betaFeedback

WorksAgricultural and Biological Sciences

Can a self-sustaining chemical network emerge due to just a few catalysts?

Abstract

Understanding how self-sustaining chemical reaction networks can emerge from simple molecules is an important question in chemical approaches to the origin of life. Reflexively autocatalytic and food-generated (RAF) sets provide a mathematical framework for investigating this question. This line of research goes back to the early work of Kauffman on autocatalytic reaction networks and was subsequently developed through the probabilistic framework of Mossel and Steel. Further contributions by Hordijk, Steel, and collaborators extended the theory to more heterogeneous forms of catalysis, including reaction-dependent probabilities or power-law distributions of catalytic activity. We introduce a stochastic model that combines heterogeneous molecular catalytic activity with reaction-dependent molecule-reaction compatibility. By assigning random catalytic weights to molecules and incorporating compatibility through an explicit formulation of catalysis probabilities, our model provides a concrete and flexible setting that captures both sources of heterogeneity while remaining mathematically tractable. We establish novel and general criteria for RAF formation and nonformation. We further demonstrate that classical phase-transition criteria need not remain valid in the presence of reaction-dependent catalysis. For heavy-tailed catalytic weights, we identify and rigorously quantify an additional phase transition. In models with finite-mean catalytic weights, RAF formation and nonformation are governed by a catalytic intensity parameter. For infinite-mean weights, however, we show that RAFs may emerge with probability tending to one even when the catalytic intensity parameter tends to zero.

Introduction

A self-sustaining chemical reaction system is one in which molecules are produced from available resources through reactions that collectively support one another. Such systems have attracted considerable attention because they may help explain how organized chemical systems emerged during the origin of life. An interesting mathematical question is how likely a self-sustaining system is to arise when the catalysis network is random. That is, for a fixed set of possible reactions, if the catalytic relationships between molecules and reactions are randomly determined, how likely is it that a self-sustaining system exists? This question has been studied in a series of pioneering works by Mossel, Steel, Hordijk, and their collaborators, who developed mathematical frameworks for studying the emergence of reflexively autocatalytic and food-generated (RAF) sets. In particular, Mossel and Steel [13] established fundamental conditions under which RAFs emerge with high probability in random polymer models. Our study extends this line of research by introducing random molecular weights that represent differences in catalytic activity among molecules, while allowing the probability of catalysis to depend on both the molecule and the reaction. In this setting, we provide new insights into the possibility that self-sustaining systems can emerge even when catalytic activity is concentrated in a small number of molecules.

The emergence of self-sustaining chemical reaction networks has long been considered a potentially important step in the origin of life. Kauffman [8] proposed that sufficiently complex molecular systems could give rise to collectively autocatalytic networks, an idea subsequently explored in polymer-based models of chemical reactions. Experimental studies have demonstrated self-sustaining replication and cooperative catalytic networks in RNA systems. More recent research has investigated autocatalysis in connection with molecular replication, catalytic regulation, spatial organization, and the organization of prebiotic chemical systems [12, 10, 15, 16, 20, 1]. These developments motivate the mathematical investigation of the conditions under which autocatalytic networks can emerge.

Among the many previous studies, several papers are particularly relevant to our work. Mossel and Steel [13] provided a rigorous framework for random catalytic networks with independent catalysis events and reaction-independent catalysis probabilities for each molecule. They established fundamental bounds on the amount of catalysis required for RAF formation as the maximum polymer length increases. Hordijk, Wills and Steel [7] extended the investigation to reaction-dependent catalysis mechanisms in the binary polymer model. Hordijk and Steel [6] studied heterogeneous catalytic activity with each molecule assigned to a random number of reactions that it catalyzes, while the reactions are selected uniformly. They showed that the distribution of catalytic activity can significantly affect the probability and size of RAFs, while universal linear order bounds on the average catalytic activity required for RAF formation continue to hold. Phase transitions in RAF probability with respect to the catalytic probability was also studied in Varanasi and Korenaga [19]. See § 2.2.1 for more details about comparison to previous studies.

The main structure of the model used in this paper follows the frameworks developed by Mossel and Steel [13] and Hordijk and Steel [6]. In the reaction-independent models considered in [13, 6], the probability that a molecule catalyzes a reaction depends on the molecule but not on the particular reaction. We extend this framework by considering intensity parameters, which determine the probability of catalysis and depend on both the catalyst and the reaction. We also introduce random molecular weights to represent heterogeneity in catalytic activity among molecules. Our extension is analogous to the extension from the Erdős–Rényi random graph model to inhomogeneous random graph models, where the probability of an edge depends on the properties of the vertices. In our setting, the molecules and reactions form two classes of vertices, and their catalytic relationships are determined by reaction-dependent intensities and molecular weights.

Our main results provide new phase-transition criteria for the probability of existence of a RAF in terms of the catalytic intensities and molecular weights. In particular, we identify regimes in which P(RAF)→1\mathbb{P}(\mathrm{RAF}) \to1 and regimes in which P(RAF)→0\mathbb{P}(\mathrm{RAF}) \to0 as the maximum polymer length nn tends to infinity (see Theorem 1). These results reveal phenomena that are qualitatively different from those characterized by earlier models. In particular, Mossel and Steel [13] established fundamental criteria for RAF formation in terms of the expected number of reactions catalyzed by individual molecules. Our results extend their analysis to a more general setting by introducing random molecular weights and allowing catalysis probabilities to depend on both molecules and reactions.

One particularly interesting consequence is that a RAF can emerge with probability tending to one even in regimes where catalytic activity is concentrated in a small number of molecules (see Figure 2). This can happen even with a single molecule (see Proposition 10). This consequence is consistent with the numerical simulations shown in Hordijk and Steel [6]. In particular, heavy-tailed molecular weights allow a few molecules to have substantially greater catalytic influence than the others, while the overall intensity of catalysis may decrease with the system size. Nevertheless, sufficiently strong and appropriately distributed catalytic interactions can ensure that P(RAF)→1\mathbb{P}(\mathrm{RAF}) \to1 as n→∞n \to\infty. This suggests that the emergence of collectively autocatalytic systems does not necessarily require catalytic activity to be distributed broadly among molecules. Instead, a small number of highly active molecules may play a decisive role in supporting the emergence of a RAF. Although our model does not directly describe the chemical dynamics of early life, this phenomenon provides a possible mathematical perspective on how organized autocatalytic systems might emerge from heterogeneous molecular populations.

Examples of reaction-independent and reaction-dependent networks containing RAFs

Figure 2. Examples of networks containing RAFs in the supercritical regime of Theorem 1, with λn→0\lambda_n\to0 as n→∞n\to\infty. The left and right panels correspond to the reaction-independent and reaction-dependent kernels, respectively. For a reaction rr of the form u+v→uvu+v\to uv or uv→u+vuv\to u+v, its size ∣r∣|r| is defined as the length ∣uv∣|uv| of the polymer uvuv. Molecule and reaction nodes are vertically arranged according to their sizes. “RAF catalyst molecules” are molecules that catalyze at least one reaction in the RAF. The “degree” of a molecule denotes the number of reactions in the RAF that it catalyzes.

One of the main mathematical challenges is the dependence of catalysis probabilities on both catalysts and reactions, together with an additional source of randomness in the model arising from molecular weights. Our proofs of the RAF formation criteria follow the general constructive approach of Mossel and Steel [13], which was subsequently extended to reaction-dependent catalysis mechanisms by Hordijk, Wills and Steel [7]. The key observation in our weighted model is that, conditional on the molecular weights, catalysis events are independent, and the probability that a given reaction rr receives no catalyst is exactly exp⁡(−λnMn(r))\exp(-\lambda_n M_n(r)), where Mn(r)M_n(r) denotes its weighted catalytic mass (see (1)). This allows us to formulate general criteria for RAF formation in terms of the asymptotic behavior of these masses. The resulting criteria are independent of the specific weight distribution, so that the subsequent analysis of heavy-tailed weights reduces to establishing the corresponding asymptotic properties of the catalytic masses.

The main novelty of our results lies in their formulation in terms of weighted catalytic masses, which provides general criteria for RAF formation that simultaneously accommodate reaction-specific catalysis and heterogeneous molecular catalytic activity. This framework allows us to identify and mathematically quantify phase transitions, in particular for heavy-tailed catalytic weights, and to distinguish different regimes of RAF formation. Beyond these phase-transition results, our analysis also reveals limitations of earlier criteria. In particular, we show that reaction-dependent catalysis alone already provides counterexamples to extending the Mossel–Steel criteria beyond their original assumptions. Thus, reaction specificity and molecular heterogeneity are not merely additional features of random catalytic models; they can fundamentally affect the probability of RAF formation.

Model and main results

We begin this section by introducing the fundamental definitions and notation used throughout the paper.

Catalytic reaction systems and autocatalytic sets

The catalytic reaction systems in this paper have the same format and structure as shown in [13].

Fix a finite alphabet A={a1,…,ak}\mathcal{A} = \{a_1,\ldots,a_k\} with k≥2k \ge2. For n∈Nn \in\mathbb{N}, let Xn:=⋃j=1nAjX_n := \bigcup_{j=1}^{n}\mathcal{A}^j be the set of all non-empty strings over A\mathcal{A} of length at most nn. Its cardinality is ∣Xn∣=∑j=1nkj=kn+1−kk−1|X_n| = \sum_{j=1}^{n} k^j = \frac{k^{n+1}-k}{k-1}. For x∈Xnx \in X_n, write ∣x∣|x| for its length (or size). For example, if x=a1a2x = a_1a_2, then ∣x∣=2|x| = 2.

The deterministic reaction system. Whenever u,v∈Xnu,v \in X_n satisfy ∣u∣+∣v∣≤n|u|+|v|\le n, we allow the ligation reaction u+v⟶uvu+v\longrightarrow uv, where uvuv denotes concatenation. The set of forward ligation reactions is Rn+:={(u,v;uv):u,v∈Xn, ∣u∣+∣v∣≤n}R_n^+ := \{(u,v;uv): u,v\in X_n,\ |u|+|v|\le n\}. Associated with each ligation reaction is the reverse cleavage reaction uv⟶u+vuv\longrightarrow u+v. Accordingly, define

Rn−:={(uv;u,v):(u,v;uv)∈Rn+},Rn:=Rn+∪Rn−.R_n^- := \{(uv;u,v):(u,v;uv)\in R_n^+\},\qquad R_n:=R_n^+\cup R_n^-.

Forward and reverse reactions are regarded as distinct directed reactions. When only ligation reactions are required, we restrict attention to Rn+R_n^+. For each r∈Rnr\in R_n, let ρ(r)⊆Xn\rho(r)\subseteq X_n and π(r)⊆Xn\pi(r)\subseteq X_n denote its sets of reactants and products, respectively. Thus

ρ((u,v;uv))={u,v},π((u,v;uv))={uv},\rho((u,v;uv))=\{u,v\},\qquad\pi((u,v;uv))=\{uv\},
ρ((uv;u,v))={uv},π((uv;u,v))={u,v}.\rho((uv;u,v))=\{uv\},\qquad\pi((uv;u,v))=\{u,v\}.

These sets record molecule types rather than stoichiometric multiplicities; the generation conditions below concern the availability of these types.

Food set and FF-generation. Fix an integer t≥1t\ge1, independently of nn, and henceforth assume n>tn>t. Define the food set by

Fn:={x∈Xn:∣x∣≤t}=Xt.F_n:=\{x\in X_n:|x|\le t\}=X_t.

In particular, FnF_n is independent of nn for n>tn>t. For R′⊆RnR'\subseteq R_n, define the closure of FnF_n under R′R', denoted cl⁡R′(Fn)\operatorname{cl}_{R'}(F_n), to be the smallest subset Y⊆XnY\subseteq X_n satisfying Fn⊆YF_n\subseteq Y and

ρ(r)⊆Y⟹π(r)⊆Yfor every r∈R′.\rho(r)\subseteq Y\quad\Longrightarrow\quad\pi(r)\subseteq Y\qquad\text{for every }r\in R'.

Equivalently, cl⁡R′(Fn)\operatorname{cl}_{R'}(F_n) is the collection of molecules that can be generated from the food set by successively applying reactions from R′R'. A reaction set R′⊆RnR'\subseteq R_n is called FF-generated if

ρ(r)⊆cl⁡R′(Fn)for every r∈R′.\rho(r)\subseteq\operatorname{cl}_{R'}(F_n)\qquad\text{for every }r\in R'.

Catalysis configurations. A catalysis configuration is a subset Cn⊆Xn×RnC_n\subseteq X_n\times R_n. We interpret (x,r)∈Cn(x,r)\in C_n as the statement that molecule xx catalyses reaction rr. Equivalently, introduce the indicators Ix,r:=1{(x,r)∈Cn}I_{x,r}:=\mathbf{1}_{\{(x,r)\in C_n\}}, x∈Xnx\in X_n, r∈Rnr\in R_n. The definitions below apply to any catalysis configuration; its probability law will be specified subsequently.

Reflexively autocatalytic and food-generated (RAF) sets. Given a catalysis configuration CnC_n, a non-empty reaction set R′⊆RnR'\subseteq R_n is called reflexively autocatalytic and food-generated, abbreviated RAFRAF, if the following two conditions hold.

  1. FF-generation. Every reactant required by R′R' can be generated from the food set:

ρ(r)⊆cl⁡R′(Fn)for all r∈R′.\rho(r)\subseteq\operatorname{cl}_{R'}(F_n)\qquad\text{for all }r\in R'.
  1. Reflexive autocatalysis. Every reaction in R′R' is catalysed by at least one molecule available from the food set or generated by R′R':

∀r∈R′,∃x∈cl⁡R′(Fn) such that Ix,r=1.\forall r\in R',\qquad\exists x\in\operatorname{cl}_{R'}(F_n)\text{ such that }I_{x,r}=1.

When the catalysis configuration is random, we write RAFnRAF_n for the event that the resulting catalytic reaction system contains at least one RAF. For a system restricted to forward reactions, the same definitions apply with RnR_n replaced by Rn+R_n^+. See Figure 1 for examples of a RAF set and a non-RAF set.

Examples of a RAF and a non-RAF reaction set in a binary polymer system

Figure 1. Examples of a RAF and a non-RAF reaction set in a binary polymer system. Solid black arrows indicate reactions, while orange dashed arrows indicate catalysis. (i) R′={r1,r2}R' = \{r_1,r_2\} is a RAF: both reactions are FF-generated and catalyzed within cl⁡R′(F)\operatorname{cl}_{R'}(F). (ii) R′={r1,r2}R' = \{r_1,r_2\} is not a RAF: the reactant cc of r2r_2 is not in cl⁡R′(F)\operatorname{cl}_{R'}(F), so R′R' is not FF-generated. Moreover, r2r_2 is catalyzed only by cc, which is also outside the closure. Hence both RAF conditions fail.

A random catalysis model with heterogeneous molecular activity

We now introduce a new model of random catalysis. To account for heterogeneity in catalytic activity, associate with each molecule x∈Xnx \in X_n a random weight WxW_x. Assume that {Wx:x∈Xn}\{W_x : x \in X_n\} are independent and identically distributed copies of a random variable WW satisfying P(0<W<∞)=1\mathbb{P}(0 < W < \infty) = 1. The weight WxW_x represents the intrinsic catalytic activity of molecule xx. No moment assumptions on WW are imposed at this stage. Let λn∈[0,∞)\lambda_n \in[0,\infty) be a deterministic intensity parameter, and let Kn:Xn×Rn⟶[0,∞)K_n : X_n \times R_n \longrightarrow[0,\infty) be a deterministic kernel taking finite values. The quantity Kn(x,r)K_n(x,r) describes the compatibility of molecule xx with reaction rr. Define the sigma-algebra generated by the molecular weights as Wn:=σ(Wx:x∈Xn)\mathcal{W}_n := \sigma(W_x : x \in X_n). Conditional on Wn\mathcal{W}_n, the catalysis indicators {Ix,r:(x,r)∈Xn×Rn}\{I_{x,r} : (x,r) \in X_n \times R_n\} are mutually independent, with

P(Ix,r=1∣Wn)=1−exp⁡{−λnWxKn(x,r)}\mathbb{P}(I_{x,r}=1 \mid\mathcal{W}_n) = 1-\exp\{-\lambda_n W_x K_n(x,r)\}

almost surely.

Note that our model relaxes both assumptions (R1) and (R2) of Mossel and Steel [13]. Their assumption (R1) requires the catalysis indicators to be mutually independent. Assumption (R2) requires the probability that a molecule catalyses a forward reaction to depend only on the molecule xx. Observe that both of these assumptions need not hold in our model. In particular independence of Ix,rI_{x,r} only hold conditioned on Wn\mathcal{W}_n.

References

  1. [1]A. Despons. Nonequilibrium properties of autocatalytic networks. Physical Review E, 111(1):014414, 2025.DOI
  2. [2]R. Golnik, T. Gatter, W. Hordijk, P. F. Stadler, and N. Vassena. Bridging two theoretical frameworks of autocatalysis: Raf sets and stoichiometric autocatalysis. arXiv preprint arXiv:2605.25523, 2026.arxiv.org/abs/2605.25523
  3. [3]R. Golnik, T. Gatter, W. Hordijk, P. F. Stadler, and N. Vassena. Minimality in reflexive and stoichiometric autocatalysis. arXiv preprint arXiv:2609.23602, 2026.arxiv.org/abs/2609.23602
  4. [4]R. Golnik, T. Gatter, P. F. Stadler, and N. Vassena. Enumeration of autocatalytic subsystems in large chemical reaction networks. Journal of Chemical Theory and Computation, 22(10):4888, 2026.DOI
  5. [5]R. Golnik, N. Vassena, P. F. Stadler, and T. Gatter. Using mr-chordless circuits for efficient enumeration of autocatalytic cores in large chemical reaction networks. Journal of Cheminformatics, 18(1):117, 2026.DOI
  6. [6]W. Hordijk and M. Steel. Autocatalytic sets in polymer networks with variable catalysis distributions. Journal of Mathematical Chemistry, 54:1997–2021, 2016.arxiv.org/abs/1605.03919
  7. [7]W. Hordijk, P. R. Wills, and M. Steel. Autocatalytic sets and biological specificity. Bulletin of Mathematical Biology, 76(1):201–224, 2014.arxiv.org/abs/1307.2860
  8. [8]S. A. Kauffman. Autocatalytic sets of proteins. Journal of Theoretical Biology, 119(1):1–24, 1986.
  9. [10]D. V. Kriukov, J. Huskens, and A. S. Y. Wong. Exploring the programmability of autocatalytic chemical reaction networks. Nature communications, 15(1):8289, 2024.DOI
  10. [12]H. Lu, A. Blokhuis, R. Turk-MacLeod, J. Karuppusamy, A. Franconi, G. Woronoff, C. Jeancolas, A. Abrishamkar, E. Loire, F. Ferrage, et al. Small-molecule autocatalysis drives compartment growth, competition and reproduction. Nature Chemistry, 16(1):70–78, 2024.DOI
  11. [13]E. Mossel and M. Steel. Random biochemical networks: The probability of self-sustaining autocatalysis. Journal of Theoretical Biology, 233(3):327–336, 2005.pubmed.ncbi.nlm.nih.gov/15652142
  12. [14]I. Norros and H. Reittu. On a conditionally poissonian graph process. Advances in Applied Probability, 38(1):59–75, 2006.DOI
  13. [15]A. M. Plum, C. P. Kempes, Z. Peng, and D. A. Baum. Spatial structure supports diversity in prebiotic autocatalytic chemical ecosystems. npj Complexity, 2:21, 2025.DOI
  14. [16]Y. Sakref and O. Rivoire. Design principles, growth laws, and competition of minimal autocatalysts. Communications Chemistry, 7:239, 2024.arxiv.org/abs/2403.19047
  15. [17]M. Steel. The emergence of a self-catalysing structure in abstract origin-of-life models. Applied Mathematics Letters, 13(3):91–95, 2000.DOI
  16. [18]N. Vaidya, M. L. Manapat, I. A. Chen, R. Xulvi-Brunet, E. J. Hayden, and N. Lehman. Spontaneous network formation among cooperative rna replicators. Nature, 491(7422):72–77, 2012.DOI
  17. [19]V. Varanasi and J. Korenaga. Emergence of autocatalysis in prebiotic reaction networks. Physical Review E, 113(4):044304, 2026.DOI
  18. [20]S. A. Zorc and R. N. Roy. Origin and influence of autocatalytic reaction networks at the advent of the RNA world. RNA Biology, 21(1):1023–1037, 2024.

Paper details

vlnc:W_k7enkdkp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY-SA 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Agricultural and Biological Sciences

Not recorded for this paper

View statement
Looking for related papers…
Contents