LLRnet/PRPnet rally April 4th-11th

[Lennart]

That was a fun rally

I found more primes then expected.


How many did you find on llrnet

Lennart

[vaughan]

I agree that race was close. I had an Intel quad 9450 drop out due to heat exhaustion overnight so its a warning it needs the dust bunnies cleaned out of the heat sink. Computer room got to 40C according to the max/min thermometer so the air con is back on (huh, global warming, bah its computer room warming!)

Nice gauntlet Lumiukko

Takes me back a few years when you blitzed me in Riesel Sieve.

Hey Gary, Kar_bon or Max how many work units do you need to crunch on "NotVeryManyPrimesLeftBehind" to find one now? I went the whole rally and didn't find one.

[mdettweiler]

Originally Posted by vaughan

Hey Gary, Kar_bon or Max how many work units do you need to crunch on "NotVeryManyPrimesLeftBehind" to find one now? I went the whole rally and didn't find one.

For the 7th Drive at current sieve depth and n-level, you can statistically expect one prime every 10,000 candidates tested. That is, after you've tested 10,000 candidates there's a ~62% chance you'll have found a prime. (Gary can explain the mathematical basis for all this in much greater detail than I if you're interested.)

Of course, note that this is just a statistical expectation based on an assumed random distribution of primes; in reality, while this model does fit quite nicely over the big picture, the specific distribution of primes can vary quite a bit. It has been my (and others') observation that primes tend to come in "bunches", i.e. there will often be dense clusters and large droughts rather than a more even distribution; though at this point there is no mathematical basis for this theory and it's just a popular "hunch". Assuming the "bunch theory" is true, then port 9000 happened to hit a few bunches and 3000 hit all the droughts in between. And since you happened to be on port 3000, you were out of luck (as was everybody else on 3000 except Gary and Karsten, who found one prime each and probably came out right on or just a little under the statistical expectation).

[kar_bon]

Here're some countings:

candidates tested all over: 235868
- with PRPnet: 172365
- with LLRnet: 63503


... and it seems, Gary and I have found only 2 primes with LLRnet, the remaining 19 (if I counted right) with PRPnet.

So 21 primes in about 230,000 pairs is in the range, Max said (one prime in 10,000 pairs).

But 2 primes in 60,000 pairs is the old problem with them: estimated but not guaranteed!

Nice work of all! Thanks.

[vaughan]

So I did approx 20,000 pairs to find no primes in this rally; need to run maybe another 10,000 pairs to try to stay on the average for port 3000.

I'm not giving up until I find one.

[gd_barnes]

Why is this big deal being made by several people over PRPnet finding so many more primes than LLRnet? I recall a recent rally where it was just the opposite. It all evens out in the long run. It's just randomness.

Also, reference clumps followed by droughts and vice versa in something that is randomly distributed: In the famous words of one well-known mathetmatician here: pure jibberish. A clump can be just as easily followed by another clump instead of a drought, which makes it even a larger clump. The same applies to droughts. Any attempt to start testing after a long drought in anticipation of a coming clump is akin to the gambler's fallacy.

[gd_barnes]

Originally Posted by kar_bon

Here're some countings:

candidates tested all over: 235868
- with PRPnet: 172365
- with LLRnet: 63503


... and it seems, Gary and I have found only 2 primes with LLRnet, the remaining 19 (if I counted right) with PRPnet.

So 21 primes in about 230,000 pairs is in the range, Max said (one prime in 10,000 pairs).

But 2 primes in 60,000 pairs is the old problem with them: estimated but not guaranteed!

Nice work of all! Thanks.

Only 20 primes were found with the rally. The 21st found by Tim was over an hour after the rally was over.

[AMDave]

Originally Posted by vaughan

So I did approx 20,000 pairs to find no primes in this rally; need to run maybe another 10,000 pairs to try to stay on the average for port 3000.

I'm not giving up until I find one.

I with you there. Me too.

Using the new drives chart, I estimated there should be another 22-ish primes left in this range between now and 1M.

/edit - Make that 21 now that Ototero just got a lucky strike. So he's one up on us.. PS - The grape-vine said Ototero was going on a trip / may have already left, so he may not be able to report it immediately although he said he should be able to get online at some points in his travel. - edit/

I'm quite impressed that I got the asynchronous JSON queries to work on the new Drives chart. And the colours are not always the same depending on which drive query finishes first. It gives a nice effect while the chart is building.
I like it.
I believe we are watching Drive#7
http://www.noprimeleftbehind.net/sta...drive_progress

PS - Kudos to kar_bon for the drives table on his web site. Most helpful.

[vaughan]

Dave that is a pretty chart!

Congrats to Ototero. Primes are scarce on port 3000 at the moment.

[gd_barnes]

Originally Posted by AMDave

I with you there. Me too.

Using the new drives chart, I estimated there should be another 22-ish primes left in this range between now and 1M.

/edit - Make that 21 now that Ototero just got a lucky strike. So he's one up on us.. PS - The grape-vine said Ototero was going on a trip / may have already left, so he may not be able to report it immediately although he said he should be able to get online at some points in his travel. - edit/

I'm quite impressed that I got the asynchronous JSON queries to work on the new Drives chart. And the colours are not always the same depending on which drive query finishes first. It gives a nice effect while the chart is building.
I like it.
I believe we are watching Drive#7
http://www.noprimeleftbehind.net/sta...drive_progress

PS - Kudos to kar_bon for the drives table on his web site. Most helpful.

In IE9, I get a blank page followed by a "noprimeleftbehind is not responding due to a long running script" error followed by the page completely locking up. So not working for me.

[mdettweiler]

Originally Posted by gd_barnes

Also, reference clumps followed by droughts and vice versa in something that is randomly distributed: In the famous words of one well-known mathetmatician here: pure jibberish. A clump can be just as easily followed by another clump instead of a drought, which makes it even a larger clump. The same applies to droughts. Any attempt to start testing after a long drought in anticipation of a coming clump is akin to the gambler's fallacy.

Well, to more precisely elaborate on what I meant, I wasn't meaning to say that a clump is necessarily "due" after a long drought...just that, quite often, once you first find one, you may well have hit a clump. (i.e., you can't use it to predict anything until you're right on top of it) But indeed, you're right, it's all very unmathematical...in other words, "jibberish".