From: Spike (spike66@comcast.net)
Date: Fri Jul 18 2003 - 12:58:25 MDT
Perhaps many of you will recall the excellent
pitch by Greg Stock at Extro4 regarding the
effect of cash prizes on scientific research.
I suggested at that time that we have an excellent
test case with the collaborative online background
computing effort, GIMPS. With GIMPS we have an
hourly-updated measure of participation, so it
is idea for testing Dr. Stock's theories. My
notion at the time was that not only the amount
of the prize but also the actual award structure
that is of primary interest.
This contention has been wonderfully verified
by an observation I have made over the past
few months: As a result of the Electronic
Frontier Foundation's cash reward, the total
computational thru-put of GIMPS has risen
steadily, but ironically the productivity (output)
has begun to actually decline. {8-[
This paradox is a result of the structure of
the prize. EFF offered $50k for the first
million digit prime, $100k for the first 10 million
digit prime, $150k for the first 1E8 and $200k
for the first billion digit. EFF offered no
explanation for the odd prize structure,
but he who pays the piper calls the tune.
The million digit $50k prize was claimed
in June 1999. The GIMPS project progressed steadily
to where it is now searching in the 6 million
digit range, but currently more and more of the
new accounts, signing on with the newer and
faster computers, are jumping ship and searching
up in the 10 million digit range. A 2GHz processor
can complete a 1E7 search in a reasonable time
(about a month) and a one-in-50,000 chance at
$100k is worth a couple bucks in a sense, so
why not collect the free lottery tickets?
This strategy makes the probability of
of finding a record prime on any given day
actually decline, since fewer candidates can be
checked (each one takes longer) and each candidate
has a lower probability of being prime (approximately
in proportion to the number of digits). Furthermore,
if the prize is claimed before the rest of the GIMPS
effort can catch up (extremely likely), it may be many
years hence before the two Mersenne primes hiding in
the 6 to 10 million digit range are discovered.
Furthermore the EFF prize will surely create an
even larger unsearched gap in the 10 million to
100 million digit range, with few volunteers willing
to search for the biggies that will not set any
records, resulting in little fame or riches.
So while the EFF prize may have provided financial
motivation in some cases, it has kinda messed up
this one area of research. One could say prime
number research has been EFFed, in a way perhaps
analogous to the way modern medical research is
spending more resources trying to cure impotence
and baldness than on life extension, for that is
where the money is.
spike
PS does anyone here have Dr. Stock's email @?
Please forward this note to him. s
This archive was generated by hypermail 2.1.5 : Fri Jul 18 2003 - 15:10:21 MDT