Escolar Documentos
Profissional Documentos
Cultura Documentos
North-Holland
C O M P U T A T I O N AT T H E E D G E OF CHAOS:
P H A S E T R A N S I T I O N S AND E M E R G E N T C O M P U T A T I O N
Chris G. L A N G T O N
Complex Systems Group, Theoretical Division, Los Alamos National Laboratory, Los Alamos, N M 87454, USA
In order for computation to emerge spontaneously and become an important factor in the dynamics of a system, the
material substrate must support the primitive functions required for computation: the transmission, storage, and modification
of information. Under what conditions might we expect physical systems to support such computational primitives?
This paper presents research on cellular automata which suggests that the optimal conditions for the support of information
transmission, storage, and modification, are achieved in the vicinity of a phase transition. We observe surprising similarities
between the behaviors of computations and systems near phase transitions, finding analogs of computational complexity
classes and the halting problem within the phenomenology of phase transitions.
We conclude that there is a fundamental connection between computation and phase transitions, especially second-order or
"critical" transitions, and discuss some of the implications for our understanding of nature if such a connection is borne out.
a for each automaton consists of the set of possi- scheme would partition the space of CA rules in
ble neighborhood states: a = Z s. Letting K = I~;I such a manner that rules from the same partition
(the number of cell states) the size of a is equal to would support similar dynamics. Such an ordering
the number of possible neighborhood states on ~Nr would allow us to observe the way in
which the dynamical behaviors of CAs vary from
l a l = IAI = I ~ S l = K s . partition to partition.
T o define a transition function A, one must The location in this space of the partitions
associate a unique next state in ~ with each supporting the transmission, modification, and
possible neighborhood state. Since there are K = storage of information, relative to the location of
12:1 choices of state to assign as the next state for partitions supporting other possible dynamical be-
each of the IZSl possible neighborhood states, haviors should provide us with insight into the
there are K (rN) possible transition functions A conditions under which we should expect compu-
that can be defined. We use the notation ~ f f to tation to emerge in CAs.
refer to the set of all possible transition functions
2.1. The ~ parameter
A which can be defined using N neighbors and K
states.
We will consider only a subspace of ~ , char-
acterized by the parameter h [18, 17].
1.5. Example The ?~ parameter is defined as follows. We pick
an arbitrary state s ~ ~, and call it the quiescent
Consider a two-dimensional cellular automaton state Sq. Let there be n transitions to this special
using 8 states per cell, a rectangular lattice, and quiescent state in a transition function A. Let the
the five-cell neighborhood template shown above. remaining K S - n transitions in A be filled by
Here K = 8 and N = 5, so IAI = K s = 85 = 32768 picking randomly and uniformly over the other
and there are thus 32 768 possible neighborhood K - 1 states in Z - Sq. Then
states. For each of these, there is a choice of 8
states as the next cell state under za, so there are KS-n
K txN) = I ~ l = 8 (aS) --103000o possible transition x IC s (1)
functions using the 5-cell neighborhood template
If n = K s, then all of the transitions in the rule
with 8 states per cell, an exceedingly large num-
table will be to the quiescent state Sq and X = 0.0.
ber.
If n = 0, then there will be no transitions to Sq
and ~ = 1.0. When all states are represented
equally in the rule table, then X = 1.0 - 1 / K .
2. Parameterizing the space of CA rules The parameter values X = 0.0 and ~ = 1 . 0 -
1 / K represent the most homogeneous and the
~ Nr, the set of possible transition functions A most heterogeneous rule tables, respectively. The
for a CA of K states and N neighbors, is fixed behavior in which we will be interested is captured
once we have chosen the number of states per cell between these two parameter values. Therefore,
and the neighborhood template. However, there is we experiment primarily with h in this range.
no intrinsic order within ~ ; it is a large, undif-
ferentiated space of CA rules. 2.2. Searching CA space with the ~ parameter
Imposing a structure on this undifferentiated
space of CA rules allows us to define a natural In the following, we use the ?~ parameter as a
ordering on the rules, and provides us with an means of sampling ~ in an ordered manner. We
index into the rule space. The ideal ordering do this by stepping through the range 0.0 < 2~ <
C. Langton / Computation at the edge of chaos 15
On the second view, the initial configuration These fundamental properties must be provided
itself constitutes a computer, and the transition by any dynamical system if it is to support compu-
function is seen as the "physics" obeyed by the tation. Taken together, they require that any dy-
parts of this embedded computer. The algorithm namical system supporting computation must
being run and the data being manipulated are exhibit arbitrarily large correlation lengths in space
functions of the precise state of the initial config- and time. These correlation lengths must be poten-
uration of the embedded computer. In the most tially infinite, but not necessarily so. Codd [5]
general case, the initial configuration will consti- refers to this situation as one in which the propa-
tute a universal computer. gation of information must be unbounded in prin-
We can always take the first point of view, but ciple but boundable in practice.
what we are interested in here is the question:
when is it possible- even necessary- to adopt the 2.6. Wolfram's quafitative CA classes
second point of view to understand the dynamics
of a CA? Wolfram [29] has proposed the following four
That CAs are capable of supporting universal qualitative classes of CA behavior:
computation has been known since their invention
by Ulam and von Neumann in the late 40's. Von
Class I evolves to a homogeneous state.
Neumann's proof of the possibility of machine
Class II evolves to simple separated periodic
self-reproduction involves the demonstration of
structures.
the existence of a universal computer/constructor
Class III yields chaotic aperiodic patterns.
in a 29-state CA [26]. Since then, Codd [5], Smith
Class IV yields complex patterns of localized
[24], Conway and co-workers [2], Fredkin and
structures.
Toffoli [7]- to name but a few- have found much
simpler CA rules supporting universal computa-
tion. Wolfram finds the following analogs for his
All of these proofs involve the embedding of a classes of cellular automaton behaviors in the field
computer within the CA, or at least they show that of dynamical systems.
all of the important parts of such a computer
could be implemented and that those parts are Class I cellular automata evolve to limit points.
sufficient to construct a computer. Some of these Class II cellular automata evolve to limit cycles.
proofs involve the construction of Turing ma- Class III cellular automata evolve to chaotic
chines, others involve the construction of stored- behavior of the kind associated with strange at-
program computers. tractors.
All of these constructs rely on three fundamen- Class IV cellular automata "effectively have very
tal features of the dynamics supported by the long transients ".
underlying transition function physics. First, the
physics must support the storage of information, This association of class IV CAs with "very
which means that the dynamics must preserve long transients" will figure "critically" in what
local state information for arbitrarily long times. follows.
Second, the physics must support the transmission Wolfram suggests that class IV CAs are capable
of information, which means that the dynamics of supporting computation, even universal compu-
must provide for the propagation of information tation, and that it is this capacity that makes
in the form of signals over arbitrarily long dis- their behavior so complex. This paper supports
tances. Third, stored and transmitted information Wolfram's hypothesis, and offers an explanation
must be able to interact with one another, result- for both the existence of these classes and their
ing in a possible modification of one or the other. relationship to one another.
C. Langton / Computation at the edge of chaos 17
In their surveys of 1D and 2D CAs, Packard cannot have X = 0.00 exactly. The primary fea-
and Wolfram [23] hypothesized that class IV CAs tures observed as we vary X throughout its range
constitute a set of measure 0. This means that are itemized below.
class IV behaviors should be infinitely hard to find
in the "thermodynamic limit" of an infinitely large X = 0.00 All dynamical activity dies out after a
CA rule space. However, it turns out that they are single time step, leaving the arrays uni-
not hard to find in rule spaces that are far from form in state Sq. The area of dynamical
the thermodynamic limit. By locating class IV activity has collapsed to zero.
behaviors in these non-limiting rule spaces and X = 0.05 The dynamics reaches the uniform Sq
tracking the manner in which they become vanish- fixed point after approximately 2 time
ingly rare as one goes to larger rule spaces, we can steps.
derive a general theory about where to locate rules X = 0.10 The homogeneous fixed point is reached
likely to support computation in any CA rule after 3 or 4 time steps.
space. X -- 0.15 The homogeneous fixed point is reached
after 4 or 5 time steps.
X = 0.20 The dynamics reaches a periodic struc-
3. Qualitative overview of CA dynamics ture which will persist forever (fig. 1,
X = 0.20). Transients have increased to
In this section, we present a series of examples 7 to 10 time steps as well. Note that the
illustrating the changes observed in the dynamical evolution does not necessarily lead to
behavior of one-dimensional CAs as we alter the periodic dynamics (fig. 2, X = 0.20).
parameter throughout its range using the table- = 0.25 Structures of period 1 appear. Thus,
walk-through method. For these CAs, K = 4, N = there are now three different possible
5 (i.e. two cells on the left and two cells on the outcomes for the ultimate dynamics of
right are included in the neighborhood template). the system, depending on the initial
The arrays consist of 128 sites connected in a state. The dynamics may reach a homo-
circle, resulting in periodic boundary conditions. geneous fixed point consisting entirely
Each array is started from a random initial con- of state Sq, or it may reach a heteroge-
figuration on the top line, and successive lines neous fixed point consisting mostly of
show successive time steps in the evolution. cells in state Sq with a sprinkling of
For each value of X, we show two evolutions. cells stuck in one of the other states, or
The arrays in fig. 1 are started from a uniform it may settle down to periodic behavior.
random initial configuration over all 128 sites, Notice that the transients have length-
while those in fig. 2 are started from configura- ened even more.
tions whose sites are all 0, with the exception of a -- 0.30 Transients have lengthened again.
patch of 20 randomized sites in the middle. X ~ 0.35 Transient length has grown signifi-
Fig. 1 illustrates the kinds of structures that cantly, and a new kind of periodic
develop, as well as the typical transient times structure with a longer period has ap-
before these structures are achieved. Fig. 2 illus- peared (fig. 1, X = 0.35). Most of the
trates the relative spread or collapse of the area of previous structures are still possible,
dynamical activity with time. For those values of hence the spectrum of dynamical possi-
exhibiting long transients, we have reduced the bilities is broadening.
scale of the arrays in order to display longer = 0.40 Transient length has increased to about
evolutions. 60 time steps, and a structure has ap-
We start with X--0.00. Note that under the peared with a period of about 40 time
strong quiescence condition mentioned above we steps. The area of dynamical activity is
18 C. Langton /Computation at the edge of chaos
A = 0.00 A = 0.05
A = 0.10 A = 0.15
A = 0.20
Fig. 1. Evolutions of one-dimensional, K = 4, N = 5 CAs from fully random initial configurations over 0.0 < h < 0.75. As A is
increased the structures become more complicated, and the transients grow in length until they become arbitrarily long at ~ -- 0.50.
For 0.50 < X < 0.75, the transient lengths decrease with increasing ~, as indicated by the arrows to the fight of the evolutions.
still collapsing down onto isolated peri- several cycles of its period, it is appar-
odic configurations. ent that the whole structure is moving
--- 0.45 Transient length has increased to al- to the left, and so this pattern will not
most 1000 time steps (fig. 1, X = 0.45). recur precisely in its same position until
Here, the structure on the right appears it has cycled at least once around the
to be periodic, with a period of about array. Furthermore, as it propagates to
100 time steps. However, after viewing the left, this structure eventually annihi-
C. Langton // Computation at the edge of chaos 19
A = 0.25 A = 0.30
" -
•, -,,~,.~-i~'~,, .-r,.-~ r~,n '."..
A = 0.35 A = 0.40
i .i *, ~
• " :': :i
Fig. 1. Continued
~2 f
"i
r P
Fig. 1. Continued
dynamical activity expands rather than come so long that,- for all practical pur-
contracts with time. There are, however, p o s e s - they are the steady state behav-
large fluctuations in the area covered by ior of the system over any period of
dynamical activity, and it is these fluc- time for which we can observe them.
tuations which lead to the eventual col- Whereas before, the dynamics eventu-
lapse of the dynamics. ally settled down to periodic behavior,
= 0.55 We have entered a new dynamical we are now in a regime in which the
regime in which the transients have be- dynamics typically settles down to ef-
C. Langton /Computation at the edge of chaos 21
A= 0.60 ,\ = 0.65
.... ,%
~.; , ~ + . ~
, ~".,
:~_~,-,-,
i~'~.:.~-
~.~._~
,
-r-.,-.-_-
--~,
~x..-N~,
~
.. ~ -_
~.v.~-..-~,'_-,,,-:
,
-~.'-~.=-~-';,"~. '%.11
A =0.70 A =0.75
~ : ~- ~
" ~i ;.-"~'T.~';-~,'r;
- ; ;: .'i->/?.
J 'I 1 t ~r" ~ - q , " _,~J :-" ~'. ,',L " .~ -~-.-,¢ ::
-~ j r~g.~,~._-~.-~-~ . . . . : n % ; - : :
•".0 ~" .~i.
~ -'-jk''tr:.. '~-,.,. ~...?.,, ,,, ,
t-, ~ ,, .~- ~,:.,~ _ ~,.~__~-.:':
•."~. . e ~'~'~
.... • "-,,.,
" M"
-.,~~ ,~... -:-:
~ - .....
.":~ .~ "~ ,
~y,- ,- ,. ,,, ,'-_ . , ~ , : ;. . .-..~,,,.'v ~ n
Fig. 1. Continued
22 C. Langton / Computation at the edge of chaos
..&m~
o",~"~ '~'rt~*'~
,.dr
T[
Fig. 2. Evolutions of one-dimensional, K = 4, N = 5 CAs from partially random initial configurations over 0.0 < X < 0.75. This
series illustrates the change in the rate of spread of the dynamics from negative for X < 0.45, to positive for X > 0.45. For 2, = 0.45,
the dynamics is balanced between collapse and expansion, giving rise to particle-like solitary waves.
f e c t i v e l y chaotic b e h a v i o r . F u r t h e r m o r e , o c c u p a t i o n d e n s i t y h a s s e t t l e d d o w n to
the previous trend of transient length w i t h i n 1% o f its l o n g - t i m e a v e r a g e . N o t e
increasing w i t h i n c r e a s i n g )t is r e v e r s e d . t h a t the a r e a o f d y n a m i c a l a c t i v i t y ex-
T h e a r r o w to t h e r i g h t o f t h e e v o l u t i o n s p a n d s m o r e r a p i d l y w i t h time.
o f figs. 1, )t = 0 . 5 5 - 0 . 7 5 i n d i c a t e s the h -- 0.60 T h e d y n a m i c s are q u i t e c h a o t i c , a n d t h e
approximate time by which t h e site- t r a n s i e n t l e n g t h to " t y p i c a l " c h a o t i c be-
C. Langton /Computation at the edge of chaos 23
Fig. 2. Continued
havior has decreased significantly. The one cell per time step in each direction,
area of dynamical activity expands more approximately half of the maximum
rapidly with time. possible rate for this neighborhood tem-
~ 0.65 Typical chaotic behavior is achieved in plate.
only 10 time steps or so. The area of = 0.70 Fully developed chaotic behavior is
dynamical activity is expanding at about reached in only 2 time steps. The area
24 C. Langton /Computation at the edge of chaos
X = 0.60 ~\ = 0.65
=0.70 X = 0.75
•- ~ J -
Fig. 2. Continued
15000 "l""l .... I'"'1 .... r 108 ' ' ' I ' ' ' ' I ' ' ' ' I ' ' ' ' I .... i ' ' '
"r .... I .... i'r,'l'Tf I
I0 r
10000
bO
10 e
5000 100000
ooooooo
ooooo:oi -0
-0
IIIrl
.1
.... I,,,ll
.2 .3
.... I ....
.4
I
.5
''lIlrIlllllllltllll,I
.6 .7 .8 .9
i0000
i000
-0
, ,, ,
100
I, ,, ,I
200
, , , , I ,
300 400
,, ,I , ,, ,I
500
, , ,
600
Number of Ceils
F i g . 3. A v e r a g e t r a n s i e n t l e n g t h as a f u n c t i o n o f h in a n a r r a y F i g . 4. G r o w t h o f a v e r a g e transients as a f u n c t i o n of array
o f 1 2 8 cells. size f o r h = 0.50.
First, transients grow rapidly in the vicinity of more, transient times exhibit decreasing depen-
the transition between ordered and disordered dy- dence on array size as X is increased past the
namics, a phenomenon known in the study of transition point. By the time all states are repre-
phase transitions as critical slowing down. The sented uniformly in the transition t a b l e - a t X =
relationship between transient length and 2~ is 0.75 in this c a s e - the transient lengths exhibit no
plotted in fig. 3. dependence on array s i z e - j u s t as was the case for
Second, the size of the array has an effect on the low values of 2~.
dynamics only for intermediate values of ~. For Third, the overall evolutionary pattern in time
low values of X, array size has no discernible appears more random as 2~--* 0.75. This observa-
effect on transient length. Not until X = 0.45 do tion is borne out by various entropy and correla-
we begin to see a small difference in the transient tion measures (see section 5). h = 0.75 represents
length as the size of the array is increased. For the state of maximal dynamical disorder.
= 0.50, however, array size has a significant ef- Fourth, the transition region supports both static
fect on the transient length. The growth of tran- and propagating structures (fig. 1, X = 0.45.) These
sient length as a function of array size for X = 0.50 particle-like structures are essentially solitary
is plotted in fig. 4. The essentially linear relation- waves, quasi-periodic patterns of state change,
ship on this log-normal plot suggests that transient w h i c h - l i k e the "gliders" in Conway's Game of
length depends exponentially on array size at ~, = Life [8]- propagate through the array, constantly
0.50. As we continue to raise X beyond 0.50, moving with respect to the fixed background of
although the dynamics is now settling down to the lattice. The ~, value for the Game of Life
effectively chaotic behavior instead of periodic (>'Li(e = 0.273) lies within the transition region for
behavior, the transient lengths are getting shorter K=2, N-9 2D CAs. Fig. 5 traces the time
with increasing X, rather than longer. A number evolution of an array of 512 sites, and shows that
of statistical measures (see ref. [17]) reveal that the the rule governing the behavior of fig. 1, ~, = 0.45
time it takes to reach "typical" behavior decreases supports several different kinds of particles, which
as ~ increases past the transition point. Further- interact with each other and with static periodic
26 c. Langton / Computation at the edge of chaos
4.1. Complications
3.5 -.... [ .... I " ' 1 ' " I .... I ' " ' 1 ' " ' 1 " " 1 .... I .... I ' " ~ variability as ~ is raised from - 0 . 6 to its
maximum value of 0.875.
3
I 1111~4'" ........ Two other features of this plot deserve special
2.5 i lli III!!!!1' mention. First, the abrupt cutoff of low H values
, i! at ), = 0 . 6 corresponds to the site-percolation
2 . I:~II ~i ' i
threshold Pc -- 0.59 for this neighborhood
template. Thus, we may suppose that, since ~ is a
1.5
dynamical analog of the site occupation prob-
i ability P, the dynamical percolation threshold for
a particular neighborhood template is bounded
.5 above by the static percolation threshold Pc- This
is borne out by experiments with other neighbor-
0
hood templates. For instance, the 9-neighbor
-.5 i, , , I , , , , I , , , , I , , , , I , , , , I , , , , I , , , , I , , , , I .... I .... I,, template exhibits a sharp cutoff at ~ = 0.4, which
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 corresponds well with the site percolation
X
threshold Pc = 0.402 for this lattice.
Fig. 6. A v e r a g e single cell e n t r o p y H over h space for a p p r o x - The second feature is the "ceiling" of the gap at
i m a t e l y 1 0 0 0 0 C A runs. Each point represents a different H = 0.84. This turns out to be the average entropy
t r a n s i t i o n function.
value for one of the most commonly occurring
chaotic rules. In such rules the dynamics has
5.1. Measures of complexity
collapsed onto only two s t a t e s - s q and one
other - and the rule is such that a mostly quiescent
The measures employed were chosen for their
neighborhood containing one non-quiescent state
collective ability to reveal the presence of informa-
maps to that non-quiescent state. In 1D CAs, such
tion in its various forms within CA dynamics.
rules give rise to the familiar triangular fractal
pattern known as the Sierpifiski gasket. There are
5.1.1. Shannon entropy
many ways to achieve such rules, and they can be
We use Shannon's entropy H to measure basic
achieved at very low )~ values. Most of the low-X
information capacitY. For a discrete process A o f
chaotic rules are of this type.
K states#2:
The entropy data of fig. 6 suggest an anomaly at
K
intermediate parameter values, possibly a phase
H ( A ) = - Y'~ pilogpi. (2)
i=1
transition between two kinds of dynamics. Since
there seems to be a discrete jump between low and
Fig. 6 shows the average entropy per cell, H, as
high entropy values, the evidence points to a
a function of ~ for approximately 10 000 CA runs.
first-order transition, similar to that observed
The random-table method was employed, so each
between the solid and fluid phases of matter.
point represents a distinct random transition table.
However, the fact that the gap is not completely
First, note the overall envelope of the data and
empty suggests the possibility of second-order
the large variance at most ~ points. Second, note
transitions as well.
the sparsely populated gap over 0.0 < ~ < 0.6 and
The table-walk-through method of varying
between 0.0 < H _< 0.84. This distribution appears
reveals more details of the structure of the entropy
to be bimodal, suggesting the presence of a phase
data. Fig. 7 shows four superimposed examples of
transition. Third. note the rapid decrease in
the change in the average cell entropy as we vary
"*2Throughout, log is t a k e n to the base 2, thus the u n i t s are the ~ value of a table. Notice that in each of the
bits. four cases the entropy remains fairly close to zero
28 C. Langton / Computation at the edge of chaos
~, 2.5
2.5
o
2
"~ 1.5
1.5
1
1
"~ .5
.5
F i g . 7. S u p e r p o s i t i o n o f 4 t r a n s i t i o n e v e n t s . N o t e t h e d i f f e r e n t
F i g . 8. S u p e r p o s i t i o n o f 50 t r a n s i t i o n e v e n t s , s h o w i n g t h e in-
k v a l u e s at w h i c h t h e t r a n s i t i o n s t a k e p l a c e .
t e r n a l s t r u c t u r e o f fig. 6.
3.5 ' I ' ' ' ' I ' ' ' ' I ' ' '
1 ,,,i,,,,i,~,,i,,,,l~,,,i,,,,i,,,,i,,,,i,,,,l~,,i,,~,z
.9
3
.8
2.5 .7
a= .6 - _
2
,~ .5
> 1.5 >~ .4
<
.3
I
.2
, :;q,,.
.1
...... i !!:;i:i ~
.5
- = "~'!I!! iil l Ii
'' mlHiiIlfiih,,m,,,, ..........
0
-0 ,,,I .... I .... I .... I .... I .... I .... I .... I .... I .... I,~
--.1
-1 -.5 0 .5 I .1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9
A X
Fig. 10. Superposition of 50 transition events lined up by A?~. Fig. 11. Average mutual information between a cell and itself
Compare with fig. 8. at the next time step.
The mutual information is a simple function of approach to effectively random dynamics. The lack
the individual cell entropies, H(A) and H(B), and of correlation between even adjacent cells at high
the entropy of the two cells considered as a joint h means that cells are acting as if they were
process, H(A, B): independent of each other, even though they are
causally connected. The resulting global dynamics
I(A;B) = H ( A ) +H(B) - H(A,B). (3) is the same as if each cell picked its next state at
uniform random from among the K states, with
This is a measure of the degree to which the no consideration of the states of its neighbors.
state of cell A is correlated with the state of cell B, This kind of global dynamics is predictable in the
and vice versa. same statistical sense that an ideal gas is globally
Fig. 11 shows the average mutual information predictable. In fact it is appropriate to view this
between a cell and itself at the next time step. dynamical regime as a hot gas of randomly flipping
Note the tight convergence to low values of the cells.
mutual information for high )~ and the location of Fig. 13 shows the average mutual information
the highest values. curves for several different temporal and spatial
The increase of the mutual information in a separations. Note that the decay in both time and
particular region is evidence that the correlation space is slowest in the middle region.
length is growing in that region, further evidence At intermediate )~ values, the dynamics support
for a phase transition. the preservation of information locally, as indi-
Fig. 12 shows the behavior of the average mutual cated in the peak in correlations between distinct
information as 9~ is varied, both against ?~ and cells. If cells are cooperatively engaged in the
A?t. The average mutual information is essentially support of a computation, they must exhibit
zero below the transition point, it jumps to a s o m e - b u t not too m u c h - c o r r e l a t i o n in their
moderate value at the transition, and then decays behaviors. If the correlations are too strong, then
slowly with increasing ~. The jump in the mutual the cells are overly dependent, with one mimicing
information clearly indicates the onset of the the other - not a cooperative computational
chaotic regime, and the decaying tail indicates the enterprise.
30 C. Langton / Computation at the edge of chaos
, , , , i , i , , i , ,
0
J
CD
<
< 0
.1 , , i t l t l , l [ l l Jlt,ll
-0 .1 .2 .3 .4 .5 .6 .7 .8 .9 -.5 0 .5
X AX
SPACE
1 2 3
F i g . 13. D e c a y o f a v e r a g e m u t u a l i n f o r m a t i o n i n s p a c e a n d time.
C. Langton /Computation at the edge of chaos 31
On the other hand, if the correlations are too measures is plotted in fig. 14. Again, we see clear
small, then the cells are overly independent, and evidence of a phase transition.
again, they cannot cooperate in a computational The envelope of the relationship is bounded
enterprise, as each cell does something totally below the transition by the linear bound that H
unpredictable in response to the state of the other. places on the mutual information. All of the points
Correlations in behavior imply a kind of common on this line are for periodic CAs. This line inter-
code, or protocol, by which changes of state in one sects the curve bounding the envelope above the
cell can be recognized and understood by the transition at an entropy value H c ---0.32 on the
other as a meaningful signal. With no correlations normalized entropy scale.
in behavior, there can be no common code with This is a very informative plot. There is a clear,
which to communicate information. sharply defined maximum value of mutual infor-
mation at a specific value of the entropy, and the
mutual information falls off rapidly on either side.
6. Mutual information and entropy This seems to imply that there is an optimal work-
ing entropy at which CAs exhibit large spatial and
It is often useful to examine the way in which temporal correlations. Why should this be the
observed measures behave when plotted against case?
one another, effectively removing the (possibly Briefly, information storage involves lowering
unnatural) ordering imposed by the control pa- entropy while information transmission involves
rameter. raising entropy [10]. In order to compute, a system
Of the measures we have looked at, the most must do both, and therefore must effect a trade-off
informative pair when plotted against each other between high and low operating entropy. It would
are the mutual information and the average single seem from the work reported here that this trade-
cell entropy. The relationship between these two off is optimized in the vicinity of a phase transi-
tion.
A similar relationship has been observed by
1.I
Crutchfield at Berkeley in his work on the transi-
1
tion to chaos in continuous dynamical systems [6].
.9
This relationship is illustrated in fig. 15. Briefly,
.8 the ordinate of this p l o t - C - i s a measure of the
.7 size of the minimal finite state machine required
to recognize strings of l's and O's generated by a
~ .5 dynamical system (the logistic map, in this case)
o when these strings are characterized by the nor-
.3 malized per-symbol entropy listed on the abcissa.
.2
The observance of this same fundamental en-
tropy/complexity relationship in these different
.1
classes of dynamical systems is very exciting.
o
These relationships support the view that, rather
-.1
than increasing monotonically with random-
I 0 .i .2 .3 .4 .5 .6 .7 .8 .9 1 1.1
normalized H n e s s - a s is the case for the usual measures of
complexity, such as that of Chaitin and Kol-
Fig. 14. Average mutual information versus average single cell mogorov [4, 16]-complexity increases with ran-
entropy H. The mutual information in this case is computed
between a cell and itself at the next time step. The entropy is donmess only up to a p o i n t - a phase transition-
normalized to 1.0. after which complexity decreases with further in-
32 C. Langton / Computation at the edge of chaos
HIGH
Zk
C1
LOW
| | •
If Wolfram is correct in attributing the capacity lem" is undecidable: there exist computations for
for universal computation to class IV CAs, then which it is not possible to decide whether or not
when we locate class IV CAs at a phase transition, they will halt.
we are also locating universal computation at a Thus, with respect to our ability to decide the
phase transition. ultimate outcome of computations, there are es-
sentially three possibilities: we can determine that
7.2. Complexity classes they will halt, we can determine that they will not
halt, or we cannot determine whether or not they
One obvious property of computations for which will halt.
we would like to find an analog in phase-transition As we have seen, there are three similar possibil-
phenomena is the existence of the various com- ities for the ultimate outcome of the evolutions of
plexity classes. Some computations may be per- CAs. CAs below the transition point rapidly
formed using an amount of time or space which is "freeze up" into short-period behavior from any
only a l i n e a r - or even a c o n s t a n t - function of the possible initial configuration. On the other hand,
"size" of the input, while other computations ex- CAs above the transition point will never freeze
hibit polynomial, or even exponential dependence into periodic behavior, settling down rapidly in-
[9]. Where can we find a natural analog of these stead to chaotic behavior. Thus, we can predict
complexity classes within the phenomenology of the ultimate dynamics of CAs away from the
phase transitions? transition point with a high degree of certainty.
The obvious answer is in the divergence of For CAs in the vicinity of the transition, how-
transient times as one approaches the phase tran- ever, both of these ultimate dynamical outcomes
sition. As illustrated in the qualitative dynamics of are possible, and because of the extended tran-
1D CAs, for ~, values far from the transition sients, it will be "effectively" undecidable whether
point, transients die out in time which is indepen- a particular rule operating on a particular initial
dent of the size of the array. As X approaches the configuration will ultimately lead to a frozen state
transition point, transients begin to show more or not for this range of ~,.
and more dependence on array size. For values of Thus, we can identify a natural analog of
very near a "critical" transition, this size depen- Turing's Halting problem in what we call the
dence appears to be exponential or worse. This is Freezing problem: for an arbitrary CA in the vicin-
true whether we approach the transition from the ity of the transition point, will the dynamics ulti-
ordered regime or the disordered regime, which mately "freeze up" into short-period behavior or
suggests that in addition to the familiar complex- not? It is quite likely that the freezing problem is
ity-class hierarchy for halting computations, there undecidable.
should be a similar complexity-class hierarchy for
non-halting computations.
This last point brings up another property of Let us now lay out in general outline an inter-
computation which should be reflected in phase- pretation that will tie together all of these dis-
transition dynamics. parate phenomena into a coherent picture of the
Some computations halt, and some do not. For nature of computation. The reader should bear in
some computations, we can decide whether or not mind that this interpretation, although strongly
they will halt. However, Turing demonstrated that supported by evidence, is only a conjecture at this
for certain classes of machines this "halting prob- point; many details remain to be worked out.
34 C. Langton /Computation at the edge of chaos
8.1. Solids, fluids, and dynamics system in the vicinity of a solid/liquid or a liq-
uid/vapor transition.
We propose that the solid and fluid phases of
matter, with which we are so familiar from every- 8.2. Related work
day experience, are much more fundamental as-
pects of nature than we have supposed them to be. Others have been working on the problem of
Rather than merely being possible states of mat- finding structure in the rule spaces of cellular
ter, they constitute two fundamental universality automata and other, similar spatially distributed
classes of dynamical behavior. dynamical systems.
We know solids and fluids primarily as states of In my initial investigations with the )~ parame-
matter because up until quite recently, everything ter [18], I suggested that Wolfram's class IV CAs
that exhibited dynamical behavior was made up of constituted a transition between class II and class
some kind of material. Now, however, with the III, that is, between periodic and chaotic dynam-
availability of computers, we are able to experi- ics.
ment with dynamics abstracted from any particu- Kauffman [14, 13] has investigated a class of
lar material substrate. The findings reported in related dynamical systems known as Boolean nets,
this paper suggest that for dynamical systems in in which he finds a similar phase transition be-
general- whether purely formal or manifestly ma- tween ordered and disordered dynamics.
terial- there are primarily only two ultimate dy- Vichniac, Tamayo and Hartman [25] discovered
namical possibilities. that the Wolfram classes could be recovered by
However, these two universality classes are sep- varying the frequency of two simple rules in an
arated by a phase transition. The dynamics of inhomogeneous cellular automaton. They also sug-
systems within this transition region-especially gested a relation between critical slowing down
the "critical" systems- appear to support the ba- and the halting problem.
sic mechanisms necessary for information trans- Packard and Li [20] have mapped out the space
mission, storage, and modification, and therefore of "elementary" K = 2, N = 3, 1D CAs fairly
provide the capacity for emergent computation. completely, using a parameterization scheme simi-
Thus, a third possibility is that systems can be lar to ~.
constructed in such a way that they manage to Packard [22] has also performed an interesting
avoid either of the two primary dynamical out- series of experiments in which he "adapts" CA
comes by maintaining themselves on indefinitely rules by selecting for certain behaviors. He finds
extended transients. an initially random population of rules will drift
It is a system's capacity for supporting a dy- towards the phase-transition region. His interpre-
namics of information that allows complex behav- tation of this phenomenon is that it is easier to
ior in the vicinity of a phase transition. This in find rules which will compute the desired
turn allows for ihe possibility of the freezing behavior-by making use of a general computa-
problem. Since computers and computations are tional capacity-than it is to find rules that are
specific instances of material and formal systems "hard-wired" to produce only the desired behav-
respectively, they are also ultimately bound by ior.
these universalicty classes. Therefore, if this inter- McIntosh [21] has applied the mean-field ap-
pretation is correct, the halting problem can be proach of Gutowitz [12, 11] and suggests that the
seen as a specific instance of the more general Wolfram classes can be distinguished on the basis
freezing problem for dynamical systems. We can of simple features of the mean field theory curves.
therefore view computations as special instances Wootters [30] has applied mean-field theory to
of the kinds of processes that occur in a physical explain the results from the ~ parameter, and has
c. Langton /Computation at the edge of chaos 35
been able to reproduce many of the features of organized critical system, although he does not
fig. 6. bring Life's computational capacity into the dis-
Together with Crutchfield's work mentioned cussion.
earlier, these results collectively point to the exis- Finally, what are the implications for under-
tence of a phase transition in the spectrum of standing the origin and evolution of life? One of
dynamical systems, and also suggest that the com- the most exciting implications of this point of
plex dynamics of systems in the vicinity of a phase view is that life had its origin in just these kinds
transition rest on a fundamental capacity for pro- of extended transient dynamics. Looking at a liv-
cessing information. ing cell, one finds phase-transition phenomena
everywhere. The point of view advocated here
8.3. Questions would suggest that we ourselves are examples of
the kind of "computation" that can emerge in the
There are many questions that need to be ad- vicinity of a phase transition given enough time.
dressed. For instance, can the "fluid" dynamical Now nature is not so beneficient as to maintain
systems be further divided up into "gases" and conditions at or near a phase transition forever.
"liquids"? There is some evidence for both Therefore, in order to survive, the early extended
solid/liquid and liquid/gas transitions in the space transient systems that were the precursors of life
of CAs [17]. as we now know it had to gain control over their
How might these issues be addressed by statisti- own dynamical state. They had to learn to main-
cal mechanics, which has been very effective in tain themselves on these extended transients in the
treating phase transitions in general? Can analogs face of fluctuating environmental parameters, and
for temperature, pressure, volume, and energy be to steer a delicate course between too much order
found? There is some evidence that equivalent and too much chaos, the Scylla and Charybdis of
measures can be defined [6, 17]. On the other dynamical systems. Such transient systems must
hand, it is possible that statistical mechanics alone have "discovered" how to make use of their intrin-
will not be able to fully treat phase-transition sic information processing capability in order to
phenomena without being augmented by ideas sense and respond to their local environment.
from the theory of computation. Evolution has been the process by which such
What are the implications for optimization tech- systems have managed to gain local control over
niques such as simulated annealing [15], which call more and more of the environmental variables
for extended stays in the vicinity of the freezing affecting their ability to maintain themselves on
point? It is interesting that this is the very point at extended transients with essentially open futures.
which we would expect information processing to
emerge spontaneously within the system being an-
n e a l e d - suggesting that the real reason for hover-
9. Conclusion
ing in the vicinity of the freezing point is to allow
the system to compute its own solution via an
V o n Neumann observed that#3:
emergent computation.
" T h e r e is thus this completely decisive property
How are the notions reported here related to
of complexity, that there exists a critical size be-
Bak's self-organized criticality [1]? In many ways,
low which the process of synthesis is degenerative,
it seems that Bak has discovered that dynamical
but above which the phenomenon of synthesis, if
systems can be made to boil when driven in the
right way, which is a phenomenon we would
¢*3john von Neumann, in his 1949 University of Illinois
expect at a phase transition. In fact, Bak has lectures on the Theory and Organization of Complicated Au-
suggested that Conway's game of Life is a self- tomata [26].
36 C. Langton /Computation at the edge of chaos
properly arranged, can become explosive, in other [2] E. Berlekamp, J.H. Conway and R. Guy, Winning Ways
for Your Mathematical Plays (Academic Press, New York,
words, where syntheses of automata can proceed 1982).
in such a manner that each automaton will pro- [3] A.W. Burks, Essays on Cellular Automata (University of
duce other automata which are more complex and Illinois Press, Urbana, IL, 1970).
[4] G. Chaitin, J. Assoc. Comput. Mach. 13 (1966)-145.
of higher potentialities than itself".
[5] E.F. Codd, Cellular Automata (Academic Press, New
York, 1968).
Although we are using a slightly different sense [6] J.P. Crutchfield and K. Young, Computation at the onset
of "complexity" than von Neumann, the results of of chaos, in: Complexity, Entropy, and Physics of Infor-
mation, ed. W. Zurek (Addison-Wesley, Reading, MA,
this paper support his observation. More impor-
1990).
tantly, however, we suggest that a similar observa- [7] E. Fredkin and T. Toffoli, Conservative logic, Int. J.
tion can be made in the case of too much "com- Theor. Phys. 21 (1982) 219-253.
plexity": aboue a certain level of "complexity", [8] M. Gardner, Mathematical games: The fantastic combina-
tions of John Conway's new solitaire game 'Life', Sci. Am.
the process of synthesis is also degenerative. 223(4) (October 1979) 120-123.
In other words, we find that there exist an upper [9] M.R. Garey and D.S. Johnson, Computers and In-
limit as well as a lower limit on the "complexity" tractability (Freeman, San Fransisco, 1979).
[10] L.L. Gatlin, Information Theory and the Living System
of a system if the process of synthesis is to be
(Columbia Univ. Press, New York, 1972).
non-degenerative, constructive, or open ended. We [11] H.A. Gutowitz, A hierarchical classification of cellular
also find that these upper and lower bounds seem automata, in: Proceedings of the 1989 Cellular Automata
Workshop, ed. H.A. Gutowitz (North-Holland, Amster-
to be fairly close together and are located in the
dam, 1990), Physica D, to be published.
vicinity of a phase transition. [12] H.A. Gutowitz, J.D. Victor and B.W. Knight, Local struc-
As the systems near the phase transition exhibit ture theory for cellular automata, Physica D 28 (1987)
a range of behaviors which reflects the phe- 18-48.
[13] S.A. Kauffman, Emergent properties in random complex
nomenology of computations surprisingly well, we automata, Physica D 10 (1984) 145-156.
suggest that we can locate computation within the [14] S.A. Kauffman, Metabolic stability and epigenesis in ran-
spectrum of dynamical behaviors at a phase tran- domly constructed genetic nets, J. Theor. Biol. 22 (1969)
437-467.
sition here at the "edge of chaos".
[15] S. Kirkpatrick, C.D. Gelatt and M.P. Vecchi, Optimiza-
tion by simulated annealing, Science 220 (1983) 671-680.
[16] A.N. Kolmogorov, Prob. Inf. Transm. 1 (1965) 1.
[17] C.G. Langton, Computation at the edge of chaos, Ph.D.
Acknowledgements
Thesis, University of Michigan (1990).
[18] C.G. Langton, Studying artificial life with cellular au-
Many people have contributed to the ideas pre- tomata, Physica D 22 (1986) 120-149.
sented here. I have benefitted greatly from conver- [19] W. Li, Analyzing Complex Systems, Ph.D. Thesis,
Columbia University (1989).
sations with Richard Bagley, Jim Crutchfield, [20] W. Li and N.H. Packard, Structure of elementary cellular
D o y n e Farmer, Howard Gutowitz, Hyman automata rule-space, Complex Systems, submitted for
Hartman, Stuart Kauffman, Wentian Li, Norman publication (1990).
[21] H.V. McIntosh, in: Proceedings of the 1989 Cellular Au-
Packard, Steen Rasmussen, Rob Shaw and Bill
tomata Workshop, ed. H.A. Gutowitz (North-Holland,
Wooters. Stephanie Forrest has been a long time Amsterdam), Physica D, to be published.
intellectual companion and critic, and served as [22] N.H. Packard, Adaptation toward the edge of chaos,
midwife in the delivery of this paper. Technical Report, Center for Complex Systems Research,
University of Illinois, CCSR-88-5 (1988).
[23] N.H. Packard and S. Wolfram, Two-dimensional cellular
automata, J. Stat. Phys. 38 (1985) 901.
[24] A.R. Smith III, Simple computation-universal cellular
References
spaces, J. Assoc. Comput. Mach. 18 (1971) 339-353.
[25] G.Y. Vichniac, P. Tamayo and H. Hartman, Annealed
[1] P. Bak, C. Tang and K. Wiesenfeld, Self-organized criti- and quenched inhomogeneous cellular automata, J. Stat.
cality, Phys. Rev. A 38 (1988) 364-374. Phys. 45 (1986) 875-883.
C. Langton /Computation at the edge of chaos 37
[26] J. von Neumann, Theory of self-reproducing automata, Automata (World Scientific, Singapore, 1986).
1949 University of Illinois Lectures on the Theory and [29] S. Wolfram, Universality and complexity in cellular au-
Organization of Complicated Automata, ed. A.W. Burks tomata, Physica D 10 (1984) 1-35.
(University of Illinois Press, Urbana, IL, 1966). [30] W.T. Wootters and C.G. Langton, Is there a sharp phase
[27] S. Wolfram, Statistical mechanics of cellular automata, transition for deterministic cellular automata'?., in: Pro-
Rev. Mod. Phys. 55 (1983) 601-644. ceedings of the 1989 Cellular Automata Workshop, ed.
[28] S. Wolfram, ed., Theory and Applications of Cellular H.A. Gutowitz, to appear in Physica D (1990).