Contribution
MacArthur (1969–70) showed that equilibria of his consumer–resource model minimize a quadratic in the consumer abundances, and could not interpret the objective outside one special case. This paper moves the optimization to the space of environmental states, states exactly when an optimization principle exists in contemporary niche theory (interaction symmetry, up to a state-dependent rescaling of abundances), and gives the objective one reading for all such models: how far the community pushes the environment from where it would rest without consumers.
Key insight
Species are constraints. Each species’ zero-net-growth isocline bounds the region of environments in which nobody can grow; the community settles where the environment is as close as possible to its unperturbed supply point inside that region, and the species that persist are the constraints that bind, at abundances equal to their Lagrange multipliers.
Assumptions
- Niche-theory form: dN_i/dt = N_i g_i(R), dR_α/dt = h_α(R) + Σ_i N_i q_iα(R); all interactions mediated by the environment R.
- Symmetry: dg_i/d(a_j N_j) = dg_j/d(a_i N_i) for some positive, possibly state-dependent a_i. Appendix B shows this holds iff q_iα = −a_i(R) b_α(R) ∂g_i/∂R_α with a diagonal b (generically), and then h_α/b_α = −∂d/∂R_α for some d.
- A stable supply point R0 of the intrinsic environmental dynamics, b_α > 0, and invertibility conditions used in Appendix B.
- Monostability for the assembly result.
Key results
- MEPP. Equilibria satisfy 0 = a_i N_i g_i, 0 = h_α/b_α − Σ a_i N_i ∂g_i/∂R_α, g_i ≤ 0, a_i N_i ≥ 0: the KKT conditions for minimizing d(R0, R) subject to g_i(R) ≤ 0, with a_i N_i the multipliers.
- The measures. Logistic resources: d = ½ Σ w_α r_α K_α⁻¹ (R_α − K_α)², with R ≥ 0 imposed explicitly (eq. 14). Interacting plants, when r_α = r and w_α ∝ a_α: d ∝ (1 − Σ a_α R_α)² (eq. 17). Predators: an added Σ u_a P_a, symmetric only if e_i = e0/η_i, which corrects MacArthur’s claim that the model is generically symmetric (eq. 21). Chemostat: a weighted Kullback–Leibler divergence, τ⁻¹ Σ w_α[R0_α ln(R0_α/R_α) − (R0_α − R_α)], which needs no explicit feasibility constraint (eq. 24).
- Asymmetric models. Byproduct secretion and Liebig’s law break symmetry; an equilibrium still minimizes the chemostat d about a shifted, self-consistent supply point R̃0, found by iteration (eqs. 27, A26).
- Rotifers. For two Brachionus species on two algae, the symmetry condition (maximal carbon uptake independent of food) holds for B. calyciflorus within error and only at low food for B. rubens; three fitted parameters per species give good fits (Table 2).
- Assembly. Every successful invasion adds a constraint, so the minimized d cannot fall: “community assembly and evolution are unidirectional”, and limit cycles or chaos in composition are ruled out in this class.
- Duality with MacArthur (Appendix A): with resource extinction handled through R̄(N) = max(0, K(1 − r⁻¹ Σ N c)), MacArthur’s Q(N) is the Lagrange dual of MEPP. MacArthur’s sum over all resources is wrong when any resource goes extinct; the correct Q is piecewise.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Symmetric environmentally mediated interactions are equivalent to impact vectors of the form −a b ∂g/∂R | moderate | Appendix B, with genericity and invertibility assumptions |
| C2 | In symmetric niche models the uninvadable equilibrium minimizes d over the uninvadable region | strong | KKT argument; numerical agreement in all examples; checked on Fig. 1 |
| C3 | d is non-decreasing under successive invasions | strong, given monostability and that the pool’s species stay as constraints | constraint argument |
| C4 | MacArthur’s principle is the Lagrange dual of MEPP | strong | Appendix A |
| C5 | d correlates positively with species richness and could explain diversity–productivity relations | weak | discussion only |
Concepts
- Supply point R0: the environmental state the intrinsic dynamics rest at without consumers.
- ZNGI: the surface g_i(R) = 0; the outer envelope of the pool’s ZNGIs bounds the uninvadable region Ω.
- Environmental perturbation d(R0, R): the function whose gradient, rescaled by b, is the supply vector; zero at R0.
- Interaction symmetry: equal effects of a rescaled abundance change of i on j and of j on i, mediated through the environment.
Connections
Gatto (1982, 1990) saw the rescaling freedom and used the KKT conditions for MacArthur’s model; this paper generalizes his. The companion LIT-1231 runs the same duality from the optimization side, for quadratic programming. Chizat and Bach’s optimality condition for measures (LIT-1232) is the continuum form of the same KKT structure.
Bearing on the record
- CLAIM-369 is the population-space side of this paper’s result. Its objective ½‖b − Un‖² + cᵀn is MacArthur’s Q with resources eliminated, and its KKT points are the uninvadable equilibria. Two things this reading adds. First, read as the fast-resource reduction of a consumer–resource model, the claim’s A = UᵀU and r = Uᵀb − c hold only while every resource stays positive. Once a resource is driven extinct, its row drops out and the effective objective is MacArthur’s piecewise Q (Appendix A), a point MacArthur himself missed. The claim states the Lotka–Volterra form directly, so it is not wrong, but a third boundary belongs beside its two: the reduction from resources. Second, the dual (environment-space) reading gives the claim’s objective a physical meaning: the perturbation the active components impose on the supply.
- QUESTION-083, QUESTION-132: in this class, assembly is monotone in d. Differentiated specialists that each recover from rarity are exactly the binding constraints, and no assembly sequence can undo an invasion except by killing species. That is a structural limit on “revisable” authority in any analogue built on symmetric resource competition: revision needs an external removal or an asymmetry.
- No instruction for machine-learning practice; nothing for the anthology.
Limitations
- Symmetry is the whole hinge. The asymmetric extension needs the equilibrium to construct d, so it is a reformulation, not a prediction.
- The monotone-assembly result also needs monostability, and needs every species once introduced to remain in the pool as a constraint; it says nothing about systems with alternative stable states, which Liebig’s law “generically” has.
- The rotifer test checks the symmetry condition on growth curves; the zooplankton assembly experiment of Figure 5 is hypothetical.
- Many displayed equations in Appendices B and C did not extract legibly from the arXiv PDF and were followed from the prose.
- No errors of substance found. Small slips: “a optimization principle”, “resource contribute”.
Open questions
- For asymmetric niche models with alternative stable states, is there any monotone quantity of assembly?
- Does d, measured in a real community, rise with richness as C5 says?