ConwayLife.com

• Supports arbitrary isotropic 2-state Moore-neighbourhood cellular automata (not just B3/S23);
• Makes use of top-tier modern SAT solvers (namely Armin Biere's cadical and kissat);
• Uses the 'multi-armed bandit' notion from reinforcement learning to learn which of the SAT solvers are best for which subproblems;
• More efficient, being written in C++ and statically linked against the solvers instead of spawning subprocesses and using named pipes;
• Better method of avoiding redundant repetition of work;
• More user-friendly: can digest partials as RLEs (oriented to move between north and north-west) without requiring manual truncation or running that awkward golly2ikpx script.
• Better lattice basis: it tries to grow spaceships in the direction exactly antiparallel to the direction of travel;
• Internal priority heap: it tries to extend the partials with the narrowest backends first (as they're more likely to admit a completion), using length as a tie-breaker (to prefer shorter completions);
• Adaptive symmetry switching: for orthogonal velocities, it can find spaceships which mix and match components of different symmetries (such as a c/3 turtle with an asymmetric tagalong);
• Integrated apgsearch: you can use the familiar -n and/or -k flags from apgsearch, and it will run every phase of every intermediate partial through apgsearch (without requiring Unix pipes!) and upload the results to Catagolue's ikpx2_stdin symmetry. This is useful for finding higher-period spaceships and puffers travelling at the target speed.

x = 21, y = 14, rule = B3-kq4j/S2-i34q
3bo13bo$b2ob2o9b2ob2o$b2ob2o9b2ob2o$bo3bo9bo3bo$2o3b2o7b2o3b2o2$2o3b2o
2b3o2b2o3b2o$2bobo3b2ob2o3bobo$2bo4bo5bo4bo$4bobo2bobo2bobo$8bo3bo$6bo
7bo$5bobo5bobo$6bo7bo!

x = 13, y = 41, rule = B3-kq4j/S2-i34q
b4ob3o$b2obob2obo$obo3b4o$bo2b3o$2bo4bo$b2obo3bo$b2o2bo3bo$7bobo$5bo$4b2o$4b
2o$5b2o$6bo$9bo$3b3obob2o$2bobo$7bo$5b2o2$5bo4bo$5bob2ob2o$11b2o2$4b3ob3o$4b
3obobo$6bo$6bo$5bobo$5bobo2$4b2ob2o$6bo$3b2o3b2o$3b2obob2o$4bo3bo3$4bo$3b2o$
2bobo$3bo!

wwei23 wrote: October 13th, 2020, 11:29 pm It works like a charm, but: From my own runs of ikpx2, doesn't it print far less than 10000 partials in a reasonable amount of time? How would we reach 10000 partials to upload an ikpx2_stdin haul?
EDIT: I'm just going to try it and hope for the best.

wwei23 wrote: October 14th, 2020, 8:12 am This program is so much faster than JLS or gfind! Why haven't c/8 and 3c/8 orthogonal spaceships been found yet?
Also, where are all the higher-period spaceships? I thought there were supposed to be high-period spaceships on the ikpx2_stdin symmetry.
https://catagolue.hatsya.com/census/b3s2/ikpx2_stdin

Moosey wrote: October 14th, 2020, 7:29 pm (1) because small ones don't exist (or they would have already been found
(2) because none were found [thinking emoji]

wwei23 wrote: October 14th, 2020, 7:30 pm 1. I mean that I'm surprised that no one has found c/8 or 3c/8 orthogonal with this program because it's so fast.

Moosey wrote: October 14th, 2020, 7:35 pmwhich makes it difficult to find smaller ships since it can't be told to stay at a width.

LaundryPizza03 wrote: October 16th, 2020, 4:54 pm Would it be possible to apply symmetry to diagonal ships as well? It would speed up diagonal searches like this one.

calcyman wrote: October 17th, 2020, 5:59 am(Adding symmetry to odd-period diagonal searches would be extremely difficult.)

wwei23 wrote: October 17th, 2020, 8:26 am

calcyman wrote: October 17th, 2020, 5:59 am(Adding symmetry to odd-period diagonal searches would be extremely difficult.)

Out of curiosity, why so?
Also, could ikpx2 be adapted to oscillator searches as a "0c spaceship"?

• (x, y, t) is a function of [(x+u, y+v, t-1) for u in [-1, 0, 1] for v in [-1, 0, 1]] according to the cellular automaton;
• (x, y, t) is always equal to (x+a, y+b, t+p);

calcyman wrote: October 17th, 2020, 7:39 pm Maybe clone the repository again? It looks like your submodules are somehow confused.

silversmith wrote: October 18th, 2020, 12:59 pm Is there by any chance a way to save completed ships or pause when a ship is completed? I am running a long search and it would be a shame to just lose a result due being unable to find it.

./ikpx2 [OPTIONS] | tee -i progress.txt

calcyman wrote: October 18th, 2020, 5:26 pm

Code: Select all

./ikpx2 [OPTIONS] | tee -i progress.txt

will make a persistent copy of the terminal output that you can later search through.

calcyman wrote: October 17th, 2020, 11:35 amThe 'torsion-free' case where gcd(a, b, p) = 1 is conceptually much simpler, because it means that the patterns just live on an ordinary grid isomorphic to Z^2; ikpx2 was written using this assumption for the sake of simplicity, and as such can't find oscillators.

MathAndCode wrote: October 19th, 2020, 9:20 pm

calcyman wrote: October 17th, 2020, 11:35 amThe 'torsion-free' case where gcd(a, b, p) = 1 is conceptually much simpler, because it means that the patterns just live on an ordinary grid isomorphic to Z^2; ikpx2 was written using this assumption for the sake of simplicity, and as such can't find oscillators.

So then shouldn't that prevent it from finding spaceships like the XWSSes? (This is not intended to imply that the program should not be able to find XWSSes but instead intended to imply that it should be able to find oscillators.) I think that adaptive symmetry switching should allow it to find both as long as the oscillator has a stator.

MathAndCode wrote: October 19th, 2020, 9:20 pm(This is not intended to imply that the program should not be able to find XWSSes but instead intended to imply that it should be able to find oscillators.)

calcyman wrote: October 20th, 2020, 4:07 am Correct; it can't find XWSSes or oscillators.

wwei23 wrote: October 20th, 2020, 8:38 am

calcyman wrote: October 20th, 2020, 4:07 am Correct; it can't find XWSSes or oscillators.

?????

calcyman wrote: October 20th, 2020, 4:07 amCorrect; it can't find XWSSes or oscillators.