simsim314 wrote:Every slow salvo recipe will just work 5 times faster, and that means that Silver's cost should be count as SL count (or more precisely slow salvo glider synth count) X 400 ticks (glider/tick). And syringe cost should be count as #SLX70.
That means if we take replication time as the final goal, syringe with eater2 will replicate 5 times faster, i.e. allowing much "uglier" solutions and larger SL counts for any particular problem.
Yes, I keep going back and forth about this. If there was something like a million-dollar prize for the replicator that completed a replication cycle in the smallest number of ticks, then this all makes complete sense.
The winning design might be a fairly complicated multi-channel device with multiple construction arms running in parallel. That way you could build not just with slow 90-degree glider pairs, but with maybe six or eight gliders colliding at a time. Maybe there would be dedicated programmable shotguns that could build a syringe, let's say, or some complex piece of general-purpose logic circuitry, at very high speed.
An idea that Paul Chapman came up with a decade ago was to build an entire replicator out of a single standardized part, as much as possible -- imagine some arrangement of different orientations of glider reflectors, so that you could get a left or right or 180-degree reflection, or various kinds of logic gates, just by sending gliders in at different locations. Then the construction instructions would be relatively tiny -- just a list of all the different locations to place these Big Logic Gate Things. Could probably store that list in a static tape instead of a circulating loop.
-- It seems like a nice idea until you start to try to implement it. One of the problems is that you still have to store the recipe for a Big Logic Gate Thing somewhere somehow, probably in a circulating glider loop.
Anyway, that design is aimed more at simplifying and compartmentalizing the construction data, rather than minimizing the total number of ticks needed to replicate. After I thought about it for a while, neither of those design goals seemed like the most useful one.
The problem with a cost metric like (number-of-still-lifes x ticks-per-glider) is twofold:
1) If a recipe is five times as complex, then (as a rough rule of thumb) the pattern will actually run in Golly at about a fifth of the speed. Tighter-packed patterns tend to fill up memory with more different hashtiles, so if there are more gliders in the recipe loop, Golly will have to do more work. Just as a practical consideration with the simulation software that we have at the moment, the signal-packing gains are roughly balanced out by the slower simulation speed.
2) These theoretical tightly-packed recipes with five times as many gliders seem to be at least five times harder to design and troubleshoot, in practice, than the simple Spartan U.C. designs. We have the Gemini, the linear propagator, the spiral-growth pattern, and some progress toward a Demonoid spaceship and a quadratic-growth replicator. Nobody has actually managed to design any useful self-constructing circuitry containing eater2s or other similar large still lifes, just because it's a significantly harder problem.
For example, it's a lot of work to find the minimal recipe for eight different orientations of a boojum reflector... and even more work to find an actual use for a boojum reflector in self-constructing circuitry, especially since it doesn't even allow a factor-of-three improvement in compression over Silver reflectors. The final payoff for adding boojum-reflector construction capability might be a reduction in replication time of just 20%, or 10%, or 1% or no improvement at all.
A Spartan Snark, or a really nice slow-salvo or glider-pair construction of the current Snark, or even better a Spartan G-to-H, would
change the cost calculation really significantly. Until then, it seems like it would take an order of magnitude more research time to put together a design that would be many times more complex (in terms of number of gliders in the recipe)... and it will only end up running at the same speed or slower in Golly anyway, never mind what the generation count says.
I may be overly biased by my own particular experience, of course! Of the three linear propagator patterns
, the slowest version in terms of number of ticks to replicate is the one that runs fastest in Golly, by far.