Biologically inspired replication of spatial patterns

with reversible and additive cellular automata

Vladimir Garcı́a-Morales
Departament de Fı́sica de la Terra i Termodinàmica,
Universitat de València, E-46100 Burjassot, Spain∗

In this article, the replication of arbitrary patterns by reversible and additive cellular automata
is reported. The orbit of an 1D deterministic cellular automaton operating on p symbols that is
both additive and reversible is explicitly given in terms of coefficients that appear in the theory
of Gegenbauer polynomials. It is shown that if p is and odd prime, the pattern formed after
(p − 1)/2 time steps from any arbitrary initial condition (spatially confined to a region of side less
than p) replicates after p+(p−1)/2 time steps in a way that resembles budding in biological systems.

I. INTRODUCTION x0−d+1 , x0−d+2 , . . . , x00 , . . . , x0d−1 . . . surrounded by ze-

ros the solution is a convolution of the solution for
Cellular automata (CAs) are discrete dynamical a single seed with this more general initial condition
systems in which a discrete lattice of p symbols, that and one has
can be labelled with the set Ap = {0, 1, . . . p − 1} d−1  
X t
called the alphabet, is iteratively updated accord- xit = xi−k mod p (3)
ing to a specified local rule [1–3]. Locally, each site i−k 0
k=−d+1
i ∈ Z in the lattice is characterized by the variable
xit ∈ Ap at time t. An additive cellular automaton ACAs have found many technological applications
(ACA) of radius (l + r)/2 is a CA whose update rule in, for example, image processing [7] and the synthe-
is a linear function modulo p of the dynamical states sis of cryptographic interleaved sequences [8]. ACAs
xi+l
t , xt
i+l−1
, . . . xit , . . . , xi−r+1
t , xi−r
t . The spatiotem-
were the subject of intense research in the late 60’s
poral evolutions of ACAs are well known to give rise [9] and during the 70’s because of the discovery by
to nested patterns and fractals (when infinite lattices Amoroso and Cooper [10] of their ability to replicate
are considered). any pattern. A series of works were published where
Spatiotemporal evolutions of ACAs behave simi- this work was extended to several dimensions [11], ar-
larly to the solutions of linear partial differential equa- bitrary neighborhood indices [12] and reproduction in
tions [1]. For example, ACAs admit analogs of Green’s quiescent environments [13]. Barto [14] provided a
functions: given an initial condition, the resulting evo- nice summary of these works and showed the connec-
lution can be found by means of a convolution of the tion between the replication problem and local trans-
discrete evolution of a single non-zero site (analogous formations that are linear over fields of nonzero char-
to an integral kernel) with the initial condition [1]. acteristic. This series of works culminated with the
For example, the ACA of radius 1/2 given by article by Ito, Osato and Nasu [15] who generalized
these results beyond finite fields to finite rings, estab-
xit+1 = xit + xi+1
t mod p (1) lishing rigorous results on the surjectivity of additive
cellular automata rules in those cases. The crucial
has solution work of Martin, Odlyzko and Wolfram [16] established
 some algebraic properties of ACAs, showing also the
t connection of the spatiotemporal evolution of the lat-
xit = mod p (2)
i ter with generating functions. ACAs are straightfor-
wardly related to arithmetic triangles and simplices
for an initial condition consisting of a single seed (e.g. Pascal pyramids) modulo an integer number p
of value one at i = 0 and zero elsewhere, i.e. [17, 18].
xi0 = δ0i . This corresponds to the Pascal trian- The replication property in ACAs can be simply
gle modulo p [4–6]. For an arbitrary initial condi- understood from Eq. (2). The ACA starts with a
tion, consisting of an array of 2d − 1 nonzero sites single seed at i = 0. At time p, two seeds are formed
at i = 0 and i = p, since
   
p p
∗ Electronic = =1 (4)
address: [email protected] 0 p
 2

and, if p is prime, orbit. This part draws heavily on previous work by

  Dilcher [21] from which it constitutes an application.
p It is needed to obtain our main, original result, which
= 0 mod p if 1 ≤ i ≤ p − 1 (5)
i is presented in Section III: We establish the replica-
tion property of the RACA under general initial con-
This latter equation holds because ditions. Finally, in Section IV we discuss our result,
  presenting some simulations to illustrate it visually, as
p p! well as some generalizations.
= (6)
i i!(p − i)!

is divisible by p because the numerator contains a fac-

II. THE CA MAP AND ITS EXACT
tor p that cannot be divided by the factors in the de-
SPATIOTEMPORAL EVOLUTION
nominator (because of p being prime). Two seeds sep-
arated by sites in the quiescent state are thus formed
at t = p and these initiate spatiotemporal evolutions The spatiotemporal dynamics of the RACA here
that are similar to the ones found for t ≤ p − 1. If considered is given by the map
one thinks in the patterns arising from the spatiotem-
poral evolution of the ACA as modeling the shapes of xit+1 = xi+1
t + xit + xi−1
t − xit−1 mod p (7)
organisms, this process does not resemble any replica-
tion process found in biological systems. where i ∈ Z labels the site on a 1D discrete lattice
Although replication is a transparent property of and t ∈ [−1, ∞), t ∈ Z is the discrete time. The
ACAs, this property is highly nontrivial for non- dynamic variable xit is restricted to an integer value
additive CA [19]. It is also highly nontrivial for re- in the interval [0, p − 1], where p shall be considered
versible ACAs (RACAs), since initial conditions have an odd prime number. The initial condition at times
to be specified not only at an initial time t = 0 but also t = −1 and t = 0 is assumed to be the same, i.e.
on a previous time t = −1 and the resulting combi- xi−1 = xi0 a crucial condition that is needed for the
natorial solution is much more complex. A reversible result. We first study the solution of the map for the
CA is a CA in which every configuration has a unique initial conditions xi−1 = xi0 = δi0 consisting of a single
predecessor. Reversible CAs pose formidable mathe- site with value 1 surrounded by zeros. The result for
matical problems and, it is known, for example, that this particular case easily generalizes to an arbitrary
it is undecidable to determine whether a given CA initial condition by means of the convolution property.
in two or more dimensions is reversible [20]. RACAs The reversibility of the map is clear from the fact that
share the nice convolution property of ACAs but their it remains invariant after exchanging t + 1 ↔ t − 1.
spatiotemporal evolution is much more complex and Let us first leave aside the mod p operation in Eq.
their ability to replicate patterns has never been re- (7). Then, the map
ported before, to the best of our knowledge.
In this work we report on a robust replication prop- xit+1 = xi+1
t + xit + xi−1
t − xit−1 (8)
erty for certain RACAs under very general initial con-
ditions. We first consider 1D RACAs acting on a local with the initial conditions xi−1 = xi0 = δi0 can be
neighborhood of unit radius (three sites) and deter- expressed in terms of a generating function as
ministically operating over a prime number p of sym- ∞ X t
bols. Our result shows that replication in RACAs is, 1 − zy X
= 1 + xit z t−i y t (9)
indeed, possible and much more subtle than in ACAs. 1 − y(1 + z + z 2 ) + y 2 z 2 t=1 i=−t
Furthermore, the replication process resembles bud-
ding in biological systems: a new pattern/organism This is obtained by formally expanding the l.h.s. of
develops from a bud at one particular site. When ma- Eq. (9) in powers of z and y and equating powers
ture, the buds develop into tiny individuals detaching of the latter quantities with same exponents on both
from the parent body and becoming new independent sides.
individuals. Our main result, Theorem 2, establishes For the generating function
the mathematical details of this replication process.
Our result can be easily generalized to a larger num- 1 X∞ X t
1,1 t−i t
ber of dimensions. = Ct,i z y (10)
1 − y(1 + z + z 2 ) + y 2 z 2
The outline of this article is as follows. In Section t=0 i=−t
II we present the RACA under study and establish
1,1
a rigorous mathematical result that yields its exact Dilcher [21] found that the coefficients Ct,i are given
 3

by the expression III. MAIN RESULT: REPLICATION OF

SPATIALLY EXTENDED STRUCTURES
t−|i| t−|i|−2s
b 2 c b 2 c    
1,1
X X t − s 2j + |i| t − 2s
Ct,i = (−1)s It is interesting to compare the spatiotemporal evo-
s j 2j + |i|
s=0 j=0 lution of the map Eq. (8) before and after the mod p
(11) operation. This is shown in Fig. 1 where both evo-
(see Theorem 1 in [21], with ν = λ = 1 as defined lutions are shown. Without performing the mod p
there). These coefficients, related to the theory of operation, the values obtained by the recurrence in
Gegenbauer (ultraspherical) polynomials [21], obey a Eq. (8) with xi0 = xi−1 = δi,0 arrange in the arith-
recurrence relation similar to Eq. (8) metic triangle shown in Fig. 1 a). There, each ele-
ment in the tth row is the sum of the three closest
1,1 1,1 1,1 1,1 1,1
Ct+1,i = Ct,i+1 + Ct,i + Ct,i−1 − Ct−1,i (12) elements in the (t − 1)th row, minus the closest ele-
ment in the (t − 2)th row. Remarkably, if one now
We then have, by using Eq. (10) picks any p prime, and performs the mod p opera-
tion, the structure originated after (p−1)/2 time steps
1 − zy is duplicated at p + (p − 1)/2 time steps as can be seen
1 − y(1 + z + z 2 ) + y 2 z 2 in Fig. 1 b) for p = 5. The two copies of the structure
∞ X
X t are neatly separated by sites in the quiescent state.
1,1 1,1 t−i t
= [Ct,i − zyCt,i ]z y For other initial conditions which are nonzero only for
t=0 i=−t sites within a region of size lower than (p − 1)/2, an
∞ X
X t analogous behavior is observed. Considering two di-
1,1 1,1
= 1+ [Ct,i − Ct−1,i ]z t−i y t mensions and other neighborhoods the same fact is
t=1 i=−t computationally observed.
∞ X
X t We now mathematically substantiate this fact. We
= 1+ xjt z t−i y t (13) first study the orbit of the RACA and prove two
t=1 i=−t lemmas before our main result, the theorem on the
replication of complex structures. Lemma 1 estab-
where Eq. (9) has also been used. We thus find that, lishes some overall symmetries of the arithmetic tri-
for t ≥ 1 angle arising from the spatiotemporal evolution of the
RACA. Lemma 2 is necessary to establish that the
xjt = Ct,i
1,1 1,1
− Ct−1,i mod p (14) sites separating the two copies of the same structure
are in the quiescent state.
for initial conditions xi−1 = xi0 = δi0 . We gather now
all above observations. Lemma 1: The following relationships, for xit given
by Eq. (16), hold
Theorem 1: The solution for the orbit xit of (i) xit = x−i
t
the one-dimensional reversible cellular automaton i+p
(ii) xt+p = xit = x−i−p
t+p for 0 ≤ t ≤ p − 1 and |i| ≤ t
rule (iii) xit = 0 for |i| > t
xit+1 = xi+1
t + xit + xi−1
t − xit−1 mod p (15)
Proof: Result (i) is trivial to prove from the symme-
1,1 1,1
where xi0= xi−1
= δi,0 , δi,j is the Kronecker delta and try relationships of the coefficients Ct,i = Ct,−i in
p is any odd prime number, is given for t ≥ 1 by: Eqs. Eq.(16) and (17) since all the i-dependence of
the latter is through the absolute value |i|.
1,1 1,1
xit = Ct,i − Ct−1,i mod p (16) Result (ii) is a remarkable property of Eqs.(16) and
(17). To prove it we first recall Lucas’ correspondence
where theorem. This theorem, proved in [22], establishes
that if two integer numbers n and m have the base p
t−|i| t−|i|
b 2 c b 2 −sc     representation n = n0 + n1 p + ... + nk pk , m = m0 +
1,1
X X t − s 2j + |i| t − 2s
Ct,i = (−1)s m1 p + ... + mk pk with k integer, then
s=0 j=0
s j 2j + |i|
  k  
(17) n Y nh
mod p = mod p (18)
with b...c denoting the floor function. m mh
h=0

We now observe that, since 0 ≤ s ≤ smax < t ≤

 4

FIG. 1: Arithmetic triangle formed by the values obtained by iterating (a) the map given by Eq. (8), (b) the CA rule Eq.(15) with
xi0 = xi−1 = δi,0 for p = 5. The structure formed at t = (p − 1)/2 = 2 (shown in a green box) duplicates at time t = p + (p − 1)/2 = 7, both
structures being separated by sites in the quiescent state. This duplication process grows out of “buds” that are created at time t = p = 7.

p−1, 0 ≤ j ≤ smax < t ≤ p−1, 0 ≤ 2j +|i| ≤ t ≤ p−1 where Eqs. (19), (20) and (21) have been used to
and 0 ≤ t − s ≤ t ≤ p − 1, we have, from Lucas’ replace the binomial coefficients within the sum. We
theorem thus have as well
xi+p 1,1 1,1
    
t−s+p t−s 1 t+p = Ct+p,i+p − Ct−1+p,i+p mod p
mod p = mod p
s s 0 1,1 1,1
  = Ct,i − Ct−1,i mod p
t−s
= mod p (19) = xit (23)
s
as we wanted to prove. The second equality in (ii)

t − 2s + p
 
t − 2s
 
1 comes then after applying result (i).
mod p = mod p Result (iii) is directly obtained from Eq. (17),
2j + |i| + p 2j + |i| 1 1,1
  since Ct,i = 0 for |i| > t. 2
t − 2s
= mod p (20)
2j + |i|
Lemma 2: Let |i| ≤ (p − 1)/2 and p > 1 be any odd
and
     prime number. Then
2j + |i| + p 2j + |i| 1 1,1 1,1
mod p = mod p Cp+ p−1 mod p = Cp+ p−1 mod p (24)
j j 0 2 ,i −1,i 2
 
2j + |i| and, therefore, for the map given by Eq. (15), we
= mod p (21)
j have that xip+(p−1)/2 = 0, ∀|i| ≤ (p − 1)/2.
We thus have, from Eq. (17) Proof: From Eq. (17), we have, for t = T :=
1,1 p + (p − 1)/2,
Ct+p,i+p mod p =
−|i|
t−|i|
b 2 c b
t−|i|−2s
2 c    b TX
2 c b T −|i|
2 −sc
s t−s+p 2j + |i| + p
X X 1,1
X
= (−1) × CT,i = cT,i (s, j) (25)
s=0 j=0
s j s=0 j=0
 
t − 2s + p where
× mod p
2j + |i| + p    
s T − s 2j + |i| T − 2s
t−|i|
b 2 cb
t−|i|−2s
c cT,i (s, j) := (−1) (26)
2    s j 2j + |i|
s t−s 2j + |i|
X X
= (−1) × (T − s)!
s=0 j=0
s j = (−1)s (27)
  s!j!(j + |i|)! (T − 2s − 2j − |i|)!
t − 2s
× mod p Clearly, from Eq. (27) all cT,i (s, j)’s for which
2j + |i|
1,1
= Ct,i mod p (22) T − p < 2s ≤ p − 1 (28)
 5

do not contribute to the sum in Eq. (25) because from which, by comparing with Eq. (30),
they are equal to 0 mod p. This can be seen from
the fact that the factorial (T − s)! in the numera- cT,i (s0 , j) mod p = cT −1,i (s00 , j) mod p (34)
tor of cT,i (s, j) contains a factor p that is absent in
the denominator, since s < p, j < p jand because k and, therefore
|i| ≤ (p − 1)/2 we also have j + |i| ≤ T +|i| 2 < p. XX
1,1
Thus, the only terms that contribute to the sum in CT,i mod p = cT,i (s0 , j)
j fork which 0 ≤ 2s ≤ T − p and
Eq. (25) are those s0 j
T +|i|
XX
p + 1 ≤ 2s ≤ 2 . If one considers now time
2 = cT −1,i (s00 , j)
T − 1, these inequalities for the values s00 j
j of s readk now
0 ≤ 2s ≤ T − p − 1 and p − 1 ≤ 2s ≤ 2 T −1+|i|
2 . This = CT1,1
−1,i mod p (35)
0 0
means that for any s = s such that cT,i (s , j) is not
1,1
divisible by p a bijection can be established to another as we wanted to prove. Now, since xiT = CT,i −CT1,1
−1,i
value s00 = s0 + p−12 mod p for which cT −1,i (s00 , j) is mod p, we finally have
divisible by p neither. We now show that cT,i (s0 , j)
mod p = cT −1,i (s00 , j) mod p. Note that, by using p−1
xip+ p−1 = 0 mod p ∀|i| ≤ (36)
the symmetries of binomial coefficients and Lucas’ cor- 2 2
respondence theorem we have, on one hand,
and the proof of the Lemma is completed. 2
cT,i (s0 , j) mod p = We now state and give a proof of the main result
T − s0 2j + |i| T − 2s0
   
s0 of this article.
= (−1) mod p
T − 2s0 j 2j + |i|
p + T − s0 2j + |i| T − 2s0
   
0
= (−1)s mod p Theorem 2 (Replication of spatially extended
T − 2s0 j 2j + |i|
structures): Let an arbitrary initial condition be
(29) given for an integer variable xit ∈ [0, p − 1], with p and
and, on the other hand, using again Lucas’ correspon- odd prime at t = 0 and t = −1, which is nonzero only
dence theorem twice on a collection of 2δ + 1 < p adjacent sites centered at
i = 0 and such that x−δ −δ
0 = x−1 , x0
−δ+1
= x−δ+1
−1 , . . .,
cT −1,i (s00 , j) mod p = 0 0 δ−1 δ−1 δ δ
x0 = x−1 , . . ., x0 = x−1 , x0 = x−1 . Then, at time
2p − 1 − s0 2j + |i| T − 2s0 t00 = p + (p − 1)/2, two copies arise that are identical
   
0
+ p−1
= (−1)s 2
to the pattern obtained at time t0 = (p − 1)/2. The
T − 2s0 j 2j + |i|
copies are separated by sites in the quiescent state
mod p (30)
(i.e. sites with value 0).
We now note that
Proof: Because of the linearity of the map, Eq.
p + T − s0 (p + T − s0 ) (p + T − s0 − 1)
 
= (31) (15), the orbit is given by the superposition (convo-
T − 2s0 p + s0 (p + s0 − 1) lution) of arithmetic triangles obtained in Theorem 1
(2p − s0 ) 2p − 1 − s0
 
with the initial condition as
... ×
p + s0 − p−1 T − 2s0


2 δ  
X 1,1 1,1
and, therefore, xit = x0i−k Ct,i−k − Ct−1,i−k mod p (37)
k=−δ
p + T − s0 2p − s0 − 1
   
p−1
mod p = (−1) 2
T − 2s0 T − 2s0 1,1
with Ct,i given by Eq. (17). At time t = p−1
a
2
mod p (32) pattern is obtained
By replacing this result in Eq. (29), we obtain δ  
X
0
cT,i (s , j)
mod p = xip−1 = xi−k
0 C 1,1
p−1
,i−k
− C 1,1
p−1
−1,i−k
mod p
2 2 2
k=−δ
2p − s0 − 1 2j + |i| T − 2s0
   
0 p−1
(−1)s + 2 (38)
T − 2s0 j 2j + |i| Because of Lemma 1, (i) and (ii) two copies of this
mod p (33) pattern are produced at time t = p + p−1
2 . Let |i| ≤ t,
 6

FIG. 2: Spatiotemporal evolution of map Eq. (41) for the initial condition shown in the leftmost panel and p = 7. Subsequent time steps
are shown and it is observed that the pattern obtained at t = (p − 1)/2 = 3 is replicated at time step t = p + (p − 1)/2 = 10, yielding two
copies separated by sites in the quiescent state (dark blue).

i.e. −t ≤ i ≤ t. The two copies are given by solution of the above map is simply given for t ≥ 1 as

δ   δ
X  
xi,j xi,j−k 1,1 1,1
X
xi+p
p+ p−1
= xi−k
0 C 1,1
p−1
,i−k
− C 1,1
p−1
−1,i−k t = 0 Ct,j−k − Ct−1,j−k (42)
2 2 2
k=−δ k=−δ
mod p (39)
Theorem 2 can be applied in this case and, for a
x−i−p
p+ p−1
= i+p
xp+ p−1 (40) given odd prime p the structure obtained at time step
2 2
(p−1)/2 will yield two copies at time p+(p−1)/2 sep-
and are centered at positions p and −p, respectively, arated by sites in the quiescent state. This is shown in
occupying positions [p − p−1 p−1
2 − δ, p + 2 + δ] and
Figure 2 for p = 7 and an initial condition contained
p−1 p−1
[−p − 2 − δ, −p + 2 + δ] . These structures are in a rectangle satisfying the conditions of Theorem 2.
externally surrounded by sites in the quiescent state Computer simulations show that this replication
(because of Lemma 1 (iii)). The sites occupying posi- process extend to von Neumann and Moore neigh-
tions [− p−1 p−1
2 +δ, 2 −δ] are also in the quiescent state,
borhoods. In two dimensions the RACA for a von
because of Lemma 2. Note that the properties of the Neumann neighborhood has the form
1,1
coefficients Ct,i when taken modulo p alone suffice to
establish all this. 2 xi,j i,j−1
t+1 = xt + xi,j i,j+1
t + xt + xi−1,j
t + xi+1,j
t − xi,j
t−1
mod p (43)

IV. DISCUSSION AND MODIFICATIONS In this case, the pattern obtained after (p − 1)/2 time
steps starting from an arbitrary initial condition of
side, at most p, is replicated at time T = p+(p−1)/2.
The above result can be generalized to two and However, four copies are produced for the pattern in
three dimensions. If two dimensions are considered, this case, that compare to the two copies produced by
but the dynamics takes place only along one dimen- the 1D RACA, Eq. (15). This is observed in Fig. 3
sion, the map admits a straigtforward generalization in which the spatiotemporal evolution of Eq. (43) is
as shown for p = 13, starting from an initial condition
xi,j i,j
−1 = x0 contained in a small rectangle, as shown
xi,j i,j−1
t+1 = xt + xi,j i,j+1
t + xt − xi,j
t−1 mod p (41) in the leftmost, uppermost panel. Subsequent time
steps are shown and it is observed that four copies are
Here i, j ∈ Z give the coordinates of a site on a dis- produced at t = p + (p − 1)/2 = 19 of the pattern
crete, planar lattice. The dynamics takes place only obtained at t = (p − 1)/2 = 6 through a process that
along the j direction. By selecting an initial condition resembles budding in biological systems. Four buds
xi,j i,j
−1 = x0 where the non-zero sites are contained on a are originated at t = p = 13 and these grow until
rectangle and such that along the j direction the con- the patterns representing the mature individuals are
ditions of Theorem 2 are satisfied (the nonzero sites obtained, being then separated from the main body
occupy a region of side δ such that 2δ + 1 < p), the by sites in the quiescent state.
 7

FIG. 3: Spatiotemporal evolution of map Eq. (43) for the initial condition at t = −1 and t = 0 shown in the leftmost and uppermost panel
and p = 13. Subsequent time steps (indicated in the panels) are shown and it is observed that the pattern obtained at t = (p − 1)/2 = 6 is
replicated at time t = p + (p − 1)/2 = 19 from four ’buds’ generated at t = p, yielding four copies separated by sites in the quiescent state
(dark blue).

For a Moore neighborhood the RACA (p − 1)/2 replicates at time T = p + (p − 1)/2. This
1 1   time, eight copies of the pattern are obtained.
X X
xi,j
t+1 = xi−k,j−m
t − xi,j
t−1 mod p (44) Let f (xt ) denote the right hand side of Eqs. (15),
k=−1 m=−1
(41), (43) or (44). Modifications can be introduced to
also satisfies the fact that the pattern obtained at time any of these maps by following the methods in [23] to
 8

create (conditional) predictability, mutations on the V. CONCLUSIONS

patterns, or fixed points. For example, if we wish that
the map given by Eq. (41) does not evolve further
after the replication has taken place at T = p + (p −
1)/2 (so that the replicated structures are fixed points) In this article it has been rigorously shown that
we can modify the map as follows certain RACAs are able to replicate spatial patterns
found in their spatiotemporal evolution in a way that
resembles budding in biological systems. The RACAs
 
xi,j i,j
t+1 = xt + xt
i,j−1
+ xi,j+1
t − xi,j
t−1 (45)
  are computationally inexpensive and using both large
p−1 1 lattices and a prime number p large, spectacular re-
×B t − ,p + mod p sults can be obtained that can be used in the math-
2 2
ematical modeling of multiplication in biological sys-
where the boxcar function tems: even when three-dimensional systems (cells) are
  involved, the dynamics takes place along an axis in
1 x+y x−y one dimension and the extension of Eq. (41) to 3D is
B(x, y) := − (46)
2 |x + y| |x − y| straightforward.
was introduced in [5] to formulate a universal map for
cellular automata. Eq. (45) behaves as Eq. (41) for The RACAs here studied can, indeed, be adapted
t ≤ T and yields xi,j i,j to model the main features of life [24]: (a) Multiplica-
t+1 = xt for t > T . Note that
the creation of this fixed point involves breaking the tion (or replication), i.e. the ability of an individual
reversibility of Eq. (41) since, by its very definition, a to produce two; (b) Heredity, i.e. there are different
reversible CA does not allow for fixed points. kinds of individuals and these produce offsprings like
Let St := xi,j i,j−1
+ xi,j+1 themselves and (c) Variation: Heredity is not perfect
t + xt t . The following mod-
ification of Eq. (41) so that, occassionally, the replicated structures have
mutations. We have shown that the RACAs here pre-
sented satisfy property (a) naturally giving rise to the
 
xi,j i,j
t+1 = xt + xt
i,j−1
+ xi,j+1
t − xi,j
t−1 (47)
   replicated patterns in a way that resembles biological
1 processes. Property (b) is also naturally incorporated,
× 1 − B St , mod p since different initial conditions will lead to different
2
patterns at time (p−1)/2 from which copies like them-
has a behavior similar to Eq. (41) except in those selves will be produced at time p + (p − 1)/2. Finally,
places where xi,j
t = xt
i,j−1
= xi,j+1
t = St = 0, in which property (c) can be achieved by means of appropriate
i,j
case, from Eq. (47) xt+1 = 0 regardless of the value of adaptations of the RACAs, using the methods in [23],
as explained in the modification given by Eq. (47) of
xi,j
t−1 . Again this breaks the reversibility of the rule: the original RACA, Eq. (41).
If a quiescent neighborhood is attained reversibly, the
quiescent state is not left. This is a realistic, physical
modification because once the dynamics leads to a It is also possible to embed the RACA here pre-
quiescent neighborhood it is physical to require that sented in continuous structures (coupled map lattices
no activity will arise ex nihilo. Once structures are [25]) so that, although the dynamics is still dictated by
separated, they will not coalesce again in the next the discrete evolution of the RACA, continuous struc-
time step. It is observed in the simulations that, for tures emerge and replicate. This embedding, called
certain initial conditions and prime numbers of the nonlinear Bκ -embeddings allow to construct complex
form p = 4n + 3, n ∈ N defects are introduced in shapes out of a discrete amount of information [26]. If
the replication process leading to “mutations” in the the shapes are constructed at each time of the RACA
copies of the patterns replicated. evolution, continuous complex structures can be thus
replicated. These lattice functions for CAs will be
discussed elsewhere.

[1] S. Wolfram, A New Kind of Science. (Champaign, IL, (CRC Press, Boca Raton, FL, 1994).
Wolfram Media, 2002). [3] B. Voorhees, Additive Cellular Automata. In: Mey-
[2] A. Adamatzky, Identification of cellular automata. ers R. (eds) Encyclopedia of Complexity and Systems
 9

Science. (Springer, New York, NY, 2009). [16] O. Martin, A. M. Odlyzko, and S. Wolfram, Commun.
[4] S. Wolfram, Amer. Math. Mon. 91 (1984) 566-571. Math. Phys. 93 (1984) 219.
[5] V. Garcı́a-Morales, Phys. Lett. A 376 (2012) 2645. [17] B. A. Bondarenko, Generalized Pascal Triangles
[6] V. Garcı́a-Morales, Phys. Lett. A 377 (2013) 276. and Pyramids (The Fibonacci Association, Dalhousie,
[7] P. P. Chaudhuri, D. R. Chowdhuri, S. Nandi and Nova Scotia, Canada, 1993).
S. Chattopadyhay, Additive cellular automata. The- [18] H.-O. Peitgen, H. Jürgens and D. Saupe, Chaos and
ory and applications. Vol. 1. (Wiley, New York, NY, Fractals: New Frontiers of Science (Springer, New
1997). York, 2004), pp. 377-224.
[8] A. Fúster-Sabater and P. Caballero-Gil, Appl. Math. [19] J. Gravner, G. Gliner and M. Pelfrey, Physica D 240
Lett. 22 (2009) 1518-1524. (2011) 1460-1474.
[9] H. Yamada and S. Amoroso, Inform. Contr. 14 (1969) [20] J. Kari, Physica D 45 (1990) 379-385.
299. [21] K. Dilcher, Siam J. Math. Anal. 19 (1988) 473-483.
[10] S. Amoroso and G. Cooper, J. Comput. Syst. Sci. 5 [22] N. J. Fine, Amer. Math. Mon. 54 (1947) 589.
(1971) 455. [23] V. Garcı́a-Morales, Phys. Rev. E 88 (2013) 042814.
[11] T. J. Ostrand, J. Comput. Syst. Sci. 5 (1971) 623. [24] J. Maynard Smith and E. Szathmáry, The origins of
[12] W. L. Hamilton and J. R. Mertens, J. Comput. Syst. life: From the birth of life to the origin of language
Sci. 10 (1975) 248. (Oxford University Press, Oxford, UK, 1999).
[13] P. G. Anderson, J. Comput. Syst. Sci. 12 (1976) 394. [25] V. Garcı́a-Morales, J. Phys. A: Math. Theor. 49,
[14] A. G. Barto, J. Comput. Syst. Sci. 16 (1978) 445. 295101 (2016).
[15] M. Ito, N. Osato, and M. Nasu, J. Comput. Syst. Sci. [26] V. Garcı́a-Morales, Chaos. Sol. Fract. 152 (2021)
27 (1983) 125. 111426.