Notes on Riemann Integration 2

Let us solve a few problems

1) Using definition show that the floor function is R-integrable on [0,3]. Also find its
value.

Recall that floor function is defined as ⌊𝑥⌋ = greatest integer ≤ 𝑥 .

Define partition
1 2 𝑛 2𝑛 3𝑛
𝑃𝑛 = {0, 𝑛 , 𝑛 , … , 𝑛 , … , 𝑛 , … , 𝑛 }
Observe that
𝑀𝑘 = 𝑚𝑘 = 0 ∀ 𝑘 = 1,2, … , 𝑛 − 1
𝑀𝑘 = 𝑚𝑘 = 1 ∀ 𝑘 = 𝑛 + 1, … ,2𝑛 − 1
𝑀𝑘 = 𝑚𝑘 = 2 ∀ 𝑘 = 2𝑛 + 1, … ,3𝑛 − 1
𝑀𝑛 = 1, 𝑚𝑛 = 0, 𝑀2𝑛 = 2, 𝑚2𝑛 = 1,
𝑀3𝑛 = 3, 𝑚3𝑛 = 2,

Hence
3
𝑈(𝑃𝑛 , ⌊ ⌋) = 3 + 𝑛 ,
𝐿(𝑃𝑛 , ⌊ ⌋) = 3

But, as we know,
𝐿(𝑃𝑛 , ⌊ ⌋) ≤ 𝐿(⌊ ⌋) ≤ 𝑈(⌊ ⌋) ≤ 𝑈(𝑃𝑛 , ⌊ ⌋)
3
⇒ 3 ≤ 𝐿(⌊ ⌋) ≤ 𝑈(⌊ ⌋) ≤ 3 +
𝑛
∀𝑛 ∈ℕ
Hence, we must have
𝐿(⌊ ⌋) = 𝑈(⌊ ⌋) = 3

i.e., floor function is R-integrable on [0,3] and the value of the integral is 3

2) If 𝑓: [𝑎, 𝑏] → ℝ is a bounded function for which it is possible to find a sequence

(𝑃𝑛 ) of partitions of [𝑎, 𝑏] such that (𝑈(𝑃𝑛 , 𝑓) − 𝐿(𝑃𝑛 , 𝑓)) converges to 0, then
𝑏
prove that 𝑓 is R-integrable and ∫𝑎 𝑓 can be obtained by calculating limit of any of
the sequences (𝑈(𝑃𝑛 , 𝑓)), (𝐿(𝑃𝑛 , 𝑓)).

Suppose it is possible to find such a sequence of partitions.

Now, for every 𝜀 > 0, ∃ 𝑛0 ∈ ℕ such that |𝑈(𝑃𝑛 , 𝑓) − 𝐿(𝑃𝑛 , 𝑓) − 0| < 𝜀 ∀ 𝑛 ≥ 𝑛0 .
In particular taking 𝑛 = 𝑛0 there is a partition 𝑃𝑛0 for which Riemann’s criterion is
satisfied. Hence 𝑓 is R-integrable on [𝑎, 𝑏].
Moreover, since 𝐿(𝑃𝑛 , 𝑓) ≤ 𝐿(𝑓) = 𝑈(𝑓) ≤ 𝑈(𝑃𝑛 , 𝑓) and (𝑈(𝑃𝑛 , 𝑓) − 𝐿(𝑃𝑛 , 𝑓))
converges to zero, we must have
𝑏
lim 𝐿(𝑃𝑛 , 𝑓) = lim 𝑈(𝑃𝑛 , 𝑓) = 𝐿(𝑓) = 𝑈(𝑓) = ∫𝑎 𝑓
𝑛→∞ 𝑛→∞
 More rigorously, observe that both (𝑈(𝑃𝑛 , 𝑓)), (𝐿(𝑃𝑛 , 𝑓)) are bounded sequences
and as such they contain convergent subsequences. We may pass on to such
subsequence if needed. Then we can apply sandwich theorem.

3) Using Riemann’s criterion, show that the function 𝑓(𝑥) = 𝑥 3 is R-integrable on

[0,2]. Also find its value.

Let 𝜀 > 0. We will find a partition to show that Riemann’s criterion is satisfied.
2 4 2𝑛
We will take 𝑃𝑛 = {0, 𝑛 , 𝑛 , … , 𝑛 = 2}. We also observe that the function is strictly
increasing, hence
2𝑘 2𝑘−2
𝑀𝑘 = 𝑓(𝑥𝑘 ) = 𝑓 ( 𝑛 ) , 𝑚𝑘 = 𝑓(𝑥𝑘−1 ) = 𝑓 ( 𝑛 ) ∀ 𝑘 = 1,2, … , 𝑛. Thus,
𝑈(𝑃𝑛 , 𝑓) − 𝐿(𝑃𝑛 , 𝑓) = ∑𝑛𝑘=1 𝑀𝑘 ∆𝑘 − ∑𝑛𝑘=1 𝑚𝑘 ∆𝑘 = ∆𝑘 ∑𝑛𝑘=1(𝑀𝑘 − 𝑚𝑘 )
2
= (𝑓(𝑥1 ) − 𝑓(𝑥0 ) + 𝑓(𝑥2 ) − 𝑓(𝑥1 ) + ⋯ + 𝑓(𝑥𝑛 ) − 𝑓(𝑥𝑛−1 ))
𝑛
2 2(23 − 03 ) 16
= (𝑓(𝑥𝑛 ) − 𝑓(𝑥0 )) = = →0
𝑛 𝑛 𝑛
16
As such by selecting 𝑛 > 𝜀 we can show 𝑈(𝑃𝑛 , 𝑓) − 𝐿(𝑃𝑛 , 𝑓) < 𝜀
Thus 𝑓 satisfies Riemann’s criterion and it is R-integrable. To calculate the integral,
we need to find lim 𝑈(𝑃𝑛 , 𝑓).
𝑛→∞
2 2𝑘 3 16 16 𝑛2 (𝑛+1)2
Now, 𝑈(𝑃𝑛 , 𝑓) = ∑𝑛𝑘=1 𝑀𝑘 ∆𝑘 = 𝑛 ∑𝑛𝑘=1 ( 𝑛 ) = 𝑛3 ∑𝑛𝑘=1 𝑘 3 = 𝑛4 × →4
4
𝑏
Hence ∫𝑎 𝑓 = 4.

Result 8: If value of a R-integrable function is changed at finitely many points, then

its R-integrability remains unaffected.

Proof: Let 𝑓: [𝑎, 𝑏] → ℝ be R-integrable.

Let 𝑐 ∈ [𝑎, 𝑏] and g: [𝑎, 𝑏] → ℝ be defined as
𝑓(𝑥) ,𝑥 ≠ 𝑐
𝑔(𝑥) = { where, 𝑑 ≠ 𝑓(𝑐)
𝑑 ,𝑥 = 𝑐
𝑏 𝑏
We will prove that 𝑔 is R-integrable on [𝑎, 𝑏] and ∫𝑎 𝑔 = ∫𝑎 𝑓

Assume first that 𝑐 ∈ (𝑎, 𝑏) and let 𝜀 > 0.

There exists a partition of [𝑎, 𝑏], say
𝑃 = {𝑎 = 𝑎0 , 𝑎1 , … , 𝑐 = 𝑎𝑖 , … , 𝑎𝑛 = 𝑏} such that
𝜀
𝑈(𝑃, 𝑓) − 𝐿(𝑃, 𝑓) <
2
We have assumed w.l.g. that 𝑐 = 𝑎𝑖 is a point in 𝑃 .
We choose m∈ ℕ sufficiently large, so that
1 1
(i) 𝑎𝑖−1 < 𝑐 − 𝑚 < 𝑎𝑖 = 𝑐 < 𝑐 + 𝑚 < 𝑎𝑖+1
2 sup{|𝑓(𝑥) − 𝑔(𝑥)| | 𝑥 ∈ [𝑎, 𝑏]} 𝜀
(ii) <
𝑚 4
By Archimedean property it is always possible to choose such a natural number.
Now, let
1 1
𝑄 = {𝑎0 , 𝑎1 , … , 𝑎𝑖−1 , 𝑐 − 𝑚 , 𝑐, 𝑐 + 𝑚 , … , 𝑎𝑛 = 𝑏}.
It is a refinement of 𝑃.
 𝜀
Hence, 𝑈(𝑄, 𝑓) − 𝐿(𝑄, 𝑓) <
2
We would like to compute 𝑈(𝑄, 𝑔) − 𝐿(𝑄, 𝑔)
Now, 𝑈(𝑄, 𝑔) and 𝑈(𝑄, 𝑓) may differ only due to the fact that 𝑔(𝑐) = 𝑑 ≠ 𝑓(𝑐)
1
and 𝑐 is a member of only 2 subintervals, each of length 𝑛 . The difference in
2 sup{|𝑓(𝑥) − 𝑔(𝑥)||𝑥 ∈ [𝑎, 𝑏]}
𝑈(𝑄, 𝑔) and 𝑈(𝑄, 𝑓) must therefore be ≤ .
𝑚
𝜀 𝜀
Thus, we must have 𝑈(𝑄, 𝑓) − 4 ≤ 𝑈(𝑄, 𝑔) ≤ 𝑈(𝑄, 𝑓) + 4
Similarly, the difference in 𝑈(𝑄, 𝑔) and 𝑈(𝑄, 𝑓) must be
2 sup{|𝑓(𝑥) − 𝑔(𝑥)||𝑥 ∈ [𝑎, 𝑏]}
≤ and, we must have
𝑚
𝜀 𝜀
𝐿(𝑄, 𝑓) − 4 ≤ 𝐿(𝑄, 𝑔) ≤ 𝐿(𝑄, 𝑓) + 4
𝜀 𝜀 𝜀 𝜀 𝜀
Hence 𝑈(𝑄, 𝑔) − 𝐿(𝑄, 𝑔) ≤ 𝑈(𝑄, 𝑓) − 𝐿(𝑄, 𝑓) + 4 + 4 < 2 + 4 + 4 = 𝜀
Therefore, 𝑔 is integrable on [𝑎, 𝑏]
Now, on each m∈ ℕ, sufficiently large as above, we have
𝑏 𝑏 𝑏 𝑏
|∫ 𝑓 − ∫ 𝑔| = |∫ 𝑓 − 𝑔| ≤ ∫ |𝑓 − 𝑔|
𝑎 𝑎 𝑎 𝑎
1 1
𝑐− 𝑐+ 𝑏
𝑛 𝑚
=∫ |𝑓 − 𝑔| + ∫ |𝑓 − 𝑔| + ∫ |𝑓 − 𝑔|
1 1
𝑎 𝑐− 𝑐+
𝑛 𝑚
1
𝑐+
𝑚
= 0+∫ |𝑓 − 𝑔| + 0
1
𝑐−
𝑚
1
𝑐+
𝑚
≤∫ sup{|𝑓(𝑥) − 𝑔(𝑥)||𝑥 ∈ [𝑎, 𝑏]}
1
𝑐−
𝑚
1
𝑐+
𝑚
= sup{|𝑓(𝑥) − 𝑔(𝑥)||𝑥 ∈ [𝑎, 𝑏]} ∫ 1
1
𝑐−
𝑚
2
= sup{|𝑓(𝑥) − 𝑔(𝑥)||𝑥 ∈ [𝑎, 𝑏]} → 0
𝑚
𝑏 𝑏
Hence, |∫𝑎 𝑓 − ∫𝑎 𝑔| = 0
𝑏 𝑏
i.e., ∫𝑎 𝑓 = ∫𝑎 𝑔
In case 𝑐 = 𝑎 or 𝑐 = 𝑏, the argument is even simpler.
In the argument so far, the functions differ in only one point. But for finitely many
points, the argument can be used repeatedly, proving the required result.

Result 9: If 𝑓: [𝑎, 𝑏] → ℝ is a bounded function which is continuous except at

finitely many points, then it is R-integrable.

Outline for a Proof: Let 𝑥0 < 𝑥1 <, … , < 𝑥𝑛 be all points of discontinuity of 𝑓.
First, consider the case 𝑎 = 𝑥1 , 𝑏 = 𝑥𝑛
In each of the interval [𝑥𝑖−1 , 𝑥𝑖 ], define a function 𝑔𝑖 as
𝑓(𝑥) , 𝑥𝑖−1 < 𝑥 < 𝑥𝑖
lim 𝑓(𝑥) , 𝑥 = 𝑥𝑖−1
𝑔𝑖 (𝑥) = { 𝑥→𝑥𝑖−1 +
lim 𝑓(𝑥) , 𝑥 = 𝑥𝑖
𝑥→𝑥𝑖 −
Then it is continuous and hence it is R-integrable, but then 𝑓 is obtained by making
changes at most two points, hence 𝑓 is R-integrable on each [𝑥𝑖−1 , 𝑥𝑖 ].
Can you prove that 𝑓 is R-integrable on [𝑎, 𝑏]? This is somewhat like proving
converse of property (vi).
Try for an argument.

We present here, an alternate proof.

Let 𝑓: [𝑎, 𝑏] → ℝ is a bounded function which is continuous except at finitely many
points.
Let these points be 𝑥1 <, … , < 𝑥𝑛 and suppose all these points are within (𝑎, 𝑏).
We will use the Riemann’s criterion.
𝜀 𝑥 −𝑎 𝑥 −𝑥 𝑥 −𝑥 𝑥 −𝑥 𝑏−𝑥
Let 𝜀 > 0, Take 𝛿 < 4𝑛(𝑀−𝑚) , 12 , 2 2 1 , 3 2 2 , … , 𝑛 2 𝑛−1 , 2 𝑛 where,
𝑀 = sup{𝑓(𝑥)|𝑥 ∈ [𝑎, 𝑏]} , 𝑚 = inf{𝑓(𝑥)|𝑥 ∈ [𝑎, 𝑏]}

Now label the points

𝑎 = 𝑎0 , 𝑎1 = 𝑥1 − 𝛿, 𝑎2 = 𝑥1 + 𝛿, 𝑎3 = 𝑥2 − 𝛿,
𝑎4 = 𝑥2 + 𝛿, 𝑎5 = 𝑥3 − 𝛿, …,
𝑎2𝑛 = 𝑥𝑛 + 𝛿, 𝑎2𝑛+1 = 𝑏
This gives us a partition 𝑃 = {𝑎0 , 𝑎1 , … , 𝑎2𝑛+1 } of the interval [𝑎, 𝑏]. Refer to the
figure below.

(In this figure, 𝑛 = 3, 2𝑛 + 1 = 7)

Now, on [𝑎0 , 𝑎1 ], 𝑓 is continuous, hence it is R-ntegrable and we can find a

𝜀
partition 𝑃0 of [𝑎0 , 𝑎1 ] such that 𝑈(𝑃0 , 𝑓) − 𝐿(𝑃0 , 𝑓) < 2𝑛+2
𝜀
But we can also find a partition 𝑃1 of [𝑎2 , 𝑎3 ] such that 𝑈(𝑃1 , 𝑓) − 𝐿(𝑃1 , 𝑓) < 2𝑛+2 ,
and in fact for each 𝑘 = 1,2, … , 𝑛 we can find a partition 𝑃𝑘 of [𝑎2𝑘 , 𝑎2𝑘+1 ] such
𝜀
that 𝑈(𝑃𝑘 , 𝑓) − 𝐿(𝑃𝑘 , 𝑓) < 2𝑛+2
Also, let 𝑄𝑘 = {𝑎2𝑘−1 , 𝑎2𝑘 } be the partition of the subinterval ⌈𝑎2𝑘−1 , 𝑎2𝑘 ⌉ for each
𝑘 = 1,2, … , 𝑛.
Now, we consider all points in 𝑃0 , 𝑃1 , … 𝑃𝑛 together in ascending order and find a
partition 𝑄 of [𝑎, 𝑏] and consider 𝑈(𝑄, 𝑓) − 𝐿(𝑄, 𝑓).
This is a big summation which can be roughly broken into two parts, say Σ1 , Σ2
where,
Σ1 = summation over the subintervals [𝑎1 , 𝑎2 ], [𝑎3 , 𝑎4 ], [𝑎5 , 𝑎6 ], … [𝑎2𝑛−1 , 𝑎2𝑛 ],
i.e., Σ1 = ∑𝑛𝑘=1(𝑈(𝑄𝑘 , 𝑓) − 𝐿(𝑄𝑘 , 𝑓))
and Σ2 = ∑𝑛𝑘=0(𝑈(𝑃𝑘 , 𝑓) − 𝐿(𝑃𝑘 , 𝑓))
We can observe that each 𝑈(𝑄𝑘 , 𝑓) − 𝐿(𝑄𝑘 , 𝑓) ≤ 2𝛿(𝑀 − 𝑚) and hence,
Σ1 = ∑𝑛𝑘=1(𝑈(𝑄𝑘 , 𝑓) − 𝐿(𝑄𝑘 , 𝑓))
≤ ∑𝑛𝑘=1 2𝛿(𝑀 − 𝑚) < 2𝑛𝛿(𝑀 − 𝑚)
𝜀
<2
 On the other hand,
Σ2 = ∑𝑛𝑘=0(𝑈(𝑃𝑘 , 𝑓) − 𝐿(𝑃𝑘 , 𝑓))
𝜀 𝜀
< ∑𝑛𝑘=0 2𝑛+2 = 2
𝜀 𝜀
Hence 𝑈(𝑄, 𝑓) − 𝐿(𝑄, 𝑓) = Σ1 + Σ2 < 2 + 2 = 𝜀
Thus, by Riemann’s criterion 𝑓 is R-integrable over [𝑎, 𝑏] .
For finding the integral, we can find it over the intervals [𝑎, 𝑥1 ], [𝑥1 , 𝑥2 ], … , [𝑥𝑛 , 𝑏].
In each of these intervals we can find value of the integral by definition, or by
choosing a convenient sequence of partitions.

A quite interesting problem is …

4) Prove that the function 𝑓: [0,1] → ℝ given by

0 , 𝑥=0
1 1
1 , 𝑥∈[ , ]
𝑓(𝑥) = 2𝑛 2𝑛 − 1
1 1
−1 , 𝑥 ∈ ( , )
{ 2𝑛 + 1 2𝑛
is a R-integrable function.
(Observe that this function has infinitely many points of discontinuity, yet its R-
integrability can be checked by a technique similar to the one used above.)

1 𝜀
Let 𝜀 > 0. Take 𝑛 ∈ ℕ such that 𝑛 < 4 .
1 1 1 1 1
Consider a partition 𝑃𝑛 = {0, 2𝑛 , 2𝑛−1 , 2𝑛−2 , … , 3 , 2 , 1}.
Now choose 𝛿 > 0 such that
1 1 1 1
(i) +𝛿 < − 𝛿 ∈ (2𝑛 , 2𝑛−1) and
2𝑛 2𝑛−1
𝜀
(ii) 4𝑛𝛿 < 4
We now consider a partition
1 1 1 1 1 1 1 1
𝑄𝑛 = {0, 2𝑛 , 2𝑛 + 𝛿, 2𝑛−1 − 𝛿, 2𝑛−1 + 𝛿, 2𝑛−2 − 𝛿, … , 3 + 𝛿, 2 − 𝛿, 2 + 𝛿, 1 − 𝛿, 1}
and calculate 𝑈(𝑄𝑛 , 𝑓) − 𝐿(𝑄𝑛 , 𝑓).
1 1 1 1
On each subinterval of the type [2𝑘 + 𝛿, 2𝑘−1 − 𝛿] , [2𝑘−1 + 𝛿, 2𝑘−2 − 𝛿] , 𝑓 is
constant, so also it is constant on the last subinterval, viz [1 − 𝛿, 1]and on these
subintervals, supremum and infimum of 𝑓 are equal (either 1
Or −1 ) while on the other intervals, supremum is 1, infimum is −1. On calculation,
we see that 𝑈(𝑄𝑛 , 𝑓) − 𝐿(𝑄𝑛 , 𝑓)
1 1 1
< 2 (2𝑛 − 0) + 2 (2𝑛 + 𝛿 − 2𝑛) + 0 + 2(2𝛿) + 0 + 2(2𝛿) + ⋯ + 2(2𝛿) + 0
1
< 𝑛 + 8𝑛𝛿 < 𝜀
(Write a partition for small values of 𝑛 and verify this)
This proves that by Riemann’s criterion, the function is R-integrable.
Notice that even though there are infinitely many points where the function is
discontinuous, we have managed to put all but finitely many points of these in the
1
subinterval [0, 2𝑛]. Rest of the discontinuities are included each within a small
interval of length 2𝛿 .
 1
If one uses the idea that on [0, ] the function has many discontinuities but, on
2𝑛
1
the interval [ 2𝑛 , 1] it has only finitely many discontinuities so it is R-integrable on
1 1
[ 2𝑛 , 1], then for a given 𝜀 > 0 one can first find a partition 𝑃 of [ 2𝑛 , 1] having
𝜀 1
𝑈(𝑃, 𝑓) − 𝐿(𝑃, 𝑓) < 2 and then join this partition with the part [0, 2𝑛] to make a
partition 𝑄 of [0,1] . This would make the solution shorter.

Can we calculate the integral? Of course, by making 𝑛 very large and 𝛿 very small,
we can imagine that the integral would be nothing but sum of the series
1 1 1 1 1
− + − + −⋯
2 3 4 5 6

5) 𝑏
If 𝑓: [𝑎, 𝑏] → ℝ is continuous non negative function, then prove that ∫𝑎 𝑓 ≥ 0 and
𝑏
∫𝑎 𝑓 = 0 if and only if 𝑓(𝑥) = 0, ∀ 𝑥 ∈ [𝑎, 𝑏] i.e., 𝑓 vanishes on the entire interval.

By continuity, the function is seen to be R-integrable, while for any partition, the
upper Riemann sum for that partition would be ≥ 0. So, the upper integral of the
𝑏
function would be ≥ 0. This proves ∫𝑎 𝑓 ≥ 0.
Now if the function vanishes on the entire interval then it is the constant zero
function, whose upper and lower Riemann sums remain zero w.r.t. any partition.
So the integral is zero.
On the other hand, suppose 𝑓 does not vanish at at least one point, say 𝑝 in the
interval. W.l.g. let this be other than the endpoints. Now, 𝑓(𝑝) > 0 , hence we can
find 𝛿 > 0 such that (𝑝 − 𝛿, 𝑝 + 𝛿) ⊆ [𝑎, 𝑏] and ∀ 𝑥 ∈ (𝑝 − 𝛿, 𝑝 + 𝛿),
𝑓(𝑝) 𝑓(𝑝)
|𝑓(𝑥) − 𝑓(𝑝)| < ⇒ 𝑓(𝑝) > ∀ 𝑥 ∈ (𝑝 − 𝛿, 𝑝 + 𝛿)
2 2
Now, we use properties of Riemann integrable functions to say that
𝑏 𝑝−𝛿 𝑝+𝛿 𝑏
∫ 𝑓=∫ 𝑓+∫ 𝑓+∫ 𝑓
𝑎 𝑎 𝑝−𝛿 𝑝+𝛿
The first and third integral on RHS are ≥ 0 while in the second integral, the
𝑓(𝑝)
integrand takes values > 2 throughout (𝑝 − 𝛿, 𝑝 + 𝛿) and hence for any
𝑓(𝑝)
partition 𝑃 of [𝑝 − 𝛿, 𝑝 + 𝛿], we must have 𝐿(𝑃, 𝑓) ≥ 2𝛿 > 0 . Hence RHS > 0
2
𝑏
and so, ∫𝑎 𝑓 > 0 as required.

6) If the hypothesis of continuity of 𝑓 over the interval is dropped from the above
problem, then with the help of an example, show that it is possible to have
𝑏
∫𝑎 𝑓 = 0 even if 𝑓 does not vanish on the entire interval.

Several examples are there. Simplest perhaps is the identity function on [−1,1].
1 , 𝑥 ∈ [0,1]
Even simpler, is the function 𝑓(𝑥) = { .
−1 , 𝑥 ∈ [−1,0)

1
7) Evaluate ∫−1 𝑓 where, 𝑓(𝑥) = 𝑥 2 . (Exercise)
 𝜋
8) Evaluate ∫02 𝑓 where, 𝑓(𝑥) = sin 𝑥. (Exercise)
Hint: As the function is monotonic increasing, it is R-integrable. To find the integral,
we have to divide the interval into 𝑛 equal parts, take the upper Riemann sum and
take limit as 𝑛 → ∞ . Now, to find the sum you have to use a property of the
𝜃 1
cos −cos(𝑛+ )𝜃
trigonometric sum ∑𝑛𝑘=1 sin 𝑘𝜃 = 2
𝜃
2
2 sin
2

Some terminology:
Let 𝑓: [𝑎, 𝑏] → ℝ be a bounded function.
The quantity 𝑈(𝑓) is also called as the upper Riemann integral of 𝑓 on [𝑎, 𝑏] .
Similarly, the quantity 𝐿(𝑓) is also called as the lower Riemann integral of 𝑓 on
[𝑎, 𝑏] .
𝑓 is R-integrable on [𝑎, 𝑏] is also called as 𝑓 is R-integrable over [𝑎, 𝑏].
𝑏 𝑏
In that case ∫𝑎 𝑓 can also be written as ∫𝑎 𝑓(𝑥)𝑑𝑥
This notation is quite useful when we are referring to a function not by its name,
𝑏 𝑏 𝜋
but by its formula, e.g., ∫𝑎 2𝑥 𝑑𝑥 , ∫𝑎 𝑥 2 𝑑𝑥 , ∫0 sin 2𝑥 𝑑𝑥 etc.
If 𝑃 = {𝑎0 , 𝑎1 , … , 𝑎𝑛 , } then max{|𝑎𝑖 , −𝑎𝑖−1 , ||𝑖 = 1,2, … , 𝑛} = ‖𝑃‖ is called the
mesh of 𝑃. It is the length of the largest subinterval in a partition.

9) (This in fact, we would also like to label as a result)

Result 10: If 𝑓 is R-integrable on [𝑎, 𝑏] then so is 𝑓 2

Proof: Observe first, that since 𝑓 is R-integrable we can say |𝑓| is R-integrable and
2
𝑓 2 (𝑥) = (𝑓(𝑥)) = |𝑓(𝑥)|2 ∀ 𝑥 ∈ [𝑎, 𝑏]
Also, as the squaring function is strictly increasing function, for each partition
𝑃 = {𝑎0 , 𝑎1 , … , 𝑎𝑛 } of [𝑎, 𝑏], if 𝑀𝑘 , 𝑚𝑘 have usual meanings for |𝑓| and
∀ 𝑘 = 1, … , 𝑛 we have
𝑀𝑘′ = sup{(𝑓(𝑥))2 | 𝑥 ∈ [𝑎𝑘−1 , 𝑎𝑘 ]} = 𝑀𝑘2 and similarly
𝑚𝑘′ = inf{(𝑓(𝑥))2 | 𝑥 ∈ [𝑎𝑘−1 , 𝑎𝑘 ]} = 𝑚𝑘2
Let 𝑀 = sup{|𝑓(𝑥)| |𝑥 ∈ [𝑎, 𝑏]} then 𝑀𝑘 , 𝑚𝑘 ≤ 𝑀 ∀ 𝑘
Now, let 𝜀 > 0
By Riemann’s criterion there exists a partition 𝑃 = {𝑎0 , 𝑎1 , … , 𝑎𝑛 } of [𝑎, 𝑏], such
𝜀
that 𝑈(𝑃, |𝑓|) − 𝐿(𝑃, |𝑓|) < 2𝑀 .
But then
𝑈(𝑃, 𝑓 2 ) − 𝐿(𝑃, 𝑓 2 ) = ∑𝑛𝑘=1(𝑀𝑘′ − 𝑚𝑘′ ) ∆𝑘
= ∑𝑛𝑘=1(𝑀𝑘2 − 𝑚𝑘2 ) ∆𝑘 = ∑𝑛𝑘=1(𝑀𝑘 − 𝑚𝑘 )(𝑀𝑘 + 𝑚𝑘 ) ∆𝑘
≤ ∑𝑛𝑘=1(𝑀𝑘 − 𝑚𝑘 )(2𝑀) ∆𝑘 = 2𝑀 ∑𝑛𝑘=1(𝑀𝑘 − 𝑚𝑘 ) ∆𝑘
𝜀
= 2𝑀(𝑈(𝑃, |𝑓|) − 𝐿(𝑃, |𝑓|)) < 2𝑀 2𝑀 = 𝜀
This proves 𝑓 2 is R-integrable on [𝑎, 𝑏]

10) Prove that R-integrability of 𝑓 2 does not imply 𝑓 is R-integrable.

1 , 𝑥 ∈ 𝑄 ∩ [0,1]
Consider the function 𝑓(𝑥) = {
−1 , 𝑥 ∈ (ℝ − ℚ) ∩ [0,1]
 It is not R-integrable, as for any partition 𝑃 of [0,1] , 𝑈(𝑃, 𝑓) = 1, 𝐿(𝑃, 𝑓) = −1
⇒ 𝑈(𝑓) = 1, 𝐿(𝑓) = −1 are not equal.
But, (𝑓(𝑥))2 = 1 ∀ 𝑥 ∈ [0.1] ⇒ 𝑓 2 is a constant function and hence it is R-
integrable.

11) Result 11: If 𝑓, 𝑔 are R-integrable on [𝑎, 𝑏] then so is their product, 𝑓𝑔 .

Proof: Let 𝑓, 𝑔 be R-integrable on [𝑎, 𝑏]. Then by using linearity properties of the
functions 𝑓 + 𝑔, 𝑓 − 𝑔 are also R-integrable and using result 10, the functions
(𝑓 + 𝑔)2 , (𝑓 − 𝑔)2 are R-integrable which means using linearity properties again,
(𝑓+𝑔)2 −(𝑓−𝑔)2
𝑓𝑔 = is R-integrable on [𝑎, 𝑏].
4

1
12) What can be said about the function 𝑓 ? At least if, we assume that 𝑓 is continuous
and does not vanish anywhere in an interval, can we say something about the
reciprocal of 𝑓?
More precisely prove the following
1
Result 12: If 𝑓 is continuous on [𝑎, 𝑏], 𝑓(𝑥) ≠ 0 ∀ 𝑥 ∈ [𝑎, 𝑏] then prove that 𝑓 is an
R-integrable function on [𝑎, 𝑏] .
Proof: As continuous functions have the intermediate value property; it is clear
that either 𝑓(𝑥) > 0 ∀ 𝑥 ∈ [𝑎, 𝑏] or 𝑓(𝑥) < 0 ∀ 𝑥 ∈ [𝑎, 𝑏]
Assume w.l.g. that 𝑓(𝑥) > 0 ∀ 𝑥 ∈ [𝑎, 𝑏].
As [𝑎, 𝑏] is a closed and bounded interval, 𝑓 attains maximum and minimum
values, say 𝑀 ≥ 𝑚 > 0.
1 1 1 1
As such, is defined and ≤ (𝑥) ≤ ∀𝑥 ∈ [𝑎, 𝑏]
𝑓 𝑀 𝑓 𝑚
Thus, it is a bounded and continuous function on [𝑎, 𝑏], so it is R-integrable.

Even if 𝑓 is an R-integrable function which is not continuous, but if it does not

change sign and remains bounded away from 0, we can still prove R-integrability of
1
using Riemann’s criterion. Try this as a problem.
𝑓

13) Finally, can something be said about inverse of a function? Well, the following can
definitely be said.
Result13: Suppose 𝑓 is continuous and strictly increasing on [𝑎, 𝑏] then 𝑓 −1 is
defined and it is R-integrable on [𝑓(𝑎), 𝑓(𝑏)] .
To prove this, we only need to prove that a continuous, strictly increasing function
defined on an interval is invertible and the inverse function is also continuous.
Continuity of the inverse function will then be sufficient to prove its R-integrability.
Try this.
A sharper result can in fact be proved.
Result: Suppose 𝑓 is continuous and strictly increasing on [𝑎, 𝑏] then 𝑓 −1 is defined
and it is R-integrable on [𝑓(𝑎), 𝑓(𝑏)] . Further,
𝑏 𝑓(𝑏)

∫ 𝑓 + ∫ 𝑓 −1 = 𝑏𝑓(𝑏) − 𝑎𝑓(𝑎)
𝑎 𝑓(𝑎)
 This allows us to compute the integral of the inverse function in terms of the
integral of the given invertible function.
Proof: Suppose 𝑓 is continuous and strictly increasing on [𝑎, 𝑏]. Then its image
must be some [𝑚. 𝑀] and it must be injective, for if not, then there must be
𝑥1 < 𝑥2 with 𝑓(𝑥1 ) = 𝑓(𝑥2 ) in the domain which contradicts the strictly increasing
nature of 𝑓.
Therefore 𝑔(𝑦) = 𝑓 −1 (𝑦) is defined on [𝑚. 𝑀] with 𝑓(𝑎) = 𝑚, 𝑓(𝑏) = 𝑀 and
whenever
𝑚 ≤ 𝑦1 < 𝑦2 ≤ 𝑀, let 𝑥1 = 𝑔(𝑦1 ), 𝑥2 = 𝑔(𝑦2 ).
We must have 𝑥1 < 𝑥2 , i.e., 𝑔 is strictly increasing on [𝑚. 𝑀]. But then it is
monotonic, therefore it is R-integrable. Now, let 𝑃 = {𝑥0 , 𝑥1 , … , 𝑥𝑛 } be any
partition of [𝑎, 𝑏] and Q= {𝑦0 , 𝑦1 , … , 𝑦𝑛 } be corresponding partition of [𝑚. 𝑀] with
𝑦𝑘 = 𝑓(𝑥𝑘 ) ∀ 𝑘 = 1, … , 𝑛. In fact, all partitions of [𝑎, 𝑏] are in one to one
correspondence with partitions of [𝑚. 𝑀] and we have
𝑈(𝑄, 𝑔) + 𝐿(𝑃, 𝑓)
= 𝑔(𝑦1 )(𝑦1 − 𝑦0 ) + ⋯ + 𝑔(𝑦𝑛 )(𝑦𝑛 − 𝑦𝑛−1 )
+𝑓(𝑥0 )(𝑥1 − 𝑥0 ) + ⋯ + 𝑓(𝑥𝑛−1 )(𝑥𝑛 − 𝑥𝑛−1 )
= 𝑥1 (𝑦1 − 𝑦0 ) + 𝑥2 (𝑦2 − 𝑦1 ) + ⋯ + 𝑥𝑛 (𝑦𝑛 − 𝑦𝑛−1 )
+𝑦0 (𝑥1 − 𝑥0 ) + 𝑦2 (𝑥2 − 𝑥1 ) + ⋯ + 𝑦𝑛 (𝑥𝑛 − 𝑥𝑛−1 )
= 𝑥𝑛 𝑦𝑛 − 𝑥0 𝑦0 = 𝑏𝑀 − 𝑎𝑚
Thus, 𝑈(𝑔) ≤ 𝑈(𝑄, 𝑔) = 𝑏𝑀 − 𝑎𝑚 − 𝐿(𝑃, 𝑓) ⇒ 𝑈(𝑔) + 𝐿(𝑃, 𝑓) ≤ 𝑏𝑀 − 𝑎𝑚
for every partition 𝑃 of [𝑎, 𝑏]
⇒ 𝑈(𝑔) + 𝐿(𝑓) ≤ 𝑏𝑀 − 𝑎𝑚
Similarly, 𝐿(𝑓) ≥ 𝐿(𝑃, 𝑓) = 𝑏𝑀 − 𝑎𝑚 − 𝑈(𝑄, 𝑔) for every partition Q of [𝑚, 𝑀]
⇒ 𝑈(𝑔) + 𝐿(𝑓) ≥ 𝑏𝑀 − 𝑎𝑚
Hence 𝑈(𝑔) + 𝐿(𝑓) = 𝑏𝑀 − 𝑎𝑚

1
14) Calculate ∫0 tan−1 𝑥 𝑑𝑥.
𝜋
We know that 𝑓(𝑥) = tan 𝑥 is strictly increasing on [0, 4 ] and hence
1 𝜋⁄
−1
4 𝜋
∫ tan 𝑦 𝑑𝑦 + ∫ tan 𝑥 𝑑𝑥 = ×1−0×0
0 0 4
𝜋⁄ 𝜋⁄
4
Now, ∫0 4 tan 𝑥 𝑑𝑥 = [log sec 𝑥]0
𝜋
= log sec − log sec 0
4
= log √2 − log 1 = log √2
1 𝜋
Hence, ∫0 tan−1 𝑦 𝑑𝑦 = 4 − log √2

3
15) Let 𝑓(𝑥) = 1 + 𝑥 + 𝑥 2 ∀ 𝑥 ∈ [0,1]. Find ∫0 𝑓 −1 (𝑥) 𝑑𝑥
(Verify that 𝑓 is invertible and use the result.)

Result: (Schwartz’ inequality)

Let 𝑓, 𝑔 be R-integrable on [𝑎, 𝑏] then
𝑏 2
𝑏 𝑏
(∫ 𝑓𝑔) ≤ (∫ 𝑓 2 ) (∫ 𝑔2 )
𝑎 𝑎
𝑎
 We do not go into the proof of this, but for known functions, you can verify this.
Recall the Cauchy-Schwartz inequality of real numbers states that
if 𝑎1 , 𝑏1 , 𝑎2 , 𝑏2 , … , 𝑎𝑛 , 𝑏𝑛 are real numbers, then
𝑛 2 𝑛 𝑛

(∑ 𝑎𝑘 𝑏𝑘 ) ≤ (∑ 𝑎𝑘 ) (∑ 𝑏𝑘 2 ) 2

𝑘=1 𝑘=1 𝑘=1

Do you see any similarity between the two?

16) Verify Schwartz’ inequality for 𝑥, 𝑥 2 on [0,1]

For simplicity of calculations, we will assume fundamental theorem of calculus and
calculate the definite integrals using antiderivatives.
Let 𝑓(𝑥) = 𝑥, 𝑔(𝑥) = 𝑥 2 .
𝑏 2 𝑏 𝑏
We need to verify (∫𝑎 𝑓𝑔) ≤ (∫𝑎 𝑓 2 ) (∫𝑎 𝑔2 )
1 2 1
LHS = (∫0 𝑥 3 𝑑𝑥) = 16
1 2 1 2 1 1 1
RHS = (∫0 𝑥 2 𝑑𝑥) (∫0 𝑥 4 𝑑𝑥) = 3 × 5 = 15
Thus, LHS ≤ RHS as required.

The Schwartz’ inequality plays an important role in the study of Fourier series, as
we define an ‘inner product’ on the vector space of all 2𝜋 −periodic functions on
say [−𝜋, 𝜋] as
1 𝜋
〈𝑓, 𝑔〉 = ∫−𝜋 𝑓(𝑥)𝑔(𝑥)𝑑𝑥 which induces a ‘norm’ on this vector space. In this
𝜋
normed vector space, the set {cos 𝑚𝑥 , sin 𝑛𝑥 |𝑚, 𝑛 ∈ ℕ} along with the constant
function 1, behaves almost like a basis and hence any 2𝜋 −periodic function 𝑓(𝑥)
can be expressed as
𝑎0 ∞ ∞
𝑓(𝑥) ≈ + ∑ 𝑎𝑛 cos 𝑛𝑥 + ∑ 𝑏𝑛 sin 𝑛𝑥
2 𝑛=1 𝑛=1
Where the scalars are calculated as
1 𝜋 1 𝜋 1 𝜋
𝑎0 = 𝜋 ∫−𝜋 𝑓(𝑥)𝑑𝑥 , 𝑎𝑛 = 𝜋 ∫−𝜋 𝑓(𝑥) cos 𝑛𝑥 𝑑𝑥 and 𝑏𝑛 = 𝜋 ∫−𝜋 𝑓(𝑥) sin 𝑛𝑥 𝑑𝑥 .

We now come to the fundamental theorem of Calculus.

First, suppose 𝑓 is a R-integrable function on [𝑎, 𝑏]

𝑥
Define 𝑔(𝑥) = 𝑔(𝑎) + ∫𝑎 𝑓 , where 𝑔(𝑎) is a convenient constant.
𝑥
For simplicity, we can use the notations 𝑔(𝑥) = ∫𝑎 𝑓(𝑡) 𝑑𝑡
Is it a good function? How good it is? Is it continuous? is it differentiable? What is
the derivative?
𝑥 𝑝 𝑥
Let 𝑝 be fixed in [𝑎, 𝑏] and 𝑥 is near 𝑝. Then 𝑔(𝑥) − 𝑔(𝑝) = ∫𝑎 𝑓 − ∫𝑎 𝑓 = ∫𝑝 𝑓
As 𝑓 is bounded, let 𝑀 = sup{|𝑓(𝑥)| | 𝑥 ∈ [𝑎, 𝑏]}
Unless 𝑓 ≡ 0, we must have 𝑀 > 0
𝜀
For every 𝜀 > 0, take 𝛿 = 𝑀 and observe that
𝑥
|𝑥 − 𝑝| < 𝛿 ⇒ |𝑔(𝑥) − 𝑔(𝑝)| = |∫ 𝑓| ≤ |𝑥 − 𝑝|𝑀 < 𝛿𝑀 = 𝜀
𝑝
Thus, 𝑔 is continuous.
In case 𝑓 ≡ 0, 𝑔 is constant, so it is continuous.
To check whether it is differentiable, we should try to calculate the derivative.
𝑓(𝑥)−𝑓(𝑝) 1 𝑥
So, suppose 𝑝 < 𝑏 and calculate lim = lim ∫ 𝑓.
𝑥→𝑝+ 𝑥−𝑝 𝑥→𝑝+ 𝑥−𝑝 𝑝
How will you tackle with this problem?
In general, 𝑔 need not be differentiable, as for example if we consider the function
1 , 𝑥 ∈ [0,1]
𝑓(𝑥) = {
2 , 𝑥 ∈ (1,2]
Define
𝑥
𝑔(𝑥) = ∫ 𝑓
0
We will show that 𝑔 is not differentiable at 1
Let ℎ > 0 and consider
1 1+ℎ 1 1+ℎ
𝑔(1 + ℎ) − 𝑔(1) ∫0 𝑓 + ∫1 𝑓 − ∫0 𝑓 ∫1 𝑓
= =
ℎ ℎ ℎ
In [1,1 + ℎ], 𝑓 is almost constant, in fact replacing its value at 1 by 𝑓(1) = 2, in
place of 𝑓(1) = 1, the integral does not change. It can be calculated as
2(1 + ℎ − 1) = 2ℎ
𝑔(1+ℎ)−𝑔(1) 2ℎ
Hence lim = lim ℎ = 2
ℎ→0+ ℎ ℎ→0+
On the other hand, if we take the limit from left,
1−ℎ 1−ℎ 1
𝑔(1+ℎ)−𝑔(1) 𝑔(1−ℎ)−𝑔(1) ∫0 𝑓 −∫0 𝑓 −∫1−ℎ 𝑓
lim = lim = lim
ℎ→0− ℎ ℎ→0+ −ℎ ℎ→0+ −ℎ
1 1
− ∫1−ℎ 𝑓 ∫1−ℎ 𝑓
= lim = lim
ℎ→0+ −ℎ ℎ→0+ ℎ
Again, as 𝑓 is constant on [1 − ℎ, 1], the integral can be evaluated as ℎ and hence
ℎ
the left derivative of 𝑔 at 1 is lim ℎ = 1.
ℎ→0−
This shows that 𝑔 is not differentiable.

In this example, the function 𝑓 was discontinuous. However, if we insist on

continuity of 𝑓, the integrand, then we have

Result 14: (Fundamental theorem of calculus in first form) Let 𝑓 be a continuous

function on [𝑎, 𝑏] and define a function
𝑥

𝐹(𝑥) = ∫ 𝑓(𝑡) 𝑑𝑡 ∀ 𝑥 ∈ [𝑎, 𝑏]

𝑎
then, 𝐹 is differentiable on [𝑎, 𝑏] and
𝐹 ′ (𝑥) = 𝑓(𝑥) , ∀ 𝑥 ∈ [𝑎, 𝑏]

Proof: As 𝑓 is continuous function on [𝑎, 𝑏], a closed and bounded interval, it is

uniformly continuous and hence ∀ 𝜀 > 0, ∃ 𝛿 > 0 such that whenever 𝑠, 𝑡 ∈ [𝑎, 𝑏]
and |𝑠 − 𝑡| < 𝛿 ⇒ |𝑓(𝑠) − 𝑓(𝑡)| < 𝜀
Now, suppose 𝑥 ∈ (𝑎, 𝑏) and 0 < |ℎ| < 𝛿 is sufficiently small, so that
𝑥 + ℎ ∈ (𝑎, 𝑏) then for ℎ > 0,
𝐹(𝑥+ℎ)−𝐹(𝑥)
| ℎ
− 𝑓(𝑥)|
1 𝑥+ℎ 𝑥
= |∫ 𝑓(𝑡)𝑑𝑡 − ∫𝑎 𝑓(𝑡)𝑑𝑡 − ℎ𝑓(𝑥)|
ℎ 𝑎
 1 𝑥+ℎ 𝑥+ℎ
= |∫𝑥 𝑓(𝑡)𝑑𝑡 − ∫𝑥 𝑓(𝑥)𝑑𝑡|
ℎ
1 𝑥+ℎ
= ℎ |∫𝑥 (𝑓(𝑡) − 𝑓(𝑥))𝑑𝑡|
1 𝑥+ℎ 1 𝑥+ℎ 𝜀ℎ
≤ ∫ |𝑓(𝑡)
ℎ 𝑥
− 𝑓(𝑥)| ≤ ℎ ∫𝑥 𝜀= ℎ
=𝜀
𝐹(𝑥+ℎ)−𝐹(𝑥) 1 𝑥+ℎ 𝑥
If ℎ < 0, then | − 𝑓(𝑥)| = −ℎ |∫𝑎 𝑓(𝑡)𝑑𝑡 − ∫𝑎 𝑓(𝑡)𝑑𝑡 − ℎ𝑓(𝑥)|
ℎ
1 𝑥+ℎ 𝑥+ℎ 1 𝑥
= |∫ 𝑓(𝑡)𝑑𝑡 − ∫𝑥 𝑓(𝑥)𝑑𝑡| = −ℎ |− ∫𝑥+ℎ(𝑓(𝑡) − 𝑓(𝑥))𝑑𝑡|
−ℎ 𝑥
1 𝑥 1 𝑥 −𝜀ℎ
≤ ∫ |𝑓(𝑡) − 𝑓(𝑥)| ≤ −ℎ ∫𝑥+ℎ 𝜀 = −ℎ = 𝜀
−ℎ 𝑥+ℎ
As 𝜀 can be taken arbitrarily small, this in fact proves that 𝐹 is differentiable, and
𝐹 ′ (𝑥) = 𝑓(𝑥)
In case, 𝑥 = 𝑎, we take ℎ > 0 and in case, 𝑥 = 𝑏, we take ℎ < 0 , thereby proving
the result for each 𝑥 ∈ [𝑎, 𝑏].

Result 15:(corollary) Every continuous function defined over a closed and bounded
interval must have at least one primitive (antiderivative).
Proof: evident from above theorem.

In fact, if 𝐹 is an antiderivative of 𝑓 on [𝑎, 𝑏] obtained as above, then any function

defined by 𝑔(𝑥) = 𝑔(𝑎) + 𝐹(𝑥) where 𝑔(𝑎) is an antiderivative of 𝑓 on [𝑎, 𝑏].

Result 16: (Second form of fundamental theorem of Calculus) If 𝑓 is continuous on

an interval and if 𝑎 is any point in that interval, then 𝑓 has a unique antiderivative
𝑥
𝐹 on that interval which satisfies 𝐹(𝑥) = ∫𝑎 𝑓(𝑡)𝑑𝑡 for all 𝑥 in the interval.

This can be proved in many ways, though as a step in proving this, we can use
Cauchy’s theorem which was popularly called as second form of fundamental
theorem.
Result 17: (Cauchy’s theorem)
Let 𝑓 be R-integrable on [𝑎, 𝑏] and suppose 𝐹 is a primitive of 𝑓.
𝑏
Then 𝐹(𝑏) − 𝐹(𝑎) = ∫𝑎 𝑓

Proof: Consider a partition 𝑃 = {𝑎0 , 𝑎1 , … , 𝑎𝑛 } of [𝑎, 𝑏].

On each subinterval [𝑎𝑘−1 , 𝑎𝑘 ] we can apply LMVT to 𝑓 and establish existence of
some 𝑡𝑘 ∈ (𝑎𝑘−1 , 𝑎𝑘 ) such that
𝐹(𝑎𝑘 ) − 𝐹(𝑎𝑘−1 ) = (𝑎𝑘 − 𝑎𝑘−1 )𝑓(𝑡𝑘 )
But then in usual notations, 𝑚𝑘 ≤ 𝑓(𝑡𝑘 ) ≤ 𝑀𝑘
Hence,
𝐿(𝑃, 𝑓) = ∑𝑛𝑘=1 𝑚𝑘 (𝑎𝑘 − 𝑎𝑘−1 )
≤ ∑𝑛𝑘=1 𝑓(𝑡𝑘 )(𝑎𝑘 − 𝑎𝑘−1 )
= ∑𝑛𝑘=1 𝐹(𝑎𝑘 ) − 𝐹(𝑎𝑘−1 ) = 𝐹(𝑏) − 𝐹(𝑎)
Therefore, 𝐿(𝑓) ≤ 𝐹(𝑏) − 𝐹(𝑎)
Similarly, we can prove 𝑈(𝑓) ≥ 𝐹(𝑏) − 𝐹(𝑎)
Since 𝑓 is R-integrable on [𝑎, 𝑏] we must have
𝑏
𝑈(𝑓) = 𝐿(𝑓) = ∫ 𝑓 = 𝐹(𝑏) − 𝐹(𝑎)
𝑎
This proves the result.
Further, if 𝑎 ≤ 𝑥 ≤ 𝑏, R-integrability of 𝑓 on [𝑎, 𝑏] implies that on [𝑎, 𝑥], so
𝑥
∫𝑎 𝑓(𝑡)𝑑𝑡 = 𝐹(𝑥) − 𝐹(𝑎)
𝑥
⇒ 𝐹(𝑥) = 𝐹(𝑎) + ∫𝑎 𝑓(𝑡)𝑑𝑡
If above hypotheses stand for an interval on which 𝐹 happens to be a primitive of 𝑓
and 𝑥 < 𝑎, then conventionally, we can write
𝑥 𝑎
∫𝑎 𝑓(𝑡)𝑑𝑡 = − ∫𝑥 𝑓(𝑡)𝑑𝑡 = −(𝐹(𝑎) − 𝐹(𝑥)) which gives us
𝑥
∫𝑎 𝑓(𝑡)𝑑𝑡 = 𝐹(𝑥) − 𝐹(𝑎) as required.
Now, in particular, if 𝑓(𝑥) = 𝐹 ′ (𝑥) = 0 on an interval, then the constant function
𝑥
zero is R-integrable and 𝐹(𝑥) − 𝐹(𝑎) = ∫𝑎 0𝑑𝑡 = 0 ⇒ 𝐹 is constant on the
interval. Using these facts, we can now prove the statement of result 16.

As observed from the proof of Cauchy’s form of fundamental theorem of calculus,

if 𝑓 is Riemann integrable on [𝑎, 𝑏] and for any partition 𝑃 of [𝑎, 𝑏] in 𝑛 parts, if we
pick any 𝑡𝑘 ∈ [𝑎𝑘−1 , 𝑎𝑘 ] we still get
𝐿(𝑃, 𝑓) = ∑𝑛𝑘=1 𝑚𝑘 (𝑎𝑘 − 𝑎𝑘−1 )
≤ ∑𝑛𝑘=1 𝑓(𝑡𝑘 )(𝑎𝑘 − 𝑎𝑘−1 )
≤ ∑𝑛𝑘=1 𝑀𝑘 (𝑎𝑘 − 𝑎𝑘−1 ) = 𝑈(𝑃, 𝑓)
This means if we know that 𝑓 is R-integrable on [𝑎, 𝑏], then to calculate the
integral, we can use any convenient sequence of partitions and in that, we need
not find 𝑀𝑘 , 𝑚𝑘 for each subinterval, but any choice of 𝑡𝑘 ∈ [𝑎𝑘−1 , 𝑎𝑘 ] will give us a
value 𝑓(𝑡𝑘 ) and we can write
𝑏
∫𝑎 𝑓 = lim ∑𝑛𝑘=1 𝑓(𝑡𝑘 )∆𝑘
𝑛→∞

We now present Riemann’s original idea of integration as follows:

Let for any partition 𝑃 of an interval into 𝑛 parts, by mesh of 𝑃 , we mean the
quantity ‖𝑃‖ = max {∆𝑘 |𝑘 = 1,2, … , 𝑛} .
Let 𝑓 be a bounded real valued function on [𝑎, 𝑏]. We say, 𝑓 is R-integrable on
[𝑎, 𝑏] if ∃ 𝐼 ∈ ℝ such that ∀ 𝜀 > 0, ∃ 𝛿 > 0 such that for any partition 𝑃 =
{𝑎0 , 𝑎1 , … , 𝑎𝑛 } of [𝑎, 𝑏] with ‖𝑃‖ < 𝛿 and for any choice of 𝑡𝑘 ∈ [𝑎𝑘−1 , 𝑎𝑘 ], we
have
𝑛

|𝐼 − ∑ 𝑓(𝑡𝑘 )∆𝑘 | < 𝜀

𝑘=1
The quantity 𝐼 is called an integral of 𝑓 on [𝑎, 𝑏].
(Of course, such a quantity will be unique, if it exists)

Integration as a limit of a sum:

If we take an R-integrable function 𝑓 on [𝑎, 𝑏] and a sequence of partitions (𝑃𝑛 )

each, into 𝑛 parts, with a choice of convenient 𝑡𝑘 ∈ [𝑎𝑘−1 , 𝑎𝑘 ] so that the sums
∑𝑛𝑘=1 𝑓(𝑡𝑘 ) ∆𝑘 form a convergent sequence, then the limit of this sequence must
𝑏
be ∫𝑎 𝑓
Based on this idea, we have a few problems.
17) 1 𝑛+𝑘
Evaluate lim ∑𝑛𝑘=1
𝑛→∞ 𝑛 𝑛

Take 𝑓: [1,2] → ℝ as 𝑓(𝑥) = 𝑥 and consider the partition

1 2 𝑛
𝑃𝑛 = {1, 1 + 𝑛 , 1 + 𝑛 , … ,1 + 𝑛 = 2}
1 𝑛+𝑘−1 𝑛+𝑘
Now, width of each subinterval is 𝑛 and each 𝑘 𝑡ℎ subinterval is just [ , ]
𝑛 𝑛
𝑛+𝑘 𝑛+𝑘−1 𝑛+𝑘
If we choose 𝑡𝑘 = ∈[ , ] ∀ 𝑘 = 1,2, … , 𝑛 then observe that
𝑛 𝑛 𝑛
1 𝑛+𝑘 1 𝑛+𝑘 1 𝑛+𝑘 1
∑𝑛𝑘=1 = 𝑛 ∑𝑛𝑘=1 ( 𝑛 ) 𝑛 = ∑𝑛𝑘=1 𝑓 ( 𝑛 ) 𝑛
𝑛 𝑛
= ∑𝑛𝑘=1 𝑓(𝑡𝑘 )∆𝑘 = 𝑈(𝑃𝑛 , 𝑓)
As the function is R-integrable, we know that
𝑛 𝑛 2
1 𝑛+𝑘
lim ∑ = lim ∑ 𝑓(𝑡𝑘 )∆𝑘 = ∫ 𝑥 𝑑𝑥
𝑛→∞ 𝑛 𝑛 𝑛→∞ 1
𝑘=1 𝑘=1
𝑥2
Now, an antiderivative of 𝑥 is and by fundamental theorem of calculus, we can
2
2
2 𝑥2 3
compute ∫1 𝑥 𝑑𝑥 = [ 2 ] = 2
1

1 𝑛
18) Evaluate lim 𝑛 ∑𝑛𝑘=1 𝑛+𝑘 .
𝑛→∞

1
We now take the function 𝑓(𝑥) = 𝑥 ∀ 𝑥 ∈ [1,2]
Again, with the partition
1 2 𝑛 1 𝑛
𝑃𝑛 = {1, 1 + 𝑛 , 1 + 𝑛 , … ,1 + 𝑛 = 2} we observe that 𝑛 ∑𝑛𝑘=1 𝑛+𝑘 = 𝑈(𝑃𝑛 , 𝑓) and
hence
1 𝑛 2 𝑑𝑥
lim ∑𝑛𝑘=1 = ∫1 = log 2 − log 1 = log 2 .
𝑛→∞ 𝑛 𝑛+𝑘 𝑥

1 𝑥 2
19) Evaluate lim 𝑥 ∫0 𝑒 𝑡 𝑑𝑡
𝑥→0

𝑥 2 2
Let 𝐹(𝑥) = ∫0 𝑒 𝑡 𝑑𝑡 The function 𝑓(𝑥) = 𝑒 𝑥 is a continuous function defined on
ℝ and therefore by fundamental theorem of calculus, we must have
𝐹 ′ (𝑥) = 𝑓(𝑥) ∀ 𝑥 ∈ ℝ .
𝑥 2 0 2
′ (0) 𝐹(𝑥)−𝐹(0) ∫0 𝑒 𝑡 𝑑𝑡−∫0 𝑒 𝑡 𝑑𝑡
Now, 𝐹 = lim = lim
𝑥→0 𝑥−0 𝑥→0 𝑥
𝑥 𝑡2
1 02
= lim ∫ 𝑒 𝑑𝑡 = 𝑓(0) = 𝑒 =1
𝑥→0 𝑥 0

1 𝑥 2
20) Evaluate lim 𝑥 ∫100 𝑒 𝑡 𝑑𝑡 (Exercise)
𝑥→0

21) 𝑑 𝑥2 2
Evaluate 𝑑𝑥 ∫0 𝑒 𝑡 𝑑𝑡

𝑥 2
Define 𝑓(𝑥) = 𝑥 2 , 𝜑(𝑥) = ∫0 𝑒 𝑡 𝑑𝑡 ∀ 𝑥 ∈ ℝ.
2
On one hand, by fundamental theorem of calculus we know that 𝜑 ′ (𝑥) = 𝑒 𝑥 ,
𝑥2 2
while we also know that for 𝐹(𝑥) = ∫0 𝑒 𝑡 𝑑𝑡, 𝐹(𝑥) = 𝜑𝑜𝑓(𝑥)
 Hence, by chain rule of differentiation,
𝐹 ′ (𝑥) = 𝜑′(𝑓(𝑥))𝑓 ′ (𝑥)
2
Now, 𝑓(𝑥) = 𝑥 2 ⇒ 𝑓 ′ (𝑥) = 2𝑥, and 𝜑 ′ (𝑥) = 𝑒 𝑥
2 2 4
⇒ 𝜑 ′ (𝑓(𝑥)) = 𝜑 ′ (𝑥 2 ) = 𝑒 (𝑥 ) = 𝑒 𝑥
4
⇒ 𝐹 ′ (𝑥) = 𝜑 ′ (𝑓(𝑥))𝑓 ′ (𝑥) = 2𝑥𝑒 𝑥

A caution to be taken with conditions/hypotheses in a theorem:

The first form of fundamental theorem requires the integrand to be a continuous

function, while in Cauchy’s theorem, the integrand is a Riemann Integrable
function which has a primitive.
What happens, if in first form, we take a Riemann integrable function which may
not be continuous, then can we expect to prove existence of a primitive function?
An example we have discussed earlier is:
1 , 𝑥 ∈ [0,1]
𝑓(𝑥) = {
2 , 𝑥 ∈ (1,2]
𝑥
This is integrable, though not continuous. So, we can not say 𝑔(𝑥) = ∫𝑎 𝑓(𝑡)𝑑𝑡 is a
primitive. In fact, it is not even differentiable. But one can say that within [0,1] 𝑓
has a primitive which is given by 𝑔.
We now take example of a function which is R-integrable over every interval, but
does not have a primitive on any interval.
𝑎
22) For a rational number 𝑟, we know that it can be expressed in many ways as 𝑟 = 𝑏 .
𝑝
Let 𝑞 be a form of 𝑟, where 𝑞 > 0 and gcd(𝑝, 𝑞) = 1. Then there is a unique way in
which 𝑟 can be expressed in such form. We will call it as the lowest form of the
rational number.
2 25 −5
e.g., lowest form of 2 is 1 and lowest form of − 15 is 3 .

Now define a function on ℝ as

1 𝑝
, 𝑥 ∈ ℚ, 𝑥 = in lowest form
𝑓(𝑥) = { 𝑞 𝑞
0 , 𝑥 ∈ℝ−ℚ
This function is R-integrable over any interval, though it does not have a primitive.
For simplicity, we present a proof for the interval [0,1].

2
Consider 𝜀 > 0. By Archimedean property, choose 𝑛 ∈ ℕ, 𝑛 > 𝜀 .
Now, if 𝑥 is an irrational number in [0,1] then 𝑓(𝑥) = 0. While, if it is rational, then
𝑝
its lowest form would be some 𝑞 . We will say 𝑥 is ‘bad’ if 𝑞 < 𝑛 and it is ‘good’ if
𝑞 ≥ 𝑛 . Verify that there are only a finitely many bad numbers in [0,1].

Now, these bad numbers can be taken in ascending order, and we can choose a
partition 𝑃 of [0,1] in which these bad numbers occur in alternate subintervals,
𝜀
whose total length can be made less than 2 . (we have done such a thing before).
It is now left to verify that 𝑈(𝑃, 𝑓) − 𝐿(𝑃, 𝑓) < 𝜀 . Can you verify this?
 1
This proves that the function 𝑓 is R-integrable on [0,1] and ∫0 𝑓 = 0 . In fact, we
𝑥
have also proved that ∫0 𝑓 = 0 ∀ 𝑥 ∈ [0,1].
Now, suppose 𝑓 has a primitive over [0,1], say 𝐹. Then by Cauchy.s theorem, we
have
𝑥
𝐹(𝑥) − 𝐹(0) − ∫ 𝑓 = 0
0
⇒ 𝐹(𝑥) = 𝐹(0) ∀ 𝑥 ∈ [0,1]
i.e., 𝐹 ′ (𝑥) = 0 ∀ 𝑥 ∈ [0,1] whereas, 𝐹 ′ (𝑥) = 𝑓(𝑥) as 𝐹 is a primitive of 𝑓 as per
our assumption.
This is a contradiction and therefore 𝑓 does not have a primitive.

23) On the other hand, Cauchy’s theorem requires the integrand function 𝑓to be an
R-integrable function. If we just take a function 𝐹 which is differentiable on an
interval and take 𝑓 = 𝐹′ on that interval then can we not believe that 𝑓 being the
derivative of some function, must be an R-integrable function? Can we conclude
𝑏
𝐹(𝑏) − 𝐹(𝑎) = ∫ 𝑓 ?
𝑎
Answer is NO. WE need to establish the R-integrability of 𝑓 before drawing that
conclusion. 𝑓 may not even be bounded on the interval.
Consider the function given by
4 1
𝑥 3 sin ( ) , 𝑥>0
𝐹(𝑥) = { 𝑥
0 , 𝑥=0
We can prove that 𝐹 is differentiable.
Clearly, in (0. ∞) ∪ (−∞, 0) we have
4
′ (𝑥)
4 1 1 𝑥3 1
𝐹 = 𝑥 3 sin ( ) − 2 cos ( )
3 𝑥 𝑥 𝑥
while,
4 1
𝐹(ℎ) − 𝐹(0) ℎ3 sin ( )
lim = lim ℎ = lim ℎ13 sin (1) = 0
ℎ→0 ℎ ℎ→0 ℎ ℎ→0 ℎ
Now that proves 𝐹 is differentiable and
4 1 1 −2 1
′ (𝑥) 𝑥 3 sin (𝑥) − 𝑥 3 cos (𝑥) , 𝑥≠0
𝑓(𝑥) = 𝐹 ={ 3
0 , 𝑥=0

Now, this function is unbounded in every interval containing zero. This can be seen
−2
1 1 1 3
by taking 𝑥𝑛 = 2𝑛𝜋 ∀ 𝑛 ∈ ℕ and the calculation 𝑓(𝑥𝑛 ) = 𝑓 (2𝑛𝜋) = − (2𝑛𝜋)
1
shows that for large values of 𝑛, 𝑓(𝑥𝑛 ) → −∞. In fact, 𝑓 (2𝑛𝜋+𝜋) → ∞.
An unbounded function can not be R-intgrable as per our definitions.
(This 𝐹 in fact works as an example of a differentiable function whose derivative is
not continuous.)

This shows that the hypotheses in the statement of a theorem can not be relaxed
in general.
24) 𝑑 𝑥3 1
Evaluate 𝑑𝑥 ∫0 𝑑𝑡
1+sin2 𝑡

As before, define
𝑥 1
𝜑(𝑥) = ∫0 1+sin2𝑡 𝑑𝑡, 𝑓(𝑥) = 𝑥 3 and
𝑥3
1
2
𝐹(𝑥) = ∫
𝑑𝑡 = 𝜑𝑜𝑓(𝑥)
0 1 + sin 𝑡
Apply fundamental theorem of calculus as well as the chain rule for differentiation.

25) If 𝑎, 𝑏 are any two distinct positive real numbers then prove that the value of the
integral
𝜋 𝑥 𝑑𝑥 𝜋2 𝜋2
∫0 lies between 2𝑎 and 2𝑏
𝑎 cos2 𝑥+𝑏 sin2 𝑥

w.l.g let 𝑎 > 𝑏 > 0

𝜋 𝜋2
Observe that ∫0 𝑥 𝑑𝑥 = 2 .
Now, 𝑎 cos2 𝑥 + 𝑏 sin2 𝑥 = 𝑎 cos2 𝑥 + 𝑏 − 𝑏 cos 2 𝑥 = (𝑎 − 𝑏)cos2 𝑥 + 𝑏 ≥ 𝑏
1 1 𝑥 𝑥
⇒ 𝑎 cos2𝑥+𝑏 sin2𝑥 ≤ 𝑏 ⇒ 𝑎 cos2𝑥+𝑏 sin2𝑥 ≤ 𝑏 ∀ 𝑥 ∈ [0, 𝜋] , while
𝑎 cos2 𝑥 + 𝑏 sin2 𝑥 = 𝑎 − (𝑎 − 𝑏)sin2 𝑥 ≤ 𝑎
1 1 𝑥 𝑥
⇒ 𝑎 cos2𝑥+𝑏 sin2𝑥 ≥ 𝑎 ⇒ 𝑎 cos2𝑥+𝑏 sin2𝑥 ≥ 𝑎 ∀ 𝑥 ∈ [0, 𝜋]
𝑥 𝑥 𝑥
Thus, 𝑎 ≤ 𝑎 cos2𝑥+𝑏 sin2𝑥 ≤ 𝑏 ∀ 𝑥 ∈ [0, 𝜋] and hence
𝜋 𝜋 𝜋
𝑥 𝑑𝑥 𝑥 𝑑𝑥 𝑥 𝑑𝑥
∫ ≤∫ 2 2
≤∫
0 𝑎 0 𝑎 cos 𝑥 + 𝑏 sin 𝑥 0 𝑏
Evaluating the integrals on the two extremes, we get the required result.
The case 𝑏 > 𝑎 > 0 is similar.
𝑏 𝑏
We have used the property 𝑓(𝑥) ≥ 𝑔(𝑥) ∀ 𝑥 ∈ [𝑎, 𝑏] ⇒ ∫𝑎 𝑓 ≥ ∫𝑎 𝑔 as well as the
fundamental theorem of calculus, when we finally evaluate the integrals.

𝑒−1 1 2
26) Show that < ∫0 𝑒 −𝑥 𝑑𝑥 < 1
2𝑒

2 2
Observe that ∀ 𝑥 ∈ [0,1]; 𝑥𝑒 −𝑥 ≤ 𝑒 −𝑥 ≤ 1
1 2
Now, consider ∫0 (1 − 𝑒 −𝑥 )𝑑𝑥 . The integrand in this is a continuous nonnegative
function and it does not vanish everywhere in [0,1], hence the integral will be
positive.
1 2 1 1 2
Thus, ∫0 (1 − 𝑒 −𝑥 )𝑑𝑥 > 0 ⇒ ∫0 1 𝑑𝑥 − ∫0 𝑒 −𝑥 𝑑𝑥 > 0 This establishes one part.
1 2 1 2
For the other one, similar argument yields ∫0 𝑒 −𝑥 𝑑𝑥 > ∫0 𝑥𝑒 −𝑥 𝑑𝑥 .
2
1 2 𝑒 −𝑥
We can evaluate the integral ∫0 𝑥𝑒 −𝑥 𝑑𝑥 using the fact that − is an
2
−𝑥 2
antiderivative of the function 𝑥𝑒 which is a continuous, hence an R-integrable
function on [0,1].
By Cauchy’s form of fundamental theorem of calculus,
2 1
1 2 𝑒 −𝑥 𝑒−1
∫0 𝑥𝑒 −𝑥 𝑑𝑥 = [− 2
] = 2𝑒
.
0
27) 1 1⁄2 𝑑𝑥 𝜋
Prove that 2 < ∫0 <
√1−𝑥 4 6
1
In the interval [0, 2] observe that √1 − 𝑥 2 ≤ √1 − 𝑥 4 ≤ 1.
1 1 1 1
⇒ ≤ ≤ ∀ 𝑥 ∈ [0, ]
1 √1 − 𝑥 4 √1 − 𝑥 2 2
1 1 1
Now, consider the functions √1−𝑥 4 − 1 and √1−𝑥 2 − √1−𝑥 4 . Both these functions are
1
continuous nonnegative functions and they do not vanish everywhere in [0, 2].
1⁄2 1 1⁄2 1 1
Hence ∫0 (√1−𝑥 4 − 1) 𝑑𝑥 , ∫0 (√1−𝑥 2 − √1−𝑥4 ) 𝑑𝑥 > 0.
1⁄2 𝑑𝑥 1⁄2 1⁄2 𝑑𝑥 1⁄2 𝑑𝑥
But this means ∫0 √1−𝑥 4 − ∫0 𝑑𝑥 > 0 and ∫0 √1−𝑥 2 − ∫0 √1−𝑥4 > 0.
We now use fundamental theorem Cauchy form to get the result.
(Write the details)