ftp://gimps:[EMAIL PROTECTED]/home/gimps/programs.txt identifies the
program codes as follows:
J3,64,John Sweeney. Mac version 1.4 - Bug fix. prev versions can give bad
data
WP1,64,Woltman - PFA version - Win 95
X,64,Richard Crandall program and its successors - UNIX
There is a J4.
Also: WQ1,64,Woltman - Better error checking - Win 95
What were the residues produced by these? They are not listed in
ftp://gimps:[EMAIL PROTECTED]/home/gimps/lucas_v.txt (the double-check file)
I recall George Woltman giving a figure that about 0.5% of LLtests are wrong.
If 1 in 200 or so residues are incorrect, and double checks are completely
independent, then it makes sense that 1 in 40,000 double checked exponents
could have both checks wrong; then a triple check would match neither of
the previous results. (It is a reflection of awesome reliability of code
and hardware, given the number of hours or days that go into a single
residue.) Coincidentally we have about 40,200 double-checks completed.
Given a 3-way tie, the 4th test should (usually) tell which residue is
correct, if the programs used are consistent enough in design that the
residues should match. (See the thread about differing start values
causing residues to vary.)
The most significant point though would be that if all 3 are producing
nonzero residues, we're pretty sure it's composite.
How many cases of disagreeing triple checks are there to this point?
Ken
At 11:04 AM 1998/10/13 +0200, [EMAIL PROTECTED]
(Marc-Etienne Vargenau) wrote:
>The status page tells us that all exponents below 1 318 200 have been
>double checked.
>We will soon hit cases like
>1419967,wayney,WP1
>1419967,SB,X
>1419967,MV1,J3
>(in file hrf3)
>where three different programs (Woltman, mersenne1 and MacLucas) give
>three different results. How will this be resolved? Has one of the
>programs a known bug in this area?
>
>Regards,
>
>Marc-Etienne
>
