Saturday, June 4, 2016

a computer science paper

Summary of “Divide and Conquer”

Why was Thomas Nicely so obsessed with finding twin primes? It was the early 90’s and public access to computers with powerful calculation capabilities was finally becoming a reality. Nicely, a professor of mathematics at Lynchburg College, was weighing the feasibility of several different “unfinished” mathematical queries. He was looking for a problem much too difficult to solve with pen and paper or even the most powerful calculator—yet would not depend on a difficult to access supercomputer either. With innovations in desktop computing, such as Intel’s Pentium chip, Nicely set out to elaborate on a mathematical problem by harnessing, as well as challenging, the computer processing capabilities of the era. His research was “to serve as a model for other people working in small college environments.” Of the many mathematical theories that needed further exploration, the question of whether the occurrence of twin primes continued infinitely as you counted up and at what exact number did the sum of the reciprocals of the twin primes converge (known as Brun’s sum) was the tantalizing question Nicely wanted to answer. In 1993, Nicely decided to solve Brun’s sum to such a precision that would have been impossible even half a decade before.

Twin primes are consecutive pairs of odd primes: 3 and 5, 5 and 7, or 11 and 13 are the first few of an infinite set of examples. Brun’s sum was an estimation of the occurrence of twin primes. But, because humans are only capable of counting and keeping track of a finite set of numbers and the occurrence of twin primes can be sporadic, Brun’s sum was not as precise as it could be. Why is this important, you may ask? Well, let’s start this mathematical conundrum at the beginning—300 BC. The “father of Geometry”, Euclid of Alexandria, proves that there are,
in fact, an infinite number of primes! Fast forward two millennia to Europe in 1896 --Charles-Jean Étienne Gustave Nicolas de la Vallée Poussin of Belgium and Jacques Hadamard of France independently proved the Prime Number Theorem. This theorem describes how prime numbers are distributed among the positive integers. It posits that primes become less common the higher you count. The Prime Number Theorem answered many of the unsolved mysteries of prime numbers…but the solution to the enigma of twin primes remained out of the grasp for mathematicians of the era. I

n Norway, 1919, twenty years later, Viggo Brun postulated an upper bound for twin primes; yet no lower bounds were found to prove twin primes occurred in an infinite sequence. Brun also developed the titular Brun’s sum. The Brun sum states that if you reciprocate all the twin primes and add them together – the sum is finite. Furthermore, the infinite series converges to some number “B.” In 1974, Shanks and Wrench Jr. -- naval mathematicians working in Bethesda, Maryland, tested Brun’s sum by investigating all the twin primes in the set of the first two million prime numbers. A year later, Australian mathematician Richard Brent tracked down all the twin primes of the first one-hundred billion prime numbers! Brent found an astonishing 224,376,048 twin primes and obtained a numerical estimate of the convergence of Brun’s sum: 1.90216054. As amazing as that was, Nicely knew he could take the next leap in calculating an even more precise Brun’s sum—he was going to calculated the set of prime numbers up into the trillions!

However, Nicely’s quest was not without difficulties. Because of the fallibility of new technology and the dilution of accountability of software development within massive corporations, imperfect products were sometimes released. Enter Intel. Intel released a faster and more powerful Pentium chip in 1993. In less than 2 years, nearly one million Pentiums had been sold at a retail price of $1,000 each. About 18 months after its release, a previously unknown flaw in the processor’s floating point unit was finally discovered by the company.

However, Intel swept this discovery under the rug. They figured no user would be using the processor for the kind of high-stakes, ultra-precise computation that would generate a meaningful error. They also did not want to give out replacements and refunds for the near million units that were already out on the market!

When Thomas Nicely discovered that Intel had not issued a recall, he decided: No more Mr. Nicely. Nicely confronted Intel directly- when he was ignored he took his shocking discovery to the Internet. Word of mouth spread and public opinion of Intel fell. The consumers demanded their money back. Intel tried to argue that the error only interfered in a small minority of very specific problems. Yet, the consumers insisted and Intel caved to public demand. They issued a refund of the faulty Pentium chip at the consumer’s request—and fortunately, the blow to Intel’s reputation was short lived. In fact, it seemed to have the opposite effect. The chip was a hot seller that Christmas and by mid-1995—tens of millions units had been sold.

Discussion

The article “Divide and Conquer” is an important read for Computer Science students because it is both a history piece capturing the growing pains of an industry in the early 90’s as well as a cautionary tale alerting the public to the surprising fallibility of computers. The quote: “In a sense, all computers do division wrong,” is a shocking realization. I, myself, am so awed by the beyond human processing capabilities of computers that I inherently bestow upon them supernatural powers. But then to see that long division is a relatively complex process for
computers and that they can only process a finite amount of decimal points—it brings the truth of computers out to light.

The descriptions of how Nicely went about solving Brun’s sum, despite setbacks, could be very eye opening for computer science students. The amount of trouble Nicely went through to ensure that safeguards were in effect and no unexpected error from the hardware would result in a large loss of data was very smart. Carpenters have a saying, “Measure twice, cut once.” Meaning, that if you do more pre-planning work than you even think is necessary, it can really save you time and headache further down the road.

The safeguards Nicely enacted paid off in spades. For example, he would have two different computers with different processing equipment (either a ‘486’ or a Pentium chip) compute the same set of data—any discrepancies between the two was a red flag that either one, the other, or BOTH computers were making processing errors! Another example, Nicely would store the results outputted by each computer at the end of each billion numbers. That way, if two different machines settled on a different end value, Nicely could look back through the data to find the range where the two number sets started to differ. Another security measure Nicely took to ensure accuracy, is he would check his output against previously published result. Furthermore, Nicely decided to compute Brun’s sum using two different methods. One was to use the computer’s floating point unit to calculate each reciprocal to nineteen digits and the other was a ultra-precision algorithm that measured up to twenty-six (and later fifty-three) digits accuracy! Also, like any good scientist, Nicely had a healthy skepticism…he did not just take it for granted that machines will behave the way you want
them too. And that scientific-method mentality can be an inspiration for anyone breaking new ground for any perplexing un-solved problem.

References
Cipra, B. (1995). Divide and Conquer. What’s Happening in Mathematical Sciences
Volume 3.

No comments:

Post a Comment