Re: LIFE on a finite grid

From: David Hillman (dhist1+@pitt.edu)
Date: Fri May 21 1993 - 22:07:54 UTC


In article <13594@tiglio.rl.ac.uk> arno@inf.rl.ac.uk (Arno Siebes) writes:

>The second question, which is related to the first, what is
>known about a Life-like system, with a finite grid as above, but
>in which the topology can be changed (that is, cells can acquire
>new neighbours and discard other)?

Can't help with your first question (universal computation).  As for
changing "topology": some work has been done in this area.  First, note
that a change in grid structure need not be seen as a change in topology.
One may, for example, triangulate a sphere in many different ways; all may
be seen as discrete models of a sphere.  Imagine an initial triangulation
of a sphere in which, say, each vertex is colored black or white, and a
local law in which the triangulation changes.  The grid structure changes
but the topology (sphere) remains the same.  To distinguish the two cases,
call the change in grid structure a change in geometry.  Then, yes, laws exist
in which the geometry changes over time; and laws also surely exist in
which the topology changes over time (though I have no examples of one).

Now much work has been done concerning ONE-dimensional cellular-automaton-
like systems in which the geometry changes.  These are L-systems.  In one
dimension there is not much way in which the geometry can change.  One has
a state of the universe consisting of a loop of cells.  The geometry changes
by changing the number of cells in the loop.  It is easy to construct a
local law which does this.  L-systems typically work with strings instead
of loops.  Most work on L-systems concerns laws with neighborhood size one.
These are non-interactive systems: a single cell at time t produces zero
or more cells; the state at time t+1 is obtained by stringing these cells
together in the obvious way.  But there has also been work done in which the
neighborhood size is bigger than one; these can have interactions and are
thus more like your usual cellular automata.

Extensions have been worked on also in higher dimensions: L-systems have
been generalized in several ways to laws on arbitrary graphs, or to laws
on specific sorts of 2-dimensional graphs.  As far as I know, these
extensions all involve laws which are NOT interactive (I might easily not
know, however).  Historically, the primary motivations for studying
L-systems seem to be linguistic and biological: thus, e.g., a lot of
emphasis on growth.  One can easily have growth without having any
interactions between the pieces of the universe.  And it's easier not to
have them.  Hence, as they stand I assume that these higher-dimensional
models do not behaviorally resemble typical higher-dimensional CAs. 

Many books have been written about L-systems, by (or edited by) Lindenmeyer
(who invented them in the 1960s), Rozenberg, Salomaa and probably others

Note one difficulty in higher dimensions: typically one wishes to have a
finite set of allowed neighborhoods; one must construct the law in such a
way that it guarantees that any allowed state at time t leads to an allowed
state at time t+1 (one in which all neighborhoods are in this finite set).

In the CA literature the only thread I know of addressing the question of
changing geometries begins with an article called "Structurally Dynamic
Cellular Automata" by Andrew Ilachinski and Paul Halpern in Complex
Systems 1 (1987) 503-527.  (There are several follow-up papers written or
co-written by Halpern.)  They discuss the problem in general and then offer
an example in which there is a 2-D grid, the vertices are fixed over time
but the edges can move around (thus the neighborhoods of the vertices
change).

Hope this helps.

David Hillman
dhist1+@pitt.edu


This archive was generated by hypermail 2.1.7 : Tue Oct 14 2003 - 21:44:14 UTC