Prime Numbers
Prime Numbers Prime Numbers
4 Chapter 1 PRIMES! of the most recent “largest known primes” performed thus: The primality proof/disproof for a single candidate 2 q − 1 required in 2004 about one CPUweek, on a typical modern PC (see continually updating website [Woltman 2000]). By contrast, a number of order 2 20000000 would have required, just a decade earlier, perhaps a decade of a typical PC’s CPU time! Of course, both machine and algorithm advances are responsible for this performance offset. To convey again an idea of scale: At the start of the 21st century, a typical workstation equipped with the right software can multiply together two numbers, each with a million decimal digits, in a fraction of a second. As explained at the end of Section 9.5.2, appropriate cluster hardware can now multiply two numbers each of a billion digits in roughly one minute. The special Mersenne form 2 q − 1 of such numbers renders primality proofs feasible. For Mersenne numbers we have the very speedy Lucas– Lehmer test, discussed in Chapter 4. What about primes of no special form— shall we say “random” primes? Primality proofs can be effected these days for such primes having a few thousand digits. Much of the implementation work has been pioneered by F. Morain, who applied ideas of A. Atkin and others to develop an efficient elliptic curve primality proving (ECPP) method, along with a newer “fastECPP” method, discussed in Chapter 7. A typically impressive ECPP result at the turn of the century was the proof that (2 7331 − 1)/458072843161, possessed of 2196 decimal digits, is prime (by Mayer and Morain; see [Morain 1998]). A sensational announcement in July 2004 by Franke, Kleinjung, Morain, and Wirth is that, thanks to fastECPP, the Leyland number 4405 2638 + 2638 4405 , having 15071 decimal digits, is now proven prime. Alongside these modern factoring achievements and prime-number analyses there stand a great many record-breaking attempts geared to yet more specialized cases. From time to time we see new largest twin primes (pairs of primes p, p+2), an especially long arithmetic progression {p, p+d,...,p+kd} of primes, or spectacular cases of primes falling in other particular patterns. There are searches for primes we expect some day to find but have not yet found (such as new instances of the so-called Wieferich, Wilson, or Wall–Sun– Sun primes). In various sections of this book we refer to a few of these many endeavors, especially when the computational issues at hand lie within the scope of the book. Details and special cases aside, the reader should be aware that there is a widespread “culture” of computational research. For a readable and entertaining account of prime number and factoring “records,” see, for example, [Ribenboim 1996] as well as the popular and thorough newsletter of S. Wagstaff, Jr., on state-of-the-art factorizations of Cunningham numbers (numbers of the form b n ± 1forb ≤ 12). A summary of this newsletter is kept at the website [Wagstaff 2004]. Some new factorization records as of this (early 2005) writing are the following:
1.1 Problems and progress 5 The Pollard-(p − 1) method (see our Section 5.4) was used in 2003 by P. Zimmermann to find 57-digit factors of two separate numbers, namely 6 396 + 1 and 11 260 +1. There are recent successes for the elliptic-curve method (ECM) (see our Section 7.4.1), namely, a 57-digit factor of 2 997 − 1 (see [Wagstaff 2004]), a 58-digit factor of 8 · 10 141 − 1 found in 2003 by R. Backstrom, and a 59digit factor of 10 233 − 1 found in 2005 by B. Dodson. (It is surprising, and historically rare over the last decade, that the (p − 1) method be anywhere near the ECM in the size of record factors.) In late 2001, the quadratic sieve (QS) (see our Section 6.1), actually a threelarge-prime variant, factored a 135-digit composite piece of 2 803 − 2 402 +1. This seems to have been in some sense a “last gasp” for QS, being as the more modern NFS and SNFS have dominated for numbers of this size. The general-purpose number field sieve (GNFS) has, as we mentioned earlier, factored the 174-digit number RSA-576. For numbers of special form, the special number field sieve (SNFS) (see our Section 6.2.7) has factored numbers beyond 200 digits, the record currently being the 248-digit number 2 821 +2 411 +1. Details in regard to some such record factorizations can be found in the aforementioned Wagstaff newsletter. Elsewhere in the present book, for example after Algorithm 7.4.4 and at other similar junctures, one finds older records from our 1st edition; we have left these intact because of their historical importance. After all, one wants not only to see progress, but also track it. Here at the dawn of the 21st century, vast distributed computations are not uncommon. A good lay reference is [Peterson 2000]. Another lay treatment about large-number achievements is [Crandall 1997a]. In the latter exposition appears an estimate that answers roughly the question, “How many computing operations have been performed by all machines across all of world history?” One is speaking of fundamental operations such as logical “and” as well as “add,” “multiply,” and so on. The answer is relevant for various issues raised in the present book, and could be called the “mole rule.” To put it roughly, right around the turn of the century (2000 ad), about one mole—that is, the Avogadro number 6 · 10 23 of chemistry, call it 10 24 —is the total operation count for all machines for all of history. In spite of the usual mystery and awe that surrounds the notion of industrial and government supercomputing, it is the huge collection of personal computers that allows this 10 24 ,this mole. The relevance is that a task such as trial dividing an integer N ≈ 10 50 directly for prime factors is hopeless in the sense that one would essentially have to replicate the machine effort of all time. To convey an idea of scale, a typical instance of the deepest factoring or primality-proving runs of the modern era involves perhaps 10 16 to 10 18 machine operations. Similarly, a fulllength graphically rendered synthetic movie of today—for example, the 2003 Pixar/Disney movie Finding Nemo—involves operation counts in the 10 18 range. It is amusing that for this kind of Herculean machine effort one may either obtain a single answer (a factor, maybe even a single “prime/composite”
- Page 2 and 3: Prime Numbers
- Page 4 and 5: Richard Crandall Center for Advance
- Page 6 and 7: Preface In this volume we have ende
- Page 8 and 9: Preface ix Examples of computationa
- Page 10 and 11: Contents Preface vii 1 PRIMES! 1 1.
- Page 12 and 13: CONTENTS xiii 4.2.2 An improved n +
- Page 14 and 15: CONTENTS xv 8.5.2 The Shor quantum
- Page 16 and 17: 2 Chapter 1 PRIMES! where exponents
- Page 20 and 21: 6 Chapter 1 PRIMES! decision bit) o
- Page 22 and 23: 8 Chapter 1 PRIMES! 1.1.4 Asymptoti
- Page 24 and 25: 10 Chapter 1 PRIMES! naive subrouti
- Page 26 and 27: 12 Chapter 1 PRIMES! x π(x) 10 2 1
- Page 28 and 29: 14 Chapter 1 PRIMES! Erdős-Turán
- Page 30 and 31: 16 Chapter 1 PRIMES! Let’s hear i
- Page 32 and 33: 18 Chapter 1 PRIMES! Conjecture 1.2
- Page 34 and 35: 20 Chapter 1 PRIMES! It is not hard
- Page 36 and 37: 22 Chapter 1 PRIMES! 1.3 Primes of
- Page 38 and 39: 24 Chapter 1 PRIMES! 9.5.18 and Alg
- Page 40 and 41: 26 Chapter 1 PRIMES! Mersenne conje
- Page 42 and 43: 28 Chapter 1 PRIMES! Again, as with
- Page 44 and 45: 30 Chapter 1 PRIMES! then taken aga
- Page 46 and 47: 32 Chapter 1 PRIMES! McIntosh has p
- Page 48 and 49: 34 Chapter 1 PRIMES! have identitie
- Page 50 and 51: 36 Chapter 1 PRIMES! This theorem i
- Page 52 and 53: 38 Chapter 1 PRIMES! conjectured. T
- Page 54 and 55: 40 Chapter 1 PRIMES! Definition 1.4
- Page 56 and 57: 42 Chapter 1 PRIMES! on the left co
- Page 58 and 59: 44 Chapter 1 PRIMES! sums. These su
- Page 60 and 61: 46 Chapter 1 PRIMES! Vinogradov’s
- Page 62 and 63: 48 Chapter 1 PRIMES! been beyond re
- Page 64 and 65: 50 Chapter 1 PRIMES! prime? What is
- Page 66 and 67: 52 Chapter 1 PRIMES! This kind of c
4 Chapter 1 PRIMES!<br />
of the most recent “largest known primes” performed thus: The primality<br />
proof/disproof for a single candidate 2 q − 1 required in 2004 about one CPUweek,<br />
on a typical modern PC (see continually updating website [Woltman<br />
2000]). By contrast, a number of order 2 20000000 would have required, just<br />
a decade earlier, perhaps a decade of a typical PC’s CPU time! Of course,<br />
both machine and algorithm advances are responsible for this performance<br />
offset. To convey again an idea of scale: At the start of the 21st century, a<br />
typical workstation equipped with the right software can multiply together<br />
two numbers, each with a million decimal digits, in a fraction of a second. As<br />
explained at the end of Section 9.5.2, appropriate cluster hardware can now<br />
multiply two numbers each of a billion digits in roughly one minute.<br />
The special Mersenne form 2 q − 1 of such numbers renders primality<br />
proofs feasible. For Mersenne numbers we have the very speedy Lucas–<br />
Lehmer test, discussed in Chapter 4. What about primes of no special form—<br />
shall we say “random” primes? Primality proofs can be effected these days<br />
for such primes having a few thousand digits. Much of the implementation<br />
work has been pioneered by F. Morain, who applied ideas of A. Atkin<br />
and others to develop an efficient elliptic curve primality proving (ECPP)<br />
method, along with a newer “fastECPP” method, discussed in Chapter 7. A<br />
typically impressive ECPP result at the turn of the century was the proof<br />
that (2 7331 − 1)/458072843161, possessed of 2196 decimal digits, is prime (by<br />
Mayer and Morain; see [Morain 1998]). A sensational announcement in July<br />
2004 by Franke, Kleinjung, Morain, and Wirth is that, thanks to fastECPP,<br />
the Leyland number<br />
4405 2638 + 2638 4405 ,<br />
having 15071 decimal digits, is now proven prime.<br />
Alongside these modern factoring achievements and prime-number analyses<br />
there stand a great many record-breaking attempts geared to yet more<br />
specialized cases. From time to time we see new largest twin primes (pairs of<br />
primes p, p+2), an especially long arithmetic progression {p, p+d,...,p+kd}<br />
of primes, or spectacular cases of primes falling in other particular patterns.<br />
There are searches for primes we expect some day to find but have not yet<br />
found (such as new instances of the so-called Wieferich, Wilson, or Wall–Sun–<br />
Sun primes). In various sections of this book we refer to a few of these many<br />
endeavors, especially when the computational issues at hand lie within the<br />
scope of the book.<br />
Details and special cases aside, the reader should be aware that there<br />
is a widespread “culture” of computational research. For a readable and<br />
entertaining account of prime number and factoring “records,” see, for<br />
example, [Ribenboim 1996] as well as the popular and thorough newsletter<br />
of S. Wagstaff, Jr., on state-of-the-art factorizations of Cunningham numbers<br />
(numbers of the form b n ± 1forb ≤ 12). A summary of this newsletter is<br />
kept at the website [Wagstaff 2004]. Some new factorization records as of this<br />
(early 2005) writing are the following: