The Calculation

of 
Paul Butterfoss
Zachary Tseng
Math 251 – Extra Credit Paper

3.1415926535897932384626433832795028841971693993751058209749445923078164062862
089986280348253421170679821480865132823066470938446095505822317253594081284811
174502841027019385211055596446229489549303819644288109756659334461284756482337
867831652712019091456485669234603486104543266482133936072602491412737245870066
063155881748815209209628292540917153643678925903600113305305488204665213841469
519415116094330572703657595919530921861173819326117931051185480744623799627495
673518857527248912279381830119491298336733624406566430860213949463952247371907
021798609437027705392171762931767523846748184676694051320005681271452635608277
857713427577896091736371787214684409012249534301465495853710507922796892589235
420199561121290219608640344181598136297747713099605187072113499999983729780499
510597317328160963185950244594553469083026425223082533446850352619311881710100
031378387528865875332083814206171776691473035982534904287554687311595628638823
537875937519577818577805321712268066130019278766111959092164201989380952572010
654858632788659361533818279682303019520353018529689957736225994138912497217752
834791315155748572424541506959508295331168617278558890750983817546374649393192
550604009277016711390098488240128583616035637076601047101819429555961989467678
374494482553797747268471040475346462080466842590694912933136770289891521047521
620569660240580381501935112533824300355876402474964732639141992726042699227967
823547816360093417216412199245863150302861829745557067498385054945885869269956
909272107975093029553211653449872027559602364806654991198818347977535663698074
265425278625518184175746728909777727938000816470600161452491921732172147723501
414419735685481613611573525521334757418494684385233239073941433345477624168625
189835694855620992192221842725502542568876717904946016534668049886272327917860
857843838279679766814541009538837863609506800642251252051173929848960841284886
269456042419652850222106611863067442786220391949450471237137869609563643719172
874677646575739624138908658326459958133904780275900994657640789512694683983525
957098258226205224894077267194782684826014769909026401363944374553050682034962
524517493996514314298091906592509372216964615157098583874105978859597729754989
301617539284681382686838689427741559918559252459539594310499725246808459872736
446958486538367362226260991246080512438843904512441365497627807977156914359977
001296160894416948685558484063534220722258284886481584560285060168427394522674
6767889525213852254995466672782398645659611635
Paul Butterfoss
Zachary Tseng
Math 251

Pi () is probably the most well known mathematical symbol. It deals with everything
from trigonometric functions to three-dimensional spheres. One of its most interesting
characteristics is that it is a non-repeating decimal, a totally random series of numbers with no
apparent pattern. Then how is pi calculated? That is what I plan to discus in the following
pages.

One method for calculating pi involves the MacLaurin series of the inverse tangent
function (1).

x 2 n 1 x 3 x 5 x 7 x 9 x11
tan (x) =  ( 1) n
-1
= x     …… (1)
n 0 2n  1 3 5 7 9 11

The MacLaurin series for a function (3) is found by taking the Taylor series (2) and
evaluating for the special case a=0.

f ( n ) (a)
f (x) =  ( x  a) n
n0 n !
f ' (a) f '' ( a ) f ''' ( a )
= f (a) + ( x  a) + ( x  a) 2 + ( x  a ) 3 ……. (2)
1! 2! 3!

 (n)
f (0) f ' (0) f '' (0)
f (x) = 
n 0 n!
( x) n = f (0) +
1!
( x) +
2!
( x ) 2 ……. (3)

When x=1 the MacLaurin series for tan-1(x) is equal to the series expansion for /4. This
is called the Leibniz series (4). (This formula is attributed to Leibniz, although it now seems that
James Gregory (1638-1675) first discovered it).

12 n 1 1 1 1 1 1
tan-1(1) =  (1) n
n 0 2n  1
= 1      ……..
3 5 7 9 11
(4)

To compute the number of terms of the Leibniz series needed to reach /4 to a certain
accuracy, we must use the Alternating Series Estimation Theorem (5).

If s = (-1)n-1bn is the sum of an alternating series that satisfies

(a) 0  bn+1  bn and (b) lim b

n 
n 0

then |Rn| = |s – sn|  bn+1 (5)

 In order to reach 6-digit accuracy, bn+1 must be  0.000001. Therefore, the greatest
possible value for bn+1 is 1/1000001 (6).

12 n 1 1
bn+1 = (1) n  (6)
2n  1 1000001

Therefore n must be  500000. Since n=0 is the first term, n=500000 is the 5000001
1
term. But does not affect the accuracy so there must be exactly 500000 terms in order
1000001
to reach 6-digit accuracy.

Using this series, it takes an enormous amount of terms to simply get 6-digit accuracy.
Therefore we should try to find another series that converges to /4 much faster. One possible
way is through the formula below (7).

4tan-1(1/5) – tan-1(1/239) = /4 (7)

To prove (7), we could use this formula (8) but the coefficient of 4 in front of the inverse
tangent function causes a problem.

x y
( )
-1 -1
tan (x) - tan (y) = tan 1  xy
-1
(8)

Therefore we must incorporate the double angle formula for tangent (9).

2 tan x
tan 2x = (9)
1  tan 2 x

Substituting u = tan x (x = tan-1 u) into this formula we get the following:

2u 2u
tan 1 ( ) -1 tan 1 ( )
x= 1 u2 or tan u = 1 u2 (10)
2 2

1 2u
2 tan-1 u = tan ( ) (11)
1 u2

This gives us a coefficient of only 2 however so we need to multiply both sides by 2.

1 2u
4 tan-1 u = 2 tan ( ) (12)
1 u2
Notice however that the right side of equation 12 is the same as the left side of equation
11 with a different value. Therefore we can substitute (2u/1-u2) for u in equation 11.
 Then we get an equation for changing an inverse tangent function with a coefficient of 1
into an inverse tangent function with a coefficient of 4.

2u
2( )
-1 tan 1
( 1 u2
4 tan u = ) (13)
2u 2
1 ( )
1 u2

Using this equation and plugging in 1/5 for u from the equation for /4 (7), we can get a
value for 4tan-1u without the coefficient of 4. This will allow us to use equation 8 to prove
equation 7.

2(0.2)
2( )
1 1  (0.2) 2
4 tan (1/5) = tan (
-1
= tan-1 (120/119) (14)
2(0.2) 2
1 ( )
1  (0.2) 2

Then we can substitute this value into equation 7 to get a new equation that is in the form
of equation 8.

tan-1(120/119) - tan-1(1/239) = /4 (15)

Then using equation 8 and the above values for x and y we can prove that this is true and
that it does equal /4.

120 1

( 119 239 )
120 1
1 *
tan -1 119 239 = tan-1(1) = /4 (16)

Once we have proven that equation 7 is true, a new series expansion for /4 can be found
by using equation 7 and equation 4.


(1/5) 2n 1 
(1/239) 2n 1
4 -  = /4
n 0 2n  1 n 0 2n  1

(17)

Again to determine accuracy to a certain number of digits, in this case 6, we must utilize
equation 5. Like in equation 6, bn+1 must be  1/1000001 to achieve 6-digit accuracy. But unlike
equation 4, using this new series expansion we reach 6-digit accuracy much faster. It takes only
4 terms before we have 6-digit accuracy.
 
(1/5) 2(4) 1 
(1/239) 2(4) 1
4 -   1/1000001
n 0 2( 4)  1 n 0 2( 4)  1
(18)

Since this series expansion for tan-1x allows us to reach higher accuracy much faster, it is
the more efficient of the two series. It is therefore a better series expansion for calculating .

Example: Calculation of Pi to 707-digit accuracy (like William Shanks):

To determine this series to 707-digit accuracy we basically follow the same procedure as
above. We find that we need only 504 terms before we can achieve 707-digit accuracy.


(1/5) 2(504) 1 
(1/239) 2(504)1
4 -   1x10-707
n 0 2(504)  1 n 0 2(504)  1
(19)

The History of the Calculation of Pi

Anyone who knows the simple formula for the area of a circle (Area = pi times the square

of the radius) can compute pi from that relationship (Alfeld 1). However, calculating it

accurately using this method requires much time and patience. Still, this approach to computing

was used for over 3500 years until the late 17th century when the much more efficient calculus-

based series expansions became available.

About two thousand years B.C., the first calculations of pi were made, probably by the

Egyptians and Babylonians. A document entitled the Rhind Papyrus, dated 1650 BC, shows that

the Egyptians obtained the value of pi to be (4/3)^4; the Babylonians earlier approximated a

value of 3 1/8 for pi. At about the same time, it is believed that the Indians were using the square

root of 10. These early approximations of pi were not very accurate as they were correct to only

1 decimal place (A Brief History 1).

 The next major step towards a more accurate value of pi was taken by the

Greek mathematician Archimedes who decided to take up the problem in about

250 BC. He observed that area of a unit circle (a circle with radius 1) equals the

exact value of pi. By first finding the area of a square inscribed in the circle, then

a pentagon, and so on up through 96-sided polygon, Archimedes continuously

found better and better approximations for pi. With each polygon having more

sides (close to becoming a circle), he came closer and closer to finding an exact value of pi. He

finally discovered that 3 10/71 < pi < 3 10/70. Many people often use the latter value, 22/7,

when dealing with pi (when it does not require precise calculations). In fact, it is used so often

that some people actually think that it is the exact value of pi (A Brief History 1).

Following in Archimedes’ footsteps in 150 AD was Ptolemy of Alexandria. He

estimated the value of pi to be 377/120. The Chinese were a part of this pi race also. In 500 AD

Tsi Ch’ung-Chi gave the value of pi at 355/113. These estimations are correct by 3 and 6

decimal places respectively (A Brief History 1).

Other contributors include:

 Al'Khwarizmi (c. 800 ) 3.1416
 Al'Kashi (c. 1430) 14 places
 Viète (1540-1603) 9 places
 Roomen (1561-1615) 17 places
 Ludolph Van Ceulen (c. 1600) 34 places
(used Archimedes’s method and a -sided polygon)
 After Van Ceulen, we encounter the method that we used to solve the problem in the

previous pages. Wrongly entitled the Leibniz series, it was actually created by James Gregory.

Using the two formulas above (like we did today), John Machin (c. 1706) calculated 100

decimal digits of . (Carothers 1)

Another interesting mathematical formula for pi was discovered during the European

Renaissance by Wallis (1616-1703).

Here the two products continue indefinitely and they finally converge to

the value of pi over 2 (A Brief History 1).

William Shanks (c. 1807) determined the first 707 digits of . This amazing feat took him

over 15 years! Sadly however, only 527 of Shanks' digits were correct. Even after publishing

them 3 times, he had mistakes. Each time he corrected errors in the previously published digits,

new ones crept in. Ironically, his first set of calculations proved to be the most accurate

(Carothers 1).

After the invention of desktop calculators, D. F. Ferguson (c. 1947) raised the total to 808

(accurate) decimal digits. In fact, it was Ferguson who discovered the errors in Shanks'

calculations (Carothers 1).

 The current record for most pi digits is held by Yasumasa Kanada and Daisuke Takahashi

from the University of Tokyo with 51 billion digits of pi (51,539,600,000 decimal digits to be

precise) (Alfeld 1). Now that’s a lot of pi. 

(The number on the cover is pi to 2,462 decimal places)

 Works Cited

Alfeld, Peter. “Archimedes and the Computation of Pi.” 01 April 1999. University of Utah

05 December 2000 <http://www.math.utah.edu/~alfeld/Archimedes/Archimedes.html>.

“A Brief History of Pi.” The Pi Project. Univ. of Texas. 05 December 2000

<http://www.ma.utexas.edu/users/tyilk/PiProj/PiHistory.htm>.

“Calculating Pi.” The Pi Project. Univ. of Texas. 05 December 2000

<http://www.ma.utexas.edu/users/tyilk/PiProj/PiCalc.htm>.

Carothers, Neal. “The Precomputer History of .” 05 December 2000

<http://ernie.bgsu.edu/~carother/pi/Pi2.html>.