Why I Fail at Making Discoveries

[Freywa]

This may be a little tangential here, but I am pretty much a loser when it comes to discoveries. I have no clear reason why, but I've been around for some years now and still have nothing to my name. I think it's because I have little computing power to work with (I use a laptop) and I can't distinguish patterns clearly. What is wrong with me exactly and have I been of any use?

[turtleguy1134]

Well, discoveries are hard to make. GoL has been around for decades, all the good stuff you can get with a laptop and limited programming abilities has already been taken.

[flipper77]

GoL has been around for decades, all the good stuff you can get with a laptop and limited programming abilities has already been taken.

Well, I wouldn't necessarily say that. I used to think I'd never find anything that would be even remotely significant for years until I decided to run symmetrical soups in apgsearch, in which only a few people had explored beforehand, and when you usually combine your curiosity to explore with quite a lot of persistence and knowledge, you'll eventually find something, like how I found the pufferfish. I found that because I really wanted to explore patterns with some type of symmetries, since symmetrical patterns typically make certain objects of interest much more likely. A good place to start if you want a chance to discover something is try something very few other people have tried themselves, and it most likely will require you to do something you've never done before(before the pufferfish, I knew almost nothing about Python, and now I know enough to write code to do what I want mostly), so if that's the case, you should go for it.

I've only made 1 somewhat significant discovery so far, so perhaps other people who've made more discoveries can add their own advice.

[Extrementhusiast]

I would say the main reason why I am where I am right now is by thinking about things in a different way from everybody else. Earlier on (which might be more relevant), I went out of my way to make things that hadn't been made yet, like the stable MWSS heisenburp.

Here's something to do: make something that generates a number sequence (the OEIS is a good place to find sequences).

[Kazyan]

Life is all one big curiosity, so what seems to work is exploring things that seem like they should be doable with current research, but haven't been looked at yet because they aren't obviously useful to e.g. Herschel circuitry. apgsearch opened up ways to find "natural" syntheses, so that's what several of us did shortly after calcyman completed it.

I haven't done anything major, so you might want to take others' advice instead, but...just look around the LifeWiki and ask questions.

One thing I've been wondering that you could look into: is there a Twin Bees Shuttle stabilization out there that emits gliders without needing a second Twin Bees? The shuttle makes a really big spark, after all, and we have Bellman now. If the answer is "yes", p46 tech could become significantly more compact.

[dvgrn]

This may be a little tangential here, but I am pretty much a loser when it comes to discoveries. I have no clear reason why, but I've been around for some years now and still have nothing to my name. I think it's because I have little computing power to work with (I use a laptop) and I can't distinguish patterns clearly.

Oddly enough, I think the vast majority of Life discoveries have been made with no more than a modern laptop's worth of computing power. (I can admit to thinking about getting apgsearch running on a high-performance cluster this semester, but I haven't actually done it yet...!)

Not so sure about the "can't distinguish patterns clearly" part. Do you mean things like knowing a B-to-Herschel sequence from a LOM on first glance? Lots of people are in the same boat there. It's much easier to develop a specialty or two and avoid worrying too much about other people's areas of expertise. And developing this kind of expertise definitely takes time on the order of years -- as well as a certain determined lack of perspective about the ultimate relevance of the whole line of inquiry. Not everyone can manage to sustain this lack of perspective for very long, and that’s probably a good thing.

If it's any encouragement, my own first interesting Conway's Life discovery was a lucky accident followed by about two weeks of obsessive late nights, running searches on what would now be considered very slow computers (this was 2001). I didn't know enough about C get Paul Callahan's 'ptbsearch' to get it to compile right away, and was puzzled about how to run it anyway. So I used Gabriel Nivasch's 'catalyst' instead, and it happened to be enough to find a "small" 180-degree reflector.

After that I spent several years, off and on, looking for a 90-degree reflector. But my beginner's luck had run out, and it seemed as if everyone else knew a lot more than I did and that I was never going to catch up. I still haven't caught up entirely, but I remember it took over a year before it even seemed like I could contribute anything to a discussion. That's improved over time, at least... or possibly I just talk too much these days.

What is wrong with me exactly and have I been of any use?

I doubt there's anything wrong in particular -- you may just not have found an irresistible research topic yet. You might never find one, but not to worry: as I hinted above, productivity on the Conway’s Life realm doesn’t correlate with anything useful in Real Life, anyway. You’ve contributed to quite a few forum discussions in the past few years, and it’s generally a good thing to have people around who can see the difference between interesting new results and enthusiastic-newbie contributions.

(And you originated the "Synthesizing Oscillators" thread, which triggered a fair fraction of recent new discoveries -- heh, that should be enough fame-by-association to last you a few more years, anyway…!)

I found a nice quote from Bill Gosper in Alex Bellos' recent book, Alex Through the Looking Glass (or The Grapes of Math in the U.S.):

"People who try [to make patterns] quickly realize how difficult it is. You have to be almost insanely manic to concentrate hard enough."

followed quickly by

"Life is an inexhaustible supply of questions and problems."

New Life discoveries seem to be made for the most part by people who are temporarily obsessed with a problem and can't think about anything else. Even that doesn't always work, but it certainly improves the odds.

GoL has been around for decades, all the good stuff you can get with a laptop and limited programming abilities has already been taken.

This is a very plausible explanation, and in some limited areas maybe it's true -- nobody is likely to find any new 18-cell still lifes in B3/S23, or new 9-cell methuselahs, or what have you. But the funny thing is that people have been saying that since at least 1971... and nobody found any new Conway's Life spaceships until 1989, and the first odd-period glider gun had to wait until 1995 (!?!)

Each new discovery has opened up more possibilities than it has closed off, by and large. There are certainly lots of unexplored search spaces in Conway's Life that are still within easy reach of a laptop. I suppose the recent pufferfish fuse discovery is a good example of a newly available and accessible search space.

Life is all one big curiosity, so what seems to work is exploring things that seem like they should be doable with current research, but haven't been looked at yet...

I'm afraid most of the examples I can think of immediately are in my own little ridiculously specialized patch of expertise, though. Here's a random sample. You know how universal constructors can use slow salvos of gliders to construct copies of themselves, or other things. Chris Cain came up with a large number of LWSS recipes for the 10hd diagonal Geminoid universal constructor, so it would now be fairly efficient to run a U.C. that shoots slow LWSSes instead of slow gliders. In some situations it could be very useful to do an incremental construction orthogonally instead of diagonally.

Oddly enough, I don't think anyone has ever even started to look into whether slow LWSSes colliding with a block can reliably construct any object. Another question is whether slow LWSSes can reliably be used to destroy existing stable objects. Maybe *WSS collisions are more explosive than gliders, or something like that -- can't just assume that they'll work the same way.

If that doesn't strike your fancy, just keep looking around for a problem that presents what Bill Gosper calls an “insurmountable opportunity” -- a puzzle that you can’t help but try to solve. Something will show up eventually...!

[simsim314]

GOL is for long time already not a place for discovery geniuses.

First place to start is to learn coding for golly. Learn the existing utilities the community uses, like gencols, Hersrch, LifeAPI to name a few.

Then many times you don't need to discover anything - you just need to use the tools, be patient, do a lot of searches, and generally work hard and step by step. The recent waterbear is great example of "discovery". There was nothing really to discover, only work hard.

I would bring an example from my search: for long time there were no good p1 *WSS edge shooting examples for slow salvo construction. So I made a simple golly brute force search, that checked all 3 SL patterns, and checked whether there is an edge shooter. The script took me half hour to write. The script was running for two weeks, which is much less than writing some more complex and smarter utility, but gave me what I needed.

It's less about "discovering" something, and more about using the existing tools wisely, together with hard work and open mind. Experience is also important of course.

[knightlife]

It is still possible to make discoveries by being extremely observant and being able to recognize patterns that can be ignored because they are commonplace. Endless patience and trial and error are necessary if searching manually, which is what I have preferred to do so far, just because it's more fun than programming for me. I like to observe patterns evolve and can avoid futile searches just by being observant. However, the discoveries are few in the areas that others have already investigated, as already mentioned. Of course, programs can be "observant" too and do all the tedious searching and is really the way to go now in order to find things faster. I am starting to do more programming myself.

[simsim314]

I think the best observant "by hand" GOL explorer today is Extrementhusiast. And he is still uses search utilities, and have a lot of experience, which many of us don't.

Still there is limitation of things you can do by hand - and there are many things can be done by simple patience.

You want "discover" GOL patterns? You can start from reducing the size of many glider guns known today. dvgrn suggested 3-4 times a lot of ideas how to reduce the size for many of the known guns, and offered his advice on the topic. No one "jumped" on the opportunity to discover new smaller glider guns - I'm not sure what you look and hope to achieve talking about discoveries, but a little bit of work could give you a lot of very nice "middle range" discoveries. A bit more hard work, and you can do something like quadratic replicator, adjustable speed spaceship, or C compiler (from C to GOL).

I would suggest to start from things known to be possible to do, and just make them. Recently chric_c (and me) built HBK gun. This was known to be possible, there was nothing "miraculous" about the HBK gun, and it took a bit less than a week to build.

So I'm not sure what you think "discoveries" are, but a little bit of patience, learn how to code and use existing search utilities is the certain way to go, for new discoveries. And if you didn't discover any, just start from simple things, and move from there.

[knightlife]

There is the accidental discovery and then there is the engineering feat. Either way you may need a little luck!

[simsim314]

Either way you may need a little luck

Or you might be smart to have enough "degrees of freedom" and have statistics on your side.

I find myself many times doing statistical estimations - it's not exactly luck, of course luck can definitely help from time to time. But if all you do is trust on luck, you won't get far.

Let's check the "greatest" discoveries in GOL:

Gosper gun - shuttles were place one near the other. It's statistics.
Non trivial spaceships - special search utility.
Gemini - hard work and design.
Stable reflector - special search utility.
Herschel Conduits - special search utility.

Of course they all have their share of luck, intuition and a lot of GOL experience. Even the puffer-fish for example, couldn't be found without cyclamen soup search utility. It's all comes to tweaking and playing with GOL patterns, using maximum of the utilities available.

[simsim314]

GoL has been around for decades, all the good stuff you can get with a laptop and limited programming abilities has already been taken.

I've recently discovered 3 gliders infinite growth. Don't you think 3 glider collision is something that supposed to be discover 30 years ago? Well I can't find it anywhere... probably it's actually my discovery, in 2014.

I also discovered around dozen more 10 cells infinite growth patterns that weren't known, with laptop and primitive golly scripting.

It's maybe a little bit disappointing, but not too many people are interested in GOL, and if they do, not always they search everything in the best way possible. They sometimes miss stuff, as being the first they don't care to look for all the tiny detail, and all the edge cases.

Also my guess is that we're missing an order of magnitude in Herschel conduits, and some trivial glider to Herschel converter, that will reduce all known circuitry an order of magnitude. This could be found on laptop with basic programming skills, but we need to be smarter in our searches.

You know forget me, half bakery glider reaction - don't you think someone in the history of GOL should have found it in 40 years? How complex is it to collide glider into still life? But the fact is only bellman found it, and no one tried to collide gliders into less common SLs in all 40 years.

[codeholic]

chric_c

cyclamen

You're a master of misspelled nicknames!

You know forget me, half bakery glider reaction - don't you think someone in the history of GOL should have found it in 40 years? How complex is it to collide glider into still life? But the fact is only bellman found it

That's not true. The half-bakery reaction with glider has been known since ages. I found that two of these reactions can produce an extra glider in 2014, using nothing more than gencols.

[Kazyan]

Also my guess is that we're missing an order of magnitude in Herschel conduits, and some trivial glider to Herschel converter, that will reduce all known circuitry an order of magnitude. This could be found on laptop with basic programming skills, but we need to be smarter in our searches.

Herschel conduits could be expanded heavily with Bellman's new realm of catalysts, but I think the most glaring omission is a small H-to*WSS terminator. We have over a dozen H-to-G terminators--why not search for a LWSS? It would be kind of a big deal for circuitry.

I've been fiddling with a manual search to find something that could be hacked into that--filling a random 20k x 5 rectangle, zooming out, letting it run until I can clearly see the c/2 dots flying out orthogonally, and then investigating. Usually they're nothing interesting, but sometimes I'll see an interesting reaction. (I don't know if there's a good way to automate this.)

Code: Select all

x = 13, y = 15, rule = B3/S23
10bo$2b2o4b2ob2o$bobo7b2o$3o2$8bobo$11bo$7b2o2bo$5bobo2bo$4b2o2bo$3b2o
2b2o$3bob3o$3bo3bo$4bobo$5bo!

I wonder if Bellman can complete it with something along these lines:

Code: Select all

x = 12, y = 15, rule = B3/S23
5b2o3b2o$5bobobobo$7bobo$6b2ob2o$2b2o$bobo$3o6bo$8bobo$9bo3$3b2o$3bo$
4b3o$6bo!

[simsim314]

Don't forget we now have LifeAPI. I was experimenting with CUDA as well. Imagine 3SL 10x10 Search within a minute.

You could find all the catalysts + emitting stuff just with pure brute force, in reasonable time frame.

Making it a bit smarter - could bring us 4-5 catalysts.

The options are endless with new search tools, and new ideas on how to use them. New hardware is also available now. GPU acceleration, could benefit GOL community as well as distributed searches.

I'm certain that Bellman search in lines of Snark can bring us very fast G->H as well. Unfortunately this won't work for replicator circuitry that need simple components, but for many things it could work pretty well.

EDIT BTW does anyone has compiled bellman for windows?

[simsim314]

You're a master of misspelled nicknames!

Woops well, you know I have problem with typos...

That's not true. The half-bakery reaction with glider has been known since ages. I found that two of these reactions can produce an extra glider in 2014, using nothing more than gencols.

Didn't know that... Anyway this was luck but... you still used search utility. And you do it over and over again, I see you use gencols and python scripts a lot, and find some interesting stuff.

I think it's combination of experience, basic GOL curiosity, a lot of searches and creative usage of available tools, and a bit of luck. And of course this is even before we start to talk about design, which is simply basic hard and systematic work.

Bellman is a tool from 2013 which wasn't used enough to search stuff. So I would say it's still has a lot of potential, and you need only laptop, not even coding abilities. Did anyone tried to use Bellman to search for H conduits for example?

[Scorbie]

EDIT BTW does anyone has compiled bellman for windows?

I did manage to compile it with MinGW, but I couldn't find the gd library made into windows so I removed the code involving gd.
(I use cygwin for the complete version)

EDIT: removed ffsll() function definition. The user would have to define it oneself...
EDIT^2: Thanks to simsim314, replaced ffsll() with __builtin_ffsll().

[Sphenocorona]

I've compiled Bellman on windows, but I used the utility Cygwin to do so, so I'm not entirely sure it would run from the built-in windows command line.

[Scorbie]

oops. I used the ffsll() source from linux which says that it's part of the GNU C Library. (On universe.h, line 159)
If that is against copyright problems, then I think I should remove that attatchment. Could someone tell me about copyright issues here?

EDIT: And it says that it's for AMD x86-64 so you might have to change that part that's applyable to your computer...

[simsim314]

I used the ffsll() source from linux which says that it's part of the GNU C Library

In windows gcc you have __builtin_ffsll() instead. Comes with any gcc compiler.

I did manage to compile it with MinGW

Thx a lot! It worked for me with gcc.

EDIT I also needed to add #include "universe.h" in bitwise.h, and for some reason compile twice - probably some issue on my machine. Anyway it worked.
----

Off topic We must have better support for community tools. Compilation should be straightforward, and precompiled x86 and x64 should be available for all major platforms.

If people are willing to join this support project, I can take the windows versions on myself.

We need the following tools to be available:

1. gencols.
2. bellman.
3. lifesrc (or gfind and other tools similar tools).
4. hersrch.
5. catalyst and catgl.

Please add more tools you think need to be in development and support.

We need good introduction documentation, and versions for all platforms with compilation instructions. It would be also nice to add UI for the tools (multi platforms), as the current command line approach is kinda scary.

We also must have people that can tweak the source code, add (at least simple) functionality, and solve basic bugs.

If someone wants to benefit GOL community, this kind of support would be a great step. GOL is not only about discoveries.

[codeholic]

How about making the portal with GitHub pages? We could make a site for conwaylife organization. I think it is more reliable than paid hostings (primarily because they tend to be shut down when a responsible person does not pay )

[simsim314]

@codeholic I'm not sure how your comment is related.

I was just saying something is missing in the simplicity and user friendliness of the existing search tools.

Three major things should be improved:

1. The tools should be supported. Bugs and requests should be reported, improved, and fixed.
2. Distribution versions for major OS should be simply available.
3. UI and preferably golly interface should be added.

And finally documentation and nice tutorials should be added. Most of the tools do have some starting point, but still I would like to see more tutorials, more cases explored and explained and more discussions, on what is missing and how to tweak the existing tools.

We have it all more or less, and I know you try to do your best, but I think the community is very focused on discoveries, and creating new things instead of support, documentation and user friendliness.

EDIT Woops I got what you were saying. Yes it's great idea!

[Scorbie]

@simsim314
Really nice idea! I was struggling with the user-unfriendly UI, still learning to use a program efficiently, (and gave up learning some search tools, like ofind.) and struggled with compilation thru windows. (Although I did compile what I need, in the end.)

Well, they're not that popular, but two of my favorite programs are dr2 and randomagar. Would it be okay if they're on the list?

EDIT: and many thanks to Sokwe for sharing the program!

[simsim314]

Every search tool is welcomed.

As long as there are people willing to share their experience, improve, support, and generally add user friendliness - I think the tool should be part of the project (that actually didn't started yet).

codeholic is doing pretty good job, but the "general" user friendliness of the utilities we have today is simply "awful". The only real exception is golly. And golly has all mentioned above (nice UI, documentation, rapid development, binaries for all major OS including android, nice collections) and more (like introduction tutorial into golly source code).

[Sokwe]

I did manage to compile [Bellman] with MinGW, but I couldn't find the gd library made into windows so I removed the code involving gd.

I managed to compile Bellman for Windows, which I posted here.

I think it would be a good idea to organize all of the publicly available search utilities, as was suggested by several people above. We should maybe move that discussion to Codeholic's Organization on GitHub topic.