December PRIME 2004 k = 28433

[jjjjL]

!!!1!!

See homepage http://www.seventeenorbust.com/

:D

Cheers,
Louie

[Frodo42]

:cheers:

Great.

Wondering why no one saw this one coming ... I guess we were to preoccupied with all other things to notice the small signs. Well done Louie and David ... this time we didn't catch it :D

It wasn't one of the heaviest K's butit's going to mean a boost to the project in many ways.

7 down, 10 to go ... 10 or bust :p

[kugano]

The only person who could've known is the person who found it... we didn't remove that multiplier from the queue until about 5 minutes before the announcement was posted. You guys caught on to that telltale way too quickly last year and we didn't want to give it away prematurely =)

[ceselb]

Woot, nice.
http://ceselb.myftp.org/dj_dancefloor_smiley.gif

[tqft]

I hope so as I just submitted a story to slashdot on it - probably get rejected though.

[kugano]

Yeah... we're not really pushing the story yet, since we haven't contacted the lucky user. We do know who it is, we just haven't got hold of them yet. Once we have all those details we'll put together a more formal release and start passing it around.

[scrap]

Did you notice ? Decembre is always the mouth we dicover primes. This is strange :bouncy:

[MikeH]

Great job everyone. Very big pat on the back.

I guess that put's us at no. 4 in the largest primes (http://primes.utm.edu/primes/search.php?Number=200) behind M39.

Siever's and P-1 factorer's should check out this post (http://www.free-dc.org/forum/showthread.php?s=&threadid=8240) for sob.dat and scoring updates.

[Mystwalker]

Originally posted by scrap
Did you notice ? Decembre is always the mouth we dicover primes. This is strange :bouncy:

The first one came in November - apart from that, yes, very strange. But who cares?!? :D

:|party|: :|party|: :elephant: :elephant: :elephant: :elephant: :|party|: :|party|:

Interesting coincidence:

- the new PRP client is out, greatly suited for new members
- k = 5359 has been finally completely checked below n = 5054502 recently

Any more conspiracy theories? :crazy:

[kugano]

We're kinda surprised at the December coincidence, too. But there may be a rational explanation for it as a self-reinforcing phenomenon. Because the last 2 years also gave us primes in December, a lot of people were anxious (jokingly or otherwise) to see one this month, too. So with all the "it's that time of year" gauntlets, more work got done in December than in other months, making the prime-find more likely.

Next year it'll probably only be "worse" as more people will expect a prime in December and make the project output surge even more.

[vjs]

Any predictions for where we will be by then?



Prime Predicted N range Prime Date
1st 800K-1.1M prime found: 46157•2^698207+1 27 November 2002
2nd 1.1M-1.6M prime found: 44131•2^995972+1 6 December 2002
3rd 1.6M-2.2M prime found: 65567•2^1013803+1 3 December 2002
4th 2.2M-3.1M prime found: 69109•2^1157446+1 7 December 2002
5th 3.1M-4.4M prime found: 54767•2^1337287+1 22 December 2002
6th 4.4M-6.5M prime found: 5359•2^5054502+1 6 December 2003
7th 6.5M-10.1M prime found: 28433•2^7830457+1 30 December 2004
8th 10.1M-16.8M
9th 16.8M-29.5M
10th 29.5M-57M
11th 57M-127M
12th 127M-344M
13th 344M-1.2G
14th 1.2G-7.2G
15th 7.2G-108G
16th 108G-18T
17th above 18T


I guess it sort of shows how these predicitions are off, I'd rather go by the trend of, bunch of primes grouped together by n, 698207 and 5054502, being the odd balls.

[Mystwalker]

Originally posted by vjs
Any predictions for where we will be by then?

I think we should be around 11M, so estimations won't contradict a new prime. :)

[vjs]

O.K. I'll make an estimation...

Progress around 250-260 tests per day... a prime in december 2005 that 330-360, days from now.

So roughly 82500-93600 test from now...

Estimate 60 tests eliminated by sieve/P1 per day. Humm, roughly ... 2,500 tests eliminated within or before active testing. All client progress users etc in the 250-260 tests etc.

Put's us somewhere in the 11.1m to 11.6 range for december 2005!!!

Well looks like our next prime could easily be sometime december 05...

I personally think that we may actually be supprised and find another well before then. It would be great to get another in the low 8's, it's not impossible. Then we probably wouldn't find another until around 18m if not later.

[wblipp]

Originally posted by vjs



Prime Predicted N range Prime Date
1st 800K-1.1M prime found: 46157•2^698207+1 27 November 2002
2nd 1.1M-1.6M prime found: 44131•2^995972+1 6 December 2002
3rd 1.6M-2.2M prime found: 65567•2^1013803+1 3 December 2002
4th 2.2M-3.1M prime found: 69109•2^1157446+1 7 December 2002
5th 3.1M-4.4M prime found: 54767•2^1337287+1 22 December 2002
6th 4.4M-6.5M prime found: 5359•2^5054502+1 6 December 2003
7th 6.5M-10.1M prime found: 28433•2^7830457+1 30 December 2004
8th 10.1M-16.8M
9th 16.8M-29.5M
10th 29.5M-57M
11th 57M-127M
12th 127M-344M
13th 344M-1.2G
14th 1.2G-7.2G
15th 7.2G-108G
16th 108G-18T
17th above 18T



Looks like remarkably lucky forecasting!

I made this forecast on May 5th 2003, when we were analyzing why the sixth prime was so long coming. Using these figures, I argued that the sixth was on time, it was just that the fifth had been early. Most people didn't like the forecast, but with six and seven both coming inside their windows, the methodology is looking good.

William

[vjs]

My hat's off to you

:allhail:

It has become a staple prediction around here.

I don't like the

16th 108G-18T
17th above 18T

But hey... lets hope your predictions break down around the 10th or 11th and all are found by then... :D

[kugano]

Congratulations to Team Prime Rib!!! (http://www.free-dc.org/forum/showthread.php?s=&threadid=8248)

[Keroberts1]

actually i believe the reasoning for such high predictions for the final primes is because of the low proth weight K values. Being that we've been finding the lower ones first we can probably expect the final date to move up. Probably can expect it to signifigantly rise if 67607 is found next. Please correct me if I'm wrong I'd love ot hear from Wblipp on this and to see if this discovery has effected the estimates.


Also a question i had.

I heard some mention a while back about possibly sending out a bulk e-mail to old participants. Perhaps wiht this new discovery we should ty this. Between the new client and the new prime we should have what it takes to pull alot of people who used ot play with us back into the circle.

[vjs]

Originally posted by Keroberts1
actually i believe the reasoning for such high predictions for the final primes is because of the low proth weight K values. Being that we've been finding the lower ones first we can probably expect the final date to move up.

Wow excellent point but I'm not sure if I understand correctly, by saying up you mean sooner right?



K Probability
4847 0.98860
5359 0.99469 Prime Found
10223 0.99539
19249 0.85992
21181 0.98736
22699 0.84058
24737 0.98830
27653 0.94479
28433 0.91430 (most recent prime)
33661 0.98703
44131 0.99661 Prime Found
46157 0.87826 Prime Found
54767 0.99464 Prime Found
55459 0.99832
65567 0.83838 Prime Found
67607 0.79880
69109 0.97235 Prime Found


If you sort them by probability



K Probability
55459 0.99832
44131 0.99661 Prime Found
10223 0.99539
5359 0.99469 Prime Found
54767 0.99464 Prime Found
4847 0.98860
24737 0.98830
33661 0.98703
21181 0.98736
69109 0.97235 Prime Found
27653 0.94479
28433 0.91430 (most recent prime)
46157 0.87826 Prime Found
19249 0.85992
22699 0.84058
65567 0.83838 Prime Found
67607 0.79880


It seems to be sort of random, in other words we have removed the same amount of low probabilty as high probablility primes. I wonder what this really means...

For example if our last prime is found at 100's of G it would take alot more k/n pair testing if we eliminated the low probability primes first. (I believe the probablilites were based upon how many k/n pairs there are per n=1m range.

So is it better for the project as a whole to eliminate the low probablility first (no we can't choose) or last? Of course I'm not suggesting doing any particular k first since we don't know where primes are etc.

Interesting

[kugano]

I haven't gone through wblipp's calculations in much detail, so take this with a grain of salt. But my take on the whole thing is that regardless of how technically accurate the math is, we've proven there's a pretty low correlation between reality and the model. Wblipp's methods may be sound, and when you look at the "big picture" (all k, not just our remaining candidates), they may give excellent results. But we're only looking at only a tiny slice of the big picture, as through a magnifying glass. Any empirical model, no matter how well thought out, has a bit of error. The model is very good for predicting patterns and trends in the whole, but very bad for predicting specific events, as we're trying to do with our remaining multipliers. I think at this point in the lifetime of the Sierpinski problem, no model will give us any reliable predictions as to how the rest of it will go.

[vjs]

Agree there is also the coin flip analogy... what are the chances of flipping 4 and 5 heads in a row?

(0.5)^4 = 6.25%
(0.5)^5 = 3%

So what is the probablity of flipping the 5th head once you flip the fourth?

Still 50%

I find it very difficult to believe that we won't find all of the primes except for maybe 1 or 2 before 100m. From a project stand point once we pass GIMPS (n=~26-27m) we will probably also have alot more users as well. Largest prime etc.

n=33.4m doesn't sound unreasonable anymore and also something to look forward to. Remember when n=13m primes seemed huge... perhaps that account should also be killed now.

[kugano]

Of course, I remember when n = 1,000,000 seemed huge... =)

The latest prime will help, but I think the altogether best thing we can do to bring in more power is to fix the client (I know, I know, don't say it), and then to update the website. Right now people have to bend over backwards to use the client in all but the most mundane environments, and that's not doing much for user retention. And the website could be a lot more friendly. Right now it probably scares the hell out of anyone who isn't a math or DC junkie. Not the way to attract the average Internet user.

[wblipp]

I share the hope that the model breaks down soon, and we find all the remaining primes much sooner than the model predicts.

The model basically says that candidates have a probability of being prime that follows the Prime Number Theorem adjusted for the Proth Weight of the k-value. This is a model of gazillions of very low probability events. The variability of results is huge in such models - it has to be counted as a lucky break that the last two results came within the ranges predicted.

I don't update the model when primes are found. Updating the model encourages belief that the model has greater predictive ability than is justified, and leads to counter intuitive effects when looking at fine grain predictions. I'll make an exception to this when we find a prime for 67607 - this k value has the lowest weight and it dominates the probability that we must go so high to find all seventeen primes.

[Frodo42]

What will happen if I expire a test for k=28433 ... will it just be moved into a new que?

I am considering to move some of my slow P3's to sieving and since I need to expire some tests to do that I thought it would be best to expire tests with k=28433, but if they just get moved to another que, then I might as well finish them and expire some of the ones that has not been prp'd as much (prp'iing one test with n~8000000 takes something like 1.5 months on these machines right now).

[vjs]

Frodo,

If your actually considering sieving instead of picking up a new test for that p3 I'd say dump it if it were less than 50%.

In that time 0.75 month you would probably eliminate at least one test that would enter prp anyways, IMHO, P3 or < shouldn't maineffort prp anyways there are better things for them to do.

Sieve, P-1 if it has alot of memory, even double-check.

[Vato]

I run 5 machines that are P3s or Athlons that many sievers keep saying should be moved to sieving or P-1. I just don't have the time for manual intervention - so more automation may convince me. But at current n, that's >80 PRPs completed per year - is that really so unworthwhile? (With tongue firmly in cheek) ;)

[olaright]

Bravo for the new prime!

:elephant: :elephant: :elephant: :elephant: :elephant: :elephant:
:|party|: :|party|: :|party|: :|party|: :|party|: :|party|: :|party|:
:banana: :banana: :banana: :banana: :banana: :banana:

Is it going to be included in Chris Caldwell page?

[wirthi]

It is as far as I see, if you mean THIS (http://primes.utm.edu/bios/page.php?id=429) page.

On that page one can read: "As of January of 2005, Seventeen or Bust has eliminated ten of those seventeen candidates. The project might now be styled "Ten or Bust," but the original name will be kept for consistency."

That does not sound correct to me. SOB has eliminated SEVEN, so now we can call it "Ten or Bust". Or am I missing something? Perhaps one of the admins ought to change that side (you have the password guys)

edit: typo

[Nuri]

you're right.. ;)

[kugano]

Oy! My fault. Fixed, thanks for the heads-up.

Yes, we've been in contact with Caldwell for some time. We had asked him to wait until we had verified the prime with a different program and on different hardware before adding it to his database. This took until last Thursday to finish. We sent him the all-clear but haven't gotten a response yet. He's probably just busy or out of town.