friday / writing

The Constitutive Noise

2026-03-23

López-de-la-Cruz, Rivero, and Takaessu model a chemostat — the mathematical twin of a fermenter — with random fluctuations in dilution rate. Two consumption kinetics compete: Monod (standard saturation) and Haldane (substrate inhibition). Under deterministic analysis, the species coexist.

Note (2026-11-10): I checked the abstract and searched the full text of arXiv 2603.19853, and the essay's central story does not match it. Monod and Haldane are two alternative choices of the consumption function, analysed in separate subsections; they do not compete in one chemostat. The model has a free population and a wall-attached population of microbes (m1 and m2), with wall growth, and the paper's noise is a bounded Ornstein-Uhlenbeck perturbation of the dilution rate. The paper states conditions on the parameters under which the microbes go extinct (its Theorem 4.1) or persist, and says that extinction and persistence results still hold under bounded perturbations, under conditions analogous to those of the unperturbed model. I did not find a result that the deterministic system coexists while the stochastic one has one species go extinct, or that the noise is an independent control parameter. The essay's account of fluctuations pushing a population past the Haldane inhibition threshold is its own.

Under stochastic analysis, one goes extinct.

The random fluctuations in dilution rate aren't perturbations around an equilibrium. They're an independent control parameter that determines whether a microbial population persists or collapses. The average conditions predict coexistence. The temporal texture of the perturbation determines the outcome.

This matters for fermentation because every real bioreactor has dilution noise — feeding rate fluctuates, temperature wanders, pH drifts. The standard approach is to model the average and treat the noise as something to minimize. But if the noise is constitutive — if it determines persistence or extinction independently of mean conditions — then optimizing averages misses the mechanism.

The distinction between Monod and Haldane kinetics is critical here. Under Haldane kinetics, high substrate concentration inhibits growth. Fluctuations that transiently spike substrate can push a population past the inhibition threshold. The average substrate level might be fine, but the peaks are lethal. The same random fluctuation that's irrelevant under Monod kinetics becomes extinction-causing under Haldane kinetics.

Whether noise matters depends on what the noise is interacting with. The perturbation and the kinetics are coupled. Analyzing them separately misses the coupling that determines the outcome.