MATHEMATICS SL INVESTIGATION

Methods of approximating sin(x) as an algebraic function

1
Introduction

The function 𝑠𝑖𝑛(𝑥) is an elementary trigonometric relation often defined in terms of

the unit circle as the vertical distance between the x-axis and the point on the unit circle that

meets a line subtended by angle 𝑥 (in radians). It exhibits the property of periodicity in that

its values and the shape of its graph are repeated after an interval of 2𝜋. This makes the

function an essential tool in developing mathematical models and generalised formulas for

periodic phenomena such as temperature cycles or tidal wave behaviour. However, deriving

values for 𝑠𝑖𝑛(𝑥) is surprisingly difficult without the use of a calculator or other

computerised machinery. Unlike algebraic functions, which can be expressed in terms of a

finite combination1 of algebraic operations, 𝑠𝑖𝑛(𝑥) is a transcendental function; it can

only be defined in terms of other trigonometric expressions, or an infinite series of

polynomials. For centuries, solving for sines often meant tediously examining an elaborate

trigonometric table and finding the corresponding value of 𝑠𝑖𝑛(𝑥) for each required instance

of 𝑥. This is an impractical process and having to “search” for the values of a function

seems counter-intuitive to the search for generalisation and broad applicability in

mathematics.

In my IB physics class, we were taught to replace the function 𝑠𝑖𝑛(𝑥) with simply the

angle 𝑥 itself for any “small” values of 𝑥 in radians. It allowed us to derive a much simpler

formula for wave diffraction without compromising the theoretical accuracy of the formula.

I became increasingly curious about what “small” truly meant, and in my quest to

understand the limits of this approximation, I realised that 𝑠𝑖𝑛(𝑥) — or any other function

— could actually be algebraically approximated in a number of ways. This investigation

will explore the mathematical accuracy, rationale, and practical limitations of a few key

methods of approximating the function 𝑠𝑖𝑛(𝑥). The success of each approximation depends

1
“Transcendental Function.” Encyclopædia Britannica, Encyclopædia Britannica, Inc., 15 Apr. 2011,
www.britannica.com/science/transcendental-function. Accessed 27 Jan. 2020.
2
on its ability to map 𝑠𝑖𝑛(𝑥) to a simpler algebraic form that would enable direct calculation

and allow us to make less obtrusive models without compromising significant accuracy.

Substituting x for sin(x): “The Small-Angle Approximation”

The aforementioned method of approximating 𝑠𝑖𝑛(𝑥) to 𝑥 for “small angles” is a

common practice in mechanical physics. The accepted rationale behind this substitution is

that the graph of 𝑠𝑖𝑛(𝑥) roughly coincides with the graph of 𝑓(𝑥) = 𝑥 for angles very close

to zero. Plotting the graph of both functions in the domain −2𝜋 to 2𝜋 shows that they are

seemingly comparable around the origin.

Approximation

sin(𝑥)

Graph 1: The small-angle approximation and sin(x)

However, visual similarity at the low resolution of this document is hardly rigorous

mathematical confirmation. We can plot an error squared function for the approximation by

taking the square of the difference between the two functions: (𝑠𝑖𝑛(𝑥) − 𝑥)! .

3
 Graph 2: The squared error in the linear approximation (sin(x) - x) 2

" "
We observe that the squared error is less than 0.00001 for angles between − #! and #!,

which would imply that this linear approximation is accurate up to two decimal places in

this domain. It is important to note that the error reduces consistently as 𝑥 approaches zero,

so the smaller the angle, the better the approximation. In the motion of a simple pendulum,
"
where maximum angular displacement 𝜃 is most often below #!, using this approximation

can give us a simpler equation for calculating the acceleration of the pendulum bob (where

L is the length of the pendulum cord).

The theoretical formula is:

$
𝑎 = −𝑔𝑠𝑖𝑛(% )

Using the small-angle approximation, we eliminate the sine function entirely, leaving

us with:
$
𝑎 = −𝑔(% )

Although our function loses its mathematical accuracy to a certain extent, it is now

possible to calculate a number of physical quantities with simple algebraic operations,

including the acceleration of the bob and the force exerted by it in its direction of motion. In

cases such as this, where the domain of 𝑥 is known to be restricted to values near zero (𝑥 →

4
0), equating 𝑠𝑖𝑛(𝑥) to 𝑥 can be practically useful. The definition of “small angle” — the

range in which this approximation can be used — depends on the level of accuracy that is

appropriate or necessary for a given situation but taking smaller angles will reduce the error

in the approximation increasingly as 𝑥 approaches zero.

Simple Quadratic Approximation

The key issue with using linear approximations for trigonometric functions is that they

cannot mimic the curvature of the graphs of such functions, and thus an approximation such
"
as 𝑠𝑖𝑛(𝑥) = 𝑥 becomes extremely inaccurate as 𝑠𝑖𝑛(𝑥) approaches its first maximum at ! .

We can obtain a simpler function that approximates the parabolic shape of 𝑠𝑖𝑛(𝑥) between

0 and 𝜋 by using quadratics instead2.

We know that any quadratic function can be expressed in the form 𝑎𝑥 ! + 𝑏𝑥 + 𝑐,

where 𝑎, 𝑏 and 𝑐 are the coefficients of the expression and 𝑎 ≠ 0. The graph of 𝑠𝑖𝑛(𝑥) in

the domain 0 to 𝜋 resembles a downward facing parabola, with roots 0 and 𝜋, and a
"
maximum at !. Using this information, we obtain 3 simultaneous equations that we can

solve for 𝑎, 𝑏 and 𝑐:

0 = 𝑎(0) + 𝑏(0) + 𝑐
" "
1 = 𝑎(! )! + 𝑏( !) + 𝑐

0 = 𝑎(𝜋)! + 𝑏(𝜋) + 𝑐
&' '
Which gives us the values: 𝑎 = , 𝑏 = ", 𝑐 = 0. Plugging these into the quadratic
"!

formula, the equation becomes:

&'( ! '(
𝑦= + +0
"! "

Using 𝜋 ! as a common denominator, we get:

2
Berry, Nick. “Approximating the Sine Function.” Data Genetics, 1 July 2019,
datagenetics.com/blog/july12019/index.html. Accessed 3 Dec. 2019.
5
 𝑥
4𝑥(1 − )
→𝑦= 𝜋
𝜋
4𝑥(𝜋 − 𝑥)
→𝑦=
𝜋!

Approximation

sin(𝑥)

Graph 3: The simple quadratic approximation plotted against sin(x)

Plotting the two functions together in Graph 3, we can see that the approximated

parabola does roughly resemble the shape of the sine graph between 0 and 𝜋, with the two

graphs becoming closer to each other at the 3 points we used to formulate the approximation
"
— (0,0), ( !, 1) and (𝜋, 0).

Graph 4: Squared-error in the quadratic approximation

6
 The peak value of the error graph in Graph 4 is 0.003137 between 0 and 𝜋, showing

that this approximation is fairly accurate up to 1 decimal place in the chosen domain. The

fixed points we used to derive the quadratic function are still present in this approximation,
"
which means that it will be even more accurate for values surrounding 0, and 𝜋. The
!

fundamental issue with using such an approximation is that it would become highly

inaccurate beyond this domain due to the nature of quadratic functions — while a pure sine

function has an endless (infinite) number of upward and downward curves, a quadratic

function will only formulate one parabola, and will move away from the x-axis continuously

for any values of 𝑥 beyond that. This method would be appropriate and efficient for

approximating 𝑠𝑖𝑛(𝑥) in certain cases where the range of the function required is relatively

small (up to 𝜋) as quadratic graphs only have two roots, but they can roughly mimic the

shape of the sine curve in any one direction. In order to form approximations with more

roots and thus cover a larger domain, we must use polynomials of a higher degree.

Taylor Polynomials and Higher Degree Approximations

The Taylor series, also known as the Maclaurin series when it is centred at zero, is a

method of representing any function as an infinite sum of algebraic terms using various

derivatives of the original function3. Theoretically, the infinitely long Taylor series of a

given function will be identical to the original function, but if we take a limited number of

terms in the series (or “truncate” it), we can obtain useful approximations. The formula for

calculating Taylor polynomials is as follows:

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

3
“Taylor Series Approximation.” Brilliant Math & Science Wiki, brilliant.org/wiki/taylor-series-
approximation/. Accessed 12 Feb. 2020.
7
 Unlike quadratic functions, which can only produce a single parabolic curve, using the

Taylor series can produce higher degree polynomial approximations. Let us denote the

Taylor polynomial for 𝑠𝑖𝑛(𝑥) where 𝑎 = 0:

𝑐𝑜𝑠(0) 𝑠𝑖𝑛(0) 𝑐𝑜𝑠(0)
𝑠𝑖𝑛(0) + (𝑥 − 0) − (𝑥 − 0)( − (𝑥 − 0)) +. ..
1! 2! 3!

As any term involving 𝑠𝑖𝑛(0) will have a value of zero, the series becomes:
𝑥) 𝑥* 𝑥+
sin(𝑥) = 𝑥 − + − +⋯
3! 5! 7!

We can see that the first term in this series is in fact 𝑥, the small-angle approximation

for 𝑠𝑖𝑛(𝑥). Plotting the first 5 terms of this function over the graph of 𝑠𝑖𝑛(𝑥) allows us to

observe an intriguing pattern:

Color Terms
1

sin(𝑥)

Graph 5: The first five Taylor polynomials for sin(x) centered at 0

Each additional term adds two new roots to the approximation and makes it closer in

shape to the pure sine function. The higher degree approximations also appear to have a

higher range of Y-values in which they are accurate. Seemingly, taking a higher degree

approximation would allow us to gain a more accurate and versatile approximation. Let us

test this by plotting the absolute error function in each of these polynomial approximations

in Graph 6:

8
 Graph 6: The squared error in the Taylor polynomials

Although all of our Taylor approximations have an absolute error of zero for 𝑥 = 0 (as

this is the value on which our series is centred), the polynomials of a higher degree are

observed not only to have larger domains where they are relatively accurate, but also a

lower absolute error for all values of 𝑥. That is to say, including more terms of the Taylor

series in our approximation will give us a value that is closer to the pure sine function.

Bhaskara I’s Approximation

Bhaskara I was a seventh-century mathematician4 who derived one of the earliest yet

most elegant approximations of 𝑠𝑖𝑛(𝑥). It is a rational function composed of a ratio of two

quadratic functions, and it is known to provide accurate approximations of trigonometric

values despite its simple form. Bhaskara I’s formula can be written as follows:

16𝑥(𝜋 − 𝑥)
𝑠𝑖𝑛(𝑥) ≈
5𝜋 ! − 4𝑥(𝜋 − 𝑥)

4
O' Connor, J. J., and E. F. Robertson. “Bhaskara I.” Bhaskara I (about 600 - about 680), mathshistory.st-
andrews.ac.uk/Biographies/Bhaskara_I.html. Accessed 3 Dec. 2019.
9
 Bhaskara I never provided a detailed rationale or proof for his formula, but it is indeed

possible to derive using algebraic methods. This formula can be seen to bear resemblance to

the previously derived quadratic approximation — both functions contain the quadratic

relation 4𝑥(𝜋 − 𝑥) in their numerator, which means that Bhaskara I’s approximation can be

expressed as a scaled expression5 of our original quadratic approximation. This essentially

involves adding a “scale factor” to our original formula that increases its accuracy by using

another set of known values for 𝑠𝑖𝑛(𝑥).

Since the scaling function is a quadratic, we start with:

4𝑥 (𝜋 − 𝑥) 4𝑥 (𝜋 − 𝑥 )
𝜋 ! 𝜋!
=
𝑓 (𝑥 ) 𝑎𝑥 ! + 𝑏𝑥 + 𝑐

As the value of our quadratic approximation was exactly equal to the actual value of
,
𝑠𝑖𝑛(𝑥) at ( , we know that the scale function will be equal to 1 at this point. To find 𝑎, 𝑏 and

𝑐 , we can take two other points on the graph of 𝑠𝑖𝑛(𝑥) that we know to be inexact in our
, *,
approximation: and . By scaling our original quadratic approximation, we can ensure
- -

that the new formula returns the exact values for 𝑠𝑖𝑛(𝑥) at these two points, which means

that it will be more accurate than the previous one and the shape of its graph will coincide

more closely with 𝑠𝑖𝑛(𝑥).

, *, .
Using the information 𝑠𝑖𝑛( ) = 𝑠𝑖𝑛( ) = , we can derive our scaling function. The
- - (
/0(,10) *
function returns a value of for both of these values of 𝑥, which means that our
,! 2
.3 .
scale factor must be for the approximation to equal . This gives us 3 simultaneous
2 (

equations for 𝑓(𝑥):

.3 *, *,
= 𝑎( - )( + 𝑏( - ) + 𝑐
2

.3 , ,
= 𝑎( - )( + 𝑏( - ) + 𝑐
2

5
Berry, Nick. “Approximating the Sine Function.” Data Genetics, 1 July 2019,
datagenetics.com/blog/july12019/index.html. Accessed 3 Dec. 2019.
10
 , ,
1 = 𝑎( )( + 𝑏( ) + 𝑐
( (

Solving for 𝑎, 𝑏 and 𝑐 :

. . *
𝑎 = ,! , 𝑏 = − ,, 𝑐 = /

Plugging the values into the scaling function, we get:

4𝑥(𝜋 − 𝑥)
𝜋!
1 ! 1 5
𝑥 − 𝑥+
𝜋! 𝜋 4
This formula simplifies to:
"#$(&'$) "#$(&'$)
𝑠𝑖𝑛(𝑥) ≈ =
)$! ')&$*+&! +&! ')$(&'$)

Plotting Bhaskara’s approximation over the true sine function gives us two graphs that

are nearly indistinguishable in their shape between 0 and 𝜋 (Graph 7). The error function

has a maximum value of 2.663 × 10&) (Graph 8), which means that it will be accurate up to

nearly 3 decimal places.

Approximation

sin(𝑥)

Graph 7: Bhaskara I's approximation plotted against sin(x)

The limitations of Bhaskara’s approximation are similar to the ones faced by the

simple quadratic method — a parabolic approximation cannot represent a wide range of

values for 𝑠𝑖𝑛(𝑥), it remains limited to a maximum range of 𝜋, after which its inaccuracy

11
 Graph 8: Squared-error in Bhaskara I's approximation

increases at a very large rate. As such, Bhaskara’s approximation would provide us with a

higher level of accuracy than our simple quadratic approximation or a comparable Taylor

polynomial in the limited range of 0 to 𝜋, but it would not be appropriate in other cases.

Conclusion

I began this investigation with a certain curiosity about the nature of transcendental

functions and exploring the various methods in which we can approximate trigonometric

expressions in algebraic terms helped me understand how such functions fundamentally

differ from their simpler non-transcendental counterparts. While we may take the ability to

calculate trigonometric values for granted due to widespread access to digital calculators,

attempting to evaluate them manually using algebraic approximations can often hinder the

accuracy of our calculation. We can improve the accuracy and range of such approximations

by either using higher degree approximations or scaling our approximations closer to the

values of the real function, but there is a constant trade-off — making our approximations

more complicated makes them inherently less useful as it becomes harder to calculate them

manually.

12
 The infinitely repetitive nature of the trigonometric 𝑠𝑖𝑛(𝑥) was one that turned out to

be impossible to replicate in terms of finite algebraic terms, but the Taylor series provides

us with a versatile solution that can be expanded to any required range. The fact that

Bhaskara I’s method could evaluate a trigonometric function to a high level of accuracy

centuries before the invention of the first calculators showcases the effectiveness of

analogue mathematical methods despite the algebraically “inexpressible” nature of

transcendental functions. Ultimately, these approximations bring to light the advantages and

limitations of algebraic functions — their ability to both be directly calculated and to

approximate the shapes of other relations and patterns is a powerful mathematical tool that

allows us to better understand and utilise even the most elusive mathematical concepts, but

they are inhibited by their finite nature and the need to be explicitly defined.

These limitations suggest further scope for exploration in developing trigonometric

approximations — we could use computer programming or equations in terms of mod

functions to generalise such approximations to a much larger range of values for 𝑥. We can

also derive approximations for other trigonometric functions such as cos(𝑥 ) and tan (𝑥)

using trigonometric identities. Other lines of exploration may include comparing these

elementary methods to advanced calculation algorithms such as CORDIC6, which is

commonly used in graphing calculators to compute trigonometric functions.

6
Sultan, Alan. “CORDIC: How Hand Calculators Calculate.” The College Mathematics Journal, vol. 40, no. 2, 2017,
pp. 87-92, doi:10.1080/07468342.2009.11922342. Accessed 12 Feb. 2020.

13
 Works Cited

Electronic Sources

Berry, Nick. “Approximating the Sine Function.” Data Genetics, 1 July 2019,

datagenetics.com/blog/july12019/index.html. Accessed 3 Dec. 2019.

O' Connor, J. J., and E. F. Robertson. “Bhaskara I.” Bhaskara I (about 600 - about 680),

mathshistory.st-andrews.ac.uk/Biographies/Bhaskara_I.html. Accessed 3 Dec. 2019.

Sultan, Alan. “CORDIC: How Hand Calculators Calculate.” The College Mathematics Journal, vol.

40, no. 2, 2017, pp. 87-92, doi:10.1080/07468342.2009.11922342. Accessed 12 Feb. 2020.

“Taylor Series Approximation.” Brilliant Math & Science Wiki, brilliant.org/wiki/taylor-series-

approximation/. Accessed 12 Feb. 2020.

“Transcendental Function.” Encyclopædia Britannica, Encyclopædia Britannica, Inc., 15 Apr. 2011,

www.britannica.com/science/transcendental-function. Accessed 27 Jan. 2020.

14