Brian J Beesley writes:
> It would also - improperly - catch those of us that can't use the
> automatic networking but whose machines are slower than the estimate
> for some reason. A participant without email access that sends me
> USMail with his data now and then is about to get caught by this for
> an exponent that he's working on; 60 days simply isn't going to be
> enough even though the prediction was for less than 40 days.
Not neccessarily.
Yes, it would, because only the automatic networking code in Prime95
is supported directly by the server. Since all my activitiy for the
person noted above is via the manual web forms, his status report says
the last activity by his account was in July even though I have
submitted a new residue and a new factor already in October. I had to
ask Scott to manually add 30 days to the predictions to prevent it
from timing out an exponent part way thru the LL test. The only
reason I won't have to do that repeatedly is Prime95's new ability to
do double check LL tests of exponents already tested by Prime95.
What I meant was, if the last date any message was received about an
exponent is more than 4 months ago, reassign it, even if the expected
completion date hasn't yet arrived.
Yes, I agree that something like this has to be done. But it still
causes problems, as described above.
> Also, P-1 factoring, using Factor98, is still useful, there are still
> many exponents under 1000 that noone is working on, and it will work
> on exponents thru 170,000. Reserve exponents for P-1 by sending me
> email; I also have some Factor98 save files for many exponents.
Perhaps you should publicise this a bit...
That was part of the reason I mentioned it, actually. Where else
would it make sense to publicize it?
Does ECM find only prime factors, or is it just that it is so much
more likely to find factors which are numerically small, so that
factors composed of a product of two or more largish primes are most
unlikely to be discovered without discovering the prime factors?
The mers package's ecm3 and ecmfactor only print prime factors because
they check the primality of factors found and factor them in turn if
they fail a pseudo-prime test. But they generally only find composite
factors of Mersennes early on, with low bounds and no already known
factors.
How do you verify that an arbitary number with ~1000 digits really is
prime? (Let alone a number ten times that length!)
There are programs out there, notably Morain's ECPP (Elliptic Curve
Primality Prover), that provide primality proofs. I don't have one
myself, so I don't know a lot about them.
I think the rules would have to state that the "prize" factor must
be found using one of a set of specific programs, which must supply
the "s number" so that the calculation can be verified.
"S number"? And factors can be confirmed quickly and easily, much
easier than large primes; what does it matter how it was produced as
long as it's a new factor, previously unknown?
Will
