos2 z
cos z dz cos zdz d (sin z) 1
ˆ ˆ ˆ
I= = = = +C =
tan2 z cos2 z sin2 z sin2 z sin z
r p
p 1 1 + t2
2
= 1 + cot z + C = 1+ + C = +C
tan2 z t
´q 3
Exercise 6. I = (x2 1) dx
 CHAPTER 2

Definite Integrals

Evaluating definite integrals using changes of variable requires more care. In

this case we need to take into account the behavior of the substitute function on
a particular interval where the integral is being calculated. The following theorem
which we state without proof gives the sufficient conditions for a valid substitute.
Theorem. If function x = (t) satisfies the following conditions:
1) (t) is a continuous one-to-one function defined on the interval [↵, ] and
having a continuous derivative there.
2) values of (t) are contained within the interval [a, b]
3) (↵) = a and ( ) = b
then for every function f (x) continuous on [a, b] the following formula for the
change of variable in a definite integral holds:
ˆ b ˆ
f (x) dx = f [ (t)] 0 (t) dt
a ↵

Instead of the direct change x = (t) and inverse change t = (x) is used. In
calculations it is convenient to use monotonic continuously differentiable functions
in substitutes. Possible pitfalls shall be described in the following examples.
Example 12.
2
dx
ˆ
I=
2 4 + x2
Formal change of variable x = 1
t leads to a wrong answer:

1 1
2
dx dt dt
ˆ ˆ 2
ˆ 2
I= = 1 = =
2 4 + x2 1
2
t2 1+ t2
1
2
4t2+1
1
1 2
⇡
= arctan 2t =
2 1 4
2

which is impossible, because the integrand is strictly positive. It happens be-

cause the substitute function is not differentiable at the origin. The right answer
is, of course:

1 h⇡ ⇣ ⇡ ⌘i ⇡
2 2
dx 1 x
ˆ
I= = arctan = =
2 4 + x2 2 2 2 2 4 4 4

Another possible mistake resulting from inverse substitutes t = (t) can occur
if x = (t) is multi-valued.
14
 2. DEFINITE INTEGRALS 15

Example 13. ˆ 3
2
I= (x 2) dx
0
2
Formal application of the substitutept = (x 2) will result in a wrong answer,
because the
p inverse function x = 2 ± t is double-valued. One p of its branches
x1 = 2 t cannot take values x > 2 and the second one x2 = 2 + t cannot take
values x < 2. In order to get the correct answer it is necessary to split the integral
in two parts:
ˆ 2 ˆ 3
2 2
I= (x 2) dx + (x 2) dx
0 2
p p
Then let x = 2 t in the first one and x = 2 + t in the second one to obtain:
ˆ 2 ˆ 0
dt 1 4p 8
ˆ
2
I1 = (x 2) dx = t p = tdt =
0 4 2 t 2 0 3
ˆ 3 ˆ 1 ˆ 1p
2 dt 1 1
I2 = (x 2) dx = t p = tdt =
2 0 2 t 2 0 3
Therefore, I = I1 + I2 = 3 + 3 = 3. This result can be verified by direct
8 1

calculation:
ˆ 3 3 3
2 (x 2) 1 8
I= (x 2) d (x 2) = = + =3
0 3 3 3
0

It is possible to use functions that are not monotonic on the interval, however
care must be taken to choose the correct values of the new limits.
Example 14. p
3
dx
ˆ 2
I= p
1
2
x 1 x2
Use the substitute x = sin t which is notpmonotonic. New limits of integration
can be found from equations 12 = sin t and 23 = sin t. Possible values are t1 = ⇡6
and t2 =h ⇡3 , or it1 = 5⇡6 and t2 = 3 . In both cases x = sin t takes all values on the
2⇡
p ⇥ ⇤ ⇥ ⇤
interval 12 , 23 and is monotonic on both ⇡6 , ⇡3 and 5⇡ 6 , 3 . The result will be
2⇡

the same in both cases:

ˆ p23 ˆ ⇡3 ˆ ⇡3
dx cos tdt dt
I= p = = =
1
2
x 1 x 2 ⇡
6
sin t cos t ⇡
6
sin t
⇡ p
t 3 ⇡ ⇡ 2+ 3
= ln tan = ln tan ln tan = ln p
2 ⇡ 6 12 3
6
⇥ 5⇡ 2⇡ ⇤
On the other hand, taking into account that on 6 , 3 cos t is negative, we
obtain:
ˆ p23 ˆ 2⇡ ˆ 5⇡
dx 3 cos tdt 6 dt
I= p = = =
1
2
x 1 x 2 5⇡
6
sin t ( cos t) 2⇡
3
sin t
5⇡ p
t 6 5⇡ ⇡ 2+ 3
= ln tan = ln tan ln tan = ln p
2 2⇡ 12 3 3
3
 2. DEFINITE INTEGRALS 16

Note that we may not take the limits to be t1 = 5⇡ 6 andh t2 p

= i⇡3 , because as t
⇥ ⇡ 5⇡ ⇤
varies on 3 , 6 the values of x = sin t exceed the limits of 12 , 23 .
´ 2⇡
Exercise 7. Consider the following integral: I = 0 5 2dxcos x . Substituting
x = tan 2t we obtain:
ˆ 0
2dt
I= ⇣ ⌘ =0
2
0 (1 + t2 ) 5 3 11+tt2
What is wrong here?
Example 15. Choosing the correct sign is important as it always is when
dealing with radicals. ˆ 0
dx
I= p
1 2x 3 3x2 + 1
2
We begin by factorizing the polynomial under the radical:

2x3 3x2 + 1 = 2x3 2x2 x2 + 1 = 2x2 (x 1) (x 1)(x + 1) =

2 2
= (x 1)(2x x 1) = (x 1)(x x + x2 1) =
= (x 1)(x(x 1) + (x 1)(x + 1)) = (x 1)(x 1)(2x + 1) =
2
= (x 1) (2x + 1)

0
dx
ˆ
p I=
(x 1)2 (2x + 1) 1
2

For 12 < x < 0 we have 32 < x 1 < 1 and 0 < 2x + 1 < 1 we need to
choose the negative sign when taking a square root of the quadratic term. Inserting
the radical under d we obtain:
ˆ 0 ˆ 0 p
dx d( 2x + 1)
I= p =
1 2x + 1(1 x) 1 1 x
p 2 2

Now t = 2x + 1 seems an appropriate substitute:

1 1 1
dt dt 2 dt
ˆ ˆ ˆ
I= t2 1
= 3 t2
= t2
=
1
0 2 0 2 2
3 0 1 3
p ˆ 1 d( pt3 ) p 1 t
1 p 1 1
=2 3 = 2 3 tanh p = 2 3 tanh p
0 1 ( pt3 )2 3 0 3
Changing the integration variable in conjunction with some trigonometric iden-
tities can simplify calculations in certain cases.
ˆ ⇡
d
I=
⇡ 1
2
✏ cos2
In this case we can use the complimentary formula sin ⇡
2 = cos followed
by expressing the sine in terms of the tangent of a half-angle:
2 tan 2
sin =
1 + tan2 2
 2. DEFINITE INTEGRALS 17

Then let:
du
u = tan , d =
2 1 + u2
⇡ ⇡
⇡ ⇡
d d d
ˆ ˆ ˆ 2
2
I= = = d =
⇡
2
1 ✏ cos2 ⇡
2
1 ✏ sin2 ⇡
2 0 1 ✏ sin2
1 ˆ 1 1
p
du du 1 d u 1 ✏
ˆ ˆ
= ⇣ ⇣ ⌘⌘ = =p p 2 =
0 2
(1 + u ) 1 ✏ 1 1 0 1 + (1 ✏) u2 1 ✏ 0 1+ u 1 ✏
1+u2
1 p 1 ⇡ 1
=p arctan u 1 ✏ 0
= p
1 ✏ 2 1 ✏
Many integrals that cannot be expressed in terms of the so-called elementary
functions (essentially rational functions, radicals, exponent and their combinations)
give rise to new functions. Some of these are of such importance in theoretical
studies and applications that they have to be considered and used in equal rights
with elementary functions. Some of the most prominent representative of these
special functions are the gamma-function (x) and its close relative beta-function
B (x, y). There is a huge number of identities involving these functions. We will
only concentrate on a few properties of the beta-function and its applications to
calculating integrals involving trigonometric functions.
One of the definitions of the beta-functions is given by the following integral:
ˆ 1
q 1
B (p, q) = xp 1 (1 x) dx
0
Alternative form can be obtained by substituting:
ˆ 1
tp 1
B (p, q) = p+q dt
0 (1 + t)
For applications to integrands involving trigonometric functions the following
useful representation is obtained by letting x = sin2 :
ˆ ⇡2
B (p, q) = 2 cos2q 1 sin2p 1 d
0
The relation that makes beta-function particularly useful for calculations is the
one that links in to gamma-function:
(p) (q)
B (p, q) =
(p + q)
where (x) can be represented as the following real integral:
ˆ 1
(x) = tx 1 e t dt
0
With the technique of casting the expressions in the most convenient form
to fit a certain pattern we can often get the result using only one of the above
representations.
Example 16. ˆ ⇡
2
I= sin4 ✓ cos2 ✓d✓
0
 2. DEFINITE INTEGRALS 18

In the same fashion as we used to add and subtract artificial terms to split the
integrand in several parts we can pull out the factors and add coefficients to achieve
the desired form of the expression:

⇡ ⇡
1
ˆ 2
ˆ 2
4
I= sin ✓ cos ✓d✓ = 2
sin3 ✓ cos ✓2 sin ✓ cos ✓d✓ =
0 2 0
⇡
1 1 1 3
ˆ 2 3 1
ˆ
1
2 2 2 2 2
= sin ✓ 1 sin ✓ d sin ✓ = t 2 (1 t) 2 dt =
2 0 2 0
1 ✓ ◆ 5 3
1 1 5 3 ⇡
ˆ
5 3
1 1 2 2
= t2 (1 t) 2 dt = B , = =
2 0 2 2 2 2 (4) 32
⇡
Exercise 8. I = sin↵ 1
´
0
2
d
We conclude this series of examples by demonstrating how several different
approaches can be applied to the same. The actual choice may be dictated by a
wider context of the calculations, ease of manipulations, or just personal preference.
Then we shall look at changing of variables in a slightly different aspect.
Example 17.
1
x3 dx
ˆ
p I=
0 1 x2
This integral can be calculated using a trigonometric substitute which is left to
the reader. Instead we shall handle it by factoring out the appropriate power of x:

1 ˆ 1 2
x3 dx x xdx 1 1 x2 d(x2 ) 1 1 tdt
ˆ ˆ ˆ
I= p = p = p = p =
0 1 x2 0 1 x2 2 0 1 x2 2 0 1 t
1 1 (1 t 1)dt 1 1 dt 1 1p
ˆ ˆ ˆ
= p = p 1 tdt =
2 0 1 t 2 0 1 t 2 0
p 1 1 2
= 1 t|10 + (1 t)3/2 |10 = 1 =
3 3 3
Another path would be using integration by parts:

ˆ
x3 dx 1 ˆ 1 2
x xdx
ˆ 1 p
I= p = p = x2 d( 1 x2 ) =
0 1 x2 0 1 x2 0
p ˆ 1p
2 2
= x2 1 x2 |10 + 1 x2 2xdx = (1 x2 )3/2 |10 =
0 3 3
Finally we can combine rational and hyperbolic substitutes to arrive at the
same result:

1
x=
t
1
dt
ˆ
I= p
1 t4 t2 1

t = cosh z
 2. DEFINITE INTEGRALS 19

1 ˆ 1 ˆ 1
dt sinh zdz dz
ˆ
I= p = 4
= 4
=
cosh 1 1 (cosh z) sinh z cosh 1 1 (cosh z)
4
t t 2 1
1
ˆ 1 ˆ 1 ˆ 1
((cosh z)2 (sinh z)2 )dz dz (tanh z)2 dz
= = =
cosh 1 1 (cosh z)4 cosh 1 1 (cosh z)
2
cosh 1 1 (cosh z)
2
ˆ 1 ˆ 1 1
1 (tanh z)3
= d (tanh z) (tanh z)2 d (tanh z) = tanh z|cosh 1 1 =
cosh 1 1 cosh 1 1 3 cosh 1 1
1 2
=1 =
3 3
We finish this chapter with a brief look at the method of defining functions as
inversions of integrals. In fact the functions that we shall consider were the first
example of such approach. These are called lemniscate functions. They arise in
the process of calculating arc length of the 4th order algebraic curve called the
lemniscate which is given by the following equation:
2
x2 + y 2 = a 2 x2 y2
or in polar coordinates:

⇢2 = a2 cos 2
Lemniscate functions are defined in terms of the following integrals:
ˆ sinlemn ˆ 1
dt dt
= p = p
1 t 4 1 t4
0 coslemn
Similarity of the notation to the circle functions sin and cos is not accidental.
Polar equation:
⇢ = asinlemn
describes a lemniscate (⇢ = a sin gives a circle). Let us see facts about these
functions can be derived using only the above representations. Differentiating both
sides of the first integral by we obtain:
1
1 = sinlemn0 p
1 sinlemn2
sinlemn02 = 1 sinlemn4
Differentiating by again:
2sinlemn0 sinlemn00 = 4sinlemn3 sinlemn0
sinlemn00 = 2sinlemn3
Hence we obtain the differential equation for the lemniscate functions:
y 00 = 2y 3
which could be equally obtained from the second integral.
Now we shall find an algebraic relation between lemniscate functions. Rewrite
the integral defining coslemn as follows:
ˆ 1 ˆ 1
dt dt
= p = q
1 t 4 1 t 2
coslemn coslemn 2
2 (1 + t )
1+t

Now substitute:
 2. DEFINITE INTEGRALS 20

1 z2
t2 =
1 + z2
2z(1 + z 2 ) 2z(1 z2)
2tdt = dz
(1 + z 2 )2
2z 3 dx
tdt =
(1 + z 2 )2
r
1 + z 2 2z 3 dz
dt =
1 z 2 (1 + z 2 )2
Inserting these expressions in the integral and observing the new limits we
obtain:
r
0
r 1 coslemn2
2 3
1 (1 + z ) 1+ z2 2z dz dz
ˆ ˆ
1+coslemn2
= r = q =
1 coslemn2 z 2z 2 1 z 2 (1 + z 2 )2 1 z2
+ z2)
1+z 2 (1
1+coslemn2
0
r r
1 coslemn2 ˆ 1 coslemn22
dz dt
ˆ
1+coslemn2 1+coslemn
= p = p
1 z 4 1 t4
0 0
Comparing the last result with the definition of sinlemn we deduce:

1 coslemn2
sinlemn2 =
1 + coslemn2
Conversely

1 sinlemn2
coslemn2 =
1 + sinlemn2
Now we can calculate the derivatives of the lemniscate functions:
s
q
0 4 1 sinlemn2
sinlemn = 1 sinlemn = (1+sinlemn2 ) = sinlemn2 (1+sinlemn2 )
1 + sinlemn2
Formula for coslemn can be deduced in a similar manner, however we shall first
establish one more relation. Rewriting the definitions in the following form

x 1
dt dt
ˆ ˆ
1 1
= sinlemn x= p = coslemn x= p
0 1 t4 x 1 t4
And observing that
1
dt
ˆ
1 1
+ = sinlemn x + coslemn x= p =C
0 1 t4
inverting the integrals

x = sinlemn

x = coslemn ¯
 2. DEFINITE INTEGRALS 21

we deduce the following complimentary formula:

sinlemn = coslemn(C )
and hence obtain the derivative of coslemn:

coslemn0 = sinlemn (1 + coslemn2 )

Constant C can be calculated using beta-function mentioned in the previous
section with the substitute u = 1 t4 :

0
1 1 1
ˆ ˆ
1 3 1 3
C= u 2 (1 u) 4 du =
u 2 (1 u) 4 du =
4 1 4 0
1 1
1 1 1 1 (2) (4) [ ( 1 )]2
= B( , ) = 3 = p 4p
4 2 4 4 (4) 4 2 ⇡
Finally we can calculate the indefinite integral of sinlemn:

tdt 1 d(t2 )
ˆ ˆ ˆ ˆ
1
sinlemn d = td(sinlemn t) = p = p =
1 t 4 2 1 (t2 )2 )
arcsin t2 arcsin sinlemn2
= +C = +C
2 2
Exercise 9. Deduce the identities
sin2 + cos2 =1
⇡
sin = cos 2
0
(sin ) = cos
0
(cos ) = sin
using the following definitions:
ˆ sin ˆ 1
dt dt
= p = p
1 t 2 1 t2
0 cos
 CHAPTER 3

Differential Equations

Although our main focus has remained on calculating integrals, similar tech-
niques often work well for certain types of integrable ordinary differential equations.
We shall touch upon two aspects of manipulating expressions in differential equa-
tions. Changes of dependent, or dependent variable, or both also often help to cast
the equation into a simple form, or separate variables. Another useful technique
which demonstrates the use of the method of inserting expressions under d is known
as the method of integrating factor.
Example 18. Find the general solution of the following equation.
a
y=p
1 + y 02
The form of the denominator suggest that a trigonometric substitute might
work here. Indeed:
y 0 = tan
(3.0.15) y = a cos
dy sin
dx = 0 = a d = a cos d
y tan
Integration gives:
(3.0.16) x + C = a sin
In fact we have already obtained the solution in the parametric form. Squaring
and adding (3.0.15) and (3.0.16):
2
(x + C) + y 2 = a2
p dy
3
Exercise 10. (y x) 1 + x2 dx = 1 + y2 2

The general method of solving equations

M (x, y) dx + N (x, y) dy = 0
using the integrating multiplier requires finding a function µ which when multiplied
by the left-hand side turns it to an exact differential:
du = µM dx + µN dy
Since in this representation:
@u @u
= µM = µN
@x @x
equality of the mixed derivatives requires that:
@ @
(µM ) = (µN )
@y @x
22
 3. DIFFERENTIAL EQUATIONS 23

which leads to a partial differential equation for µ:

@ ln µ @ ln µ @M @N
N M =
@x @y @y @x
It can be solved in particular whenµ is a function of only one variable, say x:
@M @N
@ ln µ @y @x
=
@x N
The latter equation can be solved if the right-hand side is a function only of x.
In some cases, however, by inspection of the given equation it is possible to cast it
into the desired form without reverting to the general method.
Example 19.
1 x2 y dx + x2 (y x) dy = 0
dx + x2 ydy x3 dy x2 ydx = 0
✓ ◆
2 dx
x + ydy x2 (xdy + ydx) = 0
x2
✓ ◆
1 y2
d + d (xy) = 0
x 2
✓ ◆
1 y2
d + xy = 0
x 2
1 y2
+ xy = C
x 2
Even if the integrating factor is a function of both variables it is possible to de-
termine it by using formulae of the derivative of a product and ratio in a backwards
manner.
Example 20.
xdx + ydy + x (xdy ydx) = 0
1 xdy ydx
d x2 + y 2 + x3 =0
2 x2
1 ⇣y⌘
d x2 + y 2 + x3 d =0
2 x
We need to absorb x3 to arrive at an exact differential. Moreover, the inte-
grating factor should be a function of x2 + y 2 which is homogeneous in x and y, so
should work well for the second term as well. Combining these considerations we
3
see that µ (x, y) = x2 + y 2 2 should be an appropriate choice:
y
1 d x2 + y 2 d x
+⇣ ⌘ 32 = 0
2 (x2 + y 2 ) 32 y2
1+ x2

y
1 d x2 + y 2 d
ˆ ˆ
x
3 + ⇣ ⌘ 32 = 0
2 (x2 + y 2 ) 2 y2
1+ x2
 3. DIFFERENTIAL EQUATIONS 24

The second integral can be transformed as follows::

1
dt dt dt
ˆ ˆ ˆ
t
3 = 3 = 3 2
=
2
(1 + t ) 2 1
t3 1 + t 2 2 1 + t2 2 t
1
✓ ◆
1
1 1 d 1 + t12
ˆ ˆ
t
= 3 d = 3 =
1 + t12 2 t 2 1 + t12 2
1 t
= 1 = p
1 + 12 2 1 + t2
t
Finally, the general solution is:
1 y
p +p =C
2
x +y 2 x + y2
2

y 1
p =C
x2 + y 2
Exercise 11. x2 + y 2 + 1 dx 2xydy = 0
There is a simple fact about the integrating factor which leads to a very useful
practical trick making its application easier. Like other methods discussed above
this one is aimed at breaking down the original expression into blocks for each
of which the solution can be made clear. The fact is that there exist classes of
integrating factors. Indeed, let µ be an integrating factor of the equation
(3.0.17) M dx + N dy = 0
and U (x, y) = C be its general solution. Let be an arbitrary differentiable
function. Then
(U ) µ (M dx + N dy) = (U ) dU = d
where ˆ
(U ) = (U ) dU
Therefore µ (U ) is also an integrating factor. Let µ1 be another integrating
factor of the same equation. Then
dU = µ (M dx + N dy) = µM dx + µN dy
dV = µ1 (M dx + N dy) = µM dx + µN dy
Therefore
@U @U @V @V
µM = µN = µ1 M = µ1 N =
@x @y @x @y
thus
@U @U
@x @y
@V @V =0
@x @y
Hence, U and V are functionally dependent:
V = (U )
µ1 (M dx + N dy) = dV = (U ) dU = 0 (U ) µ (M dx + N dy)
0

Therefore each integrating factor of (3.0.17) has the form (U ) µ. It follows that
if we find two independent integrating factors that differ more than by a constant
multiplier, then their ratio gives the general solution of (3.0.17). We shall now
 3. DIFFERENTIAL EQUATIONS 25

demonstrate how to use the general form of the integrating factor in calculations.
The trick is to split the terms in parts for each of which the integrating factor can
be easily found. Then write out the most general forms for all of them and by
manipulating the arbitrary functions try to cast them to the same form.
Example 21.
(x2 y 2 1)dy + 2xy 3 dx = 0
We begin by rearranging the terms:

(x2 y 2 dy + 2xy 3 dx) dy = 0

The second term has a trivial integrating factor µ = 1 leading to general so-
lution y = C. Hence the most general integrating factor will be µ1 = 1 (y). For
the first term the integrating factor will obviously be x21y3 since it separates the
variables:
dy dx
+2 =0
y x
x2 y = C
Hence, the most general factor will have the form µ2 = x21y3 1 x2 y . We want
µ2 to be independent of x, hence we let 1 (t) = t which gives µ2 = y12 . This is the
integrating factor that suits both parts:

dy
(x2 dy + 2xydx) =0
y2
1
x2 y + =C
y
Exercise 12. x y 2 dx + 2xydy = 0
We conclude this chapter by demonstrating another example of using the prop-
erties of d as operator. This is a quite simple problem, but instead of following the
standard methods we shall use a slightly unconventional notation
Example 22.
dy 2
=1+
dx x+y
Add dx
dx = 1 to both sides and use linearity:
dy dx 2
+ =2+
dx dx x+y
d(x + y) 1 + (x + y)
=2
dx x+y
Separate new variables x and x + y

(x + y)d(x + y)
= 2dx
1 + (x + y)
Add artificial terms:

[ 1 + 1 + (x + y)] d(x + y)
= 2dx
1 + (x + y)
 3. DIFFERENTIAL EQUATIONS 26

d(1 + x + y)
+ d (x + y) = 2dx
1 + (x + y)
d(1 + x + y)
= d(y x)
1 + (x + y)
ln |1 + x + y| = (y x) + ln C

1 + x + y = Ce(y x)
 Bibliography

[1] Maron I.A. Differential and Integral Calculus in Examples and Exercises. Moscow: Glavnaya
Redaktsiya Fiziko-Matematicheskoy Literatury, 1970.
[2] Kiselev A.I., Krasnov M.L., Makarenko G.I. Problems in Ordinary Differential Equations.
Moscow: Vyshaya Shkola, 2nd ed., 1967.
[3] Kolmogorov A.N., Fomin S.V. Elements of the Theory of Functions and Functional Analysis.
Moscow: Glavnaya Redaktsiya Fiziko-Matematicheskoy Literatury, 1968.
[4] Kolobov A.N., Cherenkova L.P., Selected Chapters in Higher Mathematics. Special Courses.
Special Functions. Methods of Mathematical Physics. Minsk: Vysheyshaya Shkola, 1967.
[5] Stepanov V.V. A Course of Differential Equations. Leningrad: Gosudarstvennoe Izdatelstvo
Tekhniko-Teoreticheskoy Literatury, 1950.
[6] Whittakker E.T., Watson G.N., A Course of Modern Analysis. Cambridge: Cambridge Uni-
versity Press, 1920.

27