Introduction

The most prominent feature of life on Earth is its remarkable species diversity: countless macro- and micro-species fill every corner on land and in the water (Pennisi, 2005; Hoorn et al., 2010; Vargas et al., 2015; Daniel, 2005). In tropical forests, thousands of plant and vertebrate species coexist (Hoorn et al., 2010). Within a gram of soil, the number of microbial species is estimated to be 2,000-18,000 (Daniel, 2005). In the photic zone of the world ocean, there are roughly 150,000 eukaryotic plankton species (Vargas et al., 2015). Explaining the astonishing biodiversity is a major focus in ecology (Pennisi, 2005). A great challenge stems from the well-known competitive exclusion principle (CEP): two species competing for a single type of resources cannot coexist at constant population densities (Gause, 1934; Hardin, 1960), or generically, the number of consumer species cannot exceed that of resources at steady state (MacArthur and Levins, 1964; Levin, 1970; McGehee and Armstrong, 1977). On the contrary, in the paradox of plankton, a limited type of resources support hundreds or more coexisting species of phytoplankton (Hutchinson, 1961). Then, how can plankton and many other organisms somehow liberate the constraint of CEP?

Ever since MacArthur and Levin proposed the classical mathematical proof for CEP in the 1960s (MacArthur and Levins , 1964), various mechanisms have been put forward to overcome the limits set by CEP (Chesson , 2000). Some suggest that the system never approaches a steady state where the CEP applies, due to temporal variations (Hutchinson , 1961; Levins , 1979), spatial heterogeneity (Levin , 1974), or species’ self-organized dynamics (Koch , 1974; Huisman and Weissing, 1999). Others consider factors such as toxins (Lczárán et al., 2002), cross-feeding (Goyal and Maslov., 2018; Goldford et al., 2018; Niehaus et al., 2019), spatial circulation (Martín et al., 2020; Gupta et al., 2021), kill the winner (Thingstad, 2000), pack hunting (Wang and Liu, 2020), collective behavior (Dalziel et al., 2021), metabolic trade-offs (Posfai et al., 2017; Weiner et al., 2019), co-evolution (Xue and Goldenfeld, 2017), and other complex interactions among the species (Beddington, 1975; DeAngelis et al., 1975; Arditi and Ginzburg , 1989; Kelsic et al., 2015; Grilli et al., 2017; Ratzke et al., 2020). However, questions remain as to what determines species diversity in nature, especially for quasi-well-mixed systems such as that of the plankton (Pennisi, 2005; Sunagawa et al., 2020).

In this work, we present a mechanistic model of predator interference that extends the classical Beddington-DeAngelis (B-D) phenomenological model (Beddington, 1975; DeAngelis et al., 1975) with a detailed consideration of pairwise encounters. The intraspecific interference among consumer individuals effectively constitutes a negative feedback loop, enabling a wide range of consumer species to coexist with only one or a few types of resources. The coexistence state is resistant to stochasticity and can hence be realized in practice. Our model is broadly applicable and can be used to explain biodiversity in many ecosystems. In particular, it naturally explains species coexistence in classical experiments that invalidate CEP (Ayala, 1969; Park, 1954) and quantitatively illustrates the S-shaped pattern of the rank-abundance curves in an extensive spectrum of ecological communities, ranging from the communities of ocean plankton worldwide (Fuhrman et al., 2008; Ser-Giacomi et al., 2018), tropical river fishes from Argentina (Cody and Smallwood, 1996), forest bats of Trinidad (Clarke et al., 2005), rainforest trees (Hubbell, 2001), birds (Terborgh et al., 1990; Martínez et al., 2023), butterflies (Devries et al., 1997) in Amazonia, to those of desert bees (Hubbell, 2001) in Utah and lizards from South Australia (Cody and Smallwood, 1996).

Results

A generic model of pairwise encounter

Predator interference, i.e., the pairwise encounter among consumer individuals, is commonly described by the B-D model (Beddington, 1975; DeAngelis et al., 1975). From the mechanistic perspective, however, the functional response of the B-D model can be formally derived from the consumption between a consumer species and a resource species without involving any form of predator interference (Wang and Liu, 2020; Huisman and Boer, 1997) (see Appendix II B). To resolve this issue, we consider a mechanistic model of pairwise encounters (Fig. 1A). Specifically, SC consumer species competing for SR resource species . The consumers are biotic, while the resources can be either biotic or abiotic. For simplicity, we assume that all species are motile and each moves at a certain speed, namely, for consumer species Ci (i = 1, …, SC) and for resource species Rl, (l = 1, …, SR). For abiotic resources, they cannot propel themselves, yet may passively drift due to environmental factors. Each consumer is free to feast on one or multiple types of resources, while consumers do not directly interact with one another other than pairwise encounters.

Then, we proceed to explicitly consider the population structure of the consumers and resources: some are wandering around freely, taking Brownian motions; others are encountering with one another, forming ephemeral entangled pairs. Specifically, when a consumer individual Ci and a resource Rl, get close in space within a distance of (Fig. 1A), the consumer can chase the resource and form a chasing pair: (Fig. 1 B), where the superscript “(P)” represents pair. The resource can either escape with rate dil, or be caught and consumed by the consumer with rate kil. Meanwhile, when a Ci individual encounters another consumer Cj (j = 1, …, SC) of the same or different species within a distance of (Fig. 1A), they can stare at, fight against or play with each other and thus form an interference pair: (Fig. 1 B). This paired state is evanescent, and the two consumers separate from each other with rate d’il. For simplicity, we assume that all and are identical, respectively, i.e., ∀i, j, l, and .

In a well-mixed system with the size of L2 (Appendix-fig. 1), the encounter rates among the species, ail, and a’ij (Fig. 1B), can be obtained using the mean-field approximation: and (see Materials and Methods, and Appendix-fig. 1 for details). Then, we proceed to analyze scenarios involving different types of pairwise encounters. For the scenario involving only chasing pair, the population dynamics can be described as follows:

where , gl is an unspecified function, the superscript “(F)” represents the freely wandering population, Di denotes the mortality rate of Ci, and wil is the mass conversion ratio (Wang and Liu, 2020) from resource Ri to consumer Ci. With the integration of intraspecific predator interference, we combine Eq. 1 and the following equation:

where a’i = a’ii, d’i = d’ii, and represents the intraspecific interference pair. For the scenario involving chasing pair and interspecific interference, we combine Eq. 1 with the following equation:

where stands for the interspecific interference pair. In the scenario where chasing pair and both intra- and inter-specific interference are all relevant, we combine Eqs. 13, and the populations of consumers and resources are given by and , respectively.

Generically, the consumption and interference processes are much quicker compared to the birth and death processes. Thus, in derivation of the functional response, (Rl,Ci) = kllxil/Ci, the consumption and interference processes are supposed to be in fast equilibrium. In all scenarios involving different types of pairwise encounters, the functional response in the B-D model is a good approximation only for a special case with dil ≈ 0 and (Appendix-fig. 2, see Appendix II for details).

To facilitate further analysis, we assume that the population dynamics of the resources follows the same construction rule as that in MacArthur’s consumer-resource model (MacArthur, 1970; Chesson, 1990). Then,

In the absence of consumers, biotic resources exhibit logistic growth. Here, and represent the intrinsic growth rate and the carrying capacity of species Rl. For abiotic resources, stands for the external resource supply rate of Rl , and is the abundance of Rl at steady state without consumers. For simplicity, we focus our analysis on abiotic resources, although all results generally apply to biotic resources as well. By applying dimensional analysis, we render all parameters dimensionless (see Appendix VI). For convenience, we retain the same notations below, with all parameters considered dimensionless unless otherwise specified.

Intraspecific predator interference facilitates species coexistence and breaks CEP

To clarify the specific mechanisms that can facilitate species coexistence, we systematically investigate scenarios involving different forms of pairwise encounters in a simple case with SC = 2 and SR = 1. To simplify the notations, we omit the subscript/superscript “l” since SR = 1. For clarity, we assign each consumer species of unique competitiveness by setting that the mortality rate Di is the only parameter that varies with the consumer species.

A model of pairwise encounters may naturally break CEP.

(A) Ageneric model of SC consumer species feeding on SR resource species with pairwise encounters. (B) The well-mixed model of (A). (C-E) Time courses of two consumer species competing for one resource species. (F-H) Positive solutions to the steady-state equations: = 0 (orange surface), Ċ1 = 0 (blue surface), Ċ2 = 0 (green surface). The black/red dot represents the unstable/stable fixed point, while the dotted lines in (E) are the analytical solutions of the steady-state abundances (marked with superscript “(A)”). See Appendix VII for simulation details.

First, we conduct the analysis in a deterministic framework with ordinary differential equations (ODEs). In the scenario involving only chasing pair, consumer species cannot coexist at steady state except for special parameter settings (sets of measure zero) (Wang and Liu, 2020). In practice, if all species coexist, the steady-state equations of the consumer species (Ċi = 0) yield fi(R(F)) ≡ R(F)/(R(F) + Ki) = Di (i = 1,2), with Ki = (di + ki)/ai, which corresponds to two parallel surfaces in the (C1, C2, R) coordinates, making steady coexistence impossible (Wang and Liu, 2020) (Fig. 1C, F).

Meanwhile, in the scenario involving chasing pair and interspecific interference, if all species coexist, the steady-state equations correspond to three non-parallel surfaces Ω’i (R,C1, C2) = Di (i = 1,2), G’(R, C1, C2) = 0 (Fig. 1G and Appendix-fig. 3C, see Appendix IV for details), which can intersect at a common point (fixed point). However, this fixed point is unstable (Fig. 1G, Appendixfig. 3A), and thus one of the consumer species is doom to extinct (Fig. 1 D).

Next, we turn to the scenario involving chasing pair and intraspecific interference. Likewise, steady coexistence requires that three non-parallel surfaces Ωi (R, C1, C2) = Di (i = 1, 2), G (R, C1, C2) = 0 (Fig. 1H and Appendix-fig. 3D, see Appendix III for details) cross at a common point. Indeed, this naturally happens, and encouragingly the fixed point can be stable. Therefore, two consumer species may stably coexist at steady state with only one type of resources, which obviously breaks CEP (Fig. 1E Appendix-fig. 4A). In fact, the coexisting state is globally attractive (Appendix-fig. 4A), and there exists a non-zero volume of parameter space where the two consumer species stably coexist at constant population densities (Appendix-fig. 4B), demonstrating that the violation of CEP does not depend on special parameter settings. We further consider the scenario involving chasing pair and both intra- and inter-specific interference (Appendix-fig. 5). Much as expected, the species coexistence behavior is very similar to that without interspecific interference.

Intraspecific predator interference facilitates species coexistence regardless of stochasticity.

(A-B) Time courses of the species abundances simulated with ODEs, SSA, or IBM. (C) Snapshots of the IBM simulations. (D-E) A model of intraspecific predator interference explains two classical laboratory experiments that invalidate CEP. (D) In Ayala’s experiment, two Drosophila species coexist with one type of resources within a laboratory bottle (Ayala, 1969). (E) In Park’s experiment, two Tribolium species coexist for two years with one type of food (flour) within a lab (Park, 1954).

Intraspecific interference promotes biodiversity in the presence of stochasticity

Stochasticity is ubiquitous in nature. However, it is prone to jeopardize species coexistence (Xue and Goldenfeld, 2017). Influential mechanisms such as “kill the winner” fail when stochasticity is incorporated (Xue and Goldenfeld, 2017). Consistent with this, we observe that two notable cases of oscillating coexistence (Koch , 1974; Huisman and Weissing, 1999) turn into species extinction when stochasticity is introduced (Appendix-fig. 6A-B), where we simulate the models with stochastic simulation algorithm (SSA) (Gillespie, 2007) and adopt the same parameters as those in the original references (Koch , 1974; Huisman and Weissing, 1999).

Then, we proceed to investigate the impact of stochasticity on our model using SSA (Gillespie, 2007). In the scenario involving chasing pair and intraspecific interference, species may coexist indefinitely in the SSA simulations (Fig. 2A and Appendix-fig. 4C). In fact, the parameter region for species coexistence in this scenario is rather similar between the SSA and ODEs studies (Appendixfig. 6C-D). Similarly, in the scenario involving chasing pair and both inter- and intra-specific interference, all species may coexist indefinitely in company with stochasticity (Appendix-fig. 5D).

To further mimic a real ecosystem, we resort to individual-based modeling (IBM) (Grimm and Railsback, 2013; Vetsigian, 2017), an essentially stochastic simulation method. In the simple case of SC = 2 and SR = 1, we simulate the time evolution of a 2-D squared system in a size of L2 with periodic boundary conditions (see Materials and Methods for details). In the scenario involving chasing pair and intraspecific interference, two consumer species coexist for long with only one type of resources in the IBM simulations (Fig. 2B-C). Together with the SSA simulation studies, it is obvious that intraspecific interference promotes species coexistence along with stochasticity.

Comparison with experimental studies that reject CEP

In practice, two classical studies (Ayala, 1969; Park, 1954) reported that, in their respective laboratory systems, two species of insects coexisted for roughly years or more with only one type of resources. Evidently, these two experiments (Ayala, 1969; Park, 1954) are incompatible with CEP, while factors such as temporal variations, spatial heterogeneity, cross-feeding, etc. are clearly not involved in such systems. As intraspecific fighting is prevalent among insects (Boomsma et al., 2005; Dankert et al., 2009; Chen et al., 2002), we apply the model involving chasing pair and intraspecific interference to simulate the two systems. Overall, our SSA results show good consistency with those of the experiments (Fig. 2D-E, see also Appendix-figs. 6C-D, 7). The fluctuations in experimental time series can be mainly accounted by stochasticity.

A handful of resource species can support an unexpected wide range of consumer species regardless of stochasticity

To resolve the puzzle stated in the paradox of the plankton, we analyze the generic case where SC consumers species compete for SR resource species (with SC > SR) within the scenario involving chasing pair and intraspecific interference. The population dynamics is described by equations combining Eqs. 1, 2, 4. As with the cases above, each consumer species is assigned a unique competitiveness through a distinctive Di (i = 1, —, SC).

Strikingly, a plethora of consumer species may coexist at steady state with only one resource species (SC ≫ SR, SR = 1) in the ODEs simulations, and crucially, the facilitated biodiversity can still be maintained in the SSA simulations. The long-term coexistence behavior are exemplified in Fig. 3 and Appendix-fig. 810, involving simulations with or without stochasticity. The number of consumer species in long-term coexistence can be up to hundreds or more (Fig. 3 and Appendixfig. 8). To mimic the real ecosystems, we further analyze the cases with more than one type of resources, such as systems with SR = 3 (SC ≫ SR). Just like the case of SR = 1 (SC ≫ SR), an extensive range of consumer species may coexist indefinitely regardless of stochasticity (Fig. 3 and Appendix-fig. 1114).

Intuitive understanding: an underlying negative feedback loop

For the case with only one resource species (SR = 1), if the total population size of the resources is much larger than that of the consumers (i.e., ), the functional response kixi/Ct and the steady-state population of each consumer and resource species can be obtained analytically (see Appendix III B-C for details). In fact, the functional response of a consumer species (e.g., Ci) is negatively correlated with its own population size:

where βa’i/d’i. The analytical steady-state solutions are highly consistent with the numerical results (Fig. 1E and Appendix-fig. 3E-F) and can even quantitatively predict the coexistence region of the parameter space (Appendix-fig. 3F).

Intuitively, the mechanisms of how intraspecific interference facilitates species coexistence can be understood from the underlying negative feedback loop. Specifically, for consumer species of higher competitiveness (e.g., Ci) in an ecological community, as the population size of Ci increases during competition, a larger portion of Ci individuals are then engaged in intraspecific interference pairs which are temporarily absent from hunting (see Eq. S59 and Appendix-fig. 15A-B). Consequently, the fraction of Ci individuals within chasing pairs decreases (see Eq. S59 and Appendix-fig. 15A-B) and thus form a self-inhibiting negative feedback loop through the functional response (see Eq. 5 and Appendix-fig. 15C). This negative feedback loop prevents further increases in Ci populations, results in an overall balance among the consumer species, and thus promotes biodiversity (see Appendix III C for details).

The S shape pattern of the rank-abundance curves in a broad range of ecological communities

As mentioned above, a prominent feature of biodiversity is that the species’ rank-abundance curves follow a universal S-shaped pattern in the linear-log plot across a broad spectrum of ecological communities (Fuhrman et al., 2008; Ser-Giacomi et al., 2018; Cody and Smallwood , 1996; Terborgh et al., 1990; Martinez et al., 2023; Clarke et al., 2005; Hubbell, 2001; Devries et al., 1997). Previously, this pattern was mostly explained by the neutral theory (Hubbell, 2001), which requires special parameter settings that all consumer species share identical fitness. To resolve this issue, we apply the model involving chasing pair and intraspecific interference to simulate the ecological communities, where one or three types of resources support a large number of consumer species (SC ≫ SR). In each model system, the mortality rates of consumer species follow a Gaussian distribution where the coefficient of variation was taken round 0.3 (Menon et al., 2003) (see Appendix VII for details). For a broad array of the ecological communities, the rank-abundance curves obtained from the long-term coexisting state of both the ODEs and SSA simulation studies agree quantitatively with those of experiments (Fig. 3C-D, see also Appendix-figs. 814), sharing roughly equal Shannon entropies and mostly being regarded as identical distributions in the Kolmogorov-Smirnov (K-S) statistical test (with a significance threshold of 0.05). Still, there is a noticeable discrepancy between the experimental data and SSA studies in terms of the species’ absolute abundances (e.g., Appendix-fig. 8C): those with experimental abundances less than 10 tend to extinct in the SSA simulations. This is due to the fact that the recorded individuals in an experimental sample are just a tiny portion of that in the real ecological system, whereas the species population size in a natural community is certainly much larger than 10.

Discussion

The conflict between the CEP and biodiversity, exemplified by the paradox of the plankton (Hutchinson, 1961), is a long-standing puzzle in ecology. Although many mechanisms have been proposed to overcome the limit set by CEP (Hutchinson , 1961; Chesson , 2000; Levins , 1979; Levin , 1974; Koch, 1974; Huisman and Weissing, 1999; Lczaran et al., 2002; Goyal and Maslov., 2018; Goldford et al., 2018; Martín et al., 2020; Gupta et al., 2021; Thingstad, 2000; Wang and Liu, 2020; Dalziel et al., 2021; Posfai et al., 2017; Weiner et al., 2019; Xue and Goldenfeld, 2017; Beddington, 1975; DeAngelis et al., 1975; Arditi and Ginzburg, 1989; Kelsic et al., 2015; Grilli et al., 2017; Ratzke et al., 2020), it is still unclear how plankton and many other organisms can flout CEP and maintain biodiversity in quasi-well-mixed natural ecosystems. To address this issue, we investigate a mechanistic model with detailed consideration of pairwise encounters. Using numerical simulations combined with mathematical analysis, we identify that the intraspecific interference among the consumer individuals can promote a wide range of consumer species to coexist indefinitely with only one or a handful of resource species through the underlying negative feedback loop. By applying the above analysis to real ecological systems, our model naturally explains two classical experiments that reject CEP (Ayala, 1969; Park, 1954), and quantitatively illustrates the universal S-shaped pattern of the rank-abundance curves for a broad range of ecological communities (Fuhrman et al., 2008; Ser-Giacomi et al., 2018; Cody and Smallwood, 1996; Terborgh et al., 1990; Martínez et al., 2023; Clarke et al., 2005; Hubbell, 2001; Devries et al., 1997).

Intraspecific interference enables a wide range of consumer species to coexist with only one or a handful of resource species.

(A-B) Representative time courses simulated with ODEs and SSA. (C-D) A model of intraspecific predator interference illustrates the S-shaped pattern of the species’ rank-abundance curves across different ecological communities. The solid icons represent the experimental data (marked with “Exp”) reported in existing studies (Fuhrman et al., 2008; Cody and Smallwood, 1996; Terborgh et al., 1990; Martínez et al., 2023; Clarke et al., 2005; Devries et al., 1997; Hubbell, 2001), where the bird community data were collected longitudinally in 1982 and 2018 (Terborgh et al., 1990; Martínez et al., 2023). The ODEs and SSA results were constructed from timestamp t = 1.0 × 105 in the time series. In the K-S test, the probabilities (p values) that the simulation results and the corresponding experimental data come from the same distributions are: , , , ; , , , ; , . See Appendix VII for simulation details and the Shannon entropies.

In fact, predator interference has been introduced long ago by the B-D model (Beddington, 1975; DeAngelis et al., 1975). However, the functional response of the B-D model involving intraspecific interference can be formally derived from the scenario involving only chasing pair without predator interference (Wang and Liu, 2020; Huisman and Boer , 1997) (see Eqs. S8 and S24). Therefore, it is questionable regarding the validity of applying the B-D model to break CEP. From mechanistic perspective, we resolve these issues and show that B-D model corresponds to a special case of our mechanistic model yet without the escape rate (Appendix-fig. 2, see Appendix II for details).

Our model is broadly applicable to explain biodiversity in many ecosystems. In practice, many more factors are potentially involved, and special attention is required to disentangle confounding factors. In microbial systems, complex interactions are commonly involved (Goyal and Maslov., 2018; Goldford et al., 2018; Hu et al., 2022), and species’ preference for food is shaped by the evolution course and environmental history (Wang et al., 2019). It is still highly challenging to fully explain how organisms evolve and maintain biodiversity in diverse ecosystems.

Methods and Materials

Derivation of the encounter rates with the mean-field approximation

In the model depicted in Fig. 1A, consumers and resources move randomly in space, which can be regarded as Brownian motions. At moment t, a consumer individual of species Ci (i = 1, —, SC) moves at speed with velocity , while a resource individual of species Rl (l = 1, …, SR) moves at speed with velocity . Here and are two invariants, while the directions of and change constantly. The relative velocity between the two individuals is , with a relative speed of . Then, , where represents the angle between and . This system is homogeneous, thus, , where the overline stands for the temporal average. Then, we obtain the average relative speed between the Ci and Rl, individuals: . Likewise, the average relative speed between the Ci and Cj individuals is . Evidently, . Meanwhile, the concentrations of species Ci and Rl, in a squared system with a length of L are and , while those of the freely wandering Ci and Rl, are and .

Then, we use the mean-field approximation to calculate the encounter rates ail and a’ij in the well-mixed system. In particular, we estimate ail by tracking a randomly chosen consumer individual from species Ci and counting its encounter frequency with the freely wandering individuals from resource species Rl, (Appendix-fig. 1). At any moment, the consumer individual may form a chasing pair with a Rl, individual within a radius of (Fig. 1A). Over a time interval of Δt, the number of encounters between the consumer individual and Rl individuals can be estimated by the encounter area and the concentration , which takes the value of (see Appendix-fig. 1). Combined with , for all freely wandering Ci individuals, the number of their encounters with R(F) during interval Δt is . Meanwhile, in the ODEs, this corresponds to . Comparing both terms above, for chasing pair, we have = . Likewise, for interference pair, we obtain In particular, .

Stochastic simulations

To investigate the impact of stochasticity on species coexistence, we use stochastic simulation algorithm (SSA) (Gillespie, 2007) and individual-based modeling (IBM) (Vetsigian, 2017; Grimm and Railsback, 2013) in simulating the stochastic process. In the SSA studies, we follow the standard Gillespie algorithm and simulation procedures (Gillespie, 2007).

In the IBM studies, we consider a 2D squared system in a length of L with periodic boundary conditions. In the case of SC = 2 and SR = 1, consumers of species Ci, (i = 1, 2) move at speed , while the resources move at speed vR. The unit length is Δl = 1, while all individuals move probabilistically. Specifically, when Δt is small so that , Ci individuals jump a unit length with the probability . In practice, we simulate the temporal evolution of the model system following the procedures below.

Initialization

We choose the initial position for each individual randomly from a uniform distribution in the squared space, which round to the nearest integer point in the x-y coordinates.

Moving

We choose the destination of a movement randomly from four directions (x-positive, x-negative, y-positive, y-negative) following a uniform distribution. The consumers and resources jump a unit length with probabilities and , respectively.

Forming pairs

When a Ci individual and a resource individual get close in space within a distance of r(C), they form a chasing pair. Meanwhile, when two consumer individuals Ci and Cj stand within a distance of r(I), they form an interference pair.

Dissociation

We update the system with a small time step Δt so that diΔt, kiΔt, d’ijΔt ≪ 1 (i, j = 1,2). In practice, a random number ζ is sampled from a uniform distribution between 0 and 1, i.e., U(0, 1). If ζ < diΔt or ζ < d’jΔt, then, the chasing pair or interference pair dissociates into two separated individuals. One occupies the original position, while the other individual gets just out of the encounter radius in a uniformly distributed random angle. For a chasing pair, if diΔt < ζ < (di + kit, then, the consumer catch the resource, and the biomass of the resource flows into the consumer populations (updated according to the birth procedure), while the consumer individual occupies the original position. Finally, if ζ > (di + kit or ζ > d’jΔt, the chasing pair or interference pair maintain the current status.

Birth and death

For each species, we use two separate counters with decimal precision to record the contributions of the birth and death processes, which both accumulate in each time step. The incremental integer part of the counter will trigger updates in this run. Specifically, a newborn would join the system following the initialization procedure in a birth action, while an unfortunate target would be randomly chosen from the living population in a death action.

Acknowledgements

We thank Roy Kishony, Eric D. Kelsic, Yang-Yu Liu and Fan Zhong for helpful discussions. This work was supported by National Natural Science Foundation of China (Grant No.12004443), Guangzhou Municipal Innovation Fund (Grant No.202102020284) and the Hundred Talents Program of Sun Yat-sen University.

Author contributions

X.W. conceived and designed the project; J.K., S.Z. and X. W. performed research; J.K. and X. W. analyzed data and wrote the paper.

Data and materials availability

All study data are included in the article and/or appendices.

Appendix I The classical proof of Competitive Exclusion Principle (CEP)

In the 1960s, MacArthur (MacArthur and Levins, 1964) and Levin (Levin, 1970) put forward the classical mathematical proof of CEP. We rephrase their idea in the simple case of SC = 2 and SR = 1, i.e., two consumer species C1 and C2 competing for one resource species R. In practice, this proof can be generalized into higher dimensions with several consumer and resource species. The population dynamics of the system can be described as follows:

Here Ci and R represent the population abundances of consumers and resources, respectively, while the functional forms of fi(R) and g(R, C1, C2) are unspecific. Di stands for the mortality rate of the species Ci. If all consumer species can coexist at steady state, then fi (R)/Di = 1 (i = 1, 2). In a 2-D representation, this requires that three lines y = fi (R)/Di (i = 1, 2) and y = 1 share a common point, which is commonly impossible unless the model parameters satisfy special constraint (sets of Lebesgue measure zero). In a 3-D representation, the two planes corresponding to fi (R)/Di = 1 (i = 1, 2) are parallel, and hence do not share a common point (see Ref. (Wang and Liu, 2020) for details).

Appendix II Comparison of the functional response with Beddington-DeAngelis (B-D) model

A B-D model

In 1975, Beddington proposed a mathematical model (Beddington, 1975) to describe the influence of predator interference on the functional response with hand-waving derivations. In the same year, DeAngelis and his colleagues considered a related question and put forward a similar model (DeAngelis et al., 1975). Essentially, both models are phenomenological, and they were called B-D model in the subsequent studies. In practice, the B-D model can be extended into scenarios involving different types of pairwise encounters with Beddington’s modelling method. In this section, we systematically compare the functional response in B-D model with that of our mechanistic model in all the relevant scenarios.

Recalling Beddington’s analysis, the model (Beddington, 1975) consists of one consumer species C and one resource species R (SC = 1, SR = 1). In a well-mixed system, an individual consumer meets a resource with rate a, while encounters another consumer with rate a’. There are two other phenomenological parameters in this model, namely, the handling time th and the wasting time tw. Both can be determined by specifying the scenario and using statistical physics modeling analysis. In fact, Beddington analyzed the searching efficiency ΞB-D rather than the functional response B-D, yet both can be reciprocally derived with ΞB-DB-D/R. Here R stands for the population abundance of the resources, and the specific form of ΞB-D is (Beddington, 1975):

where C’ = C − 1, and C stands for the population abundance of the consumes. Generally, C ≫ 1, and thus C’ ≈ C.

B Scenario involving only chasing pair

Here we consider the scenario involving only chasing pair for the simple case with one consumer species C and one resource species R (SC = 1, SR = 1). When an individual consumer is chasing a resource, they form a chasing pair:

where the superscript “(F)” stands for populations that are freely wandering, and “(+)” signifies gaining biomass (we count C(F) (+) as C(F). C(P)R(P) represents chasing pair (where “(P)” signifies pair), denoted as x. a, d and k stand for encounter rate, escape rate and capture rate, respectively. Hence, the total number of consumers and resources are C ≡ C(F) + x and RR(F) + x. Then, the population dynamics of the system follows:

Here the functional form of g(R, x, C) is unspecific, while D and w represent the mortality rate of the consumer species and biomass conversion ratio (Wang and Liu, 2020), respectively. Since consumption process is generically much faster than the birth/death process, in deriving the functional response, the consumption process is supposed to be in fast equilibrium (i.e., = 0). Then, we can solve for x with:

where , and then,

By definition, the functional response and search efficiency are:

Hence, we obtain the functional response and search efficiency in this chasing-pair scenario:

Since , using first order approximations in Eq. S7, we obtain . Then the functional response and search efficiency are:

Evidently, there is no predator interference within the chasing-pair scenario, yet the functional response form is identical to the B-D model involving intraspecific interference (see Eq. S2). Meanwhile, using first order approximations in the denominator of Eq. S5, we have . Hence,

In the case that RC, then RC > x = RR(F). By applying RR(F) in Eq. S3, we obtain . Then,

To compare these functional responses with that of the B-D model, we determine the parameters th and tw in the B-D model by calculating their ensemble average values in a stochastic framework. Using the properties of waiting time distribution in the Poisson process, we obtain and (in the chasing-pair scenario, a’ = 0). By substituting these calculations into Eq. S2, we have

In the special case with d = 0 and RC, the B-D model is consistent with our mechanistic model: ΞB-D(R, C) = ΞCP(R, C)(4). Outside the special region, however, the discrepancy can be considerably large (see Appendix-fig. 2A-B for the comparison).

C Scenario involving chasing pair and intraspecific interference

Here we consider the scenario with additional involvement of intraspecific interference in the simple case of SC = 1 and SR = 1

Here C(P)C(P) stands for the intraspecific predator interference pair, denoted as y; a’ and d’ represent the encounter rate and separation rate of the interference pair, respectively. Then, the total population of consumers and resources are CC(F) + x + 2y and RR(F) + x. Hence the population dynamics of the consumers and resources can be described as follows:

The consumption process and interference process are supposed to be in fast equilibrium (i.e., = 0, = 0), then we can solve for x with:

where ϕ0 = —CR2, ϕ1 = 2CR + KR + R2, ϕ2 = 2βK2KC — 2R, with β = a’/d’. The discriminant of Eq. S13 (denoted as ∧) is

with ψ = ϕ1 — (ϕ2)2/3 and φ = ϕ0ϕ1ϕ2/3 + 2(ϕ2)3/27. When ∧ < 0, there are one real solution x(1) and two complex solutions x(2), x(3), which are

where (i stands for the imaginary unit), , and . On the other hand, when ∧ > 0, there are three real solutions x(1), x(2), and x(3), which are

where ψ’ = (-4ψ/3)1/2 and φ’ = arccos(-(-ψ/3)-3/2φ/2)/3. Note that x ∈ [0, min(R, C)], then we obtain the exact feasible solution of x (denoted as xext), and hence the functional response and search efficiency are

In the case of RC, then RR(F) = x < C CR, and thus R(F)R. Still, the consumption process is supposed to be in fast equilibrium (i.e., = 0, = 0), and then we obtain

Consequently,

When or 8βC/(1 + R/K)2 ≫ 1, using first order approximations in the denominator of Eq. S18, we have

and then,

In the case that 8βC/(1 + R/K)2 ≫ 1, using first order approximations in Eq. S18, we obtain

and thus,

Meanwhile, the B-D model only fits to the cases with d = 0. By calculating the average values of th and tw in the stochastic framework, we have , . Thus, we obtain the searching efficiency and functional response in the B-D model:

Overall, the searching efficiency (and the functional response) of the B-D model is quite different from either the rigorous form Ξintra(R, C)(1), the quasi rigorous form Ξintra(R, C)(2), or the more simplified forms Ξintra(R, C)(3) and Ξintra(R, C)(4) (Appendix-fig. 2C-D). Still, there is a region where the discrepancies can be small, namely d ≈ 0 and RC (Appendix-fig. 2C-D). Intuitively, when and d = 0, then Consequently, if , then . In this case, the difference between and Ξintra(R, C)(3) is small.

In fact, the above analysis also applies to cases with more than one types of consumer species (i.e., for cases with SC > 1).

D Scenario involving chasing pair and interspecific interference

Next, we consider the scenario involving chasing pair and interspecific interference in the case of SC = 2 and SR = 1:

Here stands for the interspecific interference pair, denoted as z; a12 and d12 represent the encounter rate and separation rate of the interference pair, respectively. Then, the total population of consumers and resources are and . The population dynamics of the consumers and resources follows:

where the functional form of g(R, x1, x2, C1, C2) is unspecific, while Di and wi represents the mortality rates of the two consumers species and biomass conversion ratios. Still, the consumption/interference process is supposed to be in fast equilibrium, i.e., i = 0, ż = 0. In the case that RC1 + C2 > x1 + x2, by applying R(F)R, we obtain

Then, the searching efficiencies and functional responses are:

Since , by applying first order approximation to the denominator of Eq. S26, we obtain:

and the searching efficiencies and functional responses are

Likewise, the B-D model only fits to cases with d = 0. By calculating the average values in a stochastic framework, we obtain , (i = 1, 2). Then, we obtain the searching efficiencies in the B-D model:

Consequently, the functional responses in the B-D model are:

Evidently, the searching efficiencies in the B-D model are overall different from either the quasi rigorous form Ξi(R, C1, C2)1, or the simplified form Ξi(R, C1, C2)2 (Appendix-fig. 2E-F). Still, the discrepancy can be small when d ≈ 0 and RC (Appendix-fig. 2E-F). Intuitively, when , we have

Thus, if (i = 1, 2), then . In this case, the difference between and is small.

Appendix III Scenario involving chasing pair and intraspecific interference

A Two consumers species competing for one resource species

We consider the scenario involving chasing pair and intraspecific interference in the simple case of SC = 2 and SR = 1:

Here, the variables and parameters are just extended from the case of SC = 1 and SR = 1 (see Appendix II. C). The total number of consumers and resources are and . Hence, the population dynamics of the consumers and resources can be described as follows:

The functional form of g(R, x1, x2, C1, C2) is unspecified. For simplicity, we limit our analysis to abiotic resources, while all results generically apply to biotic resources. Besides, we define Ki ≡ (di + ki)/ai, αiDi/(wiki), and βi = a’i/d’i (i = 1,2). At steady state, from i = 0, i = 0, we have

Note that , and . Then,

By substituting Eq. S35a into Eq. S35b, we have

Then, we can present with C1, C2 and R (i = 1, 2). By further combining with Eqs. S34, S35a and S36a, we express R(F), xi, and yi using C1, C2 and R. In particular, for xi, we have:

If all species coexist, then the steady-state equations of Ċi = 0 (i = 1, 2) and = 0 are:

where G(R,C1, C2) ≡ g(R,u1(R,C1, C2),u2(R,C1, C2),C1, C2), and . In practice, Eq. S38 corresponds to three unparallel surfaces, which share a common point (Fig. 1H and Appendix-fig. 3D). Importantly, the fixed point can be stable, and hence two consumer species may coexist at constant population densities.

1 Stability analysis of the fixed-point solution

We use linear stability analysis to study the local stability of the fixed point. Specifically, for an arbitrary fixed point E(x1,x2,y1,y2,C1,C2,R), only when all the eigenvalues (defined as λi, i = 1, …, 7) of the Jacobian matrix at point E own negative real parts would the point be locally stable.

To investigate whether there exists a non-zero measure parameter region for species coexistence, we set Di (i = 1, 2) to be the only parameter that varies with species C1 and C2, and then Δ ≡ (D1D2)/D2 reflects the completive difference between the two consumer species. As shown in Appendix-fig. 4B, the region below the blue surface and above the red surface corresponds to stable coexistence. Thus, there exists a non-zero measure parameter region to promote species coexistence, which breaks CEP.

2 Analytical solutions of the species abundances at steady state

At steady state, since i = i = Ċi = 0 (i =1, 2), then,

Meanwhile, , and Ci, R > 0 (i = 1, 2). Then, we have

If the resource species owns a much larger population abundance than the consumers (i.e., RC1+C2), then R x1+x2, and R(F)R. Thus,

By further assuming that the population dynamics of the resources follow identical construction rule as the MacArthur’s consumer-resource model (MacArthur, 1970), we have

Since = 0, then

where and .

Eqs. S41, S43 are the analytical solutions of species abundances at steady state when RC1 + C2. As shown in Fig. 1E, the analytical solutions agree well with the numerical results (the exact solutions). To conduct a systematic comparison for different model parameters, we assign Di (i = 1, 2) to be the only parameter varying with species C1 and C2 (D1 > D2), and define Δ ≡ (D1D2)/D2 as the competitive difference between the two consumer species. The comparison between analytical solutions and numerical results is shown in Appendix-fig. 3E. Clearly, they are close to each other, exhibiting very good consistency.

Furthermore, we test if the parameter region for species coexistence is predictable using the analytical solutions. Since Di (i = 1, 2) is the only parameter that varies with the two-consumer species, the supremum of the tolerated competitive difference for species coexistence (defined as ) corresponds to the steady-state solutions that satisfy R,C2 > 0 and C1 = 0+, where 0+ stands for the infinitesimal positive number. To calculate the analytical solutions at the upper surface of the coexistence region, where and C1 = 0+, we further combine Eq. S41 and then obtain (note that R > 0)

Meanwhile,

Combining Eqs. S43S45, we have

where . When RC1 + C2, the comparison of obtained from analytical solutions with that from numerical results (the exact solutions) are shown in Appendix-fig. 3F, which overall exhibits good consistency.

B SC consumers species competing for SR resources species

Here we consider the scenario involving chasing pair and intraspecific interference for the generic case with SC types of consumers and SR types of resources. Then, the population dynamics of the system can be described as follows:

Note that Eq. S47 is identical with Eqs. 12, and we use the same variables and parameters as that in the main text. Then, the populations of the consumers and resources are and . For convenience, we define and .

1 Analytical solutions of species abundances at steady state

At steady state, from , i =0, and , we have,

Meanwhile , and note that Ci > 0, thus

Combined with Eq. S49, and then

We further assume that the specific function of satisfies Eq. 4, i.e.,

By combining Eqs. S48, S49 and S51, we have

If the population abundance of each resource species is much more than the total population of all consumers (i.e., , then and . Thus,

with l = 1, … , SR. Eq. S53 is a set of second-order algebraic differential equations, which is clearly solvable.

In fact, when SR = 1, SC ≥ 1, and , we can explicitly present the analytical solution of the steady-state species abundances. To simplify the notations, we omit the “l” in the sub-/super-scripts since SR = 1.Then, we have

Here and .

C Intuitive understanding: an underlying negative feedback loop

Intuitively, how can intraspecific predator interference promote biodiversity? Here we solve this question by considering the case that SC types of consumers compete for one resource species. The population dynamics of the system are described in Eqs. S47 and S51 with SR = 1. To simplify the notations, we omit the “l” in the subscript since SR = 1. The consumption process and interference process are supposed to be in fast equilibrium (i.e., i = 0, i = 0). Then, we have a set of equations to solve for xi and yi given the population size of each species:

In the first three sub-equations of Eq. S55, by getting rids of , we have,

Then, by regarding R(F) as a temporary parameter, we solve for xi and yi:

If the total population size of the resources is much larger than that of consumers (i.e., ), then and R(F) ≈ R, and thus we get the analytical expressions of xi and yi

Note that the fraction of Ci individuals engaged in chasing pairs is xi/Ci, while that for individuals trapped in intraspecific interference pairs is yi/Ci. With Eq. S58, it is straightforward to obtain these fractions:

where both xi/Ci and yi/Ci are bivariate functions of R and Ci. From Eq. S59, it is clear that for a given population size of the resource species, yi/Ci is a monotonously increasing function of Ci, while xi/Ci is a monotonously decreasing function of Ci. In Appendix-fig. 15A-B, we see that the analytical results are highly consistence with the exact numerical solutions. By definition, the functional response of Ci species is kixi/Ci, and thus,

Evidently, the function response of Ci species is negatively correlated with the population size of itself, which effectively constitutes a self-inhibiting negative feedback loop (Appendix-fig. 15C).

Then, we have a simple intuitive understanding of species coexistence through the mechanism of intraspecific interference. In an ecological community, consumer species that of higher/lower competitiveness tend to increase/decrease their population size in the competition process. Without intraspecific interference, the increasing/decreasing trend would continue until the system obeys CEP. In the scenario involving intraspecific interference, however, for species of higher competitiveness (e.g., Ci), with the increase of Ci’s population size, a larger portion of Ci individuals are then engaged in intraspecific interference pair which are temporarily absent from hunting (Appendix-fig. 15A-B). Consequently, the functional response of Ci drops, which prevents further increase of Ci‘s population size, results in an overall balance among the consumer species, and thus promotes species coexistence.

Appendix IV Scenario involving chasing pair and interspecific interference

Here we consider the scenario involving chasing pair and interspecific interference in the case of SC = 2 and SR = 1 , with all settings follow that depicted in Appendix II. D. Then, , and the population dynamics follows (identical with Eq. S25):

Here the functional form of g(R, x1, x2, C1, C2) is unspecified. For convenience, we define Ki ≡ (di + ki)/ai, αiDi/(wiki)(i = 1, 2), and γ = a12/d12. At steady state, from i = 0(i = 1,2) and ż = 0, we have

Note that and RR(F) + x1 + x2, then,

Then, we can express , and R(F) with C1, C2 and R. Combined with Eq. S62, xi and z can also be expressed using C1, C2 and R. In particular, for xi, we have

If all species coexist, by defining , then, the steady-state equations of Ċi = 0 (i = 1,2) and i = 0 are:

where G’(R, C1,C2) ≡ g(R, u1(R, C1,C2), u2(R, C1,C2),C1,C2).

Here, Eq. S65 corresponds to three unparallel surfaces and share a common point (Fig. 1G and Appendix-fig 3A). However, all the fixed points are unstable (Appendix-fig. 3C), and hence the consumer species cannot stably coexist at steady state (Fig. 1D).

A Analytical results of the fixed-point solution

We proceed to investigate the unstable fixed point where R, C1, C2 > 0. From i = 0 (i = 1,2), ż = 0, Ċi = 0, and note that , we have

Since Ci > 0, then

If RC1 + C2, then Rx1 + x2 and R(F)R, we have

Still, we assume that the population dynamics of the resource species follows Eq. S42. At the fixed point, = 0. We have

Combined with Eq. S68, we can solve for R

where and .

Eqs. S68, S70 are the analytical solutions of the fixed point when RC1 + C2. As shown in Appendix-fig. 3B, the analytical predictions agree well with the numerical results (the exact solutions).

Appendix V Scenario involving chasing pair and both intra- and inter-specific interference

Here we consider the scenario involving chasing pair and both intra- and inter-specific interference in the simple case of SC = 2 and SR = 1:

We adopt the same notations as that depicted in Appendix III. A and Appendix IV. Then, and R = R(F) + x1 + x2, and the population dynamics of the system can be described as follows:

Here, the functional form of g(R, x1, x2, C1, C2) follows Eq. S42. For convenience, we define Ki ≡ (di + ki)/ai, αiDi/(wiki), βia’i/d’i, and γa12/d12, (i = 1,2). At steady state, from i = 0, i = 0, ż = 0, and Ċi = 0, (i = 1, 2), we have

Combined with , and since Ci > 0(i = 1, 2), then,

A Analytical solutions of species abundances at steady state

If RC1 + C2, then Rx1 + x2 and thus R(F)R. Combined with Eq. S73, we obtain

Using = 0 and R > 0, we have

where and . Eqs. S74S75 are the analytical solutions of the species abundances at steady state when RC1 + C2. As shown in Appendix-fig. 5E, the analytical calculations agree well with the numerical results (the exact solutions).

B Stability analysis of the coexisting state

In the scenario involving chasing pair and both intra- and inter-specific interference, the behavior of species coexistence is similar to that without interspecific interference. Evidently, the influence of interspecific interference would be negligible if d12 is extremely large, and vice versa for intraspecific interference if both d1 and d2 are tremendous. In the deterministic framework, the two-consumer species may coexist at constant population densities (Appendix-fig. 5B), and the fixed points are globally attracting (Appendix-fig. 5C). Furthermore, there is a non-zero measure of parameter set where both consumer species can coexist at steady state with only one type of resources (Appendix-fig. 5A). In the stochastic framework, just as the scenario involving chasing pair and intraspecific interference, the coexistence state can be maintained along with stochasticity (Appendix-fig. 5D).

Appendix VI Dimensional analysis for the scenario involving chasing pair and both intra- and inter-specific interference

The population dynamics of the system involving chasing pair and both intra- and inter-specific interference are shown in Eqs. 14:

with l = 1, … , SR; i,j = 1, … , SC, and i ≠ = j. Here and represent the population abundances of the consumers and resources in the system. In fact, there are already several dimensionless variables and parameter in Eq. S76, namely xil, yi, zij, , , Ci, Rl, wil, . To make all terms dimensionless, we define , where and is a reducible dimensionless parameter which is freely to take any positive values. Besides, we define dimensionless parameters , , , , , , , and . By substituting all the dimensionless terms into Eq. S76, we have

For convenience, we omit the notation “a” and use dimensionless variables and parameters in the simulation studies unless otherwise specified.

Appendix VII Simulation details of the main text figures

In Fig. 1C, F: ai = 0.1, di = 0.5, wi = 0.1, ki = 0.1 (i = 1, 2); D1 = 0.002, D2 = 0.001, K0 = 5, Ra = 0.05. In Fig. 1D, G: ai = 0.02, a’ij = 0.021, di = 0.5, d’ij = 0.01, wi = 0.08, ki = 0.03, i, j = 1,2,ij, D2 = 0.001, D1 = 0.0011, K0 = 20, Ra = 0.01. In Fig. 1E, H: ai = 0.5, a’i = 0.525, di = 0.5, d’i = 0.5, wi = 0.2, ki = 0.4 (i = 1, 2), D1 = 0.022, D2 = 0.020, K0 = 10, Ra = 0.1. Fig. 1C, F were calculated or simulated from Eqs. 1, 4. Fig. 1D, G were calculated or simulated from Eqs. 1, 3, 4. Fig. 1E, H were calculated or simulated from Eqs. 1, 2, 4. The analytical solutions in Fig. 1E were calculated from Eqs. S41 and S43.

In Fig. 2A: ai = 0.02, a’i = 0.025, di = 0.7, d’i = 0.7, wi = 0.4, ki = 0.05 (i = 1, 2); D1 = 0.0160, D2 = 0.0171, K0 = 2000, Ra = 5.5. In Fig. 2B-C: L = 100, r(C) = 5, r(I) = 5, = 1, vR = 0.1, ai = 0.2010, a’i = 0.2828, d’i = 0.8, di = 0.7, wi = 0.33, ki = 0.2 (i = 1, 2); D1 = 0.0605, D2 = 0.0600, K0 = 1000, Ra = 100. In Fig. 2D: ai = 0.3, a’i = 0.33, wi = 0.018, ki = 4.8, d’i = 5, di = 5.5 (i = 1, 2); D2 = 0.010, Ra = 35, K0 = 10000, D1 = 0.011. In Fig. 2E: wi = 0.02, ki = 4.5, d’i = 4, di = 4.5 (i = 1, 2); D2 = 0.010, Ra = 35, K0 = 10000, ai = 0.2, ai = 0.24 (i = 1, 2); D1 = 0.0120. In Fig. 2D-E: τ = 0.4Day (see Appendix VI). Fig. 2A-E were simulated from Eqs. 1, 2, 4. See Appendix-fig. 7C, E for the long-term time series of all species in Fig. 2D-E, respectively.

Model settings in Fig. 3A-B, D (plankton): ail = 0.1, a’i = 0.125, dil = 0.5, d’i = 0.2, wil = 0.3, kil = 0.2, = 8 × 104, = 5 × 104, = 3 × 104, = 280, = 200, = 150, Di = 0.03 × N(1,0.25) (i = 1, …, SC, l = 1, …, SR), SC = 140 and SR = 3. Model settings in Fig. 3C (bird): ai = 0.1, a’i = 0.125, di = 0.5, d’i = 0.5, wi = 0.3, ki = 0.2, Di = 0.02 × N(1, 0.28) (i = 1, …, SC); Ra = 110, K0 = 105, SC = 250 and SR = 1. Model settings in Fig. 3C (fish): ai = 0.1, a’i = 0.14, di = 0.5, d’i = 0.5, wi = 0.2, ki = 0.1, Di = 0.015 × N(1, 0.32) (i = 1, …, 45); Ra = 550, K0 = 106, SC = 45 and SR = 1. Model settings in Fig. 3C (butterfly): ai = 0.1, a’i = 0.125, di = 0.5, d’i = 0.3, wi = 0.3, ki = 0.2, Di = 0.034 × N(1, 0.35) (i = 1, …, SC); Ra = 300, K0 = 105, SC = 150 and SR = 1. Model settings in Fig. 3D (bat): ai = 0.1, a’i = 0.125, di = 0.5, d’i = 0.5, wi = 0.2, ki = 0.1, Di = 0.013 × N(1, 0.34) (i = 1, …, SC); Ra = 250, K0 = 106, SC = 40 and SR = 1. Model settings in Fig. 3D (lizard): ai = 0.1, a’i = 0.125, di = 0.5, d’i = 0.5, wi = 0.2, ki = 0.1, Di = 0.014 × N(1, 0.34) (i = 1, …, SC); Ra = 250, K0 = 106, SC = 55 and SR = 1. In Fig. 3A-D, the mortality rate Di (i = 1, …, SC) is the only parameter that varies with the consumer species, which was randomly sampled from a Gaussian distribution N(μ, σ), where μ and σ are the mean and standard deviation of the distribution. The coefficient of variation of the mortality rates (i.e., σ/μ) was chosen to be around 0.3, or more precisely, the best-fit in the range of 0.15–0.43. This range was estimated from experimental results (Menon et al., 2003) using the two-sigma rule. These settings for the mortality rates also apply to those in Appendix-figs. 814. Fig. 3A-D were simulated from Eqs. 1, 2, 4. See Appendix-figs. 14C, 10K, C, D, H, I, J, Fig. 3A, Fig. 3B for the time series of Fig. 3C ), 3C (), 3C (), 3D (), 3D (), 3D (), 3D (), 3D () and 3D (), respectively. The Shannon entropies of the experimental data and simulation results for each ecological community are: = 5.67(6.79), = 6.63(6.79), = 4.78(4.12), = 3.78(3.40); = 3.00(2.95, 2.84), = 4.05(3.57, 3.50); = 4.68(6.43, 6.48). Here the Shannon entropy , where Pi is the probability that a consumer individual belongs to species Ci.

Estimation of the encounter rates with the mean-field approximations.

To calculate ail in the chasing pair, we suppose that all individuals of species Rl stand still while a Ci individual moves at the speed of (the relative speed). Over a time interval of Δt, the length of zigzag trajectory of the Ci individual is approximately , while the encounter area (marked with dashed lines) is estimated to be . Then, we can estimate the encounter rate ail using the encounter area and concentrations of the species (see Materials and Methods for details). Similarly, we can estimate a’ij in the interference pair.

Functional response in scenarios involving different types of pairwise encounter.

(A-B) In the scenario involving only chasing pair, the red surface corresponds to the B-D model (calculated from Eq.S10), while the green surface represents the exact solutions to our mechanistic model (calculated from Eq. S7). The magenta (calculated from Eq. S8) and blue (calculated from Eq. S9) surfaces represent the approximate solutions to our model. (C-D) In the scenario involving chasing pair and intraspecific interference, the red surface corresponds to the B-D model (calculated from Eq.S24), while the green surface represents the exact solutions to our mechanistic model (calculated from Eq. S17). The blue surface (calculated from Eq. S19) and the magenta surface (calculated from Eq. S21) represent the quasi-rigorous and the approximate solutions to our model, respectively. (E-F) In the scenario involving chasing pair and interspecific interference, the red surface corresponds to the B-D model (calculated from Eq. S31), while the green surface represents the quasi-rigorous solutions to our mechanistic model (calculated from Eq. S27). The blue surface (calculated from Eq. S29) represents the approximate solutions to our model. In (A-B): k = 0.1, a = 0.25. In (A): d = 0. (see Appendix II B). In (C-D): a = 0.1, k = 0.1, d’ = 0.1, a’ = 0.12. In (C): d = 0 (see Appendix II C). In (E-F): a1 = a2 = 0.1, k1 = k2 = 0.1, d12 = 0.1, a12 = 0.12. In (E): di = 0 (i = 1,2) (see Appendix II D).

Fixed point solutions in scenarios involving chasing pair and different types of predator interference.

Here, SC = 2 and SR = 1. Di (i = 1, 2) is the only parameter varying with the consumer species (with D1 > D2), and Δ = (D1D2)/D2 represents the competitive difference between the two species. (A-C) Scenario involving chasing pair and interspecific interference. (D-F) Scenario involving chasing pair and intraspecific interference. (A, D) Positive solutions to the steady-state equations: = 0 (orange surface), Ċ1 = 0 (blue surface), Ċ2 = 0 (green surface). The intersection point marked by black/red dots is an unstable/stable fixed point. (B, E) Comparisons between the numerical results and analytical solutions of the species abundances at fixed points. Color bars are analytical solutions while hollow bars are numerical results. The analytical solutions in (B) and (E) (marked with superscript “(A)”) were calculated from Eqs. S68, S70 and Eqs. S41, S43, respectively. (C) In this scenario, there is no parameter region for stable coexistence. The region below the red surface and above Δ = 0 represents unstable fixed points. (F) Comparisons between the numerical results and analytical solutions of the coexistence region. Here represents the maximum competitive difference tolerated for species coexistence. The red and cyan surfaces represent the analytical solutions (calculated from Eq. S46) and numerical results, respectively. The numerical results in (A-C) and (D-F) were calculated from Eqs. S42, S61 and Eqs. S33, S42, respectively. In (A): a1 = a2 = 0.05, d1 = d2 = 0.05, K0 = 20, a12 = 0.06, d12 = 0.01, k1 = k2 = 0.02, w1 = w2 = 0.08, D1 = 0.0011, D2 = 0.001, Ra = 0.01. In (B): a1 = a2 = 0.04, d1 = d2 = 0.2, K0 = 10, a12 = 0.048, d12 = 0.001, k1 = k2 = 0.1, w1 = w2 = 0.3, D2 = 0.0008, Ra = 0.2. In (C): a1 = a2 = 0.1, K0 = 100, k1 = k2 = 0.1, w1 = w2 = 0.1, D2 = 0.001, a12 = 0.12, Ra = 0.05. In (D): a1 = a2 = 0.5, a1 = a2 = 0.625, d1 = d2 = 0.5, d1 = d2 = 0.5, k1 = k2 = 0.4, D2 = 0.02, w1 = w2 = 0.5, K0 = 10, D1 = 1.2D2, Ra = 0.1. In (E): a1 = a2 = 0.1, a1 = a2 = 0.12, k1 = k2 = 0.12, w1 = w2 = 0.3, D2 = 0.02, K0 = 100, Ra = 0.8, d1 = d2 = 0.5, d1 = d2 = 0.05. In (F): a1 = a2 = 0.5, k1 = k2 = 0.2, w1 = w2 = 0.2, D2 = 0.008, K0 = 60, Ra = 0.8, d1 = d2 = 0.8.

Intraspecific predator interference facilitates species coexistence regardless of stochasticity.

Here we consider the case of SC = 2, SR = 1. (A) A representative trajectory of species coexistence in the phase space simulated with ODEs. The fixed point (shown in red) is stable and globally attractive. (B) A 3D phase diagram in the ODEs studies. Di (i = 1, 2) is the only parameter that varies with the two consumer species, and Δ ≡ (D1D2)/D2 measures the competitive difference between the two species. The parameter region below the blue surface yet above the red surface represents stable coexistence. The region below the red surface and above Δ = 0 represents unstable fixed points (an empty set). (C) Time courses of the species abundances simulated with ODEs or SSA. (D) Representative trajectories of species coexistence in the phase space simulated with SSA. The coexistence state is stable and globally attractive (see (C) for time courses, SSA results). (A-D) were calculated or simulated from Eqs. 1, 2, 4. In (A): ai = 0.1, a’i = 0.125, di = 0.1, d’i = 0.05, wi = 0.1, ki = 0.1 (i = 1, 2); D1 = 0.0035, D2 = 0.0038, K0 = 100, Ra = 0.3. In (B): ai = 0.1, di = 0.1, wi = 0.1, ki = 0.1 (i = 1, 2); K0 = 100, D2 = 0.001, Δ = (D1D2)/D2, K0 = 100, Ra = 0.1. In (C-D): ai = 0.02, a’i = 0.025, di = 0.7, d’i = 0.7, wi = 0.4, ki = 0.05 (i = 1, 2); D1 = 0.0160, D2 = 0.0171, K0 = 2000, Ra = 5.5.

Outcomes of two consumers species competing for one resource species involving chasing pair and intra- and inter-specific interference.

Here, Di (i = 1, 2) is the only parameter that varies with the consumer species (with D1 > D2), and Δ = (D1D2)/D2 measures the competitive difference between the two species. (A-C, E) ODEs results. (D) ODEs and SSA results. (A) A 3D phase diagram. The parameter region below the blue surface yet above the red surface represents stable coexistence, while that below the red surface and above Δ = 0 represents unstable fixed points (an empty set). (B) Time courses of the species abundances. Two consumer species may coexist with one type of resources at steady state. (C) Representative tra jectories of species coexistence in the phase space. The fixed point (shown in red) is stable and globally attractive. (D) Consumer species may coexist indefinitely with the resources regardless of stochasticity. (E) Comparisons between numerical results and analytical solutions of the species abundances at fixed points. Color bars are analytical solutions while hollow bars are numerical results. The analytical solutions (marked with superscript “(A)”) were calculated from Eqs. S74S75. The numerical results in (A-E) were calculated or simulated from Eqs. 14. In (A-C): ai = a2 = 0.1, a1 = a2 = 0.12, ki = k2 = 0.1, wi = w2 = 0.1, D2 = 0.004, K0 = 100, a12 = 0.12, Ra = 0.8, d12 = 0.5. In (D): a1 = a2 = 0.1, a1 = a2 = 0.14, k1 = k2 = 0.12, w1 = w2 = 0.15, D1 = 0.0125, D2 = 0.012, K0 = 300, Ra = 5.5, d1 = d2 = 0.3, d1 = d2 = 0.5, a12 = 0.14, d12 = 5. In (E): a1 = a2 = 0.05, a1 = a2 = 0.06, k1 = k2 = 0.1, w1 = w2 = 0.2, D2 = 0.008, K0 = 100, Ra = 0.8, d1 = d2 = 0.5, d1 = d2 = 0.15, a12 = 0.06, d12 = 0.0005.

The influence of stochasticity on species coexistence.

(A-B) Stochasticity jeopardizes species coexistence. Koch’s model (Koch, 1974) and Huisman-Weissing model (Huisman and Weissing, 1999) were simulated with SSA using the same parameter settings as their deterministic model. Nevertheless, both cases of oscillating coexistence are vulnerable to stochasticity. See Ref. (Koch, 1974) and (Huisman and Weissing, 1999) for the parameters. (C-D) Phase diagrams in the scenario involving chasing pair and intraspecific interference. Here, SC = 2 and SR = 1. Di (i = 1, 2) is the only parameter varying with the consumer species (with D1 > D2), and Δ = (D1D2)/D2 represents the competitive difference between the two species. (C) The ODEs results. (D) The SSA results (with the same parameter region as (C)). The species’ coexisting fraction in each pixel was calculated from 16 random repeats. (C) and (D) were calculated from Eqs. 1, 2, 4. In (C-D): ai = 0.1, a’i = 0.125, di = 0.5, wi = 0.1, ki = 0.1 (i = 1, 2); K0 = 100, Ra = 5, D2 = 0.0014.

A model of intraspecific predator interference explains two classical experiments that invalidate CEP.

(A-B) The solid icons represent the experimental data, which are connected by the dotted lines for the sake of visibility. The solid lines stand for the SSA simulation results. In all cases, two consumer species coexist for long with only one type of resources. (A) In Ayala’s experiment (Ayala, 1969), two Drosophila species (consumers), D. serrata and D. pseudoobscura, compete for the same type of abiotic resources within a laboratory bottle. (B) In Park’s experiment (Park, 1954), two Tribolium species, T. confusum and T. castaneum, compete for the same food (flour). (C-F) Time courses of the species abundances in the scenario involving chasing pair and intraspecific interference. The time series in (C-F) correspond to the long-term version of that shown in Fig. 2D, Appendix-fig. 7A, Fig. 2E, Appendix-fig. 7B, respectively. (A-F) were simulated from Eqs. 1, 2, 4. In (A): ai = 0.3, a’i = 0.33, wi = 0.018, ki = 4.8, d’i = 5, di = 5.5 (i = 1, 2); D2 = 0.010, Ra = 35, K0 = 10000, D1 = 0.0132. In (B): wi = 0.02, ki = 4.5, d’i = 4, di = 4.5 (i = 1, 2); D2 = 0.010, Ra = 35, K0 = 10000, ai = 0.3, a’i = 0.36 (i = 1, 2); D1 = 0.0122. In(A-B): τ = 0.4Day (see Appendix VI).

A model of intraspecific interference semi-quantitatively illustrates the rank-abundance curve of a plankton community (SCSR).

(A-B) Intraspecific interference enables a wide range of consumers species to coexist with one type of resources. (A) Time courses of the species abundances simulated with ODEs. (B) Time series of the species abundances simulated with SSA (with theh same as parameter settings as (A)). (C) The rank-abundance curve of a plankton community. The solid dots represent the experimental data (marked with “Exp”) reported in a recent study (Ser-Giacomi et al., 2018) (TARA_139.SUR.180.2000.DNA), while the hollow dots and those with “+” center are the ODEs and SSA results constructed from timestamp t =5.0 × 105 in the time series (see (A) and (B)), respectively. The Shannon entropies of the experimental data and simulation results for the plankton community are: = 2.85(2.18, 2.00). In the model settings, SC = 200 and SR = 1. Di (i = 1, …, SC) is the only parameter that varies with the consumer species, which was randomly drawn from a Gaussian distribution N(μ, σ). Here, μ and σ are the mean and standard deviation of the distribution. The numerical results in (A-C) were simulated from Eqs. 1,2,4. In (A-C): ai = 0.1, a’i = 0.125, d’i = 0.5, di = 0.2, wi = 0.2, ki = 0.1, K0 = 105 , Di = N(1, 0.38) × 0.008 (i = 1, …, 200), Ra = 150.

A model of intraspecific interference illustrates the rank-abundance curves across different ecological communities (SCSR).

The solid dots represent the experimental data (marked with “Exp”) reported in existing studies (Hubbell, 2001), (Holmes et al., 1986), (Cody et al., 1996), while the hollow dots and those with “+” center are the ODEs and SSA results constructed from timestamp t = 1.0 × 105 in the time series (see Appendix-fig. 10A-G), respectively. In the model settings, SR = 1, SC = 20 (in (A)), 35 (in (B)) or 45 (in (C)). Di (i = 1, …, SC) is the only parameter varying with the consumer species, which was randomly drawn from a Gaussian distribution. The Shannon entropies of the experimental data and simulation results for each ecological community are: = 2.98(3.06, 2.98), = 4.04(4.02, 4.02), = 3.78(3.40, 3.41). In the Kolmogorov-Smirnov (K-S) test, the probabilities (p values) that the simulation results and the corresponding experimental data come from identical distributions are: = 0.89, = 0.88; = 0.47, = 0.75; = 0.88, = 0.77. With a significance threshold of 0.05, none of the p values suggest there exists a statistically significant difference. The numerical results in (A-C) were simulated from Eqs. 1, 2, 4. In (A): ai = 0.1, a’i = 0.125, di = 0.5, d’i = 0.6, wi = 0.22, ki = 0.1, Di = 0.016 × N(1, 0.35) (i = 1, …, 20); Ra = 350, K0 = 106. In (B): ai = 0.1, a’i = 0.125, di = 0.5, d’i = 0.6, wi = 0.22, ki = 0.1, Di = 0.012 × N(1, 0.35) (i = 1, …, 35); Ra = 350, K0 = 106. In (C): ai = 0.1, a’i = 0.14, di = 0.5, d’i = 0.5, wi = 0.2, ki = 0.1, Di = 0.015 × N(1, 0.32) (i = 1, …, 45); Ra = 550, K0 = 106.

Time courses of the species abundances in the scenario involving chasing pair and intraspecific interference.

The time series in (A, E), (B, F), (C, G), (D, H), (I, J) and (K) correspond to that shown in Appendixfig. 9A-C, Fig. 3D (bat), Fig. 3D (lizard) and Fig. 3C (butterfly), respectively.

A model of intraspecific interference illustrates the rank-abundance curves across different ecological communities (SCSR).

The solid dots represent the experimental data (marked with “Exp”) reported in existing studies (Hubbell, 2001), (Holmes et al., 1986), (Cody et al., 1996), (Clarke et al., 2005), while the hollow dots and those with “+” center are the ODEs and SSA results constructed from timestamp t = 1.0 × 105 in the time series (see Appendix-fig. 13), respectively. In the model settings, SR = 3, SC = 20 (in (A)), 35 (in (B)), 40 (in (C)), 45 (in (D)) or 50 (in (E)). Di (i = 1, …, SC) is the only parameter varying with the consumer species, which was randomly drawn from a Gaussian distribution. The Shannon entropies of the experimental data and simulation results for each ecological community are: = 2.98(2.98, 3.26), = 4.04(4.34, 4.35), = 3.00(3.00, 3.00), = 3.78(3.28, 3.64), = 4.05(3.94, 3.94). In the K-S test, the p values that the simulation results and the corresponding experimental data come from identical distributions are: = 0.59, = 0.43, = 0.47, = 0.33, = 0.42, = 0.27, = 0.22, = 0.06, = 0.56, = 0.36. With a significance threshold of 0.05, none of the p values suggest there exists a statistically significant difference. The numerical results in (A-E) were simulated from Eqs. 1, 2, 4. In (A-E): ail = 0.1, a’i = 0.125, dil = 0.5. In (a): d’i = 0.3, wil = 0.2, kil = 0.12, , , , Di = 0.021 × N(1, 0.28) (i = 1, …, 20, l = 1, 2, 3); , , . In (B): d’i = 0.6, wil = 0.2, kil = 0.12, = 8 × 104, , , Di = 0.017 × N(1, 0.3) (i = 1, …, 35, l = 1,2, 3); , = 160, . In (C): d’i = 0.4, wil = 0.3, kil = 0.12, , , , Di = 0.023 × N(1, 0.34) (i = 1, …, 40, l = 1, 2, 3); , , . In (d): d’i = 0.3, wil = 0.3, kil = 0.12, , = 5 × 104, , Di = 0.027 × N(1, 0.32) (i = 1, …, 45, l = 1, 2, 3); = 80, , . In (E): d’i = 0.3, wil = 0.3, kil = 0.2, , , , Di = 0.034 × N(1, 0.34) (i = 1, …, 50, l = 1, 2, 3); , , .

A model of intraspecific interference illustrates the rank-abundance curves across different plankton communities (SCSR).

The solid dots represent the experimental data (marked with “Exp”) reported in a recent study (Fuhrman et al., 2008), while the hollow dots and those with “+” center are the ODEs and SSA results constructed from timestamp t = 1.0 × 105 in the time series (see Appendix-fig. 13), respectively. The plankton community data were obtained separately from the Norwegian Sea (NS) and Pacific Station (PS). In the model settings, SR = 1 (in (B-C)), 3 (in (A)); SC = 50 (in (A, C)), 150 (in (B)). Di (i = 1, …, SC) is the only parameter varying with the consumer species, which was randomly drawn from a Gaussian distribution. The Shannon entropies of the experimental data and simulation results for each plankton community are: for SR = 3; = 4.68(6.53, 6.16), for SR. In the K-S test, the p values that the simulation results and the corresponding experimental data come from identical distributions are: , for SR = 3; ; , , for SR = 1. With a significance threshold of 0.05, none of the p values suggest there exists a statistically significant difference. The numerical results in (A-C) were simulated from Eqs. 1, 2, 4. In (A): ail = 0.1, a’i = 0.125, dil = 0.5, d’i = 0.2, wil = 0.3, kil = 0.2, = 8 × 104, , , = 280, , , Di = 0.035 × N(1, 0.25) (i = 1, …,50, l = 1, 2, 3). In (b): ai = 0.1, a’i = 0.125, di = 0.3, d’i = 0.3, wi = 0.3, ki = 0.2, Di = 0.025 × N(1, 0.25) (i = 1, …, 150, l = 1, 2, 3); Ra = 350, K0 = 104. In (c): ai = 0.1, a’i = 0.125, di = 0.3, d’i = 0.3, wi = 0.3, ki = 0.2, Ra = 350, K0 = 104, Di = 0.03 × N(1, 0.27) (i = 1, …, SC).

Time courses of the species abundances in the scenario involving chasing pair and intraspecific interference.

The time series in (A, E), (B, F), (C, G), (D, H), (I, J), (K, L), (M, N) and (O, P) correspond to that shown in Appendix-figs. 11A-E and 12A-C, respectively.

A model of intraspecific interference illustrates the rank-abundance curve of a bird community (SCSR).

(A) The rank-abundance curve. The solid dots represent the bird community data (marked with “Exp”) collected longitudinally within the same Amazonian region in 1982 (blue) and 2018 (cyan) (Terborgh et al., 1990), (Martínez et al., 2023). The hollow dots are the ODEs results constructed from timestamp t = 1.0 × 105 in the time series (see (B)). (B) Time courses of the species abundances simulated with ODEs. In the model settings, SR = 3 and SC = 250. Di (i = 1, …, SC) is the only parameter varying with the consumer species, which was randomly drawn from a Gaussian distribution. The Shannon entropies of the experimental data and simulation results for the bird community are . In the K-S test, the p values that the simulation results and the corresponding experimental data come from identical distributions are: , . With a significance threshold of 0.05, none of the p values suggest there exists a statistically significant difference. (C) Time courses of the species abundances simulated with ODEs corresponding to Fig. 3A (bird), and the simulation parameters are the same as Fig. 3A (bird). The numerical results in (A-C) were simulated from Eqs. 1, 2, 4. In (A-B): ail = 0.1, a’i = 0.125, dil = 0.5, d’i = 0.6, wil = 0.3, kil = 0.2, Di = 0.032 × N(1, 0.17) (i = 1, …,250, l = 1, 2, 3); , , , , , .

Intraspecific interference results in an underlying negative feedback loop and thus promotes biodiversity.

(A-B) The fraction of consumer individuals engaged in pairwise encounter. Here, SC = 40 and SR = 1. xi represents , and yi represents . Xi/Ci and yi/Ci stand for the fractions of consumer individuals within a chasing pair and an interference pair, respectively. The numerical results were calculated from Eq. S55, while the analytical solutions (marked with superscript “(A)”) were calculated from Eq. S59. The orange surface in (A) is an overlap of the red and yellow surfaces. (C) The formation of intraspecific interference results in a self-inhibiting negative feedback loop. In (A-B): ai = 0.0015, a’i = 0.0021, di = 0.1, d’i = 0.05, ki = 5. In (B): R = 2000.