20190801, 02:32  #1 
Mar 2004
775_{8} Posts 
Question on how to read worktodo
Hi all! I'm trying to interpret the description of the PRP section of the worktodo, but ran into a question I couldn't answer and asked someone who wasn't sure either.
Here is the PRP description from the readme: Code:
The PRP choice, available from the menus only in the Mac OS X version, lets you do a probable prime test on numbers of the form k*b^n+c. On all OSes, you can edit the worktodo.txt file directly. For example add: PRP=k,b,n,c[,how_far_factored,tests_saved][,prp_base,residue_type][,"commaseparatedlistofknownfactors"] where the how_far_factored and tests_saved values are used to pick optimal bounds for P1 factoring prior to running the PRP test. Code:
PRP=HEX,1,2,332XXXXXX,1,77,2 //from my worktodo.txt file PRP=HEX,1,2,332XXXXXX,1,79,0,3, //from my account online. Shoutouts to ATH!! UNRELATED NOTE: It says PRP is only available in the Mac OS X version. That's a typo, I assume? 
20190801, 03:11  #2 
Apr 2019
11001101_{2} Posts 
Hi, check this reference post:
https://www.mersenneforum.org/showpo...8&postcount=22 Looks like that column would be "tests_saved", which is also explained in more detail in the same post. 
20190801, 04:14  #3  
Mar 2004
775_{8} Posts 
Quote:
Last fiddled with by JuanTutors on 20190801 at 04:14 

20190801, 06:13  #4 
Romulan Interpreter
Jun 2011
Thailand
5·1,877 Posts 
The "tests saved" basically tells to TF and P1 if that is a firsttime LL/PRP test or a doublecheck, so a value other than 1 or 2 makes no sense, unless you do P1 "for a living" hehe. Basically, if you find a factor (by TF or P1, in this case) before any LL/PRP is done, you save two tests (the first time test and the double check test), and if you find a factor after LL/PRP first time test is done, you save one test, the DC test.
Now, if you save more tests, like 50 days of work, instead of less tests, like 25 days of work only, it worth doing some more TF or P1. Because if you find a factor, you save 50 days of (yours or someone else's) work, then you can afford doing 23 days P1, for example, but if you only save 25 days, better do only one day of P1. You got the idea. This information (tests saved) is used mainly by P1 when selects the bounds, i.e. you go with 34% chances to find a factor (and work one day to save 25) or with 56% (and work 2 days to save 50). Or waste your time, if unlucky, hehe. In the past, the "tests saved" was also used to select the bitlevel, when you do TF. Doing TF for first test was one additional bitlevel compared with doing TF after first test, and before DC (because you save 2 tests, instead of 1, and the amount of TF work doubles with each bitlevel, see the graphs from this page, the red and orange lines, for more explanation). But now nobody is doing TF with the CPU, since the GPUs are hundred of times faster for TF. So, from the POV of doing LL/PRP, those parameters are not relevant. Now, some people like to find factors and will TF/PM1 exponents for which both FT and DC was done (i.e.mersenne numbers we know they are not prime, but we don't know their factors). Doing that, you save no tests, and such activity is not useful for GIMPS, whose main goal is finding primes. But hey, finding factors is cute From the same reason, some guys will do P1 "forever", what we jokingly call "for a living", they pick an exponent and increase those boundaries until they find that record factor, or they get bored and move to another exponent. When prime95 choose the bounds for the P1, it will consider how many tests are saved, but it has no idea that the number can only be 1 or 2. So, if you want to do less P1, or more P1, you can modify this value. For P95, that is just a number used in the formula for B1 and B2, and and if you put there 5 or 7, P95 will think that you "save a lot of tests" (even if this is just a figure of speech, in the best case, statistically, you can only save about 2.03 test, the decimals come from the percent of tests returning errors, for which we need to do triple and quadruple checks) and will use higher B1/B2, increasing your chance to find that record factor (and wasting your more time into the test). We do not remember the largest value you can use, it is either 5 or 9, after which there is no visible effect in B1 or B2, but we remember it was a single digit value, in the past when we were doing tons of P1, we used to edit these values by hand before putting them in the worktodo, but since the GPU era, we don't do much P1 with the CPU. Last fiddled with by LaurV on 20190801 at 06:44 
20190801, 14:30  #5 
Mar 2004
111111101_{2} Posts 
@laurv so that 2 after the 1 means, "Computer, do enough P1 factoring to save 2 tests worth of work statistically"?

20190801, 16:01  #6 
Mar 2004
509 Posts 
Also I'm still not clear why the worktodo file I got automatically has a 2 but the online one has a 0.

20190801, 16:45  #7  
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
5×19×53 Posts 
Quote:
Computer, evaluate many combinations of P1 bounds and pick the combination that looks to promise the greatest savings of overall computing effort, statistically, given time to run P1, time to run primality tests, and probabilities of finding factors as a function of those bounds, saving tests_saved=2 primality tests per factored Mersenne number. Minimize the function, something like, in the usual sequence of events, Sum_probable_effort = (TF effort as function i(exponent,bitlevel,etc) + (probability of no TF factor found as j(exponent,bitlevel,etc)) * [ (P1 factoring effort as g(B1,B2,exponent,etc)) + (probability of no P1 factor found as f(B1,B2,exponent,etc)) * tests_saved * (effort per primality test as h(exponent,etc)) ] If the sequence differs, the function to be minimized differs. In reality, tests_saved is a fraction more than 1 or 2 due to error rates, and none of the effort functions are tidy explicit formulas. TF effort is approximately proportional to 2^{bitlevel}/exponent, primality testing is approximately proportional to exponent^{2.1}, P1 optimized bounds tend to approximately track the order of primality testing. See https://www.mersenneforum.org/showpo...27&postcount=2, particularly page 6 of the attachment. The vertical space between the two fits there is probable computing effort saved by P1. An important caveat to that savings is if the bounds used are not adequate to retire the P1 task for the exponent in the ~97% probability noP1factorfound case (such as the bounds were constrained by limited RAM), someone may need to run another, and that increases total P1 cost. Inadequate P1 bounds are to be avoided, in my opinion. Last fiddled with by kriesel on 20190801 at 16:55 

20190801, 16:54  #8  
Apr 2019
205_{10} Posts 
To just directly quote the reference post this time:
Quote:
If you are only looking at PRP lines, I think it would just be extra information and shouldn't affect how the PRP job itself is done. Also you can look at the history of your specific exponents and see exactly what work has been done if you are curious: https://www.mersenne.org/M332XXXXXX Last fiddled with by hansl on 20190801 at 16:56 

20190801, 16:59  #9 
Mar 2004
775_{8} Posts 
Cool, thanks for the explanation. I think I'll leave the "2" instead of the "0" for a slightly larger chance of finding a factor, just for my personal preference. I think I get why some people put that number all the way up to "9", but I'm not about that life ðŸ˜‚

20190801, 17:07  #10  
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
5·19·53 Posts 
Quote:
https://www.mersenne.org/report_expo...exp_hi=&full=1 click the exponent value under Mn, goes to https://www.mersenne.ca/exponent/332419523 GPU72 TF level + GPU72 B1 and B2 exceeded, good enough. Final TF levels, P1, and PRP all are being done on a gpu, so the gpu targets apply. I go by the GPU72 line for gpu work and primenet line for cpu work. Last fiddled with by kriesel on 20190801 at 17:07 

20190801, 17:07  #11  
Mar 2004
509 Posts 
Quote:
(If it's a bug in the system, perhaps it's because I read somewhere that Prime95 does P1 factoring before a second round of factoring for 100M digit mersennes?) 

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
PLEASE READ BEFORE POSTING!  ewmayer  Information & Answers  1  20190405 03:13 
How I read ECM  lukerichards  Lounge  2  20190318 11:16 
I am sorry please read this  meeztamike  Miscellaneous Math  3  20060103 01:47 
chance of finding a factor?......Read me read me read me :)  Firedog18  Software  9  20030725 17:10 
Please read!!!!!  andi314  Lone Mersenne Hunters  1  20030220 13:53 