Long-lived methuselahs

[simeks]

I found a record-breaking methuselah, 23801 generations from an initial population of 12 cells in a 8-by-7 bounding box:

Code: Select all

x = 8, y = 7, rule = LifeHistory
7.A$7.A$4A3.A$A2$.A$3A!


[Rhombic]

When I saw the title, I didn't expect anything interesting, but then I saw that it was simeks!

Congratulations!!
I suggest -entirely up to simeks- naming in some way after acorn? (walnut, peanut...)

[Goldtiger997]

Congratulations simeks!

How did you find it?

When I saw the title, I didn't expect something as good as this!

[gmc_nxtman]

Wow! Very nice find, did you use a script?

[Kazyan]

A quick search with WLS suggests that this should be the canonical form; there are no easily-found 12-cell predecessors. Good find!

[BlinkerSpawn]

It's certainly impressive (nice job on finding it) but a quick perusing of the table of methuselahs suggest that it doesn't actually break any records or obsolete any other methuselahs except for Blom. Excluding that one, all shorter-lived methuselahs have a smaller bounding box (excl. Bunnies 9) and lower initial population (excl. 23334M), and in both exceptions the methuselahs are equal. Yes, it's now the longest-lived 12-cell pattern but it's also two rows (40%!) larger and the margin of "victory" isn't very great.
tl;dr It's a nice find but I wouldn't go as far as to say "record-breaking".
However, we can now say this statement on the wiki is true:

This list may change frequently as new methuselahs are discovered.

[Mr. Missed Her]

I suggest -entirely up to simeks- naming in some way after acorn? (walnut, peanut...)

We could call it "mustard seed." I remember Jesus using its high tree size/seed size ratio in some kind of parable.

[simeks]

Thanks all!
I found the methuselah using a new C search program that uses my GoLGrid library.

I'm currently running exhaustive searches in small bounding boxes up to 12 initial cells. The search is complete for all bounding boxes up to 7-by-7. 8-by-5 is also complete, but 8-by-6, 8-by-7 och 8-by-8 are still running, so the reported methuselah is from a search that isn't complete yet.

... Excluding that one, all shorter-lived methuselahs have a smaller bounding box (excl. Bunnies 9) and lower initial population (excl. 23334M), and in both exceptions the methuselahs are equal. ...

Actually, 23334M has an 11 cell form, as I reported here a while back.

Here's an unusual 17778 generations methuselah from the search, the interesting part starts around generation 15600:

Code: Select all

x = 64, y = 64, rule = LifeHistory
4.A2$3.2A.A$3A$4.A$.A$3A!

[simeks]

Turns out the Blom methuselah has a form with the same population and one generation shorter life-span (23313), but with a bounding box of only 11-by-3 instead of 12-by-5:

Code: Select all

x = 11, y = 3, rule = LifeHistory
3A3.2A2.A$A2.A4.3A$A8.A!


EDIT March 19, 2017:
Found a form of Blom with 12 on-cells instead of 13. Bounding box 13-by-4 and life-span 23315:

Code: Select all

x = 13, y = 4, rule = LifeHistory
11.2A$5A3.A$9.A.A$4.A5.A!


EDIT: April 17, 2017:
There is also a version of Blom with 14 on-cells in a 7-by-6 bounding box with a life-span of 23354 generations, which means it surpasses 23334M in longevity:

Code: Select all

x = 7, y = 6, rule = LifeHistory
3.2A$3.A2.A$3.A.A$.A3.A$3A$A3.A.A!


[simeks]

My ongoing methuselah search found a 71 generation predecessor of Lidka:
It has a population of 13 cells in a 9-by-6 bounding box and lasts for 29126 generations.
Lidka has a population of 13 cells in a 15-by-9 bounding box and lasts for 29055 generation.

Code: Select all

x = 9, y = 6, rule = LifeHistory
6.A$6.3A2$3.2A3.A$3.A4.A$3A5.A!


[drc]

Anywhere this script is available for download?

[simeks]

Anywhere this script is available for download?

I haven't posted it yet... I'll keep searching for a few weeks more I think, then I'll release a version.

Here's another 13 cell result in a small bounding box (8-by-7). The lifespan is 24827 generation:

Code: Select all

x = 8, y = 7, rule = LifeHistory
6.A$3.A.A$A4.A$.2A$2.A.A$3.A$5.3A!


[gameoflifeboy]

My ongoing methuselah search found a 71 generation predecessor of Lidka:
It has a population of 13 cells in a 9-by-6 bounding box and lasts for 29126 generations.
Lidka has a population of 13 cells in a 15-by-9 bounding box and lasts for 29055 generation.

Code: Select all

x = 9, y = 6, rule = LifeHistory
6.A$6.3A2$3.2A3.A$3.A4.A$3A5.A!


Nice! I remember there being a 13-cell predecessor of Lidka here, but that lasted for only 29101 generations.

Kind of sad how that thread never got more popular, as it introduced the longest living 13-cell pattern in a 13-by-13 box (until now), and the longest living 13-cell pattern in a 4-by-20 box.

[simeks]

Surprising and long-lasting (17670 generations) Bunnies variant:

Code: Select all

x = 6, y = 6, rule = LifeHistory
2.A.2A$A.A$A2.2A$A.2A.A$5.A$A.A2.A!


35 generations Blom predecessor:

Code: Select all

x = 6, y = 6, rule = LifeHistory
3.2A$3A$3.A$A4.A$A2.3A$A.2A.A!


[_zM]

I vote for Bunnies in Space (or Rabbits in Space)

[drc]

Surprising and long-lasting (17670 generations) Bunnies variant:

Code: Select all

x = 6, y = 6, rule = LifeHistory
2.A.2A$A.A$A2.2A$A.2A.A$5.A$A.A2.A!


*before pattern starts: Oh, look, another methuselah. I'm assuming it will just do the same exact stuff that Bunnies does!*
*after pattern starts: HolycrepeanLWSS*

[simeks]

A new record for such a small bounding box: 25567 generations from a 6-by-6 pattern. Population is 18 cells:

Code: Select all

x = 64, y = 64, rule = LifeHistory
4.2A$.A3.A$2A3.A$.4A$A3.2A$A.3A!


EDIT March 30, 2017:
Another +25000 result in a small bounding box, (25306 generations, 14 on-cells in a 10-by-4 bounding box):

Code: Select all

x = 10, y = 4, rule = LifeHistory
5.A3.A$.A6.2A$.A2.3A2.A$A.A2.A2.A!


[BlinkerSpawn]

Have you tried looking for long-lasting 1xn or 2xn patterns?

[simeks]

Have you tried looking for long-lasting 1xn or 2xn patterns?

Yes, I've searched all n-by-2 patterns up to n=16 for all on-bit counts. The longest-lasting is this (23662 gens, 20 on-cells):

Code: Select all

x = 16, y = 2, rule = LifeHistory
A2.3A.A.4A.2A$3A2.3A5.3A!


There is also a version of Blom in a 15-by-2 box (23317 gens, 18 on-cells):

Code: Select all

x = 15, y = 2, rule = LifeHistory
.2A.A4.6A$A.A2.4A.A.A.A!


I'm currently searching n-by-1 patterns. The best so far is in a 38-by-1 box, but the search for 38-by-1 patterns isn't complete yet (14911 gens, 27 on-cells):

Code: Select all

x = 38, y = 1, rule = LifeHistory
4A2.4A3.5A2.5A2.3A2.6A!


EDIT: April 17, 2017:
I've completed the search for n-by-1 patterns up to n=40 for all on-bit counts. The pattern above is still the most long-lived found.
Second place (39-by-1, 29 on-cells, 14605 gens):

Code: Select all

x = 39, y = 1, rule = LifeHistory
3A3.9A2.4A.6A2.4A2.3A!


Third place (40-by-1, 25 on-cells, 13783 gens):

Code: Select all

x = 40, y = 1, rule = LifeHistory
3A5.3A3.5A3.4A2.6A2.4A!


[simeks]

An unexpected result - a methuselah in a 6-by-6 bounding box lasting 27406 generations. Population count is 20:

Code: Select all

x = 6, y = 6, rule = LifeHistory
A4.A$A.4A$A3.A$A.2A$.A2.2A$2A.3A!


EDIT April 17, 2017:
I've completed the search in a 6-by-6 bounding box for all on-bit counts, and the pattern above is the most long-lived found.

EDIT May 14, 2017:
Found a methuselah lasting 26659 generations with 16 on-cells in a 8-by-5 bounding box:

Code: Select all

x = 8, y = 5, rule = LifeHistory
4.A$2.A.4A$.2A$3.A2.A$6A!


EDIT May 26, 2017:
While completing the search of patterns in a 11-by-3 bounding box I found a methuselah lasting 25415 generations with 23 initial on-cells. This makes this the longest running methuselah in a bounding box with an area of 33 cells or less:

Code: Select all

x = 11, y = 3, rule = LifeHistory
.4A3.A.A$2A.A.A2.3A$9A.A!


[simeks]

A new runner-up for a methuselah in a small bounding box after Lidka. 18 on-cells in a 12-by-3 bounding box and 27732 generations:

Code: Select all

x = 12, y = 3, rule = LifeHistory
.3A3.5A$6.A.A.A$4A2.2A3.A!


[testitemqlstudop]

Will you release the script now?

[simeks]

Will you release the script now?

I'm currently working on a new version of this script that will be a lot faster, but I can release the version I've been using for the results reported in this thread.

Compile with "./mknative" for Linux, or use the precompiled Windows versions, meth_h4_256.exe if it doesn't complain about your CPU, or meth_h4_128.exe otherwise.

Usage is:
meth_h4_256 <seed width> <seed height> <oncell count> <min reported lifespan> <seed on * 1000000> [<state filename base>]

For example:
> meth_h4_256 7 5 12 14000 0 >7-5-12.txt

to search for methuselahs with a bounding box of 7-by-5 with 12 on-cells, and report all seeds that live for 14000 generations or more.
Results will be in 7-5-12.txt, but methuselahs will be also be reported to console output during the search.

The program used 512 Mbyte memory for hash tables. The hash tables will be cleared of older contents if they ever run out.

Use <seed on * 1000000> to resume a search from the middle of the search space. This will be very slow at the start because the hash table remembering seen patterns will be empty.
If used, <state filename base> will write a lot of data to a file every 90 minutes, but there's no way to import state right now, so it's not of much use

It will only report results with exactly the requested bounding box size and on-cell count. Only the first seed found that evolves the same way will be reported, it may not be the most long-lived version of that methuselah in that bounding box, for example these two patterns have 17 cells in an 8-by-5 bounding box and evolves the same, but the lifespan differs by 13 generations. Only the first one seen of these would be reported:

Code: Select all

x = 56, y = 5, rule = LifeHistory
5.A.A42.2A3.A$.A2.4A40.2A3.3A$2A5.A42.2A.A.A$3.A2.A42.A5.A$4A3.A41.2A
3.A!


Reported lifespans are not totally accurate and need to be verified manually. Some false positives are possible from switch engines and patterns that extend very far to one side, like this:

Code: Select all

2b2o2bo$obo2b2o$3bo$5b3o$bob5o!


Patterns with period 14 oscillators will be reported as "Pattern reached MAX_GENS" - mostly relevant for 1 cell high seeds.

[testitemqlstudop]

error running ./mknative on MacOs Seirra:

Code: Select all

clang: error: unknown argument: '-fno-tree-loop-distribute-patterns'
clang: error: unknown argument: '-fno-tree-loop-distribute-patterns'
Hao-FengjiematoMacBook-Air:meth_h4 qlstudio$

[testitemqlstudop]

It's ok now I installed debian. Thanks! (Also, how long should it take before a successful 8*8 grid 30 on cell search take? What is the most suggested grid-oncell pair?)