"Re-doing" P-1 work

[Kevin]

Is no credit for P-1 factoring given if the exponent has already had P-1 factoring done on it? I have some expired exponents that had low P-1 factoring bounds, so I decided it would be best to do them again with much improved bounds before running an LL test. I took an exponent that was previously at B1=495000 and redid P-1 testing so it's at B1=390000, B2=10432500. The exponent status page shows the new bounds under history, but didn't update the P-1 row and didn't give me any credit for it. I didn't have any problem getting credit for the previous P-1 test, but I don't know if that exponent had any P-1 already done on it or not.


prime.log

[Tue Nov 25 03:51:12 2008 - ver 25.7]
Sending result to server: UID: kmd/kmd09, M29665073 completed P-1, B1=345000, B2=8711250, E=6, Wd8: FF1B54AC, AID: 4A87D0B94AC08E148BD82519560BEA9E

URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=ar&g=6de69d06e2405f2138fc73a980eff532&k=4A87D0B94AC08E148BD82519560BEA9E&m=UID:+kmd/kmd09,+M29665073+completed+P-1,+B1=345000,+B2=8711250,+E=6,+Wd8:+FF1B54AC,+AID:+4A87D0B94AC08E148BD82519560BEA9E%0A&r=5&d=0&A=1&b=2&n=29665073&c=-1&B1=345000&B2=8711250&fftlen=1572864&ss=62701&sh=ECCE46A54DAAFAAC5B2B136B0693264D
RESPONSE:
pnErrorResult=0
pnErrorDetail=CPU credit is 0.9478 GHz-days.
==END==

PrimeNet success code with additional info:
CPU credit is 0.9478 GHz-days.
[Tue Nov 25 22:16:28 2008 - ver 25.7]
Sending result to server: UID: kmd/kmd09, M33053023 completed P-1, B1=390000, B2=10432500, E=6, Wd8: 579DDD05, AID: 2D0DB9F00992B6626C7572116618F532

URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=ar&g=6de69d06e2405f2138fc73a980eff532&k=2D0DB9F00992B6626C7572116618F532&m=UID:+kmd/kmd09,+M33053023+completed+P-1,+B1=390000,+B2=10432500,+E=6,+Wd8:+579DDD05,+AID:+2D0DB9F00992B6626C7572116618F532%0A&r=5&d=0&A=1&b=2&n=33053023&c=-1&B1=390000&B2=10432500&fftlen=1835008&ss=44975&sh=92BEDB4B0268711BD3A138333905C8BA
RESPONSE:
pnErrorResult=0
pnErrorDetail=SUCCESS
==END==

[S485122]

From my experience and some other posts on the forum I think you get credit if both your bounds are higher than the previous ones. When I have to LL test an exponent for which the previous P-1 bounds are low I use a "Pfactor=1,2,exponent,-1,HowFarFactored,4" workunit on a v24 client. (V24 because of the memory problems of v25 that is unable to use more than 903 MB out if 8192 MB.) I am under the impression that one gets to much credit : it should be 'credit for the P-1 work just returned' minus 'credit for P-1 work to the previous bounds' but I get the credit I would earn if no P-1 factoring had been done before.

Jacob

[cheesehead]

Originally Posted by Kevin

I took an exponent that was previously at B1=495000 and redid P-1 testing so it's at B1=390000, B2=10432500.

Jacob's almost right about higher bounds: the actual P-1 ranking algorithm is that highest B1 is first, then only if there are multiple results with that same highest B1 is B2 considered. As long as the ranking is one-dimensional according only to the B1/B2 bounds, that's the way it has to be. So, unfortunately, your B1=390000, B2=10432500 result was not considered an extension past the previous B1=495000 result, ranked by a single-dimensional measure of deepest P-1 done so far.

Even though your search may have had a higher probability of finding a factor than the first search, the way the database ranks P-1 results does not use a factor-probability factor or formula.

If P-1 search depth were portrayed two-dimensionally, it would be possible to represent the combined work done on your exponent by a superposition of a rectangle from x-y coordinates (0,0) to the diagonally opposite corner of (495000,495000) and another rectangle from (0,0) to a corner at (390000, 10432500), with sides parallel to the coordinate axes in each case. The resulting six-sided figure with corners at (0,0), (495000,0), (495000,495000), (390000,495000), (390000,10432500) and (0,10432500) would be a more informative way to represent the portion of P-1 search space that had been covered.

(Technical complications: There'd really need to be shading of the portion above y=495000 because your stage 2 would have found a factor there only if it were the only factor in that region. Then if Brent-Suyama had been used, there'd have to be a cross-hatched and shaded rectangle stretching from (0,10432500) to (390000, some-multiple-of-10432500) in addition. But they're beyond the scope of this thread.)

The way the highest P-1 limit is displayed now is an ordered number pair of (the highest B1, the highest B2 done with that highest B1).

The exponent status page shows the new bounds under history, but didn't update the P-1 row and didn't give me any credit for it.

(* sigh *) Tsk, tsk. The sad outcome of 1-D projection display of a 2-D truth, like a dimension-lower Mercator projection for P-1 results. (* sigh *)