1

Topic: Taylor Series

Subject: Math AA HL
Research question:
How accurate is the Taylor Series approximation to the
original function?

Word count: 3974

 2

Table of contents:

Introduction…………………………………………………………………………………………………………………………………………………….3

Section 1 - Understanding the mathematical concepts…………………………………………………………………………………..…4

Section 2 - Special cases…………………………………………………………………………………………………………………………………13

Section 3 - Evaluating arctan⁡(𝑥 3 )………………………………………………………………………………………………………………….21

Conclusion.……………………………………………………………………………………………………………………………………………..…….24

Work cited.……………………………………………………………………………………………………………………………….…………...….….26
 3

Introduction
The Taylor series is a powerful mathematical tool used to approximate functions by a series of

polynomials. It is named after the mathematician Brook Taylor, who first introduced the idea in

his 1715 paper "Methodus Incrementorum Directa et Inversa" (Direct and Inverse Methods of

Incrementation). The Taylor series is a central concept in calculus, and its applications can be

seen in various fields, including physics and engineering.

I came across this topic when watching a math related YouTube video by 3Blue1Brown. I was

extremely fascinated by the fact that functions can be approximated using polynomials. I later

started wondering how accurate the approximation is or whether there is a gap? Or do the two

lines touch? This led to coming up with this topic and writing an extended essay about it.

This essay will explore the accuracy of the Taylor Series when approximating its original function

through the application of integral calculus to calculate the area between the original function

and its Taylor Series within a certain range. The smaller the area indicates a higher accuracy in

the approximation.
 4

Section 1 – Understanding the mathematical concept

What is the Taylor Series?

In mathematics, the Taylor Series or Taylor expansion of a function is an infinite sum of terms

that are expressed in terms of the function’s derivatives at a single point. In a more compact

sigma notation, the Taylor Series formula can be written as following:

∞
𝑓 (𝑛) (𝑎)
∑ (𝑥 − 𝑎)𝑛
𝑛!
𝑛=0

where n! denotes the factorial of n, and 𝑓 (𝑛) (𝑎) denotes the nth derivative of 𝑓 at the point 𝑎.

The expanded form of the Taylor Series looks like this:

∞
𝑓 (𝑛) (𝑎) 𝑓 ′ (𝑎) 𝑓 ′′ (𝑎) 𝑓 ′′′ (𝑎)
∑ (𝑥 − 𝑎)𝑛 = 𝑓(𝑎) + (𝑥 − 𝑎) + (𝑥 − 𝑎)2 + (𝑥 − 𝑎)3 + ⋯
𝑛! 1! 2! 3!
𝑛=0

In the expanded form, the point of the Taylor Series is to write any function in the form of a

polynomial, which when graphed can approximate the original function to a high accuracy within

a certain range. For some functions such as sin 𝑥, cos⁡(𝑥), or 𝑒 𝑥 ,the more terms added will lead

to a more accurate approximation and expands the range of accurate approximation. However,

for numerous other functions, adding more terms will only improve the accuracy within a certain

range otherwise called the radius of convergence. There is an issue of inaccuracy when the values

of 𝑥 evaluated at a is outside the radius of convergence which leads to the function diverging to

infinity. Both the functions and the radius of convergence will be explored later in this essay.
 5

Where is derived from?

The Taylor Series is derived from what is known as the “power series”. The power series is a tool

used commonly to analyze data and functions, because it allows any function to be written in the

form of a polynomial.

Consider a function that has a 𝑓 that has a power series at 𝑥 = 𝑎. Then the series would look like

this:

∑ 𝑐𝑛 (𝑥 − 𝑎)𝑛 = 𝑐0 + 𝑐1 (𝑥 − 𝑎) + 𝑐2 (𝑥 − 𝑎)2 + ⋯
𝑛=0

If the series is a representation of the function 𝑓 at 𝑥 = 𝑎, certainly we would like to evaluate

the series 𝑓(𝑥) = 𝑎. Evaluating the series at 𝑥 = 𝑎, it can be seen:

∑ 𝑐𝑛 (𝑥 − 𝑎)𝑛 = 𝑐0 + 𝑐1 (𝑎 − 𝑎) + 𝑐2 (𝑎 − 𝑎)2 + ⋯ = 𝑐0
𝑛=0

Thus, the series equals 𝑓(𝑎) if the coefficient 𝑐0 = 𝑓(𝑎). Additionally, we want the first derivate

of the power series to equal 𝑓 ′ (𝑎) at 𝑥 = 𝑎.

∞
𝑑
(∑ 𝑐𝑛 (𝑥 − 𝑎)𝑛 ) = 𝑐1 + 2𝑐2 (𝑥 − 𝑎) + 3𝑐3 (𝑥 − 𝑎)2 + ⋯
𝑑𝑥
𝑛=0

and when evaluated at 𝑥 = 𝑎, it’s simply equal to 𝑐1 . Following the same logic, the second

derivative of this power series evaluated at 𝑥 = 𝑎 is equal to 2𝑐2 and the third derivative is equal

to 6𝑐3 . More generally, we see that if 𝑓 has a power series representation at 𝑥 = 𝑎, then the

coefficients should be:

 6

𝑓 (𝑛) (𝑎)
𝑐𝑛 =
𝑛!

The 𝑐𝑛 gotten by evaluating infinite derivative iterations of a power series then evaluating⁡𝑥 at

the same value as 𝑎 which leads to getting the formula for the leading coefficient of each term

in a Taylor polynomial.

The Taylor Series of 𝑐𝑜𝑠(𝑥 )

Figure 1. 𝒇(𝒙) = 𝒄𝒐𝒔(𝒙) graph.

Suppose we want to build a Taylor Series for the function 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥) at the point 𝑥 = 0. The

first equation which comes to mind is a quadratic equation in the form of 𝑃(𝑥) = 𝑐0 + 𝑐1 𝑥 +

𝑐2 𝑥 2 . When 𝑐𝑜𝑠(𝑥) is evaluated at 𝑥 = 0, it is simply 1. When 𝑃(𝑥) is evaluated at 0, the function

is equal to 𝑐0 , which must equal 1. The first derivative of 𝑐𝑜𝑠(𝑥) = −𝑠𝑖𝑛(𝑥), and when evaluated
 7

at 𝑥 = 0 it’s 0. Derivative of 𝑃(𝑥) is 𝑐1 +2𝑐2 𝑥 at 𝑥 = 0 is 𝑐1 which has complete control of the

derivative of our approximation around 𝑥 = 0 . Setting it equal to 0 ensures that our

approximation has the same derivative as 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥), and hence the same tangent slope.

This leaves us free to change 𝑐2 , but the value and slope of our polynomial at 𝑥 = 0 are locked in

place to match that of 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥). The cosine graph curves downwards above 𝑥 = 0 which

means it has a negative second derivative. The second derivative of 𝑐𝑜𝑠(𝑥) is −𝑐𝑜𝑠(𝑥) which is

equal to −1 at 𝑥 = 0. The second derivative of 𝑃(𝑥) is 2𝑐2 and to make sure that the second

1
derivative of this polynomial is equal to −1, 2𝑐2 must equal 0, making 𝑐2 = − 2. This gives us the

1
approximation 𝑃2 (𝑥) = 1 + 0𝑥 + (− 2) 𝑥 2 .

Evaluating the original function and first

derivative at 𝑥 = 0 gave us the tangent

line at 𝑦 = 0 . The second derivative

ensured the negative rate of change

within a certain interval around 𝑥 = 0

was similar giving us the best quadratic

estimate of 𝑐𝑜𝑠(𝑥).

𝒙𝟐
Figure 2. Graphs of⁡𝒇(𝒙) = 𝒄𝒐𝒔(𝒙), the tangent line 𝒚 = 𝟏, and the quadric approximation 𝑷𝟐 (𝒙) = 𝟏⁡– .
𝟐

Now suppose we want to add more terms in order to to increase the accuracy of the

1
approximation. We add a term 𝑐3 𝑥 3 to 𝑃(𝑥) which is now equal to 1 − 2 𝑥 2 + 𝑐3 𝑥 3 . When we
 8

𝑑3 𝑃
find the 3rd derivative of 𝑐𝑜𝑠(𝑥) which is sin(𝑥) and evaluated at 𝑥 = 0 is 0. 𝑑𝑥 3 (𝑥) = 6𝑐3 . It

1
means the constant 𝑐3 = 0.⁡ In other words, 1 − 2 𝑥 2 is both the best quadratic and cubic

𝑑4 𝑐𝑜𝑠
approximation. An improvement can be made by adding a fourth order term, 𝑐4 𝑥 4 . (𝑥) is
𝑑𝑥 4

itself, 𝑐𝑜𝑠(𝑥), and 𝑐𝑜𝑠(0) = 1. The 4th derivative of our polynomial is 24𝑐4 . If we want the

1
derivatives to match, 𝑐4 must equal 24, creating the best quartic approximation of 𝑐𝑜𝑠(𝑥) which

𝑥2 𝑥4
is 𝑃4 (𝑥) = 1 − + 24.
2

By doing higher order derivatives,

we find that the derivatives of

𝑐𝑜𝑠(𝑥) repeat every 4 iterations

and at every odd 𝑛𝑡ℎ ⁡iteration, the

value is 0 when evaluated at 0. This

Figure 3. Graphs of 𝒇(𝒙) = 𝒄𝒐𝒔(𝒙) and 𝑷𝟒 (𝒙) = 𝟏 −

𝒙𝟐
𝟐
+
𝒙𝟒
𝟐𝟒
. explains why there isn’t a cubic

term, and won’t be a quintic, or septic. For 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥), of we want to expand the range of

accuracy of the polynomial approximation, the function after the 4 th degree term would look as

following:

𝑥 2 𝑥 4 𝑥 6 𝑥 8 𝑥10 𝑥12
𝑃𝑛 (𝑥) = 1 − + − + − + −⋯
2 24 6! 8! 10! 12!
 9

When comparing the two graphs,

it’s visible that the range of

accuracy increased twice from less

than 𝑥 = 1.5⁡(Figure 3) to over 𝑥 =

4 (Figure 4)

𝒙𝟐 𝒙𝟒 𝒙𝟔 𝒙𝟖 𝒙𝟏𝟎 𝒙𝟏𝟐
Figure 4. Graphs of 𝒇(𝒙) = 𝒄𝒐𝒔(𝒙) and 𝑷𝟏𝟐 (𝒙) = 𝟏 − + − + − +
𝟐 𝟐𝟒 𝟔! 𝟖! 𝟏𝟎! 𝟏𝟐!

How is error relevant and calculated?

Now that the fundamentals of Taylor Series are explored, we continue to evaluate the error. The

error in this context is the difference between two curves. In order to achieve this, integral

calculus is needed to calculate the area between two curves within a certain interval. The smaller

the area indicates that the approximation is more accurate. An issue with error calculation is

choosing boundaries. From the function 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥) which we used previously, we found out

that the more terms a polynomial has, the greater the range of accuracy of the approximation. It

𝑥2 𝑥4
means, when originally, we had the quartic approximation, 1 − + 24 , the range of 𝑥
2

boundaries we could pick to calculate the area could be (−1.5, 1.5) or (−2,2). However, when

𝑥2 𝑥4
the more accurate approximation is used, which is a polynomial of 12th degree, 1 − + 24 −
2

𝑥6 𝑥8 𝑥 10 𝑥 12
+ − 10! + 12! , the 𝑥 range of accuracy increase by nearly 3 times. When we use the same
6! 8!

boundaries, the area is equal to a value near 0 proving that this polynomial is much more accurate.
 10

But suppose we have a polynomial of 20th degree, then whatever boundary we chose for the 12th

degree polynomial is going to be cancelled out and this phenomenon will continue infinitely. A

solution to this problem is to compare the approximation to the next degree approximation, and

not a polynomial 8 degrees higher than the original.

The formula for calculating the area between two curves is this:

𝑏
𝐴 = |∫ (𝑓(𝑥) − 𝑔(𝑥))𝑑𝑥|
𝑎

where 𝑎 is the lower 𝑥 boundary, 𝑏 is the upper 𝑥 boundary, 𝑓(𝑥) is the function above and

𝑔(𝑥) is the function below. It is in absolute value brackets, because under certain circumstances,

the answer gotten could be negative, yet an area cannot be negative in any way.

Calculating the error between the original function of 𝑐𝑜𝑠(𝑥 ) and its
Taylor polynomials
Now that the Taylor Series, Taylor Series of 𝑐𝑜𝑠(𝑥), and error between two curves have all been

explored, we can try to calculate the accuracy of the polynomial approximation of 𝑐𝑜𝑠(𝑥). I

decided to go with the value of 0.0001⁡𝑢𝑛𝑖𝑡𝑠 2 as a base number for the accurate approximation

of the original function. After evaluating the area, if the area is smaller than that value, the Taylor

polynomial will be deemed an accurate approximation. There are 2 possible ways to achieve such

a result. One is having a fixed number of terms in the Taylor polynomial and then decreasing the

range at which it is evaluated. The second way is to fix the interval at which the polynomial is

evaluated but adding on more terms to the polynomial. Here, I picked a fixed interval and added

on more terms to the polynomial. I begin with going back to Figure 3, where both 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥)
 11

𝑥2 𝑥4 𝜋 𝜋
and 𝑃4 (𝑥) = 1 − + 24 are graphed. A reasonable 𝑥⁡interval to pick would be (− 2 , 2 ) since it’s
2

where the approximation starts to drift away and when working with 𝑐𝑜𝑠(𝑥) and 𝑠𝑖𝑛(𝑥) it’s

easier to calculate with 𝜋. Here we get:

𝜋
2
𝑥2 𝑥4
𝐴 = ∫ [(1 − + ) − cos(𝑥)]
2 24
𝜋
−
2

𝜋
𝑥3 𝑥5
𝐴=𝑥− + − sin(𝑥)| 2𝜋
6 120 −2

𝜋 𝜋3 𝜋5 𝜋 𝜋3 𝜋5
𝐴=( − + − 1) − (− + − + 1)
2 48 3840 2 48 3840

𝐴 = 0.00905⁡𝑢𝑛𝑖𝑡𝑠 2

0.00905⁡𝑢𝑛𝑖𝑡𝑠 2 is the area between the original function 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥) and its Taylor

𝑥2 𝑥4
polynomial 𝑃4 (𝑥) = 1 − + 24. The number seems quite small and maybe even insignificant.
2

𝑥2 𝑥4 𝑥6
However, when we add just one more term to the Taylor polynomial 𝑃6 (𝑥) = 1 − + 24 − ,
2 6!

and calculate the area between 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥) and the polynomial, the following result is

reached:

𝜋
2
𝑥2 𝑥4 𝑥6
𝐴 = ∫ [cos(𝑥) − (1 − + − )⁡] 𝑑𝑥
2 24 720
𝜋
−
2

𝜋
𝑥3 𝑥5 𝑥7
𝐴 = |[sin(𝑥) − (𝑥 − + − )]| 2
6 120 5040 − 𝜋
2
 12

𝐴 = 0.000314⁡𝑢𝑛𝑖𝑡𝑠 2

By just adding one more term, the area was decreased by 96.5%. However, this value is still not

the desired one, therefore we must add another term and evaluate the area:

𝜋
2
𝑥2 𝑥4 𝑥6 𝑥8
𝐴 = ∫ |[(1 − + − + ) − cos⁡(𝑥)]| 𝑑𝑥
2! 4! 6! 8!
𝜋
−
2

𝐴 = 0.000007086⁡𝑢𝑛𝑖𝑡𝑠 2

With 5 terms in our Taylor polynomial of 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥), the area between the approximation

and the original function decreased below the desired value of 0.0001⁡𝑢𝑛𝑖𝑡𝑠 2 . Naturally, with

other functions and their own Taylor polynomials, different amounts of terms will need to be

added in order to achieve an area below 0.0001⁡𝑢𝑛𝑖𝑡𝑠 2 . It doesn’t mean all Taylor polynomials

to the 5th degree will achieve the same result.

Now we begin to explore the other method of achieving accuracy, which is leaving the polynomial

untouched, but evaluating the integral at different decreasing boundary, the Taylor polynomial

𝑥2 𝑥4 𝜋 𝜋
of 𝑓(𝑥) = cos(𝑥) is 𝑃4 (𝑥) = 1 − + 24 evaluated at (− 2 , 2 ). For the calculations to look a bit
2

𝜋 𝜋
nicer, we leave the intervals in terms of radians and shrink it to (− , ).
3 3

𝜋
3
𝑥2 𝑥4
𝐴 = ∫ ((1 − + ) − cos(𝑥)) 𝑑𝑥
2 24
−𝜋/3

𝐴 = 0.00053976𝑢𝑛𝑖𝑡𝑠 2
 13

𝜋 𝜋
The value we got isn’t the desired one, so we continue shrinking the x interval to (− 4 , 4 ).

𝜋
4
𝑥2 𝑥4
𝐴 = ∫ ((1 − + ) − cos(𝑥)) 𝑑𝑥
2 24
𝜋
−
4

𝐴 = 0.00007253𝑢𝑛𝑖𝑡𝑠 2

Both methods work when it comes to measuring how accurate the Taylor polynomial estimation

is. The number will just be different for other functions. There is no special pattern.

Section 2 – Special cases

Evaluating the function 𝑓(𝑥 ) = 𝑒 𝑥
The reason the function ⁡𝑒 𝑥 is in the special cases section is because when we deduce it’s Taylor

Series, it touches upon another mathematical concept which proves the accuracy of the

approximation without the use of integral calculus to calculate the area between two curves. For

starters, when we compute the derivatives of 𝑒 𝑥 we find that it’s all the same as the original

function. When we evaluate it at 𝑥 = 0 all derivatives equal 1. Thus, the Taylor Series of

𝑓(𝑥) = 𝑒 𝑥 looks like the following:

𝑥2 𝑥3 𝑥4
𝑃(𝑥) = 1 + 𝑥 + + + + ⋯⁡
2! 3! 4!
 14

Figure 5. 𝒇(𝒙) = 𝒆𝒙 and its Taylor Series

The graph shows a fourth-degree polynomial that estimates the function of 𝑒 𝑥 . As more terms

are added, the approximation will become infinitely closer to the original graph.

When we evaluate the Taylor Series at 𝑥 = 1:

1
12 13
𝑇𝑓 (1) = 1 + 1 + + + ⋯
2! 3!

we find that the answer converges to the actual value of e which is 2.71828… In this case, to

calculating the accuracy with area is not necessary, because for any input of 𝑥, the answer will

always converge to the value of 𝑒 𝑥 (for whatever 𝑥 is). Here, the approximation is more accurate

when more terms are added to the Taylor Series. It makes sense to add an infinite number of

terms in order to get the most accurate approximation. This is also true when we deduce the

Taylor Series itself at different 𝑥 values. Suppose at 𝑥 = 2, our Taylor Series would look as such:
 15

7.39 7.39 7.39

𝑃(𝑥) = 7.39 + 7.39(𝑥 − 2) + (𝑥 − 2)2 + (𝑥 − 2)3 + (𝑥 − 2)4 + ⋯
2 3! 4!

Figure 6. 𝒇(𝒙) = 𝒆𝒙 and its Taylor polynomials to the 4th, 10th, 20th, and 50th degree

The graph shows:

𝑓(𝑥) = 𝑒 𝑥

7.39 7.39 7.39

𝑃4 (𝑥) = 7.39 + 7.39(𝑥 − 2) + (𝑥 − 2)2 + (𝑥 − 2)3 + (𝑥 − 2)4
2 6 24

10
7.39
𝑃10 (𝑥) = 7.39 + ∑ (𝑥 − 2)𝑛
𝑛!
𝑛=1

20
7.39
𝑃20 (𝑥) = 7.39 + 7.39 + ∑ (𝑥 − 2)𝑛
𝑛!
𝑛=1

50
7.39
𝑃50 (𝑥) = 7.39 + 7.39 + ∑ (𝑥 − 2)𝑛
𝑛!
𝑛=1
 16

As more terms are added, the approximation converges to the value of 𝑒 2 thus, making its Taylor

Series more and more accurate.

Evaluating 𝑙𝑛(𝑥)
The concept of adding infinite number of terms to get the best approximation is true for a few

other important functions other than 𝑒 𝑥 such as 𝑐𝑜𝑠(𝑥)⁡𝑜𝑟⁡𝑠𝑖𝑛(𝑥). However, this concept is

what connects us to the next topic about intervals. For most of the functions, the approximation

is only accurate within a certain interval. When we work out the Taylor Series 𝑙𝑛(𝑥)⁡around the

input 𝑥 = 1, it looks like this:

(𝑥 − 1)2 (𝑥 − 1)3 (𝑥 − 1)4

𝑇𝑓 (𝑥) = (𝑥 − 1) − + − +⋯
2 3 4

Figure 7. 𝒍𝒏(𝒙), its Taylor polynomial to the 4th degree and 9th degree
 17

Graph represents:
𝑙𝑛(𝑥)

(𝑥 − 1)2 (𝑥 − 1)3 (𝑥 − 1)4

(𝑥 − 1) − + −
2 3 4

(𝑥 − 1)2 (𝑥 − 1)3 (𝑥 − 1)4 (𝑥 − 1)5 (𝑥 − 1)6 (𝑥 − 1)7 (𝑥 − 1)8 (𝑥 − 1)9

(𝑥 − 1) − + − + − + − +
2 3 4 5 6 7 8 9

When we analyze the 𝑛𝑡ℎ degree polynomial for 𝑙𝑛(𝑥), we find that it is only accurate within a

certain 𝑥 range, which in this case is (0,2). When we pick any 𝑥 value within this interval, the

series converge to 𝑙𝑛(𝑥) for whatever 𝑥⁡is. However, when we pick a value outside the range,

then the series fails to approach anything. As we add more and more terms, the sum just bounces

widely up and down. In this case, when adding more terms to a series doesn’t approach anything

means it is divergent. This interval can be shifted when we evaluate the function at different 𝑥

values. Suppose we get the Taylor Series at 𝑥 = 3 , it would mean the accurate range of

approximation is between (2,4). Anything exceeding those values will diverge.

The interval between the 𝑥 value we evaluated the series at and the limit, is called a “radius of

convergence”. It shows the values around the 𝑥⁡value at which the series was evaluated to depict

at what point is convergence and from what point is starts to diverge.

When we move the 𝑥 value to 𝑎 = 2, both the Taylor Series and radius of convergence look

different. First, let’s find the Taylor Series of 𝑙𝑛(𝑥) at 𝑎 = 2. Constructing this table will help

putting the series together:

𝑛 𝑓 𝑛 (𝑥) 𝑓 𝑛 (𝑎) 𝑛!

0 ln⁡(𝑥) ln⁡(2) 1
 18

1 1 1 1
𝑥 2

2 1 1 2
− −
𝑥2 4

3 2 1 6
𝑥3 4

4 6 3 24
− −
𝑥4 8

When put as a Taylor polynomial of 4th degree, it looks like this:

𝑥 − 2 (𝑥 − 2)2 (𝑥 − 2)3 (𝑥 − 2)4

𝑃4 (2) = ln(2) + − + −
2 8 24 64

To find the radius of convergence, we must do a ratio test:

𝑎𝑛+1
lim | |=𝐿
𝑛→∞ 𝑎𝑛

To make this operation easier, we must make the polynomial into a sum (series in terms of sigma).

The Taylor series now appears as such:

∞
(𝑥 − 2)𝑛
ln(2) + ∑(−1)(𝑛+1) ⁡ ⁡
2𝑛 𝑛
𝑛=1

Doing the ratio test, all terms of 𝑛 now must be substituted by 𝑛 + 1, then the new equation will

be divided by the original one. The ratio test looks like the following:

(𝑥 − 2)𝑛+1 2𝑛 𝑛
𝐿 = lim |(−1)𝑛+2 ⁡ |
𝑛→∞ 2𝑛+1 (𝑛 + 1) (𝑥 − 2)𝑛 (−1)𝑛+1

After simplification:
 19

𝑥−2 𝑛
𝐿 = |− | lim | |
2 𝑛→∞ 𝑛 + 1

1
𝐿= (𝑥 − 2)
2

According to the rules of radius of convergence, when we get a limit (𝐿) in the form of

1
|𝑥 − 𝑐| < 1, where 𝑐 is the center point, to find 𝑐, the equation must equal 0, which in this case
𝑅

is at⁡𝑥 = 2. Then |𝑥 − 𝑐| < 𝑅, where R is the radius of convergence. Our R=2. After, in order to

find the full 𝑥 range in which the function diverges, we must evaluate −𝑅 < 𝑥 − 𝑐 < 𝑅, so −2 <

𝑥 − 2 < 2. Simplifying to 0 < 𝑥 < 4.

𝒙−𝟐 (𝒙−𝟐)𝟐 (𝒙−𝟐)𝟑 (𝒙−𝟒)𝟒

Figure 8. 𝒇(𝒙) = 𝒍𝒏(𝒙) and 𝑷𝟒 (𝟐) = 𝒍𝒏(𝟐) + − + −
𝟐 𝟖 𝟐𝟒 𝟔𝟒

This trend will continue with an increase in the value of 𝑥. When we evaluate the function at 𝑎 =

3,⁡the radius of convergence is going to be 3 and the domain at which the function converges is
 20

(0,6). 0 is always going to be the bottom domain because the function infinitely approaches 0 as

𝑙𝑛(𝑥) is graphed.

Why do functions converge?

All functions have one main rule: the y value is dependent on the x value. This means all functions

written in the form of a polynomial will drift off into positive infinite values of 𝑦 or −𝑦, since 𝑥

value is chosen at one’s discretion. The function of 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥) or 𝑓(𝑥) = 𝑠𝑖𝑛(𝑥) which we

have explored before have a repetitive pattern of going up and down. This means that the

polynomials will infinitely expand due to increasing chosen value where the 𝑦 depends on the 𝑥.

So far, we have explored 3 functions: 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥), 𝑓(𝑥) = 𝑒 𝑥 , and 𝑓(𝑥) = 𝑙𝑛(𝑥). We found

out that 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥) will converge for any value of 𝑥 and the more terms we add to the Taylor

polynomial of 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥) the more accurate the approximation and so does the span of the

approximation increase. This is because 𝑓(𝑥) = 𝑐𝑜𝑠(𝑥) in the form of a graph is a repeating

pattern. The maximums and minimums will stay the same as the graph approaches ∞, or −∞.

Therefore, for any value of 𝑥, the Taylor polynomial will always converge to that value at which

𝑓(𝑥) = 𝑐𝑜𝑠(𝑥) is evaluated at. When it comes to 𝑒 𝑥 , it’s not quite as similar, but it also converges

at any 𝑥 value. This is because its derivatives remain the same no matter the number of iterations

is done. When the derivative at any given point is the same, we find ourselves building a Taylor

polynomial which will converge to whatever value we evaluate 𝑒 𝑥 at. In contrast, 𝑙𝑛(𝑥) doesn’t

work in the same fashion. ln(𝑥) drifts off into infinite values of 𝑥 which a polynomial is unable to

achieve, therefore evaluating the function at greater 𝑥 values will always increase the radius of

convergence; however, it will never converge infinitely at one chosen 𝑥 value.

 21

Section 3 - Evaluating 𝐚𝐫𝐜𝐭𝐚𝐧(𝐱 𝟑 )

The graph of 𝑎𝑟𝑐𝑡𝑎𝑛 is a function like 𝑓(𝑥) = 𝑙𝑛(𝑥) where the graph drifts off to infinity along

the 𝑥-axis rather than the 𝑦-axis. This means the function is likely to have a radius of convergence,

since all functions written in the form of a polynomial drifts to infinity vertically along the 𝑦-axis.

After following all steps, the Taylor polynomial to the 4th degree of 𝑎𝑟𝑐𝑡𝑎𝑛(𝑥 3 ) at 𝑥 = 1 is:

𝜋 3(𝑥 − 1) 3(𝑥 − 1)2 7(𝑥 − 1)3

𝑇𝑓 (1) = + − − + 3(𝑥 − 1)4
4 2 4 4

𝝅 𝟑 𝟑 𝟕
Figure 9. 𝒇(𝒙) = 𝒂𝒓𝒄𝒕𝒂𝒏𝒙𝟑and 𝑻𝒇 (𝟏) = + (𝒙 − 𝟏) − (𝒙 − 𝟏)𝟐 − (𝒙 − 𝟏)𝟑 + 𝟑(𝒙 − 𝟏)𝟒
𝟒 𝟐 𝟒 𝟒
 22

Logically, the next step is to find the radius of convergence, however the Taylor polynomial above

cannot be written in the form of a sigma notation, therefore we must evaluate the Taylor Series

of 𝑓(𝑥) = 𝑎𝑐𝑡𝑎𝑛(𝑥 3 ) at 𝑥 = 0 in which case we get a very nice Taylor polynomials:

𝑥 9 𝑥15 𝑥 21 𝑥 27 𝑥 33
𝑇𝑓 (0) = 𝑥 3 − + − + − −⋯
3 5 7 9 11

Figure 10. 𝒇(𝒙) = 𝒂𝒓𝒄𝒕𝒂𝒏(𝒙𝟑 ) and its Taylor Series (𝑻𝒇 (𝟎))
 23

A clear pattern can be seen in both the denominator and numerator which is very convenient to

write the series in sigma notation to calculate the radius of convergence:

∞
(−1)𝑛 (𝑥 3 )1+2𝑛
𝑇𝑓 (0) = ∑
1 + 2𝑛
𝑛=0

As mentioned above, to find the radius of convergence, a ratio test must be done:

(−1)𝑛+1 (𝑥 3 )2𝑛+1 2𝑛 + 1
𝐿 = lim | ⁡|
𝑛→∞ 2𝑛 + 3 (−1)𝑛 (𝑥 3 )2𝑛+1

2𝑛 + 1
𝐿 = |−𝑥 6 | lim | |
𝑛→∞ 2𝑛 + 3

𝐿 = |𝑥 6 |

The limit we got is |𝑥 6 |, which means there is convergence when |𝑥 6 | < 1. Such a result indicates

that the radius of convergence is −1 < 𝑥 < 1.

Since a clear convergence interval has been established, determining the accuracy of the Taylor

Series by shrinking or expanding the interval will not be necessary because we want to find out

how accurate the series is within the radius of convergence. Accuracy will solely depend on the

number of terms in the polynomial. The desired area between the curves remains untouched at

0.0001⁡𝑢𝑛𝑖𝑡𝑠 2 .

There is one problem with this graph. That is at positive 𝑥 values, the Taylor Series lies above

𝑓(𝑥) = 𝑎𝑟𝑐𝑡𝑎𝑛(𝑥 3 ). However, at negative 𝑥 values 𝑓(𝑥) = 𝑎𝑟𝑐𝑡𝑎𝑛(𝑥 3 ) lies above its Taylor

Series. This means it’s possible to calculate either two separate integrals (from -1 to 0 and from

0 to 1) or choose either negative or positive side to calculate the area and multiply it by 2, since
 24

the graph is symmetrical. The latter method is much simpler since it doesn’t involve calculating

two integrals.

1
𝑥 9 𝑥15 𝑥 21 𝑥 27 𝑥 33
𝐴 = ∫ |[(𝑥 3 − + − + − ) − (arctan(𝑥 3 ))]| 𝑑𝑥
3 5 7 9 11
0

𝐴 = 0.00110445 × 2 = 0.0022089⁡𝑢𝑛𝑖𝑡𝑠 2

After testing numerous amounts of testing, the Taylor polynomial to the 29th degree will result

in an area less than 0.0001⁡𝑢𝑛𝑖𝑡𝑠 2 :

1 29
(−1)𝑛 (𝑥 3 )1+2𝑛
𝐴 = ∫ |[(∑ ) − (arctan(𝑥 3 ))]| 𝑑𝑥
1 + 2𝑛
0 𝑛=0

𝐴 = 0.0000492175 × 2 = 0.0000984⁡𝑢𝑛𝑖𝑡𝑠 2

A desired area smaller than 0.0001⁡𝑢𝑛𝑖𝑡𝑠 2 was gotten after writing 29 terms in the

approximation for 𝑓(𝑥) = 𝑎𝑟𝑐𝑡𝑎𝑛(𝑥 3 ).

Conclusion
This essay explored the accuracy of Taylor polynomials, and it was found that there are two ways

to achieve high accuracy: to move the boundaries at which the areas are calculated, or to increase

the number of terms in the polynomial approximation. However, using Taylor Series it is

impossible to achieve definite accuracy where the area between the two curves is equal to zero,

since a polynomial can only mimic the shape of the original function. The Taylor Series does not

have a lot of applications in the real world. It’s mostly used in university level mathematics and
 25

physics. Generally, it is a powerful tool to use when approximating functions, but there are many

complications. They include calculating a massive number of terms in order to get an

approximation. There is also the issue of convergence. Many Taylor Series have a radius of

convergence which limits the use of the concept. I explored the general concept of this and

explored its accuracy. For accuracy I picked an easier way of calculating, which is using integrals

and finding the area between two curves. There is a more professional way to find the error in

the Taylor Series, which is called the Taylor Remainder Theorem, however it’s more university

level mathematics so I did not go down that path. It’s also the reason I had to pick the value

0.0001⁡𝑢𝑛𝑖𝑡𝑠 2 as a basis to evaluate the accuracy. Naturally, this essay only explored a small

proportion of the aspects of Taylor Series. There are many more which can be implemented to

maximize the use of Taylor Series.

 26

Bibliography
"Area between Two Curves." Youtube, uploaded by The Organic Chemistry Tutor, 20 Apr. 2021,

www.youtube.com/watch?v=kgg5Rspf1Js&ab_channel=TheOrganicChemistryTutor. Accessed 14

Apr. 2023.

“Website title missing.” Derivative Calculator. www.derivative-calculator.net/. Accessed 14 May 2023.

Desmos. www.desmos.com/calculator/7ojxm6gygn. Accessed 14 May 2023.

Gilbert Stang, and Edwin Jed Herman. "Area between Curves." Find the Name of the Actual Book and cite this

as a book on a website. LibreTexts,

math.libretexts.org/Bookshelves/Calculus/Calculus_(OpenStax)/06%3A_Applications_of_Integration/

6.01%3A_Areas_between_Curves#:~:text=To%20find%20the%20area%20between%20two%20curve

s%20defined%20by%20functions,the%20difference%20of%20the%20functions. Accessed 14 Apr.

2023.

Integral Calculator. www.integral-calculator.com/. Accessed 14 May 2023.

"Power Series - Finding the Radius and Interval of Convergence - Calculus 2." Youtube, uploaded by The

Organic Chemistry Tutor, 1 Apr. 2018, www.youtube.com/watch?v=EGni2-

m5yxM&ab_channel=TheOrganicChemistryTutor. Accessed 16 Apr. 2023.

Strang, Gilbert, and Edwin Jed Herman. "Taylor and Maclaurin Series." LibreTexts,

math.libretexts.org/Bookshelves/Calculus/Calculus_(OpenStax)/10%3A_Power_Series/10.03%3A_Ta

ylor_and_Maclaurin_Series. Accessed 14 Apr. 2023.

StudySmarter. www.studysmarter.co.uk/explanations/math/calculus/taylor-series/. Accessed 12 Apr. 2023.

"Taylor Series | Chapter 11, Essence of Calculus." Youtube, uploaded by 3Blue1Brown, 7 May 2017,

www.youtube.com/watch?v=3d6DsjIBzJ4&ab_channel=3Blue1Brown. Accessed 13 Apr. 2023.

27