1: \documentclass[12pt]{article}
2: \usepackage{graphicx}
3: \usepackage{amsmath, amssymb,bbm}
4: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
5: %%%%%%%%%THIS IS ONLY FOR SCREEN PDF
6: % \usepackage[screen,panelright,gray,paneltoc]{pdfscreen}
7: % \margins{0.75in}{0.75in}{0.75in}{0.75in}
8: % \screensize{6.25in}{8in}
9: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
10: % Package hfstyle just defines address and abstract fields, changes
11: % margins, and modifies section headings a bit, but does not do
12: % anything "unorthodox". The paper will compile without this package,
13: % if you remove \oneaddress and \Abstact, and replace them with standard
14: % latex constructs
15: \usepackage{hfstyle}
16:
17: \newtheorem{proposition}{Proposition}
18: \newtheorem{conjecture}{Conjecture}
19: \newtheorem{theorem}{Theorem}
20: \newcommand{\ZZ}{{\mathbbm{Z}}}
21: \newcommand{\NN}{{\mathbbm{N}}}
22: \newcommand{\RR}{{\mathbbm{R}}}
23:
24: \begin{document}
25: \author{Henryk Fuk\'s
26: \oneaddress{
27: Department of Mathematics\\
28: Brock University\\
29: St. Catharines, Ontario L2S 3A1, Canada\\
30: \texttt{hfuks@brocku.ca}}}
31: \title{Second order additive invariants in elementary cellular automata}
32: \Abstract{
33: We investigate second order additive invariants in elementary cellular automata rules.
34: Fundamental diagrams of rules which possess additive invariants are either linear or exhibit
35: singularities similar to singularities of rules with first-order invariant. Only rules which have exactly one
36: invariants exhibit singularities. At the singularity, the current decays to its equilibrium value as a power law $t^{\alpha}$,
37: and the value of the exponent $\alpha$ obtained from numerical simulations is very close
38: to $-1/2$. This is in agreements with values previously reported for number-conserving rules,
39: and leads to a conjecture that regardless of the order of the invariant,
40: exponent $\alpha$ seems to have a universal value of $1/2$.
41: } \maketitle
42:
43: \section{Introduction}
44: Cellular automata (CA) are often described as systems of cells in a regular lattice
45: updated synchronously according to a local interaction rule.
46: An interesting subclass of CA consists of rules possessing an additive
47: invariant. The simplest of such invariants is the total number of
48: sites in a particular state. CA with such invariant, often called
49: ``conservative CA'' or ``number-conserving CA'',
50: generated a lot of interest in recent years
51: \cite{Pivato02,Moreira03,Durand2003,Formenti2003,Morita2001}.
52: Number-conserving CA can be viewed as a
53: system of interacting and moving particles, where in the case of a
54: binary rule, 1's represent sites occupied by particles, and 0's
55: represent empty sites. The flux or current of particles in equilibrium
56: depends only on their density, which is invariant. The graph of the
57: current as a function of density characterizes many features of the
58: flow, and is therefore called the fundamental diagram.
59:
60: For a majority of number-conserving CA rules, fundamental diagrams are piecewise-linear,
61: usually possessing one or more ``sharp corners'' or singularities.
62: There exist a strong evidence of universal behavior at singularities,
63: as reported in \cite{paper19,paper25}.
64:
65: Since number-conserving CA are simplest rules with additive invariants,
66: it would be interesting to consider higher order invariants and CA rules
67: with such invariants. In 1991, Hattori and Takesue performed
68: extensive study of additive invariants in discrete-time lattice
69: dynamical systems \cite{Hattori91}, not necessarily restricted to CA. They derived
70: very general existence conditions, and applied them to elementary CA
71: rules as well as reversible CA. In this paper, we will use their results to study
72: second-order invariants in elementary rules, focusing mainly on
73: fundamental diagrams.
74:
75:
76: \section{Number-conserving cellular automata}
77:
78: In what follows, we will assume that the dynamics takes place on
79: one-dimensional lattice of length $L$ with periodic boundary
80: conditions. Let $s_i(t)$ denote the state of the lattice site $i$ at
81: time $t$, where $i\in \{ 0, 1, \ldots, L - 1 \}$, $t \in \mathbbm{N}$.
82: All operations on spatial indices $i$ are assumed to be modulo $L$. We
83: will further assume that $s_i(t) \in \{ 0, 1\}$, and we will say that
84: the site $i$ is occupied (empty) at time $t$ if $s_i ( t ) = 1$ ($s_i
85: ( t ) = 0$).
86:
87: Let $l$ and $r$ be two integers such that $l \leq 0 \leq r$, and let
88: $n =r-l+1$. The set $\{ s_{i + l} ( t ), s_{i + l + 1} ( t ), \ldots,
89: s_{i + r} ( t )\}$ will be called the \textit{neighbourhood} of the
90: site $s_i ( t )$. Let $f$ be a function $f : \{ 0, 1 \}^n \rightarrow
91: \{ 0, 1 \}$, also called a \emph{local function} The update rule for
92: the cellular automaton is given by
93: \begin{equation}
94: \label{cadef} s_i ( t + 1 )_{} = f ( s_{i + l} ( t ), s_{i + l + 1} ( t ),
95: \ldots, s_{i + r} ( t ) ) .
96: \end{equation}
97:
98: In \cite{Hattori91}, the concept of additive invariant for CA
99: has been introduced. Let $\alpha$ be a non-negative integer, and let
100: $\xi=\xi(x_0,x_1,\ldots, x_{\alpha})$ be a function of $\alpha +1$
101: variables taking values in~$\RR$. We say that $\xi$
102: is a density function of an additive conserved quantity
103: if for every positive integer $L$ and for every
104: initial condition $(s_0(0),s_1(0),\ldots,s_{L-1}(0)) \in \{ 0, 1 \}^L$
105: we have
106: \begin{equation} \label{defadditive}
107: \sum_{i=0}^{L-1} \xi(s_i(t),s_{i+1}(t),\ldots, s_{i+\alpha}(t))=
108: \sum_{i=0}^{L-1} \xi(s_i(t+1),s_{i+1}(t+1),\ldots, s_{i+\alpha}(t+1))
109: \end{equation}
110: for all $t \in \NN$.
111: For simplicity, if the above condition is satisfied, we will say that $\xi$
112: is an additive invariant of $f$. It is often more convenient to write (\ref{defadditive})
113: using the function $G$ defined as
114: \begin{equation}
115: G(x_0,x_1,\ldots,x_{\alpha+n-1})=
116: \xi(f(x_0,x_1,\ldots,x_{n-1}), f(x_1,x_2,\ldots,x_{n}),\ldots,
117: f(x_\alpha,x_{\alpha+1},\ldots,x_{\alpha + n-1})).
118: \end{equation}
119: With this notation, $\xi$ is an additive invariant of $f$ if
120: \begin{equation}
121: \sum_{i=0}^{L-1} G(x_i,x_{i+1},\ldots,x_{i+\alpha+n-1})=
122: \sum_{i=0}^{L-1} \xi(x_i,x_{i+1},\ldots, x_{i+\alpha})
123: \end{equation}
124: for every positive integer $L$ and for all $x_0,x_1,\ldots, x_{L-1} \in \{0,1\}$.
125:
126: In recent years, many authors studied the case of the simplest additive invariant,
127: with $\alpha=0$ and $\xi(x_0)=x_0$. For this invariant, the equation (\ref{defadditive})
128: becomes
129: \begin{equation}
130: \sum_{i=0}^{L-1} s_i(t)=
131: \sum_{i=0}^{L-1} s_i(t+1),
132: \end{equation}
133: which means that the CA rule posessing this invariant conserves the number
134: of sites in state~1. Such rules are often referred to as number-conserving
135: rules. Among elementary CA, i.e. those with $l=-1$ and $r=1$, there are only
136: five number-conserving rules. Three of these are trivial, namely the identity
137: rule 204 and two shifts 170 and 240. Two remaining rules, 184 and 226, are equivalent under the
138: spatial reflection. Rule 184, which is a discrete version
139: of the totally asymmetric exclusion process, has been extensively studied
140: \cite{Krug88,Nagatani95,Nagel96,Belitsky98,paper11,NishinariT98,BelitskyKNS01,Blank03},
141: and many rigorous result regarding its dynamics have been established.
142:
143: Hattori and Takesue \cite{Hattori91} established a very general
144: result which we will write here in a somewhat simplified form, taking into account that
145: this paper is concerned with binary rules only.
146:
147: \begin{theorem}[Hattori \& Takesue '91]
148: Let $\xi(x_0,x_1,\ldots,x_\alpha)$ be a function of $\alpha + 1$ variables.
149: Then $\xi$ is a density function of an additive conserved quantity
150: under the time evolution of cellular automaton rule (\ref{cadef})
151: if and only if the condition
152: \begin{equation} \label{gencondition}
153: G(x_0,x_1,\ldots,x_{\alpha+n-1})-\xi(x_{-l},x_{-l+1},\ldots,x_{\alpha-l})
154: =J(x_0,x_1,\ldots,x_{\alpha+n-2})
155: -J(x_1,x_2,\ldots,x_{\alpha+n-1})
156: \end{equation}
157: holds for all $x_0$, $x_1$, $\ldots$, $x_{\alpha+n-1} \in \{0,1\}$, where
158: the quantity $J$, to be referred to as the \emph{current}, is defined by
159: \begin{equation} \label{gencurrentdef}
160: J(x_0,x_1,\ldots,x_{\alpha+n-2})=
161: -\sum_{i=0}^{\alpha+n-2} G(\overbrace{0,0,\ldots,0}^{\alpha+n-1-i},x_0,x_1,\ldots,x_i)
162: +\sum_{i=-l-n+2}^{\alpha-l} \xi(\overbrace{0,0,\ldots,0}^{\alpha+1-i},x_0,x_1,\ldots,x_{i-1}).
163: \end{equation}
164: \end{theorem}
165:
166: The following convention is used in the definition of $J$:
167: \begin{equation}
168: \xi(\overbrace{0,0,\ldots,0}^{\alpha+1-i},x_0,x_1,\ldots,x_{i-1})=
169: \xi(x_{i-\alpha-1},x_{i-\alpha},\ldots,x_{i-1}) \mbox{\,\,\,\,\,if $i \geq \alpha+1$},
170: \end{equation}
171: and
172: \begin{equation}
173: \xi(\overbrace{0,0,\ldots,0}^{\alpha+1-i},x_0,x_1,\ldots,x_{i-1})=\xi(0,0,\ldots,0)
174: \mbox{\,\,\,\,\,if $i \leq 0$}.
175: \end{equation}
176:
177:
178: The equation (\ref{gencondition}) can be interpreted in a similar way as a
179: conservation law in a continuous, one dimensional physical system. In
180: such system, let $\rho(x,t)$ denote the density of some material at
181: point $x$ and time $t$, and let $j(x,t)$ be the current (flux) of this
182: material at point $x$ and time $t$. A conservation law states that the
183: rate of change of the total amount of material contained in a fixed
184: domain is equal to the flux of that material across the surface of the
185: domain. The differential form of this condition can be written as
186: \begin{equation}\label{concons}
187: \frac{\partial \rho}{\partial t}=-\frac{\partial j}{\partial
188: x}.
189: \end{equation}
190: Since in our case $\xi$ is the the density of an additive conserved quantity, the left hand side of
191: (\ref{gencondition}) is simply the change of density in a
192: single time step, so that (\ref{gencondition}) is an obvious
193: discrete analog of the current conservation law (\ref{concons})
194: with $J$ playing the role of the current.
195:
196:
197: Let us now assume that the initial configuration has been generated
198: from some translation-invariant distribution $\mu$. We define the expected value of $\xi$ at site $i$ as
199: \begin{equation}
200: \rho(i,t)=E_\mu \left[ \xi(s_i(t),s_{i+1}(t),\ldots,s_{i+\alpha}(t)) \right].
201: \end{equation}
202: Since the initial distribution is $i$-independent, we expect that
203: $\rho(i,t)$ also does not depend on $i$, and we will therefore define
204: $\rho(t)=\rho(i,t)$. Furthermore, since $\xi$ is density function of a conserved
205: quantity, $\rho(t)$ is $t$-independent, so we define $\rho=\rho(t)$.
206: The expected value of the
207: current $J(s_{i+l}(t),s_{i+l+1}(t),\ldots,s_{i+r-1}(t))$ will also be
208: $i$-independent, so we can define the expected current as
209: \begin{equation} \label{defexpcurrent}
210: j(\rho,t)=E_\mu
211: \big(J(s_{i+l}(t),s_{i+l+1}(t),\ldots,s_{i+r-1}(t))\big).
212: \end{equation}
213:
214: The graph of the equilibrium current $j(\rho,\infty)=\lim_{t
215: \rightarrow \infty} j(\rho,t)$ versus the density $\rho$ is known as
216: the fundamental diagram.
217:
218:
219:
220: \section{Number-conserving nearest-neighbour rules}
221: In order to illustrate the theorem of the previous section, we will first consider
222: the case of number-conserving nearest-neighbour rules, i.e., $\alpha=0$ and $\xi(x_0)=x_0$,
223: $l=-1$, $n=3$. Condition (\ref{gencondition}) becomes
224: \begin{equation}
225: G ( x_0, x_1, x_2) - \xi(x_{1}) = J ( x_0, x_1)- J (x_1, x_2),
226: \end{equation}
227: where
228: \begin{equation}
229: J(x_0,x_1)=
230: % - \sum_{i=0}^{1}
231: %G(\underbrace{0,0,\ldots,0}_{2-i},x_0,x_1,\ldots,x_{i}) +
232: %\sum_{i=0}^{1} \xi(\underbrace{0,0,\ldots,0}_{1-i},x_0,x_1,\ldots,x_{i-1})
233: =-G(0,0,x_0)-G(0,x_0,x_1) + \xi(0)+\xi(x_0)
234: \end{equation}
235: Obviously, $G(x_0,x_1,x_2)=f(x_0,x_1,x_2)$, thus the current becomes
236: $J(x_0,x_1)=-f(0,0,x_0)-f(0,x_0,x_1) - x_0$, and the conservation condition
237: takes the form
238: \begin{equation} \label{nnsimpleconscond}
239: f( x_0, x_1, x_2) - x_{1} = J ( x_0, x_1)- J (x_1, x_2).
240: \end{equation}
241: As mentioned earlier, rule 184 and its spatial reflection are the only non-trivial
242: elementary CA rules satisfying (\ref{nnsimpleconscond}). For rule 184, $f$ is defined by
243: $f(x_0,x_1,x_2)=x_0-x_0 x_1 +x_1 x_2$
244: and the current can be written as $J(x_0,x_1)=x_0 (1-x_1)$.
245: It is possible to show \cite{paper11} that the equilibrium current for this rule is given by
246: \begin{equation}
247: j(\rho,\infty)= \left\{ \begin{array}{ll}
248: \rho & \mbox{if $\rho<1/2$}, \\
249: 1-\rho & \mbox{otherwise}.
250: \end{array}
251: \right.
252: \end{equation}
253:
254: Since number-conserving CA rules conserve the number of occupied
255: sites, we can label each occupied site (or ``particle'') with an
256: integer $k\in\mathbbm{Z}$, such that the closest particle to the right
257: of particle $k$ is labeled $k+1$. If $y_k(t)$ denotes the position of
258: particle $k$ at time $t$, the configuration of the particle system at
259: time $t$ is described by the increasing bisequence
260: $\{y_k(t)\}_{k=-\infty}^{\infty}$. We can then specify how the
261: position of the particle at the time step $t+1$ depends on positions
262: of the particle and its neighbours at the time step $t$. For example,
263: for rule 184 one obtains
264: \begin{equation} \label{r184-00}
265: y_k(t+1)= y_n(t) + \min\{y_{k+1}(t) - y_k(t)-1,1\}.
266: \end{equation}
267: Equation (\ref{r184-00}) is sometimes referred to as the motion
268: representation. The motion representation is analogous to Lagrange
269: representation of the fluid flow, in which we observe individual
270: particles and follow their trajectories \cite{MatsukidairaN03}.
271: It turns out that the motion representation can be constructed for
272: arbitrary number-conserving CA rule by employing algorithm described
273: in \cite{paper10}.
274:
275: \section{Second-order invariants}
276: In what follows, we will referr to the number of variables of $\xi$ as the order
277: of the invariant, equal to $\alpha+1$.
278: Since the invariant of $\alpha=0$ and corresponding fundamental diagrams have been extensively
279: studied, we will explore the case of $\alpha=1$, i.e., second order invariants,
280: using the method of \cite{Hattori91}.
281:
282: The arguments $x_0,x_1$ of the density function take values in the set $\{0,1\}$, and therefore $\xi$ can be defined
283: in terms of four parameters
284: \begin{equation}
285: \xi(0,0)=c_{00}, \,\,
286: \xi(0,1)=c_{01}, \,\,
287: \xi(1,0)=c_{10}, \,\,
288: \xi(1,1)=c_{11},
289: \end{equation}
290: where $c_{00},c_{01},c_{10},c_{11} \in \mathbbm{R}$. This can be also expressed as
291: \begin{eqnarray*}
292: \xi(x_0,x_1)=c_{00} (1-x_0)(1-x_1)+
293: c_{01} (1-x_0) x_1+
294: c_{10} x_0 (1-x_1)+
295: c_{11} x_0 x_1\\
296: =c_{00} + (c_{10}-c_{00})x_0 + (c_{01}-c_{00})x_1 + (c_{00}-c_{01}-c_{10}+c_{11})x_0 x_1.
297: \end{eqnarray*}
298: The constant term does not bring anything new, so we can set $c_{00}=0$. Moreover,
299: note that for any function $g(x)$ and any $x_0,x_1,\ldots,x_{L-1}\in \{0,1\}$
300: we have
301: \begin{equation}
302: \sum_{i=0}^{L-1}\xi(x_i,x_{i+1})=\sum_{i=0}^{L-1}\big(\xi(x_i,x_{i+1}) + g(x_i) - g(x_{i+1})\big ),
303: \end{equation}
304: which means that if $\xi(x_0,x_1)$ is a density function of some conserved additive quantity,
305: then $\widehat{F}(x_0,x_1)=\xi(x_0,x_1)+g(x_0)-g(x_1)$ is also a density function of a conserved additive
306: quantity. To remove this ambiguity, we will require that $\xi(0,x_1)=0$, similarly as done in~\cite{Hattori91}.
307: This yields $c_{01}=0$, and we are left with $\xi$ depending on two parameters only
308: \begin{equation}
309: \xi(x_0,x_1)=c_{10}x_0 + (c_{11}-c_{10})x_0 x_1.
310: \end{equation}
311: Defining $a_1=-c_{10}$, $a_2=c_{11}-c_{10}$, we arrive at the final parameterization of $\xi$
312: \begin{equation}
313: \xi(x_0,x_1)=a_1 x_0 + a_2 x_0 x_1, \mbox{\,\,\,} a_1,a_2 \in \mathbbm{R}.
314: \end{equation}
315:
316:
317: For $\alpha=1$, $l=-1$, and $n=3$, eq. (\ref{gencondition}) becomes
318: \begin{equation} \label{seconordcond}
319: G ( x_0, x_1, x_2,x_3) - \xi(x_{1},x_2) = J ( x_0, x_1, x_2)- J (x_1, x_2, x_3),
320: \end{equation}
321: where
322: \begin{eqnarray*}
323: G ( x_0, x_1, x_2,x_3)&=&\xi(f( x_0, x_1, x_2), f(x_1, x_2,x_3))\\
324: J ( x_0, x_1, x_2) &=& -G(0,0,0,x_0)-G(0,0,x_0,x_1) - G(0,x_0,x_1,x_2)\\
325: &&\mbox{\,}+\xi(0,0)+\xi(0,x_0)+\xi(x_0,x_1).
326: \end{eqnarray*}
327: Since $\xi(0,0)=\xi(0,x_0)=0$, the formula for current simplifies to
328: \begin{equation}
329: J ( x_0, x_1, x_2) = -G(0,0,0,x_0)-G(0,0,x_0,x_1) - G(0,x_0,x_1,x_2) +\xi(x_0,x_1).
330: \end{equation}
331: For a given elementary CA rule $f(x_0,x_1,x_2)$, one can write eq. (\ref{seconordcond})
332: for all $2^4$ combinations of values of the variables $x_0,x_1,x_2,x_3 \in \{0,1\}$, thus obtaining
333: an overdetermined linear system of 16 equations with two unknowns $a_1,a_2$.
334: This system is homogeneous, therefore the solution, if it exists, is not unique.
335: That is, if $(a_1,a_2)$ is a solution, then $(c a_1, c a_2)$ is also a solution for any $c \in \mathbbm{R}$.
336: We will normalize the solution so that the first non-zero number in the pair $(a_1,a_2)$ is
337: set to be equal to 1.
338:
339: Solving these equations for all ``minimal'' CA rules\footnote{Elementary CA rules fall into
340: 88 equivalence classes with respect to the group of transformations generated by the spatial
341: reflection and the Boolean conjugacy. Minimally-numbered element of each class are known as
342: ``minimal rules''.}, one finds that for most CA rules
343: solutions do not exist. Remaining CA rules can be divided into two classes.
344: The first class contains rules 204 and 170, and for these rules, any pair $(a_1,a_2)$
345: is a solution. We will not be concerned with these rules, since they exhibit trivial dynamics.
346: The second class consists of 10 rules for which a unique solution exists (up to the normalization described earlier).
347: These rules are 12, 14, 15, 34, 35, 42, 43, 51, 140, 142, and 200, as reported in \cite{Hattori91}.
348:
349: Table 1 shows the density function $\xi$ and the current $J$ for all of them. The formulas for the current
350: have been obtained using the HCELL C++ library for cellular automata developed by the author.
351: \begin{table}
352: \begin{center}
353: \begin{tabular}{|c|c|c|}\hline
354: Rule number & $\xi(x_0,x_1,x_2)$ & $J(x_0,x_1)$ \\ \hline
355: 12 & $ x_0-x_0 x_1$ & $ -x_0 x_1$ \\ \hline
356: 14 & $ x_0-x_0 x_1$ & $ -x_0 x_1$ \\ \hline
357: 15 & $ x_0-x_0 x_1$ & $ -x_0 x_1$ \\ \hline
358: 34 & $ x_0-x_0 x_1$ & $ -x_1$ \\ \hline
359: 35 & $ x_0-x_0 x_1$ & $ -x_1$ \\ \hline
360: 42 & $ x_0-x_0 x_1$ & $ -x_1+x_1 x_2-x_0 x_1 x_2$ \\ \hline
361: 43 & $ x_0-x_0 x_1$ & $ -x_1+x_1 x_2-x_0 x_1 x_2$ \\ \hline
362: 51 & $ x_0-x_0 x_1$ & $ -x_1$ \\ \hline
363: 140 & $ x_0-x_0 x_1$ & $ -x_0 x_1+x_0 x_1 x_2$ \\ \hline
364: 142 & $ x_0-x_0 x_1$ & $ -x_0 x_1+x_0 x_1 x_2$ \\ \hline
365: 200 & $ x_0 x_1 $ & $ 0$ \\ \hline
366: \end{tabular}
367: \end{center}
368: \caption{Density of the invariant $\xi$ and the current $J$ for all non-trivial elementary
369: CA with second order invariants.}
370: \end{table}
371:
372: \section{Fundamental diagrams}
373: In order to construct fundamental diagrams for rules of Table 1, we first note that
374: the current for rule 200 is identically equal to zero, thus the equilibrium current
375: $j(\rho, \infty)=0$. The graph of $j(\rho, \infty)$ vs. $\rho$ for this rule is, therefore,
376: not interesting.
377:
378: For all other rules, the density of the invariant is given by the same function
379: $\xi(x_0,x_1)= x_0-x_0 x_1$. This means that rules 12, 14, 15, 34, 35, 42, 43, 51, 140, and 142
380: conserve the number of blocks ``10'' in the configuration. In order to construct
381: their fundamental diagrams, we have to be able to create an initial configuration with a given
382: number of pairs ``10''. Construction of a configuration of length $L$ with exactly
383: $m$ pairs ``10'' can proceed according to the following algorithm.
384: We start with an array of $L$ integers, $\{s_i(0)\}_{i=0}^{L-1}$.
385: \begin{enumerate}
386: \item Set $s_i(0)=0$ for all $i=0,1,\ldots,L-1$.
387: \item Place the symbol ``C'' at randomly selected site of the array.
388: Then place another symbol ``C'' at another site randomly selected among all remaining empty
389: sites. Repeat this procedure until you place exactly $2m$ symbols ``C''.
390: \item Let $x=0$. Starting from $i=0$, traverse the array filling it with
391: $x$ values. Every time when you encounter $C$, set $x:=1-x$. Stop when you reach the end
392: of the array.
393: \end{enumerate}
394: The average density of the invariant $\xi(x_0,x_1)= x_0-x_0 x_1$ for the configuration obtained
395: with the above algorithm will be
396: \begin{equation}
397: \rho_{av} = \frac{1}{L}\sum_{i=0}^{L-1} \xi(s_i(t),s_{i+1}(t)) =\frac{m}{L},
398: \end{equation}
399: and it will be independent of $t$. We can also define average current at time $t$
400: for a configuration with average density $\rho_{av}$ of the invariant as
401: \begin{equation}
402: j_{av}(\rho_{av},t)=\frac{1}{L}\sum_{i=0}^{L-1} J(s_i(t),s_{i+1}(t),s_{i+2}(t)).
403: \end{equation}
404: Graph of $j_{av}(\rho_{av},t)$ vs. $\rho_{av}$ for very large $t$ will approximate the graph
405: of $j(\rho,\infty)$ (given by eq. \ref{defexpcurrent}) vs. $\rho$, i.e., the fundamental diagram.
406:
407: For six rules from Table 1, the fundamental diagram is strictly linear, and the following
408: expressions for current can be conjectured based on numerical experiments.
409: \begin{eqnarray*}
410: \mbox{Rule 12:\,\, } j(\rho,\infty)&=&0\\
411: \mbox{Rule 15:\,\, } j(\rho,\infty)&=&\rho - 1/2\\
412: \mbox{Rule 34:\,\, } j(\rho,\infty)&=&-\rho\\
413: \mbox{Rule 42:\,\, } j(\rho,\infty)&=&-\rho\\
414: \mbox{Rule 51:\,\, } j(\rho,\infty)&=&-1/2\\
415: \mbox{Rule 140:\,\, } j(\rho,\infty)&=&0
416: \end{eqnarray*}
417:
418: The remaining four rules are more interesting, as they exhibit singularities in fundamental
419: diagrams, as shown in Figure 1.
420: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
421: \begin{figure}
422: \begin{center}
423: \includegraphics[scale=0.55]{fig1.eps}
424: \end{center}
425: \caption{Fundamental diagrams for rules 14, 35, 43, 142. Singularities are denoted by green circles.
426: Diagrams have been obtained using 50000 lattice sites after 50000 iterations.}
427: \label{fig1}
428: \end{figure}
429: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
430: It is remarkable that singularities are present only in fundamental diagram of those particular
431: rules. In \cite{Hattori91}, authors searched for invariants of up to seventh order for all ``minimal''
432: elementary CA. According to the table published in their paper, rules $14, 35, 43, 142$ do not
433: have any other invariant except $\xi(x_0,x_1)=x_0-x_0 x_1$, in contrast to remaining rules of Table 1,
434: which also posses other higher order invariants. It seems that singularities in the fundamental diagram can appear
435: only in rules which have only one invariant, just like rule 184, which only has first order invariant
436: $\xi(x_0)=x_0$.
437:
438: \section{Convergence to equilibrium}
439: In number-conserving cellular automata, singularities of the fundamental diagram exhibit
440: critical behavior. This can be illustrated by introducing the the decay time defined as
441: \begin{equation} \label{tau}
442: \tau(\rho) = \sum_{t=0}^{\infty} |j(\rho,t)-j(\rho,\infty)|.
443: \end{equation}
444: If the decay of $j(\rho, t)$ toward its equilibrium value $j(\rho,\infty)$ is of power-law type, the above
445: sum diverges.
446: For all rules in Figure 1, we have performed computer simulations
447: to estimate $\tau$. The value of
448: $\tau$ has been estimated by measuring $j_{av}(\rho,t)$ for
449: $t=0,1,\ldots,1000$, and truncating the sum (\ref{tau}) at
450: $t=1000$. Figure 2 shows a typical graph of $\tau$ as a function of
451: $\rho$, obtained for rule 42. Comparing Figures 1c and Figure 2 we clearly see that $\tau$ diverges at
452: the critical point of rule 42, which occurs at $\rho=0.25$. We will denote this value by $\rho_c$.
453: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
454: \begin{figure}
455: \begin{center}
456: \includegraphics[scale=1.0]{fig2.eps}
457: \end{center}
458: \caption{Decay time vs. density of the invariant for rule 43.}
459: \label{fig2}
460: \end{figure}
461: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
462: Assuming that $|j(\rho_c,\infty)-j(\rho_c,t)| \sim t^{-\alpha}$, we have
463: determined the exponent $\alpha$ as the slope of the straight line
464: which best fits the logarithmic plot of $|j(\rho_c,\infty)-j(\rho_c,t)|$
465: vs. time $t$. Example of such a plot, again for rule 42, is shown in Figure 3.
466: Table 2 shows values of the exponent $\alpha$ for critical points of all
467: four rules of Figure 1.
468: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
469: \begin{figure}
470: \begin{center}
471: \includegraphics[scale=1.0]{fig3.eps}
472: \end{center}
473: \caption{Logarithmic plot of $|j(\rho_c,\infty)-j(\rho_c,t)|$ as a function of time for rule 43.
474: Data points ($+$) represent
475: computer simulations, while the dashed line represents the best
476: fit.}
477: \label{fig3}
478: \end{figure}
479: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
480: \begin{table}
481: \begin{center}
482: \begin{tabular}{|c|c|c|}\hline
483: Rule number & $(\rho_c,j(\rho_c,\infty))$ & $\alpha$ \\ \hline
484: 14 & $(1/4,-1/4)$ & $-0.504$ \\ \hline
485: 35 & $(1/3,-1/3)$ & $-0.472$\\ \hline
486: 43 & $(1/4,-1/4)$ & $-0.492$ \\ \hline
487: 142 & $(1/4,-1/4)$ & $-0.502$ \\ \hline
488: \end{tabular}
489: \end{center}
490: \caption{Values of the exponent $\alpha$ at the critical point for
491: elementary CA rules with second-order additive invariant.}
492: \end{table}
493: %
494:
495: Exponent $\alpha$ is known to be equal to exactly $1/2$ for rule 184 and its generalizations, and rigorous
496: proof of this fact exists \cite{paper11}.
497: Extensive numerical experiments support the conjecture that for all rules with first-order invariant
498: the value $\tau=1/2$ is universal, in the case of both piecewise linear \cite{paper19}
499: and nonlinear \cite{paper19} fundamental diagrams. Table 2 provides evidence that
500: a more general conjecture may be valid: regardless of the order of the invariant,
501: exponent $\alpha$ seems to have universal value of $1/2$. In the next section we will
502: offer some justification for this conjecture for rules with second-order invariants.
503: \section{Dynamics of localized structures}
504: In rule 184, the power-law convergence of the current toward its equilibrium value
505: is related to the dynamics of this rule, which resembles ballistic annihilation.
506: The spatiotemporal patter generated by rule 184 can be understood as propagation of
507: two types of localized structures, shown in Figure 4.
508: \begin{figure}
509: \begin{center}
510: \includegraphics[scale=0.5]{fig4.eps}
511: \end{center}
512: \caption{Spatiotemporal pattern generated by rule 184. Initial configuration
513: is represented by the top row, where black squares represent 1's, and white spaces represent 0's.
514: Consecutive configurations are plotted as consecutive rows. Two types of defects
515: of ``A'' and ``B'' type are visible, annihilating in the circled spot.}
516: \label{fig4}
517: \end{figure}
518: These two types of structures, marked with letters ``A'' and ``B'', propagate in opposite
519: directions and annihilate upon collision. At the critical point, the number of ``A'' defects
520: in the initial configuration is the same as the number of ``B'' defects, and it takes long time
521: for all of them to disappear, hence the ``critical slowing down'', or power-law convergence is observed. Detailed analysis
522: of this process \cite{paper11} leads to the exact formula for the current $j(\rho, t)$, which
523: in the limit of large $t$ and using de Moivre-Laplace limit theorem
524: leads to $j(\rho_c,\infty)-j(\rho_c,t) \sim t^{-1/2}$. Here,
525: by $f(t)\sim g(t)$ we mean that $\lim_{t\to\infty} f(t)/g(t)$
526: exists and is different from $0$.
527:
528: Dynamics of rules 14, 35, 43, 142 resembles rule 184 very strongly, as can be seen
529: in Figure 5, which shows spatiotemporal patterns at the critical point for all
530: four rules.
531: \begin{figure}
532: \begin{center}
533: \includegraphics[scale=0.7]{fig5.eps}
534: \end{center}
535: \caption{Spatiotemporal pattern generated by rules 14, 35, 43, 142
536: at their critical points.
537: Density of the invariant $\xi$ equals $1/4$ for rules 14, 43, 142,
538: and $1/3$ for rule 35.
539: }
540: \label{fig5}
541: \end{figure}
542: In all four cases, localized propagating structures moving in opposite directions and annihilating upon collision are visible.
543: In fact, for two of these rules, it is possible to establish direct relationship with rule 184.
544: In order to do this, we will define superposition of two rules as
545: \begin{equation}
546: (f \circ g) (x_0,x_1,x_2,x_3,x_4)
547: =f(g(x_0,x_1,x_2),g(x_1,x_2,x_3),g(x_2,x_3,x_4)).
548: \end{equation}
549: If $h \circ f = h \circ g$, then following \cite{Boccara93} we will say that
550: $g$ is a transform of rule $f$ by $h$. If by $f_k$ we denote local function of rule $k$,
551: one can show \cite{Boccara93} that
552: \begin{eqnarray}
553: f_{60} \circ f_{43} = f_{184} \circ f_{60},\\
554: f_{60} \circ f_{142} = f_{226} \circ f_{60}.
555: \end{eqnarray}
556: This means that there exists a local mapping (rule 60) which transforms rule 43 into rule 184, and rule 142 into rule 226 (recall that rule 226 is
557: the image of rule 184 under spatial reflection). Similarity of dynamics of rules 43, 142 to rule 184
558: is, therefore, not surprising.
559:
560: \section{Conclusion}
561: We investigated second order additive invariants in elementary cellular automata rules. We found that
562: fundamental diagrams of rules which possess additive invariant are either linear or exhibit
563: singularities similar to singularities of rules with first-order invariant. Singularities can appear only
564: in rules with exactly one invariant. At the critical
565: density of the invariant, the current decays to its equilibrium value as a power law $t^{\alpha}$,
566: and the value of the exponent $\alpha$ obtained from numerical simulations is very close
567: to $-1/2$. This indicates that regardless of the order of the invariant, the dynamics of
568: CA rules with invariants is very similar.
569:
570: Since rules 43 and 142 can be transformed into rules 184 and 226 by a surjective local transformation,
571: it should be possible to obtain for them rigorous formulas for the expected value of the current at
572: arbitrary time, similarly as it has been done for rule 184 and its generalizations \cite{paper11}. Such formula
573: could then used to compute the exact value of the exponent $\alpha$. For rules 14 and 35 no such local transformation
574: exists, nevertheless they exhibit localized propagating structures strikingly similar to
575: structures of rule 184, so exact calculation of the current might be possible too. This problem is
576: currently under investigation and will be reported elsewhere.
577:
578:
579:
580: \vskip 1cm
581: \noindent \textbf{Acknowledgements:} The author
582: acknowledges financial support from NSERC (Natural Sciences and
583: Engineering Research Council of Canada) in the form of the
584: Discovery Grant.
585:
586:
587: \providecommand{\href}[2]{#2}\begingroup\raggedright\begin{thebibliography}{10}
588:
589: \bibitem{Pivato02}
590: M.~Pivato, ``Conservation laws in cellular automata,'' {\em Nonlinearity} {\bf
591: 15} (2002) 1781--1793,
592: \href{http://xxx.lanl.gov/abs/arXiv:math.DS/0111014}{{\tt
593: arXiv:math.DS/0111014}}.
594:
595: \bibitem{Moreira03}
596: A.~Moreira, ``Universality and decidability of number-conserving cellular
597: automata,'' {\em Theor. Comput. Sci.} {\bf 292} (2003) 711--721.
598:
599: \bibitem{Durand2003}
600: B.~Durand, E.~Formenti, and Z.~R{\'o}ka, ``Number-conserving cellular automata
601: {I}: decidability,'' {\em Theoretical Computer Science} {\bf 299} (2003)
602: 523--535.
603:
604: \bibitem{Formenti2003}
605: E.~Formenti and A.~Grange, ``Number conserving cellular automata {II}:
606: dynamics,'' {\em Theoretical Computer Science} {\bf 304} (2003) 269--290.
607:
608: \bibitem{Morita2001}
609: K.~Morita and K.~Imai, ``Number-conserving reversible cellular automata and
610: their computation-universality,'' {\em Theoretical Informatics and
611: Applications} {\bf 35} (2001) 239--258.
612:
613: \bibitem{paper19}
614: H.~Fuk{\'s} and N.~Boccara, ``Convergence to equilibrium in a class of
615: interacting particle systems,'' {\em Phys. Rev. E} {\bf 64} (2001) 016117,
616: \href{http://xxx.lanl.gov/abs/arXiv:nlin.CG/0101037}{{\tt
617: arXiv:nlin.CG/0101037}}.
618:
619: \bibitem{paper25}
620: H.~Fuk{\'s}, ``Critical behaviour of number-conserving cellular automata with
621: nonlinear fundamental diagrams,'' {\em J. Stat. Mech.: Theor. Exp.} (2004).
622: art. no. P07005.
623:
624: \bibitem{Hattori91}
625: T.~Hattori and S.~Takesue, ``Additive conserved quantities in discrete-time
626: lattice dynamical systems,'' {\em Physica D} {\bf 49} (1991) 295--322.
627:
628: \bibitem{Krug88}
629: J.~Krug and H.~Spohn, ``Universality classes for deterministic surface
630: growth,'' {\em Phys. Rev. A} {\bf 38} (1988) 4271--4283.
631:
632: \bibitem{Nagatani95}
633: T.~Nagatani, ``Creation and annihilation of traffic jams in a stochastic
634: assymetric exclusion model with open boundaries: a computer simulation,''
635: {\em J. Phys. A: Math. Gen.} {\bf 28} (1999) 7079--7088.
636:
637: \bibitem{Nagel96}
638: K.~Nagel, ``Particle hopping models and traffic flow theory,'' {\em Phys. Rev.
639: E} {\bf 53} (1996) 4655--4672,
640: \href{http://xxx.lanl.gov/abs/arXiv:cond-mat/9509075}{{\tt
641: arXiv:cond-mat/9509075}}.
642:
643: \bibitem{Belitsky98}
644: V.~Belitsky and P.~A. Ferrari, ``Invariant measures and convergence for
645: cellular automaton 184 and related processes,''
646: \href{http://xxx.lanl.gov/abs/math.PR/9811103}{{\tt math.PR/9811103}}.
647: Preprint.
648:
649: \bibitem{paper11}
650: H.~Fuk{\'s}, ``Exact results for deterministic cellular automata traffic
651: models,'' {\em Phys. Rev. E} {\bf 60} (1999) 197--202,
652: \href{http://xxx.lanl.gov/abs/arXiv:comp-gas/9902001}{{\tt
653: arXiv:comp-gas/9902001}}.
654:
655: \bibitem{NishinariT98}
656: K.~Nishinari and D.~Takahashi, ``Analytical properties of ultradiscrete burgers
657: equation and rule-184 cellular automaton,'' {\em J. Phys. A-Math. Gen.} {\bf
658: 31} (1998) 5439--5450.
659:
660: \bibitem{BelitskyKNS01}
661: V.~Belitsky, J.~Krug, E.~J. Neves, and G.~M. Schutz, ``A cellular automaton
662: model for two-lane traffic,'' {\em J. Stat. Phys.} {\bf 103} (2001) 945--971.
663:
664: \bibitem{Blank03}
665: M.~Blank, ``Ergodic properties of a simple deterministic traffic flow model,''
666: {\em J. Stat. Phys.} {\bf 111} (2003) 903--930.
667:
668: \bibitem{MatsukidairaN03}
669: J.~Matsukidaira and K.~Nishinari, ``Euler-{L}agrange correspondence of cellular
670: automaton for traffic-flow models,'' {\em Phys. Rev. Lett.} {\bf 90} (2003)
671: art. no.--088701.
672:
673: \bibitem{paper10}
674: H.~Fuk{\'s}, ``A class of cellular automata equivalent to deterministic
675: particle systems,'' in {\em Hydrodynamic Limits and Related Topics}, A.~T.~L.
676: S.~Feng and R.~S. Varadhan, eds., Fields Institute Communications Series.
677: \newblock AMS, Providence, RI, 2000.
678: \newblock \href{http://xxx.lanl.gov/abs/arXiv:nlin.CG/0207047}{{\tt
679: arXiv:nlin.CG/0207047}}.
680:
681: \bibitem{Boccara93}
682: N.~Boccara, ``Transformations of one-dimensional cellular automaton rules by
683: translation-invariant local surjective mappings,'' {\em Physica D} {\bf 68}
684: (1992) 416--426.
685:
686: \end{thebibliography}\endgroup
687:
688: \end{document}
689: