1: \documentclass{elsart}
2: \usepackage[dvips]{graphics}
3: \parindent=1em
4: \begin{document}
5: \runauthor{Cicero, Caesar and Vergil}
6:
7: \begin{frontmatter}
8:
9: \title{
10: A New Parameter $F$ to Classify Cellular Automata Rule Table Space and
11: a Phase Diagram in $\lambda-F$ Plane
12: %Structure of Rule Tables and Phase Diagram of One-Dimensional
13: %Cellular Automata\\
14: %(Revised)
15: }
16:
17: \author{Sunao Sakai and Megumi Kanno}
18:
19: \address{
20: Faculty of Education, Yamagata University, Yamagata,
21: 990-8560, Japan}
22: \thanks{E-mail: sakai@e.yamagata-u.ac.jp}
23:
24: \begin{abstract}
25: It is shown that for the N-neighbor and K-state cellular automata, the
26: class II, class
27: III and class IV patterns coexist at least in the range
28: $\frac{1}{K} \le \lambda \le 1-\frac{1}{K} $.
29: The mechanism which determines the difference between the pattern
30: classes at a fixed $\lambda$ is found, and it is studied
31: quantitatively by introducing a new
32: parameter $F$. Using the parameter F and $\lambda$, the phase diagram
33: of cellular automata is obtained for 5-neighbor and 4-state cellular
34: automata.\\
35:
36: \vspace*{0.3cm}
37: \noindent
38: PACS: 89.75.-k Complex Systems
39: \end{abstract}
40: \maketitle
41: \end{frontmatter}
42:
43: \section{Introduction}
44: Cellular automata (CA) has been one of the most
45: studied fields in the research of
46: complex systems.
47: Various patterns has been generated
48: by choosing the rule tables.
49: Wolfram\cite{wolfram} has classified these patterns into four rough
50: categories:
51: class I (homogeneous), class II (periodic), class III (chaos) and
52: class IV (edge of chaos).
53: The class IV patterns have been the most interesting target for the
54: study of CA, because it provides us with an example of the
55: self-organization in a simple
56: system and it is argued that the possibility of
57: computation is realized by the complexity at the edge of
58: chaos\cite{wolframs,langton,mitchell}.\\
59: \indent
60: A more detailed classification of CA, has been carried out mainly for
61: the elementary cellular automata (3-neighbor and 2-state
62: CA)\cite{hanson,wuensche}, in which the pattern is studied quite
63: accurately for each rule table.
64: However the number of rule table in
65: N-neighbor and K-state cellular automata grows like $K^{K^{N}}$. Therefore
66: except for a few smallest combinations of the $N$ and $K$, the numbers
67: of the rule tables become so large that studies of the CA dynamics for
68: all rule tables are impossible even with the fastest supercomputers. \\
69: \indent
70: On the other hand the rule table of edge of
71: chaos is rare in the whole CA rule table space, therefore
72: it is important to find a set of parameters by which the pattern
73: classes could be classified, and to determine a
74: phase diagram of CA in these parameter space, even if it is a
75: qualitative one.\\
76: \indent
77: Langton has introduced $\lambda$ parameter and
78: argued that as $\lambda$ increases the pattern class
79: changes from class I to class II and then to class
80: III. And in many cases, class IV behavior is
81: observed between class II and class III
82: patterns\cite{langton0,langton,langton2}.
83: The $\lambda$ parameter represents rough behavior of CA in the rule table
84: space, but finally
85: does not sufficiently classify the quantitative behavior of CA.
86: %\cite{langton,langton2}.
87: It is well known that different pattern classes coexist at the same
88: $\lambda$.
89: Which of these pattern classes is chosen, depends on
90: the random number.
91: The reason or mechanism for this is not yet known;
92: we have no way to control the pattern classes at fixed $\lambda$.
93: And the transitions between a periodic to
94: chaotic pattern classes are observed in a rather wide range of $\lambda$.
95: In Ref.\cite{langton2}, a schematic phase-diagram was sketched. However
96: a vertical axis was not specified.
97: Therefore, it is has been thought
98: that more parameters are necessary to arrive at a more quantitative
99: understanding of the rule table space of the CA.\\
100: \indent
101: In this article, we will report a mechanism which discriminates the
102: pattern classes at a fixed $\lambda$. The mechanism is closely related
103: to the structure of the rule tables and is expressed by the
104: numbers of rules which breaks
105: strings of quiescent state.
106: For N-neighbor and K-state CA, it is found that in the region
107: $1/K \le \lambda \le 1-1/K$, the class II, class III and class IV
108: pattern classes coexist.\\
109: \indent
110: This property is studied quantitatively by
111: introducing a new parameter $F$, which is taken to be orthogonal to
112: $\lambda$.
113: In the region $1/K \le \lambda \le 1-1/K$,
114: the maximum of $F$
115: correspond to class III rule tables while minimum of $F$, to class II
116: or class I rule tables. Therefore
117: the transition of the pattern classes takes place somewhere between
118: these two limits without fail.
119: If we determine the region of $F$, where the transitions of the pattern
120: classes take place, we could obtain the phase diagram in $\lambda-F$
121: plane. \\
122: \indent
123: The determination of the phase diagram is carried out for
124: 5-neighbor
125: and 4-state CA. In this case, phase boundary is not sharp but has some
126: range in $\lambda-F$ plane. The region has a gentle slope
127: as a function of $\lambda$, and
128: extends over the range
129: $0.2 \le \lambda \le 0.8$. This means
130: that for this CA the edge of chaos could be found at least in this
131: range in $\lambda$.\\
132: \indent
133: In section 2, we will briefly summarize our notations and present
134: a key discovery, which leads us to the understanding of the structure
135: of the rule table and pattern classes. It strongly suggested that the
136: rules which break
137: strings of the quiescent states play an important role for
138: the pattern classes.\\
139: \indent
140: In section 3, we classify rule tables according to the
141: destruction and construction of strings of the quiescent states,
142: and carry out the replacements of the rules to
143: change the chaotic pattern class into periodic one
144: and vice versa while keeping $\lambda$ fixed.
145: The reason why the patten classes changes by the replacements is
146: discussed, and we will show that in the region $1/K \le \lambda \le
147: 1-1/K$, the change of the pattern classes takes place without fail
148: by the replacements. \\
149: \indent
150: In section 4, the result obtained in section 3 is studied
151: quantitatively by introducing a new parameter $F$.
152: Using $F$ and $\lambda$, we
153: determine the phase diagram in the $\lambda-F$ plane for 5-neighbor
154: and 4-state CA. \\
155: \indent
156: Section 5 is devoted to conclusions and discussions.
157:
158: \section{Summary of CA and a key discovery}
159: \subsection{ Summary of Cellular Automata }
160: In order to make our arguments concrete, we focus mainly on
161: the one-dimensional 5-neighbor and 4-state CA in the following,
162: however, the qualitative conclusions hold true for other CAs.
163: This point will be discussed in subsections 3.2.\\
164: \indent
165: We will briefly summarize our notation of CA\cite{wolfram,langton}.
166: In our study, the site consists of
167: 150 cells having the periodic boundary condition. The states are denoted
168: as $s(t,i)$.
169: The $t$ represents the time step which takes an integer value,
170: and the $i$ is the position of cells which range from $0$ to
171: $149$.
172: The $s(t,i)$ takes values $0,1,2,$ and $3$, and the state $0$ is taken
173: to be the quiescent state.
174: The set of the states $s(t,i)$
175: at the same $t$ is called the configuration. \\
176: \indent
177: The configuration at time $t+1$
178: is determined by that of time $t$ by using following local relation,
179: \begin{equation}
180: s(t+1,i) = T(s(t,i-2),s(t,i-1),s(t,i),s(t,i+1),s(t,i+2)).
181: \label{table}
182: \end{equation}
183: The set of the mappings
184: \begin{equation}
185: T(\mu,\nu,\kappa,\rho,\sigma)=\eta,(\mu,\nu,etc.= 0,1,2,3)
186: \label{r_table}
187: \end{equation}
188: is called the
189: rule table. The rule table consists of $4^5$ mappings,
190: which are selected from a total of $4^{1024}$ elements.\\
191: \indent
192: The $\lambda$ parameter is defined as\cite{langton}
193: \begin{equation}
194: \lambda=\frac{N_{h}}{1024},
195: \label{lambda}
196: \end{equation}
197: where $N_{h}$ is the number in which $\eta$ in Eq.\ref{r_table} is not
198: equal to $0$. In other words
199: the $\lambda$ is the probability that the rules do not select the
200: quiescent state in next time step. In the following we set the
201: rule tables randomly with the probability $\lambda$.
202: We choose $1024-N_{h}$ rules randomly, and set $\eta=0$ in the
203: right hand side of Eq. \ref{lambda}.
204: For the rest of the $N_{h}$ rules, the $\eta$ picks up the number $1,2,3$
205: randomly.
206: The initial configurations are also set randomly.\\
207: \indent
208: The time sequence of the configurations
209: is called a pattern. The patterns are classified
210: roughly into four classes established by Wolfram\cite{wolfram}.
211: It has been known that as the
212: $\lambda$ increases the most frequently generated
213: patterns change from homogeneous (class I) to
214: periodic (class II) and then to chaotic (class III),
215: and at the region between class II and class III, the edge of
216: chaos (class IV) is located.
217: \subsection{Correlation between pattern classes and rules
218: which break strings of quiescent state at a fixed $\lambda$}
219: In order to find the reason why the different pattern classes are
220: generated with the same $\lambda$,
221: we have started to collect rule tables of
222: different pattern classes, and tried to
223: find the differences between them.
224: We have fixed at $\lambda=0.44$ $(N_{h}=450)$,
225: because at this point the chaotic, edge of chaos, and periodic patterns
226: are generated with a similar ratio.
227: By changing the random number, we have
228: gathered a few tens of the rule tables and classified them
229: into chaotic, edge of chaos, and periodic ones.\\
230: \indent
231: In this article, a pattern is considered the edge of chaos when its
232: transient length\cite{langton} is longer
233: than $3000$ time steps.\\
234: \indent
235: First, we study whether or not the pattern classes are sensitive to the
236: initial configurations.
237: We fix the rule table and change the initial configurations.
238: The details of the patterns depend on the initial configurations, but
239: the pattern classes are not changed\cite{wolfram}.
240: Thus the difference of the pattern classes is due to the differences
241: in the rule tables, and the target of our inquiry has to do with the
242: differences between them.\\
243: \indent
244: For a little while, we do not impose a quiescent
245: condition (QC)\cite{langton},\\
246: $T(0,0,0,0,0)=0$,
247: because without this condition,
248: the structure of the rule table becomes more transparent.
249: This point will be discussed at the footnote 4 in section 4.\\
250: \indent
251: After some trial and error, we have found a strong correlation between
252: the pattern classes and the QC.
253: For class II patterns,
254: the probability of the rule table,
255: which satisfies the QC is much larger than that of the class III
256: patterns. This correlation has suggested that the rule
257: $T(0,0,0,0,0)=h,\hspace{0.2cm} h \neq 0$, which breaks the string of the
258: quiescent states with length 5,
259: pushes the pattern toward chaos.
260: We anticipate that the similar situation will hold for the strings of
261: quiescent states with length 4.\\
262: \indent
263: We go back to the usual definitions of CA. In the following we discuss
264: CA under QC, $T(0,0,0,0,0)=0$.
265: We study the correlation between the number of the rules of
266: Eq.\ref{d4} and the pattern classes:
267: \begin{equation}
268: T(0,0,0,0,i)=h , \hspace {0.5cm} T(i,0,0,0,0)=h,(i,h=1,2,3).
269: \label{d4}
270: \end{equation}
271: \indent
272: These rules break length 4 strings of the quiescent states, and
273: will also push the pattern toward chaos
274: \footnote{Similar ideas had
275: been noticed by Wolfram and Suzudo with the
276: arguments of the unbounded growth\cite{wolfram} and
277: expandability\cite{suzudo}.\\}.
278: We denote the total number of rules of Eq.\ref{d4} in a rule table
279: as $N_4$.
280: We have collected 30 rule tables and grouped them by the number $N_4$.
281: We have 4 rule tables with $N_4 \ge 4$ , 13 rule tables with $N_4=3$, 9
282: rule tables with $N_4=2$ and 4 rule tables with $N_4 \le 1$.
283: When $ N_4 \ge 4$, all rule tables generate
284: chaotic patterns, while when $N_4 \le 1$, only
285: periodic ones are generated.
286: At $N_4=3$ and $N_4=2$, chaotic, edge of chaos, and periodic
287: patterns coexist.
288: Examples are shown in the Fig.\ref{pattern_at_lambda0.44}.
289: The coexistence of three pattern classes at $N_4=3$ is seen in
290: Fig.\ref{pattern_at_lambda0.44}(b),
291: Fig.\ref{pattern_at_lambda0.44}(c) and Fig.\ref{pattern_at_lambda0.44}(d)
292: and that of $N_4=2$ is exhibited in
293: Fig.\ref{pattern_at_lambda0.44}(e),
294: Fig.\ref{pattern_at_lambda0.44}(f) and Fig.\ref{pattern_at_lambda0.44}(g).\\
295: \begin{figure}
296: \begin{center}
297: \scalebox{0.58}{ { \includegraphics{Fig1.eps} } }
298: \caption{
299: The pattern classes at $\lambda=0.44$. The quiescent state is
300: shown by white dot, while other states are indicated by black
301: point. Fig.\ref{pattern_at_lambda0.44}(a) corresponds to $N_4=4$,
302: Fig.\ref{pattern_at_lambda0.44}(b), Fig.\ref{pattern_at_lambda0.44}(c),
303: and Fig.\ref{pattern_at_lambda0.44}(d), to $N_4=3$,
304: Fig.\ref{pattern_at_lambda0.44}(e), Fig.\ref{pattern_at_lambda0.44}(f), and
305: Fig.\ref{pattern_at_lambda0.44}(g), to $N_4=2$, and
306: Fig.\ref{pattern_at_lambda0.44}(h) corresponds to $N_4=1$.
307: }
308: \label{pattern_at_lambda0.44}
309: \end{center}
310: \end{figure}
311: \indent
312: The strong correlation between $N_4$
313: %, the number of rules of
314: %Eq.\ref{d4} in a rule table,
315: and the pattern classes has been
316: observed in this case too, as anticipated.
317: These discoveries have provided us with a key hint leading us to the
318: hypothesis
319: that the rules, which break strings of the quiescent states, will
320: play a major role in the pattern classes.
321:
322: \section{Structure of rule table and pattern classes}
323: \subsection{Structure of rule table and replacement experiment}
324: In order to test the hypothesis of the previous section,
325: we classify the rules into four groups according to the
326: operation on strings of the quiescent states.
327: In the following, Greek characters in the rules
328: represent groups $0,1,2,3$ while Roman, represent groups $1,2,3$.\\
329: \noindent
330: Group 1: $T(\mu,\nu,0,\rho,\sigma)=h$.\\
331: The rules in this group break strings of the quiescent states.\\
332: Group 2: $T(\mu,\nu,0,\rho,\sigma)=0$. \\
333: The rules of this group conserve them.\\
334: Group 3: $T(\mu,\nu,i,\rho,\sigma)=0$.\\
335: The rules of this group develop them.\\
336: Group 4: $T(\mu,\nu,i,\rho,\sigma)=l$. \\
337: The rules in this group do not affect string of quiescent states in next
338: time step.\\
339: \indent
340: The sum of the numbers of the group 1 and group 2 rules is 256, while that of
341: group 3 and group 4 rules is 768. The number of each group of rules
342: included in the rule table is determined mainly by the probability
343: $\lambda$, therefore it suffers from fluctuation due to randomness.\\
344: \indent
345: The group 1 rules are further classified into five types according to
346: the length of string of quiescent states, which they break. These are
347: shown in Table \ref{destruc}.
348: \renewcommand{\arraystretch}{0.9}
349: \begin{table}[h]
350: \caption{ The classification of the rules in group 1.}
351: \vspace*{0.3cm}
352: \label{destruc}
353: \begin{center}
354: \begin{tabular}{|c|c|c|c|c|c|c|}
355: \hline
356: \multicolumn{1}{|c|}{type} &
357: \multicolumn{1}{|c|}{Total Number} &
358: \multicolumn{1}{|c|}{Name}&
359: \multicolumn{1}{|c|}{Replacement}\\
360: \hline
361: $T(0,0,0,0,0)=h$ &1 &D5 &RP5,RC5\\
362: \hline
363: $T(0,0,0,0,i)=h$ &3 &D4 &RP4,RC4\\
364: $T(i,0,0,0,0)=h$ &3 & & \\
365: \hline
366: $T(0,0,0,i,\sigma)=h$ &12 & & \\
367: $T(i,0,0,0,m)=h$ &9 &D3 &RP3,RC3\\
368: $T(\mu,j,0,0,0)=h$ &12 & & \\
369: \hline
370: $T(\mu,j,0,0,m)=h$ &36 &D2 &RP2,RC2 \\
371: $T(i,0,0,l,\sigma)=h$ &36 & & \\
372: \hline
373: $T(\mu,j,0,l,\sigma)=h$ &144 &D1 &RP1,RC1\\
374: \hline
375: \end{tabular}
376: \end{center}
377: \end{table}
378: \vspace{0.5cm}
379: \\
380: The D5 rule is always excluded from rule tables by the
381: quiescent condition.\\
382: \indent
383: Our hypothesis presented at the end of the
384: section 2 is expressed more quantitatively as follows;
385: the numbers of the D4, D3, D2, and D1 rules shown in Table
386: \ref{destruc} will mainly determine the pattern classes.\\
387: \indent
388: In order to test this hypothesis we artificially change the numbers of
389: these rules in Table \ref{destruc} while keeping
390: the $\lambda$ fixed.
391: For D4 rules, we carry out the replacements defined by the following
392: equations,
393: \begin{equation}
394: \begin{array}{ll}
395: T(0,0,0,0,i)=h \rightarrow T(0,0,0,0,i)=0, \\
396: or \hspace {0.1cm}T(i,0,0,0,0)=h \rightarrow T(i,0,0,0,0)=0, \\
397: T(\mu,\nu,j,\rho,\sigma)=0 \rightarrow T(\mu,\nu,j,\rho,\sigma)=l,\\
398: \label{toperio}
399: \end{array}
400: \end{equation}
401: where except for $h$, the groups $\mu$, $\nu$, $\rho$, $\sigma$, $j$ and $l$
402: are selected randomly.
403: Similarly the replacements are generalized for D3, D2, and D1 rules
404: , which are denoted as RP4 to RP1 in Table \ref{destruc}.
405: They change the rules of group 1 to
406: that of group 2 together with group 3 to group 4 and are expected
407: to push the rule table toward the periodic direction. \\
408: \indent
409: The reverse replacements for D4 are
410: \begin{equation}
411: \begin{array}{ll}
412: T(0,0,0,0,i)=0 \rightarrow T(0,0,0,0,i)=h, \\
413: or \hspace {0.1cm}T(i,0,0,0,0)=0 \rightarrow T(i,0,0,0,0)=h, \\
414: T(\mu,\nu,j,\rho,\sigma)=l \rightarrow T(\mu,\nu,j,\rho,\sigma)=0,\\
415: \label{tochaos}
416: \end{array}
417: \end{equation}
418: which will push the rule table toward the chaotic direction. In this case,
419: the groups $h$, $\mu$, $\nu$, $j$, $\rho$, and $\sigma$ are
420: selected randomly.
421: Similarly we introduce the replacements for D3, D2, and D1, which will
422: be called RC4 to RC1 in the following.\\
423: \indent
424: By the replacement of RP4 to RP1 or
425: RC4 to RC1,
426: we change the numbers of tfhe rules in Table
427: \ref{destruc} while keeping the $\lambda$ fixed. We denote these
428: numbers $N_4$, $N_3$, $N_2$, and $N_1$
429: for D4, D3, D2, and D1 rules, respectively.
430: The examples of the replacement experiments are shown in
431: Fig.\ref{replace1}.\\
432: \begin{figure}
433: \begin{center}
434: \scalebox{0.58}{ { \includegraphics{Fig2.eps} } }
435: \caption{
436: Example of the replacement experiments at $\lambda=0.6$.
437: Fig.\ref{replace1}(a) is obtained randomly with probability
438: $\lambda=0.6$, by the method explained in subsection 2.1.
439: Fig.\ref{replace1}(b) to Fig.\ref{replace1}(h) are obtained by the
440: replacements of the rules
441: of Fig.\ref{replace1}(a), which are summarized in Table \ref{table2}.
442: }
443: \label{replace1}
444: \end{center}
445: \end{figure}
446: \renewcommand{\arraystretch}{0.9}
447: \begin{table}[h]
448: \begin{center}
449: \caption{ The numbers of the rules and the replacements for each
450: figures shown in Fig.\ref{replace1}}
451: \label{table2}
452: \vspace*{0.3cm}
453: \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|}
454: \hline
455: \multicolumn{1}{|c|}{Figure} &
456: \multicolumn{1}{|c|}{$N_4$}&
457: \multicolumn{1}{|c|}{$N_3$}&
458: \multicolumn{1}{|c|}{$N_2$}&
459: \multicolumn{1}{|c|}{$N_1$}&
460: \multicolumn{1}{|c|}{RP4}&
461: \multicolumn{1}{|c|}{RP3}&
462: \multicolumn{1}{|c|}{RP2}&
463: \multicolumn{1}{|c|}{RP1}\\
464: \hline
465: Fig.\ref{replace1}(a) &3 &22 &53 &96 &0 &0 &0 &0\\
466: \hline
467: Fig.\ref{replace1}(b) &0 &14 &53 &96 &3 &8 &0 &0\\
468: \hline
469: Fig.\ref{replace1}(c) &0 &13 &53 &96 &3 &9 &0 &0\\
470: \hline
471: Fig.\ref{replace1}(d) &0 &12 &53 &96 &3 &10 &0 &0\\
472: \hline
473: Fig.\ref{replace1}(e) &1 &7 &53 &96 &2 &15 &0 &0\\
474: \hline
475: Fig.\ref{replace1}(f) &1 &6 &53 &96 &2 &16 &0 &0\\
476: \hline
477: Fig.\ref{replace1}(g) &2 &1 &53 &96 &1 &21 &0 &0\\
478: \hline
479: Fig.\ref{replace1}(h) &2 &0 &53 &96 &1 &22 &0 &0\\
480: \hline
481: \end{tabular}
482: \end{center}
483: \vspace{0.5cm}
484: \vspace{0.5cm}
485: \end{table}
486: \indent
487: The rule table of Fig.\ref{replace1}(a) is obtained randomly with probability
488: $\lambda=0.6$. At $\lambda=0.6$ most of the randomly created rule tables
489: generate chaotic patterns. Fig.\ref{replace1}(b) is obtained by
490: the 3 RP4s and 8
491: RP3s, and the numbers of the rules become $N_4=0$ and $N_3=14$,
492: respectively. These numbers are summarized in Table \ref{table2}.
493: By this replacements, Fig.\ref{replace1}(b) shows an edge of chaos behavior.
494: When $N_4=0$, until $N_3=22$ to $N_3=15$, the rule tables generate
495: chaotic patterns.
496: Fig.\ref{replace1}(c) is obtained by one more RP3 replacements for
497: Fig.\ref{replace1}(b) rule table. It shows a
498: periodic pattern with a rather long transient length. One more
499: replacement of RP3 for the Fig.\ref{replace1}(c) rule table is shown in
500: Fig.\ref{replace1}(d), where the transient length
501: becomes shorter.\\
502: \indent
503: Similar replacement experiments for $N_4=1$ and
504: $N_4=2$ cases are shown in Fig.\ref{replace1}(e), Fig.\ref{replace1}(f), and
505: Fig.\ref{replace1}(g), Fig.\ref{replace1}(h), respectively.
506: In these cases, until $N_3 \ge 8$ and $N_3 \ge 2$, rule tables generate
507: class III patterns at $N_4=1$ and $N_2=2$, respectively.
508: In these examples, edges of chaos are observed between classes III and
509: II behaviors.\\
510: \indent
511: For each $\lambda=0.8$, $0.75$, $0.7$, $0.6$, $0.5$, $0.4$, $0.3$
512: and $0.2$ point, we have carried out
513: a several hundreds to a few thousands replacement experiments.
514: In these replacements, we have
515: succeeded in changing chaotic rule tables to a periodic ones by the
516: replacements of RP4 to RP2, and vice versa by RC4 to RC2.
517: For the replacements RP4, RP3 and RP2, we have observed no example that
518: the rule
519: table moves to chaotic direction. The converse is true for RC4, RC3 and RC2.
520: Exceptions are observed only in the replacements RP1 and RC1, which
521: will be discussed in section 4. \\
522:
523: \subsection{Chaotic and periodic limit at fixed $\lambda$ in
524: N-neighbor and K-state CA}
525: Let us study the effects of the replacements theoretically in the
526: N-neighbor and K-state CA.
527: In this general case too, the rule tables are classified into
528: four groups.
529: We denote the number of the group 1 rules as $N(g1)$, and similarly for the
530: numbers of the other groups.
531: These numbers satisfy the following sum rules.
532: \begin{equation}
533: \begin{array}{ll}
534: N(g2)+N(g3)=K^{N}(1-\lambda), \hspace {0.5cm}
535: N(g1)+N(g4)=K^{N}\lambda, \\
536: N(g1)+N(g2)=K^{N-1}, \hspace {1.4cm}
537: N(g3)+N(g4)=K^{N-1}(K-1).
538: \label{sum_rule1}
539: \end{array}
540: \end{equation}
541: \indent
542: The individual numbers $N(gi)$ are determined by the probability
543: $\lambda$. They are summarized in the Table \ref{N-K}.
544: They suffer from fluctuations due to
545: random number, however in this subsection, we neglect the fluctuations.\\
546: \renewcommand{\arraystretch}{0.9}
547: \begin{table}[h]
548: \caption{ The classification of N-neighbor and K-state CA rules into
549: four groups. The $\mu_{i}$ represent 0 to K-1 and $h$,$i$ and $l$,
550: 1 to K-1.}
551: \vspace*{0.3cm}
552: \label{N-K}
553: \begin{center}
554: \begin{tabular}{|c|c|c|c|c|c|c|}
555: \hline
556: \multicolumn{1}{|c|}{ } &
557: \multicolumn{1}{|c|}{rule } &
558: \multicolumn{1}{|c|}{N(gi)}\\
559: \hline
560: group 1 &$T(\mu_1,\mu_2,...,0,...,\mu_{N})=h$ &$K^{N-1}\lambda$\\
561: \hline
562: group 2 &$T(\mu_1,\mu_2,...,0,...,\mu_{N})=0$ &$K^{N-1}(1-\lambda)$\\
563: \hline
564: group 3 &$T(\mu_1,\mu_2,...,i,...,\mu_{N})=0$ &$K^{N-1}(K-1)(1-\lambda)$\\
565: \hline
566: group 4 &$T(\mu_1,\mu_2,...,i,...,\mu_{N})=l$ &$K^{N-1}(K-1)\lambda$\\
567: \hline
568: \end{tabular}
569: \end{center}
570: \end{table}
571: \vspace{0.5cm}
572: \indent
573: The replacements to decrease the number of the group 1 rule
574: while keeping the $\lambda$ fixed are given by,
575: \begin{equation}
576: \begin{array}{ll}
577: N(g1) \rightarrow N(g1)-1,\hspace {0.5cm} N(g2) \rightarrow N(g2)+1\\
578: N(g3) \rightarrow N(g3)-1,\hspace {0.5cm} N(g4) \rightarrow N(g4)+1\\
579: \label{del_num_RP}
580: \end{array}
581: \end{equation}
582: \indent
583: In the case of 5-neighbor and 4-state CA, they correspond to
584: RP4 to RP1.\\
585: \indent
586: These replacements stop either when $N(g1)=0$ or $N(g3)=0$ is reached.
587: Therefore when $N(g1) \le N(g3)$, which corresponds to $\lambda \le
588: (1-\frac{1}{K})$ in $\lambda$,
589: all the group 1 rules are replaced by the group 2 rules.
590: In this limit, quiescent
591: states at time $t$ will never be changed, because there is no rule
592: which
593: converts them to other states, while the group 3 rules have a chance to
594: create a new quiescent state in the next time step. Thus
595: the number of quiescent states at time t is a non-decreasing function
596: of t;
597: therefore, the pattern class should be class I (homogeneous) or
598: class II (periodic), which we call periodic limit. Therefore
599: the replacements of Eq.\ref{del_num_RP} push the rule table toward the
600: periodic limit.\\
601: \indent
602: Let us discuss the reverse replacements of Eq.\ref{del_num_RP}.
603: In these replacements, if $N(g2) \le N(g4)$,
604: all group 2 rules are replaced by
605: the group 1 rules, except for the quiescent condition.
606: In this extreme reverse case,
607: all the quiescent states at time t are
608: converted to other states in next time step, while group 3 rules
609: create them at different places.
610: Then this will most probably develop into chaotic patterns. This limit
611: will be called chaotic limit, which is reached in the region
612: $\frac{1}{K} \le \lambda$.
613: We should like to say that
614: atypical rule table and initial condition might
615: generate a periodic patterns even in this limit. But in this article, these
616: exceptional cases are neglected.\\
617: \indent
618: Therefore in the following region,
619: \begin{equation}
620: \frac{1}{K} \le \lambda \le 1-\frac{1}{K},
621: \label{2-limit}
622: \end{equation}
623: all the rule tables are located between
624: these two limit, and by the
625: replacements of Eq. \ref{del_num_RP} and their reverse ones, the changes
626: of the pattern classes take place without fail.
627: This explains the validity of the hypothesis
628: of previous section.
629: And we have found a method to control the pattern classes at
630: fixed $\lambda$.
631: %The details of the replacements of Eq. \ref{del_num_RP}
632: %depend on the models. In the 5-neighbor and 4-state CA they have been
633: %RP4 to RP1 and RC4 to RC1.
634:
635: \section{Phase diagram of 5-neighbor and 4-state CA in $\lambda$-F plane}
636: In the previous section, we have found that rule table is located
637: somewhere between chaotic limit and periodic limit,
638: in the region $\frac{1}{K} \le \lambda \le 1-\frac{1}{K}$.
639: In order to express
640: the position of the rule table quantitatively, we introduce new
641: parameter F, which provides us with a new axis ($F$-axis) orthogonal to
642: $\lambda$. Minimum of $F$ is the periodic limit, while maximum of it
643: corresponds to chaotic limit. In this section, we
644: determine the parameter $F$, for 5-neighbor and 4-state CA.\\
645: \indent
646: As a first approximation, the parameter $F$ is taken to be
647: be a function of
648: the numbers of the rules D4, D3, D2 and D1, which have been denoted as
649: $N_4$, $N_3$, $N_2$ and $N_1$, respectively.
650: We proceed to
651: determine $F(N_4,N_3,N_2,N_1)$
652: by applying simplest approximations and assumptions\\
653: \indent
654: We have observed in the replacement experiments, that the position of the
655: rule table in $F$ moves toward chaotic direction, when $N_4$ or $N_3$ or $N_2$
656: increases. Examples are shown in Fig. \ref{replace1} and Table
657: \ref{table2}. However for D1, replacements RP1 and RC1 sometimes look
658: like random walk on $F$-axis, around the region where the transition of
659: the pattern class is taken place\footnote{
660: The effect of D1 rule is to change an isolates quiescent state to
661: other states in the next time step. This effects
662: may easily be compensated by the creation of
663: quiescent states by group 3 rules in one time step. This may
664: be a reason that
665: the replacements of RP1 and RC1 some times look like random walk.}.
666: Therefore $F$ will be a complicated function of $N_1$ and determination
667: of it will be difficult.\\
668: \indent
669: However the number $N_1$ is rather large,
670: therefore we apply mean field approximation for $N_1$.
671: We replace $N_1$ by its average, and measure $F$ from this
672: background; namely the $N_1$ dependence is dropped from $F$ and take
673: $F=0$ at $N_4=N_3=N_2=0$.\\
674: \indent
675: Here, we apply Taylor series expansion for $F$ at this point, and
676: approximate it by the linear terms in $N_4$, $N_3$ and $N_2$. \\
677: \begin{equation}
678: F(N_4,N_3,N_2) \simeq c_4 N_4 + c_3 N_3 + c_2 N_2
679: \label{taylor}.
680: \end{equation}
681: where $c_4= \partial F/\partial N_4$, similar for $c_3$ and $c_2$.
682: They represent the strength of the effects of the rules
683: D4, D3, and D2 to push the rule table toward chaotic direction.
684: This definition is symbolic, because $N_4$ is discrete.\\
685: \indent
686: The measure in the $F$ is still arbitrary. We fix it in the unit
687: where the increase in one unit of $N_4$ results in the change of $F$ in one
688: unit. This corresponds to divide $F$ in Eq.\ref{taylor} by $c_4$, and to
689: express it by the ratio $c_3/c_4$ ($r_3$) and $c_2/c_4$ ($r_2$).\\
690: \indent
691: Before we proceed to determine $r_3$ and $r_2$, let us interpret the
692: parameter $F$ geometrically. Most generally, the rule tables are
693: classified in 1024-dimensional space in this CA. The rule tables, at
694: the boundary of the class III and class II
695: pattern classes at fixed $\lambda$, form
696: a hyper-surface in this space.
697: We map the points on hyper-surface
698: into 3-dimensional $(N_4,N_3,N_2)$ space.
699: They will be located in some region in the
700: 3-dimensional space. We introduce a surface $F(N_4,N_3,N_2)=\Phi$
701: in order to line up these points. $F$-axis is a normal line
702: of the surface $F(N_4,N_3,N_2)=\Phi$.
703: In Eq.\ref{taylor}, we
704: approximate it by a plane. \\
705: \indent
706: Our strategy to determine $r_3$ and $r_2$
707: is to find the regression plane in $N_4$, $N_3$, and $N_2$ space.
708: It is equivalent to fix the $F$-axis in such a way that
709: the projection of the distribution of transition points on $F$-axis,
710: ($F_{crit}$) looks as narrow as possible.
711: The quality of our approximations and
712: assumptions reflects the width of the distribution of
713: $F_{crit}$.\\
714: \indent
715: In the least square method, our problem is formulated to find $r_3$
716: and $r_2$, which minimize the quantity,
717: \begin{equation}
718: s(r_3,r_2)=
719: \frac{1}{c_{4}^{2}}\sum_{i,j}(F_{crit}^i(N_4^i,N_3^i,N_2^i)
720: - F_{crit}^j(N_4^j,N_3^j,N_2^j))^{2} \label{ansatz},
721: \end{equation}
722: where $i$ and $j$ label the rule tables on the hyper-surface
723: mapped into 3-dimensional space.
724: We solve the equations,
725: $\partial S/\partial r_3=0$ and $\partial S/\partial r_2=0$,
726: which are
727: \begin{equation}
728: \begin{array}{llll}
729: \displaystyle
730: r_3 \sum_{i,j} (\delta N_3^{i,j})^2 +
731: r_2 \sum_{i,j} \delta N_2^{i,j}\delta N_3^{i,j}
732: =- \sum_{i,j} \delta N_4^{i,j}\delta N_3^{i,j}, \\
733: \displaystyle
734: r_3 \sum_{i,j} \delta N_3^{i,j} \delta N_2^{i,j} +
735: r_2 \sum_{i,j} (\delta N_2^{i,j})^2
736: =- \sum_{i,j} \delta N_4^{i,j}\delta N_2^{i,j}, \\
737: \label{sol_r3}
738: \end{array}
739: \end{equation}
740: where $\delta N_4^{i,j}=N_{4}^{i}-N_{4}^{j}$, similar for
741: $\delta N_3^{i,j}$ and $\delta N_2^{i,j}$.\\
742: \indent
743: By artificially carrying out the replacements of RP4, RP3 and RP2 or
744: RC4, RC3 and RC2,
745: we look for critical combinations of $N_4$, $N_3$, and $N_2$,
746: at which the change of the pattern classes are observed.
747: %These points are the samples of rule tables to determine
748: %$r_3$ and $r_2$ in Eq. \ref{sol_r3}.
749: The examples of the critical combinations are presented in the lines
750: of Fig.\ref{replace1}(b),
751: Fig.\ref{replace1}(e), and
752: Fig.\ref{replace1}(g) of Table \ref{table2}.\\
753: \indent
754: %The critical combinations are obtained for each
755: %$\lambda=$0.2, 0.3,0.4, 0.5,0.6, 0.7, 0.75 and 0.8 points.
756: The numbers of the critical combinations ($Nc^{tot}$), which are used
757: to determine $r_3$ and $r_2$ in Eq. \ref{sol_r3}, are summarized in
758: the Table \ref{relative_force}.
759: The $r_3$ and $r_2$ are determined for each $\lambda$.
760: Their results are also shown in Table \ref{relative_force},
761: where the errors are estimated by the jackknife method.\\
762: \indent
763: The results show that the coefficients are positive, and satisfy the
764: order,
765: \begin{equation}
766: c_4 > c_3 > c_2.
767: \label{order_force}
768: \end{equation}
769: It means that the effects to move the rule table toward chaotic limit
770: on the
771: $F$-axis are stronger for the rules which break longer strings of the
772: quiescent states. \\
773: \indent
774: The order in Eq.\ref{order_force} is understood
775: by the following intuitive arguments.
776: If six D4 rules are included in the rule table, the string of the
777: quiescent states with length 5 will not develop. Similarly,
778: if 33 D3 rules are present in the rule table, no length
779: 4 string of the quiescent states could be made. These are roughly
780: similar situations for pattern classes.
781: Thus the strength of the D3 rules will be roughly
782: equal to $6/33$ of that
783: of D4 rules, similar for the strength of the D2 and D1 rules
784: \footnote{If we do not impose the quiescent
785: condition, Eq.\ref{order_force} becomes
786: $$c_5 > c_4 > c_3 > c_2.$$
787: Therefore the correlation between
788: pattern classes and the existence of D5 rule is stronger than the
789: correlation between those and the number of D4 rules.
790: If we start our
791: study within the quiescent condition, we may make a longer
792: detour to
793: find the hypothesis of section 2 and get the qualitative conclusion of
794: section 3.}.
795: \renewcommand{\arraystretch}{0.9}\\
796: \begin{table}[t]
797: \caption{The number of the critical combinations $Nc^{tot}$ and the
798: coefficients $r_3$ and $r_2$.
799: \vspace*{0.3cm}}
800: \label{relative_force}
801: \begin{center}
802: \begin{tabular}{|c|c|c|c|c|c|c|}
803: \hline
804: \multicolumn{1}{|c|}{$\lambda$} &
805: \multicolumn{1}{|c|}{ } &
806: \multicolumn{1}{|c|}{Relative Strength} &
807: \multicolumn{1}{|c|}{Error}&
808: \multicolumn{1}{|c|}{$Nc^{tot}$}\\
809: \hline
810: 0.2& $r_{3}$ &0.1083 &0.0020 &119\\
811: & $r_{2}$ &0.0153 &0.0025 & \\
812: \hline
813: 0.3& $r_{3}$ &0.1083 &0.0028 &90\\
814: & $r_{2}$ &0.0069 &0.0021 & \\
815: \hline
816: 0.4& $r_{3}$ &0.1182 &0.0074 &58\\
817: & $r_{2}$ &0.0165 &0.0072 & \\
818: \hline
819: 0.5& $r_{3}$ &0.1023 &0.0047 &61\\
820: & $r_{2}$ &0.0262 &0.0010 & \\
821: \hline
822: 0.6& $r_{3}$ &0.1288 &0.0019 &101\\
823: & $r_{2}$ &0.0216 &0.0013 & \\
824: \hline
825: 0.7& $r_{3}$ &0.1631 &0.0063 &92\\
826: & $r_{2}$ &0.0342 &0.0042 & \\
827: \hline
828: 0.75& $r_{3}$ &0.1505 &0.0114 &92\\
829: & $r_{2}$ &0.0419 &0.0065 & \\
830: \hline
831: 0.8& $r_{3}$ &0.1315 &0.0168 &89\\
832: & $r_{2}$ &0.0420 &0.0038 & \\
833: \hline
834: \end{tabular}
835: \end{center}
836: \end{table}
837: \begin{figure}
838: \begin{center}
839: \scalebox{0.7}{ { \includegraphics{Fig3.eps} } }
840: \caption{ Distribution of critical combinations $F_{crit}$ at $\lambda=0.6$.
841: Fig.\ref{phase_diagram_0.6}a shows 101 individual $F_{crit}$ points.
842: %for individual critical combinations.
843: We label these 101 points by $Nc$.
844: In Fig.\ref{phase_diagram_0.6}b,
845: number distribution of the critical combinations in
846: the bin is displayed. The region of distribution of
847: $F_{crit}$ is divided into 6 bins.}
848: \label{phase_diagram_0.6}
849: \end{center}
850: \end{figure}
851: \indent
852: Using these results for $r_3$ and $r_2$, we
853: calculate $F$s for each critical combination ($F_{crit}$).
854: For $\lambda=0.6$, they are
855: shown in Fig.\ref{phase_diagram_0.6}a for 101 data points.
856: The number distributions of them are displayed in
857: Fig.\ref{phase_diagram_0.6}b.
858: From these 101 data, we calculate average,
859: $<F_{crit}>$ and standard
860: deviation $\sigma$ of the distribution of $F_{crit}$.
861: Similar calculations are carried out for other $\lambda$
862: points.
863: The results for $<F_{crit}>$, $<F_{crit}> \pm \sigma$ and maximum of
864: $F$ ($F_{max}$) determined by the coefficients $r_3$ and $r_2$
865: are shown in the Fig.\ref{phase_diagram}.
866: This is a phase diagram in $\lambda-F$ plane.\\
867: \indent
868: The phase diagram is natural in the sense that as $\lambda$ increases,
869: the proportion of the chaotic rule table region increases in the
870: total $F$ range.
871: The two
872: lines $<F_{crit}> \pm \sigma$ in the figure, indicate that
873: about $68\%$ of the critical combinations are located in this region, if
874: the normal distribution is assumed for $F_{crit}$. We call this region
875: as transition region. We think that the class II and
876: class III patterns would not be separated by a line in
877: $\lambda-F$ plane. Because the transition region in
878: Fig. \ref{phase_diagram} is a mapping of the hyper-surface, which
879: separates the two pattern classes at a fixed $\lambda$ in 1024
880: dimensional space, into the normal line of the regression plane
881: in the 3-dimensional space.
882: Therefore the two parameters will not be enough to draw the
883: boundary by a line.\\
884: \indent
885: However we should like to notice that in spite of the simple
886: approximations and assumptions,
887: the $F_{crit}$s distribute within a rather narrow portion of total
888: $F$ range as shown in Fig. \ref{phase_diagram}.
889: At $\lambda=0.6$, the total range of
890: $F$ is $ 0 \le F \le 11.8$.
891: All the $F_{crit}$ distribute within the region $1.5 \le F \le
892: 4.48$, which is roughly $1/4$ of the total region.
893: The region $<F_{crit}> \pm \sigma$ occupies only about $11\%$ of whole range
894: of $F$. The similar situation is observed
895: for the other $\lambda$ points.\\
896: \indent
897: The Fig.\ref{phase_diagram} is still qualitative but
898: rule table space of CA is classified much better by using
899: $\lambda$ and F. And it
900: provides us with a deeper understanding of the structure of the CA rule
901: table space.
902: The existence of the transition region in $0.2 \le
903: \lambda \le 0.8$, means that the edge of the chaos could be found
904: at least in this region, which is rather wide range in $\lambda$. \\
905: \begin{figure}
906: \begin{center}
907: \scalebox{0.9}{ { \includegraphics{Fig4.eps} } }
908: \caption{
909: Phase diagram of the 5-neighbor and 4-state cellular automata.
910: The $<F_{Crit}>/c_{4}$ and the $F_{max}/c_{4}$ are shown with filled circle.
911: The
912: $<F_{crit}>/c_{4} + \sigma$ is shown with down triangle and the
913: $<F_{crit}>/c_{4} - \sigma$, with up triangle. They are joined by lines.
914: A schematic sketch of $F_{min}$ in $0.75 \le \lambda$, and $F_{max}$ in
915: $\lambda \le 0.25$ are also shown by the
916: dotted lines. They are drawn based on the arguments in the text.
917: }
918: \label{phase_diagram}
919: \end{center}
920: \end{figure}
921: %\begin{figure}
922: %\begin{center}
923: %\scalebox{0.7}{ { \includegraphics{Fig5.eps} } }
924: %\caption{ Distribution of class III(filled circle) and class II(open circle)
925: %rule tables around transition region.
926: %Fig.\ref{c-p-distribution}a shows 137 chaotic and 147 periodic rule tables
927: %at $\lambda=0.6$.
928: %They are labeled by $Nc$.
929: %In Fig.\ref{c-p-distribution}b,
930: %number distribution of the chaotic rule tables in
931: %the bin is displayed by solid line, and that of periodic rule tables by
932: %dotted line. The range of $F$ and the bin size are same as
933: %those of Fig. \ref{phase_diagram_0.6}.}
934: %\label{c-p-distribution}
935: %\end{center}
936: %\end{figure}
937: \indent
938: Let us discuss the distribution of the rule tables around the
939: transition region.
940: All the class IV rule tables are located around transition region
941: as Fig. \ref{phase_diagram_0.6} shows,
942: but converse is not true.
943: In this region, the three pattern classes coexist and
944: the number densities of the
945: class II plus class III rule table are considerably
946: larger than that of class IV rule tables.
947: Outside of the transition region, the probability of finding the class IV
948: rule table decreases rapidly and
949: the pattern classes began to be
950: classified only by the position in the $\lambda-F$ plane;
951: in the $<F_{crit}>+\sigma \le F$ region, class III rule tables
952: dominates
953: while in the $F \le <F_{crit}>-\sigma$ region, class II or class I
954: CA dominates.
955: The similar distributions of the rule tables are observed at
956: $\lambda=0.8$, 0.75, 0.7, 0.5, 0.4, 0.3 and 0.2.\\
957: \indent
958: We proceed to the investigation of the end points of the transition region.
959: In the subsection 3.2, we have find the region of $\lambda$, where
960: both chaotic limit and periodic limit coexist.
961: In the 5-neighbor and K-state CA, it is $1/4 \le \lambda \le 3/4$.
962: %But we have already found that the transition region do exist in wider
963: %$\lambda$ region.\\
964: \indent
965: In $\lambda < 0.25$ region, not all the group 2 rules could be replaced
966: by the group 1 rules. Therefore the maximum of $N(g1)$
967: could not become 256, and it deceases to zero as $\lambda$ approaches to
968: zero. Then the maximum of $F$, ($F_{max}$) also decreases to
969: zero toward $\lambda=0$.\\
970: \indent
971: Conversely in $\lambda > 0.75$ region, not all the group 1 rules could be
972: replaced by the group 2 rules. The minimum of
973: N(g1) and therefore the minimum of $F$, ($F_{min}$) could not becomes 0.
974: The line $F_{min}$ increases until $\lambda=1$.
975: In Fig.\ref{phase_diagram}, we have schematically shown the $F_{max}$
976: and $F_{min}$ lines with dotted lines.\\
977: \indent
978: The points where the transition region crosses the $F_{min}$ and
979: $F_{max}$ lines determine the end points of the coexistence of the two
980: limits, and also the end points of existence of the class IV
981: rule tables.
982: We have already found that the transition region is found in the region
983: $0.2 \le \lambda \le 0.8$. Therefore the crossing points are
984: outside of this region in the case of 5-neighbor 4-state CA.
985: \\
986: \section{Conclusions and discussions}
987: At a fixed $\lambda$, the patten classes of the CA could
988: be controlled by the numbers of the
989: group 1 rules, which has been denoted by $N(g1)$.
990: The maximum of $N(g1)(=K^{N-1}-1)$ corresponds to chaotic limit,
991: and $N(g1)=0$
992: , to the periodic limit.
993: In the N-neighbor and K-state CA, these two limit exist
994: in the region $\frac{1}{K} \le \lambda \le 1-\frac{1}{K}$.
995: Therefore in this
996: $\lambda$ region, we could control the patten classes by changing $N(g1)$
997: without fail. The method for it, is the replacements of
998: Eq. \ref{del_num_RP}.
999: This provides us with a
1000: new method to obtain the rule table of edge of chaos.
1001: However, we should like to comment that the fluctuation due to the
1002: random number will make the statements less rigorous. \\
1003: \indent
1004: This property is studied quantitatively by introducing a new
1005: parameter $F$.
1006: The maximum of the $F$ corresponds to
1007: the chaotic limit and minimum of it,
1008: the periodic limit. By changing the $F$ using the replacements, we
1009: could find the region, where the transition of
1010: the class II and class III patterns takes place. Thus we could
1011: obtain the phase diagram of the CA in $\lambda-F$ plane.\\
1012: \indent
1013: In this article, we have applied the analysis to 5-neighbor and
1014: 4-state CA. In this case the group 1 rules are further classified into
1015: 5 types as shown in Table \ref{destruc}, and the phase
1016: diagram is obtained in Fig.\ref{phase_diagram}.
1017: The transition region
1018: has a rather gentle
1019: slope as a function of $\lambda$, and it extend at least from
1020: $\lambda=0.2$ until $\lambda=0.8$, which is wider than
1021: $\frac{1}{4} \le \lambda \le 1-\frac{1}{4}$.
1022: This explains why the edge of chaos is found in the wide range
1023: in $\lambda$ for this CA.
1024: It may be interesting whether or not the slope of the transition region
1025: depends on the models.\\
1026: \indent
1027: In the replacement experiments,
1028: we have found the edge of chaos (very long transient lengths) in many
1029: cases. The examples are shown in Fig.\ref{replace1}.
1030: Sometimes they are
1031: observed in a rather wide range in $N_3$ or $N_2$.
1032: This indicates that in
1033: many cases, the transitions are second-order like. But the widths in the
1034: ranges of $N_3$ or $N_2$ are different from each other, and there are
1035: cases where the widths
1036: are less than one unit in the replacement of RP2 (first-order like).
1037: It is very interesting to investigate under what condition
1038: the transition becomes first-order like or second-order like.
1039: The mechanism of the difference in the nature of the transition is an open
1040: problem and
1041: may be studied by taking into account effects of group 2, 3, and 4
1042: rules. In these studies another new parameters may be found and a more
1043: quantitative phase diagram may be obtained.\\
1044: \indent
1045: It is also interesting to compare the detailed dynamics of class IV CA at
1046: different points of rule table space in $\lambda-F$ plane by applying
1047: Wuencshe's method\cite{wuensche} or by using computational
1048: mechanics\cite{hanson}.\\
1049: \indent
1050: These issues together with finding the points where the transition
1051: region crosses $F_{max}$ and $F_{min}$ lines in Fig.\ref{phase_diagram},
1052: and the nature of the phase transition at these points will be addressed
1053: in the forthcoming publications.
1054:
1055: \begin{thebibliography}{9}
1056: \bibitem{wolfram} S. Wolfram, Physica D 10(1984) 1-35.
1057: \bibitem{wolframs} S. Wolfram, Physica Scripta T9(1985) 170-185.
1058: \bibitem{langton} C.G.Langton, Physica D 42(1990) 12-37.
1059: \bibitem{mitchell} M. Mitchell, J.P.Crutchfield and P.T. Hraber,
1060: Santa Fe Institute Studies in the Science of Complexity, Proceedings
1061: Volume 19. Reading, MA:, Addison-Wesley.
1062: online paper, http://www.santafe.edu/~mm/
1063: paper-abstracts.html\#dyn-comp-edge.
1064: \bibitem{hanson} J.E.Hanson and J.P.Crutchfield, Physica D
1065: 103(1997),169-189.
1066: \bibitem{wuensche} A. Wuensche, Complexity Vol.4/No.3(1999) 47-66.
1067: \bibitem{langton0} C.G.Langton, Physica D 22(1986) 120-149.
1068: \bibitem{langton2} W.LI, N.H.Packard and C.G.Langton, Physica
1069: D45(1990) 77-94.
1070: \bibitem{suzudo} T. Suzudo, Crystallisation of Two-Dimensional
1071: Cellular Automata, Complexity International, Vol. 6(1999).
1072: on line journal,\\ http://www.csu.edu.au/ci/vol06/suzudo/suzudo.html.
1073: See appendix.
1074: \end{thebibliography}
1075: \end{document}
1076: \end
1077: