mersenneforum.org > Math searching for Mersenne primes
 Register FAQ Search Today's Posts Mark Forums Read

 2007-08-13, 00:39 #1 davieddy     "Lucan" Dec 2006 England 6,451 Posts searching for Mersenne primes You have two choices: 1) a program that takes 30 days with an x% chance of failing 2) a program that takes 40 days which is effectively reliable. At what x do you opt for 2 rather than 1 ? David Last fiddled with by davieddy on 2007-08-13 at 00:41
 2007-08-13, 01:07 #2 Kevin     Aug 2002 Ann Arbor, MI 433 Posts With x% chance of failure each test run, the expected number of runs to get a correct result is 1/x. The break even point is when 30/x=40, or x=75%. I'm fairly confident this still mostly works even with the application to GIMPS and the system of double-checks in place. However, I have a feeling many people here would place additional value on having things done right the first time and not having to deal with unnecessary double checks due to errors. If this question happens to be related to OCing, I'd only push it far enough to still get consistently correct results.
2007-08-13, 01:29   #3
davieddy

"Lucan"
Dec 2006
England

6,451 Posts

Quote:
 Originally Posted by Kevin With x% chance of failure each test run, the expected number of runs to get a correct result is 1/x. The break even point is when 30/x=40, or x=75%.
I don't think this is right

David

Last fiddled with by davieddy on 2007-08-13 at 01:31

2007-08-13, 02:33   #4
Jens K Andersen

Feb 2006
Denmark

2·5·23 Posts

Quote:
 Originally Posted by Kevin With x% chance of failure each test run, the expected number of runs to get a correct result is 1/x. The break even point is when 30/x=40, or x=75%.
No. With x% chance of failure, there is 1-x% chance of a correct result. The expected number of runs to get a correct result is 1/(1-x%). The break even point is when 30/(1-x%)=40, or x=25.

 2007-08-13, 05:33 #5 Orgasmic Troll Cranksta Rap Ayatollah     Jul 2003 64110 Posts This is assuming that you only care about the expected value
2007-08-13, 13:39   #6
R.D. Silverman

Nov 2003

723210 Posts

Quote:
 Originally Posted by Jens K Andersen No. With x% chance of failure, there is 1-x% chance of a correct result. The expected number of runs to get a correct result is 1/(1-x%). The break even point is when 30/(1-x%)=40, or x=25.
This assumes that your risk aversion function (the unit linear loss function in
Bayesian statistics) is linear. This is also known in economics as a utility
function.

 2007-08-15, 13:25 #7 davieddy     "Lucan" Dec 2006 England 6,451 Posts The (ill-disguised) follow up question is: What exponent size reduces the reliability of a 2048K FFT LLtest to 75% ? David
 2007-08-21, 04:51 #8 Kevin     Aug 2002 Ann Arbor, MI 433 Posts Goddamnit, I'm screwed when school starts in two weeks if I can't even get that right .

 Similar Threads Thread Thread Starter Forum Replies Last Post joblack Lounge 20 2009-01-05 15:18 flosculus Information & Answers 6 2008-11-10 18:59 optim PrimeNet 13 2004-07-09 13:51 Erasmus Factoring 3 2004-05-14 09:26 daxm Miscellaneous Math 5 2003-07-20 19:32

All times are UTC. The time now is 06:49.

Tue Nov 24 06:49:21 UTC 2020 up 75 days, 4 hrs, 4 users, load averages: 1.42, 1.72, 1.75