![]() |
|
|
#23 | |
|
Nov 2003
22×5×373 Posts |
Quote:
Once again we have a difficulty with semantics. For a *given* integer 2^p-1 it is prime with probability exactly 0 or 1. It is either prime or it isn't. For a *randomly* chosen value of X in some interval X \in [A, B], such that X has no prime factors up to k the probability is (as you say) given by Mertens' theorem *provided* that k is not too large relative to X. Once q1, q2 > log(X), the probabilities that q1 or q2 might divide X are NO LONGER INDEPENDENT (because now product of p_i, p_i < log(X) becomes greater than X) and Merten's theorem becomes innacurate. [i.e. one can't have all the primes divide X because their product is too large. They get in each other's way] One can't prove the PNT by sieve methods because the fundamental lemma of the sieve gets in the way. Compare the PNT with Mertens' theorem where the product is taken over all primes up to sqrt(N). You get different answers because Mertens' theorem is not valid beyond log(N). [Actually, a weighted version of the sieve can take this to log^r N for small integer r > 1] But this is for randomly chosen X and *not* for X of special form such as 2^p-1. We have no proof on the infinitude of Mersenne primes precisely because we have no theorems saying that numbers of this form have the same divisibility properties as random integers. We have no proof that Mersenne numbers behave as random integers do. We can only state that *heuristically*, for a randomly chosen 2^p-1 [in some bounded interval] that its divisors behave in a way that lets us apply Mertens' theorem. We run into the same problem in the time complexity analysis of NFS. We *assume* that integers of the form (-b)^n f(-a/b) [for irreducible polynomial f of degree n] have the same smoothness properties as random integers of the same size. But their is no PROOF. Which is why the time complexity estimates for QS and NFS are *heuristics*. |
|
|
|
|
|
|
#24 |
|
"Lucan"
Dec 2006
England
2×3×13×83 Posts |
The card at the bottom of a deck is either the ace of spades or it's not.
If I inspect the top n cards and no ace of spades appears, the probability of the bottom card being the ace of spades is 1/(52-n). It seems you wish to dispute something about this assertion. |
|
|
|
|
|
#25 | |
|
Nov 2003
11101001001002 Posts |
Quote:
The same does NOT apply to integers of special form (such as Mersenne Numbers). And you seemingly fail to understand the limitations of Mertens' theorem and my statement that once you have sieved (or trial divided) sufficiently many primes, that subsequent sieving or trials results in the probabilities no longer being independent. Go read Halberstam & Richert's book on sieve methods. [but be careful, the book is dense and obtuse; I've kicked it across the room in frustration more than once]. |
|
|
|
|
|
|
#26 |
|
"Lucan"
Dec 2006
England
2·3·13·83 Posts |
The limitations of Merten's theorem are obvious, but you
seem to be implying that it is meaningless to speak of a probability other than 0 or 1. |
|
|
|
|
|
#27 |
|
"Phil"
Sep 2002
Tracktown, U.S.A.
3·373 Posts |
But by your previous logic, the probability of there being a factor less than X must be either 0 or 1, i.e., it either exists or does not.
|
|
|
|
|
|
#28 |
|
Feb 2007
24×33 Posts |
Prime95 says things like
"The chances that the exponent you test yields a prime are about 1 in xxxxxxxxx." This IS a probability, for a GIVEN exponent p. Last fiddled with by m_f_h on 2007-03-09 at 20:58 |
|
|
|
|
|
#29 | |
|
Nov 2003
22×5×373 Posts |
Quote:
For a *randomly chosen* integer, drawn uniformly at random from some interval, then we can say something about the p.d.f for its factors. Otherwise: A more accurate statement would be that *ECM* is a random decision procedure that sometimes makes errors and we can compute the probability that this *decision procedure* is in error when it declares that, e.g. no factor less than X exists. |
|
|
|
|
|
|
#30 | |
|
Nov 2003
22×5×373 Posts |
Quote:
number to test. (2) It assumes mathematical facts that are not proven; that Mersenne primes behave as other, uniformly chosen random integers. But we do not know this to be the case. We know the probability of a uniformly random chosen integer being prime by the PNT. No such theorem exists for Mersenne numbers. |
|
|
|
|
|
|
#31 | |
|
Feb 2007
24·33 Posts |
Quote:
But the statement from Prime95 I refer to IS specific to the selected exponent! It uses information about its size, previous LL tests if any, previously done factoring etc. (I wonder why it does not just use the simple binary information which is its primality...) |
|
|
|
|
|
|
#32 | ||||||
|
"Richard B. Woods"
Aug 2002
Wisconsin USA
22×3×641 Posts |
I think you are failing to distinguish between a priori probability and a posteriori probability.
Those you "don't understand" are referring to probability, given only pre-test knowledge: a priori probability. That is, if we expect 1 Mersenne prime in a range of 100,000 exponents, the a priori chance of any specific Mersenne exponent's being that of a prime is 1/100000. Only after we have tested them do we find that, in fact (a posteriori), 3 (for example) of them give Mersenne primes and 99997 denote composites. In that example, the a posteriori probabilities were 1 for three of the exponents, and 0 for 99997 of them. Quote:
Quote:
Quote:
Quote:
E.g., out of all 2^p-1, with p in a given range, and with no factors less than X, and with the experiential knowledge gained from complete tests of other Mersenne numbers or from proven theorems about Mersenne numbers, we can predict that, on average, Z% of such exponents will eventually (after complete testing) turn out to correspond to Mersenne primes. Z is an a priori probability. Quote:
Quote:
The LL test is a single test. Trial division on one potential factor is a single test. Trial division on multiple potential factors consists of multiple tests. Each completed single test adds to our knowledge and enables us to refine our a priori estimate of probability of primality even though we have not yet performed all possible trial divisions (up to the square root). Last fiddled with by cheesehead on 2007-03-10 at 00:54 |
||||||
|
|
|
|
|
#33 | |||||||||
|
"Richard B. Woods"
Aug 2002
Wisconsin USA
22×3×641 Posts |
Some of you knowledgable folks have been indulging in the bad habit of attributing lesser intellect to those whom you do not understand.
I suggest that if you "do not understand" someone, you need to remember that you may be part of the problem! Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Well, that's convenient, isn't it? A quick, simple explanation ... gratifying ... almost ... instantly. Avoids all that tedious analysis of whether "instant gratification" actually has anything to do with a novice's confusion between TF and LL, doesn't it? :-) Quote:
Why spend time figuring out a reason for a novice's error that is respectful of him/her (and, BTW, is a better fit to the observed facts)? So much easier to chime in with the "instant" answer. Quote:
- - - - Y'all oughta be ashamed of yerselves. |
|||||||||
|
|
|
![]() |
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| Mlucas on ubuntu | Damian | Mlucas | 17 | 2017-11-13 18:12 |
| MLucas on IBM Mainframe | Lorenzo | Mlucas | 52 | 2016-03-13 08:45 |
| Mlucas on HP-UX/PA-RISC setup question | smoky | Mlucas | 14 | 2009-05-05 15:40 |
| mlucas on sun | delta_t | Mlucas | 14 | 2007-10-04 05:45 |
| A quick question regarding iterations in Mlucas... | DeadSpam | Mlucas | 4 | 2007-03-05 14:06 |