TECHNICAL NEWS RELEASE  (29 Sep 1999)

DEEPEST COMPUTATION IN HISTORY, FOR A YES/NO ANSWER

Contact:

Dr. Richard Crandall
Director
Center for Advanced Computation
Reed College, Portland Oregon
(503) 777-7255
email: [EMAIL PROTECTED]

What is believed to be the deepest computation -- for a simple "yes/no" or  
"1-bit" answer -- in history has just been completed by a team of three  
investigators: Ernst Mayer, formerly of Case Western Reserve University,  
Cleveland, Ohio; Jason Papadopoulos of the University of Maryland, College  
Park, Maryland;  and Richard Crandall of the Center for Advanced Computation, 
 
Reed College, Portland, Oregon.

The computation involves a gargantuan, mysterious number
called F24, the twenty-fourth Fermat number. F24 is over 5 million decimal  
digits in length, and the three investigators have answered the question: "is 
 
F24 a prime number?".  The answer, based on their intensive computation, is  
"no."  This means that there must exist proper factors of F24, though not a  
single explicit factor is yet known.  (See end of article for background on 
the  
celebrated Fermat numbers.)

Mayer and Papadopoulos used independent, floating-point
"wavefront" implementations of the rigorous, classical Pepin primality proof; 
 
which runs were completed on 27 and 31 Aug 1999 respectively, ending up in  
complete agreement on the final Pepin residue, said residue not equal to (-1) 
 
as required  for  primality.  During these "wavefront" runs Crandall used a  
pure-integer convolution scheme in parallel mode (i.e. running on many  
computers simultaneously), to check the periodically deposited wavefront  
residues. With this integer verification, the proof is considered rigorous:  
there is no doubt that F24 is composite.
The mathematical details will be published later (a preprint of the three  
authors' paper is available at www.perfsci.com/free/techpapers).  Many  
colleagues of the three investigators contributed to this massive
computational project (see below for detailed acknowledgements).

F24 = 2^(2^24) + 1 at over 5 million digits dwarfs the current largest known
prime 2^6972593-1, which is a "mere" 2 million digits (see www.mersenne.org,
www.perfsci.com, www.entropia.com).
To convey the scale of the computation, consider that the Pixar-Disney movie
"A Bug's Life" needed about 10^17 (one hundred quadrillion) computer 
operations
for its complete rendering, yet essentially the same number of operations went
into the F24 proof. So the amusing notion is: for 10^17 operations you can
either get a feature-length state-of-the-art synthetic movie, or for similar
computational cost you can get a 1-bit answer (prime/not prime).

Fermat numbers are numbers of the form Fn = 2^(2^n) + 1. Written in binary
the n-th Fermat number consists of a binary one, followed by 2^n zeros and a
trailing one. For example, in binary F2 = 100001 and F4 = 100000000000000001.
Each time you increase the index n by one, the number of binary zeros, and
thus the number of digits (in either binary or decimal form) also roughly  
doubles.  P. Fermat conjectured in the early 1600's that each of the Fn is  
prime.  He had in hand the first five examples F0 = 3, F1 = 5, F2 = 17, F3 =  
257 and F4 = 65537, each of which being indeed prime. However, unlike his  
celebrated "last theorem" recently proved by A. Wiles, Fermat's conjecture  
regarding the primality of the Fn turns out to be about as wrong as can be. 
Not  
a single prime Fermat number is known beyond F4.  For example, F5 = 
4294967297  
is divisible by 641, and every other Fermat number has either exhibited  
factors, or remains of unknown character (prime/composite).

When factoring algorithms fail to produce an explicit factor, the Fermat
number in question can still be subjected to a Pepin test, a deterministic 
test
of primality. This test requires a sequence of squarings of numbers, a member
of the sequence being generally as large as the Fermat number under test,
and one must do as many such squarings as there are binary zeros in the
Fermat number in question. The primality test for F24 thus requires
16777215 squarings, each such squaring being of a number over five
million decimal digits in length. Even though it is now generally believed
that are no more prime Fermat numbers beyond those found by Fermat himself,
the testing of these numbers has proved to be a valuable exercise, since each
new test tends to occur, for the given era, at the edge of feasibility on  
state-of-art  computer hardware, not to mention at the fringe of algorithm
development.

There have also been important theoretical and algorithmic advances spurred by
such work, and many of the fundamental algorithms used for the Fermat numbers
are also widely used in other areas - for example, the two floating-point
Pepin tests of F24 each used an efficient squaring algorithm based on a
procedure called the fast Fourier transform (FFT), which is ubiquitous in the
field of electronic signal processing. The pure-integer convolution
that verified the Pepin test also has wide application in arbitrary-precision
arithmetic.  In developing their separate implementations, each member of the 
 
team (Mayer, Papadopoulos, Crandall) made advances in the matter of  
implementation of the FFT and convolution algorithms on modern microprocessor 
 
architectures.

We note that F22 was resolved (as composite) in 1993.  Various Fermat numbers
beyond F24 have been attributed at least one explicitly known small factor
(see http://vamri.xray.ufl.edu/proths/fermat.html for the current 
computational
knowledge pertaining to Fermat numbers), so that the next unresolved case
is the monstrous F31 which (at over 600 million decimal digits), even with the
aforementioned algorithmic advances, is out of reach with current technology.
We estimate 10000 years processing on hardware of current vintage be required 
 
to resolve F31.  However, history is replete with underestimates on future  
machinery and ingenuity.  We are confident that with ever-faster processors 
and  
further algorithmic advances, in particular those aimed at implementation of  
the Pepin test on massively parallel computer hardware, a test of F31 may 
come  
within the next two decades.  And this perhaps surprisingly optimistic 
estimate
is made even without the benefit of quantum computers, nanotechnology, and
so on; breakthroughs in any of these areas could alter the assailability of 
F31
dramatically.

Background on investigators:

Ernst W. Mayer was until recently on the faculty of Engineering of Case
Western Reserve University in Cleveland, Ohio. He is currently working
as a freelance computational number theorist and algorithm developer
out of his home in Cupertino, Calfornia.

Jason S. Papadopoulos is a graduate student in the Electrical Engineering
school at the University of Maryland, College Park. His interests include
cryptography, computational number theory, and high-performance scientific
computing. Presently he works for 3S Group Inc, a Vienna, VA-based firm
specializing in encryption hardware.

Richard E. Crandall, author, lecturer and computationalist,
is the Director of the Center for Advanced Computation,
Reed College, holding also the title of Apple Distinguished Scientist.
His algorithms have previously been used to resolve the character
of F22, and to discover record primes such as the last several
largest-known Mersenne primes.


Acknowledgements:

The three investigators gratefully acknowledge the theoretical and 
engineering  
contributions from J. Buhler, H. Lenstra, J. Klivington, C. Pomerance, J. 
Selfridge, 
P. Montgomery, G. Woltman. C. Curry, P. Wellin, R. Knapp, S. Wolfram.
Alex Kruppa and the staff of the Infohalle der Fakultaet fuer Informatik an 
der
Technischen Universitaet Muenchen lent considerable processing power and 
personal maintenance time to the integer proof runs.  The first-completed 
floating-
point run was performed on a Silicon Graphics R10000 Octane workstation - the
investigators thank J. Alexander of Case Western Reserve University for his
generous donation of machine time. Thanks are also due to 3S Group 
Incorporated
for the extended use of a Sun UltraSPARC-1, used as the second floating-point
machine. The Number Theory Foundation donated additional machinery for the 
final
stages of the pure-integer proof, and a key segment of said proof
was carried out on Apple Computer's newly announced G4 processor, which
provides giga-op-level performance.  A Reed College team of staff and 
students:
N. Essy, B. Hanson, C. Chen, J. Dodson, R. Richter, W. Cooley, J. Heilman,
D. Turner, (and from Univ. Georgia) C. Gunning finished in glorious and 
selfless  
fashion the last stages of the rigorous integer proof.

For more information:

Consult the algorithm website www.perfsci.com

_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ      -- http://www.tasam.com/~lrwiman/FAQ-mers

Reply via email to