A Common Book of

3.1415926535897932384626433832795028841971693993751058209749445923078164062862 089986280348253421170679...
The mysterious and wonderful -- Philip J. Davis In recent years, the computation of the expansion of -- David H. Bailey It requires a mere 39 digits of in order to compute the circumference of a circle of radius (an upper bound on the distance travelled by a particle moving at the speed of light for 20 billion years, and as such an upper bound for the radius of the universe) with an error of less than meters (a lower bound for the radius of a hydrogen atom). -- Jonathan and Peter Borwein has assumed the role of a standard test of computer integrity. is reduced to a gargle that helps computing machines clear their throats.

The number has been the subject of a great deal of mathematical (and popular) folklore. It's been worshipped, maligned, and misunderstood. Overestimated, underestimated, and legislated. Of interest to scholars, crackpots, and everyday people. Pretty amazing accomplishments for a number! The next few pages will attempt to teach you a few facts about . You will find here, among other things, a brief history of extended precison approximations of , including Archimedes' method for estimating , a page full of "oh, wow!" formulas used to estimate over the centuries, and a brief look at a modern algorithm used to compute . I also have a list of references for further reading and a list of other pages devoted to pi on the Web. Before we begin, it might not hurt to remind you that is defined as the (constant) ratio of the circumference to the diameter in any circle. In other words, the circumference and diameter of every circle are known to be related by . It's not hard to see (using only elementary geometry) that is bigger than 3 but less than 4.

The Precomputer History of

That the ratio of circumference to diameter is the same (and roughly equal to 3) for all circles has been accepted as "fact" for centuries; at least 4000 years, as far as I can determine. (But knowing why this is true, as well as knowing the exact value of this ratio, is another story.) The "easy" history of concerns the ongoing story of our attempts to improve upon our estimates of . This page offers a brief survey of a few of the more famous early approximations to . The value of given in the Rhynd Papyrus (c. 2000 BC) is

Various Babylonian and Egyptian writings suggest that each of the values

were used (in different circumstances, of course). The Bible (c. 950 BC, 1 Kings 7:23) and the Talmud both (implicitly) give the value simply as 3. Archimedes of Syracuse (240 BC), using a 96-sided polygon and his method of exhaustion, showed that

and so his error was no more than

The important feature of Archimedes' accomplishment is not that he was able to give such an accurate estimate, but rather that his methods could be used to obtain any number of digits of . In fact, Archimedes' method of exhaustion would prove to be the basis for nearly all such calculations for over 1800 years. Over 700 years later, Tsu Chung-Chih (c. 480) improved upon Archimedes' estimate by giving the familiar value

which agrees with the actual value of to 6 places. Many years later, Ludolph van Ceulen (c. 1610) gave an estimate that was accurate to 34 decimal places using Archimedes' method (based on a -sided polygon). The digits were later used to adorn his tombstone.

The next era in the history of the extended calculation of was ushered in by James Gregory (c. 1671), who provided us with the series

Using Gregory's series in conjunction with the identity

John Machin (c. 1706) calculated 100 decimal digits of . Methods similar to Machin's would remain in vogue for over 200 years. William Shanks (c. 1807) churned out the first 707 digits of . This feat took Shanks over 15 years -- in other words, he averaged only about one decimal digit per week! Sadly, only 527 of Shanks' digits were correct. In fact, Shanks published his calculations 3 times, each time correcting errors in the previously published digits, and each time new errors crept in. As it happened, his first set of values proved to be the most accurate. In 1844, Johann Dase (a.k.a., Zacharias Dahse), a calculating prodigy (or "idiot savant") hired by the Hamburg Academy of Sciences on Gauss's recommendation, computed to 200 decimal places in less than two months. In the era of the desktop calculator (and the early calculators truly required an entire desktop!), 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. Today, of course, in the era of the supercomputer, hundreds of millions of digits are known. The evolution of the machine-assisted approximations to is summarized on a table on the next page.

The Postcomputer History of

To further highlight the improvements in our abilities to compute in recent years, consider this: The 1961 computation of 100,000 decimal digits of required roughly 105,000 full-precision operations, while a modern algorithm, devised by Jonathan and Peter Borwein in 1984, takes only 112 full-precision operations to achieve the same accuracy. A mere 8 iterations of their algorithm (roughly 56 operations) will produce 694 digits of (thus reducing Wm. Shanks' 15 year calculation to a matter of seconds).

Archimedes' Method of Exhaustion

In Measurement of the Circle, the great Archimedes (c. 287--212 BC) found an approximation for the circumference of a circle of a given radius. Since we know that the circumference and diameter of any circle are related by the formula , this means that if we start with a circle of diameter 1, then Archimedes' approximation for actually provides an approximation for . Archimedes' idea was to approximate the circle using both inscribed and circumscribed (regular) polygons. Below are pictured inscribed and circumscribed octagons.

More generally, we would consider inscribed and circumscribed -gons. The inscribed -gon has sides, each of the same length , and the circumscribed -gon has sides, each of the same length . (In truth, we should consider simplicity, forego this extra generality.) -gons, where M is a positive integer. But, for

The perimeter of the inscribed -gon, which we denote by , and the perimeter of the circumscribed -gon, which we denote by , are approximations for and so, in this case, are also approximations for :

By means of geometric (and what we would now call trigonometric) arguments, Archimedes was able to derive iterative formulas for and , which are reminiscent of the Babylonian algorithm for computing square roots.

Archimedes' Method, Part II

Recall that we want to estimate the circumference of a circle of diameter 1 (which we know to be ). For each = 2, 3, 4, ..., we inscribe and circumscribe regular polygons having sides.

Recall, too, that the perimeters

and

satisfy

In order to generate an iterative formula for the perimeters, we use a bit of geometry:

If we denote the central angle in our -gon by 2 , then right triangle with hypoteneuse 1. Hence, = sin . Next we rotate our picture and concentrate on .

is the side opposite the angle in a

This time our right triangle has Hence, = tan .

as the side opposite the angle , and

as the adjacent side.

Archimedes' iterative formulas for

and

will now follow from two trig identities.

Archimedes' Method, Part III

Recall our construction:

We have seen that = sin and sin( ) and = tan( ).

= tan . Since we have used

-gons, it follows that

In order to relate

and

to

and

, we use two trig identities:

and

In terms of the perimeters of our polygons, this means:

and

or

Notice that is the harmonic mean of . (Sound familiar?) Finally, let's look at a simple example.

and

, while

is the geometric mean of

and

An Example
Let's try Archimedes' algorithm

starting with inscribed and circumscribed squares

Here are the results of 12 iterations of the algorithm.

2 3 4 5 6 7 8 9 10 11 12 13 14

2.8284271247461901 3.0614674589207182 3.1214451522580523 3.1365484905459393 3.1403311569547529 3.1412772509327729 3.1415138011443011 3.1415729403670914 3.1415877252771597 3.1415914215112 3.1415923455701177 3.1415925765848727 3.141592634338563

4 3.3137084989847604 3.1825978780745281 3.1517249074292561 3.1441183852459043 3.1422236299424568 3.1417503691689665 3.1416320807031818 3.1416025102568089 3.1415951177495891 3.1415932696293073 3.1415928075996446 3.1415926920922544

These computations were done using a spreadsheet with only limited accuracy (ostensibly, 15 decimal places). Nevertheless, notice that either of the last two entries agree with the actual value of to at least 6 places. Another half dozen iterations would yield to 9 places. Not bad! A better starting estimate would obviously have helped our situation here. Archimedes started with regular hexagons (an inscribed perimeter of 3 and a circumscribed perimeter of ). Archimedes' algorithm falls is one of a larger class of related algorithms.

Related Algorithms
In the many years since Archimedes' discovery, algorithms of a similar type have come up repeatedly. Perhaps no one understood them so well as Carl Friedrich Gauss (1777--1855); Gauss considered the following iteration scheme: Start with and . For , define

[This same algorithm is sometimes called Borchardt's algorithm. It is equivalent to Archimedes' algorithm; just substitute and .] Gauss was considering a difficult problem when he encountered this sequence. He asked his teacher, Pfaff, about the sequence; Pfaff showed that for any positive and , the common limit of the two sequences is

Favorite Formulas for

= 3.14159265358979323846264338327950288419716939937510582097494459230781640628620899 86280348253421170679...

"Name" Formulas
(Wallis, 1655)

(Brouncker, 1658)

(Newton, 1665)

(Gregory, 1671) (Leibniz, 1674)

(Euler, 1748)

Integral Formulas

Series

Odds and Ends

And everyone's favorite:

Geometry Review
We have used several simple facts here: 1. A triangle inscribed in a semicircle, as shown below, is a right triangle.

2. Given triangle inscribed in semicircle is twice the angle .

, as shown below, the central angle

Geometry Review, Part I

1. A triangle inscribed in a semicircle is a right triangle. In the picture below we want to show that angle is a right angle. One way to see this is to take advantage of Cartesian coordinates: Here we've identified our semicircle with the top half of the graph of .

To check that is a right angle, we will show that the Pythagorean theorem is satisfied for this triangle (with the diameter of our circle as the hypoteneuse of the right triangle). In terms of our coordinates:

Geometry Review, Part II

2. Given triangle DAC inscribed in semicircle is twice the angle . , as shown below, the central angle

To prove this, we use the fact that triangle sum to .

is a right angle and the fact that the angles in any

Summing the angles in triangle BDC, we get

, or

Since triangle ADB is isosceles (two of its legs are radii of the circle), the missing angle must be equal to .