← News
Pollution Wins Pollution Wins Planet

The secret to a steadier ecosystem may be that its species grow up

The secret to a steadier ecosystem may be that its species grow up
More Complexity Can Stay Stable Model finding
Strong Competition Across Life Stages Key condition

A community can become more stable when its members stop behaving like interchangeable units. In a mathematical model, allowing juveniles and adults to interact differently with other species expands the range of ecological complexity that can remain stable. But the benefit has a condition: life stages within a species must also compete strongly with one another (Garmire and O’Dwyer, 2026).

That result addresses a stubborn puzzle. Classic ecological theory says that as a community gains species and interactions, it should become harder to keep stable. Yet forests, lakes and other diverse ecosystems persist through disturbances. One reason may be that nature’s species are not uniform blocks. A young fish may eat different prey, face different predators and occupy a different habitat than the adult it will become. Those differences, this study suggests, can help a whole community absorb complexity.

The finding is a theoretical result, not a field demonstration that age structure will rescue any particular ecosystem. The authors build a model and map its local stability boundary: the point beyond which small disturbances no longer fade away near an equilibrium. Their contribution is to show, mathematically, how life-stage differences can shift that boundary—and why the internal relationships among a species’ own stages matter too.

The Science

The work by Grace Garmire and James P. O’Dwyer, of the University of Illinois Urbana-Champaign, revisits a question that has shaped theoretical ecology for decades: how can highly diverse communities persist if complexity itself tends to destabilize them? The paper is an arXiv preprint, so its findings should be read as a contribution to ongoing research rather than as a settled ecological rule (Garmire and O’Dwyer, 2026).

The classic starting point is Robert May’s 1972 model. May imagined a large community in which species interact through a web of randomly distributed effects. In the model’s simplest form, each species is a single population: a species has one abundance, one self-regulating effect, and one average effect on every other species. That is a useful simplification, but it erases the transformations that happen during an organism’s life.

May’s stability criterion combines three ingredients: the number of species, the fraction of possible interactions that actually occur, and the typical strength of those interactions. Their combined measure is called complexity, written as , where is the number of species, is interaction connectance, and describes variation in interaction strength. In the large-community limit, the classical model predicts stability when ; above that threshold, instability becomes overwhelmingly likely. The threshold is not a claim that real ecosystems all have one universal tipping point. It is a benchmark produced by a deliberately general model.

To test what that benchmark misses, Garmire and O’Dwyer give each species two life stages. Stage one matures into stage two; stage two can reproduce and generate individuals in stage one. Each stage also has mortality and density regulation. The model tracks the abundance of each stage separately, so a community with species now has population variables.

The central change is that an interaction can depend on the stages involved. An adult of one species might compete with a juvenile of another in a way that differs from the interaction between the two adults. The researchers tune how similar those stage-specific interactions are using a correlation parameter, . When , stages interact in the same way—the model collapses toward the assumption that life stages are functionally alike. Lower correlation means more variation across stages; at zero, stage-specific interaction effects are uncorrelated in the model.

A second ingredient is competition between stages of the same species. The researchers compare models where these within-species, cross-stage effects are strongly negative with a case where they are instead drawn randomly. This contrast asks whether merely splitting species into more population groups is enough, or whether a particular demographic structure is doing the stabilizing work.

Table 1: Model types used for interaction matrix
Table 1: Model types used for interaction matrix Source: Grace Garmire, James P. O'Dwyer

The stability test is local. The authors calculate a Jacobian—a matrix summarizing how each population’s growth changes when any population is nudged—and ask whether the system returns toward its equilibrium after a small perturbation. An equilibrium here means a long-term average state, not a claim that real populations sit motionless. The team combines analytical predictions from random matrix theory with numerical simulations across randomly generated interaction networks.

What They Found

The headline result is that stage-dependent interactions can expand the stable parameter space. As the interactions experienced by different stages become less alike, the model can support greater complexity before local stability is lost. In ecological terms, different stages can create distinct routes through the interaction network: pressure on one stage does not necessarily translate into the same pressure on another.

But heterogeneity by itself is not the whole story. The authors report that the boost depends on strong competition between life stages within a species. When that competition is relaxed and cross-stage interactions are random, the stabilizing effect is not equivalent to the result for strongly competing stages. This condition is important: the study does not say that adding more demographic categories automatically stabilizes a community. It identifies a combination—stage-specific effects between species, alongside strong within-species competition—that broadens the stability boundary.

The effect also has limits. The authors say stage structure does not abolish the destabilizing influence of complexity. More species and stronger interactions still eventually push communities toward instability. The result is an expanded boundary, not immunity from it. Their numerical analysis summarizes that boundary with , the complexity at which half of randomly drawn interaction networks remain stable. Figure 4 compares this threshold across community sizes, life-cycle rates and degrees of stage correlation. Its caption reports that at large system size and large demographic rates, the simulations approach the analytical predictions; lower correlation can raise the stability threshold, while small demographic rates can weaken the benefit.

Figure 4: 
We numerically compute K50, the threshold at which 50% of random realizations of the interaction matrix remain stable across values of S, ρ, and Δ. (a) At large Δ=1000, dependence on system size S is shown for different values of correlation ρ and for the case of 𝐁ind. (b) Similarly for a large system with S=1000 species, dependence on Δ is shown for different values of correlation ρ and for the case of 𝐁ind. At large S and Δ, the threshold for stability approaches the analytical prediction for each case. In (b), for decreasing correlation ρ, larger Δ is required to use the large Δ limit, but even at finite Δ values we see the increasing stability limit of complexity with decreased ρ. Standard error is computed in (a) and (b) using a bootstrapping method, but is smaller than the marker size for most points. Dashed lines are used to connect points in order to illustrate general trends. (c) Stability threshold K50 is computed across a sweep of ρ, Δ for a system of size S=500. While there is no Δ dependence for ρ=1, at large but fractional values of ρ and small Δ, stability is decreased as demographic rates are too small to counteract inter-specific interactions. For values of Δ>0.5, decreasing ρ is increasingly stabilizing. At large Δ and the minimum value of ρ=−13, the threshold for asymptotic stability diverges.
Figure 4: We numerically compute K50, the threshold at which 50% of random realizations of the interaction matrix remain stable across values of S, ρ, and Δ. (a) At large Δ=1000, dependence on system size S is shown for different values of correlation ρ and for the case of 𝐁ind. (b) Similarly for a large system with S=1000 species, dependence on Δ is shown for different values of correlation ρ and for the case of 𝐁ind. At large S and Δ, the threshold for stability approaches the analytical prediction for each case. In (b), for decreasing correlation ρ, larger Δ is required to use the large Δ limit, but even at finite Δ values we see the increasing stability limit of complexity with decreased ρ. Standard error is computed in (a) and (b) using a bootstrapping method, but is smaller than the marker size for most points. Dashed lines are used to connect points in order to illustrate general trends. (c) Stability threshold K50 is computed across a sweep of ρ, Δ for a system of size S=500. While there is no Δ dependence for ρ=1, at large but fractional values of ρ and small Δ, stability is decreased as demographic rates are too small to counteract inter-specific interactions. For values of Δ>0.5, decreasing ρ is increasingly stabilizing. At large Δ and the minimum value of ρ=−13, the threshold for asymptotic stability diverges. Source: Grace Garmire, James P. O'Dwyer

The paper’s model also makes clear why the speed of a life cycle matters. The parameter sets a minimum scale for maturation and reproduction rates. When those rates are too small relative to the interaction effects, demographic links may not counteract the destabilizing dynamics quickly enough. For some fractional values of , the paper reports lower stability at small , with the pattern changing as demographic rates rise. That caveat makes the result more biologically meaningful than a simple “more variation is better” rule: what stages do and how quickly organisms move between them both matter in the equations.

The reported limiting case is striking. With two stages, the smallest mathematically allowed correlation is ; at that value and large , the caption says the threshold for asymptotic stability diverges. In plain language, under this model’s limiting assumptions, the analysis does not find a finite complexity threshold beyond which stability is lost. That is a mathematical edge case—not a prediction that a natural ecosystem can host unlimited complexity. The model’s simplifying assumptions and the meaning of “large” demographic rates constrain how literally to interpret it.

For comparison, the single-stage framework has the familiar large-community boundary . The expanded model’s important result is not a replacement universal number, but that the boundary can move depending on how stage-specific interactions covary and how strongly stages compete within a species. The figures use system sizes such as and to examine that approach to the large-community limit; the supplied paper text does not provide the plotted point values, so the charts below summarize reported benchmarks and parameter settings rather than reconstructing the figure’s curves.

Classical benchmark for community complexity

The one-stage May model predicts stability in the large-community limit below K=1. This is the baseline criterion, not a measured ecological value.

Classical benchmark for community complexity
LabelValue
Stability boundary K1 K

Allowed stage-correlation range in the two-stage model

For two life stages, the paper gives a maximum correlation of 1 and a minimum of -1/3. The minimum is the model’s most anticorrelated limit.

Allowed stage-correlation range in the two-stage model
LabelValue
Minimum ρ (two stages)-0.3333333333 correlation
Maximum ρ1 correlation

Community sizes highlighted in the stability sweep

The Figure 4 caption identifies systems with 500 species in the correlation–demographic-rate sweep and 1,000 species in the demographic-rate comparison. These are reported simulation settings, not plotted stability outcomes.

Community sizes highlighted in the stability sweep
LabelValue
Correlation and Δ sweep500 species
Δ comparison1,000 species

Why This Changes Things

The result gives a plausible mathematical route between two familiar observations: simple models often predict fragile complex communities, while real ecosystems can be diverse and persistent. It does not overturn May’s insight. Instead, it asks whether the units in May’s model—species treated as homogeneous populations—are too coarse to capture structures that matter for stability.

Life-stage differences are not an obscure detail. Young and adult organisms can use different resources, encounter different predators or compete in different places. A species that shifts its diet as it grows may connect different parts of a food web at different points in its life. These ontogenetic changes—the changes an organism undergoes as it develops—can alter who affects whom, and how strongly. The model suggests that such functional diversity may make the whole interaction network less uniform and, under particular conditions, more stable.

The emphasis on competition among stages adds a useful complication. Juveniles and adults are not just two versions of the same ecological actor; they can also draw on shared resources or otherwise constrain one another. Strong negative interactions between stages may counterbalance the extra degrees of freedom created by treating those stages separately. In that sense, a species’ internal structure can stabilize its external relationships. The result resonates with broader ecological work emphasizing that stability depends not simply on how many species or links exist, but on how those links are arranged.

This perspective could matter for conservation, especially where human pressures alter the age or size structure of populations. Harvesting large adults, for example, can change the balance between stages; habitat fragmentation or warming can alter survival and maturation. The paper does not test these interventions, so it cannot tell a manager that protecting a particular life stage will stabilize a real ecosystem. It does suggest that models used to assess such risks may miss a relevant mechanism if they represent each species with one average population.

That is a practical modeling point, not a promise of easy prediction. Adding stages also increases the number of rates and interactions that must be estimated. The paper itself notes how difficult it is to measure all ecological interactions in a diverse community. The framework therefore helps identify what may matter, while leaving open how to estimate it reliably in forests, lakes, fisheries or microbial systems.

What’s Next

The biggest next step is empirical: determine whether real communities show the combination that matters in the model. Researchers could measure how interaction effects differ between juvenile and adult stages, estimate competition among those stages, and ask whether communities with stronger stage-specific variation are more resilient to perturbations. Such tests would need to distinguish stability near an average state from other meanings of resilience, including recovery after a large shock or resistance to species loss.

The model also makes simplifying choices that future work can relax. The main analysis considers two stages and, in the described results, full connectance; real food webs have missing links, varied life histories and interaction types that are not random. The authors report supplementary analysis for lower connectance, but translating the framework to particular ecosystems will require more realistic networks, environmental fluctuations and potentially changing demographic rates. They also assume a feasible equilibrium in order to isolate local stability. Whether every species can persist at positive abundance is a separate question, and an ecosystem can fail that test even if its equilibrium would be locally stable.

A further question is how far the result extends beyond local asymptotic stability. A system can return after a small perturbation yet respond very differently to a severe drought, an invasive species or a rapid climate shift. Nor does the mathematical threshold alone say whether a community will keep delivering services people rely on. Linking stage structure to persistence, recovery time, and ecosystem function would make the work more directly useful for conservation decisions.

Still, the study shifts the question in a productive way. Instead of asking only how many species a community can contain before complexity overwhelms it, ecologists can ask how development changes the interaction network—and when internal competition helps that variation stabilize the whole. If future data support the model’s conditions, life history may turn out to be more than a detail to add after the food web is drawn. It may be part of the architecture that lets a complex community endure.