P-1 bounds calculation question

axn · 2006-01-04, 18:10

I was trying on Prime95's Pfactor option to do some P-1 on numbers of the form k*2^n+1. The following are some of the bounds that the program used. All numbers used 34,1.5 parameters (factored to 34 bits, one factor saves 1.5 tests).

980535*2^438222+1 completed P-1, B1=15000, B2=105000, Wc1: 0B7EC43F
980535*2^450435+1 completed P-1, B1=15000, B2=108750, Wc1: 0BC8F208
980535*2^460337+1 completed P-1, B1=10000, B2=97500, Wc1: 0BFAB234
980535*2^498747+1 completed P-1, B1=10000, B2=105000, Wc1: 0D05D273

P95 was given 200M of memory (day & night).

I am really confused! AFAIK, all else being the same, bigger numbers should get higher B1/B2 bounds. But here, P95 goes down from a higher B1 to a lower one!

Anybody has got any ideas? Geroge?!

PS:- There were many more P-1 tests I ran. The four shown above are just representative. All of them used 48k zero-padded FFT. Each P-1 took < 1 min on P4 2.8GHz.

Prime95 · 2006-01-04, 20:40

It sure looks like a minor bug. I can't tell without stepping through the bounds-generating code.

Your logic that the PRP test takes longer for bigger numbers, thus a factor is worth more, thus deeper P-1 is required is correct.

If it isn't a bug, then all I can think of is these B1/B2 combinations are real close and it is a rounding error introducing seemingly illogical choices.

It would be great if akruppa, the P-1 bounds expert bar none, would review the Mersenne bounds generating code - hint, hint.

2006-03-04, 16:49

Hi,
How can i calculate such "thing"
2^4637852?
What software shal i use?

2006-01-04, 18:10	#1
axn Jun 2003 5,051 Posts	P-1 bounds calculation question I was trying on Prime95's Pfactor option to do some P-1 on numbers of the form k2^n+1. The following are some of the bounds that the program used. All numbers used 34,1.5 parameters (factored to 34 bits, one factor saves 1.5 tests). 9805352^438222+1 completed P-1, B1=15000, B2=105000, Wc1: 0B7EC43F 9805352^450435+1 completed P-1, B1=15000, B2=108750, Wc1: 0BC8F208 9805352^460337+1 completed P-1, B1=10000, B2=97500, Wc1: 0BFAB234 980535*2^498747+1 completed P-1, B1=10000, B2=105000, Wc1: 0D05D273 P95 was given 200M of memory (day & night). I am really confused! AFAIK, all else being the same, bigger numbers should get higher B1/B2 bounds. But here, P95 goes down from a higher B1 to a lower one! Anybody has got any ideas? Geroge?! PS:- There were many more P-1 tests I ran. The four shown above are just representative. All of them used 48k zero-padded FFT. Each P-1 took < 1 min on P4 2.8GHz.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
P-1 bounds calculation for Wagstaff numbers	axn	Software	10	2018-07-24 06:27
Question on P-1 bounds	NBtarheel_33	Math	1	2016-05-09 13:10
Question about parameters and bounds	Yamato	GMP-ECM	0	2007-01-06 15:29
Question about lower bounds for prime forms	jasong	Miscellaneous Math	3	2006-01-20 22:00
Question about factoring bounds	drew	Software	4	2005-11-22 03:23

2006-01-04, 20:40	#2
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 7526₁₀ Posts	It sure looks like a minor bug. I can't tell without stepping through the bounds-generating code. Your logic that the PRP test takes longer for bigger numbers, thus a factor is worth more, thus deeper P-1 is required is correct. If it isn't a bug, then all I can think of is these B1/B2 combinations are real close and it is a rounding error introducing seemingly illogical choices. It would be great if akruppa, the P-1 bounds expert bar none, would review the Mersenne bounds generating code - hint, hint.

2006-03-04, 16:49	#3
TolyaAN 1011001101000₂ Posts	Hi, How can i calculate such "thing" 2^4637852? What software shal i use?