Reliable Cellular Automata with Self-Organization

Let \mathbb{R} be the set of real numbers, and \mathbb{Z}_{m}.cap.z@\mathbb{Z}_{m}.))) the set of remainders modulo m. For m=\infty, this is the set \mathbb{Z} of integers.

We will use the notation

for what usually is written as f(n)=O(g(n)): that is for the fact that f(n)\le c g(n) for some constant c. We write f(n)\stackrel{{}_*}{=} g(n) if f(n)\stackrel{{}_*}{<} g(n) and g(n)\le f(n).

The standard mathematical notation for open intervals and for pairs is the same; we hope that the context will make the meaning always clear. We will use the same notation for intervals of integers as for those of real numbers: the context will make it clear, whether [a,b] or [a,b]\cap \mathbb{Z} is understood. Given a set A of space or space-time and a real number c, we write

and \overline{A} for its closure: for example, \overline{(a,b]}=[a,b]. Given two space-time sets A,B, we denote

Some lists of assertions are denoted by (a), (b), \dots and some by (1), (2), \dots. The attempted convention is that for a list properties that all hold or are required (conjunction) the items are labeled with (a),(b), \dots while if the list is a list of several possible cases (disjunction) then the items are labeled with (1), (2), \dots.

Maxima and minima will sometimes be denoted by \vee and \wedge"+@\vee,\wedge.))). We will write \log\log.))) for \log_{2}. For two strings u,v, we will denote by "+@\sqcup.)))

their concatenation.

The set {\Lambda}.cap@{\Lambda}.))) of sites has the form \mathbb{Z}_{m} for finite or infinite m. This will mean that in the finite case, we take periodic boundary conditions. In a space-time vector (x,t), we will always write the space coordinate first. For a space-time set E, we will denote its space- and time projections by .greek@\boldsymbol{\pi}_{\text{s}}, \boldsymbol{\pi}_{\text{t}}.)))

respectively. We will have a finite set {\mathbb{S}}.cap.s@{\mathbb{S}}.))) of states, the potential states of each site. A configuration is a function .greek@\xi(x).)))

for x\in{\Lambda}. Here, \xi(x) is the state of site x.

The time of work of our cellular automata will be the interval \left[0,\infty\right). Our space-time is given by

A history is a space-time function \eta(x,t).greek.func@\eta(x,t).))) which for each t defines a configuration. If in a history \eta we have \eta(x,v)=s_{2} and \eta(x,t)=s_{1}\ne s_{2} for all t<v sufficiently close to v then we can say that there was a switch from state s_{1} to state s_{2} at time v. For ordinary discrete-time cellular automata, we allow only histories in which all switching times are natural numbers 0,1,2,\dots. The time 0 is considered a switching time. If there is an \varepsilon such that \eta(c,t) is constant for a-\varepsilon<t<a then this constant value will be denoted by -@\eta(c, a-).)))

The subconfiguration \xi(D') of a configuration \xi defined on D\supseteq D' is the restriction of \xi to D'. Sometimes, we write

for the sub-configuration over the space-time set V.

A deterministic cellular automaton \ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}, {\Lambda}).)))

is determined by a transition function \ensuremath{\mathord{\textit{Tr\/}}}:{\mathbb{S}}^{3}\to{\mathbb{S}}\ensuremath{\mathord{\textit{Tr\/}}}, \overline{\ensuremath{\mathord{\textit{Tr\/}}}}.))) and the set {\Lambda} of sites. We will omit {\Lambda} from the notation when it is obvious from the context. A history \eta is a trajectory of this automaton if

holds for all x,t with t>0. For a history \eta let us write

for the “intended” value of \eta(x,t).

We will deal, from now on, only with cellular automata in which the set {\mathbb{S}} of local states consists of binary strings of some fixed length \ensuremath{\mathord{\textit{Cap\/}}}={\lVert {\mathbb{S}}\rVert}"|@{\lVert {\mathbb{S}}\rVert}.))) called the capacity of the sites. Thus, if the automaton has 16 possible states then its states can be considered binary strings of length 4.

Nonempty subsets of the set \mathopen\{0,\dots,{\lVert {\mathbb{S}}\rVert}-1\mathclose\} will be called fields. Some of these subsets will have special names. Let \ensuremath{\mathord{\textsf{\slshape{All\/}}}}.)))

If s=\mathopen(\verythinmathskips(i) : i\in\ensuremath{\mathord{\textsf{\slshape{All\/}}}}\mskip 1mu\mathclose) is a bit string and \ensuremath{\mathord{\textsf{\slshape{F\/}}}}=\mathopen\{i_{1},\dots,i_{k}\mathclose\} is a field with i_{j}<i_{j+1} then we will write .@s.\ensuremath{\mathord{\textsf{\slshape{F\/}}}}.)))

for the bit string that is called field \ensuremath{\mathord{\textsf{\slshape{F\/}}}} of the state. The field consisting of the bits s(i),s(i+1),\dots,s(j-1) of the state will be denoted

The array obtained from joining the same fields (say the Mail field) of all the different cells is called a track, (say the Mail track). (The terminology recalls tracks of a magnetic tape.) We will generally define fields that are either disjoint or contained in each other (but there may be some exceptions.) We will call the number of bits |\ensuremath{\mathord{\textsf{\slshape{F\/}}}}| in a field its \emph{width}, and and the proportion |F|/\ensuremath{\mathord{\textit{Cap\/}}} its relative width. The width and relative width of a track is defined to be the same as that of the corresponding field.

Given a transition function \ensuremath{\mathord{\textit{Tr\/}}}, we say that state s' is a legal successor of state s if there are states r,t with s'=\ensuremath{\mathord{\textit{Tr\/}}}(r,s,t). We define

if this holds and 0 if it does not.

A random history is a pair (\mu,\eta).greek@\mu.))) where \mu is a probability measure over some measurable space (\Omega,\mathcal{A}).greek@\Omega,\omega.))).cap@\mathcal{A},\mathcal{A}(W),\mathcal{A}_{t}.))) together with a measurable function \eta(x,t,\omega) which is a history for every value of \omega\in\Omega. We will generally omit \omega from the arguments of \eta. When we omit the mention of \mu we will use \operatorname{\mathsf P}\operatorname{\mathsf P}.))) to denote it. If it does not lead to confusion, for some property of the form \mathopen\{\mskip 1mu\eta\in R\mskip 1mu\mathclose\}, the quantity \mu\mathopen\{\mskip 1mu\omega : \eta(\cdot,\cdot,\omega)\in R\mskip 1mu\mathclose\} will be written as usual, as

We will denote the expected value of f with respect to \mu by :cap @\operatorname{\mathsf{E}}_{\mu} f.)))

where we will omit \mu when it is clear from the context.

A function f(\eta) with values 0, 1 (that is an indicator function) and measurable in the \sigma-algebra \mathcal{A} will be called an event function over \mathcal{A}. Let W be any subset of space-time that is the union of some rectangles. Then

denotes the \sigma-algebra generated by events of the form

for s\in{\mathbb{S}}, (x,t_{i})\in W.

For all times t, we assume the existence of a \sigma-algebra

The system \{\mathcal{A}_{t}\} is required to be increasing:

The function t\mapsto\mathcal{A}_{t} is our filtration.

A transition matrix \mathbf{P}(s, (r_{-1},r_{0},r_{1})).cap@\mathbf{P}(s, (r_{-1},r_{0},r_{1})).))) is a function \mathbf{P}:{\mathbb{S}}^{4}\to[0,1] with the property \sum_{s}\mathbf{P}(s,{\mathbf{r}})=1. For an arbitrary history \eta, and space-time point (x,t), denote

We will omit the parameter \eta when it is clear from the context.

A probabilistic cellular automaton \ensuremath{\mathord{\textit{PCA\/}}}(\mathbf{P}, {\Lambda}).)))

is characterized by saying which random histories are considered trajectories. Now a trajectory is not a single history (sample path) but a distribution over histories that satisfies the following condition, saying that the random history \eta is a trajectory if and only if the following holds. Let s_{1},\dots, s_{n}\in{\mathbb{S}}.

This expression uses the notion of conditional probability over sigma-algebras. For a more elementary statement of the same property, let x_{0},\dots,x_{n+1} be given with x_{i+1}=x_{i}+1. Let us fix an arbitrary history \zeta and an arbitrary event \mathcal{H}\ni\zeta in \mathcal{A}_{t-1}. Then we require

A probabilistic cellular automaton is noisy if \mathbf{P}(s,{\mathbf{r}})>0 for all s,{\mathbf{r}}. Bandwidth can be defined for transition probabilities just as for transition functions.

Let us call a transition rate matrix a function \mathbf{R}:{\mathbb{S}}^{4}\to\left[0,\infty\right),:cap@\mathbf{R}(s, {\mathbf{r}}).))) written as

with the normalization property \mathbf{R}(r_{0}, \mathopen( r_{-1},r_{0},r_{1}\mathclose))=0 for all {\mathbf{r}}. Its elements are called the transition rates.

Consider a generalization

of probabilistic cellular automata in which the sites are at positions i{B} for some fixed {B} called the body size and integers i, and the switching times are at 0,\delta,2\delta,3\delta,\dots for some small positive \delta:greek@\delta.))).

The random process defined by the above procedure will be denoted by \ensuremath{\mathord{\textit{CCA\/}}}(\mathbf{R}, {\Lambda}).)))

and called the (continuous-time) interacting particle system with the given rate matrix. We call this system noisy if \mathbf{R}(s,{\mathbf{r}})>0 for all s\ne r_{0}.

We will say that a random history (\mu,\eta) is a trajectory of the \varepsilon-perturbation \ensuremath{\mathord{\textit{CA\/}}}_{\varepsilon}(\ensuremath{\mathord{\textit{Tr\/}}}, {\Lambda}).)))

of the transition function \ensuremath{\mathord{\textit{Tr\/}}} if the following holds. For all x_{0},\dots x_{n+1},t with x_{i+1}=x_{i}+1 for all 0<i_{1}<\cdots<i_{k}<n,

Here is the same property without using conditional probability over sigma-algebras: for events \mathcal{H} in \mathcal{A}_{<(t-1)} with \mu(\mathcal{H})>0,

By the \varepsilon-perturbation of a continuous-time interacting particle system with transition rates given by \mathbf{R}(s,{\mathbf{r}}), we understand the following: in the above construction of a process, perturb the matrix elements \mathbf{R}(s,{\mathbf{r}}) by some arbitrary amounts smaller than \varepsilon.

Suppose that the bit string that is a local state has some field \ensuremath{\mathord{\textsf{\slshape{F\/}}}} (it can for example be the first two bits of the state). We will say that \ensuremath{\mathord{\textit{Tr\/}}} remembers field \ensuremath{\mathord{\textsf{\slshape{F\/}}}} over an infinite set of sites {\Lambda} if there is an \varepsilon>0 such that for each string s\in\{0,1\}^{|\ensuremath{\mathord{\textsf{\slshape{F\/}}}}|} there is a configuration \xi_{s} such that for all trajectories (\mu,\eta) of the \varepsilon-perturbation \ensuremath{\mathord{\textit{CA\/}}}_{\varepsilon}(\ensuremath{\mathord{\textit{Tr\/}}}, {\Lambda}) with \eta(\cdot,0)=\xi_{s}, for all x,t we have

We define similarly the notions of remembering a field for a probabilistic transition matrix \mathbf{P} and a probabilistic transition rate matrix \mathbf{R}.

Let us call a configuration \xi homogeneous if there is a state q\in{\mathbb{S}} such that \xi(x)=q for all x. We say that \ensuremath{\mathord{\textit{Tr\/}}} remembers \ensuremath{\mathord{\textsf{\slshape{F\/}}}} in a self-organizing way if the initial configurations \xi_{s} can be made homogeneous.

Let us call the set .greek.cap@\Sigma_{0}.)))

the standard alphabet.

Each field \ensuremath{\mathord{\textsf{\slshape{F\/}}}} of a cell state such that the field size is even, can be considered not only a binary string but a string of (half as many) symbols in the standard alphabet. If r,s are strings in (\Sigma_{0})^{n} then "<@\preceq.)))

will mean that s(i)=r(i) for all 0\le i<n such that r(i)\ne *. For functions f, g, with values in \Sigma_{0}^{*} we will write f\preceq g if for all s we have f(s)\preceq g(s).

Let \ensuremath{\mathord{\textit{Tr\/}}} be any deterministic transition function with states in some alphabet \mathbb{A} with distinguished fields \ensuremath{\mathord{\textsf{\slshape{Input\/}}}}, \ensuremath{\mathord{\textsf{\slshape{Output\/}}}} in the standard alphabet. We say that \ensuremath{\mathord{\textit{Tr\/}}} has monotonic output if for all trajectories \eta of \ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}) we have

We call the transition function \ensuremath{\mathord{\textit{Tr\/}}}, all of whose fields have even size (that is they are in the standard alphabet) together with some distinguished fields \ensuremath{\mathord{\textsf{\slshape{Input\/}}}}, \ensuremath{\mathord{\textsf{\slshape{Output\/}}}}, a standard computing transition function if it

We will call a cellular automaton with such a transition function a standard computing medium. Our construction uses for this a special field called

It can take a constant number of values; in one case, only -1,0,1.

Let \Sigma_{0} be the standard input/output alphabet, and {\mathbb{S}} a set of states with {\mathbb{S}}.\ensuremath{\mathord{\textsf{\slshape{Input\/}}}}=\Sigma_{0}, having another field \ensuremath{\mathord{\textsf{\slshape{Color\/}}}}\in\{-1,0,1\}, and a distinguished “latent” state s_{0}. Let \psi_{*}:\Sigma_{0}^{*}\to{\mathbb{S}}^{*} be an encoding. For any finite string \rho\in\Sigma_{0}^{n}, let n'=|\psi_{*}(\rho)|. For any finite or infinite N>|\rho'|, we construct the initial configuration \ensuremath{\mathord{\textit{Init\/}}}_{\psi_{*}}(\rho).)))

as follows. Define \xi by setting \xi(x)=\psi_{*}(\rho)(x) for each x\in\left[0,n'\right), and s_{0} for all other x. In case N=\infty, obtain \xi' from \xi by setting

When the space is the finite one \mathbb{Z}_{N} then, with r=(N-n)/2, set \xi'(x).\ensuremath{\mathord{\textsf{\slshape{Color\/}}}}=-1 for -r\le x<0, and 1 for n\le x < n+r.

Let x be a site and {Q} a positive integer. The set of {Q}.cap@{Q}.))) sites x+i for i\in\left[0,{Q}\right) will be called the {Q}-colony with base x, and site x+i will be said to have address i in this colony.

An address field is a field of our cells that we will denote \ensuremath{\mathord{\textsf{\slshape{Addr\/}}}}\ensuremath{\mathord{\textsf{\slshape{Addr\/}}}}.))): we will always have it when we speak about colonies. The value \xi(x).\ensuremath{\mathord{\textsf{\slshape{Addr\/}}}} is a binary string which will be interpreted as an integer in \left[0,{Q}\right).

For a parameter {U} .cap@{U}.))) that we will fix in each case, let us call {U} steps of work of a colony a work period.

Let us denote by M_{1} the fault-tolerant cellular automaton to be built.

A code \varphi.greek@\varphi_{*},\varphi_{*}.))) between two sets R,S is a pair (\varphi^{*},\varphi_{*}) where \varphi_{*}: R\to S is the encoding function and \varphi^{*}: S\to R is the decoding function, and the relation

holds. We will be particularly interested in the example when for a positive integer {Q} called the block size and some finite sets {\mathbb{S}}_{1},{\mathbb{S}}_{2} we have R={\mathbb{S}}_{2}, S={\mathbb{S}}_{1}^{{Q}}. Such a code is called a block code. In a block code, strings of the form \varphi_{*}(r) are called codewords. The elements of a codeword s=\varphi_{*}(r) are numbered as s(0),\dots,s({Q}-1).

The codes \varphi between sets R, S used in our simulations will have a feature similar to the acceptance and rejection of Example 4.7. The set R will always have a subset of symbols called vacant, and among them a distinguished special symbol \ensuremath{\mathord{\textit{Vac\/}}}\ensuremath{\mathord{\textit{Vac\/}}}.))). An element s\in S will be called accepted by the decoding if \varphi^{*}(s)\ne \ensuremath{\mathord{\textit{Vac\/}}}, otherwise it is called rejected.

Let \varphi be a block code from {\mathbb{S}}_{2} to {\mathbb{S}}_{1}^{{Q}}, and \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1},\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{2} be fields of {\mathbb{S}}_{i}. We will say that \varphi aggregates field \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1} into \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{2} if

Hence for u,v\in{\mathbb{S}}_{2} the assumption u.\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{2}=v.\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{2} implies that for all a\in\left[0,{Q}\right) we have

Assume that code \varphi aggregates field \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1} into \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{2}. We say that field \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1} controls address a for code \varphi via a function \gamma:{\mathbb{S}}_{1}.\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1}\to{\mathbb{S}}_{1} if in all codewords w of the form w=\varphi_{*}(u), we have w(a)=\gamma(w(a).\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1}).

A dwell period of \eta is a tuple (x,s,t_{1},t_{2}) such that x is a site, s is a nonvacant state, and 0\le t_{1}<t_{2} are switching times with \eta(x,t_{1})=s. The rectangle \left[x,x+{B} \right)\times \left[t_{1},t_{2}\right) is the space-time body of the dwell period.

A configuration is a lattice configuration if all cells are at sites of the form i{B} for integers i. We can also talk about lattice histories: these have space-time bodies of the form

for integers i,j<k.

A deterministic cellular automaton \ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}, {B} , {T} , {\Lambda}).)))

is determined by parameters {B} ,{T} >0 and a transition function \ensuremath{\mathord{\textit{Tr\/}}}:{\mathbb{S}}^{3}\to{\mathbb{S}}. We may omit some obvious arguments from this notation. A lattice history \eta with parameters {B} ,{T} is a trajectory of this automaton if

holds for all x,t with t\ge {T}. For a history \eta let us write

A medium is said to have a variable period if in its histories, not all dwell periods have necessarily the same size.

A block code will be called overlap-free if for every string \mathopen( s(0),\dots s(n-1)\mathclose), and all i\le n-{Q}, if both strings \mathopen( s(0),\dots,s({Q} -1)\mathclose) and \mathopen( s(i+1),\dots,s(i+{Q} -1)\mathclose) are accepted then i\ge {Q}.

Suppose that a configuration \xi of M_{2}, which is a cellular medium \ensuremath{\mathord{\textit{AMed\/}}}({Q} {B} ), is given. Then we define the encoded configuration \xi_{*}=\varphi_{*}(\xi) of M_{1}, which is an \ensuremath{\mathord{\textit{AMed\/}}}({B} ), by setting for each cell x of \xi and 0\le i < {Q},

Suppose that a configuration \xi of M_{1} is given. We define the decoded configuration \xi^{*}=\varphi^{*}(\xi) of M_{2} as follows: for site x, set \xi'(x)=\varphi^{*}(s) where

We define \xi^{*}(x)=\xi'(x) if the latter is vacant or there is no y closer than {Q} {B} to x with \xi'(y) non-vacant; otherwise \xi^{*} = \ensuremath{\mathord{\textit{Vac\/}}}.

A configuration \xi is called a code configuration if \varphi_{*}(\varphi^{*}(\xi))=\xi and the decoded configuration \varphi^{*}(\xi) covers the space with adjacent cells.

Suppose that M_{1} and M_{2} are deterministic cellular automata where M_{i}=\ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}_{i},{B} _{i},{T}_{i}), and \varphi is a block code with

The decoding function may be as simple as in Example 4.7: there is an \ensuremath{\mathord{\textsf{\slshape{Info\/}}}} track and once the colony is accepted the decoding function depends only on this part of the information in it.

For each history \eta of M_{1}, we define the history \eta^{*}=\varphi^{*}(\eta) of M_{2} by setting

We will say that the code \varphi is a simulation if for each configuration \xi of M_{2}, for the (unique) trajectory \eta of M_{1}, such that \eta(\cdot,0)=\varphi_{*}(\xi), the history \eta^{*} is a trajectory of M_{2}.

A transition function \ensuremath{\mathord{\textit{Tr\/}}} is intrinsically universal if for every other transition function \ensuremath{\mathord{\textit{Tr\/}}}' there are {Q} ,{U} and a block code \varphi such that \varphi is a block simulation of \ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}',{Q} ,{U} ) by \ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}},1,1).

A simulation of medium M_{2} by medium M_{1} having the same system \{\mathcal{A}_{t}\}_{t\ge 0} of \sigma-algebras is given by a pair .greek.cap@\Phi^{*}.)))

where \Phi^{*} is a mapping of the set of histories of M_{1} into those of M_{2} (the decoding), \varphi_{*} is a mapping of the set of configurations of M_{2} to the set of configurations of M_{1} (the encoding for initialization). We will assume to \varphi_{*} always belongs also a decoding function \varphi^{*}, so they form a code \varphi=(\varphi_{*},\varphi^{*}). Denote

We require that for each configuration \xi of M_{2}, each trajectory of \eta of M_{1} with \eta(\cdot,0) = \varphi_{*}(\xi) the decoded history\eta^{*} is a trajectory of M_{2} with \eta^{*}(\cdot,0)=\xi.

A simulation will be called non-anticipating if for each t the random function \eta^{*}(\cdot,t) is measurable in the \sigma-algebra \mathcal{A}_{t} (this says that it does not depend on the future). Let \sigma be the shift on one-dimensional configurations and histories: that is (\sigma\xi)(x)=\xi(x+1), (\sigma\eta)(x,t)=\eta(x+1,t). A simulation (\varphi,\Phi^{*}) is called shift-invariant if both \varphi and \Phi^{*} commute with the shift. It is called local if there is a finite space-time rectangle V^{*}=I\times (u,0].cap@V^{*}.))) with u\ge 0 such that \Phi^{*}(\eta)(w,t) depends only on \eta((w,t)+V^{*}). A local simulation is both non-anticipating and shift-invariant.

All our simulations will be local, unless stated otherwise. It follows from the non-anticipating property that \eta^{*}(\cdot,0) depends only on \eta(\cdot,0), and therefore the simulation always defines a decoding function

on configurations (as already stipulated in Definition 4.23).

If u=0 and so the configuration \eta^{*}(\cdot,t) depends only on the configuration \eta(\cdot,t), then the simulation will be called memoryless.

Let \Phi=(\varphi,\Phi^{*}) be a simulation as in (4.23) whose encoding \varphi_{*} is a block code with block size {Q}, between abstract cellular media M_{k} (k=1,2). Let D_{k}>0 (k=1,2) be some parameters, and \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1} a field of the medium M_{1}. We say that \Phi has the \varepsilon-trickle-down property of field \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k+1} into field \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1} with respect to D_{1}, D_{2} if the following holds for every configuration \xi of M_{2} and every trajectory (\mu,\eta) of M_{1} with \eta(\cdot, 0)=\varphi_{*}(\xi).

Recall the notation \preceq in Definition 3.4. Let \eta^{1}=\eta, \eta^{2}=\Phi^{*}(\eta). Let x_{1},x_{2} be sites where x_{1} has address a in the {Q}-colony with base x_{2}, let t_{0} be some time. For k=1,2 let \mathcal{E}_{k}(x) be the event that \eta^{k}(x,t) is non-vacant and \eta^{k}(x,t).\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k} does not change during (t_{0}-D_{k},t_{0}+D_{k}]. Let \mathcal{E}'_{1} be the event that \mathcal{E}_{1}(x_{1}) holds and also

during this time. Then \operatorname{\mathsf P}( \mathcal{E}_{2}(x_{2})\setminus\mathcal{E}')<\varepsilon.

A cellular medium M is called initially stable for an initial configuration \xi with parameters (\varepsilon,{T} ) if for each trajectory \eta of M with \eta(\cdot,0)=\xi, for each site x, we have

If \varphi,\psi are two codes then \varphi\circ\psi is defined by (\varphi\circ\psi)_{*}(\xi)=\varphi_{*}(\psi_{*}(\xi)) and (\varphi\circ\psi)^{*}(\zeta)=\psi^{*}(\varphi^{*}(\zeta)). It is assumed that \xi and \zeta are here configurations of the appropriate cellular automata, that is the cell body sizes are in the corresponding relation. The code \varphi\circ\psi"+@\circ.))) is called the composition of \varphi and \psi.

A hierarchical code is given by a sequence .greek.hier@\varphi_{k*}.))) {\mathbb{S}}_{k},\varphi_{k*}, (k\ge 1) where {\mathbb{S}}_{k} is an alphabet and \varphi_{k*} is a code on configurations. In the present work we will also assume that \varphi_{k*} always comes from a block code

.cap@{Q}_{k}.))) Since {\mathbb{S}}_{k} and {Q}_{k} are implicitly defined by \varphi_{k*} we can refer to the code as just (\varphi_{k*})_{k\ge 1}. Consider a hierarchical code (\varphi_{k*})_{k\ge 1} with state sequence ({\mathbb{S}}_{k}) and block sizes ({Q}_{k}). If for each {\mathbb{S}}_{k} a field \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k} is also given, and the code \varphi_{k*} is aggregating from \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k} to \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k+1} as in Definition 4.9, then the hierarchical code (\varphi_{k*})_{k\ge 1} will be called aggregating for the field sequence (\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k}). If also \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k} controls address 1 via a function \gamma_{k} (as by Definition 4.10) then sequence of triples

will be called a centrally aggregating hierarchical code, or shortly, a code complex.

Given a hierarchical code as above, for a configuration \xi we define the higher-order configurations \xi^{k} by the recursion

The configuration \xi is called a code configuration for the given hierarchical code if \xi^{k} is a code configuration for the code \varphi_{k} for each k in the sense of Definition 4.18.

Suppose that a code configuration \xi of a hierarchical code (\varphi_{k})_{k\ge 1} with block sizes {Q}_{k} is given. The sizes of the higher-order cells are defined by .cap@B_{k}.))) {B} _{1}= 1, {B} _{k+1} = Q_{k}{B}_{k}. For a finite or infinite n, let .cap@K_{0}(n).)))

If n<\infty then this is the largest k for which a k-cell can fit into \mathbb{Z}_{n}. As long as k< K_{0}(n) the origin of our space {\Lambda} will be contained in some cell x_{k} of level k. Since it is in our power to define the code configuration \xi, for simplicity we will always arrange that x_{k} is the second cell (the cell with address 1) of the colony coding cell x_{k+1}. In general, we could use any addresses 0<a_{k}<{Q}-1 instead of setting them always to 1, but this is not important now. Setting a_{k}=1 gives

A finite or infinite sequence (s_{1},s_{2},\dots) will be called fitted to a code complex \mathbf{H} as in (5.3) if

for all k.

Let \rho\in ({\mathbb{S}}_{1}.\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1})^{\mathbb{Z}} be an infinite sequence, we define its encoding \psi_{*}(\rho) as follows. Define for each k the following segment of \rho using the notation of Definition 5.5:

The sequence s_{k}=\gamma_{k}(\rho_{k}), k=1,2,\dots is fitted, allowing via Proposition 5.7 to encode \rho into an infinite code configuration

Given a code complex \mathbf{H} as above and a string \rho\in\Sigma_{0}^{n} we define its encoding \xi=\psi_{*}(\rho)\in{\mathbb{S}}^{\infty} which has the property that except for an interval of size \le 2 n, \xi(x)=q_{{\mathbb{S}}} with q_{{\mathbb{S}}} as in Definition 3.6. Let k=K_{0}(n) as in (5.5). Create the string

where -x _{k} stars come at the front and the l stars at the back are the fewest to make |\rho'|=m{B}_{k} an integer multiple of {B}_{k}. Now create a string r_{k}\in {\mathbb{S}}_{k}^{m} as follows. First let r'_{k}=q_{{\mathbb{S}}}^{m}, and then obtain r_{k} by replacing the track r'_{k}.\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k} in it with \rho'. Define the string r_{k-1}=\varphi_{i*}(r_{k})\in{\mathbb{S}}_{k-1}^{m{Q}_{k-1}}. Similarly, for each 1\le i<k let r_{i}=\varphi_{(i+1)*}(r_{i+1}). By definition of the codes \varphi_{i*}, string r_{i} consists of m{Q}_{k-1}{Q}_{k-2}\cdots{Q}_{i} symbols of {\mathbb{S}}_{i}. Then r_{1} consists of m{B}_{k} symbols of {\mathbb{S}}={\mathbb{S}}_{1}. Finally obtain \xi as follows: \xi(x)=r(x-x_{k}) for 0\le x-x_{k}<m{B}_{k} and q_{{\mathbb{S}}} otherwise.

Suppose that a sequence M_{1},M_{2},\dots of abstract media is given along with simulations \Phi_{1},\Phi_{2},\dots such that \Phi_{k} is a simulation of M_{k+1} by M_{k}. Such a system will be called an amplifier.

Suppose that an amplifier \mathbf{\Psi}=(M_{k},\Phi_{k})_{k\ge 1} is given along with the sequences of parameters D_{k},

\varepsilon''_{k}, \varepsilon_{k}, and fields \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k}. Assume that each \varphi_{k} aggregates field \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k} into \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k+1}, as in Definition 4.9.

1. \mathbf{\Psi} has the trickle-down property if for each k, the simulation \Phi_{k} has the trickle-down property of \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k+1} to \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k} with parameters D_{k}, D_{k+1} and \varepsilon''_{k}, as in Definition 4.25.

2. Suppose that the sequence \varepsilon_{k}>0, k=1,2,\dots is given with \sum_{k}\varepsilon_{k}<1/6, and also a configuration \xi of M_{1} that is a code configuration of the code sequence (\varphi_{k}) (as introduced in Definition 5.3). We say that \mathbf{\Psi} has the initial stability property with respect to the initial configuration \xi and parameters ((\varepsilon_{k},{T}_{k}))_{k\ge 1} if for each k the medium M_{k} has the initial stability property for initial configuration \xi^{k} and parameters (\varepsilon_{k},2D_{k}).

A block code \varphi=(\varphi_{*}, \varphi^{*}) with \varphi^{*}: R^{{Q} }\to S will be called is t-error-correcting if for all x\in\Sigma, y\in\Sigma^{\\Q } we have \varphi^{*}(y)=x whenever y differs from \varphi_{*}(x) in at most t symbols. It is (\beta,t)-burst-error-correcting if for all x\in S, y\in R^{\\Q } we have \varphi^{*}(y)=x whenever y differs from \varphi_{*}(x) in at most t intervals of size \le\beta. For such a code, we will say that a word y\in R^{\\Q } is (\beta, r)-compliant if it differs from a codeword of the code by at most r intervals of size \le\beta.

Assume that strings R and S are in \{0,1\}^{*}. In the context of error-correcting codes, elements of R will be called the symbols of the codewords, as opposed to bits, since each symbol may consist of several bits.

Let {\Lambda}(n).cap.s@{\Lambda}(n).))) be an increasing sequence of finite subsets of {\Lambda} with \bigcup_{n}{\Lambda}(n)={\Lambda}, for example

if {\Lambda}=\mathbb{Z}^{d}. We can view an element {\mathbf{s}} of {\mathbb{S}}^{{\Lambda}(n)} as a vector whose components are indexed with the elements of {\Lambda}(n). For a measure \nu over configurations, denote :greek@\nu({\mathbf{s}}).)))

For n=0, we have the special case \nu(s)=\nu\mathopen\{\mskip 1mu\xi : \xi(0)=s\mskip 1mu\mathclose\}.

If (\mu,\eta) is a random trajectory of a probabilistic cellular automaton then let \mu^{t}.greek@\mu^{t}.))) be the distribution of the configuration \eta(\cdot,t).

A sequence \nu_{k} of measures over {\mathbb{S}}^{{\Lambda}} weakly converges to measure \nu if \lim_{k}\nu_{k}({\mathbf{s}})=\nu({\mathbf{s}}) for all n and for all {\mathbf{s}}\in{\mathbb{S}}^{{\Lambda}(n)}.

Assuming for simplicity, {\Lambda}=\mathbb{Z}, we define the Markov operator P on measures by

where the summation goes over all possible strings \mathbf{r}=(r_{-n},\dots,r_{n})\in{\mathbb{S}}^{2n+1}. Given a Markov operator, we call a measure \alpha over configurations invariant if P\alpha=\alpha.

A probabilistic cellular automaton is called ergodic if

A deterministic cellular automaton R:cap@R.))) given by Toom can be defined as follows. We start from a two-dimensional deterministic cellular automaton R with set of states \{0,1\}, by the neighborhood H= \mathopen\{(0,0),(0,1),(1,0)\mathclose\}. The transition function \ensuremath{\mathord{\textit{Tr\/}}}_R(x_{1},x_{2},x_{3}) is the majority of x_{1},x_{2},x_{3}.

Let us define the variation distance for measures \mu and \nu are restricted to {\mathbb{S}}^{{\Lambda}(n)}: d^{n}(\mu, \nu).)))

For the set of sites .cap@{\Lambda}_{m}.)))

where m can be finite or infinite, suppose that the local transition matrix (for simplicity, with nearest-neighbor interaction) gives rise to an ergodic Markov process with Markov operator P_{m}.cap@P_{m}.))). Let :cap@D^{n}_{m}(t).)))

where the dependence on the transition matrix is understood but is not shown.

We will call the function r_{m}(n,\delta).)))

the relaxation time.

We can call the Markov operator P strongly not forgetful if it has two disjoint (weakly) closed translation-invariant sets of measures.

Let us be given an arbitrary one-dimensional transition function \ensuremath{\mathord{\textit{Tr\/}}} and the integers N,L. We define the three-dimensional transition function \ensuremath{\mathord{\textit{Tr\/}}}' as follows. The interaction neighborhood is H\times\{-1,0,1\} with the neighborhood H defined in Section ? above. The transition function \ensuremath{\mathord{\textit{Tr\/}}}' says: in order to obtain your state at time t+1, first apply the transition function R in each plane defined by fixing the third coordinate. Then, apply transition function \ensuremath{\mathord{\textit{Tr\/}}} on each line obtained by fixing the first and second coordinates. (The papers [GacsReif3dim88] and [Gacs2dim89] use a neighborhood instead of H that makes some proofs easier: the northern, south-eastern and south-western neighbors.)

For an integer m, define the space {\Lambda} =\mathbb{Z}_{m}^{2}\times \mathbb{Z}_{N}. For a trajectory \xi of \ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}) on \mathbb{Z}_{N}, define the trajectory \zeta of \ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}') on {\Lambda} by

Thus, each cell of \ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}) is repeated m^{2} times, on a whole “plane” (a torus for finite m) in {\Lambda}.

A random history \eta will be called a trajectory of medium M given in (?) if for each time u\ge 0, for each set of disjoint local conditions

with (\cmforall{i})\inf\boldsymbol{\pi}_{\text{t}} V_{i}\ge u (where \inf\emptyset=\infty), we have

In a probabilistic cellular automaton \ensuremath{\mathord{\textit{PCA\/}}}(\mathbf{P},{B} ,{T} ) for each set of states E and state vector {\mathbf{r}}=(r_{-1},r_{0},r_{1}) let \mathbf{P}(E,{\mathbf{r}})=\sum_{q\in E}\mathbf{P}(q,{\mathbf{r}}). Let \mathbf{K}.cap@\mathbf{K}.))) be the set of functions K:{\mathbb{S}}^{3}\to 2^{{\mathbb{S}}} such that

For a space-time trajectory \eta and a K\in\mathbf{K} let us write .cap@\overline{K}(x,t,\eta).)))

For all K',K''\in\mathbf{K} with K'\supseteq K'' (that is K'({\mathbf{r}})\supseteq K''({\mathbf{r}}) for all {\mathbf{r}}), let .cap@c(K',K'').)))

be the maximum conditional probability, over all {\mathbf{r}}, of getting into the set K''({\mathbf{r}}) provided we get into the set K'({\mathbf{r}}).

In case of a continuous-time cellular automaton \ensuremath{\mathord{\textit{CCA\/}}}(\mathbf{R},{B} ), for any functions K,K',K''\in\mathbf{K} with \mathbf{R}(K,{\mathbf{r}}) = \sum_{q\in K}\mathbf{R}(q,{\mathbf{r}}) we define the corresponding quantity c(K',K'').)))

A real random variable \sigma with values in \left[0,\infty\right) is a stopping time if for each t the event \{\sigma\le t\} is in \mathcal{A}_{t}. Let \mathcal{A}_{\sigma}.cap@\mathcal{A}_{\sigma}.))) be the algebra of events A with the property that A\cap\{\sigma\le t\}\in \mathcal{A}_{t}.

For some time interval J\subset(0,\infty], assume that there is a stopping time \sigma\in J, and random variables \alpha(\eta)\in\mathbb{T}, V(\eta)\in\ensuremath{\mathord{\textit{Rectangles\/}}} .cap@V^{t}(\eta).))) measurable in \mathcal{A}_{\sigma}, with

The tuple (J,\sigma,\alpha(\eta),V(\eta)) will be called a constraint process.

A random history \eta will be called a strong trajectory if for each system of disjoint constraint processes

(where V_{i}(\eta) are disjoint for each \eta), h an event function over \mathcal{A}_{<\boldsymbol{\pi}_{\text{t}} J}, we have

Let M, M^{*} be media where the functions g,b define the constraints on M, and g^{*},b^{*} define those for M^{*}, and let \varphi^{*} be a mapping from the histories of M into those of M^{*}: we will write \eta^{*}=\varphi^{*}(\eta). We say that \varphi^{*} is a canonical simulation map if the following holds. Let (\alpha,V) be any local condition for M^{*} with \alpha\in\mathbb{T}^{*}. Then there is a system of constraint processes \beta_{i,j}(\alpha).))) W^{t}_{i,j}(\eta).))) \tau_{i,j}.)))

with i=1,\dots,n(\alpha,V), j=1,\dots,k(\alpha,V,i) with the following properties. For each fixed i, the subsystem containing elements for that i forms a system of disjoint constraint processes with W_{i,j}(\eta)\subseteq V, and for almost all \eta we have

We will call a canonical simulation injective if the mapping \varphi^{*} is the identity, and the functions b_{2}(\alpha),{g}_{2}(\alpha,\cdot,\cdot) differ from b_{1},{g}_{1} only in that they are defined for some additional types \alpha, too.

For some function K\in\mathbf{K} let a_{-1}(K), a_{1}(K).)))

Let us call a time u of the history \eta a K-switching-time of a cell x if for all t<u sufficiently close to u we have \eta(x,t)\notin \overline{K}(x,t,\eta) and \eta(x,u)\in \overline{K}(x,t,\eta), that is we have just jumped into the target set determined by K(\cdot). For some rational constant D>0, integers k\ge 0 and j=-1,1 let .greek@\alpha(K,D,k,j).)))

be the event function of type \alpha(K,D,k,j) for j=-1 [j=1] for the event that site x has at most [at least] k times during (u,u+D], that are K-switching times.

We say that the number {T_{\bullet}}>0:cap@{T_{\bullet}},T^{\bullet}.))) is a dwell period lower bound of a history \eta if no dwell period of \eta is shorter than {T_{\bullet}}.

The medium -var.)))

is defined as follows. The set {\mathbb{S}} of states is implicit in the transition function therefore from now on, it will be omitted. We have dwell period lower and upper bounds {T_{\bullet}}\leT^{\bullet} and a failure probability bound \varepsilon>0. The local state, as in Example 2.15, is a record with two fields, \ensuremath{\mathord{\textsf{\slshape{Det\/}}}} and \ensuremath{\mathord{\textsf{\slshape{Rand\/}}}} where \ensuremath{\mathord{\textsf{\slshape{Rand\/}}}} consists of a single bit. To simplify the upper-bounding of dwell periods, we assume that the transition function has the property

that is the deterministic part of a cell will change at every transition. In a more general setting later this will be called “time marking”, see Condition 10.9. For a history \eta, site x let a \ge 0 be a rational number. If a>0 then let \sigma_{1}=\sigma_{1}(x,a,\eta), and \sigma_{2} be the first and second switching times t>a of \eta in x. If a=0 then let \sigma_{1}=0 and let \sigma_{2} be the first switching time >0. Let us list the different types of local condition.

We define the totally asynchronous cellular automaton \ensuremath{\mathord{\textit{ACA\/}}}.)))

associated with transition function \ensuremath{\mathord{\textit{Tr\/}}} as follows: \eta is a trajectory if for all x,t we have either \eta(x,t+1)=\eta(x,t) or the usual

A site x is free in a configuration \xi if \ensuremath{\mathord{\textit{Tr\/}}}(\xi(x-1),\xi(x),\xi(x+1))\ne \xi(x). The set of free sites will be denoted by L(\xi):cap@L(\xi).))). For a space configuration \xi and a set E of sites, let us define the new configuration \ensuremath{\mathord{\textit{Tr\/}}}(\xi,E)\ensuremath{\mathord{\textit{Tr\/}}}(\xi,E).))) by

Let the update set .cap@U(t,\eta).)))

be the set of sites x with \eta(x,t+1)\ne \eta(x,t).

For any set A, let us use the indicator function .greek@\chi(x,A).)))

For given history \eta, we define the function \tau(x,t)=\tau(x,t,\eta):greek@\tau(x,t).))) as follows:

We can call \tau(x,t) the effective age of x in \eta at time t: this is the number of effective updatings that x underwent until time t.

Given a transition function \ensuremath{\mathord{\textit{Tr\/}}} and an initial configuration \xi, we say that the function has invariant histories if there is a function \zeta(x,u)=\zeta(x,u,\xi) such that for all trajectories \eta(x,t) of \ensuremath{\mathord{\textit{ACA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}) with \eta(\cdot,0)=\xi we have

Denote

We call a transition function \ensuremath{\mathord{\textit{Tr\/}}} commutative if for all configurations \xi and all distinct pairs x,y\in L(\xi) we have \ensuremath{\mathord{\textit{Tr\/}}}(\xi,\{x\},\{y\}) = \ensuremath{\mathord{\textit{Tr\/}}}(\xi,\{y\},\{x\}).

We will denote the smallest absolute-value remainders \mskip-\medmuskip\mkern5mu\mathbin {\operator@font amod}\penalty900 \mkern5mu\mskip-\medmuskip.)))

with respect to a positive integer m>2, defined by the requirement -m/2< b \mskip-\medmuskip\mkern5mu\mathbin {\operator@font amod}\penalty900 \mkern5mu\mskip-\medmuskip m \le m/2.

The commutative medium of the above proof will be called the marching soldiers scheme.

For a standard computing medium with transition function \ensuremath{\mathord{\textit{Tr\/}}}, integers s and t and string X consider a trajectory \eta of \ensuremath{\mathord{\textit{CA\/}}}(\ensuremath{\mathord{\textit{Tr\/}}}) over the rectangle [0,s]\times[0,t] with an initial configuration in which \eta(0,0)=\eta(s,0)=\#\cdots \#, further X is on the input track on the interval [1,s-1] (padded with *’s to fill up [1,s-1]), *’s on the output track and 0’s on the other tracks. This defines a trajectory \eta since the \#’s on the input field in cells 0 and s imply that the cells outside the interval [0,s] will have no effect. Assume that, at time t, the \ensuremath{\mathord{\textsf{\slshape{Output\/}}}} track has no * on the interval [1,s-1]. The string w on the \ensuremath{\mathord{\textsf{\slshape{Output\/}}}} track on [1,s-1] will be called the result of the computation, denoted by \ensuremath{\mathord{\textit{Tr\/}}}(X; s, t).)))

The standard computing transition function \ensuremath{\mathord{\textit{Tr\/}}}_{0} is efficiently universal if for every standard computing transition function \ensuremath{\mathord{\textit{Tr\/}}}, there is a string \ensuremath{\mathord{\texttt{\slshape{prog\/}}}} and constants c,d such that for all strings X and positive integers s,t we have \ensuremath{\mathord{\texttt{\slshape{prog\/}}}}.)))

whenever the left-hand side is defined.

We fix an efficiently universal standard computing transition function \ensuremath{\mathord{\textit{Univ\/}}}.)))

for the definition of various string functions. In view of Theorem 13, we can (and will) also assume that \ensuremath{\mathord{\textit{Univ\/}}} is commutative as in Definition 8.8. For a string function f(X) defined over some domain E we will say that \ensuremath{\mathord{\texttt{\slshape{prog\/}}}} is a program for f with time complexity bound t and space complexity bound s over E if the relation

holds.

For a field \ensuremath{\mathord{\textsf{\slshape{F\/}}}}, if \eta is the current configuration and x is a site then we will sometimes write

We will write .greek@\vartheta_{j}(x).)))

for the site j steps from site x. In a semi-formal description of a rule, {\mathbf{x}}{\mathbf{x}}.))) refers to the cell to which it is applied (the “current” cell). In the condition as well as the action of the rule, a field \ensuremath{\mathord{\textsf{\slshape{F\/}}}}({\mathbf{x}})=\eta.\ensuremath{\mathord{\textsf{\slshape{F\/}}}}({\mathbf{x}}) will simply be written as \ensuremath{\mathord{\textsf{\slshape{F\/}}}}. Denote .cap@\ensuremath{\mathord{\textsf{\slshape{F\/}}}}^{j}.)))

An elementary rule has the following form:

A rule is a sequence of elementary rules. The transition function will be defined by a set of rules. In case some rules contradict each other, they will be ordered by explicit priority. Rules will sometimes be associated with certain fields. The lowest-priority rule referring to a field is called the default.

When a sequence of rules is combined, this is a parallel combination: all the rules apply to the same transition, and do not refer to consecutive times. Sometimes we may use the notation "+@\texttt{;},\hbox{$\mskip\medmuskip\parallel\mskip\medmuskip$}.)))

for a combination of n rules. As a shorthand, we may also combine rules in series: the rule

asks for carrying out the rules \ensuremath{\mathord{\texttt{\slshape{R\/}}}}_{1} and \ensuremath{\mathord{\texttt{\slshape{R\/}}}}_{2} consecutively. This can be understood in terms of a field \ensuremath{\mathord{\textsf{\slshape{Age\/}}}} which we are always going to have and can be replaced with a conditional.

In both parallel and serial combination, we will refer to the rules \ensuremath{\mathord{\texttt{\slshape{R\/}}}}_{i} as sub-rules, when they will always be invoked (“called”) by other rules. Still, for a better flow of discussion we will frequently refer to sub-rules as just “rules”. Sub-rules may have parameters that are specified at the time of the call. Rule P(i) with parameter i can be viewed as a different rule for each possible value of i.

Sometimes the conditions of a conditional statement are indexed in a regular way, then we may use the shorthand \textbf{cfor}\textbf{cfor}.))) as below:

So this rule tests the conditions in order, carries out the command of the first satisfied condition and skips the rest. Several conditions that are to be tested one after the other will be combined using the construct cond. This can be seen as an alternative to the “if\dotsthen\dotselsif\dots” construct.

Both the parallel and the consecutive combination of rules can be over an indexed set, and then we get the constructs pfor and for:

Here, \texttt{pfor}.))) the rule is carried out simultaneously for all fields indexed by the indicated values k. for.))) On the other hand in the \textbf{for} example the application is consecutive:

Another use of \textbf{for} can be this:

which, started at some \ensuremath{\mathord{\textsf{\slshape{Age\/}}}}=t will mean that the commands {\mathopen\langle \cdots\mathclose\rangle} will run under the condition t_{1}\le\ensuremath{\mathord{\textsf{\slshape{Age\/}}}}<t+n. repeat.))) The repeat construct could be translated into a for:

We may use temporary constants or functions in the course of defining a rule, as in this example:

let.)))

A location is defined by a pair (\ensuremath{\mathord{\textsf{\slshape{F\/}}}},I) where \ensuremath{\mathord{\textsf{\slshape{F\/}}}} is a field, and I is an interval of addresses in the colony, or by a constant-size union of such pairs (\ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{j},I_{j}). We will denote the track \ensuremath{\mathord{\textsf{\slshape{F\/}}}} over interval I as .cap@\ensuremath{\mathord{\textsf{\slshape{F\/}}}}(I).)))

The name of a location will generally begin with an underscore, like

The set of addresses (typically an interval) of a location L will be denoted by \ensuremath{\mathord{\textit{space\/}}}(L).

The location on the \ensuremath{\mathord{\textsf{\slshape{Info\/}}}} track containing the represented string is denoted by \ensuremath{\mathord{\textsf{\slshape{\_Info\/}}}}..loc@\ensuremath{\mathord{\textsf{\slshape{\_Info\/}}}}.)))

The rule language can contain some definitions of names for constant strings of symbols, of the form \ensuremath{\mathord{\textit{Param\/}}}.)))

where s_{i} are some strings. The names \ensuremath{\mathord{\textit{Param\/}}}_{i} are allowed to be used in the rules. This has the advantage that even if we use, say, \ensuremath{\mathord{\textit{Param\/}}}_{1} twice in a rule, this does not make the rule much longer, even if the value of the parameter is a long string.

Parameters will have descriptive names: say, we might write \ensuremath{\mathord{\textit{Height\/}}} for \ensuremath{\mathord{\textit{Param\/}}}_{1}, when \ensuremath{\mathord{\textit{Param\/}}}_{1} indicates the level in a hierarchy of simulations. Parameters will always be used via the function

which writes the string value of parameter \ensuremath{\mathord{\textit{Param\/}}} to location \ensuremath{\mathord{\textit{loc\/}}}. This can be implemented as a subrule, via the primitive access function

referring to the jth symbol of parameter i. There will also be a special parameter:

whose value is the string that is the sequence of all the rules. This will implement self-reference in the style of Remark 9.4.

The rule \ensuremath{\mathord{\texttt{\slshape{Retrieve\/}}}} uses the locations \ensuremath{\mathord{\textsf{\slshape{\_Retrieved\/}}}}_{m} to deposit the string it retrieves from the neighbor colony number m\in\{-1,0,1\}.

The description \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{1}(\left[a_{1},a_{2}\right)) of a location can fit into a single field \ensuremath{\mathord{\textsf{\slshape{G\/}}}}_{1} of some cell. If for example argument \ensuremath{\mathord{\textit{loc\/}}}_{1} of the rule \ensuremath{\mathord{\texttt{\slshape{Copy\/}}}}(m,\ensuremath{\mathord{\textit{loc\/}}}_{1},\ensuremath{\mathord{\textit{loc\/}}}_{2}) is given by a field \ensuremath{\mathord{\textsf{\slshape{G\/}}}}_{1} this way, then we write \ensuremath{\mathord{\texttt{\slshape{Copy\/}}}}(m,\ensuremath{\mathord{\textsf{\slshape{G\/}}}}_{1}*,\ensuremath{\mathord{\textit{loc\/}}}_{2}). This will only happen if m=0, that is the copying proceeds within one colony. It will be assumed that field \ensuremath{\mathord{\textsf{\slshape{G\/}}}}_{1} of each cell of the colony contains the same information \ensuremath{\mathord{\textit{loc\/}}}_{1}. Therefore the rule can be written just as above, except that some of its parameters are read now from the field \ensuremath{\mathord{\textsf{\slshape{G\/}}}}_{1}.

Here are a few locations used in the rule \ensuremath{\mathord{\texttt{\slshape{Eval\/}}}}:

The encoding \varphi_{*} of a cell state v of M_{2} into a colony of M_{1} is defined as follows. The string v is written into \ensuremath{\mathord{\textsf{\slshape{\_Info\/}}}}. The \ensuremath{\mathord{\textsf{\slshape{Cpt\/}}}} track and the mail tracks are set to all 1’s. Each \ensuremath{\mathord{\textsf{\slshape{Age\/}}}} field is set to 0. For all i, the \ensuremath{\mathord{\textsf{\slshape{Addr\/}}}} field of cell i of the colony is set to i.

For the media to be defined, robust media, we introduce a special set of states \ensuremath{\mathord{\textit{Bad\/}}}\subseteq\ensuremath{\mathord{\textit{Vacant\/}}}\ensuremath{\mathord{\textit{Bad\/}}}.))). For a history \eta we define the damage set \ensuremath{\mathord{\textit{Damage\/}}}.)))

For a space configuration, the damage is defined similarly. For a site x a time interval I is damage-free if \eta(x,\cdot) is not in \ensuremath{\mathord{\textit{Bad\/}}} during I.

We will use the following constants:

Let us define the rectangle

then the corresponding rectangle V^{*} is defined using {B} ^{*},T^{\bullet *}. Two space-time points are said to be too close if they are contained in a single copy (space-time translation) of \betaV. Two sets A_{1},A_{2} are too far if no copy of \gammaV^{*} intersects both. A subset A of the damage is called an island if it is covered by a translation of \betaV and is too far from the rest of damage. In a simulation \eta^{*}=\Phi^{*}(\eta) between two robust media, the set \ensuremath{\mathord{\textit{Damage\/}}}(\eta^{*}) is obtained after we remove all islands of \ensuremath{\mathord{\textit{Damage\/}}}(\eta). We call this definition of \ensuremath{\mathord{\textit{Damage\/}}}(\eta^{*}) the damage map of simulation \Phi^{*}.

Let us fix a history \eta.

For a (non-vacant) cell x, in direction j=\pm 1 there is at most one cell at distance smaller than 2{B}: if it exists and there is no damage between it and x then we denote it also by \vartheta_{j}(x,t,\eta). We will omit \eta, and sometimes even t, from the arguments when it is obvious from the context. By convention, whenever \vartheta_{j}(x,t,\eta) is undefined then let \eta(\vartheta_{j}(x,t,\eta),t)=\ensuremath{\mathord{\textit{Vac\/}}}.

The transition function in a robust medium has the form

Here {\mathbf{s}} = (s_{-1},s_{0},s_{1}), {\mathbf{a}} =(a_{-1},a_{1}) have the following meaning. s_{0} is the state of the cell whose transition will happen, and for j=\pm 1 s_{j} is the state of its (not necessarily adjacent) neighbors in direction j. Further a_{j}=1 if the neighbor in direction j is adjacent and 0 otherwise. The transition does not depend on a_{j} if either s_{j} or s_{0} is vacant. Note that a_{j} is in fact a_{j}(x,t,\eta), but we may omit the arguments that are obvious from the context. In analogy with (2.4), let

The symbol \eta will be omitted from \ensuremath{\mathord{\textit{Tr\/}}}(\eta,x,t) when it is obvious from the context.

A robust medium will be denoted by \ensuremath{\mathord{\textit{Rob\/}}}.)))

The first bit of the state of a cell will be the field \ensuremath{\mathord{\textsf{\slshape{Rand\/}}}}; we will write

Recall the form of the transition function in (10.7). We will call a pair (j,v)\in\{-1,1\}\times{\mathbb{S}}, a potential creator of a non-vacant state s if s=\ensuremath{\mathord{\textit{Tr\/}}}({\mathbf{u}},{\mathbf{a}}) where u_{j}=v, u_{0}= u_{-j}=\ensuremath{\mathord{\textit{Vac\/}}}. The expectation is that an adjacent cell with state s will be created by the cell with state v determined by the state u_{j} of the creator independently of the state u_{-j}.

Consider a history \eta, with a switching time \sigma of cell x when \eta(x,t) turns from vacant to non-vacant. We call y=x+j{B} a creator of x for time \sigma if (j,\eta(y,\sigma-)) is a potential creator of \eta(x,\sigma), further \eta(y,t) is non-vacant for t\in(\sigma-{T_{\bullet}}/2,\sigma+{T_{\bullet}}). (So a creator is supposed to survive the creation.) On the other hand, if there is no t in (\sigma-{T_{\bullet}}/2,\sigma] when the interval\left[x-2{B},x+3{B}\right) intersects the body of any cell, we will say that the cell x emerged at time \sigma.

For a history \eta, cell x and number a \ge 0 let

be defined as follows. The values \sigma_{1},\sigma_{2} are the first two switching times of x in (a,a+2T^{\bullet}], but \sigma_{2} is defined only if \eta(x,\sigma_{1}) is non-vacant. 0 is considered a switching time if \eta(x,0) is not vacant. On the other hand, \sigma_{0} is the first switching time of x after a+T^{\bullet} but defined only if \eta(x,a+T^{\bullet}) is vacant. Whenever we have an event function {g}(\alpha,W,\eta) in whose definition \sigma_{2} occurs, this function is always understood to have value 0 if \sigma_{2} is not defined (similarly with \sigma_{0}).

[Rectangles used in the computation property]

Let .cap@W_{0}(x, a), W_{1}(x, a).)))

We will say that damage affects cell (x,t) directly, if it intersects with I_{0}(x) at time t, with I_{j}(x) as in Definition 10.14. It affects a cell via neighbors, if it intersects I_{1}(x).

Only field \ensuremath{\mathord{\textsf{\slshape{Det\/}}}} will be subdivided into subfields: therefore subfields \eta.\ensuremath{\mathord{\textsf{\slshape{Det\/}}}}.\ensuremath{\mathord{\textsf{\slshape{F\/}}}} will be denoted simply as \eta.\ensuremath{\mathord{\textsf{\slshape{F\/}}}}, without danger of confusion.

Consider an amplifier (M_{k},\Phi_{k}) where \Phi_{k}=(\varphi_{k*},\Phi^{*}_{k}). For a sequence of fields (\ensuremath{\mathord{\textsf{\slshape{Payload\/}}}}_{k}) and parameters D_{k},\varepsilon''_{k}, our amplifiers will use a code complex

as in (5.3), and they will have in addition the trickle-down property as in Definition 5.11 for the fields \ensuremath{\mathord{\textsf{\slshape{F\/}}}}_{k}=\ensuremath{\mathord{\textsf{\slshape{Payload\/}}}}_{k}. We will call such an amplifier an amplifier complex.

In a robust medium M, recall the transition function t=\ensuremath{\mathord{\textit{Tr\/}}}({\mathbf{r}},{\mathbf{a}}) from (10.7). with input vector {\mathbf{r}}=(r_{-1},r_{0},r_{1}) and the adjacency bits {\mathbf{a}}=(a_{-1},a_{1}). We will say that it carries the payload transition function \ensuremath{\mathord{\textit{Pl-trans\/}}}-trans@\ensuremath{\mathord{\textit{Pl-trans\/}}}.))) if the condition t\ne\ensuremath{\mathord{\textit{Vac\/}}}, r_{0}.\ensuremath{\mathord{\textsf{\slshape{Kind\/}}}}=\ensuremath{\mathord{\textit{Germ\/}}}, {\mathbf{a}}=(1,1) implies t.\ensuremath{\mathord{\textsf{\slshape{Payload\/}}}} = \ensuremath{\mathord{\textit{Pl-trans\/}}}({\mathbf{r}}.\ensuremath{\mathord{\textsf{\slshape{Payload\/}}}}).

The string \ensuremath{\mathord{\textit{prog\/}}} will be called a uniform program (with complexity coefficient c) for the functions f_{k} if there is some constant c such that it computes on the universal computing medium \ensuremath{\mathord{\textit{Univ\/}}} the value f_{k}({{\mathbf{r}}}) from k,{{\mathbf{r}}} with space- and time-complexities bounded by c(\log k + |{\mathbf{r}}|) where |{\mathbf{r}}| is the total number of bits in the arguments {\mathbf{r}}. If a sequence (f_{k}) of functions has a uniform program it is called uniform. As a special case, a sequence of constants c_{k} is uniform if it is computable with space- and time complexities c\log k from the index k.

Field \ensuremath{\mathord{\textsf{\slshape{Info\/}}}} consists of the subfields \ensuremath{\mathord{\textsf{\slshape{Info\/}}}}.\ensuremath{\mathord{\textsf{\slshape{Payload\/}}}}, \ensuremath{\mathord{\textsf{\slshape{Info\/}}}}.\ensuremath{\mathord{\textsf{\slshape{Pl-redun\/}}}}, \ensuremath{\mathord{\textsf{\slshape{Info\/}}}}.\ensuremath{\mathord{\textsf{\slshape{Util\/}}}} (referred to mostly as just \ensuremath{\mathord{\textsf{\slshape{Payload\/}}}}, \ensuremath{\mathord{\textsf{\slshape{Pl-redun\/}}}}, \ensuremath{\mathord{\textsf{\slshape{Util\/}}}}). .main@\ensuremath{\mathord{\textsf{\slshape{Info\/}}}}.\ensuremath{\mathord{\textsf{\slshape{Payload\/}}}}, \ensuremath{\mathord{\textsf{\slshape{Info\/}}}}.\ensuremath{\mathord{\textsf{\slshape{Pl-redun\/}}}}, \ensuremath{\mathord{\textsf{\slshape{Info\/}}}}.\ensuremath{\mathord{\textsf{\slshape{Util\/}}}}.))) The track \ensuremath{\mathord{\textsf{\slshape{Payload\/}}}} contains the intended original information, The track \ensuremath{\mathord{\textsf{\slshape{Pl-redun\/}}}} contains “error check bits” for the track \ensuremath{\mathord{\textsf{\slshape{Payload\/}}}} as in Example 5.14, while the \ensuremath{\mathord{\textsf{\slshape{Util\/}}}} track contains other parts of the represented information of the simulated colony (along with its own error checks).

Recall the transition function \ensuremath{\mathord{\textit{Tr\/}}}({\mathbf{r}},{\mathbf{a}}) where {\mathbf{a}}=(a_{-1},a_{1}) says whether the left and right neighbor is adjacent. In our rule language we will refer to a_{j} as

The values of the address field \ensuremath{\mathord{\textsf{\slshape{Addr\/}}}} will vary in \left[-{Q} ,2{Q} \right). The colony of a cell x is the set C=\mathopen\{\verythinmathskipz+i{B} : i\in\left[0,{Q}\right)\mskip 1mu\mathclose\} whose starting cell is z = x-(\ensuremath{\mathord{\textsf{\slshape{Addr\/}}}}(x) \bmod {Q} ) {B}. We will also say that x belongs to colony C. The originating colony of a cell x is the colony D whose starting cell is x-\ensuremath{\mathord{\textsf{\slshape{Addr\/}}}}(x) {B}. We will also say that x originates at colony D. The cells whose colony is their originating colony, will be called inner cells, the other ones will be called outer cells. Two neighbor cells will be called space-consistent if they originate at the same colony.

Cells will be of a few different kinds, distinguished by the field \ensuremath{\mathord{\textsf{\slshape{Kind\/}}}}.)))

The possible kinds are listed in (12.4) below. \ensuremath{\mathord{\textit{Latent\/}}}.))) \ensuremath{\mathord{\textit{Germ\/}}}.)))\ensuremath{\mathord{\textit{Channel\/}}}.)))\ensuremath{\mathord{\textit{Growth\/}}}.))) \ensuremath{\mathord{\textit{Member\/}}}\ensuremath{\mathord{\textit{Member\/}}}.))) The kind of non-latent, non-germ, cells is determined by \ensuremath{\mathord{\textsf{\slshape{Age\/}}}} and \ensuremath{\mathord{\textsf{\slshape{Addr\/}}}}. Since it will be possible to determine from the age of a cell whether it has kind \ensuremath{\mathord{\textit{Channel\/}}} or \ensuremath{\mathord{\textit{Growth\/}}}, We will sometimes use therefore just one value \ensuremath{\mathord{\textit{Ext\/}}}_{j}.)))

in place of \ensuremath{\mathord{\textit{Growth\/}}}_{j} and \ensuremath{\mathord{\textit{Channel\/}}}_{j}. The kinds of cells are ordered by a property called strength as follows:

If a cell needs to be created, overlapping in body a another one then if the new cell is stronger, the weaker one will be erased. This is the only way a cell can turn vacant see the later Condition 10.11 (Cling-to-Life). Let us discuss each kind.

• The relation \ensuremath{\mathord{\textsf{\slshape{Kind\/}}}}(x,t)=\ensuremath{\mathord{\textit{Vac\/}}} means that there is no cell at site x at time t.

• A cell will be called dead if it is vacant or latent, and live otherwise. Killing a cell x means turning it latent. Erasing it means turning it vacant.

• The member cells are “inner” cells as defined above; they are the strongest kind in order to maintain the integrity of colonies.

• Cells of kind \ensuremath{\mathord{\textit{Ext\/}}}_{j} are outer cells. A right outer cell has addresses \ge {Q}, and a left outer cell has addresses <0.

• Left channel cells are weaker than right channel cells which are weaker than growth cells. This way, a right-growing channel will erase the left-growing one if it is in its way, and information need not be transmitted between two channels. Also, left channel cells can grow only until address -{Q} +1, and right ones only until 2{Q} -2: thus a channel cannot be extended to cover a full colony.

• Germ cells exist at the highest level of the hierarchy, in the sense that they are not part of any colony simulating some higher-level cell. They also have the function to carry out a computation, the payload of the whole construction. The member cells on lower levels will receive the results of this computation by a feature of the simulation called trickle-down. Germ cells have addresses in \left[-2{Q} ,3{Q} \right).

The interval before age \ensuremath{\mathinner{\textsf{\upshape{compute-start}}}} is called the retrieval period: then the colony retrieves information from neighbor colonies. The constant K will make it substantially longer than the transition period (\ensuremath{\mathinner{\textsf{\upshape{grow-start}}}},{U}] at the end of the work period when the information to be sent to neighbor colonies would change. The age intervals (0,\ensuremath{\mathinner{\textsf{\upshape{compute-start}}}}] and (\ensuremath{\mathinner{\textsf{\upshape{grow-start}}}},\ensuremath{\mathinner{\textsf{\upshape{grow-end}}}}] for non-germ outer cells and (0,\ensuremath{\mathinner{\textsf{\upshape{grow-end}}}}] for germ cells will be called expansion periods. Cells in their expansion period are called expansion cells.

The application of the rule \ensuremath{\mathord{\texttt{\slshape{Finish\/}}}} will also be called a cut.

Recall the definition of \lambda in (?). We introduce some delay constants {p}_{0} < {p}_{1} < {p}_{2} < {p}_{3} {p}_{i}.))) whose value will be defined below, further :greek@\tau_{i}.)))

Certain fields, called locally maintained, will be kept constant over the colony, for most of the work period. Each such field \ensuremath{\mathord{\textsf{\slshape{F\/}}}} has a default value. The updating will of such a field always happen at a specific age called the update age. Let

be the (constant) number of update ages for locally maintained fields. A cell’s locally maintained field \ensuremath{\mathord{\textsf{\slshape{F\/}}}} is said to be stable if its age is at distance \ge \Delta from the update age, where this constant is defined below in (?). An age n is called stable for \ensuremath{\mathord{\textsf{\slshape{F\/}}}} if \ensuremath{\mathord{\textsf{\slshape{F\/}}}} is stable at age n. It is called simply stable if it is stable for all locally maintained fields, and is also at distance \ge \Delta from ages 0, \ensuremath{\mathinner{\textsf{\upshape{compute-start}}}} and \ensuremath{\mathinner{\textsf{\upshape{grow-end}}}}.

Two neighbor cells x<y with colors c,d will be expected to have fitting colors. This is expressed by the relation \ensuremath{\mathord{\textit{Fit\/}}}_{1}(c,d) which is 1 if the colors fit and 0 otherwise. Let

We will use two examples of the fitting relation:

Two cells x and y=x\pm {B} with |\ensuremath{\mathord{\textsf{\slshape{Addr\/}}}}(x)-{\lfloor {Q} /2\rfloor}|<|\ensuremath{\mathord{\textsf{\slshape{Addr\/}}}}(y)-{\lfloor {Q} /2\rfloor}| belong to the same work period if

They straddle a work period boundary if \ensuremath{\mathord{\textsf{\slshape{Age\/}}}}(x)=0, \ensuremath{\mathord{\textsf{\slshape{Age\/}}}}(y)={U} -1. If one of these cases holds then we will say they are time-consistent.

Two cells x,x+j{B} for j\in\{-1,1\} are siblings if

one of the following properties holds.

with value defined in (?). An interval of cells in which the adjacent cells are siblings will be called a domain. A domain of size n will also be called a n-support of its members. For integer k>0, let us call two cells x,x+k{B} relatives if we can change the states of cells x+j{B} for j=1,\dots,k-1 in such a way that the cells x,x+{B} ,\dots,x+k{B} become a domain. A colony with starting cell x will be called full if it is covered by a domain in such a way that \ensuremath{\mathord{\textsf{\slshape{Addr\/}}}}(x+i{B} )\equiv i\pmod {Q}.

Let us call the left depth of a cell x the size of the largest domain of the form \{x-i {B} , x-(i-1){B} ,\dots,x\} within a single colony containing x. The right depth is defined similarly. For example, \ensuremath{\mathord{\textit{depth\/}}}_{-1}(x)>1 \ensuremath{\mathord{\textit{depth\/}}}.))) just means that x has a left sibling.

For a locally maintained field \ensuremath{\mathord{\textsf{\slshape{F\/}}}} we define the left depth \ensuremath{\mathord{\textit{depth\/}}}_{j,\ensuremath{\mathord{\textsf{\slshape{F\/}}}}}(x) for \ensuremath{\mathord{\textsf{\slshape{F\/}}}} of a cell x the size of the largest interval of the form \{x-i {B} , x-(i-1){B} ,\dots,x\} within a single colony containing x and having constant value of \ensuremath{\mathord{\textsf{\slshape{F\/}}}}.

A cell will always try to keep track of its depths, up to the maximum possible value. It will use the fields \ensuremath{\mathord{\textsf{\slshape{Depth\/}}}}.)))

So if for example left depth is larger than \ensuremath{\mathinner{\textsf{\upshape{max-depth}}}} then the cell may only see \ensuremath{\mathord{\textsf{\slshape{Depth\/}}}}_{-1}=\ensuremath{\mathinner{\textsf{\upshape{max-depth}}}}.

There is

a field \ensuremath{\mathord{\textsf{\slshape{Frozen\/}}}}.)))

with the property that when \ensuremath{\mathord{\textsf{\slshape{Frozen\/}}}}\ne 0 then \ensuremath{\mathord{\textsf{\slshape{Age\/}}}} will not be changed. A cell with \ensuremath{\mathord{\textsf{\slshape{Frozen\/}}}}\ne 0 is called frozen.

Suppose that cell x has no siblings in direction j, that is it has \ensuremath{\mathord{\textit{depth\/}}}_{j}(x)=1. It will be called a protected edge in that direction if it is some legitimate boundary in that direction, in the sense described below; otherwise, it will be called an exposed edge. Recall expansion periods and cells in Definition 12.7. Essentially, growth edges are protected during the expansion period or if the growth reached a colony end. Here is a list of the types of protected edges, for non-germ cells. For germ cells, the definition will be given in Section ?.

• Member Colony end-cell towards j.

• Expansion cell During the expansion period, while j is the direction of expansion. After the expansion period, if the cell is a growth cell and a colony end-cell towards j.

In a rule, the condition \ensuremath{\mathord{\textit{Xposed\/}}}_{j}.)))

means that the cell {\mathbf{x}} applying the rule is an exposed edge in direction j.

A multi-domain is one of the following kinds of set:

We will have some one-bit fields \ensuremath{\mathord{\textsf{\slshape{Dying\/}}}}.)))

with default value 0. We will say that a cell x is dying if \ensuremath{\mathord{\textsf{\slshape{Dying\/}}}}_{0}(x)=1.

For a relation R, let "<@a\stackrel{\scriptstyle R}{<} b.)))

mean “a<b or (a=b and R holds)”.

Let j\in\{-1,1\}. The field \ensuremath{\mathord{\textsf{\slshape{Kind\/}}}}_{j} shows the kind of the cell intended to be created in the adjacent neighbor in direction j. Let

If a latent cell x came to life by the rule \ensuremath{\mathord{\texttt{\slshape{Adapt\/}}}}(j) then we will say that it has been animated. The neighbor in direction j will be called its parent. (The latent cell could have been created earlier by a neighbor, which we could call a “creating parent”.) In case the same result could also have arisen using the neighbor on the other side then the parent cell is defined as the one closer to the center of its colony (or to the left if the animated cell is the center).

Let {e}_{j}.)))

A cell x is a colony endcell in direction j if \ensuremath{\mathord{\textsf{\slshape{Addr\/}}}}(x)\equiv {e}_{j} \pmod {Q}.

We have a field called

Its default value is 0. A cell with \ensuremath{\mathord{\textsf{\slshape{Healing\/}}}}>0 will be called a healing cell. The field

will limit the healing process in time.

For times t<u, assume cell x is live at t and its body is damage-free during [t,u], and there are no switching times in the open interval (t,u). Then we say that point (x,t) is connected by a vertical link to point (x,u). If one end of a vertical link is not a switching time, then the link is called short. If cells x,x+{B} are siblings at time t or (t-) such that \left[x-{B} ,x+2{B} \right) is damage-free during (t-T^{\bullet},t], then the points (x,t), (x+{B} ,t) are said to be connected by a horizontal link .

If point (y,t') is, according to case ? of Lemma 14.18 (Parent) a parent of point (x,u), we will say that (y,t') is connected by a parental (maternal or paternal) link to point (x,u). A link is a link of one of these kinds. A link is steep or, equivalently, slow if it is a non-short vertical link or a parental link. (Since time is the second coordinate, steepness of a line is synonymous to slowness of the movement of a point along it.)

A sequence (x_{0},t_{0}), \dots, (x_{n},t_{n}) with t_{i}\le t_{i+1} such that subsequent points are connected by links, is called a path.

A path with only steep links is steep, or slow. A backward path is the reversed reading of a forward path, backwards in time. The adjective “forward” or “backward” will be omitted, when it is obvious from the context. For a path P=(x_{0},t_{0}),\dots,(x_{n},t_{n}) and t\in[t_{0},t_{n}], let :cap@P(t).)))

be x_{i} for the smallest i with t\in[t_{i},t_{i+1}].

A cell (x,t) will be called traceable if I_{0}(x) is damage-free during (t-4\ensuremath{\mathinner{\textsf{\upshape{purge-t}}}},t].

A backward path is called a trace-back path if every link in it is chosen to be a vertical or a parental link if this is possible, with horizontal links only as needed in Lemma 15.5 (Skirting).

Let P=\mathopen( c_{0},\dots,c_{n}\mathclose) be a forward path. The age progress of P is defined as

Consider an open interval G=(l,r) at some time t, where 0 < r - l is divisible by {B}.

We say that G is a right gap with size |G|=r-l-{B} if every traceable cell in G can be space-consistent with r only if it is a germ cell. The right age of the gap G is is the upper bound of the age of these germ cells (it is 0 if the gap contains no germ cells). If G is contained in the colony of r then it is called an interior gap.

A right bad gap is a gap of right-age

containing an interior gap of size \ge\ensuremath{\mathinner{\textsf{\upshape{gap-lb}}}}{B}. If the gap itself is not interior then its right edge is required to be an exposed right outer cell. Left gaps are defined similarly.