Species Limits: Constraints in Complex Systems

• Species Limits are constraints defined by ecological interactions, resource dynamics, and stochastic effects that restrict the feasible configurations of species in a system.
• They quantify limits on coexistence, packing, and evolutionary change across models, employing frameworks like Lotka–Volterra dynamics and consumer–resource theory.
• This concept unifies diverse scientific contexts—from microbial ecology to high-energy physics—by linking system stability to effective degrees of freedom and boundary conditions.

“Species limits” is not a single technical notion across the research literature. In the sources considered here, it denotes several distinct but structurally related kinds of constraints: ecological limits on which species can coexist and at what abundances in microbial or competitive communities; packing bounds on how many species can occupy available niches; rate limits on how fast species-level traits can evolve; fixation-time limits on how long species can coexist in advected populations; asymptotic limits for the number of sampled species in Poisson–Dirichlet models; and, in high-energy theory, the “species scale,” the cutoff associated with the number of light degrees of freedom. A common theme is that the admissible state space of a many-component system is restricted by interaction structure, resource dynamics, stochasticity, or the counting of effective degrees of freedom (Fisher et al., 2014, Cui et al., 2019, García-Pintos, 2022, Maller et al., 2023, Aoufia et al., 2024).

1. Ecological limits as interaction-constrained coexistence

In the gut-microbiome setting, “species limits” are formulated as ecological constraints on which species can coexist and at what abundances, given their interaction structure (Fisher et al., 2014). The underlying model is a stochastic, discrete-time Lotka–Volterra model for absolute abundances $x_i(t)$,

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$

with $\delta t=1$ for daily sampling, interaction coefficients $c_{ij}$, equilibrium abundances $\langle x_j\rangle$, and multiplicative stochastic factor $\eta_i(t)$ satisfying $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$ (Fisher et al., 2014). After taking logs,

$$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$

so the interaction matrix $C=(c_{ij})$ determines equilibria, stability, and the feasible region of community compositions (Fisher et al., 2014).

The same paper emphasizes that metagenomic studies typically observe relative abundances $\tilde x_i(t)$ satisfying

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$0

This compositional constraint makes the design matrix singular and induces spurious negative correlations, so abundances are restricted to a simplex rather than an unconstrained Euclidean state space (Fisher et al., 2014). In this sense, the simplex is a geometrical expression of global resource or space limitation, while the discrete-time Lotka–Volterra parameters encode which community states are dynamically achievable and stable.

To infer such limits from data, the paper introduces LIMITS, “Learning Interactions from MIcrobial Time Series,” which combines sparse linear regression with bootstrap aggregation to infer a discrete-time Lotka–Volterra model from metagenomic time series (Fisher et al., 2014). The method is designed to address three obstacles stated explicitly in the abstract: correlation does not imply interaction, the sum constraint on relative abundances makes inference difficult, and experimental uncertainty or OTU mis-assignment biases interaction estimates (Fisher et al., 2014). Synthetic tests showed reliable recovery of interaction topology, and application to two human gut microbiomes found that the inferred interaction networks differed substantially between individuals (Fisher et al., 2014).

This usage makes “species limits” fundamentally dynamical. The sign and magnitude of $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$1 determine whether species $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$2 pushes species $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$3 upward or downward, and therefore constrain persistence, exclusion, and equilibrium abundance. A plausible implication is that community individuality can be framed as an individual-specific feasible and stable region in abundance space.

2. Keystone control, species packing, and coexistence bounds

The same gut-microbiome analysis uses a network-based notion of keystone species: species with disproportionately many and/or strong outgoing interactions, even when they are only moderately abundant (Fisher et al., 2014). In the two inferred gut networks, the dominant nodes were distinct: Bacteroides fragilis in one individual and Bacteroides stercosis in the other (Fisher et al., 2014). For the 14-species set analyzed, most species had $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$4–$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$5 outgoing interactions in one individual and $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$6–$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$7 in the other, whereas the two keystone candidates had $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$8 and $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$9 outgoing interactions, respectively (Fisher et al., 2014). Because such nodes alter the net growth of many partners, they effectively set coexistence and abundance limits for the broader community.

A different formulation of coexistence limits appears in generalized consumer–resource theory. For $\delta t=1$0 consumers and $\delta t=1$1 resources, the model

$\delta t=1$2

distinguishes self-renewing resources, $\delta t=1$3, from externally supplied resources, $\delta t=1$4 (Cui et al., 2019). In the externally supplied case, the steady-state resource level is

$\delta t=1$5

The classical competitive-exclusion argument yields $\delta t=1$6, where $\delta t=1$7 is the number of surviving species (Cui et al., 2019). However, in the high-diversity random setting analyzed by the cavity method, externally supplied resources satisfy the stronger bound

$\delta t=1$8

under the stated generic assumptions, whereas self-renewing resources allow $\delta t=1$9, with $c_{ij}$0 the fraction of non-extinct resources (Cui et al., 2019).

This distinction leads to an explicit classification. “Isostatic species packing” denotes systems that can saturate $c_{ij}$1; “hypostatic species packing” denotes systems with a stricter upper bound $c_{ij}$2; and “non-generic (overpacked) species packing” denotes cases with $c_{ij}$3 due to hard metabolic tradeoffs such as $c_{ij}$4 with very small or zero noise (Cui et al., 2019). The species-packing ratio $c_{ij}$5 therefore becomes a quantitative species limit controlled not only by niche count but by resource dynamics.

A complementary high-dimensional viewpoint is developed for generalized Lotka–Volterra communities with orthogonally invariant interaction matrices (Martínez, 2024). In the unstructured model,

$c_{ij}$6

the fraction of surviving species is $c_{ij}$7, determined self-consistently through a truncated-Gaussian fixed-point equation

$c_{ij}$8

together with closure relations involving the spectral functional $c_{ij}$9 of the interaction matrix (Martínez, 2024). Stability is controlled by the upper edge of the spectrum of the reduced interaction matrix,

$\langle x_j\rangle$0

Here, species limits are emergent community-level constraints imposed by the interaction spectrum and self-regulation rather than by explicit niche counting (Martínez, 2024). The paper further shows that extinctions transform the spectrum of the surviving community according to

$\langle x_j\rangle$1

so increased extinction drives the reduced system toward more universal random-matrix behavior (Martínez, 2024).

3. Strong competition, segregation, and geometric limit configurations

A more literal spatial version of species limits arises in strongly competing reaction–diffusion systems. For four competing species on a bounded planar domain $\langle x_j\rangle$2,

$\langle x_j\rangle$3

with nonnegative segregated boundary data $\langle x_j\rangle$4, the limit $\langle x_j\rangle$5 induces spatial segregation: coexistence regions shrink and the species supports become disjoint in the limit (Lanzara et al., 2018). The limiting configuration $\langle x_j\rangle$6 is unique, belongs to the segregated class $\langle x_j\rangle$7, and is also the unique minimizer of the Dirichlet energy

$\langle x_j\rangle$8

among segregated states with the prescribed boundary conditions (Lanzara et al., 2018).

For a segregated configuration $\langle x_j\rangle$9, the nodal regions are

$\eta_i(t)$0

and the multiplicity of a point is

$\eta_i(t)$1

A central structural result states that exactly one of two possibilities occurs: either there is a single point $\eta_i(t)$2 with $\eta_i(t)$3, a 4-point where all four species concur, or there are precisely two distinct points $\eta_i(t)$4 with multiplicity $\eta_i(t)$5, two 3-points where only three species concur (Lanzara et al., 2018). No other global arrangement of multiplicity-$\eta_i(t)$6 points is possible.

The 4-point case is characterized by a harmonic function built from alternating signs: $\eta_i(t)$7 If the limiting configuration has a 4-point, then

$\eta_i(t)$8

Conversely, if $\eta_i(t)$9 has an interior critical point $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$0 with $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$1, then $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$2 and $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$3 is a 4-point (Lanzara et al., 2018). The paper also gives necessary and sufficient integral conditions for a prescribed point $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$4 to be a 4-point, expressed via the conformal automorphism $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$5 and three vanishing boundary integrals (Lanzara et al., 2018).

This setting gives species limits a geometric content. Under extreme competition, only certain territorial partitions are admissible, and the admissible multiple-point topology is rigid. The authors note that 4-point configurations satisfy codimension-3 conditions on the boundary data and are therefore nongeneric, whereas two 3-point configurations are the generic outcome (Lanzara et al., 2018). A plausible implication is that strong competition produces robust segregation patterns rather than arbitrary coexistence mosaics.

4. Limits on evolutionary change and coexistence time

In evolutionary theory, “species limits” can refer to upper bounds on how fast quantitative traits can change. In a population of types $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$6 with counts $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$7, frequencies $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$8, growth rates $\ln \eta_i(t)\sim\mathcal N(0,\sigma^2)$9, and trait values $$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$0, the mean trait is

$$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$1

The continuous-time Price equation is

$$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$2

Using Cauchy–Schwarz,

$$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$3

This is the general rate limit derived in “Limits on the Evolutionary Rates of Biological Traits” (García-Pintos, 2022). It bounds the evolutionary component of trait change by the product of trait variability and growth-rate variability.

Under replicator dynamics,

$$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$4

with $$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$5, the bound becomes

$$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$6

For intrinsic traits with $$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$7,

$$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$8

The same framework yields a generalized Fisher theorem,

$$\ln x_i(t+1)-\ln x_i(t)=\zeta_i(t)+\sum_j c_{ij}(x_j(t)-\langle x_j\rangle),$$9

and, with mutation, the bound

$C=(c_{ij})$0

where $C=(c_{ij})$1 is the mutation-only distribution (García-Pintos, 2022). With genetic drift, the paper derives an integrated inequality whose combined form is

$C=(c_{ij})$2

These results apply explicitly to trait dynamics within or across species and to the evolution of bacterial strains (García-Pintos, 2022). Species limits here are rate limits: ceilings on evolutionary change set by variance structure.

A different temporal notion appears in chaotic-flow models of neutral coexistence. In a two-species voter-model process on a dynamic interaction network generated by advection, the mean fixation time $C=(c_{ij})$3 is shortest in both the static-network and well-mixed limits but maximal at intermediate Damköhler number $C=(c_{ij})$4, where flow and evolutionary time scales are comparable (Galla et al., 2016). In the well-mixed limit,

$C=(c_{ij})$5

while on the static $C=(c_{ij})$6 lattice with $C=(c_{ij})$7,

$C=(c_{ij})$8

for the parameters studied (Galla et al., 2016). The slowdown at intermediate $C=(c_{ij})$9 is attributed to flow-induced modularization of the interaction graph, quantified by

$\tilde x_i(t)$0

where $\tilde x_i(t)$1 is the fraction of active links and $\tilde x_i(t)$2 the fraction of one type (Galla et al., 2016). Here, species limits are coexistence-time limits controlled by the balance between advection and evolutionary turnover.

5. Sampling limits: $\tilde x_i(t)$3, $\tilde x_i(t)$4, and asymptotic species counts

In species-sampling theory, “species limits” concern the asymptotic behavior of the number of observed species $\tilde x_i(t)$5 and the frequency-of-frequencies vector

$\tilde x_i(t)$6

under Poisson–Dirichlet and generalized Poisson–Dirichlet priors (Maller et al., 2023). The occupancy identity

$\tilde x_i(t)$7

always holds, and

$\tilde x_i(t)$8

For PD$\tilde x_i(t)$9, based on the gamma subordinator and Ewens’ sampling formula, the asymptotic behavior is logarithmic: $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$00 For each fixed $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$01,

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$02

and Theorem 4.1 shows that the finite vector $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$03 is asymptotically independent of the centered-and-scaled $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$04 (Maller et al., 2023). This is the asymptotic independence highlighted in the abstract.

For PD$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$05, PD$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$06, and PD$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$07, the growth is instead power-law: $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$08 with the limit given by the Mittag–Leffler law in the PD$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$09 and PD$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$10 cases (Maller et al., 2023). For fixed $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$11, the asymptotic proportions of classes of size $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$12 are centered at

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$13

and the conditional fluctuations are Gaussian: $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$14 where $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$15 is the covariance matrix defined in the paper in terms of the $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$16 (Maller et al., 2023). Thus the finite frequency spectrum is conditionally normal given the total number of species.

This contrast is structurally important. In PD$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$17, total diversity and the finite-dimensional spectrum asymptotically decouple. In the $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$18 models, the spectrum and $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$19 remain coupled through the random scale $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$20, although the conditional law is asymptotically Gaussian (Maller et al., 2023). A plausible implication is that “species limits” in sampling theory separate into logarithmic and power-law regimes according to whether the governing partition law is gamma-type or stable-type.

6. Species scale in perturbative string theory and multi-species dark matter

In perturbative string theory, the relevant notion is the “species scale,” the EFT scale associated with the number of light degrees of freedom. In $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$21 dimensions, the paper “Species scale, worldsheet CFTs and emergent geometry” defines the counting species scale $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$22 by

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$23

in Planck units, with $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$24 the number of species lighter than $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$25 (Aoufia et al., 2024). Using an exponential regulator, the equivalent definition is

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$26

The paper distinguishes $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$27, the UV cutoff $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$28, and the quantum-gravity scale $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$29, emphasizing the hierarchy

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$30

In decompactification with $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$31 large internal dimensions and Kaluza–Klein gap $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$32, Weyl’s law yields

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$33

so $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$34 equals the higher-dimensional Planck scale in that regime (Aoufia et al., 2024).

The same paper shows, in perturbative type II vacua, that a vanishing cutoff in Planck units is equivalent to the appearance of an infinite tower of light states, and that under mild assumptions the cutoff scales with the spectral gap of the internal CFT in the same way as in decompactification or emergent-string limits (Aoufia et al., 2024). This is a different use of “species limits”: not a bound on biological coexistence, but a limit on the validity of gravitational EFT imposed by the number of light species.

A second physics example comes from multi-species wave dark matter. For $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$35 nonrelativistic scalar fields $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$36, the total density contrast is

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$37

and the total power spectrum decomposes as

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$38

for arbitrary numbers of component species, density fractions, and initial field power spectra (Amin et al., 20 Oct 2025). The isocurvature piece is

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$39

and the characteristic species-dependent suppression scales include the wave Jeans scale

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$40

the free-streaming scale $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$41, the warm Jeans scale $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$42, and the white-noise scale

$$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$43

for Gaussian initial spectra (Amin et al., 20 Oct 2025).

In this setting, species limits concern which combinations of masses $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$44, initial spectra $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$45, and fractions $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$46 are compatible with the observed growth of structure. The paper states that the framework includes cold and warm wave dark matter, globally or locally misaligned scalar fields, multi-component fields with spin $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$47, and cold and warm particle dark matter in the appropriate limits (Amin et al., 20 Oct 2025). A plausible implication is that the effective number of observationally relevant dark-matter species is the number of components whose Jeans, free-streaming, or Poisson scales lie in the observable range and whose $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$48 are large enough to affect the total power spectrum.

7. Unifying theme

Across these domains, “species limits” consistently denotes a restriction on admissible multiplicity, abundance, rate, or scale. In microbial ecology it is the feasible and stable region of a community under an inferred interaction matrix (Fisher et al., 2014). In consumer–resource and generalized Lotka–Volterra theory it is the maximal richness or survivor fraction compatible with resource dynamics or spectral stability (Cui et al., 2019, Martínez, 2024). In strongly competing PDE systems it is the limited set of segregated topologies compatible with the infinite-competition limit (Lanzara et al., 2018). In evolutionary theory it is the variance-controlled ceiling on trait change (García-Pintos, 2022), and in chaotic flows it is the longest coexistence time attainable when advection and evolutionary time scales are balanced (Galla et al., 2016). In species-sampling models it is the limiting law of $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$49 and $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$50 as $$x_i(t + \delta t) = \eta_i(t)\, x_i(t)\, \exp\!\Big( \delta t \sum_j c_{ij} (x_j(t) - \langle x_j \rangle) \Big),$$51 (Maller et al., 2023). In high-energy and cosmological physics it is the cutoff or clustering behavior imposed by the number and properties of effective species (Aoufia et al., 2024, Amin et al., 20 Oct 2025).

This suggests that the term is best understood not as a taxonomic boundary but as a family of constraint concepts. The precise object being limited varies—coexistence, packing, segregation pattern, evolutionary speed, sampling diversity, or EFT validity—but the technical role is similar: a many-component system is not free to explore all nominal configurations, because dynamics, geometry, statistics, or mode counting restrict the reachable set.