1. Introduction
The main result of this work is the definition and analysis of a discrete random dynamical system leading to a distribution related to Catalan number recurrence. It is part of the program for linking discrete random dynamical systems to integer sequences [
1]. The motivation for the program results from previous works [
2,
3] on probabilistic cellular automata [
4] leading to Catalan and Motzkin numbers [
5,
6]. These form one more interpretation of each these two integer sequences, as originating from the limiting distributions for discrete dynamical systems.
The achieved results suggest that for other integer sequences also, appropriate dynamical systems can be defined to derive the respective recurrences. However, apart from the examples of Catalan and Motzkin numbers, they have not been defined so far. The implementation of this program creates an interesting link between integer sequences and dynamical systems.
Integer sequences can arise from dynamical systems in the contexts of closed orbits of a point under the action of map iteration (see, for example, [
7,
8,
9]), and from queueing theory (see [
10]). Our approach focuses on the construction of a random dynamical system, which is described by a (stationary state) distribution that is given by the recurrence associated with a specific integer sequence.
The achieved results [
2,
3] suggest that for other integer sequences also, appropriate dynamical systems can be defined to derive the respective recurrences. However, apart from the examples of Catalan and Motzkin numbers, they have not been defined so far. The implementation of this program creates an interesting link between integer sequences and dynamical systems.
The dynamical system is inspired by a very simplified view of earthquakes as driven by slow accumulation of energy (possibly generated by respective slow motions in the earth’s crust) and its abrupt releases in the form of quakes. In particular, in the case of a toy model of earthquakes in the form of Random Domino Automaton (see [
11] and the references therein), it was discovered that the model generates a size distribution of clusters, which, after re-scaling, coincides with Motzkin number recurrence [
6]. A similar system was proposed for Catalan numbers [
3]. It should be emphasized that the previous work was framed using physical terms—here, we present a more rigorous, analytical approach.
This article defines a new system for which—unlike the system [
3]—there are no space correlations. Relevant stationary-state variables are introduced and show how the equations they are supposed to fulfill result from counting all possibilities for all system states. The mean-field-type assumption is strictly formulated. Small-size terms, which are included in the group size distribution equation, disappear in the infinite size limit
. Finally, the solution to a special case is briefly discussed.
The model of aggregation and separation of individuals (which can be identified with portions of energy in the context of earthquakes) considered in this article can be illustratively described as follows. Consider the merger dynamics for N units, each of which can be in two mutually exclusive states: prone to merger, or to separation. Separated individuals are always solitary, unlike the aggregated individuals which occur in clusters of sizes . Random changes that occur in discrete time steps can only be as follows: a separating unit can turn into an aggregating unit with a certain constant probability, or a group of (only aggregating) units can either merge with another group, or separate. Both of these possibilities occur with probability depending on the size of the group.
What is the group size distribution in this process? Can it be calculated for the steady state? The short answer to this last question is yes, but the solution is a bit more complex than it might seem at first glance. In particular, Catalan number recurrence is an approximate solution. The nature of this approximation will be explained in detail in the text below. This article defines the system and analyzes it in a detailed way. Obviously, the distribution depends on the choice of model parameters; that is, two functions and one number.
The solution to the problem for the stationary state is given in the form of a recurrence equation derived with an assumption, which can be considered a kind of mean-field approximation. It has also been shown that the proposed derivation of the equations is internally consistent, i.e., the approximate group size distribution equations are consistent with two exact equations, namely for the proportions of aggregating and separating units and for the number of groups. Moreover, for a particular choice of parameters, the recurrence equation is reduced, in the limit of the infinite size of the system
N, to the form of Catalan number recurrence [
12].
The plan of the paper is as follows:
Section 2 provides the necessary notation and definitions;
Section 3 analyzes the system as a Markov process, introduces stationary variables and specifies the mean-field type approximation that is used later in the text;
Section 4 contains the derivation of the equations for the stationary state of the process; then,
Section 5 is devoted to the special case leading to Catalan number recurrence; finally,
Section 6 contains the possible interpretation of the process in physical terms and finishes with comments.
2. Definition of the System
In this section, we define the system of arbitrary size N and its evolution during discrete time steps. Dynamics of the groups (clusters) is expressed in terms of partitions.
2.1. The System
We start from N individuals, which can be in one of two states.
Definition 1 (States of elements).
Let denotes the set of the N first integers , and , where s and a are formal symbols. Then,defines the state of the element at time . The label s stands for separating state and the label a stands for aggregating state—the meaning of these terms is given in Definition 8. Since aggregative individuals occur in clusters, we introduce the notion of partitions for a set of indices of aggregative individuals—a group is formed by those individuals, for which indices belong to a block of the partition (at a given time step).
Definition 3 (State of the system).
Denote by the set of all possible partitions of The state of the system at the time t, , is given byThe block j of the partition is denoted by . Note that partitions are well-defined. In the following, we use the notation . Thus
Definition 4 (Size of a block). The size of a block is equal to .
Definition 5 (The number of blocks).
The number of blocks is denoted by . For the partition given by Formula (3), the number of blocks is . The most basic characteristic of the system is the ratio of the number of aggregative individuals to the number of separative ones, or equivalently, the density of aggregative units as given below.
Definition 6 (Density).
The density of the system is given by the ratio of the number of aggregative elements, i.e., those with , to the number of all elements 2.2. Evolution Rule
The evolution of the system depends on parameters, i.e., two functions and one number—the probability of merging and probability of separation , both depending on the size of respective cluster, and also on the probability of the transition of an individual from the separative to aggregative state.
Definition 7 (Dynamic parameters). The dynamic parameters are functions and are satisfying for all .
In the following, we use short notation and .
After introducing the above definitions, we can precisely determine the rules for the evolution of the system.
Definition 8 (Evolution rule).
The evolution of the system is given bywhere f is defined by the following rule. For a time t, choose a number from —with equal probability for each one—and denote it by j.(a)
If , then one from the two following excluding options (a1) and (a2) happens, with respective probabilities ν and , where is a fixed real number.
(a2)(b)
If , then it belongs to some block, say, of size . Then, one from the three following excluding options (b1), (b2) and (b3) is chosen, with respective probabilities , , and . (b1)
For consisting of at least two blocks, another block of the partition is chosen in such a way that every remaining block (i.e., different from ) has the same chance, and thenFor consisting of only one block(b2)(b3) The process related to the point (a1) is called transition, the one related to (b1) is called merging or aggregation and the process defined in (b2) is called separation.
4. Equations for the Stationary State of the System
For a system in a steady state, it is possible to derive equations balancing the “gains” and “losses” occurring during evolution. Below are the appropriate derivations for density, number of blocks and number of blocks of a specific size.
4.1. Density
Proof. The expected value for the number of elements changing state from aggregative to separative—which is possible due to separation rule (b2) of Definition 8 only—in a single time step, for any state
, is given by
Thus, multiplying by respective probability
and summing for all states
we arrive at
where we use Equation (
20).
The expected value for the number of elements changing state from separative to aggregative—which is possible due to transition rule (a1) of Definition 8 only—in a single time step, for any state
, is given by
The weighted sum for all states gives
where we use Equation (
18) and the property
. □
We emphasize that the Equation (
22) is exact—there were no approximations in its derivation.
4.2. The Number of Blocks
Proof. In a single time step, the number of blocks may increase—due to the transition rule (a1) of Definition 8—by one only, and the respective expected value is given by Formula (
26).
The number of blocks can decrease also by one only—due to the merging (b1) and separation (b2) rules of Definition 8. The respective contributions for a state
are
and next, after summation over all states
weighted by
, become
Hence, we obtain the Equation (
27). □
The Equation (
27) is exact, like Equation (
22). They have the same right-hand side, thus one obtains the following relation (
30), where the density is eliminated.
Proof. It follows directly from the Equations (
22) and (
27). □
4.3. The Number of Blocks of Size m
Similar reasoning as above can be applied to the number of blocks of a given size. It appears that the respective formulas depends on parity of the size. Thus, we introduce the following notation.
Proof. A decrease in the number of blocks of size
m can appear due to the merging (b1) and separation rules (b2) of Definition 8. For (b1), it may happen that
and
, thus the contributions to the expected value for any state
with
are
States with
and
do not contribute in this way. In effect, the contribution is
The mean-field approximation given by Equation (
21) leads to
and summation over all states
weighted by
leads to
where we included states with
and
, assuming that respective
s are negligible. This assumption is justified for large system, similarly to the mean-field approximation.
Next, the contribution of separation (b2) is similar to the above
An increase in the number of blocks of size
m can appear due to the rule (b1) of Definition 8, with
. In a way similar to above (i.e., writing appropriate contributions for a state
, applying the mean-field approximation (
21) and summing over all states
), one can obtain the following:
for
m odd, the contribution for a state
is
and for
m even, the contribution is
Hence, we obtain the Formula (
32). □
Since the total number of blocks is the sum of the number of blocks of all sizes, i.e., , we obtain the following fact.
Corollary 3. Equation (32) implies Equation (27). Proof. The result comes from summing Equation (
32) for all
and the following identities
□
Moreover, since the density of the system depends on the number of blocks of particular sizes, i.e., , then, similarly as before, we obtain the following fact.
Corollary 4. Equation (32) implies Equation (22). Proof. The result comes from summing Equation (
32) multiplied by respective
m for all
and the following identities
and a straightforward equality
□
Equations (
27) and (
22) may be understood as exact formulas for the first and the second moment of the distribution
weighted by functions
and
, respectively. The approximate Equation (
32) gives the distribution, for which the first and the second moments are exact.