home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.theory.cell-automata
- Path: sparky!uunet!secapl!Cookie!frank
- From: frank@Cookie.secapl.com (Frank Adams)
- Subject: Re: Undecidability of configurations/progression
- Message-ID: <1992Nov11.173807.100765@Cookie.secapl.com>
- Date: Wed, 11 Nov 1992 17:38:07 GMT
- References: <1992Nov8.210220.29731@hellgate.utah.edu> <POLLACK.92Nov9133140@dendrite.cis.ohio-state.edu>
- Organization: Security APL, Inc.
- Lines: 59
-
- In article <POLLACK.92Nov9133140@dendrite.cis.ohio-state.edu> pollack@cis.ohio-state.edu writes:
- >>2) Whether a configuration is garden of eden (as demonstrated by previous
- >> poster)
- >
- >there are 3 different problems:
- >
- >1) Given a SPECIFIC finite configuration, we can run a deterministic
- >limited-neighborhood CA BACKWARDS one step. Generally, one would get
- >lots of previous states, but a GoE would have 0 previous states.
- >The size of the previous states would expand at the speed of light.
-
- This depends on exactly how you define a GoE pattern. If I have a pattern
- which fits in an n by m rectangle, there are two questions I can ask:
-
- (1) is there any previous pattern whose next step, restricted to that
- rectangle, is the specified pattern?
-
- (2) is there any previous pattern whose next step is the specified pattern,
- with everything outside the rectangle being blank?
-
- Question (1) can be answered as described; however I believe question (2) is
- the standard definition of a GoE. And question (2) is, in fact,
- undecidable.
-
- (*Very* brief outline of proof: one can make a CA for a given Turing machine
- such that a cell disappears iff it is (locally) part of a computation of the
- TM (with successive times in adjacent rows). A row will be GoE iff the
- computation starting with it does not halt. Hence the GoE problem is
- reducible to the halting problem. (Note that if you allow infinite
- patterns, you can simply refuse to erase the halting state, inverting the
- result: a pattern is GoE iff the computation starting with it *does* halt.))
-
- >2) Finding a GoE which gives rise to a PARTICULAR configuration is a
- >much harder problem, as it would require searching through the space
- >generated by operator 1. Unless the question were asked: "is this
- >configuration less than K generations from a GoE", it is undecideable.
-
- It is undecideable anyhow.
-
- >3) finally, whether or not any GoE's EXIST for a particular CA rule
- >might be even harder.
-
- However, if I remember the Scientific American article where I first
- encountered GoE patterns, this problem is easy: a CA has a GoE pattern iff
- it is not invertible. An invertible CA A is one where there is an inverse
- CA B such that if A takes pattern X to pattern Y, B takes Y to X, and vice
- versa.
-
- >A more interesting question might involve whether the current state in
- >the smallest bounding box, could have arisen from a "smaller" state.
- >
- >Then one could invert CA's to build data compression systems:
- >
- > Start with 1011 and run rule 247 for 16325 generations,
- > then run rule 343 for 932 generations, and ta da: the Mona Lisa!
-
- You can't compress the data to less than it's information-theoretic content.
- A "data-compression" scheme which doesn't attack the redundencies in the
- original message is not likely to accomplish anything.
-