This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

[1,2]Franco Bagnoli

\equalcont

These authors contributed equally to this work.

\equalcont

These authors contributed equally to this work.

[1]Department Physics and Astronomy and CSDC, University of Florence, Via G. Sansone,1, Sesto Fiorentino, 50019, Italy

2]INFN, Sect. Florence

3] Institute of Optics and Precision Mechanics, University of Setif 1, Algeria

4]Université de Lorraine, CNRS, Inria, LORIA, Nancy, 54000, France

Regional Controllability of Cellular Automata as a SAT Problem

franco.bagnoli@unifi.it    Sara Dridi sara.dridi@univ-setif.dz    Nazim Fatès nazim.fates@loria.fr * [ [ [
Abstract

Controllability, one of the fundamental concepts in control theory, consists in guiding a system from an initial state to a desired one within a limited (and possibly minimum) time interval. When the objective is limited to a specific sub-region of the system’s domain, the concept is referred to as regional controllability.

We examine this notion in the context of Boolean one-dimensional cellular automata of finite length. Depending on the local evolution rule, we investigate whether it is possible to control the evolution of the system by imposing particular values on the boundary conditions. This approach is related to key dynamical properties of CA, specifically chain transitivity and chain mixing. We show that the control problem can be formulated as a Boolean satisfiability (SAT) problem and can thus be addressed using SAT solvers. We also show how finding shortest paths in the configuration graph allows to determine controllability properties. From our observations we can state that only peripherally-linear rules are fully controllable, while for other rules, the reachability ratio, that is, the fraction of controllable pairs of initial and final configurations, is vanishing when the system size grows.

keywords:
Cellular automata, regional controllability, boolean satisfiability problem, chain transitive, chain mixing.

1 Introduction

Control theory is a branch of mathematics and engineering that deals with the behaviour of dynamical systems and how this behaviour can be modified with some influence that takes the form of a feedback. In other words, to control a system means to influence its behaviour so as to steer it to a desired state. Control theory is widely used in large number of fields ranging from aerospace, robotics, electrical engineering to social science.

Many studies have explored the control of dynamical systems with continuous variables and continuous-time evolution addressing both finite- and infinite-dimensional cases. These systems are typically modelled and analysed using differential equations and partial differential equations [1, 2].

Controllability, introduced by Kalman in 1960, is one of the fundamental concepts in control theory. It explores whether a system can be guided from any initial state to a desired state within a predefined time interval [0,T][0,T]. Since the, this concept has been extensively studied for finite-dimensional systems [3] and infinite-dimensional systems described by partial differential equations (PDE) [4, 5].

More recently, controllability has been studied in the context of cellular automata (CA – acronym also used to define single cellular automaton), which are considered as potential alternatives to classical models based on partial differential equations, as they can effectively capture non-linear phenomena through simple local rules.

Cellular automata are discrete dynamical systems regarded as the simplest models of spatially extended systems which can offer an effective framework for describing complex phenomena. They consist of three components: a grid of cells, each taking a state from a finite set, a neighbourhood (for each cell) and a local transition function. The CA paradigm has been successfully applied to a wide range of fields, including biology, chemistry, physics, and ecology, as evidenced by an extensive literature on the subject [6, 7, 8]. For a general overview, see the proceedings of the ACRI conference [9].

The focus of this paper is on a specific aspect of controllability, known as regional controllability, where the goal is to achieve a desired objective only on a part of the whole domain by applying actions on its boundaries. Regional controllability of cellular automata focuses on the ability to guide the state of a system toward desired configurations within a specific subregion of the entire domain. This concept was introduced by El Jai and Zerrik in (1993-1995) and well studied in a wide range of works in the context of distributed parameter systems described by partial differential equations. It is particularly relevant for this type of systems to address situations where full domain control is either unachievable or unnecessary, while control within a specific region remains achievable.

When switching from a continuous description (PDE) to a discrete one (coupled maps or CA), we have also to introduce the system size nn as a relevant parameter.

In the context of CA, different characterization results have been proposed to extend or substitute the widely used Kalman criterion and the regional controllability problem was analysed with various methods to establish new criteria for the case of cellular automata and discrete complex systems [10].

Our research primarily focused on the regional controllability of Boolean CA, demonstrating its validity through the use of Markov chains and graph theory tools [11, 12, 13]; see Ref. [14] for a general overview.

The problem of regional controllability has been addressed with an approach based on the Kalman condition [15, 16]. For one-dimensional deterministic cellular automata, the problem has been explored using the concept of Boolean derivatives [17] and the probabilistic case has been investigated with a Markov-chain approach [18, 19]. In a recent work, a novel characterization of controllability and regional controllability based on symbolic dynamics was introduced [20]. This study establishes that the regional controllability for every nn (where nn is the size of the controlled region) and controllability of cellular automata are equivalent to the topological properties, namely those of chain transitivity and chain mixing.

In another recent work, a preimage algorithm was used to determine whether a desired configuration can be reached from an initial configuration, using a characterization tool known as the controllability tree [21]. This paper builds upon these findings.

In this work, we deepen our exploration of the regional controllability of elementary cellular automata by formulating this problem as a satisfiability problem (SAT). We also provide new techniques such as finding the minimum path in the directed configuration graph.

The structure of the paper is as follows. Section 2 provides an overview of the definitions of elementary cellular automata, introduces the regional controllability problem and the connection between chain transitivity, chain mixing and regional controllability. Section 3 formulates the problem of regional controllability for one-dimensional cellular automata as a SAT problem. Furthermore, this method can also be used to explore key dynamical properties of cellular automata, particularly chain transitivity and chain mixing. Section 4 is devoted to the generation of preimages of a given configuration, given the controls, and in Section 5, we present another technique for CA control, focusing on the process of finding the shortest path in the directed configuration graph, achieved by constructing trees of images and preimages corresponding to the starting and target configurations for all possible control inputs. Finally, conclusions are drawn in Section 6.

2 Definitions of the problem and mathematical properties

2.1 Elementary Cellular Automata

Elementary cellular automata (ECA – acronym also used to define a single automaton) are discrete dynamical systems for which the state of each cell only takes two values and is determined at each time step by its own state and the state of left and right neighbours. The evolution of the states of all cells occurs in parallel.

We first consider a one-dimensional infinite set of cells. The state of each cell ii\in\mathbb{Z} at time tt is given by a variable xit{0,1}x_{i}^{t}\in\{0,1\}. Mathematically, the evolution of the cells is defined in terms of a local function f{0,1}3{0,1}:f\{0,1\}^{3}\rightarrow\{0,1\}:

i,xit+1=f(xi1t,xit,xi+1t).\forall i\in\mathbb{Z},x^{t+1}_{i}=f(x^{t}_{i-1},x^{t}_{i},x^{t}_{i+1}).

For a given time tt, we denote by 𝒙t=(xit)i\boldsymbol{x}^{t}=(x^{t}_{i})_{i\in\mathbb{Z}} the sequence of all cell states, i.e., a configuration. In the rest of the paper we consider only finite configurations since we will focus on a particular set of cells, or a region, not taking into consideration what happens outside it.

Table 1: The look-up table of Rules 150 and 22. 𝒩\mathcal{N} is the neighbourhood in the base-two representation, 𝒩(10)\mathcal{N}^{(10)} is the same in then base-ten representation. The column f(150)f^{(150)} is the output of Rule 150 and D𝒩(150)D^{(150)}_{\mathcal{N}} the derivative of the Rule 150 in zero with respect to the ones in the neighbourhood in the base-two representation, same for f(22)f^{(22)} and D(22)D^{(22)} for Rule 22.
𝒩=(x1,x0,x1)\mathcal{N}=(x_{-1},x_{0},x_{1}) 𝒩(10)=4x1+2x0+x1\mathcal{N}^{(10)}=4x_{-1}+2x_{0}+x_{1} f(150)(𝒩)f^{(150)}(\mathcal{N}) D𝒩(150)D^{(150)}_{\mathcal{N}} f(22)(𝒩)f^{(22)}(\mathcal{N}) D𝒩(22)D^{(22)}_{\mathcal{N}}
0,0,00,0,0 0 0 0 0 0
0,0,10,0,1 11 11 11 11 11
0,1,00,1,0 22 11 11 11 11
0,1,10,1,1 33 0 0 0 0
1,0,01,0,0 44 11 11 11 11
1,0,11,0,1 55 0 0 0 0
1,1,01,1,0 66 0 0 0 0
1,1,11,1,1 77 11 0 0 11

Since there are 23=82^{3}=8 different neighbourhood states, there are 28=2562^{8}=256 different ECA rules. It is usual to associate to each ECA ff its decimal code WW, defined by W(f)=f(0,0,0)20++f(1,1,1)27W(f)=f(0,0,0)2^{0}+\dots+f(1,1,1)2^{7}  [22, 23]. This amounts to writing the digits of the transition table f(0,0,0)f(0,0,0)f(1,1,1)f(1,1,1) and converting the binary number to a decimal number. As an example, Rule 150 corresponds to the array 1,0,0,1,0,1,1,01,0,0,1,0,1,1,0, which is equal to 150 in base ten (see Table 1).

One can apply two symmetries on the rule, the left-right inversion and the one-zero exchange, which gives 88 classes. The rules with smallest decimal code in a class are called minimal CA and it is usual to consider only these rules for studying ECA (see the left column of Table LABEL:tab:fraction).

2.2 Definition of the regional controllability problem

Refer to caption
Figure 1: General view of the regional controllability problem of one-dimensional CA via boundary actions for n=6n=6 controlled cells. The initial and final configurations are respectively in yellow and cyan. Time goes from bottom to top. Colour Online

In our problem, the definition of a regionally-controllable CA will be set for an arbitrary number of cells, with the goal of determining the asymptotic behaviour. For a given number cells nn, we will focus on the region represented by the set of nn cells indexed from 1 to nn and try to control the evolution of this region, without setting in advance the number of steps TT. We are also interested in looking for the minimal value of T(𝒙,𝒚)T(\boldsymbol{x},\boldsymbol{y}) for a given pair of configurations (𝒙,𝒚)(\boldsymbol{x},\boldsymbol{y}) and the minimal value for a region size nn

T(n)=min{T(𝒙,𝒚):𝒙,𝒚{0,1}n},T(n)=\min\{T(\boldsymbol{x},\boldsymbol{y}):\boldsymbol{x},\boldsymbol{y}\in\{0,1\}^{n}\},

where {0,1}n\{0,1\}^{n} denotes the set of finite words (configurations) of length nn over {0,1}\{0,1\},

Given a number of cells nn and two finite sub-configurations 𝒙=(x1,,xn)\boldsymbol{x}=(x_{1},\dots,x_{n}) and 𝒚=(y1,,yn){0,1}n\boldsymbol{y}=(y_{1},\dots,y_{n})\in\{0,1\}^{n}, our objective is to determine if there exists an appropriate control sequence on the boundary cells, namely cell 0 and cell n+1n+1, such that the system will evolve from 𝒙\boldsymbol{x} to 𝒚\boldsymbol{y} in a finite number of time steps (see Fig. 1 for an illustration).

Mathematically, this amounts to saying that:

A CA is regionally controllable, if there exists N+N\in\mathbb{N}^{+} such that for every nNn\geq N and for every pair of configurations 𝐱=𝐱0,𝐲{0,1}n\boldsymbol{x}=\boldsymbol{x}^{0},\boldsymbol{y}\in\{0,1\}^{n}, there exists T>0T>0 and a control vector 𝐮=(u0,,uT1)\boldsymbol{u}=(u^{0},\dots,u^{T-1}) where ut=(x0t,xn+1t)u^{t}=(x_{0}^{t},x_{n+1}^{t}) such that

𝒙T=𝒚,\boldsymbol{x}^{T}=\boldsymbol{y},

that is, 𝐲\boldsymbol{y} is reachable from 𝐱\boldsymbol{x} in TT steps, with

t{1,,T},i{1,,n},xit+1=f(xi1t,xit,xi+1t).\forall t\in\{1,\dots,T\},\forall i\in\{1,\dots,n\},\,\,x^{t+1}_{i}=f(x^{t}_{i-1},x^{t}_{i},x^{t}_{i+1}).

In the case where a rule is not regionally controllable, we are interested in determining the reachability ratio ρ(n){\rho}(n), that is, the ratio of controllable pairs of initial and final configurations of size nn:

ρ(n)=122ncard{(𝒙,𝒚){0,1}n×{0,1}n:𝒚 is reachable from 𝒙}.{\rho}(n)=\frac{1}{2^{2n}}\operatorname{card}\{(\boldsymbol{x},\boldsymbol{y})\in\{0,1\}^{n}\times\{0,1\}^{n}:\boldsymbol{y}\text{ is reachable from }\boldsymbol{x}\}.

2.3 Peripheral linearity: a sufficient condition to be controllable

Let us now examine under which conditions elementary cellular automata can be controlled.

To this end, we introduce the notion of a derivative of a Boolean function f(x,y)f(x,y) [24] as:

fx=f(x1,y)f(x,y),\frac{\partial f}{\partial x}=f(x\oplus 1,y)\oplus f(x,y),

where \oplus stands for the sum modulo two (equivalent to an exclusive OR). This definition obeys many standard properties of derivatives, like the chain rule [25]. It is possible also to define higher-order derivatives.

By means of Boolean derivatives in zero one can obtain the Ring Sum Expansion of a function [26]:

f(x0,x1,x2)=D0D1x0D2x1D4x2D3x0x1D5x0x2D6x1x2D7x0x1x2,\begin{split}f(x_{0},x_{1},x_{2})=&D_{0}\oplus D_{1}x_{0}\oplus D_{2}x_{1}\oplus D_{4}x_{2}\oplus D_{3}x_{0}x_{1}\oplus\\ &D_{5}x_{0}x_{2}\oplus D_{6}x_{1}x_{2}\oplus D_{7}x_{0}x_{1}x_{2},\end{split} (1)

where DiD_{i} is the derivative of ff with respect to the bits that have value 1 in the binary representation of ii.

For instance

D5=D1,0,1=2fx0x2(0)=f(1,0,1)f(1,0,0)f(0,0,1)f(0,0,0).D_{5}=D_{1,0,1}=\frac{\partial^{2}f}{\partial x_{0}\partial x_{2}}(0)=f(1,0,1)\oplus f(1,0,0)\oplus f(0,0,1)\oplus f(0,0,0).

Rules that have all derivatives of order greater than one equal to zero (i.e., D3=D5=D6=D7=0D_{3}=D_{5}=D_{6}=D_{7}=0) are called affine, and linear if f(0,0,0)=0f(0,0,0)=0. For instance, Rule 150 is a linear rule since it can be written as

f(150)(x0,x1,x2)=x0x1x2,f^{(150)}(x_{0},x_{1},x_{2})=x_{0}\oplus x_{1}\oplus x_{2},

while Rule 22, although quite similar to Rule 150 (see Table. 1), is not affine:

f(22)(x0,x1,x2)=x0x1x2x0x1x2.f^{(22)}(x_{0},x_{1},x_{2})=x_{0}\oplus x_{1}\oplus x_{2}\oplus x_{0}x_{1}x_{2}.

The Boolean derivatives in zero and the affinity of all minimal ECA are reported in Table 2 of Ref. [21].

We can now define the property of linearity with respect to the periphery of the neighbourhood, which, for ECA, amounts to saying that there exists a function gg such that:

f(x0,x1,x2)=x0g(x1,x2)orf(x0,x1,x2)=g(x0,x1)x2.f(x_{0},x_{1},x_{2})=x_{0}\oplus g(x_{1},x_{2})\qquad\mathrm{or}\qquad f(x_{0},x_{1},x_{2})=g(x_{0},x_{1})\oplus x_{2}.

In term of Boolean derivatives, a rule is peripherally-linear if D1=1D_{1}=1 and D3=D5=D7=0D_{3}=D_{5}=D_{7}=0 or D4=1D_{4}=1 and D5=D6=D7=0D_{5}=D_{6}=D_{7}=0. In the ECA space, the peripherally-linear rules are Rules 15, 30, 45, 60, 90, 105, 106, 150, 154, and 170.

As shown in Refs. [17, 15], peripherally-linear rules are fully controllable.

2.4 Chain transitivity, chain mixing and regional controllability

Let us recall the definitions of ε\varepsilon-chains, chain transitivity and chain mixing.

Let ε>0\varepsilon>0 and let 𝒙,𝒚A\boldsymbol{x},\boldsymbol{y}\in A^{\mathbb{Z}} where AA is the alphabet. An ε\varepsilon-chain (ε\varepsilon-pseudo-orbit) from 𝒙\boldsymbol{x} to 𝒚\boldsymbol{y} is a finite sequence of configurations {𝒙0,𝒙1,,𝒙t}\{\boldsymbol{x}^{0},\boldsymbol{x}^{1},\dots,\boldsymbol{x}^{t}\} with 𝒙0=𝒙\boldsymbol{x}^{0}=\boldsymbol{x} and 𝒙T=𝒚\boldsymbol{x}^{T}=\boldsymbol{y}, such that for T>0T>0 and d(F(𝒙t),𝒙t+1)<εd(F(\boldsymbol{x}^{t}),\boldsymbol{x}^{t+1})<\varepsilon, for t{0,,T1}t\in\{0,\dots,T-1\}, where the distance dd is defined as:

d(𝒙,𝒚)=2min{|k|:xkyk}d(\boldsymbol{x},\boldsymbol{y})=2^{-\mathrm{min}\left\{|k|:x_{k}\neq y_{k}\right\}}

and F:AAF:A^{\mathbb{Z}}\rightarrow A^{\mathbb{Z}} is the global transition function.

A cellular automaton is chain-transitive if for all 𝒙,𝒚A\boldsymbol{x},\boldsymbol{y}\in A^{\mathbb{Z}} and ε>0\varepsilon>0 there exists an ε\varepsilon-chain from 𝒙\boldsymbol{x} to 𝒚\boldsymbol{y} [27]. Similarly, a cellular automaton is considered to be chain-mixing if, for any two configurations 𝒙,𝒚A\boldsymbol{x},\boldsymbol{y}\in A^{\mathbb{Z}} and ε>0\varepsilon>0 there exists T>0T>0 such that for all tTt\geq T, there exists an ε\varepsilon-chain of length tt from 𝒙\boldsymbol{x} to 𝒚\boldsymbol{y} [27].

Recall that regional controllability refers to the ability to steer a dynamical system in a specific region from any initial configuration to any desired configuration in that region within a finite time TT using appropriate inputs (control). Meanwhile, chain transitivity means that for any two points in the state space, one can find a sequence of pseudo-orbits (arbitrary small jumps) connecting them. The connection between chain transitivity, chain mixing and regional controllability of cellular automata was investigated and a proof has been recently proposed to establish an equivalence between regional controllability for every nn ( where nn is the size of the controlled region), the chain-transitivity and chain-mixing properties of one-dimensional cellular automata [20]. It has been shown that a CA is regionally controllable for every nn if and only if it is chain-transitive and chain-mixing.

The link between these concepts can be understood through graph theory, by modelling controlled cellular automata as directed graph, where nodes denotes the finite configurations in AnA^{n} and the arcs represent transitions between them, governed by system dynamics or external inputs. In graph terms, the regional controllability concept corresponds to strong connectivity of the state transition graph meaning that any state can be reached from any other state via a directed path between nodes [11, 10]. It has been shown that when this connectivity property is verified for every nn, it is equivalent to chain transitivity and chain mixing [20]. For a finite cellular automaton the relationship also remains valid.

This means that the value of ϵ\epsilon controls the number of cells nn that we have in our system and that applying the control sufficiently far from the centre cell (say at distance n/2n/2) guarantees that we obtain a valid ε\varepsilon-chain. Mathematically, let us consider a configuration 𝒙t\boldsymbol{x}^{t} and its image 𝒙t+1=F(𝒙t)\boldsymbol{x}^{t+1}=F(\boldsymbol{x}^{t}). Applying a control on 𝒙t+1\boldsymbol{x}^{t+1} on cells n/2-n/2 and n/2n/2 transforms this configuration into a configuration 𝒚\boldsymbol{y}. It is easy to see that we have d(F(𝒙t),𝒚)12n/2d(F(\boldsymbol{x}^{t}),\boldsymbol{y})\leq\frac{1}{2^{n/2}}, which exactly corresponds to the notion of ϵ\epsilon-chain for ϵ=12n/2\epsilon=\frac{1}{2^{n/2}}. In other words, we need to apply the control at a larger distance as the value of ϵ\epsilon gets smaller in the ϵ\epsilon-chain.

According to the equivalence presented in Ref. [20], the methods used to investigate the regional controllability problem can thus also be used to verify the chain transitivity and chain mixing properties.

Refer to caption
Figure 2: One step of the generation of the preimages (control tree) of configuration 0101 for Rule 150. Since Rule 150 is (doubly) peripherally linear, there is no failures nor additional forking.

3 Modelling the control problem as a SAT problem

Refer to caption
Refer to caption
Figure 3: (left) Modelling of the controllability problem with Rule 30, initial condition 010001, final condition 101010 and T=7T=7 time steps. Time goes from bottom to top. The control variables are shown in green and the controlled variables in pink. The values of two control variables x07x^{7}_{0} and x77x^{7}_{7} are irrelevant and are left in grey. (right) A solution that was found by the SAT solver. White and blue respectively correspond to 0 and 1, which are respectively associated to False (F) and True (T). Color online

We can now investigate whether the control problem is solvable for non-peripherally linear rules. In this section, we reformulate the regional controllability of cellular automata problem as a SAT problem.

3.1 Modelling the problem

Recall that we have nn cells in the target region with indices ranging from 1 to nn, and two external cells with index 0 and n+1n+1 as the border controls. In the CA world, xit{0,1}x_{i}^{t}\in\{0,1\} describe the state of cell i{1,n}i\in\{1,n\} at time tt and x0tx_{0}^{t} and xn+1tx^{t}_{n+1} respectively describe the state of left and right controls at time tt. To go to the SAT universe, we map each binary value xitx_{i}^{t} to a Boolean value bitb_{i}^{t}, with tt\in\mathbb{N} and i{0,,n}i\in\{0,\dots,n\} such that bit=(xit)b_{i}^{t}={\mathcal{B}}(x_{i}^{t}), where {\mathcal{B}} is the function from {0,1}\{0,1\} to {False,True}\{\texttt{False},\texttt{True}\} such that (0)=False{\mathcal{B}}(0)=\texttt{False} and (1)=True{\mathcal{B}}(1)=\texttt{True}.

Recall that we fix the values 𝒙=(x1,,xn)\boldsymbol{x}=(x_{1},\dots,x_{n}) and that we want to reach 𝒚=(y1,,yn)\boldsymbol{y}=(y_{1},\dots,y_{n}) in TT time steps. This is equivalent to saying that there exists two sequences (𝒙0t)t{0,,T1}(\boldsymbol{x}_{0}^{t})_{t\in\{0,\dots,T-1\}} and (𝒙n+1t)t{0,,T1}(\boldsymbol{x}_{n+1}^{t})_{t\in\{0,\dots,T-1\}} such that:

t{0,,T1},i{1,,n},xit+1=f(xi1t,xit,xi+1t).\forall t\in\{0,\dots,T-1\},\forall i\in\{1,\dots,n\},\,\,x_{i}^{t+1}=f(x_{i-1}^{t},x_{i}^{t},x_{i+1}^{t}). (2)

and xiT=yix_{i}^{T}=y_{i} for all i{1,,n}i\in\{1,\dots,n\}, that is, 𝒚\boldsymbol{y} is reachable from 𝒙\boldsymbol{x} in TT steps.

We now need to transform this mathematical relationship into a Boolean formula. Such a formula is usually expressed as a combination of variables, or atoms, and operators. A clause is a special case of formula where atoms and their negations are combined using only OR operators, for instance f(a,b,c)=a¬bcf(a,b,c)=a\vee\neg b\vee c. A formula is in conjunctive normal form (CNF) if it is expressed as a conjunction of clauses, for instance: f(a,b,c,d)=(a¬b)(b¬c¬d)f(a,b,c,d)=(a\vee\neg b)\wedge(b\vee\neg c\vee\neg d). SAT solvers take a CNF as an input and try to compute an assignment of the variables that makes the formula true, or, when this is not possible, try to find a proof that there this no such assignment.

To encode the CNF that verifies the regional controllability problem, we proceed in two steps (see Fig. 3).

a) For each ii and tt, we encode the condition xit+1=f(xi1t,xit,xi+1t)x_{i}^{t+1}=f(x_{i-1}^{t},x_{i}^{t},x_{i+1}^{t}) as a CNF composed of eight clauses which use the four variables bit+1b_{i}^{t+1}, bi1tb_{i-1}^{t}, bitb_{i}^{t}, and bi+1tb_{i+1}^{t}.

Indeed, each neighbourhood state will generate a clause:

(xi1t,xit,xi+1t)=(0,0,0)xit+1=f(0,0,0),(xi1t,xit,xi+1t)=(0,0,1)xit+1=f(0,0,1),,(xi1t,xit,xi+1t)=(1,1,1)xit+1=f(1,1,1).\begin{split}(x_{i-1}^{t},x_{i}^{t},x_{i+1}^{t})&=(0,0,0)\implies x_{i}^{t+1}=f(0,0,0),\\ (x_{i-1}^{t},x_{i}^{t},x_{i+1}^{t})&=(0,0,1)\implies x_{i}^{t+1}=f(0,0,1),\dots,\\ (x_{i-1}^{t},x_{i}^{t},x_{i+1}^{t})&=(1,1,1)\implies x_{i}^{t+1}=f(1,1,1).\end{split}

As ABA\implies B is equivalent to ¬AB\neg A\vee B, the conditions becomes

xi1t0xit0xi+1t0xit+1=f(0,0,0),,x_{i-1}^{t}\neq 0\vee x_{i}^{t}\neq 0\vee x_{i+1}^{t}\neq 0\vee x_{i}^{t+1}=f(0,0,0),\dots,

that is,

xi1t=1xit=1xi+1t=1xit+1=f(0,0,0),x_{i-1}^{t}=1\vee x_{i}^{t}=1\vee x_{i+1}^{t}=1\vee x_{i}^{t+1}=f(0,0,0),\dots

which translates into

bi1tbitbi+1tC[bit+1,f(0,0,0)],b_{i-1}^{t}\vee b_{i}^{t}\vee b_{i+1}^{t}\vee C[b_{i}^{t+1},f(0,0,0)]\dots,

where C[b,1]=bC[b,1]=b and C[b,0]=¬bC[b,0]=\neg b.

We thus translate the nTn\cdot T transitions described in Eq. (2) into 8nT8\cdot n\cdot T clauses.

b) In a second step, we encode the initial and final conditions (bi0)(b_{i}^{0}) and (biT)(b_{i}^{T}) for i{1,,n}i\in\{1,\dots,n\} and let the control variables free of any constraint.

By combining the conditions obtained in step a) and step b), we obtain a CNF formula Φ\Phi with 8nT+2n8nT+2n clauses and such that Φ\Phi is satisfiable if and only if 𝒚\boldsymbol{y} is reachable from 𝒙\boldsymbol{x} in TT time steps.

3.2 Experimental results

Table 2: Table showing the controllability statistics for the 88 ECA for n=40n=40, T=100T=100 and 100 random pairs of initial and final configurations. columns : ρ~\tilde{\rho} is the average reachability ; time is the cumulative CPU time ; R, C, D respectively represent the cumulative number of restarts, conflicts and decisions of the SAT solver.
ECA ρ~\tilde{\rho} time R C D ECA ρ~\tilde{\rho} time R C D
0 0.00 1.874 0 0 0 1 0.00 1.870 0 0 0
2 0.00 1.866 0 0 0 3 0.01 1.838 1 0 1
4 0.00 1.834 0 0 0 5 0.00 1.864 0 0 0
6 0.00 1.922 0 0 0 7 0.00 1.765 3 39 143
8 0.00 1.808 0 0 0 9 0.00 1.942 0 0 0
10 0.00 1.985 0 0 0 11 0.00 1.870 0 0 0
12 0.00 1.757 0 0 0 13 0.00 2.073 0 0 0
14 0.01 1.825 1 14 451 15 1.00 2.190 100 0 100
18 0.00 1.949 0 0 0 19 0.00 1.867 0 0 0
22 0.07 6.727 929 281123 780831 23 0.00 1.852 0 0 0
24 0.00 2.018 0 0 0 25 0.00 2.182 11 151 838
26 0.05 2.085 42 4554 28485 27 0.00 1.916 0 0 0
28 0.00 1.930 0 0 0 29 0.00 2.140 0 0 0
30 1.00 3.158 100 0 100 32 0.00 2.095 0 0 0
33 0.00 2.202 0 0 0 34 0.00 2.008 0 0 0
35 0.00 1.987 0 0 0 36 0.00 2.161 0 0 0
37 0.00 2.142 23 163 495 38 0.00 2.092 0 0 0
40 0.00 2.072 0 0 0 41 0.01 3.488 221 72085 177628
42 0.04 1.937 4 0 4 43 0.01 1.849 1 41 2190
44 0.00 1.795 0 0 0 45 1.00 2.506 100 0 100
46 0.00 1.867 0 0 0 50 0.00 1.923 0 0 0
51 0.00 1.760 0 0 0 54 0.00 1.947 3 148 1023
56 0.00 1.813 0 0 0 57 0.00 1.870 0 0 0
58 0.00 1.898 0 0 0 60 1.00 2.093 100 0 100
62 0.00 2.061 10 555 1883 72 0.00 1.722 0 0 0
73 0.00 1.809 0 0 0 74 0.00 1.866 3 18 37
76 0.00 1.763 0 0 0 77 0.00 1.790 0 0 0
78 0.00 1.759 0 0 0 90 1.00 2.117 100 0 100
94 0.00 1.777 0 0 0 104 0.00 1.811 0 0 0
105 1.00 2.485 100 0 100 106 1.00 2.553 100 0 100
108 0.00 1.914 0 0 0 110 0.04 7.009 667 222617 499441
122 0.00 1.874 0 0 0 126 0.03 3.445 249 62888 183969
128 0.00 1.795 0 0 0 130 0.00 1.864 0 0 0
132 0.00 1.800 0 0 0 134 0.00 1.804 5 93 394
136 0.00 1.852 0 0 0 138 0.00 1.783 0 0 0
140 0.00 1.729 0 0 0 142 0.00 1.776 0 0 0
146 0.01 2.093 39 5636 25835 150 1.00 2.498 100 0 100
152 0.00 1.842 0 0 0 154 1.00 2.608 100 0 100
156 0.00 1.899 0 0 0 160 0.00 1.805 0 0 0
162 0.00 1.907 0 0 0 164 0.00 1.810 0 0 0
168 0.00 1.864 0 0 0 170 1.00 2.095 100 0 100
172 0.00 1.787 0 0 0 178 0.00 1.773 0 0 0
184 0.00 1.765 0 0 0 200 0.00 1.780 0 0 0
204 0.00 1.814 0 0 0 232 0.00 1.832 0 0 0

We tested the solving abilities of the SAT solvers with the minisat solver [28]. We used a simple script to generate the formula seen above in the form of a CNF expressed in the popular DIMACS format. For a fixed value of nn and TT, we randomly generated a pair of initial and final configurations 𝒙,𝒚{0,1}n\boldsymbol{x},\boldsymbol{y}\in\{0,1\}^{n} and asked the solver whether 𝒚\boldsymbol{y} was reachable from 𝒙\boldsymbol{x} with a control sequence of length TT.

Table 2 presents the results for the 88 minimal ECA for the setting n=40n=40, T=100T=100 and a sample of 100 random pairs of initial and final configurations. The second column (ρ~\tilde{\rho}) presents the ratio of cases where a solution was found, that is, a rough estimate of the reachability ratio ρ(n){\rho}(n). We observe that ρ~\tilde{\rho} is either equal to 1 or is rather small (less than 10%). The case where ρ~\tilde{\rho} equals one corresponds exactly to the ten peripherally-linear rules 15, 30, 45, 60, 90, 105, 106, 150, 154, and 170. For the other rules, we issue the following conjecture :

The reachability ratio ρ(n){\rho}(n) tends to zero when the number of cells nn tends to infinity for all the ECA rules which are not peripherally-linear.

This conjecture is supported by the data shown on Table 3, where one can clearly see that the estimated reachability ratio ρ~(n)\tilde{\rho}(n) rapidly decreases with nn, and becomes of the order of a percent when nn is equal to 50. It is an open question to obtain more precise scaling laws with nn, and a good estimate of the time needed to reach a configuration from another.

Table 2 also presents the CPU time that is used to compute the presence or the absence of a solution, and the number of restarts, conflicts and decisions that were taken by the SAT solver. These figures are only presented to furnish a rough estimate of the difficulty of analysis that was met for each rule. It should be noted that we did not encounter any case where the solver was unable to provide an answer, either positive or negative and that the answer was given quite rapidly (\approx 20 ms per solution on average).

It is also remarkable that the among the rules for which the number of restarts is important, one finds the famous Rule 110, which is known to be Turing-universal. Of course, there is no direct correlation between the computational complexity of a rule and the difficulty to decide its controllability, but we can safely state that the rules which exhibit the richest panel of behaviours (gliders, collisions generating other gliders, etc.) are among the most difficult ones to analyse by the SAT solvers.

In order to gain insights on the relationships between regional controllability and the difficulty to find a solution or to prove that there is none, we now explore a technique of tree-building in order to find the shortest paths between two configurations.

Table 3: Estimation of the reachability ratio ρ(n){\rho}(n) for values of nn ranging from 10 to 50, for Rule 22 and Rule 110 and different values of TT. Statistics on 100 random pairs of initial and final random configurations.
nn
ECA T 10 20 30 40 50
22 100 0.83 0.38 0.15 0.03 0.01
22 200 0.84 0.46 0.11 0.04 0.01
22 400 0.89 0.43 0.17 0.03 0.01
110 100 0.72 0.38 0.17 0.03 0.01
110 200 0.75 0.36 0.15 0.07 0.00
110 400 0.75 0.31 0.13 0.06 0.01
Refer to caption
Figure 4: One step of the generation of the preimages (control tree) of configuration 0101 for Rule 22. Since Rule 22 is not peripherally-linear, some configurations do not have preimages and others have more than one preimage.

4 Generating preimages

Let us sketch the idea of generating the preimage (z0,z1,z2,,zn,zn+1)(z_{0},z_{1},z_{2},\dots,z_{n},z_{n+1}) of a given configuration (x1,x2,xn)(x_{1},x_{2},\dots x_{n}), for all possible controls z0z_{0} and zn+1z_{n+1}. This method will be exploited in the next section to examine the control tree.

The construction of the preimage is performed one site after the other. We start by inserting the two sites z0z_{0} ad z1z_{1}. Since there are n+2n+2 items in the preimage (z0,z1,z2,,zn,zn+1)(z_{0},z_{1},z_{2},\dots,z_{n},z_{n+1}), and only nn in (x1,x2,,xn)(x_{1},x_{2},\dots,x_{n}), we have to iterate over all the four values of z0z_{0} and z1z_{1}.

We can then select the value(s) of z2z_{2} such that

x1=f(z0,z1,z2).x_{1}=f(z_{0},z_{1},z_{2}).

It may happen that there are no values allowed, just one or two.

In the first case, we abort the reconstruction and pass to the next pair of z0z_{0} and z1z_{1} values, if available, or declare that there is no preimage. In the second case, we can proceed with the neighbouring site (z3z_{3}). In the third case, we have to fork the procedure for the two values of z2z_{2} and continue to the next site (z3z_{3}). This procedure continues until we generate zn+1z_{n+1}.

Instead of scanning the whole look-up table (which can be costly for large neighbourhoods), one can take profit of the ring sum expansion, Eq. (1)

xi=D0D1zi1D2ziD4zi+1D3zi1ziD5zi1zi+1D6zizi+1D7zi1zizi+1,\begin{split}x_{i}=&D_{0}\oplus D_{1}z_{i-1}\oplus D_{2}z_{i}\oplus D_{4}z_{i+1}\oplus D_{3}z_{i-1}z_{i}\oplus\\ &D_{5}z_{i-1}z_{i+1}\oplus D_{6}z_{i}z_{i+1}\oplus D_{7}z_{i-1}z_{i}z_{i+1},\end{split}

to calculate directly zi+1z_{i+1} knowing zi1z_{i-1} and ziz_{i}, using the formula:

zi+1(D4D5zi1D6ziD7zi1zi)=xiD0D1zi1D2ziD3zi1zi,\begin{split}z_{i+1}(D_{4}&\oplus D_{5}z_{i-1}\oplus D_{6}z_{i}\oplus D_{7}z_{i-1}z_{i})=\\ &x_{i}\oplus D_{0}\oplus D_{1}z_{i-1}\oplus D_{2}z_{i}\oplus D_{3}z_{i-1}z_{i},\end{split}

as illustrated in Fig. 4 for Rule 22.

If D4D5zi1D6ziD7zi1zi=1D_{4}\oplus D_{5}z_{i-1}\oplus D_{6}z_{i}\oplus D_{7}z_{i-1}z_{i}=1 then zi+1=xiD0D1zi1D2ziD3zi1ziz_{i+1}=x_{i}\oplus D_{0}\oplus D_{1}z_{i-1}\oplus D_{2}z_{i}\oplus D_{3}z_{i-1}z_{i}, otherwise, if xi=D0D1zi1D2ziD3zi1zix_{i}=D_{0}\oplus D_{1}z_{i-1}\oplus D_{2}z_{i}\oplus D_{3}z_{i-1}z_{i}, one has to fork for zi1=0,1z_{i-1}=0,1, else the recursion fails.

An example of this procedure is reported in Figure 2 for a peripherally-linear rule (Rule 150) and in Figure 4 for a nonlinear rule (Rule 22).

Let us illustrate in detail the operation: the evolution of a rule which is right-peripheral, i.e.,

x=g(x,x)x+x^{\prime}=g(x_{-},x)\oplus x_{+}

can be inverted, giving

x+=xg(x,x),x_{+}=x^{\prime}\oplus g(x_{-},x),

and in this case it is evident that the preimage is always unique.

For instance, for Rule 150

x=xxx+x+=xxx.x^{\prime}=x_{-}\oplus x\oplus x_{+}\qquad\Rightarrow\qquad x_{+}=x^{\prime}\oplus x\oplus x_{-}.

On the other side, for nonlinear rules like Rule 22

x=xxx+xxx+x+(1xx)=xxx+,x^{\prime}=x_{-}\oplus x\oplus x_{+}\oplus x_{-}xx_{+}\qquad\Rightarrow\qquad x_{+}(1\oplus x_{-}x)=x^{\prime}\oplus x_{-}\oplus x_{+},

we have three possibilities: if 1xx=11\oplus x_{-}x=1; one can obtain x+x_{+} uniquely, otherwise, either we have zero or two possibilities. In the example of Rule 22, 1xx=01\oplus x_{-}x=0 implies x=x=1x_{-}=x=1 and looking in the Look-up table (Table 1) we can only have x=0x^{\prime}=0, so, if this is the case, we have either both values x+=0x_{+}=0 and x+=1x_{+}=1 (if x=0x^{\prime}=0), or none.

Refer to caption
Figure 5: The two minimal control paths from configuration 10 to configuration 4 for Rule 30 and n=3n=3. The forward edges are in green and the backward ones in red. The connecting configurations 6 and 7 are marked in blue. To the left the full control graph (controls values are not shown, for each link there could be more than one control possible).

5 An alternative approach:Finding the shortest path in the control tree

We present here an alternative method to test the reachability of 𝒚\boldsymbol{y} from 𝒙\boldsymbol{x}: we simultaneously explore both the tree of images starting from 𝒙\boldsymbol{x}, with all the possible controls, and the tree of preimages starting from 𝒚\boldsymbol{y}, until these two trees meet in at least one configuration, thus indicating the shortest control that transforms 𝒙\boldsymbol{x} into 𝒚\boldsymbol{y}.

Finding a control 𝒍=(x00,x01,x0T1)\boldsymbol{l}=(x_{0}^{0},x_{0}^{1},\dots x_{0}^{T-1}) and 𝒓=(xn+10,xn+11,,xn+1T1)\boldsymbol{r}=(x_{n+1}^{0},x_{n+1}^{1},\dots,x_{n+1}^{T-1}) driving the system from configuration 𝒙=𝒙0=(x10,x20,xn0)\boldsymbol{x}=\boldsymbol{x}^{0}=(x^{0}_{1},x^{0}_{2}\dots,x^{0}_{n}) at time t=0t=0 to configuration 𝒚=𝒙T=(x1T,x2T,xnT)\boldsymbol{y}=\boldsymbol{x}^{T}=(x^{T}_{1},x^{T}_{2}\dots,x^{T}_{n}) at time TT can be done by looking for the shortest path in the tree generated by all possible pairs of values for the control at each time step, forward in time starting from 𝒙\boldsymbol{x}, or in the tree of all possible preimages backward in time starting from configuration 𝒚\boldsymbol{y}.

Going forward in time, each node can in principle generate four branches, for the four possible values of the left-right pair, but it may happen that more than one control pair gives the same configuration. Going backward in time, it may also happen that a given control pair gives no configuration, or, as we have seen, more branches have to be followed while generating preimages.

Refer to caption
Figure 6: Average number of nodes NN visited to determine the absence of a control for the 88 minimal rules, scaled by the the total number of configurations (2n2^{n}) vs. the reachability ratio ρ{\rho}. Peripherally-linear rules have ρ=1{\rho}=1 and N=0N=0. Data from Table LABEL:tab:fraction.

Depending on the branching of the trees, it may be more efficient to generate more forward or backward levels. However, an acceptable strategy consists in generating one level at a time for the forward and the backward tree, until the same configuration appears at the end of both, marking a shortest path from 𝒙\boldsymbol{x} to 𝒚\boldsymbol{y}. This procedure is indicated schematically in Figure 5.

An estimation of the reachability ratio ρ{\rho} for different lengths and all minimal elementary cellular automaton is reported in Table LABEL:tab:fraction. One can see that peripherally-linear ECA are always fully controllable, while the fraction of controllable pairs for the other ECA diminishes with the length of the configuration, therefore indicating that the boundary control is almost impossible for large lengths for non-peripherally-linear rules. This confirms the observations made in Section 3.2 that for the ECA that are not peripherally-linear, the controllability of randomly chosen pairs tends to zero as the size of the system grows.

This technique can help us compute an estimation of the reachability ratio ρ(n){\rho}(n). We can also estimate the average complexity of finding all controls by counting the average number of nodes explored in the control tree before declaring that there is no possible control. This complexity is zero for peripherally-linear rules (since they are always controllable), but also for CA like Rule 0, which have only one possible configuration in their image (all-zero). This also holds for all the rules that have a reachability ratio very near to zero, see Table LABEL:tab:fraction.

In Figure 6 we show that there is a nice relationship between the reachability ratio ρ{\rho} and the average complexity, scaled by the total number of configurations (2n2^{n}). This behaviour is reminiscent of the complexity phase transitions of SAT problems [29, 30, 31, 32].

6 Conclusions

In control theory, systems are classically defined by a set of differential equations and the control is applied with a feedback system. In this work, we extended this framework by tackling the question of how to control a spatially-extended discrete dynamical system, namely an elementary cellular automaton, by changing the state of only two boundary cells at each time step to influence a region of the entire domain.

We showed that the problem could adequately by translated into a Boolean CNF formula and fed to a SAT solver to effectively obtain a solution when the control is indeed possible, or a proof that it is impossible when this the case. From a concrete point of view, the control sequence is obtained rapidly, even for a system of the order of a hundred cells and a few hundred time steps. Interestingly enough, we observed that the difficulty to find a solution was somehow related to the computational complexity of the rules. As an alternative, we showed that one could also search for the shortest path between an initial and a final configuration by progressively extending trees in the directed graph of the transitions between configurations.

From a theoretical point view, we showed the equivalence of regional controllability for every nn with being chain-transitive and chain-mixing as presented in Ref. [20].

This means that our techniques may also apply to verify such properties in various other cases. More generally, a logical next step would be to explore how our work extends to cellular automata with higher dimensions, a greater number of states, or even to non-uniform rules (applying different local rules in a SAT solver and in the search of the shortest path can be done directly without any additional effort).

We also observed that in the ECA that are not peripherally-linear, the probability to reach a final configuration from an initial configuration vanishes when the system’s size grows and when the configurations are chosen randomly. It would be interesting to prove this property formally. In fact, a more precise description would quantify how the size of the communication classes of the transition graph scale as a function of the system’s size. The techniques that were applied for fully asynchronous ECA [33] may also apply here and the use of SAT solvers may be of great help to derive such formal proofs of reachability.

Table 4: Numerical estimation ρ~\tilde{\rho} of the reachability ratio and number NN of nodes to be visited to check for the absence of control for all minimal ECA RR and various configuration sizes nn. The data are obtained sampling over 10410^{4} random pairs of configurations.
RR n=9n=9 n=11n=11 n=13n=13 n=15n=15
ρ~\tilde{\rho} N ρ~\tilde{\rho} N ρ~\tilde{\rho} N ρ~\tilde{\rho} N
0 0.002 0.9956 0.0011 0.9985 0 0.9997 0 0.9999
1 0.0112 9.8817 0.0021 10.2039 0.0005 10.2994 0.0003 9.1978
2 0.0801 37.6393 0.0426 84.2125 0.0208 185.0444 0.0137 400.4257
3 0.3936 121.053 0.2907 436.7135 0.2303 1459.9432 0.1783 4800.218
4 0.0026 2.5131 0.0007 2.7917 0.0002 2.7316 0.0001 3.3823
5 0.0132 12.9694 0.0031 14.1058 0.0006 13.7366 0 14.5212
6 0.3918 124.9131 0.2638 401.5478 0.1666 1172.6215 0.107 3171.2119
7 0.2093 90.0002 0.1223 226.8278 0.0655 552.3652 0.0417 1302.8289
8 0.0035 3.6936 0.0013 4.0129 0.0003 4.1582 0 4.4093
9 0.3893 122.727 0.2378 376.5136 0.141 978.621 0.0774 2346.7942
10 0.195 83.5645 0.1316 236.9626 0.0902 649.5169 0.0507 1775.1363
11 0.327 112.401 0.2183 349.5516 0.1468 1011.7967 0.0917 2837.0574
12 0.003 2.6059 0.0007 2.7152 0.0004 2.8586 0 3.0346
13 0.0156 14.6131 0.0038 17.273 0.0014 19.732 0.0004 20.4391
14 0.4133 123.18 0.2794 423.1061 0.1938 1293.3974 0.1304 3763.2499
15 1 0 1 0 1 0 1 0
18 0.4262 125.7239 0.3267 443.4179 0.234 1497.6062 0.1713 4780.7041
19 0.0083 7.4527 0.0018 7.93 0.0004 7.9429 0.0002 8.3817
22 0.8586 62.3896 0.7928 333.8604 0.7057 1692.3004 0.6131 7699.2784
23 0.0126 12.1348 0.0031 13.4406 0.0008 14.4739 0.0005 14.9865
24 0.0911 43.8604 0.0515 97.4022 0.0257 213.4277 0.0142 461.6439
25 0.7168 103.373 0.5606 502.6101 0.4128 1971.0832 0.2882 6698.8188
26 0.7686 90.9939 0.6906 438.8354 0.6189 1900.8206 0.5183 8258.4423
27 0.5841 126.0192 0.4693 501.9368 0.3645 1862.9912 0.2838 6487.5229
28 0.0235 20.7356 0.0058 26.5314 0.0024 29.8961 0.0003 34.9573
29 0.0156 13.5261 0.0039 14.127 0.0013 14.6097 0.0002 14.692
30 1 0 1 0 1 0 1 0
32 0.0062 5.1255 0.0029 5.1161 0.0003 5.5633 0 5.319
33 0.0447 40.8937 0.0128 47.7303 0.0028 50.5435 0.001 50.9947
34 0.1698 73.7404 0.1125 206.6842 0.0776 562.5955 0.0541 1510.5548
35 0.5773 124.4786 0.4492 509.7556 0.3446 1888.0347 0.2731 6487.2539
36 0.0064 6.3585 0.0022 7.1627 0.0005 7.0318 0.0003 7.1574
37 0.9829 8.5978 0.8781 220.7768 0.7562 1512.4127 0.5589 8097.3136
38 0.3943 121.9057 0.2901 410.0903 0.2061 1316.2389 0.1464 4065.0292
40 0.0201 17.7929 0.0063 20.9039 0.0011 21.9891 0.0003 28.0374
41 0.8068 79.0688 0.6983 437.369 0.6055 1966.9365 0.5038 8234.82
42 0.5302 128.4775 0.4543 505.6233 0.3875 1920.5669 0.323 7182.0698
43 0.6385 118.688 0.506 503.6215 0.37 1943.5089 0.2789 6608.7907
44 0.0145 13.4614 0.0037 15.1588 0.0014 15.5487 0.0002 16.7039
45 1 0 1 0 1 0 1 0
46 0.1373 60.7865 0.0762 138.4218 0.0383 307.8265 0.0203 671.4806
50 0.0126 12.9995 0.0046 14.382 0.0009 15.2791 0.0003 16.3391
51 0.004 1.992 0.0009 1.9982 0.0003 1.9994 0 2
54 0.5991 123.0332 0.4806 512.1205 0.3817 1928.2614 0.2934 6796.8576
56 0.3266 110.1283 0.2159 345.4543 0.1452 1004.3413 0.1011 2798.7363
57 0.1427 66.3875 0.0595 138.6571 0.0368 269.1887 0.0166 522.0983
58 0.2261 91.8859 0.1241 227.1284 0.0685 522.7471 0.0352 1159.563
60 1 0 1 0 1 0 1 0
62 0.6806 115.2031 0.3775 511.3079 0.2104 1427.2255 0.1077 3330.3698
72 0.0059 6.6752 0.0014 7.1114 0.0004 7.5419 0.0001 8.7902
73 0.4241 153.995 0.2697 539.3759 0.1559 1611.5 0.0902 4281.3997
74 0.6732 112.9072 0.5043 513.9198 0.3615 1857.4982 0.2324 6055.0327
76 0.0027 1.9973 0.0008 2.1884 0.0003 2.3443 0 2.4521
77 0.0076 7.4463 0.0024 8.0492 0.0006 8.4614 0.0002 9.1361
78 0.0212 17.3528 0.0051 20.4336 0.002 23.2771 0.0005 24.8262
90 1 0 1 0 1 0 1 0
94 0.2209 119.264 0.1172 304.6733 0.0527 682.9739 0.0266 1295.4297
104 0.0594 47.047 0.016 62.2863 0.0051 73.0639 0.0012 74.4246
105 1 0 1 0 1 0 1 0
106 1 0 1 0 1 0 1 0
108 0.0133 11.6458 0.0033 12.9831 0.0014 14.3384 0.0001 15.3613
110 0.7892 85.207 0.7088 422.3375 0.6229 1918.6175 0.5404 8131.7471
122 0.7727 90.6083 0.6454 467.5212 0.515 2016.0089 0.3819 7780.4991
126 0.3602 115.4745 0.2572 389.2605 0.1823 1232.7578 0.1309 3752.0127
128 0.0042 4.767 0.0014 5.1223 0.0004 5.3654 0.0001 5.8271
130 0.0865 41.9356 0.0468 89.3382 0.0231 190.9837 0.0128 407.73
132 0.0086 7.517 0.0021 8.0448 0.0008 8.9295 0.0002 9.4588
134 0.4792 127.2468 0.3204 448.0948 0.2053 1372.3773 0.1371 3794.8008
136 0.0081 6.9283 0.0023 7.6935 0.001 9.3916 0.0001 8.8709
138 0.3932 121.1175 0.3025 429.4487 0.232 1456.7279 0.1803 4788.5434
140 0.0065 5.4538 0.0014 6.5541 0.0004 7.4921 0.0002 8.3589
142 0.6478 115.6484 0.4983 511.3795 0.376 1925.1316 0.2793 6605.3952
146 0.4318 126.3201 0.3181 451.5111 0.2391 1490.3106 0.1689 4798.1028
150 1 0 1 0 1 0 1 0
152 0.1055 57.6982 0.0559 126.3235 0.0308 271.1771 0.0155 580.6825
154 1 0 1 0 1 0 1 0
156 0.02 18.6955 0.0066 23.3457 0.0024 27.5274 0.0004 31.1357
160 0.019 14.9671 0.005 16.1459 0.0014 17.7011 0.0003 20.3167
162 0.1731 76.9699 0.1151 210.7525 0.0764 568.7733 0.0534 1517.3141
164 0.1176 64.8285 0.0527 130.9186 0.0232 228.8644 0.0075 371.3313
168 0.0949 67.665 0.0405 135.6538 0.0168 258.6673 0.0085 440.1266
170 1 0 1 0 1 0 1 0
172 0.0485 37.6656 0.0175 63.234 0.0068 99.9733 0.0026 151.9821
178 0.0135 11.9561 0.0034 13.2167 0.0016 14.1006 0.0001 13.7332
184 0.5655 126.3228 0.4373 500.8263 0.334 1775.3973 0.2422 5955.3257
200 0.0031 2.3074 0.0003 2.4128 0.0001 2.5936 0.0001 2.9415
204 0.002 0 0.0005 0 0.0004 0 0 0
232 0.0079 7.309 0.0024 8.2605 0.0007 8.5996 0.0001 8.5114

References

  • \bibcommenthead
  • Lions [1991] Lions, J.-L.: Exact controllability for distributed systems. some trends and some problems. In: Applied and Industrial Mathematics, pp. 59–84 (1991). https://doi.org/10.1007/978-94-009-1908-2_7
  • Curtain and Zwart [2012] Curtain, R.F., Zwart, H.: An Introduction to Infinite-dimensional Linear Systems Theory vol. 21. Springer Science & Business Media, New York, USA (2012). https://doi.org/10.1007/978-1-4612-4224-6
  • Sontag [2013] Sontag, E.D.: Mathematical Control Theory: Deterministic Finite Dimensional Systems vol. 6. Springer Science & Business Media, New York, USA (2013). https://doi.org/10.1007/978-1-4612-0577-7
  • Lions and Lelong [1968] Lions, J.-L., Lelong, P.: Contrôle optimal de systèmes gouvernés par des équations aux dérivées partielles. Collection "Études mathématiques". Dunod, Paris, France (1968)
  • Lions [1986] Lions, J.-L.: Controlabilité exacte des systèmes distribués: remarques sur la théorie générale et les applications. In: Analysis and Optimization of Systems: Proceedings of the Seventh International Conference on Analysis and Optimization of Systems, Antibes, June 25-27, 1986, pp. 3–14 (1986). Springer
  • Ermentrout and Edelstein-Keshet [1993] Ermentrout, G.B., Edelstein-Keshet, L.: Cellular automata approaches to biological modeling. Journal of Theoretical Biology 160(1), 97–133 (1993)
  • Kier et al. [2005] Kier, L.B., Seybold, P.G., Cheng, C.-K.: Cellular Automata Modeling of Chemical Systems. Springer, Dordrecht, Netherlands (2005). https://doi.org/10.1007/1-4020-3690-6
  • Xiao et al. [2005] Xiao, X., Shao, S., Ding, Y., Huang, Z., Chen, X., Chou, K.-C.: Using cellular automata to generate image representation for biological sequences. Amino acids 28, 29–35 (2005) https://doi.org/10.1007/s00726-004-0154-9
  • [9] ACRI: International Conference on Cellular Automata for Research and Industry. Lecture Notes in Computer Science LNCS 2493, 3305, 4173, 5191, 6350, 7495, 8751, 9863, 11115, 12599, 13402, 14978, https://link.springer.com/conference/acri. [Online; accessed 3-March-2025] (2002-2024)
  • Dridi [2019] Dridi, S.: Recent advances in regional controllability of cellular automata. PhD thesis, Perpignan (2019)
  • Dridi et al. [2019a] Dridi, S., El Yacoubi, S., Bagnoli, F., Fontaine, A.: A graph theory approach for regional controllability of boolean cellular automata. International Journal of Parallel, Emergent and Distributed Systems, 1–15 (2019) https://doi.org/10.1080/17445760.2019.1608442
  • Dridi et al. [2019b] Dridi, S., Bagnoli, F., El Yacoubi, S.: Markov chains approach for regional controllability of deterministic cellular automata, via boundary actions. Journal of Cellular Automata 14 (2019)
  • Dridi et al. [2020] Dridi, S., El Yacoubi, S., Bagnoli, F.: Boundary regional controllability of linear Boolean cellular automata using Markov chain, pp. 37–48. Springer, Cham, Switzerland (2020). https://doi.org/10.1007/978-3-030-26149-8_4
  • Bagnoli [2025] Bagnoli, F.: In: Adamatzky, A., Sirakoulis, G.C., Martinez, G.J. (eds.) Synchronization and Control of Cellular Automata. Springer, Cham, Switzerland (2025)
  • Dridi et al. [2022] Dridi, S., El Yacoubi, S., Bagnoli, F.: Kalman condition and new algorithm approach for regional controllability of peripherally-linear elementary cellular automata via boundary actions. J. Cell. Autom. 16(3-4), 173–195 (2022)
  • El Yacoubi et al. [2021] El Yacoubi, S., Plénet, T., Dridi, S., Bagnoli, F., Lefèvre, L., Raïevsky, C.: Some control and observation issues in cellular automata. Complex Systems 30(3), 391–413 (2021) https://doi.org/10.25088/ComplexSystems.30.3.391
  • Bagnoli et al. [2018a] Bagnoli, F., El Yacoubi, S., Rechtman, R.: Toward a boundary regional control problem for boolean cellular automata. Natural Computing 17(3), 479–486 (2018) https://doi.org/10.1007/s11047-017-9626-1
  • Bagnoli et al. [2018b] Bagnoli, F., Dridi, S., El Yacoubi, S., Rechtman, R.: Regional control of probabilistic cellular automata. In: International Conference on Cellular Automata, pp. 243–254 (2018). https://doi.org/10.1007/978-3-319-99813-8_22 . Springer
  • Bagnoli et al. [2019] Bagnoli, F., Dridi, S., El Yacoubi, S., Rechtman, R.: Optimal and suboptimal regional control of probabilistic cellular automata. Natural Computing 18(4), 845–853 (2019) https://doi.org/%****␣VF-NC-FB.bbl␣Line␣325␣****10.1007/s11047-019-09763-5
  • Dridi [2025] Dridi, S.: New characterization of regional controllability and controllability of deterministic cellular automata via topological and symbolic dynamics notions. arXiv preprint arXiv:2501.02622 (2025)
  • Dridi et al. [2024] Dridi, S., Bagnoli, F., El Yacoubi, S.: Regional Controllability of Cellular Automata Through Preimages, pp. 22–33. Springer, Cham, Switzerland (2024). https://doi.org/10.1007/978-3-031-71552-5_3
  • Wolfram [1983] Wolfram, S.: Statistical mechanics of cellular automata. Reviews of modern physics 55(3), 601 (1983) https://doi.org/10.1103/revmodphys.55.601
  • Wolfram [2002] Wolfram, S.: A New Kind of Science. Wolfram Media, Champaign, IL (2002). https://www.wolframscience.com
  • Vichniac [1990] Vichniac, G.Y.: Boolean derivatives on cellular automata. Physica D: Nonlinear Phenomena 45(1–3), 63–74 (1990) https://doi.org/10.1016/0167-2789(90)90174-n
  • Bagnoli [1992] Bagnoli, F.: Boolean derivatives and computation of cellular automata. International Journal of Modern Physics C 3(02), 307–320 (1992) https://doi.org/10.1142/s0129183192000257
  • Wegener [1987] Wegener, I.: The Complexity of Boolean Functions. Applicable Theory in Computer Science. Vieweg & Teubner, Wiesbaden, Germany (1987)
  • Kurka [2003] Kurka, P.: Topological and Symbolic Dynamics. Société mathématique de France, Paris, France (2003)
  • Eén and Sörensson [2003] Eén, N., Sörensson, N.: The MiniSat solver page. http://minisat.se/. [Online; accessed 16-March-2025] (2003)
  • Hayes [1997] Hayes, B.: Computing science: Can’t get no satisfaction. American scientist 85(2), 108–112 (1997)
  • Cheeseman et al. [1991] Cheeseman, P., Kanefsky, B., Taylor, W.M.: Where the really hard problems are. In: Proceedings of the 12th International Joint Conference on Artificial Intelligence - Volume 1. IJCAI’91, pp. 331–337. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA (1991). https://doi.org/10.5555/1631171.1631221
  • Kirkpatrick and Selman [1994] Kirkpatrick, S., Selman, B.: Critical behavior in the satisfiability of random boolean expressions. Science 264(5163), 1297–1301 (1994) https://doi.org/10.1126/science.264.5163.1297
  • Hogg et al. [1996] Hogg, T., Huberman, B.A., Williams, C.: Frontiers in Problem Solving: Phase Transitions and Complexity. Special issue of Artificial Intelligence, Vol. 81 (1996)
  • Roy et al. [2024] Roy, S., Fatès, N., Das, S.: Reversibility of Elementary Cellular Automata with fully asynchronous updating: An analysis of the rules with partial recurrence. Theoretical Computer Science 1011, 114721 (2024) https://doi.org/10.1016/J.TCS.2024.114721