Criticality conditions in the Derrida–Retaux model with a random number of terms

Alexey Lotnikov, Anna Kotova · 2025

Read Criticality conditions in the Derrida–Retaux model with a random number of terms by Alexey Lotnikov, Anna Kotova (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-02535.txt.

Introduction

Alexey Lotnikov and Anna Kotova study a branching math model in which each step sums a random number of copies, in a paper from two thousand twenty-five. A fixed amount is then subtracted, and negative results are cut to zero. Over many steps, the mean value either fades away or stays positive when scaled.

The scaled limit acts as the model's energy. The authors give clear conditions, based on the starting spread and the count spread, for each outcome. One outcome marks the fading, low regime. The other marks the regime where energy survives. The conditions settle when the model sits at its critical edge.

Catalog record

CREATORAlexey Lotnikov, Anna Kotova
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE15 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-02535

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Criticality conditions in the Derrida–Retaux model with a random number of terms Alexey Lotnikov, Anna Kotova Abstract Abstract The article considers the Derrida–Retaux model with a random number of terms, i.e. a sequence of integer random variables defined by the relations X_{n+1}=(X_{n}^{(1)}+X_{n}^{(2)}+...+X_{n}^{(N_{n})}-a)^{+} , n\geq 0 , where X_{n}^{j} are independent copies of X_{n} , the values of N_{j} are independent and identically distributed, a is a positive integer. The energy in the model is defined as Q:=\lim\limits_{n\to\infty}\frac{\mathbb{E}(X_{n})}{(\mathbb{E}N_{1})^{n}} . We present sufficient conditions (in terms of distributions of X_{0} and N_{1} ) for subcritical ( Q=0 ) and supercritical ( Q>0 ) regimes of model behavior. 1 Introduction 1.1 Derrida — Retaux model The theory of fluctuations in chaotic systems is a branch of physics that studies the behavior of systems in which chaos and randomness are present. The purpose of the theory is to understand how fluctuations or small changes in the properties of a system can lead to global changes in its overall behavior. In particular, two states may be distinguished in the system and transitions between them are studied. The Derrida — Retaux model [ 3 ] in a simplified form describes such transitions (depinning transition) between certain states called "pinned" and "unpinned", which we do not need to describe. Similar relationships have also been studied in the context of physical [ 4 ] and mathematical studies [ 1 ] . The Derrida — Retaux model with a random number of terms, which is studied in this paper, is formulated as follows. The model parameters are an integer a>0 and two integer random variables: X_{0}\geq 0 – the initial value, and N\geq 1 – the number of terms. It is assumed that X_{0} has a finite first moment and is not a constant, as well as \mathbb{P}(N>1)>0 . The functioning of the system is determined by the recurrence relation: \displaystyle X_{n+1}=(X_{n}^{(1)}+X_{n}^{(2)}+...+X_{n}^{(N_{n+1})}-a)^{+}, (1) where X_{n}^{(1)} and X_{n}^{(2)} are independent copies of X_{n} , and \forall\ x\in\mathbb{R} , x^{+}:=\max(x,0) is the positive part of x . Let us note that \mathbb{E}(\sum\limits_{j=1}^{N_{n+1}}X_{n}^{(j)}-a)\leq\mathbb{E}(\sum\limits_{j=1}^{N_{n+1}}X_{n}^{(j)}-a)^{+}\leq\mathbb{E}\sum\limits_{j=1}^{N_{n+1}}X_{n}^{(j)}. Also \mathbb{E}\sum\limits_{j=1}^{N_{n+1}}X_{n}^{(j)}=\mathbb{E}N_{n+1}\cdot\mathbb{E}X_{n}=\mathbb{E}N\cdot\mathbb{E}X_{n} , therefore \mathbb{E}N\cdot\mathbb{E}X_{n}-a\leq\mathbb{E}X_{n+1}\leq\mathbb{E}N\cdot\mathbb{E}X_{n}, \cfrac{\mathbb{E}(X_{n+1})}{(\mathbb{E}N)^{n+1}}\leq\cfrac{\mathbb{E}(X_{n})}{(\mathbb{E}N)^{n}}\quad\text{ and }\quad\cfrac{\mathbb{E}(X_{n})-\cfrac{a}{\mathbb{E}N-1}}{(\mathbb{E}N)^{n}}\leq\cfrac{\mathbb{E}(X_{n+1})-\cfrac{a}{\mathbb{E}N-1}}{(\mathbb{E}N)^{n+1}}. Thus, the following limit is well-defined \displaystyle{Q}=\lim\limits_{n\rightarrow\infty}\downarrow\cfrac{\mathbb{E}(X_{n})}{(\mathbb{E}N)^{n}}=\lim\limits_{n\rightarrow\infty}\uparrow\cfrac{\mathbb{E}(X_{n})-\cfrac{a}{\mathbb{E}N-1}}{(\mathbb{E}N)^{n}}. (2) The parameter Q is commonly referred to as free energy, and the recurrence relation described above was first introduced in the article by Derrida and Retaux [ 3 ] and serves to describe the Depinning transition process. Studying this process is important in both mathematics and physics. Of particular interest is the dependence of the energy value Q on the initial data — the random variable X_{0} and a . The parameter a is referred to as the tax. Depending on the value of Q , two cases are distinguished: supercritical ( Q > 0) and subcritical ( Q = 0). A critical case is also distinguished, which essentially <<separetes>> the situations described above. The most interesting questions regarding this model arise in critical cases or cases close to them. The recurrence relation (1) with parameter a = 1 and determined numbers of terms was considered in the article [ 2 ] . This article provides necessary and sufficient conditions on the random variable X_{0} under which supercritical, subcritical, and critical cases occur, respectively. We attempted to generalize the results obtained in this article to arbitrary a\geq 1 and random numbers of terms. We found sufficient conditions on the random variable X_{0} guaranteeing subcriticality or supercriticality. 2 Main result The main result of our work is the following theorem. Let F_{0},G be the moment generating function of X_{0} and N respectively. Theorem 1 . Let it be D_{0}(s,m)=(m-1)sF^{\prime}_{0}(s)-aF_{0}(s) . 1) If D_{0}(\mathbb{E}N^{\frac{1}{a}},\mathbb{E}N)>0,\text{ then }Q>0 . 2) Let \exists M:\mathbb{P}(N\leq M)=1 . If D_{0}(1+\frac{M-1}{a},M)<0,\text{ then }Q=0. Let us emphasize that the first point is true without assuming that the random variable N is bounded. The result of Theorem 1 is interesting even if the number of terms is fixed, i.e. \mathbb{P}(N=n)=1 for some n . In this case, Theorem 1 reduces to the following result. Theorem 2 . Let it be D_{0}(s)=(n-1)sF^{\prime}_{0}(s)-aF_{0}(s) . 1) If D_{0}(n^{\frac{1}{a}})>0,\text{ then }Q>0 . 2) If D_{0}(1+\frac{n-1}{a})<0,\text{ then }Q=0 . If a=1 , then the conditions of points 1 and 2 stick together, and we get the result from [ 2 ] . If a>1 , then Theorem 2 is a new result. 3 Proof of the main result 3.1 Moment generating functions and their evolution Let F_{n}(s) be the moment generating function of X_{n} . Let us rewrite the recurrence relation ( 1 ) in terms of generating functions: \displaystyle F_{n+1}(s) \displaystyle= \displaystyle\frac{G(F_{n}(s))-\sum\limits_{p=0}^{a-1}\frac{s^{p}}{p!}(G(F_{n})^{(p)}(0))}{s^{a}}+\sum_{p=0}^{a-1}\frac{G(F_{n})^{(p)}(0)}{p!} (3) \displaystyle= \displaystyle\frac{G(F_{n}(s))}{s^{a}}+\sum_{p=0}^{a-1}(G(F_{n}))^{(p)}(0)\cfrac{1}{p!}\left(1-\frac{1}{s^{a-p}}\right). By differentiating equality ( 3 ) we get \displaystyle F^{\prime}_{n+1}(s)=\frac{G^{\prime}(F_{n}(s))F^{\prime}_{n}(s)}{s^{a}}-a\frac{G(F_{n}(s))}{s^{a+1}}+\sum_{p=0}^{a-1}G(F_{n})^{(p)}(0)\cdot\cfrac{a-p}{p!}\cdot\cfrac{1}{s^{a-p+1}}\text{.} (4) We will use these formulas a lot. 3.2 Proof of the first point of Theorem 1 The proof relies on two lemmas. First, we will prove the result using them, and the proof of the lemmas themselves will be presented below. Lemma 1 . If the assumption of the first statement of the theorem is true, then there exists 1<s<(\mathbb{E}N)^{\frac{1}{a}} , such that (\mathbb{E}N-1)\mathbb{E}(X_{n}s^{X_{n}})-a\mathbb{E}(s^{X_{n}})\rightarrow\infty\text{ as }n\rightarrow\infty. In terms of generating functions, Lemma 1 can be represented as follows: there is 1<s<(\mathbb{E}N)^{\frac{1}{a}} such that (\mathbb{E}N-1)sF^{\prime}_{n}(s)-aF_{n}(s)\rightarrow\infty\text{ as }n\rightarrow\infty. Lemma 2 . If Q=0 , then \sup\limits_{n\geq 0}\mathbb{E}(X_{n}s^{X_{n}})<\infty with 0<s<(\mathbb{E}N)^{\frac{1}{a}} . Now let us apply these lemmas to prove the statement. Let a\mathbb{E}(((\mathbb{E}N)^{\frac{1}{a}})^{X_{0}})<(\mathbb{E}N-1)\mathbb{E}(X_{0}((\mathbb{E}N)^{\frac{1}{a}})^{X_{0}}) ; we will show that Q>0 . By Lemma 1, there is 1<s<(\mathbb{E}N)^{\frac{1}{a}} , such that (\mathbb{E}N-1)\mathbb{E}(X_{n}s^{X_{n}})-a\mathbb{E}(s^{X_{n}})\rightarrow\infty,\text{ as $n$}\rightarrow\infty, therefore (\mathbb{E}N-1)\mathbb{E}(X_{n}s^{X_{n}})\rightarrow\infty\text{ as $n$}\rightarrow\infty . But if Q=0 , then by Lemma 2 \sup\limits_{n\geq 0}\mathbb{E}(X_{n}s^{X_{n}})<\infty . Hence Q>0 . Now we will provide proof of the lemmas. Proof of Lemma 1 By ( 4 ) s(\mathbb{E}N-1)F^{\prime}_{n+1}(s)=\frac{(\mathbb{E}N-1)sG^{\prime}(F_{n}(s))F^{\prime}_{n}(s)}{s^{a}}-a(\mathbb{E}N-1)\frac{G(F_{n}(s))}{s^{a}}+\sum_{p=0}^{a-1}(\mathbb{E}N-1)G(F_{n})^{(p)}(0)\cdot\cfrac{a-p}{p!}\cdot\cfrac{1}{s^{a-p}}\text{.} Let us give the lower bound for G^{\prime}(F_{n}(s)) . Let us prove the inequality: for any v\geq 1 , vG^{\prime}(v)\geq\mathbb{E}N\cdot G(v) . In terms of generating functions, it means: \mathbb{E}(Nv^{N})\geq\mathbb{E}N\cdot E(v^{N}) . For any independent copies N_{1},N_{2} of the random variable N the following is true: If (N_{1}-N_{2})(v^{N_{1}}-v^{N_{2}})\geq 0 , then 0\leq\mathbb{E}[(N_{1}-N_{2})(v^{N_{1}}-v^{N_{2}})]=2(\mathbb{E}(Nv^{N})-\mathbb{E}N\mathbb{E}v^{N}) . We note that for any v\geq 1 : (N_{1}-N_{2})(v^{N_{1}}-v^{N_{2}})\geq 0 . Thus, for any v\geq 1 , vG^{\prime}(v)\geq\mathbb{E}N\cdot G(v) . Therefore, F_{n}(s)G^{\prime}(F_{n}(s))\geq\mathbb{E}N\cdot G(F_{n}(s)) . Let us substitute this relation into the previous inequality: s(\mathbb{E}N-1)F^{\prime}_{n+1}(s)\geq\frac{\mathbb{E}N(\mathbb{E}N-1)sG(F_{n}(s))F^{\prime}_{n}(s)}{F_{n}(s)s^{a}}-a(\mathbb{E}N-1)\frac{G(F_{n}(s))}{s^{a}}+\sum_{p=0}^{a-1}(\mathbb{E}N-1)G(F_{n})^{(p)}(0)\cdot\cfrac{a-p}{p!}\cdot\cfrac{1}{s^{a-p}}\text{.} Therefore, s(\mathbb{E}N-1)F^{\prime}_{n+1}(s)-aF_{n+1}(s) \geq\frac{\mathbb{E}N\cdot G(F_{n}(s))}{F_{n}(s)s^{a}}\cdot\left(s(\mathbb{E}(N)-1)F^{\prime}_{n}(s)-aF_{n}(s)\right)+\sum_{p=0}^{a-1}G(F_{n})^{(p)}(0)\cdot\cfrac{1}{p!}\cdot\cfrac{(a-p)(\mathbb{E}N-1)+a-as^{a-p}}{s^{a-p}}. [formula] Let us prove that (a-p)(\mathbb{E}N-1)+a-as^{a-p}\geq 0 for p=0,1,\dots,a-1 and s\leq\mathbb{E}N^{\frac{1}{a}} . Rewrite this as: y(\mathbb{E}N-1)+a\geq as^{y}\text{, where }y=a-p>0. [formula] Let us introduce two functions f_{1}(y)=a(\mathbb{E}N)^{\frac{y}{a}}\text{ and }f_{2}(y)=y(\mathbb{E}(N)-1)+a . These functions are equal at y=0 and y=a and the function f_{1} is concave, while f_{2} is linear. Therefore, f_{2}(y)\geq f_{1}(y),\quad 0\leq y\leq a, namely, y(\mathbb{E}N-1)+a\geq a(\mathbb{E}N)^{\frac{y}{a}}\geq as^{y}. We have proved inequality (6) , hence the sum in (5) is non-negative. We conclude that (\mathbb{E}N-1)sF^{\prime}_{n+1}(s)-aF_{n+1}(s)\geq\frac{\mathbb{E}N\cdot G(F_{n}(s))}{F_{n}(s)s^{a}}[(\mathbb{E}N-1)sF^{\prime}_{n}(s)-aF_{n}(s)]. Let G(v)=\sum\limits_{k=1}^{\infty}a_{k}v^{k} , where a_{k}\geq 0 ( N>0 ). So \frac{G(v)}{v}=\sum\limits_{k=1}^{\infty}a_{k}v^{k-1}\geq\sum\limits_{k=1}^{\infty}a_{k}1^{k-1}=1 . This means that (\mathbb{E}N-1)sF^{\prime}_{n+1}(s)-aF_{n+1}(s)\geq\left(\frac{\mathbb{E}N}{s^{a}}\right)^{n}[(\mathbb{E}N-1)sF^{\prime}_{0}(s)-aF_{0}(s)]. Let p=\frac{\mathbb{E}N}{s^{a}}>1 , C_{1}=(\mathbb{E}N-1)sF^{\prime}_{0}(s)-aF_{0}(s) , then (\mathbb{E}N-1)sF^{\prime}_{n+1}(s)-aF_{n+1}(s)\geq p^{n}C_{1}, which goes to +\infty , if C_{1}>0 . As by assumption D_{0}(\mathbb{E}(N)^{\frac{1}{a}},\mathbb{E}N)>0 , hence, for s close to \mathbb{E}N^{\frac{1}{a}} we have D_{0}(s,\mathbb{E}N)>0 . Proof of Lemma 2 Let us fix k\geq 1 and n\geq 0. Consider the following inequality X_{n+k}\geq\sum_{i=1}^{T_{k}}\mathds{1}_{X_{n}^{(i)}\geq ak+1}, where X_{n}^{(i)},i\geq 1, are independent copies of X_{n} , and T_{k} is a random variable equal to the number of ancestors, when building our counting tree, at depth k . It follows, from independence and

Read the complete text