The Math of Microbial Survival: Why Most Species in a Community Must Die

In a handful of soil teeming beneath your feet, hundreds of microbial species compete for survival. They consume resources, grow, reproduce—and then release metabolic waste products that other species can consume in turn. This invisible economy of give-and-take underlies everything from the fertility of farmland to the health of your gut microbiome. Yet the mathematical models scientists have used for decades to predict how these communities behave have systematically failed to account for one crucial feature of real microbial life: the metabolites that one species excretes become food for another. A new theoretical framework, published by researchers at the University of Fribourg, the University of Lausanne, and the University of Applied Sciences Western Switzerland, now provides rigorous mathematical ground for understanding why this omission matters—and what it costs us in our ability to predict, and perhaps engineer, microbial ecosystems.

The gap is more than academic. Microbial communities drive biogeochemical cycles, protect crops from disease, and maintain human health. Understanding their dynamics could help us design more resilient agricultural systems, develop targeted microbiome interventions, or predict how marine microbial communities will respond to warming oceans. But if the models we've been relying on are fundamentally mis-specified, our predictions will be systematically wrong.

The new paper, "Stability and Feasibility of Microbial Consumer-Resource Model" by Louis Faul, Xavier Richard, Jan Roelof van der Meer, and Christian Mazza, doesn't just identify this problem—it solves it. By rigorously extending the classical consumer-resource framework to include what biologists call cross-feeding, the researchers establish mathematical conditions under which microbial communities can persist, and under which they will collapse into extinction. Their key result is stark: when the number of resource species falls below the number of consumer species, the community cannot persist. Full stop. The math admits no exceptions.

But the paper offers hope alongside this warning. When survivors persist in numbers smaller than the available resources—a kind of competitive winnowing that has long been observed in nature—those surviving communities turn out to be generically stable. They don't teeter on the edge of collapse. They settle into equilibria that resist perturbation. This is the paradox at the heart of the work: the same mathematical forces that drive mass extinction also create the conditions for long-term stability among the survivors.

The Science

To understand what Faul and colleagues accomplished, you need to understand what came before—and why it fell short.

The story begins with Robert MacArthur, a theoretical ecologist who died far too young in 1972 at age 42. MacArthur was one of the architects of modern population biology, and in the late 1950s and 1960s he developed what became known as the Consumer-Resource Model (CRM). The basic idea was elegant: species consume resources, grow, and die. Resources are supplied from outside (think of nutrients washing into a stream, or sunlight striking a leaf) and are consumed by species at rates determined by how efficiently each species can find and metabolize each resource.

MacArthur's genius was in recognizing that this system operates on two fundamentally different timescales. Resources—sugars, amino acids, minerals—turn over quickly. A molecule of glucose in a bacterial environment might be consumed and processed in seconds or minutes. Species, by contrast, grow and die more slowly. A bacterial population might take hours or days to double. MacArthur exploited this separation of scales by assuming that resources are always at equilibrium—always in a steady state—while species abundances change gradually. This "slow-fast approximation" reduced the full dynamical system to something mathematically tractable: a set of equations describing how species compete for resources that are always in balance.

The resulting framework became the workhorse of theoretical ecology. When you plug in parameters describing consumption rates, mortality, and resource supply, you get predictions about which species will survive, how abundant they'll be, and how the community will respond to perturbations. For decades, ecologists used variations of this model to understand everything from competition among forest trees to predator-prey dynamics.

But there was always a problem that the model couldn't quite explain. Field observations and laboratory experiments consistently showed that microbial communities produce more biomass than MacArthur's model predicted. If you set up a chemostat—a controlled environment where nutrients flow in and waste flows out—and introduce a community of bacterial species, the total mass of living organisms that accumulates is consistently larger than the model says should be possible. Something was missing.

The missing piece, researchers eventually realized, was cross-feeding. Real microbes don't just consume resources and die. They consume resources, grow, and excrete metabolic byproducts—waste products that, from one species' perspective, are actually valuable nutrients. A species that can't metabolize a particular sugar directly might thrive on the fermentation products released by another species that can. Metabolic pathways interweave. Species cooperate as well as compete. And this cooperation creates new resources that the original model never accounted for.

In 2019, a team led by Joshua Goldford and others (with important contributions by Marsland and colleagues) formalized this intuition in the Microbial Consumer-Resource Model, or MiCRM. The core innovation was simple but profound: instead of assuming that all energy extracted from a resource goes directly into species growth, the MiCRM allows a fraction of that energy to be shunted back into the environment as metabolic byproducts. These byproducts become resources for other species, creating a network of interdependence that the original MacArthur model couldn't capture.

When this cross-feeding is incorporated, the discrepancy disappears. Communities produce the observed biomass. The model fits the data. But there's a catch: the mathematics become substantially more complex. The simple Lyapunov function that guaranteed stability in MacArthur's model no longer applies. The system is no longer obviously tractable. And the question of whether communities governed by the MiCRM are stable—whether they persist or collapse—becomes genuinely uncertain.

This is exactly the problem Faul and colleagues set out to solve. Their paper applies the slow-fast approximation to the MiCRM, showing how the system behaves when resources reach equilibrium faster than species change. They analyze feasibility—whether positive equilibrium states exist—persistence—whether those equilibria can be maintained—and stability—whether small perturbations push the system back toward equilibrium or trigger collapse. They also extend the analysis to a stochastic setting, where randomness in births and deaths introduces genuine extinction events rather than mathematical equilibria.

The research draws on tools from several mathematical domains: random matrix theory (to understand what happens when interaction parameters are chosen at random, as they would be in nature), spectral theory (to determine stability from the eigenvalues of linearized systems), and the theory of Markov chains (to model stochastic dynamics). The authors—mathematicians and microbiologists collaborating across institutions—combine formal proof with biological insight.

What They Found

The paper's central result concerns what happens when resources are fewer than species. Let M denote the number of resource species and S the number of consumer species. The key inequality is M < S: fewer resource types than species trying to consume them.

Under this condition, the researchers prove that the slow dynamics of the MiCRM almost surely fail to have a feasible positive equilibrium. "Almost surely" is a technical term from probability theory, but its meaning is intuitive: when you pick the interaction parameters at random from continuous distributions—as would happen in natural communities—the probability of finding a stable coexistence state is zero. Not low. Zero.

This result emerges from a simple linear algebra fact. When you derive the equilibrium conditions for the slow species dynamics, you end up with a linear system. The dimension of its solution space is bounded by M, the number of resources. If S > M, the system is underdetermined—a vector in ℝ^S cannot generally be solved by M constraints. A random vector of mortality rates, drawn from continuous distributions, will almost never lie in the column space of the consumption matrix. No solution means no positive equilibrium. No positive equilibrium means species go extinct.

The implication is mass extinction—not metaphorical, but mathematical. When resources are limiting and cross-feeding is incorporated, competitive pressures become so intense that most species cannot persist. The community undergoes what the researchers call "non-persistence": not merely that some species die out, but that the entire dynamical system lacks any state where all species coexist at positive abundances.

This finding validates and extends earlier work by researchers including Mehta and colleagues, who used methods from statistical physics to argue that the ratio of surviving species to resources should satisfy a species packing bound: S*/M ≤ 1/2. In the large-system limit, no more than half the resource diversity can be translated into species diversity. Faul and colleagues' result is sharper: when S > M, persistence is impossible. The bound becomes exact rather than asymptotic.

But the paper's most surprising finding concerns what happens after the extinction events. Once species have been winnowed down to a surviving community of size S* < M, the dynamics change dramatically. The researchers show that such equilibria are generically stable.

This is a critical distinction. Generic stability means that stability is the typical case, not a special arrangement requiring fine-tuning. If you pick parameters at random, you'll land on a stable equilibrium, not an unstable one. The competition that drove mass extinction now creates a kind of robustness. The community settles into a configuration where each surviving species can persist, and where small perturbations—slight changes in resource supply, minor shifts in consumption rates—are absorbed without triggering further collapse.

To establish this result, the researchers exploited a connection between the resource dynamics and chemical reaction networks. The fast subsystem, they show, is equivalent to the mass-action kinetics of a first-order reaction network with a specific algebraic structure. These networks have a property called zero deficiency—a technical condition that ensures the steady-state distribution, when modeled as a stochastic process, follows a Poisson distribution. The Poisson steady state is not merely a mathematical convenience; it reflects the fundamental randomness of molecular encounters in real biochemical systems.

This connection to chemistry led to the paper's final contribution: a stochastic version of the MiCRM that incorporates the randomness inherent in biological systems. Real populations don't follow deterministic differential equations. Individual organisms are born and die at random times. Resources are consumed and produced stochastically. The researchers developed a continuous-time Markov chain model that captures these fluctuations, and they analyzed the resulting extinction dynamics.

The stochastic analysis reveals a tension between theoretical stability and empirical extinction. In the deterministic limit, surviving communities are stable. But when randomness is introduced, extinction events become possible—not because the equilibrium is unstable, but because stochastic fluctuations can, given enough time, push a species below the threshold of recovery. The probability of such events depends on system size, on the magnitude of fluctuations, and on the details of the interaction network. The researchers provide tools for analyzing these probabilities, connecting large-deviation theory to the Markov chain framework.

Why This Changes Things

The implications of this work extend across multiple fields, from theoretical ecology to applied microbiology.

For ecologists, the paper resolves a long-standing tension in the literature. The classical MacArthur model predicts that competition should lead to the principle of competitive exclusion: in a stable environment with limited resources, species that are too similar in their resource requirements cannot coexist. One will always outcompete the other. But natural communities often harbor far more species than the number of distinct resources would seem to allow. This "paradox of the plankton"—G. Evelyn Hutchinson's famous 1961 question about how dozens of phytoplankton species coexist while competing for a handful of nutrients—has resisted resolution for decades.

The MiCRM, as analyzed by Faul and colleagues, offers a way forward. Cross-feeding creates an effectively larger resource space. When species can consume each other's metabolic products, the number of distinct ecological niches expands. The system can support more species than the naive resource count would suggest. But the researchers' persistence condition—M < S implies non-persistence—shows that this flexibility is not unlimited. When species become too numerous relative to resources, extinction is inevitable. The paradox is partially resolved: coexistence is possible, but only within bounds.

For microbiologists, the work provides a theoretical foundation for understanding the observed structure of microbial communities. Laboratory experiments consistently show that microbial communities cluster around particular configurations—sets of dominant species that persist across environments and timescales. The stability result suggests why: these configurations are attractors of a stable dynamical system, not random fluctuations. Understanding which attractors are possible, and what determines which one a community settles into, becomes a question of analyzing the structure of the consumption and cross-feeding matrices.

The species packing bound S*/M ≤ 1/2 also has practical implications. It suggests that microbial communities are, in a specific sense, under-full. Most resource diversity is not converted into species diversity. This gap could reflect historical contingencies—communities that haven't yet explored the full range of possible niches—or fundamental constraints that the model doesn't capture. Distinguishing between these possibilities requires both theoretical work like this paper and empirical studies of community assembly.

Perhaps most intriguingly, the work opens doors for engineers and biotechnologists who want to design microbial communities with specific properties. If we understand the stability landscape of the MiCRM, we can ask which perturbations will be absorbed and which will trigger collapse. We can identify species that are keystone—whose presence or absence disproportionately affects community structure. And we can design interventions, whether in the form of nutrient amendments or targeted species introductions, that steer communities toward desired attractors.

The connection to chemical reaction networks is particularly valuable here. The zero-deficiency property of the resource dynamics means that the steady-state distribution of resource abundances follows a known form. This gives engineers a tool for predicting how resources will be distributed in equilibrium, and how that distribution will change if consumption patterns shift. In a bio-reactor or a soil system, being able to predict resource distributions is a first step toward predicting community composition.

The stochastic analysis adds a crucial dimension of realism. In small populations, drift matters. A species that would persist indefinitely in a infinite-population model might go extinct in a finite population simply due to bad luck in birth-death events. The researchers' Markov chain framework allowsquantification of this risk. For applications where maintaining species diversity is important—say, in designing probiotics that should persist in the gut—the probability of extinction over relevant timescales becomes a design constraint.

What's Next

Faul and colleagues have laid mathematical groundwork, but many questions remain open.

The most pressing involves the assumptions underlying the analysis. The paper analyzes a version of the MiCRM with uniform cross-feeding, where metabolite release is equally distributed among all possible resource types. Real cross-feeding is almost certainly more structured. Species might preferentially release particular metabolites. Different resource types might be metabolized at different rates. The assumption of uniformity simplifies the mathematics, but it's not biologically motivated. Understanding whether the qualitative results—non-persistence when M < S, generic stability of surviving communities—hold under more realistic cross-feeding structures is an important next step.

The paper also assumes that interaction parameters are chosen independently at random. This is a standard assumption in theoretical ecology when direct measurements are unavailable, but it's almost certainly wrong. Real species don't consume resources independently of each other. Metabolic pathways create correlations. Phylogenetically related species often share similar consumption profiles. Whether the results are robust to realistic correlation structures—whether the non-persistence result survives, or whether correlated interactions create pockets of feasibility—is unknown.

The stochastic analysis is the most speculative part of the paper, and the least developed. The researchers propose a framework for analyzing extinction events in stochastic versions of the MiCRM, but they don't provide a complete characterization of extinction probabilities. In particular, the relationship between large-deviation rates and biological parameters—the things an experimentalist or engineer might actually measure—remains to be worked out. Connecting the Markov chain analysis to observable quantities like species abundances and resource concentrations would make the theory more directly applicable.

There's also a question of timescales that the paper doesn't fully address. The slow-fast approximation assumes that resources reach equilibrium much faster than species change. But real microbial communities might experience fluctuations on intermediate timescales—resource pulses, for instance, that don't allow equilibrium to be reached before conditions change again. The behavior of the MiCRM under non-equilibrium resource dynamics is an open question.

Finally, there's the profound question of what "stability" means in this context. Mathematical stability, as analyzed in the paper, means that small perturbations decay over time. But ecological systems face ongoing perturbations, not single shocks. A community might be mathematically stable yet experience persistent fluctuations that prevent any species from reaching equilibrium abundance. The relationship between stability in the mathematical sense and robustness in the ecological sense—between the equilibrium analysis and the actual dynamics of a community living in a changing world—isn't resolved by this paper.

These open questions shouldn't obscure the magnitude of what Faul and colleagues have accomplished. They have taken a model that was known to be empirically inadequate, extended it to incorporate a crucial biological mechanism, and derived rigorous results about persistence, feasibility, and stability. They have connected the MiCRM to well-understood mathematical structures—chemical reaction networks, random matrix theory, Markov chains—opening the door to the full arsenal of analytical tools from these fields. And they have identified a fundamental constraint on microbial community structure: when resources run out, species run out.

The practical implications will take years to fully explore. But the direction is clear. If we want to understand, predict, or engineer microbial communities, we need models that capture cross-feeding. And if we want those models to be tractable, we need the kind of rigorous analysis that Faul and colleagues have provided. The invisible economy of metabolites beneath our feet is more complex than MacArthur imagined. But with tools like the MiCRM, we're beginning to map its contours.