home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.theory.cell-automata
- Path: sparky!uunet!snorkelwacker.mit.edu!bloom-beacon!INTERNET!dont-send-mail-to-path-lines
- From: 100020.2727@CompuServe.COM (Andrew Wuensche)
- Subject: basins of attraction
- Message-ID: <921119191627_100020.2727_EHF46-1@CompuServe.COM>
- Sender: root@athena.mit.edu (Wizard A. Root)
- Organization: The Internet
- Distribution: inet
- Date: Thu, 19 Nov 1992 19:16:28 GMT
- Lines: 133
-
- re. message from Harold V. McIntosh;
-
- >Well, I have ordered a copy of the book; delivery has been promised
- >for January or February. In the meantime, it is only possible to
- >conjecture the contents of the book...
-
- Sorry you have to wait so long. I hope others have found it easier to
- get hold of my book !? It came out in the USA in early Oct.
- At the risk of an overlong message, I will try to reply to your
- comments, and also indicate the basis of the reverse algorithm.
-
- > ...... and the relationship of the procedures to those
- >of E. F. Moore, G. A. Hedlund, M. Nasu, Erica Jen, or even myself.
-
- > 1) Unless the counterimages are balanced, there must be a
- >Garden-of-Eden. Presumably this coincides with your 'limited
- >pre-image rules' remark above, and immediately foredooms most automata
- >to have Gardens of Eden.
-
- > 2) In my early calculations of the variance in the number of
- >ancestors, there appeared to be a gap between the rules of
- >zero-variance and the others? Have you observed such an effect? Is it
- >real (I later decided that it probably wasn't)?
-
- The reverse algorithm for generating pre-images described in my book
- [1] relates to Erica Jen's notion of deterministic structures [2].
- Entries in the rule table, say for 3-neighbour (Wolfram's elementary)
- rules are grouped into pairs of neighbourhoods that differ only by
- their rightmost value. If a pair has different outputs, ie
-
- a1,a2,0 --> T and a1,a2,1 --> not T
-
- then given the values of a1,a2 and T, a3 is determined. If all pairs in
- the rule table have "left deterministic permutations" as above then
- the set of pre-images of any state is easily found by assuming the
- start of a candidate pre-image as a1,a2...... Successive values are
- determined to the right. As there are 4 assumed starts, the maximum
- number of pre-images is 4 for any array size (in general, 2^(n-1) where
- n is neighbourhood size). A candidate pre-image can be disqualified if
- it fails a periodic boundary condition check.
-
- A pair having the same output is an "ambiguous permutations" ie.
-
- a1,a2,0 --> T and a1,a2,1 --> T
-
- and implies the "excluded permutations"
-
- [a1,a2,0 --> not T] and [a1,a2,1 --> not T]
-
- In this case given a1,a2 and T, a3 = 0 or 1 with equal validity
- (potentially increacing the number of pre-images); the occurrence of
- a1,a2 and not T disqualify the candidate pre-image, potentialy reducing
- the number of pre-images.
-
- This whole argument applies equaly in the opposite direction, from the
- right. Rule tables consisting entirely of deterministic permutations,
- either left or right, are "one-way limited pre-image rules". If BOTH
- left and right deterministic permutations fill the rule table (two-way
- limited pre-image rules, for example n=3 rules 90, 105, 60, n=5
- totalistic code 21, the "additive rules") then the pre-imaging (or in
- degree) of every state turns out to be invariably equal when the basin
- of attraction field is drawn; I assume this what you mean by
- "zero-varience". There is proof in Martin et al.[3].
- One-way limited pre-image rules (either left or right, not both)
- must have pre-imaging less than the theoretical maximum of 2^(n-1), for
- example n=3 rules 30 and 45 (the proof is given in my the book). These
- rules result in many states having only one pre-image ("balanced"?),
- and thus their basins of attraction tend to have very long transients
- and attractor cycles, corresponding to "chaotic" behaviour.
- Most rule tables have a mixture of deterministic and
- (ambiguous/excluded) permutations both left and right, which set the
- rule's Z parameter. The Z parameter predicts typical basin topology,
- and the density of G of E states in state space.
-
- >It is well established that for one dimensional cellular automata,
- >Gardens of Eden are absolutely computable; difficulties lie in two
- >dimensions and beyond.
-
- Can you give a reference ?
-
- >In fact, there is no need to restrict to periodic boundary
- >conditions; the procedures serve as well for quiescence at infinity or
- >for boundaryless (that is, unrestrictedly infinite) automata.
-
- Periodic boundary conditions are not required to decide if a state
- is a G of E by my reverse algorithm. The occurrence of an excluded
- permutation in a candidate pre-image segment disqualifies it. If all
- 2^(n-1) assumed starts to candidate pre-images of a given state lead to
- disqualification, the state is a G of E. But given a CA with periodic
- boundary conditions, a candidate pre-image may be disqualified for the
- only reason that it does not comply with periodic boundary
- conditions.
-
- By the way, in more recent work I have developed a GENERAL reverse
- algorithm that computes the pre-images (if any), and basins of
- attraction, of Stuart Kauffman's "random Boolean networks", which may
- have any arbitrary wiring and/or rule scheme for each cell. I argue
- that the basins of attraction (which may be "sculpted" by a learning
- algorithm) represent the networks dynamic memory[4,5].
- The wiring may be set to conform to 1,2 or 3-D finite CA
- architecture, allowing the pre-images as well as G of E states of
- higher dimensional CA to be computed without exhaustive testing.
-
- Andy Wuensche contact address:
- Santa Fe Institute and 48 Esmond Road, London W4 1JQ
- University of Sussex tel 081 995 8893, fax 081 742 2178
- wuensch@santafe.edu email 100020.2727@compuserve.com
-
-
- [1] "THE GLOBAL DYNAMICS OF CELLULAR AUTOMATA, An Atlas of Basin of
- Attraction Fields of One-Dimensional Cellular Automata"
- Andrew Wuensche and Mike Lesser, (Diskette included)
- foreword by Christopher Langton.
- Santa Fe Institute Studies in the Sciences of Complexity, Reference Vol
- 1, Addison-Wesley, Reading MA, phone:(800) 447 2226, IBSN 0-201-55740-1
- 1992 Hardcover, 250 pages, $53.75
-
- [2] Jen,E., "Global properties of Cellular Automata", J. of Stat.
- Phys. Vol.43. Nos 1/2. April 1986.
-
- [3]. Martin.O., A.M.Odlysko, and S.Wolfram. "Algebraic Prorerties of
- Cellular Automata". Comm.Maths.Phys.93(1984),219-258.
-
- [4] Wuensche,A., "The Ghost in the Machine; Basin of Attraction Fields
- of Disordered Cellular Automata Networks", Santa Fe Institute Working
- Paper 92-04-017, 1992. Submited to the proceedings of Alife III.
-
- [5]. Wuensche,A., "Basins of Attraction in Disordered Networks", In:
- Artificial Neural Networks 2, Ed. I.Alexander and J.Taylor, Elsevier,
- Amsterdam, 1325-1344, 1992.
-
-
-
-