The current busy beaver candidate on 6 states, with the original binary alphabet configuration, produces about 10^18267 1's, according to the wiki page on Busy Beaver. I could not find any working links to research that further. My question is, what methods were applied to emulate such a long running Turing machine? I am also interested in general methods that could produce that kind of a result, even if they were not applied in this case.
Asked
Active
Viewed 732 times
5
-
1See whether http://goanna.cs.rmit.edu.au/~jah/busybeaver/seminarMay05.html gets you anywhere --- it does seem to discuss search strategies for good candidates. – Gerry Myerson Mar 23 '15 at 10:43
1 Answers
5
Heiner Marxen provides a description of several techniques used to simulate extremely long-running Turing machines: http://www.drb.insel.de/~heiner/BB/mabu90.html
First some notation: A TM configuration can be represented like: 10 B> 111101
. Which means that it is in state B
and the head is over the 1
at position 3 from the left (the symbol that the >
points to). If this TM has a transition (B, 1) -> (0, R, B)
then the next configuration will be 100 B> 11101
.
There are specifically 3 techniques:
- Tape compression: Represent a long tape as
1 0 B> 1^18 0 1^2
instead of10 B> 111111111111111111011
. In addition to potential space savings, this can lead to time savings when you have transitions like(B, 1) -> (0, R, B)
which can jump over the entire stack of symbols:1 0 B> 1^18 0 1^2
->1 0^19 B> 0 1^2
. - Macro machine: Instead of considering the TMs actions on each individual symbol, think of it as acting on blocks of N symbols. This allows us to compress the tape like:
11 10 A> 01^6 11
. Many TMs which do not compress well with (1) will with some Macro machine for some block size N. - Proof system: Finally, by automatically detecting, proving and applying generic rules to specific configurations, you can get significant time improvements for some machines (including all record-holding TMs). An example rule might be:
100^a B> 010^(b+1)
->100^(a+2b) B> 010^1
for anya,b >= 0
inb^2 - b + 4
steps.

sligocki
- 249